+// 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_io.h"
+#include "journal_sb.h"
+#include "journal_seq_blacklist.h"
+#include "replicas.h"
+#include "quota.h"
#include "super-io.h"
#include "super.h"
#include "vstructs.h"
+#include "counters.h"
#include <linux/backing-dev.h>
#include <linux/sort.h>
-static int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *);
-static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *,
- struct bch_replicas_cpu *);
-static int bch2_sb_disk_groups_to_cpu(struct bch_fs *);
-
-/* superblock fields (optional/variable size sections: */
+#include <trace/events/bcachefs.h>
const char * const bch2_sb_fields[] = {
#define x(name, nr) #name,
NULL
};
-#define x(f, nr) \
-static const char *bch2_sb_validate_##f(struct bch_sb *, struct bch_sb_field *);
- BCH_SB_FIELDS()
-#undef x
-
-struct bch_sb_field_ops {
- const char * (*validate)(struct bch_sb *, struct bch_sb_field *);
-};
-
-static const struct bch_sb_field_ops bch2_sb_field_ops[] = {
-#define x(f, nr) \
- [BCH_SB_FIELD_##f] = { \
- .validate = bch2_sb_validate_##f, \
- },
- BCH_SB_FIELDS()
-#undef x
-};
-
-static const char *bch2_sb_field_validate(struct bch_sb *sb,
- struct bch_sb_field *f)
-
-{
- unsigned type = le32_to_cpu(f->type);
-
- return type < BCH_SB_FIELD_NR
- ? bch2_sb_field_ops[type].validate(sb, f)
- : NULL;
-}
+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,
enum bch_sb_field_type type)
return NULL;
}
-static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb *sb,
- struct bch_sb_field *f,
- unsigned u64s)
+static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
+ struct bch_sb_field *f,
+ unsigned u64s)
{
unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
+ unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
+
+ BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
- if (!f) {
- f = vstruct_last(sb);
+ 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);
f->type = 0;
void *src, *dst;
src = vstruct_end(f);
- f->u64s = cpu_to_le32(u64s);
- dst = vstruct_end(f);
- memmove(dst, src, vstruct_end(sb) - src);
+ if (u64s) {
+ f->u64s = cpu_to_le32(u64s);
+ dst = vstruct_end(f);
+ } else {
+ dst = f;
+ }
+
+ memmove(dst, src, vstruct_end(sb->sb) - src);
if (dst > src)
memset(src, 0, dst - src);
}
- le32_add_cpu(&sb->u64s, u64s - old_u64s);
+ 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: */
if (!IS_ERR_OR_NULL(sb->bdev))
blkdev_put(sb->bdev, sb->mode);
- free_pages((unsigned long) sb->sb, sb->page_order);
+ kfree(sb->sb);
memset(sb, 0, sizeof(*sb));
}
-static int __bch2_super_realloc(struct bch_sb_handle *sb, unsigned order)
+int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
{
+ size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
+ size_t new_buffer_size;
struct bch_sb *new_sb;
struct bio *bio;
- if (sb->page_order >= order && sb->sb)
- return 0;
-
- if (dynamic_fault("bcachefs:add:super_realloc"))
- return -ENOMEM;
-
- bio = bio_kmalloc(GFP_KERNEL, 1 << order);
- if (!bio)
- return -ENOMEM;
+ if (sb->bdev)
+ new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
- if (sb->bio)
- bio_put(sb->bio);
- sb->bio = bio;
+ new_buffer_size = roundup_pow_of_two(new_bytes);
- new_sb = (void *) __get_free_pages(GFP_KERNEL, order);
- if (!new_sb)
- return -ENOMEM;
+ if (sb->sb && sb->buffer_size >= new_buffer_size)
+ return 0;
- if (sb->sb)
- memcpy(new_sb, sb->sb, PAGE_SIZE << sb->page_order);
+ if (sb->have_layout) {
+ u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
- free_pages((unsigned long) sb->sb, sb->page_order);
- sb->sb = new_sb;
+ if (new_bytes > max_bytes) {
+ char buf[BDEVNAME_SIZE];
- sb->page_order = order;
+ pr_err("%s: superblock too big: want %zu but have %llu",
+ bdevname(sb->bdev, buf), new_bytes, max_bytes);
+ return -ENOSPC;
+ }
+ }
- return 0;
-}
+ if (sb->buffer_size >= new_buffer_size && sb->sb)
+ return 0;
-static int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
-{
- u64 new_bytes = __vstruct_bytes(struct bch_sb, u64s);
- u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
+ if (dynamic_fault("bcachefs:add:super_realloc"))
+ return -ENOMEM;
- if (new_bytes > max_bytes) {
- char buf[BDEVNAME_SIZE];
+ if (sb->have_bio) {
+ bio = bio_kmalloc(GFP_KERNEL,
+ DIV_ROUND_UP(new_buffer_size, PAGE_SIZE));
+ if (!bio)
+ return -ENOMEM;
- pr_err("%s: superblock too big: want %llu but have %llu",
- bdevname(sb->bdev, buf), new_bytes, max_bytes);
- return -ENOSPC;
+ if (sb->bio)
+ bio_put(sb->bio);
+ sb->bio = bio;
}
- return __bch2_super_realloc(sb, get_order(new_bytes));
-}
-
-static int bch2_fs_sb_realloc(struct bch_fs *c, unsigned u64s)
-{
- u64 bytes = __vstruct_bytes(struct bch_sb, u64s);
- struct bch_sb *sb;
- unsigned order = get_order(bytes);
-
- if (c->disk_sb && order <= c->disk_sb_order)
- return 0;
-
- sb = (void *) __get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
- if (!sb)
+ new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
+ if (!new_sb)
return -ENOMEM;
- if (c->disk_sb)
- memcpy(sb, c->disk_sb, PAGE_SIZE << c->disk_sb_order);
-
- free_pages((unsigned long) c->disk_sb, c->disk_sb_order);
+ sb->sb = new_sb;
+ sb->buffer_size = new_buffer_size;
- c->disk_sb = sb;
- c->disk_sb_order = order;
return 0;
}
if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
return NULL;
- f = __bch2_sb_field_resize(sb->sb, f, u64s);
- f->type = cpu_to_le32(type);
- return f;
-}
-
-struct bch_sb_field *bch2_fs_sb_field_resize(struct bch_fs *c,
- enum bch_sb_field_type type,
- unsigned u64s)
-{
- struct bch_sb_field *f = bch2_sb_field_get(c->disk_sb, type);
- ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
- ssize_t d = -old_u64s + u64s;
- struct bch_dev *ca;
- unsigned i;
-
- lockdep_assert_held(&c->sb_lock);
+ if (sb->fs_sb) {
+ struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
+ struct bch_dev *ca;
+ unsigned i;
- if (bch2_fs_sb_realloc(c, le32_to_cpu(c->disk_sb->u64s) + d))
- return NULL;
+ lockdep_assert_held(&c->sb_lock);
- /* XXX: we're not checking that offline device have enough space */
+ /* 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(ca, c, i) {
+ struct bch_sb_handle *sb = &ca->disk_sb;
- if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
- percpu_ref_put(&ca->ref);
- return NULL;
+ if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
+ percpu_ref_put(&ca->ref);
+ return NULL;
+ }
}
}
- f = __bch2_sb_field_resize(c->disk_sb, f, u64s);
- f->type = cpu_to_le32(type);
+ f = bch2_sb_field_get(sb->sb, type);
+ f = __bch2_sb_field_resize(sb, f, u64s);
+ if (f)
+ f->type = cpu_to_le32(type);
return f;
}
BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
}
-static const char *validate_sb_layout(struct bch_sb_layout *layout)
+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";
+ if (uuid_le_cmp(layout->magic, BCACHE_MAGIC)) {
+ pr_buf(out, "Not a bcachefs superblock layout");
+ return -EINVAL;
+ }
- if (layout->layout_type != 0)
- return "Invalid superblock layout type";
+ if (layout->layout_type != 0) {
+ pr_buf(out, "Invalid superblock layout type %u",
+ layout->layout_type);
+ return -EINVAL;
+ }
- if (!layout->nr_superblocks)
- return "Invalid superblock layout: no superblocks";
+ if (!layout->nr_superblocks) {
+ pr_buf(out, "Invalid superblock layout: no superblocks");
