?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 6.13-rc7
[linux.git] / fs / bcachefs / btree_node_scan.c
blob30131c3bdd970f5a20ba76039b0f8832dc8f0d11
1 // SPDX-License-Identifier: GPL-2.0
2
3 #include "bcachefs.h"
4 #include "btree_cache.h"
5 #include "btree_io.h"
6 #include "btree_journal_iter.h"
7 #include "btree_node_scan.h"
8 #include "btree_update_interior.h"
9 #include "buckets.h"
10 #include "error.h"
11 #include "journal_io.h"
12 #include "recovery_passes.h"
13
14 #include <linux/kthread.h>
15 #include <linux/sort.h>
16
17 struct find_btree_nodes_worker {
18 struct closure *cl;
19 struct find_btree_nodes *f;
20 struct bch_dev *ca;
21 };
22
23 static void found_btree_node_to_text(struct printbuf *out, struct bch_fs *c, const struct found_btree_node *n)
24 {
25 prt_printf(out, "%s l=%u seq=%u journal_seq=%llu cookie=%llx ",
26 bch2_btree_id_str(n->btree_id), n->level, n->seq,
27 n->journal_seq, n->cookie);
28 bch2_bpos_to_text(out, n->min_key);
29 prt_str(out, "-");
30 bch2_bpos_to_text(out, n->max_key);
31
32 if (n->range_updated)
33 prt_str(out, " range updated");
34 if (n->overwritten)
35 prt_str(out, " overwritten");
36
37 for (unsigned i = 0; i < n->nr_ptrs; i++) {
38 prt_char(out, ' ');
39 bch2_extent_ptr_to_text(out, c, n->ptrs + i);
40 }
41 }
42
43 static void found_btree_nodes_to_text(struct printbuf *out, struct bch_fs *c, found_btree_nodes nodes)
44 {
45 printbuf_indent_add(out, 2);
46 darray_for_each(nodes, i) {
47 found_btree_node_to_text(out, c, i);
48 prt_newline(out);
49 }
50 printbuf_indent_sub(out, 2);
51 }
52
53 static void found_btree_node_to_key(struct bkey_i *k, const struct found_btree_node *f)
54 {
55 struct bkey_i_btree_ptr_v2 *bp = bkey_btree_ptr_v2_init(k);
56
57 set_bkey_val_u64s(&bp->k, sizeof(struct bch_btree_ptr_v2) / sizeof(u64) + f->nr_ptrs);
58 bp->k.p = f->max_key;
59 bp->v.seq = cpu_to_le64(f->cookie);
60 bp->v.sectors_written = 0;
61 bp->v.flags = 0;
62 bp->v.sectors_written = cpu_to_le16(f->sectors_written);
63 bp->v.min_key = f->min_key;
64 SET_BTREE_PTR_RANGE_UPDATED(&bp->v, f->range_updated);
65 memcpy(bp->v.start, f->ptrs, sizeof(struct bch_extent_ptr) * f->nr_ptrs);
66 }
67
68 static inline u64 bkey_journal_seq(struct bkey_s_c k)
69 {
70 switch (k.k->type) {
71 case KEY_TYPE_inode_v3:
72 return le64_to_cpu(bkey_s_c_to_inode_v3(k).v->bi_journal_seq);
73 default:
74 return 0;
75 }
76 }
77
78 static bool found_btree_node_is_readable(struct btree_trans *trans,
79 struct found_btree_node *f)
80 {
81 struct { __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX); } tmp;
82
83 found_btree_node_to_key(&tmp.k, f);
84
85 struct btree *b = bch2_btree_node_get_noiter(trans, &tmp.k, f->btree_id, f->level, false);
86 bool ret = !IS_ERR_OR_NULL(b);
87 if (!ret)
88 return ret;
89
90 f->sectors_written = b->written;
91 f->journal_seq = le64_to_cpu(b->data->keys.journal_seq);
92
93 struct bkey_s_c k;
94 struct bkey unpacked;
95 struct btree_node_iter iter;
96 for_each_btree_node_key_unpack(b, k, &iter, &unpacked)
97 f->journal_seq = max(f->journal_seq, bkey_journal_seq(k));
98
99 six_unlock_read(&b->c.lock);
100
101 /*
102 * We might update this node's range; if that happens, we need the node
103 * to be re-read so the read path can trim keys that are no longer in
104 * this node
105 */
106 if (b != btree_node_root(trans->c, b))
107 bch2_btree_node_evict(trans, &tmp.k);
108 return ret;
109 }
110
111 static int found_btree_node_cmp_cookie(const void *_l, const void *_r)
112 {
113 const struct found_btree_node *l = _l;
114 const struct found_btree_node *r = _r;
115
116 return cmp_int(l->btree_id, r->btree_id) ?:
117 cmp_int(l->level, r->level) ?:
118 cmp_int(l->cookie, r->cookie);
119 }
120
121 /*
122 * Given two found btree nodes, if their sequence numbers are equal, take the
123 * one that's readable:
124 */
125 static int found_btree_node_cmp_time(const struct found_btree_node *l,
126 const struct found_btree_node *r)
127 {
128 return cmp_int(l->seq, r->seq) ?:
129 cmp_int(l->journal_seq, r->journal_seq);
130 }
131
132 static int found_btree_node_cmp_pos(const void *_l, const void *_r)
133 {
134 const struct found_btree_node *l = _l;
135 const struct found_btree_node *r = _r;
136
137 return cmp_int(l->btree_id, r->btree_id) ?:
138 -cmp_int(l->level, r->level) ?:
139 bpos_cmp(l->min_key, r->min_key) ?:
140 -found_btree_node_cmp_time(l, r);
141 }
142
143 static void try_read_btree_node(struct find_btree_nodes *f, struct bch_dev *ca,
144 struct bio *bio, struct btree_node *bn, u64 offset)
145 {
146 struct bch_fs *c = container_of(f, struct bch_fs, found_btree_nodes);
147
148 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ);
149 bio->bi_iter.bi_sector = offset;
150 bch2_bio_map(bio, bn, PAGE_SIZE);
151
152 submit_bio_wait(bio);
153 if (bch2_dev_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_read,
154 "IO error in try_read_btree_node() at %llu: %s",
155 offset, bch2_blk_status_to_str(bio->bi_status)))
156 return;
157
158 if (le64_to_cpu(bn->magic) != bset_magic(c))
159 return;
160
161 if (bch2_csum_type_is_encryption(BSET_CSUM_TYPE(&bn->keys))) {
162 struct nonce nonce = btree_nonce(&bn->keys, 0);
163 unsigned bytes = (void *) &bn->keys - (void *) &bn->flags;
164
165 bch2_encrypt(c, BSET_CSUM_TYPE(&bn->keys), nonce, &bn->flags, bytes);
166 }
167
168 if (btree_id_is_alloc(BTREE_NODE_ID(bn)))
169 return;
170
171 if (BTREE_NODE_LEVEL(bn) >= BTREE_MAX_DEPTH)
172 return;
173
174 if (BTREE_NODE_ID(bn) >= BTREE_ID_NR_MAX)
175 return;
176
177 rcu_read_lock();
178 struct found_btree_node n = {
179 .btree_id = BTREE_NODE_ID(bn),
180 .level = BTREE_NODE_LEVEL(bn),
181 .seq = BTREE_NODE_SEQ(bn),
182 .cookie = le64_to_cpu(bn->keys.seq),
183 .min_key = bn->min_key,
184 .max_key = bn->max_key,
185 .nr_ptrs = 1,
186 .ptrs[0].type = 1 << BCH_EXTENT_ENTRY_ptr,
187 .ptrs[0].offset = offset,
188 .ptrs[0].dev = ca->dev_idx,
189 .ptrs[0].gen = bucket_gen_get(ca, sector_to_bucket(ca, offset)),
190 };
191 rcu_read_unlock();
192
193 if (bch2_trans_run(c, found_btree_node_is_readable(trans, &n))) {
194 mutex_lock(&f->lock);
195 if (BSET_BIG_ENDIAN(&bn->keys) != CPU_BIG_ENDIAN) {
196 bch_err(c, "try_read_btree_node() can't handle endian conversion");
197 f->ret = -EINVAL;
198 goto unlock;
199 }
200
201 if (darray_push(&f->nodes, n))
202 f->ret = -ENOMEM;
203 unlock:
204 mutex_unlock(&f->lock);
205 }
206 }
207
208 static int read_btree_nodes_worker(void *p)
209 {
210 struct find_btree_nodes_worker *w = p;
211 struct bch_fs *c = container_of(w->f, struct bch_fs, found_btree_nodes);
212 struct bch_dev *ca = w->ca;
213 void *buf = (void *) __get_free_page(GFP_KERNEL);
214 struct bio *bio = bio_alloc(NULL, 1, 0, GFP_KERNEL);
215 unsigned long last_print = jiffies;
216
217 if (!buf || !bio) {
218 bch_err(c, "read_btree_nodes_worker: error allocating bio/buf");
219 w->f->ret = -ENOMEM;
220 goto err;
221 }
222
223 for (u64 bucket = ca->mi.first_bucket; bucket < ca->mi.nbuckets; bucket++)
224 for (unsigned bucket_offset = 0;
225 bucket_offset + btree_sectors(c) <= ca->mi.bucket_size;
226 bucket_offset += btree_sectors(c)) {
227 if (time_after(jiffies, last_print + HZ * 30)) {
228 u64 cur_sector = bucket * ca->mi.bucket_size + bucket_offset;
229 u64 end_sector = ca->mi.nbuckets * ca->mi.bucket_size;
230
231 bch_info(ca, "%s: %2u%% done", __func__,
232 (unsigned) div64_u64(cur_sector * 100, end_sector));
233 last_print = jiffies;
234 }
235
236 u64 sector = bucket * ca->mi.bucket_size + bucket_offset;
237
238 if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_mi_btree_bitmap &&
239 !bch2_dev_btree_bitmap_marked_sectors(ca, sector, btree_sectors(c)))
240 continue;
241
242 try_read_btree_node(w->f, ca, bio, buf, sector);
243 }
244 err:
245 bio_put(bio);
246 free_page((unsigned long) buf);
247 percpu_ref_get(&ca->io_ref);
248 closure_put(w->cl);
249 kfree(w);
250 return 0;
251 }
252
253 static int read_btree_nodes(struct find_btree_nodes *f)
254 {
255 struct bch_fs *c = container_of(f, struct bch_fs, found_btree_nodes);
256 struct closure cl;
257 int ret = 0;
258
259 closure_init_stack(&cl);
260
261 for_each_online_member(c, ca) {
262 if (!(ca->mi.data_allowed & BIT(BCH_DATA_btree)))
263 continue;
264
265 struct find_btree_nodes_worker *w = kmalloc(sizeof(*w), GFP_KERNEL);
266 struct task_struct *t;
267
268 if (!w) {
269 percpu_ref_put(&ca->io_ref);
270 ret = -ENOMEM;
271 goto err;
272 }
273
274 percpu_ref_get(&ca->io_ref);
275 closure_get(&cl);
276 w->cl = &cl;
277 w->f = f;
278 w->ca = ca;
279
280 t = kthread_run(read_btree_nodes_worker, w, "read_btree_nodes/%s", ca->name);
281 ret = PTR_ERR_OR_ZERO(t);
282 if (ret) {
283 percpu_ref_put(&ca->io_ref);
284 closure_put(&cl);
285 f->ret = ret;
286 bch_err(c, "error starting kthread: %i", ret);
287 break;
288 }
289 }
290 err:
291 closure_sync(&cl);
292 return f->ret ?: ret;
293 }
294
295 static void bubble_up(struct found_btree_node *n, struct found_btree_node *end)
296 {
297 while (n + 1 < end &&
298 found_btree_node_cmp_pos(n, n + 1) > 0) {
299 swap(n[0], n[1]);
300 n++;
301 }
302 }
303
304 static int handle_overwrites(struct bch_fs *c,
305 struct found_btree_node *start,
306 struct found_btree_node *end)
307 {
308 struct found_btree_node *n;
309 again:
310 for (n = start + 1;
311 n < end &&
312 n->btree_id == start->btree_id &&
313 n->level == start->level &&
314 bpos_lt(n->min_key, start->max_key);
315 n++) {
316 int cmp = found_btree_node_cmp_time(start, n);
317
318 if (cmp > 0) {
319 if (bpos_cmp(start->max_key, n->max_key) >= 0)
320 n->overwritten = true;
321 else {
322 n->range_updated = true;
323 n->min_key = bpos_successor(start->max_key);
324 n->range_updated = true;
325 bubble_up(n, end);
326 goto again;
327 }
328 } else if (cmp < 0) {
329 BUG_ON(bpos_cmp(n->min_key, start->min_key) <= 0);
330
331 start->max_key = bpos_predecessor(n->min_key);
332 start->range_updated = true;
333 } else if (n->level) {
334 n->overwritten = true;
335 } else {
336 if (bpos_cmp(start->max_key, n->max_key) >= 0)
337 n->overwritten = true;
338 else {
339 n->range_updated = true;
340 n->min_key = bpos_successor(start->max_key);
341 n->range_updated = true;
342 bubble_up(n, end);
343 goto again;
344 }
345 }
346 }
347
348 return 0;
349 }
350
351 int bch2_scan_for_btree_nodes(struct bch_fs *c)
352 {
353 struct find_btree_nodes *f = &c->found_btree_nodes;
354 struct printbuf buf = PRINTBUF;
355 size_t dst;
356 int ret = 0;
357
358 if (f->nodes.nr)
359 return 0;
360
361 mutex_init(&f->lock);
362
363 ret = read_btree_nodes(f);
364 if (ret)
365 return ret;
366
367 if (!f->nodes.nr) {
368 bch_err(c, "%s: no btree nodes found", __func__);
369 ret = -EINVAL;
370 goto err;
371 }
372
373 if (0 && c->opts.verbose) {
374 printbuf_reset(&buf);
375 prt_printf(&buf, "%s: nodes found:\n", __func__);
376 found_btree_nodes_to_text(&buf, c, f->nodes);
377 bch2_print_string_as_lines(KERN_INFO, buf.buf);
378 }
379
380 sort(f->nodes.data, f->nodes.nr, sizeof(f->nodes.data[0]), found_btree_node_cmp_cookie, NULL);
381
382 dst = 0;
383 darray_for_each(f->nodes, i) {
384 struct found_btree_node *prev = dst ? f->nodes.data + dst - 1 : NULL;
385
386 if (prev &&
387 prev->cookie == i->cookie) {
388 if (prev->nr_ptrs == ARRAY_SIZE(prev->ptrs)) {
389 bch_err(c, "%s: found too many replicas for btree node", __func__);
390 ret = -EINVAL;
391 goto err;
392 }
393 prev->ptrs[prev->nr_ptrs++] = i->ptrs[0];
394 } else {
395 f->nodes.data[dst++] = *i;
396 }
397 }
398 f->nodes.nr = dst;
399
400 sort(f->nodes.data, f->nodes.nr, sizeof(f->nodes.data[0]), found_btree_node_cmp_pos, NULL);
401
402 if (0 && c->opts.verbose) {
403 printbuf_reset(&buf);
404 prt_printf(&buf, "%s: nodes after merging replicas:\n", __func__);
405 found_btree_nodes_to_text(&buf, c, f->nodes);
406 bch2_print_string_as_lines(KERN_INFO, buf.buf);
407 }
408
409 dst = 0;
410 darray_for_each(f->nodes, i) {
411 if (i->overwritten)
412 continue;
413
414 ret = handle_overwrites(c, i, &darray_top(f->nodes));
415 if (ret)
416 goto err;
417
418 BUG_ON(i->overwritten);
419 f->nodes.data[dst++] = *i;
420 }
421 f->nodes.nr = dst;
422
423 if (c->opts.verbose) {
424 printbuf_reset(&buf);
425 prt_printf(&buf, "%s: nodes found after overwrites:\n", __func__);
426 found_btree_nodes_to_text(&buf, c, f->nodes);
427 bch2_print_string_as_lines(KERN_INFO, buf.buf);
428 }
429
430 eytzinger0_sort(f->nodes.data, f->nodes.nr, sizeof(f->nodes.data[0]), found_btree_node_cmp_pos, NULL);
431 err:
432 printbuf_exit(&buf);
433 return ret;
434 }
435
436 static int found_btree_node_range_start_cmp(const void *_l, const void *_r)
437 {
438 const struct found_btree_node *l = _l;
439 const struct found_btree_node *r = _r;
440
441 return cmp_int(l->btree_id, r->btree_id) ?:
442 -cmp_int(l->level, r->level) ?:
443 bpos_cmp(l->max_key, r->min_key);
444 }
445
446 #define for_each_found_btree_node_in_range(_f, _search, _idx) \
447 for (size_t _idx = eytzinger0_find_gt((_f)->nodes.data, (_f)->nodes.nr, \
448 sizeof((_f)->nodes.data[0]), \
449 found_btree_node_range_start_cmp, &search); \
450 _idx < (_f)->nodes.nr && \
451 (_f)->nodes.data[_idx].btree_id == _search.btree_id && \
452 (_f)->nodes.data[_idx].level == _search.level && \
453 bpos_lt((_f)->nodes.data[_idx].min_key, _search.max_key); \
454 _idx = eytzinger0_next(_idx, (_f)->nodes.nr))
455
456 bool bch2_btree_node_is_stale(struct bch_fs *c, struct btree *b)
457 {
458 struct find_btree_nodes *f = &c->found_btree_nodes;
459
460 struct found_btree_node search = {
461 .btree_id = b->c.btree_id,
462 .level = b->c.level,
463 .min_key = b->data->min_key,
464 .max_key = b->key.k.p,
465 };
466
467 for_each_found_btree_node_in_range(f, search, idx)
468 if (f->nodes.data[idx].seq > BTREE_NODE_SEQ(b->data))
469 return true;
470 return false;
471 }
472
473 bool bch2_btree_has_scanned_nodes(struct bch_fs *c, enum btree_id btree)
474 {
475 struct found_btree_node search = {
476 .btree_id = btree,
477 .level = 0,
478 .min_key = POS_MIN,
479 .max_key = SPOS_MAX,
480 };
481
482 for_each_found_btree_node_in_range(&c->found_btree_nodes, search, idx)
483 return true;
484 return false;
485 }
486
487 int bch2_get_scanned_nodes(struct bch_fs *c, enum btree_id btree,
488 unsigned level, struct bpos node_min, struct bpos node_max)
489 {
490 if (btree_id_is_alloc(btree))
491 return 0;
492
493 struct find_btree_nodes *f = &c->found_btree_nodes;
494
495 int ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_scan_for_btree_nodes);
496 if (ret)
497 return ret;
498
499 if (c->opts.verbose) {
500 struct printbuf buf = PRINTBUF;
501
502 prt_printf(&buf, "recovering %s l=%u ", bch2_btree_id_str(btree), level);
503 bch2_bpos_to_text(&buf, node_min);
504 prt_str(&buf, " - ");
505 bch2_bpos_to_text(&buf, node_max);
506
507 bch_info(c, "%s(): %s", __func__, buf.buf);
508 printbuf_exit(&buf);
509 }
510
511 struct found_btree_node search = {
512 .btree_id = btree,
513 .level = level,
514 .min_key = node_min,
515 .max_key = node_max,
516 };
517
518 for_each_found_btree_node_in_range(f, search, idx) {
519 struct found_btree_node n = f->nodes.data[idx];
520
521 n.range_updated |= bpos_lt(n.min_key, node_min);
522 n.min_key = bpos_max(n.min_key, node_min);
523
524 n.range_updated |= bpos_gt(n.max_key, node_max);
525 n.max_key = bpos_min(n.max_key, node_max);
526
527 struct { __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX); } tmp;
528
529 found_btree_node_to_key(&tmp.k, &n);
530
531 struct printbuf buf = PRINTBUF;
532 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&tmp.k));
533 bch_verbose(c, "%s(): recovering %s", __func__, buf.buf);
534 printbuf_exit(&buf);
535
536 BUG_ON(bch2_bkey_validate(c, bkey_i_to_s_c(&tmp.k), BKEY_TYPE_btree, 0));
537
538 ret = bch2_journal_key_insert(c, btree, level + 1, &tmp.k);
539 if (ret)
540 return ret;
541 }
542
543 return 0;
544 }
545
546 void bch2_find_btree_nodes_exit(struct find_btree_nodes *f)
547 {
548 darray_exit(&f->nodes);
549 }
Linux Kernel