// SPDX-License-Identifier: GPL-2.0
#include "bcachefs.h"
-#include "buckets.h"
#include "checksum.h"
+#include "counters.h"
#include "disk_groups.h"
#include "ec.h"
#include "error.h"
-#include "io.h"
#include "journal.h"
+#include "journal_sb.h"
#include "journal_seq_blacklist.h"
+#include "recovery.h"
#include "replicas.h"
#include "quota.h"
+#include "sb-clean.h"
+#include "sb-downgrade.h"
+#include "sb-errors.h"
+#include "sb-members.h"
#include "super-io.h"
#include "super.h"
+#include "trace.h"
#include "vstructs.h"
#include <linux/backing-dev.h>
#include <linux/sort.h>
+static const struct blk_holder_ops bch2_sb_handle_bdev_ops = {
+};
+
+struct bch2_metadata_version {
+ u16 version;
+ const char *name;
+};
+
+static const struct bch2_metadata_version bch2_metadata_versions[] = {
+#define x(n, v) { \
+ .version = v, \
+ .name = #n, \
+},
+ BCH_METADATA_VERSIONS()
+#undef x
+};
+
+void bch2_version_to_text(struct printbuf *out, unsigned v)
+{
+ const char *str = "(unknown version)";
+
+ for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
+ if (bch2_metadata_versions[i].version == v) {
+ str = bch2_metadata_versions[i].name;
+ break;
+ }
+
+ prt_printf(out, "%u.%u: %s", BCH_VERSION_MAJOR(v), BCH_VERSION_MINOR(v), str);
+}
+
+unsigned bch2_latest_compatible_version(unsigned v)
+{
+ if (!BCH_VERSION_MAJOR(v))
+ return v;
+
+ for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
+ if (bch2_metadata_versions[i].version > v &&
+ BCH_VERSION_MAJOR(bch2_metadata_versions[i].version) ==
+ BCH_VERSION_MAJOR(v))
+ v = bch2_metadata_versions[i].version;
+
+ return v;
+}
+
const char * const bch2_sb_fields[] = {
#define x(name, nr) #name,
BCH_SB_FIELDS()
NULL
};
-static const char *bch2_sb_field_validate(struct bch_sb *,
- struct bch_sb_field *);
+static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
+ struct printbuf *);
-struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
+struct bch_sb_field *bch2_sb_field_get_id(struct bch_sb *sb,
enum bch_sb_field_type type)
{
- struct bch_sb_field *f;
-
/* XXX: need locking around superblock to access optional fields */
vstruct_for_each(sb, f)
unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
- BUG_ON(get_order(__vstruct_bytes(struct bch_sb, sb_u64s)) >
- sb->page_order);
+ BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
if (!f && !u64s) {
/* nothing to do: */
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);
+ struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
if (f)
__bch2_sb_field_resize(sb, f, 0);
void bch2_free_super(struct bch_sb_handle *sb)
{
- if (sb->bio)
- bio_put(sb->bio);
+ kfree(sb->bio);
if (!IS_ERR_OR_NULL(sb->bdev))
- blkdev_put(sb->bdev, sb->mode);
+ blkdev_put(sb->bdev, sb->holder);
+ kfree(sb->holder);
+ kfree(sb->sb_name);
- free_pages((unsigned long) sb->sb, sb->page_order);
+ kfree(sb->sb);
memset(sb, 0, sizeof(*sb));
}
int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
{
size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
- unsigned order = get_order(new_bytes);
+ size_t new_buffer_size;
struct bch_sb *new_sb;
struct bio *bio;
- if (sb->sb && sb->page_order >= order)
+ if (sb->bdev)
+ new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
+
+ new_buffer_size = roundup_pow_of_two(new_bytes);
+
+ if (sb->sb && sb->buffer_size >= new_buffer_size)
return 0;
- if (sb->have_layout) {
+ if (sb->sb && sb->have_layout) {
u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
if (new_bytes > max_bytes) {
- char buf[BDEVNAME_SIZE];
+ struct printbuf buf = PRINTBUF;
- pr_err("%s: superblock too big: want %zu but have %llu",
- bdevname(sb->bdev, buf), new_bytes, max_bytes);
- return -ENOSPC;
+ prt_bdevname(&buf, sb->bdev);
+ prt_printf(&buf, ": superblock too big: want %zu but have %llu", new_bytes, max_bytes);
+ pr_err("%s", buf.buf);
+ printbuf_exit(&buf);
+ return -BCH_ERR_ENOSPC_sb;
}
}
- if (sb->page_order >= order && sb->sb)
+ if (sb->buffer_size >= new_buffer_size && sb->sb)
return 0;
if (dynamic_fault("bcachefs:add:super_realloc"))
- return -ENOMEM;
+ return -BCH_ERR_ENOMEM_sb_realloc_injected;
+
+ new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
+ if (!new_sb)
+ return -BCH_ERR_ENOMEM_sb_buf_realloc;
+
+ sb->sb = new_sb;
if (sb->have_bio) {
- bio = bio_kmalloc(GFP_KERNEL, 1 << order);
+ unsigned nr_bvecs = buf_pages(sb->sb, new_buffer_size);
+
+ bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
if (!bio)
- return -ENOMEM;
+ return -BCH_ERR_ENOMEM_sb_bio_realloc;
+
+ bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
- if (sb->bio)
- bio_put(sb->bio);
+ kfree(sb->bio);
sb->bio = bio;
}
- new_sb = (void *) __get_free_pages(GFP_NOFS|__GFP_ZERO, order);
- if (!new_sb)
- return -ENOMEM;
-
- if (sb->sb)
- memcpy(new_sb, sb->sb, PAGE_SIZE << sb->page_order);
-
- free_pages((unsigned long) sb->sb, sb->page_order);
- sb->sb = new_sb;
-
- sb->page_order = order;
+ sb->buffer_size = new_buffer_size;
return 0;
}
-struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
+struct bch_sb_field *bch2_sb_field_resize_id(struct bch_sb_handle *sb,
enum bch_sb_field_type type,
unsigned u64s)
{
- struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
+ struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
ssize_t d = -old_u64s + u64s;
if (sb->fs_sb) {
struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
- struct bch_dev *ca;
- unsigned i;
lockdep_assert_held(&c->sb_lock);
/* XXX: we're not checking that offline device have enough space */
- for_each_online_member(ca, c, i) {
- struct bch_sb_handle *sb = &ca->disk_sb;
+ for_each_online_member(c, ca) {
+ struct bch_sb_handle *dev_sb = &ca->disk_sb;
- if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
+ if (bch2_sb_realloc(dev_sb, le32_to_cpu(dev_sb->sb->u64s) + d)) {
percpu_ref_put(&ca->ref);
return NULL;
