#include "debug.h"
#include "error.h"
#include "extents.h"
-#include "io.h"
+#include "io_write.h"
#include "journal_reclaim.h"
#include "journal_seq_blacklist.h"
+#include "recovery.h"
#include "super-io.h"
+#include "trace.h"
#include <linux/sched/mm.h>
-#include <trace/events/bcachefs.h>
void bch2_btree_node_io_unlock(struct btree *b)
{
void bch2_btree_node_io_lock(struct btree *b)
{
- BUG_ON(lock_class_is_held(&bch2_btree_node_lock_key));
+ bch2_assert_btree_nodes_not_locked();
wait_on_bit_lock_io(&b->flags, BTREE_NODE_write_in_flight,
TASK_UNINTERRUPTIBLE);
void bch2_btree_node_wait_on_read(struct btree *b)
{
- BUG_ON(lock_class_is_held(&bch2_btree_node_lock_key));
+ bch2_assert_btree_nodes_not_locked();
wait_on_bit_io(&b->flags, BTREE_NODE_read_in_flight,
TASK_UNINTERRUPTIBLE);
void bch2_btree_node_wait_on_write(struct btree *b)
{
- BUG_ON(lock_class_is_held(&bch2_btree_node_lock_key));
+ bch2_assert_btree_nodes_not_locked();
wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight,
TASK_UNINTERRUPTIBLE);
if (start == end)
return;
- for (p = start, k = bkey_next(start);
+ for (p = start, k = bkey_p_next(start);
k != end;
- p = k, k = bkey_next(k)) {
+ p = k, k = bkey_p_next(k)) {
struct bkey l = bkey_unpack_key(b, p);
struct bkey r = bkey_unpack_key(b, k);
- BUG_ON(bpos_cmp(l.p, bkey_start_pos(&r)) >= 0);
+ BUG_ON(bpos_ge(l.p, bkey_start_pos(&r)));
}
#endif
}
{
struct bkey_packed *k;
- for (k = i->start; k != vstruct_last(i); k = bkey_next(k))
+ for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k))
k->needs_whiteout = v;
}
p = vpmalloc(size, __GFP_NOWARN|GFP_NOWAIT);
if (!p) {
*used_mempool = true;
- p = mempool_alloc(&c->btree_bounce_pool, GFP_NOIO);
+ p = mempool_alloc(&c->btree_bounce_pool, GFP_NOFS);
}
memalloc_nofs_restore(flags);
return p;
for (k = unwritten_whiteouts_start(c, b);
k != unwritten_whiteouts_end(c, b);
- k = bkey_next(k))
+ k = bkey_p_next(k))
*--ptrs = k;
sort_bkey_ptrs(b, ptrs, ptrs_end - ptrs);
while (ptrs != ptrs_end) {
bkey_copy(k, *ptrs);
- k = bkey_next(k);
+ k = bkey_p_next(k);
ptrs++;
}
out = i->start;
for (k = start; k != end; k = n) {
- n = bkey_next(k);
+ n = bkey_p_next(k);
if (!bkey_deleted(k)) {
bkey_copy(out, k);
- out = bkey_next(out);
+ out = bkey_p_next(out);
} else {
BUG_ON(k->needs_whiteout);
}
bool filter_whiteouts)
{
struct btree_node *out;
- struct sort_iter sort_iter;
+ struct sort_iter_stack sort_iter;
struct bset_tree *t;
struct bset *start_bset = bset(b, &b->set[start_idx]);
bool used_mempool = false;
bool sorting_entire_node = start_idx == 0 &&
end_idx == b->nsets;
- sort_iter_init(&sort_iter, b);
+ sort_iter_stack_init(&sort_iter, b);
for (t = b->set + start_idx;
t < b->set + end_idx;
t++) {
u64s += le16_to_cpu(bset(b, t)->u64s);
- sort_iter_add(&sort_iter,
+ sort_iter_add(&sort_iter.iter,
btree_bkey_first(b, t),
btree_bkey_last(b, t));
}
start_time = local_clock();
- u64s = bch2_sort_keys(out->keys.start, &sort_iter, filter_whiteouts);
+ u64s = bch2_sort_keys(out->keys.start, &sort_iter.iter, filter_whiteouts);
out->keys.u64s = cpu_to_le16(u64s);
start_bset->journal_seq = cpu_to_le64(seq);
if (sorting_entire_node) {
- unsigned u64s = le16_to_cpu(out->keys.u64s);
+ u64s = le16_to_cpu(out->keys.u64s);
BUG_ON(bytes != btree_bytes(c));
bch2_verify_btree_nr_keys(dst);
}
-#define SORT_CRIT (4096 / sizeof(u64))
-
/*
* We're about to add another bset to the btree node, so if there's currently
* too many bsets - sort some of them together:
struct btree_node_entry *bne;
bool reinit_iter = false;
- EBUG_ON(!(b->c.lock.state.seq & 1));
+ EBUG_ON(!six_lock_counts(&b->c.lock).n[SIX_LOCK_write]);
BUG_ON(bset_written(b, bset(b, &b->set[1])));
BUG_ON(btree_node_just_written(b));
prt_printf(out, "%s level %u/%u\n ",
bch2_btree_ids[b->c.btree_id],
b->c.level,
- c->btree_roots[b->c.btree_id].level);
+ bch2_btree_id_root(c, b->c.btree_id)->level);
bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
}
struct btree *b, struct bset *i,
unsigned offset, int write)
{
- prt_printf(out, bch2_log_msg(c, ""));
- if (!write)
- prt_str(out, "error validating btree node ");
- else
- prt_str(out, "corrupt btree node before write ");
+ prt_printf(out, bch2_log_msg(c, "%s"),
+ write == READ
+ ? "error validating btree node "
+ : "corrupt btree node before write ");
if (ca)
prt_printf(out, "on %s ", ca->name);
prt_printf(out, "at btree ");
prt_str(out, ": ");
}
-enum btree_err_type {
- BTREE_ERR_FIXABLE,
- BTREE_ERR_WANT_RETRY,
- BTREE_ERR_MUST_RETRY,
- BTREE_ERR_FATAL,
-};
+__printf(8, 9)
+static int __btree_err(int ret,
+ struct bch_fs *c,
+ struct bch_dev *ca,
+ struct btree *b,
+ struct bset *i,
+ int write,
+ bool have_retry,
+ const char *fmt, ...)
