1 /* $NetBSD: perfuse.c,v 1.26 2012/03/21 10:10:36 matt Exp $ */ 2 3 /*- 4 * Copyright (c) 2010-2011 Emmanuel Dreyfus. All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 16 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 17 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 18 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 19 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 20 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 21 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 22 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 23 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 24 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 25 * POSSIBILITY OF SUCH DAMAGE. 26 */ 27 28 #include <stdio.h> 29 #include <unistd.h> 30 #include <stdlib.h> 31 #include <fcntl.h> 32 #include <string.h> 33 #include <errno.h> 34 #include <puffs.h> 35 #include <sys/types.h> 36 #include <sys/mman.h> 37 #include <sys/resource.h> 38 #include <sys/socket.h> 39 #include <sys/extattr.h> 40 #include <sys/un.h> 41 #include <machine/vmparam.h> 42 43 #define LIBPERFUSE 44 #include "perfuse.h" 45 #include "perfuse_if.h" 46 #include "perfuse_priv.h" 47 48 int perfuse_diagflags = 0; /* global used only in DPRINTF/DERR/DWARN */ 49 extern char **environ; 50 51 static struct perfuse_state *init_state(void); 52 static int get_fd(const char *); 53 54 55 static struct perfuse_state * 56 init_state(void) 57 { 58 struct perfuse_state *ps; 59 char opts[1024]; 60 61 if ((ps = malloc(sizeof(*ps))) == NULL) 62 DERR(EX_OSERR, "%s: malloc failed", __func__); 63 64 (void)memset(ps, 0, sizeof(*ps)); 65 ps->ps_max_write = UINT_MAX; 66 ps->ps_max_readahead = UINT_MAX; 67 TAILQ_INIT(&ps->ps_trace); 68 69 /* 70 * Most of the time, access() is broken because the filesystem 71 * performs the check with root privileges. glusterfs will do that 72 * if the Linux-specific setfsuid() is missing, for instance. 73 */ 74 ps->ps_flags |= PS_NO_ACCESS; 75 76 /* 77 * This is a temporary way to toggle access and creat usage. 78 * It would be nice if that could be provided as mount options, 79 * but that will not be obvious to do. 80 */ 81 if (getenv_r("PERFUSE_OPTIONS", opts, sizeof(opts)) != -1) { 82 char *optname; 83 char *last; 84 85 for ((optname = strtok_r(opts, ",", &last)); 86 optname != NULL; 87 (optname = strtok_r(NULL, ",", &last))) { 88 if (strcmp(optname, "enable_access") == 0) 89 ps->ps_flags &= ~PS_NO_ACCESS; 90 91 if (strcmp(optname, "disable_access") == 0) 92 ps->ps_flags |= PS_NO_ACCESS; 93 94 if (strcmp(optname, "enable_creat") == 0) 95 ps->ps_flags &= ~PS_NO_CREAT; 96 97 if (strcmp(optname, "disable_creat") == 0) 98 ps->ps_flags |= PS_NO_CREAT; 99 } 100 } 101 102 103 return ps; 104 } 105 106 107 static int 108 get_fd(const char *data) 109 { 110 char *string; 111 const char fdopt[] = "fd="; 112 char *lastp; 113 char *opt; 