Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * fs/fs-writeback.c |
| 3 | * |
| 4 | * Copyright (C) 2002, Linus Torvalds. |
| 5 | * |
| 6 | * Contains all the functions related to writing back and waiting |
| 7 | * upon dirty inodes against superblocks, and writing back dirty |
| 8 | * pages against inodes. ie: data writeback. Writeout of the |
| 9 | * inode itself is not handled here. |
| 10 | * |
Francois Cami | e1f8e87 | 2008-10-15 22:01:59 -0700 | [diff] [blame] | 11 | * 10Apr2002 Andrew Morton |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 12 | * Split out of fs/inode.c |
| 13 | * Additions for address_space-based writeback |
| 14 | */ |
| 15 | |
| 16 | #include <linux/kernel.h> |
Jens Axboe | f5ff842 | 2007-09-21 09:19:54 +0200 | [diff] [blame] | 17 | #include <linux/module.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include <linux/spinlock.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 19 | #include <linux/slab.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | #include <linux/sched.h> |
| 21 | #include <linux/fs.h> |
| 22 | #include <linux/mm.h> |
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> |
| 25 | #include <linux/freezer.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | #include <linux/writeback.h> |
| 27 | #include <linux/blkdev.h> |
| 28 | #include <linux/backing-dev.h> |
| 29 | #include <linux/buffer_head.h> |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 30 | #include <linux/tracepoint.h> |
David Howells | 07f3f05 | 2006-09-30 20:52:18 +0200 | [diff] [blame] | 31 | #include "internal.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | |
Jens Axboe | d0bceac | 2009-05-18 08:20:32 +0200 | [diff] [blame] | 33 | /* |
Wu Fengguang | bc31b86 | 2012-01-07 20:41:55 -0600 | [diff] [blame^] | 34 | * 4MB minimal write chunk size |
| 35 | */ |
| 36 | #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) |
| 37 | |
| 38 | /* |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 39 | * Passed into wb_writeback(), essentially a subset of writeback_control |
| 40 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 41 | struct wb_writeback_work { |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 42 | long nr_pages; |
| 43 | struct super_block *sb; |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 44 | unsigned long *older_than_this; |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 45 | enum writeback_sync_modes sync_mode; |
Wu Fengguang | 6e6938b | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 46 | unsigned int tagged_writepages:1; |
H Hartley Sweeten | 52957fe | 2010-04-01 20:36:30 -0500 | [diff] [blame] | 47 | unsigned int for_kupdate:1; |
| 48 | unsigned int range_cyclic:1; |
| 49 | unsigned int for_background:1; |
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 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 56 | /* |
| 57 | * Include the creation of the trace points after defining the |
| 58 | * wb_writeback_work structure so that the definition remains local to this |
| 59 | * file. |
| 60 | */ |
| 61 | #define CREATE_TRACE_POINTS |
| 62 | #include <trace/events/writeback.h> |
| 63 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 64 | /* |
| 65 | * We don't actually have pdflush, but this one is exported though /proc... |
| 66 | */ |
| 67 | int nr_pdflush_threads; |
| 68 | |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 69 | /** |
| 70 | * writeback_in_progress - determine whether there is writeback in progress |
| 71 | * @bdi: the device's backing_dev_info structure. |
| 72 | * |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 73 | * Determine whether there is writeback waiting to be handled against a |
| 74 | * backing device. |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 75 | */ |
| 76 | int writeback_in_progress(struct backing_dev_info *bdi) |
| 77 | { |
Jan Kara | 81d73a3 | 2010-08-11 14:17:44 -0700 | [diff] [blame] | 78 | return test_bit(BDI_writeback_running, &bdi->state); |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 79 | } |
| 80 | |
Jan Kara | 692ebd1 | 2010-09-21 11:51:01 +0200 | [diff] [blame] | 81 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
| 82 | { |
| 83 | struct super_block *sb = inode->i_sb; |
Jan Kara | 692ebd1 | 2010-09-21 11:51:01 +0200 | [diff] [blame] | 84 | |
Christoph Hellwig | aaead25 | 2010-10-04 14:25:33 +0200 | [diff] [blame] | 85 | if (strcmp(sb->s_type->name, "bdev") == 0) |
| 86 | return inode->i_mapping->backing_dev_info; |
| 87 | |
| 88 | return sb->s_bdi; |
Jan Kara | 692ebd1 | 2010-09-21 11:51:01 +0200 | [diff] [blame] | 89 | } |
| 90 | |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 91 | static inline struct inode *wb_inode(struct list_head *head) |
| 92 | { |
| 93 | return list_entry(head, struct inode, i_wb_list); |
| 94 | } |
| 95 | |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 96 | /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ |
| 97 | static void bdi_wakeup_flusher(struct backing_dev_info *bdi) |
Nick Piggin | 4195f73 | 2009-05-28 09:01:15 +0200 | [diff] [blame] | 98 | { |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 99 | if (bdi->wb.task) { |
| 100 | wake_up_process(bdi->wb.task); |
| 101 | } else { |
| 102 | /* |
| 103 | * The bdi thread isn't there, wake up the forker thread which |
| 104 | * will create and run it. |
| 105 | */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 106 | wake_up_process(default_backing_dev_info.wb.task); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 107 | } |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 108 | } |
| 109 | |
| 110 | static void bdi_queue_work(struct backing_dev_info *bdi, |
| 111 | struct wb_writeback_work *work) |
| 112 | { |
| 113 | trace_writeback_queue(bdi, work); |
| 114 | |
| 115 | spin_lock_bh(&bdi->wb_lock); |
| 116 | list_add_tail(&work->list, &bdi->work_list); |
| 117 | if (!bdi->wb.task) |
| 118 | trace_writeback_nothread(bdi, work); |
| 119 | bdi_wakeup_flusher(bdi); |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 120 | spin_unlock_bh(&bdi->wb_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 121 | } |
| 122 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 123 | static void |
| 124 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 125 | bool range_cyclic, enum wb_reason reason) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 126 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 127 | struct wb_writeback_work *work; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 128 | |
Jens Axboe | bcddc3f | 2009-09-13 20:07:36 +0200 | [diff] [blame] | 129 | /* |
| 130 | * This is WB_SYNC_NONE writeback, so if allocation fails just |
| 131 | * wakeup the thread for old dirty data writeback |
| 132 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 133 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
| 134 | if (!work) { |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 135 | if (bdi->wb.task) { |
| 136 | trace_writeback_nowork(bdi); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 137 | wake_up_process(bdi->wb.task); |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 138 | } |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 139 | return; |
Jens Axboe | bcddc3f | 2009-09-13 20:07:36 +0200 | [diff] [blame] | 140 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 141 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 142 | work->sync_mode = WB_SYNC_NONE; |
| 143 | work->nr_pages = nr_pages; |
| 144 | work->range_cyclic = range_cyclic; |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 145 | work->reason = reason; |
Christoph Hellwig | f0fad8a | 2009-09-11 09:47:56 +0200 | [diff] [blame] | 146 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 147 | bdi_queue_work(bdi, work); |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 148 | } |
| 149 | |
| 150 | /** |
| 151 | * bdi_start_writeback - start writeback |
| 152 | * @bdi: the backing device to write from |
| 153 | * @nr_pages: the number of pages to write |
Marcos Paulo de Souza | 786228a | 2011-11-23 20:56:45 +0800 | [diff] [blame] | 154 | * @reason: reason why some writeback work was initiated |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 155 | * |
| 156 | * Description: |
| 157 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 158 | * started when this function returns, we make no guarantees on |
Jens Axboe | 0e3c9a2 | 2010-06-01 11:08:43 +0200 | [diff] [blame] | 159 | * completion. Caller need not hold sb s_umount semaphore. |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 160 | * |
| 161 | */ |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 162 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
| 163 | enum wb_reason reason) |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 164 | { |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 165 | __bdi_start_writeback(bdi, nr_pages, true, reason); |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 166 | } |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 167 | |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 168 | /** |
| 169 | * bdi_start_background_writeback - start background writeback |
| 170 | * @bdi: the backing device to write from |
| 171 | * |
| 172 | * Description: |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 173 | * This makes sure WB_SYNC_NONE background writeback happens. When |
| 174 | * this function returns, it is only guaranteed that for given BDI |
| 175 | * some IO is happening if we are over background dirty threshold. |
| 176 | * Caller need not hold sb s_umount semaphore. |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 177 | */ |
| 178 | void bdi_start_background_writeback(struct backing_dev_info *bdi) |
| 179 | { |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 180 | /* |
| 181 | * We just wake up the flusher thread. It will perform background |
| 182 | * writeback as soon as there is no other work to do. |
| 183 | */ |
Wu Fengguang | 71927e8 | 2011-01-13 15:45:46 -0800 | [diff] [blame] | 184 | trace_writeback_wake_background(bdi); |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 185 | spin_lock_bh(&bdi->wb_lock); |
| 186 | bdi_wakeup_flusher(bdi); |
| 187 | spin_unlock_bh(&bdi->wb_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 188 | } |
| 189 | |
| 190 | /* |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 191 | * Remove the inode from the writeback list it is on. |
| 192 | */ |
| 193 | void inode_wb_list_del(struct inode *inode) |
| 194 | { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 195 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 196 | |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 197 | spin_lock(&bdi->wb.list_lock); |
| 198 | list_del_init(&inode->i_wb_list); |
| 199 | spin_unlock(&bdi->wb.list_lock); |
| 200 | } |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 201 | |
| 202 | /* |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 203 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the |
| 204 | * furthest end of its superblock's dirty-inode list. |
| 205 | * |
| 206 | * 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] | 207 | * 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] | 208 | * the case then the inode must have been redirtied while it was being written |
| 209 | * out and we don't reset its dirtied_when. |
| 210 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 211 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 212 | { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 213 | assert_spin_locked(&wb->list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 214 | if (!list_empty(&wb->b_dirty)) { |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 215 | struct inode *tail; |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 216 | |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 217 | tail = wb_inode(wb->b_dirty.next); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 218 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 219 | inode->dirtied_when = jiffies; |
| 220 | } |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 221 | list_move(&inode->i_wb_list, &wb->b_dirty); |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 222 | } |
| 223 | |
| 224 | /* |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 225 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 226 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 227 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 228 | { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 229 | assert_spin_locked(&wb->list_lock); |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 230 | list_move(&inode->i_wb_list, &wb->b_more_io); |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 231 | } |
| 232 | |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 233 | static void inode_sync_complete(struct inode *inode) |
| 234 | { |
| 235 | /* |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 236 | * Prevent speculative execution through |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 237 | * spin_unlock(&wb->list_lock); |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 238 | */ |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 239 | |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 240 | smp_mb(); |
| 241 | wake_up_bit(&inode->i_state, __I_SYNC); |
| 242 | } |
| 243 | |
Jeff Layton | d2caa3c | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 244 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
| 245 | { |
| 246 | bool ret = time_after(inode->dirtied_when, t); |
| 247 | #ifndef CONFIG_64BIT |
| 248 | /* |
| 249 | * For inodes being constantly redirtied, dirtied_when can get stuck. |
| 250 | * It _appears_ to be in the future, but is actually in distant past. |
| 251 | * This test is necessary to prevent such wrapped-around relative times |
Jens Axboe | 5b0830c | 2009-09-23 19:37:09 +0200 | [diff] [blame] | 252 | * from permanently stopping the whole bdi writeback. |
Jeff Layton | d2caa3c | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 253 | */ |
| 254 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); |
| 255 | #endif |
| 256 | return ret; |
| 257 | } |
| 258 | |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 259 | /* |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 260 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. |
| 261 | */ |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 262 | static int move_expired_inodes(struct list_head *delaying_queue, |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 263 | struct list_head *dispatch_queue, |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 264 | struct wb_writeback_work *work) |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 265 | { |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 266 | LIST_HEAD(tmp); |
| 267 | struct list_head *pos, *node; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 268 | struct super_block *sb = NULL; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 269 | struct inode *inode; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 270 | int do_sb_sort = 0; |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 271 | int moved = 0; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 272 | |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 273 | while (!list_empty(delaying_queue)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 274 | inode = wb_inode(delaying_queue->prev); |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 275 | if (work->older_than_this && |
| 276 | inode_dirtied_after(inode, *work->older_than_this)) |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 277 | break; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 278 | if (sb && sb != inode->i_sb) |
| 279 | do_sb_sort = 1; |
| 280 | sb = inode->i_sb; |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 281 | list_move(&inode->i_wb_list, &tmp); |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 282 | moved++; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 283 | } |
| 284 | |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 285 | /* just one sb in list, splice to dispatch_queue and we're done */ |
| 286 | if (!do_sb_sort) { |
| 287 | list_splice(&tmp, dispatch_queue); |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 288 | goto out; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 289 | } |
| 290 | |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 291 | /* Move inodes from one superblock together */ |
| 292 | while (!list_empty(&tmp)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 293 | sb = wb_inode(tmp.prev)->i_sb; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 294 | list_for_each_prev_safe(pos, node, &tmp) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 295 | inode = wb_inode(pos); |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 296 | if (inode->i_sb == sb) |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 297 | list_move(&inode->i_wb_list, dispatch_queue); |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 298 | } |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 299 | } |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 300 | out: |
| 301 | return moved; |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 302 | } |
| 303 | |
| 304 | /* |
| 305 | * Queue all expired dirty inodes for io, eldest first. |
Wu Fengguang | 4ea879b | 2010-08-11 14:17:42 -0700 | [diff] [blame] | 306 | * Before |
| 307 | * newly dirtied b_dirty b_io b_more_io |
| 308 | * =============> gf edc BA |
| 309 | * After |
| 310 | * newly dirtied b_dirty b_io b_more_io |
| 311 | * =============> g fBAedc |
| 312 | * | |
| 313 | * +--> dequeue for IO |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 314 | */ |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 315 | 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] | 316 | { |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 317 | int moved; |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 318 | assert_spin_locked(&wb->list_lock); |
Wu Fengguang | 4ea879b | 2010-08-11 14:17:42 -0700 | [diff] [blame] | 319 | list_splice_init(&wb->b_more_io, &wb->b_io); |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 320 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); |
| 321 | trace_writeback_queue_io(wb, work, moved); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 322 | } |
| 323 | |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 324 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 325 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 326 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 327 | return inode->i_sb->s_op->write_inode(inode, wbc); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 328 | return 0; |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 329 | } |
| 330 | |
| 331 | /* |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 332 | * Wait for writeback on an inode to complete. |
| 333 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 334 | static void inode_wait_for_writeback(struct inode *inode, |
| 335 | struct bdi_writeback *wb) |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 336 | { |
| 337 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); |
| 338 | wait_queue_head_t *wqh; |
| 339 | |
| 340 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 341 | while (inode->i_state & I_SYNC) { |
| 342 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 343 | spin_unlock(&wb->list_lock); |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 344 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 345 | spin_lock(&wb->list_lock); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 346 | spin_lock(&inode->i_lock); |
Richard Kennedy | 58a9d3d | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 347 | } |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 348 | } |
| 349 | |
| 350 | /* |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 351 | * Write out an inode's dirty pages. Called under wb->list_lock and |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 352 | * inode->i_lock. Either the caller has an active reference on the inode or |
| 353 | * the inode has I_WILL_FREE set. |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 354 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 355 | * If `wait' is set, wait on the writeout. |
| 356 | * |
| 357 | * The whole writeout design is quite complex and fragile. We want to avoid |
| 358 | * starvation of particular inodes when others are being redirtied, prevent |
| 359 | * livelocks, etc. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 360 | */ |
| 361 | static int |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 362 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, |
| 363 | struct writeback_control *wbc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 364 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 365 | struct address_space *mapping = inode->i_mapping; |
Wu Fengguang | 251d6a4 | 2010-12-01 17:33:37 -0600 | [diff] [blame] | 366 | long nr_to_write = wbc->nr_to_write; |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 367 | unsigned dirty; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | int ret; |
| 369 | |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 370 | assert_spin_locked(&wb->list_lock); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 371 | assert_spin_locked(&inode->i_lock); |
| 372 | |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 373 | if (!atomic_read(&inode->i_count)) |
| 374 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); |
| 375 | else |
| 376 | WARN_ON(inode->i_state & I_WILL_FREE); |
| 377 | |
| 378 | if (inode->i_state & I_SYNC) { |
| 379 | /* |
| 380 | * If this inode is locked for writeback and we are not doing |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 381 | * writeback-for-data-integrity, move it to b_more_io so that |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 382 | * writeback can proceed with the other inodes on s_io. |
| 383 | * |
| 384 | * We'll have another go at writing back this inode when we |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 385 | * completed a full scan of b_io. |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 386 | */ |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 387 | if (wbc->sync_mode != WB_SYNC_ALL) { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 388 | requeue_io(inode, wb); |
Wu Fengguang | 251d6a4 | 2010-12-01 17:33:37 -0600 | [diff] [blame] | 389 | trace_writeback_single_inode_requeue(inode, wbc, |
| 390 | nr_to_write); |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 391 | return 0; |
| 392 | } |
| 393 | |
| 394 | /* |
| 395 | * It's a data-integrity sync. We must wait. |
| 396 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 397 | inode_wait_for_writeback(inode, wb); |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 398 | } |
| 399 | |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 400 | BUG_ON(inode->i_state & I_SYNC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 401 | |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 402 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 403 | inode->i_state |= I_SYNC; |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 404 | inode->i_state &= ~I_DIRTY_PAGES; |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 405 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 406 | spin_unlock(&wb->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 407 | |
| 408 | ret = do_writepages(mapping, wbc); |
| 409 | |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 410 | /* |
| 411 | * Make sure to wait on the data before writing out the metadata. |
| 412 | * This is important for filesystems that modify metadata on data |
| 413 | * I/O completion. |
| 414 | */ |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 415 | if (wbc->sync_mode == WB_SYNC_ALL) { |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 416 | int err = filemap_fdatawait(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 417 | if (ret == 0) |
| 418 | ret = err; |
| 419 | } |
| 420 | |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 421 | /* |
| 422 | * Some filesystems may redirty the inode during the writeback |
| 423 | * due to delalloc, clear dirty metadata flags right before |
| 424 | * write_inode() |
| 425 | */ |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 426 | spin_lock(&inode->i_lock); |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 427 | dirty = inode->i_state & I_DIRTY; |
| 428 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 429 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 430 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
| 431 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 432 | int err = write_inode(inode, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 433 | if (ret == 0) |
| 434 | ret = err; |
| 435 | } |
| 436 | |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 437 | spin_lock(&wb->list_lock); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 438 | spin_lock(&inode->i_lock); |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 439 | inode->i_state &= ~I_SYNC; |
Al Viro | a4ffdde | 2010-06-02 17:38:30 -0400 | [diff] [blame] | 440 | if (!(inode->i_state & I_FREEING)) { |
Wu Fengguang | 94c3dcb | 2011-04-27 19:05:21 -0600 | [diff] [blame] | 441 | /* |
| 442 | * Sync livelock prevention. Each inode is tagged and synced in |
| 443 | * one shot. If still dirty, it will be redirty_tail()'ed below. |
| 444 | * Update the dirty time to prevent enqueue and sync it again. |
| 445 | */ |
| 446 | if ((inode->i_state & I_DIRTY) && |
| 447 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) |
| 448 | inode->dirtied_when = jiffies; |
| 449 | |
Wu Fengguang | 23539af | 2010-08-11 14:17:41 -0700 | [diff] [blame] | 450 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 451 | /* |
| 452 | * We didn't write back all the pages. nfs_writepages() |
Wu Fengguang | a50aeb4 | 2010-08-11 14:17:43 -0700 | [diff] [blame] | 453 | * sometimes bales out without doing anything. |
Andrew Morton | 1b43ef9 | 2007-10-16 23:30:35 -0700 | [diff] [blame] | 454 | */ |
Wu Fengguang | a50aeb4 | 2010-08-11 14:17:43 -0700 | [diff] [blame] | 455 | inode->i_state |= I_DIRTY_PAGES; |
| 456 | if (wbc->nr_to_write <= 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 457 | /* |
Wu Fengguang | a50aeb4 | 2010-08-11 14:17:43 -0700 | [diff] [blame] | 458 | * slice used up: queue for next turn |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 459 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 460 | requeue_io(inode, wb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 461 | } else { |
| 462 | /* |
Wu Fengguang | a50aeb4 | 2010-08-11 14:17:43 -0700 | [diff] [blame] | 463 | * Writeback blocked by something other than |
| 464 | * congestion. Delay the inode for some time to |
| 465 | * avoid spinning on the CPU (100% iowait) |
| 466 | * retrying writeback of the dirty page/inode |
| 467 | * that cannot be performed immediately. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 468 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 469 | redirty_tail(inode, wb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | } |
Wu Fengguang | 23539af | 2010-08-11 14:17:41 -0700 | [diff] [blame] | 471 | } else if (inode->i_state & I_DIRTY) { |
| 472 | /* |
| 473 | * Filesystems can dirty the inode during writeback |
| 474 | * operations, such as delayed allocation during |
| 475 | * submission or metadata updates after data IO |
| 476 | * completion. |
| 477 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 478 | redirty_tail(inode, wb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 479 | } else { |
| 480 | /* |
Nick Piggin | 9e38d86 | 2010-10-23 06:55:17 -0400 | [diff] [blame] | 481 | * The inode is clean. At this point we either have |
| 482 | * a reference to the inode or it's on it's way out. |
| 483 | * No need to add it back to the LRU. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 484 | */ |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 485 | list_del_init(&inode->i_wb_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 486 | } |
| 487 | } |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 488 | inode_sync_complete(inode); |
Wu Fengguang | 251d6a4 | 2010-12-01 17:33:37 -0600 | [diff] [blame] | 489 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 490 | return ret; |
| 491 | } |
| 492 | |
Wu Fengguang | 1a12d8b | 2010-08-29 13:28:09 -0600 | [diff] [blame] | 493 | static long writeback_chunk_size(struct backing_dev_info *bdi, |
| 494 | struct wb_writeback_work *work) |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 495 | { |
| 496 | long pages; |
| 497 | |
| 498 | /* |
| 499 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty |
| 500 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX |
| 501 | * here avoids calling into writeback_inodes_wb() more than once. |
| 502 | * |
| 503 | * The intended call sequence for WB_SYNC_ALL writeback is: |
| 504 | * |
| 505 | * wb_writeback() |
| 506 | * writeback_sb_inodes() <== called only once |
| 507 | * write_cache_pages() <== called once for each inode |
| 508 | * (quickly) tag currently dirty pages |
| 509 | * (maybe slowly) sync all tagged pages |
| 510 | */ |
| 511 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) |
| 512 | pages = LONG_MAX; |
Wu Fengguang | 1a12d8b | 2010-08-29 13:28:09 -0600 | [diff] [blame] | 513 | else { |
| 514 | pages = min(bdi->avg_write_bandwidth / 2, |
| 515 | global_dirty_limit / DIRTY_SCOPE); |
| 516 | pages = min(pages, work->nr_pages); |
| 517 | pages = round_down(pages + MIN_WRITEBACK_PAGES, |
| 518 | MIN_WRITEBACK_PAGES); |
| 519 | } |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 520 | |
| 521 | return pages; |
| 522 | } |
| 523 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 524 | /* |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 525 | * Write a portion of b_io inodes which belong to @sb. |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 526 | * |
| 527 | * If @only_this_sb is true, then find and write all such |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 528 | * inodes. Otherwise write only ones which go sequentially |
| 529 | * in reverse order. |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 530 | * |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 531 | * Return the number of pages and/or inodes written. |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 532 | */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 533 | static long writeback_sb_inodes(struct super_block *sb, |
| 534 | struct bdi_writeback *wb, |
| 535 | struct wb_writeback_work *work) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 536 | { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 537 | struct writeback_control wbc = { |
| 538 | .sync_mode = work->sync_mode, |
| 539 | .tagged_writepages = work->tagged_writepages, |
| 540 | .for_kupdate = work->for_kupdate, |
| 541 | .for_background = work->for_background, |
| 542 | .range_cyclic = work->range_cyclic, |
| 543 | .range_start = 0, |
| 544 | .range_end = LLONG_MAX, |
| 545 | }; |
| 546 | unsigned long start_time = jiffies; |
| 547 | long write_chunk; |
| 548 | long wrote = 0; /* count both pages and inodes */ |
| 549 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 550 | while (!list_empty(&wb->b_io)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 551 | struct inode *inode = wb_inode(wb->b_io.prev); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 552 | |
| 553 | if (inode->i_sb != sb) { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 554 | if (work->sb) { |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 555 | /* |
| 556 | * We only want to write back data for this |
| 557 | * superblock, move all inodes not belonging |
| 558 | * to it back onto the dirty list. |
| 559 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 560 | redirty_tail(inode, wb); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 561 | continue; |
| 562 | } |
| 563 | |
| 564 | /* |
| 565 | * The inode belongs to a different superblock. |
| 566 | * Bounce back to the caller to unpin this and |
| 567 | * pin the next superblock. |
| 568 | */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 569 | break; |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 570 | } |
| 571 | |
Christoph Hellwig | 9843b76 | 2010-10-24 19:40:46 +0200 | [diff] [blame] | 572 | /* |
| 573 | * Don't bother with new inodes or inodes beeing freed, first |
| 574 | * kind does not need peridic writeout yet, and for the latter |
| 575 | * kind writeout is handled by the freer. |
| 576 | */ |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 577 | spin_lock(&inode->i_lock); |
Christoph Hellwig | 9843b76 | 2010-10-24 19:40:46 +0200 | [diff] [blame] | 578 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 579 | spin_unlock(&inode->i_lock); |
Wu Fengguang | fcc5c22 | 2011-07-11 23:08:50 -0700 | [diff] [blame] | 580 | redirty_tail(inode, wb); |
Nick Piggin | 7ef0d73 | 2009-03-12 14:31:38 -0700 | [diff] [blame] | 581 | continue; |
| 582 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 583 | __iget(inode); |
Wu Fengguang | 1a12d8b | 2010-08-29 13:28:09 -0600 | [diff] [blame] | 584 | write_chunk = writeback_chunk_size(wb->bdi, work); |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 585 | wbc.nr_to_write = write_chunk; |
| 586 | wbc.pages_skipped = 0; |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 587 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 588 | writeback_single_inode(inode, wb, &wbc); |
| 589 | |
| 590 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
| 591 | wrote += write_chunk - wbc.nr_to_write; |
| 592 | if (!(inode->i_state & I_DIRTY)) |
| 593 | wrote++; |
| 594 | if (wbc.pages_skipped) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 595 | /* |
| 596 | * writeback is not making progress due to locked |
| 597 | * buffers. Skip this inode for now. |
| 598 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 599 | redirty_tail(inode, wb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 600 | } |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 601 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 602 | spin_unlock(&wb->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 603 | iput(inode); |
OGAWA Hirofumi | 4ffc844 | 2006-03-25 03:07:44 -0800 | [diff] [blame] | 604 | cond_resched(); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 605 | spin_lock(&wb->list_lock); |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 606 | /* |
| 607 | * bail out to wb_writeback() often enough to check |
| 608 | * background threshold and other termination conditions. |
| 609 | */ |
| 610 | if (wrote) { |
| 611 | if (time_is_before_jiffies(start_time + HZ / 10UL)) |
| 612 | break; |
| 613 | if (work->nr_pages <= 0) |
| 614 | break; |
Fengguang Wu | 8bc3be2 | 2008-02-04 22:29:36 -0800 | [diff] [blame] | 615 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 616 | } |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 617 | return wrote; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 618 | } |
Nick Piggin | 38f2197 | 2009-01-06 14:40:25 -0800 | [diff] [blame] | 619 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 620 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
| 621 | struct wb_writeback_work *work) |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 622 | { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 623 | unsigned long start_time = jiffies; |
| 624 | long wrote = 0; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 625 | |
| 626 | while (!list_empty(&wb->b_io)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 627 | struct inode *inode = wb_inode(wb->b_io.prev); |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 628 | struct super_block *sb = inode->i_sb; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 629 | |
Dave Chinner | 12ad3ab | 2011-07-08 14:14:41 +1000 | [diff] [blame] | 630 | if (!grab_super_passive(sb)) { |
Wu Fengguang | 0e99581 | 2011-07-29 22:14:35 -0600 | [diff] [blame] | 631 | /* |
| 632 | * grab_super_passive() may fail consistently due to |
| 633 | * s_umount being grabbed by someone else. Don't use |
| 634 | * requeue_io() to avoid busy retrying the inode/sb. |
| 635 | */ |
| 636 | redirty_tail(inode, wb); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 637 | continue; |
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 | wrote += writeback_sb_inodes(sb, wb, work); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 640 | drop_super(sb); |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 641 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 642 | /* refer to the same tests at the end of writeback_sb_inodes */ |
| 643 | if (wrote) { |
| 644 | if (time_is_before_jiffies(start_time + HZ / 10UL)) |
| 645 | break; |
| 646 | if (work->nr_pages <= 0) |
| 647 | break; |
| 648 | } |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 649 | } |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 650 | /* Leave any unwritten inodes on b_io */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 651 | return wrote; |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 652 | } |
| 653 | |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 654 | long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
| 655 | enum wb_reason reason) |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 656 | { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 657 | struct wb_writeback_work work = { |
| 658 | .nr_pages = nr_pages, |
| 659 | .sync_mode = WB_SYNC_NONE, |
| 660 | .range_cyclic = 1, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 661 | .reason = reason, |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 662 | }; |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 663 | |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 664 | spin_lock(&wb->list_lock); |
Wu Fengguang | 424b351 | 2010-07-21 20:11:53 -0600 | [diff] [blame] | 665 | if (list_empty(&wb->b_io)) |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 666 | queue_io(wb, &work); |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 667 | __writeback_inodes_wb(wb, &work); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 668 | spin_unlock(&wb->list_lock); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 669 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 670 | return nr_pages - work.nr_pages; |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 671 | } |
| 672 | |
Wu Fengguang | b00949a | 2010-11-18 14:38:33 -0600 | [diff] [blame] | 673 | static bool over_bground_thresh(struct backing_dev_info *bdi) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 674 | { |
| 675 | unsigned long background_thresh, dirty_thresh; |
| 676 | |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 677 | global_dirty_limits(&background_thresh, &dirty_thresh); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 678 | |
Wu Fengguang | b00949a | 2010-11-18 14:38:33 -0600 | [diff] [blame] | 679 | if (global_page_state(NR_FILE_DIRTY) + |
| 680 | global_page_state(NR_UNSTABLE_NFS) > background_thresh) |
| 681 | return true; |
| 682 | |
| 683 | if (bdi_stat(bdi, BDI_RECLAIMABLE) > |
| 684 | bdi_dirty_limit(bdi, background_thresh)) |
| 685 | return true; |
| 686 | |
| 687 | return false; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 688 | } |
| 689 | |
| 690 | /* |
Wu Fengguang | e98be2d | 2010-08-29 11:22:30 -0600 | [diff] [blame] | 691 | * Called under wb->list_lock. If there are multiple wb per bdi, |
| 692 | * only the flusher working on the first wb should do it. |
| 693 | */ |
| 694 | static void wb_update_bandwidth(struct bdi_writeback *wb, |
| 695 | unsigned long start_time) |
| 696 | { |
Wu Fengguang | af6a311 | 2011-10-03 20:46:17 -0600 | [diff] [blame] | 697 | __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); |
Wu Fengguang | e98be2d | 2010-08-29 11:22:30 -0600 | [diff] [blame] | 698 | } |
| 699 | |
| 700 | /* |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 701 | * Explicit flushing or periodic writeback of "old" data. |
| 702 | * |
| 703 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
| 704 | * dirtying-time in the inode's address_space. So this periodic writeback code |
| 705 | * just walks the superblock inode list, writing back any inodes which are |
| 706 | * older than a specific point in time. |
| 707 | * |
| 708 | * Try to run once per dirty_writeback_interval. But if a writeback event |
| 709 | * takes longer than a dirty_writeback_interval interval, then leave a |
| 710 | * one-second gap. |
| 711 | * |
| 712 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
| 713 | * all dirty pages if they are all attached to "old" mappings. |
| 714 | */ |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 715 | static long wb_writeback(struct bdi_writeback *wb, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 716 | struct wb_writeback_work *work) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 717 | { |
Wu Fengguang | e98be2d | 2010-08-29 11:22:30 -0600 | [diff] [blame] | 718 | unsigned long wb_start = jiffies; |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 719 | long nr_pages = work->nr_pages; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 720 | unsigned long oldest_jif; |
Jan Kara | a5989bd | 2009-09-16 19:22:48 +0200 | [diff] [blame] | 721 | struct inode *inode; |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 722 | long progress; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 723 | |
Wu Fengguang | e185dda | 2011-04-23 11:26:07 -0600 | [diff] [blame] | 724 | oldest_jif = jiffies; |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 725 | work->older_than_this = &oldest_jif; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 726 | |
Wu Fengguang | e8dfc30 | 2011-04-21 12:06:32 -0600 | [diff] [blame] | 727 | spin_lock(&wb->list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 728 | for (;;) { |
| 729 | /* |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 730 | * Stop writeback when nr_pages has been consumed |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 731 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 732 | if (work->nr_pages <= 0) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 733 | break; |
| 734 | |
| 735 | /* |
Jan Kara | aa373cf | 2011-01-13 15:45:47 -0800 | [diff] [blame] | 736 | * Background writeout and kupdate-style writeback may |
| 737 | * run forever. Stop them if there is other work to do |
| 738 | * so that e.g. sync can proceed. They'll be restarted |
| 739 | * after the other works are all done. |
| 740 | */ |
| 741 | if ((work->for_background || work->for_kupdate) && |
| 742 | !list_empty(&wb->bdi->work_list)) |
| 743 | break; |
| 744 | |
| 745 | /* |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 746 | * For background writeout, stop when we are below the |
| 747 | * background dirty threshold |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 748 | */ |
Wu Fengguang | b00949a | 2010-11-18 14:38:33 -0600 | [diff] [blame] | 749 | if (work->for_background && !over_bground_thresh(wb->bdi)) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 750 | break; |
| 751 | |
Jan Kara | 1bc36b6 | 2011-10-19 11:44:41 +0200 | [diff] [blame] | 752 | /* |
| 753 | * Kupdate and background works are special and we want to |
| 754 | * include all inodes that need writing. Livelock avoidance is |
| 755 | * handled by these works yielding to any other work so we are |
| 756 | * safe. |
| 757 | */ |
Wu Fengguang | ba9aa83 | 2010-07-21 20:32:30 -0600 | [diff] [blame] | 758 | if (work->for_kupdate) { |
| 759 | oldest_jif = jiffies - |
| 760 | msecs_to_jiffies(dirty_expire_interval * 10); |
Jan Kara | 1bc36b6 | 2011-10-19 11:44:41 +0200 | [diff] [blame] | 761 | } else if (work->for_background) |
| 762 | oldest_jif = jiffies; |
Dave Chinner | 028c2dd | 2010-07-07 13:24:07 +1000 | [diff] [blame] | 763 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 764 | trace_writeback_start(wb->bdi, work); |
Wu Fengguang | e8dfc30 | 2011-04-21 12:06:32 -0600 | [diff] [blame] | 765 | if (list_empty(&wb->b_io)) |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 766 | queue_io(wb, work); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 767 | if (work->sb) |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 768 | progress = writeback_sb_inodes(work->sb, wb, work); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 769 | else |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 770 | progress = __writeback_inodes_wb(wb, work); |
| 771 | trace_writeback_written(wb->bdi, work); |
Dave Chinner | 028c2dd | 2010-07-07 13:24:07 +1000 | [diff] [blame] | 772 | |
Wu Fengguang | e98be2d | 2010-08-29 11:22:30 -0600 | [diff] [blame] | 773 | wb_update_bandwidth(wb, wb_start); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 774 | |
| 775 | /* |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 776 | * Did we write something? Try for more |
Wu Fengguang | e6fb6da | 2010-07-22 10:23:44 -0600 | [diff] [blame] | 777 | * |
| 778 | * Dirty inodes are moved to b_io for writeback in batches. |
| 779 | * The completion of the current batch does not necessarily |
| 780 | * mean the overall work is done. So we keep looping as long |
| 781 | * as made some progress on cleaning pages or inodes. |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 782 | */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 783 | if (progress) |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 784 | continue; |
| 785 | /* |
Wu Fengguang | e6fb6da | 2010-07-22 10:23:44 -0600 | [diff] [blame] | 786 | * No more inodes for IO, bail |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 787 | */ |
Wu Fengguang | b7a2441 | 2010-07-21 22:19:51 -0600 | [diff] [blame] | 788 | if (list_empty(&wb->b_more_io)) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 789 | break; |
| 790 | /* |
Jens Axboe | 8010c3b | 2009-09-15 20:04:57 +0200 | [diff] [blame] | 791 | * Nothing written. Wait for some inode to |
| 792 | * become available for writeback. Otherwise |
| 793 | * we'll just busyloop. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 794 | */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 795 | if (!list_empty(&wb->b_more_io)) { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 796 | trace_writeback_wait(wb->bdi, work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 797 | inode = wb_inode(wb->b_more_io.prev); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 798 | spin_lock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 799 | inode_wait_for_writeback(inode, wb); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 800 | spin_unlock(&inode->i_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 801 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 802 | } |
Wu Fengguang | e8dfc30 | 2011-04-21 12:06:32 -0600 | [diff] [blame] | 803 | spin_unlock(&wb->list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 804 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 805 | return nr_pages - work->nr_pages; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 806 | } |
| 807 | |
| 808 | /* |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 809 | * 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] | 810 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 811 | static struct wb_writeback_work * |
Minchan Kim | 08852b6 | 2010-08-03 12:51:16 +0200 | [diff] [blame] | 812 | get_next_work_item(struct backing_dev_info *bdi) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 813 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 814 | struct wb_writeback_work *work = NULL; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 815 | |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 816 | spin_lock_bh(&bdi->wb_lock); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 817 | if (!list_empty(&bdi->work_list)) { |
| 818 | work = list_entry(bdi->work_list.next, |
| 819 | struct wb_writeback_work, list); |
| 820 | list_del_init(&work->list); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 821 | } |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 822 | spin_unlock_bh(&bdi->wb_lock); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 823 | return work; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 824 | } |
| 825 | |
Linus Torvalds | cdf01dd | 2010-10-30 08:55:52 -0700 | [diff] [blame] | 826 | /* |
| 827 | * Add in the number of potentially dirty inodes, because each inode |
| 828 | * write can dirty pagecache in the underlying blockdev. |
| 829 | */ |
| 830 | static unsigned long get_nr_dirty_pages(void) |
| 831 | { |
| 832 | return global_page_state(NR_FILE_DIRTY) + |
| 833 | global_page_state(NR_UNSTABLE_NFS) + |
| 834 | get_nr_dirty_inodes(); |
| 835 | } |
| 836 | |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 837 | static long wb_check_background_flush(struct bdi_writeback *wb) |
| 838 | { |
Wu Fengguang | b00949a | 2010-11-18 14:38:33 -0600 | [diff] [blame] | 839 | if (over_bground_thresh(wb->bdi)) { |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 840 | |
| 841 | struct wb_writeback_work work = { |
| 842 | .nr_pages = LONG_MAX, |
| 843 | .sync_mode = WB_SYNC_NONE, |
| 844 | .for_background = 1, |
| 845 | .range_cyclic = 1, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 846 | .reason = WB_REASON_BACKGROUND, |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 847 | }; |
| 848 | |
| 849 | return wb_writeback(wb, &work); |
| 850 | } |
| 851 | |
| 852 | return 0; |
| 853 | } |
| 854 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 855 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
| 856 | { |
| 857 | unsigned long expired; |
| 858 | long nr_pages; |
| 859 | |
Jens Axboe | 69b62d0 | 2010-05-17 12:51:03 +0200 | [diff] [blame] | 860 | /* |
| 861 | * When set to zero, disable periodic writeback |
| 862 | */ |
| 863 | if (!dirty_writeback_interval) |
| 864 | return 0; |
| 865 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 866 | expired = wb->last_old_flush + |
| 867 | msecs_to_jiffies(dirty_writeback_interval * 10); |
| 868 | if (time_before(jiffies, expired)) |
| 869 | return 0; |
| 870 | |
| 871 | wb->last_old_flush = jiffies; |
Linus Torvalds | cdf01dd | 2010-10-30 08:55:52 -0700 | [diff] [blame] | 872 | nr_pages = get_nr_dirty_pages(); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 873 | |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 874 | if (nr_pages) { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 875 | struct wb_writeback_work work = { |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 876 | .nr_pages = nr_pages, |
| 877 | .sync_mode = WB_SYNC_NONE, |
| 878 | .for_kupdate = 1, |
| 879 | .range_cyclic = 1, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 880 | .reason = WB_REASON_PERIODIC, |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 881 | }; |
| 882 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 883 | return wb_writeback(wb, &work); |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 884 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 885 | |
| 886 | return 0; |
| 887 | } |
| 888 | |
| 889 | /* |
| 890 | * Retrieve work items and do the writeback they describe |
| 891 | */ |
| 892 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) |
| 893 | { |
| 894 | struct backing_dev_info *bdi = wb->bdi; |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 895 | struct wb_writeback_work *work; |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 896 | long wrote = 0; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 897 | |
Jan Kara | 81d73a3 | 2010-08-11 14:17:44 -0700 | [diff] [blame] | 898 | set_bit(BDI_writeback_running, &wb->bdi->state); |
Minchan Kim | 08852b6 | 2010-08-03 12:51:16 +0200 | [diff] [blame] | 899 | while ((work = get_next_work_item(bdi)) != NULL) { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 900 | /* |
| 901 | * Override sync mode, in case we must wait for completion |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 902 | * because this thread is exiting now. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 903 | */ |
| 904 | if (force_wait) |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 905 | work->sync_mode = WB_SYNC_ALL; |
| 906 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 907 | trace_writeback_exec(bdi, work); |
| 908 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 909 | wrote += wb_writeback(wb, work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 910 | |
| 911 | /* |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 912 | * Notify the caller of completion if this is a synchronous |
| 913 | * work item, otherwise just free it. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 914 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 915 | if (work->done) |
| 916 | complete(work->done); |
| 917 | else |
| 918 | kfree(work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 919 | } |
| 920 | |
| 921 | /* |
| 922 | * Check for periodic writeback, kupdated() style |
| 923 | */ |
| 924 | wrote += wb_check_old_data_flush(wb); |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 925 | wrote += wb_check_background_flush(wb); |
Jan Kara | 81d73a3 | 2010-08-11 14:17:44 -0700 | [diff] [blame] | 926 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 927 | |
| 928 | return wrote; |
| 929 | } |
| 930 | |
| 931 | /* |
| 932 | * Handle writeback of dirty data for the device backed by this bdi. Also |
| 933 | * wakes up periodically and does kupdated style flushing. |
| 934 | */ |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 935 | int bdi_writeback_thread(void *data) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 936 | { |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 937 | struct bdi_writeback *wb = data; |
| 938 | struct backing_dev_info *bdi = wb->bdi; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 939 | long pages_written; |
| 940 | |
Peter Zijlstra | 766f916 | 2010-10-26 14:22:45 -0700 | [diff] [blame] | 941 | current->flags |= PF_SWAPWRITE; |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 942 | set_freezable(); |
Artem Bityutskiy | ecd5840 | 2010-07-25 14:29:18 +0300 | [diff] [blame] | 943 | wb->last_active = jiffies; |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 944 | |
| 945 | /* |
| 946 | * Our parent may run at a different priority, just set us to normal |
| 947 | */ |
| 948 | set_user_nice(current, 0); |
| 949 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 950 | trace_writeback_thread_start(bdi); |
| 951 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 952 | while (!kthread_should_stop()) { |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 953 | /* |
| 954 | * Remove own delayed wake-up timer, since we are already awake |
| 955 | * and we'll take care of the preriodic write-back. |
| 956 | */ |
| 957 | del_timer(&wb->wakeup_timer); |
| 958 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 959 | pages_written = wb_do_writeback(wb, 0); |
| 960 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 961 | trace_writeback_pages_written(pages_written); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 962 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 963 | if (pages_written) |
Artem Bityutskiy | ecd5840 | 2010-07-25 14:29:18 +0300 | [diff] [blame] | 964 | wb->last_active = jiffies; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 965 | |
Artem Bityutskiy | 297252c | 2010-07-25 14:29:15 +0300 | [diff] [blame] | 966 | set_current_state(TASK_INTERRUPTIBLE); |
J. Bruce Fields | b76b401 | 2010-08-28 08:52:10 +0200 | [diff] [blame] | 967 | if (!list_empty(&bdi->work_list) || kthread_should_stop()) { |
Artem Bityutskiy | 297252c | 2010-07-25 14:29:15 +0300 | [diff] [blame] | 968 | __set_current_state(TASK_RUNNING); |
| 969 | continue; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 970 | } |
| 971 | |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 972 | if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 973 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 974 | else { |
| 975 | /* |
| 976 | * We have nothing to do, so can go sleep without any |
| 977 | * timeout and save power. When a work is queued or |
| 978 | * something is made dirty - we will be woken up. |
| 979 | */ |
Artem Bityutskiy | 297252c | 2010-07-25 14:29:15 +0300 | [diff] [blame] | 980 | schedule(); |
Jens Axboe | f9eadbb | 2010-05-18 14:31:45 +0200 | [diff] [blame] | 981 | } |
Jens Axboe | 69b62d0 | 2010-05-17 12:51:03 +0200 | [diff] [blame] | 982 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 983 | try_to_freeze(); |
| 984 | } |
| 985 | |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 986 | /* Flush any work that raced with us exiting */ |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 987 | if (!list_empty(&bdi->work_list)) |
| 988 | wb_do_writeback(wb, 1); |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 989 | |
| 990 | trace_writeback_thread_stop(bdi); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 991 | return 0; |
| 992 | } |
| 993 | |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 994 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 995 | /* |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 996 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
| 997 | * the whole world. |
| 998 | */ |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 999 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1000 | { |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 1001 | struct backing_dev_info *bdi; |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 1002 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1003 | if (!nr_pages) { |
| 1004 | nr_pages = global_page_state(NR_FILE_DIRTY) + |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1005 | global_page_state(NR_UNSTABLE_NFS); |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 1006 | } |
| 1007 | |
| 1008 | rcu_read_lock(); |
| 1009 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
| 1010 | if (!bdi_has_dirty_io(bdi)) |
| 1011 | continue; |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1012 | __bdi_start_writeback(bdi, nr_pages, false, reason); |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 1013 | } |
| 1014 | rcu_read_unlock(); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1015 | } |
| 1016 | |
| 1017 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
| 1018 | { |
| 1019 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { |
| 1020 | struct dentry *dentry; |
| 1021 | const char *name = "?"; |
| 1022 | |
| 1023 | dentry = d_find_alias(inode); |
| 1024 | if (dentry) { |
| 1025 | spin_lock(&dentry->d_lock); |
| 1026 | name = (const char *) dentry->d_name.name; |
| 1027 | } |
| 1028 | printk(KERN_DEBUG |
| 1029 | "%s(%d): dirtied inode %lu (%s) on %s\n", |
| 1030 | current->comm, task_pid_nr(current), inode->i_ino, |
| 1031 | name, inode->i_sb->s_id); |
| 1032 | if (dentry) { |
| 1033 | spin_unlock(&dentry->d_lock); |
| 1034 | dput(dentry); |
| 1035 | } |
| 1036 | } |
| 1037 | } |
| 1038 | |
| 1039 | /** |
| 1040 | * __mark_inode_dirty - internal function |
| 1041 | * @inode: inode to mark |
| 1042 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) |
| 1043 | * Mark an inode as dirty. Callers should use mark_inode_dirty or |
| 1044 | * mark_inode_dirty_sync. |
| 1045 | * |
| 1046 | * Put the inode on the super block's dirty list. |
| 1047 | * |
| 1048 | * CAREFUL! We mark it dirty unconditionally, but move it onto the |
| 1049 | * dirty list only if it is hashed or if it refers to a blockdev. |
| 1050 | * If it was not hashed, it will never be added to the dirty list |
| 1051 | * even if it is later hashed, as it will have been marked dirty already. |
| 1052 | * |
| 1053 | * In short, make sure you hash any inodes _before_ you start marking |
| 1054 | * them dirty. |
| 1055 | * |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1056 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
| 1057 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of |
| 1058 | * the kernel-internal blockdev inode represents the dirtying time of the |
| 1059 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use |
| 1060 | * page->mapping->host, so the page-dirtying time is recorded in the internal |
| 1061 | * blockdev inode. |
| 1062 | */ |
| 1063 | void __mark_inode_dirty(struct inode *inode, int flags) |
| 1064 | { |
| 1065 | struct super_block *sb = inode->i_sb; |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1066 | struct backing_dev_info *bdi = NULL; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1067 | |
| 1068 | /* |
| 1069 | * Don't do this for I_DIRTY_PAGES - that doesn't actually |
| 1070 | * dirty the inode itself |
| 1071 | */ |
| 1072 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { |
| 1073 | if (sb->s_op->dirty_inode) |
Christoph Hellwig | aa38572 | 2011-05-27 06:53:02 -0400 | [diff] [blame] | 1074 | sb->s_op->dirty_inode(inode, flags); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1075 | } |
| 1076 | |
| 1077 | /* |
| 1078 | * make sure that changes are seen by all cpus before we test i_state |
| 1079 | * -- mikulas |
| 1080 | */ |
| 1081 | smp_mb(); |
| 1082 | |
| 1083 | /* avoid the locking if we can */ |
| 1084 | if ((inode->i_state & flags) == flags) |
| 1085 | return; |
| 1086 | |
| 1087 | if (unlikely(block_dump)) |
| 1088 | block_dump___mark_inode_dirty(inode); |
| 1089 | |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1090 | spin_lock(&inode->i_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1091 | if ((inode->i_state & flags) != flags) { |
| 1092 | const int was_dirty = inode->i_state & I_DIRTY; |
| 1093 | |
| 1094 | inode->i_state |= flags; |
| 1095 | |
| 1096 | /* |
| 1097 | * If the inode is being synced, just update its dirty state. |
| 1098 | * The unlocker will place the inode on the appropriate |
| 1099 | * superblock list, based upon its state. |
| 1100 | */ |
| 1101 | if (inode->i_state & I_SYNC) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1102 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1103 | |
| 1104 | /* |
| 1105 | * Only add valid (hashed) inodes to the superblock's |
| 1106 | * dirty list. Add blockdev inodes as well. |
| 1107 | */ |
| 1108 | if (!S_ISBLK(inode->i_mode)) { |
Al Viro | 1d3382c | 2010-10-23 15:19:20 -0400 | [diff] [blame] | 1109 | if (inode_unhashed(inode)) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1110 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1111 | } |
Al Viro | a4ffdde | 2010-06-02 17:38:30 -0400 | [diff] [blame] | 1112 | if (inode->i_state & I_FREEING) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1113 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1114 | |
| 1115 | /* |
| 1116 | * If the inode was already on b_dirty/b_io/b_more_io, don't |
| 1117 | * reposition it (that would break b_dirty time-ordering). |
| 1118 | */ |
| 1119 | if (!was_dirty) { |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 1120 | bool wakeup_bdi = false; |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1121 | bdi = inode_to_bdi(inode); |
Jens Axboe | 500b067 | 2009-09-09 09:10:25 +0200 | [diff] [blame] | 1122 | |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1123 | if (bdi_cap_writeback_dirty(bdi)) { |
| 1124 | WARN(!test_bit(BDI_registered, &bdi->state), |
| 1125 | "bdi-%s not registered\n", bdi->name); |
| 1126 | |
| 1127 | /* |
| 1128 | * If this is the first dirty inode for this |
| 1129 | * bdi, we have to wake-up the corresponding |
| 1130 | * bdi thread to make sure background |
| 1131 | * write-back happens later. |
| 1132 | */ |
| 1133 | if (!wb_has_dirty_io(&bdi->wb)) |
| 1134 | wakeup_bdi = true; |
Jens Axboe | 500b067 | 2009-09-09 09:10:25 +0200 | [diff] [blame] | 1135 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1136 | |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 1137 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1138 | spin_lock(&bdi->wb.list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1139 | inode->dirtied_when = jiffies; |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 1140 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1141 | spin_unlock(&bdi->wb.list_lock); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 1142 | |
| 1143 | if (wakeup_bdi) |
| 1144 | bdi_wakeup_thread_delayed(bdi); |
| 1145 | return; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1146 | } |
| 1147 | } |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1148 | out_unlock_inode: |
| 1149 | spin_unlock(&inode->i_lock); |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1150 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1151 | } |
| 1152 | EXPORT_SYMBOL(__mark_inode_dirty); |
| 1153 | |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1154 | /* |
| 1155 | * Write out a superblock's list of dirty inodes. A wait will be performed |
| 1156 | * upon no inodes, all inodes or the final one, depending upon sync_mode. |
| 1157 | * |
| 1158 | * If older_than_this is non-NULL, then only write out inodes which |
| 1159 | * had their first dirtying at a time earlier than *older_than_this. |
| 1160 | * |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1161 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
| 1162 | * This function assumes that the blockdev superblock's inodes are backed by |
| 1163 | * a variety of queues, so all inodes are searched. For other superblocks, |
| 1164 | * assume that all inodes are backed by the same queue. |
| 1165 | * |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1166 | * The inodes to be written are parked on bdi->b_io. They are moved back onto |
| 1167 | * bdi->b_dirty as they are selected for writing. This way, none can be missed |
| 1168 | * on the writer throttling path, and we get decent balancing between many |
| 1169 | * throttled threads: we don't want them all piling up on inode_sync_wait. |
| 1170 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1171 | static void wait_sb_inodes(struct super_block *sb) |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1172 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1173 | struct inode *inode, *old_inode = NULL; |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1174 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1175 | /* |
| 1176 | * We need to be protected against the filesystem going from |
| 1177 | * r/o to r/w or vice versa. |
| 1178 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1179 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1180 | |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1181 | spin_lock(&inode_sb_list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1182 | |
| 1183 | /* |
| 1184 | * Data integrity sync. Must wait for all pages under writeback, |
| 1185 | * because there may have been pages dirtied before our sync |
| 1186 | * call, but which had writeout started before we write it out. |
| 1187 | * In which case, the inode may not be on the dirty list, but |
| 1188 | * we still have to wait for that writeout. |
| 1189 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1190 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1191 | struct address_space *mapping = inode->i_mapping; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1192 | |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1193 | spin_lock(&inode->i_lock); |
| 1194 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || |
| 1195 | (mapping->nrpages == 0)) { |
| 1196 | spin_unlock(&inode->i_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1197 | continue; |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1198 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1199 | __iget(inode); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1200 | spin_unlock(&inode->i_lock); |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1201 | spin_unlock(&inode_sb_list_lock); |
| 1202 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1203 | /* |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1204 | * We hold a reference to 'inode' so it couldn't have been |
| 1205 | * removed from s_inodes list while we dropped the |
| 1206 | * inode_sb_list_lock. We cannot iput the inode now as we can |
| 1207 | * be holding the last reference and we cannot iput it under |
| 1208 | * inode_sb_list_lock. So we keep the reference and iput it |
| 1209 | * later. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1210 | */ |
| 1211 | iput(old_inode); |
| 1212 | old_inode = inode; |
| 1213 | |
| 1214 | filemap_fdatawait(mapping); |
| 1215 | |
| 1216 | cond_resched(); |
Nick Piggin | 38f2197 | 2009-01-06 14:40:25 -0800 | [diff] [blame] | 1217 | |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1218 | spin_lock(&inode_sb_list_lock); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1219 | } |
Dave Chinner | 55fa609 | 2011-03-22 22:23:40 +1100 | [diff] [blame] | 1220 | spin_unlock(&inode_sb_list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1221 | iput(old_inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1222 | } |
| 1223 | |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1224 | /** |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1225 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1226 | * @sb: the superblock |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1227 | * @nr: the number of pages to write |
Marcos Paulo de Souza | 786228a | 2011-11-23 20:56:45 +0800 | [diff] [blame] | 1228 | * @reason: reason why some writeback work initiated |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1229 | * |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1230 | * Start writeback on some inodes on this super_block. No guarantees are made |
| 1231 | * 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] | 1232 | * for IO completion of submitted IO. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1233 | */ |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1234 | void writeback_inodes_sb_nr(struct super_block *sb, |
| 1235 | unsigned long nr, |
| 1236 | enum wb_reason reason) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1237 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1238 | DECLARE_COMPLETION_ONSTACK(done); |
| 1239 | struct wb_writeback_work work = { |
Wu Fengguang | 6e6938b | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 1240 | .sb = sb, |
| 1241 | .sync_mode = WB_SYNC_NONE, |
| 1242 | .tagged_writepages = 1, |
| 1243 | .done = &done, |
| 1244 | .nr_pages = nr, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1245 | .reason = reason, |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1246 | }; |
Jens Axboe | 0e3c9a2 | 2010-06-01 11:08:43 +0200 | [diff] [blame] | 1247 | |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1248 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1249 | bdi_queue_work(sb->s_bdi, &work); |
| 1250 | wait_for_completion(&done); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1251 | } |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1252 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
| 1253 | |
| 1254 | /** |
| 1255 | * writeback_inodes_sb - writeback dirty inodes from given super_block |
| 1256 | * @sb: the superblock |
Marcos Paulo de Souza | 786228a | 2011-11-23 20:56:45 +0800 | [diff] [blame] | 1257 | * @reason: reason why some writeback work was initiated |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1258 | * |
| 1259 | * Start writeback on some inodes on this super_block. No guarantees are made |
| 1260 | * on how many (if any) will be written, and this function does not wait |
| 1261 | * for IO completion of submitted IO. |
| 1262 | */ |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1263 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1264 | { |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1265 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1266 | } |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1267 | EXPORT_SYMBOL(writeback_inodes_sb); |
| 1268 | |
| 1269 | /** |
Eric Sandeen | 17bd55d | 2009-12-23 07:57:07 -0500 | [diff] [blame] | 1270 | * writeback_inodes_sb_if_idle - start writeback if none underway |
| 1271 | * @sb: the superblock |
Marcos Paulo de Souza | 786228a | 2011-11-23 20:56:45 +0800 | [diff] [blame] | 1272 | * @reason: reason why some writeback work was initiated |
Eric Sandeen | 17bd55d | 2009-12-23 07:57:07 -0500 | [diff] [blame] | 1273 | * |
| 1274 | * Invoke writeback_inodes_sb if no writeback is currently underway. |
| 1275 | * Returns 1 if writeback was started, 0 if not. |
| 1276 | */ |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1277 | int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason) |
Eric Sandeen | 17bd55d | 2009-12-23 07:57:07 -0500 | [diff] [blame] | 1278 | { |
| 1279 | if (!writeback_in_progress(sb->s_bdi)) { |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1280 | down_read(&sb->s_umount); |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1281 | writeback_inodes_sb(sb, reason); |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1282 | up_read(&sb->s_umount); |
Eric Sandeen | 17bd55d | 2009-12-23 07:57:07 -0500 | [diff] [blame] | 1283 | return 1; |
| 1284 | } else |
| 1285 | return 0; |
| 1286 | } |
| 1287 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); |
| 1288 | |
| 1289 | /** |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1290 | * writeback_inodes_sb_if_idle - start writeback if none underway |
| 1291 | * @sb: the superblock |
| 1292 | * @nr: the number of pages to write |
Marcos Paulo de Souza | 786228a | 2011-11-23 20:56:45 +0800 | [diff] [blame] | 1293 | * @reason: reason why some writeback work was initiated |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1294 | * |
| 1295 | * Invoke writeback_inodes_sb if no writeback is currently underway. |
| 1296 | * Returns 1 if writeback was started, 0 if not. |
| 1297 | */ |
| 1298 | int writeback_inodes_sb_nr_if_idle(struct super_block *sb, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1299 | unsigned long nr, |
| 1300 | enum wb_reason reason) |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1301 | { |
| 1302 | if (!writeback_in_progress(sb->s_bdi)) { |
| 1303 | down_read(&sb->s_umount); |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1304 | writeback_inodes_sb_nr(sb, nr, reason); |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 1305 | up_read(&sb->s_umount); |
| 1306 | return 1; |
| 1307 | } else |
| 1308 | return 0; |
| 1309 | } |
| 1310 | EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); |
| 1311 | |
| 1312 | /** |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1313 | * sync_inodes_sb - sync sb inode pages |
| 1314 | * @sb: the superblock |
| 1315 | * |
| 1316 | * This function writes and waits on any dirty inode belonging to this |
Stefan Hajnoczi | cb9ef8d | 2011-01-13 15:47:26 -0800 | [diff] [blame] | 1317 | * super_block. |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1318 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1319 | void sync_inodes_sb(struct super_block *sb) |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1320 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1321 | DECLARE_COMPLETION_ONSTACK(done); |
| 1322 | struct wb_writeback_work work = { |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1323 | .sb = sb, |
| 1324 | .sync_mode = WB_SYNC_ALL, |
| 1325 | .nr_pages = LONG_MAX, |
| 1326 | .range_cyclic = 0, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1327 | .done = &done, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1328 | .reason = WB_REASON_SYNC, |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1329 | }; |
| 1330 | |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1331 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
| 1332 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1333 | bdi_queue_work(sb->s_bdi, &work); |
| 1334 | wait_for_completion(&done); |
| 1335 | |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1336 | wait_sb_inodes(sb); |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1337 | } |
| 1338 | EXPORT_SYMBOL(sync_inodes_sb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1339 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1340 | /** |
Andrea Arcangeli | 7f04c26 | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 1341 | * write_inode_now - write an inode to disk |
| 1342 | * @inode: inode to write to disk |
| 1343 | * @sync: whether the write should be synchronous or not |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1344 | * |
Andrea Arcangeli | 7f04c26 | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 1345 | * This function commits an inode to disk immediately if it is dirty. This is |
| 1346 | * primarily needed by knfsd. |
| 1347 | * |
| 1348 | * 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] | 1349 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1350 | int write_inode_now(struct inode *inode, int sync) |
| 1351 | { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1352 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1353 | int ret; |
| 1354 | struct writeback_control wbc = { |
| 1355 | .nr_to_write = LONG_MAX, |
Mike Galbraith | 18914b1 | 2008-02-08 04:20:23 -0800 | [diff] [blame] | 1356 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 1357 | .range_start = 0, |
| 1358 | .range_end = LLONG_MAX, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1359 | }; |
| 1360 | |
| 1361 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) |
Andrew Morton | 49364ce | 2005-11-07 00:59:15 -0800 | [diff] [blame] | 1362 | wbc.nr_to_write = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1363 | |
| 1364 | might_sleep(); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1365 | spin_lock(&wb->list_lock); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 1366 | spin_lock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1367 | ret = writeback_single_inode(inode, wb, &wbc); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 1368 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1369 | spin_unlock(&wb->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1370 | if (sync) |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 1371 | inode_sync_wait(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1372 | return ret; |
| 1373 | } |
| 1374 | EXPORT_SYMBOL(write_inode_now); |
| 1375 | |
| 1376 | /** |
| 1377 | * sync_inode - write an inode and its pages to disk. |
| 1378 | * @inode: the inode to sync |
| 1379 | * @wbc: controls the writeback mode |
| 1380 | * |
| 1381 | * sync_inode() will write an inode and its pages to disk. It will also |
| 1382 | * correctly update the inode on its superblock's dirty inode lists and will |
| 1383 | * update inode->i_state. |
| 1384 | * |
| 1385 | * The caller must have a ref on the inode. |
| 1386 | */ |
| 1387 | int sync_inode(struct inode *inode, struct writeback_control *wbc) |
| 1388 | { |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1389 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1390 | int ret; |
| 1391 | |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1392 | spin_lock(&wb->list_lock); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 1393 | spin_lock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1394 | ret = writeback_single_inode(inode, wb, wbc); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 1395 | spin_unlock(&inode->i_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1396 | spin_unlock(&wb->list_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1397 | return ret; |
| 1398 | } |
| 1399 | EXPORT_SYMBOL(sync_inode); |
Christoph Hellwig | c376501 | 2010-10-06 10:48:20 +0200 | [diff] [blame] | 1400 | |
| 1401 | /** |
Andrew Morton | c691b9d | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 1402 | * sync_inode_metadata - write an inode to disk |
Christoph Hellwig | c376501 | 2010-10-06 10:48:20 +0200 | [diff] [blame] | 1403 | * @inode: the inode to sync |
| 1404 | * @wait: wait for I/O to complete. |
| 1405 | * |
Andrew Morton | c691b9d | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 1406 | * Write an inode to disk and adjust its dirty state after completion. |
Christoph Hellwig | c376501 | 2010-10-06 10:48:20 +0200 | [diff] [blame] | 1407 | * |
| 1408 | * Note: only writes the actual inode, no associated data or other metadata. |
| 1409 | */ |
| 1410 | int sync_inode_metadata(struct inode *inode, int wait) |
| 1411 | { |
| 1412 | struct writeback_control wbc = { |
| 1413 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, |
| 1414 | .nr_to_write = 0, /* metadata-only */ |
| 1415 | }; |
| 1416 | |
| 1417 | return sync_inode(inode, &wbc); |
| 1418 | } |
| 1419 | EXPORT_SYMBOL(sync_inode_metadata); |