3 * Handle communication with knfsd internal cache
5 * We open /proc/net/rpc/{auth.unix.ip,nfsd.export,nfsd.fh}/channel
6 * and listen for requests (using my_svc_run)
14 #include <sys/types.h>
15 #include <sys/select.h>
19 #include <netinet/in.h>
20 #include <arpa/inet.h>
34 #include "pseudoflavors.h"
37 #include "blkid/blkid.h"
41 * Invoked by RPC service loop
43 void cache_set_fds(fd_set *fdset);
44 int cache_process_req(fd_set *readfds);
58 * Support routines for text-based upcalls.
59 * Fields are separated by spaces.
60 * Fields are either mangled to quote space tab newline slosh with slosh
61 * or a hexified with a leading \x
62 * Record is terminated with newline.
65 static int cache_export_ent(char *domain, struct exportent *exp, char *p);
67 #define INITIAL_MANAGED_GROUPS 100
71 extern int use_ipaddr;
73 static void auth_unix_ip(FILE *f)
77 * Ignore if class != "nfsd"
78 * Otherwise find domainname and write back:
80 * "nfsd" IP-ADDR expiry domainname
84 char ipaddr[INET6_ADDRSTRLEN];
86 struct addrinfo *tmp = NULL;
87 struct addrinfo *ai = NULL;
88 if (readline(fileno(f), &lbuf, &lbuflen) != 1)
91 xlog(D_CALL, "auth_unix_ip: inbuf '%s'", lbuf);
95 if (qword_get(&cp, class, 20) <= 0 ||
96 strcmp(class, "nfsd") != 0)
99 if (qword_get(&cp, ipaddr, sizeof(ipaddr)) <= 0)
102 tmp = host_pton(ipaddr);
108 /* addr is a valid, interesting address, find the domain name... */
110 ai = client_resolve(tmp->ai_addr);
111 client = client_compose(ai);
116 qword_print(f, "nfsd");
117 qword_print(f, ipaddr);
118 qword_printuint(f, time(0) + DEFAULT_TTL);
120 qword_print(f, ipaddr);
122 qword_print(f, *client?client:"DEFAULT");
124 xlog(D_CALL, "auth_unix_ip: client %p '%s'", client, client?client: "DEFAULT");
129 static void auth_unix_gid(FILE *f)
134 * uid expiry count list of group ids
138 static gid_t *groups = NULL;
139 static int groups_len = 0;
145 if (groups_len == 0) {
146 groups = malloc(sizeof(gid_t) * INITIAL_MANAGED_GROUPS);
150 groups_len = INITIAL_MANAGED_GROUPS;
153 ngroups = groups_len;
155 if (readline(fileno(f), &lbuf, &lbuflen) != 1)
159 if (qword_get_uint(&cp, &uid) != 0)
166 rv = getgrouplist(pw->pw_name, pw->pw_gid, groups, &ngroups);
167 if (rv == -1 && ngroups >= groups_len) {
168 more_groups = realloc(groups, sizeof(gid_t)*ngroups);
172 groups = more_groups;
173 groups_len = ngroups;
174 rv = getgrouplist(pw->pw_name, pw->pw_gid,
179 qword_printuint(f, uid);
180 qword_printuint(f, time(0) + DEFAULT_TTL);
182 qword_printuint(f, ngroups);
183 for (i=0; i<ngroups; i++)
184 qword_printuint(f, groups[i]);
186 qword_printuint(f, 0);
191 static const char *get_uuid_blkdev(char *path)
193 /* We set *safe if we know that we need the
194 * fsid from statfs too.
196 static blkid_cache cache = NULL;
199 blkid_tag_iterate iter;
202 const char *val, *uuid = NULL;
205 blkid_get_cache(&cache, NULL);
207 if (stat(path, &stb) != 0)
209 devname = blkid_devno_to_devname(stb.st_dev);
212 dev = blkid_get_dev(cache, devname, BLKID_DEV_NORMAL);
216 iter = blkid_tag_iterate_begin(dev);
219 while (blkid_tag_next(iter, &type, &val) == 0) {
220 if (strcmp(type, "UUID") == 0)
222 if (strcmp(type, "TYPE") == 0 &&
223 strcmp(val, "btrfs") == 0) {
228 blkid_tag_iterate_end(iter);
232 #define get_uuid_blkdev(path) (NULL)
235 static int get_uuid(const char *val, int uuidlen, char *u)
237 /* extract hex digits from uuidstr and compose a uuid
238 * of the given length (max 16), xoring bytes to make
243 memset(u, 0, uuidlen);
244 for ( ; *val ; val++) {
265 static int uuid_by_path(char *path, int type, int uuidlen, char *uuid)
267 /* get a uuid for the filesystem found at 'path'.
268 * There are several possible ways of generating the
270 * Type 0 is used for new filehandles, while other types
271 * may be used to interpret old filehandle - to ensure smooth
273 * We return 1 if a uuid was found (and it might be worth
274 * trying the next type) or 0 if no more uuid types can be
278 /* Possible sources of uuid are
282 * On some filesystems (e.g. vfat) the statfs64 uuid is simply an
283 * encoding of the device that the filesystem is mounted from, so
284 * it we be very bad to use that (as device numbers change). blkid
286 * On other filesystems (e.g. btrfs) the statfs64 uuid contains
287 * important info that the blkid uuid cannot contain: This happens
288 * when multiple subvolumes are exported (they have the same
289 * blkid uuid but different statfs64 uuids).
290 * We rely on get_uuid_blkdev *knowing* which is which and not returning
291 * a uuid for filesystems where the statfs64 uuid is better.
296 const char *blkid_val;
299 blkid_val = get_uuid_blkdev(path);
301 if (statfs64(path, &st) == 0 &&
302 (st.f_fsid.__val[0] || st.f_fsid.__val[1]))
303 snprintf(fsid_val, 17, "%08x%08x",
304 st.f_fsid.__val[0], st.f_fsid.__val[1]);
308 if (blkid_val && (type--) == 0)
310 else if (fsid_val[0] && (type--) == 0)
315 get_uuid(val, uuidlen, uuid);
319 /* Iterate through /etc/mtab, finding mountpoints
320 * at or below a given path
322 static char *next_mnt(void **v, char *p)
328 f = setmntent("/etc/mtab", "r");
332 while ((me = getmntent(f)) != NULL &&
333 (strncmp(me->mnt_dir, p, l) != 0 ||
334 me->mnt_dir[l] != '/'))
344 /* True iff e1 is a child of e2 and e2 has crossmnt set: */
345 static bool subexport(struct exportent *e1, struct exportent *e2)
347 char *p1 = e1->e_path, *p2 = e2->e_path;
350 return e2->e_flags & NFSEXP_CROSSMOUNT
351 && strncmp(p1, p2, l2) == 0
355 static void nfsd_fh(FILE *f)
358 * domain fsidtype fsid
359 * interpret fsid, find export point and options, and write:
360 * domain fsidtype fsid expiry path
366 unsigned int dev, major=0, minor=0;
367 unsigned int inode=0;
368 unsigned long long inode64;
369 unsigned int fsidnum=0;
371 struct exportent *found = NULL;
372 struct addrinfo *ai = NULL;
373 char *found_path = NULL;
380 if (readline(fileno(f), &lbuf, &lbuflen) != 1)
383 xlog(D_CALL, "nfsd_fh: inbuf '%s'", lbuf);
387 dom = malloc(strlen(cp));
390 if (qword_get(&cp, dom, strlen(cp)) <= 0)
392 if (qword_get_int(&cp, &fsidtype) != 0)
394 if (fsidtype < 0 || fsidtype > 7)
395 goto out; /* unknown type */
396 if ((fsidlen = qword_get(&cp, fsid, 32)) <= 0)
399 case FSID_DEV: /* 4 bytes: 2 major, 2 minor, 4 inode */
402 memcpy(&dev, fsid, 4);
403 memcpy(&inode, fsid+4, 4);
404 major = ntohl(dev)>>16;
405 minor = ntohl(dev) & 0xFFFF;
408 case FSID_NUM: /* 4 bytes - fsid */
411 memcpy(&fsidnum, fsid, 4);
414 case FSID_MAJOR_MINOR: /* 12 bytes: 4 major, 4 minor, 4 inode
415 * This format is never actually used but was
416 * an historical accident
420 memcpy(&dev, fsid, 4); major = ntohl(dev);
421 memcpy(&dev, fsid+4, 4); minor = ntohl(dev);
422 memcpy(&inode, fsid+8, 4);
425 case FSID_ENCODE_DEV: /* 8 bytes: 4 byte packed device number, 4 inode */
426 /* This is *host* endian, not net-byte-order, because
427 * no-one outside this host has any business interpreting it
431 memcpy(&dev, fsid, 4);
432 memcpy(&inode, fsid+4, 4);
433 major = (dev & 0xfff00) >> 8;
434 minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
437 case FSID_UUID4_INUM: /* 4 byte inode number and 4 byte uuid */
440 memcpy(&inode, fsid, 4);
444 case FSID_UUID8: /* 8 byte uuid */
450 case FSID_UUID16: /* 16 byte uuid */
456 case FSID_UUID16_INUM: /* 8 byte inode number and 16 byte uuid */
459 memcpy(&inode64, fsid, 8);
468 /* Now determine export point for this fsid/domain */
469 for (i=0 ; i < MCL_MAXTYPES; i++) {
470 nfs_export *next_exp;
471 for (exp = exportlist[i].p_head; exp; exp = next_exp) {
477 if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT) {
478 static nfs_export *prev = NULL;
479 static void *mnt = NULL;
483 path = next_mnt(&mnt, exp->m_export.e_path);
485 next_exp = exp->m_next;
493 path = exp->m_export.e_path;
497 path = exp->m_export.e_path;
498 next_exp = exp->m_next;
501 if (!use_ipaddr && !client_member(dom, exp->m_client->m_hostname))
503 if (exp->m_export.e_mountpoint &&
504 !is_mountpoint(exp->m_export.e_mountpoint[0]?
505 exp->m_export.e_mountpoint:
506 exp->m_export.e_path))
508 if (stat(path, &stb) != 0)
510 if (!S_ISDIR(stb.st_mode) && !S_ISREG(stb.st_mode)) {
515 case FSID_MAJOR_MINOR:
516 case FSID_ENCODE_DEV:
517 if (stb.st_ino != inode)
519 if (major != major(stb.st_dev) ||
520 minor != minor(stb.st_dev))
524 if (((exp->m_export.e_flags & NFSEXP_FSID) == 0 ||
525 exp->m_export.e_fsid != fsidnum))
528 case FSID_UUID4_INUM:
529 case FSID_UUID16_INUM:
530 if (stb.st_ino != inode)
535 if (!is_mountpoint(path))
538 if (exp->m_export.e_uuid)
539 get_uuid(exp->m_export.e_uuid,
543 uuid_by_path(path, type, uuidlen, u);
545 if (memcmp(u, fhuuid, uuidlen) == 0)
548 if (memcmp(u, fhuuid, uuidlen) != 0)
554 struct addrinfo *tmp;
555 tmp = host_pton(dom);
558 ai = client_resolve(tmp->ai_addr);
561 if (!client_check(exp->m_client, ai))
564 if (!found || subexport(&exp->m_export, found)) {
565 found = &exp->m_export;
567 found_path = strdup(path);
568 if (found_path == NULL)
570 } else if (strcmp(found->e_path, exp->m_export.e_path)
571 && !subexport(found, &exp->m_export))
573 xlog(L_WARNING, "%s and %s have same filehandle for %s, using first",
574 found_path, path, dom);
579 found->e_mountpoint &&
580 !is_mountpoint(found->e_mountpoint[0]?
