b01c0bdb4e944286a03664c33c53bb922999e04a
[nfs-utils.git] / utils / mountd / cache.c
1
2 /*
3  * Handle communication with knfsd internal cache
4  *
5  * We open /proc/net/rpc/{auth.unix.ip,nfsd.export,nfsd.fh}/channel
6  * and listen for requests (using my_svc_run)
7  * 
8  */
9
10 #ifdef HAVE_CONFIG_H
11 #include <config.h>
12 #endif
13
14 #include <sys/types.h>
15 #include <sys/select.h>
16 #include <sys/stat.h>
17 #include <sys/vfs.h>
18 #include <time.h>
19 #include <netinet/in.h>
20 #include <arpa/inet.h>
21 #include <unistd.h>
22 #include <fcntl.h>
23 #include <errno.h>
24 #include <ctype.h>
25 #include <pwd.h>
26 #include <grp.h>
27 #include <mntent.h>
28 #include "misc.h"
29 #include "nfslib.h"
30 #include "exportfs.h"
31 #include "mountd.h"
32 #include "xmalloc.h"
33 #include "fsloc.h"
34 #include "pseudoflavors.h"
35
36 #ifdef USE_BLKID
37 #include "blkid/blkid.h"
38 #endif
39
40 /*
41  * Invoked by RPC service loop
42  */
43 void    cache_set_fds(fd_set *fdset);
44 int     cache_process_req(fd_set *readfds);
45
46 enum nfsd_fsid {
47         FSID_DEV = 0,
48         FSID_NUM,
49         FSID_MAJOR_MINOR,
50         FSID_ENCODE_DEV,
51         FSID_UUID4_INUM,
52         FSID_UUID8,
53         FSID_UUID16,
54         FSID_UUID16_INUM,
55 };
56
57 /*
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.
63  *
64  */
65 static int cache_export_ent(char *domain, struct exportent *exp, char *p);
66
67 #define INITIAL_MANAGED_GROUPS 100
68
69 char *lbuf  = NULL;
70 int lbuflen = 0;
71 extern int use_ipaddr;
72
73 static void auth_unix_ip(FILE *f)
74 {
75         /* requests are
76          *  class IP-ADDR
77          * Ignore if class != "nfsd"
78          * Otherwise find domainname and write back:
79          *
80          *  "nfsd" IP-ADDR expiry domainname
81          */
82         char *cp;
83         char class[20];
84         char ipaddr[INET6_ADDRSTRLEN];
85         char *client = NULL;
86         struct addrinfo *tmp = NULL;
87         struct addrinfo *ai = NULL;
88         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
89                 return;
90
91         xlog(D_CALL, "auth_unix_ip: inbuf '%s'", lbuf);
92
93         cp = lbuf;
94
95         if (qword_get(&cp, class, 20) <= 0 ||
96             strcmp(class, "nfsd") != 0)
97                 return;
98
99         if (qword_get(&cp, ipaddr, sizeof(ipaddr)) <= 0)
100                 return;
101
102         tmp = host_pton(ipaddr);
103         if (tmp == NULL)
104                 return;
105
106         auth_reload();
107
108         /* addr is a valid, interesting address, find the domain name... */
109         if (!use_ipaddr) {
110                 ai = client_resolve(tmp->ai_addr);
111                 client = client_compose(ai);
112                 freeaddrinfo(ai);
113         }
114         freeaddrinfo(tmp);
115
116         qword_print(f, "nfsd");
117         qword_print(f, ipaddr);
118         qword_printuint(f, time(0) + DEFAULT_TTL);
119         if (use_ipaddr)
120                 qword_print(f, ipaddr);
121         else if (client)
122                 qword_print(f, *client?client:"DEFAULT");
123         qword_eol(f);
124         xlog(D_CALL, "auth_unix_ip: client %p '%s'", client, client?client: "DEFAULT");
125
126         free(client);
127 }
128
129 static void auth_unix_gid(FILE *f)
130 {
131         /* Request are
132          *  uid
133          * reply is
134          *  uid expiry count list of group ids
135          */
136         uid_t uid;
137         struct passwd *pw;
138         static gid_t *groups = NULL;
139         static int groups_len = 0;
140         gid_t *more_groups;
141         int ngroups;
142         int rv, i;
143         char *cp;
144
145         if (groups_len == 0) {
146                 groups = malloc(sizeof(gid_t) * INITIAL_MANAGED_GROUPS);
147                 if (!groups)
148                         return;
149
150                 groups_len = INITIAL_MANAGED_GROUPS;
151         }
152
153         ngroups = groups_len;
154
155         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
156                 return;
157
158         cp = lbuf;
159         if (qword_get_uint(&cp, &uid) != 0)
160                 return;
161
162         pw = getpwuid(uid);
163         if (!pw)
164                 rv = -1;
165         else {
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);
169                         if (!more_groups)
170                                 rv = -1;
171                         else {
172                                 groups = more_groups;
173                                 groups_len = ngroups;
174                                 rv = getgrouplist(pw->pw_name, pw->pw_gid,
175                                                   groups, &ngroups);
176                         }
177                 }
178         }
179         qword_printuint(f, uid);
180         qword_printuint(f, time(0) + DEFAULT_TTL);
181         if (rv >= 0) {
182                 qword_printuint(f, ngroups);
183                 for (i=0; i<ngroups; i++)
184                         qword_printuint(f, groups[i]);
185         } else
186                 qword_printuint(f, 0);
187         qword_eol(f);
188 }
189
190 #if USE_BLKID
191 static const char *get_uuid_blkdev(char *path)
192 {
193         /* We set *safe if we know that we need the
194          * fsid from statfs too.
