s64 *i_sectors_delta_total,
bool check_enospc)
{
- struct btree_iter inode_iter;
- struct bch_inode_unpacked inode_u;
+ struct btree_iter inode_iter = { NULL };
struct bpos next_pos;
bool usage_increasing;
s64 i_sectors_delta = 0, disk_sectors_delta = 0;
return ret;
}
- ret = bch2_inode_peek(trans, &inode_iter, &inode_u, inum,
- BTREE_ITER_INTENT);
- if (ret)
- return ret;
+ if (new_i_size || i_sectors_delta) {
+ struct bkey_s_c k;
+ struct bkey_s_c_inode_v3 inode;
+ struct bkey_i_inode_v3 *new_inode;
+ bool i_size_update;
+
+ bch2_trans_iter_init(trans, &inode_iter, BTREE_ID_inodes,
+ SPOS(0, inum.inum, iter->snapshot),
+ BTREE_ITER_INTENT|BTREE_ITER_CACHED);
+ k = bch2_btree_iter_peek_slot(&inode_iter);
+ ret = bkey_err(k);
+ if (unlikely(ret))
+ goto err;
+
+ ret = bkey_is_inode(k.k) ? 0 : -ENOENT;
+ if (unlikely(ret))
+ goto err;
+
+ if (unlikely(k.k->type != KEY_TYPE_inode_v3)) {
+ k = bch2_inode_to_v3(trans, k);
+ ret = bkey_err(k);
+ if (unlikely(ret))
+ goto err;
+ }
+
+ inode = bkey_s_c_to_inode_v3(k);
+ i_size_update = !(le64_to_cpu(inode.v->bi_flags) & BCH_INODE_I_SIZE_DIRTY) &&
+ new_i_size > le64_to_cpu(inode.v->bi_size);
- if (!(inode_u.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
- new_i_size > inode_u.bi_size)
- inode_u.bi_size = new_i_size;
+ if (!i_sectors_delta && !i_size_update)
+ goto no_inode_update;
- inode_u.bi_sectors += i_sectors_delta;
+ new_inode = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
+ ret = PTR_ERR_OR_ZERO(new_inode);
+ if (unlikely(ret))
+ goto err;
+ bkey_reassemble(&new_inode->k_i, k);
+
+ if (i_size_update)
+ new_inode->v.bi_size = cpu_to_le64(new_i_size);
+
+ le64_add_cpu(&new_inode->v.bi_sectors, i_sectors_delta);
+ ret = bch2_trans_update(trans, &inode_iter, &new_inode->k_i, 0);
+ if (unlikely(ret))
+ goto err;
+ }
+no_inode_update:
ret = bch2_trans_update(trans, iter, k, 0) ?:
- bch2_inode_write(trans, &inode_iter, &inode_u) ?:
bch2_trans_commit(trans, disk_res, journal_seq,
BTREE_INSERT_NOCHECK_RW|
BTREE_INSERT_NOFAIL);
- bch2_trans_iter_exit(trans, &inode_iter);
-
- if (ret)
- return ret;
+ if (unlikely(ret))
+ goto err;
if (i_sectors_delta_total)
*i_sectors_delta_total += i_sectors_delta;
bch2_btree_iter_set_pos(iter, next_pos);
-
- return 0;
+err:
+ bch2_trans_iter_exit(trans, &inode_iter);
+ return ret;
}
/*
- * Returns -EINTR if we had to drop locks:
+ * Returns -BCH_ERR_transacton_restart if we had to drop locks:
*/
int bch2_fpunch_at(struct btree_trans *trans, struct btree_iter *iter,
subvol_inum inum, u64 end,
int ret = 0, ret2 = 0;
u32 snapshot;
- while (!ret || ret == -EINTR) {
+ while (!ret ||
+ bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
struct disk_reservation disk_res =
bch2_disk_reservation_init(c, 0);
struct bkey_i delete;
bch2_trans_iter_exit(&trans, &iter);
bch2_trans_exit(&trans);
- return ret == -EINTR ? 0 : ret;
+ if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
+ ret = 0;
+
+ return ret;
}
int bch2_write_index_default(struct bch_write_op *op)
{
struct bch_fs *c = op->c;
struct bkey_buf sk;
- struct open_bucket *ec_ob = ec_open_bucket(c, &op->open_buckets);
struct keylist *keys = &op->insert_keys;
struct bkey_i *k = bch2_keylist_front(keys);
struct btree_trans trans;
ret = bch2_subvolume_get_snapshot(&trans, inum.subvol,
&sk.k->k.p.snapshot);
- if (ret == -EINTR)
+ if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
continue;
if (ret)
break;
op->flags & BCH_WRITE_CHECK_ENOSPC);
bch2_trans_iter_exit(&trans, &iter);
- if (ret == -EINTR)
+ if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
continue;
if (ret)
break;
- if (ec_ob)
- bch2_ob_add_backpointer(c, ec_ob, &sk.k->k);
-
if (bkey_cmp(iter.pos, k->k.p) >= 0)
bch2_keylist_pop_front(&op->insert_keys);
else
ca = bch_dev_bkey_exists(c, ptr->dev);
if (to_entry(ptr + 1) < ptrs.end) {
- n = to_wbio(bio_clone_fast(&wbio->bio, GFP_NOIO,
- &ca->replica_set));
+ n = to_wbio(bio_alloc_clone(NULL, &wbio->bio,
+ GFP_NOIO, &ca->replica_set));
n->bio.bi_end_io = wbio->bio.bi_end_io;
n->bio.bi_private = wbio->bio.bi_private;
}
}
-/**
- * bch_write_index - after a write, update index to point to new data
- */
-static void __bch2_write_index(struct bch_write_op *op)
+static noinline int bch2_write_drop_io_error_ptrs(struct bch_write_op *op)
{
- struct bch_fs *c = op->c;
struct keylist *keys = &op->insert_keys;
struct bch_extent_ptr *ptr;
- struct bkey_i *src, *dst = keys->keys, *n, *k;
- unsigned dev;
- int ret;
+ struct bkey_i *src, *dst = keys->keys, *n;
for (src = keys->keys; src != keys->top; src = n) {
n = bkey_next(src);
bch2_bkey_drop_ptrs(bkey_i_to_s(src), ptr,
test_bit(ptr->dev, op->failed.d));
- if (!bch2_bkey_nr_ptrs(bkey_i_to_s_c(src))) {
- ret = -EIO;
- goto err;
- }
+ if (!bch2_bkey_nr_ptrs(bkey_i_to_s_c(src)))
+ return -EIO;
}
if (dst != src)
}
keys->top = dst;
+ return 0;
+}
+
+/**
+ * bch_write_index - after a write, update index to point to new data
+ */
+static void __bch2_write_index(struct bch_write_op *op)
+{
+ struct bch_fs *c = op->c;
+ struct keylist *keys = &op->insert_keys;
+ struct bkey_i *k;
+ unsigned dev;
+ int ret;
+
+ if (unlikely(op->flags & BCH_WRITE_IO_ERROR)) {
+ ret = bch2_write_drop_io_error_ptrs(op);
+ if (ret)
+ goto err;
+ }
/*
* probably not the ideal place to hook this in, but I don't
u64 sectors_start = keylist_sectors(keys);
int ret = op->index_update_fn(op);
- BUG_ON(ret == -EINTR);
+ BUG_ON(bch2_err_matches(ret, BCH_ERR_transaction_restart));
BUG_ON(keylist_sectors(keys) && !ret);
op->written += sectors_start - keylist_sectors(keys);
if (ret) {
bch_err_inum_ratelimited(c, op->pos.inode,
- "write error %i from btree update", ret);
+ "write error while doing btree update: %s", bch2_err_str(ret));
op->error = ret;
}
}
op->pos.inode,
op->pos.offset - bio_sectors(bio), /* XXX definitely wrong */
"data write error: %s",
- bch2_blk_status_to_str(bio->bi_status)))
+ bch2_blk_status_to_str(bio->bi_status))) {
set_bit(wbio->dev, op->failed.d);
+ op->flags |= BCH_WRITE_IO_ERROR;
+ }
if (wbio->have_ioref) {
bch2_latency_acct(ca, wbio->submit_time, WRITE);
pages = min(pages, BIO_MAX_VECS);
- bio = bio_alloc_bioset(GFP_NOIO, pages, &c->bio_write);
+ bio = bio_alloc_bioset(NULL, pages, 0,
+ GFP_NOIO, &c->bio_write);
wbio = wbio_init(bio);
wbio->put_bio = true;
/* copy WRITE_SYNC flag */
struct bch_fs *c = op->c;
struct nonce nonce = extent_nonce(op->version, op->crc);
struct bch_csum csum;
+ int ret;
if (!bch2_csum_type_is_encryption(op->crc.csum_type))
return 0;
if (bch2_crc_cmp(op->crc.csum, csum))
return -EIO;
- bch2_encrypt_bio(c, op->crc.csum_type, nonce, &op->wbio.bio);
+ ret = bch2_encrypt_bio(c, op->crc.csum_type, nonce, &op->wbio.bio);
op->crc.csum_type = 0;
op->crc.csum = (struct bch_csum) { 0, 0 };
- return 0;
+ return ret;
}
static enum prep_encoded_ret {
saved_iter = dst->bi_iter;
do {
- struct bch_extent_crc_unpacked crc =
- (struct bch_extent_crc_unpacked) { 0 };
+ struct bch_extent_crc_unpacked crc = { 0 };
struct bversion version = op->version;
size_t dst_len, src_len;
!crc_is_compressed(crc) &&
bch2_csum_type_is_encryption(op->crc.csum_type) ==
bch2_csum_type_is_encryption(op->csum_type)) {
+ u8 compression_type = crc.compression_type;
+ u16 nonce = crc.nonce;
/*
* Note: when we're using rechecksum(), we need to be
* checksumming @src because it has all the data our
bio_sectors(src) - (src_len >> 9),
op->csum_type))
goto csum_err;
+ /*
+ * rchecksum_bio sets compression_type on crc from op->crc,
+ * this isn't always correct as sometimes we're changing
+ * an extent from uncompressed to incompressible.
+ */
+ crc.compression_type = compression_type;
+ crc.nonce = nonce;
} else {
if ((op->flags & BCH_WRITE_DATA_ENCODED) &&
bch2_rechecksum_bio(c, src, version, op->crc,
crc.live_size = src_len >> 9;
swap(dst->bi_iter.bi_size, dst_len);
- bch2_encrypt_bio(c, op->csum_type,
- extent_nonce(version, crc), dst);
+ ret = bch2_encrypt_bio(c, op->csum_type,
+ extent_nonce(version, crc), dst);
+ if (ret)
+ goto err;
+
crc.csum = bch2_checksum_bio(c, op->csum_type,
extent_nonce(version, crc), dst);
crc.csum_type = op->csum_type;
*_dst = dst;
return more;
csum_err:
- bch_err(c, "error verifying existing checksum while "
- "rewriting existing data (memory corruption?)");
+ bch_err(c, "error verifying existing checksum while rewriting existing data (memory corruption?)");
ret = -EIO;
err:
if (to_wbio(dst)->bounce)
struct bch_write_op *op = container_of(cl, struct bch_write_op, cl);
struct bch_fs *c = op->c;
struct write_point *wp;
- struct bio *bio;
+ struct bio *bio = NULL;
bool skip_put = true;
unsigned nofs_flags;
int ret;
BKEY_EXTENT_U64s_MAX))
goto flush_io;
- if ((op->flags & BCH_WRITE_FROM_INTERNAL) &&
- percpu_ref_is_dying(&c->writes)) {
- ret = -EROFS;
- goto err;
- }
-
/*
* The copygc thread is now global, which means it's no longer
* freeing up space on specific disks, which means that
BCH_WRITE_ONLY_SPECIFIED_DEVS)) ? NULL : cl);
EBUG_ON(!wp);
- if (unlikely(IS_ERR(wp))) {
- if (unlikely(PTR_ERR(wp) != -EAGAIN)) {
+ if (IS_ERR(wp)) {
+ if (unlikely(wp != ERR_PTR(-EAGAIN))) {
ret = PTR_ERR(wp);
goto err;
}
}
if (c->opts.nochanges ||
- !percpu_ref_tryget(&c->writes)) {
+ !percpu_ref_tryget_live(&c->writes)) {
op->error = -EROFS;
goto err;
}
+ this_cpu_add(c->counters[BCH_COUNTER_io_write], bio_sectors(bio));
bch2_increment_clock(c, bio_sectors(bio), WRITE);
data_len = min_t(u64, bio->bi_iter.bi_size,
struct rhash_head hash;
struct bpos pos;
- struct migrate_write write;
+ struct data_update write;
struct bio_vec bi_inline_vecs[0]; /* must be last */
};
bch2_time_stats_update(&c->times[BCH_TIME_data_promote],
op->start_time);
- bch2_bio_free_pages_pool(c, &op->write.op.wbio.bio);
+ bch2_data_update_exit(&op->write);
promote_free(c, op);
}
static void promote_start(struct promote_op *op, struct bch_read_bio *rbio)
{
- struct bch_fs *c = rbio->c;
struct closure *cl = &op->cl;
struct bio *bio = &op->write.op.wbio.bio;
- trace_promote(&rbio->bio);
+ trace_and_count(op->write.op.c, read_promote, &rbio->bio);
/* we now own pages: */
BUG_ON(!rbio->bounce);
sizeof(struct bio_vec) * rbio->bio.bi_vcnt);
swap(bio->bi_vcnt, rbio->bio.bi_vcnt);
- bch2_migrate_read_done(&op->write, rbio);
-
closure_init(cl, NULL);
- closure_call(&op->write.op.cl, bch2_write, c->btree_update_wq, cl);
+ bch2_data_update_read_done(&op->write, rbio->pick.crc, cl);
closure_return_with_destructor(cl, promote_done);
}
unsigned pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
int ret;
- if (!percpu_ref_tryget(&c->writes))
+ if (!percpu_ref_tryget_live(&c->writes))
return NULL;
op = kzalloc(sizeof(*op) + sizeof(struct bio_vec) * pages, GFP_NOIO);
goto err;
rbio_init(&(*rbio)->bio, opts);
- bio_init(&(*rbio)->bio, (*rbio)->bio.bi_inline_vecs, pages);
+ bio_init(&(*rbio)->bio, NULL, (*rbio)->bio.bi_inline_vecs, pages, 0);
if (bch2_bio_alloc_pages(&(*rbio)->bio, sectors << 9,
GFP_NOIO))
goto err;
bio = &op->write.op.wbio.bio;
- bio_init(bio, bio->bi_inline_vecs, pages);
+ bio_init(bio, NULL, bio->bi_inline_vecs, pages, 0);
- ret = bch2_migrate_write_init(c, &op->write,
+ ret = bch2_data_update_init(c, &op->write,
writepoint_hashed((unsigned long) current),
opts,
- DATA_PROMOTE,
- (struct data_opts) {
+ (struct data_update_opts) {
.target = opts.promote_target,
- .nr_replicas = 1,
+ .extra_replicas = 1,
+ .write_flags = BCH_WRITE_ALLOC_NOWAIT|BCH_WRITE_CACHED,
},
btree_id, k);
BUG_ON(ret);
};
struct bch_io_failures failed = { .nr = 0 };
- trace_read_retry(&rbio->bio);
+ trace_and_count(c, read_retry, &rbio->bio);
if (rbio->retry == READ_RETRY_AVOID)
bch2_mark_io_failure(&failed, &rbio->pick);
struct nonce nonce = extent_nonce(rbio->version, crc);
unsigned nofs_flags;
struct bch_csum csum;
+ int ret;
nofs_flags = memalloc_nofs_save();
crc.live_size = bvec_iter_sectors(rbio->bvec_iter);
if (crc_is_compressed(crc)) {
- bch2_encrypt_bio(c, crc.csum_type, nonce, src);
+ ret = bch2_encrypt_bio(c, crc.csum_type, nonce, src);
+ if (ret)
+ goto decrypt_err;
+
if (bch2_bio_uncompress(c, src, dst, dst_iter, crc))
goto decompression_err;
} else {
BUG_ON(src->bi_iter.bi_size < dst_iter.bi_size);
src->bi_iter.bi_size = dst_iter.bi_size;
- bch2_encrypt_bio(c, crc.csum_type, nonce, src);
+ ret = bch2_encrypt_bio(c, crc.csum_type, nonce, src);
+ if (ret)
+ goto decrypt_err;
if (rbio->bounce) {
struct bvec_iter src_iter = src->bi_iter;
* Re encrypt data we decrypted, so it's consistent with
* rbio->crc:
*/
- bch2_encrypt_bio(c, crc.csum_type, nonce, src);
+ ret = bch2_encrypt_bio(c, crc.csum_type, nonce, src);
+ if (ret)
+ goto decrypt_err;
+
promote_start(rbio->promote, rbio);
rbio->promote = NULL;
}
}
bch2_dev_inum_io_error(ca, rbio->read_pos.inode, (u64) rbio->bvec_iter.bi_sector,
- "data checksum error: expected %0llx:%0llx got %0llx:%0llx (type %u)",
+ "data checksum error: expected %0llx:%0llx got %0llx:%0llx (type %s)",
rbio->pick.crc.csum.hi, rbio->pick.crc.csum.lo,
- csum.hi, csum.lo, crc.csum_type);
+ csum.hi, csum.lo, bch2_csum_types[crc.csum_type]);
bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR);
goto out;
decompression_err:
"decompression error");
bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR);
goto out;
+decrypt_err:
+ bch_err_inum_ratelimited(c, rbio->read_pos.inode,
+ "decrypt error");
+ bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR);
+ goto out;
}
static void bch2_read_endio(struct bio *bio)
if (((rbio->flags & BCH_READ_RETRY_IF_STALE) && race_fault()) ||
ptr_stale(ca, &rbio->pick.ptr)) {
- atomic_long_inc(&c->read_realloc_races);
+ trace_and_count(c, read_reuse_race, &rbio->bio);
if (rbio->flags & BCH_READ_RETRY_IF_STALE)
bch2_rbio_error(rbio, READ_RETRY, BLK_STS_AGAIN);
}
if (rbio->narrow_crcs ||
+ rbio->promote ||
crc_is_compressed(rbio->pick.crc) ||
bch2_csum_type_is_encryption(rbio->pick.crc.csum_type))
context = RBIO_CONTEXT_UNBOUND, wq = system_unbound_wq;
struct bch_fs *c = trans->c;
struct bch_dev *ca = bch_dev_bkey_exists(c, ptr.dev);
struct btree_iter iter;
- char buf[200];
+ struct printbuf buf = PRINTBUF;
int ret;
- bch2_bkey_val_to_text(&PBUF(buf), c, k);
- bch2_fs_inconsistent(c, "Attempting to read from stale dirty pointer: %s", buf);
-
bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
- POS(ptr.dev, PTR_BUCKET_NR(ca, &ptr)),
+ PTR_BUCKET_POS(c, &ptr),
BTREE_ITER_CACHED);
+ prt_printf(&buf, "Attempting to read from stale dirty pointer:");
+ printbuf_indent_add(&buf, 2);
+ prt_newline(&buf);
+
+ bch2_bkey_val_to_text(&buf, c, k);
+ prt_newline(&buf);
+
+ prt_printf(&buf, "memory gen: %u", *bucket_gen(ca, iter.pos.offset));
+
ret = lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_slot(&iter)));
- if (ret)
- return;
+ if (!ret) {
+ prt_newline(&buf);
+ bch2_bkey_val_to_text(&buf, c, k);
+ }
+
+ bch2_fs_inconsistent(c, "%s", buf.buf);
- bch2_bkey_val_to_text(&PBUF(buf), c, k);
- bch_err(c, "%s", buf);
- bch_err(c, "memory gen: %u", *bucket_gen(ca, iter.pos.offset));
bch2_trans_iter_exit(trans, &iter);
+ printbuf_exit(&buf);
}
int __bch2_read_extent(struct btree_trans *trans, struct bch_read_bio *orig,
ca = bch_dev_bkey_exists(c, pick.ptr.dev);
- if (!pick.ptr.cached &&
+ /*
+ * Stale dirty pointers are treated as IO errors, but @failed isn't
+ * allocated unless we're in the retry path - so if we're not in the
+ * retry path, don't check here, it'll be caught in bch2_read_endio()
+ * and we'll end up in the retry path:
+ */
+ if ((flags & BCH_READ_IN_RETRY) &&
+ !pick.ptr.cached &&
unlikely(ptr_stale(ca, &pick.ptr))) {
read_from_stale_dirty_pointer(trans, k, pick.ptr);
bch2_mark_io_failure(failed, &pick);
} else if (bounce) {
unsigned sectors = pick.crc.compressed_size;
- rbio = rbio_init(bio_alloc_bioset(GFP_NOIO,
+ rbio = rbio_init(bio_alloc_bioset(NULL,
DIV_ROUND_UP(sectors, PAGE_SECTORS),
+ 0,
+ GFP_NOIO,
&c->bio_read_split),
orig->opts);
* from the whole bio, in which case we don't want to retry and
* lose the error)
*/
- rbio = rbio_init(bio_clone_fast(&orig->bio, GFP_NOIO,
- &c->bio_read_split),
+ rbio = rbio_init(bio_alloc_clone(NULL, &orig->bio, GFP_NOIO,
+ &c->bio_read_split),
orig->opts);
rbio->bio.bi_iter = iter;
rbio->split = true;
rbio->bio.bi_end_io = bch2_read_endio;
if (rbio->bounce)
- trace_read_bounce(&rbio->bio);
+ trace_and_count(c, read_bounce, &rbio->bio);
+ this_cpu_add(c->counters[BCH_COUNTER_io_read], bio_sectors(&rbio->bio));
bch2_increment_clock(c, bio_sectors(&rbio->bio), READ);
/*
if (!(flags & (BCH_READ_IN_RETRY|BCH_READ_LAST_FRAGMENT))) {
bio_inc_remaining(&orig->bio);
- trace_read_split(&orig->bio);
+ trace_and_count(c, read_split, &orig->bio);
}
if (!rbio->pick.idx) {
* read_extent -> io_time_reset may cause a transaction restart
* without returning an error, we need to check for that here:
*/
- if (!bch2_trans_relock(&trans)) {
- ret = -EINTR;
+ ret = bch2_trans_relock(&trans);
+ if (ret)
break;
- }
bch2_btree_iter_set_pos(&iter,
POS(inum.inum, bvec_iter.bi_sector));
err:
bch2_trans_iter_exit(&trans, &iter);
- if (ret == -EINTR || ret == READ_RETRY || ret == READ_RETRY_AVOID)
+ if (bch2_err_matches(ret, BCH_ERR_transaction_restart) ||
+ ret == READ_RETRY ||
+ ret == READ_RETRY_AVOID)
goto retry;
bch2_trans_exit(&trans);