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 | 07f3f05 | 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; |
| 38 | enum writeback_sync_modes sync_mode; |
Wu Fengguang | ac69306 | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 39 | unsigned int tagged_writepages:1; |
H Hartley Sweeten | 52957fe | 2010-04-01 20:36:30 -0500 | [diff] [blame] | 40 | unsigned int for_kupdate:1; |
| 41 | unsigned int range_cyclic:1; |
| 42 | unsigned int for_background:1; |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 43 | |
Jens Axboe | 8010c3b | 2009-09-15 20:04:57 +0200 | [diff] [blame] | 44 | struct list_head list; /* pending work list */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 45 | struct completion *done; /* set if the caller waits */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 46 | }; |
| 47 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 48 | /* |
| 49 | * Include the creation of the trace points after defining the |
| 50 | * wb_writeback_work structure so that the definition remains local to this |
| 51 | * file. |
| 52 | */ |
| 53 | #define CREATE_TRACE_POINTS |
| 54 | #include <trace/events/writeback.h> |
| 55 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 56 | /* |
| 57 | * We don't actually have pdflush, but this one is exported though /proc... |
| 58 | */ |
| 59 | int nr_pdflush_threads; |
| 60 | |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 61 | /** |
| 62 | * writeback_in_progress - determine whether there is writeback in progress |
| 63 | * @bdi: the device's backing_dev_info structure. |
| 64 | * |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 65 | * Determine whether there is writeback waiting to be handled against a |
| 66 | * backing device. |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 67 | */ |
| 68 | int writeback_in_progress(struct backing_dev_info *bdi) |
| 69 | { |
Jan Kara | 81d73a3 | 2010-08-11 14:17:44 -0700 | [diff] [blame] | 70 | return test_bit(BDI_writeback_running, &bdi->state); |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 71 | } |
| 72 | |
Jan Kara | 692ebd1 | 2010-09-21 11:51:01 +0200 | [diff] [blame] | 73 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
| 74 | { |
| 75 | struct super_block *sb = inode->i_sb; |
Jan Kara | 692ebd1 | 2010-09-21 11:51:01 +0200 | [diff] [blame] | 76 | |
Christoph Hellwig | aaead25 | 2010-10-04 14:25:33 +0200 | [diff] [blame] | 77 | if (strcmp(sb->s_type->name, "bdev") == 0) |
| 78 | return inode->i_mapping->backing_dev_info; |
| 79 | |
| 80 | return sb->s_bdi; |
Jan Kara | 692ebd1 | 2010-09-21 11:51:01 +0200 | [diff] [blame] | 81 | } |
| 82 | |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 83 | static inline struct inode *wb_inode(struct list_head *head) |
| 84 | { |
| 85 | return list_entry(head, struct inode, i_wb_list); |
| 86 | } |
| 87 | |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 88 | /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ |
| 89 | static void bdi_wakeup_flusher(struct backing_dev_info *bdi) |
Nick Piggin | 4195f73 | 2009-05-28 09:01:15 +0200 | [diff] [blame] | 90 | { |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 91 | if (bdi->wb.task) { |
| 92 | wake_up_process(bdi->wb.task); |
| 93 | } else { |
| 94 | /* |
| 95 | * The bdi thread isn't there, wake up the forker thread which |
| 96 | * will create and run it. |
| 97 | */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 98 | wake_up_process(default_backing_dev_info.wb.task); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 99 | } |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 100 | } |
| 101 | |
| 102 | static void bdi_queue_work(struct backing_dev_info *bdi, |
| 103 | struct wb_writeback_work *work) |
| 104 | { |
| 105 | trace_writeback_queue(bdi, work); |
| 106 | |
| 107 | spin_lock_bh(&bdi->wb_lock); |
| 108 | list_add_tail(&work->list, &bdi->work_list); |
| 109 | if (!bdi->wb.task) |
| 110 | trace_writeback_nothread(bdi, work); |
| 111 | bdi_wakeup_flusher(bdi); |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 112 | spin_unlock_bh(&bdi->wb_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 113 | } |
| 114 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 115 | static void |
| 116 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 117 | bool range_cyclic) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 118 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 119 | struct wb_writeback_work *work; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 120 | |
Jens Axboe | bcddc3f | 2009-09-13 20:07:36 +0200 | [diff] [blame] | 121 | /* |
| 122 | * This is WB_SYNC_NONE writeback, so if allocation fails just |
| 123 | * wakeup the thread for old dirty data writeback |
| 124 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 125 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
| 126 | if (!work) { |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 127 | if (bdi->wb.task) { |
| 128 | trace_writeback_nowork(bdi); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 129 | wake_up_process(bdi->wb.task); |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 130 | } |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 131 | return; |
Jens Axboe | bcddc3f | 2009-09-13 20:07:36 +0200 | [diff] [blame] | 132 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 133 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 134 | work->sync_mode = WB_SYNC_NONE; |
| 135 | work->nr_pages = nr_pages; |
| 136 | work->range_cyclic = range_cyclic; |
Christoph Hellwig | f0fad8a | 2009-09-11 09:47:56 +0200 | [diff] [blame] | 137 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 138 | bdi_queue_work(bdi, work); |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 139 | } |
| 140 | |
| 141 | /** |
| 142 | * bdi_start_writeback - start writeback |
| 143 | * @bdi: the backing device to write from |
| 144 | * @nr_pages: the number of pages to write |
| 145 | * |
| 146 | * Description: |
| 147 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 148 | * started when this function returns, we make no guarantees on |
Jens Axboe | 0e3c9a2 | 2010-06-01 11:08:43 +0200 | [diff] [blame] | 149 | * completion. Caller need not hold sb s_umount semaphore. |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 150 | * |
| 151 | */ |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 152 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 153 | { |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 154 | __bdi_start_writeback(bdi, nr_pages, true); |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 155 | } |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 156 | |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 157 | /** |
| 158 | * bdi_start_background_writeback - start background writeback |
| 159 | * @bdi: the backing device to write from |
| 160 | * |
| 161 | * Description: |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 162 | * This makes sure WB_SYNC_NONE background writeback happens. When |
| 163 | * this function returns, it is only guaranteed that for given BDI |
| 164 | * some IO is happening if we are over background dirty threshold. |
| 165 | * Caller need not hold sb s_umount semaphore. |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 166 | */ |
| 167 | void bdi_start_background_writeback(struct backing_dev_info *bdi) |
| 168 | { |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 169 | /* |
| 170 | * We just wake up the flusher thread. It will perform background |
| 171 | * writeback as soon as there is no other work to do. |
| 172 | */ |
Wu Fengguang | 71927e8 | 2011-01-13 15:45:46 -0800 | [diff] [blame] | 173 | trace_writeback_wake_background(bdi); |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 174 | spin_lock_bh(&bdi->wb_lock); |
| 175 | bdi_wakeup_flusher(bdi); |
| 176 | spin_unlock_bh(&bdi->wb_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | } |
| 178 | |
| 179 | /* |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 180 | * Remove the inode from the writeback list it is on. |
| 181 | */ |
| 182 | void inode_wb_list_del(struct inode *inode) |
| 183 | { |
| 184 | spin_lock(&inode_wb_list_lock); |
| 185 | list_del_init(&inode->i_wb_list); |
| 186 | spin_unlock(&inode_wb_list_lock); |
| 187 | } |
| 188 | |
| 189 | |
| 190 | /* |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 191 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the |
| 192 | * furthest end of its superblock's dirty-inode list. |
| 193 | * |
| 194 | * 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] | 195 | * 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] | 196 | * the case then the inode must have been redirtied while it was being written |
| 197 | * out and we don't reset its dirtied_when. |
| 198 | */ |
| 199 | static void redirty_tail(struct inode *inode) |
| 200 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 201 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 202 | |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 203 | assert_spin_locked(&inode_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 | */ |
Ken Chen | 0e0f4fc | 2007-10-16 23:30:38 -0700 | [diff] [blame] | 217 | static void requeue_io(struct inode *inode) |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 218 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 219 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
| 220 | |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 221 | assert_spin_locked(&inode_wb_list_lock); |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 222 | list_move(&inode->i_wb_list, &wb->b_more_io); |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 223 | } |
| 224 | |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 225 | static void inode_sync_complete(struct inode *inode) |
| 226 | { |
| 227 | /* |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 228 | * Prevent speculative execution through |
| 229 | * spin_unlock(&inode_wb_list_lock); |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 230 | */ |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 231 | |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 232 | smp_mb(); |
| 233 | wake_up_bit(&inode->i_state, __I_SYNC); |
| 234 | } |
| 235 | |
Jeff Layton | d2caa3c | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 236 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
| 237 | { |
| 238 | bool ret = time_after(inode->dirtied_when, t); |
| 239 | #ifndef CONFIG_64BIT |
| 240 | /* |
| 241 | * For inodes being constantly redirtied, dirtied_when can get stuck. |
| 242 | * It _appears_ to be in the future, but is actually in distant past. |
| 243 | * This test is necessary to prevent such wrapped-around relative times |
Jens Axboe | 5b0830c | 2009-09-23 19:37:09 +0200 | [diff] [blame] | 244 | * from permanently stopping the whole bdi writeback. |
Jeff Layton | d2caa3c | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 245 | */ |
| 246 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); |
| 247 | #endif |
| 248 | return ret; |
| 249 | } |
| 250 | |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 251 | /* |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 252 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. |
| 253 | */ |
| 254 | static void move_expired_inodes(struct list_head *delaying_queue, |
| 255 | struct list_head *dispatch_queue, |
| 256 | unsigned long *older_than_this) |
| 257 | { |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 258 | LIST_HEAD(tmp); |
| 259 | struct list_head *pos, *node; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 260 | struct super_block *sb = NULL; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 261 | struct inode *inode; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 262 | int do_sb_sort = 0; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 263 | |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 264 | while (!list_empty(delaying_queue)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 265 | inode = wb_inode(delaying_queue->prev); |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 266 | if (older_than_this && |
Jeff Layton | d2caa3c | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 267 | inode_dirtied_after(inode, *older_than_this)) |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 268 | break; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 269 | if (sb && sb != inode->i_sb) |
| 270 | do_sb_sort = 1; |
| 271 | sb = inode->i_sb; |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 272 | list_move(&inode->i_wb_list, &tmp); |
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); |
| 278 | return; |
| 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 | } |
| 290 | } |
| 291 | |
| 292 | /* |
| 293 | * Queue all expired dirty inodes for io, eldest first. |
Wu Fengguang | 4ea879b | 2010-08-11 14:17:42 -0700 | [diff] [blame] | 294 | * Before |
| 295 | * newly dirtied b_dirty b_io b_more_io |
| 296 | * =============> gf edc BA |
| 297 | * After |
| 298 | * newly dirtied b_dirty b_io b_more_io |
| 299 | * =============> g fBAedc |
| 300 | * | |
| 301 | * +--> dequeue for IO |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 302 | */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 303 | 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] | 304 | { |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 305 | assert_spin_locked(&inode_wb_list_lock); |
Wu Fengguang | 4ea879b | 2010-08-11 14:17:42 -0700 | [diff] [blame] | 306 | list_splice_init(&wb->b_more_io, &wb->b_io); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 307 | move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 308 | } |
| 309 | |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 310 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 311 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 312 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 313 | return inode->i_sb->s_op->write_inode(inode, wbc); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 314 | return 0; |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 315 | } |
| 316 | |
| 317 | /* |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 318 | * Wait for writeback on an inode to complete. |
| 319 | */ |
| 320 | static void inode_wait_for_writeback(struct inode *inode) |
| 321 | { |
| 322 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); |
| 323 | wait_queue_head_t *wqh; |
| 324 | |
| 325 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 326 | while (inode->i_state & I_SYNC) { |
| 327 | spin_unlock(&inode->i_lock); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 328 | spin_unlock(&inode_wb_list_lock); |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 329 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 330 | spin_lock(&inode_wb_list_lock); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 331 | spin_lock(&inode->i_lock); |
Richard Kennedy | 58a9d3d | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 332 | } |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 333 | } |
| 334 | |
| 335 | /* |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 336 | * Write out an inode's dirty pages. Called under inode_wb_list_lock and |
| 337 | * inode->i_lock. Either the caller has an active reference on the inode or |
| 338 | * the inode has I_WILL_FREE set. |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 339 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 340 | * If `wait' is set, wait on the writeout. |
| 341 | * |
| 342 | * The whole writeout design is quite complex and fragile. We want to avoid |
| 343 | * starvation of particular inodes when others are being redirtied, prevent |
| 344 | * livelocks, etc. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 345 | */ |
| 346 | static int |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 347 | writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 348 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 349 | struct address_space *mapping = inode->i_mapping; |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 350 | unsigned dirty; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 351 | int ret; |
| 352 | |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 353 | assert_spin_locked(&inode_wb_list_lock); |
| 354 | assert_spin_locked(&inode->i_lock); |
| 355 | |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 356 | if (!atomic_read(&inode->i_count)) |
| 357 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); |
| 358 | else |
| 359 | WARN_ON(inode->i_state & I_WILL_FREE); |
| 360 | |
| 361 | if (inode->i_state & I_SYNC) { |
| 362 | /* |
| 363 | * If this inode is locked for writeback and we are not doing |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 364 | * writeback-for-data-integrity, move it to b_more_io so that |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 365 | * writeback can proceed with the other inodes on s_io. |
| 366 | * |
| 367 | * We'll have another go at writing back this inode when we |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 368 | * completed a full scan of b_io. |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 369 | */ |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 370 | if (wbc->sync_mode != WB_SYNC_ALL) { |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 371 | requeue_io(inode); |
| 372 | return 0; |
| 373 | } |
| 374 | |
| 375 | /* |
| 376 | * It's a data-integrity sync. We must wait. |
| 377 | */ |
| 378 | inode_wait_for_writeback(inode); |
| 379 | } |
| 380 | |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 381 | BUG_ON(inode->i_state & I_SYNC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 382 | |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 383 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 384 | inode->i_state |= I_SYNC; |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 385 | inode->i_state &= ~I_DIRTY_PAGES; |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 386 | spin_unlock(&inode->i_lock); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 387 | spin_unlock(&inode_wb_list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 388 | |
