mountd: clean up rmtab handling
Subsystems (2)
cover mountd: clean up rmtab handling
We have had a few reports from customers that showmount -a is just plain broken. The manpage says that it should "List both the client hostname and mounted directory in host:dir format". Instead, when used with recent kernels, the output can look something like this (after mounts from several hosts):
*:/foo
*:/scratch
*,foo.bar.baz:/scratch
10.10.1.41:*
172.16.33.192:*
172.16.33.192:*,foo.bar.baz
The output here varies somewhat depending on how the export table on the server is configured, but the key is that we're getting multiple lines here for every mount (some that don't even contain a path), and the hostname field is often not a hostname or IP address.
The problem seems to be in how the rmtab is being managed and in how mountd handles the hostname of the client when checking permissions and such.
The following two patches remedy this. They both should only affect the case where new_cache is set.
-- Jeff
-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
Quoted 2 lines · Jeff Layton
We have had a few reports from customers that showmount -a is just plain broken.
...
Quoted 4 lines
The following two patches remedy this. They both should only affect the case where
new_cacheis set.-- Jeff
It's been a little over a month since I originally posted these patches and haven't received much comment though Neil and I have traded some emails off list concerning it. I'd like to move the discussion back to the mailing list to get a wider audience and see if we can come to some sort of resolution.
When we last communicated, he mentioned a scheme whereby we could generate the showmount -a info on the fly rather than trying to maintain the rmtab. That sounds like a good approach, but I don't see how that can work without some changes in how the caches are managed.
Neil, would it be reasonable to commit these patches now to fix the immediate problem and then work on the new scheme that synthesizes the info out of the kernel caches like you proposed?
-- Jeff
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
Quoted 17 lines · On Friday December 1, [email protected] wrote:
Quoted 2 lines · Jeff Layton
We have had a few reports from customers that showmount -a is just plain broken.
...
Quoted 4 lines
The following two patches remedy this. They both should only affect the case where
new_cacheis set.-- Jeff
It's been a little over a month since I originally posted these patches and haven't received much comment though Neil and I have traded some emails off list concerning it. I'd like to move the discussion back to the mailing list to get a wider audience and see if we can come to some sort of resolution.
Quoted 8 lines
When we last communicated, he mentioned a scheme whereby we could generate the showmount -a info on the fly rather than trying to maintain the rmtab. That sounds like a good approach, but I don't see how that can work without some changes in how the caches are managed.
Neil, would it be reasonable to commit these patches now to fix the immediate problem and then work on the new scheme that synthesizes the info out of the kernel caches like you proposed?
I'm not convinced... I don't like the idea of mapping an IP address to a hostname and then just working with the hostname. Because it is really IP addresses that you trust, not host names (in the case of multi-homed hosts particularly).
The following patch is a first cut at using the kernel caches for providing the information for "showmount -a".
Note that this does *not* list precisely those hosts which currently have filesystems mounted, and precisely which filesystems. That information is simply not available reliably.
What it does do is list all the
IP-address exportpoint
combinations that could possibly have been used in the last half hour.
As an example of how this will be less precise than you might like, suppose you export
/foo @hostgroup()
/bar @hostgroup()
and suppose 1.2.3.4 and 5.6.7.8 both map to hosts in @hostgroup.
Support further than 1.2.3.4 has mounted and accessed /foo recently, and 5.6.7.8 has mounted and accessed /bar recently.
Then the kernel will know:
1.2.3.4 -> @hostgroup
5.6.7.8 -> @hostgroup
@hostgroup can access /foo
@hostgroup can access /bar
Then this patch will cause "showmount -a" to report
1.2.3.4:/foo
1.2.3.4:/bar
5.6.7.8:/foo
5.6.7.8:/bar
@hostgroup:/foo
@hostgroup:/bar
It would be possible (thought I would rather make it optional) to give the kernel more fine detail. i.e. not group multiple IP addresses under the one client name but give each IP address it's own name. This you would have
1.2.3.4 -> 1.2.3.4
5.6.7.8 -> 5.6.7.8
1.2.3.4 can access /foo
5.6.7.8 can access /barand "showmount -a" would then list
1.2.3.4:/foo
5.6.7.8:/barbut again this would only be if there were accesses in the last half hour (though that could be configurable too I guess).
Finally we could still leave the authentication the way it is and just fix up the rmtab entries.
So whenever mountd gives info to the kernel, whether due to a mount request or an up-call, we write an appropriate IP:/dir to rmtab, and whenever an unmount request is received, we deleted the corresponding request.
We could do this ignoring the current count in rmtab, and still have a fairly reliable system. We would probably want to erase rmtab at boot time to stop it from collection cruft.
Actually, that sounds like it could be the best idea. I'll have a go at that, but I'll post this patch (That gets info from the in-kernel cache) anyway.
NeilBrown
support/nfs/cacheio.c +3 -3
@@ -140,7 +140,7 @@ int qword_get(char **bpp, char *dest, int bufsize) } }else {/* text with \nnn octal quoting */ -while (*bp !=' ' && *bp !='\n' && *bp && len < bufsize-1 ) { +while (*bp !=' ' && *bp !='\t' && *bp !='\n' && *bp && len < bufsize-1 ) { if (*bp == '\\' && isodigit(bp[1]) && (bp[1] <= '3') && isodigit(bp[2]) &&
@@ -158,9 +158,9 @@ int qword_get(char **bpp, char *dest, int bufsize) } } -if (*bp !=' ' && *bp !='\n' && *bp !='\0' ) +if (*bp !=' ' && *bp !='\t' && *bp !='\n' && *bp !='\0' )return -1 ; -while (*bp ==' ' ) bp++; +while (*bp ==' ' || *bp =='\t' ) bp++; *bpp = bp; *dest ='\0' ;return len;
utils/mountd/cache.c +81 -8
116 lines not on this page · raw message
utils/mountd/mountd.c +5 -1
@@ -198,7 +198,11 @@ mount_dump_1_svc(struct svc_req *rqstp, void *argp, mountlist *res)struct sockaddr_in *addr = (struct sockaddr_in *)svc_getcaller (rqstp->rq_xprt); -if ((*res =mountlist_list ()) ==NULL ) +if (new_cache) + *res =cache_mountlist_list (); +else + *res =mountlist_list (); +if (*res ==NULL )xlog (L_WARNING ,"dump request from %s failed." ,inet_ntoa (addr->sin_addr));
utils/mountd/mountd.h +1 -0
@@ -49,6 +49,7 @@ void mountlist_add(char *host, const char *path);void mountlist_del (char *host,const char *path);void mountlist_del_all (struct sockaddr_in *sin);mountlist mountlist_list (void ); +mountlist cache_mountlist_list (void );#endif /* MOUNTD_H */
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
Quoted 11 lines · On Monday December 4, [email protected] wrote:
Finally we could still leave the authentication the way it is and just fix up the rmtab entries.
So whenever mountd gives info to the kernel, whether due to a mount request or an up-call, we write an appropriate IP:/dir to rmtab, and whenever an unmount request is received, we deleted the corresponding request.
