Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * fs/fs-writeback.c |
| 3 | * |
| 4 | * Copyright (C) 2002, Linus Torvalds. |
| 5 | * |
| 6 | * Contains all the functions related to writing back and waiting |
| 7 | * upon dirty inodes against superblocks, and writing back dirty |
| 8 | * pages against inodes. ie: data writeback. Writeout of the |
| 9 | * inode itself is not handled here. |
| 10 | * |
Francois Cami | e1f8e87 | 2008-10-15 22:01:59 -0700 | [diff] [blame] | 11 | * 10Apr2002 Andrew Morton |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 12 | * Split out of fs/inode.c |
| 13 | * Additions for address_space-based writeback |
| 14 | */ |
| 15 | |
| 16 | #include <linux/kernel.h> |
Paul Gortmaker | 630d9c4 | 2011-11-16 23:57:37 -0500 | [diff] [blame] | 17 | #include <linux/export.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include <linux/spinlock.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 19 | #include <linux/slab.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | #include <linux/sched.h> |
| 21 | #include <linux/fs.h> |
| 22 | #include <linux/mm.h> |
Wu Fengguang | bc31b86 | 2012-01-07 20:41:55 -0600 | [diff] [blame] | 23 | #include <linux/pagemap.h> |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 24 | #include <linux/kthread.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include <linux/writeback.h> |
| 26 | #include <linux/blkdev.h> |
| 27 | #include <linux/backing-dev.h> |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 28 | #include <linux/tracepoint.h> |
Al Viro | 719ea2f | 2013-09-29 11:24:49 -0400 | [diff] [blame] | 29 | #include <linux/device.h> |
Tejun Heo | 21c6321 | 2015-05-28 14:50:49 -0400 | [diff] [blame] | 30 | #include <linux/memcontrol.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 | */ |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 36 | #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_SHIFT - 10)) |
Wu Fengguang | bc31b86 | 2012-01-07 20:41:55 -0600 | [diff] [blame] | 37 | |
Tejun Heo | cc395d7 | 2015-05-22 17:13:58 -0400 | [diff] [blame] | 38 | struct wb_completion { |
| 39 | atomic_t cnt; |
| 40 | }; |
| 41 | |
Wu Fengguang | bc31b86 | 2012-01-07 20:41:55 -0600 | [diff] [blame] | 42 | /* |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 43 | * Passed into wb_writeback(), essentially a subset of writeback_control |
| 44 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 45 | struct wb_writeback_work { |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 46 | long nr_pages; |
| 47 | struct super_block *sb; |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 48 | unsigned long *older_than_this; |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 49 | enum writeback_sync_modes sync_mode; |
Wu Fengguang | 6e6938b | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 50 | unsigned int tagged_writepages:1; |
H Hartley Sweeten | 52957fe | 2010-04-01 20:36:30 -0500 | [diff] [blame] | 51 | unsigned int for_kupdate:1; |
| 52 | unsigned int range_cyclic:1; |
| 53 | unsigned int for_background:1; |
Dave Chinner | 7747bd4 | 2013-07-02 22:38:35 +1000 | [diff] [blame] | 54 | unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ |
Tejun Heo | ac7b19a | 2015-05-22 17:13:57 -0400 | [diff] [blame] | 55 | unsigned int auto_free:1; /* free on completion */ |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 56 | enum wb_reason reason; /* why was writeback initiated? */ |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 57 | |
Jens Axboe | 8010c3b | 2009-09-15 20:04:57 +0200 | [diff] [blame] | 58 | struct list_head list; /* pending work list */ |
Tejun Heo | cc395d7 | 2015-05-22 17:13:58 -0400 | [diff] [blame] | 59 | struct wb_completion *done; /* set if the caller waits */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 60 | }; |
| 61 | |
Theodore Ts'o | a2f4870 | 2015-03-17 12:23:19 -0400 | [diff] [blame] | 62 | /* |
Tejun Heo | cc395d7 | 2015-05-22 17:13:58 -0400 | [diff] [blame] | 63 | * If one wants to wait for one or more wb_writeback_works, each work's |
| 64 | * ->done should be set to a wb_completion defined using the following |
| 65 | * macro. Once all work items are issued with wb_queue_work(), the caller |
| 66 | * can wait for the completion of all using wb_wait_for_completion(). Work |
| 67 | * items which are waited upon aren't freed automatically on completion. |
| 68 | */ |
| 69 | #define DEFINE_WB_COMPLETION_ONSTACK(cmpl) \ |
| 70 | struct wb_completion cmpl = { \ |
| 71 | .cnt = ATOMIC_INIT(1), \ |
| 72 | } |
| 73 | |
| 74 | |
| 75 | /* |
Theodore Ts'o | a2f4870 | 2015-03-17 12:23:19 -0400 | [diff] [blame] | 76 | * If an inode is constantly having its pages dirtied, but then the |
| 77 | * updates stop dirtytime_expire_interval seconds in the past, it's |
| 78 | * possible for the worst case time between when an inode has its |
| 79 | * timestamps updated and when they finally get written out to be two |
| 80 | * dirtytime_expire_intervals. We set the default to 12 hours (in |
| 81 | * seconds), which means most of the time inodes will have their |
| 82 | * timestamps written to disk after 12 hours, but in the worst case a |
| 83 | * few inodes might not their timestamps updated for 24 hours. |
| 84 | */ |
| 85 | unsigned int dirtytime_expire_interval = 12 * 60 * 60; |
| 86 | |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 87 | static inline struct inode *wb_inode(struct list_head *head) |
| 88 | { |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 89 | return list_entry(head, struct inode, i_io_list); |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 90 | } |
| 91 | |
Wu Fengguang | 15eb77a | 2012-01-17 11:18:56 -0600 | [diff] [blame] | 92 | /* |
| 93 | * Include the creation of the trace points after defining the |
| 94 | * wb_writeback_work structure and inline functions so that the definition |
| 95 | * remains local to this file. |
| 96 | */ |
| 97 | #define CREATE_TRACE_POINTS |
| 98 | #include <trace/events/writeback.h> |
| 99 | |
Steven Whitehouse | 774016b | 2014-02-06 15:47:47 +0000 | [diff] [blame] | 100 | EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); |
| 101 | |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 102 | static bool wb_io_lists_populated(struct bdi_writeback *wb) |
| 103 | { |
| 104 | if (wb_has_dirty_io(wb)) { |
| 105 | return false; |
| 106 | } else { |
| 107 | set_bit(WB_has_dirty_io, &wb->state); |
Tejun Heo | 95a46c6 | 2015-05-22 17:13:47 -0400 | [diff] [blame] | 108 | WARN_ON_ONCE(!wb->avg_write_bandwidth); |
Tejun Heo | 766a9d6 | 2015-05-22 17:13:46 -0400 | [diff] [blame] | 109 | atomic_long_add(wb->avg_write_bandwidth, |
| 110 | &wb->bdi->tot_write_bandwidth); |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 111 | return true; |
| 112 | } |
| 113 | } |
| 114 | |
| 115 | static void wb_io_lists_depopulated(struct bdi_writeback *wb) |
| 116 | { |
| 117 | if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) && |
Tejun Heo | 766a9d6 | 2015-05-22 17:13:46 -0400 | [diff] [blame] | 118 | list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) { |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 119 | clear_bit(WB_has_dirty_io, &wb->state); |
Tejun Heo | 95a46c6 | 2015-05-22 17:13:47 -0400 | [diff] [blame] | 120 | WARN_ON_ONCE(atomic_long_sub_return(wb->avg_write_bandwidth, |
| 121 | &wb->bdi->tot_write_bandwidth) < 0); |
Tejun Heo | 766a9d6 | 2015-05-22 17:13:46 -0400 | [diff] [blame] | 122 | } |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 123 | } |
| 124 | |
| 125 | /** |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 126 | * inode_io_list_move_locked - move an inode onto a bdi_writeback IO list |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 127 | * @inode: inode to be moved |
| 128 | * @wb: target bdi_writeback |
| 129 | * @head: one of @wb->b_{dirty|io|more_io} |
| 130 | * |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 131 | * Move @inode->i_io_list to @list of @wb and set %WB_has_dirty_io. |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 132 | * Returns %true if @inode is the first occupant of the !dirty_time IO |
| 133 | * lists; otherwise, %false. |
| 134 | */ |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 135 | static bool inode_io_list_move_locked(struct inode *inode, |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 136 | struct bdi_writeback *wb, |
| 137 | struct list_head *head) |
| 138 | { |
| 139 | assert_spin_locked(&wb->list_lock); |
| 140 | |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 141 | list_move(&inode->i_io_list, head); |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 142 | |
| 143 | /* dirty_time doesn't count as dirty_io until expiration */ |
| 144 | if (head != &wb->b_dirty_time) |
| 145 | return wb_io_lists_populated(wb); |
| 146 | |
| 147 | wb_io_lists_depopulated(wb); |
| 148 | return false; |
| 149 | } |
| 150 | |
| 151 | /** |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 152 | * inode_io_list_del_locked - remove an inode from its bdi_writeback IO list |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 153 | * @inode: inode to be removed |
| 154 | * @wb: bdi_writeback @inode is being removed from |
| 155 | * |
| 156 | * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and |
| 157 | * clear %WB_has_dirty_io if all are empty afterwards. |
| 158 | */ |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 159 | static void inode_io_list_del_locked(struct inode *inode, |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 160 | struct bdi_writeback *wb) |
| 161 | { |
| 162 | assert_spin_locked(&wb->list_lock); |
| 163 | |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 164 | list_del_init(&inode->i_io_list); |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 165 | wb_io_lists_depopulated(wb); |
| 166 | } |
| 167 | |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 168 | static void wb_wakeup(struct bdi_writeback *wb) |
Jan Kara | 5acda9d | 2014-04-03 14:46:23 -0700 | [diff] [blame] | 169 | { |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 170 | spin_lock_bh(&wb->work_lock); |
| 171 | if (test_bit(WB_registered, &wb->state)) |
| 172 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
| 173 | spin_unlock_bh(&wb->work_lock); |
Jan Kara | 5acda9d | 2014-04-03 14:46:23 -0700 | [diff] [blame] | 174 | } |
| 175 | |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 176 | static void wb_queue_work(struct bdi_writeback *wb, |
| 177 | struct wb_writeback_work *work) |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 178 | { |
Tejun Heo | 5634cc2 | 2015-08-18 14:54:56 -0700 | [diff] [blame] | 179 | trace_writeback_queue(wb, work); |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 180 | |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 181 | spin_lock_bh(&wb->work_lock); |
Tejun Heo | 8a1270c | 2015-08-18 14:54:53 -0700 | [diff] [blame] | 182 | if (!test_bit(WB_registered, &wb->state)) |
Jan Kara | 5acda9d | 2014-04-03 14:46:23 -0700 | [diff] [blame] | 183 | goto out_unlock; |
Tejun Heo | cc395d7 | 2015-05-22 17:13:58 -0400 | [diff] [blame] | 184 | if (work->done) |
| 185 | atomic_inc(&work->done->cnt); |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 186 | list_add_tail(&work->list, &wb->work_list); |
| 187 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
Jan Kara | 5acda9d | 2014-04-03 14:46:23 -0700 | [diff] [blame] | 188 | out_unlock: |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 189 | spin_unlock_bh(&wb->work_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | } |
| 191 | |
Tejun Heo | cc395d7 | 2015-05-22 17:13:58 -0400 | [diff] [blame] | 192 | /** |
| 193 | * wb_wait_for_completion - wait for completion of bdi_writeback_works |
| 194 | * @bdi: bdi work items were issued to |
| 195 | * @done: target wb_completion |
| 196 | * |
| 197 | * Wait for one or more work items issued to @bdi with their ->done field |
| 198 | * set to @done, which should have been defined with |
| 199 | * DEFINE_WB_COMPLETION_ONSTACK(). This function returns after all such |
| 200 | * work items are completed. Work items which are waited upon aren't freed |
| 201 | * automatically on completion. |
| 202 | */ |
| 203 | static void wb_wait_for_completion(struct backing_dev_info *bdi, |
| 204 | struct wb_completion *done) |
| 205 | { |
| 206 | atomic_dec(&done->cnt); /* put down the initial count */ |
| 207 | wait_event(bdi->wb_waitq, !atomic_read(&done->cnt)); |
| 208 | } |
| 209 | |
Tejun Heo | 703c270 | 2015-05-22 17:13:44 -0400 | [diff] [blame] | 210 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 211 | |
Tejun Heo | 2a81490 | 2015-05-28 14:50:51 -0400 | [diff] [blame] | 212 | /* parameters for foreign inode detection, see wb_detach_inode() */ |
| 213 | #define WB_FRN_TIME_SHIFT 13 /* 1s = 2^13, upto 8 secs w/ 16bit */ |
| 214 | #define WB_FRN_TIME_AVG_SHIFT 3 /* avg = avg * 7/8 + new * 1/8 */ |
| 215 | #define WB_FRN_TIME_CUT_DIV 2 /* ignore rounds < avg / 2 */ |
| 216 | #define WB_FRN_TIME_PERIOD (2 * (1 << WB_FRN_TIME_SHIFT)) /* 2s */ |
| 217 | |
| 218 | #define WB_FRN_HIST_SLOTS 16 /* inode->i_wb_frn_history is 16bit */ |
| 219 | #define WB_FRN_HIST_UNIT (WB_FRN_TIME_PERIOD / WB_FRN_HIST_SLOTS) |
| 220 | /* each slot's duration is 2s / 16 */ |
| 221 | #define WB_FRN_HIST_THR_SLOTS (WB_FRN_HIST_SLOTS / 2) |
| 222 | /* if foreign slots >= 8, switch */ |
| 223 | #define WB_FRN_HIST_MAX_SLOTS (WB_FRN_HIST_THR_SLOTS / 2 + 1) |
| 224 | /* one round can affect upto 5 slots */ |
| 225 | |
Tejun Heo | a1a0e23 | 2016-02-29 18:28:53 -0500 | [diff] [blame] | 226 | static atomic_t isw_nr_in_flight = ATOMIC_INIT(0); |
| 227 | static struct workqueue_struct *isw_wq; |
| 228 | |
Tejun Heo | 21c6321 | 2015-05-28 14:50:49 -0400 | [diff] [blame] | 229 | void __inode_attach_wb(struct inode *inode, struct page *page) |
| 230 | { |
| 231 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
| 232 | struct bdi_writeback *wb = NULL; |
| 233 | |
| 234 | if (inode_cgwb_enabled(inode)) { |
| 235 | struct cgroup_subsys_state *memcg_css; |
| 236 | |
| 237 | if (page) { |
| 238 | memcg_css = mem_cgroup_css_from_page(page); |
| 239 | wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); |
| 240 | } else { |
| 241 | /* must pin memcg_css, see wb_get_create() */ |
| 242 | memcg_css = task_get_css(current, memory_cgrp_id); |
| 243 | wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); |
| 244 | css_put(memcg_css); |
| 245 | } |
| 246 | } |
| 247 | |
| 248 | if (!wb) |
| 249 | wb = &bdi->wb; |
| 250 | |
| 251 | /* |
| 252 | * There may be multiple instances of this function racing to |
| 253 | * update the same inode. Use cmpxchg() to tell the winner. |
| 254 | */ |
| 255 | if (unlikely(cmpxchg(&inode->i_wb, NULL, wb))) |
| 256 | wb_put(wb); |
| 257 | } |
| 258 | |
Tejun Heo | 703c270 | 2015-05-22 17:13:44 -0400 | [diff] [blame] | 259 | /** |
Tejun Heo | 87e1d78 | 2015-05-28 14:50:52 -0400 | [diff] [blame] | 260 | * locked_inode_to_wb_and_lock_list - determine a locked inode's wb and lock it |
| 261 | * @inode: inode of interest with i_lock held |
| 262 | * |
| 263 | * Returns @inode's wb with its list_lock held. @inode->i_lock must be |
| 264 | * held on entry and is released on return. The returned wb is guaranteed |
| 265 | * to stay @inode's associated wb until its list_lock is released. |
| 266 | */ |
| 267 | static struct bdi_writeback * |
| 268 | locked_inode_to_wb_and_lock_list(struct inode *inode) |
| 269 | __releases(&inode->i_lock) |
| 270 | __acquires(&wb->list_lock) |
| 271 | { |
| 272 | while (true) { |
| 273 | struct bdi_writeback *wb = inode_to_wb(inode); |
| 274 | |
| 275 | /* |
| 276 | * inode_to_wb() association is protected by both |
| 277 | * @inode->i_lock and @wb->list_lock but list_lock nests |
| 278 | * outside i_lock. Drop i_lock and verify that the |
| 279 | * association hasn't changed after acquiring list_lock. |
| 280 | */ |
| 281 | wb_get(wb); |
| 282 | spin_unlock(&inode->i_lock); |
| 283 | spin_lock(&wb->list_lock); |
Tejun Heo | 87e1d78 | 2015-05-28 14:50:52 -0400 | [diff] [blame] | 284 | |
Tejun Heo | aaa2cac | 2015-05-28 14:50:55 -0400 | [diff] [blame] | 285 | /* i_wb may have changed inbetween, can't use inode_to_wb() */ |
Tejun Heo | 614a4e3 | 2016-03-18 13:50:03 -0400 | [diff] [blame] | 286 | if (likely(wb == inode->i_wb)) { |
| 287 | wb_put(wb); /* @inode already has ref */ |
| 288 | return wb; |
| 289 | } |
Tejun Heo | 87e1d78 | 2015-05-28 14:50:52 -0400 | [diff] [blame] | 290 | |
| 291 | spin_unlock(&wb->list_lock); |
Tejun Heo | 614a4e3 | 2016-03-18 13:50:03 -0400 | [diff] [blame] | 292 | wb_put(wb); |
Tejun Heo | 87e1d78 | 2015-05-28 14:50:52 -0400 | [diff] [blame] | 293 | cpu_relax(); |
| 294 | spin_lock(&inode->i_lock); |
| 295 | } |
| 296 | } |
| 297 | |
| 298 | /** |
| 299 | * inode_to_wb_and_lock_list - determine an inode's wb and lock it |
| 300 | * @inode: inode of interest |
| 301 | * |
| 302 | * Same as locked_inode_to_wb_and_lock_list() but @inode->i_lock isn't held |
| 303 | * on entry. |
| 304 | */ |
| 305 | static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode) |
| 306 | __acquires(&wb->list_lock) |
| 307 | { |
| 308 | spin_lock(&inode->i_lock); |
| 309 | return locked_inode_to_wb_and_lock_list(inode); |
| 310 | } |
| 311 | |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 312 | struct inode_switch_wbs_context { |
| 313 | struct inode *inode; |
| 314 | struct bdi_writeback *new_wb; |
| 315 | |
| 316 | struct rcu_head rcu_head; |
| 317 | struct work_struct work; |
| 318 | }; |
| 319 | |
| 320 | static void inode_switch_wbs_work_fn(struct work_struct *work) |
| 321 | { |
| 322 | struct inode_switch_wbs_context *isw = |
| 323 | container_of(work, struct inode_switch_wbs_context, work); |
| 324 | struct inode *inode = isw->inode; |
Tejun Heo | d10c809 | 2015-05-28 14:50:56 -0400 | [diff] [blame] | 325 | struct address_space *mapping = inode->i_mapping; |
| 326 | struct bdi_writeback *old_wb = inode->i_wb; |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 327 | struct bdi_writeback *new_wb = isw->new_wb; |
Tejun Heo | d10c809 | 2015-05-28 14:50:56 -0400 | [diff] [blame] | 328 | struct radix_tree_iter iter; |
| 329 | bool switched = false; |
| 330 | void **slot; |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 331 | |
| 332 | /* |
| 333 | * By the time control reaches here, RCU grace period has passed |
| 334 | * since I_WB_SWITCH assertion and all wb stat update transactions |
| 335 | * between unlocked_inode_to_wb_begin/end() are guaranteed to be |
| 336 | * synchronizing against mapping->tree_lock. |
Tejun Heo | d10c809 | 2015-05-28 14:50:56 -0400 | [diff] [blame] | 337 | * |
| 338 | * Grabbing old_wb->list_lock, inode->i_lock and mapping->tree_lock |
| 339 | * gives us exclusion against all wb related operations on @inode |
| 340 | * including IO list manipulations and stat updates. |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 341 | */ |
Tejun Heo | d10c809 | 2015-05-28 14:50:56 -0400 | [diff] [blame] | 342 | if (old_wb < new_wb) { |
| 343 | spin_lock(&old_wb->list_lock); |
| 344 | spin_lock_nested(&new_wb->list_lock, SINGLE_DEPTH_NESTING); |
| 345 | } else { |
| 346 | spin_lock(&new_wb->list_lock); |
| 347 | spin_lock_nested(&old_wb->list_lock, SINGLE_DEPTH_NESTING); |
| 348 | } |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 349 | spin_lock(&inode->i_lock); |
Tejun Heo | d10c809 | 2015-05-28 14:50:56 -0400 | [diff] [blame] | 350 | spin_lock_irq(&mapping->tree_lock); |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 351 | |
Tejun Heo | d10c809 | 2015-05-28 14:50:56 -0400 | [diff] [blame] | 352 | /* |
| 353 | * Once I_FREEING is visible under i_lock, the eviction path owns |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 354 | * the inode and we shouldn't modify ->i_io_list. |
Tejun Heo | d10c809 | 2015-05-28 14:50:56 -0400 | [diff] [blame] | 355 | */ |
| 356 | if (unlikely(inode->i_state & I_FREEING)) |
| 357 | goto skip_switch; |
| 358 | |
| 359 | /* |
| 360 | * Count and transfer stats. Note that PAGECACHE_TAG_DIRTY points |
| 361 | * to possibly dirty pages while PAGECACHE_TAG_WRITEBACK points to |
| 362 | * pages actually under underwriteback. |
| 363 | */ |
| 364 | radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, 0, |
| 365 | PAGECACHE_TAG_DIRTY) { |
| 366 | struct page *page = radix_tree_deref_slot_protected(slot, |
| 367 | &mapping->tree_lock); |
| 368 | if (likely(page) && PageDirty(page)) { |
| 369 | __dec_wb_stat(old_wb, WB_RECLAIMABLE); |
| 370 | __inc_wb_stat(new_wb, WB_RECLAIMABLE); |
| 371 | } |
| 372 | } |
| 373 | |
| 374 | radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, 0, |
| 375 | PAGECACHE_TAG_WRITEBACK) { |
| 376 | struct page *page = radix_tree_deref_slot_protected(slot, |
| 377 | &mapping->tree_lock); |
| 378 | if (likely(page)) { |
| 379 | WARN_ON_ONCE(!PageWriteback(page)); |
| 380 | __dec_wb_stat(old_wb, WB_WRITEBACK); |
| 381 | __inc_wb_stat(new_wb, WB_WRITEBACK); |
| 382 | } |
| 383 | } |
| 384 | |
| 385 | wb_get(new_wb); |
| 386 | |
| 387 | /* |
| 388 | * Transfer to @new_wb's IO list if necessary. The specific list |
| 389 | * @inode was on is ignored and the inode is put on ->b_dirty which |
| 390 | * is always correct including from ->b_dirty_time. The transfer |
| 391 | * preserves @inode->dirtied_when ordering. |
| 392 | */ |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 393 | if (!list_empty(&inode->i_io_list)) { |
Tejun Heo | d10c809 | 2015-05-28 14:50:56 -0400 | [diff] [blame] | 394 | struct inode *pos; |
| 395 | |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 396 | inode_io_list_del_locked(inode, old_wb); |
Tejun Heo | d10c809 | 2015-05-28 14:50:56 -0400 | [diff] [blame] | 397 | inode->i_wb = new_wb; |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 398 | list_for_each_entry(pos, &new_wb->b_dirty, i_io_list) |
Tejun Heo | d10c809 | 2015-05-28 14:50:56 -0400 | [diff] [blame] | 399 | if (time_after_eq(inode->dirtied_when, |
| 400 | pos->dirtied_when)) |
| 401 | break; |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 402 | inode_io_list_move_locked(inode, new_wb, pos->i_io_list.prev); |
Tejun Heo | d10c809 | 2015-05-28 14:50:56 -0400 | [diff] [blame] | 403 | } else { |
| 404 | inode->i_wb = new_wb; |
| 405 | } |
| 406 | |
| 407 | /* ->i_wb_frn updates may race wbc_detach_inode() but doesn't matter */ |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 408 | inode->i_wb_frn_winner = 0; |
| 409 | inode->i_wb_frn_avg_time = 0; |
| 410 | inode->i_wb_frn_history = 0; |
Tejun Heo | d10c809 | 2015-05-28 14:50:56 -0400 | [diff] [blame] | 411 | switched = true; |
| 412 | skip_switch: |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 413 | /* |
| 414 | * Paired with load_acquire in unlocked_inode_to_wb_begin() and |
| 415 | * ensures that the new wb is visible if they see !I_WB_SWITCH. |
| 416 | */ |
| 417 | smp_store_release(&inode->i_state, inode->i_state & ~I_WB_SWITCH); |
| 418 | |
Tejun Heo | d10c809 | 2015-05-28 14:50:56 -0400 | [diff] [blame] | 419 | spin_unlock_irq(&mapping->tree_lock); |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 420 | spin_unlock(&inode->i_lock); |
Tejun Heo | d10c809 | 2015-05-28 14:50:56 -0400 | [diff] [blame] | 421 | spin_unlock(&new_wb->list_lock); |
| 422 | spin_unlock(&old_wb->list_lock); |
| 423 | |
| 424 | if (switched) { |
| 425 | wb_wakeup(new_wb); |
| 426 | wb_put(old_wb); |
| 427 | } |
| 428 | wb_put(new_wb); |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 429 | |
| 430 | iput(inode); |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 431 | kfree(isw); |
Tejun Heo | a1a0e23 | 2016-02-29 18:28:53 -0500 | [diff] [blame] | 432 | |
| 433 | atomic_dec(&isw_nr_in_flight); |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 434 | } |
| 435 | |
| 436 | static void inode_switch_wbs_rcu_fn(struct rcu_head *rcu_head) |
| 437 | { |
| 438 | struct inode_switch_wbs_context *isw = container_of(rcu_head, |
| 439 | struct inode_switch_wbs_context, rcu_head); |
| 440 | |
| 441 | /* needs to grab bh-unsafe locks, bounce to work item */ |
| 442 | INIT_WORK(&isw->work, inode_switch_wbs_work_fn); |
Tejun Heo | a1a0e23 | 2016-02-29 18:28:53 -0500 | [diff] [blame] | 443 | queue_work(isw_wq, &isw->work); |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 444 | } |
| 445 | |
| 446 | /** |
| 447 | * inode_switch_wbs - change the wb association of an inode |
| 448 | * @inode: target inode |
| 449 | * @new_wb_id: ID of the new wb |
| 450 | * |
| 451 | * Switch @inode's wb association to the wb identified by @new_wb_id. The |
| 452 | * switching is performed asynchronously and may fail silently. |
| 453 | */ |
| 454 | static void inode_switch_wbs(struct inode *inode, int new_wb_id) |
| 455 | { |
| 456 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
| 457 | struct cgroup_subsys_state *memcg_css; |
| 458 | struct inode_switch_wbs_context *isw; |
| 459 | |
| 460 | /* noop if seems to be already in progress */ |
| 461 | if (inode->i_state & I_WB_SWITCH) |
| 462 | return; |
| 463 | |
| 464 | isw = kzalloc(sizeof(*isw), GFP_ATOMIC); |
| 465 | if (!isw) |
| 466 | return; |
| 467 | |
| 468 | /* find and pin the new wb */ |
| 469 | rcu_read_lock(); |
| 470 | memcg_css = css_from_id(new_wb_id, &memory_cgrp_subsys); |
| 471 | if (memcg_css) |
| 472 | isw->new_wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); |
| 473 | rcu_read_unlock(); |
| 474 | if (!isw->new_wb) |
| 475 | goto out_free; |
| 476 | |
| 477 | /* while holding I_WB_SWITCH, no one else can update the association */ |
| 478 | spin_lock(&inode->i_lock); |
Tejun Heo | a1a0e23 | 2016-02-29 18:28:53 -0500 | [diff] [blame] | 479 | if (!(inode->i_sb->s_flags & MS_ACTIVE) || |
| 480 | inode->i_state & (I_WB_SWITCH | I_FREEING) || |
| 481 | inode_to_wb(inode) == isw->new_wb) { |
| 482 | spin_unlock(&inode->i_lock); |
| 483 | goto out_free; |
| 484 | } |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 485 | inode->i_state |= I_WB_SWITCH; |
Tahsin Erdogan | 7452495 | 2016-06-16 05:15:33 -0700 | [diff] [blame] | 486 | __iget(inode); |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 487 | spin_unlock(&inode->i_lock); |
| 488 | |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 489 | isw->inode = inode; |
| 490 | |
Tejun Heo | a1a0e23 | 2016-02-29 18:28:53 -0500 | [diff] [blame] | 491 | atomic_inc(&isw_nr_in_flight); |
| 492 | |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 493 | /* |
| 494 | * In addition to synchronizing among switchers, I_WB_SWITCH tells |
| 495 | * the RCU protected stat update paths to grab the mapping's |
| 496 | * tree_lock so that stat transfer can synchronize against them. |
| 497 | * Let's continue after I_WB_SWITCH is guaranteed to be visible. |
| 498 | */ |
| 499 | call_rcu(&isw->rcu_head, inode_switch_wbs_rcu_fn); |
| 500 | return; |
| 501 | |
| 502 | out_free: |
| 503 | if (isw->new_wb) |
| 504 | wb_put(isw->new_wb); |
| 505 | kfree(isw); |
| 506 | } |
| 507 | |
Tejun Heo | 87e1d78 | 2015-05-28 14:50:52 -0400 | [diff] [blame] | 508 | /** |
Tejun Heo | b16b1de | 2015-06-02 08:39:48 -0600 | [diff] [blame] | 509 | * wbc_attach_and_unlock_inode - associate wbc with target inode and unlock it |
| 510 | * @wbc: writeback_control of interest |
| 511 | * @inode: target inode |
| 512 | * |
| 513 | * @inode is locked and about to be written back under the control of @wbc. |
| 514 | * Record @inode's writeback context into @wbc and unlock the i_lock. On |
| 515 | * writeback completion, wbc_detach_inode() should be called. This is used |
| 516 | * to track the cgroup writeback context. |
| 517 | */ |
| 518 | void wbc_attach_and_unlock_inode(struct writeback_control *wbc, |
| 519 | struct inode *inode) |
| 520 | { |
Tejun Heo | dd73e4b | 2015-06-16 18:48:30 -0400 | [diff] [blame] | 521 | if (!inode_cgwb_enabled(inode)) { |
| 522 | spin_unlock(&inode->i_lock); |
| 523 | return; |
| 524 | } |
| 525 | |
Tejun Heo | b16b1de | 2015-06-02 08:39:48 -0600 | [diff] [blame] | 526 | wbc->wb = inode_to_wb(inode); |
Tejun Heo | 2a81490 | 2015-05-28 14:50:51 -0400 | [diff] [blame] | 527 | wbc->inode = inode; |
| 528 | |
| 529 | wbc->wb_id = wbc->wb->memcg_css->id; |
| 530 | wbc->wb_lcand_id = inode->i_wb_frn_winner; |
| 531 | wbc->wb_tcand_id = 0; |
| 532 | wbc->wb_bytes = 0; |
| 533 | wbc->wb_lcand_bytes = 0; |
| 534 | wbc->wb_tcand_bytes = 0; |
| 535 | |
Tejun Heo | b16b1de | 2015-06-02 08:39:48 -0600 | [diff] [blame] | 536 | wb_get(wbc->wb); |
| 537 | spin_unlock(&inode->i_lock); |
Tejun Heo | e8a7abf | 2015-05-28 14:50:57 -0400 | [diff] [blame] | 538 | |
| 539 | /* |
| 540 | * A dying wb indicates that the memcg-blkcg mapping has changed |
| 541 | * and a new wb is already serving the memcg. Switch immediately. |
| 542 | */ |
| 543 | if (unlikely(wb_dying(wbc->wb))) |
| 544 | inode_switch_wbs(inode, wbc->wb_id); |
Tejun Heo | b16b1de | 2015-06-02 08:39:48 -0600 | [diff] [blame] | 545 | } |
| 546 | |
| 547 | /** |
Tejun Heo | 2a81490 | 2015-05-28 14:50:51 -0400 | [diff] [blame] | 548 | * wbc_detach_inode - disassociate wbc from inode and perform foreign detection |
| 549 | * @wbc: writeback_control of the just finished writeback |
Tejun Heo | b16b1de | 2015-06-02 08:39:48 -0600 | [diff] [blame] | 550 | * |
| 551 | * To be called after a writeback attempt of an inode finishes and undoes |
| 552 | * wbc_attach_and_unlock_inode(). Can be called under any context. |
Tejun Heo | 2a81490 | 2015-05-28 14:50:51 -0400 | [diff] [blame] | 553 | * |
| 554 | * As concurrent write sharing of an inode is expected to be very rare and |
| 555 | * memcg only tracks page ownership on first-use basis severely confining |
| 556 | * the usefulness of such sharing, cgroup writeback tracks ownership |
| 557 | * per-inode. While the support for concurrent write sharing of an inode |
| 558 | * is deemed unnecessary, an inode being written to by different cgroups at |
| 559 | * different points in time is a lot more common, and, more importantly, |
| 560 | * charging only by first-use can too readily lead to grossly incorrect |
| 561 | * behaviors (single foreign page can lead to gigabytes of writeback to be |
| 562 | * incorrectly attributed). |
| 563 | * |
| 564 | * To resolve this issue, cgroup writeback detects the majority dirtier of |
| 565 | * an inode and transfers the ownership to it. To avoid unnnecessary |
| 566 | * oscillation, the detection mechanism keeps track of history and gives |
| 567 | * out the switch verdict only if the foreign usage pattern is stable over |
| 568 | * a certain amount of time and/or writeback attempts. |
| 569 | * |
| 570 | * On each writeback attempt, @wbc tries to detect the majority writer |
| 571 | * using Boyer-Moore majority vote algorithm. In addition to the byte |
| 572 | * count from the majority voting, it also counts the bytes written for the |
| 573 | * current wb and the last round's winner wb (max of last round's current |
| 574 | * wb, the winner from two rounds ago, and the last round's majority |
| 575 | * candidate). Keeping track of the historical winner helps the algorithm |
| 576 | * to semi-reliably detect the most active writer even when it's not the |
| 577 | * absolute majority. |
| 578 | * |
| 579 | * Once the winner of the round is determined, whether the winner is |
| 580 | * foreign or not and how much IO time the round consumed is recorded in |
| 581 | * inode->i_wb_frn_history. If the amount of recorded foreign IO time is |
| 582 | * over a certain threshold, the switch verdict is given. |
Tejun Heo | b16b1de | 2015-06-02 08:39:48 -0600 | [diff] [blame] | 583 | */ |
| 584 | void wbc_detach_inode(struct writeback_control *wbc) |
| 585 | { |
Tejun Heo | 2a81490 | 2015-05-28 14:50:51 -0400 | [diff] [blame] | 586 | struct bdi_writeback *wb = wbc->wb; |
| 587 | struct inode *inode = wbc->inode; |
Tejun Heo | dd73e4b | 2015-06-16 18:48:30 -0400 | [diff] [blame] | 588 | unsigned long avg_time, max_bytes, max_time; |
| 589 | u16 history; |
Tejun Heo | 2a81490 | 2015-05-28 14:50:51 -0400 | [diff] [blame] | 590 | int max_id; |
| 591 | |
Tejun Heo | dd73e4b | 2015-06-16 18:48:30 -0400 | [diff] [blame] | 592 | if (!wb) |
| 593 | return; |
| 594 | |
| 595 | history = inode->i_wb_frn_history; |
| 596 | avg_time = inode->i_wb_frn_avg_time; |
| 597 | |
Tejun Heo | 2a81490 | 2015-05-28 14:50:51 -0400 | [diff] [blame] | 598 | /* pick the winner of this round */ |
| 599 | if (wbc->wb_bytes >= wbc->wb_lcand_bytes && |
| 600 | wbc->wb_bytes >= wbc->wb_tcand_bytes) { |
| 601 | max_id = wbc->wb_id; |
| 602 | max_bytes = wbc->wb_bytes; |
| 603 | } else if (wbc->wb_lcand_bytes >= wbc->wb_tcand_bytes) { |
| 604 | max_id = wbc->wb_lcand_id; |
| 605 | max_bytes = wbc->wb_lcand_bytes; |
| 606 | } else { |
| 607 | max_id = wbc->wb_tcand_id; |
| 608 | max_bytes = wbc->wb_tcand_bytes; |
| 609 | } |
| 610 | |
| 611 | /* |
| 612 | * Calculate the amount of IO time the winner consumed and fold it |
| 613 | * into the running average kept per inode. If the consumed IO |
| 614 | * time is lower than avag / WB_FRN_TIME_CUT_DIV, ignore it for |
| 615 | * deciding whether to switch or not. This is to prevent one-off |
| 616 | * small dirtiers from skewing the verdict. |
| 617 | */ |
| 618 | max_time = DIV_ROUND_UP((max_bytes >> PAGE_SHIFT) << WB_FRN_TIME_SHIFT, |
| 619 | wb->avg_write_bandwidth); |
| 620 | if (avg_time) |
| 621 | avg_time += (max_time >> WB_FRN_TIME_AVG_SHIFT) - |
| 622 | (avg_time >> WB_FRN_TIME_AVG_SHIFT); |
| 623 | else |
| 624 | avg_time = max_time; /* immediate catch up on first run */ |
| 625 | |
| 626 | if (max_time >= avg_time / WB_FRN_TIME_CUT_DIV) { |
| 627 | int slots; |
| 628 | |
| 629 | /* |
| 630 | * The switch verdict is reached if foreign wb's consume |
| 631 | * more than a certain proportion of IO time in a |
| 632 | * WB_FRN_TIME_PERIOD. This is loosely tracked by 16 slot |
| 633 | * history mask where each bit represents one sixteenth of |
| 634 | * the period. Determine the number of slots to shift into |
| 635 | * history from @max_time. |
| 636 | */ |
| 637 | slots = min(DIV_ROUND_UP(max_time, WB_FRN_HIST_UNIT), |
| 638 | (unsigned long)WB_FRN_HIST_MAX_SLOTS); |
| 639 | history <<= slots; |
| 640 | if (wbc->wb_id != max_id) |
| 641 | history |= (1U << slots) - 1; |
| 642 | |
| 643 | /* |
| 644 | * Switch if the current wb isn't the consistent winner. |
| 645 | * If there are multiple closely competing dirtiers, the |
| 646 | * inode may switch across them repeatedly over time, which |
| 647 | * is okay. The main goal is avoiding keeping an inode on |
| 648 | * the wrong wb for an extended period of time. |
| 649 | */ |
Tejun Heo | 682aa8e | 2015-05-28 14:50:53 -0400 | [diff] [blame] | 650 | if (hweight32(history) > WB_FRN_HIST_THR_SLOTS) |
| 651 | inode_switch_wbs(inode, max_id); |
Tejun Heo | 2a81490 | 2015-05-28 14:50:51 -0400 | [diff] [blame] | 652 | } |
| 653 | |
| 654 | /* |
| 655 | * Multiple instances of this function may race to update the |
| 656 | * following fields but we don't mind occassional inaccuracies. |
| 657 | */ |
| 658 | inode->i_wb_frn_winner = max_id; |
| 659 | inode->i_wb_frn_avg_time = min(avg_time, (unsigned long)U16_MAX); |
| 660 | inode->i_wb_frn_history = history; |
| 661 | |
Tejun Heo | b16b1de | 2015-06-02 08:39:48 -0600 | [diff] [blame] | 662 | wb_put(wbc->wb); |
| 663 | wbc->wb = NULL; |
| 664 | } |
| 665 | |
| 666 | /** |
Tejun Heo | 2a81490 | 2015-05-28 14:50:51 -0400 | [diff] [blame] | 667 | * wbc_account_io - account IO issued during writeback |
| 668 | * @wbc: writeback_control of the writeback in progress |
| 669 | * @page: page being written out |
| 670 | * @bytes: number of bytes being written out |
| 671 | * |
| 672 | * @bytes from @page are about to written out during the writeback |
| 673 | * controlled by @wbc. Keep the book for foreign inode detection. See |
| 674 | * wbc_detach_inode(). |
| 675 | */ |
| 676 | void wbc_account_io(struct writeback_control *wbc, struct page *page, |
| 677 | size_t bytes) |
| 678 | { |
| 679 | int id; |
| 680 | |
| 681 | /* |
| 682 | * pageout() path doesn't attach @wbc to the inode being written |
| 683 | * out. This is intentional as we don't want the function to block |
| 684 | * behind a slow cgroup. Ultimately, we want pageout() to kick off |
| 685 | * regular writeback instead of writing things out itself. |
| 686 | */ |
| 687 | if (!wbc->wb) |
| 688 | return; |
| 689 | |
Tejun Heo | 2a81490 | 2015-05-28 14:50:51 -0400 | [diff] [blame] | 690 | id = mem_cgroup_css_from_page(page)->id; |
Tejun Heo | 2a81490 | 2015-05-28 14:50:51 -0400 | [diff] [blame] | 691 | |
| 692 | if (id == wbc->wb_id) { |
| 693 | wbc->wb_bytes += bytes; |
| 694 | return; |
| 695 | } |
| 696 | |
| 697 | if (id == wbc->wb_lcand_id) |
| 698 | wbc->wb_lcand_bytes += bytes; |
| 699 | |
| 700 | /* Boyer-Moore majority vote algorithm */ |
| 701 | if (!wbc->wb_tcand_bytes) |
| 702 | wbc->wb_tcand_id = id; |
| 703 | if (id == wbc->wb_tcand_id) |
| 704 | wbc->wb_tcand_bytes += bytes; |
| 705 | else |
| 706 | wbc->wb_tcand_bytes -= min(bytes, wbc->wb_tcand_bytes); |
| 707 | } |
Tejun Heo | 5aa2a96 | 2015-07-23 14:27:09 -0400 | [diff] [blame] | 708 | EXPORT_SYMBOL_GPL(wbc_account_io); |
Tejun Heo | 2a81490 | 2015-05-28 14:50:51 -0400 | [diff] [blame] | 709 | |
| 710 | /** |
Tejun Heo | 703c270 | 2015-05-22 17:13:44 -0400 | [diff] [blame] | 711 | * inode_congested - test whether an inode is congested |
Tejun Heo | 60292bc | 2015-08-18 14:54:54 -0700 | [diff] [blame] | 712 | * @inode: inode to test for congestion (may be NULL) |
Tejun Heo | 703c270 | 2015-05-22 17:13:44 -0400 | [diff] [blame] | 713 | * @cong_bits: mask of WB_[a]sync_congested bits to test |
| 714 | * |
| 715 | * Tests whether @inode is congested. @cong_bits is the mask of congestion |
| 716 | * bits to test and the return value is the mask of set bits. |
| 717 | * |
| 718 | * If cgroup writeback is enabled for @inode, the congestion state is |
| 719 | * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg |
| 720 | * associated with @inode is congested; otherwise, the root wb's congestion |
| 721 | * state is used. |
Tejun Heo | 60292bc | 2015-08-18 14:54:54 -0700 | [diff] [blame] | 722 | * |
| 723 | * @inode is allowed to be NULL as this function is often called on |
| 724 | * mapping->host which is NULL for the swapper space. |
Tejun Heo | 703c270 | 2015-05-22 17:13:44 -0400 | [diff] [blame] | 725 | */ |
| 726 | int inode_congested(struct inode *inode, int cong_bits) |
| 727 | { |
Tejun Heo | 5cb8b82 | 2015-05-28 14:50:54 -0400 | [diff] [blame] | 728 | /* |
| 729 | * Once set, ->i_wb never becomes NULL while the inode is alive. |
| 730 | * Start transaction iff ->i_wb is visible. |
| 731 | */ |
Tejun Heo | aaa2cac | 2015-05-28 14:50:55 -0400 | [diff] [blame] | 732 | if (inode && inode_to_wb_is_valid(inode)) { |
Tejun Heo | 5cb8b82 | 2015-05-28 14:50:54 -0400 | [diff] [blame] | 733 | struct bdi_writeback *wb; |
| 734 | bool locked, congested; |
| 735 | |
| 736 | wb = unlocked_inode_to_wb_begin(inode, &locked); |
| 737 | congested = wb_congested(wb, cong_bits); |
| 738 | unlocked_inode_to_wb_end(inode, locked); |
| 739 | return congested; |
Tejun Heo | 703c270 | 2015-05-22 17:13:44 -0400 | [diff] [blame] | 740 | } |
| 741 | |
| 742 | return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); |
| 743 | } |
| 744 | EXPORT_SYMBOL_GPL(inode_congested); |
| 745 | |
Tejun Heo | f2b6512 | 2015-05-22 17:13:55 -0400 | [diff] [blame] | 746 | /** |
| 747 | * wb_split_bdi_pages - split nr_pages to write according to bandwidth |
| 748 | * @wb: target bdi_writeback to split @nr_pages to |
| 749 | * @nr_pages: number of pages to write for the whole bdi |
| 750 | * |
| 751 | * Split @wb's portion of @nr_pages according to @wb's write bandwidth in |
| 752 | * relation to the total write bandwidth of all wb's w/ dirty inodes on |
| 753 | * @wb->bdi. |
| 754 | */ |
| 755 | static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) |
| 756 | { |
| 757 | unsigned long this_bw = wb->avg_write_bandwidth; |
| 758 | unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth); |
| 759 | |
| 760 | if (nr_pages == LONG_MAX) |
| 761 | return LONG_MAX; |
| 762 | |
| 763 | /* |
| 764 | * This may be called on clean wb's and proportional distribution |
| 765 | * may not make sense, just use the original @nr_pages in those |
| 766 | * cases. In general, we wanna err on the side of writing more. |
| 767 | */ |
| 768 | if (!tot_bw || this_bw >= tot_bw) |
| 769 | return nr_pages; |
| 770 | else |
| 771 | return DIV_ROUND_UP_ULL((u64)nr_pages * this_bw, tot_bw); |
| 772 | } |
| 773 | |
Tejun Heo | db12536 | 2015-05-22 17:14:01 -0400 | [diff] [blame] | 774 | /** |
Tejun Heo | db12536 | 2015-05-22 17:14:01 -0400 | [diff] [blame] | 775 | * bdi_split_work_to_wbs - split a wb_writeback_work to all wb's of a bdi |
| 776 | * @bdi: target backing_dev_info |
| 777 | * @base_work: wb_writeback_work to issue |
| 778 | * @skip_if_busy: skip wb's which already have writeback in progress |
| 779 | * |
| 780 | * Split and issue @base_work to all wb's (bdi_writeback's) of @bdi which |
| 781 | * have dirty inodes. If @base_work->nr_page isn't %LONG_MAX, it's |
| 782 | * distributed to the busy wbs according to each wb's proportion in the |
| 783 | * total active write bandwidth of @bdi. |
| 784 | */ |
| 785 | static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, |
| 786 | struct wb_writeback_work *base_work, |
| 787 | bool skip_if_busy) |
| 788 | { |
Tejun Heo | b817525 | 2015-10-02 14:47:05 -0400 | [diff] [blame] | 789 | struct bdi_writeback *last_wb = NULL; |
Tejun Heo | b33e18f | 2015-10-27 14:19:39 +0900 | [diff] [blame] | 790 | struct bdi_writeback *wb = list_entry(&bdi->wb_list, |
| 791 | struct bdi_writeback, bdi_node); |
Tejun Heo | db12536 | 2015-05-22 17:14:01 -0400 | [diff] [blame] | 792 | |
| 793 | might_sleep(); |
Tejun Heo | db12536 | 2015-05-22 17:14:01 -0400 | [diff] [blame] | 794 | restart: |
| 795 | rcu_read_lock(); |
Tejun Heo | b817525 | 2015-10-02 14:47:05 -0400 | [diff] [blame] | 796 | list_for_each_entry_continue_rcu(wb, &bdi->wb_list, bdi_node) { |
Tejun Heo | 8a1270c | 2015-08-18 14:54:53 -0700 | [diff] [blame] | 797 | DEFINE_WB_COMPLETION_ONSTACK(fallback_work_done); |
| 798 | struct wb_writeback_work fallback_work; |
| 799 | struct wb_writeback_work *work; |
| 800 | long nr_pages; |
| 801 | |
Tejun Heo | b817525 | 2015-10-02 14:47:05 -0400 | [diff] [blame] | 802 | if (last_wb) { |
| 803 | wb_put(last_wb); |
| 804 | last_wb = NULL; |
| 805 | } |
| 806 | |
Tejun Heo | 006a097 | 2015-08-25 14:11:52 -0400 | [diff] [blame] | 807 | /* SYNC_ALL writes out I_DIRTY_TIME too */ |
| 808 | if (!wb_has_dirty_io(wb) && |
| 809 | (base_work->sync_mode == WB_SYNC_NONE || |
| 810 | list_empty(&wb->b_dirty_time))) |
| 811 | continue; |
| 812 | if (skip_if_busy && writeback_in_progress(wb)) |
Tejun Heo | db12536 | 2015-05-22 17:14:01 -0400 | [diff] [blame] | 813 | continue; |
| 814 | |
Tejun Heo | 8a1270c | 2015-08-18 14:54:53 -0700 | [diff] [blame] | 815 | nr_pages = wb_split_bdi_pages(wb, base_work->nr_pages); |
| 816 | |
| 817 | work = kmalloc(sizeof(*work), GFP_ATOMIC); |
| 818 | if (work) { |
| 819 | *work = *base_work; |
| 820 | work->nr_pages = nr_pages; |
| 821 | work->auto_free = 1; |
| 822 | wb_queue_work(wb, work); |
| 823 | continue; |
Tejun Heo | db12536 | 2015-05-22 17:14:01 -0400 | [diff] [blame] | 824 | } |
Tejun Heo | 8a1270c | 2015-08-18 14:54:53 -0700 | [diff] [blame] | 825 | |
| 826 | /* alloc failed, execute synchronously using on-stack fallback */ |
| 827 | work = &fallback_work; |
| 828 | *work = *base_work; |
| 829 | work->nr_pages = nr_pages; |
| 830 | work->auto_free = 0; |
| 831 | work->done = &fallback_work_done; |
| 832 | |
| 833 | wb_queue_work(wb, work); |
| 834 | |
Tejun Heo | b817525 | 2015-10-02 14:47:05 -0400 | [diff] [blame] | 835 | /* |
| 836 | * Pin @wb so that it stays on @bdi->wb_list. This allows |
| 837 | * continuing iteration from @wb after dropping and |
| 838 | * regrabbing rcu read lock. |
| 839 | */ |
| 840 | wb_get(wb); |
| 841 | last_wb = wb; |
| 842 | |
Tejun Heo | 8a1270c | 2015-08-18 14:54:53 -0700 | [diff] [blame] | 843 | rcu_read_unlock(); |
| 844 | wb_wait_for_completion(bdi, &fallback_work_done); |
| 845 | goto restart; |
Tejun Heo | db12536 | 2015-05-22 17:14:01 -0400 | [diff] [blame] | 846 | } |
| 847 | rcu_read_unlock(); |
Tejun Heo | b817525 | 2015-10-02 14:47:05 -0400 | [diff] [blame] | 848 | |
| 849 | if (last_wb) |
| 850 | wb_put(last_wb); |
Tejun Heo | db12536 | 2015-05-22 17:14:01 -0400 | [diff] [blame] | 851 | } |
| 852 | |
Tejun Heo | a1a0e23 | 2016-02-29 18:28:53 -0500 | [diff] [blame] | 853 | /** |
| 854 | * cgroup_writeback_umount - flush inode wb switches for umount |
| 855 | * |
| 856 | * This function is called when a super_block is about to be destroyed and |
| 857 | * flushes in-flight inode wb switches. An inode wb switch goes through |
| 858 | * RCU and then workqueue, so the two need to be flushed in order to ensure |
| 859 | * that all previously scheduled switches are finished. As wb switches are |
| 860 | * rare occurrences and synchronize_rcu() can take a while, perform |
| 861 | * flushing iff wb switches are in flight. |
| 862 | */ |
| 863 | void cgroup_writeback_umount(void) |
| 864 | { |
| 865 | if (atomic_read(&isw_nr_in_flight)) { |
| 866 | synchronize_rcu(); |
| 867 | flush_workqueue(isw_wq); |
| 868 | } |
| 869 | } |
| 870 | |
| 871 | static int __init cgroup_writeback_init(void) |
| 872 | { |
| 873 | isw_wq = alloc_workqueue("inode_switch_wbs", 0, 0); |
| 874 | if (!isw_wq) |
| 875 | return -ENOMEM; |
| 876 | return 0; |
| 877 | } |
| 878 | fs_initcall(cgroup_writeback_init); |
| 879 | |
Tejun Heo | f2b6512 | 2015-05-22 17:13:55 -0400 | [diff] [blame] | 880 | #else /* CONFIG_CGROUP_WRITEBACK */ |
| 881 | |
Tejun Heo | 87e1d78 | 2015-05-28 14:50:52 -0400 | [diff] [blame] | 882 | static struct bdi_writeback * |
| 883 | locked_inode_to_wb_and_lock_list(struct inode *inode) |
| 884 | __releases(&inode->i_lock) |
| 885 | __acquires(&wb->list_lock) |
| 886 | { |
| 887 | struct bdi_writeback *wb = inode_to_wb(inode); |
| 888 | |
| 889 | spin_unlock(&inode->i_lock); |
| 890 | spin_lock(&wb->list_lock); |
| 891 | return wb; |
| 892 | } |
| 893 | |
| 894 | static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode) |
| 895 | __acquires(&wb->list_lock) |
| 896 | { |
| 897 | struct bdi_writeback *wb = inode_to_wb(inode); |
| 898 | |
| 899 | spin_lock(&wb->list_lock); |
| 900 | return wb; |
| 901 | } |
| 902 | |
Tejun Heo | f2b6512 | 2015-05-22 17:13:55 -0400 | [diff] [blame] | 903 | static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) |
| 904 | { |
| 905 | return nr_pages; |
| 906 | } |
| 907 | |
Tejun Heo | db12536 | 2015-05-22 17:14:01 -0400 | [diff] [blame] | 908 | static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, |
| 909 | struct wb_writeback_work *base_work, |
| 910 | bool skip_if_busy) |
| 911 | { |
| 912 | might_sleep(); |
| 913 | |
Tejun Heo | 006a097 | 2015-08-25 14:11:52 -0400 | [diff] [blame] | 914 | if (!skip_if_busy || !writeback_in_progress(&bdi->wb)) { |
Tejun Heo | db12536 | 2015-05-22 17:14:01 -0400 | [diff] [blame] | 915 | base_work->auto_free = 0; |
Tejun Heo | db12536 | 2015-05-22 17:14:01 -0400 | [diff] [blame] | 916 | wb_queue_work(&bdi->wb, base_work); |
| 917 | } |
| 918 | } |
| 919 | |
Tejun Heo | 703c270 | 2015-05-22 17:13:44 -0400 | [diff] [blame] | 920 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
| 921 | |
Tejun Heo | c00ddad | 2015-05-22 17:13:51 -0400 | [diff] [blame] | 922 | void wb_start_writeback(struct bdi_writeback *wb, long nr_pages, |
| 923 | bool range_cyclic, enum wb_reason reason) |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 924 | { |
Tejun Heo | c00ddad | 2015-05-22 17:13:51 -0400 | [diff] [blame] | 925 | struct wb_writeback_work *work; |
| 926 | |
| 927 | if (!wb_has_dirty_io(wb)) |
| 928 | return; |
| 929 | |
| 930 | /* |
| 931 | * This is WB_SYNC_NONE writeback, so if allocation fails just |
| 932 | * wakeup the thread for old dirty data writeback |
| 933 | */ |
Tetsuo Handa | 78ebc2f | 2016-05-20 16:58:10 -0700 | [diff] [blame] | 934 | work = kzalloc(sizeof(*work), |
| 935 | GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN); |
Tejun Heo | c00ddad | 2015-05-22 17:13:51 -0400 | [diff] [blame] | 936 | if (!work) { |
Tejun Heo | 5634cc2 | 2015-08-18 14:54:56 -0700 | [diff] [blame] | 937 | trace_writeback_nowork(wb); |
Tejun Heo | c00ddad | 2015-05-22 17:13:51 -0400 | [diff] [blame] | 938 | wb_wakeup(wb); |
| 939 | return; |
| 940 | } |
| 941 | |
| 942 | work->sync_mode = WB_SYNC_NONE; |
| 943 | work->nr_pages = nr_pages; |
| 944 | work->range_cyclic = range_cyclic; |
| 945 | work->reason = reason; |
Tejun Heo | ac7b19a | 2015-05-22 17:13:57 -0400 | [diff] [blame] | 946 | work->auto_free = 1; |
Tejun Heo | c00ddad | 2015-05-22 17:13:51 -0400 | [diff] [blame] | 947 | |
| 948 | wb_queue_work(wb, work); |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 949 | } |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 950 | |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 951 | /** |
Tejun Heo | 9ecf4866 | 2015-05-22 17:13:54 -0400 | [diff] [blame] | 952 | * wb_start_background_writeback - start background writeback |
| 953 | * @wb: bdi_writback to write from |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 954 | * |
| 955 | * Description: |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 956 | * This makes sure WB_SYNC_NONE background writeback happens. When |
Tejun Heo | 9ecf4866 | 2015-05-22 17:13:54 -0400 | [diff] [blame] | 957 | * this function returns, it is only guaranteed that for given wb |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 958 | * some IO is happening if we are over background dirty threshold. |
| 959 | * Caller need not hold sb s_umount semaphore. |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 960 | */ |
Tejun Heo | 9ecf4866 | 2015-05-22 17:13:54 -0400 | [diff] [blame] | 961 | void wb_start_background_writeback(struct bdi_writeback *wb) |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 962 | { |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 963 | /* |
| 964 | * We just wake up the flusher thread. It will perform background |
| 965 | * writeback as soon as there is no other work to do. |
| 966 | */ |
Tejun Heo | 5634cc2 | 2015-08-18 14:54:56 -0700 | [diff] [blame] | 967 | trace_writeback_wake_background(wb); |
Tejun Heo | 9ecf4866 | 2015-05-22 17:13:54 -0400 | [diff] [blame] | 968 | wb_wakeup(wb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 969 | } |
| 970 | |
| 971 | /* |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 972 | * Remove the inode from the writeback list it is on. |
| 973 | */ |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 974 | void inode_io_list_del(struct inode *inode) |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 975 | { |
Tejun Heo | 87e1d78 | 2015-05-28 14:50:52 -0400 | [diff] [blame] | 976 | struct bdi_writeback *wb; |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 977 | |
Tejun Heo | 87e1d78 | 2015-05-28 14:50:52 -0400 | [diff] [blame] | 978 | wb = inode_to_wb_and_lock_list(inode); |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 979 | inode_io_list_del_locked(inode, wb); |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 980 | spin_unlock(&wb->list_lock); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 981 | } |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 982 | |
| 983 | /* |
Dave Chinner | 6c60d2b | 2016-07-26 15:21:50 -0700 | [diff] [blame] | 984 | * mark an inode as under writeback on the sb |
| 985 | */ |
| 986 | void sb_mark_inode_writeback(struct inode *inode) |
| 987 | { |
| 988 | struct super_block *sb = inode->i_sb; |
| 989 | unsigned long flags; |
| 990 | |
| 991 | if (list_empty(&inode->i_wb_list)) { |
| 992 | spin_lock_irqsave(&sb->s_inode_wblist_lock, flags); |
Brian Foster | 9a46b04f | 2016-07-26 15:21:53 -0700 | [diff] [blame] | 993 | if (list_empty(&inode->i_wb_list)) { |
Dave Chinner | 6c60d2b | 2016-07-26 15:21:50 -0700 | [diff] [blame] | 994 | list_add_tail(&inode->i_wb_list, &sb->s_inodes_wb); |
Brian Foster | 9a46b04f | 2016-07-26 15:21:53 -0700 | [diff] [blame] | 995 | trace_sb_mark_inode_writeback(inode); |
| 996 | } |
Dave Chinner | 6c60d2b | 2016-07-26 15:21:50 -0700 | [diff] [blame] | 997 | spin_unlock_irqrestore(&sb->s_inode_wblist_lock, flags); |
| 998 | } |
| 999 | } |
| 1000 | |
| 1001 | /* |
| 1002 | * clear an inode as under writeback on the sb |
| 1003 | */ |
| 1004 | void sb_clear_inode_writeback(struct inode *inode) |
| 1005 | { |
| 1006 | struct super_block *sb = inode->i_sb; |
| 1007 | unsigned long flags; |
| 1008 | |
| 1009 | if (!list_empty(&inode->i_wb_list)) { |
| 1010 | spin_lock_irqsave(&sb->s_inode_wblist_lock, flags); |
Brian Foster | 9a46b04f | 2016-07-26 15:21:53 -0700 | [diff] [blame] | 1011 | if (!list_empty(&inode->i_wb_list)) { |
| 1012 | list_del_init(&inode->i_wb_list); |
| 1013 | trace_sb_clear_inode_writeback(inode); |
| 1014 | } |
Dave Chinner | 6c60d2b | 2016-07-26 15:21:50 -0700 | [diff] [blame] | 1015 | spin_unlock_irqrestore(&sb->s_inode_wblist_lock, flags); |
