X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fbtree_io.c;h=f0cbc9158e4cbdc3f985f72c016d3bb97365823d;hb=f76d246fff1c3c5b091fb2b910067d893620dc5e;hp=177fd49d65ddc80b6c4856ce5c9e265b7c5931e8;hpb=d01f633041c50452f9e837b7cc0223e6f37d42da;p=bcachefs-tools-debian diff --git a/libbcachefs/btree_io.c b/libbcachefs/btree_io.c index 177fd49..f0cbc91 100644 --- a/libbcachefs/btree_io.c +++ b/libbcachefs/btree_io.c @@ -14,13 +14,14 @@ #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 -#include void bch2_btree_node_io_unlock(struct btree *b) { @@ -33,7 +34,7 @@ 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); @@ -53,7 +54,7 @@ void __bch2_btree_node_wait_on_write(struct btree *b) 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); @@ -61,7 +62,7 @@ void bch2_btree_node_wait_on_read(struct btree *b) 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); @@ -77,13 +78,13 @@ static void verify_no_dups(struct btree *b, 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 } @@ -92,7 +93,7 @@ static void set_needs_whiteout(struct bset *i, int v) { 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; } @@ -117,7 +118,7 @@ static void *btree_bounce_alloc(struct bch_fs *c, size_t size, 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; @@ -175,7 +176,7 @@ static void bch2_sort_whiteouts(struct bch_fs *c, struct btree *b) 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); @@ -183,8 +184,8 @@ static void bch2_sort_whiteouts(struct bch_fs *c, struct btree *b) k = new_whiteouts; while (ptrs != ptrs_end) { - bkey_copy(k, *ptrs); - k = bkey_next(k); + bkey_p_copy(k, *ptrs); + k = bkey_p_next(k); ptrs++; } @@ -256,11 +257,11 @@ static bool bch2_drop_whiteouts(struct btree *b, enum compact_mode mode) 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); + bkey_p_copy(out, k); + out = bkey_p_next(out); } else { BUG_ON(k->needs_whiteout); } @@ -291,7 +292,7 @@ static void btree_node_sort(struct bch_fs *c, struct btree *b, 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; @@ -300,13 +301,13 @@ static void btree_node_sort(struct bch_fs *c, struct btree *b, 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)); } @@ -319,7 +320,7 @@ static void btree_node_sort(struct bch_fs *c, struct btree *b, 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); @@ -335,7 +336,7 @@ static void btree_node_sort(struct bch_fs *c, struct btree *b, 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)); @@ -409,8 +410,6 @@ void bch2_btree_sort_into(struct bch_fs *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: @@ -450,6 +449,24 @@ void bch2_btree_build_aux_trees(struct btree *b) t == bset_tree_last(b)); } +/* + * If we have MAX_BSETS (3) bsets, should we sort them all down to just one? + * + * The first bset is going to be of similar order to the size of the node, the + * last bset is bounded by btree_write_set_buffer(), which is set to keep the + * memmove on insert from being too expensive: the middle bset should, ideally, + * be the geometric mean of the first and the last. + * + * Returns true if the middle bset is greater than that geometric mean: + */ +static inline bool should_compact_all(struct bch_fs *c, struct btree *b) +{ + unsigned mid_u64s_bits = + (ilog2(btree_max_u64s(c)) + BTREE_WRITE_SET_U64s_BITS) / 2; + + return bset_u64s(&b->set[1]) > 1U << mid_u64s_bits; +} + /* * @bch_btree_init_next - initialize a new (unwritten) bset that can then be * inserted into @@ -465,21 +482,16 @@ void bch2_btree_init_next(struct btree_trans *trans, struct btree *b) 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)); if (b->nsets == MAX_BSETS && - !btree_node_write_in_flight(b)) { - unsigned log_u64s[] = { - ilog2(bset_u64s(&b->set[0])), - ilog2(bset_u64s(&b->set[1])), - ilog2(bset_u64s(&b->set[2])), - }; - - if (log_u64s[1] >= (log_u64s[0] + log_u64s[2]) / 2) { - bch2_btree_node_write(c, b, SIX_LOCK_write, 0); - reinit_iter = true; - } + !btree_node_write_in_flight(b) && + should_compact_all(c, b)) { + bch2_btree_node_write(c, b, SIX_LOCK_write, + BTREE_WRITE_init_next_bset); + reinit_iter = true; } if (b->nsets == MAX_BSETS && @@ -498,94 +510,104 @@ void bch2_btree_init_next(struct btree_trans *trans, struct btree *b) bch2_trans_node_reinit_iter(trans, b); } -static void btree_pos_to_text(struct printbuf *out, struct bch_fs *c, - struct btree *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_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key)); -} - static void btree_err_msg(struct printbuf *out, struct bch_fs *c, struct bch_dev *ca, struct btree *b, struct bset *i, unsigned offset, int write) { - prt_printf(out, "error validating btree node "); - if (write) - prt_printf(out, "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 "); - btree_pos_to_text(out, c, b); + bch2_btree_pos_to_text(out, c, b); prt_printf(out, "\n node offset %u", b->written); if (i) prt_printf(out, " bset u64s %u", le16_to_cpu(i->u64s)); + prt_str(out, ": "); } -enum btree_err_type { - BTREE_ERR_FIXABLE, - BTREE_ERR_WANT_RETRY, - BTREE_ERR_MUST_RETRY, - BTREE_ERR_FATAL, -}; +__printf(9, 10) +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, + enum bch_sb_error_id err_type, + 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; + + if (ret != -BCH_ERR_btree_node_read_err_fixable) + bch2_sb_error_count(c, err_type); + + switch (ret) { + case -BCH_ERR_btree_node_read_err_fixable: + ret = bch2_fsck_err(c, FSCK_CAN_FIX, err_type, "%s", out.buf); + if (ret != -BCH_ERR_fsck_fix && + ret != -BCH_ERR_fsck_ignore) + goto fsck_err; + 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, ...) \ +#define btree_err(type, c, ca, b, i, _err_type, 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, \ + BCH_FSCK_ERR_##_err_type, \ + 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; \ } \ \ - switch (write) { \ - case READ: \ - bch_err(c, "%s", out.buf); \ - \ - 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: \ - bch_err(c, "corrupt metadata before write: %s", out.buf);\ - \ - 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) @@ -594,18 +616,16 @@ out: \ * 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; @@ -618,7 +638,7 @@ void bch2_btree_node_drop_keys_outside_node(struct btree *b) 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; @@ -635,31 +655,37 @@ void bch2_btree_node_drop_keys_outside_node(struct btree *b) 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, + btree_node_unsupported_version, + "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, + btree_node_bset_older_than_sb_min, "bset version %u older than superblock version_min %u", version, c->sb.version_min)) { mutex_lock(&c->sb_lock); @@ -668,8 +694,11 @@ static int validate_bset(struct bch_fs *c, struct bch_dev *ca, 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, + btree_node_bset_newer_than_sb, "bset version %u newer than superblock version %u", version, c->sb.version)) { mutex_lock(&c->sb_lock); @@ -679,11 +708,15 @@ static int validate_bset(struct bch_fs *c, struct bch_dev *ca, } 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, + btree_node_unsupported_version, "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, "bset past end of btree node")) { i->u64s = 0; ret = 0; @@ -691,12 +724,15 @@ static int validate_bset(struct bch_fs *c, struct bch_dev *ca, } btree_err_on(offset && !i->u64s, - BTREE_ERR_FIXABLE, c, ca, b, i, + -BCH_ERR_btree_node_read_err_fixable, + c, ca, b, i, + bset_empty, "empty bset"); - btree_err_on(BSET_OFFSET(i) && - BSET_OFFSET(i) != offset, - BTREE_ERR_WANT_RETRY, c, ca, b, i, + btree_err_on(BSET_OFFSET(i) && BSET_OFFSET(i) != offset, + -BCH_ERR_btree_node_read_err_want_retry, + c, ca, b, i, + bset_wrong_sector_offset, "bset at wrong sector offset"); if (!offset) { @@ -710,16 +746,22 @@ static int validate_bset(struct bch_fs *c, struct bch_dev *ca, /* 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, + bset_bad_seq, "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, + btree_node_bad_btree, "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, + btree_node_bad_level, "incorrect level"); if (!write) @@ -735,8 +777,10 @@ static int validate_bset(struct bch_fs *c, struct bch_dev *ca, 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, + btree_node_bad_min_key, "incorrect min_key: got %s should be %s", (printbuf_reset(&buf1), bch2_bpos_to_text(&buf1, bn->min_key), buf1.buf), @@ -744,8 +788,10 @@ static int validate_bset(struct bch_fs *c, struct bch_dev *ca, 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, + btree_node_bad_max_key, "incorrect max key %s", (printbuf_reset(&buf1), bch2_bpos_to_text(&buf1, bn->max_key), buf1.buf)); @@ -754,10 +800,14 @@ static int validate_bset(struct bch_fs *c, struct bch_dev *ca, 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, + btree_node_bad_format, + "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, @@ -776,12 +826,13 @@ static int bset_key_invalid(struct bch_fs *c, struct btree *b, struct printbuf *err) { return __bch2_bkey_invalid(c, k, btree_node_type(b), READ, err) ?: - (!updated_range ? bch2_bkey_in_btree_node(b, k, err) : 0) ?: + (!updated_range ? bch2_bkey_in_btree_node(c, b, k, err) : 0) ?: (rw == WRITE ? bch2_bkey_val_invalid(c, k, READ, err) : 0); } 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; @@ -795,18 +846,22 @@ static int validate_bset_keys(struct bch_fs *c, struct btree *b, 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, + btree_node_bkey_past_bset_end, "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, + btree_node_bkey_bad_format, "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; } @@ -822,15 +877,17 @@ static int validate_bset_keys(struct bch_fs *c, struct btree *b, printbuf_reset(&buf); if (bset_key_invalid(c, b, u.s_c, updated_range, write, &buf)) { printbuf_reset(&buf); - prt_printf(&buf, "invalid bkey: "); bset_key_invalid(c, b, u.s_c, updated_range, write, &buf); 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, + btree_node_bad_bkey, + "invalid bkey: %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; } @@ -851,16 +908,19 @@ static int validate_bset_keys(struct bch_fs *c, struct btree *b, 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, + btree_node_bkey_out_of_order, + "%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); @@ -868,7 +928,7 @@ fsck_err: } 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; @@ -880,61 +940,82 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca, 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, + btree_node_fault_injected, "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, + btree_node_bad_magic, "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, - "bad btree header: seq 0"); - if (b->key.k.type == KEY_TYPE_btree_ptr_v2) { struct bch_btree_ptr_v2 *bp = &bkey_i_to_btree_ptr_v2(&b->key)->v; + bch2_bpos_to_text(&buf, b->data->min_key); + prt_str(&buf, "-"); + bch2_bpos_to_text(&buf, b->data->max_key); + btree_err_on(b->data->keys.seq != bp->seq, - BTREE_ERR_MUST_RETRY, c, ca, b, NULL, - "got wrong btree node (seq %llx want %llx)", - b->data->keys.seq, bp->seq); + -BCH_ERR_btree_node_read_err_must_retry, + c, ca, b, NULL, + btree_node_bad_seq, + "got wrong btree node (want %llx got %llx)\n" + "got btree %s level %llu pos %s", + bp->seq, b->data->keys.seq, + bch2_btree_id_str(BTREE_NODE_ID(b->data)), + BTREE_NODE_LEVEL(b->data), + buf.buf); + } else { + btree_err_on(!b->data->keys.seq, + -BCH_ERR_btree_node_read_err_must_retry, + c, ca, b, NULL, + btree_node_bad_seq, + "bad btree header: seq 0"); } while (b->written < (ptr_written ?: btree_sectors(c))) { unsigned sectors; struct nonce nonce; - struct bch_csum csum; bool first = !b->written; + bool csum_bad; if (!b->written) { i = &b->data->keys; btree_err_on(!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i)), - BTREE_ERR_WANT_RETRY, c, ca, b, i, - "unknown checksum type %llu", - BSET_CSUM_TYPE(i)); + -BCH_ERR_btree_node_read_err_want_retry, + c, ca, b, i, + bset_unknown_csum, + "unknown checksum type %llu", BSET_CSUM_TYPE(i)); nonce = btree_nonce(i, b->written << 9); - 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, + csum_bad = bch2_crc_cmp(b->data->csum, + csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, b->data)); + if (csum_bad) + bch2_io_error(ca, BCH_MEMBER_ERROR_checksum); + + btree_err_on(csum_bad, + -BCH_ERR_btree_node_read_err_want_retry, + c, ca, b, i, + bset_bad_csum, "invalid checksum"); ret = bset_encrypt(c, i, b->written << 9); @@ -944,7 +1025,9 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca, 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_unsupported_version, "btree node does not have NEW_EXTENT_OVERWRITE set"); sectors = vstruct_sectors(b->data, c->block_bits); @@ -956,15 +1039,21 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca, break; btree_err_on(!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i)), - BTREE_ERR_WANT_RETRY, c, ca, b, i, - "unknown checksum type %llu", - BSET_CSUM_TYPE(i)); + -BCH_ERR_btree_node_read_err_want_retry, + c, ca, b, i, + bset_unknown_csum, + "unknown checksum type %llu", BSET_CSUM_TYPE(i)); nonce = btree_nonce(i, b->written << 9); - 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, + csum_bad = bch2_crc_cmp(bne->csum, + csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bne)); + if (csum_bad) + bch2_io_error(ca, BCH_MEMBER_ERROR_checksum); + + btree_err_on(csum_bad, + -BCH_ERR_btree_node_read_err_want_retry, + c, ca, b, i, + bset_bad_csum, "invalid checksum"); ret = bset_encrypt(c, i, b->written << 9); @@ -979,14 +1068,14 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca, 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; @@ -997,12 +1086,16 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca, true); btree_err_on(blacklisted && first, - BTREE_ERR_FIXABLE, c, ca, b, i, + -BCH_ERR_btree_node_read_err_fixable, + c, ca, b, i, + bset_blacklisted_journal_seq, "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, + first_bset_blacklisted_journal_seq, "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); @@ -1015,13 +1108,13 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca, 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, "btree node data missing: expected %u sectors, found %u", ptr_written, b->written); } else { @@ -1032,20 +1125,10 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca, !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, + btree_node_bset_after_end, "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); @@ -1086,12 +1169,15 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca, 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, + btree_node_bad_bkey, + "%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; @@ -1103,7 +1189,7 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca, bp.v->mem_ptr = 0; } - k = bkey_next(k); + k = bkey_p_next(k); } bch2_bset_build_aux_tree(b, b->set, false); @@ -1113,9 +1199,9 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca, 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); } @@ -1126,12 +1212,11 @@ out: 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; } @@ -1167,8 +1252,9 @@ static void btree_node_read_work(struct work_struct *work) } start: printbuf_reset(&buf); - btree_pos_to_text(&buf, c, b); - bch2_dev_io_err_on(bio->bi_status, ca, "btree read error %s for %s", + bch2_btree_pos_to_text(&buf, c, b); + bch2_dev_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_read, + "btree read error %s for %s", bch2_blk_status_to_str(bio->bi_status), buf.buf); if (rb->have_ioref) percpu_ref_put(&ca->io_ref); @@ -1181,7 +1267,7 @@ start: &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; @@ -1198,11 +1284,17 @@ start: 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_id_str(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); } @@ -1215,6 +1307,7 @@ static void btree_node_read_endio(struct bio *bio) if (rb->have_ioref) { struct bch_dev *ca = bch_dev_bkey_exists(c, rb->pick.ptr.dev); + bch2_latency_acct(ca, rb->start_time, READ); } @@ -1228,7 +1321,7 @@ struct btree_node_read_all { 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) @@ -1273,10 +1366,9 @@ static bool btree_node_has_extra_bsets(struct bch_fs *c, unsigned offset, void * return offset; } -static void btree_node_read_all_replicas_done(struct closure *cl) +static CLOSURE_CALLBACK(btree_node_read_all_replicas_done) { - struct btree_node_read_all *ra = - container_of(cl, struct btree_node_read_all, cl); + closure_type(ra, struct btree_node_read_all, cl); struct bch_fs *c = ra->c; struct btree *b = ra->b; struct printbuf buf = PRINTBUF; @@ -1286,6 +1378,7 @@ static void btree_node_read_all_replicas_done(struct closure *cl) 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]; @@ -1304,14 +1397,20 @@ static void btree_node_read_all_replicas_done(struct closure *cl) } 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_replicas_sectors_written_mismatch, "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, + btree_node_bset_after_end, "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_data_mismatch, "btree node replicas content mismatch")) dump_bset_maps = true; @@ -1372,13 +1471,15 @@ fsck_err: 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); @@ -1402,6 +1503,7 @@ static void btree_node_read_all_replicas_endio(struct bio *bio) if (rb->have_ioref) { struct bch_dev *ca = bch_dev_bkey_exists(c, rb->pick.ptr.dev); + bch2_latency_acct(ca, rb->start_time, READ); } @@ -1424,7 +1526,7 @@ static int btree_node_read_all_replicas(struct bch_fs *c, struct btree *b, bool 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; @@ -1472,7 +1574,7 @@ static int btree_node_read_all_replicas(struct bch_fs *c, struct btree *b, bool if (sync) { closure_sync(&ra->cl); - btree_node_read_all_replicas_done(&ra->cl); + btree_node_read_all_replicas_done(&ra->cl.work); } else { continue_at(&ra->cl, btree_node_read_all_replicas_done, c->io_complete_wq); @@ -1481,16 +1583,17 @@ static int btree_node_read_all_replicas(struct bch_fs *c, struct btree *b, bool return 0; } -void bch2_btree_node_read(struct bch_fs *c, struct btree *b, +void bch2_btree_node_read(struct btree_trans *trans, struct btree *b, bool sync) { + struct bch_fs *c = trans->c; struct extent_ptr_decoded pick; struct btree_read_bio *rb; struct bch_dev *ca; struct bio *bio; int ret; - trace_and_count(c, btree_node_read, c, b); + trace_and_count(c, btree_node_read, trans, b); if (bch2_verify_all_btree_replicas && !btree_node_read_all_replicas(c, b, sync)) @@ -1503,10 +1606,11 @@ void bch2_btree_node_read(struct bch_fs *c, struct btree *b, struct printbuf buf = PRINTBUF; prt_str(&buf, "btree node read error: no device to read from\n at "); - btree_pos_to_text(&buf, c, b); + bch2_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); @@ -1521,7 +1625,7 @@ void bch2_btree_node_read(struct bch_fs *c, struct btree *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; @@ -1557,9 +1661,10 @@ void bch2_btree_node_read(struct bch_fs *c, struct btree *b, } } -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; @@ -1567,12 +1672,12 @@ int bch2_btree_root_read(struct bch_fs *c, enum btree_id id, closure_init_stack(&cl); do { - ret = bch2_btree_cache_cannibalize_lock(c, &cl); + ret = bch2_btree_cache_cannibalize_lock(trans, &cl); closure_sync(&cl); } while (ret); - b = bch2_btree_node_mem_alloc(c, level != 0); - bch2_btree_cache_cannibalize_unlock(c); + b = bch2_btree_node_mem_alloc(trans, level != 0); + bch2_btree_cache_cannibalize_unlock(trans); BUG_ON(IS_ERR(b)); @@ -1581,7 +1686,7 @@ int bch2_btree_root_read(struct bch_fs *c, enum btree_id id, set_btree_node_read_in_flight(b); - bch2_btree_node_read(c, b, true); + bch2_btree_node_read(trans, b, true); if (btree_node_read_error(b)) { bch2_btree_node_hash_remove(&c->btree_cache, b); @@ -1602,8 +1707,14 @@ err: return ret; } -void bch2_btree_complete_write(struct bch_fs *c, struct btree *b, - struct btree_write *w) +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)); +} + +static void bch2_btree_complete_write(struct bch_fs *c, struct btree *b, + struct btree_write *w) { unsigned long old, new, v = READ_ONCE(b->will_make_reachable); @@ -1625,6 +1736,7 @@ static void __btree_node_write_done(struct bch_fs *c, struct btree *b) { struct btree_write *w = btree_prev_write(b); unsigned long old, new, v; + unsigned type = 0; bch2_btree_complete_write(c, b, w); @@ -1643,6 +1755,9 @@ static void __btree_node_write_done(struct bch_fs *c, struct btree *b) 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); @@ -1650,22 +1765,20 @@ static void __btree_node_write_done(struct bch_fs *c, struct btree *b) } 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); + __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) @@ -1675,7 +1788,7 @@ 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, @@ -1694,8 +1807,12 @@ static void btree_node_write_work(struct work_struct *work) } } 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| + BCH_TRANS_COMMIT_journal_reclaim| + BCH_TRANS_COMMIT_no_enospc| + BCH_TRANS_COMMIT_no_check_rw, + !wbio->wbio.failed.nr)); if (ret) goto err; } @@ -1705,7 +1822,8 @@ out: 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; } @@ -1723,7 +1841,8 @@ static void btree_node_write_endio(struct bio *bio) if (wbio->have_ioref) bch2_latency_acct(ca, wbio->submit_time, WRITE); - if (bch2_dev_io_err_on(bio->bi_status, ca, "btree write error: %s", + if (bch2_dev_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_write, + "btree write error: %s", bch2_blk_status_to_str(bio->bi_status)) || bch2_meta_write_fault("btree")) { spin_lock_irqsave(&c->btree_write_error_lock, flags); @@ -1750,6 +1869,7 @@ static int validate_bset_for_write(struct bch_fs *c, struct btree *b, 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), @@ -1761,8 +1881,8 @@ static int validate_bset_for_write(struct bch_fs *c, struct btree *b, 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(); @@ -1775,14 +1895,15 @@ static void btree_write_submit(struct work_struct *work) { 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) @@ -1792,13 +1913,14 @@ 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; bool used_mempool; unsigned long old, new; bool validate_before_checksum = false; + enum btree_write_type type = flags & BTREE_WRITE_TYPE_MASK; void *data; int ret; @@ -1834,6 +1956,10 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b, unsigned flags) 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); @@ -1845,6 +1971,8 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b, unsigned flags) if (new & (1U << BTREE_NODE_need_write)) return; do_write: + BUG_ON((type == BTREE_WRITE_initial) != (b->written == 0)); + atomic_dec(&c->btree_cache.dirty); BUG_ON(btree_node_fake(b)); @@ -1858,7 +1986,7 @@ do_write: 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) @@ -1873,7 +2001,7 @@ do_write: 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)); @@ -1887,6 +2015,29 @@ do_write: /* buffer must be a multiple of the block size */ bytes = round_up(bytes, block_bytes(c)); + if (bytes > btree_bytes(c)) { + struct printbuf buf = PRINTBUF; + + prt_printf(&buf, "btree node write bounce buffer overrun: %u > %zu\n", + bytes, btree_bytes(c)); + + prt_printf(&buf, "header: %zu\n", b->written + ? sizeof(struct btree_node) + : sizeof(struct btree_node_entry)); + prt_printf(&buf, "unwritten: %zu\n", b->whiteout_u64s * sizeof(u64)); + + for_each_bset(b, t) { + i = bset(b, t); + + if (bset_written(b, i)) + continue; + prt_printf(&buf, "bset %zu: %zu\n", t - b->set, le16_to_cpu(i->u64s) * sizeof(u64)); + } + + panic("%s", buf.buf); + printbuf_exit(&buf); + } + data = btree_bounce_alloc(c, bytes, &used_mempool); if (!b->written) { @@ -1902,16 +2053,18 @@ do_write: 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); + set_needs_whiteout(i, false); /* do we have data to write? */ @@ -1921,6 +2074,10 @@ do_write: bytes_to_write = vstruct_end(i) - data; sectors_to_write = round_up(bytes_to_write, block_bytes(c)) >> 9; + if (!b->written && + b->key.k.type == KEY_TYPE_btree_ptr_v2) + BUG_ON(btree_ptr_sectors_written(&b->key) != sectors_to_write); + memset(data + bytes_to_write, 0, (sectors_to_write << 9) - bytes_to_write); @@ -1928,9 +2085,7 @@ do_write: 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)); @@ -1990,7 +2145,7 @@ do_write: 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); @@ -2009,27 +2164,18 @@ do_write: b->written += sectors_to_write; - if (wbio->wbio.first_btree_write && - b->key.k.type == KEY_TYPE_btree_ptr_v2) - bkey_i_to_btree_ptr_v2(&b->key)->v.sectors_written = - cpu_to_le16(b->written); - if (wbio->key.k.type == KEY_TYPE_btree_ptr_v2) bkey_i_to_btree_ptr_v2(&wbio->key)->v.sectors_written = cpu_to_le16(b->written); - atomic64_inc(&c->btree_writes_nr); - atomic64_add(sectors_to_write, &c->btree_writes_sectors); + atomic64_inc(&c->btree_write_stats[type].nr); + atomic64_add(bytes_to_write, &c->btree_write_stats[type].bytes); INIT_WORK(&wbio->work, btree_write_submit); queue_work(c->io_complete_wq, &wbio->work); return; err: set_btree_node_noevict(b); - if (!b->written && - b->key.k.type == KEY_TYPE_btree_ptr_v2) - bkey_i_to_btree_ptr_v2(&b->key)->v.sectors_written = - cpu_to_le16(sectors_to_write); b->written += sectors_to_write; nowrite: btree_bounce_free(c, bytes, used_mempool, data); @@ -2151,3 +2297,33 @@ bool bch2_btree_flush_all_writes(struct bch_fs *c) { return __bch2_btree_flush_all(c, BTREE_NODE_write_in_flight); } + +static const char * const bch2_btree_write_types[] = { +#define x(t, n) [n] = #t, + BCH_BTREE_WRITE_TYPES() + NULL +}; + +void bch2_btree_write_stats_to_text(struct printbuf *out, struct bch_fs *c) +{ + printbuf_tabstop_push(out, 20); + printbuf_tabstop_push(out, 10); + + prt_tab(out); + prt_str(out, "nr"); + prt_tab(out); + prt_str(out, "size"); + prt_newline(out); + + for (unsigned i = 0; i < BTREE_WRITE_TYPE_NR; i++) { + u64 nr = atomic64_read(&c->btree_write_stats[i].nr); + u64 bytes = atomic64_read(&c->btree_write_stats[i].bytes); + + prt_printf(out, "%s:", bch2_btree_write_types[i]); + prt_tab(out); + prt_u64(out, nr); + prt_tab(out); + prt_human_readable_u64(out, nr ? div64_u64(bytes, nr) : 0); + prt_newline(out); + } +}