Merge tag 'bcachefs-2024-05-07.2' of https://evilpiepirate.org/git/bc…

…achefs

Pull bcachefs fixes from Kent Overstreet:

 - Various syzbot fixes; mainly small gaps in validation

 - Fix an integer overflow in fiemap() which was preventing filefrag
   from returning the full list of extents

 - Fix a refcounting bug on the device refcount, turned up by new
   assertions in the development branch

 - Fix a device removal/readd bug; write_super() was repeatedly dropping
   and retaking bch_dev->io_ref references

* tag 'bcachefs-2024-05-07.2' of https://evilpiepirate.org/git/bcachefs:
  bcachefs: Add missing sched_annotate_sleep() in bch2_journal_flush_seq_async()
  bcachefs: Fix race in bch2_write_super()
  bcachefs: BCH_SB_LAYOUT_SIZE_BITS_MAX
  bcachefs: Add missing skcipher_request_set_callback() call
  bcachefs: Fix snapshot_t() usage in bch2_fs_quota_read_inode()
  bcachefs: Fix shift-by-64 in bformat_needs_redo()
  bcachefs: Guard against unknown k.k->type in __bkey_invalid()
  bcachefs: Add missing validation for superblock section clean
  bcachefs: Fix assert in bch2_alloc_v4_invalid()
  bcachefs: fix overflow in fiemap
  bcachefs: Add a better limit for maximum number of buckets
  bcachefs: Fix lifetime issue in device iterator helpers
  bcachefs: Fix bch2_dev_lookup() refcounting
  bcachefs: Initialize bch_write_op->failed in inline data path
  bcachefs: Fix refcount put in sb_field_resize error path
  bcachefs: Inodes need extra padding for varint_decode_fast()
  bcachefs: Fix early error path in bch2_fs_btree_key_cache_exit()
  bcachefs: bucket_pos_to_bp_noerror()
  bcachefs: don't free error pointers
  bcachefs: Fix a scheduler splat in __bch2_next_write_buffer_flush_journal_buf()

fs/bcachefs/alloc_background.c

@@ -244,10 +244,10 @@ int bch2_alloc_v4_invalid(struct bch_fs *c, struct bkey_s_c k,
 	struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
 	int ret = 0;
 
-	bkey_fsck_err_on(alloc_v4_u64s(a.v) > bkey_val_u64s(k.k), c, err,
+	bkey_fsck_err_on(alloc_v4_u64s_noerror(a.v) > bkey_val_u64s(k.k), c, err,
 			 alloc_v4_val_size_bad,
 			 "bad val size (%u > %zu)",
-			 alloc_v4_u64s(a.v), bkey_val_u64s(k.k));
+			 alloc_v4_u64s_noerror(a.v), bkey_val_u64s(k.k));
 
 	bkey_fsck_err_on(!BCH_ALLOC_V4_BACKPOINTERS_START(a.v) &&
 			 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v), c, err,

fs/bcachefs/alloc_background.h

@@ -126,13 +126,17 @@ static inline struct bpos alloc_freespace_pos(struct bpos pos, struct bch_alloc_
 	return pos;
 }
 
-static inline unsigned alloc_v4_u64s(const struct bch_alloc_v4 *a)
+static inline unsigned alloc_v4_u64s_noerror(const struct bch_alloc_v4 *a)
 {
-	unsigned ret = (BCH_ALLOC_V4_BACKPOINTERS_START(a) ?:
+	return (BCH_ALLOC_V4_BACKPOINTERS_START(a) ?:
 			BCH_ALLOC_V4_U64s_V0) +
 		BCH_ALLOC_V4_NR_BACKPOINTERS(a) *
 		(sizeof(struct bch_backpointer) / sizeof(u64));
+}
 
+static inline unsigned alloc_v4_u64s(const struct bch_alloc_v4 *a)
+{
+	unsigned ret = alloc_v4_u64s_noerror(a);
 	BUG_ON(ret > U8_MAX - BKEY_U64s);
 	return ret;
 }

fs/bcachefs/backpointers.c

@@ -54,7 +54,7 @@ int bch2_backpointer_invalid(struct bch_fs *c, struct bkey_s_c k,
 	int ret = 0;
 
 	bkey_fsck_err_on((bp.v->bucket_offset >> MAX_EXTENT_COMPRESS_RATIO_SHIFT) >= ca->mi.bucket_size ||
-			 !bpos_eq(bp.k->p, bucket_pos_to_bp(c, bucket, bp.v->bucket_offset)),
+			 !bpos_eq(bp.k->p, bucket_pos_to_bp_noerror(ca, bucket, bp.v->bucket_offset)),
 			 c, err,
 			 backpointer_bucket_offset_wrong,
 			 "backpointer bucket_offset wrong");

fs/bcachefs/backpointers.h

@@ -45,6 +45,15 @@ static inline struct bpos bp_pos_to_bucket(const struct bch_fs *c,
 	return POS(bp_pos.inode, sector_to_bucket(ca, bucket_sector));
 }
 
+static inline struct bpos bucket_pos_to_bp_noerror(const struct bch_dev *ca,
+						   struct bpos bucket,
+						   u64 bucket_offset)
+{
+	return POS(bucket.inode,
+		   (bucket_to_sector(ca, bucket.offset) <<
+		    MAX_EXTENT_COMPRESS_RATIO_SHIFT) + bucket_offset);
+}
+
 /*
  * Convert from pos in alloc btree + bucket offset to pos in backpointer btree:
  */
@@ -53,10 +62,7 @@ static inline struct bpos bucket_pos_to_bp(const struct bch_fs *c,
 					   u64 bucket_offset)
 {
 	struct bch_dev *ca = bch_dev_bkey_exists(c, bucket.inode);
-	struct bpos ret = POS(bucket.inode,
-			      (bucket_to_sector(ca, bucket.offset) <<
-			       MAX_EXTENT_COMPRESS_RATIO_SHIFT) + bucket_offset);
-
+	struct bpos ret = bucket_pos_to_bp_noerror(ca, bucket, bucket_offset);
 	EBUG_ON(!bkey_eq(bucket, bp_pos_to_bucket(c, ret)));
 	return ret;
 }

fs/bcachefs/bcachefs_format.h

@@ -591,6 +591,12 @@ struct bch_member {
 	__le64			btree_allocated_bitmap;
 };
 
+/*
+ * This limit comes from the bucket_gens array - it's a single allocation, and
+ * kernel allocation are limited to INT_MAX
+ */
+#define BCH_MEMBER_NBUCKETS_MAX	(INT_MAX - 64)
+
 #define BCH_MEMBER_V1_BYTES	56
 
 LE64_BITMASK(BCH_MEMBER_STATE,		struct bch_member, flags,  0,  4)
@@ -897,6 +903,8 @@ unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_re
 #define BCH_SB_SECTOR			8
 #define BCH_SB_MEMBERS_MAX		64 /* XXX kill */
 
+#define BCH_SB_LAYOUT_SIZE_BITS_MAX	16 /* 32 MB */
+
 struct bch_sb_layout {
 	__uuid_t		magic;	/* bcachefs superblock UUID */
 	__u8			layout_type;

fs/bcachefs/bkey_methods.c

@@ -171,8 +171,8 @@ int __bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 	if (type >= BKEY_TYPE_NR)
 		return 0;
 
-	bkey_fsck_err_on((type == BKEY_TYPE_btree ||
-			  (flags & BKEY_INVALID_COMMIT)) &&
+	bkey_fsck_err_on(k.k->type < KEY_TYPE_MAX &&
+			 (type == BKEY_TYPE_btree || (flags & BKEY_INVALID_COMMIT)) &&
 			 !(bch2_key_types_allowed[type] & BIT_ULL(k.k->type)), c, err,
 			 bkey_invalid_type_for_btree,
 			 "invalid key type for btree %s (%s)",

fs/bcachefs/btree_key_cache.c

@@ -956,13 +956,15 @@ void bch2_fs_btree_key_cache_exit(struct btree_key_cache *bc)
 	}
 
 #ifdef __KERNEL__
-	for_each_possible_cpu(cpu) {
-		struct btree_key_cache_freelist *f =
-			per_cpu_ptr(bc->pcpu_freed, cpu);
-
-		for (i = 0; i < f->nr; i++) {
-			ck = f->objs[i];
-			list_add(&ck->list, &items);
+	if (bc->pcpu_freed) {
+		for_each_possible_cpu(cpu) {
+			struct btree_key_cache_freelist *f =
+				per_cpu_ptr(bc->pcpu_freed, cpu);
+
+			for (i = 0; i < f->nr; i++) {
+				ck = f->objs[i];
+				list_add(&ck->list, &items);
+			}
 		}
 	}
 #endif

fs/bcachefs/checksum.c

@@ -102,6 +102,7 @@ static inline int do_encrypt_sg(struct crypto_sync_skcipher *tfm,
 	int ret;
 
 	skcipher_request_set_sync_tfm(req, tfm);
+	skcipher_request_set_callback(req, 0, NULL, NULL);
 	skcipher_request_set_crypt(req, sg, sg, len, nonce.d);
 
 	ret = crypto_skcipher_encrypt(req);

fs/bcachefs/errcode.h

@@ -175,6 +175,7 @@
 	x(EINVAL,			block_size_too_small)			\
 	x(EINVAL,			bucket_size_too_small)			\
 	x(EINVAL,			device_size_too_small)			\
+	x(EINVAL,			device_size_too_big)			\
 	x(EINVAL,			device_not_a_member_of_filesystem)	\
 	x(EINVAL,			device_has_been_removed)		\
 	x(EINVAL,			device_splitbrain)			\

fs/bcachefs/fs.c

@@ -964,7 +964,6 @@ static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 	struct btree_iter iter;
 	struct bkey_s_c k;
 	struct bkey_buf cur, prev;
-	struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
 	unsigned offset_into_extent, sectors;
 	bool have_extent = false;
 	u32 snapshot;
@@ -974,6 +973,7 @@ static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 	if (ret)
 		return ret;
 
+	struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
 	if (start + len < start)
 		return -EINVAL;
 

fs/bcachefs/io_write.c

@@ -199,9 +199,6 @@ static inline int bch2_extent_update_i_size_sectors(struct btree_trans *trans,
 						    u64 new_i_size,
 						    s64 i_sectors_delta)
 {
-	struct btree_iter iter;
-	struct bkey_i *k;
-	struct bkey_i_inode_v3 *inode;
 	/*
 	 * Crazy performance optimization:
 	 * Every extent update needs to also update the inode: the inode trigger
@@ -214,25 +211,36 @@ static inline int bch2_extent_update_i_size_sectors(struct btree_trans *trans,
 	 * lost, but that's fine.
 	 */
 	unsigned inode_update_flags = BTREE_UPDATE_NOJOURNAL;
-	int ret;
 
-	k = bch2_bkey_get_mut_noupdate(trans, &iter, BTREE_ID_inodes,
+	struct btree_iter iter;
+	struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
 			      SPOS(0,
 				   extent_iter->pos.inode,
 				   extent_iter->snapshot),
 			      BTREE_ITER_CACHED);
-	ret = PTR_ERR_OR_ZERO(k);
+	int ret = bkey_err(k);
 	if (unlikely(ret))
 		return ret;
 
