// SPDX-License-Identifier: GPL-2.0
#include "bcachefs.h"
+#include "alloc_background.h"
#include "alloc_foreground.h"
#include "btree_io.h"
#include "btree_update_interior.h"
#include "buckets.h"
#include "checksum.h"
+#include "disk_groups.h"
#include "error.h"
+#include "io.h"
#include "journal.h"
#include "journal_io.h"
#include "journal_reclaim.h"
+#include "journal_seq_blacklist.h"
#include "replicas.h"
#include <trace/events/bcachefs.h>
+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 bch_fs *c,
+ struct journal_replay *i)
+{
+ 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)
+{
+ i->ignore = true;
+
+ if (!c->opts.read_entire_journal)
+ __journal_replay_free(c, i);
+}
+
struct journal_list {
struct closure cl;
+ u64 last_seq;
struct mutex lock;
- struct list_head *head;
int ret;
};
* be replayed:
*/
static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca,
- struct journal_list *jlist, struct jset *j)
+ struct journal_ptr entry_ptr,
+ struct journal_list *jlist, struct jset *j,
+ bool bad)
{
- struct journal_replay *i, *pos;
- struct list_head *where;
+ struct genradix_iter iter;
+ struct journal_replay **_i, *i, *dup;
+ struct journal_ptr *ptr;
size_t bytes = vstruct_bytes(j);
- __le64 last_seq;
- int ret;
+ u64 last_seq = !JSET_NO_FLUSH(j) ? le64_to_cpu(j->last_seq) : 0;
+ int ret = JOURNAL_ENTRY_ADD_OK;
- last_seq = !list_empty(jlist->head)
- ? list_last_entry(jlist->head, struct journal_replay,
- list)->j.last_seq
- : 0;
+ /* Is this entry older than the range we need? */
+ if (!c->opts.read_entire_journal &&
+ le64_to_cpu(j->seq) < jlist->last_seq)
+ return JOURNAL_ENTRY_ADD_OUT_OF_RANGE;
- if (!c->opts.read_entire_journal) {
- /* Is this entry older than the range we need? */
- if (le64_to_cpu(j->seq) < le64_to_cpu(last_seq)) {
- ret = JOURNAL_ENTRY_ADD_OUT_OF_RANGE;
- goto out;
- }
+ /*
+ * 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 (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;
- /* Drop entries we don't need anymore */
- list_for_each_entry_safe(i, pos, jlist->head, list) {
- if (le64_to_cpu(i->j.seq) >= le64_to_cpu(j->last_seq))
+ if (le64_to_cpu(i->j.seq) >= last_seq)
break;
- list_del(&i->list);
- kvpfree(i, offsetof(struct journal_replay, j) +
- vstruct_bytes(&i->j));
+ journal_replay_free(c, i);
}
}
- list_for_each_entry_reverse(i, jlist->head, list) {
- /* Duplicate? */
- if (le64_to_cpu(j->seq) == le64_to_cpu(i->j.seq)) {
- fsck_err_on(bytes != vstruct_bytes(&i->j) ||
- memcmp(j, &i->j, bytes), c,
+ jlist->last_seq = max(jlist->last_seq, last_seq);
+
+ _i = genradix_ptr_alloc(&c->journal_entries,
+ journal_entry_radix_idx(c, le64_to_cpu(j->seq)),
+ GFP_KERNEL);
+ if (!_i)
+ return -ENOMEM;
+
+ /*
+ * 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) {
+ 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));
+ i = dup;
goto found;
}
+ }
+
+ i = kvpmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL);
+ if (!i)
+ return -ENOMEM;
+
+ i->nr_ptrs = 0;
+ i->bad = bad;
+ i->ignore = false;
+ memcpy(&i->j, j, bytes);
- if (le64_to_cpu(j->seq) > le64_to_cpu(i->j.seq)) {
- where = &i->list;
- goto add;
+ if (dup) {
+ i->nr_ptrs = dup->nr_ptrs;
+ memcpy(i->ptrs, dup->ptrs, sizeof(dup->ptrs));
+ __journal_replay_free(c, dup);
+ }
+
+
+ *_i = i;
+found:
+ for (ptr = i->ptrs; ptr < i->ptrs + i->nr_ptrs; ptr++) {
+ if (ptr->dev == ca->dev_idx) {
+ bch_err(c, "duplicate journal entry %llu on same device",
+ le64_to_cpu(i->j.seq));
+ goto out;
}
}
- where = jlist->head;
-add:
- i = kvpmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL);
- if (!i) {
- ret = -ENOMEM;
+ if (i->nr_ptrs >= ARRAY_SIZE(i->ptrs)) {
+ bch_err(c, "found too many copies of journal entry %llu",
+ le64_to_cpu(i->j.seq));
goto out;
}
- list_add(&i->list, where);
- i->devs.nr = 0;
- memcpy(&i->j, j, bytes);
-found:
- if (!bch2_dev_list_has_dev(i->devs, ca->dev_idx))
- bch2_dev_list_add_dev(&i->devs, ca->dev_idx);
- else
- fsck_err_on(1, c, "duplicate journal entries on same device");
- ret = JOURNAL_ENTRY_ADD_OK;
+ i->ptrs[i->nr_ptrs++] = entry_ptr;
out:
fsck_err:
return ret;
#define journal_entry_err_on(cond, c, msg, ...) \
((cond) ? journal_entry_err(c, msg, ##__VA_ARGS__) : false)
-static int journal_validate_key(struct bch_fs *c, struct jset *jset,
+#define FSCK_DELETED_KEY 5
+
+static int journal_validate_key(struct bch_fs *c, const char *where,
struct jset_entry *entry,
unsigned level, enum btree_id btree_id,
- struct bkey_i *k,
- const char *type, int write)
+ struct bkey_i *k, const char *type,
+ unsigned version, int big_endian, int write)
{
void *next = vstruct_next(entry);
- const char *invalid;
- unsigned version = le32_to_cpu(jset->version);
+ struct printbuf buf = PRINTBUF;
int ret = 0;
if (journal_entry_err_on(!k->k.u64s, c,
- "invalid %s in journal: k->u64s 0", type)) {
+ "invalid %s in %s entry offset %zi/%u: k->u64s 0",
+ type, where,
+ (u64 *) k - entry->_data,
+ le16_to_cpu(entry->u64s))) {
entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
journal_entry_null_range(vstruct_next(entry), next);
- return 0;
+ return FSCK_DELETED_KEY;
}
if (journal_entry_err_on((void *) bkey_next(k) >
(void *) vstruct_next(entry), c,
- "invalid %s in journal: extends past end of journal entry",
- type)) {
+ "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))) {
entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
journal_entry_null_range(vstruct_next(entry), next);
- return 0;
+ return FSCK_DELETED_KEY;
}
if (journal_entry_err_on(k->k.format != KEY_FORMAT_CURRENT, c,
- "invalid %s in journal: bad format %u",
- type, k->k.format)) {
- le16_add_cpu(&entry->u64s, -k->k.u64s);
+ "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)) {
+ 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);
- return 0;
+ return FSCK_DELETED_KEY;
}
if (!write)
- bch2_bkey_compat(level, btree_id, version,
- JSET_BIG_ENDIAN(jset), write,
- NULL, bkey_to_packed(k));
-
- invalid = bch2_bkey_invalid(c, bkey_i_to_s_c(k),
- __btree_node_type(level, btree_id));
- if (invalid) {
- char buf[160];
-
- bch2_bkey_val_to_text(&PBUF(buf), c, bkey_i_to_s_c(k));
- mustfix_fsck_err(c, "invalid %s in journal: %s\n%s",
- type, invalid, buf);
-
- le16_add_cpu(&entry->u64s, -k->k.u64s);
+ bch2_bkey_compat(level, btree_id, version, big_endian,
+ write, NULL, bkey_to_packed(k));
+
+ 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);
+
+ bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k));
+ pr_newline(&buf);
+ bch2_bkey_invalid(c, bkey_i_to_s_c(k),
+ __btree_node_type(level, btree_id), write, &buf);
+
+ mustfix_fsck_err(c, "%s", buf.buf);
+
+ 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);
- return 0;
+
+ printbuf_exit(&buf);
+ return FSCK_DELETED_KEY;
}
if (write)
- bch2_bkey_compat(level, btree_id, version,
- JSET_BIG_ENDIAN(jset), write,
- NULL, bkey_to_packed(k));
+ bch2_bkey_compat(level, btree_id, version, big_endian,
+ write, NULL, bkey_to_packed(k));
fsck_err:
+ printbuf_exit(&buf);
return ret;
}
-static int journal_entry_validate_btree_keys(struct bch_fs *c,
- struct jset *jset,
+static int journal_entry_btree_keys_validate(struct bch_fs *c,
+ const char *where,
struct jset_entry *entry,
- int write)
+ unsigned version, int big_endian, int write)
{
- struct bkey_i *k;
+ struct bkey_i *k = entry->start;
- vstruct_for_each(entry, k) {
- int ret = journal_validate_key(c, jset, entry,
+ while (k != vstruct_last(entry)) {
+ int ret = journal_validate_key(c, where, entry,
entry->level,
entry->btree_id,
- k, "key", write);
- if (ret)
- return ret;
+ k, "key", version, big_endian, write);
+ if (ret == FSCK_DELETED_KEY)
+ continue;
+
+ k = bkey_next(k);
}
return 0;
}
-static int journal_entry_validate_btree_root(struct bch_fs *c,
- struct jset *jset,
+static void journal_entry_btree_keys_to_text(struct printbuf *out, struct bch_fs *c,
+ struct jset_entry *entry)
+{
+ struct bkey_i *k;
+ bool first = true;
+
+ vstruct_for_each(entry, k) {
+ if (!first) {
+ pr_newline(out);
+ pr_buf(out, "%s: ", bch2_jset_entry_types[entry->type]);
+ }
+ pr_buf(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_entry *entry,
- int write)
+ unsigned version, int big_endian, int write)
{
struct bkey_i *k = entry->start;
int ret = 0;
return 0;
}
- return journal_validate_key(c, jset, entry, 1, entry->btree_id, k,
- "btree root", write);
+ return journal_validate_key(c, where, entry, 1, entry->btree_id, k,
+ "btree root", version, big_endian, write);
fsck_err:
return ret;
}
-static int journal_entry_validate_prio_ptrs(struct bch_fs *c,
- struct jset *jset,
+static void journal_entry_btree_root_to_text(struct printbuf *out, struct bch_fs *c,
+ struct jset_entry *entry)
+{
+ journal_entry_btree_keys_to_text(out, c, entry);
+}
+
+static int journal_entry_prio_ptrs_validate(struct bch_fs *c,
+ const char *where,
struct jset_entry *entry,
- int write)
+ unsigned version, int big_endian, int write)
{
/* obsolete, don't care: */
return 0;
}
-static int journal_entry_validate_blacklist(struct bch_fs *c,
- struct jset *jset,
+static void journal_entry_prio_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
+ struct jset_entry *entry)
+{
+}
+
+static int journal_entry_blacklist_validate(struct bch_fs *c,
+ const char *where,
struct jset_entry *entry,
- int write)
+ unsigned version, int big_endian, int write)
{
int ret = 0;
return ret;
}
-static int journal_entry_validate_blacklist_v2(struct bch_fs *c,
- struct jset *jset,
+static void journal_entry_blacklist_to_text(struct printbuf *out, struct bch_fs *c,
+ struct jset_entry *entry)
+{
+ struct jset_entry_blacklist *bl =
+ container_of(entry, struct jset_entry_blacklist, entry);
+
+ pr_buf(out, "seq=%llu", le64_to_cpu(bl->seq));
+}
+
+static int journal_entry_blacklist_v2_validate(struct bch_fs *c,
+ const char *where,
struct jset_entry *entry,
- int write)
+ unsigned version, int big_endian, int write)
{
struct jset_entry_blacklist_v2 *bl_entry;
int ret = 0;
return ret;
}
-static int journal_entry_validate_usage(struct bch_fs *c,
- struct jset *jset,
+static void journal_entry_blacklist_v2_to_text(struct printbuf *out, struct bch_fs *c,
+ struct jset_entry *entry)
+{
+ struct jset_entry_blacklist_v2 *bl =
+ container_of(entry, struct jset_entry_blacklist_v2, entry);
+
+ pr_buf(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_entry *entry,
- int write)
+ unsigned version, int big_endian, int write)
{
struct jset_entry_usage *u =
container_of(entry, struct jset_entry_usage, entry);
return ret;
}
-static int journal_entry_validate_data_usage(struct bch_fs *c,
- struct jset *jset,
+static void journal_entry_usage_to_text(struct printbuf *out, struct bch_fs *c,
+ struct jset_entry *entry)
+{
+ struct jset_entry_usage *u =
+ container_of(entry, struct jset_entry_usage, entry);
+
+ pr_buf(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_entry *entry,
- int write)
+ unsigned version, int big_endian, int write)
{
struct jset_entry_data_usage *u =
container_of(entry, struct jset_entry_data_usage, entry);
return ret;
}
+static void journal_entry_data_usage_to_text(struct printbuf *out, struct bch_fs *c,
+ struct jset_entry *entry)
+{
+ struct jset_entry_data_usage *u =
+ 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));
+}
+
+static int journal_entry_clock_validate(struct bch_fs *c,
+ const char *where,
+ struct jset_entry *entry,
+ unsigned version, int big_endian, int write)
+{
+ struct jset_entry_clock *clock =
+ container_of(entry, struct jset_entry_clock, entry);
+ unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
+ int ret = 0;
+
+ if (journal_entry_err_on(bytes != sizeof(*clock),
+ c, "invalid journal entry clock: 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")) {
+ journal_entry_null_range(entry, vstruct_next(entry));
+ return ret;
+ }
+
+fsck_err:
+ return ret;
+}
+
+static void journal_entry_clock_to_text(struct printbuf *out, struct bch_fs *c,
+ struct jset_entry *entry)
+{
+ 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));
+}
+
+static int journal_entry_dev_usage_validate(struct bch_fs *c,
+ const char *where,
+ struct jset_entry *entry,
+ unsigned version, int big_endian, int write)
+{
+ struct jset_entry_dev_usage *u =
+ container_of(entry, struct jset_entry_dev_usage, entry);
+ unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
+ unsigned expected = sizeof(*u);
+ unsigned dev;
+ int ret = 0;
+
+ if (journal_entry_err_on(bytes < expected,
+ c, "invalid journal entry dev usage: bad size (%u < %u)",
+ bytes, expected)) {
+ journal_entry_null_range(entry, vstruct_next(entry));
+ return ret;
+ }
+
+ dev = le32_to_cpu(u->dev);
+
+ if (journal_entry_err_on(!bch2_dev_exists2(c, dev),
+ c, "invalid journal entry dev usage: 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")) {
+ journal_entry_null_range(entry, vstruct_next(entry));
+ return ret;
+ }
+
+fsck_err:
+ return ret;
+}
+
+static void journal_entry_dev_usage_to_text(struct printbuf *out, struct bch_fs *c,
+ struct jset_entry *entry)
+{
+ struct jset_entry_dev_usage *u =
+ 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));
+
+ for (i = 0; i < nr_types; i++) {
+ if (i < BCH_DATA_NR)
+ pr_buf(out, " %s", bch2_data_types[i]);
+ else
+ pr_buf(out, " (unknown data type %u)", i);
+ pr_buf(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", le64_to_cpu(u->buckets_ec));
+}
+
+static int journal_entry_log_validate(struct bch_fs *c,
+ const char *where,
+ struct jset_entry *entry,
+ unsigned version, int big_endian, int write)
+{
+ return 0;
+}
+
+static void journal_entry_log_to_text(struct printbuf *out, struct bch_fs *c,
+ struct jset_entry *entry)
+{
+ 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);
+}
+
struct jset_entry_ops {
- int (*validate)(struct bch_fs *, struct jset *,
- struct jset_entry *, int);
+ int (*validate)(struct bch_fs *, const char *,
+ struct jset_entry *, unsigned, int, int);
+ void (*to_text)(struct printbuf *, struct bch_fs *, struct jset_entry *);
};
static const struct jset_entry_ops bch2_jset_entry_ops[] = {
#define x(f, nr) \
[BCH_JSET_ENTRY_##f] = (struct jset_entry_ops) { \
- .validate = journal_entry_validate_##f, \
+ .validate = journal_entry_##f##_validate, \
+ .to_text = journal_entry_##f##_to_text, \
},
BCH_JSET_ENTRY_TYPES()
#undef x
};
-static int journal_entry_validate(struct bch_fs *c, struct jset *jset,
- struct jset_entry *entry, int write)
+int bch2_journal_entry_validate(struct bch_fs *c, const char *where,
+ 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, jset,
- entry, write)
+ ? bch2_jset_entry_ops[entry->type].validate(c, where, entry,
+ version, big_endian, write)
: 0;
}
+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]);
+ bch2_jset_entry_ops[entry->type].to_text(out, c, entry);
+ } else {
+ pr_buf(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,
"journal entry extends past end of jset")) {
break;
}
- ret = journal_entry_validate(c, jset, entry, write);
+ ret = bch2_journal_entry_validate(c, buf, entry,
+ le32_to_cpu(jset->version),
+ JSET_BIG_ENDIAN(jset), write);
if (ret)
break;
}
}
static int jset_validate(struct bch_fs *c,
+ struct bch_dev *ca,
struct jset *jset, u64 sector,
unsigned bucket_sectors_left,
unsigned sectors_read,
return JOURNAL_ENTRY_NONE;
version = le32_to_cpu(jset->version);
- if ((version != BCH_JSET_VERSION_OLD &&
- version < bcachefs_metadata_version_min) ||
- version >= bcachefs_metadata_version_max) {
- bch_err(c, "unknown journal entry version %u", jset->version);
- return BCH_FSCK_UNKNOWN_VERSION;
+ if (journal_entry_err_on((version != BCH_JSET_VERSION_OLD &&
+ version < bcachefs_metadata_version_min) ||
+ version >= bcachefs_metadata_version_max, c,
+ "%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;
+
if (journal_entry_err_on(bytes > bucket_sectors_left << 9, c,
- "journal entry too big (%zu bytes), sector %lluu",
- bytes, sector)) {
- /* XXX: note we might have missing journal entries */
- return JOURNAL_ENTRY_BAD;
+ "%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 (bytes > sectors_read << 9)
- return JOURNAL_ENTRY_REREAD;
+ if (journal_entry_err_on(!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(jset)), c,
+ "%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))) {
+ ret = JOURNAL_ENTRY_BAD;
+ goto csum_done;
+ }
- if (fsck_err_on(!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(jset)), c,
- "journal entry with unknown csum type %llu sector %lluu",
- JSET_CSUM_TYPE(jset), sector))
- return JOURNAL_ENTRY_BAD;
+ 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,
- "journal checksum bad, sector %llu", sector)) {
- /* XXX: retry IO, when we start retrying checksum errors */
- /* XXX: note we might have missing journal entries */
- return JOURNAL_ENTRY_BAD;
- }
+ "%s sector %llu seq %llu: journal checksum bad",
+ ca ? ca->name : c->name,
+ sector, le64_to_cpu(jset->seq)))
+ ret = JOURNAL_ENTRY_BAD;
- bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset),
+ ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset),
jset->encrypted_start,
vstruct_end(jset) - (void *) jset->encrypted_start);
-
- if (journal_entry_err_on(le64_to_cpu(jset->last_seq) > le64_to_cpu(jset->seq), c,
- "invalid journal entry: last_seq > seq"))
+ 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,
+ "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 0;
+ return JOURNAL_ENTRY_BAD;
+ }
fsck_err:
return ret;
}
+static int jset_validate_for_write(struct bch_fs *c, struct jset *jset)
+{
+ unsigned sectors = vstruct_sectors(jset, c->block_bits);
+
+ return jset_validate(c, NULL, jset, 0, sectors, sectors, WRITE) ?:
+ jset_validate_entries(c, jset, WRITE);
+}
+
struct journal_read_buf {
void *data;
size_t size;
bio_put(bio);
if (bch2_dev_io_err_on(ret, ca,
- "journal read from sector %llu",
+ "journal read error: sector %llu",
offset) ||
- bch2_meta_read_fault("journal"))
- return -EIO;
+ bch2_meta_read_fault("journal")) {
+ /*
+ * We don't error out of the recovery process
+ * here, since the relevant journal entry may be
+ * found on a different device, and missing or
+ * no journal entries will be handled later
+ */
+ return 0;
+ }
j = buf->data;
}
- ret = jset_validate(c, j, offset,
+ ret = jset_validate(c, ca, j, offset,
end - offset, sectors_read,
READ);
switch (ret) {
case BCH_FSCK_OK:
+ sectors = vstruct_sectors(j, c->block_bits);
break;
case JOURNAL_ENTRY_REREAD:
if (vstruct_bytes(j) > buf->size) {
case JOURNAL_ENTRY_NONE:
if (!saw_bad)
return 0;
- sectors = c->opts.block_size;
+ sectors = block_sectors(c);
goto next_block;
case JOURNAL_ENTRY_BAD:
saw_bad = true;
- sectors = c->opts.block_size;
- goto next_block;
+ /*
+ * 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;
default:
return ret;
}
ja->bucket_seq[bucket] = le64_to_cpu(j->seq);
mutex_lock(&jlist->lock);
- ret = journal_entry_add(c, ca, jlist, j);
+ ret = journal_entry_add(c, ca, (struct journal_ptr) {
+ .dev = ca->dev_idx,
+ .bucket = bucket,
+ .bucket_offset = offset -
+ bucket_to_sector(ca, ja->buckets[bucket]),
+ .sector = offset,
+ }, jlist, j, ret != 0);
mutex_unlock(&jlist->lock);
switch (ret) {
default:
return ret;
}
-
- sectors = vstruct_sectors(j, c->block_bits);
next_block:
pr_debug("next");
offset += sectors;
struct journal_device *ja =
container_of(cl, struct journal_device, read);
struct bch_dev *ca = container_of(ja, struct bch_dev, journal);
+ struct bch_fs *c = ca->fs;
struct journal_list *jlist =
container_of(cl->parent, struct journal_list, cl);
+ 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;
+ int ret = 0;
if (!ja->nr)
goto out;
* allocate
*/
while (ja->bucket_seq[ja->cur_idx] > min_seq &&
- ja->bucket_seq[ja->cur_idx] >
+ 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 = 0;
+ ja->sectors_free = ca->mi.bucket_size;
+
+ mutex_lock(&jlist->lock);
+ genradix_for_each(&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 = 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);
+ }
+ }
+ }
+ mutex_unlock(&jlist->lock);
+
+ if (ja->bucket_seq[ja->cur_idx] &&
+ ja->sectors_free == ca->mi.bucket_size) {
+ 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 + ja->nr - 1 + i) % ja->nr;
+ bch_err(c, "bucket_seq[%u] = %llu", idx, ja->bucket_seq[idx]);
+ }
+ ja->sectors_free = 0;
+ }
/*
* Set dirty_idx to indicate the entire journal is full and needs to be
ja->discard_idx = ja->dirty_idx_ondisk =
ja->dirty_idx = (ja->cur_idx + 1) % ja->nr;
out:
+ bch_verbose(c, "journal read done on device %s, ret %i", ca->name, ret);
kvpfree(buf.data, buf.size);
percpu_ref_put(&ca->io_ref);
closure_return(cl);
goto out;
}
-int bch2_journal_read(struct bch_fs *c, struct list_head *list)
+void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
+ struct journal_replay *j)
+{
+ unsigned i;
+
+ for (i = 0; i < j->nr_ptrs; i++) {
+ struct bch_dev *ca = bch_dev_bkey_exists(c, j->ptrs[i].dev);
+ u64 offset;
+
+ 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)",
+ j->ptrs[i].dev,
+ j->ptrs[i].bucket,
+ j->ptrs[i].bucket_offset,
+ j->ptrs[i].sector);
+ }
+}
+
+int bch2_journal_read(struct bch_fs *c, u64 *blacklist_seq, u64 *start_seq)
{
struct journal_list jlist;
- struct journal_replay *i;
+ 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;
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) &&
- !(bch2_dev_has_data(c, ca) & (1 << BCH_DATA_JOURNAL)))
+ if (!c->opts.fsck &&
+ !(bch2_dev_has_data(c, ca) & (1 << BCH_DATA_journal)))
continue;
- if ((ca->mi.state == BCH_MEMBER_STATE_RW ||
- ca->mi.state == BCH_MEMBER_STATE_RO) &&
+ if ((ca->mi.state == BCH_MEMBER_STATE_rw ||
+ ca->mi.state == BCH_MEMBER_STATE_ro) &&
percpu_ref_tryget(&ca->io_ref))
closure_call(&ca->journal.read,
bch2_journal_read_device,
if (jlist.ret)
return jlist.ret;
- list_for_each_entry(i, list, list) {
+ *start_seq = 0;
+
+ /*
+ * Find most recent flush entry, and ignore newer non flush entries -
+ * those entries will be blacklisted:
+ */
+ genradix_for_each_reverse(&c->journal_entries, radix_iter, _i) {
+ i = *_i;
+
+ if (!i || i->ignore)
+ continue;
+
+ if (!*start_seq)
+ *start_seq = le64_to_cpu(i->j.seq) + 1;
+
+ 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;
+ }
+
+ journal_replay_free(c, i);
+ }
+
+ 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;
+ }
+
+ /* Drop blacklisted entries and entries older than last_seq: */
+ 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) {
+ journal_replay_free(c, i);
+ continue;
+ }
+
+ 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);
+ }
+ }
+
+ /* Check for missing entries: */
+ 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));
+
+ while (seq < le64_to_cpu(i->j.seq)) {
+ u64 missing_start, missing_end;
+ struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
+
+ while (seq < le64_to_cpu(i->j.seq) &&
+ bch2_journal_seq_is_blacklisted(c, seq, false))
+ seq++;
+
+ if (seq == le64_to_cpu(i->j.seq))
+ break;
+
+ missing_start = seq;
+
+ while (seq < le64_to_cpu(i->j.seq) &&
+ !bch2_journal_seq_is_blacklisted(c, seq, false))
+ seq++;
+
+ if (prev) {
+ bch2_journal_ptrs_to_text(&buf1, c, prev);
+ pr_buf(&buf1, " size %zu", vstruct_sectors(&prev->j, c->block_bits));
+ } else
+ pr_buf(&buf1, "(none)");
+ bch2_journal_ptrs_to_text(&buf2, c, i);
+
+ missing_end = seq - 1;
+ fsck_err(c, "journal entries %llu-%llu missing! (replaying %llu-%llu)\n"
+ " prev at %s\n"
+ " next at %s",
+ missing_start, missing_end,
+ last_seq, *blacklist_seq - 1,
+ buf1.buf, buf2.buf);
+
+ printbuf_exit(&buf1);
+ printbuf_exit(&buf2);
+ }
+
+ prev = i;
+ seq++;
+ }
+
+ genradix_for_each(&c->journal_entries, radix_iter, _i) {
struct jset_entry *entry;
struct bkey_i *k, *_n;
- struct bch_replicas_padded replicas;
- char buf[80];
+ struct bch_replicas_padded replicas = {
+ .e.data_type = BCH_DATA_journal,
+ .e.nr_required = 1,
+ };
+ unsigned ptr;
+
+ i = *_i;
+ if (!i || i->ignore)
+ continue;
ret = jset_validate_entries(c, &i->j, READ);
if (ret)
- goto fsck_err;
+ goto err;
+
+ for (ptr = 0; ptr < i->nr_ptrs; ptr++)
+ replicas.e.devs[replicas.e.nr_devs++] = i->ptrs[ptr].dev;
+
+ bch2_replicas_entry_sort(&replicas.e);
/*
* If we're mounting in degraded mode - if we didn't read all
* the devices - this is wrong:
*/
- bch2_devlist_to_replicas(&replicas.e, BCH_DATA_JOURNAL, i->devs);
+ printbuf_reset(&buf);
+ 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, false), c,
- "superblock not marked as containing replicas %s",
- (bch2_replicas_entry_to_text(&PBUF(buf),
- &replicas.e), 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)
- return ret;
+ goto err;
}
for_each_jset_key(k, _n, entry, &i->j)
entries++;
}
- if (!list_empty(list)) {
- i = list_last_entry(list, struct journal_replay, list);
+ bch_info(c, "journal read done, %zu keys in %zu entries, seq %llu",
+ keys, entries, *start_seq);
- bch_info(c, "journal read done, %zu keys in %zu entries, seq %llu",
- keys, entries, le64_to_cpu(i->j.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);
return ret;
}
* it:
*/
if (!ca->mi.durability ||
- ca->mi.state != BCH_MEMBER_STATE_RW ||
+ ca->mi.state != BCH_MEMBER_STATE_rw ||
!ja->nr ||
bch2_bkey_has_device(bkey_i_to_s_c(&w->key),
ca->dev_idx) ||
sectors > ja->sectors_free)
continue;
- bch2_dev_stripe_increment(c, ca, &j->wp.stripe);
+ bch2_dev_stripe_increment(ca, &j->wp.stripe);
bch2_bkey_append_ptr(&w->key,
(struct bch_extent_ptr) {
unsigned sectors)
{
struct bch_fs *c = container_of(j, struct bch_fs, journal);
+ struct bch_devs_mask devs;
struct journal_device *ja;
struct bch_dev *ca;
struct dev_alloc_list devs_sorted;
+ unsigned target = c->opts.metadata_target ?:
+ c->opts.foreground_target;
unsigned i, replicas = 0, replicas_want =
READ_ONCE(c->opts.metadata_replicas);
rcu_read_lock();
+retry:
+ devs = target_rw_devs(c, BCH_DATA_journal, target);
- devs_sorted = bch2_dev_alloc_list(c, &j->wp.stripe,
- &c->rw_devs[BCH_DATA_JOURNAL]);
+ devs_sorted = bch2_dev_alloc_list(c, &j->wp.stripe, &devs);
__journal_write_alloc(j, w, &devs_sorted,
sectors, &replicas, replicas_want);
__journal_write_alloc(j, w, &devs_sorted,
sectors, &replicas, replicas_want);
+
+ if (replicas < replicas_want && target) {
+ /* Retry from all devices: */
+ target = 0;
+ goto retry;
+ }
done:
rcu_read_unlock();
- return replicas >= c->opts.metadata_replicas_required ? 0 : -EROFS;
-}
-
-static void journal_write_compact(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);
+ BUG_ON(bkey_val_u64s(&w->key.k) > BCH_REPLICAS_MAX);
- /* Empty entry: */
- if (!u64s)
- continue;
-
- /* 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);
+ return replicas >= c->opts.metadata_replicas_required ? 0 : -EROFS;
}
static void journal_buf_realloc(struct journal *j, struct journal_buf *buf)
return;
memcpy(new_buf, buf->data, buf->buf_size);
- kvpfree(buf->data, buf->buf_size);
- buf->data = new_buf;
- buf->buf_size = new_size;
+
+ spin_lock(&j->lock);
+ swap(buf->data, new_buf);
+ swap(buf->buf_size, new_size);
+ spin_unlock(&j->lock);
+
+ kvpfree(new_buf, new_size);
+}
+
+static inline struct journal_buf *journal_last_unwritten_buf(struct journal *j)
+{
+ return j->buf + (journal_last_unwritten_seq(j) & JOURNAL_BUF_MASK);
}
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_prev_buf(j);
- struct bch_devs_list devs =
- bch2_bkey_devs(bkey_i_to_s_c(&w->key));
+ struct journal_buf *w = journal_last_unwritten_buf(j);
struct bch_replicas_padded replicas;
- u64 seq = le64_to_cpu(w->data->seq);
- u64 last_seq = le64_to_cpu(w->data->last_seq);
+ union journal_res_state old, new;
+ u64 v, seq;
+ int err = 0;
- bch2_time_stats_update(j->write_time, j->write_start_time);
+ bch2_time_stats_update(!JSET_NO_FLUSH(w->data)
+ ? j->flush_write_time
+ : j->noflush_write_time, j->write_start_time);
- if (!devs.nr) {
+ if (!w->devs_written.nr) {
bch_err(c, "unable to write journal to sufficient devices");
- goto err;
+ err = -EIO;
+ } else {
+ bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
+ w->devs_written);
+ if (bch2_mark_replicas(c, &replicas.e))
+ err = -EIO;
}
- bch2_devlist_to_replicas(&replicas.e, BCH_DATA_JOURNAL, devs);
-
- if (bch2_mark_replicas(c, &replicas.e))
- goto err;
+ if (err)
+ bch2_fatal_error(c);
spin_lock(&j->lock);
+ seq = le64_to_cpu(w->data->seq);
+
if (seq >= j->pin.front)
- journal_seq_pin(j, seq)->devs = devs;
+ journal_seq_pin(j, seq)->devs = w->devs_written;
+
+ if (!err) {
+ if (!JSET_NO_FLUSH(w->data)) {
+ j->flushed_seq_ondisk = seq;
+ j->last_seq_ondisk = w->last_seq;
+
+ bch2_do_discards(c);
+ closure_wake_up(&c->freelist_wait);
+ }
+ } else if (!j->err_seq || seq < j->err_seq)
+ j->err_seq = seq;
j->seq_ondisk = seq;
- j->last_seq_ondisk = last_seq;
- bch2_journal_space_available(j);
/*
* Updating last_seq_ondisk may let bch2_journal_reclaim_work() discard
* Must come before signaling write completion, for
* bch2_fs_journal_stop():
*/
- mod_delayed_work(c->journal_reclaim_wq, &j->reclaim_work, 0);
-out:
+ if (j->watermark)
+ journal_reclaim_kick(&c->journal);
+
/* also must come before signalling write completion: */
closure_debug_destroy(cl);
- BUG_ON(!j->reservations.prev_buf_unwritten);
- atomic64_sub(((union journal_res_state) { .prev_buf_unwritten = 1 }).v,
- &j->reservations.counter);
+ v = atomic64_read(&j->reservations.counter);
+ do {
+ old.v = new.v = v;
+ BUG_ON(journal_state_count(new, new.unwritten_idx));
+
+ new.unwritten_idx++;
+ } while ((v = atomic64_cmpxchg(&j->reservations.counter,
+ old.v, new.v)) != old.v);
+
+ bch2_journal_space_available(j);
closure_wake_up(&w->wait);
journal_wake(j);
- if (test_bit(JOURNAL_NEED_WRITE, &j->flags))
- mod_delayed_work(system_freezable_wq, &j->write_work, 0);
+ if (!journal_state_count(new, new.unwritten_idx) &&
+ journal_last_unwritten_seq(j) <= journal_cur_seq(j)) {
+ closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL);
+ } else if (journal_last_unwritten_seq(j) == journal_cur_seq(j) &&
+ new.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL) {
+ struct journal_buf *buf = journal_cur_buf(j);
+ long delta = buf->expires - jiffies;
+
+ /*
+ * We don't close a journal entry to write it while there's
+ * previous entries still in flight - the current journal entry
+ * might want to be written now:
+ */
+
+ mod_delayed_work(c->io_complete_wq, &j->write_work, max(0L, delta));
+ }
+
spin_unlock(&j->lock);
- return;
-err:
- bch2_fatal_error(c);
- spin_lock(&j->lock);
- goto out;
}
static void journal_write_endio(struct bio *bio)
{
struct bch_dev *ca = bio->bi_private;
struct journal *j = &ca->fs->journal;
+ struct journal_buf *w = journal_last_unwritten_buf(j);
+ unsigned long flags;
- if (bch2_dev_io_err_on(bio->bi_status, ca, "journal write") ||
+ if (bch2_dev_io_err_on(bio->bi_status, ca, "error writing journal entry %llu: %s",
+ le64_to_cpu(w->data->seq),
+ bch2_blk_status_to_str(bio->bi_status)) ||
bch2_meta_write_fault("journal")) {
- struct journal_buf *w = journal_prev_buf(j);
- unsigned long flags;
-
spin_lock_irqsave(&j->err_lock, flags);
- bch2_bkey_drop_device(bkey_i_to_s(&w->key), ca->dev_idx);
+ bch2_dev_list_drop_dev(&w->devs_written, ca->dev_idx);
spin_unlock_irqrestore(&j->err_lock, flags);
}
percpu_ref_put(&ca->io_ref);
}
+static void do_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_extent_ptr *ptr;
+ struct bio *bio;
+ unsigned sectors = vstruct_sectors(w->data, c->block_bits);
+
+ extent_for_each_ptr(bkey_i_to_s_extent(&w->key), ptr) {
+ ca = bch_dev_bkey_exists(c, ptr->dev);
+ if (!percpu_ref_tryget(&ca->io_ref)) {
+ /* XXX: fix this */
+ bch_err(c, "missing device for journal write\n");
+ continue;
+ }
+
+ this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_journal],
+ sectors);
+
+ bio = ca->journal.bio;
+ bio_reset(bio);
+ bio_set_dev(bio, ca->disk_sb.bdev);
+ 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;
+
+ if (!JSET_NO_FLUSH(w->data))
+ bio->bi_opf |= REQ_FUA;
+ if (!JSET_NO_FLUSH(w->data) && !w->separate_flush)
+ bio->bi_opf |= REQ_PREFLUSH;
+
+ bch2_bio_map(bio, w->data, sectors << 9);
+
+ trace_journal_write(bio);
+ closure_bio_submit(bio, cl);
+
+ ca->journal.bucket_seq[ca->journal.cur_idx] =
+ le64_to_cpu(w->data->seq);
+ }
+
+ continue_at(cl, journal_write_done, c->io_complete_wq);
+ return;
+}
+
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_prev_buf(j);
+ struct journal_buf *w = journal_last_unwritten_buf(j);
struct jset_entry *start, *end;
struct jset *jset;
struct bio *bio;
- struct bch_extent_ptr *ptr;
+ struct printbuf journal_debug_buf = PRINTBUF;
bool validate_before_checksum = false;
- unsigned i, sectors, bytes, u64s;
+ unsigned i, sectors, bytes, u64s, nr_rw_members = 0;
int ret;
- bch2_journal_pin_put(j, le64_to_cpu(w->data->seq));
+ BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
journal_buf_realloc(j, w);
jset = w->data;
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))) {
+ w->noflush = true;
+ SET_JSET_NO_FLUSH(jset, true);
+ jset->last_seq = 0;
+ w->last_seq = 0;
+
+ j->nr_noflush_writes++;
+ } else {
+ j->last_flush_write = jiffies;
+ j->nr_flush_writes++;
+ }
+ spin_unlock(&j->lock);
+
/*
* New btree roots are set by journalling them; when the journal entry
* gets written we have to propagate them to c->btree_roots
end = bch2_btree_roots_to_journal_entries(c, jset->start, end);
- end = bch2_journal_super_entries_add_common(c, end,
- le64_to_cpu(jset->seq));
+ bch2_journal_super_entries_add_common(c, &end,
+ le64_to_cpu(jset->seq));
u64s = (u64 *) end - (u64 *) start;
BUG_ON(u64s > j->entry_u64s_reserved);
le32_add_cpu(&jset->u64s, u64s);
BUG_ON(vstruct_sectors(jset, c->block_bits) > w->sectors);
- journal_write_compact(jset);
-
- jset->read_clock = cpu_to_le16(c->bucket_clock[READ].hand);
- jset->write_clock = cpu_to_le16(c->bucket_clock[WRITE].hand);
jset->magic = cpu_to_le64(jset_magic(c));
-
- jset->version = c->sb.version < bcachefs_metadata_version_new_versioning
+ jset->version = c->sb.version < bcachefs_metadata_version_bkey_renumber
? cpu_to_le32(BCH_JSET_VERSION_OLD)
: cpu_to_le32(c->sb.version);
SET_JSET_BIG_ENDIAN(jset, CPU_BIG_ENDIAN);
SET_JSET_CSUM_TYPE(jset, bch2_meta_checksum_type(c));
+ if (!JSET_NO_FLUSH(jset) && journal_entry_empty(jset))
+ j->last_empty_seq = le64_to_cpu(jset->seq);
+
if (bch2_csum_type_is_encryption(JSET_CSUM_TYPE(jset)))
validate_before_checksum = true;
- if (le32_to_cpu(jset->version) < bcachefs_metadata_version_max)
+ if (le32_to_cpu(jset->version) < bcachefs_metadata_version_current)
validate_before_checksum = true;
if (validate_before_checksum &&
- jset_validate_entries(c, jset, WRITE))
+ jset_validate_for_write(c, jset))
goto err;
- bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset),
+ ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset),
jset->encrypted_start,
vstruct_end(jset) - (void *) jset->encrypted_start);
+ if (bch2_fs_fatal_err_on(ret, c,
+ "error decrypting journal entry: %i", ret))
+ goto err;
jset->csum = csum_vstruct(c, JSET_CSUM_TYPE(jset),
journal_nonce(jset), jset);
if (!validate_before_checksum &&
- jset_validate_entries(c, jset, WRITE))
+ jset_validate_for_write(c, jset))
goto err;
sectors = vstruct_sectors(jset, c->block_bits);
goto retry_alloc;
}
+ if (ret)
+ __bch2_journal_debug_to_text(&journal_debug_buf, j);
+
/*
* write is allocated, no longer need to account for it in
* bch2_journal_space_available():
spin_unlock(&j->lock);
if (ret) {
- bch_err(c, "Unable to allocate journal write");
+ 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, system_highpri_wq);
+ continue_at(cl, journal_write_done, c->io_complete_wq);
return;
}
- /*
- * XXX: we really should just disable the entire journal in nochanges
- * mode
- */
+ w->devs_written = bch2_bkey_devs(bkey_i_to_s_c(&w->key));
+
if (c->opts.nochanges)
goto no_io;
- extent_for_each_ptr(bkey_i_to_s_extent(&w->key), ptr) {
- ca = bch_dev_bkey_exists(c, ptr->dev);
- if (!percpu_ref_tryget(&ca->io_ref)) {
- /* XXX: fix this */
- bch_err(c, "missing device for journal write\n");
- continue;
- }
-
- this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_JOURNAL],
- sectors);
-
- bio = ca->journal.bio;
- bio_reset(bio);
- bio_set_dev(bio, ca->disk_sb.bdev);
- bio->bi_iter.bi_sector = ptr->offset;
- bio->bi_end_io = journal_write_endio;
- bio->bi_private = ca;
- bio_set_op_attrs(bio, REQ_OP_WRITE,
- REQ_SYNC|REQ_META|REQ_PREFLUSH|REQ_FUA);
- bch2_bio_map(bio, jset, sectors << 9);
-
- trace_journal_write(bio);
- closure_bio_submit(bio, cl);
+ for_each_rw_member(ca, c, i)
+ nr_rw_members++;
- ca->journal.bucket_seq[ca->journal.cur_idx] = le64_to_cpu(jset->seq);
- }
+ if (nr_rw_members > 1)
+ w->separate_flush = true;
- for_each_rw_member(ca, c, i)
- if (journal_flushes_device(ca) &&
- !bch2_bkey_has_device(bkey_i_to_s_c(&w->key), i)) {
+ 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->bi_private = ca;
closure_bio_submit(bio, cl);
}
+ }
+ continue_at(cl, do_journal_write, c->io_complete_wq);
+ return;
no_io:
- bch2_bucket_seq_cleanup(c);
-
- continue_at(cl, journal_write_done, system_highpri_wq);
+ continue_at(cl, journal_write_done, c->io_complete_wq);
return;
err:
- bch2_inconsistent_error(c);
- continue_at(cl, journal_write_done, system_highpri_wq);
+ bch2_fatal_error(c);
+ continue_at(cl, journal_write_done, c->io_complete_wq);
}
projects / bcachefs-tools-debian / blobdiff