backing-dev.h 13 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493
  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. static inline struct backing_dev_info *bdi_get(struct backing_dev_info *bdi)
  18. {
  19. kref_get(&bdi->refcnt);
  20. return bdi;
  21. }
  22. void bdi_put(struct backing_dev_info *bdi);
  23. __printf(2, 3)
  24. int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...);
  25. int bdi_register_va(struct backing_dev_info *bdi, const char *fmt,
  26. va_list args);
  27. int bdi_register_owner(struct backing_dev_info *bdi, struct device *owner);
  28. void bdi_unregister(struct backing_dev_info *bdi);
  29. struct backing_dev_info *bdi_alloc_node(gfp_t gfp_mask, int node_id);
  30. static inline struct backing_dev_info *bdi_alloc(gfp_t gfp_mask)
  31. {
  32. return bdi_alloc_node(gfp_mask, NUMA_NO_NODE);
  33. }
  34. void wb_start_writeback(struct bdi_writeback *wb, long nr_pages,
  35. bool range_cyclic, enum wb_reason reason);
  36. void wb_start_background_writeback(struct bdi_writeback *wb);
  37. void wb_workfn(struct work_struct *work);
  38. void wb_wakeup_delayed(struct bdi_writeback *wb);
  39. extern spinlock_t bdi_lock;
  40. extern struct list_head bdi_list;
  41. extern struct workqueue_struct *bdi_wq;
  42. static inline bool wb_has_dirty_io(struct bdi_writeback *wb)
  43. {
  44. return test_bit(WB_has_dirty_io, &wb->state);
  45. }
  46. static inline bool bdi_has_dirty_io(struct backing_dev_info *bdi)
  47. {
  48. /*
  49. * @bdi->tot_write_bandwidth is guaranteed to be > 0 if there are
  50. * any dirty wbs. See wb_update_write_bandwidth().
  51. */
  52. return atomic_long_read(&bdi->tot_write_bandwidth);
  53. }
  54. static inline void __add_wb_stat(struct bdi_writeback *wb,
  55. enum wb_stat_item item, s64 amount)
  56. {
  57. percpu_counter_add_batch(&wb->stat[item], amount, WB_STAT_BATCH);
  58. }
  59. static inline void inc_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  60. {
  61. __add_wb_stat(wb, item, 1);
  62. }
  63. static inline void dec_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  64. {
  65. __add_wb_stat(wb, item, -1);
  66. }
  67. static inline s64 wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  68. {
  69. return percpu_counter_read_positive(&wb->stat[item]);
  70. }
  71. static inline s64 wb_stat_sum(struct bdi_writeback *wb, enum wb_stat_item item)
  72. {
  73. return percpu_counter_sum_positive(&wb->stat[item]);
  74. }
  75. extern void wb_writeout_inc(struct bdi_writeback *wb);
  76. /*
  77. * maximal error of a stat counter.
  78. */
  79. static inline unsigned long wb_stat_error(struct bdi_writeback *wb)
  80. {
  81. #ifdef CONFIG_SMP
  82. return nr_cpu_ids * WB_STAT_BATCH;
  83. #else
  84. return 1;
  85. #endif
  86. }
  87. int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
  88. int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
  89. /*
  90. * Flags in backing_dev_info::capability
  91. *
  92. * The first three flags control whether dirty pages will contribute to the
  93. * VM's accounting and whether writepages() should be called for dirty pages
  94. * (something that would not, for example, be appropriate for ramfs)
  95. *
  96. * WARNING: these flags are closely related and should not normally be
  97. * used separately. The BDI_CAP_NO_ACCT_AND_WRITEBACK combines these
  98. * three flags into a single convenience macro.
  99. *
  100. * BDI_CAP_NO_ACCT_DIRTY: Dirty pages shouldn't contribute to accounting
  101. * BDI_CAP_NO_WRITEBACK: Don't write pages back
  102. * BDI_CAP_NO_ACCT_WB: Don't automatically account writeback pages
  103. * BDI_CAP_STRICTLIMIT: Keep number of dirty pages below bdi threshold.
  104. *
  105. * BDI_CAP_CGROUP_WRITEBACK: Supports cgroup-aware writeback.
  106. */
  107. #define BDI_CAP_NO_ACCT_DIRTY 0x00000001
  108. #define BDI_CAP_NO_WRITEBACK 0x00000002
  109. #define BDI_CAP_NO_ACCT_WB 0x00000004
  110. #define BDI_CAP_STABLE_WRITES 0x00000008
  111. #define BDI_CAP_STRICTLIMIT 0x00000010
  112. #define BDI_CAP_CGROUP_WRITEBACK 0x00000020
  113. #define BDI_CAP_NO_ACCT_AND_WRITEBACK \
  114. (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB)
  115. extern struct backing_dev_info noop_backing_dev_info;
  116. /**
  117. * writeback_in_progress - determine whether there is writeback in progress
  118. * @wb: bdi_writeback of interest
  119. *
  120. * Determine whether there is writeback waiting to be handled against a
  121. * bdi_writeback.
  122. */
  123. static inline bool writeback_in_progress(struct bdi_writeback *wb)
  124. {
  125. return test_bit(WB_writeback_running, &wb->state);
  126. }
  127. static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
  128. {
  129. struct super_block *sb;
  130. if (!inode)
  131. return &noop_backing_dev_info;
  132. sb = inode->i_sb;
  133. #ifdef CONFIG_BLOCK
  134. if (sb_is_blkdev_sb(sb))
  135. return I_BDEV(inode)->bd_bdi;
  136. #endif
  137. return sb->s_bdi;
  138. }
  139. static inline int wb_congested(struct bdi_writeback *wb, int cong_bits)
  140. {
  141. struct backing_dev_info *bdi = wb->bdi;
  142. if (bdi->congested_fn)
  143. return bdi->congested_fn(bdi->congested_data, cong_bits);
  144. return wb->congested->state & cong_bits;
  145. }
  146. long congestion_wait(int sync, long timeout);
  147. long wait_iff_congested(struct pglist_data *pgdat, int sync, long timeout);
  148. int pdflush_proc_obsolete(struct ctl_table *table, int write,
  149. void __user *buffer, size_t *lenp, loff_t *ppos);
  150. static inline bool bdi_cap_stable_pages_required(struct backing_dev_info *bdi)
  151. {
  152. return bdi->capabilities & BDI_CAP_STABLE_WRITES;
  153. }
  154. static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi)
  155. {
  156. return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK);
  157. }
  158. static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi)
  159. {
  160. return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY);
  161. }
  162. static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi)
  163. {
  164. /* Paranoia: BDI_CAP_NO_WRITEBACK implies BDI_CAP_NO_ACCT_WB */
  165. return !(bdi->capabilities & (BDI_CAP_NO_ACCT_WB |
  166. BDI_CAP_NO_WRITEBACK));
  167. }
  168. static inline bool mapping_cap_writeback_dirty(struct address_space *mapping)
  169. {
  170. return bdi_cap_writeback_dirty(inode_to_bdi(mapping->host));
  171. }
  172. static inline bool mapping_cap_account_dirty(struct address_space *mapping)
  173. {
  174. return bdi_cap_account_dirty(inode_to_bdi(mapping->host));
  175. }
  176. static inline int bdi_sched_wait(void *word)
  177. {
  178. schedule();
  179. return 0;
  180. }
  181. #ifdef CONFIG_CGROUP_WRITEBACK
  182. struct bdi_writeback_congested *
  183. wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp);
  184. void wb_congested_put(struct bdi_writeback_congested *congested);
  185. struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
  186. struct cgroup_subsys_state *memcg_css,
  187. gfp_t gfp);
  188. void wb_memcg_offline(struct mem_cgroup *memcg);
  189. void wb_blkcg_offline(struct blkcg *blkcg);
  190. int inode_congested(struct inode *inode, int cong_bits);
  191. /**
  192. * inode_cgwb_enabled - test whether cgroup writeback is enabled on an inode
  193. * @inode: inode of interest
  194. *
  195. * cgroup writeback requires support from both the bdi and filesystem.
  196. * Also, both memcg and iocg have to be on the default hierarchy. Test
  197. * whether all conditions are met.
  198. *
  199. * Note that the test result may change dynamically on the same inode
  200. * depending on how memcg and iocg are configured.
  201. */
  202. static inline bool inode_cgwb_enabled(struct inode *inode)
  203. {
  204. struct backing_dev_info *bdi = inode_to_bdi(inode);
  205. return cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
  206. cgroup_subsys_on_dfl(io_cgrp_subsys) &&
  207. bdi_cap_account_dirty(bdi) &&
  208. (bdi->capabilities & BDI_CAP_CGROUP_WRITEBACK) &&
  209. (inode->i_sb->s_iflags & SB_I_CGROUPWB);
  210. }
  211. /**
  212. * wb_find_current - find wb for %current on a bdi
  213. * @bdi: bdi of interest
  214. *
  215. * Find the wb of @bdi which matches both the memcg and blkcg of %current.
  216. * Must be called under rcu_read_lock() which protects the returend wb.
  217. * NULL if not found.
  218. */
  219. static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
  220. {
  221. struct cgroup_subsys_state *memcg_css;
  222. struct bdi_writeback *wb;
  223. memcg_css = task_css(current, memory_cgrp_id);
  224. if (!memcg_css->parent)
  225. return &bdi->wb;
  226. wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
  227. /*
  228. * %current's blkcg equals the effective blkcg of its memcg. No
  229. * need to use the relatively expensive cgroup_get_e_css().
  230. */
  231. if (likely(wb && wb->blkcg_css == task_css(current, io_cgrp_id)))
  232. return wb;
  233. return NULL;
  234. }
  235. /**
  236. * wb_get_create_current - get or create wb for %current on a bdi
  237. * @bdi: bdi of interest
  238. * @gfp: allocation mask
  239. *
  240. * Equivalent to wb_get_create() on %current's memcg. This function is
  241. * called from a relatively hot path and optimizes the common cases using
  242. * wb_find_current().
  243. */
  244. static inline struct bdi_writeback *
  245. wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
  246. {
  247. struct bdi_writeback *wb;
  248. rcu_read_lock();
  249. wb = wb_find_current(bdi);
  250. if (wb && unlikely(!wb_tryget(wb)))
  251. wb = NULL;
  252. rcu_read_unlock();
  253. if (unlikely(!wb)) {
  254. struct cgroup_subsys_state *memcg_css;
  255. memcg_css = task_get_css(current, memory_cgrp_id);
  256. wb = wb_get_create(bdi, memcg_css, gfp);
  257. css_put(memcg_css);
  258. }
  259. return wb;
  260. }
  261. /**
  262. * inode_to_wb_is_valid - test whether an inode has a wb associated
  263. * @inode: inode of interest
  264. *
  265. * Returns %true if @inode has a wb associated. May be called without any
  266. * locking.
  267. */
  268. static inline bool inode_to_wb_is_valid(struct inode *inode)
  269. {
  270. return inode->i_wb;
  271. }
  272. /**
  273. * inode_to_wb - determine the wb of an inode
  274. * @inode: inode of interest
  275. *
  276. * Returns the wb @inode is currently associated with. The caller must be
  277. * holding either @inode->i_lock, @inode->i_mapping->tree_lock, or the
  278. * associated wb's list_lock.
  279. */
  280. static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
  281. {
  282. #ifdef CONFIG_LOCKDEP
  283. WARN_ON_ONCE(debug_locks &&
  284. (!lockdep_is_held(&inode->i_lock) &&
  285. !lockdep_is_held(&inode->i_mapping->tree_lock) &&
  286. !lockdep_is_held(&inode->i_wb->list_lock)));
  287. #endif
  288. return inode->i_wb;
  289. }
  290. /**
  291. * unlocked_inode_to_wb_begin - begin unlocked inode wb access transaction
  292. * @inode: target inode
  293. * @lockedp: temp bool output param, to be passed to the end function
  294. *
  295. * The caller wants to access the wb associated with @inode but isn't
  296. * holding inode->i_lock, mapping->tree_lock or wb->list_lock. This
  297. * function determines the wb associated with @inode and ensures that the
  298. * association doesn't change until the transaction is finished with
  299. * unlocked_inode_to_wb_end().
  300. *
  301. * The caller must call unlocked_inode_to_wb_end() with *@lockdep
  302. * afterwards and can't sleep during transaction. IRQ may or may not be
  303. * disabled on return.
  304. */
  305. static inline struct bdi_writeback *
  306. unlocked_inode_to_wb_begin(struct inode *inode, bool *lockedp)
  307. {
  308. rcu_read_lock();
  309. /*
  310. * Paired with store_release in inode_switch_wb_work_fn() and
  311. * ensures that we see the new wb if we see cleared I_WB_SWITCH.
  312. */
  313. *lockedp = smp_load_acquire(&inode->i_state) & I_WB_SWITCH;
  314. if (unlikely(*lockedp))
  315. spin_lock_irq(&inode->i_mapping->tree_lock);
  316. /*
  317. * Protected by either !I_WB_SWITCH + rcu_read_lock() or tree_lock.
  318. * inode_to_wb() will bark. Deref directly.
  319. */
  320. return inode->i_wb;
  321. }
  322. /**
  323. * unlocked_inode_to_wb_end - end inode wb access transaction
  324. * @inode: target inode
  325. * @locked: *@lockedp from unlocked_inode_to_wb_begin()
  326. */
  327. static inline void unlocked_inode_to_wb_end(struct inode *inode, bool locked)
  328. {
  329. if (unlikely(locked))
  330. spin_unlock_irq(&inode->i_mapping->tree_lock);
  331. rcu_read_unlock();
  332. }
  333. #else /* CONFIG_CGROUP_WRITEBACK */
  334. static inline bool inode_cgwb_enabled(struct inode *inode)
  335. {
  336. return false;
  337. }
  338. static inline struct bdi_writeback_congested *
  339. wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp)
  340. {
  341. atomic_inc(&bdi->wb_congested->refcnt);
  342. return bdi->wb_congested;
  343. }
  344. static inline void wb_congested_put(struct bdi_writeback_congested *congested)
  345. {
  346. if (atomic_dec_and_test(&congested->refcnt))
  347. kfree(congested);
  348. }
  349. static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
  350. {
  351. return &bdi->wb;
  352. }
  353. static inline struct bdi_writeback *
  354. wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
  355. {
  356. return &bdi->wb;
  357. }
  358. static inline bool inode_to_wb_is_valid(struct inode *inode)
  359. {
  360. return true;
  361. }
  362. static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
  363. {
  364. return &inode_to_bdi(inode)->wb;
  365. }
  366. static inline struct bdi_writeback *
  367. unlocked_inode_to_wb_begin(struct inode *inode, bool *lockedp)
  368. {
  369. return inode_to_wb(inode);
  370. }
  371. static inline void unlocked_inode_to_wb_end(struct inode *inode, bool locked)
  372. {
  373. }
  374. static inline void wb_memcg_offline(struct mem_cgroup *memcg)
  375. {
  376. }
  377. static inline void wb_blkcg_offline(struct blkcg *blkcg)
  378. {
  379. }
  380. static inline int inode_congested(struct inode *inode, int cong_bits)
  381. {
  382. return wb_congested(&inode_to_bdi(inode)->wb, cong_bits);
  383. }
  384. #endif /* CONFIG_CGROUP_WRITEBACK */
  385. static inline int inode_read_congested(struct inode *inode)
  386. {
  387. return inode_congested(inode, 1 << WB_sync_congested);
  388. }
  389. static inline int inode_write_congested(struct inode *inode)
  390. {
  391. return inode_congested(inode, 1 << WB_async_congested);
  392. }
  393. static inline int inode_rw_congested(struct inode *inode)
  394. {
  395. return inode_congested(inode, (1 << WB_sync_congested) |
  396. (1 << WB_async_congested));
  397. }
  398. static inline int bdi_congested(struct backing_dev_info *bdi, int cong_bits)
  399. {
  400. return wb_congested(&bdi->wb, cong_bits);
  401. }
  402. static inline int bdi_read_congested(struct backing_dev_info *bdi)
  403. {
  404. return bdi_congested(bdi, 1 << WB_sync_congested);
  405. }
  406. static inline int bdi_write_congested(struct backing_dev_info *bdi)
  407. {
  408. return bdi_congested(bdi, 1 << WB_async_congested);
  409. }
  410. static inline int bdi_rw_congested(struct backing_dev_info *bdi)
  411. {
  412. return bdi_congested(bdi, (1 << WB_sync_congested) |
  413. (1 << WB_async_congested));
  414. }
  415. #endif /* _LINUX_BACKING_DEV_H */