blob: 8dbf6953845ba61097703006553c34d1d83bd81f [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/jbd2.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070029#include <linux/highuid.h>
30#include <linux/pagemap.h>
31#include <linux/quotaops.h>
32#include <linux/string.h>
33#include <linux/buffer_head.h>
34#include <linux/writeback.h>
Alex Tomas64769242008-07-11 19:27:31 -040035#include <linux/pagevec.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070036#include <linux/mpage.h>
37#include <linux/uio.h>
38#include <linux/bio.h>
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040039#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070040#include "xattr.h"
41#include "acl.h"
Mingming Caod2a17632008-07-14 17:52:37 -040042#include "ext4_extents.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070043
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -040044#define MPAGE_DA_EXTENT_TAIL 0x01
45
Jan Kara678aaf42008-07-11 19:27:31 -040046static inline int ext4_begin_ordered_truncate(struct inode *inode,
47 loff_t new_size)
48{
49 return jbd2_journal_begin_ordered_truncate(&EXT4_I(inode)->jinode,
50 new_size);
51}
52
Alex Tomas64769242008-07-11 19:27:31 -040053static void ext4_invalidatepage(struct page *page, unsigned long offset);
54
Dave Kleikampac27a0e2006-10-11 01:20:50 -070055/*
56 * Test whether an inode is a fast symlink.
57 */
Mingming Cao617ba132006-10-11 01:20:53 -070058static int ext4_inode_is_fast_symlink(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070059{
Mingming Cao617ba132006-10-11 01:20:53 -070060 int ea_blocks = EXT4_I(inode)->i_file_acl ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -070061 (inode->i_sb->s_blocksize >> 9) : 0;
62
63 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
64}
65
66/*
Mingming Cao617ba132006-10-11 01:20:53 -070067 * The ext4 forget function must perform a revoke if we are freeing data
Dave Kleikampac27a0e2006-10-11 01:20:50 -070068 * which has been journaled. Metadata (eg. indirect blocks) must be
69 * revoked in all cases.
70 *
71 * "bh" may be NULL: a metadata block may have been freed from memory
72 * but there may still be a record of it in the journal, and that record
73 * still needs to be revoked.
74 */
Mingming Cao617ba132006-10-11 01:20:53 -070075int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
76 struct buffer_head *bh, ext4_fsblk_t blocknr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070077{
78 int err;
79
80 might_sleep();
81
82 BUFFER_TRACE(bh, "enter");
83
84 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
85 "data mode %lx\n",
86 bh, is_metadata, inode->i_mode,
87 test_opt(inode->i_sb, DATA_FLAGS));
88
89 /* Never use the revoke function if we are doing full data
90 * journaling: there is no need to, and a V1 superblock won't
91 * support it. Otherwise, only skip the revoke on un-journaled
92 * data blocks. */
93
Mingming Cao617ba132006-10-11 01:20:53 -070094 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
95 (!is_metadata && !ext4_should_journal_data(inode))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -070096 if (bh) {
Mingming Caodab291a2006-10-11 01:21:01 -070097 BUFFER_TRACE(bh, "call jbd2_journal_forget");
Mingming Cao617ba132006-10-11 01:20:53 -070098 return ext4_journal_forget(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070099 }
100 return 0;
101 }
102
103 /*
104 * data!=journal && (is_metadata || should_journal_data(inode))
105 */
Mingming Cao617ba132006-10-11 01:20:53 -0700106 BUFFER_TRACE(bh, "call ext4_journal_revoke");
107 err = ext4_journal_revoke(handle, blocknr, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700108 if (err)
Harvey Harrison46e665e2008-04-17 10:38:59 -0400109 ext4_abort(inode->i_sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700110 "error %d when attempting revoke", err);
111 BUFFER_TRACE(bh, "exit");
112 return err;
113}
114
115/*
116 * Work out how many blocks we need to proceed with the next chunk of a
117 * truncate transaction.
118 */
119static unsigned long blocks_for_truncate(struct inode *inode)
120{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500121 ext4_lblk_t needed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700122
123 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
124
125 /* Give ourselves just enough room to cope with inodes in which
126 * i_blocks is corrupt: we've seen disk corruptions in the past
127 * which resulted in random data in an inode which looked enough
Mingming Cao617ba132006-10-11 01:20:53 -0700128 * like a regular file for ext4 to try to delete it. Things
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700129 * will go a bit crazy if that happens, but at least we should
130 * try not to panic the whole kernel. */
131 if (needed < 2)
132 needed = 2;
133
134 /* But we need to bound the transaction so we don't overflow the
135 * journal. */
Mingming Cao617ba132006-10-11 01:20:53 -0700136 if (needed > EXT4_MAX_TRANS_DATA)
137 needed = EXT4_MAX_TRANS_DATA;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700138
Mingming Cao617ba132006-10-11 01:20:53 -0700139 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700140}
141
142/*
143 * Truncate transactions can be complex and absolutely huge. So we need to
144 * be able to restart the transaction at a conventient checkpoint to make
145 * sure we don't overflow the journal.
146 *
147 * start_transaction gets us a new handle for a truncate transaction,
148 * and extend_transaction tries to extend the existing one a bit. If
149 * extend fails, we need to propagate the failure up and restart the
150 * transaction in the top-level truncate loop. --sct
151 */
152static handle_t *start_transaction(struct inode *inode)
153{
154 handle_t *result;
155
Mingming Cao617ba132006-10-11 01:20:53 -0700156 result = ext4_journal_start(inode, blocks_for_truncate(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700157 if (!IS_ERR(result))
158 return result;
159
Mingming Cao617ba132006-10-11 01:20:53 -0700160 ext4_std_error(inode->i_sb, PTR_ERR(result));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700161 return result;
162}
163
164/*
165 * Try to extend this transaction for the purposes of truncation.
166 *
167 * Returns 0 if we managed to create more room. If we can't create more
168 * room, and the transaction must be restarted we return 1.
169 */
170static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
171{
Mingming Cao617ba132006-10-11 01:20:53 -0700172 if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700173 return 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700174 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700175 return 0;
176 return 1;
177}
178
179/*
180 * Restart the transaction associated with *handle. This does a commit,
181 * so before we call here everything must be consistently dirtied against
182 * this transaction.
183 */
Mingming Cao617ba132006-10-11 01:20:53 -0700184static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700185{
186 jbd_debug(2, "restarting handle %p\n", handle);
Mingming Cao617ba132006-10-11 01:20:53 -0700187 return ext4_journal_restart(handle, blocks_for_truncate(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700188}
189
190/*
191 * Called at the last iput() if i_nlink is zero.
192 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400193void ext4_delete_inode(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700194{
195 handle_t *handle;
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400196 int err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700197
Jan Kara678aaf42008-07-11 19:27:31 -0400198 if (ext4_should_order_data(inode))
199 ext4_begin_ordered_truncate(inode, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700200 truncate_inode_pages(&inode->i_data, 0);
201
202 if (is_bad_inode(inode))
203 goto no_delete;
204
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400205 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700206 if (IS_ERR(handle)) {
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400207 ext4_std_error(inode->i_sb, PTR_ERR(handle));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700208 /*
209 * If we're going to skip the normal cleanup, we still need to
210 * make sure that the in-core orphan linked list is properly
211 * cleaned up.
212 */
Mingming Cao617ba132006-10-11 01:20:53 -0700213 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700214 goto no_delete;
215 }
216
217 if (IS_SYNC(inode))
218 handle->h_sync = 1;
219 inode->i_size = 0;
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400220 err = ext4_mark_inode_dirty(handle, inode);
221 if (err) {
222 ext4_warning(inode->i_sb, __func__,
223 "couldn't mark inode dirty (err %d)", err);
224 goto stop_handle;
225 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700226 if (inode->i_blocks)
Mingming Cao617ba132006-10-11 01:20:53 -0700227 ext4_truncate(inode);
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400228
229 /*
230 * ext4_ext_truncate() doesn't reserve any slop when it
231 * restarts journal transactions; therefore there may not be
232 * enough credits left in the handle to remove the inode from
233 * the orphan list and set the dtime field.
234 */
235 if (handle->h_buffer_credits < 3) {
236 err = ext4_journal_extend(handle, 3);
237 if (err > 0)
238 err = ext4_journal_restart(handle, 3);
239 if (err != 0) {
240 ext4_warning(inode->i_sb, __func__,
241 "couldn't extend journal (err %d)", err);
242 stop_handle:
243 ext4_journal_stop(handle);
244 goto no_delete;
245 }
246 }
247
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700248 /*
Mingming Cao617ba132006-10-11 01:20:53 -0700249 * Kill off the orphan record which ext4_truncate created.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700250 * AKPM: I think this can be inside the above `if'.
Mingming Cao617ba132006-10-11 01:20:53 -0700251 * Note that ext4_orphan_del() has to be able to cope with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700252 * deletion of a non-existent orphan - this is because we don't
Mingming Cao617ba132006-10-11 01:20:53 -0700253 * know if ext4_truncate() actually created an orphan record.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700254 * (Well, we could do this if we need to, but heck - it works)
255 */
Mingming Cao617ba132006-10-11 01:20:53 -0700256 ext4_orphan_del(handle, inode);
257 EXT4_I(inode)->i_dtime = get_seconds();
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700258
259 /*
260 * One subtle ordering requirement: if anything has gone wrong
261 * (transaction abort, IO errors, whatever), then we can still
262 * do these next steps (the fs will already have been marked as
263 * having errors), but we can't free the inode if the mark_dirty
264 * fails.
265 */
Mingming Cao617ba132006-10-11 01:20:53 -0700266 if (ext4_mark_inode_dirty(handle, inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700267 /* If that failed, just do the required in-core inode clear. */
268 clear_inode(inode);
269 else
Mingming Cao617ba132006-10-11 01:20:53 -0700270 ext4_free_inode(handle, inode);
271 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700272 return;
273no_delete:
274 clear_inode(inode); /* We must guarantee clearing of inode... */
275}
276
277typedef struct {
278 __le32 *p;
279 __le32 key;
280 struct buffer_head *bh;
281} Indirect;
282
283static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
284{
285 p->key = *(p->p = v);
286 p->bh = bh;
287}
288
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700289/**
Mingming Cao617ba132006-10-11 01:20:53 -0700290 * ext4_block_to_path - parse the block number into array of offsets
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700291 * @inode: inode in question (we are only interested in its superblock)
292 * @i_block: block number to be parsed
293 * @offsets: array to store the offsets in
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400294 * @boundary: set this non-zero if the referred-to block is likely to be
295 * followed (on disk) by an indirect block.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700296 *
Mingming Cao617ba132006-10-11 01:20:53 -0700297 * To store the locations of file's data ext4 uses a data structure common
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700298 * for UNIX filesystems - tree of pointers anchored in the inode, with
299 * data blocks at leaves and indirect blocks in intermediate nodes.
300 * This function translates the block number into path in that tree -
301 * return value is the path length and @offsets[n] is the offset of
302 * pointer to (n+1)th node in the nth one. If @block is out of range
303 * (negative or too large) warning is printed and zero returned.
304 *
305 * Note: function doesn't find node addresses, so no IO is needed. All
306 * we need to know is the capacity of indirect blocks (taken from the
307 * inode->i_sb).
308 */
309
310/*
311 * Portability note: the last comparison (check that we fit into triple
312 * indirect block) is spelled differently, because otherwise on an
313 * architecture with 32-bit longs and 8Kb pages we might get into trouble
314 * if our filesystem had 8Kb blocks. We might use long long, but that would
315 * kill us on x86. Oh, well, at least the sign propagation does not matter -
316 * i_block would have to be negative in the very beginning, so we would not
317 * get there at all.
318 */
319
Mingming Cao617ba132006-10-11 01:20:53 -0700320static int ext4_block_to_path(struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500321 ext4_lblk_t i_block,
322 ext4_lblk_t offsets[4], int *boundary)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700323{
Mingming Cao617ba132006-10-11 01:20:53 -0700324 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
325 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
326 const long direct_blocks = EXT4_NDIR_BLOCKS,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700327 indirect_blocks = ptrs,
328 double_blocks = (1 << (ptrs_bits * 2));
329 int n = 0;
330 int final = 0;
331
332 if (i_block < 0) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400333 ext4_warning(inode->i_sb, "ext4_block_to_path", "block < 0");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700334 } else if (i_block < direct_blocks) {
335 offsets[n++] = i_block;
336 final = direct_blocks;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400337 } else if ((i_block -= direct_blocks) < indirect_blocks) {
Mingming Cao617ba132006-10-11 01:20:53 -0700338 offsets[n++] = EXT4_IND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700339 offsets[n++] = i_block;
340 final = ptrs;
341 } else if ((i_block -= indirect_blocks) < double_blocks) {
Mingming Cao617ba132006-10-11 01:20:53 -0700342 offsets[n++] = EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700343 offsets[n++] = i_block >> ptrs_bits;
344 offsets[n++] = i_block & (ptrs - 1);
345 final = ptrs;
346 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
Mingming Cao617ba132006-10-11 01:20:53 -0700347 offsets[n++] = EXT4_TIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700348 offsets[n++] = i_block >> (ptrs_bits * 2);
349 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
350 offsets[n++] = i_block & (ptrs - 1);
351 final = ptrs;
352 } else {
Eric Sandeene2b46572008-01-28 23:58:27 -0500353 ext4_warning(inode->i_sb, "ext4_block_to_path",
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500354 "block %lu > max",
Eric Sandeene2b46572008-01-28 23:58:27 -0500355 i_block + direct_blocks +
356 indirect_blocks + double_blocks);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700357 }
358 if (boundary)
359 *boundary = final - 1 - (i_block & (ptrs - 1));
360 return n;
361}
362
363/**
Mingming Cao617ba132006-10-11 01:20:53 -0700364 * ext4_get_branch - read the chain of indirect blocks leading to data
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700365 * @inode: inode in question
366 * @depth: depth of the chain (1 - direct pointer, etc.)
367 * @offsets: offsets of pointers in inode/indirect blocks
368 * @chain: place to store the result
369 * @err: here we store the error value
370 *
371 * Function fills the array of triples <key, p, bh> and returns %NULL
372 * if everything went OK or the pointer to the last filled triple
373 * (incomplete one) otherwise. Upon the return chain[i].key contains
374 * the number of (i+1)-th block in the chain (as it is stored in memory,
375 * i.e. little-endian 32-bit), chain[i].p contains the address of that
376 * number (it points into struct inode for i==0 and into the bh->b_data
377 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
378 * block for i>0 and NULL for i==0. In other words, it holds the block
379 * numbers of the chain, addresses they were taken from (and where we can
380 * verify that chain did not change) and buffer_heads hosting these
381 * numbers.
382 *
383 * Function stops when it stumbles upon zero pointer (absent block)
384 * (pointer to last triple returned, *@err == 0)
385 * or when it gets an IO error reading an indirect block
386 * (ditto, *@err == -EIO)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700387 * or when it reads all @depth-1 indirect blocks successfully and finds
388 * the whole chain, all way to the data (returns %NULL, *err == 0).
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -0500389 *
390 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500391 * down_read(&EXT4_I(inode)->i_data_sem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700392 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500393static Indirect *ext4_get_branch(struct inode *inode, int depth,
394 ext4_lblk_t *offsets,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700395 Indirect chain[4], int *err)
396{
397 struct super_block *sb = inode->i_sb;
398 Indirect *p = chain;
399 struct buffer_head *bh;
400
401 *err = 0;
402 /* i_data is not going away, no lock needed */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400403 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700404 if (!p->key)
405 goto no_block;
406 while (--depth) {
407 bh = sb_bread(sb, le32_to_cpu(p->key));
408 if (!bh)
409 goto failure;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400410 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700411 /* Reader: end */
412 if (!p->key)
413 goto no_block;
414 }
415 return NULL;
416
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700417failure:
418 *err = -EIO;
419no_block:
420 return p;
421}
422
423/**
Mingming Cao617ba132006-10-11 01:20:53 -0700424 * ext4_find_near - find a place for allocation with sufficient locality
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700425 * @inode: owner
426 * @ind: descriptor of indirect block.
427 *
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000428 * This function returns the preferred place for block allocation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700429 * It is used when heuristic for sequential allocation fails.
430 * Rules are:
431 * + if there is a block to the left of our position - allocate near it.
432 * + if pointer will live in indirect block - allocate near that block.
433 * + if pointer will live in inode - allocate in the same
434 * cylinder group.
435 *
436 * In the latter case we colour the starting block by the callers PID to
437 * prevent it from clashing with concurrent allocations for a different inode
438 * in the same block group. The PID is used here so that functionally related
439 * files will be close-by on-disk.
440 *
441 * Caller must make sure that @ind is valid and will stay that way.
442 */
Mingming Cao617ba132006-10-11 01:20:53 -0700443static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700444{
Mingming Cao617ba132006-10-11 01:20:53 -0700445 struct ext4_inode_info *ei = EXT4_I(inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400446 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700447 __le32 *p;
Mingming Cao617ba132006-10-11 01:20:53 -0700448 ext4_fsblk_t bg_start;
Valerie Clement74d34872008-02-15 13:43:07 -0500449 ext4_fsblk_t last_block;
Mingming Cao617ba132006-10-11 01:20:53 -0700450 ext4_grpblk_t colour;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700451
452 /* Try to find previous block */
453 for (p = ind->p - 1; p >= start; p--) {
454 if (*p)
455 return le32_to_cpu(*p);
456 }
457
458 /* No such thing, so let's try location of indirect block */
459 if (ind->bh)
460 return ind->bh->b_blocknr;
461
462 /*
463 * It is going to be referred to from the inode itself? OK, just put it
464 * into the same cylinder group then.
465 */
Mingming Cao617ba132006-10-11 01:20:53 -0700466 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
Valerie Clement74d34872008-02-15 13:43:07 -0500467 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
468
469 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
470 colour = (current->pid % 16) *
Mingming Cao617ba132006-10-11 01:20:53 -0700471 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
Valerie Clement74d34872008-02-15 13:43:07 -0500472 else
473 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700474 return bg_start + colour;
475}
476
477/**
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000478 * ext4_find_goal - find a preferred place for allocation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700479 * @inode: owner
480 * @block: block we want
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700481 * @partial: pointer to the last triple within a chain
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700482 *
Benoit Boissinot1cc8dcf52008-04-21 22:45:55 +0000483 * Normally this function find the preferred place for block allocation,
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800484 * returns it.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700485 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500486static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800487 Indirect *partial)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700488{
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700489 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -0400490 * XXX need to get goal block from mballoc's data structures
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700491 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700492
Mingming Cao617ba132006-10-11 01:20:53 -0700493 return ext4_find_near(inode, partial);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700494}
495
496/**
Mingming Cao617ba132006-10-11 01:20:53 -0700497 * ext4_blks_to_allocate: Look up the block map and count the number
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700498 * of direct blocks need to be allocated for the given branch.
499 *
500 * @branch: chain of indirect blocks
501 * @k: number of blocks need for indirect blocks
502 * @blks: number of data blocks to be mapped.
503 * @blocks_to_boundary: the offset in the indirect block
504 *
505 * return the total number of blocks to be allocate, including the
506 * direct and indirect blocks.
507 */
Mingming Cao617ba132006-10-11 01:20:53 -0700508static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700509 int blocks_to_boundary)
510{
511 unsigned long count = 0;
512
513 /*
514 * Simple case, [t,d]Indirect block(s) has not allocated yet
515 * then it's clear blocks on that path have not allocated
516 */
517 if (k > 0) {
518 /* right now we don't handle cross boundary allocation */
519 if (blks < blocks_to_boundary + 1)
520 count += blks;
521 else
522 count += blocks_to_boundary + 1;
523 return count;
524 }
525
526 count++;
527 while (count < blks && count <= blocks_to_boundary &&
528 le32_to_cpu(*(branch[0].p + count)) == 0) {
529 count++;
530 }
531 return count;
532}
533
534/**
Mingming Cao617ba132006-10-11 01:20:53 -0700535 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700536 * @indirect_blks: the number of blocks need to allocate for indirect
537 * blocks
538 *
539 * @new_blocks: on return it will store the new block numbers for
540 * the indirect blocks(if needed) and the first direct block,
541 * @blks: on return it will store the total number of allocated
542 * direct blocks
543 */
Mingming Cao617ba132006-10-11 01:20:53 -0700544static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400545 ext4_lblk_t iblock, ext4_fsblk_t goal,
546 int indirect_blks, int blks,
547 ext4_fsblk_t new_blocks[4], int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700548{
549 int target, i;
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400550 unsigned long count = 0, blk_allocated = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700551 int index = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700552 ext4_fsblk_t current_block = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700553 int ret = 0;
554
555 /*
556 * Here we try to allocate the requested multiple blocks at once,
557 * on a best-effort basis.
558 * To build a branch, we should allocate blocks for
559 * the indirect blocks(if not allocated yet), and at least
560 * the first direct block of this branch. That's the
561 * minimum number of blocks need to allocate(required)
562 */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400563 /* first we try to allocate the indirect blocks */
564 target = indirect_blks;
565 while (target > 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700566 count = target;
567 /* allocating blocks for indirect blocks and direct blocks */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400568 current_block = ext4_new_meta_blocks(handle, inode,
569 goal, &count, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700570 if (*err)
571 goto failed_out;
572
573 target -= count;
574 /* allocate blocks for indirect blocks */
575 while (index < indirect_blks && count) {
576 new_blocks[index++] = current_block++;
577 count--;
578 }
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400579 if (count > 0) {
580 /*
581 * save the new block number
582 * for the first direct block
583 */
584 new_blocks[index] = current_block;
585 printk(KERN_INFO "%s returned more blocks than "
586 "requested\n", __func__);
587 WARN_ON(1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700588 break;
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400589 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700590 }
591
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400592 target = blks - count ;
593 blk_allocated = count;
594 if (!target)
595 goto allocated;
596 /* Now allocate data blocks */
597 count = target;
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400598 /* allocating blocks for data blocks */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400599 current_block = ext4_new_blocks(handle, inode, iblock,
600 goal, &count, err);
601 if (*err && (target == blks)) {
602 /*
603 * if the allocation failed and we didn't allocate
604 * any blocks before
605 */
606 goto failed_out;
607 }
608 if (!*err) {
609 if (target == blks) {
610 /*
611 * save the new block number
612 * for the first direct block
613 */
614 new_blocks[index] = current_block;
615 }
616 blk_allocated += count;
617 }
618allocated:
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700619 /* total number of blocks allocated for direct blocks */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400620 ret = blk_allocated;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700621 *err = 0;
622 return ret;
623failed_out:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400624 for (i = 0; i < index; i++)
Alex Tomasc9de5602008-01-29 00:19:52 -0500625 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700626 return ret;
627}
628
629/**
Mingming Cao617ba132006-10-11 01:20:53 -0700630 * ext4_alloc_branch - allocate and set up a chain of blocks.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700631 * @inode: owner
632 * @indirect_blks: number of allocated indirect blocks
633 * @blks: number of allocated direct blocks
634 * @offsets: offsets (in the blocks) to store the pointers to next.
635 * @branch: place to store the chain in.
636 *
637 * This function allocates blocks, zeroes out all but the last one,
638 * links them into chain and (if we are synchronous) writes them to disk.
639 * In other words, it prepares a branch that can be spliced onto the
640 * inode. It stores the information about that chain in the branch[], in
Mingming Cao617ba132006-10-11 01:20:53 -0700641 * the same format as ext4_get_branch() would do. We are calling it after
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700642 * we had read the existing part of chain and partial points to the last
643 * triple of that (one with zero ->key). Upon the exit we have the same
Mingming Cao617ba132006-10-11 01:20:53 -0700644 * picture as after the successful ext4_get_block(), except that in one
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700645 * place chain is disconnected - *branch->p is still zero (we did not
646 * set the last link), but branch->key contains the number that should
647 * be placed into *branch->p to fill that gap.
648 *
649 * If allocation fails we free all blocks we've allocated (and forget
650 * their buffer_heads) and return the error value the from failed
Mingming Cao617ba132006-10-11 01:20:53 -0700651 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700652 * as described above and return 0.
653 */
Mingming Cao617ba132006-10-11 01:20:53 -0700654static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400655 ext4_lblk_t iblock, int indirect_blks,
656 int *blks, ext4_fsblk_t goal,
657 ext4_lblk_t *offsets, Indirect *branch)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700658{
659 int blocksize = inode->i_sb->s_blocksize;
660 int i, n = 0;
661 int err = 0;
662 struct buffer_head *bh;
663 int num;
Mingming Cao617ba132006-10-11 01:20:53 -0700664 ext4_fsblk_t new_blocks[4];
665 ext4_fsblk_t current_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700666
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400667 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700668 *blks, new_blocks, &err);
669 if (err)
670 return err;
671
672 branch[0].key = cpu_to_le32(new_blocks[0]);
673 /*
674 * metadata blocks and data blocks are allocated.
675 */
676 for (n = 1; n <= indirect_blks; n++) {
677 /*
678 * Get buffer_head for parent block, zero it out
679 * and set the pointer to new one, then send
680 * parent to disk.
681 */
682 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
683 branch[n].bh = bh;
684 lock_buffer(bh);
685 BUFFER_TRACE(bh, "call get_create_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700686 err = ext4_journal_get_create_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700687 if (err) {
688 unlock_buffer(bh);
689 brelse(bh);
690 goto failed;
691 }
692
693 memset(bh->b_data, 0, blocksize);
694 branch[n].p = (__le32 *) bh->b_data + offsets[n];
695 branch[n].key = cpu_to_le32(new_blocks[n]);
696 *branch[n].p = branch[n].key;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400697 if (n == indirect_blks) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700698 current_block = new_blocks[n];
699 /*
700 * End of chain, update the last new metablock of
701 * the chain to point to the new allocated
702 * data blocks numbers
703 */
704 for (i=1; i < num; i++)
705 *(branch[n].p + i) = cpu_to_le32(++current_block);
706 }
707 BUFFER_TRACE(bh, "marking uptodate");
708 set_buffer_uptodate(bh);
709 unlock_buffer(bh);
710
Mingming Cao617ba132006-10-11 01:20:53 -0700711 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
712 err = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700713 if (err)
714 goto failed;
715 }
716 *blks = num;
717 return err;
718failed:
719 /* Allocation failed, free what we already allocated */
720 for (i = 1; i <= n ; i++) {
Mingming Caodab291a2006-10-11 01:21:01 -0700721 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
Mingming Cao617ba132006-10-11 01:20:53 -0700722 ext4_journal_forget(handle, branch[i].bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700723 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400724 for (i = 0; i < indirect_blks; i++)
Alex Tomasc9de5602008-01-29 00:19:52 -0500725 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700726
Alex Tomasc9de5602008-01-29 00:19:52 -0500727 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700728
729 return err;
730}
731
732/**
Mingming Cao617ba132006-10-11 01:20:53 -0700733 * ext4_splice_branch - splice the allocated branch onto inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700734 * @inode: owner
735 * @block: (logical) number of block we are adding
736 * @chain: chain of indirect blocks (with a missing link - see
Mingming Cao617ba132006-10-11 01:20:53 -0700737 * ext4_alloc_branch)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700738 * @where: location of missing link
739 * @num: number of indirect blocks we are adding
740 * @blks: number of direct blocks we are adding
741 *
742 * This function fills the missing link and does all housekeeping needed in
743 * inode (->i_blocks, etc.). In case of success we end up with the full
744 * chain to new block and return 0.
745 */
Mingming Cao617ba132006-10-11 01:20:53 -0700746static int ext4_splice_branch(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500747 ext4_lblk_t block, Indirect *where, int num, int blks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700748{
749 int i;
750 int err = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700751 ext4_fsblk_t current_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700752
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700753 /*
754 * If we're splicing into a [td]indirect block (as opposed to the
755 * inode) then we need to get write access to the [td]indirect block
756 * before the splice.
757 */
758 if (where->bh) {
759 BUFFER_TRACE(where->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700760 err = ext4_journal_get_write_access(handle, where->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700761 if (err)
762 goto err_out;
763 }
764 /* That's it */
765
766 *where->p = where->key;
767
768 /*
769 * Update the host buffer_head or inode to point to more just allocated
770 * direct blocks blocks
771 */
772 if (num == 0 && blks > 1) {
773 current_block = le32_to_cpu(where->key) + 1;
774 for (i = 1; i < blks; i++)
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400775 *(where->p + i) = cpu_to_le32(current_block++);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700776 }
777
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700778 /* We are done with atomic stuff, now do the rest of housekeeping */
779
Kalpak Shahef7f3832007-07-18 09:15:20 -0400780 inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -0700781 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700782
783 /* had we spliced it onto indirect block? */
784 if (where->bh) {
785 /*
786 * If we spliced it onto an indirect block, we haven't
787 * altered the inode. Note however that if it is being spliced
788 * onto an indirect block at the very end of the file (the
789 * file is growing) then we *will* alter the inode to reflect
790 * the new i_size. But that is not done here - it is done in
Mingming Cao617ba132006-10-11 01:20:53 -0700791 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700792 */
793 jbd_debug(5, "splicing indirect only\n");
Mingming Cao617ba132006-10-11 01:20:53 -0700794 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
795 err = ext4_journal_dirty_metadata(handle, where->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700796 if (err)
797 goto err_out;
798 } else {
799 /*
800 * OK, we spliced it into the inode itself on a direct block.
801 * Inode was dirtied above.
802 */
803 jbd_debug(5, "splicing direct\n");
804 }
805 return err;
806
807err_out:
808 for (i = 1; i <= num; i++) {
Mingming Caodab291a2006-10-11 01:21:01 -0700809 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
Mingming Cao617ba132006-10-11 01:20:53 -0700810 ext4_journal_forget(handle, where[i].bh);
Alex Tomasc9de5602008-01-29 00:19:52 -0500811 ext4_free_blocks(handle, inode,
812 le32_to_cpu(where[i-1].key), 1, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700813 }
Alex Tomasc9de5602008-01-29 00:19:52 -0500814 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700815
816 return err;
817}
818
819/*
820 * Allocation strategy is simple: if we have to allocate something, we will
821 * have to go the whole way to leaf. So let's do it before attaching anything
822 * to tree, set linkage between the newborn blocks, write them if sync is
823 * required, recheck the path, free and repeat if check fails, otherwise
824 * set the last missing link (that will protect us from any truncate-generated
825 * removals - all blocks on the path are immune now) and possibly force the
826 * write on the parent block.
827 * That has a nice additional property: no special recovery from the failed
828 * allocations is needed - we simply release blocks and do not touch anything
829 * reachable from inode.
830 *
831 * `handle' can be NULL if create == 0.
832 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700833 * return > 0, # of blocks mapped or allocated.
834 * return = 0, if plain lookup failed.
835 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -0500836 *
837 *
838 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500839 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
840 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700841 */
Mingming Cao617ba132006-10-11 01:20:53 -0700842int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500843 ext4_lblk_t iblock, unsigned long maxblocks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700844 struct buffer_head *bh_result,
845 int create, int extend_disksize)
846{
847 int err = -EIO;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500848 ext4_lblk_t offsets[4];
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700849 Indirect chain[4];
850 Indirect *partial;
Mingming Cao617ba132006-10-11 01:20:53 -0700851 ext4_fsblk_t goal;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700852 int indirect_blks;
853 int blocks_to_boundary = 0;
854 int depth;
Mingming Cao617ba132006-10-11 01:20:53 -0700855 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700856 int count = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700857 ext4_fsblk_t first_block = 0;
Mingming Cao61628a32008-07-11 19:27:31 -0400858 loff_t disksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700859
860
Alex Tomasa86c6182006-10-11 01:21:03 -0700861 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700862 J_ASSERT(handle != NULL || create == 0);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500863 depth = ext4_block_to_path(inode, iblock, offsets,
864 &blocks_to_boundary);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700865
866 if (depth == 0)
867 goto out;
868
Mingming Cao617ba132006-10-11 01:20:53 -0700869 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700870
871 /* Simplest case - block found, no allocation needed */
872 if (!partial) {
873 first_block = le32_to_cpu(chain[depth - 1].key);
874 clear_buffer_new(bh_result);
875 count++;
876 /*map more blocks*/
877 while (count < maxblocks && count <= blocks_to_boundary) {
Mingming Cao617ba132006-10-11 01:20:53 -0700878 ext4_fsblk_t blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700879
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700880 blk = le32_to_cpu(*(chain[depth-1].p + count));
881
882 if (blk == first_block + count)
883 count++;
884 else
885 break;
886 }
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -0500887 goto got_it;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700888 }
889
890 /* Next simple case - plain lookup or failed read of indirect block */
891 if (!create || err == -EIO)
892 goto cleanup;
893
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700894 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -0400895 * Okay, we need to do block allocation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700896 */
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800897 goal = ext4_find_goal(inode, iblock, partial);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700898
899 /* the number of blocks need to allocate for [d,t]indirect blocks */
900 indirect_blks = (chain + depth) - partial - 1;
901
902 /*
903 * Next look up the indirect map to count the totoal number of
904 * direct blocks to allocate for this branch.
905 */
Mingming Cao617ba132006-10-11 01:20:53 -0700906 count = ext4_blks_to_allocate(partial, indirect_blks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700907 maxblocks, blocks_to_boundary);
908 /*
Mingming Cao617ba132006-10-11 01:20:53 -0700909 * Block out ext4_truncate while we alter the tree
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700910 */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400911 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
912 &count, goal,
913 offsets + (partial - chain), partial);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700914
915 /*
Mingming Cao617ba132006-10-11 01:20:53 -0700916 * The ext4_splice_branch call will free and forget any buffers
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700917 * on the new chain if there is a failure, but that risks using
918 * up transaction credits, especially for bitmaps where the
919 * credits cannot be returned. Can we handle this somehow? We
920 * may need to return -EAGAIN upwards in the worst case. --sct
921 */
922 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -0700923 err = ext4_splice_branch(handle, inode, iblock,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700924 partial, indirect_blks, count);
925 /*
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500926 * i_disksize growing is protected by i_data_sem. Don't forget to
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700927 * protect it if you're about to implement concurrent
Mingming Cao617ba132006-10-11 01:20:53 -0700928 * ext4_get_block() -bzzz
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700929 */
Mingming Cao61628a32008-07-11 19:27:31 -0400930 if (!err && extend_disksize) {
931 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
932 if (disksize > i_size_read(inode))
933 disksize = i_size_read(inode);
934 if (disksize > ei->i_disksize)
935 ei->i_disksize = disksize;
936 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700937 if (err)
938 goto cleanup;
939
940 set_buffer_new(bh_result);
941got_it:
942 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
943 if (count > blocks_to_boundary)
944 set_buffer_boundary(bh_result);
945 err = count;
946 /* Clean up and exit */
947 partial = chain + depth - 1; /* the whole chain */
948cleanup:
949 while (partial > chain) {
950 BUFFER_TRACE(partial->bh, "call brelse");
951 brelse(partial->bh);
952 partial--;
953 }
954 BUFFER_TRACE(bh_result, "returned");
955out:
956 return err;
957}
958
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400959/*
960 * Calculate the number of metadata blocks need to reserve
961 * to allocate @blocks for non extent file based file
962 */
963static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
964{
965 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
966 int ind_blks, dind_blks, tind_blks;
967
968 /* number of new indirect blocks needed */
969 ind_blks = (blocks + icap - 1) / icap;
970
971 dind_blks = (ind_blks + icap - 1) / icap;
972
973 tind_blks = 1;
974
975 return ind_blks + dind_blks + tind_blks;
976}
977
978/*
979 * Calculate the number of metadata blocks need to reserve
980 * to allocate given number of blocks
981 */
982static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
983{
Mingming Caocd213222008-08-19 22:16:59 -0400984 if (!blocks)
985 return 0;
986
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400987 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
988 return ext4_ext_calc_metadata_amount(inode, blocks);
989
990 return ext4_indirect_calc_metadata_amount(inode, blocks);
991}
992
993static void ext4_da_update_reserve_space(struct inode *inode, int used)
994{
995 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
996 int total, mdb, mdb_free;
997
998 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
999 /* recalculate the number of metablocks still need to be reserved */
1000 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1001 mdb = ext4_calc_metadata_amount(inode, total);
1002
1003 /* figure out how many metablocks to release */
1004 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1005 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1006
Aneesh Kumar K.V6bc6e632008-10-10 09:39:00 -04001007 if (mdb_free) {
1008 /* Account for allocated meta_blocks */
1009 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001010
Aneesh Kumar K.V6bc6e632008-10-10 09:39:00 -04001011 /* update fs dirty blocks counter */
1012 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1013 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1014 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1015 }
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001016
1017 /* update per-inode reservations */
1018 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1019 EXT4_I(inode)->i_reserved_data_blocks -= used;
1020
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001021 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1022}
1023
Mingming Caof5ab0d12008-02-25 15:29:55 -05001024/*
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001025 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1026 * and returns if the blocks are already mapped.
Mingming Caof5ab0d12008-02-25 15:29:55 -05001027 *
Mingming Caof5ab0d12008-02-25 15:29:55 -05001028 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1029 * and store the allocated blocks in the result buffer head and mark it
1030 * mapped.
1031 *
1032 * If file type is extents based, it will call ext4_ext_get_blocks(),
1033 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1034 * based files
1035 *
1036 * On success, it returns the number of blocks being mapped or allocate.
1037 * if create==0 and the blocks are pre-allocated and uninitialized block,
1038 * the result buffer head is unmapped. If the create ==1, it will make sure
1039 * the buffer head is mapped.
1040 *
1041 * It returns 0 if plain look up failed (blocks have not been allocated), in
1042 * that casem, buffer head is unmapped
1043 *
1044 * It returns the error in case of allocation failure.
1045 */
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001046int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1047 unsigned long max_blocks, struct buffer_head *bh,
Mingming Caod2a17632008-07-14 17:52:37 -04001048 int create, int extend_disksize, int flag)
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001049{
1050 int retval;
Mingming Caof5ab0d12008-02-25 15:29:55 -05001051
1052 clear_buffer_mapped(bh);
1053
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001054 /*
1055 * Try to see if we can get the block without requesting
1056 * for new file system block.
1057 */
1058 down_read((&EXT4_I(inode)->i_data_sem));
1059 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1060 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1061 bh, 0, 0);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001062 } else {
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001063 retval = ext4_get_blocks_handle(handle,
1064 inode, block, max_blocks, bh, 0, 0);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001065 }
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001066 up_read((&EXT4_I(inode)->i_data_sem));
Mingming Caof5ab0d12008-02-25 15:29:55 -05001067
1068 /* If it is only a block(s) look up */
1069 if (!create)
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001070 return retval;
1071
1072 /*
Mingming Caof5ab0d12008-02-25 15:29:55 -05001073 * Returns if the blocks have already allocated
1074 *
1075 * Note that if blocks have been preallocated
1076 * ext4_ext_get_block() returns th create = 0
1077 * with buffer head unmapped.
1078 */
1079 if (retval > 0 && buffer_mapped(bh))
1080 return retval;
1081
1082 /*
1083 * New blocks allocate and/or writing to uninitialized extent
1084 * will possibly result in updating i_data, so we take
1085 * the write lock of i_data_sem, and call get_blocks()
1086 * with create == 1 flag.
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001087 */
1088 down_write((&EXT4_I(inode)->i_data_sem));
Mingming Caod2a17632008-07-14 17:52:37 -04001089
1090 /*
1091 * if the caller is from delayed allocation writeout path
1092 * we have already reserved fs blocks for allocation
1093 * let the underlying get_block() function know to
1094 * avoid double accounting
1095 */
1096 if (flag)
1097 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001098 /*
1099 * We need to check for EXT4 here because migrate
1100 * could have changed the inode type in between
1101 */
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001102 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1103 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1104 bh, create, extend_disksize);
1105 } else {
1106 retval = ext4_get_blocks_handle(handle, inode, block,
1107 max_blocks, bh, create, extend_disksize);
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -04001108
1109 if (retval > 0 && buffer_new(bh)) {
1110 /*
1111 * We allocated new blocks which will result in
1112 * i_data's format changing. Force the migrate
1113 * to fail by clearing migrate flags
1114 */
1115 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1116 ~EXT4_EXT_MIGRATE;
1117 }
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001118 }
Mingming Caod2a17632008-07-14 17:52:37 -04001119
1120 if (flag) {
1121 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1122 /*
1123 * Update reserved blocks/metadata blocks
1124 * after successful block allocation
1125 * which were deferred till now
1126 */
1127 if ((retval > 0) && buffer_delay(bh))
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001128 ext4_da_update_reserve_space(inode, retval);
Mingming Caod2a17632008-07-14 17:52:37 -04001129 }
1130
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001131 up_write((&EXT4_I(inode)->i_data_sem));
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001132 return retval;
1133}
1134
Mingming Caof3bd1f32008-08-19 22:16:03 -04001135/* Maximum number of blocks we map for direct IO at once. */
1136#define DIO_MAX_BLOCKS 4096
1137
Eric Sandeen6873fa02008-10-07 00:46:36 -04001138int ext4_get_block(struct inode *inode, sector_t iblock,
1139 struct buffer_head *bh_result, int create)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001140{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -08001141 handle_t *handle = ext4_journal_current_handle();
Jan Kara7fb54092008-02-10 01:08:38 -05001142 int ret = 0, started = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001143 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
Mingming Caof3bd1f32008-08-19 22:16:03 -04001144 int dio_credits;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001145
Jan Kara7fb54092008-02-10 01:08:38 -05001146 if (create && !handle) {
1147 /* Direct IO write... */
1148 if (max_blocks > DIO_MAX_BLOCKS)
1149 max_blocks = DIO_MAX_BLOCKS;
Mingming Caof3bd1f32008-08-19 22:16:03 -04001150 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1151 handle = ext4_journal_start(inode, dio_credits);
Jan Kara7fb54092008-02-10 01:08:38 -05001152 if (IS_ERR(handle)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001153 ret = PTR_ERR(handle);
Jan Kara7fb54092008-02-10 01:08:38 -05001154 goto out;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001155 }
Jan Kara7fb54092008-02-10 01:08:38 -05001156 started = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001157 }
1158
Jan Kara7fb54092008-02-10 01:08:38 -05001159 ret = ext4_get_blocks_wrap(handle, inode, iblock,
Mingming Caod2a17632008-07-14 17:52:37 -04001160 max_blocks, bh_result, create, 0, 0);
Jan Kara7fb54092008-02-10 01:08:38 -05001161 if (ret > 0) {
1162 bh_result->b_size = (ret << inode->i_blkbits);
1163 ret = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001164 }
Jan Kara7fb54092008-02-10 01:08:38 -05001165 if (started)
1166 ext4_journal_stop(handle);
1167out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001168 return ret;
1169}
1170
1171/*
1172 * `handle' can be NULL if create is zero
1173 */
Mingming Cao617ba132006-10-11 01:20:53 -07001174struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001175 ext4_lblk_t block, int create, int *errp)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001176{
1177 struct buffer_head dummy;
1178 int fatal = 0, err;
1179
1180 J_ASSERT(handle != NULL || create == 0);
1181
1182 dummy.b_state = 0;
1183 dummy.b_blocknr = -1000;
1184 buffer_trace_init(&dummy.b_history);
Alex Tomasa86c6182006-10-11 01:21:03 -07001185 err = ext4_get_blocks_wrap(handle, inode, block, 1,
Mingming Caod2a17632008-07-14 17:52:37 -04001186 &dummy, create, 1, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001187 /*
Mingming Cao617ba132006-10-11 01:20:53 -07001188 * ext4_get_blocks_handle() returns number of blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001189 * mapped. 0 in case of a HOLE.
1190 */
1191 if (err > 0) {
1192 if (err > 1)
1193 WARN_ON(1);
1194 err = 0;
1195 }
1196 *errp = err;
1197 if (!err && buffer_mapped(&dummy)) {
1198 struct buffer_head *bh;
1199 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1200 if (!bh) {
1201 *errp = -EIO;
1202 goto err;
1203 }
1204 if (buffer_new(&dummy)) {
1205 J_ASSERT(create != 0);
Aneesh Kumar K.Vac398492007-10-16 18:38:25 -04001206 J_ASSERT(handle != NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001207
1208 /*
1209 * Now that we do not always journal data, we should
1210 * keep in mind whether this should always journal the
1211 * new buffer as metadata. For now, regular file
Mingming Cao617ba132006-10-11 01:20:53 -07001212 * writes use ext4_get_block instead, so it's not a
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001213 * problem.
1214 */
1215 lock_buffer(bh);
1216 BUFFER_TRACE(bh, "call get_create_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001217 fatal = ext4_journal_get_create_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001218 if (!fatal && !buffer_uptodate(bh)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001219 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001220 set_buffer_uptodate(bh);
1221 }
1222 unlock_buffer(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001223 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1224 err = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001225 if (!fatal)
1226 fatal = err;
1227 } else {
1228 BUFFER_TRACE(bh, "not a new buffer");
1229 }
1230 if (fatal) {
1231 *errp = fatal;
1232 brelse(bh);
1233 bh = NULL;
1234 }
1235 return bh;
1236 }
1237err:
1238 return NULL;
1239}
1240
Mingming Cao617ba132006-10-11 01:20:53 -07001241struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001242 ext4_lblk_t block, int create, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001243{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001244 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001245
Mingming Cao617ba132006-10-11 01:20:53 -07001246 bh = ext4_getblk(handle, inode, block, create, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001247 if (!bh)
1248 return bh;
1249 if (buffer_uptodate(bh))
1250 return bh;
1251 ll_rw_block(READ_META, 1, &bh);
1252 wait_on_buffer(bh);
1253 if (buffer_uptodate(bh))
1254 return bh;
1255 put_bh(bh);
1256 *err = -EIO;
1257 return NULL;
1258}
1259
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001260static int walk_page_buffers(handle_t *handle,
1261 struct buffer_head *head,
1262 unsigned from,
1263 unsigned to,
1264 int *partial,
1265 int (*fn)(handle_t *handle,
1266 struct buffer_head *bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001267{
1268 struct buffer_head *bh;
1269 unsigned block_start, block_end;
1270 unsigned blocksize = head->b_size;
1271 int err, ret = 0;
1272 struct buffer_head *next;
1273
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001274 for (bh = head, block_start = 0;
1275 ret == 0 && (bh != head || !block_start);
1276 block_start = block_end, bh = next)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001277 {
1278 next = bh->b_this_page;
1279 block_end = block_start + blocksize;
1280 if (block_end <= from || block_start >= to) {
1281 if (partial && !buffer_uptodate(bh))
1282 *partial = 1;
1283 continue;
1284 }
1285 err = (*fn)(handle, bh);
1286 if (!ret)
1287 ret = err;
1288 }
1289 return ret;
1290}
1291
1292/*
1293 * To preserve ordering, it is essential that the hole instantiation and
1294 * the data write be encapsulated in a single transaction. We cannot
Mingming Cao617ba132006-10-11 01:20:53 -07001295 * close off a transaction and start a new one between the ext4_get_block()
Mingming Caodab291a2006-10-11 01:21:01 -07001296 * and the commit_write(). So doing the jbd2_journal_start at the start of
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001297 * prepare_write() is the right place.
1298 *
Mingming Cao617ba132006-10-11 01:20:53 -07001299 * Also, this function can nest inside ext4_writepage() ->
1300 * block_write_full_page(). In that case, we *know* that ext4_writepage()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001301 * has generated enough buffer credits to do the whole page. So we won't
1302 * block on the journal in that case, which is good, because the caller may
1303 * be PF_MEMALLOC.
1304 *
Mingming Cao617ba132006-10-11 01:20:53 -07001305 * By accident, ext4 can be reentered when a transaction is open via
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001306 * quota file writes. If we were to commit the transaction while thus
1307 * reentered, there can be a deadlock - we would be holding a quota
1308 * lock, and the commit would never complete if another thread had a
1309 * transaction open and was blocking on the quota lock - a ranking
1310 * violation.
1311 *
Mingming Caodab291a2006-10-11 01:21:01 -07001312 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001313 * will _not_ run commit under these circumstances because handle->h_ref
1314 * is elevated. We'll still have enough credits for the tiny quotafile
1315 * write.
1316 */
1317static int do_journal_get_write_access(handle_t *handle,
1318 struct buffer_head *bh)
1319{
1320 if (!buffer_mapped(bh) || buffer_freed(bh))
1321 return 0;
Mingming Cao617ba132006-10-11 01:20:53 -07001322 return ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001323}
1324
Nick Pigginbfc1af62007-10-16 01:25:05 -07001325static int ext4_write_begin(struct file *file, struct address_space *mapping,
1326 loff_t pos, unsigned len, unsigned flags,
1327 struct page **pagep, void **fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001328{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001329 struct inode *inode = mapping->host;
Andrew Morton7479d2b2007-04-01 23:49:44 -07001330 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001331 handle_t *handle;
1332 int retries = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001333 struct page *page;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001334 pgoff_t index;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001335 unsigned from, to;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001336
1337 index = pos >> PAGE_CACHE_SHIFT;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001338 from = pos & (PAGE_CACHE_SIZE - 1);
1339 to = from + len;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001340
1341retry:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001342 handle = ext4_journal_start(inode, needed_blocks);
1343 if (IS_ERR(handle)) {
1344 ret = PTR_ERR(handle);
1345 goto out;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001346 }
1347
Jan Karacf108bc2008-07-11 19:27:31 -04001348 page = __grab_cache_page(mapping, index);
1349 if (!page) {
1350 ext4_journal_stop(handle);
1351 ret = -ENOMEM;
1352 goto out;
1353 }
1354 *pagep = page;
1355
Nick Pigginbfc1af62007-10-16 01:25:05 -07001356 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1357 ext4_get_block);
1358
1359 if (!ret && ext4_should_journal_data(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001360 ret = walk_page_buffers(handle, page_buffers(page),
1361 from, to, NULL, do_journal_get_write_access);
1362 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001363
1364 if (ret) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001365 unlock_page(page);
Jan Karacf108bc2008-07-11 19:27:31 -04001366 ext4_journal_stop(handle);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001367 page_cache_release(page);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04001368 /*
1369 * block_write_begin may have instantiated a few blocks
1370 * outside i_size. Trim these off again. Don't need
1371 * i_size_read because we hold i_mutex.
1372 */
1373 if (pos + len > inode->i_size)
1374 vmtruncate(inode, inode->i_size);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001375 }
1376
Mingming Cao617ba132006-10-11 01:20:53 -07001377 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001378 goto retry;
Andrew Morton7479d2b2007-04-01 23:49:44 -07001379out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001380 return ret;
1381}
1382
Nick Pigginbfc1af62007-10-16 01:25:05 -07001383/* For write_end() in data=journal mode */
1384static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001385{
1386 if (!buffer_mapped(bh) || buffer_freed(bh))
1387 return 0;
1388 set_buffer_uptodate(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001389 return ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001390}
1391
1392/*
1393 * We need to pick up the new inode size which generic_commit_write gave us
1394 * `file' can be NULL - eg, when called from page_symlink().
1395 *
Mingming Cao617ba132006-10-11 01:20:53 -07001396 * ext4 never places buffers on inode->i_mapping->private_list. metadata
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001397 * buffers are managed internally.
1398 */
Nick Pigginbfc1af62007-10-16 01:25:05 -07001399static int ext4_ordered_write_end(struct file *file,
1400 struct address_space *mapping,
1401 loff_t pos, unsigned len, unsigned copied,
1402 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001403{
Mingming Cao617ba132006-10-11 01:20:53 -07001404 handle_t *handle = ext4_journal_current_handle();
Jan Karacf108bc2008-07-11 19:27:31 -04001405 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001406 int ret = 0, ret2;
1407
Jan Kara678aaf42008-07-11 19:27:31 -04001408 ret = ext4_jbd2_file_inode(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001409
1410 if (ret == 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001411 loff_t new_i_size;
1412
Nick Pigginbfc1af62007-10-16 01:25:05 -07001413 new_i_size = pos + copied;
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001414 if (new_i_size > EXT4_I(inode)->i_disksize) {
1415 ext4_update_i_disksize(inode, new_i_size);
1416 /* We need to mark inode dirty even if
1417 * new_i_size is less that inode->i_size
1418 * bu greater than i_disksize.(hint delalloc)
1419 */
1420 ext4_mark_inode_dirty(handle, inode);
1421 }
1422
Jan Karacf108bc2008-07-11 19:27:31 -04001423 ret2 = generic_write_end(file, mapping, pos, len, copied,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001424 page, fsdata);
Roel Kluinf8a87d82008-04-29 22:01:18 -04001425 copied = ret2;
1426 if (ret2 < 0)
1427 ret = ret2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001428 }
Mingming Cao617ba132006-10-11 01:20:53 -07001429 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001430 if (!ret)
1431 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001432
1433 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001434}
1435
Nick Pigginbfc1af62007-10-16 01:25:05 -07001436static int ext4_writeback_write_end(struct file *file,
1437 struct address_space *mapping,
1438 loff_t pos, unsigned len, unsigned copied,
1439 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001440{
Mingming Cao617ba132006-10-11 01:20:53 -07001441 handle_t *handle = ext4_journal_current_handle();
Jan Karacf108bc2008-07-11 19:27:31 -04001442 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001443 int ret = 0, ret2;
1444 loff_t new_i_size;
1445
Nick Pigginbfc1af62007-10-16 01:25:05 -07001446 new_i_size = pos + copied;
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001447 if (new_i_size > EXT4_I(inode)->i_disksize) {
1448 ext4_update_i_disksize(inode, new_i_size);
1449 /* We need to mark inode dirty even if
1450 * new_i_size is less that inode->i_size
1451 * bu greater than i_disksize.(hint delalloc)
1452 */
1453 ext4_mark_inode_dirty(handle, inode);
1454 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001455
Jan Karacf108bc2008-07-11 19:27:31 -04001456 ret2 = generic_write_end(file, mapping, pos, len, copied,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001457 page, fsdata);
Roel Kluinf8a87d82008-04-29 22:01:18 -04001458 copied = ret2;
1459 if (ret2 < 0)
1460 ret = ret2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001461
Mingming Cao617ba132006-10-11 01:20:53 -07001462 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001463 if (!ret)
1464 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001465
1466 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001467}
1468
Nick Pigginbfc1af62007-10-16 01:25:05 -07001469static int ext4_journalled_write_end(struct file *file,
1470 struct address_space *mapping,
1471 loff_t pos, unsigned len, unsigned copied,
1472 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001473{
Mingming Cao617ba132006-10-11 01:20:53 -07001474 handle_t *handle = ext4_journal_current_handle();
Nick Pigginbfc1af62007-10-16 01:25:05 -07001475 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001476 int ret = 0, ret2;
1477 int partial = 0;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001478 unsigned from, to;
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001479 loff_t new_i_size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001480
Nick Pigginbfc1af62007-10-16 01:25:05 -07001481 from = pos & (PAGE_CACHE_SIZE - 1);
1482 to = from + len;
1483
1484 if (copied < len) {
1485 if (!PageUptodate(page))
1486 copied = 0;
1487 page_zero_new_buffers(page, from+copied, to);
1488 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001489
1490 ret = walk_page_buffers(handle, page_buffers(page), from,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001491 to, &partial, write_end_fn);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001492 if (!partial)
1493 SetPageUptodate(page);
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001494 new_i_size = pos + copied;
1495 if (new_i_size > inode->i_size)
Nick Pigginbfc1af62007-10-16 01:25:05 -07001496 i_size_write(inode, pos+copied);
Mingming Cao617ba132006-10-11 01:20:53 -07001497 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001498 if (new_i_size > EXT4_I(inode)->i_disksize) {
1499 ext4_update_i_disksize(inode, new_i_size);
Mingming Cao617ba132006-10-11 01:20:53 -07001500 ret2 = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001501 if (!ret)
1502 ret = ret2;
1503 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001504
Jan Karacf108bc2008-07-11 19:27:31 -04001505 unlock_page(page);
Mingming Cao617ba132006-10-11 01:20:53 -07001506 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001507 if (!ret)
1508 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001509 page_cache_release(page);
1510
1511 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001512}
Mingming Caod2a17632008-07-14 17:52:37 -04001513
1514static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1515{
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001516 int retries = 0;
Mingming Caod2a17632008-07-14 17:52:37 -04001517 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1518 unsigned long md_needed, mdblocks, total = 0;
1519
1520 /*
1521 * recalculate the amount of metadata blocks to reserve
1522 * in order to allocate nrblocks
1523 * worse case is one extent per block
1524 */
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001525repeat:
Mingming Caod2a17632008-07-14 17:52:37 -04001526 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1527 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1528 mdblocks = ext4_calc_metadata_amount(inode, total);
1529 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1530
1531 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1532 total = md_needed + nrblocks;
1533
Aneesh Kumar K.Va30d542a2008-10-09 10:56:23 -04001534 if (ext4_claim_free_blocks(sbi, total)) {
Mingming Caod2a17632008-07-14 17:52:37 -04001535 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001536 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1537 yield();
1538 goto repeat;
1539 }
Mingming Caod2a17632008-07-14 17:52:37 -04001540 return -ENOSPC;
1541 }
Mingming Caod2a17632008-07-14 17:52:37 -04001542 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1543 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1544
1545 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1546 return 0; /* success */
1547}
1548
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001549static void ext4_da_release_space(struct inode *inode, int to_free)
Mingming Caod2a17632008-07-14 17:52:37 -04001550{
1551 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1552 int total, mdb, mdb_free, release;
1553
Mingming Caocd213222008-08-19 22:16:59 -04001554 if (!to_free)
1555 return; /* Nothing to release, exit */
1556
Mingming Caod2a17632008-07-14 17:52:37 -04001557 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Caocd213222008-08-19 22:16:59 -04001558
1559 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1560 /*
1561 * if there is no reserved blocks, but we try to free some
1562 * then the counter is messed up somewhere.
1563 * but since this function is called from invalidate
1564 * page, it's harmless to return without any action
1565 */
1566 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1567 "blocks for inode %lu, but there is no reserved "
1568 "data blocks\n", to_free, inode->i_ino);
1569 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1570 return;
1571 }
1572
Mingming Caod2a17632008-07-14 17:52:37 -04001573 /* recalculate the number of metablocks still need to be reserved */
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001574 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
Mingming Caod2a17632008-07-14 17:52:37 -04001575 mdb = ext4_calc_metadata_amount(inode, total);
1576
1577 /* figure out how many metablocks to release */
1578 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1579 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1580
Mingming Caod2a17632008-07-14 17:52:37 -04001581 release = to_free + mdb_free;
1582
Aneesh Kumar K.V6bc6e632008-10-10 09:39:00 -04001583 /* update fs dirty blocks counter for truncate case */
1584 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
Mingming Caod2a17632008-07-14 17:52:37 -04001585
1586 /* update per-inode reservations */
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001587 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1588 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
Mingming Caod2a17632008-07-14 17:52:37 -04001589
1590 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1591 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
Mingming Caod2a17632008-07-14 17:52:37 -04001592 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1593}
1594
1595static void ext4_da_page_release_reservation(struct page *page,
1596 unsigned long offset)
1597{
1598 int to_release = 0;
1599 struct buffer_head *head, *bh;
1600 unsigned int curr_off = 0;
1601
1602 head = page_buffers(page);
1603 bh = head;
1604 do {
1605 unsigned int next_off = curr_off + bh->b_size;
1606
1607 if ((offset <= curr_off) && (buffer_delay(bh))) {
1608 to_release++;
1609 clear_buffer_delay(bh);
1610 }
1611 curr_off = next_off;
1612 } while ((bh = bh->b_this_page) != head);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001613 ext4_da_release_space(page->mapping->host, to_release);
Mingming Caod2a17632008-07-14 17:52:37 -04001614}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001615
1616/*
Alex Tomas64769242008-07-11 19:27:31 -04001617 * Delayed allocation stuff
1618 */
1619
1620struct mpage_da_data {
1621 struct inode *inode;
1622 struct buffer_head lbh; /* extent of blocks */
1623 unsigned long first_page, next_page; /* extent of pages */
1624 get_block_t *get_block;
1625 struct writeback_control *wbc;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001626 int io_done;
1627 long pages_written;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001628 int retval;
Alex Tomas64769242008-07-11 19:27:31 -04001629};
1630
1631/*
1632 * mpage_da_submit_io - walks through extent of pages and try to write
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001633 * them with writepage() call back
Alex Tomas64769242008-07-11 19:27:31 -04001634 *
1635 * @mpd->inode: inode
1636 * @mpd->first_page: first page of the extent
1637 * @mpd->next_page: page after the last page of the extent
1638 * @mpd->get_block: the filesystem's block mapper function
1639 *
1640 * By the time mpage_da_submit_io() is called we expect all blocks
1641 * to be allocated. this may be wrong if allocation failed.
1642 *
1643 * As pages are already locked by write_cache_pages(), we can't use it
1644 */
1645static int mpage_da_submit_io(struct mpage_da_data *mpd)
1646{
1647 struct address_space *mapping = mpd->inode->i_mapping;
Alex Tomas64769242008-07-11 19:27:31 -04001648 int ret = 0, err, nr_pages, i;
1649 unsigned long index, end;
1650 struct pagevec pvec;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04001651 long pages_skipped;
Alex Tomas64769242008-07-11 19:27:31 -04001652
1653 BUG_ON(mpd->next_page <= mpd->first_page);
Alex Tomas64769242008-07-11 19:27:31 -04001654 pagevec_init(&pvec, 0);
1655 index = mpd->first_page;
1656 end = mpd->next_page - 1;
1657
1658 while (index <= end) {
Aneesh Kumar K.Vaf6f0292008-10-14 09:20:19 -04001659 /*
1660 * We can use PAGECACHE_TAG_DIRTY lookup here because
1661 * even though we have cleared the dirty flag on the page
1662 * We still keep the page in the radix tree with tag
1663 * PAGECACHE_TAG_DIRTY. See clear_page_dirty_for_io.
1664 * The PAGECACHE_TAG_DIRTY is cleared in set_page_writeback
1665 * which is called via the below writepage callback.
1666 */
1667 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1668 PAGECACHE_TAG_DIRTY,
1669 min(end - index,
1670 (pgoff_t)PAGEVEC_SIZE-1) + 1);
Alex Tomas64769242008-07-11 19:27:31 -04001671 if (nr_pages == 0)
1672 break;
1673 for (i = 0; i < nr_pages; i++) {
1674 struct page *page = pvec.pages[i];
1675
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04001676 pages_skipped = mpd->wbc->pages_skipped;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001677 err = mapping->a_ops->writepage(page, mpd->wbc);
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04001678 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
1679 /*
1680 * have successfully written the page
1681 * without skipping the same
1682 */
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001683 mpd->pages_written++;
Alex Tomas64769242008-07-11 19:27:31 -04001684 /*
1685 * In error case, we have to continue because
1686 * remaining pages are still locked
1687 * XXX: unlock and re-dirty them?
1688 */
1689 if (ret == 0)
1690 ret = err;
1691 }
1692 pagevec_release(&pvec);
1693 }
Alex Tomas64769242008-07-11 19:27:31 -04001694 return ret;
1695}
1696
1697/*
1698 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1699 *
1700 * @mpd->inode - inode to walk through
1701 * @exbh->b_blocknr - first block on a disk
1702 * @exbh->b_size - amount of space in bytes
1703 * @logical - first logical block to start assignment with
1704 *
1705 * the function goes through all passed space and put actual disk
1706 * block numbers into buffer heads, dropping BH_Delay
1707 */
1708static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1709 struct buffer_head *exbh)
1710{
1711 struct inode *inode = mpd->inode;
1712 struct address_space *mapping = inode->i_mapping;
1713 int blocks = exbh->b_size >> inode->i_blkbits;
1714 sector_t pblock = exbh->b_blocknr, cur_logical;
1715 struct buffer_head *head, *bh;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001716 pgoff_t index, end;
Alex Tomas64769242008-07-11 19:27:31 -04001717 struct pagevec pvec;
1718 int nr_pages, i;
1719
1720 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1721 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1722 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1723
1724 pagevec_init(&pvec, 0);
1725
1726 while (index <= end) {
1727 /* XXX: optimize tail */
1728 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1729 if (nr_pages == 0)
1730 break;
1731 for (i = 0; i < nr_pages; i++) {
1732 struct page *page = pvec.pages[i];
1733
1734 index = page->index;
1735 if (index > end)
1736 break;
1737 index++;
1738
1739 BUG_ON(!PageLocked(page));
1740 BUG_ON(PageWriteback(page));
1741 BUG_ON(!page_has_buffers(page));
1742
1743 bh = page_buffers(page);
1744 head = bh;
1745
1746 /* skip blocks out of the range */
1747 do {
1748 if (cur_logical >= logical)
1749 break;
1750 cur_logical++;
1751 } while ((bh = bh->b_this_page) != head);
1752
1753 do {
1754 if (cur_logical >= logical + blocks)
1755 break;
Alex Tomas64769242008-07-11 19:27:31 -04001756 if (buffer_delay(bh)) {
1757 bh->b_blocknr = pblock;
1758 clear_buffer_delay(bh);
Aneesh Kumar K.Vbf068ee2008-08-19 22:16:43 -04001759 bh->b_bdev = inode->i_sb->s_bdev;
1760 } else if (buffer_unwritten(bh)) {
1761 bh->b_blocknr = pblock;
1762 clear_buffer_unwritten(bh);
1763 set_buffer_mapped(bh);
1764 set_buffer_new(bh);
1765 bh->b_bdev = inode->i_sb->s_bdev;
Mingming Cao61628a32008-07-11 19:27:31 -04001766 } else if (buffer_mapped(bh))
Alex Tomas64769242008-07-11 19:27:31 -04001767 BUG_ON(bh->b_blocknr != pblock);
Alex Tomas64769242008-07-11 19:27:31 -04001768
1769 cur_logical++;
1770 pblock++;
1771 } while ((bh = bh->b_this_page) != head);
1772 }
1773 pagevec_release(&pvec);
1774 }
1775}
1776
1777
1778/*
1779 * __unmap_underlying_blocks - just a helper function to unmap
1780 * set of blocks described by @bh
1781 */
1782static inline void __unmap_underlying_blocks(struct inode *inode,
1783 struct buffer_head *bh)
1784{
1785 struct block_device *bdev = inode->i_sb->s_bdev;
1786 int blocks, i;
1787
1788 blocks = bh->b_size >> inode->i_blkbits;
1789 for (i = 0; i < blocks; i++)
1790 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1791}
1792
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001793static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
1794 sector_t logical, long blk_cnt)
1795{
1796 int nr_pages, i;
1797 pgoff_t index, end;
1798 struct pagevec pvec;
1799 struct inode *inode = mpd->inode;
1800 struct address_space *mapping = inode->i_mapping;
1801
1802 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1803 end = (logical + blk_cnt - 1) >>
1804 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1805 while (index <= end) {
1806 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1807 if (nr_pages == 0)
1808 break;
1809 for (i = 0; i < nr_pages; i++) {
1810 struct page *page = pvec.pages[i];
1811 index = page->index;
1812 if (index > end)
1813 break;
1814 index++;
1815
1816 BUG_ON(!PageLocked(page));
1817 BUG_ON(PageWriteback(page));
1818 block_invalidatepage(page, 0);
1819 ClearPageUptodate(page);
1820 unlock_page(page);
1821 }
1822 }
1823 return;
1824}
1825
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001826static void ext4_print_free_blocks(struct inode *inode)
1827{
1828 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1829 printk(KERN_EMERG "Total free blocks count %lld\n",
1830 ext4_count_free_blocks(inode->i_sb));
1831 printk(KERN_EMERG "Free/Dirty block details\n");
1832 printk(KERN_EMERG "free_blocks=%lld\n",
1833 percpu_counter_sum(&sbi->s_freeblocks_counter));
1834 printk(KERN_EMERG "dirty_blocks=%lld\n",
1835 percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1836 printk(KERN_EMERG "Block reservation details\n");
1837 printk(KERN_EMERG "i_reserved_data_blocks=%lu\n",
1838 EXT4_I(inode)->i_reserved_data_blocks);
1839 printk(KERN_EMERG "i_reserved_meta_blocks=%lu\n",
1840 EXT4_I(inode)->i_reserved_meta_blocks);
1841 return;
1842}
1843
Alex Tomas64769242008-07-11 19:27:31 -04001844/*
1845 * mpage_da_map_blocks - go through given space
1846 *
1847 * @mpd->lbh - bh describing space
1848 * @mpd->get_block - the filesystem's block mapper function
1849 *
1850 * The function skips space we know is already mapped to disk blocks.
1851 *
Alex Tomas64769242008-07-11 19:27:31 -04001852 */
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001853static int mpage_da_map_blocks(struct mpage_da_data *mpd)
Alex Tomas64769242008-07-11 19:27:31 -04001854{
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001855 int err = 0;
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001856 struct buffer_head new;
Alex Tomas64769242008-07-11 19:27:31 -04001857 struct buffer_head *lbh = &mpd->lbh;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001858 sector_t next;
Alex Tomas64769242008-07-11 19:27:31 -04001859
1860 /*
1861 * We consider only non-mapped and non-allocated blocks
1862 */
1863 if (buffer_mapped(lbh) && !buffer_delay(lbh))
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001864 return 0;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001865 new.b_state = lbh->b_state;
1866 new.b_blocknr = 0;
1867 new.b_size = lbh->b_size;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001868 next = lbh->b_blocknr;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001869 /*
1870 * If we didn't accumulate anything
1871 * to write simply return
1872 */
1873 if (!new.b_size)
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001874 return 0;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001875 err = mpd->get_block(mpd->inode, next, &new, 1);
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001876 if (err) {
1877
1878 /* If get block returns with error
1879 * we simply return. Later writepage
1880 * will redirty the page and writepages
1881 * will find the dirty page again
1882 */
1883 if (err == -EAGAIN)
1884 return 0;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001885
1886 if (err == -ENOSPC &&
1887 ext4_count_free_blocks(mpd->inode->i_sb)) {
1888 mpd->retval = err;
1889 return 0;
1890 }
1891
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001892 /*
1893 * get block failure will cause us
1894 * to loop in writepages. Because
1895 * a_ops->writepage won't be able to
1896 * make progress. The page will be redirtied
1897 * by writepage and writepages will again
1898 * try to write the same.
1899 */
1900 printk(KERN_EMERG "%s block allocation failed for inode %lu "
1901 "at logical offset %llu with max blocks "
1902 "%zd with error %d\n",
1903 __func__, mpd->inode->i_ino,
1904 (unsigned long long)next,
1905 lbh->b_size >> mpd->inode->i_blkbits, err);
1906 printk(KERN_EMERG "This should not happen.!! "
1907 "Data will be lost\n");
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001908 if (err == -ENOSPC) {
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001909 ext4_print_free_blocks(mpd->inode);
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001910 }
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001911 /* invlaidate all the pages */
1912 ext4_da_block_invalidatepages(mpd, next,
1913 lbh->b_size >> mpd->inode->i_blkbits);
1914 return err;
1915 }
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001916 BUG_ON(new.b_size == 0);
Alex Tomas64769242008-07-11 19:27:31 -04001917
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001918 if (buffer_new(&new))
1919 __unmap_underlying_blocks(mpd->inode, &new);
Alex Tomas64769242008-07-11 19:27:31 -04001920
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001921 /*
1922 * If blocks are delayed marked, we need to
1923 * put actual blocknr and drop delayed bit
1924 */
1925 if (buffer_delay(lbh) || buffer_unwritten(lbh))
1926 mpage_put_bnr_to_bhs(mpd, next, &new);
1927
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001928 return 0;
Alex Tomas64769242008-07-11 19:27:31 -04001929}
1930
Aneesh Kumar K.Vbf068ee2008-08-19 22:16:43 -04001931#define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1932 (1 << BH_Delay) | (1 << BH_Unwritten))
Alex Tomas64769242008-07-11 19:27:31 -04001933
1934/*
1935 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1936 *
1937 * @mpd->lbh - extent of blocks
1938 * @logical - logical number of the block in the file
1939 * @bh - bh of the block (used to access block's state)
1940 *
1941 * the function is used to collect contig. blocks in same state
1942 */
1943static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1944 sector_t logical, struct buffer_head *bh)
1945{
Alex Tomas64769242008-07-11 19:27:31 -04001946 sector_t next;
Mingming Cao525f4ed2008-08-19 22:15:58 -04001947 size_t b_size = bh->b_size;
1948 struct buffer_head *lbh = &mpd->lbh;
1949 int nrblocks = lbh->b_size >> mpd->inode->i_blkbits;
Alex Tomas64769242008-07-11 19:27:31 -04001950
Mingming Cao525f4ed2008-08-19 22:15:58 -04001951 /* check if thereserved journal credits might overflow */
1952 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
1953 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1954 /*
1955 * With non-extent format we are limited by the journal
1956 * credit available. Total credit needed to insert
1957 * nrblocks contiguous blocks is dependent on the
1958 * nrblocks. So limit nrblocks.
1959 */
1960 goto flush_it;
1961 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1962 EXT4_MAX_TRANS_DATA) {
1963 /*
1964 * Adding the new buffer_head would make it cross the
1965 * allowed limit for which we have journal credit
1966 * reserved. So limit the new bh->b_size
1967 */
1968 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1969 mpd->inode->i_blkbits;
1970 /* we will do mpage_da_submit_io in the next loop */
1971 }
1972 }
Alex Tomas64769242008-07-11 19:27:31 -04001973 /*
1974 * First block in the extent
1975 */
1976 if (lbh->b_size == 0) {
1977 lbh->b_blocknr = logical;
Mingming Cao525f4ed2008-08-19 22:15:58 -04001978 lbh->b_size = b_size;
Alex Tomas64769242008-07-11 19:27:31 -04001979 lbh->b_state = bh->b_state & BH_FLAGS;
1980 return;
1981 }
1982
Mingming Cao525f4ed2008-08-19 22:15:58 -04001983 next = lbh->b_blocknr + nrblocks;
Alex Tomas64769242008-07-11 19:27:31 -04001984 /*
1985 * Can we merge the block to our big extent?
1986 */
1987 if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) {
Mingming Cao525f4ed2008-08-19 22:15:58 -04001988 lbh->b_size += b_size;
Alex Tomas64769242008-07-11 19:27:31 -04001989 return;
1990 }
1991
Mingming Cao525f4ed2008-08-19 22:15:58 -04001992flush_it:
Alex Tomas64769242008-07-11 19:27:31 -04001993 /*
1994 * We couldn't merge the block to our extent, so we
1995 * need to flush current extent and start new one
1996 */
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001997 if (mpage_da_map_blocks(mpd) == 0)
1998 mpage_da_submit_io(mpd);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001999 mpd->io_done = 1;
2000 return;
Alex Tomas64769242008-07-11 19:27:31 -04002001}
2002
2003/*
2004 * __mpage_da_writepage - finds extent of pages and blocks
2005 *
2006 * @page: page to consider
2007 * @wbc: not used, we just follow rules
2008 * @data: context
2009 *
2010 * The function finds extents of pages and scan them for all blocks.
2011 */
2012static int __mpage_da_writepage(struct page *page,
2013 struct writeback_control *wbc, void *data)
2014{
2015 struct mpage_da_data *mpd = data;
2016 struct inode *inode = mpd->inode;
2017 struct buffer_head *bh, *head, fake;
2018 sector_t logical;
2019
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002020 if (mpd->io_done) {
2021 /*
2022 * Rest of the page in the page_vec
2023 * redirty then and skip then. We will
2024 * try to to write them again after
2025 * starting a new transaction
2026 */
2027 redirty_page_for_writepage(wbc, page);
2028 unlock_page(page);
2029 return MPAGE_DA_EXTENT_TAIL;
2030 }
Alex Tomas64769242008-07-11 19:27:31 -04002031 /*
2032 * Can we merge this page to current extent?
2033 */
2034 if (mpd->next_page != page->index) {
2035 /*
2036 * Nope, we can't. So, we map non-allocated blocks
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002037 * and start IO on them using writepage()
Alex Tomas64769242008-07-11 19:27:31 -04002038 */
2039 if (mpd->next_page != mpd->first_page) {
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002040 if (mpage_da_map_blocks(mpd) == 0)
2041 mpage_da_submit_io(mpd);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002042 /*
2043 * skip rest of the page in the page_vec
2044 */
2045 mpd->io_done = 1;
2046 redirty_page_for_writepage(wbc, page);
2047 unlock_page(page);
2048 return MPAGE_DA_EXTENT_TAIL;
Alex Tomas64769242008-07-11 19:27:31 -04002049 }
2050
2051 /*
2052 * Start next extent of pages ...
2053 */
2054 mpd->first_page = page->index;
2055
2056 /*
2057 * ... and blocks
2058 */
2059 mpd->lbh.b_size = 0;
2060 mpd->lbh.b_state = 0;
2061 mpd->lbh.b_blocknr = 0;
2062 }
2063
2064 mpd->next_page = page->index + 1;
2065 logical = (sector_t) page->index <<
2066 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2067
2068 if (!page_has_buffers(page)) {
2069 /*
2070 * There is no attached buffer heads yet (mmap?)
2071 * we treat the page asfull of dirty blocks
2072 */
2073 bh = &fake;
2074 bh->b_size = PAGE_CACHE_SIZE;
2075 bh->b_state = 0;
2076 set_buffer_dirty(bh);
2077 set_buffer_uptodate(bh);
2078 mpage_add_bh_to_extent(mpd, logical, bh);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002079 if (mpd->io_done)
2080 return MPAGE_DA_EXTENT_TAIL;
Alex Tomas64769242008-07-11 19:27:31 -04002081 } else {
2082 /*
2083 * Page with regular buffer heads, just add all dirty ones
2084 */
2085 head = page_buffers(page);
2086 bh = head;
2087 do {
2088 BUG_ON(buffer_locked(bh));
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002089 if (buffer_dirty(bh) &&
2090 (!buffer_mapped(bh) || buffer_delay(bh))) {
Alex Tomas64769242008-07-11 19:27:31 -04002091 mpage_add_bh_to_extent(mpd, logical, bh);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002092 if (mpd->io_done)
2093 return MPAGE_DA_EXTENT_TAIL;
2094 }
Alex Tomas64769242008-07-11 19:27:31 -04002095 logical++;
2096 } while ((bh = bh->b_this_page) != head);
2097 }
2098
2099 return 0;
2100}
2101
2102/*
2103 * mpage_da_writepages - walk the list of dirty pages of the given
2104 * address space, allocates non-allocated blocks, maps newly-allocated
2105 * blocks to existing bhs and issue IO them
2106 *
2107 * @mapping: address space structure to write
2108 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2109 * @get_block: the filesystem's block mapper function.
2110 *
2111 * This is a library function, which implements the writepages()
2112 * address_space_operation.
Alex Tomas64769242008-07-11 19:27:31 -04002113 */
2114static int mpage_da_writepages(struct address_space *mapping,
2115 struct writeback_control *wbc,
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002116 struct mpage_da_data *mpd)
Alex Tomas64769242008-07-11 19:27:31 -04002117{
Alex Tomas64769242008-07-11 19:27:31 -04002118 int ret;
2119
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002120 if (!mpd->get_block)
Alex Tomas64769242008-07-11 19:27:31 -04002121 return generic_writepages(mapping, wbc);
2122
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002123 mpd->lbh.b_size = 0;
2124 mpd->lbh.b_state = 0;
2125 mpd->lbh.b_blocknr = 0;
2126 mpd->first_page = 0;
2127 mpd->next_page = 0;
2128 mpd->io_done = 0;
2129 mpd->pages_written = 0;
2130 mpd->retval = 0;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002131
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002132 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, mpd);
Alex Tomas64769242008-07-11 19:27:31 -04002133 /*
2134 * Handle last extent of pages
2135 */
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002136 if (!mpd->io_done && mpd->next_page != mpd->first_page) {
2137 if (mpage_da_map_blocks(mpd) == 0)
2138 mpage_da_submit_io(mpd);
Alex Tomas64769242008-07-11 19:27:31 -04002139
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002140 mpd->io_done = 1;
2141 ret = MPAGE_DA_EXTENT_TAIL;
2142 }
2143 wbc->nr_to_write -= mpd->pages_written;
Alex Tomas64769242008-07-11 19:27:31 -04002144 return ret;
2145}
2146
2147/*
2148 * this is a special callback for ->write_begin() only
2149 * it's intention is to return mapped block or reserve space
2150 */
2151static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2152 struct buffer_head *bh_result, int create)
2153{
2154 int ret = 0;
2155
2156 BUG_ON(create == 0);
2157 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2158
2159 /*
2160 * first, we need to know whether the block is allocated already
2161 * preallocated blocks are unmapped but should treated
2162 * the same as allocated blocks.
2163 */
Mingming Caod2a17632008-07-14 17:52:37 -04002164 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
2165 if ((ret == 0) && !buffer_delay(bh_result)) {
2166 /* the block isn't (pre)allocated yet, let's reserve space */
Alex Tomas64769242008-07-11 19:27:31 -04002167 /*
2168 * XXX: __block_prepare_write() unmaps passed block,
2169 * is it OK?
2170 */
Mingming Caod2a17632008-07-14 17:52:37 -04002171 ret = ext4_da_reserve_space(inode, 1);
2172 if (ret)
2173 /* not enough space to reserve */
2174 return ret;
2175
Alex Tomas64769242008-07-11 19:27:31 -04002176 map_bh(bh_result, inode->i_sb, 0);
2177 set_buffer_new(bh_result);
2178 set_buffer_delay(bh_result);
2179 } else if (ret > 0) {
2180 bh_result->b_size = (ret << inode->i_blkbits);
2181 ret = 0;
2182 }
2183
2184 return ret;
2185}
Mingming Caod2a17632008-07-14 17:52:37 -04002186#define EXT4_DELALLOC_RSVED 1
Alex Tomas64769242008-07-11 19:27:31 -04002187static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
2188 struct buffer_head *bh_result, int create)
2189{
Mingming Cao61628a32008-07-11 19:27:31 -04002190 int ret;
Alex Tomas64769242008-07-11 19:27:31 -04002191 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2192 loff_t disksize = EXT4_I(inode)->i_disksize;
2193 handle_t *handle = NULL;
2194
Mingming Cao61628a32008-07-11 19:27:31 -04002195 handle = ext4_journal_current_handle();
Aneesh Kumar K.V166348d2008-09-08 23:08:40 -04002196 BUG_ON(!handle);
2197 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2198 bh_result, create, 0, EXT4_DELALLOC_RSVED);
Alex Tomas64769242008-07-11 19:27:31 -04002199 if (ret > 0) {
Aneesh Kumar K.V166348d2008-09-08 23:08:40 -04002200
Alex Tomas64769242008-07-11 19:27:31 -04002201 bh_result->b_size = (ret << inode->i_blkbits);
2202
Aneesh Kumar K.V166348d2008-09-08 23:08:40 -04002203 if (ext4_should_order_data(inode)) {
2204 int retval;
2205 retval = ext4_jbd2_file_inode(handle, inode);
2206 if (retval)
2207 /*
2208 * Failed to add inode for ordered
2209 * mode. Don't update file size
2210 */
2211 return retval;
2212 }
2213
Alex Tomas64769242008-07-11 19:27:31 -04002214 /*
2215 * Update on-disk size along with block allocation
2216 * we don't use 'extend_disksize' as size may change
2217 * within already allocated block -bzzz
2218 */
2219 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2220 if (disksize > i_size_read(inode))
2221 disksize = i_size_read(inode);
2222 if (disksize > EXT4_I(inode)->i_disksize) {
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04002223 ext4_update_i_disksize(inode, disksize);
2224 ret = ext4_mark_inode_dirty(handle, inode);
2225 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04002226 }
Alex Tomas64769242008-07-11 19:27:31 -04002227 ret = 0;
2228 }
Alex Tomas64769242008-07-11 19:27:31 -04002229 return ret;
2230}
Mingming Cao61628a32008-07-11 19:27:31 -04002231
2232static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2233{
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002234 /*
2235 * unmapped buffer is possible for holes.
2236 * delay buffer is possible with delayed allocation
2237 */
2238 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2239}
2240
2241static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2242 struct buffer_head *bh_result, int create)
2243{
2244 int ret = 0;
2245 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2246
2247 /*
2248 * we don't want to do block allocation in writepage
2249 * so call get_block_wrap with create = 0
2250 */
2251 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2252 bh_result, 0, 0, 0);
2253 if (ret > 0) {
2254 bh_result->b_size = (ret << inode->i_blkbits);
2255 ret = 0;
2256 }
2257 return ret;
Mingming Cao61628a32008-07-11 19:27:31 -04002258}
2259
Mingming Cao61628a32008-07-11 19:27:31 -04002260/*
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002261 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2262 * get called via journal_submit_inode_data_buffers (no journal handle)
2263 * get called via shrink_page_list via pdflush (no journal handle)
2264 * or grab_page_cache when doing write_begin (have journal handle)
Mingming Cao61628a32008-07-11 19:27:31 -04002265 */
Alex Tomas64769242008-07-11 19:27:31 -04002266static int ext4_da_writepage(struct page *page,
2267 struct writeback_control *wbc)
2268{
Alex Tomas64769242008-07-11 19:27:31 -04002269 int ret = 0;
Mingming Cao61628a32008-07-11 19:27:31 -04002270 loff_t size;
2271 unsigned long len;
Mingming Cao61628a32008-07-11 19:27:31 -04002272 struct buffer_head *page_bufs;
2273 struct inode *inode = page->mapping->host;
Alex Tomas64769242008-07-11 19:27:31 -04002274
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002275 size = i_size_read(inode);
2276 if (page->index == size >> PAGE_CACHE_SHIFT)
2277 len = size & ~PAGE_CACHE_MASK;
2278 else
2279 len = PAGE_CACHE_SIZE;
Alex Tomas64769242008-07-11 19:27:31 -04002280
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002281 if (page_has_buffers(page)) {
Mingming Cao61628a32008-07-11 19:27:31 -04002282 page_bufs = page_buffers(page);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002283 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2284 ext4_bh_unmapped_or_delay)) {
Mingming Cao61628a32008-07-11 19:27:31 -04002285 /*
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002286 * We don't want to do block allocation
2287 * So redirty the page and return
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -04002288 * We may reach here when we do a journal commit
2289 * via journal_submit_inode_data_buffers.
2290 * If we don't have mapping block we just ignore
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002291 * them. We can also reach here via shrink_page_list
2292 */
2293 redirty_page_for_writepage(wbc, page);
2294 unlock_page(page);
2295 return 0;
2296 }
2297 } else {
2298 /*
2299 * The test for page_has_buffers() is subtle:
2300 * We know the page is dirty but it lost buffers. That means
2301 * that at some moment in time after write_begin()/write_end()
2302 * has been called all buffers have been clean and thus they
2303 * must have been written at least once. So they are all
2304 * mapped and we can happily proceed with mapping them
2305 * and writing the page.
2306 *
2307 * Try to initialize the buffer_heads and check whether
2308 * all are mapped and non delay. We don't want to
2309 * do block allocation here.
2310 */
2311 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2312 ext4_normal_get_block_write);
2313 if (!ret) {
2314 page_bufs = page_buffers(page);
2315 /* check whether all are mapped and non delay */
2316 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2317 ext4_bh_unmapped_or_delay)) {
2318 redirty_page_for_writepage(wbc, page);
2319 unlock_page(page);
2320 return 0;
2321 }
2322 } else {
2323 /*
2324 * We can't do block allocation here
2325 * so just redity the page and unlock
2326 * and return
Mingming Cao61628a32008-07-11 19:27:31 -04002327 */
Mingming Cao61628a32008-07-11 19:27:31 -04002328 redirty_page_for_writepage(wbc, page);
2329 unlock_page(page);
2330 return 0;
2331 }
Alex Tomas64769242008-07-11 19:27:31 -04002332 }
2333
2334 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002335 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
Alex Tomas64769242008-07-11 19:27:31 -04002336 else
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002337 ret = block_write_full_page(page,
2338 ext4_normal_get_block_write,
2339 wbc);
Alex Tomas64769242008-07-11 19:27:31 -04002340
Alex Tomas64769242008-07-11 19:27:31 -04002341 return ret;
2342}
2343
Mingming Cao61628a32008-07-11 19:27:31 -04002344/*
Mingming Cao525f4ed2008-08-19 22:15:58 -04002345 * This is called via ext4_da_writepages() to
2346 * calulate the total number of credits to reserve to fit
2347 * a single extent allocation into a single transaction,
2348 * ext4_da_writpeages() will loop calling this before
2349 * the block allocation.
Mingming Cao61628a32008-07-11 19:27:31 -04002350 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002351
2352static int ext4_da_writepages_trans_blocks(struct inode *inode)
2353{
2354 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2355
2356 /*
2357 * With non-extent format the journal credit needed to
2358 * insert nrblocks contiguous block is dependent on
2359 * number of contiguous block. So we will limit
2360 * number of contiguous block to a sane value
2361 */
2362 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2363 (max_blocks > EXT4_MAX_TRANS_DATA))
2364 max_blocks = EXT4_MAX_TRANS_DATA;
2365
2366 return ext4_chunk_trans_blocks(inode, max_blocks);
2367}
Mingming Cao61628a32008-07-11 19:27:31 -04002368
Alex Tomas64769242008-07-11 19:27:31 -04002369static int ext4_da_writepages(struct address_space *mapping,
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002370 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04002371{
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002372 pgoff_t index;
2373 int range_whole = 0;
Mingming Cao61628a32008-07-11 19:27:31 -04002374 handle_t *handle = NULL;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002375 struct mpage_da_data mpd;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002376 struct inode *inode = mapping->host;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002377 int no_nrwrite_index_update;
2378 long pages_written = 0, pages_skipped;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002379 int needed_blocks, ret = 0, nr_to_writebump = 0;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002380 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
Mingming Cao61628a32008-07-11 19:27:31 -04002381
2382 /*
2383 * No pages to write? This is mainly a kludge to avoid starting
2384 * a transaction for special inodes like journal inode on last iput()
2385 * because that could violate lock ordering on umount
2386 */
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002387 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
Mingming Cao61628a32008-07-11 19:27:31 -04002388 return 0;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002389 /*
2390 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2391 * This make sure small files blocks are allocated in
2392 * single attempt. This ensure that small files
2393 * get less fragmented.
2394 */
2395 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2396 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2397 wbc->nr_to_write = sbi->s_mb_stream_request;
2398 }
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002399 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2400 range_whole = 1;
Mingming Cao61628a32008-07-11 19:27:31 -04002401
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002402 if (wbc->range_cyclic)
2403 index = mapping->writeback_index;
2404 else
2405 index = wbc->range_start >> PAGE_CACHE_SHIFT;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002406
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002407 mpd.wbc = wbc;
2408 mpd.inode = mapping->host;
2409
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002410 /*
2411 * we don't want write_cache_pages to update
2412 * nr_to_write and writeback_index
2413 */
2414 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2415 wbc->no_nrwrite_index_update = 1;
2416 pages_skipped = wbc->pages_skipped;
2417
2418 while (!ret && wbc->nr_to_write > 0) {
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002419
2420 /*
2421 * we insert one extent at a time. So we need
2422 * credit needed for single extent allocation.
2423 * journalled mode is currently not supported
2424 * by delalloc
2425 */
2426 BUG_ON(ext4_should_journal_data(inode));
Mingming Cao525f4ed2008-08-19 22:15:58 -04002427 needed_blocks = ext4_da_writepages_trans_blocks(inode);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002428
Mingming Cao61628a32008-07-11 19:27:31 -04002429 /* start a new transaction*/
2430 handle = ext4_journal_start(inode, needed_blocks);
2431 if (IS_ERR(handle)) {
2432 ret = PTR_ERR(handle);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002433 printk(KERN_EMERG "%s: jbd2_start: "
2434 "%ld pages, ino %lu; err %d\n", __func__,
2435 wbc->nr_to_write, inode->i_ino, ret);
2436 dump_stack();
Mingming Cao61628a32008-07-11 19:27:31 -04002437 goto out_writepages;
2438 }
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002439 mpd.get_block = ext4_da_get_block_write;
2440 ret = mpage_da_writepages(mapping, wbc, &mpd);
2441
Mingming Cao61628a32008-07-11 19:27:31 -04002442 ext4_journal_stop(handle);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002443
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002444 if (mpd.retval == -ENOSPC) {
2445 /* commit the transaction which would
2446 * free blocks released in the transaction
2447 * and try again
2448 */
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002449 jbd2_journal_force_commit_nested(sbi->s_journal);
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002450 wbc->pages_skipped = pages_skipped;
2451 ret = 0;
2452 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002453 /*
2454 * got one extent now try with
2455 * rest of the pages
2456 */
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002457 pages_written += mpd.pages_written;
2458 wbc->pages_skipped = pages_skipped;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002459 ret = 0;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002460 } else if (wbc->nr_to_write)
Mingming Cao61628a32008-07-11 19:27:31 -04002461 /*
2462 * There is no more writeout needed
2463 * or we requested for a noblocking writeout
2464 * and we found the device congested
2465 */
Mingming Cao61628a32008-07-11 19:27:31 -04002466 break;
Mingming Cao61628a32008-07-11 19:27:31 -04002467 }
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002468 if (pages_skipped != wbc->pages_skipped)
2469 printk(KERN_EMERG "This should not happen leaving %s "
2470 "with nr_to_write = %ld ret = %d\n",
2471 __func__, wbc->nr_to_write, ret);
Mingming Cao61628a32008-07-11 19:27:31 -04002472
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002473 /* Update index */
2474 index += pages_written;
2475 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2476 /*
2477 * set the writeback_index so that range_cyclic
2478 * mode will write it back later
2479 */
2480 mapping->writeback_index = index;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002481
Mingming Cao61628a32008-07-11 19:27:31 -04002482out_writepages:
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002483 if (!no_nrwrite_index_update)
2484 wbc->no_nrwrite_index_update = 0;
2485 wbc->nr_to_write -= nr_to_writebump;
Mingming Cao61628a32008-07-11 19:27:31 -04002486 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04002487}
2488
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002489#define FALL_BACK_TO_NONDELALLOC 1
2490static int ext4_nonda_switch(struct super_block *sb)
2491{
2492 s64 free_blocks, dirty_blocks;
2493 struct ext4_sb_info *sbi = EXT4_SB(sb);
2494
2495 /*
2496 * switch to non delalloc mode if we are running low
2497 * on free block. The free block accounting via percpu
2498 * counters can get slightly wrong with FBC_BATCH getting
2499 * accumulated on each CPU without updating global counters
2500 * Delalloc need an accurate free block accounting. So switch
2501 * to non delalloc when we are near to error range.
2502 */
2503 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2504 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2505 if (2 * free_blocks < 3 * dirty_blocks ||
2506 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
2507 /*
2508 * free block count is less that 150% of dirty blocks
2509 * or free blocks is less that watermark
2510 */
2511 return 1;
2512 }
2513 return 0;
2514}
2515
Alex Tomas64769242008-07-11 19:27:31 -04002516static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2517 loff_t pos, unsigned len, unsigned flags,
2518 struct page **pagep, void **fsdata)
2519{
Mingming Caod2a17632008-07-14 17:52:37 -04002520 int ret, retries = 0;
Alex Tomas64769242008-07-11 19:27:31 -04002521 struct page *page;
2522 pgoff_t index;
2523 unsigned from, to;
2524 struct inode *inode = mapping->host;
2525 handle_t *handle;
2526
2527 index = pos >> PAGE_CACHE_SHIFT;
2528 from = pos & (PAGE_CACHE_SIZE - 1);
2529 to = from + len;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002530
2531 if (ext4_nonda_switch(inode->i_sb)) {
2532 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2533 return ext4_write_begin(file, mapping, pos,
2534 len, flags, pagep, fsdata);
2535 }
2536 *fsdata = (void *)0;
Mingming Caod2a17632008-07-14 17:52:37 -04002537retry:
Alex Tomas64769242008-07-11 19:27:31 -04002538 /*
2539 * With delayed allocation, we don't log the i_disksize update
2540 * if there is delayed block allocation. But we still need
2541 * to journalling the i_disksize update if writes to the end
2542 * of file which has an already mapped buffer.
2543 */
2544 handle = ext4_journal_start(inode, 1);
2545 if (IS_ERR(handle)) {
2546 ret = PTR_ERR(handle);
2547 goto out;
2548 }
2549
2550 page = __grab_cache_page(mapping, index);
Eric Sandeend5a0d4f2008-08-02 18:51:06 -04002551 if (!page) {
2552 ext4_journal_stop(handle);
2553 ret = -ENOMEM;
2554 goto out;
2555 }
Alex Tomas64769242008-07-11 19:27:31 -04002556 *pagep = page;
2557
2558 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2559 ext4_da_get_block_prep);
2560 if (ret < 0) {
2561 unlock_page(page);
2562 ext4_journal_stop(handle);
2563 page_cache_release(page);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04002564 /*
2565 * block_write_begin may have instantiated a few blocks
2566 * outside i_size. Trim these off again. Don't need
2567 * i_size_read because we hold i_mutex.
2568 */
2569 if (pos + len > inode->i_size)
2570 vmtruncate(inode, inode->i_size);
Alex Tomas64769242008-07-11 19:27:31 -04002571 }
2572
Mingming Caod2a17632008-07-14 17:52:37 -04002573 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2574 goto retry;
Alex Tomas64769242008-07-11 19:27:31 -04002575out:
2576 return ret;
2577}
2578
Mingming Cao632eaea2008-07-11 19:27:31 -04002579/*
2580 * Check if we should update i_disksize
2581 * when write to the end of file but not require block allocation
2582 */
2583static int ext4_da_should_update_i_disksize(struct page *page,
2584 unsigned long offset)
2585{
2586 struct buffer_head *bh;
2587 struct inode *inode = page->mapping->host;
2588 unsigned int idx;
2589 int i;
2590
2591 bh = page_buffers(page);
2592 idx = offset >> inode->i_blkbits;
2593
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002594 for (i = 0; i < idx; i++)
Mingming Cao632eaea2008-07-11 19:27:31 -04002595 bh = bh->b_this_page;
2596
2597 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2598 return 0;
2599 return 1;
2600}
2601
Alex Tomas64769242008-07-11 19:27:31 -04002602static int ext4_da_write_end(struct file *file,
2603 struct address_space *mapping,
2604 loff_t pos, unsigned len, unsigned copied,
2605 struct page *page, void *fsdata)
2606{
2607 struct inode *inode = mapping->host;
2608 int ret = 0, ret2;
2609 handle_t *handle = ext4_journal_current_handle();
2610 loff_t new_i_size;
Mingming Cao632eaea2008-07-11 19:27:31 -04002611 unsigned long start, end;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002612 int write_mode = (int)(unsigned long)fsdata;
2613
2614 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2615 if (ext4_should_order_data(inode)) {
2616 return ext4_ordered_write_end(file, mapping, pos,
2617 len, copied, page, fsdata);
2618 } else if (ext4_should_writeback_data(inode)) {
2619 return ext4_writeback_write_end(file, mapping, pos,
2620 len, copied, page, fsdata);
2621 } else {
2622 BUG();
2623 }
2624 }
Mingming Cao632eaea2008-07-11 19:27:31 -04002625
2626 start = pos & (PAGE_CACHE_SIZE - 1);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002627 end = start + copied - 1;
Alex Tomas64769242008-07-11 19:27:31 -04002628
2629 /*
2630 * generic_write_end() will run mark_inode_dirty() if i_size
2631 * changes. So let's piggyback the i_disksize mark_inode_dirty
2632 * into that.
2633 */
2634
2635 new_i_size = pos + copied;
Mingming Cao632eaea2008-07-11 19:27:31 -04002636 if (new_i_size > EXT4_I(inode)->i_disksize) {
2637 if (ext4_da_should_update_i_disksize(page, end)) {
2638 down_write(&EXT4_I(inode)->i_data_sem);
2639 if (new_i_size > EXT4_I(inode)->i_disksize) {
2640 /*
2641 * Updating i_disksize when extending file
2642 * without needing block allocation
2643 */
2644 if (ext4_should_order_data(inode))
2645 ret = ext4_jbd2_file_inode(handle,
2646 inode);
Alex Tomas64769242008-07-11 19:27:31 -04002647
Mingming Cao632eaea2008-07-11 19:27:31 -04002648 EXT4_I(inode)->i_disksize = new_i_size;
2649 }
2650 up_write(&EXT4_I(inode)->i_data_sem);
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04002651 /* We need to mark inode dirty even if
2652 * new_i_size is less that inode->i_size
2653 * bu greater than i_disksize.(hint delalloc)
2654 */
2655 ext4_mark_inode_dirty(handle, inode);
Alex Tomas64769242008-07-11 19:27:31 -04002656 }
Mingming Cao632eaea2008-07-11 19:27:31 -04002657 }
Alex Tomas64769242008-07-11 19:27:31 -04002658 ret2 = generic_write_end(file, mapping, pos, len, copied,
2659 page, fsdata);
2660 copied = ret2;
2661 if (ret2 < 0)
2662 ret = ret2;
2663 ret2 = ext4_journal_stop(handle);
2664 if (!ret)
2665 ret = ret2;
2666
2667 return ret ? ret : copied;
2668}
2669
2670static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2671{
Alex Tomas64769242008-07-11 19:27:31 -04002672 /*
2673 * Drop reserved blocks
2674 */
2675 BUG_ON(!PageLocked(page));
2676 if (!page_has_buffers(page))
2677 goto out;
2678
Mingming Caod2a17632008-07-14 17:52:37 -04002679 ext4_da_page_release_reservation(page, offset);
Alex Tomas64769242008-07-11 19:27:31 -04002680
2681out:
2682 ext4_invalidatepage(page, offset);
2683
2684 return;
2685}
2686
2687
2688/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002689 * bmap() is special. It gets used by applications such as lilo and by
2690 * the swapper to find the on-disk block of a specific piece of data.
2691 *
2692 * Naturally, this is dangerous if the block concerned is still in the
Mingming Cao617ba132006-10-11 01:20:53 -07002693 * journal. If somebody makes a swapfile on an ext4 data-journaling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002694 * filesystem and enables swap, then they may get a nasty shock when the
2695 * data getting swapped to that swapfile suddenly gets overwritten by
2696 * the original zero's written out previously to the journal and
2697 * awaiting writeback in the kernel's buffer cache.
2698 *
2699 * So, if we see any bmap calls here on a modified, data-journaled file,
2700 * take extra steps to flush any blocks which might be in the cache.
2701 */
Mingming Cao617ba132006-10-11 01:20:53 -07002702static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002703{
2704 struct inode *inode = mapping->host;
2705 journal_t *journal;
2706 int err;
2707
Alex Tomas64769242008-07-11 19:27:31 -04002708 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2709 test_opt(inode->i_sb, DELALLOC)) {
2710 /*
2711 * With delalloc we want to sync the file
2712 * so that we can make sure we allocate
2713 * blocks for file
2714 */
2715 filemap_write_and_wait(mapping);
2716 }
2717
Mingming Cao617ba132006-10-11 01:20:53 -07002718 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002719 /*
2720 * This is a REALLY heavyweight approach, but the use of
2721 * bmap on dirty files is expected to be extremely rare:
2722 * only if we run lilo or swapon on a freshly made file
2723 * do we expect this to happen.
2724 *
2725 * (bmap requires CAP_SYS_RAWIO so this does not
2726 * represent an unprivileged user DOS attack --- we'd be
2727 * in trouble if mortal users could trigger this path at
2728 * will.)
2729 *
Mingming Cao617ba132006-10-11 01:20:53 -07002730 * NB. EXT4_STATE_JDATA is not set on files other than
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002731 * regular files. If somebody wants to bmap a directory
2732 * or symlink and gets confused because the buffer
2733 * hasn't yet been flushed to disk, they deserve
2734 * everything they get.
2735 */
2736
Mingming Cao617ba132006-10-11 01:20:53 -07002737 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2738 journal = EXT4_JOURNAL(inode);
Mingming Caodab291a2006-10-11 01:21:01 -07002739 jbd2_journal_lock_updates(journal);
2740 err = jbd2_journal_flush(journal);
2741 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002742
2743 if (err)
2744 return 0;
2745 }
2746
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002747 return generic_block_bmap(mapping, block, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002748}
2749
2750static int bget_one(handle_t *handle, struct buffer_head *bh)
2751{
2752 get_bh(bh);
2753 return 0;
2754}
2755
2756static int bput_one(handle_t *handle, struct buffer_head *bh)
2757{
2758 put_bh(bh);
2759 return 0;
2760}
2761
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002762/*
Jan Kara678aaf42008-07-11 19:27:31 -04002763 * Note that we don't need to start a transaction unless we're journaling data
2764 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2765 * need to file the inode to the transaction's list in ordered mode because if
2766 * we are writing back data added by write(), the inode is already there and if
2767 * we are writing back data modified via mmap(), noone guarantees in which
2768 * transaction the data will hit the disk. In case we are journaling data, we
2769 * cannot start transaction directly because transaction start ranks above page
2770 * lock so we have to do some magic.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002771 *
Jan Kara678aaf42008-07-11 19:27:31 -04002772 * In all journaling modes block_write_full_page() will start the I/O.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002773 *
2774 * Problem:
2775 *
Mingming Cao617ba132006-10-11 01:20:53 -07002776 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2777 * ext4_writepage()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002778 *
2779 * Similar for:
2780 *
Mingming Cao617ba132006-10-11 01:20:53 -07002781 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002782 *
Mingming Cao617ba132006-10-11 01:20:53 -07002783 * Same applies to ext4_get_block(). We will deadlock on various things like
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05002784 * lock_journal and i_data_sem
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002785 *
2786 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2787 * allocations fail.
2788 *
2789 * 16May01: If we're reentered then journal_current_handle() will be
2790 * non-zero. We simply *return*.
2791 *
2792 * 1 July 2001: @@@ FIXME:
2793 * In journalled data mode, a data buffer may be metadata against the
2794 * current transaction. But the same file is part of a shared mapping
2795 * and someone does a writepage() on it.
2796 *
2797 * We will move the buffer onto the async_data list, but *after* it has
2798 * been dirtied. So there's a small window where we have dirty data on
2799 * BJ_Metadata.
2800 *
2801 * Note that this only applies to the last partial page in the file. The
2802 * bit which block_write_full_page() uses prepare/commit for. (That's
2803 * broken code anyway: it's wrong for msync()).
2804 *
2805 * It's a rare case: affects the final partial page, for journalled data
2806 * where the file is subject to bith write() and writepage() in the same
2807 * transction. To fix it we'll need a custom block_write_full_page().
2808 * We'll probably need that anyway for journalling writepage() output.
2809 *
2810 * We don't honour synchronous mounts for writepage(). That would be
2811 * disastrous. Any write() or metadata operation will sync the fs for
2812 * us.
2813 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002814 */
Jan Kara678aaf42008-07-11 19:27:31 -04002815static int __ext4_normal_writepage(struct page *page,
Jan Karacf108bc2008-07-11 19:27:31 -04002816 struct writeback_control *wbc)
2817{
2818 struct inode *inode = page->mapping->host;
2819
2820 if (test_opt(inode->i_sb, NOBH))
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002821 return nobh_writepage(page,
2822 ext4_normal_get_block_write, wbc);
Jan Karacf108bc2008-07-11 19:27:31 -04002823 else
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002824 return block_write_full_page(page,
2825 ext4_normal_get_block_write,
2826 wbc);
Jan Karacf108bc2008-07-11 19:27:31 -04002827}
2828
Jan Kara678aaf42008-07-11 19:27:31 -04002829static int ext4_normal_writepage(struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002830 struct writeback_control *wbc)
2831{
2832 struct inode *inode = page->mapping->host;
Jan Karacf108bc2008-07-11 19:27:31 -04002833 loff_t size = i_size_read(inode);
2834 loff_t len;
2835
2836 J_ASSERT(PageLocked(page));
Jan Karacf108bc2008-07-11 19:27:31 -04002837 if (page->index == size >> PAGE_CACHE_SHIFT)
2838 len = size & ~PAGE_CACHE_MASK;
2839 else
2840 len = PAGE_CACHE_SIZE;
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002841
2842 if (page_has_buffers(page)) {
2843 /* if page has buffers it should all be mapped
2844 * and allocated. If there are not buffers attached
2845 * to the page we know the page is dirty but it lost
2846 * buffers. That means that at some moment in time
2847 * after write_begin() / write_end() has been called
2848 * all buffers have been clean and thus they must have been
2849 * written at least once. So they are all mapped and we can
2850 * happily proceed with mapping them and writing the page.
2851 */
2852 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2853 ext4_bh_unmapped_or_delay));
2854 }
Jan Karacf108bc2008-07-11 19:27:31 -04002855
2856 if (!ext4_journal_current_handle())
Jan Kara678aaf42008-07-11 19:27:31 -04002857 return __ext4_normal_writepage(page, wbc);
Jan Karacf108bc2008-07-11 19:27:31 -04002858
2859 redirty_page_for_writepage(wbc, page);
2860 unlock_page(page);
2861 return 0;
2862}
2863
2864static int __ext4_journalled_writepage(struct page *page,
2865 struct writeback_control *wbc)
2866{
2867 struct address_space *mapping = page->mapping;
2868 struct inode *inode = mapping->host;
2869 struct buffer_head *page_bufs;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002870 handle_t *handle = NULL;
2871 int ret = 0;
2872 int err;
2873
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002874 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2875 ext4_normal_get_block_write);
Jan Karacf108bc2008-07-11 19:27:31 -04002876 if (ret != 0)
2877 goto out_unlock;
2878
2879 page_bufs = page_buffers(page);
2880 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
2881 bget_one);
2882 /* As soon as we unlock the page, it can go away, but we have
2883 * references to buffers so we are safe */
2884 unlock_page(page);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002885
Mingming Cao617ba132006-10-11 01:20:53 -07002886 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002887 if (IS_ERR(handle)) {
2888 ret = PTR_ERR(handle);
Jan Karacf108bc2008-07-11 19:27:31 -04002889 goto out;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002890 }
2891
Jan Karacf108bc2008-07-11 19:27:31 -04002892 ret = walk_page_buffers(handle, page_bufs, 0,
2893 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002894
Jan Karacf108bc2008-07-11 19:27:31 -04002895 err = walk_page_buffers(handle, page_bufs, 0,
2896 PAGE_CACHE_SIZE, NULL, write_end_fn);
2897 if (ret == 0)
2898 ret = err;
Mingming Cao617ba132006-10-11 01:20:53 -07002899 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002900 if (!ret)
2901 ret = err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002902
Jan Karacf108bc2008-07-11 19:27:31 -04002903 walk_page_buffers(handle, page_bufs, 0,
2904 PAGE_CACHE_SIZE, NULL, bput_one);
2905 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2906 goto out;
2907
2908out_unlock:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002909 unlock_page(page);
Jan Karacf108bc2008-07-11 19:27:31 -04002910out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002911 return ret;
2912}
2913
Mingming Cao617ba132006-10-11 01:20:53 -07002914static int ext4_journalled_writepage(struct page *page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002915 struct writeback_control *wbc)
2916{
2917 struct inode *inode = page->mapping->host;
Jan Karacf108bc2008-07-11 19:27:31 -04002918 loff_t size = i_size_read(inode);
2919 loff_t len;
2920
2921 J_ASSERT(PageLocked(page));
Jan Karacf108bc2008-07-11 19:27:31 -04002922 if (page->index == size >> PAGE_CACHE_SHIFT)
2923 len = size & ~PAGE_CACHE_MASK;
2924 else
2925 len = PAGE_CACHE_SIZE;
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002926
2927 if (page_has_buffers(page)) {
2928 /* if page has buffers it should all be mapped
2929 * and allocated. If there are not buffers attached
2930 * to the page we know the page is dirty but it lost
2931 * buffers. That means that at some moment in time
2932 * after write_begin() / write_end() has been called
2933 * all buffers have been clean and thus they must have been
2934 * written at least once. So they are all mapped and we can
2935 * happily proceed with mapping them and writing the page.
2936 */
2937 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2938 ext4_bh_unmapped_or_delay));
2939 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002940
Mingming Cao617ba132006-10-11 01:20:53 -07002941 if (ext4_journal_current_handle())
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002942 goto no_write;
2943
Jan Karacf108bc2008-07-11 19:27:31 -04002944 if (PageChecked(page)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002945 /*
2946 * It's mmapped pagecache. Add buffers and journal it. There
2947 * doesn't seem much point in redirtying the page here.
2948 */
2949 ClearPageChecked(page);
Jan Karacf108bc2008-07-11 19:27:31 -04002950 return __ext4_journalled_writepage(page, wbc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002951 } else {
2952 /*
2953 * It may be a page full of checkpoint-mode buffers. We don't
2954 * really know unless we go poke around in the buffer_heads.
2955 * But block_write_full_page will do the right thing.
2956 */
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002957 return block_write_full_page(page,
2958 ext4_normal_get_block_write,
2959 wbc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002960 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002961no_write:
2962 redirty_page_for_writepage(wbc, page);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002963 unlock_page(page);
Jan Karacf108bc2008-07-11 19:27:31 -04002964 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002965}
2966
Mingming Cao617ba132006-10-11 01:20:53 -07002967static int ext4_readpage(struct file *file, struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002968{
Mingming Cao617ba132006-10-11 01:20:53 -07002969 return mpage_readpage(page, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002970}
2971
2972static int
Mingming Cao617ba132006-10-11 01:20:53 -07002973ext4_readpages(struct file *file, struct address_space *mapping,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002974 struct list_head *pages, unsigned nr_pages)
2975{
Mingming Cao617ba132006-10-11 01:20:53 -07002976 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002977}
2978
Mingming Cao617ba132006-10-11 01:20:53 -07002979static void ext4_invalidatepage(struct page *page, unsigned long offset)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002980{
Mingming Cao617ba132006-10-11 01:20:53 -07002981 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002982
2983 /*
2984 * If it's a full truncate we just forget about the pending dirtying
2985 */
2986 if (offset == 0)
2987 ClearPageChecked(page);
2988
Mingming Caodab291a2006-10-11 01:21:01 -07002989 jbd2_journal_invalidatepage(journal, page, offset);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002990}
2991
Mingming Cao617ba132006-10-11 01:20:53 -07002992static int ext4_releasepage(struct page *page, gfp_t wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002993{
Mingming Cao617ba132006-10-11 01:20:53 -07002994 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002995
2996 WARN_ON(PageChecked(page));
2997 if (!page_has_buffers(page))
2998 return 0;
Mingming Caodab291a2006-10-11 01:21:01 -07002999 return jbd2_journal_try_to_free_buffers(journal, page, wait);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003000}
3001
3002/*
3003 * If the O_DIRECT write will extend the file then add this inode to the
3004 * orphan list. So recovery will truncate it back to the original size
3005 * if the machine crashes during the write.
3006 *
3007 * If the O_DIRECT write is intantiating holes inside i_size and the machine
Jan Kara7fb54092008-02-10 01:08:38 -05003008 * crashes then stale disk data _may_ be exposed inside the file. But current
3009 * VFS code falls back into buffered path in that case so we are safe.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003010 */
Mingming Cao617ba132006-10-11 01:20:53 -07003011static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003012 const struct iovec *iov, loff_t offset,
3013 unsigned long nr_segs)
3014{
3015 struct file *file = iocb->ki_filp;
3016 struct inode *inode = file->f_mapping->host;
Mingming Cao617ba132006-10-11 01:20:53 -07003017 struct ext4_inode_info *ei = EXT4_I(inode);
Jan Kara7fb54092008-02-10 01:08:38 -05003018 handle_t *handle;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003019 ssize_t ret;
3020 int orphan = 0;
3021 size_t count = iov_length(iov, nr_segs);
3022
3023 if (rw == WRITE) {
3024 loff_t final_size = offset + count;
3025
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003026 if (final_size > inode->i_size) {
Jan Kara7fb54092008-02-10 01:08:38 -05003027 /* Credits for sb + inode write */
3028 handle = ext4_journal_start(inode, 2);
3029 if (IS_ERR(handle)) {
3030 ret = PTR_ERR(handle);
3031 goto out;
3032 }
Mingming Cao617ba132006-10-11 01:20:53 -07003033 ret = ext4_orphan_add(handle, inode);
Jan Kara7fb54092008-02-10 01:08:38 -05003034 if (ret) {
3035 ext4_journal_stop(handle);
3036 goto out;
3037 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003038 orphan = 1;
3039 ei->i_disksize = inode->i_size;
Jan Kara7fb54092008-02-10 01:08:38 -05003040 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003041 }
3042 }
3043
3044 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3045 offset, nr_segs,
Mingming Cao617ba132006-10-11 01:20:53 -07003046 ext4_get_block, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003047
Jan Kara7fb54092008-02-10 01:08:38 -05003048 if (orphan) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003049 int err;
3050
Jan Kara7fb54092008-02-10 01:08:38 -05003051 /* Credits for sb + inode write */
3052 handle = ext4_journal_start(inode, 2);
3053 if (IS_ERR(handle)) {
3054 /* This is really bad luck. We've written the data
3055 * but cannot extend i_size. Bail out and pretend
3056 * the write failed... */
3057 ret = PTR_ERR(handle);
3058 goto out;
3059 }
3060 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07003061 ext4_orphan_del(handle, inode);
Jan Kara7fb54092008-02-10 01:08:38 -05003062 if (ret > 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003063 loff_t end = offset + ret;
3064 if (end > inode->i_size) {
3065 ei->i_disksize = end;
3066 i_size_write(inode, end);
3067 /*
3068 * We're going to return a positive `ret'
3069 * here due to non-zero-length I/O, so there's
3070 * no way of reporting error returns from
Mingming Cao617ba132006-10-11 01:20:53 -07003071 * ext4_mark_inode_dirty() to userspace. So
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003072 * ignore it.
3073 */
Mingming Cao617ba132006-10-11 01:20:53 -07003074 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003075 }
3076 }
Mingming Cao617ba132006-10-11 01:20:53 -07003077 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003078 if (ret == 0)
3079 ret = err;
3080 }
3081out:
3082 return ret;
3083}
3084
3085/*
Mingming Cao617ba132006-10-11 01:20:53 -07003086 * Pages can be marked dirty completely asynchronously from ext4's journalling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003087 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3088 * much here because ->set_page_dirty is called under VFS locks. The page is
3089 * not necessarily locked.
3090 *
3091 * We cannot just dirty the page and leave attached buffers clean, because the
3092 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3093 * or jbddirty because all the journalling code will explode.
3094 *
3095 * So what we do is to mark the page "pending dirty" and next time writepage
3096 * is called, propagate that into the buffers appropriately.
3097 */
Mingming Cao617ba132006-10-11 01:20:53 -07003098static int ext4_journalled_set_page_dirty(struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003099{
3100 SetPageChecked(page);
3101 return __set_page_dirty_nobuffers(page);
3102}
3103
Mingming Cao617ba132006-10-11 01:20:53 -07003104static const struct address_space_operations ext4_ordered_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003105 .readpage = ext4_readpage,
3106 .readpages = ext4_readpages,
3107 .writepage = ext4_normal_writepage,
3108 .sync_page = block_sync_page,
3109 .write_begin = ext4_write_begin,
3110 .write_end = ext4_ordered_write_end,
3111 .bmap = ext4_bmap,
3112 .invalidatepage = ext4_invalidatepage,
3113 .releasepage = ext4_releasepage,
3114 .direct_IO = ext4_direct_IO,
3115 .migratepage = buffer_migrate_page,
3116 .is_partially_uptodate = block_is_partially_uptodate,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003117};
3118
Mingming Cao617ba132006-10-11 01:20:53 -07003119static const struct address_space_operations ext4_writeback_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003120 .readpage = ext4_readpage,
3121 .readpages = ext4_readpages,
3122 .writepage = ext4_normal_writepage,
3123 .sync_page = block_sync_page,
3124 .write_begin = ext4_write_begin,
3125 .write_end = ext4_writeback_write_end,
3126 .bmap = ext4_bmap,
3127 .invalidatepage = ext4_invalidatepage,
3128 .releasepage = ext4_releasepage,
3129 .direct_IO = ext4_direct_IO,
3130 .migratepage = buffer_migrate_page,
3131 .is_partially_uptodate = block_is_partially_uptodate,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003132};
3133
Mingming Cao617ba132006-10-11 01:20:53 -07003134static const struct address_space_operations ext4_journalled_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003135 .readpage = ext4_readpage,
3136 .readpages = ext4_readpages,
3137 .writepage = ext4_journalled_writepage,
3138 .sync_page = block_sync_page,
3139 .write_begin = ext4_write_begin,
3140 .write_end = ext4_journalled_write_end,
3141 .set_page_dirty = ext4_journalled_set_page_dirty,
3142 .bmap = ext4_bmap,
3143 .invalidatepage = ext4_invalidatepage,
3144 .releasepage = ext4_releasepage,
3145 .is_partially_uptodate = block_is_partially_uptodate,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003146};
3147
Alex Tomas64769242008-07-11 19:27:31 -04003148static const struct address_space_operations ext4_da_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003149 .readpage = ext4_readpage,
3150 .readpages = ext4_readpages,
3151 .writepage = ext4_da_writepage,
3152 .writepages = ext4_da_writepages,
3153 .sync_page = block_sync_page,
3154 .write_begin = ext4_da_write_begin,
3155 .write_end = ext4_da_write_end,
3156 .bmap = ext4_bmap,
3157 .invalidatepage = ext4_da_invalidatepage,
3158 .releasepage = ext4_releasepage,
3159 .direct_IO = ext4_direct_IO,
3160 .migratepage = buffer_migrate_page,
3161 .is_partially_uptodate = block_is_partially_uptodate,
Alex Tomas64769242008-07-11 19:27:31 -04003162};
3163
Mingming Cao617ba132006-10-11 01:20:53 -07003164void ext4_set_aops(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003165{
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -04003166 if (ext4_should_order_data(inode) &&
3167 test_opt(inode->i_sb, DELALLOC))
3168 inode->i_mapping->a_ops = &ext4_da_aops;
3169 else if (ext4_should_order_data(inode))
Mingming Cao617ba132006-10-11 01:20:53 -07003170 inode->i_mapping->a_ops = &ext4_ordered_aops;
Alex Tomas64769242008-07-11 19:27:31 -04003171 else if (ext4_should_writeback_data(inode) &&
3172 test_opt(inode->i_sb, DELALLOC))
3173 inode->i_mapping->a_ops = &ext4_da_aops;
Mingming Cao617ba132006-10-11 01:20:53 -07003174 else if (ext4_should_writeback_data(inode))
3175 inode->i_mapping->a_ops = &ext4_writeback_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003176 else
Mingming Cao617ba132006-10-11 01:20:53 -07003177 inode->i_mapping->a_ops = &ext4_journalled_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003178}
3179
3180/*
Mingming Cao617ba132006-10-11 01:20:53 -07003181 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003182 * up to the end of the block which corresponds to `from'.
3183 * This required during truncate. We need to physically zero the tail end
3184 * of that block so it doesn't yield old data if the file is later grown.
3185 */
Jan Karacf108bc2008-07-11 19:27:31 -04003186int ext4_block_truncate_page(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003187 struct address_space *mapping, loff_t from)
3188{
Mingming Cao617ba132006-10-11 01:20:53 -07003189 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003190 unsigned offset = from & (PAGE_CACHE_SIZE-1);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003191 unsigned blocksize, length, pos;
3192 ext4_lblk_t iblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003193 struct inode *inode = mapping->host;
3194 struct buffer_head *bh;
Jan Karacf108bc2008-07-11 19:27:31 -04003195 struct page *page;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003196 int err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003197
Jan Karacf108bc2008-07-11 19:27:31 -04003198 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3199 if (!page)
3200 return -EINVAL;
3201
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003202 blocksize = inode->i_sb->s_blocksize;
3203 length = blocksize - (offset & (blocksize - 1));
3204 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3205
3206 /*
3207 * For "nobh" option, we can only work if we don't need to
3208 * read-in the page - otherwise we create buffers to do the IO.
3209 */
3210 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
Mingming Cao617ba132006-10-11 01:20:53 -07003211 ext4_should_writeback_data(inode) && PageUptodate(page)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08003212 zero_user(page, offset, length);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003213 set_page_dirty(page);
3214 goto unlock;
3215 }
3216
3217 if (!page_has_buffers(page))
3218 create_empty_buffers(page, blocksize, 0);
3219
3220 /* Find the buffer that contains "offset" */
3221 bh = page_buffers(page);
3222 pos = blocksize;
3223 while (offset >= pos) {
3224 bh = bh->b_this_page;
3225 iblock++;
3226 pos += blocksize;
3227 }
3228
3229 err = 0;
3230 if (buffer_freed(bh)) {
3231 BUFFER_TRACE(bh, "freed: skip");
3232 goto unlock;
3233 }
3234
3235 if (!buffer_mapped(bh)) {
3236 BUFFER_TRACE(bh, "unmapped");
Mingming Cao617ba132006-10-11 01:20:53 -07003237 ext4_get_block(inode, iblock, bh, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003238 /* unmapped? It's a hole - nothing to do */
3239 if (!buffer_mapped(bh)) {
3240 BUFFER_TRACE(bh, "still unmapped");
3241 goto unlock;
3242 }
3243 }
3244
3245 /* Ok, it's mapped. Make sure it's up-to-date */
3246 if (PageUptodate(page))
3247 set_buffer_uptodate(bh);
3248
3249 if (!buffer_uptodate(bh)) {
3250 err = -EIO;
3251 ll_rw_block(READ, 1, &bh);
3252 wait_on_buffer(bh);
3253 /* Uhhuh. Read error. Complain and punt. */
3254 if (!buffer_uptodate(bh))
3255 goto unlock;
3256 }
3257
Mingming Cao617ba132006-10-11 01:20:53 -07003258 if (ext4_should_journal_data(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003259 BUFFER_TRACE(bh, "get write access");
Mingming Cao617ba132006-10-11 01:20:53 -07003260 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003261 if (err)
3262 goto unlock;
3263 }
3264
Christoph Lametereebd2aa2008-02-04 22:28:29 -08003265 zero_user(page, offset, length);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003266
3267 BUFFER_TRACE(bh, "zeroed end of block");
3268
3269 err = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07003270 if (ext4_should_journal_data(inode)) {
3271 err = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003272 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07003273 if (ext4_should_order_data(inode))
Jan Kara678aaf42008-07-11 19:27:31 -04003274 err = ext4_jbd2_file_inode(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003275 mark_buffer_dirty(bh);
3276 }
3277
3278unlock:
3279 unlock_page(page);
3280 page_cache_release(page);
3281 return err;
3282}
3283
3284/*
3285 * Probably it should be a library function... search for first non-zero word
3286 * or memcmp with zero_page, whatever is better for particular architecture.
3287 * Linus?
3288 */
3289static inline int all_zeroes(__le32 *p, __le32 *q)
3290{
3291 while (p < q)
3292 if (*p++)
3293 return 0;
3294 return 1;
3295}
3296
3297/**
Mingming Cao617ba132006-10-11 01:20:53 -07003298 * ext4_find_shared - find the indirect blocks for partial truncation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003299 * @inode: inode in question
3300 * @depth: depth of the affected branch
Mingming Cao617ba132006-10-11 01:20:53 -07003301 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003302 * @chain: place to store the pointers to partial indirect blocks
3303 * @top: place to the (detached) top of branch
3304 *
Mingming Cao617ba132006-10-11 01:20:53 -07003305 * This is a helper function used by ext4_truncate().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003306 *
3307 * When we do truncate() we may have to clean the ends of several
3308 * indirect blocks but leave the blocks themselves alive. Block is
3309 * partially truncated if some data below the new i_size is refered
3310 * from it (and it is on the path to the first completely truncated
3311 * data block, indeed). We have to free the top of that path along
3312 * with everything to the right of the path. Since no allocation
Mingming Cao617ba132006-10-11 01:20:53 -07003313 * past the truncation point is possible until ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003314 * finishes, we may safely do the latter, but top of branch may
3315 * require special attention - pageout below the truncation point
3316 * might try to populate it.
3317 *
3318 * We atomically detach the top of branch from the tree, store the
3319 * block number of its root in *@top, pointers to buffer_heads of
3320 * partially truncated blocks - in @chain[].bh and pointers to
3321 * their last elements that should not be removed - in
3322 * @chain[].p. Return value is the pointer to last filled element
3323 * of @chain.
3324 *
3325 * The work left to caller to do the actual freeing of subtrees:
3326 * a) free the subtree starting from *@top
3327 * b) free the subtrees whose roots are stored in
3328 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3329 * c) free the subtrees growing from the inode past the @chain[0].
3330 * (no partially truncated stuff there). */
3331
Mingming Cao617ba132006-10-11 01:20:53 -07003332static Indirect *ext4_find_shared(struct inode *inode, int depth,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003333 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003334{
3335 Indirect *partial, *p;
3336 int k, err;
3337
3338 *top = 0;
3339 /* Make k index the deepest non-null offest + 1 */
3340 for (k = depth; k > 1 && !offsets[k-1]; k--)
3341 ;
Mingming Cao617ba132006-10-11 01:20:53 -07003342 partial = ext4_get_branch(inode, k, offsets, chain, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003343 /* Writer: pointers */
3344 if (!partial)
3345 partial = chain + k-1;
3346 /*
3347 * If the branch acquired continuation since we've looked at it -
3348 * fine, it should all survive and (new) top doesn't belong to us.
3349 */
3350 if (!partial->key && *partial->p)
3351 /* Writer: end */
3352 goto no_top;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003353 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003354 ;
3355 /*
3356 * OK, we've found the last block that must survive. The rest of our
3357 * branch should be detached before unlocking. However, if that rest
3358 * of branch is all ours and does not grow immediately from the inode
3359 * it's easier to cheat and just decrement partial->p.
3360 */
3361 if (p == chain + k - 1 && p > chain) {
3362 p->p--;
3363 } else {
3364 *top = *p->p;
Mingming Cao617ba132006-10-11 01:20:53 -07003365 /* Nope, don't do this in ext4. Must leave the tree intact */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003366#if 0
3367 *p->p = 0;
3368#endif
3369 }
3370 /* Writer: end */
3371
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003372 while (partial > p) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003373 brelse(partial->bh);
3374 partial--;
3375 }
3376no_top:
3377 return partial;
3378}
3379
3380/*
3381 * Zero a number of block pointers in either an inode or an indirect block.
3382 * If we restart the transaction we must again get write access to the
3383 * indirect block for further modification.
3384 *
3385 * We release `count' blocks on disk, but (last - first) may be greater
3386 * than `count' because there can be holes in there.
3387 */
Mingming Cao617ba132006-10-11 01:20:53 -07003388static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3389 struct buffer_head *bh, ext4_fsblk_t block_to_free,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003390 unsigned long count, __le32 *first, __le32 *last)
3391{
3392 __le32 *p;
3393 if (try_to_extend_transaction(handle, inode)) {
3394 if (bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07003395 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
3396 ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003397 }
Mingming Cao617ba132006-10-11 01:20:53 -07003398 ext4_mark_inode_dirty(handle, inode);
3399 ext4_journal_test_restart(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003400 if (bh) {
3401 BUFFER_TRACE(bh, "retaking write access");
Mingming Cao617ba132006-10-11 01:20:53 -07003402 ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003403 }
3404 }
3405
3406 /*
3407 * Any buffers which are on the journal will be in memory. We find
Mingming Caodab291a2006-10-11 01:21:01 -07003408 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003409 * on them. We've already detached each block from the file, so
Mingming Caodab291a2006-10-11 01:21:01 -07003410 * bforget() in jbd2_journal_forget() should be safe.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003411 *
Mingming Caodab291a2006-10-11 01:21:01 -07003412 * AKPM: turn on bforget in jbd2_journal_forget()!!!
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003413 */
3414 for (p = first; p < last; p++) {
3415 u32 nr = le32_to_cpu(*p);
3416 if (nr) {
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05003417 struct buffer_head *tbh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003418
3419 *p = 0;
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05003420 tbh = sb_find_get_block(inode->i_sb, nr);
3421 ext4_forget(handle, 0, inode, tbh, nr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003422 }
3423 }
3424
Alex Tomasc9de5602008-01-29 00:19:52 -05003425 ext4_free_blocks(handle, inode, block_to_free, count, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003426}
3427
3428/**
Mingming Cao617ba132006-10-11 01:20:53 -07003429 * ext4_free_data - free a list of data blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003430 * @handle: handle for this transaction
3431 * @inode: inode we are dealing with
3432 * @this_bh: indirect buffer_head which contains *@first and *@last
3433 * @first: array of block numbers
3434 * @last: points immediately past the end of array
3435 *
3436 * We are freeing all blocks refered from that array (numbers are stored as
3437 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3438 *
3439 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3440 * blocks are contiguous then releasing them at one time will only affect one
3441 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3442 * actually use a lot of journal space.
3443 *
3444 * @this_bh will be %NULL if @first and @last point into the inode's direct
3445 * block pointers.
3446 */
Mingming Cao617ba132006-10-11 01:20:53 -07003447static void ext4_free_data(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003448 struct buffer_head *this_bh,
3449 __le32 *first, __le32 *last)
3450{
Mingming Cao617ba132006-10-11 01:20:53 -07003451 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003452 unsigned long count = 0; /* Number of blocks in the run */
3453 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3454 corresponding to
3455 block_to_free */
Mingming Cao617ba132006-10-11 01:20:53 -07003456 ext4_fsblk_t nr; /* Current block # */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003457 __le32 *p; /* Pointer into inode/ind
3458 for current block */
3459 int err;
3460
3461 if (this_bh) { /* For indirect block */
3462 BUFFER_TRACE(this_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07003463 err = ext4_journal_get_write_access(handle, this_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003464 /* Important: if we can't update the indirect pointers
3465 * to the blocks, we can't free them. */
3466 if (err)
3467 return;
3468 }
3469
3470 for (p = first; p < last; p++) {
3471 nr = le32_to_cpu(*p);
3472 if (nr) {
3473 /* accumulate blocks to free if they're contiguous */
3474 if (count == 0) {
3475 block_to_free = nr;
3476 block_to_free_p = p;
3477 count = 1;
3478 } else if (nr == block_to_free + count) {
3479 count++;
3480 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07003481 ext4_clear_blocks(handle, inode, this_bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003482 block_to_free,
3483 count, block_to_free_p, p);
3484 block_to_free = nr;
3485 block_to_free_p = p;
3486 count = 1;
3487 }
3488 }
3489 }
3490
3491 if (count > 0)
Mingming Cao617ba132006-10-11 01:20:53 -07003492 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003493 count, block_to_free_p, p);
3494
3495 if (this_bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07003496 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
Duane Griffin71dc8fb2008-07-11 19:27:31 -04003497
3498 /*
3499 * The buffer head should have an attached journal head at this
3500 * point. However, if the data is corrupted and an indirect
3501 * block pointed to itself, it would have been detached when
3502 * the block was cleared. Check for this instead of OOPSing.
3503 */
3504 if (bh2jh(this_bh))
3505 ext4_journal_dirty_metadata(handle, this_bh);
3506 else
3507 ext4_error(inode->i_sb, __func__,
3508 "circular indirect block detected, "
3509 "inode=%lu, block=%llu",
3510 inode->i_ino,
3511 (unsigned long long) this_bh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003512 }
3513}
3514
3515/**
Mingming Cao617ba132006-10-11 01:20:53 -07003516 * ext4_free_branches - free an array of branches
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003517 * @handle: JBD handle for this transaction
3518 * @inode: inode we are dealing with
3519 * @parent_bh: the buffer_head which contains *@first and *@last
3520 * @first: array of block numbers
3521 * @last: pointer immediately past the end of array
3522 * @depth: depth of the branches to free
3523 *
3524 * We are freeing all blocks refered from these branches (numbers are
3525 * stored as little-endian 32-bit) and updating @inode->i_blocks
3526 * appropriately.
3527 */
Mingming Cao617ba132006-10-11 01:20:53 -07003528static void ext4_free_branches(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003529 struct buffer_head *parent_bh,
3530 __le32 *first, __le32 *last, int depth)
3531{
Mingming Cao617ba132006-10-11 01:20:53 -07003532 ext4_fsblk_t nr;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003533 __le32 *p;
3534
3535 if (is_handle_aborted(handle))
3536 return;
3537
3538 if (depth--) {
3539 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -07003540 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003541 p = last;
3542 while (--p >= first) {
3543 nr = le32_to_cpu(*p);
3544 if (!nr)
3545 continue; /* A hole */
3546
3547 /* Go read the buffer for the next level down */
3548 bh = sb_bread(inode->i_sb, nr);
3549
3550 /*
3551 * A read failure? Report error and clear slot
3552 * (should be rare).
3553 */
3554 if (!bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07003555 ext4_error(inode->i_sb, "ext4_free_branches",
Mingming Cao2ae02102006-10-11 01:21:11 -07003556 "Read failure, inode=%lu, block=%llu",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003557 inode->i_ino, nr);
3558 continue;
3559 }
3560
3561 /* This zaps the entire block. Bottom up. */
3562 BUFFER_TRACE(bh, "free child branches");
Mingming Cao617ba132006-10-11 01:20:53 -07003563 ext4_free_branches(handle, inode, bh,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003564 (__le32 *) bh->b_data,
3565 (__le32 *) bh->b_data + addr_per_block,
3566 depth);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003567
3568 /*
3569 * We've probably journalled the indirect block several
3570 * times during the truncate. But it's no longer
3571 * needed and we now drop it from the transaction via
Mingming Caodab291a2006-10-11 01:21:01 -07003572 * jbd2_journal_revoke().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003573 *
3574 * That's easy if it's exclusively part of this
3575 * transaction. But if it's part of the committing
Mingming Caodab291a2006-10-11 01:21:01 -07003576 * transaction then jbd2_journal_forget() will simply
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003577 * brelse() it. That means that if the underlying
Mingming Cao617ba132006-10-11 01:20:53 -07003578 * block is reallocated in ext4_get_block(),
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003579 * unmap_underlying_metadata() will find this block
3580 * and will try to get rid of it. damn, damn.
3581 *
3582 * If this block has already been committed to the
3583 * journal, a revoke record will be written. And
3584 * revoke records must be emitted *before* clearing
3585 * this block's bit in the bitmaps.
3586 */
Mingming Cao617ba132006-10-11 01:20:53 -07003587 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003588
3589 /*
3590 * Everything below this this pointer has been
3591 * released. Now let this top-of-subtree go.
3592 *
3593 * We want the freeing of this indirect block to be
3594 * atomic in the journal with the updating of the
3595 * bitmap block which owns it. So make some room in
3596 * the journal.
3597 *
3598 * We zero the parent pointer *after* freeing its
3599 * pointee in the bitmaps, so if extend_transaction()
3600 * for some reason fails to put the bitmap changes and
3601 * the release into the same transaction, recovery
3602 * will merely complain about releasing a free block,
3603 * rather than leaking blocks.
3604 */
3605 if (is_handle_aborted(handle))
3606 return;
3607 if (try_to_extend_transaction(handle, inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07003608 ext4_mark_inode_dirty(handle, inode);
3609 ext4_journal_test_restart(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003610 }
3611
Alex Tomasc9de5602008-01-29 00:19:52 -05003612 ext4_free_blocks(handle, inode, nr, 1, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003613
3614 if (parent_bh) {
3615 /*
3616 * The block which we have just freed is
3617 * pointed to by an indirect block: journal it
3618 */
3619 BUFFER_TRACE(parent_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07003620 if (!ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003621 parent_bh)){
3622 *p = 0;
3623 BUFFER_TRACE(parent_bh,
Mingming Cao617ba132006-10-11 01:20:53 -07003624 "call ext4_journal_dirty_metadata");
3625 ext4_journal_dirty_metadata(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003626 parent_bh);
3627 }
3628 }
3629 }
3630 } else {
3631 /* We have reached the bottom of the tree. */
3632 BUFFER_TRACE(parent_bh, "free data blocks");
Mingming Cao617ba132006-10-11 01:20:53 -07003633 ext4_free_data(handle, inode, parent_bh, first, last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003634 }
3635}
3636
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003637int ext4_can_truncate(struct inode *inode)
3638{
3639 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3640 return 0;
3641 if (S_ISREG(inode->i_mode))
3642 return 1;
3643 if (S_ISDIR(inode->i_mode))
3644 return 1;
3645 if (S_ISLNK(inode->i_mode))
3646 return !ext4_inode_is_fast_symlink(inode);
3647 return 0;
3648}
3649
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003650/*
Mingming Cao617ba132006-10-11 01:20:53 -07003651 * ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003652 *
Mingming Cao617ba132006-10-11 01:20:53 -07003653 * We block out ext4_get_block() block instantiations across the entire
3654 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003655 * simultaneously on behalf of the same inode.
3656 *
3657 * As we work through the truncate and commmit bits of it to the journal there
3658 * is one core, guiding principle: the file's tree must always be consistent on
3659 * disk. We must be able to restart the truncate after a crash.
3660 *
3661 * The file's tree may be transiently inconsistent in memory (although it
3662 * probably isn't), but whenever we close off and commit a journal transaction,
3663 * the contents of (the filesystem + the journal) must be consistent and
3664 * restartable. It's pretty simple, really: bottom up, right to left (although
3665 * left-to-right works OK too).
3666 *
3667 * Note that at recovery time, journal replay occurs *before* the restart of
3668 * truncate against the orphan inode list.
3669 *
3670 * The committed inode has the new, desired i_size (which is the same as
Mingming Cao617ba132006-10-11 01:20:53 -07003671 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003672 * that this inode's truncate did not complete and it will again call
Mingming Cao617ba132006-10-11 01:20:53 -07003673 * ext4_truncate() to have another go. So there will be instantiated blocks
3674 * to the right of the truncation point in a crashed ext4 filesystem. But
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003675 * that's fine - as long as they are linked from the inode, the post-crash
Mingming Cao617ba132006-10-11 01:20:53 -07003676 * ext4_truncate() run will find them and release them.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003677 */
Mingming Cao617ba132006-10-11 01:20:53 -07003678void ext4_truncate(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003679{
3680 handle_t *handle;
Mingming Cao617ba132006-10-11 01:20:53 -07003681 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003682 __le32 *i_data = ei->i_data;
Mingming Cao617ba132006-10-11 01:20:53 -07003683 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003684 struct address_space *mapping = inode->i_mapping;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003685 ext4_lblk_t offsets[4];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003686 Indirect chain[4];
3687 Indirect *partial;
3688 __le32 nr = 0;
3689 int n;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003690 ext4_lblk_t last_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003691 unsigned blocksize = inode->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003692
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003693 if (!ext4_can_truncate(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003694 return;
3695
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05003696 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
Jan Karacf108bc2008-07-11 19:27:31 -04003697 ext4_ext_truncate(inode);
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05003698 return;
3699 }
Alex Tomasa86c6182006-10-11 01:21:03 -07003700
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003701 handle = start_transaction(inode);
Jan Karacf108bc2008-07-11 19:27:31 -04003702 if (IS_ERR(handle))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003703 return; /* AKPM: return what? */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003704
3705 last_block = (inode->i_size + blocksize-1)
Mingming Cao617ba132006-10-11 01:20:53 -07003706 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003707
Jan Karacf108bc2008-07-11 19:27:31 -04003708 if (inode->i_size & (blocksize - 1))
3709 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3710 goto out_stop;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003711
Mingming Cao617ba132006-10-11 01:20:53 -07003712 n = ext4_block_to_path(inode, last_block, offsets, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003713 if (n == 0)
3714 goto out_stop; /* error */
3715
3716 /*
3717 * OK. This truncate is going to happen. We add the inode to the
3718 * orphan list, so that if this truncate spans multiple transactions,
3719 * and we crash, we will resume the truncate when the filesystem
3720 * recovers. It also marks the inode dirty, to catch the new size.
3721 *
3722 * Implication: the file must always be in a sane, consistent
3723 * truncatable state while each transaction commits.
3724 */
Mingming Cao617ba132006-10-11 01:20:53 -07003725 if (ext4_orphan_add(handle, inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003726 goto out_stop;
3727
3728 /*
Mingming Cao632eaea2008-07-11 19:27:31 -04003729 * From here we block out all ext4_get_block() callers who want to
3730 * modify the block allocation tree.
3731 */
3732 down_write(&ei->i_data_sem);
Theodore Ts'ob4df2032008-08-13 21:44:34 -04003733
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04003734 ext4_discard_preallocations(inode);
Theodore Ts'ob4df2032008-08-13 21:44:34 -04003735
Mingming Cao632eaea2008-07-11 19:27:31 -04003736 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003737 * The orphan list entry will now protect us from any crash which
3738 * occurs before the truncate completes, so it is now safe to propagate
3739 * the new, shorter inode size (held for now in i_size) into the
3740 * on-disk inode. We do this via i_disksize, which is the value which
Mingming Cao617ba132006-10-11 01:20:53 -07003741 * ext4 *really* writes onto the disk inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003742 */
3743 ei->i_disksize = inode->i_size;
3744
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003745 if (n == 1) { /* direct blocks */
Mingming Cao617ba132006-10-11 01:20:53 -07003746 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
3747 i_data + EXT4_NDIR_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003748 goto do_indirects;
3749 }
3750
Mingming Cao617ba132006-10-11 01:20:53 -07003751 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003752 /* Kill the top of shared branch (not detached) */
3753 if (nr) {
3754 if (partial == chain) {
3755 /* Shared branch grows from the inode */
Mingming Cao617ba132006-10-11 01:20:53 -07003756 ext4_free_branches(handle, inode, NULL,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003757 &nr, &nr+1, (chain+n-1) - partial);
3758 *partial->p = 0;
3759 /*
3760 * We mark the inode dirty prior to restart,
3761 * and prior to stop. No need for it here.
3762 */
3763 } else {
3764 /* Shared branch grows from an indirect block */
3765 BUFFER_TRACE(partial->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07003766 ext4_free_branches(handle, inode, partial->bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003767 partial->p,
3768 partial->p+1, (chain+n-1) - partial);
3769 }
3770 }
3771 /* Clear the ends of indirect blocks on the shared branch */
3772 while (partial > chain) {
Mingming Cao617ba132006-10-11 01:20:53 -07003773 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003774 (__le32*)partial->bh->b_data+addr_per_block,
3775 (chain+n-1) - partial);
3776 BUFFER_TRACE(partial->bh, "call brelse");
3777 brelse (partial->bh);
3778 partial--;
3779 }
3780do_indirects:
3781 /* Kill the remaining (whole) subtrees */
3782 switch (offsets[0]) {
3783 default:
Mingming Cao617ba132006-10-11 01:20:53 -07003784 nr = i_data[EXT4_IND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003785 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07003786 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
3787 i_data[EXT4_IND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003788 }
Mingming Cao617ba132006-10-11 01:20:53 -07003789 case EXT4_IND_BLOCK:
3790 nr = i_data[EXT4_DIND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003791 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07003792 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
3793 i_data[EXT4_DIND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003794 }
Mingming Cao617ba132006-10-11 01:20:53 -07003795 case EXT4_DIND_BLOCK:
3796 nr = i_data[EXT4_TIND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003797 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07003798 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
3799 i_data[EXT4_TIND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003800 }
Mingming Cao617ba132006-10-11 01:20:53 -07003801 case EXT4_TIND_BLOCK:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003802 ;
3803 }
3804
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05003805 up_write(&ei->i_data_sem);
Kalpak Shahef7f3832007-07-18 09:15:20 -04003806 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07003807 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003808
3809 /*
3810 * In a multi-transaction truncate, we only make the final transaction
3811 * synchronous
3812 */
3813 if (IS_SYNC(inode))
3814 handle->h_sync = 1;
3815out_stop:
3816 /*
3817 * If this was a simple ftruncate(), and the file will remain alive
3818 * then we need to clear up the orphan record which we created above.
3819 * However, if this was a real unlink then we were called by
Mingming Cao617ba132006-10-11 01:20:53 -07003820 * ext4_delete_inode(), and we allow that function to clean up the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003821 * orphan info for us.
3822 */
3823 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07003824 ext4_orphan_del(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003825
Mingming Cao617ba132006-10-11 01:20:53 -07003826 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003827}
3828
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003829/*
Mingming Cao617ba132006-10-11 01:20:53 -07003830 * ext4_get_inode_loc returns with an extra refcount against the inode's
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003831 * underlying buffer_head on success. If 'in_mem' is true, we have all
3832 * data in memory that is needed to recreate the on-disk version of this
3833 * inode.
3834 */
Mingming Cao617ba132006-10-11 01:20:53 -07003835static int __ext4_get_inode_loc(struct inode *inode,
3836 struct ext4_iloc *iloc, int in_mem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003837{
Theodore Ts'o240799c2008-10-09 23:53:47 -04003838 struct ext4_group_desc *gdp;
3839 struct buffer_head *bh;
3840 struct super_block *sb = inode->i_sb;
3841 ext4_fsblk_t block;
3842 int inodes_per_block, inode_offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003843
Theodore Ts'o240799c2008-10-09 23:53:47 -04003844 iloc->bh = 0;
3845 if (!ext4_valid_inum(sb, inode->i_ino))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003846 return -EIO;
3847
Theodore Ts'o240799c2008-10-09 23:53:47 -04003848 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3849 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3850 if (!gdp)
3851 return -EIO;
3852
3853 /*
3854 * Figure out the offset within the block group inode table
3855 */
3856 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
3857 inode_offset = ((inode->i_ino - 1) %
3858 EXT4_INODES_PER_GROUP(sb));
3859 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3860 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3861
3862 bh = sb_getblk(sb, block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003863 if (!bh) {
Theodore Ts'o240799c2008-10-09 23:53:47 -04003864 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
3865 "inode block - inode=%lu, block=%llu",
3866 inode->i_ino, block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003867 return -EIO;
3868 }
3869 if (!buffer_uptodate(bh)) {
3870 lock_buffer(bh);
Hidehiro Kawai9c83a922008-07-26 16:39:26 -04003871
3872 /*
3873 * If the buffer has the write error flag, we have failed
3874 * to write out another inode in the same block. In this
3875 * case, we don't have to read the block because we may
3876 * read the old inode data successfully.
3877 */
3878 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3879 set_buffer_uptodate(bh);
3880
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003881 if (buffer_uptodate(bh)) {
3882 /* someone brought it uptodate while we waited */
3883 unlock_buffer(bh);
3884 goto has_buffer;
3885 }
3886
3887 /*
3888 * If we have all information of the inode in memory and this
3889 * is the only valid inode in the block, we need not read the
3890 * block.
3891 */
3892 if (in_mem) {
3893 struct buffer_head *bitmap_bh;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003894 int i, start;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003895
Theodore Ts'o240799c2008-10-09 23:53:47 -04003896 start = inode_offset & ~(inodes_per_block - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003897
3898 /* Is the inode bitmap in cache? */
Theodore Ts'o240799c2008-10-09 23:53:47 -04003899 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003900 if (!bitmap_bh)
3901 goto make_io;
3902
3903 /*
3904 * If the inode bitmap isn't in cache then the
3905 * optimisation may end up performing two reads instead
3906 * of one, so skip it.
3907 */
3908 if (!buffer_uptodate(bitmap_bh)) {
3909 brelse(bitmap_bh);
3910 goto make_io;
3911 }
Theodore Ts'o240799c2008-10-09 23:53:47 -04003912 for (i = start; i < start + inodes_per_block; i++) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003913 if (i == inode_offset)
3914 continue;
Mingming Cao617ba132006-10-11 01:20:53 -07003915 if (ext4_test_bit(i, bitmap_bh->b_data))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003916 break;
3917 }
3918 brelse(bitmap_bh);
Theodore Ts'o240799c2008-10-09 23:53:47 -04003919 if (i == start + inodes_per_block) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003920 /* all other inodes are free, so skip I/O */
3921 memset(bh->b_data, 0, bh->b_size);
3922 set_buffer_uptodate(bh);
3923 unlock_buffer(bh);
3924 goto has_buffer;
3925 }
3926 }
3927
3928make_io:
3929 /*
Theodore Ts'o240799c2008-10-09 23:53:47 -04003930 * If we need to do any I/O, try to pre-readahead extra
3931 * blocks from the inode table.
3932 */
3933 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3934 ext4_fsblk_t b, end, table;
3935 unsigned num;
3936
3937 table = ext4_inode_table(sb, gdp);
3938 /* Make sure s_inode_readahead_blks is a power of 2 */
3939 while (EXT4_SB(sb)->s_inode_readahead_blks &
3940 (EXT4_SB(sb)->s_inode_readahead_blks-1))
3941 EXT4_SB(sb)->s_inode_readahead_blks =
3942 (EXT4_SB(sb)->s_inode_readahead_blks &
3943 (EXT4_SB(sb)->s_inode_readahead_blks-1));
3944 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3945 if (table > b)
3946 b = table;
3947 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3948 num = EXT4_INODES_PER_GROUP(sb);
3949 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3950 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3951 num -= le16_to_cpu(gdp->bg_itable_unused);
3952 table += num / inodes_per_block;
3953 if (end > table)
3954 end = table;
3955 while (b <= end)
3956 sb_breadahead(sb, b++);
3957 }
3958
3959 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003960 * There are other valid inodes in the buffer, this inode
3961 * has in-inode xattrs, or we don't have this inode in memory.
3962 * Read the block from disk.
3963 */
3964 get_bh(bh);
3965 bh->b_end_io = end_buffer_read_sync;
3966 submit_bh(READ_META, bh);
3967 wait_on_buffer(bh);
3968 if (!buffer_uptodate(bh)) {
Theodore Ts'o240799c2008-10-09 23:53:47 -04003969 ext4_error(sb, __func__,
3970 "unable to read inode block - inode=%lu, "
3971 "block=%llu", inode->i_ino, block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003972 brelse(bh);
3973 return -EIO;
3974 }
3975 }
3976has_buffer:
3977 iloc->bh = bh;
3978 return 0;
3979}
3980
Mingming Cao617ba132006-10-11 01:20:53 -07003981int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003982{
3983 /* We have all inode data except xattrs in memory here. */
Mingming Cao617ba132006-10-11 01:20:53 -07003984 return __ext4_get_inode_loc(inode, iloc,
3985 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003986}
3987
Mingming Cao617ba132006-10-11 01:20:53 -07003988void ext4_set_inode_flags(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003989{
Mingming Cao617ba132006-10-11 01:20:53 -07003990 unsigned int flags = EXT4_I(inode)->i_flags;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003991
3992 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
Mingming Cao617ba132006-10-11 01:20:53 -07003993 if (flags & EXT4_SYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003994 inode->i_flags |= S_SYNC;
Mingming Cao617ba132006-10-11 01:20:53 -07003995 if (flags & EXT4_APPEND_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003996 inode->i_flags |= S_APPEND;
Mingming Cao617ba132006-10-11 01:20:53 -07003997 if (flags & EXT4_IMMUTABLE_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003998 inode->i_flags |= S_IMMUTABLE;
Mingming Cao617ba132006-10-11 01:20:53 -07003999 if (flags & EXT4_NOATIME_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004000 inode->i_flags |= S_NOATIME;
Mingming Cao617ba132006-10-11 01:20:53 -07004001 if (flags & EXT4_DIRSYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004002 inode->i_flags |= S_DIRSYNC;
4003}
4004
Jan Karaff9ddf72007-07-18 09:24:20 -04004005/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4006void ext4_get_inode_flags(struct ext4_inode_info *ei)
4007{
4008 unsigned int flags = ei->vfs_inode.i_flags;
4009
4010 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4011 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4012 if (flags & S_SYNC)
4013 ei->i_flags |= EXT4_SYNC_FL;
4014 if (flags & S_APPEND)
4015 ei->i_flags |= EXT4_APPEND_FL;
4016 if (flags & S_IMMUTABLE)
4017 ei->i_flags |= EXT4_IMMUTABLE_FL;
4018 if (flags & S_NOATIME)
4019 ei->i_flags |= EXT4_NOATIME_FL;
4020 if (flags & S_DIRSYNC)
4021 ei->i_flags |= EXT4_DIRSYNC_FL;
4022}
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004023static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4024 struct ext4_inode_info *ei)
4025{
4026 blkcnt_t i_blocks ;
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004027 struct inode *inode = &(ei->vfs_inode);
4028 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004029
4030 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4031 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4032 /* we are using combined 48 bit field */
4033 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4034 le32_to_cpu(raw_inode->i_blocks_lo);
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004035 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4036 /* i_blocks represent file system block size */
4037 return i_blocks << (inode->i_blkbits - 9);
4038 } else {
4039 return i_blocks;
4040 }
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004041 } else {
4042 return le32_to_cpu(raw_inode->i_blocks_lo);
4043 }
4044}
Jan Karaff9ddf72007-07-18 09:24:20 -04004045
David Howells1d1fe1e2008-02-07 00:15:37 -08004046struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004047{
Mingming Cao617ba132006-10-11 01:20:53 -07004048 struct ext4_iloc iloc;
4049 struct ext4_inode *raw_inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08004050 struct ext4_inode_info *ei;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004051 struct buffer_head *bh;
David Howells1d1fe1e2008-02-07 00:15:37 -08004052 struct inode *inode;
4053 long ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004054 int block;
4055
David Howells1d1fe1e2008-02-07 00:15:37 -08004056 inode = iget_locked(sb, ino);
4057 if (!inode)
4058 return ERR_PTR(-ENOMEM);
4059 if (!(inode->i_state & I_NEW))
4060 return inode;
4061
4062 ei = EXT4_I(inode);
Theodore Ts'o03010a32008-10-10 20:02:48 -04004063#ifdef CONFIG_EXT4_FS_POSIX_ACL
Mingming Cao617ba132006-10-11 01:20:53 -07004064 ei->i_acl = EXT4_ACL_NOT_CACHED;
4065 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004066#endif
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004067
David Howells1d1fe1e2008-02-07 00:15:37 -08004068 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4069 if (ret < 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004070 goto bad_inode;
4071 bh = iloc.bh;
Mingming Cao617ba132006-10-11 01:20:53 -07004072 raw_inode = ext4_raw_inode(&iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004073 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4074 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4075 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004076 if (!(test_opt(inode->i_sb, NO_UID32))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004077 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4078 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4079 }
4080 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004081
4082 ei->i_state = 0;
4083 ei->i_dir_start_lookup = 0;
4084 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4085 /* We now have enough fields to check if the inode was active or not.
4086 * This is needed because nfsd might try to access dead inodes
4087 * the test is that same one that e2fsck uses
4088 * NeilBrown 1999oct15
4089 */
4090 if (inode->i_nlink == 0) {
4091 if (inode->i_mode == 0 ||
Mingming Cao617ba132006-10-11 01:20:53 -07004092 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004093 /* this inode is deleted */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004094 brelse(bh);
David Howells1d1fe1e2008-02-07 00:15:37 -08004095 ret = -ESTALE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004096 goto bad_inode;
4097 }
4098 /* The only unlinked inodes we let through here have
4099 * valid i_mode and are being read by the orphan
4100 * recovery code: that's fine, we're about to complete
4101 * the process of deleting those. */
4102 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004103 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004104 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05004105 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
Mingming Cao9b8f1f02006-10-11 01:21:13 -07004106 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004107 cpu_to_le32(EXT4_OS_HURD)) {
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07004108 ei->i_file_acl |=
4109 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004110 }
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004111 inode->i_size = ext4_isize(raw_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004112 ei->i_disksize = inode->i_size;
4113 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4114 ei->i_block_group = iloc.block_group;
4115 /*
4116 * NOTE! The in-memory inode i_data array is in little-endian order
4117 * even on big-endian machines: we do NOT byteswap the block numbers!
4118 */
Mingming Cao617ba132006-10-11 01:20:53 -07004119 for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004120 ei->i_data[block] = raw_inode->i_block[block];
4121 INIT_LIST_HEAD(&ei->i_orphan);
4122
Eric Sandeen0040d982008-02-05 22:36:43 -05004123 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004124 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
Mingming Cao617ba132006-10-11 01:20:53 -07004125 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
Kirill Korotaeve5d28612007-06-23 17:16:51 -07004126 EXT4_INODE_SIZE(inode->i_sb)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004127 brelse(bh);
David Howells1d1fe1e2008-02-07 00:15:37 -08004128 ret = -EIO;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004129 goto bad_inode;
Kirill Korotaeve5d28612007-06-23 17:16:51 -07004130 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004131 if (ei->i_extra_isize == 0) {
4132 /* The extra space is currently unused. Use it. */
Mingming Cao617ba132006-10-11 01:20:53 -07004133 ei->i_extra_isize = sizeof(struct ext4_inode) -
4134 EXT4_GOOD_OLD_INODE_SIZE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004135 } else {
4136 __le32 *magic = (void *)raw_inode +
Mingming Cao617ba132006-10-11 01:20:53 -07004137 EXT4_GOOD_OLD_INODE_SIZE +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004138 ei->i_extra_isize;
Mingming Cao617ba132006-10-11 01:20:53 -07004139 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4140 ei->i_state |= EXT4_STATE_XATTR;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004141 }
4142 } else
4143 ei->i_extra_isize = 0;
4144
Kalpak Shahef7f3832007-07-18 09:15:20 -04004145 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4146 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4147 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4148 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4149
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004150 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4151 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4152 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4153 inode->i_version |=
4154 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4155 }
4156
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004157 if (S_ISREG(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004158 inode->i_op = &ext4_file_inode_operations;
4159 inode->i_fop = &ext4_file_operations;
4160 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004161 } else if (S_ISDIR(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004162 inode->i_op = &ext4_dir_inode_operations;
4163 inode->i_fop = &ext4_dir_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004164 } else if (S_ISLNK(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004165 if (ext4_inode_is_fast_symlink(inode))
4166 inode->i_op = &ext4_fast_symlink_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004167 else {
Mingming Cao617ba132006-10-11 01:20:53 -07004168 inode->i_op = &ext4_symlink_inode_operations;
4169 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004170 }
4171 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07004172 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004173 if (raw_inode->i_block[0])
4174 init_special_inode(inode, inode->i_mode,
4175 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4176 else
4177 init_special_inode(inode, inode->i_mode,
4178 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4179 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004180 brelse(iloc.bh);
Mingming Cao617ba132006-10-11 01:20:53 -07004181 ext4_set_inode_flags(inode);
David Howells1d1fe1e2008-02-07 00:15:37 -08004182 unlock_new_inode(inode);
4183 return inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004184
4185bad_inode:
David Howells1d1fe1e2008-02-07 00:15:37 -08004186 iget_failed(inode);
4187 return ERR_PTR(ret);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004188}
4189
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004190static int ext4_inode_blocks_set(handle_t *handle,
4191 struct ext4_inode *raw_inode,
4192 struct ext4_inode_info *ei)
4193{
4194 struct inode *inode = &(ei->vfs_inode);
4195 u64 i_blocks = inode->i_blocks;
4196 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004197
4198 if (i_blocks <= ~0U) {
4199 /*
4200 * i_blocks can be represnted in a 32 bit variable
4201 * as multiple of 512 bytes
4202 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004203 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004204 raw_inode->i_blocks_high = 0;
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004205 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
Theodore Ts'of287a1a2008-10-16 22:50:48 -04004206 return 0;
4207 }
4208 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4209 return -EFBIG;
4210
4211 if (i_blocks <= 0xffffffffffffULL) {
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004212 /*
4213 * i_blocks can be represented in a 48 bit variable
4214 * as multiple of 512 bytes
4215 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004216 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004217 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004218 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004219 } else {
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004220 ei->i_flags |= EXT4_HUGE_FILE_FL;
4221 /* i_block is stored in file system block size */
4222 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4223 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4224 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004225 }
Theodore Ts'of287a1a2008-10-16 22:50:48 -04004226 return 0;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004227}
4228
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004229/*
4230 * Post the struct inode info into an on-disk inode location in the
4231 * buffer-cache. This gobbles the caller's reference to the
4232 * buffer_head in the inode location struct.
4233 *
4234 * The caller must have write access to iloc->bh.
4235 */
Mingming Cao617ba132006-10-11 01:20:53 -07004236static int ext4_do_update_inode(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004237 struct inode *inode,
Mingming Cao617ba132006-10-11 01:20:53 -07004238 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004239{
Mingming Cao617ba132006-10-11 01:20:53 -07004240 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4241 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004242 struct buffer_head *bh = iloc->bh;
4243 int err = 0, rc, block;
4244
4245 /* For fields not not tracking in the in-memory inode,
4246 * initialise them to zero for new inodes. */
Mingming Cao617ba132006-10-11 01:20:53 -07004247 if (ei->i_state & EXT4_STATE_NEW)
4248 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004249
Jan Karaff9ddf72007-07-18 09:24:20 -04004250 ext4_get_inode_flags(ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004251 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004252 if (!(test_opt(inode->i_sb, NO_UID32))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004253 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4254 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4255/*
4256 * Fix up interoperability with old kernels. Otherwise, old inodes get
4257 * re-used with the upper 16 bits of the uid/gid intact
4258 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004259 if (!ei->i_dtime) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004260 raw_inode->i_uid_high =
4261 cpu_to_le16(high_16_bits(inode->i_uid));
4262 raw_inode->i_gid_high =
4263 cpu_to_le16(high_16_bits(inode->i_gid));
4264 } else {
4265 raw_inode->i_uid_high = 0;
4266 raw_inode->i_gid_high = 0;
4267 }
4268 } else {
4269 raw_inode->i_uid_low =
4270 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4271 raw_inode->i_gid_low =
4272 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4273 raw_inode->i_uid_high = 0;
4274 raw_inode->i_gid_high = 0;
4275 }
4276 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
Kalpak Shahef7f3832007-07-18 09:15:20 -04004277
4278 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4279 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4280 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4281 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4282
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004283 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4284 goto out_brelse;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004285 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -04004286 /* clear the migrate flag in the raw_inode */
4287 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
Mingming Cao9b8f1f02006-10-11 01:21:13 -07004288 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4289 cpu_to_le32(EXT4_OS_HURD))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07004290 raw_inode->i_file_acl_high =
4291 cpu_to_le16(ei->i_file_acl >> 32);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05004292 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004293 ext4_isize_set(raw_inode, ei->i_disksize);
4294 if (ei->i_disksize > 0x7fffffffULL) {
4295 struct super_block *sb = inode->i_sb;
4296 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4297 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4298 EXT4_SB(sb)->s_es->s_rev_level ==
4299 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4300 /* If this is the first large file
4301 * created, add a flag to the superblock.
4302 */
4303 err = ext4_journal_get_write_access(handle,
4304 EXT4_SB(sb)->s_sbh);
4305 if (err)
4306 goto out_brelse;
4307 ext4_update_dynamic_rev(sb);
4308 EXT4_SET_RO_COMPAT_FEATURE(sb,
Mingming Cao617ba132006-10-11 01:20:53 -07004309 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004310 sb->s_dirt = 1;
4311 handle->h_sync = 1;
4312 err = ext4_journal_dirty_metadata(handle,
4313 EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004314 }
4315 }
4316 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4317 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4318 if (old_valid_dev(inode->i_rdev)) {
4319 raw_inode->i_block[0] =
4320 cpu_to_le32(old_encode_dev(inode->i_rdev));
4321 raw_inode->i_block[1] = 0;
4322 } else {
4323 raw_inode->i_block[0] = 0;
4324 raw_inode->i_block[1] =
4325 cpu_to_le32(new_encode_dev(inode->i_rdev));
4326 raw_inode->i_block[2] = 0;
4327 }
Mingming Cao617ba132006-10-11 01:20:53 -07004328 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004329 raw_inode->i_block[block] = ei->i_data[block];
4330
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004331 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4332 if (ei->i_extra_isize) {
4333 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4334 raw_inode->i_version_hi =
4335 cpu_to_le32(inode->i_version >> 32);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004336 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004337 }
4338
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004339
Mingming Cao617ba132006-10-11 01:20:53 -07004340 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
4341 rc = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004342 if (!err)
4343 err = rc;
Mingming Cao617ba132006-10-11 01:20:53 -07004344 ei->i_state &= ~EXT4_STATE_NEW;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004345
4346out_brelse:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004347 brelse(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07004348 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004349 return err;
4350}
4351
4352/*
Mingming Cao617ba132006-10-11 01:20:53 -07004353 * ext4_write_inode()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004354 *
4355 * We are called from a few places:
4356 *
4357 * - Within generic_file_write() for O_SYNC files.
4358 * Here, there will be no transaction running. We wait for any running
4359 * trasnaction to commit.
4360 *
4361 * - Within sys_sync(), kupdate and such.
4362 * We wait on commit, if tol to.
4363 *
4364 * - Within prune_icache() (PF_MEMALLOC == true)
4365 * Here we simply return. We can't afford to block kswapd on the
4366 * journal commit.
4367 *
4368 * In all cases it is actually safe for us to return without doing anything,
4369 * because the inode has been copied into a raw inode buffer in
Mingming Cao617ba132006-10-11 01:20:53 -07004370 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004371 * knfsd.
4372 *
4373 * Note that we are absolutely dependent upon all inode dirtiers doing the
4374 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4375 * which we are interested.
4376 *
4377 * It would be a bug for them to not do this. The code:
4378 *
4379 * mark_inode_dirty(inode)
4380 * stuff();
4381 * inode->i_size = expr;
4382 *
4383 * is in error because a kswapd-driven write_inode() could occur while
4384 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4385 * will no longer be on the superblock's dirty inode list.
4386 */
Mingming Cao617ba132006-10-11 01:20:53 -07004387int ext4_write_inode(struct inode *inode, int wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004388{
4389 if (current->flags & PF_MEMALLOC)
4390 return 0;
4391
Mingming Cao617ba132006-10-11 01:20:53 -07004392 if (ext4_journal_current_handle()) {
Mingming Caob38bd332007-07-19 01:48:35 -07004393 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004394 dump_stack();
4395 return -EIO;
4396 }
4397
4398 if (!wait)
4399 return 0;
4400
Mingming Cao617ba132006-10-11 01:20:53 -07004401 return ext4_force_commit(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004402}
4403
4404/*
Mingming Cao617ba132006-10-11 01:20:53 -07004405 * ext4_setattr()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004406 *
4407 * Called from notify_change.
4408 *
4409 * We want to trap VFS attempts to truncate the file as soon as
4410 * possible. In particular, we want to make sure that when the VFS
4411 * shrinks i_size, we put the inode on the orphan list and modify
4412 * i_disksize immediately, so that during the subsequent flushing of
4413 * dirty pages and freeing of disk blocks, we can guarantee that any
4414 * commit will leave the blocks being flushed in an unused state on
4415 * disk. (On recovery, the inode will get truncated and the blocks will
4416 * be freed, so we have a strong guarantee that no future commit will
4417 * leave these blocks visible to the user.)
4418 *
Jan Kara678aaf42008-07-11 19:27:31 -04004419 * Another thing we have to assure is that if we are in ordered mode
4420 * and inode is still attached to the committing transaction, we must
4421 * we start writeout of all the dirty pages which are being truncated.
4422 * This way we are sure that all the data written in the previous
4423 * transaction are already on disk (truncate waits for pages under
4424 * writeback).
4425 *
4426 * Called with inode->i_mutex down.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004427 */
Mingming Cao617ba132006-10-11 01:20:53 -07004428int ext4_setattr(struct dentry *dentry, struct iattr *attr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004429{
4430 struct inode *inode = dentry->d_inode;
4431 int error, rc = 0;
4432 const unsigned int ia_valid = attr->ia_valid;
4433
4434 error = inode_change_ok(inode, attr);
4435 if (error)
4436 return error;
4437
4438 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4439 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4440 handle_t *handle;
4441
4442 /* (user+group)*(old+new) structure, inode write (sb,
4443 * inode block, ? - but truncate inode update has it) */
Mingming Cao617ba132006-10-11 01:20:53 -07004444 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4445 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004446 if (IS_ERR(handle)) {
4447 error = PTR_ERR(handle);
4448 goto err_out;
4449 }
4450 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
4451 if (error) {
Mingming Cao617ba132006-10-11 01:20:53 -07004452 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004453 return error;
4454 }
4455 /* Update corresponding info in inode so that everything is in
4456 * one transaction */
4457 if (attr->ia_valid & ATTR_UID)
4458 inode->i_uid = attr->ia_uid;
4459 if (attr->ia_valid & ATTR_GID)
4460 inode->i_gid = attr->ia_gid;
Mingming Cao617ba132006-10-11 01:20:53 -07004461 error = ext4_mark_inode_dirty(handle, inode);
4462 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004463 }
4464
Eric Sandeene2b46572008-01-28 23:58:27 -05004465 if (attr->ia_valid & ATTR_SIZE) {
4466 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4467 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4468
4469 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4470 error = -EFBIG;
4471 goto err_out;
4472 }
4473 }
4474 }
4475
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004476 if (S_ISREG(inode->i_mode) &&
4477 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4478 handle_t *handle;
4479
Mingming Cao617ba132006-10-11 01:20:53 -07004480 handle = ext4_journal_start(inode, 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004481 if (IS_ERR(handle)) {
4482 error = PTR_ERR(handle);
4483 goto err_out;
4484 }
4485
Mingming Cao617ba132006-10-11 01:20:53 -07004486 error = ext4_orphan_add(handle, inode);
4487 EXT4_I(inode)->i_disksize = attr->ia_size;
4488 rc = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004489 if (!error)
4490 error = rc;
Mingming Cao617ba132006-10-11 01:20:53 -07004491 ext4_journal_stop(handle);
Jan Kara678aaf42008-07-11 19:27:31 -04004492
4493 if (ext4_should_order_data(inode)) {
4494 error = ext4_begin_ordered_truncate(inode,
4495 attr->ia_size);
4496 if (error) {
4497 /* Do as much error cleanup as possible */
4498 handle = ext4_journal_start(inode, 3);
4499 if (IS_ERR(handle)) {
4500 ext4_orphan_del(NULL, inode);
4501 goto err_out;
4502 }
4503 ext4_orphan_del(handle, inode);
4504 ext4_journal_stop(handle);
4505 goto err_out;
4506 }
4507 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004508 }
4509
4510 rc = inode_setattr(inode, attr);
4511
Mingming Cao617ba132006-10-11 01:20:53 -07004512 /* If inode_setattr's call to ext4_truncate failed to get a
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004513 * transaction handle at all, we need to clean up the in-core
4514 * orphan list manually. */
4515 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07004516 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004517
4518 if (!rc && (ia_valid & ATTR_MODE))
Mingming Cao617ba132006-10-11 01:20:53 -07004519 rc = ext4_acl_chmod(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004520
4521err_out:
Mingming Cao617ba132006-10-11 01:20:53 -07004522 ext4_std_error(inode->i_sb, error);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004523 if (!error)
4524 error = rc;
4525 return error;
4526}
4527
Mingming Cao3e3398a2008-07-11 19:27:31 -04004528int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4529 struct kstat *stat)
4530{
4531 struct inode *inode;
4532 unsigned long delalloc_blocks;
4533
4534 inode = dentry->d_inode;
4535 generic_fillattr(inode, stat);
4536
4537 /*
4538 * We can't update i_blocks if the block allocation is delayed
4539 * otherwise in the case of system crash before the real block
4540 * allocation is done, we will have i_blocks inconsistent with
4541 * on-disk file blocks.
4542 * We always keep i_blocks updated together with real
4543 * allocation. But to not confuse with user, stat
4544 * will return the blocks that include the delayed allocation
4545 * blocks for this file.
4546 */
4547 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4548 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4549 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4550
4551 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4552 return 0;
4553}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004554
Mingming Caoa02908f2008-08-19 22:16:07 -04004555static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4556 int chunk)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004557{
Mingming Caoa02908f2008-08-19 22:16:07 -04004558 int indirects;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004559
Mingming Caoa02908f2008-08-19 22:16:07 -04004560 /* if nrblocks are contiguous */
4561 if (chunk) {
4562 /*
4563 * With N contiguous data blocks, it need at most
4564 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4565 * 2 dindirect blocks
4566 * 1 tindirect block
4567 */
4568 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4569 return indirects + 3;
4570 }
4571 /*
4572 * if nrblocks are not contiguous, worse case, each block touch
4573 * a indirect block, and each indirect block touch a double indirect
4574 * block, plus a triple indirect block
4575 */
4576 indirects = nrblocks * 2 + 1;
4577 return indirects;
4578}
Alex Tomasa86c6182006-10-11 01:21:03 -07004579
Mingming Caoa02908f2008-08-19 22:16:07 -04004580static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4581{
4582 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4583 return ext4_indirect_trans_blocks(inode, nrblocks, 0);
4584 return ext4_ext_index_trans_blocks(inode, nrblocks, 0);
4585}
4586/*
4587 * Account for index blocks, block groups bitmaps and block group
4588 * descriptor blocks if modify datablocks and index blocks
4589 * worse case, the indexs blocks spread over different block groups
4590 *
4591 * If datablocks are discontiguous, they are possible to spread over
4592 * different block groups too. If they are contiugous, with flexbg,
4593 * they could still across block group boundary.
4594 *
4595 * Also account for superblock, inode, quota and xattr blocks
4596 */
4597int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4598{
4599 int groups, gdpblocks;
4600 int idxblocks;
4601 int ret = 0;
4602
4603 /*
4604 * How many index blocks need to touch to modify nrblocks?
4605 * The "Chunk" flag indicating whether the nrblocks is
4606 * physically contiguous on disk
4607 *
4608 * For Direct IO and fallocate, they calls get_block to allocate
4609 * one single extent at a time, so they could set the "Chunk" flag
4610 */
4611 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4612
4613 ret = idxblocks;
4614
4615 /*
4616 * Now let's see how many group bitmaps and group descriptors need
4617 * to account
4618 */
4619 groups = idxblocks;
4620 if (chunk)
4621 groups += 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004622 else
Mingming Caoa02908f2008-08-19 22:16:07 -04004623 groups += nrblocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004624
Mingming Caoa02908f2008-08-19 22:16:07 -04004625 gdpblocks = groups;
4626 if (groups > EXT4_SB(inode->i_sb)->s_groups_count)
4627 groups = EXT4_SB(inode->i_sb)->s_groups_count;
4628 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4629 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4630
4631 /* bitmaps and block group descriptor blocks */
4632 ret += groups + gdpblocks;
4633
4634 /* Blocks for super block, inode, quota and xattr blocks */
4635 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004636
4637 return ret;
4638}
4639
4640/*
Mingming Caoa02908f2008-08-19 22:16:07 -04004641 * Calulate the total number of credits to reserve to fit
Mingming Caof3bd1f32008-08-19 22:16:03 -04004642 * the modification of a single pages into a single transaction,
4643 * which may include multiple chunks of block allocations.
Mingming Caoa02908f2008-08-19 22:16:07 -04004644 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04004645 * This could be called via ext4_write_begin()
Mingming Caoa02908f2008-08-19 22:16:07 -04004646 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04004647 * We need to consider the worse case, when
Mingming Caoa02908f2008-08-19 22:16:07 -04004648 * one new block per extent.
Mingming Caoa02908f2008-08-19 22:16:07 -04004649 */
4650int ext4_writepage_trans_blocks(struct inode *inode)
4651{
4652 int bpp = ext4_journal_blocks_per_page(inode);
4653 int ret;
4654
4655 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4656
4657 /* Account for data blocks for journalled mode */
4658 if (ext4_should_journal_data(inode))
4659 ret += bpp;
4660 return ret;
4661}
Mingming Caof3bd1f32008-08-19 22:16:03 -04004662
4663/*
4664 * Calculate the journal credits for a chunk of data modification.
4665 *
4666 * This is called from DIO, fallocate or whoever calling
4667 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4668 *
4669 * journal buffers for data blocks are not included here, as DIO
4670 * and fallocate do no need to journal data buffers.
4671 */
4672int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4673{
4674 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4675}
4676
Mingming Caoa02908f2008-08-19 22:16:07 -04004677/*
Mingming Cao617ba132006-10-11 01:20:53 -07004678 * The caller must have previously called ext4_reserve_inode_write().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004679 * Give this, we know that the caller already has write access to iloc->bh.
4680 */
Mingming Cao617ba132006-10-11 01:20:53 -07004681int ext4_mark_iloc_dirty(handle_t *handle,
4682 struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004683{
4684 int err = 0;
4685
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004686 if (test_opt(inode->i_sb, I_VERSION))
4687 inode_inc_iversion(inode);
4688
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004689 /* the do_update_inode consumes one bh->b_count */
4690 get_bh(iloc->bh);
4691
Mingming Caodab291a2006-10-11 01:21:01 -07004692 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
Mingming Cao617ba132006-10-11 01:20:53 -07004693 err = ext4_do_update_inode(handle, inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004694 put_bh(iloc->bh);
4695 return err;
4696}
4697
4698/*
4699 * On success, We end up with an outstanding reference count against
4700 * iloc->bh. This _must_ be cleaned up later.
4701 */
4702
4703int
Mingming Cao617ba132006-10-11 01:20:53 -07004704ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4705 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004706{
4707 int err = 0;
4708 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07004709 err = ext4_get_inode_loc(inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004710 if (!err) {
4711 BUFFER_TRACE(iloc->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07004712 err = ext4_journal_get_write_access(handle, iloc->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004713 if (err) {
4714 brelse(iloc->bh);
4715 iloc->bh = NULL;
4716 }
4717 }
4718 }
Mingming Cao617ba132006-10-11 01:20:53 -07004719 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004720 return err;
4721}
4722
4723/*
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004724 * Expand an inode by new_extra_isize bytes.
4725 * Returns 0 on success or negative error number on failure.
4726 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05004727static int ext4_expand_extra_isize(struct inode *inode,
4728 unsigned int new_extra_isize,
4729 struct ext4_iloc iloc,
4730 handle_t *handle)
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004731{
4732 struct ext4_inode *raw_inode;
4733 struct ext4_xattr_ibody_header *header;
4734 struct ext4_xattr_entry *entry;
4735
4736 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4737 return 0;
4738
4739 raw_inode = ext4_raw_inode(&iloc);
4740
4741 header = IHDR(inode, raw_inode);
4742 entry = IFIRST(header);
4743
4744 /* No extended attributes present */
4745 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
4746 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4747 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4748 new_extra_isize);
4749 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4750 return 0;
4751 }
4752
4753 /* try to expand with EAs present */
4754 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4755 raw_inode, handle);
4756}
4757
4758/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004759 * What we do here is to mark the in-core inode as clean with respect to inode
4760 * dirtiness (it may still be data-dirty).
4761 * This means that the in-core inode may be reaped by prune_icache
4762 * without having to perform any I/O. This is a very good thing,
4763 * because *any* task may call prune_icache - even ones which
4764 * have a transaction open against a different journal.
4765 *
4766 * Is this cheating? Not really. Sure, we haven't written the
4767 * inode out, but prune_icache isn't a user-visible syncing function.
4768 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4769 * we start and wait on commits.
4770 *
4771 * Is this efficient/effective? Well, we're being nice to the system
4772 * by cleaning up our inodes proactively so they can be reaped
4773 * without I/O. But we are potentially leaving up to five seconds'
4774 * worth of inodes floating about which prune_icache wants us to
4775 * write out. One way to fix that would be to get prune_icache()
4776 * to do a write_super() to free up some memory. It has the desired
4777 * effect.
4778 */
Mingming Cao617ba132006-10-11 01:20:53 -07004779int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004780{
Mingming Cao617ba132006-10-11 01:20:53 -07004781 struct ext4_iloc iloc;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004782 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4783 static unsigned int mnt_count;
4784 int err, ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004785
4786 might_sleep();
Mingming Cao617ba132006-10-11 01:20:53 -07004787 err = ext4_reserve_inode_write(handle, inode, &iloc);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004788 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4789 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
4790 /*
4791 * We need extra buffer credits since we may write into EA block
4792 * with this same handle. If journal_extend fails, then it will
4793 * only result in a minor loss of functionality for that inode.
4794 * If this is felt to be critical, then e2fsck should be run to
4795 * force a large enough s_min_extra_isize.
4796 */
4797 if ((jbd2_journal_extend(handle,
4798 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4799 ret = ext4_expand_extra_isize(inode,
4800 sbi->s_want_extra_isize,
4801 iloc, handle);
4802 if (ret) {
4803 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04004804 if (mnt_count !=
4805 le16_to_cpu(sbi->s_es->s_mnt_count)) {
Harvey Harrison46e665e2008-04-17 10:38:59 -04004806 ext4_warning(inode->i_sb, __func__,
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004807 "Unable to expand inode %lu. Delete"
4808 " some EAs or run e2fsck.",
4809 inode->i_ino);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04004810 mnt_count =
4811 le16_to_cpu(sbi->s_es->s_mnt_count);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004812 }
4813 }
4814 }
4815 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004816 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07004817 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004818 return err;
4819}
4820
4821/*
Mingming Cao617ba132006-10-11 01:20:53 -07004822 * ext4_dirty_inode() is called from __mark_inode_dirty()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004823 *
4824 * We're really interested in the case where a file is being extended.
4825 * i_size has been changed by generic_commit_write() and we thus need
4826 * to include the updated inode in the current transaction.
4827 *
4828 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4829 * are allocated to the file.
4830 *
4831 * If the inode is marked synchronous, we don't honour that here - doing
4832 * so would cause a commit on atime updates, which we don't bother doing.
4833 * We handle synchronous inodes at the highest possible level.
4834 */
Mingming Cao617ba132006-10-11 01:20:53 -07004835void ext4_dirty_inode(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004836{
Mingming Cao617ba132006-10-11 01:20:53 -07004837 handle_t *current_handle = ext4_journal_current_handle();
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004838 handle_t *handle;
4839
Mingming Cao617ba132006-10-11 01:20:53 -07004840 handle = ext4_journal_start(inode, 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004841 if (IS_ERR(handle))
4842 goto out;
4843 if (current_handle &&
4844 current_handle->h_transaction != handle->h_transaction) {
4845 /* This task has a transaction open against a different fs */
4846 printk(KERN_EMERG "%s: transactions do not match!\n",
Harvey Harrison46e665e2008-04-17 10:38:59 -04004847 __func__);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004848 } else {
4849 jbd_debug(5, "marking dirty. outer handle=%p\n",
4850 current_handle);
Mingming Cao617ba132006-10-11 01:20:53 -07004851 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004852 }
Mingming Cao617ba132006-10-11 01:20:53 -07004853 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004854out:
4855 return;
4856}
4857
4858#if 0
4859/*
4860 * Bind an inode's backing buffer_head into this transaction, to prevent
4861 * it from being flushed to disk early. Unlike
Mingming Cao617ba132006-10-11 01:20:53 -07004862 * ext4_reserve_inode_write, this leaves behind no bh reference and
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004863 * returns no iloc structure, so the caller needs to repeat the iloc
4864 * lookup to mark the inode dirty later.
4865 */
Mingming Cao617ba132006-10-11 01:20:53 -07004866static int ext4_pin_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004867{
Mingming Cao617ba132006-10-11 01:20:53 -07004868 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004869
4870 int err = 0;
4871 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07004872 err = ext4_get_inode_loc(inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004873 if (!err) {
4874 BUFFER_TRACE(iloc.bh, "get_write_access");
Mingming Caodab291a2006-10-11 01:21:01 -07004875 err = jbd2_journal_get_write_access(handle, iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004876 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07004877 err = ext4_journal_dirty_metadata(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004878 iloc.bh);
4879 brelse(iloc.bh);
4880 }
4881 }
Mingming Cao617ba132006-10-11 01:20:53 -07004882 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004883 return err;
4884}
4885#endif
4886
Mingming Cao617ba132006-10-11 01:20:53 -07004887int ext4_change_inode_journal_flag(struct inode *inode, int val)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004888{
4889 journal_t *journal;
4890 handle_t *handle;
4891 int err;
4892
4893 /*
4894 * We have to be very careful here: changing a data block's
4895 * journaling status dynamically is dangerous. If we write a
4896 * data block to the journal, change the status and then delete
4897 * that block, we risk forgetting to revoke the old log record
4898 * from the journal and so a subsequent replay can corrupt data.
4899 * So, first we make sure that the journal is empty and that
4900 * nobody is changing anything.
4901 */
4902
Mingming Cao617ba132006-10-11 01:20:53 -07004903 journal = EXT4_JOURNAL(inode);
Dave Hansend6995942007-07-18 08:33:51 -04004904 if (is_journal_aborted(journal))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004905 return -EROFS;
4906
Mingming Caodab291a2006-10-11 01:21:01 -07004907 jbd2_journal_lock_updates(journal);
4908 jbd2_journal_flush(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004909
4910 /*
4911 * OK, there are no updates running now, and all cached data is
4912 * synced to disk. We are now in a completely consistent state
4913 * which doesn't have anything in the journal, and we know that
4914 * no filesystem updates are running, so it is safe to modify
4915 * the inode's in-core data-journaling state flag now.
4916 */
4917
4918 if (val)
Mingming Cao617ba132006-10-11 01:20:53 -07004919 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004920 else
Mingming Cao617ba132006-10-11 01:20:53 -07004921 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
4922 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004923
Mingming Caodab291a2006-10-11 01:21:01 -07004924 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004925
4926 /* Finally we can mark the inode as dirty. */
4927
Mingming Cao617ba132006-10-11 01:20:53 -07004928 handle = ext4_journal_start(inode, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004929 if (IS_ERR(handle))
4930 return PTR_ERR(handle);
4931
Mingming Cao617ba132006-10-11 01:20:53 -07004932 err = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004933 handle->h_sync = 1;
Mingming Cao617ba132006-10-11 01:20:53 -07004934 ext4_journal_stop(handle);
4935 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004936
4937 return err;
4938}
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004939
4940static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4941{
4942 return !buffer_mapped(bh);
4943}
4944
4945int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4946{
4947 loff_t size;
4948 unsigned long len;
4949 int ret = -EINVAL;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04004950 void *fsdata;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004951 struct file *file = vma->vm_file;
4952 struct inode *inode = file->f_path.dentry->d_inode;
4953 struct address_space *mapping = inode->i_mapping;
4954
4955 /*
4956 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4957 * get i_mutex because we are already holding mmap_sem.
4958 */
4959 down_read(&inode->i_alloc_sem);
4960 size = i_size_read(inode);
4961 if (page->mapping != mapping || size <= page_offset(page)
4962 || !PageUptodate(page)) {
4963 /* page got truncated from under us? */
4964 goto out_unlock;
4965 }
4966 ret = 0;
4967 if (PageMappedToDisk(page))
4968 goto out_unlock;
4969
4970 if (page->index == size >> PAGE_CACHE_SHIFT)
4971 len = size & ~PAGE_CACHE_MASK;
4972 else
4973 len = PAGE_CACHE_SIZE;
4974
4975 if (page_has_buffers(page)) {
4976 /* return if we have all the buffers mapped */
4977 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4978 ext4_bh_unmapped))
4979 goto out_unlock;
4980 }
4981 /*
4982 * OK, we need to fill the hole... Do write_begin write_end
4983 * to do block allocation/reservation.We are not holding
4984 * inode.i__mutex here. That allow * parallel write_begin,
4985 * write_end call. lock_page prevent this from happening
4986 * on the same page though
4987 */
4988 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04004989 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004990 if (ret < 0)
4991 goto out_unlock;
4992 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04004993 len, len, page, fsdata);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004994 if (ret < 0)
4995 goto out_unlock;
4996 ret = 0;
4997out_unlock:
4998 up_read(&inode->i_alloc_sem);
4999 return ret;
5000}