| Message ID | 20200827153657.111098-4-dgilbert@redhat.com |
|---|---|
| State | New |
| Headers | show |
| Series | None | expand |
On a Thursday in 2020, Dr. David Alan Gilbert (git) wrote: >From: "Dr. David Alan Gilbert" <dgilbert@redhat.com> > >Add an option to define mappings of xattr names so that >the client and server filesystems see different views. >This can be used to have different SELinux mappings as >seen by the guest, to run the virtiofsd with less privileges >(e.g. in a case where it can't set trusted/system/security >xattrs but you want the guest to be able to), or to isolate >multiple users of the same name; e.g. trusted attributes >used by stacking overlayfs. > >A mapping engine is used wit 3 simple rules; the rules can >be combined to allow most useful mapping scenarios. >The ruleset is defined by -o xattrmap='rules...'. > >This patch doesn't use the rule maps yet. > >Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com> >--- > docs/tools/virtiofsd.rst | 55 ++++++++++++ > tools/virtiofsd/passthrough_ll.c | 148 +++++++++++++++++++++++++++++++ > 2 files changed, 203 insertions(+) > >diff --git a/docs/tools/virtiofsd.rst b/docs/tools/virtiofsd.rst >index 824e713491..2efa16d3c5 100644 >--- a/docs/tools/virtiofsd.rst >+++ b/docs/tools/virtiofsd.rst >@@ -107,6 +107,60 @@ Options > performance. ``auto`` acts similar to NFS with a 1 second metadata cache > timeout. ``always`` sets a long cache lifetime at the expense of coherency. > >+xattr-mapping >+------------- >+ >+By default the name of xattr's used by the client are passed through to the server >+file system. This can be a problem where either those xattr names are used >+by something on the server (e.g. selinux client/server confusion) or if the >+virtiofsd is running in a container with restricted priviliges where it cannot privileges >+access some attributes. >+ >+A mapping of xattr names can be made using -o xattrmap=mapping where the ``mapping`` >+string consists of a series of rules. >+ >+The first matching rule terminates the mapping. >+ >+Each rule consists of a number of fields separated with a separator that is the >+first non-white space character in the rule. This separator must then be used >+for the whole rule. >+White space may be added before and after each rule. >+Using ':' as the separator a rule is of the form: >+ >+``:scope:type:key:prepend:`` >+ >+**scope** is: >+ >+- 'client' - match 'key' against a xattr name from the client for >+ setxattr/getxattr/removexattr >+- 'server' - match 'prepend' against a xattr name from the server >+ for listxattr >+- 'all' - can be used to match both cases. >+ >+**type** is one of: >+ >+- 'prefix' - If 'key' matches the client then the 'prepend' >+ is added before the name is passed to the server. >+ For a server case, the prepend is tested and stripped >+ if matching. >+ >+- 'ok' - The attribute name is OK and passed through to >+ the server unchanged. >+ >+- 'bad' - If a client tries to use this name it's >+ denied using EPERM; when the server passes an attribute >+ name matching it's hidden. >+ >+**key** is a string tested as a prefix on an attribute name originating >+on the client. It maybe empty in which case a 'client' rule >+will always match on client names. >+ >+**prepend** is a string tested as a prefix on an attribute name originiating originating >+on the server, and used as a new prefix. It maybe empty may be >+in which case a 'server' rule will always match on all names from >+the server. >+ >+ > Examples > -------- > >@@ -123,3 +177,4 @@ Export ``/var/lib/fs/vm001/`` on vhost-user UNIX domain socket > -numa node,memdev=mem \ > ... > guest# mount -t virtiofs myfs /mnt >+ git complains about trailing whitespace at EOF Jano
On Thu, Aug 27, 2020 at 04:36:54PM +0100, Dr. David Alan Gilbert (git) wrote: > From: "Dr. David Alan Gilbert" <dgilbert@redhat.com> > > Add an option to define mappings of xattr names so that > the client and server filesystems see different views. > This can be used to have different SELinux mappings as > seen by the guest, to run the virtiofsd with less privileges > (e.g. in a case where it can't set trusted/system/security > xattrs but you want the guest to be able to), or to isolate > multiple users of the same name; e.g. trusted attributes > used by stacking overlayfs. > > A mapping engine is used wit 3 simple rules; the rules can > be combined to allow most useful mapping scenarios. > The ruleset is defined by -o xattrmap='rules...'. > > This patch doesn't use the rule maps yet. > > Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com> > --- > docs/tools/virtiofsd.rst | 55 ++++++++++++ > tools/virtiofsd/passthrough_ll.c | 148 +++++++++++++++++++++++++++++++ > 2 files changed, 203 insertions(+) > > diff --git a/docs/tools/virtiofsd.rst b/docs/tools/virtiofsd.rst > index 824e713491..2efa16d3c5 100644 > --- a/docs/tools/virtiofsd.rst > +++ b/docs/tools/virtiofsd.rst > @@ -107,6 +107,60 @@ Options > performance. ``auto`` acts similar to NFS with a 1 second metadata cache > timeout. ``always`` sets a long cache lifetime at the expense of coherency. > > +xattr-mapping > +------------- > + > +By default the name of xattr's used by the client are passed through to the server > +file system. This can be a problem where either those xattr names are used > +by something on the server (e.g. selinux client/server confusion) or if the > +virtiofsd is running in a container with restricted priviliges where it cannot > +access some attributes. > + > +A mapping of xattr names can be made using -o xattrmap=mapping where the ``mapping`` > +string consists of a series of rules. > + > +The first matching rule terminates the mapping. > + > +Each rule consists of a number of fields separated with a separator that is the > +first non-white space character in the rule. This separator must then be used > +for the whole rule. > +White space may be added before and after each rule. > +Using ':' as the separator a rule is of the form: > + > +``:scope:type:key:prepend:`` Hi David, This seems very genric and which makes it harder to understand and harder to write rules. I am wondering do we really need this degree of flexibility. Is it worth, dropping some of the requirements and simplify the syntax. - I am wonderig why do we need to allow choice of separator. - Wondering why do we need to allow separate rules for client/server. Once we start remapping something, is it not good enough that mapping be bidirectonal. - Not sure why separate notion of "bad". To me once we decide to remap something, should automatically block unprefixed version. IOW, what functionality we will lose if we just say -o remap_xattr="trusted.". This implies following. A. If client is sending any xattr prefixed with "trusted.", prefix it with "user.virtiofs". B. Server filters out anything starting with "trusted." C. If server sees "user.virtiofs.trusted." it strips "user.virtiofs". For remapping security.selinux, user could specify. -o remap_xattr="security.selinux." For nested configuration. virtiofsd at L1 will specify. -o remap_xattr="security.selinux.". And virtiofsd at L0 can specify. -o remap_xattr="user.virtiofs.security.selinux." I doubt we need to care about being able to remap xattrs of other filesystems like virtio-9p. I also have some questions about how this will be used. Overlay ------- - So for non nested guests, we can have two instances of overlay. Lets call these ovl0 and ovl1. (ovl0 being on host, and ovl1 being inside guest). Fs hierarcy might look as follows. ext4-->ovl0-->virtiofsd0-->ovl1 This case does not work by default even if virtiofsd has CAP_SYS_ADMIN by default as overlay does not allow nesting. So when ovl1 tries to set trusted.overlay, ovl0 will deny it. We could simple pass extra directory from host which does not go through overlay on host and use that as upper inside guest. ext4-->ovl0-->virtiofsd0-->ovl1 ext4-->ovl0/upper-->virtiofsd0-->ovl1 (/upper used as upper directory of ovl1) I guess remapping "trusted.overlay" will allow us not to have a separate ovl0/upper. And following itself will work. Have you tested it? Does this work. Basically we are creating nested overlay configuration with virtiofs in between. Is "trusted.overlay" only conflict. I wonder there might be others. Just that "trusted.overlay" is first failure we noticed. Nested Overlay -------------- - For now I will assume that we are using separate upper dir. ext4-->ovl0-->virtiofsd0-->ovl1-->virtiofsd1-->ovl2 ext4-->ovl0/upper1-->virtiofsd0-->ovl1(uses upper1 as