Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * fs/fs-writeback.c |
| 3 | * |
| 4 | * Copyright (C) 2002, Linus Torvalds. |
| 5 | * |
| 6 | * Contains all the functions related to writing back and waiting |
| 7 | * upon dirty inodes against superblocks, and writing back dirty |
| 8 | * pages against inodes. ie: data writeback. Writeout of the |
| 9 | * inode itself is not handled here. |
| 10 | * |
Francois Cami | e1f8e87 | 2008-10-15 22:01:59 -0700 | [diff] [blame] | 11 | * 10Apr2002 Andrew Morton |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 12 | * Split out of fs/inode.c |
| 13 | * Additions for address_space-based writeback |
| 14 | */ |
| 15 | |
| 16 | #include <linux/kernel.h> |
Jens Axboe | f5ff842 | 2007-09-21 09:19:54 +0200 | [diff] [blame] | 17 | #include <linux/module.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include <linux/spinlock.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 19 | #include <linux/slab.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | #include <linux/sched.h> |
| 21 | #include <linux/fs.h> |
| 22 | #include <linux/mm.h> |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 23 | #include <linux/kthread.h> |
| 24 | #include <linux/freezer.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include <linux/writeback.h> |
| 26 | #include <linux/blkdev.h> |
| 27 | #include <linux/backing-dev.h> |
| 28 | #include <linux/buffer_head.h> |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 29 | #include <linux/tracepoint.h> |
David Howells | 07f3f05 | 2006-09-30 20:52:18 +0200 | [diff] [blame] | 30 | #include "internal.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 31 | |
Jens Axboe | d0bceac | 2009-05-18 08:20:32 +0200 | [diff] [blame] | 32 | /* |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 33 | * Passed into wb_writeback(), essentially a subset of writeback_control |
| 34 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 35 | struct wb_writeback_work { |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 36 | long nr_pages; |
| 37 | struct super_block *sb; |
| 38 | enum writeback_sync_modes sync_mode; |
H Hartley Sweeten | 52957fe | 2010-04-01 20:36:30 -0500 | [diff] [blame] | 39 | unsigned int for_kupdate:1; |
| 40 | unsigned int range_cyclic:1; |
| 41 | unsigned int for_background:1; |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 42 | |
Jens Axboe | 8010c3b | 2009-09-15 20:04:57 +0200 | [diff] [blame] | 43 | struct list_head list; /* pending work list */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 44 | struct completion *done; /* set if the caller waits */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 45 | }; |
| 46 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 47 | /* |
| 48 | * Include the creation of the trace points after defining the |
| 49 | * wb_writeback_work structure so that the definition remains local to this |
| 50 | * file. |
| 51 | */ |
| 52 | #define CREATE_TRACE_POINTS |
| 53 | #include <trace/events/writeback.h> |
| 54 | |
| 55 | #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info) |
| 56 | |
| 57 | /* |
| 58 | * We don't actually have pdflush, but this one is exported though /proc... |
| 59 | */ |
| 60 | int nr_pdflush_threads; |
| 61 | |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 62 | /** |
| 63 | * writeback_in_progress - determine whether there is writeback in progress |
| 64 | * @bdi: the device's backing_dev_info structure. |
| 65 | * |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 66 | * Determine whether there is writeback waiting to be handled against a |
| 67 | * backing device. |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 68 | */ |
| 69 | int writeback_in_progress(struct backing_dev_info *bdi) |
| 70 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 71 | return !list_empty(&bdi->work_list); |
Adrian Bunk | f11b00f | 2008-04-29 00:58:56 -0700 | [diff] [blame] | 72 | } |
| 73 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 74 | static void bdi_queue_work(struct backing_dev_info *bdi, |
| 75 | struct wb_writeback_work *work) |
Nick Piggin | 4195f73 | 2009-05-28 09:01:15 +0200 | [diff] [blame] | 76 | { |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 77 | trace_writeback_queue(bdi, work); |
| 78 | |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 79 | spin_lock_bh(&bdi->wb_lock); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 80 | list_add_tail(&work->list, &bdi->work_list); |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 81 | if (bdi->wb.task) { |
| 82 | wake_up_process(bdi->wb.task); |
| 83 | } else { |
| 84 | /* |
| 85 | * The bdi thread isn't there, wake up the forker thread which |
| 86 | * will create and run it. |
| 87 | */ |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 88 | trace_writeback_nothread(bdi, work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 89 | wake_up_process(default_backing_dev_info.wb.task); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 90 | } |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 91 | spin_unlock_bh(&bdi->wb_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 92 | } |
| 93 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 94 | static void |
| 95 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
| 96 | bool range_cyclic, bool for_background) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 97 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 98 | struct wb_writeback_work *work; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 99 | |
Jens Axboe | bcddc3f | 2009-09-13 20:07:36 +0200 | [diff] [blame] | 100 | /* |
| 101 | * This is WB_SYNC_NONE writeback, so if allocation fails just |
| 102 | * wakeup the thread for old dirty data writeback |
| 103 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 104 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
| 105 | if (!work) { |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 106 | if (bdi->wb.task) { |
| 107 | trace_writeback_nowork(bdi); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 108 | wake_up_process(bdi->wb.task); |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 109 | } |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 110 | return; |
Jens Axboe | bcddc3f | 2009-09-13 20:07:36 +0200 | [diff] [blame] | 111 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 112 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 113 | work->sync_mode = WB_SYNC_NONE; |
| 114 | work->nr_pages = nr_pages; |
| 115 | work->range_cyclic = range_cyclic; |
| 116 | work->for_background = for_background; |
Christoph Hellwig | f0fad8a | 2009-09-11 09:47:56 +0200 | [diff] [blame] | 117 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 118 | bdi_queue_work(bdi, work); |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 119 | } |
| 120 | |
| 121 | /** |
| 122 | * bdi_start_writeback - start writeback |
| 123 | * @bdi: the backing device to write from |
| 124 | * @nr_pages: the number of pages to write |
| 125 | * |
| 126 | * Description: |
| 127 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only |
| 128 | * started when this function returns, we make no guarentees on |
Jens Axboe | 0e3c9a2 | 2010-06-01 11:08:43 +0200 | [diff] [blame] | 129 | * completion. Caller need not hold sb s_umount semaphore. |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 130 | * |
| 131 | */ |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 132 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 133 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 134 | __bdi_start_writeback(bdi, nr_pages, true, false); |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 135 | } |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 136 | |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 137 | /** |
| 138 | * bdi_start_background_writeback - start background writeback |
| 139 | * @bdi: the backing device to write from |
| 140 | * |
| 141 | * Description: |
| 142 | * This does WB_SYNC_NONE background writeback. The IO is only |
| 143 | * started when this function returns, we make no guarentees on |
| 144 | * completion. Caller need not hold sb s_umount semaphore. |
| 145 | */ |
| 146 | void bdi_start_background_writeback(struct backing_dev_info *bdi) |
| 147 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 148 | __bdi_start_writeback(bdi, LONG_MAX, true, true); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 149 | } |
| 150 | |
| 151 | /* |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 152 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the |
| 153 | * furthest end of its superblock's dirty-inode list. |
| 154 | * |
| 155 | * 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] | 156 | * 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] | 157 | * the case then the inode must have been redirtied while it was being written |
| 158 | * out and we don't reset its dirtied_when. |
| 159 | */ |
| 160 | static void redirty_tail(struct inode *inode) |
| 161 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 162 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 163 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 164 | if (!list_empty(&wb->b_dirty)) { |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 165 | struct inode *tail; |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 166 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 167 | tail = list_entry(wb->b_dirty.next, struct inode, i_list); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 168 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 169 | inode->dirtied_when = jiffies; |
| 170 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 171 | list_move(&inode->i_list, &wb->b_dirty); |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 172 | } |
| 173 | |
| 174 | /* |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 175 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 176 | */ |
Ken Chen | 0e0f4fc | 2007-10-16 23:30:38 -0700 | [diff] [blame] | 177 | static void requeue_io(struct inode *inode) |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 178 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 179 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
| 180 | |
| 181 | list_move(&inode->i_list, &wb->b_more_io); |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 182 | } |
| 183 | |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 184 | static void inode_sync_complete(struct inode *inode) |
| 185 | { |
| 186 | /* |
| 187 | * Prevent speculative execution through spin_unlock(&inode_lock); |
| 188 | */ |
| 189 | smp_mb(); |
| 190 | wake_up_bit(&inode->i_state, __I_SYNC); |
| 191 | } |
| 192 | |
Jeff Layton | d2caa3c | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 193 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
| 194 | { |
| 195 | bool ret = time_after(inode->dirtied_when, t); |
| 196 | #ifndef CONFIG_64BIT |
| 197 | /* |
| 198 | * For inodes being constantly redirtied, dirtied_when can get stuck. |
| 199 | * It _appears_ to be in the future, but is actually in distant past. |
| 200 | * This test is necessary to prevent such wrapped-around relative times |
Jens Axboe | 5b0830c | 2009-09-23 19:37:09 +0200 | [diff] [blame] | 201 | * from permanently stopping the whole bdi writeback. |
Jeff Layton | d2caa3c | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 202 | */ |
| 203 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); |
| 204 | #endif |
| 205 | return ret; |
| 206 | } |
| 207 | |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 208 | /* |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 209 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. |
| 210 | */ |
| 211 | static void move_expired_inodes(struct list_head *delaying_queue, |
| 212 | struct list_head *dispatch_queue, |
| 213 | unsigned long *older_than_this) |
| 214 | { |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 215 | LIST_HEAD(tmp); |
| 216 | struct list_head *pos, *node; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 217 | struct super_block *sb = NULL; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 218 | struct inode *inode; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 219 | int do_sb_sort = 0; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 220 | |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 221 | while (!list_empty(delaying_queue)) { |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 222 | inode = list_entry(delaying_queue->prev, struct inode, i_list); |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 223 | if (older_than_this && |
Jeff Layton | d2caa3c | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 224 | inode_dirtied_after(inode, *older_than_this)) |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 225 | break; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 226 | if (sb && sb != inode->i_sb) |
| 227 | do_sb_sort = 1; |
| 228 | sb = inode->i_sb; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 229 | list_move(&inode->i_list, &tmp); |
| 230 | } |
| 231 | |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 232 | /* just one sb in list, splice to dispatch_queue and we're done */ |
| 233 | if (!do_sb_sort) { |
| 234 | list_splice(&tmp, dispatch_queue); |
| 235 | return; |
| 236 | } |
| 237 | |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 238 | /* Move inodes from one superblock together */ |
| 239 | while (!list_empty(&tmp)) { |
| 240 | inode = list_entry(tmp.prev, struct inode, i_list); |
| 241 | sb = inode->i_sb; |
| 242 | list_for_each_prev_safe(pos, node, &tmp) { |
| 243 | inode = list_entry(pos, struct inode, i_list); |
| 244 | if (inode->i_sb == sb) |
| 245 | list_move(&inode->i_list, dispatch_queue); |
| 246 | } |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 247 | } |
| 248 | } |
| 249 | |
| 250 | /* |
| 251 | * Queue all expired dirty inodes for io, eldest first. |
| 252 | */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 253 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 254 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 255 | list_splice_init(&wb->b_more_io, wb->b_io.prev); |
| 256 | move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 257 | } |
| 258 | |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 259 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 260 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 261 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 262 | return inode->i_sb->s_op->write_inode(inode, wbc); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 263 | return 0; |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 264 | } |
| 265 | |
| 266 | /* |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 267 | * Wait for writeback on an inode to complete. |
| 268 | */ |
| 269 | static void inode_wait_for_writeback(struct inode *inode) |
| 270 | { |
| 271 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); |
| 272 | wait_queue_head_t *wqh; |
| 273 | |
| 274 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); |
Richard Kennedy | 58a9d3d | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 275 | while (inode->i_state & I_SYNC) { |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 276 | spin_unlock(&inode_lock); |
| 277 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); |
| 278 | spin_lock(&inode_lock); |
Richard Kennedy | 58a9d3d | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 279 | } |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 280 | } |
| 281 | |
| 282 | /* |
| 283 | * Write out an inode's dirty pages. Called under inode_lock. Either the |
| 284 | * caller has ref on the inode (either via __iget or via syscall against an fd) |
| 285 | * or the inode has I_WILL_FREE set (via generic_forget_inode) |
| 286 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 287 | * If `wait' is set, wait on the writeout. |
| 288 | * |
| 289 | * The whole writeout design is quite complex and fragile. We want to avoid |
| 290 | * starvation of particular inodes when others are being redirtied, prevent |
| 291 | * livelocks, etc. |
| 292 | * |
| 293 | * Called under inode_lock. |
| 294 | */ |
| 295 | static int |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 296 | writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 297 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 298 | struct address_space *mapping = inode->i_mapping; |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 299 | unsigned dirty; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 300 | int ret; |
| 301 | |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 302 | if (!atomic_read(&inode->i_count)) |
| 303 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); |
| 304 | else |
| 305 | WARN_ON(inode->i_state & I_WILL_FREE); |
| 306 | |
| 307 | if (inode->i_state & I_SYNC) { |
| 308 | /* |
| 309 | * If this inode is locked for writeback and we are not doing |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 310 | * writeback-for-data-integrity, move it to b_more_io so that |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 311 | * writeback can proceed with the other inodes on s_io. |
| 312 | * |
| 313 | * We'll have another go at writing back this inode when we |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 314 | * completed a full scan of b_io. |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 315 | */ |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 316 | if (wbc->sync_mode != WB_SYNC_ALL) { |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 317 | requeue_io(inode); |
| 318 | return 0; |
| 319 | } |
| 320 | |
| 321 | /* |
| 322 | * It's a data-integrity sync. We must wait. |
| 323 | */ |
| 324 | inode_wait_for_writeback(inode); |
| 325 | } |
| 326 | |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 327 | BUG_ON(inode->i_state & I_SYNC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 328 | |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 329 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 330 | inode->i_state |= I_SYNC; |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 331 | inode->i_state &= ~I_DIRTY_PAGES; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | spin_unlock(&inode_lock); |
| 333 | |
| 334 | ret = do_writepages(mapping, wbc); |
| 335 | |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 336 | /* |
| 337 | * Make sure to wait on the data before writing out the metadata. |
| 338 | * This is important for filesystems that modify metadata on data |
| 339 | * I/O completion. |
| 340 | */ |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 341 | if (wbc->sync_mode == WB_SYNC_ALL) { |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 342 | int err = filemap_fdatawait(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 343 | if (ret == 0) |
| 344 | ret = err; |
| 345 | } |
| 346 | |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 347 | /* |
| 348 | * Some filesystems may redirty the inode during the writeback |
| 349 | * due to delalloc, clear dirty metadata flags right before |
| 350 | * write_inode() |
| 351 | */ |
| 352 | spin_lock(&inode_lock); |
| 353 | dirty = inode->i_state & I_DIRTY; |
| 354 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); |
| 355 | spin_unlock(&inode_lock); |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 356 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
| 357 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 358 | int err = write_inode(inode, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 359 | if (ret == 0) |
| 360 | ret = err; |
| 361 | } |
| 362 | |
| 363 | spin_lock(&inode_lock); |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 364 | inode->i_state &= ~I_SYNC; |
Al Viro | a4ffdde | 2010-06-02 17:38:30 -0400 | [diff] [blame] | 365 | if (!(inode->i_state & I_FREEING)) { |
Wu Fengguang | b3af946 | 2009-09-25 06:04:10 +0200 | [diff] [blame] | 366 | if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) { |
Wu Fengguang | ae1b7f7 | 2009-09-23 20:33:42 +0800 | [diff] [blame] | 367 | /* |
Wu Fengguang | b3af946 | 2009-09-25 06:04:10 +0200 | [diff] [blame] | 368 | * More pages get dirtied by a fast dirtier. |
| 369 | */ |
| 370 | goto select_queue; |
| 371 | } else if (inode->i_state & I_DIRTY) { |
| 372 | /* |
| 373 | * At least XFS will redirty the inode during the |
| 374 | * writeback (delalloc) and on io completion (isize). |
Wu Fengguang | ae1b7f7 | 2009-09-23 20:33:42 +0800 | [diff] [blame] | 375 | */ |
| 376 | redirty_tail(inode); |
| 377 | } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 378 | /* |
| 379 | * We didn't write back all the pages. nfs_writepages() |
| 380 | * sometimes bales out without doing anything. Redirty |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 381 | * the inode; Move it from b_io onto b_more_io/b_dirty. |
Andrew Morton | 1b43ef9 | 2007-10-16 23:30:35 -0700 | [diff] [blame] | 382 | */ |
| 383 | /* |
| 384 | * akpm: if the caller was the kupdate function we put |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 385 | * this inode at the head of b_dirty so it gets first |
Andrew Morton | 1b43ef9 | 2007-10-16 23:30:35 -0700 | [diff] [blame] | 386 | * consideration. Otherwise, move it to the tail, for |
| 387 | * the reasons described there. I'm not really sure |
| 388 | * how much sense this makes. Presumably I had a good |
| 389 | * reasons for doing it this way, and I'd rather not |
| 390 | * muck with it at present. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 391 | */ |
| 392 | if (wbc->for_kupdate) { |
| 393 | /* |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 394 | * For the kupdate function we move the inode |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 395 | * to b_more_io so it will get more writeout as |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 396 | * soon as the queue becomes uncongested. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 397 | */ |
| 398 | inode->i_state |= I_DIRTY_PAGES; |
Wu Fengguang | b3af946 | 2009-09-25 06:04:10 +0200 | [diff] [blame] | 399 | select_queue: |
Fengguang Wu | 8bc3be2 | 2008-02-04 22:29:36 -0800 | [diff] [blame] | 400 | if (wbc->nr_to_write <= 0) { |
| 401 | /* |
| 402 | * slice used up: queue for next turn |
| 403 | */ |
| 404 | requeue_io(inode); |
| 405 | } else { |
| 406 | /* |
| 407 | * somehow blocked: retry later |
| 408 | */ |
| 409 | redirty_tail(inode); |
| 410 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 411 | } else { |
| 412 | /* |
| 413 | * Otherwise fully redirty the inode so that |
| 414 | * other inodes on this superblock will get some |
| 415 | * writeout. Otherwise heavy writing to one |
| 416 | * file would indefinitely suspend writeout of |
| 417 | * all the other files. |
| 418 | */ |
| 419 | inode->i_state |= I_DIRTY_PAGES; |
Andrew Morton | 1b43ef9 | 2007-10-16 23:30:35 -0700 | [diff] [blame] | 420 | redirty_tail(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 421 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 422 | } else if (atomic_read(&inode->i_count)) { |
| 423 | /* |
| 424 | * The inode is clean, inuse |
| 425 | */ |
| 426 | list_move(&inode->i_list, &inode_in_use); |
| 427 | } else { |
| 428 | /* |
| 429 | * The inode is clean, unused |
| 430 | */ |
| 431 | list_move(&inode->i_list, &inode_unused); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 432 | } |
| 433 | } |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 434 | inode_sync_complete(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 435 | return ret; |
| 436 | } |
| 437 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 438 | /* |
Christoph Hellwig | d19de7e | 2010-06-08 18:14:58 +0200 | [diff] [blame] | 439 | * For background writeback the caller does not have the sb pinned |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 440 | * before calling writeback. So make sure that we do pin it, so it doesn't |
| 441 | * go away while we are writing inodes from it. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 442 | */ |
Christoph Hellwig | d19de7e | 2010-06-08 18:14:58 +0200 | [diff] [blame] | 443 | static bool pin_sb_for_writeback(struct super_block *sb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 444 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 445 | spin_lock(&sb_lock); |
Christoph Hellwig | 29cb485 | 2010-06-09 15:31:01 +0200 | [diff] [blame] | 446 | if (list_empty(&sb->s_instances)) { |
| 447 | spin_unlock(&sb_lock); |
| 448 | return false; |
| 449 | } |
| 450 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 451 | sb->s_count++; |
Christoph Hellwig | 29cb485 | 2010-06-09 15:31:01 +0200 | [diff] [blame] | 452 | spin_unlock(&sb_lock); |
| 453 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 454 | if (down_read_trylock(&sb->s_umount)) { |
Christoph Hellwig | 29cb485 | 2010-06-09 15:31:01 +0200 | [diff] [blame] | 455 | if (sb->s_root) |
Christoph Hellwig | d19de7e | 2010-06-08 18:14:58 +0200 | [diff] [blame] | 456 | return true; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 457 | up_read(&sb->s_umount); |
| 458 | } |
Christoph Hellwig | 29cb485 | 2010-06-09 15:31:01 +0200 | [diff] [blame] | 459 | |
| 460 | put_super(sb); |
Christoph Hellwig | d19de7e | 2010-06-08 18:14:58 +0200 | [diff] [blame] | 461 | return false; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 462 | } |
| 463 | |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 464 | /* |
| 465 | * Write a portion of b_io inodes which belong to @sb. |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 466 | * |
| 467 | * If @only_this_sb is true, then find and write all such |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 468 | * inodes. Otherwise write only ones which go sequentially |
| 469 | * in reverse order. |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 470 | * |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 471 | * Return 1, if the caller writeback routine should be |
| 472 | * interrupted. Otherwise return 0. |
| 473 | */ |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 474 | static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb, |
| 475 | struct writeback_control *wbc, bool only_this_sb) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 476 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 477 | while (!list_empty(&wb->b_io)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 478 | long pages_skipped; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 479 | struct inode *inode = list_entry(wb->b_io.prev, |
| 480 | struct inode, i_list); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 481 | |
| 482 | if (inode->i_sb != sb) { |
| 483 | if (only_this_sb) { |
| 484 | /* |
| 485 | * We only want to write back data for this |
| 486 | * superblock, move all inodes not belonging |
| 487 | * to it back onto the dirty list. |
| 488 | */ |
| 489 | redirty_tail(inode); |
| 490 | continue; |
| 491 | } |
| 492 | |
| 493 | /* |
| 494 | * The inode belongs to a different superblock. |
| 495 | * Bounce back to the caller to unpin this and |
| 496 | * pin the next superblock. |
| 497 | */ |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 498 | return 0; |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 499 | } |
| 500 | |
Wu Fengguang | 84a8924 | 2009-06-16 15:33:17 -0700 | [diff] [blame] | 501 | if (inode->i_state & (I_NEW | I_WILL_FREE)) { |
Nick Piggin | 7ef0d73 | 2009-03-12 14:31:38 -0700 | [diff] [blame] | 502 | requeue_io(inode); |
| 503 | continue; |
| 504 | } |
Jeff Layton | d2caa3c | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 505 | /* |
| 506 | * Was this inode dirtied after sync_sb_inodes was called? |
| 507 | * This keeps sync from extra jobs and livelock. |
| 508 | */ |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 509 | if (inode_dirtied_after(inode, wbc->wb_start)) |
| 510 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 511 | |
Al Viro | a4ffdde | 2010-06-02 17:38:30 -0400 | [diff] [blame] | 512 | BUG_ON(inode->i_state & I_FREEING); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 513 | __iget(inode); |
| 514 | pages_skipped = wbc->pages_skipped; |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 515 | writeback_single_inode(inode, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 516 | if (wbc->pages_skipped != pages_skipped) { |
| 517 | /* |
| 518 | * writeback is not making progress due to locked |
| 519 | * buffers. Skip this inode for now. |
| 520 | */ |
Andrew Morton | f57b9b7 | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 521 | redirty_tail(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 522 | } |
| 523 | spin_unlock(&inode_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 524 | iput(inode); |
OGAWA Hirofumi | 4ffc844 | 2006-03-25 03:07:44 -0800 | [diff] [blame] | 525 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 526 | spin_lock(&inode_lock); |
Fengguang Wu | 8bc3be2 | 2008-02-04 22:29:36 -0800 | [diff] [blame] | 527 | if (wbc->nr_to_write <= 0) { |
| 528 | wbc->more_io = 1; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 529 | return 1; |
Fengguang Wu | 8bc3be2 | 2008-02-04 22:29:36 -0800 | [diff] [blame] | 530 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 531 | if (!list_empty(&wb->b_more_io)) |
Fengguang Wu | 8bc3be2 | 2008-02-04 22:29:36 -0800 | [diff] [blame] | 532 | wbc->more_io = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 533 | } |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 534 | /* b_io is empty */ |
| 535 | return 1; |
| 536 | } |
Nick Piggin | 38f2197 | 2009-01-06 14:40:25 -0800 | [diff] [blame] | 537 | |
Christoph Hellwig | 9c3a8ee | 2010-06-10 12:07:27 +0200 | [diff] [blame] | 538 | void writeback_inodes_wb(struct bdi_writeback *wb, |
| 539 | struct writeback_control *wbc) |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 540 | { |
| 541 | int ret = 0; |
Jens Axboe | 9ecc273 | 2009-09-24 15:25:11 +0200 | [diff] [blame] | 542 | |
Jan Kara | 7624ee7 | 2010-08-09 17:20:03 -0700 | [diff] [blame] | 543 | if (!wbc->wb_start) |
| 544 | wbc->wb_start = jiffies; /* livelock avoidance */ |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 545 | spin_lock(&inode_lock); |
| 546 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) |
| 547 | queue_io(wb, wbc->older_than_this); |
| 548 | |
| 549 | while (!list_empty(&wb->b_io)) { |
| 550 | struct inode *inode = list_entry(wb->b_io.prev, |
| 551 | struct inode, i_list); |
| 552 | struct super_block *sb = inode->i_sb; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 553 | |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 554 | if (!pin_sb_for_writeback(sb)) { |
| 555 | requeue_io(inode); |
| 556 | continue; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 557 | } |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 558 | ret = writeback_sb_inodes(sb, wb, wbc, false); |
| 559 | drop_super(sb); |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 560 | |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 561 | if (ret) |
| 562 | break; |
| 563 | } |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 564 | spin_unlock(&inode_lock); |
| 565 | /* Leave any unwritten inodes on b_io */ |
| 566 | } |
| 567 | |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 568 | static void __writeback_inodes_sb(struct super_block *sb, |
| 569 | struct bdi_writeback *wb, struct writeback_control *wbc) |
| 570 | { |
| 571 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
| 572 | |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 573 | spin_lock(&inode_lock); |
| 574 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) |
| 575 | queue_io(wb, wbc->older_than_this); |
| 576 | writeback_sb_inodes(sb, wb, wbc, true); |
| 577 | spin_unlock(&inode_lock); |
| 578 | } |
| 579 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 580 | /* |
| 581 | * The maximum number of pages to writeout in a single bdi flush/kupdate |
| 582 | * operation. We do this so we don't hold I_SYNC against an inode for |
| 583 | * enormous amounts of time, which would block a userspace task which has |
| 584 | * been forced to throttle against that inode. Also, the code reevaluates |
| 585 | * the dirty each time it has written this many pages. |
| 586 | */ |
| 587 | #define MAX_WRITEBACK_PAGES 1024 |
| 588 | |
| 589 | static inline bool over_bground_thresh(void) |
| 590 | { |
| 591 | unsigned long background_thresh, dirty_thresh; |
| 592 | |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 593 | global_dirty_limits(&background_thresh, &dirty_thresh); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 594 | |
| 595 | return (global_page_state(NR_FILE_DIRTY) + |
| 596 | global_page_state(NR_UNSTABLE_NFS) >= background_thresh); |
| 597 | } |
| 598 | |
| 599 | /* |
| 600 | * Explicit flushing or periodic writeback of "old" data. |
| 601 | * |
| 602 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
| 603 | * dirtying-time in the inode's address_space. So this periodic writeback code |
| 604 | * just walks the superblock inode list, writing back any inodes which are |
| 605 | * older than a specific point in time. |
| 606 | * |
| 607 | * Try to run once per dirty_writeback_interval. But if a writeback event |
| 608 | * takes longer than a dirty_writeback_interval interval, then leave a |
| 609 | * one-second gap. |
| 610 | * |
| 611 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
| 612 | * all dirty pages if they are all attached to "old" mappings. |
| 613 | */ |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 614 | static long wb_writeback(struct bdi_writeback *wb, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 615 | struct wb_writeback_work *work) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 616 | { |
| 617 | struct writeback_control wbc = { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 618 | .sync_mode = work->sync_mode, |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 619 | .older_than_this = NULL, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 620 | .for_kupdate = work->for_kupdate, |
| 621 | .for_background = work->for_background, |
| 622 | .range_cyclic = work->range_cyclic, |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 623 | }; |
| 624 | unsigned long oldest_jif; |
| 625 | long wrote = 0; |
Jan Kara | a5989bd | 2009-09-16 19:22:48 +0200 | [diff] [blame] | 626 | struct inode *inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 627 | |
| 628 | if (wbc.for_kupdate) { |
| 629 | wbc.older_than_this = &oldest_jif; |
| 630 | oldest_jif = jiffies - |
| 631 | msecs_to_jiffies(dirty_expire_interval * 10); |
| 632 | } |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 633 | if (!wbc.range_cyclic) { |
| 634 | wbc.range_start = 0; |
| 635 | wbc.range_end = LLONG_MAX; |
| 636 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 637 | |
Jan Kara | 7624ee7 | 2010-08-09 17:20:03 -0700 | [diff] [blame] | 638 | wbc.wb_start = jiffies; /* livelock avoidance */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 639 | for (;;) { |
| 640 | /* |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 641 | * Stop writeback when nr_pages has been consumed |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 642 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 643 | if (work->nr_pages <= 0) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 644 | break; |
| 645 | |
| 646 | /* |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 647 | * For background writeout, stop when we are below the |
| 648 | * background dirty threshold |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 649 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 650 | if (work->for_background && !over_bground_thresh()) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 651 | break; |
| 652 | |
| 653 | wbc.more_io = 0; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 654 | wbc.nr_to_write = MAX_WRITEBACK_PAGES; |
| 655 | wbc.pages_skipped = 0; |
Dave Chinner | 028c2dd | 2010-07-07 13:24:07 +1000 | [diff] [blame] | 656 | |
| 657 | trace_wbc_writeback_start(&wbc, wb->bdi); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 658 | if (work->sb) |
| 659 | __writeback_inodes_sb(work->sb, wb, &wbc); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 660 | else |
| 661 | writeback_inodes_wb(wb, &wbc); |
Dave Chinner | 028c2dd | 2010-07-07 13:24:07 +1000 | [diff] [blame] | 662 | trace_wbc_writeback_written(&wbc, wb->bdi); |
| 663 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 664 | work->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 665 | wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
| 666 | |
| 667 | /* |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 668 | * If we consumed everything, see if we have more |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 669 | */ |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 670 | if (wbc.nr_to_write <= 0) |
| 671 | continue; |
| 672 | /* |
| 673 | * Didn't write everything and we don't have more IO, bail |
| 674 | */ |
| 675 | if (!wbc.more_io) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 676 | break; |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 677 | /* |
| 678 | * Did we write something? Try for more |
| 679 | */ |
| 680 | if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) |
| 681 | continue; |
| 682 | /* |
| 683 | * Nothing written. Wait for some inode to |
| 684 | * become available for writeback. Otherwise |
| 685 | * we'll just busyloop. |
| 686 | */ |
| 687 | spin_lock(&inode_lock); |
| 688 | if (!list_empty(&wb->b_more_io)) { |
| 689 | inode = list_entry(wb->b_more_io.prev, |
| 690 | struct inode, i_list); |
Dave Chinner | 028c2dd | 2010-07-07 13:24:07 +1000 | [diff] [blame] | 691 | trace_wbc_writeback_wait(&wbc, wb->bdi); |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 692 | inode_wait_for_writeback(inode); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 693 | } |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 694 | spin_unlock(&inode_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 695 | } |
| 696 | |
| 697 | return wrote; |
| 698 | } |
| 699 | |
| 700 | /* |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 701 | * 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] | 702 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 703 | static struct wb_writeback_work * |
Minchan Kim | 08852b6 | 2010-08-03 12:51:16 +0200 | [diff] [blame] | 704 | get_next_work_item(struct backing_dev_info *bdi) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 705 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 706 | struct wb_writeback_work *work = NULL; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 707 | |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 708 | spin_lock_bh(&bdi->wb_lock); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 709 | if (!list_empty(&bdi->work_list)) { |
| 710 | work = list_entry(bdi->work_list.next, |
| 711 | struct wb_writeback_work, list); |
| 712 | list_del_init(&work->list); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 713 | } |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 714 | spin_unlock_bh(&bdi->wb_lock); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 715 | return work; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 716 | } |
| 717 | |
| 718 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
| 719 | { |
| 720 | unsigned long expired; |
| 721 | long nr_pages; |
| 722 | |
Jens Axboe | 69b62d0 | 2010-05-17 12:51:03 +0200 | [diff] [blame] | 723 | /* |
| 724 | * When set to zero, disable periodic writeback |
| 725 | */ |
| 726 | if (!dirty_writeback_interval) |
| 727 | return 0; |
| 728 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 729 | expired = wb->last_old_flush + |
| 730 | msecs_to_jiffies(dirty_writeback_interval * 10); |
| 731 | if (time_before(jiffies, expired)) |
| 732 | return 0; |
| 733 | |
| 734 | wb->last_old_flush = jiffies; |
| 735 | nr_pages = global_page_state(NR_FILE_DIRTY) + |
| 736 | global_page_state(NR_UNSTABLE_NFS) + |
| 737 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); |
| 738 | |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 739 | if (nr_pages) { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 740 | struct wb_writeback_work work = { |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 741 | .nr_pages = nr_pages, |
| 742 | .sync_mode = WB_SYNC_NONE, |
| 743 | .for_kupdate = 1, |
| 744 | .range_cyclic = 1, |
| 745 | }; |
| 746 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 747 | return wb_writeback(wb, &work); |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 748 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 749 | |
| 750 | return 0; |
| 751 | } |
| 752 | |
| 753 | /* |
| 754 | * Retrieve work items and do the writeback they describe |
| 755 | */ |
| 756 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) |
| 757 | { |
| 758 | struct backing_dev_info *bdi = wb->bdi; |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 759 | struct wb_writeback_work *work; |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 760 | long wrote = 0; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 761 | |
Minchan Kim | 08852b6 | 2010-08-03 12:51:16 +0200 | [diff] [blame] | 762 | while ((work = get_next_work_item(bdi)) != NULL) { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 763 | /* |
| 764 | * Override sync mode, in case we must wait for completion |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 765 | * because this thread is exiting now. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 766 | */ |
| 767 | if (force_wait) |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 768 | work->sync_mode = WB_SYNC_ALL; |
| 769 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 770 | trace_writeback_exec(bdi, work); |
| 771 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 772 | wrote += wb_writeback(wb, work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 773 | |
| 774 | /* |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 775 | * Notify the caller of completion if this is a synchronous |
| 776 | * work item, otherwise just free it. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 777 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 778 | if (work->done) |
| 779 | complete(work->done); |
| 780 | else |
| 781 | kfree(work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 782 | } |
| 783 | |
| 784 | /* |
| 785 | * Check for periodic writeback, kupdated() style |
| 786 | */ |
| 787 | wrote += wb_check_old_data_flush(wb); |
| 788 | |
| 789 | return wrote; |
| 790 | } |
| 791 | |
| 792 | /* |
| 793 | * Handle writeback of dirty data for the device backed by this bdi. Also |
| 794 | * wakes up periodically and does kupdated style flushing. |
| 795 | */ |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 796 | int bdi_writeback_thread(void *data) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 797 | { |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 798 | struct bdi_writeback *wb = data; |
| 799 | struct backing_dev_info *bdi = wb->bdi; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 800 | long pages_written; |
| 801 | |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 802 | current->flags |= PF_FLUSHER | PF_SWAPWRITE; |
| 803 | set_freezable(); |
Artem Bityutskiy | ecd5840 | 2010-07-25 14:29:18 +0300 | [diff] [blame] | 804 | wb->last_active = jiffies; |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 805 | |
| 806 | /* |
| 807 | * Our parent may run at a different priority, just set us to normal |
| 808 | */ |
| 809 | set_user_nice(current, 0); |
| 810 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 811 | trace_writeback_thread_start(bdi); |
| 812 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 813 | while (!kthread_should_stop()) { |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 814 | /* |
| 815 | * Remove own delayed wake-up timer, since we are already awake |
| 816 | * and we'll take care of the preriodic write-back. |
| 817 | */ |
| 818 | del_timer(&wb->wakeup_timer); |
| 819 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 820 | pages_written = wb_do_writeback(wb, 0); |
| 821 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 822 | trace_writeback_pages_written(pages_written); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 823 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 824 | if (pages_written) |
Artem Bityutskiy | ecd5840 | 2010-07-25 14:29:18 +0300 | [diff] [blame] | 825 | wb->last_active = jiffies; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 826 | |
Artem Bityutskiy | 297252c | 2010-07-25 14:29:15 +0300 | [diff] [blame] | 827 | set_current_state(TASK_INTERRUPTIBLE); |
| 828 | if (!list_empty(&bdi->work_list)) { |
| 829 | __set_current_state(TASK_RUNNING); |
| 830 | continue; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 831 | } |
| 832 | |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 833 | if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 834 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 835 | else { |
| 836 | /* |
| 837 | * We have nothing to do, so can go sleep without any |
| 838 | * timeout and save power. When a work is queued or |
| 839 | * something is made dirty - we will be woken up. |
| 840 | */ |
Artem Bityutskiy | 297252c | 2010-07-25 14:29:15 +0300 | [diff] [blame] | 841 | schedule(); |
Jens Axboe | f9eadbb | 2010-05-18 14:31:45 +0200 | [diff] [blame] | 842 | } |
Jens Axboe | 69b62d0 | 2010-05-17 12:51:03 +0200 | [diff] [blame] | 843 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 844 | try_to_freeze(); |
| 845 | } |
| 846 | |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 847 | /* Flush any work that raced with us exiting */ |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 848 | if (!list_empty(&bdi->work_list)) |
| 849 | wb_do_writeback(wb, 1); |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 850 | |
| 851 | trace_writeback_thread_stop(bdi); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 852 | return 0; |
| 853 | } |
| 854 | |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 855 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 856 | /* |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 857 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
| 858 | * the whole world. |
| 859 | */ |
| 860 | void wakeup_flusher_threads(long nr_pages) |
| 861 | { |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 862 | struct backing_dev_info *bdi; |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 863 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 864 | if (!nr_pages) { |
| 865 | nr_pages = global_page_state(NR_FILE_DIRTY) + |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 866 | global_page_state(NR_UNSTABLE_NFS); |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 867 | } |
| 868 | |
| 869 | rcu_read_lock(); |
| 870 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
| 871 | if (!bdi_has_dirty_io(bdi)) |
| 872 | continue; |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 873 | __bdi_start_writeback(bdi, nr_pages, false, false); |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 874 | } |
| 875 | rcu_read_unlock(); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 876 | } |
| 877 | |
| 878 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
| 879 | { |
| 880 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { |
| 881 | struct dentry *dentry; |
| 882 | const char *name = "?"; |
| 883 | |
| 884 | dentry = d_find_alias(inode); |
| 885 | if (dentry) { |
| 886 | spin_lock(&dentry->d_lock); |
| 887 | name = (const char *) dentry->d_name.name; |
| 888 | } |
| 889 | printk(KERN_DEBUG |
| 890 | "%s(%d): dirtied inode %lu (%s) on %s\n", |
| 891 | current->comm, task_pid_nr(current), inode->i_ino, |
| 892 | name, inode->i_sb->s_id); |
| 893 | if (dentry) { |
| 894 | spin_unlock(&dentry->d_lock); |
| 895 | dput(dentry); |
| 896 | } |
| 897 | } |
| 898 | } |
| 899 | |
| 900 | /** |
| 901 | * __mark_inode_dirty - internal function |
| 902 | * @inode: inode to mark |
| 903 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) |
| 904 | * Mark an inode as dirty. Callers should use mark_inode_dirty or |
| 905 | * mark_inode_dirty_sync. |
| 906 | * |
| 907 | * Put the inode on the super block's dirty list. |
| 908 | * |
| 909 | * CAREFUL! We mark it dirty unconditionally, but move it onto the |
| 910 | * dirty list only if it is hashed or if it refers to a blockdev. |
| 911 | * If it was not hashed, it will never be added to the dirty list |
| 912 | * even if it is later hashed, as it will have been marked dirty already. |
| 913 | * |
| 914 | * In short, make sure you hash any inodes _before_ you start marking |
| 915 | * them dirty. |
| 916 | * |
| 917 | * This function *must* be atomic for the I_DIRTY_PAGES case - |
| 918 | * set_page_dirty() is called under spinlock in several places. |
| 919 | * |
| 920 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
| 921 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of |
| 922 | * the kernel-internal blockdev inode represents the dirtying time of the |
| 923 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use |
| 924 | * page->mapping->host, so the page-dirtying time is recorded in the internal |
| 925 | * blockdev inode. |
| 926 | */ |
| 927 | void __mark_inode_dirty(struct inode *inode, int flags) |
| 928 | { |
| 929 | struct super_block *sb = inode->i_sb; |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 930 | struct backing_dev_info *bdi = NULL; |
| 931 | bool wakeup_bdi = false; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 932 | |
| 933 | /* |
| 934 | * Don't do this for I_DIRTY_PAGES - that doesn't actually |
| 935 | * dirty the inode itself |
| 936 | */ |
| 937 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { |
| 938 | if (sb->s_op->dirty_inode) |
| 939 | sb->s_op->dirty_inode(inode); |
| 940 | } |
| 941 | |
| 942 | /* |
| 943 | * make sure that changes are seen by all cpus before we test i_state |
| 944 | * -- mikulas |
| 945 | */ |
| 946 | smp_mb(); |
| 947 | |
| 948 | /* avoid the locking if we can */ |
| 949 | if ((inode->i_state & flags) == flags) |
| 950 | return; |
| 951 | |
| 952 | if (unlikely(block_dump)) |
| 953 | block_dump___mark_inode_dirty(inode); |
| 954 | |
| 955 | spin_lock(&inode_lock); |
| 956 | if ((inode->i_state & flags) != flags) { |
| 957 | const int was_dirty = inode->i_state & I_DIRTY; |
| 958 | |
| 959 | inode->i_state |= flags; |
| 960 | |
| 961 | /* |
| 962 | * If the inode is being synced, just update its dirty state. |
| 963 | * The unlocker will place the inode on the appropriate |
| 964 | * superblock list, based upon its state. |
| 965 | */ |
| 966 | if (inode->i_state & I_SYNC) |
| 967 | goto out; |
| 968 | |
| 969 | /* |
| 970 | * Only add valid (hashed) inodes to the superblock's |
| 971 | * dirty list. Add blockdev inodes as well. |
| 972 | */ |
| 973 | if (!S_ISBLK(inode->i_mode)) { |
| 974 | if (hlist_unhashed(&inode->i_hash)) |
| 975 | goto out; |
| 976 | } |
Al Viro | a4ffdde | 2010-06-02 17:38:30 -0400 | [diff] [blame] | 977 | if (inode->i_state & I_FREEING) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 978 | goto out; |
| 979 | |
| 980 | /* |
| 981 | * If the inode was already on b_dirty/b_io/b_more_io, don't |
| 982 | * reposition it (that would break b_dirty time-ordering). |
| 983 | */ |
| 984 | if (!was_dirty) { |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 985 | bdi = inode_to_bdi(inode); |
Jens Axboe | 500b067 | 2009-09-09 09:10:25 +0200 | [diff] [blame] | 986 | |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 987 | if (bdi_cap_writeback_dirty(bdi)) { |
| 988 | WARN(!test_bit(BDI_registered, &bdi->state), |
| 989 | "bdi-%s not registered\n", bdi->name); |
| 990 | |
| 991 | /* |
| 992 | * If this is the first dirty inode for this |
| 993 | * bdi, we have to wake-up the corresponding |
| 994 | * bdi thread to make sure background |
| 995 | * write-back happens later. |
| 996 | */ |
| 997 | if (!wb_has_dirty_io(&bdi->wb)) |
| 998 | wakeup_bdi = true; |
Jens Axboe | 500b067 | 2009-09-09 09:10:25 +0200 | [diff] [blame] | 999 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1000 | |
| 1001 | inode->dirtied_when = jiffies; |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1002 | list_move(&inode->i_list, &bdi->wb.b_dirty); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1003 | } |
| 1004 | } |
| 1005 | out: |
| 1006 | spin_unlock(&inode_lock); |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 1007 | |
| 1008 | if (wakeup_bdi) |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 1009 | bdi_wakeup_thread_delayed(bdi); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1010 | } |
| 1011 | EXPORT_SYMBOL(__mark_inode_dirty); |
| 1012 | |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1013 | /* |
| 1014 | * Write out a superblock's list of dirty inodes. A wait will be performed |
| 1015 | * upon no inodes, all inodes or the final one, depending upon sync_mode. |
| 1016 | * |
| 1017 | * If older_than_this is non-NULL, then only write out inodes which |
| 1018 | * had their first dirtying at a time earlier than *older_than_this. |
| 1019 | * |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1020 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
| 1021 | * This function assumes that the blockdev superblock's inodes are backed by |
| 1022 | * a variety of queues, so all inodes are searched. For other superblocks, |
| 1023 | * assume that all inodes are backed by the same queue. |
| 1024 | * |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1025 | * The inodes to be written are parked on bdi->b_io. They are moved back onto |
| 1026 | * bdi->b_dirty as they are selected for writing. This way, none can be missed |
| 1027 | * on the writer throttling path, and we get decent balancing between many |
| 1028 | * throttled threads: we don't want them all piling up on inode_sync_wait. |
| 1029 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1030 | static void wait_sb_inodes(struct super_block *sb) |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1031 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1032 | struct inode *inode, *old_inode = NULL; |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1033 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1034 | /* |
| 1035 | * We need to be protected against the filesystem going from |
| 1036 | * r/o to r/w or vice versa. |
| 1037 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1038 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1039 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1040 | spin_lock(&inode_lock); |
| 1041 | |
| 1042 | /* |
| 1043 | * Data integrity sync. Must wait for all pages under writeback, |
| 1044 | * because there may have been pages dirtied before our sync |
| 1045 | * call, but which had writeout started before we write it out. |
| 1046 | * In which case, the inode may not be on the dirty list, but |
| 1047 | * we still have to wait for that writeout. |
| 1048 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1049 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1050 | struct address_space *mapping; |
| 1051 | |
Al Viro | a4ffdde | 2010-06-02 17:38:30 -0400 | [diff] [blame] | 1052 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1053 | continue; |
| 1054 | mapping = inode->i_mapping; |
| 1055 | if (mapping->nrpages == 0) |
| 1056 | continue; |
| 1057 | __iget(inode); |
| 1058 | spin_unlock(&inode_lock); |
| 1059 | /* |
| 1060 | * We hold a reference to 'inode' so it couldn't have |
| 1061 | * been removed from s_inodes list while we dropped the |
| 1062 | * inode_lock. We cannot iput the inode now as we can |
| 1063 | * be holding the last reference and we cannot iput it |
| 1064 | * under inode_lock. So we keep the reference and iput |
| 1065 | * it later. |
| 1066 | */ |
| 1067 | iput(old_inode); |
| 1068 | old_inode = inode; |
| 1069 | |
| 1070 | filemap_fdatawait(mapping); |
| 1071 | |
| 1072 | cond_resched(); |
Nick Piggin | 38f2197 | 2009-01-06 14:40:25 -0800 | [diff] [blame] | 1073 | |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1074 | spin_lock(&inode_lock); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1075 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1076 | spin_unlock(&inode_lock); |
| 1077 | iput(old_inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1078 | } |
| 1079 | |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1080 | /** |
| 1081 | * writeback_inodes_sb - writeback dirty inodes from given super_block |
| 1082 | * @sb: the superblock |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1083 | * |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1084 | * Start writeback on some inodes on this super_block. No guarantees are made |
| 1085 | * on how many (if any) will be written, and this function does not wait |
| 1086 | * for IO completion of submitted IO. The number of pages submitted is |
| 1087 | * returned. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1088 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1089 | void writeback_inodes_sb(struct super_block *sb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1090 | { |
Jens Axboe | 0e3c9a2 | 2010-06-01 11:08:43 +0200 | [diff] [blame] | 1091 | unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); |
| 1092 | unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1093 | DECLARE_COMPLETION_ONSTACK(done); |
| 1094 | struct wb_writeback_work work = { |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1095 | .sb = sb, |
| 1096 | .sync_mode = WB_SYNC_NONE, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1097 | .done = &done, |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1098 | }; |
Jens Axboe | 0e3c9a2 | 2010-06-01 11:08:43 +0200 | [diff] [blame] | 1099 | |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1100 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
| 1101 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1102 | work.nr_pages = nr_dirty + nr_unstable + |
Jens Axboe | 0e3c9a2 | 2010-06-01 11:08:43 +0200 | [diff] [blame] | 1103 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); |
| 1104 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1105 | bdi_queue_work(sb->s_bdi, &work); |
| 1106 | wait_for_completion(&done); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1107 | } |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1108 | EXPORT_SYMBOL(writeback_inodes_sb); |
| 1109 | |
| 1110 | /** |
Eric Sandeen | 17bd55d | 2009-12-23 07:57:07 -0500 | [diff] [blame] | 1111 | * writeback_inodes_sb_if_idle - start writeback if none underway |
| 1112 | * @sb: the superblock |
| 1113 | * |
| 1114 | * Invoke writeback_inodes_sb if no writeback is currently underway. |
| 1115 | * Returns 1 if writeback was started, 0 if not. |
| 1116 | */ |
| 1117 | int writeback_inodes_sb_if_idle(struct super_block *sb) |
| 1118 | { |
| 1119 | if (!writeback_in_progress(sb->s_bdi)) { |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1120 | down_read(&sb->s_umount); |
Eric Sandeen | 17bd55d | 2009-12-23 07:57:07 -0500 | [diff] [blame] | 1121 | writeback_inodes_sb(sb); |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1122 | up_read(&sb->s_umount); |
Eric Sandeen | 17bd55d | 2009-12-23 07:57:07 -0500 | [diff] [blame] | 1123 | return 1; |
| 1124 | } else |
| 1125 | return 0; |
| 1126 | } |
| 1127 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); |
| 1128 | |
| 1129 | /** |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1130 | * sync_inodes_sb - sync sb inode pages |
| 1131 | * @sb: the superblock |
| 1132 | * |
| 1133 | * This function writes and waits on any dirty inode belonging to this |
| 1134 | * super_block. The number of pages synced is returned. |
| 1135 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1136 | void sync_inodes_sb(struct super_block *sb) |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1137 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1138 | DECLARE_COMPLETION_ONSTACK(done); |
| 1139 | struct wb_writeback_work work = { |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1140 | .sb = sb, |
| 1141 | .sync_mode = WB_SYNC_ALL, |
| 1142 | .nr_pages = LONG_MAX, |
| 1143 | .range_cyclic = 0, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1144 | .done = &done, |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 1145 | }; |
| 1146 | |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 1147 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
| 1148 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1149 | bdi_queue_work(sb->s_bdi, &work); |
| 1150 | wait_for_completion(&done); |
| 1151 | |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 1152 | wait_sb_inodes(sb); |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 1153 | } |
| 1154 | EXPORT_SYMBOL(sync_inodes_sb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1155 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1156 | /** |
Andrea Arcangeli | 7f04c26 | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 1157 | * write_inode_now - write an inode to disk |
| 1158 | * @inode: inode to write to disk |
| 1159 | * @sync: whether the write should be synchronous or not |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1160 | * |
Andrea Arcangeli | 7f04c26 | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 1161 | * This function commits an inode to disk immediately if it is dirty. This is |
| 1162 | * primarily needed by knfsd. |
| 1163 | * |
| 1164 | * 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] | 1165 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1166 | int write_inode_now(struct inode *inode, int sync) |
| 1167 | { |
| 1168 | int ret; |
| 1169 | struct writeback_control wbc = { |
| 1170 | .nr_to_write = LONG_MAX, |
Mike Galbraith | 18914b1 | 2008-02-08 04:20:23 -0800 | [diff] [blame] | 1171 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 1172 | .range_start = 0, |
| 1173 | .range_end = LLONG_MAX, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1174 | }; |
| 1175 | |
| 1176 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) |
Andrew Morton | 49364ce | 2005-11-07 00:59:15 -0800 | [diff] [blame] | 1177 | wbc.nr_to_write = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1178 | |
| 1179 | might_sleep(); |
| 1180 | spin_lock(&inode_lock); |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 1181 | ret = writeback_single_inode(inode, &wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1182 | spin_unlock(&inode_lock); |
| 1183 | if (sync) |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 1184 | inode_sync_wait(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1185 | return ret; |
| 1186 | } |
| 1187 | EXPORT_SYMBOL(write_inode_now); |
| 1188 | |
| 1189 | /** |
| 1190 | * sync_inode - write an inode and its pages to disk. |
| 1191 | * @inode: the inode to sync |
| 1192 | * @wbc: controls the writeback mode |
| 1193 | * |
| 1194 | * sync_inode() will write an inode and its pages to disk. It will also |
| 1195 | * correctly update the inode on its superblock's dirty inode lists and will |
| 1196 | * update inode->i_state. |
| 1197 | * |
| 1198 | * The caller must have a ref on the inode. |
| 1199 | */ |
| 1200 | int sync_inode(struct inode *inode, struct writeback_control *wbc) |
| 1201 | { |
| 1202 | int ret; |
| 1203 | |
| 1204 | spin_lock(&inode_lock); |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 1205 | ret = writeback_single_inode(inode, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1206 | spin_unlock(&inode_lock); |
| 1207 | return ret; |
| 1208 | } |
| 1209 | EXPORT_SYMBOL(sync_inode); |