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> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 37 | |
| 38 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | * After a CPU has dirtied this many pages, balance_dirty_pages_ratelimited |
| 40 | * will look to see if it needs to force writeback or throttling. |
| 41 | */ |
| 42 | static long ratelimit_pages = 32; |
| 43 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 44 | /* |
| 45 | * When balance_dirty_pages decides that the caller needs to perform some |
| 46 | * non-background writeback, this is how many pages it will attempt to write. |
| 47 | * It should be somewhat larger than RATELIMIT_PAGES to ensure that reasonably |
| 48 | * large amounts of I/O are submitted. |
| 49 | */ |
| 50 | static inline long sync_writeback_pages(void) |
| 51 | { |
| 52 | return ratelimit_pages + ratelimit_pages / 2; |
| 53 | } |
| 54 | |
| 55 | /* The following parameters are exported via /proc/sys/vm */ |
| 56 | |
| 57 | /* |
| 58 | * Start background writeback (via pdflush) at this percentage |
| 59 | */ |
Wu Fengguang | 1b5e62b | 2009-03-23 08:57:38 +0800 | [diff] [blame] | 60 | int dirty_background_ratio = 10; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | |
| 62 | /* |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 63 | * dirty_background_bytes starts at 0 (disabled) so that it is a function of |
| 64 | * dirty_background_ratio * the amount of dirtyable memory |
| 65 | */ |
| 66 | unsigned long dirty_background_bytes; |
| 67 | |
| 68 | /* |
Bron Gondwana | 195cf453 | 2008-02-04 22:29:20 -0800 | [diff] [blame] | 69 | * free highmem will not be subtracted from the total free memory |
| 70 | * for calculating free ratios if vm_highmem_is_dirtyable is true |
| 71 | */ |
| 72 | int vm_highmem_is_dirtyable; |
| 73 | |
| 74 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 75 | * The generator of dirty data starts writeback at this percentage |
| 76 | */ |
Wu Fengguang | 1b5e62b | 2009-03-23 08:57:38 +0800 | [diff] [blame] | 77 | int vm_dirty_ratio = 20; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 78 | |
| 79 | /* |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 80 | * vm_dirty_bytes starts at 0 (disabled) so that it is a function of |
| 81 | * vm_dirty_ratio * the amount of dirtyable memory |
| 82 | */ |
| 83 | unsigned long vm_dirty_bytes; |
| 84 | |
| 85 | /* |
Alexey Dobriyan | 704503d | 2009-03-31 15:23:18 -0700 | [diff] [blame] | 86 | * The interval between `kupdate'-style writebacks |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 87 | */ |
Toshiyuki Okajima | 22ef37e | 2009-05-16 22:56:28 -0700 | [diff] [blame] | 88 | unsigned int dirty_writeback_interval = 5 * 100; /* centiseconds */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 89 | |
| 90 | /* |
Alexey Dobriyan | 704503d | 2009-03-31 15:23:18 -0700 | [diff] [blame] | 91 | * The longest time for which data is allowed to remain dirty |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 92 | */ |
Toshiyuki Okajima | 22ef37e | 2009-05-16 22:56:28 -0700 | [diff] [blame] | 93 | unsigned int dirty_expire_interval = 30 * 100; /* centiseconds */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 94 | |
| 95 | /* |
| 96 | * Flag that makes the machine dump writes/reads and block dirtyings. |
| 97 | */ |
| 98 | int block_dump; |
| 99 | |
| 100 | /* |
Bart Samwel | ed5b43f | 2006-03-24 03:15:49 -0800 | [diff] [blame] | 101 | * Flag that puts the machine in "laptop mode". Doubles as a timeout in jiffies: |
| 102 | * 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] | 103 | */ |
| 104 | int laptop_mode; |
| 105 | |
| 106 | EXPORT_SYMBOL(laptop_mode); |
| 107 | |
| 108 | /* End of sysctl-exported parameters */ |
| 109 | |
| 110 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 111 | /* |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 112 | * Scale the writeback cache size proportional to the relative writeout speeds. |
| 113 | * |
| 114 | * We do this by keeping a floating proportion between BDIs, based on page |
| 115 | * writeback completions [end_page_writeback()]. Those devices that write out |
| 116 | * pages fastest will get the larger share, while the slower will get a smaller |
| 117 | * share. |
| 118 | * |
| 119 | * We use page writeout completions because we are interested in getting rid of |
| 120 | * dirty pages. Having them written out is the primary goal. |
| 121 | * |
| 122 | * We introduce a concept of time, a period over which we measure these events, |
| 123 | * because demand can/will vary over time. The length of this period itself is |
| 124 | * measured in page writeback completions. |
| 125 | * |
| 126 | */ |
| 127 | static struct prop_descriptor vm_completions; |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 128 | static struct prop_descriptor vm_dirties; |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 129 | |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 130 | /* |
| 131 | * couple the period to the dirty_ratio: |
| 132 | * |
| 133 | * period/2 ~ roundup_pow_of_two(dirty limit) |
| 134 | */ |
| 135 | static int calc_period_shift(void) |
| 136 | { |
| 137 | unsigned long dirty_total; |
| 138 | |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 139 | if (vm_dirty_bytes) |
| 140 | dirty_total = vm_dirty_bytes / PAGE_SIZE; |
| 141 | else |
| 142 | dirty_total = (vm_dirty_ratio * determine_dirtyable_memory()) / |
| 143 | 100; |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 144 | return 2 + ilog2(dirty_total - 1); |
| 145 | } |
| 146 | |
| 147 | /* |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 148 | * update the period when the dirty threshold changes. |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 149 | */ |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 150 | static void update_completion_period(void) |
| 151 | { |
| 152 | int shift = calc_period_shift(); |
| 153 | prop_change_shift(&vm_completions, shift); |
| 154 | prop_change_shift(&vm_dirties, shift); |
| 155 | } |
| 156 | |
| 157 | int dirty_background_ratio_handler(struct ctl_table *table, int write, |
| 158 | struct file *filp, void __user *buffer, size_t *lenp, |
| 159 | loff_t *ppos) |
| 160 | { |
| 161 | int ret; |
| 162 | |
| 163 | ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos); |
| 164 | if (ret == 0 && write) |
| 165 | dirty_background_bytes = 0; |
| 166 | return ret; |
| 167 | } |
| 168 | |
| 169 | int dirty_background_bytes_handler(struct ctl_table *table, int write, |
| 170 | struct file *filp, void __user *buffer, size_t *lenp, |
| 171 | loff_t *ppos) |
| 172 | { |
| 173 | int ret; |
| 174 | |
| 175 | ret = proc_doulongvec_minmax(table, write, filp, buffer, lenp, ppos); |
| 176 | if (ret == 0 && write) |
| 177 | dirty_background_ratio = 0; |
| 178 | return ret; |
| 179 | } |
| 180 | |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 181 | int dirty_ratio_handler(struct ctl_table *table, int write, |
| 182 | struct file *filp, void __user *buffer, size_t *lenp, |
| 183 | loff_t *ppos) |
| 184 | { |
| 185 | int old_ratio = vm_dirty_ratio; |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 186 | int ret; |
| 187 | |
| 188 | ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 189 | if (ret == 0 && write && vm_dirty_ratio != old_ratio) { |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 190 | update_completion_period(); |
| 191 | vm_dirty_bytes = 0; |
| 192 | } |
| 193 | return ret; |
| 194 | } |
| 195 | |
| 196 | |
| 197 | int dirty_bytes_handler(struct ctl_table *table, int write, |
| 198 | struct file *filp, void __user *buffer, size_t *lenp, |
| 199 | loff_t *ppos) |
| 200 | { |
Sven Wegener | fc3501d | 2009-02-11 13:04:23 -0800 | [diff] [blame] | 201 | unsigned long old_bytes = vm_dirty_bytes; |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 202 | int ret; |
| 203 | |
| 204 | ret = proc_doulongvec_minmax(table, write, filp, buffer, lenp, ppos); |
| 205 | if (ret == 0 && write && vm_dirty_bytes != old_bytes) { |
| 206 | update_completion_period(); |
| 207 | vm_dirty_ratio = 0; |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 208 | } |
| 209 | return ret; |
| 210 | } |
| 211 | |
| 212 | /* |
| 213 | * Increment the BDI's writeout completion count and the global writeout |
| 214 | * completion count. Called from test_clear_page_writeback(). |
| 215 | */ |
| 216 | static inline void __bdi_writeout_inc(struct backing_dev_info *bdi) |
| 217 | { |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 218 | __prop_inc_percpu_max(&vm_completions, &bdi->completions, |
| 219 | bdi->max_prop_frac); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 220 | } |
| 221 | |
Miklos Szeredi | dd5656e | 2008-04-30 00:54:37 -0700 | [diff] [blame] | 222 | void bdi_writeout_inc(struct backing_dev_info *bdi) |
| 223 | { |
| 224 | unsigned long flags; |
| 225 | |
| 226 | local_irq_save(flags); |
| 227 | __bdi_writeout_inc(bdi); |
| 228 | local_irq_restore(flags); |
| 229 | } |
| 230 | EXPORT_SYMBOL_GPL(bdi_writeout_inc); |
| 231 | |
Nick Piggin | 1cf6e7d | 2009-02-18 14:48:18 -0800 | [diff] [blame] | 232 | void task_dirty_inc(struct task_struct *tsk) |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 233 | { |
| 234 | prop_inc_single(&vm_dirties, &tsk->dirties); |
| 235 | } |
| 236 | |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 237 | /* |
| 238 | * Obtain an accurate fraction of the BDI's portion. |
| 239 | */ |
| 240 | static void bdi_writeout_fraction(struct backing_dev_info *bdi, |
| 241 | long *numerator, long *denominator) |
| 242 | { |
| 243 | if (bdi_cap_writeback_dirty(bdi)) { |
| 244 | prop_fraction_percpu(&vm_completions, &bdi->completions, |
| 245 | numerator, denominator); |
| 246 | } else { |
| 247 | *numerator = 0; |
| 248 | *denominator = 1; |
| 249 | } |
| 250 | } |
| 251 | |
| 252 | /* |
| 253 | * Clip the earned share of dirty pages to that which is actually available. |
| 254 | * This avoids exceeding the total dirty_limit when the floating averages |
| 255 | * fluctuate too quickly. |
| 256 | */ |
H Hartley Sweeten | dcf975d | 2009-06-16 15:31:44 -0700 | [diff] [blame] | 257 | static void clip_bdi_dirty_limit(struct backing_dev_info *bdi, |
| 258 | unsigned long dirty, unsigned long *pbdi_dirty) |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 259 | { |
H Hartley Sweeten | dcf975d | 2009-06-16 15:31:44 -0700 | [diff] [blame] | 260 | unsigned long avail_dirty; |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 261 | |
H Hartley Sweeten | dcf975d | 2009-06-16 15:31:44 -0700 | [diff] [blame] | 262 | avail_dirty = global_page_state(NR_FILE_DIRTY) + |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 263 | global_page_state(NR_WRITEBACK) + |
Miklos Szeredi | fc3ba69 | 2008-04-30 00:54:38 -0700 | [diff] [blame] | 264 | global_page_state(NR_UNSTABLE_NFS) + |
H Hartley Sweeten | dcf975d | 2009-06-16 15:31:44 -0700 | [diff] [blame] | 265 | global_page_state(NR_WRITEBACK_TEMP); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 266 | |
H Hartley Sweeten | dcf975d | 2009-06-16 15:31:44 -0700 | [diff] [blame] | 267 | if (avail_dirty < dirty) |
| 268 | avail_dirty = dirty - avail_dirty; |
| 269 | else |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 270 | avail_dirty = 0; |
| 271 | |
| 272 | avail_dirty += bdi_stat(bdi, BDI_RECLAIMABLE) + |
| 273 | bdi_stat(bdi, BDI_WRITEBACK); |
| 274 | |
| 275 | *pbdi_dirty = min(*pbdi_dirty, avail_dirty); |
| 276 | } |
| 277 | |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 278 | static inline void task_dirties_fraction(struct task_struct *tsk, |
| 279 | long *numerator, long *denominator) |
| 280 | { |
| 281 | prop_fraction_single(&vm_dirties, &tsk->dirties, |
| 282 | numerator, denominator); |
| 283 | } |
| 284 | |
| 285 | /* |
| 286 | * scale the dirty limit |
| 287 | * |
| 288 | * task specific dirty limit: |
| 289 | * |
| 290 | * dirty -= (dirty/8) * p_{t} |
| 291 | */ |
H Hartley Sweeten | dcf975d | 2009-06-16 15:31:44 -0700 | [diff] [blame] | 292 | static void task_dirty_limit(struct task_struct *tsk, unsigned long *pdirty) |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 293 | { |
| 294 | long numerator, denominator; |
H Hartley Sweeten | dcf975d | 2009-06-16 15:31:44 -0700 | [diff] [blame] | 295 | unsigned long dirty = *pdirty; |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 296 | u64 inv = dirty >> 3; |
| 297 | |
| 298 | task_dirties_fraction(tsk, &numerator, &denominator); |
| 299 | inv *= numerator; |
| 300 | do_div(inv, denominator); |
| 301 | |
| 302 | dirty -= inv; |
| 303 | if (dirty < *pdirty/2) |
| 304 | dirty = *pdirty/2; |
| 305 | |
| 306 | *pdirty = dirty; |
| 307 | } |
| 308 | |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 309 | /* |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 310 | * |
| 311 | */ |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 312 | static unsigned int bdi_min_ratio; |
| 313 | |
| 314 | int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio) |
| 315 | { |
| 316 | int ret = 0; |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 317 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 318 | spin_lock(&bdi_lock); |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 319 | if (min_ratio > bdi->max_ratio) { |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 320 | ret = -EINVAL; |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 321 | } else { |
| 322 | min_ratio -= bdi->min_ratio; |
| 323 | if (bdi_min_ratio + min_ratio < 100) { |
| 324 | bdi_min_ratio += min_ratio; |
| 325 | bdi->min_ratio += min_ratio; |
| 326 | } else { |
| 327 | ret = -EINVAL; |
| 328 | } |
| 329 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 330 | spin_unlock(&bdi_lock); |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 331 | |
| 332 | return ret; |
| 333 | } |
| 334 | |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 335 | int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned max_ratio) |
| 336 | { |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 337 | int ret = 0; |
| 338 | |
| 339 | if (max_ratio > 100) |
| 340 | return -EINVAL; |
| 341 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 342 | spin_lock(&bdi_lock); |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 343 | if (bdi->min_ratio > max_ratio) { |
| 344 | ret = -EINVAL; |
| 345 | } else { |
| 346 | bdi->max_ratio = max_ratio; |
| 347 | bdi->max_prop_frac = (PROP_FRAC_BASE * max_ratio) / 100; |
| 348 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 349 | spin_unlock(&bdi_lock); |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 350 | |
| 351 | return ret; |
| 352 | } |
| 353 | EXPORT_SYMBOL(bdi_set_max_ratio); |
| 354 | |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 355 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 356 | * Work out the current dirty-memory clamping and background writeout |
| 357 | * thresholds. |
| 358 | * |
| 359 | * The main aim here is to lower them aggressively if there is a lot of mapped |
| 360 | * memory around. To avoid stressing page reclaim with lots of unreclaimable |
| 361 | * pages. It is better to clamp down on writers than to start swapping, and |
| 362 | * performing lots of scanning. |
| 363 | * |
| 364 | * We only allow 1/2 of the currently-unmapped memory to be dirtied. |
| 365 | * |
| 366 | * We don't permit the clamping level to fall below 5% - that is getting rather |
| 367 | * excessive. |
| 368 | * |
| 369 | * We make sure that the background writeout level is below the adjusted |
| 370 | * clamping level. |
| 371 | */ |
Christoph Lameter | 1b42446 | 2007-05-06 14:48:59 -0700 | [diff] [blame] | 372 | |
| 373 | static unsigned long highmem_dirtyable_memory(unsigned long total) |
| 374 | { |
| 375 | #ifdef CONFIG_HIGHMEM |
| 376 | int node; |
| 377 | unsigned long x = 0; |
| 378 | |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 379 | for_each_node_state(node, N_HIGH_MEMORY) { |
Christoph Lameter | 1b42446 | 2007-05-06 14:48:59 -0700 | [diff] [blame] | 380 | struct zone *z = |
| 381 | &NODE_DATA(node)->node_zones[ZONE_HIGHMEM]; |
| 382 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 383 | x += zone_page_state(z, NR_FREE_PAGES) + zone_lru_pages(z); |
Christoph Lameter | 1b42446 | 2007-05-06 14:48:59 -0700 | [diff] [blame] | 384 | } |
| 385 | /* |
| 386 | * Make sure that the number of highmem pages is never larger |
| 387 | * than the number of the total dirtyable memory. This can only |
| 388 | * occur in very strange VM situations but we want to make sure |
| 389 | * that this does not occur. |
| 390 | */ |
| 391 | return min(x, total); |
| 392 | #else |
| 393 | return 0; |
| 394 | #endif |
| 395 | } |
| 396 | |
