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 (fnieo.namelen > sizeof(name)) 394 return (EINVAL); 395 396 if ((err = uiomove(name, fnieo.namelen, uio)) != 0) 397 return (err); 398 name[fnieo.namelen] = '\0'; 399 /* fusefs does not cache "." or ".." entries */ 400 if (strncmp(name, ".", sizeof(".")) == 0 || 401 strncmp(name, "..", sizeof("..")) == 0) 402 return (0); 403 404 if (fnieo.parent == FUSE_ROOT_ID) 405 err = VFS_ROOT(mp, LK_SHARED, &dvp); 406 else 407 err = fuse_internal_get_cached_vnode( mp, fnieo.parent, 408 LK_SHARED, &dvp); 409 /* 410 * If dvp is not in the cache, then it must've been reclaimed. And 411 * since fuse_vnop_reclaim does a cache_purge, name's entry must've 412 * been invalidated already. So we can safely return if dvp == NULL 413 */ 414 if (err != 0 || dvp == NULL) 415 return (err); 416 /* 417 * XXX we can't check dvp's generation because the FUSE invalidate 418 * entry message doesn't include it. Worse case is that we invalidate 419 * an entry that didn't need to be invalidated. 420 */ 421 422 cn.cn_nameiop = LOOKUP; 423 cn.cn_flags = 0; /* !MAKEENTRY means free cached entry */ 424 cn.cn_thread = curthread; 425 cn.cn_cred = curthread->td_ucred; 426 cn.cn_lkflags = LK_SHARED; 427 cn.cn_pnbuf = NULL; 428 cn.cn_nameptr = name; 429 cn.cn_namelen = fnieo.namelen; 430 err = cache_lookup(dvp, &vp, &cn, NULL, NULL); 431 MPASS(err == 0); 432 fuse_vnode_clear_attr_cache(dvp); 433 vput(dvp); 434 return (0); 435 } 436 437 int 438 fuse_internal_invalidate_inode(struct mount *mp, struct uio *uio) 439 { 440 struct fuse_notify_inval_inode_out fniio; 441 struct vnode *vp; 442 int err; 443 444 if ((err = uiomove(&fniio, sizeof(fniio), uio)) != 0) 445 return (err); 446 447 if (fniio.ino == FUSE_ROOT_ID) 448 err = VFS_ROOT(mp, LK_EXCLUSIVE, &vp); 449 else 450 err = fuse_internal_get_cached_vnode(mp, fniio.ino, LK_SHARED, 451 &vp); 452 if (err != 0 || vp == NULL) 453 return (err); 454 /* 455 * XXX we can't check vp's generation because the FUSE invalidate 456 * entry message doesn't include it. Worse case is that we invalidate 457 * an inode that didn't need to be invalidated. 458 */ 459 460 /* 461 * Flush and invalidate buffers if off >= 0. Technically we only need 462 * to flush and invalidate the range of offsets [off, off + len), but 463 * for simplicity's sake we do everything. 464 */ 465 if (fniio.off >= 0) 466 fuse_io_invalbuf(vp, curthread); 467 fuse_vnode_clear_attr_cache(vp); 468 vput(vp); 469 return (0); 470 } 471 472 /* mknod */ 473 int 474 fuse_internal_mknod(struct vnode *dvp, struct vnode **vpp, 475 struct componentname *cnp, struct vattr *vap) 476 { 477 struct fuse_data *data; 478 struct fuse_mknod_in fmni; 479 size_t insize; 480 481 data = fuse_get_mpdata(dvp->v_mount); 482 483 fmni.mode = MAKEIMODE(vap->va_type, vap->va_mode); 484 fmni.rdev = vap->va_rdev; 485 if (fuse_libabi_geq(data, 7, 12)) { 486 insize = sizeof(fmni); 487 fmni.umask = curthread->td_proc->p_fd->fd_cmask; 488 } else { 489 insize = FUSE_COMPAT_MKNOD_IN_SIZE; 490 } 491 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKNOD, &fmni, 492 insize, vap->va_type)); 493 } 494 495 /* readdir */ 496 497 int 498 fuse_internal_readdir(struct vnode *vp, 499 struct uio *uio, 500 off_t startoff, 501 struct fuse_filehandle *fufh, 502 struct fuse_iov *cookediov, 503 int *ncookies, 504 u_long *cookies) 505 { 506 int err = 0; 507 struct fuse_dispatcher fdi; 508 struct fuse_read_in *fri = NULL; 509 int fnd_start; 510 511 if (uio_resid(uio) == 0) 512 return 0; 513 fdisp_init(&fdi, 0); 514 515 /* 516 * Note that we DO NOT have a UIO_SYSSPACE here (so no need for p2p 517 * I/O). 518 */ 519 520 /* 521 * fnd_start is set non-zero once the offset in the directory gets 522 * to the startoff. This is done because directories must be read 523 * from the beginning (offset == 0) when fuse_vnop_readdir() needs 524 * to do an open of the directory. 525 * If it is not set non-zero here, it will be set non-zero in 526 * fuse_internal_readdir_processdata() when uio_offset == startoff. 527 */ 528 fnd_start = 0; 529 if (uio->uio_offset == startoff) 530 fnd_start = 1; 531 while (uio_resid(uio) > 0) { 532 fdi.iosize = sizeof(*fri); 533 if (fri == NULL) 534 fdisp_make_vp(&fdi, FUSE_READDIR, vp, NULL, NULL); 535 else 536 fdisp_refresh_vp(&fdi, FUSE_READDIR, vp, NULL, NULL); 537 538 fri = fdi.indata; 539 fri->fh = fufh->fh_id; 540 fri->offset = uio_offset(uio); 541 fri->size = MIN(uio->uio_resid, 542 fuse_get_mpdata(vp->v_mount)->max_read); 543 544 if ((err = fdisp_wait_answ(&fdi))) 545 break; 546 if ((err = fuse_internal_readdir_processdata(uio, startoff, 547 &fnd_start, fri->size, fdi.answ, fdi.iosize, cookediov, 548 ncookies, &cookies))) 549 break; 550 } 551 552 fdisp_destroy(&fdi); 553 return ((err == -1) ? 0 : err); 554 } 555 556 /* 557 * Return -1 to indicate that this readdir is finished, 0 if it copied 558 * all the directory data read in and it may be possible to read more 559 * and greater than 0 for a failure. 560 */ 561 int 562 fuse_internal_readdir_processdata(struct uio *uio, 563 off_t startoff, 564 int *fnd_start, 565 size_t reqsize, 566 void *buf, 567 size_t bufsize, 568 struct fuse_iov *cookediov, 569 int *ncookies, 570 u_long **cookiesp) 571 { 572 int err = 0; 573 int bytesavail; 574 size_t freclen; 575 576 struct dirent *de; 577 struct fuse_dirent *fudge; 578 u_long *cookies; 579 580 cookies = *cookiesp; 581 if (bufsize < FUSE_NAME_OFFSET) 582 return -1; 583 for (;;) { 584 if (bufsize < FUSE_NAME_OFFSET) { 585 err = -1; 586 break; 587 } 588 fudge = (struct fuse_dirent *)buf; 589 freclen = FUSE_DIRENT_SIZE(fudge); 590 591 if (bufsize < freclen) { 592 /* 593 * This indicates a partial directory entry at the 594 * end of the directory data. 595 */ 596 err = -1; 597 break; 598 } 599 #ifdef ZERO_PAD_INCOMPLETE_BUFS 600 if (isbzero(buf, FUSE_NAME_OFFSET)) { 601 err = -1; 602 break; 603 } 604 #endif 605 606 if (!fudge->namelen || fudge->namelen > MAXNAMLEN) { 607 err = EINVAL; 608 break; 609 } 610 bytesavail = GENERIC_DIRSIZ((struct pseudo_dirent *) 611 &fudge->namelen); 612 613 if (bytesavail > uio_resid(uio)) { 614 /* Out of space for the dir so we are done. */ 615 err = -1; 616 break; 617 } 618 /* 619 * Don't start to copy the directory entries out until 620 * the requested offset in the directory is found. 621 */ 622 if (*fnd_start != 0) { 623 fiov_adjust(cookediov, bytesavail); 624 bzero(cookediov->base, bytesavail); 625 626 de = (struct dirent *)cookediov->base; 627 de->d_fileno = fudge->ino; 628 de->d_reclen = bytesavail; 629 de->d_type = fudge->type; 630 de->d_namlen = fudge->namelen; 631 memcpy((char *)cookediov->base + sizeof(struct dirent) - 632 MAXNAMLEN - 1, 633 (char *)buf + FUSE_NAME_OFFSET, fudge->namelen); 634 dirent_terminate(de); 635 636 err = uiomove(cookediov->base, cookediov->len, uio); 637 if (err) 638 break; 639 if (cookies != NULL) { 640 if (*ncookies == 0) { 641 err = -1; 642 break; 643 } 644 *cookies = fudge->off; 645 cookies++; 646 (*ncookies)--; 647 } 648 } else if (startoff == fudge->off) 649 *fnd_start = 1; 650 buf = (char *)buf + freclen; 651 bufsize -= freclen; 652 uio_setoffset(uio, fudge->off); 653 } 654 *cookiesp = cookies; 655 656 return err; 657 } 658 659 /* remove */ 660 661 int 662 fuse_internal_remove(struct vnode *dvp, 663 struct vnode *vp, 664 struct componentname *cnp, 665 enum fuse_opcode op) 666 { 667 struct fuse_dispatcher fdi; 668 nlink_t nlink; 669 int err = 0; 670 671 fdisp_init(&fdi, cnp->cn_namelen + 1); 672 fdisp_make_vp(&fdi, op, dvp, cnp->cn_thread, cnp->cn_cred); 673 674 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen); 675 ((char *)fdi.indata)[cnp->cn_namelen] = '\0'; 676 677 err = fdisp_wait_answ(&fdi); 678 fdisp_destroy(&fdi); 679 680 if (err) 681 return (err); 682 683 /* 684 * Access the cached nlink even if the attr cached has expired. If 685 * it's inaccurate, the worst that will happen is: 686 * 1) We'll recycle the vnode even though the file has another link we 687 * don't know about, costing a bit of cpu time, or 688 * 2) We won't recycle the vnode even though all of its links are gone. 689 * It will linger around until vnlru reclaims it, costing a bit of 690 * temporary memory. 691 */ 692 nlink = VTOFUD(vp)->cached_attrs.va_nlink--; 693 694 /* 695 * Purge the parent's attribute cache because the daemon 696 * should've updated its mtime and ctime. 697 */ 698 fuse_vnode_clear_attr_cache(dvp); 699 700 /* NB: nlink could be zero if it was never cached */ 701 if (nlink <= 1 || vnode_vtype(vp) == VDIR) { 702 fuse_internal_vnode_disappear(vp); 703 } else { 704 cache_purge(vp); 705 fuse_vnode_update(vp, FN_CTIMECHANGE); 706 } 707 708 return err; 709 } 710 711 /* rename */ 712 713 int 714 fuse_internal_rename(struct vnode *fdvp, 715 struct componentname *fcnp, 716 struct vnode *tdvp, 717 struct componentname *tcnp) 718 { 719 struct fuse_dispatcher fdi; 720 struct fuse_rename_in *fri; 721 int err = 0; 722 723 fdisp_init(&fdi, sizeof(*fri) + fcnp->cn_namelen + tcnp->cn_namelen + 2); 724 fdisp_make_vp(&fdi, FUSE_RENAME, fdvp, tcnp->cn_thread, tcnp->cn_cred); 725 726 fri = fdi.indata; 727 fri->newdir = VTOI(tdvp); 728 memcpy((char *)fdi.indata + sizeof(*fri), fcnp->cn_nameptr, 729 fcnp->cn_namelen); 730 ((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen] = '\0'; 731 memcpy((char *)fdi.indata + sizeof(*fri) + fcnp->cn_namelen + 1, 732 tcnp->cn_nameptr, tcnp->cn_namelen); 733 ((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen + 734 tcnp->cn_namelen + 1] = '\0'; 735 736 err = fdisp_wait_answ(&fdi); 737 fdisp_destroy(&fdi); 738 return err; 739 } 740 741 /* strategy */ 742 743 /* entity creation */ 744 745 void 746 fuse_internal_newentry_makerequest(struct mount *mp, 747 uint64_t dnid, 748 struct componentname *cnp, 749 enum fuse_opcode op, 750 void *buf, 751 size_t bufsize, 752 struct fuse_dispatcher *fdip) 753 { 754 fdip->iosize = bufsize + cnp->cn_namelen + 1; 755 756 fdisp_make(fdip, op, mp, dnid, cnp->cn_thread, cnp->cn_cred); 757 memcpy(fdip->indata, buf, bufsize); 758 memcpy((char *)fdip->indata + bufsize, cnp->cn_nameptr, cnp->cn_namelen); 759 ((char *)fdip->indata)[bufsize + cnp->cn_namelen] = '\0'; 760 } 761 762 int 763 fuse_internal_newentry_core(struct vnode *dvp, 764 struct vnode **vpp, 765 struct componentname *cnp, 766 enum vtype vtyp, 767 struct fuse_dispatcher *fdip) 768 { 769 int err = 0; 770 struct fuse_entry_out *feo; 771 struct mount *mp = vnode_mount(dvp); 772 773 if ((err = fdisp_wait_answ(fdip))) { 774 return err; 775 } 776 feo = fdip->answ; 777 778 if ((err = fuse_internal_checkentry(feo, vtyp))) { 779 return err; 780 } 781 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vtyp); 782 if (err) { 783 fuse_internal_forget_send(mp, cnp->cn_thread, cnp->cn_cred, 784 feo->nodeid, 1); 785 return err; 786 } 787 788 /* 789 * Purge the parent's attribute cache because the daemon should've 790 * updated its mtime and ctime 791 */ 792 fuse_vnode_clear_attr_cache(dvp); 793 794 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid, 795 feo->attr_valid_nsec, NULL); 796 797 return err; 798 } 799 800 int 801 fuse_internal_newentry(struct vnode *dvp, 802 struct vnode **vpp, 803 struct componentname *cnp, 804 enum fuse_opcode op, 805 void *buf, 806 size_t bufsize, 807 enum vtype vtype) 808 { 809 int err; 810 struct fuse_dispatcher fdi; 811 struct mount *mp = vnode_mount(dvp); 812 813 fdisp_init(&fdi, 0); 814 fuse_internal_newentry_makerequest(mp, VTOI(dvp), cnp, op, buf, 815 bufsize, &fdi); 816 err = fuse_internal_newentry_core(dvp, vpp, cnp, vtype, &fdi); 817 fdisp_destroy(&fdi); 818 819 return err; 820 } 821 822 /* entity destruction */ 823 824 int 825 fuse_internal_forget_callback(struct fuse_ticket *ftick, struct uio *uio) 826 { 827 fuse_internal_forget_send(ftick->tk_data->mp, curthread, NULL, 828 ((struct fuse_in_header *)ftick->tk_ms_fiov.base)->nodeid, 1); 829 830 return 0; 831 } 832 833 void 834 fuse_internal_forget_send(struct mount *mp, 835 struct thread *td, 836 struct ucred *cred, 837 uint64_t nodeid, 838 uint64_t nlookup) 839 { 840 841 struct fuse_dispatcher fdi; 842 struct fuse_forget_in *ffi; 843 844 /* 845 * KASSERT(nlookup > 0, ("zero-times forget for vp #%llu", 846 * (long long unsigned) nodeid)); 847 */ 848 849 fdisp_init(&fdi, sizeof(*ffi)); 850 fdisp_make(&fdi, FUSE_FORGET, mp, nodeid, td, cred); 851 852 ffi = fdi.indata; 853 ffi->nlookup = nlookup; 854 855 fuse_insert_message(fdi.tick, false); 856 fdisp_destroy(&fdi); 857 } 858 859 /* Fetch the vnode's attributes from the daemon*/ 860 int 861 fuse_internal_do_getattr(struct vnode *vp, struct vattr *vap, 862 struct ucred *cred, struct thread *td) 863 { 864 struct fuse_dispatcher fdi; 865 struct fuse_vnode_data *fvdat = VTOFUD(vp); 866 struct fuse_getattr_in *fgai; 867 struct fuse_attr_out *fao; 868 off_t old_filesize = fvdat->cached_attrs.va_size; 869 struct timespec old_ctime = fvdat->cached_attrs.va_ctime; 870 struct timespec old_mtime = fvdat->cached_attrs.va_mtime; 871 enum vtype vtyp; 872 int err; 873 874 fdisp_init(&fdi, sizeof(*fgai)); 875 fdisp_make_vp(&fdi, FUSE_GETATTR, vp, td, cred); 876 fgai = fdi.indata; 877 /* 878 * We could look up a file handle and set it in fgai->fh, but that 879 * involves extra runtime work and I'm unaware of any file systems that 880 * care. 881 */ 882 fgai->getattr_flags = 0; 883 if ((err = fdisp_wait_answ(&fdi))) { 884 if (err == ENOENT) 885 fuse_internal_vnode_disappear(vp); 886 goto out; 887 } 888 889 fao = (struct fuse_attr_out *)fdi.answ; 890 vtyp = IFTOVT(fao->attr.mode); 891 if (fvdat->flag & FN_SIZECHANGE) 892 fao->attr.size = old_filesize; 893 if (fvdat->flag & FN_CTIMECHANGE) { 894 fao->attr.ctime = old_ctime.tv_sec; 895 fao->attr.ctimensec = old_ctime.tv_nsec; 896 } 897 if (fvdat->flag & FN_MTIMECHANGE) { 898 fao->attr.mtime = old_mtime.tv_sec; 899 fao->attr.mtimensec = old_mtime.tv_nsec; 900 } 901 fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid, 902 fao->attr_valid_nsec, vap); 903 if (vtyp != vnode_vtype(vp)) { 904 fuse_internal_vnode_disappear(vp); 905 err = ENOENT; 906 } 907 908 out: 909 fdisp_destroy(&fdi); 910 return err; 911 } 912 913 /* Read a vnode's attributes from cache or fetch them from the fuse daemon */ 914 int 915 fuse_internal_getattr(struct vnode *vp, struct vattr *vap, struct ucred *cred, 916 struct thread *td) 917 { 918 struct vattr *attrs; 919 920 if ((attrs = VTOVA(vp)) != NULL) { 921 *vap = *attrs; /* struct copy */ 922 return 0; 923 } 924 925 return fuse_internal_do_getattr(vp, vap, cred, td); 926 } 927 928 void 929 fuse_internal_vnode_disappear(struct vnode *vp) 930 { 931 struct fuse_vnode_data *fvdat = VTOFUD(vp); 932 933 ASSERT_VOP_ELOCKED(vp, "fuse_internal_vnode_disappear"); 934 fvdat->flag |= FN_REVOKED; 935 cache_purge(vp); 936 } 937 938 /* fuse start/stop */ 939 940 int 941 fuse_internal_init_callback(struct fuse_ticket *tick, struct uio *uio) 942 { 943 int err = 0; 944 struct fuse_data *data = tick->tk_data; 945 struct fuse_init_out *fiio; 946 947 if ((err = tick->tk_aw_ohead.error)) { 948 goto out; 949 } 950 if ((err = fticket_pull(tick, uio))) { 951 goto out; 952 } 953 fiio = fticket_resp(tick)->base; 954 955 data->fuse_libabi_major = fiio->major; 956 data->fuse_libabi_minor = fiio->minor; 957 if (!fuse_libabi_geq(data, 7, 4)) { 958 /* 959 * With a little work we could support servers as old as 7.1. 960 * But there would be little payoff. 961 */ 962 SDT_PROBE2(fusefs, , internal, trace, 1, 963 "userpace version too low"); 964 err = EPROTONOSUPPORT; 965 goto out; 966 } 967 968 if (fuse_libabi_geq(data, 7, 5)) { 969 if (fticket_resp(tick)->len == sizeof(struct fuse_init_out) || 970 fticket_resp(tick)->len == FUSE_COMPAT_22_INIT_OUT_SIZE) { 971 data->max_write = fiio->max_write; 972 if (fiio->flags & FUSE_ASYNC_READ) 973 data->dataflags |= FSESS_ASYNC_READ; 974 if (fiio->flags & FUSE_POSIX_LOCKS) 975 data->dataflags |= FSESS_POSIX_LOCKS; 976 if (fiio->flags & FUSE_EXPORT_SUPPORT) 977 data->dataflags |= FSESS_EXPORT_SUPPORT; 978 /* 979 * Don't bother to check FUSE_BIG_WRITES, because it's 980 * redundant with max_write 981 */ 982 /* 983 * max_background and congestion_threshold are not 984 * implemented 985 */ 986 } else { 987 err = EINVAL; 988 } 989 } else { 990 /* Old fixed values */ 991 data->max_write = 4096; 992 } 993 994 if (fuse_libabi_geq(data, 7, 6)) 995 data->max_readahead_blocks = fiio->max_readahead / maxbcachebuf; 996 997 if (!fuse_libabi_geq(data, 7, 7)) 998 fsess_set_notimpl(data->mp, FUSE_INTERRUPT); 999 1000 if (!fuse_libabi_geq(data, 7, 8)) { 1001 fsess_set_notimpl(data->mp, FUSE_BMAP); 1002 fsess_set_notimpl(data->mp, FUSE_DESTROY); 1003 } 1004 1005 if (fuse_libabi_geq(data, 7, 23) && fiio->time_gran >= 1 && 1006 fiio->time_gran <= 1000000000) 1007 data->time_gran = fiio->time_gran; 1008 else 1009 data->time_gran = 1; 1010 1011 if (!fuse_libabi_geq(data, 7, 23)) 1012 data->cache_mode = fuse_data_cache_mode; 1013 else if (fiio->flags & FUSE_WRITEBACK_CACHE) 1014 data->cache_mode = FUSE_CACHE_WB; 1015 else 1016 data->cache_mode = FUSE_CACHE_WT; 1017 1018 out: 1019 if (err) { 1020 fdata_set_dead(data); 1021 } 1022 FUSE_LOCK(); 1023 data->dataflags |= FSESS_INITED; 1024 wakeup(&data->ticketer); 1025 FUSE_UNLOCK(); 1026 1027 return 0; 1028 } 1029 1030 void 1031 fuse_internal_send_init(struct fuse_data *data, struct thread *td) 1032 { 1033 struct fuse_init_in *fiii; 1034 struct fuse_dispatcher fdi; 1035 1036 fdisp_init(&fdi, sizeof(*fiii)); 1037 fdisp_make(&fdi, FUSE_INIT, data->mp, 0, td, NULL); 1038 fiii = fdi.indata; 1039 fiii->major = FUSE_KERNEL_VERSION; 1040 fiii->minor = FUSE_KERNEL_MINOR_VERSION; 1041 /* 1042 * fusefs currently reads ahead no more than one cache block at a time. 1043 * See fuse_read_biobackend 1044 */ 1045 fiii->max_readahead = maxbcachebuf; 1046 /* 1047 * Unsupported features: 1048 * FUSE_FILE_OPS: No known FUSE server or client supports it 1049 * FUSE_ATOMIC_O_TRUNC: our VFS cannot support it 1050 * FUSE_DONT_MASK: unlike Linux, FreeBSD always applies the umask, even 1051 * when default ACLs are in use. 1052 * FUSE_SPLICE_WRITE, FUSE_SPLICE_MOVE, FUSE_SPLICE_READ: FreeBSD 1053 * doesn't have splice(2). 1054 * FUSE_FLOCK_LOCKS: not yet implemented 1055 * FUSE_HAS_IOCTL_DIR: not yet implemented 1056 * FUSE_AUTO_INVAL_DATA: not yet implemented 1057 * FUSE_DO_READDIRPLUS: not yet implemented 1058 * FUSE_READDIRPLUS_AUTO: not yet implemented 1059 * FUSE_ASYNC_DIO: not yet implemented 1060 * FUSE_NO_OPEN_SUPPORT: not yet implemented 1061 */ 1062 fiii->flags = FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_EXPORT_SUPPORT 1063 | FUSE_BIG_WRITES | FUSE_WRITEBACK_CACHE; 1064 1065 fuse_insert_callback(fdi.tick, fuse_internal_init_callback); 1066 fuse_insert_message(fdi.tick, false); 1067 fdisp_destroy(&fdi); 1068 } 1069 1070 /* 1071 * Send a FUSE_SETATTR operation with no permissions checks. If cred is NULL, 1072 * send the request with root credentials 1073 */ 1074 int fuse_internal_setattr(struct vnode *vp, struct vattr *vap, 1075 struct thread *td, struct ucred *cred) 1076 { 1077 struct fuse_vnode_data *fvdat; 1078 struct fuse_dispatcher fdi; 1079 struct fuse_setattr_in *fsai; 1080 struct mount *mp; 1081 pid_t pid = td->td_proc->p_pid; 1082 struct fuse_data *data; 1083 int dataflags; 1084 int err = 0; 1085 enum vtype vtyp; 1086 int sizechanged = -1; 1087 uint64_t newsize = 0; 1088 1089 mp = vnode_mount(vp); 1090 fvdat = VTOFUD(vp); 1091 data = fuse_get_mpdata(mp); 1092 dataflags = data->dataflags; 1093 1094 fdisp_init(&fdi, sizeof(*fsai)); 1095 fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred); 1096 if (!cred) { 1097 fdi.finh->uid = 0; 1098 fdi.finh->gid = 0; 1099 } 1100 fsai = fdi.indata; 1101 fsai->valid = 0; 1102 1103 if (vap->va_uid != (uid_t)VNOVAL) { 1104 fsai->uid = vap->va_uid; 1105 fsai->valid |= FATTR_UID; 1106 } 1107 if (vap->va_gid != (gid_t)VNOVAL) { 1108 fsai->gid = vap->va_gid; 1109 fsai->valid |= FATTR_GID; 1110 } 1111 if (vap->va_size != VNOVAL) { 1112 struct fuse_filehandle *fufh = NULL; 1113 1114 /*Truncate to a new value. */ 1115 fsai->size = vap->va_size; 1116 sizechanged = 1; 1117 newsize = vap->va_size; 1118 fsai->valid |= FATTR_SIZE; 1119 1120 fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); 1121 if (fufh) { 1122 fsai->fh = fufh->fh_id; 1123 fsai->valid |= FATTR_FH; 1124 } 1125 VTOFUD(vp)->flag &= ~FN_SIZECHANGE; 1126 } 1127 if (vap->va_atime.tv_sec != VNOVAL) { 1128 fsai->atime = vap->va_atime.tv_sec; 1129 fsai->atimensec = vap->va_atime.tv_nsec; 1130 fsai->valid |= FATTR_ATIME; 1131 if (vap->va_vaflags & VA_UTIMES_NULL) 1132 fsai->valid |= FATTR_ATIME_NOW; 1133 } 1134 if (vap->va_mtime.tv_sec != VNOVAL) { 1135 fsai->mtime = vap->va_mtime.tv_sec; 1136 fsai->mtimensec = vap->va_mtime.tv_nsec; 1137 fsai->valid |= FATTR_MTIME; 1138 if (vap->va_vaflags & VA_UTIMES_NULL) 1139 fsai->valid |= FATTR_MTIME_NOW; 1140 } else if (fvdat->flag & FN_MTIMECHANGE) { 1141 fsai->mtime = fvdat->cached_attrs.va_mtime.tv_sec; 1142 fsai->mtimensec = fvdat->cached_attrs.va_mtime.tv_nsec; 1143 fsai->valid |= FATTR_MTIME; 1144 } 1145 if (fuse_libabi_geq(data, 7, 23) && fvdat->flag & FN_CTIMECHANGE) { 1146 fsai->ctime = fvdat->cached_attrs.va_ctime.tv_sec; 1147 fsai->ctimensec = fvdat->cached_attrs.va_ctime.tv_nsec; 1148 fsai->valid |= FATTR_CTIME; 1149 } 1150 if (vap->va_mode != (mode_t)VNOVAL) { 1151 fsai->mode = vap->va_mode & ALLPERMS; 1152 fsai->valid |= FATTR_MODE; 1153 } 1154 if (!fsai->valid) { 1155 goto out; 1156 } 1157 1158 if ((err = fdisp_wait_answ(&fdi))) 1159 goto out; 1160 vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode); 1161 1162 if (vnode_vtype(vp) != vtyp) { 1163 if (vnode_vtype(vp) == VNON && vtyp != VNON) { 1164 SDT_PROBE2(fusefs, , internal, trace, 1, "FUSE: Dang! " 1165 "vnode_vtype is VNON and vtype isn't."); 1166 } else { 1167 /* 1168 * STALE vnode, ditch 1169 * 1170 * The vnode has changed its type "behind our back". 1171 * There's nothing really we can do, so let us just 1172 * force an internal revocation and tell the caller to 1173 * try again, if interested. 1174 */ 1175 fuse_internal_vnode_disappear(vp); 1176 err = EAGAIN; 1177 } 1178 } 1179 if (err == 0) { 1180 struct fuse_attr_out *fao = (struct fuse_attr_out*)fdi.answ; 1181 fuse_vnode_undirty_cached_timestamps(vp); 1182 fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid, 1183 fao->attr_valid_nsec, NULL); 1184 } 1185 1186 out: 1187 fdisp_destroy(&fdi); 1188 return err; 1189 } 1190 1191 #ifdef ZERO_PAD_INCOMPLETE_BUFS 1192 static int 1193 isbzero(void *buf, size_t len) 1194 { 1195 int i; 1196 1197 for (i = 0; i < len; i++) { 1198 if (((char *)buf)[i]) 1199 return (0); 1200 } 1201 1202 return (1); 1203 } 1204 1205 #endif 1206 1207 void 1208 fuse_internal_init(void) 1209 { 1210 fuse_lookup_cache_misses = counter_u64_alloc(M_WAITOK); 1211 fuse_lookup_cache_hits = counter_u64_alloc(M_WAITOK); 1212 } 1213 1214 void 1215 fuse_internal_destroy(void) 1216 { 1217 counter_u64_free(fuse_lookup_cache_hits); 1218 counter_u64_free(fuse_lookup_cache_misses); 1219 } 1220