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/param.h> 64 #include <sys/module.h> 65 #include <sys/systm.h> 66 #include <sys/errno.h> 67 #include <sys/kernel.h> 68 #include <sys/conf.h> 69 #include <sys/filio.h> 70 #include <sys/uio.h> 71 #include <sys/malloc.h> 72 #include <sys/queue.h> 73 #include <sys/limits.h> 74 #include <sys/lock.h> 75 #include <sys/rwlock.h> 76 #include <sys/sx.h> 77 #include <sys/proc.h> 78 #include <sys/mount.h> 79 #include <sys/vnode.h> 80 #include <sys/namei.h> 81 #include <sys/extattr.h> 82 #include <sys/stat.h> 83 #include <sys/unistd.h> 84 #include <sys/filedesc.h> 85 #include <sys/file.h> 86 #include <sys/fcntl.h> 87 #include <sys/dirent.h> 88 #include <sys/bio.h> 89 #include <sys/buf.h> 90 #include <sys/sysctl.h> 91 #include <sys/vmmeter.h> 92 93 #include <vm/vm.h> 94 #include <vm/vm_extern.h> 95 #include <vm/pmap.h> 96 #include <vm/vm_map.h> 97 #include <vm/vm_page.h> 98 #include <vm/vm_param.h> 99 #include <vm/vm_object.h> 100 #include <vm/vm_pager.h> 101 #include <vm/vnode_pager.h> 102 #include <vm/vm_object.h> 103 104 #include "fuse.h" 105 #include "fuse_file.h" 106 #include "fuse_internal.h" 107 #include "fuse_ipc.h" 108 #include "fuse_node.h" 109 #include "fuse_io.h" 110 111 #include <sys/priv.h> 112 113 /* Maximum number of hardlinks to a single FUSE file */ 114 #define FUSE_LINK_MAX UINT32_MAX 115 116 SDT_PROVIDER_DECLARE(fusefs); 117 /* 118 * Fuse trace probe: 119 * arg0: verbosity. Higher numbers give more verbose messages 120 * arg1: Textual message 121 */ 122 SDT_PROBE_DEFINE2(fusefs, , vnops, trace, "int", "char*"); 123 124 /* vnode ops */ 125 static vop_access_t fuse_vnop_access; 126 static vop_advlock_t fuse_vnop_advlock; 127 static vop_allocate_t fuse_vnop_allocate; 128 static vop_bmap_t fuse_vnop_bmap; 129 static vop_close_t fuse_fifo_close; 130 static vop_close_t fuse_vnop_close; 131 static vop_copy_file_range_t fuse_vnop_copy_file_range; 132 static vop_create_t fuse_vnop_create; 133 static vop_deallocate_t fuse_vnop_deallocate; 134 static vop_deleteextattr_t fuse_vnop_deleteextattr; 135 static vop_fdatasync_t fuse_vnop_fdatasync; 136 static vop_fsync_t fuse_vnop_fsync; 137 static vop_getattr_t fuse_vnop_getattr; 138 static vop_getextattr_t fuse_vnop_getextattr; 139 static vop_inactive_t fuse_vnop_inactive; 140 static vop_ioctl_t fuse_vnop_ioctl; 141 static vop_link_t fuse_vnop_link; 142 static vop_listextattr_t fuse_vnop_listextattr; 143 static vop_lookup_t fuse_vnop_lookup; 144 static vop_mkdir_t fuse_vnop_mkdir; 145 static vop_mknod_t fuse_vnop_mknod; 146 static vop_open_t fuse_vnop_open; 147 static vop_pathconf_t fuse_vnop_pathconf; 148 static vop_read_t fuse_vnop_read; 149 static vop_readdir_t fuse_vnop_readdir; 150 static vop_readlink_t fuse_vnop_readlink; 151 static vop_reclaim_t fuse_vnop_reclaim; 152 static vop_remove_t fuse_vnop_remove; 153 static vop_rename_t fuse_vnop_rename; 154 static vop_rmdir_t fuse_vnop_rmdir; 155 static vop_setattr_t fuse_vnop_setattr; 156 static vop_setextattr_t fuse_vnop_setextattr; 157 static vop_strategy_t fuse_vnop_strategy; 158 static vop_symlink_t fuse_vnop_symlink; 159 static vop_write_t fuse_vnop_write; 160 static vop_getpages_t fuse_vnop_getpages; 161 static vop_print_t fuse_vnop_print; 162 static vop_vptofh_t fuse_vnop_vptofh; 163 164 struct vop_vector fuse_fifoops = { 165 .vop_default = &fifo_specops, 166 .vop_access = fuse_vnop_access, 167 .vop_close = fuse_fifo_close, 168 .vop_fsync = fuse_vnop_fsync, 169 .vop_getattr = fuse_vnop_getattr, 170 .vop_inactive = fuse_vnop_inactive, 171 .vop_pathconf = fuse_vnop_pathconf, 172 .vop_print = fuse_vnop_print, 173 .vop_read = VOP_PANIC, 174 .vop_reclaim = fuse_vnop_reclaim, 175 .vop_setattr = fuse_vnop_setattr, 176 .vop_write = VOP_PANIC, 177 .vop_vptofh = fuse_vnop_vptofh, 178 }; 179 VFS_VOP_VECTOR_REGISTER(fuse_fifoops); 180 181 struct vop_vector fuse_vnops = { 182 .vop_allocate = fuse_vnop_allocate, 183 .vop_default = &default_vnodeops, 184 .vop_access = fuse_vnop_access, 185 .vop_advlock = fuse_vnop_advlock, 186 .vop_bmap = fuse_vnop_bmap, 187 .vop_close = fuse_vnop_close, 188 .vop_copy_file_range = fuse_vnop_copy_file_range, 189 .vop_create = fuse_vnop_create, 190 .vop_deallocate = fuse_vnop_deallocate, 191 .vop_deleteextattr = fuse_vnop_deleteextattr, 192 .vop_fsync = fuse_vnop_fsync, 193 .vop_fdatasync = fuse_vnop_fdatasync, 194 .vop_getattr = fuse_vnop_getattr, 195 .vop_getextattr = fuse_vnop_getextattr, 196 .vop_inactive = fuse_vnop_inactive, 197 .vop_ioctl = fuse_vnop_ioctl, 198 .vop_link = fuse_vnop_link, 199 .vop_listextattr = fuse_vnop_listextattr, 200 .vop_lookup = fuse_vnop_lookup, 201 .vop_mkdir = fuse_vnop_mkdir, 202 .vop_mknod = fuse_vnop_mknod, 203 .vop_open = fuse_vnop_open, 204 .vop_pathconf = fuse_vnop_pathconf, 205 /* 206 * TODO: implement vop_poll after upgrading to protocol 7.21. 207 * FUSE_POLL was added in protocol 7.11, but it's kind of broken until 208 * 7.21, which adds the ability for the client to choose which poll 209 * events it wants, and for a client to deregister a file handle 210 */ 211 .vop_read = fuse_vnop_read, 212 .vop_readdir = fuse_vnop_readdir, 213 .vop_readlink = fuse_vnop_readlink, 214 .vop_reclaim = fuse_vnop_reclaim, 215 .vop_remove = fuse_vnop_remove, 216 .vop_rename = fuse_vnop_rename, 217 .vop_rmdir = fuse_vnop_rmdir, 218 .vop_setattr = fuse_vnop_setattr, 219 .vop_setextattr = fuse_vnop_setextattr, 220 .vop_strategy = fuse_vnop_strategy, 221 .vop_symlink = fuse_vnop_symlink, 222 .vop_write = fuse_vnop_write, 223 .vop_getpages = fuse_vnop_getpages, 224 .vop_print = fuse_vnop_print, 225 .vop_vptofh = fuse_vnop_vptofh, 226 }; 227 VFS_VOP_VECTOR_REGISTER(fuse_vnops); 228 229 /* Check permission for extattr operations, much like extattr_check_cred */ 230 static int 231 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred, 232 struct thread *td, accmode_t accmode) 233 { 234 struct mount *mp = vnode_mount(vp); 235 struct fuse_data *data = fuse_get_mpdata(mp); 236 int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS; 237 238 /* 239 * Kernel-invoked always succeeds. 240 */ 241 if (cred == NOCRED) 242 return (0); 243 244 /* 245 * Do not allow privileged processes in jail to directly manipulate 246 * system attributes. 247 */ 248 switch (ns) { 249 case EXTATTR_NAMESPACE_SYSTEM: 250 if (default_permissions) { 251 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM)); 252 } 253 return (0); 254 case EXTATTR_NAMESPACE_USER: 255 if (default_permissions) { 256 return (fuse_internal_access(vp, accmode, td, cred)); 257 } 258 return (0); 259 default: 260 return (EPERM); 261 } 262 } 263 264 /* Get a filehandle for a directory */ 265 static int 266 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp, 267 struct ucred *cred, pid_t pid) 268 { 269 if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0) 270 return 0; 271 return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid); 272 } 273 274 /* Send FUSE_FLUSH for this vnode */ 275 static int 276 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag) 277 { 278 struct fuse_flush_in *ffi; 279 struct fuse_filehandle *fufh; 280 struct fuse_dispatcher fdi; 281 struct thread *td = curthread; 282 struct mount *mp = vnode_mount(vp); 283 int err; 284 285 if (fsess_not_impl(vnode_mount(vp), FUSE_FLUSH)) 286 return 0; 287 288 err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid); 289 if (err) 290 return err; 291 292 fdisp_init(&fdi, sizeof(*ffi)); 293 fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred); 294 ffi = fdi.indata; 295 ffi->fh = fufh->fh_id; 296 /* 297 * If the file has a POSIX lock then we're supposed to set lock_owner. 298 * If not, then lock_owner is undefined. So we may as well always set 299 * it. 300 */ 301 ffi->lock_owner = td->td_proc->p_pid; 302 303 err = fdisp_wait_answ(&fdi); 304 if (err == ENOSYS) { 305 fsess_set_notimpl(mp, FUSE_FLUSH); 306 err = 0; 307 } 308 fdisp_destroy(&fdi); 309 return err; 310 } 311 312 /* Close wrapper for fifos. */ 313 static int 314 fuse_fifo_close(struct vop_close_args *ap) 315 { 316 return (fifo_specops.vop_close(ap)); 317 } 318 319 /* Invalidate a range of cached data, whether dirty of not */ 320 static int 321 fuse_inval_buf_range(struct vnode *vp, off_t filesize, off_t start, off_t end) 322 { 323 struct buf *bp; 324 daddr_t left_lbn, end_lbn, right_lbn; 325 off_t new_filesize; 326 int iosize, left_on, right_on, right_blksize; 327 328 iosize = fuse_iosize(vp); 329 left_lbn = start / iosize; 330 end_lbn = howmany(end, iosize); 331 left_on = start & (iosize - 1); 332 if (left_on != 0) { 333 bp = getblk(vp, left_lbn, iosize, PCATCH, 0, 0); 334 if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyend >= left_on) { 335 /* 336 * Flush the dirty buffer, because we don't have a 337 * byte-granular way to record which parts of the 338 * buffer are valid. 339 */ 340 bwrite(bp); 341 if (bp->b_error) 342 return (bp->b_error); 343 } else { 344 brelse(bp); 345 } 346 } 347 right_on = end & (iosize - 1); 348 if (right_on != 0) { 349 right_lbn = end / iosize; 350 new_filesize = MAX(filesize, end); 351 right_blksize = MIN(iosize, new_filesize - iosize * right_lbn); 352 bp = getblk(vp, right_lbn, right_blksize, PCATCH, 0, 0); 353 if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyoff < right_on) { 354 /* 355 * Flush the dirty buffer, because we don't have a 356 * byte-granular way to record which parts of the 357 * buffer are valid. 