Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Uwe Zeisberger | f30c226 | 2006-10-03 23:01:26 +0200 | [diff] [blame] | 2 | * mm/page-writeback.c |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3 | * |
| 4 | * Copyright (C) 2002, Linus Torvalds. |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 5 | * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6 | * |
| 7 | * Contains functions related to writing back dirty pages at the |
| 8 | * address_space level. |
| 9 | * |
Francois Cami | e1f8e87 | 2008-10-15 22:01:59 -0700 | [diff] [blame] | 10 | * 10Apr2002 Andrew Morton |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 11 | * Initial version |
| 12 | */ |
| 13 | |
| 14 | #include <linux/kernel.h> |
| 15 | #include <linux/module.h> |
| 16 | #include <linux/spinlock.h> |
| 17 | #include <linux/fs.h> |
| 18 | #include <linux/mm.h> |
| 19 | #include <linux/swap.h> |
| 20 | #include <linux/slab.h> |
| 21 | #include <linux/pagemap.h> |
| 22 | #include <linux/writeback.h> |
| 23 | #include <linux/init.h> |
| 24 | #include <linux/backing-dev.h> |
Andrew Morton | 55e829a | 2006-12-10 02:19:27 -0800 | [diff] [blame] | 25 | #include <linux/task_io_accounting_ops.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | #include <linux/blkdev.h> |
| 27 | #include <linux/mpage.h> |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 28 | #include <linux/rmap.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | #include <linux/percpu.h> |
| 30 | #include <linux/notifier.h> |
| 31 | #include <linux/smp.h> |
| 32 | #include <linux/sysctl.h> |
| 33 | #include <linux/cpu.h> |
| 34 | #include <linux/syscalls.h> |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 35 | #include <linux/buffer_head.h> |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 36 | #include <linux/pagevec.h> |
Dave Chinner | 028c2dd | 2010-07-07 13:24:07 +1000 | [diff] [blame] | 37 | #include <trace/events/writeback.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 38 | |
| 39 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 40 | * After a CPU has dirtied this many pages, balance_dirty_pages_ratelimited |
| 41 | * will look to see if it needs to force writeback or throttling. |
| 42 | */ |
| 43 | static long ratelimit_pages = 32; |
| 44 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 45 | /* |
| 46 | * When balance_dirty_pages decides that the caller needs to perform some |
| 47 | * non-background writeback, this is how many pages it will attempt to write. |
Wu Fengguang | 3a2e9a5 | 2009-09-23 21:56:00 +0800 | [diff] [blame] | 48 | * It should be somewhat larger than dirtied pages to ensure that reasonably |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 49 | * large amounts of I/O are submitted. |
| 50 | */ |
Wu Fengguang | 3a2e9a5 | 2009-09-23 21:56:00 +0800 | [diff] [blame] | 51 | static inline long sync_writeback_pages(unsigned long dirtied) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 52 | { |
Wu Fengguang | 3a2e9a5 | 2009-09-23 21:56:00 +0800 | [diff] [blame] | 53 | if (dirtied < ratelimit_pages) |
| 54 | dirtied = ratelimit_pages; |
| 55 | |
| 56 | return dirtied + dirtied / 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 57 | } |
| 58 | |
| 59 | /* The following parameters are exported via /proc/sys/vm */ |
| 60 | |
| 61 | /* |
Jens Axboe | 5b0830c | 2009-09-23 19:37:09 +0200 | [diff] [blame] | 62 | * Start background writeback (via writeback threads) at this percentage |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 63 | */ |
Wu Fengguang | 1b5e62b | 2009-03-23 08:57:38 +0800 | [diff] [blame] | 64 | int dirty_background_ratio = 10; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 65 | |
| 66 | /* |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 67 | * dirty_background_bytes starts at 0 (disabled) so that it is a function of |
| 68 | * dirty_background_ratio * the amount of dirtyable memory |
| 69 | */ |
| 70 | unsigned long dirty_background_bytes; |
| 71 | |
| 72 | /* |
Bron Gondwana | 195cf45 | 2008-02-04 22:29:20 -0800 | [diff] [blame] | 73 | * free highmem will not be subtracted from the total free memory |
| 74 | * for calculating free ratios if vm_highmem_is_dirtyable is true |
| 75 | */ |
| 76 | int vm_highmem_is_dirtyable; |
| 77 | |
| 78 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 79 | * The generator of dirty data starts writeback at this percentage |
| 80 | */ |
Wu Fengguang | 1b5e62b | 2009-03-23 08:57:38 +0800 | [diff] [blame] | 81 | int vm_dirty_ratio = 20; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 82 | |
| 83 | /* |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 84 | * vm_dirty_bytes starts at 0 (disabled) so that it is a function of |
| 85 | * vm_dirty_ratio * the amount of dirtyable memory |
| 86 | */ |
| 87 | unsigned long vm_dirty_bytes; |
| 88 | |
| 89 | /* |
Alexey Dobriyan | 704503d | 2009-03-31 15:23:18 -0700 | [diff] [blame] | 90 | * The interval between `kupdate'-style writebacks |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 91 | */ |
Toshiyuki Okajima | 22ef37e | 2009-05-16 22:56:28 -0700 | [diff] [blame] | 92 | unsigned int dirty_writeback_interval = 5 * 100; /* centiseconds */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 93 | |
| 94 | /* |
Alexey Dobriyan | 704503d | 2009-03-31 15:23:18 -0700 | [diff] [blame] | 95 | * The longest time for which data is allowed to remain dirty |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | */ |
Toshiyuki Okajima | 22ef37e | 2009-05-16 22:56:28 -0700 | [diff] [blame] | 97 | unsigned int dirty_expire_interval = 30 * 100; /* centiseconds */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 98 | |
| 99 | /* |
| 100 | * Flag that makes the machine dump writes/reads and block dirtyings. |
| 101 | */ |
| 102 | int block_dump; |
| 103 | |
| 104 | /* |
Bart Samwel | ed5b43f | 2006-03-24 03:15:49 -0800 | [diff] [blame] | 105 | * Flag that puts the machine in "laptop mode". Doubles as a timeout in jiffies: |
| 106 | * a full sync is triggered after this time elapses without any disk activity. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 107 | */ |
| 108 | int laptop_mode; |
| 109 | |
| 110 | EXPORT_SYMBOL(laptop_mode); |
| 111 | |
| 112 | /* End of sysctl-exported parameters */ |
| 113 | |
| 114 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 115 | /* |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 116 | * Scale the writeback cache size proportional to the relative writeout speeds. |
| 117 | * |
| 118 | * We do this by keeping a floating proportion between BDIs, based on page |
| 119 | * writeback completions [end_page_writeback()]. Those devices that write out |
| 120 | * pages fastest will get the larger share, while the slower will get a smaller |
| 121 | * share. |
| 122 | * |
| 123 | * We use page writeout completions because we are interested in getting rid of |
| 124 | * dirty pages. Having them written out is the primary goal. |
| 125 | * |
| 126 | * We introduce a concept of time, a period over which we measure these events, |
| 127 | * because demand can/will vary over time. The length of this period itself is |
| 128 | * measured in page writeback completions. |
| 129 | * |
| 130 | */ |
| 131 | static struct prop_descriptor vm_completions; |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 132 | static struct prop_descriptor vm_dirties; |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 133 | |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 134 | /* |
| 135 | * couple the period to the dirty_ratio: |
| 136 | * |
| 137 | * period/2 ~ roundup_pow_of_two(dirty limit) |
| 138 | */ |
| 139 | static int calc_period_shift(void) |
| 140 | { |
| 141 | unsigned long dirty_total; |
| 142 | |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 143 | if (vm_dirty_bytes) |
| 144 | dirty_total = vm_dirty_bytes / PAGE_SIZE; |
| 145 | else |
| 146 | dirty_total = (vm_dirty_ratio * determine_dirtyable_memory()) / |
| 147 | 100; |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 148 | return 2 + ilog2(dirty_total - 1); |
| 149 | } |
| 150 | |
| 151 | /* |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 152 | * update the period when the dirty threshold changes. |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 153 | */ |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 154 | static void update_completion_period(void) |
| 155 | { |
| 156 | int shift = calc_period_shift(); |
| 157 | prop_change_shift(&vm_completions, shift); |
| 158 | prop_change_shift(&vm_dirties, shift); |
| 159 | } |
| 160 | |
| 161 | int dirty_background_ratio_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 162 | void __user *buffer, size_t *lenp, |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 163 | loff_t *ppos) |
| 164 | { |
| 165 | int ret; |
| 166 | |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 167 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 168 | if (ret == 0 && write) |
| 169 | dirty_background_bytes = 0; |
| 170 | return ret; |
| 171 | } |
| 172 | |
| 173 | int dirty_background_bytes_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 174 | void __user *buffer, size_t *lenp, |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 175 | loff_t *ppos) |
| 176 | { |
| 177 | int ret; |
| 178 | |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 179 | ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 180 | if (ret == 0 && write) |
| 181 | dirty_background_ratio = 0; |
| 182 | return ret; |
| 183 | } |
| 184 | |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 185 | int dirty_ratio_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 186 | void __user *buffer, size_t *lenp, |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 187 | loff_t *ppos) |
| 188 | { |
| 189 | int old_ratio = vm_dirty_ratio; |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 190 | int ret; |
