X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fjournal_io.c;h=38458ab0013d049d6b7cc3c6db990be1a09e2c95;hb=6b1f79d5df9f2735192ed1a40c711cf131d4f43e;hp=2b1974a9f360e943e59a408ccbbcb9cc2c262f4b;hpb=3867739e566d7f0df17ebed5e26c84ce88a90d40;p=bcachefs-tools-debian diff --git a/libbcachefs/journal_io.c b/libbcachefs/journal_io.c index 2b1974a..38458ab 100644 --- a/libbcachefs/journal_io.c +++ b/libbcachefs/journal_io.c @@ -14,8 +14,7 @@ #include "journal_reclaim.h" #include "journal_seq_blacklist.h" #include "replicas.h" - -#include +#include "trace.h" static struct nonce journal_nonce(const struct jset *jset) { @@ -119,7 +118,7 @@ static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca, journal_entry_radix_idx(c, le64_to_cpu(j->seq)), GFP_KERNEL); if (!_i) - return -ENOMEM; + return -BCH_ERR_ENOMEM_journal_entry_add; /* * Duplicate journal entries? If so we want the one that didn't have a @@ -149,12 +148,12 @@ static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca, replace: i = kvpmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL); if (!i) - return -ENOMEM; + return -BCH_ERR_ENOMEM_journal_entry_add; i->nr_ptrs = 0; i->csum_good = entry_ptr.csum_good; i->ignore = false; - memcpy(&i->j, j, bytes); + unsafe_memcpy(&i->j, j, bytes, "embedded variable length struct"); i->ptrs[i->nr_ptrs++] = entry_ptr; if (dup) { @@ -341,7 +340,7 @@ static int journal_entry_btree_keys_validate(struct bch_fs *c, int ret = journal_validate_key(c, jset, entry, entry->level, entry->btree_id, - k, version, big_endian, write); + k, version, big_endian, write|BKEY_INVALID_FROM_JOURNAL); if (ret == FSCK_DELETED_KEY) continue; @@ -357,7 +356,7 @@ static void journal_entry_btree_keys_to_text(struct printbuf *out, struct bch_fs struct bkey_i *k; bool first = true; - vstruct_for_each(entry, k) { + jset_entry_for_each_key(entry, k) { if (!first) { prt_newline(out); prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]); @@ -662,7 +661,8 @@ static int journal_entry_overwrite_validate(struct bch_fs *c, struct jset_entry *entry, unsigned version, int big_endian, int write) { - return journal_entry_btree_keys_validate(c, jset, entry, version, big_endian, write); + return journal_entry_btree_keys_validate(c, jset, entry, + version, big_endian, READ); } static void journal_entry_overwrite_to_text(struct printbuf *out, struct bch_fs *c, @@ -835,12 +835,12 @@ static int journal_read_buf_realloc(struct journal_read_buf *b, /* the bios are sized for this many pages, max: */ if (new_size > JOURNAL_ENTRY_SIZE_MAX) - return -ENOMEM; + return -BCH_ERR_ENOMEM_journal_read_buf_realloc; new_size = roundup_pow_of_two(new_size); n = kvpmalloc(new_size, GFP_KERNEL); if (!n) - return -ENOMEM; + return -BCH_ERR_ENOMEM_journal_read_buf_realloc; kvpfree(b->data, b->size); b->data = n; @@ -1080,7 +1080,10 @@ void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c, } } -int bch2_journal_read(struct bch_fs *c, u64 *blacklist_seq, u64 *start_seq) +int bch2_journal_read(struct bch_fs *c, + u64 *last_seq, + u64 *blacklist_seq, + u64 *start_seq) { struct journal_list jlist; struct journal_replay *i, **_i, *prev = NULL; @@ -1089,7 +1092,7 @@ int bch2_journal_read(struct bch_fs *c, u64 *blacklist_seq, u64 *start_seq) unsigned iter; struct printbuf buf = PRINTBUF; bool degraded = false, last_write_torn = false; - u64 seq, last_seq = 0; + u64 seq; int ret = 0; closure_init_stack(&jlist.cl); @@ -1118,15 +1121,13 @@ int bch2_journal_read(struct bch_fs *c, u64 *blacklist_seq, u64 *start_seq) if (jlist.ret) return jlist.ret; + *last_seq = 0; *start_seq = 0; *blacklist_seq = 0; /* * Find most recent flush entry, and ignore newer non flush entries - * those entries will be blacklisted: - * - * - * XXX check for torn write on last journal entry */ genradix_for_each_reverse(&c->journal_entries, radix_iter, _i) { int write = READ; @@ -1140,13 +1141,13 @@ int bch2_journal_read(struct bch_fs *c, u64 *blacklist_seq, u64 *start_seq) *blacklist_seq = *start_seq = le64_to_cpu(i->j.seq) + 1; if (JSET_NO_FLUSH(&i->j)) { - journal_replay_free(c, i); + i->ignore = true; continue; } if (!last_write_torn && !i->csum_good) { last_write_torn = true; - journal_replay_free(c, i); + i->ignore = true; continue; } @@ -1157,7 +1158,7 @@ int bch2_journal_read(struct bch_fs *c, u64 *blacklist_seq, u64 *start_seq) le64_to_cpu(i->j.seq))) i->j.last_seq = i->j.seq; - last_seq = le64_to_cpu(i->j.last_seq); + *last_seq = le64_to_cpu(i->j.last_seq); *blacklist_seq = le64_to_cpu(i->j.seq) + 1; break; } @@ -1167,13 +1168,13 @@ int bch2_journal_read(struct bch_fs *c, u64 *blacklist_seq, u64 *start_seq) return 0; } - if (!last_seq) { + if (!*last_seq) { fsck_err(c, "journal read done, but no entries found after dropping non-flushes"); return 0; } bch_info(c, "journal read done, replaying entries %llu-%llu", - last_seq, *blacklist_seq - 1); + *last_seq, *blacklist_seq - 1); if (*start_seq != *blacklist_seq) bch_info(c, "dropped unflushed entries %llu-%llu", @@ -1187,7 +1188,7 @@ int bch2_journal_read(struct bch_fs *c, u64 *blacklist_seq, u64 *start_seq) continue; seq = le64_to_cpu(i->j.seq); - if (seq < last_seq) { + if (seq < *last_seq) { journal_replay_free(c, i); continue; } @@ -1195,13 +1196,12 @@ int bch2_journal_read(struct bch_fs *c, u64 *blacklist_seq, u64 *start_seq) if (bch2_journal_seq_is_blacklisted(c, seq, true)) { fsck_err_on(!JSET_NO_FLUSH(&i->j), c, "found blacklisted journal entry %llu", seq); - - journal_replay_free(c, i); + i->ignore = true; } } /* Check for missing entries: */ - seq = last_seq; + seq = *last_seq; genradix_for_each(&c->journal_entries, radix_iter, _i) { i = *_i; @@ -1239,7 +1239,7 @@ int bch2_journal_read(struct bch_fs *c, u64 *blacklist_seq, u64 *start_seq) " prev at %s\n" " next at %s", missing_start, missing_end, - last_seq, *blacklist_seq - 1, + *last_seq, *blacklist_seq - 1, buf1.buf, buf2.buf); printbuf_exit(&buf1); @@ -1338,8 +1338,7 @@ static void __journal_write_alloc(struct journal *j, if (!ca->mi.durability || ca->mi.state != BCH_MEMBER_STATE_rw || !ja->nr || - bch2_bkey_has_device(bkey_i_to_s_c(&w->key), - ca->dev_idx) || + bch2_bkey_has_device_c(bkey_i_to_s_c(&w->key), ca->dev_idx) || sectors > ja->sectors_free) continue; @@ -1498,6 +1497,8 @@ static void journal_write_done(struct closure *cl) bch2_do_discards(c); closure_wake_up(&c->freelist_wait); + + bch2_reset_alloc_cursors(c); } } else if (!j->err_seq || seq < j->err_seq) j->err_seq = seq; @@ -1619,6 +1620,52 @@ static void do_journal_write(struct closure *cl) return; } +static void bch2_journal_entries_postprocess(struct bch_fs *c, struct jset *jset) +{ + struct jset_entry *i, *next, *prev = NULL; + + /* + * Simple compaction, dropping empty jset_entries (from journal + * reservations that weren't fully used) and merging jset_entries that + * can be. + * + * If we wanted to be really fancy here, we could sort all the keys in + * the jset and drop keys that were overwritten - probably not worth it: + */ + vstruct_for_each_safe(jset, i, next) { + unsigned u64s = le16_to_cpu(i->u64s); + + /* Empty entry: */ + if (!u64s) + continue; + + if (i->type == BCH_JSET_ENTRY_btree_root) + bch2_journal_entry_to_btree_root(c, i); + + /* Can we merge with previous entry? */ + if (prev && + i->btree_id == prev->btree_id && + i->level == prev->level && + i->type == prev->type && + i->type == BCH_JSET_ENTRY_btree_keys && + le16_to_cpu(prev->u64s) + u64s <= U16_MAX) { + memmove_u64s_down(vstruct_next(prev), + i->_data, + u64s); + le16_add_cpu(&prev->u64s, u64s); + continue; + } + + /* Couldn't merge, move i into new position (after prev): */ + prev = prev ? vstruct_next(prev) : jset->start; + if (i != prev) + memmove_u64s_down(prev, i, jset_u64s(u64s)); + } + + prev = prev ? vstruct_next(prev) : jset->start; + jset->u64s = cpu_to_le32((u64 *) prev - jset->_data); +} + void bch2_journal_write(struct closure *cl) { struct journal *j = container_of(cl, struct journal, io); @@ -1690,7 +1737,7 @@ void bch2_journal_write(struct closure *cl) * entry: */ - bch2_journal_entries_to_btree_roots(c, jset); + bch2_journal_entries_postprocess(c, jset); start = end = vstruct_last(jset);