git: 1d99e8d9a37e - main - p9fs: Use UNLINKAT instead of REMOVE to implement removals

The branch main has been updated by markj:

URL: https://cgit.FreeBSD.org/src/commit/?id=1d99e8d9a37e1ba528628186df57b79fe74e196c

commit 1d99e8d9a37e1ba528628186df57b79fe74e196c
Author:     Mark Johnston <markj@FreeBSD.org>
AuthorDate: 2024-11-07 18:23:32 +0000
Commit:     Mark Johnston <markj@FreeBSD.org>
CommitDate: 2024-11-07 18:23:32 +0000

    p9fs: Use UNLINKAT instead of REMOVE to implement removals
    
    REMOVE doesn't work properly in the face of hard links.  Use UNLINKAT
    instead, which is implemented by qemu and bhyve and lets the client
    specify the name being removed.
    
    PR:             282432
    Reviewed by:    dfr
    Differential Revision:  https://reviews.freebsd.org/D47438
---
 sys/fs/p9fs/p9_client.c   | 21 +++++++++++++++++++++
 sys/fs/p9fs/p9_client.h   |  1 +
 sys/fs/p9fs/p9_protocol.h |  2 ++
 sys/fs/p9fs/p9fs_vnops.c  | 21 ++++++++++++++-------
 4 files changed, 38 insertions(+), 7 deletions(-)

diff --git a/sys/fs/p9fs/p9_client.c b/sys/fs/p9fs/p9_client.c
index 45d24870306c..547de98c4c03 100644
--- a/sys/fs/p9fs/p9_client.c
+++ b/sys/fs/p9fs/p9_client.c
@@ -669,6 +669,27 @@ p9_client_remove(struct p9_fid *fid)
 	return (error);
 }
 
+int
+p9_client_unlink(struct p9_fid *dfid, const char *name, int32_t flags)
+{
+	int error;
+	struct p9_client *clnt;
+	struct p9_req_t *req;
+
+	error = 0;
+	clnt = dfid->clnt;
+
+	req = p9_client_request(clnt, P9PROTO_TUNLINKAT, &error, "dsd",
+	    dfid->fid, name, flags);
+	if (error != 0) {
+		P9_DEBUG(PROTO, "RUNLINKAT fid %d\n", dfid->fid);
+		return (error);
+	}
+
+	p9_free_req(clnt, req);
+	return (error);
+}
+
 /* Inform the file server that the current file represented by fid is no longer
  * needed by the client. Any allocated fid on the server needs a clunk to be
  * destroyed.
diff --git a/sys/fs/p9fs/p9_client.h b/sys/fs/p9fs/p9_client.h
index 8597c0732ba3..79d402c174e8 100644
--- a/sys/fs/p9fs/p9_client.h
+++ b/sys/fs/p9fs/p9_client.h
@@ -140,6 +140,7 @@ int p9_client_write(struct p9_fid *fid, uint64_t offset, uint32_t count, char *d
 int p9_client_file_create(struct p9_fid *fid, char *name, uint32_t perm, int mode,
     char *extension);
 int p9_client_remove(struct p9_fid *fid);
+int p9_client_unlink(struct p9_fid *dfid, const char *name, int32_t flags);
 int p9_dirent_read(struct p9_client *clnt, char *buf, int start, int len,
     struct p9_dirent *dirent);
 int p9_client_statfs(struct p9_fid *fid, struct p9_statfs *stat);
diff --git a/sys/fs/p9fs/p9_protocol.h b/sys/fs/p9fs/p9_protocol.h
index ddd8571adc8d..7ffd7dd67bcf 100644
--- a/sys/fs/p9fs/p9_protocol.h
+++ b/sys/fs/p9fs/p9_protocol.h
@@ -267,6 +267,8 @@ struct p9_iattr_dotl {
 
 #define P9PROTO_TGETATTR_BLK		512
 
+#define	P9PROTO_UNLINKAT_REMOVEDIR	0x200
+
 /* PDU buffer used for SG lists. */
 struct p9_buffer {
 	uint32_t size;
diff --git a/sys/fs/p9fs/p9fs_vnops.c b/sys/fs/p9fs/p9fs_vnops.c
index 14951e96b59d..cfb50c704019 100644
--- a/sys/fs/p9fs/p9fs_vnops.c
+++ b/sys/fs/p9fs/p9fs_vnops.c
@@ -1497,7 +1497,8 @@ out:
  * After that, does a node metadata cleanup on client side.
  */
 static int
-remove_common(struct p9fs_node *np, struct ucred *cred)
+remove_common(struct p9fs_node *dnp, struct p9fs_node *np, const char *name,
+    struct ucred *cred)
 {
 	int error;
 	struct p9fs_session *vses;
@@ -1508,21 +1509,23 @@ remove_common(struct p9fs_node *np, struct ucred *cred)
 	vses = np->p9fs_ses;
 	vp = P9FS_NTOV(np);
 
-	vfid = p9fs_get_fid(vses->clnt, np, cred, VFID, -1, &error);
+	vfid = p9fs_get_fid(vses->clnt, dnp, cred, VFID, -1, &error);
 	if (error != 0)
 		return (error);
 
-	error = p9_client_remove(vfid);
+	error = p9_client_unlink(vfid, name,
+	    np->v_node->v_type == VDIR ? P9PROTO_UNLINKAT_REMOVEDIR : 0);
 	if (error != 0)
 		return (error);
 
 	/* Remove all non-open fids associated with the vp */
-	p9fs_fid_remove_all(np, TRUE);
+	if (np->inode.i_links_count == 1)
+		p9fs_fid_remove_all(np, TRUE);
 
 	/* Invalidate all entries of vnode from name cache and hash list. */
 	cache_purge(vp);
-
 	vfs_hash_remove(vp);
+
 	np->flags |= P9FS_NODE_DELETED;
 
 	return (error);
@@ -1537,8 +1540,10 @@ p9fs_remove(struct vop_remove_args *ap)
 	struct vnode *dvp;
 	struct p9fs_node *dnp;
 	struct p9fs_inode *dinode;
+	struct componentname *cnp;
 	int error;
 
+	cnp = ap->a_cnp;
 	vp = ap->a_vp;
 	np = P9FS_VTON(vp);
 	dvp = ap->a_dvp;
@@ -1550,7 +1555,7 @@ p9fs_remove(struct vop_remove_args *ap)
 	if (vp->v_type == VDIR)
 		return (EISDIR);
 
-	error = remove_common(np, ap->a_cnp->cn_cred);
+	error = remove_common(dnp, np, cnp->cn_nameptr, cnp->cn_cred);
 	if (error == 0)
 		P9FS_DECR_LINKS(dinode);
 
@@ -1566,8 +1571,10 @@ p9fs_rmdir(struct vop_rmdir_args *ap)
 	struct vnode *dvp;
 	struct p9fs_node *dnp;
 	struct p9fs_inode *dinode;
+	struct componentname *cnp;
 	int error;
 
+	cnp = ap->a_cnp;
 	vp = ap->a_vp;
 	np = P9FS_VTON(vp);
 	dvp = ap->a_dvp;
@@ -1576,7 +1583,7 @@ p9fs_rmdir(struct vop_rmdir_args *ap)
 
 	P9_DEBUG(VOPS, "%s: vp %p node %p \n", __func__, vp, np);
 
-	error = remove_common(np, ap->a_cnp->cn_cred);
+	error = remove_common(dnp, np, cnp->cn_nameptr, cnp->cn_cred);
 	if (error == 0)
 		P9FS_DECR_LINKS(dinode);