Commit | Line | Data |
---|---|---|
1c6fdbd8 KO |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | ||
3 | #include "bcachefs.h" | |
7b3f84ea | 4 | #include "alloc_foreground.h" |
1c6fdbd8 KO |
5 | #include "btree_gc.h" |
6 | #include "btree_update.h" | |
7ef2a73a | 7 | #include "btree_update_interior.h" |
1c6fdbd8 | 8 | #include "buckets.h" |
4628529f | 9 | #include "disk_groups.h" |
1c6fdbd8 KO |
10 | #include "inode.h" |
11 | #include "io.h" | |
12 | #include "journal_reclaim.h" | |
13 | #include "keylist.h" | |
14 | #include "move.h" | |
15 | #include "replicas.h" | |
16 | #include "super-io.h" | |
17 | #include "trace.h" | |
18 | ||
19 | #include <linux/ioprio.h> | |
20 | #include <linux/kthread.h> | |
21 | ||
22 | #define SECTORS_IN_FLIGHT_PER_DEVICE 2048 | |
23 | ||
24 | struct moving_io { | |
25 | struct list_head list; | |
26 | struct closure cl; | |
27 | bool read_completed; | |
28 | ||
29 | unsigned read_sectors; | |
30 | unsigned write_sectors; | |
31 | ||
32 | struct bch_read_bio rbio; | |
33 | ||
34 | struct migrate_write write; | |
35 | /* Must be last since it is variable size */ | |
36 | struct bio_vec bi_inline_vecs[0]; | |
37 | }; | |
38 | ||
39 | struct moving_context { | |
40 | /* Closure for waiting on all reads and writes to complete */ | |
41 | struct closure cl; | |
42 | ||
43 | struct bch_move_stats *stats; | |
44 | ||
45 | struct list_head reads; | |
46 | ||
47 | /* in flight sectors: */ | |
48 | atomic_t read_sectors; | |
49 | atomic_t write_sectors; | |
50 | ||
51 | wait_queue_head_t wait; | |
52 | }; | |
53 | ||
54 | static int bch2_migrate_index_update(struct bch_write_op *op) | |
55 | { | |
56 | struct bch_fs *c = op->c; | |
0564b167 KO |
57 | struct btree_trans trans; |
58 | struct btree_iter *iter; | |
1c6fdbd8 KO |
59 | struct migrate_write *m = |
60 | container_of(op, struct migrate_write, op); | |
61 | struct keylist *keys = &op->insert_keys; | |
1c6fdbd8 KO |
62 | int ret = 0; |
63 | ||
0564b167 KO |
64 | bch2_trans_init(&trans, c); |
65 | ||
66 | iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, | |
67 | bkey_start_pos(&bch2_keylist_front(keys)->k), | |
68 | BTREE_ITER_SLOTS|BTREE_ITER_INTENT); | |
1c6fdbd8 KO |
69 | |
70 | while (1) { | |
0564b167 | 71 | struct bkey_s_c k = bch2_btree_iter_peek_slot(iter); |
1c6fdbd8 KO |
72 | struct bkey_i_extent *insert, *new = |
73 | bkey_i_to_extent(bch2_keylist_front(keys)); | |
74 | BKEY_PADDED(k) _new, _insert; | |
1742237b KO |
75 | const union bch_extent_entry *entry; |
76 | struct extent_ptr_decoded p; | |
1c6fdbd8 KO |
77 | bool did_work = false; |
78 | int nr; | |
79 | ||
0564b167 KO |
80 | ret = btree_iter_err(k); |
81 | if (ret) | |
1c6fdbd8 | 82 | break; |
1c6fdbd8 KO |
83 | |
84 | if (bversion_cmp(k.k->version, new->k.version) || | |
85 | !bkey_extent_is_data(k.k) || | |
86 | !bch2_extent_matches_ptr(c, bkey_s_c_to_extent(k), | |
87 | m->ptr, m->offset)) | |
88 | goto nomatch; | |
89 | ||
90 | if (m->data_cmd == DATA_REWRITE && | |
91 | !bch2_extent_has_device(bkey_s_c_to_extent(k), | |
92 | m->data_opts.rewrite_dev)) | |
93 | goto nomatch; | |
94 | ||
95 | bkey_reassemble(&_insert.k, k); | |
96 | insert = bkey_i_to_extent(&_insert.k); | |
97 | ||
98 | bkey_copy(&_new.k, bch2_keylist_front(keys)); | |
99 | new = bkey_i_to_extent(&_new.k); | |
100 | ||
0564b167 | 101 | bch2_cut_front(iter->pos, &insert->k_i); |
1c6fdbd8 KO |
