1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2007-2009 Google Inc. and Amit Singh 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions are 9 * met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above 14 * copyright notice, this list of conditions and the following disclaimer 15 * in the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Google Inc. nor the names of its 18 * contributors may be used to endorse or promote products derived from 19 * this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Copyright (C) 2005 Csaba Henk. 34 * All rights reserved. 35 * 36 * Copyright (c) 2019 The FreeBSD Foundation 37 * 38 * Portions of this software were developed by BFF Storage Systems, LLC under 39 * sponsorship from the FreeBSD Foundation. 40 * 41 * Redistribution and use in source and binary forms, with or without 42 * modification, are permitted provided that the following conditions 43 * are met: 44 * 1. Redistributions of source code must retain the above copyright 45 * notice, this list of conditions and the following disclaimer. 46 * 2. Redistributions in binary form must reproduce the above copyright 47 * notice, this list of conditions and the following disclaimer in the 48 * documentation and/or other materials provided with the distribution. 49 * 50 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND 51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 53 * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE 54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 60 * SUCH DAMAGE. 61 */ 62 63 #include <sys/cdefs.h> 64 __FBSDID("$FreeBSD$"); 65 66 #include <sys/param.h> 67 #include <sys/systm.h> 68 #include <sys/counter.h> 69 #include <sys/module.h> 70 #include <sys/errno.h> 71 #include <sys/kernel.h> 72 #include <sys/conf.h> 73 #include <sys/uio.h> 74 #include <sys/malloc.h> 75 #include <sys/queue.h> 76 #include <sys/lock.h> 77 #include <sys/mutex.h> 78 #include <sys/sdt.h> 79 #include <sys/sx.h> 80 #include <sys/proc.h> 81 #include <sys/mount.h> 82 #include <sys/vnode.h> 83 #include <sys/namei.h> 84 #include <sys/stat.h> 85 #include <sys/unistd.h> 86 #include <sys/filedesc.h> 87 #include <sys/file.h> 88 #include <sys/fcntl.h> 89 #include <sys/dirent.h> 90 #include <sys/bio.h> 91 #include <sys/buf.h> 92 #include <sys/sysctl.h> 93 #include <sys/priv.h> 94 95 #include "fuse.h" 96 #include "fuse_file.h" 97 #include "fuse_internal.h" 98 #include "fuse_io.h" 99 #include "fuse_ipc.h" 100 #include "fuse_node.h" 101 #include "fuse_file.h" 102 103 SDT_PROVIDER_DECLARE(fusefs); 104 /* 105 * Fuse trace probe: 106 * arg0: verbosity. Higher numbers give more verbose messages 107 * arg1: Textual message 108 */ 109 SDT_PROBE_DEFINE2(fusefs, , internal, trace, "int", "char*"); 110 111 #ifdef ZERO_PAD_INCOMPLETE_BUFS 112 static int isbzero(void *buf, size_t len); 113 114 #endif 115 116 counter_u64_t fuse_lookup_cache_hits; 117 counter_u64_t fuse_lookup_cache_misses; 118 119 SYSCTL_COUNTER_U64(_vfs_fusefs_stats, OID_AUTO, lookup_cache_hits, CTLFLAG_RD, 120 &fuse_lookup_cache_hits, "number of positive cache hits in lookup"); 121 122 SYSCTL_COUNTER_U64(_vfs_fusefs_stats, OID_AUTO, lookup_cache_misses, CTLFLAG_RD, 123 &fuse_lookup_cache_misses, "number of cache misses in lookup"); 124 125 int 126 fuse_internal_get_cached_vnode(struct mount* mp, ino_t ino, int flags, 127 struct vnode **vpp) 128 { 129 struct bintime now; 130 struct thread *td = curthread; 131 uint64_t nodeid = ino; 132 int error; 133 134 *vpp = NULL; 135 136 error = vfs_hash_get(mp, fuse_vnode_hash(nodeid), flags, td, vpp, 137 fuse_vnode_cmp, &nodeid); 138 if (error) 139 return error; 140 /* 141 * Check the entry cache timeout. We have to do this within fusefs 142 * instead of by using cache_enter_time/cache_lookup because those 143 * routines are only intended to work with pathnames, not inodes 144 */ 145 if (*vpp != NULL) { 146 getbinuptime(&now); 147 if (bintime_cmp(&(VTOFUD(*vpp)->entry_cache_timeout), &now, >)){ 148 counter_u64_add(fuse_lookup_cache_hits, 1); 149 return 0; 150 } else { 151 /* Entry cache timeout */ 152 counter_u64_add(fuse_lookup_cache_misses, 1); 153 cache_purge(*vpp); 154 vput(*vpp); 155 *vpp = NULL; 156 } 157 } 158 return 0; 159 } 160 161 /* Synchronously send a FUSE_ACCESS operation */ 162 int 163 fuse_internal_access(struct vnode *vp, 164 accmode_t mode, 165 struct thread *td, 166 struct ucred *cred) 167 { 168 int err = 0; 169 uint32_t mask = F_OK; 170 int dataflags; 171 int vtype; 172 struct mount *mp; 173 struct fuse_dispatcher fdi; 174 struct fuse_access_in *fai; 175 struct fuse_data *data; 176 177 mp = vnode_mount(vp); 178 vtype = vnode_vtype(vp); 179 180 data = fuse_get_mpdata(mp); 181 dataflags = data->dataflags; 182 183 if (mode == 0) 184 return 0; 185 186 if (mode & VMODIFY_PERMS && vfs_isrdonly(mp)) { 187 switch (vp->v_type) { 188 case VDIR: 189 /* FALLTHROUGH */ 190 case VLNK: 191 /* FALLTHROUGH */ 192 case VREG: 193 return EROFS; 194 default: 195 break; 196 } 197 } 198 199 /* Unless explicitly permitted, deny everyone except the fs owner. */ 200 if (!(dataflags & FSESS_DAEMON_CAN_SPY)) { 201 if (fuse_match_cred(data->daemoncred, cred)) 202 return EPERM; 203 } 204 205 if (dataflags & FSESS_DEFAULT_PERMISSIONS) { 206 struct vattr va; 207 208 fuse_internal_getattr(vp, &va, cred, td); 209 return vaccess(vp->v_type, va.va_mode, va.va_uid, 210 va.va_gid, mode, cred, NULL); 211 } 212 213 if (!fsess_isimpl(mp, FUSE_ACCESS)) 214 return 0; 215 216 if ((mode & (VWRITE | VAPPEND | VADMIN)) != 0) 217 mask |= W_OK; 218 if ((mode & VREAD) != 0) 219 mask |= R_OK; 220 if ((mode & VEXEC) != 0) 221 mask |= X_OK; 222 223 fdisp_init(&fdi, sizeof(*fai)); 224 fdisp_make_vp(&fdi, FUSE_ACCESS, vp, td, cred); 225 226 fai = fdi.indata; 227 fai->mask = mask; 228 229 err = fdisp_wait_answ(&fdi); 230 fdisp_destroy(&fdi); 231 232 if (err == ENOSYS) { 233 fsess_set_notimpl(mp, FUSE_ACCESS); 234 err = 0; 235 } 236 return err; 237 } 238 239 /* 240 * Cache FUSE attributes from attr, in attribute cache associated with vnode 241 * 'vp'. Optionally, if argument 'vap' is not NULL, store a copy of the 242 * converted attributes there as well. 243 * 244 * If the nominal attribute cache TTL is zero, do not cache on the 'vp' (but do 245 * return the result to the caller). 246 */ 247 void 248 fuse_internal_cache_attrs(struct vnode *vp, struct fuse_attr *attr, 249 uint64_t attr_valid, uint32_t attr_valid_nsec, struct vattr *vap) 250 { 251 struct mount *mp; 252 struct fuse_vnode_data *fvdat; 253 struct fuse_data *data; 254 struct vattr *vp_cache_at; 255 256 mp = vnode_mount(vp); 257 fvdat = VTOFUD(vp); 258 data = fuse_get_mpdata(mp); 259 260 ASSERT_VOP_ELOCKED(vp, "fuse_internal_cache_attrs"); 261 262 fuse_validity_2_bintime(attr_valid, attr_valid_nsec, 263 &fvdat->attr_cache_timeout); 264 265 /* Fix our buffers if the filesize changed without us knowing */ 266 if (vnode_isreg(vp) && attr->size != fvdat->cached_attrs.va_size) { 267 (void)fuse_vnode_setsize(vp, attr->size); 268 fvdat->cached_attrs.va_size = attr->size; 269 } 270 271 if (attr_valid > 0 || attr_valid_nsec > 0) 272 vp_cache_at = &(fvdat->cached_attrs); 273 else if (vap != NULL) 274 vp_cache_at = vap; 275 else 276 return; 277 278 vattr_null(vp_cache_at); 279 vp_cache_at->va_fsid = mp->mnt_stat.f_fsid.val[0]; 280 vp_cache_at->va_fileid = attr->ino; 281 vp_cache_at->va_mode = attr->mode & ~S_IFMT; 282 vp_cache_at->va_nlink = attr->nlink; 283 vp_cache_at->va_uid = attr->uid; 284 vp_cache_at->va_gid = attr->gid; 285 vp_cache_at->va_rdev = attr->rdev; 286 vp_cache_at->va_size = attr->size; 287 /* XXX on i386, seconds are truncated to 32 bits */ 288 vp_cache_at->va_atime.tv_sec = attr->atime; 289 vp_cache_at->va_atime.tv_nsec = attr->atimensec; 290 vp_cache_at->va_mtime.tv_sec = attr->mtime; 291 vp_cache_at->va_mtime.tv_nsec = attr->mtimensec; 292 vp_cache_at->va_ctime.tv_sec = attr->ctime; 293 vp_cache_at->va_ctime.tv_nsec = attr->ctimensec; 294 if (fuse_libabi_geq(data, 7, 9) && attr->blksize > 0) 295 vp_cache_at->va_blocksize = attr->blksize; 296 else 297 vp_cache_at->va_blocksize = PAGE_SIZE; 298 vp_cache_at->va_type = IFTOVT(attr->mode); 299 vp_cache_at->va_bytes = attr->blocks * S_BLKSIZE; 300 vp_cache_at->va_flags = 0; 301 302 if (vap != vp_cache_at && vap != NULL) 303 memcpy(vap, vp_cache_at, sizeof(*vap)); 304 } 305 306 307 /* fsync */ 308 309 int 310 fuse_internal_fsync_callback(struct fuse_ticket *tick, struct uio *uio) 311 { 312 if (tick->tk_aw_ohead.error == ENOSYS) { 313 fsess_set_notimpl(tick->tk_data->mp, fticket_opcode(tick)); 314 } 315 return 0; 316 } 317 318 int 319 fuse_internal_fsync(struct vnode *vp, 320 struct thread *td, 321 int waitfor, 322 bool datasync) 323 { 324 struct fuse_fsync_in *ffsi = NULL; 325 struct fuse_dispatcher fdi; 326 struct fuse_filehandle *fufh; 327 struct fuse_vnode_data *fvdat = VTOFUD(vp); 328 struct mount *mp = vnode_mount(vp); 329 int op = FUSE_FSYNC; 330 int err = 0; 331 332 if (!fsess_isimpl(vnode_mount(vp), 333 (vnode_vtype(vp) == VDIR ? FUSE_FSYNCDIR : FUSE_FSYNC))) { 334 return 0; 335 } 336 if (vnode_isdir(vp)) 337 op = FUSE_FSYNCDIR; 338 339 if (!fsess_isimpl(mp, op)) 340 return 0; 341 342 fdisp_init(&fdi, sizeof(*ffsi)); 343 /* 344 * fsync every open file handle for this file, because we can't be sure 345 * which file handle the caller is really referring to. 346 */ 347 LIST_FOREACH(fufh, &fvdat->handles, next) { 348 if (ffsi == NULL) 349 fdisp_make_vp(&fdi, op, vp, td, NULL); 350 else 351 fdisp_refresh_vp(&fdi, op, vp, td, NULL); 352 ffsi = fdi.indata; 353 ffsi->fh = fufh->fh_id; 354 ffsi->fsync_flags = 0; 355 356 if (datasync) 357 ffsi->fsync_flags = 1; 358 359 if (waitfor == MNT_WAIT) { 360 err = fdisp_wait_answ(&fdi); 361 } else { 362 fuse_insert_callback(fdi.tick, 363 fuse_internal_fsync_callback); 364 fuse_insert_message(fdi.tick, false); 365 } 366 if (err == ENOSYS) { 367 /* ENOSYS means "success, and don't call again" */ 368 fsess_set_notimpl(mp, op); 369 err = 0; 370 break; 371 } 372 } 373 fdisp_destroy(&fdi); 374 375 return err; 376 } 377 378 /* Asynchronous invalidation */ 379 SDT_PROBE_DEFINE2(fusefs, , internal, invalidate_cache_hit, 380 "struct vnode*", "struct vnode*"); 381 int 382 fuse_internal_invalidate_entry(struct mount *mp, struct uio *uio) 383 { 384 struct fuse_notify_inval_entry_out fnieo; 385 struct componentname cn; 386 struct vnode *dvp, *vp; 387 char name[PATH_MAX]; 388 int err; 389 390 if ((err = uiomove(&fnieo, sizeof(fnieo), uio)) != 0) 391 return (err); 392 393 if ((err = uiomove(name, fnieo.namelen, uio)) != 0) 394 return (err); 395 name[fnieo.namelen] = '\0'; 396 /* fusefs does not cache "." or ".." entries */ 397 if (strncmp(name, ".", sizeof(".")) == 0 || 398 strncmp(name, "..", sizeof("..")) == 0) 399 return (0); 400 401 if (fnieo.parent == FUSE_ROOT_ID) 402 err = VFS_ROOT(mp, LK_SHARED, &dvp); 403 else 404 err = fuse_internal_get_cached_vnode( mp, fnieo.parent, 405 LK_SHARED, &dvp); 406 /* 407 * If dvp is not in the cache, then it must've been reclaimed. And 408 * since fuse_vnop_reclaim does a cache_purge, name's entry must've 409 * been invalidated already. So we can safely return if dvp == NULL 410 */ 411 if (err != 0 || dvp == NULL) 412 return (err); 413 /* 414 * XXX we can't check dvp's generation because the FUSE invalidate 415 * entry message doesn't include it. Worse case is that we invalidate 416 * an entry that didn't need to be invalidated. 417 */ 418 419 cn.cn_nameiop = LOOKUP; 420 cn.cn_flags = 0; /* !MAKEENTRY means free cached entry */ 421 cn.cn_thread = curthread; 422 cn.cn_cred = curthread->td_ucred; 423 cn.cn_lkflags = LK_SHARED; 424 cn.cn_pnbuf = NULL; 425 cn.cn_nameptr = name; 426 cn.cn_namelen = fnieo.namelen; 427 err = cache_lookup(dvp, &vp, &cn, NULL, NULL); 428 MPASS(err == 0); 429 fuse_vnode_clear_attr_cache(dvp); 430 vput(dvp); 431 return (0); 432 } 433 434 int 435 fuse_internal_invalidate_inode(struct mount *mp, struct uio *uio) 436 { 437 struct fuse_notify_inval_inode_out fniio; 438 struct vnode *vp; 439 int err; 440 441 if ((err = uiomove(&fniio, sizeof(fniio), uio)) != 0) 442 return (err); 443 444 if (fniio.ino == FUSE_ROOT_ID) 445 err = VFS_ROOT(mp, LK_EXCLUSIVE, &vp); 446 else 447 err = fuse_internal_get_cached_vnode(mp, fniio.ino, LK_SHARED, 448 &vp); 449 if (err != 0 || vp == NULL) 450 return (err); 451 /* 452 * XXX we can't check vp's generation because the FUSE invalidate 453 * entry message doesn't include it. Worse case is that we invalidate 454 * an inode that didn't need to be invalidated. 455 */ 456 457 /* 458 * Flush and invalidate buffers if off >= 0. Technically we only need 459 * to flush and invalidate the range of offsets [off, off + len), but 460 * for simplicity's sake we do everything. 461 */ 462 if (fniio.off >= 0) 463 fuse_io_invalbuf(vp, curthread); 464 fuse_vnode_clear_attr_cache(vp); 465 vput(vp); 466 return (0); 467 } 468 469 /* mknod */ 470 int 471 fuse_internal_mknod(struct vnode *dvp, struct vnode **vpp, 472 struct componentname *cnp, struct vattr *vap) 473 { 474 struct fuse_data *data; 475 struct fuse_mknod_in fmni; 476 size_t insize; 477 478 data = fuse_get_mpdata(dvp->v_mount); 479 480 fmni.mode = MAKEIMODE(vap->va_type, vap->va_mode); 481 fmni.rdev = vap->va_rdev; 482 if (fuse_libabi_geq(data, 7, 12)) { 483 insize = sizeof(fmni); 484 fmni.umask = curthread->td_proc->p_fd->fd_cmask; 485 } else { 486 insize = FUSE_COMPAT_MKNOD_IN_SIZE; 487 } 488 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKNOD, &fmni, 489 insize, vap->va_type)); 490 } 491 492 /* readdir */ 493 494 int 495 fuse_internal_readdir(struct vnode *vp, 496 struct uio *uio, 497 off_t startoff, 498 struct fuse_filehandle *fufh, 499 struct fuse_iov *cookediov, 500 int *ncookies, 501 u_long *cookies) 502 { 503 int err = 0; 504 struct fuse_dispatcher fdi; 505 struct fuse_read_in *fri = NULL; 506 int fnd_start; 507 508 if (uio_resid(uio) == 0) 509 return 0; 510 fdisp_init(&fdi, 0); 511 512 /* 513 * Note that we DO NOT have a UIO_SYSSPACE here (so no need for p2p 514 * I/O). 515 */ 516 517 /* 518 * fnd_start is set non-zero once the offset in the directory gets 519 * to the startoff. This is done because directories must be read 520 * from the beginning (offset == 0) when fuse_vnop_readdir() needs 521 * to do an open of the directory. 522 * If it is not set non-zero here, it will be set non-zero in 523 * fuse_internal_readdir_processdata() when uio_offset == startoff. 524 */ 525 fnd_start = 0; 526 if (uio->uio_offset == startoff) 527 fnd_start = 1; 528 while (uio_resid(uio) > 0) { 529 fdi.iosize = sizeof(*fri); 530 if (fri == NULL) 531 fdisp_make_vp(&fdi, FUSE_READDIR, vp, NULL, NULL); 532 else 533 fdisp_refresh_vp(&fdi, FUSE_READDIR, vp, NULL, NULL); 534 535 fri = fdi.indata; 536 fri->fh = fufh->fh_id; 537 fri->offset = uio_offset(uio); 538 fri->size = MIN(uio->uio_resid, 539 fuse_get_mpdata(vp->v_mount)->max_read); 540 541 if ((err = fdisp_wait_answ(&fdi))) 542 break; 543 if ((err = fuse_internal_readdir_processdata(uio, startoff, 544 &fnd_start, fri->size, fdi.answ, fdi.iosize, cookediov, 545 ncookies, &cookies))) 546 break; 547 } 548 549 fdisp_destroy(&fdi); 550 return ((err == -1) ? 0 : err); 551 } 552 553 /* 554 * Return -1 to indicate that this readdir is finished, 0 if it copied 555 * all the directory data read in and it may be possible to read more 556 * and greater than 0 for a failure. 557 */ 558 int 559 fuse_internal_readdir_processdata(struct uio *uio, 560 off_t startoff, 561 int *fnd_start, 562 size_t reqsize, 563 void *buf, 564 size_t bufsize, 565 struct fuse_iov *cookediov, 566 int *ncookies, 567 u_long **cookiesp) 568 { 569 int err = 0; 570 int bytesavail; 571 size_t freclen; 572 573 struct dirent *de; 574 struct fuse_dirent *fudge; 575 u_long *cookies; 576 577 cookies = *cookiesp; 578 if (bufsize < FUSE_NAME_OFFSET) 579 return -1; 580 for (;;) { 581 if (bufsize < FUSE_NAME_OFFSET) { 582 err = -1; 583 break; 584 } 585 fudge = (struct fuse_dirent *)buf; 586 freclen = FUSE_DIRENT_SIZE(fudge); 587 588 if (bufsize < freclen) { 589 /* 590 * This indicates a partial directory entry at the 591 * end of the directory data. 592 */ 593 err = -1; 594 break; 595 } 596 #ifdef ZERO_PAD_INCOMPLETE_BUFS 597 if (isbzero(buf, FUSE_NAME_OFFSET)) { 598 err = -1; 599 break; 600 } 601 #endif 602 603 if (!fudge->namelen || fudge->namelen > MAXNAMLEN) { 604 err = EINVAL; 605 break; 606 } 607 bytesavail = GENERIC_DIRSIZ((struct pseudo_dirent *) 608 &fudge->namelen); 609 610 if (bytesavail > uio_resid(uio)) { 611 /* Out of space for the dir so we are done. */ 612 err = -1; 613 break; 614 } 615 /* 616 * Don't start to copy the directory entries out until 617 * the requested offset in the directory is found. 618 */ 619 if (*fnd_start != 0) { 620 fiov_adjust(cookediov, bytesavail); 621 bzero(cookediov->base, bytesavail); 622 623 de = (struct dirent *)cookediov->base; 624 de->d_fileno = fudge->ino; 625 de->d_reclen = bytesavail; 626 de->d_type = fudge->type; 627 de->d_namlen = fudge->namelen; 628 memcpy((char *)cookediov->base + sizeof(struct dirent) - 629 MAXNAMLEN - 1, 630 (char *)buf + FUSE_NAME_OFFSET, fudge->namelen); 631 dirent_terminate(de); 632 633 err = uiomove(cookediov->base, cookediov->len, uio); 634 if (err) 635 break; 636 if (cookies != NULL) { 637 if (*ncookies == 0) { 638 err = -1; 639 break; 640 } 641 *cookies = fudge->off; 642 cookies++; 643 (*ncookies)--; 644 } 645 } else if (startoff == fudge->off) 646 *fnd_start = 1; 647 buf = (char *)buf + freclen; 648 bufsize -= freclen; 649 uio_setoffset(uio, fudge->off); 650 } 651 *cookiesp = cookies; 652 653 return err; 654 } 655 656 /* remove */ 657 658 int 659 fuse_internal_remove(struct vnode *dvp, 660 struct vnode *vp, 661 struct componentname *cnp, 662 enum fuse_opcode op) 663 { 664 struct fuse_dispatcher fdi; 665 nlink_t nlink; 666 int err = 0; 667 668 fdisp_init(&fdi, cnp->cn_namelen + 1); 669 fdisp_make_vp(&fdi, op, dvp, cnp->cn_thread, cnp->cn_cred); 670 671 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen); 672 ((char *)fdi.indata)[cnp->cn_namelen] = '\0'; 673 674 err = fdisp_wait_answ(&fdi); 675 fdisp_destroy(&fdi); 676 677 if (err) 678 return (err); 679 680 /* 681 * Access the cached nlink even if the attr cached has expired. If 682 * it's inaccurate, the worst that will happen is: 683 * 1) We'll recycle the vnode even though the file has another link we 684 * don't know about, costing a bit of cpu time, or 685 * 2) We won't recycle the vnode even though all of its links are gone. 686 * It will linger around until vnlru reclaims it, costing a bit of 687 * temporary memory. 688 */ 689 nlink = VTOFUD(vp)->cached_attrs.va_nlink--; 690 691 /* 692 * Purge the parent's attribute cache because the daemon 693 * should've updated its mtime and ctime. 694 */ 695 fuse_vnode_clear_attr_cache(dvp); 696 697 /* NB: nlink could be zero if it was never cached */ 698 if (nlink <= 1 || vnode_vtype(vp) == VDIR) { 699 fuse_internal_vnode_disappear(vp); 700 } else { 701 cache_purge(vp); 702 fuse_vnode_update(vp, FN_CTIMECHANGE); 703 } 704 705 return err; 706 } 707 708 /* rename */ 709 710 int 711 fuse_internal_rename(struct vnode *fdvp, 712 struct componentname *fcnp, 713 struct vnode *tdvp, 714 struct componentname *tcnp) 715 { 716 struct fuse_dispatcher fdi; 717 struct fuse_rename_in *fri; 718 int err = 0; 719 720 fdisp_init(&fdi, sizeof(*fri) + fcnp->cn_namelen + tcnp->cn_namelen + 2); 721 fdisp_make_vp(&fdi, FUSE_RENAME, fdvp, tcnp->cn_thread, tcnp->cn_cred); 722 723 fri = fdi.indata; 724 fri->newdir = VTOI(tdvp); 725 memcpy((char *)fdi.indata + sizeof(*fri), fcnp->cn_nameptr, 726 fcnp->cn_namelen); 727 ((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen] = '\0'; 728 memcpy((char *)fdi.indata + sizeof(*fri) + fcnp->cn_namelen + 1, 729 tcnp->cn_nameptr, tcnp->cn_namelen); 730 ((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen + 731 tcnp->cn_namelen + 1] = '\0'; 732 733 err = fdisp_wait_answ(&fdi); 734 fdisp_destroy(&fdi); 735 return err; 736 } 737 738 /* strategy */ 739 740 /* entity creation */ 741 742 void 743 fuse_internal_newentry_makerequest(struct mount *mp, 744 uint64_t dnid, 745 struct componentname *cnp, 746 enum fuse_opcode op, 747 void *buf, 748 size_t bufsize, 749 struct fuse_dispatcher *fdip) 750 { 751 fdip->iosize = bufsize + cnp->cn_namelen + 1; 752 753 fdisp_make(fdip, op, mp, dnid, cnp->cn_thread, cnp->cn_cred); 754 memcpy(fdip->indata, buf, bufsize); 755 memcpy((char *)fdip->indata + bufsize, cnp->cn_nameptr, cnp->cn_namelen); 756 ((char *)fdip->indata)[bufsize + cnp->cn_namelen] = '\0'; 757 } 758 759 int 760 fuse_internal_newentry_core(struct vnode *dvp, 761 struct vnode **vpp, 762 struct componentname *cnp, 763 enum vtype vtyp, 764 struct fuse_dispatcher *fdip) 765 { 766 int err = 0; 767 struct fuse_entry_out *feo; 768 struct mount *mp = vnode_mount(dvp); 769 770 if ((err = fdisp_wait_answ(fdip))) { 771 return err; 772 } 773 feo = fdip->answ; 774 775 if ((err = fuse_internal_checkentry(feo, vtyp))) { 776 return err; 777 } 778 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vtyp); 779 if (err) { 780 fuse_internal_forget_send(mp, cnp->cn_thread, cnp->cn_cred, 781 feo->nodeid, 1); 782 return err; 783 } 784 785 /* 786 * Purge the parent's attribute cache because the daemon should've 787 * updated its mtime and ctime 788 */ 789 fuse_vnode_clear_attr_cache(dvp); 790 791 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid, 792 feo->attr_valid_nsec, NULL); 793 794 return err; 795 } 796 797 int 798 fuse_internal_newentry(struct vnode *dvp, 799 struct vnode **vpp, 800 struct componentname *cnp, 801 enum fuse_opcode op, 802 void *buf, 803 size_t bufsize, 804 enum vtype vtype) 805 { 806 int err; 807 struct fuse_dispatcher fdi; 808 struct mount *mp = vnode_mount(dvp); 809 810 fdisp_init(&fdi, 0); 811 fuse_internal_newentry_makerequest(mp, VTOI(dvp), cnp, op, buf, 812 bufsize, &fdi); 813 err = fuse_internal_newentry_core(dvp, vpp, cnp, vtype, &fdi); 814 fdisp_destroy(&fdi); 815 816 return err; 817 } 818 819 /* entity destruction */ 820 821 int 822 fuse_internal_forget_callback(struct fuse_ticket *ftick, struct uio *uio) 823 { 824 fuse_internal_forget_send(ftick->tk_data->mp, curthread, NULL, 825 ((struct fuse_in_header *)ftick->tk_ms_fiov.base)->nodeid, 1); 826 827 return 0; 828 } 829 830 void 831 fuse_internal_forget_send(struct mount *mp, 832 struct thread *td, 833 struct ucred *cred, 834 uint64_t nodeid, 835 uint64_t nlookup) 836 { 837 838 struct fuse_dispatcher fdi; 839 struct fuse_forget_in *ffi; 840 841 /* 842 * KASSERT(nlookup > 0, ("zero-times forget for vp #%llu", 843 * (long long unsigned) nodeid)); 844 */ 845 846 fdisp_init(&fdi, sizeof(*ffi)); 847 fdisp_make(&fdi, FUSE_FORGET, mp, nodeid, td, cred); 848 849 ffi = fdi.indata; 850 ffi->nlookup = nlookup; 851 852 fuse_insert_message(fdi.tick, false); 853 fdisp_destroy(&fdi); 854 } 855 856 /* Fetch the vnode's attributes from the daemon*/ 857 int 858 fuse_internal_do_getattr(struct vnode *vp, struct vattr *vap, 859 struct ucred *cred, struct thread *td) 860 { 861 struct fuse_dispatcher fdi; 862 struct fuse_vnode_data *fvdat = VTOFUD(vp); 863 struct fuse_getattr_in *fgai; 864 struct fuse_attr_out *fao; 865 off_t old_filesize = fvdat->cached_attrs.va_size; 866 struct timespec old_ctime = fvdat->cached_attrs.va_ctime; 867 struct timespec old_mtime = fvdat->cached_attrs.va_mtime; 868 enum vtype vtyp; 869 int err; 870 871 fdisp_init(&fdi, sizeof(*fgai)); 872 fdisp_make_vp(&fdi, FUSE_GETATTR, vp, td, cred); 873 fgai = fdi.indata; 874 /* 875 * We could look up a file handle and set it in fgai->fh, but that 876 * involves extra runtime work and I'm unaware of any file systems that 877 * care. 878 */ 879 fgai->getattr_flags = 0; 880 if ((err = fdisp_wait_answ(&fdi))) { 881 if (err == ENOENT) 882 fuse_internal_vnode_disappear(vp); 883 goto out; 884 } 885 886 fao = (struct fuse_attr_out *)fdi.answ; 887 vtyp = IFTOVT(fao->attr.mode); 888 if (fvdat->flag & FN_SIZECHANGE) 889 fao->attr.size = old_filesize; 890 if (fvdat->flag & FN_CTIMECHANGE) { 891 fao->attr.ctime = old_ctime.tv_sec; 892 fao->attr.ctimensec = old_ctime.tv_nsec; 893 } 894 if (fvdat->flag & FN_MTIMECHANGE) { 895 fao->attr.mtime = old_mtime.tv_sec; 896 fao->attr.mtimensec = old_mtime.tv_nsec; 897 } 898 fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid, 899 fao->attr_valid_nsec, vap); 900 if (vtyp != vnode_vtype(vp)) { 901 fuse_internal_vnode_disappear(vp); 902 err = ENOENT; 903 } 904 905 out: 906 fdisp_destroy(&fdi); 907 return err; 908 } 909 910 /* Read a vnode's attributes from cache or fetch them from the fuse daemon */ 911 int 912 fuse_internal_getattr(struct vnode *vp, struct vattr *vap, struct ucred *cred, 913 struct thread *td) 914 { 915 struct vattr *attrs; 916 917 if ((attrs = VTOVA(vp)) != NULL) { 918 *vap = *attrs; /* struct copy */ 919 return 0; 920 } 921 922 return fuse_internal_do_getattr(vp, vap, cred, td); 923 } 924 925 void 926 fuse_internal_vnode_disappear(struct vnode *vp) 927 { 928 struct fuse_vnode_data *fvdat = VTOFUD(vp); 929 930 ASSERT_VOP_ELOCKED(vp, "fuse_internal_vnode_disappear"); 931 fvdat->flag |= FN_REVOKED; 932 cache_purge(vp); 933 } 934 935 /* fuse start/stop */ 936 937 int 938 fuse_internal_init_callback(struct fuse_ticket *tick, struct uio *uio) 939 { 940 int err = 0; 941 struct fuse_data *data = tick->tk_data; 942 struct fuse_init_out *fiio; 943 944 if ((err = tick->tk_aw_ohead.error)) { 945 goto out; 946 } 947 if ((err = fticket_pull(tick, uio))) { 948 goto out; 949 } 950 fiio = fticket_resp(tick)->base; 951 952 data->fuse_libabi_major = fiio->major; 953 data->fuse_libabi_minor = fiio->minor; 954 if (!fuse_libabi_geq(data, 7, 4)) { 955 /* 956 * With a little work we could support servers as old as 7.1. 957 * But there would be little payoff. 958 */ 959 SDT_PROBE2(fusefs, , internal, trace, 1, 960 "userpace version too low"); 961 err = EPROTONOSUPPORT; 962 goto out; 963 } 964 965 if (fuse_libabi_geq(data, 7, 5)) { 966 if (fticket_resp(tick)->len == sizeof(struct fuse_init_out) || 967 fticket_resp(tick)->len == FUSE_COMPAT_22_INIT_OUT_SIZE) { 968 data->max_write = fiio->max_write; 969 if (fiio->flags & FUSE_ASYNC_READ) 970 data->dataflags |= FSESS_ASYNC_READ; 971 if (fiio->flags & FUSE_POSIX_LOCKS) 972 data->dataflags |= FSESS_POSIX_LOCKS; 973 if (fiio->flags & FUSE_EXPORT_SUPPORT) 974 data->dataflags |= FSESS_EXPORT_SUPPORT; 975 /* 976 * Don't bother to check FUSE_BIG_WRITES, because it's 977 * redundant with max_write 978 */ 979 /* 980 * max_background and congestion_threshold are not 981 * implemented 982 */ 983 } else { 984 err = EINVAL; 985 } 986 } else { 987 /* Old fixed values */ 988 data->max_write = 4096; 989 } 990 991 if (fuse_libabi_geq(data, 7, 6)) 992 data->max_readahead_blocks = fiio->max_readahead / maxbcachebuf; 993 994 if (!fuse_libabi_geq(data, 7, 7)) 995 fsess_set_notimpl(data->mp, FUSE_INTERRUPT); 996 997 if (!fuse_libabi_geq(data, 7, 8)) { 998 fsess_set_notimpl(data->mp, FUSE_BMAP); 999 fsess_set_notimpl(data->mp, FUSE_DESTROY); 1000 } 1001 1002 if (fuse_libabi_geq(data, 7, 23) && fiio->time_gran >= 1 && 1003 fiio->time_gran <= 1000000000) 1004 data->time_gran = fiio->time_gran; 1005 else 1006 data->time_gran = 1; 1007 1008 if (!fuse_libabi_geq(data, 7, 23)) 1009 data->cache_mode = fuse_data_cache_mode; 1010 else if (fiio->flags & FUSE_WRITEBACK_CACHE) 1011 data->cache_mode = FUSE_CACHE_WB; 1012 else 1013 data->cache_mode = FUSE_CACHE_WT; 1014 1015 out: 1016 if (err) { 1017 fdata_set_dead(data); 1018 } 1019 FUSE_LOCK(); 1020 data->dataflags |= FSESS_INITED; 1021 wakeup(&data->ticketer); 1022 FUSE_UNLOCK(); 1023 1024 return 0; 1025 } 1026 1027 void 1028 fuse_internal_send_init(struct fuse_data *data, struct thread *td) 1029 { 1030 struct fuse_init_in *fiii; 1031 struct fuse_dispatcher fdi; 1032 1033 fdisp_init(&fdi, sizeof(*fiii)); 1034 fdisp_make(&fdi, FUSE_INIT, data->mp, 0, td, NULL); 1035 fiii = fdi.indata; 1036 fiii->major = FUSE_KERNEL_VERSION; 1037 fiii->minor = FUSE_KERNEL_MINOR_VERSION; 1038 /* 1039 * fusefs currently reads ahead no more than one cache block at a time. 1040 * See fuse_read_biobackend 1041 */ 1042 fiii->max_readahead = maxbcachebuf; 1043 /* 1044 * Unsupported features: 1045 * FUSE_FILE_OPS: No known FUSE server or client supports it 1046 * FUSE_ATOMIC_O_TRUNC: our VFS cannot support it 1047 * FUSE_DONT_MASK: unlike Linux, FreeBSD always applies the umask, even 1048 * when default ACLs are in use. 1049 * FUSE_SPLICE_WRITE, FUSE_SPLICE_MOVE, FUSE_SPLICE_READ: FreeBSD 1050 * doesn't have splice(2). 1051 * FUSE_FLOCK_LOCKS: not yet implemented 1052 * FUSE_HAS_IOCTL_DIR: not yet implemented 1053 * FUSE_AUTO_INVAL_DATA: not yet implemented 1054 * FUSE_DO_READDIRPLUS: not yet implemented 1055 * FUSE_READDIRPLUS_AUTO: not yet implemented 1056 * FUSE_ASYNC_DIO: not yet implemented 1057 * FUSE_NO_OPEN_SUPPORT: not yet implemented 1058 */ 1059 fiii->flags = FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_EXPORT_SUPPORT 1060 | FUSE_BIG_WRITES | FUSE_WRITEBACK_CACHE; 1061 1062 fuse_insert_callback(fdi.tick, fuse_internal_init_callback); 1063 fuse_insert_message(fdi.tick, false); 1064 fdisp_destroy(&fdi); 1065 } 1066 1067 /* 1068 * Send a FUSE_SETATTR operation with no permissions checks. If cred is NULL, 1069 * send the request with root credentials 1070 */ 1071 int fuse_internal_setattr(struct vnode *vp, struct vattr *vap, 1072 struct thread *td, struct ucred *cred) 1073 { 1074 struct fuse_vnode_data *fvdat; 1075 struct fuse_dispatcher fdi; 1076 struct fuse_setattr_in *fsai; 1077 struct mount *mp; 1078 pid_t pid = td->td_proc->p_pid; 1079 struct fuse_data *data; 1080 int dataflags; 1081 int err = 0; 1082 enum vtype vtyp; 1083 int sizechanged = -1; 1084 uint64_t newsize = 0; 1085 1086 mp = vnode_mount(vp); 1087 fvdat = VTOFUD(vp); 1088 data = fuse_get_mpdata(mp); 1089 dataflags = data->dataflags; 1090 1091 fdisp_init(&fdi, sizeof(*fsai)); 1092 fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred); 1093 if (!cred) { 1094 fdi.finh->uid = 0; 1095 fdi.finh->gid = 0; 1096 } 1097 fsai = fdi.indata; 1098 fsai->valid = 0; 1099 1100 if (vap->va_uid != (uid_t)VNOVAL) { 1101 fsai->uid = vap->va_uid; 1102 fsai->valid |= FATTR_UID; 1103 } 1104 if (vap->va_gid != (gid_t)VNOVAL) { 1105 fsai->gid = vap->va_gid; 1106 fsai->valid |= FATTR_GID; 1107 } 1108 if (vap->va_size != VNOVAL) { 1109 struct fuse_filehandle *fufh = NULL; 1110 1111 /*Truncate to a new value. */ 1112 fsai->size = vap->va_size; 1113 sizechanged = 1; 1114 newsize = vap->va_size; 1115 fsai->valid |= FATTR_SIZE; 1116 1117 fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); 1118 if (fufh) { 1119 fsai->fh = fufh->fh_id; 1120 fsai->valid |= FATTR_FH; 1121 } 1122 VTOFUD(vp)->flag &= ~FN_SIZECHANGE; 1123 } 1124 if (vap->va_atime.tv_sec != VNOVAL) { 1125 fsai->atime = vap->va_atime.tv_sec; 1126 fsai->atimensec = vap->va_atime.tv_nsec; 1127 fsai->valid |= FATTR_ATIME; 1128 if (vap->va_vaflags & VA_UTIMES_NULL) 1129 fsai->valid |= FATTR_ATIME_NOW; 1130 } 1131 if (vap->va_mtime.tv_sec != VNOVAL) { 1132 fsai->mtime = vap->va_mtime.tv_sec; 1133 fsai->mtimensec = vap->va_mtime.tv_nsec; 1134 fsai->valid |= FATTR_MTIME; 1135 if (vap->va_vaflags & VA_UTIMES_NULL) 1136 fsai->valid |= FATTR_MTIME_NOW; 1137 } else if (fvdat->flag & FN_MTIMECHANGE) { 1138 fsai->mtime = fvdat->cached_attrs.va_mtime.tv_sec; 1139 fsai->mtimensec = fvdat->cached_attrs.va_mtime.tv_nsec; 1140 fsai->valid |= FATTR_MTIME; 1141 } 1142 if (fuse_libabi_geq(data, 7, 23) && fvdat->flag & FN_CTIMECHANGE) { 1143 fsai->ctime = fvdat->cached_attrs.va_ctime.tv_sec; 1144 fsai->ctimensec = fvdat->cached_attrs.va_ctime.tv_nsec; 1145 fsai->valid |= FATTR_CTIME; 1146 } 1147 if (vap->va_mode != (mode_t)VNOVAL) { 1148 fsai->mode = vap->va_mode & ALLPERMS; 1149 fsai->valid |= FATTR_MODE; 1150 } 1151 if (!fsai->valid) { 1152 goto out; 1153 } 1154 1155 if ((err = fdisp_wait_answ(&fdi))) 1156 goto out; 1157 vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode); 1158 1159 if (vnode_vtype(vp) != vtyp) { 1160 if (vnode_vtype(vp) == VNON && vtyp != VNON) { 1161 SDT_PROBE2(fusefs, , internal, trace, 1, "FUSE: Dang! " 1162 "vnode_vtype is VNON and vtype isn't."); 1163 } else { 1164 /* 1165 * STALE vnode, ditch 1166 * 1167 * The vnode has changed its type "behind our back". 1168 * There's nothing really we can do, so let us just 1169 * force an internal revocation and tell the caller to 1170 * try again, if interested. 1171 */ 1172 fuse_internal_vnode_disappear(vp); 1173 err = EAGAIN; 1174 } 1175 } 1176 if (err == 0) { 1177 struct fuse_attr_out *fao = (struct fuse_attr_out*)fdi.answ; 1178 fuse_vnode_undirty_cached_timestamps(vp); 1179 fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid, 1180 fao->attr_valid_nsec, NULL); 1181 } 1182 1183 out: 1184 fdisp_destroy(&fdi); 1185 return err; 1186 } 1187 1188 #ifdef ZERO_PAD_INCOMPLETE_BUFS 1189 static int 1190 isbzero(void *buf, size_t len) 1191 { 1192 int i; 1193 1194 for (i = 0; i < len; i++) { 1195 if (((char *)buf)[i]) 1196 return (0); 1197 } 1198 1199 return (1); 1200 } 1201 1202 #endif 1203 1204 void 1205 fuse_internal_init(void) 1206 { 1207 fuse_lookup_cache_misses = counter_u64_alloc(M_WAITOK); 1208 fuse_lookup_cache_hits = counter_u64_alloc(M_WAITOK); 1209 } 1210 1211 void 1212 fuse_internal_destroy(void) 1213 { 1214 counter_u64_free(fuse_lookup_cache_hits); 1215 counter_u64_free(fuse_lookup_cache_misses); 1216 } 1217