1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef __LINUX_BACKING_DEV_DEFS_H 3 #define __LINUX_BACKING_DEV_DEFS_H 4 5 #include <linux/list.h> 6 #include <linux/radix-tree.h> 7 #include <linux/rbtree.h> 8 #include <linux/spinlock.h> 9 #include <linux/percpu_counter.h> 10 #include <linux/percpu-refcount.h> 11 #include <linux/flex_proportions.h> 12 #include <linux/timer.h> 13 #include <linux/workqueue.h> 14 #include <linux/kref.h> 15 #include <linux/refcount.h> 16 17 struct page; 18 struct device; 19 struct dentry; 20 21 /* 22 * Bits in bdi_writeback.state 23 */ 24 enum wb_state { 25 WB_registered, /* bdi_register() was done */ 26 WB_writeback_running, /* Writeback is in progress */ 27 WB_has_dirty_io, /* Dirty inodes on ->b_{dirty|io|more_io} */ 28 WB_start_all, /* nr_pages == 0 (all) work pending */ 29 }; 30 31 enum wb_congested_state { 32 WB_async_congested, /* The async (write) queue is getting full */ 33 WB_sync_congested, /* The sync queue is getting full */ 34 }; 35 36 typedef int (congested_fn)(void *, int); 37 38 enum wb_stat_item { 39 WB_RECLAIMABLE, 40 WB_WRITEBACK, 41 WB_DIRTIED, 42 WB_WRITTEN, 43 NR_WB_STAT_ITEMS 44 }; 45 46 #define WB_STAT_BATCH (8*(1+ilog2(nr_cpu_ids))) 47 48 /* 49 * why some writeback work was initiated 50 */ 51 enum wb_reason { 52 WB_REASON_BACKGROUND, 53 WB_REASON_VMSCAN, 54 WB_REASON_SYNC, 55 WB_REASON_PERIODIC, 56 WB_REASON_LAPTOP_TIMER, 57 WB_REASON_FREE_MORE_MEM, 58 WB_REASON_FS_FREE_SPACE, 59 /* 60 * There is no bdi forker thread any more and works are done 61 * by emergency worker, however, this is TPs userland visible 62 * and we'll be exposing exactly the same information, 63 * so it has a mismatch name. 64 */ 65 WB_REASON_FORKER_THREAD, 66 WB_REASON_FOREIGN_FLUSH, 67 68 WB_REASON_MAX, 69 }; 70 71 struct wb_completion { 72 atomic_t cnt; 73 wait_queue_head_t *waitq; 74 }; 75 76 #define __WB_COMPLETION_INIT(_waitq) \ 77 (struct wb_completion){ .cnt = ATOMIC_INIT(1), .waitq = (_waitq) } 78 79 /* 80 * If one wants to wait for one or more wb_writeback_works, each work's 81 * ->done should be set to a wb_completion defined using the following 82 * macro. Once all work items are issued with wb_queue_work(), the caller 83 * can wait for the completion of all using wb_wait_for_completion(). Work 84 * items which are waited upon aren't freed automatically on completion. 85 */ 86 #define WB_COMPLETION_INIT(bdi) __WB_COMPLETION_INIT(&(bdi)->wb_waitq) 87 88 #define DEFINE_WB_COMPLETION(cmpl, bdi) \ 89 struct wb_completion cmpl = WB_COMPLETION_INIT(bdi) 90 91 /* 92 * For cgroup writeback, multiple wb's may map to the same blkcg. Those 93 * wb's can operate mostly independently but should share the congested 94 * state. To facilitate such sharing, the congested state is tracked using 95 * the following struct which is created on demand, indexed by blkcg ID on 96 * its bdi, and refcounted. 97 */ 98 struct bdi_writeback_congested { 99 unsigned long state; /* WB_[a]sync_congested flags */ 100 refcount_t refcnt; /* nr of attached wb's and blkg */ 101 102 #ifdef CONFIG_CGROUP_WRITEBACK 103 struct backing_dev_info *__bdi; /* the associated bdi, set to NULL 104 * on bdi unregistration. For memcg-wb 105 * internal use only! */ 106 int blkcg_id; /* ID of the associated blkcg */ 107 struct rb_node rb_node; /* on bdi->cgwb_congestion_tree */ 108 #endif 109 }; 110 111 /* 112 * Each wb (bdi_writeback) can perform writeback operations, is measured 113 * and throttled, independently. Without cgroup writeback, each bdi 114 * (bdi_writeback) is served by its embedded bdi->wb. 115 * 116 * On the default hierarchy, blkcg implicitly enables memcg. This allows 117 * using memcg's page ownership for attributing writeback IOs, and every 118 * memcg - blkcg combination can be served by its own wb by assigning a 119 * dedicated wb to each memcg, which enables isolation across different 120 * cgroups and propagation of IO back pressure down from the IO layer upto 121 * the tasks which are generating the dirty pages to be written back. 122 * 123 * A cgroup wb is indexed on its bdi by the ID of the associated memcg, 124 * refcounted with the number of inodes attached to it, and pins the memcg 125 * and the corresponding blkcg. As the corresponding blkcg for a memcg may 126 * change as blkcg is disabled and enabled higher up in the hierarchy, a wb 127 * is tested for blkcg after lookup and removed from index on mismatch so 128 * that a new wb for the combination can be created. 129 */ 130 struct bdi_writeback { 131 struct backing_dev_info *bdi; /* our parent bdi */ 132 133 unsigned long state; /* Always use atomic bitops on this */ 134 unsigned long last_old_flush; /* last old data flush */ 135 136 struct list_head b_dirty; /* dirty inodes */ 137 struct list_head b_io; /* parked for writeback */ 138 struct list_head b_more_io; /* parked for more writeback */ 139 struct list_head b_dirty_time; /* time stamps are dirty */ 140 spinlock_t list_lock; /* protects the b_* lists */ 141 142 struct percpu_counter stat[NR_WB_STAT_ITEMS]; 143 144 struct bdi_writeback_congested *congested; 145 146 unsigned long bw_time_stamp; /* last time write bw is updated */ 147 unsigned long dirtied_stamp; 148 unsigned long written_stamp; /* pages written at bw_time_stamp */ 149 unsigned long write_bandwidth; /* the estimated write bandwidth */ 150 unsigned long avg_write_bandwidth; /* further smoothed write bw, > 0 */ 151 152 /* 153 * The base dirty throttle rate, re-calculated on every 200ms. 154 * All the bdi tasks' dirty rate will be curbed under it. 155 * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit 156 * in small steps and is much more smooth/stable than the latter. 157 */ 158 unsigned long dirty_ratelimit; 159 unsigned long balanced_dirty_ratelimit; 160 161 struct fprop_local_percpu completions; 162 int dirty_exceeded; 163 enum wb_reason start_all_reason; 164 165 spinlock_t work_lock; /* protects work_list & dwork scheduling */ 166 struct list_head work_list; 167 struct delayed_work dwork; /* work item used for writeback */ 168 169 unsigned long dirty_sleep; /* last wait */ 170 171 struct list_head bdi_node; /* anchored at bdi->wb_list */ 172 173 #ifdef CONFIG_CGROUP_WRITEBACK 174 struct percpu_ref refcnt; /* used only for !root wb's */ 175 struct fprop_local_percpu memcg_completions; 176 struct cgroup_subsys_state *memcg_css; /* the associated memcg */ 177 struct cgroup_subsys_state *blkcg_css; /* and blkcg */ 178 struct list_head memcg_node; /* anchored at memcg->cgwb_list */ 179 struct list_head blkcg_node; /* anchored at blkcg->cgwb_list */ 180 181 union { 182 struct work_struct release_work; 183 struct rcu_head rcu; 184 }; 185 #endif 186 }; 187 188 struct backing_dev_info { 189 u64 id; 190 struct rb_node rb_node; /* keyed by ->id */ 191 struct list_head bdi_list; 192 unsigned long ra_pages; /* max readahead in PAGE_SIZE units */ 193 unsigned long io_pages; /* max allowed IO size */ 194 congested_fn *congested_fn; /* Function pointer if device is md/dm */ 195 void *congested_data; /* Pointer to aux data for congested func */ 196 197 const char *name; 198 199 struct kref refcnt; /* Reference counter for the structure */ 200 unsigned int capabilities; /* Device capabilities */ 201 unsigned int min_ratio; 202 unsigned int max_ratio, max_prop_frac; 203 204 /* 205 * Sum of avg_write_bw of wbs with dirty inodes. > 0 if there are 206 * any dirty wbs, which is depended upon by bdi_has_dirty(). 207 */ 208 atomic_long_t tot_write_bandwidth; 209 210 struct bdi_writeback wb; /* the root writeback info for this bdi */ 211 struct list_head wb_list; /* list of all wbs */ 212 #ifdef CONFIG_CGROUP_WRITEBACK 213 struct radix_tree_root cgwb_tree; /* radix tree of active cgroup wbs */ 214 struct rb_root cgwb_congested_tree; /* their congested states */ 215 struct mutex cgwb_release_mutex; /* protect shutdown of wb structs */ 216 struct rw_semaphore wb_switch_rwsem; /* no cgwb switch while syncing */ 217 #else 218 struct bdi_writeback_congested *wb_congested; 219 #endif 220 wait_queue_head_t wb_waitq; 221 222 struct device *dev; 223 struct device *owner; 224 225 struct timer_list laptop_mode_wb_timer; 226 227 #ifdef CONFIG_DEBUG_FS 228 struct dentry *debug_dir; 229 #endif 230 }; 231 232 enum { 233 BLK_RW_ASYNC = 0, 234 BLK_RW_SYNC = 1, 235 }; 236 237 void clear_wb_congested(struct bdi_writeback_congested *congested, int sync); 238 void set_wb_congested(struct bdi_writeback_congested *congested, int sync); 239 240 static inline void clear_bdi_congested(struct backing_dev_info *bdi, int sync) 241 { 242 clear_wb_congested(bdi->wb.congested, sync); 243 } 244 245 static inline void set_bdi_congested(struct backing_dev_info *bdi, int sync) 246 { 247 set_wb_congested(bdi->wb.congested, sync); 248 } 249 250 struct wb_lock_cookie { 251 bool locked; 252 unsigned long flags; 253 }; 254 255 #ifdef CONFIG_CGROUP_WRITEBACK 256 257 /** 258 * wb_tryget - try to increment a wb's refcount 259 * @wb: bdi_writeback to get 260 */ 261 static inline bool wb_tryget(struct bdi_writeback *wb) 262 { 263 if (wb != &wb->bdi->wb) 264 return percpu_ref_tryget(&wb->refcnt); 265 return true; 266 } 267 268 /** 269 * wb_get - increment a wb's refcount 270 * @wb: bdi_writeback to get 271 */ 272 static inline void wb_get(struct bdi_writeback *wb) 273 { 274 if (wb != &wb->bdi->wb) 275 percpu_ref_get(&wb->refcnt); 276 } 277 278 /** 279 * wb_put - decrement a wb's refcount 280 * @wb: bdi_writeback to put 281 */ 282 static inline void wb_put(struct bdi_writeback *wb) 283 { 284 if (WARN_ON_ONCE(!wb->bdi)) { 285 /* 286 * A driver bug might cause a file to be removed before bdi was 287 * initialized. 288 */ 289 return; 290 } 291 292 if (wb != &wb->bdi->wb) 293 percpu_ref_put(&wb->refcnt); 294 } 295 296 /** 297 * wb_dying - is a wb dying? 298 * @wb: bdi_writeback of interest 299 * 300 * Returns whether @wb is unlinked and being drained. 301 */ 302 static inline bool wb_dying(struct bdi_writeback *wb) 303 { 304 return percpu_ref_is_dying(&wb->refcnt); 305 } 306 307 #else /* CONFIG_CGROUP_WRITEBACK */ 308 309 static inline bool wb_tryget(struct bdi_writeback *wb) 310 { 311 return true; 312 } 313 314 static inline void wb_get(struct bdi_writeback *wb) 315 { 316 } 317 318 static inline void wb_put(struct bdi_writeback *wb) 319 { 320 } 321 322 static inline bool wb_dying(struct bdi_writeback *wb) 323 { 324 return false; 325 } 326 327 #endif /* CONFIG_CGROUP_WRITEBACK */ 328 329 #endif /* __LINUX_BACKING_DEV_DEFS_H */ 330