blob: feaa82fe629d067e0900744fcbb50da2768b0183 [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 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -070015 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
17 *
Mingming Cao617ba132006-10-11 01:20:53 -070018 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
Dave Kleikampac27a0e2006-10-11 01:20:50 -070019 */
20
Dave Kleikampac27a0e2006-10-11 01:20:50 -070021#include <linux/fs.h>
22#include <linux/time.h>
Mingming Caodab291a2006-10-11 01:21:01 -070023#include <linux/jbd2.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070024#include <linux/highuid.h>
25#include <linux/pagemap.h>
26#include <linux/quotaops.h>
27#include <linux/string.h>
28#include <linux/buffer_head.h>
29#include <linux/writeback.h>
Alex Tomas64769242008-07-11 19:27:31 -040030#include <linux/pagevec.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070031#include <linux/mpage.h>
Duane Griffine83c1392008-12-19 20:47:15 +000032#include <linux/namei.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070033#include <linux/uio.h>
34#include <linux/bio.h>
Mingming Cao4c0425f2009-09-28 15:48:41 -040035#include <linux/workqueue.h>
Jiaying Zhang744692d2010-03-04 16:14:02 -050036#include <linux/kernel.h>
Andrew Morton6db26ff2011-01-12 16:59:13 -080037#include <linux/printk.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090038#include <linux/slab.h>
Theodore Ts'oa8901d32010-12-17 10:40:47 -050039#include <linux/ratelimit.h>
Theodore Ts'o9bffad12009-06-17 11:48:11 -040040
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040041#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070042#include "xattr.h"
43#include "acl.h"
Theodore Ts'o9f125d62011-06-27 19:16:04 -040044#include "truncate.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070045
Theodore Ts'o9bffad12009-06-17 11:48:11 -040046#include <trace/events/ext4.h>
47
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -040048#define MPAGE_DA_EXTENT_TAIL 0x01
49
Jan Kara678aaf42008-07-11 19:27:31 -040050static inline int ext4_begin_ordered_truncate(struct inode *inode,
51 loff_t new_size)
52{
Theodore Ts'o7ff9c072010-11-08 13:51:33 -050053 trace_ext4_begin_ordered_truncate(inode, new_size);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -050054 /*
55 * If jinode is zero, then we never opened the file for
56 * writing, so there's no need to call
57 * jbd2_journal_begin_ordered_truncate() since there's no
58 * outstanding writes we need to flush.
59 */
60 if (!EXT4_I(inode)->jinode)
61 return 0;
62 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
63 EXT4_I(inode)->jinode,
64 new_size);
Jan Kara678aaf42008-07-11 19:27:31 -040065}
66
Alex Tomas64769242008-07-11 19:27:31 -040067static void ext4_invalidatepage(struct page *page, unsigned long offset);
Theodore Ts'ocb20d512010-10-27 21:30:09 -040068static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
69 struct buffer_head *bh_result, int create);
70static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
71static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
72static int __ext4_journalled_writepage(struct page *page, unsigned int len);
73static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
Eric Sandeen5f163cc2012-01-04 22:33:28 -050074static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
75 struct inode *inode, struct page *page, loff_t from,
76 loff_t length, int flags);
Alex Tomas64769242008-07-11 19:27:31 -040077
Dave Kleikampac27a0e2006-10-11 01:20:50 -070078/*
79 * Test whether an inode is a fast symlink.
80 */
Mingming Cao617ba132006-10-11 01:20:53 -070081static int ext4_inode_is_fast_symlink(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070082{
Mingming Cao617ba132006-10-11 01:20:53 -070083 int ea_blocks = EXT4_I(inode)->i_file_acl ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -070084 (inode->i_sb->s_blocksize >> 9) : 0;
85
86 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
87}
88
89/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -070090 * Restart the transaction associated with *handle. This does a commit,
91 * so before we call here everything must be consistently dirtied against
92 * this transaction.
93 */
Aneesh Kumar K.Vfa5d1112009-11-02 18:50:49 -050094int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
Jan Kara487caee2009-08-17 22:17:20 -040095 int nblocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070096{
Jan Kara487caee2009-08-17 22:17:20 -040097 int ret;
98
99 /*
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400100 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
Jan Kara487caee2009-08-17 22:17:20 -0400101 * moment, get_block can be called only for blocks inside i_size since
102 * page cache has been already dropped and writes are blocked by
103 * i_mutex. So we can safely drop the i_data_sem here.
104 */
Frank Mayhar03901312009-01-07 00:06:22 -0500105 BUG_ON(EXT4_JOURNAL(inode) == NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700106 jbd_debug(2, "restarting handle %p\n", handle);
Jan Kara487caee2009-08-17 22:17:20 -0400107 up_write(&EXT4_I(inode)->i_data_sem);
Amir Goldstein8e8eaab2011-02-27 23:32:12 -0500108 ret = ext4_journal_restart(handle, nblocks);
Jan Kara487caee2009-08-17 22:17:20 -0400109 down_write(&EXT4_I(inode)->i_data_sem);
Aneesh Kumar K.Vfa5d1112009-11-02 18:50:49 -0500110 ext4_discard_preallocations(inode);
Jan Kara487caee2009-08-17 22:17:20 -0400111
112 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700113}
114
115/*
116 * Called at the last iput() if i_nlink is zero.
117 */
Al Viro0930fcc2010-06-07 13:16:22 -0400118void ext4_evict_inode(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700119{
120 handle_t *handle;
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400121 int err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700122
Theodore Ts'o7ff9c072010-11-08 13:51:33 -0500123 trace_ext4_evict_inode(inode);
Jiaying Zhang2581fdc2011-08-13 12:17:13 -0400124
Jiaying Zhang2581fdc2011-08-13 12:17:13 -0400125 ext4_ioend_wait(inode);
126
Al Viro0930fcc2010-06-07 13:16:22 -0400127 if (inode->i_nlink) {
Jan Kara2d859db2011-07-26 09:07:11 -0400128 /*
129 * When journalling data dirty buffers are tracked only in the
130 * journal. So although mm thinks everything is clean and
131 * ready for reaping the inode might still have some pages to
132 * write in the running transaction or waiting to be
133 * checkpointed. Thus calling jbd2_journal_invalidatepage()
134 * (via truncate_inode_pages()) to discard these buffers can
135 * cause data loss. Also even if we did not discard these
136 * buffers, we would have no way to find them after the inode
137 * is reaped and thus user could see stale data if he tries to
138 * read them before the transaction is checkpointed. So be
139 * careful and force everything to disk here... We use
140 * ei->i_datasync_tid to store the newest transaction
141 * containing inode's data.
142 *
143 * Note that directories do not have this problem because they
144 * don't use page cache.
145 */
146 if (ext4_should_journal_data(inode) &&
147 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
148 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
149 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
150
151 jbd2_log_start_commit(journal, commit_tid);
152 jbd2_log_wait_commit(journal, commit_tid);
153 filemap_write_and_wait(&inode->i_data);
154 }
Al Viro0930fcc2010-06-07 13:16:22 -0400155 truncate_inode_pages(&inode->i_data, 0);
156 goto no_delete;
157 }
158
Christoph Hellwig907f4552010-03-03 09:05:06 -0500159 if (!is_bad_inode(inode))
Christoph Hellwig871a2932010-03-03 09:05:07 -0500160 dquot_initialize(inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -0500161
Jan Kara678aaf42008-07-11 19:27:31 -0400162 if (ext4_should_order_data(inode))
163 ext4_begin_ordered_truncate(inode, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700164 truncate_inode_pages(&inode->i_data, 0);
165
166 if (is_bad_inode(inode))
167 goto no_delete;
168
Theodore Ts'o9f125d62011-06-27 19:16:04 -0400169 handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700170 if (IS_ERR(handle)) {
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400171 ext4_std_error(inode->i_sb, PTR_ERR(handle));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700172 /*
173 * If we're going to skip the normal cleanup, we still need to
174 * make sure that the in-core orphan linked list is properly
175 * cleaned up.
176 */
Mingming Cao617ba132006-10-11 01:20:53 -0700177 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700178 goto no_delete;
179 }
180
181 if (IS_SYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -0500182 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700183 inode->i_size = 0;
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400184 err = ext4_mark_inode_dirty(handle, inode);
185 if (err) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500186 ext4_warning(inode->i_sb,
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400187 "couldn't mark inode dirty (err %d)", err);
188 goto stop_handle;
189 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700190 if (inode->i_blocks)
Mingming Cao617ba132006-10-11 01:20:53 -0700191 ext4_truncate(inode);
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400192
193 /*
194 * ext4_ext_truncate() doesn't reserve any slop when it
195 * restarts journal transactions; therefore there may not be
196 * enough credits left in the handle to remove the inode from
197 * the orphan list and set the dtime field.
198 */
Frank Mayhar03901312009-01-07 00:06:22 -0500199 if (!ext4_handle_has_enough_credits(handle, 3)) {
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400200 err = ext4_journal_extend(handle, 3);
201 if (err > 0)
202 err = ext4_journal_restart(handle, 3);
203 if (err != 0) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500204 ext4_warning(inode->i_sb,
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400205 "couldn't extend journal (err %d)", err);
206 stop_handle:
207 ext4_journal_stop(handle);
Theodore Ts'o45388212010-07-29 15:06:10 -0400208 ext4_orphan_del(NULL, inode);
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400209 goto no_delete;
210 }
211 }
212
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700213 /*
Mingming Cao617ba132006-10-11 01:20:53 -0700214 * Kill off the orphan record which ext4_truncate created.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700215 * AKPM: I think this can be inside the above `if'.
Mingming Cao617ba132006-10-11 01:20:53 -0700216 * Note that ext4_orphan_del() has to be able to cope with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700217 * deletion of a non-existent orphan - this is because we don't
Mingming Cao617ba132006-10-11 01:20:53 -0700218 * know if ext4_truncate() actually created an orphan record.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700219 * (Well, we could do this if we need to, but heck - it works)
220 */
Mingming Cao617ba132006-10-11 01:20:53 -0700221 ext4_orphan_del(handle, inode);
222 EXT4_I(inode)->i_dtime = get_seconds();
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700223
224 /*
225 * One subtle ordering requirement: if anything has gone wrong
226 * (transaction abort, IO errors, whatever), then we can still
227 * do these next steps (the fs will already have been marked as
228 * having errors), but we can't free the inode if the mark_dirty
229 * fails.
230 */
Mingming Cao617ba132006-10-11 01:20:53 -0700231 if (ext4_mark_inode_dirty(handle, inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700232 /* If that failed, just do the required in-core inode clear. */
Al Viro0930fcc2010-06-07 13:16:22 -0400233 ext4_clear_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700234 else
Mingming Cao617ba132006-10-11 01:20:53 -0700235 ext4_free_inode(handle, inode);
236 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700237 return;
238no_delete:
Al Viro0930fcc2010-06-07 13:16:22 -0400239 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700240}
241
Dmitry Monakhova9e7f442009-12-14 15:21:14 +0300242#ifdef CONFIG_QUOTA
243qsize_t *ext4_get_reserved_space(struct inode *inode)
Mingming Cao60e58e02009-01-22 18:13:05 +0100244{
Dmitry Monakhova9e7f442009-12-14 15:21:14 +0300245 return &EXT4_I(inode)->i_reserved_quota;
Mingming Cao60e58e02009-01-22 18:13:05 +0100246}
Dmitry Monakhova9e7f442009-12-14 15:21:14 +0300247#endif
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500248
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400249/*
250 * Calculate the number of metadata blocks need to reserve
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500251 * to allocate a block located at @lblock
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400252 */
Theodore Ts'o01f49d02011-01-10 12:13:03 -0500253static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400254{
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400255 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500256 return ext4_ext_calc_metadata_amount(inode, lblock);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400257
Amir Goldstein8bb2b242011-06-27 17:10:28 -0400258 return ext4_ind_calc_metadata_amount(inode, lblock);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400259}
260
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500261/*
262 * Called with i_data_sem down, which is important since we can call
263 * ext4_discard_preallocations() from here.
264 */
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500265void ext4_da_update_reserve_space(struct inode *inode,
266 int used, int quota_claim)
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400267{
268 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500269 struct ext4_inode_info *ei = EXT4_I(inode);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400270
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500271 spin_lock(&ei->i_block_reservation_lock);
Aditya Kalid8990242011-09-09 19:18:51 -0400272 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500273 if (unlikely(used > ei->i_reserved_data_blocks)) {
274 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
275 "with only %d reserved data blocks\n",
276 __func__, inode->i_ino, used,
277 ei->i_reserved_data_blocks);
278 WARN_ON(1);
279 used = ei->i_reserved_data_blocks;
Aneesh Kumar K.V6bc6e632008-10-10 09:39:00 -0400280 }
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400281
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500282 /* Update per-inode reservations */
283 ei->i_reserved_data_blocks -= used;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500284 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
Theodore Ts'o57042652011-09-09 18:56:51 -0400285 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400286 used + ei->i_allocated_meta_blocks);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500287 ei->i_allocated_meta_blocks = 0;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500288
289 if (ei->i_reserved_data_blocks == 0) {
290 /*
291 * We can release all of the reserved metadata blocks
292 * only when we have written all of the delayed
293 * allocation blocks.
294 */
Theodore Ts'o57042652011-09-09 18:56:51 -0400295 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400296 ei->i_reserved_meta_blocks);
Theodore Ts'oee5f4d92010-01-01 02:36:15 -0500297 ei->i_reserved_meta_blocks = 0;
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500298 ei->i_da_metadata_calc_len = 0;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500299 }
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400300 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao60e58e02009-01-22 18:13:05 +0100301
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400302 /* Update quota subsystem for data blocks */
303 if (quota_claim)
Aditya Kali7b415bf2011-09-09 19:04:51 -0400304 dquot_claim_block(inode, EXT4_C2B(sbi, used));
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400305 else {
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500306 /*
307 * We did fallocate with an offset that is already delayed
308 * allocated. So on delayed allocated writeback we should
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400309 * not re-claim the quota for fallocated blocks.
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500310 */
Aditya Kali7b415bf2011-09-09 19:04:51 -0400311 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500312 }
Aneesh Kumar K.Vd6014302009-03-27 22:36:43 -0400313
314 /*
315 * If we have done all the pending block allocations and if
316 * there aren't any writers on the inode, we can discard the
317 * inode's preallocations.
318 */
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500319 if ((ei->i_reserved_data_blocks == 0) &&
320 (atomic_read(&inode->i_writecount) == 0))
Aneesh Kumar K.Vd6014302009-03-27 22:36:43 -0400321 ext4_discard_preallocations(inode);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400322}
323
Theodore Ts'oe29136f2010-06-29 12:54:28 -0400324static int __check_block_validity(struct inode *inode, const char *func,
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400325 unsigned int line,
326 struct ext4_map_blocks *map)
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400327{
Theodore Ts'o24676da2010-05-16 21:00:00 -0400328 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
329 map->m_len)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400330 ext4_error_inode(inode, func, line, map->m_pblk,
331 "lblock %lu mapped to illegal pblock "
332 "(length %d)", (unsigned long) map->m_lblk,
333 map->m_len);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400334 return -EIO;
335 }
336 return 0;
337}
338
Theodore Ts'oe29136f2010-06-29 12:54:28 -0400339#define check_block_validity(inode, map) \
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400340 __check_block_validity((inode), __func__, __LINE__, (map))
Theodore Ts'oe29136f2010-06-29 12:54:28 -0400341
Mingming Caof5ab0d12008-02-25 15:29:55 -0500342/*
Theodore Ts'o1f945332009-09-30 22:57:41 -0400343 * Return the number of contiguous dirty pages in a given inode
344 * starting at page frame idx.
Theodore Ts'o55138e02009-09-29 13:31:31 -0400345 */
346static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
347 unsigned int max_pages)
348{
349 struct address_space *mapping = inode->i_mapping;
350 pgoff_t index;
351 struct pagevec pvec;
352 pgoff_t num = 0;
353 int i, nr_pages, done = 0;
354
355 if (max_pages == 0)
356 return 0;
357 pagevec_init(&pvec, 0);
358 while (!done) {
359 index = idx;
360 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
361 PAGECACHE_TAG_DIRTY,
362 (pgoff_t)PAGEVEC_SIZE);
363 if (nr_pages == 0)
364 break;
365 for (i = 0; i < nr_pages; i++) {
366 struct page *page = pvec.pages[i];
367 struct buffer_head *bh, *head;
368
369 lock_page(page);
370 if (unlikely(page->mapping != mapping) ||
371 !PageDirty(page) ||
372 PageWriteback(page) ||
373 page->index != idx) {
374 done = 1;
375 unlock_page(page);
376 break;
377 }
Theodore Ts'o1f945332009-09-30 22:57:41 -0400378 if (page_has_buffers(page)) {
379 bh = head = page_buffers(page);
380 do {
381 if (!buffer_delay(bh) &&
382 !buffer_unwritten(bh))
383 done = 1;
384 bh = bh->b_this_page;
385 } while (!done && (bh != head));
386 }
Theodore Ts'o55138e02009-09-29 13:31:31 -0400387 unlock_page(page);
388 if (done)
389 break;
390 idx++;
391 num++;
Eric Sandeen659c6002010-10-27 21:30:03 -0400392 if (num >= max_pages) {
393 done = 1;
Theodore Ts'o55138e02009-09-29 13:31:31 -0400394 break;
Eric Sandeen659c6002010-10-27 21:30:03 -0400395 }
Theodore Ts'o55138e02009-09-29 13:31:31 -0400396 }
397 pagevec_release(&pvec);
398 }
399 return num;
400}
401
402/*
Aditya Kali5356f262011-09-09 19:20:51 -0400403 * Sets the BH_Da_Mapped bit on the buffer heads corresponding to the given map.
404 */
405static void set_buffers_da_mapped(struct inode *inode,
406 struct ext4_map_blocks *map)
407{
408 struct address_space *mapping = inode->i_mapping;
409 struct pagevec pvec;
410 int i, nr_pages;
411 pgoff_t index, end;
412
413 index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
414 end = (map->m_lblk + map->m_len - 1) >>
415 (PAGE_CACHE_SHIFT - inode->i_blkbits);
416
417 pagevec_init(&pvec, 0);
418 while (index <= end) {
419 nr_pages = pagevec_lookup(&pvec, mapping, index,
420 min(end - index + 1,
421 (pgoff_t)PAGEVEC_SIZE));
422 if (nr_pages == 0)
423 break;
424 for (i = 0; i < nr_pages; i++) {
425 struct page *page = pvec.pages[i];
426 struct buffer_head *bh, *head;
427
428 if (unlikely(page->mapping != mapping) ||
429 !PageDirty(page))
430 break;
431
432 if (page_has_buffers(page)) {
433 bh = head = page_buffers(page);
434 do {
435 set_buffer_da_mapped(bh);
436 bh = bh->b_this_page;
437 } while (bh != head);
438 }
439 index++;
440 }
441 pagevec_release(&pvec);
442 }
443}
444
445/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400446 * The ext4_map_blocks() function tries to look up the requested blocks,
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400447 * and returns if the blocks are already mapped.
Mingming Caof5ab0d12008-02-25 15:29:55 -0500448 *
Mingming Caof5ab0d12008-02-25 15:29:55 -0500449 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
450 * and store the allocated blocks in the result buffer head and mark it
451 * mapped.
452 *
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400453 * If file type is extents based, it will call ext4_ext_map_blocks(),
454 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
Mingming Caof5ab0d12008-02-25 15:29:55 -0500455 * based files
456 *
457 * On success, it returns the number of blocks being mapped or allocate.
458 * if create==0 and the blocks are pre-allocated and uninitialized block,
459 * the result buffer head is unmapped. If the create ==1, it will make sure
460 * the buffer head is mapped.
461 *
462 * It returns 0 if plain look up failed (blocks have not been allocated), in
Tao Madf3ab172011-10-08 15:53:49 -0400463 * that case, buffer head is unmapped
Mingming Caof5ab0d12008-02-25 15:29:55 -0500464 *
465 * It returns the error in case of allocation failure.
466 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400467int ext4_map_blocks(handle_t *handle, struct inode *inode,
468 struct ext4_map_blocks *map, int flags)
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500469{
470 int retval;
Mingming Caof5ab0d12008-02-25 15:29:55 -0500471
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400472 map->m_flags = 0;
473 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
474 "logical block %lu\n", inode->i_ino, flags, map->m_len,
475 (unsigned long) map->m_lblk);
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500476 /*
Theodore Ts'ob920c752009-05-14 00:54:29 -0400477 * Try to see if we can get the block without requesting a new
478 * file system block.
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500479 */
480 down_read((&EXT4_I(inode)->i_data_sem));
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400481 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
Dmitry Monakhova4e5d882011-10-25 08:15:12 -0400482 retval = ext4_ext_map_blocks(handle, inode, map, flags &
483 EXT4_GET_BLOCKS_KEEP_SIZE);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500484 } else {
Dmitry Monakhova4e5d882011-10-25 08:15:12 -0400485 retval = ext4_ind_map_blocks(handle, inode, map, flags &
486 EXT4_GET_BLOCKS_KEEP_SIZE);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500487 }
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500488 up_read((&EXT4_I(inode)->i_data_sem));
Mingming Caof5ab0d12008-02-25 15:29:55 -0500489
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400490 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
Theodore Ts'oe29136f2010-06-29 12:54:28 -0400491 int ret = check_block_validity(inode, map);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400492 if (ret != 0)
493 return ret;
494 }
495
Mingming Caof5ab0d12008-02-25 15:29:55 -0500496 /* If it is only a block(s) look up */
Theodore Ts'oc2177052009-05-14 00:58:52 -0400497 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500498 return retval;
499
500 /*
Mingming Caof5ab0d12008-02-25 15:29:55 -0500501 * Returns if the blocks have already allocated
502 *
503 * Note that if blocks have been preallocated
Tao Madf3ab172011-10-08 15:53:49 -0400504 * ext4_ext_get_block() returns the create = 0
Mingming Caof5ab0d12008-02-25 15:29:55 -0500505 * with buffer head unmapped.
506 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400507 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
Mingming Caof5ab0d12008-02-25 15:29:55 -0500508 return retval;
509
510 /*
Aneesh Kumar K.V2a8964d2009-05-14 17:05:39 -0400511 * When we call get_blocks without the create flag, the
512 * BH_Unwritten flag could have gotten set if the blocks
513 * requested were part of a uninitialized extent. We need to
514 * clear this flag now that we are committed to convert all or
515 * part of the uninitialized extent to be an initialized
516 * extent. This is because we need to avoid the combination
517 * of BH_Unwritten and BH_Mapped flags being simultaneously
518 * set on the buffer_head.
519 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400520 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
Aneesh Kumar K.V2a8964d2009-05-14 17:05:39 -0400521
522 /*
Mingming Caof5ab0d12008-02-25 15:29:55 -0500523 * New blocks allocate and/or writing to uninitialized extent
524 * will possibly result in updating i_data, so we take
525 * the write lock of i_data_sem, and call get_blocks()
526 * with create == 1 flag.
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500527 */
528 down_write((&EXT4_I(inode)->i_data_sem));
Mingming Caod2a17632008-07-14 17:52:37 -0400529
530 /*
531 * if the caller is from delayed allocation writeout path
532 * we have already reserved fs blocks for allocation
533 * let the underlying get_block() function know to
534 * avoid double accounting
535 */
Theodore Ts'oc2177052009-05-14 00:58:52 -0400536 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
Theodore Ts'of2321092011-01-10 12:12:36 -0500537 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500538 /*
539 * We need to check for EXT4 here because migrate
540 * could have changed the inode type in between
541 */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400542 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400543 retval = ext4_ext_map_blocks(handle, inode, map, flags);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500544 } else {
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400545 retval = ext4_ind_map_blocks(handle, inode, map, flags);
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -0400546
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400547 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -0400548 /*
549 * We allocated new blocks which will result in
550 * i_data's format changing. Force the migrate
551 * to fail by clearing migrate flags
552 */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -0500553 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -0400554 }
Mingming Caod2a17632008-07-14 17:52:37 -0400555
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500556 /*
557 * Update reserved blocks/metadata blocks after successful
558 * block allocation which had been deferred till now. We don't
559 * support fallocate for non extent files. So we can update
560 * reserve space here.
561 */
562 if ((retval > 0) &&
Aneesh Kumar K.V1296cc82010-01-15 01:27:59 -0500563 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500564 ext4_da_update_reserve_space(inode, retval, 1);
565 }
Aditya Kali5356f262011-09-09 19:20:51 -0400566 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
Theodore Ts'of2321092011-01-10 12:12:36 -0500567 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -0400568
Aditya Kali5356f262011-09-09 19:20:51 -0400569 /* If we have successfully mapped the delayed allocated blocks,
570 * set the BH_Da_Mapped bit on them. Its important to do this
571 * under the protection of i_data_sem.
