X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fjournal_io.c;h=b455ef041dfefd6b6ab1c7bae0608406d2decba9;hb=5ef62f56ab50c5799f713e3a42f5c7ad7e8283d3;hp=cbde21a4c5479dd9f3583fe125ed415118f886fe;hpb=790ca9522a96efe321aae36fb0d7f4e437110b0f;p=bcachefs-tools-debian diff --git a/libbcachefs/journal_io.c b/libbcachefs/journal_io.c index cbde21a..b455ef0 100644 --- a/libbcachefs/journal_io.c +++ b/libbcachefs/journal_io.c @@ -14,15 +14,41 @@ #include "journal_reclaim.h" #include "journal_seq_blacklist.h" #include "replicas.h" +#include "trace.h" -#include +static struct nonce journal_nonce(const struct jset *jset) +{ + return (struct nonce) {{ + [0] = 0, + [1] = ((__le32 *) &jset->seq)[0], + [2] = ((__le32 *) &jset->seq)[1], + [3] = BCH_NONCE_JOURNAL, + }}; +} + +static bool jset_csum_good(struct bch_fs *c, struct jset *j) +{ + return bch2_checksum_type_valid(c, JSET_CSUM_TYPE(j)) && + !bch2_crc_cmp(j->csum, + csum_vstruct(c, JSET_CSUM_TYPE(j), journal_nonce(j), j)); +} + +static inline u32 journal_entry_radix_idx(struct bch_fs *c, u64 seq) +{ + return (seq - c->journal_entries_base_seq) & (~0U >> 1); +} -static void __journal_replay_free(struct journal_replay *i) +static void __journal_replay_free(struct bch_fs *c, + struct journal_replay *i) { - list_del(&i->list); + struct journal_replay **p = + genradix_ptr(&c->journal_entries, + journal_entry_radix_idx(c, le64_to_cpu(i->j.seq))); + + BUG_ON(*p != i); + *p = NULL; kvpfree(i, offsetof(struct journal_replay, j) + vstruct_bytes(&i->j)); - } static void journal_replay_free(struct bch_fs *c, struct journal_replay *i) @@ -30,13 +56,13 @@ static void journal_replay_free(struct bch_fs *c, struct journal_replay *i) i->ignore = true; if (!c->opts.read_entire_journal) - __journal_replay_free(i); + __journal_replay_free(c, i); } struct journal_list { struct closure cl; + u64 last_seq; struct mutex lock; - struct list_head *head; int ret; }; @@ -49,93 +75,104 @@ struct journal_list { */ static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca, struct journal_ptr entry_ptr, - struct journal_list *jlist, struct jset *j, - bool bad) + struct journal_list *jlist, struct jset *j) { - struct journal_replay *i, *pos, *dup = NULL; + struct genradix_iter iter; + struct journal_replay **_i, *i, *dup; struct journal_ptr *ptr; - struct list_head *where; size_t bytes = vstruct_bytes(j); - u64 last_seq = 0; + u64 last_seq = !JSET_NO_FLUSH(j) ? le64_to_cpu(j->last_seq) : 0; int ret = JOURNAL_ENTRY_ADD_OK; - list_for_each_entry_reverse(i, jlist->head, list) { - if (!JSET_NO_FLUSH(&i->j)) { - last_seq = le64_to_cpu(i->j.last_seq); - break; - } - } - /* Is this entry older than the range we need? */ if (!c->opts.read_entire_journal && - le64_to_cpu(j->seq) < last_seq) { - ret = JOURNAL_ENTRY_ADD_OUT_OF_RANGE; - goto out; - } + le64_to_cpu(j->seq) < jlist->last_seq) + return JOURNAL_ENTRY_ADD_OUT_OF_RANGE; + + /* + * genradixes are indexed by a ulong, not a u64, so we can't index them + * by sequence number directly: Assume instead that they will all fall + * within the range of +-2billion of the filrst one we find. + */ + if (!c->journal_entries_base_seq) + c->journal_entries_base_seq = max_t(s64, 1, le64_to_cpu(j->seq) - S32_MAX); /* Drop entries we don't need anymore */ - if (!JSET_NO_FLUSH(j)) { - list_for_each_entry_safe(i, pos, jlist->head, list) { - if (le64_to_cpu(i->j.seq) >= le64_to_cpu(j->last_seq)) + if (last_seq > jlist->last_seq && !c->opts.read_entire_journal) { + genradix_for_each_from(&c->journal_entries, iter, _i, + journal_entry_radix_idx(c, jlist->last_seq)) { + i = *_i; + + if (!i || i->ignore) + continue; + + if (le64_to_cpu(i->j.seq) >= last_seq) break; journal_replay_free(c, i); } } - list_for_each_entry_reverse(i, jlist->head, list) { - if (le64_to_cpu(j->seq) > le64_to_cpu(i->j.seq)) { - where = &i->list; - goto add; - } - } + jlist->last_seq = max(jlist->last_seq, last_seq); - where = jlist->head; -add: - dup = where->next != jlist->head - ? container_of(where->next, struct journal_replay, list) - : NULL; - - if (dup && le64_to_cpu(j->seq) != le64_to_cpu(dup->j.seq)) - dup = NULL; + _i = genradix_ptr_alloc(&c->journal_entries, + journal_entry_radix_idx(c, le64_to_cpu(j->seq)), + GFP_KERNEL); + if (!_i) + return -BCH_ERR_ENOMEM_journal_entry_add; /* * Duplicate journal entries? If so we want the one that didn't have a * checksum error: */ + dup = *_i; if (dup) { - if (dup->bad) { - /* we'll replace @dup: */ - } else if (bad) { + if (bytes == vstruct_bytes(&dup->j) && + !memcmp(j, &dup->j, bytes)) { i = dup; goto found; - } else { - fsck_err_on(bytes != vstruct_bytes(&dup->j) || - memcmp(j, &dup->j, bytes), c, - "found duplicate but non identical journal entries (seq %llu)", - le64_to_cpu(j->seq)); + } + + if (!entry_ptr.csum_good) { i = dup; goto found; } - } - i = kvpmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL); - if (!i) { - ret = -ENOMEM; - goto out; + if (!dup->csum_good) + goto replace; + + fsck_err(c, "found duplicate but non identical journal entries (seq %llu)", + le64_to_cpu(j->seq)); + i = dup; + goto found; } +replace: + i = kvpmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL); + if (!i) + return -BCH_ERR_ENOMEM_journal_entry_add; - i->nr_ptrs = 0; - i->bad = bad; + 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) { - i->nr_ptrs = dup->nr_ptrs; - memcpy(i->ptrs, dup->ptrs, sizeof(dup->ptrs)); - __journal_replay_free(dup); + if (dup->nr_ptrs >= ARRAY_SIZE(dup->ptrs)) { + bch_err(c, "found too many copies of journal entry %llu", + le64_to_cpu(i->j.seq)); + dup->nr_ptrs = ARRAY_SIZE(dup->ptrs) - 1; + } + + /* The first ptr should represent the jset we kept: */ + memcpy(i->ptrs + i->nr_ptrs, + dup->ptrs, + sizeof(dup->ptrs[0]) * dup->nr_ptrs); + i->nr_ptrs += dup->nr_ptrs; + __journal_replay_free(c, dup); } - list_add(&i->list, where); + *_i = i; + return 0; found: for (ptr = i->ptrs; ptr < i->ptrs + i->nr_ptrs; ptr++) { if (ptr->dev == ca->dev_idx) { @@ -157,16 +194,6 @@ fsck_err: return ret; } -static struct nonce journal_nonce(const struct jset *jset) -{ - return (struct nonce) {{ - [0] = 0, - [1] = ((__le32 *) &jset->seq)[0], - [2] = ((__le32 *) &jset->seq)[1], - [3] = BCH_NONCE_JOURNAL, - }}; -} - /* this fills in a range with empty jset_entries: */ static void journal_entry_null_range(void *start, void *end) { @@ -180,66 +207,84 @@ static void journal_entry_null_range(void *start, void *end) #define JOURNAL_ENTRY_NONE 6 #define JOURNAL_ENTRY_BAD 7 -#define journal_entry_err(c, msg, ...) \ +static void journal_entry_err_msg(struct printbuf *out, + struct jset *jset, + struct jset_entry *entry) +{ + prt_str(out, "invalid journal entry "); + if (entry) + prt_printf(out, "%s ", bch2_jset_entry_types[entry->type]); + + if (!jset) + prt_printf(out, "in superblock"); + else if (!entry) + prt_printf(out, "at seq %llu", le64_to_cpu(jset->seq)); + else + prt_printf(out, "at offset %zi/%u seq %llu", + (u64 *) entry - jset->_data, + le32_to_cpu(jset->u64s), + le64_to_cpu(jset->seq)); + prt_str(out, ": "); +} + +#define journal_entry_err(c, jset, entry, msg, ...) \ ({ \ + struct printbuf buf = PRINTBUF; \ + \ + journal_entry_err_msg(&buf, jset, entry); \ + prt_printf(&buf, msg, ##__VA_ARGS__); \ + \ switch (write) { \ case READ: \ - mustfix_fsck_err(c, msg, ##__VA_ARGS__); \ + mustfix_fsck_err(c, "%s", buf.buf); \ break; \ case WRITE: \ - bch_err(c, "corrupt metadata before write:\n" \ - msg, ##__VA_ARGS__); \ + bch_err(c, "corrupt metadata before write: %s\n", buf.buf);\ if (bch2_fs_inconsistent(c)) { \ - ret = BCH_FSCK_ERRORS_NOT_FIXED; \ + ret = -BCH_ERR_fsck_errors_not_fixed; \ goto fsck_err; \ } \ break; \ } \ + \ + printbuf_exit(&buf); \ true; \ }) -#define journal_entry_err_on(cond, c, msg, ...) \ - ((cond) ? journal_entry_err(c, msg, ##__VA_ARGS__) : false) +#define journal_entry_err_on(cond, c, jset, entry, msg, ...) \ + ((cond) ? journal_entry_err(c, jset, entry, msg, ##__VA_ARGS__) : false) #define FSCK_DELETED_KEY 5 -static int journal_validate_key(struct bch_fs *c, const char *where, +static int journal_validate_key(struct bch_fs *c, + struct jset *jset, struct jset_entry *entry, unsigned level, enum btree_id btree_id, - struct bkey_i *k, const char *type, + struct bkey_i *k, unsigned version, int big_endian, int write) { void *next = vstruct_next(entry); struct printbuf buf = PRINTBUF; int ret = 0; - if (journal_entry_err_on(!k->k.u64s, c, - "invalid %s in %s entry offset %zi/%u: k->u64s 0", - type, where, - (u64 *) k - entry->_data, - le16_to_cpu(entry->u64s))) { + if (journal_entry_err_on(!k->k.u64s, c, jset, entry, "k->u64s 0")) { entry->u64s = cpu_to_le16((u64 *) k - entry->_data); journal_entry_null_range(vstruct_next(entry), next); return FSCK_DELETED_KEY; } if (journal_entry_err_on((void *) bkey_next(k) > - (void *) vstruct_next(entry), c, - "invalid %s in %s entry offset %zi/%u: extends past end of journal entry", - type, where, - (u64 *) k - entry->_data, - le16_to_cpu(entry->u64s))) { + (void *) vstruct_next(entry), + c, jset, entry, + "extends past end of journal entry")) { entry->u64s = cpu_to_le16((u64 *) k - entry->_data); journal_entry_null_range(vstruct_next(entry), next); return FSCK_DELETED_KEY; } - if (journal_entry_err_on(k->k.format != KEY_FORMAT_CURRENT, c, - "invalid %s in %s entry offset %zi/%u: bad format %u", - type, where, - (u64 *) k - entry->_data, - le16_to_cpu(entry->u64s), - k->k.format)) { + if (journal_entry_err_on(k->k.format != KEY_FORMAT_CURRENT, + c, jset, entry, + "bad format %u", k->k.format)) { le16_add_cpu(&entry->u64s, -((u16) k->k.u64s)); memmove(k, bkey_next(k), next - (void *) bkey_next(k)); journal_entry_null_range(vstruct_next(entry), next); @@ -253,15 +298,16 @@ static int journal_validate_key(struct bch_fs *c, const char *where, if (bch2_bkey_invalid(c, bkey_i_to_s_c(k), __btree_node_type(level, btree_id), write, &buf)) { printbuf_reset(&buf); - pr_buf(&buf, "invalid %s in %s entry offset %zi/%u:", - type, where, - (u64 *) k - entry->_data, - le16_to_cpu(entry->u64s)); - pr_newline(&buf); - pr_indent_push(&buf, 2); + prt_printf(&buf, "invalid journal entry %s at offset %zi/%u seq %llu:", + bch2_jset_entry_types[entry->type], + (u64 *) entry - jset->_data, + le32_to_cpu(jset->u64s), + le64_to_cpu(jset->seq)); + prt_newline(&buf); + printbuf_indent_add(&buf, 2); bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k)); - pr_newline(&buf); + prt_newline(&buf); bch2_bkey_invalid(c, bkey_i_to_s_c(k), __btree_node_type(level, btree_id), write, &buf); @@ -284,17 +330,17 @@ fsck_err: } static int journal_entry_btree_keys_validate(struct bch_fs *c, - const char *where, + struct jset *jset, struct jset_entry *entry, unsigned version, int big_endian, int write) { struct bkey_i *k = entry->start; while (k != vstruct_last(entry)) { - int ret = journal_validate_key(c, where, entry, + int ret = journal_validate_key(c, jset, entry, entry->level, entry->btree_id, - k, "key", version, big_endian, write); + k, version, big_endian, write|BKEY_INVALID_FROM_JOURNAL); if (ret == FSCK_DELETED_KEY) continue; @@ -310,19 +356,19 @@ 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) { - pr_newline(out); - pr_buf(out, "%s: ", bch2_jset_entry_types[entry->type]); + prt_newline(out); + prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]); } - pr_buf(out, "btree=%s l=%u ", bch2_btree_ids[entry->btree_id], entry->level); + prt_printf(out, "btree=%s l=%u ", bch2_btree_ids[entry->btree_id], entry->level); bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(k)); first = false; } } static int journal_entry_btree_root_validate(struct bch_fs *c, - const char *where, + struct jset *jset, struct jset_entry *entry, unsigned version, int big_endian, int write) { @@ -330,7 +376,8 @@ static int journal_entry_btree_root_validate(struct bch_fs *c, int ret = 0; if (journal_entry_err_on(!entry->u64s || - le16_to_cpu(entry->u64s) != k->k.u64s, c, + le16_to_cpu(entry->u64s) != k->k.u64s, + c, jset, entry, "invalid btree root journal entry: wrong number of keys")) { void *next = vstruct_next(entry); /* @@ -343,8 +390,8 @@ static int journal_entry_btree_root_validate(struct bch_fs *c, return 0; } - return journal_validate_key(c, where, entry, 1, entry->btree_id, k, - "btree root", version, big_endian, write); + return journal_validate_key(c, jset, entry, 1, entry->btree_id, k, + version, big_endian, write); fsck_err: return ret; } @@ -356,7 +403,7 @@ static void journal_entry_btree_root_to_text(struct printbuf *out, struct bch_fs } static int journal_entry_prio_ptrs_validate(struct bch_fs *c, - const char *where, + struct jset *jset, struct jset_entry *entry, unsigned version, int big_endian, int write) { @@ -370,13 +417,14 @@ static void journal_entry_prio_ptrs_to_text(struct printbuf *out, struct bch_fs } static int journal_entry_blacklist_validate(struct bch_fs *c, - const char *where, + struct jset *jset, struct jset_entry *entry, unsigned version, int big_endian, int write) { int ret = 0; - if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 1, c, + if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 1, + c, jset, entry, "invalid journal seq blacklist entry: bad size")) { journal_entry_null_range(entry, vstruct_next(entry)); } @@ -390,18 +438,19 @@ static void journal_entry_blacklist_to_text(struct printbuf *out, struct bch_fs struct jset_entry_blacklist *bl = container_of(entry, struct jset_entry_blacklist, entry); - pr_buf(out, "seq=%llu", le64_to_cpu(bl->seq)); + prt_printf(out, "seq=%llu", le64_to_cpu(bl->seq)); } static int journal_entry_blacklist_v2_validate(struct bch_fs *c, - const char *where, + struct jset *jset, struct jset_entry *entry, unsigned version, int big_endian, int write) { struct jset_entry_blacklist_v2 *bl_entry; int ret = 0; - if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 2, c, + if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 2, + c, jset, entry, "invalid journal seq blacklist entry: bad size")) { journal_entry_null_range(entry, vstruct_next(entry)); goto out; @@ -410,7 +459,8 @@ static int journal_entry_blacklist_v2_validate(struct bch_fs *c, bl_entry = container_of(entry, struct jset_entry_blacklist_v2, entry); if (journal_entry_err_on(le64_to_cpu(bl_entry->start) > - le64_to_cpu(bl_entry->end), c, + le64_to_cpu(bl_entry->end), + c, jset, entry, "invalid journal seq blacklist entry: start > end")) { journal_entry_null_range(entry, vstruct_next(entry)); } @@ -425,13 +475,13 @@ static void journal_entry_blacklist_v2_to_text(struct printbuf *out, struct bch_ struct jset_entry_blacklist_v2 *bl = container_of(entry, struct jset_entry_blacklist_v2, entry); - pr_buf(out, "start=%llu end=%llu", + prt_printf(out, "start=%llu end=%llu", le64_to_cpu(bl->start), le64_to_cpu(bl->end)); } static int journal_entry_usage_validate(struct bch_fs *c, - const char *where, + struct jset *jset, struct jset_entry *entry, unsigned version, int big_endian, int write) { @@ -441,7 +491,7 @@ static int journal_entry_usage_validate(struct bch_fs *c, int ret = 0; if (journal_entry_err_on(bytes < sizeof(*u), - c, + c, jset, entry, "invalid journal entry usage: bad size")) { journal_entry_null_range(entry, vstruct_next(entry)); return ret; @@ -457,13 +507,13 @@ static void journal_entry_usage_to_text(struct printbuf *out, struct bch_fs *c, struct jset_entry_usage *u = container_of(entry, struct jset_entry_usage, entry); - pr_buf(out, "type=%s v=%llu", + prt_printf(out, "type=%s v=%llu", bch2_fs_usage_types[u->entry.btree_id], le64_to_cpu(u->v)); } static int journal_entry_data_usage_validate(struct bch_fs *c, - const char *where, + struct jset *jset, struct jset_entry *entry, unsigned version, int big_endian, int write) { @@ -474,7 +524,7 @@ static int journal_entry_data_usage_validate(struct bch_fs *c, if (journal_entry_err_on(bytes < sizeof(*u) || bytes < sizeof(*u) + u->r.nr_devs, - c, + c, jset, entry, "invalid journal entry usage: bad size")) { journal_entry_null_range(entry, vstruct_next(entry)); return ret; @@ -491,11 +541,11 @@ static void journal_entry_data_usage_to_text(struct printbuf *out, struct bch_fs container_of(entry, struct jset_entry_data_usage, entry); bch2_replicas_entry_to_text(out, &u->r); - pr_buf(out, "=%llu", le64_to_cpu(u->v)); + prt_printf(out, "=%llu", le64_to_cpu(u->v)); } static int journal_entry_clock_validate(struct bch_fs *c, - const char *where, + struct jset *jset, struct jset_entry *entry, unsigned version, int big_endian, int write) { @@ -505,13 +555,13 @@ static int journal_entry_clock_validate(struct bch_fs *c, int ret = 0; if (journal_entry_err_on(bytes != sizeof(*clock), - c, "invalid journal entry clock: bad size")) { + c, jset, entry, "bad size")) { journal_entry_null_range(entry, vstruct_next(entry)); return ret; } if (journal_entry_err_on(clock->rw > 1, - c, "invalid journal entry clock: bad rw")) { + c, jset, entry, "bad rw")) { journal_entry_null_range(entry, vstruct_next(entry)); return ret; } @@ -526,11 +576,11 @@ static void journal_entry_clock_to_text(struct printbuf *out, struct bch_fs *c, struct jset_entry_clock *clock = container_of(entry, struct jset_entry_clock, entry); - pr_buf(out, "%s=%llu", clock->rw ? "write" : "read", le64_to_cpu(clock->time)); + prt_printf(out, "%s=%llu", clock->rw ? "write" : "read", le64_to_cpu(clock->time)); } static int journal_entry_dev_usage_validate(struct bch_fs *c, - const char *where, + struct jset *jset, struct jset_entry *entry, unsigned version, int big_endian, int write) { @@ -542,7 +592,7 @@ static int journal_entry_dev_usage_validate(struct bch_fs *c, int ret = 0; if (journal_entry_err_on(bytes < expected, - c, "invalid journal entry dev usage: bad size (%u < %u)", + c, jset, entry, "bad size (%u < %u)", bytes, expected)) { journal_entry_null_range(entry, vstruct_next(entry)); return ret; @@ -551,13 +601,13 @@ static int journal_entry_dev_usage_validate(struct bch_fs *c, dev = le32_to_cpu(u->dev); if (journal_entry_err_on(!bch2_dev_exists2(c, dev), - c, "invalid journal entry dev usage: bad dev")) { + c, jset, entry, "bad dev")) { journal_entry_null_range(entry, vstruct_next(entry)); return ret; } if (journal_entry_err_on(u->pad, - c, "invalid journal entry dev usage: bad pad")) { + c, jset, entry, "bad pad")) { journal_entry_null_range(entry, vstruct_next(entry)); return ret; } @@ -573,26 +623,24 @@ static void journal_entry_dev_usage_to_text(struct printbuf *out, struct bch_fs container_of(entry, struct jset_entry_dev_usage, entry); unsigned i, nr_types = jset_entry_dev_usage_nr_types(u); - pr_buf(out, "dev=%u", le32_to_cpu(u->dev)); + prt_printf(out, "dev=%u", le32_to_cpu(u->dev)); for (i = 0; i < nr_types; i++) { if (i < BCH_DATA_NR) - pr_buf(out, " %s", bch2_data_types[i]); + prt_printf(out, " %s", bch2_data_types[i]); else - pr_buf(out, " (unknown data type %u)", i); - pr_buf(out, ": buckets=%llu sectors=%llu fragmented=%llu", + prt_printf(out, " (unknown data type %u)", i); + prt_printf(out, ": buckets=%llu sectors=%llu fragmented=%llu", le64_to_cpu(u->d[i].buckets), le64_to_cpu(u->d[i].sectors), le64_to_cpu(u->d[i].fragmented)); } - pr_buf(out, " buckets_ec: %llu buckets_unavailable: %llu", - le64_to_cpu(u->buckets_ec), - le64_to_cpu(u->buckets_unavailable)); + prt_printf(out, " buckets_ec: %llu", le64_to_cpu(u->buckets_ec)); } static int journal_entry_log_validate(struct bch_fs *c, - const char *where, + struct jset *jset, struct jset_entry *entry, unsigned version, int big_endian, int write) { @@ -605,11 +653,26 @@ static void journal_entry_log_to_text(struct printbuf *out, struct bch_fs *c, struct jset_entry_log *l = container_of(entry, struct jset_entry_log, entry); unsigned bytes = vstruct_bytes(entry) - offsetof(struct jset_entry_log, d); - pr_buf(out, "%.*s", bytes, l->d); + prt_printf(out, "%.*s", bytes, l->d); +} + +static int journal_entry_overwrite_validate(struct bch_fs *c, + struct jset *jset, + struct jset_entry *entry, + unsigned version, int big_endian, int 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, + struct jset_entry *entry) +{ + journal_entry_btree_keys_to_text(out, c, entry); } struct jset_entry_ops { - int (*validate)(struct bch_fs *, const char *, + int (*validate)(struct bch_fs *, struct jset *, struct jset_entry *, unsigned, int, int); void (*to_text)(struct printbuf *, struct bch_fs *, struct jset_entry *); }; @@ -624,12 +687,13 @@ static const struct jset_entry_ops bch2_jset_entry_ops[] = { #undef x }; -int bch2_journal_entry_validate(struct bch_fs *c, const char *where, +int bch2_journal_entry_validate(struct bch_fs *c, + struct jset *jset, struct jset_entry *entry, unsigned version, int big_endian, int write) { return entry->type < BCH_JSET_ENTRY_NR - ? bch2_jset_entry_ops[entry->type].validate(c, where, entry, + ? bch2_jset_entry_ops[entry->type].validate(c, jset, entry, version, big_endian, write) : 0; } @@ -638,34 +702,28 @@ void bch2_journal_entry_to_text(struct printbuf *out, struct bch_fs *c, struct jset_entry *entry) { if (entry->type < BCH_JSET_ENTRY_NR) { - pr_buf(out, "%s: ", bch2_jset_entry_types[entry->type]); + prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]); bch2_jset_entry_ops[entry->type].to_text(out, c, entry); } else { - pr_buf(out, "(unknown type %u)", entry->type); + prt_printf(out, "(unknown type %u)", entry->type); } } static int jset_validate_entries(struct bch_fs *c, struct jset *jset, int write) { - char buf[100]; struct jset_entry *entry; int ret = 0; vstruct_for_each(jset, entry) { - scnprintf(buf, sizeof(buf), "jset %llu entry offset %zi/%u", - le64_to_cpu(jset->seq), - (u64 *) entry - jset->_data, - le32_to_cpu(jset->u64s)); - if (journal_entry_err_on(vstruct_next(entry) > - vstruct_last(jset), c, + vstruct_last(jset), c, jset, entry, "journal entry extends past end of jset")) { jset->u64s = cpu_to_le32((u64 *) entry - jset->_data); break; } - ret = bch2_journal_entry_validate(c, buf, entry, + ret = bch2_journal_entry_validate(c, jset, entry, le32_to_cpu(jset->version), JSET_BIG_ENDIAN(jset), write); if (ret) @@ -678,12 +736,8 @@ fsck_err: static int jset_validate(struct bch_fs *c, struct bch_dev *ca, struct jset *jset, u64 sector, - unsigned bucket_sectors_left, - unsigned sectors_read, int write) { - size_t bytes = vstruct_bytes(jset); - struct bch_csum csum; unsigned version; int ret = 0; @@ -693,72 +747,80 @@ static int jset_validate(struct bch_fs *c, version = le32_to_cpu(jset->version); if (journal_entry_err_on((version != BCH_JSET_VERSION_OLD && version < bcachefs_metadata_version_min) || - version >= bcachefs_metadata_version_max, c, + version >= bcachefs_metadata_version_max, + c, jset, NULL, "%s sector %llu seq %llu: unknown journal entry version %u", ca ? ca->name : c->name, sector, le64_to_cpu(jset->seq), version)) { /* don't try to continue: */ - return EINVAL; + return -EINVAL; } - if (bytes > (sectors_read << 9) && - sectors_read < bucket_sectors_left) - return JOURNAL_ENTRY_REREAD; - - if (journal_entry_err_on(bytes > bucket_sectors_left << 9, c, - "%s sector %llu seq %llu: journal entry too big (%zu bytes)", - ca ? ca->name : c->name, - sector, le64_to_cpu(jset->seq), bytes)) { - ret = JOURNAL_ENTRY_BAD; - le32_add_cpu(&jset->u64s, - -((bytes - (bucket_sectors_left << 9)) / 8)); - } - - if (journal_entry_err_on(!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(jset)), c, + if (journal_entry_err_on(!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(jset)), + c, jset, NULL, "%s sector %llu seq %llu: journal entry with unknown csum type %llu", ca ? ca->name : c->name, sector, le64_to_cpu(jset->seq), - JSET_CSUM_TYPE(jset))) { + JSET_CSUM_TYPE(jset))) ret = JOURNAL_ENTRY_BAD; - goto csum_done; - } - - if (write) - goto csum_done; - csum = csum_vstruct(c, JSET_CSUM_TYPE(jset), journal_nonce(jset), jset); - if (journal_entry_err_on(bch2_crc_cmp(csum, jset->csum), c, - "%s sector %llu seq %llu: journal checksum bad", - ca ? ca->name : c->name, - sector, le64_to_cpu(jset->seq))) - ret = JOURNAL_ENTRY_BAD; - - ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset), - jset->encrypted_start, - vstruct_end(jset) - (void *) jset->encrypted_start); - bch2_fs_fatal_err_on(ret, c, - "error decrypting journal entry: %i", ret); -csum_done: /* last_seq is ignored when JSET_NO_FLUSH is true */ if (journal_entry_err_on(!JSET_NO_FLUSH(jset) && - le64_to_cpu(jset->last_seq) > le64_to_cpu(jset->seq), c, + le64_to_cpu(jset->last_seq) > le64_to_cpu(jset->seq), + c, jset, NULL, "invalid journal entry: last_seq > seq (%llu > %llu)", le64_to_cpu(jset->last_seq), le64_to_cpu(jset->seq))) { jset->last_seq = jset->seq; return JOURNAL_ENTRY_BAD; } + + ret = jset_validate_entries(c, jset, write); fsck_err: return ret; } -static int jset_validate_for_write(struct bch_fs *c, struct jset *jset) +static int jset_validate_early(struct bch_fs *c, + struct bch_dev *ca, + struct jset *jset, u64 sector, + unsigned bucket_sectors_left, + unsigned sectors_read) { - unsigned sectors = vstruct_sectors(jset, c->block_bits); + size_t bytes = vstruct_bytes(jset); + unsigned version; + int write = READ; + int ret = 0; + + if (le64_to_cpu(jset->magic) != jset_magic(c)) + return JOURNAL_ENTRY_NONE; + + version = le32_to_cpu(jset->version); + if (journal_entry_err_on((version != BCH_JSET_VERSION_OLD && + version < bcachefs_metadata_version_min) || + version >= bcachefs_metadata_version_max, + c, jset, NULL, + "%s sector %llu seq %llu: unknown journal entry version %u", + ca ? ca->name : c->name, + sector, le64_to_cpu(jset->seq), + version)) { + /* don't try to continue: */ + return -EINVAL; + } + + if (bytes > (sectors_read << 9) && + sectors_read < bucket_sectors_left) + return JOURNAL_ENTRY_REREAD; - return jset_validate(c, NULL, jset, 0, sectors, sectors, WRITE) ?: - jset_validate_entries(c, jset, WRITE); + if (journal_entry_err_on(bytes > bucket_sectors_left << 9, + c, jset, NULL, + "%s sector %llu seq %llu: journal entry too big (%zu bytes)", + ca ? ca->name : c->name, + sector, le64_to_cpu(jset->seq), bytes)) + le32_add_cpu(&jset->u64s, + -((bytes - (bucket_sectors_left << 9)) / 8)); +fsck_err: + return ret; } struct journal_read_buf { @@ -773,12 +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; @@ -797,7 +859,7 @@ static int journal_read_bucket(struct bch_dev *ca, unsigned sectors, sectors_read = 0; u64 offset = bucket_to_sector(ca, ja->buckets[bucket]), end = offset + ca->mi.bucket_size; - bool saw_bad = false; + bool saw_bad = false, csum_good; int ret = 0; pr_debug("reading %u", bucket); @@ -805,20 +867,20 @@ static int journal_read_bucket(struct bch_dev *ca, while (offset < end) { if (!sectors_read) { struct bio *bio; + unsigned nr_bvecs; reread: sectors_read = min_t(unsigned, end - offset, buf->size >> 9); + nr_bvecs = buf_pages(buf->data, sectors_read << 9); + + bio = bio_kmalloc(nr_bvecs, GFP_KERNEL); + bio_init(bio, ca->disk_sb.bdev, bio->bi_inline_vecs, nr_bvecs, REQ_OP_READ); - bio = bio_kmalloc(GFP_KERNEL, - buf_pages(buf->data, - sectors_read << 9)); - bio_set_dev(bio, ca->disk_sb.bdev); - bio->bi_iter.bi_sector = offset; - bio_set_op_attrs(bio, REQ_OP_READ, 0); + bio->bi_iter.bi_sector = offset; bch2_bio_map(bio, buf->data, sectors_read << 9); ret = submit_bio_wait(bio); - bio_put(bio); + kfree(bio); if (bch2_dev_io_err_on(ret, ca, "journal read error: sector %llu", @@ -836,11 +898,10 @@ reread: j = buf->data; } - ret = jset_validate(c, ca, j, offset, - end - offset, sectors_read, - READ); + ret = jset_validate_early(c, ca, j, offset, + end - offset, sectors_read); switch (ret) { - case BCH_FSCK_OK: + case 0: sectors = vstruct_sectors(j, c->block_bits); break; case JOURNAL_ENTRY_REREAD: @@ -854,17 +915,13 @@ reread: case JOURNAL_ENTRY_NONE: if (!saw_bad) return 0; - sectors = block_sectors(c); - goto next_block; - case JOURNAL_ENTRY_BAD: - saw_bad = true; /* * On checksum error we don't really trust the size * field of the journal entry we read, so try reading * again at next block boundary: */ sectors = block_sectors(c); - break; + goto next_block; default: return ret; } @@ -880,14 +937,25 @@ reread: ja->bucket_seq[bucket] = le64_to_cpu(j->seq); + csum_good = jset_csum_good(c, j); + if (!csum_good) + saw_bad = true; + + ret = bch2_encrypt(c, JSET_CSUM_TYPE(j), journal_nonce(j), + j->encrypted_start, + vstruct_end(j) - (void *) j->encrypted_start); + bch2_fs_fatal_err_on(ret, c, + "error decrypting journal entry: %i", ret); + mutex_lock(&jlist->lock); ret = journal_entry_add(c, ca, (struct journal_ptr) { + .csum_good = csum_good, .dev = ca->dev_idx, .bucket = bucket, .bucket_offset = offset - bucket_to_sector(ca, ja->buckets[bucket]), .sector = offset, - }, jlist, j, ret != 0); + }, jlist, j); mutex_unlock(&jlist->lock); switch (ret) { @@ -916,9 +984,9 @@ static void bch2_journal_read_device(struct closure *cl) struct bch_fs *c = ca->fs; struct journal_list *jlist = container_of(cl->parent, struct journal_list, cl); - struct journal_replay *r; + struct journal_replay *r, **_r; + struct genradix_iter iter; struct journal_read_buf buf = { NULL, 0 }; - u64 min_seq = U64_MAX; unsigned i; int ret = 0; @@ -937,40 +1005,27 @@ static void bch2_journal_read_device(struct closure *cl) goto err; } - /* Find the journal bucket with the highest sequence number: */ - for (i = 0; i < ja->nr; i++) { - if (ja->bucket_seq[i] > ja->bucket_seq[ja->cur_idx]) - ja->cur_idx = i; - - min_seq = min(ja->bucket_seq[i], min_seq); - } - - /* - * If there's duplicate journal entries in multiple buckets (which - * definitely isn't supposed to happen, but...) - make sure to start - * cur_idx at the last of those buckets, so we don't deadlock trying to - * allocate - */ - while (ja->bucket_seq[ja->cur_idx] > min_seq && - ja->bucket_seq[ja->cur_idx] == - ja->bucket_seq[(ja->cur_idx + 1) % ja->nr]) - ja->cur_idx = (ja->cur_idx + 1) % ja->nr; - ja->sectors_free = ca->mi.bucket_size; mutex_lock(&jlist->lock); - list_for_each_entry(r, jlist->head, list) { + genradix_for_each_reverse(&c->journal_entries, iter, _r) { + r = *_r; + + if (!r) + continue; + for (i = 0; i < r->nr_ptrs; i++) { - if (r->ptrs[i].dev == ca->dev_idx && - sector_to_bucket(ca, r->ptrs[i].sector) == ja->buckets[ja->cur_idx]) { - unsigned wrote = (r->ptrs[i].sector % ca->mi.bucket_size) + + if (r->ptrs[i].dev == ca->dev_idx) { + unsigned wrote = bucket_remainder(ca, r->ptrs[i].sector) + vstruct_sectors(&r->j, c->block_bits); - ja->sectors_free = min(ja->sectors_free, - ca->mi.bucket_size - wrote); + ja->cur_idx = r->ptrs[i].bucket; + ja->sectors_free = ca->mi.bucket_size - wrote; + goto found; } } } +found: mutex_unlock(&jlist->lock); if (ja->bucket_seq[ja->cur_idx] && @@ -978,7 +1033,7 @@ static void bch2_journal_read_device(struct closure *cl) bch_err(c, "ja->sectors_free == ca->mi.bucket_size"); bch_err(c, "cur_idx %u/%u", ja->cur_idx, ja->nr); for (i = 0; i < 3; i++) { - unsigned idx = ja->cur_idx - 1 + i; + unsigned idx = (ja->cur_idx + ja->nr - 1 + i) % ja->nr; bch_err(c, "bucket_seq[%u] = %llu", idx, ja->bucket_seq[idx]); } ja->sectors_free = 0; @@ -1016,8 +1071,8 @@ void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c, div64_u64_rem(j->ptrs[i].sector, ca->mi.bucket_size, &offset); if (i) - pr_buf(out, " "); - pr_buf(out, "%u:%u:%u (sector %llu)", + prt_printf(out, " "); + prt_printf(out, "%u:%u:%u (sector %llu)", j->ptrs[i].dev, j->ptrs[i].bucket, j->ptrs[i].bucket_offset, @@ -1025,26 +1080,28 @@ void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c, } } -int bch2_journal_read(struct bch_fs *c, struct list_head *list, - 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, *t; + struct journal_replay *i, **_i, *prev = NULL; + struct genradix_iter radix_iter; struct bch_dev *ca; unsigned iter; struct printbuf buf = PRINTBUF; - size_t keys = 0, entries = 0; - bool degraded = false; - u64 seq, last_seq = 0; + bool degraded = false, last_write_torn = false; + u64 seq; int ret = 0; closure_init_stack(&jlist.cl); mutex_init(&jlist.lock); - jlist.head = list; + jlist.last_seq = 0; jlist.ret = 0; for_each_member_device(ca, c, iter) { - if (!test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) && + if (!c->opts.fsck && !(bch2_dev_has_data(c, ca) & (1 << BCH_DATA_journal))) continue; @@ -1064,44 +1121,74 @@ int bch2_journal_read(struct bch_fs *c, struct list_head *list, if (jlist.ret) return jlist.ret; - if (list_empty(list)) { - bch_info(c, "journal read done, but no entries found"); - return 0; - } - - i = list_last_entry(list, struct journal_replay, list); - *start_seq = le64_to_cpu(i->j.seq) + 1; + *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: */ - list_for_each_entry_safe_reverse(i, t, list, list) { - if (i->ignore) + genradix_for_each_reverse(&c->journal_entries, radix_iter, _i) { + int write = READ; + + i = *_i; + + if (!i || i->ignore) continue; - if (!JSET_NO_FLUSH(&i->j)) { - last_seq = le64_to_cpu(i->j.last_seq); - *blacklist_seq = le64_to_cpu(i->j.seq) + 1; - break; + if (!*start_seq) + *blacklist_seq = *start_seq = le64_to_cpu(i->j.seq) + 1; + + if (JSET_NO_FLUSH(&i->j)) { + i->ignore = true; + continue; + } + + if (!last_write_torn && !i->csum_good) { + last_write_torn = true; + i->ignore = true; + continue; } - journal_replay_free(c, i); + if (journal_entry_err_on(le64_to_cpu(i->j.last_seq) > le64_to_cpu(i->j.seq), + c, &i->j, NULL, + "invalid journal entry: last_seq > seq (%llu > %llu)", + le64_to_cpu(i->j.last_seq), + le64_to_cpu(i->j.seq))) + i->j.last_seq = i->j.seq; + + *last_seq = le64_to_cpu(i->j.last_seq); + *blacklist_seq = le64_to_cpu(i->j.seq) + 1; + break; } - if (!last_seq) { + if (!*start_seq) { + bch_info(c, "journal read done, but no entries found"); + return 0; + } + + if (!*last_seq) { fsck_err(c, "journal read done, but no entries found after dropping non-flushes"); - ret = -1; - goto err; + return 0; } + bch_info(c, "journal read done, replaying entries %llu-%llu", + *last_seq, *blacklist_seq - 1); + + if (*start_seq != *blacklist_seq) + bch_info(c, "dropped unflushed entries %llu-%llu", + *blacklist_seq, *start_seq - 1); + /* Drop blacklisted entries and entries older than last_seq: */ - list_for_each_entry_safe(i, t, list, list) { - if (i->ignore) + genradix_for_each(&c->journal_entries, radix_iter, _i) { + i = *_i; + + if (!i || i->ignore) continue; seq = le64_to_cpu(i->j.seq); - if (seq < last_seq) { + if (seq < *last_seq) { journal_replay_free(c, i); continue; } @@ -1109,15 +1196,16 @@ int bch2_journal_read(struct bch_fs *c, struct list_head *list, 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; - list_for_each_entry(i, list, list) { - if (i->ignore) + seq = *last_seq; + genradix_for_each(&c->journal_entries, radix_iter, _i) { + i = *_i; + + if (!i || i->ignore) continue; BUG_ON(seq > le64_to_cpu(i->j.seq)); @@ -1139,13 +1227,11 @@ int bch2_journal_read(struct bch_fs *c, struct list_head *list, !bch2_journal_seq_is_blacklisted(c, seq, false)) seq++; - if (i->list.prev != list) { - struct journal_replay *p = list_prev_entry(i, list); - - bch2_journal_ptrs_to_text(&buf1, c, p); - pr_buf(&buf1, " size %zu", vstruct_sectors(&p->j, c->block_bits)); + if (prev) { + bch2_journal_ptrs_to_text(&buf1, c, prev); + prt_printf(&buf1, " size %zu", vstruct_sectors(&prev->j, c->block_bits)); } else - pr_buf(&buf1, "(none)"); + prt_printf(&buf1, "(none)"); bch2_journal_ptrs_to_text(&buf2, c, i); missing_end = seq - 1; @@ -1153,29 +1239,43 @@ int bch2_journal_read(struct bch_fs *c, struct list_head *list, " 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); printbuf_exit(&buf2); } + prev = i; seq++; } - list_for_each_entry(i, list, list) { - struct jset_entry *entry; - struct bkey_i *k, *_n; + genradix_for_each(&c->journal_entries, radix_iter, _i) { struct bch_replicas_padded replicas = { .e.data_type = BCH_DATA_journal, .e.nr_required = 1, }; unsigned ptr; - if (i->ignore) + i = *_i; + if (!i || i->ignore) continue; - ret = jset_validate_entries(c, &i->j, READ); + for (ptr = 0; ptr < i->nr_ptrs; ptr++) { + struct bch_dev *ca = bch_dev_bkey_exists(c, i->ptrs[ptr].dev); + + if (!i->ptrs[ptr].csum_good) + bch_err_dev_offset(ca, i->ptrs[ptr].sector, + "invalid journal checksum, seq %llu%s", + le64_to_cpu(i->j.seq), + i->csum_good ? " (had good copy on another device)" : ""); + } + + ret = jset_validate(c, + bch_dev_bkey_exists(c, i->ptrs[0].dev), + &i->j, + i->ptrs[0].sector, + READ); if (ret) goto err; @@ -1193,26 +1293,14 @@ int bch2_journal_read(struct bch_fs *c, struct list_head *list, bch2_replicas_entry_to_text(&buf, &replicas.e); if (!degraded && - (test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) || - fsck_err_on(!bch2_replicas_marked(c, &replicas.e), c, - "superblock not marked as containing replicas %s", - buf.buf))) { + fsck_err_on(!bch2_replicas_marked(c, &replicas.e), c, + "superblock not marked as containing replicas %s", + buf.buf)) { ret = bch2_mark_replicas(c, &replicas.e); if (ret) goto err; } - - for_each_jset_key(k, _n, entry, &i->j) - keys++; - entries++; } - - bch_info(c, "journal read done, %zu keys in %zu entries, seq %llu", - keys, entries, *start_seq); - - if (*start_seq != *blacklist_seq) - bch_info(c, "dropped unflushed entries %llu-%llu", - *blacklist_seq, *start_seq - 1); err: fsck_err: printbuf_exit(&buf); @@ -1250,8 +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; @@ -1375,7 +1462,6 @@ static void journal_write_done(struct closure *cl) struct journal *j = container_of(cl, struct journal, io); struct bch_fs *c = container_of(j, struct bch_fs, journal); struct journal_buf *w = journal_last_unwritten_buf(j); - struct bch_replicas_padded replicas; union journal_res_state old, new; u64 v, seq; int err = 0; @@ -1387,13 +1473,7 @@ static void journal_write_done(struct closure *cl) if (!w->devs_written.nr) { bch_err(c, "unable to write journal to sufficient devices"); err = -EIO; - } else { - bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal, - w->devs_written); - if (bch2_mark_replicas(c, &replicas.e)) - err = -EIO; } - if (err) bch2_fatal_error(c); @@ -1410,6 +1490,8 @@ 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; @@ -1423,7 +1505,8 @@ static void journal_write_done(struct closure *cl) * Must come before signaling write completion, for * bch2_fs_journal_stop(): */ - journal_reclaim_kick(&c->journal); + if (j->watermark) + journal_reclaim_kick(&c->journal); /* also must come before signalling write completion: */ closure_debug_destroy(cl); @@ -1504,12 +1587,10 @@ static void do_journal_write(struct closure *cl) sectors); bio = ca->journal.bio; - bio_reset(bio); - bio_set_dev(bio, ca->disk_sb.bdev); + bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META); bio->bi_iter.bi_sector = ptr->offset; bio->bi_end_io = journal_write_endio; bio->bi_private = ca; - bio->bi_opf = REQ_OP_WRITE|REQ_SYNC|REQ_META; BUG_ON(bio->bi_iter.bi_sector == ca->prev_journal_sector); ca->prev_journal_sector = bio->bi_iter.bi_sector; @@ -1521,7 +1602,7 @@ static void do_journal_write(struct closure *cl) bch2_bio_map(bio, w->data, sectors << 9); - trace_journal_write(bio); + trace_and_count(c, journal_write, bio); closure_bio_submit(bio, cl); ca->journal.bucket_seq[ca->journal.cur_idx] = @@ -1532,12 +1613,59 @@ 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); struct bch_fs *c = container_of(j, struct bch_fs, journal); struct bch_dev *ca; struct journal_buf *w = journal_last_unwritten_buf(j); + struct bch_replicas_padded replicas; struct jset_entry *start, *end; struct jset *jset; struct bio *bio; @@ -1554,20 +1682,42 @@ void bch2_journal_write(struct closure *cl) j->write_start_time = local_clock(); spin_lock(&j->lock); - if (bch2_journal_error(j) || - w->noflush || - (!w->must_flush && - (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) && - test_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags))) { + + /* + * If the journal is in an error state - we did an emergency shutdown - + * we prefer to continue doing journal writes. We just mark them as + * noflush so they'll never be used, but they'll still be visible by the + * list_journal tool - this helps in debugging. + * + * There's a caveat: the first journal write after marking the + * superblock dirty must always be a flush write, because on startup + * from a clean shutdown we didn't necessarily read the journal and the + * new journal write might overwrite whatever was in the journal + * previously - we can't leave the journal without any flush writes in + * it. + * + * So if we're in an error state, and we're still starting up, we don't + * write anything at all. + */ + if (!test_bit(JOURNAL_NEED_FLUSH_WRITE, &j->flags) && + (bch2_journal_error(j) || + w->noflush || + (!w->must_flush && + (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) && + test_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags)))) { w->noflush = true; SET_JSET_NO_FLUSH(jset, true); jset->last_seq = 0; w->last_seq = 0; j->nr_noflush_writes++; - } else { + } else if (!bch2_journal_error(j)) { j->last_flush_write = jiffies; j->nr_flush_writes++; + clear_bit(JOURNAL_NEED_FLUSH_WRITE, &j->flags); + } else { + spin_unlock(&j->lock); + goto err; } spin_unlock(&j->lock); @@ -1581,7 +1731,7 @@ 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); @@ -1593,7 +1743,16 @@ void bch2_journal_write(struct closure *cl) BUG_ON(u64s > j->entry_u64s_reserved); le32_add_cpu(&jset->u64s, u64s); - BUG_ON(vstruct_sectors(jset, c->block_bits) > w->sectors); + + sectors = vstruct_sectors(jset, c->block_bits); + bytes = vstruct_bytes(jset); + + if (sectors > w->sectors) { + bch2_fs_fatal_error(c, "aieeee! journal write overran available space, %zu > %u (extra %u reserved %u/%u)", + vstruct_bytes(jset), w->sectors << 9, + u64s, w->u64s_reserved, j->entry_u64s_reserved); + goto err; + } jset->magic = cpu_to_le64(jset_magic(c)); jset->version = c->sb.version < bcachefs_metadata_version_bkey_renumber @@ -1613,7 +1772,7 @@ void bch2_journal_write(struct closure *cl) validate_before_checksum = true; if (validate_before_checksum && - jset_validate_for_write(c, jset)) + jset_validate(c, NULL, jset, 0, WRITE)) goto err; ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset), @@ -1627,13 +1786,9 @@ void bch2_journal_write(struct closure *cl) journal_nonce(jset), jset); if (!validate_before_checksum && - jset_validate_for_write(c, jset)) + jset_validate(c, NULL, jset, 0, WRITE)) goto err; - sectors = vstruct_sectors(jset, c->block_bits); - BUG_ON(sectors > w->sectors); - - bytes = vstruct_bytes(jset); memset((void *) jset + bytes, 0, (sectors << 9) - bytes); retry_alloc: @@ -1666,9 +1821,7 @@ retry_alloc: bch_err(c, "Unable to allocate journal write:\n%s", journal_debug_buf.buf); printbuf_exit(&journal_debug_buf); - bch2_fatal_error(c); - continue_at(cl, journal_write_done, c->io_complete_wq); - return; + goto err; } w->devs_written = bch2_bkey_devs(bkey_i_to_s_c(&w->key)); @@ -1682,14 +1835,22 @@ retry_alloc: if (nr_rw_members > 1) w->separate_flush = true; + /* + * Mark journal replicas before we submit the write to guarantee + * recovery will find the journal entries after a crash. + */ + bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal, + w->devs_written); + ret = bch2_mark_replicas(c, &replicas.e); + if (ret) + goto err; + if (!JSET_NO_FLUSH(jset) && w->separate_flush) { for_each_rw_member(ca, c, i) { percpu_ref_get(&ca->io_ref); bio = ca->journal.bio; - bio_reset(bio); - bio_set_dev(bio, ca->disk_sb.bdev); - bio->bi_opf = REQ_OP_FLUSH; + bio_reset(bio, ca->disk_sb.bdev, REQ_OP_FLUSH); bio->bi_end_io = journal_write_endio; bio->bi_private = ca; closure_bio_submit(bio, cl);