114 int fd = -1; 115 116 if ((string = strdup(data)) == NULL) 117 return -1; 118 119 for (opt = strtok_r(string, ",", &lastp); 120 opt != NULL; 121 opt = strtok_r(NULL, ",", &lastp)) { 122 if (strncmp(opt, fdopt, strlen(fdopt)) == 0) { 123 fd = atoi(opt + strlen(fdopt)); 124 break; 125 } 126 } 127 128 /* 129 * No file descriptor found 130 */ 131 if (fd == -1) 132 errno = EINVAL; 133 134 free(string); 135 return fd; 136 137 } 138 139 int 140 perfuse_open(const char *path, int flags, mode_t mode) 141 { 142 int sv[2]; 143 struct sockaddr_un sun; 144 struct sockaddr *sa; 145 char progname[] = _PATH_PERFUSED; 146 char minus_i[] = "-i"; 147 char fdstr[16]; 148 char *const argv[] = { progname, minus_i, fdstr, NULL}; 149 uint32_t opt; 150 uint32_t optlen; 151 int sock_type = SOCK_SEQPACKET; 152 153 if (strcmp(path, _PATH_FUSE) != 0) 154 return open(path, flags, mode); 155 156 /* 157 * Try SOCK_SEQPACKET then SOCK_DGRAM if unavailable 158 */ 159 if ((sv[0] = socket(PF_LOCAL, SOCK_SEQPACKET, 0)) == -1) { 160 sock_type = SOCK_DGRAM; 161 DWARNX("SEQPACKET local sockets unavailable, using less " 162 "reliable DGRAM sockets. Expect file operation hangs."); 163 164 if ((sv[0] = socket(PF_LOCAL, SOCK_DGRAM, 0)) == -1) { 165 #ifdef PERFUSE_DEBUG 166 DWARN("%s: %d socket failed", __func__, __LINE__); 167 #endif 168 return -1; 169 } 170 } 171 172 /* 173 * Set a buffer lentgh large enough so that any FUSE packet 174 * will fit. 175 */ 176 opt = (uint32_t)FUSE_BUFSIZE; 177 optlen = sizeof(opt); 178 if (setsockopt(sv[0], SOL_SOCKET, SO_SNDBUF, &opt, optlen) != 0) 179 DWARN("%s: setsockopt SO_SNDBUF to %d failed", __func__, opt); 180 181 if (setsockopt(sv[0], SOL_SOCKET, SO_RCVBUF, &opt, optlen) != 0) 182 DWARN("%s: setsockopt SO_RCVBUF to %d failed", __func__, opt); 183 184 sa = (struct sockaddr *)(void *)&sun; 185 sun.sun_len = sizeof(sun); 186 sun.sun_family = AF_LOCAL; 187 (void)strcpy(sun.sun_path, path); 188 189 if (connect(sv[0], sa, (socklen_t)sun.sun_len) == 0) 190 return sv[0]; 191 192 /* 193 * Attempt to run perfused on our own 194 * if it does not run yet; In that case 195 * we will talk using a socketpair 196 * instead of /dev/fuse. 197 */ 198 if (socketpair(PF_LOCAL, sock_type, 0, sv) != 0) { 199 DWARN("%s:%d: socketpair failed", __func__, __LINE__); 200 return -1; 201 } 202 203 /* 204 * Set a buffer lentgh large enough so that any FUSE packet 205 * will fit. 206 */ 207 opt = (uint32_t)(4 * FUSE_BUFSIZE); 208 optlen = sizeof(opt); 209 if (setsockopt(sv[0], SOL_SOCKET, SO_SNDBUF, &opt, optlen) != 0) 210 DWARN("%s: setsockopt SO_SNDBUF to %d failed", __func__, opt); 211 212 if (setsockopt(sv[0], SOL_SOCKET, SO_RCVBUF, &opt, optlen) != 0) 213 DWARN("%s: setsockopt SO_RCVBUF to %d failed", __func__, opt); 214 215 if (setsockopt(sv[1], SOL_SOCKET, SO_SNDBUF, &opt, optlen) != 0) 216 DWARN("%s: setsockopt SO_SNDBUF to %d