102 | bch2_cut_back(new->k.p, &insert->k); |
103 | bch2_cut_back(insert->k.p, &new->k); | |
104 | ||
a2753581 | 105 | if (m->data_cmd == DATA_REWRITE) |
26609b61 KO |
106 | bch2_bkey_drop_device(extent_i_to_s(insert).s, |
107 | m->data_opts.rewrite_dev); | |
1c6fdbd8 | 108 | |
1742237b KO |
109 | extent_for_each_ptr_decode(extent_i_to_s(new), p, entry) { |
110 | if (bch2_extent_has_device(extent_i_to_s_c(insert), p.ptr.dev)) { | |
1c6fdbd8 KO |
111 | /* |
112 | * raced with another move op? extent already | |
113 | * has a pointer to the device we just wrote | |
114 | * data to | |
115 | */ | |
116 | continue; | |
117 | } | |
118 | ||
71c9e0ba | 119 | bch2_extent_ptr_decoded_append(insert, &p); |
1c6fdbd8 KO |
120 | did_work = true; |
121 | } | |
122 | ||
123 | if (!did_work) | |
124 | goto nomatch; | |
125 | ||
126 | bch2_extent_narrow_crcs(insert, | |
127 | (struct bch_extent_crc_unpacked) { 0 }); | |
128 | bch2_extent_normalize(c, extent_i_to_s(insert).s); | |
129 | bch2_extent_mark_replicas_cached(c, extent_i_to_s(insert), | |
130 | op->opts.background_target, | |
131 | op->opts.data_replicas); | |
132 | ||
133 | /* | |
db636adb KO |
134 | * If we're not fully overwriting @k, and it's compressed, we |
135 | * need a reservation for all the pointers in @insert | |
1c6fdbd8 | 136 | */ |
26609b61 | 137 | nr = bch2_bkey_nr_dirty_ptrs(bkey_i_to_s_c(&insert->k_i)) - |
db636adb KO |
138 | m->nr_ptrs_reserved; |
139 | ||
140 | if (insert->k.size < k.k->size && | |
141 | bch2_extent_is_compressed(k) && | |
142 | nr > 0) { | |
1c6fdbd8 KO |
143 | ret = bch2_disk_reservation_add(c, &op->res, |
144 | keylist_sectors(keys) * nr, 0); | |
145 | if (ret) | |
146 | goto out; | |
147 | ||
148 | m->nr_ptrs_reserved += nr; | |
149 | goto next; | |
150 | } | |
151 | ||
0564b167 KO |
152 | bch2_trans_update(&trans, |
153 | BTREE_INSERT_ENTRY(iter, &insert->k_i)); | |
154 | ||
155 | ret = bch2_trans_commit(&trans, &op->res, | |
fc3268c1 | 156 | op_journal_seq(op), |
1c6fdbd8 KO |
157 | BTREE_INSERT_ATOMIC| |
158 | BTREE_INSERT_NOFAIL| | |
159 | BTREE_INSERT_USE_RESERVE| | |
0564b167 | 160 | m->data_opts.btree_insert_flags); |
1c6fdbd8 KO |
161 | if (!ret) |
162 | atomic_long_inc(&c->extent_migrate_done); | |
163 | if (ret == -EINTR) | |
164 | ret = 0; | |
165 | if (ret) | |
166 | break; | |
167 | next: | |
0564b167 | 168 | while (bkey_cmp(iter->pos, bch2_keylist_front(keys)->k.p) >= 0) { |
1c6fdbd8 KO |
169 | bch2_keylist_pop_front(keys); |
170 | if (bch2_keylist_empty(keys)) | |
171 | goto out; | |
172 | } | |
173 | ||
0564b167 | 174 | bch2_cut_front(iter->pos, bch2_keylist_front(keys)); |
1c6fdbd8 KO |
175 | continue; |
176 | nomatch: | |
177 | if (m->ctxt) | |
0564b167 | 178 | atomic64_add(k.k->p.offset - iter->pos.offset, |
1c6fdbd8 KO |
179 | &m->ctxt->stats->sectors_raced); |
180 | atomic_long_inc(&c->extent_migrate_raced); | |
181 | trace_move_race(&new->k); | |
0564b167 | 182 | bch2_btree_iter_next_slot(iter); |
1c6fdbd8 KO |
183 | goto next; |
184 | } | |
185 | out: | |
0564b167 | 186 | bch2_trans_exit(&trans); |
1c6fdbd8 KO |
187 | return ret; |
188 | } | |
189 | ||
190 | void bch2_migrate_read_done(struct migrate_write *m, struct bch_read_bio *rbio) | |
191 | { | |
192 | /* write bio must own pages: */ | |
193 | BUG_ON(!m->op.wbio.bio.bi_vcnt); | |
194 | ||
195 | m->ptr = rbio->pick.ptr; | |
196 | m->offset = rbio->pos.offset - rbio->pick.crc.offset; | |
