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> |
Paul Gortmaker | 630d9c4 | 2011-11-16 23:57:37 -0500 | [diff] [blame] | 17 | #include <linux/export.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> |
Wu Fengguang | bc31b86 | 2012-01-07 20:41:55 -0600 | [diff] [blame] | 23 | #include <linux/pagemap.h> |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 24 | #include <linux/kthread.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> |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 28 | #include <linux/tracepoint.h> |
Al Viro | 719ea2f | 2013-09-29 11:24:49 -0400 | [diff] [blame] | 29 | #include <linux/device.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 | /* |
Wu Fengguang | bc31b86 | 2012-01-07 20:41:55 -0600 | [diff] [blame] | 33 | * 4MB minimal write chunk size |
| 34 | */ |
| 35 | #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) |
| 36 | |
| 37 | /* |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 38 | * Passed into wb_writeback(), essentially a subset of writeback_control |
| 39 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 40 | struct wb_writeback_work { |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 41 | long nr_pages; |
| 42 | struct super_block *sb; |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 43 | unsigned long *older_than_this; |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 44 | enum writeback_sync_modes sync_mode; |
Wu Fengguang | 6e6938b | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 45 | unsigned int tagged_writepages:1; |
H Hartley Sweeten | 52957fe | 2010-04-01 20:36:30 -0500 | [diff] [blame] | 46 | unsigned int for_kupdate:1; |
| 47 | unsigned int range_cyclic:1; |
| 48 | unsigned int for_background:1; |
Dave Chinner | 7747bd4 | 2013-07-02 22:38:35 +1000 | [diff] [blame] | 49 | unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 50 | enum wb_reason reason; /* why was writeback initiated? */ |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 51 | |
Jens Axboe | 8010c3b | 2009-09-15 20:04:57 +0200 | [diff] [blame] | 52 | struct list_head list; /* pending work list */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 53 | struct completion *done; /* set if the caller waits */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 54 | }; |
| 55 | |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 56 | /** |
| 57 | * writeback_in_progress - determine whether there is writeback in progress |
| 58 | * @bdi: the device's backing_dev_info structure. |
| 59 | * |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 60 | * Determine whether there is writeback waiting to be handled against a |
| 61 | * backing device. |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 62 | */ |
| 63 | int writeback_in_progress(struct backing_dev_info *bdi) |
| 64 | { |
Jan Kara | 81d73a3 | 2010-08-11 14:17:44 -0700 | [diff] [blame] | 65 | return test_bit(BDI_writeback_running, &bdi->state); |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 66 | } |
Theodore Ts'o | 00d4e73 | 2012-09-19 22:42:36 -0400 | [diff] [blame] | 67 | EXPORT_SYMBOL(writeback_in_progress); |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 68 | |
Jan Kara | 692ebd1 | 2010-09-21 11:51:01 +0200 | [diff] [blame] | 69 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
| 70 | { |
| 71 | struct super_block *sb = inode->i_sb; |
Jan Kara | 692ebd1 | 2010-09-21 11:51:01 +0200 | [diff] [blame] | 72 | |
Jan Kara | a885599 | 2013-07-09 22:36:45 +0800 | [diff] [blame] | 73 | if (sb_is_blkdev_sb(sb)) |
Christoph Hellwig | aaead25 | 2010-10-04 14:25:33 +0200 | [diff] [blame] | 74 | return inode->i_mapping->backing_dev_info; |
| 75 | |
| 76 | return sb->s_bdi; |
Jan Kara | 692ebd1 | 2010-09-21 11:51:01 +0200 | [diff] [blame] | 77 | } |
| 78 | |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 79 | static inline struct inode *wb_inode(struct list_head *head) |
| 80 | { |
| 81 | return list_entry(head, struct inode, i_wb_list); |
| 82 | } |
| 83 | |
Wu Fengguang | 15eb77a | 2012-01-17 11:18:56 -0600 | [diff] [blame] | 84 | /* |
| 85 | * Include the creation of the trace points after defining the |
| 86 | * wb_writeback_work structure and inline functions so that the definition |
| 87 | * remains local to this file. |
| 88 | */ |
| 89 | #define CREATE_TRACE_POINTS |
| 90 | #include <trace/events/writeback.h> |
| 91 | |
Steven Whitehouse | 774016b | 2014-02-06 15:47:47 +0000 | [diff] [blame] | 92 | EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); |
| 93 | |
Jan Kara | 5acda9d | 2014-04-03 14:46:23 -0700 | [diff] [blame] | 94 | static void bdi_wakeup_thread(struct backing_dev_info *bdi) |
| 95 | { |
| 96 | spin_lock_bh(&bdi->wb_lock); |
| 97 | if (test_bit(BDI_registered, &bdi->state)) |
| 98 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); |
| 99 | spin_unlock_bh(&bdi->wb_lock); |
| 100 | } |
| 101 | |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 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); |
Jan Kara | 5acda9d | 2014-04-03 14:46:23 -0700 | [diff] [blame] | 108 | if (!test_bit(BDI_registered, &bdi->state)) { |
| 109 | if (work->done) |
| 110 | complete(work->done); |
| 111 | goto out_unlock; |
| 112 | } |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 113 | list_add_tail(&work->list, &bdi->work_list); |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 114 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); |
Jan Kara | 5acda9d | 2014-04-03 14:46:23 -0700 | [diff] [blame] | 115 | out_unlock: |
| 116 | spin_unlock_bh(&bdi->wb_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 117 | } |
| 118 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 119 | static void |
| 120 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 121 | bool range_cyclic, enum wb_reason reason) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 122 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 123 | struct wb_writeback_work *work; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 124 | |
Jens Axboe | bcddc3f | 2009-09-13 20:07:36 +0200 | [diff] [blame] | 125 | /* |
| 126 | * This is WB_SYNC_NONE writeback, so if allocation fails just |
| 127 | * wakeup the thread for old dirty data writeback |
| 128 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 129 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
| 130 | if (!work) { |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 131 | trace_writeback_nowork(bdi); |
Jan Kara | 5acda9d | 2014-04-03 14:46:23 -0700 | [diff] [blame] | 132 | bdi_wakeup_thread(bdi); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 133 | return; |
Jens Axboe | bcddc3f | 2009-09-13 20:07:36 +0200 | [diff] [blame] | 134 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 135 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 136 | work->sync_mode = WB_SYNC_NONE; |
| 137 | work->nr_pages = nr_pages; |
| 138 | work->range_cyclic = range_cyclic; |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 139 | work->reason = reason; |
Christoph Hellwig | f0fad8a | 2009-09-11 09:47:56 +0200 | [diff] [blame] | 140 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 141 | bdi_queue_work(bdi, work); |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 142 | } |
| 143 | |
| 144 | /** |
| 145 | * bdi_start_writeback - start writeback |
| 146 | * @bdi: the backing device to write from |
| 147 | * @nr_pages: the number of pages to write |
Marcos Paulo de Souza | 786228a | 2011-11-23 20:56:45 +0800 | [diff] [blame] | 148 | * @reason: reason why some writeback work was initiated |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 149 | * |
| 150 | * Description: |
| 151 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 152 | * started when this function returns, we make no guarantees on |
Jens Axboe | 0e3c9a2 | 2010-06-01 11:08:43 +0200 | [diff] [blame] | 153 | * completion. Caller need not hold sb s_umount semaphore. |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 154 | * |
| 155 | */ |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 156 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
| 157 | enum wb_reason reason) |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 158 | { |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 159 | __bdi_start_writeback(bdi, nr_pages, true, reason); |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 160 | } |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 161 | |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 162 | /** |
| 163 | * bdi_start_background_writeback - start background writeback |
| 164 | * @bdi: the backing device to write from |
| 165 | * |
| 166 | * Description: |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 167 | * This makes sure WB_SYNC_NONE background writeback happens. When |
| 168 | * this function returns, it is only guaranteed that for given BDI |
| 169 | * some IO is happening if we are over background dirty threshold. |
| 170 | * Caller need not hold sb s_umount semaphore. |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 171 | */ |
| 172 | void bdi_start_background_writeback(struct backing_dev_info *bdi) |
| 173 | { |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 174 | /* |
| 175 | * We just wake up the flusher thread. It will perform background |
| 176 | * writeback as soon as there is no other work to do. |
| 177 | */ |
Wu Fengguang | 71927e8 | 2011-01-13 15:45:46 -0800 | [diff] [blame] | 178 | trace_writeback_wake_background(bdi); |
Jan Kara | 5acda9d | 2014-04-03 14:46:23 -0700 | [diff] [blame] | 179 | bdi_wakeup_thread(bdi); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | } |
| 181 | |
| 182 | /* |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 183 | * Remove the inode from the writeback list it is on. |
| 184 | */ |
| 185 | void inode_wb_list_del(struct inode *inode) |
| 186 | { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 187 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 188 | |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 189 | spin_lock(&bdi->wb.list_lock); |
| 190 | list_del_init(&inode->i_wb_list); |
| 191 | spin_unlock(&bdi->wb.list_lock); |
| 192 | } |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 193 | |
| 194 | /* |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 195 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the |
| 196 | * furthest end of its superblock's dirty-inode list. |
| 197 | * |
| 198 | * 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] | 199 | * 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] | 200 | * the case then the inode must have been redirtied while it was being written |
| 201 | * out and we don't reset its dirtied_when. |
| 202 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 203 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 204 | { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 205 | assert_spin_locked(&wb->list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 206 | if (!list_empty(&wb->b_dirty)) { |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 207 | struct inode *tail; |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 208 | |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 209 | tail = wb_inode(wb->b_dirty.next); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 210 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 211 | inode->dirtied_when = jiffies; |
| 212 | } |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 213 | list_move(&inode->i_wb_list, &wb->b_dirty); |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 214 | } |
| 215 | |
| 216 | /* |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 217 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
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 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 220 | { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 221 | assert_spin_locked(&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 | { |
Jan Kara | 365b94ae | 2012-05-03 14:47:55 +0200 | [diff] [blame] | 227 | inode->i_state &= ~I_SYNC; |
Jan Kara | 4eff96d | 2012-11-26 16:29:51 -0800 | [diff] [blame] | 228 | /* If inode is clean an unused, put it into LRU now... */ |
| 229 | inode_add_lru(inode); |
Jan Kara | 365b94ae | 2012-05-03 14:47:55 +0200 | [diff] [blame] | 230 | /* Waiters must see I_SYNC cleared before being woken up */ |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 231 | smp_mb(); |
| 232 | wake_up_bit(&inode->i_state, __I_SYNC); |
| 233 | } |
| 234 | |
Jeff Layton | d2caa3c5 | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 235 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
| 236 | { |
| 237 | bool ret = time_after(inode->dirtied_when, t); |
| 238 | #ifndef CONFIG_64BIT |
| 239 | /* |
| 240 | * For inodes being constantly redirtied, dirtied_when can get stuck. |
| 241 | * It _appears_ to be in the future, but is actually in distant past. |
| 242 | * This test is necessary to prevent such wrapped-around relative times |
Jens Axboe | 5b0830c | 2009-09-23 19:37:09 +0200 | [diff] [blame] | 243 | * from permanently stopping the whole bdi writeback. |
Jeff Layton | d2caa3c5 | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 244 | */ |
| 245 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); |
| 246 | #endif |
| 247 | return ret; |
| 248 | } |
| 249 | |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 250 | #define EXPIRE_DIRTY_ATIME 0x0001 |
| 251 | |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 252 | /* |
Wang Sheng-Hui | 0e2f2b2 | 2012-09-11 08:28:18 +0800 | [diff] [blame] | 253 | * Move expired (dirtied before work->older_than_this) dirty inodes from |
Jan Kara | 697e6fe | 2012-03-09 07:26:22 -0800 | [diff] [blame] | 254 | * @delaying_queue to @dispatch_queue. |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 255 | */ |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 256 | static int move_expired_inodes(struct list_head *delaying_queue, |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 257 | struct list_head *dispatch_queue, |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 258 | int flags, |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 259 | struct wb_writeback_work *work) |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 260 | { |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 261 | unsigned long *older_than_this = NULL; |
| 262 | unsigned long expire_time; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 263 | LIST_HEAD(tmp); |
| 264 | struct list_head *pos, *node; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 265 | struct super_block *sb = NULL; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 266 | struct inode *inode; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 267 | int do_sb_sort = 0; |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 268 | int moved = 0; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 269 | |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 270 | if ((flags & EXPIRE_DIRTY_ATIME) == 0) |
| 271 | older_than_this = work->older_than_this; |
| 272 | else if ((work->reason == WB_REASON_SYNC) == 0) { |
| 273 | expire_time = jiffies - (HZ * 86400); |
| 274 | older_than_this = &expire_time; |
| 275 | } |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 276 | while (!list_empty(delaying_queue)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 277 | inode = wb_inode(delaying_queue->prev); |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 278 | if (older_than_this && |
| 279 | inode_dirtied_after(inode, *older_than_this)) |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 280 | break; |
Jan Kara | a885599 | 2013-07-09 22:36:45 +0800 | [diff] [blame] | 281 | list_move(&inode->i_wb_list, &tmp); |
| 282 | moved++; |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 283 | if (flags & EXPIRE_DIRTY_ATIME) |
| 284 | set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); |
Jan Kara | a885599 | 2013-07-09 22:36:45 +0800 | [diff] [blame] | 285 | if (sb_is_blkdev_sb(inode->i_sb)) |
| 286 | continue; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 287 | if (sb && sb != inode->i_sb) |
| 288 | do_sb_sort = 1; |
| 289 | sb = inode->i_sb; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 290 | } |
| 291 | |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 292 | /* just one sb in list, splice to dispatch_queue and we're done */ |
| 293 | if (!do_sb_sort) { |
| 294 | list_splice(&tmp, dispatch_queue); |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 295 | goto out; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 296 | } |
| 297 | |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 298 | /* Move inodes from one superblock together */ |
| 299 | while (!list_empty(&tmp)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 300 | sb = wb_inode(tmp.prev)->i_sb; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 301 | list_for_each_prev_safe(pos, node, &tmp) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 302 | inode = wb_inode(pos); |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 303 | if (inode->i_sb == sb) |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 304 | list_move(&inode->i_wb_list, dispatch_queue); |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 305 | } |
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 | out: |
| 308 | return moved; |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 309 | } |
| 310 | |
| 311 | /* |
| 312 | * Queue all expired dirty inodes for io, eldest first. |
Wu Fengguang | 4ea879b | 2010-08-11 14:17:42 -0700 | [diff] [blame] | 313 | * Before |
| 314 | * newly dirtied b_dirty b_io b_more_io |
| 315 | * =============> gf edc BA |
| 316 | * After |
| 317 | * newly dirtied b_dirty b_io b_more_io |
| 318 | * =============> g fBAedc |
| 319 | * | |
| 320 | * +--> dequeue for IO |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 321 | */ |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 322 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 323 | { |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 324 | int moved; |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 325 | |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 326 | assert_spin_locked(&wb->list_lock); |
Wu Fengguang | 4ea879b | 2010-08-11 14:17:42 -0700 | [diff] [blame] | 327 | list_splice_init(&wb->b_more_io, &wb->b_io); |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 328 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); |
| 329 | moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, |
| 330 | EXPIRE_DIRTY_ATIME, work); |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 331 | trace_writeback_queue_io(wb, work, moved); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 332 | } |
| 333 | |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 334 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 335 | { |
Tejun Heo | 9fb0a7d | 2013-01-11 13:06:37 -0800 | [diff] [blame] | 336 | int ret; |
| 337 | |
| 338 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { |
| 339 | trace_writeback_write_inode_start(inode, wbc); |
| 340 | ret = inode->i_sb->s_op->write_inode(inode, wbc); |
| 341 | trace_writeback_write_inode(inode, wbc); |
| 342 | return ret; |
| 343 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 344 | return 0; |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 345 | } |
| 346 | |
| 347 | /* |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 348 | * Wait for writeback on an inode to complete. Called with i_lock held. |
| 349 | * Caller must make sure inode cannot go away when we drop i_lock. |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 350 | */ |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 351 | static void __inode_wait_for_writeback(struct inode *inode) |
| 352 | __releases(inode->i_lock) |
| 353 | __acquires(inode->i_lock) |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 354 | { |
| 355 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); |
| 356 | wait_queue_head_t *wqh; |
| 357 | |
| 358 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 359 | while (inode->i_state & I_SYNC) { |
| 360 | spin_unlock(&inode->i_lock); |
NeilBrown | 7431620 | 2014-07-07 15:16:04 +1000 | [diff] [blame] | 361 | __wait_on_bit(wqh, &wq, bit_wait, |
| 362 | TASK_UNINTERRUPTIBLE); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 363 | spin_lock(&inode->i_lock); |
Richard Kennedy | 58a9d3d8 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 364 | } |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 365 | } |
| 366 | |
| 367 | /* |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 368 | * Wait for writeback on an inode to complete. Caller must have inode pinned. |
| 369 | */ |
| 370 | void inode_wait_for_writeback(struct inode *inode) |
| 371 | { |
| 372 | spin_lock(&inode->i_lock); |
| 373 | __inode_wait_for_writeback(inode); |
| 374 | spin_unlock(&inode->i_lock); |
| 375 | } |
| 376 | |
| 377 | /* |
| 378 | * Sleep until I_SYNC is cleared. This function must be called with i_lock |
| 379 | * held and drops it. It is aimed for callers not holding any inode reference |
| 380 | * so once i_lock is dropped, inode can go away. |
| 381 | */ |
| 382 | static void inode_sleep_on_writeback(struct inode *inode) |
| 383 | __releases(inode->i_lock) |
| 384 | { |
| 385 | DEFINE_WAIT(wait); |