| 1016 | } |
| 1017 | } |
| 1018 | |
| 1019 | /* |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 1020 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the |
| 1021 | * furthest end of its superblock's dirty-inode list. |
| 1022 | * |
| 1023 | * 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] | 1024 | * 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] | 1025 | * the case then the inode must have been redirtied while it was being written |
| 1026 | * out and we don't reset its dirtied_when. |
| 1027 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1028 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 1029 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1030 | if (!list_empty(&wb->b_dirty)) { |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1031 | struct inode *tail; |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 1032 | |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 1033 | tail = wb_inode(wb->b_dirty.next); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1034 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 1035 | inode->dirtied_when = jiffies; |
| 1036 | } |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 1037 | inode_io_list_move_locked(inode, wb, &wb->b_dirty); |
Andrew Morton | 6610a0b | 2007-10-16 23:30:32 -0700 | [diff] [blame] | 1038 | } |
| 1039 | |
| 1040 | /* |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1041 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 1042 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1043 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 1044 | { |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 1045 | inode_io_list_move_locked(inode, wb, &wb->b_more_io); |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 1046 | } |
| 1047 | |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 1048 | static void inode_sync_complete(struct inode *inode) |
| 1049 | { |
Jan Kara | 365b94ae | 2012-05-03 14:47:55 +0200 | [diff] [blame] | 1050 | inode->i_state &= ~I_SYNC; |
Jan Kara | 4eff96d | 2012-11-26 16:29:51 -0800 | [diff] [blame] | 1051 | /* If inode is clean an unused, put it into LRU now... */ |
| 1052 | inode_add_lru(inode); |
Jan Kara | 365b94ae | 2012-05-03 14:47:55 +0200 | [diff] [blame] | 1053 | /* Waiters must see I_SYNC cleared before being woken up */ |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 1054 | smp_mb(); |
| 1055 | wake_up_bit(&inode->i_state, __I_SYNC); |
| 1056 | } |
| 1057 | |
Jeff Layton | d2caa3c5 | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 1058 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
| 1059 | { |
| 1060 | bool ret = time_after(inode->dirtied_when, t); |
| 1061 | #ifndef CONFIG_64BIT |
| 1062 | /* |
| 1063 | * For inodes being constantly redirtied, dirtied_when can get stuck. |
| 1064 | * It _appears_ to be in the future, but is actually in distant past. |
| 1065 | * This test is necessary to prevent such wrapped-around relative times |
Jens Axboe | 5b0830c | 2009-09-23 19:37:09 +0200 | [diff] [blame] | 1066 | * from permanently stopping the whole bdi writeback. |
Jeff Layton | d2caa3c5 | 2009-04-02 16:56:37 -0700 | [diff] [blame] | 1067 | */ |
| 1068 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); |
| 1069 | #endif |
| 1070 | return ret; |
| 1071 | } |
| 1072 | |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1073 | #define EXPIRE_DIRTY_ATIME 0x0001 |
| 1074 | |
Andrew Morton | c986d1e | 2007-10-16 23:30:34 -0700 | [diff] [blame] | 1075 | /* |
Wang Sheng-Hui | 0e2f2b2 | 2012-09-11 08:28:18 +0800 | [diff] [blame] | 1076 | * Move expired (dirtied before work->older_than_this) dirty inodes from |
Jan Kara | 697e6fe | 2012-03-09 07:26:22 -0800 | [diff] [blame] | 1077 | * @delaying_queue to @dispatch_queue. |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 1078 | */ |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 1079 | static int move_expired_inodes(struct list_head *delaying_queue, |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 1080 | struct list_head *dispatch_queue, |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1081 | int flags, |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 1082 | struct wb_writeback_work *work) |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 1083 | { |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1084 | unsigned long *older_than_this = NULL; |
| 1085 | unsigned long expire_time; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 1086 | LIST_HEAD(tmp); |
| 1087 | struct list_head *pos, *node; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 1088 | struct super_block *sb = NULL; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 1089 | struct inode *inode; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 1090 | int do_sb_sort = 0; |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 1091 | int moved = 0; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 1092 | |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1093 | if ((flags & EXPIRE_DIRTY_ATIME) == 0) |
| 1094 | older_than_this = work->older_than_this; |
Theodore Ts'o | a2f4870 | 2015-03-17 12:23:19 -0400 | [diff] [blame] | 1095 | else if (!work->for_sync) { |
| 1096 | expire_time = jiffies - (dirtytime_expire_interval * HZ); |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1097 | older_than_this = &expire_time; |
| 1098 | } |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 1099 | while (!list_empty(delaying_queue)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 1100 | inode = wb_inode(delaying_queue->prev); |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1101 | if (older_than_this && |
| 1102 | inode_dirtied_after(inode, *older_than_this)) |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 1103 | break; |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 1104 | list_move(&inode->i_io_list, &tmp); |
Jan Kara | a885599 | 2013-07-09 22:36:45 +0800 | [diff] [blame] | 1105 | moved++; |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1106 | if (flags & EXPIRE_DIRTY_ATIME) |
| 1107 | set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); |
Jan Kara | a885599 | 2013-07-09 22:36:45 +0800 | [diff] [blame] | 1108 | if (sb_is_blkdev_sb(inode->i_sb)) |
| 1109 | continue; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 1110 | if (sb && sb != inode->i_sb) |
| 1111 | do_sb_sort = 1; |
| 1112 | sb = inode->i_sb; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 1113 | } |
| 1114 | |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 1115 | /* just one sb in list, splice to dispatch_queue and we're done */ |
| 1116 | if (!do_sb_sort) { |
| 1117 | list_splice(&tmp, dispatch_queue); |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 1118 | goto out; |
Jens Axboe | cf13730 | 2009-09-24 15:12:57 +0200 | [diff] [blame] | 1119 | } |
| 1120 | |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 1121 | /* Move inodes from one superblock together */ |
| 1122 | while (!list_empty(&tmp)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 1123 | sb = wb_inode(tmp.prev)->i_sb; |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 1124 | list_for_each_prev_safe(pos, node, &tmp) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 1125 | inode = wb_inode(pos); |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 1126 | if (inode->i_sb == sb) |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 1127 | list_move(&inode->i_io_list, dispatch_queue); |
Shaohua Li | 5c03449 | 2009-09-24 14:42:33 +0200 | [diff] [blame] | 1128 | } |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 1129 | } |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 1130 | out: |
| 1131 | return moved; |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 1132 | } |
| 1133 | |
| 1134 | /* |
| 1135 | * Queue all expired dirty inodes for io, eldest first. |
Wu Fengguang | 4ea879b | 2010-08-11 14:17:42 -0700 | [diff] [blame] | 1136 | * Before |
| 1137 | * newly dirtied b_dirty b_io b_more_io |
| 1138 | * =============> gf edc BA |
| 1139 | * After |
| 1140 | * newly dirtied b_dirty b_io b_more_io |
| 1141 | * =============> g fBAedc |
| 1142 | * | |
| 1143 | * +--> dequeue for IO |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 1144 | */ |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 1145 | 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] | 1146 | { |
Wu Fengguang | e84d0a4 | 2011-04-23 12:27:27 -0600 | [diff] [blame] | 1147 | int moved; |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1148 | |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1149 | assert_spin_locked(&wb->list_lock); |
Wu Fengguang | 4ea879b | 2010-08-11 14:17:42 -0700 | [diff] [blame] | 1150 | list_splice_init(&wb->b_more_io, &wb->b_io); |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1151 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); |
| 1152 | moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, |
| 1153 | EXPIRE_DIRTY_ATIME, work); |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 1154 | if (moved) |
| 1155 | wb_io_lists_populated(wb); |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 1156 | trace_writeback_queue_io(wb, work, moved); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1157 | } |
| 1158 | |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 1159 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1160 | { |
Tejun Heo | 9fb0a7d | 2013-01-11 13:06:37 -0800 | [diff] [blame] | 1161 | int ret; |
| 1162 | |
| 1163 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { |
| 1164 | trace_writeback_write_inode_start(inode, wbc); |
| 1165 | ret = inode->i_sb->s_op->write_inode(inode, wbc); |
| 1166 | trace_writeback_write_inode(inode, wbc); |
| 1167 | return ret; |
| 1168 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1169 | return 0; |
Fengguang Wu | 2c13657 | 2007-10-16 23:30:39 -0700 | [diff] [blame] | 1170 | } |
| 1171 | |
| 1172 | /* |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 1173 | * Wait for writeback on an inode to complete. Called with i_lock held. |
| 1174 | * Caller must make sure inode cannot go away when we drop i_lock. |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 1175 | */ |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 1176 | static void __inode_wait_for_writeback(struct inode *inode) |
| 1177 | __releases(inode->i_lock) |
| 1178 | __acquires(inode->i_lock) |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 1179 | { |
| 1180 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); |
| 1181 | wait_queue_head_t *wqh; |
| 1182 | |
| 1183 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1184 | while (inode->i_state & I_SYNC) { |
| 1185 | spin_unlock(&inode->i_lock); |
NeilBrown | 7431620 | 2014-07-07 15:16:04 +1000 | [diff] [blame] | 1186 | __wait_on_bit(wqh, &wq, bit_wait, |
| 1187 | TASK_UNINTERRUPTIBLE); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1188 | spin_lock(&inode->i_lock); |
Richard Kennedy | 58a9d3d8 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 1189 | } |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 1190 | } |
| 1191 | |
| 1192 | /* |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 1193 | * Wait for writeback on an inode to complete. Caller must have inode pinned. |
| 1194 | */ |
| 1195 | void inode_wait_for_writeback(struct inode *inode) |
| 1196 | { |
| 1197 | spin_lock(&inode->i_lock); |
| 1198 | __inode_wait_for_writeback(inode); |
| 1199 | spin_unlock(&inode->i_lock); |
| 1200 | } |
| 1201 | |
| 1202 | /* |
| 1203 | * Sleep until I_SYNC is cleared. This function must be called with i_lock |
| 1204 | * held and drops it. It is aimed for callers not holding any inode reference |
| 1205 | * so once i_lock is dropped, inode can go away. |
| 1206 | */ |
| 1207 | static void inode_sleep_on_writeback(struct inode *inode) |
| 1208 | __releases(inode->i_lock) |
| 1209 | { |
| 1210 | DEFINE_WAIT(wait); |
| 1211 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); |
| 1212 | int sleep; |
| 1213 | |
| 1214 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); |
| 1215 | sleep = inode->i_state & I_SYNC; |
| 1216 | spin_unlock(&inode->i_lock); |
| 1217 | if (sleep) |
| 1218 | schedule(); |
| 1219 | finish_wait(wqh, &wait); |
| 1220 | } |
| 1221 | |
| 1222 | /* |
Jan Kara | ccb26b5 | 2012-05-03 14:47:58 +0200 | [diff] [blame] | 1223 | * Find proper writeback list for the inode depending on its current state and |
| 1224 | * possibly also change of its state while we were doing writeback. Here we |
| 1225 | * handle things such as livelock prevention or fairness of writeback among |
| 1226 | * inodes. This function can be called only by flusher thread - noone else |
| 1227 | * processes all inodes in writeback lists and requeueing inodes behind flusher |
| 1228 | * thread's back can have unexpected consequences. |
| 1229 | */ |
| 1230 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, |
| 1231 | struct writeback_control *wbc) |
| 1232 | { |
| 1233 | if (inode->i_state & I_FREEING) |
| 1234 | return; |
| 1235 | |
| 1236 | /* |
| 1237 | * Sync livelock prevention. Each inode is tagged and synced in one |
| 1238 | * shot. If still dirty, it will be redirty_tail()'ed below. Update |
| 1239 | * the dirty time to prevent enqueue and sync it again. |
| 1240 | */ |
| 1241 | if ((inode->i_state & I_DIRTY) && |
| 1242 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) |
| 1243 | inode->dirtied_when = jiffies; |
| 1244 | |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1245 | if (wbc->pages_skipped) { |
| 1246 | /* |
| 1247 | * writeback is not making progress due to locked |
| 1248 | * buffers. Skip this inode for now. |
| 1249 | */ |
| 1250 | redirty_tail(inode, wb); |
| 1251 | return; |
| 1252 | } |
| 1253 | |
Jan Kara | ccb26b5 | 2012-05-03 14:47:58 +0200 | [diff] [blame] | 1254 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
| 1255 | /* |
| 1256 | * We didn't write back all the pages. nfs_writepages() |
| 1257 | * sometimes bales out without doing anything. |
| 1258 | */ |
| 1259 | if (wbc->nr_to_write <= 0) { |
| 1260 | /* Slice used up. Queue for next turn. */ |
| 1261 | requeue_io(inode, wb); |
| 1262 | } else { |
| 1263 | /* |
| 1264 | * Writeback blocked by something other than |
| 1265 | * congestion. Delay the inode for some time to |
| 1266 | * avoid spinning on the CPU (100% iowait) |
| 1267 | * retrying writeback of the dirty page/inode |
| 1268 | * that cannot be performed immediately. |
| 1269 | */ |
| 1270 | redirty_tail(inode, wb); |
| 1271 | } |
| 1272 | } else if (inode->i_state & I_DIRTY) { |
| 1273 | /* |
| 1274 | * Filesystems can dirty the inode during writeback operations, |
| 1275 | * such as delayed allocation during submission or metadata |
| 1276 | * updates after data IO completion. |
| 1277 | */ |
| 1278 | redirty_tail(inode, wb); |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1279 | } else if (inode->i_state & I_DIRTY_TIME) { |
Theodore Ts'o | a2f4870 | 2015-03-17 12:23:19 -0400 | [diff] [blame] | 1280 | inode->dirtied_when = jiffies; |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 1281 | inode_io_list_move_locked(inode, wb, &wb->b_dirty_time); |
Jan Kara | ccb26b5 | 2012-05-03 14:47:58 +0200 | [diff] [blame] | 1282 | } else { |
| 1283 | /* The inode is clean. Remove from writeback lists. */ |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 1284 | inode_io_list_del_locked(inode, wb); |
Jan Kara | ccb26b5 | 2012-05-03 14:47:58 +0200 | [diff] [blame] | 1285 | } |
| 1286 | } |
| 1287 | |
| 1288 | /* |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1289 | * Write out an inode and its dirty pages. Do not update the writeback list |
| 1290 | * linkage. That is left to the caller. The caller is also responsible for |
| 1291 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1292 | */ |
| 1293 | static int |
Yan Hong | cd8ed2a | 2012-10-08 16:33:45 -0700 | [diff] [blame] | 1294 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1295 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1296 | struct address_space *mapping = inode->i_mapping; |
Wu Fengguang | 251d6a4 | 2010-12-01 17:33:37 -0600 | [diff] [blame] | 1297 | long nr_to_write = wbc->nr_to_write; |
Christoph Hellwig | 01c0319 | 2009-06-08 13:35:40 +0200 | [diff] [blame] | 1298 | unsigned dirty; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1299 | int ret; |
| 1300 | |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1301 | WARN_ON(!(inode->i_state & I_SYNC)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1302 | |
Tejun Heo | 9fb0a7d | 2013-01-11 13:06:37 -0800 | [diff] [blame] | 1303 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); |
| 1304 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1305 | ret = do_writepages(mapping, wbc); |
| 1306 | |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 1307 | /* |
| 1308 | * Make sure to wait on the data before writing out the metadata. |
| 1309 | * This is important for filesystems that modify metadata on data |
Dave Chinner | 7747bd4 | 2013-07-02 22:38:35 +1000 | [diff] [blame] | 1310 | * I/O completion. We don't do it for sync(2) writeback because it has a |
| 1311 | * separate, external IO completion path and ->sync_fs for guaranteeing |
| 1312 | * inode metadata is written back correctly. |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 1313 | */ |
Dave Chinner | 7747bd4 | 2013-07-02 22:38:35 +1000 | [diff] [blame] | 1314 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 1315 | int err = filemap_fdatawait(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1316 | if (ret == 0) |
| 1317 | ret = err; |
| 1318 | } |
| 1319 | |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 1320 | /* |