upperdir) ext4-->ovl0/upper2-->virtiofsd0-->ovl1-->virtiofsd1-->ovl2 (users upper2 as upper dir) Basically create two directories upper1 and upper2 on regular filesystem say ext4/xfs. Bind mount them on ovl0/upper1 and ovl0/upper2 respectively. And now ovl1 uses ovl0/upper1 as upperdir and ovl2 uses ovl0/upper2 as upperdir. This should make sure ovl0, ovl1 and ovl2 are not nested from sharing upper perspective. Now virtiofsd1 will run with '-o remap_xattr="trusted.overlay"' and virtiofsd0 will run with '-o remap_xattr="user.virtiofs.trusted.overlay"' Just trying to wrap my head around how our use cases will use this new remapping xattr thing. Thanks Vivek > + > +**scope** is: > + > +- 'client' - match 'key' against a xattr name from the client for > + setxattr/getxattr/removexattr > +- 'server' - match 'prepend' against a xattr name from the server > + for listxattr > +- 'all' - can be used to match both cases. > + > +**type** is one of: > + > +- 'prefix' - If 'key' matches the client then the 'prepend' > + is added before the name is passed to the server. > + For a server case, the prepend is tested and stripped > + if matching. > + > +- 'ok' - The attribute name is OK and passed through to > + the server unchanged. > + > +- 'bad' - If a client tries to use this name it's > + denied using EPERM; when the server passes an attribute > + name matching it's hidden. > + > +**key** is a string tested as a prefix on an attribute name originating > +on the client. It maybe empty in which case a 'client' rule > +will always match on client names. > + > +**prepend** is a string tested as a prefix on an attribute name originiating > +on the server, and used as a new prefix. It maybe empty > +in which case a 'server' rule will always match on all names from > +the server. > + > + > Examples > -------- > > @@ -123,3 +177,4 @@ Export ``/var/lib/fs/vm001/`` on vhost-user UNIX domain socket > -numa node,memdev=mem \ > ... > guest# mount -t virtiofs myfs /mnt > + > diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passthrough_ll.c > index 083d17a960..00e96a10cd 100644 > --- a/tools/virtiofsd/passthrough_ll.c > +++ b/tools/virtiofsd/passthrough_ll.c > @@ -64,6 +64,7 @@ > #include <syslog.h> > #include <unistd.h> > > +#include "qemu/cutils.h" > #include "passthrough_helpers.h" > #include "passthrough_seccomp.h" > > @@ -144,6 +145,7 @@ struct lo_data { > int flock; > int posix_lock; > int xattr; > + char *xattrmap; > char *source; > char *modcaps; > double timeout; > @@ -171,6 +173,7 @@ static const struct fuse_opt lo_opts[] = { > { "no_posix_lock", offsetof(struct lo_data, posix_lock), 0 }, > { "xattr", offsetof(struct lo_data, xattr), 1 }, > { "no_xattr", offsetof(struct lo_data, xattr), 0 }, > + { "xattrmap=%s", offsetof(struct lo_data, xattrmap), 0 }, > { "modcaps=%s", offsetof(struct lo_data, modcaps), 0 }, > { "timeout=%lf", offsetof(struct lo_data, timeout), 0 }, > { "timeout=", offsetof(struct lo_data, timeout_set), 1 }, > @@ -2003,6 +2006,146 @@ static void lo_flock(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi, > fuse_reply_err(req, res == -1 ? errno : 0); > } > > +typedef struct xattr_map_entry { > + const char *key; > + const char *prepend; > + unsigned int flags; > +} XattrMapEntry; > + > +/* > + * Exit; process attribute unmodified if matched. > + * An empty key applies to all. > + */ > +#define XATTR_MAP_FLAG_END_OK (1 << 0) > +/* > + * The attribute is unwanted; > + * EPERM on write hidden on read. > + */ > +#define XATTR_MAP_FLAG_END_BAD (1 << 1) > +/* > + * For attr that start with 'key' prepend 'prepend' > + * 'key' maybe empty to prepend for all attrs > + * key is defined from set/remove point of view. > + * Automatically reversed on read > + */ > +#define XATTR_MAP_FLAG_PREFIX (1 << 2) > +/* Apply rule to get/set/remove */ > +#define XATTR_MAP_FLAG_CLIENT (1 << 16) > +/* Apply rule to list */ > +#define XATTR_MAP_FLAG_SERVER (1 << 17) > +/* Apply rule to all */ > +#define XATTR_MAP_FLAG_ALL (XATTR_MAP_FLAG_SERVER | XATTR_MAP_FLAG_CLIENT) > + > +static XattrMapEntry *xattr_map_list; > + > +static XattrMapEntry *parse_xattrmap(const char *map) > +{ > + XattrMapEntry *res = NULL; > + size_t nentries = 0; > + const char *tmp; > + > + while (*map) { > + char sep; > + > + if (isspace(*map)) { > + map++; > + continue; > + } > + /* The separator is the first non-space of the rule */ > + sep = *map++; > + if (!sep) { > + break; > + } > + > + /* Allocate some space for the rule */ > + res = g_realloc_n(res, ++nentries, sizeof(XattrMapEntry)); > + res[nentries - 1].flags = 0; > + > + if (strstart(map, "client", &map)) { > + res[nentries - 1].flags |= XATTR_MAP_FLAG_CLIENT; > + } else if (strstart(map, "server", &map)) { > + res[nentries - 1].flags |= XATTR_MAP_FLAG_SERVER; > + } else if (strstart(map, "all", &map)) { > + res[nentries - 1].flags |= XATTR_MAP_FLAG_ALL; > + } else { > + fuse_log(FUSE_LOG_ERR, > + "%s: Unexpected scope;" > + " Expecting 'client', 'server', or 'all', in rule %zu\n", > + __func__, nentries); > + exit(1); > + } > + > + > + if (*map != sep) { > + fuse_log(FUSE_LOG_ERR, > + "%s: Expecting '%c' found '%c'" > + " after scope in rule %zu\n", > + __func__, sep, *map, nentries + 1); > + exit(1); > + } > + /* Skip the separator, now at the start of the 'type' */ > + map++; > + > + /* Start of 'type' */ > + if (strstart(map, "prefix", &map)) { > + res[nentries - 1].flags |= XATTR_MAP_FLAG_PREFIX; > + } else if (strstart(map, "ok", &map)) { > + res[nentries - 1].flags |= XATTR_MAP_FLAG_END_OK; > + } else if (strstart(map, "bad", &map)) { > + res[nentries - 1].flags |= XATTR_MAP_FLAG_END_BAD; > + } else { > + fuse_log(FUSE_LOG_ERR, > + "%s: Unexpected type;" > + "Expecting 'prefix', 'ok', or 'bad' in rule %zu\n", > + __func__, nentries); > + exit(1); > + } > + > + if (*map++ != sep) { > + fuse_log(FUSE_LOG_ERR, > + "%s: Missing '%c' at end of type field of rule %zu\n", > + __func__, sep, nentries); > + exit(1); > + } > + > + /* At start of 'key' field */ > + tmp = strchr(map, sep); > + if (!tmp) { > + fuse_log(FUSE_LOG_ERR, > + "%s: Missing '%c' at end of key field of rule %zu", > + __func__, sep, nentries); > + exit(1); > + } > + res[nentries - 1].key = g_strndup(map, tmp - map); > + map = tmp + 1; > + > + /* At start of 'prepend' field */ > + tmp = strchr(map, sep); > + if (!tmp) { > + fuse_log(FUSE_LOG_ERR, > + "%s: Missing '%c' at end of prepend field of rule %zu", > + __func__, sep, nentries); > + exit(1); > + } > + res[nentries - 1].prepend = g_strndup(map, tmp - map); > + map = tmp + 1; > + /* End of rule - go around again for another rule */ > + } > + > + if (!nentries) { > + fuse_log(FUSE_LOG_ERR, "Empty xattr map\n"); > + exit(1); > + } > + > + /* Add a terminator to error in cases the user hasn't specified */ > + res = g_realloc_n(res, ++nentries, sizeof(XattrMapEntry)); > + res[nentries - 1].flags = XATTR_MAP_FLAG_ALL | XATTR_MAP_FLAG_END_BAD; > + res[nentries - 1].key = g_strdup(""); > + res[nentries - 1].prepend = g_strdup(""); > + > + return res; > +} > + > static void lo_getxattr(fuse_req_t req, fuse_ino_t ino, const char *name, > size_t size) > { > @@ -2909,6 +3052,11 @@ int main(int argc, char *argv[]) > } else { > lo.source = strdup("/"); > } > + > + if (lo.xattrmap) { > + xattr_map_list = parse_xattrmap(lo.xattrmap); > + } > + > if (!lo.timeout_set) { > switch (lo.cache) { > case CACHE_NONE: > -- > 2.26.2 > > _______________________________________________ > Virtio-fs mailing list > Virtio-fs@redhat.com > https://www.redhat.com/mailman/listinfo/virtio-fs
* Vivek Goyal (vgoyal@redhat.com) wrote: > On Thu, Aug 27, 2020 at 04:36:54PM +0100, Dr. David Alan Gilbert (git) wrote: > > From: "Dr. David Alan Gilbert" <dgilbert@redhat.com> > > > > Add an option to define mappings of xattr names so that > > the client and server filesystems see different views. > > This can be used to have different SELinux mappings as > > seen by the guest, to run the virtiofsd with less privileges > > (e.g. in a case where it can't set trusted/system/security > > xattrs but you want the guest to be able to), or to isolate > > multiple users of the same name; e.g. trusted attributes > > used by stacking overlayfs. > > > > A mapping engine is used wit 3 simple rules; the rules can > > be combined to allow most useful mapping scenarios. > > The ruleset is defined by -o xattrmap='rules...'. > > > > This patch doesn't use the rule maps yet. > > > > Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com> > > --- > > docs/tools/virtiofsd.rst | 55 ++++++++++++ > > tools/virtiofsd/passthrough_ll.c | 148 +++++++++++++++++++++++++++++++ > > 2 files changed, 203 insertions(+) > > > > diff --git a/docs/tools/virtiofsd.rst b/docs/tools/virtiofsd.rst > > index 824e713491..2efa16d3c5 100644 > > --- a/docs/tools/virtiofsd.rst > > +++ b/docs/tools/virtiofsd.rst > > @@ -107,6 +107,60 @@ Options > > performance. ``auto`` acts similar to NFS with a 1 second metadata cache > > timeout. ``always`` sets a long cache lifetime at the expense of coherency. > > > > +xattr-mapping > > +------------- > > + > > +By default the name of xattr's used by the client are passed through to the server > > +file system. This can be a problem where either those xattr names are used > > +by something on the server (e.g. selinux client/server confusion) or if the > > +virtiofsd is running in a container with restricted priviliges where it cannot > > +access some attributes. > > + > > +A mapping of xattr names can be made using -o xattrmap=mapping where the ``mapping`` > > +string consists of a series of rules. > > + > > +The first matching rule terminates the mapping. > > + > > +Each rule consists of a number of fields separated with a separator that is the > > +first non-white space character in the rule. This separator must then be used > > +for the whole rule. > > +White space may be added before and after each rule. > > +Using ':' as the separator a rule is of the form: > > + > > +``:scope:type:key:prepend:`` > > Hi David, > > This seems very genric and which makes it harder to understand and > harder to write rules. I am wondering do we really need this degree > of flexibility. Is it worth, dropping some of the requirements > and simplify the syntax. I'm wondering perhaps if we could solve this by adding sugared simple versions but leaving the flexible syntax for those who need it. > - I am wonderig why do we need to allow choice of separator. I didnd't have that at first, but it was simple to add and solves the problem of if you have the separator in the string you want to substitute. > - Wondering why do we need to allow separate rules for client/server. > Once we start remapping something, is it not good enough that > mapping be bidirectonal. > > - Not sure why separate notion of "bad". To me once we decide to > remap something, should automatically block unprefixed version. I wanted to be able to block things rather than remap; for example just to block 'trusted.' > IOW, what functionality we will lose if we just say > > -o remap_xattr="trusted.". > > This implies following. > > A. If client is sending any xattr prefixed with "trusted.", prefix it > with "user.virtiofs". > > B. Server filters out anything starting with "trusted." > > C. If server sees "user.virtiofs.trusted." it strips "user.virtiofs". Don't forget you also have to stop the client explicitly sending 'user.virtiofs.trusted'; that would let an unpriv client process overwrite the prefixed name. > For remapping security.selinux, user could specify. > > -o remap_xattr="security.selinux." > > For nested configuration. virtiofsd at L1 will specify. > > -o remap_xattr="security.selinux.". > > And virtiofsd at L0 can specify. > > -o remap_xattr="user.virtiofs.security.selinux." I think you're saying that means it needs to know if it's L0 or L1 which is a shame; ideally you'd be able to have something that transparently worked at either. In your scheme how do I do both the 'trusted.' and 'security.selinux.' stuff? > I doubt we need to care about being able to remap xattrs of > other filesystems like virtio-9p. Well that's the thing; there's at least 9p and crosvm's setup; both of which are different, and it would make sense if someone wanted to transition their existing on disk container to a virtiofs setup from a 9p setup without having to change all their xattr's. That was my main reason for wanting the flexibility. > I also have some questions about how this will be used. > > Overlay > ------- > - So for non nested guests, we can have two instances of overlay. Lets > call these ovl0 and ovl1. (ovl0 being on host, and ovl1 being inside > guest). Fs hierarcy might look as follows. > > ext4-->ovl0-->virtiofsd0-->ovl1 > > This case does not work by default even if virtiofsd has CAP_SYS_ADMIN > by default as overlay does not allow nesting. So when ovl1 tries to > set trusted.overlay, ovl0 will deny it. > > We could simple pass extra directory from host which does not go through > overlay on host and use that as upper inside guest. > > ext4-->ovl0-->virtiofsd0-->ovl1 > ext4-->ovl0/upper-->virtiofsd0-->ovl1 > (/upper used as upper directory of ovl1) If I understand correctly that does mean that the L1 has to understand it's an L1 and do things differently. > I guess remapping "trusted.overlay" will allow us not to have a separate > ovl0/upper. And following itself will work. Have you tested it? Does > this work. Not tried, but that is my hope. > Basically we are creating nested overlay configuration with > virtiofs in between. Is "trusted.overlay" only conflict. I wonder > there might be others. Just that "trusted.overlay" is first failure > we noticed. I think there's a whole bunch of trusted.overlay.* stuff but I didn't find anything else (I think jt's a define as the prefix). Note also, that if someone has an existing fuse-overlayfs setup that nested by using user.fuseoverlayfs you might be able to use the rule system to map it back. > > Nested Overlay > -------------- > - For now I will assume that we are using separate upper dir. > > ext4-->ovl0-->virtiofsd0-->ovl1-->virtiofsd1-->ovl2 > ext4-->ovl0/upper1-->virtiofsd0-->ovl1(uses upper1 as upperdir) > ext4-->ovl0/upper2-->virtiofsd0-->ovl1-->virtiofsd1-->ovl2 (users upper2 > as upper dir) > > Basically create two directories upper1 and upper2 on regular filesystem > say ext4/xfs. Bind mount them on ovl0/upper1 and ovl0/upper2 respectively. > And now ovl1 uses ovl0/upper1 as upperdir and ovl2 uses ovl0/upper2 as > upperdir. This should make sure ovl0, ovl1 and ovl2 are not nested from > sharing upper perspective. > > Now virtiofsd1 will run with '-o remap_xattr="trusted.overlay"' and > virtiofsd0 will run with '-o remap_xattr="user.virtiofs.trusted.overlay"' You could tell both layers the same thing; prefix/strip everything with user.virtiofs. and then you can do the same thing at both layers and they don't need to know which layer they're at. Dave > Just trying to wrap my head around how our use cases will use this new > remapping xattr thing. > > Thanks > Vivek > > > > > + > > +**scope** is: > > + > > +- 'client' - match 'key' against a xattr name from the client for > > + setxattr/getxattr/removexattr > > +- 'server' - match 'prepend' against a xattr name from the server > > + for listxattr > > +- 'all' - can be used to match both cases. > > + > > +**type** is one of: > > + > > +- 'prefix' - If 'key' matches the client then the 'prepend' > > + is added before the name is passed to the server. > > + For a server case, the prepend is tested and stripped > > + if matching. > > + > > +- 'ok' - The attribute name is OK and passed through to > > + the server unchanged. > > + > > +- 'bad' - If a client tries to use this name it's > > + denied using EPERM; when the server passes an attribute > > + name matching it's hidden. > > + > > +**key** is a string tested as a prefix on an attribute name originating > > +on the client. It maybe empty in which case a 'client' rule > > +will always match on client names. > > + > > +**prepend** is a string tested as a prefix on an attribute name originiating > > +on the server, and used as a new prefix. It maybe empty > > +in which case a 'server' rule will always match on all names from > > +the server. > > + > > + > > Examples > > -------- > > > > @@ -123,3 +177,4 @@ Export ``/var/lib/fs/vm001/`` on vhost-user UNIX domain socket > > -numa node,memdev=mem \ > > ... > > guest# mount -t virtiofs myfs /mnt > > + > > diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passthrough_ll.c > > index 083d17a960..00e96a10cd 100644 > > --- a/tools/virtiofsd/passthrough_ll.c > > +++ b/tools/virtiofsd/passthrough_ll.c > > @@ -64,6 +64,7 @@ > > #include <syslog.h> > > #include <unistd.h> > > > > +#include "qemu/cutils.h" > > #include "passthrough_helpers.h" > > #include "passthrough_seccomp.h" > > > > @@ -144,6 +145,7 @@ struct lo_data { > > int flock; > > int posix_lock; > > int xattr; > > + char *xattrmap; > > char *source; > > char *modcaps; > > double timeout; > > @@ -171,6 +173,7 @@ static const struct fuse_opt lo_opts[] = { > > { "no_posix_lock", offsetof(struct lo_data, posix_lock), 0 }, > > { "xattr", offsetof(struct lo_data, xattr), 1 }, > > { "no_xattr", offsetof(struct lo_data, xattr), 0 }, > > + { "xattrmap=%s", offsetof(struct lo_data, xattrmap), 0 }, > > { "modcaps=%s", offsetof(struct lo_data, modcaps), 0 }, > > { "timeout=%lf", offsetof(struct lo_data, timeout), 0 }, > > { "timeout=", offsetof(struct lo_data, timeout_set), 1 }, > > @@ -2003,6 +2006,146 @@ static void lo_flock(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi, > > fuse_reply_err(req, res == -1 ? errno : 0); > > } > > > > +typedef struct xattr_map_entry { > > + const char *key; > > + const char *prepend; > > + unsigned int flags; > > +} XattrMapEntry; > > + > > +/* > > + * Exit; process attribute unmodified if matched. > > + * An empty key applies to all. > > + */ > > +#define XATTR_MAP_FLAG_END_OK (1 << 0) > > +/* > > + * The attribute is unwanted; > > + * EPERM on write hidden on read. > > + */ > > +#define XATTR_MAP_FLAG_END_BAD (1 << 1) > > +/* > > + * For attr that start with 'key' prepend 'prepend' > > + * 'key' maybe empty to prepend for all attrs > > + * key is defined from set/remove point of view. > > + * Automatically reversed on read > > + */ > > +#define XATTR_MAP_FLAG_PREFIX (1 << 2) > > +/* Apply rule to get/set/remove */ > > +#define XATTR_MAP_FLAG_CLIENT (1 << 16) > > +/* Apply rule to list */ > > +#define XATTR_MAP_FLAG_SERVER (1 << 17) > > +/* Apply rule to all */ > > +#define XATTR_MAP_FLAG_ALL (XATTR_MAP_FLAG_SERVER | XATTR_MAP_FLAG_CLIENT) > > + > > +static XattrMapEntry *xattr_map_list; > > + > > +static XattrMapEntry *parse_xattrmap(const char *map) > > +{ > > + XattrMapEntry *res = NULL; > > + size_t nentries = 0; > > + const char *tmp; > > + > > + while (*map) { > > + char sep; > > + > > + if (isspace(*map)) { > > + map++; > > + continue; > > + } > > + /* The separator is the first non-space of the rule */ > > + sep = *map++; > > + if (!sep) { > > + break; > > + } > > + > > + /* Allocate some space for the rule */ > > + res = g_realloc_n(res, ++nentries, sizeof(XattrMapEntry)); > > + res[nentries - 1].flags = 0; > > + > > + if (strstart(map, "client", &map)) { > > + res[nentries - 1].flags |= XATTR_MAP_FLAG_CLIENT; > > + } else if (strstart(map, "server", &map)) { > > + res[nentries - 1].flags |= XATTR_MAP_FLAG_SERVER; > > + } else if (strstart(map, "all", &map)) { > > + res[nentries - 1].flags |= XATTR_MAP_FLAG_ALL; > > + } else { > > + fuse_log(FUSE_LOG_ERR, > > + "%s: Unexpected scope;" > > + " Expecting 'client', 'server', or 'all', in rule %zu\n", > > + __func__, nentries); > > + exit(1); > > + } > > + > > + > > + if (*map != sep) { > > + fuse_log(FUSE_LOG_ERR, > > + "%s: Expecting '%c' found '%c'" > > + " after scope in rule %zu\n", > > + __func__, sep, *map, nentries + 1); > > + exit(1); > > + } > > + /* Skip the separator, now at the start of the 'type' */ > > + map++; > > + > > + /* Start of 'type' */ > > + if (strstart(map, "prefix", &map)) { > > + res[nentries - 1].flags |= XATTR_MAP_FLAG_PREFIX; > > + } else if (strstart(map, "ok", &map)) { > > + res[nentries - 1].flags |= XATTR_MAP_FLAG_END_OK; > > + } else if (strstart(map, "bad", &map)) { > > + res[nentries - 1].flags |= XATTR_MAP_FLAG_END_BAD; > > + } else { > > + fuse_log(FUSE_LOG_ERR, > > + "%s: Unexpected type;" > > + "Expecting 'prefix', 'ok', or 'bad' in rule %zu\n", > > + __func__, nentries); > > + exit(1); > > + } > > + > > + if (*map++ != sep) { > > + fuse_log(FUSE_LOG_ERR, > > + "%s: Missing '%c' at end of type field of rule %zu\n", > > + __func__, sep, nentries); > > + exit(1); > > + } > > + > > + /* At start of 'key' field */ > > + tmp = strchr(map, sep); > > + if (!tmp) { > > + fuse_log(FUSE_LOG_ERR, > > + "%s: Missing '%c' at end of key field of rule %zu", > > + __func__, sep, nentries); > > + exit(1); > > + } > > + res[nentries - 1].key = g_strndup(map, tmp - map); > > + map = tmp + 1; > > + > > + /* At start of 'prepend' field */ > > + tmp = strchr(map, sep); > > + if (!tmp) { > > + fuse_log(FUSE_LOG_ERR, > > + "%s: Missing '%c' at end of prepend field of rule %zu", > > + __func__, sep, nentries); > > + exit(1); > > + } > > + res[nentries - 1].prepend = g_strndup(map, tmp - map); > > + map = tmp + 1; > > + /* End of rule - go around again for another rule */ > > + } > > + > > + if (!nentries) { > > + fuse_log(FUSE_LOG_ERR, "Empty xattr map\n"); > > + exit(1); > > + } > > + > > + /* Add a terminator to error in cases the user hasn't specified */ > > + res = g_realloc_n(res, ++nentries, sizeof(XattrMapEntry)); > > + res[nentries - 1].flags = XATTR_MAP_FLAG_ALL | XATTR_MAP_FLAG_END_BAD; > > + res[nentries - 1].key = g_strdup(""); > > + res[nentries - 1].prepend = g_strdup(""); > > + > > + return res; > > +} > > + > > static void lo_getxattr(fuse_req_t req, fuse_ino_t ino, const char *name, > > size_t size) > > { > > @@ -2909,6 +3052,11 @@ int main(int argc, char *argv[]) > > } else { > > lo.source = strdup("/"); > > } > > + > > + if (lo.xattrmap) { > > + xattr_map_list = parse_xattrmap(lo.xattrmap); > > + } > > + > > if (!lo.timeout_set) { > > switch (lo.cache) { > > case CACHE_NONE: > > -- > > 2.26.2 > > > > _______________________________________________ > > Virtio-fs mailing list > > Virtio-fs@redhat.com > > https://www.redhat.com/mailman/listinfo/virtio-fs
On 2020-08-27 at 17:36 CEST, Dr. David Alan Gilbert (git) wrote... > From: "Dr. David Alan Gilbert" <dgilbert@redhat.com> > > Add an option to define mappings of xattr names so that > the client and server filesystems see different views. > This can be used to have different SELinux mappings as > seen by the guest, to run the virtiofsd with less privileges > (e.g. in a case where it can't set trusted/system/security > xattrs but you want the guest to be able to), or to isolate > multiple users of the same name; e.g. trusted attributes > used by stacking overlayfs. > > A mapping engine is used wit 3 simple rules; the rules can > be combined to allow most useful mapping scenarios. > The ruleset is defined by -o xattrmap='rules...'. > > This patch doesn't use the rule maps yet. > > Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com> > --- > docs/tools/virtiofsd.rst | 55 ++++++++++++ > tools/virtiofsd/passthrough_ll.c | 148 +++++++++++++++++++++++++++++++ > 2 files changed, 203 insertions(+) > > diff --git a/docs/tools/virtiofsd.rst b/docs/tools/virtiofsd.rst > index 824e713491..2efa16d3c5 100644 > --- a/docs/tools/virtiofsd.rst > +++ b/docs/tools/virtiofsd.rst > @@ -107,6 +107,60 @@ Options > performance. ``auto`` acts similar to NFS with a 1 second metadata cache > timeout. ``always`` sets a long cache lifetime at the expense of coherency. > > +xattr-mapping > +------------- > + > +By default the name of xattr's used by the client are passed through to the server > +file system. This can be a problem where either those xattr names are used > +by something on the server (e.g. selinux client/server confusion) or if the > +virtiofsd is running in a container with restricted priviliges where it cannot > +access some attributes. > + > +A mapping of xattr names can be made using -o xattrmap=mapping where the ``mapping`` > +string consists of a series of rules. > + > +The first matching rule terminates the mapping. > + > +Each rule consists of a number of fields separated with a separator that is the > +first non-white space character in the rule. This separator must then be used > +for the whole rule. > +White space may be added before and after each rule. > +Using ':' as the separator a rule is of the form: > + > +``:scope:type:key:prepend:`` > + > +**scope** is: > + > +- 'client' - match 'key' against a xattr name from the client for > + setxattr/getxattr/removexattr > +- 'server' - match 'prepend' against a xattr name from the server > + for listxattr > +- 'all' - can be used to match both cases. > + > +**type** is one of: > + > +- 'prefix' - If 'key' matches the client then the 'prepend' > + is added before the name is passed to the server. > + For a server case, the prepend is tested and stripped > + if matching. > + > +- 'ok' - The attribute name is OK and passed through to > + the server unchanged. > + > +- 'bad' - If a client tries to use this name it's > + denied using EPERM; when the server passes an attribute > + name matching it's hidden. > + > +**key** is a string tested as a prefix on an attribute name originating > +on the client. It maybe empty in which case a 'client' rule > +will always match on client names. > + > +**prepend** is a string tested as a prefix on an attribute name originiating > +on