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 1/4 region under the global dirty thresh is for smooth dirty throttling: 21 * 22 * (thresh - thresh/DIRTY_FULL_SCOPE, thresh) 23 * 24 * Further beyond, all dirtier tasks will enter a loop waiting (possibly long 25 * time) for the dirty pages to drop, unless written enough pages. 26 * 27 * The global dirty threshold is normally equal to the global dirty limit, 28 * except when the system suddenly allocates a lot of anonymous memory and 29 * knocks down the global dirty threshold quickly, in which case the global 30 * dirty limit will follow down slowly to prevent livelocking all dirtier tasks. 31 */ 32 #define DIRTY_SCOPE 8 33 #define DIRTY_FULL_SCOPE (DIRTY_SCOPE / 2) 34 35 struct backing_dev_info; 36 37 /* 38 * fs/fs-writeback.c 39 */ 40 enum writeback_sync_modes { 41 WB_SYNC_NONE, /* Don't wait on anything */ 42 WB_SYNC_ALL, /* Wait on every mapping */ 43 }; 44 45 /* 46 * A control structure which tells the writeback code what to do. These are 47 * always on the stack, and hence need no locking. They are always initialised 48 * in a manner such that unspecified fields are set to zero. 49 */ 50 struct writeback_control { 51 long nr_to_write; /* Write this many pages, and decrement 52 this for each page written */ 53 long pages_skipped; /* Pages which were not written */ 54 55 /* 56 * For a_ops->writepages(): if start or end are non-zero then this is 57 * a hint that the filesystem need only write out the pages inside that 58 * byterange. The byte at `end' is included in the writeout request. 59 */ 60 loff_t range_start; 61 loff_t range_end; 62 63 enum writeback_sync_modes sync_mode; 64 65 unsigned for_kupdate:1; /* A kupdate writeback */ 66 unsigned for_background:1; /* A background writeback */ 67 unsigned tagged_writepages:1; /* tag-and-write to avoid livelock */ 68 unsigned for_reclaim:1; /* Invoked from the page allocator */ 69 unsigned range_cyclic:1; /* range_start is cyclic */ 70 unsigned for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ 71 unsigned unpinned_fscache_wb:1; /* Cleared I_PINNING_FSCACHE_WB */ 72 73 /* 74 * When writeback IOs are bounced through async layers, only the 75 * initial synchronous phase should be accounted towards inode 76 * cgroup ownership arbitration to avoid confusion. Later stages 77 * can set the following flag to disable the accounting. 78 */ 79 unsigned no_cgroup_owner:1; 80 81 unsigned punt_to_cgroup:1; /* cgrp punting, see __REQ_CGROUP_PUNT */ 82 83 /* To enable batching of swap writes to non-block-device backends, 84 * "plug" can be set point to a 'struct swap_iocb *'. When all swap 85 * writes have been submitted, if with swap_iocb is not NULL, 86 * swap_write_unplug() should be called. 87 */ 88 struct swap_iocb **swap_plug; 89 90 #ifdef CONFIG_CGROUP_WRITEBACK 91 struct bdi_writeback *wb; /* wb this writeback is issued under */ 92 struct inode *inode; /* inode being written out */ 93 94 /* foreign inode detection, see wbc_detach_inode() */ 95 int wb_id; /* current wb id */ 96 int wb_lcand_id; /* last foreign candidate wb id */ 97 int wb_tcand_id; /* this foreign candidate wb id */ 98 size_t wb_bytes; /* bytes written by current wb */ 99 size_t wb_lcand_bytes; /* bytes written by last candidate */ 100 size_t wb_tcand_bytes; /* bytes written by this candidate */ 101 #endif 102 }; 103 104 static inline int wbc_to_write_flags(struct writeback_control *wbc) 105 { 106 int flags = 0; 107 108 if (wbc->punt_to_cgroup) 109 flags = REQ_CGROUP_PUNT; 110 111 if (wbc->sync_mode == WB_SYNC_ALL) 112 flags |= REQ_SYNC; 113 else if (wbc->for_kupdate || wbc->for_background) 114 flags |= REQ_BACKGROUND; 115 116 return flags; 117 } 118 119 #ifdef CONFIG_CGROUP_WRITEBACK 120 #define wbc_blkcg_css(wbc) \ 121 ((wbc)->wb ? (wbc)->wb->blkcg_css : blkcg_root_css) 122 #else 123 #define wbc_blkcg_css(wbc) (blkcg_root_css) 124 #endif /* CONFIG_CGROUP_WRITEBACK */ 125 126 /* 127 * A wb_domain represents a domain that wb's (bdi_writeback's) belong to 128 * and are measured against each other in. There always is one global 129 * domain, global_wb_domain, that every wb in the system is a member of. 130 * This allows measuring the relative bandwidth of each wb to distribute 131 * dirtyable memory accordingly. 132 */ 133 struct wb_domain { 134 spinlock_t lock; 135 136 /* 137 * Scale the writeback cache size proportional to the relative 138 * writeout speed. 139 * 140 * We do this by keeping a floating proportion between BDIs, based 141 * on page writeback completions [end_page_writeback()]. Those 142 * devices that write out pages fastest will get the larger share, 143 * while the slower will get a smaller share. 144 * 145 * We use page writeout completions because we are interested in 146 * getting rid of dirty pages. Having them written out is the 147 * primary goal. 148 * 149 * We introduce a concept of time, a period over which we measure 150 * these events, because demand can/will vary over time. The length 151 * of this period itself is measured in page writeback completions. 152 */ 153 struct fprop_global completions; 154 struct timer_list period_timer; /* timer for aging of completions */ 155 unsigned long period_time; 156 157 /* 158 * The dirtyable memory and dirty threshold could be suddenly 159 * knocked down by a large amount (eg. on the startup of KVM in a 160 * swapless system). This may throw the system into deep dirty 161 * exceeded state and throttle heavy/light dirtiers alike. To 162 * retain good responsiveness, maintain global_dirty_limit for 163 * tracking slowly down to the knocked down dirty threshold. 164 * 165 * Both fields are protected by ->lock. 166 */ 167 unsigned long dirty_limit_tstamp; 168 unsigned long dirty_limit; 169 }; 170 171 /** 172 * wb_domain_size_changed - memory available to a wb_domain has changed 173 * @dom: wb_domain of interest 174 * 175 * This function should be called when the amount of memory available to 176 * @dom has changed. It resets @dom's dirty limit parameters to prevent 177 * the past values which don't match the current configuration from skewing 178 * dirty throttling. Without this, when memory size of a wb_domain is 179 * greatly reduced, the dirty throttling logic may allow too many pages to 180 * be dirtied leading to consecutive unnecessary OOMs and may get stuck in 181 * that situation. 182 */ 183 static inline void wb_domain_size_changed(struct wb_domain *dom) 184 { 185 spin_lock(&dom->lock); 186 dom->dirty_limit_tstamp = jiffies; 187 dom->dirty_limit = 0; 188 spin_unlock(&dom->lock); 189 } 190 191 /* 192 * fs/fs-writeback.c 193 */ 194 struct bdi_writeback; 195 void writeback_inodes_sb(struct super_block *, enum wb_reason reason); 196 void writeback_inodes_sb_nr(struct super_block *, unsigned long nr, 197 enum wb_reason reason); 198 void try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason); 199 void sync_inodes_sb(struct super_block *); 200 void wakeup_flusher_threads(enum wb_reason reason); 201 void wakeup_flusher_threads_bdi(struct backing_dev_info *bdi, 202 enum wb_reason reason); 203 void inode_wait_for_writeback(struct inode *inode); 204 void inode_io_list_del(struct inode *inode); 205 206 /* writeback.h requires fs.h; it, too, is not included from here. */ 207 static inline void wait_on_inode(struct inode *inode) 208 { 209 might_sleep(); 210 wait_on_bit(&inode->i_state, __I_NEW, TASK_UNINTERRUPTIBLE); 211 } 212 213 #ifdef CONFIG_CGROUP_WRITEBACK 214 215 #include <linux/cgroup.h> 216 #include <linux/bio.h> 217 218 void __inode_attach_wb(struct inode *inode, struct page *page); 219 void wbc_attach_and_unlock_inode(struct writeback_control *wbc, 220 struct inode *inode) 221 __releases(&inode->i_lock); 222 void wbc_detach_inode(struct writeback_control *wbc); 223 void wbc_account_cgroup_owner(struct writeback_control *wbc, struct page *page, 224 size_t bytes); 225 int cgroup_writeback_by_id(u64 bdi_id, int memcg_id, 226 enum wb_reason reason, struct wb_completion *done); 227 void cgroup_writeback_umount(void); 228 bool cleanup_offline_cgwb(struct bdi_writeback *wb); 229 230 /** 231 * inode_attach_wb - associate an inode with its wb 232 * @inode: inode of interest 233 * @page: page being dirtied (may be NULL) 234 * 235 * If @inode doesn't have its wb, associate it with the wb matching the 236 * memcg of @page or, if @page is NULL, %current. May be called w/ or w/o 237 * @inode->i_lock. 238 */ 239 static inline void inode_attach_wb(struct inode *inode, struct page *page) 240 { 241 if (!inode->i_wb) 242 __inode_attach_wb(inode, page); 243 } 244 245 /** 246 * inode_detach_wb - disassociate an inode from its wb 247 * @inode: inode of interest 248 * 249 * @inode is being freed. Detach from its wb. 250 */ 251 static inline void inode_detach_wb(struct inode *inode) 252 { 253 if (inode->i_wb) { 254 WARN_ON_ONCE(!(inode->i_state & I_CLEAR)); 255 wb_put(inode->i_wb); 256 inode->i_wb = NULL; 257 } 258 } 259 260 /** 261 * wbc_attach_fdatawrite_inode - associate wbc and inode for fdatawrite 262 * @wbc: writeback_control of interest 263 * @inode: target inode 264 * 265 * This function is to be used by __filemap_fdatawrite_range(), which is an 266 * alternative entry point into writeback code, and first ensures @inode is 267 * associated with a bdi_writeback and attaches it to @wbc. 268 */ 269 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc, 270 struct inode *inode) 271 { 272 spin_lock(&inode->i_lock); 273 inode_attach_wb(inode, NULL); 274 wbc_attach_and_unlock_inode(wbc, inode); 275 } 276 277 /** 278 * wbc_init_bio - writeback specific initializtion of bio 279 * @wbc: writeback_control for the writeback in progress 280 * @bio: bio to be initialized 281 * 282 * @bio is a part of the writeback in progress controlled by @wbc. Perform 283 * writeback specific initialization. This is used to apply the cgroup 284 * writeback context. Must be called after the bio has been associated with 285 * a device. 286 */ 287 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio) 288 { 289 /* 290 * pageout() path doesn't attach @wbc to the inode being written 291 * out. This is intentional as we don't want the function to block 292 * behind a slow cgroup. Ultimately, we want pageout() to kick off 293 * regular writeback instead of writing things out itself. 294 */ 295 if (wbc->wb) 296 bio_associate_blkg_from_css(bio, wbc->wb->blkcg_css); 297 } 298 299 #else /* CONFIG_CGROUP_WRITEBACK */ 300 301 static inline void inode_attach_wb(struct inode *inode, struct page *page) 302 { 303 } 304 305 static inline void inode_detach_wb(struct inode *inode) 306 { 307 } 308 309 static inline void wbc_attach_and_unlock_inode(struct writeback_control *wbc, 310 struct inode *inode) 311 __releases(&inode->i_lock) 312 { 313 spin_unlock(&inode->i_lock); 314 } 315 316 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc, 317 struct inode *inode) 318 { 319 } 320 321 static inline void wbc_detach_inode(struct writeback_control *wbc) 322 { 323 } 324 325 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio) 326 { 327 } 328 329 static inline void wbc_account_cgroup_owner(struct writeback_control *wbc, 330 struct page *page, size_t bytes) 331 { 332 } 333 334 static inline void cgroup_writeback_umount(void) 335 { 336 } 337 338 #endif /* CONFIG_CGROUP_WRITEBACK */ 339 340 /* 341 * mm/page-writeback.c 342 */ 343 void laptop_io_completion(struct backing_dev_info *info); 344 void laptop_sync_completion(void); 345 void laptop_mode_timer_fn(struct timer_list *t); 346 bool node_dirty_ok(struct pglist_data *pgdat); 347 int wb_domain_init(struct wb_domain *dom, gfp_t gfp); 348 #ifdef CONFIG_CGROUP_WRITEBACK 349 void wb_domain_exit(struct wb_domain *dom); 350 #endif 351 352 extern struct wb_domain global_wb_domain; 353 354 /* These are exported to sysctl. */ 355 extern unsigned int dirty_writeback_interval; 356 extern unsigned int dirty_expire_interval; 357 extern unsigned int dirtytime_expire_interval; 358 extern int laptop_mode; 359 360 int dirtytime_interval_handler(struct ctl_table *table, int write, 361 void *buffer, size_t *lenp, loff_t *ppos); 362 363 void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty); 364 unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh); 365 366 void wb_update_bandwidth(struct bdi_writeback *wb); 367 void balance_dirty_pages_ratelimited(struct address_space *mapping); 368 bool wb_over_bg_thresh(struct bdi_writeback *wb); 369 370 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc, 371 void *data); 372 373 int generic_writepages(struct address_space *mapping, 374 struct writeback_control *wbc); 375 void tag_pages_for_writeback(struct address_space *mapping, 376 pgoff_t start, pgoff_t end); 377 int write_cache_pages(struct address_space *mapping, 378 struct writeback_control *wbc, writepage_t writepage, 379 void *data); 380 int do_writepages(struct address_space *mapping, struct writeback_control *wbc); 381 void writeback_set_ratelimit(void); 382 void tag_pages_for_writeback(struct address_space *mapping, 383 pgoff_t start, pgoff_t end); 384 385 bool filemap_dirty_folio(struct address_space *mapping, struct folio *folio); 386 void folio_account_redirty(struct folio *folio); 387 static inline void account_page_redirty(struct page *page) 388 { 389 folio_account_redirty(page_folio(page)); 390 } 391 bool folio_redirty_for_writepage(struct writeback_control *, struct folio *); 392 bool redirty_page_for_writepage(struct writeback_control *, struct page *); 393 394 void sb_mark_inode_writeback(struct inode *inode); 395 void sb_clear_inode_writeback(struct inode *inode); 396 397 #endif /* WRITEBACK_H */ 398