| 191 | |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 192 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 193 | if (ret == 0 && write && vm_dirty_ratio != old_ratio) { |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 194 | update_completion_period(); |
| 195 | vm_dirty_bytes = 0; |
| 196 | } |
| 197 | return ret; |
| 198 | } |
| 199 | |
| 200 | |
| 201 | int dirty_bytes_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 202 | void __user *buffer, size_t *lenp, |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 203 | loff_t *ppos) |
| 204 | { |
Sven Wegener | fc3501d | 2009-02-11 13:04:23 -0800 | [diff] [blame] | 205 | unsigned long old_bytes = vm_dirty_bytes; |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 206 | int ret; |
| 207 | |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 208 | ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 209 | if (ret == 0 && write && vm_dirty_bytes != old_bytes) { |
| 210 | update_completion_period(); |
| 211 | vm_dirty_ratio = 0; |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 212 | } |
| 213 | return ret; |
| 214 | } |
| 215 | |
| 216 | /* |
| 217 | * Increment the BDI's writeout completion count and the global writeout |
| 218 | * completion count. Called from test_clear_page_writeback(). |
| 219 | */ |
| 220 | static inline void __bdi_writeout_inc(struct backing_dev_info *bdi) |
| 221 | { |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 222 | __prop_inc_percpu_max(&vm_completions, &bdi->completions, |
| 223 | bdi->max_prop_frac); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 224 | } |
| 225 | |
Miklos Szeredi | dd5656e | 2008-04-30 00:54:37 -0700 | [diff] [blame] | 226 | void bdi_writeout_inc(struct backing_dev_info *bdi) |
| 227 | { |
| 228 | unsigned long flags; |
| 229 | |
| 230 | local_irq_save(flags); |
| 231 | __bdi_writeout_inc(bdi); |
| 232 | local_irq_restore(flags); |
| 233 | } |
| 234 | EXPORT_SYMBOL_GPL(bdi_writeout_inc); |
| 235 | |
Nick Piggin | 1cf6e7d | 2009-02-18 14:48:18 -0800 | [diff] [blame] | 236 | void task_dirty_inc(struct task_struct *tsk) |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 237 | { |
| 238 | prop_inc_single(&vm_dirties, &tsk->dirties); |
| 239 | } |
| 240 | |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 241 | /* |
| 242 | * Obtain an accurate fraction of the BDI's portion. |
| 243 | */ |
| 244 | static void bdi_writeout_fraction(struct backing_dev_info *bdi, |
| 245 | long *numerator, long *denominator) |
| 246 | { |
Wu Fengguang | 3efaf0f | 2010-12-16 22:22:00 -0600 | [diff] [blame] | 247 | prop_fraction_percpu(&vm_completions, &bdi->completions, |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 248 | numerator, denominator); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 249 | } |
| 250 | |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 251 | static inline void task_dirties_fraction(struct task_struct *tsk, |
| 252 | long *numerator, long *denominator) |
| 253 | { |
| 254 | prop_fraction_single(&vm_dirties, &tsk->dirties, |
| 255 | numerator, denominator); |
| 256 | } |
| 257 | |
| 258 | /* |
Wu Fengguang | 1babe18 | 2010-08-11 14:17:40 -0700 | [diff] [blame] | 259 | * task_dirty_limit - scale down dirty throttling threshold for one task |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 260 | * |
| 261 | * task specific dirty limit: |
| 262 | * |
| 263 | * dirty -= (dirty/8) * p_{t} |
Wu Fengguang | 1babe18 | 2010-08-11 14:17:40 -0700 | [diff] [blame] | 264 | * |
| 265 | * To protect light/slow dirtying tasks from heavier/fast ones, we start |
| 266 | * throttling individual tasks before reaching the bdi dirty limit. |
| 267 | * Relatively low thresholds will be allocated to heavy dirtiers. So when |
| 268 | * dirty pages grow large, heavy dirtiers will be throttled first, which will |
| 269 | * effectively curb the growth of dirty pages. Light dirtiers with high enough |
| 270 | * dirty threshold may never get throttled. |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 271 | */ |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 272 | static unsigned long task_dirty_limit(struct task_struct *tsk, |
| 273 | unsigned long bdi_dirty) |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 274 | { |
| 275 | long numerator, denominator; |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 276 | unsigned long dirty = bdi_dirty; |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 277 | u64 inv = dirty >> 3; |
| 278 | |
| 279 | task_dirties_fraction(tsk, &numerator, &denominator); |
| 280 | inv *= numerator; |
| 281 | do_div(inv, denominator); |
| 282 | |
| 283 | dirty -= inv; |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 284 | |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 285 | return max(dirty, bdi_dirty/2); |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 286 | } |
| 287 | |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 288 | /* |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 289 | * |
| 290 | */ |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 291 | static unsigned int bdi_min_ratio; |
| 292 | |
| 293 | int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio) |
| 294 | { |
| 295 | int ret = 0; |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 296 | |
Jens Axboe | cfc4ba5 | 2009-09-14 13:12:40 +0200 | [diff] [blame] | 297 | spin_lock_bh(&bdi_lock); |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 298 | if (min_ratio > bdi->max_ratio) { |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 299 | ret = -EINVAL; |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 300 | } else { |
| 301 | min_ratio -= bdi->min_ratio; |
| 302 | if (bdi_min_ratio + min_ratio < 100) { |
| 303 | bdi_min_ratio += min_ratio; |
| 304 | bdi->min_ratio += min_ratio; |
| 305 | } else { |
| 306 | ret = -EINVAL; |
| 307 | } |
| 308 | } |
Jens Axboe | cfc4ba5 | 2009-09-14 13:12:40 +0200 | [diff] [blame] | 309 | spin_unlock_bh(&bdi_lock); |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 310 | |
| 311 | return ret; |
| 312 | } |
| 313 | |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 314 | int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned max_ratio) |
| 315 | { |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 316 | int ret = 0; |
| 317 | |
| 318 | if (max_ratio > 100) |
| 319 | return -EINVAL; |
| 320 | |
Jens Axboe | cfc4ba5 | 2009-09-14 13:12:40 +0200 | [diff] [blame] | 321 | spin_lock_bh(&bdi_lock); |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 322 | if (bdi->min_ratio > max_ratio) { |
| 323 | ret = -EINVAL; |
| 324 | } else { |
| 325 | bdi->max_ratio = max_ratio; |
| 326 | bdi->max_prop_frac = (PROP_FRAC_BASE * max_ratio) / 100; |
| 327 | } |
Jens Axboe | cfc4ba5 | 2009-09-14 13:12:40 +0200 | [diff] [blame] | 328 | spin_unlock_bh(&bdi_lock); |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 329 | |
| 330 | return ret; |
| 331 | } |
| 332 | EXPORT_SYMBOL(bdi_set_max_ratio); |
| 333 | |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 334 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 335 | * Work out the current dirty-memory clamping and background writeout |
| 336 | * thresholds. |
| 337 | * |
| 338 | * The main aim here is to lower them aggressively if there is a lot of mapped |
| 339 | * memory around. To avoid stressing page reclaim with lots of unreclaimable |
| 340 | * pages. It is better to clamp down on writers than to start swapping, and |
| 341 | * performing lots of scanning. |
| 342 | * |
| 343 | * We only allow 1/2 of the currently-unmapped memory to be dirtied. |
| 344 | * |
| 345 | * We don't permit the clamping level to fall below 5% - that is getting rather |
| 346 | * excessive. |
| 347 | * |
| 348 | * We make sure that the background writeout level is below the adjusted |
| 349 | * clamping level. |
| 350 | */ |
Christoph Lameter | 1b42446 | 2007-05-06 14:48:59 -0700 | [diff] [blame] | 351 | |
| 352 | static unsigned long highmem_dirtyable_memory(unsigned long total) |
| 353 | { |
| 354 | #ifdef CONFIG_HIGHMEM |
| 355 | int node; |
| 356 | unsigned long x = 0; |
| 357 | |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 358 | for_each_node_state(node, N_HIGH_MEMORY) { |
Christoph Lameter | 1b42446 | 2007-05-06 14:48:59 -0700 | [diff] [blame] | 359 | struct zone *z = |
| 360 | &NODE_DATA(node)->node_zones[ZONE_HIGHMEM]; |
| 361 | |
Wu Fengguang | adea02a | 2009-09-21 17:01:42 -0700 | [diff] [blame] | 362 | x += zone_page_state(z, NR_FREE_PAGES) + |
| 363 | zone_reclaimable_pages(z); |
Christoph Lameter | 1b42446 | 2007-05-06 14:48:59 -0700 | [diff] [blame] | 364 | } |
| 365 | /* |
| 366 | * Make sure that the number of highmem pages is never larger |
| 367 | * than the number of the total dirtyable memory. This can only |
| 368 | * occur in very strange VM situations but we want to make sure |
| 369 | * that this does not occur. |
| 370 | */ |
| 371 | return min(x, total); |
| 372 | #else |
| 373 | return 0; |
| 374 | #endif |
| 375 | } |
| 376 | |
Steven Rostedt | 3eefae9 | 2008-05-12 21:21:04 +0200 | [diff] [blame] | 377 | /** |
| 378 | * determine_dirtyable_memory - amount of memory that may be used |
| 379 | * |
| 380 | * Returns the numebr of pages that can currently be freed and used |
| 381 | * by the kernel for direct mappings. |
| 382 | */ |
| 383 | unsigned long determine_dirtyable_memory(void) |
Christoph Lameter | 1b42446 | 2007-05-06 14:48:59 -0700 | [diff] [blame] | 384 | { |
| 385 | unsigned long x; |
| 386 | |
Wu Fengguang | adea02a | 2009-09-21 17:01:42 -0700 | [diff] [blame] | 387 | x = global_page_state(NR_FREE_PAGES) + global_reclaimable_pages(); |
Bron Gondwana | 195cf45 | 2008-02-04 22:29:20 -0800 | [diff] [blame] | 388 | |
| 389 | if (!vm_highmem_is_dirtyable) |
| 390 | x -= highmem_dirtyable_memory(x); |
| 391 | |
