5 #include <linux/blkdev.h>
6 #include <linux/closure.h>
7 #include <linux/errno.h>
8 #include <linux/blkdev.h>
9 #include <linux/freezer.h>
10 #include <linux/kernel.h>
11 #include <linux/llist.h>
12 #include <linux/log2.h>
13 #include <linux/ratelimit.h>
14 #include <linux/slab.h>
15 #include <linux/vmalloc.h>
16 #include <linux/workqueue.h>
18 #define PAGE_SECTOR_SHIFT (PAGE_SHIFT - 9)
19 #define PAGE_SECTORS (1UL << PAGE_SECTOR_SHIFT)
23 #ifdef CONFIG_BCACHEFS_DEBUG
25 #define EBUG_ON(cond) BUG_ON(cond)
26 #define atomic_dec_bug(v) BUG_ON(atomic_dec_return(v) < 0)
27 #define atomic_inc_bug(v, i) BUG_ON(atomic_inc_return(v) <= i)
28 #define atomic_sub_bug(i, v) BUG_ON(atomic_sub_return(i, v) < 0)
29 #define atomic_add_bug(i, v) BUG_ON(atomic_add_return(i, v) < 0)
30 #define atomic_long_dec_bug(v) BUG_ON(atomic_long_dec_return(v) < 0)
31 #define atomic_long_sub_bug(i, v) BUG_ON(atomic_long_sub_return(i, v) < 0)
32 #define atomic64_dec_bug(v) BUG_ON(atomic64_dec_return(v) < 0)
33 #define atomic64_inc_bug(v, i) BUG_ON(atomic64_inc_return(v) <= i)
34 #define atomic64_sub_bug(i, v) BUG_ON(atomic64_sub_return(i, v) < 0)
35 #define atomic64_add_bug(i, v) BUG_ON(atomic64_add_return(i, v) < 0)
37 #define memcpy(_dst, _src, _len) \
39 BUG_ON(!((void *) (_dst) >= (void *) (_src) + (_len) || \
40 (void *) (_dst) + (_len) <= (void *) (_src))); \
41 memcpy(_dst, _src, _len); \
47 #define atomic_dec_bug(v) atomic_dec(v)
48 #define atomic_inc_bug(v, i) atomic_inc(v)
49 #define atomic_sub_bug(i, v) atomic_sub(i, v)
50 #define atomic_add_bug(i, v) atomic_add(i, v)
51 #define atomic_long_dec_bug(v) atomic_long_dec(v)
52 #define atomic_long_sub_bug(i, v) atomic_long_sub(i, v)
53 #define atomic64_dec_bug(v) atomic64_dec(v)
54 #define atomic64_inc_bug(v, i) atomic64_inc(v)
55 #define atomic64_sub_bug(i, v) atomic64_sub(i, v)
56 #define atomic64_add_bug(i, v) atomic64_add(i, v)
61 #define __flatten __attribute__((flatten))
63 /* sparse doesn't know about attribute((flatten)) */
67 #ifdef __LITTLE_ENDIAN
68 #define CPU_BIG_ENDIAN 0
70 #define CPU_BIG_ENDIAN 1
75 #define type_is_exact(_val, _type) \
76 __builtin_types_compatible_p(typeof(_val), _type)
78 #define type_is(_val, _type) \
79 (__builtin_types_compatible_p(typeof(_val), _type) || \
80 __builtin_types_compatible_p(typeof(_val), const _type))
82 static inline void kvpfree(void *p, size_t size)
86 else if (is_vmalloc_addr(p))
89 free_pages((unsigned long) p, get_order(size));
93 static inline void *kvpmalloc(size_t size, gfp_t gfp_mask)
95 return size < PAGE_SIZE ? kmalloc(size, gfp_mask)
96 : (void *) __get_free_pages(gfp_mask, get_order(size))
97 ?: __vmalloc(size, gfp_mask, PAGE_KERNEL);
100 #define DECLARE_HEAP(type, name) \
106 #define init_heap(heap, _size, gfp) \
109 (heap)->size = (_size); \
110 (heap)->data = kvpmalloc((heap)->size * sizeof((heap)->data[0]),\
114 #define free_heap(heap) \
116 kvpfree((heap)->data, (heap)->size * sizeof((heap)->data[0])); \
117 (heap)->data = NULL; \
120 #define heap_swap(h, i, j) swap((h)->data[i], (h)->data[j])
122 #define heap_sift(h, i, cmp) \
126 for (; _j * 2 + 1 < (h)->used; _j = _r) { \
128 if (_r + 1 < (h)->used && \
129 cmp((h)->data[_r], (h)->data[_r + 1])) \
132 if (cmp((h)->data[_r], (h)->data[_j])) \
134 heap_swap(h, _r, _j); \
138 #define heap_sift_down(h, i, cmp) \
141 size_t p = (i - 1) / 2; \
142 if (cmp((h)->data[i], (h)->data[p])) \
144 heap_swap(h, i, p); \
149 #define heap_add(h, d, cmp) \
151 bool _r = !heap_full(h); \
153 size_t _i = (h)->used++; \
156 heap_sift_down(h, _i, cmp); \
157 heap_sift(h, _i, cmp); \
162 #define heap_del(h, i, cmp) \
166 BUG_ON(_i >= (h)->used); \
168 heap_swap(h, _i, (h)->used); \
169 heap_sift_down(h, _i, cmp); \
170 heap_sift(h, _i, cmp); \
173 #define heap_pop(h, d, cmp) \
175 bool _r = (h)->used; \
177 (d) = (h)->data[0]; \
178 heap_del(h, 0, cmp); \
183 #define heap_peek(h) \
185 EBUG_ON(!(h)->used); \
189 #define heap_full(h) ((h)->used == (h)->size)
191 #define heap_resort(heap, cmp) \
194 for (_i = (ssize_t) (heap)->used / 2 - 1; _i >= 0; --_i) \
195 heap_sift(heap, _i, cmp); \
199 * Simple array based allocator - preallocates a number of elements and you can
200 * never allocate more than that, also has no locking.
202 * Handy because if you know you only need a fixed number of elements you don't
203 * have to worry about memory allocation failure, and sometimes a mempool isn't
206 * We treat the free elements as entries in a singly linked list, and the
207 * freelist as a stack - allocating and freeing push and pop off the freelist.
210 #define DECLARE_ARRAY_ALLOCATOR(type, name, size) \
216 #define array_alloc(array) \
218 typeof((array)->freelist) _ret = (array)->freelist; \
221 (array)->freelist = *((typeof((array)->freelist) *) _ret);\
226 #define array_free(array, ptr) \
228 typeof((array)->freelist) _ptr = ptr; \
230 *((typeof((array)->freelist) *) _ptr) = (array)->freelist; \
231 (array)->freelist = _ptr; \
234 #define array_allocator_init(array) \
236 typeof((array)->freelist) _i; \
238 BUILD_BUG_ON(sizeof((array)->data[0]) < sizeof(void *)); \
239 (array)->freelist = NULL; \
241 for (_i = (array)->data; \
242 _i < (array)->data + ARRAY_SIZE((array)->data); \
244 array_free(array, _i); \
247 #define array_freelist_empty(array) ((array)->freelist == NULL)
249 #define ANYSINT_MAX(t) \
250 ((((t) 1 << (sizeof(t) * 8 - 2)) - (t) 1) * (t) 2 + (t) 1)
252 int bch2_strtoint_h(const char *, int *);
253 int bch2_strtouint_h(const char *, unsigned int *);
254 int bch2_strtoll_h(const char *, long long *);
255 int bch2_strtoull_h(const char *, unsigned long long *);
257 static inline int bch2_strtol_h(const char *cp, long *res)
259 #if BITS_PER_LONG == 32
260 return bch2_strtoint_h(cp, (int *) res);
262 return bch2_strtoll_h(cp, (long long *) res);
266 static inline int bch2_strtoul_h(const char *cp, long *res)
268 #if BITS_PER_LONG == 32
269 return bch2_strtouint_h(cp, (unsigned int *) res);
271 return bch2_strtoull_h(cp, (unsigned long long *) res);
275 #define strtoi_h(cp, res) \
276 ( type_is(*res, int) ? bch2_strtoint_h(cp, (void *) res)\
277 : type_is(*res, long) ? bch2_strtol_h(cp, (void *) res)\
278 : type_is(*res, long long) ? bch2_strtoll_h(cp, (void *) res)\
279 : type_is(*res, unsigned) ? bch2_strtouint_h(cp, (void *) res)\
280 : type_is(*res, unsigned long) ? bch2_strtoul_h(cp, (void *) res)\
281 : type_is(*res, unsigned long long) ? bch2_strtoull_h(cp, (void *) res)\
284 #define strtoul_safe(cp, var) \
287 int _r = kstrtoul(cp, 10, &_v); \
293 #define strtoul_safe_clamp(cp, var, min, max) \
296 int _r = kstrtoul(cp, 10, &_v); \
298 var = clamp_t(typeof(var), _v, min, max); \
302 #define strtoul_safe_restrict(cp, var, min, max) \
305 int _r = kstrtoul(cp, 10, &_v); \
306 if (!_r && _v >= min && _v <= max) \
313 #define snprint(buf, size, var) \
314 snprintf(buf, size, \
315 type_is(var, int) ? "%i\n" \
316 : type_is(var, unsigned) ? "%u\n" \
317 : type_is(var, long) ? "%li\n" \
318 : type_is(var, unsigned long) ? "%lu\n" \
319 : type_is(var, s64) ? "%lli\n" \
320 : type_is(var, u64) ? "%llu\n" \
321 : type_is(var, char *) ? "%s\n" \
324 ssize_t bch2_hprint(char *buf, s64 v);
326 bool bch2_is_zero(const void *, size_t);
328 ssize_t bch2_snprint_string_list(char *buf, size_t size, const char * const list[],
331 ssize_t bch2_read_string_list(const char *buf, const char * const list[]);
337 * all fields are in nanoseconds, averages are ewmas stored left shifted
342 u64 average_duration;
343 u64 average_frequency;
347 void bch2_time_stats_clear(struct time_stats *stats);
348 void __bch2_time_stats_update(struct time_stats *stats, u64 time);
349 void bch2_time_stats_update(struct time_stats *stats, u64 time);
351 static inline unsigned local_clock_us(void)
353 return local_clock() >> 10;
356 #define NSEC_PER_ns 1L
357 #define NSEC_PER_us NSEC_PER_USEC
358 #define NSEC_PER_ms NSEC_PER_MSEC
359 #define NSEC_PER_sec NSEC_PER_SEC
361 #define __print_time_stat(stats, name, stat, units) \
362 sysfs_print(name ## _ ## stat ## _ ## units, \
363 div_u64((stats)->stat >> 8, NSEC_PER_ ## units))
365 #define sysfs_print_time_stats(stats, name, \
369 __print_time_stat(stats, name, \
370 average_frequency, frequency_units); \
371 __print_time_stat(stats, name, \
372 average_duration, duration_units); \
373 sysfs_print(name ## _ ##count, (stats)->count); \
374 sysfs_print(name ## _ ##last_duration ## _ ## duration_units, \
375 div_u64((stats)->last_duration, \
376 NSEC_PER_ ## duration_units)); \
377 sysfs_print(name ## _ ##max_duration ## _ ## duration_units, \
378 div_u64((stats)->max_duration, \
379 NSEC_PER_ ## duration_units)); \
381 sysfs_print(name ## _last_ ## frequency_units, (stats)->last \
382 ? div_s64(local_clock() - (stats)->last, \
383 NSEC_PER_ ## frequency_units) \
387 #define sysfs_clear_time_stats(stats, name) \
389 if (attr == &sysfs_ ## name ## _clear) \
390 bch2_time_stats_clear(stats); \
393 #define sysfs_time_stats_attribute(name, \
396 write_attribute(name ## _clear); \
397 read_attribute(name ## _count); \
398 read_attribute(name ## _average_frequency_ ## frequency_units); \
399 read_attribute(name ## _average_duration_ ## duration_units); \
400 read_attribute(name ## _last_duration_ ## duration_units); \
401 read_attribute(name ## _max_duration_ ## duration_units); \
402 read_attribute(name ## _last_ ## frequency_units)
404 #define sysfs_time_stats_attribute_list(name, \
407 &sysfs_ ## name ## _clear, \
408 &sysfs_ ## name ## _count, \
409 &sysfs_ ## name ## _average_frequency_ ## frequency_units, \
410 &sysfs_ ## name ## _average_duration_ ## duration_units, \
411 &sysfs_ ## name ## _last_duration_ ## duration_units, \
412 &sysfs_ ## name ## _max_duration_ ## duration_units, \
413 &sysfs_ ## name ## _last_ ## frequency_units,
415 #define ewma_add(ewma, val, weight) \
417 typeof(ewma) _ewma = (ewma); \
418 typeof(weight) _weight = (weight); \
420 (((_ewma << _weight) - _ewma) + (val)) >> _weight; \
423 struct bch_ratelimit {
424 /* Next time we want to do some work, in nanoseconds */
428 * Rate at which we want to do work, in units per nanosecond
429 * The units here correspond to the units passed to
430 * bch2_ratelimit_increment()
435 static inline void bch2_ratelimit_reset(struct bch_ratelimit *d)
437 d->next = local_clock();
440 u64 bch2_ratelimit_delay(struct bch_ratelimit *);
441 void bch2_ratelimit_increment(struct bch_ratelimit *, u64);
442 int bch2_ratelimit_wait_freezable_stoppable(struct bch_ratelimit *);
444 struct bch_pd_controller {
445 struct bch_ratelimit rate;
446 unsigned long last_update;
449 s64 smoothed_derivative;
451 unsigned p_term_inverse;
455 /* for exporting to sysfs (no effect on behavior) */
457 s64 last_proportional;
461 /* If true, the rate will not increase if bch2_ratelimit_delay()
462 * is not being called often enough. */
466 void bch2_pd_controller_update(struct bch_pd_controller *, s64, s64, int);
467 void bch2_pd_controller_init(struct bch_pd_controller *);
468 size_t bch2_pd_controller_print_debug(struct bch_pd_controller *, char *);
470 #define sysfs_pd_controller_attribute(name) \
471 rw_attribute(name##_rate); \
472 rw_attribute(name##_rate_bytes); \
473 rw_attribute(name##_rate_d_term); \
474 rw_attribute(name##_rate_p_term_inverse); \
475 read_attribute(name##_rate_debug)
477 #define sysfs_pd_controller_files(name) \
478 &sysfs_##name##_rate, \
479 &sysfs_##name##_rate_bytes, \
480 &sysfs_##name##_rate_d_term, \
481 &sysfs_##name##_rate_p_term_inverse, \
482 &sysfs_##name##_rate_debug
484 #define sysfs_pd_controller_show(name, var) \
486 sysfs_hprint(name##_rate, (var)->rate.rate); \
487 sysfs_print(name##_rate_bytes, (var)->rate.rate); \
488 sysfs_print(name##_rate_d_term, (var)->d_term); \
489 sysfs_print(name##_rate_p_term_inverse, (var)->p_term_inverse); \
491 if (attr == &sysfs_##name##_rate_debug) \
492 return bch2_pd_controller_print_debug(var, buf); \
495 #define sysfs_pd_controller_store(name, var) \
497 sysfs_strtoul_clamp(name##_rate, \
498 (var)->rate.rate, 1, UINT_MAX); \
499 sysfs_strtoul_clamp(name##_rate_bytes, \
500 (var)->rate.rate, 1, UINT_MAX); \
501 sysfs_strtoul(name##_rate_d_term, (var)->d_term); \
502 sysfs_strtoul_clamp(name##_rate_p_term_inverse, \
503 (var)->p_term_inverse, 1, INT_MAX); \
506 #define __DIV_SAFE(n, d, zero) \
508 typeof(n) _n = (n); \
509 typeof(d) _d = (d); \
510 _d ? _n / _d : zero; \
513 #define DIV_SAFE(n, d) __DIV_SAFE(n, d, 0)
515 #define container_of_or_null(ptr, type, member) \
517 typeof(ptr) _ptr = ptr; \
518 _ptr ? container_of(_ptr, type, member) : NULL; \
521 #define RB_INSERT(root, new, member, cmp) \
524 struct rb_node **n = &(root)->rb_node, *parent = NULL; \
530 this = container_of(*n, typeof(*(new)), member); \
531 res = cmp(new, this); \
539 rb_link_node(&(new)->member, parent, n); \
540 rb_insert_color(&(new)->member, root); \
546 #define RB_SEARCH(root, search, member, cmp) \
548 struct rb_node *n = (root)->rb_node; \
549 typeof(&(search)) this, ret = NULL; \
553 this = container_of(n, typeof(search), member); \
554 res = cmp(&(search), this); \
566 #define RB_GREATER(root, search, member, cmp) \
568 struct rb_node *n = (root)->rb_node; \
569 typeof(&(search)) this, ret = NULL; \
573 this = container_of(n, typeof(search), member); \
574 res = cmp(&(search), this); \
584 #define RB_FIRST(root, type, member) \
585 container_of_or_null(rb_first(root), type, member)
587 #define RB_LAST(root, type, member) \
588 container_of_or_null(rb_last(root), type, member)
590 #define RB_NEXT(ptr, member) \
591 container_of_or_null(rb_next(&(ptr)->member), typeof(*ptr), member)
593 #define RB_PREV(ptr, member) \
594 container_of_or_null(rb_prev(&(ptr)->member), typeof(*ptr), member)
596 /* Does linear interpolation between powers of two */
597 static inline unsigned fract_exp_two(unsigned x, unsigned fract_bits)
599 unsigned fract = x & ~(~0 << fract_bits);
603 x += (x * fract) >> fract_bits;
608 void bch2_bio_map(struct bio *bio, void *base);
610 static inline sector_t bdev_sectors(struct block_device *bdev)
612 return bdev->bd_inode->i_size >> 9;
615 #define closure_bio_submit(bio, cl) \
621 #define kthread_wait_freezable(cond) \
625 set_current_state(TASK_INTERRUPTIBLE); \
626 if (kthread_should_stop()) { \
637 set_current_state(TASK_RUNNING); \
641 size_t bch2_rand_range(size_t);
643 void memcpy_to_bio(struct bio *, struct bvec_iter, void *);
644 void memcpy_from_bio(void *, struct bio *, struct bvec_iter);
646 static inline void __memcpy_u64s(void *dst, const void *src,
651 asm volatile("rep ; movsq"
652 : "=&c" (d0), "=&D" (d1), "=&S" (d2)
653 : "0" (u64s), "1" (dst), "2" (src)
664 static inline void memcpy_u64s(void *dst, const void *src,
667 EBUG_ON(!(dst >= src + u64s * sizeof(u64) ||
668 dst + u64s * sizeof(u64) <= src));
670 __memcpy_u64s(dst, src, u64s);
673 static inline void __memmove_u64s_down(void *dst, const void *src,
676 __memcpy_u64s(dst, src, u64s);
679 static inline void memmove_u64s_down(void *dst, const void *src,
684 __memmove_u64s_down(dst, src, u64s);
687 static inline void __memmove_u64s_up(void *_dst, const void *_src,
690 u64 *dst = (u64 *) _dst + u64s - 1;
691 u64 *src = (u64 *) _src + u64s - 1;
695 asm volatile("std ;\n"
698 : "=&c" (d0), "=&D" (d1), "=&S" (d2)
699 : "0" (u64s), "1" (dst), "2" (src)
707 static inline void memmove_u64s_up(void *dst, const void *src,
712 __memmove_u64s_up(dst, src, u64s);
715 static inline void memmove_u64s(void *dst, const void *src,
719 __memmove_u64s_down(dst, src, u64s);
721 __memmove_u64s_up(dst, src, u64s);
724 static inline struct bio_vec next_contig_bvec(struct bio *bio,
725 struct bvec_iter *iter)
727 struct bio_vec bv = bio_iter_iovec(bio, *iter);
729 bio_advance_iter(bio, iter, bv.bv_len);
730 #ifndef CONFIG_HIGHMEM
731 while (iter->bi_size) {
732 struct bio_vec next = bio_iter_iovec(bio, *iter);
734 if (page_address(bv.bv_page) + bv.bv_offset + bv.bv_len !=
735 page_address(next.bv_page) + next.bv_offset)
738 bv.bv_len += next.bv_len;
739 bio_advance_iter(bio, iter, next.bv_len);
745 #define __bio_for_each_contig_segment(bv, bio, iter, start) \
746 for (iter = (start); \
748 ((bv = next_contig_bvec((bio), &(iter))), 1);)
750 #define bio_for_each_contig_segment(bv, bio, iter) \
751 __bio_for_each_contig_segment(bv, bio, iter, (bio)->bi_iter)
753 #endif /* _BCACHE_UTIL_H */