}
}
}
- f = bch2_sb_field_get(sb->sb, type);
+ f = bch2_sb_field_get_id(sb->sb, type);
f = __bch2_sb_field_resize(sb, f, u64s);
if (f)
f->type = cpu_to_le32(type);
return f;
}
-/* Superblock validate: */
-
-static inline void __bch2_sb_layout_size_assert(void)
+struct bch_sb_field *bch2_sb_field_get_minsize_id(struct bch_sb_handle *sb,
+ enum bch_sb_field_type type,
+ unsigned u64s)
{
- BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
+ struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
+
+ if (!f || le32_to_cpu(f->u64s) < u64s)
+ f = bch2_sb_field_resize_id(sb, type, u64s);
+ return f;
}
-static const char *validate_sb_layout(struct bch_sb_layout *layout)
+/* Superblock validate: */
+
+static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
{
u64 offset, prev_offset, max_sectors;
unsigned i;
- if (uuid_le_cmp(layout->magic, BCACHE_MAGIC))
- return "Not a bcachefs superblock layout";
+ BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
- if (layout->layout_type != 0)
- return "Invalid superblock layout type";
+ if (!uuid_equal(&layout->magic, &BCACHE_MAGIC) &&
+ !uuid_equal(&layout->magic, &BCHFS_MAGIC)) {
+ prt_printf(out, "Not a bcachefs superblock layout");
+ return -BCH_ERR_invalid_sb_layout;
+ }
- if (!layout->nr_superblocks)
- return "Invalid superblock layout: no superblocks";
+ if (layout->layout_type != 0) {
+ prt_printf(out, "Invalid superblock layout type %u",
+ layout->layout_type);
+ return -BCH_ERR_invalid_sb_layout_type;
+ }
+
+ if (!layout->nr_superblocks) {
+ prt_printf(out, "Invalid superblock layout: no superblocks");
+ return -BCH_ERR_invalid_sb_layout_nr_superblocks;
+ }
- if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset))
- return "Invalid superblock layout: too many superblocks";
+ if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
+ prt_printf(out, "Invalid superblock layout: too many superblocks");
+ return -BCH_ERR_invalid_sb_layout_nr_superblocks;
+ }
max_sectors = 1 << layout->sb_max_size_bits;
for (i = 1; i < layout->nr_superblocks; i++) {
offset = le64_to_cpu(layout->sb_offset[i]);
- if (offset < prev_offset + max_sectors)
- return "Invalid superblock layout: superblocks overlap";
+ if (offset < prev_offset + max_sectors) {
+ prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
+ " (sb %u ends at %llu next starts at %llu",
+ i - 1, prev_offset + max_sectors, offset);
+ return -BCH_ERR_invalid_sb_layout_superblocks_overlap;
+ }
prev_offset = offset;
}
- return NULL;
+ return 0;
}
-const char *bch2_sb_validate(struct bch_sb_handle *disk_sb)
+static int bch2_sb_compatible(struct bch_sb *sb, struct printbuf *out)
{
- struct bch_sb *sb = disk_sb->sb;
- struct bch_sb_field *f;
- struct bch_sb_field_members *mi;
- const char *err;
- u32 version, version_min;
- u16 block_size;
+ u16 version = le16_to_cpu(sb->version);
+ u16 version_min = le16_to_cpu(sb->version_min);
+
+ if (!bch2_version_compatible(version)) {
+ prt_str(out, "Unsupported superblock version ");
+ bch2_version_to_text(out, version);
+ prt_str(out, " (min ");
+ bch2_version_to_text(out, bcachefs_metadata_version_min);
+ prt_str(out, ", max ");
+ bch2_version_to_text(out, bcachefs_metadata_version_current);
+ prt_str(out, ")");
+ return -BCH_ERR_invalid_sb_version;
+ }
+
+ if (!bch2_version_compatible(version_min)) {
+ prt_str(out, "Unsupported superblock version_min ");
+ bch2_version_to_text(out, version_min);
+ prt_str(out, " (min ");
+ bch2_version_to_text(out, bcachefs_metadata_version_min);
+ prt_str(out, ", max ");
+ bch2_version_to_text(out, bcachefs_metadata_version_current);
+ prt_str(out, ")");
+ return -BCH_ERR_invalid_sb_version;
+ }
+
+ if (version_min > version) {
+ prt_str(out, "Bad minimum version ");
+ bch2_version_to_text(out, version_min);
+ prt_str(out, ", greater than version field ");
+ bch2_version_to_text(out, version);
+ return -BCH_ERR_invalid_sb_version;
+ }
- version = le16_to_cpu(sb->version);
- version_min = version >= bcachefs_metadata_version_new_versioning
- ? le16_to_cpu(sb->version_min)
- : version;
+ return 0;
+}
- if (version >= bcachefs_metadata_version_max ||
- version_min < bcachefs_metadata_version_min)
- return "Unsupported superblock version";
+static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
+ int rw)
+{
+ struct bch_sb *sb = disk_sb->sb;
+ struct bch_sb_field_members_v1 *mi;
+ enum bch_opt_id opt_id;
+ u16 block_size;
+ int ret;
- if (version_min > version)
- return "Bad minimum version";
+ ret = bch2_sb_compatible(sb, out);
+ if (ret)
+ return ret;
if (sb->features[1] ||
- (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR)))
- return "Filesystem has incompatible features";
+ (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
+ prt_printf(out, "Filesystem has incompatible features");
+ return -BCH_ERR_invalid_sb_features;
+ }
block_size = le16_to_cpu(sb->block_size);
- if (!is_power_of_2(block_size) ||
- block_size > PAGE_SECTORS)
- return "Bad block size";
+ if (block_size > PAGE_SECTORS) {
+ prt_printf(out, "Block size too big (got %u, max %u)",
+ block_size, PAGE_SECTORS);
+ return -BCH_ERR_invalid_sb_block_size;
+ }
- if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le)))
- return "Bad user UUID";
+ if (bch2_is_zero(sb->user_uuid.b, sizeof(sb->user_uuid))) {
+ prt_printf(out, "Bad user UUID (got zeroes)");
+ return -BCH_ERR_invalid_sb_uuid;
+ }
- if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le)))
- return "Bad internal UUID";
+ if (bch2_is_zero(sb->uuid.b, sizeof(sb->uuid))) {
+ prt_printf(out, "Bad internal UUID (got zeroes)");
+ return -BCH_ERR_invalid_sb_uuid;
+ }
if (!sb->nr_devices ||
- sb->nr_devices <= sb->dev_idx ||
- sb->nr_devices > BCH_SB_MEMBERS_MAX)
- return "Bad number of member devices";
-
- if (!BCH_SB_META_REPLICAS_WANT(sb) ||
