1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * include/linux/backing-dev.h 4 * 5 * low-level device information and state which is propagated up through 6 * to high-level code. 7 */ 8 9 #ifndef _LINUX_BACKING_DEV_H 10 #define _LINUX_BACKING_DEV_H 11 12 #include <linux/kernel.h> 13 #include <linux/fs.h> 14 #include <linux/sched.h> 15 #include <linux/blkdev.h> 16 #include <linux/device.h> 17 #include <linux/writeback.h> 18 #include <linux/blk-cgroup.h> 19 #include <linux/backing-dev-defs.h> 20 #include <linux/slab.h> 21 bdi_get(struct backing_dev_info * bdi)22 static inline struct backing_dev_info *bdi_get(struct backing_dev_info *bdi) 23 { 24 kref_get(&bdi->refcnt); 25 return bdi; 26 } 27 28 struct backing_dev_info *bdi_get_by_id(u64 id); 29 void bdi_put(struct backing_dev_info *bdi); 30 31 __printf(2, 3) 32 int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...); 33 __printf(2, 0) 34 int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, 35 va_list args); 36 void bdi_set_owner(struct backing_dev_info *bdi, struct device *owner); 37 void bdi_unregister(struct backing_dev_info *bdi); 38 39 struct backing_dev_info *bdi_alloc(int node_id); 40 41 void wb_start_background_writeback(struct bdi_writeback *wb); 42 void wb_workfn(struct work_struct *work); 43 void wb_wakeup_delayed(struct bdi_writeback *wb); 44 45 void wb_wait_for_completion(struct wb_completion *done); 46 47 extern spinlock_t bdi_lock; 48 extern struct list_head bdi_list; 49 50 extern struct workqueue_struct *bdi_wq; 51 extern struct workqueue_struct *bdi_async_bio_wq; 52 wb_has_dirty_io(struct bdi_writeback * wb)53 static inline bool wb_has_dirty_io(struct bdi_writeback *wb) 54 { 55 return test_bit(WB_has_dirty_io, &wb->state); 56 } 57 bdi_has_dirty_io(struct backing_dev_info * bdi)58 static inline bool bdi_has_dirty_io(struct backing_dev_info *bdi) 59 { 60 /* 61 * @bdi->tot_write_bandwidth is guaranteed to be > 0 if there are 62 * any dirty wbs. See wb_update_write_bandwidth(). 63 */ 64 return atomic_long_read(&bdi->tot_write_bandwidth); 65 } 66 __add_wb_stat(struct bdi_writeback * wb,enum wb_stat_item item,s64 amount)67 static inline void __add_wb_stat(struct bdi_writeback *wb, 68 enum wb_stat_item item, s64 amount) 69 { 70 percpu_counter_add_batch(&wb->stat[item], amount, WB_STAT_BATCH); 71 } 72 inc_wb_stat(struct bdi_writeback * wb,enum wb_stat_item item)73 static inline void inc_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item) 74 { 75 __add_wb_stat(wb, item, 1); 76 } 77 dec_wb_stat(struct bdi_writeback * wb,enum wb_stat_item item)78 static inline void dec_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item) 79 { 80 __add_wb_stat(wb, item, -1); 81 } 82 wb_stat(struct bdi_writeback * wb,enum wb_stat_item item)83 static inline s64 wb_stat(struct bdi_writeback *wb, enum wb_stat_item item) 84 { 85 return percpu_counter_read_positive(&wb->stat[item]); 86 } 87 wb_stat_sum(struct bdi_writeback * wb,enum wb_stat_item item)88 static inline s64 wb_stat_sum(struct bdi_writeback *wb, enum wb_stat_item item) 89 { 90 return percpu_counter_sum_positive(&wb->stat[item]); 91 } 92 93 extern void wb_writeout_inc(struct bdi_writeback *wb); 94 95 /* 96 * maximal error of a stat counter. 