| 1321 | * Some filesystems may redirty the inode during the writeback |
| 1322 | * due to delalloc, clear dirty metadata flags right before |
| 1323 | * write_inode() |
| 1324 | */ |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1325 | spin_lock(&inode->i_lock); |
Tejun Heo | 9c6ac78 | 2014-10-24 15:38:21 -0400 | [diff] [blame] | 1326 | |
Dmitry Monakhov | 5547e8a | 2010-05-07 13:35:44 +0400 | [diff] [blame] | 1327 | dirty = inode->i_state & I_DIRTY; |
Theodore Ts'o | a2f4870 | 2015-03-17 12:23:19 -0400 | [diff] [blame] | 1328 | if (inode->i_state & I_DIRTY_TIME) { |
| 1329 | if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || |
Jan Kara | dc5ff2b | 2016-07-26 11:38:20 +0200 | [diff] [blame] | 1330 | wbc->sync_mode == WB_SYNC_ALL || |
Theodore Ts'o | a2f4870 | 2015-03-17 12:23:19 -0400 | [diff] [blame] | 1331 | unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || |
| 1332 | unlikely(time_after(jiffies, |
| 1333 | (inode->dirtied_time_when + |
| 1334 | dirtytime_expire_interval * HZ)))) { |
| 1335 | dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; |
| 1336 | trace_writeback_lazytime(inode); |
| 1337 | } |
| 1338 | } else |
| 1339 | inode->i_state &= ~I_DIRTY_TIME_EXPIRED; |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1340 | inode->i_state &= ~dirty; |
Tejun Heo | 9c6ac78 | 2014-10-24 15:38:21 -0400 | [diff] [blame] | 1341 | |
| 1342 | /* |
| 1343 | * Paired with smp_mb() in __mark_inode_dirty(). This allows |
| 1344 | * __mark_inode_dirty() to test i_state without grabbing i_lock - |
| 1345 | * either they see the I_DIRTY bits cleared or we see the dirtied |
| 1346 | * inode. |
| 1347 | * |
| 1348 | * I_DIRTY_PAGES is always cleared together above even if @mapping |
| 1349 | * still has dirty pages. The flag is reinstated after smp_mb() if |
| 1350 | * necessary. This guarantees that either __mark_inode_dirty() |
| 1351 | * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. |
| 1352 | */ |
| 1353 | smp_mb(); |
| 1354 | |
| 1355 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
| 1356 | inode->i_state |= I_DIRTY_PAGES; |
| 1357 | |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1358 | spin_unlock(&inode->i_lock); |
Tejun Heo | 9c6ac78 | 2014-10-24 15:38:21 -0400 | [diff] [blame] | 1359 | |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1360 | if (dirty & I_DIRTY_TIME) |
| 1361 | mark_inode_dirty_sync(inode); |
Christoph Hellwig | 26821ed | 2010-03-05 09:21:21 +0100 | [diff] [blame] | 1362 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1363 | if (dirty & ~I_DIRTY_PAGES) { |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 1364 | int err = write_inode(inode, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1365 | if (ret == 0) |
| 1366 | ret = err; |
| 1367 | } |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1368 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
| 1369 | return ret; |
| 1370 | } |
| 1371 | |
| 1372 | /* |
| 1373 | * Write out an inode's dirty pages. Either the caller has an active reference |
| 1374 | * on the inode or the inode has I_WILL_FREE set. |
| 1375 | * |
| 1376 | * This function is designed to be called for writing back one inode which |
| 1377 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() |
| 1378 | * and does more profound writeback list handling in writeback_sb_inodes(). |
| 1379 | */ |
Tejun Heo | aaf2559 | 2016-03-18 13:52:04 -0400 | [diff] [blame] | 1380 | static int writeback_single_inode(struct inode *inode, |
| 1381 | struct writeback_control *wbc) |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1382 | { |
Tejun Heo | aaf2559 | 2016-03-18 13:52:04 -0400 | [diff] [blame] | 1383 | struct bdi_writeback *wb; |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1384 | int ret = 0; |
| 1385 | |
| 1386 | spin_lock(&inode->i_lock); |
| 1387 | if (!atomic_read(&inode->i_count)) |
| 1388 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); |
| 1389 | else |
| 1390 | WARN_ON(inode->i_state & I_WILL_FREE); |
| 1391 | |
| 1392 | if (inode->i_state & I_SYNC) { |
| 1393 | if (wbc->sync_mode != WB_SYNC_ALL) |
| 1394 | goto out; |
| 1395 | /* |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 1396 | * It's a data-integrity sync. We must wait. Since callers hold |
| 1397 | * inode reference or inode has I_WILL_FREE set, it cannot go |
| 1398 | * away under us. |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1399 | */ |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 1400 | __inode_wait_for_writeback(inode); |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1401 | } |
| 1402 | WARN_ON(inode->i_state & I_SYNC); |
| 1403 | /* |
Jan Kara | f9b0e05 | 2013-12-14 04:21:26 +0800 | [diff] [blame] | 1404 | * Skip inode if it is clean and we have no outstanding writeback in |
| 1405 | * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this |
| 1406 | * function since flusher thread may be doing for example sync in |
| 1407 | * parallel and if we move the inode, it could get skipped. So here we |
| 1408 | * make sure inode is on some writeback list and leave it there unless |
| 1409 | * we have completely cleaned the inode. |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1410 | */ |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1411 | if (!(inode->i_state & I_DIRTY_ALL) && |
Jan Kara | f9b0e05 | 2013-12-14 04:21:26 +0800 | [diff] [blame] | 1412 | (wbc->sync_mode != WB_SYNC_ALL || |
| 1413 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1414 | goto out; |
| 1415 | inode->i_state |= I_SYNC; |
Tejun Heo | b16b1de | 2015-06-02 08:39:48 -0600 | [diff] [blame] | 1416 | wbc_attach_and_unlock_inode(wbc, inode); |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1417 | |
Yan Hong | cd8ed2a | 2012-10-08 16:33:45 -0700 | [diff] [blame] | 1418 | ret = __writeback_single_inode(inode, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1419 | |
Tejun Heo | b16b1de | 2015-06-02 08:39:48 -0600 | [diff] [blame] | 1420 | wbc_detach_inode(wbc); |
Tejun Heo | aaf2559 | 2016-03-18 13:52:04 -0400 | [diff] [blame] | 1421 | |
| 1422 | wb = inode_to_wb_and_lock_list(inode); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1423 | spin_lock(&inode->i_lock); |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1424 | /* |
| 1425 | * If inode is clean, remove it from writeback lists. Otherwise don't |
| 1426 | * touch it. See comment above for explanation. |
| 1427 | */ |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1428 | if (!(inode->i_state & I_DIRTY_ALL)) |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 1429 | inode_io_list_del_locked(inode, wb); |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1430 | spin_unlock(&wb->list_lock); |
Joern Engel | 1c0eeaf | 2007-10-16 23:30:44 -0700 | [diff] [blame] | 1431 | inode_sync_complete(inode); |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1432 | out: |
| 1433 | spin_unlock(&inode->i_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1434 | return ret; |
| 1435 | } |
| 1436 | |
Tejun Heo | a88a341 | 2015-05-22 17:13:28 -0400 | [diff] [blame] | 1437 | static long writeback_chunk_size(struct bdi_writeback *wb, |
Wu Fengguang | 1a12d8b | 2010-08-29 13:28:09 -0600 | [diff] [blame] | 1438 | struct wb_writeback_work *work) |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1439 | { |
| 1440 | long pages; |
| 1441 | |
| 1442 | /* |
| 1443 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty |
| 1444 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX |
| 1445 | * here avoids calling into writeback_inodes_wb() more than once. |
| 1446 | * |
| 1447 | * The intended call sequence for WB_SYNC_ALL writeback is: |
| 1448 | * |
| 1449 | * wb_writeback() |
| 1450 | * writeback_sb_inodes() <== called only once |
| 1451 | * write_cache_pages() <== called once for each inode |
| 1452 | * (quickly) tag currently dirty pages |
| 1453 | * (maybe slowly) sync all tagged pages |
| 1454 | */ |
| 1455 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) |
| 1456 | pages = LONG_MAX; |
Wu Fengguang | 1a12d8b | 2010-08-29 13:28:09 -0600 | [diff] [blame] | 1457 | else { |
Tejun Heo | a88a341 | 2015-05-22 17:13:28 -0400 | [diff] [blame] | 1458 | pages = min(wb->avg_write_bandwidth / 2, |
Tejun Heo | dcc25ae | 2015-05-22 18:23:22 -0400 | [diff] [blame] | 1459 | global_wb_domain.dirty_limit / DIRTY_SCOPE); |
Wu Fengguang | 1a12d8b | 2010-08-29 13:28:09 -0600 | [diff] [blame] | 1460 | pages = min(pages, work->nr_pages); |
| 1461 | pages = round_down(pages + MIN_WRITEBACK_PAGES, |
| 1462 | MIN_WRITEBACK_PAGES); |
| 1463 | } |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1464 | |
| 1465 | return pages; |
| 1466 | } |
| 1467 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1468 | /* |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 1469 | * Write a portion of b_io inodes which belong to @sb. |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 1470 | * |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1471 | * Return the number of pages and/or inodes written. |
Linus Torvalds | 0ba13fd | 2015-09-11 13:26:39 -0700 | [diff] [blame] | 1472 | * |
| 1473 | * NOTE! This is called with wb->list_lock held, and will |
| 1474 | * unlock and relock that for each inode it ends up doing |
| 1475 | * IO for. |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 1476 | */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1477 | static long writeback_sb_inodes(struct super_block *sb, |
| 1478 | struct bdi_writeback *wb, |
| 1479 | struct wb_writeback_work *work) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1480 | { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1481 | struct writeback_control wbc = { |
| 1482 | .sync_mode = work->sync_mode, |
| 1483 | .tagged_writepages = work->tagged_writepages, |
| 1484 | .for_kupdate = work->for_kupdate, |
| 1485 | .for_background = work->for_background, |
Dave Chinner | 7747bd4 | 2013-07-02 22:38:35 +1000 | [diff] [blame] | 1486 | .for_sync = work->for_sync, |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1487 | .range_cyclic = work->range_cyclic, |
| 1488 | .range_start = 0, |
| 1489 | .range_end = LLONG_MAX, |
| 1490 | }; |
| 1491 | unsigned long start_time = jiffies; |
| 1492 | long write_chunk; |
| 1493 | long wrote = 0; /* count both pages and inodes */ |
| 1494 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1495 | while (!list_empty(&wb->b_io)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 1496 | struct inode *inode = wb_inode(wb->b_io.prev); |
Tejun Heo | aaf2559 | 2016-03-18 13:52:04 -0400 | [diff] [blame] | 1497 | struct bdi_writeback *tmp_wb; |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 1498 | |
| 1499 | if (inode->i_sb != sb) { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1500 | if (work->sb) { |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 1501 | /* |
| 1502 | * We only want to write back data for this |
| 1503 | * superblock, move all inodes not belonging |
| 1504 | * to it back onto the dirty list. |
| 1505 | */ |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1506 | redirty_tail(inode, wb); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 1507 | continue; |
| 1508 | } |
| 1509 | |
| 1510 | /* |
| 1511 | * The inode belongs to a different superblock. |
| 1512 | * Bounce back to the caller to unpin this and |
| 1513 | * pin the next superblock. |
| 1514 | */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1515 | break; |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 1516 | } |
| 1517 | |
Christoph Hellwig | 9843b76 | 2010-10-24 19:40:46 +0200 | [diff] [blame] | 1518 | /* |
Wanpeng Li | 331cbde | 2012-06-09 11:10:55 +0800 | [diff] [blame] | 1519 | * Don't bother with new inodes or inodes being freed, first |
| 1520 | * kind does not need periodic writeout yet, and for the latter |
Christoph Hellwig | 9843b76 | 2010-10-24 19:40:46 +0200 | [diff] [blame] | 1521 | * kind writeout is handled by the freer. |
| 1522 | */ |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1523 | spin_lock(&inode->i_lock); |
Christoph Hellwig | 9843b76 | 2010-10-24 19:40:46 +0200 | [diff] [blame] | 1524 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1525 | spin_unlock(&inode->i_lock); |
Wu Fengguang | fcc5c22 | 2011-07-11 23:08:50 -0700 | [diff] [blame] | 1526 | redirty_tail(inode, wb); |
Nick Piggin | 7ef0d73 | 2009-03-12 14:31:38 -0700 | [diff] [blame] | 1527 | continue; |
| 1528 | } |
Jan Kara | cc1676d | 2012-05-03 14:47:56 +0200 | [diff] [blame] | 1529 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
| 1530 | /* |
| 1531 | * If this inode is locked for writeback and we are not |
| 1532 | * doing writeback-for-data-integrity, move it to |
| 1533 | * b_more_io so that writeback can proceed with the |
| 1534 | * other inodes on s_io. |
| 1535 | * |
| 1536 | * We'll have another go at writing back this inode |
| 1537 | * when we completed a full scan of b_io. |
| 1538 | */ |
| 1539 | spin_unlock(&inode->i_lock); |
| 1540 | requeue_io(inode, wb); |
| 1541 | trace_writeback_sb_inodes_requeue(inode); |
| 1542 | continue; |
| 1543 | } |
Jan Kara | f0d07b7 | 2012-05-03 14:47:59 +0200 | [diff] [blame] | 1544 | spin_unlock(&wb->list_lock); |
| 1545 | |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1546 | /* |
| 1547 | * We already requeued the inode if it had I_SYNC set and we |
| 1548 | * are doing WB_SYNC_NONE writeback. So this catches only the |
| 1549 | * WB_SYNC_ALL case. |
| 1550 | */ |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 1551 | if (inode->i_state & I_SYNC) { |
| 1552 | /* Wait for I_SYNC. This function drops i_lock... */ |
| 1553 | inode_sleep_on_writeback(inode); |
| 1554 | /* Inode may be gone, start again */ |
Jan Kara | ead188f | 2012-06-08 17:07:36 +0200 | [diff] [blame] | 1555 | spin_lock(&wb->list_lock); |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 1556 | continue; |
| 1557 | } |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1558 | inode->i_state |= I_SYNC; |
Tejun Heo | b16b1de | 2015-06-02 08:39:48 -0600 | [diff] [blame] | 1559 | wbc_attach_and_unlock_inode(&wbc, inode); |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 1560 | |
Tejun Heo | a88a341 | 2015-05-22 17:13:28 -0400 | [diff] [blame] | 1561 | write_chunk = writeback_chunk_size(wb, work); |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1562 | wbc.nr_to_write = write_chunk; |
| 1563 | wbc.pages_skipped = 0; |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1564 | |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 1565 | /* |
| 1566 | * We use I_SYNC to pin the inode in memory. While it is set |
| 1567 | * evict_inode() will wait so the inode cannot be freed. |
| 1568 | */ |
Yan Hong | cd8ed2a | 2012-10-08 16:33:45 -0700 | [diff] [blame] | 1569 | __writeback_single_inode(inode, &wbc); |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1570 | |
Tejun Heo | b16b1de | 2015-06-02 08:39:48 -0600 | [diff] [blame] | 1571 | wbc_detach_inode(&wbc); |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1572 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
| 1573 | wrote += write_chunk - wbc.nr_to_write; |
Chris Mason | 590dca3 | 2015-09-18 13:35:08 -0400 | [diff] [blame] | 1574 | |
| 1575 | if (need_resched()) { |
| 1576 | /* |
| 1577 | * We're trying to balance between building up a nice |
| 1578 | * long list of IOs to improve our merge rate, and |
| 1579 | * getting those IOs out quickly for anyone throttling |
| 1580 | * in balance_dirty_pages(). cond_resched() doesn't |
| 1581 | * unplug, so get our IOs out the door before we |
| 1582 | * give up the CPU. |
| 1583 | */ |
| 1584 | blk_flush_plug(current); |
| 1585 | cond_resched(); |
| 1586 | } |
| 1587 | |
Tejun Heo | aaf2559 | 2016-03-18 13:52:04 -0400 | [diff] [blame] | 1588 | /* |
| 1589 | * Requeue @inode if still dirty. Be careful as @inode may |
| 1590 | * have been switched to another wb in the meantime. |
| 1591 | */ |
| 1592 | tmp_wb = inode_to_wb_and_lock_list(inode); |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1593 | spin_lock(&inode->i_lock); |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 1594 | if (!(inode->i_state & I_DIRTY_ALL)) |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1595 | wrote++; |
Tejun Heo | aaf2559 | 2016-03-18 13:52:04 -0400 | [diff] [blame] | 1596 | requeue_inode(inode, tmp_wb, &wbc); |
Jan Kara | 4f8ad65 | 2012-05-03 14:48:00 +0200 | [diff] [blame] | 1597 | inode_sync_complete(inode); |
Dave Chinner | 0f1b1fd | 2011-03-22 22:23:43 +1100 | [diff] [blame] | 1598 | spin_unlock(&inode->i_lock); |
Chris Mason | 590dca3 | 2015-09-18 13:35:08 -0400 | [diff] [blame] | 1599 | |
Tejun Heo | aaf2559 | 2016-03-18 13:52:04 -0400 | [diff] [blame] | 1600 | if (unlikely(tmp_wb != wb)) { |
| 1601 | spin_unlock(&tmp_wb->list_lock); |
| 1602 | spin_lock(&wb->list_lock); |
| 1603 | } |
| 1604 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1605 | /* |
| 1606 | * bail out to wb_writeback() often enough to check |
| 1607 | * background threshold and other termination conditions. |
| 1608 | */ |
| 1609 | if (wrote) { |
| 1610 | if (time_is_before_jiffies(start_time + HZ / 10UL)) |
| 1611 | break; |
| 1612 | if (work->nr_pages <= 0) |
| 1613 | break; |
Fengguang Wu | 8bc3be2 | 2008-02-04 22:29:36 -0800 | [diff] [blame] | 1614 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1615 | } |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1616 | return wrote; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 1617 | } |
Nick Piggin | 38f2197 | 2009-01-06 14:40:25 -0800 | [diff] [blame] | 1618 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1619 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