| 389 | ret = do_writepages(mapping, wbc); |
| 390 | |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 391 | /* |
| 392 | * Make sure to wait on the data before writing out the metadata. |
| 393 | * This is important for filesystems that modify metadata on data |
| 394 | * I/O completion. |
| 395 | */ |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 396 | if (wbc->sync_mode == WB_SYNC_ALL) { |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 397 | int err = filemap_fdatawait(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 398 | if (ret == 0) |
| 399 | ret = err; |
| 400 | } |
| 401 | |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 402 | /* |
| 403 | * Some filesystems may redirty the inode during the writeback |
| 404 | * due to delalloc, clear dirty metadata flags right before |
| 405 | * write_inode() |
| 406 | */ |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 407 | spin_lock(&inode->i_lock); |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 408 | dirty = inode->i_state & I_DIRTY; |
| 409 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 410 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 411 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
| 412 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 413 | int err = write_inode(inode, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 414 | if (ret == 0) |
| 415 | ret = err; |
| 416 | } |
| 417 | |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 418 | spin_lock(&inode_wb_list_lock); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 419 | spin_lock(&inode->i_lock); |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 420 | inode->i_state &= ~I_SYNC; |
Al Viro | a4ffdde | 2010-06-02 17:38:30 -0400 | [diff] [blame] | 421 | if (!(inode->i_state & I_FREEING)) { |
Wu Fengguang | cfdf798 | 2011-04-27 19:05:21 -0600 | [diff] [blame] | 422 | /* |
| 423 | * Sync livelock prevention. Each inode is tagged and synced in |
| 424 | * one shot. If still dirty, it will be redirty_tail()'ed below. |
| 425 | * Update the dirty time to prevent enqueue and sync it again. |
| 426 | */ |
| 427 | if ((inode->i_state & I_DIRTY) && |
| 428 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) |
| 429 | inode->dirtied_when = jiffies; |
| 430 | |
Wu Fengguang | 23539af | 2010-08-11 14:17:41 -0700 | [diff] [blame] | 431 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 432 | /* |
| 433 | * We didn't write back all the pages. nfs_writepages() |
Wu Fengguang | a50aeb4 | 2010-08-11 14:17:43 -0700 | [diff] [blame] | 434 | * sometimes bales out without doing anything. |
Andrew Morton | 1b43ef9 | 2007-10-16 23:30:35 -0700 | [diff] [blame] | 435 | */ |
Wu Fengguang | a50aeb4 | 2010-08-11 14:17:43 -0700 | [diff] [blame] | 436 | inode->i_state |= I_DIRTY_PAGES; |
| 437 | if (wbc->nr_to_write <= 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 438 | /* |
Wu Fengguang | a50aeb4 | 2010-08-11 14:17:43 -0700 | [diff] [blame] | 439 | * slice used up: queue for next turn |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 440 | */ |
Wu Fengguang | a50aeb4 | 2010-08-11 14:17:43 -0700 | [diff] [blame] | 441 | requeue_io(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 442 | } else { |
| 443 | /* |
Wu Fengguang | a50aeb4 | 2010-08-11 14:17:43 -0700 | [diff] [blame] | 444 | * Writeback blocked by something other than |
| 445 | * congestion. Delay the inode for some time to |
| 446 | * avoid spinning on the CPU (100% iowait) |
| 447 | * retrying writeback of the dirty page/inode |
| 448 | * that cannot be performed immediately. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 449 | */ |
Andrew Morton | 1b43ef9 | 2007-10-16 23:30:35 -0700 | [diff] [blame] | 450 | redirty_tail(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 451 | } |
Wu Fengguang | 23539af | 2010-08-11 14:17:41 -0700 | [diff] [blame] | 452 | } else if (inode->i_state & I_DIRTY) { |
| 453 | /* |
| 454 | * Filesystems can dirty the inode during writeback |
| 455 | * operations, such as delayed allocation during |
| 456 | * submission or metadata updates after data IO |
| 457 | * completion. |
| 458 | */ |
| 459 | redirty_tail(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 460 | } else { |
| 461 | /* |
Nick Piggin | 9e38d86 | 2010-10-23 06:55:17 -0400 | [diff] [blame] | 462 | * The inode is clean. At this point we either have |
| 463 | * a reference to the inode or it's on it's way out. |
| 464 | * No need to add it back to the LRU. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 465 | */ |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 466 | list_del_init(&inode->i_wb_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 467 | } |
| 468 | } |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 469 | inode_sync_complete(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | return ret; |
| 471 | } |
| 472 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 473 | /* |
Christoph Hellwig | d19de7e | 2010-06-08 18:14:58 +0200 | [diff] [blame] | 474 | * For background writeback the caller does not have the sb pinned |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 475 | * before calling writeback. So make sure that we do pin it, so it doesn't |
| 476 | * go away while we are writing inodes from it. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 477 | */ |
Christoph Hellwig | d19de7e | 2010-06-08 18:14:58 +0200 | [diff] [blame] | 478 | static bool pin_sb_for_writeback(struct super_block *sb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 479 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 480 | spin_lock(&sb_lock); |
Christoph Hellwig | 29cb485 | 2010-06-09 15:31:01 +0200 | [diff] [blame] | 481 | if (list_empty(&sb->s_instances)) { |
| 482 | spin_unlock(&sb_lock); |
| 483 | return false; |
| 484 | } |
| 485 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 486 | sb->s_count++; |
Christoph Hellwig | 29cb485 | 2010-06-09 15:31:01 +0200 | [diff] [blame] | 487 | spin_unlock(&sb_lock); |
| 488 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 489 | if (down_read_trylock(&sb->s_umount)) { |
Christoph Hellwig | 29cb485 | 2010-06-09 15:31:01 +0200 | [diff] [blame] | 490 | if (sb->s_root) |
Christoph Hellwig | d19de7e | 2010-06-08 18:14:58 +0200 | [diff] [blame] | 491 | return true; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 492 | up_read(&sb->s_umount); |
| 493 | } |
Christoph Hellwig | 29cb485 | 2010-06-09 15:31:01 +0200 | [diff] [blame] | 494 | |
| 495 | put_super(sb); |
Christoph Hellwig | d19de7e | 2010-06-08 18:14:58 +0200 | [diff] [blame] | 496 | return false; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 497 | } |
| 498 | |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 499 | /* |
| 500 | * Write a portion of b_io inodes which belong to @sb. |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 501 | * |
| 502 | * If @only_this_sb is true, then find and write all such |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 503 | * inodes. Otherwise write only ones which go sequentially |
| 504 | * in reverse order. |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 505 | * |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 506 | * Return 1, if the caller writeback routine should be |
| 507 | * interrupted. Otherwise return 0. |
| 508 | */ |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 509 | static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb, |
| 510 | struct writeback_control *wbc, bool only_this_sb) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 511 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 512 | while (!list_empty(&wb->b_io)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 513 | long pages_skipped; |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 514 | struct inode *inode = wb_inode(wb->b_io.prev); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 515 | |
| 516 | if (inode->i_sb != sb) { |
| 517 | if (only_this_sb) { |
| 518 | /* |
| 519 | * We only want to write back data for this |
| 520 | * superblock, move all inodes not belonging |
| 521 | * to it back onto the dirty list. |
| 522 | */ |
| 523 | redirty_tail(inode); |
| 524 | continue; |
| 525 | } |
| 526 | |
| 527 | /* |
| 528 | * The inode belongs to a different superblock. |
| 529 | * Bounce back to the caller to unpin this and |
| 530 | * pin the next superblock. |
| 531 | */ |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 532 | return 0; |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 533 | } |
| 534 | |
Christoph Hellwig | 9843b76 | 2010-10-24 19:40:46 +0200 | [diff] [blame] | 535 | /* |
| 536 | * Don't bother with new inodes or inodes beeing freed, first |
| 537 | * kind does not need peridic writeout yet, and for the latter |
| 538 | * kind writeout is handled by the freer. |
| 539 | */ |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 540 | spin_lock(&inode->i_lock); |
Christoph Hellwig | 9843b76 | 2010-10-24 19:40:46 +0200 | [diff] [blame] | 541 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 542 | spin_unlock(&inode->i_lock); |
