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         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
  88                 return;
  89
  90         xlog(D_CALL, "auth_unix_ip: inbuf '%s'", lbuf);
  91
  92         cp = lbuf;
  93
  94         if (qword_get(&cp, class, 20) <= 0 ||
  95             strcmp(class, "nfsd") != 0)
  96                 return;
  97
  98         if (qword_get(&cp, ipaddr, sizeof(ipaddr)) <= 0)
  99                 return;
 100
 101         tmp = host_pton(ipaddr);
 102         if (tmp == NULL)
 103                 return;
 104
 105         auth_reload();
 106
 107         /* addr is a valid, interesting address, find the domain name... */
 108         if (!use_ipaddr) {
 109                 struct addrinfo *ai = NULL;
 110
 111                 ai = client_resolve(tmp->ai_addr);
 112                 if (ai) {
 113                         client = client_compose(ai);
 114                         freeaddrinfo(ai);
 115                 }
 116         }
 117         qword_print(f, "nfsd");
 118         qword_print(f, ipaddr);
 119         qword_printtimefrom(f, DEFAULT_TTL);
 120         if (use_ipaddr)
 121                 qword_print(f, ipaddr);
 122         else if (client)
 123                 qword_print(f, *client?client:"DEFAULT");
 124         qword_eol(f);
 125         xlog(D_CALL, "auth_unix_ip: client %p '%s'", client, client?client: "DEFAULT");
 126
 127         free(client);
 128         freeaddrinfo(tmp);
 129
 130 }
 131
 132 static void auth_unix_gid(FILE *f)
 133 {
 134         /* Request are
 135          *  uid
 136          * reply is
 137          *  uid expiry count list of group ids
 138          */
 139         uid_t uid;
 140         struct passwd *pw;
 141         static gid_t *groups = NULL;
 142         static int groups_len = 0;
 143         gid_t *more_groups;
 144         int ngroups;
 145         int rv, i;
 146         char *cp;
 147
 148         if (groups_len == 0) {
 149                 groups = malloc(sizeof(gid_t) * INITIAL_MANAGED_GROUPS);
 150                 if (!groups)
 151                         return;
 152
 153                 groups_len = INITIAL_MANAGED_GROUPS;
 154         }
 155
 156         ngroups = groups_len;
 157
 158         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
 159                 return;
 160
 161         cp = lbuf;
 162         if (qword_get_uint(&cp, &uid) != 0)
 163                 return;
 164
 165         pw = getpwuid(uid);
 166         if (!pw)
 167                 rv = -1;
 168         else {
 169                 rv = getgrouplist(pw->pw_name, pw->pw_gid, groups, &ngroups);
 170                 if (rv == -1 && ngroups >= groups_len) {
 171                         more_groups = realloc(groups, sizeof(gid_t)*ngroups);
 172                         if (!more_groups)
 173                                 rv = -1;
 174                         else {
 175                                 groups = more_groups;
 176                                 groups_len = ngroups;
 177                                 rv = getgrouplist(pw->pw_name, pw->pw_gid,
 178                                                   groups, &ngroups);
 179                         }
 180                 }
 181         }
 182         qword_printuint(f, uid);
 183         qword_printtimefrom(f, DEFAULT_TTL);
 184         if (rv >= 0) {
 185                 qword_printuint(f, ngroups);
 186                 for (i=0; i<ngroups; i++)
 187                         qword_printuint(f, groups[i]);
 188         } else
 189                 qword_printuint(f, 0);
 190         qword_eol(f);
 191 }
 192
 193 #if USE_BLKID
 194 static const char *get_uuid_blkdev(char *path)
 195 {
 196         /* We set *safe if we know that we need the
 197          * fsid from statfs too.
 198          */
 199         static blkid_cache cache = NULL;
 200         struct stat stb;
 201         char *devname;
 202         blkid_tag_iterate iter;
 203         blkid_dev dev;
 204         const char *type;
 205         const char *val, *uuid = NULL;
 206
 207         if (cache == NULL)
 208                 blkid_get_cache(&cache, NULL);
 209
 210         if (stat(path, &stb) != 0)
 211                 return NULL;
 212         devname = blkid_devno_to_devname(stb.st_dev);
 213         if (!devname)
 214                 return NULL;
 215         dev = blkid_get_dev(cache, devname, BLKID_DEV_NORMAL);
 216         free(devname);
 217         if (!dev)
 218                 return NULL;
 219         iter = blkid_tag_iterate_begin(dev);
 220         if (!iter)
 221                 return NULL;
 222         while (blkid_tag_next(iter, &type, &val) == 0) {
 223                 if (strcmp(type, "UUID") == 0)
 224                         uuid = val;
 225                 if (strcmp(type, "TYPE") == 0 &&
 226                     strcmp(val, "btrfs") == 0) {
 227                         uuid = NULL;
 228                         break;
 229                 }
 230         }
 231         blkid_tag_iterate_end(iter);
 232         return uuid;
 233 }
 234 #else
 235 #define get_uuid_blkdev(path) (NULL)
 236 #endif
 237
 238 static int get_uuid(const char *val, size_t uuidlen, char *u)
 239 {
 240         /* extract hex digits from uuidstr and compose a uuid
 241          * of the given length (max 16), xoring bytes to make
 242          * a smaller uuid.
 243          */
 244         size_t i = 0;
 245
 246         memset(u, 0, uuidlen);
 247         for ( ; *val ; val++) {
 248                 int c = *val;
 249                 if (!isxdigit(c))
 250                         continue;
 251                 if (isalpha(c)) {
 252                         if (isupper(c))
 253                                 c = c - 'A' + 10;
 254                         else
 255                                 c = c - 'a' + 10;
 256                 } else
 257                         c = c - '0' + 0;
 258                 if ((i&1) == 0)
 259                         c <<= 4;
 260                 u[i/2] ^= (char)c;
 261                 i++;
 262                 if (i == uuidlen*2)
 263                         i = 0;
 264         }
 265         return 1;
 266 }
 267
 268 static int uuid_by_path(char *path, int type, size_t uuidlen, char *uuid)
 269 {
 270         /* get a uuid for the filesystem found at 'path'.