the server, and used as a new prefix. It maybe empty > +in which case a 'server' rule will always match on all names from > +the server. > + > + > Examples > -------- > > @@ -123,3 +177,4 @@ Export ``/var/lib/fs/vm001/`` on vhost-user UNIX domain socket > -numa node,memdev=mem \ > ... > guest# mount -t virtiofs myfs /mnt > + > diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passthrough_ll.c > index 083d17a960..00e96a10cd 100644 > --- a/tools/virtiofsd/passthrough_ll.c > +++ b/tools/virtiofsd/passthrough_ll.c > @@ -64,6 +64,7 @@ > #include <syslog.h> > #include <unistd.h> > > +#include "qemu/cutils.h" > #include "passthrough_helpers.h" > #include "passthrough_seccomp.h" > > @@ -144,6 +145,7 @@ struct lo_data { > int flock; > int posix_lock; > int xattr; > + char *xattrmap; Who owns that field? Should it be cleaned up in fuse_lo_data_cleanup() just like source is? > char *source; > char *modcaps; > double timeout; > @@ -171,6 +173,7 @@ static const struct fuse_opt lo_opts[] = { > { "no_posix_lock", offsetof(struct lo_data, posix_lock), 0 }, > { "xattr", offsetof(struct lo_data, xattr), 1 }, > { "no_xattr", offsetof(struct lo_data, xattr), 0 }, > + { "xattrmap=%s", offsetof(struct lo_data, xattrmap), 0 }, > { "modcaps=%s", offsetof(struct lo_data, modcaps), 0 }, > { "timeout=%lf", offsetof(struct lo_data, timeout), 0 }, > { "timeout=", offsetof(struct lo_data, timeout_set), 1 }, > @@ -2003,6 +2006,146 @@ static void lo_flock(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi, > fuse_reply_err(req, res == -1 ? errno : 0); > } > > +typedef struct xattr_map_entry { > + const char *key; > + const char *prepend; > + unsigned int flags; > +} XattrMapEntry; > + > +/* > + * Exit; process attribute unmodified if matched. > + * An empty key applies to all. > + */ > +#define XATTR_MAP_FLAG_END_OK (1 << 0) > +/* > + * The attribute is unwanted; > + * EPERM on write hidden on read. > + */ > +#define XATTR_MAP_FLAG_END_BAD (1 << 1) > +/* > + * For attr that start with 'key' prepend 'prepend' > + * 'key' maybe empty to prepend for all attrs > + * key is defined from set/remove point of view. > + * Automatically reversed on read > + */ > +#define XATTR_MAP_FLAG_PREFIX (1 << 2) > +/* Apply rule to get/set/remove */ > +#define XATTR_MAP_FLAG_CLIENT (1 << 16) > +/* Apply rule to list */ > +#define XATTR_MAP_FLAG_SERVER (1 << 17) > +/* Apply rule to all */ > +#define XATTR_MAP_FLAG_ALL (XATTR_MAP_FLAG_SERVER | XATTR_MAP_FLAG_CLIENT) > + > +static XattrMapEntry *xattr_map_list; Curious why you made it a static variable and not a field in struct lo_data? > + > +static XattrMapEntry *parse_xattrmap(const char *map) > +{ > + XattrMapEntry *res = NULL; > + size_t nentries = 0; > + const char *tmp; > + > + while (*map) { > + char sep; > + > + if (isspace(*map)) { > + map++; > + continue; > + } > + /* The separator is the first non-space of the rule */ > + sep = *map++; > + if (!sep) { > + break; > + } > + > + /* Allocate some space for the rule */ > + res = g_realloc_n(res, ++nentries, sizeof(XattrMapEntry)); > + res[nentries - 1].flags = 0; I would probably create an `entry` pointer to `res[nentries - 1]` since there are 9 uses for it. > + > + if (strstart(map, "client", &map)) { > + res[nentries - 1].flags |= XATTR_MAP_FLAG_CLIENT; > + } else if (strstart(map, "server", &map)) { > + res[nentries - 1].flags |= XATTR_MAP_FLAG_SERVER; > + } else if (strstart(map, "all", &map)) { > + res[nentries - 1].flags |= XATTR_MAP_FLAG_ALL; > + } else { > + fuse_log(FUSE_LOG_ERR, > + "%s: Unexpected scope;" > + " Expecting 'client', 'server', or 'all', in rule %zu\n", > + __func__, nentries); > + exit(1); > + } > + > + > + if (*map != sep) { > + fuse_log(FUSE_LOG_ERR, > + "%s: Expecting '%c' found '%c'" > + " after scope in rule %zu\n", > + __func__, sep, *map, nentries + 1); I think it should be `nentries` here like in the others > + exit(1); > + } > + /* Skip the separator, now at the start of the 'type' */ > + map++; > + > + /* Start of 'type' */ > + if (strstart(map, "prefix", &map)) { > + res[nentries - 1].flags |= XATTR_MAP_FLAG_PREFIX; > + } else if (strstart(map, "ok", &map)) { > + res[nentries - 1].flags |= XATTR_MAP_FLAG_END_OK; > + } else if (strstart(map, "bad", &map)) { > + res[nentries - 1].flags |= XATTR_MAP_FLAG_END_BAD; > + } else { > + fuse_log(FUSE_LOG_ERR, > + "%s: Unexpected type;" > + "Expecting 'prefix', 'ok', or 'bad' in rule %zu\n", > + __func__, nentries); > + exit(1); > + } > + > + if (*map++ != sep) { > + fuse_log(FUSE_LOG_ERR, > + "%s: Missing '%c' at end of type field of rule %zu\n", > + __func__, sep, nentries); > + exit(1); > + } > + > + /* At start of 'key' field */ > + tmp = strchr(map, sep); > + if (!tmp) { > + fuse_log(FUSE_LOG_ERR, > + "%s: Missing '%c' at end of key field of rule %zu", > + __func__, sep, nentries); > + exit(1); > + } > + res[nentries - 1].key = g_strndup(map, tmp - map); > + map = tmp + 1; > + > + /* At start of 'prepend' field */ > + tmp = strchr(map, sep); > + if (!tmp) { > + fuse_log(FUSE_LOG_ERR, > + "%s: Missing '%c' at end of prepend field of rule %zu", > + __func__, sep, nentries); > + exit(1); > + } > + res[nentries - 1].prepend = g_strndup(map, tmp - map); > + map = tmp + 1; > + /* End of rule - go around again for another rule */ > + } > + > + if (!nentries) { > + fuse_log(FUSE_LOG_ERR, "Empty xattr map\n"); > + exit(1); > + } > + > + /* Add a terminator to error in cases the user hasn't specified */ > + res = g_realloc_n(res, ++nentries, sizeof(XattrMapEntry)); > + res[nentries - 1].flags = XATTR_MAP_FLAG_ALL | XATTR_MAP_FLAG_END_BAD; > + res[nentries - 1].key = g_strdup(""); > + res[nentries - 1].prepend = g_strdup(""); > + > + return res; > +} > + > static void lo_getxattr(fuse_req_t req, fuse_ino_t ino, const char *name, > size_t size) > { > @@ -2909,6 +3052,11 @@ int main(int argc, char *argv[]) > } else { > lo.source = strdup("/"); > } > + > + if (lo.xattrmap) { > + xattr_map_list = parse_xattrmap(lo.xattrmap); This is never freed. If you put the static in struct lo_data, you could naturally clean it up in fuse_lo_data_cleanup. > + } > + > if (!lo.timeout_set) { > switch (lo.cache) { > case CACHE_NONE: -- Cheers, Christophe de Dinechin (IRC c3d)
* Christophe de Dinechin (dinechin@redhat.com) wrote: > > On 2020-08-27 at 17:36 CEST, Dr. David Alan Gilbert (git) wrote... > > From: "Dr. David Alan Gilbert" <dgilbert@redhat.com> > > > > Add an option to define mappings of xattr names so that > > the client and server filesystems see different views. > > This can be used to have different SELinux mappings as > > seen by the guest, to run the virtiofsd with less privileges > > (e.g. in a case where it can't set trusted/system/security > > xattrs but you want the guest to be able to), or to isolate > > multiple users of the same name; e.g. trusted attributes > > used by stacking overlayfs. > > > > A mapping engine is used wit 3 simple rules; the rules can > > be combined to allow most useful mapping scenarios. > > The ruleset is defined by -o xattrmap='rules...'. > > > > This patch doesn't use the rule maps yet. > > > > Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com> > > --- > > docs/tools/virtiofsd.rst | 55 ++++++++++++ > > tools/virtiofsd/passthrough_ll.c | 148 +++++++++++++++++++++++++++++++ > > 2 files changed, 203 insertions(+) > > > > diff --git a/docs/tools/virtiofsd.rst b/docs/tools/virtiofsd.rst > > index 824e713491..2efa16d3c5 100644 > > --- a/docs/tools/virtiofsd.rst > > +++ b/docs/tools/virtiofsd.rst > > @@ -107,6 +107,60 @@ Options > > performance. ``auto`` acts similar to NFS with a 1 second metadata cache > > timeout. ``always`` sets a long cache lifetime at the expense of coherency. > > > > +xattr-mapping > > +------------- > > + > > +By default the name of xattr's used by the client are passed through to the server > > +file system. This can be a problem where either those xattr names are used > > +by something on the server (e.g. selinux client/server confusion) or if the > > +virtiofsd is running in a container with restricted priviliges where it cannot > > +access some attributes. > > + > > +A mapping of xattr names can be made using -o xattrmap=mapping where the ``mapping`` > > +string consists of a series of rules. > > + > > +The first matching rule terminates the mapping. > > + > > +Each rule consists of a number of fields separated with a separator that is the > > +first non-white space character in the rule. This separator must then be used > > +for the whole rule. > > +White space may be added before and after each