Christoph Lameter | 1b42446 | 2007-05-06 14:48:59 -0700 | [diff] [blame] | 392 | return x + 1; /* Ensure that we never return 0 */ |
| 393 | } |
| 394 | |
Randy Dunlap | 03ab450 | 2010-08-14 13:05:17 -0700 | [diff] [blame] | 395 | /* |
Wu Fengguang | 1babe18 | 2010-08-11 14:17:40 -0700 | [diff] [blame] | 396 | * global_dirty_limits - background-writeback and dirty-throttling thresholds |
| 397 | * |
| 398 | * Calculate the dirty thresholds based on sysctl parameters |
| 399 | * - vm.dirty_background_ratio or vm.dirty_background_bytes |
| 400 | * - vm.dirty_ratio or vm.dirty_bytes |
| 401 | * The dirty limits will be lifted by 1/4 for PF_LESS_THROTTLE (ie. nfsd) and |
Minchan Kim | ebd1373 | 2011-01-04 01:36:48 +0900 | [diff] [blame] | 402 | * real-time tasks. |
Wu Fengguang | 1babe18 | 2010-08-11 14:17:40 -0700 | [diff] [blame] | 403 | */ |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 404 | void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 405 | { |
David Rientjes | 364aeb2 | 2009-01-06 14:39:29 -0800 | [diff] [blame] | 406 | unsigned long background; |
| 407 | unsigned long dirty; |
Minchan Kim | 240c879 | 2011-01-13 15:46:27 -0800 | [diff] [blame] | 408 | unsigned long uninitialized_var(available_memory); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 409 | struct task_struct *tsk; |
| 410 | |
Minchan Kim | 240c879 | 2011-01-13 15:46:27 -0800 | [diff] [blame] | 411 | if (!vm_dirty_bytes || !dirty_background_bytes) |
| 412 | available_memory = determine_dirtyable_memory(); |
| 413 | |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 414 | if (vm_dirty_bytes) |
| 415 | dirty = DIV_ROUND_UP(vm_dirty_bytes, PAGE_SIZE); |
Wu Fengguang | 4cbec4c | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 416 | else |
| 417 | dirty = (vm_dirty_ratio * available_memory) / 100; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 418 | |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 419 | if (dirty_background_bytes) |
| 420 | background = DIV_ROUND_UP(dirty_background_bytes, PAGE_SIZE); |
| 421 | else |
| 422 | background = (dirty_background_ratio * available_memory) / 100; |
| 423 | |
| 424 | if (background >= dirty) |
| 425 | background = dirty / 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 426 | tsk = current; |
| 427 | if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) { |
| 428 | background += background / 4; |
| 429 | dirty += dirty / 4; |
| 430 | } |
| 431 | *pbackground = background; |
| 432 | *pdirty = dirty; |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 433 | } |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 434 | |
Wu Fengguang | 6f71865 | 2011-03-02 17:14:34 -0600 | [diff] [blame] | 435 | /** |
Wu Fengguang | 1babe18 | 2010-08-11 14:17:40 -0700 | [diff] [blame] | 436 | * bdi_dirty_limit - @bdi's share of dirty throttling threshold |
Wu Fengguang | 6f71865 | 2011-03-02 17:14:34 -0600 | [diff] [blame] | 437 | * @bdi: the backing_dev_info to query |
| 438 | * @dirty: global dirty limit in pages |
Wu Fengguang | 1babe18 | 2010-08-11 14:17:40 -0700 | [diff] [blame] | 439 | * |
Wu Fengguang | 6f71865 | 2011-03-02 17:14:34 -0600 | [diff] [blame] | 440 | * Returns @bdi's dirty limit in pages. The term "dirty" in the context of |
| 441 | * dirty balancing includes all PG_dirty, PG_writeback and NFS unstable pages. |
| 442 | * And the "limit" in the name is not seriously taken as hard limit in |
| 443 | * balance_dirty_pages(). |
| 444 | * |
| 445 | * It allocates high/low dirty limits to fast/slow devices, in order to prevent |
Wu Fengguang | 1babe18 | 2010-08-11 14:17:40 -0700 | [diff] [blame] | 446 | * - starving fast devices |
| 447 | * - piling up dirty pages (that will take long time to sync) on slow devices |
| 448 | * |
| 449 | * The bdi's share of dirty limit will be adapting to its throughput and |
| 450 | * bounded by the bdi->min_ratio and/or bdi->max_ratio parameters, if set. |
| 451 | */ |
| 452 | unsigned long bdi_dirty_limit(struct backing_dev_info *bdi, unsigned long dirty) |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 453 | { |
| 454 | u64 bdi_dirty; |
| 455 | long numerator, denominator; |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 456 | |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 457 | /* |
| 458 | * Calculate this BDI's share of the dirty ratio. |
| 459 | */ |
| 460 | bdi_writeout_fraction(bdi, &numerator, &denominator); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 461 | |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 462 | bdi_dirty = (dirty * (100 - bdi_min_ratio)) / 100; |
| 463 | bdi_dirty *= numerator; |
| 464 | do_div(bdi_dirty, denominator); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 465 | |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 466 | bdi_dirty += (dirty * bdi->min_ratio) / 100; |
| 467 | if (bdi_dirty > (dirty * bdi->max_ratio) / 100) |
| 468 | bdi_dirty = dirty * bdi->max_ratio / 100; |
| 469 | |
| 470 | return bdi_dirty; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 471 | } |
| 472 | |
| 473 | /* |
| 474 | * balance_dirty_pages() must be called by processes which are generating dirty |
| 475 | * data. It looks at the number of dirty pages in the machine and will force |
| 476 | * the caller to perform writeback if the system is over `vm_dirty_ratio'. |
Jens Axboe | 5b0830c | 2009-09-23 19:37:09 +0200 | [diff] [blame] | 477 | * If we're over `background_thresh' then the writeback threads are woken to |
| 478 | * perform some writeout. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 479 | */ |
Wu Fengguang | 3a2e9a5 | 2009-09-23 21:56:00 +0800 | [diff] [blame] | 480 | static void balance_dirty_pages(struct address_space *mapping, |
| 481 | unsigned long write_chunk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 482 | { |
Peter Zijlstra | 5fce25a | 2007-11-14 16:59:15 -0800 | [diff] [blame] | 483 | long nr_reclaimable, bdi_nr_reclaimable; |
| 484 | long nr_writeback, bdi_nr_writeback; |
David Rientjes | 364aeb2 | 2009-01-06 14:39:29 -0800 | [diff] [blame] | 485 | unsigned long background_thresh; |
| 486 | unsigned long dirty_thresh; |
| 487 | unsigned long bdi_thresh; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 488 | unsigned long pages_written = 0; |
Jens Axboe | 87c6a9b | 2009-09-17 19:59:14 +0200 | [diff] [blame] | 489 | unsigned long pause = 1; |
Wu Fengguang | e50e372 | 2010-08-11 14:17:37 -0700 | [diff] [blame] | 490 | bool dirty_exceeded = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 491 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
| 492 | |
| 493 | for (;;) { |
Peter Zijlstra | 5fce25a | 2007-11-14 16:59:15 -0800 | [diff] [blame] | 494 | nr_reclaimable = global_page_state(NR_FILE_DIRTY) + |
| 495 | global_page_state(NR_UNSTABLE_NFS); |
| 496 | nr_writeback = global_page_state(NR_WRITEBACK); |
| 497 | |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 498 | global_dirty_limits(&background_thresh, &dirty_thresh); |
| 499 | |
| 500 | /* |
| 501 | * Throttle it only when the background writeback cannot |
| 502 | * catch-up. This avoids (excessively) small writeouts |
| 503 | * when the bdi limits are ramping up. |
| 504 | */ |
Wu Fengguang | 4cbec4c | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 505 | if (nr_reclaimable + nr_writeback <= |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 506 | (background_thresh + dirty_thresh) / 2) |
| 507 | break; |
| 508 | |
| 509 | bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh); |
| 510 | bdi_thresh = task_dirty_limit(current, bdi_thresh); |
| 511 | |
Wu Fengguang | e50e372 | 2010-08-11 14:17:37 -0700 | [diff] [blame] | 512 | /* |
| 513 | * In order to avoid the stacked BDI deadlock we need |
| 514 | * to ensure we accurately count the 'dirty' pages when |
| 515 | * the threshold is low. |
| 516 | * |
| 517 | * Otherwise it would be possible to get thresh+n pages |
| 518 | * reported dirty, even though there are thresh-m pages |
| 519 | * actually dirty; with m+n sitting in the percpu |
| 520 | * deltas. |
| 521 | */ |
| 522 | if (bdi_thresh < 2*bdi_stat_error(bdi)) { |
| 523 | bdi_nr_reclaimable = bdi_stat_sum(bdi, BDI_RECLAIMABLE); |
| 524 | bdi_nr_writeback = bdi_stat_sum(bdi, BDI_WRITEBACK); |
| 525 | } else { |
| 526 | bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE); |
| 527 | bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK); |
| 528 | } |
Peter Zijlstra | 5fce25a | 2007-11-14 16:59:15 -0800 | [diff] [blame] | 529 | |
Wu Fengguang | e50e372 | 2010-08-11 14:17:37 -0700 | [diff] [blame] | 530 | /* |
| 531 | * The bdi thresh is somehow "soft" limit derived from the |
| 532 | * global "hard" limit. The former helps to prevent heavy IO |
| 533 | * bdi or process from holding back light ones; The latter is |
| 534 | * the last resort safeguard. |
| 535 | */ |
| 536 | dirty_exceeded = |
Wu Fengguang | 4cbec4c | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 537 | (bdi_nr_reclaimable + bdi_nr_writeback > bdi_thresh) |
| 538 | || (nr_reclaimable + nr_writeback > dirty_thresh); |
Wu Fengguang | e50e372 | 2010-08-11 14:17:37 -0700 | [diff] [blame] | 539 | |
| 540 | if (!dirty_exceeded) |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 541 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 542 | |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 543 | if (!bdi->dirty_exceeded) |
| 544 | bdi->dirty_exceeded = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 545 | |
| 546 | /* Note: nr_reclaimable denotes nr_dirty + nr_unstable. |
| 547 | * Unstable writes are a feature of certain networked |
| 548 | * filesystems (i.e. NFS) in which data may have been |
| 549 | * written to the server's write cache, but has not yet |
| 550 | * been flushed to permanent storage. |
Richard Kennedy | d7831a0 | 2009-06-30 11:41:35 -0700 | [diff] [blame] | 551 | * Only move pages to writeback if this bdi is over its |
| 552 | * threshold otherwise wait until the disk writes catch |
| 553 | * up. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 554 | */ |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame^] | 555 | trace_balance_dirty_start(bdi); |
Richard Kennedy | d7831a0 | 2009-06-30 11:41:35 -0700 | [diff] [blame] | 556 | if (bdi_nr_reclaimable > bdi_thresh) { |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame^] | 557 | pages_written += writeback_inodes_wb(&bdi->wb, |
| 558 | write_chunk); |
| 559 | trace_balance_dirty_written(bdi, pages_written); |
Wu Fengguang | e50e372 | 2010-08-11 14:17:37 -0700 | [diff] [blame] | 560 | if (pages_written >= write_chunk) |
| 561 | break; /* We've done our duty */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 562 | } |
Wu Fengguang | d153ba6 | 2010-12-21 17:24:21 -0800 | [diff] [blame] | 563 | __set_current_state(TASK_UNINTERRUPTIBLE); |
Wu Fengguang | d25105e | 2009-10-09 12:40:42 +0200 | [diff] [blame] | 564 | io_schedule_timeout(pause); |
Wu Fengguang | d46db3d | 2011-05-04 19:54:37 -0600 | [diff] [blame^] | 565 | trace_balance_dirty_wait(bdi); |
Jens Axboe | 87c6a9b | 2009-09-17 19:59:14 +0200 | [diff] [blame] | 566 | |
| 567 | /* |
| 568 | * Increase the delay for each loop, up to our previous |
| 569 | * default of taking a 100ms nap. |
| 570 | */ |
| 571 | pause <<= 1; |
| 572 | if (pause > HZ / 10) |
| 573 | pause = HZ / 10; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 574 | } |
| 575 | |
Wu Fengguang | e50e372 | 2010-08-11 14:17:37 -0700 | [diff] [blame] | 576 | if (!dirty_exceeded && bdi->dirty_exceeded) |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 577 | bdi->dirty_exceeded = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 578 | |
| 579 | if (writeback_in_progress(bdi)) |
Jens Axboe | 5b0830c | 2009-09-23 19:37:09 +0200 | [diff] [blame] | 580 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 581 | |
| 582 | /* |
| 583 | * In laptop mode, we wait until hitting the higher threshold before |
| 584 | * starting background writeout, and then write out all the way down |
| 585 | * to the lower threshold. So slow writers cause minimal disk activity. |
| 586 | * |
| 587 | * In normal mode, we start background writeout at the lower |
| 588 | * background_thresh, to keep the amount of dirty memory low. |
| 589 | */ |
| 590 | if ((laptop_mode && pages_written) || |
Wu Fengguang | e50e372 | 2010-08-11 14:17:37 -0700 | [diff] [blame] | 591 | (!laptop_mode && (nr_reclaimable > background_thresh))) |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 592 | bdi_start_background_writeback(bdi); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 593 | } |
| 594 | |
Peter Zijlstra | a200ee1 | 2007-10-08 18:54:37 +0200 | [diff] [blame] | 595 | void set_page_dirty_balance(struct page *page, int page_mkwrite) |
Peter Zijlstra | edc79b2 | 2006-09-25 23:30:58 -0700 | [diff] [blame] | 596 | { |
Peter Zijlstra | a200ee1 | 2007-10-08 18:54:37 +0200 | [diff] [blame] | 597 | if (set_page_dirty(page) || page_mkwrite) { |
Peter Zijlstra | edc79b2 | 2006-09-25 23:30:58 -0700 | [diff] [blame] | 598 | struct address_space *mapping = page_mapping(page); |
| 599 | |
| 600 | if (mapping) |
| 601 | balance_dirty_pages_ratelimited(mapping); |
| 602 | } |
| 603 | } |
| 604 | |
Tejun Heo | 245b2e7 | 2009-06-24 15:13:48 +0900 | [diff] [blame] | 605 | static DEFINE_PER_CPU(unsigned long, bdp_ratelimits) = 0; |
| 606 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | /** |
Andrew Morton | fa5a734 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 608 | * balance_dirty_pages_ratelimited_nr - balance dirty memory state |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 609 | * @mapping: address_space which was dirtied |
Martin Waitz | a580290 | 2006-04-02 13:59:55 +0200 | [diff] [blame] | 610 | * @nr_pages_dirtied: number of pages which the caller has just dirtied |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 611 | * |
| 612 | * Processes which are dirtying memory should call in here once for each page |
| 613 | * which was newly dirtied. The function will periodically check the system's |
| 614 | * dirty state and will initiate writeback if needed. |
| 615 | * |
| 616 | * On really big machines, get_writeback_state is expensive, so try to avoid |
| 617 | * calling it too often (ratelimiting). But once we're over the dirty memory |
| 618 | * limit we decrease the ratelimiting by a lot, to prevent individual processes |
| 619 | * from overshooting the limit by (ratelimit_pages) each. |
| 620 | */ |
Andrew Morton | fa5a734 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 621 | void balance_dirty_pages_ratelimited_nr(struct address_space *mapping, |
| 622 | unsigned long nr_pages_dirtied) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 623 | { |
Wu Fengguang | 36715ce | 2011-06-11 17:53:57 -0600 | [diff] [blame] | 624 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
Andrew Morton | fa5a734 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 625 | unsigned long ratelimit; |
| 626 | unsigned long *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 627 | |
Wu Fengguang | 36715ce | 2011-06-11 17:53:57 -0600 | [diff] [blame] | 628 | if (!bdi_cap_account_dirty(bdi)) |
| 629 | return; |
| 630 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 631 | ratelimit = ratelimit_pages; |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 632 | if (mapping->backing_dev_info->dirty_exceeded) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 633 | ratelimit = 8; |
| 634 | |
| 635 | /* |
| 636 | * Check the rate limiting. Also, we do not want to throttle real-time |
| 637 | * tasks in balance_dirty_pages(). Period. |
| 638 | */ |
Andrew Morton | fa5a734 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 639 | preempt_disable(); |
Tejun Heo | 245b2e7 | 2009-06-24 15:13:48 +0900 | [diff] [blame] | 640 | p = &__get_cpu_var(bdp_ratelimits); |
Andrew Morton | fa5a734 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 641 | *p += nr_pages_dirtied; |
| 642 | if (unlikely(*p >= ratelimit)) { |
Wu Fengguang | 3a2e9a5 | 2009-09-23 21:56:00 +0800 | [diff] [blame] | 643 | ratelimit = sync_writeback_pages(*p); |
Andrew Morton | fa5a734 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 644 | *p = 0; |
| 645 | preempt_enable(); |
Wu Fengguang | 3a2e9a5 | 2009-09-23 21:56:00 +0800 | [diff] [blame] | 646 | balance_dirty_pages(mapping, ratelimit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 647 | return; |
| 648 | } |
Andrew Morton | fa5a734 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 649 | preempt_enable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 650 | } |
Andrew Morton | fa5a734 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 651 | EXPORT_SYMBOL(balance_dirty_pages_ratelimited_nr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 652 | |
Andrew Morton | 232ea4d | 2007-02-28 20:13:21 -0800 | [diff] [blame] | 653 | void throttle_vm_writeout(gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 654 | { |
David Rientjes | 364aeb2 | 2009-01-06 14:39:29 -0800 | [diff] [blame] | 655 | unsigned long background_thresh; |
| 656 | unsigned long dirty_thresh; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 657 | |
| 658 | for ( ; ; ) { |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 659 | global_dirty_limits(&background_thresh, &dirty_thresh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 660 | |
| 661 | /* |
| 662 | * Boost the allowable dirty threshold a bit for page |
| 663 | * allocators so they don't get DoS'ed by heavy writers |
| 664 | */ |
| 665 | dirty_thresh += dirty_thresh / 10; /* wheeee... */ |
| 666 | |
Christoph Lameter | c24f21b | 2006-06-30 01:55:42 -0700 | [diff] [blame] | 667 | if (global_page_state(NR_UNSTABLE_NFS) + |
| 668 | global_page_state(NR_WRITEBACK) <= dirty_thresh) |
| 669 | break; |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 670 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
Fengguang Wu | 369f238 | 2007-10-16 23:30:45 -0700 | [diff] [blame] | 671 | |
| 672 | /* |
| 673 | * The caller might hold locks which can prevent IO completion |
| 674 | * or progress in the filesystem. So we cannot just sit here |
| 675 | * waiting for IO to complete. |
| 676 | */ |
| 677 | if ((gfp_mask & (__GFP_FS|__GFP_IO)) != (__GFP_FS|__GFP_IO)) |
| 678 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 679 | } |
| 680 | } |
| 681 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 682 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 683 | * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs |
| 684 | */ |
| 685 | int dirty_writeback_centisecs_handler(ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 686 | void __user *buffer, size_t *length, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 687 | { |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 688 | proc_dointvec(table, write, buffer, length, ppos); |
Jens Axboe | 6423104 | 2010-05-21 20:00:35 +0200 | [diff] [blame] | 689 | bdi_arm_supers_timer(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 690 | return 0; |
| 691 | } |
| 692 | |
Jens Axboe | c2c4986 | 2010-05-20 09:18:47 +0200 | [diff] [blame] | 693 | #ifdef CONFIG_BLOCK |
Matthew Garrett | 31373d0 | 2010-04-06 14:25:14 +0200 | [diff] [blame] | 694 | void laptop_mode_timer_fn(unsigned long data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 695 | { |
Matthew Garrett | 31373d0 | 2010-04-06 14:25:14 +0200 | [diff] [blame] | 696 | struct request_queue *q = (struct request_queue *)data; |
| 697 | int nr_pages = global_page_state(NR_FILE_DIRTY) + |
| 698 | global_page_state(NR_UNSTABLE_NFS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 699 | |
Matthew Garrett | 31373d0 | 2010-04-06 14:25:14 +0200 | [diff] [blame] | 700 | /* |
| 701 | * We want to write everything out, not just down to the dirty |
