include/linux/backing-dev.h

   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/device.h>
  16 #include <linux/writeback.h>
  17 #include <linux/backing-dev-defs.h>
  18 #include <linux/slab.h>
  19
  20 static inline struct backing_dev_info *bdi_get(struct backing_dev_info *bdi)
  21 {
  22         kref_get(&bdi->refcnt);
  23         return bdi;
  24 }
  25
  26 struct backing_dev_info *bdi_get_by_id(u64 id);
  27 void bdi_put(struct backing_dev_info *bdi);
  28
  29 __printf(2, 3)
  30 int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...);
  31 __printf(2, 0)
  32 int bdi_register_va(struct backing_dev_info *bdi, const char *fmt,
  33                     va_list args);
  34 void bdi_set_owner(struct backing_dev_info *bdi, struct device *owner);
  35 void bdi_unregister(struct backing_dev_info *bdi);
  36
  37 struct backing_dev_info *bdi_alloc(int node_id);
  38
  39 void wb_start_background_writeback(struct bdi_writeback *wb);
  40 void wb_workfn(struct work_struct *work);
  41
  42 void wb_wait_for_completion(struct wb_completion *done);
  43
  44 extern spinlock_t bdi_lock;
  45 extern struct list_head bdi_list;
  46
  47 extern struct workqueue_struct *bdi_wq;
  48
  49 static inline bool wb_has_dirty_io(struct bdi_writeback *wb)
  50 {
  51         return test_bit(WB_has_dirty_io, &wb->state);
  52 }
  53
  54 static inline bool bdi_has_dirty_io(struct backing_dev_info *bdi)
  55 {
  56         /*
  57          * @bdi->tot_write_bandwidth is guaranteed to be > 0 if there are
  58          * any dirty wbs.  See wb_update_write_bandwidth().
  59          */
  60         return atomic_long_read(&bdi->tot_write_bandwidth);
  61 }
  62
  63 static inline void wb_stat_mod(struct bdi_writeback *wb,
  64                                  enum wb_stat_item item, s64 amount)
  65 {
  66         percpu_counter_add_batch(&wb->stat[item], amount, WB_STAT_BATCH);
  67 }
  68
  69 static inline void inc_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  70 {
  71         wb_stat_mod(wb, item, 1);
  72 }
  73
  74 static inline void dec_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  75 {
  76         wb_stat_mod(wb, item, -1);
  77 }
  78
  79 static inline s64 wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  80 {
  81         return percpu_counter_read_positive(&wb->stat[item]);
  82 }
  83
  84 static inline s64 wb_stat_sum(struct bdi_writeback *wb, enum wb_stat_item item)
  85 {
  86         return percpu_counter_sum_positive(&wb->stat[item]);
  87 }
  88
  89 extern void wb_writeout_inc(struct bdi_writeback *wb);
  90
  91 /*
  92  * maximal error of a stat counter.
  93  */
  94 static inline unsigned long wb_stat_error(void)
  95 {
  96 #ifdef CONFIG_SMP
  97         return nr_cpu_ids * WB_STAT_BATCH;
  98 #else
  99         return 1;
 100 #endif
 101 }
 102
 103 /* BDI ratio is expressed as part per 1000000 for finer granularity. */
 104 #define BDI_RATIO_SCALE 10000
 105
 106 u64 bdi_get_min_bytes(struct backing_dev_info *bdi);
 107 u64 bdi_get_max_bytes(struct backing_dev_info *bdi);
 108 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
 109 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
 110 int bdi_set_min_ratio_no_scale(struct backing_dev_info *bdi, unsigned int min_ratio);
 111 int bdi_set_max_ratio_no_scale(struct backing_dev_info *bdi, unsigned int max_ratio);
 112 int bdi_set_min_bytes(struct backing_dev_info *bdi, u64 min_bytes);
 113 int bdi_set_max_bytes(struct backing_dev_info *bdi, u64 max_bytes);
 114 int bdi_set_strict_limit(struct backing_dev_info *bdi, unsigned int strict_limit);
 115
 116 /*
 117  * Flags in backing_dev_info::capability
 118  *
 119  * BDI_CAP_WRITEBACK:           Supports dirty page writeback, and dirty pages
 120  *                              should contribute to accounting
 121  * BDI_CAP_WRITEBACK_ACCT:      Automatically account writeback pages
 122  * BDI_CAP_STRICTLIMIT:         Keep number of dirty pages below bdi threshold
 123  */
 124 #define BDI_CAP_WRITEBACK               (1 << 0)
 125 #define BDI_CAP_WRITEBACK_ACCT          (1 << 1)
 126 #define BDI_CAP_STRICTLIMIT             (1 << 2)
 127
 128 extern struct backing_dev_info noop_backing_dev_info;
 129
 130 int bdi_init(struct backing_dev_info *bdi);
 131
 132 /**
 133  * writeback_in_progress - determine whether there is writeback in progress
 134  * @wb: bdi_writeback of interest
 135  *
 136  * Determine whether there is writeback waiting to be handled against a
 137  * bdi_writeback.
 138  */
 139 static inline bool writeback_in_progress(struct bdi_writeback *wb)
 140 {
 141         return test_bit(WB_writeback_running, &wb->state);
 142 }
 143
 144 struct backing_dev_info *inode_to_bdi(struct inode *inode);
 145
 146 static inline bool mapping_can_writeback(struct address_space *mapping)
 147 {
 148         return inode_to_bdi(mapping->host)->capabilities & BDI_CAP_WRITEBACK;
 149 }
 150
 151 #ifdef CONFIG_CGROUP_WRITEBACK
 152
 153 struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi,
 154                                     struct cgroup_subsys_state *memcg_css);
 155 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
 156                                     struct cgroup_subsys_state *memcg_css,
 157                                     gfp_t gfp);
 158 void wb_memcg_offline(struct mem_cgroup *memcg);
 159 void wb_blkcg_offline(struct cgroup_subsys_state *css);
 160
 161 /**
 162  * inode_cgwb_enabled - test whether cgroup writeback is enabled on an inode
 163  * @inode: inode of interest
 164  *
 165  * Cgroup writeback requires support from the filesystem.  Also, both memcg and
 166  * iocg have to be on the default hierarchy.  Test whether all conditions are
 167  * met.
 168  *
 169  * Note that the test result may change dynamically on the same inode
 170  * depending on how memcg and iocg are configured.
 171  */
 172 static inline bool inode_cgwb_enabled(struct inode *inode)
 173 {
 174         struct backing_dev_info *bdi = inode_to_bdi(inode);
 175
 176         return cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
 177                 cgroup_subsys_on_dfl(io_cgrp_subsys) &&
 178                 (bdi->capabilities & BDI_CAP_WRITEBACK) &&
 179                 (inode->i_sb->s_iflags & SB_I_CGROUPWB);
 180 }
 181
 182 /**
 183  * wb_find_current - find wb for %current on a bdi
 184  * @bdi: bdi of interest
 185  *
 186  * Find the wb of @bdi which matches both the memcg and blkcg of %current.
 187  * Must be called under rcu_read_lock() which protects the returend wb.
 188  * NULL if not found.
 189  */
 190 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
 191 {
 192         struct cgroup_subsys_state *memcg_css;
 193         struct bdi_writeback *wb;
 194
 195         memcg_css = task_css(current, memory_cgrp_id);
 196         if (!memcg_css->parent)
 197                 return &bdi->wb;
 198
 199         wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
 200
 201         /*
 202          * %current's blkcg equals the effective blkcg of its memcg.  No
 203          * need to use the relatively expensive cgroup_get_e_css().
 204          */
 205         if (likely(wb && wb->blkcg_css == task_css(current, io_cgrp_id)))
 206                 return wb;
 207         return NULL;
 208 }
 209
 210 /**
 211  * wb_get_create_current - get or create wb for %current on a bdi
 212  * @bdi: bdi of interest
 213  * @gfp: allocation mask
 214  *
 215  * Equivalent to wb_get_create() on %current's memcg.  This function is
 216  * called from a relatively hot path and optimizes the common cases using
 217  * wb_find_current().
 218  */
 219 static inline struct bdi_writeback *
 220 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
 221 {
 222         struct bdi_writeback *wb;
 223
 224         rcu_read_lock();
 225         wb = wb_find_current(bdi);
 226         if (wb && unlikely(!wb_tryget(wb)))
 227                 wb = NULL;
 228         rcu_read_unlock();
 229
 230         if (unlikely(!wb)) {
 231                 struct cgroup_subsys_state *memcg_css;
 232
 233                 memcg_css = task_get_css(current, memory_cgrp_id);
 234                 wb = wb_get_create(bdi, memcg_css, gfp);
 235                 css_put(memcg_css);
 236         }
 237         return wb;
 238 }
 239
 240 /**
 241  * inode_to_wb - determine the wb of an inode
 242  * @inode: inode of interest
 243  *
 244  * Returns the wb @inode is currently associated with.  The caller must be
 245  * holding either @inode->i_lock, the i_pages lock, or the
 246  * associated wb's list_lock.
 247  */
 248 static inline struct bdi_writeback *inode_to_wb(const struct inode *inode)
 249 {
 250 #ifdef CONFIG_LOCKDEP
 251         WARN_ON_ONCE(debug_locks &&
 252                      (!lockdep_is_held(&inode->i_lock) &&
 253                       !lockdep_is_held(&inode->i_mapping->i_pages.xa_lock) &&
 254                       !lockdep_is_held(&inode->i_wb->list_lock)));
 255 #endif
 256         return inode->i_wb;
 257 }
 258
 259 static inline struct bdi_writeback *inode_to_wb_wbc(
 260                                 struct inode *inode,
 261                                 struct writeback_control *wbc)
 262 {
 263         /*
 264          * If wbc does not have inode attached, it means cgroup writeback was
 265          * disabled when wbc started. Just use the default wb in that case.
 266          */
 267         return wbc->wb ? wbc->wb : &inode_to_bdi(inode)->wb;
 268 }
 269
 270 /**
 271  * unlocked_inode_to_wb_begin - begin unlocked inode wb access transaction
 272  * @inode: target inode
 273  * @cookie: output param, to be passed to the end function
 274  *
 275  * The caller wants to access the wb associated with @inode but isn't
 276  * holding inode->i_lock, the i_pages lock or wb->list_lock.  This
 277  * function determines the wb associated with @inode and ensures that the
 278  * association doesn't change until the transaction is finished with
 279  * unlocked_inode_to_wb_end().
 280  *
 281  * The caller must call unlocked_inode_to_wb_end() with *@cookie afterwards and
 282  * can't sleep during the transaction.  IRQs may or may not be disabled on
 283  * return.
 284  */
 285 static inline struct bdi_writeback *
 286 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
 287 {
 288         rcu_read_lock();
 289
 290         /*
 291          * Paired with store_release in inode_switch_wbs_work_fn() and
 292          * ensures that we see the new wb if we see cleared I_WB_SWITCH.
 293          */
 294         cookie->locked = smp_load_acquire(&inode->i_state) & I_WB_SWITCH;
 295
 296         if (unlikely(cookie->locked))
 297                 xa_lock_irqsave(&inode->i_mapping->i_pages, cookie->flags);
 298
 299         /*
 300          * Protected by either !I_WB_SWITCH + rcu_read_lock() or the i_pages
 301          * lock.  inode_to_wb() will bark.  Deref directly.
 302          */
 303         return inode->i_wb;
 304 }
 305
 306 /**
 307  * unlocked_inode_to_wb_end - end inode wb access transaction
 308  * @inode: target inode
 309  * @cookie: @cookie from unlocked_inode_to_wb_begin()
 310  */
 311 static inline void unlocked_inode_to_wb_end(struct inode *inode,
 312                                             struct wb_lock_cookie *cookie)
 313 {
 314         if (unlikely(cookie->locked))
 315                 xa_unlock_irqrestore(&inode->i_mapping->i_pages, cookie->flags);
 316
 317         rcu_read_unlock();
 318 }
 319
 320 #else   /* CONFIG_CGROUP_WRITEBACK */
 321
 322 static inline bool inode_cgwb_enabled(struct inode *inode)
 323 {
 324         return false;
 325 }
 326
 327 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
 328 {
 329         return &bdi->wb;
 330 }
 331
 332 static inline struct bdi_writeback *
 333 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
 334 {
 335         return &bdi->wb;
 336 }
 337
 338 static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
 339 {
 340         return &inode_to_bdi(inode)->wb;
 341 }
 342
 343 static inline struct bdi_writeback *inode_to_wb_wbc(
 344                                 struct inode *inode,
 345                                 struct writeback_control *wbc)
 346 {
 347         return inode_to_wb(inode);
 348 }
 349
 350
 351 static inline struct bdi_writeback *
 352 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
 353 {
 354         return inode_to_wb(inode);
 355 }
 356
 357 static inline void unlocked_inode_to_wb_end(struct inode *inode,
 358                                             struct wb_lock_cookie *cookie)
 359 {
 360 }
 361
 362 static inline void wb_memcg_offline(struct mem_cgroup *memcg)
 363 {
 364 }
 365
 366 static inline void wb_blkcg_offline(struct cgroup_subsys_state *css)
 367 {
 368 }
 369
 370 #endif  /* CONFIG_CGROUP_WRITEBACK */
 371
 372 const char *bdi_dev_name(struct backing_dev_info *bdi);
 373
 374 #endif  /* _LINUX_BACKING_DEV_H */