failed", __func__, opt); 217 218 if (setsockopt(sv[1], SOL_SOCKET, SO_RCVBUF, &opt, optlen) != 0) 219 DWARN("%s: setsockopt SO_RCVBUF to %d failed", __func__, opt); 220 221 /* 222 * Request peer credentials. This musr be done before first 223 * frame is sent. 224 */ 225 opt = 1; 226 optlen = sizeof(opt); 227 if (setsockopt(sv[1], 0, LOCAL_CREDS, &opt, optlen) != 0) 228 DWARN("%s: setsockopt LOCAL_CREDS failed", __func__); 229 230 (void)sprintf(fdstr, "%d", sv[1]); 231 232 switch(fork()) { 233 case -1: 234 #ifdef PERFUSE_DEBUG 235 DWARN("%s:%d: fork failed", __func__, __LINE__); 236 #endif 237 return -1; 238 /* NOTREACHED */ 239 break; 240 case 0: 241 (void)close(sv[0]); 242 (void)execve(argv[0], argv, environ); 243 #ifdef PERFUSE_DEBUG 244 DWARN("%s:%d: execve failed", __func__, __LINE__); 245 #endif 246 return -1; 247 /* NOTREACHED */ 248 break; 249 default: 250 break; 251 } 252 253 (void)close(sv[1]); 254 return sv[0]; 255 } 256 257 int 258 perfuse_mount(const char *source, const char *target, 259 const char *filesystemtype, long mountflags, const void *data) 260 { 261 int s; 262 size_t len; 263 struct perfuse_mount_out *pmo; 264 struct sockaddr_storage ss; 265 struct sockaddr_un *sun; 266 struct sockaddr *sa; 267 socklen_t sa_len; 268 size_t sock_len; 269 char *frame; 270 char *cp; 271 272 #ifdef PERFUSE_DEBUG 273 if (perfuse_diagflags & PDF_MISC) 274 DPRINTF("%s(\"%s\", \"%s\", \"%s\", 0x%lx, \"%s\")\n", 275 __func__, source, target, filesystemtype, 276 mountflags, (const char *)data); 277 #endif 278 279 if ((s = get_fd(data)) == -1) 280 return -1; 281 282 /* 283 * If we are connected to /dev/fuse, we need a second 284 * socket to get replies from perfused. 285 * XXX This socket is not removed at exit time yet 286 */ 287 sock_len = 0; 288 sa = (struct sockaddr *)(void *)&ss; 289 sun = (struct sockaddr_un *)(void *)&ss; 290 sa_len = sizeof(ss); 291 if ((getpeername(s, sa, &sa_len) == 0) && 292 (sa->sa_family = AF_LOCAL) && 293 (strcmp(sun->sun_path, _PATH_FUSE) == 0)) { 294 295 sun->sun_len = sizeof(*sun); 296 sun->sun_family = AF_LOCAL; 297 (void)sprintf(sun->sun_path, "%s/%s-%d", 298 _PATH_TMP, getprogname(), getpid()); 299 300 if (bind(s, sa, (socklen_t)sa->sa_len) != 0) 301 DERR(EX_OSERR, "%s:%d bind to \"%s\" failed", 302 __func__, __LINE__, sun->sun_path); 303 304 sock_len = strlen(sun->sun_path) + 1; 305 } 306 307 len = sizeof(*pmo); 308 len += source ? (uint32_t)strlen(source) + 1 : 0; 309 len += target ? (uint32_t)strlen(target) + 1 : 0; 310 len += filesystemtype ? (uint32_t)strlen(filesystemtype) + 1 : 0; 311 len += data ? (uint32_t)strlen(data) + 1 : 0; 312 len += sock_len; 313 314 if ((frame = malloc(len)) == NULL) { 315 #ifdef PERFUSE_DEBUG 316 if (perfuse_diagflags & PDF_MISC) 317 DWARN("%s:%d malloc failed", __func__, __LINE__); 318 #endif 319 return -1; 320 } 321 322 pmo = (struct perfuse_mount_out *)(void *)frame; 