583 /* Cannot export this yet
584 * should log a warning, but need to rate limit
585 xlog(L_WARNING, "%s not exported as %d not a mountpoint",
586 found->e_path, found->e_mountpoint);
588 /* FIXME we need to make sure we re-visit this later */
591 if (!found && dev_missing) {
592 /* The missing dev could be what we want, so just be
593 * quite rather than returning stale yet
599 if (cache_export_ent(dom, found, found_path) < 0)
603 qword_printint(f, fsidtype);
604 qword_printhex(f, fsid, fsidlen);
605 /* The fsid -> path lookup can be quite expensive as it
606 * potentially stats and reads lots of devices, and some of those
607 * might have spun-down. The Answer is not likely to
608 * change underneath us, and an 'exportfs -f' can always
609 * remove this from the kernel, so use a really log
610 * timeout. Maybe this should be configurable on the command
613 qword_printint(f, 0x7fffffff);
615 qword_print(f, found_path);
622 xlog(D_CALL, "nfsd_fh: found %p path %s", found, found ? found->e_path : NULL);
626 static void write_fsloc(FILE *f, struct exportent *ep)
628 struct servers *servers;
630 if (ep->e_fslocmethod == FSLOC_NONE)
633 servers = replicas_lookup(ep->e_fslocmethod, ep->e_fslocdata);
636 qword_print(f, "fsloc");
637 qword_printint(f, servers->h_num);
638 if (servers->h_num >= 0) {
640 for (i=0; i<servers->h_num; i++) {
641 qword_print(f, servers->h_mp[i]->h_host);
642 qword_print(f, servers->h_mp[i]->h_path);
645 qword_printint(f, servers->h_referral);
646 release_replicas(servers);
649 static void write_secinfo(FILE *f, struct exportent *ep, int flag_mask)
653 for (p = ep->e_secinfo; p->flav; p++)
655 if (p == ep->e_secinfo) {
656 /* There was no sec= option */
659 qword_print(f, "secinfo");
660 qword_printint(f, p - ep->e_secinfo);
661 for (p = ep->e_secinfo; p->flav; p++) {
662 qword_printint(f, p->flav->fnum);
663 qword_printint(f, p->flags & flag_mask);
668 static int dump_to_cache(FILE *f, char *domain, char *path, struct exportent *exp)
670 qword_print(f, domain);
671 qword_print(f, path);
673 int different_fs = strcmp(path, exp->e_path) != 0;
674 int flag_mask = different_fs ? ~NFSEXP_FSID : ~0;
676 qword_printuint(f, time(0) + exp->e_ttl);
677 qword_printint(f, exp->e_flags & flag_mask);
678 qword_printint(f, exp->e_anonuid);
679 qword_printint(f, exp->e_anongid);
680 qword_printint(f, exp->e_fsid);
682 write_secinfo(f, exp, flag_mask);
683 if (exp->e_uuid == NULL || different_fs) {
685 if (uuid_by_path(path, 0, 16, u)) {
686 qword_print(f, "uuid");
687 qword_printhex(f, u, 16);
691 get_uuid(exp->e_uuid, 16, u);
692 qword_print(f, "uuid");
693 qword_printhex(f, u, 16);
696 qword_printuint(f, time(0) + DEFAULT_TTL);
700 static int is_subdirectory(char *child, char *parent)
702 int l = strlen(parent);
704 return strcmp(child, parent) == 0
705 || (strncmp(child, parent, l) == 0 && child[l] == '/');
708 static int path_matches(nfs_export *exp, char *path)
710 if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)
711 return is_subdirectory(path, exp->m_export.e_path);
712 return strcmp(path, exp->m_export.e_path) == 0;
716 client_matches(nfs_export *exp, char *dom, struct addrinfo *ai)
719 return client_check(exp->m_client, ai);
720 return client_member(dom, exp->m_client->m_hostname);
724 export_matches(nfs_export *exp, char *dom, char *path, struct addrinfo *ai)
726 return path_matches(exp, path) && client_matches(exp, dom, ai);
730 lookup_export(char *dom, char *path, struct addrinfo *ai)
733 nfs_export *found = NULL;
737 for (i=0 ; i < MCL_MAXTYPES; i++) {
738 for (exp = exportlist[i].p_head; exp; exp = exp->m_next) {
739 if (!export_matches(exp, dom, path, ai))
747 /* Always prefer non-V4ROOT mounts */
748 if (found->m_export.e_flags & NFSEXP_V4ROOT)
751 /* If one is a CROSSMOUNT, then prefer the longest path */
752 if (((found->m_export.e_flags & NFSEXP_CROSSMOUNT) ||
753 (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)) &&
754 strlen(found->m_export.e_path) !=
755 strlen(exp->m_export.e_path)) {
757 if (strlen(exp->m_export.e_path) >
758 strlen(found->m_export.e_path)) {
764 } else if (found_type == i && found->m_warned == 0) {
765 xlog(L_WARNING, "%s exported to both %s and %s, "
766 "arbitrarily choosing options from first",
767 path, found->m_client->m_hostname, exp->m_client->m_hostname,
776 static void nfsd_export(FILE *f)
780 * determine export options and return:
781 * domain path expiry flags anonuid anongid fsid
786 nfs_export *found = NULL;
787 struct addrinfo *ai = NULL;
789 if (readline(fileno(f), &lbuf, &lbuflen) != 1)
792 xlog(D_CALL, "nfsd_export: inbuf '%s'", lbuf);
795 dom = malloc(strlen(cp));
796 path = malloc(strlen(cp));
801 if (qword_get(&cp, dom, strlen(lbuf)) <= 0)
803 if (qword_get(&cp, path, strlen(lbuf)) <= 0)
809 struct addrinfo *tmp;
810 tmp = host_pton(dom);
813 ai = client_resolve(tmp->ai_addr);
818 found = lookup_export(dom, path, ai);
821 if (dump_to_cache(f, dom, path, &found->m_export) < 0) {
823 "Cannot export %s, possibly unsupported filesystem"
824 " or fsid= required", path);
825 dump_to_cache(f, dom, path, NULL);
828 dump_to_cache(f, dom, path, NULL);
831 xlog(D_CALL, "nfsd_export: found %p path %s", found, path ? path : NULL);
833 if (path) free(path);
840 void (*cache_handle)(FILE *f);
842 char vbuf[RPC_CHAN_BUF_SIZE];
844 { "auth.unix.ip", auth_unix_ip, NULL, ""},
845 { "auth.unix.gid", auth_unix_gid, NULL, ""},
846 { "nfsd.export", nfsd_export, NULL, ""},
847 { "nfsd.fh", nfsd_fh, NULL, ""},
848 { NULL, NULL, NULL, ""}
851 extern int manage_gids;
854 * cache_open - prepare communications channels with kernel RPC caches
857 void cache_open(void)
860 for (i=0; cachelist[i].cache_name; i++ ) {
862 if (!manage_gids && cachelist[i].cache_handle == auth_unix_gid)
864 sprintf(path, "/proc/net/rpc/%s/channel", cachelist[i].cache_name);
865 cachelist[i].f = fopen(path, "r+");
866 if (cachelist[i].f != NULL) {
867 setvbuf(cachelist[i].f, cachelist[i].vbuf, _IOLBF,
874 * cache_set_fds - prepare cache file descriptors for one iteration of the service loop
875 * @fdset: pointer to fd_set to prepare
877 void cache_set_fds(fd_set *fdset)
880 for (i=0; cachelist[i].cache_name; i++) {
882 FD_SET(fileno(cachelist[i].f), fdset);
887 * cache_process_req - process any active cache file descriptors during service loop iteration
888 * @fdset: pointer to fd_set to examine for activity
890 int cache_process_req(fd_set *readfds)
894 for (i=0; cachelist[i].cache_name; i++) {
895 if (cachelist[i].f != NULL &&
896 FD_ISSET(fileno(cachelist[i].f), readfds)) {
898 cachelist[i].cache_handle(cachelist[i].f);
899 FD_CLR(fileno(cachelist[i].f), readfds);
907 * Give IP->domain and domain+path->options to kernel
908 * % echo nfsd $IP $[now+DEFAULT_TTL] $domain > /proc/net/rpc/auth.unix.ip/channel
909 * % echo $domain $path $[now+DEFAULT_TTL] $options $anonuid $anongid $fsid > /proc/net/rpc/nfsd.export/channel
912 static int cache_export_ent(char *domain, struct exportent *exp, char *path)
915 FILE *f = fopen("/proc/net/rpc/nfsd.export/channel", "w");
919 err = dump_to_cache(f, domain, exp->e_path, exp);
922 "Cannot export %s, possibly unsupported filesystem or"
923 " fsid= required", exp->e_path);
926 while (err == 0 && (exp->e_flags & NFSEXP_CROSSMOUNT) && path) {
927 /* really an 'if', but we can break out of
928 * a 'while' more easily */
929 /* Look along 'path' for other filesystems
930 * and export them with the same options
933 size_t l = strlen(exp->e_path);
936 if (strlen(path) <= l || path[l] != '/' ||
937 strncmp(exp->e_path, path, l) != 0)
939 if (stat(exp->e_path, &stb) != 0)
942 while(path[l] == '/') {
944 /* errors for submount should fail whole filesystem */
948 while (path[l] != '/' && path[l])
952 err2 = lstat(path, &stb);
956 if (stb.st_dev == dev)
960 dump_to_cache(f, domain, path, exp);
971 * cache_export - Inform kernel of a new nfs_export
972 * @exp: target nfs_export
973 * @path: NUL-terminated C string containing export path
975 int cache_export(nfs_export *exp, char *path)
977 char buf[INET6_ADDRSTRLEN];
981 f = fopen("/proc/net/rpc/auth.unix.ip/channel", "w");
986 qword_print(f, "nfsd");
988 host_ntop(get_addrlist(exp->m_client, 0), buf, sizeof(buf)));
989 qword_printuint(f, time(0) + exp->m_export.e_ttl);
990 qword_print(f, exp->m_client->m_hostname);
995 err = cache_export_ent(exp->m_client->m_hostname, &exp->m_export, path)
1001 * cache_get_filehandle - given an nfs_export, get its root filehandle
1002 * @exp: target nfs_export
1003 * @len: length of requested file handle
1004 * @p: NUL-terminated C string containing export path
1006 * Returns pointer to NFS file handle of root directory of export
1009 * echo $domain $path $length
1010 * read filehandle <&0
1011 * } <> /proc/fs/nfsd/filehandle
1014 cache_get_filehandle(nfs_export *exp, int len, char *p)
1016 FILE *f = fopen("/proc/fs/nfsd/filehandle", "r+");
1020 static struct nfs_fh_len fh;
1023 f = fopen("/proc/fs/nfs/filehandle", "r+");
1027 qword_print(f, exp->m_client->m_hostname);
1029 qword_printint(f, len);
1030 failed = qword_eol(f);
1033 failed = (fgets(buf, sizeof(buf), f) == NULL);
1037 memset(fh.fh_handle, 0, sizeof(fh.fh_handle));
1038 fh.fh_size = qword_get(&bp, (char *)fh.fh_handle, NFS3_FHSIZE);