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 static void nfsd_fh(FILE *f)
 368 {
 369         /* request are:
 370          *  domain fsidtype fsid
 371          * interpret fsid, find export point and options, and write:
 372          *  domain fsidtype fsid expiry path
 373          */
 374         char *cp;
 375         char *dom;
 376         int fsidtype;
 377         int fsidlen;
 378         unsigned long long inode64;
 379         unsigned int dev;
 380         char fsid[32];
 381         struct parsed_fsid parsed;
 382         struct exportent *found = NULL;
 383         struct addrinfo *ai = NULL;
 384         char *found_path = NULL;
 385         nfs_export *exp;
 386         int i;
 387         int dev_missing = 0;
 388
 389         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
 390                 return;
 391
 392         xlog(D_CALL, "nfsd_fh: inbuf '%s'", lbuf);
 393
 394         cp = lbuf;
 395
 396         dom = malloc(strlen(cp));
 397         if (dom == NULL)
 398                 return;
 399         if (qword_get(&cp, dom, strlen(cp)) <= 0)
 400                 goto out;
 401         if (qword_get_int(&cp, &fsidtype) != 0)
 402                 goto out;
 403         if (fsidtype < 0 || fsidtype > 7)
 404                 goto out; /* unknown type */
 405         if ((fsidlen = qword_get(&cp, fsid, 32)) <= 0)
 406                 goto out;
 407         switch(fsidtype) {
 408         case FSID_DEV: /* 4 bytes: 2 major, 2 minor, 4 inode */
 409                 if (fsidlen != 8)
 410                         goto out;
 411                 memcpy(&dev, fsid, 4);
 412                 memcpy(&parsed.inode, fsid+4, 4);
 413                 parsed.major = ntohl(dev)>>16;
 414                 parsed.minor = ntohl(dev) & 0xFFFF;
 415                 break;
 416
 417         case FSID_NUM: /* 4 bytes - fsid */
 418                 if (fsidlen != 4)
 419                         goto out;
 420                 memcpy(&parsed.fsidnum, fsid, 4);
 421                 break;
 422
 423         case FSID_MAJOR_MINOR: /* 12 bytes: 4 major, 4 minor, 4 inode
 424                  * This format is never actually used but was
 425                  * an historical accident
 426                  */
 427                 if (fsidlen != 12)
 428                         goto out;
 429                 memcpy(&dev, fsid, 4);
 430                 parsed.major = ntohl(dev);
 431                 memcpy(&dev, fsid+4, 4);
 432                 parsed.minor = ntohl(dev);
 433                 memcpy(&parsed.inode, fsid+8, 4);
 434                 break;
 435
 436         case FSID_ENCODE_DEV: /* 8 bytes: 4 byte packed device number, 4 inode */
 437                 /* This is *host* endian, not net-byte-order, because
 438                  * no-one outside this host has any business interpreting it
 439                  */
 440                 if (fsidlen != 8)
 441                         goto out;
 442                 memcpy(&dev, fsid, 4);
 443                 memcpy(&parsed.inode, fsid+4, 4);
 444                 parsed.major = (dev & 0xfff00) >> 8;
 445                 parsed.minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
 446                 break;
 447
 448         case FSID_UUID4_INUM: /* 4 byte inode number and 4 byte uuid */
 449                 if (fsidlen != 8)
 450                         goto out;
 451                 memcpy(&parsed.inode, fsid, 4);
 452                 parsed.uuidlen = 4;
 453                 parsed.fhuuid = fsid+4;
 454                 break;
 455         case FSID_UUID8: /* 8 byte uuid */
 456                 if (fsidlen != 8)
 457                         goto out;
 458                 parsed.uuidlen = 8;
 459                 parsed.fhuuid = fsid;
 460                 break;
 461         case FSID_UUID16: /* 16 byte uuid */
 462                 if (fsidlen != 16)
 463                         goto out;
 464                 parsed.uuidlen = 16;
 465                 parsed.fhuuid = fsid;
 466                 break;
 467         case FSID_UUID16_INUM: /* 8 byte inode number and 16 byte uuid */
 468                 if (fsidlen != 24)
 469                         goto out;
 470                 memcpy(&inode64, fsid, 8);
 471                 parsed.inode = inode64;
 472                 parsed.uuidlen = 16;
 473                 parsed.fhuuid = fsid+8;
 474                 break;
 475         }
 476
 477         auth_reload();
 478
 479         /* Now determine export point for this fsid/domain */
 480         for (i=0 ; i < MCL_MAXTYPES; i++) {
 481                 nfs_export *next_exp;
 482                 for (exp = exportlist[i].p_head; exp; exp = next_exp) {
 483                         struct stat stb;
 484                         char u[16];
 485                         char *path;
 486                         int type;
 487
 488                         if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT) {
 489                                 static nfs_export *prev = NULL;
 490                                 static void *mnt = NULL;
 491
 492                                 if (prev == exp) {
 493                                         /* try a submount */
 494                                         path = next_mnt(&mnt, exp->m_export.e_path);
 495                                         if (!path) {
 496                                                 next_exp = exp->m_next;
