git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/btree_write_buffer.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "btree_locking.h"
   5 #include "btree_update.h"
   6 #include "btree_update_interior.h"
   7 #include "btree_write_buffer.h"
   8 #include "error.h"
   9 #include "journal.h"
  10 #include "journal_reclaim.h"
  11
  12 #include <linux/sort.h>
  13
  14 static int btree_write_buffered_key_cmp(const void *_l, const void *_r)
  15 {
  16         const struct btree_write_buffered_key *l = _l;
  17         const struct btree_write_buffered_key *r = _r;
  18
  19         return  cmp_int(l->btree, r->btree) ?:
  20                 bpos_cmp(l->k.k.p, r->k.k.p) ?:
  21                 cmp_int(l->journal_seq, r->journal_seq) ?:
  22                 cmp_int(l->journal_offset, r->journal_offset);
  23 }
  24
  25 static int btree_write_buffered_journal_cmp(const void *_l, const void *_r)
  26 {
  27         const struct btree_write_buffered_key *l = _l;
  28         const struct btree_write_buffered_key *r = _r;
  29
  30         return  cmp_int(l->journal_seq, r->journal_seq);
  31 }
  32
  33 static int bch2_btree_write_buffer_flush_one(struct btree_trans *trans,
  34                                              struct btree_iter *iter,
  35                                              struct btree_write_buffered_key *wb,
  36                                              unsigned commit_flags,
  37                                              bool *write_locked,
  38                                              size_t *fast)
  39 {
  40         struct bch_fs *c = trans->c;
  41         struct btree_path *path;
  42         int ret;
  43
  44         ret = bch2_btree_iter_traverse(iter);
  45         if (ret)
  46                 return ret;
  47
  48         path = iter->path;
  49
  50         if (!*write_locked) {
  51                 ret = bch2_btree_node_lock_write(trans, path, &path->l[0].b->c);
  52                 if (ret)
  53                         return ret;
  54
  55                 bch2_btree_node_prep_for_write(trans, path, path->l[0].b);
  56                 *write_locked = true;
  57         }
  58
  59         if (!bch2_btree_node_insert_fits(c, path->l[0].b, wb->k.k.u64s)) {
  60                 bch2_btree_node_unlock_write(trans, path, path->l[0].b);
  61                 *write_locked = false;
  62                 goto trans_commit;
  63         }
  64
  65         bch2_btree_insert_key_leaf(trans, path, &wb->k, wb->journal_seq);
  66         (*fast)++;
  67
  68         if (path->ref > 1) {
  69                 /*
  70                  * We can't clone a path that has write locks: if the path is
  71                  * shared, unlock before set_pos(), traverse():
  72                  */
  73                 bch2_btree_node_unlock_write(trans, path, path->l[0].b);
  74                 *write_locked = false;
  75         }
  76         return 0;
  77 trans_commit:
  78         return  bch2_trans_update(trans, iter, &wb->k, 0) ?:
  79                 bch2_trans_commit(trans, NULL, NULL,
  80                                   commit_flags|
  81                                   BTREE_INSERT_NOFAIL|
  82                                   BTREE_INSERT_JOURNAL_RECLAIM);
  83 }
  84
  85 static union btree_write_buffer_state btree_write_buffer_switch(struct btree_write_buffer *wb)
  86 {
  87         union btree_write_buffer_state old, new;
  88         u64 v = READ_ONCE(wb->state.v);
  89
  90         do {
  91                 old.v = new.v = v;
  92
  93                 new.nr = 0;
  94                 new.idx++;
  95         } while ((v = atomic64_cmpxchg_acquire(&wb->state.counter, old.v, new.v)) != old.v);
  96
  97         while (old.idx == 0 ? wb->state.ref0 : wb->state.ref1)
  98                 cpu_relax();
  99
 100         smp_mb();
 101
 102         return old;
 103 }
 104
 105 int __bch2_btree_write_buffer_flush(struct btree_trans *trans, unsigned commit_flags,
 106                                     bool locked)
 107 {
 108         struct bch_fs *c = trans->c;
 109         struct journal *j = &c->journal;
 110         struct btree_write_buffer *wb = &c->btree_write_buffer;
 111         struct journal_entry_pin pin;
 112         struct btree_write_buffered_key *i, *dst, *keys;
 113         struct btree_iter iter = { NULL };
 114         size_t nr = 0, skipped = 0, fast = 0;
 115         bool write_locked = false;
 116         union btree_write_buffer_state s;
 117         int ret = 0;
 118
 119         memset(&pin, 0, sizeof(pin));
 120
 121         if (!locked && !mutex_trylock(&wb->flush_lock))
 122                 return 0;
 123
 124         bch2_journal_pin_copy(j, &pin, &wb->journal_pin, NULL);
 125         bch2_journal_pin_drop(j, &wb->journal_pin);
 126
 127         s = btree_write_buffer_switch(wb);
 128         keys = wb->keys[s.idx];
 129         nr = s.nr;
 130
 131         /*
 132          * We first sort so that we can detect and skip redundant updates, and
 133          * then we attempt to flush in sorted btree order, as this is most
 134          * efficient.
 135          *
 136          * However, since we're not flushing in the order they appear in the
 137          * journal we won't be able to drop our journal pin until everything is
 138          * flushed - which means this could deadlock the journal, if we weren't
 139          * passing BTREE_INSERT_JORUNAL_RECLAIM. This causes the update to fail
 140          * if it would block taking a journal reservation.
 141          *
 142          * If that happens, we sort them by the order they appeared in the
 143          * journal - after dropping redundant entries - and then restart
 144          * flushing, this time dropping journal pins as we go.
 145          */
 146
 147         sort(keys, nr, sizeof(keys[0]),
 148              btree_write_buffered_key_cmp, NULL);
 149
 150         for (i = keys; i < keys + nr; i++) {
 151                 if (i + 1 < keys + nr &&
 152                     i[0].btree == i[1].btree &&
 153                     bpos_eq(i[0].k.k.p, i[1].k.k.p)) {
 154                         skipped++;
 155                         continue;
 156                 }
 157
 158                 if (write_locked &&
 159                     (iter.path->btree_id != i->btree ||
 160                      bpos_gt(i->k.k.p, iter.path->l[0].b->key.k.p))) {
 161                         bch2_btree_node_unlock_write(trans, iter.path, iter.path->l[0].b);
 162                         write_locked = false;
 163                 }
 164
 165                 if (!iter.path || iter.path->btree_id != i->btree) {
 166                         bch2_trans_iter_exit(trans, &iter);
 167                         bch2_trans_iter_init(trans, &iter, i->btree, i->k.k.p, BTREE_ITER_INTENT);
 168                 }
 169
 170                 bch2_btree_iter_set_pos(&iter, i->k.k.p);
 171                 iter.path->preserve = false;
 172
 173                 do {
 174                         ret = bch2_btree_write_buffer_flush_one(trans, &iter, i,
 175                                                 commit_flags, &write_locked, &fast);
 176                         if (!write_locked)
 177                                 bch2_trans_begin(trans);
 178                 } while (bch2_err_matches(ret, BCH_ERR_transaction_restart));
 179
 180                 if (ret)
 181                         break;
 182         }
 183
 184         if (write_locked)
 185                 bch2_btree_node_unlock_write(trans, iter.path, iter.path->l[0].b);
 186         bch2_trans_iter_exit(trans, &iter);
 187
 188         trace_write_buffer_flush(trans, nr, skipped, fast, wb->size);
 189
 190         if (ret == -BCH_ERR_journal_reclaim_would_deadlock)
 191                 goto slowpath;
 192
 193         bch2_fs_fatal_err_on(ret, c, "%s: insert error %s", __func__, bch2_err_str(ret));
 194 out:
 195         bch2_journal_pin_drop(j, &pin);
 196         mutex_unlock(&wb->flush_lock);
 197         return ret;
 198 slowpath:
 199         trace_write_buffer_flush_slowpath(trans, i - keys, nr);
 200
 201         dst = keys;
 202         for (; i < keys + nr; i++) {
 203                 if (i + 1 < keys + nr &&
 204                     i[0].btree == i[1].btree &&
 205                     bpos_eq(i[0].k.k.p, i[1].k.k.p))
 206                         continue;
 207
 208                 *dst = *i;
 209                 dst++;
 210         }
 211         nr = dst - keys;
 212
 213         sort(keys, nr, sizeof(keys[0]),
 214              btree_write_buffered_journal_cmp,
 215              NULL);
 216
 217         for (i = keys; i < keys + nr; i++) {
 218                 if (i->journal_seq > pin.seq) {
 219                         struct journal_entry_pin pin2;
 220
 221                         memset(&pin2, 0, sizeof(pin2));
 222
 223                         bch2_journal_pin_add(j, i->journal_seq, &pin2, NULL);
 224                         bch2_journal_pin_drop(j, &pin);
 225                         bch2_journal_pin_copy(j, &pin, &pin2, NULL);
 226                         bch2_journal_pin_drop(j, &pin2);
 227                 }
 228
 229                 ret = commit_do(trans, NULL, NULL,
 230                                 commit_flags|
 231                                 BTREE_INSERT_NOFAIL|
 232                                 BTREE_INSERT_JOURNAL_RECLAIM|
 233                                 JOURNAL_WATERMARK_reserved,
 234                                 __bch2_btree_insert(trans, i->btree, &i->k, 0));
 235                 if (bch2_fs_fatal_err_on(ret, c, "%s: insert error %s", __func__, bch2_err_str(ret)))
 236                         break;
 237         }
 238
 239         goto out;
 240 }
 241
 242 int bch2_btree_write_buffer_flush_sync(struct btree_trans *trans)
 243 {
 244         bch2_trans_unlock(trans);
 245         mutex_lock(&trans->c->btree_write_buffer.flush_lock);
 246         return __bch2_btree_write_buffer_flush(trans, 0, true);
 247 }
 248
 249 int bch2_btree_write_buffer_flush(struct btree_trans *trans)
 250 {
 251         return __bch2_btree_write_buffer_flush(trans, 0, false);
 252 }
 253
 254 static int bch2_btree_write_buffer_journal_flush(struct journal *j,
 255                                 struct journal_entry_pin *_pin, u64 seq)
 256 {
 257         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 258         struct btree_write_buffer *wb = &c->btree_write_buffer;
 259
 260         mutex_lock(&wb->flush_lock);
 261
 262         return bch2_trans_run(c,
 263                         __bch2_btree_write_buffer_flush(&trans, BTREE_INSERT_NOCHECK_RW, true));
 264 }
 265
 266 static inline u64 btree_write_buffer_ref(int idx)
 267 {
 268         return ((union btree_write_buffer_state) {
 269                 .ref0 = idx == 0,
 270                 .ref1 = idx == 1,
 271         }).v;
 272 }
 273
 274 int bch2_btree_insert_keys_write_buffer(struct btree_trans *trans)
 275 {
 276         struct bch_fs *c = trans->c;
 277         struct btree_write_buffer *wb = &c->btree_write_buffer;
 278         struct btree_write_buffered_key *i;
 279         union btree_write_buffer_state old, new;
 280         int ret = 0;
 281         u64 v;
 282
 283         trans_for_each_wb_update(trans, i) {
 284                 EBUG_ON(i->k.k.u64s > BTREE_WRITE_BUFERED_U64s_MAX);
 285
 286                 i->journal_seq          = trans->journal_res.seq;
 287                 i->journal_offset       = trans->journal_res.offset;
 288         }
 289
 290         preempt_disable();
 291         v = READ_ONCE(wb->state.v);
 292         do {
 293                 old.v = new.v = v;
 294
 295                 new.v += btree_write_buffer_ref(new.idx);
 296                 new.nr += trans->nr_wb_updates;
 297                 if (new.nr > wb->size) {
 298                         ret = -BCH_ERR_btree_insert_need_flush_buffer;
 299                         goto out;
 300                 }
 301         } while ((v = atomic64_cmpxchg_acquire(&wb->state.counter, old.v, new.v)) != old.v);
 302
 303         memcpy(wb->keys[new.idx] + old.nr,
 304                trans->wb_updates,
 305                sizeof(trans->wb_updates[0]) * trans->nr_wb_updates);
 306
 307         bch2_journal_pin_add(&c->journal, trans->journal_res.seq, &wb->journal_pin,
 308                              bch2_btree_write_buffer_journal_flush);
 309
 310         atomic64_sub_return_release(btree_write_buffer_ref(new.idx), &wb->state.counter);
 311 out:
 312         preempt_enable();
 313         return ret;
 314 }
 315
 316 void bch2_fs_btree_write_buffer_exit(struct bch_fs *c)
 317 {
 318         struct btree_write_buffer *wb = &c->btree_write_buffer;
 319
 320         BUG_ON(wb->state.nr && !bch2_journal_error(&c->journal));
 321
 322         kvfree(wb->keys[1]);
 323         kvfree(wb->keys[0]);
 324 }
 325
 326 int bch2_fs_btree_write_buffer_init(struct bch_fs *c)
 327 {
 328         struct btree_write_buffer *wb = &c->btree_write_buffer;
 329
 330         mutex_init(&wb->flush_lock);
 331         wb->size = c->opts.btree_write_buffer_size;
 332
 333         wb->keys[0] = kvmalloc_array(wb->size, sizeof(*wb->keys[0]), GFP_KERNEL);
 334         wb->keys[1] = kvmalloc_array(wb->size, sizeof(*wb->keys[1]), GFP_KERNEL);
 335         if (!wb->keys[0] || !wb->keys[1])
 336                 return -ENOMEM;
 337
 338         return 0;
 339 }