+{
+ struct printbuf out = PRINTBUF;
+ va_list args;
-enum btree_validate_ret {
- BTREE_RETRY_READ = 64,
-};
+ btree_err_msg(&out, c, ca, b, i, b->written, write);
+
+ va_start(args, fmt);
+ prt_vprintf(&out, fmt, args);
+ va_end(args);
+
+ if (write == WRITE) {
+ bch2_print_string_as_lines(KERN_ERR, out.buf);
+ ret = c->opts.errors == BCH_ON_ERROR_continue
+ ? 0
+ : -BCH_ERR_fsck_errors_not_fixed;
+ goto out;
+ }
+
+ if (!have_retry && ret == -BCH_ERR_btree_node_read_err_want_retry)
+ ret = -BCH_ERR_btree_node_read_err_fixable;
+ if (!have_retry && ret == -BCH_ERR_btree_node_read_err_must_retry)
+ ret = -BCH_ERR_btree_node_read_err_bad_node;
+
+ switch (ret) {
+ case -BCH_ERR_btree_node_read_err_fixable:
+ mustfix_fsck_err(c, "%s", out.buf);
+ ret = -BCH_ERR_fsck_fix;
+ break;
+ case -BCH_ERR_btree_node_read_err_want_retry:
+ case -BCH_ERR_btree_node_read_err_must_retry:
+ bch2_print_string_as_lines(KERN_ERR, out.buf);
+ break;
+ case -BCH_ERR_btree_node_read_err_bad_node:
+ bch2_print_string_as_lines(KERN_ERR, out.buf);
+ bch2_topology_error(c);
+ ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology) ?: -EIO;
+ break;
+ case -BCH_ERR_btree_node_read_err_incompatible:
+ bch2_print_string_as_lines(KERN_ERR, out.buf);
+ ret = -BCH_ERR_fsck_errors_not_fixed;
+ break;
+ default:
+ BUG();
+ }
+out:
+fsck_err:
+ printbuf_exit(&out);
+ return ret;
+}
#define btree_err(type, c, ca, b, i, msg, ...) \
({ \
- __label__ out; \
- struct printbuf out = PRINTBUF; \
- \
- btree_err_msg(&out, c, ca, b, i, b->written, write); \
- prt_printf(&out, msg, ##__VA_ARGS__); \
+ int _ret = __btree_err(type, c, ca, b, i, write, have_retry, msg, ##__VA_ARGS__);\
\
- if (type == BTREE_ERR_FIXABLE && \
- write == READ && \
- !test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags)) { \
- mustfix_fsck_err(c, "%s", out.buf); \
- goto out; \
+ if (_ret != -BCH_ERR_fsck_fix) { \
+ ret = _ret; \
+ goto fsck_err; \
} \
\
- bch2_print_string_as_lines(KERN_ERR, out.buf); \
- \
- switch (write) { \
- case READ: \
- switch (type) { \
- case BTREE_ERR_FIXABLE: \
- ret = -BCH_ERR_fsck_errors_not_fixed; \
- goto fsck_err; \
- case BTREE_ERR_WANT_RETRY: \
- if (have_retry) { \
- ret = BTREE_RETRY_READ; \
- goto fsck_err; \
- } \
- break; \
- case BTREE_ERR_MUST_RETRY: \
- ret = BTREE_RETRY_READ; \
- goto fsck_err; \
- case BTREE_ERR_FATAL: \
- ret = -BCH_ERR_fsck_errors_not_fixed; \
- goto fsck_err; \
- } \
- break; \
- case WRITE: \
- if (bch2_fs_inconsistent(c)) { \
- ret = -BCH_ERR_fsck_errors_not_fixed; \
- goto fsck_err; \
- } \
- break; \
- } \
-out: \
- printbuf_exit(&out); \
- true; \
+ *saw_error = true; \
})
#define btree_err_on(cond, ...) ((cond) ? btree_err(__VA_ARGS__) : false)
* When btree topology repair changes the start or end of a node, that might
* mean we have to drop keys that are no longer inside the node:
*/
+__cold
void bch2_btree_node_drop_keys_outside_node(struct btree *b)
{
struct bset_tree *t;
- struct bkey_s_c k;
- struct bkey unpacked;
- struct btree_node_iter iter;
for_each_bset(b, t) {
struct bset *i = bset(b, t);
struct bkey_packed *k;
- for (k = i->start; k != vstruct_last(i); k = bkey_next(k))
+ for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k))
if (bkey_cmp_left_packed(b, k, &b->data->min_key) >= 0)
break;
set_btree_bset_end(b, t);
}
- for (k = i->start; k != vstruct_last(i); k = bkey_next(k))
+ for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k))
if (bkey_cmp_left_packed(b, k, &b->data->max_key) > 0)
break;
bch2_bset_set_no_aux_tree(b, b->set);
bch2_btree_build_aux_trees(b);
+ struct bkey_s_c k;
+ struct bkey unpacked;
+ struct btree_node_iter iter;
for_each_btree_node_key_unpack(b, k, &iter, &unpacked) {
- BUG_ON(bpos_cmp(k.k->p, b->data->min_key) < 0);
- BUG_ON(bpos_cmp(k.k->p, b->data->max_key) > 0);
+ BUG_ON(bpos_lt(k.k->p, b->data->min_key));
+ BUG_ON(bpos_gt(k.k->p, b->data->max_key));
}
}
static int validate_bset(struct bch_fs *c, struct bch_dev *ca,
struct btree *b, struct bset *i,
unsigned offset, unsigned sectors,
- int write, bool have_retry)
+ int write, bool have_retry, bool *saw_error)
{
unsigned version = le16_to_cpu(i->version);
- const char *err;
struct printbuf buf1 = PRINTBUF;
struct printbuf buf2 = PRINTBUF;
int ret = 0;
- btree_err_on((version != BCH_BSET_VERSION_OLD &&
- version < bcachefs_metadata_version_min) ||
- version >= bcachefs_metadata_version_max,
- BTREE_ERR_FATAL, c, ca, b, i,
- "unsupported bset version");
+ btree_err_on(!bch2_version_compatible(version),
+ -BCH_ERR_btree_node_read_err_incompatible, c, ca, b, i,
+ "unsupported bset version %u.%u",
+ BCH_VERSION_MAJOR(version),
+ BCH_VERSION_MINOR(version));
if (btree_err_on(version < c->sb.version_min,
- BTREE_ERR_FIXABLE, c, NULL, b, i,
+ -BCH_ERR_btree_node_read_err_fixable, c, NULL, b, i,
"bset version %u older than superblock version_min %u",
version, c->sb.version_min)) {
mutex_lock(&c->sb_lock);
mutex_unlock(&c->sb_lock);
}
- if (btree_err_on(version > c->sb.version,
- BTREE_ERR_FIXABLE, c, NULL, b, i,
+ if (btree_err_on(BCH_VERSION_MAJOR(version) >
+ BCH_VERSION_MAJOR(c->sb.version),
+ -BCH_ERR_btree_node_read_err_fixable, c, NULL, b, i,
"bset version %u newer than superblock version %u",
version, c->sb.version)) {
mutex_lock(&c->sb_lock);
}
btree_err_on(BSET_SEPARATE_WHITEOUTS(i),
- BTREE_ERR_FATAL, c, ca, b, i,
+ -BCH_ERR_btree_node_read_err_incompatible, c, ca, b, i,
"BSET_SEPARATE_WHITEOUTS no longer supported");
if (btree_err_on(offset + sectors > btree_sectors(c),
- BTREE_ERR_FIXABLE, c, ca, b, i,
+ -BCH_ERR_btree_node_read_err_fixable, c, ca, b, i,
"bset past end of btree node")) {
i->u64s = 0;
ret = 0;
}
btree_err_on(offset && !i->u64s,
- BTREE_ERR_FIXABLE, c, ca, b, i,
+ -BCH_ERR_btree_node_read_err_fixable, c, ca, b, i,
"empty bset");
btree_err_on(BSET_OFFSET(i) &&
BSET_OFFSET(i) != offset,
- BTREE_ERR_WANT_RETRY, c, ca, b, i,
+ -BCH_ERR_btree_node_read_err_want_retry, c, ca, b, i,
"bset at wrong sector offset");
if (!offset) {
/* XXX endianness */
btree_err_on(bp->seq != bn->keys.seq,
- BTREE_ERR_MUST_RETRY, c, ca, b, NULL,
+ -BCH_ERR_btree_node_read_err_must_retry, c, ca, b, NULL,
"incorrect sequence number (wrong btree node)");
}
btree_err_on(BTREE_NODE_ID(bn) != b->c.btree_id,
- BTREE_ERR_MUST_RETRY, c, ca, b, i,
+ -BCH_ERR_btree_node_read_err_must_retry, c, ca, b, i,
"incorrect btree id");
btree_err_on(BTREE_NODE_LEVEL(bn) != b->c.level,
- BTREE_ERR_MUST_RETRY, c, ca, b, i,
+ -BCH_ERR_btree_node_read_err_must_retry, c, ca, b, i,
"incorrect level");
if (!write)
b->data->max_key = b->key.k.p;
}
- btree_err_on(bpos_cmp(b->data->min_key, bp->min_key),
- BTREE_ERR_MUST_RETRY, c, ca, b, NULL,
+ btree_err_on(!bpos_eq(b->data->min_key, bp->min_key),
+ -BCH_ERR_btree_node_read_err_must_retry, c, ca, b, NULL,
"incorrect min_key: got %s should be %s",
(printbuf_reset(&buf1),
bch2_bpos_to_text(&buf1, bn->min_key), buf1.buf),
bch2_bpos_to_text(&buf2, bp->min_key), buf2.buf));
}
- btree_err_on(bpos_cmp(bn->max_key, b->key.k.p),
- BTREE_ERR_MUST_RETRY, c, ca, b, i,
+ btree_err_on(!bpos_eq(bn->max_key, b->key.k.p),
+ -BCH_ERR_btree_node_read_err_must_retry, c, ca, b, i,
"incorrect max key %s",
(printbuf_reset(&buf1),
bch2_bpos_to_text(&buf1, bn->max_key), buf1.buf));
compat_btree_node(b->c.level, b->c.btree_id, version,
BSET_BIG_ENDIAN(i), write, bn);
- err = bch2_bkey_format_validate(&bn->format);
- btree_err_on(err,
- BTREE_ERR_FATAL, c, ca, b, i,
- "invalid bkey format: %s", err);
+ btree_err_on(bch2_bkey_format_invalid(c, &bn->format, write, &buf1),
+ -BCH_ERR_btree_node_read_err_bad_node, c, ca, b, i,
+ "invalid bkey format: %s\n %s", buf1.buf,
+ (printbuf_reset(&buf2),
+ bch2_bkey_format_to_text(&buf2, &bn->format), buf2.buf));
+ printbuf_reset(&buf1);
compat_bformat(b->c.level, b->c.btree_id, version,
BSET_BIG_ENDIAN(i), write,
}
static int validate_bset_keys(struct bch_fs *c, struct btree *b,
- struct bset *i, int write, bool have_retry)
+ struct bset *i, int write,
+ bool have_retry, bool *saw_error)
{
unsigned version = le16_to_cpu(i->version);
struct bkey_packed *k, *prev = NULL;
struct bkey_s u;
struct bkey tmp;
- if (btree_err_on(bkey_next(k) > vstruct_last(i),
- BTREE_ERR_FIXABLE, c, NULL, b, i,
+ if (btree_err_on(bkey_p_next(k) > vstruct_last(i),
+ -BCH_ERR_btree_node_read_err_fixable, c, NULL, b, i,
"key extends past end of bset")) {
i->u64s = cpu_to_le16((u64 *) k - i->_data);
break;
}
if (btree_err_on(k->format > KEY_FORMAT_CURRENT,
- BTREE_ERR_FIXABLE, c, NULL, b, i,
+ -BCH_ERR_btree_node_read_err_fixable, c, NULL, b, i,
"invalid bkey format %u", k->format)) {
i->u64s = cpu_to_le16(le16_to_cpu(i->u64s) - k->u64s);
- memmove_u64s_down(k, bkey_next(k),
+ memmove_u64s_down(k, bkey_p_next(k),
(u64 *) vstruct_end(i) - (u64 *) k);
continue;
}
prt_printf(&buf, "\n ");
bch2_bkey_val_to_text(&buf, c, u.s_c);
- btree_err(BTREE_ERR_FIXABLE, c, NULL, b, i, "%s", buf.buf);
+ btree_err(-BCH_ERR_btree_node_read_err_fixable, c, NULL, b, i, "%s", buf.buf);
i->u64s = cpu_to_le16(le16_to_cpu(i->u64s) - k->u64s);
- memmove_u64s_down(k, bkey_next(k),
+ memmove_u64s_down(k, bkey_p_next(k),
(u64 *) vstruct_end(i) - (u64 *) k);
continue;
}
bch2_dump_bset(c, b, i, 0);
- if (btree_err(BTREE_ERR_FIXABLE, c, NULL, b, i, "%s", buf.buf)) {
+ if (btree_err(-BCH_ERR_btree_node_read_err_fixable, c, NULL, b, i, "%s", buf.buf)) {
i->u64s = cpu_to_le16(le16_to_cpu(i->u64s) - k->u64s);
- memmove_u64s_down(k, bkey_next(k),
+ memmove_u64s_down(k, bkey_p_next(k),
(u64 *) vstruct_end(i) - (u64 *) k);
continue;
}
}
prev = k;
- k = bkey_next(k);
+ k = bkey_p_next(k);
}
fsck_err:
printbuf_exit(&buf);
}
int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
- struct btree *b, bool have_retry)
+ struct btree *b, bool have_retry, bool *saw_error)
{
struct btree_node_entry *bne;
struct sort_iter *iter;
bool updated_range = b->key.k.type == KEY_TYPE_btree_ptr_v2 &&
BTREE_PTR_RANGE_UPDATED(&bkey_i_to_btree_ptr_v2(&b->key)->v);
unsigned u64s;
- unsigned blacklisted_written, nonblacklisted_written = 0;
unsigned ptr_written = btree_ptr_sectors_written(&b->key);
struct printbuf buf = PRINTBUF;
- int ret, retry_read = 0, write = READ;
+ int ret = 0, retry_read = 0, write = READ;
b->version_ondisk = U16_MAX;
/* We might get called multiple times on read retry: */
b->written = 0;
- iter = mempool_alloc(&c->fill_iter, GFP_NOIO);
- sort_iter_init(iter, b);
- iter->size = (btree_blocks(c) + 1) * 2;
+ iter = mempool_alloc(&c->fill_iter, GFP_NOFS);
+ sort_iter_init(iter, b, (btree_blocks(c) + 1) * 2);
if (bch2_meta_read_fault("btree"))
- btree_err(BTREE_ERR_MUST_RETRY, c, ca, b, NULL,
+ btree_err(-BCH_ERR_btree_node_read_err_must_retry, c, ca, b, NULL,
"dynamic fault");
btree_err_on(le64_to_cpu(b->data->magic) != bset_magic(c),
- BTREE_ERR_MUST_RETRY, c, ca, b, NULL,
+ -BCH_ERR_btree_node_read_err_must_retry, c, ca, b, NULL,