rule. > > +Using ':' as the separator a rule is of the form: > > + > > +``:scope:type:key:prepend:`` > > + > > +**scope** is: > > + > > +- 'client' - match 'key' against a xattr name from the client for > > + setxattr/getxattr/removexattr > > +- 'server' - match 'prepend' against a xattr name from the server > > + for listxattr > > +- 'all' - can be used to match both cases. > > + > > +**type** is one of: > > + > > +- 'prefix' - If 'key' matches the client then the 'prepend' > > + is added before the name is passed to the server. > > + For a server case, the prepend is tested and stripped > > + if matching. > > + > > +- 'ok' - The attribute name is OK and passed through to > > + the server unchanged. > > + > > +- 'bad' - If a client tries to use this name it's > > + denied using EPERM; when the server passes an attribute > > + name matching it's hidden. > > + > > +**key** is a string tested as a prefix on an attribute name originating > > +on the client. It maybe empty in which case a 'client' rule > > +will always match on client names. > > + > > +**prepend** is a string tested as a prefix on an attribute name originiating > > +on the server, and used as a new prefix. It maybe empty > > +in which case a 'server' rule will always match on all names from > > +the server. > > + > > + > > Examples > > -------- > > > > @@ -123,3 +177,4 @@ Export ``/var/lib/fs/vm001/`` on vhost-user UNIX domain socket > > -numa node,memdev=mem \ > > ... > > guest# mount -t virtiofs myfs /mnt > > + > > diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passthrough_ll.c > > index 083d17a960..00e96a10cd 100644 > > --- a/tools/virtiofsd/passthrough_ll.c > > +++ b/tools/virtiofsd/passthrough_ll.c > > @@ -64,6 +64,7 @@ > > #include <syslog.h> > > #include <unistd.h> > > > > +#include "qemu/cutils.h" > > #include "passthrough_helpers.h" > > #include "passthrough_seccomp.h" > > > > @@ -144,6 +145,7 @@ struct lo_data { > > int flock; > > int posix_lock; > > int xattr; > > + char *xattrmap; > > Who owns that field? Should it be cleaned up in fuse_lo_data_cleanup() just like > source is? Done. > > char *source; > > char *modcaps; > > double timeout; > > @@ -171,6 +173,7 @@ static const struct fuse_opt lo_opts[] = { > > { "no_posix_lock", offsetof(struct lo_data, posix_lock), 0 }, > > { "xattr", offsetof(struct lo_data, xattr), 1 }, > > { "no_xattr", offsetof(struct lo_data, xattr), 0 }, > > + { "xattrmap=%s", offsetof(struct lo_data, xattrmap), 0 }, > > { "modcaps=%s", offsetof(struct lo_data, modcaps), 0 }, > > { "timeout=%lf", offsetof(struct lo_data, timeout), 0 }, > > { "timeout=", offsetof(struct lo_data, timeout_set), 1 }, > > @@ -2003,6 +2006,146 @@ static void lo_flock(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi, > > fuse_reply_err(req, res == -1 ? errno : 0); > > } > > > > +typedef struct xattr_map_entry { > > + const char *key; > > + const char *prepend; > > + unsigned int flags; > > +} XattrMapEntry; > > + > > +/* > > + * Exit; process attribute unmodified if matched. > > + * An empty key applies to all. > > + */ > > +#define XATTR_MAP_FLAG_END_OK (1 << 0) > > +/* > > + * The attribute is unwanted; > > + * EPERM on write hidden on read. > > + */ > > +#define XATTR_MAP_FLAG_END_BAD (1 << 1) > > +/* > > + * For attr that start with 'key' prepend 'prepend' > > + * 'key' maybe empty to prepend for all attrs > > + * key is defined from set/remove point of view. > > + * Automatically reversed on read > > + */ > > +#define XATTR_MAP_FLAG_PREFIX (1 << 2) > > +/* Apply rule to get/set/remove */ > > +#define XATTR_MAP_FLAG_CLIENT (1 << 16) > > +/* Apply rule to list */ > > +#define XATTR_MAP_FLAG_SERVER (1 << 17) > > +/* Apply rule to all */ > > +#define XATTR_MAP_FLAG_ALL (XATTR_MAP_FLAG_SERVER | XATTR_MAP_FLAG_CLIENT) > > + > > +static XattrMapEntry *xattr_map_list; > > Curious why you made it a static variable and not a field in struct lo_data? Done. > > + > > +static XattrMapEntry *parse_xattrmap(const char *map) > > +{ > > + XattrMapEntry *res = NULL; > > + size_t nentries = 0; > > + const char *tmp; > > + > > + while (*map) { > > + char sep; > > + > > + if (isspace(*map)) { > > + map++; > > + continue; > > + } > > + /* The separator is the first non-space of the rule */ > > + sep = *map++; > > + if (!sep) { > > + break; > > + } > > + > > + /* Allocate some space for the rule */ > > + res = g_realloc_n(res, ++nentries, sizeof(XattrMapEntry)); > > + res[nentries - 1].flags = 0; > > I would probably create an `entry` pointer to `res[nentries - 1]` > since there are 9 uses for it. I've reworked that whole bit; we've now got a temporary and a function that adds an entry. > > + > > + if (strstart(map, "client", &map)) { > > + res[nentries - 1].flags |= XATTR_MAP_FLAG_CLIENT; > > + } else if (strstart(map, "server", &map)) { > > + res[nentries - 1].flags |= XATTR_MAP_FLAG_SERVER; > > + } else if (strstart(map, "all", &map)) { > > + res[nentries - 1].flags |= XATTR_MAP_FLAG_ALL; > > + } else { > > + fuse_log(FUSE_LOG_ERR, > > + "%s: Unexpected scope;" > > + " Expecting 'client', 'server', or 'all', in rule %zu\n", > > + __func__, nentries); > > + exit(1); > > + } > > + > > + > > + if (*map != sep) { > > + fuse_log(FUSE_LOG_ERR, > > + "%s: Expecting '%c' found '%c'" > > + " after scope in rule %zu\n", > > + __func__, sep, *map, nentries + 1); > > I think it should be `nentries` here like in the others Done. > > + exit(1); > > + } > > + /* Skip the separator, now at the start of the 'type' */ > > + map++; > > + > > + /* Start of 'type' */ > > + if (strstart(map, "prefix", &map)) { > > + res[nentries - 1].flags |= XATTR_MAP_FLAG_PREFIX; > > + } else if (strstart(map, "ok", &map)) { > > + res[nentries - 1].flags |= XATTR_MAP_FLAG_END_OK; > > + } else if (strstart(map, "bad", &map)) { > > + res[nentries - 1].flags |= XATTR_MAP_FLAG_END_BAD; > > + } else { > > + fuse_log(FUSE_LOG_ERR, > > + "%s: Unexpected type;" > > + "Expecting 'prefix', 'ok', or 'bad' in rule %zu\n", > > + __func__, nentries); > > + exit(1); > > + } > > + > > + if (*map++ != sep) { > > + fuse_log(FUSE_LOG_ERR, > > + "%s: Missing '%c' at end of type field of rule %zu\n", > > + __func__, sep, nentries); > > + exit(1); > > + } > > + > > + /* At start of 'key' field */ > > + tmp = strchr(map, sep); > > + if (!tmp) { > > + fuse_log(FUSE_LOG_ERR, > > + "%s: Missing '%c' at end of key field of rule %zu", > > + __func__, sep, nentries); > > + exit(1); > > + } > > + res[nentries - 1].key = g_strndup(map, tmp - map); > > + map = tmp + 1; > > + > > + /* At start of 'prepend' field */ > > + tmp = strchr(map, sep); > > + if (!tmp) { > > + fuse_log(FUSE_LOG_ERR, > > + "%s: Missing '%c' at end of prepend field of rule %zu", > > + __func__, sep, nentries); > > + exit(1); > > + } > > + res[nentries - 1].prepend = g_strndup(map, tmp - map); > > + map = tmp + 1; > > + /* End of rule - go around again for another rule */ > > + } > > + > > + if (!nentries) { > > + fuse_log(FUSE_LOG_ERR, "Empty xattr map\n"); > > + exit(1); > > + } > > + > > + /* Add a terminator to error in cases the user hasn't specified */ > > + res = g_realloc_n(res, ++nentries, sizeof(XattrMapEntry)); > > + res[nentries - 1].flags = XATTR_MAP_FLAG_ALL | XATTR_MAP_FLAG_END_BAD; > > + res[nentries - 1].key = g_strdup(""); > > + res[nentries - 1].prepend = g_strdup(""); > > + > > + return res; > > +} > > + > > static void lo_getxattr(fuse_req_t req, fuse_ino_t ino, const char *name, > > size_t size) > > { > > @@ -2909,6 +3052,11 @@ int main(int argc, char *argv[]) > > } else { > > lo.source = strdup("/"); > > } > > + > > + if (lo.xattrmap) { > > + xattr_map_list = parse_xattrmap(lo.xattrmap); > > This is never freed. If you put the static in struct lo_data, you could > naturally clean it up in fuse_lo_data_cleanup. Cleanup added. Dave > > + } > > + > > if (!lo.timeout_set) { > > switch (lo.cache) { > > case CACHE_NONE: > > > -- > Cheers, > Christophe de Dinechin (IRC c3d)
