1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * include/linux/writeback.h 4 */ 5 #ifndef WRITEBACK_H 6 #define WRITEBACK_H 7 8 #include <linux/sched.h> 9 #include <linux/workqueue.h> 10 #include <linux/fs.h> 11 #include <linux/flex_proportions.h> 12 #include <linux/backing-dev-defs.h> 13 #include <linux/blk_types.h> 14 15 struct bio; 16 17 DECLARE_PER_CPU(int, dirty_throttle_leaks); 18 19 /* 20 * The global dirty threshold is normally equal to the global dirty limit, 21 * except when the system suddenly allocates a lot of anonymous memory and 22 * knocks down the global dirty threshold quickly, in which case the global 23 * dirty limit will follow down slowly to prevent livelocking all dirtier tasks. 24 */ 25 #define DIRTY_SCOPE 8 26 27 struct backing_dev_info; 28 29 /* 30 * fs/fs-writeback.c 31 */ 32 enum writeback_sync_modes { 33 WB_SYNC_NONE, /* Don't wait on anything */ 34 WB_SYNC_ALL, /* Wait on every mapping */ 35 }; 36 37 /* 38 * A control structure which tells the writeback code what to do. These are 39 * always on the stack, and hence need no locking. They are always initialised 40 * in a manner such that unspecified fields are set to zero. 41 */ 42 struct writeback_control { 43 long nr_to_write; /* Write this many pages, and decrement 44 this for each page written */ 45 long pages_skipped; /* Pages which were not written */ 46 47 /* 48 * For a_ops->writepages(): if start or end are non-zero then this is 49 * a hint that the filesystem need only write out the pages inside that 50 * byterange. The byte at `end' is included in the writeout request. 51 */ 52 loff_t range_start; 53 loff_t range_end; 54 55 enum writeback_sync_modes sync_mode; 56 57 unsigned for_kupdate:1; /* A kupdate writeback */ 58 unsigned for_background:1; /* A background writeback */ 59 unsigned tagged_writepages:1; /* tag-and-write to avoid livelock */ 60 unsigned for_reclaim:1; /* Invoked from the page allocator */ 61 unsigned range_cyclic:1; /* range_start is cyclic */ 62 unsigned for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ 63 unsigned unpinned_fscache_wb:1; /* Cleared I_PINNING_FSCACHE_WB */ 64 65 /* 66 * When writeback IOs are bounced through async layers, only the 67 * initial synchronous phase should be accounted towards inode 68 * cgroup ownership arbitration to avoid confusion. Later stages 69 * can set the following flag to disable the accounting. 70 */ 71 unsigned no_cgroup_owner:1; 72 73 unsigned punt_to_cgroup:1; /* cgrp punting, see __REQ_CGROUP_PUNT */ 74 75 /* To enable batching of swap writes to non-block-device backends, 76 * "plug" can be set point to a 'struct swap_iocb *'. When all swap 77 * writes have been submitted, if with swap_iocb is not NULL, 78 * swap_write_unplug() should be called. 79 */ 80 struct swap_iocb **swap_plug; 81 82 #ifdef CONFIG_CGROUP_WRITEBACK 83 struct bdi_writeback *wb; /* wb this writeback is issued under */ 84 struct inode *inode; /* inode being written out */ 85 86 /* foreign inode detection, see wbc_detach_inode() */ 87 int wb_id; /* current wb id */ 88 int wb_lcand_id; /* last foreign candidate wb id */ 89 int wb_tcand_id; /* this foreign candidate wb id */ 90 size_t wb_bytes; /* bytes written by current wb */ 91 size_t wb_lcand_bytes; /* bytes written by last candidate */ 92 size_t wb_tcand_bytes; /* bytes written by this candidate */ 93 #endif 94 }; 95 96 static inline blk_opf_t wbc_to_write_flags(struct writeback_control *wbc) 97 { 98 blk_opf_t flags = 0; 99 100 if (wbc->punt_to_cgroup) 101 flags = REQ_CGROUP_PUNT; 102 103 if (wbc->sync_mode == WB_SYNC_ALL) 104 flags |= REQ_SYNC; 105 else if (wbc->for_kupdate || wbc->for_background) 106 flags |= REQ_BACKGROUND; 107 108 return flags; 109 } 110 111 #ifdef CONFIG_CGROUP_WRITEBACK 112 #define wbc_blkcg_css(wbc) \ 113 ((wbc)->wb ? (wbc)->wb->blkcg_css : blkcg_root_css) 114 #else 115 #define wbc_blkcg_css(wbc) (blkcg_root_css) 116 #endif /* CONFIG_CGROUP_WRITEBACK */ 117 118 /* 119 * A wb_domain represents a domain that wb's (bdi_writeback's) belong to 120 * and are measured against each other in. There always is one global 121 * domain, global_wb_domain, that every wb in the system is a member of. 122 * This allows measuring the relative bandwidth of each wb to distribute 123 * dirtyable memory accordingly. 124 */ 125 struct wb_domain { 126 spinlock_t lock; 127 128 /* 129 * Scale the writeback cache size proportional to the relative 130 * writeout speed. 131 * 132 * We do this by keeping a floating proportion between BDIs, based 133 * on page writeback completions [end_page_writeback()]. Those 134 * devices that write out pages fastest will get the larger share, 135 * while the slower will get a smaller share. 136 * 137 * We use page writeout completions because we are interested in 138 * getting rid of dirty pages. Having them written out is the 139 * primary goal. 140 * 141 * We introduce a concept of time, a period over which we measure 142 * these events, because demand can/will vary over time. The length 143 * of this period itself is measured in page writeback completions. 144 */ 145 struct fprop_global completions; 146 struct timer_list period_timer; /* timer for aging of completions */ 147 unsigned long period_time; 148 149 /* 150 * The dirtyable memory and dirty threshold could be suddenly 151 * knocked down by a large amount (eg. on the startup of KVM in a 152 * swapless system). This may throw the system into deep dirty 153 * exceeded state and throttle heavy/light dirtiers alike. To 154 * retain good responsiveness, maintain global_dirty_limit for 155 * tracking slowly down to the knocked down dirty threshold. 156 * 157 * Both fields are protected by ->lock. 158 */ 159 unsigned long dirty_limit_tstamp; 160 unsigned long dirty_limit; 161 }; 162 163 /** 164 * wb_domain_size_changed - memory available to a wb_domain has changed 165 * @dom: wb_domain of interest 166 * 167 * This function should be called when the amount of memory available to 168 * @dom has changed. It resets @dom's dirty limit parameters to prevent 169 * the past values which don't match the current configuration from skewing 170 * dirty throttling. Without this, when memory size of a wb_domain is 171 * greatly reduced, the dirty throttling logic may allow too many pages to 172 * be dirtied leading to consecutive unnecessary OOMs and may get stuck in 173 * that situation. 174 */ 175 static inline void wb_domain_size_changed(struct wb_domain *dom) 176 { 177 spin_lock(&dom->lock); 178 dom->dirty_limit_tstamp = jiffies; 179 dom->dirty_limit = 0; 180 spin_unlock(&dom->lock); 181 } 182 183 /* 184 * fs/fs-writeback.c 185 */ 186 struct bdi_writeback; 187 void writeback_inodes_sb(struct super_block *, enum wb_reason reason); 188 void writeback_inodes_sb_nr(struct super_block *, unsigned long nr, 189 enum wb_reason reason); 190 void try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason); 191 void sync_inodes_sb(struct super_block *); 192 void wakeup_flusher_threads(enum wb_reason reason); 193 void wakeup_flusher_threads_bdi(struct backing_dev_info *bdi, 194 enum wb_reason reason); 195 void inode_wait_for_writeback(struct inode *inode); 196 void inode_io_list_del(struct inode *inode); 197 198 /* writeback.h requires fs.h; it, too, is not included from here. */ 199 static inline void wait_on_inode(struct inode *inode) 200 { 201 might_sleep(); 202 wait_on_bit(&inode->i_state, __I_NEW, TASK_UNINTERRUPTIBLE); 203 } 204 205 #ifdef CONFIG_CGROUP_WRITEBACK 206 207 #include <linux/cgroup.h> 208 #include <linux/bio.h> 209 210 void __inode_attach_wb(struct inode *inode, struct page *page); 211 void wbc_attach_and_unlock_inode(struct writeback_control *wbc, 212 struct inode *inode) 213 __releases(&inode->i_lock); 214 void wbc_detach_inode(struct writeback_control *wbc); 215 void wbc_account_cgroup_owner(struct writeback_control *wbc, struct page *page, 216 size_t bytes); 217 int cgroup_writeback_by_id(u64 bdi_id, int memcg_id, 218 enum wb_reason reason, struct wb_completion *done); 219 void cgroup_writeback_umount(void); 220 bool cleanup_offline_cgwb(struct bdi_writeback *wb); 221 222 /** 223 * inode_attach_wb - associate an inode with its wb 224 * @inode: inode of interest 225 * @page: page being dirtied (may be NULL) 226 * 227 * If @inode doesn't have its wb, associate it with the wb matching the 228 * memcg of @page or, if @page is NULL, %current. May be called w/ or w/o 229 * @inode->i_lock. 230 */ 231 static inline void inode_attach_wb(struct inode *inode, struct page *page) 232 { 233 if (!inode->i_wb) 234 __inode_attach_wb(inode, page); 235 } 236 237 /** 238 * inode_detach_wb - disassociate an inode from its wb 239 * @inode: inode of interest 240 * 241 * @inode is being freed. Detach from its wb. 242 */ 243 static inline void inode_detach_wb(struct inode *inode) 244 { 245 if (inode->i_wb) { 246 WARN_ON_ONCE(!(inode->i_state & I_CLEAR)); 247 wb_put(inode->i_wb); 248 inode->i_wb = NULL; 249 } 250 } 251 252 /** 253 * wbc_attach_fdatawrite_inode - associate wbc and inode for fdatawrite 254 * @wbc: writeback_control of interest 255 * @inode: target inode 256 * 257 * This function is to be used by __filemap_fdatawrite_range(), which is an 258 * alternative entry point into writeback code, and first ensures @inode is 259 * associated with a bdi_writeback and attaches it to @wbc. 260 */ 261 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc, 262 struct inode *inode) 263 { 264 spin_lock(&inode->i_lock); 265 inode_attach_wb(inode, NULL); 266 wbc_attach_and_unlock_inode(wbc, inode); 267 } 268 269 /** 270 * wbc_init_bio - writeback specific initializtion of bio 271 * @wbc: writeback_control for the writeback in progress 272 * @bio: bio to be initialized 273 * 274 * @bio is a part of the writeback in progress controlled by @wbc. Perform 275 * writeback specific initialization. This is used to apply the cgroup 276 * writeback context. Must be called after the bio has been associated with 277 * a device. 278 */ 279 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio) 280 { 281 /* 282 * pageout() path doesn't attach @wbc to the inode being written 283 * out. This is intentional as we don't want the function to block 284 * behind a slow cgroup. Ultimately, we want pageout() to kick off 285 * regular writeback instead of writing things out itself. 286 */ 287 if (wbc->wb) 288 bio_associate_blkg_from_css(bio, wbc->wb->blkcg_css); 289 } 290 291 #else /* CONFIG_CGROUP_WRITEBACK */ 292 293 static inline void inode_attach_wb(struct inode *inode, struct page *page) 294 { 295 } 296 297 static inline void inode_detach_wb(struct inode *inode) 298 { 299 } 300 301 static inline void wbc_attach_and_unlock_inode(struct writeback_control *wbc, 302 struct inode *inode) 303 __releases(&inode->i_lock) 304 { 305 spin_unlock(&inode->i_lock); 306 } 307 308 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc, 309 struct inode *inode) 310 { 311 } 312 313 static inline void wbc_detach_inode(struct writeback_control *wbc) 314 { 315 } 316 317 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio) 318 { 319 } 320 321 static inline void wbc_account_cgroup_owner(struct writeback_control *wbc, 322 struct page *page, size_t bytes) 323 { 324 } 325 326 static inline void cgroup_writeback_umount(void) 327 { 328 } 329 330 #endif /* CONFIG_CGROUP_WRITEBACK */ 331 332 /* 333 * mm/page-writeback.c 334 */ 335 void laptop_io_completion(struct backing_dev_info *info); 336 void laptop_sync_completion(void); 337 void laptop_mode_timer_fn(struct timer_list *t); 338 bool node_dirty_ok(struct pglist_data *pgdat); 339 int wb_domain_init(struct wb_domain *dom, gfp_t gfp); 340 #ifdef CONFIG_CGROUP_WRITEBACK 341 void wb_domain_exit(struct wb_domain *dom); 342 #endif 343 344 extern struct wb_domain global_wb_domain; 345 346 /* These are exported to sysctl. */ 347 extern unsigned int dirty_writeback_interval; 348 extern unsigned int dirty_expire_interval; 349 extern unsigned int dirtytime_expire_interval; 350 extern int laptop_mode; 351 352 int dirtytime_interval_handler(struct ctl_table *table, int write, 353 void *buffer, size_t *lenp, loff_t *ppos); 354 355 void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty); 356 unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh); 357 358 void wb_update_bandwidth(struct bdi_writeback *wb); 359 360 /* Invoke balance dirty pages in async mode. */ 361 #define BDP_ASYNC 0x0001 362 363 void balance_dirty_pages_ratelimited(struct address_space *mapping); 364 int balance_dirty_pages_ratelimited_flags(struct address_space *mapping, 365 unsigned int flags); 366 367 bool wb_over_bg_thresh(struct bdi_writeback *wb); 368 369 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc, 370 void *data); 371 372 int generic_writepages(struct address_space *mapping, 373 struct writeback_control *wbc); 374 void tag_pages_for_writeback(struct address_space *mapping, 375 pgoff_t start, pgoff_t end); 376 int write_cache_pages(struct address_space *mapping, 377 struct writeback_control *wbc, writepage_t writepage, 378 void *data); 379 int do_writepages(struct address_space *mapping, struct writeback_control *wbc); 380 void writeback_set_ratelimit(void); 381 void tag_pages_for_writeback(struct address_space *mapping, 382 pgoff_t start, pgoff_t end); 383 384 bool filemap_dirty_folio(struct address_space *mapping, struct folio *folio); 385 void folio_account_redirty(struct folio *folio); 386 static inline void account_page_redirty(struct page *page) 387 { 388 folio_account_redirty(page_folio(page)); 389 } 390 bool folio_redirty_for_writepage(struct writeback_control *, struct folio *); 391 bool redirty_page_for_writepage(struct writeback_control *, struct page *); 392 393 void sb_mark_inode_writeback(struct inode *inode); 394 void sb_clear_inode_writeback(struct inode *inode); 395 396 #endif /* WRITEBACK_H */ 397