572 */
573 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
574 set_buffers_da_mapped(inode, map);
575 }
576
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500577 up_write((&EXT4_I(inode)->i_data_sem));
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400578 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
Theodore Ts'oe29136f2010-06-29 12:54:28 -0400579 int ret = check_block_validity(inode, map);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400580 if (ret != 0)
581 return ret;
582 }
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500583 return retval;
584}
585
Mingming Caof3bd1f32008-08-19 22:16:03 -0400586/* Maximum number of blocks we map for direct IO at once. */
587#define DIO_MAX_BLOCKS 4096
588
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400589static int _ext4_get_block(struct inode *inode, sector_t iblock,
590 struct buffer_head *bh, int flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700591{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -0800592 handle_t *handle = ext4_journal_current_handle();
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400593 struct ext4_map_blocks map;
Jan Kara7fb54092008-02-10 01:08:38 -0500594 int ret = 0, started = 0;
Mingming Caof3bd1f32008-08-19 22:16:03 -0400595 int dio_credits;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700596
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400597 map.m_lblk = iblock;
598 map.m_len = bh->b_size >> inode->i_blkbits;
599
600 if (flags && !handle) {
Jan Kara7fb54092008-02-10 01:08:38 -0500601 /* Direct IO write... */
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400602 if (map.m_len > DIO_MAX_BLOCKS)
603 map.m_len = DIO_MAX_BLOCKS;
604 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
Mingming Caof3bd1f32008-08-19 22:16:03 -0400605 handle = ext4_journal_start(inode, dio_credits);
Jan Kara7fb54092008-02-10 01:08:38 -0500606 if (IS_ERR(handle)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700607 ret = PTR_ERR(handle);
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400608 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700609 }
Jan Kara7fb54092008-02-10 01:08:38 -0500610 started = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700611 }
612
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400613 ret = ext4_map_blocks(handle, inode, &map, flags);
Jan Kara7fb54092008-02-10 01:08:38 -0500614 if (ret > 0) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400615 map_bh(bh, inode->i_sb, map.m_pblk);
616 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
617 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
Jan Kara7fb54092008-02-10 01:08:38 -0500618 ret = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700619 }
Jan Kara7fb54092008-02-10 01:08:38 -0500620 if (started)
621 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700622 return ret;
623}
624
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400625int ext4_get_block(struct inode *inode, sector_t iblock,
626 struct buffer_head *bh, int create)
627{
628 return _ext4_get_block(inode, iblock, bh,
629 create ? EXT4_GET_BLOCKS_CREATE : 0);
630}
631
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700632/*
633 * `handle' can be NULL if create is zero
634 */
Mingming Cao617ba132006-10-11 01:20:53 -0700635struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500636 ext4_lblk_t block, int create, int *errp)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700637{
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400638 struct ext4_map_blocks map;
639 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700640 int fatal = 0, err;
641
642 J_ASSERT(handle != NULL || create == 0);
643
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400644 map.m_lblk = block;
645 map.m_len = 1;
646 err = ext4_map_blocks(handle, inode, &map,
647 create ? EXT4_GET_BLOCKS_CREATE : 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700648
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400649 if (err < 0)
650 *errp = err;
651 if (err <= 0)
652 return NULL;
653 *errp = 0;
654
655 bh = sb_getblk(inode->i_sb, map.m_pblk);
656 if (!bh) {
657 *errp = -EIO;
658 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700659 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400660 if (map.m_flags & EXT4_MAP_NEW) {
661 J_ASSERT(create != 0);
662 J_ASSERT(handle != NULL);
663
664 /*
665 * Now that we do not always journal data, we should
666 * keep in mind whether this should always journal the
667 * new buffer as metadata. For now, regular file
668 * writes use ext4_get_block instead, so it's not a
669 * problem.
670 */
671 lock_buffer(bh);
672 BUFFER_TRACE(bh, "call get_create_access");
673 fatal = ext4_journal_get_create_access(handle, bh);
674 if (!fatal && !buffer_uptodate(bh)) {
675 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
676 set_buffer_uptodate(bh);
677 }
678 unlock_buffer(bh);
679 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
680 err = ext4_handle_dirty_metadata(handle, inode, bh);
681 if (!fatal)
682 fatal = err;
683 } else {
684 BUFFER_TRACE(bh, "not a new buffer");
685 }
686 if (fatal) {
687 *errp = fatal;
688 brelse(bh);
689 bh = NULL;
690 }
691 return bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700692}
693
Mingming Cao617ba132006-10-11 01:20:53 -0700694struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500695 ext4_lblk_t block, int create, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700696{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400697 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700698
Mingming Cao617ba132006-10-11 01:20:53 -0700699 bh = ext4_getblk(handle, inode, block, create, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700700 if (!bh)
701 return bh;
702 if (buffer_uptodate(bh))
703 return bh;
Christoph Hellwig65299a32011-08-23 14:50:29 +0200704 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700705 wait_on_buffer(bh);
706 if (buffer_uptodate(bh))
707 return bh;
708 put_bh(bh);
709 *err = -EIO;
710 return NULL;
711}
712
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400713static int walk_page_buffers(handle_t *handle,
714 struct buffer_head *head,
715 unsigned from,
716 unsigned to,
717 int *partial,
718 int (*fn)(handle_t *handle,
719 struct buffer_head *bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700720{
721 struct buffer_head *bh;
722 unsigned block_start, block_end;
723 unsigned blocksize = head->b_size;
724 int err, ret = 0;
725 struct buffer_head *next;
726
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400727 for (bh = head, block_start = 0;
728 ret == 0 && (bh != head || !block_start);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400729 block_start = block_end, bh = next) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700730 next = bh->b_this_page;
731 block_end = block_start + blocksize;
732 if (block_end <= from || block_start >= to) {
733 if (partial && !buffer_uptodate(bh))
734 *partial = 1;
735 continue;
736 }
737 err = (*fn)(handle, bh);
738 if (!ret)
739 ret = err;
740 }
741 return ret;
742}
743
744/*
745 * To preserve ordering, it is essential that the hole instantiation and
746 * the data write be encapsulated in a single transaction. We cannot
Mingming Cao617ba132006-10-11 01:20:53 -0700747 * close off a transaction and start a new one between the ext4_get_block()
Mingming Caodab291a2006-10-11 01:21:01 -0700748 * and the commit_write(). So doing the jbd2_journal_start at the start of
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700749 * prepare_write() is the right place.
750 *
Mingming Cao617ba132006-10-11 01:20:53 -0700751 * Also, this function can nest inside ext4_writepage() ->
752 * block_write_full_page(). In that case, we *know* that ext4_writepage()
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700753 * has generated enough buffer credits to do the whole page. So we won't
754 * block on the journal in that case, which is good, because the caller may
755 * be PF_MEMALLOC.
756 *
Mingming Cao617ba132006-10-11 01:20:53 -0700757 * By accident, ext4 can be reentered when a transaction is open via
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700758 * quota file writes. If we were to commit the transaction while thus
759 * reentered, there can be a deadlock - we would be holding a quota
760 * lock, and the commit would never complete if another thread had a
761 * transaction open and was blocking on the quota lock - a ranking
762 * violation.
763 *
Mingming Caodab291a2006-10-11 01:21:01 -0700764 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700765 * will _not_ run commit under these circumstances because handle->h_ref
766 * is elevated. We'll still have enough credits for the tiny quotafile
767 * write.
768 */
769static int do_journal_get_write_access(handle_t *handle,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400770 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700771{
Jan Kara56d35a42010-08-05 14:41:42 -0400772 int dirty = buffer_dirty(bh);
773 int ret;
774
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700775 if (!buffer_mapped(bh) || buffer_freed(bh))
776 return 0;
Jan Kara56d35a42010-08-05 14:41:42 -0400777 /*
Christoph Hellwigebdec242010-10-06 10:47:23 +0200778 * __block_write_begin() could have dirtied some buffers. Clean
Jan Kara56d35a42010-08-05 14:41:42 -0400779 * the dirty bit as jbd2_journal_get_write_access() could complain
780 * otherwise about fs integrity issues. Setting of the dirty bit
Christoph Hellwigebdec242010-10-06 10:47:23 +0200781 * by __block_write_begin() isn't a real problem here as we clear
Jan Kara56d35a42010-08-05 14:41:42 -0400782 * the bit before releasing a page lock and thus writeback cannot
783 * ever write the buffer.
784 */
785 if (dirty)
786 clear_buffer_dirty(bh);
787 ret = ext4_journal_get_write_access(handle, bh);
788 if (!ret && dirty)
789 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
790 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700791}
792
Jiaying Zhang744692d2010-03-04 16:14:02 -0500793static int ext4_get_block_write(struct inode *inode, sector_t iblock,
794 struct buffer_head *bh_result, int create);
Nick Pigginbfc1af62007-10-16 01:25:05 -0700795static int ext4_write_begin(struct file *file, struct address_space *mapping,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400796 loff_t pos, unsigned len, unsigned flags,
797 struct page **pagep, void **fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700798{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400799 struct inode *inode = mapping->host;
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -0400800 int ret, needed_blocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700801 handle_t *handle;
802 int retries = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400803 struct page *page;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400804 pgoff_t index;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400805 unsigned from, to;
Nick Pigginbfc1af62007-10-16 01:25:05 -0700806
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400807 trace_ext4_write_begin(inode, pos, len, flags);
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -0400808 /*
809 * Reserve one block more for addition to orphan list in case
810 * we allocate blocks but write fails for some reason
811 */
812 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400813 index = pos >> PAGE_CACHE_SHIFT;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400814 from = pos & (PAGE_CACHE_SIZE - 1);
815 to = from + len;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700816
817retry:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400818 handle = ext4_journal_start(inode, needed_blocks);
819 if (IS_ERR(handle)) {
820 ret = PTR_ERR(handle);
821 goto out;
Nick Pigginbfc1af62007-10-16 01:25:05 -0700822 }
823
Jan Karaebd36102009-02-22 21:09:59 -0500824 /* We cannot recurse into the filesystem as the transaction is already
825 * started */
826 flags |= AOP_FLAG_NOFS;
827
Nick Piggin54566b22009-01-04 12:00:53 -0800828 page = grab_cache_page_write_begin(mapping, index, flags);
Jan Karacf108bc2008-07-11 19:27:31 -0400829 if (!page) {
830 ext4_journal_stop(handle);
831 ret = -ENOMEM;
832 goto out;
833 }
834 *pagep = page;
835
Jiaying Zhang744692d2010-03-04 16:14:02 -0500836 if (ext4_should_dioread_nolock(inode))
Christoph Hellwig6e1db882010-06-04 11:29:57 +0200837 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
Jiaying Zhang744692d2010-03-04 16:14:02 -0500838 else
Christoph Hellwig6e1db882010-06-04 11:29:57 +0200839 ret = __block_write_begin(page, pos, len, ext4_get_block);
Nick Pigginbfc1af62007-10-16 01:25:05 -0700840
841 if (!ret && ext4_should_journal_data(inode)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700842 ret = walk_page_buffers(handle, page_buffers(page),
843 from, to, NULL, do_journal_get_write_access);
844 }
Nick Pigginbfc1af62007-10-16 01:25:05 -0700845
846 if (ret) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400847 unlock_page(page);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400848 page_cache_release(page);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -0400849 /*
Christoph Hellwig6e1db882010-06-04 11:29:57 +0200850 * __block_write_begin may have instantiated a few blocks
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -0400851 * outside i_size. Trim these off again. Don't need
852 * i_size_read because we hold i_mutex.
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -0400853 *
854 * Add inode to orphan list in case we crash before
855 * truncate finishes
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -0400856 */
Jan Karaffacfa72009-07-13 16:22:22 -0400857 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -0400858 ext4_orphan_add(handle, inode);
859
860 ext4_journal_stop(handle);
861 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -0500862 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400863 /*
Jan Karaffacfa72009-07-13 16:22:22 -0400864 * If truncate failed early the inode might
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -0400865 * still be on the orphan list; we need to
866 * make sure the inode is removed from the
867 * orphan list in that case.
868 */
869 if (inode->i_nlink)
870 ext4_orphan_del(NULL, inode);
871 }
Nick Pigginbfc1af62007-10-16 01:25:05 -0700872 }
873
Mingming Cao617ba132006-10-11 01:20:53 -0700874 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700875 goto retry;
Andrew Morton7479d2b2007-04-01 23:49:44 -0700876out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700877 return ret;
878}
879
Nick Pigginbfc1af62007-10-16 01:25:05 -0700880/* For write_end() in data=journal mode */
881static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700882{
883 if (!buffer_mapped(bh) || buffer_freed(bh))
884 return 0;
885 set_buffer_uptodate(bh);
Frank Mayhar03901312009-01-07 00:06:22 -0500886 return ext4_handle_dirty_metadata(handle, NULL, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700887}
888
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -0400889static int ext4_generic_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400890 struct address_space *mapping,
891 loff_t pos, unsigned len, unsigned copied,
892 struct page *page, void *fsdata)
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -0400893{
894 int i_size_changed = 0;
895 struct inode *inode = mapping->host;
896 handle_t *handle = ext4_journal_current_handle();
897
898 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
899
900 /*
901 * No need to use i_size_read() here, the i_size
902 * cannot change under us because we hold i_mutex.
903 *
904 * But it's important to update i_size while still holding page lock:
905 * page writeout could otherwise come in and zero beyond i_size.
906 */
907 if (pos + copied > inode->i_size) {
908 i_size_write(inode, pos + copied);
909 i_size_changed = 1;
910 }
911
912 if (pos + copied > EXT4_I(inode)->i_disksize) {
913 /* We need to mark inode dirty even if
914 * new_i_size is less that inode->i_size
915 * bu greater than i_disksize.(hint delalloc)
916 */
917 ext4_update_i_disksize(inode, (pos + copied));
918 i_size_changed = 1;
919 }
920 unlock_page(page);
921 page_cache_release(page);
922
923 /*
924 * Don't mark the inode dirty under page lock. First, it unnecessarily
925 * makes the holding time of page lock longer. Second, it forces lock
926 * ordering of page lock and transaction start for journaling
927 * filesystems.
928 */
929 if (i_size_changed)
930 ext4_mark_inode_dirty(handle, inode);
931
932 return copied;
933}
934
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700935/*
936 * We need to pick up the new inode size which generic_commit_write gave us
937 * `file' can be NULL - eg, when called from page_symlink().
938 *
Mingming Cao617ba132006-10-11 01:20:53 -0700939 * ext4 never places buffers on inode->i_mapping->private_list. metadata
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700940 * buffers are managed internally.
941 */
Nick Pigginbfc1af62007-10-16 01:25:05 -0700942static int ext4_ordered_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400943 struct address_space *mapping,
944 loff_t pos, unsigned len, unsigned copied,
945 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700946{
Mingming Cao617ba132006-10-11 01:20:53 -0700947 handle_t *handle = ext4_journal_current_handle();
Jan Karacf108bc2008-07-11 19:27:31 -0400948 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700949 int ret = 0, ret2;
950
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400951 trace_ext4_ordered_write_end(inode, pos, len, copied);
Jan Kara678aaf42008-07-11 19:27:31 -0400952 ret = ext4_jbd2_file_inode(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700953
954 if (ret == 0) {
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -0400955 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
Nick Pigginbfc1af62007-10-16 01:25:05 -0700956 page, fsdata);
Roel Kluinf8a87d82008-04-29 22:01:18 -0400957 copied = ret2;
Jan Karaffacfa72009-07-13 16:22:22 -0400958 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -0400959 /* if we have allocated more blocks and copied
960 * less. We will have blocks allocated outside
961 * inode->i_size. So truncate them
962 */
963 ext4_orphan_add(handle, inode);
Roel Kluinf8a87d82008-04-29 22:01:18 -0400964 if (ret2 < 0)
965 ret = ret2;
Akira Fujita09e08342011-10-20 18:56:10 -0400966 } else {
967 unlock_page(page);
968 page_cache_release(page);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700969 }
Akira Fujita09e08342011-10-20 18:56:10 -0400970
Mingming Cao617ba132006-10-11 01:20:53 -0700971 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700972 if (!ret)
973 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -0700974
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -0400975 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -0500976 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400977 /*
Jan Karaffacfa72009-07-13 16:22:22 -0400978 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -0400979 * on the orphan list; we need to make sure the inode
980 * is removed from the orphan list in that case.
981 */
982 if (inode->i_nlink)
983 ext4_orphan_del(NULL, inode);
984 }
985
986
Nick Pigginbfc1af62007-10-16 01:25:05 -0700987 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700988}
989
Nick Pigginbfc1af62007-10-16 01:25:05 -0700990static int ext4_writeback_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400991 struct address_space *mapping,
992 loff_t pos, unsigned len, unsigned copied,
993 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700994{
Mingming Cao617ba132006-10-11 01:20:53 -0700995 handle_t *handle = ext4_journal_current_handle();
Jan Karacf108bc2008-07-11 19:27:31 -0400996 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700997 int ret = 0, ret2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700998
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400999 trace_ext4_writeback_write_end(inode, pos, len, copied);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001000 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001001 page, fsdata);
Roel Kluinf8a87d82008-04-29 22:01:18 -04001002 copied = ret2;
Jan Karaffacfa72009-07-13 16:22:22 -04001003 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001004 /* if we have allocated more blocks and copied
1005 * less. We will have blocks allocated outside
1006 * inode->i_size. So truncate them
1007 */
1008 ext4_orphan_add(handle, inode);
1009
Roel Kluinf8a87d82008-04-29 22:01:18 -04001010 if (ret2 < 0)
1011 ret = ret2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001012
Mingming Cao617ba132006-10-11 01:20:53 -07001013 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001014 if (!ret)
1015 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001016
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001017 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -05001018 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001019 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001020 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001021 * on the orphan list; we need to make sure the inode
1022 * is removed from the orphan list in that case.
1023 */
1024 if (inode->i_nlink)
1025 ext4_orphan_del(NULL, inode);
1026 }
1027
Nick Pigginbfc1af62007-10-16 01:25:05 -07001028 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001029}
1030
Nick Pigginbfc1af62007-10-16 01:25:05 -07001031static int ext4_journalled_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001032 struct address_space *mapping,
1033 loff_t pos, unsigned len, unsigned copied,
1034 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001035{
Mingming Cao617ba132006-10-11 01:20:53 -07001036 handle_t *handle = ext4_journal_current_handle();
Nick Pigginbfc1af62007-10-16 01:25:05 -07001037 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001038 int ret = 0, ret2;
1039 int partial = 0;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001040 unsigned from, to;
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001041 loff_t new_i_size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001042
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001043 trace_ext4_journalled_write_end(inode, pos, len, copied);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001044 from = pos & (PAGE_CACHE_SIZE - 1);
1045 to = from + len;
1046
Curt Wohlgemuth441c8502011-08-13 11:25:18 -04001047 BUG_ON(!ext4_handle_valid(handle));
1048
Nick Pigginbfc1af62007-10-16 01:25:05 -07001049 if (copied < len) {
1050 if (!PageUptodate(page))
1051 copied = 0;
1052 page_zero_new_buffers(page, from+copied, to);
1053 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001054
1055 ret = walk_page_buffers(handle, page_buffers(page), from,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001056 to, &partial, write_end_fn);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001057 if (!partial)
1058 SetPageUptodate(page);
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001059 new_i_size = pos + copied;
1060 if (new_i_size > inode->i_size)
Nick Pigginbfc1af62007-10-16 01:25:05 -07001061 i_size_write(inode, pos+copied);
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05001062 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
Jan Kara2d859db2011-07-26 09:07:11 -04001063 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001064 if (new_i_size > EXT4_I(inode)->i_disksize) {
1065 ext4_update_i_disksize(inode, new_i_size);
Mingming Cao617ba132006-10-11 01:20:53 -07001066 ret2 = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001067 if (!ret)
1068 ret = ret2;
1069 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001070
Jan Karacf108bc2008-07-11 19:27:31 -04001071 unlock_page(page);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001072 page_cache_release(page);
Jan Karaffacfa72009-07-13 16:22:22 -04001073 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001074 /* if we have allocated more blocks and copied
1075 * less. We will have blocks allocated outside
1076 * inode->i_size. So truncate them
1077 */
1078 ext4_orphan_add(handle, inode);
1079
Mingming Cao617ba132006-10-11 01:20:53 -07001080 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001081 if (!ret)
1082 ret = ret2;
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001083 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -05001084 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001085 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001086 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001087 * on the orphan list; we need to make sure the inode
1088 * is removed from the orphan list in that case.
1089 */
1090 if (inode->i_nlink)
1091 ext4_orphan_del(NULL, inode);
1092 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001093
1094 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001095}
Mingming Caod2a17632008-07-14 17:52:37 -04001096
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001097/*
Aditya Kali7b415bf2011-09-09 19:04:51 -04001098 * Reserve a single cluster located at lblock
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001099 */
Theodore Ts'o01f49d02011-01-10 12:13:03 -05001100static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
Mingming Caod2a17632008-07-14 17:52:37 -04001101{
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001102 int retries = 0;
Mingming Cao60e58e02009-01-22 18:13:05 +01001103 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001104 struct ext4_inode_info *ei = EXT4_I(inode);
Aditya Kali7b415bf2011-09-09 19:04:51 -04001105 unsigned int md_needed;
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001106 int ret;
Mingming Caod2a17632008-07-14 17:52:37 -04001107
1108 /*
1109 * recalculate the amount of metadata blocks to reserve
1110 * in order to allocate nrblocks
1111 * worse case is one extent per block
1112 */
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001113repeat:
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001114 spin_lock(&ei->i_block_reservation_lock);
Aditya Kali7b415bf2011-09-09 19:04:51 -04001115 md_needed = EXT4_NUM_B2C(sbi,
1116 ext4_calc_metadata_amount(inode, lblock));
Theodore Ts'of8ec9d62010-01-01 01:00:21 -05001117 trace_ext4_da_reserve_space(inode, md_needed);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001118 spin_unlock(&ei->i_block_reservation_lock);
Mingming Caod2a17632008-07-14 17:52:37 -04001119
Mingming Cao60e58e02009-01-22 18:13:05 +01001120 /*
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001121 * We will charge metadata quota at writeout time; this saves
1122 * us from metadata over-estimation, though we may go over by
1123 * a small amount in the end. Here we just reserve for data.
Mingming Cao60e58e02009-01-22 18:13:05 +01001124 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04001125 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001126 if (ret)
1127 return ret;
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001128 /*
1129 * We do still charge estimated metadata to the sb though;
1130 * we cannot afford to run out of free blocks.
1131 */
Theodore Ts'oe7d5f312011-09-09 19:14:51 -04001132 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
Aditya Kali7b415bf2011-09-09 19:04:51 -04001133 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001134 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1135 yield();
1136 goto repeat;
1137 }
Mingming Caod2a17632008-07-14 17:52:37 -04001138 return -ENOSPC;
1139 }
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001140 spin_lock(&ei->i_block_reservation_lock);
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001141 ei->i_reserved_data_blocks++;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001142 ei->i_reserved_meta_blocks += md_needed;
1143 spin_unlock(&ei->i_block_reservation_lock);
Dmitry Monakhov39bc6802009-12-10 16:36:27 +00001144
Mingming Caod2a17632008-07-14 17:52:37 -04001145 return 0; /* success */
1146}
1147
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001148static void ext4_da_release_space(struct inode *inode, int to_free)
Mingming Caod2a17632008-07-14 17:52:37 -04001149{
1150 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001151 struct ext4_inode_info *ei = EXT4_I(inode);
Mingming Caod2a17632008-07-14 17:52:37 -04001152
Mingming Caocd213222008-08-19 22:16:59 -04001153 if (!to_free)
1154 return; /* Nothing to release, exit */
1155
Mingming Caod2a17632008-07-14 17:52:37 -04001156 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Caocd213222008-08-19 22:16:59 -04001157
Li Zefan5a58ec82010-05-17 02:00:00 -04001158 trace_ext4_da_release_space(inode, to_free);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001159 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
Mingming Caocd213222008-08-19 22:16:59 -04001160 /*
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001161 * if there aren't enough reserved blocks, then the
1162 * counter is messed up somewhere. Since this
1163 * function is called from invalidate page, it's
1164 * harmless to return without any action.
Mingming Caocd213222008-08-19 22:16:59 -04001165 */
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001166 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1167 "ino %lu, to_free %d with only %d reserved "
1168 "data blocks\n", inode->i_ino, to_free,
1169 ei->i_reserved_data_blocks);
1170 WARN_ON(1);
1171 to_free = ei->i_reserved_data_blocks;
1172 }
1173 ei->i_reserved_data_blocks -= to_free;
1174
1175 if (ei->i_reserved_data_blocks == 0) {
1176 /*
1177 * We can release all of the reserved metadata blocks
1178 * only when we have written all of the delayed
1179 * allocation blocks.