323 pmo->pmo_len = (uint32_t)len; 324 pmo->pmo_error = 0; 325 pmo->pmo_unique = (uint64_t)-1; 326 (void)strcpy(pmo->pmo_magic, PERFUSE_MOUNT_MAGIC); 327 328 pmo->pmo_source_len = source ? (uint32_t)strlen(source) + 1 : 0; 329 pmo->pmo_target_len = target ? (uint32_t)strlen(target) + 1: 0; 330 pmo->pmo_filesystemtype_len = 331 filesystemtype ? (uint32_t)strlen(filesystemtype) + 1 : 0; 332 pmo->pmo_mountflags = (uint32_t)mountflags; 333 pmo->pmo_data_len = data ? (uint32_t)strlen(data) + 1 : 0; 334 pmo->pmo_sock_len = (uint32_t)sock_len; 335 336 cp = (char *)(void *)(pmo + 1); 337 338 if (source) { 339 (void)strcpy(cp, source); 340 cp += pmo->pmo_source_len; 341 } 342 343 if (target) { 344 (void)strcpy(cp, target); 345 cp += pmo->pmo_target_len; 346 } 347 348 if (filesystemtype) { 349 (void)strcpy(cp, filesystemtype); 350 cp += pmo->pmo_filesystemtype_len; 351 } 352 353 if (data) { 354 (void)strcpy(cp, data); 355 cp += pmo->pmo_data_len; 356 } 357 358 if (sock_len != 0) { 359 (void)strcpy(cp, sun->sun_path); 360 cp += pmo->pmo_sock_len; 361 } 362 363 if (send(s, frame, len, MSG_NOSIGNAL) != (ssize_t)len) { 364 #ifdef PERFUSE_DEBUG 365 DWARN("%s:%d sendto failed", __func__, __LINE__); 366 #endif 367 return -1; 368 } 369 370 return 0; 371 } 372 373 374 uint64_t 375 perfuse_next_unique(struct puffs_usermount *pu) 376 { 377 struct perfuse_state *ps; 378 379 ps = puffs_getspecific(pu); 380 381 return ps->ps_unique++; 382 } 383 384 struct puffs_usermount * 385 perfuse_init(struct perfuse_callbacks *pc, struct perfuse_mount_info *pmi) 386 { 387 struct perfuse_state *ps; 388 struct puffs_usermount *pu; 389 struct puffs_ops *pops; 390 const char *source = _PATH_PUFFS; 391 char *fstype; 392 unsigned int puffs_flags; 393 struct puffs_node *pn_root; 394 struct puffs_pathobj *po_root; 395 struct rlimit rl; 396 397 /* 398 * perfused can grow quite large, let assume there's enough ram ... 399 */ 400 if (getrlimit(RLIMIT_DATA, &rl) < 0) { 401 DERR(EX_OSERR, "%s: getrlimit failed: %s", __func__, 402 strerror(errno)); 403 } else { 404 rl.rlim_cur = rl.rlim_max; 405 if (setrlimit(RLIMIT_DATA, &rl) < 0) { 406 DERR(EX_OSERR, "%s: setrlimit failed: %s", __func__, 407 strerror(errno)); 408 } 409 } 410 411 412 ps = init_state(); 413 ps->ps_owner_uid = pmi->pmi_uid; 414 415 if (pmi->pmi_source) { 416 if ((ps->ps_source = strdup(pmi->pmi_source)) == NULL) 417 DERR(EX_OSERR, "%s: strdup failed", __func__); 418 419 source = ps->ps_source; 420 } 421 422 if (pmi->pmi_filesystemtype) { 423 size_t len; 424 425 ps->ps_filesystemtype = strdup(pmi->pmi_filesystemtype); 426 if (ps->ps_filesystemtype == NULL) 427 DERR(EX_OSERR, "%s: strdup failed", __func__); 428 429 len = sizeof("perfuse|") + strlen(ps->ps_filesystemtype) + 1; 430 if ((fstype = malloc(len)) == NULL) 431 DERR(EX_OSERR, "%s: malloc failed", __func__); 432 433 (void)sprintf(fstype, "perfuse|%s", ps->ps_filesystemtype); 434 } else { 435 if ((fstype = strdup("perfuse")) == NULL) 436 DERR(EX_OSERR, "%s: strdup failed", __func__); 437 } 438 439 if ((ps->ps_target = strdup(pmi->pmi_target)) == NULL) 440 DERR(EX_OSERR, "%s: strdup failed", __func__); 441 442 ps->ps_mountflags = pmi->pmi_mountflags; 443 444 /* 445 * Some options are forbidden for non root users 446 */ 447 if (ps->ps_owner_uid != 0) 448 ps->ps_mountflags |= MNT_NOSUID|MNT_NODEV; 449 450 PUFFSOP_INIT(pops); 451 PUFFSOP_SET(pops, perfuse, fs, unmount); 452 PUFFSOP_SET(pops, perfuse, fs, statvfs); 453 PUFFSOP_SET(pops, perfuse, fs, sync); 454 PUFFSOP_SET(pops, perfuse, node, lookup); 455 PUFFSOP_SET(pops, perfuse, node, create); 456 PUFFSOP_SET(pops, perfuse, node, mknod); 457 PUFFSOP_SET(pops, perfuse, node, open); 458 PUFFSOP_SET(pops, perfuse, node, close); 459 PUFFSOP_SET(pops, perfuse, node, access); 460 PUFFSOP_SET(pops, perfuse, node, getattr); 461 PUFFSOP_SET(pops, perfuse, node, setattr); 462 PUFFSOP_SET(pops, perfuse, node, poll); 463 #if 0 464 PUFFSOP_SET(pops, perfuse, node, mmap); 465 #endif 466 PUFFSOP_SET(pops, perfuse, node, fsync); 467 PUFFSOP_SET(pops, perfuse, node, seek); 468 PUFFSOP_SET(pops, perfuse, node, remove); 469 PUFFSOP_SET(pops, perfuse, node, link); 470 PUFFSOP_SET(pops, perfuse, node, rename); 471 PUFFSOP_SET(pops, perfuse, node, mkdir); 472 PUFFSOP_SET(pops, perfuse, node, rmdir); 473 PUFFSOP_SET(pops, perfuse, node, symlink); 474 PUFFSOP_SET(pops, perfuse, node, readdir); 475 PUFFSOP_SET(pops, perfuse, node, readlink); 476 PUFFSOP_SET(pops, perfuse, node, reclaim); 477 PUFFSOP_SET(pops, perfuse, node, inactive); 478 PUFFSOP_SET(pops, perfuse, node, print); 479 PUFFSOP_SET(pops, perfuse, node, advlock); 480 PUFFSOP_SET(pops, perfuse, node, read); 481 PUFFSOP_SET(pops, perfuse, node, write); 482 #ifdef PUFFS_EXTNAMELEN 483 PUFFSOP_SET(pops, perfuse, node, getextattr); 484 PUFFSOP_SET(pops, perfuse, node, setextattr); 485 PUFFSOP_SET(pops, perfuse, node, listextattr); 486 PUFFSOP_SET(pops, perfuse, node, deleteextattr); 487 #endif /* PUFFS_EXTNAMELEN */ 488 489 /* 490 * We used to have PUFFS_KFLAG_WTCACHE here, which uses the 491 * page cache (highly desirable to get mmap(2)), but still sends 492 * all writes to the filesystem. In fact it does not send the 493 * data written, but the pages that contain it. 494 * 495 * There is a nasty bug hidden somewhere, possibly in libpuffs' 496 * VOP_FSYNC, which sends an asynchronous PUFFS_SETATTR that 497 * update file size. When writes are in progress, it will cause 498 * the file to be truncated and we get a zero-filled chunk at the 499 * beginning of a page. Removing PUFFS_KFLAG_WTCACHE fixes that 500 * problem. 501 * 502 * The other consequences are that changes will not be propagated 503 * immediatly to the filesystem, and we get a huge performance gain 504 * because much less requests are sent. A test case for the above 505 * mentioned bug got its execution time slashed by factor 50. 