Merge tag 'bcachefs-2024-09-09' of git://evilpiepirate.org/bcachefs

Pull bcachefs fixes from Kent Overstreet:

 - fix ca->io_ref usage; analagous to previous patch doing that for main
   discard path

 - cond_resched() in __journal_keys_sort(), cutting down on "hung task"
   warnings when journal is big

 - rest of basic BCH_SB_MEMBER_INVALID support

 - and the critical one: don't delete open files in online fsck, this
   was causing the "dirent points to inode that doesn't point back"
   inconsistencies some users were seeing

* tag 'bcachefs-2024-09-09' of git://evilpiepirate.org/bcachefs:
  bcachefs: Don't delete open files in online fsck
  bcachefs: fix btree_key_cache sysfs knob
  bcachefs: More BCH_SB_MEMBER_INVALID support
  bcachefs: Simplify bch2_bkey_drop_ptrs()
  bcachefs: Add a cond_resched() to __journal_keys_sort()
  bcachefs: Fix ca->io_ref usage

fs/bcachefs/alloc_background.c

@@ -1968,8 +1968,8 @@ static void bch2_do_discards_fast_work(struct work_struct *work)
 			break;
 	}
 
-	bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast);
 	percpu_ref_put(&ca->io_ref);
+	bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast);
 }
 
 static void bch2_discard_one_bucket_fast(struct bch_dev *ca, u64 bucket)
@@ -1979,18 +1979,18 @@ static void bch2_discard_one_bucket_fast(struct bch_dev *ca, u64 bucket)
 	if (discard_in_flight_add(ca, bucket, false))
 		return;
 
-	if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE))
+	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_discard_fast))
 		return;
 
-	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_discard_fast))
-		goto put_ioref;
+	if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE))
+		goto put_ref;
 
 	if (queue_work(c->write_ref_wq, &ca->discard_fast_work))
 		return;
 
-	bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast);
-put_ioref:
 	percpu_ref_put(&ca->io_ref);
+put_ref:
+	bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast);
 }
 
 static int invalidate_one_bucket(struct btree_trans *trans,
@@ -2132,26 +2132,26 @@ static void bch2_do_invalidates_work(struct work_struct *work)
 	bch2_trans_iter_exit(trans, &iter);
 err:
 	bch2_trans_put(trans);
-	bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
 	percpu_ref_put(&ca->io_ref);
+	bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
 }
 
 void bch2_dev_do_invalidates(struct bch_dev *ca)
 {
 	struct bch_fs *c = ca->fs;
 
-	if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE))
+	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate))
 		return;
 
-	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate))
-		goto put_ioref;
+	if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE))
+		goto put_ref;
 
 	if (queue_work(c->write_ref_wq, &ca->invalidate_work))
 		return;
 
-	bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
-put_ioref:
 	percpu_ref_put(&ca->io_ref);
+put_ref:
+	bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
 }
 
 void bch2_do_invalidates(struct bch_fs *c)

fs/bcachefs/btree_journal_iter.c

@@ -530,6 +530,8 @@ static void __journal_keys_sort(struct journal_keys *keys)
 {
 	sort(keys->data, keys->nr, sizeof(keys->data[0]), journal_sort_key_cmp, NULL);
 
+	cond_resched();
+
 	struct journal_key *dst = keys->data;
 
 	darray_for_each(*keys, src) {

fs/bcachefs/buckets.c

@@ -100,12 +100,13 @@ static int bch2_check_fix_ptr(struct btree_trans *trans,
 
 	struct bch_dev *ca = bch2_dev_tryget(c, p.ptr.dev);
 	if (!ca) {
-		if (fsck_err(trans, ptr_to_invalid_device,
-			     "pointer to missing device %u\n"
-			     "while marking %s",
-			     p.ptr.dev,
-			     (printbuf_reset(&buf),
-			      bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
+		if (fsck_err_on(p.ptr.dev != BCH_SB_MEMBER_INVALID,
+				trans, ptr_to_invalid_device,
+				"pointer to missing device %u\n"
+				"while marking %s",
+				p.ptr.dev,
+				(printbuf_reset(&buf),
+				 bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
 			*do_update = true;
 		return 0;
 	}
@@ -562,7 +563,7 @@ static int bch2_trigger_pointer(struct btree_trans *trans,
 	struct bch_fs *c = trans->c;
 	struct bch_dev *ca = bch2_dev_tryget(c, p.ptr.dev);
 	if (unlikely(!ca)) {
-		if (insert)
+		if (insert && p.ptr.dev != BCH_SB_MEMBER_INVALID)
 			ret = -EIO;
 		goto err;
 	}

fs/bcachefs/ec.h

@@ -97,7 +97,9 @@ static inline bool __bch2_ptr_matches_stripe(const struct bch_extent_ptr *stripe
 					     const struct bch_extent_ptr *data_ptr,
 					     unsigned sectors)
 {
-	return  data_ptr->dev    == stripe_ptr->dev &&
+	return  (data_ptr->dev    == stripe_ptr->dev ||
+		 data_ptr->dev    == BCH_SB_MEMBER_INVALID ||
+		 stripe_ptr->dev  == BCH_SB_MEMBER_INVALID) &&
 		data_ptr->gen    == stripe_ptr->gen &&
 		data_ptr->offset >= stripe_ptr->offset &&
 		data_ptr->offset  < stripe_ptr->offset + sectors;

fs/bcachefs/extents.c

@@ -781,14 +781,17 @@ static union bch_extent_entry *extent_entry_prev(struct bkey_ptrs ptrs,
 /*
  * Returns pointer to the next entry after the one being dropped:
  */
-union bch_extent_entry *bch2_bkey_drop_ptr_noerror(struct bkey_s k,
-						   struct bch_extent_ptr *ptr)
+void bch2_bkey_drop_ptr_noerror(struct bkey_s k, struct bch_extent_ptr *ptr)
 {
 	struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
 	union bch_extent_entry *entry = to_entry(ptr), *next;
-	union bch_extent_entry *ret = entry;
 	bool drop_crc = true;
 
+	if (k.k->type == KEY_TYPE_stripe) {
+		ptr->dev = BCH_SB_MEMBER_INVALID;
+		return;
+	}
+
 	EBUG_ON(ptr < &ptrs.start->ptr ||
 		ptr >= &ptrs.end->ptr);
 	EBUG_ON(ptr->type != 1 << BCH_EXTENT_ENTRY_ptr);
@@ -811,21 +814,16 @@ union bch_extent_entry *bch2_bkey_drop_ptr_noerror(struct bkey_s k,
 			break;
 
 		if ((extent_entry_is_crc(entry) && drop_crc) ||
-		    extent_entry_is_stripe_ptr(entry)) {
-			ret = (void *) ret - extent_entry_bytes(entry);
+		    extent_entry_is_stripe_ptr(entry))
 			extent_entry_drop(k, entry);
-		}
 	}
-
-	return ret;
 }
 
-union bch_extent_entry *bch2_bkey_drop_ptr(struct bkey_s k,
-					   struct bch_extent_ptr *ptr)
+void bch2_bkey_drop_ptr(struct bkey_s k, struct bch_extent_ptr *ptr)
 {
 	bool have_dirty = bch2_bkey_dirty_devs(k.s_c).nr;
-	union bch_extent_entry *ret =
-		bch2_bkey_drop_ptr_noerror(k, ptr);
+
+	bch2_bkey_drop_ptr_noerror(k, ptr);
 
 	/*
 	 * If we deleted all the dirty pointers and there's still cached
@@ -837,14 +835,10 @@ union bch_extent_entry *bch2_bkey_drop_ptr(struct bkey_s k,
 	    !bch2_bkey_dirty_devs(k.s_c).nr) {
 		k.k->type = KEY_TYPE_error;
 		set_bkey_val_u64s(k.k, 0);
-		ret = NULL;
 	} else if (!bch2_bkey_nr_ptrs(k.s_c)) {
 		k.k->type = KEY_TYPE_deleted;
 		set_bkey_val_u64s(k.k, 0);
-		ret = NULL;
 	}
-
-	return ret;
 }
 
 void bch2_bkey_drop_device(struct bkey_s k, unsigned dev)

fs/bcachefs/extents.h

@@ -649,26 +649,21 @@ static inline void bch2_bkey_append_ptr(struct bkey_i *k, struct bch_extent_ptr
 
 void bch2_extent_ptr_decoded_append(struct bkey_i *,
 				    struct extent_ptr_decoded *);
-union bch_extent_entry *bch2_bkey_drop_ptr_noerror(struct bkey_s,
-						   struct bch_extent_ptr *);
-union bch_extent_entry *bch2_bkey_drop_ptr(struct bkey_s,
-					   struct bch_extent_ptr *);
+void bch2_bkey_drop_ptr_noerror(struct bkey_s, struct bch_extent_ptr *);
+void bch2_bkey_drop_ptr(struct bkey_s, struct bch_extent_ptr *);
 
 #define bch2_bkey_drop_ptrs(_k, _ptr, _cond)				\
 do {									\
-	struct bkey_ptrs _ptrs = bch2_bkey_ptrs(_k);			\
+	__label__ _again;						\
+	struct bkey_ptrs _ptrs;						\
+_again:									\
+	_ptrs = bch2_bkey_ptrs(_k);					\
 									\
-	struct bch_extent_ptr *_ptr = &_ptrs.start->ptr;		\
-									\
-	while ((_ptr = bkey_ptr_next(_ptrs, _ptr))) {			\
+	bkey_for_each_ptr(_ptrs, _ptr)					\
 		if (_cond) {						\
-			_ptr = (void *) bch2_bkey_drop_ptr(_k, _ptr);	\
-			_ptrs = bch2_bkey_ptrs(_k);			\
-			continue;					\
+			bch2_bkey_drop_ptr(_k, _ptr);			\
+			goto _again;					\
 		}							\
-									\
-		(_ptr)++;						\
-	}								\
 } while (0)
 
 bool bch2_bkey_matches_ptr(struct bch_fs *, struct bkey_s_c,

fs/bcachefs/fs.c

@@ -177,6 +177,14 @@ static unsigned bch2_inode_hash(subvol_inum inum)
 	return jhash_3words(inum.subvol, inum.inum >> 32, inum.inum, JHASH_INITVAL);
 }
 
+struct bch_inode_info *__bch2_inode_hash_find(struct bch_fs *c, subvol_inum inum)
+{
+	return to_bch_ei(ilookup5_nowait(c->vfs_sb,
+					 bch2_inode_hash(inum),
+					 bch2_iget5_test,
+					 &inum));
+}
+
 static struct bch_inode_info *bch2_inode_insert(struct bch_fs *c, struct bch_inode_info *inode)
 {
 	subvol_inum inum = inode_inum(inode);

fs/bcachefs/fs.h

@@ -56,6 +56,8 @@ static inline subvol_inum inode_inum(struct bch_inode_info *inode)
 	};
 }
 
+struct bch_inode_info *__bch2_inode_hash_find(struct bch_fs *, subvol_inum);
+
 /*
  * Set if we've gotten a btree error for this inode, and thus the vfs inode and
  * btree inode may be inconsistent:
@@ -194,6 +196,11 @@ int bch2_vfs_init(void);
 
 #define bch2_inode_update_after_write(_trans, _inode, _inode_u, _fields)	({ do {} while (0); })
 
+static inline struct bch_inode_info *__bch2_inode_hash_find(struct bch_fs *c, subvol_inum inum)
+{
+	return NULL;
+}
+
 static inline void bch2_evict_subvolume_inodes(struct bch_fs *c,
 					       snapshot_id_list *s) {}
 static inline void bch2_vfs_exit(void) {}

fs/bcachefs/fsck.c

@@ -8,6 +8,7 @@
 #include "darray.h"
 #include "dirent.h"
 #include "error.h"
+#include "fs.h"
 #include "fs-common.h"
 #include "fsck.h"
 #include "inode.h"
@@ -962,6 +963,22 @@ static int check_inode_dirent_inode(struct btree_trans *trans, struct bkey_s_c i
 	return ret;
 }
 
+static bool bch2_inode_open(struct bch_fs *c, struct bpos p)
+{
+	subvol_inum inum = {
+		.subvol = snapshot_t(c, p.snapshot)->subvol,
+		.inum	= p.offset,
+	};
+
+	/* snapshot tree corruption, can't safely delete */
+	if (!inum.subvol) {
+		bch_err_ratelimited(c, "%s(): snapshot %u has no subvol", __func__, p.snapshot);
+		return true;
+	}
+
+	return __bch2_inode_hash_find(c, inum) != NULL;
+}
+
 static int check_inode(struct btree_trans *trans,
 		       struct btree_iter *iter,
 		       struct bkey_s_c k,
@@ -1040,6 +1057,7 @@ static int check_inode(struct btree_trans *trans,
 	}
 
 	if (u.bi_flags & BCH_INODE_unlinked &&
+	    !bch2_inode_open(c, k.k->p) &&
 	    (!c->sb.clean ||
 	     fsck_err(trans, inode_unlinked_but_clean,
 		      "filesystem marked clean, but inode %llu unlinked",

fs/bcachefs/replicas.c

@@ -796,7 +796,7 @@ bool bch2_have_enough_devs(struct bch_fs *c, struct bch_devs_mask devs,
 			nr_online += test_bit(e->devs[i], devs.d);
 
 			struct bch_dev *ca = bch2_dev_rcu(c, e->devs[i]);
-			nr_failed += ca && ca->mi.state == BCH_MEMBER_STATE_failed;
+			nr_failed += !ca || ca->mi.state == BCH_MEMBER_STATE_failed;
 		}
 		rcu_read_unlock();
 

fs/bcachefs/sysfs.c

@@ -461,7 +461,7 @@ STORE(bch2_fs)
 
 		sc.gfp_mask = GFP_KERNEL;
 		sc.nr_to_scan = strtoul_or_return(buf);
-		c->btree_key_cache.shrink->scan_objects(c->btree_cache.shrink, &sc);
+		c->btree_key_cache.shrink->scan_objects(c->btree_key_cache.shrink, &sc);
 	}
 
 	if (attr == &sysfs_trigger_gc)