358 */ 359 bwrite(bp); 360 if (bp->b_error) 361 return (bp->b_error); 362 } else { 363 brelse(bp); 364 } 365 } 366 367 v_inval_buf_range(vp, left_lbn, end_lbn, iosize); 368 return (0); 369 } 370 371 372 /* Send FUSE_LSEEK for this node */ 373 static int 374 fuse_vnop_do_lseek(struct vnode *vp, struct thread *td, struct ucred *cred, 375 pid_t pid, off_t *offp, int whence) 376 { 377 struct fuse_dispatcher fdi; 378 struct fuse_filehandle *fufh; 379 struct fuse_lseek_in *flsi; 380 struct fuse_lseek_out *flso; 381 struct mount *mp = vnode_mount(vp); 382 int err; 383 384 ASSERT_VOP_LOCKED(vp, __func__); 385 386 err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid); 387 if (err) 388 return (err); 389 fdisp_init(&fdi, sizeof(*flsi)); 390 fdisp_make_vp(&fdi, FUSE_LSEEK, vp, td, cred); 391 flsi = fdi.indata; 392 flsi->fh = fufh->fh_id; 393 flsi->offset = *offp; 394 flsi->whence = whence; 395 err = fdisp_wait_answ(&fdi); 396 if (err == ENOSYS) { 397 fsess_set_notimpl(mp, FUSE_LSEEK); 398 } else if (err == 0) { 399 fsess_set_impl(mp, FUSE_LSEEK); 400 flso = fdi.answ; 401 *offp = flso->offset; 402 } 403 fdisp_destroy(&fdi); 404 405 return (err); 406 } 407 408 /* 409 struct vnop_access_args { 410 struct vnode *a_vp; 411 #if VOP_ACCESS_TAKES_ACCMODE_T 412 accmode_t a_accmode; 413 #else 414 int a_mode; 415 #endif 416 struct ucred *a_cred; 417 struct thread *a_td; 418 }; 419 */ 420 static int 421 fuse_vnop_access(struct vop_access_args *ap) 422 { 423 struct vnode *vp = ap->a_vp; 424 int accmode = ap->a_accmode; 425 struct ucred *cred = ap->a_cred; 426 427 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp)); 428 429 int err; 430 431 if (fuse_isdeadfs(vp)) { 432 if (vnode_isvroot(vp)) { 433 return 0; 434 } 435 return ENXIO; 436 } 437 if (!(data->dataflags & FSESS_INITED)) { 438 if (vnode_isvroot(vp)) { 439 if (priv_check_cred(cred, PRIV_VFS_ADMIN) || 440 (fuse_match_cred(data->daemoncred, cred) == 0)) { 441 return 0; 442 } 443 } 444 return EBADF; 445 } 446 if (vnode_islnk(vp)) { 447 return 0; 448 } 449 450 err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred); 451 return err; 452 } 453 454 /* 455 * struct vop_advlock_args { 456 * struct vop_generic_args a_gen; 457 * struct vnode *a_vp; 458 * void *a_id; 459 * int a_op; 460 * struct flock *a_fl; 461 * int a_flags; 462 * } 463 */ 464 static int 465 fuse_vnop_advlock(struct vop_advlock_args *ap) 466 { 467 struct vnode *vp = ap->a_vp; 468 struct flock *fl = ap->a_fl; 469 struct thread *td = curthread; 470 struct ucred *cred = td->td_ucred; 471 pid_t pid = td->td_proc->p_pid; 472 struct fuse_filehandle *fufh; 473 struct fuse_dispatcher fdi; 474 struct fuse_lk_in *fli; 475 struct fuse_lk_out *flo; 476 struct vattr vattr; 477 enum fuse_opcode op; 478 off_t size, start; 479 int dataflags, err; 480 int flags = ap->a_flags; 481 482 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags; 483 484 if (fuse_isdeadfs(vp)) { 485 return ENXIO; 486 } 487 488 switch(ap->a_op) { 489 case F_GETLK: 490 op = FUSE_GETLK; 491 break; 492 case F_SETLK: 493 if (flags & F_WAIT) 494 op = FUSE_SETLKW; 495 else 496 op = FUSE_SETLK; 497 break; 498 case F_UNLCK: 499 op = FUSE_SETLK; 500 break; 501 default: 502 return EINVAL; 503 } 504 505 if (!(dataflags & FSESS_POSIX_LOCKS)) 506 return vop_stdadvlock(ap); 507 /* FUSE doesn't properly support flock until protocol 7.17 */ 508 if (flags & F_FLOCK) 509 return vop_stdadvlock(ap); 510 511 vn_lock(vp, LK_SHARED | LK_RETRY); 512 513 switch (fl->l_whence) { 514 case SEEK_SET: 515 case SEEK_CUR: 516 /* 517 * Caller is responsible for adding any necessary offset 518 * when SEEK_CUR is used. 519 */ 520 start = fl->l_start; 521 break; 522 523 case SEEK_END: 524 err = fuse_internal_getattr(vp, &vattr, cred, td); 525 if (err) 526 goto out; 527 size = vattr.va_size; 528 if (size > OFF_MAX || 529 (fl->l_start > 0 && size > OFF_MAX - fl->l_start)) { 530 err = EOVERFLOW; 531 goto out; 532 } 533 start = size + fl->l_start; 534 break; 535 536 default: 537 return (EINVAL); 538 } 539 540 err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid); 541 if (err) 542 goto out; 543 544 fdisp_init(&fdi, sizeof(*fli)); 545 546 fdisp_make_vp(&fdi, op, vp, td, cred); 547 fli = fdi.indata; 548 fli->fh = fufh->fh_id; 549 fli->owner = td->td_proc->p_pid; 550 fli->lk.start = start; 551 if (fl->l_len != 0) 552 fli->lk.end = start + fl->l_len - 1; 553 else 554 fli->lk.end = INT64_MAX; 555 fli->lk.type = fl->l_type; 556 fli->lk.pid = td->td_proc->p_pid; 557 558 err = fdisp_wait_answ(&fdi); 559 fdisp_destroy(&fdi); 560 561 if (err == 0 && op == FUSE_GETLK) { 562 flo = fdi.answ; 563 fl->l_type = flo->lk.type; 564 fl->l_whence = SEEK_SET; 565 if (flo->lk.type != F_UNLCK) { 566 fl->l_pid = flo->lk.pid; 567 fl->l_start = flo->lk.start; 568 if (flo->lk.end == INT64_MAX) 569 fl->l_len = 0; 570 else 571 fl->l_len = flo->lk.end - flo->lk.start + 1; 572 fl->l_start = flo->lk.start; 573 } 574 } 575 576 out: 577 VOP_UNLOCK(vp); 578 return err; 579 } 580 581 static int 582 fuse_vnop_allocate(struct vop_allocate_args *ap) 583 { 584 struct vnode *vp = ap->a_vp; 585 off_t *len = ap->a_len; 586 off_t *offset = ap->a_offset; 587 struct ucred *cred = ap->a_cred; 588 struct fuse_filehandle *fufh; 589 struct mount *mp = vnode_mount(vp); 590 struct fuse_dispatcher fdi; 591 struct fuse_fallocate_in *ffi; 592 struct uio io; 593 pid_t pid = curthread->td_proc->p_pid; 594 struct fuse_vnode_data *fvdat = VTOFUD(vp); 595 off_t filesize; 596 int err; 597 598 if (fuse_isdeadfs(vp)) 599 return (ENXIO); 600 601 switch (vp->v_type) { 602 case VFIFO: 603 return (ESPIPE); 604 case VLNK: 605 case VREG: 606 if (vfs_isrdonly(mp)) 607 return (EROFS); 608 break; 609 default: 610 return (ENODEV); 611 } 612 613 if (vfs_isrdonly(mp)) 614 return (EROFS); 615 616 if (fsess_not_impl(mp, FUSE_FALLOCATE)) 617 return (EINVAL); 618 619 io.uio_offset = *offset; 620 io.uio_resid = *len; 621 err = vn_rlimit_fsize(vp, &io, curthread); 622 if (err) 623 return (err); 624 625 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); 626 if (err) 627 return (err); 628 629 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE); 630 631 err = fuse_vnode_size(vp, &filesize, cred, curthread); 632 if (err) 633 return (err); 634 fuse_inval_buf_range(vp, filesize, *offset, *offset + *len); 635 636 fdisp_init(&fdi, sizeof(*ffi)); 637 fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred); 638 ffi = fdi.indata; 639 ffi->fh = fufh->fh_id; 640 ffi->offset = *offset; 641 ffi->length = *len; 642 ffi->mode = 0; 643 err = fdisp_wait_answ(&fdi); 644 645 if (err == ENOSYS) { 646 fsess_set_notimpl(mp, FUSE_FALLOCATE); 647 err = EINVAL; 648 } else if (err == EOPNOTSUPP) { 649 /* 650 * The file system server does not support FUSE_FALLOCATE with 651 * the supplied mode for this particular file. 652 */ 653 err = EINVAL; 654 } else if (!err) { 655 *offset += *len; 656 *len = 0; 657 fuse_vnode_undirty_cached_timestamps(vp, false); 658 fuse_internal_clear_suid_on_write(vp, cred, curthread); 659 if (*offset > fvdat->cached_attrs.va_size) { 660 fuse_vnode_setsize(vp, *offset, false); 661 getnanouptime(&fvdat->last_local_modify); 662 } 663 } 664 665 fdisp_destroy(&fdi); 666 return (err); 667 } 668 669 /* { 670 struct vnode *a_vp; 671 daddr_t a_bn; 672 struct bufobj **a_bop; 673 daddr_t *a_bnp; 674 int *a_runp; 675 int *a_runb; 676 } */ 677 static int 678 fuse_vnop_bmap(struct vop_bmap_args *ap) 679 { 680 struct vnode *vp = ap->a_vp; 681 struct bufobj **bo = ap->a_bop; 682 struct thread *td = curthread; 683 struct mount *mp; 684 struct fuse_dispatcher fdi; 685 struct fuse_bmap_in *fbi; 686 struct fuse_bmap_out *fbo; 687 struct fuse_data *data; 688 struct fuse_vnode_data *fvdat = VTOFUD(vp); 689 uint64_t biosize; 690 off_t fsize; 691 daddr_t lbn = ap->a_bn; 692 daddr_t *pbn = ap->a_bnp; 693 int *runp = ap->a_runp; 694 int *runb = ap->a_runb; 695 int error = 0; 696 int maxrun; 697 698 if (fuse_isdeadfs(vp)) { 699 return ENXIO; 700 } 701 702 mp = vnode_mount(vp); 703 data = fuse_get_mpdata(mp); 704 biosize = fuse_iosize(vp); 705 maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1, 706 data->max_readahead_blocks); 707 708 if (bo != NULL) 709 *bo = &vp->v_bufobj; 710 711 /* 712 * The FUSE_BMAP operation does not include the runp and runb 713 * variables, so we must guess. Report nonzero contiguous runs so 714 * cluster_read will combine adjacent reads. It's worthwhile to reduce 715 * upcalls even if we don't know the true physical layout of the file. 716 * 717 * FUSE file systems may opt out of read clustering in two ways: 718 * * mounting with -onoclusterr 719 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT 720 */ 721 if (runb != NULL) 722 *runb = MIN(lbn, maxrun); 723 if (runp != NULL && maxrun == 0) 724 *runp = 0; 725 else if (runp != NULL) { 726 /* 727 * If the file's size is cached, use that value to calculate 728 * runp, even if the cache is expired. runp is only advisory, 729 * and the risk of getting it wrong is not worth the cost of 730 * another upcall. 731 */ 732 if (fvdat->cached_attrs.va_size != VNOVAL) 733 fsize = fvdat->cached_attrs.va_size; 734 else 735 error = fuse_vnode_size(vp, &fsize, td->td_ucred, td); 736 if (error == 0) 737 *runp = MIN(MAX(0, fsize / (off_t)biosize - lbn - 1), 738 maxrun); 739 else 740 *runp = 0; 741 } 742 743 if (fsess_maybe_impl(mp, FUSE_BMAP)) { 744 fdisp_init(&fdi, sizeof(*fbi)); 745 fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred); 746 fbi = fdi.indata; 747 fbi->block = lbn; 748 fbi->blocksize = biosize; 749 error = fdisp_wait_answ(&fdi); 750 if (error == ENOSYS) { 751 fdisp_destroy(&fdi); 752 fsess_set_notimpl(mp, FUSE_BMAP); 753 error = 0; 754 } else { 755 fbo = fdi.answ; 756 if (error == 0 && pbn != NULL) 757 *pbn = fbo->block; 758 fdisp_destroy(&fdi); 759 return error; 760 } 761 } 762 763 /* If the daemon doesn't support BMAP, make up a sensible default */ 764 if (pbn != NULL) 765 *pbn = lbn * btodb(biosize); 766 return (error); 767 } 768 769 /* 770 struct vop_close_args { 771 struct vnode *a_vp; 772 int a_fflag; 773 struct ucred *a_cred; 774 struct thread *a_td; 775 }; 776 */ 777 static int 778 fuse_vnop_close(struct vop_close_args *ap) 779 { 780 struct vnode *vp = ap->a_vp; 781 struct mount *mp = vnode_mount(vp); 782 struct ucred *cred = ap->a_cred; 783 int fflag = ap->a_fflag; 784 struct thread *td = ap->a_td; 785 pid_t pid = td->td_proc->p_pid; 786 struct fuse_vnode_data *fvdat = VTOFUD(vp); 787 int err = 0; 788 789 if (fuse_isdeadfs(vp)) 790 return 0; 791 if (vnode_isdir(vp)) 792 return 0; 793 if (fflag & IO_NDELAY) 794 return 0; 795 796 err = fuse_flush(vp, cred, pid, fflag); 797 if (err == 0 && (fvdat->flag & FN_ATIMECHANGE) && !vfs_isrdonly(mp)) { 798 struct vattr vap; 799 struct fuse_data *data; 800 int dataflags; 801 int access_e = 0; 802 803 data = fuse_get_mpdata(mp); 804 dataflags = data->dataflags; 805 if (dataflags & FSESS_DEFAULT_PERMISSIONS) { 806 struct vattr va; 807 808 fuse_internal_getattr(vp, &va, cred, td); 809 access_e = vaccess(vp->v_type, va.va_mode, va.va_uid, 810 va.va_gid, VWRITE, cred); 811 } 812 if (access_e == 0) { 813 VATTR_NULL(&vap); 814 vap.va_atime = fvdat->cached_attrs.va_atime; 815 /* 816 * Ignore errors setting when setting atime. That 817 * should not cause close(2) to fail. 818 */ 819 fuse_internal_setattr(vp, &vap, td, NULL); 820 } 821 } 822 /* TODO: close the file handle, if we're sure it's no longer used */ 823 if ((fvdat->flag & FN_SIZECHANGE) != 0) { 824 fuse_vnode_savesize(vp, cred, td->td_proc->p_pid); 825 } 826 return err; 827 } 828 829 /* 830 struct vop_copy_file_range_args { 831 struct vop_generic_args a_gen; 832 struct vnode *a_invp; 833 off_t *a_inoffp; 834 struct vnode *a_outvp; 835 off_t *a_outoffp; 836 size_t *a_lenp; 837 unsigned int a_flags; 838 struct ucred *a_incred; 839 struct ucred *a_outcred; 840 struct thread *a_fsizetd; 841 } 842 */ 843 static int 844 fuse_vnop_copy_file_range(struct vop_copy_file_range_args *ap) 845 { 846 struct vnode *invp = ap->a_invp; 847 struct vnode *outvp = ap->a_outvp; 848 struct mount *mp = vnode_mount(invp); 849 struct fuse_vnode_data *outfvdat = VTOFUD(outvp); 850 struct fuse_dispatcher fdi; 851 struct fuse_filehandle *infufh, *outfufh; 852 struct fuse_copy_file_range_in *fcfri; 853 struct ucred *incred = ap->a_incred; 854 struct ucred *outcred = ap->a_outcred; 855 struct fuse_write_out *fwo; 856 struct thread *td; 857 struct uio io; 858 off_t outfilesize; 859 ssize_t r = 0; 860 pid_t pid; 861 int err; 862 863 err = ENOSYS; 864 if (mp == NULL || mp != vnode_mount(outvp)) 865 goto fallback; 866 867 if (incred->cr_uid != outcred->cr_uid) 868 goto fallback; 869 870 if (incred->cr_groups[0] != outcred->cr_groups[0]) 871 goto fallback; 872 873 /* Caller busied mp, mnt_data can be safely accessed. */ 874 if (fsess_not_impl(mp, FUSE_COPY_FILE_RANGE)) 875 goto fallback; 876 877 if (ap->a_fsizetd == NULL) 878 td = curthread; 879 else 880 td = ap->a_fsizetd; 881 pid = td->td_proc->p_pid; 882 883 vn_lock_pair(invp, false, LK_SHARED, outvp, false, LK_EXCLUSIVE); 884 if (invp->v_data == NULL || outvp->v_data == NULL) { 885 err = EBADF; 886 goto unlock; 887 } 888 889 err = fuse_filehandle_getrw(invp, FREAD, &infufh, incred, pid); 890 if (err) 891 goto unlock; 892 893 err = fuse_filehandle_getrw(outvp, FWRITE, &outfufh, outcred, pid); 894 if (err) 895 goto unlock; 896 897 io.uio_resid = *ap->a_lenp; 898 if (ap->a_fsizetd) { 899 io.uio_offset = *ap->a_outoffp; 900 err = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd); 901 if (err != 0) 902 goto unlock; 903 } 904 905 err = fuse_vnode_size(outvp, &outfilesize, outcred, curthread); 906 if (err) 907 goto unlock; 908 909 err = fuse_inval_buf_range(outvp, outfilesize, *ap->a_outoffp, 910 *ap->a_outoffp + io.uio_resid); 911 if (err) 912 goto unlock; 913 914 fdisp_init(&fdi, sizeof(*fcfri)); 915 fdisp_make_vp(&fdi, FUSE_COPY_FILE_RANGE, invp, td, incred); 916 fcfri = fdi.indata; 917 fcfri->fh_in = infufh->fh_id; 918 fcfri->off_in = *ap->a_inoffp; 919 fcfri->nodeid_out = VTOI(outvp); 920 fcfri->fh_out = outfufh->fh_id; 921 fcfri->off_out = *ap->a_outoffp; 922 fcfri->len = io.uio_resid; 923 fcfri->flags = 0; 924 925 err = fdisp_wait_answ(&fdi); 926 if (err == 0) { 927 fwo = fdi.answ; 928 *ap->a_lenp = fwo->size; 929 *ap->a_inoffp += fwo->size; 930 *ap->a_outoffp += fwo->size; 931 fuse_internal_clear_suid_on_write(outvp, outcred, td); 932 if (*ap->a_outoffp > outfvdat->cached_attrs.va_size) { 933 fuse_vnode_setsize(outvp, *ap->a_outoffp, false); 934 getnanouptime(&outfvdat->last_local_modify); 935 } 936 fuse_vnode_update(invp, FN_ATIMECHANGE); 937 fuse_vnode_update(outvp, FN_MTIMECHANGE | FN_CTIMECHANGE); 938 } 939 fdisp_destroy(&fdi); 940 941 unlock: 942 if (invp != outvp) 943 VOP_UNLOCK(invp); 944 VOP_UNLOCK(outvp); 945 946 if (err == ENOSYS) 947 fsess_set_notimpl(mp, FUSE_COPY_FILE_RANGE); 948 fallback: 949 950 /* 951 * No need to call vn_rlimit_fsizex_res before return, since the uio is 952 * local. 953 */ 954 return (err); 955 } 956 957 static void 958 fdisp_make_mknod_for_fallback( 959 struct fuse_dispatcher *fdip, 960 struct componentname *cnp, 961 struct vnode *dvp, 962 uint64_t parentnid, 963 struct thread *td, 964 struct ucred *cred, 965 mode_t mode, 966 enum fuse_opcode *op) 967 { 968 struct fuse_mknod_in *fmni; 969 970 fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1); 971 *op = FUSE_MKNOD; 972 fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred); 973 fmni = fdip->indata; 974 fmni->mode = mode; 975 fmni->rdev = 0; 976 memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr, 977 cnp->cn_namelen); 978 ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0'; 979 } 980 /* 981 struct vnop_create_args { 982 struct vnode *a_dvp; 983 struct vnode **a_vpp; 984 struct componentname *a_cnp; 985 struct vattr *a_vap; 986 }; 987 */ 988 static int 989 fuse_vnop_create(struct vop_create_args *ap) 990 { 991 struct vnode *dvp = ap->a_dvp; 992 struct vnode **vpp = ap->a_vpp; 993 struct componentname *cnp = ap->a_cnp; 994 struct vattr *vap = ap->a_vap; 995 struct thread *td = curthread; 996 struct ucred *cred = cnp->cn_cred; 997 998 struct fuse_data *data; 999 struct fuse_create_in *fci; 1000 struct fuse_entry_out *feo; 1001 struct fuse_open_out *foo; 1002 struct fuse_dispatcher fdi, fdi2; 1003 struct fuse_dispatcher *fdip = &fdi; 1004 struct fuse_dispatcher *fdip2 = NULL; 1005 1006 int err; 1007 1008 struct mount *mp = vnode_mount(dvp); 1009 data = fuse_get_mpdata(mp); 1010 uint64_t parentnid = VTOFUD(dvp)->nid; 1011 mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode); 1012 enum fuse_opcode op; 1013 int flags; 1014 1015 if (fuse_isdeadfs(dvp)) 1016 return ENXIO; 1017 1018 /* FUSE expects sockets to be created with FUSE_MKNOD */ 1019 if (vap->va_type == VSOCK) 1020 return fuse_internal_mknod(dvp, vpp, cnp, vap); 1021 1022 /* 1023 * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a 1024 * writable mode makes sense, and we might as well include readability 1025 * too. 1026 */ 1027 flags = O_RDWR; 1028 1029 bzero(&fdi, sizeof(fdi)); 1030 1031 if (vap->va_type != VREG) 1032 return (EINVAL); 1033 1034 if (fsess_not_impl(mp, FUSE_CREATE) || vap->va_type == VSOCK) { 1035 /* Fallback to FUSE_MKNOD/FUSE_OPEN */ 1036 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td, 1037 cred, mode, &op); 1038 } else { 1039 /* Use FUSE_CREATE */ 1040 size_t insize; 1041 1042 op = FUSE_CREATE; 1043 fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1); 1044 fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred); 1045 fci = fdip->indata; 1046 fci->mode = mode; 1047 fci->flags = O_CREAT | flags; 1048 if (fuse_libabi_geq(data, 7, 12)) { 1049 insize = sizeof(*fci); 1050 fci->umask = td->td_proc->p_pd->pd_cmask; 1051 } else { 1052 insize = sizeof(struct fuse_open_in); 1053 } 1054 1055 memcpy((char *)fdip->indata + insize, cnp->cn_nameptr, 1056 cnp->cn_namelen); 1057 ((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0'; 1058 } 1059 1060 err = fdisp_wait_answ(fdip); 1061 1062 if (err) { 1063 if (err == ENOSYS && op == FUSE_CREATE) { 1064 fsess_set_notimpl(mp, FUSE_CREATE); 1065 fdisp_destroy(fdip); 1066 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, 1067 parentnid, td, cred, mode, &op); 1068 err = fdisp_wait_answ(fdip); 1069 } 1070 if (err) 1071 goto out; 1072 } 1073 1074 feo = fdip->answ; 1075 1076 if ((err = fuse_internal_checkentry(feo, vap->va_type))) { 1077 goto out; 1078 } 1079 1080 if (op == FUSE_CREATE) { 1081 if (fuse_libabi_geq(data, 7, 9)) 1082 foo = (struct fuse_open_out*)(feo + 1); 1083 else 1084 foo = (struct fuse_open_out*)((char*)feo + 1085 FUSE_COMPAT_ENTRY_OUT_SIZE); 1086 } else { 1087 /* Issue a separate FUSE_OPEN */ 1088 struct fuse_open_in *foi; 1089 1090 fdip2 = &fdi2; 1091 fdisp_init(fdip2, sizeof(*foi)); 1092 fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td, 1093 cred); 1094 foi = fdip2->indata; 1095 foi->flags = flags; 1096 err = fdisp_wait_answ(fdip2); 1097 if (err) 1098 goto out; 1099 foo = fdip2->answ; 1100 } 1101 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type); 1102 if (err) { 1103 struct fuse_release_in *fri; 1104 uint64_t nodeid = feo->nodeid; 1105 uint64_t fh_id = foo->fh; 1106 1107 fdisp_destroy(fdip); 1108 fdisp_init(fdip, sizeof(*fri)); 1109 fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred); 1110 fri = fdip->indata; 1111 fri->fh = fh_id; 1112 fri->flags = flags; 1113 fuse_insert_callback(fdip->tick, fuse_internal_forget_callback); 1114 fuse_insert_message(fdip->tick, false); 1115 goto out; 1116 } 1117 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create"); 1118 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid, 1119 feo->attr_valid_nsec, NULL, true); 1120 1121 fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo); 1122 fuse_vnode_open(*vpp, foo->open_flags, td); 1123 /* 1124 * Purge the parent's attribute cache because the daemon should've 1125 * updated its mtime and ctime 1126 */ 1127 fuse_vnode_clear_attr_cache(dvp); 1128 cache_purge_negative(dvp); 1129 1130 out: 1131 if (fdip2) 1132 fdisp_destroy(fdip2); 1133 fdisp_destroy(fdip); 1134 return err; 1135 } 1136 1137 /* 1138 struct vnop_fdatasync_args { 1139 struct vop_generic_args a_gen; 1140 struct vnode * a_vp; 1141 struct thread * a_td; 1142 }; 1143 */ 1144 static int 1145 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap) 1146 { 1147 struct vnode *vp = ap->a_vp; 1148 struct thread *td = ap->a_td; 1149 int waitfor = MNT_WAIT; 1150 1151 int err = 0; 1152 1153 if (fuse_isdeadfs(vp)) { 1154 return 0; 1155 } 1156 if ((err = vop_stdfdatasync_buf(ap))) 1157 return err; 1158 1159 return fuse_internal_fsync(vp, td, waitfor, true); 1160 } 1161 1162 /* 1163 struct vnop_fsync_args { 1164 struct vop_generic_args a_gen; 1165 struct vnode * a_vp; 1166 int a_waitfor; 1167 struct thread * a_td; 1168 }; 1169 */ 1170 static int 1171 fuse_vnop_fsync(struct vop_fsync_args *ap) 1172 { 1173 struct vnode *vp = ap->a_vp; 1174 struct thread *td = ap->a_td; 1175 int waitfor = ap->a_waitfor; 1176 int err = 0; 1177 1178 if (fuse_isdeadfs(vp)) { 1179 return 0; 1180 } 1181 if ((err = vop_stdfsync(ap))) 1182 return err; 1183 1184 return fuse_internal_fsync(vp, td, waitfor, false); 1185 } 1186 1187 /* 1188 struct vnop_getattr_args { 1189 struct vnode *a_vp; 1190 struct vattr *a_vap; 1191 struct ucred *a_cred; 1192 struct thread *a_td; 1193 }; 1194 */ 1195 static int 1196 fuse_vnop_getattr(struct vop_getattr_args *ap) 1197 { 1198 struct vnode *vp = ap->a_vp; 1199 struct vattr *vap = ap->a_vap; 1200 struct ucred *cred = ap->a_cred; 1201 struct thread *td = curthread; 1202 1203 int err = 0; 1204 int dataflags; 1205 1206 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags; 1207 1208 /* Note that we are not bailing out on a dead file system just yet. */ 1209 1210 if (!(dataflags & FSESS_INITED)) { 1211 if (!vnode_isvroot(vp)) { 1212 fdata_set_dead(fuse_get_mpdata(vnode_mount(vp))); 1213 err = ENOTCONN; 1214 return err; 1215 } else { 1216 goto fake; 1217 } 1218 } 1219 err = fuse_internal_getattr(vp, vap, cred, td); 1220 if (err == ENOTCONN && vnode_isvroot(vp)) { 1221 /* see comment in fuse_vfsop_statfs() */ 1222 goto fake; 1223 } else { 1224 return err; 1225 } 1226 1227 fake: 1228 bzero(vap, sizeof(*vap)); 1229 vap->va_type = vnode_vtype(vp); 1230 1231 return 0; 1232 } 1233 1234 /* 1235 struct vnop_inactive_args { 1236 struct vnode *a_vp; 1237 }; 1238 */ 1239 static int 1240 fuse_vnop_inactive(struct vop_inactive_args *ap) 1241 { 1242 struct vnode *vp = ap->a_vp; 1243 struct thread *td = curthread; 1244 1245 struct fuse_vnode_data *fvdat = VTOFUD(vp); 1246 struct fuse_filehandle *fufh, *fufh_tmp; 1247 1248 int need_flush = 1; 1249 1250 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) { 1251 if (need_flush && vp->v_type == VREG) { 1252 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) { 1253 fuse_vnode_savesize(vp, NULL, 0); 1254 } 1255 if ((fvdat->flag & FN_REVOKED) != 0) 1256 fuse_io_invalbuf(vp, td); 1257 else 1258 fuse_io_flushbuf(vp, MNT_WAIT, td); 1259 need_flush = 0; 1260 } 1261 fuse_filehandle_close(vp, fufh, td, NULL); 1262 } 1263 1264 if ((fvdat->flag & FN_REVOKED) != 0) 1265 vrecycle(vp); 1266 1267 return 0; 1268 } 1269 1270 /* 1271 struct vnop_ioctl_args { 1272 struct vnode *a_vp; 1273 u_long a_command; 1274 caddr_t a_data; 1275 int a_fflag; 1276 struct ucred *a_cred; 1277 struct thread *a_td; 1278 }; 1279 */ 1280 static int 1281 fuse_vnop_ioctl(struct vop_ioctl_args *ap) 1282 { 1283 struct vnode *vp = ap->a_vp; 1284 struct mount *mp = vnode_mount(vp); 1285 struct ucred *cred = ap->a_cred; 1286 off_t *offp; 1287 pid_t pid = ap->a_td->td_proc->p_pid; 1288 int err; 1289 1290 switch (ap->a_command) { 1291 case FIOSEEKDATA: 1292 case FIOSEEKHOLE: 1293 /* Call FUSE_LSEEK, if we can, or fall back to vop_stdioctl */ 1294 if (fsess_maybe_impl(mp, FUSE_LSEEK)) { 1295 int whence; 1296 1297 offp = ap->a_data; 1298 if (ap->a_command == FIOSEEKDATA) 1299 whence = SEEK_DATA; 1300 else 1301 whence = SEEK_HOLE; 1302 1303 vn_lock(vp, LK_SHARED | LK_RETRY); 1304 err = fuse_vnop_do_lseek(vp, ap->a_td, cred, pid, offp, 1305 whence); 1306 VOP_UNLOCK(vp); 1307 } 1308 if (fsess_not_impl(mp, FUSE_LSEEK)) 1309 err = vop_stdioctl(ap); 1310 break; 1311 default: 1312 /* TODO: implement FUSE_IOCTL */ 1313 err = ENOTTY; 1314 break; 1315 } 1316 return (err); 1317 } 1318 1319 1320 /* 1321 struct vnop_link_args { 1322 struct vnode *a_tdvp; 1323 struct vnode *a_vp; 1324 struct componentname *a_cnp; 1325 }; 1326 */ 1327 static int 1328 fuse_vnop_link(struct vop_link_args *ap) 1329 { 1330 struct vnode *vp = ap->a_vp; 1331 struct vnode *tdvp = ap->a_tdvp; 1332 struct componentname *cnp = ap->a_cnp; 1333 1334 struct vattr *vap = VTOVA(vp); 1335 1336 struct fuse_dispatcher fdi; 1337 struct fuse_entry_out *feo; 1338 struct fuse_link_in fli; 1339 1340 int err; 1341 1342 if (fuse_isdeadfs(vp)) { 1343 return ENXIO; 1344 } 1345 if (vnode_mount(tdvp) != vnode_mount(vp)) { 1346 return EXDEV; 1347 } 1348 1349 /* 1350 * This is a seatbelt check to protect naive userspace filesystems from 1351 * themselves and the limitations of the FUSE IPC protocol. If a 1352 * filesystem does not allow attribute caching, assume it is capable of 1353 * validating that nlink does not overflow. 1354 */ 1355 if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX) 1356 return EMLINK; 1357 fli.oldnodeid = VTOI(vp); 1358 1359 fdisp_init(&fdi, 0); 1360 fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp, 1361 FUSE_LINK, &fli, sizeof(fli), &fdi); 1362 if ((err = fdisp_wait_answ(&fdi))) { 1363 goto out; 1364 } 1365 feo = fdi.answ; 1366 1367 if (fli.oldnodeid != feo->nodeid) { 1368 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp)); 1369 fuse_warn(data, FSESS_WARN_ILLEGAL_INODE, 1370 "Assigned wrong inode for a hard link."); 1371 fuse_vnode_clear_attr_cache(vp); 1372 fuse_vnode_clear_attr_cache(tdvp); 1373 err = EIO; 1374 goto out; 1375 } 1376 1377 err = fuse_internal_checkentry(feo, vnode_vtype(vp)); 1378 if (!err) { 1379 /* 1380 * Purge the parent's attribute cache because the daemon 1381 * should've updated its mtime and ctime 1382 */ 1383 fuse_vnode_clear_attr_cache(tdvp); 1384 fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid, 1385 feo->attr_valid_nsec, NULL, true); 1386 } 1387 out: 1388 fdisp_destroy(&fdi); 1389 return err; 1390 } 1391 1392 struct fuse_lookup_alloc_arg { 1393 struct fuse_entry_out *feo; 1394 struct componentname *cnp; 1395 uint64_t nid; 1396 __enum_uint8(vtype) vtyp; 1397 }; 1398 1399 /* Callback for vn_get_ino */ 1400 static int 1401 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp) 1402 { 1403 struct fuse_lookup_alloc_arg *flaa = arg; 1404 1405 return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp, 1406 flaa->vtyp); 1407 } 1408 1409 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup, 1410 "int", "struct timespec*", "struct timespec*"); 1411 /* 1412 struct vnop_lookup_args { 1413 struct vnodeop_desc *a_desc; 1414 struct vnode *a_dvp; 1415 struct vnode **a_vpp; 1416 struct componentname *a_cnp; 1417 }; 1418 */ 1419 int 1420 fuse_vnop_lookup(struct vop_lookup_args *ap) 1421 { 1422 struct vnode *dvp = ap->a_dvp; 1423 struct vnode **vpp = ap->a_vpp; 1424 struct componentname *cnp = ap->a_cnp; 1425 struct thread *td = curthread; 1426 struct ucred *cred = cnp->cn_cred; 1427 struct timespec now; 1428 1429 int nameiop = cnp->cn_nameiop; 1430 int flags = cnp->cn_flags; 1431 int islastcn = flags & ISLASTCN; 1432 struct mount *mp = vnode_mount(dvp); 1433 struct fuse_data *data = fuse_get_mpdata(mp); 1434 int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS; 1435 bool is_dot; 1436 1437 int err = 0; 1438 int lookup_err = 0; 1439 struct vnode *vp = NULL; 1440 1441 struct fuse_dispatcher fdi; 1442 bool did_lookup = false; 1443 struct fuse_entry_out *feo = NULL; 1444 __enum_uint8(vtype) vtyp; /* vnode type of target */ 1445 1446 uint64_t nid; 1447 1448 if (fuse_isdeadfs(dvp)) { 1449 *vpp = NULL; 1450 return ENXIO; 1451 } 1452 if (!vnode_isdir(dvp)) 1453 return ENOTDIR; 1454 1455 if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP)) 1456 return EROFS; 1457 1458 if ((cnp->cn_flags & NOEXECCHECK) != 0) 1459 cnp->cn_flags &= ~NOEXECCHECK; 1460 else if ((err = fuse_internal_access(dvp, VEXEC, td, cred))) 1461 return err; 1462 1463 is_dot = cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.'; 1464 if ((flags & ISDOTDOT) && !(data->dataflags & FSESS_EXPORT_SUPPORT)) 1465 { 1466 if (!(VTOFUD(dvp)->flag & FN_PARENT_NID)) { 1467 /* 1468 * Since the file system doesn't support ".." lookups, 1469 * we have no way to find this entry. 1470 */ 1471 return ESTALE; 1472 } 1473 nid = VTOFUD(dvp)->parent_nid; 1474 if (nid == 0) 1475 return ENOENT; 1476 /* .. is obviously a directory */ 1477 vtyp = VDIR; 1478 } else if (is_dot) { 1479 nid = VTOI(dvp); 1480 /* . is obviously a directory */ 1481 vtyp = VDIR; 1482 } else { 1483 struct timespec timeout; 1484 int ncpticks; /* here to accommodate for API contract */ 1485 1486 err = cache_lookup(dvp, vpp, cnp, &timeout, &ncpticks); 1487 getnanouptime(&now); 1488 SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now); 1489 switch (err) { 1490 case -1: /* positive match */ 1491 if (timespeccmp(&timeout, &now, >)) { 1492 counter_u64_add(fuse_lookup_cache_hits, 1); 1493 } else { 1494 /* Cache timeout */ 1495 counter_u64_add(fuse_lookup_cache_misses, 1); 1496 bintime_clear( 1497 &VTOFUD(*vpp)->entry_cache_timeout); 1498 cache_purge(*vpp); 1499 if (dvp != *vpp) 1500 vput(*vpp); 1501 else 1502 vrele(*vpp); 1503 *vpp = NULL; 1504 break; 1505 } 1506 return 0; 1507 1508 case 0: /* no match in cache */ 1509 counter_u64_add(fuse_lookup_cache_misses, 1); 1510 break; 1511 1512 case ENOENT: /* negative match */ 1513 if (timespeccmp(&timeout, &now, <=)) { 1514 /* Cache timeout */ 1515 cache_purge_negative(dvp); 1516 break; 1517 } 1518 /* fall through */ 1519 default: 1520 return err; 1521 } 1522 1523 fdisp_init(&fdi, cnp->cn_namelen + 1); 1524 fdisp_make(&fdi, FUSE_LOOKUP, mp, VTOI(dvp), td, cred); 1525 1526 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen); 1527 ((char *)fdi.indata)[cnp->cn_namelen] = '\0'; 1528 lookup_err = fdisp_wait_answ(&fdi); 1529 did_lookup = true; 1530 1531 if (!lookup_err) { 1532 /* lookup call succeeded */ 1533 feo = (struct fuse_entry_out *)fdi.answ; 1534 nid = feo->nodeid; 1535 if (nid == 0) { 1536 /* zero nodeid means ENOENT and cache it */ 1537 struct timespec timeout; 1538 1539 fdi.answ_stat = ENOENT; 1540 lookup_err = ENOENT; 1541 if (cnp->cn_flags & MAKEENTRY) { 1542 fuse_validity_2_timespec(feo, &timeout); 1543 /* Use the same entry_time for .. as for 1544 * the file itself. That doesn't honor 1545 * exactly what the fuse server tells 1546 * us, but to do otherwise would require 1547 * another cache lookup at this point. 1548 */ 1549 struct timespec *dtsp = NULL; 1550 cache_enter_time(dvp, *vpp, cnp, 1551 &timeout, dtsp); 1552 } 1553 } 1554 vtyp = IFTOVT(feo->attr.mode); 1555 } 1556 if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) { 1557 fdisp_destroy(&fdi); 1558 return lookup_err; 1559 } 1560 } 1561 /* lookup_err, if non-zero, must be ENOENT at this point */ 1562 1563 if (lookup_err) { 1564 /* Entry not found */ 1565 if ((nameiop == CREATE || nameiop == RENAME) && islastcn) { 1566 if (default_permissions) 1567 err = fuse_internal_access(dvp, VWRITE, td, 1568 cred); 1569 else 1570 err = 0; 1571 if (!err) { 1572 err = EJUSTRETURN; 1573 } 1574 } else { 1575 err = ENOENT; 1576 } 1577 } else { 1578 /* Entry was found */ 1579 if (flags & ISDOTDOT) { 1580 struct fuse_lookup_alloc_arg flaa; 1581 1582 flaa.nid = nid; 1583 flaa.feo = feo; 1584 flaa.cnp = cnp; 1585 flaa.vtyp = vtyp; 1586 err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0, 1587 &vp); 1588 *vpp = vp; 1589 } else if (nid == VTOI(dvp)) { 1590 if (is_dot) { 1591 vref(dvp); 1592 *vpp = dvp; 1593 } else { 1594 fuse_warn(fuse_get_mpdata(mp), 1595 FSESS_WARN_ILLEGAL_INODE, 1596 "Assigned same inode to both parent and " 1597 "child."); 1598 err = EIO; 1599 } 1600 1601 } else { 1602 struct fuse_vnode_data *fvdat; 1603 1604 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp, 1605 &vp, cnp, vtyp); 1606 if (err) 1607 goto out; 1608 *vpp = vp; 1609 fvdat = VTOFUD(vp); 1610 1611 MPASS(feo != NULL); 1612 if (timespeccmp(&now, &fvdat->last_local_modify, >)) { 1613 /* 1614 * Attributes from the server are definitely 1615 * newer than the last attributes we sent to 1616 * the server, so cache them. 1617 */ 1618 fuse_internal_cache_attrs(*vpp, &feo->attr, 1619 feo->attr_valid, feo->attr_valid_nsec, 1620 NULL, true); 1621 } 1622 fuse_validity_2_bintime(feo->entry_valid, 1623 feo->entry_valid_nsec, 1624 &fvdat->entry_cache_timeout); 1625 1626 if ((nameiop == DELETE || nameiop == RENAME) && 1627 islastcn && default_permissions) 1628 { 1629 struct vattr dvattr; 1630 1631 err = fuse_internal_access(dvp, VWRITE, td, 1632 cred); 1633 if (err != 0) 1634 goto out; 1635 /* 1636 * if the parent's sticky bit is set, check 1637 * whether we're allowed to remove the file. 1638 * Need to figure out the vnode locking to make 1639 * this work. 1640 */ 1641 fuse_internal_getattr(dvp, &dvattr, cred, td); 1642 if ((dvattr.va_mode & S_ISTXT) && 1643 fuse_internal_access(dvp, VADMIN, td, 1644 cred) && 1645 fuse_internal_access(*vpp, VADMIN, td, 1646 cred)) { 1647 err = EPERM; 1648 goto out; 1649 } 1650 } 1651 } 1652 } 1653 out: 1654 if (err) { 1655 if (vp != NULL && dvp != vp) 1656 vput(vp); 1657 else if (vp != NULL) 1658 vrele(vp); 1659 *vpp = NULL; 1660 } 1661 if (did_lookup) 1662 fdisp_destroy(&fdi); 1663 1664 return err; 1665 } 1666 1667 /* 1668 struct vnop_mkdir_args { 1669 struct vnode *a_dvp; 1670 struct vnode **a_vpp; 1671 struct componentname *a_cnp; 1672 struct vattr *a_vap; 1673 }; 1674 */ 1675 static int 1676 fuse_vnop_mkdir(struct vop_mkdir_args *ap) 1677 { 1678 struct vnode *dvp = ap->a_dvp; 1679 struct vnode **vpp = ap->a_vpp; 1680 struct componentname *cnp = ap->a_cnp; 1681 struct vattr *vap = ap->a_vap; 1682 1683 struct fuse_mkdir_in fmdi; 1684 1685 if (fuse_isdeadfs(dvp)) { 1686 return ENXIO; 1687 } 1688 fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode); 1689 fmdi.umask = curthread->td_proc->p_pd->pd_cmask; 1690 1691 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi, 1692 sizeof(fmdi), VDIR)); 1693 } 1694 1695 /* 1696 struct vnop_mknod_args { 1697 struct vnode *a_dvp; 1698 struct vnode **a_vpp; 1699 struct componentname *a_cnp; 1700 struct vattr *a_vap; 1701 }; 1702 */ 1703 static int 1704 fuse_vnop_mknod(struct vop_mknod_args *ap) 1705 { 1706 1707 struct vnode *dvp = ap->a_dvp; 1708 struct vnode **vpp = ap->a_vpp; 1709 struct componentname *cnp = ap->a_cnp; 1710 struct vattr *vap = ap->a_vap; 1711 1712 if (fuse_isdeadfs(dvp)) 1713 return ENXIO; 1714 1715 return fuse_internal_mknod(dvp, vpp, cnp, vap); 1716 } 1717 1718 /* 1719 struct vop_open_args { 1720 struct vnode *a_vp; 1721 int a_mode; 1722 struct ucred *a_cred; 1723 struct thread *a_td; 1724 int a_fdidx; / struct file *a_fp; 1725 }; 1726 */ 1727 static int 1728 fuse_vnop_open(struct vop_open_args *ap) 1729 { 1730 struct vnode *vp = ap->a_vp; 1731 int a_mode = ap->a_mode; 1732 struct thread *td = ap->a_td; 1733 struct ucred *cred = ap->a_cred; 1734 pid_t pid = td->td_proc->p_pid; 1735 1736 if (fuse_isdeadfs(vp)) 1737 return ENXIO; 1738 if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO) 1739 return (EOPNOTSUPP); 1740 if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0) 1741 return EINVAL; 1742 1743 if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) { 1744 fuse_vnode_open(vp, 0, td); 1745 return 0; 1746 } 1747 1748 return fuse_filehandle_open(vp, a_mode, NULL, td, cred); 1749 } 1750 1751 static int 1752 fuse_vnop_pathconf(struct vop_pathconf_args *ap) 1753 { 1754 struct vnode *vp = ap->a_vp; 1755 struct mount *mp; 1756 1757 switch (ap->a_name) { 1758 case _PC_FILESIZEBITS: 1759 *ap->a_retval = 64; 1760 return (0); 1761 case _PC_NAME_MAX: 1762 *ap->a_retval = NAME_MAX; 1763 return (0); 1764 case _PC_LINK_MAX: 1765 *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX); 1766 return (0); 1767 case _PC_SYMLINK_MAX: 1768 *ap->a_retval = MAXPATHLEN; 1769 return (0); 1770 case _PC_NO_TRUNC: 1771 *ap->a_retval = 1; 1772 return (0); 1773 case _PC_MIN_HOLE_SIZE: 1774 /* 1775 * The FUSE protocol provides no mechanism for a server to 1776 * report _PC_MIN_HOLE_SIZE. It's a protocol bug. Instead, 1777 * return EINVAL if the server does not support FUSE_LSEEK, or 1778 * 1 if it does. 1779 */ 1780 mp = vnode_mount(vp); 1781 if (!fsess_is_impl(mp, FUSE_LSEEK) && 1782 !fsess_not_impl(mp, FUSE_LSEEK)) { 1783 off_t offset = 0; 1784 1785 /* Issue a FUSE_LSEEK to find out if it's implemented */ 1786 fuse_vnop_do_lseek(vp, curthread, curthread->td_ucred, 1787 curthread->td_proc->p_pid, &offset, SEEK_DATA); 1788 } 1789 1790 if (fsess_is_impl(mp, FUSE_LSEEK)) { 1791 *ap->a_retval = 1; 1792 return (0); 1793 } else { 1794 /* 1795 * Probably FUSE_LSEEK is not implemented. It might 1796 * be, if the FUSE_LSEEK above returned an error like 1797 * EACCES, but in that case we can't tell, so it's 1798 * safest to report EINVAL anyway. 1799 */ 1800 return (EINVAL); 1801 } 1802 default: 1803 return (vop_stdpathconf(ap)); 1804 } 1805 } 1806 1807 SDT_PROBE_DEFINE3(fusefs, , vnops, filehandles_closed, "struct vnode*", 1808 "struct uio*", "struct ucred*"); 1809 /* 1810 struct vnop_read_args { 1811 struct vnode *a_vp; 1812 struct uio *a_uio; 1813 int a_ioflag; 1814 struct ucred *a_cred; 1815 }; 1816 */ 1817 static int 1818 fuse_vnop_read(struct vop_read_args *ap) 1819 { 1820 struct vnode *vp = ap->a_vp; 1821 struct uio *uio = ap->a_uio; 1822 int ioflag = ap->a_ioflag; 1823 struct ucred *cred = ap->a_cred; 1824 pid_t pid = curthread->td_proc->p_pid; 1825 struct fuse_filehandle *fufh; 1826 int err; 1827 bool closefufh = false, directio; 1828 1829 MPASS(vp->v_type == VREG || vp->v_type == VDIR); 1830 1831 if (fuse_isdeadfs(vp)) { 1832 return ENXIO; 1833 } 1834 1835 if (VTOFUD(vp)->flag & FN_DIRECTIO) { 1836 ioflag |= IO_DIRECT; 1837 } 1838 1839 err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid); 1840 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) { 1841 /* 1842 * nfsd will do I/O without first doing VOP_OPEN. We 1843 * must implicitly open the file here 1844 */ 1845 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred); 1846 closefufh = true; 1847 } 1848 if (err) { 1849 SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred); 1850 return err; 1851 } 1852 1853 /* 1854 * Ideally, when the daemon asks for direct io at open time, the 1855 * standard file flag should be set according to this, so that would 1856 * just change the default mode, which later on could be changed via 1857 * fcntl(2). 1858 * But this doesn't work, the O_DIRECT flag gets cleared at some point 1859 * (don't know where). So to make any use of the Fuse direct_io option, 1860 * we hardwire it into the file's private data (similarly to Linux, 1861 * btw.). 1862 */ 1863 directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp)); 1864 1865 fuse_vnode_update(vp, FN_ATIMECHANGE); 1866 if (directio) { 1867 SDT_PROBE2(fusefs, , vnops, trace, 1, "direct read of vnode"); 1868 err = fuse_read_directbackend(vp, uio, cred, fufh); 1869 } else { 1870 SDT_PROBE2(fusefs, , vnops, trace, 1, "buffered read of vnode"); 1871 err = fuse_read_biobackend(vp, uio, ioflag, cred, fufh, pid); 1872 } 1873 1874 if (closefufh) 1875 fuse_filehandle_close(vp, fufh, curthread, cred); 1876 1877 return (err); 1878 } 1879 1880 /* 1881 struct vnop_readdir_args { 1882 struct vnode *a_vp; 1883 struct uio *a_uio; 1884 struct ucred *a_cred; 1885 int *a_eofflag; 1886 int *a_ncookies; 1887 uint64_t **a_cookies; 1888 }; 1889 */ 1890 static int 1891 fuse_vnop_readdir(struct vop_readdir_args *ap) 1892 { 1893 struct vnode *vp = ap->a_vp; 1894 struct uio *uio = ap->a_uio; 1895 struct ucred *cred = ap->a_cred; 1896 struct fuse_filehandle *fufh = NULL; 1897 struct mount *mp = vnode_mount(vp); 1898 struct fuse_iov cookediov; 1899 int err = 0; 1900 uint64_t *cookies; 1901 ssize_t tresid; 1902 int ncookies; 1903 bool closefufh = false; 1904 pid_t pid = curthread->td_proc->p_pid; 1905 1906 if (ap->a_eofflag) 1907 *ap->a_eofflag = 0; 1908 if (fuse_isdeadfs(vp)) { 1909 return ENXIO; 1910 } 1911 if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */ 1912 (uio_resid(uio) < sizeof(struct dirent))) { 1913 return EINVAL; 1914 } 1915 1916 tresid = uio->uio_resid; 1917 err = fuse_filehandle_get_dir(vp, &fufh, cred, pid); 1918 if (err == EBADF && mp->mnt_flag & MNT_EXPORTED) { 1919 KASSERT(fuse_get_mpdata(mp)->dataflags 1920 & FSESS_NO_OPENDIR_SUPPORT, 1921 ("FUSE file systems that don't set " 1922 "FUSE_NO_OPENDIR_SUPPORT should not be exported")); 1923 /* 1924 * nfsd will do VOP_READDIR without first doing VOP_OPEN. We 1925 * must implicitly open the directory here. 1926 */ 1927 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred); 1928 closefufh = true; 1929 } 1930 if (err) 1931 return (err); 1932 if (ap->a_ncookies != NULL) { 1933 ncookies = uio->uio_resid / 1934 (offsetof(struct dirent, d_name) + 4) + 1; 1935 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK); 1936 *ap->a_ncookies = ncookies; 1937 *ap->a_cookies = cookies; 1938 } else { 1939 ncookies = 0; 1940 cookies = NULL; 1941 } 1942 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1) 1943 fiov_init(&cookediov, DIRCOOKEDSIZE); 1944 1945 err = fuse_internal_readdir(vp, uio, fufh, &cookediov, 1946 &ncookies, cookies); 1947 1948 fiov_teardown(&cookediov); 1949 if (closefufh) 1950 fuse_filehandle_close(vp, fufh, curthread, cred); 1951 1952 if (ap->a_ncookies != NULL) { 1953 if (err == 0) { 1954 *ap->a_ncookies -= ncookies; 1955 } else { 1956 free(*ap->a_cookies, M_TEMP); 1957 *ap->a_ncookies = 0; 1958 *ap->a_cookies = NULL; 1959 } 1960 } 1961 if (err == 0 && tresid == uio->uio_resid) 1962 *ap->a_eofflag = 1; 1963 1964 return err; 1965 } 1966 1967 /* 1968 struct vnop_readlink_args { 1969 struct vnode *a_vp; 1970 struct uio *a_uio; 1971 struct ucred *a_cred; 1972 }; 1973 */ 1974 static int 1975 fuse_vnop_readlink(struct vop_readlink_args *ap) 1976 { 1977 struct vnode *vp = ap->a_vp; 1978 struct uio *uio = ap->a_uio; 1979 struct ucred *cred = ap->a_cred; 1980 1981 struct fuse_dispatcher fdi; 1982 int err; 1983 1984 if (fuse_isdeadfs(vp)) { 1985 return ENXIO; 1986 } 1987 if (!vnode_islnk(vp)) { 1988 return EINVAL; 1989 } 1990 fdisp_init(&fdi, 0); 1991 err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred); 1992 if (err) { 1993 goto out; 1994 } 1995 if (strnlen(fdi.answ, fdi.iosize) + 1 < fdi.iosize) { 1996 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp)); 1997 fuse_warn(data, FSESS_WARN_READLINK_EMBEDDED_NUL, 1998 "Returned an embedded NUL from FUSE_READLINK."); 1999 err = EIO; 2000 goto out; 2001 } 2002 if (((char *)fdi.answ)[0] == '/' && 2003 fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) { 2004 char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname; 2005 2006 err = uiomove(mpth, strlen(mpth), uio); 2007 } 2008 if (!err) { 2009 err = uiomove(fdi.answ, fdi.iosize, uio); 2010 } 2011 out: 2012 fdisp_destroy(&fdi); 2013 return err; 2014 } 2015 2016 /* 2017 struct vnop_reclaim_args { 2018 struct vnode *a_vp; 2019 }; 2020 */ 2021 static int 2022 fuse_vnop_reclaim(struct vop_reclaim_args *ap) 2023 { 2024 struct vnode *vp = ap->a_vp; 2025 struct thread *td = curthread; 2026 struct fuse_vnode_data *fvdat = VTOFUD(vp); 2027 struct fuse_filehandle *fufh, *fufh_tmp; 2028 2029 if (!fvdat) { 2030 panic("FUSE: no vnode data during recycling"); 2031 } 2032 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) { 2033 printf("FUSE: vnode being reclaimed with open fufh " 2034 "(type=%#x)", fufh->fufh_type); 2035 fuse_filehandle_close(vp, fufh, td, NULL); 2036 } 2037 2038 if (VTOI(vp) == 1) { 2039 /* 2040 * Don't send FUSE_FORGET for the root inode, because 2041 * we never send FUSE_LOOKUP for it (see 2042 * fuse_vfsop_root) and we don't want the server to see 2043 * mismatched lookup counts. 2044 */ 2045 struct fuse_data *data; 2046 struct vnode *vroot; 2047 2048 data = fuse_get_mpdata(vnode_mount(vp)); 2049 FUSE_LOCK(); 2050 vroot = data->vroot; 2051 data->vroot = NULL; 2052 FUSE_UNLOCK(); 2053 if (vroot) 2054 vrele(vroot); 2055 } else if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) { 2056 fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp), 2057 fvdat->nlookup); 2058 } 2059 cache_purge(vp); 2060 vfs_hash_remove(vp); 2061 fuse_vnode_destroy(vp); 2062 2063 return 0; 2064 } 2065 2066 /* 2067 struct vnop_remove_args { 2068 struct vnode *a_dvp; 2069 struct vnode *a_vp; 2070 struct componentname *a_cnp; 2071 }; 2072 */ 2073 static int 2074 fuse_vnop_remove(struct vop_remove_args *ap) 2075 { 2076 struct vnode *dvp = ap->a_dvp; 2077 struct vnode *vp = ap->a_vp; 2078 struct componentname *cnp = ap->a_cnp; 2079 2080 int err; 2081 2082 if (fuse_isdeadfs(vp)) { 2083 return ENXIO; 2084 } 2085 if (vnode_isdir(vp)) { 2086 return EPERM; 2087 } 2088 2089 err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK); 2090 2091 return err; 2092 } 2093 2094 /* 2095 struct vnop_rename_args { 2096 struct vnode *a_fdvp; 2097 struct vnode *a_fvp; 2098 struct componentname *a_fcnp; 2099 struct vnode *a_tdvp; 2100 struct vnode *a_tvp; 2101 struct componentname *a_tcnp; 2102 }; 2103 */ 2104 static int 2105 fuse_vnop_rename(struct vop_rename_args *ap) 2106 { 2107 struct vnode *fdvp = ap->a_fdvp; 2108 struct vnode *fvp = ap->a_fvp; 2109 struct componentname *fcnp = ap->a_fcnp; 2110 struct vnode *tdvp = ap->a_tdvp; 2111 struct vnode *tvp = ap->a_tvp; 2112 struct componentname *tcnp = ap->a_tcnp; 2113 struct fuse_data *data; 2114 bool newparent = fdvp != tdvp; 2115 bool isdir = fvp->v_type == VDIR; 2116 int err = 0; 2117 2118 if (fuse_isdeadfs(fdvp)) { 2119 return ENXIO; 2120 } 2121 if (fvp->v_mount != tdvp->v_mount || 2122 (tvp && fvp->v_mount != tvp->v_mount)) { 2123 SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename"); 2124 err = EXDEV; 2125 goto out; 2126 } 2127 cache_purge(fvp); 2128 2129 /* 2130 * FUSE library is expected to check if target directory is not 2131 * under the source directory in the file system tree. 2132 * Linux performs this check at VFS level. 2133 */ 2134 /* 2135 * If source is a directory, and it will get a new parent, user must 2136 * have write permission to it, so ".." can be modified. 2137 */ 2138 data = fuse_get_mpdata(vnode_mount(tdvp)); 2139 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) { 2140 err = fuse_internal_access(fvp, VWRITE, 2141 curthread, tcnp->cn_cred); 2142 if (err) 2143 goto out; 2144 } 2145 sx_xlock(&data->rename_lock); 2146 err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp); 2147 if (err == 0) { 2148 if (tdvp != fdvp) 2149 fuse_vnode_setparent(fvp, tdvp); 2150 if (tvp != NULL) 2151 fuse_vnode_setparent(tvp, NULL); 2152 } 2153 sx_unlock(&data->rename_lock); 2154 2155 if (tvp != NULL && tvp != fvp) { 2156 cache_purge(tvp); 2157 } 2158 if (vnode_isdir(fvp)) { 2159 if (((tvp != NULL) && vnode_isdir(tvp)) || vnode_isdir(fvp)) { 2160 cache_purge(tdvp); 2161 } 2162 cache_purge(fdvp); 2163 } 2164 out: 2165 if (tdvp == tvp) { 2166 vrele(tdvp); 2167 } else { 2168 vput(tdvp); 2169 } 2170 if (tvp != NULL) { 2171 vput(tvp); 2172 } 2173 vrele(fdvp); 2174 vrele(fvp); 2175 2176 return err; 2177 } 2178 2179 /* 2180 struct vnop_rmdir_args { 2181 struct vnode *a_dvp; 2182 struct vnode *a_vp; 2183 struct componentname *a_cnp; 2184 } *ap; 2185 */ 2186 static int 2187 fuse_vnop_rmdir(struct vop_rmdir_args *ap) 2188 { 2189 struct vnode *dvp = ap->a_dvp; 2190 struct vnode *vp = ap->a_vp; 2191 2192 int err; 2193 2194 if (fuse_isdeadfs(vp)) { 2195 return ENXIO; 2196 } 2197 if (VTOFUD(vp) == VTOFUD(dvp)) { 2198 return EINVAL; 2199 } 2200 err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR); 2201 2202 return err; 2203 } 2204 2205 /* 2206 struct vnop_setattr_args { 2207 struct vnode *a_vp; 2208 struct vattr *a_vap; 2209 struct ucred *a_cred; 2210 struct thread *a_td; 2211 }; 2212 */ 2213 static int 2214 fuse_vnop_setattr(struct vop_setattr_args *ap) 2215 { 2216 struct vnode *vp = ap->a_vp; 2217 struct vattr *vap = ap->a_vap; 2218 struct ucred *cred = ap->a_cred; 2219 struct thread *td = curthread; 2220 struct mount *mp; 2221 struct fuse_data *data; 2222 struct vattr old_va; 2223 int dataflags; 2224 int err = 0, err2; 2225 accmode_t accmode = 0; 2226 bool checkperm; 2227 bool drop_suid = false; 2228 2229 mp = vnode_mount(vp); 2230 data = fuse_get_mpdata(mp); 2231 dataflags = data->dataflags; 2232 checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS; 2233 2234 if (fuse_isdeadfs(vp)) { 2235 return ENXIO; 2236 } 2237 2238 if (vap->va_uid != (uid_t)VNOVAL) { 2239 if (checkperm) { 2240 /* Only root may change a file's owner */ 2241 err = priv_check_cred(cred, PRIV_VFS_CHOWN); 2242 if (err) { 2243 /* As a special case, allow the null chown */ 2244 err2 = fuse_internal_getattr(vp, &old_va, cred, 2245 td); 2246 if (err2) 2247 return (err2); 2248 if (vap->va_uid != old_va.va_uid) 2249 return err; 2250 else 2251 accmode |= VADMIN; 2252 drop_suid = true; 2253 } else 2254 accmode |= VADMIN; 2255 } else 2256 accmode |= VADMIN; 2257 } 2258 if (vap->va_gid != (gid_t)VNOVAL) { 2259 if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN)) 2260 drop_suid = true; 2261 if (checkperm && !groupmember(vap->va_gid, cred)) 2262 { 2263 /* 2264 * Non-root users may only chgrp to one of their own 2265 * groups 2266 */ 2267 err = priv_check_cred(cred, PRIV_VFS_CHOWN); 2268 if (err) { 2269 /* As a special case, allow the null chgrp */ 2270 err2 = fuse_internal_getattr(vp, &old_va, cred, 2271 td); 2272 if (err2) 2273 return (err2); 2274 if (vap->va_gid != old_va.va_gid) 2275 return err; 2276 accmode |= VADMIN; 2277 } else 2278 accmode |= VADMIN; 2279 } else 2280 accmode |= VADMIN; 2281 } 2282 if (vap->va_size != VNOVAL) { 2283 switch (vp->v_type) { 2284 case VDIR: 2285 return (EISDIR); 2286 case VLNK: 2287 case VREG: 2288 if (vfs_isrdonly(mp)) 2289 return (EROFS); 2290 err = vn_rlimit_trunc(vap->va_size, td); 2291 if (err) 2292 return (err); 2293 break; 2294 default: 2295 /* 2296 * According to POSIX, the result is unspecified 2297 * for file types other than regular files, 2298 * directories and shared memory objects. We 2299 * don't support shared memory objects in the file 2300 * system, and have dubious support for truncating 2301 * symlinks. Just ignore the request in other cases. 2302 */ 2303 return (0); 2304 } 2305 /* Don't set accmode. Permission to trunc is checked upstack */ 2306 } 2307 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 2308 if (vap->va_vaflags & VA_UTIMES_NULL) 2309 accmode |= VWRITE; 2310 else 2311 accmode |= VADMIN; 2312 } 2313 if (drop_suid) { 2314 if (vap->va_mode != (mode_t)VNOVAL) 2315 vap->va_mode &= ~(S_ISUID | S_ISGID); 2316 else { 2317 err = fuse_internal_getattr(vp, &old_va, cred, td); 2318 if (err) 2319 return (err); 2320 vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID); 2321 } 2322 } 2323 if (vap->va_mode != (mode_t)VNOVAL) { 2324 /* Only root may set the sticky bit on non-directories */ 2325 if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT) 2326 && priv_check_cred(cred, PRIV_VFS_STICKYFILE)) 2327 return EFTYPE; 2328 if (checkperm && (vap->va_mode & S_ISGID)) { 2329 err = fuse_internal_getattr(vp, &old_va, cred, td); 2330 if (err) 2331 return (err); 2332 if (!groupmember(old_va.va_gid, cred)) { 2333 err = priv_check_cred(cred, PRIV_VFS_SETGID); 2334 if (err) 2335 return (err); 2336 } 2337 } 2338 accmode |= VADMIN; 2339 } 2340 2341 if (vfs_isrdonly(mp)) 2342 return EROFS; 2343 2344 if (checkperm) { 2345 err = fuse_internal_access(vp, accmode, td, cred); 2346 } else { 2347 err = 0; 2348 } 2349 if (err) 2350 return err; 2351 else 2352 return fuse_internal_setattr(vp, vap, td, cred); 2353 } 2354 2355 /* 2356 struct vnop_strategy_args { 2357 struct vnode *a_vp; 2358 struct buf *a_bp; 2359 }; 2360 */ 2361 static int 2362 fuse_vnop_strategy(struct vop_strategy_args *ap) 2363 { 2364 struct vnode *vp = ap->a_vp; 2365 struct buf *bp = ap->a_bp; 2366 2367 if (!vp || fuse_isdeadfs(vp)) { 2368 bp->b_ioflags |= BIO_ERROR; 2369 bp->b_error = ENXIO; 2370 bufdone(bp); 2371 return 0; 2372 } 2373 2374 /* 2375 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags. 2376 * fuse_io_strategy sets bp's error fields 2377 */ 2378 (void)fuse_io_strategy(vp, bp); 2379 2380 return 0; 2381 } 2382 2383 /* 2384 struct vnop_symlink_args { 2385 struct vnode *a_dvp; 2386 struct vnode **a_vpp; 2387 struct componentname *a_cnp; 2388 struct vattr *a_vap; 2389 char *a_target; 2390 }; 2391 */ 2392 static int 2393 fuse_vnop_symlink(struct vop_symlink_args *ap) 2394 { 2395 struct vnode *dvp = ap->a_dvp; 2396 struct vnode **vpp = ap->a_vpp; 2397 struct componentname *cnp = ap->a_cnp; 2398 const char *target = ap->a_target; 2399 2400 struct fuse_dispatcher fdi; 2401 2402 int err; 2403 size_t len; 2404 2405 if (fuse_isdeadfs(dvp)) { 2406 return ENXIO; 2407 } 2408 /* 2409 * Unlike the other creator type calls, here we have to create a message 2410 * where the name of the new entry comes first, and the data describing 2411 * the entry comes second. 2412 * Hence we can't rely on our handy fuse_internal_newentry() routine, 2413 * but put together the message manually and just call the core part. 2414 */ 2415 2416 len = strlen(target) + 1; 2417 fdisp_init(&fdi, len + cnp->cn_namelen + 1); 2418 fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL); 2419 2420 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen); 2421 ((char *)fdi.indata)[cnp->cn_namelen] = '\0'; 2422 memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len); 2423 2424 err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi); 2425 fdisp_destroy(&fdi); 2426 return err; 2427 } 2428 2429 /* 2430 struct vnop_write_args { 2431 struct vnode *a_vp; 2432 struct uio *a_uio; 2433 int a_ioflag; 2434 struct ucred *a_cred; 2435 }; 2436 */ 2437 static int 2438 fuse_vnop_write(struct vop_write_args *ap) 2439 { 2440 struct vnode *vp = ap->a_vp; 2441 struct uio *uio = ap->a_uio; 2442 int ioflag = ap->a_ioflag; 2443 struct ucred *cred = ap->a_cred; 2444 pid_t pid = curthread->td_proc->p_pid; 2445 struct fuse_filehandle *fufh; 2446 int err; 2447 bool closefufh = false, directio; 2448 2449 MPASS(vp->v_type == VREG || vp->v_type == VDIR); 2450 2451 if (fuse_isdeadfs(vp)) { 2452 return ENXIO; 2453 } 2454 2455 if (VTOFUD(vp)->flag & FN_DIRECTIO) { 2456 ioflag |= IO_DIRECT; 2457 } 2458 2459 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); 2460 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) { 2461 /* 2462 * nfsd will do I/O without first doing VOP_OPEN. We 2463 * must implicitly open the file here 2464 */ 2465 err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred); 2466 closefufh = true; 2467 } 2468 if (err) { 2469 SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred); 2470 return err; 2471 } 2472 2473 /* 2474 * Ideally, when the daemon asks for direct io at open time, the 2475 * standard file flag should be set according to this, so that would 2476 * just change the default mode, which later on could be changed via 2477 * fcntl(2). 2478 * But this doesn't work, the O_DIRECT flag gets cleared at some point 2479 * (don't know where). So to make any use of the Fuse direct_io option, 2480 * we hardwire it into the file's private data (similarly to Linux, 2481 * btw.). 2482 */ 2483 directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp)); 2484 2485 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE); 2486 if (directio) { 2487 off_t start, end, filesize; 2488 bool pages = (ioflag & IO_VMIO) != 0; 2489 2490 SDT_PROBE2(fusefs, , vnops, trace, 1, "direct write of vnode"); 2491 2492 err = fuse_vnode_size(vp, &filesize, cred, curthread); 2493 if (err) 2494 goto out; 2495 2496 start = uio->uio_offset; 2497 end = start + uio->uio_resid; 2498 if (!pages) { 2499 err = fuse_inval_buf_range(vp, filesize, start, 2500 end); 2501 if (err) 2502 goto out; 2503 } 2504 err = fuse_write_directbackend(vp, uio, cred, fufh, 2505 filesize, ioflag, pages); 2506 } else { 2507 SDT_PROBE2(fusefs, , vnops, trace, 1, 2508 "buffered write of vnode"); 2509 if (!fsess_opt_writeback(vnode_mount(vp))) 2510 ioflag |= IO_SYNC; 2511 err = fuse_write_biobackend(vp, uio, cred, fufh, ioflag, pid); 2512 } 2513 fuse_internal_clear_suid_on_write(vp, cred, uio->uio_td); 2514 2515 out: 2516 if (closefufh) 2517 fuse_filehandle_close(vp, fufh, curthread, cred); 2518 2519 return (err); 2520 } 2521 2522 static daddr_t 2523 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off) 2524 { 2525 const int biosize = fuse_iosize(vp); 2526 2527 return (off / biosize); 2528 } 2529 2530 static int 2531 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *blksz) 2532 { 2533 off_t filesize; 2534 int err; 2535 const int biosize = fuse_iosize(vp); 2536 2537 err = fuse_vnode_size(vp, &filesize, NULL, NULL); 2538 if (err) { 2539 /* This will turn into a SIGBUS */ 2540 return (EIO); 2541 } else if ((off_t)lbn * biosize >= filesize) { 2542 *blksz = 0; 2543 } else if ((off_t)(lbn + 1) * biosize > filesize) { 2544 *blksz = filesize - (off_t)lbn *biosize; 2545 } else { 2546 *blksz = biosize; 2547 } 2548 return (0); 2549 } 2550 2551 /* 2552 struct vnop_getpages_args { 2553 struct vnode *a_vp; 2554 vm_page_t *a_m; 2555 int a_count; 2556 int a_reqpage; 2557 }; 2558 */ 2559 static int 2560 fuse_vnop_getpages(struct vop_getpages_args *ap) 2561 { 2562 struct vnode *vp = ap->a_vp; 2563 2564 if (!fsess_opt_mmap(vnode_mount(vp))) { 2565 SDT_PROBE2(fusefs, , vnops, trace, 1, 2566 "called on non-cacheable vnode??