- BCH_SB_META_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
- return "Invalid number of metadata replicas";
-
- if (!BCH_SB_META_REPLICAS_REQ(sb) ||
- BCH_SB_META_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
- return "Invalid number of metadata replicas";
-
- if (!BCH_SB_DATA_REPLICAS_WANT(sb) ||
- BCH_SB_DATA_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
- return "Invalid number of data replicas";
-
- if (!BCH_SB_DATA_REPLICAS_REQ(sb) ||
- BCH_SB_DATA_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
- return "Invalid number of data replicas";
+ sb->nr_devices > BCH_SB_MEMBERS_MAX) {
+ prt_printf(out, "Bad number of member devices %u (max %u)",
+ sb->nr_devices, BCH_SB_MEMBERS_MAX);
+ return -BCH_ERR_invalid_sb_too_many_members;
+ }
- if (BCH_SB_META_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
- return "Invalid metadata checksum type";
+ if (sb->dev_idx >= sb->nr_devices) {
+ prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
+ sb->dev_idx, sb->nr_devices);
+ return -BCH_ERR_invalid_sb_dev_idx;
+ }
- if (BCH_SB_DATA_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
- return "Invalid metadata checksum type";
+ if (!sb->time_precision ||
+ le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
+ prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
+ le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
+ return -BCH_ERR_invalid_sb_time_precision;
+ }
- if (BCH_SB_COMPRESSION_TYPE(sb) >= BCH_COMPRESSION_OPT_NR)
- return "Invalid compression type";
+ if (rw == READ) {
+ /*
+ * Been seeing a bug where these are getting inexplicably
+ * zeroed, so we're now validating them, but we have to be
+ * careful not to preven people's filesystems from mounting:
+ */
+ if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
+ SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
+ if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
+ SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
+
+ if (!BCH_SB_VERSION_UPGRADE_COMPLETE(sb))
+ SET_BCH_SB_VERSION_UPGRADE_COMPLETE(sb, le16_to_cpu(sb->version));
+ }
- if (!BCH_SB_BTREE_NODE_SIZE(sb))
- return "Btree node size not set";
+ for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
+ const struct bch_option *opt = bch2_opt_table + opt_id;
- if (!is_power_of_2(BCH_SB_BTREE_NODE_SIZE(sb)))
- return "Btree node size not a power of two";
+ if (opt->get_sb != BCH2_NO_SB_OPT) {
+ u64 v = bch2_opt_from_sb(sb, opt_id);
- if (BCH_SB_GC_RESERVE(sb) < 5)
- return "gc reserve percentage too small";
+ prt_printf(out, "Invalid option ");
+ ret = bch2_opt_validate(opt, v, out);
+ if (ret)
+ return ret;
- if (!sb->time_precision ||
- le32_to_cpu(sb->time_precision) > NSEC_PER_SEC)
- return "invalid time precision";
+ printbuf_reset(out);
+ }
+ }
/* validate layout */
- err = validate_sb_layout(&sb->layout);
- if (err)
- return err;
+ ret = validate_sb_layout(&sb->layout, out);
+ if (ret)
+ return ret;
vstruct_for_each(sb, f) {
- if (!f->u64s)
- return "Invalid superblock: invalid optional field";
+ if (!f->u64s) {
+ prt_printf(out, "Invalid superblock: optional field with size 0 (type %u)",
+ le32_to_cpu(f->type));
+ return -BCH_ERR_invalid_sb_field_size;
+ }
- if (vstruct_next(f) > vstruct_last(sb))
- return "Invalid superblock: invalid optional field";
+ if (vstruct_next(f) > vstruct_last(sb)) {
+ prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
+ le32_to_cpu(f->type));
+ return -BCH_ERR_invalid_sb_field_size;
+ }
}
/* members must be validated first: */
- mi = bch2_sb_get_members(sb);
- if (!mi)
- return "Invalid superblock: member info area missing";
+ mi = bch2_sb_field_get(sb, members_v1);
+ if (!mi) {
+ prt_printf(out, "Invalid superblock: member info area missing");
+ return -BCH_ERR_invalid_sb_members_missing;
+ }
- err = bch2_sb_field_validate(sb, &mi->field);
- if (err)
- return err;
+ ret = bch2_sb_field_validate(sb, &mi->field, out);
+ if (ret)
+ return ret;
vstruct_for_each(sb, f) {
- if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
+ if (le32_to_cpu(f->type) == BCH_SB_FIELD_members_v1)
continue;
- err = bch2_sb_field_validate(sb, f);
- if (err)
- return err;
+ ret = bch2_sb_field_validate(sb, f, out);
+ if (ret)
+ return ret;
}
- return NULL;
+ return 0;
}
/* device open: */
+static unsigned long le_ulong_to_cpu(unsigned long v)
+{
+ return sizeof(unsigned long) == 8
+ ? le64_to_cpu(v)
+ : le32_to_cpu(v);
+}
+
+static void le_bitvector_to_cpu(unsigned long *dst, unsigned long *src, unsigned nr)
+{
+ BUG_ON(nr & (BITS_PER_TYPE(long) - 1));
+
+ for (unsigned i = 0; i < BITS_TO_LONGS(nr); i++)
+ dst[i] = le_ulong_to_cpu(src[i]);
+}
+
static void bch2_sb_update(struct bch_fs *c)
{
struct bch_sb *src = c->disk_sb.sb;
- struct bch_sb_field_members *mi = bch2_sb_get_members(src);
- struct bch_dev *ca;
- unsigned i;
lockdep_assert_held(&c->sb_lock);
c->sb.uuid = src->uuid;
c->sb.user_uuid = src->user_uuid;
c->sb.version = le16_to_cpu(src->version);
+ c->sb.version_min = le16_to_cpu(src->version_min);
+ c->sb.version_upgrade_complete = BCH_SB_VERSION_UPGRADE_COMPLETE(src);
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.encoded_extent_max= 1 << BCH_SB_ENCODED_EXTENT_MAX_BITS(src);
- c->sb.time_base_lo = le64_to_cpu(src->time_base_lo);
+
+ c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
+ c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
+
+ /* XXX this is wrong, we need a 96 or 128 bit integer type */
+ c->sb.time_base_lo = div_u64(le64_to_cpu(src->time_base_lo),
+ c->sb.nsec_per_time_unit);
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);
+ memset(c->sb.errors_silent, 0, sizeof(c->sb.errors_silent));
+
+ struct bch_sb_field_ext *ext = bch2_sb_field_get(src, ext);
+ if (ext)
+ le_bitvector_to_cpu(c->sb.errors_silent, (void *) ext->errors_silent,
+ sizeof(c->sb.errors_silent) * 8);
