1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 2
3 #include "bcachefs.h" 3 #include "bcachefs.h"
4 #include "bkey_methods.h" 4 #include "bkey_methods.h"
5 #include "bkey_buf.h" 5 #include "bkey_buf.h"
6 #include "btree_cache.h" 6 #include "btree_cache.h"
7 #include "btree_iter.h" 7 #include "btree_iter.h"
8 #include "btree_journal_iter.h" 8 #include "btree_journal_iter.h"
9 #include "btree_key_cache.h" 9 #include "btree_key_cache.h"
10 #include "btree_locking.h" 10 #include "btree_locking.h"
11 #include "btree_update.h" 11 #include "btree_update.h"
12 #include "debug.h" 12 #include "debug.h"
13 #include "error.h" 13 #include "error.h"
14 #include "extents.h" 14 #include "extents.h"
15 #include "journal.h" 15 #include "journal.h"
16 #include "journal_io.h" 16 #include "journal_io.h"
17 #include "replicas.h" 17 #include "replicas.h"
18 #include "snapshot.h" 18 #include "snapshot.h"
19 #include "trace.h" 19 #include "trace.h"
20 20
21 #include <linux/random.h> 21 #include <linux/random.h>
22 #include <linux/prefetch.h> 22 #include <linux/prefetch.h>
23 23
24 static inline void btree_path_list_remove(stru 24 static inline void btree_path_list_remove(struct btree_trans *, struct btree_path *);
25 static inline void btree_path_list_add(struct 25 static inline void btree_path_list_add(struct btree_trans *,
26 btree_path_idx_t, btre 26 btree_path_idx_t, btree_path_idx_t);
27 27
28 static inline unsigned long btree_iter_ip_allo 28 static inline unsigned long btree_iter_ip_allocated(struct btree_iter *iter)
29 { 29 {
30 #ifdef TRACK_PATH_ALLOCATED 30 #ifdef TRACK_PATH_ALLOCATED
31 return iter->ip_allocated; 31 return iter->ip_allocated;
32 #else 32 #else
33 return 0; 33 return 0;
34 #endif 34 #endif
35 } 35 }
36 36
37 static btree_path_idx_t btree_path_alloc(struc 37 static btree_path_idx_t btree_path_alloc(struct btree_trans *, btree_path_idx_t);
38 static void bch2_trans_srcu_lock(struct btree_ 38 static void bch2_trans_srcu_lock(struct btree_trans *);
39 39
40 static inline int __btree_path_cmp(const struc 40 static inline int __btree_path_cmp(const struct btree_path *l,
41 enum btree_ 41 enum btree_id r_btree_id,
42 bool 42 bool r_cached,
43 struct bpos 43 struct bpos r_pos,
44 unsigned 44 unsigned r_level)
45 { 45 {
46 /* 46 /*
47 * Must match lock ordering as defined 47 * Must match lock ordering as defined by __bch2_btree_node_lock:
48 */ 48 */
49 return cmp_int(l->btree_id, r_btre 49 return cmp_int(l->btree_id, r_btree_id) ?:
50 cmp_int((int) l->cached, 50 cmp_int((int) l->cached, (int) r_cached) ?:
51 bpos_cmp(l->pos, r_pos) 51 bpos_cmp(l->pos, r_pos) ?:
52 -cmp_int(l->level, r_leve 52 -cmp_int(l->level, r_level);
53 } 53 }
54 54
55 static inline int btree_path_cmp(const struct 55 static inline int btree_path_cmp(const struct btree_path *l,
56 const struct 56 const struct btree_path *r)
57 { 57 {
58 return __btree_path_cmp(l, r->btree_id 58 return __btree_path_cmp(l, r->btree_id, r->cached, r->pos, r->level);
59 } 59 }
60 60
61 static inline struct bpos bkey_successor(struc 61 static inline struct bpos bkey_successor(struct btree_iter *iter, struct bpos p)
62 { 62 {
63 /* Are we iterating over keys in all s 63 /* Are we iterating over keys in all snapshots? */
64 if (iter->flags & BTREE_ITER_all_snaps 64 if (iter->flags & BTREE_ITER_all_snapshots) {
65 p = bpos_successor(p); 65 p = bpos_successor(p);
66 } else { 66 } else {
67 p = bpos_nosnap_successor(p); 67 p = bpos_nosnap_successor(p);
68 p.snapshot = iter->snapshot; 68 p.snapshot = iter->snapshot;
69 } 69 }
70 70
71 return p; 71 return p;
72 } 72 }
73 73
74 static inline struct bpos bkey_predecessor(str 74 static inline struct bpos bkey_predecessor(struct btree_iter *iter, struct bpos p)
75 { 75 {
76 /* Are we iterating over keys in all s 76 /* Are we iterating over keys in all snapshots? */
77 if (iter->flags & BTREE_ITER_all_snaps 77 if (iter->flags & BTREE_ITER_all_snapshots) {
78 p = bpos_predecessor(p); 78 p = bpos_predecessor(p);
79 } else { 79 } else {
80 p = bpos_nosnap_predecessor(p) 80 p = bpos_nosnap_predecessor(p);
81 p.snapshot = iter->snapshot; 81 p.snapshot = iter->snapshot;
82 } 82 }
83 83
84 return p; 84 return p;
85 } 85 }
86 86
87 static inline struct bpos btree_iter_search_ke 87 static inline struct bpos btree_iter_search_key(struct btree_iter *iter)
88 { 88 {
89 struct bpos pos = iter->pos; 89 struct bpos pos = iter->pos;
90 90
91 if ((iter->flags & BTREE_ITER_is_exten 91 if ((iter->flags & BTREE_ITER_is_extents) &&
92 !bkey_eq(pos, POS_MAX)) 92 !bkey_eq(pos, POS_MAX))
93 pos = bkey_successor(iter, pos 93 pos = bkey_successor(iter, pos);
94 return pos; 94 return pos;
95 } 95 }
96 96
97 static inline bool btree_path_pos_before_node( 97 static inline bool btree_path_pos_before_node(struct btree_path *path,
98 98 struct btree *b)
99 { 99 {
100 return bpos_lt(path->pos, b->data->min 100 return bpos_lt(path->pos, b->data->min_key);
101 } 101 }
102 102
103 static inline bool btree_path_pos_after_node(s 103 static inline bool btree_path_pos_after_node(struct btree_path *path,
104 s 104 struct btree *b)
105 { 105 {
106 return bpos_gt(path->pos, b->key.k.p); 106 return bpos_gt(path->pos, b->key.k.p);
107 } 107 }
108 108
109 static inline bool btree_path_pos_in_node(stru 109 static inline bool btree_path_pos_in_node(struct btree_path *path,
110 stru 110 struct btree *b)
111 { 111 {
112 return path->btree_id == b->c.btree_id 112 return path->btree_id == b->c.btree_id &&
113 !btree_path_pos_before_node(pa 113 !btree_path_pos_before_node(path, b) &&
114 !btree_path_pos_after_node(pat 114 !btree_path_pos_after_node(path, b);
115 } 115 }
116 116
117 /* Btree iterator: */ 117 /* Btree iterator: */
118 118
119 #ifdef CONFIG_BCACHEFS_DEBUG 119 #ifdef CONFIG_BCACHEFS_DEBUG
120 120
121 static void bch2_btree_path_verify_cached(stru 121 static void bch2_btree_path_verify_cached(struct btree_trans *trans,
122 stru 122 struct btree_path *path)
123 { 123 {
124 struct bkey_cached *ck; 124 struct bkey_cached *ck;
125 bool locked = btree_node_locked(path, 125 bool locked = btree_node_locked(path, 0);
126 126
127 if (!bch2_btree_node_relock(trans, pat 127 if (!bch2_btree_node_relock(trans, path, 0))
128 return; 128 return;
129 129
130 ck = (void *) path->l[0].b; 130 ck = (void *) path->l[0].b;
131 BUG_ON(ck->key.btree_id != path->btree 131 BUG_ON(ck->key.btree_id != path->btree_id ||
132 !bkey_eq(ck->key.pos, path->pos 132 !bkey_eq(ck->key.pos, path->pos));
133 133
134 if (!locked) 134 if (!locked)
135 btree_node_unlock(trans, path, 135 btree_node_unlock(trans, path, 0);
136 } 136 }
137 137
138 static void bch2_btree_path_verify_level(struc 138 static void bch2_btree_path_verify_level(struct btree_trans *trans,
139 struct btree_p 139 struct btree_path *path, unsigned level)
140 { 140 {
141 struct btree_path_level *l; 141 struct btree_path_level *l;
142 struct btree_node_iter tmp; 142 struct btree_node_iter tmp;
143 bool locked; 143 bool locked;
144 struct bkey_packed *p, *k; 144 struct bkey_packed *p, *k;
145 struct printbuf buf1 = PRINTBUF; 145 struct printbuf buf1 = PRINTBUF;
146 struct printbuf buf2 = PRINTBUF; 146 struct printbuf buf2 = PRINTBUF;
147 struct printbuf buf3 = PRINTBUF; 147 struct printbuf buf3 = PRINTBUF;
148 const char *msg; 148 const char *msg;
149 149
150 if (!bch2_debug_check_iterators) 150 if (!bch2_debug_check_iterators)
151 return; 151 return;
152 152
153 l = &path->l[level]; 153 l = &path->l[level];
154 tmp = l->iter; 154 tmp = l->iter;
155 locked = btree_node_locked(path, leve 155 locked = btree_node_locked(path, level);
156 156
157 if (path->cached) { 157 if (path->cached) {
158 if (!level) 158 if (!level)
159 bch2_btree_path_verify 159 bch2_btree_path_verify_cached(trans, path);
160 return; 160 return;
161 } 161 }
162 162
163 if (!btree_path_node(path, level)) 163 if (!btree_path_node(path, level))
164 return; 164 return;
165 165
166 if (!bch2_btree_node_relock_notrace(tr 166 if (!bch2_btree_node_relock_notrace(trans, path, level))
167 return; 167 return;
168 168
169 BUG_ON(!btree_path_pos_in_node(path, l 169 BUG_ON(!btree_path_pos_in_node(path, l->b));
170 170
171 bch2_btree_node_iter_verify(&l->iter, 171 bch2_btree_node_iter_verify(&l->iter, l->b);
172 172
173 /* 173 /*
174 * For interior nodes, the iterator wi 174 * For interior nodes, the iterator will have skipped past deleted keys:
175 */ 175 */
176 p = level 176 p = level
177 ? bch2_btree_node_iter_prev(&t 177 ? bch2_btree_node_iter_prev(&tmp, l->b)
178 : bch2_btree_node_iter_prev_al 178 : bch2_btree_node_iter_prev_all(&tmp, l->b);
179 k = bch2_btree_node_iter_peek_all(&l-> 179 k = bch2_btree_node_iter_peek_all(&l->iter, l->b);
180 180
181 if (p && bkey_iter_pos_cmp(l->b, p, &p 181 if (p && bkey_iter_pos_cmp(l->b, p, &path->pos) >= 0) {
182 msg = "before"; 182 msg = "before";
183 goto err; 183 goto err;
184 } 184 }
185 185
186 if (k && bkey_iter_pos_cmp(l->b, k, &p 186 if (k && bkey_iter_pos_cmp(l->b, k, &path->pos) < 0) {
187 msg = "after"; 187 msg = "after";
188 goto err; 188 goto err;
189 } 189 }
190 190
191 if (!locked) 191 if (!locked)
192 btree_node_unlock(trans, path, 192 btree_node_unlock(trans, path, level);
193 return; 193 return;
194 err: 194 err:
195 bch2_bpos_to_text(&buf1, path->pos); 195 bch2_bpos_to_text(&buf1, path->pos);
196 196
197 if (p) { 197 if (p) {
198 struct bkey uk = bkey_unpack_k 198 struct bkey uk = bkey_unpack_key(l->b, p);
199 199
200 bch2_bkey_to_text(&buf2, &uk); 200 bch2_bkey_to_text(&buf2, &uk);
201 } else { 201 } else {
202 prt_printf(&buf2, "(none)"); 202 prt_printf(&buf2, "(none)");
203 } 203 }
204 204
205 if (k) { 205 if (k) {
206 struct bkey uk = bkey_unpack_k 206 struct bkey uk = bkey_unpack_key(l->b, k);
207 207
208 bch2_bkey_to_text(&buf3, &uk); 208 bch2_bkey_to_text(&buf3, &uk);
209 } else { 209 } else {
210 prt_printf(&buf3, "(none)"); 210 prt_printf(&buf3, "(none)");
211 } 211 }
212 212
213 panic("path should be %s key at level 213 panic("path should be %s key at level %u:\n"
214 "path pos %s\n" 214 "path pos %s\n"
215 "prev key %s\n" 215 "prev key %s\n"
216 "cur key %s\n", 216 "cur key %s\n",
217 msg, level, buf1.buf, buf2.buf, 217 msg, level, buf1.buf, buf2.buf, buf3.buf);
218 } 218 }
219 219
220 static void bch2_btree_path_verify(struct btre 220 static void bch2_btree_path_verify(struct btree_trans *trans,
221 struct btre 221 struct btree_path *path)
222 { 222 {
223 struct bch_fs *c = trans->c; 223 struct bch_fs *c = trans->c;
224 224
225 for (unsigned i = 0; i < (!path->cache 225 for (unsigned i = 0; i < (!path->cached ? BTREE_MAX_DEPTH : 1); i++) {
226 if (!path->l[i].b) { 226 if (!path->l[i].b) {
227 BUG_ON(!path->cached & 227 BUG_ON(!path->cached &&
228 bch2_btree_id_r 228 bch2_btree_id_root(c, path->btree_id)->b->c.level > i);
229 break; 229 break;
230 } 230 }
231 231
232 bch2_btree_path_verify_level(t 232 bch2_btree_path_verify_level(trans, path, i);
233 } 233 }
234 234
235 bch2_btree_path_verify_locks(path); 235 bch2_btree_path_verify_locks(path);
236 } 236 }
237 237
238 void bch2_trans_verify_paths(struct btree_tran 238 void bch2_trans_verify_paths(struct btree_trans *trans)
239 { 239 {
240 struct btree_path *path; 240 struct btree_path *path;
241 unsigned iter; 241 unsigned iter;
242 242
243 trans_for_each_path(trans, path, iter) 243 trans_for_each_path(trans, path, iter)
244 bch2_btree_path_verify(trans, 244 bch2_btree_path_verify(trans, path);
245 } 245 }
246 246
247 static void bch2_btree_iter_verify(struct btre 247 static void bch2_btree_iter_verify(struct btree_iter *iter)
248 { 248 {
249 struct btree_trans *trans = iter->tran 249 struct btree_trans *trans = iter->trans;
250 250
251 BUG_ON(!!(iter->flags & BTREE_ITER_cac 251 BUG_ON(!!(iter->flags & BTREE_ITER_cached) != btree_iter_path(trans, iter)->cached);
252 252
253 BUG_ON((iter->flags & BTREE_ITER_is_ex 253 BUG_ON((iter->flags & BTREE_ITER_is_extents) &&
254 (iter->flags & BTREE_ITER_all_s 254 (iter->flags & BTREE_ITER_all_snapshots));
255 255
256 BUG_ON(!(iter->flags & BTREE_ITER_snap 256 BUG_ON(!(iter->flags & BTREE_ITER_snapshot_field) &&
257 (iter->flags & BTREE_ITER_all_s 257 (iter->flags & BTREE_ITER_all_snapshots) &&
258 !btree_type_has_snapshot_field( 258 !btree_type_has_snapshot_field(iter->btree_id));
259 259
260 if (iter->update_path) 260 if (iter->update_path)
261 bch2_btree_path_verify(trans, 261 bch2_btree_path_verify(trans, &trans->paths[iter->update_path]);
262 bch2_btree_path_verify(trans, btree_it 262 bch2_btree_path_verify(trans, btree_iter_path(trans, iter));
263 } 263 }
264 264
265 static void bch2_btree_iter_verify_entry_exit( 265 static void bch2_btree_iter_verify_entry_exit(struct btree_iter *iter)
266 { 266 {
267 BUG_ON((iter->flags & BTREE_ITER_filte 267 BUG_ON((iter->flags & BTREE_ITER_filter_snapshots) &&
268 !iter->pos.snapshot); 268 !iter->pos.snapshot);
269 269
270 BUG_ON(!(iter->flags & BTREE_ITER_all_ 270 BUG_ON(!(iter->flags & BTREE_ITER_all_snapshots) &&
271 iter->pos.snapshot != iter->sna 271 iter->pos.snapshot != iter->snapshot);
272 272
273 BUG_ON(bkey_lt(iter->pos, bkey_start_p 273 BUG_ON(bkey_lt(iter->pos, bkey_start_pos(&iter->k)) ||
274 bkey_gt(iter->pos, iter->k.p)); 274 bkey_gt(iter->pos, iter->k.p));
275 } 275 }
276 276
277 static int bch2_btree_iter_verify_ret(struct b 277 static int bch2_btree_iter_verify_ret(struct btree_iter *iter, struct bkey_s_c k)
278 { 278 {
279 struct btree_trans *trans = iter->tran 279 struct btree_trans *trans = iter->trans;
280 struct btree_iter copy; 280 struct btree_iter copy;
281 struct bkey_s_c prev; 281 struct bkey_s_c prev;
282 int ret = 0; 282 int ret = 0;
283 283
284 if (!bch2_debug_check_iterators) 284 if (!bch2_debug_check_iterators)
285 return 0; 285 return 0;
286 286
287 if (!(iter->flags & BTREE_ITER_filter_ 287 if (!(iter->flags & BTREE_ITER_filter_snapshots))
288 return 0; 288 return 0;
289 289
290 if (bkey_err(k) || !k.k) 290 if (bkey_err(k) || !k.k)
291 return 0; 291 return 0;
292 292
293 BUG_ON(!bch2_snapshot_is_ancestor(tran 293 BUG_ON(!bch2_snapshot_is_ancestor(trans->c,
294 iter 294 iter->snapshot,
295 k.k- 295 k.k->p.snapshot));
296 296
297 bch2_trans_iter_init(trans, ©, ite 297 bch2_trans_iter_init(trans, ©, iter->btree_id, iter->pos,
298 BTREE_ITER_nopres 298 BTREE_ITER_nopreserve|
299 BTREE_ITER_all_sn 299 BTREE_ITER_all_snapshots);
300 prev = bch2_btree_iter_prev(©); 300 prev = bch2_btree_iter_prev(©);
301 if (!prev.k) 301 if (!prev.k)
302 goto out; 302 goto out;
303 303
304 ret = bkey_err(prev); 304 ret = bkey_err(prev);
305 if (ret) 305 if (ret)
306 goto out; 306 goto out;
307 307
308 if (bkey_eq(prev.k->p, k.k->p) && 308 if (bkey_eq(prev.k->p, k.k->p) &&
309 bch2_snapshot_is_ancestor(trans->c 309 bch2_snapshot_is_ancestor(trans->c, iter->snapshot,
310 prev.k-> 310 prev.k->p.snapshot) > 0) {
311 struct printbuf buf1 = PRINTBU 311 struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
312 312
313 bch2_bkey_to_text(&buf1, k.k); 313 bch2_bkey_to_text(&buf1, k.k);
314 bch2_bkey_to_text(&buf2, prev. 314 bch2_bkey_to_text(&buf2, prev.k);
315 315
316 panic("iter snap %u\n" 316 panic("iter snap %u\n"
317 "k %s\n" 317 "k %s\n"
318 "prev %s\n", 318 "prev %s\n",
319 iter->snapshot, 319 iter->snapshot,
320 buf1.buf, buf2.buf); 320 buf1.buf, buf2.buf);
321 } 321 }
322 out: 322 out:
323 bch2_trans_iter_exit(trans, ©); 323 bch2_trans_iter_exit(trans, ©);
324 return ret; 324 return ret;
325 } 325 }
326 326
327 void bch2_assert_pos_locked(struct btree_trans 327 void bch2_assert_pos_locked(struct btree_trans *trans, enum btree_id id,
328 struct bpos pos) !! 328 struct bpos pos, bool key_cache)
329 { 329 {
330 bch2_trans_verify_not_unlocked(trans); 330 bch2_trans_verify_not_unlocked(trans);
331 331
332 struct btree_path *path; 332 struct btree_path *path;
333 struct trans_for_each_path_inorder_ite 333 struct trans_for_each_path_inorder_iter iter;
334 struct printbuf buf = PRINTBUF; 334 struct printbuf buf = PRINTBUF;
335 335
336 btree_trans_sort_paths(trans); 336 btree_trans_sort_paths(trans);
337 337
338 trans_for_each_path_inorder(trans, pat 338 trans_for_each_path_inorder(trans, path, iter) {
339 if (path->btree_id != id || !! 339 int cmp = cmp_int(path->btree_id, id) ?:
340 !btree_node_locked(path, 0 !! 340 cmp_int(path->cached, key_cache);
>> 341
>> 342 if (cmp > 0)
>> 343 break;
>> 344 if (cmp < 0)
>> 345 continue;
>> 346
>> 347 if (!btree_node_locked(path, 0) ||
341 !path->should_be_locked) 348 !path->should_be_locked)
342 continue; 349 continue;
343 350
344 if (!path->cached) { !! 351 if (!key_cache) {
345 if (bkey_ge(pos, path- 352 if (bkey_ge(pos, path->l[0].b->data->min_key) &&
346 bkey_le(pos, path- 353 bkey_le(pos, path->l[0].b->key.k.p))
347 return; 354 return;
348 } else { 355 } else {
349 if (bkey_eq(pos, path- 356 if (bkey_eq(pos, path->pos))
350 return; 357 return;
351 } 358 }
352 } 359 }
353 360
354 bch2_dump_trans_paths_updates(trans); 361 bch2_dump_trans_paths_updates(trans);
355 bch2_bpos_to_text(&buf, pos); 362 bch2_bpos_to_text(&buf, pos);
356 363
357 panic("not locked: %s %s\n", bch2_btre !! 364 panic("not locked: %s %s%s\n",
>> 365 bch2_btree_id_str(id), buf.buf,
>> 366 key_cache ? " cached" : "");
358 } 367 }
359 368
360 #else 369 #else
361 370
362 static inline void bch2_btree_path_verify_leve 371 static inline void bch2_btree_path_verify_level(struct btree_trans *trans,
363 372 struct btree_path *path, unsigned l) {}
364 static inline void bch2_btree_path_verify(stru 373 static inline void bch2_btree_path_verify(struct btree_trans *trans,
365 stru 374 struct btree_path *path) {}
366 static inline void bch2_btree_iter_verify(stru 375 static inline void bch2_btree_iter_verify(struct btree_iter *iter) {}
367 static inline void bch2_btree_iter_verify_entr 376 static inline void bch2_btree_iter_verify_entry_exit(struct btree_iter *iter) {}
368 static inline int bch2_btree_iter_verify_ret(s 377 static inline int bch2_btree_iter_verify_ret(struct btree_iter *iter, struct bkey_s_c k) { return 0; }
369 378
370 #endif 379 #endif
371 380
372 /* Btree path: fixups after btree updates */ 381 /* Btree path: fixups after btree updates */
373 382
374 static void btree_node_iter_set_set_pos(struct 383 static void btree_node_iter_set_set_pos(struct btree_node_iter *iter,
375 struct 384 struct btree *b,
376 struct 385 struct bset_tree *t,
377 struct 386 struct bkey_packed *k)
378 { 387 {
379 struct btree_node_iter_set *set; 388 struct btree_node_iter_set *set;
380 389
381 btree_node_iter_for_each(iter, set) 390 btree_node_iter_for_each(iter, set)
382 if (set->end == t->end_offset) 391 if (set->end == t->end_offset) {
383 set->k = __btree_node_ 392 set->k = __btree_node_key_to_offset(b, k);
384 bch2_btree_node_iter_s 393 bch2_btree_node_iter_sort(iter, b);
385 return; 394 return;
386 } 395 }
387 396
388 bch2_btree_node_iter_push(iter, b, k, 397 bch2_btree_node_iter_push(iter, b, k, btree_bkey_last(b, t));
389 } 398 }
390 399
391 static void __bch2_btree_path_fix_key_modified 400 static void __bch2_btree_path_fix_key_modified(struct btree_path *path,
392 401 struct btree *b,
393 402 struct bkey_packed *where)
394 { 403 {
395 struct btree_path_level *l = &path->l[ 404 struct btree_path_level *l = &path->l[b->c.level];
396 405
397 if (where != bch2_btree_node_iter_peek 406 if (where != bch2_btree_node_iter_peek_all(&l->iter, l->b))
398 return; 407 return;
399 408
400 if (bkey_iter_pos_cmp(l->b, where, &pa 409 if (bkey_iter_pos_cmp(l->b, where, &path->pos) < 0)
401 bch2_btree_node_iter_advance(& 410 bch2_btree_node_iter_advance(&l->iter, l->b);
402 } 411 }
403 412
404 void bch2_btree_path_fix_key_modified(struct b 413 void bch2_btree_path_fix_key_modified(struct btree_trans *trans,
405 struct b 414 struct btree *b,
406 struct b 415 struct bkey_packed *where)
407 { 416 {
408 struct btree_path *path; 417 struct btree_path *path;
409 unsigned i; 418 unsigned i;
410 419
411 trans_for_each_path_with_node(trans, b 420 trans_for_each_path_with_node(trans, b, path, i) {
412 __bch2_btree_path_fix_key_modi 421 __bch2_btree_path_fix_key_modified(path, b, where);
413 bch2_btree_path_verify_level(t 422 bch2_btree_path_verify_level(trans, path, b->c.level);
414 } 423 }
415 } 424 }
416 425
417 static void __bch2_btree_node_iter_fix(struct 426 static void __bch2_btree_node_iter_fix(struct btree_path *path,
418 struct 427 struct btree *b,
419 struct 428 struct btree_node_iter *node_iter,
420 struct 429 struct bset_tree *t,
421 struct 430 struct bkey_packed *where,
422 unsigne 431 unsigned clobber_u64s,
423 unsigne 432 unsigned new_u64s)
424 { 433 {
425 const struct bkey_packed *end = btree_ 434 const struct bkey_packed *end = btree_bkey_last(b, t);
426 struct btree_node_iter_set *set; 435 struct btree_node_iter_set *set;
427 unsigned offset = __btree_node_key_to_ 436 unsigned offset = __btree_node_key_to_offset(b, where);
428 int shift = new_u64s - clobber_u64s; 437 int shift = new_u64s - clobber_u64s;
429 unsigned old_end = t->end_offset - shi 438 unsigned old_end = t->end_offset - shift;
430 unsigned orig_iter_pos = node_iter->da 439 unsigned orig_iter_pos = node_iter->data[0].k;
431 bool iter_current_key_modified = 440 bool iter_current_key_modified =
432 orig_iter_pos >= offset && 441 orig_iter_pos >= offset &&
433 orig_iter_pos <= offset + clob 442 orig_iter_pos <= offset + clobber_u64s;
434 443
435 btree_node_iter_for_each(node_iter, se 444 btree_node_iter_for_each(node_iter, set)
436 if (set->end == old_end) 445 if (set->end == old_end)
437 goto found; 446 goto found;
438 447
439 /* didn't find the bset in the iterato 448 /* didn't find the bset in the iterator - might have to readd it: */
440 if (new_u64s && 449 if (new_u64s &&
441 bkey_iter_pos_cmp(b, where, &path- 450 bkey_iter_pos_cmp(b, where, &path->pos) >= 0) {
442 bch2_btree_node_iter_push(node 451 bch2_btree_node_iter_push(node_iter, b, where, end);
443 goto fixup_done; 452 goto fixup_done;
444 } else { 453 } else {
445 /* Iterator is after key that 454 /* Iterator is after key that changed */
446 return; 455 return;
447 } 456 }
448 found: 457 found:
449 set->end = t->end_offset; 458 set->end = t->end_offset;
450 459
451 /* Iterator hasn't gotten to the key t 460 /* Iterator hasn't gotten to the key that changed yet: */
452 if (set->k < offset) 461 if (set->k < offset)
453 return; 462 return;
454 463
455 if (new_u64s && 464 if (new_u64s &&
456 bkey_iter_pos_cmp(b, where, &path- 465 bkey_iter_pos_cmp(b, where, &path->pos) >= 0) {
457 set->k = offset; 466 set->k = offset;
458 } else if (set->k < offset + clobber_u 467 } else if (set->k < offset + clobber_u64s) {
459 set->k = offset + new_u64s; 468 set->k = offset + new_u64s;
460 if (set->k == set->end) 469 if (set->k == set->end)
461 bch2_btree_node_iter_s 470 bch2_btree_node_iter_set_drop(node_iter, set);
462 } else { 471 } else {
463 /* Iterator is after key that 472 /* Iterator is after key that changed */
464 set->k = (int) set->k + shift; 473 set->k = (int) set->k + shift;
465 return; 474 return;
466 } 475 }
467 476
468 bch2_btree_node_iter_sort(node_iter, b 477 bch2_btree_node_iter_sort(node_iter, b);
469 fixup_done: 478 fixup_done:
470 if (node_iter->data[0].k != orig_iter_ 479 if (node_iter->data[0].k != orig_iter_pos)
471 iter_current_key_modified = tr 480 iter_current_key_modified = true;
472 481
473 /* 482 /*
474 * When a new key is added, and the no 483 * When a new key is added, and the node iterator now points to that
475 * key, the iterator might have skippe 484 * key, the iterator might have skipped past deleted keys that should
476 * come after the key the iterator now 485 * come after the key the iterator now points to. We have to rewind to
477 * before those deleted keys - otherwi 486 * before those deleted keys - otherwise
478 * bch2_btree_node_iter_prev_all() bre 487 * bch2_btree_node_iter_prev_all() breaks:
479 */ 488 */
480 if (!bch2_btree_node_iter_end(node_ite 489 if (!bch2_btree_node_iter_end(node_iter) &&
481 iter_current_key_modified && 490 iter_current_key_modified &&
482 b->c.level) { 491 b->c.level) {
483 struct bkey_packed *k, *k2, *p 492 struct bkey_packed *k, *k2, *p;
484 493
485 k = bch2_btree_node_iter_peek_ 494 k = bch2_btree_node_iter_peek_all(node_iter, b);
486 495
487 for_each_bset(b, t) { 496 for_each_bset(b, t) {
488 bool set_pos = false; 497 bool set_pos = false;
489 498
490 if (node_iter->data[0] 499 if (node_iter->data[0].end == t->end_offset)
491 continue; 500 continue;
492 501
493 k2 = bch2_btree_node_i 502 k2 = bch2_btree_node_iter_bset_pos(node_iter, b, t);
494 503
495 while ((p = bch2_bkey_ 504 while ((p = bch2_bkey_prev_all(b, t, k2)) &&
496 bkey_iter_cmp(b 505 bkey_iter_cmp(b, k, p) < 0) {
497 k2 = p; 506 k2 = p;
498 set_pos = true 507 set_pos = true;
499 } 508 }
500 509
501 if (set_pos) 510 if (set_pos)
502 btree_node_ite 511 btree_node_iter_set_set_pos(node_iter,
503 512 b, t, k2);
504 } 513 }
505 } 514 }
506 } 515 }
507 516
508 void bch2_btree_node_iter_fix(struct btree_tra 517 void bch2_btree_node_iter_fix(struct btree_trans *trans,
509 struct btree_pat 518 struct btree_path *path,
510 struct btree *b, 519 struct btree *b,
511 struct btree_nod 520 struct btree_node_iter *node_iter,
512 struct bkey_pack 521 struct bkey_packed *where,
513 unsigned clobber 522 unsigned clobber_u64s,
514 unsigned new_u64 523 unsigned new_u64s)
515 { 524 {
516 struct bset_tree *t = bch2_bkey_to_bse 525 struct bset_tree *t = bch2_bkey_to_bset_inlined(b, where);
517 struct btree_path *linked; 526 struct btree_path *linked;
518 unsigned i; 527 unsigned i;
519 528
520 if (node_iter != &path->l[b->c.level]. 529 if (node_iter != &path->l[b->c.level].iter) {
521 __bch2_btree_node_iter_fix(pat 530 __bch2_btree_node_iter_fix(path, b, node_iter, t,
522 whe 531 where, clobber_u64s, new_u64s);
523 532
524 if (bch2_debug_check_iterators 533 if (bch2_debug_check_iterators)
525 bch2_btree_node_iter_v 534 bch2_btree_node_iter_verify(node_iter, b);
526 } 535 }
527 536
528 trans_for_each_path_with_node(trans, b 537 trans_for_each_path_with_node(trans, b, linked, i) {
529 __bch2_btree_node_iter_fix(lin 538 __bch2_btree_node_iter_fix(linked, b,
530 &li 539 &linked->l[b->c.level].iter, t,
531 whe 540 where, clobber_u64s, new_u64s);
532 bch2_btree_path_verify_level(t 541 bch2_btree_path_verify_level(trans, linked, b->c.level);
533 } 542 }
534 } 543 }
535 544
536 /* Btree path level: pointer to a particular b 545 /* Btree path level: pointer to a particular btree node and node iter */
537 546
538 static inline struct bkey_s_c __btree_iter_unp 547 static inline struct bkey_s_c __btree_iter_unpack(struct bch_fs *c,
539 548 struct btree_path_level *l,
540 549 struct bkey *u,
541 550 struct bkey_packed *k)
542 { 551 {
543 if (unlikely(!k)) { 552 if (unlikely(!k)) {
544 /* 553 /*
545 * signal to bch2_btree_iter_p 554 * signal to bch2_btree_iter_peek_slot() that we're currently at
546 * a hole 555 * a hole
547 */ 556 */
548 u->type = KEY_TYPE_deleted; 557 u->type = KEY_TYPE_deleted;
549 return bkey_s_c_null; 558 return bkey_s_c_null;
550 } 559 }
551 560
552 return bkey_disassemble(l->b, k, u); 561 return bkey_disassemble(l->b, k, u);
553 } 562 }
554 563
555 static inline struct bkey_s_c btree_path_level 564 static inline struct bkey_s_c btree_path_level_peek_all(struct bch_fs *c,
556 565 struct btree_path_level *l,
557 566 struct bkey *u)
558 { 567 {
559 return __btree_iter_unpack(c, l, u, 568 return __btree_iter_unpack(c, l, u,
560 bch2_btree_node_iter_p 569 bch2_btree_node_iter_peek_all(&l->iter, l->b));
561 } 570 }
562 571
563 static inline struct bkey_s_c btree_path_level 572 static inline struct bkey_s_c btree_path_level_peek(struct btree_trans *trans,
564 573 struct btree_path *path,
565 574 struct btree_path_level *l,
566 575 struct bkey *u)
567 { 576 {
568 struct bkey_s_c k = __btree_iter_unpac 577 struct bkey_s_c k = __btree_iter_unpack(trans->c, l, u,
569 bch2_btree_node_iter_p 578 bch2_btree_node_iter_peek(&l->iter, l->b));
570 579
571 path->pos = k.k ? k.k->p : l->b->key.k 580 path->pos = k.k ? k.k->p : l->b->key.k.p;
572 trans->paths_sorted = false; 581 trans->paths_sorted = false;
573 bch2_btree_path_verify_level(trans, pa 582 bch2_btree_path_verify_level(trans, path, l - path->l);
574 return k; 583 return k;
575 } 584 }
576 585
577 static inline struct bkey_s_c btree_path_level 586 static inline struct bkey_s_c btree_path_level_prev(struct btree_trans *trans,
578 587 struct btree_path *path,
579 588 struct btree_path_level *l,
580 589 struct bkey *u)
581 { 590 {
582 struct bkey_s_c k = __btree_iter_unpac 591 struct bkey_s_c k = __btree_iter_unpack(trans->c, l, u,
583 bch2_btree_node_iter_p 592 bch2_btree_node_iter_prev(&l->iter, l->b));
584 593
585 path->pos = k.k ? k.k->p : l->b->data- 594 path->pos = k.k ? k.k->p : l->b->data->min_key;
586 trans->paths_sorted = false; 595 trans->paths_sorted = false;
587 bch2_btree_path_verify_level(trans, pa 596 bch2_btree_path_verify_level(trans, path, l - path->l);
588 return k; 597 return k;
589 } 598 }
590 599
591 static inline bool btree_path_advance_to_pos(s 600 static inline bool btree_path_advance_to_pos(struct btree_path *path,
592 s 601 struct btree_path_level *l,
593 i 602 int max_advance)
594 { 603 {
595 struct bkey_packed *k; 604 struct bkey_packed *k;
596 int nr_advanced = 0; 605 int nr_advanced = 0;
597 606
598 while ((k = bch2_btree_node_iter_peek_ 607 while ((k = bch2_btree_node_iter_peek_all(&l->iter, l->b)) &&
599 bkey_iter_pos_cmp(l->b, k, &pat 608 bkey_iter_pos_cmp(l->b, k, &path->pos) < 0) {
600 if (max_advance > 0 && nr_adva 609 if (max_advance > 0 && nr_advanced >= max_advance)
601 return false; 610 return false;
602 611
603 bch2_btree_node_iter_advance(& 612 bch2_btree_node_iter_advance(&l->iter, l->b);
604 nr_advanced++; 613 nr_advanced++;
605 } 614 }
606 615
607 return true; 616 return true;
608 } 617 }
609 618
610 static inline void __btree_path_level_init(str 619 static inline void __btree_path_level_init(struct btree_path *path,
611 uns 620 unsigned level)
612 { 621 {
613 struct btree_path_level *l = &path->l[ 622 struct btree_path_level *l = &path->l[level];
614 623
615 bch2_btree_node_iter_init(&l->iter, l- 624 bch2_btree_node_iter_init(&l->iter, l->b, &path->pos);
616 625
617 /* 626 /*
618 * Iterators to interior nodes should 627 * Iterators to interior nodes should always be pointed at the first non
619 * whiteout: 628 * whiteout:
620 */ 629 */
621 if (level) 630 if (level)
622 bch2_btree_node_iter_peek(&l-> 631 bch2_btree_node_iter_peek(&l->iter, l->b);
623 } 632 }
624 633
625 void bch2_btree_path_level_init(struct btree_t 634 void bch2_btree_path_level_init(struct btree_trans *trans,
626 struct btree_p 635 struct btree_path *path,
627 struct btree * 636 struct btree *b)
628 { 637 {
629 BUG_ON(path->cached); 638 BUG_ON(path->cached);
630 639
631 EBUG_ON(!btree_path_pos_in_node(path, 640 EBUG_ON(!btree_path_pos_in_node(path, b));
632 641
633 path->l[b->c.level].lock_seq = six_loc 642 path->l[b->c.level].lock_seq = six_lock_seq(&b->c.lock);
634 path->l[b->c.level].b = b; 643 path->l[b->c.level].b = b;
635 __btree_path_level_init(path, b->c.lev 644 __btree_path_level_init(path, b->c.level);
636 } 645 }
637 646
638 /* Btree path: fixups after btree node updates 647 /* Btree path: fixups after btree node updates: */
639 648
640 static void bch2_trans_revalidate_updates_in_n 649 static void bch2_trans_revalidate_updates_in_node(struct btree_trans *trans, struct btree *b)
641 { 650 {
642 struct bch_fs *c = trans->c; 651 struct bch_fs *c = trans->c;
643 652
644 trans_for_each_update(trans, i) 653 trans_for_each_update(trans, i)
645 if (!i->cached && 654 if (!i->cached &&
646 i->level == b->c.level 655 i->level == b->c.level &&
647 i->btree_id == b->c.btree_ 656 i->btree_id == b->c.btree_id &&
648 bpos_cmp(i->k->k.p, b->dat 657 bpos_cmp(i->k->k.p, b->data->min_key) >= 0 &&
649 bpos_cmp(i->k->k.p, b->dat 658 bpos_cmp(i->k->k.p, b->data->max_key) <= 0) {
650 i->old_v = bch2_btree_ 659 i->old_v = bch2_btree_path_peek_slot(trans->paths + i->path, &i->old_k).v;
651 660
652 if (unlikely(trans->jo 661 if (unlikely(trans->journal_replay_not_finished)) {
653 struct bkey_i 662 struct bkey_i *j_k =
654 bch2_j 663 bch2_journal_keys_peek_slot(c, i->btree_id, i->level,
655 664 i->k->k.p);
656 665
657 if (j_k) { 666 if (j_k) {
658 i->old 667 i->old_k = j_k->k;
659 i->old 668 i->old_v = &j_k->v;
660 } 669 }
661 } 670 }
662 } 671 }
663 } 672 }
664 673
665 /* 674 /*
666 * A btree node is being replaced - update the 675 * A btree node is being replaced - update the iterator to point to the new
667 * node: 676 * node:
668 */ 677 */
669 void bch2_trans_node_add(struct btree_trans *t 678 void bch2_trans_node_add(struct btree_trans *trans,
670 struct btree_path *pa 679 struct btree_path *path,
671 struct btree *b) 680 struct btree *b)
672 { 681 {
673 struct btree_path *prev; 682 struct btree_path *prev;
674 683
675 BUG_ON(!btree_path_pos_in_node(path, b 684 BUG_ON(!btree_path_pos_in_node(path, b));
676 685
677 while ((prev = prev_btree_path(trans, 686 while ((prev = prev_btree_path(trans, path)) &&
678 btree_path_pos_in_node(prev, b) 687 btree_path_pos_in_node(prev, b))
679 path = prev; 688 path = prev;
680 689
681 for (; 690 for (;
682 path && btree_path_pos_in_node(pa 691 path && btree_path_pos_in_node(path, b);
683 path = next_btree_path(trans, pat 692 path = next_btree_path(trans, path))
684 if (path->uptodate == BTREE_IT 693 if (path->uptodate == BTREE_ITER_UPTODATE && !path->cached) {
685 enum btree_node_locked 694 enum btree_node_locked_type t =
686 btree_lock_wan 695 btree_lock_want(path, b->c.level);
687 696
688 if (t != BTREE_NODE_UN 697 if (t != BTREE_NODE_UNLOCKED) {
689 btree_node_unl 698 btree_node_unlock(trans, path, b->c.level);
690 six_lock_incre 699 six_lock_increment(&b->c.lock, (enum six_lock_type) t);
691 mark_btree_nod 700 mark_btree_node_locked(trans, path, b->c.level, t);
692 } 701 }
693 702
694 bch2_btree_path_level_ 703 bch2_btree_path_level_init(trans, path, b);
695 } 704 }
696 705
697 bch2_trans_revalidate_updates_in_node( 706 bch2_trans_revalidate_updates_in_node(trans, b);
698 } 707 }
699 708
700 /* 709 /*
701 * A btree node has been modified in such a wa 710 * A btree node has been modified in such a way as to invalidate iterators - fix
702 * them: 711 * them:
703 */ 712 */
704 void bch2_trans_node_reinit_iter(struct btree_ 713 void bch2_trans_node_reinit_iter(struct btree_trans *trans, struct btree *b)
705 { 714 {
706 struct btree_path *path; 715 struct btree_path *path;
707 unsigned i; 716 unsigned i;
708 717
709 trans_for_each_path_with_node(trans, b 718 trans_for_each_path_with_node(trans, b, path, i)
710 __btree_path_level_init(path, 719 __btree_path_level_init(path, b->c.level);
711 720
712 bch2_trans_revalidate_updates_in_node( 721 bch2_trans_revalidate_updates_in_node(trans, b);
713 } 722 }
714 723
715 /* Btree path: traverse, set_pos: */ 724 /* Btree path: traverse, set_pos: */
716 725
717 static inline int btree_path_lock_root(struct 726 static inline int btree_path_lock_root(struct btree_trans *trans,
718 struct 727 struct btree_path *path,
719 unsigne 728 unsigned depth_want,
720 unsigne 729 unsigned long trace_ip)
721 { 730 {
722 struct bch_fs *c = trans->c; 731 struct bch_fs *c = trans->c;
723 struct btree *b, **rootp = &bch2_btree 732 struct btree *b, **rootp = &bch2_btree_id_root(c, path->btree_id)->b;
724 enum six_lock_type lock_type; 733 enum six_lock_type lock_type;
725 unsigned i; 734 unsigned i;
726 int ret; 735 int ret;
727 736
728 EBUG_ON(path->nodes_locked); 737 EBUG_ON(path->nodes_locked);
729 738
730 while (1) { 739 while (1) {
731 b = READ_ONCE(*rootp); 740 b = READ_ONCE(*rootp);
732 path->level = READ_ONCE(b->c.l 741 path->level = READ_ONCE(b->c.level);
733 742
734 if (unlikely(path->level < dep 743 if (unlikely(path->level < depth_want)) {
735 /* 744 /*
736 * the root is at a lo 745 * the root is at a lower depth than the depth we want:
737 * got to the end of t 746 * got to the end of the btree, or we're walking nodes
738 * greater than some d 747 * greater than some depth and there are no nodes >=
739 * that depth 748 * that depth
740 */ 749 */
741 path->level = depth_wa 750 path->level = depth_want;
742 for (i = path->level; 751 for (i = path->level; i < BTREE_MAX_DEPTH; i++)
743 path->l[i].b = 752 path->l[i].b = NULL;
744 return 1; 753 return 1;
745 } 754 }
746 755
747 lock_type = __btree_lock_want( 756 lock_type = __btree_lock_want(path, path->level);
748 ret = btree_node_lock(trans, p 757 ret = btree_node_lock(trans, path, &b->c,
749 path->le 758 path->level, lock_type, trace_ip);
750 if (unlikely(ret)) { 759 if (unlikely(ret)) {
751 if (bch2_err_matches(r 760 if (bch2_err_matches(ret, BCH_ERR_lock_fail_root_changed))
752 continue; 761 continue;
753 if (bch2_err_matches(r 762 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
754 return ret; 763 return ret;
755 BUG(); 764 BUG();
756 } 765 }
757 766
758 if (likely(b == READ_ONCE(*roo 767 if (likely(b == READ_ONCE(*rootp) &&
759 b->c.level == path- 768 b->c.level == path->level &&
760 !race_fault())) { 769 !race_fault())) {
761 for (i = 0; i < path-> 770 for (i = 0; i < path->level; i++)
762 path->l[i].b = 771 path->l[i].b = ERR_PTR(-BCH_ERR_no_btree_node_lock_root);
763 path->l[path->level].b 772 path->l[path->level].b = b;
764 for (i = path->level + 773 for (i = path->level + 1; i < BTREE_MAX_DEPTH; i++)
765 path->l[i].b = 774 path->l[i].b = NULL;
766 775
767 mark_btree_node_locked 776 mark_btree_node_locked(trans, path, path->level,
768 777 (enum btree_node_locked_type) lock_type);
769 bch2_btree_path_level_ 778 bch2_btree_path_level_init(trans, path, b);
770 return 0; 779 return 0;
771 } 780 }
772 781
773 six_unlock_type(&b->c.lock, lo 782 six_unlock_type(&b->c.lock, lock_type);
774 } 783 }
775 } 784 }
776 785
777 noinline 786 noinline
778 static int btree_path_prefetch(struct btree_tr 787 static int btree_path_prefetch(struct btree_trans *trans, struct btree_path *path)
779 { 788 {
780 struct bch_fs *c = trans->c; 789 struct bch_fs *c = trans->c;
781 struct btree_path_level *l = path_l(pa 790 struct btree_path_level *l = path_l(path);
782 struct btree_node_iter node_iter = l-> 791 struct btree_node_iter node_iter = l->iter;
783 struct bkey_packed *k; 792 struct bkey_packed *k;
784 struct bkey_buf tmp; 793 struct bkey_buf tmp;
785 unsigned nr = test_bit(BCH_FS_started, 794 unsigned nr = test_bit(BCH_FS_started, &c->flags)
786 ? (path->level > 1 ? 0 : 2) 795 ? (path->level > 1 ? 0 : 2)
787 : (path->level > 1 ? 1 : 16); 796 : (path->level > 1 ? 1 : 16);
788 bool was_locked = btree_node_locked(pa 797 bool was_locked = btree_node_locked(path, path->level);
789 int ret = 0; 798 int ret = 0;
790 799
791 bch2_bkey_buf_init(&tmp); 800 bch2_bkey_buf_init(&tmp);
792 801
793 while (nr-- && !ret) { 802 while (nr-- && !ret) {
794 if (!bch2_btree_node_relock(tr 803 if (!bch2_btree_node_relock(trans, path, path->level))
795 break; 804 break;
796 805
797 bch2_btree_node_iter_advance(& 806 bch2_btree_node_iter_advance(&node_iter, l->b);
798 k = bch2_btree_node_iter_peek( 807 k = bch2_btree_node_iter_peek(&node_iter, l->b);
799 if (!k) 808 if (!k)
800 break; 809 break;
801 810
802 bch2_bkey_buf_unpack(&tmp, c, 811 bch2_bkey_buf_unpack(&tmp, c, l->b, k);
803 ret = bch2_btree_node_prefetch 812 ret = bch2_btree_node_prefetch(trans, path, tmp.k, path->btree_id,
804 813 path->level - 1);
805 } 814 }
806 815
807 if (!was_locked) 816 if (!was_locked)
808 btree_node_unlock(trans, path, 817 btree_node_unlock(trans, path, path->level);
809 818
810 bch2_bkey_buf_exit(&tmp, c); 819 bch2_bkey_buf_exit(&tmp, c);
811 return ret; 820 return ret;
812 } 821 }
813 822
814 static int btree_path_prefetch_j(struct btree_ 823 static int btree_path_prefetch_j(struct btree_trans *trans, struct btree_path *path,
815 struct btree_ 824 struct btree_and_journal_iter *jiter)
816 { 825 {
817 struct bch_fs *c = trans->c; 826 struct bch_fs *c = trans->c;
818 struct bkey_s_c k; 827 struct bkey_s_c k;
819 struct bkey_buf tmp; 828 struct bkey_buf tmp;
820 unsigned nr = test_bit(BCH_FS_started, 829 unsigned nr = test_bit(BCH_FS_started, &c->flags)
821 ? (path->level > 1 ? 0 : 2) 830 ? (path->level > 1 ? 0 : 2)
822 : (path->level > 1 ? 1 : 16); 831 : (path->level > 1 ? 1 : 16);
823 bool was_locked = btree_node_locked(pa 832 bool was_locked = btree_node_locked(path, path->level);
824 int ret = 0; 833 int ret = 0;
825 834
826 bch2_bkey_buf_init(&tmp); 835 bch2_bkey_buf_init(&tmp);
827 836
828 while (nr-- && !ret) { 837 while (nr-- && !ret) {
829 if (!bch2_btree_node_relock(tr 838 if (!bch2_btree_node_relock(trans, path, path->level))
830 break; 839 break;
831 840
832 bch2_btree_and_journal_iter_ad 841 bch2_btree_and_journal_iter_advance(jiter);
833 k = bch2_btree_and_journal_ite 842 k = bch2_btree_and_journal_iter_peek(jiter);
834 if (!k.k) 843 if (!k.k)
835 break; 844 break;
836 845
837 bch2_bkey_buf_reassemble(&tmp, 846 bch2_bkey_buf_reassemble(&tmp, c, k);
838 ret = bch2_btree_node_prefetch 847 ret = bch2_btree_node_prefetch(trans, path, tmp.k, path->btree_id,
839 848 path->level - 1);
840 } 849 }
841 850
842 if (!was_locked) 851 if (!was_locked)
843 btree_node_unlock(trans, path, 852 btree_node_unlock(trans, path, path->level);
844 853
845 bch2_bkey_buf_exit(&tmp, c); 854 bch2_bkey_buf_exit(&tmp, c);
846 return ret; 855 return ret;
847 } 856 }
848 857
849 static noinline void btree_node_mem_ptr_set(st 858 static noinline void btree_node_mem_ptr_set(struct btree_trans *trans,
850 st 859 struct btree_path *path,
851 un 860 unsigned plevel, struct btree *b)
852 { 861 {
853 struct btree_path_level *l = &path->l[ 862 struct btree_path_level *l = &path->l[plevel];
854 bool locked = btree_node_locked(path, 863 bool locked = btree_node_locked(path, plevel);
855 struct bkey_packed *k; 864 struct bkey_packed *k;
856 struct bch_btree_ptr_v2 *bp; 865 struct bch_btree_ptr_v2 *bp;
857 866
858 if (!bch2_btree_node_relock(trans, pat 867 if (!bch2_btree_node_relock(trans, path, plevel))
859 return; 868 return;
860 869
861 k = bch2_btree_node_iter_peek_all(&l-> 870 k = bch2_btree_node_iter_peek_all(&l->iter, l->b);
862 BUG_ON(k->type != KEY_TYPE_btree_ptr_v 871 BUG_ON(k->type != KEY_TYPE_btree_ptr_v2);
863 872
864 bp = (void *) bkeyp_val(&l->b->format, 873 bp = (void *) bkeyp_val(&l->b->format, k);
865 bp->mem_ptr = (unsigned long)b; 874 bp->mem_ptr = (unsigned long)b;
866 875
867 if (!locked) 876 if (!locked)
868 btree_node_unlock(trans, path, 877 btree_node_unlock(trans, path, plevel);
869 } 878 }
870 879
871 static noinline int btree_node_iter_and_journa 880 static noinline int btree_node_iter_and_journal_peek(struct btree_trans *trans,
872 881 struct btree_path *path,
873 882 unsigned flags,
874 883 struct bkey_buf *out)
875 { 884 {
876 struct bch_fs *c = trans->c; 885 struct bch_fs *c = trans->c;
877 struct btree_path_level *l = path_l(pa 886 struct btree_path_level *l = path_l(path);
878 struct btree_and_journal_iter jiter; 887 struct btree_and_journal_iter jiter;
879 struct bkey_s_c k; 888 struct bkey_s_c k;
880 int ret = 0; 889 int ret = 0;
881 890
882 __bch2_btree_and_journal_iter_init_nod 891 __bch2_btree_and_journal_iter_init_node_iter(trans, &jiter, l->b, l->iter, path->pos);
883 892
884 k = bch2_btree_and_journal_iter_peek(& 893 k = bch2_btree_and_journal_iter_peek(&jiter);
885 if (!k.k) { <<
886 struct printbuf buf = PRINTBUF <<
887 <<
888 prt_str(&buf, "node not found <<
889 bch2_bpos_to_text(&buf, path-> <<
890 prt_str(&buf, " at btree "); <<
891 bch2_btree_pos_to_text(&buf, c <<
892 <<
893 ret = bch2_fs_topology_error(c <<
894 printbuf_exit(&buf); <<
895 goto err; <<
896 } <<
897 894
898 bch2_bkey_buf_reassemble(out, c, k); 895 bch2_bkey_buf_reassemble(out, c, k);
899 896
900 if ((flags & BTREE_ITER_prefetch) && 897 if ((flags & BTREE_ITER_prefetch) &&
901 c->opts.btree_node_prefetch) 898 c->opts.btree_node_prefetch)
902 ret = btree_path_prefetch_j(tr 899 ret = btree_path_prefetch_j(trans, path, &jiter);
903 900
904 err: <<
905 bch2_btree_and_journal_iter_exit(&jite 901 bch2_btree_and_journal_iter_exit(&jiter);
906 return ret; 902 return ret;
907 } 903 }
908 904
909 static __always_inline int btree_path_down(str 905 static __always_inline int btree_path_down(struct btree_trans *trans,
910 str 906 struct btree_path *path,
911 uns 907 unsigned flags,
912 uns 908 unsigned long trace_ip)
913 { 909 {
914 struct bch_fs *c = trans->c; 910 struct bch_fs *c = trans->c;
915 struct btree_path_level *l = path_l(pa 911 struct btree_path_level *l = path_l(path);
916 struct btree *b; 912 struct btree *b;
917 unsigned level = path->level - 1; 913 unsigned level = path->level - 1;
918 enum six_lock_type lock_type = __btree 914 enum six_lock_type lock_type = __btree_lock_want(path, level);
919 struct bkey_buf tmp; 915 struct bkey_buf tmp;
920 int ret; 916 int ret;
921 917
922 EBUG_ON(!btree_node_locked(path, path- 918 EBUG_ON(!btree_node_locked(path, path->level));
923 919
924 bch2_bkey_buf_init(&tmp); 920 bch2_bkey_buf_init(&tmp);
925 921
926 if (unlikely(trans->journal_replay_not 922 if (unlikely(trans->journal_replay_not_finished)) {
927 ret = btree_node_iter_and_jour 923 ret = btree_node_iter_and_journal_peek(trans, path, flags, &tmp);
928 if (ret) 924 if (ret)
929 goto err; 925 goto err;
930 } else { 926 } else {
931 struct bkey_packed *k = bch2_b 927 struct bkey_packed *k = bch2_btree_node_iter_peek(&l->iter, l->b);
932 if (!k) { 928 if (!k) {
933 struct printbuf buf = 929 struct printbuf buf = PRINTBUF;
934 930
935 prt_str(&buf, "node no 931 prt_str(&buf, "node not found at pos ");
936 bch2_bpos_to_text(&buf 932 bch2_bpos_to_text(&buf, path->pos);
937 prt_str(&buf, " within 933 prt_str(&buf, " within parent node ");
938 bch2_bkey_val_to_text( 934 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&l->b->key));
939 935
940 bch2_fs_fatal_error(c, 936 bch2_fs_fatal_error(c, "%s", buf.buf);
941 printbuf_exit(&buf); 937 printbuf_exit(&buf);
942 ret = -BCH_ERR_btree_n 938 ret = -BCH_ERR_btree_need_topology_repair;
943 goto err; 939 goto err;
944 } 940 }
945 941
946 bch2_bkey_buf_unpack(&tmp, c, 942 bch2_bkey_buf_unpack(&tmp, c, l->b, k);
947 943
948 if ((flags & BTREE_ITER_prefet 944 if ((flags & BTREE_ITER_prefetch) &&
949 c->opts.btree_node_prefetc 945 c->opts.btree_node_prefetch) {
950 ret = btree_path_prefe 946 ret = btree_path_prefetch(trans, path);
951 if (ret) 947 if (ret)
952 goto err; 948 goto err;
953 } 949 }
954 } 950 }
955 951
956 b = bch2_btree_node_get(trans, path, t 952 b = bch2_btree_node_get(trans, path, tmp.k, level, lock_type, trace_ip);
957 ret = PTR_ERR_OR_ZERO(b); 953 ret = PTR_ERR_OR_ZERO(b);
958 if (unlikely(ret)) 954 if (unlikely(ret))
959 goto err; 955 goto err;
960 956
961 if (likely(!trans->journal_replay_not_ 957 if (likely(!trans->journal_replay_not_finished &&
962 tmp.k->k.type == KEY_TYPE_b 958 tmp.k->k.type == KEY_TYPE_btree_ptr_v2) &&
963 unlikely(b != btree_node_mem_ptr(t 959 unlikely(b != btree_node_mem_ptr(tmp.k)))
964 btree_node_mem_ptr_set(trans, 960 btree_node_mem_ptr_set(trans, path, level + 1, b);
965 961
966 if (btree_node_read_locked(path, level 962 if (btree_node_read_locked(path, level + 1))
967 btree_node_unlock(trans, path, 963 btree_node_unlock(trans, path, level + 1);
968 964
969 mark_btree_node_locked(trans, path, le 965 mark_btree_node_locked(trans, path, level,
970 (enum btree_nod 966 (enum btree_node_locked_type) lock_type);
971 path->level = level; 967 path->level = level;
972 bch2_btree_path_level_init(trans, path 968 bch2_btree_path_level_init(trans, path, b);
973 969
974 bch2_btree_path_verify_locks(path); 970 bch2_btree_path_verify_locks(path);
975 err: 971 err:
976 bch2_bkey_buf_exit(&tmp, c); 972 bch2_bkey_buf_exit(&tmp, c);
977 return ret; 973 return ret;
978 } 974 }
979 975
980 static int bch2_btree_path_traverse_all(struct 976 static int bch2_btree_path_traverse_all(struct btree_trans *trans)
981 { 977 {
982 struct bch_fs *c = trans->c; 978 struct bch_fs *c = trans->c;
983 struct btree_path *path; 979 struct btree_path *path;
984 unsigned long trace_ip = _RET_IP_; 980 unsigned long trace_ip = _RET_IP_;
985 unsigned i; 981 unsigned i;
986 int ret = 0; 982 int ret = 0;
987 983
988 if (trans->in_traverse_all) 984 if (trans->in_traverse_all)
989 return -BCH_ERR_transaction_re 985 return -BCH_ERR_transaction_restart_in_traverse_all;
990 986
991 trans->in_traverse_all = true; 987 trans->in_traverse_all = true;
992 retry_all: 988 retry_all:
993 trans->restarted = 0; 989 trans->restarted = 0;
994 trans->last_restarted_ip = 0; 990 trans->last_restarted_ip = 0;
995 991
996 trans_for_each_path(trans, path, i) 992 trans_for_each_path(trans, path, i)
997 path->should_be_locked = false 993 path->should_be_locked = false;
998 994
999 btree_trans_sort_paths(trans); 995 btree_trans_sort_paths(trans);
1000 996
1001 bch2_trans_unlock(trans); 997 bch2_trans_unlock(trans);
1002 cond_resched(); 998 cond_resched();
1003 trans_set_locked(trans); 999 trans_set_locked(trans);
1004 1000
1005 if (unlikely(trans->memory_allocation 1001 if (unlikely(trans->memory_allocation_failure)) {
1006 struct closure cl; 1002 struct closure cl;
1007 1003
1008 closure_init_stack(&cl); 1004 closure_init_stack(&cl);
1009 1005
1010 do { 1006 do {
1011 ret = bch2_btree_cach 1007 ret = bch2_btree_cache_cannibalize_lock(trans, &cl);
1012 closure_sync(&cl); 1008 closure_sync(&cl);
1013 } while (ret); 1009 } while (ret);
1014 } 1010 }
1015 1011
1016 /* Now, redo traversals in correct or 1012 /* Now, redo traversals in correct order: */
1017 i = 0; 1013 i = 0;
1018 while (i < trans->nr_sorted) { 1014 while (i < trans->nr_sorted) {
1019 btree_path_idx_t idx = trans- 1015 btree_path_idx_t idx = trans->sorted[i];
1020 1016
1021 /* 1017 /*
1022 * Traversing a path can caus 1018 * Traversing a path can cause another path to be added at about
1023 * the same position: 1019 * the same position:
1024 */ 1020 */
1025 if (trans->paths[idx].uptodat 1021 if (trans->paths[idx].uptodate) {
1026 __btree_path_get(tran !! 1022 __btree_path_get(&trans->paths[idx], false);
1027 ret = bch2_btree_path 1023 ret = bch2_btree_path_traverse_one(trans, idx, 0, _THIS_IP_);
1028 __btree_path_put(tran !! 1024 __btree_path_put(&trans->paths[idx], false);
1029 1025
1030 if (bch2_err_matches( 1026 if (bch2_err_matches(ret, BCH_ERR_transaction_restart) ||
1031 bch2_err_matches( 1027 bch2_err_matches(ret, ENOMEM))
1032 goto retry_al 1028 goto retry_all;
1033 if (ret) 1029 if (ret)
1034 goto err; 1030 goto err;
1035 } else { 1031 } else {
1036 i++; 1032 i++;
1037 } 1033 }
1038 } 1034 }
1039 1035
1040 /* 1036 /*
1041 * We used to assert that all paths h 1037 * We used to assert that all paths had been traversed here
1042 * (path->uptodate < BTREE_ITER_NEED_ 1038 * (path->uptodate < BTREE_ITER_NEED_TRAVERSE); however, since
1043 * path->should_be_locked is not set 1039 * path->should_be_locked is not set yet, we might have unlocked and
1044 * then failed to relock a path - tha 1040 * then failed to relock a path - that's fine.
1045 */ 1041 */
1046 err: 1042 err:
1047 bch2_btree_cache_cannibalize_unlock(t 1043 bch2_btree_cache_cannibalize_unlock(trans);
1048 1044
1049 trans->in_traverse_all = false; 1045 trans->in_traverse_all = false;
1050 1046
1051 trace_and_count(c, trans_traverse_all 1047 trace_and_count(c, trans_traverse_all, trans, trace_ip);
1052 return ret; 1048 return ret;
1053 } 1049 }
1054 1050
1055 static inline bool btree_path_check_pos_in_no 1051 static inline bool btree_path_check_pos_in_node(struct btree_path *path,
1056 1052 unsigned l, int check_pos)
1057 { 1053 {
1058 if (check_pos < 0 && btree_path_pos_b 1054 if (check_pos < 0 && btree_path_pos_before_node(path, path->l[l].b))
1059 return false; 1055 return false;
1060 if (check_pos > 0 && btree_path_pos_a 1056 if (check_pos > 0 && btree_path_pos_after_node(path, path->l[l].b))
1061 return false; 1057 return false;
1062 return true; 1058 return true;
1063 } 1059 }
1064 1060
1065 static inline bool btree_path_good_node(struc 1061 static inline bool btree_path_good_node(struct btree_trans *trans,
1066 struc 1062 struct btree_path *path,
1067 unsig 1063 unsigned l, int check_pos)
1068 { 1064 {
1069 return is_btree_node(path, l) && 1065 return is_btree_node(path, l) &&
1070 bch2_btree_node_relock(trans, 1066 bch2_btree_node_relock(trans, path, l) &&
1071 btree_path_check_pos_in_node( 1067 btree_path_check_pos_in_node(path, l, check_pos);
1072 } 1068 }
1073 1069
1074 static void btree_path_set_level_down(struct 1070 static void btree_path_set_level_down(struct btree_trans *trans,
1075 struct 1071 struct btree_path *path,
1076 unsigne 1072 unsigned new_level)
1077 { 1073 {
1078 unsigned l; 1074 unsigned l;
1079 1075
1080 path->level = new_level; 1076 path->level = new_level;
1081 1077
1082 for (l = path->level + 1; l < BTREE_M 1078 for (l = path->level + 1; l < BTREE_MAX_DEPTH; l++)
1083 if (btree_lock_want(path, l) 1079 if (btree_lock_want(path, l) == BTREE_NODE_UNLOCKED)
1084 btree_node_unlock(tra 1080 btree_node_unlock(trans, path, l);
1085 1081
1086 btree_path_set_dirty(path, BTREE_ITER 1082 btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
1087 bch2_btree_path_verify(trans, path); 1083 bch2_btree_path_verify(trans, path);
1088 } 1084 }
1089 1085
1090 static noinline unsigned __btree_path_up_unti 1086 static noinline unsigned __btree_path_up_until_good_node(struct btree_trans *trans,
1091 1087 struct btree_path *path,
1092 1088 int check_pos)
1093 { 1089 {
1094 unsigned i, l = path->level; 1090 unsigned i, l = path->level;
1095 again: 1091 again:
1096 while (btree_path_node(path, l) && 1092 while (btree_path_node(path, l) &&
1097 !btree_path_good_node(trans, p 1093 !btree_path_good_node(trans, path, l, check_pos))
1098 __btree_path_set_level_up(tra 1094 __btree_path_set_level_up(trans, path, l++);
1099 1095
1100 /* If we need intent locks, take them 1096 /* If we need intent locks, take them too: */
1101 for (i = l + 1; 1097 for (i = l + 1;
1102 i < path->locks_want && btree_pa 1098 i < path->locks_want && btree_path_node(path, i);
1103 i++) 1099 i++)
1104 if (!bch2_btree_node_relock(t 1100 if (!bch2_btree_node_relock(trans, path, i)) {
1105 while (l <= i) 1101 while (l <= i)
1106 __btree_path_ 1102 __btree_path_set_level_up(trans, path, l++);
1107 goto again; 1103 goto again;
1108 } 1104 }
1109 1105
1110 return l; 1106 return l;
1111 } 1107 }
1112 1108
1113 static inline unsigned btree_path_up_until_go 1109 static inline unsigned btree_path_up_until_good_node(struct btree_trans *trans,
1114 1110 struct btree_path *path,
1115 1111 int check_pos)
1116 { 1112 {
1117 return likely(btree_node_locked(path, 1113 return likely(btree_node_locked(path, path->level) &&
1118 btree_path_check_pos_in 1114 btree_path_check_pos_in_node(path, path->level, check_pos))
1119 ? path->level 1115 ? path->level
1120 : __btree_path_up_until_good_ 1116 : __btree_path_up_until_good_node(trans, path, check_pos);
1121 } 1117 }
1122 1118
1123 /* 1119 /*
1124 * This is the main state machine for walking 1120 * This is the main state machine for walking down the btree - walks down to a
1125 * specified depth 1121 * specified depth
1126 * 1122 *
1127 * Returns 0 on success, -EIO on error (error 1123 * Returns 0 on success, -EIO on error (error reading in a btree node).
1128 * 1124 *
1129 * On error, caller (peek_node()/peek_key()) 1125 * On error, caller (peek_node()/peek_key()) must return NULL; the error is
1130 * stashed in the iterator and returned from 1126 * stashed in the iterator and returned from bch2_trans_exit().
1131 */ 1127 */
1132 int bch2_btree_path_traverse_one(struct btree 1128 int bch2_btree_path_traverse_one(struct btree_trans *trans,
1133 btree_path_i 1129 btree_path_idx_t path_idx,
1134 unsigned fla 1130 unsigned flags,
1135 unsigned lon 1131 unsigned long trace_ip)
1136 { 1132 {
1137 struct btree_path *path = &trans->pat 1133 struct btree_path *path = &trans->paths[path_idx];
1138 unsigned depth_want = path->level; 1134 unsigned depth_want = path->level;
1139 int ret = -((int) trans->restarted); 1135 int ret = -((int) trans->restarted);
1140 1136
1141 if (unlikely(ret)) 1137 if (unlikely(ret))
1142 goto out; 1138 goto out;
1143 1139
1144 if (unlikely(!trans->srcu_held)) 1140 if (unlikely(!trans->srcu_held))
1145 bch2_trans_srcu_lock(trans); 1141 bch2_trans_srcu_lock(trans);
1146 1142
1147 trace_btree_path_traverse_start(trans <<
1148 <<
1149 /* 1143 /*
1150 * Ensure we obey path->should_be_loc 1144 * Ensure we obey path->should_be_locked: if it's set, we can't unlock
1151 * and re-traverse the path without a 1145 * and re-traverse the path without a transaction restart:
1152 */ 1146 */
1153 if (path->should_be_locked) { 1147 if (path->should_be_locked) {
1154 ret = bch2_btree_path_relock( 1148 ret = bch2_btree_path_relock(trans, path, trace_ip);
1155 goto out; 1149 goto out;
1156 } 1150 }
1157 1151
1158 if (path->cached) { 1152 if (path->cached) {
1159 ret = bch2_btree_path_travers 1153 ret = bch2_btree_path_traverse_cached(trans, path, flags);
1160 goto out; 1154 goto out;
1161 } 1155 }
1162 1156
1163 path = &trans->paths[path_idx]; 1157 path = &trans->paths[path_idx];
1164 1158
1165 if (unlikely(path->level >= BTREE_MAX 1159 if (unlikely(path->level >= BTREE_MAX_DEPTH))
1166 goto out_uptodate; 1160 goto out_uptodate;
1167 1161
1168 path->level = btree_path_up_until_goo 1162 path->level = btree_path_up_until_good_node(trans, path, 0);
1169 unsigned max_level = path->level; 1163 unsigned max_level = path->level;
1170 1164
1171 EBUG_ON(btree_path_node(path, path->l 1165 EBUG_ON(btree_path_node(path, path->level) &&
1172 !btree_node_locked(path, path 1166 !btree_node_locked(path, path->level));
1173 1167
1174 /* 1168 /*
1175 * Note: path->nodes[path->level] may 1169 * Note: path->nodes[path->level] may be temporarily NULL here - that
1176 * would indicate to other code that 1170 * would indicate to other code that we got to the end of the btree,
1177 * here it indicates that relocking t 1171 * here it indicates that relocking the root failed - it's critical that
1178 * btree_path_lock_root() comes next 1172 * btree_path_lock_root() comes next and that it can't fail
1179 */ 1173 */
1180 while (path->level > depth_want) { 1174 while (path->level > depth_want) {
1181 ret = btree_path_node(path, p 1175 ret = btree_path_node(path, path->level)
1182 ? btree_path_down(tra 1176 ? btree_path_down(trans, path, flags, trace_ip)
1183 : btree_path_lock_roo 1177 : btree_path_lock_root(trans, path, depth_want, trace_ip);
1184 if (unlikely(ret)) { 1178 if (unlikely(ret)) {
1185 if (ret == 1) { 1179 if (ret == 1) {
1186 /* 1180 /*
1187 * No nodes a 1181 * No nodes at this level - got to the end of
1188 * the btree: 1182 * the btree:
1189 */ 1183 */
1190 ret = 0; 1184 ret = 0;
1191 goto out; 1185 goto out;
1192 } 1186 }
1193 1187
1194 __bch2_btree_path_unl 1188 __bch2_btree_path_unlock(trans, path);
1195 path->level = depth_w 1189 path->level = depth_want;
1196 path->l[path->level]. 1190 path->l[path->level].b = ERR_PTR(ret);
1197 goto out; 1191 goto out;
1198 } 1192 }
1199 } 1193 }
1200 1194
1201 if (unlikely(max_level > path->level) 1195 if (unlikely(max_level > path->level)) {
1202 struct btree_path *linked; 1196 struct btree_path *linked;
1203 unsigned iter; 1197 unsigned iter;
1204 1198
1205 trans_for_each_path_with_node 1199 trans_for_each_path_with_node(trans, path_l(path)->b, linked, iter)
1206 for (unsigned j = pat 1200 for (unsigned j = path->level + 1; j < max_level; j++)
1207 linked->l[j] 1201 linked->l[j] = path->l[j];
1208 } 1202 }
1209 1203
1210 out_uptodate: 1204 out_uptodate:
1211 path->uptodate = BTREE_ITER_UPTODATE; 1205 path->uptodate = BTREE_ITER_UPTODATE;
1212 trace_btree_path_traverse_end(trans, <<
1213 out: 1206 out:
1214 if (bch2_err_matches(ret, BCH_ERR_tra 1207 if (bch2_err_matches(ret, BCH_ERR_transaction_restart) != !!trans->restarted)
1215 panic("ret %s (%i) trans->res 1208 panic("ret %s (%i) trans->restarted %s (%i)\n",
1216 bch2_err_str(ret), ret, 1209 bch2_err_str(ret), ret,
1217 bch2_err_str(trans->res 1210 bch2_err_str(trans->restarted), trans->restarted);
1218 bch2_btree_path_verify(trans, path); 1211 bch2_btree_path_verify(trans, path);
1219 return ret; 1212 return ret;
1220 } 1213 }
1221 1214
1222 static inline void btree_path_copy(struct btr 1215 static inline void btree_path_copy(struct btree_trans *trans, struct btree_path *dst,
1223 struct btree_path 1216 struct btree_path *src)
1224 { 1217 {
1225 unsigned i, offset = offsetof(struct 1218 unsigned i, offset = offsetof(struct btree_path, pos);
1226 1219
1227 memcpy((void *) dst + offset, 1220 memcpy((void *) dst + offset,
1228 (void *) src + offset, 1221 (void *) src + offset,
1229 sizeof(struct btree_path) - of 1222 sizeof(struct btree_path) - offset);
1230 1223
1231 for (i = 0; i < BTREE_MAX_DEPTH; i++) 1224 for (i = 0; i < BTREE_MAX_DEPTH; i++) {
1232 unsigned t = btree_node_locke 1225 unsigned t = btree_node_locked_type(dst, i);
1233 1226
1234 if (t != BTREE_NODE_UNLOCKED) 1227 if (t != BTREE_NODE_UNLOCKED)
1235 six_lock_increment(&d 1228 six_lock_increment(&dst->l[i].b->c.lock, t);
1236 } 1229 }
1237 } 1230 }
1238 1231
1239 static btree_path_idx_t btree_path_clone(stru 1232 static btree_path_idx_t btree_path_clone(struct btree_trans *trans, btree_path_idx_t src,
1240 bool 1233 bool intent, unsigned long ip)
1241 { 1234 {
1242 btree_path_idx_t new = btree_path_all 1235 btree_path_idx_t new = btree_path_alloc(trans, src);
1243 btree_path_copy(trans, trans->paths + 1236 btree_path_copy(trans, trans->paths + new, trans->paths + src);
1244 __btree_path_get(trans, trans->paths !! 1237 __btree_path_get(trans->paths + new, intent);
1245 #ifdef TRACK_PATH_ALLOCATED 1238 #ifdef TRACK_PATH_ALLOCATED
1246 trans->paths[new].ip_allocated = ip; 1239 trans->paths[new].ip_allocated = ip;
1247 #endif 1240 #endif
1248 return new; 1241 return new;
1249 } 1242 }
1250 1243
1251 __flatten 1244 __flatten
1252 btree_path_idx_t __bch2_btree_path_make_mut(s 1245 btree_path_idx_t __bch2_btree_path_make_mut(struct btree_trans *trans,
1253 btree_path_idx_t path 1246 btree_path_idx_t path, bool intent, unsigned long ip)
1254 { 1247 {
1255 struct btree_path *old = trans->paths !! 1248 __btree_path_put(trans->paths + path, intent);
1256 __btree_path_put(trans, trans->paths <<
1257 path = btree_path_clone(trans, path, 1249 path = btree_path_clone(trans, path, intent, ip);
1258 trace_btree_path_clone(trans, old, tr <<
1259 trans->paths[path].preserve = false; 1250 trans->paths[path].preserve = false;
1260 return path; 1251 return path;
1261 } 1252 }
1262 1253
1263 btree_path_idx_t __must_check 1254 btree_path_idx_t __must_check
1264 __bch2_btree_path_set_pos(struct btree_trans 1255 __bch2_btree_path_set_pos(struct btree_trans *trans,
1265 btree_path_idx_t pa 1256 btree_path_idx_t path_idx, struct bpos new_pos,
1266 bool intent, unsign 1257 bool intent, unsigned long ip)
1267 { 1258 {
1268 int cmp = bpos_cmp(new_pos, trans->pa 1259 int cmp = bpos_cmp(new_pos, trans->paths[path_idx].pos);
1269 1260
1270 bch2_trans_verify_not_in_restart(tran 1261 bch2_trans_verify_not_in_restart(trans);
1271 EBUG_ON(!trans->paths[path_idx].ref); 1262 EBUG_ON(!trans->paths[path_idx].ref);
1272 1263
1273 trace_btree_path_set_pos(trans, trans <<
1274 <<
1275 path_idx = bch2_btree_path_make_mut(t 1264 path_idx = bch2_btree_path_make_mut(trans, path_idx, intent, ip);
1276 1265
1277 struct btree_path *path = trans->path 1266 struct btree_path *path = trans->paths + path_idx;
1278 path->pos = new_pos; 1267 path->pos = new_pos;
1279 trans->paths_sorted = false; 1268 trans->paths_sorted = false;
1280 1269
1281 if (unlikely(path->cached)) { 1270 if (unlikely(path->cached)) {
1282 btree_node_unlock(trans, path 1271 btree_node_unlock(trans, path, 0);
1283 path->l[0].b = ERR_PTR(-BCH_E 1272 path->l[0].b = ERR_PTR(-BCH_ERR_no_btree_node_up);
1284 btree_path_set_dirty(path, BT 1273 btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
1285 goto out; 1274 goto out;
1286 } 1275 }
1287 1276
1288 unsigned level = btree_path_up_until_ 1277 unsigned level = btree_path_up_until_good_node(trans, path, cmp);
1289 1278
1290 if (btree_path_node(path, level)) { 1279 if (btree_path_node(path, level)) {
1291 struct btree_path_level *l = 1280 struct btree_path_level *l = &path->l[level];
1292 1281
1293 BUG_ON(!btree_node_locked(pat 1282 BUG_ON(!btree_node_locked(path, level));
1294 /* 1283 /*
1295 * We might have to skip over 1284 * We might have to skip over many keys, or just a few: try
1296 * advancing the node iterato 1285 * advancing the node iterator, and if we have to skip over too
1297 * many keys just reinit it ( 1286 * many keys just reinit it (or if we're rewinding, since that
1298 * is expensive). 1287 * is expensive).