-	if (unlikely(k->k.type != KEY_TYPE_inode_v3)) {
-		k = bch2_inode_to_v3(trans, k);
-		ret = PTR_ERR_OR_ZERO(k);
+	/*
+	 * varint_decode_fast(), in the inode .invalid method, reads up to 7
+	 * bytes past the end of the buffer:
+	 */
+	struct bkey_i *k_mut = bch2_trans_kmalloc_nomemzero(trans, bkey_bytes(k.k) + 8);
+	ret = PTR_ERR_OR_ZERO(k_mut);
+	if (unlikely(ret))
+		goto err;
+
+	bkey_reassemble(k_mut, k);
+
+	if (unlikely(k_mut->k.type != KEY_TYPE_inode_v3)) {
+		k_mut = bch2_inode_to_v3(trans, k_mut);
+		ret = PTR_ERR_OR_ZERO(k_mut);
 		if (unlikely(ret))
 			goto err;
 	}
 
-	inode = bkey_i_to_inode_v3(k);
+	struct bkey_i_inode_v3 *inode = bkey_i_to_inode_v3(k_mut);
 
 	if (!(le64_to_cpu(inode->v.bi_flags) & BCH_INODE_i_size_dirty) &&
 	    new_i_size > le64_to_cpu(inode->v.bi_size)) {
@@ -1505,6 +1513,8 @@ static void bch2_write_data_inline(struct bch_write_op *op, unsigned data_len)
 	unsigned sectors;
 	int ret;
 
+	memset(&op->failed, 0, sizeof(op->failed));
+
 	op->flags |= BCH_WRITE_WROTE_DATA_INLINE;
 	op->flags |= BCH_WRITE_DONE;
 

fs/bcachefs/journal.c

@@ -706,6 +706,12 @@ int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
 
 		spin_unlock(&j->lock);
 
+		/*
+		 * We're called from bch2_journal_flush_seq() -> wait_event();
+		 * but this might block. We won't usually block, so we won't
+		 * livelock:
+		 */
+		sched_annotate_sleep();
 		ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 		if (ret)
 			return ret;
@@ -870,6 +876,8 @@ static struct journal_buf *__bch2_next_write_buffer_flush_journal_buf(struct jou
 {
 	struct journal_buf *ret = NULL;
 
+	/* We're inside wait_event(), but using mutex_lock(: */
+	sched_annotate_sleep();
 	mutex_lock(&j->buf_lock);
 	spin_lock(&j->lock);
 	max_seq = min(max_seq, journal_cur_seq(j));

fs/bcachefs/move.c

@@ -968,24 +968,30 @@ static bool migrate_btree_pred(struct bch_fs *c, void *arg,
 	return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
 }
 
+/*
+ * Ancient versions of bcachefs produced packed formats which could represent
+ * keys that the in memory format cannot represent; this checks for those
+ * formats so we can get rid of them.
+ */
 static bool bformat_needs_redo(struct bkey_format *f)
 {
-	unsigned i;
-
-	for (i = 0; i < f->nr_fields; i++) {
+	for (unsigned i = 0; i < f->nr_fields; i++) {
+		unsigned f_bits = f->bits_per_field[i];
 		unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
 		u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
 		u64 field_offset = le64_to_cpu(f->field_offset[i]);
 
-		if (f->bits_per_field[i] > unpacked_bits)
+		if (f_bits > unpacked_bits)
 			return true;
 
-		if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
+		if ((f_bits == unpacked_bits) && field_offset)
 			return true;
 
-		if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
-		     unpacked_mask) <
-		    field_offset)
+		u64 f_mask = f_bits
+			? ~((~0ULL << (f_bits - 1)) << 1)
+			: 0;
+
+		if (((field_offset + f_mask) & unpacked_mask) < field_offset)
 			return true;
 	}
 

fs/bcachefs/quota.c

@@ -560,13 +560,11 @@ static int bch2_fs_quota_read_inode(struct btree_trans *trans,
 	struct bch_fs *c = trans->c;
 	struct bch_inode_unpacked u;
 	struct bch_snapshot_tree s_t;
-	int ret;
+	u32 tree = bch2_snapshot_tree(c, k.k->p.snapshot);
 
-	ret = bch2_snapshot_tree_lookup(trans,
-			bch2_snapshot_tree(c, k.k->p.snapshot), &s_t);
+	int ret = bch2_snapshot_tree_lookup(trans, tree, &s_t);
 	bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
-			"%s: snapshot tree %u not found", __func__,
-			snapshot_t(c, k.k->p.snapshot)->tree);
+			"%s: snapshot tree %u not found", __func__, tree);
 	if (ret)
 		return ret;
 

fs/bcachefs/recovery.c

@@ -902,7 +902,8 @@ int bch2_fs_recovery(struct bch_fs *c)
 		bch2_journal_keys_put_initial(c);
 		bch2_find_btree_nodes_exit(&c->found_btree_nodes);
 	}
-	kfree(clean);
+	if (!IS_ERR(clean))
+		kfree(clean);
 
 	if (!ret &&
 	    test_bit(BCH_FS_need_delete_dead_snapshots, &c->flags) &&

fs/bcachefs/sb-clean.c

@@ -278,6 +278,17 @@ static int bch2_sb_clean_validate(struct bch_sb *sb,
 		return -BCH_ERR_invalid_sb_clean;
 	}
 
+	for (struct jset_entry *entry = clean->start;
+	     entry != vstruct_end(&clean->field);
+	     entry = vstruct_next(entry)) {
+		if ((void *) vstruct_next(entry) > vstruct_end(&clean->field)) {
+			prt_str(err, "entry type ");
+			bch2_prt_jset_entry_type(err, le16_to_cpu(entry->type));
+			prt_str(err, " overruns end of section");
+			return -BCH_ERR_invalid_sb_clean;
+		}
+	}
+
 	return 0;
 }
 
@@ -295,6 +306,9 @@ static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
 	for (entry = clean->start;
 	     entry != vstruct_end(&clean->field);
 	     entry = vstruct_next(entry)) {
+		if ((void *) vstruct_next(entry) > vstruct_end(&clean->field))
+			break;
+
 		if (entry->type == BCH_JSET_ENTRY_btree_keys &&
 		    !entry->u64s)
 			continue;

fs/bcachefs/sb-members.c

@@ -124,9 +124,9 @@ static int validate_member(struct printbuf *err,
 			   struct bch_sb *sb,
 			   int i)
 {
-	if (le64_to_cpu(m.nbuckets) > LONG_MAX) {
-		prt_printf(err, "device %u: too many buckets (got %llu, max %lu)",
-			   i, le64_to_cpu(m.nbuckets), LONG_MAX);
+	if (le64_to_cpu(m.nbuckets) > BCH_MEMBER_NBUCKETS_MAX) {
+		prt_printf(err, "device %u: too many buckets (got %llu, max %u)",
+			   i, le64_to_cpu(m.nbuckets), BCH_MEMBER_NBUCKETS_MAX);
 		return -BCH_ERR_invalid_sb_members;
 	}
 

fs/bcachefs/sb-members.h

@@ -107,10 +107,10 @@ static inline struct bch_dev *__bch2_next_dev(struct bch_fs *c, struct bch_dev *
 
 static inline struct bch_dev *bch2_get_next_dev(struct bch_fs *c, struct bch_dev *ca)
 {
+	rcu_read_lock();
 	if (ca)
 		percpu_ref_put(&ca->ref);
 
-	rcu_read_lock();
 	if ((ca = __bch2_next_dev(c, ca, NULL)))
 		percpu_ref_get(&ca->ref);
 	rcu_read_unlock();
@@ -132,10 +132,10 @@ static inline struct bch_dev *bch2_get_next_online_dev(struct bch_fs *c,
 						       struct bch_dev *ca,
 						       unsigned state_mask)
 {
+	rcu_read_lock();
 	if (ca)
 		percpu_ref_put(&ca->io_ref);
 
-	rcu_read_lock();
 	while ((ca = __bch2_next_dev(c, ca, NULL)) &&
 	       (!((1 << ca->mi.state) & state_mask) ||
 		!percpu_ref_tryget(&ca->io_ref)))