Aditya Kali7b415bf2011-09-09 19:04:51 -04001180 * Note that in case of bigalloc, i_reserved_meta_blocks,
1181 * i_reserved_data_blocks, etc. refer to number of clusters.
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001182 */
Theodore Ts'o57042652011-09-09 18:56:51 -04001183 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001184 ei->i_reserved_meta_blocks);
Theodore Ts'oee5f4d92010-01-01 02:36:15 -05001185 ei->i_reserved_meta_blocks = 0;
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001186 ei->i_da_metadata_calc_len = 0;
Mingming Caocd213222008-08-19 22:16:59 -04001187 }
1188
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001189 /* update fs dirty data blocks counter */
Theodore Ts'o57042652011-09-09 18:56:51 -04001190 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
Mingming Caod2a17632008-07-14 17:52:37 -04001191
Mingming Caod2a17632008-07-14 17:52:37 -04001192 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao60e58e02009-01-22 18:13:05 +01001193
Aditya Kali7b415bf2011-09-09 19:04:51 -04001194 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
Mingming Caod2a17632008-07-14 17:52:37 -04001195}
1196
1197static void ext4_da_page_release_reservation(struct page *page,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001198 unsigned long offset)
Mingming Caod2a17632008-07-14 17:52:37 -04001199{
1200 int to_release = 0;
1201 struct buffer_head *head, *bh;
1202 unsigned int curr_off = 0;
Aditya Kali7b415bf2011-09-09 19:04:51 -04001203 struct inode *inode = page->mapping->host;
1204 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1205 int num_clusters;
Mingming Caod2a17632008-07-14 17:52:37 -04001206
1207 head = page_buffers(page);
1208 bh = head;
1209 do {
1210 unsigned int next_off = curr_off + bh->b_size;
1211
1212 if ((offset <= curr_off) && (buffer_delay(bh))) {
1213 to_release++;
1214 clear_buffer_delay(bh);
Aditya Kali5356f262011-09-09 19:20:51 -04001215 clear_buffer_da_mapped(bh);
Mingming Caod2a17632008-07-14 17:52:37 -04001216 }
1217 curr_off = next_off;
1218 } while ((bh = bh->b_this_page) != head);
Aditya Kali7b415bf2011-09-09 19:04:51 -04001219
1220 /* If we have released all the blocks belonging to a cluster, then we
1221 * need to release the reserved space for that cluster. */
1222 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1223 while (num_clusters > 0) {
1224 ext4_fsblk_t lblk;
1225 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1226 ((num_clusters - 1) << sbi->s_cluster_bits);
1227 if (sbi->s_cluster_ratio == 1 ||
1228 !ext4_find_delalloc_cluster(inode, lblk, 1))
1229 ext4_da_release_space(inode, 1);
1230
1231 num_clusters--;
1232 }
Mingming Caod2a17632008-07-14 17:52:37 -04001233}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001234
1235/*
Alex Tomas64769242008-07-11 19:27:31 -04001236 * Delayed allocation stuff
1237 */
1238
Alex Tomas64769242008-07-11 19:27:31 -04001239/*
1240 * mpage_da_submit_io - walks through extent of pages and try to write
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001241 * them with writepage() call back
Alex Tomas64769242008-07-11 19:27:31 -04001242 *
1243 * @mpd->inode: inode
1244 * @mpd->first_page: first page of the extent
1245 * @mpd->next_page: page after the last page of the extent
Alex Tomas64769242008-07-11 19:27:31 -04001246 *
1247 * By the time mpage_da_submit_io() is called we expect all blocks
1248 * to be allocated. this may be wrong if allocation failed.
1249 *
1250 * As pages are already locked by write_cache_pages(), we can't use it
1251 */
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001252static int mpage_da_submit_io(struct mpage_da_data *mpd,
1253 struct ext4_map_blocks *map)
Alex Tomas64769242008-07-11 19:27:31 -04001254{
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001255 struct pagevec pvec;
1256 unsigned long index, end;
1257 int ret = 0, err, nr_pages, i;
1258 struct inode *inode = mpd->inode;
1259 struct address_space *mapping = inode->i_mapping;
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001260 loff_t size = i_size_read(inode);
Theodore Ts'o3ecdb3a2010-10-27 21:30:10 -04001261 unsigned int len, block_start;
1262 struct buffer_head *bh, *page_bufs = NULL;
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001263 int journal_data = ext4_should_journal_data(inode);
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001264 sector_t pblock = 0, cur_logical = 0;
Theodore Ts'obd2d0212010-10-27 21:30:10 -04001265 struct ext4_io_submit io_submit;
Alex Tomas64769242008-07-11 19:27:31 -04001266
1267 BUG_ON(mpd->next_page <= mpd->first_page);
Theodore Ts'obd2d0212010-10-27 21:30:10 -04001268 memset(&io_submit, 0, sizeof(io_submit));
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001269 /*
1270 * We need to start from the first_page to the next_page - 1
1271 * to make sure we also write the mapped dirty buffer_heads.
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001272 * If we look at mpd->b_blocknr we would only be looking
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001273 * at the currently mapped buffer_heads.
1274 */
Alex Tomas64769242008-07-11 19:27:31 -04001275 index = mpd->first_page;
1276 end = mpd->next_page - 1;
1277
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001278 pagevec_init(&pvec, 0);
Alex Tomas64769242008-07-11 19:27:31 -04001279 while (index <= end) {
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001280 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
Alex Tomas64769242008-07-11 19:27:31 -04001281 if (nr_pages == 0)
1282 break;
1283 for (i = 0; i < nr_pages; i++) {
Theodore Ts'o97498952011-02-26 14:08:01 -05001284 int commit_write = 0, skip_page = 0;
Alex Tomas64769242008-07-11 19:27:31 -04001285 struct page *page = pvec.pages[i];
1286
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001287 index = page->index;
1288 if (index > end)
1289 break;
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001290
1291 if (index == size >> PAGE_CACHE_SHIFT)
1292 len = size & ~PAGE_CACHE_MASK;
1293 else
1294 len = PAGE_CACHE_SIZE;
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001295 if (map) {
1296 cur_logical = index << (PAGE_CACHE_SHIFT -
1297 inode->i_blkbits);
1298 pblock = map->m_pblk + (cur_logical -
1299 map->m_lblk);
1300 }
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001301 index++;
1302
1303 BUG_ON(!PageLocked(page));
1304 BUG_ON(PageWriteback(page));
1305
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001306 /*
1307 * If the page does not have buffers (for
1308 * whatever reason), try to create them using
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001309 * __block_write_begin. If this fails,
Theodore Ts'o97498952011-02-26 14:08:01 -05001310 * skip the page and move on.
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001311 */
1312 if (!page_has_buffers(page)) {
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001313 if (__block_write_begin(page, 0, len,
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001314 noalloc_get_block_write)) {
Theodore Ts'o97498952011-02-26 14:08:01 -05001315 skip_page:
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001316 unlock_page(page);
1317 continue;
1318 }
1319 commit_write = 1;
1320 }
Theodore Ts'o3ecdb3a2010-10-27 21:30:10 -04001321
1322 bh = page_bufs = page_buffers(page);
1323 block_start = 0;
1324 do {
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001325 if (!bh)
Theodore Ts'o97498952011-02-26 14:08:01 -05001326 goto skip_page;
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001327 if (map && (cur_logical >= map->m_lblk) &&
1328 (cur_logical <= (map->m_lblk +
1329 (map->m_len - 1)))) {
1330 if (buffer_delay(bh)) {
1331 clear_buffer_delay(bh);
1332 bh->b_blocknr = pblock;
1333 }
Aditya Kali5356f262011-09-09 19:20:51 -04001334 if (buffer_da_mapped(bh))
1335 clear_buffer_da_mapped(bh);
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001336 if (buffer_unwritten(bh) ||
1337 buffer_mapped(bh))
1338 BUG_ON(bh->b_blocknr != pblock);
1339 if (map->m_flags & EXT4_MAP_UNINIT)
1340 set_buffer_uninit(bh);
1341 clear_buffer_unwritten(bh);
1342 }
1343
Yongqiang Yang13a79a42011-12-13 21:51:55 -05001344 /*
1345 * skip page if block allocation undone and
1346 * block is dirty
1347 */
1348 if (ext4_bh_delay_or_unwritten(NULL, bh))
Theodore Ts'o97498952011-02-26 14:08:01 -05001349 skip_page = 1;
Theodore Ts'o3ecdb3a2010-10-27 21:30:10 -04001350 bh = bh->b_this_page;
1351 block_start += bh->b_size;
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001352 cur_logical++;
1353 pblock++;
1354 } while (bh != page_bufs);
1355
Theodore Ts'o97498952011-02-26 14:08:01 -05001356 if (skip_page)
1357 goto skip_page;
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001358
1359 if (commit_write)
1360 /* mark the buffer_heads as dirty & uptodate */
1361 block_commit_write(page, 0, len);
1362
Theodore Ts'o97498952011-02-26 14:08:01 -05001363 clear_page_dirty_for_io(page);
Theodore Ts'obd2d0212010-10-27 21:30:10 -04001364 /*
1365 * Delalloc doesn't support data journalling,
1366 * but eventually maybe we'll lift this
1367 * restriction.
1368 */
1369 if (unlikely(journal_data && PageChecked(page)))
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001370 err = __ext4_journalled_writepage(page, len);
Theodore Ts'o14490322010-12-14 15:27:50 -05001371 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
Theodore Ts'obd2d0212010-10-27 21:30:10 -04001372 err = ext4_bio_write_page(&io_submit, page,
1373 len, mpd->wbc);
Theodore Ts'o9dd75f12011-08-13 12:58:21 -04001374 else if (buffer_uninit(page_bufs)) {
1375 ext4_set_bh_endio(page_bufs, inode);
1376 err = block_write_full_page_endio(page,
1377 noalloc_get_block_write,
1378 mpd->wbc, ext4_end_io_buffer_write);
1379 } else
Theodore Ts'o14490322010-12-14 15:27:50 -05001380 err = block_write_full_page(page,
1381 noalloc_get_block_write, mpd->wbc);
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001382
1383 if (!err)
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001384 mpd->pages_written++;
Alex Tomas64769242008-07-11 19:27:31 -04001385 /*
1386 * In error case, we have to continue because
1387 * remaining pages are still locked
Alex Tomas64769242008-07-11 19:27:31 -04001388 */
1389 if (ret == 0)
1390 ret = err;
1391 }
1392 pagevec_release(&pvec);
1393 }
Theodore Ts'obd2d0212010-10-27 21:30:10 -04001394 ext4_io_submit(&io_submit);
Alex Tomas64769242008-07-11 19:27:31 -04001395 return ret;
1396}
1397
Curt Wohlgemuthc7f59382011-02-26 12:27:52 -05001398static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001399{
1400 int nr_pages, i;
1401 pgoff_t index, end;
1402 struct pagevec pvec;
1403 struct inode *inode = mpd->inode;
1404 struct address_space *mapping = inode->i_mapping;
1405
Curt Wohlgemuthc7f59382011-02-26 12:27:52 -05001406 index = mpd->first_page;
1407 end = mpd->next_page - 1;
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001408 while (index <= end) {
1409 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1410 if (nr_pages == 0)
1411 break;
1412 for (i = 0; i < nr_pages; i++) {
1413 struct page *page = pvec.pages[i];
Jan Kara9b1d09982010-03-03 16:19:32 -05001414 if (page->index > end)
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001415 break;
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001416 BUG_ON(!PageLocked(page));
1417 BUG_ON(PageWriteback(page));
1418 block_invalidatepage(page, 0);
1419 ClearPageUptodate(page);
1420 unlock_page(page);
1421 }
Jan Kara9b1d09982010-03-03 16:19:32 -05001422 index = pvec.pages[nr_pages - 1]->index + 1;
1423 pagevec_release(&pvec);
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001424 }
1425 return;
1426}
1427
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001428static void ext4_print_free_blocks(struct inode *inode)
1429{
1430 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o16939182009-09-26 17:43:59 -04001431 printk(KERN_CRIT "Total free blocks count %lld\n",
Theodore Ts'o5dee5432011-09-09 19:10:51 -04001432 EXT4_C2B(EXT4_SB(inode->i_sb),
1433 ext4_count_free_clusters(inode->i_sb)));
Theodore Ts'o16939182009-09-26 17:43:59 -04001434 printk(KERN_CRIT "Free/Dirty block details\n");
1435 printk(KERN_CRIT "free_blocks=%lld\n",
Theodore Ts'o57042652011-09-09 18:56:51 -04001436 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1437 percpu_counter_sum(&sbi->s_freeclusters_counter)));
Theodore Ts'o16939182009-09-26 17:43:59 -04001438 printk(KERN_CRIT "dirty_blocks=%lld\n",
Aditya Kali7b415bf2011-09-09 19:04:51 -04001439 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1440 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
Theodore Ts'o16939182009-09-26 17:43:59 -04001441 printk(KERN_CRIT "Block reservation details\n");
1442 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
1443 EXT4_I(inode)->i_reserved_data_blocks);
1444 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
1445 EXT4_I(inode)->i_reserved_meta_blocks);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001446 return;
1447}
1448
Theodore Ts'ob920c752009-05-14 00:54:29 -04001449/*
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001450 * mpage_da_map_and_submit - go through given space, map them
1451 * if necessary, and then submit them for I/O
Alex Tomas64769242008-07-11 19:27:31 -04001452 *
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001453 * @mpd - bh describing space
Alex Tomas64769242008-07-11 19:27:31 -04001454 *
1455 * The function skips space we know is already mapped to disk blocks.
1456 *
Alex Tomas64769242008-07-11 19:27:31 -04001457 */
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001458static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
Alex Tomas64769242008-07-11 19:27:31 -04001459{
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04001460 int err, blks, get_blocks_flags;
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001461 struct ext4_map_blocks map, *mapp = NULL;
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001462 sector_t next = mpd->b_blocknr;
1463 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1464 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1465 handle_t *handle = NULL;
Alex Tomas64769242008-07-11 19:27:31 -04001466
1467 /*
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001468 * If the blocks are mapped already, or we couldn't accumulate
1469 * any blocks, then proceed immediately to the submission stage.
Alex Tomas64769242008-07-11 19:27:31 -04001470 */
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001471 if ((mpd->b_size == 0) ||
1472 ((mpd->b_state & (1 << BH_Mapped)) &&
1473 !(mpd->b_state & (1 << BH_Delay)) &&
1474 !(mpd->b_state & (1 << BH_Unwritten))))
1475 goto submit_io;
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001476
1477 handle = ext4_journal_current_handle();
1478 BUG_ON(!handle);
1479
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -04001480 /*
Eric Sandeen79e83032010-07-27 11:56:07 -04001481 * Call ext4_map_blocks() to allocate any delayed allocation
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04001482 * blocks, or to convert an uninitialized extent to be
1483 * initialized (in the case where we have written into
1484 * one or more preallocated blocks).
1485 *
1486 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1487 * indicate that we are on the delayed allocation path. This
1488 * affects functions in many different parts of the allocation
1489 * call path. This flag exists primarily because we don't
Eric Sandeen79e83032010-07-27 11:56:07 -04001490 * want to change *many* call functions, so ext4_map_blocks()
Theodore Ts'of2321092011-01-10 12:12:36 -05001491 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04001492 * inode's allocation semaphore is taken.
1493 *
1494 * If the blocks in questions were delalloc blocks, set
1495 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1496 * variables are updated after the blocks have been allocated.
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -04001497 */
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001498 map.m_lblk = next;
1499 map.m_len = max_blocks;
Aneesh Kumar K.V1296cc82010-01-15 01:27:59 -05001500 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
Jiaying Zhang744692d2010-03-04 16:14:02 -05001501 if (ext4_should_dioread_nolock(mpd->inode))
1502 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04001503 if (mpd->b_state & (1 << BH_Delay))
Aneesh Kumar K.V1296cc82010-01-15 01:27:59 -05001504 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1505
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001506 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001507 if (blks < 0) {
Eric Sandeene3570632010-07-27 11:56:08 -04001508 struct super_block *sb = mpd->inode->i_sb;
1509
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001510 err = blks;
Theodore Ts'oed5bde02009-02-23 10:48:07 -05001511 /*
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001512 * If get block returns EAGAIN or ENOSPC and there
Theodore Ts'o97498952011-02-26 14:08:01 -05001513 * appears to be free blocks we will just let
1514 * mpage_da_submit_io() unlock all of the pages.
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001515 */
1516 if (err == -EAGAIN)
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001517 goto submit_io;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001518
Theodore Ts'o5dee5432011-09-09 19:10:51 -04001519 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001520 mpd->retval = err;
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001521 goto submit_io;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001522 }
1523
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001524 /*
Theodore Ts'oed5bde02009-02-23 10:48:07 -05001525 * get block failure will cause us to loop in
1526 * writepages, because a_ops->writepage won't be able
1527 * to make progress. The page will be redirtied by
1528 * writepage and writepages will again try to write
1529 * the same.
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001530 */
Eric Sandeene3570632010-07-27 11:56:08 -04001531 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1532 ext4_msg(sb, KERN_CRIT,
1533 "delayed block allocation failed for inode %lu "
1534 "at logical offset %llu with max blocks %zd "
1535 "with error %d", mpd->inode->i_ino,
1536 (unsigned long long) next,
1537 mpd->b_size >> mpd->inode->i_blkbits, err);
1538 ext4_msg(sb, KERN_CRIT,
1539 "This should not happen!! Data will be lost\n");
1540 if (err == -ENOSPC)
1541 ext4_print_free_blocks(mpd->inode);
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001542 }
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001543 /* invalidate all the pages */
Curt Wohlgemuthc7f59382011-02-26 12:27:52 -05001544 ext4_da_block_invalidatepages(mpd);
Curt Wohlgemuthe0fd9b92011-02-26 12:25:52 -05001545
1546 /* Mark this page range as having been completed */
1547 mpd->io_done = 1;
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001548 return;
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001549 }
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001550 BUG_ON(blks == 0);
1551
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001552 mapp = &map;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001553 if (map.m_flags & EXT4_MAP_NEW) {
1554 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1555 int i;
Alex Tomas64769242008-07-11 19:27:31 -04001556
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001557 for (i = 0; i < map.m_len; i++)
1558 unmap_underlying_metadata(bdev, map.m_pblk + i);
Alex Tomas64769242008-07-11 19:27:31 -04001559
Theodore Ts'odecbd912011-09-06 02:37:06 -04001560 if (ext4_should_order_data(mpd->inode)) {
1561 err = ext4_jbd2_file_inode(handle, mpd->inode);
Kazuya Mio8de49e62011-10-20 19:23:08 -04001562 if (err) {
Theodore Ts'odecbd912011-09-06 02:37:06 -04001563 /* Only if the journal is aborted */
Kazuya Mio8de49e62011-10-20 19:23:08 -04001564 mpd->retval = err;
1565 goto submit_io;
1566 }
Theodore Ts'odecbd912011-09-06 02:37:06 -04001567 }
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001568 }
1569
1570 /*
Jan Kara03f5d8b2009-06-09 00:17:05 -04001571 * Update on-disk size along with block allocation.
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001572 */
1573 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1574 if (disksize > i_size_read(mpd->inode))
1575 disksize = i_size_read(mpd->inode);
1576 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1577 ext4_update_i_disksize(mpd->inode, disksize);
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001578 err = ext4_mark_inode_dirty(handle, mpd->inode);
1579 if (err)
1580 ext4_error(mpd->inode->i_sb,
1581 "Failed to mark inode %lu dirty",
1582 mpd->inode->i_ino);
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001583 }
1584
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001585submit_io:
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001586 mpage_da_submit_io(mpd, mapp);
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001587 mpd->io_done = 1;
Alex Tomas64769242008-07-11 19:27:31 -04001588}
1589
Aneesh Kumar K.Vbf068ee2008-08-19 22:16:43 -04001590#define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1591 (1 << BH_Delay) | (1 << BH_Unwritten))
Alex Tomas64769242008-07-11 19:27:31 -04001592
1593/*
1594 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1595 *
1596 * @mpd->lbh - extent of blocks
1597 * @logical - logical number of the block in the file
1598 * @bh - bh of the block (used to access block's state)
1599 *
1600 * the function is used to collect contig. blocks in same state
1601 */
1602static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001603 sector_t logical, size_t b_size,
1604 unsigned long b_state)
Alex Tomas64769242008-07-11 19:27:31 -04001605{
Alex Tomas64769242008-07-11 19:27:31 -04001606 sector_t next;
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001607 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
Alex Tomas64769242008-07-11 19:27:31 -04001608
Eric Sandeenc445e3e2010-05-16 04:00:00 -04001609 /*
1610 * XXX Don't go larger than mballoc is willing to allocate
1611 * This is a stopgap solution. We eventually need to fold
1612 * mpage_da_submit_io() into this function and then call
Eric Sandeen79e83032010-07-27 11:56:07 -04001613 * ext4_map_blocks() multiple times in a loop
Eric Sandeenc445e3e2010-05-16 04:00:00 -04001614 */
1615 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1616 goto flush_it;
1617
Mingming Cao525f4ed2008-08-19 22:15:58 -04001618 /* check if thereserved journal credits might overflow */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001619 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
Mingming Cao525f4ed2008-08-19 22:15:58 -04001620 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1621 /*
1622 * With non-extent format we are limited by the journal
1623 * credit available. Total credit needed to insert
1624 * nrblocks contiguous blocks is dependent on the
1625 * nrblocks. So limit nrblocks.
1626 */
1627 goto flush_it;
1628 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1629 EXT4_MAX_TRANS_DATA) {
1630 /*
1631 * Adding the new buffer_head would make it cross the
1632 * allowed limit for which we have journal credit
1633 * reserved. So limit the new bh->b_size
1634 */
1635 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1636 mpd->inode->i_blkbits;
1637 /* we will do mpage_da_submit_io in the next loop */
1638 }
1639 }
Alex Tomas64769242008-07-11 19:27:31 -04001640 /*
1641 * First block in the extent
1642 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001643 if (mpd->b_size == 0) {
1644 mpd->b_blocknr = logical;
1645 mpd->b_size = b_size;
1646 mpd->b_state = b_state & BH_FLAGS;
Alex Tomas64769242008-07-11 19:27:31 -04001647 return;
1648 }
1649
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001650 next = mpd->b_blocknr + nrblocks;
Alex Tomas64769242008-07-11 19:27:31 -04001651 /*
1652 * Can we merge the block to our big extent?
1653 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001654 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1655 mpd->b_size += b_size;
Alex Tomas64769242008-07-11 19:27:31 -04001656 return;
1657 }
1658
Mingming Cao525f4ed2008-08-19 22:15:58 -04001659flush_it:
Alex Tomas64769242008-07-11 19:27:31 -04001660 /*
1661 * We couldn't merge the block to our extent, so we
1662 * need to flush current extent and start new one
1663 */
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001664 mpage_da_map_and_submit(mpd);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001665 return;
Alex Tomas64769242008-07-11 19:27:31 -04001666}
1667
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04001668static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04001669{
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04001670 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04001671}
1672
Alex Tomas64769242008-07-11 19:27:31 -04001673/*
Aditya Kali5356f262011-09-09 19:20:51 -04001674 * This function is grabs code from the very beginning of
1675 * ext4_map_blocks, but assumes that the caller is from delayed write
1676 * time. This function looks up the requested blocks and sets the
1677 * buffer delay bit under the protection of i_data_sem.
1678 */
1679static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1680 struct ext4_map_blocks *map,
1681 struct buffer_head *bh)
1682{
1683 int retval;
1684 sector_t invalid_block = ~((sector_t) 0xffff);
1685
1686 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1687 invalid_block = ~0;
1688
1689 map->m_flags = 0;
1690 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1691 "logical block %lu\n", inode->i_ino, map->m_len,
1692 (unsigned long) map->m_lblk);
1693 /*
1694 * Try to see if we can get the block without requesting a new
1695 * file system block.
1696 */
1697 down_read((&EXT4_I(inode)->i_data_sem));
1698 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1699 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1700 else
1701 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1702
1703 if (retval == 0) {
1704 /*
1705 * XXX: __block_prepare_write() unmaps passed block,
1706 * is it OK?
1707 */
1708 /* If the block was allocated from previously allocated cluster,
1709 * then we dont need to reserve it again. */
1710 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1711 retval = ext4_da_reserve_space(inode, iblock);
1712 if (retval)
1713 /* not enough space to reserve */
1714 goto out_unlock;
1715 }
1716
1717 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1718 * and it should not appear on the bh->b_state.
1719 */
1720 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1721
1722 map_bh(bh, inode->i_sb, invalid_block);
1723 set_buffer_new(bh);
1724 set_buffer_delay(bh);
1725 }
1726
1727out_unlock:
1728 up_read((&EXT4_I(inode)->i_data_sem));
1729
1730 return retval;
1731}
1732
1733/*
Theodore Ts'ob920c752009-05-14 00:54:29 -04001734 * This is a special get_blocks_t callback which is used by
1735 * ext4_da_write_begin(). It will either return mapped block or
1736 * reserve space for a single block.
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04001737 *
1738 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1739 * We also have b_blocknr = -1 and b_bdev initialized properly
1740 *
1741 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1742 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1743 * initialized properly.
Alex Tomas64769242008-07-11 19:27:31 -04001744 */
1745static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001746 struct buffer_head *bh, int create)
Alex Tomas64769242008-07-11 19:27:31 -04001747{
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001748 struct ext4_map_blocks map;
Alex Tomas64769242008-07-11 19:27:31 -04001749 int ret = 0;
1750
1751 BUG_ON(create == 0);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001752 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1753
1754 map.m_lblk = iblock;
1755 map.m_len = 1;
Alex Tomas64769242008-07-11 19:27:31 -04001756
1757 /*
1758 * first, we need to know whether the block is allocated already
1759 * preallocated blocks are unmapped but should treated
1760 * the same as allocated blocks.
1761 */
Aditya Kali5356f262011-09-09 19:20:51 -04001762 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1763 if (ret <= 0)
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001764 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04001765
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001766 map_bh(bh, inode->i_sb, map.m_pblk);
1767 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1768
1769 if (buffer_unwritten(bh)) {
1770 /* A delayed write to unwritten bh should be marked
1771 * new and mapped. Mapped ensures that we don't do
1772 * get_block multiple times when we write to the same
1773 * offset and new ensures that we do proper zero out
1774 * for partial write.
1775 */
1776 set_buffer_new(bh);
Theodore Ts'oc8205632011-04-10 22:30:07 -04001777 set_buffer_mapped(bh);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001778 }
1779 return 0;
Alex Tomas64769242008-07-11 19:27:31 -04001780}
Mingming Cao61628a32008-07-11 19:27:31 -04001781
Theodore Ts'ob920c752009-05-14 00:54:29 -04001782/*
1783 * This function is used as a standard get_block_t calback function
1784 * when there is no desire to allocate any blocks. It is used as a
Christoph Hellwigebdec242010-10-06 10:47:23 +02001785 * callback function for block_write_begin() and block_write_full_page().
Christoph Hellwig206f7ab2010-06-14 14:42:49 -04001786 * These functions should only try to map a single block at a time.
Theodore Ts'ob920c752009-05-14 00:54:29 -04001787 *
1788 * Since this function doesn't do block allocations even if the caller
1789 * requests it by passing in create=1, it is critically important that
1790 * any caller checks to make sure that any buffer heads are returned
1791 * by this function are either all already mapped or marked for
Christoph Hellwig206f7ab2010-06-14 14:42:49 -04001792 * delayed allocation before calling block_write_full_page(). Otherwise,
1793 * b_blocknr could be left unitialized, and the page write functions will
1794 * be taken by surprise.
Theodore Ts'ob920c752009-05-14 00:54:29 -04001795 */
1796static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04001797 struct buffer_head *bh_result, int create)
1798{
Theodore Ts'oa2dc52b2009-05-12 13:51:29 -04001799 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001800 return _ext4_get_block(inode, iblock, bh_result, 0);
Mingming Cao61628a32008-07-11 19:27:31 -04001801}
1802
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001803static int bget_one(handle_t *handle, struct buffer_head *bh)
1804{
1805 get_bh(bh);
1806 return 0;
1807}
1808
1809static int bput_one(handle_t *handle, struct buffer_head *bh)
1810{
1811 put_bh(bh);
1812 return 0;
1813}
1814
1815static int __ext4_journalled_writepage(struct page *page,
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001816 unsigned int len)
1817{
1818 struct address_space *mapping = page->mapping;
1819 struct inode *inode = mapping->host;
1820 struct buffer_head *page_bufs;
1821 handle_t *handle = NULL;
1822 int ret = 0;
1823 int err;
1824
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001825 ClearPageChecked(page);
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001826 page_bufs = page_buffers(page);
1827 BUG_ON(!page_bufs);
1828 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
1829 /* As soon as we unlock the page, it can go away, but we have
1830 * references to buffers so we are safe */
1831 unlock_page(page);
1832
1833 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1834 if (IS_ERR(handle)) {
1835 ret = PTR_ERR(handle);
1836 goto out;
1837 }
1838
Curt Wohlgemuth441c8502011-08-13 11:25:18 -04001839 BUG_ON(!ext4_handle_valid(handle));
1840
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001841 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1842 do_journal_get_write_access);
1843
1844 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1845 write_end_fn);
1846 if (ret == 0)
1847 ret = err;
Jan Kara2d859db2011-07-26 09:07:11 -04001848 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001849 err = ext4_journal_stop(handle);
1850 if (!ret)
1851 ret = err;
1852
1853 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05001854 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001855out:
1856 return ret;
1857}
1858
Jiaying Zhang744692d2010-03-04 16:14:02 -05001859static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
1860static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
1861
Mingming Cao61628a32008-07-11 19:27:31 -04001862/*
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001863 * Note that we don't need to start a transaction unless we're journaling data
1864 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1865 * need to file the inode to the transaction's list in ordered mode because if
1866 * we are writing back data added by write(), the inode is already there and if
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001867 * we are writing back data modified via mmap(), no one guarantees in which
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001868 * transaction the data will hit the disk. In case we are journaling data, we
1869 * cannot start transaction directly because transaction start ranks above page
1870 * lock so we have to do some magic.
1871 *
Theodore Ts'ob920c752009-05-14 00:54:29 -04001872 * This function can get called via...
1873 * - ext4_da_writepages after taking page lock (have journal handle)
1874 * - journal_submit_inode_data_buffers (no journal handle)
1875 * - shrink_page_list via pdflush (no journal handle)
1876 * - grab_page_cache when doing write_begin (have journal handle)
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001877 *
1878 * We don't do any block allocation in this function. If we have page with
1879 * multiple blocks we need to write those buffer_heads that are mapped. This
1880 * is important for mmaped based write. So if we do with blocksize 1K
1881 * truncate(f, 1024);
1882 * a = mmap(f, 0, 4096);
1883 * a[0] = 'a';
1884 * truncate(f, 4096);
1885 * we have in the page first buffer_head mapped via page_mkwrite call back
Paul Bolle90802ed2011-12-05 13:00:34 +01001886 * but other buffer_heads would be unmapped but dirty (dirty done via the
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001887 * do_wp_page). So writepage should write the first block. If we modify
1888 * the mmap area beyond 1024 we will again get a page_fault and the
1889 * page_mkwrite callback will do the block allocation and mark the
1890 * buffer_heads mapped.
1891 *
1892 * We redirty the page if we have any buffer_heads that is either delay or
1893 * unwritten in the page.
1894 *
1895 * We can get recursively called as show below.
1896 *
1897 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1898 * ext4_writepage()
1899 *
1900 * But since we don't do any block allocation we should not deadlock.
1901 * Page also have the dirty flag cleared so we don't get recurive page_lock.
Mingming Cao61628a32008-07-11 19:27:31 -04001902 */
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001903static int ext4_writepage(struct page *page,
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001904 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04001905{
Theodore Ts'oa42afc52010-10-27 21:30:09 -04001906 int ret = 0, commit_write = 0;
Mingming Cao61628a32008-07-11 19:27:31 -04001907 loff_t size;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001908 unsigned int len;
Jiaying Zhang744692d2010-03-04 16:14:02 -05001909 struct buffer_head *page_bufs = NULL;
Mingming Cao61628a32008-07-11 19:27:31 -04001910 struct inode *inode = page->mapping->host;
Alex Tomas64769242008-07-11 19:27:31 -04001911
Lukas Czernera9c667f2011-06-06 09:51:52 -04001912 trace_ext4_writepage(page);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04001913 size = i_size_read(inode);
1914 if (page->index == size >> PAGE_CACHE_SHIFT)
1915 len = size & ~PAGE_CACHE_MASK;
1916 else
1917 len = PAGE_CACHE_SIZE;
Alex Tomas64769242008-07-11 19:27:31 -04001918
Theodore Ts'oa42afc52010-10-27 21:30:09 -04001919 /*
1920 * If the page does not have buffers (for whatever reason),
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001921 * try to create them using __block_write_begin. If this
Theodore Ts'oa42afc52010-10-27 21:30:09 -04001922 * fails, redirty the page and move on.
1923 */
Theodore Ts'ob1142e82010-10-28 17:33:57 -04001924 if (!page_has_buffers(page)) {
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001925 if (__block_write_begin(page, 0, len,
Theodore Ts'oa42afc52010-10-27 21:30:09 -04001926 noalloc_get_block_write)) {
1927 redirty_page:
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04001928 redirty_page_for_writepage(wbc, page);
1929 unlock_page(page);
1930 return 0;
1931 }
Theodore Ts'oa42afc52010-10-27 21:30:09 -04001932 commit_write = 1;
1933 }
1934 page_bufs = page_buffers(page);
1935 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1936 ext4_bh_delay_or_unwritten)) {
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04001937 /*
Theodore Ts'ob1142e82010-10-28 17:33:57 -04001938 * We don't want to do block allocation, so redirty
1939 * the page and return. We may reach here when we do
1940 * a journal commit via journal_submit_inode_data_buffers.
Mel Gorman966dbde2011-10-31 17:07:48 -07001941 * We can also reach here via shrink_page_list but it
1942 * should never be for direct reclaim so warn if that
1943 * happens
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04001944 */
Mel Gorman966dbde2011-10-31 17:07:48 -07001945 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
1946 PF_MEMALLOC);
Theodore Ts'oa42afc52010-10-27 21:30:09 -04001947 goto redirty_page;
1948 }
1949 if (commit_write)
Aneesh Kumar K.Ved9b3e32008-11-07 09:06:45 -05001950 /* now mark the buffer_heads as dirty and uptodate */
Aneesh Kumar K.Vb767e782009-06-04 08:06:06 -04001951 block_commit_write(page, 0, len);
Alex Tomas64769242008-07-11 19:27:31 -04001952
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001953 if (PageChecked(page) && ext4_should_journal_data(inode))
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001954 /*
1955 * It's mmapped pagecache. Add buffers and journal it. There
1956 * doesn't seem much point in redirtying the page here.
1957 */
Wu Fengguang3f0ca302009-11-24 11:15:44 -05001958 return __ext4_journalled_writepage(page, len);
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001959
Theodore Ts'oa42afc52010-10-27 21:30:09 -04001960 if (buffer_uninit(page_bufs)) {
Jiaying Zhang744692d2010-03-04 16:14:02 -05001961 ext4_set_bh_endio(page_bufs, inode);
1962 ret = block_write_full_page_endio(page, noalloc_get_block_write,
1963 wbc, ext4_end_io_buffer_write);
1964 } else
Theodore Ts'ob920c752009-05-14 00:54:29 -04001965 ret = block_write_full_page(page, noalloc_get_block_write,
1966 wbc);
Alex Tomas64769242008-07-11 19:27:31 -04001967
Alex Tomas64769242008-07-11 19:27:31 -04001968 return ret;
1969}
1970
Mingming Cao61628a32008-07-11 19:27:31 -04001971/*
Mingming Cao525f4ed2008-08-19 22:15:58 -04001972 * This is called via ext4_da_writepages() to
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001973 * calculate the total number of credits to reserve to fit
Mingming Cao525f4ed2008-08-19 22:15:58 -04001974 * a single extent allocation into a single transaction,
1975 * ext4_da_writpeages() will loop calling this before
1976 * the block allocation.
Mingming Cao61628a32008-07-11 19:27:31 -04001977 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04001978
1979static int ext4_da_writepages_trans_blocks(struct inode *inode)
1980{
1981 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
1982
1983 /*
1984 * With non-extent format the journal credit needed to
1985 * insert nrblocks contiguous block is dependent on
1986 * number of contiguous block. So we will limit
1987 * number of contiguous block to a sane value
1988 */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001989 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
Mingming Cao525f4ed2008-08-19 22:15:58 -04001990 (max_blocks > EXT4_MAX_TRANS_DATA))
1991 max_blocks = EXT4_MAX_TRANS_DATA;
1992
1993 return ext4_chunk_trans_blocks(inode, max_blocks);
1994}
Mingming Cao61628a32008-07-11 19:27:31 -04001995
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04001996/*
1997 * write_cache_pages_da - walk the list of dirty pages of the given
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05001998 * address space and accumulate pages that need writing, and call
Theodore Ts'o168fc022011-02-26 14:09:20 -05001999 * mpage_da_map_and_submit to map a single contiguous memory region
2000 * and then write them.
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002001 */
2002static int write_cache_pages_da(struct address_space *mapping,
2003 struct writeback_control *wbc,
Eric Sandeen72f84e62010-10-27 21:30:13 -04002004 struct mpage_da_data *mpd,
2005 pgoff_t *done_index)
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002006{
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002007 struct buffer_head *bh, *head;
Theodore Ts'o168fc022011-02-26 14:09:20 -05002008 struct inode *inode = mapping->host;
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002009 struct pagevec pvec;
2010 unsigned int nr_pages;
2011 sector_t logical;
2012 pgoff_t index, end;
2013 long nr_to_write = wbc->nr_to_write;
2014 int i, tag, ret = 0;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002015
Theodore Ts'o168fc022011-02-26 14:09:20 -05002016 memset(mpd, 0, sizeof(struct mpage_da_data));
2017 mpd->wbc = wbc;
2018 mpd->inode = inode;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002019 pagevec_init(&pvec, 0);
2020 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2021 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2022
Wu Fengguang6e6938b2010-06-06 10:38:15 -06002023 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
Eric Sandeen5b41d922010-10-27 21:30:13 -04002024 tag = PAGECACHE_TAG_TOWRITE;
2025 else
2026 tag = PAGECACHE_TAG_DIRTY;
2027
Eric Sandeen72f84e62010-10-27 21:30:13 -04002028 *done_index = index;
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002029 while (index <= end) {
Eric Sandeen5b41d922010-10-27 21:30:13 -04002030 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002031 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2032 if (nr_pages == 0)
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002033 return 0;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002034
2035 for (i = 0; i < nr_pages; i++) {
2036 struct page *page = pvec.pages[i];
2037
2038 /*
2039 * At this point, the page may be truncated or
2040 * invalidated (changing page->mapping to NULL), or
2041 * even swizzled back from swapper_space to tmpfs file
2042 * mapping. However, page->index will not change
2043 * because we have a reference on the page.
2044 */
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002045 if (page->index > end)
2046 goto out;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002047
Eric Sandeen72f84e62010-10-27 21:30:13 -04002048 *done_index = page->index + 1;
2049
Theodore Ts'o78aaced2011-02-26 14:09:14 -05002050 /*
2051 * If we can't merge this page, and we have
2052 * accumulated an contiguous region, write it
2053 */
2054 if ((mpd->next_page != page->index) &&
2055 (mpd->next_page != mpd->first_page)) {
2056 mpage_da_map_and_submit(mpd);
2057 goto ret_extent_tail;
2058 }
2059
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002060 lock_page(page);
2061
2062 /*
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002063 * If the page is no longer dirty, or its
2064 * mapping no longer corresponds to inode we
2065 * are writing (which means it has been
2066 * truncated or invalidated), or the page is
2067 * already under writeback and we are not
2068 * doing a data integrity writeback, skip the page
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002069 */
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002070 if (!PageDirty(page) ||
2071 (PageWriteback(page) &&
2072 (wbc->sync_mode == WB_SYNC_NONE)) ||
2073 unlikely(page->mapping != mapping)) {
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002074 unlock_page(page);
2075 continue;
2076 }
2077
Darrick J. Wong7cb1a532011-05-18 13:53:20 -04002078 wait_on_page_writeback(page);
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002079 BUG_ON(PageWriteback(page));
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002080
Theodore Ts'o168fc022011-02-26 14:09:20 -05002081 if (mpd->next_page != page->index)
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002082 mpd->first_page = page->index;
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002083 mpd->next_page = page->index + 1;
2084 logical = (sector_t) page->index <<
2085 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2086
2087 if (!page_has_buffers(page)) {
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002088 mpage_add_bh_to_extent(mpd, logical,
2089 PAGE_CACHE_SIZE,
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002090 (1 << BH_Dirty) | (1 << BH_Uptodate));
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002091 if (mpd->io_done)
2092 goto ret_extent_tail;
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002093 } else {
2094 /*
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002095 * Page with regular buffer heads,
2096 * just add all dirty ones
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002097 */
2098 head = page_buffers(page);
2099 bh = head;
2100 do {
2101 BUG_ON(buffer_locked(bh));
2102 /*
2103 * We need to try to allocate
2104 * unmapped blocks in the same page.
2105 * Otherwise we won't make progress
2106 * with the page in ext4_writepage
2107 */
2108 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2109 mpage_add_bh_to_extent(mpd, logical,
2110 bh->b_size,
2111 bh->b_state);
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002112 if (mpd->io_done)
2113 goto ret_extent_tail;
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002114 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2115 /*
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002116 * mapped dirty buffer. We need
2117 * to update the b_state
2118 * because we look at b_state
2119 * in mpage_da_map_blocks. We
2120 * don't update b_size because
2121 * if we find an unmapped
2122 * buffer_head later we need to
2123 * use the b_state flag of that
2124 * buffer_head.
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002125 */
2126 if (mpd->b_size == 0)
2127 mpd->b_state = bh->b_state & BH_FLAGS;
2128 }
2129 logical++;
2130 } while ((bh = bh->b_this_page) != head);
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002131 }
2132
2133 if (nr_to_write > 0) {
2134 nr_to_write--;
2135 if (nr_to_write == 0 &&
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002136 wbc->sync_mode == WB_SYNC_NONE)
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002137 /*
2138 * We stop writing back only if we are
2139 * not doing integrity sync. In case of
2140 * integrity sync we have to keep going
2141 * because someone may be concurrently
2142 * dirtying pages, and we might have
2143 * synced a lot of newly appeared dirty
2144 * pages, but have not synced all of the
2145 * old dirty pages.
2146 */
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002147 goto out;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002148 }
2149 }
2150 pagevec_release(&pvec);
2151 cond_resched();
2152 }
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002153 return 0;
2154ret_extent_tail:
2155 ret = MPAGE_DA_EXTENT_TAIL;
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002156out:
2157 pagevec_release(&pvec);
2158 cond_resched();
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002159 return ret;
2160}
2161
2162
Alex Tomas64769242008-07-11 19:27:31 -04002163static int ext4_da_writepages(struct address_space *mapping,
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002164 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04002165{
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002166 pgoff_t index;
2167 int range_whole = 0;
Mingming Cao61628a32008-07-11 19:27:31 -04002168 handle_t *handle = NULL;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002169 struct mpage_da_data mpd;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002170 struct inode *inode = mapping->host;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002171 int pages_written = 0;
Theodore Ts'o55138e02009-09-29 13:31:31 -04002172 unsigned int max_pages;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002173 int range_cyclic, cycled = 1, io_done = 0;
Theodore Ts'o55138e02009-09-29 13:31:31 -04002174 int needed_blocks, ret = 0;
2175 long desired_nr_to_write, nr_to_writebump = 0;
Theodore Ts'ode89de62009-08-31 17:00:59 -04002176 loff_t range_start = wbc->range_start;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002177 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
Eric Sandeen72f84e62010-10-27 21:30:13 -04002178 pgoff_t done_index = 0;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002179 pgoff_t end;
Shaohua Li1bce63d2011-10-18 10:55:51 -04002180 struct blk_plug plug;
Mingming Cao61628a32008-07-11 19:27:31 -04002181
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002182 trace_ext4_da_writepages(inode, wbc);
Theodore Ts'oba80b102009-01-03 20:03:21 -05002183
Mingming Cao61628a32008-07-11 19:27:31 -04002184 /*
2185 * No pages to write? This is mainly a kludge to avoid starting
2186 * a transaction for special inodes like journal inode on last iput()
2187 * because that could violate lock ordering on umount
2188 */
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002189 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
Mingming Cao61628a32008-07-11 19:27:31 -04002190 return 0;
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002191
2192 /*
2193 * If the filesystem has aborted, it is read-only, so return
2194 * right away instead of dumping stack traces later on that
2195 * will obscure the real source of the problem. We test
Theodore Ts'o4ab2f152009-06-13 10:09:36 -04002196 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002197 * the latter could be true if the filesystem is mounted
2198 * read-only, and in that case, ext4_da_writepages should
2199 * *never* be called, so if that ever happens, we would want
2200 * the stack trace.
2201 */
Theodore Ts'o4ab2f152009-06-13 10:09:36 -04002202 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002203 return -EROFS;
2204
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002205 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2206 range_whole = 1;
Mingming Cao61628a32008-07-11 19:27:31 -04002207
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002208 range_cyclic = wbc->range_cyclic;
2209 if (wbc->range_cyclic) {
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002210 index = mapping->writeback_index;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002211 if (index)
2212 cycled = 0;
2213 wbc->range_start = index << PAGE_CACHE_SHIFT;
2214 wbc->range_end = LLONG_MAX;
2215 wbc->range_cyclic = 0;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002216 end = -1;
2217 } else {
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002218 index = wbc->range_start >> PAGE_CACHE_SHIFT;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002219 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2220 }
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002221
Theodore Ts'o55138e02009-09-29 13:31:31 -04002222 /*
2223 * This works around two forms of stupidity. The first is in
2224 * the writeback code, which caps the maximum number of pages
2225 * written to be 1024 pages. This is wrong on multiple
2226 * levels; different architectues have a different page size,
2227 * which changes the maximum amount of data which gets
2228 * written. Secondly, 4 megabytes is way too small. XFS
2229 * forces this value to be 16 megabytes by multiplying
2230 * nr_to_write parameter by four, and then relies on its
2231 * allocator to allocate larger extents to make them
2232 * contiguous. Unfortunately this brings us to the second
2233 * stupidity, which is that ext4's mballoc code only allocates
2234 * at most 2048 blocks. So we force contiguous writes up to
2235 * the number of dirty blocks in the inode, or
2236 * sbi->max_writeback_mb_bump whichever is smaller.
2237 */
2238 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
Eric Sandeenb443e732010-10-27 21:30:03 -04002239 if (!range_cyclic && range_whole) {
2240 if (wbc->nr_to_write == LONG_MAX)
2241 desired_nr_to_write = wbc->nr_to_write;
2242 else
2243 desired_nr_to_write = wbc->nr_to_write * 8;
2244 } else
Theodore Ts'o55138e02009-09-29 13:31:31 -04002245 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2246 max_pages);
2247 if (desired_nr_to_write > max_pages)
2248 desired_nr_to_write = max_pages;
2249
2250 if (wbc->nr_to_write < desired_nr_to_write) {
2251 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2252 wbc->nr_to_write = desired_nr_to_write;
2253 }
2254
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002255retry:
Wu Fengguang6e6938b2010-06-06 10:38:15 -06002256 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
Eric Sandeen5b41d922010-10-27 21:30:13 -04002257 tag_pages_for_writeback(mapping, index, end);
2258
Shaohua Li1bce63d2011-10-18 10:55:51 -04002259 blk_start_plug(&plug);
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002260 while (!ret && wbc->nr_to_write > 0) {
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002261
2262 /*
2263 * we insert one extent at a time. So we need
2264 * credit needed for single extent allocation.
2265 * journalled mode is currently not supported
2266 * by delalloc
2267 */
2268 BUG_ON(ext4_should_journal_data(inode));
Mingming Cao525f4ed2008-08-19 22:15:58 -04002269 needed_blocks = ext4_da_writepages_trans_blocks(inode);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002270
Mingming Cao61628a32008-07-11 19:27:31 -04002271 /* start a new transaction*/
2272 handle = ext4_journal_start(inode, needed_blocks);
2273 if (IS_ERR(handle)) {
2274 ret = PTR_ERR(handle);
Theodore Ts'o16939182009-09-26 17:43:59 -04002275 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
Curt Wohlgemuthfbe845d2010-05-16 13:00:00 -04002276 "%ld pages, ino %lu; err %d", __func__,
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002277 wbc->nr_to_write, inode->i_ino, ret);
Namjae Jeon3c1fcb22011-11-07 11:01:13 -05002278 blk_finish_plug(&plug);
Mingming Cao61628a32008-07-11 19:27:31 -04002279 goto out_writepages;
2280 }
Theodore Ts'of63e6002009-02-23 16:42:39 -05002281
2282 /*
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002283 * Now call write_cache_pages_da() to find the next
Theodore Ts'of63e6002009-02-23 16:42:39 -05002284 * contiguous region of logical blocks that need
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002285 * blocks to be allocated by ext4 and submit them.
Theodore Ts'of63e6002009-02-23 16:42:39 -05002286 */
Eric Sandeen72f84e62010-10-27 21:30:13 -04002287 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
Theodore Ts'of63e6002009-02-23 16:42:39 -05002288 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002289 * If we have a contiguous extent of pages and we
Theodore Ts'of63e6002009-02-23 16:42:39 -05002290 * haven't done the I/O yet, map the blocks and submit
2291 * them for I/O.
