+// SPDX-License-Identifier: GPL-2.0
#include "bcachefs.h"
+#include "btree_update_interior.h"
+#include "buckets.h"
#include "checksum.h"
+#include "disk_groups.h"
+#include "ec.h"
#include "error.h"
#include "io.h"
#include "journal.h"
+#include "journal_seq_blacklist.h"
+#include "replicas.h"
+#include "quota.h"
#include "super-io.h"
#include "super.h"
#include "vstructs.h"
#include <linux/backing-dev.h>
#include <linux/sort.h>
-static int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *);
-static const char *bch2_sb_validate_replicas(struct bch_sb *);
+const char * const bch2_sb_fields[] = {
+#define x(name, nr) #name,
+ BCH_SB_FIELDS()
+#undef x
+ NULL
+};
-static inline void __bch2_sb_layout_size_assert(void)
-{
- BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
-}
+static const char *bch2_sb_field_validate(struct bch_sb *,
+ struct bch_sb_field *);
struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
enum bch_sb_field_type type)
return NULL;
}
-void bch2_free_super(struct bcache_superblock *sb)
+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(get_order(__vstruct_bytes(struct bch_sb, sb_u64s)) >
+ sb->page_order);
+
+ 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;
+ } else {
+ void *src, *dst;
+
+ src = vstruct_end(f);
+
+ if (u64s) {
+ f->u64s = cpu_to_le32(u64s);
+ dst = vstruct_end(f);
+ } else {
+ dst = f;
+ }
+
+ memmove(dst, src, vstruct_end(sb->sb) - src);
+
+ if (dst > src)
+ memset(src, 0, dst - src);
+ }
+
+ sb->sb->u64s = cpu_to_le32(sb_u64s);
+
+ 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: */
+
+void bch2_free_super(struct bch_sb_handle *sb)
{
if (sb->bio)
bio_put(sb->bio);
memset(sb, 0, sizeof(*sb));
}
-static int __bch2_super_realloc(struct bcache_superblock *sb, unsigned order)
+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);
struct bch_sb *new_sb;
struct bio *bio;
+ if (sb->sb && sb->page_order >= order)
+ return 0;
+
+ if (sb->have_layout) {
+ u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
+
+ if (new_bytes > max_bytes) {
+ char buf[BDEVNAME_SIZE];
+
+ pr_err("%s: superblock too big: want %zu but have %llu",
+ bdevname(sb->bdev, buf), new_bytes, max_bytes);
+ return -ENOSPC;
+ }
+ }
+
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->have_bio) {
+ bio = bio_kmalloc(GFP_KERNEL, 1 << order);
+ if (!bio)
+ return -ENOMEM;
- if (sb->bio)
- bio_put(sb->bio);
- sb->bio = bio;
+ if (sb->bio)
+ bio_put(sb->bio);
+ sb->bio = bio;
+ }
- new_sb = (void *) __get_free_pages(GFP_KERNEL, order);
+ new_sb = (void *) __get_free_pages(GFP_NOFS|__GFP_ZERO, order);
if (!new_sb)
return -ENOMEM;
return 0;
}
-static int bch2_sb_realloc(struct bcache_superblock *sb, unsigned u64s)
-{
- u64 new_bytes = __vstruct_bytes(struct bch_sb, u64s);
- u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
-
- if (new_bytes > max_bytes) {
- char buf[BDEVNAME_SIZE];
-
- pr_err("%s: superblock too big: want %llu but have %llu",
- bdevname(sb->bdev, buf), new_bytes, max_bytes);
- return -ENOSPC;
- }
-
- 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)
- 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);
-
- c->disk_sb = sb;
- c->disk_sb_order = order;
- return 0;
-}
-
-static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb *sb,
- struct bch_sb_field *f,
- unsigned u64s)
-{
- unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
-
- if (!f) {
- f = vstruct_last(sb);
- memset(f, 0, sizeof(u64) * u64s);
- f->u64s = cpu_to_le32(u64s);
- f->type = 0;
- } else {
- 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 (dst > src)
- memset(src, 0, dst - src);
- }
-
- le32_add_cpu(&sb->u64s, u64s - old_u64s);
-
- return f;
-}
-
-struct bch_sb_field *bch2_sb_field_resize(struct bcache_superblock *sb,
- enum bch_sb_field_type type,
- unsigned u64s)
+struct bch_sb_field *bch2_sb_field_resize(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);
ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 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 = type;
- return f;
-}
+ if (sb->fs_sb) {
+ struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
+ struct bch_dev *ca;
+ unsigned i;
-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);
- lockdep_assert_held(&c->sb_lock);
+ /* XXX: we're not checking that offline device have enough space */
- if (bch2_fs_sb_realloc(c, le32_to_cpu(c->disk_sb->u64s) + d))
- return NULL;
-
- /* XXX: we're not checking that offline device have enough space */
-
- for_each_online_member(ca, c, i) {
- struct bcache_superblock *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 = 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;
}
+/* Superblock validate: */
+
+static inline void __bch2_sb_layout_size_assert(void)
+{
+ BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
+}
+
static const char *validate_sb_layout(struct bch_sb_layout *layout)
{
u64 offset, prev_offset, max_sectors;
return NULL;
}
-static int u64_cmp(const void *_l, const void *_r)
-{
- u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
-
- return l < r ? -1 : l > r ? 1 : 0;
-}
-
-const char *bch2_sb_validate_journal(struct bch_sb *sb,
- struct bch_member_cpu mi)
-{
- struct bch_sb_field_journal *journal;
- 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 (b[0] < mi.first_bucket)
- goto err;
-
- err = "journal bucket past end of device";
- if (b[nr - 1] >= mi.nbuckets)
- goto err;
-
- err = "duplicate journal buckets";
- for (i = 0; i + 1 < nr; i++)
- if (b[i] == b[i + 1])
- goto err;
-
- err = NULL;
-err:
- kfree(b);
- return err;
-}
-
-static const char *bch2_sb_validate_members(struct bch_sb *sb)
-{
- struct bch_sb_field_members *mi;
- unsigned i;
-
- mi = bch2_sb_get_members(sb);
- if (!mi)
- return "Invalid superblock: member info area missing";
-
- 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++) {
- if (!bch2_dev_exists(sb, mi, i))
- continue;
-
- if (le16_to_cpu(mi->members[i].bucket_size) <
- BCH_SB_BTREE_NODE_SIZE(sb))
- return "bucket size smaller than btree node size";
- }
-
- return NULL;
-}
-
-const char *bch2_sb_validate(struct bcache_superblock *disk_sb)
+const char *bch2_sb_validate(struct bch_sb_handle *disk_sb)
{
struct bch_sb *sb = disk_sb->sb;
struct bch_sb_field *f;
- struct bch_sb_field_members *sb_mi;
- struct bch_member_cpu mi;
+ struct bch_sb_field_members *mi;
const char *err;
+ u32 version, version_min;
u16 block_size;
- switch (le64_to_cpu(sb->version)) {
- case BCACHE_SB_VERSION_CDEV_V4:
- break;
- default:
- return"Unsupported superblock version";
- }
+ version = le16_to_cpu(sb->version);
+ version_min = version >= bcachefs_metadata_version_new_versioning
+ ? le16_to_cpu(sb->version_min)
+ : version;
- if (BCH_SB_INITIALIZED(sb) &&
- le64_to_cpu(sb->version) != BCACHE_SB_VERSION_CDEV_V4)
+ if (version >= bcachefs_metadata_version_max ||
+ version_min < bcachefs_metadata_version_min)
return "Unsupported superblock version";
+ if (version_min > version)
+ return "Bad minimum version";
+
+ if (sb->features[1] ||
+ (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR)))
+ return "Filesystem has incompatible features";
+
block_size = le16_to_cpu(sb->block_size);
if (!is_power_of_2(block_size) ||
if (!sb->nr_devices ||
sb->nr_devices <= sb->dev_idx ||
sb->nr_devices > BCH_SB_MEMBERS_MAX)
- return "Bad cache device number in set";
+ return "Bad number of member devices";
if (!BCH_SB_META_REPLICAS_WANT(sb) ||
BCH_SB_META_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
if (!BCH_SB_DATA_REPLICAS_REQ(sb) ||
BCH_SB_DATA_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
- return "Invalid number of metadata replicas";
+ return "Invalid number of data replicas";
+
+ if (BCH_SB_META_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
+ return "Invalid metadata checksum type";
+
+ if (BCH_SB_DATA_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
+ return "Invalid metadata checksum type";
+
+ if (BCH_SB_COMPRESSION_TYPE(sb) >= BCH_COMPRESSION_OPT_NR)
+ return "Invalid compression type";
if (!BCH_SB_BTREE_NODE_SIZE(sb))
return "Btree node size not set";
if (!is_power_of_2(BCH_SB_BTREE_NODE_SIZE(sb)))
return "Btree node size not a power of two";
- if (BCH_SB_BTREE_NODE_SIZE(sb) > BTREE_NODE_SIZE_MAX)
- return "Btree node size too large";
-
if (BCH_SB_GC_RESERVE(sb) < 5)
return "gc reserve percentage too small";
if (vstruct_next(f) > vstruct_last(sb))
return "Invalid superblock: invalid optional field";
-
- if (le32_to_cpu(f->type) >= BCH_SB_FIELD_NR)
- return "Invalid superblock: unknown optional field type";
}
- err = bch2_sb_validate_members(sb);
- if (err)
- return err;
-
- sb_mi = bch2_sb_get_members(sb);
- mi = bch2_mi_to_cpu(sb_mi->members + sb->dev_idx);
-
- if (mi.nbuckets > LONG_MAX)
- return "Too many buckets";
-
- if (mi.nbuckets - mi.first_bucket < 1 << 10)
- return "Not enough buckets";
-
- if (!is_power_of_2(mi.bucket_size) ||
- mi.bucket_size < PAGE_SECTORS ||
- mi.bucket_size < block_size)
- return "Bad bucket size";
-
- if (get_capacity(disk_sb->bdev->bd_disk) <
- mi.bucket_size * mi.nbuckets)
- return "Invalid superblock: device too small";
+ /* members must be validated first: */
+ mi = bch2_sb_get_members(sb);
+ if (!mi)
+ return "Invalid superblock: member info area missing";
- err = bch2_sb_validate_journal(sb, mi);
+ err = bch2_sb_field_validate(sb, &mi->field);
if (err)
return err;
- err = bch2_sb_validate_replicas(sb);
- if (err)
- return err;
+ 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;
+ }
return NULL;
}
/* device open: */
-static const char *bch2_blkdev_open(const char *path, fmode_t mode,
- void *holder, struct block_device **ret)
-{
- struct block_device *bdev;
-
- *ret = NULL;
- bdev = blkdev_get_by_path(path, mode, holder);
- if (bdev == ERR_PTR(-EBUSY))
- return "device busy";
-
- if (IS_ERR(bdev))
- return "failed to open device";
-
- if (mode & FMODE_WRITE)
- bdev_get_queue(bdev)->backing_dev_info->capabilities
- |= BDI_CAP_STABLE_WRITES;
-
- *ret = bdev;
- return NULL;
-}
-
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.block_size = le16_to_cpu(src->block_size);
- c->sb.btree_node_size = BCH_SB_BTREE_NODE_SIZE(src);
+ c->sb.version = le16_to_cpu(src->version);
c->sb.nr_devices = src->nr_devices;
c->sb.clean = BCH_SB_CLEAN(src);
- c->sb.str_hash_type = BCH_SB_STR_HASH_TYPE(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.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 (i == BCH_SB_FIELD_journal)
continue;
- dst_f = bch2_sb_field_get(dst, 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);
- if (bch2_fs_sb_realloc(c, le32_to_cpu(src->u64s) - journal_u64s))
- return -ENOMEM;
+ 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)
return ret;
+ ret = bch2_sb_disk_groups_to_cpu(c);
+ if (ret)
+ return ret;
+
bch2_sb_update(c);
return 0;
}
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 bcache_superblock *sb, u64 offset)
+static const char *read_one_super(struct bch_sb_handle *sb, u64 offset)
{
struct bch_csum csum;
size_t bytes;
- unsigned order;
reread:
bio_reset(sb->bio);
- sb->bio->bi_bdev = sb->bdev;
+ bio_set_dev(sb->bio, sb->bdev);
sb->bio->bi_iter.bi_sector = offset;
- sb->bio->bi_iter.bi_size = PAGE_SIZE << sb->page_order;
bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
- bch2_bio_map(sb->bio, sb->sb);
+ bch2_bio_map(sb->bio, sb->sb, PAGE_SIZE << sb->page_order);
if (submit_bio_wait(sb->bio))
return "IO error";
if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
return "Not a bcachefs superblock";
- if (le64_to_cpu(sb->sb->version) != BCACHE_SB_VERSION_CDEV_V4)
+ if (le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_min ||
+ le16_to_cpu(sb->sb->version) >= bcachefs_metadata_version_max)
return "Unsupported superblock version";
bytes = vstruct_bytes(sb->sb);
if (bytes > 512 << sb->sb->layout.sb_max_size_bits)
return "Bad superblock: too big";
- order = get_order(bytes);
- if (order > sb->page_order) {
- if (__bch2_super_realloc(sb, order))
+ if (get_order(bytes) > sb->page_order) {
+ if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
return "cannot allocate memory";
goto reread;
}
/* XXX: verify MACs */
csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
- (struct nonce) { 0 }, sb->sb);
+ null_nonce(), sb->sb);
if (bch2_crc_cmp(csum, sb->sb->csum))
return "bad checksum reading superblock";
+ sb->seq = le64_to_cpu(sb->sb->seq);
+
return NULL;
}
-const char *bch2_read_super(struct bcache_superblock *sb,
- struct bch_opts opts,
- const char *path)
+int bch2_read_super(const char *path, struct bch_opts *opts,
+ struct bch_sb_handle *sb)
{
- u64 offset = opt_defined(opts.sb) ? opts.sb : BCH_SB_SECTOR;
+ u64 offset = opt_get(*opts, sb);
struct bch_sb_layout layout;
const char *err;
- unsigned i;
+ __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_defined(opts.noexcl) && opts.noexcl))
+ if (!opt_get(*opts, noexcl))
sb->mode |= FMODE_EXCL;
- if (!(opt_defined(opts.nochanges) && opts.nochanges))
+ if (!opt_get(*opts, nochanges))
sb->mode |= FMODE_WRITE;
- err = bch2_blkdev_open(path, sb->mode, sb, &sb->bdev);
- if (err)
- return err;
+ sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
+ if (IS_ERR(sb->bdev) &&
+ PTR_ERR(sb->bdev) == -EACCES &&
+ opt_get(*opts, read_only)) {
+ sb->mode &= ~FMODE_WRITE;
+
+ sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
+ if (!IS_ERR(sb->bdev))
+ opt_set(*opts, nochanges, true);
+ }
+
+ if (IS_ERR(sb->bdev)) {
+ ret = PTR_ERR(sb->bdev);
+ goto out;
+ }
err = "cannot allocate memory";
- if (__bch2_super_realloc(sb, 0))
+ ret = bch2_sb_realloc(sb, 0);
+ if (ret)
goto err;
+ ret = -EFAULT;
err = "dynamic fault";
if (bch2_fs_init_fault("read_super"))
goto err;
+ ret = -EINVAL;
err = read_one_super(sb, offset);
if (!err)
goto got_super;
- if (offset != BCH_SB_SECTOR) {
- pr_err("error reading superblock: %s", err);
+ if (opt_defined(*opts, sb))
goto err;
- }
pr_err("error reading default superblock: %s", err);
* superblocks:
*/
bio_reset(sb->bio);
- sb->bio->bi_bdev = sb->bdev;
+ 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))
if (err)
goto err;
- for (i = 0; i < layout.nr_superblocks; i++) {
- u64 offset = le64_to_cpu(layout.sb_offset[i]);
+ for (i = layout.sb_offset;
+ i < layout.sb_offset + layout.nr_superblocks; i++) {
+ offset = le64_to_cpu(*i);
- if (offset == BCH_SB_SECTOR)
+ if (offset == opt_get(*opts, sb))
continue;
err = read_one_super(sb, offset);
if (!err)
goto got_super;
}
+
+ ret = -EINVAL;
goto err;
-got_super:
- pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
- le64_to_cpu(sb->sb->version),
- le64_to_cpu(sb->sb->flags),
- le64_to_cpu(sb->sb->seq),
- le16_to_cpu(sb->sb->u64s));
+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))
goto err;
- return NULL;
+ ret = 0;
+ sb->have_layout = true;
+out:
+ pr_verbose_init(*opts, "ret %i", ret);
+ return ret;
err:
bch2_free_super(sb);
- return err;
+ pr_err("error reading superblock: %s", err);
+ goto out;
}
/* write superblock: */
/* XXX: return errors directly */
- if (bch2_dev_io_err_on(bio->bi_error, 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;
SET_BCH_SB_CSUM_TYPE(sb, c->opts.metadata_checksum);
sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
- (struct nonce) { 0 }, sb);
+ null_nonce(), sb);
bio_reset(bio);
- bio->bi_bdev = ca->disk_sb.bdev;
+ 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],
+ bio_sectors(bio));
percpu_ref_get(&ca->io_ref);
closure_bio_submit(bio, &c->sb_write);
}
-void bch2_write_super(struct bch_fs *c)
+int bch2_write_super(struct bch_fs *c)
{
struct closure *cl = &c->sb_write;
struct bch_dev *ca;
const char *err;
struct bch_devs_mask sb_written;
bool wrote, can_mount_without_written, can_mount_with_written;
+ int ret = 0;
lockdep_assert_held(&c->sb_lock);
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);
for_each_online_member(ca, c, i)
bch2_sb_from_fs(c, ca);
err = bch2_sb_validate(&ca->disk_sb);
if (err) {
bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
+ ret = -1;
goto out;
}
}
- if (c->opts.nochanges ||
- test_bit(BCH_FS_ERROR, &c->flags))
+ if (c->opts.nochanges)
goto out;
for_each_online_member(ca, c, i) {
ca->sb_write_error = 0;
}
+ for_each_online_member(ca, c, i)
+ read_back_super(c, ca);
+ closure_sync(cl);
+
+ for_each_online_member(ca, c, i) {
+ if (!ca->sb_write_error &&
+ ca->disk_sb.seq !=
+ le64_to_cpu(ca->sb_read_scratch->seq)) {
+ bch2_fs_fatal_error(c,
+ "Superblock modified by another process");
+ percpu_ref_put(&ca->io_ref);
+ ret = -EROFS;
+ goto out;
+ }
+ }
+
do {
wrote = false;
for_each_online_member(ca, c, i)
- if (sb < ca->disk_sb.sb->layout.nr_superblocks) {
+ if (!ca->sb_write_error &&
+ sb < ca->disk_sb.sb->layout.nr_superblocks) {
write_one_super(c, ca, sb);
wrote = true;
}
sb++;
} while (wrote);
- for_each_online_member(ca, c, i)
+ for_each_online_member(ca, c, i) {
if (ca->sb_write_error)
__clear_bit(ca->dev_idx, sb_written.d);
+ else
+ ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
+ }
- nr_wrote = bitmap_weight(sb_written.d, BCH_SB_MEMBERS_MAX);
+ nr_wrote = dev_mask_nr(&sb_written);
can_mount_with_written =
- bch2_have_enough_devs(c,
- __bch2_replicas_status(c, sb_written),
- BCH_FORCE_IF_DEGRADED);
+ bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
+ BCH_FORCE_IF_DEGRADED);
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(c,
- __bch2_replicas_status(c, sb_written),
- BCH_FORCE_IF_DEGRADED);
+ bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
+ BCH_FORCE_IF_DEGRADED);
/*
* 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_without_written &&
+ !can_mount_with_written), c,
+ "Unable to write superblock to sufficient devices"))
+ ret = -1;
out:
/* Make new options visible after they're persistent: */
bch2_sb_update(c);
+ return ret;
}
-/* replica information: */
-
-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;
-}
-
-static inline struct bch_replicas_entry *
-replicas_entry_next(struct bch_replicas_entry *i)
-{
- return (void *) i + offsetof(struct bch_replicas_entry, devs) + i->nr;
-}
-
-#define for_each_replicas_entry(_r, _i) \
- for (_i = (_r)->entries; \
- (void *) (_i) < vstruct_end(&(_r)->field) && (_i)->data_type;\
- (_i) = replicas_entry_next(_i))
-
-static inline bool replicas_test_dev(struct bch_replicas_cpu_entry *e,
- unsigned dev)
+void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
{
- return (e->devs[dev >> 3] & (1 << (dev & 7))) != 0;
-}
+ mutex_lock(&c->sb_lock);
+ if (!(c->sb.features & (1ULL << feat))) {
+ c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
-static inline void replicas_set_dev(struct bch_replicas_cpu_entry *e,
- unsigned dev)
-{
- e->devs[dev >> 3] |= 1 << (dev & 7);
+ bch2_write_super(c);
+ }
+ mutex_unlock(&c->sb_lock);
}
-static inline unsigned replicas_dev_slots(struct bch_replicas_cpu *r)
-{
- return (r->entry_size -
- offsetof(struct bch_replicas_cpu_entry, devs)) * 8;
-}
+/* BCH_SB_FIELD_journal: */
-static void bch2_sb_replicas_nr_entries(struct bch_sb_field_replicas *r,
- unsigned *nr,
- unsigned *bytes,
- unsigned *max_dev)
+static int u64_cmp(const void *_l, const void *_r)
{
- struct bch_replicas_entry *i;
- unsigned j;
-
- *nr = 0;
- *bytes = sizeof(*r);
- *max_dev = 0;
-
- 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)++;
- }
+ u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
- *bytes = (void *) i - (void *) r;
+ return l < r ? -1 : l > r ? 1 : 0;
}
-static struct bch_replicas_cpu *
-__bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r)
+static const char *bch2_sb_validate_journal(struct bch_sb *sb,
+ struct bch_sb_field *f)
{
- 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);
+ 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;
- cpu_r = kzalloc(sizeof(struct bch_replicas_cpu) +
- nr * entry_size, GFP_NOIO);
- if (!cpu_r)
+ journal = bch2_sb_get_journal(sb);
+ if (!journal)
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;
+ nr = bch2_nr_journal_buckets(journal);
+ if (!nr)
+ return NULL;
- 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]);
+ b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
+ if (!b)
+ return "cannot allocate memory";
- src = replicas_entry_next(src);
- dst = (void *) dst + entry_size;
- }
- }
+ for (i = 0; i < nr; i++)
+ b[i] = le64_to_cpu(journal->buckets[i]);
- eytzinger0_sort(cpu_r->entries,
- cpu_r->nr,
- cpu_r->entry_size,
- memcmp, NULL);
- return cpu_r;
-}
+ sort(b, nr, sizeof(u64), u64_cmp, NULL);
-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;
+ err = "journal bucket at sector 0";
+ if (!b[0])
+ goto err;
- lockdep_assert_held(&c->sb_lock);
+ err = "journal bucket before first bucket";
+ if (m && b[0] < le16_to_cpu(m->first_bucket))
+ goto err;
- sb_r = bch2_sb_get_replicas(c->disk_sb);
- cpu_r = __bch2_sb_replicas_to_cpu_replicas(sb_r);
- if (!cpu_r)
- return -ENOMEM;
+ err = "journal bucket past end of device";
+ if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
+ goto err;
- old_r = c->replicas;
- rcu_assign_pointer(c->replicas, cpu_r);
- if (old_r)
- kfree_rcu(old_r, rcu);
+ err = "duplicate journal buckets";
+ for (i = 0; i + 1 < nr; i++)
+ if (b[i] == b[i + 1])
+ goto err;
- return 0;
+ err = NULL;
+err:
+ kfree(b);
+ return err;
}
-static void bkey_to_replicas(struct bkey_s_c_extent e,
- enum bch_data_types data_type,
- struct bch_replicas_cpu_entry *r,
- unsigned *max_dev)
-{
- const struct bch_extent_ptr *ptr;
-
- BUG_ON(!data_type ||
- data_type == BCH_DATA_SB ||
- data_type >= BCH_DATA_NR);
-
- memset(r, 0, sizeof(*r));
- r->data_type = data_type;
+static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
+ .validate = bch2_sb_validate_journal,
+};
- *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);
- }
-}
+/* BCH_SB_FIELD_members: */
-/*
- * for when gc of replica information is in progress:
- */
-static int bch2_update_gc_replicas(struct bch_fs *c,
- struct bch_replicas_cpu *gc_r,
- struct bkey_s_c_extent e,
- enum bch_data_types data_type)
+static const char *bch2_sb_validate_members(struct bch_sb *sb,
+ struct bch_sb_field *f)
{
- struct bch_replicas_cpu_entry new_e;
- struct bch_replicas_cpu *new;
- unsigned i, nr, entry_size, max_dev;
+ struct bch_sb_field_members *mi = field_to_type(f, members);
+ struct bch_member *m;
- bkey_to_replicas(e, data_type, &new_e, &max_dev);
-
- entry_size = offsetof(struct bch_replicas_cpu_entry, devs) +
- DIV_ROUND_UP(max_dev + 1, 8);
- entry_size = max(entry_size, gc_r->entry_size);
- nr = gc_r->nr + 1;
-
- new = kzalloc(sizeof(struct bch_replicas_cpu) +
- nr * entry_size, GFP_NOIO);
- if (!new)
- return -ENOMEM;
-
- new->nr = nr;
- new->entry_size = entry_size;
-
- for (i = 0; i < gc_r->nr; i++)
- memcpy(cpu_replicas_entry(new, i),
- cpu_replicas_entry(gc_r, i),
- gc_r->entry_size);
+ if ((void *) (mi->members + sb->nr_devices) >
+ vstruct_end(&mi->field))
+ return "Invalid superblock: bad member info";
- memcpy(cpu_replicas_entry(new, nr - 1),
- &new_e,
- new->entry_size);
+ for (m = mi->members;
+ m < mi->members + sb->nr_devices;
+ m++) {
+ if (!bch2_member_exists(m))
+ continue;
- eytzinger0_sort(new->entries,
- new->nr,
- new->entry_size,
- memcmp, NULL);
+ if (le64_to_cpu(m->nbuckets) > LONG_MAX)
+ return "Too many buckets";
- rcu_assign_pointer(c->replicas_gc, new);
- kfree_rcu(gc_r, rcu);
- return 0;
-}
+ if (le64_to_cpu(m->nbuckets) -
+ le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
+ return "Not enough buckets";
-static bool replicas_has_extent(struct bch_replicas_cpu *r,
- struct bkey_s_c_extent e,
- enum bch_data_types data_type)
-{
- struct bch_replicas_cpu_entry search;
- unsigned max_dev;
+ if (le16_to_cpu(m->bucket_size) <
+ le16_to_cpu(sb->block_size))
+ return "bucket size smaller than block size";
- bkey_to_replicas(e, data_type, &search, &max_dev);
+ if (le16_to_cpu(m->bucket_size) <
+ BCH_SB_BTREE_NODE_SIZE(sb))
+ return "bucket size smaller than btree node size";
+ }
- return max_dev < replicas_dev_slots(r) &&
- eytzinger0_find(r->entries, r->nr,
- r->entry_size,
- memcmp, &search) < r->nr;
+ return NULL;
}
-bool bch2_sb_has_replicas(struct bch_fs *c, struct bkey_s_c_extent e,
- enum bch_data_types data_type)
-{
- bool ret;
-
- rcu_read_lock();
- ret = replicas_has_extent(rcu_dereference(c->replicas),
- e, data_type);
- rcu_read_unlock();
+static const struct bch_sb_field_ops bch_sb_field_ops_members = {
+ .validate = bch2_sb_validate_members,
+};
- return ret;
-}
+/* BCH_SB_FIELD_crypt: */
-noinline
-static int bch2_check_mark_super_slowpath(struct bch_fs *c,
- struct bkey_s_c_extent e,
- enum bch_data_types data_type)
+static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
+ struct bch_sb_field *f)
{
- struct bch_replicas_cpu *gc_r;
- const struct bch_extent_ptr *ptr;
- struct bch_sb_field_replicas *sb_r;
- struct bch_replicas_entry *new_entry;
- unsigned new_entry_bytes, new_u64s, nr, bytes, max_dev;
- int ret = 0;
-
- mutex_lock(&c->sb_lock);
-
- gc_r = rcu_dereference_protected(c->replicas_gc,
- lockdep_is_held(&c->sb_lock));
- if (gc_r &&
- !replicas_has_extent(gc_r, e, data_type)) {
- ret = bch2_update_gc_replicas(c, gc_r, e, data_type);
- if (ret)
- goto err;
- }
-
- /* recheck, might have raced */
- if (bch2_sb_has_replicas(c, e, data_type)) {
- mutex_unlock(&c->sb_lock);
- return 0;
- }
+ struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
- new_entry_bytes = sizeof(struct bch_replicas_entry) +
- bch2_extent_nr_dirty_ptrs(e.s_c);
+ if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
+ return "invalid field crypt: wrong size";
- sb_r = bch2_sb_get_replicas(c->disk_sb);
+ if (BCH_CRYPT_KDF_TYPE(crypt))
+ return "invalid field crypt: bad kdf type";
- bch2_sb_replicas_nr_entries(sb_r, &nr, &bytes, &max_dev);
-
- new_u64s = DIV_ROUND_UP(bytes + new_entry_bytes, sizeof(u64));
-
- sb_r = bch2_fs_sb_resize_replicas(c,
- DIV_ROUND_UP(sizeof(*sb_r) + bytes + new_entry_bytes,
- sizeof(u64)));
- if (!sb_r) {
- ret = -ENOSPC;
- goto err;
- }
-
- new_entry = (void *) sb_r + bytes;
- new_entry->data_type = data_type;
- new_entry->nr = 0;
-
- extent_for_each_ptr(e, ptr)
- if (!ptr->cached)
- new_entry->devs[new_entry->nr++] = ptr->dev;
-
- ret = bch2_sb_replicas_to_cpu_replicas(c);
- if (ret) {
- memset(new_entry, 0,
- vstruct_end(&sb_r->field) - (void *) new_entry);
- goto err;
- }
-
- bch2_write_super(c);
-err:
- mutex_unlock(&c->sb_lock);
- return ret;
+ return NULL;
}
-int bch2_check_mark_super(struct bch_fs *c, struct bkey_s_c_extent e,
- enum bch_data_types data_type)
-{
- struct bch_replicas_cpu *gc_r;
- bool marked;
+static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
+ .validate = bch2_sb_validate_crypt,
+};
- rcu_read_lock();
- marked = replicas_has_extent(rcu_dereference(c->replicas),
- e, data_type) &&
- (!(gc_r = rcu_dereference(c->replicas_gc)) ||
- replicas_has_extent(gc_r, e, data_type));
- rcu_read_unlock();
+/* BCH_SB_FIELD_clean: */
- if (marked)
- return 0;
+void bch2_sb_clean_renumber(struct bch_sb_field_clean *clean, int write)
+{
+ struct jset_entry *entry;
- return bch2_check_mark_super_slowpath(c, e, data_type);
+ 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);
}
-struct replicas_status __bch2_replicas_status(struct bch_fs *c,
- struct bch_devs_mask online_devs)
+int bch2_fs_mark_dirty(struct bch_fs *c)
{
- 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;
-
- rcu_read_lock();
- r = rcu_dereference(c->replicas);
- dev_slots = min_t(unsigned, replicas_dev_slots(r), c->sb.nr_devices);
-
- for (i = 0; i < r->nr; i++) {
- e = cpu_replicas_entry(r, i);
-
- BUG_ON(e->data_type >= ARRAY_SIZE(ret.replicas));
-
- nr_online = nr_offline = 0;
-
- for (dev = 0; dev < dev_slots; dev++) {
- if (!replicas_test_dev(e, dev))
- continue;
-
- if (test_bit(dev, online_devs.d))
- nr_online++;
- else
- nr_offline++;
- }
-
- ret.replicas[e->data_type].nr_online =
- min(ret.replicas[e->data_type].nr_online,
- nr_online);
+ int ret;
- ret.replicas[e->data_type].nr_offline =
- max(ret.replicas[e->data_type].nr_offline,
- nr_offline);
- }
+ /*
+ * Unconditionally write superblock, to verify it hasn't changed before
+ * we go rw:
+ */
- rcu_read_unlock();
+ mutex_lock(&c->sb_lock);
+ SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
+ c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_extent_overwrite;
+ c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_extents_above_btree_updates;
+ c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_btree_updates_journalled;
+ ret = bch2_write_super(c);
+ mutex_unlock(&c->sb_lock);
return ret;
}
-struct replicas_status bch2_replicas_status(struct bch_fs *c)
+static void
+entry_init_u64s(struct jset_entry *entry, unsigned u64s)
{
- return __bch2_replicas_status(c, bch2_online_devs(c));
-}
-
-bool bch2_have_enough_devs(struct bch_fs *c,
- struct replicas_status s,
- unsigned flags)
-{
- if ((s.replicas[BCH_DATA_JOURNAL].nr_offline ||
- s.replicas[BCH_DATA_BTREE].nr_offline) &&
- !(flags & BCH_FORCE_IF_METADATA_DEGRADED))
- return false;
-
- if ((!s.replicas[BCH_DATA_JOURNAL].nr_online ||
- !s.replicas[BCH_DATA_BTREE].nr_online) &&
- !(flags & BCH_FORCE_IF_METADATA_LOST))
- return false;
+ memset(entry, 0, u64s * sizeof(u64));
- if (s.replicas[BCH_DATA_USER].nr_offline &&
- !(flags & BCH_FORCE_IF_DATA_DEGRADED))
- return false;
-
- if (!s.replicas[BCH_DATA_USER].nr_online &&
- !(flags & BCH_FORCE_IF_DATA_LOST))
- return false;
-
- return true;
+ /*
+ * The u64s field counts from the start of data, ignoring the shared
+ * fields.
+ */
+ entry->u64s = u64s - 1;
}
-unsigned bch2_replicas_online(struct bch_fs *c, bool meta)
+static void
+entry_init_size(struct jset_entry *entry, size_t size)
{
- 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;
+ unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
+ entry_init_u64s(entry, u64s);
}
-unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
+struct jset_entry *
+bch2_journal_super_entries_add_common(struct bch_fs *c,
+ struct jset_entry *entry,
+ u64 journal_seq)
{
- struct bch_replicas_cpu_entry *e;
- struct bch_replicas_cpu *r;
- unsigned i, ret = 0;
-
- rcu_read_lock();
- r = rcu_dereference(c->replicas);
-
- if (ca->dev_idx >= replicas_dev_slots(r))
- goto out;
+ unsigned i;
- for (i = 0; i < r->nr; i++) {
- e = cpu_replicas_entry(r, i);
+ percpu_down_write(&c->mark_lock);
- if (replicas_test_dev(e, ca->dev_idx)) {
- ret |= 1 << e->data_type;
- break;
- }
+ 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);
}
-out:
- rcu_read_unlock();
- return ret;
-}
+ {
+ struct jset_entry_usage *u =
+ container_of(entry, struct jset_entry_usage, entry);
-static const char *bch2_sb_validate_replicas(struct bch_sb *sb)
-{
- struct bch_sb_field_members *mi;
- struct bch_sb_field_replicas *sb_r;
- struct bch_replicas_cpu *cpu_r = NULL;
- struct bch_replicas_entry *e;
- const char *err;
- unsigned i;
+ 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);
- mi = bch2_sb_get_members(sb);
- sb_r = bch2_sb_get_replicas(sb);
- if (!sb_r)
- return NULL;
+ entry = vstruct_next(entry);
+ }
- for_each_replicas_entry(sb_r, e) {
- err = "invalid replicas entry: invalid data type";
- if (e->data_type >= BCH_DATA_NR)
- goto err;
+ {
+ struct jset_entry_usage *u =
+ container_of(entry, struct jset_entry_usage, entry);
- err = "invalid replicas entry: too many devices";
- if (e->nr >= BCH_REPLICAS_MAX)
- goto err;
+ 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));
- err = "invalid replicas entry: invalid device";
- for (i = 0; i < e->nr; i++)
- if (!bch2_dev_exists(sb, mi, e->devs[i]))
- goto err;
+ entry = vstruct_next(entry);
}
- err = "cannot allocate memory";
- cpu_r = __bch2_sb_replicas_to_cpu_replicas(sb_r);
- if (!cpu_r)
- goto err;
+ 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]);
- sort_cmp_size(cpu_r->entries,
- cpu_r->nr,
- cpu_r->entry_size,
- memcmp, NULL);
+ entry = vstruct_next(entry);
+ }
- 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(entry, struct jset_entry_data_usage, entry);
- BUG_ON(memcmp(l, r, cpu_r->entry_size) > 0);
+ 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));
- err = "duplicate replicas entry";
- if (!memcmp(l, r, cpu_r->entry_size))
- goto err;
+ entry = vstruct_next(entry);
}
- err = NULL;
-err:
- kfree(cpu_r);
- return err;
+ percpu_up_write(&c->mark_lock);
+
+ return entry;
}
-int bch2_replicas_gc_end(struct bch_fs *c, int err)
+void bch2_fs_mark_clean(struct bch_fs *c)
{
- struct bch_sb_field_replicas *sb_r;
- struct bch_replicas_cpu *r, *old_r;
- struct bch_replicas_entry *dst_e;
- size_t i, j, bytes, dev_slots;
- int ret = 0;
-
- lockdep_assert_held(&c->replicas_gc_lock);
+ 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;
- r = rcu_dereference_protected(c->replicas_gc,
- lockdep_is_held(&c->sb_lock));
-
- if (err) {
- rcu_assign_pointer(c->replicas_gc, NULL);
- kfree_rcu(r, rcu);
- goto err;
- }
-
- dev_slots = replicas_dev_slots(r);
-
- bytes = sizeof(struct bch_sb_field_replicas);
+ SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
- for (i = 0; i < r->nr; i++) {
- struct bch_replicas_cpu_entry *e =
- cpu_replicas_entry(r, i);
+ c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
+ c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA;
+ c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_extents_above_btree_updates);
+ c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_btree_updates_journalled);
- bytes += sizeof(struct bch_replicas_entry);
- for (j = 0; j < r->entry_size - 1; j++)
- bytes += hweight8(e->devs[j]);
- }
+ u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
- sb_r = bch2_fs_sb_resize_replicas(c,
- DIV_ROUND_UP(sizeof(*sb_r) + bytes, sizeof(u64)));
- if (!sb_r) {
- ret = -ENOSPC;
- goto err;
+ 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;
}
- memset(&sb_r->entries, 0,
- vstruct_end(&sb_r->field) -
- (void *) &sb_r->entries);
+ 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);
- dst_e = sb_r->entries;
- for (i = 0; i < r->nr; i++) {
- struct bch_replicas_cpu_entry *src_e =
- cpu_replicas_entry(r, i);
+ /* Trying to catch outstanding bug: */
+ BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
- dst_e->data_type = src_e->data_type;
+ entry = sb_clean->start;
+ entry = bch2_journal_super_entries_add_common(c, entry, 0);
+ entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
+ BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
- for (j = 0; j < dev_slots; j++)
- if (replicas_test_dev(src_e, j))
- dst_e->devs[dst_e->nr++] = j;
+ memset(entry, 0,
+ vstruct_end(&sb_clean->field) - (void *) entry);
- dst_e = replicas_entry_next(dst_e);
- }
-
- old_r = rcu_dereference_protected(c->replicas,
- lockdep_is_held(&c->sb_lock));
- rcu_assign_pointer(c->replicas, r);
- rcu_assign_pointer(c->replicas_gc, NULL);
- kfree_rcu(old_r, rcu);
+ 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);
-err:
+out:
mutex_unlock(&c->sb_lock);
- return ret;
}
-int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
+static const char *bch2_sb_validate_clean(struct bch_sb *sb,
+ struct bch_sb_field *f)
{
- struct bch_replicas_cpu *r, *src;
- unsigned i;
+ struct bch_sb_field_clean *clean = field_to_type(f, clean);
- lockdep_assert_held(&c->replicas_gc_lock);
+ if (vstruct_bytes(&clean->field) < sizeof(*clean))
+ return "invalid field crypt: wrong size";
- mutex_lock(&c->sb_lock);
- BUG_ON(c->replicas_gc);
+ return NULL;
+}
- src = rcu_dereference_protected(c->replicas,
- lockdep_is_held(&c->sb_lock));
+static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
+ .validate = bch2_sb_validate_clean,
+};
- r = kzalloc(sizeof(struct bch_replicas_cpu) +
- src->nr * src->entry_size, GFP_NOIO);
- if (!r) {
- mutex_unlock(&c->sb_lock);
- return -ENOMEM;
- }
+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
+};
- r->entry_size = src->entry_size;
- r->nr = 0;
+static const char *bch2_sb_field_validate(struct bch_sb *sb,
+ struct bch_sb_field *f)
+{
+ unsigned type = le32_to_cpu(f->type);
- for (i = 0; i < src->nr; i++) {
- struct bch_replicas_cpu_entry *dst_e =
- cpu_replicas_entry(r, r->nr);
- struct bch_replicas_cpu_entry *src_e =
- cpu_replicas_entry(src, i);
+ return type < BCH_SB_FIELD_NR
+ ? bch2_sb_field_ops[type]->validate(sb, f)
+ : NULL;
+}
- if (!(src_e->data_type & typemask)) {
- memcpy(dst_e, src_e, r->entry_size);
- r->nr++;
- }
- }
+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;
- eytzinger0_sort(r->entries,
- r->nr,
- r->entry_size,
- memcmp, NULL);
+ if (ops)
+ pr_buf(out, "%s", bch2_sb_fields[type]);
+ else
+ pr_buf(out, "(unknown field %u)", type);
- rcu_assign_pointer(c->replicas_gc, r);
- mutex_unlock(&c->sb_lock);
+ pr_buf(out, " (size %llu):", vstruct_bytes(f));
- return 0;
+ if (ops && ops->to_text)
+ bch2_sb_field_ops[type]->to_text(out, sb, f);
}
projects / bcachefs-tools-debian / blobdiff