We could do this ignoring the current count in rmtab, and still have a fairly reliable system. We would probably want to erase rmtab at boot time to stop it from collection cruft.
Quoted 3 lines
Actually, that sounds like it could be the best idea. I'll have a go at that, but I'll post this patch (That gets info from the in-kernel cache) anyway.
I couldn't quite do that the way I had hoped as the info isn't available at the right places.
However here is another patch for consideration.
It tries a bit harder to keep vaguely sensible information in rmtab and if you run mounted with "-I", then "showmount -a" will list IP and export-point of all current mounts ... as best as possible. e.g. if you mount the same filesystem twice on the one client, then it will appear once in "showmount -a". If you then unmount one of them, it will disappear. But as soon as you access it again it will re-appear.
I think this is about the best I can do.
Does it meet the needs?
NeilBrown
support/export/client.c +23 -0
@@ -241,6 +241,7 @@ client_find(struct hostent *hp)* */ static char *add_name (char *old,char *add); +extern int use_ip_address;char *client_compose (struct in_addr addr)
@@ -249,6 +250,9 @@ client_compose(struct in_addr addr)char *name =NULL ;int i; +if (use_ip_address) +return strdup (inet_ntoa (addr)); +if (clientlist[MCL_WILDCARD ] || clientlist[MCL_NETGROUP ]) he =get_reliable_hostbyaddr ((const char *)&addr,sizeof (addr),AF_INET );if (he ==NULL )
@@ -284,6 +288,25 @@ client_member(char *client, char *name)return 0 ; } +int client_matches (char *client,nfs_client *clp) +{ +struct hostent host; +struct in_addr *alist[2 ]; +struct in_addr a; +if (! use_ip_address) +return client_member (client, clp->m_hostname); + + host.h_addrtype =AF_INET ; + host.h_length =4 ; + host.h_name = client; + host.h_addr_list = (char *)alist; + alist[0 ] = &a; + alist[1 ] =NULL ; +inet_aton (client, &a); + +return client_check (clp, &host); +} +int name_cmp (char *a,char *b)
support/include/exportfs.h +1 -0
@@ -57,6 +57,7 @@ void client_release(nfs_client *);void client_freeall (void );char *client_compose (struct in_addr addr);int client_member (char *client,char *name); +int client_matches (char *client,nfs_client *clp);int export_read (char *fname);void export_add (nfs_export *);
utils/exportfs/exportfs.c +1 -0
@@ -33,6 +33,7 @@ static void dump(int verbose);static void error (nfs_export *exp,int err);static void usage (void ); +int use_ip_address;/* Should not get used... */ int main (int argc,char **argv)
utils/mountd/auth.c +1 -1
@@ -96,7 +96,7 @@ auth_authenticate_internal(char *what, struct sockaddr_in *caller, exp =NULL ;for (i =0 ; !exp && i <MCL_MAXTYPES ; i++)for (exp = exportlist[i]; exp; exp = exp->m_next) { -if (!client_member (my_client.m_hostname, exp->m_client->m_hostname)) +if (!client_matches (my_client.m_hostname, exp->m_client))continue ;if (strcmp (path, exp->m_export.e_path))continue ;
utils/mountd/cache.c +19 -8
@@ -83,9 +83,6 @@ void auth_unix_ip(FILE *f)qword_print (f, *client?client:"DEFAULT" );qword_eol (f); -if (client &&strcmp (ipaddr, client)) -mountlist_add (ipaddr, *client?client:"DEFAULT" ); -if (client)free (client); }
@@ -174,7 +171,7 @@ void nfsd_fh(FILE *f) for (exp = exportlist[i]; exp; exp = exp->m_next) { struct stat stb; - if (!client_member(dom, exp->m_client->m_hostname)) + if (!client_matches(dom, exp->m_client)) continue; if (exp->m_export.e_mountpoint && !is_mountpoint(exp->m_export.e_mountpoint[0]?
@@ -273,7 +270,7 @@ void nfsd_export(FILE *f)/* now find flags for this export point in this domain */ for (i=0 ; i <MCL_MAXTYPES ; i++) {for (exp = exportlist[i]; exp; exp = exp->m_next) { -if (!client_member (dom, exp->m_client->m_hostname)) +if (!client_matches (dom, exp->m_client))continue ;if (strcmp (path, exp->m_export.e_path))continue ;
@@ -377,6 +374,23 @@ void cache_export_ent(char *domain, struct exportent *exp)mountlist_add (domain, exp->e_path); } +void cache_export_flush (char *domain,struct exportent *exp) +{ + +FILE *f =fopen ("/proc/net/rpc/nfsd.export/channel" ,"w" ); +if (!f) +return ; + +qword_print (f, domain); +qword_print (f, exp->e_path); +qword_printint (f,time (0 )-60 ); +qword_eol (f); + +fclose (f); + +mountlist_del (domain, exp->e_path,1 ); +} +void cache_export (nfs_export *exp) {FILE *f;
@@ -393,9 +407,6 @@ void cache_export(nfs_export *exp)fclose (f); -if (strcmp (inet_ntoa (exp->m_client->m_addrlist[0 ]), exp->m_client->m_hostname)) -mountlist_add (inet_ntoa (exp->m_client->m_addrlist[0 ]), exp->m_client->m_hostname); -cache_export_ent (exp->m_client->m_hostname, &exp->m_export); }
utils/mountd/mountd.c +10 -5
@@ -30,6 +30,7 @@extern void cache_open (void );extern struct nfs_fh_len *cache_get_filehandle (nfs_export *exp,int len,char *p);extern void cache_export (nfs_export *exp); +extern void cache_export_flush (char *domain,struct exportent *exp);extern void my_svc_run (void );
@@ -38,6 +39,7 @@ static exports get_exportlist(void);static struct nfs_fh_len *get_rootfh (struct svc_req *,dirpath *,mountstat3 *,int v3);int new_cache =0 ; +int use_ip_address =0 ;/* PRC: a high-availability callout program can be specified with -H * When this is done, the program will receive callouts whenever clients
@@ -66,6 +68,7 @@ static struct option longopts[] = { "ha-callout", 1, 0, 'H' }, { "state-directory-path", 1, 0, 's' }, { "num-threads", 1, 0, 't' }, + { "use-ip-addresses", 0, 0, 'I'}, { NULL, 0, 0, 0 } };
@@ -226,11 +229,10 @@ mount_umnt_1_svc(struct svc_req *rqstp, dirpath *argp, void *resp)return 1 ; }if (new_cache) { -if (strcmp (inet_ntoa (exp->m_client->m_addrlist[0 ]), exp->m_client->m_hostname)) -mountlist_del (inet_ntoa (exp->m_client->m_addrlist[0 ]), exp->m_client->m_hostname); -mountlist_del (exp->m_client->m_hostname, p); +cache_export_flush (exp->m_client->m_hostname, + &exp->m_export); }else { -mountlist_del (exp->m_client->m_hostname, p); +mountlist_del (exp->m_client->m_hostname, p,0 );export_reset (exp); }return 1 ;
@@ -558,8 +560,11 @@ main(int argc, char **argv)/* Parse the command line options and arguments. */ opterr =0 ; -while ((c =getopt_long (argc, argv,"o:nFd:f:p:P:hH:N:V:vs:t:" , longopts,NULL )) !=EOF ) +while ((c =getopt_long (argc, argv,"Io:nFd:f:p:P:hH:N:V:vs:t:" , longopts,NULL )) !=EOF )switch (c) { +case 'I' : + use_ip_address =1 ; +break ;case 'o' : descriptors =atoi (optarg);if (descriptors <=0 ) {
utils/mountd/mountd.h +1 -1
@@ -46,7 +46,7 @@ nfs_export * auth_authenticate(char *what, struct sockaddr_in *sin,void auth_export (nfs_export *exp);void mountlist_add (char *host,const char *path); -void mountlist_del (char *host,const char *path); +void mountlist_del (char *host,const char *path,int force);void mountlist_del_all (struct sockaddr_in *sin);mountlist mountlist_list (void );
utils/mountd/rmtab.c +5 -2
@@ -11,6 +11,7 @@#endif #include <sys/types.h> +#include <time.h> #include <sys/stat.h> #include <unistd.h> #include <netinet/in.h>
@@ -89,7 +90,7 @@ mountlist_add(char *host, const char *path) }void -mountlist_del (char *hname,const char *path) +mountlist_del (char *hname,const char *path,int force) {struct rmtabent *rep;FILE *fp;
@@ -115,7 +116,7 @@ mountlist_del(char *hname, const char *path)/* PRC: do the HA callout: */ ha_callout ("unmount" , rep->r_client, rep->r_path, rep->r_count); } -if (!match || rep->r_count) +if (!match || (!force && rep->r_count))fputrmtabent (fp, rep,NULL ); }if (slink_safe_rename (_PATH_RMTABTMP, _PATH_RMTAB) <0 ) {
@@ -200,6 +201,8 @@ mountlist_list(void)xfree (m); } last_mtime = stb.st_mtime; +if (last_mtime ==time (0 )) + last_mtime--;setrmtabent ("r" );while ((rep =getrmtabent (1 ,NULL )) !=NULL ) {
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
Quoted 8 lines · Neil Brown
It tries a bit harder to keep vaguely sensible information in rmtab and if you run mounted with "-I", then "showmount -a" will list IP and export-point of all current mounts ... as best as possible. e.g. if you mount the same filesystem twice on the one client, then it will appear once in "showmount -a". If you then unmount one of them, it will disappear. But as soon as you access it again it will re-appear.
Quoted 4 lines
I think this is about the best I can do.
Does it meet the needs?
I think we need to fix the default case here instead of adding a workaround flag. We should also strive to make it report hostnames rather than ip addresses if possible. I don't see a need for the -I flag here.
How about this patch instead? It combines some of my earlier patches with the one you just proposed. I did some very cursory testing and it seems to work correctly, populating rmtab with hostname:/path instead of ipaddr:/path.
--- nfs-utils-1.0.10/support/export/client.c.rmtab 2006-08-07
02:40:50.000000000 -0400
+++ nfs-utils-1.0.10/support/export/client.c 2006-12-04
20:33:51.000000000 -0500
@@ -234,102 +234,15 @@
return NULL;
}
-/*
- * Find client name given an IP address
- * This is found by gathering all known names that match that IP address,
- * sorting them and joining them with '+'
- *
- */
-static char *add_name(char *old, char *add);
-
-char *
-client_compose(struct in_addr addr)
+struct hostent *
+client_resolve(struct in_addr addr)
{
struct hostent *he = NULL;
- char *name = NULL;
- int i;
- if (clientlist[MCL_WILDCARD] || clientlist[MCL_NETGROUP])
- he = get_reliable_hostbyaddr((const char*)&addr, sizeof(addr), AF_INET);
+ he = get_reliable_hostbyaddr((const char*)&addr, sizeof(addr), AF_INET);
if (he == NULL)
he = get_hostent((const char*)&addr, sizeof(addr), AF_INET);
-
- for (i = 0 ; i < MCL_MAXTYPES; i++) {
- nfs_client *clp;
- for (clp = clientlist[i]; clp ; clp = clp->m_next) {
- if (!client_check(clp, he))
- continue;
- name = add_name(name, clp->m_hostname);
- }
- }
- return name;
-}
-
-int
-client_member(char *client, char *name)
-{
- /* check if "client" (a ',' separated list of names)
- * contains 'name' as a member
- */
- int l = strlen(name);
- while (*client) {
- if (strncmp(client, name, l) == 0 &&
- (client[l] == ',' || client[l] == '\0'))
- return 1;
- client = strchr(client, ',');
- if (client == NULL)
- return 0;
- client++;
- }
- return 0;
-}
-
-
-int
-name_cmp(char *a, char *b)
-{
- /* compare strings a and b, but only upto ',' in a */
- while (*a && *b && *a != ',' && *a == *b)
- a++, b++;
- if (!*b && (!*a || !a == ',') )
- return 0;
- if (!*b) return 1;
- if (!*a || *a == ',') return -1;
- return *a - *b;
-}
-
-static char *
-add_name(char *old, char *add)
-{
- int len = strlen(add)+2;
- char *new;
- char *cp;
- if (old) len += strlen(old);
-
- new = malloc(len);
- if (!new) {
- free(old);
- return NULL;
- }
- cp = old;
- while (cp && *cp && name_cmp(cp, add) < 0) {
- /* step cp forward over a name */
- char *e = strchr(cp, ',');
- if (e)
- cp = e+1;
- else
- cp = cp + strlen(cp);
- }
- strncpy(new, old, cp-old);
- new[cp-old] = 0;
- if (cp != old && !*cp)
- strcat(new, ",");
- strcat(new, add);
- if (cp && *cp) {
- strcat(new, ",");
- strcat(new, cp);
- }
- return new;
+ return he;
}
/*
--- nfs-utils-1.0.10/support/include/exportfs.h.rmtab 2006-08-07
02:40:50.000000000 -0400
+++ nfs-utils-1.0.10/support/include/exportfs.h 2006-12-04
20:38:46.000000000 -0500
@@ -55,8 +55,7 @@
int client_match(nfs_client *, char *hname);
void client_release(nfs_client *);
void client_freeall(void);
-char * client_compose(struct in_addr addr);
-int client_member(char *client, char *name);
+struct hostent * client_resolve(struct in_addr addr);
int export_read(char *fname);
void export_add(nfs_export *);
--- nfs-utils-1.0.10/utils/mountd/auth.c.rmtab 2006-08-07
02:40:50.000000000 -0400
+++ nfs-utils-1.0.10/utils/mountd/auth.c 2006-12-04 20:33:51.000000000 -0500
@@ -76,27 +76,16 @@
if (new_cache) {
int i;
/* return static nfs_export with details filled in */
- if (my_client.m_naddr != 1 ||
- my_client.m_addrlist[0].s_addr != caller->sin_addr.s_addr) {
- /* different client to last time, so do a lookup */
- char *n;
- my_client.m_naddr = 0;
- my_client.m_addrlist[0] = caller->sin_addr;
- n = client_compose(caller->sin_addr);
- *error = unknown_host;
- if (!n)
- return NULL;
- strcpy(my_client.m_hostname, *n?n:"DEFAULT");
- free(n);
- my_client.m_naddr = 1;
- }
-
+ my_client.m_addrlist[0] = caller->sin_addr;
+ my_client.m_naddr = 1;
+ strncpy(my_client.m_hostname, hp->h_name,
+ sizeof(my_client.m_hostname));
my_exp.m_client = &my_client;
exp = NULL;
for (i = 0; !exp && i < MCL_MAXTYPES; i++)
for (exp = exportlist[i]; exp; exp = exp->m_next) {
- if (!client_member(my_client.m_hostname, exp->m_client->m_hostname))
+ if (!client_check(exp->m_client, hp))
continue;
if (strcmp(path, exp->m_export.e_path))
continue;
--- nfs-utils-1.0.10/utils/mountd/cache.c.rmtab 2006-08-07
02:40:50.000000000 -0400
+++ nfs-utils-1.0.10/utils/mountd/cache.c 2006-12-04 20:37:24.000000000
-0500
@@ -53,8 +53,8 @@
char *cp;
char class[20];
char ipaddr[20];
- char *client;
struct in_addr addr;
+ struct hostent *he;
if (readline(fileno(f), &lbuf, &lbuflen) != 1)
return;
@@ -73,21 +73,16 @@
auth_reload();
/* addr is a valid, interesting address, find the domain name... */
- client = client_compose(addr);
+ he = client_resolve(addr);
-
qword_print(f, "nfsd");
qword_print(f, ipaddr);
qword_printint(f, time(0)+30*60);
- if (client)
- qword_print(f, *client?client:"DEFAULT");
+ if (he) {
+ qword_print(f, *he->h_name?he->h_name:"DEFAULT");
+ free(he);
+ }
qword_eol(f);
-
- if (client && strcmp(ipaddr, client))
- mountlist_add(ipaddr, *client?client:"DEFAULT");
-
- if (client) free(client);
-
}
void nfsd_fh(FILE *f)
@@ -109,6 +104,7 @@
nfs_export *exp;
int i;
int dev_missing = 0;
+ struct hostent *hp = NULL;
if (readline(fileno(f), &lbuf, &lbuflen) != 1)
return;
@@ -170,11 +166,12 @@
auth_reload();
/* Now determine export point for this fsid/domain */
+ hp = gethostbyname(dom);
for (i=0 ; i < MCL_MAXTYPES; i++) {
for (exp = exportlist[i]; exp; exp = exp->m_next) {
struct stat stb;
- if (!client_member(dom, exp->m_client->m_hostname))
+ if (!client_check(exp->m_client, hp))
continue;
if (exp->m_export.e_mountpoint &&
!is_mountpoint(exp->m_export.e_mountpoint[0]?
@@ -251,6 +248,7 @@
int i;
char *dom, *path;
nfs_export *exp, *found = NULL;
+ struct hostent *hp;
if (readline(fileno(f), &lbuf, &lbuflen) != 1)
@@ -271,9 +269,10 @@
auth_reload();
/* now find flags for this export point in this domain */
+ hp = gethostbyname(dom);
for (i=0 ; i < MCL_MAXTYPES; i++) {
for (exp = exportlist[i]; exp; exp = exp->m_next) {
- if (!client_member(dom, exp->m_client->m_hostname))
+ if (!client_check(exp->m_client, hp))
continue;
if (strcmp(path, exp->m_export.e_path))
continue;
@@ -377,6 +376,23 @@
mountlist_add(domain, exp->e_path);
}
+void cache_export_flush(char *domain, struct exportent *exp) +{
+
+ FILE *f = fopen("/proc/net/rpc/nfsd.export/channel", "w");
+ if (!f)
+ return;
+
+ qword_print(f, domain);
+ qword_print(f, exp->e_path);
+ qword_printint(f, time(0)-60);
+ qword_eol(f);
+
+ fclose(f);
+
+ mountlist_del(domain, exp->e_path, 1);
+}
+
void cache_export(nfs_export *exp)
{
FILE *f;
@@ -393,9 +409,6 @@
fclose(f);
- if (strcmp(inet_ntoa(exp->m_client->m_addrlist[0]),exp->m_client->m_hostname))
- mountlist_add(inet_ntoa(exp->m_client->m_addrlist[0]),exp->m_client->m_hostname);
-
cache_export_ent(exp->m_client->m_hostname, &exp->m_export);
}
--- nfs-utils-1.0.10/utils/mountd/mountd.c.rmtab 2006-08-07
02:40:50.000000000 -0400
+++ nfs-utils-1.0.10/utils/mountd/mountd.c 2006-12-04 20:33:51.000000000
-0500
@@ -30,6 +30,7 @@
extern void cache_open(void);
extern struct nfs_fh_len *cache_get_filehandle(nfs_export *exp, int
len, char *p);
extern void cache_export(nfs_export *exp);
+extern void cache_export_flush(char *domain, struct exportent *exp);
extern void my_svc_run(void);
@@ -226,11 +227,10 @@
return 1;
}
if (new_cache) {
- if (strcmp(inet_ntoa(exp->m_client->m_addrlist[0]),exp->m_client->m_hostname))
- mountlist_del(inet_ntoa(exp->m_client->m_addrlist[0]),exp->m_client->m_hostname);
- mountlist_del(exp->m_client->m_hostname, p);
+ cache_export_flush(exp->m_client->m_hostname,
+ &exp->m_export);
} else {
- mountlist_del(exp->m_client->m_hostname, p);
+ mountlist_del(exp->m_client->m_hostname, p, 0);
export_reset (exp);
}
return 1;
--- nfs-utils-1.0.10/utils/mountd/mountd.h.rmtab 2006-08-07
02:40:50.000000000 -0400
+++ nfs-utils-1.0.10/utils/mountd/mountd.h 2006-12-04 20:33:51.000000000
-0500
@@ -46,7 +46,7 @@
void auth_export(nfs_export *exp);
void mountlist_add(char *host, const char *path);
-void mountlist_del(char *host, const char *path);
+void mountlist_del(char *host, const char *path, int force);
void mountlist_del_all(struct sockaddr_in *sin);
mountlist mountlist_list(void);
--- nfs-utils-1.0.10/utils/mountd/rmtab.c.rmtab 2006-08-07
02:40:50.000000000 -0400
+++ nfs-utils-1.0.10/utils/mountd/rmtab.c 2006-12-04 20:33:51.000000000
-0500
@@ -11,6 +11,7 @@
#endif
#include <sys/types.h>
+#include <time.h>
#include <sys/stat.h>
#include <unistd.h>
#include <netinet/in.h>
@@ -89,7 +90,7 @@
}
void
-mountlist_del(char *hname, const char *path)
+mountlist_del(char *hname, const char *path, int force)
{
struct rmtabent *rep;
FILE *fp;
@@ -115,7 +116,7 @@
/* PRC: do the HA callout: */ ha_callout("unmount", rep->r_client, rep->r_path, rep->r_count);
}
- if (!match || rep->r_count) + if (!match || (!force && rep->r_count))
fputrmtabent(fp, rep, NULL);
}
if (slink_safe_rename(_PATH_RMTABTMP, _PATH_RMTAB) < 0) {
@@ -200,6 +201,8 @@
xfree(m);
}
last_mtime = stb.st_mtime;
+ if (last_mtime == time(0))
+ last_mtime--;
setrmtabent("r");
while ((rep = getrmtabent(1, NULL)) != NULL) {-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
Quoted 19 lines · Jeff Layton
Quoted 8 lines · Neil Brown
It tries a bit harder to keep vaguely sensible information in rmtab and if you run mounted with "-I", then "showmount -a" will list IP and export-point of all current mounts ... as best as possible. e.g. if you mount the same filesystem twice on the one client, then it will appear once in "showmount -a". If you then unmount one of them, it will disappear. But as soon as you access it again it will re-appear.
Quoted 4 lines
I think this is about the best I can do.
Does it meet the needs?
I think we need to fix the default case here instead of adding a workaround flag. We should also strive to make it report hostnames rather than ip addresses if possible. I don't see a need for the -I flag here.
Quoted 5 lines
How about this patch instead? It combines some of my earlier patches with the one you just proposed. I did some very cursory testing and it seems to work correctly, populating rmtab with hostname:/path instead of ipaddr:/path.
Apologies, my mailer munged that last patch. Here it is again...
support/export/client.c renamed from support/export/client.c.rmtab +4 -91
@@ -234,102 +234,15 @@return NULL ; } -/* -* Find client name given an IP address -* This is found by gathering all known names that match that IP address, -* sorting them and joining them with '+' -* -*/ -static char *add_name (char *old,char *add); - -char * -client_compose (struct in_addr addr) +struct hostent * +client_resolve (struct in_addr addr) {struct hostent *he =NULL ; -char *name =NULL ; -int i; -if (clientlist[MCL_WILDCARD ] || clientlist[MCL_NETGROUP ]) - he =get_reliable_hostbyaddr ((const char *)&addr,sizeof (addr),AF_INET ); + he =get_reliable_hostbyaddr ((const char *)&addr,sizeof (addr),AF_INET );if (he ==NULL ) he =get_hostent ((const char *)&addr,sizeof (addr),AF_INET ); - -for (i =0 ; i <MCL_MAXTYPES ; i++) { -nfs_client *clp; -for (clp = clientlist[i]; clp ; clp = clp->m_next) { -if (!client_check (clp, he)) -continue ; - name =add_name (name, clp->m_hostname); - } - } -return name; -} - -int -client_member (char *client,char *name) -{ -/* check if "client" (a ',' separated list of names) -* contains 'name' as a member -*/ -int l =strlen (name); -while (*client) { -if (strncmp (client, name, l) ==0 && - (client[l] ==',' || client[l] =='\0' )) -return 1 ; - client =strchr (client,',' ); -if (client ==NULL ) -return 0 ; - client++; - } -return 0 ; -} - - -int -name_cmp (char *a,char *b) -{ -/* compare strings a and b, but only upto ',' in a */ -while (*a && *b && *a !=',' && *a == *b) - a++, b++; -if (!*b && (!*a || !a ==',' ) ) -return 0 ; -if (!*b)return 1 ; -if (!*a || *a ==',' )return -1 ; -return *a - *b; -} - -static char * -add_name (char *old,char *add) -{ -int len =strlen (add)+2 ; -char *new; -char *cp; -if (old) len +=strlen (old); - - new =malloc (len); -if (!new) { -free (old); -return NULL ; - } - cp = old; -while (cp && *cp &&name_cmp (cp, add) <0 ) { -/* step cp forward over a name */ -char *e =strchr (cp,',' ); -if (e) - cp = e+1 ; -else - cp = cp +strlen (cp); - } -strncpy (new, old, cp-old); - new[cp-old] =0 ; -if (cp != old && !*cp) -strcat (new,"," ); -strcat (new, add); -if (cp && *cp) { -strcat (new,"," ); -strcat (new, cp); - } -return new; +return he; }/*
support/include/exportfs.h renamed from support/include/exportfs.h.rmtab +1 -2
@@ -55,8 +55,7 @@int client_match (nfs_client *,char *hname);void client_release (nfs_client *);void client_freeall (void ); -char *client_compose (struct in_addr addr); -int client_member (char *client,char *name); +struct hostent *client_resolve (struct in_addr addr);int export_read (char *fname);void export_add (nfs_export *);
utils/mountd/auth.c renamed from utils/mountd/auth.c.rmtab +5 -16
@@ -76,27 +76,16 @@if (new_cache) {int i;/* return static nfs_export with details filled in */ -if (my_client.m_naddr !=1 || - my_client.m_addrlist[0 ].s_addr != caller->sin_addr.s_addr) { -/* different client to last time, so do a lookup */ -char *n; - my_client.m_naddr =0 ; - my_client.m_addrlist[0 ] = caller->sin_addr; - n =client_compose (caller->sin_addr); - *error = unknown_host; -if (!n) -return NULL ; -strcpy (my_client.m_hostname, *n?n:"DEFAULT" ); -free (n); - my_client.m_naddr =1 ; - } - + my_client.m_addrlist[0 ] = caller->sin_addr; + my_client.m_naddr =1 ; +strncpy (my_client.m_hostname, hp->h_name, +sizeof (my_client.m_hostname)); my_exp.m_client = &my_client; exp =NULL ;for (i =0 ; !exp && i <MCL_MAXTYPES ; i++)for (exp = exportlist[i]; exp; exp = exp->m_next) { -if (!client_member (my_client.m_hostname, exp->m_client->m_hostname)) +if (!client_check (exp->m_client, hp))continue ;if (strcmp (path, exp->m_export.e_path))continue ;
utils/mountd/cache.c renamed from utils/mountd/cache.c.rmtab +29 -16
105 lines not on this page · raw message
utils/mountd/mountd.c renamed from utils/mountd/mountd.c.rmtab +4 -4
@@ -30,6 +30,7 @@extern void cache_open (void );extern struct nfs_fh_len *cache_get_filehandle (nfs_export *exp,int len,char *p);extern void cache_export (nfs_export *exp); +extern void cache_export_flush (char *domain,struct exportent *exp);extern void my_svc_run (void );
@@ -226,11 +227,10 @@return 1 ; }if (new_cache) { -if (strcmp (inet_ntoa (exp->m_client->m_addrlist[0 ]), exp->m_client->m_hostname)) -mountlist_del (inet_ntoa (exp->m_client->m_addrlist[0 ]), exp->m_client->m_hostname); -mountlist_del (exp->m_client->m_hostname, p); +cache_export_flush (exp->m_client->m_hostname, + &exp->m_export); }else { -mountlist_del (exp->m_client->m_hostname, p); +mountlist_del (exp->m_client->m_hostname, p,0 );export_reset (exp); }return 1 ;
utils/mountd/mountd.h renamed from utils/mountd/mountd.h.rmtab +1 -1
@@ -46,7 +46,7 @@void auth_export (nfs_export *exp);void mountlist_add (char *host,const char *path); -void mountlist_del (char *host,const char *path); +void mountlist_del (char *host,const char *path,int force);void mountlist_del_all (struct sockaddr_in *sin);mountlist mountlist_list (void );
utils/mountd/rmtab.c renamed from utils/mountd/rmtab.c.rmtab +5 -2
@@ -11,6 +11,7 @@#endif #include <sys/types.h> +#include <time.h> #include <sys/stat.h> #include <unistd.h> #include <netinet/in.h>
@@ -89,7 +90,7 @@ }void -mountlist_del (char *hname,const char *path) +mountlist_del (char *hname,const char *path,int force) {struct rmtabent *rep;FILE *fp;
@@ -115,7 +116,7 @@/* PRC: do the HA callout: */ ha_callout ("unmount" , rep->r_client, rep->r_path, rep->r_count); } -if (!match || rep->r_count) +if (!match || (!force && rep->r_count))fputrmtabent (fp, rep,NULL ); }if (slink_safe_rename (_PATH_RMTABTMP, _PATH_RMTAB) <0 ) {
@@ -200,6 +201,8 @@xfree (m); } last_mtime = stb.st_mtime; +if (last_mtime ==time (0 )) + last_mtime--;setrmtabent ("r" );while ((rep =getrmtabent (1 ,NULL )) !=NULL ) {
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
Quoted 8 lines · Jeff Layton
Quoted 5 lines
How about this patch instead? It combines some of my earlier patches with the one you just proposed. I did some very cursory testing and it seems to work correctly, populating rmtab with hostname:/path instead of ipaddr:/path.
Apologies, my mailer munged that last patch. Here it is again...
Hi Neil. Any thoughts on the patch I sent out on Dec 4th?
Thanks,
Jeff
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
Quoted 6 lines · Neil Brown
I'm not convinced... I don't like the idea of mapping an IP address to a hostname and then just working with the hostname. Because it is really IP addresses that you trust, not host names (in the case of multi-homed hosts particularly).
The patch I posted doesn't do that though. Perhaps I didn't explain it well enough...
All that patch does is get rid of the caching of the hostname list in my_client. Currently, what happens is that we build a comma separated list of "hostnames" and stuff that into my_client.m_hostname. We build this comma-separated list via client_compose, which just calls client_check repeatedly to see what hostnames this address matches.
The patch I posted still uses the exact same decision making process to see if an address matches an nfs_client entry (client_check). It just no longer does this in two stages via the comma-separated list.
The patch I posted should make no change in behavior of whether a host is allowed or denied, aside from the particular case that we already discussed where a cached my_client hostname list is no longer correct.
All that said, I'm OK with pulling this info out of the kernel caches instead. I'll have a look over the latest patch that you sent and see if it does what we need.
Thanks,
Jeff
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
Quoted 17 lines · On Monday December 4, [email protected] wrote:
Quoted 6 lines · Neil Brown
I'm not convinced... I don't like the idea of mapping an IP address to a hostname and then just working with the hostname. Because it is really IP addresses that you trust, not host names (in the case of multi-homed hosts particularly).
The patch I posted doesn't do that though. Perhaps I didn't explain it well enough...
All that patch does is get rid of the caching of the hostname list in
my_client. Currently, what happens is that we build a comma separated list of "hostnames" and stuff that into my_client.m_hostname. We build this comma-separated list viaclient_compose, which just callsclient_checkrepeatedly to see what hostnames this address matches.
Quoted 7 lines
The patch I posted still uses the exact same decision making process to see if an address matches an
nfs_cliententry (client_check). It just no longer does this in two stages via the comma-separated list.The patch I posted should make no change in behavior of whether a host is allowed or denied, aside from the particular case that we already discussed where a cached
my_clienthostname list is no longer correct.
I don't agree that it makes no change.
At one level this is clear from the fact that it loses information. auth_unix_ip (which gets called when the kernel finds an unknown IP address and wants a name for it) always provides a mapping from the IP address to a hostname (As returned by the DNS) - if such a hostname is available. If the host is multi-homed (multiple IP addresses for the one host), this might lose information.
Consider a situation where we have one subnet that is physically very secure, and we export some filesystem (rw,no_root_squash) to that subnet, and another subnet that is physically less secure and we export the filesystem (ro,root_squash) to that subnet.
Suppose further there is some host - foo - with an interface on both subnets.
Now attacker Alice sets up a machine on the less secure subnet and injects packets from foo's IP address (on that subnet). They get routed to the fileserver which maps the IP address to "foo". When you then call client_check, both of the exports could match, and you might end up giving internal access to this external packets. Not good.
So I'm not happy with any patch that removes client_compose and client_member in favour of client_resolve.
To the big-picture issue - you want "showmount -a" to return "the right thing". I don't believe there is a universal "the right thing" as various issues are fairly poorly defined. That is part of why I was think of adding flags to mountd so you could choose your poison.
The two main areas of uncertainty in "showmount" output are:
- how to identify the host - IP address or host name. The former is
more precise in the case of multi-homed hosts. The latter is
possibly more common.
- How current should the information be? rmtab cannot be kept reliably uptodate and tends to grow - there can easily be hosts listed there that do not exist any more. Conversely the kernel cache information is relatively short term, and there could easily be clients that have the filesystem mounted but that are not listed in the cache.
So: what exactly do you - or your customers - want?
My impression is that most people don't want anything as "showmount -a" will have been returned the "interesting" information that you mention for some years and this is the first time it has even been mentioned.
Some people might want IP addresses for precision. Others might want host names so that it works with current tools.
Some people might want "who has mounted this since I last manually flushed the rmtab file". Others might want "who has accessed this in the last X minutes".
And if you want precise info from the kernel cache, you need to make them bigger than they currently need to be (The 'exports' list only need to be about the size of your /etc/exports file, but to get precise information, it will need to be roughly the size of the number of active clients).
So I really think that any solution to this issue needs to involve configuration options.
I would like to see mountd work internally with IP addresses, and map them to hostnames only as they are being returned to 'showmount' if that is what is wanted.
I would like to see mountd usually using a minimal cache, and only adding the extra information when it is actually requested. I can imagine that you might even want to added timestamps to the rmtab file which are updated when the kernel cache is updated, and the relevant lines are ignored if older than some set time.
But we really need a precise answer to "what are the actual requirements that someone has on 'showmount -a'??"
NeilBrown
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
Quoted 23 lines · Neil Brown
I don't agree that it makes no change.