Steven Rostedt | 3eefae9 | 2008-05-12 21:21:04 +0200 | [diff] [blame] | 397 | /** |
| 398 | * determine_dirtyable_memory - amount of memory that may be used |
| 399 | * |
| 400 | * Returns the numebr of pages that can currently be freed and used |
| 401 | * by the kernel for direct mappings. |
| 402 | */ |
| 403 | unsigned long determine_dirtyable_memory(void) |
Christoph Lameter | 1b42446 | 2007-05-06 14:48:59 -0700 | [diff] [blame] | 404 | { |
| 405 | unsigned long x; |
| 406 | |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 407 | x = global_page_state(NR_FREE_PAGES) + global_lru_pages(); |
Bron Gondwana | 195cf453 | 2008-02-04 22:29:20 -0800 | [diff] [blame] | 408 | |
| 409 | if (!vm_highmem_is_dirtyable) |
| 410 | x -= highmem_dirtyable_memory(x); |
| 411 | |
Christoph Lameter | 1b42446 | 2007-05-06 14:48:59 -0700 | [diff] [blame] | 412 | return x + 1; /* Ensure that we never return 0 */ |
| 413 | } |
| 414 | |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 415 | void |
David Rientjes | 364aeb2 | 2009-01-06 14:39:29 -0800 | [diff] [blame] | 416 | get_dirty_limits(unsigned long *pbackground, unsigned long *pdirty, |
| 417 | unsigned long *pbdi_dirty, struct backing_dev_info *bdi) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 418 | { |
David Rientjes | 364aeb2 | 2009-01-06 14:39:29 -0800 | [diff] [blame] | 419 | unsigned long background; |
| 420 | unsigned long dirty; |
Christoph Lameter | 1b42446 | 2007-05-06 14:48:59 -0700 | [diff] [blame] | 421 | unsigned long available_memory = determine_dirtyable_memory(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 422 | struct task_struct *tsk; |
| 423 | |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 424 | if (vm_dirty_bytes) |
| 425 | dirty = DIV_ROUND_UP(vm_dirty_bytes, PAGE_SIZE); |
| 426 | else { |
| 427 | int dirty_ratio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 428 | |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 429 | dirty_ratio = vm_dirty_ratio; |
| 430 | if (dirty_ratio < 5) |
| 431 | dirty_ratio = 5; |
| 432 | dirty = (dirty_ratio * available_memory) / 100; |
| 433 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 434 | |
David Rientjes | 2da0299 | 2009-01-06 14:39:31 -0800 | [diff] [blame] | 435 | if (dirty_background_bytes) |
| 436 | background = DIV_ROUND_UP(dirty_background_bytes, PAGE_SIZE); |
| 437 | else |
| 438 | background = (dirty_background_ratio * available_memory) / 100; |
| 439 | |
| 440 | if (background >= dirty) |
| 441 | background = dirty / 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 442 | tsk = current; |
| 443 | if (tsk->flags & PF_LESS_THROTTLE || rt_task(tsk)) { |
| 444 | background += background / 4; |
| 445 | dirty += dirty / 4; |
| 446 | } |
| 447 | *pbackground = background; |
| 448 | *pdirty = dirty; |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 449 | |
| 450 | if (bdi) { |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 451 | u64 bdi_dirty; |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 452 | long numerator, denominator; |
| 453 | |
| 454 | /* |
| 455 | * Calculate this BDI's share of the dirty ratio. |
| 456 | */ |
| 457 | bdi_writeout_fraction(bdi, &numerator, &denominator); |
| 458 | |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 459 | bdi_dirty = (dirty * (100 - bdi_min_ratio)) / 100; |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 460 | bdi_dirty *= numerator; |
| 461 | do_div(bdi_dirty, denominator); |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 462 | bdi_dirty += (dirty * bdi->min_ratio) / 100; |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 463 | if (bdi_dirty > (dirty * bdi->max_ratio) / 100) |
| 464 | bdi_dirty = dirty * bdi->max_ratio / 100; |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 465 | |
| 466 | *pbdi_dirty = bdi_dirty; |
| 467 | clip_bdi_dirty_limit(bdi, dirty, pbdi_dirty); |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 468 | task_dirty_limit(current, pbdi_dirty); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 469 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | } |
| 471 | |
| 472 | /* |
| 473 | * balance_dirty_pages() must be called by processes which are generating dirty |
| 474 | * data. It looks at the number of dirty pages in the machine and will force |
| 475 | * the caller to perform writeback if the system is over `vm_dirty_ratio'. |
| 476 | * If we're over `background_thresh' then pdflush is woken to perform some |
| 477 | * writeout. |
| 478 | */ |
| 479 | static void balance_dirty_pages(struct address_space *mapping) |
| 480 | { |
Peter Zijlstra | 5fce25a | 2007-11-14 16:59:15 -0800 | [diff] [blame] | 481 | long nr_reclaimable, bdi_nr_reclaimable; |
| 482 | long nr_writeback, bdi_nr_writeback; |
David Rientjes | 364aeb2 | 2009-01-06 14:39:29 -0800 | [diff] [blame] | 483 | unsigned long background_thresh; |
| 484 | unsigned long dirty_thresh; |
| 485 | unsigned long bdi_thresh; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 486 | unsigned long pages_written = 0; |
| 487 | unsigned long write_chunk = sync_writeback_pages(); |
| 488 | |
| 489 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
| 490 | |
| 491 | for (;;) { |
| 492 | struct writeback_control wbc = { |
| 493 | .bdi = bdi, |
| 494 | .sync_mode = WB_SYNC_NONE, |
| 495 | .older_than_this = NULL, |
| 496 | .nr_to_write = write_chunk, |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 497 | .range_cyclic = 1, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 498 | }; |
| 499 | |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 500 | get_dirty_limits(&background_thresh, &dirty_thresh, |
| 501 | &bdi_thresh, bdi); |
Peter Zijlstra | 5fce25a | 2007-11-14 16:59:15 -0800 | [diff] [blame] | 502 | |
| 503 | nr_reclaimable = global_page_state(NR_FILE_DIRTY) + |
| 504 | global_page_state(NR_UNSTABLE_NFS); |
| 505 | nr_writeback = global_page_state(NR_WRITEBACK); |
| 506 | |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 507 | bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE); |
| 508 | bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK); |
Peter Zijlstra | 5fce25a | 2007-11-14 16:59:15 -0800 | [diff] [blame] | 509 | |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 510 | if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh) |
| 511 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 512 | |
Peter Zijlstra | 5fce25a | 2007-11-14 16:59:15 -0800 | [diff] [blame] | 513 | /* |
| 514 | * Throttle it only when the background writeback cannot |
| 515 | * catch-up. This avoids (excessively) small writeouts |
| 516 | * when the bdi limits are ramping up. |
| 517 | */ |
| 518 | if (nr_reclaimable + nr_writeback < |
| 519 | (background_thresh + dirty_thresh) / 2) |
| 520 | break; |
| 521 | |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 522 | if (!bdi->dirty_exceeded) |
| 523 | bdi->dirty_exceeded = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 524 | |
| 525 | /* Note: nr_reclaimable denotes nr_dirty + nr_unstable. |
| 526 | * Unstable writes are a feature of certain networked |
| 527 | * filesystems (i.e. NFS) in which data may have been |
| 528 | * written to the server's write cache, but has not yet |
| 529 | * been flushed to permanent storage. |
Richard Kennedy | d7831a0 | 2009-06-30 11:41:35 -0700 | [diff] [blame] | 530 | * Only move pages to writeback if this bdi is over its |
| 531 | * threshold otherwise wait until the disk writes catch |
| 532 | * up. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 533 | */ |
Richard Kennedy | d7831a0 | 2009-06-30 11:41:35 -0700 | [diff] [blame] | 534 | if (bdi_nr_reclaimable > bdi_thresh) { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 535 | writeback_inodes_wbc(&wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 536 | pages_written += write_chunk - wbc.nr_to_write; |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 537 | get_dirty_limits(&background_thresh, &dirty_thresh, |
| 538 | &bdi_thresh, bdi); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 539 | } |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 540 | |
| 541 | /* |
| 542 | * In order to avoid the stacked BDI deadlock we need |
| 543 | * to ensure we accurately count the 'dirty' pages when |
| 544 | * the threshold is low. |
| 545 | * |
| 546 | * Otherwise it would be possible to get thresh+n pages |
| 547 | * reported dirty, even though there are thresh-m pages |
| 548 | * actually dirty; with m+n sitting in the percpu |
| 549 | * deltas. |
| 550 | */ |
| 551 | if (bdi_thresh < 2*bdi_stat_error(bdi)) { |
| 552 | bdi_nr_reclaimable = bdi_stat_sum(bdi, BDI_RECLAIMABLE); |
| 553 | bdi_nr_writeback = bdi_stat_sum(bdi, BDI_WRITEBACK); |
| 554 | } else if (bdi_nr_reclaimable) { |
| 555 | bdi_nr_reclaimable = bdi_stat(bdi, BDI_RECLAIMABLE); |
| 556 | bdi_nr_writeback = bdi_stat(bdi, BDI_WRITEBACK); |
| 557 | } |
| 558 | |
| 559 | if (bdi_nr_reclaimable + bdi_nr_writeback <= bdi_thresh) |
| 560 | break; |
| 561 | if (pages_written >= write_chunk) |
| 562 | break; /* We've done our duty */ |
| 563 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 564 | schedule_timeout(1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 565 | } |
| 566 | |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 567 | if (bdi_nr_reclaimable + bdi_nr_writeback < bdi_thresh && |
| 568 | bdi->dirty_exceeded) |
| 569 | bdi->dirty_exceeded = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 570 | |
| 571 | if (writeback_in_progress(bdi)) |
| 572 | return; /* pdflush is already working this queue */ |
| 573 | |
| 574 | /* |
| 575 | * In laptop mode, we wait until hitting the higher threshold before |
| 576 | * starting background writeout, and then write out all the way down |
| 577 | * to the lower threshold. So slow writers cause minimal disk activity. |
| 578 | * |
| 579 | * In normal mode, we start background writeout at the lower |
| 580 | * background_thresh, to keep the amount of dirty memory low. |
| 581 | */ |
| 582 | if ((laptop_mode && pages_written) || |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 583 | (!laptop_mode && ((nr_writeback = global_page_state(NR_FILE_DIRTY) |
| 584 | + global_page_state(NR_UNSTABLE_NFS)) |
| 585 | > background_thresh))) { |
| 586 | struct writeback_control wbc = { |
| 587 | .bdi = bdi, |
| 588 | .sync_mode = WB_SYNC_NONE, |
| 589 | .nr_to_write = nr_writeback, |
| 590 | }; |
| 591 | |
| 592 | |
| 593 | bdi_start_writeback(&wbc); |
| 594 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 595 | } |
| 596 | |
Peter Zijlstra | a200ee1 | 2007-10-08 18:54:37 +0200 | [diff] [blame] | 597 | void set_page_dirty_balance(struct page *page, int page_mkwrite) |
Peter Zijlstra | edc79b2 | 2006-09-25 23:30:58 -0700 | [diff] [blame] | 598 | { |
Peter Zijlstra | a200ee1 | 2007-10-08 18:54:37 +0200 | [diff] [blame] | 599 | if (set_page_dirty(page) || page_mkwrite) { |
Peter Zijlstra | edc79b2 | 2006-09-25 23:30:58 -0700 | [diff] [blame] | 600 | struct address_space *mapping = page_mapping(page); |
| 601 | |
| 602 | if (mapping) |
| 603 | balance_dirty_pages_ratelimited(mapping); |
| 604 | } |
| 605 | } |
| 606 | |
Tejun Heo | 245b2e7 | 2009-06-24 15:13:48 +0900 | [diff] [blame] | 607 | static DEFINE_PER_CPU(unsigned long, bdp_ratelimits) = 0; |
| 608 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 609 | /** |
Andrew Morton | fa5a734 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 610 | * balance_dirty_pages_ratelimited_nr - balance dirty memory state |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 611 | * @mapping: address_space which was dirtied |
Martin Waitz | a580290 | 2006-04-02 13:59:55 +0200 | [diff] [blame] | 612 | * @nr_pages_dirtied: number of pages which the caller has just dirtied |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 613 | * |
| 614 | * Processes which are dirtying memory should call in here once for each page |
| 615 | * which was newly dirtied. The function will periodically check the system's |
| 616 | * dirty state and will initiate writeback if needed. |
| 617 | * |
| 618 | * On really big machines, get_writeback_state is expensive, so try to avoid |
| 619 | * calling it too often (ratelimiting). But once we're over the dirty memory |
| 620 | * limit we decrease the ratelimiting by a lot, to prevent individual processes |
| 621 | * from overshooting the limit by (ratelimit_pages) each. |
| 622 | */ |
Andrew Morton | fa5a734 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 623 | void balance_dirty_pages_ratelimited_nr(struct address_space *mapping, |
| 624 | unsigned long nr_pages_dirtied) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 625 | { |
Andrew Morton | fa5a734 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 626 | unsigned long ratelimit; |
| 627 | unsigned long *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 628 | |
| 629 | ratelimit = ratelimit_pages; |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 630 | if (mapping->backing_dev_info->dirty_exceeded) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 631 | ratelimit = 8; |
| 632 | |
| 633 | /* |
| 634 | * Check the rate limiting. Also, we do not want to throttle real-time |
| 635 | * tasks in balance_dirty_pages(). Period. |
| 636 | */ |
Andrew Morton | fa5a734 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 637 | preempt_disable(); |
Tejun Heo | 245b2e7 | 2009-06-24 15:13:48 +0900 | [diff] [blame] | 638 | p = &__get_cpu_var(bdp_ratelimits); |
Andrew Morton | fa5a734 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 639 | *p += nr_pages_dirtied; |
| 640 | if (unlikely(*p >= ratelimit)) { |
| 641 | *p = 0; |
| 642 | preempt_enable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 643 | balance_dirty_pages(mapping); |
| 644 | return; |
| 645 | } |
Andrew Morton | fa5a734 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 646 | preempt_enable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 647 | } |
Andrew Morton | fa5a734 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 648 | EXPORT_SYMBOL(balance_dirty_pages_ratelimited_nr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 649 | |
Andrew Morton | 232ea4d | 2007-02-28 20:13:21 -0800 | [diff] [blame] | 650 | void throttle_vm_writeout(gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 651 | { |
David Rientjes | 364aeb2 | 2009-01-06 14:39:29 -0800 | [diff] [blame] | 652 | unsigned long background_thresh; |
| 653 | unsigned long dirty_thresh; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 654 | |
| 655 | for ( ; ; ) { |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 656 | get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 657 | |
| 658 | /* |
| 659 | * Boost the allowable dirty threshold a bit for page |
| 660 | * allocators so they don't get DoS'ed by heavy writers |
| 661 | */ |
| 662 | dirty_thresh += dirty_thresh / 10; /* wheeee... */ |
| 663 | |
Christoph Lameter | c24f21b | 2006-06-30 01:55:42 -0700 | [diff] [blame] | 664 | if (global_page_state(NR_UNSTABLE_NFS) + |
| 665 | global_page_state(NR_WRITEBACK) <= dirty_thresh) |
| 666 | break; |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 667 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
Fengguang Wu | 369f238 | 2007-10-16 23:30:45 -0700 | [diff] [blame] | 668 | |
| 669 | /* |
| 670 | * The caller might hold locks which can prevent IO completion |
| 671 | * or progress in the filesystem. So we cannot just sit here |
| 672 | * waiting for IO to complete. |
| 673 | */ |
| 674 | if ((gfp_mask & (__GFP_FS|__GFP_IO)) != (__GFP_FS|__GFP_IO)) |
| 675 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 676 | } |
| 677 | } |
| 678 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 679 | static void laptop_timer_fn(unsigned long unused); |
| 680 | |
Ingo Molnar | 8d06afa | 2005-09-09 13:10:40 -0700 | [diff] [blame] | 681 | static DEFINE_TIMER(laptop_mode_wb_timer, laptop_timer_fn, 0, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 682 | |
| 683 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 684 | * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs |
| 685 | */ |
| 686 | int dirty_writeback_centisecs_handler(ctl_table *table, int write, |
Andrew Morton | 3e733f0 | 2007-07-15 23:41:05 -0700 | [diff] [blame] | 687 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 688 | { |
Alexey Dobriyan | 704503d | 2009-03-31 15:23:18 -0700 | [diff] [blame] | 689 | proc_dointvec(table, write, file, buffer, length, ppos); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 690 | return 0; |
| 691 | } |
| 692 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 693 | static void do_laptop_sync(struct work_struct *work) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 694 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 695 | wakeup_flusher_threads(0); |
| 696 | kfree(work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 697 | } |
| 698 | |
| 699 | static void laptop_timer_fn(unsigned long unused) |
| 700 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 701 | struct work_struct *work; |
| 702 | |
| 703 | work = kmalloc(sizeof(*work), GFP_ATOMIC); |
| 704 | if (work) { |
| 705 | INIT_WORK(work, do_laptop_sync); |
| 706 | schedule_work(work); |
| 707 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 708 | } |
| 709 | |
| 710 | /* |
| 711 | * We've spun up the disk and we're in laptop mode: schedule writeback |
| 712 | * of all dirty data a few seconds from now. If the flush is already scheduled |
| 713 | * then push it back - the user is still using the disk. |
| 714 | */ |
| 715 | void laptop_io_completion(void) |
| 716 | { |
Bart Samwel | ed5b43f | 2006-03-24 03:15:49 -0800 | [diff] [blame] | 717 | mod_timer(&laptop_mode_wb_timer, jiffies + laptop_mode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 718 | } |
| 719 | |
| 720 | /* |
| 721 | * We're in laptop mode and we've just synced. The sync's writes will have |
| 722 | * caused another writeback to be scheduled by laptop_io_completion. |
| 723 | * Nothing needs to be written back anymore, so we unschedule the writeback. |
| 724 | */ |
| 725 | void laptop_sync_completion(void) |
| 726 | { |
| 727 | del_timer(&laptop_mode_wb_timer); |
| 728 | } |
| 729 | |
| 730 | /* |
| 731 | * If ratelimit_pages is too high then we can get into dirty-data overload |
| 732 | * if a large number of processes all perform writes at the same time. |
| 733 | * If it is too low then SMP machines will call the (expensive) |
| 734 | * get_writeback_state too often. |
| 735 | * |
| 736 | * Here we set ratelimit_pages to a level which ensures that when all CPUs are |
| 737 | * dirtying in parallel, we cannot go more than 3% (1/32) over the dirty memory |
| 738 | * thresholds before writeback cuts in. |
| 739 | * |
| 740 | * But the limit should not be set too high. Because it also controls the |
| 741 | * amount of memory which the balance_dirty_pages() caller has to write back. |
| 742 | * If this is too large then the caller will block on the IO queue all the |
| 743 | * time. So limit it to four megabytes - the balance_dirty_pages() caller |
| 744 | * will write six megabyte chunks, max. |
| 745 | */ |
| 746 | |
Chandra Seetharaman | 2d1d43f | 2006-09-29 02:01:25 -0700 | [diff] [blame] | 747 | void writeback_set_ratelimit(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 748 | { |
Chandra Seetharaman | 40c99aa | 2006-09-29 02:01:24 -0700 | [diff] [blame] | 749 | ratelimit_pages = vm_total_pages / (num_online_cpus() * 32); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 750 | if (ratelimit_pages < 16) |
| 751 | ratelimit_pages = 16; |
| 752 | if (ratelimit_pages * PAGE_CACHE_SIZE > 4096 * 1024) |
| 753 | ratelimit_pages = (4096 * 1024) / PAGE_CACHE_SIZE; |
| 754 | } |
| 755 | |
Chandra Seetharaman | 26c2143 | 2006-06-27 02:54:10 -0700 | [diff] [blame] | 756 | static int __cpuinit |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 757 | ratelimit_handler(struct notifier_block *self, unsigned long u, void *v) |
| 758 | { |
Chandra Seetharaman | 2d1d43f | 2006-09-29 02:01:25 -0700 | [diff] [blame] | 759 | writeback_set_ratelimit(); |
Paul E. McKenney | aa0f030 | 2007-02-10 01:46:37 -0800 | [diff] [blame] | 760 | return NOTIFY_DONE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 761 | } |
| 762 | |
Chandra Seetharaman | 74b85f3 | 2006-06-27 02:54:09 -0700 | [diff] [blame] | 763 | static struct notifier_block __cpuinitdata ratelimit_nb = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 764 | .notifier_call = ratelimit_handler, |
| 765 | .next = NULL, |
| 766 | }; |
| 767 | |
| 768 | /* |
Linus Torvalds | dc6e29d | 2007-01-29 16:37:38 -0800 | [diff] [blame] | 769 | * Called early on to tune the page writeback dirty limits. |
| 770 | * |
| 771 | * We used to scale dirty pages according to how total memory |
| 772 | * related to pages that could be allocated for buffers (by |
| 773 | * comparing nr_free_buffer_pages() to vm_total_pages. |
| 774 | * |
| 775 | * However, that was when we used "dirty_ratio" to scale with |
| 776 | * all memory, and we don't do that any more. "dirty_ratio" |
| 777 | * is now applied to total non-HIGHPAGE memory (by subtracting |
| 778 | * totalhigh_pages from vm_total_pages), and as such we can't |
| 779 | * get into the old insane situation any more where we had |
| 780 | * large amounts of dirty pages compared to a small amount of |
| 781 | * non-HIGHMEM memory. |
| 782 | * |
| 783 | * But we might still want to scale the dirty_ratio by how |
| 784 | * much memory the box has.. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 785 | */ |
| 786 | void __init page_writeback_init(void) |
| 787 | { |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 788 | int shift; |
| 789 | |
Chandra Seetharaman | 2d1d43f | 2006-09-29 02:01:25 -0700 | [diff] [blame] | 790 | writeback_set_ratelimit(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 791 | register_cpu_notifier(&ratelimit_nb); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 792 | |
| 793 | shift = calc_period_shift(); |
| 794 | prop_descriptor_init(&vm_completions, shift); |
Peter Zijlstra | 3e26c14 | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 795 | prop_descriptor_init(&vm_dirties, shift); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 796 | } |
| 797 | |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 798 | /** |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 799 | * 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] | 800 | * @mapping: address space structure to write |
| 801 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 802 | * @writepage: function called for each page |
| 803 | * @data: data passed to writepage function |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 804 | * |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 805 | * 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] | 806 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, |
| 807 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() |
| 808 | * and msync() need to guarantee that all the data which was dirty at the time |
| 809 | * the call was made get new I/O started against them. If wbc->sync_mode is |
| 810 | * WB_SYNC_ALL then we were called for data integrity and we must wait for |
| 811 | * existing IO to complete. |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 812 | */ |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 813 | int write_cache_pages(struct address_space *mapping, |
| 814 | struct writeback_control *wbc, writepage_t writepage, |
| 815 | void *data) |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 816 | { |
| 817 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
| 818 | int ret = 0; |
| 819 | int done = 0; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 820 | struct pagevec pvec; |
| 821 | int nr_pages; |
Nick Piggin | 31a1266 | 2009-01-06 14:39:04 -0800 | [diff] [blame] | 822 | pgoff_t uninitialized_var(writeback_index); |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 823 | pgoff_t index; |
| 824 | pgoff_t end; /* Inclusive */ |
Nick Piggin | bd19e01 | 2009-01-06 14:39:06 -0800 | [diff] [blame] | 825 | pgoff_t done_index; |
Nick Piggin | 31a1266 | 2009-01-06 14:39:04 -0800 | [diff] [blame] | 826 | int cycled; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 827 | int range_whole = 0; |
Aneesh Kumar K.V | 17bc6c3 | 2008-10-16 10:09:17 -0400 | [diff] [blame] | 828 | long nr_to_write = wbc->nr_to_write; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 829 | |
| 830 | if (wbc->nonblocking && bdi_write_congested(bdi)) { |
| 831 | wbc->encountered_congestion = 1; |
| 832 | return 0; |
| 833 | } |
| 834 | |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 835 | pagevec_init(&pvec, 0); |
| 836 | if (wbc->range_cyclic) { |
Nick Piggin | 31a1266 | 2009-01-06 14:39:04 -0800 | [diff] [blame] | 837 | writeback_index = mapping->writeback_index; /* prev offset */ |
| 838 | index = writeback_index; |
| 839 | if (index == 0) |
| 840 | cycled = 1; |
| 841 | else |
| 842 | cycled = 0; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 843 | end = -1; |
| 844 | } else { |
| 845 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
| 846 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
| 847 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
| 848 | range_whole = 1; |
Nick Piggin | 31a1266 | 2009-01-06 14:39:04 -0800 | [diff] [blame] | 849 | cycled = 1; /* ignore range_cyclic tests */ |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 850 | } |
| 851 | retry: |
Nick Piggin | bd19e01 | 2009-01-06 14:39:06 -0800 | [diff] [blame] | 852 | done_index = index; |
Nick Piggin | 5a3d5c9 | 2009-01-06 14:39:09 -0800 | [diff] [blame] | 853 | while (!done && (index <= end)) { |
| 854 | int i; |
| 855 | |
| 856 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, |
| 857 | PAGECACHE_TAG_DIRTY, |
| 858 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
| 859 | if (nr_pages == 0) |
| 860 | break; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 861 | |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 862 | for (i = 0; i < nr_pages; i++) { |
| 863 | struct page *page = pvec.pages[i]; |
| 864 | |
Nick Piggin | d5482cd | 2009-01-06 14:39:11 -0800 | [diff] [blame] | 865 | /* |
| 866 | * At this point, the page may be truncated or |
| 867 | * invalidated (changing page->mapping to NULL), or |
| 868 | * even swizzled back from swapper_space to tmpfs file |
| 869 | * mapping. However, page->index will not change |
| 870 | * because we have a reference on the page. |
| 871 | */ |
| 872 | if (page->index > end) { |
| 873 | /* |
| 874 | * can't be range_cyclic (1st pass) because |
| 875 | * end == -1 in that case. |
| 876 | */ |
| 877 | done = 1; |
| 878 | break; |
| 879 | } |
| 880 | |
Nick Piggin | bd19e01 | 2009-01-06 14:39:06 -0800 | [diff] [blame] | 881 | done_index = page->index + 1; |
| 882 | |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 883 | lock_page(page); |
| 884 | |
Nick Piggin | 5a3d5c9 | 2009-01-06 14:39:09 -0800 | [diff] [blame] | 885 | /* |
| 886 | * Page truncated or invalidated. We can freely skip it |
| 887 | * then, even for data integrity operations: the page |
| 888 | * has disappeared concurrently, so there could be no |
| 889 | * real expectation of this data interity operation |
| 890 | * even if there is now a new, dirty page at the same |
| 891 | * pagecache address. |
| 892 | */ |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 893 | if (unlikely(page->mapping != mapping)) { |
Nick Piggin | 5a3d5c9 | 2009-01-06 14:39:09 -0800 | [diff] [blame] | 894 | continue_unlock: |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 895 | unlock_page(page); |
| 896 | continue; |
| 897 | } |
| 898 | |
Nick Piggin | 515f4a0 | 2009-01-06 14:39:10 -0800 | [diff] [blame] | 899 | if (!PageDirty(page)) { |
| 900 | /* someone wrote it for us */ |
| 901 | goto continue_unlock; |
| 902 | } |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 903 | |
Nick Piggin | 515f4a0 | 2009-01-06 14:39:10 -0800 | [diff] [blame] | 904 | if (PageWriteback(page)) { |
| 905 | if (wbc->sync_mode != WB_SYNC_NONE) |
| 906 | wait_on_page_writeback(page); |
| 907 | else |
| 908 | goto continue_unlock; |
| 909 | } |
| 910 | |
| 911 | BUG_ON(PageWriteback(page)); |
| 912 | if (!clear_page_dirty_for_io(page)) |
Nick Piggin | 5a3d5c9 | 2009-01-06 14:39:09 -0800 | [diff] [blame] | 913 | goto continue_unlock; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 914 | |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 915 | ret = (*writepage)(page, wbc, data); |
Nick Piggin | 0026677 | 2009-01-06 14:39:06 -0800 | [diff] [blame] | 916 | if (unlikely(ret)) { |
| 917 | if (ret == AOP_WRITEPAGE_ACTIVATE) { |
| 918 | unlock_page(page); |
| 919 | ret = 0; |
| 920 | } else { |
| 921 | /* |
| 922 | * done_index is set past this page, |
| 923 | * so media errors will not choke |
| 924 | * background writeout for the entire |
| 925 | * file. This has consequences for |
| 926 | * range_cyclic semantics (ie. it may |
| 927 | * not be suitable for data integrity |
| 928 | * writeout). |
| 929 | */ |
| 930 | done = 1; |
| 931 | break; |
| 932 | } |
| 933 | } |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 934 | |
Federico Cuello | 89e1219 | 2009-02-11 13:04:39 -0800 | [diff] [blame] | 935 | if (nr_to_write > 0) { |
Artem Bityutskiy | dcf6a79 | 2009-02-02 18:33:49 +0200 | [diff] [blame] | 936 | nr_to_write--; |
Federico Cuello | 89e1219 | 2009-02-11 13:04:39 -0800 | [diff] [blame] | 937 | if (nr_to_write == 0 && |
| 938 | wbc->sync_mode == WB_SYNC_NONE) { |
| 939 | /* |
| 940 | * We stop writing back only if we are |
| 941 | * not doing integrity sync. In case of |
| 942 | * integrity sync we have to keep going |
| 943 | * because someone may be concurrently |
| 944 | * dirtying pages, and we might have |
| 945 | * synced a lot of newly appeared dirty |
| 946 | * pages, but have not synced all of the |
| 947 | * old dirty pages. |
| 948 | */ |
| 949 | done = 1; |
| 950 | break; |
| 951 | } |
Nick Piggin | 05fe478 | 2009-01-06 14:39:08 -0800 | [diff] [blame] | 952 | } |
Artem Bityutskiy | dcf6a79 | 2009-02-02 18:33:49 +0200 | [diff] [blame] | 953 | |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 954 | if (wbc->nonblocking && bdi_write_congested(bdi)) { |
| 955 | wbc->encountered_congestion = 1; |
| 956 | done = 1; |
Andrew Morton | 82fd1a9 | 2009-01-06 14:39:11 -0800 | [diff] [blame] | 957 | break; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 958 | } |
| 959 | } |
| 960 | pagevec_release(&pvec); |
| 961 | cond_resched(); |
| 962 | } |
Nick Piggin | 3a4c680 | 2009-02-12 04:34:23 +0100 | [diff] [blame] | 963 | if (!cycled && !done) { |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 964 | /* |
Nick Piggin | 31a1266 | 2009-01-06 14:39:04 -0800 | [diff] [blame] | 965 | * range_cyclic: |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 966 | * We hit the last page and there is more work to be done: wrap |
| 967 | * back to the start of the file |
| 968 | */ |
Nick Piggin | 31a1266 | 2009-01-06 14:39:04 -0800 | [diff] [blame] | 969 | cycled = 1; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 970 | index = 0; |
Nick Piggin | 31a1266 | 2009-01-06 14:39:04 -0800 | [diff] [blame] | 971 | end = writeback_index - 1; |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 972 | goto retry; |
| 973 | } |
Aneesh Kumar K.V | 17bc6c3 | 2008-10-16 10:09:17 -0400 | [diff] [blame] | 974 | if (!wbc->no_nrwrite_index_update) { |
| 975 | if (wbc->range_cyclic || (range_whole && nr_to_write > 0)) |
Nick Piggin | bd19e01 | 2009-01-06 14:39:06 -0800 | [diff] [blame] | 976 | mapping->writeback_index = done_index; |
Aneesh Kumar K.V | 17bc6c3 | 2008-10-16 10:09:17 -0400 | [diff] [blame] | 977 | wbc->nr_to_write = nr_to_write; |
| 978 | } |
Aneesh Kumar K.V | 06d6cf6 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 979 | |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 980 | return ret; |
| 981 | } |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 982 | EXPORT_SYMBOL(write_cache_pages); |
| 983 | |
| 984 | /* |
| 985 | * Function used by generic_writepages to call the real writepage |
| 986 | * function and set the mapping flags on error |
| 987 | */ |
| 988 | static int __writepage(struct page *page, struct writeback_control *wbc, |
| 989 | void *data) |
| 990 | { |
| 991 | struct address_space *mapping = data; |
| 992 | int ret = mapping->a_ops->writepage(page, wbc); |
| 993 | mapping_set_error(mapping, ret); |
| 994 | return ret; |
| 995 | } |
| 996 | |
| 997 | /** |
| 998 | * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them. |
| 999 | * @mapping: address space structure to write |
| 1000 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write |
| 1001 | * |
| 1002 | * This is a library function, which implements the writepages() |
| 1003 | * address_space_operation. |
| 1004 | */ |
| 1005 | int generic_writepages(struct address_space *mapping, |
| 1006 | struct writeback_control *wbc) |
| 1007 | { |
| 1008 | /* deal with chardevs and other special file */ |
| 1009 | if (!mapping->a_ops->writepage) |
| 1010 | return 0; |
| 1011 | |
| 1012 | return write_cache_pages(mapping, wbc, __writepage, mapping); |
| 1013 | } |
David Howells | 811d736 | 2006-08-29 19:06:09 +0100 | [diff] [blame] | 1014 | |
| 1015 | EXPORT_SYMBOL(generic_writepages); |