+ return -EINVAL;
+ }
- 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)) {
+ pr_buf(out, "Invalid superblock layout: too many superblocks");
+ return -EINVAL;
+ }
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) {
+ pr_buf(out, "Invalid superblock layout: superblocks overlap\n"
+ " (sb %u ends at %llu next starts at %llu",
+ i - 1, prev_offset + max_sectors, offset);
+ return -EINVAL;
+ }
prev_offset = offset;
}
- return NULL;
+ return 0;
}
-const char *bch2_sb_validate(struct bch_sb_handle *disk_sb)
+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 *f;
struct bch_sb_field_members *mi;
- const char *err;
+ enum bch_opt_id opt_id;
+ u32 version, version_min;
u16 block_size;
+ int ret;
- 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_bkey_renumber
+ ? 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) {
+ pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
+ version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
+ return -EINVAL;
}
- block_size = le16_to_cpu(sb->block_size);
-
- if (!is_power_of_2(block_size) ||
- block_size > PAGE_SECTORS)
- return "Bad block size";
+ if (version_min < bcachefs_metadata_version_min) {
+ pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
+ version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
+ return -EINVAL;
+ }
- if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le)))
- return "Bad user UUID";
+ if (version_min > version) {
+ pr_buf(out, "Bad minimum version %u, greater than version field %u",
+ version_min, version);
+ return -EINVAL;
+ }
- if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le)))
- return "Bad internal UUID";
+ if (sb->features[1] ||
+ (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
+ pr_buf(out, "Filesystem has incompatible features");
+ return -EINVAL;
+ }
- if (!sb->nr_devices ||
- sb->nr_devices <= sb->dev_idx ||
- sb->nr_devices > BCH_SB_MEMBERS_MAX)
- return "Bad number of member devices";
+ block_size = le16_to_cpu(sb->block_size);
- if (!BCH_SB_META_REPLICAS_WANT(sb) ||
- BCH_SB_META_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
- return "Invalid number of metadata replicas";
+ if (block_size > PAGE_SECTORS) {
+ pr_buf(out, "Block size too big (got %u, max %u)",
+ block_size, PAGE_SECTORS);
+ return -EINVAL;
+ }
- if (!BCH_SB_META_REPLICAS_REQ(sb) ||
- BCH_SB_META_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
- return "Invalid number of metadata replicas";
+ if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le))) {
+ pr_buf(out, "Bad user UUID (got zeroes)");
+ return -EINVAL;
+ }
- if (!BCH_SB_DATA_REPLICAS_WANT(sb) ||
- BCH_SB_DATA_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
- return "Invalid number of data replicas";
+ if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le))) {
+ pr_buf(out, "Bad intenal UUID (got zeroes)");
+ return -EINVAL;
+ }
- if (!BCH_SB_DATA_REPLICAS_REQ(sb) ||
- BCH_SB_DATA_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
- return "Invalid number of data replicas";
+ if (!sb->nr_devices ||
+ sb->nr_devices > BCH_SB_MEMBERS_MAX) {
+ pr_buf(out, "Bad number of member devices %u (max %u)",
+ sb->nr_devices, BCH_SB_MEMBERS_MAX);
+ return -EINVAL;
+ }
- if (BCH_SB_META_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
- return "Invalid metadata checksum type";
+ if (sb->dev_idx >= sb->nr_devices) {
+ pr_buf(out, "Bad dev_idx (got %u, nr_devices %u)",
+ sb->dev_idx, sb->nr_devices);
+ return -EINVAL;
+ }
- 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) {
+ pr_buf(out, "Invalid time precision: %u (min 1, max %lu)",
+ le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
+ return -EINVAL;
+ }
- 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'r 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_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";
+ pr_buf(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) {
+ pr_buf(out, "Invalid superblock: optional with size 0 (type %u)",
+ le32_to_cpu(f->type));
+ return -EINVAL;
+ }
- if (vstruct_next(f) > vstruct_last(sb))
- return "Invalid superblock: invalid optional field";
+ if (vstruct_next(f) > vstruct_last(sb)) {
+ pr_buf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
+ le32_to_cpu(f->type));
+ return -EINVAL;
+ }
}
/* members must be validated first: */
mi = bch2_sb_get_members(sb);
- if (!mi)
- return "Invalid superblock: member info area missing";
+ if (!mi) {
+ pr_buf(out, "Invalid superblock: member info area missing");
+ return -EINVAL;
+ }
- 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)
continue;
- err = bch2_sb_field_validate(sb, f);
- if (err)
- return err;
+ ret = bch2_sb_field_validate(sb, f, out);
+ if (ret)
+ return ret;
}
- 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;
+ return 0;
}
/* device open: */
static void bch2_sb_update(struct bch_fs *c)
{
- struct bch_sb *src = c->disk_sb;
+ 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;
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.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);
}
-/* doesn't copy member info */
-static void __copy_super(struct bch_sb *dst, struct bch_sb *src)
+static void __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;
+ 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 ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
continue;
- dst_f = bch2_sb_field_get(dst, le32_to_cpu(src_f->type));
- dst_f = __bch2_sb_field_resize(dst, dst_f,
- le32_to_cpu(src_f->u64s));
+ 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,
+ 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));
}
}
lockdep_assert_held(&c->sb_lock);
- ret = bch2_fs_sb_realloc(c, le32_to_cpu(src->u64s) - journal_u64s);
+ ret = bch2_sb_realloc(&c->disk_sb,
+ le32_to_cpu(src->u64s) - journal_u64s);
if (ret)
return ret;
- __copy_super(c->disk_sb, src);
+ __copy_super(&c->disk_sb, src);
ret = bch2_sb_replicas_to_cpu_replicas(c);
if (ret)
int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
{
- struct bch_sb *src = c->disk_sb, *dst = ca->disk_sb.sb;
+ 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
if (ret)
return ret;
- __copy_super(dst, src);
+ __copy_super(&ca->disk_sb, src);
return 0;
}
/* 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;
+ u32 version, version_min;
size_t bytes;
- unsigned order;
+ int ret;
reread:
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, sb->buffer_size);
+
+ ret = submit_bio_wait(sb->bio);
+ if (ret) {
+ pr_buf(err, "IO error: %i", ret);
+ return ret;
+ }
+
+ if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC)) {
+ pr_buf(err, "Not a bcachefs superblock");
+ return -EINVAL;
+ }
- if (submit_bio_wait(sb->bio))
- return "IO error";
+ version = le16_to_cpu(sb->sb->version);
+ version_min = version >= bcachefs_metadata_version_bkey_renumber
+ ? le16_to_cpu(sb->sb->version_min)
+ : version;
- if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
- return "Not a bcachefs superblock";
+ if (version >= bcachefs_metadata_version_max) {
+ pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
+ version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
+ return -EINVAL;
+ }
- 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 (version_min < bcachefs_metadata_version_min) {
+ pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
+ version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
+ return -EINVAL;
+ }
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) {
+ pr_buf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
+ bytes, 512UL << sb->sb->layout.sb_max_size_bits);
+ return -EINVAL;
+ }
- order = get_order(bytes);
- if (order > sb->page_order) {
- if (__bch2_super_realloc(sb, order))
- return "cannot allocate memory";
+ if (bytes > sb->buffer_size) {
+ if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
+ return -ENOMEM;
goto reread;
}
- if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR)
- return "unknown csum type";
+ if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
+ pr_buf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
+ return -EINVAL;
+ }
/* 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";
+ if (bch2_crc_cmp(csum, sb->sb->csum)) {
+ pr_buf(err, "bad checksum");
+ return -EINVAL;
+ }
+
+ sb->seq = le64_to_cpu(sb->sb->seq);
- return NULL;
+ return 0;
}
int bch2_read_super(const char *path, struct bch_opts *opts,
{
u64 offset = opt_get(*opts, sb);
struct bch_sb_layout layout;
- const char *err;
+ struct printbuf err = PRINTBUF;
__le64 *i;
int ret;
pr_verbose_init(*opts, "");
memset(sb, 0, sizeof(*sb));
- sb->mode = FMODE_READ;
+ sb->mode = FMODE_READ;
+ sb->have_bio = true;
if (!opt_get(*opts, noexcl))
sb->mode |= FMODE_EXCL;
goto out;
}
- err = "cannot allocate memory";
- ret = __bch2_super_realloc(sb, 0);
- if (ret)
+ ret = bch2_sb_realloc(sb, 0);
+ if (ret) {
+ pr_buf(&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")) {
+ pr_buf(&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);
+ printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
+ path, err.buf);
+ printbuf_reset(&err);
/*
* Error reading primary superblock - read location of backup
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))
+ ret = submit_bio_wait(sb->bio);
+ if (ret) {
+ pr_buf(&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)) {
+ pr_buf(&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 = -EINVAL;
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",
+ 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",
+ path, err.buf);
+err_no_print:
bch2_free_super(sb);
- pr_err("error reading superblock: %s", err);
goto 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;
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->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;
+ struct printbuf err = PRINTBUF;
unsigned i, sb = 0, nr_wrote;
- const char *err;
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_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));
- le64_add_cpu(&c->disk_sb->seq, 1);
+ 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);
+ 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);
for_each_online_member(ca, c, i)
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);
+ 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 ||
- test_bit(BCH_FS_ERROR, &c->flags))
+ if (c->opts.nochanges)
+ goto out;
+
+ /*
+ * 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;
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)
+ continue;
+
+ if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
+ bch2_fs_fatal_error(c,
+ "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;
+ 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 = -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);
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++)
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
* 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_with_written ||
+ (can_mount_without_written &&
+ !can_mount_with_written), c,
+ "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;
}
-/* BCH_SB_FIELD_journal: */
-
-static int u64_cmp(const void *_l, const void *_r)
+void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
{
- u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
+ mutex_lock(&c->sb_lock);
+ if (!(c->sb.features & (1ULL << feat))) {
+ c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
- return l < r ? -1 : l > r ? 1 : 0;
+ bch2_write_super(c);
+ }
+ mutex_unlock(&c->sb_lock);
}
-static const char *bch2_sb_validate_journal(struct bch_sb *sb,
- struct bch_sb_field *f)
+/* BCH_SB_FIELD_members: */
+
+static int bch2_sb_members_validate(struct bch_sb *sb,
+ struct bch_sb_field *f,
+ struct printbuf *err)
{
- 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;
+ struct bch_sb_field_members *mi = field_to_type(f, members);
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";
+ if ((void *) (mi->members + sb->nr_devices) >
+ vstruct_end(&mi->field)) {
+ pr_buf(err, "too many devices for section size");
+ return -EINVAL;
+ }
- for (i = 0; i < nr; i++)
- b[i] = le64_to_cpu(journal->buckets[i]);
+ for (i = 0; i < sb->nr_devices; i++) {
+ struct bch_member *m = mi->members + i;
- sort(b, nr, sizeof(u64), u64_cmp, NULL);
+ if (!bch2_member_exists(m))
+ continue;
- err = "journal bucket at sector 0";
- if (!b[0])
- goto err;
+ if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
+ pr_buf(err, "device %u: too many buckets (got %llu, max %lu)",
+ i, le64_to_cpu(m->nbuckets), LONG_MAX);
+ return -EINVAL;
+ }
- err = "journal bucket before first bucket";
- if (m && b[0] < le16_to_cpu(m->first_bucket))
- goto err;
+ if (le64_to_cpu(m->nbuckets) -
+ le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
+ pr_buf(err, "device %u: not enough buckets (got %llu, max %u)",
+ i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
+ return -EINVAL;
+ }
- err = "journal bucket past end of device";
- if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
- goto err;
+ if (le16_to_cpu(m->bucket_size) <
+ le16_to_cpu(sb->block_size)) {
+ pr_buf(err, "device %u: bucket size %u smaller than block size %u",
+ i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
+ return -EINVAL;
+ }
- err = "duplicate journal buckets";
- for (i = 0; i + 1 < nr; i++)
- if (b[i] == b[i + 1])
- goto err;
+ if (le16_to_cpu(m->bucket_size) <
+ BCH_SB_BTREE_NODE_SIZE(sb)) {
+ pr_buf(err, "device %u: bucket size %u smaller than btree node size %llu",
+ i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
+ return -EINVAL;
+ }
+ }
- err = NULL;
-err:
- kfree(b);
- return err;
+ return 0;
}
-/* BCH_SB_FIELD_members: */
-
-static const char *bch2_sb_validate_members(struct bch_sb *sb,
- struct bch_sb_field *f)
+static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
+ struct bch_sb_field *f)
{
struct bch_sb_field_members *mi = field_to_type(f, members);
- struct bch_member *m;
+ struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
+ unsigned i;
- if ((void *) (mi->members + sb->nr_devices) >
- vstruct_end(&mi->field))
- return "Invalid superblock: bad member info";
+ for (i = 0; i < sb->nr_devices; i++) {
+ struct bch_member *m = mi->members + i;
+ unsigned data_have = bch2_sb_dev_has_data(sb, i);
+ u64 bucket_size = le16_to_cpu(m->bucket_size);
+ u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
- for (m = mi->members;
- m < mi->members + sb->nr_devices;
- m++) {
if (!bch2_member_exists(m))
continue;
- if (le64_to_cpu(m->nbuckets) > LONG_MAX)
