Quelle  sb-members.c   Sprache: C

// SPDX-License-Identifier: GPL-2.0

#include "bcachefs.h"
#include "btree_cache.h"
#include "disk_groups.h"
#include "error.h"
#include "opts.h"
#include "recovery_passes.h"
#include "replicas.h"
#include "sb-members.h"
#include "super-io.h"

int bch2_dev_missing_bkey(struct bch_fs *c, struct bkey_s_c k, unsigned dev)
{
 struct printbuf buf = PRINTBUF;
 bch2_log_msg_start(c, &buf);

 prt_printf(&buf, "pointer to nonexistent device %u in key\n", dev);
 bch2_bkey_val_to_text(&buf, c, k);

 bool print = bch2_count_fsck_err(c, ptr_to_invalid_device, &buf);

 int ret = bch2_run_explicit_recovery_pass(c, &buf,
     BCH_RECOVERY_PASS_check_allocations, 0);

 if (print)
  bch2_print_str(c, KERN_ERR, buf.buf);
 printbuf_exit(&buf);
 return ret;
}

void bch2_dev_missing_atomic(struct bch_fs *c, unsigned dev)
{
 if (dev != BCH_SB_MEMBER_INVALID)
  bch2_fs_inconsistent(c, "pointer to nonexistent device %u", dev);
}

void bch2_dev_bucket_missing(struct bch_dev *ca, u64 bucket)
{
 bch2_fs_inconsistent(ca->fs,
  "pointer to nonexistent bucket %llu on device %s (valid range %u-%llu)",
  bucket, ca->name, ca->mi.first_bucket, ca->mi.nbuckets);
}

#define x(t, n, ...) [n] = #t,
static const char * const bch2_iops_measurements[] = {
 BCH_IOPS_MEASUREMENTS()
 NULL
};

char * const bch2_member_error_strs[] = {
 BCH_MEMBER_ERROR_TYPES()
 NULL
};
#undef x

/* Code for bch_sb_field_members_v1: */

struct bch_member *bch2_members_v2_get_mut(struct bch_sb *sb, int i)
{
 return __bch2_members_v2_get_mut(bch2_sb_field_get(sb, members_v2), i);
}

static struct bch_member members_v2_get(struct bch_sb_field_members_v2 *mi, int i)
{
 struct bch_member ret, *p = __bch2_members_v2_get_mut(mi, i);
 memset(&ret, 0, sizeof(ret));
 memcpy(&ret, p, min_t(size_t, le16_to_cpu(mi->member_bytes), sizeof(ret)));
 return ret;
}

static struct bch_member *members_v1_get_mut(struct bch_sb_field_members_v1 *mi, int i)
{
 return (void *) mi->_members + (i * BCH_MEMBER_V1_BYTES);
}

static struct bch_member members_v1_get(struct bch_sb_field_members_v1 *mi, int i)
{
 struct bch_member ret, *p = members_v1_get_mut(mi, i);
 memset(&ret, 0, sizeof(ret));
 memcpy(&ret, p, min_t(size_t, BCH_MEMBER_V1_BYTES, sizeof(ret)));
 return ret;
}

struct bch_member bch2_sb_member_get(struct bch_sb *sb, int i)
{
 struct bch_sb_field_members_v2 *mi2 = bch2_sb_field_get(sb, members_v2);
 if (mi2)
  return members_v2_get(mi2, i);
 struct bch_sb_field_members_v1 *mi1 = bch2_sb_field_get(sb, members_v1);
 return members_v1_get(mi1, i);
}

static int sb_members_v2_resize_entries(struct bch_fs *c)
{
 struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);

 if (le16_to_cpu(mi->member_bytes) < sizeof(struct bch_member)) {
  unsigned u64s = DIV_ROUND_UP((sizeof(*mi) + sizeof(mi->_members[0]) *
           c->disk_sb.sb->nr_devices), 8);

  mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
  if (!mi)
   return bch_err_throw(c, ENOSPC_sb_members_v2);

  for (int i = c->disk_sb.sb->nr_devices - 1; i >= 0; --i) {
   void *dst = (void *) mi->_members + (i * sizeof(struct bch_member));
   memmove(dst, __bch2_members_v2_get_mut(mi, i), le16_to_cpu(mi->member_bytes));
   memset(dst + le16_to_cpu(mi->member_bytes),
          0, (sizeof(struct bch_member) - le16_to_cpu(mi->member_bytes)));
  }
  mi->member_bytes = cpu_to_le16(sizeof(struct bch_member));
 }
 return 0;
}

int bch2_sb_members_v2_init(struct bch_fs *c)
{
 struct bch_sb_field_members_v1 *mi1;
 struct bch_sb_field_members_v2 *mi2;

 if (!bch2_sb_field_get(c->disk_sb.sb, members_v2)) {
  mi2 = bch2_sb_field_resize(&c->disk_sb, members_v2,
    DIV_ROUND_UP(sizeof(*mi2) +
          sizeof(struct bch_member) * c->sb.nr_devices,
          sizeof(u64)));
  mi1 = bch2_sb_field_get(c->disk_sb.sb, members_v1);
  memcpy(&mi2->_members[0], &mi1->_members[0],
         BCH_MEMBER_V1_BYTES * c->sb.nr_devices);
  memset(&mi2->pad[0], 0, sizeof(mi2->pad));
  mi2->member_bytes = cpu_to_le16(BCH_MEMBER_V1_BYTES);
 }

 return sb_members_v2_resize_entries(c);
}

int bch2_sb_members_cpy_v2_v1(struct bch_sb_handle *disk_sb)
{
 struct bch_sb_field_members_v1 *mi1;
 struct bch_sb_field_members_v2 *mi2;

 if (BCH_SB_VERSION_INCOMPAT(disk_sb->sb) > bcachefs_metadata_version_extent_flags) {
  bch2_sb_field_resize(disk_sb, members_v1, 0);
  return 0;
 }

 mi1 = bch2_sb_field_resize(disk_sb, members_v1,
   DIV_ROUND_UP(sizeof(*mi1) + BCH_MEMBER_V1_BYTES *
         disk_sb->sb->nr_devices, sizeof(u64)));
 if (!mi1)
  return -BCH_ERR_ENOSPC_sb_members;

 mi2 = bch2_sb_field_get(disk_sb->sb, members_v2);

 for (unsigned i = 0; i < disk_sb->sb->nr_devices; i++)
  memcpy(members_v1_get_mut(mi1, i), __bch2_members_v2_get_mut(mi2, i), BCH_MEMBER_V1_BYTES);

 return 0;
}

static int validate_member(struct printbuf *err,
      struct bch_member m,
      struct bch_sb *sb,
      int i)
{
 if (le64_to_cpu(m.nbuckets) > BCH_MEMBER_NBUCKETS_MAX) {
  prt_printf(err, "device %u: too many buckets (got %llu, max %u)",
      i, le64_to_cpu(m.nbuckets), BCH_MEMBER_NBUCKETS_MAX);
  return -BCH_ERR_invalid_sb_members;
 }

 if (le64_to_cpu(m.nbuckets) -
     le16_to_cpu(m.first_bucket) < BCH_MIN_NR_NBUCKETS) {
  prt_printf(err, "device %u: not enough buckets (got %llu, max %u)",
      i, le64_to_cpu(m.nbuckets), BCH_MIN_NR_NBUCKETS);
  return -BCH_ERR_invalid_sb_members;
 }

 if (le16_to_cpu(m.bucket_size) <
     le16_to_cpu(sb->block_size)) {
  prt_printf(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 -BCH_ERR_invalid_sb_members;
 }

 if (le16_to_cpu(m.bucket_size) <
     BCH_SB_BTREE_NODE_SIZE(sb)) {
  prt_printf(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 -BCH_ERR_invalid_sb_members;
 }

 if (m.btree_bitmap_shift >= BCH_MI_BTREE_BITMAP_SHIFT_MAX) {
  prt_printf(err, "device %u: invalid btree_bitmap_shift %u", i, m.btree_bitmap_shift);
  return -BCH_ERR_invalid_sb_members;
 }

 if (BCH_MEMBER_FREESPACE_INITIALIZED(&m) &&
     sb->features[0] & cpu_to_le64(BIT_ULL(BCH_FEATURE_no_alloc_info))) {
  prt_printf(err, "device %u: freespace initialized but fs has no alloc info", i);
  return -BCH_ERR_invalid_sb_members;
 }

 return 0;
}

static void member_to_text(struct printbuf *out,
      struct bch_member m,
      struct bch_sb_field_disk_groups *gi,
      struct bch_sb *sb,
      int 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;

 if (!bch2_member_alive(&m))
  return;

 prt_printf(out, "Device:\t%u\n", i);

 printbuf_indent_add(out, 2);

 prt_printf(out, "Label:\t");
 if (BCH_MEMBER_GROUP(&m))
  bch2_disk_path_to_text_sb(out, sb,
    BCH_MEMBER_GROUP(&m) - 1);
 else
  prt_printf(out, "(none)");
 prt_newline(out);

 prt_printf(out, "UUID:\t");
 pr_uuid(out, m.uuid.b);
 prt_newline(out);

 prt_printf(out, "Size:\t");
 prt_units_u64(out, device_size << 9);
 prt_newline(out);

 for (unsigned i = 0; i < BCH_MEMBER_ERROR_NR; i++)
  prt_printf(out, "%s errors:\t%llu\n", bch2_member_error_strs[i], le64_to_cpu(m.errors[i]));

 for (unsigned i = 0; i < BCH_IOPS_NR; i++)
  prt_printf(out, "%s iops:\t%u\n", bch2_iops_measurements[i], le32_to_cpu(m.iops[i]));

 prt_printf(out, "Bucket size:\t");
 prt_units_u64(out, bucket_size << 9);
 prt_newline(out);

 prt_printf(out, "First bucket:\t%u\n", le16_to_cpu(m.first_bucket));
 prt_printf(out, "Buckets:\t%llu\n", le64_to_cpu(m.nbuckets));

 prt_printf(out, "Last mount:\t");
 if (m.last_mount)
  bch2_prt_datetime(out, le64_to_cpu(m.last_mount));
 else
  prt_printf(out, "(never)");
 prt_newline(out);

 prt_printf(out, "Last superblock write:\t%llu\n", le64_to_cpu(m.seq));

 prt_printf(out, "State:\t%s\n",
     BCH_MEMBER_STATE(&m) < BCH_MEMBER_STATE_NR
     ? bch2_member_states[BCH_MEMBER_STATE(&m)]
     : "unknown");

 prt_printf(out, "Data allowed:\t");
 if (BCH_MEMBER_DATA_ALLOWED(&m))
  prt_bitflags(out, __bch2_data_types, BCH_MEMBER_DATA_ALLOWED(&m));
 else
  prt_printf(out, "(none)");
 prt_newline(out);

 prt_printf(out, "Has data:\t");
 if (data_have)
  prt_bitflags(out, __bch2_data_types, data_have);
 else
  prt_printf(out, "(none)");
 prt_newline(out);

 prt_printf(out, "Btree allocated bitmap blocksize:\t");
 if (m.btree_bitmap_shift < 64)
  prt_units_u64(out, 1ULL << m.btree_bitmap_shift);
 else
  prt_printf(out, "(invalid shift %u)", m.btree_bitmap_shift);
 prt_newline(out);

 prt_printf(out, "Btree allocated bitmap:\t");
 bch2_prt_u64_base2_nbits(out, le64_to_cpu(m.btree_allocated_bitmap), 64);
 prt_newline(out);

 prt_printf(out, "Durability:\t%llu\n", BCH_MEMBER_DURABILITY(&m) ? BCH_MEMBER_DURABILITY(&m) - 1 : 1);

 prt_printf(out, "Discard:\t%llu\n", BCH_MEMBER_DISCARD(&m));
 prt_printf(out, "Freespace initialized:\t%llu\n", BCH_MEMBER_FREESPACE_INITIALIZED(&m));
 prt_printf(out, "Resize on mount:\t%llu\n", BCH_MEMBER_RESIZE_ON_MOUNT(&m));

 printbuf_indent_sub(out, 2);
}

static int bch2_sb_members_v1_validate(struct bch_sb *sb, struct bch_sb_field *f,
    enum bch_validate_flags flags, struct printbuf *err)
{
 struct bch_sb_field_members_v1 *mi = field_to_type(f, members_v1);
 unsigned i;

 if ((void *) members_v1_get_mut(mi, sb->nr_devices) > vstruct_end(&mi->field)) {
  prt_printf(err, "too many devices for section size");
  return -BCH_ERR_invalid_sb_members;
 }

 for (i = 0; i < sb->nr_devices; i++) {
  struct bch_member m = members_v1_get(mi, i);

  int ret = validate_member(err, m, sb, i);
  if (ret)
   return ret;
 }

 return 0;
}

static void bch2_sb_members_v1_to_text(struct printbuf *out, struct bch_sb *sb,
           struct bch_sb_field *f)
{
 struct bch_sb_field_members_v1 *mi = field_to_type(f, members_v1);
 struct bch_sb_field_disk_groups *gi = bch2_sb_field_get(sb, disk_groups);

 if (vstruct_end(&mi->field) <= (void *) &mi->_members[0]) {
  prt_printf(out, "field ends before start of entries");
  return;
 }

 unsigned nr = (vstruct_end(&mi->field) - (void *) &mi->_members[0]) / sizeof(mi->_members[0]);
 if (nr != sb->nr_devices)
  prt_printf(out, "nr_devices mismatch: have %i entries, should be %u", nr, sb->nr_devices);

 for (unsigned i = 0; i < min(sb->nr_devices, nr); i++)
  member_to_text(out, members_v1_get(mi, i), gi, sb, i);
}

const struct bch_sb_field_ops bch_sb_field_ops_members_v1 = {
 .validate = bch2_sb_members_v1_validate,
 .to_text = bch2_sb_members_v1_to_text,
};

static void bch2_sb_members_v2_to_text(struct printbuf *out, struct bch_sb *sb,
           struct bch_sb_field *f)
{
 struct bch_sb_field_members_v2 *mi = field_to_type(f, members_v2);
 struct bch_sb_field_disk_groups *gi = bch2_sb_field_get(sb, disk_groups);

 if (vstruct_end(&mi->field) <= (void *) &mi->_members[0]) {
  prt_printf(out, "field ends before start of entries");
  return;
 }

 if (!le16_to_cpu(mi->member_bytes)) {
  prt_printf(out, "member_bytes 0");
  return;
 }

 unsigned nr = (vstruct_end(&mi->field) - (void *) &mi->_members[0]) / le16_to_cpu(mi->member_bytes);
 if (nr != sb->nr_devices)
  prt_printf(out, "nr_devices mismatch: have %i entries, should be %u", nr, sb->nr_devices);

 /*
 * We call to_text() on superblock sections that haven't passed
 * validate, so we can't trust sb->nr_devices.
 */

 for (unsigned i = 0; i < min(sb->nr_devices, nr); i++)
  member_to_text(out, members_v2_get(mi, i), gi, sb, i);
}

static int bch2_sb_members_v2_validate(struct bch_sb *sb, struct bch_sb_field *f,
    enum bch_validate_flags flags, struct printbuf *err)
{
 struct bch_sb_field_members_v2 *mi = field_to_type(f, members_v2);
 size_t mi_bytes = (void *) __bch2_members_v2_get_mut(mi, sb->nr_devices) -
  (void *) mi;

 if (mi_bytes > vstruct_bytes(&mi->field)) {
  prt_printf(err, "section too small (%zu > %zu)",
      mi_bytes, vstruct_bytes(&mi->field));
  return -BCH_ERR_invalid_sb_members;
 }

 for (unsigned i = 0; i < sb->nr_devices; i++) {
  int ret = validate_member(err, members_v2_get(mi, i), sb, i);
  if (ret)
   return ret;
 }

 return 0;
}

const struct bch_sb_field_ops bch_sb_field_ops_members_v2 = {
 .validate = bch2_sb_members_v2_validate,
 .to_text = bch2_sb_members_v2_to_text,
};

void bch2_sb_members_from_cpu(struct bch_fs *c)
{
 struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);

 guard(rcu)();
 for_each_member_device_rcu(c, ca, NULL) {
  struct bch_member *m = __bch2_members_v2_get_mut(mi, ca->dev_idx);

  for (unsigned e = 0; e < BCH_MEMBER_ERROR_NR; e++)
   m->errors[e] = cpu_to_le64(atomic64_read(&ca->errors[e]));
 }
}

void bch2_dev_io_errors_to_text(struct printbuf *out, struct bch_dev *ca)
{
 struct bch_fs *c = ca->fs;
 struct bch_member m;

 mutex_lock(&ca->fs->sb_lock);
 m = bch2_sb_member_get(c->disk_sb.sb, ca->dev_idx);
 mutex_unlock(&ca->fs->sb_lock);

 printbuf_tabstop_push(out, 12);

 prt_str(out, "IO errors since filesystem creation");
 prt_newline(out);

 printbuf_indent_add(out, 2);
 for (unsigned i = 0; i < BCH_MEMBER_ERROR_NR; i++)
  prt_printf(out, "%s:\t%llu\n", bch2_member_error_strs[i], atomic64_read(&ca->errors[i]));
 printbuf_indent_sub(out, 2);

 prt_str(out, "IO errors since ");
 bch2_pr_time_units(out, (ktime_get_real_seconds() - le64_to_cpu(m.errors_reset_time)) * NSEC_PER_SEC);
 prt_str(out, " ago");
 prt_newline(out);

 printbuf_indent_add(out, 2);
 for (unsigned i = 0; i < BCH_MEMBER_ERROR_NR; i++)
  prt_printf(out, "%s:\t%llu\n", bch2_member_error_strs[i],
      atomic64_read(&ca->errors[i]) - le64_to_cpu(m.errors_at_reset[i]));
 printbuf_indent_sub(out, 2);
}

void bch2_dev_errors_reset(struct bch_dev *ca)
{
 struct bch_fs *c = ca->fs;
 struct bch_member *m;

 mutex_lock(&c->sb_lock);
 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
 for (unsigned i = 0; i < ARRAY_SIZE(m->errors_at_reset); i++)
  m->errors_at_reset[i] = cpu_to_le64(atomic64_read(&ca->errors[i]));
 m->errors_reset_time = cpu_to_le64(ktime_get_real_seconds());

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

/*
 * Per member "range has btree nodes" bitmap:
 *
 * This is so that if we ever have to run the btree node scan to repair we don't
 * have to scan full devices:
 */

bool bch2_dev_btree_bitmap_marked(struct bch_fs *c, struct bkey_s_c k)
{
 guard(rcu)();
 bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
  struct bch_dev *ca = bch2_dev_rcu(c, ptr->dev);
  if (ca &&
      !bch2_dev_btree_bitmap_marked_sectors(ca, ptr->offset, btree_sectors(c)))
   return false;
 }
 return true;
}

static void __bch2_dev_btree_bitmap_mark(struct bch_sb_field_members_v2 *mi, unsigned dev,
    u64 start, unsigned sectors)
{
 struct bch_member *m = __bch2_members_v2_get_mut(mi, dev);
 u64 bitmap = le64_to_cpu(m->btree_allocated_bitmap);

 u64 end = start + sectors;

 int resize = ilog2(roundup_pow_of_two(end)) - (m->btree_bitmap_shift + 6);
 if (resize > 0) {
  u64 new_bitmap = 0;

  for (unsigned i = 0; i < 64; i++)
   if (bitmap & BIT_ULL(i))
    new_bitmap |= BIT_ULL(i >> resize);
  bitmap = new_bitmap;
  m->btree_bitmap_shift += resize;
 }

 BUG_ON(m->btree_bitmap_shift >= BCH_MI_BTREE_BITMAP_SHIFT_MAX);
 BUG_ON(end > 64ULL << m->btree_bitmap_shift);

 for (unsigned bit = start >> m->btree_bitmap_shift;
      (u64) bit << m->btree_bitmap_shift < end;
      bit++)
  bitmap |= BIT_ULL(bit);

 m->btree_allocated_bitmap = cpu_to_le64(bitmap);
}

void bch2_dev_btree_bitmap_mark(struct bch_fs *c, struct bkey_s_c k)
{
 lockdep_assert_held(&c->sb_lock);

 struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
 bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
  if (!bch2_member_exists(c->disk_sb.sb, ptr->dev))
   continue;

  __bch2_dev_btree_bitmap_mark(mi, ptr->dev, ptr->offset, btree_sectors(c));
 }
}

unsigned bch2_sb_nr_devices(const struct bch_sb *sb)
{
 unsigned nr = 0;

 for (unsigned i = 0; i < sb->nr_devices; i++)
  nr += bch2_member_exists((struct bch_sb *) sb, i);
 return nr;
}

int bch2_sb_member_alloc(struct bch_fs *c)
{
 unsigned dev_idx = c->sb.nr_devices;
 struct bch_sb_field_members_v2 *mi;
 unsigned nr_devices;
 unsigned u64s;
 int best = -1;
 u64 best_last_mount = 0;
 unsigned nr_deleted = 0;

 if (dev_idx < BCH_SB_MEMBERS_MAX)
  goto have_slot;

 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++) {
  /* eventually BCH_SB_MEMBERS_MAX will be raised */
  if (dev_idx == BCH_SB_MEMBER_INVALID)
   continue;

  struct bch_member m = bch2_sb_member_get(c->disk_sb.sb, dev_idx);

  nr_deleted += uuid_equal(&m.uuid, &BCH_SB_MEMBER_DELETED_UUID);

  if (!bch2_is_zero(&m.uuid, sizeof(m.uuid)))
   continue;

  u64 last_mount = le64_to_cpu(m.last_mount);
  if (best < 0 || last_mount < best_last_mount) {
   best = dev_idx;
   best_last_mount = last_mount;
  }
 }
 if (best >= 0) {
  dev_idx = best;
  goto have_slot;
 }

 if (nr_deleted)
  bch_err(c, "unable to allocate new member, but have %u deleted: run fsck",
   nr_deleted);

 return -BCH_ERR_ENOSPC_sb_members;
have_slot:
 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);

 mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
 u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
       le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));

 mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
 if (!mi)
  return -BCH_ERR_ENOSPC_sb_members;

 c->disk_sb.sb->nr_devices = nr_devices;
 return dev_idx;
}

void bch2_sb_members_clean_deleted(struct bch_fs *c)
{
 mutex_lock(&c->sb_lock);
 bool write_sb = false;

 for (unsigned i = 0; i < c->sb.nr_devices; i++) {
  struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, i);

  if (uuid_equal(&m->uuid, &BCH_SB_MEMBER_DELETED_UUID)) {
   memset(&m->uuid, 0, sizeof(m->uuid));
   write_sb = true;
  }
 }

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