| 1016 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1017 | int do_writepages(struct address_space *mapping, struct writeback_control *wbc) |
| 1018 | { |
Andrew Morton | 22905f7 | 2005-11-16 15:07:01 -0800 | [diff] [blame] | 1019 | int ret; |
| 1020 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1021 | if (wbc->nr_to_write <= 0) |
| 1022 | return 0; |
Andrew Morton | 22905f7 | 2005-11-16 15:07:01 -0800 | [diff] [blame] | 1023 | wbc->for_writepages = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1024 | if (mapping->a_ops->writepages) |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 1025 | ret = mapping->a_ops->writepages(mapping, wbc); |
Andrew Morton | 22905f7 | 2005-11-16 15:07:01 -0800 | [diff] [blame] | 1026 | else |
| 1027 | ret = generic_writepages(mapping, wbc); |
| 1028 | wbc->for_writepages = 0; |
| 1029 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1030 | } |
| 1031 | |
| 1032 | /** |
| 1033 | * 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] | 1034 | * @page: the page to write |
| 1035 | * @wait: if true, wait on writeout |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1036 | * |
| 1037 | * The page must be locked by the caller and will be unlocked upon return. |
| 1038 | * |
| 1039 | * write_one_page() returns a negative error code if I/O failed. |
| 1040 | */ |
| 1041 | int write_one_page(struct page *page, int wait) |
| 1042 | { |
| 1043 | struct address_space *mapping = page->mapping; |
| 1044 | int ret = 0; |
| 1045 | struct writeback_control wbc = { |
| 1046 | .sync_mode = WB_SYNC_ALL, |
| 1047 | .nr_to_write = 1, |
| 1048 | }; |
| 1049 | |
| 1050 | BUG_ON(!PageLocked(page)); |
| 1051 | |
| 1052 | if (wait) |
| 1053 | wait_on_page_writeback(page); |
| 1054 | |
| 1055 | if (clear_page_dirty_for_io(page)) { |
| 1056 | page_cache_get(page); |
| 1057 | ret = mapping->a_ops->writepage(page, &wbc); |
| 1058 | if (ret == 0 && wait) { |
| 1059 | wait_on_page_writeback(page); |
| 1060 | if (PageError(page)) |
| 1061 | ret = -EIO; |
| 1062 | } |
| 1063 | page_cache_release(page); |
| 1064 | } else { |
| 1065 | unlock_page(page); |
| 1066 | } |
| 1067 | return ret; |
| 1068 | } |
| 1069 | EXPORT_SYMBOL(write_one_page); |
| 1070 | |
| 1071 | /* |
Ken Chen | 7671932 | 2007-02-10 01:43:15 -0800 | [diff] [blame] | 1072 | * For address_spaces which do not use buffers nor write back. |
| 1073 | */ |
| 1074 | int __set_page_dirty_no_writeback(struct page *page) |
| 1075 | { |
| 1076 | if (!PageDirty(page)) |
| 1077 | SetPageDirty(page); |
| 1078 | return 0; |
| 1079 | } |
| 1080 | |
| 1081 | /* |
Edward Shishkin | e3a7cca | 2009-03-31 15:19:39 -0700 | [diff] [blame] | 1082 | * Helper function for set_page_dirty family. |
| 1083 | * NOTE: This relies on being atomic wrt interrupts. |
| 1084 | */ |
| 1085 | void account_page_dirtied(struct page *page, struct address_space *mapping) |
| 1086 | { |
| 1087 | if (mapping_cap_account_dirty(mapping)) { |
| 1088 | __inc_zone_page_state(page, NR_FILE_DIRTY); |
| 1089 | __inc_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); |
| 1090 | task_dirty_inc(current); |
| 1091 | task_io_account_write(PAGE_CACHE_SIZE); |
| 1092 | } |
| 1093 | } |
| 1094 | |
| 1095 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1096 | * For address_spaces which do not use buffers. Just tag the page as dirty in |
| 1097 | * its radix tree. |
| 1098 | * |
| 1099 | * This is also used when a single buffer is being dirtied: we want to set the |
| 1100 | * page dirty in that case, but not all the buffers. This is a "bottom-up" |
| 1101 | * dirtying, whereas __set_page_dirty_buffers() is a "top-down" dirtying. |
| 1102 | * |
| 1103 | * Most callers have locked the page, which pins the address_space in memory. |
| 1104 | * But zap_pte_range() does not lock the page, however in that case the |
| 1105 | * mapping is pinned by the vma's ->vm_file reference. |
| 1106 | * |
| 1107 | * 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] | 1108 | * mapping by re-checking page_mapping() inside tree_lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1109 | */ |
| 1110 | int __set_page_dirty_nobuffers(struct page *page) |
| 1111 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1112 | if (!TestSetPageDirty(page)) { |
| 1113 | struct address_space *mapping = page_mapping(page); |
| 1114 | struct address_space *mapping2; |
| 1115 | |
Andrew Morton | 8c08540 | 2006-12-10 02:19:24 -0800 | [diff] [blame] | 1116 | if (!mapping) |
| 1117 | return 1; |
| 1118 | |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 1119 | spin_lock_irq(&mapping->tree_lock); |
Andrew Morton | 8c08540 | 2006-12-10 02:19:24 -0800 | [diff] [blame] | 1120 | mapping2 = page_mapping(page); |
| 1121 | if (mapping2) { /* Race with truncate? */ |
| 1122 | BUG_ON(mapping2 != mapping); |
Nick Piggin | 787d221 | 2007-07-17 04:03:34 -0700 | [diff] [blame] | 1123 | WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page)); |
Edward Shishkin | e3a7cca | 2009-03-31 15:19:39 -0700 | [diff] [blame] | 1124 | account_page_dirtied(page, mapping); |
Andrew Morton | 8c08540 | 2006-12-10 02:19:24 -0800 | [diff] [blame] | 1125 | radix_tree_tag_set(&mapping->page_tree, |
| 1126 | page_index(page), PAGECACHE_TAG_DIRTY); |
| 1127 | } |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 1128 | spin_unlock_irq(&mapping->tree_lock); |
Andrew Morton | 8c08540 | 2006-12-10 02:19:24 -0800 | [diff] [blame] | 1129 | if (mapping->host) { |
| 1130 | /* !PageAnon && !swapper_space */ |
| 1131 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1132 | } |
Andrew Morton | 4741c9f | 2006-03-24 03:18:11 -0800 | [diff] [blame] | 1133 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1134 | } |
Andrew Morton | 4741c9f | 2006-03-24 03:18:11 -0800 | [diff] [blame] | 1135 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1136 | } |
| 1137 | EXPORT_SYMBOL(__set_page_dirty_nobuffers); |
| 1138 | |
| 1139 | /* |
| 1140 | * When a writepage implementation decides that it doesn't want to write this |
| 1141 | * page for some reason, it should redirty the locked page via |
| 1142 | * redirty_page_for_writepage() and it should then unlock the page and return 0 |
| 1143 | */ |
| 1144 | int redirty_page_for_writepage(struct writeback_control *wbc, struct page *page) |
| 1145 | { |
| 1146 | wbc->pages_skipped++; |
| 1147 | return __set_page_dirty_nobuffers(page); |
| 1148 | } |
| 1149 | EXPORT_SYMBOL(redirty_page_for_writepage); |
| 1150 | |
| 1151 | /* |
| 1152 | * If the mapping doesn't provide a set_page_dirty a_op, then |
| 1153 | * just fall through and assume that it wants buffer_heads. |
| 1154 | */ |
Nick Piggin | 1cf6e7d | 2009-02-18 14:48:18 -0800 | [diff] [blame] | 1155 | int set_page_dirty(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1156 | { |
| 1157 | struct address_space *mapping = page_mapping(page); |
| 1158 | |
| 1159 | if (likely(mapping)) { |
| 1160 | int (*spd)(struct page *) = mapping->a_ops->set_page_dirty; |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 1161 | #ifdef CONFIG_BLOCK |
| 1162 | if (!spd) |
| 1163 | spd = __set_page_dirty_buffers; |
| 1164 | #endif |
| 1165 | return (*spd)(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1166 | } |
Andrew Morton | 4741c9f | 2006-03-24 03:18:11 -0800 | [diff] [blame] | 1167 | if (!PageDirty(page)) { |
| 1168 | if (!TestSetPageDirty(page)) |
| 1169 | return 1; |
| 1170 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1171 | return 0; |
| 1172 | } |
| 1173 | EXPORT_SYMBOL(set_page_dirty); |
| 1174 | |
| 1175 | /* |
| 1176 | * set_page_dirty() is racy if the caller has no reference against |
| 1177 | * page->mapping->host, and if the page is unlocked. This is because another |
| 1178 | * CPU could truncate the page off the mapping and then free the mapping. |
| 1179 | * |
| 1180 | * Usually, the page _is_ locked, or the caller is a user-space process which |
| 1181 | * holds a reference on the inode by having an open file. |
| 1182 | * |
| 1183 | * In other cases, the page should be locked before running set_page_dirty(). |
| 1184 | */ |
| 1185 | int set_page_dirty_lock(struct page *page) |
| 1186 | { |
| 1187 | int ret; |
| 1188 | |
Nick Piggin | db37648 | 2006-09-25 23:31:24 -0700 | [diff] [blame] | 1189 | lock_page_nosync(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1190 | ret = set_page_dirty(page); |
| 1191 | unlock_page(page); |
| 1192 | return ret; |
| 1193 | } |
| 1194 | EXPORT_SYMBOL(set_page_dirty_lock); |
| 1195 | |
| 1196 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1197 | * Clear a page's dirty flag, while caring for dirty memory accounting. |
| 1198 | * Returns true if the page was previously dirty. |
| 1199 | * |
| 1200 | * This is for preparing to put the page under writeout. We leave the page |
| 1201 | * tagged as dirty in the radix tree so that a concurrent write-for-sync |
| 1202 | * can discover it via a PAGECACHE_TAG_DIRTY walk. The ->writepage |
| 1203 | * implementation will run either set_page_writeback() or set_page_dirty(), |
| 1204 | * at which stage we bring the page's dirty flag and radix-tree dirty tag |
| 1205 | * back into sync. |
| 1206 | * |
| 1207 | * This incoherency between the page's dirty flag and radix-tree tag is |
| 1208 | * unfortunate, but it only exists while the page is locked. |
| 1209 | */ |
| 1210 | int clear_page_dirty_for_io(struct page *page) |
| 1211 | { |
| 1212 | struct address_space *mapping = page_mapping(page); |
| 1213 | |
Nick Piggin | 7935289 | 2007-07-19 01:47:22 -0700 | [diff] [blame] | 1214 | BUG_ON(!PageLocked(page)); |
| 1215 | |
Fengguang Wu | fe3cba1 | 2007-07-19 01:48:07 -0700 | [diff] [blame] | 1216 | ClearPageReclaim(page); |
Linus Torvalds | 7658cc2 | 2006-12-29 10:00:58 -0800 | [diff] [blame] | 1217 | if (mapping && mapping_cap_account_dirty(mapping)) { |
| 1218 | /* |
| 1219 | * Yes, Virginia, this is indeed insane. |
| 1220 | * |
| 1221 | * We use this sequence to make sure that |
| 1222 | * (a) we account for dirty stats properly |
| 1223 | * (b) we tell the low-level filesystem to |
| 1224 | * mark the whole page dirty if it was |
| 1225 | * dirty in a pagetable. Only to then |
| 1226 | * (c) clean the page again and return 1 to |
| 1227 | * cause the writeback. |
| 1228 | * |
| 1229 | * This way we avoid all nasty races with the |
| 1230 | * dirty bit in multiple places and clearing |
| 1231 | * them concurrently from different threads. |
| 1232 | * |
| 1233 | * Note! Normally the "set_page_dirty(page)" |
| 1234 | * has no effect on the actual dirty bit - since |
| 1235 | * that will already usually be set. But we |
| 1236 | * need the side effects, and it can help us |
| 1237 | * avoid races. |
| 1238 | * |
| 1239 | * We basically use the page "master dirty bit" |
| 1240 | * as a serialization point for all the different |
| 1241 | * threads doing their things. |
Linus Torvalds | 7658cc2 | 2006-12-29 10:00:58 -0800 | [diff] [blame] | 1242 | */ |
| 1243 | if (page_mkclean(page)) |
| 1244 | set_page_dirty(page); |
Nick Piggin | 7935289 | 2007-07-19 01:47:22 -0700 | [diff] [blame] | 1245 | /* |
| 1246 | * We carefully synchronise fault handlers against |
| 1247 | * installing a dirty pte and marking the page dirty |
| 1248 | * at this point. We do this by having them hold the |
| 1249 | * page lock at some point after installing their |
| 1250 | * pte, but before marking the page dirty. |
| 1251 | * Pages are always locked coming in here, so we get |
| 1252 | * the desired exclusion. See mm/memory.c:do_wp_page() |
| 1253 | * for more comments. |
| 1254 | */ |
Linus Torvalds | 7658cc2 | 2006-12-29 10:00:58 -0800 | [diff] [blame] | 1255 | if (TestClearPageDirty(page)) { |
Andrew Morton | 8c08540 | 2006-12-10 02:19:24 -0800 | [diff] [blame] | 1256 | dec_zone_page_state(page, NR_FILE_DIRTY); |
Peter Zijlstra | c9e51e4 | 2007-10-16 23:25:47 -0700 | [diff] [blame] | 1257 | dec_bdi_stat(mapping->backing_dev_info, |
| 1258 | BDI_RECLAIMABLE); |
Linus Torvalds | 7658cc2 | 2006-12-29 10:00:58 -0800 | [diff] [blame] | 1259 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1260 | } |
Linus Torvalds | 7658cc2 | 2006-12-29 10:00:58 -0800 | [diff] [blame] | 1261 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1262 | } |
Linus Torvalds | 7658cc2 | 2006-12-29 10:00:58 -0800 | [diff] [blame] | 1263 | return TestClearPageDirty(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1264 | } |
Hans Reiser | 58bb01a | 2005-11-18 01:10:53 -0800 | [diff] [blame] | 1265 | EXPORT_SYMBOL(clear_page_dirty_for_io); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1266 | |
| 1267 | int test_clear_page_writeback(struct page *page) |
| 1268 | { |
| 1269 | struct address_space *mapping = page_mapping(page); |
| 1270 | int ret; |
| 1271 | |
| 1272 | if (mapping) { |
Peter Zijlstra | 69cb51d | 2007-10-16 23:25:48 -0700 | [diff] [blame] | 1273 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1274 | unsigned long flags; |
| 1275 | |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 1276 | spin_lock_irqsave(&mapping->tree_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1277 | ret = TestClearPageWriteback(page); |
Peter Zijlstra | 69cb51d | 2007-10-16 23:25:48 -0700 | [diff] [blame] | 1278 | if (ret) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1279 | radix_tree_tag_clear(&mapping->page_tree, |
| 1280 | page_index(page), |
| 1281 | PAGECACHE_TAG_WRITEBACK); |
Miklos Szeredi | e4ad08f | 2008-04-30 00:54:37 -0700 | [diff] [blame] | 1282 | if (bdi_cap_account_writeback(bdi)) { |
Peter Zijlstra | 69cb51d | 2007-10-16 23:25:48 -0700 | [diff] [blame] | 1283 | __dec_bdi_stat(bdi, BDI_WRITEBACK); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 1284 | __bdi_writeout_inc(bdi); |
| 1285 | } |
Peter Zijlstra | 69cb51d | 2007-10-16 23:25:48 -0700 | [diff] [blame] | 1286 | } |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 1287 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1288 | } else { |
| 1289 | ret = TestClearPageWriteback(page); |
| 1290 | } |
Andrew Morton | d688abf | 2007-07-19 01:49:17 -0700 | [diff] [blame] | 1291 | if (ret) |
| 1292 | dec_zone_page_state(page, NR_WRITEBACK); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1293 | return ret; |
| 1294 | } |
| 1295 | |
| 1296 | int test_set_page_writeback(struct page *page) |
| 1297 | { |
| 1298 | struct address_space *mapping = page_mapping(page); |
| 1299 | int ret; |
| 1300 | |
| 1301 | if (mapping) { |
Peter Zijlstra | 69cb51d | 2007-10-16 23:25:48 -0700 | [diff] [blame] | 1302 | struct backing_dev_info *bdi = mapping->backing_dev_info; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1303 | unsigned long flags; |
| 1304 | |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 1305 | spin_lock_irqsave(&mapping->tree_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1306 | ret = TestSetPageWriteback(page); |
Peter Zijlstra | 69cb51d | 2007-10-16 23:25:48 -0700 | [diff] [blame] | 1307 | if (!ret) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1308 | radix_tree_tag_set(&mapping->page_tree, |
| 1309 | page_index(page), |
| 1310 | PAGECACHE_TAG_WRITEBACK); |
Miklos Szeredi | e4ad08f | 2008-04-30 00:54:37 -0700 | [diff] [blame] | 1311 | if (bdi_cap_account_writeback(bdi)) |
Peter Zijlstra | 69cb51d | 2007-10-16 23:25:48 -0700 | [diff] [blame] | 1312 | __inc_bdi_stat(bdi, BDI_WRITEBACK); |
| 1313 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1314 | if (!PageDirty(page)) |
| 1315 | radix_tree_tag_clear(&mapping->page_tree, |
| 1316 | page_index(page), |
| 1317 | PAGECACHE_TAG_DIRTY); |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 1318 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1319 | } else { |
| 1320 | ret = TestSetPageWriteback(page); |
| 1321 | } |
Andrew Morton | d688abf | 2007-07-19 01:49:17 -0700 | [diff] [blame] | 1322 | if (!ret) |
| 1323 | inc_zone_page_state(page, NR_WRITEBACK); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1324 | return ret; |
| 1325 | |
| 1326 | } |
| 1327 | EXPORT_SYMBOL(test_set_page_writeback); |
| 1328 | |
| 1329 | /* |
Nick Piggin | 0012818 | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 1330 | * 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] | 1331 | * passed tag. |
| 1332 | */ |
| 1333 | int mapping_tagged(struct address_space *mapping, int tag) |
| 1334 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1335 | int ret; |
Nick Piggin | 0012818 | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 1336 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1337 | ret = radix_tree_tagged(&mapping->page_tree, tag); |
Nick Piggin | 0012818 | 2007-10-16 01:24:40 -0700 | [diff] [blame] | 1338 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1339 | return ret; |
| 1340 | } |
| 1341 | EXPORT_SYMBOL(mapping_tagged); |