On Fri, Sep 18, 2020 at 06:38:38PM +0100, Dr. David Alan Gilbert wrote: > * Vivek Goyal (vgoyal@redhat.com) wrote: > > On Thu, Aug 27, 2020 at 04:36:54PM +0100, Dr. David Alan Gilbert (git) wrote: > > > From: "Dr. David Alan Gilbert" <dgilbert@redhat.com> > > > > > > Add an option to define mappings of xattr names so that > > > the client and server filesystems see different views. > > > This can be used to have different SELinux mappings as > > > seen by the guest, to run the virtiofsd with less privileges > > > (e.g. in a case where it can't set trusted/system/security > > > xattrs but you want the guest to be able to), or to isolate > > > multiple users of the same name; e.g. trusted attributes > > > used by stacking overlayfs. > > > > > > A mapping engine is used wit 3 simple rules; the rules can > > > be combined to allow most useful mapping scenarios. > > > The ruleset is defined by -o xattrmap='rules...'. > > > > > > This patch doesn't use the rule maps yet. > > > > > > Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com> > > > --- > > > docs/tools/virtiofsd.rst | 55 ++++++++++++ > > > tools/virtiofsd/passthrough_ll.c | 148 +++++++++++++++++++++++++++++++ > > > 2 files changed, 203 insertions(+) > > > > > > diff --git a/docs/tools/virtiofsd.rst b/docs/tools/virtiofsd.rst > > > index 824e713491..2efa16d3c5 100644 > > > --- a/docs/tools/virtiofsd.rst > > > +++ b/docs/tools/virtiofsd.rst > > > @@ -107,6 +107,60 @@ Options > > > performance. ``auto`` acts similar to NFS with a 1 second metadata cache > > > timeout. ``always`` sets a long cache lifetime at the expense of coherency. > > > > > > +xattr-mapping > > > +------------- > > > + > > > +By default the name of xattr's used by the client are passed through to the server > > > +file system. This can be a problem where either those xattr names are used > > > +by something on the server (e.g. selinux client/server confusion) or if the > > > +virtiofsd is running in a container with restricted priviliges where it cannot > > > +access some attributes. > > > + > > > +A mapping of xattr names can be made using -o xattrmap=mapping where the ``mapping`` > > > +string consists of a series of rules. > > > + > > > +The first matching rule terminates the mapping. > > > + > > > +Each rule consists of a number of fields separated with a separator that is the > > > +first non-white space character in the rule. This separator must then be used > > > +for the whole rule. > > > +White space may be added before and after each rule. > > > +Using ':' as the separator a rule is of the form: > > > + > > > +``:scope:type:key:prepend:`` > > > > Hi David, > > > > This seems very genric and which makes it harder to understand and > > harder to write rules. I am wondering do we really need this degree > > of flexibility. Is it worth, dropping some of the requirements > > and simplify the syntax. > > I'm wondering perhaps if we could solve this by adding sugared simple > versions but leaving the flexible syntax for those who need it. I guess fair enough. This syntax is so generic (hence complex), that its not my first choice. But if others feel the need of such a generic mechanism, I am not going to get in the way. > > > - I am wonderig why do we need to allow choice of separator. > > I didnd't have that at first, but it was simple to add and solves > the problem of if you have the separator in the string you want to > substitute. > > > - Wondering why do we need to allow separate rules for client/server. > > Once we start remapping something, is it not good enough that > > mapping be bidirectonal. > > > > - Not sure why separate notion of "bad". To me once we decide to > > remap something, should automatically block unprefixed version. > > I wanted to be able to block things rather than remap; for example > just to block 'trusted.' > > > IOW, what functionality we will lose if we just say > > > > -o remap_xattr="trusted.". > > > > This implies following. > > > > A. If client is sending any xattr prefixed with "trusted.", prefix it > > with "user.virtiofs". > > > > B. Server filters out anything starting with "trusted." > > > > C. If server sees "user.virtiofs.trusted." it strips "user.virtiofs". > [ I had missed reading this email. Looking at it now. ] > Don't forget you also have to stop the client explicitly sending > 'user.virtiofs.trusted'; that would let an unpriv client process > overwrite the prefixed name. Fair enough. Yes, we will have to block this as part of "remap_xattr" semantics. > > > For remapping security.selinux, user could specify. > > > > -o remap_xattr="security.selinux." > > > > For nested configuration. virtiofsd at L1 will specify. > > > > -o remap_xattr="security.selinux.". > > > > And virtiofsd at L0 can specify. > > > > -o remap_xattr="user.virtiofs.security.selinux." > > I think you're saying that means it needs to know if it's L0 or L1 > which is a shame; ideally you'd be able to have something that > transparently worked at either. I will be nice to avoid knowing level information because it is ugly. But how to avoid it? Either don't use virtiofs in stacking configurations. But if we have to, how to avoid it? > > In your scheme how do I do both the 'trusted.' and 'security.selinux.' > stuff? We can allow specifying multiple "remap_xattr" or allow multiple rules in single option separted by ":". Say -o remap_xattr="security.selinux.:trusted." > > > I doubt we need to care about being able to remap xattrs of > > other filesystems like virtio-9p. > > Well that's the thing; there's at least 9p and crosvm's setup; both > of which are different, and it would make sense if someone wanted > to transition their existing on disk container to a virtiofs setup > from a 9p setup without having to change all their xattr's. > That was my main reason for wanting the flexibility. > > > I also have some questions about how this will be used. > > > > Overlay > > ------- > > - So for non nested guests, we can have two instances of overlay. Lets > > call these ovl0 and ovl1. (ovl0 being on host, and ovl1 being inside > > guest). Fs hierarcy might look as follows. > > > > ext4-->ovl0-->virtiofsd0-->ovl1 > > > > This case does not work by default even if virtiofsd has CAP_SYS_ADMIN > > by default as overlay does not allow nesting. So when ovl1 tries to > > set trusted.overlay, ovl0 will deny it. > > > > We could simple pass extra directory from host which does not go through > > overlay on host and use that as upper inside guest. > > > > ext4-->ovl0-->virtiofsd0-->ovl1 > > ext4-->ovl0/upper-->virtiofsd0-->ovl1 > > (/upper used as upper directory of ovl1) > > If I understand correctly that does mean that the L1 has to understand > it's an L1 and do things differently. Yes. I could not find a way to avoid it. > > > I guess remapping "trusted.overlay" will allow us not to have a separate > > ovl0/upper. And following itself will work. Have you tested it? Does > > this work. > > Not tried, but that is my hope. > > > Basically we are creating nested overlay configuration with > > virtiofs in between. Is "trusted.overlay" only conflict. I wonder > > there might be others. Just that "trusted.overlay" is first failure > > we noticed. > > I think there's a whole bunch of trusted.overlay.* stuff but I didn't > find anything else (I think jt's a define as the prefix). > Note also, that if someone has an existing fuse-overlayfs setup that > nested by using user.fuseoverlayfs you might be able to use the rule > system to map it back. > > > > > Nested Overlay > > -------------- > > - For now I will assume that we are using separate upper dir. > > > > ext4-->ovl0-->virtiofsd0-->ovl1-->virtiofsd1-->ovl2 > > ext4-->ovl0/upper1-->virtiofsd0-->ovl1(uses upper1 as upperdir) > > ext4-->ovl0/upper2-->virtiofsd0-->ovl1-->virtiofsd1-->ovl2 (users upper2 > > as upper dir) > > > > Basically create two directories upper1 and upper2 on regular filesystem > > say ext4/xfs. Bind mount them on ovl0/upper1 and ovl0/upper2 respectively. > > And now ovl1 uses ovl0/upper1 as upperdir and ovl2 uses ovl0/upper2 as > > upperdir. This should make sure ovl0, ovl1 and ovl2 are not nested from > > sharing upper perspective. > > > > Now virtiofsd1 will run with '-o remap_xattr="trusted.overlay"' and > > virtiofsd0 will run with '-o remap_xattr="user.virtiofs.trusted.overlay"' > > You could tell both layers the same thing; prefix/strip everything with > user.virtiofs. and then you can do the same thing at both layers and > they don't need to know which layer they're at. Right that should work. But this will prefix user.virtiofs for every xattr. If we want to prefix it on a specific xattr, then it become little more tricky. Because inner most layer will do setxattr(trusted.foo.) and next layer will do setxattr(user.virtiofs.trusted.foo). So question is what's the common syntax which works for both. I have been thinking little more of nested overlay use case. If we make sure that "upper" for each overlay instance is separate (and not coming from stacked overlay), then following should work well for nested configurations. To reiterate, I think something like this should work for nested overlay configuration. - On host, create lower0, upper0, upper1 and upper2 dirs. mkdir -p lower0 lower0/lower1 work0 upper1 work1 upper2 work2 ovl0 - Create ovl0 on host (L0) mount -t overlay -o lowerdir=lower0,upperdir=upper0,workdir=work0 none ovl0 - Bind mount upper and work dirs for ovl1 and ovl2. mkdir -p ovl0/upper1 ovl0/work1 ovl0/upper2 ovl0/work2 mount --bind upper1 ovl0/upper1 mount --bind work1 ovl0/work1 mount --bind upper2 ovl0/upper2 mount --bind work2 ovl0/work2 - Run virtiofsd0 on ovl0 with "-o remap_xattr="trusted.overlay." - Inside L1 guest say ovl0 is mounted at "/". Create second overlay. mkdir ovl1 mount -t overlay -o lowerdir=lower1,upperdir=upper1,workdir=work1 none ovl1 - Bind mount upper2 and work2 inside ovl1. mkdir -p ovl1/upper2 ovl1/work2 mount --bind upper2 ovl1/upper2 mount --bind work2 ovl1/work2 - Run virtiofsd1 on ovl1 with -o "remap_xattr="trusted.overlay." - Inisde L2 guest say virtiofs is mounted at "/" mkdir ovl2 mount -t overlay -o lowerdir=lower1,upperdir=upper2,workdir=work2 none ovl2 Now upper for overlayfs are not stacked through overlayfs file systems. Thanks Vivek