At one level this is clear from the fact that it loses information.auth_unix_ip(which gets called when the kernel finds an unknown IP address and wants a name for it) always provides a mapping from the IP address to a hostname (As returned by the DNS) - if such a hostname is available. If the host is multi-homed (multiple IP addresses for the one host), this might lose information.Consider a situation where we have one subnet that is physically very secure, and we export some filesystem (rw,no_root_squash) to that subnet, and another subnet that is physically less secure and we export the filesystem (ro,root_squash) to that subnet.
Suppose further there is some host - foo - with an interface on both subnets.
Now attacker Alice sets up a machine on the less secure subnet and injects packets from foo's IP address (on that subnet). They get routed to the fileserver which maps the IP address to "foo". When you then call
client_check, both of the exports could match, and you might end up giving internal access to this external packets. Not good.
Quoted 3 lines
So I'm not happy with any patch that removes
client_composeandclient_memberin favour ofclient_resolve.
Here's where I'm confused. I don't see how the patch I proposed loses any information:
In the old method, we get an IP address and we build a list of client hostnames (in no particular order), using repeated calls to the client_check function. We then roll through the list of exports and make repeated calls to client_member to compare the m_client of the export against the list. If it doesn't match, we go to the next export.
With my patch, we don't bother building a list, and simply call client_check repeatedly within the loop to do the comparison. This seems to me to be functionally equivalent. Since the list is built solely on the result of client_check calls, and the order there doesn't seem to matter, then I don't see what information that it could carry that we don't get by just calling client_check directly.
So I don't see how the patch I proposed would leave us any more vulnerable to the situation you describe than what's already in place. The patch I proposed *does* remove one vulnerability as well. The situation where my_client has a list of hostnames that are no longer valid due to DNS or netgroup changes.
Quoted 9 lines
To the big-picture issue - you want "showmount -a" to return "the right thing". I don't believe there is a universal "the right thing" as various issues are fairly poorly defined. That is part of why I was think of adding flags to mountd so you could choose your poison.
The two main areas of uncertainty in "showmount" output are:
- how to identify the host - IP address or host name. The former is
more precise in the case of multi-homed hosts. The latter is
possibly more common.
Very true. Perhaps IP addresses would be better overall. It's certainly preferable to the current situation where "hostnames" can look like:
*,foo.example.com,192.168.10.0/255.255.255.0
Probably having it command line switchable is reasonable as well, but it seems like if people *don't* choose to have it report IP addresses, then it should give actual, resolvable, hostnames and not the comma-separated lists that we get now.
Quoted 7 lines
- How current should the information be? rmtab cannot be kept reliably uptodate and tends to grow - there can easily be hosts listed there that do not exist any more. Conversely the kernel cache information is relatively short term, and there could easily be clients that have the filesystem mounted but that are not listed in the cache.
Quoted 6 lines
So: what exactly do you - or your customers - want?
My impression is that most people don't want anything as "showmount -a" will have been returned the "interesting" information that you mention for some years and this is the first time it has even been mentioned.
The problem reports I have so far have really to do with the fact that the info returned by showmount -a doesn't really reflect anything comprehensible. The lack of reports is probably more reflective of the fact that rather few people bother to run showmount -a since it is traditionally unreliable (and not just on Linux). I'm sure, though, that once we get over the hurdle of having output that matches what's in the manpage, we'll start getting problem reports about what the info actually represents.
In my opinion, it would be ideal to have it list all active mounts. Now comes the question -- what constitutes an "active" mount? We really don't have a reliable way to track that, since a host can go idle and not send out any packets for a long time, and clients can spontaneously reboot and not remount the filesystem. The design of NFS makes this very difficult to reconcile.
As you said, neither the rmtab or the kernel cache is an ideal place to pull this info. So, my inclincation is to stick with the rmtab, and simply have it track what we know is trackable -- mount and unmount calls into mountd. I think having false positives is preferable to possibly having some mounts that are not reflected at all because they have gone idle.
Either way, a manpage update is probably also in order to outline the folly of depending on showmount -a :-).
Thanks,
Jeff
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
Quoted 14 lines · On Sunday December 10, [email protected] wrote:
Quoted 3 lines · Neil Brown
So I'm not happy with any patch that removes
client_composeandclient_memberin favour ofclient_resolve.Here's where I'm confused. I don't see how the patch I proposed loses any information:
In the old method, we get an IP address and we build a list of client hostnames (in no particular order), using repeated calls to the
client_checkfunction. We then roll through the list of exports and make repeated calls toclient_memberto compare them_clientof the export against the list. If it doesn't match, we go to the next export.
Quoted 6 lines
With my patch, we don't bother building a list, and simply call
client_checkrepeatedly within the loop to do the comparison. This seems to me to be functionally equivalent. Since the list is built solely on the result ofclient_checkcalls, and the order there doesn't seem to matter, then I don't see what information that it could carry that we don't get by just callingclient_checkdirectly.
Suppose your /etc/exports contains:
/baz 192.168.1.0/24(rw,no_root_squash)
/baz 192.168.2.0/24(ro,root_squash)
and your host 'foo' has two IP addresses: 192.168.1.10 and 192.168.2.10.
Suppose a request comes from 192.168.2.10 to access a file in /baz.
Old code:
Kernel asks for "192.168.2.10".
This goes to auth_unix_ip (in cache.c) which calls client_compose on that address. client_compose will return 192.168.2.0/24 so kernel is told
192.168.2.10 -> 192.168.2.0/24
Then kernel asks "how can 192.168.2.0/24 access /baz". This goes to nfsd_export which checks each export and the first (and only) one that matches is 192.168.2.0/24 so it answers "ro,root_squash" as you would hope.
New code:
Kernel asks for "192.168.2.10"
This goes to auth_unix_ip which calls client_resolve which calls get_reliable_hostbyaddr which returns "foo". So kernel is told
192.168.2.10 -> foo
Then kernel asks "how can foo access /baz".
This goes to nfsd_export which calls gethostbyname on 'foo' getting { "foo", 192.168.1.10, 192.168.2.10} and then calls client_check against each export. If the order is unchanged from the /etc/exports file, the first match will be
/baz 192.168.1.0/24(rw,no_root_squash)
so that will be returned to the kernel. Not what you would hope.
Does that make sense?
Quoted 5 lines
So I don't see how the patch I proposed would leave us any more vulnerable to the situation you describe than what's already in place. The patch I proposed *does* remove one vulnerability as well. The situation where
my_clienthas a list of hostnames that are no longer valid due to DNS or netgroup changes.
I agree that the my_client caching needs to be fixed.
Quoted 6 lines
As you said, neither the rmtab or the kernel cache is an ideal place to pull this info. So, my inclincation is to stick with the rmtab, and simply have it track what we know is trackable -- mount and unmount calls into mountd. I think having false positives is preferable to possibly having some mounts that are not reflected at all because they have gone idle.
Ok, how about this as a way forward.
1/ remove that 'my_client' caching.