| 702 | * threshold |
| 703 | */ |
Matthew Garrett | 31373d0 | 2010-04-06 14:25:14 +0200 | [diff] [blame] | 704 | if (bdi_has_dirty_io(&q->backing_dev_info)) |
Christoph Hellwig | c544419 | 2010-06-08 18:15:15 +0200 | [diff] [blame] | 705 | bdi_start_writeback(&q->backing_dev_info, nr_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 706 | } |
| 707 | |
| 708 | /* |
| 709 | * We've spun up the disk and we're in laptop mode: schedule writeback |
| 710 | * of all dirty data a few seconds from now. If the flush is already scheduled |
| 711 | * then push it back - the user is still using the disk. |
| 712 | */ |
Matthew Garrett | 31373d0 | 2010-04-06 14:25:14 +0200 | [diff] [blame] | 713 | void laptop_io_completion(struct backing_dev_info *info) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 714 | { |
Matthew Garrett | 31373d0 | 2010-04-06 14:25:14 +0200 | [diff] [blame] | 715 | mod_timer(&info->laptop_mode_wb_timer, jiffies + laptop_mode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 716 | } |
| 717 | |
| 718 | /* |
| 719 | * We're in laptop mode and we've just synced. The sync's writes will have |
| 720 | * caused another writeback to be scheduled by laptop_io_completion. |
| 721 | * Nothing needs to be written back anymore, so we unschedule the writeback. |
| 722 | */ |
| 723 | void laptop_sync_completion(void) |
| 724 | { |
Matthew Garrett | 31373d0 | 2010-04-06 14:25:14 +0200 | [diff] [blame] | 725 | struct backing_dev_info *bdi; |
| 726 | |
| 727 | rcu_read_lock(); |
| 728 | |
| 729 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) |
| 730 | del_timer(&bdi->laptop_mode_wb_timer); |
| 731 | |
| 732 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 733 | } |
Jens Axboe | c2c4986 | 2010-05-20 09:18:47 +0200 | [diff] [blame] | 734 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 735 | |
| 736 | /* |
| 737 | * If ratelimit_pages is too high then we can get into dirty-data overload |
| 738 | * if a large number of processes all perform writes at the same time. |
| 739 | * If it is too low then SMP machines will call the (expensive) |
| 740 | * get_writeback_state too often. |
| 741 | * |
| 742 | * Here we set ratelimit_pages to a level which ensures that when all CPUs are |
| 743 | * dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory |
| 744 | * thresholds before writeback cuts in. |
| 745 | * |
| 746 | * But the limit should not be set too high. Because it also controls the |
| 747 | * amount of memory which the balance_dirty_pages() caller has to write back. |
| 748 | * If this is too large then the caller will block on the IO queue all the |
| 749 | * time. So limit it to four megabytes - the balance_dirty_pages() caller |
| 750 | * will write six megabyte chunks, max. |
| 751 | */ |
| 752 | |
Chandra Seetharaman | 2d1d43f | 2006-09-29 02:01:25 -0700 | [diff] [blame] | 753 | void writeback_set_ratelimit(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 754 | { |
Chandra Seetharaman | 40c99aa | 2006-09-29 02:01:24 -0700 | [diff] [blame] | 755 | ratelimit_pages = vm_total_pages / (num_online_cpus() * 32); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 756 | if (ratelimit_pages < 16) |
| 757 | ratelimit_pages = 16; |
| 758 | if (ratelimit_pages * PAGE_CACHE_SIZE > 4096 * 1024) |
| 759 | ratelimit_pages = (4096 * 1024) / PAGE_CACHE_SIZE; |
| 760 | } |
| 761 | |
Chandra Seetharaman | 26c2143 | 2006-06-27 02:54:10 -0700 | [diff] [blame] | 762 | static int __cpuinit |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 763 | ratelimit_handler(struct notifier_block *self, unsigned long u, void *v) |
| 764 | { |
Chandra Seetharaman | 2d1d43f | 2006-09-29 02:01:25 -0700 | [diff] [blame] | 765 | writeback_set_ratelimit(); |
Paul E. McKenney | aa0f030 | 2007-02-10 01:46:37 -0800 | [diff] [blame] | 766 | return NOTIFY_DONE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 767 | } |
| 768 | |
Chandra Seetharaman | 74b85f3 | 2006-06-27 02:54:09 -0700 | [diff] [blame] | 769 | static struct notifier_block __cpuinitdata ratelimit_nb = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 770 | .notifier_call = ratelimit_handler, |
| 771 | .next = NULL, |
| 772 | }; |
| 773 | |
| 774 | /* |
Linus Torvalds | dc6e29d | 2007-01-29 16:37:38 -0800 | [diff] [blame] | 775 | * Called early on to tune the page writeback dirty limits. |
| 776 | * |
| 777 | * We used to scale dirty pages according to how total memory |
| 778 | * related to pages that could be allocated for buffers (by |
| 779 | * comparing nr_free_buffer_pages() to vm_total_pages. |
| 780 | * |
| 781 | * However, that was when we used "dirty_ratio" to scale with |
| 782 | * all memory, and we don't do that any more. "dirty_ratio" |
| 783 | * is now applied to total non-HIGHPAGE memory (by subtracting |
| 784 | * totalhigh_pages from vm_total_pages), and as such we can't |
| 785 | * get into the old insane situation any more where we had |
| 786 | * large amounts of dirty pages compared to a small amount of |
| 787 | * non-HIGHMEM memory. |
| 788 | * |
| 789 | * But we might still want to scale the dirty_ratio by how |
| 790 | * much memory the box has.. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 791 | */ |
| 792 | void __init page_writeback_init(void) |
| 793 | { |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 794 | int shift; |
| 795 | |
Chandra Seetharaman | 2d1d43f | 2006-09-29 02:01:25 -0700 | [diff] [blame] | 796 | writeback_set_ratelimit(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | register_cpu_notifier(&ratelimit_nb); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 798 | |
| 799 | shift = calc_period_shift(); |
| 800 | prop_descriptor_init(&vm_completions, shift); |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 801 | prop_descriptor_init(&vm_dirties, shift); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 802 | } |
| 803 | |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 804 | /** |
Jan Kara | f446daa | 2010-08-09 17:19:12 -0700 | [diff] [blame] | 805 | * tag_pages_for_writeback - tag pages to be written by write_cache_pages |
| 806 | * @mapping: address space structure to write |
| 807 | * @start: starting page index |
| 808 | * @end: ending page index (inclusive) |
| 809 | * |
| 810 | * This function scans the page range from @start to @end (inclusive) and tags |
| 811 | * all pages that have DIRTY tag set with a special TOWRITE tag. The idea is |
| 812 | * that write_cache_pages (or whoever calls this function) will then use |
| 813 | * TOWRITE tag to identify pages eligible for writeback. This mechanism is |
| 814 | * used to avoid livelocking of writeback by a process steadily creating new |
| 815 | * dirty pages in the file (thus it is important for this function to be quick |
| 816 | * so that it can tag pages faster than a dirtying process can create them). |
| 817 | */ |
| 818 | /* |
| 819 | * We tag pages in batches of WRITEBACK_TAG_BATCH to reduce tree_lock latency. |
| 820 | */ |
Jan Kara | f446daa | 2010-08-09 17:19:12 -0700 | [diff] [blame] | 821 | void tag_pages_for_writeback(struct address_space *mapping, |
| 822 | pgoff_t start, pgoff_t end) |
| 823 | { |
Randy Dunlap | 3c111a0 | 2010-08-11 14:17:30 -0700 | [diff] [blame] | 824 | #define WRITEBACK_TAG_BATCH 4096 |
Jan Kara | f446daa | 2010-08-09 17:19:12 -0700 | [diff] [blame] | 825 | unsigned long tagged; |
| 826 | |
| 827 | do { |
| 828 | spin_lock_irq(&mapping->tree_lock); |
| 829 | tagged = radix_tree_range_tag_if_tagged(&mapping->page_tree, |
| 830 | &start, end, WRITEBACK_TAG_BATCH, |
| 831 | PAGECACHE_TAG_DIRTY, PAGECACHE_TAG_TOWRITE); |
| 832 | spin_unlock_irq(&mapping->tree_lock); |
| 833 | WARN_ON_ONCE(tagged > WRITEBACK_TAG_BATCH); |
| 834 | cond_resched(); |
Jan Kara | d5ed3a4 | 2010-08-19 14:13:33 -0700 | [diff] [blame] | 835 | /* We check 'start' to handle wrapping when end == ~0UL */ |
| 836 | } while (tagged >= WRITEBACK_TAG_BATCH && start); |
Jan Kara | f446daa | 2010-08-09 17:19:12 -0700 | [diff] [blame] | 837 | } |
| 838 | EXPORT_SYMBOL(tag_pages_for_writeback); |
| 839 | |
| 840 | /** |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 841 | * write_cache_pages - walk the list of dirty pages of the given address space and write all of them. |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 842 | * @mapping: address space structure to write |
| 843 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 844 | * @writepage: function called for each page |
| 845 | * @data: data passed to writepage function |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 846 | * |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 847 | * If a page is already under I/O, write_cache_pages() skips it, even |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 848 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, |
| 849 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() |
| 850 | * and msync() need to guarantee that all the data which was dirty at the time |
| 851 | * the call was made get new I/O started against them. If wbc->sync_mode is |
| 852 | * WB_SYNC_ALL then we were called for data integrity and we must wait for |
| 853 | * existing IO to complete. |
Jan Kara | f446daa | 2010-08-09 17:19:12 -0700 | [diff] [blame] | 854 | * |
| 855 | * To avoid livelocks (when other process dirties new pages), we first tag |
| 856 | * pages which should be written back with TOWRITE tag and only then start |
| 857 | * writing them. For data-integrity sync we have to be careful so that we do |
| 858 | * not miss some pages (e.g., because some other process has cleared TOWRITE |
| 859 | * tag we set). The rule we follow is that TOWRITE tag can be cleared only |
| 860 | * by the process clearing the DIRTY tag (and submitting the page for IO). |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 861 | */ |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 862 | int write_cache_pages(struct address_space *mapping, |