+
+ for_each_member_device(c, ca) {
+ struct bch_member m = bch2_sb_member_get(src, ca->dev_idx);
+ ca->mi = bch2_mi_to_cpu(&m);
+ }
}
-/* doesn't copy member info */
-static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
+static int __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
{
struct bch_sb_field *src_f, *dst_f;
struct bch_sb *dst = dst_handle->sb;
dst->time_base_lo = src->time_base_lo;
dst->time_base_hi = src->time_base_hi;
dst->time_precision = src->time_precision;
+ dst->write_time = src->write_time;
memcpy(dst->flags, src->flags, sizeof(dst->flags));
memcpy(dst->features, src->features, sizeof(dst->features));
memcpy(dst->compat, src->compat, sizeof(dst->compat));
for (i = 0; i < BCH_SB_FIELD_NR; i++) {
- if (i == BCH_SB_FIELD_journal)
+ int d;
+
+ if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
continue;
- src_f = bch2_sb_field_get(src, i);
- dst_f = bch2_sb_field_get(dst, i);
+ src_f = bch2_sb_field_get_id(src, i);
+ dst_f = bch2_sb_field_get_id(dst, i);
+
+ d = (src_f ? le32_to_cpu(src_f->u64s) : 0) -
+ (dst_f ? le32_to_cpu(dst_f->u64s) : 0);
+ if (d > 0) {
+ int ret = bch2_sb_realloc(dst_handle,
+ le32_to_cpu(dst_handle->sb->u64s) + d);
+
+ if (ret)
+ return ret;
+
+ dst = dst_handle->sb;
+ dst_f = bch2_sb_field_get_id(dst, i);
+ }
+
dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
src_f ? le32_to_cpu(src_f->u64s) : 0);
if (src_f)
memcpy(dst_f, src_f, vstruct_bytes(src_f));
}
+
+ return 0;
}
int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
{
- struct bch_sb_field_journal *journal_buckets =
- bch2_sb_get_journal(src);
- unsigned journal_u64s = journal_buckets
- ? le32_to_cpu(journal_buckets->field.u64s)
- : 0;
int ret;
lockdep_assert_held(&c->sb_lock);
- ret = bch2_sb_realloc(&c->disk_sb,
- le32_to_cpu(src->u64s) - journal_u64s);
- if (ret)
- return ret;
-
- __copy_super(&c->disk_sb, src);
-
- ret = bch2_sb_replicas_to_cpu_replicas(c);
- if (ret)
- return ret;
-
- ret = bch2_sb_disk_groups_to_cpu(c);
+ ret = bch2_sb_realloc(&c->disk_sb, 0) ?:
+ __copy_super(&c->disk_sb, src) ?:
+ bch2_sb_replicas_to_cpu_replicas(c) ?:
+ bch2_sb_disk_groups_to_cpu(c);
if (ret)
return ret;
int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
{
- struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
- struct bch_sb_field_journal *journal_buckets =
- bch2_sb_get_journal(dst);
- unsigned journal_u64s = journal_buckets
- ? le32_to_cpu(journal_buckets->field.u64s)
- : 0;
- unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
- int ret;
-
- ret = bch2_sb_realloc(&ca->disk_sb, u64s);
- if (ret)
- return ret;
-
- __copy_super(&ca->disk_sb, src);
- return 0;
+ return __copy_super(&ca->disk_sb, c->disk_sb.sb);
}
/* read superblock: */
-static const char *read_one_super(struct bch_sb_handle *sb, u64 offset)
+static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
{
- struct bch_csum csum;
size_t bytes;
+ int ret;
reread:
- bio_reset(sb->bio);
- bio_set_dev(sb->bio, sb->bdev);
+ bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
sb->bio->bi_iter.bi_sector = offset;
- bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
- bch2_bio_map(sb->bio, sb->sb, PAGE_SIZE << sb->page_order);
+ bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
- if (submit_bio_wait(sb->bio))
- return "IO error";
+ ret = submit_bio_wait(sb->bio);
+ if (ret) {
+ prt_printf(err, "IO error: %i", ret);
+ return ret;
+ }
- if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
- return "Not a bcachefs superblock";
+ if (!uuid_equal(&sb->sb->magic, &BCACHE_MAGIC) &&
+ !uuid_equal(&sb->sb->magic, &BCHFS_MAGIC)) {
+ prt_str(err, "Not a bcachefs superblock (got magic ");
+ pr_uuid(err, sb->sb->magic.b);
+ prt_str(err, ")");
+ return -BCH_ERR_invalid_sb_magic;
+ }
- 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";
+ ret = bch2_sb_compatible(sb->sb, err);
+ if (ret)
+ return ret;
bytes = vstruct_bytes(sb->sb);
- if (bytes > 512 << sb->sb->layout.sb_max_size_bits)
- return "Bad superblock: too big";
+ if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
+ prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
+ bytes, 512UL << sb->sb->layout.sb_max_size_bits);
+ return -BCH_ERR_invalid_sb_too_big;
+ }
- if (get_order(bytes) > sb->page_order) {
- if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
- return "cannot allocate memory";
+ if (bytes > sb->buffer_size) {
+ ret = bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s));
+ if (ret)
+ return ret;
goto reread;
}
- if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR)
- return "unknown csum type";
+ enum bch_csum_type csum_type = BCH_SB_CSUM_TYPE(sb->sb);
+ if (csum_type >= BCH_CSUM_NR) {
+ prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
+ return -BCH_ERR_invalid_sb_csum_type;
+ }
/* XXX: verify MACs */
- csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
- null_nonce(), sb->sb);
-
- if (bch2_crc_cmp(csum, sb->sb->csum))
- return "bad checksum reading superblock";
+ struct bch_csum csum = csum_vstruct(NULL, csum_type, null_nonce(), sb->sb);
+ if (bch2_crc_cmp(csum, sb->sb->csum)) {
+ bch2_csum_err_msg(err, csum_type, sb->sb->csum, csum);
+ return -BCH_ERR_invalid_sb_csum;
+ }
sb->seq = le64_to_cpu(sb->sb->seq);
- return NULL;
+ return 0;
}
-int bch2_read_super(const char *path, struct bch_opts *opts,
- struct bch_sb_handle *sb)
+static int __bch2_read_super(const char *path, struct bch_opts *opts,
+ struct bch_sb_handle *sb, bool ignore_notbchfs_msg)
{
u64 offset = opt_get(*opts, sb);
struct bch_sb_layout layout;
- const char *err;
+ struct printbuf err = PRINTBUF;
+ struct printbuf err2 = PRINTBUF;
__le64 *i;
int ret;
-
- pr_verbose_init(*opts, "");
-
+#ifndef __KERNEL__
+retry:
+#endif
memset(sb, 0, sizeof(*sb));
- sb->mode = FMODE_READ;
+ sb->mode = BLK_OPEN_READ;
sb->have_bio = true;