197 | m->op.devs_have = rbio->devs_have; | |
198 | m->op.pos = rbio->pos; | |
199 | m->op.version = rbio->version; | |
200 | m->op.crc = rbio->pick.crc; | |
201 | m->op.wbio.bio.bi_iter.bi_size = m->op.crc.compressed_size << 9; | |
202 | ||
203 | if (bch2_csum_type_is_encryption(m->op.crc.csum_type)) { | |
204 | m->op.nonce = m->op.crc.nonce + m->op.crc.offset; | |
205 | m->op.csum_type = m->op.crc.csum_type; | |
206 | } | |
207 | ||
208 | if (m->data_cmd == DATA_REWRITE) | |
209 | bch2_dev_list_drop_dev(&m->op.devs_have, m->data_opts.rewrite_dev); | |
210 | } | |
211 | ||
212 | int bch2_migrate_write_init(struct bch_fs *c, struct migrate_write *m, | |
213 | struct write_point_specifier wp, | |
214 | struct bch_io_opts io_opts, | |
215 | enum data_cmd data_cmd, | |
216 | struct data_opts data_opts, | |
217 | struct bkey_s_c k) | |
218 | { | |
219 | int ret; | |
220 | ||
221 | m->data_cmd = data_cmd; | |
222 | m->data_opts = data_opts; | |
223 | m->nr_ptrs_reserved = 0; | |
224 | ||
225 | bch2_write_op_init(&m->op, c, io_opts); | |
226 | m->op.compression_type = | |
227 | bch2_compression_opt_to_type[io_opts.background_compression ?: | |
228 | io_opts.compression]; | |
229 | m->op.target = data_opts.target, | |
230 | m->op.write_point = wp; | |
231 | ||
232 | if (m->data_opts.btree_insert_flags & BTREE_INSERT_USE_RESERVE) | |
233 | m->op.alloc_reserve = RESERVE_MOVINGGC; | |
234 | ||
235 | m->op.flags |= BCH_WRITE_ONLY_SPECIFIED_DEVS| | |
236 | BCH_WRITE_PAGES_STABLE| | |
237 | BCH_WRITE_PAGES_OWNED| | |
1d25849c | 238 | BCH_WRITE_DATA_ENCODED; |
1c6fdbd8 KO |
239 | |
240 | m->op.nr_replicas = 1; | |
241 | m->op.nr_replicas_required = 1; | |
242 | m->op.index_update_fn = bch2_migrate_index_update; | |
243 | ||
244 | switch (data_cmd) { | |
245 | case DATA_ADD_REPLICAS: { | |
db636adb KO |
246 | /* |
247 | * DATA_ADD_REPLICAS is used for moving data to a different | |
248 | * device in the background, and due to compression the new copy | |
249 | * might take up more space than the old copy: | |
250 | */ | |
251 | #if 0 | |
1c6fdbd8 | 252 | int nr = (int) io_opts.data_replicas - |
26609b61 | 253 | bch2_bkey_nr_dirty_ptrs(k); |
db636adb KO |
254 | #endif |
255 | int nr = (int) io_opts.data_replicas; | |
1c6fdbd8 KO |
256 | |
257 | if (nr > 0) { | |
258 | m->op.nr_replicas = m->nr_ptrs_reserved = nr; | |
259 | ||
260 | ret = bch2_disk_reservation_get(c, &m->op.res, | |
261 | k.k->size, m->op.nr_replicas, 0); | |
262 | if (ret) | |
263 | return ret; | |
264 | } | |
265 | break; | |
266 | } | |
4628529f KO |
267 | case DATA_REWRITE: { |
268 | const union bch_extent_entry *entry; | |
269 | struct extent_ptr_decoded p; | |
270 | unsigned compressed_sectors = 0; | |
271 | ||
272 | extent_for_each_ptr_decode(bkey_s_c_to_extent(k), p, entry) | |
273 | if (!p.ptr.cached && | |
274 | p.crc.compression_type != BCH_COMPRESSION_NONE && | |
275 | bch2_dev_in_target(c, p.ptr.dev, data_opts.target)) | |
276 | compressed_sectors += p.crc.compressed_size; | |
277 | ||
278 | if (compressed_sectors) { | |
279 | ret = bch2_disk_reservation_add(c, &m->op.res, | |
280 | compressed_sectors, | |
281 | BCH_DISK_RESERVATION_NOFAIL); | |
282 | if (ret) | |
283 | return ret; | |
284 | } | |
1c6fdbd8 | 285 | break; |
4628529f | 286 | } |
1c6fdbd8 KO |
287 | case DATA_PROMOTE: |
288 | m->op.flags |= BCH_WRITE_ALLOC_NOWAIT; | |
