Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 1 | |
| 2 | #include <linux/wait.h> |
| 3 | #include <linux/backing-dev.h> |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 4 | #include <linux/kthread.h> |
| 5 | #include <linux/freezer.h> |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 6 | #include <linux/fs.h> |
Jens Axboe | 2616015 | 2009-03-17 09:35:06 +0100 | [diff] [blame] | 7 | #include <linux/pagemap.h> |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 8 | #include <linux/mm.h> |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 9 | #include <linux/sched.h> |
| 10 | #include <linux/module.h> |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 11 | #include <linux/writeback.h> |
| 12 | #include <linux/device.h> |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 13 | #include <trace/events/writeback.h> |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 14 | |
Jens Axboe | c3c5320 | 2010-04-22 11:37:01 +0200 | [diff] [blame] | 15 | static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0); |
| 16 | |
Jens Axboe | 2616015 | 2009-03-17 09:35:06 +0100 | [diff] [blame] | 17 | struct backing_dev_info default_backing_dev_info = { |
Jens Axboe | d993831 | 2009-06-12 14:45:52 +0200 | [diff] [blame] | 18 | .name = "default", |
Jens Axboe | 2616015 | 2009-03-17 09:35:06 +0100 | [diff] [blame] | 19 | .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE, |
| 20 | .state = 0, |
| 21 | .capabilities = BDI_CAP_MAP_COPY, |
Jens Axboe | 2616015 | 2009-03-17 09:35:06 +0100 | [diff] [blame] | 22 | }; |
| 23 | EXPORT_SYMBOL_GPL(default_backing_dev_info); |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 24 | |
Jörn Engel | 5129a46 | 2010-04-25 08:54:42 +0200 | [diff] [blame] | 25 | struct backing_dev_info noop_backing_dev_info = { |
| 26 | .name = "noop", |
Jan Kara | 976e48f | 2010-09-21 11:48:55 +0200 | [diff] [blame] | 27 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
Jörn Engel | 5129a46 | 2010-04-25 08:54:42 +0200 | [diff] [blame] | 28 | }; |
| 29 | EXPORT_SYMBOL_GPL(noop_backing_dev_info); |
| 30 | |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 31 | static struct class *bdi_class; |
Jens Axboe | cfc4ba5 | 2009-09-14 13:12:40 +0200 | [diff] [blame] | 32 | |
| 33 | /* |
| 34 | * bdi_lock protects updates to bdi_list and bdi_pending_list, as well as |
| 35 | * reader side protection for bdi_pending_list. bdi_list has RCU reader side |
| 36 | * locking. |
| 37 | */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 38 | DEFINE_SPINLOCK(bdi_lock); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 39 | LIST_HEAD(bdi_list); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 40 | LIST_HEAD(bdi_pending_list); |
| 41 | |
| 42 | static struct task_struct *sync_supers_tsk; |
| 43 | static struct timer_list sync_supers_timer; |
| 44 | |
| 45 | static int bdi_sync_supers(void *); |
| 46 | static void sync_supers_timer_fn(unsigned long); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 47 | |
Miklos Szeredi | 76f1418 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 48 | #ifdef CONFIG_DEBUG_FS |
| 49 | #include <linux/debugfs.h> |
| 50 | #include <linux/seq_file.h> |
| 51 | |
| 52 | static struct dentry *bdi_debug_root; |
| 53 | |
| 54 | static void bdi_debug_init(void) |
| 55 | { |
| 56 | bdi_debug_root = debugfs_create_dir("bdi", NULL); |
| 57 | } |
| 58 | |
| 59 | static int bdi_debug_stats_show(struct seq_file *m, void *v) |
| 60 | { |
| 61 | struct backing_dev_info *bdi = m->private; |
Christoph Hellwig | c1955ce | 2010-06-19 23:08:06 +0200 | [diff] [blame] | 62 | struct bdi_writeback *wb = &bdi->wb; |
David Rientjes | 364aeb2 | 2009-01-06 14:39:29 -0800 | [diff] [blame] | 63 | unsigned long background_thresh; |
| 64 | unsigned long dirty_thresh; |
| 65 | unsigned long bdi_thresh; |
Gustavo F. Padovan | 345227d | 2011-05-20 21:23:37 +0200 | [diff] [blame] | 66 | unsigned long nr_dirty, nr_io, nr_more_io; |
Jens Axboe | f09b00d | 2009-05-25 09:08:21 +0200 | [diff] [blame] | 67 | struct inode *inode; |
| 68 | |
Gustavo F. Padovan | 345227d | 2011-05-20 21:23:37 +0200 | [diff] [blame] | 69 | nr_dirty = nr_io = nr_more_io = 0; |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 70 | spin_lock(&inode_wb_list_lock); |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 71 | list_for_each_entry(inode, &wb->b_dirty, i_wb_list) |
Christoph Hellwig | c1955ce | 2010-06-19 23:08:06 +0200 | [diff] [blame] | 72 | nr_dirty++; |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 73 | list_for_each_entry(inode, &wb->b_io, i_wb_list) |
Christoph Hellwig | c1955ce | 2010-06-19 23:08:06 +0200 | [diff] [blame] | 74 | nr_io++; |
Nick Piggin | 7ccf19a | 2010-10-21 11:49:30 +1100 | [diff] [blame] | 75 | list_for_each_entry(inode, &wb->b_more_io, i_wb_list) |
Christoph Hellwig | c1955ce | 2010-06-19 23:08:06 +0200 | [diff] [blame] | 76 | nr_more_io++; |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 77 | spin_unlock(&inode_wb_list_lock); |
Miklos Szeredi | 76f1418 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 78 | |
Wu Fengguang | 16c4042 | 2010-08-11 14:17:39 -0700 | [diff] [blame] | 79 | global_dirty_limits(&background_thresh, &dirty_thresh); |
| 80 | bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh); |
Miklos Szeredi | 76f1418 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 81 | |
| 82 | #define K(x) ((x) << (PAGE_SHIFT - 10)) |
| 83 | seq_printf(m, |
| 84 | "BdiWriteback: %8lu kB\n" |
| 85 | "BdiReclaimable: %8lu kB\n" |
| 86 | "BdiDirtyThresh: %8lu kB\n" |
| 87 | "DirtyThresh: %8lu kB\n" |
Jens Axboe | f09b00d | 2009-05-25 09:08:21 +0200 | [diff] [blame] | 88 | "BackgroundThresh: %8lu kB\n" |
Jens Axboe | f09b00d | 2009-05-25 09:08:21 +0200 | [diff] [blame] | 89 | "b_dirty: %8lu\n" |
| 90 | "b_io: %8lu\n" |
| 91 | "b_more_io: %8lu\n" |
| 92 | "bdi_list: %8u\n" |
Christoph Hellwig | c1955ce | 2010-06-19 23:08:06 +0200 | [diff] [blame] | 93 | "state: %8lx\n", |
Miklos Szeredi | 76f1418 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 94 | (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)), |
| 95 | (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)), |
Jens Axboe | f09b00d | 2009-05-25 09:08:21 +0200 | [diff] [blame] | 96 | K(bdi_thresh), K(dirty_thresh), |
Christoph Hellwig | c1955ce | 2010-06-19 23:08:06 +0200 | [diff] [blame] | 97 | K(background_thresh), nr_dirty, nr_io, nr_more_io, |
| 98 | !list_empty(&bdi->bdi_list), bdi->state); |
Miklos Szeredi | 76f1418 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 99 | #undef K |
| 100 | |
| 101 | return 0; |
| 102 | } |
| 103 | |
| 104 | static int bdi_debug_stats_open(struct inode *inode, struct file *file) |
| 105 | { |
| 106 | return single_open(file, bdi_debug_stats_show, inode->i_private); |
| 107 | } |
| 108 | |
| 109 | static const struct file_operations bdi_debug_stats_fops = { |
| 110 | .open = bdi_debug_stats_open, |
| 111 | .read = seq_read, |
| 112 | .llseek = seq_lseek, |
| 113 | .release = single_release, |
| 114 | }; |
| 115 | |
| 116 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) |
| 117 | { |
| 118 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); |
| 119 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, |
| 120 | bdi, &bdi_debug_stats_fops); |
| 121 | } |
| 122 | |
| 123 | static void bdi_debug_unregister(struct backing_dev_info *bdi) |
| 124 | { |
| 125 | debugfs_remove(bdi->debug_stats); |
| 126 | debugfs_remove(bdi->debug_dir); |
| 127 | } |
| 128 | #else |
| 129 | static inline void bdi_debug_init(void) |
| 130 | { |
| 131 | } |
| 132 | static inline void bdi_debug_register(struct backing_dev_info *bdi, |
| 133 | const char *name) |
| 134 | { |
| 135 | } |
| 136 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) |
| 137 | { |
| 138 | } |
| 139 | #endif |
| 140 | |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 141 | static ssize_t read_ahead_kb_store(struct device *dev, |
| 142 | struct device_attribute *attr, |
| 143 | const char *buf, size_t count) |
| 144 | { |
| 145 | struct backing_dev_info *bdi = dev_get_drvdata(dev); |
| 146 | char *end; |
| 147 | unsigned long read_ahead_kb; |
| 148 | ssize_t ret = -EINVAL; |
| 149 | |
| 150 | read_ahead_kb = simple_strtoul(buf, &end, 10); |
| 151 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { |
| 152 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); |
| 153 | ret = count; |
| 154 | } |
| 155 | return ret; |
| 156 | } |
| 157 | |
| 158 | #define K(pages) ((pages) << (PAGE_SHIFT - 10)) |
| 159 | |
| 160 | #define BDI_SHOW(name, expr) \ |
| 161 | static ssize_t name##_show(struct device *dev, \ |
| 162 | struct device_attribute *attr, char *page) \ |
| 163 | { \ |
| 164 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ |
| 165 | \ |
| 166 | return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \ |
| 167 | } |
| 168 | |
| 169 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) |
| 170 | |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 171 | static ssize_t min_ratio_store(struct device *dev, |
| 172 | struct device_attribute *attr, const char *buf, size_t count) |
| 173 | { |
| 174 | struct backing_dev_info *bdi = dev_get_drvdata(dev); |
| 175 | char *end; |
| 176 | unsigned int ratio; |
| 177 | ssize_t ret = -EINVAL; |
| 178 | |
| 179 | ratio = simple_strtoul(buf, &end, 10); |
| 180 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { |
| 181 | ret = bdi_set_min_ratio(bdi, ratio); |
| 182 | if (!ret) |
| 183 | ret = count; |
| 184 | } |
| 185 | return ret; |
| 186 | } |
| 187 | BDI_SHOW(min_ratio, bdi->min_ratio) |
| 188 | |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 189 | static ssize_t max_ratio_store(struct device *dev, |
| 190 | struct device_attribute *attr, const char *buf, size_t count) |
| 191 | { |
| 192 | struct backing_dev_info *bdi = dev_get_drvdata(dev); |
| 193 | char *end; |
| 194 | unsigned int ratio; |
| 195 | ssize_t ret = -EINVAL; |
| 196 | |
| 197 | ratio = simple_strtoul(buf, &end, 10); |
| 198 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { |
| 199 | ret = bdi_set_max_ratio(bdi, ratio); |
| 200 | if (!ret) |
| 201 | ret = count; |
| 202 | } |
| 203 | return ret; |
| 204 | } |
| 205 | BDI_SHOW(max_ratio, bdi->max_ratio) |
| 206 | |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 207 | #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store) |
| 208 | |
| 209 | static struct device_attribute bdi_dev_attrs[] = { |
| 210 | __ATTR_RW(read_ahead_kb), |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 211 | __ATTR_RW(min_ratio), |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 212 | __ATTR_RW(max_ratio), |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 213 | __ATTR_NULL, |
| 214 | }; |
| 215 | |
| 216 | static __init int bdi_class_init(void) |
| 217 | { |
| 218 | bdi_class = class_create(THIS_MODULE, "bdi"); |
Anton Blanchard | 1442145 | 2010-04-02 09:46:55 +0200 | [diff] [blame] | 219 | if (IS_ERR(bdi_class)) |
| 220 | return PTR_ERR(bdi_class); |
| 221 | |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 222 | bdi_class->dev_attrs = bdi_dev_attrs; |
Miklos Szeredi | 76f1418 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 223 | bdi_debug_init(); |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 224 | return 0; |
| 225 | } |
Miklos Szeredi | 76f1418 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 226 | postcore_initcall(bdi_class_init); |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 227 | |
Jens Axboe | 2616015 | 2009-03-17 09:35:06 +0100 | [diff] [blame] | 228 | static int __init default_bdi_init(void) |
| 229 | { |
| 230 | int err; |
| 231 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 232 | sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers"); |
| 233 | BUG_ON(IS_ERR(sync_supers_tsk)); |
| 234 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 235 | setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0); |
Jens Axboe | 6423104 | 2010-05-21 20:00:35 +0200 | [diff] [blame] | 236 | bdi_arm_supers_timer(); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 237 | |
Jens Axboe | 2616015 | 2009-03-17 09:35:06 +0100 | [diff] [blame] | 238 | err = bdi_init(&default_backing_dev_info); |
| 239 | if (!err) |
| 240 | bdi_register(&default_backing_dev_info, NULL, "default"); |
Jan Kara | 976e48f | 2010-09-21 11:48:55 +0200 | [diff] [blame] | 241 | err = bdi_init(&noop_backing_dev_info); |
Jens Axboe | 2616015 | 2009-03-17 09:35:06 +0100 | [diff] [blame] | 242 | |
| 243 | return err; |
| 244 | } |
| 245 | subsys_initcall(default_bdi_init); |
| 246 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 247 | int bdi_has_dirty_io(struct backing_dev_info *bdi) |
| 248 | { |
| 249 | return wb_has_dirty_io(&bdi->wb); |
| 250 | } |
| 251 | |
| 252 | static void bdi_flush_io(struct backing_dev_info *bdi) |
| 253 | { |
| 254 | struct writeback_control wbc = { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 255 | .sync_mode = WB_SYNC_NONE, |
| 256 | .older_than_this = NULL, |
| 257 | .range_cyclic = 1, |
| 258 | .nr_to_write = 1024, |
| 259 | }; |
| 260 | |
Christoph Hellwig | 9c3a8ee | 2010-06-10 12:07:27 +0200 | [diff] [blame] | 261 | writeback_inodes_wb(&bdi->wb, &wbc); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 262 | } |
| 263 | |
| 264 | /* |
Artem Bityutskiy | 6f904ff | 2010-07-25 14:29:11 +0300 | [diff] [blame] | 265 | * kupdated() used to do this. We cannot do it from the bdi_forker_thread() |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 266 | * or we risk deadlocking on ->s_umount. The longer term solution would be |
| 267 | * to implement sync_supers_bdi() or similar and simply do it from the |
Artem Bityutskiy | 6f904ff | 2010-07-25 14:29:11 +0300 | [diff] [blame] | 268 | * bdi writeback thread individually. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 269 | */ |
| 270 | static int bdi_sync_supers(void *unused) |
| 271 | { |
| 272 | set_user_nice(current, 0); |
| 273 | |
| 274 | while (!kthread_should_stop()) { |
| 275 | set_current_state(TASK_INTERRUPTIBLE); |
| 276 | schedule(); |
| 277 | |
| 278 | /* |
| 279 | * Do this periodically, like kupdated() did before. |
| 280 | */ |
| 281 | sync_supers(); |
| 282 | } |
| 283 | |
| 284 | return 0; |
| 285 | } |
| 286 | |
Jens Axboe | 6423104 | 2010-05-21 20:00:35 +0200 | [diff] [blame] | 287 | void bdi_arm_supers_timer(void) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 288 | { |
| 289 | unsigned long next; |
| 290 | |
Jens Axboe | 6423104 | 2010-05-21 20:00:35 +0200 | [diff] [blame] | 291 | if (!dirty_writeback_interval) |
| 292 | return; |
| 293 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 294 | next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies; |
| 295 | mod_timer(&sync_supers_timer, round_jiffies_up(next)); |
| 296 | } |
| 297 | |
| 298 | static void sync_supers_timer_fn(unsigned long unused) |
| 299 | { |
| 300 | wake_up_process(sync_supers_tsk); |
Jens Axboe | 6423104 | 2010-05-21 20:00:35 +0200 | [diff] [blame] | 301 | bdi_arm_supers_timer(); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 302 | } |
| 303 | |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 304 | static void wakeup_timer_fn(unsigned long data) |
| 305 | { |
| 306 | struct backing_dev_info *bdi = (struct backing_dev_info *)data; |
| 307 | |
| 308 | spin_lock_bh(&bdi->wb_lock); |
| 309 | if (bdi->wb.task) { |
Artem Bityutskiy | 6033202 | 2010-07-25 14:29:24 +0300 | [diff] [blame] | 310 | trace_writeback_wake_thread(bdi); |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 311 | wake_up_process(bdi->wb.task); |
| 312 | } else { |
| 313 | /* |
| 314 | * When bdi tasks are inactive for long time, they are killed. |
| 315 | * In this case we have to wake-up the forker thread which |
| 316 | * should create and run the bdi thread. |
| 317 | */ |
Artem Bityutskiy | 6033202 | 2010-07-25 14:29:24 +0300 | [diff] [blame] | 318 | trace_writeback_wake_forker_thread(bdi); |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 319 | wake_up_process(default_backing_dev_info.wb.task); |
| 320 | } |
| 321 | spin_unlock_bh(&bdi->wb_lock); |
| 322 | } |
| 323 | |
| 324 | /* |
| 325 | * This function is used when the first inode for this bdi is marked dirty. It |
| 326 | * wakes-up the corresponding bdi thread which should then take care of the |
| 327 | * periodic background write-out of dirty inodes. Since the write-out would |
| 328 | * starts only 'dirty_writeback_interval' centisecs from now anyway, we just |
| 329 | * set up a timer which wakes the bdi thread up later. |
| 330 | * |
| 331 | * Note, we wouldn't bother setting up the timer, but this function is on the |
| 332 | * fast-path (used by '__mark_inode_dirty()'), so we save few context switches |
| 333 | * by delaying the wake-up. |
| 334 | */ |
| 335 | void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi) |
| 336 | { |
| 337 | unsigned long timeout; |
| 338 | |
| 339 | timeout = msecs_to_jiffies(dirty_writeback_interval * 10); |
| 340 | mod_timer(&bdi->wb.wakeup_timer, jiffies + timeout); |
| 341 | } |
| 342 | |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 343 | /* |
| 344 | * Calculate the longest interval (jiffies) bdi threads are allowed to be |
| 345 | * inactive. |
| 346 | */ |
| 347 | static unsigned long bdi_longest_inactive(void) |
| 348 | { |
| 349 | unsigned long interval; |
| 350 | |
| 351 | interval = msecs_to_jiffies(dirty_writeback_interval * 10); |
| 352 | return max(5UL * 60 * HZ, interval); |
| 353 | } |
| 354 | |
Artem Bityutskiy | 6f904ff | 2010-07-25 14:29:11 +0300 | [diff] [blame] | 355 | static int bdi_forker_thread(void *ptr) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 356 | { |
| 357 | struct bdi_writeback *me = ptr; |
| 358 | |
Peter Zijlstra | 766f916 | 2010-10-26 14:22:45 -0700 | [diff] [blame] | 359 | current->flags |= PF_SWAPWRITE; |
Christoph Hellwig | c1955ce | 2010-06-19 23:08:06 +0200 | [diff] [blame] | 360 | set_freezable(); |
| 361 | |
| 362 | /* |
| 363 | * Our parent may run at a different priority, just set us to normal |
| 364 | */ |
| 365 | set_user_nice(current, 0); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 366 | |
| 367 | for (;;) { |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 368 | struct task_struct *task = NULL; |
Artem Bityutskiy | 78c40cb | 2010-07-25 14:29:17 +0300 | [diff] [blame] | 369 | struct backing_dev_info *bdi; |
Artem Bityutskiy | adf3924 | 2010-07-25 14:29:19 +0300 | [diff] [blame] | 370 | enum { |
| 371 | NO_ACTION, /* Nothing to do */ |
| 372 | FORK_THREAD, /* Fork bdi thread */ |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 373 | KILL_THREAD, /* Kill inactive bdi thread */ |
Artem Bityutskiy | adf3924 | 2010-07-25 14:29:19 +0300 | [diff] [blame] | 374 | } action = NO_ACTION; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 375 | |
| 376 | /* |
| 377 | * Temporary measure, we want to make sure we don't see |
| 378 | * dirty data on the default backing_dev_info |
| 379 | */ |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 380 | if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) { |
| 381 | del_timer(&me->wakeup_timer); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 382 | wb_do_writeback(me, 0); |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 383 | } |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 384 | |
Jens Axboe | cfc4ba5 | 2009-09-14 13:12:40 +0200 | [diff] [blame] | 385 | spin_lock_bh(&bdi_lock); |
Artem Bityutskiy | c5f7ad2 | 2010-07-25 14:29:13 +0300 | [diff] [blame] | 386 | set_current_state(TASK_INTERRUPTIBLE); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 387 | |
Artem Bityutskiy | 78c40cb | 2010-07-25 14:29:17 +0300 | [diff] [blame] | 388 | list_for_each_entry(bdi, &bdi_list, bdi_list) { |
Artem Bityutskiy | adf3924 | 2010-07-25 14:29:19 +0300 | [diff] [blame] | 389 | bool have_dirty_io; |
| 390 | |
| 391 | if (!bdi_cap_writeback_dirty(bdi) || |
| 392 | bdi_cap_flush_forker(bdi)) |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 393 | continue; |
| 394 | |
Artem Bityutskiy | 080dcec | 2010-07-25 14:29:16 +0300 | [diff] [blame] | 395 | WARN(!test_bit(BDI_registered, &bdi->state), |
| 396 | "bdi %p/%s is not registered!\n", bdi, bdi->name); |
| 397 | |
Artem Bityutskiy | adf3924 | 2010-07-25 14:29:19 +0300 | [diff] [blame] | 398 | have_dirty_io = !list_empty(&bdi->work_list) || |
| 399 | wb_has_dirty_io(&bdi->wb); |
Artem Bityutskiy | 78c40cb | 2010-07-25 14:29:17 +0300 | [diff] [blame] | 400 | |
| 401 | /* |
Artem Bityutskiy | adf3924 | 2010-07-25 14:29:19 +0300 | [diff] [blame] | 402 | * If the bdi has work to do, but the thread does not |
| 403 | * exist - create it. |
Artem Bityutskiy | 78c40cb | 2010-07-25 14:29:17 +0300 | [diff] [blame] | 404 | */ |
Artem Bityutskiy | adf3924 | 2010-07-25 14:29:19 +0300 | [diff] [blame] | 405 | if (!bdi->wb.task && have_dirty_io) { |
| 406 | /* |
| 407 | * Set the pending bit - if someone will try to |
| 408 | * unregister this bdi - it'll wait on this bit. |
| 409 | */ |
| 410 | set_bit(BDI_pending, &bdi->state); |
| 411 | action = FORK_THREAD; |
| 412 | break; |
| 413 | } |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 414 | |
Jens Axboe | 6bf05d0 | 2010-08-04 13:34:31 +0200 | [diff] [blame] | 415 | spin_lock(&bdi->wb_lock); |
| 416 | |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 417 | /* |
| 418 | * If there is no work to do and the bdi thread was |
| 419 | * inactive long enough - kill it. The wb_lock is taken |
| 420 | * to make sure no-one adds more work to this bdi and |
| 421 | * wakes the bdi thread up. |
| 422 | */ |
| 423 | if (bdi->wb.task && !have_dirty_io && |
| 424 | time_after(jiffies, bdi->wb.last_active + |
| 425 | bdi_longest_inactive())) { |
| 426 | task = bdi->wb.task; |
| 427 | bdi->wb.task = NULL; |
| 428 | spin_unlock(&bdi->wb_lock); |
| 429 | set_bit(BDI_pending, &bdi->state); |
| 430 | action = KILL_THREAD; |
| 431 | break; |
| 432 | } |
Jens Axboe | 6bf05d0 | 2010-08-04 13:34:31 +0200 | [diff] [blame] | 433 | spin_unlock(&bdi->wb_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 434 | } |
Artem Bityutskiy | 080dcec | 2010-07-25 14:29:16 +0300 | [diff] [blame] | 435 | spin_unlock_bh(&bdi_lock); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 436 | |
Artem Bityutskiy | c4ec790 | 2010-07-25 14:29:14 +0300 | [diff] [blame] | 437 | /* Keep working if default bdi still has things to do */ |
| 438 | if (!list_empty(&me->bdi->work_list)) |
| 439 | __set_current_state(TASK_RUNNING); |
| 440 | |
Artem Bityutskiy | adf3924 | 2010-07-25 14:29:19 +0300 | [diff] [blame] | 441 | switch (action) { |
| 442 | case FORK_THREAD: |
| 443 | __set_current_state(TASK_RUNNING); |
Artem Bityutskiy | 6628bc7 | 2010-08-27 09:15:09 +0200 | [diff] [blame] | 444 | task = kthread_create(bdi_writeback_thread, &bdi->wb, |
| 445 | "flush-%s", dev_name(bdi->dev)); |
Artem Bityutskiy | adf3924 | 2010-07-25 14:29:19 +0300 | [diff] [blame] | 446 | if (IS_ERR(task)) { |
| 447 | /* |
| 448 | * If thread creation fails, force writeout of |
| 449 | * the bdi from the thread. |
| 450 | */ |
| 451 | bdi_flush_io(bdi); |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 452 | } else { |
| 453 | /* |
| 454 | * The spinlock makes sure we do not lose |
| 455 | * wake-ups when racing with 'bdi_queue_work()'. |
Artem Bityutskiy | 6628bc7 | 2010-08-27 09:15:09 +0200 | [diff] [blame] | 456 | * And as soon as the bdi thread is visible, we |
| 457 | * can start it. |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 458 | */ |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 459 | spin_lock_bh(&bdi->wb_lock); |
Artem Bityutskiy | adf3924 | 2010-07-25 14:29:19 +0300 | [diff] [blame] | 460 | bdi->wb.task = task; |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 461 | spin_unlock_bh(&bdi->wb_lock); |
Artem Bityutskiy | 6628bc7 | 2010-08-27 09:15:09 +0200 | [diff] [blame] | 462 | wake_up_process(task); |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 463 | } |
| 464 | break; |
| 465 | |
| 466 | case KILL_THREAD: |
| 467 | __set_current_state(TASK_RUNNING); |
| 468 | kthread_stop(task); |
Artem Bityutskiy | adf3924 | 2010-07-25 14:29:19 +0300 | [diff] [blame] | 469 | break; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 470 | |
Artem Bityutskiy | adf3924 | 2010-07-25 14:29:19 +0300 | [diff] [blame] | 471 | case NO_ACTION: |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 472 | if (!wb_has_dirty_io(me) || !dirty_writeback_interval) |
| 473 | /* |
| 474 | * There are no dirty data. The only thing we |
| 475 | * should now care about is checking for |
| 476 | * inactive bdi threads and killing them. Thus, |
| 477 | * let's sleep for longer time, save energy and |
| 478 | * be friendly for battery-driven devices. |
| 479 | */ |
| 480 | schedule_timeout(bdi_longest_inactive()); |
Jens Axboe | 6423104 | 2010-05-21 20:00:35 +0200 | [diff] [blame] | 481 | else |
Artem Bityutskiy | 253c34e | 2010-07-25 14:29:21 +0300 | [diff] [blame] | 482 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 483 | try_to_freeze(); |
Artem Bityutskiy | adf3924 | 2010-07-25 14:29:19 +0300 | [diff] [blame] | 484 | /* Back to the main loop */ |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 485 | continue; |
| 486 | } |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 487 | |
| 488 | /* |
| 489 | * Clear pending bit and wakeup anybody waiting to tear us down. |
| 490 | */ |
| 491 | clear_bit(BDI_pending, &bdi->state); |
| 492 | smp_mb__after_clear_bit(); |
| 493 | wake_up_bit(&bdi->state, BDI_pending); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 494 | } |
| 495 | |
| 496 | return 0; |
| 497 | } |
| 498 | |
Jens Axboe | cfc4ba5 | 2009-09-14 13:12:40 +0200 | [diff] [blame] | 499 | /* |
| 500 | * Remove bdi from bdi_list, and ensure that it is no longer visible |
| 501 | */ |
| 502 | static void bdi_remove_from_list(struct backing_dev_info *bdi) |
| 503 | { |
| 504 | spin_lock_bh(&bdi_lock); |
| 505 | list_del_rcu(&bdi->bdi_list); |
| 506 | spin_unlock_bh(&bdi_lock); |
| 507 | |
| 508 | synchronize_rcu(); |
| 509 | } |
| 510 | |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 511 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
| 512 | const char *fmt, ...) |
| 513 | { |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 514 | va_list args; |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 515 | struct device *dev; |
| 516 | |
Andrew Morton | 69fc208 | 2008-12-09 13:14:06 -0800 | [diff] [blame] | 517 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
Artem Bityutskiy | c284de6 | 2010-07-25 14:29:25 +0300 | [diff] [blame] | 518 | return 0; |
Kay Sievers | f1d0b06 | 2008-12-02 10:31:50 -0800 | [diff] [blame] | 519 | |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 520 | va_start(args, fmt); |
Greg Kroah-Hartman | 19051c5 | 2008-05-15 13:44:08 -0700 | [diff] [blame] | 521 | dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args); |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 522 | va_end(args); |
Artem Bityutskiy | c284de6 | 2010-07-25 14:29:25 +0300 | [diff] [blame] | 523 | if (IS_ERR(dev)) |
| 524 | return PTR_ERR(dev); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 525 | |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 526 | bdi->dev = dev; |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 527 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 528 | /* |
| 529 | * Just start the forker thread for our default backing_dev_info, |
| 530 | * and add other bdi's to the list. They will get a thread created |
| 531 | * on-demand when they need it. |
| 532 | */ |
| 533 | if (bdi_cap_flush_forker(bdi)) { |
| 534 | struct bdi_writeback *wb = &bdi->wb; |
| 535 | |
Artem Bityutskiy | 6f904ff | 2010-07-25 14:29:11 +0300 | [diff] [blame] | 536 | wb->task = kthread_run(bdi_forker_thread, wb, "bdi-%s", |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 537 | dev_name(dev)); |
Artem Bityutskiy | c284de6 | 2010-07-25 14:29:25 +0300 | [diff] [blame] | 538 | if (IS_ERR(wb->task)) |
| 539 | return PTR_ERR(wb->task); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 540 | } |
| 541 | |
| 542 | bdi_debug_register(bdi, dev_name(dev)); |
Jens Axboe | 500b067 | 2009-09-09 09:10:25 +0200 | [diff] [blame] | 543 | set_bit(BDI_registered, &bdi->state); |
Artem Bityutskiy | c284de6 | 2010-07-25 14:29:25 +0300 | [diff] [blame] | 544 | |
| 545 | spin_lock_bh(&bdi_lock); |
| 546 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); |
| 547 | spin_unlock_bh(&bdi_lock); |
| 548 | |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 549 | trace_writeback_bdi_register(bdi); |
Artem Bityutskiy | c284de6 | 2010-07-25 14:29:25 +0300 | [diff] [blame] | 550 | return 0; |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 551 | } |
| 552 | EXPORT_SYMBOL(bdi_register); |
| 553 | |
| 554 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) |
| 555 | { |
| 556 | return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); |
| 557 | } |
| 558 | EXPORT_SYMBOL(bdi_register_dev); |
| 559 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 560 | /* |
| 561 | * Remove bdi from the global list and shutdown any threads we have running |
| 562 | */ |
| 563 | static void bdi_wb_shutdown(struct backing_dev_info *bdi) |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 564 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 565 | if (!bdi_cap_writeback_dirty(bdi)) |
| 566 | return; |
| 567 | |
| 568 | /* |
Artem Bityutskiy | fff5b85 | 2010-07-25 14:29:20 +0300 | [diff] [blame] | 569 | * Make sure nobody finds us on the bdi_list anymore |
| 570 | */ |
| 571 | bdi_remove_from_list(bdi); |
| 572 | |
| 573 | /* |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 574 | * If setup is pending, wait for that to complete first |
| 575 | */ |
| 576 | wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait, |
| 577 | TASK_UNINTERRUPTIBLE); |
| 578 | |
| 579 | /* |
Christoph Hellwig | c1955ce | 2010-06-19 23:08:06 +0200 | [diff] [blame] | 580 | * Finally, kill the kernel thread. We don't need to be RCU |
Romit Dasgupta | c62b17a | 2009-11-12 13:08:11 +0100 | [diff] [blame] | 581 | * safe anymore, since the bdi is gone from visibility. Force |
| 582 | * unfreeze of the thread before calling kthread_stop(), otherwise |
| 583 | * it would never exet if it is currently stuck in the refrigerator. |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 584 | */ |
Christoph Hellwig | c1955ce | 2010-06-19 23:08:06 +0200 | [diff] [blame] | 585 | if (bdi->wb.task) { |
| 586 | thaw_process(bdi->wb.task); |
| 587 | kthread_stop(bdi->wb.task); |
Romit Dasgupta | c62b17a | 2009-11-12 13:08:11 +0100 | [diff] [blame] | 588 | } |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 589 | } |
| 590 | |
Jens Axboe | 592b09a | 2009-10-29 11:46:12 +0100 | [diff] [blame] | 591 | /* |
| 592 | * This bdi is going away now, make sure that no super_blocks point to it |
| 593 | */ |
| 594 | static void bdi_prune_sb(struct backing_dev_info *bdi) |
| 595 | { |
| 596 | struct super_block *sb; |
| 597 | |
| 598 | spin_lock(&sb_lock); |
| 599 | list_for_each_entry(sb, &super_blocks, s_list) { |
| 600 | if (sb->s_bdi == bdi) |
Jens Axboe | 95f2860 | 2011-03-17 11:13:12 +0100 | [diff] [blame] | 601 | sb->s_bdi = &default_backing_dev_info; |
Jens Axboe | 592b09a | 2009-10-29 11:46:12 +0100 | [diff] [blame] | 602 | } |
| 603 | spin_unlock(&sb_lock); |
| 604 | } |
| 605 | |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 606 | void bdi_unregister(struct backing_dev_info *bdi) |
| 607 | { |
| 608 | if (bdi->dev) { |
Peter Zijlstra | 650957d | 2011-07-25 17:11:57 -0700 | [diff] [blame^] | 609 | bdi_set_min_ratio(bdi, 0); |
Dave Chinner | 455b286 | 2010-07-07 13:24:06 +1000 | [diff] [blame] | 610 | trace_writeback_bdi_unregister(bdi); |
Jens Axboe | 8c4db33 | 2009-11-03 20:18:44 +0100 | [diff] [blame] | 611 | bdi_prune_sb(bdi); |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 612 | del_timer_sync(&bdi->wb.wakeup_timer); |
Jens Axboe | 8c4db33 | 2009-11-03 20:18:44 +0100 | [diff] [blame] | 613 | |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 614 | if (!bdi_cap_flush_forker(bdi)) |
| 615 | bdi_wb_shutdown(bdi); |
Miklos Szeredi | 76f1418 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 616 | bdi_debug_unregister(bdi); |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 617 | device_unregister(bdi->dev); |
| 618 | bdi->dev = NULL; |
| 619 | } |
| 620 | } |
| 621 | EXPORT_SYMBOL(bdi_unregister); |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 622 | |
Artem Bityutskiy | 6467716 | 2010-07-25 14:29:22 +0300 | [diff] [blame] | 623 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) |
| 624 | { |
| 625 | memset(wb, 0, sizeof(*wb)); |
| 626 | |
| 627 | wb->bdi = bdi; |
| 628 | wb->last_old_flush = jiffies; |
| 629 | INIT_LIST_HEAD(&wb->b_dirty); |
| 630 | INIT_LIST_HEAD(&wb->b_io); |
| 631 | INIT_LIST_HEAD(&wb->b_more_io); |
| 632 | setup_timer(&wb->wakeup_timer, wakeup_timer_fn, (unsigned long)bdi); |
| 633 | } |
| 634 | |
Peter Zijlstra | b2e8fb6 | 2007-10-16 23:25:47 -0700 | [diff] [blame] | 635 | int bdi_init(struct backing_dev_info *bdi) |
| 636 | { |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 637 | int i, err; |
Peter Zijlstra | b2e8fb6 | 2007-10-16 23:25:47 -0700 | [diff] [blame] | 638 | |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 639 | bdi->dev = NULL; |
| 640 | |
Peter Zijlstra | 189d3c4 | 2008-04-30 00:54:35 -0700 | [diff] [blame] | 641 | bdi->min_ratio = 0; |
Peter Zijlstra | a42dde0 | 2008-04-30 00:54:36 -0700 | [diff] [blame] | 642 | bdi->max_ratio = 100; |
| 643 | bdi->max_prop_frac = PROP_FRAC_BASE; |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 644 | spin_lock_init(&bdi->wb_lock); |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 645 | INIT_LIST_HEAD(&bdi->bdi_list); |
Jens Axboe | 03ba378 | 2009-09-09 09:08:54 +0200 | [diff] [blame] | 646 | INIT_LIST_HEAD(&bdi->work_list); |
| 647 | |
| 648 | bdi_wb_init(&bdi->wb, bdi); |
| 649 | |
Peter Zijlstra | b2e8fb6 | 2007-10-16 23:25:47 -0700 | [diff] [blame] | 650 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { |
Peter Zijlstra | ea31951 | 2008-12-26 15:08:55 +0100 | [diff] [blame] | 651 | err = percpu_counter_init(&bdi->bdi_stat[i], 0); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 652 | if (err) |
| 653 | goto err; |
| 654 | } |
| 655 | |
| 656 | bdi->dirty_exceeded = 0; |
| 657 | err = prop_local_init_percpu(&bdi->completions); |
| 658 | |
| 659 | if (err) { |
| 660 | err: |
Denis Cheng | 4b01a0b | 2007-12-04 23:45:07 -0800 | [diff] [blame] | 661 | while (i--) |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 662 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
Peter Zijlstra | b2e8fb6 | 2007-10-16 23:25:47 -0700 | [diff] [blame] | 663 | } |
| 664 | |
| 665 | return err; |
| 666 | } |
| 667 | EXPORT_SYMBOL(bdi_init); |
| 668 | |
| 669 | void bdi_destroy(struct backing_dev_info *bdi) |
| 670 | { |
| 671 | int i; |
| 672 | |
Jens Axboe | ce5f8e7 | 2009-09-14 12:57:56 +0200 | [diff] [blame] | 673 | /* |
| 674 | * Splice our entries to the default_backing_dev_info, if this |
| 675 | * bdi disappears |
| 676 | */ |
| 677 | if (bdi_has_dirty_io(bdi)) { |
| 678 | struct bdi_writeback *dst = &default_backing_dev_info.wb; |
| 679 | |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 680 | spin_lock(&inode_wb_list_lock); |
Jens Axboe | ce5f8e7 | 2009-09-14 12:57:56 +0200 | [diff] [blame] | 681 | list_splice(&bdi->wb.b_dirty, &dst->b_dirty); |
| 682 | list_splice(&bdi->wb.b_io, &dst->b_io); |
| 683 | list_splice(&bdi->wb.b_more_io, &dst->b_more_io); |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 684 | spin_unlock(&inode_wb_list_lock); |
Jens Axboe | ce5f8e7 | 2009-09-14 12:57:56 +0200 | [diff] [blame] | 685 | } |
Jens Axboe | 66f3b8e | 2009-09-02 09:19:46 +0200 | [diff] [blame] | 686 | |
Peter Zijlstra | cf0ca9f | 2008-04-30 00:54:32 -0700 | [diff] [blame] | 687 | bdi_unregister(bdi); |
| 688 | |
Peter Zijlstra | b2e8fb6 | 2007-10-16 23:25:47 -0700 | [diff] [blame] | 689 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) |
| 690 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 691 | |
| 692 | prop_local_destroy_percpu(&bdi->completions); |
Peter Zijlstra | b2e8fb6 | 2007-10-16 23:25:47 -0700 | [diff] [blame] | 693 | } |
| 694 | EXPORT_SYMBOL(bdi_destroy); |
| 695 | |
Jens Axboe | c3c5320 | 2010-04-22 11:37:01 +0200 | [diff] [blame] | 696 | /* |
| 697 | * For use from filesystems to quickly init and register a bdi associated |
| 698 | * with dirty writeback |
| 699 | */ |
| 700 | int bdi_setup_and_register(struct backing_dev_info *bdi, char *name, |
| 701 | unsigned int cap) |
| 702 | { |
| 703 | char tmp[32]; |
| 704 | int err; |
| 705 | |
| 706 | bdi->name = name; |
| 707 | bdi->capabilities = cap; |
| 708 | err = bdi_init(bdi); |
| 709 | if (err) |
| 710 | return err; |
| 711 | |
| 712 | sprintf(tmp, "%.28s%s", name, "-%d"); |
| 713 | err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq)); |
| 714 | if (err) { |
| 715 | bdi_destroy(bdi); |
| 716 | return err; |
| 717 | } |
| 718 | |
| 719 | return 0; |
| 720 | } |
| 721 | EXPORT_SYMBOL(bdi_setup_and_register); |
| 722 | |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 723 | static wait_queue_head_t congestion_wqh[2] = { |
| 724 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), |
| 725 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) |
| 726 | }; |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 727 | static atomic_t nr_bdi_congested[2]; |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 728 | |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 729 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 730 | { |
| 731 | enum bdi_state bit; |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 732 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 733 | |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 734 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 735 | if (test_and_clear_bit(bit, &bdi->state)) |
| 736 | atomic_dec(&nr_bdi_congested[sync]); |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 737 | smp_mb__after_clear_bit(); |
| 738 | if (waitqueue_active(wqh)) |
| 739 | wake_up(wqh); |
| 740 | } |
| 741 | EXPORT_SYMBOL(clear_bdi_congested); |
| 742 | |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 743 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 744 | { |
| 745 | enum bdi_state bit; |
| 746 | |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 747 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 748 | if (!test_and_set_bit(bit, &bdi->state)) |
| 749 | atomic_inc(&nr_bdi_congested[sync]); |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 750 | } |
| 751 | EXPORT_SYMBOL(set_bdi_congested); |
| 752 | |
| 753 | /** |
| 754 | * congestion_wait - wait for a backing_dev to become uncongested |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 755 | * @sync: SYNC or ASYNC IO |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 756 | * @timeout: timeout in jiffies |
| 757 | * |
| 758 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit |
| 759 | * write congestion. If no backing_devs are congested then just wait for the |
| 760 | * next write to be completed. |
| 761 | */ |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 762 | long congestion_wait(int sync, long timeout) |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 763 | { |
| 764 | long ret; |
Mel Gorman | 52bb919 | 2010-10-26 14:21:41 -0700 | [diff] [blame] | 765 | unsigned long start = jiffies; |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 766 | DEFINE_WAIT(wait); |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 767 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 768 | |
| 769 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); |
| 770 | ret = io_schedule_timeout(timeout); |
| 771 | finish_wait(wqh, &wait); |
Mel Gorman | 52bb919 | 2010-10-26 14:21:41 -0700 | [diff] [blame] | 772 | |
| 773 | trace_writeback_congestion_wait(jiffies_to_usecs(timeout), |
| 774 | jiffies_to_usecs(jiffies - start)); |
| 775 | |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 776 | return ret; |
| 777 | } |
| 778 | EXPORT_SYMBOL(congestion_wait); |
Peter Zijlstra | 04fbfdc | 2007-10-16 23:25:50 -0700 | [diff] [blame] | 779 | |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 780 | /** |
| 781 | * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes |
| 782 | * @zone: A zone to check if it is heavily congested |
| 783 | * @sync: SYNC or ASYNC IO |
| 784 | * @timeout: timeout in jiffies |
| 785 | * |
| 786 | * In the event of a congested backing_dev (any backing_dev) and the given |
| 787 | * @zone has experienced recent congestion, this waits for up to @timeout |
| 788 | * jiffies for either a BDI to exit congestion of the given @sync queue |
| 789 | * or a write to complete. |
| 790 | * |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 791 | * In the absence of zone congestion, cond_resched() is called to yield |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 792 | * the processor if necessary but otherwise does not sleep. |
| 793 | * |
| 794 | * The return value is 0 if the sleep is for the full timeout. Otherwise, |
| 795 | * it is the number of jiffies that were still remaining when the function |
| 796 | * returned. return_value == timeout implies the function did not sleep. |
| 797 | */ |
| 798 | long wait_iff_congested(struct zone *zone, int sync, long timeout) |
| 799 | { |
| 800 | long ret; |
| 801 | unsigned long start = jiffies; |
| 802 | DEFINE_WAIT(wait); |
| 803 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
| 804 | |
| 805 | /* |
| 806 | * If there is no congestion, or heavy congestion is not being |
| 807 | * encountered in the current zone, yield if necessary instead |
| 808 | * of sleeping on the congestion queue |
| 809 | */ |
| 810 | if (atomic_read(&nr_bdi_congested[sync]) == 0 || |
| 811 | !zone_is_reclaim_congested(zone)) { |
| 812 | cond_resched(); |
| 813 | |
| 814 | /* In case we scheduled, work out time remaining */ |
| 815 | ret = timeout - (jiffies - start); |
| 816 | if (ret < 0) |
| 817 | ret = 0; |
| 818 | |
| 819 | goto out; |
| 820 | } |
| 821 | |
| 822 | /* Sleep until uncongested or a write happens */ |
| 823 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); |
| 824 | ret = io_schedule_timeout(timeout); |
| 825 | finish_wait(wqh, &wait); |
| 826 | |
| 827 | out: |
| 828 | trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout), |
| 829 | jiffies_to_usecs(jiffies - start)); |
| 830 | |
| 831 | return ret; |
| 832 | } |
| 833 | EXPORT_SYMBOL(wait_iff_congested); |