| 386 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); |
| 387 | int sleep; |
| 388 | |
| 389 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); |
| 390 | sleep = inode->i_state & I_SYNC; |
| 391 | spin_unlock(&inode->i_lock); |
| 392 | if (sleep) |
| 393 | schedule(); |
| 394 | finish_wait(wqh, &wait); |
| 395 | } |
| 396 | |
| 397 | /* |
Jan Kara | ccb26b5 | 2012-05-03 14:47:58 +0200 | [diff] [blame] | 398 | * Find proper writeback list for the inode depending on its current state and |
| 399 | * possibly also change of its state while we were doing writeback. Here we |
| 400 | * handle things such as livelock prevention or fairness of writeback among |
| 401 | * inodes. This function can be called only by flusher thread - noone else |
| 402 | * processes all inodes in writeback lists and requeueing inodes behind flusher |
| 403 | * thread's back can have unexpected consequences. |
| 404 | */ |
| 405 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, |
| 406 | struct writeback_control *wbc) |
| 407 | { |
| 408 | if (inode->i_state & I_FREEING) |
| 409 | return; |
| 410 | |
| 411 | /* |
| 412 | * Sync livelock prevention. Each inode is tagged and synced in one |
| 413 | * shot. If still dirty, it will be redirty_tail()'ed below. Update |
| 414 | * the dirty time to prevent enqueue and sync it again. |
| 415 | */ |
| 416 | if ((inode->i_state & I_DIRTY) && |
| 417 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) |
| 418 | inode->dirtied_when = jiffies; |
| 419 | |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 420 | if (wbc->pages_skipped) { |
| 421 | /* |
| 422 | * writeback is not making progress due to locked |
| 423 | * buffers. Skip this inode for now. |
| 424 | */ |
| 425 | redirty_tail(inode, wb); |
| 426 | return; |
| 427 | } |
| 428 | |
Jan Kara | ccb26b5 | 2012-05-03 14:47:58 +0200 | [diff] [blame] | 429 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
| 430 | /* |
| 431 | * We didn't write back all the pages. nfs_writepages() |
| 432 | * sometimes bales out without doing anything. |
| 433 | */ |
| 434 | if (wbc->nr_to_write <= 0) { |
| 435 | /* Slice used up. Queue for next turn. */ |
| 436 | requeue_io(inode, wb); |
| 437 | } else { |
| 438 | /* |
| 439 | * Writeback blocked by something other than |
| 440 | * congestion. Delay the inode for some time to |
| 441 | * avoid spinning on the CPU (100% iowait) |
| 442 | * retrying writeback of the dirty page/inode |
| 443 | * that cannot be performed immediately. |
| 444 | */ |
| 445 | redirty_tail(inode, wb); |
| 446 | } |
| 447 | } else if (inode->i_state & I_DIRTY) { |
| 448 | /* |
| 449 | * Filesystems can dirty the inode during writeback operations, |
| 450 | * such as delayed allocation during submission or metadata |
| 451 | * updates after data IO completion. |
| 452 | */ |
| 453 | redirty_tail(inode, wb); |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 454 | } else if (inode->i_state & I_DIRTY_TIME) { |
| 455 | list_move(&inode->i_wb_list, &wb->b_dirty_time); |
Jan Kara | ccb26b5 | 2012-05-03 14:47:58 +0200 | [diff] [blame] | 456 | } else { |
| 457 | /* The inode is clean. Remove from writeback lists. */ |
| 458 | list_del_init(&inode->i_wb_list); |
| 459 | } |
| 460 | } |
| 461 | |
| 462 | /* |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 463 | * Write out an inode and its dirty pages. Do not update the writeback list |
| 464 | * linkage. That is left to the caller. The caller is also responsible for |
| 465 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 466 | */ |
| 467 | static int |
Yan Hong | cd8ed2a | 2012-10-08 16:33:45 -0700 | [diff] [blame] | 468 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 469 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | struct address_space *mapping = inode->i_mapping; |
Wu Fengguang | 251d6a4 | 2010-12-01 17:33:37 -0600 | [diff] [blame] | 471 | long nr_to_write = wbc->nr_to_write; |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 472 | unsigned dirty; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 473 | int ret; |
| 474 | |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 475 | WARN_ON(!(inode->i_state & I_SYNC)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 476 | |
Tejun Heo | 9fb0a7d | 2013-01-11 13:06:37 -0800 | [diff] [blame] | 477 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); |
| 478 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 479 | ret = do_writepages(mapping, wbc); |
| 480 | |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 481 | /* |
| 482 | * Make sure to wait on the data before writing out the metadata. |
| 483 | * This is important for filesystems that modify metadata on data |
Dave Chinner | 7747bd4 | 2013-07-02 22:38:35 +1000 | [diff] [blame] | 484 | * I/O completion. We don't do it for sync(2) writeback because it has a |
| 485 | * separate, external IO completion path and ->sync_fs for guaranteeing |
| 486 | * inode metadata is written back correctly. |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 487 | */ |
Dave Chinner | 7747bd4 | 2013-07-02 22:38:35 +1000 | [diff] [blame] | 488 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 489 | int err = filemap_fdatawait(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 490 | if (ret == 0) |
| 491 | ret = err; |
| 492 | } |
| 493 | |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 494 | /* |
| 495 | * Some filesystems may redirty the inode during the writeback |
| 496 | * due to delalloc, clear dirty metadata flags right before |
| 497 | * write_inode() |
| 498 | */ |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 499 | spin_lock(&inode->i_lock); |
Tejun Heo | 9c6ac78 | 2014-10-24 15:38:21 -0400 | [diff] [blame] | 500 | |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 501 | dirty = inode->i_state & I_DIRTY; |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 502 | if (((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) && |
| 503 | (inode->i_state & I_DIRTY_TIME)) || |
| 504 | (inode->i_state & I_DIRTY_TIME_EXPIRED)) { |
| 505 | dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; |
| 506 | trace_writeback_lazytime(inode); |
| 507 | } |
| 508 | inode->i_state &= ~dirty; |
Tejun Heo | 9c6ac78 | 2014-10-24 15:38:21 -0400 | [diff] [blame] | 509 | |
| 510 | /* |
| 511 | * Paired with smp_mb() in __mark_inode_dirty(). This allows |
| 512 | * __mark_inode_dirty() to test i_state without grabbing i_lock - |
| 513 | * either they see the I_DIRTY bits cleared or we see the dirtied |
| 514 | * inode. |
| 515 | * |
| 516 | * I_DIRTY_PAGES is always cleared together above even if @mapping |
| 517 | * still has dirty pages. The flag is reinstated after smp_mb() if |
| 518 | * necessary. This guarantees that either __mark_inode_dirty() |
| 519 | * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. |
| 520 | */ |
| 521 | smp_mb(); |
| 522 | |
| 523 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
| 524 | inode->i_state |= I_DIRTY_PAGES; |
| 525 | |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 526 | spin_unlock(&inode->i_lock); |
Tejun Heo | 9c6ac78 | 2014-10-24 15:38:21 -0400 | [diff] [blame] | 527 | |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 528 | if (dirty & I_DIRTY_TIME) |
| 529 | mark_inode_dirty_sync(inode); |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 530 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 531 | if (dirty & ~I_DIRTY_PAGES) { |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 532 | int err = write_inode(inode, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 533 | if (ret == 0) |
| 534 | ret = err; |
| 535 | } |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 536 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
| 537 | return ret; |
| 538 | } |
| 539 | |
| 540 | /* |
| 541 | * Write out an inode's dirty pages. Either the caller has an active reference |
| 542 | * on the inode or the inode has I_WILL_FREE set. |
| 543 | * |
| 544 | * This function is designed to be called for writing back one inode which |
| 545 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() |
| 546 | * and does more profound writeback list handling in writeback_sb_inodes(). |
| 547 | */ |
| 548 | static int |
| 549 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, |
| 550 | struct writeback_control *wbc) |
| 551 | { |
| 552 | int ret = 0; |
| 553 | |
| 554 | spin_lock(&inode->i_lock); |
| 555 | if (!atomic_read(&inode->i_count)) |
| 556 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); |
| 557 | else |
| 558 | WARN_ON(inode->i_state & I_WILL_FREE); |
| 559 | |
| 560 | if (inode->i_state & I_SYNC) { |
| 561 | if (wbc->sync_mode != WB_SYNC_ALL) |
| 562 | goto out; |
| 563 | /* |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 564 | * It's a data-integrity sync. We must wait. Since callers hold |
| 565 | * inode reference or inode has I_WILL_FREE set, it cannot go |
| 566 | * away under us. |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 567 | */ |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 568 | __inode_wait_for_writeback(inode); |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 569 | } |
| 570 | WARN_ON(inode->i_state & I_SYNC); |
| 571 | /* |
Jan Kara | f9b0e05 | 2013-12-14 04:21:26 +0800 | [diff] [blame] | 572 | * Skip inode if it is clean and we have no outstanding writeback in |
| 573 | * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this |
| 574 | * function since flusher thread may be doing for example sync in |
| 575 | * parallel and if we move the inode, it could get skipped. So here we |
| 576 | * make sure inode is on some writeback list and leave it there unless |
| 577 | * we have completely cleaned the inode. |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 578 | */ |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 579 | if (!(inode->i_state & I_DIRTY_ALL) && |
Jan Kara | f9b0e05 | 2013-12-14 04:21:26 +0800 | [diff] [blame] | 580 | (wbc->sync_mode != WB_SYNC_ALL || |
| 581 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 582 | goto out; |
| 583 | inode->i_state |= I_SYNC; |
| 584 | spin_unlock(&inode->i_lock); |
| 585 | |
Yan Hong | cd8ed2a | 2012-10-08 16:33:45 -0700 | [diff] [blame] | 586 | ret = __writeback_single_inode(inode, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 587 | |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 588 | spin_lock(&wb->list_lock); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 589 | spin_lock(&inode->i_lock); |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 590 | /* |
| 591 | * If inode is clean, remove it from writeback lists. Otherwise don't |
| 592 | * touch it. See comment above for explanation. |
| 593 | */ |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 594 | if (!(inode->i_state & I_DIRTY_ALL)) |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 595 | list_del_init(&inode->i_wb_list); |
| 596 | spin_unlock(&wb->list_lock); |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 597 | inode_sync_complete(inode); |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 598 | out: |
| 599 | spin_unlock(&inode->i_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 600 | return ret; |
| 601 | } |
| 602 | |
Wu Fengguang | 1a12d8b | 2010-08-29 13:28:09 -0600 | [diff] [blame] | 603 | static long writeback_chunk_size(struct backing_dev_info *bdi, |
| 604 | struct wb_writeback_work *work) |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 605 | { |
| 606 | long pages; |
| 607 | |
| 608 | /* |
| 609 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty |
| 610 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX |
| 611 | * here avoids calling into writeback_inodes_wb() more than once. |
| 612 | * |
| 613 | * The intended call sequence for WB_SYNC_ALL writeback is: |
| 614 | * |
| 615 | * wb_writeback() |
| 616 | * writeback_sb_inodes() <== called only once |
| 617 | * write_cache_pages() <== called once for each inode |
| 618 | * (quickly) tag currently dirty pages |
| 619 | * (maybe slowly) sync all tagged pages |
| 620 | */ |
| 621 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) |
| 622 | pages = LONG_MAX; |
Wu Fengguang | 1a12d8b | 2010-08-29 13:28:09 -0600 | [diff] [blame] | 623 | else { |
| 624 | pages = min(bdi->avg_write_bandwidth / 2, |
| 625 | global_dirty_limit / DIRTY_SCOPE); |
| 626 | pages = min(pages, work->nr_pages); |
| 627 | pages = round_down(pages + MIN_WRITEBACK_PAGES, |
| 628 | MIN_WRITEBACK_PAGES); |
| 629 | } |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 630 | |
| 631 | return pages; |
| 632 | } |
| 633 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 634 | /* |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 635 | * Write a portion of b_io inodes which belong to @sb. |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 636 | * |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 637 | * Return the number of pages and/or inodes written. |
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 | static long writeback_sb_inodes(struct super_block *sb, |
| 640 | struct bdi_writeback *wb, |
| 641 | struct wb_writeback_work *work) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 642 | { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 643 | struct writeback_control wbc = { |
| 644 | .sync_mode = work->sync_mode, |
| 645 | .tagged_writepages = work->tagged_writepages, |
| 646 | .for_kupdate = work->for_kupdate, |
| 647 | .for_background = work->for_background, |
Dave Chinner | 7747bd4 | 2013-07-02 22:38:35 +1000 | [diff] [blame] | 648 | .for_sync = work->for_sync, |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 649 | .range_cyclic = work->range_cyclic, |
| 650 | .range_start = 0, |
| 651 | .range_end = LLONG_MAX, |
| 652 | }; |
| 653 | unsigned long start_time = jiffies; |
| 654 | long write_chunk; |
| 655 | long wrote = 0; /* count both pages and inodes */ |
| 656 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 657 | while (!list_empty(&wb->b_io)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 658 | struct inode *inode = wb_inode(wb->b_io.prev); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 659 | |
| 660 | if (inode->i_sb != sb) { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 661 | if (work->sb) { |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 662 | /* |
| 663 | * We only want to write back data for this |
| 664 | * superblock, move all inodes not belonging |
| 665 | * to it back onto the dirty list. |
| 666 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 667 | redirty_tail(inode, wb); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 668 | continue; |
| 669 | } |
| 670 | |
| 671 | /* |
| 672 | * The inode belongs to a different superblock. |
| 673 | * Bounce back to the caller to unpin this and |
| 674 | * pin the next superblock. |
| 675 | */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 676 | break; |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 677 | } |
| 678 | |
Christoph Hellwig | 9843b76 | 2010-10-24 19:40:46 +0200 | [diff] [blame] | 679 | /* |
Wanpeng Li | 331cbde | 2012-06-09 11:10:55 +0800 | [diff] [blame] | 680 | * Don't bother with new inodes or inodes being freed, first |
| 681 | * kind does not need periodic writeout yet, and for the latter |
Christoph Hellwig | 9843b76 | 2010-10-24 19:40:46 +0200 | [diff] [blame] | 682 | * kind writeout is handled by the freer. |
| 683 | */ |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 684 | spin_lock(&inode->i_lock); |
Christoph Hellwig | 9843b76 | 2010-10-24 19:40:46 +0200 | [diff] [blame] | 685 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 686 | spin_unlock(&inode->i_lock); |
Wu Fengguang | fcc5c22 | 2011-07-11 23:08:50 -0700 | [diff] [blame] | 687 | redirty_tail(inode, wb); |
Nick Piggin | 7ef0d73 | 2009-03-12 14:31:38 -0700 | [diff] [blame] | 688 | continue; |
| 689 | } |
Jan Kara | cc1676d | 2012-05-03 14:47:56 +0200 | [diff] [blame] | 690 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
| 691 | /* |
| 692 | * If this inode is locked for writeback and we are not |
| 693 | * doing writeback-for-data-integrity, move it to |
| 694 | * b_more_io so that writeback can proceed with the |
| 695 | * other inodes on s_io. |
| 696 | * |
| 697 | * We'll have another go at writing back this inode |
| 698 | * when we completed a full scan of b_io. |
| 699 | */ |
| 700 | spin_unlock(&inode->i_lock); |
| 701 | requeue_io(inode, wb); |
| 702 | trace_writeback_sb_inodes_requeue(inode); |
| 703 | continue; |
| 704 | } |
Jan Kara | f0d07b7 | 2012-05-03 14:47:59 +0200 | [diff] [blame] | 705 | spin_unlock(&wb->list_lock); |
| 706 | |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 707 | /* |
| 708 | * We already requeued the inode if it had I_SYNC set and we |
| 709 | * are doing WB_SYNC_NONE writeback. So this catches only the |
| 710 | * WB_SYNC_ALL case. |
| 711 | */ |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 712 | if (inode->i_state & I_SYNC) { |
| 713 | /* Wait for I_SYNC. This function drops i_lock... */ |
| 714 | inode_sleep_on_writeback(inode); |
| 715 | /* Inode may be gone, start again */ |
Jan Kara | ead188f | 2012-06-08 17:07:36 +0200 | [diff] [blame] | 716 | spin_lock(&wb->list_lock); |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 717 | continue; |
| 718 | } |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 719 | inode->i_state |= I_SYNC; |
| 720 | spin_unlock(&inode->i_lock); |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 721 | |
Wu Fengguang | 1a12d8b | 2010-08-29 13:28:09 -0600 | [diff] [blame] | 722 | write_chunk = writeback_chunk_size(wb->bdi, work); |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 723 | wbc.nr_to_write = write_chunk; |
| 724 | wbc.pages_skipped = 0; |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 725 | |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 726 | /* |
| 727 | * We use I_SYNC to pin the inode in memory. While it is set |
| 728 | * evict_inode() will wait so the inode cannot be freed. |
| 729 | */ |
Yan Hong | cd8ed2a | 2012-10-08 16:33:45 -0700 | [diff] [blame] | 730 | __writeback_single_inode(inode, &wbc); |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 731 | |
| 732 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
| 733 | wrote += write_chunk - wbc.nr_to_write; |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 734 | spin_lock(&wb->list_lock); |
| 735 | spin_lock(&inode->i_lock); |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 736 | if (!(inode->i_state & I_DIRTY_ALL)) |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 737 | wrote++; |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 738 | requeue_inode(inode, wb, &wbc); |
| 739 | inode_sync_complete(inode); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 740 | spin_unlock(&inode->i_lock); |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 741 | cond_resched_lock(&wb->list_lock); |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 742 | /* |
| 743 | * bail out to wb_writeback() often enough to check |
| 744 | * background threshold and other termination conditions. |
| 745 | */ |
| 746 | if (wrote) { |
| 747 | if (time_is_before_jiffies(start_time + HZ / 10UL)) |
| 748 | break; |
| 749 | if (work->nr_pages <= 0) |
| 750 | break; |
Fengguang Wu | 8bc3be2 | 2008-02-04 22:29:36 -0800 | [diff] [blame] | 751 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 752 | } |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 753 | return wrote; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 754 | } |
Nick Piggin | 38f2197 | 2009-01-06 14:40:25 -0800 | [diff] [blame] | 755 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 756 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
| 757 | struct wb_writeback_work *work) |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 758 | { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 759 | unsigned long start_time = jiffies; |
| 760 | long wrote = 0; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 761 | |
| 762 | while (!list_empty(&wb->b_io)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 763 | struct inode *inode = wb_inode(wb->b_io.prev); |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 764 | struct super_block *sb = inode->i_sb; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 765 | |
Dave Chinner | 12ad3ab | 2011-07-08 14:14:41 +1000 | [diff] [blame] | 766 | if (!grab_super_passive(sb)) { |
Wu Fengguang | 0e99581 | 2011-07-29 22:14:35 -0600 | [diff] [blame] | 767 | /* |
| 768 | * grab_super_passive() may fail consistently due to |
| 769 | * s_umount being grabbed by someone else. Don't use |
| 770 | * requeue_io() to avoid busy retrying the inode/sb. |
| 771 | */ |
| 772 | redirty_tail(inode, wb); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 773 | continue; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 774 | } |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 775 | wrote += writeback_sb_inodes(sb, wb, work); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 776 | drop_super(sb); |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 777 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 778 | /* refer to the same tests at the end of writeback_sb_inodes */ |
| 779 | if (wrote) { |
| 780 | if (time_is_before_jiffies(start_time + HZ / 10UL)) |
| 781 | break; |
| 782 | if (work->nr_pages <= 0) |
| 783 | break; |
| 784 | } |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 785 | } |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 786 | /* Leave any unwritten inodes on b_io */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 787 | return wrote; |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 788 | } |
| 789 | |
Wanpeng Li | 7d9f073 | 2013-09-11 14:22:40 -0700 | [diff] [blame] | 790 | static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 791 | enum wb_reason reason) |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 792 | { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 793 | struct wb_writeback_work work = { |
| 794 | .nr_pages = nr_pages, |
| 795 | .sync_mode = WB_SYNC_NONE, |
| 796 | .range_cyclic = 1, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 797 | .reason = reason, |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 798 | }; |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 799 | |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 800 | spin_lock(&wb->list_lock); |
Wu Fengguang | 424b351 | 2010-07-21 20:11:53 -0600 | [diff] [blame] | 801 | if (list_empty(&wb->b_io)) |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 802 | queue_io(wb, &work); |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 803 | __writeback_inodes_wb(wb, &work); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 804 | spin_unlock(&wb->list_lock); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 805 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 806 | return nr_pages - work.nr_pages; |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 807 | } |
| 808 | |
Wu Fengguang | b00949a | 2010-11-18 14:38:33 -0600 | [diff] [blame] | 809 | static bool over_bground_thresh(struct backing_dev_info *bdi) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 810 | { |
| 811 | unsigned long background_thresh, dirty_thresh; |
| 812 | |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 813 | global_dirty_limits(&background_thresh, &dirty_thresh); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 814 | |
Wu Fengguang | b00949a | 2010-11-18 14:38:33 -0600 | [diff] [blame] | 815 | if (global_page_state(NR_FILE_DIRTY) + |
| 816 | global_page_state(NR_UNSTABLE_NFS) > background_thresh) |
| 817 | return true; |
| 818 | |
| 819 | if (bdi_stat(bdi, BDI_RECLAIMABLE) > |
| 820 | bdi_dirty_limit(bdi, background_thresh)) |
| 821 | return true; |
| 822 | |
| 823 | return false; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 824 | } |
| 825 | |
| 826 | /* |
Wu Fengguang | e98be2d | 2010-08-29 11:22:30 -0600 | [diff] [blame] | 827 | * Called under wb->list_lock. If there are multiple wb per bdi, |
| 828 | * only the flusher working on the first wb should do it. |
| 829 | */ |
| 830 | static void wb_update_bandwidth(struct bdi_writeback *wb, |
| 831 | unsigned long start_time) |
| 832 | { |
Wu Fengguang | af6a311 | 2011-10-03 20:46:17 -0600 | [diff] [blame] | 833 | __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); |
Wu Fengguang | e98be2d | 2010-08-29 11:22:30 -0600 | [diff] [blame] | 834 | } |
| 835 | |
| 836 | /* |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 837 | * Explicit flushing or periodic writeback of "old" data. |
| 838 | * |
| 839 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
| 840 | * dirtying-time in the inode's address_space. So this periodic writeback code |
| 841 | * just walks the superblock inode list, writing back any inodes which are |
| 842 | * older than a specific point in time. |
| 843 | * |
| 844 | * Try to run once per dirty_writeback_interval. But if a writeback event |
| 845 | * takes longer than a dirty_writeback_interval interval, then leave a |
| 846 | * one-second gap. |
| 847 | * |
| 848 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
| 849 | * all dirty pages if they are all attached to "old" mappings. |
| 850 | */ |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 851 | static long wb_writeback(struct bdi_writeback *wb, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 852 | struct wb_writeback_work *work) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 853 | { |
Wu Fengguang | e98be2d | 2010-08-29 11:22:30 -0600 | [diff] [blame] | 854 | unsigned long wb_start = jiffies; |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 855 | long nr_pages = work->nr_pages; |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 856 | unsigned long oldest_jif; |
Jan Kara | a5989bd | 2009-09-16 19:22:48 +0200 | [diff] [blame] | 857 | struct inode *inode; |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 858 | long progress; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 859 | |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 860 | oldest_jif = jiffies; |
| 861 | work->older_than_this = &oldest_jif; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 862 | |
Wu Fengguang | e8dfc30 | 2011-04-21 12:06:32 -0600 | [diff] [blame] | 863 | spin_lock(&wb->list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 864 | for (;;) { |
| 865 | /* |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 866 | * Stop writeback when nr_pages has been consumed |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 867 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 868 | if (work->nr_pages <= 0) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 869 | break; |
| 870 | |
| 871 | /* |
Jan Kara | aa373cf | 2011-01-13 15:45:47 -0800 | [diff] [blame] | 872 | * Background writeout and kupdate-style writeback may |
| 873 | * run forever. Stop them if there is other work to do |
| 874 | * so that e.g. sync can proceed. They'll be restarted |
| 875 | * after the other works are all done. |
| 876 | */ |
| 877 | if ((work->for_background || work->for_kupdate) && |
| 878 | !list_empty(&wb->bdi->work_list)) |
| 879 | break; |
| 880 | |
| 881 | /* |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 882 | * For background writeout, stop when we are below the |
| 883 | * background dirty threshold |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 884 | */ |
Wu Fengguang | b00949a | 2010-11-18 14:38:33 -0600 | [diff] [blame] | 885 | if (work->for_background && !over_bground_thresh(wb->bdi)) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 886 | break; |
| 887 | |
Jan Kara | 1bc36b6 | 2011-10-19 11:44:41 +0200 | [diff] [blame] | 888 | /* |
| 889 | * Kupdate and background works are special and we want to |
| 890 | * include all inodes that need writing. Livelock avoidance is |
| 891 | * handled by these works yielding to any other work so we are |
| 892 | * safe. |
| 893 | */ |
Wu Fengguang | ba9aa83 | 2010-07-21 20:32:30 -0600 | [diff] [blame] | 894 | if (work->for_kupdate) { |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 895 | oldest_jif = jiffies - |
Wu Fengguang | ba9aa83 | 2010-07-21 20:32:30 -0600 | [diff] [blame] | 896 | msecs_to_jiffies(dirty_expire_interval * 10); |
Jan Kara | 1bc36b6 | 2011-10-19 11:44:41 +0200 | [diff] [blame] | 897 | } else if (work->for_background) |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 898 | oldest_jif = jiffies; |
Dave Chinner | 028c2dd | 2010-07-07 13:24:07 +1000 | [diff] [blame] | 899 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 900 | trace_writeback_start(wb->bdi, work); |
Wu Fengguang | e8dfc30 | 2011-04-21 12:06:32 -0600 | [diff] [blame] | 901 | if (list_empty(&wb->b_io)) |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 902 | queue_io(wb, work); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 903 | if (work->sb) |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 904 | progress = writeback_sb_inodes(work->sb, wb, work); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 905 | else |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 906 | progress = __writeback_inodes_wb(wb, work); |
| 907 | trace_writeback_written(wb->bdi, work); |
Dave Chinner | 028c2dd | 2010-07-07 13:24:07 +1000 | [diff] [blame] | 908 | |
Wu Fengguang | e98be2d | 2010-08-29 11:22:30 -0600 | [diff] [blame] | 909 | wb_update_bandwidth(wb, wb_start); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 910 | |
| 911 | /* |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 912 | * Did we write something? Try for more |
Wu Fengguang | e6fb6da | 2010-07-22 10:23:44 -0600 | [diff] [blame] | 913 | * |
| 914 | * Dirty inodes are moved to b_io for writeback in batches. |
| 915 | * The completion of the current batch does not necessarily |
| 916 | * mean the overall work is done. So we keep looping as long |
| 917 | * as made some progress on cleaning pages or inodes. |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 918 | */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 919 | if (progress) |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 920 | continue; |
| 921 | /* |
Wu Fengguang | e6fb6da | 2010-07-22 10:23:44 -0600 | [diff] [blame] | 922 | * No more inodes for IO, bail |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 923 | */ |
Wu Fengguang | b7a2441 | 2010-07-21 22:19:51 -0600 | [diff] [blame] | 924 | if (list_empty(&wb->b_more_io)) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 925 | break; |
| 926 | /* |
Jens Axboe | 8010c3b | 2009-09-15 20:04:57 +0200 | [diff] [blame] | 927 | * Nothing written. Wait for some inode to |
| 928 | * become available for writeback. Otherwise |
| 929 | * we'll just busyloop. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 930 | */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 931 | if (!list_empty(&wb->b_more_io)) { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 932 | trace_writeback_wait(wb->bdi, work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 933 | inode = wb_inode(wb->b_more_io.prev); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 934 | spin_lock(&inode->i_lock); |
Jan Kara | f0d07b7 | 2012-05-03 14:47:59 +0200 | [diff] [blame] | 935 | spin_unlock(&wb->list_lock); |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 936 | /* This function drops i_lock... */ |
| 937 | inode_sleep_on_writeback(inode); |
Jan Kara | f0d07b7 | 2012-05-03 14:47:59 +0200 | [diff] [blame] | 938 | spin_lock(&wb->list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 939 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 940 | } |
Wu Fengguang | e8dfc30 | 2011-04-21 12:06:32 -0600 | [diff] [blame] | 941 | spin_unlock(&wb->list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 942 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 943 | return nr_pages - work->nr_pages; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 944 | } |
| 945 | |
| 946 | /* |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 947 | * 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] | 948 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 949 | static struct wb_writeback_work * |
Minchan Kim | 08852b6 | 2010-08-03 12:51:16 +0200 | [diff] [blame] | 950 | get_next_work_item(struct backing_dev_info *bdi) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 951 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 952 | struct wb_writeback_work *work = NULL; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 953 | |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 954 | spin_lock_bh(&bdi->wb_lock); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 955 | if (!list_empty(&bdi->work_list)) { |
| 956 | work = list_entry(bdi->work_list.next, |
| 957 | struct wb_writeback_work, list); |
| 958 | list_del_init(&work->list); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 959 | } |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 960 | spin_unlock_bh(&bdi->wb_lock); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 961 | return work; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 962 | } |
| 963 | |
Linus Torvalds | cdf01dd | 2010-10-30 08:55:52 -0700 | [diff] [blame] | 964 | /* |
| 965 | * Add in the number of potentially dirty inodes, because each inode |
| 966 | * write can dirty pagecache in the underlying blockdev. |
| 967 | */ |
| 968 | static unsigned long get_nr_dirty_pages(void) |
| 969 | { |
| 970 | return global_page_state(NR_FILE_DIRTY) + |
| 971 | global_page_state(NR_UNSTABLE_NFS) + |
| 972 | get_nr_dirty_inodes(); |
| 973 | } |
| 974 | |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 975 | static long wb_check_background_flush(struct bdi_writeback *wb) |
| 976 | { |
Wu Fengguang | b00949a | 2010-11-18 14:38:33 -0600 | [diff] [blame] | 977 | if (over_bground_thresh(wb->bdi)) { |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 978 | |
| 979 | struct wb_writeback_work work = { |
| 980 | .nr_pages = LONG_MAX, |
| 981 | .sync_mode = WB_SYNC_NONE, |
| 982 | .for_background = 1, |
| 983 | .range_cyclic = 1, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 984 | .reason = WB_REASON_BACKGROUND, |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 985 | }; |
| 986 | |
| 987 | return wb_writeback(wb, &work); |
| 988 | } |
| 989 | |
| 990 | return 0; |
| 991 | } |
| 992 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 993 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
| 994 | { |
| 995 | unsigned long expired; |
| 996 | long nr_pages; |
| 997 | |
Jens Axboe | 69b62d0 | 2010-05-17 12:51:03 +0200 | [diff] [blame] | 998 | /* |
| 999 | * When set to zero, disable periodic writeback |
| 1000 | */ |
| 1001 | if (!dirty_writeback_interval) |
| 1002 | return 0; |
| 1003 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1004 | expired = wb->last_old_flush + |
| 1005 | msecs_to_jiffies(dirty_writeback_interval * 10); |
| 1006 | if (time_before(jiffies, expired)) |
| 1007 | return 0; |
| 1008 | |
| 1009 | wb->last_old_flush = jiffies; |
Linus Torvalds | cdf01dd | 2010-10-30 08:55:52 -0700 | [diff] [blame] | 1010 | nr_pages = get_nr_dirty_pages(); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1011 | |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 1012 | if (nr_pages) { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1013 | struct wb_writeback_work work = { |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 1014 | .nr_pages = nr_pages, |
| 1015 | .sync_mode = WB_SYNC_NONE, |
| 1016 | .for_kupdate = 1, |
| 1017 | .range_cyclic = 1, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1018 | .reason = WB_REASON_PERIODIC, |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 1019 | }; |
| 1020 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1021 | return wb_writeback(wb, &work); |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 1022 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1023 | |
| 1024 | return 0; |
| 1025 | } |
| 1026 | |
| 1027 | /* |
| 1028 | * Retrieve work items and do the writeback they describe |
| 1029 | */ |
Wanpeng Li | 25d130b | 2013-07-08 16:00:14 -0700 | [diff] [blame] | 1030 | static long wb_do_writeback(struct bdi_writeback *wb) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1031 | { |
| 1032 | struct backing_dev_info *bdi = wb->bdi; |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1033 | struct wb_writeback_work *work; |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 1034 | long wrote = 0; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1035 | |
Jan Kara | 81d73a3 | 2010-08-11 14:17:44 -0700 | [diff] [blame] | 1036 | set_bit(BDI_writeback_running, &wb->bdi->state); |
Minchan Kim | 08852b6 | 2010-08-03 12:51:16 +0200 | [diff] [blame] | 1037 | while ((work = get_next_work_item(bdi)) != NULL) { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1038 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 1039 | trace_writeback_exec(bdi, work); |
| 1040 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1041 | wrote += wb_writeback(wb, work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1042 | |
| 1043 | /* |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1044 | * Notify the caller of completion if this is a synchronous |
| 1045 | * work item, otherwise just free it. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1046 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1047 | if (work->done) |
| 1048 | complete(work->done); |
| 1049 | else |
| 1050 | kfree(work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1051 | } |
| 1052 | |
| 1053 | /* |
| 1054 | * Check for periodic writeback, kupdated() style |
| 1055 | */ |
| 1056 | wrote += wb_check_old_data_flush(wb); |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 1057 | wrote += wb_check_background_flush(wb); |
Jan Kara | 81d73a3 | 2010-08-11 14:17:44 -0700 | [diff] [blame] | 1058 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1059 | |
| 1060 | return wrote; |
| 1061 | } |
| 1062 | |
| 1063 | /* |
| 1064 | * Handle writeback of dirty data for the device backed by this bdi. Also |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1065 | * reschedules periodically and does kupdated style flushing. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1066 | */ |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1067 | void bdi_writeback_workfn(struct work_struct *work) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1068 | { |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1069 | struct bdi_writeback *wb = container_of(to_delayed_work(work), |
| 1070 | struct bdi_writeback, dwork); |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 1071 | struct backing_dev_info *bdi = wb->bdi; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1072 | long pages_written; |
| 1073 | |
Tejun Heo | ef3b101 | 2013-04-30 15:27:24 -0700 | [diff] [blame] | 1074 | set_worker_desc("flush-%s", dev_name(bdi->dev)); |
Peter Zijlstra | 766f916 | 2010-10-26 14:22:45 -0700 | [diff] [blame] | 1075 | current->flags |= PF_SWAPWRITE; |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 1076 | |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1077 | if (likely(!current_is_workqueue_rescuer() || |
Jan Kara | 5acda9d | 2014-04-03 14:46:23 -0700 | [diff] [blame] | 1078 | !test_bit(BDI_registered, &bdi->state))) { |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 1079 | /* |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1080 | * The normal path. Keep writing back @bdi until its |
| 1081 | * work_list is empty. Note that this path is also taken |
| 1082 | * if @bdi is shutting down even when we're running off the |
| 1083 | * rescuer as work_list needs to be drained. |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 1084 | */ |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1085 | do { |
Wanpeng Li | 25d130b | 2013-07-08 16:00:14 -0700 | [diff] [blame] | 1086 | pages_written = wb_do_writeback(wb); |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1087 | trace_writeback_pages_written(pages_written); |
| 1088 | } while (!list_empty(&bdi->work_list)); |
| 1089 | } else { |
| 1090 | /* |
| 1091 | * bdi_wq can't get enough workers and we're running off |
| 1092 | * the emergency worker. Don't hog it. Hopefully, 1024 is |
| 1093 | * enough for efficient IO. |
| 1094 | */ |
| 1095 | pages_written = writeback_inodes_wb(&bdi->wb, 1024, |
| 1096 | WB_REASON_FORKER_THREAD); |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 1097 | trace_writeback_pages_written(pages_written); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1098 | } |
| 1099 | |
Derek Basehore | 6ca738d | 2014-04-03 14:46:22 -0700 | [diff] [blame] | 1100 | if (!list_empty(&bdi->work_list)) |
| 1101 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
| 1102 | else if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
| 1103 | bdi_wakeup_thread_delayed(bdi); |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 1104 | |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1105 | current->flags &= ~PF_SWAPWRITE; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1106 | } |
| 1107 | |
| 1108 | /* |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1109 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
| 1110 | * the whole world. |
| 1111 | */ |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1112 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1113 | { |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 1114 | struct backing_dev_info *bdi; |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 1115 | |
Jan Kara | 47df3dd | 2013-09-11 14:22:22 -0700 | [diff] [blame] | 1116 | if (!nr_pages) |
| 1117 | nr_pages = get_nr_dirty_pages(); |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 1118 | |
| 1119 | rcu_read_lock(); |
| 1120 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
| 1121 | if (!bdi_has_dirty_io(bdi)) |
| 1122 | continue; |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1123 | __bdi_start_writeback(bdi, nr_pages, false, reason); |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 1124 | } |
| 1125 | rcu_read_unlock(); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1126 | } |
| 1127 | |
| 1128 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
| 1129 | { |
| 1130 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { |
| 1131 | struct dentry *dentry; |
| 1132 | const char *name = "?"; |
| 1133 | |
| 1134 | dentry = d_find_alias(inode); |
| 1135 | if (dentry) { |
| 1136 | spin_lock(&dentry->d_lock); |
| 1137 | name = (const char *) dentry->d_name.name; |
| 1138 | } |
| 1139 | printk(KERN_DEBUG |
| 1140 | "%s(%d): dirtied inode %lu (%s) on %s\n", |
| 1141 | current->comm, task_pid_nr(current), inode->i_ino, |
| 1142 | name, inode->i_sb->s_id); |
| 1143 | if (dentry) { |
| 1144 | spin_unlock(&dentry->d_lock); |
| 1145 | dput(dentry); |
| 1146 | } |
| 1147 | } |
| 1148 | } |
| 1149 | |
| 1150 | /** |
| 1151 | * __mark_inode_dirty - internal function |
| 1152 | * @inode: inode to mark |
| 1153 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) |
| 1154 | * Mark an inode as dirty. Callers should use mark_inode_dirty or |
| 1155 | * mark_inode_dirty_sync. |
| 1156 | * |
| 1157 | * Put the inode on the super block's dirty list. |
| 1158 | * |
| 1159 | * CAREFUL! We mark it dirty unconditionally, but move it onto the |
| 1160 | * dirty list only if it is hashed or if it refers to a blockdev. |
| 1161 | * If it was not hashed, it will never be added to the dirty list |
| 1162 | * even if it is later hashed, as it will have been marked dirty already. |
| 1163 | * |
| 1164 | * In short, make sure you hash any inodes _before_ you start marking |
| 1165 | * them dirty. |
| 1166 | * |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1167 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
| 1168 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of |
| 1169 | * the kernel-internal blockdev inode represents the dirtying time of the |
| 1170 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use |
| 1171 | * page->mapping->host, so the page-dirtying time is recorded in the internal |
| 1172 | * blockdev inode. |
| 1173 | */ |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 1174 | #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1175 | void __mark_inode_dirty(struct inode *inode, int flags) |
| 1176 | { |
| 1177 | struct super_block *sb = inode->i_sb; |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1178 | struct backing_dev_info *bdi = NULL; |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 1179 | int dirtytime; |
| 1180 | |
| 1181 | trace_writeback_mark_inode_dirty(inode, flags); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1182 | |
| 1183 | /* |
| 1184 | * Don't do this for I_DIRTY_PAGES - that doesn't actually |
| 1185 | * dirty the inode itself |
| 1186 | */ |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 1187 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { |
Tejun Heo | 9fb0a7d | 2013-01-11 13:06:37 -0800 | [diff] [blame] | 1188 | trace_writeback_dirty_inode_start(inode, flags); |
| 1189 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1190 | if (sb->s_op->dirty_inode) |
Christoph Hellwig | aa38572 | 2011-05-27 06:53:02 -0400 | [diff] [blame] | 1191 | sb->s_op->dirty_inode(inode, flags); |
Tejun Heo | 9fb0a7d | 2013-01-11 13:06:37 -0800 | [diff] [blame] | 1192 | |
| 1193 | trace_writeback_dirty_inode(inode, flags); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1194 | } |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 1195 | if (flags & I_DIRTY_INODE) |
| 1196 | flags &= ~I_DIRTY_TIME; |
| 1197 | dirtytime = flags & I_DIRTY_TIME; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1198 | |
| 1199 | /* |
Tejun Heo | 9c6ac78 | 2014-10-24 15:38:21 -0400 | [diff] [blame] | 1200 | * Paired with smp_mb() in __writeback_single_inode() for the |
| 1201 | * following lockless i_state test. See there for details. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1202 | */ |
| 1203 | smp_mb(); |
| 1204 | |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 1205 | if (((inode->i_state & flags) == flags) || |
| 1206 | (dirtytime && (inode->i_state & I_DIRTY_INODE))) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1207 | return; |
| 1208 | |
| 1209 | if (unlikely(block_dump)) |
| 1210 | block_dump___mark_inode_dirty(inode); |
| 1211 | |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1212 | spin_lock(&inode->i_lock); |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 1213 | if (dirtytime && (inode->i_state & I_DIRTY_INODE)) |
| 1214 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1215 | if ((inode->i_state & flags) != flags) { |
| 1216 | const int was_dirty = inode->i_state & I_DIRTY; |
| 1217 | |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 1218 | if (flags & I_DIRTY_INODE) |
| 1219 | inode->i_state &= ~I_DIRTY_TIME; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1220 | inode->i_state |= flags; |
| 1221 | |
| 1222 | /* |
| 1223 | * If the inode is being synced, just update its dirty state. |
| 1224 | * The unlocker will place the inode on the appropriate |
| 1225 | * superblock list, based upon its state. |
| 1226 | */ |
| 1227 | if (inode->i_state & I_SYNC) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1228 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1229 | |
| 1230 | /* |
| 1231 | * Only add valid (hashed) inodes to the superblock's |
| 1232 | * dirty list. Add blockdev inodes as well. |
| 1233 | */ |
| 1234 | if (!S_ISBLK(inode->i_mode)) { |
Al Viro | 1d3382cb | 2010-10-23 15:19:20 -0400 | [diff] [blame] | 1235 | if (inode_unhashed(inode)) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1236 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1237 | } |
Al Viro | a4ffdde | 2010-06-02 17:38:30 -0400 | [diff] [blame] | 1238 | if (inode->i_state & I_FREEING) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1239 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1240 | |
| 1241 | /* |
| 1242 | * If the inode was already on b_dirty/b_io/b_more_io, don't |
| 1243 | * reposition it (that would break b_dirty time-ordering). |
| 1244 | */ |
| 1245 | if (!was_dirty) { |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 1246 | bool wakeup_bdi = false; |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1247 | bdi = inode_to_bdi(inode); |
Jens Axboe | 500b067 | 2009-09-09 09:10:25 +0200 | [diff] [blame] | 1248 | |
Junxiao Bi | 146d700 | 2013-09-11 14:23:04 -0700 | [diff] [blame] | 1249 | spin_unlock(&inode->i_lock); |
| 1250 | spin_lock(&bdi->wb.list_lock); |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1251 | if (bdi_cap_writeback_dirty(bdi)) { |
| 1252 | WARN(!test_bit(BDI_registered, &bdi->state), |
| 1253 | "bdi-%s not registered\n", bdi->name); |
| 1254 | |
| 1255 | /* |
| 1256 | * If this is the first dirty inode for this |
| 1257 | * bdi, we have to wake-up the corresponding |
| 1258 | * bdi thread to make sure background |
| 1259 | * write-back happens later. |
| 1260 | */ |
| 1261 | if (!wb_has_dirty_io(&bdi->wb)) |
| 1262 | wakeup_bdi = true; |
Jens Axboe | 500b067 | 2009-09-09 09:10:25 +0200 | [diff] [blame] | 1263 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1264 | |
| 1265 | inode->dirtied_when = jiffies; |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 1266 | list_move(&inode->i_wb_list, dirtytime ? |
| 1267 | &bdi->wb.b_dirty_time : &bdi->wb.b_dirty); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1268 | spin_unlock(&bdi->wb.list_lock); |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame^] | 1269 | trace_writeback_dirty_inode_enqueue(inode); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 1270 | |
| 1271 | if (wakeup_bdi) |
| 1272 | bdi_wakeup_thread_delayed(bdi); |
| 1273 | return; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1274 | } |
| 1275 | } |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1276 | out_unlock_inode: |
| 1277 | spin_unlock(&inode->i_lock); |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1278 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1279 | } |
| 1280 | EXPORT_SYMBOL(__mark_inode_dirty); |
| 1281 | |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1282 | static void wait_sb_inodes(struct super_block *sb) |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1283 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1284 | struct inode *inode, *old_inode = NULL; |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1285 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1286 | /* |
| 1287 | * We need to be protected against the filesystem going from |
| 1288 | * r/o to r/w or vice versa. |
| 1289 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1290 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1291 | |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1292 | spin_lock(&inode_sb_list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1293 | |
| 1294 | /* |
| 1295 | * Data integrity sync. Must wait for all pages under writeback, |
| 1296 | * because there may have been pages dirtied before our sync |
| 1297 | * call, but which had writeout started before we write it out. |
| 1298 | * In which case, the inode may not be on the dirty list, but |
| 1299 | * we still have to wait for that writeout. |
| 1300 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1301 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1302 | struct address_space *mapping = inode->i_mapping; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1303 | |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1304 | spin_lock(&inode->i_lock); |
| 1305 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || |
| 1306 | (mapping->nrpages == 0)) { |
| 1307 | spin_unlock(&inode->i_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1308 | continue; |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1309 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1310 | __iget(inode); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1311 | spin_unlock(&inode->i_lock); |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1312 | spin_unlock(&inode_sb_list_lock); |
| 1313 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1314 | /* |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1315 | * We hold a reference to 'inode' so it couldn't have been |
| 1316 | * removed from s_inodes list while we dropped the |
| 1317 | * inode_sb_list_lock. We cannot iput the inode now as we can |
| 1318 | * be holding the last reference and we cannot iput it under |
| 1319 | * inode_sb_list_lock. So we keep the reference and iput it |
| 1320 | * later. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1321 | */ |
| 1322 | iput(old_inode); |
| 1323 | old_inode = inode; |
| 1324 | |
| 1325 | filemap_fdatawait(mapping); |
| 1326 | |
| 1327 | cond_resched(); |
Nick Piggin | 38f2197 | 2009-01-06 14:40:25 -0800 | [diff] [blame] | 1328 | |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1329 | spin_lock(&inode_sb_list_lock); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1330 | } |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1331 | spin_unlock(&inode_sb_list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1332 | iput(old_inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1333 | } |
| 1334 | |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1335 | /** |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1336 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1337 | * @sb: the superblock |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1338 | * @nr: the number of pages to write |
Marcos Paulo de Souza | 786228a | 2011-11-23 20:56:45 +0800 | [diff] [blame] | 1339 | * @reason: reason why some writeback work initiated |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1340 | * |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1341 | * Start writeback on some inodes on this super_block. No guarantees are made |
| 1342 | * 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] | 1343 | * for IO completion of submitted IO. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1344 | */ |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1345 | void writeback_inodes_sb_nr(struct super_block *sb, |
| 1346 | unsigned long nr, |
| 1347 | enum wb_reason reason) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1348 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1349 | DECLARE_COMPLETION_ONSTACK(done); |
| 1350 | struct wb_writeback_work work = { |
Wu Fengguang | 6e6938b | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 1351 | .sb = sb, |
| 1352 | .sync_mode = WB_SYNC_NONE, |
| 1353 | .tagged_writepages = 1, |
| 1354 | .done = &done, |
| 1355 | .nr_pages = nr, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1356 | .reason = reason, |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1357 | }; |
Jens Axboe | 0e3c9a2 | 2010-06-01 11:08:43 +0200 | [diff] [blame] | 1358 | |
Jan Kara | 6eedc70 | 2012-07-03 16:45:27 +0200 | [diff] [blame] | 1359 | if (sb->s_bdi == &noop_backing_dev_info) |
| 1360 | return; |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1361 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1362 | bdi_queue_work(sb->s_bdi, &work); |
| 1363 | wait_for_completion(&done); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1364 | } |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1365 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
| 1366 | |
| 1367 | /** |
| 1368 | * writeback_inodes_sb - writeback dirty inodes from given super_block |
| 1369 | * @sb: the superblock |
Marcos Paulo de Souza | 786228a | 2011-11-23 20:56:45 +0800 | [diff] [blame] | 1370 | * @reason: reason why some writeback work was initiated |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1371 | * |
| 1372 | * Start writeback on some inodes on this super_block. No guarantees are made |
| 1373 | * on how many (if any) will be written, and this function does not wait |
| 1374 | * for IO completion of submitted IO. |
| 1375 | */ |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1376 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1377 | { |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1378 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1379 | } |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1380 | EXPORT_SYMBOL(writeback_inodes_sb); |
| 1381 | |
| 1382 | /** |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 1383 | * try_to_writeback_inodes_sb_nr - try to start writeback if none underway |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1384 | * @sb: the superblock |
| 1385 | * @nr: the number of pages to write |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 1386 | * @reason: the reason of writeback |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1387 | * |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 1388 | * Invoke writeback_inodes_sb_nr if no writeback is currently underway. |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1389 | * Returns 1 if writeback was started, 0 if not. |
| 1390 | */ |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 1391 | int try_to_writeback_inodes_sb_nr(struct super_block *sb, |
| 1392 | unsigned long nr, |
| 1393 | enum wb_reason reason) |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1394 | { |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 1395 | if (writeback_in_progress(sb->s_bdi)) |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1396 | return 1; |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 1397 | |
| 1398 | if (!down_read_trylock(&sb->s_umount)) |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1399 | return 0; |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 1400 | |
| 1401 | writeback_inodes_sb_nr(sb, nr, reason); |
| 1402 | up_read(&sb->s_umount); |
| 1403 | return 1; |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1404 | } |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 1405 | EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); |
| 1406 | |
| 1407 | /** |
| 1408 | * try_to_writeback_inodes_sb - try to start writeback if none underway |
| 1409 | * @sb: the superblock |
| 1410 | * @reason: reason why some writeback work was initiated |
| 1411 | * |
| 1412 | * Implement by try_to_writeback_inodes_sb_nr() |
| 1413 | * Returns 1 if writeback was started, 0 if not. |
| 1414 | */ |
| 1415 | int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
| 1416 | { |
| 1417 | return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
| 1418 | } |
| 1419 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1420 | |
| 1421 | /** |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1422 | * sync_inodes_sb - sync sb inode pages |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 1423 | * @sb: the superblock |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1424 | * |
| 1425 | * This function writes and waits on any dirty inode belonging to this |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 1426 | * super_block. |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1427 | */ |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 1428 | void sync_inodes_sb(struct super_block *sb) |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1429 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1430 | DECLARE_COMPLETION_ONSTACK(done); |
| 1431 | struct wb_writeback_work work = { |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1432 | .sb = sb, |
| 1433 | .sync_mode = WB_SYNC_ALL, |
| 1434 | .nr_pages = LONG_MAX, |
| 1435 | .range_cyclic = 0, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1436 | .done = &done, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1437 | .reason = WB_REASON_SYNC, |
Dave Chinner | 7747bd4 | 2013-07-02 22:38:35 +1000 | [diff] [blame] | 1438 | .for_sync = 1, |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1439 | }; |
| 1440 | |
Jan Kara | 6eedc70 | 2012-07-03 16:45:27 +0200 | [diff] [blame] | 1441 | /* Nothing to do? */ |
| 1442 | if (sb->s_bdi == &noop_backing_dev_info) |
| 1443 | return; |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1444 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
| 1445 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1446 | bdi_queue_work(sb->s_bdi, &work); |
| 1447 | wait_for_completion(&done); |
| 1448 | |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1449 | wait_sb_inodes(sb); |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1450 | } |
| 1451 | EXPORT_SYMBOL(sync_inodes_sb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1452 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1453 | /** |
Andrea Arcangeli | 7f04c26 | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 1454 | * write_inode_now - write an inode to disk |
| 1455 | * @inode: inode to write to disk |
| 1456 | * @sync: whether the write should be synchronous or not |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1457 | * |
Andrea Arcangeli | 7f04c26 | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 1458 | * This function commits an inode to disk immediately if it is dirty. This is |
| 1459 | * primarily needed by knfsd. |
| 1460 | * |
| 1461 | * 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] | 1462 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1463 | int write_inode_now(struct inode *inode, int sync) |
| 1464 | { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1465 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1466 | struct writeback_control wbc = { |
| 1467 | .nr_to_write = LONG_MAX, |
Mike Galbraith | 18914b1 | 2008-02-08 04:20:23 -0800 | [diff] [blame] | 1468 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 1469 | .range_start = 0, |
| 1470 | .range_end = LLONG_MAX, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1471 | }; |
| 1472 | |
| 1473 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) |
Andrew Morton | 49364ce | 2005-11-07 00:59:15 -0800 | [diff] [blame] | 1474 | wbc.nr_to_write = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1475 | |
| 1476 | might_sleep(); |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1477 | return writeback_single_inode(inode, wb, &wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1478 | } |
| 1479 | EXPORT_SYMBOL(write_inode_now); |
| 1480 | |
| 1481 | /** |
| 1482 | * sync_inode - write an inode and its pages to disk. |
| 1483 | * @inode: the inode to sync |
| 1484 | * @wbc: controls the writeback mode |
| 1485 | * |
| 1486 | * sync_inode() will write an inode and its pages to disk. It will also |
| 1487 | * correctly update the inode on its superblock's dirty inode lists and will |
| 1488 | * update inode->i_state. |
| 1489 | * |
| 1490 | * The caller must have a ref on the inode. |
| 1491 | */ |
| 1492 | int sync_inode(struct inode *inode, struct writeback_control *wbc) |
| 1493 | { |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1494 | return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1495 | } |
| 1496 | EXPORT_SYMBOL(sync_inode); |
Christoph Hellwig | c3765016 | 2010-10-06 10:48:20 +0200 | [diff] [blame] | 1497 | |
| 1498 | /** |
Andrew Morton | c691b9d | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 1499 | * sync_inode_metadata - write an inode to disk |
Christoph Hellwig | c3765016 | 2010-10-06 10:48:20 +0200 | [diff] [blame] | 1500 | * @inode: the inode to sync |
| 1501 | * @wait: wait for I/O to complete. |
| 1502 | * |
Andrew Morton | c691b9d | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 1503 | * Write an inode to disk and adjust its dirty state after completion. |
Christoph Hellwig | c3765016 | 2010-10-06 10:48:20 +0200 | [diff] [blame] | 1504 | * |
| 1505 | * Note: only writes the actual inode, no associated data or other metadata. |
| 1506 | */ |
| 1507 | int sync_inode_metadata(struct inode *inode, int wait) |
| 1508 | { |
| 1509 | struct writeback_control wbc = { |
| 1510 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, |
| 1511 | .nr_to_write = 0, /* metadata-only */ |
| 1512 | }; |
| 1513 | |
| 1514 | return sync_inode(inode, &wbc); |
| 1515 | } |
| 1516 | EXPORT_SYMBOL(sync_inode_metadata); |