Under normal circumstances each device is given a part of the total write-back cache that relates to its current avg writeout speed in relation to the other devices. min_ratio - allows one to assign a minimum portion of the write-back cache to a particular device. This is useful in situations where you might want to provide a minimum QoS. (One request for this feature came from flash based storage people who wanted to avoid writing out at all costs - they of course needed some pdflush hacks as well) max_ratio - allows one to assign a maximum portion of the dirty limit to a particular device. This is useful in situations where you want to avoid one device taking all or most of the write-back cache. Eg. an NFS mount that is prone to get stuck, or a FUSE mount which you don't trust to play fair. Add "min_ratio" to /sys/class/bdi. This indicates the minimum percentage of the global dirty threshold allocated to this bdi. [mszeredi@suse.cz] - fix parsing in min_ratio_store() - document new sysfs attribute
Unified · Split
| @@ -44,3 +44,9 @@ bdi_dirty_kb (read-only) · 43 unmodified lines | |||
| 44 | Current threshold on this BDI for reclaimable + writeback | 44 | Current threshold on this BDI for reclaimable + writeback |
| 45 | memory | 45 | memory |
| 46 | 46 | ||
| 47 | min_ratio (read-write) | ||
| 48 | |||
| 49 | Minimal percentage of global dirty threshold allocated to this | ||
| 50 | bdi. If the value written to this file would make the the sum | ||
| 51 | of all min_ratio values exceed 100, then EINVAL is returned. | ||
| 52 | The default is zero | ||
| @@ -51,6 +51,8 @@ struct backing_dev_info { · 50 unmodified lines | |||
| 51 | struct prop_local_percpu completions; | 51 | struct prop_local_percpu completions; |
| 52 | int dirty_exceeded; | 52 | int dirty_exceeded; |
| 53 | 53 | ||
| 54 | unsigned int min_ratio; | ||
| 55 | |||
| 54 | struct device *dev; | 56 | struct device *dev; |
| 55 | }; | 57 | }; |
| 56 | 58 | ||
| @@ -137,6 +139,8 @@ static inline unsigned long bdi_stat_error(struct backing_dev_info *bdi) · 80 unmodified lines | |||
| 137 | #endif | 139 | #endif |
| 138 | } | 140 | } |
| 139 | 141 | ||
| 142 | int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio); | ||
| 143 | |||
| 140 | /* | 144 | /* |
| 141 | * Flags in backing_dev_info::capability | 145 | * Flags in backing_dev_info::capability |
| 142 | * - The first two flags control whether dirty pages will contribute to the | 146 | * - The first two flags control whether dirty pages will contribute to the |
| @@ -55,6 +55,24 @@ static inline unsigned long get_dirty(struct backing_dev_info *bdi, int i) · 54 unmodified lines | |||
| 55 | BDI_SHOW(dirty_kb, K(get_dirty(bdi, 1))) | 55 | BDI_SHOW(dirty_kb, K(get_dirty(bdi, 1))) |
| 56 | BDI_SHOW(bdi_dirty_kb, K(get_dirty(bdi, 2))) | 56 | BDI_SHOW(bdi_dirty_kb, K(get_dirty(bdi, 2))) |
| 57 | 57 | ||
| 58 | static ssize_t min_ratio_store(struct device *dev, | ||
| 59 | struct device_attribute *attr, const char *buf, size_t count) | ||
| 60 | { | ||
| 61 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | ||
| 62 | char *end; | ||
| 63 | unsigned int ratio; | ||
| 64 | ssize_t ret = -EINVAL; | ||
| 65 | |||
| 66 | ratio = simple_strtoul(buf, &end, 10); | ||
| 67 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | ||
| 68 | ret = bdi_set_min_ratio(bdi, ratio); | ||
| 69 | if (!ret) | ||
| 70 | ret = count; | ||
| 71 | } | ||
| 72 | return ret; | ||
| 73 | } | ||
| 74 | BDI_SHOW(min_ratio, bdi->min_ratio) | ||
| 75 | |||
| 58 | #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store) | 76 | #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store) |
| 59 | 77 | ||
| 60 | static struct device_attribute bdi_dev_attrs[] = { | 78 | static struct device_attribute bdi_dev_attrs[] = { |
| @@ -63,6 +81,7 @@ static struct device_attribute bdi_dev_attrs[] = { · 2 unmodified lines | |||
| 63 | __ATTR_RO(writeback_kb), | 81 | __ATTR_RO(writeback_kb), |
| 64 | __ATTR_RO(dirty_kb), | 82 | __ATTR_RO(dirty_kb), |
| 65 | __ATTR_RO(bdi_dirty_kb), | 83 | __ATTR_RO(bdi_dirty_kb), |
| 84 | __ATTR_RW(min_ratio), | ||
| 66 | __ATTR_NULL, | 85 | __ATTR_NULL, |
| 67 | }; | 86 | }; |
| 68 | 87 | ||
| @@ -127,6 +146,8 @@ int bdi_init(struct backing_dev_info *bdi) · 58 unmodified lines | |||
| 127 | 146 | ||
| 128 | bdi->dev = NULL; | 147 | bdi->dev = NULL; |
| 129 | 148 | ||
| 149 | bdi->min_ratio = 0; | ||
| 150 | |||
| 130 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { | 151 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { |
| 131 | err = percpu_counter_init_irq(&bdi->bdi_stat[i], 0); | 152 | err = percpu_counter_init_irq(&bdi->bdi_stat[i], 0); |
| 132 | if (err) | 153 | if (err) |
| @@ -242,6 +242,29 @@ static void task_dirty_limit(struct task_struct *tsk, long *pdirty) · 241 unmodified lines | |||
| 242 | *pdirty = dirty; | 242 | *pdirty = dirty; |
| 243 | } | 243 | } |
| 244 | 244 | ||
| 245 | /* | ||
| 246 | * | ||
| 247 | */ | ||
| 248 | static DEFINE_SPINLOCK(bdi_lock); | ||
| 249 | static unsigned int bdi_min_ratio; | ||
| 250 | |||
| 251 | int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio) | ||
| 252 | { | ||
| 253 | int ret = 0; | ||
| 254 | unsigned long flags; | ||
| 255 | |||
| 256 | spin_lock_irqsave(&bdi_lock, flags); | ||
| 257 | min_ratio -= bdi->min_ratio; | ||
| 258 | if (bdi_min_ratio + min_ratio < 100) { | ||
| 259 | bdi_min_ratio += min_ratio; | ||
| 260 | bdi->min_ratio += min_ratio; | ||
| 261 | } else | ||
| 262 | ret = -EINVAL; | ||
| 263 | spin_unlock_irqrestore(&bdi_lock, flags); | ||
| 264 | |||
| 265 | return ret; | ||
| 266 | } | ||
| 267 | |||
| 245 | /* | 268 | /* |
| 246 | * Work out the current dirty-memory clamping and background writeout | 269 | * Work out the current dirty-memory clamping and background writeout |
| 247 | * thresholds. | 270 | * thresholds. |
| @@ -330,7 +353,7 @@ get_dirty_limits(long *pbackground, long *pdirty, long *pbdi_dirty, · 82 unmodified lines | |||
| 330 | *pdirty = dirty; | 353 | *pdirty = dirty; |
| 331 | 354 | ||
| 332 | if (bdi) { | 355 | if (bdi) { |
| 333 | u64 bdi_dirty = dirty; | 356 | u64 bdi_dirty; |
| 334 | long numerator, denominator; | 357 | long numerator, denominator; |
| 335 | 358 | ||
| 336 | /* | 359 | /* |
| @@ -338,8 +361,10 @@ get_dirty_limits(long *pbackground, long *pdirty, long *pbdi_dirty, · 1 unmodified line | |||
| 338 | */ | 361 | */ |
| 339 | bdi_writeout_fraction(bdi, &numerator, &denominator); | 362 | bdi_writeout_fraction(bdi, &numerator, &denominator); |
| 340 | 363 | ||
| 364 | bdi_dirty = (dirty * (100 - bdi_min_ratio)) / 100; | ||
| 341 | bdi_dirty *= numerator; | 365 | bdi_dirty *= numerator; |
| 342 | do_div(bdi_dirty, denominator); | 366 | do_div(bdi_dirty, denominator); |
| 367 | bdi_dirty += (dirty * bdi->min_ratio) / 100; | ||
| 343 | 368 | ||
| 344 | *pbdi_dirty = bdi_dirty; | 369 | *pbdi_dirty = bdi_dirty; |
| 345 | clip_bdi_dirty_limit(bdi, dirty, pbdi_dirty); | 370 | clip_bdi_dirty_limit(bdi, dirty, pbdi_dirty); |