| 1620 | struct wb_writeback_work *work) |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 1621 | { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1622 | unsigned long start_time = jiffies; |
| 1623 | long wrote = 0; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 1624 | |
| 1625 | while (!list_empty(&wb->b_io)) { |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 1626 | struct inode *inode = wb_inode(wb->b_io.prev); |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 1627 | struct super_block *sb = inode->i_sb; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 1628 | |
Konstantin Khlebnikov | eb6ef3d | 2015-02-19 20:19:35 +0300 | [diff] [blame] | 1629 | if (!trylock_super(sb)) { |
Wu Fengguang | 0e995816 | 2011-07-29 22:14:35 -0600 | [diff] [blame] | 1630 | /* |
Konstantin Khlebnikov | eb6ef3d | 2015-02-19 20:19:35 +0300 | [diff] [blame] | 1631 | * trylock_super() may fail consistently due to |
Wu Fengguang | 0e995816 | 2011-07-29 22:14:35 -0600 | [diff] [blame] | 1632 | * s_umount being grabbed by someone else. Don't use |
| 1633 | * requeue_io() to avoid busy retrying the inode/sb. |
| 1634 | */ |
| 1635 | redirty_tail(inode, wb); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 1636 | continue; |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 1637 | } |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1638 | wrote += writeback_sb_inodes(sb, wb, work); |
Konstantin Khlebnikov | eb6ef3d | 2015-02-19 20:19:35 +0300 | [diff] [blame] | 1639 | up_read(&sb->s_umount); |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 1640 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1641 | /* refer to the same tests at the end of writeback_sb_inodes */ |
| 1642 | if (wrote) { |
| 1643 | if (time_is_before_jiffies(start_time + HZ / 10UL)) |
| 1644 | break; |
| 1645 | if (work->nr_pages <= 0) |
| 1646 | break; |
| 1647 | } |
Edward Shishkin | f11c9c5 | 2010-03-11 14:09:47 -0800 | [diff] [blame] | 1648 | } |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1649 | /* Leave any unwritten inodes on b_io */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1650 | return wrote; |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1651 | } |
| 1652 | |
Wanpeng Li | 7d9f073 | 2013-09-11 14:22:40 -0700 | [diff] [blame] | 1653 | static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1654 | enum wb_reason reason) |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 1655 | { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1656 | struct wb_writeback_work work = { |
| 1657 | .nr_pages = nr_pages, |
| 1658 | .sync_mode = WB_SYNC_NONE, |
| 1659 | .range_cyclic = 1, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1660 | .reason = reason, |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1661 | }; |
Linus Torvalds | 505a666 | 2015-09-11 13:37:19 -0700 | [diff] [blame] | 1662 | struct blk_plug plug; |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 1663 | |
Linus Torvalds | 505a666 | 2015-09-11 13:37:19 -0700 | [diff] [blame] | 1664 | blk_start_plug(&plug); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1665 | spin_lock(&wb->list_lock); |
Wu Fengguang | 424b351 | 2010-07-21 20:11:53 -0600 | [diff] [blame] | 1666 | if (list_empty(&wb->b_io)) |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 1667 | queue_io(wb, &work); |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1668 | __writeback_inodes_wb(wb, &work); |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 1669 | spin_unlock(&wb->list_lock); |
Linus Torvalds | 505a666 | 2015-09-11 13:37:19 -0700 | [diff] [blame] | 1670 | blk_finish_plug(&plug); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 1671 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1672 | return nr_pages - work.nr_pages; |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 1673 | } |
| 1674 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1675 | /* |
| 1676 | * Explicit flushing or periodic writeback of "old" data. |
| 1677 | * |
| 1678 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
| 1679 | * dirtying-time in the inode's address_space. So this periodic writeback code |
| 1680 | * just walks the superblock inode list, writing back any inodes which are |
| 1681 | * older than a specific point in time. |
| 1682 | * |
| 1683 | * Try to run once per dirty_writeback_interval. But if a writeback event |
| 1684 | * takes longer than a dirty_writeback_interval interval, then leave a |
| 1685 | * one-second gap. |
| 1686 | * |
| 1687 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
| 1688 | * all dirty pages if they are all attached to "old" mappings. |
| 1689 | */ |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 1690 | static long wb_writeback(struct bdi_writeback *wb, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1691 | struct wb_writeback_work *work) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1692 | { |
Wu Fengguang | e98be2d | 2010-08-29 11:22:30 -0600 | [diff] [blame] | 1693 | unsigned long wb_start = jiffies; |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1694 | long nr_pages = work->nr_pages; |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 1695 | unsigned long oldest_jif; |
Jan Kara | a5989bd | 2009-09-16 19:22:48 +0200 | [diff] [blame] | 1696 | struct inode *inode; |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1697 | long progress; |
Linus Torvalds | 505a666 | 2015-09-11 13:37:19 -0700 | [diff] [blame] | 1698 | struct blk_plug plug; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1699 | |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 1700 | oldest_jif = jiffies; |
| 1701 | work->older_than_this = &oldest_jif; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1702 | |
Linus Torvalds | 505a666 | 2015-09-11 13:37:19 -0700 | [diff] [blame] | 1703 | blk_start_plug(&plug); |
Wu Fengguang | e8dfc30 | 2011-04-21 12:06:32 -0600 | [diff] [blame] | 1704 | spin_lock(&wb->list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1705 | for (;;) { |
| 1706 | /* |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 1707 | * Stop writeback when nr_pages has been consumed |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1708 | */ |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1709 | if (work->nr_pages <= 0) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1710 | break; |
| 1711 | |
| 1712 | /* |
Jan Kara | aa373cf | 2011-01-13 15:45:47 -0800 | [diff] [blame] | 1713 | * Background writeout and kupdate-style writeback may |
| 1714 | * run forever. Stop them if there is other work to do |
| 1715 | * so that e.g. sync can proceed. They'll be restarted |
| 1716 | * after the other works are all done. |
| 1717 | */ |
| 1718 | if ((work->for_background || work->for_kupdate) && |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 1719 | !list_empty(&wb->work_list)) |
Jan Kara | aa373cf | 2011-01-13 15:45:47 -0800 | [diff] [blame] | 1720 | break; |
| 1721 | |
| 1722 | /* |
Wu Fengguang | d3ddec7 | 2009-09-23 20:33:40 +0800 | [diff] [blame] | 1723 | * For background writeout, stop when we are below the |
| 1724 | * background dirty threshold |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1725 | */ |
Tejun Heo | aa661bb | 2015-05-22 18:23:31 -0400 | [diff] [blame] | 1726 | if (work->for_background && !wb_over_bg_thresh(wb)) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1727 | break; |
| 1728 | |
Jan Kara | 1bc36b6 | 2011-10-19 11:44:41 +0200 | [diff] [blame] | 1729 | /* |
| 1730 | * Kupdate and background works are special and we want to |
| 1731 | * include all inodes that need writing. Livelock avoidance is |
| 1732 | * handled by these works yielding to any other work so we are |
| 1733 | * safe. |
| 1734 | */ |
Wu Fengguang | ba9aa83 | 2010-07-21 20:32:30 -0600 | [diff] [blame] | 1735 | if (work->for_kupdate) { |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 1736 | oldest_jif = jiffies - |
Wu Fengguang | ba9aa83 | 2010-07-21 20:32:30 -0600 | [diff] [blame] | 1737 | msecs_to_jiffies(dirty_expire_interval * 10); |
Jan Kara | 1bc36b6 | 2011-10-19 11:44:41 +0200 | [diff] [blame] | 1738 | } else if (work->for_background) |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 1739 | oldest_jif = jiffies; |
Dave Chinner | 028c2dd | 2010-07-07 13:24:07 +1000 | [diff] [blame] | 1740 | |
Tejun Heo | 5634cc2 | 2015-08-18 14:54:56 -0700 | [diff] [blame] | 1741 | trace_writeback_start(wb, work); |
Wu Fengguang | e8dfc30 | 2011-04-21 12:06:32 -0600 | [diff] [blame] | 1742 | if (list_empty(&wb->b_io)) |
Curt Wohlgemuth | ad4e38d | 2011-10-07 21:51:56 -0600 | [diff] [blame] | 1743 | queue_io(wb, work); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1744 | if (work->sb) |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1745 | progress = writeback_sb_inodes(work->sb, wb, work); |
Christoph Hellwig | edadfb1 | 2010-06-10 12:07:54 +0200 | [diff] [blame] | 1746 | else |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1747 | progress = __writeback_inodes_wb(wb, work); |
Tejun Heo | 5634cc2 | 2015-08-18 14:54:56 -0700 | [diff] [blame] | 1748 | trace_writeback_written(wb, work); |
Dave Chinner | 028c2dd | 2010-07-07 13:24:07 +1000 | [diff] [blame] | 1749 | |
Wu Fengguang | e98be2d | 2010-08-29 11:22:30 -0600 | [diff] [blame] | 1750 | wb_update_bandwidth(wb, wb_start); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1751 | |
| 1752 | /* |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 1753 | * Did we write something? Try for more |
Wu Fengguang | e6fb6da | 2010-07-22 10:23:44 -0600 | [diff] [blame] | 1754 | * |
| 1755 | * Dirty inodes are moved to b_io for writeback in batches. |
| 1756 | * The completion of the current batch does not necessarily |
| 1757 | * mean the overall work is done. So we keep looping as long |
| 1758 | * as made some progress on cleaning pages or inodes. |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 1759 | */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1760 | if (progress) |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 1761 | continue; |
| 1762 | /* |
Wu Fengguang | e6fb6da | 2010-07-22 10:23:44 -0600 | [diff] [blame] | 1763 | * No more inodes for IO, bail |
Jens Axboe | 71fd05a | 2009-09-23 19:32:26 +0200 | [diff] [blame] | 1764 | */ |
Wu Fengguang | b7a2441 | 2010-07-21 22:19:51 -0600 | [diff] [blame] | 1765 | if (list_empty(&wb->b_more_io)) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1766 | break; |
| 1767 | /* |
Jens Axboe | 8010c3b | 2009-09-15 20:04:57 +0200 | [diff] [blame] | 1768 | * Nothing written. Wait for some inode to |
| 1769 | * become available for writeback. Otherwise |
| 1770 | * we'll just busyloop. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1771 | */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1772 | if (!list_empty(&wb->b_more_io)) { |
Tejun Heo | 5634cc2 | 2015-08-18 14:54:56 -0700 | [diff] [blame] | 1773 | trace_writeback_wait(wb, work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1774 | inode = wb_inode(wb->b_more_io.prev); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 1775 | spin_lock(&inode->i_lock); |
Jan Kara | f0d07b7 | 2012-05-03 14:47:59 +0200 | [diff] [blame] | 1776 | spin_unlock(&wb->list_lock); |
Jan Kara | 169ebd9 | 2012-05-03 14:48:03 +0200 | [diff] [blame] | 1777 | /* This function drops i_lock... */ |
| 1778 | inode_sleep_on_writeback(inode); |
Jan Kara | f0d07b7 | 2012-05-03 14:47:59 +0200 | [diff] [blame] | 1779 | spin_lock(&wb->list_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1780 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1781 | } |
Wu Fengguang | e8dfc30 | 2011-04-21 12:06:32 -0600 | [diff] [blame] | 1782 | spin_unlock(&wb->list_lock); |
Linus Torvalds | 505a666 | 2015-09-11 13:37:19 -0700 | [diff] [blame] | 1783 | blk_finish_plug(&plug); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1784 | |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame] | 1785 | return nr_pages - work->nr_pages; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1786 | } |
| 1787 | |
| 1788 | /* |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1789 | * 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] | 1790 | */ |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 1791 | static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1792 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1793 | struct wb_writeback_work *work = NULL; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1794 | |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 1795 | spin_lock_bh(&wb->work_lock); |
| 1796 | if (!list_empty(&wb->work_list)) { |
| 1797 | work = list_entry(wb->work_list.next, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1798 | struct wb_writeback_work, list); |
| 1799 | list_del_init(&work->list); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1800 | } |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 1801 | spin_unlock_bh(&wb->work_lock); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1802 | return work; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1803 | } |
| 1804 | |
Linus Torvalds | cdf01dd | 2010-10-30 08:55:52 -0700 | [diff] [blame] | 1805 | /* |
| 1806 | * Add in the number of potentially dirty inodes, because each inode |
| 1807 | * write can dirty pagecache in the underlying blockdev. |
| 1808 | */ |
| 1809 | static unsigned long get_nr_dirty_pages(void) |
| 1810 | { |
Mel Gorman | 11fb998 | 2016-07-28 15:46:20 -0700 | [diff] [blame] | 1811 | return global_node_page_state(NR_FILE_DIRTY) + |
| 1812 | global_node_page_state(NR_UNSTABLE_NFS) + |
Linus Torvalds | cdf01dd | 2010-10-30 08:55:52 -0700 | [diff] [blame] | 1813 | get_nr_dirty_inodes(); |
| 1814 | } |
| 1815 | |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 1816 | static long wb_check_background_flush(struct bdi_writeback *wb) |
| 1817 | { |
Tejun Heo | aa661bb | 2015-05-22 18:23:31 -0400 | [diff] [blame] | 1818 | if (wb_over_bg_thresh(wb)) { |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 1819 | |
| 1820 | struct wb_writeback_work work = { |
| 1821 | .nr_pages = LONG_MAX, |
| 1822 | .sync_mode = WB_SYNC_NONE, |
| 1823 | .for_background = 1, |
| 1824 | .range_cyclic = 1, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1825 | .reason = WB_REASON_BACKGROUND, |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 1826 | }; |
| 1827 | |
| 1828 | return wb_writeback(wb, &work); |
| 1829 | } |
| 1830 | |
| 1831 | return 0; |
| 1832 | } |
| 1833 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1834 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
| 1835 | { |
| 1836 | unsigned long expired; |
| 1837 | long nr_pages; |
| 1838 | |
Jens Axboe | 69b62d0 | 2010-05-17 12:51:03 +0200 | [diff] [blame] | 1839 | /* |
| 1840 | * When set to zero, disable periodic writeback |
| 1841 | */ |
| 1842 | if (!dirty_writeback_interval) |
| 1843 | return 0; |
| 1844 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1845 | expired = wb->last_old_flush + |
| 1846 | msecs_to_jiffies(dirty_writeback_interval * 10); |
| 1847 | if (time_before(jiffies, expired)) |
| 1848 | return 0; |
| 1849 | |
| 1850 | wb->last_old_flush = jiffies; |
Linus Torvalds | cdf01dd | 2010-10-30 08:55:52 -0700 | [diff] [blame] | 1851 | nr_pages = get_nr_dirty_pages(); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1852 | |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 1853 | if (nr_pages) { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1854 | struct wb_writeback_work work = { |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 1855 | .nr_pages = nr_pages, |
| 1856 | .sync_mode = WB_SYNC_NONE, |
| 1857 | .for_kupdate = 1, |
| 1858 | .range_cyclic = 1, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1859 | .reason = WB_REASON_PERIODIC, |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 1860 | }; |
| 1861 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1862 | return wb_writeback(wb, &work); |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 1863 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1864 | |
| 1865 | return 0; |
| 1866 | } |
| 1867 | |
| 1868 | /* |
| 1869 | * Retrieve work items and do the writeback they describe |
| 1870 | */ |
Wanpeng Li | 25d130b | 2013-07-08 16:00:14 -0700 | [diff] [blame] | 1871 | static long wb_do_writeback(struct bdi_writeback *wb) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1872 | { |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1873 | struct wb_writeback_work *work; |
Jens Axboe | c4a77a6 | 2009-09-16 15:18:25 +0200 | [diff] [blame] | 1874 | long wrote = 0; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1875 | |
Tejun Heo | 4452226 | 2015-05-22 17:13:26 -0400 | [diff] [blame] | 1876 | set_bit(WB_writeback_running, &wb->state); |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 1877 | while ((work = get_next_work_item(wb)) != NULL) { |
Tejun Heo | cc395d7 | 2015-05-22 17:13:58 -0400 | [diff] [blame] | 1878 | struct wb_completion *done = work->done; |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1879 | |
Tejun Heo | 5634cc2 | 2015-08-18 14:54:56 -0700 | [diff] [blame] | 1880 | trace_writeback_exec(wb, work); |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 1881 | |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1882 | wrote += wb_writeback(wb, work); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1883 | |
Tejun Heo | 8a1270c | 2015-08-18 14:54:53 -0700 | [diff] [blame] | 1884 | if (work->auto_free) |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 1885 | kfree(work); |
Tejun Heo | cc395d7 | 2015-05-22 17:13:58 -0400 | [diff] [blame] | 1886 | if (done && atomic_dec_and_test(&done->cnt)) |
| 1887 | wake_up_all(&wb->bdi->wb_waitq); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1888 | } |
| 1889 | |
| 1890 | /* |
| 1891 | * Check for periodic writeback, kupdated() style |
| 1892 | */ |
| 1893 | wrote += wb_check_old_data_flush(wb); |
Jan Kara | 6585027 | 2011-01-13 15:45:44 -0800 | [diff] [blame] | 1894 | wrote += wb_check_background_flush(wb); |
Tejun Heo | 4452226 | 2015-05-22 17:13:26 -0400 | [diff] [blame] | 1895 | clear_bit(WB_writeback_running, &wb->state); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1896 | |
| 1897 | return wrote; |
| 1898 | } |
| 1899 | |
| 1900 | /* |
| 1901 | * Handle writeback of dirty data for the device backed by this bdi. Also |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1902 | * reschedules periodically and does kupdated style flushing. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1903 | */ |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 1904 | void wb_workfn(struct work_struct *work) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1905 | { |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1906 | struct bdi_writeback *wb = container_of(to_delayed_work(work), |
| 1907 | struct bdi_writeback, dwork); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1908 | long pages_written; |
| 1909 | |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 1910 | set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); |
Peter Zijlstra | 766f916 | 2010-10-26 14:22:45 -0700 | [diff] [blame] | 1911 | current->flags |= PF_SWAPWRITE; |
Christoph Hellwig | 0824390 | 2010-06-19 23:08:22 +0200 | [diff] [blame] | 1912 | |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1913 | if (likely(!current_is_workqueue_rescuer() || |
Tejun Heo | 4452226 | 2015-05-22 17:13:26 -0400 | [diff] [blame] | 1914 | !test_bit(WB_registered, &wb->state))) { |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 1915 | /* |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 1916 | * The normal path. Keep writing back @wb until its |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1917 | * work_list is empty. Note that this path is also taken |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 1918 | * if @wb is shutting down even when we're running off the |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1919 | * rescuer as work_list needs to be drained. |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 1920 | */ |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1921 | do { |
Wanpeng Li | 25d130b | 2013-07-08 16:00:14 -0700 | [diff] [blame] | 1922 | pages_written = wb_do_writeback(wb); |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1923 | trace_writeback_pages_written(pages_written); |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 1924 | } while (!list_empty(&wb->work_list)); |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1925 | } else { |
| 1926 | /* |
| 1927 | * bdi_wq can't get enough workers and we're running off |
| 1928 | * the emergency worker. Don't hog it. Hopefully, 1024 is |
| 1929 | * enough for efficient IO. |
| 1930 | */ |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 1931 | pages_written = writeback_inodes_wb(wb, 1024, |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1932 | WB_REASON_FORKER_THREAD); |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 1933 | trace_writeback_pages_written(pages_written); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1934 | } |
| 1935 | |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 1936 | if (!list_empty(&wb->work_list)) |
Derek Basehore | 6ca738d | 2014-04-03 14:46:22 -0700 | [diff] [blame] | 1937 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
| 1938 | else if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
Tejun Heo | f0054bb | 2015-05-22 17:13:30 -0400 | [diff] [blame] | 1939 | wb_wakeup_delayed(wb); |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 1940 | |
Tejun Heo | 839a8e8 | 2013-04-01 19:08:06 -0700 | [diff] [blame] | 1941 | current->flags &= ~PF_SWAPWRITE; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1942 | } |
| 1943 | |
| 1944 | /* |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1945 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
| 1946 | * the whole world. |
| 1947 | */ |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 1948 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1949 | { |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 1950 | struct backing_dev_info *bdi; |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 1951 | |
Jan Kara | 47df3dd | 2013-09-11 14:22:22 -0700 | [diff] [blame] | 1952 | if (!nr_pages) |
| 1953 | nr_pages = get_nr_dirty_pages(); |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 1954 | |
| 1955 | rcu_read_lock(); |
Tejun Heo | f2b6512 | 2015-05-22 17:13:55 -0400 | [diff] [blame] | 1956 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
| 1957 | struct bdi_writeback *wb; |
Tejun Heo | f2b6512 | 2015-05-22 17:13:55 -0400 | [diff] [blame] | 1958 | |
| 1959 | if (!bdi_has_dirty_io(bdi)) |
| 1960 | continue; |
| 1961 | |
Tejun Heo | b817525 | 2015-10-02 14:47:05 -0400 | [diff] [blame] | 1962 | list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node) |
Tejun Heo | f2b6512 | 2015-05-22 17:13:55 -0400 | [diff] [blame] | 1963 | wb_start_writeback(wb, wb_split_bdi_pages(wb, nr_pages), |
| 1964 | false, reason); |
| 1965 | } |
Christoph Hellwig | b8c2f34 | 2010-06-08 18:15:07 +0200 | [diff] [blame] | 1966 | rcu_read_unlock(); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 1967 | } |
| 1968 | |
Theodore Ts'o | a2f4870 | 2015-03-17 12:23:19 -0400 | [diff] [blame] | 1969 | /* |
| 1970 | * Wake up bdi's periodically to make sure dirtytime inodes gets |
| 1971 | * written back periodically. We deliberately do *not* check the |
| 1972 | * b_dirtytime list in wb_has_dirty_io(), since this would cause the |
| 1973 | * kernel to be constantly waking up once there are any dirtytime |
| 1974 | * inodes on the system. So instead we define a separate delayed work |
| 1975 | * function which gets called much more rarely. (By default, only |
| 1976 | * once every 12 hours.) |
| 1977 | * |
| 1978 | * If there is any other write activity going on in the file system, |
| 1979 | * this function won't be necessary. But if the only thing that has |
| 1980 | * happened on the file system is a dirtytime inode caused by an atime |
| 1981 | * update, we need this infrastructure below to make sure that inode |
| 1982 | * eventually gets pushed out to disk. |
| 1983 | */ |
| 1984 | static void wakeup_dirtytime_writeback(struct work_struct *w); |
| 1985 | static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); |
| 1986 | |
| 1987 | static void wakeup_dirtytime_writeback(struct work_struct *w) |
| 1988 | { |
| 1989 | struct backing_dev_info *bdi; |
| 1990 | |
| 1991 | rcu_read_lock(); |
| 1992 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
Tejun Heo | 001fe6f | 2015-05-22 17:13:56 -0400 | [diff] [blame] | 1993 | struct bdi_writeback *wb; |
Tejun Heo | 001fe6f | 2015-05-22 17:13:56 -0400 | [diff] [blame] | 1994 | |
Tejun Heo | b817525 | 2015-10-02 14:47:05 -0400 | [diff] [blame] | 1995 | list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node) |
Tejun Heo | 6fdf860 | 2015-09-29 12:47:51 -0400 | [diff] [blame] | 1996 | if (!list_empty(&wb->b_dirty_time)) |
| 1997 | wb_wakeup(wb); |
Theodore Ts'o | a2f4870 | 2015-03-17 12:23:19 -0400 | [diff] [blame] | 1998 | } |
| 1999 | rcu_read_unlock(); |
| 2000 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); |
| 2001 | } |
| 2002 | |
| 2003 | static int __init start_dirtytime_writeback(void) |
| 2004 | { |
| 2005 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); |
| 2006 | return 0; |
| 2007 | } |
| 2008 | __initcall(start_dirtytime_writeback); |
| 2009 | |
Theodore Ts'o | 1efff91 | 2015-03-17 12:23:32 -0400 | [diff] [blame] | 2010 | int dirtytime_interval_handler(struct ctl_table *table, int write, |
| 2011 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 2012 | { |
| 2013 | int ret; |
| 2014 | |
| 2015 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
| 2016 | if (ret == 0 && write) |
| 2017 | mod_delayed_work(system_wq, &dirtytime_work, 0); |
| 2018 | return ret; |
| 2019 | } |
| 2020 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2021 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
| 2022 | { |
| 2023 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { |
| 2024 | struct dentry *dentry; |
| 2025 | const char *name = "?"; |
| 2026 | |
| 2027 | dentry = d_find_alias(inode); |
| 2028 | if (dentry) { |
| 2029 | spin_lock(&dentry->d_lock); |
| 2030 | name = (const char *) dentry->d_name.name; |
| 2031 | } |
| 2032 | printk(KERN_DEBUG |
| 2033 | "%s(%d): dirtied inode %lu (%s) on %s\n", |
| 2034 | current->comm, task_pid_nr(current), inode->i_ino, |
| 2035 | name, inode->i_sb->s_id); |
| 2036 | if (dentry) { |
| 2037 | spin_unlock(&dentry->d_lock); |
| 2038 | dput(dentry); |
| 2039 | } |
| 2040 | } |
| 2041 | } |
| 2042 | |
| 2043 | /** |
| 2044 | * __mark_inode_dirty - internal function |
| 2045 | * @inode: inode to mark |
| 2046 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) |
| 2047 | * Mark an inode as dirty. Callers should use mark_inode_dirty or |
| 2048 | * mark_inode_dirty_sync. |
| 2049 | * |
| 2050 | * Put the inode on the super block's dirty list. |
| 2051 | * |
| 2052 | * CAREFUL! We mark it dirty unconditionally, but move it onto the |
| 2053 | * dirty list only if it is hashed or if it refers to a blockdev. |
| 2054 | * If it was not hashed, it will never be added to the dirty list |
| 2055 | * even if it is later hashed, as it will have been marked dirty already. |
| 2056 | * |
| 2057 | * In short, make sure you hash any inodes _before_ you start marking |
| 2058 | * them dirty. |
| 2059 | * |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2060 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
| 2061 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of |
| 2062 | * the kernel-internal blockdev inode represents the dirtying time of the |
| 2063 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use |
| 2064 | * page->mapping->host, so the page-dirtying time is recorded in the internal |
| 2065 | * blockdev inode. |
| 2066 | */ |
| 2067 | void __mark_inode_dirty(struct inode *inode, int flags) |
| 2068 | { |
Randy Dunlap | dbce03b | 2015-11-09 14:58:00 -0800 | [diff] [blame] | 2069 | #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2070 | struct super_block *sb = inode->i_sb; |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 2071 | int dirtytime; |
| 2072 | |
| 2073 | trace_writeback_mark_inode_dirty(inode, flags); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2074 | |
| 2075 | /* |
| 2076 | * Don't do this for I_DIRTY_PAGES - that doesn't actually |
| 2077 | * dirty the inode itself |
| 2078 | */ |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 2079 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { |
Tejun Heo | 9fb0a7d | 2013-01-11 13:06:37 -0800 | [diff] [blame] | 2080 | trace_writeback_dirty_inode_start(inode, flags); |
| 2081 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2082 | if (sb->s_op->dirty_inode) |
Christoph Hellwig | aa38572 | 2011-05-27 06:53:02 -0400 | [diff] [blame] | 2083 | sb->s_op->dirty_inode(inode, flags); |
Tejun Heo | 9fb0a7d | 2013-01-11 13:06:37 -0800 | [diff] [blame] | 2084 | |
| 2085 | trace_writeback_dirty_inode(inode, flags); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2086 | } |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 2087 | if (flags & I_DIRTY_INODE) |
| 2088 | flags &= ~I_DIRTY_TIME; |
| 2089 | dirtytime = flags & I_DIRTY_TIME; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2090 | |
| 2091 | /* |
Tejun Heo | 9c6ac78 | 2014-10-24 15:38:21 -0400 | [diff] [blame] | 2092 | * Paired with smp_mb() in __writeback_single_inode() for the |
| 2093 | * following lockless i_state test. See there for details. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2094 | */ |
| 2095 | smp_mb(); |
| 2096 | |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 2097 | if (((inode->i_state & flags) == flags) || |
| 2098 | (dirtytime && (inode->i_state & I_DIRTY_INODE))) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2099 | return; |
| 2100 | |
San Mehat | 2dbb855 | 2009-03-21 18:48:54 -0700 | [diff] [blame^] | 2101 | if (unlikely(block_dump > 1)) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2102 | block_dump___mark_inode_dirty(inode); |
| 2103 | |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 2104 | spin_lock(&inode->i_lock); |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 2105 | if (dirtytime && (inode->i_state & I_DIRTY_INODE)) |
| 2106 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2107 | if ((inode->i_state & flags) != flags) { |
| 2108 | const int was_dirty = inode->i_state & I_DIRTY; |
| 2109 | |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 2110 | inode_attach_wb(inode, NULL); |
| 2111 | |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 2112 | if (flags & I_DIRTY_INODE) |
| 2113 | inode->i_state &= ~I_DIRTY_TIME; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2114 | inode->i_state |= flags; |
| 2115 | |
| 2116 | /* |
| 2117 | * If the inode is being synced, just update its dirty state. |
| 2118 | * The unlocker will place the inode on the appropriate |
| 2119 | * superblock list, based upon its state. |
| 2120 | */ |
| 2121 | if (inode->i_state & I_SYNC) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 2122 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2123 | |
| 2124 | /* |
| 2125 | * Only add valid (hashed) inodes to the superblock's |
| 2126 | * dirty list. Add blockdev inodes as well. |
| 2127 | */ |
| 2128 | if (!S_ISBLK(inode->i_mode)) { |
Al Viro | 1d3382cb | 2010-10-23 15:19:20 -0400 | [diff] [blame] | 2129 | if (inode_unhashed(inode)) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 2130 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2131 | } |
Al Viro | a4ffdde | 2010-06-02 17:38:30 -0400 | [diff] [blame] | 2132 | if (inode->i_state & I_FREEING) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 2133 | goto out_unlock_inode; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2134 | |
| 2135 | /* |
| 2136 | * If the inode was already on b_dirty/b_io/b_more_io, don't |
| 2137 | * reposition it (that would break b_dirty time-ordering). |
| 2138 | */ |
| 2139 | if (!was_dirty) { |
Tejun Heo | 87e1d78 | 2015-05-28 14:50:52 -0400 | [diff] [blame] | 2140 | struct bdi_writeback *wb; |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 2141 | struct list_head *dirty_list; |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 2142 | bool wakeup_bdi = false; |
Jens Axboe | 500b067 | 2009-09-09 09:10:25 +0200 | [diff] [blame] | 2143 | |
Tejun Heo | 87e1d78 | 2015-05-28 14:50:52 -0400 | [diff] [blame] | 2144 | wb = locked_inode_to_wb_and_lock_list(inode); |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 2145 | |
Tejun Heo | 0747259 | 2015-05-22 17:14:02 -0400 | [diff] [blame] | 2146 | WARN(bdi_cap_writeback_dirty(wb->bdi) && |
| 2147 | !test_bit(WB_registered, &wb->state), |
| 2148 | "bdi-%s not registered\n", wb->bdi->name); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2149 | |
| 2150 | inode->dirtied_when = jiffies; |
Theodore Ts'o | a2f4870 | 2015-03-17 12:23:19 -0400 | [diff] [blame] | 2151 | if (dirtytime) |
| 2152 | inode->dirtied_time_when = jiffies; |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 2153 | |
Theodore Ts'o | a2f4870 | 2015-03-17 12:23:19 -0400 | [diff] [blame] | 2154 | if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) |
Tejun Heo | 0747259 | 2015-05-22 17:14:02 -0400 | [diff] [blame] | 2155 | dirty_list = &wb->b_dirty; |
Theodore Ts'o | a2f4870 | 2015-03-17 12:23:19 -0400 | [diff] [blame] | 2156 | else |
Tejun Heo | 0747259 | 2015-05-22 17:14:02 -0400 | [diff] [blame] | 2157 | dirty_list = &wb->b_dirty_time; |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 2158 | |
Dave Chinner | c7f5408 | 2015-03-04 14:07:22 -0500 | [diff] [blame] | 2159 | wakeup_bdi = inode_io_list_move_locked(inode, wb, |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 2160 | dirty_list); |
| 2161 | |
Tejun Heo | 0747259 | 2015-05-22 17:14:02 -0400 | [diff] [blame] | 2162 | spin_unlock(&wb->list_lock); |
Theodore Ts'o | 0ae45f6 | 2015-02-02 00:37:00 -0500 | [diff] [blame] | 2163 | trace_writeback_dirty_inode_enqueue(inode); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 2164 | |
Tejun Heo | d6c10f1 | 2015-05-22 17:13:45 -0400 | [diff] [blame] | 2165 | /* |
| 2166 | * If this is the first dirty inode for this bdi, |
| 2167 | * we have to wake-up the corresponding bdi thread |
| 2168 | * to make sure background write-back happens |
| 2169 | * later. |
| 2170 | */ |
Tejun Heo | 0747259 | 2015-05-22 17:14:02 -0400 | [diff] [blame] | 2171 | if (bdi_cap_writeback_dirty(wb->bdi) && wakeup_bdi) |
| 2172 | wb_wakeup_delayed(wb); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 2173 | return; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2174 | } |
| 2175 | } |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 2176 | out_unlock_inode: |
| 2177 | spin_unlock(&inode->i_lock); |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 2178 | |
Randy Dunlap | dbce03b | 2015-11-09 14:58:00 -0800 | [diff] [blame] | 2179 | #undef I_DIRTY_INODE |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2180 | } |
| 2181 | EXPORT_SYMBOL(__mark_inode_dirty); |
| 2182 | |
Dave Chinner | e97fedb | 2015-03-04 13:40:00 -0500 | [diff] [blame] | 2183 | /* |
| 2184 | * The @s_sync_lock is used to serialise concurrent sync operations |
| 2185 | * to avoid lock contention problems with concurrent wait_sb_inodes() calls. |
| 2186 | * Concurrent callers will block on the s_sync_lock rather than doing contending |