"bad magic: want %llx, got %llx",
bset_magic(c), le64_to_cpu(b->data->magic));
btree_err_on(!b->data->keys.seq,
- BTREE_ERR_MUST_RETRY, c, ca, b, NULL,
+ -BCH_ERR_btree_node_read_err_must_retry, c, ca, b, NULL,
"bad btree header: seq 0");
if (b->key.k.type == KEY_TYPE_btree_ptr_v2) {
&bkey_i_to_btree_ptr_v2(&b->key)->v;
btree_err_on(b->data->keys.seq != bp->seq,
- BTREE_ERR_MUST_RETRY, c, ca, b, NULL,
+ -BCH_ERR_btree_node_read_err_must_retry, c, ca, b, NULL,
"got wrong btree node (seq %llx want %llx)",
b->data->keys.seq, bp->seq);
}
i = &b->data->keys;
btree_err_on(!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i)),
- BTREE_ERR_WANT_RETRY, c, ca, b, i,
+ -BCH_ERR_btree_node_read_err_want_retry, c, ca, b, i,
"unknown checksum type %llu",
BSET_CSUM_TYPE(i));
csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, b->data);
btree_err_on(bch2_crc_cmp(csum, b->data->csum),
- BTREE_ERR_WANT_RETRY, c, ca, b, i,
+ -BCH_ERR_btree_node_read_err_want_retry, c, ca, b, i,
"invalid checksum");
ret = bset_encrypt(c, i, b->written << 9);
btree_err_on(btree_node_type_is_extents(btree_node_type(b)) &&
!BTREE_NODE_NEW_EXTENT_OVERWRITE(b->data),
- BTREE_ERR_FATAL, c, NULL, b, NULL,
+ -BCH_ERR_btree_node_read_err_incompatible, c, NULL, b, NULL,
"btree node does not have NEW_EXTENT_OVERWRITE set");
sectors = vstruct_sectors(b->data, c->block_bits);
break;
btree_err_on(!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i)),
- BTREE_ERR_WANT_RETRY, c, ca, b, i,
+ -BCH_ERR_btree_node_read_err_want_retry, c, ca, b, i,
"unknown checksum type %llu",
BSET_CSUM_TYPE(i));
csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bne);
btree_err_on(bch2_crc_cmp(csum, bne->csum),
- BTREE_ERR_WANT_RETRY, c, ca, b, i,
+ -BCH_ERR_btree_node_read_err_want_retry, c, ca, b, i,
"invalid checksum");
ret = bset_encrypt(c, i, b->written << 9);
le16_to_cpu(i->version));
ret = validate_bset(c, ca, b, i, b->written, sectors,
- READ, have_retry);
+ READ, have_retry, saw_error);
if (ret)
goto fsck_err;
if (!b->written)
btree_node_set_format(b, b->data->format);
- ret = validate_bset_keys(c, b, i, READ, have_retry);
+ ret = validate_bset_keys(c, b, i, READ, have_retry, saw_error);
if (ret)
goto fsck_err;
true);
btree_err_on(blacklisted && first,
- BTREE_ERR_FIXABLE, c, ca, b, i,
+ -BCH_ERR_btree_node_read_err_fixable, c, ca, b, i,
"first btree node bset has blacklisted journal seq (%llu)",
le64_to_cpu(i->journal_seq));
btree_err_on(blacklisted && ptr_written,
- BTREE_ERR_FIXABLE, c, ca, b, i,
+ -BCH_ERR_btree_node_read_err_fixable, c, ca, b, i,
"found blacklisted bset (journal seq %llu) in btree node at offset %u-%u/%u",
le64_to_cpu(i->journal_seq),
b->written, b->written + sectors, ptr_written);
sort_iter_add(iter,
vstruct_idx(i, 0),
vstruct_last(i));
-
- nonblacklisted_written = b->written;
}
if (ptr_written) {
btree_err_on(b->written < ptr_written,
- BTREE_ERR_WANT_RETRY, c, ca, b, NULL,
+ -BCH_ERR_btree_node_read_err_want_retry, c, ca, b, NULL,
"btree node data missing: expected %u sectors, found %u",
ptr_written, b->written);
} else {
!bch2_journal_seq_is_blacklisted(c,
le64_to_cpu(bne->keys.journal_seq),
true),
- BTREE_ERR_WANT_RETRY, c, ca, b, NULL,
+ -BCH_ERR_btree_node_read_err_want_retry, c, ca, b, NULL,
"found bset signature after last bset");
-
- /*
- * Blacklisted bsets are those that were written after the most recent
- * (flush) journal write. Since there wasn't a flush, they may not have
- * made it to all devices - which means we shouldn't write new bsets
- * after them, as that could leave a gap and then reads from that device
- * wouldn't find all the bsets in that btree node - which means it's
- * important that we start writing new bsets after the most recent _non_
- * blacklisted bset:
- */
- blacklisted_written = b->written;
- b->written = nonblacklisted_written;
}
sorted = btree_bounce_alloc(c, btree_bytes(c), &used_mempool);
prt_printf(&buf, "\n ");
bch2_bkey_val_to_text(&buf, c, u.s_c);
- btree_err(BTREE_ERR_FIXABLE, c, NULL, b, i, "%s", buf.buf);
+ btree_err(-BCH_ERR_btree_node_read_err_fixable, c, NULL, b, i, "%s", buf.buf);
btree_keys_account_key_drop(&b->nr, 0, k);
i->u64s = cpu_to_le16(le16_to_cpu(i->u64s) - k->u64s);
- memmove_u64s_down(k, bkey_next(k),
+ memmove_u64s_down(k, bkey_p_next(k),
(u64 *) vstruct_end(i) - (u64 *) k);
set_btree_bset_end(b, b->set);
continue;
bp.v->mem_ptr = 0;
}
- k = bkey_next(k);
+ k = bkey_p_next(k);
}
bch2_bset_build_aux_tree(b, b->set, false);
btree_node_reset_sib_u64s(b);
bkey_for_each_ptr(bch2_bkey_ptrs(bkey_i_to_s(&b->key)), ptr) {
- struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
+ struct bch_dev *ca2 = bch_dev_bkey_exists(c, ptr->dev);
- if (ca->mi.state != BCH_MEMBER_STATE_rw)
+ if (ca2->mi.state != BCH_MEMBER_STATE_rw)
set_btree_node_need_rewrite(b);
}
printbuf_exit(&buf);
return retry_read;
fsck_err:
- if (ret == BTREE_RETRY_READ) {
+ if (ret == -BCH_ERR_btree_node_read_err_want_retry ||
+ ret == -BCH_ERR_btree_node_read_err_must_retry)
retry_read = 1;
- } else {
- bch2_inconsistent_error(c);
+ else
set_btree_node_read_error(b);
- }
goto out;
}
&failed, &rb->pick) > 0;
if (!bio->bi_status &&
- !bch2_btree_node_read_done(c, ca, b, can_retry)) {
+ !bch2_btree_node_read_done(c, ca, b, can_retry, &saw_error)) {
if (retry)
bch_info(c, "retry success");
break;
bch2_time_stats_update(&c->times[BCH_TIME_btree_node_read],
rb->start_time);
bio_put(&rb->bio);
- printbuf_exit(&buf);
- if (saw_error && !btree_node_read_error(b))
+ if (saw_error && !btree_node_read_error(b)) {
+ printbuf_reset(&buf);
+ bch2_bpos_to_text(&buf, b->key.k.p);
+ bch_info(c, "%s: rewriting btree node at btree=%s level=%u %s due to error",
+ __func__, bch2_btree_ids[b->c.btree_id], b->c.level, buf.buf);
+
bch2_btree_node_rewrite_async(c, b);
+ }
+ printbuf_exit(&buf);
clear_btree_node_read_in_flight(b);
wake_up_bit(&b->flags, BTREE_NODE_read_in_flight);
}
unsigned nr;
void *buf[BCH_REPLICAS_MAX];
struct bio *bio[BCH_REPLICAS_MAX];
- int err[BCH_REPLICAS_MAX];
+ blk_status_t err[BCH_REPLICAS_MAX];
};
static unsigned btree_node_sectors_written(struct bch_fs *c, void *data)
unsigned i, written = 0, written2 = 0;
__le64 seq = b->key.k.type == KEY_TYPE_btree_ptr_v2
? bkey_i_to_btree_ptr_v2(&b->key)->v.seq : 0;
+ bool _saw_error = false, *saw_error = &_saw_error;
for (i = 0; i < ra->nr; i++) {
struct btree_node *bn = ra->buf[i];
}
written2 = btree_node_sectors_written(c, ra->buf[i]);
- if (btree_err_on(written2 != written, BTREE_ERR_FIXABLE, c, NULL, b, NULL,
+ if (btree_err_on(written2 != written, -BCH_ERR_btree_node_read_err_fixable, c, NULL, b, NULL,
"btree node sectors written mismatch: %u != %u",
written, written2) ||
btree_err_on(btree_node_has_extra_bsets(c, written2, ra->buf[i]),
- BTREE_ERR_FIXABLE, c, NULL, b, NULL,
+ -BCH_ERR_btree_node_read_err_fixable, c, NULL, b, NULL,
"found bset signature after last bset") ||
btree_err_on(memcmp(ra->buf[best], ra->buf[i], written << 9),
- BTREE_ERR_FIXABLE, c, NULL, b, NULL,
+ -BCH_ERR_btree_node_read_err_fixable, c, NULL, b, NULL,
"btree node replicas content mismatch"))
dump_bset_maps = true;
if (best >= 0) {
memcpy(b->data, ra->buf[best], btree_bytes(c));
- ret = bch2_btree_node_read_done(c, NULL, b, false);
+ ret = bch2_btree_node_read_done(c, NULL, b, false, saw_error);
} else {
ret = -1;
}
if (ret)
set_btree_node_read_error(b);
+ else if (*saw_error)
+ bch2_btree_node_rewrite_async(c, b);
for (i = 0; i < ra->nr; i++) {
mempool_free(ra->buf[i], &c->btree_bounce_pool);
ra = kzalloc(sizeof(*ra), GFP_NOFS);
if (!ra)
- return -ENOMEM;
+ return -BCH_ERR_ENOMEM_btree_node_read_all_replicas;
closure_init(&ra->cl, NULL);
ra->c = c;
btree_pos_to_text(&buf, c, b);
bch_err(c, "%s", buf.buf);
- if (test_bit(BCH_FS_TOPOLOGY_REPAIR_DONE, &c->flags))
+ if (c->recovery_passes_explicit & BIT_ULL(BCH_RECOVERY_PASS_check_topology) &&
+ c->curr_recovery_pass > BCH_RECOVERY_PASS_check_topology)
bch2_fatal_error(c);
set_btree_node_read_error(b);
bio = bio_alloc_bioset(NULL,
buf_pages(b->data, btree_bytes(c)),
REQ_OP_READ|REQ_SYNC|REQ_META,
- GFP_NOIO,
+ GFP_NOFS,
&c->btree_bio);
rb = container_of(bio, struct btree_read_bio, bio);
rb->c = c;
}
}
-int bch2_btree_root_read(struct bch_fs *c, enum btree_id id,
- const struct bkey_i *k, unsigned level)
+static int __bch2_btree_root_read(struct btree_trans *trans, enum btree_id id,
+ const struct bkey_i *k, unsigned level)
{
+ struct bch_fs *c = trans->c;
struct closure cl;
struct btree *b;
int ret;
closure_sync(&cl);
} while (ret);
- b = bch2_btree_node_mem_alloc(c, level != 0);
+ b = bch2_btree_node_mem_alloc(trans, level != 0);
bch2_btree_cache_cannibalize_unlock(c);
BUG_ON(IS_ERR(b));
return ret;
}
+int bch2_btree_root_read(struct bch_fs *c, enum btree_id id,
+ const struct bkey_i *k, unsigned level)
+{
+ return bch2_trans_run(c, __bch2_btree_root_read(trans, id, k, level));
+}
+
void bch2_btree_complete_write(struct bch_fs *c, struct btree *b,
struct btree_write *w)
{
{
struct btree_write *w = btree_prev_write(b);
unsigned long old, new, v;
+ unsigned type = 0;
bch2_btree_complete_write(c, b, w);
new |= (1U << BTREE_NODE_write_in_flight_inner);
new |= (1U << BTREE_NODE_just_written);
new ^= (1U << BTREE_NODE_write_idx);
+
+ type = new & BTREE_WRITE_TYPE_MASK;
+ new &= ~BTREE_WRITE_TYPE_MASK;
} else {
new &= ~(1U << BTREE_NODE_write_in_flight);
new &= ~(1U << BTREE_NODE_write_in_flight_inner);
} while ((v = cmpxchg(&b->flags, old, new)) != old);
if (new & (1U << BTREE_NODE_write_in_flight))
- __bch2_btree_node_write(c, b, BTREE_WRITE_ALREADY_STARTED|b->write_type);
+ __bch2_btree_node_write(c, b, BTREE_WRITE_ALREADY_STARTED|type);
else
wake_up_bit(&b->flags, BTREE_NODE_write_in_flight);
}
static void btree_node_write_done(struct bch_fs *c, struct btree *b)
{
- struct btree_trans trans;
-
- bch2_trans_init(&trans, c, 0, 0);
+ struct btree_trans *trans = bch2_trans_get(c);
- btree_node_lock_nopath_nofail(&trans, &b->c, SIX_LOCK_read);
+ btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read);
__btree_node_write_done(c, b);
six_unlock_read(&b->c.lock);
- bch2_trans_exit(&trans);
+ bch2_trans_put(trans);
}
static void btree_node_write_work(struct work_struct *work)
struct bch_fs *c = wbio->wbio.c;
struct btree *b = wbio->wbio.bio.bi_private;
struct bch_extent_ptr *ptr;
- int ret;
+ int ret = 0;
btree_bounce_free(c,
wbio->data_bytes,
}
} else {
ret = bch2_trans_do(c, NULL, NULL, 0,
- bch2_btree_node_update_key_get_iter(&trans, b, &wbio->key,
- !wbio->wbio.failed.nr));
+ bch2_btree_node_update_key_get_iter(trans, b, &wbio->key,
+ BCH_WATERMARK_reclaim|
+ BTREE_INSERT_JOURNAL_RECLAIM|
+ BTREE_INSERT_NOFAIL|
+ BTREE_INSERT_NOCHECK_RW,
+ !wbio->wbio.failed.nr));
if (ret)
goto err;
}
return;
err:
set_btree_node_noevict(b);
- bch2_fs_fatal_error(c, "fatal error writing btree node");
+ if (!bch2_err_matches(ret, EROFS))
+ bch2_fs_fatal_error(c, "fatal error writing btree node: %s", bch2_err_str(ret));
goto out;
}
struct bset *i, unsigned sectors)
{
struct printbuf buf = PRINTBUF;
+ bool saw_error;
int ret;
ret = bch2_bkey_invalid(c, bkey_i_to_s_c(&b->key),
if (ret)
return ret;
- ret = validate_bset_keys(c, b, i, WRITE, false) ?:
- validate_bset(c, NULL, b, i, b->written, sectors, WRITE, false);
+ ret = validate_bset_keys(c, b, i, WRITE, false, &saw_error) ?:
+ validate_bset(c, NULL, b, i, b->written, sectors, WRITE, false, &saw_error);
if (ret) {
bch2_inconsistent_error(c);
dump_stack();
{
struct btree_write_bio *wbio = container_of(work, struct btree_write_bio, work);
struct bch_extent_ptr *ptr;
- __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX) tmp;
+ BKEY_PADDED_ONSTACK(k, BKEY_BTREE_PTR_VAL_U64s_MAX) tmp;
bkey_copy(&tmp.k, &wbio->key);
bkey_for_each_ptr(bch2_bkey_ptrs(bkey_i_to_s(&tmp.k)), ptr)
ptr->offset += wbio->sector_offset;
- bch2_submit_wbio_replicas(&wbio->wbio, wbio->wbio.c, BCH_DATA_btree, &tmp.k);
+ bch2_submit_wbio_replicas(&wbio->wbio, wbio->wbio.c, BCH_DATA_btree,
+ &tmp.k, false);
}
void __bch2_btree_node_write(struct bch_fs *c, struct btree *b, unsigned flags)
struct bset *i;
struct btree_node *bn = NULL;
struct btree_node_entry *bne = NULL;
- struct sort_iter sort_iter;
+ struct sort_iter_stack sort_iter;
struct nonce nonce;
unsigned bytes_to_write, sectors_to_write, bytes, u64s;
u64 seq = 0;
if (old & (1 << BTREE_NODE_write_in_flight))
return;
+ if (flags & BTREE_WRITE_ONLY_IF_NEED)
+ type = new & BTREE_WRITE_TYPE_MASK;
+ new &= ~BTREE_WRITE_TYPE_MASK;
+
new &= ~(1 << BTREE_NODE_dirty);
new &= ~(1 << BTREE_NODE_need_write);
new |= (1 << BTREE_NODE_write_in_flight);
if (new & (1U << BTREE_NODE_need_write))
return;
do_write:
- if ((flags & BTREE_WRITE_ONLY_IF_NEED))
- type = b->write_type;
- b->write_type = 0;
-
BUG_ON((type == BTREE_WRITE_initial) != (b->written == 0));
atomic_dec(&c->btree_cache.dirty);
bch2_sort_whiteouts(c, b);
- sort_iter_init(&sort_iter, b);
+ sort_iter_stack_init(&sort_iter, b);
bytes = !b->written
? sizeof(struct btree_node)
continue;
bytes += le16_to_cpu(i->u64s) * sizeof(u64);
- sort_iter_add(&sort_iter,
+ sort_iter_add(&sort_iter.iter,
btree_bkey_first(b, t),
btree_bkey_last(b, t));
seq = max(seq, le64_to_cpu(i->journal_seq));
i->journal_seq = cpu_to_le64(seq);
i->u64s = 0;
- sort_iter_add(&sort_iter,
+ sort_iter_add(&sort_iter.iter,
unwritten_whiteouts_start(c, b),
unwritten_whiteouts_end(c, b));
SET_BSET_SEPARATE_WHITEOUTS(i, false);
b->whiteout_u64s = 0;
- u64s = bch2_sort_keys(i->start, &sort_iter, false);
+ u64s = bch2_sort_keys(i->start, &sort_iter.iter, false);
le16_add_cpu(&i->u64s, u64s);
BUG_ON(!b->written && i->u64s != b->data->keys.u64s);
BUG_ON(BSET_BIG_ENDIAN(i) != CPU_BIG_ENDIAN);
BUG_ON(i->seq != b->data->keys.seq);
- i->version = c->sb.version < bcachefs_metadata_version_bkey_renumber
- ? cpu_to_le16(BCH_BSET_VERSION_OLD)
- : cpu_to_le16(c->sb.version);
+ i->version = cpu_to_le16(c->sb.version);
SET_BSET_OFFSET(i, b->written);
SET_BSET_CSUM_TYPE(i, bch2_meta_checksum_type(c));
wbio = container_of(bio_alloc_bioset(NULL,
buf_pages(data, sectors_to_write << 9),
REQ_OP_WRITE|REQ_META,
- GFP_NOIO,
+ GFP_NOFS,
&c->btree_bio),
struct btree_write_bio, wbio.bio);
wbio_init(&wbio->wbio.bio);
return __bch2_btree_flush_all(c, BTREE_NODE_write_in_flight);
}
-const char * const bch2_btree_write_types[] = {
+static const char * const bch2_btree_write_types[] = {
#define x(t, n) [n] = #t,
BCH_BTREE_WRITE_TYPES()
NULL
projects / bcachefs-tools-debian / blobdiff