506 * 507 * PUFFS_KFLAG_NOCACHE_NAME is required so that we can see changes 508 * done by other machines in networked filesystems. 509 */ 510 puffs_flags = PUFFS_KFLAG_NOCACHE_NAME; 511 512 if (perfuse_diagflags & PDF_PUFFS) 513 puffs_flags |= PUFFS_FLAG_OPDUMP; 514 515 if ((pu = puffs_init(pops, source, fstype, ps, puffs_flags)) == NULL) 516 DERR(EX_OSERR, "%s: puffs_init failed", __func__); 517 518 ps->ps_pu = pu; 519 520 /* 521 * Setup filesystem root 522 */ 523 pn_root = perfuse_new_pn(pu, "", NULL); 524 PERFUSE_NODE_DATA(pn_root)->pnd_nodeid = FUSE_ROOT_ID; 525 PERFUSE_NODE_DATA(pn_root)->pnd_parent = pn_root; 526 puffs_setroot(pu, pn_root); 527 ps->ps_fsid = pn_root->pn_va.va_fsid; 528 529 po_root = puffs_getrootpathobj(pu); 530 if ((po_root->po_path = strdup("/")) == NULL) 531 DERRX(EX_OSERR, "perfuse_mount_start() failed"); 532 533 po_root->po_len = 1; 534 puffs_path_buildhash(pu, po_root); 535 536 puffs_vattr_null(&pn_root->pn_va); 537 pn_root->pn_va.va_type = VDIR; 538 pn_root->pn_va.va_mode = 0755; 539 pn_root->pn_va.va_fileid = FUSE_ROOT_ID; 540 541 ps->ps_root = pn_root; 542 543 /* 544 * Callbacks 545 */ 546 ps->ps_new_msg = pc->pc_new_msg; 547 ps->ps_xchg_msg = pc->pc_xchg_msg; 548 ps->ps_destroy_msg = pc->pc_destroy_msg; 549 ps->ps_get_inhdr = pc->pc_get_inhdr; 550 ps->ps_get_inpayload = pc->pc_get_inpayload; 551 ps->ps_get_outhdr = pc->pc_get_outhdr; 552 ps->ps_get_outpayload = pc->pc_get_outpayload; 553 ps->ps_umount = pc->pc_umount; 554 555 return pu; 556 } 557 558 void 559 perfuse_setspecific(struct puffs_usermount *pu, void *priv) 560 { 561 struct perfuse_state *ps; 562 563 ps = puffs_getspecific(pu); 564 ps->ps_private = priv; 565 566 return; 567 } 568 569 void * 570 perfuse_getspecific(struct puffs_usermount *pu) 571 { 572 struct perfuse_state *ps; 573 574 ps = puffs_getspecific(pu); 575 576 return ps->ps_private; 577 } 578 579 int 580 perfuse_inloop(struct puffs_usermount *pu) 581 { 582 struct perfuse_state *ps; 583 584 ps = puffs_getspecific(pu); 585 586 return ps->ps_flags & PS_INLOOP; 587 } 588 589 int 590 perfuse_mainloop(struct puffs_usermount *pu) 591 { 592 struct perfuse_state *ps; 593 594 ps = puffs_getspecific(pu); 595 596 ps->ps_flags |= PS_INLOOP; 597 if (puffs_mainloop(ps->ps_pu) != 0) { 598 DERR(EX_OSERR, "%s: failed", __func__); 599 return -1; 600 } 601 602 /* 603 * Normal exit after unmount 604 */ 605 return 0; 606 } 607 608 /* ARGSUSED0 */ 609 uint64_t 610 perfuse_get_nodeid(struct puffs_usermount *pu, puffs_cookie_t opc) 611 { 612 return PERFUSE_NODE_DATA(opc)->pnd_nodeid; 613 } 614 615 int 616 perfuse_unmount(struct puffs_usermount *pu) 617 { 618 struct perfuse_state *ps; 619 620 ps = puffs_getspecific(pu); 621 622 return unmount(ps->ps_target, MNT_FORCE); 623 } 624