289 | m->op.flags |= BCH_WRITE_CACHED; | |
290 | break; | |
291 | default: | |
292 | BUG(); | |
293 | } | |
294 | ||
295 | return 0; | |
296 | } | |
297 | ||
298 | static void move_free(struct closure *cl) | |
299 | { | |
300 | struct moving_io *io = container_of(cl, struct moving_io, cl); | |
301 | struct moving_context *ctxt = io->write.ctxt; | |
302 | struct bvec_iter_all iter; | |
303 | struct bio_vec *bv; | |
304 | ||
305 | bch2_disk_reservation_put(io->write.op.c, &io->write.op.res); | |
306 | ||
307 | bio_for_each_segment_all(bv, &io->write.op.wbio.bio, iter) | |
308 | if (bv->bv_page) | |
309 | __free_page(bv->bv_page); | |
310 | ||
311 | wake_up(&ctxt->wait); | |
312 | ||
313 | kfree(io); | |
314 | } | |
315 | ||
316 | static void move_write_done(struct closure *cl) | |
317 | { | |
318 | struct moving_io *io = container_of(cl, struct moving_io, cl); | |
319 | ||
320 | atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors); | |
321 | closure_return_with_destructor(cl, move_free); | |
322 | } | |
323 | ||
324 | static void move_write(struct closure *cl) | |
325 | { | |
326 | struct moving_io *io = container_of(cl, struct moving_io, cl); | |
327 | ||
328 | if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) { | |
329 | closure_return_with_destructor(cl, move_free); | |
330 | return; | |
331 | } | |
332 | ||
333 | bch2_migrate_read_done(&io->write, &io->rbio); | |
334 | ||
335 | atomic_add(io->write_sectors, &io->write.ctxt->write_sectors); | |
336 | closure_call(&io->write.op.cl, bch2_write, NULL, cl); | |
337 | continue_at(cl, move_write_done, NULL); | |
338 | } | |
339 | ||
340 | static inline struct moving_io *next_pending_write(struct moving_context *ctxt) | |
341 | { | |
342 | struct moving_io *io = | |
343 | list_first_entry_or_null(&ctxt->reads, struct moving_io, list); | |
344 | ||
345 | return io && io->read_completed ? io : NULL; | |
346 | } | |
347 | ||
348 | static void move_read_endio(struct bio *bio) | |
349 | { | |
350 | struct moving_io *io = container_of(bio, struct moving_io, rbio.bio); | |
351 | struct moving_context *ctxt = io->write.ctxt; | |
352 | ||
353 | atomic_sub(io->read_sectors, &ctxt->read_sectors); | |
354 | io->read_completed = true; | |
355 | ||
356 | if (next_pending_write(ctxt)) | |
357 | wake_up(&ctxt->wait); | |
358 | ||
359 | closure_put(&ctxt->cl); | |
360 | } | |
361 | ||
362 | static void do_pending_writes(struct moving_context *ctxt) | |
363 | { | |
364 | struct moving_io *io; | |
365 | ||
366 | while ((io = next_pending_write(ctxt))) { | |
367 | list_del(&io->list); | |
368 | closure_call(&io->cl, move_write, NULL, &ctxt->cl); | |
369 | } | |
370 | } | |
371 | ||
372 | #define move_ctxt_wait_event(_ctxt, _cond) \ | |
373 | do { \ | |
374 | do_pending_writes(_ctxt); \ | |
375 | \ | |
376 | if (_cond) \ | |
377 | break; \ | |
378 | __wait_event((_ctxt)->wait, \ | |
379 | next_pending_write(_ctxt) || (_cond)); \ | |
380 | } while (1) | |
381 | ||
382 | static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt) | |
383 | { | |
384 | unsigned sectors_pending = atomic_read(&ctxt->write_sectors); | |
385 | ||
386 | move_ctxt_wait_event(ctxt, | |
387 | !atomic_read(&ctxt->write_sectors) || | |
388 | atomic_read(&ctxt->write_sectors) != sectors_pending); | |
389 | } | |
390 | ||
391 | static int bch2_move_extent(struct bch_fs *c, | |
392 | struct moving_context *ctxt, | |
393 | struct write_point_specifier wp, | |
394 | struct bch_io_opts io_opts, | |
395 | struct bkey_s_c_extent e, | |
396 | enum data_cmd data_cmd, | |
397 | struct data_opts data_opts) | |
398 | { | |
399 | struct moving_io *io; | |
1742237b KO |
400 | const union bch_extent_entry *entry; |
401 | struct extent_ptr_decoded p; | |
1c6fdbd8 KO |
402 | unsigned sectors = e.k->size, pages; |
403 | int ret = -ENOMEM; | |
404 | ||
405 | move_ctxt_wait_event(ctxt, | |
406 | atomic_read(&ctxt->write_sectors) < | |
407 | SECTORS_IN_FLIGHT_PER_DEVICE); | |
408 | ||
409 | move_ctxt_wait_event(ctxt, | |
410 | atomic_read(&ctxt->read_sectors) < | |
411 | SECTORS_IN_FLIGHT_PER_DEVICE); | |
412 | ||
413 | /* write path might have to decompress data: */ | |
1742237b KO |
414 | extent_for_each_ptr_decode(e, p, entry) |
415 | sectors = max_t(unsigned, sectors, p.crc.uncompressed_size); | |
1c6fdbd8 KO |
416 | |
417 | pages = DIV_ROUND_UP(sectors, PAGE_SECTORS); | |
418 | io = kzalloc(sizeof(struct moving_io) + | |
419 | sizeof(struct bio_vec) * pages, GFP_KERNEL); | |
420 | if (!io) | |
421 | goto err; | |
422 | ||
423 | io->write.ctxt = ctxt; | |
424 | io->read_sectors = e.k->size; | |
425 | io->write_sectors = e.k->size; | |
426 | ||
427 | bio_init(&io->write.op.wbio.bio, NULL, io->bi_inline_vecs, pages, 0); | |
428 | bio_set_prio(&io->write.op.wbio.bio, | |
429 | IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0)); | |
430 | ||
431 | if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9, | |
432 | GFP_KERNEL)) | |
433 | goto err_free; | |
434 | ||
435 | io->rbio.opts = io_opts; | |
436 | bio_init(&io->rbio.bio, NULL, io->bi_inline_vecs, pages, 0); | |
437 | io->rbio.bio.bi_vcnt = pages; | |
438 | bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0)); | |
439 | io->rbio.bio.bi_iter.bi_size = sectors << 9; | |
440 | ||
441 | io->rbio.bio.bi_opf = REQ_OP_READ; | |
442 | io->rbio.bio.bi_iter.bi_sector = bkey_start_offset(e.k); | |
443 | io->rbio.bio.bi_end_io = move_read_endio; | |
444 | ||
445 | ret = bch2_migrate_write_init(c, &io->write, wp, io_opts, | |
446 | data_cmd, data_opts, e.s_c); | |
447 | if (ret) | |
448 | goto err_free_pages; | |
449 | ||
450 | atomic64_inc(&ctxt->stats->keys_moved); | |
451 | atomic64_add(e.k->size, &ctxt->stats->sectors_moved); | |
452 | ||
453 | trace_move_extent(e.k); | |
454 | ||
455 | atomic_add(io->read_sectors, &ctxt->read_sectors); | |
456 | list_add_tail(&io->list, &ctxt->reads); | |
457 | ||
458 | /* | |
459 | * dropped by move_read_endio() - guards against use after free of | |
460 | * ctxt when doing wakeup | |
461 | */ | |
462 | closure_get(&ctxt->cl); | |
463 | bch2_read_extent(c, &io->rbio, e.s_c, | |
464 | BCH_READ_NODECODE| | |
465 | BCH_READ_LAST_FRAGMENT); | |
466 | return 0; | |
467 | err_free_pages: | |
468 | bio_free_pages(&io->write.op.wbio.bio); | |
469 | err_free: | |
470 | kfree(io); | |
471 | err: | |
472 | trace_move_alloc_fail(e.k); | |
473 | return ret; | |
474 | } | |
475 | ||
476 | int bch2_move_data(struct bch_fs *c, | |
477 | struct bch_ratelimit *rate, | |
478 | struct write_point_specifier wp, | |
479 | struct bpos start, | |
480 | struct bpos end, | |
481 | move_pred_fn pred, void *arg, | |
482 | struct bch_move_stats *stats) | |
483 | { | |
484 | bool kthread = (current->flags & PF_KTHREAD) != 0; | |
485 | struct moving_context ctxt = { .stats = stats }; | |
