1 files changed, 223 insertions, 85 deletions

diff --git a/libbcachefs/fs.c b/libbcachefs/fs.c
index 0422dcab..f1472e91 100644
--- a/libbcachefs/fs.c
+++ b/libbcachefs/fs.c
@@ -53,7 +53,7 @@ static void bch2_vfs_inode_init(struct btree_trans *, subvol_inum,
 				struct bch_subvolume *);
 
 /* Set VFS inode flags from bcachefs inode: */
-static inline void bch2_inode_flags_to_vfs(struct bch_inode_info *inode)
+static inline void bch2_inode_flags_to_vfs(struct bch_fs *c, struct bch_inode_info *inode)
 {
 	static const __maybe_unused unsigned bch_flags_to_vfs[] = {
 		[__BCH_INODE_sync]		= S_SYNC,
@@ -64,8 +64,10 @@ static inline void bch2_inode_flags_to_vfs(struct bch_inode_info *inode)
 
 	set_flags(bch_flags_to_vfs, inode->ei_inode.bi_flags, inode->v.i_flags);
 
-	if (inode->ei_inode.bi_casefold)
+	if (bch2_inode_casefold(c, &inode->ei_inode))
 		inode->v.i_flags |= S_CASEFOLD;
+	else
+		inode->v.i_flags &= ~S_CASEFOLD;
 }
 
 void bch2_inode_update_after_write(struct btree_trans *trans,
@@ -96,7 +98,7 @@ void bch2_inode_update_after_write(struct btree_trans *trans,
 
 	inode->ei_inode		= *bi;
 
-	bch2_inode_flags_to_vfs(inode);
+	bch2_inode_flags_to_vfs(c, inode);
 }
 
 int __must_check bch2_write_inode(struct bch_fs *c,
@@ -647,13 +649,18 @@ static struct bch_inode_info *bch2_lookup_trans(struct btree_trans *trans,
 			const struct qstr *name)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter dirent_iter = {};
 	subvol_inum inum = {};
 	struct printbuf buf = PRINTBUF;
 
+	struct qstr lookup_name;
+	int ret = bch2_maybe_casefold(trans, dir_hash_info, name, &lookup_name);
+	if (ret)
+		return ERR_PTR(ret);
+
+	struct btree_iter dirent_iter = {};
 	struct bkey_s_c k = bch2_hash_lookup(trans, &dirent_iter, bch2_dirent_hash_desc,
-					     dir_hash_info, dir, name, 0);
-	int ret = bkey_err(k);
+					     dir_hash_info, dir, &lookup_name, 0);
+	ret = bkey_err(k);
 	if (ret)
 		return ERR_PTR(ret);
 
@@ -841,6 +848,9 @@ int __bch2_unlink(struct inode *vdir, struct dentry *dentry,
 		 */
 		set_nlink(&inode->v, 0);
 	}
+
+	if (IS_CASEFOLDED(vdir))
+		d_invalidate(dentry);
 err:
 	bch2_trans_put(trans);
 	bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
@@ -1251,10 +1261,20 @@ static int bch2_tmpfile(struct mnt_idmap *idmap,
 	return finish_open_simple(file, 0);
 }
 
+struct bch_fiemap_extent {
+	struct bkey_buf	kbuf;
+	unsigned	flags;
+};
+
 static int bch2_fill_extent(struct bch_fs *c,
 			    struct fiemap_extent_info *info,
-			    struct bkey_s_c k, unsigned flags)
+			    struct bch_fiemap_extent *fe)
 {
+	struct bkey_s_c k = bkey_i_to_s_c(fe->kbuf.k);
+	unsigned flags = fe->flags;
+
+	BUG_ON(!k.k->size);
+
 	if (bkey_extent_is_direct_data(k.k)) {
 		struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 		const union bch_extent_entry *entry;
@@ -1307,110 +1327,223 @@ static int bch2_fill_extent(struct bch_fs *c,
 	}
 }
 
-static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
-		       u64 start, u64 len)
+/*
+ * Scan a range of an inode for data in pagecache.
+ *
+ * Intended to be retryable, so don't modify the output params until success is
+ * imminent.
+ */
+static int
+bch2_fiemap_hole_pagecache(struct inode *vinode, u64 *start, u64 *end,
+			   bool nonblock)
 {
-	struct bch_fs *c = vinode->i_sb->s_fs_info;
-	struct bch_inode_info *ei = to_bch_ei(vinode);
-	struct btree_trans *trans;
-	struct btree_iter iter;
-	struct bkey_s_c k;
-	struct bkey_buf cur, prev;
-	bool have_extent = false;
-	int ret = 0;
+	loff_t	dstart, dend;
 
-	ret = fiemap_prep(&ei->v, info, start, &len, FIEMAP_FLAG_SYNC);
-	if (ret)
+	dstart = bch2_seek_pagecache_data(vinode, *start, *end, 0, nonblock);
+	if (dstart < 0)
+		return dstart;
+
+	if (dstart == *end) {
+		*start = dstart;
+		return 0;
+	}
+
+	dend = bch2_seek_pagecache_hole(vinode, dstart, *end, 0, nonblock);
+	if (dend < 0)
+		return dend;
+
+	/* race */
+	BUG_ON(dstart == dend);
+
+	*start = dstart;
+	*end = dend;
+	return 0;
+}
+
+/*
+ * Scan a range of pagecache that corresponds to a file mapping hole in the
+ * extent btree. If data is found, fake up an extent key so it looks like a
+ * delalloc extent to the rest of the fiemap processing code.
+ */
+static int
+bch2_next_fiemap_pagecache_extent(struct btree_trans *trans, struct bch_inode_info *inode,
+				  u64 start, u64 end, struct bch_fiemap_extent *cur)
+{
+	struct bch_fs		*c = trans->c;
+	struct bkey_i_extent	*delextent;
+	struct bch_extent_ptr	ptr = {};
+	loff_t			dstart = start << 9, dend = end << 9;
+	int			ret;
+
+	/*
+	 * We hold btree locks here so we cannot block on folio locks without
+	 * dropping trans locks first. Run a nonblocking scan for the common
+	 * case of no folios over holes and fall back on failure.
+	 *
+	 * Note that dropping locks like this is technically racy against
+	 * writeback inserting to the extent tree, but a non-sync fiemap scan is
+	 * fundamentally racy with writeback anyways. Therefore, just report the
+	 * range as delalloc regardless of whether we have to cycle trans locks.
+	 */
+	ret = bch2_fiemap_hole_pagecache(&inode->v, &dstart, &dend, true);
+	if (ret == -EAGAIN)
+		ret = drop_locks_do(trans,
+			bch2_fiemap_hole_pagecache(&inode->v, &dstart, &dend, false));
+	if (ret < 0)
 		return ret;
 
-	struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
-	if (start + len < start)
-		return -EINVAL;
+	/*
+	 * Create a fake extent key in the buffer. We have to add a dummy extent
+	 * pointer for the fill code to add an extent entry. It's explicitly
+	 * zeroed to reflect delayed allocation (i.e. phys offset 0).
+	 */
+	bch2_bkey_buf_realloc(&cur->kbuf, c, sizeof(*delextent) / sizeof(u64));
+	delextent = bkey_extent_init(cur->kbuf.k);
+	delextent->k.p = POS(inode->ei_inum.inum, dend >> 9);
+	delextent->k.size = (dend - dstart) >> 9;
+	bch2_bkey_append_ptr(&delextent->k_i, ptr);
 
-	start >>= 9;
+	cur->flags = FIEMAP_EXTENT_DELALLOC;
 
-	bch2_bkey_buf_init(&cur);
-	bch2_bkey_buf_init(&prev);
-	trans = bch2_trans_get(c);
+	return 0;
+}
 
+static int bch2_next_fiemap_extent(struct btree_trans *trans,
+				   struct bch_inode_info *inode,
+				   u64 start, u64 end,
+				   struct bch_fiemap_extent *cur)
+{
+	u32 snapshot;
+	int ret = bch2_subvolume_get_snapshot(trans, inode->ei_inum.subvol, &snapshot);
+	if (ret)
+		return ret;
+
+	struct btree_iter iter;
 	bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
-			     POS(ei->v.i_ino, start), 0);
+			     SPOS(inode->ei_inum.inum, start, snapshot), 0);
 
-	while (!ret || bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
-		enum btree_id data_btree = BTREE_ID_extents;
+	struct bkey_s_c k =
+		bch2_btree_iter_peek_max(trans, &iter, POS(inode->ei_inum.inum, end));
+	ret = bkey_err(k);
+	if (ret)
+		goto err;
 
-		bch2_trans_begin(trans);
+	ret = bch2_next_fiemap_pagecache_extent(trans, inode, start, end, cur);
+	if (ret)
+		goto err;
 
-		u32 snapshot;
-		ret = bch2_subvolume_get_snapshot(trans, ei->ei_inum.subvol, &snapshot);
-		if (ret)
-			continue;
+	struct bpos pagecache_start = bkey_start_pos(&cur->kbuf.k->k);
 
-		bch2_btree_iter_set_snapshot(trans, &iter, snapshot);
+	/*
+	 * Does the pagecache or the btree take precedence?
+	 *
+	 * It _should_ be the pagecache, so that we correctly report delalloc
+	 * extents when dirty in the pagecache (we're COW, after all).
+	 *
+	 * But we'd have to add per-sector writeback tracking to
+	 * bch_folio_state, otherwise we report delalloc extents for clean
+	 * cached data in the pagecache.
+	 *
+	 * We should do this, but even then fiemap won't report stable mappings:
+	 * on bcachefs data moves around in the background (copygc, rebalance)
+	 * and we don't provide a way for userspace to lock that out.
+	 */
+	if (k.k &&
+	    bkey_le(bpos_max(iter.pos, bkey_start_pos(k.k)),
+		    pagecache_start)) {
+		bch2_bkey_buf_reassemble(&cur->kbuf, trans->c, k);
+		bch2_cut_front(iter.pos, cur->kbuf.k);
+		bch2_cut_back(POS(inode->ei_inum.inum, end), cur->kbuf.k);
+		cur->flags = 0;
+	} else if (k.k) {
+		bch2_cut_back(bkey_start_pos(k.k), cur->kbuf.k);
+	}
 
-		k = bch2_btree_iter_peek_max(trans, &iter, end);
-		ret = bkey_err(k);
+	if (cur->kbuf.k->k.type == KEY_TYPE_reflink_p) {
+		unsigned sectors = cur->kbuf.k->k.size;
+		s64 offset_into_extent = 0;
+		enum btree_id data_btree = BTREE_ID_extents;
+		ret = bch2_read_indirect_extent(trans, &data_btree, &offset_into_extent,
+						&cur->kbuf);
 		if (ret)
-			continue;
+			goto err;
 
-		if (!k.k)
-			break;
+		struct bkey_i *k = cur->kbuf.k;
+		sectors = min_t(unsigned, sectors, k->k.size - offset_into_extent);
 
-		if (!bkey_extent_is_data(k.k) &&
-		    k.k->type != KEY_TYPE_reservation) {
-			bch2_btree_iter_advance(trans, &iter);
-			continue;
-		}
+		bch2_cut_front(POS(k->k.p.inode,
+				   bkey_start_offset(&k->k) + offset_into_extent),
+			       k);
+		bch2_key_resize(&k->k, sectors);
+		k->k.p = iter.pos;
+		k->k.p.offset += k->k.size;
+	}
+err:
+	bch2_trans_iter_exit(trans, &iter);
+	return ret;
+}
 
-		s64 offset_into_extent	= iter.pos.offset - bkey_start_offset(k.k);
-		unsigned sectors	= k.k->size - offset_into_extent;
+static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
+		       u64 start, u64 len)
+{
+	struct bch_fs *c = vinode->i_sb->s_fs_info;
+	struct bch_inode_info *ei = to_bch_ei(vinode);
+	struct btree_trans *trans;
+	struct bch_fiemap_extent cur, prev;
+	int ret = 0;
+
+	ret = fiemap_prep(&ei->v, info, start, &len, 0);
+	if (ret)
+		return ret;
+
+	if (start + len < start)
+		return -EINVAL;
+
+	start >>= 9;
+	u64 end = (start + len) >> 9;
 
-		bch2_bkey_buf_reassemble(&cur, c, k);
+	bch2_bkey_buf_init(&cur.kbuf);
+	bch2_bkey_buf_init(&prev.kbuf);
+	bkey_init(&prev.kbuf.k->k);
 
-		ret = bch2_read_indirect_extent(trans, &data_btree,
-					&offset_into_extent, &cur);
+	trans = bch2_trans_get(c);
+
+	while (start < end) {
+		ret = lockrestart_do(trans,
+			bch2_next_fiemap_extent(trans, ei, start, end, &cur));
 		if (ret)
-			continue;
+			goto err;
 
-		k = bkey_i_to_s_c(cur.k);
-		bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
+		BUG_ON(bkey_start_offset(&cur.kbuf.k->k) < start);
+		BUG_ON(cur.kbuf.k->k.p.offset > end);
 
-		sectors = min_t(unsigned, sectors, k.k->size - offset_into_extent);
+		if (bkey_start_offset(&cur.kbuf.k->k) == end)
+			break;
 
-		bch2_cut_front(POS(k.k->p.inode,
-				   bkey_start_offset(k.k) +
-				   offset_into_extent),
-			       cur.k);
-		bch2_key_resize(&cur.k->k, sectors);
-		cur.k->k.p = iter.pos;
-		cur.k->k.p.offset += cur.k->k.size;
+		start = cur.kbuf.k->k.p.offset;
 
-		if (have_extent) {
+		if (!bkey_deleted(&prev.kbuf.k->k)) {
 			bch2_trans_unlock(trans);
-			ret = bch2_fill_extent(c, info,
-					bkey_i_to_s_c(prev.k), 0);
+			ret = bch2_fill_extent(c, info, &prev);
 			if (ret)
-				break;
+				goto err;
 		}
 
-		bkey_copy(prev.k, cur.k);
-		have_extent = true;
-
-		bch2_btree_iter_set_pos(trans, &iter,
-			POS(iter.pos.inode, iter.pos.offset + sectors));
+		bch2_bkey_buf_copy(&prev.kbuf, c, cur.kbuf.k);
+		prev.flags = cur.flags;
 	}
-	bch2_trans_iter_exit(trans, &iter);
 
-	if (!ret && have_extent) {
+	if (!bkey_deleted(&prev.kbuf.k->k)) {
 		bch2_trans_unlock(trans);
-		ret = bch2_fill_extent(c, info, bkey_i_to_s_c(prev.k),
-				       FIEMAP_EXTENT_LAST);
+		prev.flags |= FIEMAP_EXTENT_LAST;
+		ret = bch2_fill_extent(c, info, &prev);
 	}
-
+err:
 	bch2_trans_put(trans);
-	bch2_bkey_buf_exit(&cur, c);
-	bch2_bkey_buf_exit(&prev, c);
-	return ret < 0 ? ret : 0;
+	bch2_bkey_buf_exit(&cur.kbuf, c);
+	bch2_bkey_buf_exit(&prev.kbuf, c);
+
+	return bch2_err_class(ret < 0 ? ret : 0);
 }
 
 static const struct vm_operations_struct bch_vm_ops = {
@@ -1487,13 +1620,14 @@ static int bch2_fileattr_get(struct dentry *dentry,
 			     struct fileattr *fa)
 {
 	struct bch_inode_info *inode = to_bch_ei(d_inode(dentry));
+	struct bch_fs *c = inode->v.i_sb->s_fs_info;
 
 	fileattr_fill_xflags(fa, map_flags(bch_flags_to_xflags, inode->ei_inode.bi_flags));
 
 	if (inode->ei_inode.bi_fields_set & (1 << Inode_opt_project))
 		fa->fsx_xflags |= FS_XFLAG_PROJINHERIT;
 
-	if (inode->ei_inode.bi_casefold)
+	if (bch2_inode_casefold(c, &inode->ei_inode))
 		fa->flags |= FS_CASEFOLD_FL;
 
 	fa->fsx_projid = inode->ei_qid.q[QTYP_PRJ];
@@ -1526,7 +1660,7 @@ static int fssetxattr_inode_update_fn(struct btree_trans *trans,
 	    (s->flags & (BCH_INODE_nodump|BCH_INODE_noatime)) != s->flags)
 		return -EINVAL;
 
-	if (s->casefold != bi->bi_casefold) {
+	if (s->casefold != bch2_inode_casefold(c, bi)) {
 #ifdef CONFIG_UNICODE
 		int ret = 0;
 		/* Not supported on individual files. */
@@ -1547,9 +1681,8 @@ static int fssetxattr_inode_update_fn(struct btree_trans *trans,
 
 		bch2_check_set_feature(c, BCH_FEATURE_casefolding);
 
-		bi->bi_casefold = s->casefold;
-		bi->bi_fields_set &= ~BIT(Inode_opt_casefold);
-		bi->bi_fields_set |= s->casefold << Inode_opt_casefold;
+		bi->bi_casefold = s->casefold + 1;
+		bi->bi_fields_set |= BIT(Inode_opt_casefold);
 
 #else
 		printk(KERN_ERR "Cannot use casefolding on a kernel without CONFIG_UNICODE\n");
@@ -2330,7 +2463,7 @@ static int bch2_fs_get_tree(struct fs_context *fc)
 	struct inode *vinode;
 	struct bch2_opts_parse *opts_parse = fc->fs_private;
 	struct bch_opts opts = opts_parse->opts;
-	darray_str devs;
+	darray_const_str devs;
 	darray_fs devs_to_fs = {};
 	int ret;
 
@@ -2354,7 +2487,7 @@ static int bch2_fs_get_tree(struct fs_context *fc)
 	if (!IS_ERR(sb))
 		goto got_sb;
 
-	c = bch2_fs_open(devs.data, devs.nr, opts);
+	c = bch2_fs_open(&devs, &opts);
 	ret = PTR_ERR_OR_ZERO(c);
 	if (ret)
 		goto err;
@@ -2445,6 +2578,11 @@ got_sb:
 	if (ret)
 		goto err_put_super;
 
+#ifdef CONFIG_UNICODE
+	sb->s_encoding = c->cf_encoding;
+#endif
+	generic_set_sb_d_ops(sb);
+
 	vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_SUBVOL_INUM);
 	ret = PTR_ERR_OR_ZERO(vinode);
 	bch_err_msg(c, ret, "mounting: error getting root inode");