1 /* $NetBSD: perfuse.c,v 1.24 2011/12/28 17:33:53 manu 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(data) 109 const char *data; 110 { 111 char *string; 112 const char fdopt[] = "fd="; 113 char *lastp; 114 char *opt; 115 int fd = -1; 116 117 if ((string = strdup(data)) == NULL) 118 return -1; 119 120 for (opt = strtok_r(string, ",", &lastp); 121 opt != NULL; 122 opt = strtok_r(NULL, ",", &lastp)) { 123 if (strncmp(opt, fdopt, strlen(fdopt)) == 0) { 124 fd = atoi(opt + strlen(fdopt)); 125 break; 126 } 127 } 128 129 /* 130 * No file descriptor found 131 */ 132 if (fd == -1) 133 errno = EINVAL; 134 135 free(string); 136 return fd; 137 138 } 139 140 int 141 perfuse_open(path, flags, mode) 142 const char *path; 143 int flags; 144 mode_t mode; 145 { 146 int sv[2]; 147 struct sockaddr_un sun; 148 struct sockaddr *sa; 149 char progname[] = _PATH_PERFUSED; 150 char minus_i[] = "-i"; 151 char fdstr[16]; 152 char *const argv[] = { progname, minus_i, fdstr, NULL}; 153 uint32_t opt; 154 uint32_t optlen; 155 int sock_type = SOCK_SEQPACKET; 156 157 if (strcmp(path, _PATH_FUSE) != 0) 158 return open(path, flags, mode); 159 160 /* 161 * Try SOCK_SEQPACKET then SOCK_DGRAM if unavailable 162 */ 163 if ((sv[0] = socket(PF_LOCAL, SOCK_SEQPACKET, 0)) == -1) { 164 sock_type = SOCK_DGRAM; 165 DWARNX("SEQPACKET local sockets unavailable, using less " 166 "reliable DGRAM sockets. Expect file operation hangs."); 167 168 if ((sv[0] = socket(PF_LOCAL, SOCK_DGRAM, 0)) == -1) { 169 #ifdef PERFUSE_DEBUG 170 DWARN("%s: %d socket failed", __func__, __LINE__); 171 #endif 172 return -1; 173 } 174 } 175 176 /* 177 * Set a buffer lentgh large enough so that any FUSE packet 178 * will fit. 179 */ 180 opt = (uint32_t)FUSE_BUFSIZE; 181 optlen = sizeof(opt); 182 if (setsockopt(sv[0], SOL_SOCKET, SO_SNDBUF, &opt, optlen) != 0) 183 DWARN("%s: setsockopt SO_SNDBUF to %d failed", __func__, opt); 184 185 if (setsockopt(sv[0], SOL_SOCKET, SO_RCVBUF, &opt, optlen) != 0) 186 DWARN("%s: setsockopt SO_RCVBUF to %d failed", __func__, opt); 187 188 sa = (struct sockaddr *)(void *)&sun; 189 sun.sun_len = sizeof(sun); 190 sun.sun_family = AF_LOCAL; 191 (void)strcpy(sun.sun_path, path); 192 193 if (connect(sv[0], sa, (socklen_t)sun.sun_len) == 0) 194 return sv[0]; 195 196 /* 197 * Attempt to run perfused on our own 198 * if it does not run yet; In that case 199 * we will talk using a socketpair 200 * instead of /dev/fuse. 201 */ 202 if (socketpair(PF_LOCAL, sock_type, 0, sv) != 0) { 203 DWARN("%s:%d: socketpair failed", __func__, __LINE__); 204 return -1; 205 } 206 207 /* 208 * Set a buffer lentgh large enough so that any FUSE packet 209 * will fit. 210 */ 211 opt = (uint32_t)(4 * FUSE_BUFSIZE); 212 optlen = sizeof(opt); 213 if (setsockopt(sv[0], SOL_SOCKET, SO_SNDBUF, &opt, optlen) != 0) 214 DWARN("%s: setsockopt SO_SNDBUF to %d failed", __func__, opt); 215 216 if (setsockopt(sv[0], SOL_SOCKET, SO_RCVBUF, &opt, optlen) != 0) 217 DWARN("%s: setsockopt SO_RCVBUF to %d failed", __func__, opt); 218 219 if (setsockopt(sv[1], SOL_SOCKET, SO_SNDBUF, &opt, optlen) != 0) 220 DWARN("%s: setsockopt SO_SNDBUF to %d failed", __func__, opt); 221 222 if (setsockopt(sv[1], SOL_SOCKET, SO_RCVBUF, &opt, optlen) != 0) 223 DWARN("%s: setsockopt SO_RCVBUF to %d failed", __func__, opt); 224 225 /* 226 * Request peer credentials. This musr be done before first 227 * frame is sent. 228 */ 229 opt = 1; 230 optlen = sizeof(opt); 231 if (setsockopt(sv[1], 0, LOCAL_CREDS, &opt, optlen) != 0) 232 DWARN("%s: setsockopt LOCAL_CREDS failed", __func__); 233 234 (void)sprintf(fdstr, "%d", sv[1]); 235 236 switch(fork()) { 237 case -1: 238 #ifdef PERFUSE_DEBUG 239 DWARN("%s:%d: fork failed", __func__, __LINE__); 240 #endif 241 return -1; 242 /* NOTREACHED */ 243 break; 244 case 0: 245 (void)execve(argv[0], argv, environ); 246 #ifdef PERFUSE_DEBUG 247 DWARN("%s:%d: execve failed", __func__, __LINE__); 248 #endif 249 return -1; 250 /* NOTREACHED */ 251 break; 252 default: 253 break; 254 } 255 256 return sv[0]; 257 } 258 259 int 260 perfuse_mount(source, target, filesystemtype, mountflags, data) 261 const char *source; 262 const char *target; 263 const char *filesystemtype; 264 long mountflags; 265 const void *data; 266 { 267 int s; 268 size_t len; 269 struct perfuse_mount_out *pmo; 270 struct sockaddr_storage ss; 271 struct sockaddr_un *sun; 272 struct sockaddr *sa; 273 socklen_t sa_len; 274 size_t sock_len; 275 char *frame; 276 char *cp; 277 278 #ifdef PERFUSE_DEBUG 279 if (perfuse_diagflags & PDF_MISC) 280 DPRINTF("%s(\"%s\", \"%s\", \"%s\", 0x%lx, \"%s\")\n", 281 __func__, source, target, filesystemtype, 282 mountflags, (const char *)data); 283 #endif 284 285 if ((s = get_fd(data)) == -1) 286 return -1; 287 288 /* 289 * If we are connected to /dev/fuse, we need a second 290 * socket to get replies from perfused. 291 * XXX This socket is not removed at exit time yet 292 */ 293 sock_len = 0; 294 sa = (struct sockaddr *)(void *)&ss; 295 sun = (struct sockaddr_un *)(void *)&ss; 296 sa_len = sizeof(ss); 297 if ((getpeername(s, sa, &sa_len) == 0) && 298 (sa->sa_family = AF_LOCAL) && 299 (strcmp(sun->sun_path, _PATH_FUSE) == 0)) { 300 301 sun->sun_len = sizeof(*sun); 302 sun->sun_family = AF_LOCAL; 303 (void)sprintf(sun->sun_path, "%s/%s-%d", 304 _PATH_TMP, getprogname(), getpid()); 305 306 if (bind(s, sa, (socklen_t)sa->sa_len) != 0) 307 DERR(EX_OSERR, "%s:%d bind to \"%s\" failed", 308 __func__, __LINE__, sun->sun_path); 309 310 sock_len = strlen(sun->sun_path) + 1; 311 } 312 313 len = sizeof(*pmo); 314 len += source ? (uint32_t)strlen(source) + 1 : 0; 315 len += target ? (uint32_t)strlen(target) + 1 : 0; 316 len += filesystemtype ? (uint32_t)strlen(filesystemtype) + 1 : 0; 317 len += data ? (uint32_t)strlen(data) + 1 : 0; 318 len += sock_len; 319 320 if ((frame = malloc(len)) == NULL) { 321 #ifdef PERFUSE_DEBUG 322 if (perfuse_diagflags & PDF_MISC) 323 DWARN("%s:%d malloc failed", __func__, __LINE__); 324 #endif 325 return -1; 326 } 327 328 pmo = (struct perfuse_mount_out *)(void *)frame; 329 pmo->pmo_len = (uint32_t)len; 330 pmo->pmo_error = 0; 331 pmo->pmo_unique = (uint64_t)-1; 332 (void)strcpy(pmo->pmo_magic, PERFUSE_MOUNT_MAGIC); 333 334 pmo->pmo_source_len = source ? (uint32_t)strlen(source) + 1 : 0; 335 pmo->pmo_target_len = target ? (uint32_t)strlen(target) + 1: 0; 336 pmo->pmo_filesystemtype_len = 337 filesystemtype ? (uint32_t)strlen(filesystemtype) + 1 : 0; 338 pmo->pmo_mountflags = (uint32_t)mountflags; 339 pmo->pmo_data_len = data ? (uint32_t)strlen(data) + 1 : 0; 340 pmo->pmo_sock_len = (uint32_t)sock_len; 341 342 cp = (char *)(void *)(pmo + 1); 343 344 if (source) { 345 (void)strcpy(cp, source); 346 cp += pmo->pmo_source_len; 347 } 348 349 if (target) { 350 (void)strcpy(cp, target); 351 cp += pmo->pmo_target_len; 352 } 353 354 if (filesystemtype) { 355 (void)strcpy(cp, filesystemtype); 356 cp += pmo->pmo_filesystemtype_len; 357 } 358 359 if (data) { 360 (void)strcpy(cp, data); 361 cp += pmo->pmo_data_len; 362 } 363 364 if (sock_len != 0) { 365 (void)strcpy(cp, sun->sun_path); 366 cp += pmo->pmo_sock_len; 367 } 368 369 if (send(s, frame, len, MSG_NOSIGNAL) != (ssize_t)len) { 370 #ifdef PERFUSE_DEBUG 371 DWARN("%s:%d sendto failed", __func__, __LINE__); 372 #endif 373 return -1; 374 } 375 376 return 0; 377 } 378 379 380 uint64_t 381 perfuse_next_unique(pu) 382 struct puffs_usermount *pu; 383 { 384 struct perfuse_state *ps; 385 386 ps = puffs_getspecific(pu); 387 388 return ps->ps_unique++; 389 } 390 391 struct puffs_usermount * 392 perfuse_init(pc, pmi) 393 struct perfuse_callbacks *pc; 394 struct perfuse_mount_info *pmi; 395 { 396 struct perfuse_state *ps; 397 struct puffs_usermount *pu; 398 struct puffs_ops *pops; 399 const char *source = _PATH_PUFFS; 400 char *fstype; 401 unsigned int puffs_flags; 402 struct puffs_node *pn_root; 403 struct puffs_pathobj *po_root; 404 struct rlimit rl; 405 406 /* 407 * perfused can grow quite large, let assume there's enough ram ... 408 */ 409 if (getrlimit(RLIMIT_DATA, &rl) < 0) { 410 DERR(EX_OSERR, "%s: getrlimit failed: %s", __func__, 411 strerror(errno)); 412 } else { 413 rl.rlim_cur = rl.rlim_max; 414 if (setrlimit(RLIMIT_DATA, &rl) < 0) { 415 DERR(EX_OSERR, "%s: setrlimit failed: %s", __func__, 416 strerror(errno)); 417 } 418 } 419 420 421 ps = init_state(); 422 ps->ps_owner_uid = pmi->pmi_uid; 423 424 if (pmi->pmi_source) { 425 if ((ps->ps_source = strdup(pmi->pmi_source)) == NULL) 426 DERR(EX_OSERR, "%s: strdup failed", __func__); 427 428 source = ps->ps_source; 429 } 430 431 if (pmi->pmi_filesystemtype) { 432 size_t len; 433 434 ps->ps_filesystemtype = strdup(pmi->pmi_filesystemtype); 435 if (ps->ps_filesystemtype == NULL) 436 DERR(EX_OSERR, "%s: strdup failed", __func__); 437 438 len = sizeof("perfuse|") + strlen(ps->ps_filesystemtype) + 1; 439 if ((fstype = malloc(len)) == NULL) 440 DERR(EX_OSERR, "%s: malloc failed", __func__); 441 442 (void)sprintf(fstype, "perfuse|%s", ps->ps_filesystemtype); 443 } else { 444 if ((fstype = strdup("perfuse")) == NULL) 445 DERR(EX_OSERR, "%s: strdup failed", __func__); 446 } 447 448 if ((ps->ps_target = strdup(pmi->pmi_target)) == NULL) 449 DERR(EX_OSERR, "%s: strdup failed", __func__); 450 451 ps->ps_mountflags = pmi->pmi_mountflags; 452 453 /* 454 * Some options are forbidden for non root users 455 */ 456 if (ps->ps_owner_uid != 0) 457 ps->ps_mountflags |= MNT_NOSUID|MNT_NODEV; 458 459 PUFFSOP_INIT(pops); 460 PUFFSOP_SET(pops, perfuse, fs, unmount); 461 PUFFSOP_SET(pops, perfuse, fs, statvfs); 462 PUFFSOP_SET(pops, perfuse, fs, sync); 463 PUFFSOP_SET(pops, perfuse, node, lookup); 464 PUFFSOP_SET(pops, perfuse, node, create); 465 PUFFSOP_SET(pops, perfuse, node, mknod); 466 PUFFSOP_SET(pops, perfuse, node, open); 467 PUFFSOP_SET(pops, perfuse, node, close); 468 PUFFSOP_SET(pops, perfuse, node, access); 469 PUFFSOP_SET(pops, perfuse, node, getattr); 470 PUFFSOP_SET(pops, perfuse, node, setattr); 471 PUFFSOP_SET(pops, perfuse, node, poll); 472 #if 0 473 PUFFSOP_SET(pops, perfuse, node, mmap); 474 #endif 475 PUFFSOP_SET(pops, perfuse, node, fsync); 476 PUFFSOP_SET(pops, perfuse, node, seek); 477 PUFFSOP_SET(pops, perfuse, node, remove); 478 PUFFSOP_SET(pops, perfuse, node, link); 479 PUFFSOP_SET(pops, perfuse, node, rename); 480 PUFFSOP_SET(pops, perfuse, node, mkdir); 481 PUFFSOP_SET(pops, perfuse, node, rmdir); 482 PUFFSOP_SET(pops, perfuse, node, symlink); 483 PUFFSOP_SET(pops, perfuse, node, readdir); 484 PUFFSOP_SET(pops, perfuse, node, readlink); 485 PUFFSOP_SET(pops, perfuse, node, reclaim); 486 PUFFSOP_SET(pops, perfuse, node, inactive); 487 PUFFSOP_SET(pops, perfuse, node, print); 488 PUFFSOP_SET(pops, perfuse, node, advlock); 489 PUFFSOP_SET(pops, perfuse, node, read); 490 PUFFSOP_SET(pops, perfuse, node, write); 491 #ifdef PUFFS_EXTNAMELEN 492 PUFFSOP_SET(pops, perfuse, node, getextattr); 493 PUFFSOP_SET(pops, perfuse, node, setextattr); 494 PUFFSOP_SET(pops, perfuse, node, listextattr); 495 PUFFSOP_SET(pops, perfuse, node, deleteextattr); 496 #endif /* PUFFS_EXTNAMELEN */ 497 498 /* 499 * We used to have PUFFS_KFLAG_WTCACHE here, which uses the 500 * page cache (highly desirable to get mmap(2)), but still sends 501 * all writes to the filesystem. In fact it does not send the 502 * data written, but the pages that contain it. 503 * 504 * There is a nasty bug hidden somewhere, possibly in libpuffs' 505 * VOP_FSYNC, which sends an asynchronous PUFFS_SETATTR that 506 * update file size. When writes are in progress, it will cause 507 * the file to be truncated and we get a zero-filled chunk at the 508 * beginning of a page. Removing PUFFS_KFLAG_WTCACHE fixes that 509 * problem. 510 * 511 * The other consequences are that changes will not be propagated 512 * immediatly to the filesystem, and we get a huge performance gain 513 * because much less requests are sent. A test case for the above 514 * mentioned bug got its execution time slashed by factor 50. 515 * 516 * PUFFS_KFLAG_NOCACHE_NAME is required so that we can see changes 517 * done by other machines in networked filesystems. 518 */ 519 puffs_flags = PUFFS_KFLAG_NOCACHE_NAME; 520 521 if (perfuse_diagflags & PDF_PUFFS) 522 puffs_flags |= PUFFS_FLAG_OPDUMP; 523 524 if ((pu = puffs_init(pops, source, fstype, ps, puffs_flags)) == NULL) 525 DERR(EX_OSERR, "%s: puffs_init failed", __func__); 526 527 ps->ps_pu = pu; 528 529 /* 530 * Setup filesystem root 531 */ 532 pn_root = perfuse_new_pn(pu, "", NULL); 533 PERFUSE_NODE_DATA(pn_root)->pnd_nodeid = FUSE_ROOT_ID; 534 PERFUSE_NODE_DATA(pn_root)->pnd_parent = pn_root; 535 puffs_setroot(pu, pn_root); 536 ps->ps_fsid = pn_root->pn_va.va_fsid; 537 538 po_root = puffs_getrootpathobj(pu); 539 if ((po_root->po_path = strdup("/")) == NULL) 540 DERRX(EX_OSERR, "perfuse_mount_start() failed"); 541 542 po_root->po_len = 1; 543 puffs_path_buildhash(pu, po_root); 544 545 puffs_vattr_null(&pn_root->pn_va); 546 pn_root->pn_va.va_type = VDIR; 547 pn_root->pn_va.va_mode = 0755; 548 pn_root->pn_va.va_fileid = FUSE_ROOT_ID; 549 550 ps->ps_root = pn_root; 551 552 /* 553 * Callbacks 554 */ 555 ps->ps_new_msg = pc->pc_new_msg; 556 ps->ps_xchg_msg = pc->pc_xchg_msg; 557 ps->ps_destroy_msg = pc->pc_destroy_msg; 558 ps->ps_get_inhdr = pc->pc_get_inhdr; 559 ps->ps_get_inpayload = pc->pc_get_inpayload; 560 ps->ps_get_outhdr = pc->pc_get_outhdr; 561 ps->ps_get_outpayload = pc->pc_get_outpayload; 562 ps->ps_umount = pc->pc_umount; 563 564 return pu; 565 } 566 567 void 568 perfuse_setspecific(pu, priv) 569 struct puffs_usermount *pu; 570 void *priv; 571 { 572 struct perfuse_state *ps; 573 574 ps = puffs_getspecific(pu); 575 ps->ps_private = priv; 576 577 return; 578 } 579 580 void * 581 perfuse_getspecific(pu) 582 struct puffs_usermount *pu; 583 { 584 struct perfuse_state *ps; 585 586 ps = puffs_getspecific(pu); 587 588 return ps->ps_private; 589 } 590 591 int 592 perfuse_inloop(pu) 593 struct puffs_usermount *pu; 594 { 595 struct perfuse_state *ps; 596 597 ps = puffs_getspecific(pu); 598 599 return ps->ps_flags & PS_INLOOP; 600 } 601 602 int 603 perfuse_mainloop(pu) 604 struct puffs_usermount *pu; 605 { 606 struct perfuse_state *ps; 607 608 ps = puffs_getspecific(pu); 609 610 ps->ps_flags |= PS_INLOOP; 611 if (puffs_mainloop(ps->ps_pu) != 0) { 612 DERR(EX_OSERR, "%s: failed", __func__); 613 return -1; 614 } 615 616 /* 617 * Normal exit after unmount 618 */ 619 return 0; 620 } 621 622 /* ARGSUSED0 */ 623 uint64_t 624 perfuse_get_nodeid(pu, opc) 625 struct puffs_usermount *pu; 626 puffs_cookie_t opc; 627 { 628 return PERFUSE_NODE_DATA(opc)->pnd_nodeid; 629 } 630 631 int 632 perfuse_unmount(pu) 633 struct puffs_usermount *pu; 634 { 635 struct perfuse_state *ps; 636 637 ps = puffs_getspecific(pu); 638 639 return unmount(ps->ps_target, MNT_FORCE); 640 } 641