2/ Remove the updates of rmtab on client upcalls and simply do rmtab
updates with hostname (from get_reliable_hostbyaddr) and path name from MOUNT or UMOUNT requests
3/ Add -I flag which:
passes IP address rather than client_compose name to kernel and adds rmtab entries on kernel upcalls. It also hard-removes rmtab entries on 'UMOUNT' and flush the kernel cache so that the next access for that host/path causes an upcall.
4/ Add a -? flag (haven't chosen a letter yet) which:
implies -I, but when asked for a 'DUMP', it calls gethosytbyname on each IP address to ge a hostname.
Then the default would be vaguely usable, and more precise information would be available (at a cost).
Quoted 3 lines
Either way, a manpage update is probably also in order to outline the folly of depending on showmount -a :-).
Yes, I would happily accept a patch making such a change.
NeilBrown
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
thanks for opening up this mail.
this is really my old problem, I used showmount -a to get who is mounted but its not reporting correctly. It includes the previous mounts until everything is messed up and you can no longer trust anything. I hope this can be improved at least. :-)
I would prefer IP address than hostnames.
warren
Quoted 118 lines · Neil Brown
Quoted 6 lines · On Sunday December 10, [email protected] wrote:
Quoted 3 lines · Neil Brown
So I'm not happy with any patch that removes
client_composeandclient_memberin favour ofclient_resolve.Here's where I'm confused. I don't see how the patch I proposed loses
any
Quoted 4 lines
information:
In the old method, we get an IP address and we build a list of client hostnames (in no particular order), using repeated calls to the
client_checkQuoted 5 lines
function. We then roll through the list of exports and make repeated calls to
client_memberto compare them_clientof the export against the list. If it doesn't match, we go to the next export.With my patch, we don't bother building a list, and simply call
client_checkQuoted 3 lines
repeatedly within the loop to do the comparison. This seems to me to be functionally equivalent. Since the list is built solely on the result of
client_checkcalls, and the order there doesn't seem to matter, then Idon't
Quoted 1 line
see what information that it could carry that we don't get by just
calling
Quoted 1 line
client_checkdirectly.Suppose your
/etc/exportscontains:/baz 192.168.1.0/24(rw,no_root_squash)
/baz 192.168.2.0/24(ro,root_squash)and your host 'foo' has two IP addresses: 192.168.1.10 and 192.168.2.10.
Suppose a request comes from 192.168.2.10 to access a file in /baz.
Old code:
Kernel asks for "192.168.2.10".
This goes toauth_unix_ip(incache.c) which callsclient_composeon that address.client_composewill return 192.168.2.0/24 so kernel is told192.168.2.10 -> 192.168.2.0/24Then kernel asks "how can 192.168.2.0/24 access /baz". This goes to
nfsd_exportwhich checks each export and the first (and only) one that matches is 192.168.2.0/24 so it answers "ro,root_squash" as you would hope.New code:
Kernel asks for "192.168.2.10"
This goes toauth_unix_ipwhich callsclient_resolvewhich callsget_reliable_hostbyaddrwhich returns "foo". So kernel is told192.168.2.10 -> fooThen kernel asks "how can foo access /baz".
This goes tonfsd_exportwhich calls gethostbyname on 'foo' getting { "foo", 192.168.1.10, 192.168.2.10} and then callsclient_checkagainst each export. If the order is unchanged from the/etc/exportsfile, the first match will be/baz 192.168.1.0/24(rw,no_root_squash)so that will be returned to the kernel. Not what you would hope.
Does that make sense?
Quoted 2 lines
So I don't see how the patch I proposed would leave us any more
vulnerable to
Quoted 1 line
the situation you describe than what's already in place. The patch I
proposed
Quoted 1 line
*does* remove one vulnerability as well. The situation where
my_clienthas
Quoted 1 line
a list of hostnames that are no longer valid due to DNS or netgroup
changes.
I agree that the
my_clientcaching needs to be fixed.Quoted 2 lines
As you said, neither the rmtab or the kernel cache is an ideal place to
pull
Quoted 1 line
this info. So, my inclincation is to stick with the rmtab, and simply
have it
Quoted 1 line
track what we know is trackable -- mount and unmount calls into mountd.
I
Quoted 1 line
think having false positives is preferable to possibly having some
mounts
Quoted 1 line
that are not reflected at all because they have gone idle.
Ok, how about this as a way forward.
1/ remove that '
my_client' caching.
2/ Remove the updates of rmtab on client upcalls and simply do rmtab
updates with hostname (fromget_reliable_hostbyaddr) and path name from MOUNT or UMOUNT requests
3/ Add -I flag which:
passes IP address rather thanclient_composename to kernel and adds rmtab entries on kernel upcalls. It also hard-removes rmtab entries on 'UMOUNT' and flush the kernel cache so that the next access for that host/path causes an upcall.
4/ Add a -? flag (haven't chosen a letter yet) which:
implies -I, but when asked for a 'DUMP', it calls gethosytbyname on each IP address to ge a hostname.Then the default would be vaguely usable, and more precise information would be available (at a cost).
Quoted 2 lines
Either way, a manpage update is probably also in order to outline the
folly
Quoted 1 line
of depending on showmount -a :-).
Yes, I would happily accept a patch making such a change.
NeilBrown
Quoted 10 lines
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
Quoted 2 lines · Neil Brown
Does that make sense?
Yes. Thanks for explaining that -- it makes total sense now. I was focusing on auth_authenticate, which I don't think would behave differently, but you're clearly correct that we'd need to ensure the same for the kernel caches.
Quoted 21 lines
Quoted 5 lines
As you said, neither the rmtab or the kernel cache is an ideal place to pull this info. So, my inclincation is to stick with the rmtab, and simply have it track what we know is trackable -- mount and unmount calls into mountd. I think having false positives is preferable to possibly having some mounts that are not reflected at all because they have gone idle.
Ok, how about this as a way forward.
1/ remove that '
my_client' caching.
2/ Remove the updates of rmtab on client upcalls and simply do rmtab
updates with hostname (fromget_reliable_hostbyaddr) and path name from MOUNT or UMOUNT requests
3/ Add -I flag which:
passes IP address rather thanclient_composename to kernel and adds rmtab entries on kernel upcalls. It also hard-removes rmtab entries on 'UMOUNT' and flush the kernel cache so that the next access for that host/path causes an upcall.
4/ Add a -? flag (haven't chosen a letter yet) which:
implies -I, but when asked for a 'DUMP', it calls gethosytbyname on each IP address to ge a hostname.
Quoted 3 lines
Then the default would be vaguely usable, and more precise information would be available (at a cost).
At first glance, that sounds reasonable. I'll have a closer look at the code and your idea and see if I can come up with a patch.
Quoted 5 lines
Quoted 2 lines
Either way, a manpage update is probably also in order to outline the folly of depending on showmount -a :-).
Yes, I would happily accept a patch making such a change.
Once we come to consensus on a code patch, I'll have a look at what manpage changes need to be made.
Thanks,
Jeff
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs