Merge tag 'bcachefs-2024-05-30' of https://evilpiepirate.org/git/bcachefs

Pull bcachefs fixes from Kent Overstreet:
 "Assorted odds and ends...

   - two downgrade fixes

   - a couple snapshot deletion and repair fixes, thanks to noradtux for
     finding these and providing the image to debug them

   - a couple assert fixes

   - convert to folio helper, from Matthew

   - some improved error messages

   - bit of code reorganization (just moving things around); doing this
     while things are quiet so I'm not rebasing fixes past reorgs

   - don't return -EROFS on inconsistency error in recovery, this
     confuses util-linux and has it retry the mount

   - fix failure to return error on misaligned dio write; reported as an
     issue with coreutils shred"

* tag 'bcachefs-2024-05-30' of https://evilpiepirate.org/git/bcachefs: (21 commits)
  bcachefs: Fix failure to return error on misaligned dio write
  bcachefs: Don't return -EROFS from mount on inconsistency error
  bcachefs: Fix uninitialized var warning
  bcachefs: Split out sb-errors_format.h
  bcachefs: Split out journal_seq_blacklist_format.h
  bcachefs: Split out replicas_format.h
  bcachefs: Split out disk_groups_format.h
  bcachefs: split out sb-downgrade_format.h
  bcachefs: split out sb-members_format.h
  bcachefs: Better fsck error message for key version
  bcachefs: btree_gc can now handle unknown btrees
  bcachefs: add missing MODULE_DESCRIPTION()
  bcachefs: Fix setting of downgrade recovery passes/errors
  bcachefs: Run check_key_has_snapshot in snapshot_delete_keys()
  bcachefs: Refactor delete_dead_snapshots()
  bcachefs: Fix locking assert
  bcachefs: Fix lookup_first_inode() when inode_generations are present
  bcachefs: Plumb bkey into __btree_err()
  bcachefs: Use copy_folio_from_iter_atomic()
  bcachefs: Fix sb-downgrade validation
  ...

fs/bcachefs/backpointers.c

@@ -690,7 +690,7 @@ static int check_extent_to_backpointers(struct btree_trans *trans,
 
 	ptrs = bch2_bkey_ptrs_c(k);
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
-		struct bpos bucket_pos;
+		struct bpos bucket_pos = POS_MIN;
 		struct bch_backpointer bp;
 
 		if (p.ptr.cached)

fs/bcachefs/bcachefs.h

@@ -457,6 +457,7 @@ enum bch_time_stats {
 };
 
 #include "alloc_types.h"
+#include "btree_gc_types.h"
 #include "btree_types.h"
 #include "btree_node_scan_types.h"
 #include "btree_write_buffer_types.h"
@@ -488,49 +489,6 @@ enum bch_time_stats {
 
 struct btree;
 
-enum gc_phase {
-	GC_PHASE_NOT_RUNNING,
-	GC_PHASE_START,
-	GC_PHASE_SB,
-
-	GC_PHASE_BTREE_stripes,
-	GC_PHASE_BTREE_extents,
-	GC_PHASE_BTREE_inodes,
-	GC_PHASE_BTREE_dirents,
-	GC_PHASE_BTREE_xattrs,
-	GC_PHASE_BTREE_alloc,
-	GC_PHASE_BTREE_quotas,
-	GC_PHASE_BTREE_reflink,
-	GC_PHASE_BTREE_subvolumes,
-	GC_PHASE_BTREE_snapshots,
-	GC_PHASE_BTREE_lru,
-	GC_PHASE_BTREE_freespace,
-	GC_PHASE_BTREE_need_discard,
-	GC_PHASE_BTREE_backpointers,
-	GC_PHASE_BTREE_bucket_gens,
-	GC_PHASE_BTREE_snapshot_trees,
-	GC_PHASE_BTREE_deleted_inodes,
-	GC_PHASE_BTREE_logged_ops,
-	GC_PHASE_BTREE_rebalance_work,
-	GC_PHASE_BTREE_subvolume_children,
-
-	GC_PHASE_PENDING_DELETE,
-};
-
-struct gc_pos {
-	enum gc_phase		phase;
-	u16			level;
-	struct bpos		pos;
-};
-
-struct reflink_gc {
-	u64		offset;
-	u32		size;
-	u32		refcount;
-};
-
-typedef GENRADIX(struct reflink_gc) reflink_gc_table;
-
 struct io_count {
 	u64			sectors[2][BCH_DATA_NR];
 };

fs/bcachefs/bcachefs_format.h

@@ -503,16 +503,22 @@ struct bch_sb_field {
 
 #include "alloc_background_format.h"
 #include "extents_format.h"
-#include "reflink_format.h"
 #include "ec_format.h"
-#include "inode_format.h"
 #include "dirent_format.h"
-#include "xattr_format.h"
-#include "quota_format.h"
+#include "disk_groups_format.h"
+#include "inode_format.h"
+#include "journal_seq_blacklist_format.h"
 #include "logged_ops_format.h"
+#include "quota_format.h"
+#include "reflink_format.h"
+#include "replicas_format.h"
 #include "snapshot_format.h"
 #include "subvolume_format.h"
 #include "sb-counters_format.h"
+#include "sb-downgrade_format.h"
+#include "sb-errors_format.h"
+#include "sb-members_format.h"
+#include "xattr_format.h"
 
 enum bch_sb_field_type {
 #define x(f, nr)	BCH_SB_FIELD_##f = nr,
@@ -545,107 +551,6 @@ struct bch_sb_field_journal_v2 {
 	}			d[];
 };
 