 497                                                 prev = NULL;
 498                                                 continue;
 499                                         }
 500                                         next_exp = exp;
 501                                 } else {
 502                                         prev = exp;
 503                                         mnt = NULL;
 504                                         path = exp->m_export.e_path;
 505                                         next_exp = exp;
 506                                 }
 507                         } else {
 508                                 path = exp->m_export.e_path;
 509                                 next_exp = exp->m_next;
 510                         }
 511
 512                         if (!use_ipaddr && !client_member(dom, exp->m_client->m_hostname))
 513                                 continue;
 514                         if (exp->m_export.e_mountpoint &&
 515                             !is_mountpoint(exp->m_export.e_mountpoint[0]?
 516                                            exp->m_export.e_mountpoint:
 517                                            exp->m_export.e_path))
 518                                 dev_missing ++;
 519                         if (stat(path, &stb) != 0)
 520                                 continue;
 521                         if (!S_ISDIR(stb.st_mode) && !S_ISREG(stb.st_mode)) {
 522                                 continue;
 523                         }
 524                         switch(fsidtype){
 525                         case FSID_DEV:
 526                         case FSID_MAJOR_MINOR:
 527                         case FSID_ENCODE_DEV:
 528                                 if (stb.st_ino != parsed.inode)
 529                                         continue;
 530                                 if (parsed.major != major(stb.st_dev) ||
 531                                     parsed.minor != minor(stb.st_dev))
 532                                         continue;
 533                                 break;
 534                         case FSID_NUM:
 535                                 if (((exp->m_export.e_flags & NFSEXP_FSID) == 0 ||
 536                                      exp->m_export.e_fsid != parsed.fsidnum))
 537                                         continue;
 538                                 break;
 539                         case FSID_UUID4_INUM:
 540                         case FSID_UUID16_INUM:
 541                                 if (stb.st_ino != parsed.inode)
 542                                         continue;
 543                                 goto check_uuid;
 544                         case FSID_UUID8:
 545                         case FSID_UUID16:
 546                                 if (!is_mountpoint(path))
 547                                         continue;
 548                         check_uuid:
 549                                 if (exp->m_export.e_uuid)
 550                                         get_uuid(exp->m_export.e_uuid,
 551                                                  parsed.uuidlen, u);
 552                                 else
 553                                         for (type = 0;
 554                                              uuid_by_path(path, type, parsed.uuidlen, u);
 555                                              type++)
 556                                                 if (memcmp(u, parsed.fhuuid, parsed.uuidlen) == 0)
 557                                                         break;
 558
 559                                 if (memcmp(u, parsed.fhuuid, parsed.uuidlen) != 0)
 560                                         continue;
 561                                 break;
 562                         }
 563                         if (use_ipaddr) {
 564                                 if (ai == NULL) {
 565                                         struct addrinfo *tmp;
 566                                         tmp = host_pton(dom);
 567                                         if (tmp == NULL)
 568                                                 goto out;
 569                                         ai = client_resolve(tmp->ai_addr);
 570                                         freeaddrinfo(tmp);
 571                                 }
 572                                 if (!client_check(exp->m_client, ai))
 573                                         continue;
 574                         }
 575                         if (!found || subexport(&exp->m_export, found)) {
 576                                 found = &exp->m_export;
 577                                 free(found_path);
 578                                 found_path = strdup(path);
 579                                 if (found_path == NULL)
 580                                         goto out;
 581                         } else if (strcmp(found->e_path, exp->m_export.e_path)
 582                                    && !subexport(found, &exp->m_export))
 583                         {
 584                                 xlog(L_WARNING, "%s and %s have same filehandle for %s, using first",
 585                                      found_path, path, dom);
 586                         }
 587                 }
 588         }
 589         if (found &&
 590             found->e_mountpoint &&
 591             !is_mountpoint(found->e_mountpoint[0]?
