Update bcachefs sources to 386f00b639 bcachefs: Snapshot creation, deletion

1 files changed, 210 insertions, 72 deletions

diff --git a/libbcachefs/fs-common.c b/libbcachefs/fs-common.c
index 6bc82559..3e8e3c5b 100644
--- a/libbcachefs/fs-common.c
+++ b/libbcachefs/fs-common.c
@@ -6,82 +6,186 @@
 #include "dirent.h"
 #include "fs-common.h"
 #include "inode.h"
+#include "subvolume.h"
 #include "xattr.h"
 
 #include <linux/posix_acl.h>
 
-int bch2_create_trans(struct btree_trans *trans, u64 dir_inum,
+static inline int is_subdir_for_nlink(struct bch_inode_unpacked *inode)
+{
+	return S_ISDIR(inode->bi_mode) && !inode->bi_subvol;
+}
+
+int bch2_create_trans(struct btree_trans *trans,
+		      subvol_inum dir,
 		      struct bch_inode_unpacked *dir_u,
 		      struct bch_inode_unpacked *new_inode,
 		      const struct qstr *name,
 		      uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
 		      struct posix_acl *default_acl,
-		      struct posix_acl *acl)
+		      struct posix_acl *acl,
+		      subvol_inum snapshot_src,
+		      unsigned flags)
 {
 	struct bch_fs *c = trans->c;
 	struct btree_iter dir_iter = { NULL };
 	struct btree_iter inode_iter = { NULL };
-	struct bch_hash_info hash = bch2_hash_info_init(c, new_inode);
+	subvol_inum new_inum = dir;
 	u64 now = bch2_current_time(c);
 	u64 cpu = raw_smp_processor_id();
-	u64 dir_offset = 0;
+	u64 dir_target;
+	u32 snapshot;
+	unsigned dir_type = mode_to_type(mode);
 	int ret;
 
-	ret = bch2_inode_peek(trans, &dir_iter, dir_u, dir_inum, BTREE_ITER_INTENT);
+	ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
 	if (ret)
 		goto err;
 
-	bch2_inode_init_late(new_inode, now, uid, gid, mode, rdev, dir_u);
-
-	if (!name)
-		new_inode->bi_flags |= BCH_INODE_UNLINKED;
-
-	ret = bch2_inode_create(trans, &inode_iter, new_inode, U32_MAX, cpu);
+	ret = bch2_inode_peek(trans, &dir_iter, dir_u, dir, BTREE_ITER_INTENT);
 	if (ret)
 		goto err;
 
-	if (default_acl) {
-		ret = bch2_set_acl_trans(trans, new_inode, &hash,
-					 default_acl, ACL_TYPE_DEFAULT);
+	if (!(flags & BCH_CREATE_SNAPSHOT)) {
+		/* Normal create path - allocate a new inode: */
+		bch2_inode_init_late(new_inode, now, uid, gid, mode, rdev, dir_u);
+
+		if (flags & BCH_CREATE_TMPFILE)
+			new_inode->bi_flags |= BCH_INODE_UNLINKED;
+
+		ret = bch2_inode_create(trans, &inode_iter, new_inode, snapshot, cpu);
 		if (ret)
 			goto err;
+
+		snapshot_src = (subvol_inum) { 0 };
+	} else {
+		/*
+		 * Creating a snapshot - we're not allocating a new inode, but
+		 * we do have to lookup the root inode of the subvolume we're
+		 * snapshotting and update it (in the new snapshot):
+		 */
+
+		if (!snapshot_src.inum) {
+			/* Inode wasn't specified, just snapshot: */
+			struct btree_iter subvol_iter;
+			struct bkey_s_c k;
+
+			bch2_trans_iter_init(trans, &subvol_iter, BTREE_ID_subvolumes,
+					     POS(0, snapshot_src.subvol), 0);
+			k = bch2_btree_iter_peek_slot(&subvol_iter);
+
+			ret = bkey_err(k);
+			if (!ret && k.k->type != KEY_TYPE_subvolume) {
+				bch_err(c, "subvolume %u not found",
+					snapshot_src.subvol);
+				ret = -ENOENT;
+			}
+
+			if (!ret)
+				snapshot_src.inum = le64_to_cpu(bkey_s_c_to_subvolume(k).v->inode);
+			bch2_trans_iter_exit(trans, &subvol_iter);
+
+			if (ret)
+				goto err;
+		}
+
+		ret = bch2_inode_peek(trans, &inode_iter, new_inode, snapshot_src,
+				      BTREE_ITER_INTENT);
+		if (ret)
+			goto err;
+
+		if (new_inode->bi_subvol != snapshot_src.subvol) {
+			/* Not a subvolume root: */
+			ret = -EINVAL;
+			goto err;
+		}
+
+		/*
+		 * If we're not root, we have to own the subvolume being
+		 * snapshotted:
+		 */
+		if (uid && new_inode->bi_uid != uid) {
+			ret = -EPERM;
+			goto err;
+		}
+
+		flags |= BCH_CREATE_SUBVOL;
 	}
 
-	if (acl) {
-		ret = bch2_set_acl_trans(trans, new_inode, &hash,
-					 acl, ACL_TYPE_ACCESS);
+	new_inum.inum	= new_inode->bi_inum;
+	dir_target	= new_inode->bi_inum;
+
+	if (flags & BCH_CREATE_SUBVOL) {
+		u32 new_subvol, dir_snapshot;
+
+		ret = bch2_subvolume_create(trans, new_inode->bi_inum,
+					    snapshot_src.subvol,
+					    &new_subvol, &snapshot,
+					    (flags & BCH_CREATE_SNAPSHOT_RO) != 0);
 		if (ret)
 			goto err;
+
+		new_inode->bi_parent_subvol	= dir.subvol;
+		new_inode->bi_subvol		= new_subvol;
+		new_inum.subvol			= new_subvol;
+		dir_target			= new_subvol;
+		dir_type			= DT_SUBVOL;
+
+		ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &dir_snapshot);
+		if (ret)
+			goto err;
+
+		bch2_btree_iter_set_snapshot(&dir_iter, dir_snapshot);
+		ret = bch2_btree_iter_traverse(&dir_iter);
+		if (ret)
+			goto err;
+	}
+
+	if (!(flags & BCH_CREATE_SNAPSHOT)) {
+		if (default_acl) {
+			ret = bch2_set_acl_trans(trans, new_inum, new_inode,
+						 default_acl, ACL_TYPE_DEFAULT);
+			if (ret)
+				goto err;
+		}
+
+		if (acl) {
+			ret = bch2_set_acl_trans(trans, new_inum, new_inode,
+						 acl, ACL_TYPE_ACCESS);
+			if (ret)
+				goto err;
+		}
 	}
 
-	if (name) {
+	if (!(flags & BCH_CREATE_TMPFILE)) {
 		struct bch_hash_info dir_hash = bch2_hash_info_init(c, dir_u);
-		dir_u->bi_mtime = dir_u->bi_ctime = now;
+		u64 dir_offset;
 
-		if (S_ISDIR(new_inode->bi_mode))
+		if (is_subdir_for_nlink(new_inode))
 			dir_u->bi_nlink++;
+		dir_u->bi_mtime = dir_u->bi_ctime = now;
 
 		ret = bch2_inode_write(trans, &dir_iter, dir_u);
 		if (ret)
 			goto err;
 
-		ret = bch2_dirent_create(trans, dir_inum, &dir_hash,
-					 mode_to_type(new_inode->bi_mode),
-					 name, new_inode->bi_inum,
+		ret = bch2_dirent_create(trans, dir, &dir_hash,
+					 dir_type,
+					 name,
+					 dir_target,
 					 &dir_offset,
 					 BCH_HASH_SET_MUST_CREATE);
 		if (ret)
 			goto err;
-	}
 
-	if (c->sb.version >= bcachefs_metadata_version_inode_backpointers) {
-		new_inode->bi_dir		= dir_u->bi_inum;
-		new_inode->bi_dir_offset	= dir_offset;
+		if (c->sb.version >= bcachefs_metadata_version_inode_backpointers) {
+			new_inode->bi_dir		= dir_u->bi_inum;
+			new_inode->bi_dir_offset	= dir_offset;
+		}
 	}
 
-	/* XXX use bch2_btree_iter_set_snapshot() */
-	inode_iter.snapshot = U32_MAX;
-	bch2_btree_iter_set_pos(&inode_iter, SPOS(0, new_inode->bi_inum, U32_MAX));
+	inode_iter.flags &= ~BTREE_ITER_ALL_SNAPSHOTS;
+	bch2_btree_iter_set_snapshot(&inode_iter, snapshot);
 
 	ret   = bch2_btree_iter_traverse(&inode_iter) ?:
 		bch2_inode_write(trans, &inode_iter, new_inode);
@@ -91,9 +195,10 @@ err:
 	return ret;
 }
 
-int bch2_link_trans(struct btree_trans *trans, u64 dir_inum,
-		    u64 inum, struct bch_inode_unpacked *dir_u,
-		    struct bch_inode_unpacked *inode_u, const struct qstr *name)
+int bch2_link_trans(struct btree_trans *trans,
+		    subvol_inum dir,  struct bch_inode_unpacked *dir_u,
+		    subvol_inum inum, struct bch_inode_unpacked *inode_u,
+		    const struct qstr *name)
 {
 	struct bch_fs *c = trans->c;
 	struct btree_iter dir_iter = { NULL };
@@ -103,6 +208,9 @@ int bch2_link_trans(struct btree_trans *trans, u64 dir_inum,
 	u64 dir_offset = 0;
 	int ret;
 
+	if (dir.subvol != inum.subvol)
+		return -EXDEV;
+
 	ret = bch2_inode_peek(trans, &inode_iter, inode_u, inum, BTREE_ITER_INTENT);
 	if (ret)
 		goto err;
@@ -110,7 +218,7 @@ int bch2_link_trans(struct btree_trans *trans, u64 dir_inum,
 	inode_u->bi_ctime = now;
 	bch2_inode_nlink_inc(inode_u);
 
-	ret = bch2_inode_peek(trans, &dir_iter, dir_u, dir_inum, BTREE_ITER_INTENT);
+	ret = bch2_inode_peek(trans, &dir_iter, dir_u, dir, BTREE_ITER_INTENT);
 	if (ret)
 		goto err;
 
@@ -118,15 +226,15 @@ int bch2_link_trans(struct btree_trans *trans, u64 dir_inum,
 
 	dir_hash = bch2_hash_info_init(c, dir_u);
 
-	ret = bch2_dirent_create(trans, dir_inum, &dir_hash,
+	ret = bch2_dirent_create(trans, dir, &dir_hash,
 				 mode_to_type(inode_u->bi_mode),
-				 name, inum, &dir_offset,
+				 name, inum.inum, &dir_offset,
 				 BCH_HASH_SET_MUST_CREATE);
 	if (ret)
 		goto err;
 
 	if (c->sb.version >= bcachefs_metadata_version_inode_backpointers) {
-		inode_u->bi_dir		= dir_inum;
+		inode_u->bi_dir		= dir.inum;
 		inode_u->bi_dir_offset	= dir_offset;
 	}
 
@@ -139,55 +247,83 @@ err:
 }
 
 int bch2_unlink_trans(struct btree_trans *trans,
-		      u64 dir_inum, struct bch_inode_unpacked *dir_u,
+		      subvol_inum dir,
+		      struct bch_inode_unpacked *dir_u,
 		      struct bch_inode_unpacked *inode_u,
-		      const struct qstr *name)
+		      const struct qstr *name,
+		      int deleting_snapshot)
 {
 	struct bch_fs *c = trans->c;
 	struct btree_iter dir_iter = { NULL };
 	struct btree_iter dirent_iter = { NULL };
 	struct btree_iter inode_iter = { NULL };
 	struct bch_hash_info dir_hash;
-	u64 inum, now = bch2_current_time(c);
+	subvol_inum inum;
+	u64 now = bch2_current_time(c);
 	struct bkey_s_c k;
 	int ret;
 
-	ret = bch2_inode_peek(trans, &dir_iter, dir_u, dir_inum, BTREE_ITER_INTENT);
+	ret = bch2_inode_peek(trans, &dir_iter, dir_u, dir, BTREE_ITER_INTENT);
 	if (ret)
 		goto err;
 
 	dir_hash = bch2_hash_info_init(c, dir_u);
 
-	ret = __bch2_dirent_lookup_trans(trans, &dirent_iter, dir_inum, &dir_hash,
-					 name, BTREE_ITER_INTENT);
+	ret = __bch2_dirent_lookup_trans(trans, &dirent_iter, dir, &dir_hash,
+					 name, &inum, BTREE_ITER_INTENT);
 	if (ret)
 		goto err;
 
-	k = bch2_btree_iter_peek_slot(&dirent_iter);
-	ret = bkey_err(k);
+	ret = bch2_inode_peek(trans, &inode_iter, inode_u, inum,
+			      BTREE_ITER_INTENT);
 	if (ret)
 		goto err;
 
-	inum = le64_to_cpu(bkey_s_c_to_dirent(k).v->d_inum);
-
-	ret = bch2_inode_peek(trans, &inode_iter, inode_u, inum, BTREE_ITER_INTENT);
-	if (ret)
+	if (deleting_snapshot == 1 && !inode_u->bi_subvol) {
+		ret = -ENOENT;
 		goto err;
+	}
+
+	if (deleting_snapshot <= 0 && S_ISDIR(inode_u->bi_mode)) {
+		ret = bch2_empty_dir_trans(trans, inum);
+		if (ret)
+			goto err;
+	}
+
+	if (inode_u->bi_subvol) {
+		ret = bch2_subvolume_delete(trans, inode_u->bi_subvol,
+					    deleting_snapshot);
+		if (ret)
+			goto err;
+
+		k = bch2_btree_iter_peek_slot(&dirent_iter);
+		ret = bkey_err(k);
+		if (ret)
+			goto err;
+
+		/*
+		 * If we're deleting a subvolume, we need to really delete the
+		 * dirent, not just emit a whiteout in the current snapshot:
+		 */
+		bch2_btree_iter_set_snapshot(&dirent_iter, k.k->p.snapshot);
+		ret = bch2_btree_iter_traverse(&dirent_iter);
+		if (ret)
+			goto err;
+	}
 
-	if (inode_u->bi_dir		== k.k->p.inode &&
-	    inode_u->bi_dir_offset	== k.k->p.offset) {
+	if (inode_u->bi_dir		== dirent_iter.pos.inode &&
+	    inode_u->bi_dir_offset	== dirent_iter.pos.offset) {
 		inode_u->bi_dir		= 0;
 		inode_u->bi_dir_offset	= 0;
 	}
 
 	dir_u->bi_mtime = dir_u->bi_ctime = inode_u->bi_ctime = now;
-	dir_u->bi_nlink -= S_ISDIR(inode_u->bi_mode);
+	dir_u->bi_nlink -= is_subdir_for_nlink(inode_u);
 	bch2_inode_nlink_dec(inode_u);
 
-	ret =   (S_ISDIR(inode_u->bi_mode)
-		 ? bch2_empty_dir_trans(trans, inum)
-		 : 0) ?:
-		bch2_dirent_delete_at(trans, &dir_hash, &dirent_iter) ?:
+	ret =   bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
+				    &dir_hash, &dirent_iter,
+				    BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
 		bch2_inode_write(trans, &dir_iter, dir_u) ?:
 		bch2_inode_write(trans, &inode_iter, inode_u);
 err:
@@ -222,8 +358,8 @@ bool bch2_reinherit_attrs(struct bch_inode_unpacked *dst_u,
 }
 
 int bch2_rename_trans(struct btree_trans *trans,
-		      u64 src_dir, struct bch_inode_unpacked *src_dir_u,
-		      u64 dst_dir, struct bch_inode_unpacked *dst_dir_u,
+		      subvol_inum src_dir, struct bch_inode_unpacked *src_dir_u,
+		      subvol_inum dst_dir, struct bch_inode_unpacked *dst_dir_u,
 		      struct bch_inode_unpacked *src_inode_u,
 		      struct bch_inode_unpacked *dst_inode_u,
 		      const struct qstr *src_name,
@@ -236,7 +372,8 @@ int bch2_rename_trans(struct btree_trans *trans,
 	struct btree_iter src_inode_iter = { NULL };
 	struct btree_iter dst_inode_iter = { NULL };
 	struct bch_hash_info src_hash, dst_hash;
-	u64 src_inode, src_offset, dst_inode, dst_offset;
+	subvol_inum src_inum, dst_inum;
+	u64 src_offset, dst_offset;
 	u64 now = bch2_current_time(c);
 	int ret;
 
@@ -247,7 +384,8 @@ int bch2_rename_trans(struct btree_trans *trans,
 
 	src_hash = bch2_hash_info_init(c, src_dir_u);
 
-	if (dst_dir != src_dir) {
+	if (dst_dir.inum	!= src_dir.inum ||
+	    dst_dir.subvol	!= src_dir.subvol) {
 		ret = bch2_inode_peek(trans, &dst_dir_iter, dst_dir_u, dst_dir,
 				      BTREE_ITER_INTENT);
 		if (ret)
@@ -262,19 +400,19 @@ int bch2_rename_trans(struct btree_trans *trans,
 	ret = bch2_dirent_rename(trans,
 				 src_dir, &src_hash,
 				 dst_dir, &dst_hash,
-				 src_name, &src_inode, &src_offset,
-				 dst_name, &dst_inode, &dst_offset,
+				 src_name, &src_inum, &src_offset,
+				 dst_name, &dst_inum, &dst_offset,
 				 mode);
 	if (ret)
 		goto err;
 
-	ret = bch2_inode_peek(trans, &src_inode_iter, src_inode_u, src_inode,
+	ret = bch2_inode_peek(trans, &src_inode_iter, src_inode_u, src_inum,
 			      BTREE_ITER_INTENT);
 	if (ret)
 		goto err;
 
-	if (dst_inode) {
-		ret = bch2_inode_peek(trans, &dst_inode_iter, dst_inode_u, dst_inode,
+	if (dst_inum.inum) {
+		ret = bch2_inode_peek(trans, &dst_inode_iter, dst_inode_u, dst_inum,
 				      BTREE_ITER_INTENT);
 		if (ret)
 			goto err;
@@ -305,7 +443,7 @@ int bch2_rename_trans(struct btree_trans *trans,
 		}
 
 		if (S_ISDIR(dst_inode_u->bi_mode) &&
-		    bch2_empty_dir_trans(trans, dst_inode)) {
+		    bch2_empty_dir_trans(trans, dst_inum)) {
 			ret = -ENOTEMPTY;
 			goto err;
 		}
@@ -324,12 +462,12 @@ int bch2_rename_trans(struct btree_trans *trans,
 		goto err;
 	}
 
-	if (S_ISDIR(src_inode_u->bi_mode)) {
+	if (is_subdir_for_nlink(src_inode_u)) {
 		src_dir_u->bi_nlink--;
 		dst_dir_u->bi_nlink++;
 	}
 
-	if (dst_inode && S_ISDIR(dst_inode_u->bi_mode)) {
+	if (dst_inum.inum && is_subdir_for_nlink(dst_inode_u)) {
 		dst_dir_u->bi_nlink--;
 		src_dir_u->bi_nlink += mode == BCH_RENAME_EXCHANGE;
 	}
@@ -340,22 +478,22 @@ int bch2_rename_trans(struct btree_trans *trans,
 	src_dir_u->bi_mtime		= now;
 	src_dir_u->bi_ctime		= now;
 
-	if (src_dir != dst_dir) {
+	if (src_dir.inum != dst_dir.inum) {
 		dst_dir_u->bi_mtime	= now;
 		dst_dir_u->bi_ctime	= now;
 	}
 
 	src_inode_u->bi_ctime		= now;
 
-	if (dst_inode)
+	if (dst_inum.inum)
 		dst_inode_u->bi_ctime	= now;
 
 	ret =   bch2_inode_write(trans, &src_dir_iter, src_dir_u) ?:
-		(src_dir != dst_dir
+		(src_dir.inum != dst_dir.inum
 		 ? bch2_inode_write(trans, &dst_dir_iter, dst_dir_u)
 		 : 0 ) ?:
 		bch2_inode_write(trans, &src_inode_iter, src_inode_u) ?:
-		(dst_inode
+		(dst_inum.inum
 		 ? bch2_inode_write(trans, &dst_inode_iter, dst_inode_u)
 		 : 0 );
 err: