Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * fs/fs-writeback.c |
| 3 | * |
| 4 | * Copyright (C) 2002, Linus Torvalds. |
| 5 | * |
| 6 | * Contains all the functions related to writing back and waiting |
| 7 | * upon dirty inodes against superblocks, and writing back dirty |
| 8 | * pages against inodes. ie: data writeback. Writeout of the |
| 9 | * inode itself is not handled here. |
| 10 | * |
Francois Cami | e1f8e87 | 2008-10-15 22:01:59 -0700 | [diff] [blame] | 11 | * 10Apr2002 Andrew Morton |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 12 | * Split out of fs/inode.c |
| 13 | * Additions for address_space-based writeback |
| 14 | */ |
| 15 | |
| 16 | #include <linux/kernel.h> |
Jens Axboe | f5ff842 | 2007-09-21 09:19:54 +0200 | [diff] [blame] | 17 | #include <linux/module.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include <linux/spinlock.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 19 | #include <linux/slab.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | #include <linux/sched.h> |
| 21 | #include <linux/fs.h> |
| 22 | #include <linux/mm.h> |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 23 | #include <linux/kthread.h> |
| 24 | #include <linux/freezer.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include <linux/writeback.h> |
| 26 | #include <linux/blkdev.h> |
| 27 | #include <linux/backing-dev.h> |
| 28 | #include <linux/buffer_head.h> |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 29 | #include <linux/tracepoint.h> |
David Howells | 07f3f05c | 2006-09-30 20:52:18 +0200 | [diff] [blame] | 30 | #include "internal.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 31 | |
Jens Axboe | d0bceac | 2009-05-18 08:20:32 +0200 | [diff] [blame] | 32 | /* |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 33 | * Passed into wb_writeback(), essentially a subset of writeback_control |
| 34 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 35 | struct wb_writeback_work { |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 36 | long nr_pages; |
| 37 | struct super_block *sb; |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 38 | unsigned long *older_than_this; |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 39 | enum writeback_sync_modes sync_mode; |
Wu Fengguang | 6e6938b | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 40 | unsigned int tagged_writepages:1; |
H Hartley Sweeten | 52957fe | 2010-04-01 20:36:30 -0500 | [diff] [blame] | 41 | unsigned int for_kupdate:1; |
| 42 | unsigned int range_cyclic:1; |
| 43 | unsigned int for_background:1; |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 44 | |
Jens Axboe | 8010c3b | 2009-09-15 20:04:57 +0200 | [diff] [blame] | 45 | struct list_head list; /* pending work list */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 46 | struct completion *done; /* set if the caller waits */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 47 | }; |
| 48 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 49 | /* |
| 50 | * Include the creation of the trace points after defining the |
| 51 | * wb_writeback_work structure so that the definition remains local to this |
| 52 | * file. |
| 53 | */ |
| 54 | #define CREATE_TRACE_POINTS |
| 55 | #include <trace/events/writeback.h> |
| 56 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 57 | /* |
| 58 | * We don't actually have pdflush, but this one is exported though /proc... |
| 59 | */ |
| 60 | int nr_pdflush_threads; |
| 61 | |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 62 | /** |
| 63 | * writeback_in_progress - determine whether there is writeback in progress |
| 64 | * @bdi: the device's backing_dev_info structure. |
| 65 | * |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 66 | * Determine whether there is writeback waiting to be handled against a |
| 67 | * backing device. |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 68 | */ |
| 69 | int writeback_in_progress(struct backing_dev_info *bdi) |
| 70 | { |
Jan Kara | 81d73a3 | 2010-08-11 14:17:44 -0700 | [diff] [blame] | 71 | return test_bit(BDI_writeback_running, &bdi->state); |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 72 | } |
| 73 | |
Jan Kara | 692ebd1 | 2010-09-21 11:51:01 +0200 | [diff] [blame] | 74 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
| 75 | { |
| 76 | struct super_block *sb = inode->i_sb; |
Jan Kara | 692ebd1 | 2010-09-21 11:51:01 +0200 | [diff] [blame] | 77 | |
Christoph Hellwig | aaead25 | 2010-10-04 14:25:33 +0200 | [diff] [blame] | 78 | if (strcmp(sb->s_type->name, "bdev") == 0) |
| 79 | return inode->i_mapping->backing_dev_info; |
| 80 | |
| 81 | return sb->s_bdi; |
Jan Kara | 692ebd1 | 2010-09-21 11:51:01 +0200 | [diff] [blame] | 82 | } |
| 83 | |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 84 | static inline struct inode *wb_inode(struct list_head *head) |
| 85 | { |
| 86 | return list_entry(head, struct inode, i_wb_list); |
| 87 | } |
| 88 | |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 89 | /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ |
| 90 | static void bdi_wakeup_flusher(struct backing_dev_info *bdi) |
Nick Piggin | 4195f73 | 2009-05-28 09:01:15 +0200 | [diff] [blame] | 91 | { |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 92 | if (bdi->wb.task) { |
| 93 | wake_up_process(bdi->wb.task); |
| 94 | } else { |
| 95 | /* |
| 96 | * The bdi thread isn't there, wake up the forker thread which |
| 97 | * will create and run it. |
| 98 | */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 99 | wake_up_process(default_backing_dev_info.wb.task); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 100 | } |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 101 | } |
| 102 | |
| 103 | static void bdi_queue_work(struct backing_dev_info *bdi, |
| 104 | struct wb_writeback_work *work) |
| 105 | { |
| 106 | trace_writeback_queue(bdi, work); |
| 107 | |
| 108 | spin_lock_bh(&bdi->wb_lock); |
| 109 | list_add_tail(&work->list, &bdi->work_list); |
| 110 | if (!bdi->wb.task) |
| 111 | trace_writeback_nothread(bdi, work); |
| 112 | bdi_wakeup_flusher(bdi); |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 113 | spin_unlock_bh(&bdi->wb_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 114 | } |
| 115 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 116 | static void |
| 117 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 118 | bool range_cyclic) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 119 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 120 | struct wb_writeback_work *work; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 121 | |
Jens Axboe | bcddc3f | 2009-09-13 20:07:36 +0200 | [diff] [blame] | 122 | /* |
| 123 | * This is WB_SYNC_NONE writeback, so if allocation fails just |
| 124 | * wakeup the thread for old dirty data writeback |
| 125 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 126 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
| 127 | if (!work) { |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 128 | if (bdi->wb.task) { |
| 129 | trace_writeback_nowork(bdi); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 130 | wake_up_process(bdi->wb.task); |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 131 | } |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 132 | return; |
Jens Axboe | bcddc3f | 2009-09-13 20:07:36 +0200 | [diff] [blame] | 133 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 134 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 135 | work->sync_mode = WB_SYNC_NONE; |
| 136 | work->nr_pages = nr_pages; |
| 137 | work->range_cyclic = range_cyclic; |
Christoph Hellwig | f0fad8a | 2009-09-11 09:47:56 +0200 | [diff] [blame] | 138 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 139 | bdi_queue_work(bdi, work); |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 140 | } |
| 141 | |
| 142 | /** |
| 143 | * bdi_start_writeback - start writeback |
| 144 | * @bdi: the backing device to write from |
| 145 | * @nr_pages: the number of pages to write |
| 146 | * |
| 147 | * Description: |
| 148 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 149 | * started when this function returns, we make no guarantees on |
Jens Axboe | 0e3c9a2 | 2010-06-01 11:08:43 +0200 | [diff] [blame] | 150 | * completion. Caller need not hold sb s_umount semaphore. |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 151 | * |
| 152 | */ |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 153 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 154 | { |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 155 | __bdi_start_writeback(bdi, nr_pages, true); |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 156 | } |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 157 | |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 158 | /** |
| 159 | * bdi_start_background_writeback - start background writeback |
| 160 | * @bdi: the backing device to write from |
| 161 | * |
| 162 | * Description: |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 163 | * This makes sure WB_SYNC_NONE background writeback happens. When |
| 164 | * this function returns, it is only guaranteed that for given BDI |
| 165 | * some IO is happening if we are over background dirty threshold. |
| 166 | * Caller need not hold sb s_umount semaphore. |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 167 | */ |
| 168 | void bdi_start_background_writeback(struct backing_dev_info *bdi) |
| 169 | { |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 170 | /* |
| 171 | * We just wake up the flusher thread. It will perform background |
| 172 | * writeback as soon as there is no other work to do. |
| 173 | */ |
Wu Fengguang | 71927e8 | 2011-01-13 15:45:46 -0800 | [diff] [blame] | 174 | trace_writeback_wake_background(bdi); |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 175 | spin_lock_bh(&bdi->wb_lock); |
| 176 | bdi_wakeup_flusher(bdi); |
| 177 | spin_unlock_bh(&bdi->wb_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 178 | } |
| 179 | |
| 180 | /* |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 181 | * Remove the inode from the writeback list it is on. |
| 182 | */ |
| 183 | void inode_wb_list_del(struct inode *inode) |
| 184 | { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 185 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 186 | |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 187 | spin_lock(&bdi->wb.list_lock); |
| 188 | list_del_init(&inode->i_wb_list); |
| 189 | spin_unlock(&bdi->wb.list_lock); |
| 190 | } |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 191 | |
| 192 | /* |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 193 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the |
| 194 | * furthest end of its superblock's dirty-inode list. |
| 195 | * |
| 196 | * Before stamping the inode's ->dirtied_when, we check to see whether it is |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 197 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 198 | * the case then the inode must have been redirtied while it was being written |
| 199 | * out and we don't reset its dirtied_when. |
| 200 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 201 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 202 | { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 203 | assert_spin_locked(&wb->list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 204 | if (!list_empty(&wb->b_dirty)) { |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 205 | struct inode *tail; |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 206 | |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 207 | tail = wb_inode(wb->b_dirty.next); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 208 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 209 | inode->dirtied_when = jiffies; |
| 210 | } |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 211 | list_move(&inode->i_wb_list, &wb->b_dirty); |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 212 | } |
| 213 | |
| 214 | /* |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 215 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 216 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 217 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 218 | { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 219 | assert_spin_locked(&wb->list_lock); |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 220 | list_move(&inode->i_wb_list, &wb->b_more_io); |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 221 | } |
| 222 | |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 223 | static void inode_sync_complete(struct inode *inode) |
| 224 | { |
| 225 | /* |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 226 | * Prevent speculative execution through |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 227 | * spin_unlock(&wb->list_lock); |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 228 | */ |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 229 | |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 230 | smp_mb(); |
| 231 | wake_up_bit(&inode->i_state, __I_SYNC); |
| 232 | } |
| 233 | |
Jeff Layton | d2caa3c5 | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 234 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
| 235 | { |
| 236 | bool ret = time_after(inode->dirtied_when, t); |
| 237 | #ifndef CONFIG_64BIT |
| 238 | /* |
| 239 | * For inodes being constantly redirtied, dirtied_when can get stuck. |
| 240 | * It _appears_ to be in the future, but is actually in distant past. |
| 241 | * This test is necessary to prevent such wrapped-around relative times |
Jens Axboe | 5b0830c | 2009-09-23 19:37:09 +0200 | [diff] [blame] | 242 | * from permanently stopping the whole bdi writeback. |
Jeff Layton | d2caa3c5 | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 243 | */ |
| 244 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); |
| 245 | #endif |
| 246 | return ret; |
| 247 | } |
| 248 | |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 249 | /* |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 250 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. |
| 251 | */ |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 252 | static int move_expired_inodes(struct list_head *delaying_queue, |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 253 | struct list_head *dispatch_queue, |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 254 | unsigned long *older_than_this) |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 255 | { |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 256 | LIST_HEAD(tmp); |
| 257 | struct list_head *pos, *node; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 258 | struct super_block *sb = NULL; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 259 | struct inode *inode; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 260 | int do_sb_sort = 0; |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 261 | int moved = 0; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 262 | |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 263 | while (!list_empty(delaying_queue)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 264 | inode = wb_inode(delaying_queue->prev); |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 265 | if (older_than_this && |
Jeff Layton | d2caa3c5 | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 266 | inode_dirtied_after(inode, *older_than_this)) |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 267 | break; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 268 | if (sb && sb != inode->i_sb) |
| 269 | do_sb_sort = 1; |
| 270 | sb = inode->i_sb; |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 271 | list_move(&inode->i_wb_list, &tmp); |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 272 | moved++; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 273 | } |
| 274 | |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 275 | /* just one sb in list, splice to dispatch_queue and we're done */ |
| 276 | if (!do_sb_sort) { |
| 277 | list_splice(&tmp, dispatch_queue); |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 278 | goto out; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 279 | } |
| 280 | |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 281 | /* Move inodes from one superblock together */ |
| 282 | while (!list_empty(&tmp)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 283 | sb = wb_inode(tmp.prev)->i_sb; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 284 | list_for_each_prev_safe(pos, node, &tmp) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 285 | inode = wb_inode(pos); |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 286 | if (inode->i_sb == sb) |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 287 | list_move(&inode->i_wb_list, dispatch_queue); |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 288 | } |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 289 | } |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 290 | out: |
| 291 | return moved; |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 292 | } |
| 293 | |
| 294 | /* |
| 295 | * Queue all expired dirty inodes for io, eldest first. |
Wu Fengguang | 4ea879b | 2010-08-11 14:17:42 -0700 | [diff] [blame] | 296 | * Before |
| 297 | * newly dirtied b_dirty b_io b_more_io |
| 298 | * =============> gf edc BA |
| 299 | * After |
| 300 | * newly dirtied b_dirty b_io b_more_io |
| 301 | * =============> g fBAedc |
| 302 | * | |
| 303 | * +--> dequeue for IO |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 304 | */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 305 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 306 | { |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 307 | int moved; |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 308 | assert_spin_locked(&wb->list_lock); |
Wu Fengguang | 4ea879b | 2010-08-11 14:17:42 -0700 | [diff] [blame] | 309 | list_splice_init(&wb->b_more_io, &wb->b_io); |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 310 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); |
| 311 | trace_writeback_queue_io(wb, older_than_this, moved); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 312 | } |
| 313 | |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 314 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 315 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 316 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 317 | return inode->i_sb->s_op->write_inode(inode, wbc); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 318 | return 0; |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 319 | } |
| 320 | |
| 321 | /* |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 322 | * Wait for writeback on an inode to complete. |
| 323 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 324 | static void inode_wait_for_writeback(struct inode *inode, |
| 325 | struct bdi_writeback *wb) |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 326 | { |
| 327 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); |
| 328 | wait_queue_head_t *wqh; |
| 329 | |
| 330 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 331 | while (inode->i_state & I_SYNC) { |
| 332 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 333 | spin_unlock(&wb->list_lock); |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 334 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 335 | spin_lock(&wb->list_lock); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 336 | spin_lock(&inode->i_lock); |
Richard Kennedy | 58a9d3d8 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 337 | } |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 338 | } |
| 339 | |
| 340 | /* |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 341 | * Write out an inode's dirty pages. Called under wb->list_lock and |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 342 | * inode->i_lock. Either the caller has an active reference on the inode or |
| 343 | * the inode has I_WILL_FREE set. |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 344 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 345 | * If `wait' is set, wait on the writeout. |
| 346 | * |
| 347 | * The whole writeout design is quite complex and fragile. We want to avoid |
| 348 | * starvation of particular inodes when others are being redirtied, prevent |
| 349 | * livelocks, etc. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 350 | */ |
| 351 | static int |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 352 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, |
| 353 | struct writeback_control *wbc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 354 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 355 | struct address_space *mapping = inode->i_mapping; |
Wu Fengguang | 251d6a4 | 2010-12-01 17:33:37 -0600 | [diff] [blame] | 356 | long nr_to_write = wbc->nr_to_write; |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 357 | unsigned dirty; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 358 | int ret; |
| 359 | |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 360 | assert_spin_locked(&wb->list_lock); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 361 | assert_spin_locked(&inode->i_lock); |
| 362 | |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 363 | if (!atomic_read(&inode->i_count)) |
| 364 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); |
| 365 | else |
| 366 | WARN_ON(inode->i_state & I_WILL_FREE); |
| 367 | |
| 368 | if (inode->i_state & I_SYNC) { |
| 369 | /* |
| 370 | * If this inode is locked for writeback and we are not doing |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 371 | * writeback-for-data-integrity, move it to b_more_io so that |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 372 | * writeback can proceed with the other inodes on s_io. |
| 373 | * |
| 374 | * We'll have another go at writing back this inode when we |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 375 | * completed a full scan of b_io. |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 376 | */ |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 377 | if (wbc->sync_mode != WB_SYNC_ALL) { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 378 | requeue_io(inode, wb); |
Wu Fengguang | 251d6a4 | 2010-12-01 17:33:37 -0600 | [diff] [blame] | 379 | trace_writeback_single_inode_requeue(inode, wbc, |
| 380 | nr_to_write); |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 381 | return 0; |
| 382 | } |
| 383 | |
| 384 | /* |
| 385 | * It's a data-integrity sync. We must wait. |
| 386 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 387 | inode_wait_for_writeback(inode, wb); |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 388 | } |
| 389 | |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 390 | BUG_ON(inode->i_state & I_SYNC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 391 | |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 392 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 393 | inode->i_state |= I_SYNC; |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 394 | inode->i_state &= ~I_DIRTY_PAGES; |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 395 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 396 | spin_unlock(&wb->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 397 | |
| 398 | ret = do_writepages(mapping, wbc); |
| 399 | |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 400 | /* |
| 401 | * Make sure to wait on the data before writing out the metadata. |
| 402 | * This is important for filesystems that modify metadata on data |
| 403 | * I/O completion. |
| 404 | */ |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 405 | if (wbc->sync_mode == WB_SYNC_ALL) { |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 406 | int err = filemap_fdatawait(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 407 | if (ret == 0) |
| 408 | ret = err; |
| 409 | } |
| 410 | |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 411 | /* |
| 412 | * Some filesystems may redirty the inode during the writeback |
| 413 | * due to delalloc, clear dirty metadata flags right before |
| 414 | * write_inode() |
| 415 | */ |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 416 | spin_lock(&inode->i_lock); |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 417 | dirty = inode->i_state & I_DIRTY; |
| 418 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 419 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 420 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
| 421 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 422 | int err = write_inode(inode, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 423 | if (ret == 0) |
| 424 | ret = err; |
| 425 | } |
| 426 | |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 427 | spin_lock(&wb->list_lock); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 428 | spin_lock(&inode->i_lock); |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 429 | inode->i_state &= ~I_SYNC; |
Al Viro | a4ffdde | 2010-06-02 17:38:30 -0400 | [diff] [blame] | 430 | if (!(inode->i_state & I_FREEING)) { |
Wu Fengguang | 94c3dcb | 2011-04-27 19:05:21 -0600 | [diff] [blame] | 431 | /* |
| 432 | * Sync livelock prevention. Each inode is tagged and synced in |
| 433 | * one shot. If still dirty, it will be redirty_tail()'ed below. |
| 434 | * Update the dirty time to prevent enqueue and sync it again. |
| 435 | */ |
| 436 | if ((inode->i_state & I_DIRTY) && |
| 437 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) |
| 438 | inode->dirtied_when = jiffies; |
| 439 | |
Wu Fengguang | 23539af | 2010-08-11 14:17:41 -0700 | [diff] [blame] | 440 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 441 | /* |
| 442 | * We didn't write back all the pages. nfs_writepages() |
Wu Fengguang | a50aeb4 | 2010-08-11 14:17:43 -0700 | [diff] [blame] | 443 | * sometimes bales out without doing anything. |
Andrew Morton | 1b43ef9 | 2007-10-16 23:30:35 -0700 | [diff] [blame] | 444 | */ |
Wu Fengguang | a50aeb4 | 2010-08-11 14:17:43 -0700 | [diff] [blame] | 445 | inode->i_state |= I_DIRTY_PAGES; |
| 446 | if (wbc->nr_to_write <= 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 447 | /* |
Wu Fengguang | a50aeb4 | 2010-08-11 14:17:43 -0700 | [diff] [blame] | 448 | * slice used up: queue for next turn |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 449 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 450 | requeue_io(inode, wb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 451 | } else { |
| 452 | /* |
Wu Fengguang | a50aeb4 | 2010-08-11 14:17:43 -0700 | [diff] [blame] | 453 | * Writeback blocked by something other than |
| 454 | * congestion. Delay the inode for some time to |
| 455 | * avoid spinning on the CPU (100% iowait) |
| 456 | * retrying writeback of the dirty page/inode |
| 457 | * that cannot be performed immediately. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 458 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 459 | redirty_tail(inode, wb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 460 | } |
Wu Fengguang | 23539af | 2010-08-11 14:17:41 -0700 | [diff] [blame] | 461 | } else if (inode->i_state & I_DIRTY) { |
| 462 | /* |
| 463 | * Filesystems can dirty the inode during writeback |
| 464 | * operations, such as delayed allocation during |
| 465 | * submission or metadata updates after data IO |
| 466 | * completion. |
| 467 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 468 | redirty_tail(inode, wb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 469 | } else { |
| 470 | /* |
Nick Piggin | 9e38d86 | 2010-10-23 06:55:17 -0400 | [diff] [blame] | 471 | * The inode is clean. At this point we either have |
| 472 | * a reference to the inode or it's on it's way out. |
| 473 | * No need to add it back to the LRU. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 474 | */ |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 475 | list_del_init(&inode->i_wb_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 476 | } |
| 477 | } |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 478 | inode_sync_complete(inode); |
Wu Fengguang | 251d6a4 | 2010-12-01 17:33:37 -0600 | [diff] [blame] | 479 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 480 | return ret; |
| 481 | } |
| 482 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 483 | /* |
Christoph Hellwig | d19de7e | 2010-06-08 18:14:58 +0200 | [diff] [blame] | 484 | * For background writeback the caller does not have the sb pinned |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 485 | * before calling writeback. So make sure that we do pin it, so it doesn't |
| 486 | * go away while we are writing inodes from it. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 487 | */ |
Christoph Hellwig | d19de7e | 2010-06-08 18:14:58 +0200 | [diff] [blame] | 488 | static bool pin_sb_for_writeback(struct super_block *sb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 489 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 490 | spin_lock(&sb_lock); |
Christoph Hellwig | 29cb4859 | 2010-06-09 15:31:01 +0200 | [diff] [blame] | 491 | if (list_empty(&sb->s_instances)) { |
| 492 | spin_unlock(&sb_lock); |
| 493 | return false; |
| 494 | } |
| 495 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 496 | sb->s_count++; |
Christoph Hellwig | 29cb4859 | 2010-06-09 15:31:01 +0200 | [diff] [blame] | 497 | spin_unlock(&sb_lock); |
| 498 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 499 | if (down_read_trylock(&sb->s_umount)) { |
Christoph Hellwig | 29cb4859 | 2010-06-09 15:31:01 +0200 | [diff] [blame] | 500 | if (sb->s_root) |
Christoph Hellwig | d19de7e | 2010-06-08 18:14:58 +0200 | [diff] [blame] | 501 | return true; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 502 | up_read(&sb->s_umount); |
| 503 | } |
Christoph Hellwig | 29cb4859 | 2010-06-09 15:31:01 +0200 | [diff] [blame] | 504 | |
| 505 | put_super(sb); |
Christoph Hellwig | d19de7e | 2010-06-08 18:14:58 +0200 | [diff] [blame] | 506 | return false; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 507 | } |
| 508 | |
Wu Fengguang | 1a12d8b | 2010-08-29 13:28:09 -0600 | [diff] [blame^] | 509 | static long writeback_chunk_size(struct backing_dev_info *bdi, |
| 510 | struct wb_writeback_work *work) |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 511 | { |
| 512 | long pages; |
| 513 | |
| 514 | /* |
| 515 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty |
| 516 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX |
| 517 | * here avoids calling into writeback_inodes_wb() more than once. |
| 518 | * |
| 519 | * The intended call sequence for WB_SYNC_ALL writeback is: |
| 520 | * |
| 521 | * wb_writeback() |
| 522 | * writeback_sb_inodes() <== called only once |
| 523 | * write_cache_pages() <== called once for each inode |
| 524 | * (quickly) tag currently dirty pages |
| 525 | * (maybe slowly) sync all tagged pages |
| 526 | */ |
| 527 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) |
| 528 | pages = LONG_MAX; |
Wu Fengguang | 1a12d8b | 2010-08-29 13:28:09 -0600 | [diff] [blame^] | 529 | else { |
| 530 | pages = min(bdi->avg_write_bandwidth / 2, |
| 531 | global_dirty_limit / DIRTY_SCOPE); |
| 532 | pages = min(pages, work->nr_pages); |
| 533 | pages = round_down(pages + MIN_WRITEBACK_PAGES, |
| 534 | MIN_WRITEBACK_PAGES); |
| 535 | } |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 536 | |
| 537 | return pages; |
| 538 | } |
| 539 | |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 540 | /* |
| 541 | * Write a portion of b_io inodes which belong to @sb. |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 542 | * |
| 543 | * If @only_this_sb is true, then find and write all such |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 544 | * inodes. Otherwise write only ones which go sequentially |
| 545 | * in reverse order. |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 546 | * |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 547 | * Return the number of pages and/or inodes written. |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 548 | */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 549 | static long writeback_sb_inodes(struct super_block *sb, |
| 550 | struct bdi_writeback *wb, |
| 551 | struct wb_writeback_work *work) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 552 | { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 553 | struct writeback_control wbc = { |
| 554 | .sync_mode = work->sync_mode, |
| 555 | .tagged_writepages = work->tagged_writepages, |
| 556 | .for_kupdate = work->for_kupdate, |
| 557 | .for_background = work->for_background, |
| 558 | .range_cyclic = work->range_cyclic, |
| 559 | .range_start = 0, |
| 560 | .range_end = LLONG_MAX, |
| 561 | }; |
| 562 | unsigned long start_time = jiffies; |
| 563 | long write_chunk; |
| 564 | long wrote = 0; /* count both pages and inodes */ |
| 565 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 566 | while (!list_empty(&wb->b_io)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 567 | struct inode *inode = wb_inode(wb->b_io.prev); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 568 | |
| 569 | if (inode->i_sb != sb) { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 570 | if (work->sb) { |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 571 | /* |
| 572 | * We only want to write back data for this |
| 573 | * superblock, move all inodes not belonging |
| 574 | * to it back onto the dirty list. |
| 575 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 576 | redirty_tail(inode, wb); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 577 | continue; |
| 578 | } |
| 579 | |
| 580 | /* |
| 581 | * The inode belongs to a different superblock. |
| 582 | * Bounce back to the caller to unpin this and |
| 583 | * pin the next superblock. |
| 584 | */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 585 | break; |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 586 | } |
| 587 | |
Christoph Hellwig | 9843b76 | 2010-10-24 19:40:46 +0200 | [diff] [blame] | 588 | /* |
| 589 | * Don't bother with new inodes or inodes beeing freed, first |
| 590 | * kind does not need peridic writeout yet, and for the latter |
| 591 | * kind writeout is handled by the freer. |
| 592 | */ |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 593 | spin_lock(&inode->i_lock); |
Christoph Hellwig | 9843b76 | 2010-10-24 19:40:46 +0200 | [diff] [blame] | 594 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 595 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 596 | requeue_io(inode, wb); |
Nick Piggin | 7ef0d73 | 2009-03-12 14:31:38 -0700 | [diff] [blame] | 597 | continue; |
| 598 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 599 | __iget(inode); |
Wu Fengguang | 1a12d8b | 2010-08-29 13:28:09 -0600 | [diff] [blame^] | 600 | write_chunk = writeback_chunk_size(wb->bdi, work); |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 601 | wbc.nr_to_write = write_chunk; |
| 602 | wbc.pages_skipped = 0; |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 603 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 604 | writeback_single_inode(inode, wb, &wbc); |
| 605 | |
| 606 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
| 607 | wrote += write_chunk - wbc.nr_to_write; |
| 608 | if (!(inode->i_state & I_DIRTY)) |
| 609 | wrote++; |
| 610 | if (wbc.pages_skipped) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 611 | /* |
| 612 | * writeback is not making progress due to locked |
| 613 | * buffers. Skip this inode for now. |
| 614 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 615 | redirty_tail(inode, wb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 616 | } |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 617 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 618 | spin_unlock(&wb->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 619 | iput(inode); |
OGAWA Hirofumi | 4ffc844 | 2006-03-25 03:07:44 -0800 | [diff] [blame] | 620 | cond_resched(); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 621 | spin_lock(&wb->list_lock); |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 622 | /* |
| 623 | * bail out to wb_writeback() often enough to check |
| 624 | * background threshold and other termination conditions. |
| 625 | */ |
| 626 | if (wrote) { |
| 627 | if (time_is_before_jiffies(start_time + HZ / 10UL)) |
| 628 | break; |
| 629 | if (work->nr_pages <= 0) |
| 630 | break; |
| 631 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 632 | } |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 633 | return wrote; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 634 | } |
Nick Piggin | 38f2197 | 2009-01-06 14:40:25 -0800 | [diff] [blame] | 635 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 636 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
| 637 | struct wb_writeback_work *work) |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 638 | { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 639 | unsigned long start_time = jiffies; |
| 640 | long wrote = 0; |
Jens Axboe | 9ecc273 | 2009-09-24 15:25:11 +0200 | [diff] [blame] | 641 | |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 642 | while (!list_empty(&wb->b_io)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 643 | struct inode *inode = wb_inode(wb->b_io.prev); |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 644 | struct super_block *sb = inode->i_sb; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 645 | |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 646 | if (!pin_sb_for_writeback(sb)) { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 647 | requeue_io(inode, wb); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 648 | continue; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 649 | } |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 650 | wrote += writeback_sb_inodes(sb, wb, work); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 651 | drop_super(sb); |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 652 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 653 | /* refer to the same tests at the end of writeback_sb_inodes */ |
| 654 | if (wrote) { |
| 655 | if (time_is_before_jiffies(start_time + HZ / 10UL)) |
| 656 | break; |
| 657 | if (work->nr_pages <= 0) |
| 658 | break; |
| 659 | } |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 660 | } |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 661 | /* Leave any unwritten inodes on b_io */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 662 | return wrote; |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 663 | } |
| 664 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 665 | long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages) |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 666 | { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 667 | struct wb_writeback_work work = { |
| 668 | .nr_pages = nr_pages, |
| 669 | .sync_mode = WB_SYNC_NONE, |
| 670 | .range_cyclic = 1, |
| 671 | }; |
| 672 | |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 673 | spin_lock(&wb->list_lock); |
Wu Fengguang | 424b351 | 2010-07-21 20:11:53 -0600 | [diff] [blame] | 674 | if (list_empty(&wb->b_io)) |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 675 | queue_io(wb, NULL); |
| 676 | __writeback_inodes_wb(wb, &work); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 677 | spin_unlock(&wb->list_lock); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 678 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 679 | return nr_pages - work.nr_pages; |
| 680 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 681 | |
| 682 | static inline bool over_bground_thresh(void) |
| 683 | { |
| 684 | unsigned long background_thresh, dirty_thresh; |
| 685 | |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 686 | global_dirty_limits(&background_thresh, &dirty_thresh); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 687 | |
| 688 | return (global_page_state(NR_FILE_DIRTY) + |
Wu Fengguang | 4cbec4c | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 689 | global_page_state(NR_UNSTABLE_NFS) > background_thresh); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 690 | } |
| 691 | |
| 692 | /* |
Wu Fengguang | e98be2d | 2010-08-29 11:22:30 -0600 | [diff] [blame] | 693 | * Called under wb->list_lock. If there are multiple wb per bdi, |
| 694 | * only the flusher working on the first wb should do it. |
| 695 | */ |
| 696 | static void wb_update_bandwidth(struct bdi_writeback *wb, |
| 697 | unsigned long start_time) |
| 698 | { |
Wu Fengguang | c42843f | 2011-03-02 15:54:09 -0600 | [diff] [blame] | 699 | __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, start_time); |
Wu Fengguang | e98be2d | 2010-08-29 11:22:30 -0600 | [diff] [blame] | 700 | } |
| 701 | |
| 702 | /* |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 703 | * Explicit flushing or periodic writeback of "old" data. |
| 704 | * |
| 705 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
| 706 | * dirtying-time in the inode's address_space. So this periodic writeback code |
| 707 | * just walks the superblock inode list, writing back any inodes which are |
| 708 | * older than a specific point in time. |
| 709 | * |
| 710 | * Try to run once per dirty_writeback_interval. But if a writeback event |
| 711 | * takes longer than a dirty_writeback_interval interval, then leave a |
| 712 | * one-second gap. |
| 713 | * |
| 714 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
| 715 | * all dirty pages if they are all attached to "old" mappings. |
| 716 | */ |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 717 | static long wb_writeback(struct bdi_writeback *wb, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 718 | struct wb_writeback_work *work) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 719 | { |
Wu Fengguang | e98be2d | 2010-08-29 11:22:30 -0600 | [diff] [blame] | 720 | unsigned long wb_start = jiffies; |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 721 | long nr_pages = work->nr_pages; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 722 | unsigned long oldest_jif; |
Jan Kara | a5989bd | 2009-09-16 19:22:48 +0200 | [diff] [blame] | 723 | struct inode *inode; |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 724 | long progress; |
Jan Kara | b9543da | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 725 | |
Wu Fengguang | e185dda | 2011-04-23 11:26:07 -0600 | [diff] [blame] | 726 | oldest_jif = jiffies; |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 727 | work->older_than_this = &oldest_jif; |
Wu Fengguang | e185dda | 2011-04-23 11:26:07 -0600 | [diff] [blame] | 728 | |
Wu Fengguang | e8dfc30 | 2011-04-21 12:06:32 -0600 | [diff] [blame] | 729 | spin_lock(&wb->list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 730 | for (;;) { |
| 731 | /* |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 732 | * Stop writeback when nr_pages has been consumed |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 733 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 734 | if (work->nr_pages <= 0) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 735 | break; |
| 736 | |
| 737 | /* |
Jan Kara | aa373cf | 2011-01-13 15:45:47 -0800 | [diff] [blame] | 738 | * Background writeout and kupdate-style writeback may |
| 739 | * run forever. Stop them if there is other work to do |
| 740 | * so that e.g. sync can proceed. They'll be restarted |
| 741 | * after the other works are all done. |
| 742 | */ |
| 743 | if ((work->for_background || work->for_kupdate) && |
| 744 | !list_empty(&wb->bdi->work_list)) |
| 745 | break; |
| 746 | |
| 747 | /* |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 748 | * For background writeout, stop when we are below the |
| 749 | * background dirty threshold |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 750 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 751 | if (work->for_background && !over_bground_thresh()) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 752 | break; |
| 753 | |
Wu Fengguang | ba9aa83 | 2010-07-21 20:32:30 -0600 | [diff] [blame] | 754 | if (work->for_kupdate) { |
| 755 | oldest_jif = jiffies - |
| 756 | msecs_to_jiffies(dirty_expire_interval * 10); |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 757 | work->older_than_this = &oldest_jif; |
Wu Fengguang | ba9aa83 | 2010-07-21 20:32:30 -0600 | [diff] [blame] | 758 | } |
| 759 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 760 | trace_writeback_start(wb->bdi, work); |
Wu Fengguang | e8dfc30 | 2011-04-21 12:06:32 -0600 | [diff] [blame] | 761 | if (list_empty(&wb->b_io)) |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 762 | queue_io(wb, work->older_than_this); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 763 | if (work->sb) |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 764 | progress = writeback_sb_inodes(work->sb, wb, work); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 765 | else |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 766 | progress = __writeback_inodes_wb(wb, work); |
| 767 | trace_writeback_written(wb->bdi, work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 768 | |
Wu Fengguang | e98be2d | 2010-08-29 11:22:30 -0600 | [diff] [blame] | 769 | wb_update_bandwidth(wb, wb_start); |
| 770 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 771 | /* |
Wu Fengguang | e6fb6da | 2010-07-22 10:23:44 -0600 | [diff] [blame] | 772 | * Did we write something? Try for more |
| 773 | * |
| 774 | * Dirty inodes are moved to b_io for writeback in batches. |
| 775 | * The completion of the current batch does not necessarily |
| 776 | * mean the overall work is done. So we keep looping as long |
| 777 | * as made some progress on cleaning pages or inodes. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 778 | */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 779 | if (progress) |
Wu Fengguang | cb9bd11 | 2010-07-21 22:50:57 -0600 | [diff] [blame] | 780 | continue; |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 781 | /* |
Wu Fengguang | e6fb6da | 2010-07-22 10:23:44 -0600 | [diff] [blame] | 782 | * No more inodes for IO, bail |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 783 | */ |
Wu Fengguang | b7a2441 | 2010-07-21 22:19:51 -0600 | [diff] [blame] | 784 | if (list_empty(&wb->b_more_io)) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 785 | break; |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 786 | /* |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 787 | * Nothing written. Wait for some inode to |
| 788 | * become available for writeback. Otherwise |
| 789 | * we'll just busyloop. |
| 790 | */ |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 791 | if (!list_empty(&wb->b_more_io)) { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 792 | trace_writeback_wait(wb->bdi, work); |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 793 | inode = wb_inode(wb->b_more_io.prev); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 794 | spin_lock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 795 | inode_wait_for_writeback(inode, wb); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 796 | spin_unlock(&inode->i_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 797 | } |
| 798 | } |
Wu Fengguang | e8dfc30 | 2011-04-21 12:06:32 -0600 | [diff] [blame] | 799 | spin_unlock(&wb->list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 800 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 801 | return nr_pages - work->nr_pages; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 802 | } |
| 803 | |
| 804 | /* |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 805 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 806 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 807 | static struct wb_writeback_work * |
Minchan Kim | 08852b6 | 2010-08-03 12:51:16 +0200 | [diff] [blame] | 808 | get_next_work_item(struct backing_dev_info *bdi) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 809 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 810 | struct wb_writeback_work *work = NULL; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 811 | |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 812 | spin_lock_bh(&bdi->wb_lock); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 813 | if (!list_empty(&bdi->work_list)) { |
| 814 | work = list_entry(bdi->work_list.next, |
| 815 | struct wb_writeback_work, list); |
| 816 | list_del_init(&work->list); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 817 | } |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 818 | spin_unlock_bh(&bdi->wb_lock); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 819 | return work; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 820 | } |
| 821 | |
Linus Torvalds | cdf01dd | 2010-10-30 08:55:52 -0700 | [diff] [blame] | 822 | /* |
| 823 | * Add in the number of potentially dirty inodes, because each inode |
| 824 | * write can dirty pagecache in the underlying blockdev. |
| 825 | */ |
| 826 | static unsigned long get_nr_dirty_pages(void) |
| 827 | { |
| 828 | return global_page_state(NR_FILE_DIRTY) + |
| 829 | global_page_state(NR_UNSTABLE_NFS) + |
| 830 | get_nr_dirty_inodes(); |
| 831 | } |
| 832 | |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 833 | static long wb_check_background_flush(struct bdi_writeback *wb) |
| 834 | { |
| 835 | if (over_bground_thresh()) { |
| 836 | |
| 837 | struct wb_writeback_work work = { |
| 838 | .nr_pages = LONG_MAX, |
| 839 | .sync_mode = WB_SYNC_NONE, |
| 840 | .for_background = 1, |
| 841 | .range_cyclic = 1, |
| 842 | }; |
| 843 | |
| 844 | return wb_writeback(wb, &work); |
| 845 | } |
| 846 | |
| 847 | return 0; |
| 848 | } |
| 849 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 850 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
| 851 | { |
| 852 | unsigned long expired; |
| 853 | long nr_pages; |
| 854 | |
Jens Axboe | 69b62d0 | 2010-05-17 12:51:03 +0200 | [diff] [blame] | 855 | /* |
| 856 | * When set to zero, disable periodic writeback |
| 857 | */ |
| 858 | if (!dirty_writeback_interval) |
| 859 | return 0; |
| 860 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 861 | expired = wb->last_old_flush + |
| 862 | msecs_to_jiffies(dirty_writeback_interval * 10); |
| 863 | if (time_before(jiffies, expired)) |
| 864 | return 0; |
| 865 | |
| 866 | wb->last_old_flush = jiffies; |
Linus Torvalds | cdf01dd | 2010-10-30 08:55:52 -0700 | [diff] [blame] | 867 | nr_pages = get_nr_dirty_pages(); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 868 | |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 869 | if (nr_pages) { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 870 | struct wb_writeback_work work = { |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 871 | .nr_pages = nr_pages, |
| 872 | .sync_mode = WB_SYNC_NONE, |
| 873 | .for_kupdate = 1, |
| 874 | .range_cyclic = 1, |
| 875 | }; |
| 876 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 877 | return wb_writeback(wb, &work); |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 878 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 879 | |
| 880 | return 0; |
| 881 | } |
| 882 | |
| 883 | /* |
| 884 | * Retrieve work items and do the writeback they describe |
| 885 | */ |
| 886 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) |
| 887 | { |
| 888 | struct backing_dev_info *bdi = wb->bdi; |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 889 | struct wb_writeback_work *work; |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 890 | long wrote = 0; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 891 | |
Jan Kara | 81d73a3 | 2010-08-11 14:17:44 -0700 | [diff] [blame] | 892 | set_bit(BDI_writeback_running, &wb->bdi->state); |
Minchan Kim | 08852b6 | 2010-08-03 12:51:16 +0200 | [diff] [blame] | 893 | while ((work = get_next_work_item(bdi)) != NULL) { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 894 | /* |
| 895 | * Override sync mode, in case we must wait for completion |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 896 | * because this thread is exiting now. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 897 | */ |
| 898 | if (force_wait) |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 899 | work->sync_mode = WB_SYNC_ALL; |
| 900 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 901 | trace_writeback_exec(bdi, work); |
| 902 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 903 | wrote += wb_writeback(wb, work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 904 | |
| 905 | /* |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 906 | * Notify the caller of completion if this is a synchronous |
| 907 | * work item, otherwise just free it. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 908 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 909 | if (work->done) |
| 910 | complete(work->done); |
| 911 | else |
| 912 | kfree(work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 913 | } |
| 914 | |
| 915 | /* |
| 916 | * Check for periodic writeback, kupdated() style |
| 917 | */ |
| 918 | wrote += wb_check_old_data_flush(wb); |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 919 | wrote += wb_check_background_flush(wb); |
Jan Kara | 81d73a3 | 2010-08-11 14:17:44 -0700 | [diff] [blame] | 920 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 921 | |
| 922 | return wrote; |
| 923 | } |
| 924 | |
| 925 | /* |
| 926 | * Handle writeback of dirty data for the device backed by this bdi. Also |
| 927 | * wakes up periodically and does kupdated style flushing. |
| 928 | */ |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 929 | int bdi_writeback_thread(void *data) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 930 | { |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 931 | struct bdi_writeback *wb = data; |
| 932 | struct backing_dev_info *bdi = wb->bdi; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 933 | long pages_written; |
| 934 | |
Peter Zijlstra | 766f916 | 2010-10-26 14:22:45 -0700 | [diff] [blame] | 935 | current->flags |= PF_SWAPWRITE; |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 936 | set_freezable(); |
Artem Bityutskiy | ecd5840 | 2010-07-25 14:29:18 +0300 | [diff] [blame] | 937 | wb->last_active = jiffies; |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 938 | |
| 939 | /* |
| 940 | * Our parent may run at a different priority, just set us to normal |
| 941 | */ |
| 942 | set_user_nice(current, 0); |
| 943 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 944 | trace_writeback_thread_start(bdi); |
| 945 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 946 | while (!kthread_should_stop()) { |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 947 | /* |
| 948 | * Remove own delayed wake-up timer, since we are already awake |
| 949 | * and we'll take care of the preriodic write-back. |
| 950 | */ |
| 951 | del_timer(&wb->wakeup_timer); |
| 952 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 953 | pages_written = wb_do_writeback(wb, 0); |
| 954 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 955 | trace_writeback_pages_written(pages_written); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 956 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 957 | if (pages_written) |
Artem Bityutskiy | ecd5840 | 2010-07-25 14:29:18 +0300 | [diff] [blame] | 958 | wb->last_active = jiffies; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 959 | |
Artem Bityutskiy | 297252c | 2010-07-25 14:29:15 +0300 | [diff] [blame] | 960 | set_current_state(TASK_INTERRUPTIBLE); |
J. Bruce Fields | b76b401 | 2010-08-28 08:52:10 +0200 | [diff] [blame] | 961 | if (!list_empty(&bdi->work_list) || kthread_should_stop()) { |
Artem Bityutskiy | 297252c | 2010-07-25 14:29:15 +0300 | [diff] [blame] | 962 | __set_current_state(TASK_RUNNING); |
| 963 | continue; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 964 | } |
| 965 | |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 966 | if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 967 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 968 | else { |
| 969 | /* |
| 970 | * We have nothing to do, so can go sleep without any |
| 971 | * timeout and save power. When a work is queued or |
| 972 | * something is made dirty - we will be woken up. |
| 973 | */ |
Artem Bityutskiy | 297252c | 2010-07-25 14:29:15 +0300 | [diff] [blame] | 974 | schedule(); |
Jens Axboe | f9eadbb | 2010-05-18 14:31:45 +0200 | [diff] [blame] | 975 | } |
Jens Axboe | 69b62d0 | 2010-05-17 12:51:03 +0200 | [diff] [blame] | 976 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 977 | try_to_freeze(); |
| 978 | } |
| 979 | |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 980 | /* Flush any work that raced with us exiting */ |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 981 | if (!list_empty(&bdi->work_list)) |
| 982 | wb_do_writeback(wb, 1); |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 983 | |
| 984 | trace_writeback_thread_stop(bdi); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 985 | return 0; |
| 986 | } |
| 987 | |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 988 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 989 | /* |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 990 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
| 991 | * the whole world. |
| 992 | */ |
| 993 | void wakeup_flusher_threads(long nr_pages) |
| 994 | { |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 995 | struct backing_dev_info *bdi; |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 996 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 997 | if (!nr_pages) { |
| 998 | nr_pages = global_page_state(NR_FILE_DIRTY) + |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 999 | global_page_state(NR_UNSTABLE_NFS); |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 1000 | } |
| 1001 | |
| 1002 | rcu_read_lock(); |
| 1003 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
| 1004 | if (!bdi_has_dirty_io(bdi)) |
| 1005 | continue; |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 1006 | __bdi_start_writeback(bdi, nr_pages, false); |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 1007 | } |
| 1008 | rcu_read_unlock(); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1009 | } |
| 1010 | |
| 1011 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
| 1012 | { |
| 1013 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { |
| 1014 | struct dentry *dentry; |
| 1015 | const char *name = "?"; |
| 1016 | |
| 1017 | dentry = d_find_alias(inode); |
| 1018 | if (dentry) { |
| 1019 | spin_lock(&dentry->d_lock); |
| 1020 | name = (const char *) dentry->d_name.name; |
| 1021 | } |
| 1022 | printk(KERN_DEBUG |
| 1023 | "%s(%d): dirtied inode %lu (%s) on %s\n", |
| 1024 | current->comm, task_pid_nr(current), inode->i_ino, |
| 1025 | name, inode->i_sb->s_id); |
| 1026 | if (dentry) { |
| 1027 | spin_unlock(&dentry->d_lock); |
| 1028 | dput(dentry); |
| 1029 | } |
| 1030 | } |
| 1031 | } |
| 1032 | |
| 1033 | /** |
| 1034 | * __mark_inode_dirty - internal function |
| 1035 | * @inode: inode to mark |
| 1036 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) |
| 1037 | * Mark an inode as dirty. Callers should use mark_inode_dirty or |
| 1038 | * mark_inode_dirty_sync. |
| 1039 | * |
| 1040 | * Put the inode on the super block's dirty list. |
| 1041 | * |
| 1042 | * CAREFUL! We mark it dirty unconditionally, but move it onto the |
| 1043 | * dirty list only if it is hashed or if it refers to a blockdev. |
| 1044 | * If it was not hashed, it will never be added to the dirty list |
| 1045 | * even if it is later hashed, as it will have been marked dirty already. |
| 1046 | * |
| 1047 | * In short, make sure you hash any inodes _before_ you start marking |
| 1048 | * them dirty. |
| 1049 | * |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1050 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
| 1051 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of |
| 1052 | * the kernel-internal blockdev inode represents the dirtying time of the |
| 1053 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use |
| 1054 | * page->mapping->host, so the page-dirtying time is recorded in the internal |
| 1055 | * blockdev inode. |
| 1056 | */ |
| 1057 | void __mark_inode_dirty(struct inode *inode, int flags) |
| 1058 | { |
| 1059 | struct super_block *sb = inode->i_sb; |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1060 | struct backing_dev_info *bdi = NULL; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1061 | |
| 1062 | /* |
| 1063 | * Don't do this for I_DIRTY_PAGES - that doesn't actually |
| 1064 | * dirty the inode itself |
| 1065 | */ |
| 1066 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { |
| 1067 | if (sb->s_op->dirty_inode) |
Christoph Hellwig | aa38572 | 2011-05-27 06:53:02 -0400 | [diff] [blame] | 1068 | sb->s_op->dirty_inode(inode, flags); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1069 | } |
| 1070 | |
| 1071 | /* |
| 1072 | * make sure that changes are seen by all cpus before we test i_state |
| 1073 | * -- mikulas |
| 1074 | */ |
| 1075 | smp_mb(); |
| 1076 | |
| 1077 | /* avoid the locking if we can */ |
| 1078 | if ((inode->i_state & flags) == flags) |
| 1079 | return; |
| 1080 | |
| 1081 | if (unlikely(block_dump)) |
| 1082 | block_dump___mark_inode_dirty(inode); |
| 1083 | |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1084 | spin_lock(&inode->i_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1085 | if ((inode->i_state & flags) != flags) { |
| 1086 | const int was_dirty = inode->i_state & I_DIRTY; |
| 1087 | |
| 1088 | inode->i_state |= flags; |
| 1089 | |
| 1090 | /* |
| 1091 | * If the inode is being synced, just update its dirty state. |
| 1092 | * The unlocker will place the inode on the appropriate |
| 1093 | * superblock list, based upon its state. |
| 1094 | */ |
| 1095 | if (inode->i_state & I_SYNC) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1096 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1097 | |
| 1098 | /* |
| 1099 | * Only add valid (hashed) inodes to the superblock's |
| 1100 | * dirty list. Add blockdev inodes as well. |
| 1101 | */ |
| 1102 | if (!S_ISBLK(inode->i_mode)) { |
Al Viro | 1d3382cb | 2010-10-23 15:19:20 -0400 | [diff] [blame] | 1103 | if (inode_unhashed(inode)) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1104 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1105 | } |
Al Viro | a4ffdde | 2010-06-02 17:38:30 -0400 | [diff] [blame] | 1106 | if (inode->i_state & I_FREEING) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1107 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1108 | |
| 1109 | /* |
| 1110 | * If the inode was already on b_dirty/b_io/b_more_io, don't |
| 1111 | * reposition it (that would break b_dirty time-ordering). |
| 1112 | */ |
| 1113 | if (!was_dirty) { |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 1114 | bool wakeup_bdi = false; |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1115 | bdi = inode_to_bdi(inode); |
Jens Axboe | 500b067 | 2009-09-09 09:10:25 +0200 | [diff] [blame] | 1116 | |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1117 | if (bdi_cap_writeback_dirty(bdi)) { |
| 1118 | WARN(!test_bit(BDI_registered, &bdi->state), |
| 1119 | "bdi-%s not registered\n", bdi->name); |
| 1120 | |
| 1121 | /* |
| 1122 | * If this is the first dirty inode for this |
| 1123 | * bdi, we have to wake-up the corresponding |
| 1124 | * bdi thread to make sure background |
| 1125 | * write-back happens later. |
| 1126 | */ |
| 1127 | if (!wb_has_dirty_io(&bdi->wb)) |
| 1128 | wakeup_bdi = true; |
Jens Axboe | 500b067 | 2009-09-09 09:10:25 +0200 | [diff] [blame] | 1129 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1130 | |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 1131 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1132 | spin_lock(&bdi->wb.list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1133 | inode->dirtied_when = jiffies; |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 1134 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1135 | spin_unlock(&bdi->wb.list_lock); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 1136 | |
| 1137 | if (wakeup_bdi) |
| 1138 | bdi_wakeup_thread_delayed(bdi); |
| 1139 | return; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1140 | } |
| 1141 | } |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1142 | out_unlock_inode: |
| 1143 | spin_unlock(&inode->i_lock); |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1144 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1145 | } |
| 1146 | EXPORT_SYMBOL(__mark_inode_dirty); |
| 1147 | |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1148 | /* |
| 1149 | * Write out a superblock's list of dirty inodes. A wait will be performed |
| 1150 | * upon no inodes, all inodes or the final one, depending upon sync_mode. |
| 1151 | * |
| 1152 | * If older_than_this is non-NULL, then only write out inodes which |
| 1153 | * had their first dirtying at a time earlier than *older_than_this. |
| 1154 | * |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1155 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
| 1156 | * This function assumes that the blockdev superblock's inodes are backed by |
| 1157 | * a variety of queues, so all inodes are searched. For other superblocks, |
| 1158 | * assume that all inodes are backed by the same queue. |
| 1159 | * |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1160 | * The inodes to be written are parked on bdi->b_io. They are moved back onto |
| 1161 | * bdi->b_dirty as they are selected for writing. This way, none can be missed |
| 1162 | * on the writer throttling path, and we get decent balancing between many |
| 1163 | * throttled threads: we don't want them all piling up on inode_sync_wait. |
| 1164 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1165 | static void wait_sb_inodes(struct super_block *sb) |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1166 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1167 | struct inode *inode, *old_inode = NULL; |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1168 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1169 | /* |
| 1170 | * We need to be protected against the filesystem going from |
| 1171 | * r/o to r/w or vice versa. |
| 1172 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1173 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1174 | |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1175 | spin_lock(&inode_sb_list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1176 | |
| 1177 | /* |
| 1178 | * Data integrity sync. Must wait for all pages under writeback, |
| 1179 | * because there may have been pages dirtied before our sync |
| 1180 | * call, but which had writeout started before we write it out. |
| 1181 | * In which case, the inode may not be on the dirty list, but |
| 1182 | * we still have to wait for that writeout. |
| 1183 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1184 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1185 | struct address_space *mapping = inode->i_mapping; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1186 | |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1187 | spin_lock(&inode->i_lock); |
| 1188 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || |
| 1189 | (mapping->nrpages == 0)) { |
| 1190 | spin_unlock(&inode->i_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1191 | continue; |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1192 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1193 | __iget(inode); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1194 | spin_unlock(&inode->i_lock); |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1195 | spin_unlock(&inode_sb_list_lock); |
| 1196 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1197 | /* |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1198 | * We hold a reference to 'inode' so it couldn't have been |
| 1199 | * removed from s_inodes list while we dropped the |
| 1200 | * inode_sb_list_lock. We cannot iput the inode now as we can |
| 1201 | * be holding the last reference and we cannot iput it under |
| 1202 | * inode_sb_list_lock. So we keep the reference and iput it |
| 1203 | * later. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1204 | */ |
| 1205 | iput(old_inode); |
| 1206 | old_inode = inode; |
| 1207 | |
| 1208 | filemap_fdatawait(mapping); |
| 1209 | |
| 1210 | cond_resched(); |
Nick Piggin | 38f2197 | 2009-01-06 14:40:25 -0800 | [diff] [blame] | 1211 | |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1212 | spin_lock(&inode_sb_list_lock); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1213 | } |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1214 | spin_unlock(&inode_sb_list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1215 | iput(old_inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1216 | } |
| 1217 | |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1218 | /** |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1219 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1220 | * @sb: the superblock |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1221 | * @nr: the number of pages to write |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1222 | * |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1223 | * Start writeback on some inodes on this super_block. No guarantees are made |
