~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/bcachefs/btree_cache.c

Version: ~ [ linux-6.11.5 ] ~ [ linux-6.10.14 ] ~ [ linux-6.9.12 ] ~ [ linux-6.8.12 ] ~ [ linux-6.7.12 ] ~ [ linux-6.6.58 ] ~ [ linux-6.5.13 ] ~ [ linux-6.4.16 ] ~ [ linux-6.3.13 ] ~ [ linux-6.2.16 ] ~ [ linux-6.1.114 ] ~ [ linux-6.0.19 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.169 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.228 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.284 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.322 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.336 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.337 ] ~ [ linux-4.4.302 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0
  2 
  3 #include "bcachefs.h"
  4 #include "bbpos.h"
  5 #include "bkey_buf.h"
  6 #include "btree_cache.h"
  7 #include "btree_io.h"
  8 #include "btree_iter.h"
  9 #include "btree_locking.h"
 10 #include "debug.h"
 11 #include "errcode.h"
 12 #include "error.h"
 13 #include "journal.h"
 14 #include "trace.h"
 15 
 16 #include <linux/prefetch.h>
 17 #include <linux/sched/mm.h>
 18 
 19 #define BTREE_CACHE_NOT_FREED_INCREMENT(counter) \
 20 do {                                             \
 21         if (shrinker_counter)                    \
 22                 bc->not_freed_##counter++;       \
 23 } while (0)
 24 
 25 const char * const bch2_btree_node_flags[] = {
 26 #define x(f)    #f,
 27         BTREE_FLAGS()
 28 #undef x
 29         NULL
 30 };
 31 
 32 void bch2_recalc_btree_reserve(struct bch_fs *c)
 33 {
 34         unsigned i, reserve = 16;
 35 
 36         if (!c->btree_roots_known[0].b)
 37                 reserve += 8;
 38 
 39         for (i = 0; i < btree_id_nr_alive(c); i++) {
 40                 struct btree_root *r = bch2_btree_id_root(c, i);
 41 
 42                 if (r->b)
 43                         reserve += min_t(unsigned, 1, r->b->c.level) * 8;
 44         }
 45 
 46         c->btree_cache.reserve = reserve;
 47 }
 48 
 49 static inline unsigned btree_cache_can_free(struct btree_cache *bc)
 50 {
 51         return max_t(int, 0, bc->used - bc->reserve);
 52 }
 53 
 54 static void btree_node_to_freedlist(struct btree_cache *bc, struct btree *b)
 55 {
 56         if (b->c.lock.readers)
 57                 list_move(&b->list, &bc->freed_pcpu);
 58         else
 59                 list_move(&b->list, &bc->freed_nonpcpu);
 60 }
 61 
 62 static void btree_node_data_free(struct bch_fs *c, struct btree *b)
 63 {
 64         struct btree_cache *bc = &c->btree_cache;
 65 
 66         EBUG_ON(btree_node_write_in_flight(b));
 67 
 68         clear_btree_node_just_written(b);
 69 
 70         kvfree(b->data);
 71         b->data = NULL;
 72 #ifdef __KERNEL__
 73         kvfree(b->aux_data);
 74 #else
 75         munmap(b->aux_data, btree_aux_data_bytes(b));
 76 #endif
 77         b->aux_data = NULL;
 78 
 79         bc->used--;
 80 
 81         btree_node_to_freedlist(bc, b);
 82 }
 83 
 84 static int bch2_btree_cache_cmp_fn(struct rhashtable_compare_arg *arg,
 85                                    const void *obj)
 86 {
 87         const struct btree *b = obj;
 88         const u64 *v = arg->key;
 89 
 90         return b->hash_val == *v ? 0 : 1;
 91 }
 92 
 93 static const struct rhashtable_params bch_btree_cache_params = {
 94         .head_offset            = offsetof(struct btree, hash),
 95         .key_offset             = offsetof(struct btree, hash_val),
 96         .key_len                = sizeof(u64),
 97         .obj_cmpfn              = bch2_btree_cache_cmp_fn,
 98         .automatic_shrinking    = true,
 99 };
100 
101 static int btree_node_data_alloc(struct bch_fs *c, struct btree *b, gfp_t gfp)
102 {
103         BUG_ON(b->data || b->aux_data);
104 
105         b->data = kvmalloc(btree_buf_bytes(b), gfp);
106         if (!b->data)
107                 return -BCH_ERR_ENOMEM_btree_node_mem_alloc;
108 #ifdef __KERNEL__
109         b->aux_data = kvmalloc(btree_aux_data_bytes(b), gfp);
110 #else
111         b->aux_data = mmap(NULL, btree_aux_data_bytes(b),
112                            PROT_READ|PROT_WRITE|PROT_EXEC,
113                            MAP_PRIVATE|MAP_ANONYMOUS, 0, 0);
114         if (b->aux_data == MAP_FAILED)
115                 b->aux_data = NULL;
116 #endif
117         if (!b->aux_data) {
118                 kvfree(b->data);
119                 b->data = NULL;
120                 return -BCH_ERR_ENOMEM_btree_node_mem_alloc;
121         }
122 
123         return 0;
124 }
125 
126 static struct btree *__btree_node_mem_alloc(struct bch_fs *c, gfp_t gfp)
127 {
128         struct btree *b;
129 
130         b = kzalloc(sizeof(struct btree), gfp);
131         if (!b)
132                 return NULL;
133 
134         bkey_btree_ptr_init(&b->key);
135         INIT_LIST_HEAD(&b->list);
136         INIT_LIST_HEAD(&b->write_blocked);
137         b->byte_order = ilog2(c->opts.btree_node_size);
138         return b;
139 }
140 
141 struct btree *__bch2_btree_node_mem_alloc(struct bch_fs *c)
142 {
143         struct btree_cache *bc = &c->btree_cache;
144         struct btree *b;
145 
146         b = __btree_node_mem_alloc(c, GFP_KERNEL);
147         if (!b)
148                 return NULL;
149 
150         if (btree_node_data_alloc(c, b, GFP_KERNEL)) {
151                 kfree(b);
152                 return NULL;
153         }
154 
155         bch2_btree_lock_init(&b->c, 0);
156 
157         bc->used++;
158         list_add(&b->list, &bc->freeable);
159         return b;
160 }
161 
162 void bch2_btree_node_to_freelist(struct bch_fs *c, struct btree *b)
163 {
164         mutex_lock(&c->btree_cache.lock);
165         list_move(&b->list, &c->btree_cache.freeable);
166         mutex_unlock(&c->btree_cache.lock);
167 
168         six_unlock_write(&b->c.lock);
169         six_unlock_intent(&b->c.lock);
170 }
171 
172 /* Btree in memory cache - hash table */
173 
174 void bch2_btree_node_hash_remove(struct btree_cache *bc, struct btree *b)
175 {
