1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _BCACHEFS_BTREE_UPDATE_H 3 #define _BCACHEFS_BTREE_UPDATE_H 4 5 #include "btree_iter.h" 6 #include "journal.h" 7 8 struct bch_fs; 9 struct btree; 10 11 void bch2_btree_node_prep_for_write(struct btree_trans *, 12 struct btree_path *, struct btree *); 13 bool bch2_btree_bset_insert_key(struct btree_trans *, struct btree_path *, 14 struct btree *, struct btree_node_iter *, 15 struct bkey_i *); 16 17 int bch2_btree_node_flush0(struct journal *, struct journal_entry_pin *, u64); 18 int bch2_btree_node_flush1(struct journal *, struct journal_entry_pin *, u64); 19 void bch2_btree_add_journal_pin(struct bch_fs *, struct btree *, u64); 20 21 void bch2_btree_insert_key_leaf(struct btree_trans *, struct btree_path *, 22 struct bkey_i *, u64); 23 24 #define BCH_TRANS_COMMIT_FLAGS() \ 25 x(no_enospc, "don't check for enospc") \ 26 x(no_check_rw, "don't attempt to take a ref on c->writes") \ 27 x(no_journal_res, "don't take a journal reservation, instead " \ 28 "pin journal entry referred to by trans->journal_res.seq") \ 29 x(journal_reclaim, "operation required for journal reclaim; may return error" \ 30 "instead of deadlocking if BCH_WATERMARK_reclaim not specified")\ 31 x(skip_accounting_apply, "we're in journal replay - accounting updates have already been applied") 32 33 enum __bch_trans_commit_flags { 34 /* First bits for bch_watermark: */ 35 __BCH_TRANS_COMMIT_FLAGS_START = BCH_WATERMARK_BITS, 36 #define x(n, ...) __BCH_TRANS_COMMIT_##n, 37 BCH_TRANS_COMMIT_FLAGS() 38 #undef x 39 }; 40 41 enum bch_trans_commit_flags { 42 #define x(n, ...) BCH_TRANS_COMMIT_##n = BIT(__BCH_TRANS_COMMIT_##n), 43 BCH_TRANS_COMMIT_FLAGS() 44 #undef x 45 }; 46 47 void bch2_trans_commit_flags_to_text(struct printbuf *, enum bch_trans_commit_flags); 48 49 int bch2_btree_delete_at(struct btree_trans *, struct btree_iter *, unsigned); 50 int bch2_btree_delete(struct btree_trans *, enum btree_id, struct bpos, unsigned); 51 52 int bch2_btree_insert_nonextent(struct btree_trans *, enum btree_id, 53 struct bkey_i *, enum btree_iter_update_trigger_flags); 54 55 int bch2_btree_insert_trans(struct btree_trans *, enum btree_id, struct bkey_i *, 56 enum btree_iter_update_trigger_flags); 57 int bch2_btree_insert(struct bch_fs *, enum btree_id, struct bkey_i *, struct 58 disk_reservation *, int flags, enum 59 btree_iter_update_trigger_flags iter_flags); 60 61 int bch2_btree_delete_range_trans(struct btree_trans *, enum btree_id, 62 struct bpos, struct bpos, unsigned, u64 *); 63 int bch2_btree_delete_range(struct bch_fs *, enum btree_id, 64 struct bpos, struct bpos, unsigned, u64 *); 65 66 int bch2_btree_bit_mod_iter(struct btree_trans *, struct btree_iter *, bool); 67 int bch2_btree_bit_mod(struct btree_trans *, enum btree_id, struct bpos, bool); 68 int bch2_btree_bit_mod_buffered(struct btree_trans *, enum btree_id, struct bpos, bool); 69 70 static inline int bch2_btree_delete_at_buffered(struct btree_trans *trans, 71 enum btree_id btree, struct bpos pos) 72 { 73 return bch2_btree_bit_mod_buffered(trans, btree, pos, false); 74 } 75 76 int __bch2_insert_snapshot_whiteouts(struct btree_trans *, enum btree_id, 77 struct bpos, struct bpos); 78 79 /* 80 * For use when splitting extents in existing snapshots: 81 * 82 * If @old_pos is an interior snapshot node, iterate over descendent snapshot 83 * nodes: for every descendent snapshot in whiche @old_pos is overwritten and 84 * not visible, emit a whiteout at @new_pos. 85 */ 86 static inline int bch2_insert_snapshot_whiteouts(struct btree_trans *trans, 87 enum btree_id btree, 88 struct bpos old_pos, 89 struct bpos new_pos) 90 { 91 if (!btree_type_has_snapshots(btree) || 92 bkey_eq(old_pos, new_pos)) 93 return 0; 94 95 return __bch2_insert_snapshot_whiteouts(trans, btree, old_pos, new_pos); 96 } 97 98 int bch2_trans_update_extent_overwrite(struct btree_trans *, struct btree_iter *, 99 enum btree_iter_update_trigger_flags, 100 struct bkey_s_c, struct bkey_s_c); 101 102 int bch2_bkey_get_empty_slot(struct btree_trans *, struct btree_iter *, 103 enum btree_id, struct bpos); 104 105 int __must_check bch2_trans_update(struct btree_trans *, struct btree_iter *, 106 struct bkey_i *, enum btree_iter_update_trigger_flags); 107 108 struct jset_entry *__bch2_trans_jset_entry_alloc(struct btree_trans *, unsigned); 109 110 static inline struct jset_entry *btree_trans_journal_entries_top(struct btree_trans *trans) 111 { 112 return (void *) ((u64 *) trans->journal_entries + trans->journal_entries_u64s); 113 } 114 115 static inline struct jset_entry * 116 bch2_trans_jset_entry_alloc(struct btree_trans *trans, unsigned u64s) 117 { 118 if (!trans->journal_entries || 119 trans->journal_entries_u64s + u64s > trans->journal_entries_size) 120 return __bch2_trans_jset_entry_alloc(trans, u64s); 121 122 struct jset_entry *e = btree_trans_journal_entries_top(trans); 123 trans->journal_entries_u64s += u64s; 124 return e; 125 } 126 127 int bch2_btree_insert_clone_trans(struct btree_trans *, enum btree_id, struct bkey_i *); 128 129 int bch2_btree_write_buffer_insert_err(struct btree_trans *, 130 enum btree_id, struct bkey_i *); 131 132 static inline int __must_check bch2_trans_update_buffered(struct btree_trans *trans, 133 enum btree_id btree, 134 struct bkey_i *k) 135 { 136 kmsan_check_memory(k, bkey_bytes(&k->k)); 137 138 if (unlikely(!btree_type_uses_write_buffer(btree))) { 139 int ret = bch2_btree_write_buffer_insert_err(trans, btree, k); 140 dump_stack(); 141 return ret; 142 } 143 /* 144 * Most updates skip the btree write buffer until journal replay is 145 * finished because synchronization with journal replay relies on having 146 * a btree node locked - if we're overwriting a key in the journal that 147 * journal replay hasn't yet replayed, we have to mark it as 148 * overwritten. 149 * 150 * But accounting updates don't overwrite, they're deltas, and they have 151 * to be flushed to the btree strictly in order for journal replay to be 152 * able to tell which updates need to be applied: 153 */ 154 if (k->k.type != KEY_TYPE_accounting && 155 unlikely(trans->journal_replay_not_finished)) 156 return bch2_btree_insert_clone_trans(trans, btree, k); 157 158 struct jset_entry *e = bch2_trans_jset_entry_alloc(trans, jset_u64s(k->k.u64s)); 159 int ret = PTR_ERR_OR_ZERO(e); 160 if (ret) 161 return ret; 162 163 journal_entry_init(e, BCH_JSET_ENTRY_write_buffer_keys, btree, 0, k->k.u64s); 164 bkey_copy(e->start, k); 165 return 0; 166 } 167 168 void bch2_trans_commit_hook(struct btree_trans *, 169 struct btree_trans_commit_hook *); 170 int __bch2_trans_commit(struct btree_trans *, unsigned); 171 172 int bch2_trans_log_msg(struct btree_trans *, struct printbuf *); 173 __printf(2, 3) int bch2_fs_log_msg(struct bch_fs *, const char *, ...); 174 __printf(2, 3) int bch2_journal_log_msg(struct bch_fs *, const char *, ...); 175 176 /** 177 * bch2_trans_commit - insert keys at given iterator positions 178 * 179 * This is main entry point for btree updates. 180 * 181 * Return values: 182 * -EROFS: filesystem read only 183 * -EIO: journal or btree node IO error 184 */ 185 static inline int bch2_trans_commit(struct btree_trans *trans, 186 struct disk_reservation *disk_res, 187 u64 *journal_seq, 188 unsigned flags) 189 { 190 trans->disk_res = disk_res; 191 trans->journal_seq = journal_seq; 192 193 return __bch2_trans_commit(trans, flags); 194 } 195 196 #define commit_do(_trans, _disk_res, _journal_seq, _flags, _do) \ 197 lockrestart_do(_trans, _do ?: bch2_trans_commit(_trans, (_disk_res),\ 198 (_journal_seq), (_flags))) 199 200 #define nested_commit_do(_trans, _disk_res, _journal_seq, _flags, _do) \ 201 nested_lockrestart_do(_trans, _do ?: bch2_trans_commit(_trans, (_disk_res),\ 202 (_journal_seq), (_flags))) 203 204 #define bch2_trans_commit_do(_c, _disk_res, _journal_seq, _flags, _do) \ 205 bch2_trans_run(_c, commit_do(trans, _disk_res, _journal_seq, _flags, _do)) 206 207 #define trans_for_each_update(_trans, _i) \ 208 for (struct btree_insert_entry *_i = (_trans)->updates; \ 209 (_i) < (_trans)->updates + (_trans)->nr_updates; \ 210 (_i)++) 211 212 static inline void bch2_trans_reset_updates(struct btree_trans *trans) 213 { 214 trans_for_each_update(trans, i) 215 bch2_path_put(trans, i->path, true); 216 217 trans->nr_updates = 0; 218 trans->journal_entries_u64s = 0; 219 trans->hooks = NULL; 220 trans->extra_disk_res = 0; 221 } 222 223 static inline struct bkey_i *__bch2_bkey_make_mut_noupdate(struct btree_trans *trans, struct bkey_s_c k, 224 unsigned type, unsigned min_bytes) 225 { 226 unsigned bytes = max_t(unsigned, min_bytes, bkey_bytes(k.k)); 227 struct bkey_i *mut; 228 229 if (type && k.k->type != type) 230 return ERR_PTR(-ENOENT); 231 232 /* extra padding for varint_decode_fast... */ 233 mut = bch2_trans_kmalloc_nomemzero(trans, bytes + 8); 234 if (!IS_ERR(mut)) { 235 bkey_reassemble(mut, k); 236 237 if (unlikely(bytes > bkey_bytes(k.k))) { 238 memset((void *) mut + bkey_bytes(k.k), 0, 239 bytes - bkey_bytes(k.k)); 240 mut->k.u64s = DIV_ROUND_UP(bytes, sizeof(u64)); 241 } 242 } 243 return mut; 244 } 245 246 static inline struct bkey_i *bch2_bkey_make_mut_noupdate(struct btree_trans *trans, struct bkey_s_c k) 247 { 248 return __bch2_bkey_make_mut_noupdate(trans, k, 0, 0); 249 } 250 251 #define bch2_bkey_make_mut_noupdate_typed(_trans, _k, _type) \ 252 bkey_i_to_##_type(__bch2_bkey_make_mut_noupdate(_trans, _k, \ 253 KEY_TYPE_##_type, sizeof(struct bkey_i_##_type))) 254 255 static inline struct bkey_i *__bch2_bkey_make_mut(struct btree_trans *trans, struct btree_iter *iter, 256 struct bkey_s_c *k, 257 enum btree_iter_update_trigger_flags flags, 258 unsigned type, unsigned min_bytes) 259 { 260 struct bkey_i *mut = __bch2_bkey_make_mut_noupdate(trans, *k, type, min_bytes); 261 int ret; 262 263 if (IS_ERR(mut)) 264 return mut; 265 266 ret = bch2_trans_update(trans, iter, mut, flags); 267 if (ret) 268 return ERR_PTR(ret); 269 270 *k = bkey_i_to_s_c(mut); 271 return mut; 272 } 273 274 static inline struct bkey_i *bch2_bkey_make_mut(struct btree_trans *trans, 275 struct btree_iter *iter, struct bkey_s_c *k, 276 enum btree_iter_update_trigger_flags flags) 277 { 278 return __bch2_bkey_make_mut(trans, iter, k, flags, 0, 0); 279 } 280 281 #define bch2_bkey_make_mut_typed(_trans, _iter, _k, _flags, _type) \ 282 bkey_i_to_##_type(__bch2_bkey_make_mut(_trans, _iter, _k, _flags,\ 283 KEY_TYPE_##_type, sizeof(struct bkey_i_##_type))) 284 285 static inline struct bkey_i *__bch2_bkey_get_mut_noupdate(struct btree_trans *trans, 286 struct btree_iter *iter, 287 unsigned btree_id, struct bpos pos, 288 enum btree_iter_update_trigger_flags flags, 289 unsigned type, unsigned min_bytes) 290 { 291 struct bkey_s_c k = __bch2_bkey_get_iter(trans, iter, 292 btree_id, pos, flags|BTREE_ITER_intent, type); 293 struct bkey_i *ret = IS_ERR(k.k) 294 ? ERR_CAST(k.k) 295 : __bch2_bkey_make_mut_noupdate(trans, k, 0, min_bytes); 296 if (IS_ERR(ret)) 297 bch2_trans_iter_exit(trans, iter); 298 return ret; 299 } 300 301 static inline struct bkey_i *bch2_bkey_get_mut_noupdate(struct btree_trans *trans, 302 struct btree_iter *iter, 303 unsigned btree_id, struct bpos pos, 304 enum btree_iter_update_trigger_flags flags) 305 { 306 return __bch2_bkey_get_mut_noupdate(trans, iter, btree_id, pos, flags, 0, 0); 307 } 308 309 static inline struct bkey_i *__bch2_bkey_get_mut(struct btree_trans *trans, 310 struct btree_iter *iter, 311 unsigned btree_id, struct bpos pos, 312 enum btree_iter_update_trigger_flags flags, 313 unsigned type, unsigned min_bytes) 314 { 315 struct bkey_i *mut = __bch2_bkey_get_mut_noupdate(trans, iter, 316 btree_id, pos, flags|BTREE_ITER_intent, type, min_bytes); 317 int ret; 318 319 if (IS_ERR(mut)) 320 return mut; 321 322 ret = bch2_trans_update(trans, iter, mut, flags); 323 if (ret) { 324 bch2_trans_iter_exit(trans, iter); 325 return ERR_PTR(ret); 326 } 327 328 return mut; 329 } 330 331 static inline struct bkey_i *bch2_bkey_get_mut_minsize(struct btree_trans *trans, 332 struct btree_iter *iter, 333 unsigned btree_id, struct bpos pos, 334 enum btree_iter_update_trigger_flags flags, 335 unsigned min_bytes) 336 { 337 return __bch2_bkey_get_mut(trans, iter, btree_id, pos, flags, 0, min_bytes); 338 } 339 340 static inline struct bkey_i *bch2_bkey_get_mut(struct btree_trans *trans, 341 struct btree_iter *iter, 342 unsigned btree_id, struct bpos pos, 343 enum btree_iter_update_trigger_flags flags) 344 { 345 return __bch2_bkey_get_mut(trans, iter, btree_id, pos, flags, 0, 0); 346 } 347 348 #define bch2_bkey_get_mut_typed(_trans, _iter, _btree_id, _pos, _flags, _type)\ 349 bkey_i_to_##_type(__bch2_bkey_get_mut(_trans, _iter, \ 350 _btree_id, _pos, _flags, \ 351 KEY_TYPE_##_type, sizeof(struct bkey_i_##_type))) 352 353 static inline struct bkey_i *__bch2_bkey_alloc(struct btree_trans *trans, struct btree_iter *iter, 354 enum btree_iter_update_trigger_flags flags, 355 unsigned type, unsigned val_size) 356 { 357 struct bkey_i *k = bch2_trans_kmalloc(trans, sizeof(*k) + val_size); 358 int ret; 359 360 if (IS_ERR(k)) 361 return k; 362 363 bkey_init(&k->k); 364 k->k.p = iter->pos; 365 k->k.type = type; 366 set_bkey_val_bytes(&k->k, val_size); 367 368 ret = bch2_trans_update(trans, iter, k, flags); 369 if (unlikely(ret)) 370 return ERR_PTR(ret); 371 return k; 372 } 373 374 #define bch2_bkey_alloc(_trans, _iter, _flags, _type) \ 375 bkey_i_to_##_type(__bch2_bkey_alloc(_trans, _iter, _flags, \ 376 KEY_TYPE_##_type, sizeof(struct bch_##_type))) 377 378 #endif /* _BCACHEFS_BTREE_UPDATE_H */ 379