Nick Piggin | 7ef0d73 | 2009-03-12 14:31:38 -0700 | [diff] [blame] | 543 | requeue_io(inode); |
| 544 | continue; |
| 545 | } |
Christoph Hellwig | 9843b76 | 2010-10-24 19:40:46 +0200 | [diff] [blame] | 546 | |
Jeff Layton | d2caa3c | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 547 | /* |
| 548 | * Was this inode dirtied after sync_sb_inodes was called? |
| 549 | * This keeps sync from extra jobs and livelock. |
| 550 | */ |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 551 | if (inode_dirtied_after(inode, wbc->wb_start)) { |
| 552 | spin_unlock(&inode->i_lock); |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 553 | return 1; |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 554 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 555 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 556 | __iget(inode); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 557 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 558 | pages_skipped = wbc->pages_skipped; |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 559 | writeback_single_inode(inode, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 560 | if (wbc->pages_skipped != pages_skipped) { |
| 561 | /* |
| 562 | * writeback is not making progress due to locked |
| 563 | * buffers. Skip this inode for now. |
| 564 | */ |
Andrew Morton | f57b9b7 | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 565 | redirty_tail(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | } |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 567 | spin_unlock(&inode->i_lock); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 568 | spin_unlock(&inode_wb_list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 569 | iput(inode); |
OGAWA Hirofumi | 4ffc844 | 2006-03-25 03:07:44 -0800 | [diff] [blame] | 570 | cond_resched(); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 571 | spin_lock(&inode_wb_list_lock); |
Fengguang Wu | 8bc3be2 | 2008-02-04 22:29:36 -0800 | [diff] [blame] | 572 | if (wbc->nr_to_write <= 0) { |
| 573 | wbc->more_io = 1; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 574 | return 1; |
Fengguang Wu | 8bc3be2 | 2008-02-04 22:29:36 -0800 | [diff] [blame] | 575 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 576 | if (!list_empty(&wb->b_more_io)) |
Fengguang Wu | 8bc3be2 | 2008-02-04 22:29:36 -0800 | [diff] [blame] | 577 | wbc->more_io = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 578 | } |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 579 | /* b_io is empty */ |
| 580 | return 1; |
| 581 | } |
Nick Piggin | 38f2197 | 2009-01-06 14:40:25 -0800 | [diff] [blame] | 582 | |
Christoph Hellwig | 9c3a8ee | 2010-06-10 12:07:27 +0200 | [diff] [blame] | 583 | void writeback_inodes_wb(struct bdi_writeback *wb, |
| 584 | struct writeback_control *wbc) |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 585 | { |
| 586 | int ret = 0; |
Jens Axboe | 9ecc273 | 2009-09-24 15:25:11 +0200 | [diff] [blame] | 587 | |
Jan Kara | 7624ee7 | 2010-08-09 17:20:03 -0700 | [diff] [blame] | 588 | if (!wbc->wb_start) |
| 589 | wbc->wb_start = jiffies; /* livelock avoidance */ |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 590 | spin_lock(&inode_wb_list_lock); |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 591 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) |
| 592 | queue_io(wb, wbc->older_than_this); |
| 593 | |
| 594 | while (!list_empty(&wb->b_io)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 595 | struct inode *inode = wb_inode(wb->b_io.prev); |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 596 | struct super_block *sb = inode->i_sb; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 597 | |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 598 | if (!pin_sb_for_writeback(sb)) { |
| 599 | requeue_io(inode); |
| 600 | continue; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 601 | } |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 602 | ret = writeback_sb_inodes(sb, wb, wbc, false); |
| 603 | drop_super(sb); |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 604 | |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 605 | if (ret) |
| 606 | break; |
| 607 | } |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 608 | spin_unlock(&inode_wb_list_lock); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 609 | /* Leave any unwritten inodes on b_io */ |
| 610 | } |
| 611 | |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 612 | static void __writeback_inodes_sb(struct super_block *sb, |
| 613 | struct bdi_writeback *wb, struct writeback_control *wbc) |
| 614 | { |
| 615 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
| 616 | |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 617 | spin_lock(&inode_wb_list_lock); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 618 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) |
| 619 | queue_io(wb, wbc->older_than_this); |
| 620 | writeback_sb_inodes(sb, wb, wbc, true); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 621 | spin_unlock(&inode_wb_list_lock); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 622 | } |
| 623 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 624 | /* |
| 625 | * The maximum number of pages to writeout in a single bdi flush/kupdate |
| 626 | * operation. We do this so we don't hold I_SYNC against an inode for |
| 627 | * enormous amounts of time, which would block a userspace task which has |
| 628 | * been forced to throttle against that inode. Also, the code reevaluates |
| 629 | * the dirty each time it has written this many pages. |
| 630 | */ |
| 631 | #define MAX_WRITEBACK_PAGES 1024 |
| 632 | |
| 633 | static inline bool over_bground_thresh(void) |
| 634 | { |
| 635 | unsigned long background_thresh, dirty_thresh; |
| 636 | |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 637 | global_dirty_limits(&background_thresh, &dirty_thresh); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 638 | |
| 639 | return (global_page_state(NR_FILE_DIRTY) + |
Wu Fengguang | 4cbec4c | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 640 | global_page_state(NR_UNSTABLE_NFS) > background_thresh); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 641 | } |
| 642 | |
| 643 | /* |
| 644 | * Explicit flushing or periodic writeback of "old" data. |
| 645 | * |
| 646 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
| 647 | * dirtying-time in the inode's address_space. So this periodic writeback code |
| 648 | * just walks the superblock inode list, writing back any inodes which are |
| 649 | * older than a specific point in time. |
| 650 | * |
| 651 | * Try to run once per dirty_writeback_interval. But if a writeback event |
| 652 | * takes longer than a dirty_writeback_interval interval, then leave a |
| 653 | * one-second gap. |
| 654 | * |
| 655 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
| 656 | * all dirty pages if they are all attached to "old" mappings. |
| 657 | */ |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 658 | static long wb_writeback(struct bdi_writeback *wb, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 659 | struct wb_writeback_work *work) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 660 | { |
| 661 | struct writeback_control wbc = { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 662 | .sync_mode = work->sync_mode, |
Wu Fengguang | ac69306 | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 663 | .tagged_writepages = work->tagged_writepages, |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 664 | .older_than_this = NULL, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 665 | .for_kupdate = work->for_kupdate, |
| 666 | .for_background = work->for_background, |
| 667 | .range_cyclic = work->range_cyclic, |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 668 | }; |
| 669 | unsigned long oldest_jif; |
| 670 | long wrote = 0; |
Wu Fengguang | ac69306 | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 671 | long write_chunk = MAX_WRITEBACK_PAGES; |
Jan Kara | a5989bd | 2009-09-16 19:22:48 +0200 | [diff] [blame] | 672 | struct inode *inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 673 | |
| 674 | if (wbc.for_kupdate) { |
| 675 | wbc.older_than_this = &oldest_jif; |
| 676 | oldest_jif = jiffies - |
| 677 | msecs_to_jiffies(dirty_expire_interval * 10); |
| 678 | } |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 679 | if (!wbc.range_cyclic) { |
| 680 | wbc.range_start = 0; |
| 681 | wbc.range_end = LLONG_MAX; |
| 682 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 683 | |
Jan Kara | b9543da | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 684 | /* |
| 685 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty |
| 686 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX |
| 687 | * here avoids calling into writeback_inodes_wb() more than once. |
| 688 | * |
| 689 | * The intended call sequence for WB_SYNC_ALL writeback is: |
| 690 | * |
| 691 | * wb_writeback() |
| 692 | * __writeback_inodes_sb() <== called only once |
| 693 | * write_cache_pages() <== called once for each inode |
| 694 | * (quickly) tag currently dirty pages |
| 695 | * (maybe slowly) sync all tagged pages |
| 696 | */ |
Wu Fengguang | ac69306 | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 697 | if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_writepages) |
Jan Kara | b9543da | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 698 | write_chunk = LONG_MAX; |
| 699 | |
Jan Kara | 7624ee7 | 2010-08-09 17:20:03 -0700 | [diff] [blame] | 700 | wbc.wb_start = jiffies; /* livelock avoidance */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 701 | for (;;) { |
| 702 | /* |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 703 | * Stop writeback when nr_pages has been consumed |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 704 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 705 | if (work->nr_pages <= 0) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 706 | break; |
| 707 | |
| 708 | /* |
Jan Kara | aa373cf | 2011-01-13 15:45:47 -0800 | [diff] [blame] | 709 | * Background writeout and kupdate-style writeback may |
| 710 | * run forever. Stop them if there is other work to do |
| 711 | * so that e.g. sync can proceed. They'll be restarted |
| 712 | * after the other works are all done. |
| 713 | */ |
| 714 | if ((work->for_background || work->for_kupdate) && |
| 715 | !list_empty(&wb->bdi->work_list)) |
| 716 | break; |
| 717 | |
| 718 | /* |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 719 | * For background writeout, stop when we are below the |
| 720 | * background dirty threshold |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 721 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 722 | if (work->for_background && !over_bground_thresh()) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 723 | break; |
| 724 | |
| 725 | wbc.more_io = 0; |
Jan Kara | b9543da | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 726 | wbc.nr_to_write = write_chunk; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 727 | wbc.pages_skipped = 0; |
Dave Chinner | 028c2dd | 2010-07-07 13:24:07 +1000 | [diff] [blame] | 728 | |
| 729 | trace_wbc_writeback_start(&wbc, wb->bdi); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 730 | if (work->sb) |
| 731 | __writeback_inodes_sb(work->sb, wb, &wbc); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 732 | else |
| 733 | writeback_inodes_wb(wb, &wbc); |
Dave Chinner | 028c2dd | 2010-07-07 13:24:07 +1000 | [diff] [blame] | 734 | trace_wbc_writeback_written(&wbc, wb->bdi); |
| 735 | |
Jan Kara | b9543da | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 736 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
| 737 | wrote += write_chunk - wbc.nr_to_write; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 738 | |
| 739 | /* |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 740 | * If we consumed everything, see if we have more |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 741 | */ |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 742 | if (wbc.nr_to_write <= 0) |
| 743 | continue; |
| 744 | /* |
| 745 | * Didn't write everything and we don't have more IO, bail |
| 746 | */ |
| 747 | if (!wbc.more_io) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 748 | break; |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 749 | /* |
| 750 | * Did we write something? Try for more |
| 751 | */ |
Jan Kara | b9543da | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 752 | if (wbc.nr_to_write < write_chunk) |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 753 | continue; |
| 754 | /* |
| 755 | * Nothing written. Wait for some inode to |
| 756 | * become available for writeback. Otherwise |
| 757 | * we'll just busyloop. |
| 758 | */ |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 759 | spin_lock(&inode_wb_list_lock); |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 760 | if (!list_empty(&wb->b_more_io)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 761 | inode = wb_inode(wb->b_more_io.prev); |
Dave Chinner | 028c2dd | 2010-07-07 13:24:07 +1000 | [diff] [blame] | 762 | trace_wbc_writeback_wait(&wbc, wb->bdi); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 763 | spin_lock(&inode->i_lock); |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 764 | inode_wait_for_writeback(inode); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 765 | spin_unlock(&inode->i_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 766 | } |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 767 | spin_unlock(&inode_wb_list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 768 | } |
| 769 | |
| 770 | return wrote; |
| 771 | } |
| 772 | |
| 773 | /* |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 774 | * 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] | 775 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 776 | static struct wb_writeback_work * |
Minchan Kim | 08852b6 | 2010-08-03 12:51:16 +0200 | [diff] [blame] | 777 | get_next_work_item(struct backing_dev_info *bdi) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 778 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 779 | struct wb_writeback_work *work = NULL; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 780 | |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 781 | spin_lock_bh(&bdi->wb_lock); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 782 | if (!list_empty(&bdi->work_list)) { |
| 783 | work = list_entry(bdi->work_list.next, |
| 784 | struct wb_writeback_work, list); |
| 785 | list_del_init(&work->list); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 786 | } |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 787 | spin_unlock_bh(&bdi->wb_lock); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 788 | return work; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 789 | } |
| 790 | |
Linus Torvalds | cdf01dd | 2010-10-30 08:55:52 -0700 | [diff] [blame] | 791 | /* |
| 792 | * Add in the number of potentially dirty inodes, because each inode |
| 793 | * write can dirty pagecache in the underlying blockdev. |
| 794 | */ |
| 795 | static unsigned long get_nr_dirty_pages(void) |
| 796 | { |
| 797 | return global_page_state(NR_FILE_DIRTY) + |
| 798 | global_page_state(NR_UNSTABLE_NFS) + |
| 799 | get_nr_dirty_inodes(); |
| 800 | } |
| 801 | |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 802 | static long wb_check_background_flush(struct bdi_writeback *wb) |
| 803 | { |
| 804 | if (over_bground_thresh()) { |
| 805 | |
| 806 | struct wb_writeback_work work = { |
| 807 | .nr_pages = LONG_MAX, |
| 808 | .sync_mode = WB_SYNC_NONE, |
| 809 | .for_background = 1, |
| 810 | .range_cyclic = 1, |
| 811 | }; |
| 812 | |
| 813 | return wb_writeback(wb, &work); |
| 814 | } |
| 815 | |
| 816 | return 0; |
| 817 | } |
| 818 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 819 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
| 820 | { |
| 821 | unsigned long expired; |
| 822 | long nr_pages; |
| 823 | |
Jens Axboe | 69b62d0 | 2010-05-17 12:51:03 +0200 | [diff] [blame] | 824 | /* |
| 825 | * When set to zero, disable periodic writeback |
| 826 | */ |
| 827 | if (!dirty_writeback_interval) |
| 828 | return 0; |
| 829 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 830 | expired = wb->last_old_flush + |
| 831 | msecs_to_jiffies(dirty_writeback_interval * 10); |
| 832 | if (time_before(jiffies, expired)) |
| 833 | return 0; |
| 834 | |
| 835 | wb->last_old_flush = jiffies; |
Linus Torvalds | cdf01dd | 2010-10-30 08:55:52 -0700 | [diff] [blame] | 836 | nr_pages = get_nr_dirty_pages(); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 837 | |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 838 | if (nr_pages) { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 839 | struct wb_writeback_work work = { |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 840 | .nr_pages = nr_pages, |
| 841 | .sync_mode = WB_SYNC_NONE, |
| 842 | .for_kupdate = 1, |
| 843 | .range_cyclic = 1, |
| 844 | }; |
| 845 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 846 | return wb_writeback(wb, &work); |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 847 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 848 | |
| 849 | return 0; |
| 850 | } |
| 851 | |
| 852 | /* |
| 853 | * Retrieve work items and do the writeback they describe |
| 854 | */ |
| 855 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) |
| 856 | { |
| 857 | struct backing_dev_info *bdi = wb->bdi; |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 858 | struct wb_writeback_work *work; |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 859 | long wrote = 0; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 860 | |
Jan Kara | 81d73a3 | 2010-08-11 14:17:44 -0700 | [diff] [blame] | 861 | set_bit(BDI_writeback_running, &wb->bdi->state); |
Minchan Kim | 08852b6 | 2010-08-03 12:51:16 +0200 | [diff] [blame] | 862 | while ((work = get_next_work_item(bdi)) != NULL) { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 863 | /* |