- return "Too many buckets";
-
- if (le64_to_cpu(m->nbuckets) -
- le16_to_cpu(m->first_bucket) < 1 << 10)
- return "Not enough buckets";
-
- if (le16_to_cpu(m->bucket_size) <
- le16_to_cpu(sb->block_size))
- return "bucket size smaller than block size";
-
- if (le16_to_cpu(m->bucket_size) <
- BCH_SB_BTREE_NODE_SIZE(sb))
- return "bucket size smaller than btree node size";
+ pr_buf(out, "Device:");
+ pr_tab(out);
+ pr_buf(out, "%u", i);
+ pr_newline(out);
+
+ pr_indent_push(out, 2);
+
+ pr_buf(out, "UUID:");
+ pr_tab(out);
+ pr_uuid(out, m->uuid.b);
+ pr_newline(out);
+
+ pr_buf(out, "Size:");
+ pr_tab(out);
+ pr_units(out, device_size, device_size << 9);
+ pr_newline(out);
+
+ pr_buf(out, "Bucket size:");
+ pr_tab(out);
+ pr_units(out, bucket_size, bucket_size << 9);
+ pr_newline(out);
+
+ pr_buf(out, "First bucket:");
+ pr_tab(out);
+ pr_buf(out, "%u", le16_to_cpu(m->first_bucket));
+ pr_newline(out);
+
+ pr_buf(out, "Buckets:");
+ pr_tab(out);
+ pr_buf(out, "%llu", le64_to_cpu(m->nbuckets));
+ pr_newline(out);
+
+ pr_buf(out, "Last mount:");
+ pr_tab(out);
+ if (m->last_mount)
+ pr_time(out, le64_to_cpu(m->last_mount));
+ else
+ pr_buf(out, "(never)");
+ pr_newline(out);
+
+ pr_buf(out, "State:");
+ pr_tab(out);
+ pr_buf(out, "%s",
+ BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
+ ? bch2_member_states[BCH_MEMBER_STATE(m)]
+ : "unknown");
+ pr_newline(out);
+
+ pr_buf(out, "Group:");
+ pr_tab(out);
+ if (BCH_MEMBER_GROUP(m)) {
+ unsigned idx = BCH_MEMBER_GROUP(m) - 1;
+
+ if (idx < disk_groups_nr(gi))
+ pr_buf(out, "%s (%u)",
+ gi->entries[idx].label, idx);
+ else
+ pr_buf(out, "(bad disk labels section)");
+ } else {
+ pr_buf(out, "(none)");
+ }
+ pr_newline(out);
+
+ pr_buf(out, "Data allowed:");
+ pr_tab(out);
+ if (BCH_MEMBER_DATA_ALLOWED(m))
+ bch2_flags_to_text(out, bch2_data_types,
+ BCH_MEMBER_DATA_ALLOWED(m));
+ else
+ pr_buf(out, "(none)");
+ pr_newline(out);
+
+ pr_buf(out, "Has data:");
+ pr_tab(out);
+ if (data_have)
+ bch2_flags_to_text(out, bch2_data_types, data_have);
+ else
+ pr_buf(out, "(none)");
+ pr_newline(out);
+
+ pr_buf(out, "Discard:");
+ pr_tab(out);
+ pr_buf(out, "%llu", BCH_MEMBER_DISCARD(m));
+ pr_newline(out);
+
+ pr_buf(out, "Freespace initialized:");
+ pr_tab(out);
+ pr_buf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
+ pr_newline(out);
+
+ pr_indent_pop(out, 2);
}
-
- 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;
}
+static const struct bch_sb_field_ops bch_sb_field_ops_members = {
+ .validate = bch2_sb_members_validate,
+ .to_text = bch2_sb_members_to_text,
+};
+
/* BCH_SB_FIELD_crypt: */
-static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
- struct bch_sb_field *f)
+static int bch2_sb_crypt_validate(struct bch_sb *sb,
+ struct bch_sb_field *f,
+ struct printbuf *err)
{
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 (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
+ pr_buf(err, "wrong size (got %zu should be %zu)",
+ vstruct_bytes(&crypt->field), sizeof(*crypt));
+ return -EINVAL;
+ }
- if (BCH_CRYPT_KDF_TYPE(crypt))
- return "invalid field crypt: bad kdf type";
+ if (BCH_CRYPT_KDF_TYPE(crypt)) {
+ pr_buf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
+ return -EINVAL;
+ }
- return NULL;
+ return 0;
}
-/* BCH_SB_FIELD_replicas: */
+static void bch2_sb_crypt_to_text(struct printbuf *out, struct bch_sb *sb,
+ struct bch_sb_field *f)
+{
+ struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
-/* Replicas tracking - in memory: */
+ pr_buf(out, "KFD: %llu", BCH_CRYPT_KDF_TYPE(crypt));
+ pr_newline(out);
+ pr_buf(out, "scrypt n: %llu", BCH_KDF_SCRYPT_N(crypt));
+ pr_newline(out);
+ pr_buf(out, "scrypt r: %llu", BCH_KDF_SCRYPT_R(crypt));
+ pr_newline(out);
+ pr_buf(out, "scrypt p: %llu", BCH_KDF_SCRYPT_P(crypt));
+ pr_newline(out);
+}
-#define for_each_cpu_replicas_entry(_r, _i) \
- for (_i = (_r)->entries; \
- (void *) (_i) < (void *) (_r)->entries + (_r)->nr * (_r)->entry_size;\
- _i = (void *) (_i) + (_r)->entry_size)
+static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
+ .validate = bch2_sb_crypt_validate,
+ .to_text = bch2_sb_crypt_to_text,
+};
-static inline struct bch_replicas_cpu_entry *
-cpu_replicas_entry(struct bch_replicas_cpu *r, unsigned i)
-{
- return (void *) r->entries + r->entry_size * i;
-}
+/* BCH_SB_FIELD_clean: */
-static void bch2_cpu_replicas_sort(struct bch_replicas_cpu *r)
+int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
{
- eytzinger0_sort(r->entries, r->nr, r->entry_size, memcmp, NULL);
-}
+ struct jset_entry *entry;
+ int ret;
-static inline bool replicas_test_dev(struct bch_replicas_cpu_entry *e,
- unsigned dev)
-{
- return (e->devs[dev >> 3] & (1 << (dev & 7))) != 0;
-}
+ for (entry = clean->start;
+ entry < (struct jset_entry *) vstruct_end(&clean->field);
+ entry = vstruct_next(entry)) {
+ ret = bch2_journal_entry_validate(c, "superblock", entry,
+ le16_to_cpu(c->disk_sb.sb->version),
+ BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
+ write);
+ if (ret)
+ return ret;
+ }
-static inline void replicas_set_dev(struct bch_replicas_cpu_entry *e,
- unsigned dev)
-{
- e->devs[dev >> 3] |= 1 << (dev & 7);
+ return 0;
}
-static inline unsigned replicas_dev_slots(struct bch_replicas_cpu *r)
+int bch2_fs_mark_dirty(struct bch_fs *c)
{
- return (r->entry_size -
- offsetof(struct bch_replicas_cpu_entry, devs)) * 8;
-}
+ int ret;
-int bch2_cpu_replicas_to_text(struct bch_replicas_cpu *r,
- char *buf, size_t size)
-{
- char *out = buf, *end = out + size;
- struct bch_replicas_cpu_entry *e;
- bool first = true;
- unsigned i;
-
- for_each_cpu_replicas_entry(r, e) {
- bool first_e = true;
-
- if (!first)
- out += scnprintf(out, end - out, " ");
- first = false;
-
- out += scnprintf(out, end - out, "%u: [", e->data_type);
-
- for (i = 0; i < replicas_dev_slots(r); i++)
- if (replicas_test_dev(e, i)) {
- if (!first_e)
- out += scnprintf(out, end - out, " ");
- first_e = false;
- out += scnprintf(out, end - out, "%u", i);
- }
- out += scnprintf(out, end - out, "]");
- }
-
- return out - buf;
-}
-
-static inline unsigned bkey_to_replicas(struct bkey_s_c_extent e,
- enum bch_data_type data_type,
- struct bch_replicas_cpu_entry *r,
- unsigned *max_dev)
-{
- const struct bch_extent_ptr *ptr;
- unsigned nr = 0;
-
- BUG_ON(!data_type ||
- data_type == BCH_DATA_SB ||
- data_type >= BCH_DATA_NR);
-
- memset(r, 0, sizeof(*r));
- r->data_type = data_type;
-
- *max_dev = 0;
-
- extent_for_each_ptr(e, ptr)
- if (!ptr->cached) {
- *max_dev = max_t(unsigned, *max_dev, ptr->dev);
- replicas_set_dev(r, ptr->dev);
- nr++;
- }
- return nr;
-}
-
-static inline void devlist_to_replicas(struct bch_devs_list devs,
- enum bch_data_type data_type,
- struct bch_replicas_cpu_entry *r,
- unsigned *max_dev)
-{
- unsigned i;
-
- BUG_ON(!data_type ||
- data_type == BCH_DATA_SB ||
- data_type >= BCH_DATA_NR);
-
- memset(r, 0, sizeof(*r));
- r->data_type = data_type;
-
- *max_dev = 0;
-
- for (i = 0; i < devs.nr; i++) {
- *max_dev = max_t(unsigned, *max_dev, devs.devs[i]);
- replicas_set_dev(r, devs.devs[i]);
- }
-}
-
-static struct bch_replicas_cpu *
-cpu_replicas_add_entry(struct bch_replicas_cpu *old,
- struct bch_replicas_cpu_entry new_entry,
- unsigned max_dev)
-{
- struct bch_replicas_cpu *new;
- unsigned i, nr, entry_size;
-
- entry_size = offsetof(struct bch_replicas_cpu_entry, devs) +
- DIV_ROUND_UP(max_dev + 1, 8);
- entry_size = max(entry_size, old->entry_size);
- nr = old->nr + 1;
-
- new = kzalloc(sizeof(struct bch_replicas_cpu) +
- nr * entry_size, GFP_NOIO);
- if (!new)
- return NULL;
-
- new->nr = nr;
- new->entry_size = entry_size;
-
- for (i = 0; i < old->nr; i++)
- memcpy(cpu_replicas_entry(new, i),
- cpu_replicas_entry(old, i),
- min(new->entry_size, old->entry_size));
-
- memcpy(cpu_replicas_entry(new, old->nr),
- &new_entry,
- new->entry_size);
-
- bch2_cpu_replicas_sort(new);
- return new;
-}
-
-static bool replicas_has_entry(struct bch_replicas_cpu *r,
- struct bch_replicas_cpu_entry search,
- unsigned max_dev)
-{
- return max_dev < replicas_dev_slots(r) &&
- eytzinger0_find(r->entries, r->nr,
- r->entry_size,
- memcmp, &search) < r->nr;
-}
-
-noinline
-static int bch2_mark_replicas_slowpath(struct bch_fs *c,
- struct bch_replicas_cpu_entry new_entry,
- unsigned max_dev)
-{
- struct bch_replicas_cpu *old_gc, *new_gc = NULL, *old_r, *new_r = NULL;
- int ret = -ENOMEM;
+ /*
+ * Unconditionally write superblock, to verify it hasn't changed before
+ * we go rw:
+ */
mutex_lock(&c->sb_lock);
-
- old_gc = rcu_dereference_protected(c->replicas_gc,
- lockdep_is_held(&c->sb_lock));
- if (old_gc && !replicas_has_entry(old_gc, new_entry, max_dev)) {
- new_gc = cpu_replicas_add_entry(old_gc, new_entry, max_dev);
- if (!new_gc)
- goto err;
- }
-
- old_r = rcu_dereference_protected(c->replicas,
- lockdep_is_held(&c->sb_lock));
- if (!replicas_has_entry(old_r, new_entry, max_dev)) {
- new_r = cpu_replicas_add_entry(old_r, new_entry, max_dev);
- if (!new_r)
- goto err;
-
- ret = bch2_cpu_replicas_to_sb_replicas(c, new_r);
- if (ret)
- goto err;
- }
-
- /* allocations done, now commit: */
-
- if (new_r)
- bch2_write_super(c);
-
- /* don't update in memory replicas until changes are persistent */
-
- if (new_gc) {
- rcu_assign_pointer(c->replicas_gc, new_gc);
- kfree_rcu(old_gc, rcu);
- }
-
- if (new_r) {
- rcu_assign_pointer(c->replicas, new_r);
- kfree_rcu(old_r, rcu);
- }
-
- mutex_unlock(&c->sb_lock);
- return 0;
-err:
+ SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
+ c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
+ c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
+ ret = bch2_write_super(c);
mutex_unlock(&c->sb_lock);
- if (new_gc)
- kfree(new_gc);
- if (new_r)
- kfree(new_r);
- return ret;
-}
-
-int bch2_mark_replicas(struct bch_fs *c,
- enum bch_data_type data_type,
- struct bch_devs_list devs)
-{
- struct bch_replicas_cpu_entry search;
- struct bch_replicas_cpu *r, *gc_r;
- unsigned max_dev;
- bool marked;
-
- if (!devs.nr)
- return 0;
-
- BUG_ON(devs.nr >= BCH_REPLICAS_MAX);
-
- devlist_to_replicas(devs, data_type, &search, &max_dev);
-
- rcu_read_lock();
- r = rcu_dereference(c->replicas);
- gc_r = rcu_dereference(c->replicas_gc);
- marked = replicas_has_entry(r, search, max_dev) &&
- (!likely(gc_r) || replicas_has_entry(gc_r, search, max_dev));
- rcu_read_unlock();
-
- return likely(marked) ? 0
- : bch2_mark_replicas_slowpath(c, search, max_dev);
-}
-
-int bch2_mark_bkey_replicas(struct bch_fs *c,
- enum bch_data_type data_type,
- struct bkey_s_c k)
-{
- struct bch_devs_list cached = bch2_bkey_cached_devs(k);
- unsigned i;
- int ret;
-
- for (i = 0; i < cached.nr; i++)
- if ((ret = bch2_mark_replicas(c, BCH_DATA_CACHED,
- bch2_dev_list_single(cached.devs[i]))))
- return ret;
-
- return bch2_mark_replicas(c, data_type, bch2_bkey_dirty_devs(k));
-}
-
-int bch2_replicas_gc_end(struct bch_fs *c, int err)
-{
- struct bch_replicas_cpu *new_r, *old_r;
- int ret = 0;
-
- lockdep_assert_held(&c->replicas_gc_lock);
-
- mutex_lock(&c->sb_lock);
-
- new_r = rcu_dereference_protected(c->replicas_gc,
- lockdep_is_held(&c->sb_lock));
-
- if (err) {
- rcu_assign_pointer(c->replicas_gc, NULL);
- kfree_rcu(new_r, rcu);
- goto err;
- }
-
- if (bch2_cpu_replicas_to_sb_replicas(c, new_r)) {
- ret = -ENOSPC;
- goto err;
- }
-
- old_r = rcu_dereference_protected(c->replicas,
- lockdep_is_held(&c->sb_lock));
- rcu_assign_pointer(c->replicas, new_r);
- rcu_assign_pointer(c->replicas_gc, NULL);
- kfree_rcu(old_r, rcu);
-
- bch2_write_super(c);
-err:
- mutex_unlock(&c->sb_lock);
return ret;
}
-int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
-{
- struct bch_replicas_cpu *dst, *src;
- struct bch_replicas_cpu_entry *e;
-
- lockdep_assert_held(&c->replicas_gc_lock);
-
- mutex_lock(&c->sb_lock);
- BUG_ON(c->replicas_gc);
-
- src = rcu_dereference_protected(c->replicas,
- lockdep_is_held(&c->sb_lock));
-
- dst = kzalloc(sizeof(struct bch_replicas_cpu) +
- src->nr * src->entry_size, GFP_NOIO);
- if (!dst) {
- mutex_unlock(&c->sb_lock);
- return -ENOMEM;
- }
-
- dst->nr = 0;
- dst->entry_size = src->entry_size;
-
- for_each_cpu_replicas_entry(src, e)
- if (!((1 << e->data_type) & typemask))
- memcpy(cpu_replicas_entry(dst, dst->nr++),
- e, dst->entry_size);
-
- bch2_cpu_replicas_sort(dst);
-
- rcu_assign_pointer(c->replicas_gc, dst);
- mutex_unlock(&c->sb_lock);
-
- return 0;
-}
-
-/* Replicas tracking - superblock: */
-
-static void bch2_sb_replicas_nr_entries(struct bch_sb_field_replicas *r,
- unsigned *nr,
- unsigned *bytes,
- unsigned *max_dev)
+static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
{
- struct bch_replicas_entry *i;
- unsigned j;
-
- *nr = 0;
- *bytes = sizeof(*r);
- *max_dev = 0;
+ struct jset_entry *entry = *end;
+ unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
- if (!r)
- return;
-
- for_each_replicas_entry(r, i) {
- for (j = 0; j < i->nr; j++)
- *max_dev = max_t(unsigned, *max_dev, i->devs[j]);
- (*nr)++;
- }
+ memset(entry, 0, u64s * sizeof(u64));
+ /*
+ * The u64s field counts from the start of data, ignoring the shared
+ * fields.
+ */
+ entry->u64s = cpu_to_le16(u64s - 1);
- *bytes = (void *) i - (void *) r;
+ *end = vstruct_next(*end);
+ return entry;
}
-static struct bch_replicas_cpu *
-__bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r)
+void bch2_journal_super_entries_add_common(struct bch_fs *c,
+ struct jset_entry **end,
+ u64 journal_seq)
{
- struct bch_replicas_cpu *cpu_r;
- unsigned i, nr, bytes, max_dev, entry_size;
-
- bch2_sb_replicas_nr_entries(sb_r, &nr, &bytes, &max_dev);
-
- entry_size = offsetof(struct bch_replicas_cpu_entry, devs) +
- DIV_ROUND_UP(max_dev + 1, 8);
-
- cpu_r = kzalloc(sizeof(struct bch_replicas_cpu) +
- nr * entry_size, GFP_NOIO);
- if (!cpu_r)
- return NULL;
-
- cpu_r->nr = nr;
- cpu_r->entry_size = entry_size;
-
- if (nr) {
- struct bch_replicas_cpu_entry *dst =
- cpu_replicas_entry(cpu_r, 0);
- struct bch_replicas_entry *src = sb_r->entries;
+ struct bch_dev *ca;
+ unsigned i, dev;
- while (dst < cpu_replicas_entry(cpu_r, nr)) {
- dst->data_type = src->data_type;
- for (i = 0; i < src->nr; i++)
- replicas_set_dev(dst, src->devs[i]);
+ percpu_down_read(&c->mark_lock);
- src = replicas_entry_next(src);
- dst = (void *) dst + entry_size;
- }
+ 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);
}
- bch2_cpu_replicas_sort(cpu_r);
- return cpu_r;
-}
-
-static int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
-{
- struct bch_sb_field_replicas *sb_r;
- struct bch_replicas_cpu *cpu_r, *old_r;
-
- sb_r = bch2_sb_get_replicas(c->disk_sb);
- cpu_r = __bch2_sb_replicas_to_cpu_replicas(sb_r);
- if (!cpu_r)
- return -ENOMEM;
-
- old_r = rcu_dereference_check(c->replicas, lockdep_is_held(&c->sb_lock));
- rcu_assign_pointer(c->replicas, cpu_r);
- if (old_r)
- kfree_rcu(old_r, rcu);
-
- return 0;
-}
-
-static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
- struct bch_replicas_cpu *r)
-{
- struct bch_sb_field_replicas *sb_r;
- struct bch_replicas_entry *sb_e;
- struct bch_replicas_cpu_entry *e;
- size_t i, bytes;
-
- bytes = sizeof(struct bch_sb_field_replicas);
+ {
+ struct jset_entry_usage *u =
+ container_of(jset_entry_init(end, sizeof(*u)),
+ struct jset_entry_usage, entry);
- for_each_cpu_replicas_entry(r, e) {
- bytes += sizeof(struct bch_replicas_entry);
- for (i = 0; i < r->entry_size - 1; i++)
- bytes += hweight8(e->devs[i]);
+ u->entry.type = BCH_JSET_ENTRY_usage;
+ u->entry.btree_id = BCH_FS_USAGE_inodes;
+ u->v = cpu_to_le64(c->usage_base->nr_inodes);
}
- sb_r = bch2_fs_sb_resize_replicas(c,
- DIV_ROUND_UP(sizeof(*sb_r) + bytes, sizeof(u64)));
- if (!sb_r)
- return -ENOSPC;
-
- memset(&sb_r->entries, 0,
- vstruct_end(&sb_r->field) -
- (void *) &sb_r->entries);
-
- sb_e = sb_r->entries;
- for_each_cpu_replicas_entry(r, e) {
- sb_e->data_type = e->data_type;
+ {
+ struct jset_entry_usage *u =
+ container_of(jset_entry_init(end, sizeof(*u)),
+ struct jset_entry_usage, entry);
- for (i = 0; i < replicas_dev_slots(r); i++)
- if (replicas_test_dev(e, i))
- sb_e->devs[sb_e->nr++] = i;
-
- sb_e = replicas_entry_next(sb_e);
-
- BUG_ON((void *) sb_e > vstruct_end(&sb_r->field));
+ u->entry.type = BCH_JSET_ENTRY_usage;
+ u->entry.btree_id = BCH_FS_USAGE_key_version;
+ u->v = cpu_to_le64(atomic64_read(&c->key_version));
}
- return 0;
-}
-
-static const char *bch2_sb_validate_replicas(struct bch_sb *sb,
- struct bch_sb_field *f)
-{
- struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
- struct bch_sb_field_members *mi = bch2_sb_get_members(sb);
- struct bch_replicas_cpu *cpu_r = NULL;
- struct bch_replicas_entry *e;
- const char *err;
- unsigned i;
-
- for_each_replicas_entry(sb_r, e) {
- err = "invalid replicas entry: invalid data type";
- if (e->data_type >= BCH_DATA_NR)
- goto err;
-
- err = "invalid replicas entry: no devices";
- if (!e->nr)
- goto err;
-
- err = "invalid replicas entry: too many devices";
- if (e->nr >= BCH_REPLICAS_MAX)
- goto err;
+ for (i = 0; i < BCH_REPLICAS_MAX; i++) {
+ struct jset_entry_usage *u =
+ container_of(jset_entry_init(end, sizeof(*u)),
+ struct jset_entry_usage, entry);
- err = "invalid replicas entry: invalid device";
- for (i = 0; i < e->nr; i++)
- if (!bch2_dev_exists(sb, mi, e->devs[i]))
- goto err;
+ u->entry.type = BCH_JSET_ENTRY_usage;
+ u->entry.btree_id = BCH_FS_USAGE_reserved;
+ u->entry.level = i;
+ u->v = cpu_to_le64(c->usage_base->persistent_reserved[i]);
}
- err = "cannot allocate memory";
- cpu_r = __bch2_sb_replicas_to_cpu_replicas(sb_r);
- if (!cpu_r)
- goto err;
-
- sort_cmp_size(cpu_r->entries,
- cpu_r->nr,
- cpu_r->entry_size,
- memcmp, NULL);
-
- for (i = 0; i + 1 < cpu_r->nr; i++) {
- struct bch_replicas_cpu_entry *l =
- cpu_replicas_entry(cpu_r, i);
- struct bch_replicas_cpu_entry *r =
- cpu_replicas_entry(cpu_r, i + 1);
+ 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(jset_entry_init(end, sizeof(*u) + e->nr_devs),
+ struct jset_entry_data_usage, entry);
- BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
-
- err = "duplicate replicas entry";
- if (!memcmp(l, r, cpu_r->entry_size))
- goto err;
+ 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));
}
- err = NULL;
-err:
- kfree(cpu_r);
- return err;
-}
-
-int bch2_sb_replicas_to_text(struct bch_sb_field_replicas *r, char *buf, size_t size)
-{
- char *out = buf, *end = out + size;
- struct bch_replicas_entry *e;
- bool first = true;
- unsigned i;
-
- if (!r) {
- out += scnprintf(out, end - out, "(no replicas section found)");
- return out - buf;
+ for_each_member_device(ca, c, dev) {
+ unsigned b = sizeof(struct jset_entry_dev_usage) +
+ sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
+ struct jset_entry_dev_usage *u =
+ container_of(jset_entry_init(end, b),
+ struct jset_entry_dev_usage, entry);
+
+ u->entry.type = BCH_JSET_ENTRY_dev_usage;
+ u->dev = cpu_to_le32(dev);
+ u->buckets_ec = cpu_to_le64(ca->usage_base->buckets_ec);
+
+ for (i = 0; i < BCH_DATA_NR; i++) {
+ u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
+ u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
+ u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
+ }
}
- for_each_replicas_entry(r, e) {
- if (!first)
- out += scnprintf(out, end - out, " ");
- first = false;
+ percpu_up_read(&c->mark_lock);
- out += scnprintf(out, end - out, "%u: [", e->data_type);
+ for (i = 0; i < 2; i++) {
+ struct jset_entry_clock *clock =
+ container_of(jset_entry_init(end, sizeof(*clock)),
+ struct jset_entry_clock, entry);
- for (i = 0; i < e->nr; i++)
- out += scnprintf(out, end - out,
- i ? " %u" : "%u", e->devs[i]);
- out += scnprintf(out, end - out, "]");
+ clock->entry.type = BCH_JSET_ENTRY_clock;
+ clock->rw = i;
+ clock->time = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
}
-
- return out - buf;
-}
-
-/* Query replicas: */
-
-bool bch2_replicas_marked(struct bch_fs *c,
- enum bch_data_type data_type,
- struct bch_devs_list devs)
-{
- struct bch_replicas_cpu_entry search;
- unsigned max_dev;
- bool ret;
-
- if (!devs.nr)
- return true;
-
- devlist_to_replicas(devs, data_type, &search, &max_dev);
-
- rcu_read_lock();
- ret = replicas_has_entry(rcu_dereference(c->replicas),
- search, max_dev);
- rcu_read_unlock();
-
- return ret;
-}
-
-bool bch2_bkey_replicas_marked(struct bch_fs *c,
- enum bch_data_type data_type,
- struct bkey_s_c k)
-{
- struct bch_devs_list cached = bch2_bkey_cached_devs(k);
- unsigned i;
-
- for (i = 0; i < cached.nr; i++)
- if (!bch2_replicas_marked(c, BCH_DATA_CACHED,
- bch2_dev_list_single(cached.devs[i])))
- return false;
-
- return bch2_replicas_marked(c, data_type, bch2_bkey_dirty_devs(k));
}
-struct replicas_status __bch2_replicas_status(struct bch_fs *c,
- struct bch_devs_mask online_devs)
+void bch2_fs_mark_clean(struct bch_fs *c)
{
- struct bch_sb_field_members *mi;
- struct bch_replicas_cpu_entry *e;
- struct bch_replicas_cpu *r;
- unsigned i, dev, dev_slots, nr_online, nr_offline;
- struct replicas_status ret;
-
- memset(&ret, 0, sizeof(ret));
-
- for (i = 0; i < ARRAY_SIZE(ret.replicas); i++)
- ret.replicas[i].nr_online = UINT_MAX;
-
- mi = bch2_sb_get_members(c->disk_sb);
- rcu_read_lock();
-
- r = rcu_dereference(c->replicas);
- dev_slots = replicas_dev_slots(r);
-
- for_each_cpu_replicas_entry(r, e) {
- if (e->data_type >= ARRAY_SIZE(ret.replicas))
- panic("e %p data_type %u\n", e, e->data_type);
-
- nr_online = nr_offline = 0;
+ struct bch_sb_field_clean *sb_clean;
+ struct jset_entry *entry;
+ unsigned u64s;
+ int ret;
- for (dev = 0; dev < dev_slots; dev++) {
- if (!replicas_test_dev(e, dev))
- continue;
+ mutex_lock(&c->sb_lock);
+ if (BCH_SB_CLEAN(c->disk_sb.sb))
+ goto out;
- BUG_ON(!bch2_dev_exists(c->disk_sb, mi, dev));
+ SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
- if (test_bit(dev, online_devs.d))
- nr_online++;
- else
- nr_offline++;
- }
+ c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
+ c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
+ c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
+ c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
- ret.replicas[e->data_type].nr_online =
- min(ret.replicas[e->data_type].nr_online,
- nr_online);
+ u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
- ret.replicas[e->data_type].nr_offline =
- max(ret.replicas[e->data_type].nr_offline,
- nr_offline);
+ 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;
}
- rcu_read_unlock();
-
- return ret;
-}
-
-struct replicas_status bch2_replicas_status(struct bch_fs *c)
-{
- return __bch2_replicas_status(c, bch2_online_devs(c));
-}
-
-static bool have_enough_devs(struct replicas_status s,
- enum bch_data_type type,
- bool force_if_degraded,
- bool force_if_lost)
-{
- return (!s.replicas[type].nr_offline || force_if_degraded) &&
- (s.replicas[type].nr_online || force_if_lost);
-}
-
-bool bch2_have_enough_devs(struct replicas_status s, unsigned flags)
-{
- return (have_enough_devs(s, BCH_DATA_JOURNAL,
- flags & BCH_FORCE_IF_METADATA_DEGRADED,
- flags & BCH_FORCE_IF_METADATA_LOST) &&
- have_enough_devs(s, BCH_DATA_BTREE,
- flags & BCH_FORCE_IF_METADATA_DEGRADED,
- flags & BCH_FORCE_IF_METADATA_LOST) &&
- have_enough_devs(s, BCH_DATA_USER,
- flags & BCH_FORCE_IF_DATA_DEGRADED,
- flags & BCH_FORCE_IF_DATA_LOST));
-}
+ sb_clean->flags = 0;
+ sb_clean->journal_seq = cpu_to_le64(atomic64_read(&c->journal.seq));
-unsigned bch2_replicas_online(struct bch_fs *c, bool meta)
-{
- struct replicas_status s = bch2_replicas_status(c);
-
- return meta
- ? min(s.replicas[BCH_DATA_JOURNAL].nr_online,
- s.replicas[BCH_DATA_BTREE].nr_online)
- : s.replicas[BCH_DATA_USER].nr_online;
-}
+ /* Trying to catch outstanding bug: */
+ BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
-unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
-{
- struct bch_replicas_cpu_entry *e;
- struct bch_replicas_cpu *r;
- unsigned ret = 0;
+ entry = sb_clean->start;
+ 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));
- rcu_read_lock();
- r = rcu_dereference(c->replicas);
+ memset(entry, 0,
+ vstruct_end(&sb_clean->field) - (void *) entry);
- if (ca->dev_idx >= replicas_dev_slots(r))
+ /*
+ * this should be in the write path, and we should be validating every
+ * superblock section:
+ */
+ ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
+ if (ret) {
+ bch_err(c, "error writing marking filesystem clean: validate error");
goto out;
+ }
- for_each_cpu_replicas_entry(r, e)
- if (replicas_test_dev(e, ca->dev_idx))
- ret |= 1 << e->data_type;
+ bch2_write_super(c);
out:
- rcu_read_unlock();
-
- return ret;
+ mutex_unlock(&c->sb_lock);
}
-/* Quotas: */
-
-static const char *bch2_sb_validate_quota(struct bch_sb *sb,
- struct bch_sb_field *f)
+static int bch2_sb_clean_validate(struct bch_sb *sb,
+ struct bch_sb_field *f,
+ struct printbuf *err)
{
- struct bch_sb_field_quota *q = field_to_type(f, quota);
-
- if (vstruct_bytes(&q->field) != sizeof(*q))
- return "invalid field quota: wrong size";
+ struct bch_sb_field_clean *clean = field_to_type(f, clean);
- return NULL;
-}
-
-/* Disk groups: */
+ if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
+ pr_buf(err, "wrong size (got %zu should be %zu)",
+ vstruct_bytes(&clean->field), sizeof(*clean));
+ return -EINVAL;
+ }
-static int strcmp_void(const void *l, const void *r)
-{
- return strcmp(l, r);
+ return 0;
}
-static const char *bch2_sb_validate_disk_groups(struct bch_sb *sb,
- struct bch_sb_field *f)
+static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
+ struct bch_sb_field *f)
{
- struct bch_sb_field_disk_groups *groups =
- field_to_type(f, disk_groups);
- struct bch_disk_group *g;
- struct bch_sb_field_members *mi;
- struct bch_member *m;
- unsigned i, nr_groups, nr_live = 0, len;
- char **labels, *l;
- const char *err = NULL;
-
- mi = bch2_sb_get_members(sb);
- groups = bch2_sb_get_disk_groups(sb);
- nr_groups = disk_groups_nr(groups);
-
- for (m = mi->members;
- m < mi->members + sb->nr_devices;
- m++) {
- unsigned g;
-
- if (!BCH_MEMBER_GROUP(m))
+ struct bch_sb_field_clean *clean = field_to_type(f, clean);
+ struct jset_entry *entry;
+
+ pr_buf(out, "flags: %x", le32_to_cpu(clean->flags));
+ pr_newline(out);
+ pr_buf(out, "journal_seq: %llu", le64_to_cpu(clean->journal_seq));
+ pr_newline(out);
+
+ for (entry = clean->start;
+ entry != vstruct_end(&clean->field);
+ entry = vstruct_next(entry)) {
+ if (entry->type == BCH_JSET_ENTRY_btree_keys &&
+ !entry->u64s)
continue;
- g = BCH_MEMBER_GROUP(m) - 1;
-
- if (g >= nr_groups ||
- BCH_GROUP_DELETED(&groups->entries[g]))
- return "disk has invalid group";
- }
-
- if (!nr_groups)
- return NULL;
-
- labels = kcalloc(nr_groups, sizeof(char *), GFP_KERNEL);
- if (!labels)
- return "cannot allocate memory";
-
- for (g = groups->entries;
- g < groups->entries + nr_groups;
- g++) {
- if (BCH_GROUP_DELETED(g))
- continue;
-
- len = strnlen(g->label, sizeof(g->label));
-
- labels[nr_live++] = l = kmalloc(len + 1, GFP_KERNEL);
- if (!l) {
- err = "cannot allocate memory";
- goto err;
- }
-
- memcpy(l, g->label, len);
- l[len] = '\0';
+ bch2_journal_entry_to_text(out, NULL, entry);
+ pr_newline(out);
}
-
- sort(labels, nr_live, sizeof(labels[0]), strcmp_void, NULL);
-
- for (i = 0; i + 1 < nr_live; i++)
- if (!strcmp(labels[i], labels[i + 1])) {
- err = "duplicate group labels";
- goto err;
- }
-
- err = NULL;
-err:
- for (i = 0; i < nr_live; i++)
- kfree(labels[i]);
- kfree(labels);
- return err;
}
-static int bch2_sb_disk_groups_to_cpu(struct bch_fs *c)
-{
- struct bch_sb_field_members *mi;
- struct bch_sb_field_disk_groups *groups;
- struct bch_disk_groups_cpu *cpu_g, *old_g;
- unsigned i, nr_groups;
+static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
+ .validate = bch2_sb_clean_validate,
+ .to_text = bch2_sb_clean_to_text,
+};
- lockdep_assert_held(&c->sb_lock);
+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
+};
- mi = bch2_sb_get_members(c->disk_sb);
- groups = bch2_sb_get_disk_groups(c->disk_sb);
- nr_groups = disk_groups_nr(groups);
+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;
+ int ret;
- if (!groups)
+ if (type >= BCH_SB_FIELD_NR)
return 0;
- cpu_g = kzalloc(sizeof(*cpu_g) +
- sizeof(cpu_g->entries[0]) * nr_groups, GFP_KERNEL);
- if (!cpu_g)
- return -ENOMEM;
-
- cpu_g->nr = nr_groups;
-
- for (i = 0; i < nr_groups; i++) {
- struct bch_disk_group *src = &groups->entries[i];
- struct bch_disk_group_cpu *dst = &cpu_g->entries[i];
-
- dst->deleted = BCH_GROUP_DELETED(src);
- }
-
- for (i = 0; i < c->disk_sb->nr_devices; i++) {
- struct bch_member *m = mi->members + i;
- struct bch_disk_group_cpu *dst =
- &cpu_g->entries[BCH_MEMBER_GROUP(m)];
-
- if (!bch2_member_exists(m))
- continue;
-
- dst = BCH_MEMBER_GROUP(m)
- ? &cpu_g->entries[BCH_MEMBER_GROUP(m) - 1]
- : NULL;
- if (dst)
- __set_bit(i, dst->devs.d);
+ ret = bch2_sb_field_ops[type]->validate(sb, f, &field_err);
+ if (ret) {
+ pr_buf(err, "Invalid superblock section %s: %s",
+ bch2_sb_fields[type],
+ field_err.buf);
+ pr_newline(err);
+ bch2_sb_field_to_text(err, sb, f);
}
- old_g = c->disk_groups;
- rcu_assign_pointer(c->disk_groups, cpu_g);
- if (old_g)
- kfree_rcu(old_g, rcu);
-
- return 0;
+ printbuf_exit(&field_err);
+ return ret;
}
-const struct bch_devs_mask *bch2_target_to_mask(struct bch_fs *c, unsigned target)
+void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
+ struct bch_sb_field *f)
{
- struct target t = target_decode(target);
-
- switch (t.type) {
- case TARGET_DEV: {
- struct bch_dev *ca = t.dev < c->sb.nr_devices
- ? rcu_dereference(c->devs[t.dev])
- : NULL;
- return ca ? &ca->self : NULL;
- }
- case TARGET_GROUP: {
- struct bch_disk_groups_cpu *g = rcu_dereference(c->disk_groups);
-
- return t.group < g->nr && !g->entries[t.group].deleted
- ? &g->entries[t.group].devs
- : NULL;
- }
- default:
- BUG();
- }
-}
+ 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;
-int __bch2_disk_group_find(struct bch_sb_field_disk_groups *groups,
- const char *name)
-{
- unsigned i, nr_groups = disk_groups_nr(groups);
- unsigned len = strlen(name);
+ if (!out->tabstops[0])
+ out->tabstops[0] = 32;
- for (i = 0; i < nr_groups; i++) {
- struct bch_disk_group *g = groups->entries + i;
+ if (ops)
+ pr_buf(out, "%s", bch2_sb_fields[type]);
+ else
+ pr_buf(out, "(unknown field %u)", type);
- if (BCH_GROUP_DELETED(g))
- continue;
+ pr_buf(out, " (size %zu):", vstruct_bytes(f));
+ pr_newline(out);
- if (strnlen(g->label, sizeof(g->label)) == len &&
- !memcmp(name, g->label, len))
- return i;
+ if (ops && ops->to_text) {
+ pr_indent_push(out, 2);
+ bch2_sb_field_ops[type]->to_text(out, sb, f);
+ pr_indent_pop(out, 2);
}
-
- return -1;
}
-static int bch2_disk_group_find(struct bch_fs *c, const char *name)
+void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
{
- int ret;
-
- mutex_lock(&c->sb_lock);
- ret = __bch2_disk_group_find(bch2_sb_get_disk_groups(c->disk_sb), name);
- mutex_unlock(&c->sb_lock);
-
- return ret;
-}
+ unsigned i;
-int bch2_opt_target_parse(struct bch_fs *c, const char *buf, u64 *v)
-{
- struct bch_dev *ca;
- int g;
+ pr_buf(out, "Type: %u", l->layout_type);
+ pr_newline(out);
- if (!strlen(buf) || !strcmp(buf, "none")) {
- *v = 0;
- return 0;
- }
+ pr_buf(out, "Superblock max size: ");
+ pr_units(out,
+ 1 << l->sb_max_size_bits,
+ 512 << l->sb_max_size_bits);
+ pr_newline(out);
- /* Is it a device? */
- ca = bch2_dev_lookup(c, buf);
- if (!IS_ERR(ca)) {
- *v = dev_to_target(ca->dev_idx);
- percpu_ref_put(&ca->ref);
- return 0;
- }
+ pr_buf(out, "Nr superblocks: %u", l->nr_superblocks);
+ pr_newline(out);
- g = bch2_disk_group_find(c, buf);
- if (g >= 0) {
- *v = group_to_target(g);
- return 0;
+ pr_buf(out, "Offsets: ");
+ for (i = 0; i < l->nr_superblocks; i++) {
+ if (i)
+ pr_buf(out, ", ");
+ pr_buf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
}
-
- return -EINVAL;
+ pr_newline(out);
}
-int bch2_opt_target_print(struct bch_fs *c, char *buf, size_t len, u64 v)
+void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
+ bool print_layout, unsigned fields)
{
- struct target t = target_decode(v);
- int ret;
+ struct bch_sb_field_members *mi;
+ struct bch_sb_field *f;
+ u64 fields_have = 0;
+ unsigned nr_devices = 0;
- switch (t.type) {
- case TARGET_NULL:
- return scnprintf(buf, len, "none");
- case TARGET_DEV: {
- struct bch_dev *ca;
+ if (!out->tabstops[0])
+ out->tabstops[0] = 32;
- rcu_read_lock();
- ca = t.dev < c->sb.nr_devices
- ? rcu_dereference(c->devs[t.dev])
- : NULL;
+ mi = bch2_sb_get_members(sb);
+ if (mi) {
+ struct bch_member *m;
- if (ca && percpu_ref_tryget(&ca->io_ref)) {
- char b[BDEVNAME_SIZE];
+ for (m = mi->members;
+ m < mi->members + sb->nr_devices;
+ m++)
+ nr_devices += bch2_member_exists(m);
+ }
- ret = scnprintf(buf, len, "/dev/%s",
- bdevname(ca->disk_sb.bdev, b));
- percpu_ref_put(&ca->io_ref);
- } else if (ca) {
- ret = scnprintf(buf, len, "offline device %u", t.dev);
- } else {
- ret = scnprintf(buf, len, "invalid device %u", t.dev);
+ pr_buf(out, "External UUID:");
+ pr_tab(out);
+ pr_uuid(out, sb->user_uuid.b);
+ pr_newline(out);
+
+ pr_buf(out, "Internal UUID:");
+ pr_tab(out);
+ pr_uuid(out, sb->uuid.b);
+ pr_newline(out);
+
+ pr_buf(out, "Device index:");
+ pr_tab(out);
+ pr_buf(out, "%u", sb->dev_idx);
+ pr_newline(out);
+
+ pr_buf(out, "Label:");
+ pr_tab(out);
+ pr_buf(out, "%.*s", (int) sizeof(sb->label), sb->label);
+ pr_newline(out);
+
+ pr_buf(out, "Version:");
+ pr_tab(out);
+ pr_buf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version)]);
+ pr_newline(out);
+
+ pr_buf(out, "Oldest version on disk:");
+ pr_tab(out);
+ pr_buf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version_min)]);
+ pr_newline(out);
+
+ pr_buf(out, "Created:");
+ pr_tab(out);
+ if (sb->time_base_lo)
+ pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
+ else
+ pr_buf(out, "(not set)");
+ pr_newline(out);
+
+ pr_buf(out, "Sequence number:");
+ pr_tab(out);
+ pr_buf(out, "%llu", le64_to_cpu(sb->seq));
+ pr_newline(out);
+
+ pr_buf(out, "Superblock size:");
+ pr_tab(out);
+ pr_buf(out, "%zu", vstruct_bytes(sb));
+ pr_newline(out);
+
+ pr_buf(out, "Clean:");
+ pr_tab(out);
+ pr_buf(out, "%llu", BCH_SB_CLEAN(sb));
+ pr_newline(out);
+
+ pr_buf(out, "Devices:");
+ pr_tab(out);
+ pr_buf(out, "%u", nr_devices);
+ pr_newline(out);
+
+ pr_buf(out, "Sections:");
+ vstruct_for_each(sb, f)
+ fields_have |= 1 << le32_to_cpu(f->type);
+ pr_tab(out);
+ bch2_flags_to_text(out, bch2_sb_fields, fields_have);
+ pr_newline(out);
+
+ pr_buf(out, "Features:");
+ pr_tab(out);
+ bch2_flags_to_text(out, bch2_sb_features,
+ le64_to_cpu(sb->features[0]));
+ pr_newline(out);
+
+ pr_buf(out, "Compat features:");
+ pr_tab(out);
+ bch2_flags_to_text(out, bch2_sb_compat,
+ le64_to_cpu(sb->compat[0]));
+ pr_newline(out);
+
+ pr_newline(out);
+ pr_buf(out, "Options:");
+ pr_newline(out);
+ pr_indent_push(out, 2);
+ {
+ enum bch_opt_id id;
+
+ for (id = 0; id < bch2_opts_nr; id++) {
+ const struct bch_option *opt = bch2_opt_table + id;
+
+ if (opt->get_sb != BCH2_NO_SB_OPT) {
+ u64 v = bch2_opt_from_sb(sb, id);
+
+ pr_buf(out, "%s:", opt->attr.name);
+ pr_tab(out);
+ bch2_opt_to_text(out, NULL, sb, opt, v,
+ OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
+ pr_newline(out);
+ }
}
-
- rcu_read_unlock();
- break;
}
- case TARGET_GROUP: {
- struct bch_sb_field_disk_groups *groups;
- struct bch_disk_group *g;
-
- mutex_lock(&c->sb_lock);
- groups = bch2_sb_get_disk_groups(c->disk_sb);
-
- g = t.group < disk_groups_nr(groups)
- ? groups->entries + t.group
- : NULL;
- if (g && !BCH_GROUP_DELETED(g)) {
- ret = len ? min(len - 1, strnlen(g->label, sizeof(g->label))) : 0;
-
- memcpy(buf, g->label, ret);
- if (len)
- buf[ret] = '\0';
- } else {
- ret = scnprintf(buf, len, "invalid group %u", t.group);
- }
+ pr_indent_pop(out, 2);
- mutex_unlock(&c->sb_lock);
- break;
- }
- default:
- BUG();
+ if (print_layout) {
+ pr_newline(out);
+ pr_buf(out, "layout:");
+ pr_newline(out);
+ pr_indent_push(out, 2);
+ bch2_sb_layout_to_text(out, &sb->layout);
+ pr_indent_pop(out, 2);
}
- return ret;
+ vstruct_for_each(sb, f)
+ if (fields & (1 << le32_to_cpu(f->type))) {
+ pr_newline(out);
+ bch2_sb_field_to_text(out, sb, f);
+ }
}
projects / bcachefs-tools-debian / blobdiff