 271          * There are several possible ways of generating the
 272          * uuids (types).
 273          * Type 0 is used for new filehandles, while other types
 274          * may be used to interpret old filehandle - to ensure smooth
 275          * forward migration.
 276          * We return 1 if a uuid was found (and it might be worth
 277          * trying the next type) or 0 if no more uuid types can be
 278          * extracted.
 279          */
 280
 281         /* Possible sources of uuid are
 282          * - blkid uuid
 283          * - statfs64 uuid
 284          *
 285          * On some filesystems (e.g. vfat) the statfs64 uuid is simply an
 286          * encoding of the device that the filesystem is mounted from, so
 287          * it we be very bad to use that (as device numbers change).  blkid
 288          * must be preferred.
 289          * On other filesystems (e.g. btrfs) the statfs64 uuid contains
 290          * important info that the blkid uuid cannot contain:  This happens
 291          * when multiple subvolumes are exported (they have the same
 292          * blkid uuid but different statfs64 uuids).
 293          * We rely on get_uuid_blkdev *knowing* which is which and not returning
 294          * a uuid for filesystems where the statfs64 uuid is better.
 295          *
 296          */
 297         struct statfs64 st;
 298         char fsid_val[17];
 299         const char *blkid_val;
 300         const char *val;
 301
 302         blkid_val = get_uuid_blkdev(path);
 303
 304         if (statfs64(path, &st) == 0 &&
 305             (st.f_fsid.__val[0] || st.f_fsid.__val[1]))
 306                 snprintf(fsid_val, 17, "%08x%08x",
 307                          st.f_fsid.__val[0], st.f_fsid.__val[1]);
 308         else
 309                 fsid_val[0] = 0;
 310
 311         if (blkid_val && (type--) == 0)
 312                 val = blkid_val;
 313         else if (fsid_val[0] && (type--) == 0)
 314                 val = fsid_val;
 315         else
 316                 return 0;
 317
 318         get_uuid(val, uuidlen, uuid);
 319         return 1;
 320 }
 321
 322 /* Iterate through /etc/mtab, finding mountpoints
 323  * at or below a given path
 324  */
 325 static char *next_mnt(void **v, char *p)
 326 {
 327         FILE *f;
 328         struct mntent *me;
 329         size_t l = strlen(p);
 330         if (*v == NULL) {
 331                 f = setmntent("/etc/mtab", "r");
 332                 *v = f;
 333         } else
 334                 f = *v;
 335         while ((me = getmntent(f)) != NULL &&
 336                (strncmp(me->mnt_dir, p, l) != 0 ||
 337                 me->mnt_dir[l] != '/'))
 338                 ;
 339         if (me == NULL) {
 340                 endmntent(f);
 341                 *v = NULL;
 342                 return NULL;
 343         }
 344         return me->mnt_dir;
 345 }
 346
 347 /* True iff e1 is a child of e2 and e2 has crossmnt set: */
 348 static bool subexport(struct exportent *e1, struct exportent *e2)
 349 {
 350         char *p1 = e1->e_path, *p2 = e2->e_path;
 351         size_t l2 = strlen(p2);
 352
 353         return e2->e_flags & NFSEXP_CROSSMOUNT
 354                && strncmp(p1, p2, l2) == 0
 355                && p1[l2] == '/';
 356 }
 357
 358 struct parsed_fsid {
 359         int fsidtype;
 360         /* We could use a union for this, but it would be more
 361          * complicated; why bother? */
 362         unsigned int inode;
 363         unsigned int minor;
 364         unsigned int major;
 365         unsigned int fsidnum;
 366         size_t uuidlen;
 367         char *fhuuid;
 368 };
 369
 370 static int parse_fsid(int fsidtype, int fsidlen, char *fsid,
 371                 struct parsed_fsid *parsed)
 372 {
 373         unsigned int dev;
 374         unsigned long long inode64;
 375
 376         memset(parsed, 0, sizeof(*parsed));
 377         parsed->fsidtype = fsidtype;
 378         switch(fsidtype) {
 379         case FSID_DEV: /* 4 bytes: 2 major, 2 minor, 4 inode */
 380                 if (fsidlen != 8)
 381                         return -1;
 382                 memcpy(&dev, fsid, 4);
 383                 memcpy(&parsed->inode, fsid+4, 4);
 384                 parsed->major = ntohl(dev)>>16;
 385                 parsed->minor = ntohl(dev) & 0xFFFF;
 386                 break;
 387
 388         case FSID_NUM: /* 4 bytes - fsid */
 389                 if (fsidlen != 4)
 390                         return -1;
 391                 memcpy(&parsed->fsidnum, fsid, 4);
 392                 break;
 393
 394         case FSID_MAJOR_MINOR: /* 12 bytes: 4 major, 4 minor, 4 inode
 395                  * This format is never actually used but was
 396                  * an historical accident
 397                  */
 398                 if (fsidlen != 12)
 399                         return -1;
 400                 memcpy(&dev, fsid, 4);
 401                 parsed->major = ntohl(dev);
 402                 memcpy(&dev, fsid+4, 4);
 403                 parsed->minor = ntohl(dev);
 404                 memcpy(&parsed->inode, fsid+8, 4);
 405                 break;
 406
 407         case FSID_ENCODE_DEV: /* 8 bytes: 4 byte packed device number, 4 inode */
 408                 /* This is *host* endian, not net-byte-order, because
 409                  * no-one outside this host has any business interpreting it
 410                  */
 411                 if (fsidlen != 8)
 412                         return -1;
 413                 memcpy(&dev, fsid, 4);
 414                 memcpy(&parsed->inode, fsid+4, 4);
 415                 parsed->major = (dev & 0xfff00) >> 8;
 416                 parsed->minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
 417                 break;
 418
 419         case FSID_UUID4_INUM: /* 4 byte inode number and 4 byte uuid */
 420                 if (fsidlen != 8)
 421                         return -1;
 422                 memcpy(&parsed->inode, fsid, 4);
 423                 parsed->uuidlen = 4;
 424                 parsed->fhuuid = fsid+4;
 425                 break;
 426         case FSID_UUID8: /* 8 byte uuid */
 427                 if (fsidlen != 8)
 428                         return -1;
 429                 parsed->uuidlen = 8;
 430                 parsed->fhuuid = fsid;
 431                 break;
 432         case FSID_UUID16: /* 16 byte uuid */
 433                 if (fsidlen != 16)
 434                         return -1;
 435                 parsed->uuidlen = 16;
 436                 parsed->fhuuid = fsid;
 437                 break;
 438         case FSID_UUID16_INUM: /* 8 byte inode number and 16 byte uuid */
 439                 if (fsidlen != 24)
 440                         return -1;
 441                 memcpy(&inode64, fsid, 8);
 442                 parsed->inode = inode64;
 443                 parsed->uuidlen = 16;
 444                 parsed->fhuuid = fsid+8;
 445                 break;
 446         }
 447         return 0;
 448 }
 449
 450 static bool match_fsid(struct parsed_fsid *parsed, nfs_export *exp, char *path)
 451 {
 452         struct stat stb;
 453         int type;
 454         char u[16];
 455
 456         if (stat(path, &stb) != 0)
 457                 return false;
 458         if (!S_ISDIR(stb.st_mode) && !S_ISREG(stb.st_mode))
 459                 return false;
 460
 461         switch (parsed->fsidtype) {
 462         case FSID_DEV:
 463         case FSID_MAJOR_MINOR:
 464         case FSID_ENCODE_DEV:
 465                 if (stb.st_ino != parsed->inode)
 466                         return false;
 467                 if (parsed->major != major(stb.st_dev) ||
 468                     parsed->minor != minor(stb.st_dev))
 469                         return false;
 470                 return true;
 471         case FSID_NUM:
 472                 if (((exp->m_export.e_flags & NFSEXP_FSID) == 0 ||
 473                      exp->m_export.e_fsid != parsed->fsidnum))
 474                         return false;
 475                 return true;
 476         case FSID_UUID4_INUM:
 477         case FSID_UUID16_INUM:
 478                 if (stb.st_ino != parsed->inode)
 479                         return false;
 480                 goto check_uuid;
 481         case FSID_UUID8:
 482         case FSID_UUID16:
 483                 if (!is_mountpoint(path))
 484                         return false;
 485         check_uuid:
 486                 if (exp->m_export.e_uuid)
 487                         get_uuid(exp->m_export.e_uuid, parsed->uuidlen, u);
 488                 else
 489                         for (type = 0;
 490                              uuid_by_path(path, type, parsed->uuidlen, u);
 491                              type++)
 492                                 if (memcmp(u, parsed->fhuuid, parsed->uuidlen) == 0)
 493                                         return true;
 494
 495                 if (memcmp(u, parsed->fhuuid, parsed->uuidlen) != 0)
 496                         return false;
 497                 return true;
 498         }
 499         /* Well, unreachable, actually: */
 500         return false;
 501 }
 502
 503 static struct addrinfo *lookup_client_addr(char *dom)
 504 {
 505         struct addrinfo *ret;
 506         struct addrinfo *tmp;
 507
 508         dom++; /* skip initial "$" */
 509
 510         tmp = host_pton(dom);
 511         if (tmp == NULL)
 512                 return NULL;
 513         ret = client_resolve(tmp->ai_addr);
 514         freeaddrinfo(tmp);
 515         return ret;
 516 }
 517
 518 static void nfsd_fh(FILE *f)
 519 {
 520         /* request are:
 521          *  domain fsidtype fsid
 522          * interpret fsid, find export point and options, and write:
 523          *  domain fsidtype fsid expiry path
 524          */
 525         char *cp;
 526         char *dom;
 527         int fsidtype;
 528         int fsidlen;
 529         char fsid[32];
 530         struct parsed_fsid parsed;
 531         struct exportent *found = NULL;
 532         struct addrinfo *ai = NULL;
 533         char *found_path = NULL;
 534         nfs_export *exp;
 535         int i;
 536         int dev_missing = 0;
 537
 538         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
 539                 return;
 540
 541         xlog(D_CALL, "nfsd_fh: inbuf '%s'", lbuf);
 542
 543         cp = lbuf;
 544
 545         dom = malloc(strlen(cp));
 546         if (dom == NULL)
 547                 return;
 548         if (qword_get(&cp, dom, strlen(cp)) <= 0)
 549                 goto out;
 550         if (qword_get_int(&cp, &fsidtype) != 0)
 551                 goto out;
 552         if (fsidtype < 0 || fsidtype > 7)
 553                 goto out; /* unknown type */
 554         if ((fsidlen = qword_get(&cp, fsid, 32)) <= 0)
 555                 goto out;
 556         if (parse_fsid(fsidtype, fsidlen, fsid, &parsed))
 557                 goto out;
 558
 559         auth_reload();
 560
 561         if (is_ipaddr_client(dom)) {
 562                 ai = lookup_client_addr(dom);
 563                 if (!ai)
 564                         goto out;
 565         }
 566
 567         /* Now determine export point for this fsid/domain */
 568         for (i=0 ; i < MCL_MAXTYPES; i++) {
 569                 nfs_export *next_exp;
 570                 for (exp = exportlist[i].p_head; exp; exp = next_exp) {
 571                         char *path;
 572
 573                         if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT) {
 574                                 static nfs_export *prev = NULL;
 575                                 static void *mnt = NULL;
 576
 577                                 if (prev == exp) {
 578                                         /* try a submount */
 579                                         path = next_mnt(&mnt, exp->m_export.e_path);
 580                                         if (!path) {
 581                                                 next_exp = exp->m_next;
 582                                                 prev = NULL;
 583                                                 continue;
 584                                         }
 585                                         next_exp = exp;
 586                                 } else {
 587                                         prev = exp;
 588                                         mnt = NULL;
 589                                         path = exp->m_export.e_path;
 590                                         next_exp = exp;
 591                                 }
 592                         } else {
 593                                 path = exp->m_export.e_path;
 594                                 next_exp = exp->m_next;
 595                         }
 596
 597                         if (!is_ipaddr_client(dom)
 598                                         && !namelist_client_matches(exp, dom))
 599                                 continue;
 600                         if (exp->m_export.e_mountpoint &&
 601                             !is_mountpoint(exp->m_export.e_mountpoint[0]?
 602                                            exp->m_export.e_mountpoint:
 603                                            exp->m_export.e_path))
 604                                 dev_missing ++;
 605
 606                         if (!match_fsid(&parsed, exp, path))
 607                                 continue;
 608                         if (is_ipaddr_client(dom)
 609                                         && !ipaddr_client_matches(exp, ai))
 610                                 continue;
 611                         if (!found || subexport(&exp->m_export, found)) {
 612                                 found = &exp->m_export;
 613                                 free(found_path);
 614                                 found_path = strdup(path);
 615                                 if (found_path == NULL)
 616                                         goto out;
 617                         } else if (strcmp(found->e_path, exp->m_export.e_path) != 0
 618                                    && !subexport(found, &exp->m_export))
 619                         {
 620                                 xlog(L_WARNING, "%s and %s have same filehandle for %s, using first",
 621                                      found_path, path, dom);
 622                         } else {
 623                                 /* same path, if one is V4ROOT, choose the other */
 624                                 if (found->e_flags & NFSEXP_V4ROOT) {
 625                                         found = &exp->m_export;
 626                                         free(found_path);
 627                                         found_path = strdup(path);
 628                                         if (found_path == NULL)
 629                                                 goto out;
 630                                 }
 631                         }
 632                 }
 633         }
 634         if (found &&
 635             found->e_mountpoint &&
 636             !is_mountpoint(found->e_mountpoint[0]?
 637                            found->e_mountpoint:
 638                            found->e_path)) {
 639                 /* Cannot export this yet
 640                  * should log a warning, but need to rate limit
 641                    xlog(L_WARNING, "%s not exported as %d not a mountpoint",
 642                    found->e_path, found->e_mountpoint);
 643                  */
 644                 /* FIXME we need to make sure we re-visit this later */
 645                 goto out;
 646         }
 647         if (!found && dev_missing) {
 648                 /* The missing dev could be what we want, so just be
 649                  * quite rather than returning stale yet
 650                  */
 651                 goto out;
 652         }
 653
 654         if (found)
 655                 if (cache_export_ent(dom, found, found_path) < 0)
 656                         found = 0;
 657
 658         qword_print(f, dom);
 659         qword_printint(f, fsidtype);
 660         qword_printhex(f, fsid, fsidlen);
 661         /* The fsid -> path lookup can be quite expensive as it
 662          * potentially stats and reads lots of devices, and some of those
 663          * might have spun-down.  The Answer is not likely to
 664          * change underneath us, and an 'exportfs -f' can always
 665          * remove this from the kernel, so use a really log
 666          * timeout.  Maybe this should be configurable on the command
 667          * line.
 668          */
 669         qword_printint(f, 0x7fffffff);
 670         if (found)
 671                 qword_print(f, found_path);
 672         qword_eol(f);
 673  out:
 674         if (found_path)
 675                 free(found_path);
 676         freeaddrinfo(ai);
 677         free(dom);
 678         xlog(D_CALL, "nfsd_fh: found %p path %s", found, found ? found->e_path : NULL);
 679         return;
 680 }
 681
 682 static void write_fsloc(FILE *f, struct exportent *ep)
 683 {
 684         struct servers *servers;
 685
 686         if (ep->e_fslocmethod == FSLOC_NONE)
 687                 return;
 688
 689         servers = replicas_lookup(ep->e_fslocmethod, ep->e_fslocdata);
 690         if (!servers)
 691                 return;
 692         qword_print(f, "fsloc");
 693         qword_printint(f, servers->h_num);
 694         if (servers->h_num >= 0) {
 695                 int i;
 696                 for (i=0; i<servers->h_num; i++) {
 697                         qword_print(f, servers->h_mp[i]->h_host);
 698                         qword_print(f, servers->h_mp[i]->h_path);
 699                 }
 700         }
 701         qword_printint(f, servers->h_referral);
 702         release_replicas(servers);
 703 }
 704
 705 static void write_secinfo(FILE *f, struct exportent *ep, int flag_mask)
 706 {
 707         struct sec_entry *p;
 708
 709         for (p = ep->e_secinfo; p->flav; p++)
 710                 ; /* Do nothing */
 711         if (p == ep->e_secinfo) {
 712                 /* There was no sec= option */
 713                 return;
 714         }
 715         qword_print(f, "secinfo");
 716         qword_printint(f, p - ep->e_secinfo);
 717         for (p = ep->e_secinfo; p->flav; p++) {
 718                 qword_printint(f, p->flav->fnum);
 719                 qword_printint(f, p->flags & flag_mask);
 720         }
 721
 722 }
 723
 724 static int dump_to_cache(FILE *f, char *domain, char *path, struct exportent *exp)
 725 {
 726         qword_print(f, domain);
 727         qword_print(f, path);
 728         if (exp) {
 729                 int different_fs = strcmp(path, exp->e_path) != 0;
 730                 int flag_mask = different_fs ? ~NFSEXP_FSID : ~0;
 731
 732                 qword_printtimefrom(f, exp->e_ttl);
 733                 qword_printint(f, exp->e_flags & flag_mask);
 734                 qword_printint(f, exp->e_anonuid);
 735                 qword_printint(f, exp->e_anongid);
 736                 qword_printint(f, exp->e_fsid);
 737                 write_fsloc(f, exp);
 738                 write_secinfo(f, exp, flag_mask);
 739                 if (exp->e_uuid == NULL || different_fs) {
 740                         char u[16];
 741                         if (uuid_by_path(path, 0, 16, u)) {
 742                                 qword_print(f, "uuid");
 743                                 qword_printhex(f, u, 16);
 744                         }
 745                 } else {
 746                         char u[16];
 747                         get_uuid(exp->e_uuid, 16, u);
 748                         qword_print(f, "uuid");
 749                         qword_printhex(f, u, 16);
 750                 }
 751         } else
 752                 qword_printtimefrom(f, DEFAULT_TTL);
 753         return qword_eol(f);
 754 }
 755
 756 static int is_subdirectory(char *child, char *parent)
 757 {
 758         size_t l = strlen(parent);
 759
 760         return strcmp(child, parent) == 0
 761                 || (strncmp(child, parent, l) == 0 && child[l] == '/');
 762 }
 763
 764 static int path_matches(nfs_export *exp, char *path)
 765 {
 766         if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)
 767                 return is_subdirectory(path, exp->m_export.e_path);
 768         return strcmp(path, exp->m_export.e_path) == 0;
 769 }
 770
 771 static int
 772 export_matches(nfs_export *exp, char *dom, char *path, struct addrinfo *ai)
 773 {
 774         return path_matches(exp, path) && client_matches(exp, dom, ai);
 775 }
 776
 777 static nfs_export *
 778 lookup_export(char *dom, char *path, struct addrinfo *ai)
 779 {
 780         nfs_export *exp;
 781         nfs_export *found = NULL;
 782         int found_type = 0;
 783         int i;
 784
 785         for (i=0 ; i < MCL_MAXTYPES; i++) {
 786                 for (exp = exportlist[i].p_head; exp; exp = exp->m_next) {
 787                         if (!export_matches(exp, dom, path, ai))
 788                                 continue;
 789                         if (!found) {
 790                                 found = exp;
 791                                 found_type = i;
 792                                 continue;
 793                         }
 794                         /* Always prefer non-V4ROOT exports */
 795                         if (exp->m_export.e_flags & NFSEXP_V4ROOT)
 796                                 continue;
 797                         if (found->m_export.e_flags & NFSEXP_V4ROOT) {
 798                                 found = exp;
 799                                 found_type = i;
 800                                 continue;
 801                         }
 802
 803                         /* If one is a CROSSMOUNT, then prefer the longest path */
 804                         if (((found->m_export.e_flags & NFSEXP_CROSSMOUNT) ||
 805                              (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)) &&
 806                             strlen(found->m_export.e_path) !=
 807                             strlen(exp->m_export.e_path)) {
 808
 809                                 if (strlen(exp->m_export.e_path) >
 810                                     strlen(found->m_export.e_path)) {
 811                                         found = exp;
 812                                         found_type = i;
 813                                 }
 814                                 continue;
 815
 816                         } else if (found_type == i && found->m_warned == 0) {
 817                                 xlog(L_WARNING, "%s exported to both %s and %s, "
 818                                      "arbitrarily choosing options from first",
 819                                      path, found->m_client->m_hostname, exp->m_client->m_hostname,
 820                                      dom);
 821                                 found->m_warned = 1;
 822                         }
 823                 }
 824         }
 825         return found;
 826 }
 827
 828 #ifdef HAVE_NFS_PLUGIN_H
 829 #include <dlfcn.h>
 830 #include <nfs-plugin.h>
 831
 832 /*
 833  * Find the export entry for the parent of "pathname".
 834  * Caller must not free returned exportent.
 835  */
 836 static struct exportent *lookup_parent_export(char *dom,
 837                 const char *pathname, struct addrinfo *ai)
 838 {
 839         char *parent, *slash;
 840         nfs_export *result;
 841
 842         parent = strdup(pathname);
 843         if (parent == NULL) {
 844                 xlog(D_GENERAL, "%s: failed to allocate parent path buffer",
 845                         __func__);
 846                 goto out_default;
 847         }
 848         xlog(D_CALL, "%s: pathname = '%s'", __func__, pathname);
 849
 850 again:
 851         /* shorten pathname by one component */
 852         slash = strrchr(parent, '/');
 853         if (slash == NULL) {
 854                 xlog(D_GENERAL, "%s: no slash found in pathname",
 855                         __func__);
 856                 goto out_default;
 857         }
 858         *slash = '\0';
 859
 860         if (strlen(parent) == 0) {
 861                 result = lookup_export(dom, "/", ai);
 862                 if (result == NULL) {
 863                         xlog(L_ERROR, "%s: no root export found.", __func__);
 864                         goto out_default;
 865                 }
 866                 goto out;
 867         }
 868
 869         result = lookup_export(dom, parent, ai);
 870         if (result == NULL) {
 871                 xlog(D_GENERAL, "%s: lookup_export(%s) found nothing",
 872                         __func__, parent);
 873                 goto again;
 874         }
 875
 876 out:
 877         xlog(D_CALL, "%s: found export for %s", __func__, parent);
 878         free(parent);
 879         return &result->m_export;
 880
 881 out_default:
 882         free(parent);
 883         return mkexportent("*", "/", "insecure");
 884 }
 885
 886 /*
 887  * Walk through a set of FS locations and build an e_fslocdata string.
 888  * Returns true if all went to plan; otherwise, false.
 889  */
 890 static bool locations_to_fslocdata(struct jp_ops *ops,
 891                 nfs_fsloc_set_t locations, char *fslocdata,
 892                 size_t remaining, int *ttl)
 893 {
 894         char *server, *last_path, *rootpath, *ptr;
 895         _Bool seen = false;
 896
 897         last_path = NULL;
 898         rootpath = NULL;
 899         server = NULL;
 900         ptr = fslocdata;
 901         *ttl = 0;
 902
 903         for (;;) {
 904                 enum jp_status status;
 905                 int len;
 906
 907                 status = ops->jp_get_next_location(locations, &server,
 908                                                         &rootpath, ttl);
 909                 if (status == JP_EMPTY)
 910                         break;
 911                 if (status != JP_OK) {
 912                         xlog(D_GENERAL, "%s: failed to parse location: %s",
 913                                 __func__, ops->jp_error(status));
 914                         goto out_false;
 915                 }
 916                 xlog(D_GENERAL, "%s: Location: %s:%s",
 917                         __func__, server, rootpath);
 918
 919                 if (last_path && strcmp(rootpath, last_path) == 0) {
 920                         len = snprintf(ptr, remaining, "+%s", server);
 921                         if (len < 0) {
 922                                 xlog(D_GENERAL, "%s: snprintf: %m", __func__);
 923                                 goto out_false;
 924                         }
 925                         if ((size_t)len >= remaining) {
 926                                 xlog(D_GENERAL, "%s: fslocdata buffer overflow", __func__);
 927                                 goto out_false;
 928                         }
 929                         remaining -= (size_t)len;
 930                         ptr += len;
 931                 } else {
 932                         if (last_path == NULL)
 933                                 len = snprintf(ptr, remaining, "%s@%s",
 934                                                         rootpath, server);
 935                         else
 936                                 len = snprintf(ptr, remaining, ":%s@%s",
 937                                                         rootpath, server);
 938                         if (len < 0) {
 939                                 xlog(D_GENERAL, "%s: snprintf: %m", __func__);
 940                                 goto out_false;
 941                         }
 942                         if ((size_t)len >= remaining) {
 943                                 xlog(D_GENERAL, "%s: fslocdata buffer overflow",
 944                                         __func__);
 945                                 goto out_false;
 946                         }
 947                         remaining -= (size_t)len;
 948                         ptr += len;
 949                         last_path = rootpath;
 950                 }
 951
 952                 seen = true;
 953                 free(rootpath);
 954                 free(server);
 955         }
 956
 957         xlog(D_CALL, "%s: fslocdata='%s', ttl=%d",
 958                 __func__, fslocdata, *ttl);
 959         return seen;
 960
 961 out_false:
 962         free(rootpath);
 963         free(server);
 964         return false;
 965 }
 966
 967 /*
 968  * Duplicate the junction's parent's export options and graft in
 969  * the fslocdata we constructed from the locations list.
 970  */
 971 static struct exportent *create_junction_exportent(struct exportent *parent,
 972                 const char *junction, const char *fslocdata, int ttl)
 973 {
 974         static struct exportent *eep;
 975
 976         eep = (struct exportent *)malloc(sizeof(*eep));
 977         if (eep == NULL)
 978                 goto out_nomem;
 979
 980         dupexportent(eep, parent);
 981         strcpy(eep->e_path, junction);
 982         eep->e_hostname = strdup(parent->e_hostname);
 983         if (eep->e_hostname == NULL) {
 984                 free(eep);
 985                 goto out_nomem;
 986         }
 987         free(eep->e_uuid);
 988         eep->e_uuid = NULL;
 989         eep->e_ttl = (unsigned int)ttl;
 990
 991         free(eep->e_fslocdata);
 992         eep->e_fslocdata = strdup(fslocdata);
 993         if (eep->e_fslocdata == NULL) {
 994                 free(eep->e_hostname);
 995                 free(eep);
 996                 goto out_nomem;
 997         }
 998         eep->e_fslocmethod = FSLOC_REFER;
 999         return eep;
1000
1001 out_nomem:
1002         xlog(L_ERROR, "%s: No memory", __func__);
1003         return NULL;
1004 }
1005
1006 /*
1007  * Walk through the set of FS locations and build an exportent.
1008  * Returns pointer to an exportent if "junction" refers to a junction.
1009  */
1010 static struct exportent *locations_to_export(struct jp_ops *ops,
1011                 nfs_fsloc_set_t locations, const char *junction,
1012                 struct exportent *parent)
1013 {
1014         static char fslocdata[BUFSIZ];
1015         int ttl;
1016
1017         fslocdata[0] = '\0';
1018         if (!locations_to_fslocdata(ops, locations,
1019                                         fslocdata, sizeof(fslocdata), &ttl))
1020                 return NULL;
1021         return create_junction_exportent(parent, junction, fslocdata, ttl);
1022 }
1023
1024 /*
1025  * Retrieve locations information in "junction" and dump it to the
1026  * kernel.  Returns pointer to an exportent if "junction" refers
1027  * to a junction.
1028  */
1029 static struct exportent *invoke_junction_ops(void *handle, char *dom,
1030                 const char *junction, struct addrinfo *ai)
1031 {
1032         struct exportent *parent, *exp = NULL;
1033         nfs_fsloc_set_t locations;
1034         enum jp_status status;
1035         struct jp_ops *ops;
1036         char *error;
1037
1038         ops = (struct jp_ops *)dlsym(handle, "nfs_junction_ops");
1039         error = dlerror();
1040         if (error != NULL) {
1041                 xlog(D_GENERAL, "%s: dlsym(jp_junction_ops): %s",
1042                         __func__, error);
1043                 return NULL;
1044         }
1045         if (ops->jp_api_version != JP_API_VERSION) {
1046                 xlog(D_GENERAL, "%s: unrecognized junction API version: %u",
1047                         __func__, ops->jp_api_version);
1048                 return NULL;
1049         }
1050
1051         status = ops->jp_init(false);
1052         if (status != JP_OK) {
1053                 xlog(D_GENERAL, "%s: failed to resolve %s: %s",
1054                         __func__, junction, ops->jp_error(status));
1055                 return NULL;
1056         }
1057
1058         status = ops->jp_get_locations(junction, &locations);
1059         switch (status) {
1060         case JP_OK:
1061                 break;
1062         case JP_NOTJUNCTION:
1063                 xlog(D_GENERAL, "%s: %s is not a junction",
1064                         __func__, junction);
1065                 goto out;
1066         default:
1067                 xlog(L_WARNING, "Dangling junction %s: %s",
1068                         junction, ops->jp_error(status));
1069                 goto out;
1070         }
1071
1072         parent = lookup_parent_export(dom, junction, ai);
1073         if (parent == NULL)
1074                 goto out;
1075
1076         exp = locations_to_export(ops, locations, junction, parent);
1077
1078         ops->jp_put_locations(locations);
1079
1080 out:
1081         ops->jp_done();
1082         return exp;
1083 }
1084
1085 /*
1086  * Load the junction plug-in, then try to resolve "pathname".
1087  * Returns pointer to an initialized exportent if "junction"
1088  * refers to a junction, or NULL if not.
1089  */
1090 static struct exportent *lookup_junction(char *dom, const char *pathname,
1091                 struct addrinfo *ai)
1092 {
1093         struct exportent *exp;
1094         void *handle;
1095
1096         handle = dlopen("libnfsjunct.so", RTLD_NOW);
1097         if (handle == NULL) {
1098                 xlog(D_GENERAL, "%s: dlopen: %s", __func__, dlerror());
1099                 return NULL;
1100         }
1101         (void)dlerror();        /* Clear any error */
1102
1103         exp = invoke_junction_ops(handle, dom, pathname, ai);
1104
1105         /* We could leave it loaded to make junction resolution
1106          * faster next time.  However, if we want to replace the
1107          * library, that would require restarting mountd. */
1108         (void)dlclose(handle);
1109         return exp;
1110 }
1111
1112 static void lookup_nonexport(FILE *f, char *dom, char *path,
1113                 struct addrinfo *ai)
1114 {
1115         struct exportent *eep;
1116
1117         eep = lookup_junction(dom, path, ai);
1118         dump_to_cache(f, dom, path, eep);
1119         if (eep == NULL)
1120                 return;
1121         exportent_release(eep);
1122         free(eep);
1123 }
1124 #else   /* !HAVE_NFS_PLUGIN_H */
1125 static void lookup_nonexport(FILE *f, char *dom, char *path,
1126                 struct addrinfo *UNUSED(ai))
1127 {
1128         dump_to_cache(f, dom, path, NULL);
1129 }
1130 #endif  /* !HAVE_NFS_PLUGIN_H */
1131
1132 static void nfsd_export(FILE *f)
1133 {
1134         /* requests are:
1135          *  domain path
1136          * determine export options and return:
1137          *  domain path expiry flags anonuid anongid fsid
1138          */
1139
1140         char *cp;
1141         char *dom, *path;
1142         nfs_export *found = NULL;
1143         struct addrinfo *ai = NULL;
1144
1145         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
1146                 return;
1147
1148         xlog(D_CALL, "nfsd_export: inbuf '%s'", lbuf);
1149
1150         cp = lbuf;
1151         dom = malloc(strlen(cp));
1152         path = malloc(strlen(cp));
1153
1154         if (!dom || !path)
1155                 goto out;
1156
1157         if (qword_get(&cp, dom, strlen(lbuf)) <= 0)
1158                 goto out;
1159         if (qword_get(&cp, path, strlen(lbuf)) <= 0)
1160                 goto out;
1161
1162         auth_reload();
1163
1164         if (is_ipaddr_client(dom)) {
1165                 ai = lookup_client_addr(dom);
1166                 if (!ai)
1167                         goto out;
1168         }
1169
1170         found = lookup_export(dom, path, ai);
1171
1172         if (found) {
1173                 if (dump_to_cache(f, dom, path, &found->m_export) < 0) {
1174                         xlog(L_WARNING,
1175                              "Cannot export %s, possibly unsupported filesystem"
1176                              " or fsid= required", path);
1177                         dump_to_cache(f, dom, path, NULL);
1178                 }
1179         } else
1180                 lookup_nonexport(f, dom, path, ai);
1181
1182  out:
1183         xlog(D_CALL, "nfsd_export: found %p path %s", found, path ? path : NULL);
1184         if (dom) free(dom);
1185         if (path) free(path);
1186         freeaddrinfo(ai);
1187 }
1188
1189
1190 struct {
1191         char *cache_name;
1192         void (*cache_handle)(FILE *f);
1193         FILE *f;
1194         char vbuf[RPC_CHAN_BUF_SIZE];
1195 } cachelist[] = {
1196         { "auth.unix.ip", auth_unix_ip, NULL, ""},
1197         { "auth.unix.gid", auth_unix_gid, NULL, ""},
1198         { "nfsd.export", nfsd_export, NULL, ""},
1199         { "nfsd.fh", nfsd_fh, NULL, ""},
1200         { NULL, NULL, NULL, ""}
1201 };
1202
1203 extern int manage_gids;
1204
1205 /**
1206  * cache_open - prepare communications channels with kernel RPC caches
1207  *
1208  */
1209 void cache_open(void)
1210 {
1211         int i;
1212         for (i=0; cachelist[i].cache_name; i++ ) {
1213                 char path[100];
1214                 if (!manage_gids && cachelist[i].cache_handle == auth_unix_gid)
1215                         continue;
1216                 sprintf(path, "/proc/net/rpc/%s/channel", cachelist[i].cache_name);
1217                 cachelist[i].f = fopen(path, "r+");
1218                 if (cachelist[i].f != NULL) {
1219                         setvbuf(cachelist[i].f, cachelist[i].vbuf, _IOLBF,
1220                                 RPC_CHAN_BUF_SIZE);
1221                 }
1222         }
1223 }
1224
1225 /**
1226  * cache_set_fds - prepare cache file descriptors for one iteration of the service loop
1227  * @fdset: pointer to fd_set to prepare
1228  */
1229 void cache_set_fds(fd_set *fdset)
1230 {
1231         int i;
1232         for (i=0; cachelist[i].cache_name; i++) {
1233                 if (cachelist[i].f)
1234                         FD_SET(fileno(cachelist[i].f), fdset);
1235         }
1236 }
1237
1238 /**
1239  * cache_process_req - process any active cache file descriptors during service loop iteration
1240  * @fdset: pointer to fd_set to examine for activity
1241  */
1242 int cache_process_req(fd_set *readfds)
1243 {
1244         int i;
1245         int cnt = 0;
1246         for (i=0; cachelist[i].cache_name; i++) {
1247                 if (cachelist[i].f != NULL &&
1248                     FD_ISSET(fileno(cachelist[i].f), readfds)) {
1249                         cnt++;
1250                         cachelist[i].cache_handle(cachelist[i].f);
1251                         FD_CLR(fileno(cachelist[i].f), readfds);
1252                 }
1253         }
1254         return cnt;
1255 }
1256
1257
1258 /*
1259  * Give IP->domain and domain+path->options to kernel
1260  * % echo nfsd $IP  $[now+DEFAULT_TTL] $domain > /proc/net/rpc/auth.unix.ip/channel
1261  * % echo $domain $path $[now+DEFAULT_TTL] $options $anonuid $anongid $fsid > /proc/net/rpc/nfsd.export/channel
1262  */
1263
1264 static int cache_export_ent(char *domain, struct exportent *exp, char *path)
1265 {
1266         int err;
1267         FILE *f = fopen("/proc/net/rpc/nfsd.export/channel", "w");
1268         if (!f)
1269                 return -1;
1270
1271         err = dump_to_cache(f, domain, exp->e_path, exp);
1272         if (err) {
1273                 xlog(L_WARNING,
1274                      "Cannot export %s, possibly unsupported filesystem or"
1275                      " fsid= required", exp->e_path);
1276         }
1277
1278         while (err == 0 && (exp->e_flags & NFSEXP_CROSSMOUNT) && path) {
1279                 /* really an 'if', but we can break out of
1280                  * a 'while' more easily */
1281                 /* Look along 'path' for other filesystems
1282                  * and export them with the same options
1283                  */
1284                 struct stat stb;
1285                 size_t l = strlen(exp->e_path);
1286                 __dev_t dev;
1287
1288                 if (strlen(path) <= l || path[l] != '/' ||
1289                     strncmp(exp->e_path, path, l) != 0)
1290                         break;
1291                 if (stat(exp->e_path, &stb) != 0)
1292                         break;
1293                 dev = stb.st_dev;
1294                 while(path[l] == '/') {
1295                         char c;
1296                         /* errors for submount should fail whole filesystem */
1297                         int err2;
1298
1299                         l++;
1300                         while (path[l] != '/' && path[l])
1301                                 l++;
1302                         c = path[l];
1303                         path[l] = 0;
1304                         err2 = lstat(path, &stb);
1305                         path[l] = c;
1306                         if (err2 < 0)
1307                                 break;
1308                         if (stb.st_dev == dev)
1309                                 continue;
1310                         dev = stb.st_dev;
1311                         path[l] = 0;
1312                         dump_to_cache(f, domain, path, exp);
1313                         path[l] = c;
1314                 }
1315                 break;
1316         }
1317
1318         fclose(f);
1319         return err;
1320 }
1321
1322 /**
1323  * cache_export - Inform kernel of a new nfs_export
1324  * @exp: target nfs_export
1325  * @path: NUL-terminated C string containing export path
1326  */
1327 int cache_export(nfs_export *exp, char *path)
1328 {
1329         char buf[INET6_ADDRSTRLEN];
1330         int err;
1331         FILE *f;
1332
1333         f = fopen("/proc/net/rpc/auth.unix.ip/channel", "w");
1334         if (!f)
1335                 return -1;
1336
1337
1338         qword_print(f, "nfsd");
1339         qword_print(f,
1340                 host_ntop(get_addrlist(exp->m_client, 0), buf, sizeof(buf)));
1341         qword_printtimefrom(f, exp->m_export.e_ttl);
1342         qword_print(f, exp->m_client->m_hostname);
1343         err = qword_eol(f);
1344
1345         fclose(f);
1346
1347         err = cache_export_ent(exp->m_client->m_hostname, &exp->m_export, path)
1348                 || err;
1349         return err;
1350 }
1351
1352 /**
1353  * cache_get_filehandle - given an nfs_export, get its root filehandle
1354  * @exp: target nfs_export
1355  * @len: length of requested file handle
1356  * @p: NUL-terminated C string containing export path
1357  *
1358  * Returns pointer to NFS file handle of root directory of export
1359  *
1360  * {
1361  *   echo $domain $path $length
1362  *   read filehandle <&0
1363  * } <> /proc/fs/nfsd/filehandle
1364  */
1365 struct nfs_fh_len *
1366 cache_get_filehandle(nfs_export *exp, int len, char *p)
1367 {
1368         FILE *f = fopen("/proc/fs/nfsd/filehandle", "r+");
1369         char buf[200];
1370         char *bp = buf;
1371         int failed;
1372         static struct nfs_fh_len fh;
1373
1374         if (!f)
1375                 f = fopen("/proc/fs/nfs/filehandle", "r+");
1376         if (!f)
1377                 return NULL;
1378
1379         qword_print(f, exp->m_client->m_hostname);
1380         qword_print(f, p);
1381         qword_printint(f, len);
1382         failed = qword_eol(f);
1383
1384         if (!failed)
1385                 failed = (fgets(buf, sizeof(buf), f) == NULL);
1386         fclose(f);
1387         if (failed)
1388                 return NULL;
1389         memset(fh.fh_handle, 0, sizeof(fh.fh_handle));
1390         fh.fh_size = qword_get(&bp, (char *)fh.fh_handle, NFS3_FHSIZE);
1391         return &fh;
1392 }