| 864 | * Override sync mode, in case we must wait for completion |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 865 | * because this thread is exiting now. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 866 | */ |
| 867 | if (force_wait) |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 868 | work->sync_mode = WB_SYNC_ALL; |
| 869 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 870 | trace_writeback_exec(bdi, work); |
| 871 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 872 | wrote += wb_writeback(wb, work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 873 | |
| 874 | /* |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 875 | * Notify the caller of completion if this is a synchronous |
| 876 | * work item, otherwise just free it. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 877 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 878 | if (work->done) |
| 879 | complete(work->done); |
| 880 | else |
| 881 | kfree(work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 882 | } |
| 883 | |
| 884 | /* |
| 885 | * Check for periodic writeback, kupdated() style |
| 886 | */ |
| 887 | wrote += wb_check_old_data_flush(wb); |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 888 | wrote += wb_check_background_flush(wb); |
Jan Kara | 81d73a3 | 2010-08-11 14:17:44 -0700 | [diff] [blame] | 889 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 890 | |
| 891 | return wrote; |
| 892 | } |
| 893 | |
| 894 | /* |
| 895 | * Handle writeback of dirty data for the device backed by this bdi. Also |
| 896 | * wakes up periodically and does kupdated style flushing. |
| 897 | */ |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 898 | int bdi_writeback_thread(void *data) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 899 | { |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 900 | struct bdi_writeback *wb = data; |
| 901 | struct backing_dev_info *bdi = wb->bdi; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 902 | long pages_written; |
| 903 | |
Peter Zijlstra | 766f916 | 2010-10-26 14:22:45 -0700 | [diff] [blame] | 904 | current->flags |= PF_SWAPWRITE; |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 905 | set_freezable(); |
Artem Bityutskiy | ecd5840 | 2010-07-25 14:29:18 +0300 | [diff] [blame] | 906 | wb->last_active = jiffies; |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 907 | |
| 908 | /* |
| 909 | * Our parent may run at a different priority, just set us to normal |
| 910 | */ |
| 911 | set_user_nice(current, 0); |
| 912 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 913 | trace_writeback_thread_start(bdi); |
| 914 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 915 | while (!kthread_should_stop()) { |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 916 | /* |
| 917 | * Remove own delayed wake-up timer, since we are already awake |
| 918 | * and we'll take care of the preriodic write-back. |
| 919 | */ |
| 920 | del_timer(&wb->wakeup_timer); |
| 921 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 922 | pages_written = wb_do_writeback(wb, 0); |
| 923 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 924 | trace_writeback_pages_written(pages_written); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 925 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 926 | if (pages_written) |
Artem Bityutskiy | ecd5840 | 2010-07-25 14:29:18 +0300 | [diff] [blame] | 927 | wb->last_active = jiffies; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 928 | |
Artem Bityutskiy | 297252c | 2010-07-25 14:29:15 +0300 | [diff] [blame] | 929 | set_current_state(TASK_INTERRUPTIBLE); |
J. Bruce Fields | b76b401 | 2010-08-28 08:52:10 +0200 | [diff] [blame] | 930 | if (!list_empty(&bdi->work_list) || kthread_should_stop()) { |
Artem Bityutskiy | 297252c | 2010-07-25 14:29:15 +0300 | [diff] [blame] | 931 | __set_current_state(TASK_RUNNING); |
| 932 | continue; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 933 | } |
| 934 | |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 935 | if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 936 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 937 | else { |
| 938 | /* |
| 939 | * We have nothing to do, so can go sleep without any |
| 940 | * timeout and save power. When a work is queued or |
| 941 | * something is made dirty - we will be woken up. |
| 942 | */ |
Artem Bityutskiy | 297252c | 2010-07-25 14:29:15 +0300 | [diff] [blame] | 943 | schedule(); |
Jens Axboe | f9eadbb | 2010-05-18 14:31:45 +0200 | [diff] [blame] | 944 | } |
Jens Axboe | 69b62d0 | 2010-05-17 12:51:03 +0200 | [diff] [blame] | 945 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 946 | try_to_freeze(); |
| 947 | } |
| 948 | |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 949 | /* Flush any work that raced with us exiting */ |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 950 | if (!list_empty(&bdi->work_list)) |
| 951 | wb_do_writeback(wb, 1); |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 952 | |
| 953 | trace_writeback_thread_stop(bdi); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 954 | return 0; |
| 955 | } |
| 956 | |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 957 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 958 | /* |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 959 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
| 960 | * the whole world. |
| 961 | */ |
| 962 | void wakeup_flusher_threads(long nr_pages) |
| 963 | { |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 964 | struct backing_dev_info *bdi; |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 965 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 966 | if (!nr_pages) { |
| 967 | nr_pages = global_page_state(NR_FILE_DIRTY) + |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 968 | global_page_state(NR_UNSTABLE_NFS); |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 969 | } |
| 970 | |
| 971 | rcu_read_lock(); |
| 972 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
| 973 | if (!bdi_has_dirty_io(bdi)) |
| 974 | continue; |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 975 | __bdi_start_writeback(bdi, nr_pages, false); |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 976 | } |
| 977 | rcu_read_unlock(); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 978 | } |
| 979 | |
| 980 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
| 981 | { |
| 982 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { |
| 983 | struct dentry *dentry; |
| 984 | const char *name = "?"; |
| 985 | |
| 986 | dentry = d_find_alias(inode); |
| 987 | if (dentry) { |
| 988 | spin_lock(&dentry->d_lock); |
| 989 | name = (const char *) dentry->d_name.name; |
| 990 | } |
| 991 | printk(KERN_DEBUG |
| 992 | "%s(%d): dirtied inode %lu (%s) on %s\n", |
| 993 | current->comm, task_pid_nr(current), inode->i_ino, |
| 994 | name, inode->i_sb->s_id); |
| 995 | if (dentry) { |
| 996 | spin_unlock(&dentry->d_lock); |
| 997 | dput(dentry); |
| 998 | } |
| 999 | } |
| 1000 | } |
| 1001 | |
| 1002 | /** |
| 1003 | * __mark_inode_dirty - internal function |
| 1004 | * @inode: inode to mark |
| 1005 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) |
| 1006 | * Mark an inode as dirty. Callers should use mark_inode_dirty or |
| 1007 | * mark_inode_dirty_sync. |
| 1008 | * |
| 1009 | * Put the inode on the super block's dirty list. |
| 1010 | * |
| 1011 | * CAREFUL! We mark it dirty unconditionally, but move it onto the |
| 1012 | * dirty list only if it is hashed or if it refers to a blockdev. |
| 1013 | * If it was not hashed, it will never be added to the dirty list |
| 1014 | * even if it is later hashed, as it will have been marked dirty already. |
| 1015 | * |
| 1016 | * In short, make sure you hash any inodes _before_ you start marking |
| 1017 | * them dirty. |
| 1018 | * |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1019 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
| 1020 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of |
| 1021 | * the kernel-internal blockdev inode represents the dirtying time of the |
| 1022 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use |
| 1023 | * page->mapping->host, so the page-dirtying time is recorded in the internal |
| 1024 | * blockdev inode. |
| 1025 | */ |
| 1026 | void __mark_inode_dirty(struct inode *inode, int flags) |
| 1027 | { |
| 1028 | struct super_block *sb = inode->i_sb; |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1029 | struct backing_dev_info *bdi = NULL; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1030 | |
| 1031 | /* |
| 1032 | * Don't do this for I_DIRTY_PAGES - that doesn't actually |
| 1033 | * dirty the inode itself |
| 1034 | */ |
| 1035 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { |
| 1036 | if (sb->s_op->dirty_inode) |
Christoph Hellwig | aa38572 | 2011-05-27 06:53:02 -0400 | [diff] [blame] | 1037 | sb->s_op->dirty_inode(inode, flags); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1038 | } |
| 1039 | |
| 1040 | /* |
| 1041 | * make sure that changes are seen by all cpus before we test i_state |
| 1042 | * -- mikulas |
| 1043 | */ |
| 1044 | smp_mb(); |
| 1045 | |
| 1046 | /* avoid the locking if we can */ |
| 1047 | if ((inode->i_state & flags) == flags) |
| 1048 | return; |
| 1049 | |
| 1050 | if (unlikely(block_dump)) |
| 1051 | block_dump___mark_inode_dirty(inode); |
| 1052 | |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1053 | spin_lock(&inode->i_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1054 | if ((inode->i_state & flags) != flags) { |
| 1055 | const int was_dirty = inode->i_state & I_DIRTY; |
| 1056 | |
| 1057 | inode->i_state |= flags; |
| 1058 | |
| 1059 | /* |
| 1060 | * If the inode is being synced, just update its dirty state. |
| 1061 | * The unlocker will place the inode on the appropriate |
| 1062 | * superblock list, based upon its state. |
| 1063 | */ |
| 1064 | if (inode->i_state & I_SYNC) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1065 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1066 | |
| 1067 | /* |
| 1068 | * Only add valid (hashed) inodes to the superblock's |
| 1069 | * dirty list. Add blockdev inodes as well. |
| 1070 | */ |
| 1071 | if (!S_ISBLK(inode->i_mode)) { |
Al Viro | 1d3382c | 2010-10-23 15:19:20 -0400 | [diff] [blame] | 1072 | if (inode_unhashed(inode)) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1073 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1074 | } |
Al Viro | a4ffdde | 2010-06-02 17:38:30 -0400 | [diff] [blame] | 1075 | if (inode->i_state & I_FREEING) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1076 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1077 | |
| 1078 | /* |
| 1079 | * If the inode was already on b_dirty/b_io/b_more_io, don't |
| 1080 | * reposition it (that would break b_dirty time-ordering). |
| 1081 | */ |
| 1082 | if (!was_dirty) { |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 1083 | bool wakeup_bdi = false; |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1084 | bdi = inode_to_bdi(inode); |
Jens Axboe | 500b067 | 2009-09-09 09:10:25 +0200 | [diff] [blame] | 1085 | |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1086 | if (bdi_cap_writeback_dirty(bdi)) { |
| 1087 | WARN(!test_bit(BDI_registered, &bdi->state), |
| 1088 | "bdi-%s not registered\n", bdi->name); |
| 1089 | |
| 1090 | /* |
| 1091 | * If this is the first dirty inode for this |
| 1092 | * bdi, we have to wake-up the corresponding |
| 1093 | * bdi thread to make sure background |
| 1094 | * write-back happens later. |
| 1095 | */ |
| 1096 | if (!wb_has_dirty_io(&bdi->wb)) |
| 1097 | wakeup_bdi = true; |
Jens Axboe | 500b067 | 2009-09-09 09:10:25 +0200 | [diff] [blame] | 1098 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1099 | |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 1100 | spin_unlock(&inode->i_lock); |
| 1101 | spin_lock(&inode_wb_list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1102 | inode->dirtied_when = jiffies; |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 1103 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 1104 | spin_unlock(&inode_wb_list_lock); |
| 1105 | |
| 1106 | if (wakeup_bdi) |
| 1107 | bdi_wakeup_thread_delayed(bdi); |
| 1108 | return; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1109 | } |
| 1110 | } |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1111 | out_unlock_inode: |
| 1112 | spin_unlock(&inode->i_lock); |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1113 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1114 | } |
| 1115 | EXPORT_SYMBOL(__mark_inode_dirty); |
| 1116 | |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1117 | /* |
| 1118 | * Write out a superblock's list of dirty inodes. A wait will be performed |
| 1119 | * upon no inodes, all inodes or the final one, depending upon sync_mode. |
| 1120 | * |
| 1121 | * If older_than_this is non-NULL, then only write out inodes which |
| 1122 | * had their first dirtying at a time earlier than *older_than_this. |
| 1123 | * |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1124 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
| 1125 | * This function assumes that the blockdev superblock's inodes are backed by |
| 1126 | * a variety of queues, so all inodes are searched. For other superblocks, |
| 1127 | * assume that all inodes are backed by the same queue. |
| 1128 | * |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1129 | * The inodes to be written are parked on bdi->b_io. They are moved back onto |
| 1130 | * bdi->b_dirty as they are selected for writing. This way, none can be missed |
| 1131 | * on the writer throttling path, and we get decent balancing between many |
| 1132 | * throttled threads: we don't want them all piling up on inode_sync_wait. |
| 1133 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1134 | static void wait_sb_inodes(struct super_block *sb) |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1135 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1136 | struct inode *inode, *old_inode = NULL; |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1137 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1138 | /* |
| 1139 | * We need to be protected against the filesystem going from |
| 1140 | * r/o to r/w or vice versa. |
| 1141 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1142 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1143 | |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1144 | spin_lock(&inode_sb_list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1145 | |
| 1146 | /* |
| 1147 | * Data integrity sync. Must wait for all pages under writeback, |
| 1148 | * because there may have been pages dirtied before our sync |
| 1149 | * call, but which had writeout started before we write it out. |
| 1150 | * In which case, the inode may not be on the dirty list, but |
| 1151 | * we still have to wait for that writeout. |
| 1152 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1153 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1154 | struct address_space *mapping = inode->i_mapping; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1155 | |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1156 | spin_lock(&inode->i_lock); |
| 1157 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || |
| 1158 | (mapping->nrpages == 0)) { |
| 1159 | spin_unlock(&inode->i_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1160 | continue; |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1161 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1162 | __iget(inode); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1163 | spin_unlock(&inode->i_lock); |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1164 | spin_unlock(&inode_sb_list_lock); |
| 1165 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1166 | /* |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1167 | * We hold a reference to 'inode' so it couldn't have been |
| 1168 | * removed from s_inodes list while we dropped the |
| 1169 | * inode_sb_list_lock. We cannot iput the inode now as we can |
| 1170 | * be holding the last reference and we cannot iput it under |
| 1171 | * inode_sb_list_lock. So we keep the reference and iput it |
| 1172 | * later. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1173 | */ |
| 1174 | iput(old_inode); |
| 1175 | old_inode = inode; |
| 1176 | |
| 1177 | filemap_fdatawait(mapping); |
| 1178 | |
| 1179 | cond_resched(); |
Nick Piggin | 38f2197 | 2009-01-06 14:40:25 -0800 | [diff] [blame] | 1180 | |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1181 | spin_lock(&inode_sb_list_lock); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1182 | } |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1183 | spin_unlock(&inode_sb_list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1184 | iput(old_inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1185 | } |
| 1186 | |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1187 | /** |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1188 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1189 | * @sb: the superblock |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1190 | * @nr: the number of pages to write |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1191 | * |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1192 | * Start writeback on some inodes on this super_block. No guarantees are made |
| 1193 | * 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] | 1194 | * for IO completion of submitted IO. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1195 | */ |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1196 | void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1197 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1198 | DECLARE_COMPLETION_ONSTACK(done); |
| 1199 | struct wb_writeback_work work = { |
Wu Fengguang | ac69306 | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 1200 | .sb = sb, |
| 1201 | .sync_mode = WB_SYNC_NONE, |
| 1202 | .tagged_writepages = 1, |
| 1203 | .done = &done, |
