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