diff --git a/docs/tools/virtiofsd.rst b/docs/tools/virtiofsd.rst index 824e713491..2efa16d3c5 100644 --- a/docs/tools/virtiofsd.rst +++ b/docs/tools/virtiofsd.rst @@ -107,6 +107,60 @@ Options performance. ``auto`` acts similar to NFS with a 1 second metadata cache timeout. ``always`` sets a long cache lifetime at the expense of coherency. +xattr-mapping +------------- + +By default the name of xattr's used by the client are passed through to the server +file system. This can be a problem where either those xattr names are used +by something on the server (e.g. selinux client/server confusion) or if the +virtiofsd is running in a container with restricted priviliges where it cannot +access some attributes. + +A mapping of xattr names can be made using -o xattrmap=mapping where the ``mapping`` +string consists of a series of rules. + +The first matching rule terminates the mapping. + +Each rule consists of a number of fields separated with a separator that is the +first non-white space character in the rule. This separator must then be used +for the whole rule. +White space may be added before and after each rule. +Using ':' as the separator a rule is of the form: + +``:scope:type:key:prepend:`` + +**scope** is: + +- 'client' - match 'key' against a xattr name from the client for + setxattr/getxattr/removexattr +- 'server' - match 'prepend' against a xattr name from the server + for listxattr +- 'all' - can be used to match both cases. + +**type** is one of: + +- 'prefix' - If 'key' matches the client then the 'prepend' + is added before the name is passed to the server. + For a server case, the prepend is tested and stripped + if matching. + +- 'ok' - The attribute name is OK and passed through to + the server unchanged. + +- 'bad' - If a client tries to use this name it's + denied using EPERM; when the server passes an attribute + name matching it's hidden. + +**key** is a string tested as a prefix on an attribute name originating +on the client. It maybe empty in which case a 'client' rule +will always match on client names. + +**prepend** is a string tested as a prefix on an attribute name originiating +on the server, and used as a new prefix. It maybe empty +in which case a 'server' rule will always match on all names from +the server. + + Examples -------- @@ -123,3 +177,4 @@ Export ``/var/lib/fs/vm001/`` on vhost-user UNIX domain socket -numa node,memdev=mem \ ... guest# mount -t virtiofs myfs /mnt + diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passthrough_ll.c index 083d17a960..00e96a10cd 100644 --- a/tools/virtiofsd/passthrough_ll.c +++ b/tools/virtiofsd/passthrough_ll.c @@ -64,6 +64,7 @@ #include <syslog.h> #include <unistd.h> +#include "qemu/cutils.h" #include "passthrough_helpers.h" #include "passthrough_seccomp.h" @@ -144,6 +145,7 @@ struct lo_data { int flock; int posix_lock; int xattr; + char *xattrmap; char *source; char *modcaps; double timeout; @@ -171,6 +173,7 @@ static const struct fuse_opt lo_opts[] = { { "no_posix_lock", offsetof(struct lo_data, posix_lock), 0 }, { "xattr", offsetof(struct lo_data, xattr), 1 }, { "no_xattr", offsetof(struct lo_data, xattr), 0 }, + { "xattrmap=%s", offsetof(struct lo_data, xattrmap), 0 }, { "modcaps=%s", offsetof(struct lo_data, modcaps), 0 }, { "timeout=%lf", offsetof(struct lo_data, timeout), 0 }, { "timeout=", offsetof(struct lo_data, timeout_set), 1 }, @@ -2003,6 +2006,146 @@ static void lo_flock(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi, fuse_reply_err(req, res == -1 ? errno : 0); } +typedef struct xattr_map_entry { + const char *key; + const char *prepend; + unsigned int flags; +} XattrMapEntry; + +/* + * Exit; process attribute unmodified if matched. + * An empty key applies to all. + */ +#define XATTR_MAP_FLAG_END_OK (1 << 0) +/* + * The attribute is unwanted; + * EPERM on write hidden on read. + */ +#define XATTR_MAP_FLAG_END_BAD (1 << 1) +/* + * For attr that start with 'key' prepend 'prepend' + * 'key' maybe empty to prepend for all attrs + * key is defined from set/remove point of view. + * Automatically reversed on read + */ +#define XATTR_MAP_FLAG_PREFIX (1 << 2) +/* Apply rule to get/set/remove */ +#define XATTR_MAP_FLAG_CLIENT (1 << 16) +/* Apply rule to list */ +#define XATTR_MAP_FLAG_SERVER (1 << 17) +/* Apply rule to all */ +#define XATTR_MAP_FLAG_ALL (XATTR_MAP_FLAG_SERVER | XATTR_MAP_FLAG_CLIENT) + +static XattrMapEntry *xattr_map_list; + +static XattrMapEntry *parse_xattrmap(const char *map) +{ + XattrMapEntry *res = NULL; + size_t nentries = 0; + const char *tmp; + + while (*map) { + char sep; + + if (isspace(*map)) { + map++; + continue; + } + /* The separator is the first non-space of the rule */ + sep = *map++; + if (!sep) { + break; + } + + /* Allocate some space for the rule */ + res = g_realloc_n(res, ++nentries, sizeof(XattrMapEntry)); + res[nentries - 1].flags = 0; + + if (strstart(map, "client", &map)) { + res[nentries - 1].flags |= XATTR_MAP_FLAG_CLIENT; + } else if (strstart(map, "server", &map)) { + res[nentries - 1].flags |= XATTR_MAP_FLAG_SERVER; + } else if (strstart(map, "all", &map)) { + res[nentries - 1].flags |= XATTR_MAP_FLAG_ALL; + } else { + fuse_log(FUSE_LOG_ERR, + "%s: Unexpected scope;" + " Expecting 'client', 'server', or 'all', in rule %zu\n", + __func__, nentries); + exit(1); + } + + + if (*map != sep) { + fuse_log(FUSE_LOG_ERR, + "%s: Expecting '%c' found '%c'" + " after scope in rule %zu\n", + __func__, sep, *map, nentries + 1); + exit(1); + } + /* Skip the separator, now at the start of the 'type' */ + map++; + + /* Start of 'type' */ + if (strstart(map, "prefix", &map)) { + res[nentries - 1].flags |= XATTR_MAP_FLAG_PREFIX; + } else if (strstart(map, "ok", &map)) { + res[nentries - 1].flags |= XATTR_MAP_FLAG_END_OK; + } else if (strstart(map, "bad", &map)) { + res[nentries - 1].flags |= XATTR_MAP_FLAG_END_BAD; + } else { + fuse_log(FUSE_LOG_ERR, + "%s: Unexpected type;" + "Expecting 'prefix', 'ok', or 'bad' in rule %zu\n", + __func__, nentries); + exit(1); + } + + if (*map++ != sep) { + fuse_log(FUSE_LOG_ERR, + "%s: Missing '%c' at end of type field of rule %zu\n", + __func__, sep, nentries); + exit(1); + } + + /* At start of 'key' field */ + tmp = strchr(map, sep); + if (!tmp) { + fuse_log(FUSE_LOG_ERR, + "%s: Missing '%c' at end of key field of rule %zu", + __func__, sep, nentries); + exit(1); + } + res[nentries - 1].key = g_strndup(map, tmp - map); + map = tmp + 1; + + /* At start of 'prepend' field */ + tmp = strchr(map, sep); + if (!tmp) { + fuse_log(FUSE_LOG_ERR, + "%s: Missing '%c' at end of prepend field of rule %zu", + __func__, sep, nentries); + exit(1); + } + res[nentries - 1].prepend = g_strndup(map, tmp - map); + map = tmp + 1; + /* End of rule - go around again for another rule */ + } + + if (!nentries) { + fuse_log(FUSE_LOG_ERR, "Empty xattr map\n"); + exit(1); + } + + /* Add a terminator to error in cases the user hasn't specified */ + res = g_realloc_n(res, ++nentries, sizeof(XattrMapEntry)); + res[nentries - 1].flags = XATTR_MAP_FLAG_ALL | XATTR_MAP_FLAG_END_BAD; + res[nentries - 1].key = g_strdup(""); + res[nentries - 1].prepend = g_strdup(""); + + return res; +} + static void lo_getxattr(fuse_req_t req, fuse_ino_t ino, const char *name, size_t size) { @@ -2909,6 +3052,11 @@ int main(int argc, char *argv[]) } else { lo.source = strdup("/"); } + + if (lo.xattrmap) { + xattr_map_list = parse_xattrmap(lo.xattrmap); + } + if (!lo.timeout_set) { switch (lo.cache) { case CACHE_NONE: