Quelle  super-io.c   Sprache: C

// SPDX-License-Identifier: GPL-2.0

#include "bcachefs.h"
#include "checksum.h"
#include "disk_groups.h"
#include "ec.h"
#include "error.h"
#include "journal.h"
#include "journal_sb.h"
#include "journal_seq_blacklist.h"
#include "recovery_passes.h"
#include "replicas.h"
#include "quota.h"
#include "sb-clean.h"
#include "sb-counters.h"
#include "sb-downgrade.h"
#include "sb-errors.h"
#include "sb-members.h"
#include "super-io.h"
#include "super.h"
#include "trace.h"
#include "vstructs.h"

#include <linux/backing-dev.h>
#include <linux/sort.h>
#include <linux/string_choices.h>

struct bch2_metadata_version {
 u16  version;
 const char *name;
};

static const struct bch2_metadata_version bch2_metadata_versions[] = {
#define x(n, v) {  \
 .version = v,    \
 .name = #n,    \
},
 BCH_METADATA_VERSIONS()
#undef x
};

void bch2_version_to_text(struct printbuf *out, enum bcachefs_metadata_version v)
{
 const char *str = "(unknown version)";

 for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
  if (bch2_metadata_versions[i].version == v) {
   str = bch2_metadata_versions[i].name;
   break;
  }

 prt_printf(out, "%u.%u: %s", BCH_VERSION_MAJOR(v), BCH_VERSION_MINOR(v), str);
}

enum bcachefs_metadata_version bch2_latest_compatible_version(enum bcachefs_metadata_version v)
{
 if (!BCH_VERSION_MAJOR(v))
  return v;

 for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
  if (bch2_metadata_versions[i].version > v &&
      BCH_VERSION_MAJOR(bch2_metadata_versions[i].version) ==
      BCH_VERSION_MAJOR(v))
   v = bch2_metadata_versions[i].version;

 return v;
}

int bch2_set_version_incompat(struct bch_fs *c, enum bcachefs_metadata_version version)
{
 int ret = ((c->sb.features & BIT_ULL(BCH_FEATURE_incompat_version_field)) &&
     version <= c->sb.version_incompat_allowed)
  ? 0
  : -BCH_ERR_may_not_use_incompat_feature;

 mutex_lock(&c->sb_lock);
 if (!ret) {
  SET_BCH_SB_VERSION_INCOMPAT(c->disk_sb.sb,
   max(BCH_SB_VERSION_INCOMPAT(c->disk_sb.sb), version));
  bch2_write_super(c);
 } else {
  darray_for_each(c->incompat_versions_requested, i)
   if (version == *i)
    goto out;

  darray_push(&c->incompat_versions_requested, version);
  struct printbuf buf = PRINTBUF;
  prt_str(&buf, "requested incompat feature ");
  bch2_version_to_text(&buf, version);
  prt_str(&buf, " currently not enabled, allowed up to ");
  bch2_version_to_text(&buf, version);
  prt_printf(&buf, "\n set version_upgrade=incompat to enable");

  bch_notice(c, "%s", buf.buf);
  printbuf_exit(&buf);
 }

out:
 mutex_unlock(&c->sb_lock);

 return ret;
}

const char * const bch2_sb_fields[] = {
#define x(name, nr) #name,
 BCH_SB_FIELDS()
#undef x
 NULL
};

static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
      enum bch_validate_flags, struct printbuf *);

struct bch_sb_field *bch2_sb_field_get_id(struct bch_sb *sb,
          enum bch_sb_field_type type)
{
 /* XXX: need locking around superblock to access optional fields */

 vstruct_for_each(sb, f)
  if (le32_to_cpu(f->type) == type)
   return f;
 return NULL;
}

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 && !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_id(sb->sb, type);

 if (f)
  __bch2_sb_field_resize(sb, f, 0);
}

/* Superblock realloc/free: */

void bch2_free_super(struct bch_sb_handle *sb)
{
 kfree(sb->bio);
 if (!IS_ERR_OR_NULL(sb->s_bdev_file))
  bdev_fput(sb->s_bdev_file);
 kfree(sb->holder);
 kfree(sb->sb_name);

 kfree(sb->sb);
 memset(sb, 0, sizeof(*sb));
}

int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
{
 size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
 size_t new_buffer_size;
 struct bch_sb *new_sb;
 struct bio *bio;

 if (sb->bdev)
  new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));

 new_buffer_size = roundup_pow_of_two(new_bytes);

 if (sb->sb && sb->buffer_size >= new_buffer_size)
  return 0;

 if (sb->sb && sb->have_layout) {
  u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;

  if (new_bytes > max_bytes) {
   struct printbuf buf = PRINTBUF;

   prt_bdevname(&buf, sb->bdev);
   prt_printf(&buf, ": superblock too big: want %zu but have %llu", new_bytes, max_bytes);
   pr_err("%s", buf.buf);
   printbuf_exit(&buf);
   return -BCH_ERR_ENOSPC_sb;
  }
 }

 if (sb->buffer_size >= new_buffer_size && sb->sb)
  return 0;

 if (dynamic_fault("bcachefs:add:super_realloc"))
  return -BCH_ERR_ENOMEM_sb_realloc_injected;

 new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
 if (!new_sb)
  return -BCH_ERR_ENOMEM_sb_buf_realloc;

 sb->sb = new_sb;

 if (sb->have_bio) {
  unsigned nr_bvecs = buf_pages(sb->sb, new_buffer_size);

  bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
  if (!bio)
   return -BCH_ERR_ENOMEM_sb_bio_realloc;

  bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);

  kfree(sb->bio);
  sb->bio = bio;
 }

 sb->buffer_size = new_buffer_size;

 return 0;
}

struct bch_sb_field *bch2_sb_field_resize_id(struct bch_sb_handle *sb,
       enum bch_sb_field_type type,
       unsigned u64s)
{
 struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
 ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 ssize_t d = -old_u64s + u64s;

 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
  return NULL;

 if (sb->fs_sb) {
  struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);

  lockdep_assert_held(&c->sb_lock);

  /* XXX: we're not checking that offline device have enough space */

  for_each_online_member(c, ca, BCH_DEV_READ_REF_sb_field_resize) {
   struct bch_sb_handle *dev_sb = &ca->disk_sb;

   if (bch2_sb_realloc(dev_sb, le32_to_cpu(dev_sb->sb->u64s) + d)) {
    enumerated_ref_put(&ca->io_ref[READ], BCH_DEV_READ_REF_sb_field_resize);
    return NULL;
   }
  }
 }

 f = bch2_sb_field_get_id(sb->sb, type);
 f = __bch2_sb_field_resize(sb, f, u64s);
 if (f)
  f->type = cpu_to_le32(type);
 return f;
}

struct bch_sb_field *bch2_sb_field_get_minsize_id(struct bch_sb_handle *sb,
        enum bch_sb_field_type type,
        unsigned u64s)
{
 struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);

 if (!f || le32_to_cpu(f->u64s) < u64s)
  f = bch2_sb_field_resize_id(sb, type, u64s);
 return f;
}

/* Superblock validate: */

static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
{
 u64 offset, prev_offset, max_sectors;
 unsigned i;

 BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);

 if (!uuid_equal(&layout->magic, &BCACHE_MAGIC) &&
     !uuid_equal(&layout->magic, &BCHFS_MAGIC)) {
  prt_printf(out, "Not a bcachefs superblock layout");
  return -BCH_ERR_invalid_sb_layout;
 }

 if (layout->layout_type != 0) {
  prt_printf(out, "Invalid superblock layout type %u",
         layout->layout_type);
  return -BCH_ERR_invalid_sb_layout_type;
 }

 if (!layout->nr_superblocks) {
  prt_printf(out, "Invalid superblock layout: no superblocks");
  return -BCH_ERR_invalid_sb_layout_nr_superblocks;
 }

 if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
  prt_printf(out, "Invalid superblock layout: too many superblocks");
  return -BCH_ERR_invalid_sb_layout_nr_superblocks;
 }

 if (layout->sb_max_size_bits > BCH_SB_LAYOUT_SIZE_BITS_MAX) {
  prt_printf(out, "Invalid superblock layout: max_size_bits too high");
  return -BCH_ERR_invalid_sb_layout_sb_max_size_bits;
 }

 max_sectors = 1 << layout->sb_max_size_bits;

 prev_offset = le64_to_cpu(layout->sb_offset[0]);

 for (i = 1; i < layout->nr_superblocks; i++) {
  offset = le64_to_cpu(layout->sb_offset[i]);

  if (offset < prev_offset + max_sectors) {
   prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
          " (sb %u ends at %llu next starts at %llu",
          i - 1, prev_offset + max_sectors, offset);
   return -BCH_ERR_invalid_sb_layout_superblocks_overlap;
  }
  prev_offset = offset;
 }

 return 0;
}

static int bch2_sb_compatible(struct bch_sb *sb, struct printbuf *out)
{
 u16 version  = le16_to_cpu(sb->version);
 u16 version_min  = le16_to_cpu(sb->version_min);

 if (!bch2_version_compatible(version)) {
  prt_str(out, "Unsupported superblock version ");
  bch2_version_to_text(out, version);
  prt_str(out, " (min ");
  bch2_version_to_text(out, bcachefs_metadata_version_min);
  prt_str(out, ", max ");
  bch2_version_to_text(out, bcachefs_metadata_version_current);
  prt_str(out, ")");
  return -BCH_ERR_invalid_sb_version;
 }

 if (!bch2_version_compatible(version_min)) {
  prt_str(out, "Unsupported superblock version_min ");
  bch2_version_to_text(out, version_min);
  prt_str(out, " (min ");
  bch2_version_to_text(out, bcachefs_metadata_version_min);
  prt_str(out, ", max ");
  bch2_version_to_text(out, bcachefs_metadata_version_current);
  prt_str(out, ")");
  return -BCH_ERR_invalid_sb_version;
 }

 if (version_min > version) {
  prt_str(out, "Bad minimum version ");
  bch2_version_to_text(out, version_min);
  prt_str(out, ", greater than version field ");
  bch2_version_to_text(out, version);
  return -BCH_ERR_invalid_sb_version;
 }

 return 0;
}

int bch2_sb_validate(struct bch_sb *sb, u64 read_offset,
       enum bch_validate_flags flags, struct printbuf *out)
{
 enum bch_opt_id opt_id;
 int ret;

 ret = bch2_sb_compatible(sb, out);
 if (ret)
  return ret;

 u64 incompat = le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR);
 unsigned incompat_bit = 0;
 if (incompat)
  incompat_bit = __ffs64(incompat);
 else if (sb->features[1])
  incompat_bit = 64 + __ffs64(le64_to_cpu(sb->features[1]));

 if (incompat_bit) {
  prt_printf(out, "Filesystem has incompatible feature bit %u, highest supported %s (%u)",
      incompat_bit,
      bch2_sb_features[BCH_FEATURE_NR - 1],
      BCH_FEATURE_NR - 1);
  return -BCH_ERR_invalid_sb_features;
 }

 if (BCH_VERSION_MAJOR(le16_to_cpu(sb->version)) > BCH_VERSION_MAJOR(bcachefs_metadata_version_current) ||
     BCH_SB_VERSION_INCOMPAT(sb) > bcachefs_metadata_version_current) {
  prt_str(out, "Filesystem has incompatible version ");
  bch2_version_to_text(out, le16_to_cpu(sb->version));
  prt_str(out, ", current version ");
  bch2_version_to_text(out, bcachefs_metadata_version_current);
  return -BCH_ERR_invalid_sb_features;
 }

 if (bch2_is_zero(sb->user_uuid.b, sizeof(sb->user_uuid))) {
  prt_printf(out, "Bad user UUID (got zeroes)");
  return -BCH_ERR_invalid_sb_uuid;
 }

 if (bch2_is_zero(sb->uuid.b, sizeof(sb->uuid))) {
  prt_printf(out, "Bad internal UUID (got zeroes)");
  return -BCH_ERR_invalid_sb_uuid;
 }

 if (!(flags & BCH_VALIDATE_write) &&
     le64_to_cpu(sb->offset) != read_offset) {
  prt_printf(out, "Bad sb offset (got %llu, read from %llu)",
      le64_to_cpu(sb->offset), read_offset);
  return -BCH_ERR_invalid_sb_offset;
 }

 if (!sb->nr_devices ||
     sb->nr_devices > BCH_SB_MEMBERS_MAX) {
  prt_printf(out, "Bad number of member devices %u (max %u)",
         sb->nr_devices, BCH_SB_MEMBERS_MAX);
  return -BCH_ERR_invalid_sb_too_many_members;
 }

 if (sb->dev_idx >= sb->nr_devices) {
  prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
         sb->dev_idx, sb->nr_devices);
  return -BCH_ERR_invalid_sb_dev_idx;
 }

 if (!sb->time_precision ||
     le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
  prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
         le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
  return -BCH_ERR_invalid_sb_time_precision;
 }

 /* old versions didn't know to downgrade this field */
 if (BCH_SB_VERSION_INCOMPAT_ALLOWED(sb) > le16_to_cpu(sb->version))
  SET_BCH_SB_VERSION_INCOMPAT_ALLOWED(sb, le16_to_cpu(sb->version));

 if (BCH_SB_VERSION_INCOMPAT(sb) > BCH_SB_VERSION_INCOMPAT_ALLOWED(sb)) {
  prt_printf(out, "Invalid version_incompat ");
  bch2_version_to_text(out, BCH_SB_VERSION_INCOMPAT(sb));
  prt_str(out, " > incompat_allowed ");
  bch2_version_to_text(out, BCH_SB_VERSION_INCOMPAT_ALLOWED(sb));
  if (flags & BCH_VALIDATE_write)
   return -BCH_ERR_invalid_sb_version;
  else
   SET_BCH_SB_VERSION_INCOMPAT_ALLOWED(sb, BCH_SB_VERSION_INCOMPAT(sb));
 }

 if (sb->nr_devices > 1)
  SET_BCH_SB_MULTI_DEVICE(sb, true);

 if (!flags) {
  /*
 * Been seeing a bug where these are getting inexplicably
 * zeroed, so we're now validating them, but we have to be
 * careful not to preven people's filesystems from mounting:
 */
  if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
   SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
  if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
   SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);

  if (!BCH_SB_VERSION_UPGRADE_COMPLETE(sb))
   SET_BCH_SB_VERSION_UPGRADE_COMPLETE(sb, le16_to_cpu(sb->version));

  if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_disk_accounting_v2 &&
      !BCH_SB_ALLOCATOR_STUCK_TIMEOUT(sb))
   SET_BCH_SB_ALLOCATOR_STUCK_TIMEOUT(sb, 30);

  if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_disk_accounting_v2)
   SET_BCH_SB_PROMOTE_WHOLE_EXTENTS(sb, true);

  if (!BCH_SB_WRITE_ERROR_TIMEOUT(sb))
   SET_BCH_SB_WRITE_ERROR_TIMEOUT(sb, 30);

  if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_extent_flags &&
      !BCH_SB_CSUM_ERR_RETRY_NR(sb))
   SET_BCH_SB_CSUM_ERR_RETRY_NR(sb, 3);
 }

#ifdef __KERNEL__
 if (!BCH_SB_SHARD_INUMS_NBITS(sb))
  SET_BCH_SB_SHARD_INUMS_NBITS(sb, ilog2(roundup_pow_of_two(num_online_cpus())));
#endif

 for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
  const struct bch_option *opt = bch2_opt_table + opt_id;

  if (opt->get_sb) {
   u64 v = bch2_opt_from_sb(sb, opt_id, -1);

   prt_printf(out, "Invalid option ");
   ret = bch2_opt_validate(opt, v, out);
   if (ret)
    return ret;

   printbuf_reset(out);
  }
 }

 /* validate layout */
 ret = validate_sb_layout(&sb->layout, out);
 if (ret)
  return ret;

 vstruct_for_each(sb, f) {
  if (!f->u64s) {
   prt_printf(out, "Invalid superblock: optional field with size 0 (type %u)",
          le32_to_cpu(f->type));
   return -BCH_ERR_invalid_sb_field_size;
  }

  if (vstruct_next(f) > vstruct_last(sb)) {
   prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
          le32_to_cpu(f->type));
   return -BCH_ERR_invalid_sb_field_size;
  }
 }

 struct bch_sb_field *mi =
  bch2_sb_field_get_id(sb, BCH_SB_FIELD_members_v2) ?:
  bch2_sb_field_get_id(sb, BCH_SB_FIELD_members_v1);

 /* members must be validated first: */
 if (!mi) {
  prt_printf(out, "Invalid superblock: member info area missing");
  return -BCH_ERR_invalid_sb_members_missing;
 }

 ret = bch2_sb_field_validate(sb, mi, flags, out);
 if (ret)
  return ret;

 vstruct_for_each(sb, f) {
  if (le32_to_cpu(f->type) == BCH_SB_FIELD_members_v1)
   continue;

  ret = bch2_sb_field_validate(sb, f, flags, out);
  if (ret)
   return ret;
 }

 if ((flags & BCH_VALIDATE_write) &&
     bch2_sb_member_get(sb, sb->dev_idx).seq != sb->seq) {
  prt_printf(out, "Invalid superblock: member seq %llu != sb seq %llu",
      le64_to_cpu(bch2_sb_member_get(sb, sb->dev_idx).seq),
      le64_to_cpu(sb->seq));
  return -BCH_ERR_invalid_sb_members_missing;
 }

 return 0;
}

/* device open: */

static unsigned long le_ulong_to_cpu(unsigned long v)
{
 return sizeof(unsigned long) == 8
  ? le64_to_cpu(v)
  : le32_to_cpu(v);
}

static void le_bitvector_to_cpu(unsigned long *dst, unsigned long *src, unsigned nr)
{
 BUG_ON(nr & (BITS_PER_TYPE(long) - 1));

 for (unsigned i = 0; i < BITS_TO_LONGS(nr); i++)
  dst[i] = le_ulong_to_cpu(src[i]);
}

static void bch2_sb_update(struct bch_fs *c)
{
 struct bch_sb *src = c->disk_sb.sb;

 lockdep_assert_held(&c->sb_lock);

 c->sb.uuid  = src->uuid;
 c->sb.user_uuid  = src->user_uuid;
 c->sb.version  = le16_to_cpu(src->version);
 c->sb.version_incompat = BCH_SB_VERSION_INCOMPAT(src);
 c->sb.version_incompat_allowed
    = BCH_SB_VERSION_INCOMPAT_ALLOWED(src);
 c->sb.version_min = le16_to_cpu(src->version_min);
 c->sb.version_upgrade_complete = BCH_SB_VERSION_UPGRADE_COMPLETE(src);
 c->sb.nr_devices = src->nr_devices;
 c->sb.clean  = BCH_SB_CLEAN(src);
 c->sb.encryption_type = BCH_SB_ENCRYPTION_TYPE(src);

 c->sb.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.features  = le64_to_cpu(src->features[0]);
 c->sb.compat  = le64_to_cpu(src->compat[0]);
 c->sb.multi_device = BCH_SB_MULTI_DEVICE(src);

 memset(c->sb.errors_silent, 0, sizeof(c->sb.errors_silent));

 struct bch_sb_field_ext *ext = bch2_sb_field_get(src, ext);
 if (ext) {
  c->sb.recovery_passes_required =
   bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0]));

  le_bitvector_to_cpu(c->sb.errors_silent, (void *) ext->errors_silent,
        sizeof(c->sb.errors_silent) * 8);
  c->sb.btrees_lost_data = le64_to_cpu(ext->btrees_lost_data);
 }

 for_each_member_device(c, ca) {
  struct bch_member m = bch2_sb_member_get(src, ca->dev_idx);
  ca->mi = bch2_mi_to_cpu(&m);
 }
}

static int __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
{
 struct bch_sb_field *src_f, *dst_f;
 struct bch_sb *dst = dst_handle->sb;
 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->label, src->label, sizeof(dst->label));

 dst->block_size  = src->block_size;
 dst->nr_devices  = src->nr_devices;

 dst->time_base_lo = src->time_base_lo;
 dst->time_base_hi = src->time_base_hi;
 dst->time_precision = src->time_precision;
 dst->write_time  = src->write_time;

 memcpy(dst->flags, src->flags, sizeof(dst->flags));
 memcpy(dst->features, src->features, sizeof(dst->features));
 memcpy(dst->compat, src->compat, sizeof(dst->compat));

 for (i = 0; i < BCH_SB_FIELD_NR; i++) {
  int d;

  if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
   continue;

  src_f = bch2_sb_field_get_id(src, i);
  dst_f = bch2_sb_field_get_id(dst, i);

  d = (src_f ? le32_to_cpu(src_f->u64s) : 0) -
      (dst_f ? le32_to_cpu(dst_f->u64s) : 0);
  if (d > 0) {
   int ret = bch2_sb_realloc(dst_handle,
     le32_to_cpu(dst_handle->sb->u64s) + d);

   if (ret)
    return ret;

   dst = dst_handle->sb;
   dst_f = bch2_sb_field_get_id(dst, i);
  }

  dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
    src_f ? le32_to_cpu(src_f->u64s) : 0);

  if (src_f)
   memcpy(dst_f, src_f, vstruct_bytes(src_f));
 }

 return 0;
}

int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
{
 int ret;

 lockdep_assert_held(&c->sb_lock);

 ret =   bch2_sb_realloc(&c->disk_sb, 0) ?:
  __copy_super(&c->disk_sb, src) ?:
  bch2_sb_replicas_to_cpu_replicas(c) ?:
  bch2_sb_disk_groups_to_cpu(c);
 if (ret)
  return ret;

 bch2_sb_update(c);
 return 0;
}

int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
{
 return __copy_super(&ca->disk_sb, c->disk_sb.sb);
}

/* read superblock: */

static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
{
 size_t bytes;
 int ret;
reread:
 bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 sb->bio->bi_iter.bi_sector = offset;
 bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);

 ret = submit_bio_wait(sb->bio);
 if (ret) {
  prt_printf(err, "IO error: %i", ret);
  return ret;
 }

 if (!uuid_equal(&sb->sb->magic, &BCACHE_MAGIC) &&
     !uuid_equal(&sb->sb->magic, &BCHFS_MAGIC)) {
  prt_str(err, "Not a bcachefs superblock (got magic ");
  pr_uuid(err, sb->sb->magic.b);
  prt_str(err, ")");
  return -BCH_ERR_invalid_sb_magic;
 }

 ret = bch2_sb_compatible(sb->sb, err);
 if (ret)
  return ret;

 bytes = vstruct_bytes(sb->sb);

 u64 sb_size = 512ULL << min(BCH_SB_LAYOUT_SIZE_BITS_MAX, sb->sb->layout.sb_max_size_bits);
 if (bytes > sb_size) {
  prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %llu)",
      bytes, sb_size);
  return -BCH_ERR_invalid_sb_too_big;
 }

 if (bytes > sb->buffer_size) {
  ret = bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s));
  if (ret)
   return ret;
  goto reread;
 }

 enum bch_csum_type csum_type = BCH_SB_CSUM_TYPE(sb->sb);
 if (csum_type >= BCH_CSUM_NR ||
     bch2_csum_type_is_encryption(csum_type)) {
  prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
  return -BCH_ERR_invalid_sb_csum_type;
 }

 /* XXX: verify MACs */
 struct bch_csum csum = csum_vstruct(NULL, csum_type, null_nonce(), sb->sb);
 if (bch2_crc_cmp(csum, sb->sb->csum)) {
  bch2_csum_err_msg(err, csum_type, sb->sb->csum, csum);
  return -BCH_ERR_invalid_sb_csum;
 }

 sb->seq = le64_to_cpu(sb->sb->seq);

 return 0;
}

static int __bch2_read_super(const char *path, struct bch_opts *opts,
      struct bch_sb_handle *sb, bool ignore_notbchfs_msg)
{
 u64 offset = opt_get(*opts, sb);
 struct bch_sb_layout layout;
 struct printbuf err = PRINTBUF;
 struct printbuf err2 = PRINTBUF;
 __le64 *i;
 int ret;
#ifndef __KERNEL__
retry:
#endif
 memset(sb, 0, sizeof(*sb));
 sb->mode = BLK_OPEN_READ;
 sb->have_bio = true;
 sb->holder = kzalloc(sizeof(*sb->holder), GFP_KERNEL);
 if (!sb->holder)
  return -ENOMEM;

 sb->sb_name = kstrdup(path, GFP_KERNEL);
 if (!sb->sb_name) {
  ret = -ENOMEM;
  prt_printf(&err, "error allocating memory for sb_name");
  goto err;
 }

#ifndef __KERNEL__
 if (opt_get(*opts, direct_io) == false)
  sb->mode |= BLK_OPEN_BUFFERED;
#endif

 if (!opt_get(*opts, noexcl))
  sb->mode |= BLK_OPEN_EXCL;

 if (!opt_get(*opts, nochanges))
  sb->mode |= BLK_OPEN_WRITE;

 sb->s_bdev_file = bdev_file_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
 if (IS_ERR(sb->s_bdev_file) &&
     PTR_ERR(sb->s_bdev_file) == -EACCES &&
     opt_get(*opts, read_only)) {
  sb->mode &= ~BLK_OPEN_WRITE;

  sb->s_bdev_file = bdev_file_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
  if (!IS_ERR(sb->s_bdev_file))
   opt_set(*opts, nochanges, true);
 }

 if (IS_ERR(sb->s_bdev_file)) {
  ret = PTR_ERR(sb->s_bdev_file);
  prt_printf(&err, "error opening %s: %s", path, bch2_err_str(ret));
  goto err;
 }
 sb->bdev = file_bdev(sb->s_bdev_file);

 ret = bch2_sb_realloc(sb, 0);
 if (ret) {
  prt_printf(&err, "error allocating memory for superblock");
  goto err;
 }

 if (bch2_fs_init_fault("read_super")) {
  prt_printf(&err, "dynamic fault");
  ret = -EFAULT;
  goto err;
 }

 ret = read_one_super(sb, offset, &err);
 if (!ret)
  goto got_super;

 if (opt_defined(*opts, sb))
  goto err;

 prt_printf(&err2, "bcachefs (%s): error reading default superblock: %s\n",
        path, err.buf);
 if (ret == -BCH_ERR_invalid_sb_magic && ignore_notbchfs_msg)
  bch2_print_opts(opts, KERN_INFO "%s", err2.buf);
 else
  bch2_print_opts(opts, KERN_ERR "%s", err2.buf);

 printbuf_exit(&err2);
 printbuf_reset(&err);

 /*
 * Error reading primary superblock - read location of backup
 * superblocks:
 */
 bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 /*
 * use sb buffer to read layout, since sb buffer is page aligned but
 * layout won't be:
 */
 bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));

 ret = submit_bio_wait(sb->bio);
 if (ret) {
  prt_printf(&err, "IO error: %i", ret);
  goto err;
 }

 memcpy(&layout, sb->sb, sizeof(layout));
 ret = validate_sb_layout(&layout, &err);
 if (ret)
  goto err;

 for (i = layout.sb_offset;
      i < layout.sb_offset + layout.nr_superblocks; i++) {
  offset = le64_to_cpu(*i);

  if (offset == opt_get(*opts, sb)) {
   ret = -BCH_ERR_invalid;
   continue;
  }

  ret = read_one_super(sb, offset, &err);
  if (!ret)
   goto got_super;
 }

 goto err;

got_super:
 if (le16_to_cpu(sb->sb->block_size) << 9 <
     bdev_logical_block_size(sb->bdev) &&
     opt_get(*opts, direct_io)) {
#ifndef __KERNEL__
  opt_set(*opts, direct_io, false);
  bch2_free_super(sb);
  goto retry;
#endif
  prt_printf(&err, "block size (%u) smaller than device block size (%u)",
         le16_to_cpu(sb->sb->block_size) << 9,
         bdev_logical_block_size(sb->bdev));
  ret = -BCH_ERR_block_size_too_small;
  goto err;
 }

 sb->have_layout = true;

 ret = bch2_sb_validate(sb->sb, offset, 0, &err);
 if (ret) {
  bch2_print_opts(opts, KERN_ERR "bcachefs (%s): error validating superblock: %s\n",
    path, err.buf);
  goto err_no_print;
 }
out:
 printbuf_exit(&err);
 return ret;
err:
 bch2_print_opts(opts, KERN_ERR "bcachefs (%s): error reading superblock: %s\n",
   path, err.buf);
err_no_print:
 bch2_free_super(sb);
 goto out;
}

int bch2_read_super(const char *path, struct bch_opts *opts,
      struct bch_sb_handle *sb)
{
 return __bch2_read_super(path, opts, sb, false);
}

/* provide a silenced version for mount.bcachefs */

int bch2_read_super_silent(const char *path, struct bch_opts *opts,
      struct bch_sb_handle *sb)
{
 return __bch2_read_super(path, opts, sb, true);
}

/* write superblock: */

static void write_super_endio(struct bio *bio)
{
 struct bch_dev *ca = bio->bi_private;

 bch2_account_io_success_fail(ca, bio_data_dir(bio), !bio->bi_status);

 /* XXX: return errors directly */

 if (bio->bi_status) {
  bch_err_dev_ratelimited(ca, "superblock %s error: %s",
          str_write_read(bio_data_dir(bio)),
          bch2_blk_status_to_str(bio->bi_status));
  ca->sb_write_error = 1;
 }

 closure_put(&ca->fs->sb_write);
 enumerated_ref_put(&ca->io_ref[READ], BCH_DEV_READ_REF_write_super);
}

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;

 memset(ca->sb_read_scratch, 0, BCH_SB_READ_SCRATCH_BUF_SIZE);

 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 bio->bi_iter.bi_sector = le64_to_cpu(sb->layout.sb_offset[0]);
 bio->bi_end_io  = write_super_endio;
 bio->bi_private  = ca;
 bch2_bio_map(bio, ca->sb_read_scratch, BCH_SB_READ_SCRATCH_BUF_SIZE);

 this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb], bio_sectors(bio));

 enumerated_ref_get(&ca->io_ref[READ], BCH_DEV_READ_REF_write_super);
 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;
 struct bio *bio = ca->disk_sb.bio;

 sb->offset = sb->layout.sb_offset[idx];

 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, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
 bio->bi_iter.bi_sector = le64_to_cpu(sb->offset);
 bio->bi_end_io  = write_super_endio;
 bio->bi_private  = ca;
 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));

 enumerated_ref_get(&ca->io_ref[READ], BCH_DEV_READ_REF_write_super);
 closure_bio_submit(bio, &c->sb_write);
}

int bch2_write_super(struct bch_fs *c)
{
 struct closure *cl = &c->sb_write;
 struct printbuf err = PRINTBUF;
 unsigned sb = 0, nr_wrote;
 struct bch_devs_mask sb_written;
 bool wrote, can_mount_without_written, can_mount_with_written;
 unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 DARRAY(struct bch_dev *) online_devices = {};
 int ret = 0;

 trace_and_count(c, write_super, c, _RET_IP_);

 if (c->opts.degraded == BCH_DEGRADED_very)
  degraded_flags |= BCH_FORCE_IF_LOST;

 lockdep_assert_held(&c->sb_lock);

 closure_init_stack(cl);
 memset(&sb_written, 0, sizeof(sb_written));

 /*
 * Note: we do writes to RO devices here, and we might want to change
 * that in the future.
 *
 * For now, we expect to be able to call write_super() when we're not
 * yet RW:
 */
 for_each_online_member(c, ca, BCH_DEV_READ_REF_write_super) {
  ret = darray_push(&online_devices, ca);
  if (bch2_fs_fatal_err_on(ret, c, "%s: error allocating online devices", __func__)) {
   enumerated_ref_put(&ca->io_ref[READ], BCH_DEV_READ_REF_write_super);
   goto out;
  }
  enumerated_ref_get(&ca->io_ref[READ], BCH_DEV_READ_REF_write_super);
 }

 /* Make sure we're using the new magic numbers: */
 c->disk_sb.sb->magic = BCHFS_MAGIC;
 c->disk_sb.sb->layout.magic = BCHFS_MAGIC;

 le64_add_cpu(&c->disk_sb.sb->seq, 1);

 struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
 darray_for_each(online_devices, ca)
  __bch2_members_v2_get_mut(mi, (*ca)->dev_idx)->seq = c->disk_sb.sb->seq;
 c->disk_sb.sb->write_time = cpu_to_le64(ktime_get_real_seconds());

 if (test_bit(BCH_FS_error, &c->flags))
  SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 if (test_bit(BCH_FS_topology_error, &c->flags))
  SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);

 SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);

 bch2_sb_counters_from_cpu(c);
 bch2_sb_members_from_cpu(c);
 bch2_sb_members_cpy_v2_v1(&c->disk_sb);
 bch2_sb_errors_from_cpu(c);
 bch2_sb_downgrade_update(c);

 darray_for_each(online_devices, ca)
  bch2_sb_from_fs(c, (*ca));

 darray_for_each(online_devices, ca) {
  printbuf_reset(&err);

  ret = bch2_sb_validate((*ca)->disk_sb.sb, 0, BCH_VALIDATE_write, &err);
  if (ret) {
   bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
   goto out;
  }
 }

 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;

 if (le16_to_cpu(c->disk_sb.sb->version) > bcachefs_metadata_version_current) {
  struct printbuf buf = PRINTBUF;
  prt_printf(&buf, "attempting to write superblock that wasn't version downgraded (");
  bch2_version_to_text(&buf, le16_to_cpu(c->disk_sb.sb->version));
  prt_str(&buf, " > ");
  bch2_version_to_text(&buf, bcachefs_metadata_version_current);
  prt_str(&buf, ")");
  bch2_fs_fatal_error(c, ": %s", buf.buf);
  printbuf_exit(&buf);
  ret = bch_err_throw(c, sb_not_downgraded);
  goto out;
 }

 darray_for_each(online_devices, ca) {
  __set_bit((*ca)->dev_idx, sb_written.d);
  (*ca)->sb_write_error = 0;
 }

 darray_for_each(online_devices, ca)
  read_back_super(c, *ca);
 closure_sync(cl);

 darray_for_each(online_devices, cap) {
  struct bch_dev *ca = *cap;

  if (ca->sb_write_error)
   continue;

  if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
   struct printbuf buf = PRINTBUF;
   prt_char(&buf, ' ');
   prt_bdevname(&buf, ca->disk_sb.bdev);
   prt_printf(&buf,
    ": Superblock write was silently dropped! (seq %llu expected %llu)",
    le64_to_cpu(ca->sb_read_scratch->seq),
    ca->disk_sb.seq);

   if (c->opts.errors != BCH_ON_ERROR_continue &&
       c->opts.errors != BCH_ON_ERROR_fix_safe) {
    ret = bch_err_throw(c, erofs_sb_err);
    bch2_fs_fatal_error(c, "%s", buf.buf);
   } else {
    bch_err(c, "%s", buf.buf);
   }

   printbuf_exit(&buf);
  }

  if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
   struct printbuf buf = PRINTBUF;
   prt_char(&buf, ' ');
   prt_bdevname(&buf, ca->disk_sb.bdev);
   prt_printf(&buf,
    ": Superblock modified by another process (seq %llu expected %llu)",
    le64_to_cpu(ca->sb_read_scratch->seq),
    ca->disk_sb.seq);
   bch2_fs_fatal_error(c, "%s", buf.buf);
   printbuf_exit(&buf);
   ret = bch_err_throw(c, erofs_sb_err);
  }
 }

 if (ret)
  goto out;

 do {
  wrote = false;
  darray_for_each(online_devices, cap) {
   struct bch_dev *ca = *cap;
   if (!ca->sb_write_error &&
       sb < ca->disk_sb.sb->layout.nr_superblocks) {
    write_one_super(c, ca, sb);
    wrote = true;
   }
  }
  closure_sync(cl);
  sb++;
 } while (wrote);

 darray_for_each(online_devices, cap) {
  struct bch_dev *ca = *cap;
  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(c, sb_written, degraded_flags, false);

 for (unsigned i = 0; i < ARRAY_SIZE(sb_written.d); i++)
  sb_written.d[i] = ~sb_written.d[i];

 can_mount_without_written =
  bch2_have_enough_devs(c, sb_written, degraded_flags, false);

 /*
 * If we would be able to mount _without_ the devices we successfully
 * wrote superblocks to, we weren't able to write to enough devices:
 *
 * Exception: if we can mount without the successes because we haven't
 * written anything (new filesystem), we continue if we'd be able to
 * mount with the devices we did successfully write to:
 */
 if (bch2_fs_fatal_err_on(!nr_wrote ||
     !can_mount_with_written ||
     (can_mount_without_written &&
      !can_mount_with_written), c,
  ": Unable to write superblock to sufficient devices (from %ps)",
  (void *) _RET_IP_))
  ret = bch_err_throw(c, erofs_sb_err);
out:
 /* Make new options visible after they're persistent: */
 bch2_sb_update(c);
 darray_for_each(online_devices, ca)
  enumerated_ref_put(&(*ca)->io_ref[READ], BCH_DEV_READ_REF_write_super);
 darray_exit(&online_devices);
 printbuf_exit(&err);
 return ret;
}

void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
{
 mutex_lock(&c->sb_lock);
 if (!(c->sb.features & (1ULL << feat))) {
  c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);

  bch2_write_super(c);
 }
 mutex_unlock(&c->sb_lock);
}

/* Downgrade if superblock is at a higher version than currently supported: */
bool bch2_check_version_downgrade(struct bch_fs *c)
{
 bool ret = bcachefs_metadata_version_current < c->sb.version;

 lockdep_assert_held(&c->sb_lock);

 /*
 * Downgrade, if superblock is at a higher version than currently
 * supported:
 *
 * c->sb will be checked before we write the superblock, so update it as
 * well:
 */
 if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) > bcachefs_metadata_version_current)
  SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
 if (BCH_SB_VERSION_INCOMPAT_ALLOWED(c->disk_sb.sb) > bcachefs_metadata_version_current)
  SET_BCH_SB_VERSION_INCOMPAT_ALLOWED(c->disk_sb.sb, bcachefs_metadata_version_current);
 if (c->sb.version > bcachefs_metadata_version_current)
  c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
 if (c->sb.version_min > bcachefs_metadata_version_current)
  c->disk_sb.sb->version_min = cpu_to_le16(bcachefs_metadata_version_current);
 c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
 return ret;
}

void bch2_sb_upgrade(struct bch_fs *c, unsigned new_version, bool incompat)
{
 lockdep_assert_held(&c->sb_lock);

 if (BCH_VERSION_MAJOR(new_version) >
     BCH_VERSION_MAJOR(le16_to_cpu(c->disk_sb.sb->version)))
  bch2_sb_field_resize(&c->disk_sb, downgrade, 0);

 c->disk_sb.sb->version = cpu_to_le16(new_version);

 if (incompat) {
  c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
  SET_BCH_SB_VERSION_INCOMPAT_ALLOWED(c->disk_sb.sb,
   max(BCH_SB_VERSION_INCOMPAT_ALLOWED(c->disk_sb.sb), new_version));
 }
}

void bch2_sb_upgrade_incompat(struct bch_fs *c)
{
 mutex_lock(&c->sb_lock);
 if (c->sb.version == c->sb.version_incompat_allowed)
  goto unlock;

 struct printbuf buf = PRINTBUF;

 prt_str(&buf, "Now allowing incompatible features up to ");
 bch2_version_to_text(&buf, c->sb.version);
 prt_str(&buf, ", previously allowed up to ");
 bch2_version_to_text(&buf, c->sb.version_incompat_allowed);
 prt_newline(&buf);

 bch_notice(c, "%s", buf.buf);
 printbuf_exit(&buf);

 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
 SET_BCH_SB_VERSION_INCOMPAT_ALLOWED(c->disk_sb.sb,
   max(BCH_SB_VERSION_INCOMPAT_ALLOWED(c->disk_sb.sb), c->sb.version));
 bch2_write_super(c);
unlock:
 mutex_unlock(&c->sb_lock);
}

static int bch2_sb_ext_validate(struct bch_sb *sb, struct bch_sb_field *f,
    enum bch_validate_flags flags, struct printbuf *err)
{
 if (vstruct_bytes(f) < 88) {
  prt_printf(err, "field too small (%zu < %u)", vstruct_bytes(f), 88);
  return -BCH_ERR_invalid_sb_ext;
 }

 return 0;
}

static void bch2_sb_ext_to_text(struct printbuf *out, struct bch_sb *sb,
    struct bch_sb_field *f)
{
 struct bch_sb_field_ext *e = field_to_type(f, ext);

 prt_printf(out, "Recovery passes required:\t");
 prt_bitflags(out, bch2_recovery_passes,
       bch2_recovery_passes_from_stable(le64_to_cpu(e->recovery_passes_required[0])));
 prt_newline(out);

 unsigned long *errors_silent = kmalloc(sizeof(e->errors_silent), GFP_KERNEL);
 if (errors_silent) {
  le_bitvector_to_cpu(errors_silent, (void *) e->errors_silent, sizeof(e->errors_silent) * 8);

  prt_printf(out, "Errors to silently fix:\t");
  prt_bitflags_vector(out, bch2_sb_error_strs, errors_silent,
        min(BCH_FSCK_ERR_MAX, sizeof(e->errors_silent) * 8));
  prt_newline(out);

  kfree(errors_silent);
 }

 prt_printf(out, "Btrees with missing data:\t");
 prt_bitflags(out, __bch2_btree_ids, le64_to_cpu(e->btrees_lost_data));
 prt_newline(out);
}

static const struct bch_sb_field_ops bch_sb_field_ops_ext = {
 .validate = bch2_sb_ext_validate,
 .to_text = bch2_sb_ext_to_text,
};

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
};

static const struct bch_sb_field_ops bch2_sb_field_null_ops;

static const struct bch_sb_field_ops *bch2_sb_field_type_ops(unsigned type)
{
 return likely(type < ARRAY_SIZE(bch2_sb_field_ops))
  ? bch2_sb_field_ops[type]
  : &bch2_sb_field_null_ops;
}

static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
      enum bch_validate_flags flags, struct printbuf *err)
{
 unsigned type = le32_to_cpu(f->type);
 struct printbuf field_err = PRINTBUF;
 const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
 int ret;

 ret = ops->validate ? ops->validate(sb, f, flags, &field_err) : 0;
 if (ret) {
  prt_printf(err, "Invalid superblock section %s: %s",
      bch2_sb_fields[type], field_err.buf);
  prt_newline(err);
  bch2_sb_field_to_text(err, sb, f);
 }

 printbuf_exit(&field_err);
 return ret;
}

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 = bch2_sb_field_type_ops(type);

 if (!out->nr_tabstops)
  printbuf_tabstop_push(out, 32);

 if (ops->to_text)
  ops->to_text(out, sb, f);
}

void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
      struct bch_sb_field *f)
{
 unsigned type = le32_to_cpu(f->type);

 if (type < BCH_SB_FIELD_NR)
  prt_printf(out, "%s", bch2_sb_fields[type]);
 else
  prt_printf(out, "(unknown field %u)", type);

 prt_printf(out, " (size %zu):", vstruct_bytes(f));
 prt_newline(out);

 __bch2_sb_field_to_text(out, sb, f);
}

void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
{
 unsigned i;

 prt_printf(out, "Type: %u", l->layout_type);
 prt_newline(out);

 prt_str(out, "Superblock max size: ");
 prt_units_u64(out, 512 << l->sb_max_size_bits);
 prt_newline(out);

 prt_printf(out, "Nr superblocks: %u", l->nr_superblocks);
 prt_newline(out);

 prt_str(out, "Offsets: ");
 for (i = 0; i < l->nr_superblocks; i++) {
  if (i)
   prt_str(out, ", ");
  prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
 }
 prt_newline(out);
}

void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
       bool print_layout, unsigned fields)
{
 if (!out->nr_tabstops)
  printbuf_tabstop_push(out, 44);

 prt_printf(out, "External UUID:\t");
 pr_uuid(out, sb->user_uuid.b);
 prt_newline(out);

 prt_printf(out, "Internal UUID:\t");
 pr_uuid(out, sb->uuid.b);
 prt_newline(out);

 prt_printf(out, "Magic number:\t");
 pr_uuid(out, sb->magic.b);
 prt_newline(out);

 prt_printf(out, "Device index:\t%u\n", sb->dev_idx);

 prt_printf(out, "Label:\t");
 if (!strlen(sb->label))
  prt_printf(out, "(none)");
 else
  prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
 prt_newline(out);

 prt_printf(out, "Version:\t");
 bch2_version_to_text(out, le16_to_cpu(sb->version));
 prt_newline(out);

 prt_printf(out, "Incompatible features allowed:\t");
 bch2_version_to_text(out, BCH_SB_VERSION_INCOMPAT_ALLOWED(sb));
 prt_newline(out);

 prt_printf(out, "Incompatible features in use:\t");
 bch2_version_to_text(out, BCH_SB_VERSION_INCOMPAT(sb));
 prt_newline(out);

 prt_printf(out, "Version upgrade complete:\t");
 bch2_version_to_text(out, BCH_SB_VERSION_UPGRADE_COMPLETE(sb));
 prt_newline(out);

 prt_printf(out, "Oldest version on disk:\t");
 bch2_version_to_text(out, le16_to_cpu(sb->version_min));
 prt_newline(out);

 prt_printf(out, "Created:\t");
 if (sb->time_base_lo)
  bch2_prt_datetime(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
 else
  prt_printf(out, "(not set)");
 prt_newline(out);

 prt_printf(out, "Sequence number:\t");
 prt_printf(out, "%llu", le64_to_cpu(sb->seq));
 prt_newline(out);

 prt_printf(out, "Time of last write:\t");
 bch2_prt_datetime(out, le64_to_cpu(sb->write_time));
 prt_newline(out);

 prt_printf(out, "Superblock size:\t");
 prt_units_u64(out, vstruct_bytes(sb));
 prt_str(out, "/");
 prt_units_u64(out, 512ULL << sb->layout.sb_max_size_bits);
 prt_newline(out);

 prt_printf(out, "Clean:\t%llu\n", BCH_SB_CLEAN(sb));
 prt_printf(out, "Devices:\t%u\n", bch2_sb_nr_devices(sb));

 prt_printf(out, "Sections:\t");
 u64 fields_have = 0;
 vstruct_for_each(sb, f)
  fields_have |= 1 << le32_to_cpu(f->type);
 prt_bitflags(out, bch2_sb_fields, fields_have);
 prt_newline(out);

 prt_printf(out, "Features:\t");
 prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
 prt_newline(out);

 prt_printf(out, "Compat features:\t");
 prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
 prt_newline(out);

 prt_newline(out);
 prt_printf(out, "Options:");
 prt_newline(out);
 printbuf_indent_add(out, 2);
 {
  enum bch_opt_id id;

  for (id = 0; id < bch2_opts_nr; id++) {
   const struct bch_option *opt = bch2_opt_table + id;

   if (opt->get_sb) {
    u64 v = bch2_opt_from_sb(sb, id, -1);

    prt_printf(out, "%s:\t", opt->attr.name);
    bch2_opt_to_text(out, NULL, sb, opt, v,
       OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
    prt_newline(out);
   }
  }
 }

 printbuf_indent_sub(out, 2);

 if (print_layout) {
  prt_newline(out);
  prt_printf(out, "layout:");
  prt_newline(out);
  printbuf_indent_add(out, 2);
  bch2_sb_layout_to_text(out, &sb->layout);
  printbuf_indent_sub(out, 2);
 }

 vstruct_for_each(sb, f)
  if (fields & (1 << le32_to_cpu(f->type))) {
   prt_newline(out);
   bch2_sb_field_to_text(out, sb, f);
  }
}