+ sb->holder = kmalloc(1, GFP_KERNEL);
+ if (!sb->holder)
+ return -ENOMEM;
+
+ sb->sb_name = kstrdup(path, GFP_KERNEL);
+ if (!sb->sb_name)
+ return -ENOMEM;
+
+#ifndef __KERNEL__
+ if (opt_get(*opts, direct_io) == false)
+ sb->mode |= BLK_OPEN_BUFFERED;
+#endif
if (!opt_get(*opts, noexcl))
- sb->mode |= FMODE_EXCL;
+ sb->mode |= BLK_OPEN_EXCL;
if (!opt_get(*opts, nochanges))
- sb->mode |= FMODE_WRITE;
+ sb->mode |= BLK_OPEN_WRITE;
- sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
+ sb->bdev = blkdev_get_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
if (IS_ERR(sb->bdev) &&
PTR_ERR(sb->bdev) == -EACCES &&
opt_get(*opts, read_only)) {
- sb->mode &= ~FMODE_WRITE;
+ sb->mode &= ~BLK_OPEN_WRITE;
- sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
+ sb->bdev = blkdev_get_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
if (!IS_ERR(sb->bdev))
opt_set(*opts, nochanges, true);
}
goto out;
}
- err = "cannot allocate memory";
ret = bch2_sb_realloc(sb, 0);
- if (ret)
+ if (ret) {
+ prt_printf(&err, "error allocating memory for superblock");
goto err;
+ }
- ret = -EFAULT;
- err = "dynamic fault";
- if (bch2_fs_init_fault("read_super"))
+ if (bch2_fs_init_fault("read_super")) {
+ prt_printf(&err, "dynamic fault");
+ ret = -EFAULT;
goto err;
+ }
- ret = -EINVAL;
- err = read_one_super(sb, offset);
- if (!err)
+ ret = read_one_super(sb, offset, &err);
+ if (!ret)
goto got_super;
if (opt_defined(*opts, sb))
goto err;
- pr_err("error reading default superblock: %s", err);
+ prt_printf(&err2, "bcachefs (%s): error reading default superblock: %s\n",
+ path, err.buf);
+ if (ret == -BCH_ERR_invalid_sb_magic && ignore_notbchfs_msg)
+ printk(KERN_INFO "%s", err2.buf);
+ else
+ printk(KERN_ERR "%s", err2.buf);
+
+ printbuf_exit(&err2);
+ printbuf_reset(&err);
/*
* Error reading primary superblock - read location of backup
* superblocks:
*/
- bio_reset(sb->bio);
- bio_set_dev(sb->bio, sb->bdev);
+ bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
- 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, sizeof(struct bch_sb_layout));
- err = "IO error";
- if (submit_bio_wait(sb->bio))
+ ret = submit_bio_wait(sb->bio);
+ if (ret) {
+ prt_printf(&err, "IO error: %i", ret);
goto err;
+ }
memcpy(&layout, sb->sb, sizeof(layout));
- err = validate_sb_layout(&layout);
- if (err)
+ ret = validate_sb_layout(&layout, &err);
+ if (ret)
goto err;
for (i = layout.sb_offset;
if (offset == opt_get(*opts, sb))
continue;
- err = read_one_super(sb, offset);
- if (!err)
+ ret = read_one_super(sb, offset, &err);
+ if (!ret)
goto got_super;
}
- ret = -EINVAL;
goto err;
got_super:
- err = "Superblock block size smaller than device block size";
- ret = -EINVAL;
if (le16_to_cpu(sb->sb->block_size) << 9 <
- bdev_logical_block_size(sb->bdev))
+ bdev_logical_block_size(sb->bdev) &&
+ opt_get(*opts, direct_io)) {
+#ifndef __KERNEL__
+ opt_set(*opts, direct_io, false);
+ bch2_free_super(sb);
+ goto retry;
+#endif
+ prt_printf(&err, "block size (%u) smaller than device block size (%u)",
+ le16_to_cpu(sb->sb->block_size) << 9,
+ bdev_logical_block_size(sb->bdev));
+ ret = -BCH_ERR_block_size_too_small;
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;
+
+ ret = bch2_sb_validate(sb, &err, READ);
+ if (ret) {
+ printk(KERN_ERR "bcachefs (%s): error validating superblock: %s\n",
+ path, err.buf);
+ goto err_no_print;
+ }
out:
- pr_verbose_init(*opts, "ret %i", ret);
+ printbuf_exit(&err);
return ret;
err:
+ printk(KERN_ERR "bcachefs (%s): error reading superblock: %s\n",
+ path, err.buf);
+err_no_print:
bch2_free_super(sb);
- pr_err("error reading superblock: %s", err);
goto out;
}
+int bch2_read_super(const char *path, struct bch_opts *opts,
+ struct bch_sb_handle *sb)
+{
+ return __bch2_read_super(path, opts, sb, false);
+}
+
+/* provide a silenced version for mount.bcachefs */
+
+int bch2_read_super_silent(const char *path, struct bch_opts *opts,
+ struct bch_sb_handle *sb)
+{
+ return __bch2_read_super(path, opts, sb, true);
+}
+
/* write superblock: */
static void write_super_endio(struct bio *bio)
/* 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,
+ bio_data_dir(bio)
+ ? BCH_MEMBER_ERROR_write
+ : BCH_MEMBER_ERROR_read,
+ "superblock %s error: %s",
+ bio_data_dir(bio) ? "write" : "read",
+ bch2_blk_status_to_str(bio->bi_status)))
ca->sb_write_error = 1;
closure_put(&ca->fs->sb_write);
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_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
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],
+ this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
bio_sectors(bio));
percpu_ref_get(&ca->io_ref);
sb->offset = sb->layout.sb_offset[idx];
- SET_BCH_SB_CSUM_TYPE(sb, c->opts.metadata_checksum);
+ SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
null_nonce(), sb);
- 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 = le64_to_cpu(sb->offset);
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,
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);
int bch2_write_super(struct bch_fs *c)
{
struct closure *cl = &c->sb_write;
- struct bch_dev *ca;
- unsigned i, sb = 0, nr_wrote;
- const char *err;
+ struct printbuf err = PRINTBUF;
+ unsigned sb = 0, nr_wrote;
struct bch_devs_mask sb_written;
bool wrote, can_mount_without_written, can_mount_with_written;
+ unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
int ret = 0;
+ trace_and_count(c, write_super, c, _RET_IP_);
+
+ if (c->opts.very_degraded)
+ degraded_flags |= BCH_FORCE_IF_LOST;
+
lockdep_assert_held(&c->sb_lock);
closure_init_stack(cl);
memset(&sb_written, 0, sizeof(sb_written));
+ /* Make sure we're using the new magic numbers: */
+ c->disk_sb.sb->magic = BCHFS_MAGIC;
+ c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
+
le64_add_cpu(&c->disk_sb.sb->seq, 1);