| 1204 | .nr_pages = nr, |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1205 | }; |
Jens Axboe | 0e3c9a2 | 2010-06-01 11:08:43 +0200 | [diff] [blame] | 1206 | |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1207 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1208 | bdi_queue_work(sb->s_bdi, &work); |
| 1209 | wait_for_completion(&done); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1210 | } |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1211 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
| 1212 | |
| 1213 | /** |
| 1214 | * writeback_inodes_sb - writeback dirty inodes from given super_block |
| 1215 | * @sb: the superblock |
| 1216 | * |
| 1217 | * Start writeback on some inodes on this super_block. No guarantees are made |
| 1218 | * on how many (if any) will be written, and this function does not wait |
| 1219 | * for IO completion of submitted IO. |
| 1220 | */ |
| 1221 | void writeback_inodes_sb(struct super_block *sb) |
| 1222 | { |
Linus Torvalds | 925d169 | 2010-10-30 09:05:48 -0700 | [diff] [blame] | 1223 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages()); |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1224 | } |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1225 | EXPORT_SYMBOL(writeback_inodes_sb); |
| 1226 | |
| 1227 | /** |
Eric Sandeen | 17bd55d | 2009-12-23 07:57:07 -0500 | [diff] [blame] | 1228 | * writeback_inodes_sb_if_idle - start writeback if none underway |
| 1229 | * @sb: the superblock |
| 1230 | * |
| 1231 | * Invoke writeback_inodes_sb if no writeback is currently underway. |
| 1232 | * Returns 1 if writeback was started, 0 if not. |
| 1233 | */ |
| 1234 | int writeback_inodes_sb_if_idle(struct super_block *sb) |
| 1235 | { |
| 1236 | if (!writeback_in_progress(sb->s_bdi)) { |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1237 | down_read(&sb->s_umount); |
Eric Sandeen | 17bd55d | 2009-12-23 07:57:07 -0500 | [diff] [blame] | 1238 | writeback_inodes_sb(sb); |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1239 | up_read(&sb->s_umount); |
Eric Sandeen | 17bd55d | 2009-12-23 07:57:07 -0500 | [diff] [blame] | 1240 | return 1; |
| 1241 | } else |
| 1242 | return 0; |
| 1243 | } |
| 1244 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); |
| 1245 | |
| 1246 | /** |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1247 | * writeback_inodes_sb_if_idle - start writeback if none underway |
| 1248 | * @sb: the superblock |
| 1249 | * @nr: the number of pages to write |
| 1250 | * |
| 1251 | * Invoke writeback_inodes_sb if no writeback is currently underway. |
| 1252 | * Returns 1 if writeback was started, 0 if not. |
| 1253 | */ |
| 1254 | int writeback_inodes_sb_nr_if_idle(struct super_block *sb, |
| 1255 | unsigned long nr) |
| 1256 | { |
| 1257 | if (!writeback_in_progress(sb->s_bdi)) { |
| 1258 | down_read(&sb->s_umount); |
| 1259 | writeback_inodes_sb_nr(sb, nr); |
| 1260 | up_read(&sb->s_umount); |
| 1261 | return 1; |
| 1262 | } else |
| 1263 | return 0; |
| 1264 | } |
| 1265 | EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); |
| 1266 | |
| 1267 | /** |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1268 | * sync_inodes_sb - sync sb inode pages |
| 1269 | * @sb: the superblock |
| 1270 | * |
| 1271 | * This function writes and waits on any dirty inode belonging to this |
Stefan Hajnoczi | cb9ef8d | 2011-01-13 15:47:26 -0800 | [diff] [blame] | 1272 | * super_block. |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1273 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1274 | void sync_inodes_sb(struct super_block *sb) |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1275 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1276 | DECLARE_COMPLETION_ONSTACK(done); |
| 1277 | struct wb_writeback_work work = { |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1278 | .sb = sb, |
| 1279 | .sync_mode = WB_SYNC_ALL, |
| 1280 | .nr_pages = LONG_MAX, |
| 1281 | .range_cyclic = 0, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1282 | .done = &done, |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1283 | }; |
| 1284 | |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1285 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
| 1286 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1287 | bdi_queue_work(sb->s_bdi, &work); |
| 1288 | wait_for_completion(&done); |
| 1289 | |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1290 | wait_sb_inodes(sb); |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1291 | } |
| 1292 | EXPORT_SYMBOL(sync_inodes_sb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1293 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1294 | /** |
Andrea Arcangeli | 7f04c26 | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 1295 | * write_inode_now - write an inode to disk |
| 1296 | * @inode: inode to write to disk |
| 1297 | * @sync: whether the write should be synchronous or not |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1298 | * |
Andrea Arcangeli | 7f04c26 | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 1299 | * This function commits an inode to disk immediately if it is dirty. This is |
| 1300 | * primarily needed by knfsd. |
| 1301 | * |
| 1302 | * 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] | 1303 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1304 | int write_inode_now(struct inode *inode, int sync) |
| 1305 | { |
| 1306 | int ret; |
| 1307 | struct writeback_control wbc = { |
| 1308 | .nr_to_write = LONG_MAX, |
Mike Galbraith | 18914b1 | 2008-02-08 04:20:23 -0800 | [diff] [blame] | 1309 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 1310 | .range_start = 0, |
| 1311 | .range_end = LLONG_MAX, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1312 | }; |
| 1313 | |
| 1314 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) |
Andrew Morton | 49364ce | 2005-11-07 00:59:15 -0800 | [diff] [blame] | 1315 | wbc.nr_to_write = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1316 | |
| 1317 | might_sleep(); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 1318 | spin_lock(&inode_wb_list_lock); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 1319 | spin_lock(&inode->i_lock); |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 1320 | ret = writeback_single_inode(inode, &wbc); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 1321 | spin_unlock(&inode->i_lock); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 1322 | spin_unlock(&inode_wb_list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1323 | if (sync) |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 1324 | inode_sync_wait(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1325 | return ret; |
| 1326 | } |
| 1327 | EXPORT_SYMBOL(write_inode_now); |
| 1328 | |
| 1329 | /** |
| 1330 | * sync_inode - write an inode and its pages to disk. |
| 1331 | * @inode: the inode to sync |
| 1332 | * @wbc: controls the writeback mode |
| 1333 | * |
| 1334 | * sync_inode() will write an inode and its pages to disk. It will also |
| 1335 | * correctly update the inode on its superblock's dirty inode lists and will |
| 1336 | * update inode->i_state. |
| 1337 | * |
| 1338 | * The caller must have a ref on the inode. |
| 1339 | */ |
| 1340 | int sync_inode(struct inode *inode, struct writeback_control *wbc) |
| 1341 | { |
| 1342 | int ret; |
| 1343 | |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 1344 | spin_lock(&inode_wb_list_lock); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 1345 | spin_lock(&inode->i_lock); |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 1346 | ret = writeback_single_inode(inode, wbc); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 1347 | spin_unlock(&inode->i_lock); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 1348 | spin_unlock(&inode_wb_list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1349 | return ret; |
| 1350 | } |
| 1351 | EXPORT_SYMBOL(sync_inode); |
Christoph Hellwig | c376501 | 2010-10-06 10:48:20 +0200 | [diff] [blame] | 1352 | |
| 1353 | /** |
Andrew Morton | c691b9d | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 1354 | * sync_inode_metadata - write an inode to disk |
Christoph Hellwig | c376501 | 2010-10-06 10:48:20 +0200 | [diff] [blame] | 1355 | * @inode: the inode to sync |
| 1356 | * @wait: wait for I/O to complete. |
| 1357 | * |
Andrew Morton | c691b9d | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 1358 | * Write an inode to disk and adjust its dirty state after completion. |
Christoph Hellwig | c376501 | 2010-10-06 10:48:20 +0200 | [diff] [blame] | 1359 | * |
| 1360 | * Note: only writes the actual inode, no associated data or other metadata. |
| 1361 | */ |
| 1362 | int sync_inode_metadata(struct inode *inode, int wait) |
| 1363 | { |
| 1364 | struct writeback_control wbc = { |
| 1365 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, |
| 1366 | .nr_to_write = 0, /* metadata-only */ |
| 1367 | }; |
| 1368 | |
| 1369 | return sync_inode(inode, &wbc); |
| 1370 | } |
| 1371 | EXPORT_SYMBOL(sync_inode_metadata); |