486 | struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts); | |
487 | BKEY_PADDED(k) tmp; | |
488 | struct bkey_s_c k; | |
1c6fdbd8 KO |
489 | struct data_opts data_opts; |
490 | enum data_cmd data_cmd; | |
c2fcff59 | 491 | u64 delay, cur_inum = U64_MAX; |
1c6fdbd8 KO |
492 | int ret = 0, ret2; |
493 | ||
494 | closure_init_stack(&ctxt.cl); | |
495 | INIT_LIST_HEAD(&ctxt.reads); | |
496 | init_waitqueue_head(&ctxt.wait); | |
497 | ||
498 | stats->data_type = BCH_DATA_USER; | |
499 | bch2_btree_iter_init(&stats->iter, c, BTREE_ID_EXTENTS, start, | |
500 | BTREE_ITER_PREFETCH); | |
501 | ||
502 | if (rate) | |
503 | bch2_ratelimit_reset(rate); | |
504 | ||
c2fcff59 KO |
505 | while (1) { |
506 | do { | |
507 | delay = rate ? bch2_ratelimit_delay(rate) : 0; | |
508 | ||
509 | if (delay) { | |
510 | bch2_btree_iter_unlock(&stats->iter); | |
511 | set_current_state(TASK_INTERRUPTIBLE); | |
512 | } | |
513 | ||
514 | if (kthread && (ret = kthread_should_stop())) { | |
515 | __set_current_state(TASK_RUNNING); | |
516 | goto out; | |
517 | } | |
518 | ||
519 | if (delay) | |
520 | schedule_timeout(delay); | |
521 | ||
522 | if (unlikely(freezing(current))) { | |
523 | bch2_btree_iter_unlock(&stats->iter); | |
524 | move_ctxt_wait_event(&ctxt, list_empty(&ctxt.reads)); | |
c2fcff59 KO |
525 | try_to_freeze(); |
526 | } | |
527 | } while (delay); | |
1c6fdbd8 KO |
528 | peek: |
529 | k = bch2_btree_iter_peek(&stats->iter); | |
530 | if (!k.k) | |
531 | break; | |
532 | ret = btree_iter_err(k); | |
533 | if (ret) | |
534 | break; | |
535 | if (bkey_cmp(bkey_start_pos(k.k), end) >= 0) | |
536 | break; | |
537 | ||
538 | if (!bkey_extent_is_data(k.k)) | |
539 | goto next_nondata; | |
540 | ||
1c6fdbd8 KO |
541 | if (cur_inum != k.k->p.inode) { |
542 | struct bch_inode_unpacked inode; | |
543 | ||
544 | /* don't hold btree locks while looking up inode: */ | |
545 | bch2_btree_iter_unlock(&stats->iter); | |
546 | ||
547 | io_opts = bch2_opts_to_inode_opts(c->opts); | |
548 | if (!bch2_inode_find_by_inum(c, k.k->p.inode, &inode)) | |
549 | bch2_io_opts_apply(&io_opts, bch2_inode_opts_get(&inode)); | |
550 | cur_inum = k.k->p.inode; | |
551 | goto peek; | |
552 | } | |
553 | ||
26609b61 | 554 | switch ((data_cmd = pred(c, arg, k, &io_opts, &data_opts))) { |
1c6fdbd8 KO |
555 | case DATA_SKIP: |
556 | goto next; | |
557 | case DATA_SCRUB: | |
558 | BUG(); | |
559 | case DATA_ADD_REPLICAS: | |
560 | case DATA_REWRITE: | |
561 | case DATA_PROMOTE: | |
562 | break; | |
563 | default: | |
564 | BUG(); | |
565 | } | |
566 | ||
567 | /* unlock before doing IO: */ | |
568 | bkey_reassemble(&tmp.k, k); | |
569 | k = bkey_i_to_s_c(&tmp.k); | |
570 | bch2_btree_iter_unlock(&stats->iter); | |
571 | ||
572 | ret2 = bch2_move_extent(c, &ctxt, wp, io_opts, | |
573 | bkey_s_c_to_extent(k), | |
574 | data_cmd, data_opts); | |
575 | if (ret2) { | |
576 | if (ret2 == -ENOMEM) { | |
577 | /* memory allocation failure, wait for some IO to finish */ | |
578 | bch2_move_ctxt_wait_for_io(&ctxt); | |
579 | continue; | |
580 | } | |
581 | ||
582 | /* XXX signal failure */ | |
583 | goto next; | |
584 | } | |
585 | ||
586 | if (rate) | |
587 | bch2_ratelimit_increment(rate, k.k->size); | |
588 | next: | |
26609b61 | 589 | atomic64_add(k.k->size * bch2_bkey_nr_dirty_ptrs(k), |
1c6fdbd8 KO |
590 | &stats->sectors_seen); |
591 | next_nondata: | |
592 | bch2_btree_iter_next(&stats->iter); | |
593 | bch2_btree_iter_cond_resched(&stats->iter); | |
594 | } | |
c2fcff59 | 595 | out: |
1c6fdbd8 KO |
596 | bch2_btree_iter_unlock(&stats->iter); |
597 | ||
598 | move_ctxt_wait_event(&ctxt, list_empty(&ctxt.reads)); | |
599 | closure_sync(&ctxt.cl); | |
600 | ||
601 | EBUG_ON(atomic_read(&ctxt.write_sectors)); | |
602 | ||
603 | trace_move_data(c, | |
604 | atomic64_read(&stats->sectors_moved), | |
605 | atomic64_read(&stats->keys_moved)); | |
606 | ||
607 | return ret; | |
608 | } | |
609 | ||
610 | static int bch2_gc_data_replicas(struct bch_fs *c) | |
611 | { | |
612 | struct btree_iter iter; | |
613 | struct bkey_s_c k; | |
614 | int ret; | |
615 | ||
616 | mutex_lock(&c->replicas_gc_lock); | |
617 | bch2_replicas_gc_start(c, (1 << BCH_DATA_USER)|(1 << BCH_DATA_CACHED)); | |
618 | ||
619 | for_each_btree_key(&iter, c, BTREE_ID_EXTENTS, POS_MIN, | |
620 | BTREE_ITER_PREFETCH, k) { | |
26609b61 | 621 | ret = bch2_mark_bkey_replicas(c, k); |
1c6fdbd8 KO |
622 | if (ret) |
623 | break; | |
624 | } | |
625 | ret = bch2_btree_iter_unlock(&iter) ?: ret; | |
626 | ||
627 | bch2_replicas_gc_end(c, ret); | |
628 | mutex_unlock(&c->replicas_gc_lock); | |
629 | ||
630 | return ret; | |
631 | } | |
632 | ||
633 | static int bch2_gc_btree_replicas(struct bch_fs *c) | |
634 | { | |
635 | struct btree_iter iter; | |
636 | struct btree *b; | |
637 | unsigned id; | |
638 | int ret = 0; | |
639 | ||
640 | mutex_lock(&c->replicas_gc_lock); | |
641 | bch2_replicas_gc_start(c, 1 << BCH_DATA_BTREE); | |
642 | ||
643 | for (id = 0; id < BTREE_ID_NR; id++) { | |
644 | for_each_btree_node(&iter, c, id, POS_MIN, BTREE_ITER_PREFETCH, b) { | |
26609b61 | 645 | ret = bch2_mark_bkey_replicas(c, bkey_i_to_s_c(&b->key)); |
1c6fdbd8 KO |
646 | |
647 | bch2_btree_iter_cond_resched(&iter); | |
648 | } | |
649 | ||
650 | ret = bch2_btree_iter_unlock(&iter) ?: ret; | |
651 | } | |
652 | ||
653 | bch2_replicas_gc_end(c, ret); | |
654 | mutex_unlock(&c->replicas_gc_lock); | |
655 | ||
656 | return ret; | |
657 | } | |
658 | ||
659 | static int bch2_move_btree(struct bch_fs *c, | |
660 | move_pred_fn pred, | |
661 | void *arg, | |
662 | struct bch_move_stats *stats) | |
663 | { | |
664 | struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts); | |
665 | struct btree *b; | |
666 | unsigned id; | |
667 | struct data_opts data_opts; | |
668 | enum data_cmd cmd; | |
669 | int ret = 0; | |
670 | ||
671 | stats->data_type = BCH_DATA_BTREE; | |
672 | ||
673 | for (id = 0; id < BTREE_ID_NR; id++) { | |
674 | for_each_btree_node(&stats->iter, c, id, POS_MIN, BTREE_ITER_PREFETCH, b) { | |
26609b61 KO |
675 | switch ((cmd = pred(c, arg, |
676 | bkey_i_to_s_c(&b->key), | |
677 | &io_opts, &data_opts))) { | |
1c6fdbd8 KO |
678 | case DATA_SKIP: |
679 | goto next; | |
680 | case DATA_SCRUB: | |
681 | BUG(); | |
682 | case DATA_ADD_REPLICAS: | |
683 | case DATA_REWRITE: | |
684 | break; | |
685 | default: | |
686 | BUG(); | |
687 | } | |
688 | ||
689 | ret = bch2_btree_node_rewrite(c, &stats->iter, | |
690 | b->data->keys.seq, 0) ?: ret; | |
691 | next: | |
692 | bch2_btree_iter_cond_resched(&stats->iter); | |
693 | } | |
694 | ||
695 | ret = bch2_btree_iter_unlock(&stats->iter) ?: ret; | |
696 | } | |
697 | ||
698 | return ret; | |
699 | } | |
700 | ||
701 | #if 0 | |
702 | static enum data_cmd scrub_pred(struct bch_fs *c, void *arg, | |
26609b61 | 703 | struct bkey_s_c k, |
1c6fdbd8 KO |
704 | struct bch_io_opts *io_opts, |
705 | struct data_opts *data_opts) | |
706 | { | |
707 | return DATA_SCRUB; | |
708 | } | |
709 | #endif | |
710 | ||
711 | static enum data_cmd rereplicate_pred(struct bch_fs *c, void *arg, | |
26609b61 | 712 | struct bkey_s_c k, |
1c6fdbd8 KO |
713 | struct bch_io_opts *io_opts, |
714 | struct data_opts *data_opts) | |
715 | { | |
26609b61 KO |
716 | unsigned nr_good = bch2_bkey_durability(c, k); |
717 | unsigned replicas = 0; | |
718 | ||
719 | switch (k.k->type) { | |
720 | case KEY_TYPE_btree_ptr: | |
721 | replicas = c->opts.metadata_replicas; | |
722 | break; | |
723 | case KEY_TYPE_extent: | |
724 | replicas = io_opts->data_replicas; | |
725 | break; | |
726 | } | |
1c6fdbd8 KO |
727 | |
728 | if (!nr_good || nr_good >= replicas) | |
729 | return DATA_SKIP; | |
730 | ||
731 | data_opts->target = 0; | |
26609b61 | 732 | data_opts->btree_insert_flags = 0; |
1c6fdbd8 KO |
733 | return DATA_ADD_REPLICAS; |
734 | } | |
735 | ||
736 | static enum data_cmd migrate_pred(struct bch_fs *c, void *arg, | |
26609b61 | 737 | struct bkey_s_c k, |
1c6fdbd8 KO |
738 | struct bch_io_opts *io_opts, |
739 | struct data_opts *data_opts) | |
740 | { | |
741 | struct bch_ioctl_data *op = arg; | |
742 | ||
26609b61 | 743 | if (!bch2_bkey_has_device(k, op->migrate.dev)) |
1c6fdbd8 KO |
744 | return DATA_SKIP; |
745 | ||
746 | data_opts->target = 0; | |
747 | data_opts->btree_insert_flags = 0; | |
748 | data_opts->rewrite_dev = op->migrate.dev; | |
749 | return DATA_REWRITE; | |
750 | } | |
751 | ||
752 | int bch2_data_job(struct bch_fs *c, | |
753 | struct bch_move_stats *stats, | |
754 | struct bch_ioctl_data op) | |
755 | { | |
756 | int ret = 0; | |
757 | ||
758 | switch (op.op) { | |
759 | case BCH_DATA_OP_REREPLICATE: | |
760 | stats->data_type = BCH_DATA_JOURNAL; | |
761 | ret = bch2_journal_flush_device_pins(&c->journal, -1); | |
762 | ||
763 | ret = bch2_move_btree(c, rereplicate_pred, c, stats) ?: ret; | |
7ef2a73a KO |
764 | |
765 | while (1) { | |
766 | closure_wait_event(&c->btree_interior_update_wait, | |
767 | !bch2_btree_interior_updates_nr_pending(c) || | |
768 | c->btree_roots_dirty); | |
769 | if (!bch2_btree_interior_updates_nr_pending(c)) | |
770 | break; | |
771 | bch2_journal_meta(&c->journal); | |
772 | } | |
773 | ||
1c6fdbd8 KO |
774 | ret = bch2_gc_btree_replicas(c) ?: ret; |
775 | ||
776 | ret = bch2_move_data(c, NULL, | |
777 | writepoint_hashed((unsigned long) current), | |
778 | op.start, | |
779 | op.end, | |
780 | rereplicate_pred, c, stats) ?: ret; | |
781 | ret = bch2_gc_data_replicas(c) ?: ret; | |
782 | break; | |
783 | case BCH_DATA_OP_MIGRATE: | |
784 | if (op.migrate.dev >= c->sb.nr_devices) | |
785 | return -EINVAL; | |
786 | ||
787 | stats->data_type = BCH_DATA_JOURNAL; | |
788 | ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev); | |
789 | ||
790 | ret = bch2_move_btree(c, migrate_pred, &op, stats) ?: ret; | |
791 | ret = bch2_gc_btree_replicas(c) ?: ret; | |
792 | ||
793 | ret = bch2_move_data(c, NULL, | |
794 | writepoint_hashed((unsigned long) current), | |
795 | op.start, | |
796 | op.end, | |
797 | migrate_pred, &op, stats) ?: ret; | |
798 | ret = bch2_gc_data_replicas(c) ?: ret; | |
799 | break; | |
800 | default: | |
801 | ret = -EINVAL; | |
802 | } | |
803 | ||
804 | return ret; | |
805 | } |