1299 */ 1288 */
1300 if (cmp < 0 || 1289 if (cmp < 0 ||
1301 !btree_path_advance_to_po 1290 !btree_path_advance_to_pos(path, l, 8))
1302 bch2_btree_node_iter_ 1291 bch2_btree_node_iter_init(&l->iter, l->b, &path->pos);
1303 1292
1304 /* 1293 /*
1305 * Iterators to interior node 1294 * Iterators to interior nodes should always be pointed at the first non
1306 * whiteout: 1295 * whiteout:
1307 */ 1296 */
1308 if (unlikely(level)) 1297 if (unlikely(level))
1309 bch2_btree_node_iter_ 1298 bch2_btree_node_iter_peek(&l->iter, l->b);
1310 } 1299 }
1311 1300
1312 if (unlikely(level != path->level)) { 1301 if (unlikely(level != path->level)) {
1313 btree_path_set_dirty(path, BT 1302 btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
1314 __bch2_btree_path_unlock(tran 1303 __bch2_btree_path_unlock(trans, path);
1315 } 1304 }
1316 out: 1305 out:
1317 bch2_btree_path_verify(trans, path); 1306 bch2_btree_path_verify(trans, path);
1318 return path_idx; 1307 return path_idx;
1319 } 1308 }
1320 1309
1321 /* Btree path: main interface: */ 1310 /* Btree path: main interface: */
1322 1311
1323 static struct btree_path *have_path_at_pos(st 1312 static struct btree_path *have_path_at_pos(struct btree_trans *trans, struct btree_path *path)
1324 { 1313 {
1325 struct btree_path *sib; 1314 struct btree_path *sib;
1326 1315
1327 sib = prev_btree_path(trans, path); 1316 sib = prev_btree_path(trans, path);
1328 if (sib && !btree_path_cmp(sib, path) 1317 if (sib && !btree_path_cmp(sib, path))
1329 return sib; 1318 return sib;
1330 1319
1331 sib = next_btree_path(trans, path); 1320 sib = next_btree_path(trans, path);
1332 if (sib && !btree_path_cmp(sib, path) 1321 if (sib && !btree_path_cmp(sib, path))
1333 return sib; 1322 return sib;
1334 1323
1335 return NULL; 1324 return NULL;
1336 } 1325 }
1337 1326
1338 static struct btree_path *have_node_at_pos(st 1327 static struct btree_path *have_node_at_pos(struct btree_trans *trans, struct btree_path *path)
1339 { 1328 {
1340 struct btree_path *sib; 1329 struct btree_path *sib;
1341 1330
1342 sib = prev_btree_path(trans, path); 1331 sib = prev_btree_path(trans, path);
1343 if (sib && sib->level == path->level 1332 if (sib && sib->level == path->level && path_l(sib)->b == path_l(path)->b)
1344 return sib; 1333 return sib;
1345 1334
1346 sib = next_btree_path(trans, path); 1335 sib = next_btree_path(trans, path);
1347 if (sib && sib->level == path->level 1336 if (sib && sib->level == path->level && path_l(sib)->b == path_l(path)->b)
1348 return sib; 1337 return sib;
1349 1338
1350 return NULL; 1339 return NULL;
1351 } 1340 }
1352 1341
1353 static inline void __bch2_path_free(struct bt 1342 static inline void __bch2_path_free(struct btree_trans *trans, btree_path_idx_t path)
1354 { 1343 {
1355 __bch2_btree_path_unlock(trans, trans 1344 __bch2_btree_path_unlock(trans, trans->paths + path);
1356 btree_path_list_remove(trans, trans-> 1345 btree_path_list_remove(trans, trans->paths + path);
1357 __clear_bit(path, trans->paths_alloca 1346 __clear_bit(path, trans->paths_allocated);
1358 } 1347 }
1359 1348
1360 static bool bch2_btree_path_can_relock(struct 1349 static bool bch2_btree_path_can_relock(struct btree_trans *trans, struct btree_path *path)
1361 { 1350 {
1362 unsigned l = path->level; 1351 unsigned l = path->level;
1363 1352
1364 do { 1353 do {
1365 if (!btree_path_node(path, l) 1354 if (!btree_path_node(path, l))
1366 break; 1355 break;
1367 1356
1368 if (!is_btree_node(path, l)) 1357 if (!is_btree_node(path, l))
1369 return false; 1358 return false;
1370 1359
1371 if (path->l[l].lock_seq != pa 1360 if (path->l[l].lock_seq != path->l[l].b->c.lock.seq)
1372 return false; 1361 return false;
1373 1362
1374 l++; 1363 l++;
1375 } while (l < path->locks_want); 1364 } while (l < path->locks_want);
1376 1365
1377 return true; 1366 return true;
1378 } 1367 }
1379 1368
1380 void bch2_path_put(struct btree_trans *trans, 1369 void bch2_path_put(struct btree_trans *trans, btree_path_idx_t path_idx, bool intent)
1381 { 1370 {
1382 struct btree_path *path = trans->path 1371 struct btree_path *path = trans->paths + path_idx, *dup;
1383 1372
1384 if (!__btree_path_put(trans, path, in !! 1373 if (!__btree_path_put(path, intent))
1385 return; 1374 return;
1386 1375
1387 dup = path->preserve 1376 dup = path->preserve
1388 ? have_path_at_pos(trans, pat 1377 ? have_path_at_pos(trans, path)
1389 : have_node_at_pos(trans, pat 1378 : have_node_at_pos(trans, path);
1390 1379
1391 trace_btree_path_free(trans, path_idx <<
1392 <<
1393 if (!dup && !(!path->preserve && !is_ 1380 if (!dup && !(!path->preserve && !is_btree_node(path, path->level)))
1394 return; 1381 return;
1395 1382
1396 if (path->should_be_locked && !trans- 1383 if (path->should_be_locked && !trans->restarted) {
1397 if (!dup) 1384 if (!dup)
1398 return; 1385 return;
1399 1386
1400 if (!(trans->locked 1387 if (!(trans->locked
1401 ? bch2_btree_path_reloc 1388 ? bch2_btree_path_relock_norestart(trans, dup)
1402 : bch2_btree_path_can_r 1389 : bch2_btree_path_can_relock(trans, dup)))
1403 return; 1390 return;
1404 } 1391 }
1405 1392
1406 if (dup) { 1393 if (dup) {
1407 dup->preserve |= pa 1394 dup->preserve |= path->preserve;
1408 dup->should_be_locked |= pa 1395 dup->should_be_locked |= path->should_be_locked;
1409 } 1396 }
1410 1397
1411 __bch2_path_free(trans, path_idx); 1398 __bch2_path_free(trans, path_idx);
1412 } 1399 }
1413 1400
1414 static void bch2_path_put_nokeep(struct btree 1401 static void bch2_path_put_nokeep(struct btree_trans *trans, btree_path_idx_t path,
1415 bool intent) 1402 bool intent)
1416 { 1403 {
1417 if (!__btree_path_put(trans, trans->p !! 1404 if (!__btree_path_put(trans->paths + path, intent))
1418 return; 1405 return;
1419 1406
1420 __bch2_path_free(trans, path); 1407 __bch2_path_free(trans, path);
1421 } 1408 }
1422 1409
1423 void __noreturn bch2_trans_restart_error(stru 1410 void __noreturn bch2_trans_restart_error(struct btree_trans *trans, u32 restart_count)
1424 { 1411 {
1425 panic("trans->restart_count %u, shoul 1412 panic("trans->restart_count %u, should be %u, last restarted by %pS\n",
1426 trans->restart_count, restart_c 1413 trans->restart_count, restart_count,
1427 (void *) trans->last_begin_ip); 1414 (void *) trans->last_begin_ip);
1428 } 1415 }
1429 1416
1430 void __noreturn bch2_trans_in_restart_error(s 1417 void __noreturn bch2_trans_in_restart_error(struct btree_trans *trans)
1431 { 1418 {
1432 panic("in transaction restart: %s, la 1419 panic("in transaction restart: %s, last restarted by %pS\n",
1433 bch2_err_str(trans->restarted), 1420 bch2_err_str(trans->restarted),
1434 (void *) trans->last_restarted_ 1421 (void *) trans->last_restarted_ip);
1435 } 1422 }
1436 1423
1437 void __noreturn bch2_trans_unlocked_error(str 1424 void __noreturn bch2_trans_unlocked_error(struct btree_trans *trans)
1438 { 1425 {
1439 panic("trans should be locked, unlock 1426 panic("trans should be locked, unlocked by %pS\n",
1440 (void *) trans->last_unlock_ip) 1427 (void *) trans->last_unlock_ip);
1441 } 1428 }
1442 1429
1443 noinline __cold 1430 noinline __cold
1444 void bch2_trans_updates_to_text(struct printb 1431 void bch2_trans_updates_to_text(struct printbuf *buf, struct btree_trans *trans)
1445 { 1432 {
1446 prt_printf(buf, "%u transaction updat !! 1433 prt_printf(buf, "transaction updates for %s journal seq %llu\n",
1447 trans->nr_updates, trans-> !! 1434 trans->fn, trans->journal_res.seq);
1448 printbuf_indent_add(buf, 2); 1435 printbuf_indent_add(buf, 2);
1449 1436
1450 trans_for_each_update(trans, i) { 1437 trans_for_each_update(trans, i) {
1451 struct bkey_s_c old = { &i->o 1438 struct bkey_s_c old = { &i->old_k, i->old_v };
1452 1439
1453 prt_printf(buf, "update: btre 1440 prt_printf(buf, "update: btree=%s cached=%u %pS\n",
1454 bch2_btree_id_str(i->b 1441 bch2_btree_id_str(i->btree_id),
1455 i->cached, 1442 i->cached,
1456 (void *) i->ip_allocat 1443 (void *) i->ip_allocated);
1457 1444
1458 prt_printf(buf, " old "); 1445 prt_printf(buf, " old ");
1459 bch2_bkey_val_to_text(buf, tr 1446 bch2_bkey_val_to_text(buf, trans->c, old);
1460 prt_newline(buf); 1447 prt_newline(buf);
1461 1448
1462 prt_printf(buf, " new "); 1449 prt_printf(buf, " new ");
1463 bch2_bkey_val_to_text(buf, tr 1450 bch2_bkey_val_to_text(buf, trans->c, bkey_i_to_s_c(i->k));
1464 prt_newline(buf); 1451 prt_newline(buf);
1465 } 1452 }
1466 1453
1467 for (struct jset_entry *e = trans->jo 1454 for (struct jset_entry *e = trans->journal_entries;
1468 e != btree_trans_journal_entries 1455 e != btree_trans_journal_entries_top(trans);
1469 e = vstruct_next(e)) 1456 e = vstruct_next(e))
1470 bch2_journal_entry_to_text(bu 1457 bch2_journal_entry_to_text(buf, trans->c, e);
1471 1458
1472 printbuf_indent_sub(buf, 2); 1459 printbuf_indent_sub(buf, 2);
1473 } 1460 }
1474 1461
1475 noinline __cold 1462 noinline __cold
1476 void bch2_dump_trans_updates(struct btree_tra 1463 void bch2_dump_trans_updates(struct btree_trans *trans)
1477 { 1464 {
1478 struct printbuf buf = PRINTBUF; 1465 struct printbuf buf = PRINTBUF;
1479 1466
1480 bch2_trans_updates_to_text(&buf, tran 1467 bch2_trans_updates_to_text(&buf, trans);
1481 bch2_print_str(trans->c, buf.buf); !! 1468 bch2_print_string_as_lines(KERN_ERR, buf.buf);
1482 printbuf_exit(&buf); 1469 printbuf_exit(&buf);
1483 } 1470 }
1484 1471
1485 static void bch2_btree_path_to_text_short(str 1472 static void bch2_btree_path_to_text_short(struct printbuf *out, struct btree_trans *trans, btree_path_idx_t path_idx)
1486 { 1473 {
1487 struct btree_path *path = trans->path 1474 struct btree_path *path = trans->paths + path_idx;
1488 1475
1489 prt_printf(out, "path: idx %3u ref %u !! 1476 prt_printf(out, "path: idx %2u ref %u:%u %c %c %c btree=%s l=%u pos ",
1490 path_idx, path->ref, path- 1477 path_idx, path->ref, path->intent_ref,
1491 path->preserve ? 'P' : ' ' 1478 path->preserve ? 'P' : ' ',
1492 path->should_be_locked ? ' 1479 path->should_be_locked ? 'S' : ' ',
1493 path->cached ? 'C' : 'B', 1480 path->cached ? 'C' : 'B',
1494 bch2_btree_id_str(path->bt 1481 bch2_btree_id_str(path->btree_id),
1495 path->level); 1482 path->level);
1496 bch2_bpos_to_text(out, path->pos); 1483 bch2_bpos_to_text(out, path->pos);
1497 1484
1498 if (!path->cached && btree_node_locke <<
1499 prt_char(out, ' '); <<
1500 struct btree *b = path_l(path <<
1501 bch2_bpos_to_text(out, b->dat <<
1502 prt_char(out, '-'); <<
1503 bch2_bpos_to_text(out, b->key <<
1504 } <<
1505 <<
1506 #ifdef TRACK_PATH_ALLOCATED 1485 #ifdef TRACK_PATH_ALLOCATED
1507 prt_printf(out, " %pS", (void *) path 1486 prt_printf(out, " %pS", (void *) path->ip_allocated);
1508 #endif 1487 #endif
1509 } 1488 }
1510 1489
1511 static const char *btree_node_locked_str(enum 1490 static const char *btree_node_locked_str(enum btree_node_locked_type t)
1512 { 1491 {
1513 switch (t) { 1492 switch (t) {
1514 case BTREE_NODE_UNLOCKED: 1493 case BTREE_NODE_UNLOCKED:
1515 return "unlocked"; 1494 return "unlocked";
1516 case BTREE_NODE_READ_LOCKED: 1495 case BTREE_NODE_READ_LOCKED:
1517 return "read"; 1496 return "read";
1518 case BTREE_NODE_INTENT_LOCKED: 1497 case BTREE_NODE_INTENT_LOCKED:
1519 return "intent"; 1498 return "intent";
1520 case BTREE_NODE_WRITE_LOCKED: 1499 case BTREE_NODE_WRITE_LOCKED:
1521 return "write"; 1500 return "write";
1522 default: 1501 default:
1523 return NULL; 1502 return NULL;
1524 } 1503 }
1525 } 1504 }
1526 1505
1527 void bch2_btree_path_to_text(struct printbuf 1506 void bch2_btree_path_to_text(struct printbuf *out, struct btree_trans *trans, btree_path_idx_t path_idx)
1528 { 1507 {
1529 bch2_btree_path_to_text_short(out, tr 1508 bch2_btree_path_to_text_short(out, trans, path_idx);
1530 1509
1531 struct btree_path *path = trans->path 1510 struct btree_path *path = trans->paths + path_idx;
1532 1511
1533 prt_printf(out, " uptodate %u locks_w 1512 prt_printf(out, " uptodate %u locks_want %u", path->uptodate, path->locks_want);
1534 prt_newline(out); 1513 prt_newline(out);
1535 1514
1536 printbuf_indent_add(out, 2); 1515 printbuf_indent_add(out, 2);
1537 for (unsigned l = 0; l < BTREE_MAX_DE 1516 for (unsigned l = 0; l < BTREE_MAX_DEPTH; l++) {
1538 prt_printf(out, "l=%u locks % 1517 prt_printf(out, "l=%u locks %s seq %u node ", l,
1539 btree_node_locked_ 1518 btree_node_locked_str(btree_node_locked_type(path, l)),
1540 path->l[l].lock_se 1519 path->l[l].lock_seq);
1541 1520
1542 int ret = PTR_ERR_OR_ZERO(pat 1521 int ret = PTR_ERR_OR_ZERO(path->l[l].b);
1543 if (ret) 1522 if (ret)
1544 prt_str(out, bch2_err 1523 prt_str(out, bch2_err_str(ret));
1545 else 1524 else
1546 prt_printf(out, "%px" 1525 prt_printf(out, "%px", path->l[l].b);
1547 prt_newline(out); 1526 prt_newline(out);
1548 } 1527 }
1549 printbuf_indent_sub(out, 2); 1528 printbuf_indent_sub(out, 2);
1550 } 1529 }
1551 1530
1552 static noinline __cold 1531 static noinline __cold
1553 void __bch2_trans_paths_to_text(struct printb 1532 void __bch2_trans_paths_to_text(struct printbuf *out, struct btree_trans *trans,
1554 bool nosort) 1533 bool nosort)
1555 { 1534 {
1556 struct trans_for_each_path_inorder_it 1535 struct trans_for_each_path_inorder_iter iter;
1557 1536
1558 if (!nosort) 1537 if (!nosort)
1559 btree_trans_sort_paths(trans) 1538 btree_trans_sort_paths(trans);
1560 1539
1561 trans_for_each_path_idx_inorder(trans 1540 trans_for_each_path_idx_inorder(trans, iter) {
1562 bch2_btree_path_to_text_short 1541 bch2_btree_path_to_text_short(out, trans, iter.path_idx);
1563 prt_newline(out); 1542 prt_newline(out);
1564 } 1543 }
1565 } 1544 }
1566 1545
1567 noinline __cold 1546 noinline __cold
1568 void bch2_trans_paths_to_text(struct printbuf 1547 void bch2_trans_paths_to_text(struct printbuf *out, struct btree_trans *trans)
1569 { 1548 {
1570 __bch2_trans_paths_to_text(out, trans 1549 __bch2_trans_paths_to_text(out, trans, false);
1571 } 1550 }
1572 1551
1573 static noinline __cold 1552 static noinline __cold
1574 void __bch2_dump_trans_paths_updates(struct b 1553 void __bch2_dump_trans_paths_updates(struct btree_trans *trans, bool nosort)
1575 { 1554 {
1576 struct printbuf buf = PRINTBUF; 1555 struct printbuf buf = PRINTBUF;
1577 1556
1578 __bch2_trans_paths_to_text(&buf, tran 1557 __bch2_trans_paths_to_text(&buf, trans, nosort);
1579 bch2_trans_updates_to_text(&buf, tran 1558 bch2_trans_updates_to_text(&buf, trans);
1580 1559
1581 bch2_print_str(trans->c, buf.buf); !! 1560 bch2_print_string_as_lines(KERN_ERR, buf.buf);
1582 printbuf_exit(&buf); 1561 printbuf_exit(&buf);
1583 } 1562 }
1584 1563
1585 noinline __cold 1564 noinline __cold
1586 void bch2_dump_trans_paths_updates(struct btr 1565 void bch2_dump_trans_paths_updates(struct btree_trans *trans)
1587 { 1566 {
1588 __bch2_dump_trans_paths_updates(trans 1567 __bch2_dump_trans_paths_updates(trans, false);
1589 } 1568 }
1590 1569
1591 noinline __cold 1570 noinline __cold
1592 static void bch2_trans_update_max_paths(struc 1571 static void bch2_trans_update_max_paths(struct btree_trans *trans)
1593 { 1572 {
1594 struct btree_transaction_stats *s = b 1573 struct btree_transaction_stats *s = btree_trans_stats(trans);
1595 struct printbuf buf = PRINTBUF; 1574 struct printbuf buf = PRINTBUF;
1596 size_t nr = bitmap_weight(trans->path 1575 size_t nr = bitmap_weight(trans->paths_allocated, trans->nr_paths);
1597 1576
1598 bch2_trans_paths_to_text(&buf, trans) 1577 bch2_trans_paths_to_text(&buf, trans);
1599 1578
1600 if (!buf.allocation_failure) { 1579 if (!buf.allocation_failure) {
1601 mutex_lock(&s->lock); 1580 mutex_lock(&s->lock);
1602 if (nr > s->nr_max_paths) { 1581 if (nr > s->nr_max_paths) {
1603 s->nr_max_paths = nr; 1582 s->nr_max_paths = nr;
1604 swap(s->max_paths_tex 1583 swap(s->max_paths_text, buf.buf);
1605 } 1584 }
1606 mutex_unlock(&s->lock); 1585 mutex_unlock(&s->lock);
1607 } 1586 }
1608 1587
1609 printbuf_exit(&buf); 1588 printbuf_exit(&buf);
1610 1589
1611 trans->nr_paths_max = nr; 1590 trans->nr_paths_max = nr;
1612 } 1591 }
1613 1592
1614 noinline __cold 1593 noinline __cold
1615 int __bch2_btree_trans_too_many_iters(struct 1594 int __bch2_btree_trans_too_many_iters(struct btree_trans *trans)
1616 { 1595 {
1617 if (trace_trans_restart_too_many_iter 1596 if (trace_trans_restart_too_many_iters_enabled()) {
1618 struct printbuf buf = PRINTBU 1597 struct printbuf buf = PRINTBUF;
1619 1598
1620 bch2_trans_paths_to_text(&buf 1599 bch2_trans_paths_to_text(&buf, trans);
1621 trace_trans_restart_too_many_ 1600 trace_trans_restart_too_many_iters(trans, _THIS_IP_, buf.buf);
1622 printbuf_exit(&buf); 1601 printbuf_exit(&buf);
1623 } 1602 }
1624 1603
1625 count_event(trans->c, trans_restart_t 1604 count_event(trans->c, trans_restart_too_many_iters);
1626 1605
1627 return btree_trans_restart(trans, BCH 1606 return btree_trans_restart(trans, BCH_ERR_transaction_restart_too_many_iters);
1628 } 1607 }
1629 1608
1630 static noinline void btree_path_overflow(stru 1609 static noinline void btree_path_overflow(struct btree_trans *trans)
1631 { 1610 {
1632 bch2_dump_trans_paths_updates(trans); 1611 bch2_dump_trans_paths_updates(trans);
1633 bch_err(trans->c, "trans path overflo 1612 bch_err(trans->c, "trans path overflow");
1634 } 1613 }
1635 1614
1636 static noinline void btree_paths_realloc(stru 1615 static noinline void btree_paths_realloc(struct btree_trans *trans)
1637 { 1616 {
1638 unsigned nr = trans->nr_paths * 2; 1617 unsigned nr = trans->nr_paths * 2;
1639 1618
1640 void *p = kvzalloc(BITS_TO_LONGS(nr) 1619 void *p = kvzalloc(BITS_TO_LONGS(nr) * sizeof(unsigned long) +
1641 sizeof(struct btree 1620 sizeof(struct btree_trans_paths) +
1642 nr * sizeof(struct 1621 nr * sizeof(struct btree_path) +
1643 nr * sizeof(btree_p 1622 nr * sizeof(btree_path_idx_t) + 8 +
1644 nr * sizeof(struct 1623 nr * sizeof(struct btree_insert_entry), GFP_KERNEL|__GFP_NOFAIL);
1645 1624
1646 unsigned long *paths_allocated = p; 1625 unsigned long *paths_allocated = p;
1647 memcpy(paths_allocated, trans->paths_ 1626 memcpy(paths_allocated, trans->paths_allocated, BITS_TO_LONGS(trans->nr_paths) * sizeof(unsigned long));
1648 p += BITS_TO_LONGS(nr) * sizeof(unsig 1627 p += BITS_TO_LONGS(nr) * sizeof(unsigned long);
1649 1628
1650 p += sizeof(struct btree_trans_paths) 1629 p += sizeof(struct btree_trans_paths);
1651 struct btree_path *paths = p; 1630 struct btree_path *paths = p;
1652 *trans_paths_nr(paths) = nr; 1631 *trans_paths_nr(paths) = nr;
1653 memcpy(paths, trans->paths, trans->nr 1632 memcpy(paths, trans->paths, trans->nr_paths * sizeof(struct btree_path));
1654 p += nr * sizeof(struct btree_path); 1633 p += nr * sizeof(struct btree_path);
1655 1634
1656 btree_path_idx_t *sorted = p; 1635 btree_path_idx_t *sorted = p;
1657 memcpy(sorted, trans->sorted, trans-> 1636 memcpy(sorted, trans->sorted, trans->nr_sorted * sizeof(btree_path_idx_t));
1658 p += nr * sizeof(btree_path_idx_t) + 1637 p += nr * sizeof(btree_path_idx_t) + 8;
1659 1638
1660 struct btree_insert_entry *updates = 1639 struct btree_insert_entry *updates = p;
1661 memcpy(updates, trans->updates, trans 1640 memcpy(updates, trans->updates, trans->nr_paths * sizeof(struct btree_insert_entry));
1662 1641
1663 unsigned long *old = trans->paths_all 1642 unsigned long *old = trans->paths_allocated;
1664 1643
1665 rcu_assign_pointer(trans->paths_alloc 1644 rcu_assign_pointer(trans->paths_allocated, paths_allocated);
1666 rcu_assign_pointer(trans->paths, 1645 rcu_assign_pointer(trans->paths, paths);
1667 rcu_assign_pointer(trans->sorted, 1646 rcu_assign_pointer(trans->sorted, sorted);
1668 rcu_assign_pointer(trans->updates, 1647 rcu_assign_pointer(trans->updates, updates);
1669 1648
1670 trans->nr_paths = nr; 1649 trans->nr_paths = nr;
1671 1650
1672 if (old != trans->_paths_allocated) 1651 if (old != trans->_paths_allocated)
1673 kfree_rcu_mightsleep(old); 1652 kfree_rcu_mightsleep(old);
1674 } 1653 }
1675 1654
1676 static inline btree_path_idx_t btree_path_all 1655 static inline btree_path_idx_t btree_path_alloc(struct btree_trans *trans,
1677 1656 btree_path_idx_t pos)
1678 { 1657 {
1679 btree_path_idx_t idx = find_first_zer 1658 btree_path_idx_t idx = find_first_zero_bit(trans->paths_allocated, trans->nr_paths);
1680 1659
1681 if (unlikely(idx == trans->nr_paths)) 1660 if (unlikely(idx == trans->nr_paths)) {
1682 if (trans->nr_paths == BTREE_ 1661 if (trans->nr_paths == BTREE_ITER_MAX) {
1683 btree_path_overflow(t 1662 btree_path_overflow(trans);
1684 return 0; 1663 return 0;
1685 } 1664 }
1686 1665
1687 btree_paths_realloc(trans); 1666 btree_paths_realloc(trans);
1688 } 1667 }
1689 1668
1690 /* 1669 /*
1691 * Do this before marking the new pat 1670 * Do this before marking the new path as allocated, since it won't be
1692 * initialized yet: 1671 * initialized yet:
1693 */ 1672 */
1694 if (unlikely(idx > trans->nr_paths_ma 1673 if (unlikely(idx > trans->nr_paths_max))
1695 bch2_trans_update_max_paths(t 1674 bch2_trans_update_max_paths(trans);
1696 1675
1697 __set_bit(idx, trans->paths_allocated 1676 __set_bit(idx, trans->paths_allocated);
1698 1677
1699 struct btree_path *path = &trans->pat 1678 struct btree_path *path = &trans->paths[idx];
1700 path->ref = 0; 1679 path->ref = 0;
1701 path->intent_ref = 0; 1680 path->intent_ref = 0;
1702 path->nodes_locked = 0; 1681 path->nodes_locked = 0;
1703 1682
1704 btree_path_list_add(trans, pos, idx); 1683 btree_path_list_add(trans, pos, idx);
1705 trans->paths_sorted = false; 1684 trans->paths_sorted = false;
1706 return idx; 1685 return idx;
1707 } 1686 }
1708 1687
1709 btree_path_idx_t bch2_path_get(struct btree_t 1688 btree_path_idx_t bch2_path_get(struct btree_trans *trans,
1710 enum btree_id bt 1689 enum btree_id btree_id, struct bpos pos,
1711 unsigned locks_w 1690 unsigned locks_want, unsigned level,
1712 unsigned flags, 1691 unsigned flags, unsigned long ip)
1713 { 1692 {
1714 struct btree_path *path; 1693 struct btree_path *path;
1715 bool cached = flags & BTREE_ITER_cach 1694 bool cached = flags & BTREE_ITER_cached;
1716 bool intent = flags & BTREE_ITER_inte 1695 bool intent = flags & BTREE_ITER_intent;
1717 struct trans_for_each_path_inorder_it 1696 struct trans_for_each_path_inorder_iter iter;
1718 btree_path_idx_t path_pos = 0, path_i 1697 btree_path_idx_t path_pos = 0, path_idx;
1719 1698
1720 bch2_trans_verify_not_unlocked(trans) 1699 bch2_trans_verify_not_unlocked(trans);
1721 bch2_trans_verify_not_in_restart(tran 1700 bch2_trans_verify_not_in_restart(trans);
1722 bch2_trans_verify_locks(trans); 1701 bch2_trans_verify_locks(trans);
1723 1702
1724 btree_trans_sort_paths(trans); 1703 btree_trans_sort_paths(trans);
1725 1704
1726 trans_for_each_path_inorder(trans, pa 1705 trans_for_each_path_inorder(trans, path, iter) {
1727 if (__btree_path_cmp(path, 1706 if (__btree_path_cmp(path,
1728 btree_id 1707 btree_id,
1729 cached, 1708 cached,
1730 pos, 1709 pos,
1731 level) > 1710 level) > 0)
1732 break; 1711 break;
1733 1712
1734 path_pos = iter.path_idx; 1713 path_pos = iter.path_idx;
1735 } 1714 }
1736 1715
1737 if (path_pos && 1716 if (path_pos &&
1738 trans->paths[path_pos].cached 1717 trans->paths[path_pos].cached == cached &&
1739 trans->paths[path_pos].btree_id 1718 trans->paths[path_pos].btree_id == btree_id &&
1740 trans->paths[path_pos].level 1719 trans->paths[path_pos].level == level) {
1741 trace_btree_path_get(trans, t !! 1720 __btree_path_get(trans->paths + path_pos, intent);
1742 <<
1743 __btree_path_get(trans, trans <<
1744 path_idx = bch2_btree_path_se 1721 path_idx = bch2_btree_path_set_pos(trans, path_pos, pos, intent, ip);
1745 path = trans->paths + path_id 1722 path = trans->paths + path_idx;
1746 } else { 1723 } else {
1747 path_idx = btree_path_alloc(t 1724 path_idx = btree_path_alloc(trans, path_pos);
1748 path = trans->paths + path_id 1725 path = trans->paths + path_idx;
1749 1726
1750 __btree_path_get(trans, path, !! 1727 __btree_path_get(path, intent);
1751 path->pos 1728 path->pos = pos;
1752 path->btree_id 1729 path->btree_id = btree_id;
1753 path->cached 1730 path->cached = cached;
1754 path->uptodate 1731 path->uptodate = BTREE_ITER_NEED_TRAVERSE;
1755 path->should_be_locked 1732 path->should_be_locked = false;
1756 path->level 1733 path->level = level;
1757 path->locks_want 1734 path->locks_want = locks_want;
1758 path->nodes_locked 1735 path->nodes_locked = 0;
1759 for (unsigned i = 0; i < ARRA 1736 for (unsigned i = 0; i < ARRAY_SIZE(path->l); i++)
1760 path->l[i].b 1737 path->l[i].b = ERR_PTR(-BCH_ERR_no_btree_node_init);
1761 #ifdef TRACK_PATH_ALLOCATED 1738 #ifdef TRACK_PATH_ALLOCATED
1762 path->ip_allocated 1739 path->ip_allocated = ip;
1763 #endif 1740 #endif
1764 trans->paths_sorted 1741 trans->paths_sorted = false;
1765 <<
1766 trace_btree_path_alloc(trans, <<
1767 } 1742 }
1768 1743
1769 if (!(flags & BTREE_ITER_nopreserve)) 1744 if (!(flags & BTREE_ITER_nopreserve))
1770 path->preserve = true; 1745 path->preserve = true;
1771 1746
1772 if (path->intent_ref) 1747 if (path->intent_ref)
1773 locks_want = max(locks_want, 1748 locks_want = max(locks_want, level + 1);
1774 1749
1775 /* 1750 /*
1776 * If the path has locks_want greater 1751 * If the path has locks_want greater than requested, we don't downgrade
1777 * it here - on transaction restart b 1752 * it here - on transaction restart because btree node split needs to
1778 * upgrade locks, we might be putting 1753 * upgrade locks, we might be putting/getting the iterator again.
1779 * Downgrading iterators only happens 1754 * Downgrading iterators only happens via bch2_trans_downgrade(), after
1780 * a successful transaction commit. 1755 * a successful transaction commit.
1781 */ 1756 */
1782 1757
1783 locks_want = min(locks_want, BTREE_MA 1758 locks_want = min(locks_want, BTREE_MAX_DEPTH);
1784 if (locks_want > path->locks_want) 1759 if (locks_want > path->locks_want)
1785 bch2_btree_path_upgrade_noupg 1760 bch2_btree_path_upgrade_noupgrade_sibs(trans, path, locks_want, NULL);
1786 1761
1787 return path_idx; 1762 return path_idx;
1788 } 1763 }
1789 1764
1790 btree_path_idx_t bch2_path_get_unlocked_mut(s 1765 btree_path_idx_t bch2_path_get_unlocked_mut(struct btree_trans *trans,
1791 e 1766 enum btree_id btree_id,
1792 u 1767 unsigned level,
1793 s 1768 struct bpos pos)
1794 { 1769 {
1795 btree_path_idx_t path_idx = bch2_path 1770 btree_path_idx_t path_idx = bch2_path_get(trans, btree_id, pos, level + 1, level,
1796 BTREE_ITER_nopre 1771 BTREE_ITER_nopreserve|
1797 BTREE_ITER_inten 1772 BTREE_ITER_intent, _RET_IP_);
1798 path_idx = bch2_btree_path_make_mut(t 1773 path_idx = bch2_btree_path_make_mut(trans, path_idx, true, _RET_IP_);
1799 1774
1800 struct btree_path *path = trans->path 1775 struct btree_path *path = trans->paths + path_idx;
1801 bch2_btree_path_downgrade(trans, path 1776 bch2_btree_path_downgrade(trans, path);
1802 __bch2_btree_path_unlock(trans, path) 1777 __bch2_btree_path_unlock(trans, path);
1803 return path_idx; 1778 return path_idx;
1804 } 1779 }
1805 1780
1806 struct bkey_s_c bch2_btree_path_peek_slot(str 1781 struct bkey_s_c bch2_btree_path_peek_slot(struct btree_path *path, struct bkey *u)
1807 { 1782 {
1808 1783
1809 struct btree_path_level *l = path_l(p 1784 struct btree_path_level *l = path_l(path);
1810 struct bkey_packed *_k; 1785 struct bkey_packed *_k;
1811 struct bkey_s_c k; 1786 struct bkey_s_c k;
1812 1787
1813 if (unlikely(!l->b)) 1788 if (unlikely(!l->b))
1814 return bkey_s_c_null; 1789 return bkey_s_c_null;
1815 1790
1816 EBUG_ON(path->uptodate != BTREE_ITER_ 1791 EBUG_ON(path->uptodate != BTREE_ITER_UPTODATE);
1817 EBUG_ON(!btree_node_locked(path, path 1792 EBUG_ON(!btree_node_locked(path, path->level));
1818 1793
1819 if (!path->cached) { 1794 if (!path->cached) {
1820 _k = bch2_btree_node_iter_pee 1795 _k = bch2_btree_node_iter_peek_all(&l->iter, l->b);
1821 k = _k ? bkey_disassemble(l-> 1796 k = _k ? bkey_disassemble(l->b, _k, u) : bkey_s_c_null;
1822 1797
1823 EBUG_ON(k.k && bkey_deleted(k 1798 EBUG_ON(k.k && bkey_deleted(k.k) && bpos_eq(k.k->p, path->pos));
1824 1799
1825 if (!k.k || !bpos_eq(path->po 1800 if (!k.k || !bpos_eq(path->pos, k.k->p))
1826 goto hole; 1801 goto hole;
1827 } else { 1802 } else {
1828 struct bkey_cached *ck = (voi 1803 struct bkey_cached *ck = (void *) path->l[0].b;
1829 if (!ck) <<
1830 return bkey_s_c_null; <<
1831 1804
1832 EBUG_ON(path->btree_id != ck- !! 1805 EBUG_ON(ck &&
1833 !bkey_eq(path->pos, c !! 1806 (path->btree_id != ck->key.btree_id ||
>> 1807 !bkey_eq(path->pos, ck->key.pos)));
>> 1808 if (!ck || !ck->valid)
>> 1809 return bkey_s_c_null;
1834 1810
1835 *u = ck->k->k; 1811 *u = ck->k->k;
1836 k = bkey_i_to_s_c(ck->k); 1812 k = bkey_i_to_s_c(ck->k);
1837 } 1813 }
1838 1814
1839 return k; 1815 return k;
1840 hole: 1816 hole:
1841 bkey_init(u); 1817 bkey_init(u);
1842 u->p = path->pos; 1818 u->p = path->pos;
1843 return (struct bkey_s_c) { u, NULL }; 1819 return (struct bkey_s_c) { u, NULL };
1844 } 1820 }
1845 1821
1846 1822
1847 void bch2_set_btree_iter_dontneed(struct btre 1823 void bch2_set_btree_iter_dontneed(struct btree_iter *iter)
1848 { 1824 {
1849 struct btree_trans *trans = iter->tra 1825 struct btree_trans *trans = iter->trans;
1850 1826
1851 if (!iter->path || trans->restarted) 1827 if (!iter->path || trans->restarted)
1852 return; 1828 return;
1853 1829
1854 struct btree_path *path = btree_iter_ 1830 struct btree_path *path = btree_iter_path(trans, iter);
1855 path->preserve = false; 1831 path->preserve = false;
1856 if (path->ref == 1) 1832 if (path->ref == 1)
1857 path->should_be_locked = fal 1833 path->should_be_locked = false;
1858 } 1834 }
1859 /* Btree iterators: */ 1835 /* Btree iterators: */
1860 1836
1861 int __must_check 1837 int __must_check
1862 __bch2_btree_iter_traverse(struct btree_iter 1838 __bch2_btree_iter_traverse(struct btree_iter *iter)
1863 { 1839 {
1864 return bch2_btree_path_traverse(iter- 1840 return bch2_btree_path_traverse(iter->trans, iter->path, iter->flags);
1865 } 1841 }
1866 1842
1867 int __must_check 1843 int __must_check
1868 bch2_btree_iter_traverse(struct btree_iter *i 1844 bch2_btree_iter_traverse(struct btree_iter *iter)
1869 { 1845 {
1870 struct btree_trans *trans = iter->tra 1846 struct btree_trans *trans = iter->trans;
1871 int ret; 1847 int ret;
1872 1848
1873 bch2_trans_verify_not_unlocked(trans) 1849 bch2_trans_verify_not_unlocked(trans);
1874 1850
1875 iter->path = bch2_btree_path_set_pos( 1851 iter->path = bch2_btree_path_set_pos(trans, iter->path,
1876 btree 1852 btree_iter_search_key(iter),
1877 iter- 1853 iter->flags & BTREE_ITER_intent,
1878 btree 1854 btree_iter_ip_allocated(iter));
1879 1855
1880 ret = bch2_btree_path_traverse(iter-> 1856 ret = bch2_btree_path_traverse(iter->trans, iter->path, iter->flags);
1881 if (ret) 1857 if (ret)
1882 return ret; 1858 return ret;
1883 1859
1884 struct btree_path *path = btree_iter_ 1860 struct btree_path *path = btree_iter_path(trans, iter);
1885 if (btree_path_node(path, path->level 1861 if (btree_path_node(path, path->level))
1886 btree_path_set_should_be_lock !! 1862 btree_path_set_should_be_locked(path);
1887 return 0; 1863 return 0;
1888 } 1864 }
1889 1865
1890 /* Iterate across nodes (leaf and interior no 1866 /* Iterate across nodes (leaf and interior nodes) */
1891 1867
1892 struct btree *bch2_btree_iter_peek_node(struc 1868 struct btree *bch2_btree_iter_peek_node(struct btree_iter *iter)
1893 { 1869 {
1894 struct btree_trans *trans = iter->tra 1870 struct btree_trans *trans = iter->trans;
1895 struct btree *b = NULL; 1871 struct btree *b = NULL;
1896 int ret; 1872 int ret;
1897 1873
1898 EBUG_ON(trans->paths[iter->path].cach 1874 EBUG_ON(trans->paths[iter->path].cached);
1899 bch2_btree_iter_verify(iter); 1875 bch2_btree_iter_verify(iter);
1900 1876
1901 ret = bch2_btree_path_traverse(trans, 1877 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
1902 if (ret) 1878 if (ret)
1903 goto err; 1879 goto err;
1904 1880
1905 struct btree_path *path = btree_iter_ 1881 struct btree_path *path = btree_iter_path(trans, iter);
1906 b = btree_path_node(path, path->level 1882 b = btree_path_node(path, path->level);
1907 if (!b) 1883 if (!b)
1908 goto out; 1884 goto out;
1909 1885
1910 BUG_ON(bpos_lt(b->key.k.p, iter->pos) 1886 BUG_ON(bpos_lt(b->key.k.p, iter->pos));
1911 1887
1912 bkey_init(&iter->k); 1888 bkey_init(&iter->k);
1913 iter->k.p = iter->pos = b->key.k.p; 1889 iter->k.p = iter->pos = b->key.k.p;
1914 1890
1915 iter->path = bch2_btree_path_set_pos( 1891 iter->path = bch2_btree_path_set_pos(trans, iter->path, b->key.k.p,
1916 iter- 1892 iter->flags & BTREE_ITER_intent,
1917 btree 1893 btree_iter_ip_allocated(iter));
1918 btree_path_set_should_be_locked(trans !! 1894 btree_path_set_should_be_locked(btree_iter_path(trans, iter));
1919 out: 1895 out:
1920 bch2_btree_iter_verify_entry_exit(ite 1896 bch2_btree_iter_verify_entry_exit(iter);
1921 bch2_btree_iter_verify(iter); 1897 bch2_btree_iter_verify(iter);
1922 1898
1923 return b; 1899 return b;
1924 err: 1900 err:
1925 b = ERR_PTR(ret); 1901 b = ERR_PTR(ret);
1926 goto out; 1902 goto out;
1927 } 1903 }
1928 1904
1929 /* Only kept for -tools */ <<
1930 struct btree *bch2_btree_iter_peek_node_and_r 1905 struct btree *bch2_btree_iter_peek_node_and_restart(struct btree_iter *iter)
1931 { 1906 {
1932 struct btree *b; 1907 struct btree *b;
1933 1908
1934 while (b = bch2_btree_iter_peek_node( 1909 while (b = bch2_btree_iter_peek_node(iter),
1935 bch2_err_matches(PTR_ERR_OR_ZE 1910 bch2_err_matches(PTR_ERR_OR_ZERO(b), BCH_ERR_transaction_restart))
1936 bch2_trans_begin(iter->trans) 1911 bch2_trans_begin(iter->trans);
1937 1912
1938 return b; 1913 return b;
1939 } 1914 }
1940 1915
1941 struct btree *bch2_btree_iter_next_node(struc 1916 struct btree *bch2_btree_iter_next_node(struct btree_iter *iter)
1942 { 1917 {
1943 struct btree_trans *trans = iter->tra 1918 struct btree_trans *trans = iter->trans;
1944 struct btree *b = NULL; 1919 struct btree *b = NULL;
1945 int ret; 1920 int ret;
1946 1921
1947 EBUG_ON(trans->paths[iter->path].cach 1922 EBUG_ON(trans->paths[iter->path].cached);
1948 bch2_trans_verify_not_in_restart(tran 1923 bch2_trans_verify_not_in_restart(trans);
1949 bch2_btree_iter_verify(iter); 1924 bch2_btree_iter_verify(iter);
1950 1925
1951 ret = bch2_btree_path_traverse(trans, <<
1952 if (ret) <<
1953 goto err; <<
1954 <<
1955 <<
1956 struct btree_path *path = btree_iter_ 1926 struct btree_path *path = btree_iter_path(trans, iter);
1957 1927
1958 /* already at end? */ 1928 /* already at end? */
1959 if (!btree_path_node(path, path->leve 1929 if (!btree_path_node(path, path->level))
1960 return NULL; 1930 return NULL;
1961 1931
1962 /* got to end? */ 1932 /* got to end? */
1963 if (!btree_path_node(path, path->leve 1933 if (!btree_path_node(path, path->level + 1)) {
1964 btree_path_set_level_up(trans 1934 btree_path_set_level_up(trans, path);
1965 return NULL; 1935 return NULL;
1966 } 1936 }
1967 1937
1968 if (!bch2_btree_node_relock(trans, pa 1938 if (!bch2_btree_node_relock(trans, path, path->level + 1)) {
1969 __bch2_btree_path_unlock(tran 1939 __bch2_btree_path_unlock(trans, path);
1970 path->l[path->level].b 1940 path->l[path->level].b = ERR_PTR(-BCH_ERR_no_btree_node_relock);
1971 path->l[path->level + 1].b 1941 path->l[path->level + 1].b = ERR_PTR(-BCH_ERR_no_btree_node_relock);
1972 btree_path_set_dirty(path, BT 1942 btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE);
1973 trace_and_count(trans->c, tra 1943 trace_and_count(trans->c, trans_restart_relock_next_node, trans, _THIS_IP_, path);
1974 ret = btree_trans_restart(tra 1944 ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_relock);
1975 goto err; 1945 goto err;
1976 } 1946 }
1977 1947
1978 b = btree_path_node(path, path->level 1948 b = btree_path_node(path, path->level + 1);
1979 1949
1980 if (bpos_eq(iter->pos, b->key.k.p)) { 1950 if (bpos_eq(iter->pos, b->key.k.p)) {
1981 __btree_path_set_level_up(tra 1951 __btree_path_set_level_up(trans, path, path->level++);
1982 } else { 1952 } else {
1983 if (btree_lock_want(path, pat 1953 if (btree_lock_want(path, path->level + 1) == BTREE_NODE_UNLOCKED)
1984 btree_node_unlock(tra 1954 btree_node_unlock(trans, path, path->level + 1);
1985 1955
1986 /* 1956 /*
1987 * Haven't gotten to the end 1957 * Haven't gotten to the end of the parent node: go back down to
1988 * the next child node 1958 * the next child node
1989 */ 1959 */
1990 iter->path = bch2_btree_path_ 1960 iter->path = bch2_btree_path_set_pos(trans, iter->path,
1991 bpos_ 1961 bpos_successor(iter->pos),
1992 iter- 1962 iter->flags & BTREE_ITER_intent,
1993 btree 1963 btree_iter_ip_allocated(iter));
1994 1964
1995 path = btree_iter_path(trans, 1965 path = btree_iter_path(trans, iter);
1996 btree_path_set_level_down(tra 1966 btree_path_set_level_down(trans, path, iter->min_depth);
1997 1967
1998 ret = bch2_btree_path_travers 1968 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
1999 if (ret) 1969 if (ret)
2000 goto err; 1970 goto err;
2001 1971
2002 path = btree_iter_path(trans, 1972 path = btree_iter_path(trans, iter);
2003 b = path->l[path->level].b; 1973 b = path->l[path->level].b;
2004 } 1974 }
2005 1975
2006 bkey_init(&iter->k); 1976 bkey_init(&iter->k);
2007 iter->k.p = iter->pos = b->key.k.p; 1977 iter->k.p = iter->pos = b->key.k.p;
2008 1978
2009 iter->path = bch2_btree_path_set_pos( 1979 iter->path = bch2_btree_path_set_pos(trans, iter->path, b->key.k.p,
2010 iter- 1980 iter->flags & BTREE_ITER_intent,
2011 btree 1981 btree_iter_ip_allocated(iter));
2012 btree_path_set_should_be_locked(trans !! 1982 btree_path_set_should_be_locked(btree_iter_path(trans, iter));
2013 EBUG_ON(btree_iter_path(trans, iter)- 1983 EBUG_ON(btree_iter_path(trans, iter)->uptodate);
2014 out: 1984 out:
2015 bch2_btree_iter_verify_entry_exit(ite 1985 bch2_btree_iter_verify_entry_exit(iter);
2016 bch2_btree_iter_verify(iter); 1986 bch2_btree_iter_verify(iter);
2017 1987
2018 return b; 1988 return b;
2019 err: 1989 err:
2020 b = ERR_PTR(ret); 1990 b = ERR_PTR(ret);
2021 goto out; 1991 goto out;
2022 } 1992 }
2023 1993
2024 /* Iterate across keys (in leaf nodes only) * 1994 /* Iterate across keys (in leaf nodes only) */
2025 1995
2026 inline bool bch2_btree_iter_advance(struct bt 1996 inline bool bch2_btree_iter_advance(struct btree_iter *iter)
2027 { 1997 {
2028 struct bpos pos = iter->k.p; 1998 struct bpos pos = iter->k.p;
2029 bool ret = !(iter->flags & BTREE_ITER 1999 bool ret = !(iter->flags & BTREE_ITER_all_snapshots
2030 ? bpos_eq(pos, SPOS_MAX) 2000 ? bpos_eq(pos, SPOS_MAX)
2031 : bkey_eq(pos, SPOS_MAX) 2001 : bkey_eq(pos, SPOS_MAX));
2032 2002
2033 if (ret && !(iter->flags & BTREE_ITER 2003 if (ret && !(iter->flags & BTREE_ITER_is_extents))
2034 pos = bkey_successor(iter, po 2004 pos = bkey_successor(iter, pos);
2035 bch2_btree_iter_set_pos(iter, pos); 2005 bch2_btree_iter_set_pos(iter, pos);
2036 return ret; 2006 return ret;
2037 } 2007 }
2038 2008
2039 inline bool bch2_btree_iter_rewind(struct btr 2009 inline bool bch2_btree_iter_rewind(struct btree_iter *iter)
2040 { 2010 {
2041 struct bpos pos = bkey_start_pos(&ite 2011 struct bpos pos = bkey_start_pos(&iter->k);
2042 bool ret = !(iter->flags & BTREE_ITER 2012 bool ret = !(iter->flags & BTREE_ITER_all_snapshots
2043 ? bpos_eq(pos, POS_MIN) 2013 ? bpos_eq(pos, POS_MIN)
2044 : bkey_eq(pos, POS_MIN)) 2014 : bkey_eq(pos, POS_MIN));
2045 2015
2046 if (ret && !(iter->flags & BTREE_ITER 2016 if (ret && !(iter->flags & BTREE_ITER_is_extents))
2047 pos = bkey_predecessor(iter, 2017 pos = bkey_predecessor(iter, pos);
2048 bch2_btree_iter_set_pos(iter, pos); 2018 bch2_btree_iter_set_pos(iter, pos);
2049 return ret; 2019 return ret;
2050 } 2020 }
2051 2021
2052 static noinline 2022 static noinline
2053 void bch2_btree_trans_peek_prev_updates(struc 2023 void bch2_btree_trans_peek_prev_updates(struct btree_trans *trans, struct btree_iter *iter,
2054 struc 2024 struct bkey_s_c *k)
2055 { 2025 {
2056 struct bpos end = path_l(btree_iter_p 2026 struct bpos end = path_l(btree_iter_path(trans, iter))->b->data->min_key;
2057 2027
2058 trans_for_each_update(trans, i) 2028 trans_for_each_update(trans, i)
2059 if (!i->key_cache_already_flu 2029 if (!i->key_cache_already_flushed &&
2060 i->btree_id == iter->btre 2030 i->btree_id == iter->btree_id &&
2061 bpos_le(i->k->k.p, iter-> 2031 bpos_le(i->k->k.p, iter->pos) &&
2062 bpos_ge(i->k->k.p, k->k ? 2032 bpos_ge(i->k->k.p, k->k ? k->k->p : end)) {
2063 iter->k = i->k->k; 2033 iter->k = i->k->k;
2064 *k = bkey_i_to_s_c(i- 2034 *k = bkey_i_to_s_c(i->k);
2065 } 2035 }
2066 } 2036 }
2067 2037
2068 static noinline 2038 static noinline
2069 void bch2_btree_trans_peek_updates(struct btr 2039 void bch2_btree_trans_peek_updates(struct btree_trans *trans, struct btree_iter *iter,
2070 struct bke 2040 struct bkey_s_c *k)
2071 { 2041 {
2072 struct btree_path *path = btree_iter_ 2042 struct btree_path *path = btree_iter_path(trans, iter);
2073 struct bpos end = path_l(path)->b->ke 2043 struct bpos end = path_l(path)->b->key.k.p;
2074 2044
2075 trans_for_each_update(trans, i) 2045 trans_for_each_update(trans, i)
2076 if (!i->key_cache_already_flu 2046 if (!i->key_cache_already_flushed &&
2077 i->btree_id == iter->btre 2047 i->btree_id == iter->btree_id &&
2078 bpos_ge(i->k->k.p, path-> 2048 bpos_ge(i->k->k.p, path->pos) &&
2079 bpos_le(i->k->k.p, k->k ? 2049 bpos_le(i->k->k.p, k->k ? k->k->p : end)) {
2080 iter->k = i->k->k; 2050 iter->k = i->k->k;
2081 *k = bkey_i_to_s_c(i- 2051 *k = bkey_i_to_s_c(i->k);
2082 } 2052 }
2083 } 2053 }
2084 2054
2085 static noinline 2055 static noinline
2086 void bch2_btree_trans_peek_slot_updates(struc 2056 void bch2_btree_trans_peek_slot_updates(struct btree_trans *trans, struct btree_iter *iter,
2087 struc 2057 struct bkey_s_c *k)
2088 { 2058 {
2089 trans_for_each_update(trans, i) 2059 trans_for_each_update(trans, i)
2090 if (!i->key_cache_already_flu 2060 if (!i->key_cache_already_flushed &&
2091 i->btree_id == iter->btre 2061 i->btree_id == iter->btree_id &&
2092 bpos_eq(i->k->k.p, iter-> 2062 bpos_eq(i->k->k.p, iter->pos)) {
2093 iter->k = i->k->k; 2063 iter->k = i->k->k;
2094 *k = bkey_i_to_s_c(i- 2064 *k = bkey_i_to_s_c(i->k);
2095 } 2065 }
2096 } 2066 }
2097 2067
2098 static struct bkey_i *bch2_btree_journal_peek 2068 static struct bkey_i *bch2_btree_journal_peek(struct btree_trans *trans,
2099 2069 struct btree_iter *iter,
2100 2070 struct bpos end_pos)
2101 { 2071 {
2102 struct btree_path *path = btree_iter_ 2072 struct btree_path *path = btree_iter_path(trans, iter);
2103 2073
2104 return bch2_journal_keys_peek_upto(tr 2074 return bch2_journal_keys_peek_upto(trans->c, iter->btree_id,
2105 pa 2075 path->level,
2106 pa 2076 path->pos,
2107 en 2077 end_pos,
2108 &i 2078 &iter->journal_idx);
2109 } 2079 }
2110 2080
2111 static noinline 2081 static noinline
2112 struct bkey_s_c btree_trans_peek_slot_journal 2082 struct bkey_s_c btree_trans_peek_slot_journal(struct btree_trans *trans,
2113 2083 struct btree_iter *iter)
2114 { 2084 {
2115 struct btree_path *path = btree_iter_ 2085 struct btree_path *path = btree_iter_path(trans, iter);
2116 struct bkey_i *k = bch2_btree_journal 2086 struct bkey_i *k = bch2_btree_journal_peek(trans, iter, path->pos);
2117 2087
2118 if (k) { 2088 if (k) {
2119 iter->k = k->k; 2089 iter->k = k->k;
2120 return bkey_i_to_s_c(k); 2090 return bkey_i_to_s_c(k);
2121 } else { 2091 } else {
2122 return bkey_s_c_null; 2092 return bkey_s_c_null;
2123 } 2093 }
2124 } 2094 }
2125 2095
2126 static noinline 2096 static noinline
2127 struct bkey_s_c btree_trans_peek_journal(stru 2097 struct bkey_s_c btree_trans_peek_journal(struct btree_trans *trans,
2128 stru 2098 struct btree_iter *iter,
2129 stru 2099 struct bkey_s_c k)
2130 { 2100 {
2131 struct btree_path *path = btree_iter_ 2101 struct btree_path *path = btree_iter_path(trans, iter);
2132 struct bkey_i *next_journal = 2102 struct bkey_i *next_journal =
2133 bch2_btree_journal_peek(trans 2103 bch2_btree_journal_peek(trans, iter,
2134 k.k ? k.k->p 2104 k.k ? k.k->p : path_l(path)->b->key.k.p);
2135 2105
2136 if (next_journal) { 2106 if (next_journal) {
2137 iter->k = next_journal->k; 2107 iter->k = next_journal->k;
2138 k = bkey_i_to_s_c(next_journa 2108 k = bkey_i_to_s_c(next_journal);
2139 } 2109 }
2140 2110
2141 return k; 2111 return k;
2142 } 2112 }
2143 2113
2144 /* 2114 /*
2145 * Checks btree key cache for key at iter->po 2115 * Checks btree key cache for key at iter->pos and returns it if present, or
2146 * bkey_s_c_null: 2116 * bkey_s_c_null:
2147 */ 2117 */
2148 static noinline 2118 static noinline
2149 struct bkey_s_c btree_trans_peek_key_cache(st 2119 struct bkey_s_c btree_trans_peek_key_cache(struct btree_iter *iter, struct bpos pos)
2150 { 2120 {
2151 struct btree_trans *trans = iter->tra 2121 struct btree_trans *trans = iter->trans;
2152 struct bch_fs *c = trans->c; 2122 struct bch_fs *c = trans->c;
2153 struct bkey u; 2123 struct bkey u;
2154 struct bkey_s_c k; 2124 struct bkey_s_c k;
2155 int ret; 2125 int ret;
2156 2126
2157 bch2_trans_verify_not_in_restart(tran 2127 bch2_trans_verify_not_in_restart(trans);
2158 bch2_trans_verify_not_unlocked(trans) 2128 bch2_trans_verify_not_unlocked(trans);
2159 2129
2160 if ((iter->flags & BTREE_ITER_key_cac 2130 if ((iter->flags & BTREE_ITER_key_cache_fill) &&
2161 bpos_eq(iter->pos, pos)) 2131 bpos_eq(iter->pos, pos))
2162 return bkey_s_c_null; 2132 return bkey_s_c_null;
2163 2133
2164 if (!bch2_btree_key_cache_find(c, ite 2134 if (!bch2_btree_key_cache_find(c, iter->btree_id, pos))
2165 return bkey_s_c_null; 2135 return bkey_s_c_null;
2166 2136
2167 if (!iter->key_cache_path) 2137 if (!iter->key_cache_path)
2168 iter->key_cache_path = bch2_p 2138 iter->key_cache_path = bch2_path_get(trans, iter->btree_id, pos,
2169 2139 iter->flags & BTREE_ITER_intent, 0,
2170 2140 iter->flags|BTREE_ITER_cached|
2171 2141 BTREE_ITER_cached_nofill,
2172 2142 _THIS_IP_);
2173 2143
2174 iter->key_cache_path = bch2_btree_pat 2144 iter->key_cache_path = bch2_btree_path_set_pos(trans, iter->key_cache_path, pos,
2175 iter- 2145 iter->flags & BTREE_ITER_intent,
2176 btree 2146 btree_iter_ip_allocated(iter));
2177 2147
2178 ret = bch2_btree_path_traverse(tran 2148 ret = bch2_btree_path_traverse(trans, iter->key_cache_path,
2179 iter 2149 iter->flags|BTREE_ITER_cached) ?:
2180 bch2_btree_path_relock(trans, 2150 bch2_btree_path_relock(trans, btree_iter_path(trans, iter), _THIS_IP_);
2181 if (unlikely(ret)) 2151 if (unlikely(ret))
2182 return bkey_s_c_err(ret); 2152 return bkey_s_c_err(ret);
2183 2153
2184 btree_path_set_should_be_locked(trans !! 2154 btree_path_set_should_be_locked(trans->paths + iter->key_cache_path);
2185 2155
2186 k = bch2_btree_path_peek_slot(trans-> 2156 k = bch2_btree_path_peek_slot(trans->paths + iter->key_cache_path, &u);
2187 if (k.k && !bkey_err(k)) { 2157 if (k.k && !bkey_err(k)) {
2188 iter->k = u; 2158 iter->k = u;
2189 k.k = &iter->k; 2159 k.k = &iter->k;
2190 } 2160 }
2191 return k; 2161 return k;
2192 } 2162 }
2193 2163
2194 static struct bkey_s_c __bch2_btree_iter_peek 2164 static struct bkey_s_c __bch2_btree_iter_peek(struct btree_iter *iter, struct bpos search_key)
2195 { 2165 {
2196 struct btree_trans *trans = iter->tra 2166 struct btree_trans *trans = iter->trans;
2197 struct bkey_s_c k, k2; 2167 struct bkey_s_c k, k2;
2198 int ret; 2168 int ret;
2199 2169
2200 EBUG_ON(btree_iter_path(trans, iter)- 2170 EBUG_ON(btree_iter_path(trans, iter)->cached);
2201 bch2_btree_iter_verify(iter); 2171 bch2_btree_iter_verify(iter);
2202 2172
2203 while (1) { 2173 while (1) {
2204 struct btree_path_level *l; 2174 struct btree_path_level *l;
2205 2175
2206 iter->path = bch2_btree_path_ 2176 iter->path = bch2_btree_path_set_pos(trans, iter->path, search_key,
2207 iter- 2177 iter->flags & BTREE_ITER_intent,
2208 btree 2178 btree_iter_ip_allocated(iter));
2209 2179
2210 ret = bch2_btree_path_travers 2180 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
2211 if (unlikely(ret)) { 2181 if (unlikely(ret)) {
2212 /* ensure that iter-> 2182 /* ensure that iter->k is consistent with iter->pos: */
2213 bch2_btree_iter_set_p 2183 bch2_btree_iter_set_pos(iter, iter->pos);
2214 k = bkey_s_c_err(ret) 2184 k = bkey_s_c_err(ret);
2215 goto out; 2185 goto out;
2216 } 2186 }
2217 2187
2218 struct btree_path *path = btr 2188 struct btree_path *path = btree_iter_path(trans, iter);
2219 l = path_l(path); 2189 l = path_l(path);
2220 2190
2221 if (unlikely(!l->b)) { 2191 if (unlikely(!l->b)) {
2222 /* No btree nodes at 2192 /* No btree nodes at requested level: */
2223 bch2_btree_iter_set_p 2193 bch2_btree_iter_set_pos(iter, SPOS_MAX);
2224 k = bkey_s_c_null; 2194 k = bkey_s_c_null;
2225 goto out; 2195 goto out;
2226 } 2196 }
2227 2197
2228 btree_path_set_should_be_lock !! 2198 btree_path_set_should_be_locked(path);
2229 2199
2230 k = btree_path_level_peek_all 2200 k = btree_path_level_peek_all(trans->c, l, &iter->k);
2231 2201
2232 if (unlikely(iter->flags & BT 2202 if (unlikely(iter->flags & BTREE_ITER_with_key_cache) &&
2233 k.k && 2203 k.k &&
2234 (k2 = btree_trans_peek_ke 2204 (k2 = btree_trans_peek_key_cache(iter, k.k->p)).k) {
2235 k = k2; 2205 k = k2;
2236 ret = bkey_err(k); 2206 ret = bkey_err(k);
2237 if (ret) { 2207 if (ret) {
2238 bch2_btree_it 2208 bch2_btree_iter_set_pos(iter, iter->pos);
2239 goto out; 2209 goto out;
2240 } 2210 }
2241 } 2211 }
2242 2212
2243 if (unlikely(iter->flags & BT 2213 if (unlikely(iter->flags & BTREE_ITER_with_journal))
2244 k = btree_trans_peek_ 2214 k = btree_trans_peek_journal(trans, iter, k);
2245 2215
2246 if (unlikely((iter->flags & B 2216 if (unlikely((iter->flags & BTREE_ITER_with_updates) &&
2247 trans->nr_update 2217 trans->nr_updates))
2248 bch2_btree_trans_peek 2218 bch2_btree_trans_peek_updates(trans, iter, &k);
2249 2219
2250 if (k.k && bkey_deleted(k.k)) 2220 if (k.k && bkey_deleted(k.k)) {
2251 /* 2221 /*
2252 * If we've got a whi 2222 * If we've got a whiteout, and it's after the search
2253 * key, advance the s 2223 * key, advance the search key to the whiteout instead
2254 * of just after the 2224 * of just after the whiteout - it might be a btree
2255 * whiteout, with a r 2225 * whiteout, with a real key at the same position, since
2256 * in the btree delet 2226 * in the btree deleted keys sort before non deleted.
2257 */ 2227 */
2258 search_key = !bpos_eq 2228 search_key = !bpos_eq(search_key, k.k->p)
2259 ? k.k->p 2229 ? k.k->p
2260 : bpos_succes 2230 : bpos_successor(k.k->p);
2261 continue; 2231 continue;
2262 } 2232 }
2263 2233
2264 if (likely(k.k)) { 2234 if (likely(k.k)) {
2265 break; 2235 break;
2266 } else if (likely(!bpos_eq(l- 2236 } else if (likely(!bpos_eq(l->b->key.k.p, SPOS_MAX))) {
2267 /* Advance to next le 2237 /* Advance to next leaf node: */
2268 search_key = bpos_suc 2238 search_key = bpos_successor(l->b->key.k.p);
2269 } else { 2239 } else {
2270 /* End of btree: */ 2240 /* End of btree: */
2271 bch2_btree_iter_set_p 2241 bch2_btree_iter_set_pos(iter, SPOS_MAX);
2272 k = bkey_s_c_null; 2242 k = bkey_s_c_null;
2273 goto out; 2243 goto out;
2274 } 2244 }
2275 } 2245 }
2276 out: 2246 out:
2277 bch2_btree_iter_verify(iter); 2247 bch2_btree_iter_verify(iter);
2278 2248
2279 return k; 2249 return k;
2280 } 2250 }
2281 2251
2282 /** 2252 /**
2283 * bch2_btree_iter_peek_upto() - returns firs 2253 * bch2_btree_iter_peek_upto() - returns first key greater than or equal to
2284 * iterator's current position 2254 * iterator's current position
2285 * @iter: iterator to peek from 2255 * @iter: iterator to peek from
2286 * @end: search limit: returns keys le 2256 * @end: search limit: returns keys less than or equal to @end
2287 * 2257 *
2288 * Returns: key if found, or an error ext 2258 * Returns: key if found, or an error extractable with bkey_err().
2289 */ 2259 */
2290 struct bkey_s_c bch2_btree_iter_peek_upto(str 2260 struct bkey_s_c bch2_btree_iter_peek_upto(struct btree_iter *iter, struct bpos end)
2291 { 2261 {
2292 struct btree_trans *trans = iter->tra 2262 struct btree_trans *trans = iter->trans;
2293 struct bpos search_key = btree_iter_s 2263 struct bpos search_key = btree_iter_search_key(iter);
2294 struct bkey_s_c k; 2264 struct bkey_s_c k;
2295 struct bpos iter_pos; 2265 struct bpos iter_pos;
2296 int ret; 2266 int ret;
2297 2267
2298 bch2_trans_verify_not_unlocked(trans) 2268 bch2_trans_verify_not_unlocked(trans);
2299 EBUG_ON((iter->flags & BTREE_ITER_fil 2269 EBUG_ON((iter->flags & BTREE_ITER_filter_snapshots) && bkey_eq(end, POS_MAX));
2300 2270
2301 if (iter->update_path) { 2271 if (iter->update_path) {
2302 bch2_path_put_nokeep(trans, i 2272 bch2_path_put_nokeep(trans, iter->update_path,
2303 iter->fl 2273 iter->flags & BTREE_ITER_intent);
2304 iter->update_path = 0; 2274 iter->update_path = 0;
2305 } 2275 }
2306 2276
2307 bch2_btree_iter_verify_entry_exit(ite 2277 bch2_btree_iter_verify_entry_exit(iter);
2308 2278
2309 while (1) { 2279 while (1) {
2310 k = __bch2_btree_iter_peek(it 2280 k = __bch2_btree_iter_peek(iter, search_key);
2311 if (unlikely(!k.k)) 2281 if (unlikely(!k.k))
2312 goto end; 2282 goto end;
2313 if (unlikely(bkey_err(k))) 2283 if (unlikely(bkey_err(k)))
2314 goto out_no_locked; 2284 goto out_no_locked;
2315 2285
2316 /* 2286 /*
2317 * We need to check against @ 2287 * We need to check against @end before FILTER_SNAPSHOTS because
2318 * if we get to a different i 2288 * if we get to a different inode that requested we might be
2319 * seeing keys for a differen 2289 * seeing keys for a different snapshot tree that will all be
2320 * filtered out. 2290 * filtered out.
2321 * 2291 *
2322 * But we can't do the full c 2292 * But we can't do the full check here, because bkey_start_pos()
2323 * isn't monotonically increa 2293 * isn't monotonically increasing before FILTER_SNAPSHOTS, and
2324 * that's what we check again 2294 * that's what we check against in extents mode:
2325 */ 2295 */
2326 if (unlikely(!(iter->flags & 2296 if (unlikely(!(iter->flags & BTREE_ITER_is_extents)
2327 ? bkey_gt(k.k->p 2297 ? bkey_gt(k.k->p, end)
2328 : k.k->p.inode > 2298 : k.k->p.inode > end.inode))
2329 goto end; 2299 goto end;
2330 2300
2331 if (iter->update_path && 2301 if (iter->update_path &&
2332 !bkey_eq(trans->paths[ite 2302 !bkey_eq(trans->paths[iter->update_path].pos, k.k->p)) {
2333 bch2_path_put_nokeep( 2303 bch2_path_put_nokeep(trans, iter->update_path,
2334 2304 iter->flags & BTREE_ITER_intent);
2335 iter->update_path = 0 2305 iter->update_path = 0;
2336 } 2306 }
2337 2307
2338 if ((iter->flags & BTREE_ITER 2308 if ((iter->flags & BTREE_ITER_filter_snapshots) &&
2339 (iter->flags & BTREE_ITER 2309 (iter->flags & BTREE_ITER_intent) &&
2340 !(iter->flags & BTREE_ITE 2310 !(iter->flags & BTREE_ITER_is_extents) &&
2341 !iter->update_path) { 2311 !iter->update_path) {
2342 struct bpos pos = k.k 2312 struct bpos pos = k.k->p;
2343 2313
2344 if (pos.snapshot < it 2314 if (pos.snapshot < iter->snapshot) {
2345 search_key = 2315 search_key = bpos_successor(k.k->p);
2346 continue; 2316 continue;
2347 } 2317 }
2348 2318
2349 pos.snapshot = iter-> 2319 pos.snapshot = iter->snapshot;
2350 2320
2351 /* 2321 /*
2352 * advance, same as o 2322 * advance, same as on exit for iter->path, but only up
2353 * to snapshot 2323 * to snapshot
2354 */ 2324 */
2355 __btree_path_get(tran !! 2325 __btree_path_get(trans->paths + iter->path, iter->flags & BTREE_ITER_intent);
2356 iter->update_path = i 2326 iter->update_path = iter->path;
2357 2327
2358 iter->update_path = b 2328 iter->update_path = bch2_btree_path_set_pos(trans,
2359 2329 iter->update_path, pos,
2360 2330 iter->flags & BTREE_ITER_intent,
2361 2331 _THIS_IP_);
2362 ret = bch2_btree_path 2332 ret = bch2_btree_path_traverse(trans, iter->update_path, iter->flags);
2363 if (unlikely(ret)) { 2333 if (unlikely(ret)) {
2364 k = bkey_s_c_ 2334 k = bkey_s_c_err(ret);
2365 goto out_no_l 2335 goto out_no_locked;
2366 } 2336 }
2367 } 2337 }
2368 2338
2369 /* 2339 /*
2370 * We can never have a key in 2340 * We can never have a key in a leaf node at POS_MAX, so
2371 * we don't have to check the 2341 * we don't have to check these successor() calls:
2372 */ 2342 */
2373 if ((iter->flags & BTREE_ITER 2343 if ((iter->flags & BTREE_ITER_filter_snapshots) &&
2374 !bch2_snapshot_is_ancesto 2344 !bch2_snapshot_is_ancestor(trans->c,
2375 2345 iter->snapshot,
2376 2346 k.k->p.snapshot)) {
2377 search_key = bpos_suc 2347 search_key = bpos_successor(k.k->p);
2378 continue; 2348 continue;
2379 } 2349 }
2380 2350
2381 if (bkey_whiteout(k.k) && 2351 if (bkey_whiteout(k.k) &&
2382 !(iter->flags & BTREE_ITE 2352 !(iter->flags & BTREE_ITER_all_snapshots)) {
2383 search_key = bkey_suc 2353 search_key = bkey_successor(iter, k.k->p);
2384 continue; 2354 continue;
2385 } 2355 }
2386 2356
2387 /* 2357 /*
2388 * iter->pos should be monono 2358 * iter->pos should be mononotically increasing, and always be
2389 * equal to the key we just r 2359 * equal to the key we just returned - except extents can
2390 * straddle iter->pos: 2360 * straddle iter->pos:
2391 */ 2361 */
2392 if (!(iter->flags & BTREE_ITE 2362 if (!(iter->flags & BTREE_ITER_is_extents))
2393 iter_pos = k.k->p; 2363 iter_pos = k.k->p;
2394 else 2364 else
2395 iter_pos = bkey_max(i 2365 iter_pos = bkey_max(iter->pos, bkey_start_pos(k.k));
2396 2366
2397 if (unlikely(iter->flags & BT !! 2367 if (unlikely(!(iter->flags & BTREE_ITER_is_extents)
2398 iter->flags & BT !! 2368 ? bkey_gt(iter_pos, end)
2399 !! 2369 : bkey_ge(iter_pos, end)))
2400 goto end; 2370 goto end;
2401 2371
2402 break; 2372 break;
2403 } 2373 }
2404 2374
2405 iter->pos = iter_pos; 2375 iter->pos = iter_pos;
2406 2376
2407 iter->path = bch2_btree_path_set_pos( 2377 iter->path = bch2_btree_path_set_pos(trans, iter->path, k.k->p,
2408 iter->flags & 2378 iter->flags & BTREE_ITER_intent,
2409 btree_iter_ip 2379 btree_iter_ip_allocated(iter));
2410 2380
2411 btree_path_set_should_be_locked(trans !! 2381 btree_path_set_should_be_locked(btree_iter_path(trans, iter));
2412 out_no_locked: 2382 out_no_locked:
2413 if (iter->update_path) { 2383 if (iter->update_path) {
2414 ret = bch2_btree_path_relock( 2384 ret = bch2_btree_path_relock(trans, trans->paths + iter->update_path, _THIS_IP_);
2415 if (unlikely(ret)) 2385 if (unlikely(ret))
2416 k = bkey_s_c_err(ret) 2386 k = bkey_s_c_err(ret);
2417 else 2387 else
2418 btree_path_set_should !! 2388 btree_path_set_should_be_locked(trans->paths + iter->update_path);
2419 } 2389 }
2420 2390
2421 if (!(iter->flags & BTREE_ITER_all_sn 2391 if (!(iter->flags & BTREE_ITER_all_snapshots))
2422 iter->pos.snapshot = iter->sn 2392 iter->pos.snapshot = iter->snapshot;
2423 2393
2424 ret = bch2_btree_iter_verify_ret(iter 2394 ret = bch2_btree_iter_verify_ret(iter, k);
2425 if (unlikely(ret)) { 2395 if (unlikely(ret)) {
2426 bch2_btree_iter_set_pos(iter, 2396 bch2_btree_iter_set_pos(iter, iter->pos);
2427 k = bkey_s_c_err(ret); 2397 k = bkey_s_c_err(ret);
2428 } 2398 }
2429 2399
2430 bch2_btree_iter_verify_entry_exit(ite 2400 bch2_btree_iter_verify_entry_exit(iter);
2431 2401
2432 return k; 2402 return k;
2433 end: 2403 end:
2434 bch2_btree_iter_set_pos(iter, end); 2404 bch2_btree_iter_set_pos(iter, end);
2435 k = bkey_s_c_null; 2405 k = bkey_s_c_null;
2436 goto out_no_locked; 2406 goto out_no_locked;
2437 } 2407 }
2438 2408
2439 /** 2409 /**
2440 * bch2_btree_iter_next() - returns first key 2410 * bch2_btree_iter_next() - returns first key greater than iterator's current
2441 * position 2411 * position
2442 * @iter: iterator to peek from 2412 * @iter: iterator to peek from
2443 * 2413 *
2444 * Returns: key if found, or an error ext 2414 * Returns: key if found, or an error extractable with bkey_err().
2445 */ 2415 */
2446 struct bkey_s_c bch2_btree_iter_next(struct b 2416 struct bkey_s_c bch2_btree_iter_next(struct btree_iter *iter)
2447 { 2417 {
2448 if (!bch2_btree_iter_advance(iter)) 2418 if (!bch2_btree_iter_advance(iter))
2449 return bkey_s_c_null; 2419 return bkey_s_c_null;
2450 2420
2451 return bch2_btree_iter_peek(iter); 2421 return bch2_btree_iter_peek(iter);
2452 } 2422 }
2453 2423
2454 /** 2424 /**
2455 * bch2_btree_iter_peek_prev() - returns firs 2425 * bch2_btree_iter_peek_prev() - returns first key less than or equal to
2456 * iterator's current position 2426 * iterator's current position
2457 * @iter: iterator to peek from 2427 * @iter: iterator to peek from
2458 * 2428 *
2459 * Returns: key if found, or an error ext 2429 * Returns: key if found, or an error extractable with bkey_err().
2460 */ 2430 */
2461 struct bkey_s_c bch2_btree_iter_peek_prev(str 2431 struct bkey_s_c bch2_btree_iter_peek_prev(struct btree_iter *iter)
2462 { 2432 {
2463 struct btree_trans *trans = iter->tra 2433 struct btree_trans *trans = iter->trans;
2464 struct bpos search_key = iter->pos; 2434 struct bpos search_key = iter->pos;
2465 struct bkey_s_c k; 2435 struct bkey_s_c k;
2466 struct bkey saved_k; 2436 struct bkey saved_k;
2467 const struct bch_val *saved_v; 2437 const struct bch_val *saved_v;
2468 btree_path_idx_t saved_path = 0; 2438 btree_path_idx_t saved_path = 0;
2469 int ret; 2439 int ret;
2470 2440
2471 bch2_trans_verify_not_unlocked(trans) 2441 bch2_trans_verify_not_unlocked(trans);
2472 EBUG_ON(btree_iter_path(trans, iter)- 2442 EBUG_ON(btree_iter_path(trans, iter)->cached ||
2473 btree_iter_path(trans, iter)- 2443 btree_iter_path(trans, iter)->level);
2474 2444
2475 if (iter->flags & BTREE_ITER_with_jou 2445 if (iter->flags & BTREE_ITER_with_journal)
2476 return bkey_s_c_err(-BCH_ERR_ 2446 return bkey_s_c_err(-BCH_ERR_btree_iter_with_journal_not_supported);
2477 2447
2478 bch2_btree_iter_verify(iter); 2448 bch2_btree_iter_verify(iter);
2479 bch2_btree_iter_verify_entry_exit(ite 2449 bch2_btree_iter_verify_entry_exit(iter);
2480 2450
2481 if (iter->flags & BTREE_ITER_filter_s 2451 if (iter->flags & BTREE_ITER_filter_snapshots)
2482 search_key.snapshot = U32_MAX 2452 search_key.snapshot = U32_MAX;
2483 2453
2484 while (1) { 2454 while (1) {
2485 iter->path = bch2_btree_path_ 2455 iter->path = bch2_btree_path_set_pos(trans, iter->path, search_key,
2486 2456 iter->flags & BTREE_ITER_intent,
2487 2457 btree_iter_ip_allocated(iter));
2488 2458
2489 ret = bch2_btree_path_travers 2459 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
2490 if (unlikely(ret)) { 2460 if (unlikely(ret)) {
2491 /* ensure that iter-> 2461 /* ensure that iter->k is consistent with iter->pos: */
2492 bch2_btree_iter_set_p 2462 bch2_btree_iter_set_pos(iter, iter->pos);
2493 k = bkey_s_c_err(ret) 2463 k = bkey_s_c_err(ret);
2494 goto out_no_locked; 2464 goto out_no_locked;
2495 } 2465 }
2496 2466
2497 struct btree_path *path = btr 2467 struct btree_path *path = btree_iter_path(trans, iter);
2498 2468
2499 k = btree_path_level_peek(tra 2469 k = btree_path_level_peek(trans, path, &path->l[0], &iter->k);
2500 if (!k.k || 2470 if (!k.k ||
2501 ((iter->flags & BTREE_ITE 2471 ((iter->flags & BTREE_ITER_is_extents)
2502 ? bpos_ge(bkey_start_pos 2472 ? bpos_ge(bkey_start_pos(k.k), search_key)
2503 : bpos_gt(k.k->p, search 2473 : bpos_gt(k.k->p, search_key)))
2504 k = btree_path_level_ 2474 k = btree_path_level_prev(trans, path, &path->l[0], &iter->k);
2505 2475
2506 if (unlikely((iter->flags & B 2476 if (unlikely((iter->flags & BTREE_ITER_with_updates) &&
2507 trans->nr_update 2477 trans->nr_updates))
2508 bch2_btree_trans_peek 2478 bch2_btree_trans_peek_prev_updates(trans, iter, &k);
2509 2479
2510 if (likely(k.k)) { 2480 if (likely(k.k)) {
2511 if (iter->flags & BTR 2481 if (iter->flags & BTREE_ITER_filter_snapshots) {
2512 if (k.k->p.sn 2482 if (k.k->p.snapshot == iter->snapshot)
2513 goto 2483 goto got_key;
2514 2484
2515 /* 2485 /*
2516 * If we have 2486 * If we have a saved candidate, and we're no
2517 * longer at 2487 * longer at the same _key_ (not pos), return
2518 * that candi 2488 * that candidate
2519 */ 2489 */
2520 if (saved_pat 2490 if (saved_path && !bkey_eq(k.k->p, saved_k.p)) {
2521 bch2_ 2491 bch2_path_put_nokeep(trans, iter->path,
2522 2492 iter->flags & BTREE_ITER_intent);
2523 iter- 2493 iter->path = saved_path;
2524 saved 2494 saved_path = 0;
2525 iter- 2495 iter->k = saved_k;
2526 k.v 2496 k.v = saved_v;
2527 goto 2497 goto got_key;
2528 } 2498 }
2529 2499
2530 if (bch2_snap 2500 if (bch2_snapshot_is_ancestor(trans->c,
2531 2501 iter->snapshot,
2532 2502 k.k->p.snapshot)) {
2533 if (s 2503 if (saved_path)
2534 2504 bch2_path_put_nokeep(trans, saved_path,
2535 2505 iter->flags & BTREE_ITER_intent);
2536 saved 2506 saved_path = btree_path_clone(trans, iter->path,
2537 2507 iter->flags & BTREE_ITER_intent,
2538 2508 _THIS_IP_);
2539 path 2509 path = btree_iter_path(trans, iter);
2540 trace <<
2541 saved 2510 saved_k = *k.k;
2542 saved 2511 saved_v = k.v;
2543 } 2512 }
2544 2513
2545 search_key = 2514 search_key = bpos_predecessor(k.k->p);
2546 continue; 2515 continue;
2547 } 2516 }
2548 got_key: 2517 got_key:
2549 if (bkey_whiteout(k.k 2518 if (bkey_whiteout(k.k) &&
2550 !(iter->flags & B 2519 !(iter->flags & BTREE_ITER_all_snapshots)) {
2551 search_key = 2520 search_key = bkey_predecessor(iter, k.k->p);
2552 if (iter->fla 2521 if (iter->flags & BTREE_ITER_filter_snapshots)
2553 searc 2522 search_key.snapshot = U32_MAX;
2554 continue; 2523 continue;
2555 } 2524 }
2556 2525
2557 btree_path_set_should !! 2526 btree_path_set_should_be_locked(path);
2558 break; 2527 break;
2559 } else if (likely(!bpos_eq(pa 2528 } else if (likely(!bpos_eq(path->l[0].b->data->min_key, POS_MIN))) {
2560 /* Advance to previou 2529 /* Advance to previous leaf node: */
2561 search_key = bpos_pre 2530 search_key = bpos_predecessor(path->l[0].b->data->min_key);
2562 } else { 2531 } else {
2563 /* Start of btree: */ 2532 /* Start of btree: */
2564 bch2_btree_iter_set_p 2533 bch2_btree_iter_set_pos(iter, POS_MIN);
2565 k = bkey_s_c_null; 2534 k = bkey_s_c_null;
2566 goto out_no_locked; 2535 goto out_no_locked;
2567 } 2536 }
2568 } 2537 }
2569 2538
2570 EBUG_ON(bkey_gt(bkey_start_pos(k.k), 2539 EBUG_ON(bkey_gt(bkey_start_pos(k.k), iter->pos));
2571 2540
2572 /* Extents can straddle iter->pos: */ 2541 /* Extents can straddle iter->pos: */
2573 if (bkey_lt(k.k->p, iter->pos)) 2542 if (bkey_lt(k.k->p, iter->pos))
2574 iter->pos = k.k->p; 2543 iter->pos = k.k->p;
2575 2544
2576 if (iter->flags & BTREE_ITER_filter_s 2545 if (iter->flags & BTREE_ITER_filter_snapshots)
2577 iter->pos.snapshot = iter->sn 2546 iter->pos.snapshot = iter->snapshot;
2578 out_no_locked: 2547 out_no_locked:
2579 if (saved_path) 2548 if (saved_path)
2580 bch2_path_put_nokeep(trans, s 2549 bch2_path_put_nokeep(trans, saved_path, iter->flags & BTREE_ITER_intent);
2581 2550
2582 bch2_btree_iter_verify_entry_exit(ite 2551 bch2_btree_iter_verify_entry_exit(iter);
2583 bch2_btree_iter_verify(iter); 2552 bch2_btree_iter_verify(iter);
2584 2553
2585 return k; 2554 return k;
2586 } 2555 }
2587 2556
2588 /** 2557 /**
2589 * bch2_btree_iter_prev() - returns first key 2558 * bch2_btree_iter_prev() - returns first key less than iterator's current
2590 * position 2559 * position
2591 * @iter: iterator to peek from 2560 * @iter: iterator to peek from
2592 * 2561 *
2593 * Returns: key if found, or an error ext 2562 * Returns: key if found, or an error extractable with bkey_err().
2594 */ 2563 */
2595 struct bkey_s_c bch2_btree_iter_prev(struct b 2564 struct bkey_s_c bch2_btree_iter_prev(struct btree_iter *iter)
2596 { 2565 {
2597 if (!bch2_btree_iter_rewind(iter)) 2566 if (!bch2_btree_iter_rewind(iter))
2598 return bkey_s_c_null; 2567 return bkey_s_c_null;
2599 2568
2600 return bch2_btree_iter_peek_prev(iter 2569 return bch2_btree_iter_peek_prev(iter);
2601 } 2570 }
2602 2571
2603 struct bkey_s_c bch2_btree_iter_peek_slot(str 2572 struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *iter)
2604 { 2573 {
2605 struct btree_trans *trans = iter->tra 2574 struct btree_trans *trans = iter->trans;
2606 struct bpos search_key; 2575 struct bpos search_key;
2607 struct bkey_s_c k; 2576 struct bkey_s_c k;
2608 int ret; 2577 int ret;
2609 2578
2610 bch2_trans_verify_not_unlocked(trans) 2579 bch2_trans_verify_not_unlocked(trans);
2611 bch2_btree_iter_verify(iter); 2580 bch2_btree_iter_verify(iter);
2612 bch2_btree_iter_verify_entry_exit(ite 2581 bch2_btree_iter_verify_entry_exit(iter);
2613 EBUG_ON(btree_iter_path(trans, iter)- 2582 EBUG_ON(btree_iter_path(trans, iter)->level && (iter->flags & BTREE_ITER_with_key_cache));
2614 2583
2615 /* extents can't span inode numbers: 2584 /* extents can't span inode numbers: */
2616 if ((iter->flags & BTREE_ITER_is_exte 2585 if ((iter->flags & BTREE_ITER_is_extents) &&
2617 unlikely(iter->pos.offset == KEY_ 2586 unlikely(iter->pos.offset == KEY_OFFSET_MAX)) {
2618 if (iter->pos.inode == KEY_IN 2587 if (iter->pos.inode == KEY_INODE_MAX)
2619 return bkey_s_c_null; 2588 return bkey_s_c_null;
2620 2589
2621 bch2_btree_iter_set_pos(iter, 2590 bch2_btree_iter_set_pos(iter, bpos_nosnap_successor(iter->pos));
2622 } 2591 }
2623 2592
2624 search_key = btree_iter_search_key(it 2593 search_key = btree_iter_search_key(iter);
2625 iter->path = bch2_btree_path_set_pos( 2594 iter->path = bch2_btree_path_set_pos(trans, iter->path, search_key,
2626 iter- 2595 iter->flags & BTREE_ITER_intent,
2627 btree 2596 btree_iter_ip_allocated(iter));
2628 2597
2629 ret = bch2_btree_path_traverse(trans, 2598 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
2630 if (unlikely(ret)) { 2599 if (unlikely(ret)) {
2631 k = bkey_s_c_err(ret); 2600 k = bkey_s_c_err(ret);
2632 goto out_no_locked; 2601 goto out_no_locked;
2633 } 2602 }
2634 2603
2635 if ((iter->flags & BTREE_ITER_cached) 2604 if ((iter->flags & BTREE_ITER_cached) ||
2636 !(iter->flags & (BTREE_ITER_is_ex 2605 !(iter->flags & (BTREE_ITER_is_extents|BTREE_ITER_filter_snapshots))) {
2637 k = bkey_s_c_null; 2606 k = bkey_s_c_null;
2638 2607
2639 if (unlikely((iter->flags & B 2608 if (unlikely((iter->flags & BTREE_ITER_with_updates) &&
2640 trans->nr_update 2609 trans->nr_updates)) {
2641 bch2_btree_trans_peek 2610 bch2_btree_trans_peek_slot_updates(trans, iter, &k);
2642 if (k.k) 2611 if (k.k)
2643 goto out; 2612 goto out;
2644 } 2613 }
2645 2614
2646 if (unlikely(iter->flags & BT 2615 if (unlikely(iter->flags & BTREE_ITER_with_journal) &&
2647 (k = btree_trans_peek_slo 2616 (k = btree_trans_peek_slot_journal(trans, iter)).k)
2648 goto out; 2617 goto out;
2649 2618
2650 if (unlikely(iter->flags & BT 2619 if (unlikely(iter->flags & BTREE_ITER_with_key_cache) &&
2651 (k = btree_trans_peek_key 2620 (k = btree_trans_peek_key_cache(iter, iter->pos)).k) {
2652 if (!bkey_err(k)) 2621 if (!bkey_err(k))
2653 iter->k = *k. 2622 iter->k = *k.k;
2654 /* We're not returnin 2623 /* We're not returning a key from iter->path: */
2655 goto out_no_locked; 2624 goto out_no_locked;
2656 } 2625 }
2657 2626
2658 k = bch2_btree_path_peek_slot 2627 k = bch2_btree_path_peek_slot(trans->paths + iter->path, &iter->k);
2659 if (unlikely(!k.k)) 2628 if (unlikely(!k.k))
2660 goto out_no_locked; 2629 goto out_no_locked;
2661 } else { 2630 } else {
2662 struct bpos next; 2631 struct bpos next;
2663 struct bpos end = iter->pos; 2632 struct bpos end = iter->pos;
2664 2633
2665 if (iter->flags & BTREE_ITER_ 2634 if (iter->flags & BTREE_ITER_is_extents)
2666 end.offset = U64_MAX; 2635 end.offset = U64_MAX;
2667 2636
2668 EBUG_ON(btree_iter_path(trans 2637 EBUG_ON(btree_iter_path(trans, iter)->level);
2669 2638
2670 if (iter->flags & BTREE_ITER_ 2639 if (iter->flags & BTREE_ITER_intent) {
2671 struct btree_iter ite 2640 struct btree_iter iter2;
2672 2641
2673 bch2_trans_copy_iter( 2642 bch2_trans_copy_iter(&iter2, iter);
2674 k = bch2_btree_iter_p 2643 k = bch2_btree_iter_peek_upto(&iter2, end);
2675 2644
2676 if (k.k && !bkey_err( 2645 if (k.k && !bkey_err(k)) {
2677 swap(iter->ke 2646 swap(iter->key_cache_path, iter2.key_cache_path);
2678 iter->k = ite 2647 iter->k = iter2.k;
2679 k.k = &iter-> 2648 k.k = &iter->k;
2680 } 2649 }
2681 bch2_trans_iter_exit( 2650 bch2_trans_iter_exit(trans, &iter2);
2682 } else { 2651 } else {
2683 struct bpos pos = ite 2652 struct bpos pos = iter->pos;
2684 2653
2685 k = bch2_btree_iter_p 2654 k = bch2_btree_iter_peek_upto(iter, end);
2686 if (unlikely(bkey_err 2655 if (unlikely(bkey_err(k)))
2687 bch2_btree_it 2656 bch2_btree_iter_set_pos(iter, pos);
2688 else 2657 else
2689 iter->pos = p 2658 iter->pos = pos;
2690 } 2659 }
2691 2660
2692 if (unlikely(bkey_err(k))) 2661 if (unlikely(bkey_err(k)))
2693 goto out_no_locked; 2662 goto out_no_locked;
2694 2663
2695 next = k.k ? bkey_start_pos(k 2664 next = k.k ? bkey_start_pos(k.k) : POS_MAX;
2696 2665
2697 if (bkey_lt(iter->pos, next)) 2666 if (bkey_lt(iter->pos, next)) {
2698 bkey_init(&iter->k); 2667 bkey_init(&iter->k);
2699 iter->k.p = iter->pos 2668 iter->k.p = iter->pos;
2700 2669
2701 if (iter->flags & BTR 2670 if (iter->flags & BTREE_ITER_is_extents) {
2702 bch2_key_resi 2671 bch2_key_resize(&iter->k,
2703 2672 min_t(u64, KEY_SIZE_MAX,
2704 2673 (next.inode == iter->pos.inode
2705 2674 ? next.offset
2706 2675 : KEY_OFFSET_MAX) -
2707 2676 iter->pos.offset));
2708 EBUG_ON(!iter 2677 EBUG_ON(!iter->k.size);
2709 } 2678 }
2710 2679
2711 k = (struct bkey_s_c) 2680 k = (struct bkey_s_c) { &iter->k, NULL };
2712 } 2681 }
2713 } 2682 }
2714 out: 2683 out:
2715 btree_path_set_should_be_locked(trans !! 2684 btree_path_set_should_be_locked(btree_iter_path(trans, iter));
2716 out_no_locked: 2685 out_no_locked:
2717 bch2_btree_iter_verify_entry_exit(ite 2686 bch2_btree_iter_verify_entry_exit(iter);
2718 bch2_btree_iter_verify(iter); 2687 bch2_btree_iter_verify(iter);
2719 ret = bch2_btree_iter_verify_ret(iter 2688 ret = bch2_btree_iter_verify_ret(iter, k);
2720 if (unlikely(ret)) 2689 if (unlikely(ret))
2721 return bkey_s_c_err(ret); 2690 return bkey_s_c_err(ret);
2722 2691
2723 return k; 2692 return k;
2724 } 2693 }
2725 2694
2726 struct bkey_s_c bch2_btree_iter_next_slot(str 2695 struct bkey_s_c bch2_btree_iter_next_slot(struct btree_iter *iter)
2727 { 2696 {
2728 if (!bch2_btree_iter_advance(iter)) 2697 if (!bch2_btree_iter_advance(iter))
2729 return bkey_s_c_null; 2698 return bkey_s_c_null;
2730 2699
2731 return bch2_btree_iter_peek_slot(iter 2700 return bch2_btree_iter_peek_slot(iter);
2732 } 2701 }
2733 2702
2734 struct bkey_s_c bch2_btree_iter_prev_slot(str 2703 struct bkey_s_c bch2_btree_iter_prev_slot(struct btree_iter *iter)
2735 { 2704 {
2736 if (!bch2_btree_iter_rewind(iter)) 2705 if (!bch2_btree_iter_rewind(iter))
2737 return bkey_s_c_null; 2706 return bkey_s_c_null;
2738 2707
2739 return bch2_btree_iter_peek_slot(iter 2708 return bch2_btree_iter_peek_slot(iter);
2740 } 2709 }
2741 2710
2742 /* Obsolete, but still used by rust wrapper i <<
2743 struct bkey_s_c bch2_btree_iter_peek_and_rest 2711 struct bkey_s_c bch2_btree_iter_peek_and_restart_outlined(struct btree_iter *iter)
2744 { 2712 {
2745 struct bkey_s_c k; 2713 struct bkey_s_c k;
2746 2714
2747 while (btree_trans_too_many_iters(ite 2715 while (btree_trans_too_many_iters(iter->trans) ||
2748 (k = bch2_btree_iter_peek_type 2716 (k = bch2_btree_iter_peek_type(iter, iter->flags),
2749 bch2_err_matches(bkey_err(k), 2717 bch2_err_matches(bkey_err(k), BCH_ERR_transaction_restart)))
2750 bch2_trans_begin(iter->trans) 2718 bch2_trans_begin(iter->trans);
2751 2719
2752 return k; 2720 return k;
2753 } 2721 }
2754 2722
2755 /* new transactional stuff: */ 2723 /* new transactional stuff: */
2756 2724
2757 #ifdef CONFIG_BCACHEFS_DEBUG 2725 #ifdef CONFIG_BCACHEFS_DEBUG
2758 static void btree_trans_verify_sorted_refs(st 2726 static void btree_trans_verify_sorted_refs(struct btree_trans *trans)
2759 { 2727 {
2760 struct btree_path *path; 2728 struct btree_path *path;
2761 unsigned i; 2729 unsigned i;
2762 2730
2763 BUG_ON(trans->nr_sorted != bitmap_wei 2731 BUG_ON(trans->nr_sorted != bitmap_weight(trans->paths_allocated, trans->nr_paths) - 1);
2764 2732
2765 trans_for_each_path(trans, path, i) { 2733 trans_for_each_path(trans, path, i) {
2766 BUG_ON(path->sorted_idx >= tr 2734 BUG_ON(path->sorted_idx >= trans->nr_sorted);
2767 BUG_ON(trans->sorted[path->so 2735 BUG_ON(trans->sorted[path->sorted_idx] != i);
2768 } 2736 }
2769 2737
2770 for (i = 0; i < trans->nr_sorted; i++ 2738 for (i = 0; i < trans->nr_sorted; i++) {
2771 unsigned idx = trans->sorted[ 2739 unsigned idx = trans->sorted[i];
2772 2740
2773 BUG_ON(!test_bit(idx, trans-> 2741 BUG_ON(!test_bit(idx, trans->paths_allocated));
2774 BUG_ON(trans->paths[idx].sort 2742 BUG_ON(trans->paths[idx].sorted_idx != i);
2775 } 2743 }
2776 } 2744 }
2777 2745
2778 static void btree_trans_verify_sorted(struct 2746 static void btree_trans_verify_sorted(struct btree_trans *trans)
2779 { 2747 {
2780 struct btree_path *path, *prev = NULL 2748 struct btree_path *path, *prev = NULL;
2781 struct trans_for_each_path_inorder_it 2749 struct trans_for_each_path_inorder_iter iter;
2782 2750
2783 if (!bch2_debug_check_iterators) 2751 if (!bch2_debug_check_iterators)
2784 return; 2752 return;
2785 2753
2786 trans_for_each_path_inorder(trans, pa 2754 trans_for_each_path_inorder(trans, path, iter) {
2787 if (prev && btree_path_cmp(pr 2755 if (prev && btree_path_cmp(prev, path) > 0) {
2788 __bch2_dump_trans_pat 2756 __bch2_dump_trans_paths_updates(trans, true);
2789 panic("trans paths ou 2757 panic("trans paths out of order!\n");
2790 } 2758 }
2791 prev = path; 2759 prev = path;
2792 } 2760 }
2793 } 2761 }
2794 #else 2762 #else
2795 static inline void btree_trans_verify_sorted_ 2763 static inline void btree_trans_verify_sorted_refs(struct btree_trans *trans) {}
2796 static inline void btree_trans_verify_sorted( 2764 static inline void btree_trans_verify_sorted(struct btree_trans *trans) {}
2797 #endif 2765 #endif
2798 2766
2799 void __bch2_btree_trans_sort_paths(struct btr 2767 void __bch2_btree_trans_sort_paths(struct btree_trans *trans)
2800 { 2768 {
2801 int i, l = 0, r = trans->nr_sorted, i 2769 int i, l = 0, r = trans->nr_sorted, inc = 1;
2802 bool swapped; 2770 bool swapped;
2803 2771
2804 btree_trans_verify_sorted_refs(trans) 2772 btree_trans_verify_sorted_refs(trans);
2805 2773
2806 if (trans->paths_sorted) 2774 if (trans->paths_sorted)
2807 goto out; 2775 goto out;
2808 2776
2809 /* 2777 /*
2810 * Cocktail shaker sort: this is effi 2778 * Cocktail shaker sort: this is efficient because iterators will be
2811 * mostly sorted. 2779 * mostly sorted.
2812 */ 2780 */
2813 do { 2781 do {
2814 swapped = false; 2782 swapped = false;
2815 2783
2816 for (i = inc > 0 ? l : r - 2; 2784 for (i = inc > 0 ? l : r - 2;
2817 i + 1 < r && i >= l; 2785 i + 1 < r && i >= l;
2818 i += inc) { 2786 i += inc) {
2819 if (btree_path_cmp(tr 2787 if (btree_path_cmp(trans->paths + trans->sorted[i],
2820 tr 2788 trans->paths + trans->sorted[i + 1]) > 0) {
2821 swap(trans->s 2789 swap(trans->sorted[i], trans->sorted[i + 1]);
2822 trans->paths[ 2790 trans->paths[trans->sorted[i]].sorted_idx = i;
2823 trans->paths[ 2791 trans->paths[trans->sorted[i + 1]].sorted_idx = i + 1;
2824 swapped = tru 2792 swapped = true;
2825 } 2793 }
2826 } 2794 }
2827 2795
2828 if (inc > 0) 2796 if (inc > 0)
2829 --r; 2797 --r;
2830 else 2798 else
2831 l++; 2799 l++;
2832 inc = -inc; 2800 inc = -inc;
2833 } while (swapped); 2801 } while (swapped);
2834 2802
2835 trans->paths_sorted = true; 2803 trans->paths_sorted = true;
2836 out: 2804 out:
2837 btree_trans_verify_sorted(trans); 2805 btree_trans_verify_sorted(trans);
2838 } 2806 }
2839 2807
2840 static inline void btree_path_list_remove(str 2808 static inline void btree_path_list_remove(struct btree_trans *trans,
2841 str 2809 struct btree_path *path)
2842 { 2810 {
2843 EBUG_ON(path->sorted_idx >= trans->nr 2811 EBUG_ON(path->sorted_idx >= trans->nr_sorted);
2844 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS 2812 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
2845 trans->nr_sorted--; 2813 trans->nr_sorted--;
2846 memmove_u64s_down_small(trans->sorted 2814 memmove_u64s_down_small(trans->sorted + path->sorted_idx,
2847 trans->sorted 2815 trans->sorted + path->sorted_idx + 1,
2848 DIV_ROUND_UP( 2816 DIV_ROUND_UP(trans->nr_sorted - path->sorted_idx,
2849 2817 sizeof(u64) / sizeof(btree_path_idx_t)));
2850 #else 2818 #else
2851 array_remove_item(trans->sorted, tran 2819 array_remove_item(trans->sorted, trans->nr_sorted, path->sorted_idx);
2852 #endif 2820 #endif
2853 for (unsigned i = path->sorted_idx; i 2821 for (unsigned i = path->sorted_idx; i < trans->nr_sorted; i++)
2854 trans->paths[trans->sorted[i] 2822 trans->paths[trans->sorted[i]].sorted_idx = i;
2855 } 2823 }
2856 2824
2857 static inline void btree_path_list_add(struct 2825 static inline void btree_path_list_add(struct btree_trans *trans,
2858 btree_ 2826 btree_path_idx_t pos,
2859 btree_ 2827 btree_path_idx_t path_idx)
2860 { 2828 {
2861 struct btree_path *path = trans->path 2829 struct btree_path *path = trans->paths + path_idx;
2862 2830
2863 path->sorted_idx = pos ? trans->paths 2831 path->sorted_idx = pos ? trans->paths[pos].sorted_idx + 1 : trans->nr_sorted;
2864 2832
2865 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS 2833 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
2866 memmove_u64s_up_small(trans->sorted + 2834 memmove_u64s_up_small(trans->sorted + path->sorted_idx + 1,
2867 trans->sorted + 2835 trans->sorted + path->sorted_idx,
2868 DIV_ROUND_UP(tr 2836 DIV_ROUND_UP(trans->nr_sorted - path->sorted_idx,
2869 si 2837 sizeof(u64) / sizeof(btree_path_idx_t)));
2870 trans->nr_sorted++; 2838 trans->nr_sorted++;
2871 trans->sorted[path->sorted_idx] = pat 2839 trans->sorted[path->sorted_idx] = path_idx;
2872 #else 2840 #else
2873 array_insert_item(trans->sorted, tran 2841 array_insert_item(trans->sorted, trans->nr_sorted, path->sorted_idx, path_idx);
2874 #endif 2842 #endif
2875 2843
2876 for (unsigned i = path->sorted_idx; i 2844 for (unsigned i = path->sorted_idx; i < trans->nr_sorted; i++)
2877 trans->paths[trans->sorted[i] 2845 trans->paths[trans->sorted[i]].sorted_idx = i;
2878 2846
2879 btree_trans_verify_sorted_refs(trans) 2847 btree_trans_verify_sorted_refs(trans);
2880 } 2848 }
2881 2849
2882 void bch2_trans_iter_exit(struct btree_trans 2850 void bch2_trans_iter_exit(struct btree_trans *trans, struct btree_iter *iter)
2883 { 2851 {
2884 if (iter->update_path) 2852 if (iter->update_path)
2885 bch2_path_put_nokeep(trans, i 2853 bch2_path_put_nokeep(trans, iter->update_path,
2886 iter->flags & B 2854 iter->flags & BTREE_ITER_intent);
2887 if (iter->path) 2855 if (iter->path)
2888 bch2_path_put(trans, iter->pa 2856 bch2_path_put(trans, iter->path,
2889 iter->flags & B 2857 iter->flags & BTREE_ITER_intent);
2890 if (iter->key_cache_path) 2858 if (iter->key_cache_path)
2891 bch2_path_put(trans, iter->ke 2859 bch2_path_put(trans, iter->key_cache_path,
2892 iter->flags & B 2860 iter->flags & BTREE_ITER_intent);
2893 iter->path = 0; 2861 iter->path = 0;
2894 iter->update_path = 0; 2862 iter->update_path = 0;
2895 iter->key_cache_path = 0; 2863 iter->key_cache_path = 0;
2896 iter->trans = NULL; 2864 iter->trans = NULL;
2897 } 2865 }
2898 2866
2899 void bch2_trans_iter_init_outlined(struct btr 2867 void bch2_trans_iter_init_outlined(struct btree_trans *trans,
2900 struct btree_iter * 2868 struct btree_iter *iter,
2901 enum btree_id btree 2869 enum btree_id btree_id, struct bpos pos,
2902 unsigned flags) 2870 unsigned flags)
2903 { 2871 {
2904 bch2_trans_iter_init_common(trans, it 2872 bch2_trans_iter_init_common(trans, iter, btree_id, pos, 0, 0,
2905 bch2_btree_ite 2873 bch2_btree_iter_flags(trans, btree_id, flags),
2906 _RET_IP_); 2874 _RET_IP_);
2907 } 2875 }
2908 2876
2909 void bch2_trans_node_iter_init(struct btree_t 2877 void bch2_trans_node_iter_init(struct btree_trans *trans,
2910 struct btree_i 2878 struct btree_iter *iter,
2911 enum btree_id 2879 enum btree_id btree_id,
2912 struct bpos po 2880 struct bpos pos,
2913 unsigned locks 2881 unsigned locks_want,
2914 unsigned depth 2882 unsigned depth,
2915 unsigned flags 2883 unsigned flags)
2916 { 2884 {
2917 flags |= BTREE_ITER_not_extents; 2885 flags |= BTREE_ITER_not_extents;
2918 flags |= BTREE_ITER_snapshot_field; 2886 flags |= BTREE_ITER_snapshot_field;
2919 flags |= BTREE_ITER_all_snapshots; 2887 flags |= BTREE_ITER_all_snapshots;
2920 2888
2921 bch2_trans_iter_init_common(trans, it 2889 bch2_trans_iter_init_common(trans, iter, btree_id, pos, locks_want, depth,
2922 __bch2_btree_i 2890 __bch2_btree_iter_flags(trans, btree_id, flags),
2923 _RET_IP_); 2891 _RET_IP_);
2924 2892
2925 iter->min_depth = depth; 2893 iter->min_depth = depth;
2926 2894
2927 struct btree_path *path = btree_iter_ 2895 struct btree_path *path = btree_iter_path(trans, iter);
2928 BUG_ON(path->locks_want < min(locks_ 2896 BUG_ON(path->locks_want < min(locks_want, BTREE_MAX_DEPTH));
2929 BUG_ON(path->level != depth); 2897 BUG_ON(path->level != depth);
2930 BUG_ON(iter->min_depth != depth); 2898 BUG_ON(iter->min_depth != depth);
2931 } 2899 }
2932 2900
2933 void bch2_trans_copy_iter(struct btree_iter * 2901 void bch2_trans_copy_iter(struct btree_iter *dst, struct btree_iter *src)
2934 { 2902 {
2935 struct btree_trans *trans = src->tran 2903 struct btree_trans *trans = src->trans;
2936 2904
2937 *dst = *src; 2905 *dst = *src;
2938 #ifdef TRACK_PATH_ALLOCATED 2906 #ifdef TRACK_PATH_ALLOCATED
2939 dst->ip_allocated = _RET_IP_; 2907 dst->ip_allocated = _RET_IP_;
2940 #endif 2908 #endif
2941 if (src->path) 2909 if (src->path)
2942 __btree_path_get(trans, trans !! 2910 __btree_path_get(trans->paths + src->path, src->flags & BTREE_ITER_intent);
2943 if (src->update_path) 2911 if (src->update_path)
2944 __btree_path_get(trans, trans !! 2912 __btree_path_get(trans->paths + src->update_path, src->flags & BTREE_ITER_intent);
2945 dst->key_cache_path = 0; 2913 dst->key_cache_path = 0;
2946 } 2914 }
2947 2915
2948 void *__bch2_trans_kmalloc(struct btree_trans 2916 void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
2949 { 2917 {
2950 struct bch_fs *c = trans->c; 2918 struct bch_fs *c = trans->c;
2951 unsigned new_top = trans->mem_top + s 2919 unsigned new_top = trans->mem_top + size;
2952 unsigned old_bytes = trans->mem_bytes 2920 unsigned old_bytes = trans->mem_bytes;
2953 unsigned new_bytes = roundup_pow_of_t 2921 unsigned new_bytes = roundup_pow_of_two(new_top);
2954 int ret; 2922 int ret;
2955 void *new_mem; 2923 void *new_mem;
2956 void *p; 2924 void *p;
2957 2925
2958 WARN_ON_ONCE(new_bytes > BTREE_TRANS_ 2926 WARN_ON_ONCE(new_bytes > BTREE_TRANS_MEM_MAX);
2959 2927
2960 struct btree_transaction_stats *s = b 2928 struct btree_transaction_stats *s = btree_trans_stats(trans);
2961 s->max_mem = max(s->max_mem, new_byte 2929 s->max_mem = max(s->max_mem, new_bytes);
2962 2930
2963 if (trans->used_mempool) { 2931 if (trans->used_mempool) {
2964 if (trans->mem_bytes >= new_b 2932 if (trans->mem_bytes >= new_bytes)
2965 goto out_change_top; 2933 goto out_change_top;
2966 2934
2967 /* No more space from mempool 2935 /* No more space from mempool item, need malloc new one */
2968 new_mem = kmalloc(new_bytes, 2936 new_mem = kmalloc(new_bytes, GFP_NOWAIT|__GFP_NOWARN);
2969 if (unlikely(!new_mem)) { 2937 if (unlikely(!new_mem)) {
2970 bch2_trans_unlock(tra 2938 bch2_trans_unlock(trans);
2971 2939
2972 new_mem = kmalloc(new 2940 new_mem = kmalloc(new_bytes, GFP_KERNEL);
2973 if (!new_mem) 2941 if (!new_mem)
2974 return ERR_PT 2942 return ERR_PTR(-BCH_ERR_ENOMEM_trans_kmalloc);
2975 2943
2976 ret = bch2_trans_relo 2944 ret = bch2_trans_relock(trans);
2977 if (ret) { 2945 if (ret) {
2978 kfree(new_mem 2946 kfree(new_mem);
2979 return ERR_PT 2947 return ERR_PTR(ret);
2980 } 2948 }
2981 } 2949 }
2982 memcpy(new_mem, trans->mem, t 2950 memcpy(new_mem, trans->mem, trans->mem_top);
2983 trans->used_mempool = false; 2951 trans->used_mempool = false;
2984 mempool_free(trans->mem, &c-> 2952 mempool_free(trans->mem, &c->btree_trans_mem_pool);
2985 goto out_new_mem; 2953 goto out_new_mem;
2986 } 2954 }
2987 2955
2988 new_mem = krealloc(trans->mem, new_by 2956 new_mem = krealloc(trans->mem, new_bytes, GFP_NOWAIT|__GFP_NOWARN);
2989 if (unlikely(!new_mem)) { 2957 if (unlikely(!new_mem)) {
2990 bch2_trans_unlock(trans); 2958 bch2_trans_unlock(trans);
2991 2959
2992 new_mem = krealloc(trans->mem 2960 new_mem = krealloc(trans->mem, new_bytes, GFP_KERNEL);
2993 if (!new_mem && new_bytes <= 2961 if (!new_mem && new_bytes <= BTREE_TRANS_MEM_MAX) {
2994 new_mem = mempool_all 2962 new_mem = mempool_alloc(&c->btree_trans_mem_pool, GFP_KERNEL);
2995 new_bytes = BTREE_TRA 2963 new_bytes = BTREE_TRANS_MEM_MAX;
2996 memcpy(new_mem, trans 2964 memcpy(new_mem, trans->mem, trans->mem_top);
2997 trans->used_mempool = 2965 trans->used_mempool = true;
2998 kfree(trans->mem); 2966 kfree(trans->mem);
2999 } 2967 }
3000 2968
3001 if (!new_mem) 2969 if (!new_mem)
3002 return ERR_PTR(-BCH_E 2970 return ERR_PTR(-BCH_ERR_ENOMEM_trans_kmalloc);
3003 2971
3004 trans->mem = new_mem; 2972 trans->mem = new_mem;
3005 trans->mem_bytes = new_bytes; 2973 trans->mem_bytes = new_bytes;
3006 2974
3007 ret = bch2_trans_relock(trans 2975 ret = bch2_trans_relock(trans);
3008 if (ret) 2976 if (ret)
3009 return ERR_PTR(ret); 2977 return ERR_PTR(ret);
3010 } 2978 }
3011 out_new_mem: 2979 out_new_mem:
3012 trans->mem = new_mem; 2980 trans->mem = new_mem;
3013 trans->mem_bytes = new_bytes; 2981 trans->mem_bytes = new_bytes;
3014 2982
3015 if (old_bytes) { 2983 if (old_bytes) {
3016 trace_and_count(c, trans_rest 2984 trace_and_count(c, trans_restart_mem_realloced, trans, _RET_IP_, new_bytes);
3017 return ERR_PTR(btree_trans_re 2985 return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_mem_realloced));
3018 } 2986 }
3019 out_change_top: 2987 out_change_top:
3020 p = trans->mem + trans->mem_top; 2988 p = trans->mem + trans->mem_top;
3021 trans->mem_top += size; 2989 trans->mem_top += size;
3022 memset(p, 0, size); 2990 memset(p, 0, size);
3023 return p; 2991 return p;
3024 } 2992 }
3025 2993
3026 static inline void check_srcu_held_too_long(s 2994 static inline void check_srcu_held_too_long(struct btree_trans *trans)
3027 { 2995 {
3028 WARN(trans->srcu_held && time_after(j 2996 WARN(trans->srcu_held && time_after(jiffies, trans->srcu_lock_time + HZ * 10),
3029 "btree trans held srcu lock (del 2997 "btree trans held srcu lock (delaying memory reclaim) for %lu seconds",
3030 (jiffies - trans->srcu_lock_time 2998 (jiffies - trans->srcu_lock_time) / HZ);
3031 } 2999 }
3032 3000
3033 void bch2_trans_srcu_unlock(struct btree_tran 3001 void bch2_trans_srcu_unlock(struct btree_trans *trans)
3034 { 3002 {
3035 if (trans->srcu_held) { 3003 if (trans->srcu_held) {
3036 struct bch_fs *c = trans->c; 3004 struct bch_fs *c = trans->c;
3037 struct btree_path *path; 3005 struct btree_path *path;
3038 unsigned i; 3006 unsigned i;
3039 3007
3040 trans_for_each_path(trans, pa 3008 trans_for_each_path(trans, path, i)
3041 if (path->cached && ! 3009 if (path->cached && !btree_node_locked(path, 0))
3042 path->l[0].b 3010 path->l[0].b = ERR_PTR(-BCH_ERR_no_btree_node_srcu_reset);
3043 3011
3044 check_srcu_held_too_long(tran 3012 check_srcu_held_too_long(trans);
3045 srcu_read_unlock(&c->btree_tr 3013 srcu_read_unlock(&c->btree_trans_barrier, trans->srcu_idx);
3046 trans->srcu_held = false; 3014 trans->srcu_held = false;
3047 } 3015 }
3048 } 3016 }
3049 3017
3050 static void bch2_trans_srcu_lock(struct btree 3018 static void bch2_trans_srcu_lock(struct btree_trans *trans)
3051 { 3019 {
3052 if (!trans->srcu_held) { 3020 if (!trans->srcu_held) {
3053 trans->srcu_idx = srcu_read_l 3021 trans->srcu_idx = srcu_read_lock(&trans->c->btree_trans_barrier);
3054 trans->srcu_lock_time = jif 3022 trans->srcu_lock_time = jiffies;
3055 trans->srcu_held = true; 3023 trans->srcu_held = true;
3056 } 3024 }
3057 } 3025 }
3058 3026
3059 /** 3027 /**
3060 * bch2_trans_begin() - reset a transaction a 3028 * bch2_trans_begin() - reset a transaction after a interrupted attempt
3061 * @trans: transaction to reset 3029 * @trans: transaction to reset
3062 * 3030 *
3063 * Returns: current restart counter, to b 3031 * Returns: current restart counter, to be used with trans_was_restarted()
3064 * 3032 *
3065 * While iterating over nodes or updating nod 3033 * While iterating over nodes or updating nodes a attempt to lock a btree node
3066 * may return BCH_ERR_transaction_restart whe 3034 * may return BCH_ERR_transaction_restart when the trylock fails. When this
3067 * occurs bch2_trans_begin() should be called 3035 * occurs bch2_trans_begin() should be called and the transaction retried.
3068 */ 3036 */
3069 u32 bch2_trans_begin(struct btree_trans *tran 3037 u32 bch2_trans_begin(struct btree_trans *trans)
3070 { 3038 {
3071 struct btree_path *path; 3039 struct btree_path *path;
3072 unsigned i; 3040 unsigned i;
3073 u64 now; 3041 u64 now;
3074 3042
3075 bch2_trans_reset_updates(trans); 3043 bch2_trans_reset_updates(trans);
3076 3044
3077 trans->restart_count++; 3045 trans->restart_count++;
3078 trans->mem_top = 0; 3046 trans->mem_top = 0;
3079 trans->journal_entries = NUL 3047 trans->journal_entries = NULL;
3080 3048
3081 trans_for_each_path(trans, path, i) { 3049 trans_for_each_path(trans, path, i) {
3082 path->should_be_locked = fals 3050 path->should_be_locked = false;
3083 3051
3084 /* 3052 /*
3085 * If the transaction wasn't 3053 * If the transaction wasn't restarted, we're presuming to be
3086 * doing something new: dont 3054 * doing something new: dont keep iterators excpt the ones that
3087 * are in use - except for th 3055 * are in use - except for the subvolumes btree:
3088 */ 3056 */
3089 if (!trans->restarted && path 3057 if (!trans->restarted && path->btree_id != BTREE_ID_subvolumes)
3090 path->preserve = fals 3058 path->preserve = false;
3091 3059
3092 /* 3060 /*
3093 * XXX: we probably shouldn't 3061 * XXX: we probably shouldn't be doing this if the transaction
3094 * was restarted, but current 3062 * was restarted, but currently we still overflow transaction
3095 * iterators if we do that 3063 * iterators if we do that
3096 */ 3064 */
3097 if (!path->ref && !path->pres 3065 if (!path->ref && !path->preserve)
3098 __bch2_path_free(tran 3066 __bch2_path_free(trans, i);
3099 else 3067 else
3100 path->preserve = fals 3068 path->preserve = false;
3101 } 3069 }
3102 3070
3103 now = local_clock(); 3071 now = local_clock();
3104 3072
3105 if (!IS_ENABLED(CONFIG_BCACHEFS_NO_LA 3073 if (!IS_ENABLED(CONFIG_BCACHEFS_NO_LATENCY_ACCT) &&
3106 time_after64(now, trans->last_beg 3074 time_after64(now, trans->last_begin_time + 10))
3107 __bch2_time_stats_update(&btr 3075 __bch2_time_stats_update(&btree_trans_stats(trans)->duration,
3108 tran 3076 trans->last_begin_time, now);
3109 3077
3110 if (!trans->restarted && 3078 if (!trans->restarted &&
3111 (need_resched() || 3079 (need_resched() ||
3112 time_after64(now, trans->last_be 3080 time_after64(now, trans->last_begin_time + BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS))) {
3113 bch2_trans_unlock(trans); 3081 bch2_trans_unlock(trans);
3114 cond_resched(); 3082 cond_resched();
3115 now = local_clock(); 3083 now = local_clock();
3116 } 3084 }
3117 trans->last_begin_time = now; 3085 trans->last_begin_time = now;
3118 3086
3119 if (unlikely(trans->srcu_held && 3087 if (unlikely(trans->srcu_held &&
3120 time_after(jiffies, tran 3088 time_after(jiffies, trans->srcu_lock_time + msecs_to_jiffies(10))))
3121 bch2_trans_srcu_unlock(trans) 3089 bch2_trans_srcu_unlock(trans);
3122 3090
3123 trans->last_begin_ip = _RET_IP_; 3091 trans->last_begin_ip = _RET_IP_;
3124 3092
3125 trans_set_locked(trans); 3093 trans_set_locked(trans);
3126 3094
3127 if (trans->restarted) { 3095 if (trans->restarted) {
3128 bch2_btree_path_traverse_all( 3096 bch2_btree_path_traverse_all(trans);
3129 trans->notrace_relock_fail = 3097 trans->notrace_relock_fail = false;
3130 } 3098 }
3131 3099
3132 bch2_trans_verify_not_unlocked(trans) 3100 bch2_trans_verify_not_unlocked(trans);
3133 return trans->restart_count; 3101 return trans->restart_count;
3134 } 3102 }
3135 3103
3136 const char *bch2_btree_transaction_fns[BCH_TR 3104 const char *bch2_btree_transaction_fns[BCH_TRANSACTIONS_NR] = { "(unknown)" };
3137 3105
3138 unsigned bch2_trans_get_fn_idx(const char *fn 3106 unsigned bch2_trans_get_fn_idx(const char *fn)
3139 { 3107 {
3140 for (unsigned i = 0; i < ARRAY_SIZE(b 3108 for (unsigned i = 0; i < ARRAY_SIZE(bch2_btree_transaction_fns); i++)
3141 if (!bch2_btree_transaction_f 3109 if (!bch2_btree_transaction_fns[i] ||
3142 bch2_btree_transaction_fn 3110 bch2_btree_transaction_fns[i] == fn) {
3143 bch2_btree_transactio 3111 bch2_btree_transaction_fns[i] = fn;
3144 return i; 3112 return i;
3145 } 3113 }
3146 3114
3147 pr_warn_once("BCH_TRANSACTIONS_NR not 3115 pr_warn_once("BCH_TRANSACTIONS_NR not big enough!");
3148 return 0; 3116 return 0;
3149 } 3117 }
3150 3118
3151 struct btree_trans *__bch2_trans_get(struct b 3119 struct btree_trans *__bch2_trans_get(struct bch_fs *c, unsigned fn_idx)
3152 __acquires(&c->btree_trans_barrier) 3120 __acquires(&c->btree_trans_barrier)
3153 { 3121 {
3154 struct btree_trans *trans; 3122 struct btree_trans *trans;
3155 3123
3156 if (IS_ENABLED(__KERNEL__)) { 3124 if (IS_ENABLED(__KERNEL__)) {
3157 trans = this_cpu_xchg(c->btre 3125 trans = this_cpu_xchg(c->btree_trans_bufs->trans, NULL);
3158 if (trans) { 3126 if (trans) {
3159 memset(trans, 0, offs 3127 memset(trans, 0, offsetof(struct btree_trans, list));
3160 goto got_trans; 3128 goto got_trans;
3161 } 3129 }
3162 } 3130 }
3163 3131
3164 trans = mempool_alloc(&c->btree_trans 3132 trans = mempool_alloc(&c->btree_trans_pool, GFP_NOFS);
3165 memset(trans, 0, sizeof(*trans)); 3133 memset(trans, 0, sizeof(*trans));
3166 3134
3167 seqmutex_lock(&c->btree_trans_lock); 3135 seqmutex_lock(&c->btree_trans_lock);
3168 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) 3136 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
3169 struct btree_trans *pos; 3137 struct btree_trans *pos;
3170 pid_t pid = current->pid; 3138 pid_t pid = current->pid;
3171 3139
3172 trans->locking_wait.task = cu 3140 trans->locking_wait.task = current;
3173 3141
3174 list_for_each_entry(pos, &c-> 3142 list_for_each_entry(pos, &c->btree_trans_list, list) {
3175 struct task_struct *p 3143 struct task_struct *pos_task = READ_ONCE(pos->locking_wait.task);
3176 /* 3144 /*
3177 * We'd much prefer t 3145 * We'd much prefer to be stricter here and completely
3178 * disallow multiple 3146 * disallow multiple btree_trans in the same thread -
3179 * but the data move 3147 * but the data move path calls bch2_write when we
3180 * already have a btr 3148 * already have a btree_trans initialized.
3181 */ 3149 */
3182 BUG_ON(pos_task && 3150 BUG_ON(pos_task &&
3183 pid == pos_tas 3151 pid == pos_task->pid &&
3184 pos->locked); 3152 pos->locked);
3185 } 3153 }
3186 } 3154 }
3187 3155
3188 list_add(&trans->list, &c->btree_tran 3156 list_add(&trans->list, &c->btree_trans_list);
3189 seqmutex_unlock(&c->btree_trans_lock) 3157 seqmutex_unlock(&c->btree_trans_lock);
3190 got_trans: 3158 got_trans:
3191 trans->c = c; 3159 trans->c = c;
3192 trans->last_begin_time = local_clock 3160 trans->last_begin_time = local_clock();
3193 trans->fn_idx = fn_idx; 3161 trans->fn_idx = fn_idx;
3194 trans->locking_wait.task = current; 3162 trans->locking_wait.task = current;
3195 trans->journal_replay_not_finished = 3163 trans->journal_replay_not_finished =
3196 unlikely(!test_bit(JOURNAL_re 3164 unlikely(!test_bit(JOURNAL_replay_done, &c->journal.flags)) &&
3197 atomic_inc_not_zero(&c->journ 3165 atomic_inc_not_zero(&c->journal_keys.ref);
3198 trans->nr_paths = ARRAY_SIZE( 3166 trans->nr_paths = ARRAY_SIZE(trans->_paths);
3199 trans->paths_allocated = trans->_pat 3167 trans->paths_allocated = trans->_paths_allocated;
3200 trans->sorted = trans->_sor 3168 trans->sorted = trans->_sorted;
3201 trans->paths = trans->_pat 3169 trans->paths = trans->_paths;
3202 trans->updates = trans->_upd 3170 trans->updates = trans->_updates;
3203 3171
3204 *trans_paths_nr(trans->paths) = BTREE 3172 *trans_paths_nr(trans->paths) = BTREE_ITER_INITIAL;
3205 3173
3206 trans->paths_allocated[0] = 1; 3174 trans->paths_allocated[0] = 1;
3207 3175
3208 static struct lock_class_key lockdep_ <<
3209 lockdep_init_map(&trans->dep_map, "bc <<
3210 <<
3211 if (fn_idx < BCH_TRANSACTIONS_NR) { 3176 if (fn_idx < BCH_TRANSACTIONS_NR) {
3212 trans->fn = bch2_btree_transa 3177 trans->fn = bch2_btree_transaction_fns[fn_idx];
3213 3178
3214 struct btree_transaction_stat 3179 struct btree_transaction_stats *s = &c->btree_transaction_stats[fn_idx];
3215 3180
3216 if (s->max_mem) { 3181 if (s->max_mem) {
3217 unsigned expected_mem 3182 unsigned expected_mem_bytes = roundup_pow_of_two(s->max_mem);
3218 3183
3219 trans->mem = kmalloc( 3184 trans->mem = kmalloc(expected_mem_bytes, GFP_KERNEL);
3220 if (likely(trans->mem 3185 if (likely(trans->mem))
3221 trans->mem_by 3186 trans->mem_bytes = expected_mem_bytes;
3222 } 3187 }
3223 3188
3224 trans->nr_paths_max = s->nr_m 3189 trans->nr_paths_max = s->nr_max_paths;
3225 trans->journal_entries_size = 3190 trans->journal_entries_size = s->journal_entries_size;
3226 } 3191 }
3227 3192
3228 trans->srcu_idx = srcu_read_l 3193 trans->srcu_idx = srcu_read_lock(&c->btree_trans_barrier);
3229 trans->srcu_lock_time = jiffies; 3194 trans->srcu_lock_time = jiffies;
3230 trans->srcu_held = true; 3195 trans->srcu_held = true;
3231 trans_set_locked(trans); 3196 trans_set_locked(trans);
3232 3197
3233 closure_init_stack_release(&trans->re 3198 closure_init_stack_release(&trans->ref);
3234 return trans; 3199 return trans;
3235 } 3200 }
3236 3201
3237 static void check_btree_paths_leaked(struct b 3202 static void check_btree_paths_leaked(struct btree_trans *trans)
3238 { 3203 {
3239 #ifdef CONFIG_BCACHEFS_DEBUG 3204 #ifdef CONFIG_BCACHEFS_DEBUG
3240 struct bch_fs *c = trans->c; 3205 struct bch_fs *c = trans->c;
3241 struct btree_path *path; 3206 struct btree_path *path;
3242 unsigned i; 3207 unsigned i;
3243 3208
3244 trans_for_each_path(trans, path, i) 3209 trans_for_each_path(trans, path, i)
3245 if (path->ref) 3210 if (path->ref)
3246 goto leaked; 3211 goto leaked;
3247 return; 3212 return;
3248 leaked: 3213 leaked:
3249 bch_err(c, "btree paths leaked from % 3214 bch_err(c, "btree paths leaked from %s!", trans->fn);
3250 trans_for_each_path(trans, path, i) 3215 trans_for_each_path(trans, path, i)
3251 if (path->ref) 3216 if (path->ref)
3252 printk(KERN_ERR " bt 3217 printk(KERN_ERR " btree %s %pS\n",
3253 bch2_btree_id_ 3218 bch2_btree_id_str(path->btree_id),
3254 (void *) path- 3219 (void *) path->ip_allocated);
3255 /* Be noisy about this: */ 3220 /* Be noisy about this: */
3256 bch2_fatal_error(c); 3221 bch2_fatal_error(c);
3257 #endif 3222 #endif
3258 } 3223 }
3259 3224
3260 void bch2_trans_put(struct btree_trans *trans 3225 void bch2_trans_put(struct btree_trans *trans)
3261 __releases(&c->btree_trans_barrier) 3226 __releases(&c->btree_trans_barrier)
3262 { 3227 {
3263 struct bch_fs *c = trans->c; 3228 struct bch_fs *c = trans->c;
3264 3229
3265 bch2_trans_unlock(trans); 3230 bch2_trans_unlock(trans);
3266 3231
3267 trans_for_each_update(trans, i) 3232 trans_for_each_update(trans, i)
3268 __btree_path_put(trans, trans !! 3233 __btree_path_put(trans->paths + i->path, true);
3269 trans->nr_updates = 0; 3234 trans->nr_updates = 0;
3270 3235
3271 check_btree_paths_leaked(trans); 3236 check_btree_paths_leaked(trans);
3272 3237
3273 if (trans->srcu_held) { 3238 if (trans->srcu_held) {
3274 check_srcu_held_too_long(tran 3239 check_srcu_held_too_long(trans);
3275 srcu_read_unlock(&c->btree_tr 3240 srcu_read_unlock(&c->btree_trans_barrier, trans->srcu_idx);
3276 } 3241 }
3277 3242
>> 3243 if (trans->fs_usage_deltas) {
>> 3244 if (trans->fs_usage_deltas->size + sizeof(trans->fs_usage_deltas) ==
>> 3245 REPLICAS_DELTA_LIST_MAX)
>> 3246 mempool_free(trans->fs_usage_deltas,
>> 3247 &c->replicas_delta_pool);
>> 3248 else
>> 3249 kfree(trans->fs_usage_deltas);
>> 3250 }
>> 3251
3278 if (unlikely(trans->journal_replay_no 3252 if (unlikely(trans->journal_replay_not_finished))
3279 bch2_journal_keys_put(c); 3253 bch2_journal_keys_put(c);
3280 3254
3281 /* 3255 /*
3282 * trans->ref protects trans->locking 3256 * trans->ref protects trans->locking_wait.task, btree_paths array; used
3283 * by cycle detector 3257 * by cycle detector
3284 */ 3258 */
3285 closure_return_sync(&trans->ref); 3259 closure_return_sync(&trans->ref);
3286 trans->locking_wait.task = NULL; 3260 trans->locking_wait.task = NULL;
3287 3261
3288 unsigned long *paths_allocated = tran 3262 unsigned long *paths_allocated = trans->paths_allocated;
3289 trans->paths_allocated = NULL; 3263 trans->paths_allocated = NULL;
3290 trans->paths = NULL; 3264 trans->paths = NULL;
3291 3265
3292 if (paths_allocated != trans->_paths_ 3266 if (paths_allocated != trans->_paths_allocated)
3293 kvfree_rcu_mightsleep(paths_a 3267 kvfree_rcu_mightsleep(paths_allocated);
3294 3268
3295 if (trans->used_mempool) 3269 if (trans->used_mempool)
3296 mempool_free(trans->mem, &c-> 3270 mempool_free(trans->mem, &c->btree_trans_mem_pool);
3297 else 3271 else
3298 kfree(trans->mem); 3272 kfree(trans->mem);
3299 3273
3300 /* Userspace doesn't have a real perc 3274 /* Userspace doesn't have a real percpu implementation: */
3301 if (IS_ENABLED(__KERNEL__)) 3275 if (IS_ENABLED(__KERNEL__))
3302 trans = this_cpu_xchg(c->btre 3276 trans = this_cpu_xchg(c->btree_trans_bufs->trans, trans);
3303 3277
3304 if (trans) { 3278 if (trans) {
3305 seqmutex_lock(&c->btree_trans 3279 seqmutex_lock(&c->btree_trans_lock);
3306 list_del(&trans->list); 3280 list_del(&trans->list);
3307 seqmutex_unlock(&c->btree_tra 3281 seqmutex_unlock(&c->btree_trans_lock);
3308 3282
3309 mempool_free(trans, &c->btree 3283 mempool_free(trans, &c->btree_trans_pool);
3310 } 3284 }
3311 } 3285 }
3312 3286
3313 bool bch2_current_has_btree_trans(struct bch_ <<
3314 { <<
3315 seqmutex_lock(&c->btree_trans_lock); <<
3316 struct btree_trans *trans; <<
3317 bool ret = false; <<
3318 list_for_each_entry(trans, &c->btree_ <<
3319 if (trans->locking_wait.task <<
3320 trans->locked) { <<
3321 ret = true; <<
3322 break; <<
3323 } <<
3324 seqmutex_unlock(&c->btree_trans_lock) <<
3325 return ret; <<
3326 } <<
3327 <<
3328 static void __maybe_unused 3287 static void __maybe_unused
3329 bch2_btree_bkey_cached_common_to_text(struct 3288 bch2_btree_bkey_cached_common_to_text(struct printbuf *out,
3330 struct 3289 struct btree_bkey_cached_common *b)
3331 { 3290 {
3332 struct six_lock_count c = six_lock_co 3291 struct six_lock_count c = six_lock_counts(&b->lock);
3333 struct task_struct *owner; 3292 struct task_struct *owner;
3334 pid_t pid; 3293 pid_t pid;
3335 3294
3336 rcu_read_lock(); 3295 rcu_read_lock();
3337 owner = READ_ONCE(b->lock.owner); 3296 owner = READ_ONCE(b->lock.owner);
3338 pid = owner ? owner->pid : 0; 3297 pid = owner ? owner->pid : 0;
3339 rcu_read_unlock(); 3298 rcu_read_unlock();
3340 3299
3341 prt_printf(out, "\t%px %c l=%u %s:", 3300 prt_printf(out, "\t%px %c l=%u %s:", b, b->cached ? 'c' : 'b',
3342 b->level, bch2_btree_id_st 3301 b->level, bch2_btree_id_str(b->btree_id));
3343 bch2_bpos_to_text(out, btree_node_pos 3302 bch2_bpos_to_text(out, btree_node_pos(b));
3344 3303
3345 prt_printf(out, "\t locks %u:%u:%u he 3304 prt_printf(out, "\t locks %u:%u:%u held by pid %u",
3346 c.n[0], c.n[1], c.n[2], pi 3305 c.n[0], c.n[1], c.n[2], pid);
3347 } 3306 }
3348 3307
3349 void bch2_btree_trans_to_text(struct printbuf 3308 void bch2_btree_trans_to_text(struct printbuf *out, struct btree_trans *trans)
3350 { 3309 {
3351 struct btree_bkey_cached_common *b; 3310 struct btree_bkey_cached_common *b;
3352 static char lock_types[] = { 'r', 'i' 3311 static char lock_types[] = { 'r', 'i', 'w' };
3353 struct task_struct *task = READ_ONCE( 3312 struct task_struct *task = READ_ONCE(trans->locking_wait.task);
3354 unsigned l, idx; 3313 unsigned l, idx;
3355 3314
3356 /* before rcu_read_lock(): */ 3315 /* before rcu_read_lock(): */
3357 bch2_printbuf_make_room(out, 4096); 3316 bch2_printbuf_make_room(out, 4096);
3358 3317
3359 if (!out->nr_tabstops) { 3318 if (!out->nr_tabstops) {
3360 printbuf_tabstop_push(out, 16 3319 printbuf_tabstop_push(out, 16);
3361 printbuf_tabstop_push(out, 32 3320 printbuf_tabstop_push(out, 32);
3362 } 3321 }
3363 3322
3364 prt_printf(out, "%i %s\n", task ? tas 3323 prt_printf(out, "%i %s\n", task ? task->pid : 0, trans->fn);
3365 3324
3366 /* trans->paths is rcu protected vs. 3325 /* trans->paths is rcu protected vs. freeing */
3367 rcu_read_lock(); 3326 rcu_read_lock();
3368 out->atomic++; 3327 out->atomic++;
3369 3328
3370 struct btree_path *paths = rcu_derefe 3329 struct btree_path *paths = rcu_dereference(trans->paths);
3371 if (!paths) 3330 if (!paths)
3372 goto out; 3331 goto out;
3373 3332
3374 unsigned long *paths_allocated = tran 3333 unsigned long *paths_allocated = trans_paths_allocated(paths);
3375 3334
3376 trans_for_each_path_idx_from(paths_al 3335 trans_for_each_path_idx_from(paths_allocated, *trans_paths_nr(paths), idx, 1) {
3377 struct btree_path *path = pat 3336 struct btree_path *path = paths + idx;
3378 if (!path->nodes_locked) 3337 if (!path->nodes_locked)
3379 continue; 3338 continue;
3380 3339
3381 prt_printf(out, " path %u %c 3340 prt_printf(out, " path %u %c l=%u %s:",
3382 idx, 3341 idx,
3383 path->cached ? 'c' : ' 3342 path->cached ? 'c' : 'b',
3384 path->level, 3343 path->level,
3385 bch2_btree_id_str(path 3344 bch2_btree_id_str(path->btree_id));
3386 bch2_bpos_to_text(out, path-> 3345 bch2_bpos_to_text(out, path->pos);
3387 prt_newline(out); 3346 prt_newline(out);
3388 3347
3389 for (l = 0; l < BTREE_MAX_DEP 3348 for (l = 0; l < BTREE_MAX_DEPTH; l++) {
3390 if (btree_node_locked 3349 if (btree_node_locked(path, l) &&
3391 !IS_ERR_OR_NULL(b 3350 !IS_ERR_OR_NULL(b = (void *) READ_ONCE(path->l[l].b))) {
3392 prt_printf(ou 3351 prt_printf(out, " %c l=%u ",
3393 lo 3352 lock_types[btree_node_locked_type(path, l)], l);
3394 bch2_btree_bk 3353 bch2_btree_bkey_cached_common_to_text(out, b);
3395 prt_newline(o 3354 prt_newline(out);
3396 } 3355 }
3397 } 3356 }
3398 } 3357 }
3399 3358
3400 b = READ_ONCE(trans->locking); 3359 b = READ_ONCE(trans->locking);
3401 if (b) { 3360 if (b) {
3402 prt_printf(out, " blocked fo 3361 prt_printf(out, " blocked for %lluus on\n",
3403 div_u64(local_cloc 3362 div_u64(local_clock() - trans->locking_wait.start_time, 1000));
3404 prt_printf(out, " %c", loc 3363 prt_printf(out, " %c", lock_types[trans->locking_wait.lock_want]);
3405 bch2_btree_bkey_cached_common 3364 bch2_btree_bkey_cached_common_to_text(out, b);
3406 prt_newline(out); 3365 prt_newline(out);
3407 } 3366 }
3408 out: 3367 out:
3409 --out->atomic; 3368 --out->atomic;
3410 rcu_read_unlock(); 3369 rcu_read_unlock();
3411 } 3370 }
3412 3371
3413 void bch2_fs_btree_iter_exit(struct bch_fs *c 3372 void bch2_fs_btree_iter_exit(struct bch_fs *c)
3414 { 3373 {
3415 struct btree_transaction_stats *s; 3374 struct btree_transaction_stats *s;
3416 struct btree_trans *trans; 3375 struct btree_trans *trans;
3417 int cpu; 3376 int cpu;
3418 3377
3419 if (c->btree_trans_bufs) 3378 if (c->btree_trans_bufs)
3420 for_each_possible_cpu(cpu) { 3379 for_each_possible_cpu(cpu) {
3421 struct btree_trans *t 3380 struct btree_trans *trans =
3422 per_cpu_ptr(c 3381 per_cpu_ptr(c->btree_trans_bufs, cpu)->trans;
3423 3382
3424 if (trans) { 3383 if (trans) {
3425 seqmutex_lock 3384 seqmutex_lock(&c->btree_trans_lock);
3426 list_del(&tra 3385 list_del(&trans->list);
3427 seqmutex_unlo 3386 seqmutex_unlock(&c->btree_trans_lock);
3428 } 3387 }
3429 kfree(trans); 3388 kfree(trans);
3430 } 3389 }
3431 free_percpu(c->btree_trans_bufs); 3390 free_percpu(c->btree_trans_bufs);
3432 3391
3433 trans = list_first_entry_or_null(&c-> 3392 trans = list_first_entry_or_null(&c->btree_trans_list, struct btree_trans, list);
3434 if (trans) 3393 if (trans)
3435 panic("%s leaked btree_trans\ 3394 panic("%s leaked btree_trans\n", trans->fn);
3436 3395
3437 for (s = c->btree_transaction_stats; 3396 for (s = c->btree_transaction_stats;
3438 s < c->btree_transaction_stats + 3397 s < c->btree_transaction_stats + ARRAY_SIZE(c->btree_transaction_stats);
3439 s++) { 3398 s++) {
3440 kfree(s->max_paths_text); 3399 kfree(s->max_paths_text);
3441 bch2_time_stats_exit(&s->lock 3400 bch2_time_stats_exit(&s->lock_hold_times);
3442 } 3401 }
3443 3402
3444 if (c->btree_trans_barrier_initialize 3403 if (c->btree_trans_barrier_initialized) {
3445 synchronize_srcu_expedited(&c 3404 synchronize_srcu_expedited(&c->btree_trans_barrier);
3446 cleanup_srcu_struct(&c->btree 3405 cleanup_srcu_struct(&c->btree_trans_barrier);
3447 } 3406 }
3448 mempool_exit(&c->btree_trans_mem_pool 3407 mempool_exit(&c->btree_trans_mem_pool);
3449 mempool_exit(&c->btree_trans_pool); 3408 mempool_exit(&c->btree_trans_pool);
3450 } 3409 }
3451 3410
3452 void bch2_fs_btree_iter_init_early(struct bch 3411 void bch2_fs_btree_iter_init_early(struct bch_fs *c)
3453 { 3412 {
3454 struct btree_transaction_stats *s; 3413 struct btree_transaction_stats *s;
3455 3414
3456 for (s = c->btree_transaction_stats; 3415 for (s = c->btree_transaction_stats;
3457 s < c->btree_transaction_stats + 3416 s < c->btree_transaction_stats + ARRAY_SIZE(c->btree_transaction_stats);
3458 s++) { 3417 s++) {
3459 bch2_time_stats_init(&s->dura 3418 bch2_time_stats_init(&s->duration);
3460 bch2_time_stats_init(&s->lock 3419 bch2_time_stats_init(&s->lock_hold_times);
3461 mutex_init(&s->lock); 3420 mutex_init(&s->lock);
3462 } 3421 }
3463 3422
3464 INIT_LIST_HEAD(&c->btree_trans_list); 3423 INIT_LIST_HEAD(&c->btree_trans_list);
3465 seqmutex_init(&c->btree_trans_lock); 3424 seqmutex_init(&c->btree_trans_lock);
3466 } 3425 }
3467 3426
3468 int bch2_fs_btree_iter_init(struct bch_fs *c) 3427 int bch2_fs_btree_iter_init(struct bch_fs *c)
3469 { 3428 {
3470 int ret; 3429 int ret;
3471 3430
3472 c->btree_trans_bufs = alloc_percpu(st 3431 c->btree_trans_bufs = alloc_percpu(struct btree_trans_buf);
3473 if (!c->btree_trans_bufs) 3432 if (!c->btree_trans_bufs)
3474 return -ENOMEM; 3433 return -ENOMEM;
3475 3434
3476 ret = mempool_init_kmalloc_pool(&c- 3435 ret = mempool_init_kmalloc_pool(&c->btree_trans_pool, 1,
3477 siz 3436 sizeof(struct btree_trans)) ?:
3478 mempool_init_kmalloc_pool(&c- 3437 mempool_init_kmalloc_pool(&c->btree_trans_mem_pool, 1,
3479 BTR 3438 BTREE_TRANS_MEM_MAX) ?:
3480 init_srcu_struct(&c->btree_tr 3439 init_srcu_struct(&c->btree_trans_barrier);
3481 if (ret) !! 3440 if (!ret)
3482 return ret; !! 3441 c->btree_trans_barrier_initialized = true;
3483 !! 3442 return ret;
3484 /* <<
3485 * static annotation (hackily done) f <<
3486 * btree node locks: <<
3487 */ <<
3488 #ifdef CONFIG_LOCKDEP <<
3489 fs_reclaim_acquire(GFP_KERNEL); <<
3490 struct btree_trans *trans = bch2_tran <<
3491 trans_set_locked(trans); <<
3492 bch2_trans_put(trans); <<
3493 fs_reclaim_release(GFP_KERNEL); <<
3494 #endif <<
3495 <<
3496 c->btree_trans_barrier_initialized = <<
3497 return 0; <<
3498 <<
3499 } 3443 }
3500 3444
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.