 592                            found->e_mountpoint:
 593                            found->e_path)) {
 594                 /* Cannot export this yet
 595                  * should log a warning, but need to rate limit
 596                    xlog(L_WARNING, "%s not exported as %d not a mountpoint",
 597                    found->e_path, found->e_mountpoint);
 598                  */
 599                 /* FIXME we need to make sure we re-visit this later */
 600                 goto out;
 601         }
 602         if (!found && dev_missing) {
 603                 /* The missing dev could be what we want, so just be
 604                  * quite rather than returning stale yet
 605                  */
 606                 goto out;
 607         }
 608
 609         if (found)
 610                 if (cache_export_ent(dom, found, found_path) < 0)
 611                         found = 0;
 612
 613         qword_print(f, dom);
 614         qword_printint(f, fsidtype);
 615         qword_printhex(f, fsid, fsidlen);
 616         /* The fsid -> path lookup can be quite expensive as it
 617          * potentially stats and reads lots of devices, and some of those
 618          * might have spun-down.  The Answer is not likely to
 619          * change underneath us, and an 'exportfs -f' can always
 620          * remove this from the kernel, so use a really log
 621          * timeout.  Maybe this should be configurable on the command
 622          * line.
 623          */
 624         qword_printint(f, 0x7fffffff);
 625         if (found)
 626                 qword_print(f, found_path);
 627         qword_eol(f);
 628  out:
 629         if (found_path)
 630                 free(found_path);
 631         freeaddrinfo(ai);
 632         free(dom);
 633         xlog(D_CALL, "nfsd_fh: found %p path %s", found, found ? found->e_path : NULL);
 634         return;
 635 }
 636
 637 static void write_fsloc(FILE *f, struct exportent *ep)
 638 {
 639         struct servers *servers;
 640
 641         if (ep->e_fslocmethod == FSLOC_NONE)
 642                 return;
 643
 644         servers = replicas_lookup(ep->e_fslocmethod, ep->e_fslocdata);
 645         if (!servers)
 646                 return;
 647         qword_print(f, "fsloc");
 648         qword_printint(f, servers->h_num);
 649         if (servers->h_num >= 0) {
 650                 int i;
 651                 for (i=0; i<servers->h_num; i++) {
 652                         qword_print(f, servers->h_mp[i]->h_host);
 653                         qword_print(f, servers->h_mp[i]->h_path);
 654                 }
 655         }
 656         qword_printint(f, servers->h_referral);
 657         release_replicas(servers);
 658 }
 659
 660 static void write_secinfo(FILE *f, struct exportent *ep, int flag_mask)
 661 {
 662         struct sec_entry *p;
 663
 664         for (p = ep->e_secinfo; p->flav; p++)
 665                 ; /* Do nothing */
 666         if (p == ep->e_secinfo) {
 667                 /* There was no sec= option */
 668                 return;
 669         }
 670         qword_print(f, "secinfo");
 671         qword_printint(f, p - ep->e_secinfo);
 672         for (p = ep->e_secinfo; p->flav; p++) {
 673                 qword_printint(f, p->flav->fnum);
 674                 qword_printint(f, p->flags & flag_mask);
 675         }
 676
 677 }
 678
 679 static int dump_to_cache(FILE *f, char *domain, char *path, struct exportent *exp)
 680 {
 681         qword_print(f, domain);
 682         qword_print(f, path);
 683         if (exp) {
 684                 int different_fs = strcmp(path, exp->e_path) != 0;
 685                 int flag_mask = different_fs ? ~NFSEXP_FSID : ~0;
 686
 687                 qword_printuint(f, time(0) + exp->e_ttl);
 688                 qword_printint(f, exp->e_flags & flag_mask);
 689                 qword_printint(f, exp->e_anonuid);
 690                 qword_printint(f, exp->e_anongid);
 691                 qword_printint(f, exp->e_fsid);
 692                 write_fsloc(f, exp);
 693                 write_secinfo(f, exp, flag_mask);
 694                 if (exp->e_uuid == NULL || different_fs) {
 695                         char u[16];
 696                         if (uuid_by_path(path, 0, 16, u)) {
 697                                 qword_print(f, "uuid");
 698                                 qword_printhex(f, u, 16);
 699                         }
 700                 } else {
 701                         char u[16];
 702                         get_uuid(exp->e_uuid, 16, u);
 703                         qword_print(f, "uuid");
 704                         qword_printhex(f, u, 16);
 705                 }
 706         } else
 707                 qword_printuint(f, time(0) + DEFAULT_TTL);
 708         return qword_eol(f);
 709 }
 710
 711 static int is_subdirectory(char *child, char *parent)
 712 {
 713         int l = strlen(parent);
 714
 715         return strcmp(child, parent) == 0
 716                 || (strncmp(child, parent, l) == 0 && child[l] == '/');
 717 }
 718
 719 static int path_matches(nfs_export *exp, char *path)
 720 {
 721         if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)
 722                 return is_subdirectory(path, exp->m_export.e_path);
 723         return strcmp(path, exp->m_export.e_path) == 0;
 724 }
 725
 726 static int
 727 client_matches(nfs_export *exp, char *dom, struct addrinfo *ai)
 728 {
 729         if (use_ipaddr)
 730                 return client_check(exp->m_client, ai);
 731         return client_member(dom, exp->m_client->m_hostname);
 732 }
 733
 734 static int
 735 export_matches(nfs_export *exp, char *dom, char *path, struct addrinfo *ai)
 736 {
 737         return path_matches(exp, path) && client_matches(exp, dom, ai);
 738 }
 739
 740 static nfs_export *
 741 lookup_export(char *dom, char *path, struct addrinfo *ai)
 742 {
 743         nfs_export *exp;
 744         nfs_export *found = NULL;
 745         int found_type = 0;
 746         int i;
 747
 748         for (i=0 ; i < MCL_MAXTYPES; i++) {
 749                 for (exp = exportlist[i].p_head; exp; exp = exp->m_next) {
 750                         if (!export_matches(exp, dom, path, ai))
 751                                 continue;
 752                         if (!found) {
 753                                 found = exp;
 754                                 found_type = i;
 755                                 continue;
 756                         }
 757
 758                         /* Always prefer non-V4ROOT mounts */
 759                         if (found->m_export.e_flags & NFSEXP_V4ROOT)
 760                                 continue;
 761
 762                         /* If one is a CROSSMOUNT, then prefer the longest path */
 763                         if (((found->m_export.e_flags & NFSEXP_CROSSMOUNT) ||
 764                              (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)) &&
 765                             strlen(found->m_export.e_path) !=
 766                             strlen(exp->m_export.e_path)) {
 767
 768                                 if (strlen(exp->m_export.e_path) >
 769                                     strlen(found->m_export.e_path)) {
 770                                         found = exp;
 771                                         found_type = i;
 772                                 }
 773                                 continue;
 774
 775                         } else if (found_type == i && found->m_warned == 0) {
 776                                 xlog(L_WARNING, "%s exported to both %s and %s, "
 777                                      "arbitrarily choosing options from first",
 778                                      path, found->m_client->m_hostname, exp->m_client->m_hostname,
 779                                      dom);
 780                                 found->m_warned = 1;
 781                         }
 782                 }
 783         }
 784         return found;
 785 }
 786
 787 static void nfsd_export(FILE *f)
 788 {
 789         /* requests are:
 790          *  domain path
 791          * determine export options and return:
 792          *  domain path expiry flags anonuid anongid fsid
 793          */
 794
 795         char *cp;
 796         char *dom, *path;
 797         nfs_export *found = NULL;
 798         struct addrinfo *ai = NULL;
 799
 800         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
 801                 return;
 802
 803         xlog(D_CALL, "nfsd_export: inbuf '%s'", lbuf);
 804
 805         cp = lbuf;
 806         dom = malloc(strlen(cp));
 807         path = malloc(strlen(cp));
 808