2292 */
2293 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002294 mpage_da_map_and_submit(&mpd);
Theodore Ts'of63e6002009-02-23 16:42:39 -05002295 ret = MPAGE_DA_EXTENT_TAIL;
2296 }
Theodore Ts'ob3a3ca82009-08-31 23:13:11 -04002297 trace_ext4_da_write_pages(inode, &mpd);
Theodore Ts'of63e6002009-02-23 16:42:39 -05002298 wbc->nr_to_write -= mpd.pages_written;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002299
Mingming Cao61628a32008-07-11 19:27:31 -04002300 ext4_journal_stop(handle);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002301
Eric Sandeen8f64b322009-02-26 00:57:35 -05002302 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002303 /* commit the transaction which would
2304 * free blocks released in the transaction
2305 * and try again
2306 */
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002307 jbd2_journal_force_commit_nested(sbi->s_journal);
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002308 ret = 0;
2309 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002310 /*
Kazuya Mio8de49e62011-10-20 19:23:08 -04002311 * Got one extent now try with rest of the pages.
2312 * If mpd.retval is set -EIO, journal is aborted.
2313 * So we don't need to write any more.
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002314 */
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002315 pages_written += mpd.pages_written;
Kazuya Mio8de49e62011-10-20 19:23:08 -04002316 ret = mpd.retval;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002317 io_done = 1;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002318 } else if (wbc->nr_to_write)
Mingming Cao61628a32008-07-11 19:27:31 -04002319 /*
2320 * There is no more writeout needed
2321 * or we requested for a noblocking writeout
2322 * and we found the device congested
2323 */
Mingming Cao61628a32008-07-11 19:27:31 -04002324 break;
Mingming Cao61628a32008-07-11 19:27:31 -04002325 }
Shaohua Li1bce63d2011-10-18 10:55:51 -04002326 blk_finish_plug(&plug);
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002327 if (!io_done && !cycled) {
2328 cycled = 1;
2329 index = 0;
2330 wbc->range_start = index << PAGE_CACHE_SHIFT;
2331 wbc->range_end = mapping->writeback_index - 1;
2332 goto retry;
2333 }
Mingming Cao61628a32008-07-11 19:27:31 -04002334
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002335 /* Update index */
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002336 wbc->range_cyclic = range_cyclic;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002337 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2338 /*
2339 * set the writeback_index so that range_cyclic
2340 * mode will write it back later
2341 */
Eric Sandeen72f84e62010-10-27 21:30:13 -04002342 mapping->writeback_index = done_index;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002343
Mingming Cao61628a32008-07-11 19:27:31 -04002344out_writepages:
Richard Kennedy2faf2e12009-12-25 15:46:07 -05002345 wbc->nr_to_write -= nr_to_writebump;
Theodore Ts'ode89de62009-08-31 17:00:59 -04002346 wbc->range_start = range_start;
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002347 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
Mingming Cao61628a32008-07-11 19:27:31 -04002348 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04002349}
2350
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002351#define FALL_BACK_TO_NONDELALLOC 1
2352static int ext4_nonda_switch(struct super_block *sb)
2353{
2354 s64 free_blocks, dirty_blocks;
2355 struct ext4_sb_info *sbi = EXT4_SB(sb);
2356
2357 /*
2358 * switch to non delalloc mode if we are running low
2359 * on free block. The free block accounting via percpu
Eric Dumazet179f7eb2009-01-06 14:41:04 -08002360 * counters can get slightly wrong with percpu_counter_batch getting
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002361 * accumulated on each CPU without updating global counters
2362 * Delalloc need an accurate free block accounting. So switch
2363 * to non delalloc when we are near to error range.
2364 */
Theodore Ts'o57042652011-09-09 18:56:51 -04002365 free_blocks = EXT4_C2B(sbi,
2366 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2367 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002368 if (2 * free_blocks < 3 * dirty_blocks ||
Theodore Ts'odf55c992011-09-09 19:16:51 -04002369 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002370 /*
Eric Sandeenc8afb442009-12-23 07:58:12 -05002371 * free block count is less than 150% of dirty blocks
2372 * or free blocks is less than watermark
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002373 */
2374 return 1;
2375 }
Eric Sandeenc8afb442009-12-23 07:58:12 -05002376 /*
2377 * Even if we don't switch but are nearing capacity,
2378 * start pushing delalloc when 1/2 of free blocks are dirty.
2379 */
2380 if (free_blocks < 2 * dirty_blocks)
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06002381 writeback_inodes_sb_if_idle(sb, WB_REASON_FS_FREE_SPACE);
Eric Sandeenc8afb442009-12-23 07:58:12 -05002382
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002383 return 0;
2384}
2385
Alex Tomas64769242008-07-11 19:27:31 -04002386static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002387 loff_t pos, unsigned len, unsigned flags,
2388 struct page **pagep, void **fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04002389{
Eric Sandeen72b8ab92010-05-16 11:00:00 -04002390 int ret, retries = 0;
Alex Tomas64769242008-07-11 19:27:31 -04002391 struct page *page;
2392 pgoff_t index;
Alex Tomas64769242008-07-11 19:27:31 -04002393 struct inode *inode = mapping->host;
2394 handle_t *handle;
2395
2396 index = pos >> PAGE_CACHE_SHIFT;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002397
2398 if (ext4_nonda_switch(inode->i_sb)) {
2399 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2400 return ext4_write_begin(file, mapping, pos,
2401 len, flags, pagep, fsdata);
2402 }
2403 *fsdata = (void *)0;
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002404 trace_ext4_da_write_begin(inode, pos, len, flags);
Mingming Caod2a17632008-07-14 17:52:37 -04002405retry:
Alex Tomas64769242008-07-11 19:27:31 -04002406 /*
2407 * With delayed allocation, we don't log the i_disksize update
2408 * if there is delayed block allocation. But we still need
2409 * to journalling the i_disksize update if writes to the end
2410 * of file which has an already mapped buffer.
2411 */
2412 handle = ext4_journal_start(inode, 1);
2413 if (IS_ERR(handle)) {
2414 ret = PTR_ERR(handle);
2415 goto out;
2416 }
Jan Karaebd36102009-02-22 21:09:59 -05002417 /* We cannot recurse into the filesystem as the transaction is already
2418 * started */
2419 flags |= AOP_FLAG_NOFS;
Alex Tomas64769242008-07-11 19:27:31 -04002420
Nick Piggin54566b22009-01-04 12:00:53 -08002421 page = grab_cache_page_write_begin(mapping, index, flags);
Eric Sandeend5a0d4f2008-08-02 18:51:06 -04002422 if (!page) {
2423 ext4_journal_stop(handle);
2424 ret = -ENOMEM;
2425 goto out;
2426 }
Alex Tomas64769242008-07-11 19:27:31 -04002427 *pagep = page;
2428
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002429 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
Alex Tomas64769242008-07-11 19:27:31 -04002430 if (ret < 0) {
2431 unlock_page(page);
2432 ext4_journal_stop(handle);
2433 page_cache_release(page);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04002434 /*
2435 * block_write_begin may have instantiated a few blocks
2436 * outside i_size. Trim these off again. Don't need
2437 * i_size_read because we hold i_mutex.
2438 */
2439 if (pos + len > inode->i_size)
Jan Karab9a42072009-12-08 21:24:33 -05002440 ext4_truncate_failed_write(inode);
Alex Tomas64769242008-07-11 19:27:31 -04002441 }
2442
Mingming Caod2a17632008-07-14 17:52:37 -04002443 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2444 goto retry;
Alex Tomas64769242008-07-11 19:27:31 -04002445out:
2446 return ret;
2447}
2448
Mingming Cao632eaea2008-07-11 19:27:31 -04002449/*
2450 * Check if we should update i_disksize
2451 * when write to the end of file but not require block allocation
2452 */
2453static int ext4_da_should_update_i_disksize(struct page *page,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002454 unsigned long offset)
Mingming Cao632eaea2008-07-11 19:27:31 -04002455{
2456 struct buffer_head *bh;
2457 struct inode *inode = page->mapping->host;
2458 unsigned int idx;
2459 int i;
2460
2461 bh = page_buffers(page);
2462 idx = offset >> inode->i_blkbits;
2463
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002464 for (i = 0; i < idx; i++)
Mingming Cao632eaea2008-07-11 19:27:31 -04002465 bh = bh->b_this_page;
2466
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002467 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
Mingming Cao632eaea2008-07-11 19:27:31 -04002468 return 0;
2469 return 1;
2470}
2471
Alex Tomas64769242008-07-11 19:27:31 -04002472static int ext4_da_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002473 struct address_space *mapping,
2474 loff_t pos, unsigned len, unsigned copied,
2475 struct page *page, void *fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04002476{
2477 struct inode *inode = mapping->host;
2478 int ret = 0, ret2;
2479 handle_t *handle = ext4_journal_current_handle();
2480 loff_t new_i_size;
Mingming Cao632eaea2008-07-11 19:27:31 -04002481 unsigned long start, end;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002482 int write_mode = (int)(unsigned long)fsdata;
2483
2484 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2485 if (ext4_should_order_data(inode)) {
2486 return ext4_ordered_write_end(file, mapping, pos,
2487 len, copied, page, fsdata);
2488 } else if (ext4_should_writeback_data(inode)) {
2489 return ext4_writeback_write_end(file, mapping, pos,
2490 len, copied, page, fsdata);
2491 } else {
2492 BUG();
2493 }
2494 }
Mingming Cao632eaea2008-07-11 19:27:31 -04002495
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002496 trace_ext4_da_write_end(inode, pos, len, copied);
Mingming Cao632eaea2008-07-11 19:27:31 -04002497 start = pos & (PAGE_CACHE_SIZE - 1);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002498 end = start + copied - 1;
Alex Tomas64769242008-07-11 19:27:31 -04002499
2500 /*
2501 * generic_write_end() will run mark_inode_dirty() if i_size
2502 * changes. So let's piggyback the i_disksize mark_inode_dirty
2503 * into that.
2504 */
2505
2506 new_i_size = pos + copied;
Andrea Arcangeliea51d132011-12-13 21:41:15 -05002507 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
Mingming Cao632eaea2008-07-11 19:27:31 -04002508 if (ext4_da_should_update_i_disksize(page, end)) {
2509 down_write(&EXT4_I(inode)->i_data_sem);
2510 if (new_i_size > EXT4_I(inode)->i_disksize) {
2511 /*
2512 * Updating i_disksize when extending file
2513 * without needing block allocation
2514 */
2515 if (ext4_should_order_data(inode))
2516 ret = ext4_jbd2_file_inode(handle,
2517 inode);
Alex Tomas64769242008-07-11 19:27:31 -04002518
Mingming Cao632eaea2008-07-11 19:27:31 -04002519 EXT4_I(inode)->i_disksize = new_i_size;
2520 }
2521 up_write(&EXT4_I(inode)->i_data_sem);
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04002522 /* We need to mark inode dirty even if
2523 * new_i_size is less that inode->i_size
2524 * bu greater than i_disksize.(hint delalloc)
2525 */
2526 ext4_mark_inode_dirty(handle, inode);
Alex Tomas64769242008-07-11 19:27:31 -04002527 }
Mingming Cao632eaea2008-07-11 19:27:31 -04002528 }
Alex Tomas64769242008-07-11 19:27:31 -04002529 ret2 = generic_write_end(file, mapping, pos, len, copied,
2530 page, fsdata);
2531 copied = ret2;
2532 if (ret2 < 0)
2533 ret = ret2;
2534 ret2 = ext4_journal_stop(handle);
2535 if (!ret)
2536 ret = ret2;
2537
2538 return ret ? ret : copied;
2539}
2540
2541static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2542{
Alex Tomas64769242008-07-11 19:27:31 -04002543 /*
2544 * Drop reserved blocks
2545 */
2546 BUG_ON(!PageLocked(page));
2547 if (!page_has_buffers(page))
2548 goto out;
2549
Mingming Caod2a17632008-07-14 17:52:37 -04002550 ext4_da_page_release_reservation(page, offset);
Alex Tomas64769242008-07-11 19:27:31 -04002551
2552out:
2553 ext4_invalidatepage(page, offset);
2554
2555 return;
2556}
2557
Theodore Ts'occd25062009-02-26 01:04:07 -05002558/*
2559 * Force all delayed allocation blocks to be allocated for a given inode.
2560 */
2561int ext4_alloc_da_blocks(struct inode *inode)
2562{
Theodore Ts'ofb40ba02009-09-16 19:30:40 -04002563 trace_ext4_alloc_da_blocks(inode);
2564
Theodore Ts'occd25062009-02-26 01:04:07 -05002565 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2566 !EXT4_I(inode)->i_reserved_meta_blocks)
2567 return 0;
2568
2569 /*
2570 * We do something simple for now. The filemap_flush() will
2571 * also start triggering a write of the data blocks, which is
2572 * not strictly speaking necessary (and for users of
2573 * laptop_mode, not even desirable). However, to do otherwise
2574 * would require replicating code paths in:
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002575 *
Theodore Ts'occd25062009-02-26 01:04:07 -05002576 * ext4_da_writepages() ->
2577 * write_cache_pages() ---> (via passed in callback function)
2578 * __mpage_da_writepage() -->
2579 * mpage_add_bh_to_extent()
2580 * mpage_da_map_blocks()
2581 *
2582 * The problem is that write_cache_pages(), located in
2583 * mm/page-writeback.c, marks pages clean in preparation for
2584 * doing I/O, which is not desirable if we're not planning on
2585 * doing I/O at all.
2586 *
2587 * We could call write_cache_pages(), and then redirty all of
Wu Fengguang380cf092010-11-11 19:23:29 +08002588 * the pages by calling redirty_page_for_writepage() but that
Theodore Ts'occd25062009-02-26 01:04:07 -05002589 * would be ugly in the extreme. So instead we would need to
2590 * replicate parts of the code in the above functions,
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002591 * simplifying them because we wouldn't actually intend to
Theodore Ts'occd25062009-02-26 01:04:07 -05002592 * write out the pages, but rather only collect contiguous
2593 * logical block extents, call the multi-block allocator, and
2594 * then update the buffer heads with the block allocations.
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002595 *
Theodore Ts'occd25062009-02-26 01:04:07 -05002596 * For now, though, we'll cheat by calling filemap_flush(),
2597 * which will map the blocks, and start the I/O, but not
2598 * actually wait for the I/O to complete.
2599 */
2600 return filemap_flush(inode->i_mapping);
2601}
Alex Tomas64769242008-07-11 19:27:31 -04002602
2603/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002604 * bmap() is special. It gets used by applications such as lilo and by
2605 * the swapper to find the on-disk block of a specific piece of data.
2606 *
2607 * Naturally, this is dangerous if the block concerned is still in the
Mingming Cao617ba132006-10-11 01:20:53 -07002608 * journal. If somebody makes a swapfile on an ext4 data-journaling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002609 * filesystem and enables swap, then they may get a nasty shock when the
2610 * data getting swapped to that swapfile suddenly gets overwritten by
2611 * the original zero's written out previously to the journal and
2612 * awaiting writeback in the kernel's buffer cache.
2613 *
2614 * So, if we see any bmap calls here on a modified, data-journaled file,
2615 * take extra steps to flush any blocks which might be in the cache.
2616 */
Mingming Cao617ba132006-10-11 01:20:53 -07002617static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002618{
2619 struct inode *inode = mapping->host;
2620 journal_t *journal;
2621 int err;
2622
Alex Tomas64769242008-07-11 19:27:31 -04002623 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2624 test_opt(inode->i_sb, DELALLOC)) {
2625 /*
2626 * With delalloc we want to sync the file
2627 * so that we can make sure we allocate
2628 * blocks for file
2629 */
2630 filemap_write_and_wait(mapping);
2631 }
2632
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05002633 if (EXT4_JOURNAL(inode) &&
2634 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002635 /*
2636 * This is a REALLY heavyweight approach, but the use of
2637 * bmap on dirty files is expected to be extremely rare:
2638 * only if we run lilo or swapon on a freshly made file
2639 * do we expect this to happen.
2640 *
2641 * (bmap requires CAP_SYS_RAWIO so this does not
2642 * represent an unprivileged user DOS attack --- we'd be
2643 * in trouble if mortal users could trigger this path at
2644 * will.)
2645 *
Mingming Cao617ba132006-10-11 01:20:53 -07002646 * NB. EXT4_STATE_JDATA is not set on files other than
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002647 * regular files. If somebody wants to bmap a directory
2648 * or symlink and gets confused because the buffer
2649 * hasn't yet been flushed to disk, they deserve
2650 * everything they get.
2651 */
2652
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05002653 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
Mingming Cao617ba132006-10-11 01:20:53 -07002654 journal = EXT4_JOURNAL(inode);
Mingming Caodab291a2006-10-11 01:21:01 -07002655 jbd2_journal_lock_updates(journal);
2656 err = jbd2_journal_flush(journal);
2657 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002658
2659 if (err)
2660 return 0;
2661 }
2662
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002663 return generic_block_bmap(mapping, block, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002664}
2665
Mingming Cao617ba132006-10-11 01:20:53 -07002666static int ext4_readpage(struct file *file, struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002667{
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002668 trace_ext4_readpage(page);
Mingming Cao617ba132006-10-11 01:20:53 -07002669 return mpage_readpage(page, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002670}
2671
2672static int
Mingming Cao617ba132006-10-11 01:20:53 -07002673ext4_readpages(struct file *file, struct address_space *mapping,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002674 struct list_head *pages, unsigned nr_pages)
2675{
Mingming Cao617ba132006-10-11 01:20:53 -07002676 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002677}
2678
Jiaying Zhang744692d2010-03-04 16:14:02 -05002679static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
2680{
2681 struct buffer_head *head, *bh;
2682 unsigned int curr_off = 0;
2683
2684 if (!page_has_buffers(page))
2685 return;
2686 head = bh = page_buffers(page);
2687 do {
2688 if (offset <= curr_off && test_clear_buffer_uninit(bh)
2689 && bh->b_private) {
2690 ext4_free_io_end(bh->b_private);
2691 bh->b_private = NULL;
2692 bh->b_end_io = NULL;
2693 }
2694 curr_off = curr_off + bh->b_size;
2695 bh = bh->b_this_page;
2696 } while (bh != head);
2697}
2698
Mingming Cao617ba132006-10-11 01:20:53 -07002699static void ext4_invalidatepage(struct page *page, unsigned long offset)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002700{
Mingming Cao617ba132006-10-11 01:20:53 -07002701 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002702
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002703 trace_ext4_invalidatepage(page, offset);
2704
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002705 /*
Jiaying Zhang744692d2010-03-04 16:14:02 -05002706 * free any io_end structure allocated for buffers to be discarded
2707 */
2708 if (ext4_should_dioread_nolock(page->mapping->host))
2709 ext4_invalidatepage_free_endio(page, offset);
2710 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002711 * If it's a full truncate we just forget about the pending dirtying
2712 */
2713 if (offset == 0)
2714 ClearPageChecked(page);
2715
Frank Mayhar03901312009-01-07 00:06:22 -05002716 if (journal)
2717 jbd2_journal_invalidatepage(journal, page, offset);
2718 else
2719 block_invalidatepage(page, offset);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002720}
2721
Mingming Cao617ba132006-10-11 01:20:53 -07002722static int ext4_releasepage(struct page *page, gfp_t wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002723{
Mingming Cao617ba132006-10-11 01:20:53 -07002724 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002725
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002726 trace_ext4_releasepage(page);
2727
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002728 WARN_ON(PageChecked(page));
2729 if (!page_has_buffers(page))
2730 return 0;
Frank Mayhar03901312009-01-07 00:06:22 -05002731 if (journal)
2732 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2733 else
2734 return try_to_free_buffers(page);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002735}
2736
2737/*
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002738 * ext4_get_block used when preparing for a DIO write or buffer write.
2739 * We allocate an uinitialized extent if blocks haven't been allocated.
2740 * The extent will be converted to initialized after the IO is complete.
2741 */
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05002742static int ext4_get_block_write(struct inode *inode, sector_t iblock,
Mingming Cao4c0425f2009-09-28 15:48:41 -04002743 struct buffer_head *bh_result, int create)
2744{
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05002745 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002746 inode->i_ino, create);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002747 return _ext4_get_block(inode, iblock, bh_result,
2748 EXT4_GET_BLOCKS_IO_CREATE_EXT);
Mingming Cao4c0425f2009-09-28 15:48:41 -04002749}
2750
Mingming Cao4c0425f2009-09-28 15:48:41 -04002751static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
Christoph Hellwig552ef8022010-07-27 11:56:06 -04002752 ssize_t size, void *private, int ret,
2753 bool is_async)
Mingming Cao4c0425f2009-09-28 15:48:41 -04002754{
Christoph Hellwig72c50522011-06-24 14:29:48 -04002755 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
Mingming Cao4c0425f2009-09-28 15:48:41 -04002756 ext4_io_end_t *io_end = iocb->private;
2757 struct workqueue_struct *wq;
Jiaying Zhang744692d2010-03-04 16:14:02 -05002758 unsigned long flags;
2759 struct ext4_inode_info *ei;
Mingming Cao4c0425f2009-09-28 15:48:41 -04002760
Mingming4b70df12009-11-03 14:44:54 -05002761 /* if not async direct IO or dio with 0 bytes write, just return */
2762 if (!io_end || !size)
Christoph Hellwig552ef8022010-07-27 11:56:06 -04002763 goto out;
Mingming4b70df12009-11-03 14:44:54 -05002764
Zheng Liu88635ca2011-12-28 19:00:25 -05002765 ext_debug("ext4_end_io_dio(): io_end 0x%p "
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002766 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
2767 iocb->private, io_end->inode->i_ino, iocb, offset,
2768 size);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002769
Theodore Ts'ob5a7e972011-12-12 10:53:02 -05002770 iocb->private = NULL;
2771
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002772 /* if not aio dio with unwritten extents, just free io and return */
Theodore Ts'obd2d0212010-10-27 21:30:10 -04002773 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002774 ext4_free_io_end(io_end);
jiayingz@google.com (Jiaying Zhang)5b3ff232010-07-27 11:56:06 -04002775out:
2776 if (is_async)
2777 aio_complete(iocb, ret, 0);
Christoph Hellwig72c50522011-06-24 14:29:48 -04002778 inode_dio_done(inode);
jiayingz@google.com (Jiaying Zhang)5b3ff232010-07-27 11:56:06 -04002779 return;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002780 }
2781
Mingming Cao4c0425f2009-09-28 15:48:41 -04002782 io_end->offset = offset;
2783 io_end->size = size;
jiayingz@google.com (Jiaying Zhang)5b3ff232010-07-27 11:56:06 -04002784 if (is_async) {
2785 io_end->iocb = iocb;
2786 io_end->result = ret;
2787 }
Mingming Cao4c0425f2009-09-28 15:48:41 -04002788 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
2789
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002790 /* Add the io_end to per-inode completed aio dio list*/
Jiaying Zhang744692d2010-03-04 16:14:02 -05002791 ei = EXT4_I(io_end->inode);
2792 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
2793 list_add_tail(&io_end->list, &ei->i_completed_io_list);
2794 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
Eric Sandeenc999af22010-10-27 21:30:07 -04002795
2796 /* queue the work to convert unwritten extents to written */
Tejun Heo4c81f042011-11-24 19:22:24 -05002797 queue_work(wq, &io_end->work);
Christoph Hellwig72c50522011-06-24 14:29:48 -04002798
2799 /* XXX: probably should move into the real I/O completion handler */
2800 inode_dio_done(inode);
Mingming Cao4c0425f2009-09-28 15:48:41 -04002801}
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05002802
Jiaying Zhang744692d2010-03-04 16:14:02 -05002803static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
2804{
2805 ext4_io_end_t *io_end = bh->b_private;
2806 struct workqueue_struct *wq;
2807 struct inode *inode;
2808 unsigned long flags;
2809
2810 if (!test_clear_buffer_uninit(bh) || !io_end)
2811 goto out;
2812
2813 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
2814 printk("sb umounted, discard end_io request for inode %lu\n",
2815 io_end->inode->i_ino);
2816 ext4_free_io_end(io_end);
2817 goto out;
2818 }
2819
Tao Ma32c80b32011-08-13 12:30:59 -04002820 /*
2821 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2822 * but being more careful is always safe for the future change.
2823 */
Jiaying Zhang744692d2010-03-04 16:14:02 -05002824 inode = io_end->inode;
Tao Ma0edeb712011-10-31 17:30:44 -04002825 ext4_set_io_unwritten_flag(inode, io_end);
Jiaying Zhang744692d2010-03-04 16:14:02 -05002826
2827 /* Add the io_end to per-inode completed io list*/
2828 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
2829 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
2830 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
2831
2832 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
2833 /* queue the work to convert unwritten extents to written */
2834 queue_work(wq, &io_end->work);
2835out:
2836 bh->b_private = NULL;
2837 bh->b_end_io = NULL;
2838 clear_buffer_uninit(bh);
2839 end_buffer_async_write(bh, uptodate);
2840}
2841
2842static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
2843{
2844 ext4_io_end_t *io_end;
2845 struct page *page = bh->b_page;
2846 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
2847 size_t size = bh->b_size;
2848
2849retry:
2850 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
2851 if (!io_end) {
Andrew Morton6db26ff2011-01-12 16:59:13 -08002852 pr_warn_ratelimited("%s: allocation fail\n", __func__);
Jiaying Zhang744692d2010-03-04 16:14:02 -05002853 schedule();
2854 goto retry;
2855 }
2856 io_end->offset = offset;
2857 io_end->size = size;
2858 /*
2859 * We need to hold a reference to the page to make sure it
2860 * doesn't get evicted before ext4_end_io_work() has a chance
2861 * to convert the extent from written to unwritten.
2862 */
2863 io_end->page = page;
2864 get_page(io_end->page);
2865
2866 bh->b_private = io_end;
2867 bh->b_end_io = ext4_end_io_buffer_write;
2868 return 0;
2869}
2870
Mingming Cao4c0425f2009-09-28 15:48:41 -04002871/*
2872 * For ext4 extent files, ext4 will do direct-io write to holes,
2873 * preallocated extents, and those write extend the file, no need to
2874 * fall back to buffered IO.
2875 *
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04002876 * For holes, we fallocate those blocks, mark them as uninitialized
Mingming Cao4c0425f2009-09-28 15:48:41 -04002877 * If those blocks were preallocated, we mark sure they are splited, but
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04002878 * still keep the range to write as uninitialized.
Mingming Cao4c0425f2009-09-28 15:48:41 -04002879 *
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002880 * The unwrritten extents will be converted to written when DIO is completed.
2881 * For async direct IO, since the IO may still pending when return, we
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002882 * set up an end_io call back function, which will do the conversion
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002883 * when async direct IO completed.
Mingming Cao4c0425f2009-09-28 15:48:41 -04002884 *
2885 * If the O_DIRECT write will extend the file then add this inode to the
2886 * orphan list. So recovery will truncate it back to the original size
2887 * if the machine crashes during the write.
2888 *
2889 */
2890static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2891 const struct iovec *iov, loff_t offset,
2892 unsigned long nr_segs)
2893{
2894 struct file *file = iocb->ki_filp;
2895 struct inode *inode = file->f_mapping->host;
2896 ssize_t ret;
2897 size_t count = iov_length(iov, nr_segs);
2898
2899 loff_t final_size = offset + count;
2900 if (rw == WRITE && final_size <= inode->i_size) {
2901 /*
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002902 * We could direct write to holes and fallocate.
2903 *
2904 * Allocated blocks to fill the hole are marked as uninitialized
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002905 * to prevent parallel buffered read to expose the stale data
Mingming Cao4c0425f2009-09-28 15:48:41 -04002906 * before DIO complete the data IO.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002907 *
2908 * As to previously fallocated extents, ext4 get_block
Mingming Cao4c0425f2009-09-28 15:48:41 -04002909 * will just simply mark the buffer mapped but still
2910 * keep the extents uninitialized.
2911 *
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002912 * for non AIO case, we will convert those unwritten extents
2913 * to written after return back from blockdev_direct_IO.
2914 *
2915 * for async DIO, the conversion needs to be defered when
2916 * the IO is completed. The ext4 end_io callback function
2917 * will be called to take care of the conversion work.
2918 * Here for async case, we allocate an io_end structure to
2919 * hook to the iocb.
Mingming Cao4c0425f2009-09-28 15:48:41 -04002920 */
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002921 iocb->private = NULL;
2922 EXT4_I(inode)->cur_aio_dio = NULL;
2923 if (!is_sync_kiocb(iocb)) {
Jiaying Zhang744692d2010-03-04 16:14:02 -05002924 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002925 if (!iocb->private)
2926 return -ENOMEM;
2927 /*
2928 * we save the io structure for current async
Eric Sandeen79e83032010-07-27 11:56:07 -04002929 * direct IO, so that later ext4_map_blocks()
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002930 * could flag the io structure whether there
2931 * is a unwritten extents needs to be converted
2932 * when IO is completed.
2933 */
2934 EXT4_I(inode)->cur_aio_dio = iocb->private;
2935 }
2936
Christoph Hellwigaacfc19c2011-06-24 14:29:47 -04002937 ret = __blockdev_direct_IO(rw, iocb, inode,
Mingming Cao4c0425f2009-09-28 15:48:41 -04002938 inode->i_sb->s_bdev, iov,
2939 offset, nr_segs,
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05002940 ext4_get_block_write,
Christoph Hellwigaacfc19c2011-06-24 14:29:47 -04002941 ext4_end_io_dio,
2942 NULL,
2943 DIO_LOCKING | DIO_SKIP_HOLES);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002944 if (iocb->private)
2945 EXT4_I(inode)->cur_aio_dio = NULL;
2946 /*
2947 * The io_end structure takes a reference to the inode,
2948 * that structure needs to be destroyed and the
2949 * reference to the inode need to be dropped, when IO is
2950 * complete, even with 0 byte write, or failed.
2951 *
2952 * In the successful AIO DIO case, the io_end structure will be
2953 * desctroyed and the reference to the inode will be dropped
2954 * after the end_io call back function is called.
2955 *
2956 * In the case there is 0 byte write, or error case, since
2957 * VFS direct IO won't invoke the end_io call back function,
2958 * we need to free the end_io structure here.
2959 */
2960 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
2961 ext4_free_io_end(iocb->private);
2962 iocb->private = NULL;
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05002963 } else if (ret > 0 && ext4_test_inode_state(inode,
2964 EXT4_STATE_DIO_UNWRITTEN)) {
Mingming109f5562009-11-10 10:48:08 -05002965 int err;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002966 /*
2967 * for non AIO case, since the IO is already
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002968 * completed, we could do the conversion right here
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002969 */
Mingming109f5562009-11-10 10:48:08 -05002970 err = ext4_convert_unwritten_extents(inode,
2971 offset, ret);
2972 if (err < 0)
2973 ret = err;
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05002974 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
Mingming109f5562009-11-10 10:48:08 -05002975 }
Mingming Cao4c0425f2009-09-28 15:48:41 -04002976 return ret;
2977 }
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002978
2979 /* for write the the end of file case, we fall back to old way */
Mingming Cao4c0425f2009-09-28 15:48:41 -04002980 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
2981}
2982
2983static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
2984 const struct iovec *iov, loff_t offset,
2985 unsigned long nr_segs)
2986{
2987 struct file *file = iocb->ki_filp;
2988 struct inode *inode = file->f_mapping->host;
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002989 ssize_t ret;
Mingming Cao4c0425f2009-09-28 15:48:41 -04002990
Theodore Ts'o84ebd792011-08-31 11:56:51 -04002991 /*
2992 * If we are doing data journalling we don't support O_DIRECT
2993 */
2994 if (ext4_should_journal_data(inode))
2995 return 0;
2996
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002997 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04002998 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002999 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3000 else
3001 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3002 trace_ext4_direct_IO_exit(inode, offset,
3003 iov_length(iov, nr_segs), rw, ret);
3004 return ret;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003005}
3006
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003007/*
Mingming Cao617ba132006-10-11 01:20:53 -07003008 * Pages can be marked dirty completely asynchronously from ext4's journalling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003009 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3010 * much here because ->set_page_dirty is called under VFS locks. The page is
3011 * not necessarily locked.
3012 *
3013 * We cannot just dirty the page and leave attached buffers clean, because the
3014 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3015 * or jbddirty because all the journalling code will explode.
3016 *
3017 * So what we do is to mark the page "pending dirty" and next time writepage
3018 * is called, propagate that into the buffers appropriately.
3019 */
Mingming Cao617ba132006-10-11 01:20:53 -07003020static int ext4_journalled_set_page_dirty(struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003021{
3022 SetPageChecked(page);
3023 return __set_page_dirty_nobuffers(page);
3024}
3025
Mingming Cao617ba132006-10-11 01:20:53 -07003026static const struct address_space_operations ext4_ordered_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003027 .readpage = ext4_readpage,
3028 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003029 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003030 .write_begin = ext4_write_begin,
3031 .write_end = ext4_ordered_write_end,
3032 .bmap = ext4_bmap,
3033 .invalidatepage = ext4_invalidatepage,
3034 .releasepage = ext4_releasepage,
3035 .direct_IO = ext4_direct_IO,
3036 .migratepage = buffer_migrate_page,
3037 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003038 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003039};
3040
Mingming Cao617ba132006-10-11 01:20:53 -07003041static const struct address_space_operations ext4_writeback_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003042 .readpage = ext4_readpage,
3043 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003044 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003045 .write_begin = ext4_write_begin,
3046 .write_end = ext4_writeback_write_end,
3047 .bmap = ext4_bmap,
3048 .invalidatepage = ext4_invalidatepage,
3049 .releasepage = ext4_releasepage,
3050 .direct_IO = ext4_direct_IO,
3051 .migratepage = buffer_migrate_page,
3052 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003053 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003054};
3055
Mingming Cao617ba132006-10-11 01:20:53 -07003056static const struct address_space_operations ext4_journalled_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003057 .readpage = ext4_readpage,
3058 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003059 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003060 .write_begin = ext4_write_begin,
3061 .write_end = ext4_journalled_write_end,
3062 .set_page_dirty = ext4_journalled_set_page_dirty,
3063 .bmap = ext4_bmap,
3064 .invalidatepage = ext4_invalidatepage,
3065 .releasepage = ext4_releasepage,
Theodore Ts'o84ebd792011-08-31 11:56:51 -04003066 .direct_IO = ext4_direct_IO,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003067 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003068 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003069};
3070
Alex Tomas64769242008-07-11 19:27:31 -04003071static const struct address_space_operations ext4_da_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003072 .readpage = ext4_readpage,
3073 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003074 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003075 .writepages = ext4_da_writepages,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003076 .write_begin = ext4_da_write_begin,
3077 .write_end = ext4_da_write_end,
3078 .bmap = ext4_bmap,
3079 .invalidatepage = ext4_da_invalidatepage,
3080 .releasepage = ext4_releasepage,
3081 .direct_IO = ext4_direct_IO,
3082 .migratepage = buffer_migrate_page,
3083 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003084 .error_remove_page = generic_error_remove_page,
Alex Tomas64769242008-07-11 19:27:31 -04003085};
3086
Mingming Cao617ba132006-10-11 01:20:53 -07003087void ext4_set_aops(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003088{
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -04003089 if (ext4_should_order_data(inode) &&
3090 test_opt(inode->i_sb, DELALLOC))
3091 inode->i_mapping->a_ops = &ext4_da_aops;
3092 else if (ext4_should_order_data(inode))
Mingming Cao617ba132006-10-11 01:20:53 -07003093 inode->i_mapping->a_ops = &ext4_ordered_aops;
Alex Tomas64769242008-07-11 19:27:31 -04003094 else if (ext4_should_writeback_data(inode) &&
3095 test_opt(inode->i_sb, DELALLOC))
3096 inode->i_mapping->a_ops = &ext4_da_aops;
Mingming Cao617ba132006-10-11 01:20:53 -07003097 else if (ext4_should_writeback_data(inode))
3098 inode->i_mapping->a_ops = &ext4_writeback_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003099 else
Mingming Cao617ba132006-10-11 01:20:53 -07003100 inode->i_mapping->a_ops = &ext4_journalled_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003101}
3102
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003103
3104/*
3105 * ext4_discard_partial_page_buffers()
3106 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3107 * This function finds and locks the page containing the offset
3108 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3109 * Calling functions that already have the page locked should call
3110 * ext4_discard_partial_page_buffers_no_lock directly.
3111 */
3112int ext4_discard_partial_page_buffers(handle_t *handle,
3113 struct address_space *mapping, loff_t from,
3114 loff_t length, int flags)
3115{
3116 struct inode *inode = mapping->host;
3117 struct page *page;
3118 int err = 0;
3119
3120 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3121 mapping_gfp_mask(mapping) & ~__GFP_FS);
3122 if (!page)
Yongqiang Yang5129d052011-10-31 17:56:10 -04003123 return -ENOMEM;
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003124
3125 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3126 from, length, flags);
3127
3128 unlock_page(page);
3129 page_cache_release(page);
3130 return err;
3131}
3132
3133/*
3134 * ext4_discard_partial_page_buffers_no_lock()
3135 * Zeros a page range of length 'length' starting from offset 'from'.
3136 * Buffer heads that correspond to the block aligned regions of the
3137 * zeroed range will be unmapped. Unblock aligned regions
3138 * will have the corresponding buffer head mapped if needed so that
3139 * that region of the page can be updated with the partial zero out.
3140 *
3141 * This function assumes that the page has already been locked. The
3142 * The range to be discarded must be contained with in the given page.
3143 * If the specified range exceeds the end of the page it will be shortened
3144 * to the end of the page that corresponds to 'from'. This function is
3145 * appropriate for updating a page and it buffer heads to be unmapped and
3146 * zeroed for blocks that have been either released, or are going to be
3147 * released.
3148 *
3149 * handle: The journal handle
3150 * inode: The files inode
3151 * page: A locked page that contains the offset "from"
3152 * from: The starting byte offset (from the begining of the file)
3153 * to begin discarding
3154 * len: The length of bytes to discard
3155 * flags: Optional flags that may be used:
3156 *
3157 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3158 * Only zero the regions of the page whose buffer heads
3159 * have already been unmapped. This flag is appropriate
3160 * for updateing the contents of a page whose blocks may
3161 * have already been released, and we only want to zero
3162 * out the regions that correspond to those released blocks.
3163 *
3164 * Returns zero on sucess or negative on failure.
3165 */
Eric Sandeen5f163cc2012-01-04 22:33:28 -05003166static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003167 struct inode *inode, struct page *page, loff_t from,
3168 loff_t length, int flags)
3169{
3170 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3171 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3172 unsigned int blocksize, max, pos;
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003173 ext4_lblk_t iblock;
3174 struct buffer_head *bh;
3175 int err = 0;
3176
3177 blocksize = inode->i_sb->s_blocksize;
3178 max = PAGE_CACHE_SIZE - offset;
3179
3180 if (index != page->index)
3181 return -EINVAL;
3182
3183 /*
3184 * correct length if it does not fall between
3185 * 'from' and the end of the page
3186 */
3187 if (length > max || length < 0)
3188 length = max;
3189
3190 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3191
Yongqiang Yang093e6e32011-12-13 22:05:05 -05003192 if (!page_has_buffers(page))
3193 create_empty_buffers(page, blocksize, 0);
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003194
3195 /* Find the buffer that contains "offset" */
3196 bh = page_buffers(page);
3197 pos = blocksize;
3198 while (offset >= pos) {
3199 bh = bh->b_this_page;
3200 iblock++;
3201 pos += blocksize;
3202 }
3203
3204 pos = offset;
3205 while (pos < offset + length) {
Yongqiang Yange260daf2011-10-31 17:54:36 -04003206 unsigned int end_of_block, range_to_discard;
3207
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003208 err = 0;
3209
3210 /* The length of space left to zero and unmap */
3211 range_to_discard = offset + length - pos;
3212
3213 /* The length of space until the end of the block */
3214 end_of_block = blocksize - (pos & (blocksize-1));
3215
3216 /*
3217 * Do not unmap or zero past end of block
3218 * for this buffer head
3219 */
3220 if (range_to_discard > end_of_block)
3221 range_to_discard = end_of_block;
3222
3223
3224 /*
3225 * Skip this buffer head if we are only zeroing unampped
3226 * regions of the page
3227 */
3228 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3229 buffer_mapped(bh))
3230 goto next;
3231
3232 /* If the range is block aligned, unmap */
3233 if (range_to_discard == blocksize) {
3234 clear_buffer_dirty(bh);
3235 bh->b_bdev = NULL;
3236 clear_buffer_mapped(bh);
3237 clear_buffer_req(bh);
3238 clear_buffer_new(bh);
3239 clear_buffer_delay(bh);
3240 clear_buffer_unwritten(bh);
3241 clear_buffer_uptodate(bh);
3242 zero_user(page, pos, range_to_discard);
3243 BUFFER_TRACE(bh, "Buffer discarded");
3244 goto next;
3245 }
3246
3247 /*
3248 * If this block is not completely contained in the range
3249 * to be discarded, then it is not going to be released. Because
3250 * we need to keep this block, we need to make sure this part
3251 * of the page is uptodate before we modify it by writeing
3252 * partial zeros on it.
3253 */
3254 if (!buffer_mapped(bh)) {
3255 /*
3256 * Buffer head must be mapped before we can read
3257 * from the block
3258 */
3259 BUFFER_TRACE(bh, "unmapped");
3260 ext4_get_block(inode, iblock, bh, 0);
3261 /* unmapped? It's a hole - nothing to do */
3262 if (!buffer_mapped(bh)) {
3263 BUFFER_TRACE(bh, "still unmapped");
3264 goto next;
3265 }
3266 }
3267
3268 /* Ok, it's mapped. Make sure it's up-to-date */
3269 if (PageUptodate(page))
3270 set_buffer_uptodate(bh);
3271
3272 if (!buffer_uptodate(bh)) {
3273 err = -EIO;
3274 ll_rw_block(READ, 1, &bh);
3275 wait_on_buffer(bh);
3276 /* Uhhuh. Read error. Complain and punt.*/
3277 if (!buffer_uptodate(bh))
3278 goto next;
3279 }
3280
3281 if (ext4_should_journal_data(inode)) {
3282 BUFFER_TRACE(bh, "get write access");
3283 err = ext4_journal_get_write_access(handle, bh);
3284 if (err)
3285 goto next;
3286 }
3287
3288 zero_user(page, pos, range_to_discard);
3289
3290 err = 0;
3291 if (ext4_should_journal_data(inode)) {
3292 err = ext4_handle_dirty_metadata(handle, inode, bh);
Theodore Ts'odecbd912011-09-06 02:37:06 -04003293 } else
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003294 mark_buffer_dirty(bh);
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003295
3296 BUFFER_TRACE(bh, "Partial buffer zeroed");
3297next:
3298 bh = bh->b_this_page;
3299 iblock++;
3300 pos += range_to_discard;
3301 }
3302
3303 return err;
3304}
3305
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003306int ext4_can_truncate(struct inode *inode)
3307{
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003308 if (S_ISREG(inode->i_mode))
3309 return 1;
3310 if (S_ISDIR(inode->i_mode))
3311 return 1;
3312 if (S_ISLNK(inode->i_mode))
3313 return !ext4_inode_is_fast_symlink(inode);
3314 return 0;
3315}
3316
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003317/*
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003318 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3319 * associated with the given offset and length
3320 *
3321 * @inode: File inode
3322 * @offset: The offset where the hole will begin
3323 * @len: The length of the hole
3324 *
3325 * Returns: 0 on sucess or negative on failure
3326 */
3327
3328int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3329{
3330 struct inode *inode = file->f_path.dentry->d_inode;
3331 if (!S_ISREG(inode->i_mode))
3332 return -ENOTSUPP;
3333
3334 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3335 /* TODO: Add support for non extent hole punching */
3336 return -ENOTSUPP;
3337 }
3338
Theodore Ts'obab08ab2011-09-09 18:36:51 -04003339 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3340 /* TODO: Add support for bigalloc file systems */
3341 return -ENOTSUPP;
3342 }
3343
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003344 return ext4_ext_punch_hole(file, offset, length);
3345}
3346
3347/*
Mingming Cao617ba132006-10-11 01:20:53 -07003348 * ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003349 *
Mingming Cao617ba132006-10-11 01:20:53 -07003350 * We block out ext4_get_block() block instantiations across the entire
3351 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003352 * simultaneously on behalf of the same inode.
3353 *
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08003354 * As we work through the truncate and commit bits of it to the journal there
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003355 * is one core, guiding principle: the file's tree must always be consistent on
3356 * disk. We must be able to restart the truncate after a crash.
3357 *
3358 * The file's tree may be transiently inconsistent in memory (although it
3359 * probably isn't), but whenever we close off and commit a journal transaction,
3360 * the contents of (the filesystem + the journal) must be consistent and
3361 * restartable. It's pretty simple, really: bottom up, right to left (although
3362 * left-to-right works OK too).
3363 *
3364 * Note that at recovery time, journal replay occurs *before* the restart of
3365 * truncate against the orphan inode list.
3366 *
3367 * The committed inode has the new, desired i_size (which is the same as
Mingming Cao617ba132006-10-11 01:20:53 -07003368 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003369 * that this inode's truncate did not complete and it will again call
Mingming Cao617ba132006-10-11 01:20:53 -07003370 * ext4_truncate() to have another go. So there will be instantiated blocks
3371 * to the right of the truncation point in a crashed ext4 filesystem. But
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003372 * that's fine - as long as they are linked from the inode, the post-crash
Mingming Cao617ba132006-10-11 01:20:53 -07003373 * ext4_truncate() run will find them and release them.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003374 */
Mingming Cao617ba132006-10-11 01:20:53 -07003375void ext4_truncate(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003376{
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003377 trace_ext4_truncate_enter(inode);
3378
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003379 if (!ext4_can_truncate(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003380 return;
3381
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04003382 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05003383
Theodore Ts'o5534fb52009-09-17 09:34:16 -04003384 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003385 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
Theodore Ts'o7d8f9f72009-02-24 08:21:14 -05003386
Amir Goldsteinff9893d2011-06-27 16:36:31 -04003387 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Jan Karacf108bc2008-07-11 19:27:31 -04003388 ext4_ext_truncate(inode);
Amir Goldsteinff9893d2011-06-27 16:36:31 -04003389 else
3390 ext4_ind_truncate(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07003391
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003392 trace_ext4_truncate_exit(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003393}
3394
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003395/*
Mingming Cao617ba132006-10-11 01:20:53 -07003396 * ext4_get_inode_loc returns with an extra refcount against the inode's
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003397 * underlying buffer_head on success. If 'in_mem' is true, we have all
3398 * data in memory that is needed to recreate the on-disk version of this
3399 * inode.
3400 */
Mingming Cao617ba132006-10-11 01:20:53 -07003401static int __ext4_get_inode_loc(struct inode *inode,
3402 struct ext4_iloc *iloc, int in_mem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003403{
Theodore Ts'o240799c2008-10-09 23:53:47 -04003404 struct ext4_group_desc *gdp;
3405 struct buffer_head *bh;
3406 struct super_block *sb = inode->i_sb;
3407 ext4_fsblk_t block;
3408 int inodes_per_block, inode_offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003409
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05003410 iloc->bh = NULL;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003411 if (!ext4_valid_inum(sb, inode->i_ino))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003412 return -EIO;
3413
Theodore Ts'o240799c2008-10-09 23:53:47 -04003414 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3415 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3416 if (!gdp)
3417 return -EIO;
3418
3419 /*
3420 * Figure out the offset within the block group inode table
3421 */
Tao Ma00d09882011-05-09 10:26:41 -04003422 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003423 inode_offset = ((inode->i_ino - 1) %
3424 EXT4_INODES_PER_GROUP(sb));
3425 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3426 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3427
3428 bh = sb_getblk(sb, block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003429 if (!bh) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04003430 EXT4_ERROR_INODE_BLOCK(inode, block,
3431 "unable to read itable block");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003432 return -EIO;
3433 }
3434 if (!buffer_uptodate(bh)) {
3435 lock_buffer(bh);
Hidehiro Kawai9c83a922008-07-26 16:39:26 -04003436
3437 /*
3438 * If the buffer has the write error flag, we have failed
3439 * to write out another inode in the same block. In this
3440 * case, we don't have to read the block because we may
3441 * read the old inode data successfully.
3442 */
3443 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3444 set_buffer_uptodate(bh);
3445
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003446 if (buffer_uptodate(bh)) {
3447 /* someone brought it uptodate while we waited */
3448 unlock_buffer(bh);
3449 goto has_buffer;
3450 }
3451
3452 /*
3453 * If we have all information of the inode in memory and this
3454 * is the only valid inode in the block, we need not read the
3455 * block.
3456 */
3457 if (in_mem) {
3458 struct buffer_head *bitmap_bh;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003459 int i, start;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003460
Theodore Ts'o240799c2008-10-09 23:53:47 -04003461 start = inode_offset & ~(inodes_per_block - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003462
3463 /* Is the inode bitmap in cache? */
Theodore Ts'o240799c2008-10-09 23:53:47 -04003464 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003465 if (!bitmap_bh)
3466 goto make_io;
3467
3468 /*
3469 * If the inode bitmap isn't in cache then the
3470 * optimisation may end up performing two reads instead
3471 * of one, so skip it.
3472 */
3473 if (!buffer_uptodate(bitmap_bh)) {
3474 brelse(bitmap_bh);
3475 goto make_io;
3476 }
Theodore Ts'o240799c2008-10-09 23:53:47 -04003477 for (i = start; i < start + inodes_per_block; i++) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003478 if (i == inode_offset)
3479 continue;
Mingming Cao617ba132006-10-11 01:20:53 -07003480 if (ext4_test_bit(i, bitmap_bh->b_data))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003481 break;
3482 }
3483 brelse(bitmap_bh);
Theodore Ts'o240799c2008-10-09 23:53:47 -04003484 if (i == start + inodes_per_block) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003485 /* all other inodes are free, so skip I/O */
3486 memset(bh->b_data, 0, bh->b_size);
3487 set_buffer_uptodate(bh);
3488 unlock_buffer(bh);
3489 goto has_buffer;
3490 }
3491 }
3492
3493make_io:
3494 /*
Theodore Ts'o240799c2008-10-09 23:53:47 -04003495 * If we need to do any I/O, try to pre-readahead extra
3496 * blocks from the inode table.
3497 */
3498 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3499 ext4_fsblk_t b, end, table;
3500 unsigned num;
3501
3502 table = ext4_inode_table(sb, gdp);
Theodore Ts'ob713a5e2009-03-31 09:11:14 -04003503 /* s_inode_readahead_blks is always a power of 2 */
Theodore Ts'o240799c2008-10-09 23:53:47 -04003504 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3505 if (table > b)
3506 b = table;
3507 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3508 num = EXT4_INODES_PER_GROUP(sb);
3509 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3510 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05003511 num -= ext4_itable_unused_count(sb, gdp);
Theodore Ts'o240799c2008-10-09 23:53:47 -04003512 table += num / inodes_per_block;
3513 if (end > table)
3514 end = table;
3515 while (b <= end)
3516 sb_breadahead(sb, b++);
3517 }
3518
3519 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003520 * There are other valid inodes in the buffer, this inode
3521 * has in-inode xattrs, or we don't have this inode in memory.
3522 * Read the block from disk.
3523 */
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003524 trace_ext4_load_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003525 get_bh(bh);
3526 bh->b_end_io = end_buffer_read_sync;
Christoph Hellwig65299a32011-08-23 14:50:29 +02003527 submit_bh(READ | REQ_META | REQ_PRIO, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003528 wait_on_buffer(bh);
3529 if (!buffer_uptodate(bh)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04003530 EXT4_ERROR_INODE_BLOCK(inode, block,
3531 "unable to read itable block");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003532 brelse(bh);
3533 return -EIO;
3534 }
3535 }
3536has_buffer:
3537 iloc->bh = bh;
3538 return 0;
3539}
3540
Mingming Cao617ba132006-10-11 01:20:53 -07003541int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003542{
3543 /* We have all inode data except xattrs in memory here. */
Mingming Cao617ba132006-10-11 01:20:53 -07003544 return __ext4_get_inode_loc(inode, iloc,
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003545 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003546}
3547
Mingming Cao617ba132006-10-11 01:20:53 -07003548void ext4_set_inode_flags(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003549{
Mingming Cao617ba132006-10-11 01:20:53 -07003550 unsigned int flags = EXT4_I(inode)->i_flags;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003551
3552 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
Mingming Cao617ba132006-10-11 01:20:53 -07003553 if (flags & EXT4_SYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003554 inode->i_flags |= S_SYNC;
Mingming Cao617ba132006-10-11 01:20:53 -07003555 if (flags & EXT4_APPEND_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003556 inode->i_flags |= S_APPEND;
Mingming Cao617ba132006-10-11 01:20:53 -07003557 if (flags & EXT4_IMMUTABLE_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003558 inode->i_flags |= S_IMMUTABLE;
Mingming Cao617ba132006-10-11 01:20:53 -07003559 if (flags & EXT4_NOATIME_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003560 inode->i_flags |= S_NOATIME;
Mingming Cao617ba132006-10-11 01:20:53 -07003561 if (flags & EXT4_DIRSYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003562 inode->i_flags |= S_DIRSYNC;
3563}
3564
Jan Karaff9ddf72007-07-18 09:24:20 -04003565/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3566void ext4_get_inode_flags(struct ext4_inode_info *ei)
3567{
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04003568 unsigned int vfs_fl;
3569 unsigned long old_fl, new_fl;
Jan Karaff9ddf72007-07-18 09:24:20 -04003570
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04003571 do {
3572 vfs_fl = ei->vfs_inode.i_flags;
3573 old_fl = ei->i_flags;
3574 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3575 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3576 EXT4_DIRSYNC_FL);
3577 if (vfs_fl & S_SYNC)
3578 new_fl |= EXT4_SYNC_FL;
3579 if (vfs_fl & S_APPEND)
3580 new_fl |= EXT4_APPEND_FL;
3581 if (vfs_fl & S_IMMUTABLE)
3582 new_fl |= EXT4_IMMUTABLE_FL;
3583 if (vfs_fl & S_NOATIME)
3584 new_fl |= EXT4_NOATIME_FL;
3585 if (vfs_fl & S_DIRSYNC)
3586 new_fl |= EXT4_DIRSYNC_FL;
3587 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
Jan Karaff9ddf72007-07-18 09:24:20 -04003588}
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003589
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003590static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003591 struct ext4_inode_info *ei)
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003592{
3593 blkcnt_t i_blocks ;
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05003594 struct inode *inode = &(ei->vfs_inode);
3595 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003596
3597 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3598 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3599 /* we are using combined 48 bit field */
3600 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3601 le32_to_cpu(raw_inode->i_blocks_lo);
Theodore Ts'o07a03822010-06-14 09:54:48 -04003602 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05003603 /* i_blocks represent file system block size */
3604 return i_blocks << (inode->i_blkbits - 9);
3605 } else {
3606 return i_blocks;
3607 }
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003608 } else {
3609 return le32_to_cpu(raw_inode->i_blocks_lo);
3610 }
3611}
Jan Karaff9ddf72007-07-18 09:24:20 -04003612
David Howells1d1fe1e2008-02-07 00:15:37 -08003613struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003614{
Mingming Cao617ba132006-10-11 01:20:53 -07003615 struct ext4_iloc iloc;
3616 struct ext4_inode *raw_inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08003617 struct ext4_inode_info *ei;
David Howells1d1fe1e2008-02-07 00:15:37 -08003618 struct inode *inode;
Jan Karab436b9b2009-12-08 23:51:10 -05003619 journal_t *journal = EXT4_SB(sb)->s_journal;
David Howells1d1fe1e2008-02-07 00:15:37 -08003620 long ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003621 int block;
3622
David Howells1d1fe1e2008-02-07 00:15:37 -08003623 inode = iget_locked(sb, ino);
3624 if (!inode)
3625 return ERR_PTR(-ENOMEM);
3626 if (!(inode->i_state & I_NEW))
3627 return inode;
3628
3629 ei = EXT4_I(inode);
Peter Huewe7dc57612011-02-21 21:01:42 -05003630 iloc.bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003631
David Howells1d1fe1e2008-02-07 00:15:37 -08003632 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3633 if (ret < 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003634 goto bad_inode;
Mingming Cao617ba132006-10-11 01:20:53 -07003635 raw_inode = ext4_raw_inode(&iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003636 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3637 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3638 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003639 if (!(test_opt(inode->i_sb, NO_UID32))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003640 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3641 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3642 }
Miklos Szeredibfe86842011-10-28 14:13:29 +02003643 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003644
Theodore Ts'o353eb832011-01-10 12:18:25 -05003645 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003646 ei->i_dir_start_lookup = 0;
3647 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3648 /* We now have enough fields to check if the inode was active or not.
3649 * This is needed because nfsd might try to access dead inodes
3650 * the test is that same one that e2fsck uses
3651 * NeilBrown 1999oct15
3652 */
3653 if (inode->i_nlink == 0) {
3654 if (inode->i_mode == 0 ||
Mingming Cao617ba132006-10-11 01:20:53 -07003655 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003656 /* this inode is deleted */
David Howells1d1fe1e2008-02-07 00:15:37 -08003657 ret = -ESTALE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003658 goto bad_inode;
3659 }
3660 /* The only unlinked inodes we let through here have
3661 * valid i_mode and are being read by the orphan
3662 * recovery code: that's fine, we're about to complete
3663 * the process of deleting those. */
3664 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003665 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003666 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05003667 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
Theodore Ts'oa9e81742009-04-24 16:11:18 -04003668 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07003669 ei->i_file_acl |=
3670 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05003671 inode->i_size = ext4_isize(raw_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003672 ei->i_disksize = inode->i_size;
Dmitry Monakhova9e7f442009-12-14 15:21:14 +03003673#ifdef CONFIG_QUOTA
3674 ei->i_reserved_quota = 0;
3675#endif
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003676 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3677 ei->i_block_group = iloc.block_group;
Theodore Ts'oa4912122009-03-12 12:18:34 -04003678 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003679 /*
3680 * NOTE! The in-memory inode i_data array is in little-endian order
3681 * even on big-endian machines: we do NOT byteswap the block numbers!
3682 */
Mingming Cao617ba132006-10-11 01:20:53 -07003683 for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003684 ei->i_data[block] = raw_inode->i_block[block];
3685 INIT_LIST_HEAD(&ei->i_orphan);
3686
Jan Karab436b9b2009-12-08 23:51:10 -05003687 /*
3688 * Set transaction id's of transactions that have to be committed
3689 * to finish f[data]sync. We set them to currently running transaction
3690 * as we cannot be sure that the inode or some of its metadata isn't
3691 * part of the transaction - the inode could have been reclaimed and
3692 * now it is reread from disk.
3693 */
3694 if (journal) {
3695 transaction_t *transaction;
3696 tid_t tid;
3697
Theodore Ts'oa931da62010-08-03 21:35:12 -04003698 read_lock(&journal->j_state_lock);
Jan Karab436b9b2009-12-08 23:51:10 -05003699 if (journal->j_running_transaction)
3700 transaction = journal->j_running_transaction;
3701 else
3702 transaction = journal->j_committing_transaction;
3703 if (transaction)
3704 tid = transaction->t_tid;
3705 else
3706 tid = journal->j_commit_sequence;
Theodore Ts'oa931da62010-08-03 21:35:12 -04003707 read_unlock(&journal->j_state_lock);
Jan Karab436b9b2009-12-08 23:51:10 -05003708 ei->i_sync_tid = tid;
3709 ei->i_datasync_tid = tid;
3710 }
3711
Eric Sandeen0040d982008-02-05 22:36:43 -05003712 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003713 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
Mingming Cao617ba132006-10-11 01:20:53 -07003714 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
Kirill Korotaeve5d28612007-06-23 17:16:51 -07003715 EXT4_INODE_SIZE(inode->i_sb)) {
David Howells1d1fe1e2008-02-07 00:15:37 -08003716 ret = -EIO;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003717 goto bad_inode;
Kirill Korotaeve5d28612007-06-23 17:16:51 -07003718 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003719 if (ei->i_extra_isize == 0) {
3720 /* The extra space is currently unused. Use it. */
Mingming Cao617ba132006-10-11 01:20:53 -07003721 ei->i_extra_isize = sizeof(struct ext4_inode) -
3722 EXT4_GOOD_OLD_INODE_SIZE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003723 } else {
3724 __le32 *magic = (void *)raw_inode +
Mingming Cao617ba132006-10-11 01:20:53 -07003725 EXT4_GOOD_OLD_INODE_SIZE +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003726 ei->i_extra_isize;
Mingming Cao617ba132006-10-11 01:20:53 -07003727 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003728 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003729 }
3730 } else
3731 ei->i_extra_isize = 0;
3732
Kalpak Shahef7f3832007-07-18 09:15:20 -04003733 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3734 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3735 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3736 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3737
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05003738 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3739 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3740 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3741 inode->i_version |=
3742 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
3743 }
3744
Theodore Ts'oc4b5a612009-04-24 18:45:35 -04003745 ret = 0;
Theodore Ts'o485c26e2009-04-24 13:43:20 -04003746 if (ei->i_file_acl &&
Theodore Ts'o10329882009-11-15 15:29:56 -05003747 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04003748 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
3749 ei->i_file_acl);
Theodore Ts'o485c26e2009-04-24 13:43:20 -04003750 ret = -EIO;
3751 goto bad_inode;
Theodore Ts'o07a03822010-06-14 09:54:48 -04003752 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
Theodore Ts'oc4b5a612009-04-24 18:45:35 -04003753 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3754 (S_ISLNK(inode->i_mode) &&
3755 !ext4_inode_is_fast_symlink(inode)))
3756 /* Validate extent which is part of inode */
3757 ret = ext4_ext_check_inode(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003758 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
Thiemo Nagelfe2c8192009-03-31 08:36:10 -04003759 (S_ISLNK(inode->i_mode) &&
3760 !ext4_inode_is_fast_symlink(inode))) {
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003761 /* Validate block references which are part of inode */
Theodore Ts'o1f7d1e72011-06-27 19:16:02 -04003762 ret = ext4_ind_check_inode(inode);
Thiemo Nagelfe2c8192009-03-31 08:36:10 -04003763 }
Theodore Ts'o567f3e92009-11-14 08:19:05 -05003764 if (ret)
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003765 goto bad_inode;
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -04003766
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003767 if (S_ISREG(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07003768 inode->i_op = &ext4_file_inode_operations;
3769 inode->i_fop = &ext4_file_operations;
3770 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003771 } else if (S_ISDIR(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07003772 inode->i_op = &ext4_dir_inode_operations;
3773 inode->i_fop = &ext4_dir_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003774 } else if (S_ISLNK(inode->i_mode)) {
Duane Griffine83c1392008-12-19 20:47:15 +00003775 if (ext4_inode_is_fast_symlink(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07003776 inode->i_op = &ext4_fast_symlink_inode_operations;
Duane Griffine83c1392008-12-19 20:47:15 +00003777 nd_terminate_link(ei->i_data, inode->i_size,
3778 sizeof(ei->i_data) - 1);
3779 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07003780 inode->i_op = &ext4_symlink_inode_operations;
3781 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003782 }
Theodore Ts'o563bdd62009-03-26 00:06:19 -04003783 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3784 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07003785 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003786 if (raw_inode->i_block[0])
3787 init_special_inode(inode, inode->i_mode,
3788 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3789 else
3790 init_special_inode(inode, inode->i_mode,
3791 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
Theodore Ts'o563bdd62009-03-26 00:06:19 -04003792 } else {
Theodore Ts'o563bdd62009-03-26 00:06:19 -04003793 ret = -EIO;
Theodore Ts'o24676da2010-05-16 21:00:00 -04003794 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
Theodore Ts'o563bdd62009-03-26 00:06:19 -04003795 goto bad_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003796 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003797 brelse(iloc.bh);
Mingming Cao617ba132006-10-11 01:20:53 -07003798 ext4_set_inode_flags(inode);
David Howells1d1fe1e2008-02-07 00:15:37 -08003799 unlock_new_inode(inode);
3800 return inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003801
3802bad_inode:
Theodore Ts'o567f3e92009-11-14 08:19:05 -05003803 brelse(iloc.bh);
David Howells1d1fe1e2008-02-07 00:15:37 -08003804 iget_failed(inode);
3805 return ERR_PTR(ret);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003806}
3807
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003808static int ext4_inode_blocks_set(handle_t *handle,
3809 struct ext4_inode *raw_inode,
3810 struct ext4_inode_info *ei)
3811{
3812 struct inode *inode = &(ei->vfs_inode);
3813 u64 i_blocks = inode->i_blocks;
3814 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003815
3816 if (i_blocks <= ~0U) {
3817 /*
3818 * i_blocks can be represnted in a 32 bit variable
3819 * as multiple of 512 bytes
3820 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05003821 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003822 raw_inode->i_blocks_high = 0;
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04003823 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Theodore Ts'of287a1a2008-10-16 22:50:48 -04003824 return 0;
3825 }
3826 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
3827 return -EFBIG;
3828
3829 if (i_blocks <= 0xffffffffffffULL) {
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003830 /*
3831 * i_blocks can be represented in a 48 bit variable
3832 * as multiple of 512 bytes
3833 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05003834 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003835 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04003836 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003837 } else {
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04003838 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05003839 /* i_block is stored in file system block size */
3840 i_blocks = i_blocks >> (inode->i_blkbits - 9);
3841 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3842 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003843 }
Theodore Ts'of287a1a2008-10-16 22:50:48 -04003844 return 0;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003845}
3846
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003847/*
3848 * Post the struct inode info into an on-disk inode location in the
3849 * buffer-cache. This gobbles the caller's reference to the
3850 * buffer_head in the inode location struct.
3851 *
3852 * The caller must have write access to iloc->bh.
3853 */
Mingming Cao617ba132006-10-11 01:20:53 -07003854static int ext4_do_update_inode(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003855 struct inode *inode,
Frank Mayhar830156c2009-09-29 10:07:47 -04003856 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003857{
Mingming Cao617ba132006-10-11 01:20:53 -07003858 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
3859 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003860 struct buffer_head *bh = iloc->bh;
3861 int err = 0, rc, block;
3862
3863 /* For fields not not tracking in the in-memory inode,
3864 * initialise them to zero for new inodes. */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003865 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
Mingming Cao617ba132006-10-11 01:20:53 -07003866 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003867
Jan Karaff9ddf72007-07-18 09:24:20 -04003868 ext4_get_inode_flags(ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003869 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003870 if (!(test_opt(inode->i_sb, NO_UID32))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003871 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
3872 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
3873/*
3874 * Fix up interoperability with old kernels. Otherwise, old inodes get
3875 * re-used with the upper 16 bits of the uid/gid intact
3876 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003877 if (!ei->i_dtime) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003878 raw_inode->i_uid_high =
3879 cpu_to_le16(high_16_bits(inode->i_uid));
3880 raw_inode->i_gid_high =
3881 cpu_to_le16(high_16_bits(inode->i_gid));
3882 } else {
3883 raw_inode->i_uid_high = 0;
3884 raw_inode->i_gid_high = 0;
3885 }
3886 } else {
3887 raw_inode->i_uid_low =
3888 cpu_to_le16(fs_high2lowuid(inode->i_uid));
3889 raw_inode->i_gid_low =
3890 cpu_to_le16(fs_high2lowgid(inode->i_gid));
3891 raw_inode->i_uid_high = 0;
3892 raw_inode->i_gid_high = 0;
3893 }
3894 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
Kalpak Shahef7f3832007-07-18 09:15:20 -04003895
3896 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
3897 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
3898 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
3899 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
3900
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003901 if (ext4_inode_blocks_set(handle, raw_inode, ei))
3902 goto out_brelse;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003903 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
Theodore Ts'o353eb832011-01-10 12:18:25 -05003904 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
Mingming Cao9b8f1f02006-10-11 01:21:13 -07003905 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
3906 cpu_to_le32(EXT4_OS_HURD))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07003907 raw_inode->i_file_acl_high =
3908 cpu_to_le16(ei->i_file_acl >> 32);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05003909 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05003910 ext4_isize_set(raw_inode, ei->i_disksize);
3911 if (ei->i_disksize > 0x7fffffffULL) {
3912 struct super_block *sb = inode->i_sb;
3913 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
3914 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
3915 EXT4_SB(sb)->s_es->s_rev_level ==
3916 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
3917 /* If this is the first large file
3918 * created, add a flag to the superblock.
3919 */
3920 err = ext4_journal_get_write_access(handle,
3921 EXT4_SB(sb)->s_sbh);
3922 if (err)
3923 goto out_brelse;
3924 ext4_update_dynamic_rev(sb);
3925 EXT4_SET_RO_COMPAT_FEATURE(sb,
Mingming Cao617ba132006-10-11 01:20:53 -07003926 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05003927 sb->s_dirt = 1;
Frank Mayhar03901312009-01-07 00:06:22 -05003928 ext4_handle_sync(handle);
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05003929 err = ext4_handle_dirty_metadata(handle, NULL,
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05003930 EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003931 }
3932 }
3933 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
3934 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
3935 if (old_valid_dev(inode->i_rdev)) {
3936 raw_inode->i_block[0] =
3937 cpu_to_le32(old_encode_dev(inode->i_rdev));
3938 raw_inode->i_block[1] = 0;
3939 } else {
3940 raw_inode->i_block[0] = 0;
3941 raw_inode->i_block[1] =
3942 cpu_to_le32(new_encode_dev(inode->i_rdev));
3943 raw_inode->i_block[2] = 0;
3944 }
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003945 } else
3946 for (block = 0; block < EXT4_N_BLOCKS; block++)
3947 raw_inode->i_block[block] = ei->i_data[block];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003948
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05003949 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
3950 if (ei->i_extra_isize) {
3951 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3952 raw_inode->i_version_hi =
3953 cpu_to_le32(inode->i_version >> 32);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003954 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05003955 }
3956
Frank Mayhar830156c2009-09-29 10:07:47 -04003957 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05003958 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
Frank Mayhar830156c2009-09-29 10:07:47 -04003959 if (!err)
3960 err = rc;
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003961 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003962
Jan Karab436b9b2009-12-08 23:51:10 -05003963 ext4_update_inode_fsync_trans(handle, inode, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003964out_brelse:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003965 brelse(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07003966 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003967 return err;
3968}
3969
3970/*
Mingming Cao617ba132006-10-11 01:20:53 -07003971 * ext4_write_inode()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003972 *
3973 * We are called from a few places:
3974 *
3975 * - Within generic_file_write() for O_SYNC files.
3976 * Here, there will be no transaction running. We wait for any running
3977 * trasnaction to commit.
3978 *
3979 * - Within sys_sync(), kupdate and such.
3980 * We wait on commit, if tol to.
3981 *
3982 * - Within prune_icache() (PF_MEMALLOC == true)
3983 * Here we simply return. We can't afford to block kswapd on the
3984 * journal commit.
3985 *
3986 * In all cases it is actually safe for us to return without doing anything,
3987 * because the inode has been copied into a raw inode buffer in
Mingming Cao617ba132006-10-11 01:20:53 -07003988 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003989 * knfsd.
3990 *
3991 * Note that we are absolutely dependent upon all inode dirtiers doing the
3992 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3993 * which we are interested.
3994 *
3995 * It would be a bug for them to not do this. The code:
3996 *
3997 * mark_inode_dirty(inode)
3998 * stuff();
3999 * inode->i_size = expr;
4000 *
4001 * is in error because a kswapd-driven write_inode() could occur while
4002 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4003 * will no longer be on the superblock's dirty inode list.
4004 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004005int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004006{
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004007 int err;
4008
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004009 if (current->flags & PF_MEMALLOC)
4010 return 0;
4011
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004012 if (EXT4_SB(inode->i_sb)->s_journal) {
4013 if (ext4_journal_current_handle()) {
4014 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4015 dump_stack();
4016 return -EIO;
4017 }
4018
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004019 if (wbc->sync_mode != WB_SYNC_ALL)
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004020 return 0;
4021
4022 err = ext4_force_commit(inode->i_sb);
4023 } else {
4024 struct ext4_iloc iloc;
4025
Curt Wohlgemuth8b472d72010-04-03 16:45:06 -04004026 err = __ext4_get_inode_loc(inode, &iloc, 0);
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004027 if (err)
4028 return err;
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004029 if (wbc->sync_mode == WB_SYNC_ALL)
Frank Mayhar830156c2009-09-29 10:07:47 -04004030 sync_dirty_buffer(iloc.bh);
4031 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04004032 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4033 "IO error syncing inode");
Frank Mayhar830156c2009-09-29 10:07:47 -04004034 err = -EIO;
4035 }
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04004036 brelse(iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004037 }
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004038 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004039}
4040
4041/*
Mingming Cao617ba132006-10-11 01:20:53 -07004042 * ext4_setattr()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004043 *
4044 * Called from notify_change.
4045 *
4046 * We want to trap VFS attempts to truncate the file as soon as
4047 * possible. In particular, we want to make sure that when the VFS
4048 * shrinks i_size, we put the inode on the orphan list and modify
4049 * i_disksize immediately, so that during the subsequent flushing of
4050 * dirty pages and freeing of disk blocks, we can guarantee that any
4051 * commit will leave the blocks being flushed in an unused state on
4052 * disk. (On recovery, the inode will get truncated and the blocks will
4053 * be freed, so we have a strong guarantee that no future commit will
4054 * leave these blocks visible to the user.)
4055 *
Jan Kara678aaf42008-07-11 19:27:31 -04004056 * Another thing we have to assure is that if we are in ordered mode
4057 * and inode is still attached to the committing transaction, we must
4058 * we start writeout of all the dirty pages which are being truncated.
4059 * This way we are sure that all the data written in the previous
4060 * transaction are already on disk (truncate waits for pages under
4061 * writeback).
