QEMU-Devel Archive on lore.kernel.org
 help / color / Atom feed
From: Greg Kurz <groug@kaod.org>
To: qemu-devel@nongnu.org
Cc: Miklos Szeredi <miklos@szeredi.hu>, Greg Kurz <groug@kaod.org>,
	"Dr. David Alan Gilbert" <dgilbert@redhat.com>,
	virtio-fs@redhat.com, Stefan Hajnoczi <stefanha@redhat.com>,
	Vivek Goyal <vgoyal@redhat.com>
Subject: [for-6.1 2/2] virtiofsd: Add support for FUSE_SYNCFS request
Date: Mon, 19 Apr 2021 17:11:42 +0200
Message-ID: <20210419151142.276439-3-groug@kaod.org> (raw)
In-Reply-To: <20210419151142.276439-1-groug@kaod.org>

Honor the expected behavior of syncfs() to synchronously flush all
data and metadata on linux systems. Like the ->sync_fs() superblock
operation in the linux kernel, FUSE_SYNCFS has a 'wait' argument that
tells whether the server should wait for outstanding I/Os to complete
before replying to the client. Anything virtiofsd can do to flush
the caches implies blocking syscalls, so nothing is done if waiting
isn't requested.

Flushing is done with syncfs(). This is suboptimal as it will also
flush writes performed by any other process on the same file system,
and thus add an unbounded time penalty to syncfs(). This may be
optimized in the future, but enforce correctness first.

Signed-off-by: Greg Kurz <groug@kaod.org>
---
 tools/virtiofsd/fuse_lowlevel.c       | 19 ++++++++++++++++++
 tools/virtiofsd/fuse_lowlevel.h       | 13 ++++++++++++
 tools/virtiofsd/passthrough_ll.c      | 29 +++++++++++++++++++++++++++
 tools/virtiofsd/passthrough_seccomp.c |  1 +
 4 files changed, 62 insertions(+)

diff --git a/tools/virtiofsd/fuse_lowlevel.c b/tools/virtiofsd/fuse_lowlevel.c
index 58e32fc96369..2d0c47a7a60e 100644
--- a/tools/virtiofsd/fuse_lowlevel.c
+++ b/tools/virtiofsd/fuse_lowlevel.c
@@ -1870,6 +1870,24 @@ static void do_lseek(fuse_req_t req, fuse_ino_t nodeid,
     }
 }
 
+static void do_syncfs(fuse_req_t req, fuse_ino_t nodeid,
+                      struct fuse_mbuf_iter *iter)
+{
+    struct fuse_syncfs_in *arg;
+
+    arg = fuse_mbuf_iter_advance(iter, sizeof(*arg));
+    if (!arg) {
+        fuse_reply_err(req, EINVAL);
+        return;
+    }
+
+    if (req->se->op.syncfs) {
+        req->se->op.syncfs(req, arg->wait);
+    } else {
+        fuse_reply_err(req, ENOSYS);
+    }
+}
+
 static void do_init(fuse_req_t req, fuse_ino_t nodeid,
                     struct fuse_mbuf_iter *iter)
 {
@@ -2267,6 +2285,7 @@ static struct {
     [FUSE_RENAME2] = { do_rename2, "RENAME2" },
     [FUSE_COPY_FILE_RANGE] = { do_copy_file_range, "COPY_FILE_RANGE" },
     [FUSE_LSEEK] = { do_lseek, "LSEEK" },
+    [FUSE_SYNCFS] = { do_syncfs, "SYNCFS" },
 };
 
 #define FUSE_MAXOP (sizeof(fuse_ll_ops) / sizeof(fuse_ll_ops[0]))
diff --git a/tools/virtiofsd/fuse_lowlevel.h b/tools/virtiofsd/fuse_lowlevel.h
index 3bf786b03485..b5ac42c31799 100644
--- a/tools/virtiofsd/fuse_lowlevel.h
+++ b/tools/virtiofsd/fuse_lowlevel.h
@@ -1225,6 +1225,19 @@ struct fuse_lowlevel_ops {
      */
     void (*lseek)(fuse_req_t req, fuse_ino_t ino, off_t off, int whence,
                   struct fuse_file_info *fi);
+
+    /**
+     * Synchronize file system content
+     *
+     * If this request is answered with an error code of ENOSYS,
+     * this is treated as success and future calls to syncfs() will
+     * succeed automatically without being sent to the filesystem
+     * process.
+     *
+     * @param req request handle
+     * @param wait whether to wait for outstanding I/Os to complete
+     */
+    void (*syncfs)(fuse_req_t req, int wait);
 };
 
 /**
diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passthrough_ll.c
index 1553d2ef454f..6b66f3208be0 100644
--- a/tools/virtiofsd/passthrough_ll.c
+++ b/tools/virtiofsd/passthrough_ll.c
@@ -3124,6 +3124,34 @@ static void lo_lseek(fuse_req_t req, fuse_ino_t ino, off_t off, int whence,
     }
 }
 
+static void lo_syncfs(fuse_req_t req, int wait)
+{
+    if (wait) {
+        struct lo_data *lo = lo_data(req);
+        int fd, ret;
+
+        fd = lo_inode_open(lo, &lo->root, O_RDONLY);
+        if (fd < 0) {
+            fuse_reply_err(req, errno);
+            return;
+        }
+
+        /*
+         * FIXME: this is suboptimal because it will also flush unrelated
+         *        writes not coming from the client. This can dramatically
+         *        increase the time spent in syncfs() if some process is
+         *        writing lots of data on the same filesystem as virtiofsd.
+         */
+        ret = syncfs(fd);
+        /* syncfs() never fails on a valid fd */
+        assert(ret == 0);
+
+        close(fd);
+    }
+
+    fuse_reply_err(req, 0);
+}
+
 static void lo_destroy(void *userdata)
 {
     struct lo_data *lo = (struct lo_data *)userdata;
@@ -3184,6 +3212,7 @@ static struct fuse_lowlevel_ops lo_oper = {
     .copy_file_range = lo_copy_file_range,
 #endif
     .lseek = lo_lseek,
+    .syncfs = lo_syncfs,
     .destroy = lo_destroy,
 };
 
diff --git a/tools/virtiofsd/passthrough_seccomp.c b/tools/virtiofsd/passthrough_seccomp.c
index 62441cfcdb95..343188447901 100644
--- a/tools/virtiofsd/passthrough_seccomp.c
+++ b/tools/virtiofsd/passthrough_seccomp.c
@@ -107,6 +107,7 @@ static const int syscall_allowlist[] = {
     SCMP_SYS(set_robust_list),
     SCMP_SYS(setxattr),
     SCMP_SYS(symlinkat),
+    SCMP_SYS(syncfs),
     SCMP_SYS(time), /* Rarely needed, except on static builds */
     SCMP_SYS(tgkill),
     SCMP_SYS(unlinkat),
-- 
2.26.3



  parent reply index

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-04-19 15:11 [for-6.1 0/2] " Greg Kurz
2021-04-19 15:11 ` [for-6.1 1/2] Update linux headers to 5.12-rc8 + FUSE_SYNCFS Greg Kurz
2021-04-19 15:11 ` Greg Kurz [this message]
2021-04-20 18:57   ` [Virtio-fs] [for-6.1 2/2] virtiofsd: Add support for FUSE_SYNCFS request Vivek Goyal
2021-04-21  7:44     ` Greg Kurz

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20210419151142.276439-3-groug@kaod.org \
    --to=groug@kaod.org \
    --cc=dgilbert@redhat.com \
    --cc=miklos@szeredi.hu \
    --cc=qemu-devel@nongnu.org \
    --cc=stefanha@redhat.com \
    --cc=vgoyal@redhat.com \
    --cc=virtio-fs@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

QEMU-Devel Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/qemu-devel/0 qemu-devel/git/0.git
	git clone --mirror https://lore.kernel.org/qemu-devel/1 qemu-devel/git/1.git
	git clone --mirror https://lore.kernel.org/qemu-devel/2 qemu-devel/git/2.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 qemu-devel qemu-devel/ https://lore.kernel.org/qemu-devel \
		qemu-devel@nongnu.org
	public-inbox-index qemu-devel

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.nongnu.qemu-devel


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git