 809         if (!dom || !path)
 810                 goto out;
 811
 812         if (qword_get(&cp, dom, strlen(lbuf)) <= 0)
 813                 goto out;
 814         if (qword_get(&cp, path, strlen(lbuf)) <= 0)
 815                 goto out;
 816
 817         auth_reload();
 818
 819         if (use_ipaddr) {
 820                 struct addrinfo *tmp;
 821                 tmp = host_pton(dom);
 822                 if (tmp == NULL)
 823                         goto out;
 824                 ai = client_resolve(tmp->ai_addr);
 825                 freeaddrinfo(tmp);
 826                         goto out;
 827         }
 828
 829         found = lookup_export(dom, path, ai);
 830
 831         if (found) {
 832                 if (dump_to_cache(f, dom, path, &found->m_export) < 0) {
 833                         xlog(L_WARNING,
 834                              "Cannot export %s, possibly unsupported filesystem"
 835                              " or fsid= required", path);
 836                         dump_to_cache(f, dom, path, NULL);
 837                 }
 838         } else {
 839                 dump_to_cache(f, dom, path, NULL);
 840         }
 841  out:
 842         xlog(D_CALL, "nfsd_export: found %p path %s", found, path ? path : NULL);
 843         if (dom) free(dom);
 844         if (path) free(path);
 845         freeaddrinfo(ai);
 846 }
 847
 848
 849 struct {
 850         char *cache_name;
 851         void (*cache_handle)(FILE *f);
 852         FILE *f;
 853         char vbuf[RPC_CHAN_BUF_SIZE];
 854 } cachelist[] = {
 855         { "auth.unix.ip", auth_unix_ip, NULL, ""},
 856         { "auth.unix.gid", auth_unix_gid, NULL, ""},
 857         { "nfsd.export", nfsd_export, NULL, ""},
 858         { "nfsd.fh", nfsd_fh, NULL, ""},
 859         { NULL, NULL, NULL, ""}
 860 };
 861
 862 extern int manage_gids;
 863
 864 /**
 865  * cache_open - prepare communications channels with kernel RPC caches
 866  *
 867  */
 868 void cache_open(void)
 869 {
 870         int i;
 871         for (i=0; cachelist[i].cache_name; i++ ) {
 872                 char path[100];
 873                 if (!manage_gids && cachelist[i].cache_handle == auth_unix_gid)
 874                         continue;
 875                 sprintf(path, "/proc/net/rpc/%s/channel", cachelist[i].cache_name);
 876                 cachelist[i].f = fopen(path, "r+");
 877                 if (cachelist[i].f != NULL) {
 878                         setvbuf(cachelist[i].f, cachelist[i].vbuf, _IOLBF,
 879                                 RPC_CHAN_BUF_SIZE);
 880                 }
 881         }
 882 }
 883
 884 /**
 885  * cache_set_fds - prepare cache file descriptors for one iteration of the service loop
 886  * @fdset: pointer to fd_set to prepare
 887  */
 888 void cache_set_fds(fd_set *fdset)
 889 {
 890         int i;
 891         for (i=0; cachelist[i].cache_name; i++) {
 892                 if (cachelist[i].f)
 893                         FD_SET(fileno(cachelist[i].f), fdset);
 894         }
 895 }
 896
 897 /**
 898  * cache_process_req - process any active cache file descriptors during service loop iteration
 899  * @fdset: pointer to fd_set to examine for activity
 900  */
 901 int cache_process_req(fd_set *readfds)
 902 {
 903         int i;
 904         int cnt = 0;
 905         for (i=0; cachelist[i].cache_name; i++) {
 906                 if (cachelist[i].f != NULL &&
 907                     FD_ISSET(fileno(cachelist[i].f), readfds)) {
 908                         cnt++;
 909                         cachelist[i].cache_handle(cachelist[i].f);
 910                         FD_CLR(fileno(cachelist[i].f), readfds);
 911                 }
 912         }
 913         return cnt;
 914 }
 915
 916
 917 /*
 918  * Give IP->domain and domain+path->options to kernel
 919  * % echo nfsd $IP  $[now+DEFAULT_TTL] $domain > /proc/net/rpc/auth.unix.ip/channel
 920  * % echo $domain $path $[now+DEFAULT_TTL] $options $anonuid $anongid $fsid > /proc/net/rpc/nfsd.export/channel
 921  */
 922
 923 static int cache_export_ent(char *domain, struct exportent *exp, char *path)
 924 {
 925         int err;
 926         FILE *f = fopen("/proc/net/rpc/nfsd.export/channel", "w");
 927         if (!f)
 928                 return -1;
 929
 930         err = dump_to_cache(f, domain, exp->e_path, exp);
 931         if (err) {
 932                 xlog(L_WARNING,
 933                      "Cannot export %s, possibly unsupported filesystem or"
 934                      " fsid= required", exp->e_path);
 935         }
 936
 937         while (err == 0 && (exp->e_flags & NFSEXP_CROSSMOUNT) && path) {
 938                 /* really an 'if', but we can break out of
 939                  * a 'while' more easily */
 940                 /* Look along 'path' for other filesystems
 941                  * and export them with the same options
 942                  */
 943                 struct stat stb;
 944                 size_t l = strlen(exp->e_path);
 945                 __dev_t dev;
 946
 947                 if (strlen(path) <= l || path[l] != '/' ||
 948                     strncmp(exp->e_path, path, l) != 0)
 949                         break;
 950                 if (stat(exp->e_path, &stb) != 0)
 951                         break;
 952                 dev = stb.st_dev;
 953                 while(path[l] == '/') {
 954                         char c;
 955                         /* errors for submount should fail whole filesystem */
 956                         int err2;
 957
 958                         l++;
 959                         while (path[l] != '/' && path[l])
 960                                 l++;
 961                         c = path[l];
 962                         path[l] = 0;
 963                         err2 = lstat(path, &stb);
 964                         path[l] = c;
 965                         if (err2 < 0)
 966                                 break;
 967                         if (stb.st_dev == dev)
 968                                 continue;
 969                         dev = stb.st_dev;
 970                         path[l] = 0;
 971                         dump_to_cache(f, domain, path, exp);
 972                         path[l] = c;
 973                 }
 974                 break;
 975         }
 976
 977         fclose(f);
 978         return err;
 979 }
 980
 981 /**
 982  * cache_export - Inform kernel of a new nfs_export
 983  * @exp: target nfs_export
 984  * @path: NUL-terminated C string containing export path
 985  */
 986 int cache_export(nfs_export *exp, char *path)
 987 {
 988         char buf[INET6_ADDRSTRLEN];
 989         int err;
 990         FILE *f;
 991
 992         f = fopen("/proc/net/rpc/auth.unix.ip/channel", "w");
 993         if (!f)
 994                 return -1;
 995
 996
 997         qword_print(f, "nfsd");
 998         qword_print(f,
 999                 host_ntop(get_addrlist(exp->m_client, 0), buf, sizeof(buf)));
1000         qword_printuint(f, time(0) + exp->m_export.e_ttl);
1001         qword_print(f, exp->m_client->m_hostname);
1002         err = qword_eol(f);
1003
1004         fclose(f);
1005
1006         err = cache_export_ent(exp->m_client->m_hostname, &exp->m_export, path)
1007                 || err;
1008         return err;
1009 }
1010
1011 /**
1012  * cache_get_filehandle - given an nfs_export, get its root filehandle
1013  * @exp: target nfs_export
1014  * @len: length of requested file handle
1015  * @p: NUL-terminated C string containing export path
1016  *
1017  * Returns pointer to NFS file handle of root directory of export
1018  *
1019  * {
1020  *   echo $domain $path $length
1021  *   read filehandle <&0
1022  * } <> /proc/fs/nfsd/filehandle
1023  */
1024 struct nfs_fh_len *
1025 cache_get_filehandle(nfs_export *exp, int len, char *p)
1026 {
1027         FILE *f = fopen("/proc/fs/nfsd/filehandle", "r+");
1028         char buf[200];
1029         char *bp = buf;
1030         int failed;
1031         static struct nfs_fh_len fh;
1032
1033         if (!f)
1034                 f = fopen("/proc/fs/nfs/filehandle", "r+");
1035         if (!f)
1036                 return NULL;
1037
1038         qword_print(f, exp->m_client->m_hostname);
1039         qword_print(f, p);
1040         qword_printint(f, len);
1041         failed = qword_eol(f);
1042
1043         if (!failed)
1044                 failed = (fgets(buf, sizeof(buf), f) == NULL);
1045         fclose(f);
1046         if (failed)
1047                 return NULL;
1048         memset(fh.fh_handle, 0, sizeof(fh.fh_handle));
1049         fh.fh_size = qword_get(&bp, (char *)fh.fh_handle, NFS3_FHSIZE);
1050         return &fh;
1051 }