4062 *
4063 * Called with inode->i_mutex down.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004064 */
Mingming Cao617ba132006-10-11 01:20:53 -07004065int ext4_setattr(struct dentry *dentry, struct iattr *attr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004066{
4067 struct inode *inode = dentry->d_inode;
4068 int error, rc = 0;
Dmitry Monakhov3d287de2010-10-27 22:08:46 -04004069 int orphan = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004070 const unsigned int ia_valid = attr->ia_valid;
4071
4072 error = inode_change_ok(inode, attr);
4073 if (error)
4074 return error;
4075
Dmitry Monakhov12755622010-04-08 22:04:20 +04004076 if (is_quota_modification(inode, attr))
Christoph Hellwig871a2932010-03-03 09:05:07 -05004077 dquot_initialize(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004078 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4079 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4080 handle_t *handle;
4081
4082 /* (user+group)*(old+new) structure, inode write (sb,
4083 * inode block, ? - but truncate inode update has it) */
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05004084 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
Dmitry Monakhov194074a2009-12-08 22:42:28 -05004085 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004086 if (IS_ERR(handle)) {
4087 error = PTR_ERR(handle);
4088 goto err_out;
4089 }
Christoph Hellwigb43fa822010-03-03 09:05:03 -05004090 error = dquot_transfer(inode, attr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004091 if (error) {
Mingming Cao617ba132006-10-11 01:20:53 -07004092 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004093 return error;
4094 }
4095 /* Update corresponding info in inode so that everything is in
4096 * one transaction */
4097 if (attr->ia_valid & ATTR_UID)
4098 inode->i_uid = attr->ia_uid;
4099 if (attr->ia_valid & ATTR_GID)
4100 inode->i_gid = attr->ia_gid;
Mingming Cao617ba132006-10-11 01:20:53 -07004101 error = ext4_mark_inode_dirty(handle, inode);
4102 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004103 }
4104
Eric Sandeene2b46572008-01-28 23:58:27 -05004105 if (attr->ia_valid & ATTR_SIZE) {
Christoph Hellwig562c72a2011-06-24 14:29:45 -04004106 inode_dio_wait(inode);
4107
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004108 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
Eric Sandeene2b46572008-01-28 23:58:27 -05004109 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4110
Theodore Ts'o0c095c72010-07-27 11:56:06 -04004111 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4112 return -EFBIG;
Eric Sandeene2b46572008-01-28 23:58:27 -05004113 }
4114 }
4115
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004116 if (S_ISREG(inode->i_mode) &&
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004117 attr->ia_valid & ATTR_SIZE &&
Theodore Ts'o072bd7e2011-05-23 15:13:02 -04004118 (attr->ia_size < inode->i_size)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004119 handle_t *handle;
4120
Mingming Cao617ba132006-10-11 01:20:53 -07004121 handle = ext4_journal_start(inode, 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004122 if (IS_ERR(handle)) {
4123 error = PTR_ERR(handle);
4124 goto err_out;
4125 }
Dmitry Monakhov3d287de2010-10-27 22:08:46 -04004126 if (ext4_handle_valid(handle)) {
4127 error = ext4_orphan_add(handle, inode);
4128 orphan = 1;
4129 }
Mingming Cao617ba132006-10-11 01:20:53 -07004130 EXT4_I(inode)->i_disksize = attr->ia_size;
4131 rc = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004132 if (!error)
4133 error = rc;
Mingming Cao617ba132006-10-11 01:20:53 -07004134 ext4_journal_stop(handle);
Jan Kara678aaf42008-07-11 19:27:31 -04004135
4136 if (ext4_should_order_data(inode)) {
4137 error = ext4_begin_ordered_truncate(inode,
4138 attr->ia_size);
4139 if (error) {
4140 /* Do as much error cleanup as possible */
4141 handle = ext4_journal_start(inode, 3);
4142 if (IS_ERR(handle)) {
4143 ext4_orphan_del(NULL, inode);
4144 goto err_out;
4145 }
4146 ext4_orphan_del(handle, inode);
Dmitry Monakhov3d287de2010-10-27 22:08:46 -04004147 orphan = 0;
Jan Kara678aaf42008-07-11 19:27:31 -04004148 ext4_journal_stop(handle);
4149 goto err_out;
4150 }
4151 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004152 }
4153
Theodore Ts'o072bd7e2011-05-23 15:13:02 -04004154 if (attr->ia_valid & ATTR_SIZE) {
4155 if (attr->ia_size != i_size_read(inode)) {
4156 truncate_setsize(inode, attr->ia_size);
4157 ext4_truncate(inode);
4158 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
4159 ext4_truncate(inode);
4160 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004161
Christoph Hellwig10257742010-06-04 11:30:02 +02004162 if (!rc) {
4163 setattr_copy(inode, attr);
4164 mark_inode_dirty(inode);
4165 }
4166
4167 /*
4168 * If the call to ext4_truncate failed to get a transaction handle at
4169 * all, we need to clean up the in-core orphan list manually.
4170 */
Dmitry Monakhov3d287de2010-10-27 22:08:46 -04004171 if (orphan && inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07004172 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004173
4174 if (!rc && (ia_valid & ATTR_MODE))
Mingming Cao617ba132006-10-11 01:20:53 -07004175 rc = ext4_acl_chmod(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004176
4177err_out:
Mingming Cao617ba132006-10-11 01:20:53 -07004178 ext4_std_error(inode->i_sb, error);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004179 if (!error)
4180 error = rc;
4181 return error;
4182}
4183
Mingming Cao3e3398a2008-07-11 19:27:31 -04004184int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4185 struct kstat *stat)
4186{
4187 struct inode *inode;
4188 unsigned long delalloc_blocks;
4189
4190 inode = dentry->d_inode;
4191 generic_fillattr(inode, stat);
4192
4193 /*
4194 * We can't update i_blocks if the block allocation is delayed
4195 * otherwise in the case of system crash before the real block
4196 * allocation is done, we will have i_blocks inconsistent with
4197 * on-disk file blocks.
4198 * We always keep i_blocks updated together with real
4199 * allocation. But to not confuse with user, stat
4200 * will return the blocks that include the delayed allocation
4201 * blocks for this file.
4202 */
Mingming Cao3e3398a2008-07-11 19:27:31 -04004203 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
Mingming Cao3e3398a2008-07-11 19:27:31 -04004204
4205 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4206 return 0;
4207}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004208
Mingming Caoa02908f2008-08-19 22:16:07 -04004209static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4210{
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004211 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Amir Goldstein8bb2b242011-06-27 17:10:28 -04004212 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
Theodore Ts'oac51d832008-11-06 16:49:36 -05004213 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
Mingming Caoa02908f2008-08-19 22:16:07 -04004214}
Theodore Ts'oac51d832008-11-06 16:49:36 -05004215
Mingming Caoa02908f2008-08-19 22:16:07 -04004216/*
4217 * Account for index blocks, block groups bitmaps and block group
4218 * descriptor blocks if modify datablocks and index blocks
4219 * worse case, the indexs blocks spread over different block groups
4220 *
4221 * If datablocks are discontiguous, they are possible to spread over
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02004222 * different block groups too. If they are contiuguous, with flexbg,
Mingming Caoa02908f2008-08-19 22:16:07 -04004223 * they could still across block group boundary.
4224 *
4225 * Also account for superblock, inode, quota and xattr blocks
4226 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04004227static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
Mingming Caoa02908f2008-08-19 22:16:07 -04004228{
Theodore Ts'o8df96752009-05-01 08:50:38 -04004229 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4230 int gdpblocks;
Mingming Caoa02908f2008-08-19 22:16:07 -04004231 int idxblocks;
4232 int ret = 0;
4233
4234 /*
4235 * How many index blocks need to touch to modify nrblocks?
4236 * The "Chunk" flag indicating whether the nrblocks is
4237 * physically contiguous on disk
4238 *
4239 * For Direct IO and fallocate, they calls get_block to allocate
4240 * one single extent at a time, so they could set the "Chunk" flag
4241 */
4242 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4243
4244 ret = idxblocks;
4245
4246 /*
4247 * Now let's see how many group bitmaps and group descriptors need
4248 * to account
4249 */
4250 groups = idxblocks;
4251 if (chunk)
4252 groups += 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004253 else
Mingming Caoa02908f2008-08-19 22:16:07 -04004254 groups += nrblocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004255
Mingming Caoa02908f2008-08-19 22:16:07 -04004256 gdpblocks = groups;
Theodore Ts'o8df96752009-05-01 08:50:38 -04004257 if (groups > ngroups)
4258 groups = ngroups;
Mingming Caoa02908f2008-08-19 22:16:07 -04004259 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4260 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4261
4262 /* bitmaps and block group descriptor blocks */
4263 ret += groups + gdpblocks;
4264
4265 /* Blocks for super block, inode, quota and xattr blocks */
4266 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004267
4268 return ret;
4269}
4270
4271/*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004272 * Calculate the total number of credits to reserve to fit
Mingming Caof3bd1f32008-08-19 22:16:03 -04004273 * the modification of a single pages into a single transaction,
4274 * which may include multiple chunks of block allocations.
Mingming Caoa02908f2008-08-19 22:16:07 -04004275 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04004276 * This could be called via ext4_write_begin()
Mingming Caoa02908f2008-08-19 22:16:07 -04004277 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04004278 * We need to consider the worse case, when
Mingming Caoa02908f2008-08-19 22:16:07 -04004279 * one new block per extent.
Mingming Caoa02908f2008-08-19 22:16:07 -04004280 */
4281int ext4_writepage_trans_blocks(struct inode *inode)
4282{
4283 int bpp = ext4_journal_blocks_per_page(inode);
4284 int ret;
4285
4286 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4287
4288 /* Account for data blocks for journalled mode */
4289 if (ext4_should_journal_data(inode))
4290 ret += bpp;
4291 return ret;
4292}
Mingming Caof3bd1f32008-08-19 22:16:03 -04004293
4294/*
4295 * Calculate the journal credits for a chunk of data modification.
4296 *
4297 * This is called from DIO, fallocate or whoever calling
Eric Sandeen79e83032010-07-27 11:56:07 -04004298 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
Mingming Caof3bd1f32008-08-19 22:16:03 -04004299 *
4300 * journal buffers for data blocks are not included here, as DIO
4301 * and fallocate do no need to journal data buffers.
4302 */
4303int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4304{
4305 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4306}
4307
Mingming Caoa02908f2008-08-19 22:16:07 -04004308/*
Mingming Cao617ba132006-10-11 01:20:53 -07004309 * The caller must have previously called ext4_reserve_inode_write().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004310 * Give this, we know that the caller already has write access to iloc->bh.
4311 */
Mingming Cao617ba132006-10-11 01:20:53 -07004312int ext4_mark_iloc_dirty(handle_t *handle,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004313 struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004314{
4315 int err = 0;
4316
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004317 if (test_opt(inode->i_sb, I_VERSION))
4318 inode_inc_iversion(inode);
4319
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004320 /* the do_update_inode consumes one bh->b_count */
4321 get_bh(iloc->bh);
4322
Mingming Caodab291a2006-10-11 01:21:01 -07004323 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
Frank Mayhar830156c2009-09-29 10:07:47 -04004324 err = ext4_do_update_inode(handle, inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004325 put_bh(iloc->bh);
4326 return err;
4327}
4328
4329/*
4330 * On success, We end up with an outstanding reference count against
4331 * iloc->bh. This _must_ be cleaned up later.
4332 */
4333
4334int
Mingming Cao617ba132006-10-11 01:20:53 -07004335ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4336 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004337{
Frank Mayhar03901312009-01-07 00:06:22 -05004338 int err;
4339
4340 err = ext4_get_inode_loc(inode, iloc);
4341 if (!err) {
4342 BUFFER_TRACE(iloc->bh, "get_write_access");
4343 err = ext4_journal_get_write_access(handle, iloc->bh);
4344 if (err) {
4345 brelse(iloc->bh);
4346 iloc->bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004347 }
4348 }
Mingming Cao617ba132006-10-11 01:20:53 -07004349 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004350 return err;
4351}
4352
4353/*
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004354 * Expand an inode by new_extra_isize bytes.
4355 * Returns 0 on success or negative error number on failure.
4356 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05004357static int ext4_expand_extra_isize(struct inode *inode,
4358 unsigned int new_extra_isize,
4359 struct ext4_iloc iloc,
4360 handle_t *handle)
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004361{
4362 struct ext4_inode *raw_inode;
4363 struct ext4_xattr_ibody_header *header;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004364
4365 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4366 return 0;
4367
4368 raw_inode = ext4_raw_inode(&iloc);
4369
4370 header = IHDR(inode, raw_inode);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004371
4372 /* No extended attributes present */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004373 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4374 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004375 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4376 new_extra_isize);
4377 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4378 return 0;
4379 }
4380
4381 /* try to expand with EAs present */
4382 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4383 raw_inode, handle);
4384}
4385
4386/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004387 * What we do here is to mark the in-core inode as clean with respect to inode
4388 * dirtiness (it may still be data-dirty).
4389 * This means that the in-core inode may be reaped by prune_icache
4390 * without having to perform any I/O. This is a very good thing,
4391 * because *any* task may call prune_icache - even ones which
4392 * have a transaction open against a different journal.
4393 *
4394 * Is this cheating? Not really. Sure, we haven't written the
4395 * inode out, but prune_icache isn't a user-visible syncing function.
4396 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4397 * we start and wait on commits.
4398 *
4399 * Is this efficient/effective? Well, we're being nice to the system
4400 * by cleaning up our inodes proactively so they can be reaped
4401 * without I/O. But we are potentially leaving up to five seconds'
4402 * worth of inodes floating about which prune_icache wants us to
4403 * write out. One way to fix that would be to get prune_icache()
4404 * to do a write_super() to free up some memory. It has the desired
4405 * effect.
4406 */
Mingming Cao617ba132006-10-11 01:20:53 -07004407int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004408{
Mingming Cao617ba132006-10-11 01:20:53 -07004409 struct ext4_iloc iloc;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004410 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4411 static unsigned int mnt_count;
4412 int err, ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004413
4414 might_sleep();
Theodore Ts'o7ff9c072010-11-08 13:51:33 -05004415 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
Mingming Cao617ba132006-10-11 01:20:53 -07004416 err = ext4_reserve_inode_write(handle, inode, &iloc);
Frank Mayhar03901312009-01-07 00:06:22 -05004417 if (ext4_handle_valid(handle) &&
4418 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004419 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004420 /*
4421 * We need extra buffer credits since we may write into EA block
4422 * with this same handle. If journal_extend fails, then it will
4423 * only result in a minor loss of functionality for that inode.
4424 * If this is felt to be critical, then e2fsck should be run to
4425 * force a large enough s_min_extra_isize.
4426 */
4427 if ((jbd2_journal_extend(handle,
4428 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4429 ret = ext4_expand_extra_isize(inode,
4430 sbi->s_want_extra_isize,
4431 iloc, handle);
4432 if (ret) {
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004433 ext4_set_inode_state(inode,
4434 EXT4_STATE_NO_EXPAND);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04004435 if (mnt_count !=
4436 le16_to_cpu(sbi->s_es->s_mnt_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05004437 ext4_warning(inode->i_sb,
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004438 "Unable to expand inode %lu. Delete"
4439 " some EAs or run e2fsck.",
4440 inode->i_ino);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04004441 mnt_count =
4442 le16_to_cpu(sbi->s_es->s_mnt_count);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004443 }
4444 }
4445 }
4446 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004447 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07004448 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004449 return err;
4450}
4451
4452/*
Mingming Cao617ba132006-10-11 01:20:53 -07004453 * ext4_dirty_inode() is called from __mark_inode_dirty()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004454 *
4455 * We're really interested in the case where a file is being extended.
4456 * i_size has been changed by generic_commit_write() and we thus need
4457 * to include the updated inode in the current transaction.
4458 *
Christoph Hellwig5dd40562010-03-03 09:05:00 -05004459 * Also, dquot_alloc_block() will always dirty the inode when blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004460 * are allocated to the file.
4461 *
4462 * If the inode is marked synchronous, we don't honour that here - doing
4463 * so would cause a commit on atime updates, which we don't bother doing.
4464 * We handle synchronous inodes at the highest possible level.
4465 */
Christoph Hellwigaa385722011-05-27 06:53:02 -04004466void ext4_dirty_inode(struct inode *inode, int flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004467{
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004468 handle_t *handle;
4469
Mingming Cao617ba132006-10-11 01:20:53 -07004470 handle = ext4_journal_start(inode, 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004471 if (IS_ERR(handle))
4472 goto out;
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04004473
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04004474 ext4_mark_inode_dirty(handle, inode);
4475
Mingming Cao617ba132006-10-11 01:20:53 -07004476 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004477out:
4478 return;
4479}
4480
4481#if 0
4482/*
4483 * Bind an inode's backing buffer_head into this transaction, to prevent
4484 * it from being flushed to disk early. Unlike
Mingming Cao617ba132006-10-11 01:20:53 -07004485 * ext4_reserve_inode_write, this leaves behind no bh reference and
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004486 * returns no iloc structure, so the caller needs to repeat the iloc
4487 * lookup to mark the inode dirty later.
4488 */
Mingming Cao617ba132006-10-11 01:20:53 -07004489static int ext4_pin_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004490{
Mingming Cao617ba132006-10-11 01:20:53 -07004491 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004492
4493 int err = 0;
4494 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07004495 err = ext4_get_inode_loc(inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004496 if (!err) {
4497 BUFFER_TRACE(iloc.bh, "get_write_access");
Mingming Caodab291a2006-10-11 01:21:01 -07004498 err = jbd2_journal_get_write_access(handle, iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004499 if (!err)
Frank Mayhar03901312009-01-07 00:06:22 -05004500 err = ext4_handle_dirty_metadata(handle,
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05004501 NULL,
Frank Mayhar03901312009-01-07 00:06:22 -05004502 iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004503 brelse(iloc.bh);
4504 }
4505 }
Mingming Cao617ba132006-10-11 01:20:53 -07004506 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004507 return err;
4508}
4509#endif
4510
Mingming Cao617ba132006-10-11 01:20:53 -07004511int ext4_change_inode_journal_flag(struct inode *inode, int val)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004512{
4513 journal_t *journal;
4514 handle_t *handle;
4515 int err;
4516
4517 /*
4518 * We have to be very careful here: changing a data block's
4519 * journaling status dynamically is dangerous. If we write a
4520 * data block to the journal, change the status and then delete
4521 * that block, we risk forgetting to revoke the old log record
4522 * from the journal and so a subsequent replay can corrupt data.
4523 * So, first we make sure that the journal is empty and that
4524 * nobody is changing anything.
4525 */
4526
Mingming Cao617ba132006-10-11 01:20:53 -07004527 journal = EXT4_JOURNAL(inode);
Frank Mayhar03901312009-01-07 00:06:22 -05004528 if (!journal)
4529 return 0;
Dave Hansend6995942007-07-18 08:33:51 -04004530 if (is_journal_aborted(journal))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004531 return -EROFS;
Yongqiang Yang2aff57b2011-12-28 12:02:13 -05004532 /* We have to allocate physical blocks for delalloc blocks
4533 * before flushing journal. otherwise delalloc blocks can not
4534 * be allocated any more. even more truncate on delalloc blocks
4535 * could trigger BUG by flushing delalloc blocks in journal.
4536 * There is no delalloc block in non-journal data mode.
4537 */
4538 if (val && test_opt(inode->i_sb, DELALLOC)) {
4539 err = ext4_alloc_da_blocks(inode);
4540 if (err < 0)
4541 return err;
4542 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004543
Mingming Caodab291a2006-10-11 01:21:01 -07004544 jbd2_journal_lock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004545
4546 /*
4547 * OK, there are no updates running now, and all cached data is
4548 * synced to disk. We are now in a completely consistent state
4549 * which doesn't have anything in the journal, and we know that
4550 * no filesystem updates are running, so it is safe to modify
4551 * the inode's in-core data-journaling state flag now.
4552 */
4553
4554 if (val)
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004555 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
Yongqiang Yang5872dda2011-12-28 13:55:51 -05004556 else {
4557 jbd2_journal_flush(journal);
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004558 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
Yongqiang Yang5872dda2011-12-28 13:55:51 -05004559 }
Mingming Cao617ba132006-10-11 01:20:53 -07004560 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004561
Mingming Caodab291a2006-10-11 01:21:01 -07004562 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004563
4564 /* Finally we can mark the inode as dirty. */
4565
Mingming Cao617ba132006-10-11 01:20:53 -07004566 handle = ext4_journal_start(inode, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004567 if (IS_ERR(handle))
4568 return PTR_ERR(handle);
4569
Mingming Cao617ba132006-10-11 01:20:53 -07004570 err = ext4_mark_inode_dirty(handle, inode);
Frank Mayhar03901312009-01-07 00:06:22 -05004571 ext4_handle_sync(handle);
Mingming Cao617ba132006-10-11 01:20:53 -07004572 ext4_journal_stop(handle);
4573 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004574
4575 return err;
4576}
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004577
4578static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4579{
4580 return !buffer_mapped(bh);
4581}
4582
Nick Pigginc2ec1752009-03-31 15:23:21 -07004583int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004584{
Nick Pigginc2ec1752009-03-31 15:23:21 -07004585 struct page *page = vmf->page;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004586 loff_t size;
4587 unsigned long len;
Jan Kara9ea7df52011-06-24 14:29:41 -04004588 int ret;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004589 struct file *file = vma->vm_file;
4590 struct inode *inode = file->f_path.dentry->d_inode;
4591 struct address_space *mapping = inode->i_mapping;
Jan Kara9ea7df52011-06-24 14:29:41 -04004592 handle_t *handle;
4593 get_block_t *get_block;
4594 int retries = 0;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004595
4596 /*
Jan Kara9ea7df52011-06-24 14:29:41 -04004597 * This check is racy but catches the common case. We rely on
4598 * __block_page_mkwrite() to do a reliable check.
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004599 */
Jan Kara9ea7df52011-06-24 14:29:41 -04004600 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
4601 /* Delalloc case is easy... */
4602 if (test_opt(inode->i_sb, DELALLOC) &&
4603 !ext4_should_journal_data(inode) &&
4604 !ext4_nonda_switch(inode->i_sb)) {
4605 do {
4606 ret = __block_page_mkwrite(vma, vmf,
4607 ext4_da_get_block_prep);
4608 } while (ret == -ENOSPC &&
4609 ext4_should_retry_alloc(inode->i_sb, &retries));
4610 goto out_ret;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004611 }
Darrick J. Wong0e499892011-05-18 13:55:20 -04004612
4613 lock_page(page);
Jan Kara9ea7df52011-06-24 14:29:41 -04004614 size = i_size_read(inode);
4615 /* Page got truncated from under us? */
4616 if (page->mapping != mapping || page_offset(page) > size) {
4617 unlock_page(page);
4618 ret = VM_FAULT_NOPAGE;
4619 goto out;
Darrick J. Wong0e499892011-05-18 13:55:20 -04004620 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004621
4622 if (page->index == size >> PAGE_CACHE_SHIFT)
4623 len = size & ~PAGE_CACHE_MASK;
4624 else
4625 len = PAGE_CACHE_SIZE;
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04004626 /*
Jan Kara9ea7df52011-06-24 14:29:41 -04004627 * Return if we have all the buffers mapped. This avoids the need to do
4628 * journal_start/journal_stop which can block and take a long time
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04004629 */
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004630 if (page_has_buffers(page)) {
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004631 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04004632 ext4_bh_unmapped)) {
Jan Kara9ea7df52011-06-24 14:29:41 -04004633 /* Wait so that we don't change page under IO */
4634 wait_on_page_writeback(page);
4635 ret = VM_FAULT_LOCKED;
4636 goto out;
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04004637 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004638 }
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04004639 unlock_page(page);
Jan Kara9ea7df52011-06-24 14:29:41 -04004640 /* OK, we need to fill the hole... */
4641 if (ext4_should_dioread_nolock(inode))
4642 get_block = ext4_get_block_write;
4643 else
4644 get_block = ext4_get_block;
4645retry_alloc:
4646 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4647 if (IS_ERR(handle)) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07004648 ret = VM_FAULT_SIGBUS;
Jan Kara9ea7df52011-06-24 14:29:41 -04004649 goto out;
4650 }
4651 ret = __block_page_mkwrite(vma, vmf, get_block);
4652 if (!ret && ext4_should_journal_data(inode)) {
4653 if (walk_page_buffers(handle, page_buffers(page), 0,
4654 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4655 unlock_page(page);
4656 ret = VM_FAULT_SIGBUS;
Yongqiang Yangfcbb5512011-10-26 05:00:19 -04004657 ext4_journal_stop(handle);
Jan Kara9ea7df52011-06-24 14:29:41 -04004658 goto out;
4659 }
4660 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4661 }
4662 ext4_journal_stop(handle);
4663 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4664 goto retry_alloc;
4665out_ret:
4666 ret = block_page_mkwrite_return(ret);
4667out:
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004668 return ret;
4669}