| 863 | struct writeback_control *wbc, writepage_t writepage, |
| 864 | void *data) |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 865 | { |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 866 | int ret = 0; |
| 867 | int done = 0; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 868 | struct pagevec pvec; |
| 869 | int nr_pages; |
Nick Piggin | 31a1266 | 2009-01-06 14:39:04 -0800 | [diff] [blame] | 870 | pgoff_t uninitialized_var(writeback_index); |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 871 | pgoff_t index; |
| 872 | pgoff_t end; /* Inclusive */ |
Nick Piggin | bd19e01 | 2009-01-06 14:39:06 -0800 | [diff] [blame] | 873 | pgoff_t done_index; |
Nick Piggin | 31a1266 | 2009-01-06 14:39:04 -0800 | [diff] [blame] | 874 | int cycled; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 875 | int range_whole = 0; |
Jan Kara | f446daa | 2010-08-09 17:19:12 -0700 | [diff] [blame] | 876 | int tag; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 877 | |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 878 | pagevec_init(&pvec, 0); |
| 879 | if (wbc->range_cyclic) { |
Nick Piggin | 31a1266 | 2009-01-06 14:39:04 -0800 | [diff] [blame] | 880 | writeback_index = mapping->writeback_index; /* prev offset */ |
| 881 | index = writeback_index; |
| 882 | if (index == 0) |
| 883 | cycled = 1; |
| 884 | else |
| 885 | cycled = 0; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 886 | end = -1; |
| 887 | } else { |
| 888 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
| 889 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
| 890 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
| 891 | range_whole = 1; |
Nick Piggin | 31a1266 | 2009-01-06 14:39:04 -0800 | [diff] [blame] | 892 | cycled = 1; /* ignore range_cyclic tests */ |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 893 | } |
Wu Fengguang | 6e6938b | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 894 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
Jan Kara | f446daa | 2010-08-09 17:19:12 -0700 | [diff] [blame] | 895 | tag = PAGECACHE_TAG_TOWRITE; |
| 896 | else |
| 897 | tag = PAGECACHE_TAG_DIRTY; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 898 | retry: |
Wu Fengguang | 6e6938b | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 899 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
Jan Kara | f446daa | 2010-08-09 17:19:12 -0700 | [diff] [blame] | 900 | tag_pages_for_writeback(mapping, index, end); |
Nick Piggin | bd19e01 | 2009-01-06 14:39:06 -0800 | [diff] [blame] | 901 | done_index = index; |
Nick Piggin | 5a3d5c9 | 2009-01-06 14:39:09 -0800 | [diff] [blame] | 902 | while (!done && (index <= end)) { |
| 903 | int i; |
| 904 | |
Jan Kara | f446daa | 2010-08-09 17:19:12 -0700 | [diff] [blame] | 905 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
Nick Piggin | 5a3d5c9 | 2009-01-06 14:39:09 -0800 | [diff] [blame] | 906 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
| 907 | if (nr_pages == 0) |
| 908 | break; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 909 | |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 910 | for (i = 0; i < nr_pages; i++) { |
| 911 | struct page *page = pvec.pages[i]; |
| 912 | |
Nick Piggin | d5482cd | 2009-01-06 14:39:11 -0800 | [diff] [blame] | 913 | /* |
| 914 | * At this point, the page may be truncated or |
| 915 | * invalidated (changing page->mapping to NULL), or |
| 916 | * even swizzled back from swapper_space to tmpfs file |
| 917 | * mapping. However, page->index will not change |
| 918 | * because we have a reference on the page. |
| 919 | */ |
| 920 | if (page->index > end) { |
| 921 | /* |
| 922 | * can't be range_cyclic (1st pass) because |
| 923 | * end == -1 in that case. |
| 924 | */ |
| 925 | done = 1; |
| 926 | break; |
| 927 | } |
| 928 | |
Jun'ichi Nomura | cf15b07 | 2011-03-22 16:33:40 -0700 | [diff] [blame] | 929 | done_index = page->index; |
Nick Piggin | bd19e01 | 2009-01-06 14:39:06 -0800 | [diff] [blame] | 930 | |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 931 | lock_page(page); |
| 932 | |
Nick Piggin | 5a3d5c9 | 2009-01-06 14:39:09 -0800 | [diff] [blame] | 933 | /* |
| 934 | * Page truncated or invalidated. We can freely skip it |
| 935 | * then, even for data integrity operations: the page |
| 936 | * has disappeared concurrently, so there could be no |
| 937 | * real expectation of this data interity operation |
| 938 | * even if there is now a new, dirty page at the same |
| 939 | * pagecache address. |
| 940 | */ |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 941 | if (unlikely(page->mapping != mapping)) { |
Nick Piggin | 5a3d5c9 | 2009-01-06 14:39:09 -0800 | [diff] [blame] | 942 | continue_unlock: |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 943 | unlock_page(page); |
| 944 | continue; |
| 945 | } |
| 946 | |
Nick Piggin | 515f4a0 | 2009-01-06 14:39:10 -0800 | [diff] [blame] | 947 | if (!PageDirty(page)) { |
| 948 | /* someone wrote it for us */ |
| 949 | goto continue_unlock; |
| 950 | } |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 951 | |
Nick Piggin | 515f4a0 | 2009-01-06 14:39:10 -0800 | [diff] [blame] | 952 | if (PageWriteback(page)) { |
| 953 | if (wbc->sync_mode != WB_SYNC_NONE) |
| 954 | wait_on_page_writeback(page); |
| 955 | else |
| 956 | goto continue_unlock; |
| 957 | } |
| 958 | |
| 959 | BUG_ON(PageWriteback(page)); |
| 960 | if (!clear_page_dirty_for_io(page)) |
Nick Piggin | 5a3d5c9 | 2009-01-06 14:39:09 -0800 | [diff] [blame] | 961 | goto continue_unlock; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 962 | |
Dave Chinner | 9e09438 | 2010-07-07 13:24:08 +1000 | [diff] [blame] | 963 | trace_wbc_writepage(wbc, mapping->backing_dev_info); |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 964 | ret = (*writepage)(page, wbc, data); |
Nick Piggin | 0026677 | 2009-01-06 14:39:06 -0800 | [diff] [blame] | 965 | if (unlikely(ret)) { |
| 966 | if (ret == AOP_WRITEPAGE_ACTIVATE) { |
| 967 | unlock_page(page); |
| 968 | ret = 0; |
| 969 | } else { |
| 970 | /* |
| 971 | * done_index is set past this page, |
| 972 | * so media errors will not choke |
| 973 | * background writeout for the entire |
| 974 | * file. This has consequences for |
| 975 | * range_cyclic semantics (ie. it may |
| 976 | * not be suitable for data integrity |
| 977 | * writeout). |
| 978 | */ |
Jun'ichi Nomura | cf15b07 | 2011-03-22 16:33:40 -0700 | [diff] [blame] | 979 | done_index = page->index + 1; |
Nick Piggin | 0026677 | 2009-01-06 14:39:06 -0800 | [diff] [blame] | 980 | done = 1; |
| 981 | break; |
| 982 | } |
Dave Chinner | 0b56492 | 2010-06-09 10:37:18 +1000 | [diff] [blame] | 983 | } |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 984 | |
Dave Chinner | 546a192 | 2010-08-24 11:44:34 +1000 | [diff] [blame] | 985 | /* |
| 986 | * We stop writing back only if we are not doing |
| 987 | * integrity sync. In case of integrity sync we have to |
| 988 | * keep going until we have written all the pages |
| 989 | * we tagged for writeback prior to entering this loop. |
| 990 | */ |
| 991 | if (--wbc->nr_to_write <= 0 && |
| 992 | wbc->sync_mode == WB_SYNC_NONE) { |
| 993 | done = 1; |
| 994 | break; |
Nick Piggin | 05fe478 | 2009-01-06 14:39:08 -0800 | [diff] [blame] | 995 | } |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 996 | } |
| 997 | pagevec_release(&pvec); |
| 998 | cond_resched(); |
| 999 | } |
Nick Piggin | 3a4c680 | 2009-02-12 04:34:23 +0100 | [diff] [blame] | 1000 | if (!cycled && !done) { |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 1001 | /* |
Nick Piggin | 31a1266 | 2009-01-06 14:39:04 -0800 | [diff] [blame] | 1002 | * range_cyclic: |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 1003 | * We hit the last page and there is more work to be done: wrap |
| 1004 | * back to the start of the file |
| 1005 | */ |
Nick Piggin | 31a1266 | 2009-01-06 14:39:04 -0800 | [diff] [blame] | 1006 | cycled = 1; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 1007 | index = 0; |
Nick Piggin | 31a1266 | 2009-01-06 14:39:04 -0800 | [diff] [blame] | 1008 | end = writeback_index - 1; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 1009 | goto retry; |
| 1010 | } |
Dave Chinner | 0b56492 | 2010-06-09 10:37:18 +1000 | [diff] [blame] | 1011 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
| 1012 | mapping->writeback_index = done_index; |
Aneesh Kumar K.V | 06d6cf6 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1013 | |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 1014 | return ret; |
| 1015 | } |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 1016 | EXPORT_SYMBOL(write_cache_pages); |
| 1017 | |
| 1018 | /* |
| 1019 | * Function used by generic_writepages to call the real writepage |
| 1020 | * function and set the mapping flags on error |
| 1021 | */ |
| 1022 | static int __writepage(struct page *page, struct writeback_control *wbc, |
| 1023 | void *data) |
| 1024 | { |
| 1025 | struct address_space *mapping = data; |
| 1026 | int ret = mapping->a_ops->writepage(page, wbc); |
| 1027 | mapping_set_error(mapping, ret); |
| 1028 | return ret; |
| 1029 | } |
| 1030 | |
| 1031 | /** |
| 1032 | * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them. |
| 1033 | * @mapping: address space structure to write |
| 1034 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write |
| 1035 | * |
| 1036 | * This is a library function, which implements the writepages() |
| 1037 | * address_space_operation. |
| 1038 | */ |
| 1039 | int generic_writepages(struct address_space *mapping, |
| 1040 | struct writeback_control *wbc) |
| 1041 | { |
Shaohua Li | 9b6096a | 2011-03-17 10:47:06 +0100 | [diff] [blame] | 1042 | struct blk_plug plug; |
| 1043 | int ret; |
