From: Richard Weinberger <richard@nod.at>
To: linux-nfs@vger.kernel.org
Cc: bfields@fieldses.org, luis.turcitu@appsbroker.com,
chris.chilvers@appsbroker.com, david.young@appsbroker.com,
daire@dneg.com, david.oberhollenzer@sigma-star.at,
david@sigma-star.at, goliath@sigma-star.at,
trond.myklebust@hammerspace.com, anna.schumaker@netapp.com,
Richard Weinberger <richard@nod.at>
Subject: [RFC PATCH] mountd: export: Deal with NFS filesystems
Date: Mon, 31 Jan 2022 11:43:16 +0100 [thread overview]
Message-ID: <20220131104316.10357-1-richard@nod.at> (raw)
When a NFS filesystem is exported usually NFSD refuses
to export the filesystem because mountd was unable to provide
an UUID and the kernel cannot derive a dev id from the NFS client
super block.
To deal with this situation, teach uuid_by_path() how to generate
an UUID from a NFS filesystem.
Using /proc/fs/nfsfs/volumes it is possible to find the NFS fsid
from the backend and use it as seed for mountd's UUID mechanism.
Signed-off-by: Richard Weinberger <richard@nod.at>
---
Hello,
This is the userspace side to make cross mounts work for NFS re-exports.
The basic idea is that mountd takes the NFS fsid from the backend server
and feeds into mountd'd existing UUID mechanism.
Getting the NFS fsid is currently a bit clumsy, mountd has to scan
/proc/fs/nfsfs/volumes. With the upcoming fsinfo() system call this
information could be transferred in a much simpler way.
e.g. NFS client stores the fsid in its sb->s_uuid and fsinfo() exposes
sb->s_uuid to userspace.
This approach works for v3-to-v4 and v4-to-v4 re-exports.
v3-to-v3 does *not* work, the reason is that nfs_encode_fh() aborts due
to not enough space in the fhandle.
I'm still investigating on how to improve this case.
Currently I’m toying around with a NFS fsid to export FSID_NUM mappings
scheme which are stored to disk.
Another possibility is a new fsid type for NFSD, where the raw NFS fsid
from the backend server is mirrored. Although such a mirror mode makes
local exports impossible.
At the end of the day this is something the admin has to decide.
I don't think there will be a generic solution which satisfies all needs.
Looking forward for your feedback. :-)
Thanks,
//richard
---
support/export/cache.c | 65 +++++++++++++++++++++++++++++++++++++++---
1 file changed, 61 insertions(+), 4 deletions(-)
diff --git a/support/export/cache.c b/support/export/cache.c
index a5823e92e9f2..053ad86619b8 100644
--- a/support/export/cache.c
+++ b/support/export/cache.c
@@ -331,6 +331,51 @@ static const unsigned long nonblkid_filesystems[] = {
0 /* last */
};
+static int get_uuid_from_fsid(char *path, char *uuid_str, size_t len)
+{
+ unsigned int min_dev, maj_dev, min_fsid, maj_fsid;
+ int rc, n, found = 0, header_seen = 0;
+ struct stat stb;
+ FILE *nfsfs_fd;
+ char line[128];
+
+ rc = nfsd_path_stat(path, &stb);
+ if (rc) {
+ xlog(L_WARNING, "Unable to stat %s", path);
+ return 0;
+ }
+
+ nfsfs_fd = fopen("/proc/fs/nfsfs/volumes", "r");
+ if (nfsfs_fd == NULL) {
+ xlog(L_WARNING, "Unable to open nfsfs volume file: %m");
+ return 0;
+ }
+
+ while (fgets(line, sizeof(line), nfsfs_fd) != NULL) {
+ if (!header_seen) {
+ header_seen = 1;
+ continue;
+ }
+ n = sscanf(line, "v%*u %*x %*u %u:%u %x:%x %*s", &maj_dev,
+ &min_dev, &maj_fsid, &min_fsid);
+
+ if (n != 4) {
+ xlog(L_WARNING, "Unable to parse nfsfs volume line: %d, %s", n, line);
+ continue;
+ }
+
+ if (makedev(maj_dev, min_dev) == stb.st_dev) {
+ found = 1;
+ snprintf(uuid_str, len, "%08x%08x", maj_fsid, min_fsid);
+ break;
+ }
+ }
+
+ fclose(nfsfs_fd);
+
+ return found;
+}
+
static int uuid_by_path(char *path, int type, size_t uuidlen, char *uuid)
{
/* get a uuid for the filesystem found at 'path'.
@@ -362,7 +407,7 @@ static int uuid_by_path(char *path, int type, size_t uuidlen, char *uuid)
*/
struct statfs64 st;
char fsid_val[17];
- const char *blkid_val = NULL;
+ const char *fsuuid_val = NULL;
const char *val;
int rc;
@@ -375,7 +420,19 @@ static int uuid_by_path(char *path, int type, size_t uuidlen, char *uuid)
break;
}
if (*bad == 0)
- blkid_val = get_uuid_blkdev(path);
+ fsuuid_val = get_uuid_blkdev(path);
+ else if (*bad == 0x6969 /* NFS_SUPER_MAGIC */) {
+ char tmp[17];
+ int ret = get_uuid_from_fsid(path, tmp, sizeof(tmp));
+
+ if (ret < 0) {
+ xlog(L_WARNING, "Unable to read nfsfs volume file: %i", ret);
+ } else if (ret == 0) {
+ xlog(L_WARNING, "Unable to find nfsfs volume entry for %s", path);
+ } else {
+ fsuuid_val = tmp;
+ }
+ }
}
if (rc == 0 &&
@@ -385,8 +442,8 @@ static int uuid_by_path(char *path, int type, size_t uuidlen, char *uuid)
else
fsid_val[0] = 0;
- if (blkid_val && (type--) == 0)
- val = blkid_val;
+ if (fsuuid_val && (type--) == 0)
+ val = fsuuid_val;
else if (fsid_val[0] && (type--) == 0)
val = fsid_val;
else
--
2.26.2
next reply other threads:[~2022-01-31 10:44 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-01-31 10:43 Richard Weinberger [this message]
2022-01-31 17:01 ` [RFC PATCH] mountd: export: Deal with NFS filesystems J. Bruce Fields
2022-01-31 17:15 ` Richard Weinberger
2022-01-31 20:37 ` bfields
2022-01-31 21:57 ` Richard Weinberger
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=20220131104316.10357-1-richard@nod.at \
--to=richard@nod.at \
--cc=anna.schumaker@netapp.com \
--cc=bfields@fieldses.org \
--cc=chris.chilvers@appsbroker.com \
--cc=daire@dneg.com \
--cc=david.oberhollenzer@sigma-star.at \
--cc=david.young@appsbroker.com \
--cc=david@sigma-star.at \
--cc=goliath@sigma-star.at \
--cc=linux-nfs@vger.kernel.org \
--cc=luis.turcitu@appsbroker.com \
--cc=trond.myklebust@hammerspace.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
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox