1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Moving/copying garbage collector 4 * 5 * Copyright 2012 Google, Inc. 6 */ 7 8 #include "bcachefs.h" 9 #include "alloc_background.h" 10 #include "alloc_foreground.h" 11 #include "backpointers.h" 12 #include "btree_iter.h" 13 #include "btree_update.h" 14 #include "btree_write_buffer.h" 15 #include "buckets.h" 16 #include "clock.h" 17 #include "errcode.h" 18 #include "error.h" 19 #include "lru.h" 20 #include "move.h" 21 #include "movinggc.h" 22 #include "trace.h" 23 24 #include <linux/freezer.h> 25 #include <linux/kthread.h> 26 #include <linux/math64.h> 27 #include <linux/sched/task.h> 28 #include <linux/wait.h> 29 30 struct buckets_in_flight { 31 struct rhashtable *table; 32 struct move_bucket *first; 33 struct move_bucket *last; 34 size_t nr; 35 size_t sectors; 36 37 DARRAY(struct move_bucket *) to_evacuate; 38 }; 39 40 static const struct rhashtable_params bch_move_bucket_params = { 41 .head_offset = offsetof(struct move_bucket, hash), 42 .key_offset = offsetof(struct move_bucket, k), 43 .key_len = sizeof(struct move_bucket_key), 44 .automatic_shrinking = true, 45 }; 46 47 static void move_bucket_in_flight_add(struct buckets_in_flight *list, struct move_bucket *b) 48 { 49 if (!list->first) 50 list->first = b; 51 else 52 list->last->next = b; 53 54 list->last = b; 55 list->nr++; 56 list->sectors += b->sectors; 57 } 58 59 static int bch2_bucket_is_movable(struct btree_trans *trans, 60 struct move_bucket *b, u64 time) 61 { 62 struct bch_fs *c = trans->c; 63 64 if (bch2_bucket_is_open(c, b->k.bucket.inode, b->k.bucket.offset)) 65 return 0; 66 67 struct btree_iter iter; 68 struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_alloc, 69 b->k.bucket, BTREE_ITER_cached); 70 int ret = bkey_err(k); 71 if (ret) 72 return ret; 73 74 struct bch_dev *ca = bch2_dev_bucket_tryget(c, k.k->p); 75 if (!ca) 76 goto out; 77 78 if (bch2_bucket_bitmap_test(&ca->bucket_backpointer_mismatch, b->k.bucket.offset)) 79 goto out; 80 81 if (ca->mi.state != BCH_MEMBER_STATE_rw || 82 !bch2_dev_is_online(ca)) 83 goto out; 84 85 struct bch_alloc_v4 _a; 86 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a); 87 b->k.gen = a->gen; 88 b->sectors = bch2_bucket_sectors_dirty(*a); 89 u64 lru_idx = alloc_lru_idx_fragmentation(*a, ca); 90 91 ret = lru_idx && lru_idx <= time; 92 out: 93 bch2_dev_put(ca); 94 bch2_trans_iter_exit(trans, &iter); 95 return ret; 96 } 97 98 static void move_bucket_free(struct buckets_in_flight *list, 99 struct move_bucket *b) 100 { 101 int ret = rhashtable_remove_fast(list->table, &b->hash, 102 bch_move_bucket_params); 103 BUG_ON(ret); 104 kfree(b); 105 } 106 107 static void move_buckets_wait(struct moving_context *ctxt, 108 struct buckets_in_flight *list, 109 bool flush) 110 { 111 struct move_bucket *i; 112 113 while ((i = list->first)) { 114 if (flush) 115 move_ctxt_wait_event(ctxt, !atomic_read(&i->count)); 116 117 if (atomic_read(&i->count)) 118 break; 119 120 list->first = i->next; 121 if (!list->first) 122 list->last = NULL; 123 124 list->nr--; 125 list->sectors -= i->sectors; 126 127 move_bucket_free(list, i); 128 } 129 130 bch2_trans_unlock_long(ctxt->trans); 131 } 132 133 static bool bucket_in_flight(struct buckets_in_flight *list, 134 struct move_bucket_key k) 135 { 136 return rhashtable_lookup_fast(list->table, &k, bch_move_bucket_params); 137 } 138 139 static int bch2_copygc_get_buckets(struct moving_context *ctxt, 140 struct buckets_in_flight *buckets_in_flight) 141 { 142 struct btree_trans *trans = ctxt->trans; 143 struct bch_fs *c = trans->c; 144 size_t nr_to_get = max_t(size_t, 16U, buckets_in_flight->nr / 4); 145 size_t saw = 0, in_flight = 0, not_movable = 0, sectors = 0; 146 int ret; 147 148 move_buckets_wait(ctxt, buckets_in_flight, false); 149 150 ret = bch2_btree_write_buffer_tryflush(trans); 151 if (bch2_err_matches(ret, EROFS)) 152 return ret; 153 154 if (bch2_fs_fatal_err_on(ret, c, "%s: from bch2_btree_write_buffer_tryflush()", bch2_err_str(ret))) 155 return ret; 156 157 ret = for_each_btree_key_max(trans, iter, BTREE_ID_lru, 158 lru_pos(BCH_LRU_BUCKET_FRAGMENTATION, 0, 0), 159 lru_pos(BCH_LRU_BUCKET_FRAGMENTATION, U64_MAX, LRU_TIME_MAX), 160 0, k, ({ 161 struct move_bucket b = { .k.bucket = u64_to_bucket(k.k->p.offset) }; 162 int ret2 = 0; 163 164 saw++; 165 166 ret2 = bch2_bucket_is_movable(trans, &b, lru_pos_time(k.k->p)); 167 if (ret2 < 0) 168 goto err; 169 170 if (!ret2) 171 not_movable++; 172 else if (bucket_in_flight(buckets_in_flight, b.k)) 173 in_flight++; 174 else { 175 struct move_bucket *b_i = kmalloc(sizeof(*b_i), GFP_KERNEL); 176 ret2 = b_i ? 0 : -ENOMEM; 177 if (ret2) 178 goto err; 179 180 *b_i = b; 181 182 ret2 = darray_push(&buckets_in_flight->to_evacuate, b_i); 183 if (ret2) { 184 kfree(b_i); 185 goto err; 186 } 187 188 ret2 = rhashtable_lookup_insert_fast(buckets_in_flight->table, &b_i->hash, 189 bch_move_bucket_params); 190 BUG_ON(ret2); 191 192 sectors += b.sectors; 193 } 194 195 ret2 = buckets_in_flight->to_evacuate.nr >= nr_to_get; 196 err: 197 ret2; 198 })); 199 200 pr_debug("have: %zu (%zu) saw %zu in flight %zu not movable %zu got %zu (%zu)/%zu buckets ret %i", 201 buckets_in_flight->nr, buckets_in_flight->sectors, 202 saw, in_flight, not_movable, buckets_in_flight->to_evacuate.nr, sectors, nr_to_get, ret); 203 204 return ret < 0 ? ret : 0; 205 } 206 207 noinline 208 static int bch2_copygc(struct moving_context *ctxt, 209 struct buckets_in_flight *buckets_in_flight, 210 bool *did_work) 211 { 212 struct btree_trans *trans = ctxt->trans; 213 struct bch_fs *c = trans->c; 214 struct data_update_opts data_opts = { 215 .btree_insert_flags = BCH_WATERMARK_copygc, 216 }; 217 u64 sectors_seen = atomic64_read(&ctxt->stats->sectors_seen); 218 u64 sectors_moved = atomic64_read(&ctxt->stats->sectors_moved); 219 int ret = 0; 220 221 ret = bch2_copygc_get_buckets(ctxt, buckets_in_flight); 222 if (ret) 223 goto err; 224 225 darray_for_each(buckets_in_flight->to_evacuate, i) { 226 if (kthread_should_stop() || freezing(current)) 227 break; 228 229 struct move_bucket *b = *i; 230 *i = NULL; 231 232 move_bucket_in_flight_add(buckets_in_flight, b); 233 234 ret = bch2_evacuate_bucket(ctxt, b, b->k.bucket, b->k.gen, data_opts); 235 if (ret) 236 goto err; 237 238 *did_work = true; 239 } 240 err: 241 /* no entries in LRU btree found, or got to end: */ 242 if (bch2_err_matches(ret, ENOENT)) 243 ret = 0; 244 245 if (ret < 0 && !bch2_err_matches(ret, EROFS)) 246 bch_err_msg(c, ret, "from bch2_move_data()"); 247 248 sectors_seen = atomic64_read(&ctxt->stats->sectors_seen) - sectors_seen; 249 sectors_moved = atomic64_read(&ctxt->stats->sectors_moved) - sectors_moved; 250 trace_and_count(c, copygc, c, buckets_in_flight->to_evacuate.nr, sectors_seen, sectors_moved); 251 252 darray_for_each(buckets_in_flight->to_evacuate, i) 253 if (*i) 254 move_bucket_free(buckets_in_flight, *i); 255 darray_exit(&buckets_in_flight->to_evacuate); 256 return ret; 257 } 258 259 static u64 bch2_copygc_dev_wait_amount(struct bch_dev *ca) 260 { 261 struct bch_dev_usage_full usage_full = bch2_dev_usage_full_read(ca); 262 struct bch_dev_usage usage; 263 264 for (unsigned i = 0; i < BCH_DATA_NR; i++) 265 usage.buckets[i] = usage_full.d[i].buckets; 266 267 s64 fragmented_allowed = ((__dev_buckets_available(ca, usage, BCH_WATERMARK_stripe) * 268 ca->mi.bucket_size) >> 1); 269 s64 fragmented = 0; 270 271 for (unsigned i = 0; i < BCH_DATA_NR; i++) 272 if (data_type_movable(i)) 273 fragmented += usage_full.d[i].fragmented; 274 275 return max(0LL, fragmented_allowed - fragmented); 276 } 277 278 /* 279 * Copygc runs when the amount of fragmented data is above some arbitrary 280 * threshold: 281 * 282 * The threshold at the limit - when the device is full - is the amount of space 283 * we reserved in bch2_recalc_capacity; we can't have more than that amount of 284 * disk space stranded due to fragmentation and store everything we have 285 * promised to store. 286 * 287 * But we don't want to be running copygc unnecessarily when the device still 288 * has plenty of free space - rather, we want copygc to smoothly run every so 289 * often and continually reduce the amount of fragmented space as the device 290 * fills up. So, we increase the threshold by half the current free space. 291 */ 292 u64 bch2_copygc_wait_amount(struct bch_fs *c) 293 { 294 u64 wait = U64_MAX; 295 296 guard(rcu)(); 297 for_each_rw_member_rcu(c, ca) 298 wait = min(wait, bch2_copygc_dev_wait_amount(ca)); 299 return wait; 300 } 301 302 void bch2_copygc_wait_to_text(struct printbuf *out, struct bch_fs *c) 303 { 304 printbuf_tabstop_push(out, 32); 305 prt_printf(out, "running:\t%u\n", c->copygc_running); 306 prt_printf(out, "copygc_wait:\t%llu\n", c->copygc_wait); 307 prt_printf(out, "copygc_wait_at:\t%llu\n", c->copygc_wait_at); 308 309 prt_printf(out, "Currently waiting for:\t"); 310 prt_human_readable_u64(out, max(0LL, c->copygc_wait - 311 atomic64_read(&c->io_clock[WRITE].now)) << 9); 312 prt_newline(out); 313 314 prt_printf(out, "Currently waiting since:\t"); 315 prt_human_readable_u64(out, max(0LL, 316 atomic64_read(&c->io_clock[WRITE].now) - 317 c->copygc_wait_at) << 9); 318 prt_newline(out); 319 320 bch2_printbuf_make_room(out, 4096); 321 322 struct task_struct *t; 323 out->atomic++; 324 scoped_guard(rcu) { 325 prt_printf(out, "Currently calculated wait:\n"); 326 for_each_rw_member_rcu(c, ca) { 327 prt_printf(out, " %s:\t", ca->name); 328 prt_human_readable_u64(out, bch2_copygc_dev_wait_amount(ca)); 329 prt_newline(out); 330 } 331 332 t = rcu_dereference(c->copygc_thread); 333 if (t) 334 get_task_struct(t); 335 } 336 --out->atomic; 337 338 if (t) { 339 bch2_prt_task_backtrace(out, t, 0, GFP_KERNEL); 340 put_task_struct(t); 341 } 342 } 343 344 static int bch2_copygc_thread(void *arg) 345 { 346 struct bch_fs *c = arg; 347 struct moving_context ctxt; 348 struct bch_move_stats move_stats; 349 struct io_clock *clock = &c->io_clock[WRITE]; 350 struct buckets_in_flight buckets = {}; 351 u64 last, wait; 352 353 buckets.table = kzalloc(sizeof(*buckets.table), GFP_KERNEL); 354 int ret = !buckets.table 355 ? -ENOMEM 356 : rhashtable_init(buckets.table, &bch_move_bucket_params); 357 bch_err_msg(c, ret, "allocating copygc buckets in flight"); 358 if (ret) 359 goto err; 360 361 set_freezable(); 362 363 /* 364 * Data move operations can't run until after check_snapshots has 365 * completed, and bch2_snapshot_is_ancestor() is available. 366 */ 367 kthread_wait_freezable(c->recovery.pass_done > BCH_RECOVERY_PASS_check_snapshots || 368 kthread_should_stop()); 369 370 bch2_move_stats_init(&move_stats, "copygc"); 371 bch2_moving_ctxt_init(&ctxt, c, NULL, &move_stats, 372 writepoint_ptr(&c->copygc_write_point), 373 false); 374 375 while (!ret && !kthread_should_stop()) { 376 bool did_work = false; 377 378 bch2_trans_unlock_long(ctxt.trans); 379 cond_resched(); 380 381 if (!c->opts.copygc_enabled) { 382 move_buckets_wait(&ctxt, &buckets, true); 383 kthread_wait_freezable(c->opts.copygc_enabled || 384 kthread_should_stop()); 385 } 386 387 if (unlikely(freezing(current))) { 388 move_buckets_wait(&ctxt, &buckets, true); 389 __refrigerator(false); 390 continue; 391 } 392 393 last = atomic64_read(&clock->now); 394 wait = bch2_copygc_wait_amount(c); 395 396 if (wait > clock->max_slop) { 397 c->copygc_wait_at = last; 398 c->copygc_wait = last + wait; 399 move_buckets_wait(&ctxt, &buckets, true); 400 trace_and_count(c, copygc_wait, c, wait, last + wait); 401 bch2_kthread_io_clock_wait(clock, last + wait, 402 MAX_SCHEDULE_TIMEOUT); 403 continue; 404 } 405 406 c->copygc_wait = 0; 407 408 c->copygc_running = true; 409 ret = bch2_copygc(&ctxt, &buckets, &did_work); 410 c->copygc_running = false; 411 412 wake_up(&c->copygc_running_wq); 413 414 if (!wait && !did_work) { 415 u64 min_member_capacity = bch2_min_rw_member_capacity(c); 416 417 if (min_member_capacity == U64_MAX) 418 min_member_capacity = 128 * 2048; 419 420 move_buckets_wait(&ctxt, &buckets, true); 421 bch2_kthread_io_clock_wait(clock, last + (min_member_capacity >> 6), 422 MAX_SCHEDULE_TIMEOUT); 423 } 424 } 425 426 move_buckets_wait(&ctxt, &buckets, true); 427 rhashtable_destroy(buckets.table); 428 bch2_moving_ctxt_exit(&ctxt); 429 bch2_move_stats_exit(&move_stats, c); 430 err: 431 kfree(buckets.table); 432 return ret; 433 } 434 435 void bch2_copygc_stop(struct bch_fs *c) 436 { 437 if (c->copygc_thread) { 438 kthread_stop(c->copygc_thread); 439 put_task_struct(c->copygc_thread); 440 } 441 c->copygc_thread = NULL; 442 } 443 444 int bch2_copygc_start(struct bch_fs *c) 445 { 446 struct task_struct *t; 447 int ret; 448 449 if (c->copygc_thread) 450 return 0; 451 452 if (c->opts.nochanges) 453 return 0; 454 455 if (bch2_fs_init_fault("copygc_start")) 456 return -ENOMEM; 457 458 t = kthread_create(bch2_copygc_thread, c, "bch-copygc/%s", c->name); 459 ret = PTR_ERR_OR_ZERO(t); 460 bch_err_msg(c, ret, "creating copygc thread"); 461 if (ret) 462 return ret; 463 464 get_task_struct(t); 465 466 c->copygc_thread = t; 467 wake_up_process(c->copygc_thread); 468 469 return 0; 470 } 471 472 void bch2_fs_copygc_init(struct bch_fs *c) 473 { 474 init_waitqueue_head(&c->copygc_running_wq); 475 c->copygc_running = false; 476 } 477