| 1044 | |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 1045 | /* deal with chardevs and other special file */ |
| 1046 | if (!mapping->a_ops->writepage) |
| 1047 | return 0; |
| 1048 | |
Shaohua Li | 9b6096a | 2011-03-17 10:47:06 +0100 | [diff] [blame] | 1049 | blk_start_plug(&plug); |
| 1050 | ret = write_cache_pages(mapping, wbc, __writepage, mapping); |
| 1051 | blk_finish_plug(&plug); |
| 1052 | return ret; |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 1053 | } |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 1054 | |
| 1055 | EXPORT_SYMBOL(generic_writepages); |
| 1056 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1057 | int do_writepages(struct address_space *mapping, struct writeback_control *wbc) |
| 1058 | { |
Andrew Morton | 22905f7 | 2005-11-16 15:07:01 -0800 | [diff] [blame] | 1059 | int ret; |
| 1060 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1061 | if (wbc->nr_to_write <= 0) |
| 1062 | return 0; |
| 1063 | if (mapping->a_ops->writepages) |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 1064 | ret = mapping->a_ops->writepages(mapping, wbc); |
Andrew Morton | 22905f7 | 2005-11-16 15:07:01 -0800 | [diff] [blame] | 1065 | else |
| 1066 | ret = generic_writepages(mapping, wbc); |
Andrew Morton | 22905f7 | 2005-11-16 15:07:01 -0800 | [diff] [blame] | 1067 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1068 | } |
| 1069 | |
| 1070 | /** |
| 1071 | * write_one_page - write out a single page and optionally wait on I/O |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 1072 | * @page: the page to write |
| 1073 | * @wait: if true, wait on writeout |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1074 | * |
| 1075 | * The page must be locked by the caller and will be unlocked upon return. |
| 1076 | * |
| 1077 | * write_one_page() returns a negative error code if I/O failed. |
| 1078 | */ |
| 1079 | int write_one_page(struct page *page, int wait) |
| 1080 | { |
| 1081 | struct address_space *mapping = page->mapping; |
| 1082 | int ret = 0; |
| 1083 | struct writeback_control wbc = { |
| 1084 | .sync_mode = WB_SYNC_ALL, |
| 1085 | .nr_to_write = 1, |
| 1086 | }; |
| 1087 | |
| 1088 | BUG_ON(!PageLocked(page)); |
| 1089 | |
| 1090 | if (wait) |
| 1091 | wait_on_page_writeback(page); |
| 1092 | |
| 1093 | if (clear_page_dirty_for_io(page)) { |
| 1094 | page_cache_get(page); |
| 1095 | ret = mapping->a_ops->writepage(page, &wbc); |
| 1096 | if (ret == 0 && wait) { |
| 1097 | wait_on_page_writeback(page); |
| 1098 | if (PageError(page)) |
| 1099 | ret = -EIO; |
| 1100 | } |
| 1101 | page_cache_release(page); |
| 1102 | } else { |
| 1103 | unlock_page(page); |
| 1104 | } |
| 1105 | return ret; |
| 1106 | } |
| 1107 | EXPORT_SYMBOL(write_one_page); |
| 1108 | |
| 1109 | /* |
Ken Chen | 7671932 | 2007-02-10 01:43:15 -0800 | [diff] [blame] | 1110 | * For address_spaces which do not use buffers nor write back. |
| 1111 | */ |
| 1112 | int __set_page_dirty_no_writeback(struct page *page) |
| 1113 | { |
| 1114 | if (!PageDirty(page)) |
Bob Liu | c3f0da6 | 2011-01-13 15:45:49 -0800 | [diff] [blame] | 1115 | return !TestSetPageDirty(page); |
Ken Chen | 7671932 | 2007-02-10 01:43:15 -0800 | [diff] [blame] | 1116 | return 0; |
| 1117 | } |
| 1118 | |
| 1119 | /* |
Edward Shishkin | e3a7cca | 2009-03-31 15:19:39 -0700 | [diff] [blame] | 1120 | * Helper function for set_page_dirty family. |
| 1121 | * NOTE: This relies on being atomic wrt interrupts. |
| 1122 | */ |
| 1123 | void account_page_dirtied(struct page *page, struct address_space *mapping) |
| 1124 | { |
| 1125 | if (mapping_cap_account_dirty(mapping)) { |
| 1126 | __inc_zone_page_state(page, NR_FILE_DIRTY); |
Michael Rubin | ea941f0 | 2010-10-26 14:21:35 -0700 | [diff] [blame] | 1127 | __inc_zone_page_state(page, NR_DIRTIED); |
Edward Shishkin | e3a7cca | 2009-03-31 15:19:39 -0700 | [diff] [blame] | 1128 | __inc_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); |
| 1129 | task_dirty_inc(current); |
| 1130 | task_io_account_write(PAGE_CACHE_SIZE); |
| 1131 | } |
| 1132 | } |
Michael Rubin | 679ceac | 2010-08-20 02:31:26 -0700 | [diff] [blame] | 1133 | EXPORT_SYMBOL(account_page_dirtied); |
Edward Shishkin | e3a7cca | 2009-03-31 15:19:39 -0700 | [diff] [blame] | 1134 | |
| 1135 | /* |
Michael Rubin | f629d1c | 2010-10-26 14:21:33 -0700 | [diff] [blame] | 1136 | * Helper function for set_page_writeback family. |
| 1137 | * NOTE: Unlike account_page_dirtied this does not rely on being atomic |
| 1138 | * wrt interrupts. |
| 1139 | */ |
| 1140 | void account_page_writeback(struct page *page) |
| 1141 | { |
| 1142 | inc_zone_page_state(page, NR_WRITEBACK); |
Michael Rubin | ea941f0 | 2010-10-26 14:21:35 -0700 | [diff] [blame] | 1143 | inc_zone_page_state(page, NR_WRITTEN); |
Michael Rubin | f629d1c | 2010-10-26 14:21:33 -0700 | [diff] [blame] | 1144 | } |
| 1145 | EXPORT_SYMBOL(account_page_writeback); |
| 1146 | |
| 1147 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1148 | * For address_spaces which do not use buffers. Just tag the page as dirty in |
| 1149 | * its radix tree. |
| 1150 | * |
| 1151 | * This is also used when a single buffer is being dirtied: we want to set the |
| 1152 | * page dirty in that case, but not all the buffers. This is a "bottom-up" |
| 1153 | * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying. |
| 1154 | * |
| 1155 | * Most callers have locked the page, which pins the address_space in memory. |
| 1156 | * But zap_pte_range() does not lock the page, however in that case the |
| 1157 | * mapping is pinned by the vma's ->vm_file reference. |
| 1158 | * |
| 1159 | * We take care to handle the case where the page was truncated from the |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 1160 | * mapping by re-checking page_mapping() inside tree_lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1161 | */ |
| 1162 | int __set_page_dirty_nobuffers(struct page *page) |
| 1163 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1164 | if (!TestSetPageDirty(page)) { |
| 1165 | struct address_space *mapping = page_mapping(page); |
| 1166 | struct address_space *mapping2; |
| 1167 | |
Andrew Morton | 8c08540 | 2006-12-10 02:19:24 -0800 | [diff] [blame] | 1168 | if (!mapping) |
| 1169 | return 1; |
| 1170 | |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 1171 | spin_lock_irq(&mapping->tree_lock); |
Andrew Morton | 8c08540 | 2006-12-10 02:19:24 -0800 | [diff] [blame] | 1172 | mapping2 = page_mapping(page); |
| 1173 | if (mapping2) { /* Race with truncate? */ |
| 1174 | BUG_ON(mapping2 != mapping); |
Nick Piggin | 787d221 | 2007-07-17 04:03:34 -0700 | [diff] [blame] | 1175 | WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page)); |
Edward Shishkin | e3a7cca | 2009-03-31 15:19:39 -0700 | [diff] [blame] | 1176 | account_page_dirtied(page, mapping); |
Andrew Morton | 8c08540 | 2006-12-10 02:19:24 -0800 | [diff] [blame] | 1177 | radix_tree_tag_set(&mapping->page_tree, |
| 1178 | page_index(page), PAGECACHE_TAG_DIRTY); |
| 1179 | } |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 1180 | spin_unlock_irq(&mapping->tree_lock); |
Andrew Morton | 8c08540 | 2006-12-10 02:19:24 -0800 | [diff] [blame] | 1181 | if (mapping->host) { |
| 1182 | /* !PageAnon && !swapper_space */ |
| 1183 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1184 | } |
Andrew Morton | 4741c9f | 2006-03-24 03:18:11 -0800 | [diff] [blame] | 1185 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1186 | } |
Andrew Morton | 4741c9f | 2006-03-24 03:18:11 -0800 | [diff] [blame] | 1187 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1188 | } |
| 1189 | EXPORT_SYMBOL(__set_page_dirty_nobuffers); |
| 1190 | |
| 1191 | /* |
| 1192 | * When a writepage implementation decides that it doesn't want to write this |
| 1193 | * page for some reason, it should redirty the locked page via |
| 1194 | * redirty_page_for_writepage() and it should then unlock the page and return 0 |
| 1195 | */ |
| 1196 | int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page) |
| 1197 | { |
| 1198 | wbc->pages_skipped++; |
| 1199 | return __set_page_dirty_nobuffers(page); |
| 1200 | } |
| 1201 | EXPORT_SYMBOL(redirty_page_for_writepage); |
| 1202 | |
| 1203 | /* |
Wu Fengguang | 6746aff | 2009-09-16 11:50:14 +0200 | [diff] [blame] | 1204 | * Dirty a page. |
| 1205 | * |
| 1206 | * For pages with a mapping this should be done under the page lock |
| 1207 | * for the benefit of asynchronous memory errors who prefer a consistent |
| 1208 | * dirty state. This rule can be broken in some special cases, |
| 1209 | * but should be better not to. |
| 1210 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1211 | * If the mapping doesn't provide a set_page_dirty a_op, then |
| 1212 | * just fall through and assume that it wants buffer_heads. |
| 1213 | */ |
Nick Piggin | 1cf6e7d | 2009-02-18 14:48:18 -0800 | [diff] [blame] | 1214 | int set_page_dirty(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1215 | { |
| 1216 | struct address_space *mapping = page_mapping(page); |
| 1217 | |
| 1218 | if (likely(mapping)) { |
| 1219 | int (*spd)(struct page *) = mapping->a_ops->set_page_dirty; |
Minchan Kim | 278df9f | 2011-03-22 16:32:54 -0700 | [diff] [blame] | 1220 | /* |
| 1221 | * readahead/lru_deactivate_page could remain |
| 1222 | * PG_readahead/PG_reclaim due to race with end_page_writeback |
| 1223 | * About readahead, if the page is written, the flags would be |
| 1224 | * reset. So no problem. |
| 1225 | * About lru_deactivate_page, if the page is redirty, the flag |
| 1226 | * will be reset. So no problem. but if the page is used by readahead |
| 1227 | * it will confuse readahead and make it restart the size rampup |
| 1228 | * process. But it's a trivial problem. |
| 1229 | */ |
| 1230 | ClearPageReclaim(page); |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 1231 | #ifdef CONFIG_BLOCK |
| 1232 | if (!spd) |
| 1233 | spd = __set_page_dirty_buffers; |
| 1234 | #endif |
| 1235 | return (*spd)(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1236 | } |
Andrew Morton | 4741c9f | 2006-03-24 03:18:11 -0800 | [diff] [blame] | 1237 | if (!PageDirty(page)) { |
| 1238 | if (!TestSetPageDirty(page)) |
| 1239 | return 1; |
| 1240 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1241 | return 0; |
| 1242 | } |
| 1243 | EXPORT_SYMBOL(set_page_dirty); |
| 1244 | |
| 1245 | /* |
| 1246 | * set_page_dirty() is racy if the caller has no reference against |
| 1247 | * page->mapping->host, and if the page is unlocked. This is because another |
| 1248 | * CPU could truncate the page off the mapping and then free the mapping. |
| 1249 | * |
| 1250 | * Usually, the page _is_ locked, or the caller is a user-space process which |
| 1251 | * holds a reference on the inode by having an open file. |
| 1252 | * |
| 1253 | * In other cases, the page should be locked before running set_page_dirty(). |
| 1254 | */ |
| 1255 | int set_page_dirty_lock(struct page *page) |
| 1256 | { |
| 1257 | int ret; |
| 1258 | |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 1259 | lock_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1260 | ret = set_page_dirty(page); |
| 1261 | unlock_page(page); |
| 1262 | return ret; |
| 1263 | } |
| 1264 | EXPORT_SYMBOL(set_page_dirty_lock); |
| 1265 | |
| 1266 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1267 | * Clear a page's dirty flag, while caring for dirty memory accounting. |
| 1268 | * Returns true if the page was previously dirty. |
| 1269 | * |
| 1270 | * This is for preparing to put the page under writeout. We leave the page |
| 1271 | * tagged as dirty in the radix tree so that a concurrent write-for-sync |
| 1272 | * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage |
| 1273 | * implementation will run either set_page_writeback() or set_page_dirty(), |
| 1274 | * at which stage we bring the page's dirty flag and radix-tree dirty tag |
| 1275 | * back into sync. |
| 1276 | * |
| 1277 | * This incoherency between the page's dirty flag and radix-tree tag is |
| 1278 | * unfortunate, but it only exists while the page is locked. |
| 1279 | */ |
| 1280 | int clear_page_dirty_for_io(struct page *page) |
| 1281 | { |
| 1282 | struct address_space *mapping = page_mapping(page); |
| 1283 | |
Nick Piggin | 7935289 | 2007-07-19 01:47:22 -0700 | [diff] [blame] | 1284 | BUG_ON(!PageLocked(page)); |
| 1285 | |
Linus Torvalds | 7658cc2 | 2006-12-29 10:00:58 -0800 | [diff] [blame] | 1286 | if (mapping && mapping_cap_account_dirty(mapping)) { |
| 1287 | /* |
| 1288 | * Yes, Virginia, this is indeed insane. |
| 1289 | * |
| 1290 | * We use this sequence to make sure that |
| 1291 | * (a) we account for dirty stats properly |
| 1292 | * (b) we tell the low-level filesystem to |
| 1293 | * mark the whole page dirty if it was |
| 1294 | * dirty in a pagetable. Only to then |
| 1295 | * (c) clean the page again and return 1 to |
| 1296 | * cause the writeback. |
| 1297 | * |
| 1298 | * This way we avoid all nasty races with the |
| 1299 | * dirty bit in multiple places and clearing |
| 1300 | * them concurrently from different threads. |
| 1301 | * |
| 1302 | * Note! Normally the "set_page_dirty(page)" |
| 1303 | * has no effect on the actual dirty bit - since |
| 1304 | * that will already usually be set. But we |
| 1305 | * need the side effects, and it can help us |
| 1306 | * avoid races. |
| 1307 | * |
| 1308 | * We basically use the page "master dirty bit" |
| 1309 | * as a serialization point for all the different |
| 1310 | * threads doing their things. |
Linus Torvalds | 7658cc2 | 2006-12-29 10:00:58 -0800 | [diff] [blame] | 1311 | */ |
| 1312 | if (page_mkclean(page)) |
| 1313 | set_page_dirty(page); |
Nick Piggin | 7935289 | 2007-07-19 01:47:22 -0700 | [diff] [blame] | 1314 | /* |
| 1315 | * We carefully synchronise fault handlers against |
| 1316 | * installing a dirty pte and marking the page dirty |
| 1317 | * at this point. We do this by having them hold the |
| 1318 | * page lock at some point after installing their |
| 1319 | * pte, but before marking the page dirty. |
| 1320 | * Pages are always locked coming in here, so we get |
| 1321 | * the desired exclusion. See mm/memory.c:do_wp_page() |
| 1322 | * for more comments. |
| 1323 | */ |
Linus Torvalds | 7658cc2 | 2006-12-29 10:00:58 -0800 | [diff] [blame] | 1324 | if (TestClearPageDirty(page)) { |
Andrew Morton | 8c08540 | 2006-12-10 02:19:24 -0800 | [diff] [blame] | 1325 | dec_zone_page_state(page, NR_FILE_DIRTY); |
Peter Zijlstra | c9e51e4 | 2007-10-16 23:25:47 -0700 | [diff] [blame] | 1326 | dec_bdi_stat(mapping->backing_dev_info, |
| 1327 | BDI_RECLAIMABLE); |
Linus Torvalds | 7658cc2 | 2006-12-29 10:00:58 -0800 | [diff] [blame] | 1328 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1329 | } |
Linus Torvalds | 7658cc2 | 2006-12-29 10:00:58 -0800 | [diff] [blame] | 1330 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1331 | } |
Linus Torvalds | 7658cc2 | 2006-12-29 10:00:58 -0800 | [diff] [blame] | 1332 | return TestClearPageDirty(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1333 | } |
Hans Reiser | 58bb01a | 2005-11-18 01:10:53 -0800 | [diff] [blame] | 1334 | EXPORT_SYMBOL(clear_page_dirty_for_io); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1335 | |
| 1336 | int test_clear_page_writeback(struct page *page) |
| 1337 | { |
| 1338 | struct address_space *mapping = page_mapping(page); |
| 1339 | int ret; |
| 1340 | |
| 1341 | if (mapping) { |
Peter Zijlstra | 69cb51d | 2007-10-16 23:25:48 -0700 | [diff] [blame] | 1342 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1343 | unsigned long flags; |
| 1344 | |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 1345 | spin_lock_irqsave(&mapping->tree_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1346 | ret = TestClearPageWriteback(page); |
Peter Zijlstra | 69cb51d | 2007-10-16 23:25:48 -0700 | [diff] [blame] | 1347 | if (ret) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1348 | radix_tree_tag_clear(&mapping->page_tree, |
| 1349 | page_index(page), |
| 1350 | PAGECACHE_TAG_WRITEBACK); |
Miklos Szeredi | e4ad08f | 2008-04-30 00:54:37 -0700 | [diff] [blame] | 1351 | if (bdi_cap_account_writeback(bdi)) { |
Peter Zijlstra | 69cb51d | 2007-10-16 23:25:48 -0700 | [diff] [blame] | 1352 | __dec_bdi_stat(bdi, BDI_WRITEBACK); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 1353 | __bdi_writeout_inc(bdi); |
| 1354 | } |
Peter Zijlstra | 69cb51d | 2007-10-16 23:25:48 -0700 | [diff] [blame] | 1355 | } |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 1356 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1357 | } else { |
| 1358 | ret = TestClearPageWriteback(page); |
| 1359 | } |
Andrew Morton | d688abf | 2007-07-19 01:49:17 -0700 | [diff] [blame] | 1360 | if (ret) |
| 1361 | dec_zone_page_state(page, NR_WRITEBACK); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1362 | return ret; |
| 1363 | } |
| 1364 | |
| 1365 | int test_set_page_writeback(struct page *page) |
| 1366 | { |
| 1367 | struct address_space *mapping = page_mapping(page); |
| 1368 | int ret; |
| 1369 | |
| 1370 | if (mapping) { |
Peter Zijlstra | 69cb51d | 2007-10-16 23:25:48 -0700 | [diff] [blame] | 1371 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1372 | unsigned long flags; |
| 1373 | |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 1374 | spin_lock_irqsave(&mapping->tree_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1375 | ret = TestSetPageWriteback(page); |
Peter Zijlstra | 69cb51d | 2007-10-16 23:25:48 -0700 | [diff] [blame] | 1376 | if (!ret) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1377 | radix_tree_tag_set(&mapping->page_tree, |
| 1378 | page_index(page), |
| 1379 | PAGECACHE_TAG_WRITEBACK); |
Miklos Szeredi | e4ad08f | 2008-04-30 00:54:37 -0700 | [diff] [blame] | 1380 | if (bdi_cap_account_writeback(bdi)) |
Peter Zijlstra | 69cb51d | 2007-10-16 23:25:48 -0700 | [diff] [blame] | 1381 | __inc_bdi_stat(bdi, BDI_WRITEBACK); |
| 1382 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1383 | if (!PageDirty(page)) |
| 1384 | radix_tree_tag_clear(&mapping->page_tree, |
| 1385 | page_index(page), |
| 1386 | PAGECACHE_TAG_DIRTY); |
Jan Kara | f446daa | 2010-08-09 17:19:12 -0700 | [diff] [blame] | 1387 | radix_tree_tag_clear(&mapping->page_tree, |
| 1388 | page_index(page), |
| 1389 | PAGECACHE_TAG_TOWRITE); |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 1390 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1391 | } else { |
| 1392 | ret = TestSetPageWriteback(page); |
| 1393 | } |
Andrew Morton | d688abf | 2007-07-19 01:49:17 -0700 | [diff] [blame] | 1394 | if (!ret) |
Michael Rubin | f629d1c | 2010-10-26 14:21:33 -0700 | [diff] [blame] | 1395 | account_page_writeback(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1396 | return ret; |
| 1397 | |
| 1398 | } |
| 1399 | EXPORT_SYMBOL(test_set_page_writeback); |
| 1400 | |
| 1401 | /* |
Nick Piggin | 0012818 | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 1402 | * Return true if any of the pages in the mapping are marked with the |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1403 | * passed tag. |
| 1404 | */ |
| 1405 | int mapping_tagged(struct address_space *mapping, int tag) |
| 1406 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1407 | int ret; |
Nick Piggin | 0012818 | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 1408 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1409 | ret = radix_tree_tagged(&mapping->page_tree, tag); |
Nick Piggin | 0012818 | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 1410 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1411 | return ret; |
| 1412 | } |
| 1413 | EXPORT_SYMBOL(mapping_tagged); |