- if (test_bit(BCH_FS_ERROR, &c->flags))
+ struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
+ for_each_online_member(c, ca)
+ __bch2_members_v2_get_mut(mi, ca->dev_idx)->seq = c->disk_sb.sb->seq;
+ c->disk_sb.sb->write_time = cpu_to_le64(ktime_get_real_seconds());
+
+ if (test_bit(BCH_FS_error, &c->flags))
SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
+ if (test_bit(BCH_FS_topology_error, &c->flags))
+ SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
+
+ SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
+
+ bch2_sb_counters_from_cpu(c);
+ bch2_sb_members_from_cpu(c);
+ bch2_sb_members_cpy_v2_v1(&c->disk_sb);
+ bch2_sb_errors_from_cpu(c);
+ bch2_sb_downgrade_update(c);
- for_each_online_member(ca, c, i)
+ for_each_online_member(c, ca)
bch2_sb_from_fs(c, ca);
- for_each_online_member(ca, c, i) {
- err = bch2_sb_validate(&ca->disk_sb);
- if (err) {
- bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
- ret = -1;
+ for_each_online_member(c, ca) {
+ printbuf_reset(&err);
+
+ ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
+ if (ret) {
+ bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
+ percpu_ref_put(&ca->io_ref);
goto out;
}
}
if (c->opts.nochanges)
goto out;
- for_each_online_member(ca, c, i) {
+ /*
+ * Defer writing the superblock until filesystem initialization is
+ * complete - don't write out a partly initialized superblock:
+ */
+ if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
+ goto out;
+
+ if (le16_to_cpu(c->disk_sb.sb->version) > bcachefs_metadata_version_current) {
+ struct printbuf buf = PRINTBUF;
+ prt_printf(&buf, "attempting to write superblock that wasn't version downgraded (");
+ bch2_version_to_text(&buf, le16_to_cpu(c->disk_sb.sb->version));
+ prt_str(&buf, " > ");
+ bch2_version_to_text(&buf, bcachefs_metadata_version_current);
+ prt_str(&buf, ")");
+ bch2_fs_fatal_error(c, "%s", buf.buf);
+ printbuf_exit(&buf);
+ return -BCH_ERR_sb_not_downgraded;
+ }
+
+ for_each_online_member(c, ca) {
__set_bit(ca->dev_idx, sb_written.d);
ca->sb_write_error = 0;
}
- for_each_online_member(ca, c, i)
+ for_each_online_member(c, ca)
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)) {
+ for_each_online_member(c, ca) {
+ if (ca->sb_write_error)
+ continue;
+
+ if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
bch2_fs_fatal_error(c,
- "Superblock modified by another process");
+ "Superblock write was silently dropped! (seq %llu expected %llu)",
+ le64_to_cpu(ca->sb_read_scratch->seq),
+ ca->disk_sb.seq);
percpu_ref_put(&ca->io_ref);
- ret = -EROFS;
+ ret = -BCH_ERR_erofs_sb_err;
+ goto out;
+ }
+
+ if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
+ bch2_fs_fatal_error(c,
+ "Superblock modified by another process (seq %llu expected %llu)",
+ le64_to_cpu(ca->sb_read_scratch->seq),
+ ca->disk_sb.seq);
+ percpu_ref_put(&ca->io_ref);
+ ret = -BCH_ERR_erofs_sb_err;
goto out;
}
}
do {
wrote = false;
- for_each_online_member(ca, c, i)
+ for_each_online_member(c, ca)
if (!ca->sb_write_error &&
sb < ca->disk_sb.sb->layout.nr_superblocks) {
write_one_super(c, ca, sb);
sb++;
} while (wrote);
- for_each_online_member(ca, c, i) {
+ for_each_online_member(c, ca) {
if (ca->sb_write_error)
__clear_bit(ca->dev_idx, sb_written.d);
else
nr_wrote = dev_mask_nr(&sb_written);
can_mount_with_written =
- bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
- BCH_FORCE_IF_DEGRADED);
+ bch2_have_enough_devs(c, sb_written, degraded_flags, false);
- for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
+ for (unsigned i = 0; i < ARRAY_SIZE(sb_written.d); i++)
sb_written.d[i] = ~sb_written.d[i];
can_mount_without_written =
- bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
- BCH_FORCE_IF_DEGRADED);
+ bch2_have_enough_devs(c, sb_written, degraded_flags, false);
/*
* If we would be able to mount _without_ the devices we successfully
* mount with the devices we did successfully write to:
*/
if (bch2_fs_fatal_err_on(!nr_wrote ||
+ !can_mount_with_written ||
(can_mount_without_written &&
!can_mount_with_written), c,
- "Unable to write superblock to sufficient devices"))
+ "Unable to write superblock to sufficient devices (from %ps)",
+ (void *) _RET_IP_))
ret = -1;
out:
/* Make new options visible after they're persistent: */
bch2_sb_update(c);
+ printbuf_exit(&err);
return ret;
}
mutex_unlock(&c->sb_lock);
}
-/* BCH_SB_FIELD_journal: */
-
-static int u64_cmp(const void *_l, const void *_r)
+/* Downgrade if superblock is at a higher version than currently supported: */
+bool bch2_check_version_downgrade(struct bch_fs *c)
{
- u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
+ bool ret = bcachefs_metadata_version_current < c->sb.version;
- return l < r ? -1 : l > r ? 1 : 0;
+ lockdep_assert_held(&c->sb_lock);
+
+ /*
+ * Downgrade, if superblock is at a higher version than currently
+ * supported:
+ *
+ * c->sb will be checked before we write the superblock, so update it as
+ * well:
+ */
+ if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) > bcachefs_metadata_version_current) {
+ SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
+ c->sb.version_upgrade_complete = bcachefs_metadata_version_current;
+ }
+ if (c->sb.version > bcachefs_metadata_version_current) {
+ c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
+ c->sb.version = bcachefs_metadata_version_current;
+ }
+ if (c->sb.version_min > bcachefs_metadata_version_current) {
+ c->disk_sb.sb->version_min = cpu_to_le16(bcachefs_metadata_version_current);
+ c->sb.version_min = bcachefs_metadata_version_current;
+ }
+ c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
+ return ret;
}
-static const char *bch2_sb_validate_journal(struct bch_sb *sb,
- struct bch_sb_field *f)
+void bch2_sb_upgrade(struct bch_fs *c, unsigned new_version)
{
- struct bch_sb_field_journal *journal = field_to_type(f, journal);
- struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
- const char *err;
- unsigned nr;
- unsigned i;
- u64 *b;
-
- journal = bch2_sb_get_journal(sb);
- if (!journal)
- return NULL;
-
- nr = bch2_nr_journal_buckets(journal);
- if (!nr)
- return NULL;
-
- b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
- if (!b)
- return "cannot allocate memory";
-
- for (i = 0; i < nr; i++)
- b[i] = le64_to_cpu(journal->buckets[i]);
-
- sort(b, nr, sizeof(u64), u64_cmp, NULL);
-
- err = "journal bucket at sector 0";
- if (!b[0])
- goto err;
-
- err = "journal bucket before first bucket";
- if (m && b[0] < le16_to_cpu(m->first_bucket))
- goto err;
-
- err = "journal bucket past end of device";
- if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
- goto err;
+ lockdep_assert_held(&c->sb_lock);
- err = "duplicate journal buckets";
- for (i = 0; i + 1 < nr; i++)
- if (b[i] == b[i + 1])
- goto err;
+ if (BCH_VERSION_MAJOR(new_version) >
+ BCH_VERSION_MAJOR(le16_to_cpu(c->disk_sb.sb->version)))
+ bch2_sb_field_resize(&c->disk_sb, downgrade, 0);
- err = NULL;
-err:
- kfree(b);
- return err;
+ c->disk_sb.sb->version = cpu_to_le16(new_version);
+ c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
}
-static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
- .validate = bch2_sb_validate_journal,
-};
-
-/* BCH_SB_FIELD_members: */
-
-static const char *bch2_sb_validate_members(struct bch_sb *sb,
- struct bch_sb_field *f)
+static int bch2_sb_ext_validate(struct bch_sb *sb, struct bch_sb_field *f,
+ struct printbuf *err)
{
- struct bch_sb_field_members *mi = field_to_type(f, members);
- struct bch_member *m;
+ if (vstruct_bytes(f) < 88) {
+ prt_printf(err, "field too small (%zu < %u)", vstruct_bytes(f), 88);
+ return -BCH_ERR_invalid_sb_ext;
+ }
- if ((void *) (mi->members + sb->nr_devices) >
- vstruct_end(&mi->field))
- return "Invalid superblock: bad member info";
+ return 0;
+}
- for (m = mi->members;
- m < mi->members + sb->nr_devices;
- m++) {
- if (!bch2_member_exists(m))
- continue;
+static void bch2_sb_ext_to_text(struct printbuf *out, struct bch_sb *sb,
+ struct bch_sb_field *f)
+{
+ struct bch_sb_field_ext *e = field_to_type(f, ext);
- if (le64_to_cpu(m->nbuckets) > LONG_MAX)
- return "Too many buckets";
+ prt_printf(out, "Recovery passes required:");
+ prt_tab(out);
+ prt_bitflags(out, bch2_recovery_passes,
+ bch2_recovery_passes_from_stable(le64_to_cpu(e->recovery_passes_required[0])));
+ prt_newline(out);
- if (le64_to_cpu(m->nbuckets) -
- le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
- return "Not enough buckets";
+ unsigned long *errors_silent = kmalloc(sizeof(e->errors_silent), GFP_KERNEL);
+ if (errors_silent) {
+ le_bitvector_to_cpu(errors_silent, (void *) e->errors_silent, sizeof(e->errors_silent) * 8);
- if (le16_to_cpu(m->bucket_size) <
- le16_to_cpu(sb->block_size))
- return "bucket size smaller than block size";
+ prt_printf(out, "Errors to silently fix:");
+ prt_tab(out);
+ prt_bitflags_vector(out, bch2_sb_error_strs, errors_silent, sizeof(e->errors_silent) * 8);
+ prt_newline(out);
- if (le16_to_cpu(m->bucket_size) <
- BCH_SB_BTREE_NODE_SIZE(sb))
- return "bucket size smaller than btree node size";
+ kfree(errors_silent);
}
-
- return NULL;
}
-static const struct bch_sb_field_ops bch_sb_field_ops_members = {
- .validate = bch2_sb_validate_members,
+static const struct bch_sb_field_ops bch_sb_field_ops_ext = {
+ .validate = bch2_sb_ext_validate,
+ .to_text = bch2_sb_ext_to_text,
};
-/* BCH_SB_FIELD_crypt: */
-
-static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
- struct bch_sb_field *f)
-{
- struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
-
- if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
- return "invalid field crypt: wrong size";
-
- if (BCH_CRYPT_KDF_TYPE(crypt))
- return "invalid field crypt: bad kdf type";
-
- return NULL;
-}
-
-static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
- .validate = bch2_sb_validate_crypt,
+static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
+#define x(f, nr) \
+ [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
+ BCH_SB_FIELDS()
+#undef x
};
-/* BCH_SB_FIELD_clean: */
+static const struct bch_sb_field_ops bch2_sb_field_null_ops;
-void bch2_sb_clean_renumber(struct bch_sb_field_clean *clean, int write)
+static const struct bch_sb_field_ops *bch2_sb_field_type_ops(unsigned type)
{
- 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);
+ return likely(type < ARRAY_SIZE(bch2_sb_field_ops))
+ ? bch2_sb_field_ops[type]
+ : &bch2_sb_field_null_ops;
}
-int bch2_fs_mark_dirty(struct bch_fs *c)
+static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
+ struct printbuf *err)
{
+ unsigned type = le32_to_cpu(f->type);
+ struct printbuf field_err = PRINTBUF;
+ const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
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->compat[0] &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA);
- c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_extents_above_btree_updates;
- ret = bch2_write_super(c);
- mutex_unlock(&c->sb_lock);
+ ret = ops->validate ? ops->validate(sb, f, &field_err) : 0;
+ if (ret) {
+ prt_printf(err, "Invalid superblock section %s: %s",
+ bch2_sb_fields[type], field_err.buf);
+ prt_newline(err);
+ bch2_sb_field_to_text(err, sb, f);
+ }
+ printbuf_exit(&field_err);
return ret;
}
-static void
-entry_init_u64s(struct jset_entry *entry, unsigned u64s)
+void __bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
+ struct bch_sb_field *f)
{
- memset(entry, 0, u64s * sizeof(u64));
+ unsigned type = le32_to_cpu(f->type);
+ const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
- /*
- * The u64s field counts from the start of data, ignoring the shared
- * fields.
