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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "alloc_foreground.h"
   5 #include "bkey_buf.h"
   6 #include "btree_update.h"
   7 #include "buckets.h"
   8 #include "data_update.h"
   9 #include "ec.h"
  10 #include "extents.h"
  11 #include "io.h"
  12 #include "keylist.h"
  13 #include "move.h"
  14 #include "subvolume.h"
  15
  16 #include <trace/events/bcachefs.h>
  17
  18 static int insert_snapshot_whiteouts(struct btree_trans *trans,
  19                                      enum btree_id id,
  20                                      struct bpos old_pos,
  21                                      struct bpos new_pos)
  22 {
  23         struct bch_fs *c = trans->c;
  24         struct btree_iter iter, update_iter;
  25         struct bkey_s_c k;
  26         struct snapshots_seen s;
  27         int ret;
  28
  29         if (!btree_type_has_snapshots(id))
  30                 return 0;
  31
  32         snapshots_seen_init(&s);
  33
  34         if (!bkey_cmp(old_pos, new_pos))
  35                 return 0;
  36
  37         if (!snapshot_t(c, old_pos.snapshot)->children[0])
  38                 return 0;
  39
  40         bch2_trans_iter_init(trans, &iter, id, old_pos,
  41                              BTREE_ITER_NOT_EXTENTS|
  42                              BTREE_ITER_ALL_SNAPSHOTS);
  43         while (1) {
  44 next:
  45                 k = bch2_btree_iter_prev(&iter);
  46                 ret = bkey_err(k);
  47                 if (ret)
  48                         break;
  49
  50                 if (bkey_cmp(old_pos, k.k->p))
  51                         break;
  52
  53                 if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, old_pos.snapshot)) {
  54                         struct bkey_i *update;
  55                         u32 *i;
  56
  57                         darray_for_each(s.ids, i)
  58                                 if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, *i))
  59                                         goto next;
  60
  61                         update = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
  62
  63                         ret = PTR_ERR_OR_ZERO(update);
  64                         if (ret)
  65                                 break;
  66
  67                         bkey_init(&update->k);
  68                         update->k.p = new_pos;
  69                         update->k.p.snapshot = k.k->p.snapshot;
  70
  71                         bch2_trans_iter_init(trans, &update_iter, id, update->k.p,
  72                                              BTREE_ITER_NOT_EXTENTS|
  73                                              BTREE_ITER_ALL_SNAPSHOTS|
  74                                              BTREE_ITER_INTENT);
  75                         ret   = bch2_btree_iter_traverse(&update_iter) ?:
  76                                 bch2_trans_update(trans, &update_iter, update,
  77                                           BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
  78                         bch2_trans_iter_exit(trans, &update_iter);
  79                         if (ret)
  80                                 break;
  81
  82                         ret = snapshots_seen_add(c, &s, k.k->p.snapshot);
  83                         if (ret)
  84                                 break;
  85                 }
  86         }
  87         bch2_trans_iter_exit(trans, &iter);
  88         darray_exit(&s.ids);
  89
  90         return ret;
  91 }
  92
  93 static void bch2_bkey_mark_dev_cached(struct bkey_s k, unsigned dev)
  94 {
  95         struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
  96         struct bch_extent_ptr *ptr;
  97
  98         bkey_for_each_ptr(ptrs, ptr)
  99                 if (ptr->dev == dev)
 100                         ptr->cached = true;
 101 }
 102
 103 static int bch2_data_update_index_update(struct bch_write_op *op)
 104 {
 105         struct bch_fs *c = op->c;
 106         struct btree_trans trans;
 107         struct btree_iter iter;
 108         struct data_update *m =
 109                 container_of(op, struct data_update, op);
 110         struct open_bucket *ec_ob = ec_open_bucket(c, &op->open_buckets);
 111         struct keylist *keys = &op->insert_keys;
 112         struct bkey_buf _new, _insert;
 113         int ret = 0;
 114
 115         bch2_bkey_buf_init(&_new);
 116         bch2_bkey_buf_init(&_insert);
 117         bch2_bkey_buf_realloc(&_insert, c, U8_MAX);
 118
 119         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024);
 120
 121         bch2_trans_iter_init(&trans, &iter, m->btree_id,
 122                              bkey_start_pos(&bch2_keylist_front(keys)->k),
 123                              BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 124
 125         while (1) {
 126                 struct bkey_s_c k;
 127                 struct bkey_s_c old = bkey_i_to_s_c(m->k.k);
 128                 struct bkey_i *insert;
 129                 struct bkey_i_extent *new;
 130                 const union bch_extent_entry *entry;
 131                 struct extent_ptr_decoded p;
 132                 struct bpos next_pos;
 133                 bool did_work = false;
 134                 bool should_check_enospc;
 135                 s64 i_sectors_delta = 0, disk_sectors_delta = 0;
 136                 unsigned i;
 137
 138                 bch2_trans_begin(&trans);
 139
 140                 k = bch2_btree_iter_peek_slot(&iter);
 141                 ret = bkey_err(k);
 142                 if (ret)
 143                         goto err;
 144
 145                 new = bkey_i_to_extent(bch2_keylist_front(keys));
 146
 147                 if (!bch2_extents_match(k, old))
 148                         goto nomatch;
 149
 150                 bkey_reassemble(_insert.k, k);
 151                 insert = _insert.k;
 152
 153                 bch2_bkey_buf_copy(&_new, c, bch2_keylist_front(keys));
 154                 new = bkey_i_to_extent(_new.k);
 155                 bch2_cut_front(iter.pos, &new->k_i);
 156
 157                 bch2_cut_front(iter.pos,        insert);
 158                 bch2_cut_back(new->k.p,         insert);
 159                 bch2_cut_back(insert->k.p,      &new->k_i);
 160
 161                 /*
 162                  * @old: extent that we read from
 163                  * @insert: key that we're going to update, initialized from
 164                  * extent currently in btree - same as @old unless we raced with
 165                  * other updates
 166                  * @new: extent with new pointers that we'll be adding to @insert
 167                  *
 168                  * Fist, drop rewrite_ptrs from @new:
 169                  */
 170                 i = 0;
 171                 bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs_c(old), p, entry) {
 172                         if (((1U << i) & m->data_opts.rewrite_ptrs) &&
 173                             bch2_extent_has_ptr(old, p, bkey_i_to_s_c(insert))) {
 174                                 /*
 175                                  * If we're going to be adding a pointer to the
 176                                  * same device, we have to drop the old one -
 177                                  * otherwise, we can just mark it cached:
 178                                  */
 179                                 if (bch2_bkey_has_device(bkey_i_to_s_c(&new->k_i), p.ptr.dev))
 180                                         bch2_bkey_drop_device_noerror(bkey_i_to_s(insert), p.ptr.dev);
 181                                 else
 182                                         bch2_bkey_mark_dev_cached(bkey_i_to_s(insert), p.ptr.dev);
 183                         }
 184                         i++;
 185                 }
 186
 187
 188                 /* Add new ptrs: */
 189                 extent_for_each_ptr_decode(extent_i_to_s(new), p, entry) {
 190                         if (bch2_bkey_has_device(bkey_i_to_s_c(insert), p.ptr.dev)) {
 191                                 /*
 192                                  * raced with another move op? extent already
 193                                  * has a pointer to the device we just wrote
 194                                  * data to
 195                                  */
 196                                 continue;
 197                         }
 198
 199                         bch2_extent_ptr_decoded_append(insert, &p);
 200                         did_work = true;
 201                 }
 202
 203                 if (!did_work)
 204                         goto nomatch;
 205
 206                 bch2_bkey_narrow_crcs(insert, (struct bch_extent_crc_unpacked) { 0 });
 207                 bch2_extent_normalize(c, bkey_i_to_s(insert));
 208
 209                 ret = bch2_sum_sector_overwrites(&trans, &iter, insert,
 210                                                  &should_check_enospc,
 211                                                  &i_sectors_delta,
 212                                                  &disk_sectors_delta);
 213                 if (ret)
 214                         goto err;
 215
 216                 if (disk_sectors_delta > (s64) op->res.sectors) {
 217                         ret = bch2_disk_reservation_add(c, &op->res,
 218                                                 disk_sectors_delta - op->res.sectors,
 219                                                 !should_check_enospc
 220                                                 ? BCH_DISK_RESERVATION_NOFAIL : 0);
 221                         if (ret)
 222                                 goto out;
 223                 }
 224
 225                 next_pos = insert->k.p;
 226
 227                 ret   = insert_snapshot_whiteouts(&trans, m->btree_id,
 228                                                   k.k->p, insert->k.p) ?:
 229                         bch2_trans_update(&trans, &iter, insert,
 230                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
 231                         bch2_trans_commit(&trans, &op->res,
 232                                 op_journal_seq(op),
 233                                 BTREE_INSERT_NOFAIL|
 234                                 m->data_opts.btree_insert_flags);
 235                 if (!ret) {
 236                         bch2_btree_iter_set_pos(&iter, next_pos);
 237                         atomic_long_inc(&c->extent_migrate_done);
 238                         if (ec_ob)
 239                                 bch2_ob_add_backpointer(c, ec_ob, &insert->k);
 240                 }
 241 err:
 242                 if (ret == -EINTR)
 243                         ret = 0;
 244                 if (ret)
 245                         break;
 246 next:
 247                 while (bkey_cmp(iter.pos, bch2_keylist_front(keys)->k.p) >= 0) {
 248                         bch2_keylist_pop_front(keys);
 249                         if (bch2_keylist_empty(keys))
 250                                 goto out;
 251                 }
 252                 continue;
 253 nomatch:
 254                 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
 255                         struct printbuf buf = PRINTBUF;
 256
 257                         bch2_bkey_val_to_text(&buf, c, old);
 258                         bch_info(c, "no match for %s", buf.buf);
 259                         printbuf_exit(&buf);
 260                 }
 261
 262                 if (m->ctxt) {
 263                         BUG_ON(k.k->p.offset <= iter.pos.offset);
 264                         atomic64_inc(&m->ctxt->stats->keys_raced);
 265                         atomic64_add(k.k->p.offset - iter.pos.offset,
 266                                      &m->ctxt->stats->sectors_raced);
 267                 }
 268                 atomic_long_inc(&c->extent_migrate_raced);
 269                 trace_move_race(&new->k);
 270                 bch2_btree_iter_advance(&iter);
 271                 goto next;
 272         }
 273 out:
 274         bch2_trans_iter_exit(&trans, &iter);
 275         bch2_trans_exit(&trans);
 276         bch2_bkey_buf_exit(&_insert, c);
 277         bch2_bkey_buf_exit(&_new, c);
 278         BUG_ON(ret == -EINTR);
 279         return ret;
 280 }
 281
 282 void bch2_data_update_read_done(struct data_update *m,
 283                                 struct bch_extent_crc_unpacked crc,
 284                                 struct closure *cl)
 285 {
 286         /* write bio must own pages: */
 287         BUG_ON(!m->op.wbio.bio.bi_vcnt);
 288
 289         m->op.crc = crc;
 290         m->op.wbio.bio.bi_iter.bi_size = crc.compressed_size << 9;
 291
 292         closure_call(&m->op.cl, bch2_write, NULL, cl);
 293 }
 294
 295 void bch2_data_update_exit(struct data_update *update)
 296 {
 297         struct bch_fs *c = update->op.c;
 298
 299         bch2_bkey_buf_exit(&update->k, c);
 300         bch2_disk_reservation_put(c, &update->op.res);
 301         bch2_bio_free_pages_pool(c, &update->op.wbio.bio);
 302 }
 303
 304 int bch2_data_update_init(struct bch_fs *c, struct data_update *m,
 305                           struct write_point_specifier wp,
 306                           struct bch_io_opts io_opts,
 307                           struct data_update_opts data_opts,
 308                           enum btree_id btree_id,
 309                           struct bkey_s_c k)
 310 {
 311         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 312         const union bch_extent_entry *entry;
 313         struct extent_ptr_decoded p;
 314         unsigned i, reserve_sectors = k.k->size * data_opts.extra_replicas;
 315         int ret;
 316
 317         bch2_bkey_buf_init(&m->k);
 318         bch2_bkey_buf_reassemble(&m->k, c, k);
 319         m->btree_id     = btree_id;
 320         m->data_opts    = data_opts;
 321
 322         bch2_write_op_init(&m->op, c, io_opts);
 323         m->op.pos       = bkey_start_pos(k.k);
 324         m->op.version   = k.k->version;
 325         m->op.target    = data_opts.target,
 326         m->op.write_point = wp;
 327         m->op.flags     |= BCH_WRITE_PAGES_STABLE|
 328                 BCH_WRITE_PAGES_OWNED|
 329                 BCH_WRITE_DATA_ENCODED|
 330                 BCH_WRITE_FROM_INTERNAL|
 331                 m->data_opts.write_flags;
 332         m->op.compression_type =
 333                 bch2_compression_opt_to_type[io_opts.background_compression ?:
 334                                              io_opts.compression];
 335         if (m->data_opts.btree_insert_flags & BTREE_INSERT_USE_RESERVE)
 336                 m->op.alloc_reserve = RESERVE_movinggc;
 337         m->op.index_update_fn   = bch2_data_update_index_update;
 338
 339         i = 0;
 340         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 341                 if (p.ptr.cached)
 342                         m->data_opts.rewrite_ptrs &= ~(1U << i);
 343
 344                 if (!((1U << i) & m->data_opts.rewrite_ptrs))
 345                         bch2_dev_list_add_dev(&m->op.devs_have, p.ptr.dev);
 346
 347                 if (((1U << i) & m->data_opts.rewrite_ptrs) &&
 348                     crc_is_compressed(p.crc))
 349                         reserve_sectors += k.k->size;
 350
 351                 /*
 352                  * op->csum_type is normally initialized from the fs/file's
 353                  * current options - but if an extent is encrypted, we require
 354                  * that it stays encrypted:
 355                  */
 356                 if (bch2_csum_type_is_encryption(p.crc.csum_type)) {
 357                         m->op.nonce     = p.crc.nonce + p.crc.offset;
 358                         m->op.csum_type = p.crc.csum_type;
 359                 }
 360
 361                 if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
 362                         m->op.incompressible = true;
 363
 364                 i++;
 365         }
 366
 367         if (reserve_sectors) {
 368                 ret = bch2_disk_reservation_add(c, &m->op.res, reserve_sectors,
 369                                 m->data_opts.extra_replicas
 370                                 ? 0
 371                                 : BCH_DISK_RESERVATION_NOFAIL);
 372                 if (ret)
 373                         return ret;
 374         }
 375
 376         m->op.nr_replicas = m->op.nr_replicas_required =
 377                 hweight32(m->data_opts.rewrite_ptrs) + m->data_opts.extra_replicas;
 378         return 0;
 379 }