195          */
196         static blkid_cache cache = NULL;
197         struct stat stb;
198         char *devname;
199         blkid_tag_iterate iter;
200         blkid_dev dev;
201         const char *type;
202         const char *val, *uuid = NULL;
203
204         if (cache == NULL)
205                 blkid_get_cache(&cache, NULL);
206
207         if (stat(path, &stb) != 0)
208                 return NULL;
209         devname = blkid_devno_to_devname(stb.st_dev);
210         if (!devname)
211                 return NULL;
212         dev = blkid_get_dev(cache, devname, BLKID_DEV_NORMAL);
213         free(devname);
214         if (!dev)
215                 return NULL;
216         iter = blkid_tag_iterate_begin(dev);
217         if (!iter)
218                 return NULL;
219         while (blkid_tag_next(iter, &type, &val) == 0) {
220                 if (strcmp(type, "UUID") == 0)
221                         uuid = val;
222                 if (strcmp(type, "TYPE") == 0 &&
223                     strcmp(val, "btrfs") == 0) {
224                         uuid = NULL;
225                         break;
226                 }
227         }
228         blkid_tag_iterate_end(iter);
229         return uuid;
230 }
231 #else
232 #define get_uuid_blkdev(path) (NULL)
233 #endif
234
235 static int get_uuid(const char *val, int uuidlen, char *u)
236 {
237         /* extract hex digits from uuidstr and compose a uuid
238          * of the given length (max 16), xoring bytes to make
239          * a smaller uuid.
240          */
241         int i = 0;
242         
243         memset(u, 0, uuidlen);
244         for ( ; *val ; val++) {
245                 char c = *val;
246                 if (!isxdigit(c))
247                         continue;
248                 if (isalpha(c)) {
249                         if (isupper(c))
250                                 c = c - 'A' + 10;
251                         else
252                                 c = c - 'a' + 10;
253                 } else
254                         c = c - '0' + 0;
255                 if ((i&1) == 0)
256                         c <<= 4;
257                 u[i/2] ^= c;
258                 i++;
259                 if (i == uuidlen*2)
260                         i = 0;
261         }
262         return 1;
263 }
264
265 static int uuid_by_path(char *path, int type, int uuidlen, char *uuid)
266 {
267         /* get a uuid for the filesystem found at 'path'.
268          * There are several possible ways of generating the
269          * uuids (types).
270          * Type 0 is used for new filehandles, while other types
271          * may be used to interpret old filehandle - to ensure smooth
272          * forward migration.
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
275          * extracted.
276          */
277
278         /* Possible sources of uuid are
279          * - blkid uuid
280          * - statfs64 uuid
281          *
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
285          * must be preferred.
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.
292          *
293          */
294         struct statfs64 st;
295         char fsid_val[17];
296         const char *blkid_val;
297         const char *val;
298
299         blkid_val = get_uuid_blkdev(path);
300
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]);
305         else
306                 fsid_val[0] = 0;
307
308         if (blkid_val && (type--) == 0)
309                 val = blkid_val;
310         else if (fsid_val[0] && (type--) == 0)
311                 val = fsid_val;
312         else
313                 return 0;
314
315         get_uuid(val, uuidlen, uuid);
316         return 1;
317 }
318
319 /* Iterate through /etc/mtab, finding mountpoints
320  * at or below a given path
321  */
322 static char *next_mnt(void **v, char *p)
323 {
324         FILE *f;
325         struct mntent *me;
326         int l = strlen(p);
327         if (*v == NULL) {
328                 f = setmntent("/etc/mtab", "r");
329                 *v = f;
330         } else
331                 f = *v;
332         while ((me = getmntent(f)) != NULL &&
333                (strncmp(me->mnt_dir, p, l) != 0 ||
334                 me->mnt_dir[l] != '/'))
335                 ;
336         if (me == NULL) {
337                 endmntent(f);
338                 *v = NULL;
339                 return NULL;
340         }
341         return me->mnt_dir;
342 }
343
344 /* True iff e1 is a child of e2 and e2 has crossmnt set: */
345 static bool subexport(struct exportent *e1, struct exportent *e2)
346 {
347         char *p1 = e1->e_path, *p2 = e2->e_path;
348         int l2 = strlen(p2);
349
350         return e2->e_flags & NFSEXP_CROSSMOUNT
351                && strncmp(p1, p2, l2) == 0
352                && p1[l2] == '/';
353 }
354
355 struct parsed_fsid {
356         int fsidtype;
357         /* We could use a union for this, but it would be more
358          * complicated; why bother? */
359         unsigned int inode;
360         unsigned int minor;
361         unsigned int major;
362         unsigned int fsidnum;
363         int uuidlen;
364         char *fhuuid;
365 };
366
367 int parse_fsid(int fsidtype, int fsidlen, char *fsid, struct parsed_fsid *parsed)
368 {
369         unsigned int dev;
370         unsigned long long inode64;
371
372         parsed->fsidtype = fsidtype;
373         switch(fsidtype) {
374         case FSID_DEV: /* 4 bytes: 2 major, 2 minor, 4 inode */
375                 if (fsidlen != 8)
376                         return -1;
377                 memcpy(&dev, fsid, 4);
378                 memcpy(&parsed->inode, fsid+4, 4);
379                 parsed->major = ntohl(dev)>>16;
380                 parsed->minor = ntohl(dev) & 0xFFFF;
381                 break;
382
383         case FSID_NUM: /* 4 bytes - fsid */
384                 if (fsidlen != 4)
385                         return -1;
386                 memcpy(&parsed->fsidnum, fsid, 4);
387                 break;
388
389         case FSID_MAJOR_MINOR: /* 12 bytes: 4 major, 4 minor, 4 inode 
390                  * This format is never actually used but was
391                  * an historical accident
392                  */
393                 if (fsidlen != 12)
394                         return -1;
395                 memcpy(&dev, fsid, 4);
396                 parsed->major = ntohl(dev);
397                 memcpy(&dev, fsid+4, 4);
398                 parsed->minor = ntohl(dev);
399                 memcpy(&parsed->inode, fsid+8, 4);
400                 break;
401
402         case FSID_ENCODE_DEV: /* 8 bytes: 4 byte packed device number, 4 inode */
403                 /* This is *host* endian, not net-byte-order, because
404                  * no-one outside this host has any business interpreting it
405                  */
406                 if (fsidlen != 8)
407                         return -1;
408                 memcpy(&dev, fsid, 4);
409                 memcpy(&parsed->inode, fsid+4, 4);
410                 parsed->major = (dev & 0xfff00) >> 8;
411                 parsed->minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
412                 break;
413
414         case FSID_UUID4_INUM: /* 4 byte inode number and 4 byte uuid */
415                 if (fsidlen != 8)
416                         return -1;
417                 memcpy(&parsed->inode, fsid, 4);
418                 parsed->uuidlen = 4;
419                 parsed->fhuuid = fsid+4;
420                 break;
421         case FSID_UUID8: /* 8 byte uuid */
422                 if (fsidlen != 8)
423                         return -1;
424                 parsed->uuidlen = 8;
425                 parsed->fhuuid = fsid;
426                 break;
427         case FSID_UUID16: /* 16 byte uuid */
428                 if (fsidlen != 16)
429                         return -1;
430                 parsed->uuidlen = 16;
431                 parsed->fhuuid = fsid;
432                 break;
433         case FSID_UUID16_INUM: /* 8 byte inode number and 16 byte uuid */
434                 if (fsidlen != 24)
435                         return -1;
436                 memcpy(&inode64, fsid, 8);
437                 parsed->inode = inode64;
438                 parsed->uuidlen = 16;
439                 parsed->fhuuid = fsid+8;
440                 break;
441         }
442         return 0;
443 }
444
445 static bool match_fsid(struct parsed_fsid *parsed, nfs_export *exp, char *path)
446 {
447         struct stat stb;
448         int type;
449         char u[16];
450
451         if (stat(path, &stb) != 0)
452                 return false;
453         if (!S_ISDIR(stb.st_mode) && !S_ISREG(stb.st_mode))
454                 return false;
455
456         switch (parsed->fsidtype) {
457         case FSID_DEV:
458         case FSID_MAJOR_MINOR:
459         case FSID_ENCODE_DEV:
460                 if (stb.st_ino != parsed->inode)
461                         return false;
462                 if (parsed->major != major(stb.st_dev) ||
463                     parsed->minor != minor(stb.st_dev))
464                         return false;
465                 return true;
466         case FSID_NUM:
467                 if (((exp->m_export.e_flags & NFSEXP_FSID) == 0 ||
468                      exp->m_export.e_fsid != parsed->fsidnum))
469                         return false;
470                 return true;
471         case FSID_UUID4_INUM:
472         case FSID_UUID16_INUM:
473                 if (stb.st_ino != parsed->inode)
474                         return false;
475                 goto check_uuid;
476         case FSID_UUID8:
477         case FSID_UUID16:
478                 if (!is_mountpoint(path))
479                         return false;
480         check_uuid:
481                 if (exp->m_export.e_uuid)
482                         get_uuid(exp->m_export.e_uuid, parsed->uuidlen, u);
483                 else
484                         for (type = 0;
485                              uuid_by_path(path, type, parsed->uuidlen, u);
486                              type++)
487                                 if (memcmp(u, parsed->fhuuid, parsed->uuidlen) == 0)
488                                         return true;
489
490                 if (memcmp(u, parsed->fhuuid, parsed->uuidlen) != 0)
491                         return false;
492                 return true;
493         }
494         /* Well, unreachable, actually: */
495         return false;
496 }
497
498 static void nfsd_fh(FILE *f)
499 {
500         /* request are:
501          *  domain fsidtype fsid
502          * interpret fsid, find export point and options, and write:
503          *  domain fsidtype fsid expiry path
504          */
505         char *cp;
506         char *dom;
507         int fsidtype;
508         int fsidlen;
509         char fsid[32];
510         struct parsed_fsid parsed;
511         struct exportent *found = NULL;
512         struct addrinfo *ai = NULL;
513         char *found_path = NULL;
514         nfs_export *exp;
515         int i;
516         int dev_missing = 0;
517
518         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
519                 return;
520
521         xlog(D_CALL, "nfsd_fh: inbuf '%s'", lbuf);
522
523         cp = lbuf;
524
525         dom = malloc(strlen(cp));
526         if (dom == NULL)
527                 return;
528         if (qword_get(&cp, dom, strlen(cp)) <= 0)
529                 goto out;
530         if (qword_get_int(&cp, &fsidtype) != 0)
531                 goto out;
532         if (fsidtype < 0 || fsidtype > 7)
533                 goto out; /* unknown type */
534         if ((fsidlen = qword_get(&cp, fsid, 32)) <= 0)
535                 goto out;
536         if (parse_fsid(fsidtype, fsidlen, fsid, &parsed))
537                 goto out;
538
539         auth_reload();
540
541         /* Now determine export point for this fsid/domain */
542         for (i=0 ; i < MCL_MAXTYPES; i++) {
543                 nfs_export *next_exp;
544                 for (exp = exportlist[i].p_head; exp; exp = next_exp) {
545                         char *path;
546
547                         if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT) {
548                                 static nfs_export *prev = NULL;
549                                 static void *mnt = NULL;
550                                 
551                                 if (prev == exp) {
552                                         /* try a submount */
553                                         path = next_mnt(&mnt, exp->m_export.e_path);
554                                         if (!path) {
555                                                 next_exp = exp->m_next;
556                                                 prev = NULL;
557                                                 continue;
558                                         }
559                                         next_exp = exp;
560                                 } else {
561                                         prev = exp;
562                                         mnt = NULL;
563                                         path = exp->m_export.e_path;
564                                         next_exp = exp;
565                                 }
566                         } else {
567                                 path = exp->m_export.e_path;
568                                 next_exp = exp->m_next;
569                         }
570
571                         if (!use_ipaddr && !client_member(dom, exp->m_client->m_hostname))
572                                 continue;
573                         if (exp->m_export.e_mountpoint &&
574                             !is_mountpoint(exp->m_export.e_mountpoint[0]?
575                                            exp->m_export.e_mountpoint:
576                                            exp->m_export.e_path))
577                                 dev_missing ++;
578
579                         if (!match_fsid(&parsed, exp, path))
580                                 continue;
581                         if (use_ipaddr) {
582                                 if (ai == NULL) {
583                                         struct addrinfo *tmp;
584                                         tmp = host_pton(dom);
585                                         if (tmp == NULL)
586                                                 goto out;
587                                         ai = client_resolve(tmp->ai_addr);
588                                         freeaddrinfo(tmp);
589                                 }
590                                 if (!client_check(exp->m_client, ai))
591                                         continue;
592                         }
593                         if (!found || subexport(&exp->m_export, found)) {
594                                 found = &exp->m_export;
595                                 free(found_path);
596                                 found_path = strdup(path);
597                                 if (found_path == NULL)
598                                         goto out;
599                         } else if (strcmp(found->e_path, exp->m_export.e_path)
600                                    && !subexport(found, &exp->m_export))
601                         {
602                                 xlog(L_WARNING, "%s and %s have same filehandle for %s, using first",
603                                      found_path, path, dom);
604                         }
605                 }
606         }
607         if (found && 
608             found->e_mountpoint &&
609             !is_mountpoint(found->e_mountpoint[0]?
610                            found->e_mountpoint:
611                            found->e_path)) {
612                 /* Cannot export this yet 
613                  * should log a warning, but need to rate limit
614                    xlog(L_WARNING, "%s not exported as %d not a mountpoint",
615                    found->e_path, found->e_mountpoint);
616                  */
617                 /* FIXME we need to make sure we re-visit this later */
618                 goto out;
619         }
620         if (!found && dev_missing) {
621                 /* The missing dev could be what we want, so just be
622                  * quite rather than returning stale yet
623                  */
624                 goto out;
625         }
626
627         if (found)
628                 if (cache_export_ent(dom, found, found_path) < 0)
629                         found = 0;
630
631         qword_print(f, dom);
632         qword_printint(f, fsidtype);
633         qword_printhex(f, fsid, fsidlen);
634         /* The fsid -> path lookup can be quite expensive as it
635          * potentially stats and reads lots of devices, and some of those
636          * might have spun-down.  The Answer is not likely to
637          * change underneath us, and an 'exportfs -f' can always
638          * remove this from the kernel, so use a really log
639          * timeout.  Maybe this should be configurable on the command
640          * line.
641          */
642         qword_printint(f, 0x7fffffff);
643         if (found)
644                 qword_print(f, found_path);
645         qword_eol(f);
646  out:
647         if (found_path)
648                 free(found_path);
649         freeaddrinfo(ai);
650         free(dom);
651         xlog(D_CALL, "nfsd_fh: found %p path %s", found, found ? found->e_path : NULL);
652         return;         
653 }
654
655 static void write_fsloc(FILE *f, struct exportent *ep)
656 {
657         struct servers *servers;
658
659         if (ep->e_fslocmethod == FSLOC_NONE)
660                 return;
661
662         servers = replicas_lookup(ep->e_fslocmethod, ep->e_fslocdata);
663         if (!servers)
664                 return;
665         qword_print(f, "fsloc");
666         qword_printint(f, servers->h_num);
667         if (servers->h_num >= 0) {
668                 int i;
669                 for (i=0; i<servers->h_num; i++) {
670                         qword_print(f, servers->h_mp[i]->h_host);
671                         qword_print(f, servers->h_mp[i]->h_path);
672                 }
673         }
674         qword_printint(f, servers->h_referral);
675         release_replicas(servers);
676 }
677
678 static void write_secinfo(FILE *f, struct exportent *ep, int flag_mask)
679 {
680         struct sec_entry *p;
681
682         for (p = ep->e_secinfo; p->flav; p++)
683                 ; /* Do nothing */
684         if (p == ep->e_secinfo) {
685                 /* There was no sec= option */
686                 return;
687         }
688         qword_print(f, "secinfo");
689         qword_printint(f, p - ep->e_secinfo);
690         for (p = ep->e_secinfo; p->flav; p++) {
691                 qword_printint(f, p->flav->fnum);
692                 qword_printint(f, p->flags & flag_mask);
693         }
694
695 }
696
697 static int dump_to_cache(FILE *f, char *domain, char *path, struct exportent *exp)
698 {
699         qword_print(f, domain);
700         qword_print(f, path);
701         if (exp) {
702                 int different_fs = strcmp(path, exp->e_path) != 0;
703                 int flag_mask = different_fs ? ~NFSEXP_FSID : ~0;
704
705                 qword_printuint(f, time(0) + exp->e_ttl);
706                 qword_printint(f, exp->e_flags & flag_mask);
707                 qword_printint(f, exp->e_anonuid);
708                 qword_printint(f, exp->e_anongid);
709                 qword_printint(f, exp->e_fsid);
710                 write_fsloc(f, exp);
711                 write_secinfo(f, exp, flag_mask);
712                 if (exp->e_uuid == NULL || different_fs) {
713                         char u[16];
714                         if (uuid_by_path(path, 0, 16, u)) {
715                                 qword_print(f, "uuid");
716                                 qword_printhex(f, u, 16);
717                         }
718                 } else {
719                         char u[16];
720                         get_uuid(exp->e_uuid, 16, u);
721                         qword_print(f, "uuid");
722                         qword_printhex(f, u, 16);
723                 }
724         } else
725                 qword_printuint(f, time(0) + DEFAULT_TTL);
726         return qword_eol(f);
727 }
728
729 static int is_subdirectory(char *child, char *parent)
730 {
731         int l = strlen(parent);
732
733         return strcmp(child, parent) == 0
734                 || (strncmp(child, parent, l) == 0 && child[l] == '/');
735 }
736
737 static int path_matches(nfs_export *exp, char *path)
738 {
739         if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)
740                 return is_subdirectory(path, exp->m_export.e_path);
741         return strcmp(path, exp->m_export.e_path) == 0;
742 }
743
744 static int
745 client_matches(nfs_export *exp, char *dom, struct addrinfo *ai)
746 {
747         if (use_ipaddr)
748                 return client_check(exp->m_client, ai);
749         return client_member(dom, exp->m_client->m_hostname);
750 }
751
752 static int
753 export_matches(nfs_export *exp, char *dom, char *path, struct addrinfo *ai)
754 {
755         return path_matches(exp, path) && client_matches(exp, dom, ai);
756 }
757
758 static nfs_export *
759 lookup_export(char *dom, char *path, struct addrinfo *ai)
760 {
761         nfs_export *exp;
762         nfs_export *found = NULL;
763         int found_type = 0;
764         int i;
765
766         for (i=0 ; i < MCL_MAXTYPES; i++) {
767                 for (exp = exportlist[i].p_head; exp; exp = exp->m_next) {
768                         if (!export_matches(exp, dom, path, ai))
769                                 continue;
770                         if (!found) {
771                                 found = exp;
772                                 found_type = i;
773                                 continue;
774                         }
775                         /* Always prefer non-V4ROOT exports */
776                         if (exp->m_export.e_flags & NFSEXP_V4ROOT)
777                                 continue;
778                         if (found->m_export.e_flags & NFSEXP_V4ROOT) {
779                                 found = exp;
780                                 found_type = i;
781                                 continue;
782                         }
783
784                         /* If one is a CROSSMOUNT, then prefer the longest path */
785                         if (((found->m_export.e_flags & NFSEXP_CROSSMOUNT) ||
786                              (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)) &&
787                             strlen(found->m_export.e_path) !=
788                             strlen(exp->m_export.e_path)) {
789
790                                 if (strlen(exp->m_export.e_path) >
791                                     strlen(found->m_export.e_path)) {
792                                         found = exp;
793                                         found_type = i;
794                                 }
795                                 continue;
796
797                         } else if (found_type == i && found->m_warned == 0) {
798                                 xlog(L_WARNING, "%s exported to both %s and %s, "
799                                      "arbitrarily choosing options from first",
800                                      path, found->m_client->m_hostname, exp->m_client->m_hostname,
801                                      dom);
802                                 found->m_warned = 1;
803                         }
804                 }
805         }
806         return found;
807 }
808
809 #ifdef HAVE_NFS_PLUGIN_H
810 #include <dlfcn.h>
811 #include <nfs-plugin.h>
812
813 /*
814  * Walk through a set of FS locations and build a set of export options.
815  * Returns true if all went to plan; otherwise, false.
816  */
817 static _Bool
818 locations_to_options(struct jp_ops *ops, nfs_fsloc_set_t locations,
819                 char *options, size_t remaining, int *ttl)
820 {
821         char *server, *last_path, *rootpath, *ptr;
822         _Bool seen = false;
823
824         last_path = NULL;
825         rootpath = NULL;
826         server = NULL;
827         ptr = options;
828         *ttl = 0;
829
830         for (;;) {
831                 enum jp_status status;
832                 int len;
833
834                 status = ops->jp_get_next_location(locations, &server,
835                                                         &rootpath, ttl);
836                 if (status == JP_EMPTY)
837                         break;
838                 if (status != JP_OK) {
839                         xlog(D_GENERAL, "%s: failed to parse location: %s",
840                                 __func__, ops->jp_error(status));
841                         goto out_false;
842                 }
843                 xlog(D_GENERAL, "%s: Location: %s:%s",
844                         __func__, server, rootpath);
845
846                 if (last_path && strcmp(rootpath, last_path) == 0) {
847                         len = snprintf(ptr, remaining, "+%s", server);
848                         if (len < 0) {
849                                 xlog(D_GENERAL, "%s: snprintf: %m", __func__);
850                                 goto out_false;
851                         }
852                         if ((size_t)len >= remaining) {
853                                 xlog(D_GENERAL, "%s: options buffer overflow", __func__);
854                                 goto out_false;
855                         }
856                         remaining -= (size_t)len;
857                         ptr += len;
858                 } else {
859                         if (last_path == NULL)
860                                 len = snprintf(ptr, remaining, "refer=%s@%s",
861                                                         rootpath, server);
862                         else
863                                 len = snprintf(ptr, remaining, ":%s@%s",
864                                                         rootpath, server);
865                         if (len < 0) {
866                                 xlog(D_GENERAL, "%s: snprintf: %m", __func__);
867                                 goto out_false;
868                         }
869                         if ((size_t)len >= remaining) {
870                                 xlog(D_GENERAL, "%s: options buffer overflow",
871                                         __func__);
872                                 goto out_false;
873                         }
874                         remaining -= (size_t)len;
875                         ptr += len;
876                         last_path = rootpath;
877                 }
878
879                 seen = true;
880                 free(rootpath);
881                 free(server);
882         }
883
884         xlog(D_CALL, "%s: options='%s', ttl=%d",
885                 __func__, options, *ttl);
886         return seen;
887
888 out_false:
889         free(rootpath);
890         free(server);
891         return false;
892 }
893
894 /*
895  * Walk through the set of FS locations and build an exportent.
896  * Returns pointer to an exportent if "junction" refers to a junction.
897  *
898  * Returned exportent points to static memory.
899  */
900 static struct exportent *do_locations_to_export(struct jp_ops *ops,
901                 nfs_fsloc_set_t locations, const char *junction,
902                 char *options, size_t options_len)
903 {
904         struct exportent *exp;
905         int ttl;
906
907         if (!locations_to_options(ops, locations, options, options_len, &ttl))
908                 return NULL;
909
910         exp = mkexportent("*", (char *)junction, options);
911         if (exp == NULL) {
912                 xlog(L_ERROR, "%s: Failed to construct exportent", __func__);
913                 return NULL;
914         }
915
916         exp->e_uuid = NULL;
917         exp->e_ttl = ttl;
918         return exp;
919 }
920
921 /*
922  * Convert set of FS locations to an exportent.  Returns pointer to
923  * an exportent if "junction" refers to a junction.
924  *
925  * Returned exportent points to static memory.
926  */
927 static struct exportent *locations_to_export(struct jp_ops *ops,
928                 nfs_fsloc_set_t locations, const char *junction)
929 {
930         struct exportent *exp;
931         char *options;
932
933         options = malloc(BUFSIZ);
934         if (options == NULL) {
935                 xlog(D_GENERAL, "%s: failed to allocate options buffer",
936                         __func__);
937                 return NULL;
938         }
939         options[0] = '\0';
940
941         exp = do_locations_to_export(ops, locations, junction,
942                                                 options, BUFSIZ);
943
944         free(options);
945         return exp;
946 }
947
948 /*
949  * Retrieve locations information in "junction" and dump it to the
950  * kernel.  Returns pointer to an exportent if "junction" refers
951  * to a junction.
952  *
953  * Returned exportent points to static memory.
954  */
955 static struct exportent *invoke_junction_ops(void *handle,
956                 const char *junction)
957 {
958         nfs_fsloc_set_t locations;
959         struct exportent *exp;
960         enum jp_status status;
961         struct jp_ops *ops;
962         char *error;
963
964         ops = (struct jp_ops *)dlsym(handle, "nfs_junction_ops");
965         error = dlerror();
966         if (error != NULL) {
967                 xlog(D_GENERAL, "%s: dlsym(jp_junction_ops): %s",
968                         __func__, error);
969                 return NULL;
970         }
971         if (ops->jp_api_version != JP_API_VERSION) {
972                 xlog(D_GENERAL, "%s: unrecognized junction API version: %u",
973                         __func__, ops->jp_api_version);
974                 return NULL;
975         }
976
977         status = ops->jp_init(false);
978         if (status != JP_OK) {
979                 xlog(D_GENERAL, "%s: failed to resolve %s: %s",
980                         __func__, junction, ops->jp_error(status));
981                 return NULL;
982         }
983
984         status = ops->jp_get_locations(junction, &locations);
985         if (status != JP_OK) {
986                 xlog(D_GENERAL, "%s: failed to resolve %s: %s",
987                         __func__, junction, ops->jp_error(status));
988                 return NULL;
989         }
990
991         exp = locations_to_export(ops, locations, junction);
992
993         ops->jp_put_locations(locations);
994         ops->jp_done();
995         return exp;
996 }
997
998 /*
999  * Load the junction plug-in, then try to resolve "pathname".
1000  * Returns pointer to an initialized exportent if "junction"
1001  * refers to a junction, or NULL if not.
1002  *
1003  * Returned exportent points to static memory.
1004  */
1005 static struct exportent *lookup_junction(const char *pathname)
1006 {
1007         struct exportent *exp;
1008         void *handle;
1009
1010         handle = dlopen("libnfsjunct.so", RTLD_NOW);
1011         if (handle == NULL) {
1012                 xlog(D_GENERAL, "%s: dlopen: %s", __func__, dlerror());
1013                 return NULL;
1014         }
1015         (void)dlerror();        /* Clear any error */
1016
1017         exp = invoke_junction_ops(handle, pathname);
1018
1019         /* We could leave it loaded to make junction resolution
1020          * faster next time.  However, if we want to replace the
1021          * library, that would require restarting mountd. */
1022         (void)dlclose(handle);
1023         return exp;
1024 }
1025 #else   /* !HAVE_NFS_PLUGIN_H */
1026 static inline struct exportent *lookup_junction(const char *UNUSED(pathname))
1027 {
1028         return NULL;
1029 }
1030 #endif  /* !HAVE_NFS_PLUGIN_H */
1031
1032 static void nfsd_export(FILE *f)
1033 {
1034         /* requests are:
1035          *  domain path
1036          * determine export options and return:
1037          *  domain path expiry flags anonuid anongid fsid
1038          */
1039
1040         char *cp;
1041         char *dom, *path;
1042         nfs_export *found = NULL;
1043         struct addrinfo *ai = NULL;
1044
1045         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
1046                 return;
1047
1048         xlog(D_CALL, "nfsd_export: inbuf '%s'", lbuf);
1049
1050         cp = lbuf;
1051         dom = malloc(strlen(cp));
1052         path = malloc(strlen(cp));
1053
1054         if (!dom || !path)
1055                 goto out;
1056
1057         if (qword_get(&cp, dom, strlen(lbuf)) <= 0)
1058                 goto out;
1059         if (qword_get(&cp, path, strlen(lbuf)) <= 0)
1060                 goto out;
1061
1062         auth_reload();
1063
1064         if (use_ipaddr) {
1065                 struct addrinfo *tmp;
1066                 tmp = host_pton(dom);
1067                 if (tmp == NULL)
1068                         goto out;
1069                 ai = client_resolve(tmp->ai_addr);
1070                 freeaddrinfo(tmp);
1071                         goto out;
1072         }
1073
1074         found = lookup_export(dom, path, ai);
1075
1076         if (found) {
1077                 if (dump_to_cache(f, dom, path, &found->m_export) < 0) {
1078                         xlog(L_WARNING,
1079                              "Cannot export %s, possibly unsupported filesystem"
1080                              " or fsid= required", path);
1081                         dump_to_cache(f, dom, path, NULL);
1082                 }
1083         } else {
1084                 dump_to_cache(f, dom, path, lookup_junction(path));
1085         }
1086  out:
1087         xlog(D_CALL, "nfsd_export: found %p path %s", found, path ? path : NULL);
1088         if (dom) free(dom);
1089         if (path) free(path);
1090         freeaddrinfo(ai);
1091 }
1092
1093
1094 struct {
1095         char *cache_name;
1096         void (*cache_handle)(FILE *f);
1097         FILE *f;
1098         char vbuf[RPC_CHAN_BUF_SIZE];
1099 } cachelist[] = {
1100         { "auth.unix.ip", auth_unix_ip, NULL, ""},
1101         { "auth.unix.gid", auth_unix_gid, NULL, ""},
1102         { "nfsd.export", nfsd_export, NULL, ""},
1103         { "nfsd.fh", nfsd_fh, NULL, ""},
1104         { NULL, NULL, NULL, ""}
1105 };
1106
1107 extern int manage_gids;
1108
1109 /**
1110  * cache_open - prepare communications channels with kernel RPC caches
1111  *
1112  */
1113 void cache_open(void) 
1114 {
1115         int i;
1116         for (i=0; cachelist[i].cache_name; i++ ) {
1117                 char path[100];
1118                 if (!manage_gids && cachelist[i].cache_handle == auth_unix_gid)
1119                         continue;
1120                 sprintf(path, "/proc/net/rpc/%s/channel", cachelist[i].cache_name);
1121                 cachelist[i].f = fopen(path, "r+");
1122                 if (cachelist[i].f != NULL) {
1123                         setvbuf(cachelist[i].f, cachelist[i].vbuf, _IOLBF, 
1124                                 RPC_CHAN_BUF_SIZE);
1125                 }
1126         }
1127 }
1128
1129 /**
1130  * cache_set_fds - prepare cache file descriptors for one iteration of the service loop
1131  * @fdset: pointer to fd_set to prepare
1132  */
1133 void cache_set_fds(fd_set *fdset)
1134 {
1135         int i;
1136         for (i=0; cachelist[i].cache_name; i++) {
1137                 if (cachelist[i].f)
1138                         FD_SET(fileno(cachelist[i].f), fdset);
1139         }
1140 }
1141
1142 /**
1143  * cache_process_req - process any active cache file descriptors during service loop iteration
1144  * @fdset: pointer to fd_set to examine for activity
1145  */
1146 int cache_process_req(fd_set *readfds) 
1147 {
1148         int i;
1149         int cnt = 0;
1150         for (i=0; cachelist[i].cache_name; i++) {
1151                 if (cachelist[i].f != NULL &&
1152                     FD_ISSET(fileno(cachelist[i].f), readfds)) {
1153                         cnt++;
1154                         cachelist[i].cache_handle(cachelist[i].f);
1155                         FD_CLR(fileno(cachelist[i].f), readfds);
1156                 }
1157         }
1158         return cnt;
1159 }
1160
1161
1162 /*
1163  * Give IP->domain and domain+path->options to kernel
1164  * % echo nfsd $IP  $[now+DEFAULT_TTL] $domain > /proc/net/rpc/auth.unix.ip/channel
1165  * % echo $domain $path $[now+DEFAULT_TTL] $options $anonuid $anongid $fsid > /proc/net/rpc/nfsd.export/channel
1166  */
1167
1168 static int cache_export_ent(char *domain, struct exportent *exp, char *path)
1169 {
1170         int err;
1171         FILE *f = fopen("/proc/net/rpc/nfsd.export/channel", "w");
1172         if (!f)
1173                 return -1;
1174
1175         err = dump_to_cache(f, domain, exp->e_path, exp);
1176         if (err) {
1177                 xlog(L_WARNING,
1178                      "Cannot export %s, possibly unsupported filesystem or"
1179                      " fsid= required", exp->e_path);
1180         }
1181
1182         while (err == 0 && (exp->e_flags & NFSEXP_CROSSMOUNT) && path) {
1183                 /* really an 'if', but we can break out of
1184                  * a 'while' more easily */
1185                 /* Look along 'path' for other filesystems
1186                  * and export them with the same options
1187                  */
1188                 struct stat stb;
1189                 size_t l = strlen(exp->e_path);
1190                 __dev_t dev;
1191
1192                 if (strlen(path) <= l || path[l] != '/' ||
1193                     strncmp(exp->e_path, path, l) != 0)
1194                         break;
1195                 if (stat(exp->e_path, &stb) != 0)
1196                         break;
1197                 dev = stb.st_dev;
1198                 while(path[l] == '/') {
1199                         char c;
1200                         /* errors for submount should fail whole filesystem */
1201                         int err2;
1202
1203                         l++;
1204                         while (path[l] != '/' && path[l])
1205                                 l++;
1206                         c = path[l];
1207                         path[l] = 0;
1208                         err2 = lstat(path, &stb);
1209                         path[l] = c;
1210                         if (err2 < 0)
1211                                 break;
1212                         if (stb.st_dev == dev)
1213                                 continue;
1214                         dev = stb.st_dev;
1215                         path[l] = 0;
1216                         dump_to_cache(f, domain, path, exp);
1217                         path[l] = c;
1218                 }
1219                 break;
1220         }
1221
1222         fclose(f);
1223         return err;
1224 }
1225
1226 /**
1227  * cache_export - Inform kernel of a new nfs_export
1228  * @exp: target nfs_export
1229  * @path: NUL-terminated C string containing export path
1230  */
1231 int cache_export(nfs_export *exp, char *path)
1232 {
1233         char buf[INET6_ADDRSTRLEN];
1234         int err;
1235         FILE *f;
1236
1237         f = fopen("/proc/net/rpc/auth.unix.ip/channel", "w");
1238         if (!f)
1239                 return -1;
1240
1241
1242         qword_print(f, "nfsd");
1243         qword_print(f,
1244                 host_ntop(get_addrlist(exp->m_client, 0), buf, sizeof(buf)));
1245         qword_printuint(f, time(0) + exp->m_export.e_ttl);
1246         qword_print(f, exp->m_client->m_hostname);
1247         err = qword_eol(f);
1248         
1249         fclose(f);
1250
1251         err = cache_export_ent(exp->m_client->m_hostname, &exp->m_export, path)
1252                 || err;
1253         return err;
1254 }
1255
1256 /**
1257  * cache_get_filehandle - given an nfs_export, get its root filehandle
1258  * @exp: target nfs_export
1259  * @len: length of requested file handle
1260  * @p: NUL-terminated C string containing export path
1261  *
1262  * Returns pointer to NFS file handle of root directory of export
1263  *
1264  * { 
1265  *   echo $domain $path $length 
1266  *   read filehandle <&0
1267  * } <> /proc/fs/nfsd/filehandle
1268  */
1269 struct nfs_fh_len *
1270 cache_get_filehandle(nfs_export *exp, int len, char *p)
1271 {
1272         FILE *f = fopen("/proc/fs/nfsd/filehandle", "r+");
1273         char buf[200];
1274         char *bp = buf;
1275         int failed;
1276         static struct nfs_fh_len fh;
1277
1278         if (!f)
1279                 f = fopen("/proc/fs/nfs/filehandle", "r+");
1280         if (!f)
1281                 return NULL;
1282
1283         qword_print(f, exp->m_client->m_hostname);
1284         qword_print(f, p);
1285         qword_printint(f, len); 
1286         failed = qword_eol(f);
1287         
1288         if (!failed)
1289                 failed = (fgets(buf, sizeof(buf), f) == NULL);
1290         fclose(f);
1291         if (failed)
1292                 return NULL;
1293         memset(fh.fh_handle, 0, sizeof(fh.fh_handle));
1294         fh.fh_size = qword_get(&bp, (char *)fh.fh_handle, NFS3_FHSIZE);
1295         return &fh;
1296 }