176         int ret = rhashtable_remove_fast(&bc->table, &b->hash, bch_btree_cache_params);
177 
178         BUG_ON(ret);
179 
180         /* Cause future lookups for this node to fail: */
181         b->hash_val = 0;
182 
183         if (b->c.btree_id < BTREE_ID_NR)
184                 --bc->used_by_btree[b->c.btree_id];
185 }
186 
187 int __bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b)
188 {
189         BUG_ON(b->hash_val);
190         b->hash_val = btree_ptr_hash_val(&b->key);
191 
192         int ret = rhashtable_lookup_insert_fast(&bc->table, &b->hash,
193                                                 bch_btree_cache_params);
194         if (!ret && b->c.btree_id < BTREE_ID_NR)
195                 bc->used_by_btree[b->c.btree_id]++;
196         return ret;
197 }
198 
199 int bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b,
200                                 unsigned level, enum btree_id id)
201 {
202         int ret;
203 
204         b->c.level      = level;
205         b->c.btree_id   = id;
206 
207         mutex_lock(&bc->lock);
208         ret = __bch2_btree_node_hash_insert(bc, b);
209         if (!ret)
210                 list_add_tail(&b->list, &bc->live);
211         mutex_unlock(&bc->lock);
212 
213         return ret;
214 }
215 
216 void bch2_btree_node_update_key_early(struct btree_trans *trans,
217                                       enum btree_id btree, unsigned level,
218                                       struct bkey_s_c old, struct bkey_i *new)
219 {
220         struct bch_fs *c = trans->c;
221         struct btree *b;
222         struct bkey_buf tmp;
223         int ret;
224 
225         bch2_bkey_buf_init(&tmp);
226         bch2_bkey_buf_reassemble(&tmp, c, old);
227 
228         b = bch2_btree_node_get_noiter(trans, tmp.k, btree, level, true);
229         if (!IS_ERR_OR_NULL(b)) {
230                 mutex_lock(&c->btree_cache.lock);
231 
232                 bch2_btree_node_hash_remove(&c->btree_cache, b);
233 
234                 bkey_copy(&b->key, new);
235                 ret = __bch2_btree_node_hash_insert(&c->btree_cache, b);
236                 BUG_ON(ret);
237 
238                 mutex_unlock(&c->btree_cache.lock);
239                 six_unlock_read(&b->c.lock);
240         }
241 
242         bch2_bkey_buf_exit(&tmp, c);
243 }
244 
245 __flatten
246 static inline struct btree *btree_cache_find(struct btree_cache *bc,
247                                      const struct bkey_i *k)
248 {
249         u64 v = btree_ptr_hash_val(k);
250 
251         return rhashtable_lookup_fast(&bc->table, &v, bch_btree_cache_params);
252 }
253 
254 /*
255  * this version is for btree nodes that have already been freed (we're not
256  * reaping a real btree node)
257  */
258 static int __btree_node_reclaim(struct bch_fs *c, struct btree *b, bool flush, bool shrinker_counter)
259 {
260         struct btree_cache *bc = &c->btree_cache;
261         int ret = 0;
262 
263         lockdep_assert_held(&bc->lock);
264 
265         struct bbpos pos = BBPOS(b->c.btree_id, b->key.k.p);
266 
267         u64 mask = b->c.level
268                 ? bc->pinned_nodes_interior_mask
269                 : bc->pinned_nodes_leaf_mask;
270 
271         if ((mask & BIT_ULL(b->c.btree_id)) &&
272             bbpos_cmp(bc->pinned_nodes_start, pos) < 0 &&
273             bbpos_cmp(bc->pinned_nodes_end, pos) >= 0)
274                 return -BCH_ERR_ENOMEM_btree_node_reclaim;
275 
276 wait_on_io:
277         if (b->flags & ((1U << BTREE_NODE_dirty)|
278                         (1U << BTREE_NODE_read_in_flight)|
279                         (1U << BTREE_NODE_write_in_flight))) {
280                 if (!flush) {
281                         if (btree_node_dirty(b))
282                                 BTREE_CACHE_NOT_FREED_INCREMENT(dirty);
283                         else if (btree_node_read_in_flight(b))
284                                 BTREE_CACHE_NOT_FREED_INCREMENT(read_in_flight);
285                         else if (btree_node_write_in_flight(b))
286                                 BTREE_CACHE_NOT_FREED_INCREMENT(write_in_flight);
287                         return -BCH_ERR_ENOMEM_btree_node_reclaim;
288                 }
289 
290                 /* XXX: waiting on IO with btree cache lock held */
291                 bch2_btree_node_wait_on_read(b);
292                 bch2_btree_node_wait_on_write(b);
293         }
294 
295         if (!six_trylock_intent(&b->c.lock)) {
296                 BTREE_CACHE_NOT_FREED_INCREMENT(lock_intent);
297                 return -BCH_ERR_ENOMEM_btree_node_reclaim;
298         }
299 
300         if (!six_trylock_write(&b->c.lock)) {
301                 BTREE_CACHE_NOT_FREED_INCREMENT(lock_write);
302                 goto out_unlock_intent;
303         }
304 
305         /* recheck under lock */
306         if (b->flags & ((1U << BTREE_NODE_read_in_flight)|
307                         (1U << BTREE_NODE_write_in_flight))) {
308                 if (!flush) {
309                         if (btree_node_read_in_flight(b))
310                                 BTREE_CACHE_NOT_FREED_INCREMENT(read_in_flight);
311                         else if (btree_node_write_in_flight(b))
312                                 BTREE_CACHE_NOT_FREED_INCREMENT(write_in_flight);
313                         goto out_unlock;
314                 }
315                 six_unlock_write(&b->c.lock);
316                 six_unlock_intent(&b->c.lock);
317                 goto wait_on_io;
318         }
319 
320         if (btree_node_noevict(b)) {
321                 BTREE_CACHE_NOT_FREED_INCREMENT(noevict);
322                 goto out_unlock;
323         }
324         if (btree_node_write_blocked(b)) {
325                 BTREE_CACHE_NOT_FREED_INCREMENT(write_blocked);
326                 goto out_unlock;
327         }
328         if (btree_node_will_make_reachable(b)) {
329                 BTREE_CACHE_NOT_FREED_INCREMENT(will_make_reachable);
330                 goto out_unlock;
331         }
332 
333         if (btree_node_dirty(b)) {
334                 if (!flush) {
335                         BTREE_CACHE_NOT_FREED_INCREMENT(dirty);
336                         goto out_unlock;
337                 }
338                 /*
339                  * Using the underscore version because we don't want to compact
340                  * bsets after the write, since this node is about to be evicted
341                  * - unless btree verify mode is enabled, since it runs out of
342                  * the post write cleanup:
343                  */
344                 if (bch2_verify_btree_ondisk)
345                         bch2_btree_node_write(c, b, SIX_LOCK_intent,
346                                               BTREE_WRITE_cache_reclaim);
347                 else
348                         __bch2_btree_node_write(c, b,
349                                                 BTREE_WRITE_cache_reclaim);
350 
351                 six_unlock_write(&b->c.lock);
352                 six_unlock_intent(&b->c.lock);
353                 goto wait_on_io;
354         }
355 out:
356         if (b->hash_val && !ret)
357                 trace_and_count(c, btree_cache_reap, c, b);
358         return ret;
359 out_unlock:
360         six_unlock_write(&b->c.lock);
361 out_unlock_intent:
362         six_unlock_intent(&b->c.lock);
363         ret = -BCH_ERR_ENOMEM_btree_node_reclaim;
364         goto out;
365 }
366 
367 static int btree_node_reclaim(struct bch_fs *c, struct btree *b, bool shrinker_counter)
368 {
369         return __btree_node_reclaim(c, b, false, shrinker_counter);
370 }
371 
372 static int btree_node_write_and_reclaim(struct bch_fs *c, struct btree *b)
373 {
374         return __btree_node_reclaim(c, b, true, false);
375 }
376 
377 static unsigned long bch2_btree_cache_scan(struct shrinker *shrink,
378                                            struct shrink_control *sc)
379 {
380         struct bch_fs *c = shrink->private_data;
381         struct btree_cache *bc = &c->btree_cache;
382         struct btree *b, *t;
383         unsigned long nr = sc->nr_to_scan;
384         unsigned long can_free = 0;
385         unsigned long freed = 0;
386         unsigned long touched = 0;
387         unsigned i, flags;
388         unsigned long ret = SHRINK_STOP;
389         bool trigger_writes = atomic_read(&bc->dirty) + nr >=
390                 bc->used * 3 / 4;
391 
392         if (bch2_btree_shrinker_disabled)
393                 return SHRINK_STOP;
394 
395         mutex_lock(&bc->lock);
396         flags = memalloc_nofs_save();
397 
398         /*
399          * It's _really_ critical that we don't free too many btree nodes - we
400          * have to always leave ourselves a reserve. The reserve is how we
401          * guarantee that allocating memory for a new btree node can always
402          * succeed, so that inserting keys into the btree can always succeed and
403          * IO can always make forward progress:
404          */
405         can_free = btree_cache_can_free(bc);
406         nr = min_t(unsigned long, nr, can_free);
407 
408         i = 0;
409         list_for_each_entry_safe(b, t, &bc->freeable, list) {
410                 /*
411                  * Leave a few nodes on the freeable list, so that a btree split
412                  * won't have to hit the system allocator:
413                  */
414                 if (++i <= 3)
415                         continue;
416 
417                 touched++;
418 
419                 if (touched >= nr)
420                         goto out;
421 
422                 if (!btree_node_reclaim(c, b, true)) {
423                         btree_node_data_free(c, b);
424                         six_unlock_write(&b->c.lock);
425                         six_unlock_intent(&b->c.lock);
426                         freed++;
427                         bc->freed++;
428                 }
429         }
430 restart:
431         list_for_each_entry_safe(b, t, &bc->live, list) {
432                 touched++;
433 
434                 if (btree_node_accessed(b)) {
435                         clear_btree_node_accessed(b);
436                         bc->not_freed_access_bit++;
437                 } else if (!btree_node_reclaim(c, b, true)) {
438                         freed++;
439                         btree_node_data_free(c, b);
440                         bc->freed++;
441 
442                         bch2_btree_node_hash_remove(bc, b);
443                         six_unlock_write(&b->c.lock);
444                         six_unlock_intent(&b->c.lock);
445 
446                         if (freed == nr)
447                                 goto out_rotate;
448                 } else if (trigger_writes &&
449                            btree_node_dirty(b) &&
450                            !btree_node_will_make_reachable(b) &&
451                            !btree_node_write_blocked(b) &&
452                            six_trylock_read(&b->c.lock)) {
453                         list_move(&bc->live, &b->list);
454                         mutex_unlock(&bc->lock);
455                         __bch2_btree_node_write(c, b, BTREE_WRITE_cache_reclaim);
456                         six_unlock_read(&b->c.lock);
457                         if (touched >= nr)
458                                 goto out_nounlock;
459                         mutex_lock(&bc->lock);
460                         goto restart;
461                 }
462 
463                 if (touched >= nr)
464                         break;
465         }
466 out_rotate:
467         if (&t->list != &bc->live)
468                 list_move_tail(&bc->live, &t->list);
469 out:
470         mutex_unlock(&bc->lock);
471 out_nounlock:
472         ret = freed;
473         memalloc_nofs_restore(flags);
474         trace_and_count(c, btree_cache_scan, sc->nr_to_scan, can_free, ret);
475         return ret;
476 }
477 
478 static unsigned long bch2_btree_cache_count(struct shrinker *shrink,
479                                             struct shrink_control *sc)
480 {
481         struct bch_fs *c = shrink->private_data;
482         struct btree_cache *bc = &c->btree_cache;
483 
484         if (bch2_btree_shrinker_disabled)
485                 return 0;
486 
487         return btree_cache_can_free(bc);
488 }
489 
490 void bch2_fs_btree_cache_exit(struct bch_fs *c)
491 {
492         struct btree_cache *bc = &c->btree_cache;
493         struct btree *b;
494         unsigned i, flags;
495 
496         shrinker_free(bc->shrink);
497 
498         /* vfree() can allocate memory: */
499         flags = memalloc_nofs_save();
500         mutex_lock(&bc->lock);
501 
502         if (c->verify_data)
503                 list_move(&c->verify_data->list, &bc->live);
504 
505         kvfree(c->verify_ondisk);
506 
507         for (i = 0; i < btree_id_nr_alive(c); i++) {
508                 struct btree_root *r = bch2_btree_id_root(c, i);
509 
510                 if (r->b)
511                         list_add(&r->b->list, &bc->live);
512         }
513 
514         list_splice(&bc->freeable, &bc->live);
515 
516         while (!list_empty(&bc->live)) {
517                 b = list_first_entry(&bc->live, struct btree, list);
518 
519                 BUG_ON(btree_node_read_in_flight(b) ||
520                        btree_node_write_in_flight(b));
521 
522                 btree_node_data_free(c, b);
523         }
524 
525         BUG_ON(!bch2_journal_error(&c->journal) &&
526                atomic_read(&c->btree_cache.dirty));
527 
528         list_splice(&bc->freed_pcpu, &bc->freed_nonpcpu);
529 
530         while (!list_empty(&bc->freed_nonpcpu)) {
531                 b = list_first_entry(&bc->freed_nonpcpu, struct btree, list);
532                 list_del(&b->list);
533                 six_lock_exit(&b->c.lock);
534                 kfree(b);
535         }
536 
537         mutex_unlock(&bc->lock);
538         memalloc_nofs_restore(flags);
539 
540         if (bc->table_init_done)
541                 rhashtable_destroy(&bc->table);
542 }
543 
544 int bch2_fs_btree_cache_init(struct bch_fs *c)
545 {
546         struct btree_cache *bc = &c->btree_cache;
547         struct shrinker *shrink;
548         unsigned i;
549         int ret = 0;
550 
551         ret = rhashtable_init(&bc->table, &bch_btree_cache_params);
552         if (ret)
553                 goto err;
554 
555         bc->table_init_done = true;
556 
557         bch2_recalc_btree_reserve(c);
558 
559         for (i = 0; i < bc->reserve; i++)
560                 if (!__bch2_btree_node_mem_alloc(c))
561                         goto err;
562 
563         list_splice_init(&bc->live, &bc->freeable);
564 
565         mutex_init(&c->verify_lock);
566 
567         shrink = shrinker_alloc(0, "%s-btree_cache", c->name);
568         if (!shrink)
569                 goto err;
570         bc->shrink = shrink;
571         shrink->count_objects   = bch2_btree_cache_count;
572         shrink->scan_objects    = bch2_btree_cache_scan;
573         shrink->seeks           = 4;
574         shrink->private_data    = c;
575         shrinker_register(shrink);
576 
577         return 0;
578 err:
579         return -BCH_ERR_ENOMEM_fs_btree_cache_init;
580 }
581 
582 void bch2_fs_btree_cache_init_early(struct btree_cache *bc)
583 {
584         mutex_init(&bc->lock);
585         INIT_LIST_HEAD(&bc->live);
586         INIT_LIST_HEAD(&bc->freeable);
587         INIT_LIST_HEAD(&bc->freed_pcpu);
588         INIT_LIST_HEAD(&bc->freed_nonpcpu);
589 }
590 
591 /*
592  * We can only have one thread cannibalizing other cached btree nodes at a time,
593  * or we'll deadlock. We use an open coded mutex to ensure that, which a
594  * cannibalize_bucket() will take. This means every time we unlock the root of
595  * the btree, we need to release this lock if we have it held.
596  */
597 void bch2_btree_cache_cannibalize_unlock(struct btree_trans *trans)
598 {
599         struct bch_fs *c = trans->c;
600         struct btree_cache *bc = &c->btree_cache;
601 
602         if (bc->alloc_lock == current) {
603                 trace_and_count(c, btree_cache_cannibalize_unlock, trans);
604                 bc->alloc_lock = NULL;
605                 closure_wake_up(&bc->alloc_wait);
606         }
607 }
608 
609 int bch2_btree_cache_cannibalize_lock(struct btree_trans *trans, struct closure *cl)
610 {
611         struct bch_fs *c = trans->c;
612         struct btree_cache *bc = &c->btree_cache;
613         struct task_struct *old;
614 
615         old = NULL;
616         if (try_cmpxchg(&bc->alloc_lock, &old, current) || old == current)
617                 goto success;
618 
619         if (!cl) {
620                 trace_and_count(c, btree_cache_cannibalize_lock_fail, trans);
621                 return -BCH_ERR_ENOMEM_btree_cache_cannibalize_lock;
622         }
623 
624         closure_wait(&bc->alloc_wait, cl);
625 
626         /* Try again, after adding ourselves to waitlist */
627         old = NULL;
628         if (try_cmpxchg(&bc->alloc_lock, &old, current) || old == current) {
629                 /* We raced */
630                 closure_wake_up(&bc->alloc_wait);
631                 goto success;
632         }
633 
634         trace_and_count(c, btree_cache_cannibalize_lock_fail, trans);
635         return -BCH_ERR_btree_cache_cannibalize_lock_blocked;
636 
637 success:
638         trace_and_count(c, btree_cache_cannibalize_lock, trans);
639         return 0;
640 }
641 
642 static struct btree *btree_node_cannibalize(struct bch_fs *c)
643 {
644         struct btree_cache *bc = &c->btree_cache;
645         struct btree *b;
646 
647         list_for_each_entry_reverse(b, &bc->live, list)
648                 if (!btree_node_reclaim(c, b, false))
649                         return b;
650 
651         while (1) {
652                 list_for_each_entry_reverse(b, &bc->live, list)
653                         if (!btree_node_write_and_reclaim(c, b))
654                                 return b;
655 
656                 /*
657                  * Rare case: all nodes were intent-locked.
658                  * Just busy-wait.
659                  */
660                 WARN_ONCE(1, "btree cache cannibalize failed\n");
661                 cond_resched();
662         }
663 }
664 
665 struct btree *bch2_btree_node_mem_alloc(struct btree_trans *trans, bool pcpu_read_locks)
666 {
667         struct bch_fs *c = trans->c;
668         struct btree_cache *bc = &c->btree_cache;
669         struct list_head *freed = pcpu_read_locks
670                 ? &bc->freed_pcpu
671                 : &bc->freed_nonpcpu;
672         struct btree *b, *b2;
673         u64 start_time = local_clock();
674         unsigned flags;
675 
676         flags = memalloc_nofs_save();
677         mutex_lock(&bc->lock);
678 
679         /*
680          * We never free struct btree itself, just the memory that holds the on
681          * disk node. Check the freed list before allocating a new one:
682          */
683         list_for_each_entry(b, freed, list)
684                 if (!btree_node_reclaim(c, b, false)) {
685                         list_del_init(&b->list);
686                         goto got_node;
687                 }
688 
689         b = __btree_node_mem_alloc(c, GFP_NOWAIT|__GFP_NOWARN);
690         if (!b) {
691                 mutex_unlock(&bc->lock);
692                 bch2_trans_unlock(trans);
693                 b = __btree_node_mem_alloc(c, GFP_KERNEL);
694                 if (!b)
695                         goto err;
696                 mutex_lock(&bc->lock);
697         }
698 
699         bch2_btree_lock_init(&b->c, pcpu_read_locks ? SIX_LOCK_INIT_PCPU : 0);
700 
701         BUG_ON(!six_trylock_intent(&b->c.lock));
702         BUG_ON(!six_trylock_write(&b->c.lock));
703 got_node:
704 
705         /*
706          * btree_free() doesn't free memory; it sticks the node on the end of
707          * the list. Check if there's any freed nodes there:
708          */
709         list_for_each_entry(b2, &bc->freeable, list)
710                 if (!btree_node_reclaim(c, b2, false)) {
711                         swap(b->data, b2->data);
712                         swap(b->aux_data, b2->aux_data);
713                         btree_node_to_freedlist(bc, b2);
714                         six_unlock_write(&b2->c.lock);
715                         six_unlock_intent(&b2->c.lock);
716                         goto got_mem;
717                 }
718 
719         mutex_unlock(&bc->lock);
720 
721         if (btree_node_data_alloc(c, b, GFP_NOWAIT|__GFP_NOWARN)) {
722                 bch2_trans_unlock(trans);
723                 if (btree_node_data_alloc(c, b, GFP_KERNEL|__GFP_NOWARN))
724                         goto err;
725         }
726 
727         mutex_lock(&bc->lock);
728         bc->used++;
729 got_mem:
730         mutex_unlock(&bc->lock);
731 
732         BUG_ON(btree_node_hashed(b));
733         BUG_ON(btree_node_dirty(b));
734         BUG_ON(btree_node_write_in_flight(b));
735 out:
736         b->flags                = 0;
737         b->written              = 0;
738         b->nsets                = 0;
739         b->sib_u64s[0]          = 0;
740         b->sib_u64s[1]          = 0;
741         b->whiteout_u64s        = 0;
742         bch2_btree_keys_init(b);
743         set_btree_node_accessed(b);
744 
745         bch2_time_stats_update(&c->times[BCH_TIME_btree_node_mem_alloc],
746                                start_time);
747 
748         memalloc_nofs_restore(flags);
749 
750         int ret = bch2_trans_relock(trans);
751         if (unlikely(ret)) {
752                 bch2_btree_node_to_freelist(c, b);
753                 return ERR_PTR(ret);
754         }
755 
756         return b;
757 err:
758         mutex_lock(&bc->lock);
759 
760         /* Try to cannibalize another cached btree node: */
761         if (bc->alloc_lock == current) {
762                 b2 = btree_node_cannibalize(c);
763                 clear_btree_node_just_written(b2);
764                 bch2_btree_node_hash_remove(bc, b2);
765 
766                 if (b) {
767                         swap(b->data, b2->data);
768                         swap(b->aux_data, b2->aux_data);
769                         btree_node_to_freedlist(bc, b2);
770                         six_unlock_write(&b2->c.lock);
771                         six_unlock_intent(&b2->c.lock);
772                 } else {
773                         b = b2;
774                         list_del_init(&b->list);
775                 }
776 
777                 mutex_unlock(&bc->lock);
778 
779                 trace_and_count(c, btree_cache_cannibalize, trans);
780                 goto out;
781         }
782 
783         mutex_unlock(&bc->lock);
784         memalloc_nofs_restore(flags);
785         return ERR_PTR(-BCH_ERR_ENOMEM_btree_node_mem_alloc);
786 }
787 
788 /* Slowpath, don't want it inlined into btree_iter_traverse() */
789 static noinline struct btree *bch2_btree_node_fill(struct btree_trans *trans,
790                                 struct btree_path *path,
791                                 const struct bkey_i *k,
792                                 enum btree_id btree_id,
793                                 unsigned level,
794                                 enum six_lock_type lock_type,
795                                 bool sync)
796 {
797         struct bch_fs *c = trans->c;
798         struct btree_cache *bc = &c->btree_cache;
799         struct btree *b;
800 
801         if (unlikely(level >= BTREE_MAX_DEPTH)) {
802                 int ret = bch2_fs_topology_error(c, "attempting to get btree node at level %u, >= max depth %u",
803                                                  level, BTREE_MAX_DEPTH);
804                 return ERR_PTR(ret);
805         }
806 
807         if (unlikely(!bkey_is_btree_ptr(&k->k))) {
808                 struct printbuf buf = PRINTBUF;
809                 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k));
810 
811                 int ret = bch2_fs_topology_error(c, "attempting to get btree node with non-btree key %s", buf.buf);
812                 printbuf_exit(&buf);
813                 return ERR_PTR(ret);
814         }
815 
816         if (unlikely(k->k.u64s > BKEY_BTREE_PTR_U64s_MAX)) {
817                 struct printbuf buf = PRINTBUF;
818                 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k));
819 
820                 int ret = bch2_fs_topology_error(c, "attempting to get btree node with too big key %s", buf.buf);
821                 printbuf_exit(&buf);
822                 return ERR_PTR(ret);
823         }
824 
825         /*
826          * Parent node must be locked, else we could read in a btree node that's
827          * been freed:
828          */
829         if (path && !bch2_btree_node_relock(trans, path, level + 1)) {
830                 trace_and_count(c, trans_restart_relock_parent_for_fill, trans, _THIS_IP_, path);
831                 return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_fill_relock));
832         }
833 
834         b = bch2_btree_node_mem_alloc(trans, level != 0);
835 
836         if (bch2_err_matches(PTR_ERR_OR_ZERO(b), ENOMEM)) {
837                 if (!path)
838                         return b;
839 
840                 trans->memory_allocation_failure = true;
841                 trace_and_count(c, trans_restart_memory_allocation_failure, trans, _THIS_IP_, path);
842                 return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_fill_mem_alloc_fail));
843         }
844 
845         if (IS_ERR(b))
846                 return b;
847 
848         bkey_copy(&b->key, k);
849         if (bch2_btree_node_hash_insert(bc, b, level, btree_id)) {
850                 /* raced with another fill: */
851 
852                 /* mark as unhashed... */
853                 b->hash_val = 0;
854 
855                 mutex_lock(&bc->lock);
856                 list_add(&b->list, &bc->freeable);
857                 mutex_unlock(&bc->lock);
858 
859                 six_unlock_write(&b->c.lock);
860                 six_unlock_intent(&b->c.lock);
861                 return NULL;
862         }
863 
864         set_btree_node_read_in_flight(b);
865         six_unlock_write(&b->c.lock);
866 
867         if (path) {
868                 u32 seq = six_lock_seq(&b->c.lock);
869 
870                 /* Unlock before doing IO: */
871                 six_unlock_intent(&b->c.lock);
872                 bch2_trans_unlock_noassert(trans);
873 
874                 bch2_btree_node_read(trans, b, sync);
875 
876                 int ret = bch2_trans_relock(trans);
877                 if (ret)
878                         return ERR_PTR(ret);
879 
880                 if (!sync)
881                         return NULL;
882 
883                 if (!six_relock_type(&b->c.lock, lock_type, seq))
884                         b = NULL;
885         } else {
886                 bch2_btree_node_read(trans, b, sync);
887                 if (lock_type == SIX_LOCK_read)
888                         six_lock_downgrade(&b->c.lock);
889         }
890 
891         return b;
892 }
893 
894 static noinline void btree_bad_header(struct bch_fs *c, struct btree *b)
895 {
896         struct printbuf buf = PRINTBUF;
897 
898         if (c->curr_recovery_pass <= BCH_RECOVERY_PASS_check_allocations)
899                 return;
900 
901         prt_printf(&buf,
902                "btree node header doesn't match ptr\n"
903                "btree %s level %u\n"
904                "ptr: ",
905                bch2_btree_id_str(b->c.btree_id), b->c.level);
906         bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
907 
908         prt_printf(&buf, "\nheader: btree %s level %llu\n"
909                "min ",
910                bch2_btree_id_str(BTREE_NODE_ID(b->data)),
911                BTREE_NODE_LEVEL(b->data));
912         bch2_bpos_to_text(&buf, b->data->min_key);
913 
914         prt_printf(&buf, "\nmax ");
915         bch2_bpos_to_text(&buf, b->data->max_key);
916 
917         bch2_fs_topology_error(c, "%s", buf.buf);
918 
919         printbuf_exit(&buf);
920 }
921 
922 static inline void btree_check_header(struct bch_fs *c, struct btree *b)
923 {
924         if (b->c.btree_id != BTREE_NODE_ID(b->data) ||
925             b->c.level != BTREE_NODE_LEVEL(b->data) ||
926             !bpos_eq(b->data->max_key, b->key.k.p) ||
927             (b->key.k.type == KEY_TYPE_btree_ptr_v2 &&
928              !bpos_eq(b->data->min_key,
929                       bkey_i_to_btree_ptr_v2(&b->key)->v.min_key)))
930                 btree_bad_header(c, b);
931 }
932 
933 static struct btree *__bch2_btree_node_get(struct btree_trans *trans, struct btree_path *path,
934                                            const struct bkey_i *k, unsigned level,
935                                            enum six_lock_type lock_type,
936                                            unsigned long trace_ip)
937 {
938         struct bch_fs *c = trans->c;
939         struct btree_cache *bc = &c->btree_cache;
940         struct btree *b;
941         bool need_relock = false;
942         int ret;
943 
944         EBUG_ON(level >= BTREE_MAX_DEPTH);
945 retry:
946         b = btree_cache_find(bc, k);
947         if (unlikely(!b)) {
948                 /*
949                  * We must have the parent locked to call bch2_btree_node_fill(),
950                  * else we could read in a btree node from disk that's been
951                  * freed:
952                  */
953                 b = bch2_btree_node_fill(trans, path, k, path->btree_id,
954                                          level, lock_type, true);
955                 need_relock = true;
956 
957                 /* We raced and found the btree node in the cache */
958                 if (!b)
959                         goto retry;
960 
961                 if (IS_ERR(b))
962                         return b;
963         } else {
964                 if (btree_node_read_locked(path, level + 1))
965                         btree_node_unlock(trans, path, level + 1);
966 
967                 ret = btree_node_lock(trans, path, &b->c, level, lock_type, trace_ip);
968                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
969                         return ERR_PTR(ret);
970 
971                 BUG_ON(ret);
972 
973                 if (unlikely(b->hash_val != btree_ptr_hash_val(k) ||
974                              b->c.level != level ||
975                              race_fault())) {
976                         six_unlock_type(&b->c.lock, lock_type);
977                         if (bch2_btree_node_relock(trans, path, level + 1))
978                                 goto retry;
979 
980                         trace_and_count(c, trans_restart_btree_node_reused, trans, trace_ip, path);
981                         return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_lock_node_reused));
982                 }
983 
984                 /* avoid atomic set bit if it's not needed: */
985                 if (!btree_node_accessed(b))
986                         set_btree_node_accessed(b);
987         }
988 
989         if (unlikely(btree_node_read_in_flight(b))) {
990                 u32 seq = six_lock_seq(&b->c.lock);
991 
992                 six_unlock_type(&b->c.lock, lock_type);
993                 bch2_trans_unlock(trans);
994                 need_relock = true;
995 
996                 bch2_btree_node_wait_on_read(b);
997 
998                 ret = bch2_trans_relock(trans);
999                 if (ret)
1000                         return ERR_PTR(ret);
1001 
1002                 /*
1003                  * should_be_locked is not set on this path yet, so we need to
1004                  * relock it specifically:
1005                  */
1006                 if (!six_relock_type(&b->c.lock, lock_type, seq))
1007                         goto retry;
1008         }
1009 
1010         if (unlikely(need_relock)) {
1011                 ret = bch2_trans_relock(trans) ?:
1012                         bch2_btree_path_relock_intent(trans, path);
1013                 if (ret) {
1014                         six_unlock_type(&b->c.lock, lock_type);
1015                         return ERR_PTR(ret);
1016                 }
1017         }
1018 
1019         prefetch(b->aux_data);
1020 
1021         for_each_bset(b, t) {
1022                 void *p = (u64 *) b->aux_data + t->aux_data_offset;
1023 
1024                 prefetch(p + L1_CACHE_BYTES * 0);
1025                 prefetch(p + L1_CACHE_BYTES * 1);
1026                 prefetch(p + L1_CACHE_BYTES * 2);
1027         }
1028 
1029         if (unlikely(btree_node_read_error(b))) {
1030                 six_unlock_type(&b->c.lock, lock_type);
1031                 return ERR_PTR(-BCH_ERR_btree_node_read_error);
1032         }
1033 
1034         EBUG_ON(b->c.btree_id != path->btree_id);
1035         EBUG_ON(BTREE_NODE_LEVEL(b->data) != level);
1036         btree_check_header(c, b);
1037 
1038         return b;
1039 }
1040 
1041 /**
1042  * bch2_btree_node_get - find a btree node in the cache and lock it, reading it
1043  * in from disk if necessary.
1044  *
1045  * @trans:      btree transaction object
1046  * @path:       btree_path being traversed
1047  * @k:          pointer to btree node (generally KEY_TYPE_btree_ptr_v2)
1048  * @level:      level of btree node being looked up (0 == leaf node)
1049  * @lock_type:  SIX_LOCK_read or SIX_LOCK_intent
1050  * @trace_ip:   ip of caller of btree iterator code (i.e. caller of bch2_btree_iter_peek())
1051  *
1052  * The btree node will have either a read or a write lock held, depending on
1053  * the @write parameter.
1054  *
1055  * Returns: btree node or ERR_PTR()
1056  */
1057 struct btree *bch2_btree_node_get(struct btree_trans *trans, struct btree_path *path,
1058                                   const struct bkey_i *k, unsigned level,
1059                                   enum six_lock_type lock_type,
1060                                   unsigned long trace_ip)
1061 {
1062         struct bch_fs *c = trans->c;
1063         struct btree *b;
1064         int ret;
1065 
1066         EBUG_ON(level >= BTREE_MAX_DEPTH);
1067 
1068         b = btree_node_mem_ptr(k);
1069 
1070         /*
1071          * Check b->hash_val _before_ calling btree_node_lock() - this might not
1072          * be the node we want anymore, and trying to lock the wrong node could
1073          * cause an unneccessary transaction restart:
1074          */
1075         if (unlikely(!c->opts.btree_node_mem_ptr_optimization ||
1076                      !b ||
1077                      b->hash_val != btree_ptr_hash_val(k)))
1078                 return __bch2_btree_node_get(trans, path, k, level, lock_type, trace_ip);
1079 
1080         if (btree_node_read_locked(path, level + 1))
1081                 btree_node_unlock(trans, path, level + 1);
1082 
1083         ret = btree_node_lock(trans, path, &b->c, level, lock_type, trace_ip);
1084         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1085                 return ERR_PTR(ret);
1086 
1087         BUG_ON(ret);
1088 
1089         if (unlikely(b->hash_val != btree_ptr_hash_val(k) ||
1090                      b->c.level != level ||
1091                      race_fault())) {
1092                 six_unlock_type(&b->c.lock, lock_type);
1093                 if (bch2_btree_node_relock(trans, path, level + 1))
1094                         return __bch2_btree_node_get(trans, path, k, level, lock_type, trace_ip);
1095 
1096                 trace_and_count(c, trans_restart_btree_node_reused, trans, trace_ip, path);
1097                 return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_lock_node_reused));
1098         }
1099 
1100         if (unlikely(btree_node_read_in_flight(b))) {
1101                 six_unlock_type(&b->c.lock, lock_type);
1102                 return __bch2_btree_node_get(trans, path, k, level, lock_type, trace_ip);
1103         }
1104 
1105         prefetch(b->aux_data);
1106 
1107         for_each_bset(b, t) {
1108                 void *p = (u64 *) b->aux_data + t->aux_data_offset;
1109 
1110                 prefetch(p + L1_CACHE_BYTES * 0);
1111                 prefetch(p + L1_CACHE_BYTES * 1);
1112                 prefetch(p + L1_CACHE_BYTES * 2);
1113         }
1114 
1115         /* avoid atomic set bit if it's not needed: */
1116         if (!btree_node_accessed(b))
1117                 set_btree_node_accessed(b);
1118 
1119         if (unlikely(btree_node_read_error(b))) {
1120                 six_unlock_type(&b->c.lock, lock_type);
1121                 return ERR_PTR(-BCH_ERR_btree_node_read_error);
1122         }
1123 
1124         EBUG_ON(b->c.btree_id != path->btree_id);
1125         EBUG_ON(BTREE_NODE_LEVEL(b->data) != level);
1126         btree_check_header(c, b);
1127 
1128         return b;
1129 }
1130 
1131 struct btree *bch2_btree_node_get_noiter(struct btree_trans *trans,
1132                                          const struct bkey_i *k,
1133                                          enum btree_id btree_id,
1134                                          unsigned level,
1135                                          bool nofill)
1136 {
1137         struct bch_fs *c = trans->c;
1138         struct btree_cache *bc = &c->btree_cache;
1139         struct btree *b;
1140         int ret;
1141 
1142         EBUG_ON(level >= BTREE_MAX_DEPTH);
1143 
1144         if (c->opts.btree_node_mem_ptr_optimization) {
1145                 b = btree_node_mem_ptr(k);
1146                 if (b)
1147                         goto lock_node;
1148         }
1149 retry:
1150         b = btree_cache_find(bc, k);
1151         if (unlikely(!b)) {
1152                 if (nofill)
1153                         goto out;
1154 
1155                 b = bch2_btree_node_fill(trans, NULL, k, btree_id,
1156                                          level, SIX_LOCK_read, true);
1157 
1158                 /* We raced and found the btree node in the cache */
1159                 if (!b)
1160                         goto retry;
1161 
1162                 if (IS_ERR(b) &&
1163                     !bch2_btree_cache_cannibalize_lock(trans, NULL))
1164                         goto retry;
1165 
1166                 if (IS_ERR(b))
1167                         goto out;
1168         } else {
1169 lock_node:
1170                 ret = btree_node_lock_nopath(trans, &b->c, SIX_LOCK_read, _THIS_IP_);
1171                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1172                         return ERR_PTR(ret);
1173 
1174                 BUG_ON(ret);
1175 
1176                 if (unlikely(b->hash_val != btree_ptr_hash_val(k) ||
1177                              b->c.btree_id != btree_id ||
1178                              b->c.level != level)) {
1179                         six_unlock_read(&b->c.lock);
1180                         goto retry;
1181                 }
1182         }
1183 
1184         /* XXX: waiting on IO with btree locks held: */
1185         __bch2_btree_node_wait_on_read(b);
1186 
1187         prefetch(b->aux_data);
1188 
1189         for_each_bset(b, t) {
1190                 void *p = (u64 *) b->aux_data + t->aux_data_offset;
1191 
1192                 prefetch(p + L1_CACHE_BYTES * 0);
1193                 prefetch(p + L1_CACHE_BYTES * 1);
1194                 prefetch(p + L1_CACHE_BYTES * 2);
1195         }
1196 
1197         /* avoid atomic set bit if it's not needed: */
1198         if (!btree_node_accessed(b))
1199                 set_btree_node_accessed(b);
1200 
1201         if (unlikely(btree_node_read_error(b))) {
1202                 six_unlock_read(&b->c.lock);
1203                 b = ERR_PTR(-BCH_ERR_btree_node_read_error);
1204                 goto out;
1205         }
1206 
1207         EBUG_ON(b->c.btree_id != btree_id);
1208         EBUG_ON(BTREE_NODE_LEVEL(b->data) != level);
1209         btree_check_header(c, b);
1210 out:
1211         bch2_btree_cache_cannibalize_unlock(trans);
1212         return b;
1213 }
1214 
1215 int bch2_btree_node_prefetch(struct btree_trans *trans,
1216                              struct btree_path *path,
1217                              const struct bkey_i *k,
1218                              enum btree_id btree_id, unsigned level)
1219 {
1220         struct bch_fs *c = trans->c;
1221         struct btree_cache *bc = &c->btree_cache;
1222 
1223         BUG_ON(path && !btree_node_locked(path, level + 1));
1224         BUG_ON(level >= BTREE_MAX_DEPTH);
1225 
1226         struct btree *b = btree_cache_find(bc, k);
1227         if (b)
1228                 return 0;
1229 
1230         b = bch2_btree_node_fill(trans, path, k, btree_id,
1231                                  level, SIX_LOCK_read, false);
1232         if (!IS_ERR_OR_NULL(b))
1233                 six_unlock_read(&b->c.lock);
1234         return bch2_trans_relock(trans) ?: PTR_ERR_OR_ZERO(b);
1235 }
1236 
1237 void bch2_btree_node_evict(struct btree_trans *trans, const struct bkey_i *k)
1238 {
1239         struct bch_fs *c = trans->c;
1240         struct btree_cache *bc = &c->btree_cache;
1241         struct btree *b;
1242 
1243         b = btree_cache_find(bc, k);
1244         if (!b)
1245                 return;
1246 
1247         BUG_ON(b == btree_node_root(trans->c, b));
1248 wait_on_io:
1249         /* not allowed to wait on io with btree locks held: */
1250 
1251         /* XXX we're called from btree_gc which will be holding other btree
1252          * nodes locked
1253          */
1254         __bch2_btree_node_wait_on_read(b);
1255         __bch2_btree_node_wait_on_write(b);
1256 
1257         btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent);
1258         btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_write);
1259         if (unlikely(b->hash_val != btree_ptr_hash_val(k)))
1260                 goto out;
1261 
1262         if (btree_node_dirty(b)) {
1263                 __bch2_btree_node_write(c, b, BTREE_WRITE_cache_reclaim);
1264                 six_unlock_write(&b->c.lock);
1265                 six_unlock_intent(&b->c.lock);
1266                 goto wait_on_io;
1267         }
1268 
1269         BUG_ON(btree_node_dirty(b));
1270 
1271         mutex_lock(&bc->lock);
1272         btree_node_data_free(c, b);
1273         bch2_btree_node_hash_remove(bc, b);
1274         mutex_unlock(&bc->lock);
1275 out:
1276         six_unlock_write(&b->c.lock);
1277         six_unlock_intent(&b->c.lock);
1278 }
1279 
1280 const char *bch2_btree_id_str(enum btree_id btree)
1281 {
1282         return btree < BTREE_ID_NR ? __bch2_btree_ids[btree] : "(unknown)";
1283 }
1284 
1285 void bch2_btree_id_to_text(struct printbuf *out, enum btree_id btree)
1286 {
1287         if (btree < BTREE_ID_NR)
1288                 prt_str(out, __bch2_btree_ids[btree]);
1289         else
1290                 prt_printf(out, "(unknown btree %u)", btree);
1291 }
1292 
1293 void bch2_btree_pos_to_text(struct printbuf *out, struct bch_fs *c, const struct btree *b)
1294 {
1295         prt_printf(out, "%s level %u/%u\n  ",
1296                bch2_btree_id_str(b->c.btree_id),
1297                b->c.level,
1298                bch2_btree_id_root(c, b->c.btree_id)->level);
1299         bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
1300 }
1301 
1302 void bch2_btree_node_to_text(struct printbuf *out, struct bch_fs *c, const struct btree *b)
1303 {
1304         struct bset_stats stats;
1305 
1306         memset(&stats, 0, sizeof(stats));
1307 
1308         bch2_btree_keys_stats(b, &stats);
1309 
1310         prt_printf(out, "l %u ", b->c.level);
1311         bch2_bpos_to_text(out, b->data->min_key);
1312         prt_printf(out, " - ");
1313         bch2_bpos_to_text(out, b->data->max_key);
1314         prt_printf(out, ":\n"
1315                "    ptrs: ");
1316         bch2_val_to_text(out, c, bkey_i_to_s_c(&b->key));
1317         prt_newline(out);
1318 
1319         prt_printf(out,
1320                "    format: ");
1321         bch2_bkey_format_to_text(out, &b->format);
1322 
1323         prt_printf(out,
1324                "    unpack fn len: %u\n"
1325                "    bytes used %zu/%zu (%zu%% full)\n"
1326                "    sib u64s: %u, %u (merge threshold %u)\n"
1327                "    nr packed keys %u\n"
1328                "    nr unpacked keys %u\n"
1329                "    floats %zu\n"
1330                "    failed unpacked %zu\n",
1331                b->unpack_fn_len,
1332                b->nr.live_u64s * sizeof(u64),
1333                btree_buf_bytes(b) - sizeof(struct btree_node),
1334                b->nr.live_u64s * 100 / btree_max_u64s(c),
1335                b->sib_u64s[0],
1336                b->sib_u64s[1],
1337                c->btree_foreground_merge_threshold,
1338                b->nr.packed_keys,
1339                b->nr.unpacked_keys,
1340                stats.floats,
1341                stats.failed);
1342 }
1343 
1344 static void prt_btree_cache_line(struct printbuf *out, const struct bch_fs *c,
1345                                  const char *label, unsigned nr)
1346 {
1347         prt_printf(out, "%s\t", label);
1348         prt_human_readable_u64(out, nr * c->opts.btree_node_size);
1349         prt_printf(out, " (%u)\n", nr);
1350 }
1351 
1352 void bch2_btree_cache_to_text(struct printbuf *out, const struct btree_cache *bc)
1353 {
1354         struct bch_fs *c = container_of(bc, struct bch_fs, btree_cache);
1355 
1356         if (!out->nr_tabstops)
1357                 printbuf_tabstop_push(out, 32);
1358 
1359         prt_btree_cache_line(out, c, "total:",          bc->used);
1360         prt_btree_cache_line(out, c, "nr dirty:",       atomic_read(&bc->dirty));
1361         prt_printf(out, "cannibalize lock:\t%p\n",      bc->alloc_lock);
1362         prt_newline(out);
1363 
1364         for (unsigned i = 0; i < ARRAY_SIZE(bc->used_by_btree); i++)
1365                 prt_btree_cache_line(out, c, bch2_btree_id_str(i), bc->used_by_btree[i]);
1366 
1367         prt_newline(out);
1368         prt_printf(out, "freed:\t%u\n", bc->freed);
1369         prt_printf(out, "not freed:\n");
1370         prt_printf(out, "  dirty\t%u\n", bc->not_freed_dirty);
1371         prt_printf(out, "  write in flight\t%u\n", bc->not_freed_write_in_flight);
1372         prt_printf(out, "  read in flight\t%u\n", bc->not_freed_read_in_flight);
1373         prt_printf(out, "  lock intent failed\t%u\n", bc->not_freed_lock_intent);
1374         prt_printf(out, "  lock write failed\t%u\n", bc->not_freed_lock_write);
1375         prt_printf(out, "  access bit\t%u\n", bc->not_freed_access_bit);
1376         prt_printf(out, "  no evict failed\t%u\n", bc->not_freed_noevict);
1377         prt_printf(out, "  write blocked\t%u\n", bc->not_freed_write_blocked);
1378         prt_printf(out, "  will make reachable\t%u\n", bc->not_freed_will_make_reachable);
1379 }
1380 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

sflogo.php