- */
- entry->u64s = u64s - 1;
-}
+ if (!out->nr_tabstops)
+ printbuf_tabstop_push(out, 32);
-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);
+ if (ops->to_text)
+ ops->to_text(out, sb, f);
}
-struct jset_entry *
-bch2_journal_super_entries_add_common(struct bch_fs *c,
- struct jset_entry *entry,
- u64 journal_seq)
+void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
+ struct bch_sb_field *f)
{
- struct btree_root *r;
- unsigned i;
-
- mutex_lock(&c->btree_root_lock);
-
- for (r = c->btree_roots;
- r < c->btree_roots + BTREE_ID_NR;
- r++)
- if (r->alive) {
- entry_init_u64s(entry, r->key.u64s + 1);
- entry->btree_id = r - c->btree_roots;
- entry->level = r->level;
- entry->type = BCH_JSET_ENTRY_btree_root;
- bkey_copy(&entry->start[0], &r->key);
-
- entry = vstruct_next(entry);
- }
- c->btree_roots_dirty = false;
-
- mutex_unlock(&c->btree_root_lock);
-
- percpu_down_write(&c->mark_lock);
-
- if (!journal_seq) {
- bch2_fs_usage_acc_to_base(c, 0);
- bch2_fs_usage_acc_to_base(c, 1);
- } else {
- bch2_fs_usage_acc_to_base(c, journal_seq & 1);
- }
-
- {
- 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));
+ unsigned type = le32_to_cpu(f->type);
- entry = vstruct_next(entry);
- }
+ if (type < BCH_SB_FIELD_NR)
+ prt_printf(out, "%s", bch2_sb_fields[type]);
+ else
+ prt_printf(out, "(unknown field %u)", type);
- percpu_up_write(&c->mark_lock);
+ prt_printf(out, " (size %zu):", vstruct_bytes(f));
+ prt_newline(out);
- return entry;
+ __bch2_sb_field_to_text(out, sb, f);
}
-void bch2_fs_mark_clean(struct bch_fs *c)
+void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
{
- 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;
+ unsigned i;
- SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
+ prt_printf(out, "Type: %u", l->layout_type);
+ prt_newline(out);
- 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);
+ prt_str(out, "Superblock max size: ");
+ prt_units_u64(out, 512 << l->sb_max_size_bits);
+ prt_newline(out);
- u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
+ prt_printf(out, "Nr superblocks: %u", l->nr_superblocks);
+ prt_newline(out);
- 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;
+ prt_str(out, "Offsets: ");
+ for (i = 0; i < l->nr_superblocks; i++) {
+ if (i)
+ prt_str(out, ", ");
+ prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
}
-
- 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);
- 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);
+ prt_newline(out);
}
-static const char *bch2_sb_validate_clean(struct bch_sb *sb,
- struct bch_sb_field *f)
+void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
+ bool print_layout, unsigned fields)
{
- 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,
- BCH_SB_FIELDS()
-#undef x
-};
+ u64 fields_have = 0;
+ unsigned nr_devices = 0;
+
+ if (!out->nr_tabstops)
+ printbuf_tabstop_push(out, 44);
+
+ for (int i = 0; i < sb->nr_devices; i++)
+ nr_devices += bch2_dev_exists(sb, i);
+
+ prt_printf(out, "External UUID:");
+ prt_tab(out);
+ pr_uuid(out, sb->user_uuid.b);
+ prt_newline(out);
+
+ prt_printf(out, "Internal UUID:");
+ prt_tab(out);
+ pr_uuid(out, sb->uuid.b);
+ prt_newline(out);
+
+ prt_printf(out, "Magic number:");
+ prt_tab(out);
+ pr_uuid(out, sb->magic.b);
+ prt_newline(out);
+
+ prt_str(out, "Device index:");
+ prt_tab(out);
+ prt_printf(out, "%u", sb->dev_idx);
+ prt_newline(out);
+
+ prt_str(out, "Label:");
+ prt_tab(out);
+ prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
+ prt_newline(out);
+
+ prt_str(out, "Version:");
+ prt_tab(out);
+ bch2_version_to_text(out, le16_to_cpu(sb->version));
+ prt_newline(out);
+
+ prt_str(out, "Version upgrade complete:");
+ prt_tab(out);
+ bch2_version_to_text(out, BCH_SB_VERSION_UPGRADE_COMPLETE(sb));
+ prt_newline(out);
+
+ prt_printf(out, "Oldest version on disk:");
+ prt_tab(out);
+ bch2_version_to_text(out, le16_to_cpu(sb->version_min));
+ prt_newline(out);
+
+ prt_printf(out, "Created:");
+ prt_tab(out);
+ if (sb->time_base_lo)
+ bch2_prt_datetime(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
+ else
+ prt_printf(out, "(not set)");
+ prt_newline(out);
+
+ prt_printf(out, "Sequence number:");
+ prt_tab(out);
+ prt_printf(out, "%llu", le64_to_cpu(sb->seq));
+ prt_newline(out);
+
+ prt_printf(out, "Time of last write:");
+ prt_tab(out);
+ bch2_prt_datetime(out, le64_to_cpu(sb->write_time));
+ prt_newline(out);
+
+ prt_printf(out, "Superblock size:");
+ prt_tab(out);
+ prt_printf(out, "%zu", vstruct_bytes(sb));
+ prt_newline(out);
+
+ prt_printf(out, "Clean:");
+ prt_tab(out);
+ prt_printf(out, "%llu", BCH_SB_CLEAN(sb));
+ prt_newline(out);
+
+ prt_printf(out, "Devices:");
+ prt_tab(out);
+ prt_printf(out, "%u", nr_devices);
+ prt_newline(out);
+
+ prt_printf(out, "Sections:");
+ vstruct_for_each(sb, f)
+ fields_have |= 1 << le32_to_cpu(f->type);
+ prt_tab(out);
+ prt_bitflags(out, bch2_sb_fields, fields_have);
+ prt_newline(out);
+
+ prt_printf(out, "Features:");
+ prt_tab(out);
+ prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
+ prt_newline(out);
+
+ prt_printf(out, "Compat features:");
+ prt_tab(out);
+ prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
+ prt_newline(out);
+
+ prt_newline(out);
+ prt_printf(out, "Options:");
+ prt_newline(out);
+ printbuf_indent_add(out, 2);
+ {
+ enum bch_opt_id id;
-static const char *bch2_sb_field_validate(struct bch_sb *sb,
- struct bch_sb_field *f)
-{
- unsigned type = le32_to_cpu(f->type);
+ for (id = 0; id < bch2_opts_nr; id++) {
+ const struct bch_option *opt = bch2_opt_table + id;
- return type < BCH_SB_FIELD_NR
- ? bch2_sb_field_ops[type]->validate(sb, f)
- : NULL;
-}
+ if (opt->get_sb != BCH2_NO_SB_OPT) {
+ u64 v = bch2_opt_from_sb(sb, id);
-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);
- const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
- ? bch2_sb_field_ops[type] : NULL;
+ prt_printf(out, "%s:", opt->attr.name);
+ prt_tab(out);
+ bch2_opt_to_text(out, NULL, sb, opt, v,
+ OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
+ prt_newline(out);
+ }
+ }
+ }
- if (ops)
- pr_buf(out, "%s", bch2_sb_fields[type]);
- else
- pr_buf(out, "(unknown field %u)", type);
+ printbuf_indent_sub(out, 2);
- pr_buf(out, " (size %llu):", vstruct_bytes(f));
+ if (print_layout) {
+ prt_newline(out);
+ prt_printf(out, "layout:");
+ prt_newline(out);
+ printbuf_indent_add(out, 2);
+ bch2_sb_layout_to_text(out, &sb->layout);
+ printbuf_indent_sub(out, 2);
+ }
- if (ops && ops->to_text)
- bch2_sb_field_ops[type]->to_text(out, sb, f);
+ vstruct_for_each(sb, f)
+ if (fields & (1 << le32_to_cpu(f->type))) {
+ prt_newline(out);
+ bch2_sb_field_to_text(out, sb, f);
+ }
}