-/* BCH_SB_FIELD_members_v1: */
-
-#define BCH_MIN_NR_NBUCKETS	(1 << 6)
-
-#define BCH_IOPS_MEASUREMENTS()			\
-	x(seqread,	0)			\
-	x(seqwrite,	1)			\
-	x(randread,	2)			\
-	x(randwrite,	3)
-
-enum bch_iops_measurement {
-#define x(t, n) BCH_IOPS_##t = n,
-	BCH_IOPS_MEASUREMENTS()
-#undef x
-	BCH_IOPS_NR
-};
-
-#define BCH_MEMBER_ERROR_TYPES()		\
-	x(read,		0)			\
-	x(write,	1)			\
-	x(checksum,	2)
-
-enum bch_member_error_type {
-#define x(t, n) BCH_MEMBER_ERROR_##t = n,
-	BCH_MEMBER_ERROR_TYPES()
-#undef x
-	BCH_MEMBER_ERROR_NR
-};
-
-struct bch_member {
-	__uuid_t		uuid;
-	__le64			nbuckets;	/* device size */
-	__le16			first_bucket;   /* index of first bucket used */
-	__le16			bucket_size;	/* sectors */
-	__u8			btree_bitmap_shift;
-	__u8			pad[3];
-	__le64			last_mount;	/* time_t */
-
-	__le64			flags;
-	__le32			iops[4];
-	__le64			errors[BCH_MEMBER_ERROR_NR];
-	__le64			errors_at_reset[BCH_MEMBER_ERROR_NR];
-	__le64			errors_reset_time;
-	__le64			seq;
-	__le64			btree_allocated_bitmap;
-	/*
-	 * On recovery from a clean shutdown we don't normally read the journal,
-	 * but we still want to resume writing from where we left off so we
-	 * don't overwrite more than is necessary, for list journal debugging:
-	 */
-	__le32			last_journal_bucket;
-	__le32			last_journal_bucket_offset;
-};
-
-/*
- * 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)
-/* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
-LE64_BITMASK(BCH_MEMBER_DISCARD,	struct bch_member, flags, 14, 15)
-LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED,	struct bch_member, flags, 15, 20)
-LE64_BITMASK(BCH_MEMBER_GROUP,		struct bch_member, flags, 20, 28)
-LE64_BITMASK(BCH_MEMBER_DURABILITY,	struct bch_member, flags, 28, 30)
-LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
-					struct bch_member, flags, 30, 31)
-
-#if 0
-LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS,	struct bch_member, flags[1], 0,  20);
-LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
-#endif
-
-#define BCH_MEMBER_STATES()			\
-	x(rw,		0)			\
-	x(ro,		1)			\
-	x(failed,	2)			\
-	x(spare,	3)
-
-enum bch_member_state {
-#define x(t, n) BCH_MEMBER_STATE_##t = n,
-	BCH_MEMBER_STATES()
-#undef x
-	BCH_MEMBER_STATE_NR
-};
-
-struct bch_sb_field_members_v1 {
-	struct bch_sb_field	field;
-	struct bch_member	_members[]; //Members are now variable size
-};
-
-struct bch_sb_field_members_v2 {
-	struct bch_sb_field	field;
-	__le16			member_bytes; //size of single member entry
-	u8			pad[6];
-	struct bch_member	_members[];
-};
-
 /* BCH_SB_FIELD_crypt: */
 
 struct nonce {
@@ -694,8 +599,6 @@ LE64_BITMASK(BCH_KDF_SCRYPT_N,	struct bch_sb_field_crypt, kdf_flags,  0, 16);
 LE64_BITMASK(BCH_KDF_SCRYPT_R,	struct bch_sb_field_crypt, kdf_flags, 16, 32);
 LE64_BITMASK(BCH_KDF_SCRYPT_P,	struct bch_sb_field_crypt, kdf_flags, 32, 48);
 
-/* BCH_SB_FIELD_replicas: */
-
 #define BCH_DATA_TYPES()		\
 	x(free,		0)		\
 	x(sb,		1)		\
@@ -738,50 +641,6 @@ static inline bool data_type_is_hidden(enum bch_data_type type)
 	}
 }
 
-struct bch_replicas_entry_v0 {
-	__u8			data_type;
-	__u8			nr_devs;
-	__u8			devs[];
-} __packed;
-
-struct bch_sb_field_replicas_v0 {
-	struct bch_sb_field	field;
-	struct bch_replicas_entry_v0 entries[];
-} __packed __aligned(8);
-
-struct bch_replicas_entry_v1 {
-	__u8			data_type;
-	__u8			nr_devs;
-	__u8			nr_required;
-	__u8			devs[];
-} __packed;
-
-#define replicas_entry_bytes(_i)					\
-	(offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
-
-struct bch_sb_field_replicas {
-	struct bch_sb_field	field;
-	struct bch_replicas_entry_v1 entries[];
-} __packed __aligned(8);
-
-/* BCH_SB_FIELD_disk_groups: */
-
-#define BCH_SB_LABEL_SIZE		32
-
-struct bch_disk_group {
-	__u8			label[BCH_SB_LABEL_SIZE];
-	__le64			flags[2];
-} __packed __aligned(8);
-
-LE64_BITMASK(BCH_GROUP_DELETED,		struct bch_disk_group, flags[0], 0,  1)
-LE64_BITMASK(BCH_GROUP_DATA_ALLOWED,	struct bch_disk_group, flags[0], 1,  6)
-LE64_BITMASK(BCH_GROUP_PARENT,		struct bch_disk_group, flags[0], 6, 24)
-
-struct bch_sb_field_disk_groups {
-	struct bch_sb_field	field;
-	struct bch_disk_group	entries[];
-} __packed __aligned(8);
-
 /*
  * On clean shutdown, store btree roots and current journal sequence number in
  * the superblock:
@@ -809,27 +668,6 @@ struct bch_sb_field_clean {
 	__u64			_data[];
 };
 
-struct journal_seq_blacklist_entry {
-	__le64			start;
-	__le64			end;
-};
-
-struct bch_sb_field_journal_seq_blacklist {
-	struct bch_sb_field	field;
-	struct journal_seq_blacklist_entry start[];
-};
-
-struct bch_sb_field_errors {
-	struct bch_sb_field	field;
-	struct bch_sb_field_error_entry {
-		__le64		v;
-		__le64		last_error_time;
-	}			entries[];
-};
-
-LE64_BITMASK(BCH_SB_ERROR_ENTRY_ID,	struct bch_sb_field_error_entry, v,  0, 16);
-LE64_BITMASK(BCH_SB_ERROR_ENTRY_NR,	struct bch_sb_field_error_entry, v, 16, 64);
-
 struct bch_sb_field_ext {
 	struct bch_sb_field	field;
 	__le64			recovery_passes_required[2];
@@ -837,18 +675,6 @@ struct bch_sb_field_ext {
 	__le64			btrees_lost_data;
 };
 
-struct bch_sb_field_downgrade_entry {
-	__le16			version;
-	__le64			recovery_passes[2];
-	__le16			nr_errors;
-	__le16			errors[] __counted_by(nr_errors);
-} __packed __aligned(2);
-
-struct bch_sb_field_downgrade {
-	struct bch_sb_field	field;
-	struct bch_sb_field_downgrade_entry entries[];
-};
-
 /* Superblock: */
 
 /*
@@ -909,7 +735,6 @@ unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_re
 #define bcachefs_metadata_version_current	(bcachefs_metadata_version_max - 1)
 
 #define BCH_SB_SECTOR			8
-#define BCH_SB_MEMBERS_MAX		64 /* XXX kill */
 
 #define BCH_SB_LAYOUT_SIZE_BITS_MAX	16 /* 32 MB */
 

fs/bcachefs/btree_gc.c

@@ -585,16 +585,17 @@ static int bch2_gc_mark_key(struct btree_trans *trans, enum btree_id btree_id,
 
 		if (fsck_err_on(k.k->version.lo > atomic64_read(&c->key_version), c,
 				bkey_version_in_future,
-				"key version number higher than recorded: %llu > %llu",
-				k.k->version.lo,
-				atomic64_read(&c->key_version)))
+				"key version number higher than recorded %llu\n  %s",
+				atomic64_read(&c->key_version),
+				(bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
 			atomic64_set(&c->key_version, k.k->version.lo);
 	}
 
 	if (mustfix_fsck_err_on(level && !bch2_dev_btree_bitmap_marked(c, k),
 				c, btree_bitmap_not_marked,
 				"btree ptr not marked in member info btree allocated bitmap\n  %s",
-				(bch2_bkey_val_to_text(&buf, c, k),
+				(printbuf_reset(&buf),
+				 bch2_bkey_val_to_text(&buf, c, k),
 				 buf.buf))) {
 		mutex_lock(&c->sb_lock);
 		bch2_dev_btree_bitmap_mark(c, k);
@@ -673,8 +674,7 @@ static int bch2_gc_btree(struct btree_trans *trans, enum btree_id btree, bool in
 
 static inline int btree_id_gc_phase_cmp(enum btree_id l, enum btree_id r)
 {
-	return  (int) btree_id_to_gc_phase(l) -
-		(int) btree_id_to_gc_phase(r);
+	return cmp_int(gc_btree_order(l), gc_btree_order(r));
 }
 
 static int bch2_gc_btrees(struct bch_fs *c)
@@ -711,7 +711,7 @@ static int bch2_gc_btrees(struct bch_fs *c)
 static int bch2_mark_superblocks(struct bch_fs *c)
 {
 	mutex_lock(&c->sb_lock);
-	gc_pos_set(c, gc_phase(GC_PHASE_SB));
+	gc_pos_set(c, gc_phase(GC_PHASE_sb));
 
 	int ret = bch2_trans_mark_dev_sbs_flags(c, BTREE_TRIGGER_gc);
 	mutex_unlock(&c->sb_lock);
@@ -1209,7 +1209,7 @@ int bch2_check_allocations(struct bch_fs *c)
 	if (ret)
 		goto out;
 
-	gc_pos_set(c, gc_phase(GC_PHASE_START));
+	gc_pos_set(c, gc_phase(GC_PHASE_start));
 
 	ret = bch2_mark_superblocks(c);
 	BUG_ON(ret);
@@ -1231,7 +1231,7 @@ int bch2_check_allocations(struct bch_fs *c)
 
 	percpu_down_write(&c->mark_lock);
 	/* Indicates that gc is no longer in progress: */
-	__gc_pos_set(c, gc_phase(GC_PHASE_NOT_RUNNING));
+	__gc_pos_set(c, gc_phase(GC_PHASE_not_running));
 
 	bch2_gc_free(c);
 	percpu_up_write(&c->mark_lock);

fs/bcachefs/btree_gc.h

@@ -3,6 +3,7 @@
 #define _BCACHEFS_BTREE_GC_H
 
 #include "bkey.h"
+#include "btree_gc_types.h"
 #include "btree_types.h"
 
 int bch2_check_topology(struct bch_fs *);
@@ -32,36 +33,15 @@ int bch2_check_allocations(struct bch_fs *);
 /* Position of (the start of) a gc phase: */
 static inline struct gc_pos gc_phase(enum gc_phase phase)
 {
-	return (struct gc_pos) {
-		.phase	= phase,
-		.level	= 0,
-		.pos	= POS_MIN,
-	};
-}
-
-static inline int gc_pos_cmp(struct gc_pos l, struct gc_pos r)
-{
-	return   cmp_int(l.phase, r.phase) ?:
-		-cmp_int(l.level, r.level) ?:
-		 bpos_cmp(l.pos, r.pos);
-}
-
-static inline enum gc_phase btree_id_to_gc_phase(enum btree_id id)
-{
-	switch (id) {
-#define x(name, v, ...) case BTREE_ID_##name: return GC_PHASE_BTREE_##name;
-	BCH_BTREE_IDS()
-#undef x
-	default:
-		BUG();
-	}
+	return (struct gc_pos) { .phase	= phase, };
 }
 
 static inline struct gc_pos gc_pos_btree(enum btree_id btree, unsigned level,
 					 struct bpos pos)
 {
 	return (struct gc_pos) {
-		.phase	= btree_id_to_gc_phase(btree),
+		.phase	= GC_PHASE_btree,
+		.btree	= btree,
 		.level	= level,
 		.pos	= pos,
 	};
@@ -76,6 +56,22 @@ static inline struct gc_pos gc_pos_btree_node(struct btree *b)
 	return gc_pos_btree(b->c.btree_id, b->c.level, b->key.k.p);
 }
 
+static inline int gc_btree_order(enum btree_id btree)
+{
+	if (btree == BTREE_ID_stripes)
+		return -1;
+	return btree;
+}
+
+static inline int gc_pos_cmp(struct gc_pos l, struct gc_pos r)
+{
+	return   cmp_int(l.phase, r.phase) ?:
+		 cmp_int(gc_btree_order(l.btree),
+			 gc_btree_order(r.btree)) ?:
+		-cmp_int(l.level, r.level) ?:
+		 bpos_cmp(l.pos, r.pos);
+}
+
 static inline bool gc_visited(struct bch_fs *c, struct gc_pos pos)
 {
 	unsigned seq;

fs/bcachefs/btree_gc_types.h

+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_BTREE_GC_TYPES_H
+#define _BCACHEFS_BTREE_GC_TYPES_H
+
+#include <linux/generic-radix-tree.h>
+
+enum gc_phase {
+	GC_PHASE_not_running,
+	GC_PHASE_start,
+	GC_PHASE_sb,
+	GC_PHASE_btree,
+};
+
+struct gc_pos {
+	enum gc_phase		phase:8;
+	enum btree_id		btree:8;
+	u16			level;
+	struct bpos		pos;
+};
+
+struct reflink_gc {
+	u64		offset;
+	u32		size;
+	u32		refcount;
+};
+
+typedef GENRADIX(struct reflink_gc) reflink_gc_table;
+
+#endif /* _BCACHEFS_BTREE_GC_TYPES_H */

fs/bcachefs/btree_key_cache.c

@@ -424,16 +424,16 @@ static int btree_key_cache_fill(struct btree_trans *trans,
 				goto err;
 			}
 
-			if (!bch2_btree_node_relock(trans, ck_path, 0)) {
+			ret = bch2_trans_relock(trans);
+			if (ret) {
 				kfree(new_k);
-				trace_and_count(trans->c, trans_restart_relock_key_cache_fill, trans, _THIS_IP_, ck_path);
-				ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_fill);
 				goto err;
 			}
 
-			ret = bch2_trans_relock(trans);
-			if (ret) {
+			if (!bch2_btree_node_relock(trans, ck_path, 0)) {
 				kfree(new_k);
+				trace_and_count(trans->c, trans_restart_relock_key_cache_fill, trans, _THIS_IP_, ck_path);
+				ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_fill);
 				goto err;
 			}
 		}

fs/bcachefs/buckets.c

@@ -1134,7 +1134,7 @@ static int __trigger_extent(struct btree_trans *trans,
 	r.e.nr_required	= 1;
 
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
-		s64 disk_sectors;
+		s64 disk_sectors = 0;
 		ret = bch2_trigger_pointer(trans, btree_id, level, k, p, entry, &disk_sectors, flags);
 		if (ret < 0)
 			return ret;

fs/bcachefs/disk_groups_format.h

+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_DISK_GROUPS_FORMAT_H
+#define _BCACHEFS_DISK_GROUPS_FORMAT_H
+
+#define BCH_SB_LABEL_SIZE		32
+
+struct bch_disk_group {
+	__u8			label[BCH_SB_LABEL_SIZE];
+	__le64			flags[2];
+} __packed __aligned(8);
+
+LE64_BITMASK(BCH_GROUP_DELETED,		struct bch_disk_group, flags[0], 0,  1)
+LE64_BITMASK(BCH_GROUP_DATA_ALLOWED,	struct bch_disk_group, flags[0], 1,  6)
+LE64_BITMASK(BCH_GROUP_PARENT,		struct bch_disk_group, flags[0], 6, 24)
+
+struct bch_sb_field_disk_groups {
+	struct bch_sb_field	field;
+	struct bch_disk_group	entries[];
+} __packed __aligned(8);
+
+#endif /* _BCACHEFS_DISK_GROUPS_FORMAT_H */

fs/bcachefs/ec.c

@@ -908,7 +908,7 @@ static int __ec_stripe_mem_alloc(struct bch_fs *c, size_t idx, gfp_t gfp)
 	if (!genradix_ptr_alloc(&c->stripes, idx, gfp))
 		return -BCH_ERR_ENOMEM_ec_stripe_mem_alloc;
 
-	if (c->gc_pos.phase != GC_PHASE_NOT_RUNNING &&
+	if (c->gc_pos.phase != GC_PHASE_not_running &&
 	    !genradix_ptr_alloc(&c->gc_stripes, idx, gfp))
 		return -BCH_ERR_ENOMEM_ec_stripe_mem_alloc;
 

fs/bcachefs/fs-io-buffered.c

@@ -437,8 +437,8 @@ static void bch2_writepage_io_done(struct bch_write_op *op)
 	 */
 
 	/*
-	 * PageWriteback is effectively our ref on the inode - fixup i_blocks
-	 * before calling end_page_writeback:
+	 * The writeback flag is effectively our ref on the inode -
+	 * fixup i_blocks before calling folio_end_writeback:
 	 */
 	bch2_i_sectors_acct(c, io->inode, NULL, io->op.i_sectors_delta);
 
@@ -898,7 +898,7 @@ static int __bch2_buffered_write(struct bch_inode_info *inode,
 	darray_for_each(fs, fi) {
 		f = *fi;
 		f_len = min(end, folio_end_pos(f)) - f_pos;
-		f_copied = copy_page_from_iter_atomic(&f->page, f_offset, f_len, iter);
+		f_copied = copy_folio_from_iter_atomic(f, f_offset, f_len, iter);
 		if (!f_copied) {
 			folios_trunc(&fs, fi);
 			break;

fs/bcachefs/fs-io-direct.c

@@ -609,8 +609,10 @@ ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter)
 	if (unlikely(ret))
 		goto err_put_write_ref;
 
-	if (unlikely((req->ki_pos|iter->count) & (block_bytes(c) - 1)))
+	if (unlikely((req->ki_pos|iter->count) & (block_bytes(c) - 1))) {
+		ret = -EINVAL;
 		goto err_put_write_ref;
+	}
 
 	inode_dio_begin(&inode->v);
 	bch2_pagecache_block_get(inode);

fs/bcachefs/fs.c

@@ -1939,8 +1939,7 @@ static struct dentry *bch2_mount(struct file_system_type *fs_type,
 
 	if (IS_ERR(sb)) {
 		ret = PTR_ERR(sb);
-		ret = bch2_err_class(ret);
-		return ERR_PTR(ret);
+		goto err;
 	}
 
 	c = sb->s_fs_info;
@@ -2016,6 +2015,15 @@ static struct dentry *bch2_mount(struct file_system_type *fs_type,
 err_put_super:
 	__bch2_fs_stop(c);
 	deactivate_locked_super(sb);
+err:
+	/*
+	 * On an inconsistency error in recovery we might see an -EROFS derived
+	 * errorcode (from the journal), but we don't want to return that to
+	 * userspace as that causes util-linux to retry the mount RO - which is
+	 * confusing:
+	 */
+	if (bch2_err_matches(ret, EROFS) && ret != -EROFS)
+		ret = -EIO;
 	return ERR_PTR(bch2_err_class(ret));
 }
 

fs/bcachefs/fsck.c

@@ -77,21 +77,17 @@ static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
 	struct bkey_s_c k;
 	int ret;
 
-	bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
-			     POS(0, inode_nr),
-			     BTREE_ITER_all_snapshots);
-	k = bch2_btree_iter_peek(&iter);
-	ret = bkey_err(k);
-	if (ret)
-		goto err;
-
-	if (!k.k || !bkey_eq(k.k->p, POS(0, inode_nr))) {
-		ret = -BCH_ERR_ENOENT_inode;
-		goto err;
+	for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inode_nr),
+				     BTREE_ITER_all_snapshots, k, ret) {
+		if (k.k->p.offset != inode_nr)
+			break;
+		if (!bkey_is_inode(k.k))
+			continue;
+		ret = bch2_inode_unpack(k, inode);
+		goto found;
 	}
-
-	ret = bch2_inode_unpack(k, inode);
-err:
+	ret = -BCH_ERR_ENOENT_inode;
+found:
 	bch_err_msg(trans->c, ret, "fetching inode %llu", inode_nr);
 	bch2_trans_iter_exit(trans, &iter);
 	return ret;
@@ -770,25 +766,6 @@ static int get_visible_inodes(struct btree_trans *trans,
 	return ret;
 }
 
-static int check_key_has_snapshot(struct btree_trans *trans,
-				  struct btree_iter *iter,
-				  struct bkey_s_c k)
-{
-	struct bch_fs *c = trans->c;
-	struct printbuf buf = PRINTBUF;
-	int ret = 0;
-
-	if (mustfix_fsck_err_on(!bch2_snapshot_equiv(c, k.k->p.snapshot), c,
-				bkey_in_missing_snapshot,
-				"key in missing snapshot: %s",
-				(bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
-		ret = bch2_btree_delete_at(trans, iter,
-					    BTREE_UPDATE_internal_snapshot_node) ?: 1;
-fsck_err:
-	printbuf_exit(&buf);
-	return ret;
-}
-
 static int hash_redo_key(struct btree_trans *trans,
 			 const struct bch_hash_desc desc,
 			 struct bch_hash_info *hash_info,
@@ -983,7 +960,7 @@ static int check_inode(struct btree_trans *trans,
 	bool do_update = false;
 	int ret;
 
-	ret = check_key_has_snapshot(trans, iter, k);
+	ret = bch2_check_key_has_snapshot(trans, iter, k);
 	if (ret < 0)
 		goto err;
 	if (ret)
@@ -1487,7 +1464,7 @@ static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
 	struct printbuf buf = PRINTBUF;
 	int ret = 0;
 
-	ret = check_key_has_snapshot(trans, iter, k);
+	ret = bch2_check_key_has_snapshot(trans, iter, k);
 	if (ret) {
 		ret = ret < 0 ? ret : 0;
 		goto out;
@@ -2010,7 +1987,7 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
 	struct printbuf buf = PRINTBUF;
 	int ret = 0;
 
-	ret = check_key_has_snapshot(trans, iter, k);
+	ret = bch2_check_key_has_snapshot(trans, iter, k);
 	if (ret) {
 		ret = ret < 0 ? ret : 0;
 		goto out;
@@ -2165,7 +2142,7 @@ static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
 	struct inode_walker_entry *i;
 	int ret;
 
-	ret = check_key_has_snapshot(trans, iter, k);
+	ret = bch2_check_key_has_snapshot(trans, iter, k);
 	if (ret < 0)
 		return ret;
 	if (ret)

fs/bcachefs/journal_seq_blacklist_format.h

+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_JOURNAL_SEQ_BLACKLIST_FORMAT_H
+#define _BCACHEFS_JOURNAL_SEQ_BLACKLIST_FORMAT_H
+
+struct journal_seq_blacklist_entry {
+	__le64			start;
+	__le64			end;
+};
+
+struct bch_sb_field_journal_seq_blacklist {
+	struct bch_sb_field	field;
+	struct journal_seq_blacklist_entry start[];
+};
+
+#endif /* _BCACHEFS_JOURNAL_SEQ_BLACKLIST_FORMAT_H */

fs/bcachefs/mean_and_variance_test.c

@@ -217,4 +217,5 @@ static struct kunit_suite mean_and_variance_test_suite = {
 kunit_test_suite(mean_and_variance_test_suite);
 
 MODULE_AUTHOR("Daniel B. Hill");
+MODULE_DESCRIPTION("bcachefs filesystem mean and variance unit tests");
 MODULE_LICENSE("GPL");

fs/bcachefs/replicas_format.h

+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_REPLICAS_FORMAT_H
+#define _BCACHEFS_REPLICAS_FORMAT_H
+
+struct bch_replicas_entry_v0 {
+	__u8			data_type;
+	__u8			nr_devs;
+	__u8			devs[];
+} __packed;
+
+struct bch_sb_field_replicas_v0 {
+	struct bch_sb_field	field;
+	struct bch_replicas_entry_v0 entries[];
+} __packed __aligned(8);
+
+struct bch_replicas_entry_v1 {
+	__u8			data_type;
+	__u8			nr_devs;
+	__u8			nr_required;
+	__u8			devs[];
+} __packed;
+
+struct bch_sb_field_replicas {
+	struct bch_sb_field	field;
+	struct bch_replicas_entry_v1 entries[];
+} __packed __aligned(8);
+
+#define replicas_entry_bytes(_i)					\
+	(offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
+
+#endif /* _BCACHEFS_REPLICAS_FORMAT_H */

fs/bcachefs/sb-downgrade.c

@@ -146,10 +146,17 @@ static int bch2_sb_downgrade_validate(struct bch_sb *sb, struct bch_sb_field *f,
 	for (const struct bch_sb_field_downgrade_entry *i = e->entries;
 	     (void *) i	< vstruct_end(&e->field);
 	     i = downgrade_entry_next_c(i)) {
+		/*
+		 * Careful: sb_field_downgrade_entry is only 2 byte aligned, but
+		 * section sizes are 8 byte aligned - an empty entry spanning
+		 * the end of the section is allowed (and ignored):
+		 */
+		if ((void *) &i->errors[0] > vstruct_end(&e->field))
+			break;
+
 		if (flags & BCH_VALIDATE_write &&
-		    ((void *) &i->errors[0] > vstruct_end(&e->field) ||
-		     (void *) downgrade_entry_next_c(i) > vstruct_end(&e->field))) {
-			prt_printf(err, "downgrade entry overruns end of superblock section)");
+		    (void *) downgrade_entry_next_c(i) > vstruct_end(&e->field)) {
+			prt_printf(err, "downgrade entry overruns end of superblock section");
 			return -BCH_ERR_invalid_sb_downgrade;
 		}
 

fs/bcachefs/sb-downgrade_format.h

+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_SB_DOWNGRADE_FORMAT_H
+#define _BCACHEFS_SB_DOWNGRADE_FORMAT_H
+
+struct bch_sb_field_downgrade_entry {
+	__le16			version;
+	__le64			recovery_passes[2];
+	__le16			nr_errors;
+	__le16			errors[] __counted_by(nr_errors);
+} __packed __aligned(2);
+
+struct bch_sb_field_downgrade {
+	struct bch_sb_field	field;
+	struct bch_sb_field_downgrade_entry entries[];
+};
+
+#endif /* _BCACHEFS_SB_DOWNGRADE_FORMAT_H */

fs/bcachefs/sb-members_format.h

+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_SB_MEMBERS_FORMAT_H
+#define _BCACHEFS_SB_MEMBERS_FORMAT_H
+
+/*
+ * We refer to members with bitmasks in various places - but we need to get rid
+ * of this limit:
+ */
+#define BCH_SB_MEMBERS_MAX		64
+
+#define BCH_MIN_NR_NBUCKETS	(1 << 6)
+
+#define BCH_IOPS_MEASUREMENTS()			\
+	x(seqread,	0)			\
+	x(seqwrite,	1)			\
+	x(randread,	2)			\
+	x(randwrite,	3)
+
+enum bch_iops_measurement {
+#define x(t, n) BCH_IOPS_##t = n,
+	BCH_IOPS_MEASUREMENTS()
+#undef x
+	BCH_IOPS_NR
+};
+
+#define BCH_MEMBER_ERROR_TYPES()		\
+	x(read,		0)			\
+	x(write,	1)			\
+	x(checksum,	2)
+
+enum bch_member_error_type {
+#define x(t, n) BCH_MEMBER_ERROR_##t = n,
+	BCH_MEMBER_ERROR_TYPES()
+#undef x
+	BCH_MEMBER_ERROR_NR
+};
+
+struct bch_member {
+	__uuid_t		uuid;
+	__le64			nbuckets;	/* device size */
+	__le16			first_bucket;   /* index of first bucket used */
+	__le16			bucket_size;	/* sectors */
+	__u8			btree_bitmap_shift;
+	__u8			pad[3];
+	__le64			last_mount;	/* time_t */
+
+	__le64			flags;
+	__le32			iops[4];
+	__le64			errors[BCH_MEMBER_ERROR_NR];
+	__le64			errors_at_reset[BCH_MEMBER_ERROR_NR];
+	__le64			errors_reset_time;
+	__le64			seq;
+	__le64			btree_allocated_bitmap;
+	/*
+	 * On recovery from a clean shutdown we don't normally read the journal,
+	 * but we still want to resume writing from where we left off so we
+	 * don't overwrite more than is necessary, for list journal debugging:
+	 */
+	__le32			last_journal_bucket;
+	__le32			last_journal_bucket_offset;
+};
+
+/*
+ * 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)
+/* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
+LE64_BITMASK(BCH_MEMBER_DISCARD,	struct bch_member, flags, 14, 15)
+LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED,	struct bch_member, flags, 15, 20)
+LE64_BITMASK(BCH_MEMBER_GROUP,		struct bch_member, flags, 20, 28)
+LE64_BITMASK(BCH_MEMBER_DURABILITY,	struct bch_member, flags, 28, 30)
+LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
+					struct bch_member, flags, 30, 31)
+
+#if 0
+LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS,	struct bch_member, flags[1], 0,  20);
+LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
+#endif
+
+#define BCH_MEMBER_STATES()			\
+	x(rw,		0)			\
+	x(ro,		1)			\
+	x(failed,	2)			\
+	x(spare,	3)
+
+enum bch_member_state {
+#define x(t, n) BCH_MEMBER_STATE_##t = n,
+	BCH_MEMBER_STATES()
+#undef x
+	BCH_MEMBER_STATE_NR
+};
+
+struct bch_sb_field_members_v1 {
+	struct bch_sb_field	field;
+	struct bch_member	_members[]; //Members are now variable size
+};
+
+struct bch_sb_field_members_v2 {
+	struct bch_sb_field	field;
+	__le16			member_bytes; //size of single member entry
+	u8			pad[6];
+	struct bch_member	_members[];
+};
+
+#endif /* _BCACHEFS_SB_MEMBERS_FORMAT_H */

fs/bcachefs/snapshot.c

@@ -1042,6 +1042,25 @@ int bch2_reconstruct_snapshots(struct bch_fs *c)
 	return ret;
 }
 
+int bch2_check_key_has_snapshot(struct btree_trans *trans,
+				struct btree_iter *iter,
+				struct bkey_s_c k)
+{
+	struct bch_fs *c = trans->c;
+	struct printbuf buf = PRINTBUF;
+	int ret = 0;
+
+	if (fsck_err_on(!bch2_snapshot_equiv(c, k.k->p.snapshot), c,
+			bkey_in_missing_snapshot,
+			"key in missing snapshot %s, delete?",
+			(bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
+		ret = bch2_btree_delete_at(trans, iter,
+					    BTREE_UPDATE_internal_snapshot_node) ?: 1;
+fsck_err:
+	printbuf_exit(&buf);
+	return ret;
+}
+
 /*
  * Mark a snapshot as deleted, for future cleanup:
  */
@@ -1351,35 +1370,39 @@ int bch2_snapshot_node_create(struct btree_trans *trans, u32 parent,
  * that key to snapshot leaf nodes, where we can mutate it
  */
 
-static int snapshot_delete_key(struct btree_trans *trans,
+static int delete_dead_snapshots_process_key(struct btree_trans *trans,
 			       struct btree_iter *iter,
 			       struct bkey_s_c k,
 			       snapshot_id_list *deleted,
 			       snapshot_id_list *equiv_seen,
 			       struct bpos *last_pos)
 {
+	int ret = bch2_check_key_has_snapshot(trans, iter, k);
+	if (ret)
+		return ret < 0 ? ret : 0;
+
 	struct bch_fs *c = trans->c;
 	u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
+	if (!equiv) /* key for invalid snapshot node, but we chose not to delete */
+		return 0;
 
 	if (!bkey_eq(k.k->p, *last_pos))
 		equiv_seen->nr = 0;
-	*last_pos = k.k->p;
 
-	if (snapshot_list_has_id(deleted, k.k->p.snapshot) ||
-	    snapshot_list_has_id(equiv_seen, equiv)) {
+	if (snapshot_list_has_id(deleted, k.k->p.snapshot))
 		return bch2_btree_delete_at(trans, iter,
 					    BTREE_UPDATE_internal_snapshot_node);
-	} else {
-		return snapshot_list_add(c, equiv_seen, equiv);
-	}
-}
 
-static int move_key_to_correct_snapshot(struct btree_trans *trans,
-			       struct btree_iter *iter,
-			       struct bkey_s_c k)
-{
-	struct bch_fs *c = trans->c;
-	u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
+	if (!bpos_eq(*last_pos, k.k->p) &&
+	    snapshot_list_has_id(equiv_seen, equiv))
+		return bch2_btree_delete_at(trans, iter,
+					    BTREE_UPDATE_internal_snapshot_node);
+
+	*last_pos = k.k->p;
+
+	ret = snapshot_list_add_nodup(c, equiv_seen, equiv);
+	if (ret)
+		return ret;
 
 	/*
 	 * When we have a linear chain of snapshot nodes, we consider
@@ -1389,21 +1412,20 @@ static int move_key_to_correct_snapshot(struct btree_trans *trans,
 	 *
 	 * If there are multiple keys in different snapshots at the same
 	 * position, we're only going to keep the one in the newest
-	 * snapshot - the rest have been overwritten and are redundant,
-	 * and for the key we're going to keep we need to move it to the
-	 * equivalance class ID if it's not there already.
+	 * snapshot (we delete the others above) - the rest have been
+	 * overwritten and are redundant, and for the key we're going to keep we
+	 * need to move it to the equivalance class ID if it's not there
+	 * already.
 	 */
 	if (equiv != k.k->p.snapshot) {
 		struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, k);
-		struct btree_iter new_iter;
-		int ret;
-
-		ret = PTR_ERR_OR_ZERO(new);
+		int ret = PTR_ERR_OR_ZERO(new);
 		if (ret)
 			return ret;
 
 		new->k.p.snapshot = equiv;
 
+		struct btree_iter new_iter;
 		bch2_trans_iter_init(trans, &new_iter, iter->btree_id, new->k.p,
 				     BTREE_ITER_all_snapshots|
 				     BTREE_ITER_cached|
@@ -1538,7 +1560,6 @@ int bch2_delete_dead_snapshots(struct bch_fs *c)
 	struct btree_trans *trans;
 	snapshot_id_list deleted = { 0 };
 	snapshot_id_list deleted_interior = { 0 };
-	u32 id;
 	int ret = 0;
 
 	if (!test_and_clear_bit(BCH_FS_need_delete_dead_snapshots, &c->flags))
@@ -1585,33 +1606,20 @@ int bch2_delete_dead_snapshots(struct bch_fs *c)
 	if (ret)
 		goto err;
 
-	for (id = 0; id < BTREE_ID_NR; id++) {
+	for (unsigned btree = 0; btree < BTREE_ID_NR; btree++) {
 		struct bpos last_pos = POS_MIN;
 		snapshot_id_list equiv_seen = { 0 };
 		struct disk_reservation res = { 0 };
 
-		if (!btree_type_has_snapshots(id))
-			continue;
-
-		/*
-		 * deleted inodes btree is maintained by a trigger on the inodes
-		 * btree - no work for us to do here, and it's not safe to scan
-		 * it because we'll see out of date keys due to the btree write
-		 * buffer:
-		 */
-		if (id == BTREE_ID_deleted_inodes)
+		if (!btree_type_has_snapshots(btree))
 			continue;
 
 		ret = for_each_btree_key_commit(trans, iter,
-				id, POS_MIN,
-				BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
-				&res, NULL, BCH_TRANS_COMMIT_no_enospc,
-			snapshot_delete_key(trans, &iter, k, &deleted, &equiv_seen, &last_pos)) ?:
-		      for_each_btree_key_commit(trans, iter,
-				id, POS_MIN,
+				btree, POS_MIN,
 				BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
 				&res, NULL, BCH_TRANS_COMMIT_no_enospc,
-			move_key_to_correct_snapshot(trans, &iter, k));
+			delete_dead_snapshots_process_key(trans, &iter, k, &deleted,
+							  &equiv_seen, &last_pos));
 
 		bch2_disk_reservation_put(c, &res);
 		darray_exit(&equiv_seen);

fs/bcachefs/snapshot.h

@@ -242,6 +242,7 @@ int bch2_snapshot_node_create(struct btree_trans *, u32,
 int bch2_check_snapshot_trees(struct bch_fs *);
 int bch2_check_snapshots(struct bch_fs *);
 int bch2_reconstruct_snapshots(struct bch_fs *);
+int bch2_check_key_has_snapshot(struct btree_trans *, struct btree_iter *, struct bkey_s_c);
 
 int bch2_snapshot_node_set_deleted(struct btree_trans *, u32);
 void bch2_delete_dead_snapshots_work(struct work_struct *);

fs/bcachefs/super-io.c

@@ -1132,18 +1132,12 @@ bool bch2_check_version_downgrade(struct bch_fs *c)
 	 * c->sb will be checked before we write the superblock, so update it as
 	 * well:
 	 */
-	if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) > bcachefs_metadata_version_current) {
+	if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) > bcachefs_metadata_version_current)
 		SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
-		c->sb.version_upgrade_complete = bcachefs_metadata_version_current;
-	}
-	if (c->sb.version > bcachefs_metadata_version_current) {
+	if (c->sb.version > bcachefs_metadata_version_current)
 		c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
-		c->sb.version = bcachefs_metadata_version_current;
-	}
-	if (c->sb.version_min > bcachefs_metadata_version_current) {
+	if (c->sb.version_min > bcachefs_metadata_version_current)
 		c->disk_sb.sb->version_min = cpu_to_le16(bcachefs_metadata_version_current);
-		c->sb.version_min = bcachefs_metadata_version_current;
-	}
 	c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
 	return ret;
 }

fs/bcachefs/super.c

@@ -564,7 +564,7 @@ static void __bch2_fs_free(struct bch_fs *c)
 	BUG_ON(atomic_read(&c->journal_keys.ref));
 	bch2_fs_btree_write_buffer_exit(c);
 	percpu_free_rwsem(&c->mark_lock);
-	EBUG_ON(percpu_u64_get(c->online_reserved));
+	EBUG_ON(c->online_reserved && percpu_u64_get(c->online_reserved));
 	free_percpu(c->online_reserved);
 
 	darray_exit(&c->btree_roots_extra);