blob: 2c8caa51addb40c3120d9d3c97d91a559a70814e [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>
Duane Griffine83c1392008-12-19 20:47:15 +000037#include <linux/namei.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070038#include <linux/uio.h>
39#include <linux/bio.h>
Mingming Cao4c0425f2009-09-28 15:48:41 -040040#include <linux/workqueue.h>
Theodore Ts'o9bffad12009-06-17 11:48:11 -040041
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040042#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070043#include "xattr.h"
44#include "acl.h"
Mingming Caod2a17632008-07-14 17:52:37 -040045#include "ext4_extents.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070046
Theodore Ts'o9bffad12009-06-17 11:48:11 -040047#include <trace/events/ext4.h>
48
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -040049#define MPAGE_DA_EXTENT_TAIL 0x01
50
Jan Kara678aaf42008-07-11 19:27:31 -040051static inline int ext4_begin_ordered_truncate(struct inode *inode,
52 loff_t new_size)
53{
Jan Kara7f5aa212009-02-10 11:15:34 -050054 return jbd2_journal_begin_ordered_truncate(
55 EXT4_SB(inode->i_sb)->s_journal,
56 &EXT4_I(inode)->jinode,
57 new_size);
Jan Kara678aaf42008-07-11 19:27:31 -040058}
59
Alex Tomas64769242008-07-11 19:27:31 -040060static void ext4_invalidatepage(struct page *page, unsigned long offset);
61
Dave Kleikampac27a0e2006-10-11 01:20:50 -070062/*
63 * Test whether an inode is a fast symlink.
64 */
Mingming Cao617ba132006-10-11 01:20:53 -070065static int ext4_inode_is_fast_symlink(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070066{
Mingming Cao617ba132006-10-11 01:20:53 -070067 int ea_blocks = EXT4_I(inode)->i_file_acl ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -070068 (inode->i_sb->s_blocksize >> 9) : 0;
69
70 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
71}
72
73/*
Mingming Cao617ba132006-10-11 01:20:53 -070074 * The ext4 forget function must perform a revoke if we are freeing data
Dave Kleikampac27a0e2006-10-11 01:20:50 -070075 * which has been journaled. Metadata (eg. indirect blocks) must be
76 * revoked in all cases.
77 *
78 * "bh" may be NULL: a metadata block may have been freed from memory
79 * but there may still be a record of it in the journal, and that record
80 * still needs to be revoked.
Frank Mayhar03901312009-01-07 00:06:22 -050081 *
Curt Wohlgemuthe6b5d302009-07-13 09:07:20 -040082 * If the handle isn't valid we're not journaling, but we still need to
83 * call into ext4_journal_revoke() to put the buffer head.
Dave Kleikampac27a0e2006-10-11 01:20:50 -070084 */
Mingming Cao617ba132006-10-11 01:20:53 -070085int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -040086 struct buffer_head *bh, ext4_fsblk_t blocknr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070087{
88 int err;
89
90 might_sleep();
91
92 BUFFER_TRACE(bh, "enter");
93
94 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
Theodore Ts'o7f4520c2009-06-13 10:09:41 -040095 "data mode %x\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070096 bh, is_metadata, inode->i_mode,
97 test_opt(inode->i_sb, DATA_FLAGS));
98
99 /* Never use the revoke function if we are doing full data
100 * journaling: there is no need to, and a V1 superblock won't
101 * support it. Otherwise, only skip the revoke on un-journaled
102 * data blocks. */
103
Mingming Cao617ba132006-10-11 01:20:53 -0700104 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
105 (!is_metadata && !ext4_should_journal_data(inode))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700106 if (bh) {
Mingming Caodab291a2006-10-11 01:21:01 -0700107 BUFFER_TRACE(bh, "call jbd2_journal_forget");
Mingming Cao617ba132006-10-11 01:20:53 -0700108 return ext4_journal_forget(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700109 }
110 return 0;
111 }
112
113 /*
114 * data!=journal && (is_metadata || should_journal_data(inode))
115 */
Mingming Cao617ba132006-10-11 01:20:53 -0700116 BUFFER_TRACE(bh, "call ext4_journal_revoke");
117 err = ext4_journal_revoke(handle, blocknr, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700118 if (err)
Harvey Harrison46e665e2008-04-17 10:38:59 -0400119 ext4_abort(inode->i_sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700120 "error %d when attempting revoke", err);
121 BUFFER_TRACE(bh, "exit");
122 return err;
123}
124
125/*
126 * Work out how many blocks we need to proceed with the next chunk of a
127 * truncate transaction.
128 */
129static unsigned long blocks_for_truncate(struct inode *inode)
130{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500131 ext4_lblk_t needed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700132
133 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
134
135 /* Give ourselves just enough room to cope with inodes in which
136 * i_blocks is corrupt: we've seen disk corruptions in the past
137 * which resulted in random data in an inode which looked enough
Mingming Cao617ba132006-10-11 01:20:53 -0700138 * like a regular file for ext4 to try to delete it. Things
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700139 * will go a bit crazy if that happens, but at least we should
140 * try not to panic the whole kernel. */
141 if (needed < 2)
142 needed = 2;
143
144 /* But we need to bound the transaction so we don't overflow the
145 * journal. */
Mingming Cao617ba132006-10-11 01:20:53 -0700146 if (needed > EXT4_MAX_TRANS_DATA)
147 needed = EXT4_MAX_TRANS_DATA;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700148
Mingming Cao617ba132006-10-11 01:20:53 -0700149 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700150}
151
152/*
153 * Truncate transactions can be complex and absolutely huge. So we need to
154 * be able to restart the transaction at a conventient checkpoint to make
155 * sure we don't overflow the journal.
156 *
157 * start_transaction gets us a new handle for a truncate transaction,
158 * and extend_transaction tries to extend the existing one a bit. If
159 * extend fails, we need to propagate the failure up and restart the
160 * transaction in the top-level truncate loop. --sct
161 */
162static handle_t *start_transaction(struct inode *inode)
163{
164 handle_t *result;
165
Mingming Cao617ba132006-10-11 01:20:53 -0700166 result = ext4_journal_start(inode, blocks_for_truncate(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700167 if (!IS_ERR(result))
168 return result;
169
Mingming Cao617ba132006-10-11 01:20:53 -0700170 ext4_std_error(inode->i_sb, PTR_ERR(result));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700171 return result;
172}
173
174/*
175 * Try to extend this transaction for the purposes of truncation.
176 *
177 * Returns 0 if we managed to create more room. If we can't create more
178 * room, and the transaction must be restarted we return 1.
179 */
180static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
181{
Frank Mayhar03901312009-01-07 00:06:22 -0500182 if (!ext4_handle_valid(handle))
183 return 0;
184 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700185 return 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700186 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700187 return 0;
188 return 1;
189}
190
191/*
192 * Restart the transaction associated with *handle. This does a commit,
193 * so before we call here everything must be consistently dirtied against
194 * this transaction.
195 */
Aneesh Kumar K.Vfa5d1112009-11-02 18:50:49 -0500196int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
Jan Kara487caee2009-08-17 22:17:20 -0400197 int nblocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700198{
Jan Kara487caee2009-08-17 22:17:20 -0400199 int ret;
200
201 /*
202 * Drop i_data_sem to avoid deadlock with ext4_get_blocks At this
203 * moment, get_block can be called only for blocks inside i_size since
204 * page cache has been already dropped and writes are blocked by
205 * i_mutex. So we can safely drop the i_data_sem here.
206 */
Frank Mayhar03901312009-01-07 00:06:22 -0500207 BUG_ON(EXT4_JOURNAL(inode) == NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700208 jbd_debug(2, "restarting handle %p\n", handle);
Jan Kara487caee2009-08-17 22:17:20 -0400209 up_write(&EXT4_I(inode)->i_data_sem);
210 ret = ext4_journal_restart(handle, blocks_for_truncate(inode));
211 down_write(&EXT4_I(inode)->i_data_sem);
Aneesh Kumar K.Vfa5d1112009-11-02 18:50:49 -0500212 ext4_discard_preallocations(inode);
Jan Kara487caee2009-08-17 22:17:20 -0400213
214 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700215}
216
217/*
218 * Called at the last iput() if i_nlink is zero.
219 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400220void ext4_delete_inode(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700221{
222 handle_t *handle;
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400223 int err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700224
Jan Kara678aaf42008-07-11 19:27:31 -0400225 if (ext4_should_order_data(inode))
226 ext4_begin_ordered_truncate(inode, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700227 truncate_inode_pages(&inode->i_data, 0);
228
229 if (is_bad_inode(inode))
230 goto no_delete;
231
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400232 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700233 if (IS_ERR(handle)) {
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400234 ext4_std_error(inode->i_sb, PTR_ERR(handle));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700235 /*
236 * If we're going to skip the normal cleanup, we still need to
237 * make sure that the in-core orphan linked list is properly
238 * cleaned up.
239 */
Mingming Cao617ba132006-10-11 01:20:53 -0700240 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700241 goto no_delete;
242 }
243
244 if (IS_SYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -0500245 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700246 inode->i_size = 0;
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400247 err = ext4_mark_inode_dirty(handle, inode);
248 if (err) {
249 ext4_warning(inode->i_sb, __func__,
250 "couldn't mark inode dirty (err %d)", err);
251 goto stop_handle;
252 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700253 if (inode->i_blocks)
Mingming Cao617ba132006-10-11 01:20:53 -0700254 ext4_truncate(inode);
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400255
256 /*
257 * ext4_ext_truncate() doesn't reserve any slop when it
258 * restarts journal transactions; therefore there may not be
259 * enough credits left in the handle to remove the inode from
260 * the orphan list and set the dtime field.
261 */
Frank Mayhar03901312009-01-07 00:06:22 -0500262 if (!ext4_handle_has_enough_credits(handle, 3)) {
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400263 err = ext4_journal_extend(handle, 3);
264 if (err > 0)
265 err = ext4_journal_restart(handle, 3);
266 if (err != 0) {
267 ext4_warning(inode->i_sb, __func__,
268 "couldn't extend journal (err %d)", err);
269 stop_handle:
270 ext4_journal_stop(handle);
271 goto no_delete;
272 }
273 }
274
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700275 /*
Mingming Cao617ba132006-10-11 01:20:53 -0700276 * Kill off the orphan record which ext4_truncate created.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700277 * AKPM: I think this can be inside the above `if'.
Mingming Cao617ba132006-10-11 01:20:53 -0700278 * Note that ext4_orphan_del() has to be able to cope with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700279 * deletion of a non-existent orphan - this is because we don't
Mingming Cao617ba132006-10-11 01:20:53 -0700280 * know if ext4_truncate() actually created an orphan record.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700281 * (Well, we could do this if we need to, but heck - it works)
282 */
Mingming Cao617ba132006-10-11 01:20:53 -0700283 ext4_orphan_del(handle, inode);
284 EXT4_I(inode)->i_dtime = get_seconds();
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700285
286 /*
287 * One subtle ordering requirement: if anything has gone wrong
288 * (transaction abort, IO errors, whatever), then we can still
289 * do these next steps (the fs will already have been marked as
290 * having errors), but we can't free the inode if the mark_dirty
291 * fails.
292 */
Mingming Cao617ba132006-10-11 01:20:53 -0700293 if (ext4_mark_inode_dirty(handle, inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700294 /* If that failed, just do the required in-core inode clear. */
295 clear_inode(inode);
296 else
Mingming Cao617ba132006-10-11 01:20:53 -0700297 ext4_free_inode(handle, inode);
298 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700299 return;
300no_delete:
301 clear_inode(inode); /* We must guarantee clearing of inode... */
302}
303
304typedef struct {
305 __le32 *p;
306 __le32 key;
307 struct buffer_head *bh;
308} Indirect;
309
310static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
311{
312 p->key = *(p->p = v);
313 p->bh = bh;
314}
315
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700316/**
Mingming Cao617ba132006-10-11 01:20:53 -0700317 * ext4_block_to_path - parse the block number into array of offsets
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700318 * @inode: inode in question (we are only interested in its superblock)
319 * @i_block: block number to be parsed
320 * @offsets: array to store the offsets in
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400321 * @boundary: set this non-zero if the referred-to block is likely to be
322 * followed (on disk) by an indirect block.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700323 *
Mingming Cao617ba132006-10-11 01:20:53 -0700324 * To store the locations of file's data ext4 uses a data structure common
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700325 * for UNIX filesystems - tree of pointers anchored in the inode, with
326 * data blocks at leaves and indirect blocks in intermediate nodes.
327 * This function translates the block number into path in that tree -
328 * return value is the path length and @offsets[n] is the offset of
329 * pointer to (n+1)th node in the nth one. If @block is out of range
330 * (negative or too large) warning is printed and zero returned.
331 *
332 * Note: function doesn't find node addresses, so no IO is needed. All
333 * we need to know is the capacity of indirect blocks (taken from the
334 * inode->i_sb).
335 */
336
337/*
338 * Portability note: the last comparison (check that we fit into triple
339 * indirect block) is spelled differently, because otherwise on an
340 * architecture with 32-bit longs and 8Kb pages we might get into trouble
341 * if our filesystem had 8Kb blocks. We might use long long, but that would
342 * kill us on x86. Oh, well, at least the sign propagation does not matter -
343 * i_block would have to be negative in the very beginning, so we would not
344 * get there at all.
345 */
346
Mingming Cao617ba132006-10-11 01:20:53 -0700347static int ext4_block_to_path(struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400348 ext4_lblk_t i_block,
349 ext4_lblk_t offsets[4], int *boundary)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700350{
Mingming Cao617ba132006-10-11 01:20:53 -0700351 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
352 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
353 const long direct_blocks = EXT4_NDIR_BLOCKS,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700354 indirect_blocks = ptrs,
355 double_blocks = (1 << (ptrs_bits * 2));
356 int n = 0;
357 int final = 0;
358
Roel Kluinc333e072009-08-10 22:47:22 -0400359 if (i_block < direct_blocks) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700360 offsets[n++] = i_block;
361 final = direct_blocks;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400362 } else if ((i_block -= direct_blocks) < indirect_blocks) {
Mingming Cao617ba132006-10-11 01:20:53 -0700363 offsets[n++] = EXT4_IND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700364 offsets[n++] = i_block;
365 final = ptrs;
366 } else if ((i_block -= indirect_blocks) < double_blocks) {
Mingming Cao617ba132006-10-11 01:20:53 -0700367 offsets[n++] = EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700368 offsets[n++] = i_block >> ptrs_bits;
369 offsets[n++] = i_block & (ptrs - 1);
370 final = ptrs;
371 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
Mingming Cao617ba132006-10-11 01:20:53 -0700372 offsets[n++] = EXT4_TIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700373 offsets[n++] = i_block >> (ptrs_bits * 2);
374 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
375 offsets[n++] = i_block & (ptrs - 1);
376 final = ptrs;
377 } else {
Eric Sandeene2b46572008-01-28 23:58:27 -0500378 ext4_warning(inode->i_sb, "ext4_block_to_path",
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400379 "block %lu > max in inode %lu",
380 i_block + direct_blocks +
381 indirect_blocks + double_blocks, inode->i_ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700382 }
383 if (boundary)
384 *boundary = final - 1 - (i_block & (ptrs - 1));
385 return n;
386}
387
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400388static int __ext4_check_blockref(const char *function, struct inode *inode,
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400389 __le32 *p, unsigned int max)
390{
Thiemo Nagelf73953c2009-04-07 18:46:47 -0400391 __le32 *bref = p;
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400392 unsigned int blk;
393
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400394 while (bref < p+max) {
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400395 blk = le32_to_cpu(*bref++);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400396 if (blk &&
397 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400398 blk, 1))) {
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400399 ext4_error(inode->i_sb, function,
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400400 "invalid block reference %u "
401 "in inode #%lu", blk, inode->i_ino);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400402 return -EIO;
403 }
404 }
405 return 0;
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400406}
407
408
409#define ext4_check_indirect_blockref(inode, bh) \
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400410 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400411 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
412
413#define ext4_check_inode_blockref(inode) \
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400414 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400415 EXT4_NDIR_BLOCKS)
416
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700417/**
Mingming Cao617ba132006-10-11 01:20:53 -0700418 * ext4_get_branch - read the chain of indirect blocks leading to data
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700419 * @inode: inode in question
420 * @depth: depth of the chain (1 - direct pointer, etc.)
421 * @offsets: offsets of pointers in inode/indirect blocks
422 * @chain: place to store the result
423 * @err: here we store the error value
424 *
425 * Function fills the array of triples <key, p, bh> and returns %NULL
426 * if everything went OK or the pointer to the last filled triple
427 * (incomplete one) otherwise. Upon the return chain[i].key contains
428 * the number of (i+1)-th block in the chain (as it is stored in memory,
429 * i.e. little-endian 32-bit), chain[i].p contains the address of that
430 * number (it points into struct inode for i==0 and into the bh->b_data
431 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
432 * block for i>0 and NULL for i==0. In other words, it holds the block
433 * numbers of the chain, addresses they were taken from (and where we can
434 * verify that chain did not change) and buffer_heads hosting these
435 * numbers.
436 *
437 * Function stops when it stumbles upon zero pointer (absent block)
438 * (pointer to last triple returned, *@err == 0)
439 * or when it gets an IO error reading an indirect block
440 * (ditto, *@err == -EIO)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700441 * or when it reads all @depth-1 indirect blocks successfully and finds
442 * the whole chain, all way to the data (returns %NULL, *err == 0).
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -0500443 *
444 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500445 * down_read(&EXT4_I(inode)->i_data_sem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700446 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500447static Indirect *ext4_get_branch(struct inode *inode, int depth,
448 ext4_lblk_t *offsets,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700449 Indirect chain[4], int *err)
450{
451 struct super_block *sb = inode->i_sb;
452 Indirect *p = chain;
453 struct buffer_head *bh;
454
455 *err = 0;
456 /* i_data is not going away, no lock needed */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400457 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700458 if (!p->key)
459 goto no_block;
460 while (--depth) {
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400461 bh = sb_getblk(sb, le32_to_cpu(p->key));
462 if (unlikely(!bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700463 goto failure;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400464
Thiemo Nagelfe2c8192009-03-31 08:36:10 -0400465 if (!bh_uptodate_or_lock(bh)) {
466 if (bh_submit_read(bh) < 0) {
467 put_bh(bh);
468 goto failure;
469 }
470 /* validate block references */
471 if (ext4_check_indirect_blockref(inode, bh)) {
472 put_bh(bh);
473 goto failure;
474 }
475 }
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400476
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400477 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700478 /* Reader: end */
479 if (!p->key)
480 goto no_block;
481 }
482 return NULL;
483
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700484failure:
485 *err = -EIO;
486no_block:
487 return p;
488}
489
490/**
Mingming Cao617ba132006-10-11 01:20:53 -0700491 * ext4_find_near - find a place for allocation with sufficient locality
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700492 * @inode: owner
493 * @ind: descriptor of indirect block.
494 *
Benoit Boissinot1cc8dcf2008-04-21 22:45:55 +0000495 * This function returns the preferred place for block allocation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700496 * It is used when heuristic for sequential allocation fails.
497 * Rules are:
498 * + if there is a block to the left of our position - allocate near it.
499 * + if pointer will live in indirect block - allocate near that block.
500 * + if pointer will live in inode - allocate in the same
501 * cylinder group.
502 *
503 * In the latter case we colour the starting block by the callers PID to
504 * prevent it from clashing with concurrent allocations for a different inode
505 * in the same block group. The PID is used here so that functionally related
506 * files will be close-by on-disk.
507 *
508 * Caller must make sure that @ind is valid and will stay that way.
509 */
Mingming Cao617ba132006-10-11 01:20:53 -0700510static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700511{
Mingming Cao617ba132006-10-11 01:20:53 -0700512 struct ext4_inode_info *ei = EXT4_I(inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400513 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700514 __le32 *p;
Mingming Cao617ba132006-10-11 01:20:53 -0700515 ext4_fsblk_t bg_start;
Valerie Clement74d34872008-02-15 13:43:07 -0500516 ext4_fsblk_t last_block;
Mingming Cao617ba132006-10-11 01:20:53 -0700517 ext4_grpblk_t colour;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400518 ext4_group_t block_group;
519 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700520
521 /* Try to find previous block */
522 for (p = ind->p - 1; p >= start; p--) {
523 if (*p)
524 return le32_to_cpu(*p);
525 }
526
527 /* No such thing, so let's try location of indirect block */
528 if (ind->bh)
529 return ind->bh->b_blocknr;
530
531 /*
532 * It is going to be referred to from the inode itself? OK, just put it
533 * into the same cylinder group then.
534 */
Theodore Ts'oa4912122009-03-12 12:18:34 -0400535 block_group = ei->i_block_group;
536 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
537 block_group &= ~(flex_size-1);
538 if (S_ISREG(inode->i_mode))
539 block_group++;
540 }
541 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
Valerie Clement74d34872008-02-15 13:43:07 -0500542 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
543
Theodore Ts'oa4912122009-03-12 12:18:34 -0400544 /*
545 * If we are doing delayed allocation, we don't need take
546 * colour into account.
547 */
548 if (test_opt(inode->i_sb, DELALLOC))
549 return bg_start;
550
Valerie Clement74d34872008-02-15 13:43:07 -0500551 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
552 colour = (current->pid % 16) *
Mingming Cao617ba132006-10-11 01:20:53 -0700553 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
Valerie Clement74d34872008-02-15 13:43:07 -0500554 else
555 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700556 return bg_start + colour;
557}
558
559/**
Benoit Boissinot1cc8dcf2008-04-21 22:45:55 +0000560 * ext4_find_goal - find a preferred place for allocation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700561 * @inode: owner
562 * @block: block we want
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700563 * @partial: pointer to the last triple within a chain
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700564 *
Benoit Boissinot1cc8dcf2008-04-21 22:45:55 +0000565 * Normally this function find the preferred place for block allocation,
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800566 * returns it.
Eric Sandeenfb0a3872009-09-16 14:45:10 -0400567 * Because this is only used for non-extent files, we limit the block nr
568 * to 32 bits.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700569 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500570static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400571 Indirect *partial)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700572{
Eric Sandeenfb0a3872009-09-16 14:45:10 -0400573 ext4_fsblk_t goal;
574
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700575 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -0400576 * XXX need to get goal block from mballoc's data structures
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700577 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700578
Eric Sandeenfb0a3872009-09-16 14:45:10 -0400579 goal = ext4_find_near(inode, partial);
580 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
581 return goal;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700582}
583
584/**
Mingming Cao617ba132006-10-11 01:20:53 -0700585 * ext4_blks_to_allocate: Look up the block map and count the number
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700586 * of direct blocks need to be allocated for the given branch.
587 *
588 * @branch: chain of indirect blocks
589 * @k: number of blocks need for indirect blocks
590 * @blks: number of data blocks to be mapped.
591 * @blocks_to_boundary: the offset in the indirect block
592 *
593 * return the total number of blocks to be allocate, including the
594 * direct and indirect blocks.
595 */
Theodore Ts'o498e5f22008-11-05 00:14:04 -0500596static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400597 int blocks_to_boundary)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700598{
Theodore Ts'o498e5f22008-11-05 00:14:04 -0500599 unsigned int count = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700600
601 /*
602 * Simple case, [t,d]Indirect block(s) has not allocated yet
603 * then it's clear blocks on that path have not allocated
604 */
605 if (k > 0) {
606 /* right now we don't handle cross boundary allocation */
607 if (blks < blocks_to_boundary + 1)
608 count += blks;
609 else
610 count += blocks_to_boundary + 1;
611 return count;
612 }
613
614 count++;
615 while (count < blks && count <= blocks_to_boundary &&
616 le32_to_cpu(*(branch[0].p + count)) == 0) {
617 count++;
618 }
619 return count;
620}
621
622/**
Mingming Cao617ba132006-10-11 01:20:53 -0700623 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700624 * @indirect_blks: the number of blocks need to allocate for indirect
625 * blocks
626 *
627 * @new_blocks: on return it will store the new block numbers for
628 * the indirect blocks(if needed) and the first direct block,
629 * @blks: on return it will store the total number of allocated
630 * direct blocks
631 */
Mingming Cao617ba132006-10-11 01:20:53 -0700632static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400633 ext4_lblk_t iblock, ext4_fsblk_t goal,
634 int indirect_blks, int blks,
635 ext4_fsblk_t new_blocks[4], int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700636{
Theodore Ts'o815a1132009-01-01 23:59:43 -0500637 struct ext4_allocation_request ar;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700638 int target, i;
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400639 unsigned long count = 0, blk_allocated = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700640 int index = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700641 ext4_fsblk_t current_block = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700642 int ret = 0;
643
644 /*
645 * Here we try to allocate the requested multiple blocks at once,
646 * on a best-effort basis.
647 * To build a branch, we should allocate blocks for
648 * the indirect blocks(if not allocated yet), and at least
649 * the first direct block of this branch. That's the
650 * minimum number of blocks need to allocate(required)
651 */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400652 /* first we try to allocate the indirect blocks */
653 target = indirect_blks;
654 while (target > 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700655 count = target;
656 /* allocating blocks for indirect blocks and direct blocks */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400657 current_block = ext4_new_meta_blocks(handle, inode,
658 goal, &count, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700659 if (*err)
660 goto failed_out;
661
Eric Sandeenfb0a3872009-09-16 14:45:10 -0400662 BUG_ON(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS);
663
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700664 target -= count;
665 /* allocate blocks for indirect blocks */
666 while (index < indirect_blks && count) {
667 new_blocks[index++] = current_block++;
668 count--;
669 }
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400670 if (count > 0) {
671 /*
672 * save the new block number
673 * for the first direct block
674 */
675 new_blocks[index] = current_block;
676 printk(KERN_INFO "%s returned more blocks than "
677 "requested\n", __func__);
678 WARN_ON(1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700679 break;
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400680 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700681 }
682
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400683 target = blks - count ;
684 blk_allocated = count;
685 if (!target)
686 goto allocated;
687 /* Now allocate data blocks */
Theodore Ts'o815a1132009-01-01 23:59:43 -0500688 memset(&ar, 0, sizeof(ar));
689 ar.inode = inode;
690 ar.goal = goal;
691 ar.len = target;
692 ar.logical = iblock;
693 if (S_ISREG(inode->i_mode))
694 /* enable in-core preallocation only for regular files */
695 ar.flags = EXT4_MB_HINT_DATA;
696
697 current_block = ext4_mb_new_blocks(handle, &ar, err);
Eric Sandeenfb0a3872009-09-16 14:45:10 -0400698 BUG_ON(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS);
Theodore Ts'o815a1132009-01-01 23:59:43 -0500699
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400700 if (*err && (target == blks)) {
701 /*
702 * if the allocation failed and we didn't allocate
703 * any blocks before
704 */
705 goto failed_out;
706 }
707 if (!*err) {
708 if (target == blks) {
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400709 /*
710 * save the new block number
711 * for the first direct block
712 */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400713 new_blocks[index] = current_block;
714 }
Theodore Ts'o815a1132009-01-01 23:59:43 -0500715 blk_allocated += ar.len;
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400716 }
717allocated:
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700718 /* total number of blocks allocated for direct blocks */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400719 ret = blk_allocated;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700720 *err = 0;
721 return ret;
722failed_out:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400723 for (i = 0; i < index; i++)
Alex Tomasc9de5602008-01-29 00:19:52 -0500724 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700725 return ret;
726}
727
728/**
Mingming Cao617ba132006-10-11 01:20:53 -0700729 * ext4_alloc_branch - allocate and set up a chain of blocks.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700730 * @inode: owner
731 * @indirect_blks: number of allocated indirect blocks
732 * @blks: number of allocated direct blocks
733 * @offsets: offsets (in the blocks) to store the pointers to next.
734 * @branch: place to store the chain in.
735 *
736 * This function allocates blocks, zeroes out all but the last one,
737 * links them into chain and (if we are synchronous) writes them to disk.
738 * In other words, it prepares a branch that can be spliced onto the
739 * inode. It stores the information about that chain in the branch[], in
Mingming Cao617ba132006-10-11 01:20:53 -0700740 * the same format as ext4_get_branch() would do. We are calling it after
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700741 * we had read the existing part of chain and partial points to the last
742 * triple of that (one with zero ->key). Upon the exit we have the same
Mingming Cao617ba132006-10-11 01:20:53 -0700743 * picture as after the successful ext4_get_block(), except that in one
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700744 * place chain is disconnected - *branch->p is still zero (we did not
745 * set the last link), but branch->key contains the number that should
746 * be placed into *branch->p to fill that gap.
747 *
748 * If allocation fails we free all blocks we've allocated (and forget
749 * their buffer_heads) and return the error value the from failed
Mingming Cao617ba132006-10-11 01:20:53 -0700750 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700751 * as described above and return 0.
752 */
Mingming Cao617ba132006-10-11 01:20:53 -0700753static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400754 ext4_lblk_t iblock, int indirect_blks,
755 int *blks, ext4_fsblk_t goal,
756 ext4_lblk_t *offsets, Indirect *branch)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700757{
758 int blocksize = inode->i_sb->s_blocksize;
759 int i, n = 0;
760 int err = 0;
761 struct buffer_head *bh;
762 int num;
Mingming Cao617ba132006-10-11 01:20:53 -0700763 ext4_fsblk_t new_blocks[4];
764 ext4_fsblk_t current_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700765
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400766 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700767 *blks, new_blocks, &err);
768 if (err)
769 return err;
770
771 branch[0].key = cpu_to_le32(new_blocks[0]);
772 /*
773 * metadata blocks and data blocks are allocated.
774 */
775 for (n = 1; n <= indirect_blks; n++) {
776 /*
777 * Get buffer_head for parent block, zero it out
778 * and set the pointer to new one, then send
779 * parent to disk.
780 */
781 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
782 branch[n].bh = bh;
783 lock_buffer(bh);
784 BUFFER_TRACE(bh, "call get_create_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700785 err = ext4_journal_get_create_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700786 if (err) {
Curt Wohlgemuth6487a9d2009-07-17 10:54:08 -0400787 /* Don't brelse(bh) here; it's done in
788 * ext4_journal_forget() below */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700789 unlock_buffer(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700790 goto failed;
791 }
792
793 memset(bh->b_data, 0, blocksize);
794 branch[n].p = (__le32 *) bh->b_data + offsets[n];
795 branch[n].key = cpu_to_le32(new_blocks[n]);
796 *branch[n].p = branch[n].key;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400797 if (n == indirect_blks) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700798 current_block = new_blocks[n];
799 /*
800 * End of chain, update the last new metablock of
801 * the chain to point to the new allocated
802 * data blocks numbers
803 */
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400804 for (i = 1; i < num; i++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700805 *(branch[n].p + i) = cpu_to_le32(++current_block);
806 }
807 BUFFER_TRACE(bh, "marking uptodate");
808 set_buffer_uptodate(bh);
809 unlock_buffer(bh);
810
Frank Mayhar03901312009-01-07 00:06:22 -0500811 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
812 err = ext4_handle_dirty_metadata(handle, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700813 if (err)
814 goto failed;
815 }
816 *blks = num;
817 return err;
818failed:
819 /* Allocation failed, free what we already allocated */
820 for (i = 1; i <= n ; i++) {
Mingming Caodab291a2006-10-11 01:21:01 -0700821 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
Mingming Cao617ba132006-10-11 01:20:53 -0700822 ext4_journal_forget(handle, branch[i].bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700823 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400824 for (i = 0; i < indirect_blks; i++)
Alex Tomasc9de5602008-01-29 00:19:52 -0500825 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700826
Alex Tomasc9de5602008-01-29 00:19:52 -0500827 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700828
829 return err;
830}
831
832/**
Mingming Cao617ba132006-10-11 01:20:53 -0700833 * ext4_splice_branch - splice the allocated branch onto inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700834 * @inode: owner
835 * @block: (logical) number of block we are adding
836 * @chain: chain of indirect blocks (with a missing link - see
Mingming Cao617ba132006-10-11 01:20:53 -0700837 * ext4_alloc_branch)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700838 * @where: location of missing link
839 * @num: number of indirect blocks we are adding
840 * @blks: number of direct blocks we are adding
841 *
842 * This function fills the missing link and does all housekeeping needed in
843 * inode (->i_blocks, etc.). In case of success we end up with the full
844 * chain to new block and return 0.
845 */
Mingming Cao617ba132006-10-11 01:20:53 -0700846static int ext4_splice_branch(handle_t *handle, struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400847 ext4_lblk_t block, Indirect *where, int num,
848 int blks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700849{
850 int i;
851 int err = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700852 ext4_fsblk_t current_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700853
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700854 /*
855 * If we're splicing into a [td]indirect block (as opposed to the
856 * inode) then we need to get write access to the [td]indirect block
857 * before the splice.
858 */
859 if (where->bh) {
860 BUFFER_TRACE(where->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700861 err = ext4_journal_get_write_access(handle, where->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700862 if (err)
863 goto err_out;
864 }
865 /* That's it */
866
867 *where->p = where->key;
868
869 /*
870 * Update the host buffer_head or inode to point to more just allocated
871 * direct blocks blocks
872 */
873 if (num == 0 && blks > 1) {
874 current_block = le32_to_cpu(where->key) + 1;
875 for (i = 1; i < blks; i++)
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400876 *(where->p + i) = cpu_to_le32(current_block++);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700877 }
878
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700879 /* We are done with atomic stuff, now do the rest of housekeeping */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700880 /* had we spliced it onto indirect block? */
881 if (where->bh) {
882 /*
883 * If we spliced it onto an indirect block, we haven't
884 * altered the inode. Note however that if it is being spliced
885 * onto an indirect block at the very end of the file (the
886 * file is growing) then we *will* alter the inode to reflect
887 * the new i_size. But that is not done here - it is done in
Mingming Cao617ba132006-10-11 01:20:53 -0700888 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700889 */
890 jbd_debug(5, "splicing indirect only\n");
Frank Mayhar03901312009-01-07 00:06:22 -0500891 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
892 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700893 if (err)
894 goto err_out;
895 } else {
896 /*
897 * OK, we spliced it into the inode itself on a direct block.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700898 */
Theodore Ts'o41591752009-06-15 03:41:23 -0400899 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700900 jbd_debug(5, "splicing direct\n");
901 }
902 return err;
903
904err_out:
905 for (i = 1; i <= num; i++) {
Mingming Caodab291a2006-10-11 01:21:01 -0700906 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
Mingming Cao617ba132006-10-11 01:20:53 -0700907 ext4_journal_forget(handle, where[i].bh);
Alex Tomasc9de5602008-01-29 00:19:52 -0500908 ext4_free_blocks(handle, inode,
909 le32_to_cpu(where[i-1].key), 1, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700910 }
Alex Tomasc9de5602008-01-29 00:19:52 -0500911 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700912
913 return err;
914}
915
916/*
Theodore Ts'ob920c752009-05-14 00:54:29 -0400917 * The ext4_ind_get_blocks() function handles non-extents inodes
918 * (i.e., using the traditional indirect/double-indirect i_blocks
919 * scheme) for ext4_get_blocks().
920 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700921 * Allocation strategy is simple: if we have to allocate something, we will
922 * have to go the whole way to leaf. So let's do it before attaching anything
923 * to tree, set linkage between the newborn blocks, write them if sync is
924 * required, recheck the path, free and repeat if check fails, otherwise
925 * set the last missing link (that will protect us from any truncate-generated
926 * removals - all blocks on the path are immune now) and possibly force the
927 * write on the parent block.
928 * That has a nice additional property: no special recovery from the failed
929 * allocations is needed - we simply release blocks and do not touch anything
930 * reachable from inode.
931 *
932 * `handle' can be NULL if create == 0.
933 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700934 * return > 0, # of blocks mapped or allocated.
935 * return = 0, if plain lookup failed.
936 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -0500937 *
Theodore Ts'ob920c752009-05-14 00:54:29 -0400938 * The ext4_ind_get_blocks() function should be called with
939 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
940 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
941 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
942 * blocks.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700943 */
Theodore Ts'oe4d996c2009-05-12 00:25:28 -0400944static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400945 ext4_lblk_t iblock, unsigned int maxblocks,
946 struct buffer_head *bh_result,
947 int flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700948{
949 int err = -EIO;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500950 ext4_lblk_t offsets[4];
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700951 Indirect chain[4];
952 Indirect *partial;
Mingming Cao617ba132006-10-11 01:20:53 -0700953 ext4_fsblk_t goal;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700954 int indirect_blks;
955 int blocks_to_boundary = 0;
956 int depth;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700957 int count = 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700958 ext4_fsblk_t first_block = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700959
Alex Tomasa86c6182006-10-11 01:21:03 -0700960 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
Theodore Ts'oc2177052009-05-14 00:58:52 -0400961 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500962 depth = ext4_block_to_path(inode, iblock, offsets,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400963 &blocks_to_boundary);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700964
965 if (depth == 0)
966 goto out;
967
Mingming Cao617ba132006-10-11 01:20:53 -0700968 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700969
970 /* Simplest case - block found, no allocation needed */
971 if (!partial) {
972 first_block = le32_to_cpu(chain[depth - 1].key);
973 clear_buffer_new(bh_result);
974 count++;
975 /*map more blocks*/
976 while (count < maxblocks && count <= blocks_to_boundary) {
Mingming Cao617ba132006-10-11 01:20:53 -0700977 ext4_fsblk_t blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700978
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700979 blk = le32_to_cpu(*(chain[depth-1].p + count));
980
981 if (blk == first_block + count)
982 count++;
983 else
984 break;
985 }
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -0500986 goto got_it;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700987 }
988
989 /* Next simple case - plain lookup or failed read of indirect block */
Theodore Ts'oc2177052009-05-14 00:58:52 -0400990 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700991 goto cleanup;
992
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700993 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -0400994 * Okay, we need to do block allocation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700995 */
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800996 goal = ext4_find_goal(inode, iblock, partial);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700997
998 /* the number of blocks need to allocate for [d,t]indirect blocks */
999 indirect_blks = (chain + depth) - partial - 1;
1000
1001 /*
1002 * Next look up the indirect map to count the totoal number of
1003 * direct blocks to allocate for this branch.
1004 */
Mingming Cao617ba132006-10-11 01:20:53 -07001005 count = ext4_blks_to_allocate(partial, indirect_blks,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001006 maxblocks, blocks_to_boundary);
1007 /*
Mingming Cao617ba132006-10-11 01:20:53 -07001008 * Block out ext4_truncate while we alter the tree
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001009 */
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -04001010 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001011 &count, goal,
1012 offsets + (partial - chain), partial);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001013
1014 /*
Mingming Cao617ba132006-10-11 01:20:53 -07001015 * The ext4_splice_branch call will free and forget any buffers
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001016 * on the new chain if there is a failure, but that risks using
1017 * up transaction credits, especially for bitmaps where the
1018 * credits cannot be returned. Can we handle this somehow? We
1019 * may need to return -EAGAIN upwards in the worst case. --sct
1020 */
1021 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07001022 err = ext4_splice_branch(handle, inode, iblock,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001023 partial, indirect_blks, count);
1024 else
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001025 goto cleanup;
1026
1027 set_buffer_new(bh_result);
1028got_it:
1029 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
1030 if (count > blocks_to_boundary)
1031 set_buffer_boundary(bh_result);
1032 err = count;
1033 /* Clean up and exit */
1034 partial = chain + depth - 1; /* the whole chain */
1035cleanup:
1036 while (partial > chain) {
1037 BUFFER_TRACE(partial->bh, "call brelse");
1038 brelse(partial->bh);
1039 partial--;
1040 }
1041 BUFFER_TRACE(bh_result, "returned");
1042out:
1043 return err;
1044}
1045
Mingming Cao60e58e02009-01-22 18:13:05 +01001046qsize_t ext4_get_reserved_space(struct inode *inode)
1047{
1048 unsigned long long total;
1049
1050 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1051 total = EXT4_I(inode)->i_reserved_data_blocks +
1052 EXT4_I(inode)->i_reserved_meta_blocks;
1053 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1054
1055 return total;
1056}
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001057/*
1058 * Calculate the number of metadata blocks need to reserve
1059 * to allocate @blocks for non extent file based file
1060 */
1061static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
1062{
1063 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1064 int ind_blks, dind_blks, tind_blks;
1065
1066 /* number of new indirect blocks needed */
1067 ind_blks = (blocks + icap - 1) / icap;
1068
1069 dind_blks = (ind_blks + icap - 1) / icap;
1070
1071 tind_blks = 1;
1072
1073 return ind_blks + dind_blks + tind_blks;
1074}
1075
1076/*
1077 * Calculate the number of metadata blocks need to reserve
1078 * to allocate given number of blocks
1079 */
1080static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1081{
Mingming Caocd213222008-08-19 22:16:59 -04001082 if (!blocks)
1083 return 0;
1084
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001085 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1086 return ext4_ext_calc_metadata_amount(inode, blocks);
1087
1088 return ext4_indirect_calc_metadata_amount(inode, blocks);
1089}
1090
1091static void ext4_da_update_reserve_space(struct inode *inode, int used)
1092{
1093 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1094 int total, mdb, mdb_free;
1095
1096 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1097 /* recalculate the number of metablocks still need to be reserved */
1098 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1099 mdb = ext4_calc_metadata_amount(inode, total);
1100
1101 /* figure out how many metablocks to release */
1102 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1103 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1104
Aneesh Kumar K.V6bc6e632008-10-10 09:39:00 -04001105 if (mdb_free) {
1106 /* Account for allocated meta_blocks */
1107 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001108
Aneesh Kumar K.V6bc6e632008-10-10 09:39:00 -04001109 /* update fs dirty blocks counter */
1110 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1111 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1112 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1113 }
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001114
1115 /* update per-inode reservations */
1116 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1117 EXT4_I(inode)->i_reserved_data_blocks -= used;
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001118 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao60e58e02009-01-22 18:13:05 +01001119
1120 /*
1121 * free those over-booking quota for metadata blocks
1122 */
Mingming Cao60e58e02009-01-22 18:13:05 +01001123 if (mdb_free)
1124 vfs_dq_release_reservation_block(inode, mdb_free);
Aneesh Kumar K.Vd6014302009-03-27 22:36:43 -04001125
1126 /*
1127 * If we have done all the pending block allocations and if
1128 * there aren't any writers on the inode, we can discard the
1129 * inode's preallocations.
1130 */
1131 if (!total && (atomic_read(&inode->i_writecount) == 0))
1132 ext4_discard_preallocations(inode);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001133}
1134
Theodore Ts'o80e42462009-09-08 08:21:26 -04001135static int check_block_validity(struct inode *inode, const char *msg,
1136 sector_t logical, sector_t phys, int len)
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001137{
1138 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), phys, len)) {
Theodore Ts'o80e42462009-09-08 08:21:26 -04001139 ext4_error(inode->i_sb, msg,
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001140 "inode #%lu logical block %llu mapped to %llu "
1141 "(size %d)", inode->i_ino,
1142 (unsigned long long) logical,
1143 (unsigned long long) phys, len);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001144 return -EIO;
1145 }
1146 return 0;
1147}
1148
Mingming Caof5ab0d12008-02-25 15:29:55 -05001149/*
Theodore Ts'o1f945332009-09-30 22:57:41 -04001150 * Return the number of contiguous dirty pages in a given inode
1151 * starting at page frame idx.
Theodore Ts'o55138e02009-09-29 13:31:31 -04001152 */
1153static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1154 unsigned int max_pages)
1155{
1156 struct address_space *mapping = inode->i_mapping;
1157 pgoff_t index;
1158 struct pagevec pvec;
1159 pgoff_t num = 0;
1160 int i, nr_pages, done = 0;
1161
1162 if (max_pages == 0)
1163 return 0;
1164 pagevec_init(&pvec, 0);
1165 while (!done) {
1166 index = idx;
1167 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1168 PAGECACHE_TAG_DIRTY,
1169 (pgoff_t)PAGEVEC_SIZE);
1170 if (nr_pages == 0)
1171 break;
1172 for (i = 0; i < nr_pages; i++) {
1173 struct page *page = pvec.pages[i];
1174 struct buffer_head *bh, *head;
1175
1176 lock_page(page);
1177 if (unlikely(page->mapping != mapping) ||
1178 !PageDirty(page) ||
1179 PageWriteback(page) ||
1180 page->index != idx) {
1181 done = 1;
1182 unlock_page(page);
1183 break;
1184 }
Theodore Ts'o1f945332009-09-30 22:57:41 -04001185 if (page_has_buffers(page)) {
1186 bh = head = page_buffers(page);
1187 do {
1188 if (!buffer_delay(bh) &&
1189 !buffer_unwritten(bh))
1190 done = 1;
1191 bh = bh->b_this_page;
1192 } while (!done && (bh != head));
1193 }
Theodore Ts'o55138e02009-09-29 13:31:31 -04001194 unlock_page(page);
1195 if (done)
1196 break;
1197 idx++;
1198 num++;
1199 if (num >= max_pages)
1200 break;
1201 }
1202 pagevec_release(&pvec);
1203 }
1204 return num;
1205}
1206
1207/*
Theodore Ts'o12b7ac12009-05-14 00:57:44 -04001208 * The ext4_get_blocks() function tries to look up the requested blocks,
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001209 * and returns if the blocks are already mapped.
Mingming Caof5ab0d12008-02-25 15:29:55 -05001210 *
Mingming Caof5ab0d12008-02-25 15:29:55 -05001211 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1212 * and store the allocated blocks in the result buffer head and mark it
1213 * mapped.
1214 *
1215 * If file type is extents based, it will call ext4_ext_get_blocks(),
Theodore Ts'oe4d996c2009-05-12 00:25:28 -04001216 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
Mingming Caof5ab0d12008-02-25 15:29:55 -05001217 * based files
1218 *
1219 * On success, it returns the number of blocks being mapped or allocate.
1220 * if create==0 and the blocks are pre-allocated and uninitialized block,
1221 * the result buffer head is unmapped. If the create ==1, it will make sure
1222 * the buffer head is mapped.
1223 *
1224 * It returns 0 if plain look up failed (blocks have not been allocated), in
1225 * that casem, buffer head is unmapped
1226 *
1227 * It returns the error in case of allocation failure.
1228 */
Theodore Ts'o12b7ac12009-05-14 00:57:44 -04001229int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
1230 unsigned int max_blocks, struct buffer_head *bh,
Theodore Ts'oc2177052009-05-14 00:58:52 -04001231 int flags)
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001232{
1233 int retval;
Mingming Caof5ab0d12008-02-25 15:29:55 -05001234
1235 clear_buffer_mapped(bh);
Aneesh Kumar K.V2a8964d2009-05-14 17:05:39 -04001236 clear_buffer_unwritten(bh);
Mingming Caof5ab0d12008-02-25 15:29:55 -05001237
Mingming Cao00314622009-09-28 15:49:08 -04001238 ext_debug("ext4_get_blocks(): inode %lu, flag %d, max_blocks %u,"
1239 "logical block %lu\n", inode->i_ino, flags, max_blocks,
1240 (unsigned long)block);
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001241 /*
Theodore Ts'ob920c752009-05-14 00:54:29 -04001242 * Try to see if we can get the block without requesting a new
1243 * file system block.
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001244 */
1245 down_read((&EXT4_I(inode)->i_data_sem));
1246 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1247 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
Theodore Ts'oc2177052009-05-14 00:58:52 -04001248 bh, 0);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001249 } else {
Theodore Ts'oe4d996c2009-05-12 00:25:28 -04001250 retval = ext4_ind_get_blocks(handle, inode, block, max_blocks,
Theodore Ts'oc2177052009-05-14 00:58:52 -04001251 bh, 0);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001252 }
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001253 up_read((&EXT4_I(inode)->i_data_sem));
Mingming Caof5ab0d12008-02-25 15:29:55 -05001254
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001255 if (retval > 0 && buffer_mapped(bh)) {
Theodore Ts'o80e42462009-09-08 08:21:26 -04001256 int ret = check_block_validity(inode, "file system corruption",
1257 block, bh->b_blocknr, retval);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001258 if (ret != 0)
1259 return ret;
1260 }
1261
Mingming Caof5ab0d12008-02-25 15:29:55 -05001262 /* If it is only a block(s) look up */
Theodore Ts'oc2177052009-05-14 00:58:52 -04001263 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001264 return retval;
1265
1266 /*
Mingming Caof5ab0d12008-02-25 15:29:55 -05001267 * Returns if the blocks have already allocated
1268 *
1269 * Note that if blocks have been preallocated
1270 * ext4_ext_get_block() returns th create = 0
1271 * with buffer head unmapped.
1272 */
1273 if (retval > 0 && buffer_mapped(bh))
1274 return retval;
1275
1276 /*
Aneesh Kumar K.V2a8964d2009-05-14 17:05:39 -04001277 * When we call get_blocks without the create flag, the
1278 * BH_Unwritten flag could have gotten set if the blocks
1279 * requested were part of a uninitialized extent. We need to
1280 * clear this flag now that we are committed to convert all or
1281 * part of the uninitialized extent to be an initialized
1282 * extent. This is because we need to avoid the combination
1283 * of BH_Unwritten and BH_Mapped flags being simultaneously
1284 * set on the buffer_head.
1285 */
1286 clear_buffer_unwritten(bh);
1287
1288 /*
Mingming Caof5ab0d12008-02-25 15:29:55 -05001289 * New blocks allocate and/or writing to uninitialized extent
1290 * will possibly result in updating i_data, so we take
1291 * the write lock of i_data_sem, and call get_blocks()
1292 * with create == 1 flag.
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001293 */
1294 down_write((&EXT4_I(inode)->i_data_sem));
Mingming Caod2a17632008-07-14 17:52:37 -04001295
1296 /*
1297 * if the caller is from delayed allocation writeout path
1298 * we have already reserved fs blocks for allocation
1299 * let the underlying get_block() function know to
1300 * avoid double accounting
1301 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04001302 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
Mingming Caod2a17632008-07-14 17:52:37 -04001303 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001304 /*
1305 * We need to check for EXT4 here because migrate
1306 * could have changed the inode type in between
1307 */
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001308 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1309 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
Theodore Ts'oc2177052009-05-14 00:58:52 -04001310 bh, flags);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001311 } else {
Theodore Ts'oe4d996c2009-05-12 00:25:28 -04001312 retval = ext4_ind_get_blocks(handle, inode, block,
Theodore Ts'oc2177052009-05-14 00:58:52 -04001313 max_blocks, bh, flags);
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -04001314
1315 if (retval > 0 && buffer_new(bh)) {
1316 /*
1317 * We allocated new blocks which will result in
1318 * i_data's format changing. Force the migrate
1319 * to fail by clearing migrate flags
1320 */
Theodore Ts'o1b9c12f2009-09-17 08:32:22 -04001321 EXT4_I(inode)->i_state &= ~EXT4_STATE_EXT_MIGRATE;
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -04001322 }
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001323 }
Mingming Caod2a17632008-07-14 17:52:37 -04001324
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04001325 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
Mingming Caod2a17632008-07-14 17:52:37 -04001326 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04001327
1328 /*
1329 * Update reserved blocks/metadata blocks after successful
1330 * block allocation which had been deferred till now.
1331 */
1332 if ((retval > 0) && (flags & EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE))
1333 ext4_da_update_reserve_space(inode, retval);
Mingming Caod2a17632008-07-14 17:52:37 -04001334
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -05001335 up_write((&EXT4_I(inode)->i_data_sem));
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001336 if (retval > 0 && buffer_mapped(bh)) {
Theodore Ts'o80e42462009-09-08 08:21:26 -04001337 int ret = check_block_validity(inode, "file system "
1338 "corruption after allocation",
1339 block, bh->b_blocknr, retval);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -04001340 if (ret != 0)
1341 return ret;
1342 }
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001343 return retval;
1344}
1345
Mingming Caof3bd1f32008-08-19 22:16:03 -04001346/* Maximum number of blocks we map for direct IO at once. */
1347#define DIO_MAX_BLOCKS 4096
1348
Eric Sandeen6873fa02008-10-07 00:46:36 -04001349int ext4_get_block(struct inode *inode, sector_t iblock,
1350 struct buffer_head *bh_result, int create)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001351{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -08001352 handle_t *handle = ext4_journal_current_handle();
Jan Kara7fb54092008-02-10 01:08:38 -05001353 int ret = 0, started = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001354 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
Mingming Caof3bd1f32008-08-19 22:16:03 -04001355 int dio_credits;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001356
Jan Kara7fb54092008-02-10 01:08:38 -05001357 if (create && !handle) {
1358 /* Direct IO write... */
1359 if (max_blocks > DIO_MAX_BLOCKS)
1360 max_blocks = DIO_MAX_BLOCKS;
Mingming Caof3bd1f32008-08-19 22:16:03 -04001361 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1362 handle = ext4_journal_start(inode, dio_credits);
Jan Kara7fb54092008-02-10 01:08:38 -05001363 if (IS_ERR(handle)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001364 ret = PTR_ERR(handle);
Jan Kara7fb54092008-02-10 01:08:38 -05001365 goto out;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001366 }
Jan Kara7fb54092008-02-10 01:08:38 -05001367 started = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001368 }
1369
Theodore Ts'o12b7ac12009-05-14 00:57:44 -04001370 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
Theodore Ts'oc2177052009-05-14 00:58:52 -04001371 create ? EXT4_GET_BLOCKS_CREATE : 0);
Jan Kara7fb54092008-02-10 01:08:38 -05001372 if (ret > 0) {
1373 bh_result->b_size = (ret << inode->i_blkbits);
1374 ret = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001375 }
Jan Kara7fb54092008-02-10 01:08:38 -05001376 if (started)
1377 ext4_journal_stop(handle);
1378out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001379 return ret;
1380}
1381
1382/*
1383 * `handle' can be NULL if create is zero
1384 */
Mingming Cao617ba132006-10-11 01:20:53 -07001385struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001386 ext4_lblk_t block, int create, int *errp)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001387{
1388 struct buffer_head dummy;
1389 int fatal = 0, err;
Jan Kara03f5d8b2009-06-09 00:17:05 -04001390 int flags = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001391
1392 J_ASSERT(handle != NULL || create == 0);
1393
1394 dummy.b_state = 0;
1395 dummy.b_blocknr = -1000;
1396 buffer_trace_init(&dummy.b_history);
Theodore Ts'oc2177052009-05-14 00:58:52 -04001397 if (create)
1398 flags |= EXT4_GET_BLOCKS_CREATE;
1399 err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001400 /*
Theodore Ts'oc2177052009-05-14 00:58:52 -04001401 * ext4_get_blocks() returns number of blocks mapped. 0 in
1402 * case of a HOLE.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001403 */
1404 if (err > 0) {
1405 if (err > 1)
1406 WARN_ON(1);
1407 err = 0;
1408 }
1409 *errp = err;
1410 if (!err && buffer_mapped(&dummy)) {
1411 struct buffer_head *bh;
1412 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1413 if (!bh) {
1414 *errp = -EIO;
1415 goto err;
1416 }
1417 if (buffer_new(&dummy)) {
1418 J_ASSERT(create != 0);
Aneesh Kumar K.Vac398492007-10-16 18:38:25 -04001419 J_ASSERT(handle != NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001420
1421 /*
1422 * Now that we do not always journal data, we should
1423 * keep in mind whether this should always journal the
1424 * new buffer as metadata. For now, regular file
Mingming Cao617ba132006-10-11 01:20:53 -07001425 * writes use ext4_get_block instead, so it's not a
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001426 * problem.
1427 */
1428 lock_buffer(bh);
1429 BUFFER_TRACE(bh, "call get_create_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001430 fatal = ext4_journal_get_create_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001431 if (!fatal && !buffer_uptodate(bh)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001432 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001433 set_buffer_uptodate(bh);
1434 }
1435 unlock_buffer(bh);
Frank Mayhar03901312009-01-07 00:06:22 -05001436 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1437 err = ext4_handle_dirty_metadata(handle, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001438 if (!fatal)
1439 fatal = err;
1440 } else {
1441 BUFFER_TRACE(bh, "not a new buffer");
1442 }
1443 if (fatal) {
1444 *errp = fatal;
1445 brelse(bh);
1446 bh = NULL;
1447 }
1448 return bh;
1449 }
1450err:
1451 return NULL;
1452}
1453
Mingming Cao617ba132006-10-11 01:20:53 -07001454struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001455 ext4_lblk_t block, int create, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001456{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001457 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001458
Mingming Cao617ba132006-10-11 01:20:53 -07001459 bh = ext4_getblk(handle, inode, block, create, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001460 if (!bh)
1461 return bh;
1462 if (buffer_uptodate(bh))
1463 return bh;
1464 ll_rw_block(READ_META, 1, &bh);
1465 wait_on_buffer(bh);
1466 if (buffer_uptodate(bh))
1467 return bh;
1468 put_bh(bh);
1469 *err = -EIO;
1470 return NULL;
1471}
1472
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001473static int walk_page_buffers(handle_t *handle,
1474 struct buffer_head *head,
1475 unsigned from,
1476 unsigned to,
1477 int *partial,
1478 int (*fn)(handle_t *handle,
1479 struct buffer_head *bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001480{
1481 struct buffer_head *bh;
1482 unsigned block_start, block_end;
1483 unsigned blocksize = head->b_size;
1484 int err, ret = 0;
1485 struct buffer_head *next;
1486
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001487 for (bh = head, block_start = 0;
1488 ret == 0 && (bh != head || !block_start);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001489 block_start = block_end, bh = next) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001490 next = bh->b_this_page;
1491 block_end = block_start + blocksize;
1492 if (block_end <= from || block_start >= to) {
1493 if (partial && !buffer_uptodate(bh))
1494 *partial = 1;
1495 continue;
1496 }
1497 err = (*fn)(handle, bh);
1498 if (!ret)
1499 ret = err;
1500 }
1501 return ret;
1502}
1503
1504/*
1505 * To preserve ordering, it is essential that the hole instantiation and
1506 * the data write be encapsulated in a single transaction. We cannot
Mingming Cao617ba132006-10-11 01:20:53 -07001507 * close off a transaction and start a new one between the ext4_get_block()
Mingming Caodab291a2006-10-11 01:21:01 -07001508 * and the commit_write(). So doing the jbd2_journal_start at the start of
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001509 * prepare_write() is the right place.
1510 *
Mingming Cao617ba132006-10-11 01:20:53 -07001511 * Also, this function can nest inside ext4_writepage() ->
1512 * block_write_full_page(). In that case, we *know* that ext4_writepage()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001513 * has generated enough buffer credits to do the whole page. So we won't
1514 * block on the journal in that case, which is good, because the caller may
1515 * be PF_MEMALLOC.
1516 *
Mingming Cao617ba132006-10-11 01:20:53 -07001517 * By accident, ext4 can be reentered when a transaction is open via
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001518 * quota file writes. If we were to commit the transaction while thus
1519 * reentered, there can be a deadlock - we would be holding a quota
1520 * lock, and the commit would never complete if another thread had a
1521 * transaction open and was blocking on the quota lock - a ranking
1522 * violation.
1523 *
Mingming Caodab291a2006-10-11 01:21:01 -07001524 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001525 * will _not_ run commit under these circumstances because handle->h_ref
1526 * is elevated. We'll still have enough credits for the tiny quotafile
1527 * write.
1528 */
1529static int do_journal_get_write_access(handle_t *handle,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001530 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001531{
1532 if (!buffer_mapped(bh) || buffer_freed(bh))
1533 return 0;
Mingming Cao617ba132006-10-11 01:20:53 -07001534 return ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001535}
1536
Nick Pigginbfc1af62007-10-16 01:25:05 -07001537static int ext4_write_begin(struct file *file, struct address_space *mapping,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001538 loff_t pos, unsigned len, unsigned flags,
1539 struct page **pagep, void **fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001540{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001541 struct inode *inode = mapping->host;
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001542 int ret, needed_blocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001543 handle_t *handle;
1544 int retries = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001545 struct page *page;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001546 pgoff_t index;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001547 unsigned from, to;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001548
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001549 trace_ext4_write_begin(inode, pos, len, flags);
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001550 /*
1551 * Reserve one block more for addition to orphan list in case
1552 * we allocate blocks but write fails for some reason
1553 */
1554 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001555 index = pos >> PAGE_CACHE_SHIFT;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001556 from = pos & (PAGE_CACHE_SIZE - 1);
1557 to = from + len;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001558
1559retry:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001560 handle = ext4_journal_start(inode, needed_blocks);
1561 if (IS_ERR(handle)) {
1562 ret = PTR_ERR(handle);
1563 goto out;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001564 }
1565
Jan Karaebd36102009-02-22 21:09:59 -05001566 /* We cannot recurse into the filesystem as the transaction is already
1567 * started */
1568 flags |= AOP_FLAG_NOFS;
1569
Nick Piggin54566b22009-01-04 12:00:53 -08001570 page = grab_cache_page_write_begin(mapping, index, flags);
Jan Karacf108bc2008-07-11 19:27:31 -04001571 if (!page) {
1572 ext4_journal_stop(handle);
1573 ret = -ENOMEM;
1574 goto out;
1575 }
1576 *pagep = page;
1577
Nick Pigginbfc1af62007-10-16 01:25:05 -07001578 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
Jan Karaebd36102009-02-22 21:09:59 -05001579 ext4_get_block);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001580
1581 if (!ret && ext4_should_journal_data(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001582 ret = walk_page_buffers(handle, page_buffers(page),
1583 from, to, NULL, do_journal_get_write_access);
1584 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001585
1586 if (ret) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001587 unlock_page(page);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001588 page_cache_release(page);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04001589 /*
1590 * block_write_begin may have instantiated a few blocks
1591 * outside i_size. Trim these off again. Don't need
1592 * i_size_read because we hold i_mutex.
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001593 *
1594 * Add inode to orphan list in case we crash before
1595 * truncate finishes
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04001596 */
Jan Karaffacfa72009-07-13 16:22:22 -04001597 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001598 ext4_orphan_add(handle, inode);
1599
1600 ext4_journal_stop(handle);
1601 if (pos + len > inode->i_size) {
Jan Karaffacfa72009-07-13 16:22:22 -04001602 ext4_truncate(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001603 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001604 * If truncate failed early the inode might
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001605 * still be on the orphan list; we need to
1606 * make sure the inode is removed from the
1607 * orphan list in that case.
1608 */
1609 if (inode->i_nlink)
1610 ext4_orphan_del(NULL, inode);
1611 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001612 }
1613
Mingming Cao617ba132006-10-11 01:20:53 -07001614 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001615 goto retry;
Andrew Morton7479d2b2007-04-01 23:49:44 -07001616out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001617 return ret;
1618}
1619
Nick Pigginbfc1af62007-10-16 01:25:05 -07001620/* For write_end() in data=journal mode */
1621static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001622{
1623 if (!buffer_mapped(bh) || buffer_freed(bh))
1624 return 0;
1625 set_buffer_uptodate(bh);
Frank Mayhar03901312009-01-07 00:06:22 -05001626 return ext4_handle_dirty_metadata(handle, NULL, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001627}
1628
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001629static int ext4_generic_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001630 struct address_space *mapping,
1631 loff_t pos, unsigned len, unsigned copied,
1632 struct page *page, void *fsdata)
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001633{
1634 int i_size_changed = 0;
1635 struct inode *inode = mapping->host;
1636 handle_t *handle = ext4_journal_current_handle();
1637
1638 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1639
1640 /*
1641 * No need to use i_size_read() here, the i_size
1642 * cannot change under us because we hold i_mutex.
1643 *
1644 * But it's important to update i_size while still holding page lock:
1645 * page writeout could otherwise come in and zero beyond i_size.
1646 */
1647 if (pos + copied > inode->i_size) {
1648 i_size_write(inode, pos + copied);
1649 i_size_changed = 1;
1650 }
1651
1652 if (pos + copied > EXT4_I(inode)->i_disksize) {
1653 /* We need to mark inode dirty even if
1654 * new_i_size is less that inode->i_size
1655 * bu greater than i_disksize.(hint delalloc)
1656 */
1657 ext4_update_i_disksize(inode, (pos + copied));
1658 i_size_changed = 1;
1659 }
1660 unlock_page(page);
1661 page_cache_release(page);
1662
1663 /*
1664 * Don't mark the inode dirty under page lock. First, it unnecessarily
1665 * makes the holding time of page lock longer. Second, it forces lock
1666 * ordering of page lock and transaction start for journaling
1667 * filesystems.
1668 */
1669 if (i_size_changed)
1670 ext4_mark_inode_dirty(handle, inode);
1671
1672 return copied;
1673}
1674
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001675/*
1676 * We need to pick up the new inode size which generic_commit_write gave us
1677 * `file' can be NULL - eg, when called from page_symlink().
1678 *
Mingming Cao617ba132006-10-11 01:20:53 -07001679 * ext4 never places buffers on inode->i_mapping->private_list. metadata
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001680 * buffers are managed internally.
1681 */
Nick Pigginbfc1af62007-10-16 01:25:05 -07001682static int ext4_ordered_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001683 struct address_space *mapping,
1684 loff_t pos, unsigned len, unsigned copied,
1685 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001686{
Mingming Cao617ba132006-10-11 01:20:53 -07001687 handle_t *handle = ext4_journal_current_handle();
Jan Karacf108bc2008-07-11 19:27:31 -04001688 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001689 int ret = 0, ret2;
1690
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001691 trace_ext4_ordered_write_end(inode, pos, len, copied);
Jan Kara678aaf42008-07-11 19:27:31 -04001692 ret = ext4_jbd2_file_inode(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001693
1694 if (ret == 0) {
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001695 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001696 page, fsdata);
Roel Kluinf8a87d82008-04-29 22:01:18 -04001697 copied = ret2;
Jan Karaffacfa72009-07-13 16:22:22 -04001698 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001699 /* if we have allocated more blocks and copied
1700 * less. We will have blocks allocated outside
1701 * inode->i_size. So truncate them
1702 */
1703 ext4_orphan_add(handle, inode);
Roel Kluinf8a87d82008-04-29 22:01:18 -04001704 if (ret2 < 0)
1705 ret = ret2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001706 }
Mingming Cao617ba132006-10-11 01:20:53 -07001707 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001708 if (!ret)
1709 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001710
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001711 if (pos + len > inode->i_size) {
Jan Karaffacfa72009-07-13 16:22:22 -04001712 ext4_truncate(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001713 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001714 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001715 * on the orphan list; we need to make sure the inode
1716 * is removed from the orphan list in that case.
1717 */
1718 if (inode->i_nlink)
1719 ext4_orphan_del(NULL, inode);
1720 }
1721
1722
Nick Pigginbfc1af62007-10-16 01:25:05 -07001723 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001724}
1725
Nick Pigginbfc1af62007-10-16 01:25:05 -07001726static int ext4_writeback_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001727 struct address_space *mapping,
1728 loff_t pos, unsigned len, unsigned copied,
1729 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001730{
Mingming Cao617ba132006-10-11 01:20:53 -07001731 handle_t *handle = ext4_journal_current_handle();
Jan Karacf108bc2008-07-11 19:27:31 -04001732 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001733 int ret = 0, ret2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001734
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001735 trace_ext4_writeback_write_end(inode, pos, len, copied);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001736 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001737 page, fsdata);
Roel Kluinf8a87d82008-04-29 22:01:18 -04001738 copied = ret2;
Jan Karaffacfa72009-07-13 16:22:22 -04001739 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001740 /* if we have allocated more blocks and copied
1741 * less. We will have blocks allocated outside
1742 * inode->i_size. So truncate them
1743 */
1744 ext4_orphan_add(handle, inode);
1745
Roel Kluinf8a87d82008-04-29 22:01:18 -04001746 if (ret2 < 0)
1747 ret = ret2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001748
Mingming Cao617ba132006-10-11 01:20:53 -07001749 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001750 if (!ret)
1751 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001752
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001753 if (pos + len > inode->i_size) {
Jan Karaffacfa72009-07-13 16:22:22 -04001754 ext4_truncate(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001755 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001756 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001757 * on the orphan list; we need to make sure the inode
1758 * is removed from the orphan list in that case.
1759 */
1760 if (inode->i_nlink)
1761 ext4_orphan_del(NULL, inode);
1762 }
1763
Nick Pigginbfc1af62007-10-16 01:25:05 -07001764 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001765}
1766
Nick Pigginbfc1af62007-10-16 01:25:05 -07001767static int ext4_journalled_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001768 struct address_space *mapping,
1769 loff_t pos, unsigned len, unsigned copied,
1770 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001771{
Mingming Cao617ba132006-10-11 01:20:53 -07001772 handle_t *handle = ext4_journal_current_handle();
Nick Pigginbfc1af62007-10-16 01:25:05 -07001773 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001774 int ret = 0, ret2;
1775 int partial = 0;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001776 unsigned from, to;
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001777 loff_t new_i_size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001778
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001779 trace_ext4_journalled_write_end(inode, pos, len, copied);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001780 from = pos & (PAGE_CACHE_SIZE - 1);
1781 to = from + len;
1782
1783 if (copied < len) {
1784 if (!PageUptodate(page))
1785 copied = 0;
1786 page_zero_new_buffers(page, from+copied, to);
1787 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001788
1789 ret = walk_page_buffers(handle, page_buffers(page), from,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001790 to, &partial, write_end_fn);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001791 if (!partial)
1792 SetPageUptodate(page);
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001793 new_i_size = pos + copied;
1794 if (new_i_size > inode->i_size)
Nick Pigginbfc1af62007-10-16 01:25:05 -07001795 i_size_write(inode, pos+copied);
Mingming Cao617ba132006-10-11 01:20:53 -07001796 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001797 if (new_i_size > EXT4_I(inode)->i_disksize) {
1798 ext4_update_i_disksize(inode, new_i_size);
Mingming Cao617ba132006-10-11 01:20:53 -07001799 ret2 = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001800 if (!ret)
1801 ret = ret2;
1802 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001803
Jan Karacf108bc2008-07-11 19:27:31 -04001804 unlock_page(page);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001805 page_cache_release(page);
Jan Karaffacfa72009-07-13 16:22:22 -04001806 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001807 /* if we have allocated more blocks and copied
1808 * less. We will have blocks allocated outside
1809 * inode->i_size. So truncate them
1810 */
1811 ext4_orphan_add(handle, inode);
1812
Mingming Cao617ba132006-10-11 01:20:53 -07001813 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001814 if (!ret)
1815 ret = ret2;
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001816 if (pos + len > inode->i_size) {
Jan Karaffacfa72009-07-13 16:22:22 -04001817 ext4_truncate(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001818 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001819 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001820 * on the orphan list; we need to make sure the inode
1821 * is removed from the orphan list in that case.
1822 */
1823 if (inode->i_nlink)
1824 ext4_orphan_del(NULL, inode);
1825 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001826
1827 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001828}
Mingming Caod2a17632008-07-14 17:52:37 -04001829
1830static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1831{
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001832 int retries = 0;
Mingming Cao60e58e02009-01-22 18:13:05 +01001833 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1834 unsigned long md_needed, mdblocks, total = 0;
Mingming Caod2a17632008-07-14 17:52:37 -04001835
1836 /*
1837 * recalculate the amount of metadata blocks to reserve
1838 * in order to allocate nrblocks
1839 * worse case is one extent per block
1840 */
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001841repeat:
Mingming Caod2a17632008-07-14 17:52:37 -04001842 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1843 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1844 mdblocks = ext4_calc_metadata_amount(inode, total);
1845 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1846
1847 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1848 total = md_needed + nrblocks;
1849
Mingming Cao60e58e02009-01-22 18:13:05 +01001850 /*
1851 * Make quota reservation here to prevent quota overflow
1852 * later. Real quota accounting is done at pages writeout
1853 * time.
1854 */
1855 if (vfs_dq_reserve_block(inode, total)) {
1856 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1857 return -EDQUOT;
1858 }
1859
Aneesh Kumar K.Va30d5422008-10-09 10:56:23 -04001860 if (ext4_claim_free_blocks(sbi, total)) {
Mingming Caod2a17632008-07-14 17:52:37 -04001861 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao9f0ccfd2009-09-28 15:49:52 -04001862 vfs_dq_release_reservation_block(inode, total);
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001863 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1864 yield();
1865 goto repeat;
1866 }
Mingming Caod2a17632008-07-14 17:52:37 -04001867 return -ENOSPC;
1868 }
Mingming Caod2a17632008-07-14 17:52:37 -04001869 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1870 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1871
1872 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1873 return 0; /* success */
1874}
1875
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001876static void ext4_da_release_space(struct inode *inode, int to_free)
Mingming Caod2a17632008-07-14 17:52:37 -04001877{
1878 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1879 int total, mdb, mdb_free, release;
1880
Mingming Caocd213222008-08-19 22:16:59 -04001881 if (!to_free)
1882 return; /* Nothing to release, exit */
1883
Mingming Caod2a17632008-07-14 17:52:37 -04001884 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Caocd213222008-08-19 22:16:59 -04001885
1886 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1887 /*
1888 * if there is no reserved blocks, but we try to free some
1889 * then the counter is messed up somewhere.
1890 * but since this function is called from invalidate
1891 * page, it's harmless to return without any action
1892 */
1893 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1894 "blocks for inode %lu, but there is no reserved "
1895 "data blocks\n", to_free, inode->i_ino);
1896 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1897 return;
1898 }
1899
Mingming Caod2a17632008-07-14 17:52:37 -04001900 /* recalculate the number of metablocks still need to be reserved */
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001901 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
Mingming Caod2a17632008-07-14 17:52:37 -04001902 mdb = ext4_calc_metadata_amount(inode, total);
1903
1904 /* figure out how many metablocks to release */
1905 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1906 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1907
Mingming Caod2a17632008-07-14 17:52:37 -04001908 release = to_free + mdb_free;
1909
Aneesh Kumar K.V6bc6e632008-10-10 09:39:00 -04001910 /* update fs dirty blocks counter for truncate case */
1911 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
Mingming Caod2a17632008-07-14 17:52:37 -04001912
1913 /* update per-inode reservations */
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001914 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1915 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
Mingming Caod2a17632008-07-14 17:52:37 -04001916
1917 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1918 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
Mingming Caod2a17632008-07-14 17:52:37 -04001919 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao60e58e02009-01-22 18:13:05 +01001920
1921 vfs_dq_release_reservation_block(inode, release);
Mingming Caod2a17632008-07-14 17:52:37 -04001922}
1923
1924static void ext4_da_page_release_reservation(struct page *page,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001925 unsigned long offset)
Mingming Caod2a17632008-07-14 17:52:37 -04001926{
1927 int to_release = 0;
1928 struct buffer_head *head, *bh;
1929 unsigned int curr_off = 0;
1930
1931 head = page_buffers(page);
1932 bh = head;
1933 do {
1934 unsigned int next_off = curr_off + bh->b_size;
1935
1936 if ((offset <= curr_off) && (buffer_delay(bh))) {
1937 to_release++;
1938 clear_buffer_delay(bh);
1939 }
1940 curr_off = next_off;
1941 } while ((bh = bh->b_this_page) != head);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001942 ext4_da_release_space(page->mapping->host, to_release);
Mingming Caod2a17632008-07-14 17:52:37 -04001943}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001944
1945/*
Alex Tomas64769242008-07-11 19:27:31 -04001946 * Delayed allocation stuff
1947 */
1948
Alex Tomas64769242008-07-11 19:27:31 -04001949/*
1950 * mpage_da_submit_io - walks through extent of pages and try to write
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001951 * them with writepage() call back
Alex Tomas64769242008-07-11 19:27:31 -04001952 *
1953 * @mpd->inode: inode
1954 * @mpd->first_page: first page of the extent
1955 * @mpd->next_page: page after the last page of the extent
Alex Tomas64769242008-07-11 19:27:31 -04001956 *
1957 * By the time mpage_da_submit_io() is called we expect all blocks
1958 * to be allocated. this may be wrong if allocation failed.
1959 *
1960 * As pages are already locked by write_cache_pages(), we can't use it
1961 */
1962static int mpage_da_submit_io(struct mpage_da_data *mpd)
1963{
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04001964 long pages_skipped;
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001965 struct pagevec pvec;
1966 unsigned long index, end;
1967 int ret = 0, err, nr_pages, i;
1968 struct inode *inode = mpd->inode;
1969 struct address_space *mapping = inode->i_mapping;
Alex Tomas64769242008-07-11 19:27:31 -04001970
1971 BUG_ON(mpd->next_page <= mpd->first_page);
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001972 /*
1973 * We need to start from the first_page to the next_page - 1
1974 * to make sure we also write the mapped dirty buffer_heads.
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001975 * If we look at mpd->b_blocknr we would only be looking
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001976 * at the currently mapped buffer_heads.
1977 */
Alex Tomas64769242008-07-11 19:27:31 -04001978 index = mpd->first_page;
1979 end = mpd->next_page - 1;
1980
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001981 pagevec_init(&pvec, 0);
Alex Tomas64769242008-07-11 19:27:31 -04001982 while (index <= end) {
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001983 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
Alex Tomas64769242008-07-11 19:27:31 -04001984 if (nr_pages == 0)
1985 break;
1986 for (i = 0; i < nr_pages; i++) {
1987 struct page *page = pvec.pages[i];
1988
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001989 index = page->index;
1990 if (index > end)
1991 break;
1992 index++;
1993
1994 BUG_ON(!PageLocked(page));
1995 BUG_ON(PageWriteback(page));
1996
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04001997 pages_skipped = mpd->wbc->pages_skipped;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001998 err = mapping->a_ops->writepage(page, mpd->wbc);
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04001999 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
2000 /*
2001 * have successfully written the page
2002 * without skipping the same
2003 */
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002004 mpd->pages_written++;
Alex Tomas64769242008-07-11 19:27:31 -04002005 /*
2006 * In error case, we have to continue because
2007 * remaining pages are still locked
2008 * XXX: unlock and re-dirty them?
2009 */
2010 if (ret == 0)
2011 ret = err;
2012 }
2013 pagevec_release(&pvec);
2014 }
Alex Tomas64769242008-07-11 19:27:31 -04002015 return ret;
2016}
2017
2018/*
2019 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2020 *
2021 * @mpd->inode - inode to walk through
2022 * @exbh->b_blocknr - first block on a disk
2023 * @exbh->b_size - amount of space in bytes
2024 * @logical - first logical block to start assignment with
2025 *
2026 * the function goes through all passed space and put actual disk
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002027 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
Alex Tomas64769242008-07-11 19:27:31 -04002028 */
2029static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
2030 struct buffer_head *exbh)
2031{
2032 struct inode *inode = mpd->inode;
2033 struct address_space *mapping = inode->i_mapping;
2034 int blocks = exbh->b_size >> inode->i_blkbits;
2035 sector_t pblock = exbh->b_blocknr, cur_logical;
2036 struct buffer_head *head, *bh;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002037 pgoff_t index, end;
Alex Tomas64769242008-07-11 19:27:31 -04002038 struct pagevec pvec;
2039 int nr_pages, i;
2040
2041 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2042 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2043 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2044
2045 pagevec_init(&pvec, 0);
2046
2047 while (index <= end) {
2048 /* XXX: optimize tail */
2049 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2050 if (nr_pages == 0)
2051 break;
2052 for (i = 0; i < nr_pages; i++) {
2053 struct page *page = pvec.pages[i];
2054
2055 index = page->index;
2056 if (index > end)
2057 break;
2058 index++;
2059
2060 BUG_ON(!PageLocked(page));
2061 BUG_ON(PageWriteback(page));
2062 BUG_ON(!page_has_buffers(page));
2063
2064 bh = page_buffers(page);
2065 head = bh;
2066
2067 /* skip blocks out of the range */
2068 do {
2069 if (cur_logical >= logical)
2070 break;
2071 cur_logical++;
2072 } while ((bh = bh->b_this_page) != head);
2073
2074 do {
2075 if (cur_logical >= logical + blocks)
2076 break;
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002077
2078 if (buffer_delay(bh) ||
2079 buffer_unwritten(bh)) {
2080
2081 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
2082
2083 if (buffer_delay(bh)) {
2084 clear_buffer_delay(bh);
2085 bh->b_blocknr = pblock;
2086 } else {
2087 /*
2088 * unwritten already should have
2089 * blocknr assigned. Verify that
2090 */
2091 clear_buffer_unwritten(bh);
2092 BUG_ON(bh->b_blocknr != pblock);
2093 }
2094
Mingming Cao61628a32008-07-11 19:27:31 -04002095 } else if (buffer_mapped(bh))
Alex Tomas64769242008-07-11 19:27:31 -04002096 BUG_ON(bh->b_blocknr != pblock);
Alex Tomas64769242008-07-11 19:27:31 -04002097
2098 cur_logical++;
2099 pblock++;
2100 } while ((bh = bh->b_this_page) != head);
2101 }
2102 pagevec_release(&pvec);
2103 }
2104}
2105
2106
2107/*
2108 * __unmap_underlying_blocks - just a helper function to unmap
2109 * set of blocks described by @bh
2110 */
2111static inline void __unmap_underlying_blocks(struct inode *inode,
2112 struct buffer_head *bh)
2113{
2114 struct block_device *bdev = inode->i_sb->s_bdev;
2115 int blocks, i;
2116
2117 blocks = bh->b_size >> inode->i_blkbits;
2118 for (i = 0; i < blocks; i++)
2119 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
2120}
2121
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002122static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2123 sector_t logical, long blk_cnt)
2124{
2125 int nr_pages, i;
2126 pgoff_t index, end;
2127 struct pagevec pvec;
2128 struct inode *inode = mpd->inode;
2129 struct address_space *mapping = inode->i_mapping;
2130
2131 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2132 end = (logical + blk_cnt - 1) >>
2133 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2134 while (index <= end) {
2135 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2136 if (nr_pages == 0)
2137 break;
2138 for (i = 0; i < nr_pages; i++) {
2139 struct page *page = pvec.pages[i];
2140 index = page->index;
2141 if (index > end)
2142 break;
2143 index++;
2144
2145 BUG_ON(!PageLocked(page));
2146 BUG_ON(PageWriteback(page));
2147 block_invalidatepage(page, 0);
2148 ClearPageUptodate(page);
2149 unlock_page(page);
2150 }
2151 }
2152 return;
2153}
2154
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002155static void ext4_print_free_blocks(struct inode *inode)
2156{
2157 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o16939182009-09-26 17:43:59 -04002158 printk(KERN_CRIT "Total free blocks count %lld\n",
2159 ext4_count_free_blocks(inode->i_sb));
2160 printk(KERN_CRIT "Free/Dirty block details\n");
2161 printk(KERN_CRIT "free_blocks=%lld\n",
2162 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2163 printk(KERN_CRIT "dirty_blocks=%lld\n",
2164 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2165 printk(KERN_CRIT "Block reservation details\n");
2166 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2167 EXT4_I(inode)->i_reserved_data_blocks);
2168 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2169 EXT4_I(inode)->i_reserved_meta_blocks);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002170 return;
2171}
2172
Theodore Ts'ob920c752009-05-14 00:54:29 -04002173/*
Alex Tomas64769242008-07-11 19:27:31 -04002174 * mpage_da_map_blocks - go through given space
2175 *
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002176 * @mpd - bh describing space
Alex Tomas64769242008-07-11 19:27:31 -04002177 *
2178 * The function skips space we know is already mapped to disk blocks.
2179 *
Alex Tomas64769242008-07-11 19:27:31 -04002180 */
Theodore Ts'oed5bde02009-02-23 10:48:07 -05002181static int mpage_da_map_blocks(struct mpage_da_data *mpd)
Alex Tomas64769242008-07-11 19:27:31 -04002182{
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04002183 int err, blks, get_blocks_flags;
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04002184 struct buffer_head new;
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002185 sector_t next = mpd->b_blocknr;
2186 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2187 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2188 handle_t *handle = NULL;
Alex Tomas64769242008-07-11 19:27:31 -04002189
2190 /*
2191 * We consider only non-mapped and non-allocated blocks
2192 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002193 if ((mpd->b_state & (1 << BH_Mapped)) &&
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002194 !(mpd->b_state & (1 << BH_Delay)) &&
2195 !(mpd->b_state & (1 << BH_Unwritten)))
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002196 return 0;
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002197
2198 /*
2199 * If we didn't accumulate anything to write simply return
2200 */
2201 if (!mpd->b_size)
2202 return 0;
2203
2204 handle = ext4_journal_current_handle();
2205 BUG_ON(!handle);
2206
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -04002207 /*
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04002208 * Call ext4_get_blocks() to allocate any delayed allocation
2209 * blocks, or to convert an uninitialized extent to be
2210 * initialized (in the case where we have written into
2211 * one or more preallocated blocks).
2212 *
2213 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2214 * indicate that we are on the delayed allocation path. This
2215 * affects functions in many different parts of the allocation
2216 * call path. This flag exists primarily because we don't
2217 * want to change *many* call functions, so ext4_get_blocks()
2218 * will set the magic i_delalloc_reserved_flag once the
2219 * inode's allocation semaphore is taken.
2220 *
2221 * If the blocks in questions were delalloc blocks, set
2222 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2223 * variables are updated after the blocks have been allocated.
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -04002224 */
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04002225 new.b_state = 0;
2226 get_blocks_flags = (EXT4_GET_BLOCKS_CREATE |
2227 EXT4_GET_BLOCKS_DELALLOC_RESERVE);
2228 if (mpd->b_state & (1 << BH_Delay))
2229 get_blocks_flags |= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE;
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002230 blks = ext4_get_blocks(handle, mpd->inode, next, max_blocks,
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04002231 &new, get_blocks_flags);
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002232 if (blks < 0) {
2233 err = blks;
Theodore Ts'oed5bde02009-02-23 10:48:07 -05002234 /*
2235 * If get block returns with error we simply
2236 * return. Later writepage will redirty the page and
2237 * writepages will find the dirty page again
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002238 */
2239 if (err == -EAGAIN)
2240 return 0;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002241
2242 if (err == -ENOSPC &&
Theodore Ts'oed5bde02009-02-23 10:48:07 -05002243 ext4_count_free_blocks(mpd->inode->i_sb)) {
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002244 mpd->retval = err;
2245 return 0;
2246 }
2247
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002248 /*
Theodore Ts'oed5bde02009-02-23 10:48:07 -05002249 * get block failure will cause us to loop in
2250 * writepages, because a_ops->writepage won't be able
2251 * to make progress. The page will be redirtied by
2252 * writepage and writepages will again try to write
2253 * the same.
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002254 */
Theodore Ts'o16939182009-09-26 17:43:59 -04002255 ext4_msg(mpd->inode->i_sb, KERN_CRIT,
2256 "delayed block allocation failed for inode %lu at "
2257 "logical offset %llu with max blocks %zd with "
2258 "error %d\n", mpd->inode->i_ino,
2259 (unsigned long long) next,
2260 mpd->b_size >> mpd->inode->i_blkbits, err);
2261 printk(KERN_CRIT "This should not happen!! "
2262 "Data will be lost\n");
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04002263 if (err == -ENOSPC) {
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002264 ext4_print_free_blocks(mpd->inode);
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04002265 }
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002266 /* invalidate all the pages */
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002267 ext4_da_block_invalidatepages(mpd, next,
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002268 mpd->b_size >> mpd->inode->i_blkbits);
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002269 return err;
2270 }
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002271 BUG_ON(blks == 0);
2272
2273 new.b_size = (blks << mpd->inode->i_blkbits);
Alex Tomas64769242008-07-11 19:27:31 -04002274
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002275 if (buffer_new(&new))
2276 __unmap_underlying_blocks(mpd->inode, &new);
Alex Tomas64769242008-07-11 19:27:31 -04002277
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002278 /*
2279 * If blocks are delayed marked, we need to
2280 * put actual blocknr and drop delayed bit
2281 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002282 if ((mpd->b_state & (1 << BH_Delay)) ||
2283 (mpd->b_state & (1 << BH_Unwritten)))
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002284 mpage_put_bnr_to_bhs(mpd, next, &new);
2285
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002286 if (ext4_should_order_data(mpd->inode)) {
2287 err = ext4_jbd2_file_inode(handle, mpd->inode);
2288 if (err)
2289 return err;
2290 }
2291
2292 /*
Jan Kara03f5d8b2009-06-09 00:17:05 -04002293 * Update on-disk size along with block allocation.
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04002294 */
2295 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2296 if (disksize > i_size_read(mpd->inode))
2297 disksize = i_size_read(mpd->inode);
2298 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2299 ext4_update_i_disksize(mpd->inode, disksize);
2300 return ext4_mark_inode_dirty(handle, mpd->inode);
2301 }
2302
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002303 return 0;
Alex Tomas64769242008-07-11 19:27:31 -04002304}
2305
Aneesh Kumar K.Vbf068ee2008-08-19 22:16:43 -04002306#define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2307 (1 << BH_Delay) | (1 << BH_Unwritten))
Alex Tomas64769242008-07-11 19:27:31 -04002308
2309/*
2310 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2311 *
2312 * @mpd->lbh - extent of blocks
2313 * @logical - logical number of the block in the file
2314 * @bh - bh of the block (used to access block's state)
2315 *
2316 * the function is used to collect contig. blocks in same state
2317 */
2318static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002319 sector_t logical, size_t b_size,
2320 unsigned long b_state)
Alex Tomas64769242008-07-11 19:27:31 -04002321{
Alex Tomas64769242008-07-11 19:27:31 -04002322 sector_t next;
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002323 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
Alex Tomas64769242008-07-11 19:27:31 -04002324
Mingming Cao525f4ed2008-08-19 22:15:58 -04002325 /* check if thereserved journal credits might overflow */
2326 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
2327 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2328 /*
2329 * With non-extent format we are limited by the journal
2330 * credit available. Total credit needed to insert
2331 * nrblocks contiguous blocks is dependent on the
2332 * nrblocks. So limit nrblocks.
2333 */
2334 goto flush_it;
2335 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2336 EXT4_MAX_TRANS_DATA) {
2337 /*
2338 * Adding the new buffer_head would make it cross the
2339 * allowed limit for which we have journal credit
2340 * reserved. So limit the new bh->b_size
2341 */
2342 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2343 mpd->inode->i_blkbits;
2344 /* we will do mpage_da_submit_io in the next loop */
2345 }
2346 }
Alex Tomas64769242008-07-11 19:27:31 -04002347 /*
2348 * First block in the extent
2349 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002350 if (mpd->b_size == 0) {
2351 mpd->b_blocknr = logical;
2352 mpd->b_size = b_size;
2353 mpd->b_state = b_state & BH_FLAGS;
Alex Tomas64769242008-07-11 19:27:31 -04002354 return;
2355 }
2356
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002357 next = mpd->b_blocknr + nrblocks;
Alex Tomas64769242008-07-11 19:27:31 -04002358 /*
2359 * Can we merge the block to our big extent?
2360 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002361 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2362 mpd->b_size += b_size;
Alex Tomas64769242008-07-11 19:27:31 -04002363 return;
2364 }
2365
Mingming Cao525f4ed2008-08-19 22:15:58 -04002366flush_it:
Alex Tomas64769242008-07-11 19:27:31 -04002367 /*
2368 * We couldn't merge the block to our extent, so we
2369 * need to flush current extent and start new one
2370 */
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002371 if (mpage_da_map_blocks(mpd) == 0)
2372 mpage_da_submit_io(mpd);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002373 mpd->io_done = 1;
2374 return;
Alex Tomas64769242008-07-11 19:27:31 -04002375}
2376
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04002377static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002378{
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04002379 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002380}
2381
Alex Tomas64769242008-07-11 19:27:31 -04002382/*
2383 * __mpage_da_writepage - finds extent of pages and blocks
2384 *
2385 * @page: page to consider
2386 * @wbc: not used, we just follow rules
2387 * @data: context
2388 *
2389 * The function finds extents of pages and scan them for all blocks.
2390 */
2391static int __mpage_da_writepage(struct page *page,
2392 struct writeback_control *wbc, void *data)
2393{
2394 struct mpage_da_data *mpd = data;
2395 struct inode *inode = mpd->inode;
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002396 struct buffer_head *bh, *head;
Alex Tomas64769242008-07-11 19:27:31 -04002397 sector_t logical;
2398
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002399 if (mpd->io_done) {
2400 /*
2401 * Rest of the page in the page_vec
2402 * redirty then and skip then. We will
Anand Gadiyarfd589a82009-07-16 17:13:03 +02002403 * try to write them again after
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002404 * starting a new transaction
2405 */
2406 redirty_page_for_writepage(wbc, page);
2407 unlock_page(page);
2408 return MPAGE_DA_EXTENT_TAIL;
2409 }
Alex Tomas64769242008-07-11 19:27:31 -04002410 /*
2411 * Can we merge this page to current extent?
2412 */
2413 if (mpd->next_page != page->index) {
2414 /*
2415 * Nope, we can't. So, we map non-allocated blocks
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002416 * and start IO on them using writepage()
Alex Tomas64769242008-07-11 19:27:31 -04002417 */
2418 if (mpd->next_page != mpd->first_page) {
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04002419 if (mpage_da_map_blocks(mpd) == 0)
2420 mpage_da_submit_io(mpd);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002421 /*
2422 * skip rest of the page in the page_vec
2423 */
2424 mpd->io_done = 1;
2425 redirty_page_for_writepage(wbc, page);
2426 unlock_page(page);
2427 return MPAGE_DA_EXTENT_TAIL;
Alex Tomas64769242008-07-11 19:27:31 -04002428 }
2429
2430 /*
2431 * Start next extent of pages ...
2432 */
2433 mpd->first_page = page->index;
2434
2435 /*
2436 * ... and blocks
2437 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002438 mpd->b_size = 0;
2439 mpd->b_state = 0;
2440 mpd->b_blocknr = 0;
Alex Tomas64769242008-07-11 19:27:31 -04002441 }
2442
2443 mpd->next_page = page->index + 1;
2444 logical = (sector_t) page->index <<
2445 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2446
2447 if (!page_has_buffers(page)) {
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002448 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2449 (1 << BH_Dirty) | (1 << BH_Uptodate));
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002450 if (mpd->io_done)
2451 return MPAGE_DA_EXTENT_TAIL;
Alex Tomas64769242008-07-11 19:27:31 -04002452 } else {
2453 /*
2454 * Page with regular buffer heads, just add all dirty ones
2455 */
2456 head = page_buffers(page);
2457 bh = head;
2458 do {
2459 BUG_ON(buffer_locked(bh));
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002460 /*
2461 * We need to try to allocate
2462 * unmapped blocks in the same page.
2463 * Otherwise we won't make progress
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002464 * with the page in ext4_writepage
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002465 */
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04002466 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002467 mpage_add_bh_to_extent(mpd, logical,
2468 bh->b_size,
2469 bh->b_state);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002470 if (mpd->io_done)
2471 return MPAGE_DA_EXTENT_TAIL;
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05002472 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2473 /*
2474 * mapped dirty buffer. We need to update
2475 * the b_state because we look at
2476 * b_state in mpage_da_map_blocks. We don't
2477 * update b_size because if we find an
2478 * unmapped buffer_head later we need to
2479 * use the b_state flag of that buffer_head.
2480 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05002481 if (mpd->b_size == 0)
2482 mpd->b_state = bh->b_state & BH_FLAGS;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002483 }
Alex Tomas64769242008-07-11 19:27:31 -04002484 logical++;
2485 } while ((bh = bh->b_this_page) != head);
2486 }
2487
2488 return 0;
2489}
2490
2491/*
Theodore Ts'ob920c752009-05-14 00:54:29 -04002492 * This is a special get_blocks_t callback which is used by
2493 * ext4_da_write_begin(). It will either return mapped block or
2494 * reserve space for a single block.
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002495 *
2496 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2497 * We also have b_blocknr = -1 and b_bdev initialized properly
2498 *
2499 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2500 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2501 * initialized properly.
Alex Tomas64769242008-07-11 19:27:31 -04002502 */
2503static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2504 struct buffer_head *bh_result, int create)
2505{
2506 int ret = 0;
Aneesh Kumar K.V33b98172009-05-12 14:40:37 -04002507 sector_t invalid_block = ~((sector_t) 0xffff);
2508
2509 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2510 invalid_block = ~0;
Alex Tomas64769242008-07-11 19:27:31 -04002511
2512 BUG_ON(create == 0);
2513 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2514
2515 /*
2516 * first, we need to know whether the block is allocated already
2517 * preallocated blocks are unmapped but should treated
2518 * the same as allocated blocks.
2519 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04002520 ret = ext4_get_blocks(NULL, inode, iblock, 1, bh_result, 0);
Mingming Caod2a17632008-07-14 17:52:37 -04002521 if ((ret == 0) && !buffer_delay(bh_result)) {
2522 /* the block isn't (pre)allocated yet, let's reserve space */
Alex Tomas64769242008-07-11 19:27:31 -04002523 /*
2524 * XXX: __block_prepare_write() unmaps passed block,
2525 * is it OK?
2526 */
Mingming Caod2a17632008-07-14 17:52:37 -04002527 ret = ext4_da_reserve_space(inode, 1);
2528 if (ret)
2529 /* not enough space to reserve */
2530 return ret;
2531
Aneesh Kumar K.V33b98172009-05-12 14:40:37 -04002532 map_bh(bh_result, inode->i_sb, invalid_block);
Alex Tomas64769242008-07-11 19:27:31 -04002533 set_buffer_new(bh_result);
2534 set_buffer_delay(bh_result);
2535 } else if (ret > 0) {
2536 bh_result->b_size = (ret << inode->i_blkbits);
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002537 if (buffer_unwritten(bh_result)) {
2538 /* A delayed write to unwritten bh should
2539 * be marked new and mapped. Mapped ensures
2540 * that we don't do get_block multiple times
2541 * when we write to the same offset and new
2542 * ensures that we do proper zero out for
2543 * partial write.
2544 */
Aneesh Kumar K.V9c1ee182009-05-13 18:36:58 -04002545 set_buffer_new(bh_result);
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002546 set_buffer_mapped(bh_result);
2547 }
Alex Tomas64769242008-07-11 19:27:31 -04002548 ret = 0;
2549 }
2550
2551 return ret;
2552}
Mingming Cao61628a32008-07-11 19:27:31 -04002553
Theodore Ts'ob920c752009-05-14 00:54:29 -04002554/*
2555 * This function is used as a standard get_block_t calback function
2556 * when there is no desire to allocate any blocks. It is used as a
2557 * callback function for block_prepare_write(), nobh_writepage(), and
2558 * block_write_full_page(). These functions should only try to map a
2559 * single block at a time.
2560 *
2561 * Since this function doesn't do block allocations even if the caller
2562 * requests it by passing in create=1, it is critically important that
2563 * any caller checks to make sure that any buffer heads are returned
2564 * by this function are either all already mapped or marked for
2565 * delayed allocation before calling nobh_writepage() or
2566 * block_write_full_page(). Otherwise, b_blocknr could be left
2567 * unitialized, and the page write functions will be taken by
2568 * surprise.
2569 */
2570static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002571 struct buffer_head *bh_result, int create)
2572{
2573 int ret = 0;
2574 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2575
Theodore Ts'oa2dc52b2009-05-12 13:51:29 -04002576 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2577
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002578 /*
2579 * we don't want to do block allocation in writepage
2580 * so call get_block_wrap with create = 0
2581 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04002582 ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002583 if (ret > 0) {
2584 bh_result->b_size = (ret << inode->i_blkbits);
2585 ret = 0;
2586 }
2587 return ret;
Mingming Cao61628a32008-07-11 19:27:31 -04002588}
2589
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04002590static int bget_one(handle_t *handle, struct buffer_head *bh)
2591{
2592 get_bh(bh);
2593 return 0;
2594}
2595
2596static int bput_one(handle_t *handle, struct buffer_head *bh)
2597{
2598 put_bh(bh);
2599 return 0;
2600}
2601
2602static int __ext4_journalled_writepage(struct page *page,
2603 struct writeback_control *wbc,
2604 unsigned int len)
2605{
2606 struct address_space *mapping = page->mapping;
2607 struct inode *inode = mapping->host;
2608 struct buffer_head *page_bufs;
2609 handle_t *handle = NULL;
2610 int ret = 0;
2611 int err;
2612
2613 page_bufs = page_buffers(page);
2614 BUG_ON(!page_bufs);
2615 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2616 /* As soon as we unlock the page, it can go away, but we have
2617 * references to buffers so we are safe */
2618 unlock_page(page);
2619
2620 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2621 if (IS_ERR(handle)) {
2622 ret = PTR_ERR(handle);
2623 goto out;
2624 }
2625
2626 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2627 do_journal_get_write_access);
2628
2629 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2630 write_end_fn);
2631 if (ret == 0)
2632 ret = err;
2633 err = ext4_journal_stop(handle);
2634 if (!ret)
2635 ret = err;
2636
2637 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2638 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2639out:
2640 return ret;
2641}
2642
Mingming Cao61628a32008-07-11 19:27:31 -04002643/*
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002644 * Note that we don't need to start a transaction unless we're journaling data
2645 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2646 * need to file the inode to the transaction's list in ordered mode because if
2647 * we are writing back data added by write(), the inode is already there and if
2648 * we are writing back data modified via mmap(), noone guarantees in which
2649 * transaction the data will hit the disk. In case we are journaling data, we
2650 * cannot start transaction directly because transaction start ranks above page
2651 * lock so we have to do some magic.
2652 *
Theodore Ts'ob920c752009-05-14 00:54:29 -04002653 * This function can get called via...
2654 * - ext4_da_writepages after taking page lock (have journal handle)
2655 * - journal_submit_inode_data_buffers (no journal handle)
2656 * - shrink_page_list via pdflush (no journal handle)
2657 * - grab_page_cache when doing write_begin (have journal handle)
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002658 *
2659 * We don't do any block allocation in this function. If we have page with
2660 * multiple blocks we need to write those buffer_heads that are mapped. This
2661 * is important for mmaped based write. So if we do with blocksize 1K
2662 * truncate(f, 1024);
2663 * a = mmap(f, 0, 4096);
2664 * a[0] = 'a';
2665 * truncate(f, 4096);
2666 * we have in the page first buffer_head mapped via page_mkwrite call back
2667 * but other bufer_heads would be unmapped but dirty(dirty done via the
2668 * do_wp_page). So writepage should write the first block. If we modify
2669 * the mmap area beyond 1024 we will again get a page_fault and the
2670 * page_mkwrite callback will do the block allocation and mark the
2671 * buffer_heads mapped.
2672 *
2673 * We redirty the page if we have any buffer_heads that is either delay or
2674 * unwritten in the page.
2675 *
2676 * We can get recursively called as show below.
2677 *
2678 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2679 * ext4_writepage()
2680 *
2681 * But since we don't do any block allocation we should not deadlock.
2682 * Page also have the dirty flag cleared so we don't get recurive page_lock.
Mingming Cao61628a32008-07-11 19:27:31 -04002683 */
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002684static int ext4_writepage(struct page *page,
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04002685 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04002686{
Alex Tomas64769242008-07-11 19:27:31 -04002687 int ret = 0;
Mingming Cao61628a32008-07-11 19:27:31 -04002688 loff_t size;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002689 unsigned int len;
Mingming Cao61628a32008-07-11 19:27:31 -04002690 struct buffer_head *page_bufs;
2691 struct inode *inode = page->mapping->host;
Alex Tomas64769242008-07-11 19:27:31 -04002692
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002693 trace_ext4_writepage(inode, page);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002694 size = i_size_read(inode);
2695 if (page->index == size >> PAGE_CACHE_SHIFT)
2696 len = size & ~PAGE_CACHE_MASK;
2697 else
2698 len = PAGE_CACHE_SIZE;
Alex Tomas64769242008-07-11 19:27:31 -04002699
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002700 if (page_has_buffers(page)) {
Mingming Cao61628a32008-07-11 19:27:31 -04002701 page_bufs = page_buffers(page);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002702 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04002703 ext4_bh_delay_or_unwritten)) {
Mingming Cao61628a32008-07-11 19:27:31 -04002704 /*
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002705 * We don't want to do block allocation
2706 * So redirty the page and return
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -04002707 * We may reach here when we do a journal commit
2708 * via journal_submit_inode_data_buffers.
2709 * If we don't have mapping block we just ignore
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002710 * them. We can also reach here via shrink_page_list
2711 */
2712 redirty_page_for_writepage(wbc, page);
2713 unlock_page(page);
2714 return 0;
2715 }
2716 } else {
2717 /*
2718 * The test for page_has_buffers() is subtle:
2719 * We know the page is dirty but it lost buffers. That means
2720 * that at some moment in time after write_begin()/write_end()
2721 * has been called all buffers have been clean and thus they
2722 * must have been written at least once. So they are all
2723 * mapped and we can happily proceed with mapping them
2724 * and writing the page.
2725 *
2726 * Try to initialize the buffer_heads and check whether
2727 * all are mapped and non delay. We don't want to
2728 * do block allocation here.
2729 */
Aneesh Kumar K.Vb767e782009-06-04 08:06:06 -04002730 ret = block_prepare_write(page, 0, len,
Theodore Ts'ob920c752009-05-14 00:54:29 -04002731 noalloc_get_block_write);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002732 if (!ret) {
2733 page_bufs = page_buffers(page);
2734 /* check whether all are mapped and non delay */
2735 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04002736 ext4_bh_delay_or_unwritten)) {
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002737 redirty_page_for_writepage(wbc, page);
2738 unlock_page(page);
2739 return 0;
2740 }
2741 } else {
2742 /*
2743 * We can't do block allocation here
2744 * so just redity the page and unlock
2745 * and return
Mingming Cao61628a32008-07-11 19:27:31 -04002746 */
Mingming Cao61628a32008-07-11 19:27:31 -04002747 redirty_page_for_writepage(wbc, page);
2748 unlock_page(page);
2749 return 0;
2750 }
Aneesh Kumar K.Ved9b3e32008-11-07 09:06:45 -05002751 /* now mark the buffer_heads as dirty and uptodate */
Aneesh Kumar K.Vb767e782009-06-04 08:06:06 -04002752 block_commit_write(page, 0, len);
Alex Tomas64769242008-07-11 19:27:31 -04002753 }
2754
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002755 if (PageChecked(page) && ext4_should_journal_data(inode)) {
2756 /*
2757 * It's mmapped pagecache. Add buffers and journal it. There
2758 * doesn't seem much point in redirtying the page here.
2759 */
2760 ClearPageChecked(page);
2761 return __ext4_journalled_writepage(page, wbc, len);
2762 }
2763
Alex Tomas64769242008-07-11 19:27:31 -04002764 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
Theodore Ts'ob920c752009-05-14 00:54:29 -04002765 ret = nobh_writepage(page, noalloc_get_block_write, wbc);
Alex Tomas64769242008-07-11 19:27:31 -04002766 else
Theodore Ts'ob920c752009-05-14 00:54:29 -04002767 ret = block_write_full_page(page, noalloc_get_block_write,
2768 wbc);
Alex Tomas64769242008-07-11 19:27:31 -04002769
Alex Tomas64769242008-07-11 19:27:31 -04002770 return ret;
2771}
2772
Mingming Cao61628a32008-07-11 19:27:31 -04002773/*
Mingming Cao525f4ed2008-08-19 22:15:58 -04002774 * This is called via ext4_da_writepages() to
2775 * calulate the total number of credits to reserve to fit
2776 * a single extent allocation into a single transaction,
2777 * ext4_da_writpeages() will loop calling this before
2778 * the block allocation.
Mingming Cao61628a32008-07-11 19:27:31 -04002779 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002780
2781static int ext4_da_writepages_trans_blocks(struct inode *inode)
2782{
2783 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2784
2785 /*
2786 * With non-extent format the journal credit needed to
2787 * insert nrblocks contiguous block is dependent on
2788 * number of contiguous block. So we will limit
2789 * number of contiguous block to a sane value
2790 */
2791 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2792 (max_blocks > EXT4_MAX_TRANS_DATA))
2793 max_blocks = EXT4_MAX_TRANS_DATA;
2794
2795 return ext4_chunk_trans_blocks(inode, max_blocks);
2796}
Mingming Cao61628a32008-07-11 19:27:31 -04002797
Alex Tomas64769242008-07-11 19:27:31 -04002798static int ext4_da_writepages(struct address_space *mapping,
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002799 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04002800{
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002801 pgoff_t index;
2802 int range_whole = 0;
Mingming Cao61628a32008-07-11 19:27:31 -04002803 handle_t *handle = NULL;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002804 struct mpage_da_data mpd;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002805 struct inode *inode = mapping->host;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002806 int no_nrwrite_index_update;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002807 int pages_written = 0;
2808 long pages_skipped;
Theodore Ts'o55138e02009-09-29 13:31:31 -04002809 unsigned int max_pages;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002810 int range_cyclic, cycled = 1, io_done = 0;
Theodore Ts'o55138e02009-09-29 13:31:31 -04002811 int needed_blocks, ret = 0;
2812 long desired_nr_to_write, nr_to_writebump = 0;
Theodore Ts'ode89de62009-08-31 17:00:59 -04002813 loff_t range_start = wbc->range_start;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002814 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
Mingming Cao61628a32008-07-11 19:27:31 -04002815
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002816 trace_ext4_da_writepages(inode, wbc);
Theodore Ts'oba80b102009-01-03 20:03:21 -05002817
Mingming Cao61628a32008-07-11 19:27:31 -04002818 /*
2819 * No pages to write? This is mainly a kludge to avoid starting
2820 * a transaction for special inodes like journal inode on last iput()
2821 * because that could violate lock ordering on umount
2822 */
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002823 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
Mingming Cao61628a32008-07-11 19:27:31 -04002824 return 0;
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002825
2826 /*
2827 * If the filesystem has aborted, it is read-only, so return
2828 * right away instead of dumping stack traces later on that
2829 * will obscure the real source of the problem. We test
Theodore Ts'o4ab2f152009-06-13 10:09:36 -04002830 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002831 * the latter could be true if the filesystem is mounted
2832 * read-only, and in that case, ext4_da_writepages should
2833 * *never* be called, so if that ever happens, we would want
2834 * the stack trace.
2835 */
Theodore Ts'o4ab2f152009-06-13 10:09:36 -04002836 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002837 return -EROFS;
2838
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002839 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2840 range_whole = 1;
Mingming Cao61628a32008-07-11 19:27:31 -04002841
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002842 range_cyclic = wbc->range_cyclic;
2843 if (wbc->range_cyclic) {
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002844 index = mapping->writeback_index;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002845 if (index)
2846 cycled = 0;
2847 wbc->range_start = index << PAGE_CACHE_SHIFT;
2848 wbc->range_end = LLONG_MAX;
2849 wbc->range_cyclic = 0;
2850 } else
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002851 index = wbc->range_start >> PAGE_CACHE_SHIFT;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002852
Theodore Ts'o55138e02009-09-29 13:31:31 -04002853 /*
2854 * This works around two forms of stupidity. The first is in
2855 * the writeback code, which caps the maximum number of pages
2856 * written to be 1024 pages. This is wrong on multiple
2857 * levels; different architectues have a different page size,
2858 * which changes the maximum amount of data which gets
2859 * written. Secondly, 4 megabytes is way too small. XFS
2860 * forces this value to be 16 megabytes by multiplying
2861 * nr_to_write parameter by four, and then relies on its
2862 * allocator to allocate larger extents to make them
2863 * contiguous. Unfortunately this brings us to the second
2864 * stupidity, which is that ext4's mballoc code only allocates
2865 * at most 2048 blocks. So we force contiguous writes up to
2866 * the number of dirty blocks in the inode, or
2867 * sbi->max_writeback_mb_bump whichever is smaller.
2868 */
2869 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2870 if (!range_cyclic && range_whole)
2871 desired_nr_to_write = wbc->nr_to_write * 8;
2872 else
2873 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2874 max_pages);
2875 if (desired_nr_to_write > max_pages)
2876 desired_nr_to_write = max_pages;
2877
2878 if (wbc->nr_to_write < desired_nr_to_write) {
2879 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2880 wbc->nr_to_write = desired_nr_to_write;
2881 }
2882
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002883 mpd.wbc = wbc;
2884 mpd.inode = mapping->host;
2885
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002886 /*
2887 * we don't want write_cache_pages to update
2888 * nr_to_write and writeback_index
2889 */
2890 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2891 wbc->no_nrwrite_index_update = 1;
2892 pages_skipped = wbc->pages_skipped;
2893
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002894retry:
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002895 while (!ret && wbc->nr_to_write > 0) {
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002896
2897 /*
2898 * we insert one extent at a time. So we need
2899 * credit needed for single extent allocation.
2900 * journalled mode is currently not supported
2901 * by delalloc
2902 */
2903 BUG_ON(ext4_should_journal_data(inode));
Mingming Cao525f4ed2008-08-19 22:15:58 -04002904 needed_blocks = ext4_da_writepages_trans_blocks(inode);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002905
Mingming Cao61628a32008-07-11 19:27:31 -04002906 /* start a new transaction*/
2907 handle = ext4_journal_start(inode, needed_blocks);
2908 if (IS_ERR(handle)) {
2909 ret = PTR_ERR(handle);
Theodore Ts'o16939182009-09-26 17:43:59 -04002910 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002911 "%ld pages, ino %lu; err %d\n", __func__,
2912 wbc->nr_to_write, inode->i_ino, ret);
Mingming Cao61628a32008-07-11 19:27:31 -04002913 goto out_writepages;
2914 }
Theodore Ts'of63e6002009-02-23 16:42:39 -05002915
2916 /*
2917 * Now call __mpage_da_writepage to find the next
2918 * contiguous region of logical blocks that need
2919 * blocks to be allocated by ext4. We don't actually
2920 * submit the blocks for I/O here, even though
2921 * write_cache_pages thinks it will, and will set the
2922 * pages as clean for write before calling
2923 * __mpage_da_writepage().
2924 */
2925 mpd.b_size = 0;
2926 mpd.b_state = 0;
2927 mpd.b_blocknr = 0;
2928 mpd.first_page = 0;
2929 mpd.next_page = 0;
2930 mpd.io_done = 0;
2931 mpd.pages_written = 0;
2932 mpd.retval = 0;
2933 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage,
2934 &mpd);
2935 /*
2936 * If we have a contigous extent of pages and we
2937 * haven't done the I/O yet, map the blocks and submit
2938 * them for I/O.
2939 */
2940 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2941 if (mpage_da_map_blocks(&mpd) == 0)
2942 mpage_da_submit_io(&mpd);
2943 mpd.io_done = 1;
2944 ret = MPAGE_DA_EXTENT_TAIL;
2945 }
Theodore Ts'ob3a3ca82009-08-31 23:13:11 -04002946 trace_ext4_da_write_pages(inode, &mpd);
Theodore Ts'of63e6002009-02-23 16:42:39 -05002947 wbc->nr_to_write -= mpd.pages_written;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002948
Mingming Cao61628a32008-07-11 19:27:31 -04002949 ext4_journal_stop(handle);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002950
Eric Sandeen8f64b322009-02-26 00:57:35 -05002951 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002952 /* commit the transaction which would
2953 * free blocks released in the transaction
2954 * and try again
2955 */
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002956 jbd2_journal_force_commit_nested(sbi->s_journal);
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002957 wbc->pages_skipped = pages_skipped;
2958 ret = 0;
2959 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002960 /*
2961 * got one extent now try with
2962 * rest of the pages
2963 */
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002964 pages_written += mpd.pages_written;
2965 wbc->pages_skipped = pages_skipped;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002966 ret = 0;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002967 io_done = 1;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002968 } else if (wbc->nr_to_write)
Mingming Cao61628a32008-07-11 19:27:31 -04002969 /*
2970 * There is no more writeout needed
2971 * or we requested for a noblocking writeout
2972 * and we found the device congested
2973 */
Mingming Cao61628a32008-07-11 19:27:31 -04002974 break;
Mingming Cao61628a32008-07-11 19:27:31 -04002975 }
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002976 if (!io_done && !cycled) {
2977 cycled = 1;
2978 index = 0;
2979 wbc->range_start = index << PAGE_CACHE_SHIFT;
2980 wbc->range_end = mapping->writeback_index - 1;
2981 goto retry;
2982 }
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002983 if (pages_skipped != wbc->pages_skipped)
Theodore Ts'o16939182009-09-26 17:43:59 -04002984 ext4_msg(inode->i_sb, KERN_CRIT,
2985 "This should not happen leaving %s "
2986 "with nr_to_write = %ld ret = %d\n",
2987 __func__, wbc->nr_to_write, ret);
Mingming Cao61628a32008-07-11 19:27:31 -04002988
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002989 /* Update index */
2990 index += pages_written;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002991 wbc->range_cyclic = range_cyclic;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002992 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2993 /*
2994 * set the writeback_index so that range_cyclic
2995 * mode will write it back later
2996 */
2997 mapping->writeback_index = index;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002998
Mingming Cao61628a32008-07-11 19:27:31 -04002999out_writepages:
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04003000 if (!no_nrwrite_index_update)
3001 wbc->no_nrwrite_index_update = 0;
Theodore Ts'o55138e02009-09-29 13:31:31 -04003002 if (wbc->nr_to_write > nr_to_writebump)
3003 wbc->nr_to_write -= nr_to_writebump;
Theodore Ts'ode89de62009-08-31 17:00:59 -04003004 wbc->range_start = range_start;
Theodore Ts'o9bffad12009-06-17 11:48:11 -04003005 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
Mingming Cao61628a32008-07-11 19:27:31 -04003006 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04003007}
3008
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04003009#define FALL_BACK_TO_NONDELALLOC 1
3010static int ext4_nonda_switch(struct super_block *sb)
3011{
3012 s64 free_blocks, dirty_blocks;
3013 struct ext4_sb_info *sbi = EXT4_SB(sb);
3014
3015 /*
3016 * switch to non delalloc mode if we are running low
3017 * on free block. The free block accounting via percpu
Eric Dumazet179f7eb2009-01-06 14:41:04 -08003018 * counters can get slightly wrong with percpu_counter_batch getting
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04003019 * accumulated on each CPU without updating global counters
3020 * Delalloc need an accurate free block accounting. So switch
3021 * to non delalloc when we are near to error range.
3022 */
3023 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3024 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3025 if (2 * free_blocks < 3 * dirty_blocks ||
3026 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3027 /*
3028 * free block count is less that 150% of dirty blocks
3029 * or free blocks is less that watermark
3030 */
3031 return 1;
3032 }
3033 return 0;
3034}
3035
Alex Tomas64769242008-07-11 19:27:31 -04003036static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003037 loff_t pos, unsigned len, unsigned flags,
3038 struct page **pagep, void **fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04003039{
Mingming Caod2a17632008-07-14 17:52:37 -04003040 int ret, retries = 0;
Alex Tomas64769242008-07-11 19:27:31 -04003041 struct page *page;
3042 pgoff_t index;
3043 unsigned from, to;
3044 struct inode *inode = mapping->host;
3045 handle_t *handle;
3046
3047 index = pos >> PAGE_CACHE_SHIFT;
3048 from = pos & (PAGE_CACHE_SIZE - 1);
3049 to = from + len;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04003050
3051 if (ext4_nonda_switch(inode->i_sb)) {
3052 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3053 return ext4_write_begin(file, mapping, pos,
3054 len, flags, pagep, fsdata);
3055 }
3056 *fsdata = (void *)0;
Theodore Ts'o9bffad12009-06-17 11:48:11 -04003057 trace_ext4_da_write_begin(inode, pos, len, flags);
Mingming Caod2a17632008-07-14 17:52:37 -04003058retry:
Alex Tomas64769242008-07-11 19:27:31 -04003059 /*
3060 * With delayed allocation, we don't log the i_disksize update
3061 * if there is delayed block allocation. But we still need
3062 * to journalling the i_disksize update if writes to the end
3063 * of file which has an already mapped buffer.
3064 */
3065 handle = ext4_journal_start(inode, 1);
3066 if (IS_ERR(handle)) {
3067 ret = PTR_ERR(handle);
3068 goto out;
3069 }
Jan Karaebd36102009-02-22 21:09:59 -05003070 /* We cannot recurse into the filesystem as the transaction is already
3071 * started */
3072 flags |= AOP_FLAG_NOFS;
Alex Tomas64769242008-07-11 19:27:31 -04003073
Nick Piggin54566b22009-01-04 12:00:53 -08003074 page = grab_cache_page_write_begin(mapping, index, flags);
Eric Sandeend5a0d4f2008-08-02 18:51:06 -04003075 if (!page) {
3076 ext4_journal_stop(handle);
3077 ret = -ENOMEM;
3078 goto out;
3079 }
Alex Tomas64769242008-07-11 19:27:31 -04003080 *pagep = page;
3081
3082 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
Theodore Ts'ob920c752009-05-14 00:54:29 -04003083 ext4_da_get_block_prep);
Alex Tomas64769242008-07-11 19:27:31 -04003084 if (ret < 0) {
3085 unlock_page(page);
3086 ext4_journal_stop(handle);
3087 page_cache_release(page);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04003088 /*
3089 * block_write_begin may have instantiated a few blocks
3090 * outside i_size. Trim these off again. Don't need
3091 * i_size_read because we hold i_mutex.
3092 */
3093 if (pos + len > inode->i_size)
Jan Karaffacfa72009-07-13 16:22:22 -04003094 ext4_truncate(inode);
Alex Tomas64769242008-07-11 19:27:31 -04003095 }
3096
Mingming Caod2a17632008-07-14 17:52:37 -04003097 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3098 goto retry;
Alex Tomas64769242008-07-11 19:27:31 -04003099out:
3100 return ret;
3101}
3102
Mingming Cao632eaea2008-07-11 19:27:31 -04003103/*
3104 * Check if we should update i_disksize
3105 * when write to the end of file but not require block allocation
3106 */
3107static int ext4_da_should_update_i_disksize(struct page *page,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003108 unsigned long offset)
Mingming Cao632eaea2008-07-11 19:27:31 -04003109{
3110 struct buffer_head *bh;
3111 struct inode *inode = page->mapping->host;
3112 unsigned int idx;
3113 int i;
3114
3115 bh = page_buffers(page);
3116 idx = offset >> inode->i_blkbits;
3117
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003118 for (i = 0; i < idx; i++)
Mingming Cao632eaea2008-07-11 19:27:31 -04003119 bh = bh->b_this_page;
3120
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04003121 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
Mingming Cao632eaea2008-07-11 19:27:31 -04003122 return 0;
3123 return 1;
3124}
3125
Alex Tomas64769242008-07-11 19:27:31 -04003126static int ext4_da_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003127 struct address_space *mapping,
3128 loff_t pos, unsigned len, unsigned copied,
3129 struct page *page, void *fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04003130{
3131 struct inode *inode = mapping->host;
3132 int ret = 0, ret2;
3133 handle_t *handle = ext4_journal_current_handle();
3134 loff_t new_i_size;
Mingming Cao632eaea2008-07-11 19:27:31 -04003135 unsigned long start, end;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04003136 int write_mode = (int)(unsigned long)fsdata;
3137
3138 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3139 if (ext4_should_order_data(inode)) {
3140 return ext4_ordered_write_end(file, mapping, pos,
3141 len, copied, page, fsdata);
3142 } else if (ext4_should_writeback_data(inode)) {
3143 return ext4_writeback_write_end(file, mapping, pos,
3144 len, copied, page, fsdata);
3145 } else {
3146 BUG();
3147 }
3148 }
Mingming Cao632eaea2008-07-11 19:27:31 -04003149
Theodore Ts'o9bffad12009-06-17 11:48:11 -04003150 trace_ext4_da_write_end(inode, pos, len, copied);
Mingming Cao632eaea2008-07-11 19:27:31 -04003151 start = pos & (PAGE_CACHE_SIZE - 1);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003152 end = start + copied - 1;
Alex Tomas64769242008-07-11 19:27:31 -04003153
3154 /*
3155 * generic_write_end() will run mark_inode_dirty() if i_size
3156 * changes. So let's piggyback the i_disksize mark_inode_dirty
3157 * into that.
3158 */
3159
3160 new_i_size = pos + copied;
Mingming Cao632eaea2008-07-11 19:27:31 -04003161 if (new_i_size > EXT4_I(inode)->i_disksize) {
3162 if (ext4_da_should_update_i_disksize(page, end)) {
3163 down_write(&EXT4_I(inode)->i_data_sem);
3164 if (new_i_size > EXT4_I(inode)->i_disksize) {
3165 /*
3166 * Updating i_disksize when extending file
3167 * without needing block allocation
3168 */
3169 if (ext4_should_order_data(inode))
3170 ret = ext4_jbd2_file_inode(handle,
3171 inode);
Alex Tomas64769242008-07-11 19:27:31 -04003172
Mingming Cao632eaea2008-07-11 19:27:31 -04003173 EXT4_I(inode)->i_disksize = new_i_size;
3174 }
3175 up_write(&EXT4_I(inode)->i_data_sem);
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04003176 /* We need to mark inode dirty even if
3177 * new_i_size is less that inode->i_size
3178 * bu greater than i_disksize.(hint delalloc)
3179 */
3180 ext4_mark_inode_dirty(handle, inode);
Alex Tomas64769242008-07-11 19:27:31 -04003181 }
Mingming Cao632eaea2008-07-11 19:27:31 -04003182 }
Alex Tomas64769242008-07-11 19:27:31 -04003183 ret2 = generic_write_end(file, mapping, pos, len, copied,
3184 page, fsdata);
3185 copied = ret2;
3186 if (ret2 < 0)
3187 ret = ret2;
3188 ret2 = ext4_journal_stop(handle);
3189 if (!ret)
3190 ret = ret2;
3191
3192 return ret ? ret : copied;
3193}
3194
3195static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3196{
Alex Tomas64769242008-07-11 19:27:31 -04003197 /*
3198 * Drop reserved blocks
3199 */
3200 BUG_ON(!PageLocked(page));
3201 if (!page_has_buffers(page))
3202 goto out;
3203
Mingming Caod2a17632008-07-14 17:52:37 -04003204 ext4_da_page_release_reservation(page, offset);
Alex Tomas64769242008-07-11 19:27:31 -04003205
3206out:
3207 ext4_invalidatepage(page, offset);
3208
3209 return;
3210}
3211
Theodore Ts'occd25062009-02-26 01:04:07 -05003212/*
3213 * Force all delayed allocation blocks to be allocated for a given inode.
3214 */
3215int ext4_alloc_da_blocks(struct inode *inode)
3216{
Theodore Ts'ofb40ba02009-09-16 19:30:40 -04003217 trace_ext4_alloc_da_blocks(inode);
3218
Theodore Ts'occd25062009-02-26 01:04:07 -05003219 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3220 !EXT4_I(inode)->i_reserved_meta_blocks)
3221 return 0;
3222
3223 /*
3224 * We do something simple for now. The filemap_flush() will
3225 * also start triggering a write of the data blocks, which is
3226 * not strictly speaking necessary (and for users of
3227 * laptop_mode, not even desirable). However, to do otherwise
3228 * would require replicating code paths in:
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003229 *
Theodore Ts'occd25062009-02-26 01:04:07 -05003230 * ext4_da_writepages() ->
3231 * write_cache_pages() ---> (via passed in callback function)
3232 * __mpage_da_writepage() -->
3233 * mpage_add_bh_to_extent()
3234 * mpage_da_map_blocks()
3235 *
3236 * The problem is that write_cache_pages(), located in
3237 * mm/page-writeback.c, marks pages clean in preparation for
3238 * doing I/O, which is not desirable if we're not planning on
3239 * doing I/O at all.
3240 *
3241 * We could call write_cache_pages(), and then redirty all of
3242 * the pages by calling redirty_page_for_writeback() but that
3243 * would be ugly in the extreme. So instead we would need to
3244 * replicate parts of the code in the above functions,
3245 * simplifying them becuase we wouldn't actually intend to
3246 * write out the pages, but rather only collect contiguous
3247 * logical block extents, call the multi-block allocator, and
3248 * then update the buffer heads with the block allocations.
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003249 *
Theodore Ts'occd25062009-02-26 01:04:07 -05003250 * For now, though, we'll cheat by calling filemap_flush(),
3251 * which will map the blocks, and start the I/O, but not
3252 * actually wait for the I/O to complete.
3253 */
3254 return filemap_flush(inode->i_mapping);
3255}
Alex Tomas64769242008-07-11 19:27:31 -04003256
3257/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003258 * bmap() is special. It gets used by applications such as lilo and by
3259 * the swapper to find the on-disk block of a specific piece of data.
3260 *
3261 * Naturally, this is dangerous if the block concerned is still in the
Mingming Cao617ba132006-10-11 01:20:53 -07003262 * journal. If somebody makes a swapfile on an ext4 data-journaling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003263 * filesystem and enables swap, then they may get a nasty shock when the
3264 * data getting swapped to that swapfile suddenly gets overwritten by
3265 * the original zero's written out previously to the journal and
3266 * awaiting writeback in the kernel's buffer cache.
3267 *
3268 * So, if we see any bmap calls here on a modified, data-journaled file,
3269 * take extra steps to flush any blocks which might be in the cache.
3270 */
Mingming Cao617ba132006-10-11 01:20:53 -07003271static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003272{
3273 struct inode *inode = mapping->host;
3274 journal_t *journal;
3275 int err;
3276
Alex Tomas64769242008-07-11 19:27:31 -04003277 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3278 test_opt(inode->i_sb, DELALLOC)) {
3279 /*
3280 * With delalloc we want to sync the file
3281 * so that we can make sure we allocate
3282 * blocks for file
3283 */
3284 filemap_write_and_wait(mapping);
3285 }
3286
Frank Mayhar03901312009-01-07 00:06:22 -05003287 if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003288 /*
3289 * This is a REALLY heavyweight approach, but the use of
3290 * bmap on dirty files is expected to be extremely rare:
3291 * only if we run lilo or swapon on a freshly made file
3292 * do we expect this to happen.
3293 *
3294 * (bmap requires CAP_SYS_RAWIO so this does not
3295 * represent an unprivileged user DOS attack --- we'd be
3296 * in trouble if mortal users could trigger this path at
3297 * will.)
3298 *
Mingming Cao617ba132006-10-11 01:20:53 -07003299 * NB. EXT4_STATE_JDATA is not set on files other than
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003300 * regular files. If somebody wants to bmap a directory
3301 * or symlink and gets confused because the buffer
3302 * hasn't yet been flushed to disk, they deserve
3303 * everything they get.
3304 */
3305
Mingming Cao617ba132006-10-11 01:20:53 -07003306 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
3307 journal = EXT4_JOURNAL(inode);
Mingming Caodab291a2006-10-11 01:21:01 -07003308 jbd2_journal_lock_updates(journal);
3309 err = jbd2_journal_flush(journal);
3310 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003311
3312 if (err)
3313 return 0;
3314 }
3315
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003316 return generic_block_bmap(mapping, block, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003317}
3318
Mingming Cao617ba132006-10-11 01:20:53 -07003319static int ext4_readpage(struct file *file, struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003320{
Mingming Cao617ba132006-10-11 01:20:53 -07003321 return mpage_readpage(page, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003322}
3323
3324static int
Mingming Cao617ba132006-10-11 01:20:53 -07003325ext4_readpages(struct file *file, struct address_space *mapping,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003326 struct list_head *pages, unsigned nr_pages)
3327{
Mingming Cao617ba132006-10-11 01:20:53 -07003328 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003329}
3330
Mingming Cao617ba132006-10-11 01:20:53 -07003331static void ext4_invalidatepage(struct page *page, unsigned long offset)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003332{
Mingming Cao617ba132006-10-11 01:20:53 -07003333 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003334
3335 /*
3336 * If it's a full truncate we just forget about the pending dirtying
3337 */
3338 if (offset == 0)
3339 ClearPageChecked(page);
3340
Frank Mayhar03901312009-01-07 00:06:22 -05003341 if (journal)
3342 jbd2_journal_invalidatepage(journal, page, offset);
3343 else
3344 block_invalidatepage(page, offset);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003345}
3346
Mingming Cao617ba132006-10-11 01:20:53 -07003347static int ext4_releasepage(struct page *page, gfp_t wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003348{
Mingming Cao617ba132006-10-11 01:20:53 -07003349 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003350
3351 WARN_ON(PageChecked(page));
3352 if (!page_has_buffers(page))
3353 return 0;
Frank Mayhar03901312009-01-07 00:06:22 -05003354 if (journal)
3355 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3356 else
3357 return try_to_free_buffers(page);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003358}
3359
3360/*
Mingming Cao4c0425f2009-09-28 15:48:41 -04003361 * O_DIRECT for ext3 (or indirect map) based files
3362 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003363 * If the O_DIRECT write will extend the file then add this inode to the
3364 * orphan list. So recovery will truncate it back to the original size
3365 * if the machine crashes during the write.
3366 *
3367 * If the O_DIRECT write is intantiating holes inside i_size and the machine
Jan Kara7fb54092008-02-10 01:08:38 -05003368 * crashes then stale disk data _may_ be exposed inside the file. But current
3369 * VFS code falls back into buffered path in that case so we are safe.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003370 */
Mingming Cao4c0425f2009-09-28 15:48:41 -04003371static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003372 const struct iovec *iov, loff_t offset,
3373 unsigned long nr_segs)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003374{
3375 struct file *file = iocb->ki_filp;
3376 struct inode *inode = file->f_mapping->host;
Mingming Cao617ba132006-10-11 01:20:53 -07003377 struct ext4_inode_info *ei = EXT4_I(inode);
Jan Kara7fb54092008-02-10 01:08:38 -05003378 handle_t *handle;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003379 ssize_t ret;
3380 int orphan = 0;
3381 size_t count = iov_length(iov, nr_segs);
Eric Sandeenfbbf6942009-10-02 21:20:55 -04003382 int retries = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003383
3384 if (rw == WRITE) {
3385 loff_t final_size = offset + count;
3386
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003387 if (final_size > inode->i_size) {
Jan Kara7fb54092008-02-10 01:08:38 -05003388 /* Credits for sb + inode write */
3389 handle = ext4_journal_start(inode, 2);
3390 if (IS_ERR(handle)) {
3391 ret = PTR_ERR(handle);
3392 goto out;
3393 }
Mingming Cao617ba132006-10-11 01:20:53 -07003394 ret = ext4_orphan_add(handle, inode);
Jan Kara7fb54092008-02-10 01:08:38 -05003395 if (ret) {
3396 ext4_journal_stop(handle);
3397 goto out;
3398 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003399 orphan = 1;
3400 ei->i_disksize = inode->i_size;
Jan Kara7fb54092008-02-10 01:08:38 -05003401 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003402 }
3403 }
3404
Eric Sandeenfbbf6942009-10-02 21:20:55 -04003405retry:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003406 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3407 offset, nr_segs,
Mingming Cao617ba132006-10-11 01:20:53 -07003408 ext4_get_block, NULL);
Eric Sandeenfbbf6942009-10-02 21:20:55 -04003409 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3410 goto retry;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003411
Jan Kara7fb54092008-02-10 01:08:38 -05003412 if (orphan) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003413 int err;
3414
Jan Kara7fb54092008-02-10 01:08:38 -05003415 /* Credits for sb + inode write */
3416 handle = ext4_journal_start(inode, 2);
3417 if (IS_ERR(handle)) {
3418 /* This is really bad luck. We've written the data
3419 * but cannot extend i_size. Bail out and pretend
3420 * the write failed... */
3421 ret = PTR_ERR(handle);
3422 goto out;
3423 }
3424 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07003425 ext4_orphan_del(handle, inode);
Jan Kara7fb54092008-02-10 01:08:38 -05003426 if (ret > 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003427 loff_t end = offset + ret;
3428 if (end > inode->i_size) {
3429 ei->i_disksize = end;
3430 i_size_write(inode, end);
3431 /*
3432 * We're going to return a positive `ret'
3433 * here due to non-zero-length I/O, so there's
3434 * no way of reporting error returns from
Mingming Cao617ba132006-10-11 01:20:53 -07003435 * ext4_mark_inode_dirty() to userspace. So
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003436 * ignore it.
3437 */
Mingming Cao617ba132006-10-11 01:20:53 -07003438 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003439 }
3440 }
Mingming Cao617ba132006-10-11 01:20:53 -07003441 err = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003442 if (ret == 0)
3443 ret = err;
3444 }
3445out:
3446 return ret;
3447}
3448
Mingming Cao4c0425f2009-09-28 15:48:41 -04003449static int ext4_get_block_dio_write(struct inode *inode, sector_t iblock,
3450 struct buffer_head *bh_result, int create)
3451{
3452 handle_t *handle = NULL;
3453 int ret = 0;
3454 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
3455 int dio_credits;
3456
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003457 ext4_debug("ext4_get_block_dio_write: inode %lu, create flag %d\n",
3458 inode->i_ino, create);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003459 /*
3460 * DIO VFS code passes create = 0 flag for write to
3461 * the middle of file. It does this to avoid block
3462 * allocation for holes, to prevent expose stale data
3463 * out when there is parallel buffered read (which does
3464 * not hold the i_mutex lock) while direct IO write has
3465 * not completed. DIO request on holes finally falls back
3466 * to buffered IO for this reason.
3467 *
3468 * For ext4 extent based file, since we support fallocate,
3469 * new allocated extent as uninitialized, for holes, we
3470 * could fallocate blocks for holes, thus parallel
3471 * buffered IO read will zero out the page when read on
3472 * a hole while parallel DIO write to the hole has not completed.
3473 *
3474 * when we come here, we know it's a direct IO write to
3475 * to the middle of file (<i_size)
3476 * so it's safe to override the create flag from VFS.
3477 */
3478 create = EXT4_GET_BLOCKS_DIO_CREATE_EXT;
3479
3480 if (max_blocks > DIO_MAX_BLOCKS)
3481 max_blocks = DIO_MAX_BLOCKS;
3482 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
3483 handle = ext4_journal_start(inode, dio_credits);
3484 if (IS_ERR(handle)) {
3485 ret = PTR_ERR(handle);
3486 goto out;
3487 }
3488 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
3489 create);
3490 if (ret > 0) {
3491 bh_result->b_size = (ret << inode->i_blkbits);
3492 ret = 0;
3493 }
3494 ext4_journal_stop(handle);
3495out:
3496 return ret;
3497}
3498
Mingming Cao4c0425f2009-09-28 15:48:41 -04003499static void ext4_free_io_end(ext4_io_end_t *io)
3500{
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003501 BUG_ON(!io);
3502 iput(io->inode);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003503 kfree(io);
3504}
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003505static void dump_aio_dio_list(struct inode * inode)
3506{
3507#ifdef EXT4_DEBUG
3508 struct list_head *cur, *before, *after;
3509 ext4_io_end_t *io, *io0, *io1;
3510
3511 if (list_empty(&EXT4_I(inode)->i_aio_dio_complete_list)){
3512 ext4_debug("inode %lu aio dio list is empty\n", inode->i_ino);
3513 return;
3514 }
3515
3516 ext4_debug("Dump inode %lu aio_dio_completed_IO list \n", inode->i_ino);
3517 list_for_each_entry(io, &EXT4_I(inode)->i_aio_dio_complete_list, list){
3518 cur = &io->list;
3519 before = cur->prev;
3520 io0 = container_of(before, ext4_io_end_t, list);
3521 after = cur->next;
3522 io1 = container_of(after, ext4_io_end_t, list);
3523
3524 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3525 io, inode->i_ino, io0, io1);
3526 }
3527#endif
3528}
Mingming Cao4c0425f2009-09-28 15:48:41 -04003529
3530/*
Mingming Cao4c0425f2009-09-28 15:48:41 -04003531 * check a range of space and convert unwritten extents to written.
3532 */
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003533static int ext4_end_aio_dio_nolock(ext4_io_end_t *io)
Mingming Cao4c0425f2009-09-28 15:48:41 -04003534{
Mingming Cao4c0425f2009-09-28 15:48:41 -04003535 struct inode *inode = io->inode;
3536 loff_t offset = io->offset;
3537 size_t size = io->size;
3538 int ret = 0;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003539
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003540 ext4_debug("end_aio_dio_onlock: io 0x%p from inode %lu,list->next 0x%p,"
3541 "list->prev 0x%p\n",
3542 io, inode->i_ino, io->list.next, io->list.prev);
3543
3544 if (list_empty(&io->list))
3545 return ret;
3546
3547 if (io->flag != DIO_AIO_UNWRITTEN)
3548 return ret;
3549
Mingming Cao4c0425f2009-09-28 15:48:41 -04003550 if (offset + size <= i_size_read(inode))
3551 ret = ext4_convert_unwritten_extents(inode, offset, size);
3552
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003553 if (ret < 0) {
Mingming Cao4c0425f2009-09-28 15:48:41 -04003554 printk(KERN_EMERG "%s: failed to convert unwritten"
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003555 "extents to written extents, error is %d"
3556 " io is still on inode %lu aio dio list\n",
3557 __func__, ret, inode->i_ino);
3558 return ret;
3559 }
Mingming Cao4c0425f2009-09-28 15:48:41 -04003560
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003561 /* clear the DIO AIO unwritten flag */
3562 io->flag = 0;
3563 return ret;
3564}
3565/*
3566 * work on completed aio dio IO, to convert unwritten extents to extents
3567 */
3568static void ext4_end_aio_dio_work(struct work_struct *work)
3569{
3570 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
3571 struct inode *inode = io->inode;
3572 int ret = 0;
3573
3574 mutex_lock(&inode->i_mutex);
3575 ret = ext4_end_aio_dio_nolock(io);
3576 if (ret >= 0) {
3577 if (!list_empty(&io->list))
3578 list_del_init(&io->list);
3579 ext4_free_io_end(io);
3580 }
3581 mutex_unlock(&inode->i_mutex);
3582}
3583/*
3584 * This function is called from ext4_sync_file().
3585 *
3586 * When AIO DIO IO is completed, the work to convert unwritten
3587 * extents to written is queued on workqueue but may not get immediately
3588 * scheduled. When fsync is called, we need to ensure the
3589 * conversion is complete before fsync returns.
3590 * The inode keeps track of a list of completed AIO from DIO path
3591 * that might needs to do the conversion. This function walks through
3592 * the list and convert the related unwritten extents to written.
3593 */
3594int flush_aio_dio_completed_IO(struct inode *inode)
3595{
3596 ext4_io_end_t *io;
3597 int ret = 0;
3598 int ret2 = 0;
3599
3600 if (list_empty(&EXT4_I(inode)->i_aio_dio_complete_list))
3601 return ret;
3602
3603 dump_aio_dio_list(inode);
3604 while (!list_empty(&EXT4_I(inode)->i_aio_dio_complete_list)){
3605 io = list_entry(EXT4_I(inode)->i_aio_dio_complete_list.next,
3606 ext4_io_end_t, list);
3607 /*
3608 * Calling ext4_end_aio_dio_nolock() to convert completed
3609 * IO to written.
3610 *
3611 * When ext4_sync_file() is called, run_queue() may already
3612 * about to flush the work corresponding to this io structure.
3613 * It will be upset if it founds the io structure related
3614 * to the work-to-be schedule is freed.
3615 *
3616 * Thus we need to keep the io structure still valid here after
3617 * convertion finished. The io structure has a flag to
3618 * avoid double converting from both fsync and background work
3619 * queue work.
3620 */
3621 ret = ext4_end_aio_dio_nolock(io);
3622 if (ret < 0)
3623 ret2 = ret;
3624 else
3625 list_del_init(&io->list);
3626 }
3627 return (ret2 < 0) ? ret2 : 0;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003628}
3629
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003630static ext4_io_end_t *ext4_init_io_end (struct inode *inode)
Mingming Cao4c0425f2009-09-28 15:48:41 -04003631{
3632 ext4_io_end_t *io = NULL;
3633
3634 io = kmalloc(sizeof(*io), GFP_NOFS);
3635
3636 if (io) {
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003637 igrab(inode);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003638 io->inode = inode;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003639 io->flag = 0;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003640 io->offset = 0;
3641 io->size = 0;
3642 io->error = 0;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003643 INIT_WORK(&io->work, ext4_end_aio_dio_work);
3644 INIT_LIST_HEAD(&io->list);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003645 }
3646
3647 return io;
3648}
3649
3650static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3651 ssize_t size, void *private)
3652{
3653 ext4_io_end_t *io_end = iocb->private;
3654 struct workqueue_struct *wq;
3655
Mingming4b70df12009-11-03 14:44:54 -05003656 /* if not async direct IO or dio with 0 bytes write, just return */
3657 if (!io_end || !size)
3658 return;
3659
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003660 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3661 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3662 iocb->private, io_end->inode->i_ino, iocb, offset,
3663 size);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003664
3665 /* if not aio dio with unwritten extents, just free io and return */
3666 if (io_end->flag != DIO_AIO_UNWRITTEN){
3667 ext4_free_io_end(io_end);
3668 iocb->private = NULL;
3669 return;
3670 }
3671
Mingming Cao4c0425f2009-09-28 15:48:41 -04003672 io_end->offset = offset;
3673 io_end->size = size;
3674 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3675
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003676 /* queue the work to convert unwritten extents to written */
Mingming Cao4c0425f2009-09-28 15:48:41 -04003677 queue_work(wq, &io_end->work);
3678
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003679 /* Add the io_end to per-inode completed aio dio list*/
3680 list_add_tail(&io_end->list,
3681 &EXT4_I(io_end->inode)->i_aio_dio_complete_list);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003682 iocb->private = NULL;
3683}
3684/*
3685 * For ext4 extent files, ext4 will do direct-io write to holes,
3686 * preallocated extents, and those write extend the file, no need to
3687 * fall back to buffered IO.
3688 *
3689 * For holes, we fallocate those blocks, mark them as unintialized
3690 * If those blocks were preallocated, we mark sure they are splited, but
3691 * still keep the range to write as unintialized.
3692 *
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003693 * The unwrritten extents will be converted to written when DIO is completed.
3694 * For async direct IO, since the IO may still pending when return, we
3695 * set up an end_io call back function, which will do the convertion
3696 * when async direct IO completed.
Mingming Cao4c0425f2009-09-28 15:48:41 -04003697 *
3698 * If the O_DIRECT write will extend the file then add this inode to the
3699 * orphan list. So recovery will truncate it back to the original size
3700 * if the machine crashes during the write.
3701 *
3702 */
3703static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3704 const struct iovec *iov, loff_t offset,
3705 unsigned long nr_segs)
3706{
3707 struct file *file = iocb->ki_filp;
3708 struct inode *inode = file->f_mapping->host;
3709 ssize_t ret;
3710 size_t count = iov_length(iov, nr_segs);
3711
3712 loff_t final_size = offset + count;
3713 if (rw == WRITE && final_size <= inode->i_size) {
3714 /*
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003715 * We could direct write to holes and fallocate.
3716 *
3717 * Allocated blocks to fill the hole are marked as uninitialized
Mingming Cao4c0425f2009-09-28 15:48:41 -04003718 * to prevent paralel buffered read to expose the stale data
3719 * before DIO complete the data IO.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003720 *
3721 * As to previously fallocated extents, ext4 get_block
Mingming Cao4c0425f2009-09-28 15:48:41 -04003722 * will just simply mark the buffer mapped but still
3723 * keep the extents uninitialized.
3724 *
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003725 * for non AIO case, we will convert those unwritten extents
3726 * to written after return back from blockdev_direct_IO.
3727 *
3728 * for async DIO, the conversion needs to be defered when
3729 * the IO is completed. The ext4 end_io callback function
3730 * will be called to take care of the conversion work.
3731 * Here for async case, we allocate an io_end structure to
3732 * hook to the iocb.
Mingming Cao4c0425f2009-09-28 15:48:41 -04003733 */
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003734 iocb->private = NULL;
3735 EXT4_I(inode)->cur_aio_dio = NULL;
3736 if (!is_sync_kiocb(iocb)) {
3737 iocb->private = ext4_init_io_end(inode);
3738 if (!iocb->private)
3739 return -ENOMEM;
3740 /*
3741 * we save the io structure for current async
3742 * direct IO, so that later ext4_get_blocks()
3743 * could flag the io structure whether there
3744 * is a unwritten extents needs to be converted
3745 * when IO is completed.
3746 */
3747 EXT4_I(inode)->cur_aio_dio = iocb->private;
3748 }
3749
Mingming Cao4c0425f2009-09-28 15:48:41 -04003750 ret = blockdev_direct_IO(rw, iocb, inode,
3751 inode->i_sb->s_bdev, iov,
3752 offset, nr_segs,
3753 ext4_get_block_dio_write,
3754 ext4_end_io_dio);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003755 if (iocb->private)
3756 EXT4_I(inode)->cur_aio_dio = NULL;
3757 /*
3758 * The io_end structure takes a reference to the inode,
3759 * that structure needs to be destroyed and the
3760 * reference to the inode need to be dropped, when IO is
3761 * complete, even with 0 byte write, or failed.
3762 *
3763 * In the successful AIO DIO case, the io_end structure will be
3764 * desctroyed and the reference to the inode will be dropped
3765 * after the end_io call back function is called.
3766 *
3767 * In the case there is 0 byte write, or error case, since
3768 * VFS direct IO won't invoke the end_io call back function,
3769 * we need to free the end_io structure here.
3770 */
3771 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3772 ext4_free_io_end(iocb->private);
3773 iocb->private = NULL;
Mingming5f524952009-11-10 10:48:04 -05003774 } else if (ret > 0 && (EXT4_I(inode)->i_state &
3775 EXT4_STATE_DIO_UNWRITTEN)) {
Mingming109f5562009-11-10 10:48:08 -05003776 int err;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003777 /*
3778 * for non AIO case, since the IO is already
3779 * completed, we could do the convertion right here
3780 */
Mingming109f5562009-11-10 10:48:08 -05003781 err = ext4_convert_unwritten_extents(inode,
3782 offset, ret);
3783 if (err < 0)
3784 ret = err;
Mingming5f524952009-11-10 10:48:04 -05003785 EXT4_I(inode)->i_state &= ~EXT4_STATE_DIO_UNWRITTEN;
Mingming109f5562009-11-10 10:48:08 -05003786 }
Mingming Cao4c0425f2009-09-28 15:48:41 -04003787 return ret;
3788 }
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003789
3790 /* for write the the end of file case, we fall back to old way */
Mingming Cao4c0425f2009-09-28 15:48:41 -04003791 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3792}
3793
3794static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3795 const struct iovec *iov, loff_t offset,
3796 unsigned long nr_segs)
3797{
3798 struct file *file = iocb->ki_filp;
3799 struct inode *inode = file->f_mapping->host;
3800
3801 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
3802 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3803
3804 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3805}
3806
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003807/*
Mingming Cao617ba132006-10-11 01:20:53 -07003808 * Pages can be marked dirty completely asynchronously from ext4's journalling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003809 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3810 * much here because ->set_page_dirty is called under VFS locks. The page is
3811 * not necessarily locked.
3812 *
3813 * We cannot just dirty the page and leave attached buffers clean, because the
3814 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3815 * or jbddirty because all the journalling code will explode.
3816 *
3817 * So what we do is to mark the page "pending dirty" and next time writepage
3818 * is called, propagate that into the buffers appropriately.
3819 */
Mingming Cao617ba132006-10-11 01:20:53 -07003820static int ext4_journalled_set_page_dirty(struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003821{
3822 SetPageChecked(page);
3823 return __set_page_dirty_nobuffers(page);
3824}
3825
Mingming Cao617ba132006-10-11 01:20:53 -07003826static const struct address_space_operations ext4_ordered_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003827 .readpage = ext4_readpage,
3828 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003829 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003830 .sync_page = block_sync_page,
3831 .write_begin = ext4_write_begin,
3832 .write_end = ext4_ordered_write_end,
3833 .bmap = ext4_bmap,
3834 .invalidatepage = ext4_invalidatepage,
3835 .releasepage = ext4_releasepage,
3836 .direct_IO = ext4_direct_IO,
3837 .migratepage = buffer_migrate_page,
3838 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003839 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003840};
3841
Mingming Cao617ba132006-10-11 01:20:53 -07003842static const struct address_space_operations ext4_writeback_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003843 .readpage = ext4_readpage,
3844 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003845 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003846 .sync_page = block_sync_page,
3847 .write_begin = ext4_write_begin,
3848 .write_end = ext4_writeback_write_end,
3849 .bmap = ext4_bmap,
3850 .invalidatepage = ext4_invalidatepage,
3851 .releasepage = ext4_releasepage,
3852 .direct_IO = ext4_direct_IO,
3853 .migratepage = buffer_migrate_page,
3854 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003855 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003856};
3857
Mingming Cao617ba132006-10-11 01:20:53 -07003858static const struct address_space_operations ext4_journalled_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003859 .readpage = ext4_readpage,
3860 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003861 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003862 .sync_page = block_sync_page,
3863 .write_begin = ext4_write_begin,
3864 .write_end = ext4_journalled_write_end,
3865 .set_page_dirty = ext4_journalled_set_page_dirty,
3866 .bmap = ext4_bmap,
3867 .invalidatepage = ext4_invalidatepage,
3868 .releasepage = ext4_releasepage,
3869 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003870 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003871};
3872
Alex Tomas64769242008-07-11 19:27:31 -04003873static const struct address_space_operations ext4_da_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003874 .readpage = ext4_readpage,
3875 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003876 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003877 .writepages = ext4_da_writepages,
3878 .sync_page = block_sync_page,
3879 .write_begin = ext4_da_write_begin,
3880 .write_end = ext4_da_write_end,
3881 .bmap = ext4_bmap,
3882 .invalidatepage = ext4_da_invalidatepage,
3883 .releasepage = ext4_releasepage,
3884 .direct_IO = ext4_direct_IO,
3885 .migratepage = buffer_migrate_page,
3886 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003887 .error_remove_page = generic_error_remove_page,
Alex Tomas64769242008-07-11 19:27:31 -04003888};
3889
Mingming Cao617ba132006-10-11 01:20:53 -07003890void ext4_set_aops(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003891{
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -04003892 if (ext4_should_order_data(inode) &&
3893 test_opt(inode->i_sb, DELALLOC))
3894 inode->i_mapping->a_ops = &ext4_da_aops;
3895 else if (ext4_should_order_data(inode))
Mingming Cao617ba132006-10-11 01:20:53 -07003896 inode->i_mapping->a_ops = &ext4_ordered_aops;
Alex Tomas64769242008-07-11 19:27:31 -04003897 else if (ext4_should_writeback_data(inode) &&
3898 test_opt(inode->i_sb, DELALLOC))
3899 inode->i_mapping->a_ops = &ext4_da_aops;
Mingming Cao617ba132006-10-11 01:20:53 -07003900 else if (ext4_should_writeback_data(inode))
3901 inode->i_mapping->a_ops = &ext4_writeback_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003902 else
Mingming Cao617ba132006-10-11 01:20:53 -07003903 inode->i_mapping->a_ops = &ext4_journalled_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003904}
3905
3906/*
Mingming Cao617ba132006-10-11 01:20:53 -07003907 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003908 * up to the end of the block which corresponds to `from'.
3909 * This required during truncate. We need to physically zero the tail end
3910 * of that block so it doesn't yield old data if the file is later grown.
3911 */
Jan Karacf108bc2008-07-11 19:27:31 -04003912int ext4_block_truncate_page(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003913 struct address_space *mapping, loff_t from)
3914{
Mingming Cao617ba132006-10-11 01:20:53 -07003915 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003916 unsigned offset = from & (PAGE_CACHE_SIZE-1);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003917 unsigned blocksize, length, pos;
3918 ext4_lblk_t iblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003919 struct inode *inode = mapping->host;
3920 struct buffer_head *bh;
Jan Karacf108bc2008-07-11 19:27:31 -04003921 struct page *page;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003922 int err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003923
Theodore Ts'of4a01012009-07-05 22:08:16 -04003924 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3925 mapping_gfp_mask(mapping) & ~__GFP_FS);
Jan Karacf108bc2008-07-11 19:27:31 -04003926 if (!page)
3927 return -EINVAL;
3928
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003929 blocksize = inode->i_sb->s_blocksize;
3930 length = blocksize - (offset & (blocksize - 1));
3931 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3932
3933 /*
3934 * For "nobh" option, we can only work if we don't need to
3935 * read-in the page - otherwise we create buffers to do the IO.
3936 */
3937 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
Mingming Cao617ba132006-10-11 01:20:53 -07003938 ext4_should_writeback_data(inode) && PageUptodate(page)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08003939 zero_user(page, offset, length);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003940 set_page_dirty(page);
3941 goto unlock;
3942 }
3943
3944 if (!page_has_buffers(page))
3945 create_empty_buffers(page, blocksize, 0);
3946
3947 /* Find the buffer that contains "offset" */
3948 bh = page_buffers(page);
3949 pos = blocksize;
3950 while (offset >= pos) {
3951 bh = bh->b_this_page;
3952 iblock++;
3953 pos += blocksize;
3954 }
3955
3956 err = 0;
3957 if (buffer_freed(bh)) {
3958 BUFFER_TRACE(bh, "freed: skip");
3959 goto unlock;
3960 }
3961
3962 if (!buffer_mapped(bh)) {
3963 BUFFER_TRACE(bh, "unmapped");
Mingming Cao617ba132006-10-11 01:20:53 -07003964 ext4_get_block(inode, iblock, bh, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003965 /* unmapped? It's a hole - nothing to do */
3966 if (!buffer_mapped(bh)) {
3967 BUFFER_TRACE(bh, "still unmapped");
3968 goto unlock;
3969 }
3970 }
3971
3972 /* Ok, it's mapped. Make sure it's up-to-date */
3973 if (PageUptodate(page))
3974 set_buffer_uptodate(bh);
3975
3976 if (!buffer_uptodate(bh)) {
3977 err = -EIO;
3978 ll_rw_block(READ, 1, &bh);
3979 wait_on_buffer(bh);
3980 /* Uhhuh. Read error. Complain and punt. */
3981 if (!buffer_uptodate(bh))
3982 goto unlock;
3983 }
3984
Mingming Cao617ba132006-10-11 01:20:53 -07003985 if (ext4_should_journal_data(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003986 BUFFER_TRACE(bh, "get write access");
Mingming Cao617ba132006-10-11 01:20:53 -07003987 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003988 if (err)
3989 goto unlock;
3990 }
3991
Christoph Lametereebd2aa2008-02-04 22:28:29 -08003992 zero_user(page, offset, length);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003993
3994 BUFFER_TRACE(bh, "zeroed end of block");
3995
3996 err = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07003997 if (ext4_should_journal_data(inode)) {
Frank Mayhar03901312009-01-07 00:06:22 -05003998 err = ext4_handle_dirty_metadata(handle, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003999 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07004000 if (ext4_should_order_data(inode))
Jan Kara678aaf42008-07-11 19:27:31 -04004001 err = ext4_jbd2_file_inode(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004002 mark_buffer_dirty(bh);
4003 }
4004
4005unlock:
4006 unlock_page(page);
4007 page_cache_release(page);
4008 return err;
4009}
4010
4011/*
4012 * Probably it should be a library function... search for first non-zero word
4013 * or memcmp with zero_page, whatever is better for particular architecture.
4014 * Linus?
4015 */
4016static inline int all_zeroes(__le32 *p, __le32 *q)
4017{
4018 while (p < q)
4019 if (*p++)
4020 return 0;
4021 return 1;
4022}
4023
4024/**
Mingming Cao617ba132006-10-11 01:20:53 -07004025 * ext4_find_shared - find the indirect blocks for partial truncation.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004026 * @inode: inode in question
4027 * @depth: depth of the affected branch
Mingming Cao617ba132006-10-11 01:20:53 -07004028 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004029 * @chain: place to store the pointers to partial indirect blocks
4030 * @top: place to the (detached) top of branch
4031 *
Mingming Cao617ba132006-10-11 01:20:53 -07004032 * This is a helper function used by ext4_truncate().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004033 *
4034 * When we do truncate() we may have to clean the ends of several
4035 * indirect blocks but leave the blocks themselves alive. Block is
4036 * partially truncated if some data below the new i_size is refered
4037 * from it (and it is on the path to the first completely truncated
4038 * data block, indeed). We have to free the top of that path along
4039 * with everything to the right of the path. Since no allocation
Mingming Cao617ba132006-10-11 01:20:53 -07004040 * past the truncation point is possible until ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004041 * finishes, we may safely do the latter, but top of branch may
4042 * require special attention - pageout below the truncation point
4043 * might try to populate it.
4044 *
4045 * We atomically detach the top of branch from the tree, store the
4046 * block number of its root in *@top, pointers to buffer_heads of
4047 * partially truncated blocks - in @chain[].bh and pointers to
4048 * their last elements that should not be removed - in
4049 * @chain[].p. Return value is the pointer to last filled element
4050 * of @chain.
4051 *
4052 * The work left to caller to do the actual freeing of subtrees:
4053 * a) free the subtree starting from *@top
4054 * b) free the subtrees whose roots are stored in
4055 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4056 * c) free the subtrees growing from the inode past the @chain[0].
4057 * (no partially truncated stuff there). */
4058
Mingming Cao617ba132006-10-11 01:20:53 -07004059static Indirect *ext4_find_shared(struct inode *inode, int depth,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004060 ext4_lblk_t offsets[4], Indirect chain[4],
4061 __le32 *top)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004062{
4063 Indirect *partial, *p;
4064 int k, err;
4065
4066 *top = 0;
4067 /* Make k index the deepest non-null offest + 1 */
4068 for (k = depth; k > 1 && !offsets[k-1]; k--)
4069 ;
Mingming Cao617ba132006-10-11 01:20:53 -07004070 partial = ext4_get_branch(inode, k, offsets, chain, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004071 /* Writer: pointers */
4072 if (!partial)
4073 partial = chain + k-1;
4074 /*
4075 * If the branch acquired continuation since we've looked at it -
4076 * fine, it should all survive and (new) top doesn't belong to us.
4077 */
4078 if (!partial->key && *partial->p)
4079 /* Writer: end */
4080 goto no_top;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004081 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004082 ;
4083 /*
4084 * OK, we've found the last block that must survive. The rest of our
4085 * branch should be detached before unlocking. However, if that rest
4086 * of branch is all ours and does not grow immediately from the inode
4087 * it's easier to cheat and just decrement partial->p.
4088 */
4089 if (p == chain + k - 1 && p > chain) {
4090 p->p--;
4091 } else {
4092 *top = *p->p;
Mingming Cao617ba132006-10-11 01:20:53 -07004093 /* Nope, don't do this in ext4. Must leave the tree intact */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004094#if 0
4095 *p->p = 0;
4096#endif
4097 }
4098 /* Writer: end */
4099
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004100 while (partial > p) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004101 brelse(partial->bh);
4102 partial--;
4103 }
4104no_top:
4105 return partial;
4106}
4107
4108/*
4109 * Zero a number of block pointers in either an inode or an indirect block.
4110 * If we restart the transaction we must again get write access to the
4111 * indirect block for further modification.
4112 *
4113 * We release `count' blocks on disk, but (last - first) may be greater
4114 * than `count' because there can be holes in there.
4115 */
Mingming Cao617ba132006-10-11 01:20:53 -07004116static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004117 struct buffer_head *bh,
4118 ext4_fsblk_t block_to_free,
4119 unsigned long count, __le32 *first,
4120 __le32 *last)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004121{
4122 __le32 *p;
4123 if (try_to_extend_transaction(handle, inode)) {
4124 if (bh) {
Frank Mayhar03901312009-01-07 00:06:22 -05004125 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4126 ext4_handle_dirty_metadata(handle, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004127 }
Mingming Cao617ba132006-10-11 01:20:53 -07004128 ext4_mark_inode_dirty(handle, inode);
Jan Kara487caee2009-08-17 22:17:20 -04004129 ext4_truncate_restart_trans(handle, inode,
4130 blocks_for_truncate(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004131 if (bh) {
4132 BUFFER_TRACE(bh, "retaking write access");
Mingming Cao617ba132006-10-11 01:20:53 -07004133 ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004134 }
4135 }
4136
4137 /*
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004138 * Any buffers which are on the journal will be in memory. We
4139 * find them on the hash table so jbd2_journal_revoke() will
4140 * run jbd2_journal_forget() on them. We've already detached
4141 * each block from the file, so bforget() in
4142 * jbd2_journal_forget() should be safe.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004143 *
Mingming Caodab291a2006-10-11 01:21:01 -07004144 * AKPM: turn on bforget in jbd2_journal_forget()!!!
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004145 */
4146 for (p = first; p < last; p++) {
4147 u32 nr = le32_to_cpu(*p);
4148 if (nr) {
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05004149 struct buffer_head *tbh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004150
4151 *p = 0;
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05004152 tbh = sb_find_get_block(inode->i_sb, nr);
4153 ext4_forget(handle, 0, inode, tbh, nr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004154 }
4155 }
4156
Alex Tomasc9de5602008-01-29 00:19:52 -05004157 ext4_free_blocks(handle, inode, block_to_free, count, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004158}
4159
4160/**
Mingming Cao617ba132006-10-11 01:20:53 -07004161 * ext4_free_data - free a list of data blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004162 * @handle: handle for this transaction
4163 * @inode: inode we are dealing with
4164 * @this_bh: indirect buffer_head which contains *@first and *@last
4165 * @first: array of block numbers
4166 * @last: points immediately past the end of array
4167 *
4168 * We are freeing all blocks refered from that array (numbers are stored as
4169 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4170 *
4171 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4172 * blocks are contiguous then releasing them at one time will only affect one
4173 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4174 * actually use a lot of journal space.
4175 *
4176 * @this_bh will be %NULL if @first and @last point into the inode's direct
4177 * block pointers.
4178 */
Mingming Cao617ba132006-10-11 01:20:53 -07004179static void ext4_free_data(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004180 struct buffer_head *this_bh,
4181 __le32 *first, __le32 *last)
4182{
Mingming Cao617ba132006-10-11 01:20:53 -07004183 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004184 unsigned long count = 0; /* Number of blocks in the run */
4185 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4186 corresponding to
4187 block_to_free */
Mingming Cao617ba132006-10-11 01:20:53 -07004188 ext4_fsblk_t nr; /* Current block # */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004189 __le32 *p; /* Pointer into inode/ind
4190 for current block */
4191 int err;
4192
4193 if (this_bh) { /* For indirect block */
4194 BUFFER_TRACE(this_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07004195 err = ext4_journal_get_write_access(handle, this_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004196 /* Important: if we can't update the indirect pointers
4197 * to the blocks, we can't free them. */
4198 if (err)
4199 return;
4200 }
4201
4202 for (p = first; p < last; p++) {
4203 nr = le32_to_cpu(*p);
4204 if (nr) {
4205 /* accumulate blocks to free if they're contiguous */
4206 if (count == 0) {
4207 block_to_free = nr;
4208 block_to_free_p = p;
4209 count = 1;
4210 } else if (nr == block_to_free + count) {
4211 count++;
4212 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07004213 ext4_clear_blocks(handle, inode, this_bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004214 block_to_free,
4215 count, block_to_free_p, p);
4216 block_to_free = nr;
4217 block_to_free_p = p;
4218 count = 1;
4219 }
4220 }
4221 }
4222
4223 if (count > 0)
Mingming Cao617ba132006-10-11 01:20:53 -07004224 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004225 count, block_to_free_p, p);
4226
4227 if (this_bh) {
Frank Mayhar03901312009-01-07 00:06:22 -05004228 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
Duane Griffin71dc8fb2008-07-11 19:27:31 -04004229
4230 /*
4231 * The buffer head should have an attached journal head at this
4232 * point. However, if the data is corrupted and an indirect
4233 * block pointed to itself, it would have been detached when
4234 * the block was cleared. Check for this instead of OOPSing.
4235 */
Theodore Ts'oe7f07962009-01-20 09:50:19 -05004236 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
Frank Mayhar03901312009-01-07 00:06:22 -05004237 ext4_handle_dirty_metadata(handle, inode, this_bh);
Duane Griffin71dc8fb2008-07-11 19:27:31 -04004238 else
4239 ext4_error(inode->i_sb, __func__,
4240 "circular indirect block detected, "
4241 "inode=%lu, block=%llu",
4242 inode->i_ino,
4243 (unsigned long long) this_bh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004244 }
4245}
4246
4247/**
Mingming Cao617ba132006-10-11 01:20:53 -07004248 * ext4_free_branches - free an array of branches
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004249 * @handle: JBD handle for this transaction
4250 * @inode: inode we are dealing with
4251 * @parent_bh: the buffer_head which contains *@first and *@last
4252 * @first: array of block numbers
4253 * @last: pointer immediately past the end of array
4254 * @depth: depth of the branches to free
4255 *
4256 * We are freeing all blocks refered from these branches (numbers are
4257 * stored as little-endian 32-bit) and updating @inode->i_blocks
4258 * appropriately.
4259 */
Mingming Cao617ba132006-10-11 01:20:53 -07004260static void ext4_free_branches(handle_t *handle, struct inode *inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004261 struct buffer_head *parent_bh,
4262 __le32 *first, __le32 *last, int depth)
4263{
Mingming Cao617ba132006-10-11 01:20:53 -07004264 ext4_fsblk_t nr;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004265 __le32 *p;
4266
Frank Mayhar03901312009-01-07 00:06:22 -05004267 if (ext4_handle_is_aborted(handle))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004268 return;
4269
4270 if (depth--) {
4271 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -07004272 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004273 p = last;
4274 while (--p >= first) {
4275 nr = le32_to_cpu(*p);
4276 if (!nr)
4277 continue; /* A hole */
4278
4279 /* Go read the buffer for the next level down */
4280 bh = sb_bread(inode->i_sb, nr);
4281
4282 /*
4283 * A read failure? Report error and clear slot
4284 * (should be rare).
4285 */
4286 if (!bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07004287 ext4_error(inode->i_sb, "ext4_free_branches",
Mingming Cao2ae02102006-10-11 01:21:11 -07004288 "Read failure, inode=%lu, block=%llu",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004289 inode->i_ino, nr);
4290 continue;
4291 }
4292
4293 /* This zaps the entire block. Bottom up. */
4294 BUFFER_TRACE(bh, "free child branches");
Mingming Cao617ba132006-10-11 01:20:53 -07004295 ext4_free_branches(handle, inode, bh,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004296 (__le32 *) bh->b_data,
4297 (__le32 *) bh->b_data + addr_per_block,
4298 depth);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004299
4300 /*
4301 * We've probably journalled the indirect block several
4302 * times during the truncate. But it's no longer
4303 * needed and we now drop it from the transaction via
Mingming Caodab291a2006-10-11 01:21:01 -07004304 * jbd2_journal_revoke().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004305 *
4306 * That's easy if it's exclusively part of this
4307 * transaction. But if it's part of the committing
Mingming Caodab291a2006-10-11 01:21:01 -07004308 * transaction then jbd2_journal_forget() will simply
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004309 * brelse() it. That means that if the underlying
Mingming Cao617ba132006-10-11 01:20:53 -07004310 * block is reallocated in ext4_get_block(),
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004311 * unmap_underlying_metadata() will find this block
4312 * and will try to get rid of it. damn, damn.
4313 *
4314 * If this block has already been committed to the
4315 * journal, a revoke record will be written. And
4316 * revoke records must be emitted *before* clearing
4317 * this block's bit in the bitmaps.
4318 */
Mingming Cao617ba132006-10-11 01:20:53 -07004319 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004320
4321 /*
4322 * Everything below this this pointer has been
4323 * released. Now let this top-of-subtree go.
4324 *
4325 * We want the freeing of this indirect block to be
4326 * atomic in the journal with the updating of the
4327 * bitmap block which owns it. So make some room in
4328 * the journal.
4329 *
4330 * We zero the parent pointer *after* freeing its
4331 * pointee in the bitmaps, so if extend_transaction()
4332 * for some reason fails to put the bitmap changes and
4333 * the release into the same transaction, recovery
4334 * will merely complain about releasing a free block,
4335 * rather than leaking blocks.
4336 */
Frank Mayhar03901312009-01-07 00:06:22 -05004337 if (ext4_handle_is_aborted(handle))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004338 return;
4339 if (try_to_extend_transaction(handle, inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004340 ext4_mark_inode_dirty(handle, inode);
Jan Kara487caee2009-08-17 22:17:20 -04004341 ext4_truncate_restart_trans(handle, inode,
4342 blocks_for_truncate(inode));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004343 }
4344
Alex Tomasc9de5602008-01-29 00:19:52 -05004345 ext4_free_blocks(handle, inode, nr, 1, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004346
4347 if (parent_bh) {
4348 /*
4349 * The block which we have just freed is
4350 * pointed to by an indirect block: journal it
4351 */
4352 BUFFER_TRACE(parent_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07004353 if (!ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004354 parent_bh)){
4355 *p = 0;
4356 BUFFER_TRACE(parent_bh,
Frank Mayhar03901312009-01-07 00:06:22 -05004357 "call ext4_handle_dirty_metadata");
4358 ext4_handle_dirty_metadata(handle,
4359 inode,
4360 parent_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004361 }
4362 }
4363 }
4364 } else {
4365 /* We have reached the bottom of the tree. */
4366 BUFFER_TRACE(parent_bh, "free data blocks");
Mingming Cao617ba132006-10-11 01:20:53 -07004367 ext4_free_data(handle, inode, parent_bh, first, last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004368 }
4369}
4370
Duane Griffin91ef4ca2008-07-11 19:27:31 -04004371int ext4_can_truncate(struct inode *inode)
4372{
4373 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4374 return 0;
4375 if (S_ISREG(inode->i_mode))
4376 return 1;
4377 if (S_ISDIR(inode->i_mode))
4378 return 1;
4379 if (S_ISLNK(inode->i_mode))
4380 return !ext4_inode_is_fast_symlink(inode);
4381 return 0;
4382}
4383
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004384/*
Mingming Cao617ba132006-10-11 01:20:53 -07004385 * ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004386 *
Mingming Cao617ba132006-10-11 01:20:53 -07004387 * We block out ext4_get_block() block instantiations across the entire
4388 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004389 * simultaneously on behalf of the same inode.
4390 *
4391 * As we work through the truncate and commmit bits of it to the journal there
4392 * is one core, guiding principle: the file's tree must always be consistent on
4393 * disk. We must be able to restart the truncate after a crash.
4394 *
4395 * The file's tree may be transiently inconsistent in memory (although it
4396 * probably isn't), but whenever we close off and commit a journal transaction,
4397 * the contents of (the filesystem + the journal) must be consistent and
4398 * restartable. It's pretty simple, really: bottom up, right to left (although
4399 * left-to-right works OK too).
4400 *
4401 * Note that at recovery time, journal replay occurs *before* the restart of
4402 * truncate against the orphan inode list.
4403 *
4404 * The committed inode has the new, desired i_size (which is the same as
Mingming Cao617ba132006-10-11 01:20:53 -07004405 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004406 * that this inode's truncate did not complete and it will again call
Mingming Cao617ba132006-10-11 01:20:53 -07004407 * ext4_truncate() to have another go. So there will be instantiated blocks
4408 * to the right of the truncation point in a crashed ext4 filesystem. But
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004409 * that's fine - as long as they are linked from the inode, the post-crash
Mingming Cao617ba132006-10-11 01:20:53 -07004410 * ext4_truncate() run will find them and release them.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004411 */
Mingming Cao617ba132006-10-11 01:20:53 -07004412void ext4_truncate(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004413{
4414 handle_t *handle;
Mingming Cao617ba132006-10-11 01:20:53 -07004415 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004416 __le32 *i_data = ei->i_data;
Mingming Cao617ba132006-10-11 01:20:53 -07004417 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004418 struct address_space *mapping = inode->i_mapping;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004419 ext4_lblk_t offsets[4];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004420 Indirect chain[4];
4421 Indirect *partial;
4422 __le32 nr = 0;
4423 int n;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004424 ext4_lblk_t last_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004425 unsigned blocksize = inode->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004426
Duane Griffin91ef4ca2008-07-11 19:27:31 -04004427 if (!ext4_can_truncate(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004428 return;
4429
Theodore Ts'o5534fb52009-09-17 09:34:16 -04004430 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
Theodore Ts'o7d8f9f72009-02-24 08:21:14 -05004431 ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE;
4432
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05004433 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
Jan Karacf108bc2008-07-11 19:27:31 -04004434 ext4_ext_truncate(inode);
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05004435 return;
4436 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004437
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004438 handle = start_transaction(inode);
Jan Karacf108bc2008-07-11 19:27:31 -04004439 if (IS_ERR(handle))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004440 return; /* AKPM: return what? */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004441
4442 last_block = (inode->i_size + blocksize-1)
Mingming Cao617ba132006-10-11 01:20:53 -07004443 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004444
Jan Karacf108bc2008-07-11 19:27:31 -04004445 if (inode->i_size & (blocksize - 1))
4446 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4447 goto out_stop;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004448
Mingming Cao617ba132006-10-11 01:20:53 -07004449 n = ext4_block_to_path(inode, last_block, offsets, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004450 if (n == 0)
4451 goto out_stop; /* error */
4452
4453 /*
4454 * OK. This truncate is going to happen. We add the inode to the
4455 * orphan list, so that if this truncate spans multiple transactions,
4456 * and we crash, we will resume the truncate when the filesystem
4457 * recovers. It also marks the inode dirty, to catch the new size.
4458 *
4459 * Implication: the file must always be in a sane, consistent
4460 * truncatable state while each transaction commits.
4461 */
Mingming Cao617ba132006-10-11 01:20:53 -07004462 if (ext4_orphan_add(handle, inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004463 goto out_stop;
4464
4465 /*
Mingming Cao632eaea2008-07-11 19:27:31 -04004466 * From here we block out all ext4_get_block() callers who want to
4467 * modify the block allocation tree.
4468 */
4469 down_write(&ei->i_data_sem);
Theodore Ts'ob4df2032008-08-13 21:44:34 -04004470
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004471 ext4_discard_preallocations(inode);
Theodore Ts'ob4df2032008-08-13 21:44:34 -04004472
Mingming Cao632eaea2008-07-11 19:27:31 -04004473 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004474 * The orphan list entry will now protect us from any crash which
4475 * occurs before the truncate completes, so it is now safe to propagate
4476 * the new, shorter inode size (held for now in i_size) into the
4477 * on-disk inode. We do this via i_disksize, which is the value which
Mingming Cao617ba132006-10-11 01:20:53 -07004478 * ext4 *really* writes onto the disk inode.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004479 */
4480 ei->i_disksize = inode->i_size;
4481
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004482 if (n == 1) { /* direct blocks */
Mingming Cao617ba132006-10-11 01:20:53 -07004483 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4484 i_data + EXT4_NDIR_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004485 goto do_indirects;
4486 }
4487
Mingming Cao617ba132006-10-11 01:20:53 -07004488 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004489 /* Kill the top of shared branch (not detached) */
4490 if (nr) {
4491 if (partial == chain) {
4492 /* Shared branch grows from the inode */
Mingming Cao617ba132006-10-11 01:20:53 -07004493 ext4_free_branches(handle, inode, NULL,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004494 &nr, &nr+1, (chain+n-1) - partial);
4495 *partial->p = 0;
4496 /*
4497 * We mark the inode dirty prior to restart,
4498 * and prior to stop. No need for it here.
4499 */
4500 } else {
4501 /* Shared branch grows from an indirect block */
4502 BUFFER_TRACE(partial->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07004503 ext4_free_branches(handle, inode, partial->bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004504 partial->p,
4505 partial->p+1, (chain+n-1) - partial);
4506 }
4507 }
4508 /* Clear the ends of indirect blocks on the shared branch */
4509 while (partial > chain) {
Mingming Cao617ba132006-10-11 01:20:53 -07004510 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004511 (__le32*)partial->bh->b_data+addr_per_block,
4512 (chain+n-1) - partial);
4513 BUFFER_TRACE(partial->bh, "call brelse");
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004514 brelse(partial->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004515 partial--;
4516 }
4517do_indirects:
4518 /* Kill the remaining (whole) subtrees */
4519 switch (offsets[0]) {
4520 default:
Mingming Cao617ba132006-10-11 01:20:53 -07004521 nr = i_data[EXT4_IND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004522 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07004523 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4524 i_data[EXT4_IND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004525 }
Mingming Cao617ba132006-10-11 01:20:53 -07004526 case EXT4_IND_BLOCK:
4527 nr = i_data[EXT4_DIND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004528 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07004529 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4530 i_data[EXT4_DIND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004531 }
Mingming Cao617ba132006-10-11 01:20:53 -07004532 case EXT4_DIND_BLOCK:
4533 nr = i_data[EXT4_TIND_BLOCK];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004534 if (nr) {
Mingming Cao617ba132006-10-11 01:20:53 -07004535 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4536 i_data[EXT4_TIND_BLOCK] = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004537 }
Mingming Cao617ba132006-10-11 01:20:53 -07004538 case EXT4_TIND_BLOCK:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004539 ;
4540 }
4541
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05004542 up_write(&ei->i_data_sem);
Kalpak Shahef7f3832007-07-18 09:15:20 -04004543 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07004544 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004545
4546 /*
4547 * In a multi-transaction truncate, we only make the final transaction
4548 * synchronous
4549 */
4550 if (IS_SYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -05004551 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004552out_stop:
4553 /*
4554 * If this was a simple ftruncate(), and the file will remain alive
4555 * then we need to clear up the orphan record which we created above.
4556 * However, if this was a real unlink then we were called by
Mingming Cao617ba132006-10-11 01:20:53 -07004557 * ext4_delete_inode(), and we allow that function to clean up the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004558 * orphan info for us.
4559 */
4560 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07004561 ext4_orphan_del(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004562
Mingming Cao617ba132006-10-11 01:20:53 -07004563 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004564}
4565
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004566/*
Mingming Cao617ba132006-10-11 01:20:53 -07004567 * ext4_get_inode_loc returns with an extra refcount against the inode's
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004568 * underlying buffer_head on success. If 'in_mem' is true, we have all
4569 * data in memory that is needed to recreate the on-disk version of this
4570 * inode.
4571 */
Mingming Cao617ba132006-10-11 01:20:53 -07004572static int __ext4_get_inode_loc(struct inode *inode,
4573 struct ext4_iloc *iloc, int in_mem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004574{
Theodore Ts'o240799c2008-10-09 23:53:47 -04004575 struct ext4_group_desc *gdp;
4576 struct buffer_head *bh;
4577 struct super_block *sb = inode->i_sb;
4578 ext4_fsblk_t block;
4579 int inodes_per_block, inode_offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004580
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05004581 iloc->bh = NULL;
Theodore Ts'o240799c2008-10-09 23:53:47 -04004582 if (!ext4_valid_inum(sb, inode->i_ino))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004583 return -EIO;
4584
Theodore Ts'o240799c2008-10-09 23:53:47 -04004585 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4586 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4587 if (!gdp)
4588 return -EIO;
4589
4590 /*
4591 * Figure out the offset within the block group inode table
4592 */
4593 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4594 inode_offset = ((inode->i_ino - 1) %
4595 EXT4_INODES_PER_GROUP(sb));
4596 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4597 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4598
4599 bh = sb_getblk(sb, block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004600 if (!bh) {
Theodore Ts'o240799c2008-10-09 23:53:47 -04004601 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
4602 "inode block - inode=%lu, block=%llu",
4603 inode->i_ino, block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004604 return -EIO;
4605 }
4606 if (!buffer_uptodate(bh)) {
4607 lock_buffer(bh);
Hidehiro Kawai9c83a922008-07-26 16:39:26 -04004608
4609 /*
4610 * If the buffer has the write error flag, we have failed
4611 * to write out another inode in the same block. In this
4612 * case, we don't have to read the block because we may
4613 * read the old inode data successfully.
4614 */
4615 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4616 set_buffer_uptodate(bh);
4617
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004618 if (buffer_uptodate(bh)) {
4619 /* someone brought it uptodate while we waited */
4620 unlock_buffer(bh);
4621 goto has_buffer;
4622 }
4623
4624 /*
4625 * If we have all information of the inode in memory and this
4626 * is the only valid inode in the block, we need not read the
4627 * block.
4628 */
4629 if (in_mem) {
4630 struct buffer_head *bitmap_bh;
Theodore Ts'o240799c2008-10-09 23:53:47 -04004631 int i, start;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004632
Theodore Ts'o240799c2008-10-09 23:53:47 -04004633 start = inode_offset & ~(inodes_per_block - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004634
4635 /* Is the inode bitmap in cache? */
Theodore Ts'o240799c2008-10-09 23:53:47 -04004636 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004637 if (!bitmap_bh)
4638 goto make_io;
4639
4640 /*
4641 * If the inode bitmap isn't in cache then the
4642 * optimisation may end up performing two reads instead
4643 * of one, so skip it.
4644 */
4645 if (!buffer_uptodate(bitmap_bh)) {
4646 brelse(bitmap_bh);
4647 goto make_io;
4648 }
Theodore Ts'o240799c2008-10-09 23:53:47 -04004649 for (i = start; i < start + inodes_per_block; i++) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004650 if (i == inode_offset)
4651 continue;
Mingming Cao617ba132006-10-11 01:20:53 -07004652 if (ext4_test_bit(i, bitmap_bh->b_data))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004653 break;
4654 }
4655 brelse(bitmap_bh);
Theodore Ts'o240799c2008-10-09 23:53:47 -04004656 if (i == start + inodes_per_block) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004657 /* all other inodes are free, so skip I/O */
4658 memset(bh->b_data, 0, bh->b_size);
4659 set_buffer_uptodate(bh);
4660 unlock_buffer(bh);
4661 goto has_buffer;
4662 }
4663 }
4664
4665make_io:
4666 /*
Theodore Ts'o240799c2008-10-09 23:53:47 -04004667 * If we need to do any I/O, try to pre-readahead extra
4668 * blocks from the inode table.
4669 */
4670 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4671 ext4_fsblk_t b, end, table;
4672 unsigned num;
4673
4674 table = ext4_inode_table(sb, gdp);
Theodore Ts'ob713a5e2009-03-31 09:11:14 -04004675 /* s_inode_readahead_blks is always a power of 2 */
Theodore Ts'o240799c2008-10-09 23:53:47 -04004676 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4677 if (table > b)
4678 b = table;
4679 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4680 num = EXT4_INODES_PER_GROUP(sb);
4681 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4682 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05004683 num -= ext4_itable_unused_count(sb, gdp);
Theodore Ts'o240799c2008-10-09 23:53:47 -04004684 table += num / inodes_per_block;
4685 if (end > table)
4686 end = table;
4687 while (b <= end)
4688 sb_breadahead(sb, b++);
4689 }
4690
4691 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004692 * There are other valid inodes in the buffer, this inode
4693 * has in-inode xattrs, or we don't have this inode in memory.
4694 * Read the block from disk.
4695 */
4696 get_bh(bh);
4697 bh->b_end_io = end_buffer_read_sync;
4698 submit_bh(READ_META, bh);
4699 wait_on_buffer(bh);
4700 if (!buffer_uptodate(bh)) {
Theodore Ts'o240799c2008-10-09 23:53:47 -04004701 ext4_error(sb, __func__,
4702 "unable to read inode block - inode=%lu, "
4703 "block=%llu", inode->i_ino, block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004704 brelse(bh);
4705 return -EIO;
4706 }
4707 }
4708has_buffer:
4709 iloc->bh = bh;
4710 return 0;
4711}
4712
Mingming Cao617ba132006-10-11 01:20:53 -07004713int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004714{
4715 /* We have all inode data except xattrs in memory here. */
Mingming Cao617ba132006-10-11 01:20:53 -07004716 return __ext4_get_inode_loc(inode, iloc,
4717 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004718}
4719
Mingming Cao617ba132006-10-11 01:20:53 -07004720void ext4_set_inode_flags(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004721{
Mingming Cao617ba132006-10-11 01:20:53 -07004722 unsigned int flags = EXT4_I(inode)->i_flags;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004723
4724 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
Mingming Cao617ba132006-10-11 01:20:53 -07004725 if (flags & EXT4_SYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004726 inode->i_flags |= S_SYNC;
Mingming Cao617ba132006-10-11 01:20:53 -07004727 if (flags & EXT4_APPEND_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004728 inode->i_flags |= S_APPEND;
Mingming Cao617ba132006-10-11 01:20:53 -07004729 if (flags & EXT4_IMMUTABLE_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004730 inode->i_flags |= S_IMMUTABLE;
Mingming Cao617ba132006-10-11 01:20:53 -07004731 if (flags & EXT4_NOATIME_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004732 inode->i_flags |= S_NOATIME;
Mingming Cao617ba132006-10-11 01:20:53 -07004733 if (flags & EXT4_DIRSYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004734 inode->i_flags |= S_DIRSYNC;
4735}
4736
Jan Karaff9ddf72007-07-18 09:24:20 -04004737/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4738void ext4_get_inode_flags(struct ext4_inode_info *ei)
4739{
4740 unsigned int flags = ei->vfs_inode.i_flags;
4741
4742 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4743 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4744 if (flags & S_SYNC)
4745 ei->i_flags |= EXT4_SYNC_FL;
4746 if (flags & S_APPEND)
4747 ei->i_flags |= EXT4_APPEND_FL;
4748 if (flags & S_IMMUTABLE)
4749 ei->i_flags |= EXT4_IMMUTABLE_FL;
4750 if (flags & S_NOATIME)
4751 ei->i_flags |= EXT4_NOATIME_FL;
4752 if (flags & S_DIRSYNC)
4753 ei->i_flags |= EXT4_DIRSYNC_FL;
4754}
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004755
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004756static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004757 struct ext4_inode_info *ei)
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004758{
4759 blkcnt_t i_blocks ;
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004760 struct inode *inode = &(ei->vfs_inode);
4761 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004762
4763 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4764 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4765 /* we are using combined 48 bit field */
4766 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4767 le32_to_cpu(raw_inode->i_blocks_lo);
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004768 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4769 /* i_blocks represent file system block size */
4770 return i_blocks << (inode->i_blkbits - 9);
4771 } else {
4772 return i_blocks;
4773 }
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004774 } else {
4775 return le32_to_cpu(raw_inode->i_blocks_lo);
4776 }
4777}
Jan Karaff9ddf72007-07-18 09:24:20 -04004778
David Howells1d1fe1e2008-02-07 00:15:37 -08004779struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004780{
Mingming Cao617ba132006-10-11 01:20:53 -07004781 struct ext4_iloc iloc;
4782 struct ext4_inode *raw_inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08004783 struct ext4_inode_info *ei;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004784 struct buffer_head *bh;
David Howells1d1fe1e2008-02-07 00:15:37 -08004785 struct inode *inode;
4786 long ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004787 int block;
4788
David Howells1d1fe1e2008-02-07 00:15:37 -08004789 inode = iget_locked(sb, ino);
4790 if (!inode)
4791 return ERR_PTR(-ENOMEM);
4792 if (!(inode->i_state & I_NEW))
4793 return inode;
4794
4795 ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004796
David Howells1d1fe1e2008-02-07 00:15:37 -08004797 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4798 if (ret < 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004799 goto bad_inode;
4800 bh = iloc.bh;
Mingming Cao617ba132006-10-11 01:20:53 -07004801 raw_inode = ext4_raw_inode(&iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004802 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4803 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4804 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004805 if (!(test_opt(inode->i_sb, NO_UID32))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004806 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4807 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4808 }
4809 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004810
4811 ei->i_state = 0;
4812 ei->i_dir_start_lookup = 0;
4813 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4814 /* We now have enough fields to check if the inode was active or not.
4815 * This is needed because nfsd might try to access dead inodes
4816 * the test is that same one that e2fsck uses
4817 * NeilBrown 1999oct15
4818 */
4819 if (inode->i_nlink == 0) {
4820 if (inode->i_mode == 0 ||
Mingming Cao617ba132006-10-11 01:20:53 -07004821 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004822 /* this inode is deleted */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004823 brelse(bh);
David Howells1d1fe1e2008-02-07 00:15:37 -08004824 ret = -ESTALE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004825 goto bad_inode;
4826 }
4827 /* The only unlinked inodes we let through here have
4828 * valid i_mode and are being read by the orphan
4829 * recovery code: that's fine, we're about to complete
4830 * the process of deleting those. */
4831 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004832 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004833 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05004834 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
Theodore Ts'oa9e81742009-04-24 16:11:18 -04004835 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07004836 ei->i_file_acl |=
4837 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004838 inode->i_size = ext4_isize(raw_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004839 ei->i_disksize = inode->i_size;
4840 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4841 ei->i_block_group = iloc.block_group;
Theodore Ts'oa4912122009-03-12 12:18:34 -04004842 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004843 /*
4844 * NOTE! The in-memory inode i_data array is in little-endian order
4845 * even on big-endian machines: we do NOT byteswap the block numbers!
4846 */
Mingming Cao617ba132006-10-11 01:20:53 -07004847 for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004848 ei->i_data[block] = raw_inode->i_block[block];
4849 INIT_LIST_HEAD(&ei->i_orphan);
4850
Eric Sandeen0040d982008-02-05 22:36:43 -05004851 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004852 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
Mingming Cao617ba132006-10-11 01:20:53 -07004853 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
Kirill Korotaeve5d28612007-06-23 17:16:51 -07004854 EXT4_INODE_SIZE(inode->i_sb)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004855 brelse(bh);
David Howells1d1fe1e2008-02-07 00:15:37 -08004856 ret = -EIO;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004857 goto bad_inode;
Kirill Korotaeve5d28612007-06-23 17:16:51 -07004858 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004859 if (ei->i_extra_isize == 0) {
4860 /* The extra space is currently unused. Use it. */
Mingming Cao617ba132006-10-11 01:20:53 -07004861 ei->i_extra_isize = sizeof(struct ext4_inode) -
4862 EXT4_GOOD_OLD_INODE_SIZE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004863 } else {
4864 __le32 *magic = (void *)raw_inode +
Mingming Cao617ba132006-10-11 01:20:53 -07004865 EXT4_GOOD_OLD_INODE_SIZE +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004866 ei->i_extra_isize;
Mingming Cao617ba132006-10-11 01:20:53 -07004867 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004868 ei->i_state |= EXT4_STATE_XATTR;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004869 }
4870 } else
4871 ei->i_extra_isize = 0;
4872
Kalpak Shahef7f3832007-07-18 09:15:20 -04004873 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4874 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4875 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4876 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4877
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004878 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4879 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4880 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4881 inode->i_version |=
4882 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4883 }
4884
Theodore Ts'oc4b5a612009-04-24 18:45:35 -04004885 ret = 0;
Theodore Ts'o485c26e2009-04-24 13:43:20 -04004886 if (ei->i_file_acl &&
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004887 ((ei->i_file_acl <
Theodore Ts'o485c26e2009-04-24 13:43:20 -04004888 (le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) +
4889 EXT4_SB(sb)->s_gdb_count)) ||
4890 (ei->i_file_acl >= ext4_blocks_count(EXT4_SB(sb)->s_es)))) {
4891 ext4_error(sb, __func__,
4892 "bad extended attribute block %llu in inode #%lu",
4893 ei->i_file_acl, inode->i_ino);
4894 ret = -EIO;
4895 goto bad_inode;
4896 } else if (ei->i_flags & EXT4_EXTENTS_FL) {
Theodore Ts'oc4b5a612009-04-24 18:45:35 -04004897 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4898 (S_ISLNK(inode->i_mode) &&
4899 !ext4_inode_is_fast_symlink(inode)))
4900 /* Validate extent which is part of inode */
4901 ret = ext4_ext_check_inode(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004902 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
Thiemo Nagelfe2c8192009-03-31 08:36:10 -04004903 (S_ISLNK(inode->i_mode) &&
4904 !ext4_inode_is_fast_symlink(inode))) {
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004905 /* Validate block references which are part of inode */
Thiemo Nagelfe2c8192009-03-31 08:36:10 -04004906 ret = ext4_check_inode_blockref(inode);
4907 }
4908 if (ret) {
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004909 brelse(bh);
4910 goto bad_inode;
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -04004911 }
4912
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004913 if (S_ISREG(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004914 inode->i_op = &ext4_file_inode_operations;
4915 inode->i_fop = &ext4_file_operations;
4916 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004917 } else if (S_ISDIR(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004918 inode->i_op = &ext4_dir_inode_operations;
4919 inode->i_fop = &ext4_dir_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004920 } else if (S_ISLNK(inode->i_mode)) {
Duane Griffine83c1392008-12-19 20:47:15 +00004921 if (ext4_inode_is_fast_symlink(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004922 inode->i_op = &ext4_fast_symlink_inode_operations;
Duane Griffine83c1392008-12-19 20:47:15 +00004923 nd_terminate_link(ei->i_data, inode->i_size,
4924 sizeof(ei->i_data) - 1);
4925 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07004926 inode->i_op = &ext4_symlink_inode_operations;
4927 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004928 }
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004929 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4930 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004931 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004932 if (raw_inode->i_block[0])
4933 init_special_inode(inode, inode->i_mode,
4934 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4935 else
4936 init_special_inode(inode, inode->i_mode,
4937 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004938 } else {
4939 brelse(bh);
4940 ret = -EIO;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004941 ext4_error(inode->i_sb, __func__,
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004942 "bogus i_mode (%o) for inode=%lu",
4943 inode->i_mode, inode->i_ino);
4944 goto bad_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004945 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004946 brelse(iloc.bh);
Mingming Cao617ba132006-10-11 01:20:53 -07004947 ext4_set_inode_flags(inode);
David Howells1d1fe1e2008-02-07 00:15:37 -08004948 unlock_new_inode(inode);
4949 return inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004950
4951bad_inode:
David Howells1d1fe1e2008-02-07 00:15:37 -08004952 iget_failed(inode);
4953 return ERR_PTR(ret);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004954}
4955
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004956static int ext4_inode_blocks_set(handle_t *handle,
4957 struct ext4_inode *raw_inode,
4958 struct ext4_inode_info *ei)
4959{
4960 struct inode *inode = &(ei->vfs_inode);
4961 u64 i_blocks = inode->i_blocks;
4962 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004963
4964 if (i_blocks <= ~0U) {
4965 /*
4966 * i_blocks can be represnted in a 32 bit variable
4967 * as multiple of 512 bytes
4968 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004969 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004970 raw_inode->i_blocks_high = 0;
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004971 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
Theodore Ts'of287a1a2008-10-16 22:50:48 -04004972 return 0;
4973 }
4974 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4975 return -EFBIG;
4976
4977 if (i_blocks <= 0xffffffffffffULL) {
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004978 /*
4979 * i_blocks can be represented in a 48 bit variable
4980 * as multiple of 512 bytes
4981 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004982 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004983 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004984 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004985 } else {
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004986 ei->i_flags |= EXT4_HUGE_FILE_FL;
4987 /* i_block is stored in file system block size */
4988 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4989 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4990 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004991 }
Theodore Ts'of287a1a2008-10-16 22:50:48 -04004992 return 0;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004993}
4994
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004995/*
4996 * Post the struct inode info into an on-disk inode location in the
4997 * buffer-cache. This gobbles the caller's reference to the
4998 * buffer_head in the inode location struct.
4999 *
5000 * The caller must have write access to iloc->bh.
5001 */
Mingming Cao617ba132006-10-11 01:20:53 -07005002static int ext4_do_update_inode(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005003 struct inode *inode,
Frank Mayhar830156c2009-09-29 10:07:47 -04005004 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005005{
Mingming Cao617ba132006-10-11 01:20:53 -07005006 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5007 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005008 struct buffer_head *bh = iloc->bh;
5009 int err = 0, rc, block;
5010
5011 /* For fields not not tracking in the in-memory inode,
5012 * initialise them to zero for new inodes. */
Mingming Cao617ba132006-10-11 01:20:53 -07005013 if (ei->i_state & EXT4_STATE_NEW)
5014 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005015
Jan Karaff9ddf72007-07-18 09:24:20 -04005016 ext4_get_inode_flags(ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005017 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04005018 if (!(test_opt(inode->i_sb, NO_UID32))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005019 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5020 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5021/*
5022 * Fix up interoperability with old kernels. Otherwise, old inodes get
5023 * re-used with the upper 16 bits of the uid/gid intact
5024 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04005025 if (!ei->i_dtime) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005026 raw_inode->i_uid_high =
5027 cpu_to_le16(high_16_bits(inode->i_uid));
5028 raw_inode->i_gid_high =
5029 cpu_to_le16(high_16_bits(inode->i_gid));
5030 } else {
5031 raw_inode->i_uid_high = 0;
5032 raw_inode->i_gid_high = 0;
5033 }
5034 } else {
5035 raw_inode->i_uid_low =
5036 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5037 raw_inode->i_gid_low =
5038 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5039 raw_inode->i_uid_high = 0;
5040 raw_inode->i_gid_high = 0;
5041 }
5042 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
Kalpak Shahef7f3832007-07-18 09:15:20 -04005043
5044 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5045 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5046 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5047 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5048
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05005049 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5050 goto out_brelse;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005051 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
Theodore Ts'o1b9c12f2009-09-17 08:32:22 -04005052 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
Mingming Cao9b8f1f02006-10-11 01:21:13 -07005053 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5054 cpu_to_le32(EXT4_OS_HURD))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07005055 raw_inode->i_file_acl_high =
5056 cpu_to_le16(ei->i_file_acl >> 32);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05005057 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05005058 ext4_isize_set(raw_inode, ei->i_disksize);
5059 if (ei->i_disksize > 0x7fffffffULL) {
5060 struct super_block *sb = inode->i_sb;
5061 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5062 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5063 EXT4_SB(sb)->s_es->s_rev_level ==
5064 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5065 /* If this is the first large file
5066 * created, add a flag to the superblock.
5067 */
5068 err = ext4_journal_get_write_access(handle,
5069 EXT4_SB(sb)->s_sbh);
5070 if (err)
5071 goto out_brelse;
5072 ext4_update_dynamic_rev(sb);
5073 EXT4_SET_RO_COMPAT_FEATURE(sb,
Mingming Cao617ba132006-10-11 01:20:53 -07005074 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05005075 sb->s_dirt = 1;
Frank Mayhar03901312009-01-07 00:06:22 -05005076 ext4_handle_sync(handle);
5077 err = ext4_handle_dirty_metadata(handle, inode,
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05005078 EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005079 }
5080 }
5081 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5082 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5083 if (old_valid_dev(inode->i_rdev)) {
5084 raw_inode->i_block[0] =
5085 cpu_to_le32(old_encode_dev(inode->i_rdev));
5086 raw_inode->i_block[1] = 0;
5087 } else {
5088 raw_inode->i_block[0] = 0;
5089 raw_inode->i_block[1] =
5090 cpu_to_le32(new_encode_dev(inode->i_rdev));
5091 raw_inode->i_block[2] = 0;
5092 }
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04005093 } else
5094 for (block = 0; block < EXT4_N_BLOCKS; block++)
5095 raw_inode->i_block[block] = ei->i_data[block];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005096
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05005097 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5098 if (ei->i_extra_isize) {
5099 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5100 raw_inode->i_version_hi =
5101 cpu_to_le32(inode->i_version >> 32);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005102 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05005103 }
5104
Frank Mayhar830156c2009-09-29 10:07:47 -04005105 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5106 rc = ext4_handle_dirty_metadata(handle, inode, bh);
5107 if (!err)
5108 err = rc;
Mingming Cao617ba132006-10-11 01:20:53 -07005109 ei->i_state &= ~EXT4_STATE_NEW;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005110
5111out_brelse:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04005112 brelse(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07005113 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005114 return err;
5115}
5116
5117/*
Mingming Cao617ba132006-10-11 01:20:53 -07005118 * ext4_write_inode()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005119 *
5120 * We are called from a few places:
5121 *
5122 * - Within generic_file_write() for O_SYNC files.
5123 * Here, there will be no transaction running. We wait for any running
5124 * trasnaction to commit.
5125 *
5126 * - Within sys_sync(), kupdate and such.
5127 * We wait on commit, if tol to.
5128 *
5129 * - Within prune_icache() (PF_MEMALLOC == true)
5130 * Here we simply return. We can't afford to block kswapd on the
5131 * journal commit.
5132 *
5133 * In all cases it is actually safe for us to return without doing anything,
5134 * because the inode has been copied into a raw inode buffer in
Mingming Cao617ba132006-10-11 01:20:53 -07005135 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005136 * knfsd.
5137 *
5138 * Note that we are absolutely dependent upon all inode dirtiers doing the
5139 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5140 * which we are interested.
5141 *
5142 * It would be a bug for them to not do this. The code:
5143 *
5144 * mark_inode_dirty(inode)
5145 * stuff();
5146 * inode->i_size = expr;
5147 *
5148 * is in error because a kswapd-driven write_inode() could occur while
5149 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5150 * will no longer be on the superblock's dirty inode list.
5151 */
Mingming Cao617ba132006-10-11 01:20:53 -07005152int ext4_write_inode(struct inode *inode, int wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005153{
Frank Mayhar91ac6f42009-09-09 22:33:47 -04005154 int err;
5155
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005156 if (current->flags & PF_MEMALLOC)
5157 return 0;
5158
Frank Mayhar91ac6f42009-09-09 22:33:47 -04005159 if (EXT4_SB(inode->i_sb)->s_journal) {
5160 if (ext4_journal_current_handle()) {
5161 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5162 dump_stack();
5163 return -EIO;
5164 }
5165
5166 if (!wait)
5167 return 0;
5168
5169 err = ext4_force_commit(inode->i_sb);
5170 } else {
5171 struct ext4_iloc iloc;
5172
5173 err = ext4_get_inode_loc(inode, &iloc);
5174 if (err)
5175 return err;
Frank Mayhar830156c2009-09-29 10:07:47 -04005176 if (wait)
5177 sync_dirty_buffer(iloc.bh);
5178 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5179 ext4_error(inode->i_sb, __func__,
5180 "IO error syncing inode, "
5181 "inode=%lu, block=%llu",
5182 inode->i_ino,
5183 (unsigned long long)iloc.bh->b_blocknr);
5184 err = -EIO;
5185 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005186 }
Frank Mayhar91ac6f42009-09-09 22:33:47 -04005187 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005188}
5189
5190/*
Mingming Cao617ba132006-10-11 01:20:53 -07005191 * ext4_setattr()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005192 *
5193 * Called from notify_change.
5194 *
5195 * We want to trap VFS attempts to truncate the file as soon as
5196 * possible. In particular, we want to make sure that when the VFS
5197 * shrinks i_size, we put the inode on the orphan list and modify
5198 * i_disksize immediately, so that during the subsequent flushing of
5199 * dirty pages and freeing of disk blocks, we can guarantee that any
5200 * commit will leave the blocks being flushed in an unused state on
5201 * disk. (On recovery, the inode will get truncated and the blocks will
5202 * be freed, so we have a strong guarantee that no future commit will
5203 * leave these blocks visible to the user.)
5204 *
Jan Kara678aaf42008-07-11 19:27:31 -04005205 * Another thing we have to assure is that if we are in ordered mode
5206 * and inode is still attached to the committing transaction, we must
5207 * we start writeout of all the dirty pages which are being truncated.
5208 * This way we are sure that all the data written in the previous
5209 * transaction are already on disk (truncate waits for pages under
5210 * writeback).
5211 *
5212 * Called with inode->i_mutex down.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005213 */
Mingming Cao617ba132006-10-11 01:20:53 -07005214int ext4_setattr(struct dentry *dentry, struct iattr *attr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005215{
5216 struct inode *inode = dentry->d_inode;
5217 int error, rc = 0;
5218 const unsigned int ia_valid = attr->ia_valid;
5219
5220 error = inode_change_ok(inode, attr);
5221 if (error)
5222 return error;
5223
5224 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5225 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5226 handle_t *handle;
5227
5228 /* (user+group)*(old+new) structure, inode write (sb,
5229 * inode block, ? - but truncate inode update has it) */
Mingming Cao617ba132006-10-11 01:20:53 -07005230 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
5231 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005232 if (IS_ERR(handle)) {
5233 error = PTR_ERR(handle);
5234 goto err_out;
5235 }
Jan Karaa269eb12009-01-26 17:04:39 +01005236 error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005237 if (error) {
Mingming Cao617ba132006-10-11 01:20:53 -07005238 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005239 return error;
5240 }
5241 /* Update corresponding info in inode so that everything is in
5242 * one transaction */
5243 if (attr->ia_valid & ATTR_UID)
5244 inode->i_uid = attr->ia_uid;
5245 if (attr->ia_valid & ATTR_GID)
5246 inode->i_gid = attr->ia_gid;
Mingming Cao617ba132006-10-11 01:20:53 -07005247 error = ext4_mark_inode_dirty(handle, inode);
5248 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005249 }
5250
Eric Sandeene2b46572008-01-28 23:58:27 -05005251 if (attr->ia_valid & ATTR_SIZE) {
5252 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
5253 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5254
5255 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5256 error = -EFBIG;
5257 goto err_out;
5258 }
5259 }
5260 }
5261
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005262 if (S_ISREG(inode->i_mode) &&
5263 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
5264 handle_t *handle;
5265
Mingming Cao617ba132006-10-11 01:20:53 -07005266 handle = ext4_journal_start(inode, 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005267 if (IS_ERR(handle)) {
5268 error = PTR_ERR(handle);
5269 goto err_out;
5270 }
5271
Mingming Cao617ba132006-10-11 01:20:53 -07005272 error = ext4_orphan_add(handle, inode);
5273 EXT4_I(inode)->i_disksize = attr->ia_size;
5274 rc = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005275 if (!error)
5276 error = rc;
Mingming Cao617ba132006-10-11 01:20:53 -07005277 ext4_journal_stop(handle);
Jan Kara678aaf42008-07-11 19:27:31 -04005278
5279 if (ext4_should_order_data(inode)) {
5280 error = ext4_begin_ordered_truncate(inode,
5281 attr->ia_size);
5282 if (error) {
5283 /* Do as much error cleanup as possible */
5284 handle = ext4_journal_start(inode, 3);
5285 if (IS_ERR(handle)) {
5286 ext4_orphan_del(NULL, inode);
5287 goto err_out;
5288 }
5289 ext4_orphan_del(handle, inode);
5290 ext4_journal_stop(handle);
5291 goto err_out;
5292 }
5293 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005294 }
5295
5296 rc = inode_setattr(inode, attr);
5297
Mingming Cao617ba132006-10-11 01:20:53 -07005298 /* If inode_setattr's call to ext4_truncate failed to get a
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005299 * transaction handle at all, we need to clean up the in-core
5300 * orphan list manually. */
5301 if (inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07005302 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005303
5304 if (!rc && (ia_valid & ATTR_MODE))
Mingming Cao617ba132006-10-11 01:20:53 -07005305 rc = ext4_acl_chmod(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005306
5307err_out:
Mingming Cao617ba132006-10-11 01:20:53 -07005308 ext4_std_error(inode->i_sb, error);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005309 if (!error)
5310 error = rc;
5311 return error;
5312}
5313
Mingming Cao3e3398a2008-07-11 19:27:31 -04005314int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5315 struct kstat *stat)
5316{
5317 struct inode *inode;
5318 unsigned long delalloc_blocks;
5319
5320 inode = dentry->d_inode;
5321 generic_fillattr(inode, stat);
5322
5323 /*
5324 * We can't update i_blocks if the block allocation is delayed
5325 * otherwise in the case of system crash before the real block
5326 * allocation is done, we will have i_blocks inconsistent with
5327 * on-disk file blocks.
5328 * We always keep i_blocks updated together with real
5329 * allocation. But to not confuse with user, stat
5330 * will return the blocks that include the delayed allocation
5331 * blocks for this file.
5332 */
5333 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5334 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5335 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5336
5337 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5338 return 0;
5339}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005340
Mingming Caoa02908f2008-08-19 22:16:07 -04005341static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5342 int chunk)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005343{
Mingming Caoa02908f2008-08-19 22:16:07 -04005344 int indirects;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005345
Mingming Caoa02908f2008-08-19 22:16:07 -04005346 /* if nrblocks are contiguous */
5347 if (chunk) {
5348 /*
5349 * With N contiguous data blocks, it need at most
5350 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5351 * 2 dindirect blocks
5352 * 1 tindirect block
5353 */
5354 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5355 return indirects + 3;
5356 }
5357 /*
5358 * if nrblocks are not contiguous, worse case, each block touch
5359 * a indirect block, and each indirect block touch a double indirect
5360 * block, plus a triple indirect block
5361 */
5362 indirects = nrblocks * 2 + 1;
5363 return indirects;
5364}
Alex Tomasa86c6182006-10-11 01:21:03 -07005365
Mingming Caoa02908f2008-08-19 22:16:07 -04005366static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5367{
5368 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
Theodore Ts'oac51d832008-11-06 16:49:36 -05005369 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5370 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
Mingming Caoa02908f2008-08-19 22:16:07 -04005371}
Theodore Ts'oac51d832008-11-06 16:49:36 -05005372
Mingming Caoa02908f2008-08-19 22:16:07 -04005373/*
5374 * Account for index blocks, block groups bitmaps and block group
5375 * descriptor blocks if modify datablocks and index blocks
5376 * worse case, the indexs blocks spread over different block groups
5377 *
5378 * If datablocks are discontiguous, they are possible to spread over
5379 * different block groups too. If they are contiugous, with flexbg,
5380 * they could still across block group boundary.
5381 *
5382 * Also account for superblock, inode, quota and xattr blocks
5383 */
5384int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5385{
Theodore Ts'o8df96752009-05-01 08:50:38 -04005386 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5387 int gdpblocks;
Mingming Caoa02908f2008-08-19 22:16:07 -04005388 int idxblocks;
5389 int ret = 0;
5390
5391 /*
5392 * How many index blocks need to touch to modify nrblocks?
5393 * The "Chunk" flag indicating whether the nrblocks is
5394 * physically contiguous on disk
5395 *
5396 * For Direct IO and fallocate, they calls get_block to allocate
5397 * one single extent at a time, so they could set the "Chunk" flag
5398 */
5399 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5400
5401 ret = idxblocks;
5402
5403 /*
5404 * Now let's see how many group bitmaps and group descriptors need
5405 * to account
5406 */
5407 groups = idxblocks;
5408 if (chunk)
5409 groups += 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005410 else
Mingming Caoa02908f2008-08-19 22:16:07 -04005411 groups += nrblocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005412
Mingming Caoa02908f2008-08-19 22:16:07 -04005413 gdpblocks = groups;
Theodore Ts'o8df96752009-05-01 08:50:38 -04005414 if (groups > ngroups)
5415 groups = ngroups;
Mingming Caoa02908f2008-08-19 22:16:07 -04005416 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5417 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5418
5419 /* bitmaps and block group descriptor blocks */
5420 ret += groups + gdpblocks;
5421
5422 /* Blocks for super block, inode, quota and xattr blocks */
5423 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005424
5425 return ret;
5426}
5427
5428/*
Mingming Caoa02908f2008-08-19 22:16:07 -04005429 * Calulate the total number of credits to reserve to fit
Mingming Caof3bd1f32008-08-19 22:16:03 -04005430 * the modification of a single pages into a single transaction,
5431 * which may include multiple chunks of block allocations.
Mingming Caoa02908f2008-08-19 22:16:07 -04005432 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04005433 * This could be called via ext4_write_begin()
Mingming Caoa02908f2008-08-19 22:16:07 -04005434 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04005435 * We need to consider the worse case, when
Mingming Caoa02908f2008-08-19 22:16:07 -04005436 * one new block per extent.
Mingming Caoa02908f2008-08-19 22:16:07 -04005437 */
5438int ext4_writepage_trans_blocks(struct inode *inode)
5439{
5440 int bpp = ext4_journal_blocks_per_page(inode);
5441 int ret;
5442
5443 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5444
5445 /* Account for data blocks for journalled mode */
5446 if (ext4_should_journal_data(inode))
5447 ret += bpp;
5448 return ret;
5449}
Mingming Caof3bd1f32008-08-19 22:16:03 -04005450
5451/*
5452 * Calculate the journal credits for a chunk of data modification.
5453 *
5454 * This is called from DIO, fallocate or whoever calling
Theodore Ts'o12b7ac12009-05-14 00:57:44 -04005455 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
Mingming Caof3bd1f32008-08-19 22:16:03 -04005456 *
5457 * journal buffers for data blocks are not included here, as DIO
5458 * and fallocate do no need to journal data buffers.
5459 */
5460int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5461{
5462 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5463}
5464
Mingming Caoa02908f2008-08-19 22:16:07 -04005465/*
Mingming Cao617ba132006-10-11 01:20:53 -07005466 * The caller must have previously called ext4_reserve_inode_write().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005467 * Give this, we know that the caller already has write access to iloc->bh.
5468 */
Mingming Cao617ba132006-10-11 01:20:53 -07005469int ext4_mark_iloc_dirty(handle_t *handle,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04005470 struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005471{
5472 int err = 0;
5473
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05005474 if (test_opt(inode->i_sb, I_VERSION))
5475 inode_inc_iversion(inode);
5476
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005477 /* the do_update_inode consumes one bh->b_count */
5478 get_bh(iloc->bh);
5479
Mingming Caodab291a2006-10-11 01:21:01 -07005480 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
Frank Mayhar830156c2009-09-29 10:07:47 -04005481 err = ext4_do_update_inode(handle, inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005482 put_bh(iloc->bh);
5483 return err;
5484}
5485
5486/*
5487 * On success, We end up with an outstanding reference count against
5488 * iloc->bh. This _must_ be cleaned up later.
5489 */
5490
5491int
Mingming Cao617ba132006-10-11 01:20:53 -07005492ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5493 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005494{
Frank Mayhar03901312009-01-07 00:06:22 -05005495 int err;
5496
5497 err = ext4_get_inode_loc(inode, iloc);
5498 if (!err) {
5499 BUFFER_TRACE(iloc->bh, "get_write_access");
5500 err = ext4_journal_get_write_access(handle, iloc->bh);
5501 if (err) {
5502 brelse(iloc->bh);
5503 iloc->bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005504 }
5505 }
Mingming Cao617ba132006-10-11 01:20:53 -07005506 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005507 return err;
5508}
5509
5510/*
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005511 * Expand an inode by new_extra_isize bytes.
5512 * Returns 0 on success or negative error number on failure.
5513 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05005514static int ext4_expand_extra_isize(struct inode *inode,
5515 unsigned int new_extra_isize,
5516 struct ext4_iloc iloc,
5517 handle_t *handle)
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005518{
5519 struct ext4_inode *raw_inode;
5520 struct ext4_xattr_ibody_header *header;
5521 struct ext4_xattr_entry *entry;
5522
5523 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5524 return 0;
5525
5526 raw_inode = ext4_raw_inode(&iloc);
5527
5528 header = IHDR(inode, raw_inode);
5529 entry = IFIRST(header);
5530
5531 /* No extended attributes present */
5532 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
5533 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5534 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5535 new_extra_isize);
5536 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5537 return 0;
5538 }
5539
5540 /* try to expand with EAs present */
5541 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5542 raw_inode, handle);
5543}
5544
5545/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005546 * What we do here is to mark the in-core inode as clean with respect to inode
5547 * dirtiness (it may still be data-dirty).
5548 * This means that the in-core inode may be reaped by prune_icache
5549 * without having to perform any I/O. This is a very good thing,
5550 * because *any* task may call prune_icache - even ones which
5551 * have a transaction open against a different journal.
5552 *
5553 * Is this cheating? Not really. Sure, we haven't written the
5554 * inode out, but prune_icache isn't a user-visible syncing function.
5555 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5556 * we start and wait on commits.
5557 *
5558 * Is this efficient/effective? Well, we're being nice to the system
5559 * by cleaning up our inodes proactively so they can be reaped
5560 * without I/O. But we are potentially leaving up to five seconds'
5561 * worth of inodes floating about which prune_icache wants us to
5562 * write out. One way to fix that would be to get prune_icache()
5563 * to do a write_super() to free up some memory. It has the desired
5564 * effect.
5565 */
Mingming Cao617ba132006-10-11 01:20:53 -07005566int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005567{
Mingming Cao617ba132006-10-11 01:20:53 -07005568 struct ext4_iloc iloc;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005569 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5570 static unsigned int mnt_count;
5571 int err, ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005572
5573 might_sleep();
Mingming Cao617ba132006-10-11 01:20:53 -07005574 err = ext4_reserve_inode_write(handle, inode, &iloc);
Frank Mayhar03901312009-01-07 00:06:22 -05005575 if (ext4_handle_valid(handle) &&
5576 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005577 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
5578 /*
5579 * We need extra buffer credits since we may write into EA block
5580 * with this same handle. If journal_extend fails, then it will
5581 * only result in a minor loss of functionality for that inode.
5582 * If this is felt to be critical, then e2fsck should be run to
5583 * force a large enough s_min_extra_isize.
5584 */
5585 if ((jbd2_journal_extend(handle,
5586 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5587 ret = ext4_expand_extra_isize(inode,
5588 sbi->s_want_extra_isize,
5589 iloc, handle);
5590 if (ret) {
5591 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04005592 if (mnt_count !=
5593 le16_to_cpu(sbi->s_es->s_mnt_count)) {
Harvey Harrison46e665e2008-04-17 10:38:59 -04005594 ext4_warning(inode->i_sb, __func__,
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005595 "Unable to expand inode %lu. Delete"
5596 " some EAs or run e2fsck.",
5597 inode->i_ino);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04005598 mnt_count =
5599 le16_to_cpu(sbi->s_es->s_mnt_count);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005600 }
5601 }
5602 }
5603 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005604 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07005605 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005606 return err;
5607}
5608
5609/*
Mingming Cao617ba132006-10-11 01:20:53 -07005610 * ext4_dirty_inode() is called from __mark_inode_dirty()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005611 *
5612 * We're really interested in the case where a file is being extended.
5613 * i_size has been changed by generic_commit_write() and we thus need
5614 * to include the updated inode in the current transaction.
5615 *
Jan Karaa269eb12009-01-26 17:04:39 +01005616 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005617 * are allocated to the file.
5618 *
5619 * If the inode is marked synchronous, we don't honour that here - doing
5620 * so would cause a commit on atime updates, which we don't bother doing.
5621 * We handle synchronous inodes at the highest possible level.
5622 */
Mingming Cao617ba132006-10-11 01:20:53 -07005623void ext4_dirty_inode(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005624{
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005625 handle_t *handle;
5626
Mingming Cao617ba132006-10-11 01:20:53 -07005627 handle = ext4_journal_start(inode, 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005628 if (IS_ERR(handle))
5629 goto out;
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04005630
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04005631 ext4_mark_inode_dirty(handle, inode);
5632
Mingming Cao617ba132006-10-11 01:20:53 -07005633 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005634out:
5635 return;
5636}
5637
5638#if 0
5639/*
5640 * Bind an inode's backing buffer_head into this transaction, to prevent
5641 * it from being flushed to disk early. Unlike
Mingming Cao617ba132006-10-11 01:20:53 -07005642 * ext4_reserve_inode_write, this leaves behind no bh reference and
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005643 * returns no iloc structure, so the caller needs to repeat the iloc
5644 * lookup to mark the inode dirty later.
5645 */
Mingming Cao617ba132006-10-11 01:20:53 -07005646static int ext4_pin_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005647{
Mingming Cao617ba132006-10-11 01:20:53 -07005648 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005649
5650 int err = 0;
5651 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07005652 err = ext4_get_inode_loc(inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005653 if (!err) {
5654 BUFFER_TRACE(iloc.bh, "get_write_access");
Mingming Caodab291a2006-10-11 01:21:01 -07005655 err = jbd2_journal_get_write_access(handle, iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005656 if (!err)
Frank Mayhar03901312009-01-07 00:06:22 -05005657 err = ext4_handle_dirty_metadata(handle,
5658 inode,
5659 iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005660 brelse(iloc.bh);
5661 }
5662 }
Mingming Cao617ba132006-10-11 01:20:53 -07005663 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005664 return err;
5665}
5666#endif
5667
Mingming Cao617ba132006-10-11 01:20:53 -07005668int ext4_change_inode_journal_flag(struct inode *inode, int val)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005669{
5670 journal_t *journal;
5671 handle_t *handle;
5672 int err;
5673
5674 /*
5675 * We have to be very careful here: changing a data block's
5676 * journaling status dynamically is dangerous. If we write a
5677 * data block to the journal, change the status and then delete
5678 * that block, we risk forgetting to revoke the old log record
5679 * from the journal and so a subsequent replay can corrupt data.
5680 * So, first we make sure that the journal is empty and that
5681 * nobody is changing anything.
5682 */
5683
Mingming Cao617ba132006-10-11 01:20:53 -07005684 journal = EXT4_JOURNAL(inode);
Frank Mayhar03901312009-01-07 00:06:22 -05005685 if (!journal)
5686 return 0;
Dave Hansend6995942007-07-18 08:33:51 -04005687 if (is_journal_aborted(journal))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005688 return -EROFS;
5689
Mingming Caodab291a2006-10-11 01:21:01 -07005690 jbd2_journal_lock_updates(journal);
5691 jbd2_journal_flush(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005692
5693 /*
5694 * OK, there are no updates running now, and all cached data is
5695 * synced to disk. We are now in a completely consistent state
5696 * which doesn't have anything in the journal, and we know that
5697 * no filesystem updates are running, so it is safe to modify
5698 * the inode's in-core data-journaling state flag now.
5699 */
5700
5701 if (val)
Mingming Cao617ba132006-10-11 01:20:53 -07005702 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005703 else
Mingming Cao617ba132006-10-11 01:20:53 -07005704 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5705 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005706
Mingming Caodab291a2006-10-11 01:21:01 -07005707 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005708
5709 /* Finally we can mark the inode as dirty. */
5710
Mingming Cao617ba132006-10-11 01:20:53 -07005711 handle = ext4_journal_start(inode, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005712 if (IS_ERR(handle))
5713 return PTR_ERR(handle);
5714
Mingming Cao617ba132006-10-11 01:20:53 -07005715 err = ext4_mark_inode_dirty(handle, inode);
Frank Mayhar03901312009-01-07 00:06:22 -05005716 ext4_handle_sync(handle);
Mingming Cao617ba132006-10-11 01:20:53 -07005717 ext4_journal_stop(handle);
5718 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005719
5720 return err;
5721}
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005722
5723static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5724{
5725 return !buffer_mapped(bh);
5726}
5727
Nick Pigginc2ec1752009-03-31 15:23:21 -07005728int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005729{
Nick Pigginc2ec1752009-03-31 15:23:21 -07005730 struct page *page = vmf->page;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005731 loff_t size;
5732 unsigned long len;
5733 int ret = -EINVAL;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04005734 void *fsdata;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005735 struct file *file = vma->vm_file;
5736 struct inode *inode = file->f_path.dentry->d_inode;
5737 struct address_space *mapping = inode->i_mapping;
5738
5739 /*
5740 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5741 * get i_mutex because we are already holding mmap_sem.
5742 */
5743 down_read(&inode->i_alloc_sem);
5744 size = i_size_read(inode);
5745 if (page->mapping != mapping || size <= page_offset(page)
5746 || !PageUptodate(page)) {
5747 /* page got truncated from under us? */
5748 goto out_unlock;
5749 }
5750 ret = 0;
5751 if (PageMappedToDisk(page))
5752 goto out_unlock;
5753
5754 if (page->index == size >> PAGE_CACHE_SHIFT)
5755 len = size & ~PAGE_CACHE_MASK;
5756 else
5757 len = PAGE_CACHE_SIZE;
5758
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005759 lock_page(page);
5760 /*
5761 * return if we have all the buffers mapped. This avoid
5762 * the need to call write_begin/write_end which does a
5763 * journal_start/journal_stop which can block and take
5764 * long time
5765 */
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005766 if (page_has_buffers(page)) {
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005767 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005768 ext4_bh_unmapped)) {
5769 unlock_page(page);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005770 goto out_unlock;
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005771 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005772 }
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005773 unlock_page(page);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005774 /*
5775 * OK, we need to fill the hole... Do write_begin write_end
5776 * to do block allocation/reservation.We are not holding
5777 * inode.i__mutex here. That allow * parallel write_begin,
5778 * write_end call. lock_page prevent this from happening
5779 * on the same page though
5780 */
5781 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04005782 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005783 if (ret < 0)
5784 goto out_unlock;
5785 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04005786 len, len, page, fsdata);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005787 if (ret < 0)
5788 goto out_unlock;
5789 ret = 0;
5790out_unlock:
Nick Pigginc2ec1752009-03-31 15:23:21 -07005791 if (ret)
5792 ret = VM_FAULT_SIGBUS;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005793 up_read(&inode->i_alloc_sem);
5794 return ret;
5795}