| 1224 | * on how many (if any) will be written, and this function does not wait |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1225 | * for IO completion of submitted IO. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1226 | */ |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1227 | void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1228 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1229 | DECLARE_COMPLETION_ONSTACK(done); |
| 1230 | struct wb_writeback_work work = { |
Wu Fengguang | 6e6938b | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 1231 | .sb = sb, |
| 1232 | .sync_mode = WB_SYNC_NONE, |
| 1233 | .tagged_writepages = 1, |
| 1234 | .done = &done, |
| 1235 | .nr_pages = nr, |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1236 | }; |
Jens Axboe | 0e3c9a2 | 2010-06-01 11:08:43 +0200 | [diff] [blame] | 1237 | |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1238 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1239 | bdi_queue_work(sb->s_bdi, &work); |
| 1240 | wait_for_completion(&done); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1241 | } |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1242 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
| 1243 | |
| 1244 | /** |
| 1245 | * writeback_inodes_sb - writeback dirty inodes from given super_block |
| 1246 | * @sb: the superblock |
| 1247 | * |
| 1248 | * Start writeback on some inodes on this super_block. No guarantees are made |
| 1249 | * on how many (if any) will be written, and this function does not wait |
| 1250 | * for IO completion of submitted IO. |
| 1251 | */ |
| 1252 | void writeback_inodes_sb(struct super_block *sb) |
| 1253 | { |
Linus Torvalds | 925d169 | 2010-10-30 09:05:48 -0700 | [diff] [blame] | 1254 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages()); |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1255 | } |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1256 | EXPORT_SYMBOL(writeback_inodes_sb); |
| 1257 | |
| 1258 | /** |
Eric Sandeen | 17bd55d | 2009-12-23 07:57:07 -0500 | [diff] [blame] | 1259 | * writeback_inodes_sb_if_idle - start writeback if none underway |
| 1260 | * @sb: the superblock |
| 1261 | * |
| 1262 | * Invoke writeback_inodes_sb if no writeback is currently underway. |
| 1263 | * Returns 1 if writeback was started, 0 if not. |
| 1264 | */ |
| 1265 | int writeback_inodes_sb_if_idle(struct super_block *sb) |
| 1266 | { |
| 1267 | if (!writeback_in_progress(sb->s_bdi)) { |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1268 | down_read(&sb->s_umount); |
Eric Sandeen | 17bd55d | 2009-12-23 07:57:07 -0500 | [diff] [blame] | 1269 | writeback_inodes_sb(sb); |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1270 | up_read(&sb->s_umount); |
Eric Sandeen | 17bd55d | 2009-12-23 07:57:07 -0500 | [diff] [blame] | 1271 | return 1; |
| 1272 | } else |
| 1273 | return 0; |
| 1274 | } |
| 1275 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); |
| 1276 | |
| 1277 | /** |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1278 | * writeback_inodes_sb_if_idle - start writeback if none underway |
| 1279 | * @sb: the superblock |
| 1280 | * @nr: the number of pages to write |
| 1281 | * |
| 1282 | * Invoke writeback_inodes_sb if no writeback is currently underway. |
| 1283 | * Returns 1 if writeback was started, 0 if not. |
| 1284 | */ |
| 1285 | int writeback_inodes_sb_nr_if_idle(struct super_block *sb, |
| 1286 | unsigned long nr) |
| 1287 | { |
| 1288 | if (!writeback_in_progress(sb->s_bdi)) { |
| 1289 | down_read(&sb->s_umount); |
| 1290 | writeback_inodes_sb_nr(sb, nr); |
| 1291 | up_read(&sb->s_umount); |
| 1292 | return 1; |
| 1293 | } else |
| 1294 | return 0; |
| 1295 | } |
| 1296 | EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); |
| 1297 | |
| 1298 | /** |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1299 | * sync_inodes_sb - sync sb inode pages |
| 1300 | * @sb: the superblock |
| 1301 | * |
| 1302 | * This function writes and waits on any dirty inode belonging to this |
Stefan Hajnoczi | cb9ef8d | 2011-01-13 15:47:26 -0800 | [diff] [blame] | 1303 | * super_block. |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1304 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1305 | void sync_inodes_sb(struct super_block *sb) |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1306 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1307 | DECLARE_COMPLETION_ONSTACK(done); |
| 1308 | struct wb_writeback_work work = { |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1309 | .sb = sb, |
| 1310 | .sync_mode = WB_SYNC_ALL, |
| 1311 | .nr_pages = LONG_MAX, |
| 1312 | .range_cyclic = 0, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1313 | .done = &done, |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1314 | }; |
| 1315 | |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1316 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
| 1317 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1318 | bdi_queue_work(sb->s_bdi, &work); |
| 1319 | wait_for_completion(&done); |
| 1320 | |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1321 | wait_sb_inodes(sb); |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1322 | } |
| 1323 | EXPORT_SYMBOL(sync_inodes_sb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1324 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1325 | /** |
Andrea Arcangeli | 7f04c26 | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 1326 | * write_inode_now - write an inode to disk |
| 1327 | * @inode: inode to write to disk |
| 1328 | * @sync: whether the write should be synchronous or not |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1329 | * |
Andrea Arcangeli | 7f04c26 | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 1330 | * This function commits an inode to disk immediately if it is dirty. This is |
| 1331 | * primarily needed by knfsd. |
| 1332 | * |
| 1333 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1334 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1335 | int write_inode_now(struct inode *inode, int sync) |
| 1336 | { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1337 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1338 | int ret; |
| 1339 | struct writeback_control wbc = { |
| 1340 | .nr_to_write = LONG_MAX, |
Mike Galbraith | 18914b1 | 2008-02-08 04:20:23 -0800 | [diff] [blame] | 1341 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 1342 | .range_start = 0, |
| 1343 | .range_end = LLONG_MAX, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1344 | }; |
| 1345 | |
| 1346 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) |
Andrew Morton | 49364ce | 2005-11-07 00:59:15 -0800 | [diff] [blame] | 1347 | wbc.nr_to_write = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1348 | |
| 1349 | might_sleep(); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1350 | spin_lock(&wb->list_lock); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 1351 | spin_lock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1352 | ret = writeback_single_inode(inode, wb, &wbc); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 1353 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1354 | spin_unlock(&wb->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1355 | if (sync) |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 1356 | inode_sync_wait(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1357 | return ret; |
| 1358 | } |
| 1359 | EXPORT_SYMBOL(write_inode_now); |
| 1360 | |
| 1361 | /** |
| 1362 | * sync_inode - write an inode and its pages to disk. |
| 1363 | * @inode: the inode to sync |
| 1364 | * @wbc: controls the writeback mode |
| 1365 | * |
| 1366 | * sync_inode() will write an inode and its pages to disk. It will also |
| 1367 | * correctly update the inode on its superblock's dirty inode lists and will |
| 1368 | * update inode->i_state. |
| 1369 | * |
| 1370 | * The caller must have a ref on the inode. |
| 1371 | */ |
| 1372 | int sync_inode(struct inode *inode, struct writeback_control *wbc) |
| 1373 | { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1374 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1375 | int ret; |
| 1376 | |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1377 | spin_lock(&wb->list_lock); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 1378 | spin_lock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1379 | ret = writeback_single_inode(inode, wb, wbc); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 1380 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1381 | spin_unlock(&wb->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1382 | return ret; |
| 1383 | } |
| 1384 | EXPORT_SYMBOL(sync_inode); |
Christoph Hellwig | c3765016 | 2010-10-06 10:48:20 +0200 | [diff] [blame] | 1385 | |
| 1386 | /** |
Andrew Morton | c691b9d | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 1387 | * sync_inode_metadata - write an inode to disk |
Christoph Hellwig | c3765016 | 2010-10-06 10:48:20 +0200 | [diff] [blame] | 1388 | * @inode: the inode to sync |
| 1389 | * @wait: wait for I/O to complete. |
| 1390 | * |
Andrew Morton | c691b9d | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 1391 | * Write an inode to disk and adjust its dirty state after completion. |
Christoph Hellwig | c3765016 | 2010-10-06 10:48:20 +0200 | [diff] [blame] | 1392 | * |
| 1393 | * Note: only writes the actual inode, no associated data or other metadata. |
| 1394 | */ |
| 1395 | int sync_inode_metadata(struct inode *inode, int wait) |
| 1396 | { |
| 1397 | struct writeback_control wbc = { |
| 1398 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, |
| 1399 | .nr_to_write = 0, /* metadata-only */ |
| 1400 | }; |
| 1401 | |
| 1402 | return sync_inode(inode, &wbc); |
| 1403 | } |
| 1404 | EXPORT_SYMBOL(sync_inode_metadata); |