backing-dev.h 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518
  1. /*
  2. * include/linux/backing-dev.h
  3. *
  4. * low-level device information and state which is propagated up through
  5. * to high-level code.
  6. */
  7. #ifndef _LINUX_BACKING_DEV_H
  8. #define _LINUX_BACKING_DEV_H
  9. #include <linux/kernel.h>
  10. #include <linux/fs.h>
  11. #include <linux/sched.h>
  12. #include <linux/blkdev.h>
  13. #include <linux/writeback.h>
  14. #include <linux/blk-cgroup.h>
  15. #include <linux/backing-dev-defs.h>
  16. #include <linux/slab.h>
  17. int __must_check bdi_init(struct backing_dev_info *bdi);
  18. void bdi_exit(struct backing_dev_info *bdi);
  19. __printf(3, 4)
  20. int bdi_register(struct backing_dev_info *bdi, struct device *parent,
  21. const char *fmt, ...);
  22. int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev);
  23. int bdi_register_owner(struct backing_dev_info *bdi, struct device *owner);
  24. void bdi_unregister(struct backing_dev_info *bdi);
  25. int __must_check bdi_setup_and_register(struct backing_dev_info *, char *);
  26. void bdi_destroy(struct backing_dev_info *bdi);
  27. void wb_start_writeback(struct bdi_writeback *wb, long nr_pages,
  28. bool range_cyclic, enum wb_reason reason);
  29. void wb_start_background_writeback(struct bdi_writeback *wb);
  30. void wb_workfn(struct work_struct *work);
  31. void wb_wakeup_delayed(struct bdi_writeback *wb);
  32. extern spinlock_t bdi_lock;
  33. extern struct list_head bdi_list;
  34. extern struct workqueue_struct *bdi_wq;
  35. static inline bool wb_has_dirty_io(struct bdi_writeback *wb)
  36. {
  37. return test_bit(WB_has_dirty_io, &wb->state);
  38. }
  39. static inline bool bdi_has_dirty_io(struct backing_dev_info *bdi)
  40. {
  41. /*
  42. * @bdi->tot_write_bandwidth is guaranteed to be > 0 if there are
  43. * any dirty wbs. See wb_update_write_bandwidth().
  44. */
  45. return atomic_long_read(&bdi->tot_write_bandwidth);
  46. }
  47. static inline void __add_wb_stat(struct bdi_writeback *wb,
  48. enum wb_stat_item item, s64 amount)
  49. {
  50. __percpu_counter_add(&wb->stat[item], amount, WB_STAT_BATCH);
  51. }
  52. static inline void __inc_wb_stat(struct bdi_writeback *wb,
  53. enum wb_stat_item item)
  54. {
  55. __add_wb_stat(wb, item, 1);
  56. }
  57. static inline void inc_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  58. {
  59. unsigned long flags;
  60. local_irq_save(flags);
  61. __inc_wb_stat(wb, item);
  62. local_irq_restore(flags);
  63. }
  64. static inline void __dec_wb_stat(struct bdi_writeback *wb,
  65. enum wb_stat_item item)
  66. {
  67. __add_wb_stat(wb, item, -1);
  68. }
  69. static inline void dec_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  70. {
  71. unsigned long flags;
  72. local_irq_save(flags);
  73. __dec_wb_stat(wb, item);
  74. local_irq_restore(flags);
  75. }
  76. static inline s64 wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  77. {
  78. return percpu_counter_read_positive(&wb->stat[item]);
  79. }
  80. static inline s64 __wb_stat_sum(struct bdi_writeback *wb,
  81. enum wb_stat_item item)
  82. {
  83. return percpu_counter_sum_positive(&wb->stat[item]);
  84. }
  85. static inline s64 wb_stat_sum(struct bdi_writeback *wb, enum wb_stat_item item)
  86. {
  87. s64 sum;
  88. unsigned long flags;
  89. local_irq_save(flags);
  90. sum = __wb_stat_sum(wb, item);
  91. local_irq_restore(flags);
  92. return sum;
  93. }
  94. extern void wb_writeout_inc(struct bdi_writeback *wb);
  95. /*
  96. * maximal error of a stat counter.
  97. */
  98. static inline unsigned long wb_stat_error(struct bdi_writeback *wb)
  99. {
  100. #ifdef CONFIG_SMP
  101. return nr_cpu_ids * WB_STAT_BATCH;
  102. #else
  103. return 1;
  104. #endif
  105. }
  106. int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
  107. int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
  108. /*
  109. * Flags in backing_dev_info::capability
  110. *
  111. * The first three flags control whether dirty pages will contribute to the
  112. * VM's accounting and whether writepages() should be called for dirty pages
  113. * (something that would not, for example, be appropriate for ramfs)
  114. *
  115. * WARNING: these flags are closely related and should not normally be
  116. * used separately. The BDI_CAP_NO_ACCT_AND_WRITEBACK combines these
  117. * three flags into a single convenience macro.
  118. *
  119. * BDI_CAP_NO_ACCT_DIRTY: Dirty pages shouldn't contribute to accounting
  120. * BDI_CAP_NO_WRITEBACK: Don't write pages back
  121. * BDI_CAP_NO_ACCT_WB: Don't automatically account writeback pages
  122. * BDI_CAP_STRICTLIMIT: Keep number of dirty pages below bdi threshold.
  123. *
  124. * BDI_CAP_CGROUP_WRITEBACK: Supports cgroup-aware writeback.
  125. */
  126. #define BDI_CAP_NO_ACCT_DIRTY 0x00000001
  127. #define BDI_CAP_NO_WRITEBACK 0x00000002
  128. #define BDI_CAP_NO_ACCT_WB 0x00000004
  129. #define BDI_CAP_STABLE_WRITES 0x00000008
  130. #define BDI_CAP_STRICTLIMIT 0x00000010
  131. #define BDI_CAP_CGROUP_WRITEBACK 0x00000020
  132. #define BDI_CAP_NO_ACCT_AND_WRITEBACK \
  133. (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB)
  134. extern struct backing_dev_info noop_backing_dev_info;
  135. /**
  136. * writeback_in_progress - determine whether there is writeback in progress
  137. * @wb: bdi_writeback of interest
  138. *
  139. * Determine whether there is writeback waiting to be handled against a
  140. * bdi_writeback.
  141. */
  142. static inline bool writeback_in_progress(struct bdi_writeback *wb)
  143. {
  144. return test_bit(WB_writeback_running, &wb->state);
  145. }
  146. static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
  147. {
  148. struct super_block *sb;
  149. if (!inode)
  150. return &noop_backing_dev_info;
  151. sb = inode->i_sb;
  152. #ifdef CONFIG_BLOCK
  153. if (sb_is_blkdev_sb(sb))
  154. return blk_get_backing_dev_info(I_BDEV(inode));
  155. #endif
  156. return sb->s_bdi;
  157. }
  158. static inline int wb_congested(struct bdi_writeback *wb, int cong_bits)
  159. {
  160. struct backing_dev_info *bdi = wb->bdi;
  161. if (bdi->congested_fn)
  162. return bdi->congested_fn(bdi->congested_data, cong_bits);
  163. return wb->congested->state & cong_bits;
  164. }
  165. long congestion_wait(int sync, long timeout);
  166. long wait_iff_congested(struct pglist_data *pgdat, int sync, long timeout);
  167. int pdflush_proc_obsolete(struct ctl_table *table, int write,
  168. void __user *buffer, size_t *lenp, loff_t *ppos);
  169. static inline bool bdi_cap_stable_pages_required(struct backing_dev_info *bdi)
  170. {
  171. return bdi->capabilities & BDI_CAP_STABLE_WRITES;
  172. }
  173. static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi)
  174. {
  175. return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK);
  176. }
  177. static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi)
  178. {
  179. return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY);
  180. }
  181. static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi)
  182. {
  183. /* Paranoia: BDI_CAP_NO_WRITEBACK implies BDI_CAP_NO_ACCT_WB */
  184. return !(bdi->capabilities & (BDI_CAP_NO_ACCT_WB |
  185. BDI_CAP_NO_WRITEBACK));
  186. }
  187. static inline bool mapping_cap_writeback_dirty(struct address_space *mapping)
  188. {
  189. return bdi_cap_writeback_dirty(inode_to_bdi(mapping->host));
  190. }
  191. static inline bool mapping_cap_account_dirty(struct address_space *mapping)
  192. {
  193. return bdi_cap_account_dirty(inode_to_bdi(mapping->host));
  194. }
  195. static inline int bdi_sched_wait(void *word)
  196. {
  197. schedule();
  198. return 0;
  199. }
  200. #ifdef CONFIG_CGROUP_WRITEBACK
  201. struct bdi_writeback_congested *
  202. wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp);
  203. void wb_congested_put(struct bdi_writeback_congested *congested);
  204. struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
  205. struct cgroup_subsys_state *memcg_css,
  206. gfp_t gfp);
  207. void wb_memcg_offline(struct mem_cgroup *memcg);
  208. void wb_blkcg_offline(struct blkcg *blkcg);
  209. int inode_congested(struct inode *inode, int cong_bits);
  210. /**
  211. * inode_cgwb_enabled - test whether cgroup writeback is enabled on an inode
  212. * @inode: inode of interest
  213. *
  214. * cgroup writeback requires support from both the bdi and filesystem.
  215. * Also, both memcg and iocg have to be on the default hierarchy. Test
  216. * whether all conditions are met.
  217. *
  218. * Note that the test result may change dynamically on the same inode
  219. * depending on how memcg and iocg are configured.
  220. */
  221. static inline bool inode_cgwb_enabled(struct inode *inode)
  222. {
  223. struct backing_dev_info *bdi = inode_to_bdi(inode);
  224. return cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
  225. cgroup_subsys_on_dfl(io_cgrp_subsys) &&
  226. bdi_cap_account_dirty(bdi) &&
  227. (bdi->capabilities & BDI_CAP_CGROUP_WRITEBACK) &&
  228. (inode->i_sb->s_iflags & SB_I_CGROUPWB);
  229. }
  230. /**
  231. * wb_find_current - find wb for %current on a bdi
  232. * @bdi: bdi of interest
  233. *
  234. * Find the wb of @bdi which matches both the memcg and blkcg of %current.
  235. * Must be called under rcu_read_lock() which protects the returend wb.
  236. * NULL if not found.
  237. */
  238. static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
  239. {
  240. struct cgroup_subsys_state *memcg_css;
  241. struct bdi_writeback *wb;
  242. memcg_css = task_css(current, memory_cgrp_id);
  243. if (!memcg_css->parent)
  244. return &bdi->wb;
  245. wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
  246. /*
  247. * %current's blkcg equals the effective blkcg of its memcg. No
  248. * need to use the relatively expensive cgroup_get_e_css().
  249. */
  250. if (likely(wb && wb->blkcg_css == task_css(current, io_cgrp_id)))
  251. return wb;
  252. return NULL;
  253. }
  254. /**
  255. * wb_get_create_current - get or create wb for %current on a bdi
  256. * @bdi: bdi of interest
  257. * @gfp: allocation mask
  258. *
  259. * Equivalent to wb_get_create() on %current's memcg. This function is
  260. * called from a relatively hot path and optimizes the common cases using
  261. * wb_find_current().
  262. */
  263. static inline struct bdi_writeback *
  264. wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
  265. {
  266. struct bdi_writeback *wb;
  267. rcu_read_lock();
  268. wb = wb_find_current(bdi);
  269. if (wb && unlikely(!wb_tryget(wb)))
  270. wb = NULL;
  271. rcu_read_unlock();
  272. if (unlikely(!wb)) {
  273. struct cgroup_subsys_state *memcg_css;
  274. memcg_css = task_get_css(current, memory_cgrp_id);
  275. wb = wb_get_create(bdi, memcg_css, gfp);
  276. css_put(memcg_css);
  277. }
  278. return wb;
  279. }
  280. /**
  281. * inode_to_wb_is_valid - test whether an inode has a wb associated
  282. * @inode: inode of interest
  283. *
  284. * Returns %true if @inode has a wb associated. May be called without any
  285. * locking.
  286. */
  287. static inline bool inode_to_wb_is_valid(struct inode *inode)
  288. {
  289. return inode->i_wb;
  290. }
  291. /**
  292. * inode_to_wb - determine the wb of an inode
  293. * @inode: inode of interest
  294. *
  295. * Returns the wb @inode is currently associated with. The caller must be
  296. * holding either @inode->i_lock, @inode->i_mapping->tree_lock, or the
  297. * associated wb's list_lock.
  298. */
  299. static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
  300. {
  301. #ifdef CONFIG_LOCKDEP
  302. WARN_ON_ONCE(debug_locks &&
  303. (!lockdep_is_held(&inode->i_lock) &&
  304. !lockdep_is_held(&inode->i_mapping->tree_lock) &&
  305. !lockdep_is_held(&inode->i_wb->list_lock)));
  306. #endif
  307. return inode->i_wb;
  308. }
  309. /**
  310. * unlocked_inode_to_wb_begin - begin unlocked inode wb access transaction
  311. * @inode: target inode
  312. * @lockedp: temp bool output param, to be passed to the end function
  313. *
  314. * The caller wants to access the wb associated with @inode but isn't
  315. * holding inode->i_lock, mapping->tree_lock or wb->list_lock. This
  316. * function determines the wb associated with @inode and ensures that the
  317. * association doesn't change until the transaction is finished with
  318. * unlocked_inode_to_wb_end().
  319. *
  320. * The caller must call unlocked_inode_to_wb_end() with *@lockdep
  321. * afterwards and can't sleep during transaction. IRQ may or may not be
  322. * disabled on return.
  323. */
  324. static inline struct bdi_writeback *
  325. unlocked_inode_to_wb_begin(struct inode *inode, bool *lockedp)
  326. {
  327. rcu_read_lock();
  328. /*
  329. * Paired with store_release in inode_switch_wb_work_fn() and
  330. * ensures that we see the new wb if we see cleared I_WB_SWITCH.
  331. */
  332. *lockedp = smp_load_acquire(&inode->i_state) & I_WB_SWITCH;
  333. if (unlikely(*lockedp))
  334. spin_lock_irq(&inode->i_mapping->tree_lock);
  335. /*
  336. * Protected by either !I_WB_SWITCH + rcu_read_lock() or tree_lock.
  337. * inode_to_wb() will bark. Deref directly.
  338. */
  339. return inode->i_wb;
  340. }
  341. /**
  342. * unlocked_inode_to_wb_end - end inode wb access transaction
  343. * @inode: target inode
  344. * @locked: *@lockedp from unlocked_inode_to_wb_begin()
  345. */
  346. static inline void unlocked_inode_to_wb_end(struct inode *inode, bool locked)
  347. {
  348. if (unlikely(locked))
  349. spin_unlock_irq(&inode->i_mapping->tree_lock);
  350. rcu_read_unlock();
  351. }
  352. #else /* CONFIG_CGROUP_WRITEBACK */
  353. static inline bool inode_cgwb_enabled(struct inode *inode)
  354. {
  355. return false;
  356. }
  357. static inline struct bdi_writeback_congested *
  358. wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp)
  359. {
  360. atomic_inc(&bdi->wb_congested->refcnt);
  361. return bdi->wb_congested;
  362. }
  363. static inline void wb_congested_put(struct bdi_writeback_congested *congested)
  364. {
  365. if (atomic_dec_and_test(&congested->refcnt))
  366. kfree(congested);
  367. }
  368. static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
  369. {
  370. return &bdi->wb;
  371. }
  372. static inline struct bdi_writeback *
  373. wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
  374. {
  375. return &bdi->wb;
  376. }
  377. static inline bool inode_to_wb_is_valid(struct inode *inode)
  378. {
  379. return true;
  380. }
  381. static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
  382. {
  383. return &inode_to_bdi(inode)->wb;
  384. }
  385. static inline struct bdi_writeback *
  386. unlocked_inode_to_wb_begin(struct inode *inode, bool *lockedp)
  387. {
  388. return inode_to_wb(inode);
  389. }
  390. static inline void unlocked_inode_to_wb_end(struct inode *inode, bool locked)
  391. {
  392. }
  393. static inline void wb_memcg_offline(struct mem_cgroup *memcg)
  394. {
  395. }
  396. static inline void wb_blkcg_offline(struct blkcg *blkcg)
  397. {
  398. }
  399. static inline int inode_congested(struct inode *inode, int cong_bits)
  400. {
  401. return wb_congested(&inode_to_bdi(inode)->wb, cong_bits);
  402. }
  403. #endif /* CONFIG_CGROUP_WRITEBACK */
  404. static inline int inode_read_congested(struct inode *inode)
  405. {
  406. return inode_congested(inode, 1 << WB_sync_congested);
  407. }
  408. static inline int inode_write_congested(struct inode *inode)
  409. {
  410. return inode_congested(inode, 1 << WB_async_congested);
  411. }
  412. static inline int inode_rw_congested(struct inode *inode)
  413. {
  414. return inode_congested(inode, (1 << WB_sync_congested) |
  415. (1 << WB_async_congested));
  416. }
  417. static inline int bdi_congested(struct backing_dev_info *bdi, int cong_bits)
  418. {
  419. return wb_congested(&bdi->wb, cong_bits);
  420. }
  421. static inline int bdi_read_congested(struct backing_dev_info *bdi)
  422. {
  423. return bdi_congested(bdi, 1 << WB_sync_congested);
  424. }
  425. static inline int bdi_write_congested(struct backing_dev_info *bdi)
  426. {
  427. return bdi_congested(bdi, 1 << WB_async_congested);
  428. }
  429. static inline int bdi_rw_congested(struct backing_dev_info *bdi)
  430. {
  431. return bdi_congested(bdi, (1 << WB_sync_congested) |
  432. (1 << WB_async_congested));
  433. }
  434. #endif /* _LINUX_BACKING_DEV_H */