97 */ wb_stat_error(void)98 static inline unsigned long wb_stat_error(void) 99 { 100 #ifdef CONFIG_SMP 101 return nr_cpu_ids * WB_STAT_BATCH; 102 #else 103 return 1; 104 #endif 105 } 106 107 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio); 108 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio); 109 110 /* 111 * Flags in backing_dev_info::capability 112 * 113 * BDI_CAP_WRITEBACK: Supports dirty page writeback, and dirty pages 114 * should contribute to accounting 115 * BDI_CAP_WRITEBACK_ACCT: Automatically account writeback pages 116 * BDI_CAP_STRICTLIMIT: Keep number of dirty pages below bdi threshold 117 */ 118 #define BDI_CAP_WRITEBACK (1 << 0) 119 #define BDI_CAP_WRITEBACK_ACCT (1 << 1) 120 #define BDI_CAP_STRICTLIMIT (1 << 2) 121 122 extern struct backing_dev_info noop_backing_dev_info; 123 124 /** 125 * writeback_in_progress - determine whether there is writeback in progress 126 * @wb: bdi_writeback of interest 127 * 128 * Determine whether there is writeback waiting to be handled against a 129 * bdi_writeback. 130 */ writeback_in_progress(struct bdi_writeback * wb)131 static inline bool writeback_in_progress(struct bdi_writeback *wb) 132 { 133 return test_bit(WB_writeback_running, &wb->state); 134 } 135 inode_to_bdi(struct inode * inode)136 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) 137 { 138 struct super_block *sb; 139 140 if (!inode) 141 return &noop_backing_dev_info; 142 143 sb = inode->i_sb; 144 #ifdef CONFIG_BLOCK 145 if (sb_is_blkdev_sb(sb)) 146 return I_BDEV(inode)->bd_bdi; 147 #endif 148 return sb->s_bdi; 149 } 150 wb_congested(struct bdi_writeback * wb,int cong_bits)151 static inline int wb_congested(struct bdi_writeback *wb, int cong_bits) 152 { 153 return wb->congested & cong_bits; 154 } 155 156 long congestion_wait(int sync, long timeout); 157 long wait_iff_congested(int sync, long timeout); 158 mapping_can_writeback(struct address_space * mapping)159 static inline bool mapping_can_writeback(struct address_space *mapping) 160 { 161 return inode_to_bdi(mapping->host)->capabilities & BDI_CAP_WRITEBACK; 162 } 163 bdi_sched_wait(void * word)164 static inline int bdi_sched_wait(void *word) 165 { 166 schedule(); 167 return 0; 168 } 169 170 #ifdef CONFIG_CGROUP_WRITEBACK 171 172 struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi, 173 struct cgroup_subsys_state *memcg_css); 174 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi, 175 struct cgroup_subsys_state *memcg_css, 176 gfp_t gfp); 177 void wb_memcg_offline(struct mem_cgroup *memcg); 178 void wb_blkcg_offline(struct blkcg *blkcg); 179 int inode_congested(struct inode *inode, int cong_bits); 180 181 /** 182 * inode_cgwb_enabled - test whether cgroup writeback is enabled on an inode 183 * @inode: inode of interest 184 * 185 * Cgroup writeback requires support from the filesystem. Also, both memcg and 186 * iocg have to be on the default hierarchy. Test whether all conditions are 187 * met. 188 * 189 * Note that the test result may change dynamically on the same inode 190 * depending on how memcg and iocg are configured. 191 */ inode_cgwb_enabled(struct inode * inode)192 static inline bool inode_cgwb_enabled(struct inode *inode) 193 { 194 struct backing_dev_info *bdi = inode_to_bdi(inode); 195 196 return cgroup_subsys_on_dfl(memory_cgrp_subsys) && 197 cgroup_subsys_on_dfl(io_cgrp_subsys) && 198 (bdi->capabilities & BDI_CAP_WRITEBACK) && 199 (inode->i_sb->s_iflags & SB_I_CGROUPWB); 200 } 201 202 /** 203 * wb_find_current - find wb for %current on a bdi 204 * @bdi: bdi of interest 205 * 206 * Find the wb of @bdi which matches both the memcg and blkcg of %current. 207 * Must be called under rcu_read_lock() which protects the returend wb. 208 * NULL if not found. 209 */ wb_find_current(struct backing_dev_info * bdi)210 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi) 211 { 212 struct cgroup_subsys_state *memcg_css; 213 struct bdi_writeback *wb; 214 215 memcg_css = task_css(current, memory_cgrp_id); 216 if (!memcg_css->parent) 217 return &bdi->wb; 218 219 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id); 220 221 /* 222 * %current's blkcg equals the effective blkcg of its memcg. No 223 * need to use the relatively expensive cgroup_get_e_css(). 224 */ 225 if (likely(wb && wb->blkcg_css == task_css(current, io_cgrp_id))) 226 return wb; 227 return NULL; 228 } 229 230 /** 231 * wb_get_create_current - get or create wb for %current on a bdi 232 * @bdi: bdi of interest 233 * @gfp: allocation mask 234 * 235 * Equivalent to wb_get_create() on %current's memcg. This function is 236 * called from a relatively hot path and optimizes the common cases using 237 * wb_find_current(). 238 */ 239 static inline struct bdi_writeback * wb_get_create_current(struct backing_dev_info * bdi,gfp_t gfp)240 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp) 241 { 242 struct bdi_writeback *wb; 243 244 rcu_read_lock(); 245 wb = wb_find_current(bdi); 246 if (wb && unlikely(!wb_tryget(wb))) 247 wb = NULL; 248 rcu_read_unlock(); 249 250 if (unlikely(!wb)) { 251 struct cgroup_subsys_state *memcg_css; 252 253 memcg_css = task_get_css(current, memory_cgrp_id); 254 wb = wb_get_create(bdi, memcg_css, gfp); 255 css_put(memcg_css); 256 } 257 return wb; 258 } 259 260 /** 261 * inode_to_wb_is_valid - test whether an inode has a wb associated 262 * @inode: inode of interest 263 * 264 * Returns %true if @inode has a wb associated. May be called without any 265 * locking. 266 */ inode_to_wb_is_valid(struct inode * inode)267 static inline bool inode_to_wb_is_valid(struct inode *inode) 268 { 269 return inode->i_wb; 270 } 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, the i_pages lock, or the 278 * associated wb's list_lock. 279 */ inode_to_wb(const struct inode * inode)280 static inline struct bdi_writeback *inode_to_wb(const 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->i_pages.xa_lock) && 286 !lockdep_is_held(&inode->i_wb->list_lock))); 287 #endif 288 return inode->i_wb; 289 } 290 291 /** 292 * unlocked_inode_to_wb_begin - begin unlocked inode wb access transaction 293 * @inode: target inode 294 * @cookie: output param, to be passed to the end function 295 * 296 * The caller wants to access the wb associated with @inode but isn't 297 * holding inode->i_lock, the i_pages lock or wb->list_lock. This 298 * function determines the wb associated with @inode and ensures that the 299 * association doesn't change until the transaction is finished with 300 * unlocked_inode_to_wb_end(). 301 * 302 * The caller must call unlocked_inode_to_wb_end() with *@cookie afterwards and 303 * can't sleep during the transaction. IRQs may or may not be disabled on 304 * return. 305 */ 306 static inline struct bdi_writeback * unlocked_inode_to_wb_begin(struct inode * inode,struct wb_lock_cookie * cookie)307 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie) 308 { 309 rcu_read_lock(); 310 311 /* 312 * Paired with store_release in inode_switch_wbs_work_fn() and 313 * ensures that we see the new wb if we see cleared I_WB_SWITCH. 314 */ 315 cookie->locked = smp_load_acquire(&inode->i_state) & I_WB_SWITCH; 316 317 if (unlikely(cookie->locked)) 318 xa_lock_irqsave(&inode->i_mapping->i_pages, cookie->flags); 319 320 /* 321 * Protected by either !I_WB_SWITCH + rcu_read_lock() or the i_pages 322 * lock. inode_to_wb() will bark. Deref directly. 323 */ 324 return inode->i_wb; 325 } 326 327 /** 328 * unlocked_inode_to_wb_end - end inode wb access transaction 329 * @inode: target inode 330 * @cookie: @cookie from unlocked_inode_to_wb_begin() 331 */ unlocked_inode_to_wb_end(struct inode * inode,struct wb_lock_cookie * cookie)332 static inline void unlocked_inode_to_wb_end(struct inode *inode, 333 struct wb_lock_cookie *cookie) 334 { 335 if (unlikely(cookie->locked)) 336 xa_unlock_irqrestore(&inode->i_mapping->i_pages, cookie->flags); 337 338 rcu_read_unlock(); 339 } 340 341 #else /* CONFIG_CGROUP_WRITEBACK */ 342 inode_cgwb_enabled(struct inode * inode)343 static inline bool inode_cgwb_enabled(struct inode *inode) 344 { 345 return false; 346 } 347 wb_find_current(struct backing_dev_info * bdi)348 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi) 349 { 350 return &bdi->wb; 351 } 352 353 static inline struct bdi_writeback * wb_get_create_current(struct backing_dev_info * bdi,gfp_t gfp)354 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp) 355 { 356 return &bdi->wb; 357 } 358 inode_to_wb_is_valid(struct inode * inode)359 static inline bool inode_to_wb_is_valid(struct inode *inode) 360 { 361 return true; 362 } 363 inode_to_wb(struct inode * inode)364 static inline struct bdi_writeback *inode_to_wb(struct inode *inode) 365 { 366 return &inode_to_bdi(inode)->wb; 367 } 368 369 static inline struct bdi_writeback * unlocked_inode_to_wb_begin(struct inode * inode,struct wb_lock_cookie * cookie)370 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie) 371 { 372 return inode_to_wb(inode); 373 } 374 unlocked_inode_to_wb_end(struct inode * inode,struct wb_lock_cookie * cookie)375 static inline void unlocked_inode_to_wb_end(struct inode *inode, 376 struct wb_lock_cookie *cookie) 377 { 378 } 379 wb_memcg_offline(struct mem_cgroup * memcg)380 static inline void wb_memcg_offline(struct mem_cgroup *memcg) 381 { 382 } 383 wb_blkcg_offline(struct blkcg * blkcg)384 static inline void wb_blkcg_offline(struct blkcg *blkcg) 385 { 386 } 387 inode_congested(struct inode * inode,int cong_bits)388 static inline int inode_congested(struct inode *inode, int cong_bits) 389 { 390 return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); 391 } 392 393 #endif /* CONFIG_CGROUP_WRITEBACK */ 394 inode_read_congested(struct inode * inode)395 static inline int inode_read_congested(struct inode *inode) 396 { 397 return inode_congested(inode, 1 << WB_sync_congested); 398 } 399 inode_write_congested(struct inode * inode)400 static inline int inode_write_congested(struct inode *inode) 401 { 402 return inode_congested(inode, 1 << WB_async_congested); 403 } 404 inode_rw_congested(struct inode * inode)405 static inline int inode_rw_congested(struct inode *inode) 406 { 407 return inode_congested(inode, (1 << WB_sync_congested) | 408 (1 << WB_async_congested)); 409 } 410 bdi_congested(struct backing_dev_info * bdi,int cong_bits)411 static inline int bdi_congested(struct backing_dev_info *bdi, int cong_bits) 412 { 413 return wb_congested(&bdi->wb, cong_bits); 414 } 415 bdi_read_congested(struct backing_dev_info * bdi)416 static inline int bdi_read_congested(struct backing_dev_info *bdi) 417 { 418 return bdi_congested(bdi, 1 << WB_sync_congested); 419 } 420 bdi_write_congested(struct backing_dev_info * bdi)421 static inline int bdi_write_congested(struct backing_dev_info *bdi) 422 { 423 return bdi_congested(bdi, 1 << WB_async_congested); 424 } 425 bdi_rw_congested(struct backing_dev_info * bdi)426 static inline int bdi_rw_congested(struct backing_dev_info *bdi) 427 { 428 return bdi_congested(bdi, (1 << WB_sync_congested) | 429 (1 << WB_async_congested)); 430 } 431 432 const char *bdi_dev_name(struct backing_dev_info *bdi); 433 434 #endif /* _LINUX_BACKING_DEV_H */ 435