+// SPDX-License-Identifier: GPL-2.0
#include "bcachefs.h"
+#include "btree_update_interior.h"
+#include "buckets.h"
#include "checksum.h"
#include "disk_groups.h"
+#include "ec.h"
#include "error.h"
#include "io.h"
+#include "journal.h"
+#include "journal_seq_blacklist.h"
#include "replicas.h"
#include "quota.h"
#include "super-io.h"
BUG_ON(get_order(__vstruct_bytes(struct bch_sb, sb_u64s)) >
sb->page_order);
- if (!f) {
+ if (!f && !u64s) {
+ /* nothing to do: */
+ } else if (!f) {
f = vstruct_last(sb->sb);
memset(f, 0, sizeof(u64) * u64s);
f->u64s = cpu_to_le32(u64s);
void *src, *dst;
src = vstruct_end(f);
- f->u64s = cpu_to_le32(u64s);
- dst = vstruct_end(f);
+
+ if (u64s) {
+ f->u64s = cpu_to_le32(u64s);
+ dst = vstruct_end(f);
+ } else {
+ dst = f;
+ }
memmove(dst, src, vstruct_end(sb->sb) - src);
sb->sb->u64s = cpu_to_le32(sb_u64s);
- return f;
+ return u64s ? f : NULL;
+}
+
+void bch2_sb_field_delete(struct bch_sb_handle *sb,
+ enum bch_sb_field_type type)
+{
+ struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
+
+ if (f)
+ __bch2_sb_field_resize(sb, f, 0);
}
/* Superblock realloc/free: */
struct bch_sb *new_sb;
struct bio *bio;
+ if (sb->sb && sb->page_order >= order)
+ return 0;
+
if (sb->have_layout) {
u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
sb->bio = bio;
}
- new_sb = (void *) __get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
+ new_sb = (void *) __get_free_pages(GFP_NOFS|__GFP_ZERO, order);
if (!new_sb)
return -ENOMEM;
}
}
+ f = bch2_sb_field_get(sb->sb, type);
f = __bch2_sb_field_resize(sb, f, u64s);
- f->type = cpu_to_le32(type);
+ if (f)
+ f->type = cpu_to_le32(type);
return f;
}
struct bch_sb_field *f;
struct bch_sb_field_members *mi;
const char *err;
+ u32 version, version_min;
u16 block_size;
- if (le64_to_cpu(sb->version) < BCH_SB_VERSION_MIN ||
- le64_to_cpu(sb->version) > BCH_SB_VERSION_MAX)
- return"Unsupported superblock version";
+ version = le16_to_cpu(sb->version);
+ version_min = version >= bcachefs_metadata_version_new_versioning
+ ? le16_to_cpu(sb->version_min)
+ : version;
- if (le64_to_cpu(sb->version) < BCH_SB_VERSION_EXTENT_MAX) {
- SET_BCH_SB_ENCODED_EXTENT_MAX_BITS(sb, 7);
- SET_BCH_SB_POSIX_ACL(sb, 1);
- }
+ if (version >= bcachefs_metadata_version_max ||
+ version_min < bcachefs_metadata_version_min)
+ return "Unsupported superblock version";
+
+ if (version_min > version)
+ return "Bad minimum version";
+
+ if (sb->features[1] ||
+ (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR)))
+ return "Filesystem has incompatible features";
block_size = le16_to_cpu(sb->block_size);
return err;
}
- if (le64_to_cpu(sb->version) < BCH_SB_VERSION_EXTENT_NONCE_V1 &&
- bch2_sb_get_crypt(sb) &&
- BCH_SB_INITIALIZED(sb))
- return "Incompatible extent nonces";
-
- sb->version = cpu_to_le64(BCH_SB_VERSION_MAX);
-
return NULL;
}
c->sb.uuid = src->uuid;
c->sb.user_uuid = src->user_uuid;
+ c->sb.version = le16_to_cpu(src->version);
c->sb.nr_devices = src->nr_devices;
c->sb.clean = BCH_SB_CLEAN(src);
c->sb.encryption_type = BCH_SB_ENCRYPTION_TYPE(src);
c->sb.time_base_hi = le32_to_cpu(src->time_base_hi);
c->sb.time_precision = le32_to_cpu(src->time_precision);
c->sb.features = le64_to_cpu(src->features[0]);
+ c->sb.compat = le64_to_cpu(src->compat[0]);
for_each_member_device(ca, c, i)
ca->mi = bch2_mi_to_cpu(mi->members + i);
{
struct bch_sb_field *src_f, *dst_f;
struct bch_sb *dst = dst_handle->sb;
+ unsigned i;
dst->version = src->version;
+ dst->version_min = src->version_min;
dst->seq = src->seq;
dst->uuid = src->uuid;
dst->user_uuid = src->user_uuid;
memcpy(dst->features, src->features, sizeof(dst->features));
memcpy(dst->compat, src->compat, sizeof(dst->compat));
- vstruct_for_each(src, src_f) {
- if (src_f->type == BCH_SB_FIELD_journal)
+ for (i = 0; i < BCH_SB_FIELD_NR; i++) {
+ if (i == BCH_SB_FIELD_journal)
continue;
- dst_f = bch2_sb_field_get(dst, le32_to_cpu(src_f->type));
+ src_f = bch2_sb_field_get(src, i);
+ dst_f = bch2_sb_field_get(dst, i);
dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
- le32_to_cpu(src_f->u64s));
+ src_f ? le32_to_cpu(src_f->u64s) : 0);
- memcpy(dst_f, src_f, vstruct_bytes(src_f));
+ if (src_f)
+ memcpy(dst_f, src_f, vstruct_bytes(src_f));
}
}
bio_reset(sb->bio);
bio_set_dev(sb->bio, sb->bdev);
sb->bio->bi_iter.bi_sector = offset;
- sb->bio->bi_iter.bi_size = PAGE_SIZE << sb->page_order;
bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
- bch2_bio_map(sb->bio, sb->sb);
+ bch2_bio_map(sb->bio, sb->sb, PAGE_SIZE << sb->page_order);
if (submit_bio_wait(sb->bio))
return "IO error";
if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
return "Not a bcachefs superblock";
- if (le64_to_cpu(sb->sb->version) < BCH_SB_VERSION_MIN ||
- le64_to_cpu(sb->sb->version) > BCH_SB_VERSION_MAX)
- return"Unsupported superblock version";
+ if (le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_min ||
+ le16_to_cpu(sb->sb->version) >= bcachefs_metadata_version_max)
+ return "Unsupported superblock version";
bytes = vstruct_bytes(sb->sb);
if (bch2_crc_cmp(csum, sb->sb->csum))
return "bad checksum reading superblock";
+ sb->seq = le64_to_cpu(sb->sb->seq);
+
return NULL;
}
bio_reset(sb->bio);
bio_set_dev(sb->bio, sb->bdev);
sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
- sb->bio->bi_iter.bi_size = sizeof(struct bch_sb_layout);
bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
/*
* use sb buffer to read layout, since sb buffer is page aligned but
* layout won't be:
*/
- bch2_bio_map(sb->bio, sb->sb);
+ bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
err = "IO error";
if (submit_bio_wait(sb->bio))
bdev_logical_block_size(sb->bdev))
goto err;
- if (sb->mode & FMODE_WRITE)
- bdev_get_queue(sb->bdev)->backing_dev_info->capabilities
- |= BDI_CAP_STABLE_WRITES;
ret = 0;
sb->have_layout = true;
out:
/* XXX: return errors directly */
- if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write"))
+ if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
+ bch2_blk_status_to_str(bio->bi_status)))
ca->sb_write_error = 1;
closure_put(&ca->fs->sb_write);
percpu_ref_put(&ca->io_ref);
}
+static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
+{
+ struct bch_sb *sb = ca->disk_sb.sb;
+ struct bio *bio = ca->disk_sb.bio;
+
+ bio_reset(bio);
+ bio_set_dev(bio, ca->disk_sb.bdev);
+ bio->bi_iter.bi_sector = le64_to_cpu(sb->layout.sb_offset[0]);
+ bio->bi_end_io = write_super_endio;
+ bio->bi_private = ca;
+ bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
+ bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
+
+ this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
+ bio_sectors(bio));
+
+ percpu_ref_get(&ca->io_ref);
+ closure_bio_submit(bio, &c->sb_write);
+}
+
static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
{
struct bch_sb *sb = ca->disk_sb.sb;
bio_reset(bio);
bio_set_dev(bio, ca->disk_sb.bdev);
bio->bi_iter.bi_sector = le64_to_cpu(sb->offset);
- bio->bi_iter.bi_size =
- roundup(vstruct_bytes(sb),
- bdev_logical_block_size(ca->disk_sb.bdev));
bio->bi_end_io = write_super_endio;
bio->bi_private = ca;
bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
- bch2_bio_map(bio, sb);
+ bch2_bio_map(bio, sb,
+ roundup((size_t) vstruct_bytes(sb),
+ bdev_logical_block_size(ca->disk_sb.bdev)));
- this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_SB],
+ this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
bio_sectors(bio));
percpu_ref_get(&ca->io_ref);
closure_bio_submit(bio, &c->sb_write);
}
-void bch2_write_super(struct bch_fs *c)
+int bch2_write_super(struct bch_fs *c)
{
struct closure *cl = &c->sb_write;
struct bch_dev *ca;
const char *err;
struct bch_devs_mask sb_written;
bool wrote, can_mount_without_written, can_mount_with_written;
+ int ret = 0;
lockdep_assert_held(&c->sb_lock);
le64_add_cpu(&c->disk_sb.sb->seq, 1);
+ if (test_bit(BCH_FS_ERROR, &c->flags))
+ SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
+
for_each_online_member(ca, c, i)
bch2_sb_from_fs(c, ca);
err = bch2_sb_validate(&ca->disk_sb);
if (err) {
bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
+ ret = -1;
goto out;
}
}
- if (c->opts.nochanges ||
- test_bit(BCH_FS_ERROR, &c->flags))
+ if (c->opts.nochanges)
goto out;
for_each_online_member(ca, c, i) {
ca->sb_write_error = 0;
}
+ for_each_online_member(ca, c, i)
+ read_back_super(c, ca);
+ closure_sync(cl);
+
+ for_each_online_member(ca, c, i) {
+ if (!ca->sb_write_error &&
+ ca->disk_sb.seq !=
+ le64_to_cpu(ca->sb_read_scratch->seq)) {
+ bch2_fs_fatal_error(c,
+ "Superblock modified by another process");
+ percpu_ref_put(&ca->io_ref);
+ ret = -EROFS;
+ goto out;
+ }
+ }
+
do {
wrote = false;
for_each_online_member(ca, c, i)
- if (sb < ca->disk_sb.sb->layout.nr_superblocks) {
+ if (!ca->sb_write_error &&
+ sb < ca->disk_sb.sb->layout.nr_superblocks) {
write_one_super(c, ca, sb);
wrote = true;
}
sb++;
} while (wrote);
- for_each_online_member(ca, c, i)
+ for_each_online_member(ca, c, i) {
if (ca->sb_write_error)
__clear_bit(ca->dev_idx, sb_written.d);
+ else
+ ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
+ }
nr_wrote = dev_mask_nr(&sb_written);
* written anything (new filesystem), we continue if we'd be able to
* mount with the devices we did successfully write to:
*/
- bch2_fs_fatal_err_on(!nr_wrote ||
- (can_mount_without_written &&
- !can_mount_with_written), c,
- "Unable to write superblock to sufficient devices");
+ if (bch2_fs_fatal_err_on(!nr_wrote ||
+ (can_mount_without_written &&
+ !can_mount_with_written), c,
+ "Unable to write superblock to sufficient devices"))
+ ret = -1;
out:
/* Make new options visible after they're persistent: */
bch2_sb_update(c);
+ return ret;
+}
+
+void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
+{
+ mutex_lock(&c->sb_lock);
+ if (!(c->sb.features & (1ULL << feat))) {
+ c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
+
+ bch2_write_super(c);
+ }
+ mutex_unlock(&c->sb_lock);
}
/* BCH_SB_FIELD_journal: */
return "Too many buckets";
if (le64_to_cpu(m->nbuckets) -
- le16_to_cpu(m->first_bucket) < 1 << 10)
+ le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
return "Not enough buckets";
if (le16_to_cpu(m->bucket_size) <
return "bucket size smaller than btree node size";
}
- if (le64_to_cpu(sb->version) < BCH_SB_VERSION_EXTENT_MAX)
- for (m = mi->members;
- m < mi->members + sb->nr_devices;
- m++)
- SET_BCH_MEMBER_DATA_ALLOWED(m, ~0);
-
return NULL;
}
.validate = bch2_sb_validate_crypt,
};
+/* BCH_SB_FIELD_clean: */
+
+void bch2_sb_clean_renumber(struct bch_sb_field_clean *clean, int write)
+{
+ struct jset_entry *entry;
+
+ for (entry = clean->start;
+ entry < (struct jset_entry *) vstruct_end(&clean->field);
+ entry = vstruct_next(entry))
+ bch2_bkey_renumber(BKEY_TYPE_BTREE, bkey_to_packed(entry->start), write);
+}
+
+int bch2_fs_mark_dirty(struct bch_fs *c)
+{
+ int ret;
+
+ /*
+ * Unconditionally write superblock, to verify it hasn't changed before
+ * we go rw:
+ */
+
+ mutex_lock(&c->sb_lock);
+ SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
+ c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_extent_overwrite;
+ c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_extents_above_btree_updates;
+ c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_btree_updates_journalled;
+ ret = bch2_write_super(c);
+ mutex_unlock(&c->sb_lock);
+
+ return ret;
+}
+
+static void
+entry_init_u64s(struct jset_entry *entry, unsigned u64s)
+{
+ memset(entry, 0, u64s * sizeof(u64));
+
+ /*
+ * The u64s field counts from the start of data, ignoring the shared
+ * fields.
+ */
+ entry->u64s = u64s - 1;
+}
+
+static void
+entry_init_size(struct jset_entry *entry, size_t size)
+{
+ unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
+ entry_init_u64s(entry, u64s);
+}
+
+struct jset_entry *
+bch2_journal_super_entries_add_common(struct bch_fs *c,
+ struct jset_entry *entry,
+ u64 journal_seq)
+{
+ unsigned i;
+
+ percpu_down_write(&c->mark_lock);
+
+ if (!journal_seq) {
+ for (i = 0; i < ARRAY_SIZE(c->usage); i++)
+ bch2_fs_usage_acc_to_base(c, i);
+ } else {
+ bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
+ }
+
+ {
+ struct jset_entry_usage *u =
+ container_of(entry, struct jset_entry_usage, entry);
+
+ entry_init_size(entry, sizeof(*u));
+ u->entry.type = BCH_JSET_ENTRY_usage;
+ u->entry.btree_id = FS_USAGE_INODES;
+ u->v = cpu_to_le64(c->usage_base->nr_inodes);
+
+ entry = vstruct_next(entry);
+ }
+
+ {
+ struct jset_entry_usage *u =
+ container_of(entry, struct jset_entry_usage, entry);
+
+ entry_init_size(entry, sizeof(*u));
+ u->entry.type = BCH_JSET_ENTRY_usage;
+ u->entry.btree_id = FS_USAGE_KEY_VERSION;
+ u->v = cpu_to_le64(atomic64_read(&c->key_version));
+
+ entry = vstruct_next(entry);
+ }
+
+ for (i = 0; i < BCH_REPLICAS_MAX; i++) {
+ struct jset_entry_usage *u =
+ container_of(entry, struct jset_entry_usage, entry);
+
+ entry_init_size(entry, sizeof(*u));
+ u->entry.type = BCH_JSET_ENTRY_usage;
+ u->entry.btree_id = FS_USAGE_RESERVED;
+ u->entry.level = i;
+ u->v = cpu_to_le64(c->usage_base->persistent_reserved[i]);
+
+ entry = vstruct_next(entry);
+ }
+
+ for (i = 0; i < c->replicas.nr; i++) {
+ struct bch_replicas_entry *e =
+ cpu_replicas_entry(&c->replicas, i);
+ struct jset_entry_data_usage *u =
+ container_of(entry, struct jset_entry_data_usage, entry);
+
+ entry_init_size(entry, sizeof(*u) + e->nr_devs);
+ u->entry.type = BCH_JSET_ENTRY_data_usage;
+ u->v = cpu_to_le64(c->usage_base->replicas[i]);
+ memcpy(&u->r, e, replicas_entry_bytes(e));
+
+ entry = vstruct_next(entry);
+ }
+
+ percpu_up_write(&c->mark_lock);
+
+ return entry;
+}
+
+void bch2_fs_mark_clean(struct bch_fs *c)
+{
+ struct bch_sb_field_clean *sb_clean;
+ struct jset_entry *entry;
+ unsigned u64s;
+
+ mutex_lock(&c->sb_lock);
+ if (BCH_SB_CLEAN(c->disk_sb.sb))
+ goto out;
+
+ SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
+
+ c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
+ c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA;
+ c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_extents_above_btree_updates);
+ c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_btree_updates_journalled);
+
+ u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
+
+ sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
+ if (!sb_clean) {
+ bch_err(c, "error resizing superblock while setting filesystem clean");
+ goto out;
+ }
+
+ sb_clean->flags = 0;
+ sb_clean->read_clock = cpu_to_le16(c->bucket_clock[READ].hand);
+ sb_clean->write_clock = cpu_to_le16(c->bucket_clock[WRITE].hand);
+ sb_clean->journal_seq = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
+
+ /* Trying to catch outstanding bug: */
+ BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
+
+ entry = sb_clean->start;
+ entry = bch2_journal_super_entries_add_common(c, entry, 0);
+ entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
+ BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
+
+ memset(entry, 0,
+ vstruct_end(&sb_clean->field) - (void *) entry);
+
+ if (le16_to_cpu(c->disk_sb.sb->version) <
+ bcachefs_metadata_version_bkey_renumber)
+ bch2_sb_clean_renumber(sb_clean, WRITE);
+
+ bch2_write_super(c);
+out:
+ mutex_unlock(&c->sb_lock);
+}
+
+static const char *bch2_sb_validate_clean(struct bch_sb *sb,
+ struct bch_sb_field *f)
+{
+ struct bch_sb_field_clean *clean = field_to_type(f, clean);
+
+ if (vstruct_bytes(&clean->field) < sizeof(*clean))
+ return "invalid field crypt: wrong size";
+
+ return NULL;
+}
+
+static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
+ .validate = bch2_sb_validate_clean,
+};
+
static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
#define x(f, nr) \
[BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
: NULL;
}
-size_t bch2_sb_field_to_text(char *buf, size_t size,
- struct bch_sb *sb, struct bch_sb_field *f)
+void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
+ struct bch_sb_field *f)
{
unsigned type = le32_to_cpu(f->type);
- size_t (*to_text)(char *, size_t, struct bch_sb *,
- struct bch_sb_field *) =
- type < BCH_SB_FIELD_NR
- ? bch2_sb_field_ops[type]->to_text
- : NULL;
+ const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
+ ? bch2_sb_field_ops[type] : NULL;
- if (!to_text) {
- if (size)
- buf[0] = '\0';
- return 0;
- }
+ if (ops)
+ pr_buf(out, "%s", bch2_sb_fields[type]);
+ else
+ pr_buf(out, "(unknown field %u)", type);
+
+ pr_buf(out, " (size %llu):", vstruct_bytes(f));
- return to_text(buf, size, sb, f);
+ if (ops && ops->to_text)
+ bch2_sb_field_ops[type]->to_text(out, sb, f);
}