| 2187 | * walks. The queueing maintains sync(2) required behaviour as all the IO that |
| 2188 | * has been issued up to the time this function is enter is guaranteed to be |
| 2189 | * completed by the time we have gained the lock and waited for all IO that is |
| 2190 | * in progress regardless of the order callers are granted the lock. |
| 2191 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 2192 | static void wait_sb_inodes(struct super_block *sb) |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 2193 | { |
Dave Chinner | 6c60d2b | 2016-07-26 15:21:50 -0700 | [diff] [blame] | 2194 | LIST_HEAD(sync_list); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 2195 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2196 | /* |
| 2197 | * We need to be protected against the filesystem going from |
| 2198 | * r/o to r/w or vice versa. |
| 2199 | */ |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 2200 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 2201 | |
Dave Chinner | e97fedb | 2015-03-04 13:40:00 -0500 | [diff] [blame] | 2202 | mutex_lock(&sb->s_sync_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2203 | |
| 2204 | /* |
Dave Chinner | 6c60d2b | 2016-07-26 15:21:50 -0700 | [diff] [blame] | 2205 | * Splice the writeback list onto a temporary list to avoid waiting on |
| 2206 | * inodes that have started writeback after this point. |
| 2207 | * |
| 2208 | * Use rcu_read_lock() to keep the inodes around until we have a |
| 2209 | * reference. s_inode_wblist_lock protects sb->s_inodes_wb as well as |
| 2210 | * the local list because inodes can be dropped from either by writeback |
| 2211 | * completion. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2212 | */ |
Dave Chinner | 6c60d2b | 2016-07-26 15:21:50 -0700 | [diff] [blame] | 2213 | rcu_read_lock(); |
| 2214 | spin_lock_irq(&sb->s_inode_wblist_lock); |
| 2215 | list_splice_init(&sb->s_inodes_wb, &sync_list); |
| 2216 | |
| 2217 | /* |
| 2218 | * Data integrity sync. Must wait for all pages under writeback, because |
| 2219 | * there may have been pages dirtied before our sync call, but which had |
| 2220 | * writeout started before we write it out. In which case, the inode |
| 2221 | * may not be on the dirty list, but we still have to wait for that |
| 2222 | * writeout. |
| 2223 | */ |
| 2224 | while (!list_empty(&sync_list)) { |
| 2225 | struct inode *inode = list_first_entry(&sync_list, struct inode, |
| 2226 | i_wb_list); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 2227 | struct address_space *mapping = inode->i_mapping; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2228 | |
Dave Chinner | 6c60d2b | 2016-07-26 15:21:50 -0700 | [diff] [blame] | 2229 | /* |
| 2230 | * Move each inode back to the wb list before we drop the lock |
| 2231 | * to preserve consistency between i_wb_list and the mapping |
| 2232 | * writeback tag. Writeback completion is responsible to remove |
| 2233 | * the inode from either list once the writeback tag is cleared. |
| 2234 | */ |
| 2235 | list_move_tail(&inode->i_wb_list, &sb->s_inodes_wb); |
| 2236 | |
| 2237 | /* |
| 2238 | * The mapping can appear untagged while still on-list since we |
| 2239 | * do not have the mapping lock. Skip it here, wb completion |
| 2240 | * will remove it. |
| 2241 | */ |
| 2242 | if (!mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK)) |
| 2243 | continue; |
| 2244 | |
| 2245 | spin_unlock_irq(&sb->s_inode_wblist_lock); |
| 2246 | |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 2247 | spin_lock(&inode->i_lock); |
Dave Chinner | 6c60d2b | 2016-07-26 15:21:50 -0700 | [diff] [blame] | 2248 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) { |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 2249 | spin_unlock(&inode->i_lock); |
Dave Chinner | 6c60d2b | 2016-07-26 15:21:50 -0700 | [diff] [blame] | 2250 | |
| 2251 | spin_lock_irq(&sb->s_inode_wblist_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2252 | continue; |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 2253 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2254 | __iget(inode); |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 2255 | spin_unlock(&inode->i_lock); |
Dave Chinner | 6c60d2b | 2016-07-26 15:21:50 -0700 | [diff] [blame] | 2256 | rcu_read_unlock(); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2257 | |
Junichi Nomura | aa750fd | 2015-11-05 18:47:23 -0800 | [diff] [blame] | 2258 | /* |
| 2259 | * We keep the error status of individual mapping so that |
| 2260 | * applications can catch the writeback error using fsync(2). |
| 2261 | * See filemap_fdatawait_keep_errors() for details. |
| 2262 | */ |
| 2263 | filemap_fdatawait_keep_errors(mapping); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 2264 | |
| 2265 | cond_resched(); |
Nick Piggin | 38f2197 | 2009-01-06 14:40:25 -0800 | [diff] [blame] | 2266 | |
Dave Chinner | 6c60d2b | 2016-07-26 15:21:50 -0700 | [diff] [blame] | 2267 | iput(inode); |
| 2268 | |
| 2269 | rcu_read_lock(); |
| 2270 | spin_lock_irq(&sb->s_inode_wblist_lock); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 2271 | } |
Dave Chinner | 6c60d2b | 2016-07-26 15:21:50 -0700 | [diff] [blame] | 2272 | spin_unlock_irq(&sb->s_inode_wblist_lock); |
| 2273 | rcu_read_unlock(); |
Dave Chinner | e97fedb | 2015-03-04 13:40:00 -0500 | [diff] [blame] | 2274 | mutex_unlock(&sb->s_sync_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2275 | } |
| 2276 | |
Tejun Heo | f30a7d0 | 2015-05-22 17:14:00 -0400 | [diff] [blame] | 2277 | static void __writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr, |
| 2278 | enum wb_reason reason, bool skip_if_busy) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2279 | { |
Tejun Heo | cc395d7 | 2015-05-22 17:13:58 -0400 | [diff] [blame] | 2280 | DEFINE_WB_COMPLETION_ONSTACK(done); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 2281 | struct wb_writeback_work work = { |
Wu Fengguang | 6e6938b | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 2282 | .sb = sb, |
| 2283 | .sync_mode = WB_SYNC_NONE, |
| 2284 | .tagged_writepages = 1, |
| 2285 | .done = &done, |
| 2286 | .nr_pages = nr, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 2287 | .reason = reason, |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 2288 | }; |
Tejun Heo | e797291 | 2015-05-22 17:13:48 -0400 | [diff] [blame] | 2289 | struct backing_dev_info *bdi = sb->s_bdi; |
Jens Axboe | 0e3c9a2 | 2010-06-01 11:08:43 +0200 | [diff] [blame] | 2290 | |
Tejun Heo | e797291 | 2015-05-22 17:13:48 -0400 | [diff] [blame] | 2291 | if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) |
Jan Kara | 6eedc70 | 2012-07-03 16:45:27 +0200 | [diff] [blame] | 2292 | return; |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 2293 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
Tejun Heo | f30a7d0 | 2015-05-22 17:14:00 -0400 | [diff] [blame] | 2294 | |
Tejun Heo | db12536 | 2015-05-22 17:14:01 -0400 | [diff] [blame] | 2295 | bdi_split_work_to_wbs(sb->s_bdi, &work, skip_if_busy); |
Tejun Heo | cc395d7 | 2015-05-22 17:13:58 -0400 | [diff] [blame] | 2296 | wb_wait_for_completion(bdi, &done); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2297 | } |
Tejun Heo | f30a7d0 | 2015-05-22 17:14:00 -0400 | [diff] [blame] | 2298 | |
| 2299 | /** |
| 2300 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
| 2301 | * @sb: the superblock |
| 2302 | * @nr: the number of pages to write |
| 2303 | * @reason: reason why some writeback work initiated |
| 2304 | * |
| 2305 | * Start writeback on some inodes on this super_block. No guarantees are made |
| 2306 | * on how many (if any) will be written, and this function does not wait |
| 2307 | * for IO completion of submitted IO. |
| 2308 | */ |
| 2309 | void writeback_inodes_sb_nr(struct super_block *sb, |
| 2310 | unsigned long nr, |
| 2311 | enum wb_reason reason) |
| 2312 | { |
| 2313 | __writeback_inodes_sb_nr(sb, nr, reason, false); |
| 2314 | } |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 2315 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
| 2316 | |
| 2317 | /** |
| 2318 | * writeback_inodes_sb - writeback dirty inodes from given super_block |
| 2319 | * @sb: the superblock |
Marcos Paulo de Souza | 786228a | 2011-11-23 20:56:45 +0800 | [diff] [blame] | 2320 | * @reason: reason why some writeback work was initiated |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 2321 | * |
| 2322 | * Start writeback on some inodes on this super_block. No guarantees are made |
| 2323 | * on how many (if any) will be written, and this function does not wait |
| 2324 | * for IO completion of submitted IO. |
| 2325 | */ |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 2326 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 2327 | { |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 2328 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 2329 | } |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 2330 | EXPORT_SYMBOL(writeback_inodes_sb); |
| 2331 | |
| 2332 | /** |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 2333 | * try_to_writeback_inodes_sb_nr - try to start writeback if none underway |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 2334 | * @sb: the superblock |
| 2335 | * @nr: the number of pages to write |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 2336 | * @reason: the reason of writeback |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 2337 | * |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 2338 | * Invoke writeback_inodes_sb_nr if no writeback is currently underway. |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 2339 | * Returns 1 if writeback was started, 0 if not. |
| 2340 | */ |
Tejun Heo | f30a7d0 | 2015-05-22 17:14:00 -0400 | [diff] [blame] | 2341 | bool try_to_writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr, |
| 2342 | enum wb_reason reason) |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 2343 | { |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 2344 | if (!down_read_trylock(&sb->s_umount)) |
Tejun Heo | f30a7d0 | 2015-05-22 17:14:00 -0400 | [diff] [blame] | 2345 | return false; |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 2346 | |
Tejun Heo | f30a7d0 | 2015-05-22 17:14:00 -0400 | [diff] [blame] | 2347 | __writeback_inodes_sb_nr(sb, nr, reason, true); |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 2348 | up_read(&sb->s_umount); |
Tejun Heo | f30a7d0 | 2015-05-22 17:14:00 -0400 | [diff] [blame] | 2349 | return true; |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 2350 | } |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 2351 | EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); |
| 2352 | |
| 2353 | /** |
| 2354 | * try_to_writeback_inodes_sb - try to start writeback if none underway |
| 2355 | * @sb: the superblock |
| 2356 | * @reason: reason why some writeback work was initiated |
| 2357 | * |
| 2358 | * Implement by try_to_writeback_inodes_sb_nr() |
| 2359 | * Returns 1 if writeback was started, 0 if not. |
| 2360 | */ |
Tejun Heo | f30a7d0 | 2015-05-22 17:14:00 -0400 | [diff] [blame] | 2361 | bool try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 2362 | { |
| 2363 | return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
| 2364 | } |
| 2365 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); |
Chris Mason | 3259f8b | 2010-10-29 11:16:17 -0400 | [diff] [blame] | 2366 | |
| 2367 | /** |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 2368 | * sync_inodes_sb - sync sb inode pages |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 2369 | * @sb: the superblock |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 2370 | * |
| 2371 | * This function writes and waits on any dirty inode belonging to this |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 2372 | * super_block. |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 2373 | */ |
Jan Kara | 0dc83bd | 2014-02-21 11:19:04 +0100 | [diff] [blame] | 2374 | void sync_inodes_sb(struct super_block *sb) |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 2375 | { |
Tejun Heo | cc395d7 | 2015-05-22 17:13:58 -0400 | [diff] [blame] | 2376 | DEFINE_WB_COMPLETION_ONSTACK(done); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 2377 | struct wb_writeback_work work = { |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 2378 | .sb = sb, |
| 2379 | .sync_mode = WB_SYNC_ALL, |
| 2380 | .nr_pages = LONG_MAX, |
| 2381 | .range_cyclic = 0, |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 2382 | .done = &done, |
Curt Wohlgemuth | 0e175a1 | 2011-10-07 21:54:10 -0600 | [diff] [blame] | 2383 | .reason = WB_REASON_SYNC, |
Dave Chinner | 7747bd4 | 2013-07-02 22:38:35 +1000 | [diff] [blame] | 2384 | .for_sync = 1, |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 2385 | }; |
Tejun Heo | e797291 | 2015-05-22 17:13:48 -0400 | [diff] [blame] | 2386 | struct backing_dev_info *bdi = sb->s_bdi; |
Christoph Hellwig | 3c4d716 | 2010-06-08 18:14:43 +0200 | [diff] [blame] | 2387 | |
Tejun Heo | 006a097 | 2015-08-25 14:11:52 -0400 | [diff] [blame] | 2388 | /* |
| 2389 | * Can't skip on !bdi_has_dirty() because we should wait for !dirty |
| 2390 | * inodes under writeback and I_DIRTY_TIME inodes ignored by |
| 2391 | * bdi_has_dirty() need to be written out too. |
| 2392 | */ |
| 2393 | if (bdi == &noop_backing_dev_info) |
Jan Kara | 6eedc70 | 2012-07-03 16:45:27 +0200 | [diff] [blame] | 2394 | return; |
Christoph Hellwig | cf37e97 | 2010-06-08 18:14:51 +0200 | [diff] [blame] | 2395 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
| 2396 | |
Tejun Heo | db12536 | 2015-05-22 17:14:01 -0400 | [diff] [blame] | 2397 | bdi_split_work_to_wbs(bdi, &work, false); |
Tejun Heo | cc395d7 | 2015-05-22 17:13:58 -0400 | [diff] [blame] | 2398 | wb_wait_for_completion(bdi, &done); |
Christoph Hellwig | 83ba7b0 | 2010-07-06 08:59:53 +0200 | [diff] [blame] | 2399 | |
Jens Axboe | b6e5131 | 2009-09-16 15:13:54 +0200 | [diff] [blame] | 2400 | wait_sb_inodes(sb); |
Jens Axboe | d8a8559 | 2009-09-02 12:34:32 +0200 | [diff] [blame] | 2401 | } |
| 2402 | EXPORT_SYMBOL(sync_inodes_sb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2403 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2404 | /** |
Andrea Arcangeli | 7f04c26 | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 2405 | * write_inode_now - write an inode to disk |
| 2406 | * @inode: inode to write to disk |
| 2407 | * @sync: whether the write should be synchronous or not |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2408 | * |
Andrea Arcangeli | 7f04c26 | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 2409 | * This function commits an inode to disk immediately if it is dirty. This is |
| 2410 | * primarily needed by knfsd. |
| 2411 | * |
| 2412 | * 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] | 2413 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2414 | int write_inode_now(struct inode *inode, int sync) |
| 2415 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2416 | struct writeback_control wbc = { |
| 2417 | .nr_to_write = LONG_MAX, |
Mike Galbraith | 18914b1 | 2008-02-08 04:20:23 -0800 | [diff] [blame] | 2418 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 2419 | .range_start = 0, |
| 2420 | .range_end = LLONG_MAX, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2421 | }; |
| 2422 | |
| 2423 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) |
Andrew Morton | 49364ce | 2005-11-07 00:59:15 -0800 | [diff] [blame] | 2424 | wbc.nr_to_write = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2425 | |
| 2426 | might_sleep(); |
Tejun Heo | aaf2559 | 2016-03-18 13:52:04 -0400 | [diff] [blame] | 2427 | return writeback_single_inode(inode, &wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2428 | } |
| 2429 | EXPORT_SYMBOL(write_inode_now); |
| 2430 | |
| 2431 | /** |
| 2432 | * sync_inode - write an inode and its pages to disk. |
| 2433 | * @inode: the inode to sync |
| 2434 | * @wbc: controls the writeback mode |
| 2435 | * |
| 2436 | * sync_inode() will write an inode and its pages to disk. It will also |
| 2437 | * correctly update the inode on its superblock's dirty inode lists and will |
| 2438 | * update inode->i_state. |
| 2439 | * |
| 2440 | * The caller must have a ref on the inode. |
| 2441 | */ |
| 2442 | int sync_inode(struct inode *inode, struct writeback_control *wbc) |
| 2443 | { |
Tejun Heo | aaf2559 | 2016-03-18 13:52:04 -0400 | [diff] [blame] | 2444 | return writeback_single_inode(inode, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2445 | } |
| 2446 | EXPORT_SYMBOL(sync_inode); |
Christoph Hellwig | c3765016 | 2010-10-06 10:48:20 +0200 | [diff] [blame] | 2447 | |
| 2448 | /** |
Andrew Morton | c691b9d | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 2449 | * sync_inode_metadata - write an inode to disk |
Christoph Hellwig | c3765016 | 2010-10-06 10:48:20 +0200 | [diff] [blame] | 2450 | * @inode: the inode to sync |
| 2451 | * @wait: wait for I/O to complete. |
| 2452 | * |
Andrew Morton | c691b9d | 2011-01-13 15:45:48 -0800 | [diff] [blame] | 2453 | * Write an inode to disk and adjust its dirty state after completion. |
Christoph Hellwig | c3765016 | 2010-10-06 10:48:20 +0200 | [diff] [blame] | 2454 | * |
| 2455 | * Note: only writes the actual inode, no associated data or other metadata. |
| 2456 | */ |
| 2457 | int sync_inode_metadata(struct inode *inode, int wait) |
| 2458 | { |
| 2459 | struct writeback_control wbc = { |
| 2460 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, |
| 2461 | .nr_to_write = 0, /* metadata-only */ |
| 2462 | }; |
| 2463 | |
| 2464 | return sync_inode(inode, &wbc); |
| 2465 | } |
| 2466 | EXPORT_SYMBOL(sync_inode_metadata); |