\n"); 2567 return (VM_PAGER_ERROR); 2568 } 2569 2570 return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind, 2571 ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz)); 2572 } 2573 2574 static const char extattr_namespace_separator = '.'; 2575 2576 /* 2577 struct vop_getextattr_args { 2578 struct vop_generic_args a_gen; 2579 struct vnode *a_vp; 2580 int a_attrnamespace; 2581 const char *a_name; 2582 struct uio *a_uio; 2583 size_t *a_size; 2584 struct ucred *a_cred; 2585 struct thread *a_td; 2586 }; 2587 */ 2588 static int 2589 fuse_vnop_getextattr(struct vop_getextattr_args *ap) 2590 { 2591 struct vnode *vp = ap->a_vp; 2592 struct uio *uio = ap->a_uio; 2593 struct fuse_dispatcher fdi; 2594 struct fuse_getxattr_in *get_xattr_in; 2595 struct fuse_getxattr_out *get_xattr_out; 2596 struct mount *mp = vnode_mount(vp); 2597 struct thread *td = ap->a_td; 2598 struct ucred *cred = ap->a_cred; 2599 char *prefix; 2600 char *attr_str; 2601 size_t len; 2602 int err; 2603 2604 if (fuse_isdeadfs(vp)) 2605 return (ENXIO); 2606 2607 if (fsess_not_impl(mp, FUSE_GETXATTR)) 2608 return EOPNOTSUPP; 2609 2610 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD); 2611 if (err) 2612 return err; 2613 2614 /* Default to looking for user attributes. */ 2615 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) 2616 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; 2617 else 2618 prefix = EXTATTR_NAMESPACE_USER_STRING; 2619 2620 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 2621 strlen(ap->a_name) + 1; 2622 2623 fdisp_init(&fdi, len + sizeof(*get_xattr_in)); 2624 fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred); 2625 2626 get_xattr_in = fdi.indata; 2627 /* 2628 * Check to see whether we're querying the available size or 2629 * issuing the actual request. If we pass in 0, we get back struct 2630 * fuse_getxattr_out. If we pass in a non-zero size, we get back 2631 * that much data, without the struct fuse_getxattr_out header. 2632 */ 2633 if (uio == NULL) 2634 get_xattr_in->size = 0; 2635 else 2636 get_xattr_in->size = uio->uio_resid; 2637 2638 attr_str = (char *)fdi.indata + sizeof(*get_xattr_in); 2639 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator, 2640 ap->a_name); 2641 2642 err = fdisp_wait_answ(&fdi); 2643 if (err != 0) { 2644 if (err == ENOSYS) { 2645 fsess_set_notimpl(mp, FUSE_GETXATTR); 2646 err = EOPNOTSUPP; 2647 } 2648 goto out; 2649 } 2650 2651 get_xattr_out = fdi.answ; 2652 2653 if (ap->a_size != NULL) 2654 *ap->a_size = get_xattr_out->size; 2655 2656 if (uio != NULL) 2657 err = uiomove(fdi.answ, fdi.iosize, uio); 2658 2659 out: 2660 fdisp_destroy(&fdi); 2661 return (err); 2662 } 2663 2664 /* 2665 struct vop_setextattr_args { 2666 struct vop_generic_args a_gen; 2667 struct vnode *a_vp; 2668 int a_attrnamespace; 2669 const char *a_name; 2670 struct uio *a_uio; 2671 struct ucred *a_cred; 2672 struct thread *a_td; 2673 }; 2674 */ 2675 static int 2676 fuse_vnop_setextattr(struct vop_setextattr_args *ap) 2677 { 2678 struct vnode *vp = ap->a_vp; 2679 struct uio *uio = ap->a_uio; 2680 struct fuse_dispatcher fdi; 2681 struct fuse_setxattr_in *set_xattr_in; 2682 struct mount *mp = vnode_mount(vp); 2683 struct thread *td = ap->a_td; 2684 struct ucred *cred = ap->a_cred; 2685 char *prefix; 2686 size_t len; 2687 char *attr_str; 2688 int err; 2689 2690 if (fuse_isdeadfs(vp)) 2691 return (ENXIO); 2692 2693 if (fsess_not_impl(mp, FUSE_SETXATTR)) 2694 return EOPNOTSUPP; 2695 2696 if (vfs_isrdonly(mp)) 2697 return EROFS; 2698 2699 /* Deleting xattrs must use VOP_DELETEEXTATTR instead */ 2700 if (ap->a_uio == NULL) { 2701 /* 2702 * If we got here as fallback from VOP_DELETEEXTATTR, then 2703 * return EOPNOTSUPP. 2704 */ 2705 if (fsess_not_impl(mp, FUSE_REMOVEXATTR)) 2706 return (EOPNOTSUPP); 2707 else 2708 return (EINVAL); 2709 } 2710 2711 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, 2712 VWRITE); 2713 if (err) 2714 return err; 2715 2716 /* Default to looking for user attributes. */ 2717 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) 2718 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; 2719 else 2720 prefix = EXTATTR_NAMESPACE_USER_STRING; 2721 2722 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 2723 strlen(ap->a_name) + 1; 2724 2725 fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid); 2726 fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred); 2727 2728 set_xattr_in = fdi.indata; 2729 set_xattr_in->size = uio->uio_resid; 2730 2731 attr_str = (char *)fdi.indata + sizeof(*set_xattr_in); 2732 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator, 2733 ap->a_name); 2734 2735 err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len, 2736 uio->uio_resid, uio); 2737 if (err != 0) { 2738 goto out; 2739 } 2740 2741 err = fdisp_wait_answ(&fdi); 2742 2743 if (err == ENOSYS) { 2744 fsess_set_notimpl(mp, FUSE_SETXATTR); 2745 err = EOPNOTSUPP; 2746 } 2747 if (err == ERESTART) { 2748 /* Can't restart after calling uiomove */ 2749 err = EINTR; 2750 } 2751 2752 out: 2753 fdisp_destroy(&fdi); 2754 return (err); 2755 } 2756 2757 /* 2758 * The Linux / FUSE extended attribute list is simply a collection of 2759 * NUL-terminated strings. The FreeBSD extended attribute list is a single 2760 * byte length followed by a non-NUL terminated string. So, this allows 2761 * conversion of the Linux / FUSE format to the FreeBSD format in place. 2762 * Linux attribute names are reported with the namespace as a prefix (e.g. 2763 * "user.attribute_name"), but in FreeBSD they are reported without the 2764 * namespace prefix (e.g. "attribute_name"). So, we're going from: 2765 * 2766 * user.attr_name1\0user.attr_name2\0 2767 * 2768 * to: 2769 * 2770 * <num>attr_name1<num>attr_name2 2771 * 2772 * Where "<num>" is a single byte number of characters in the attribute name. 2773 * 2774 * Args: 2775 * prefix - exattr namespace prefix string 2776 * list, list_len - input list with namespace prefixes 2777 * bsd_list, bsd_list_len - output list compatible with bsd vfs 2778 */ 2779 static int 2780 fuse_xattrlist_convert(char *prefix, const char *list, int list_len, 2781 char *bsd_list, int *bsd_list_len) 2782 { 2783 int len, pos, dist_to_next, prefix_len; 2784 2785 pos = 0; 2786 *bsd_list_len = 0; 2787 prefix_len = strlen(prefix); 2788 2789 while (pos < list_len && list[pos] != '\0') { 2790 dist_to_next = strlen(&list[pos]) + 1; 2791 if (bcmp(&list[pos], prefix, prefix_len) == 0 && 2792 list[pos + prefix_len] == extattr_namespace_separator) { 2793 len = dist_to_next - 2794 (prefix_len + sizeof(extattr_namespace_separator)) - 1; 2795 if (len >= EXTATTR_MAXNAMELEN) 2796 return (ENAMETOOLONG); 2797 2798 bsd_list[*bsd_list_len] = len; 2799 memcpy(&bsd_list[*bsd_list_len + 1], 2800 &list[pos + prefix_len + 2801 sizeof(extattr_namespace_separator)], len); 2802 2803 *bsd_list_len += len + 1; 2804 } 2805 2806 pos += dist_to_next; 2807 } 2808 2809 return (0); 2810 } 2811 2812 /* 2813 * List extended attributes 2814 * 2815 * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which 2816 * has a number of differences compared to its FreeBSD equivalent, 2817 * extattr_list_file: 2818 * 2819 * - FUSE_LISTXATTR returns all extended attributes across all namespaces, 2820 * whereas listxattr(2) only returns attributes for a single namespace 2821 * - FUSE_LISTXATTR prepends each attribute name with "namespace." 2822 * - If the provided buffer is not large enough to hold the result, 2823 * FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to 2824 * return as many results as will fit. 2825 */ 2826 /* 2827 struct vop_listextattr_args { 2828 struct vop_generic_args a_gen; 2829 struct vnode *a_vp; 2830 int a_attrnamespace; 2831 struct uio *a_uio; 2832 size_t *a_size; 2833 struct ucred *a_cred; 2834 struct thread *a_td; 2835 }; 2836 */ 2837 static int 2838 fuse_vnop_listextattr(struct vop_listextattr_args *ap) 2839 { 2840 struct vnode *vp = ap->a_vp; 2841 struct uio *uio = ap->a_uio; 2842 struct fuse_dispatcher fdi; 2843 struct fuse_listxattr_in *list_xattr_in; 2844 struct fuse_listxattr_out *list_xattr_out; 2845 struct mount *mp = vnode_mount(vp); 2846 struct thread *td = ap->a_td; 2847 struct ucred *cred = ap->a_cred; 2848 char *prefix; 2849 char *bsd_list = NULL; 2850 char *linux_list; 2851 int bsd_list_len; 2852 int linux_list_len; 2853 int err; 2854 2855 if (fuse_isdeadfs(vp)) 2856 return (ENXIO); 2857 2858 if (fsess_not_impl(mp, FUSE_LISTXATTR)) 2859 return EOPNOTSUPP; 2860 2861 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD); 2862 if (err) 2863 return err; 2864 2865 /* 2866 * Add space for a NUL and the period separator if enabled. 2867 * Default to looking for user attributes. 2868 */ 2869 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) 2870 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; 2871 else 2872 prefix = EXTATTR_NAMESPACE_USER_STRING; 2873 2874 fdisp_init(&fdi, sizeof(*list_xattr_in)); 2875 fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred); 2876 2877 /* 2878 * Retrieve Linux / FUSE compatible list size. 2879 */ 2880 list_xattr_in = fdi.indata; 2881 list_xattr_in->size = 0; 2882 2883 err = fdisp_wait_answ(&fdi); 2884 if (err != 0) { 2885 if (err == ENOSYS) { 2886 fsess_set_notimpl(mp, FUSE_LISTXATTR); 2887 err = EOPNOTSUPP; 2888 } 2889 goto out; 2890 } 2891 2892 list_xattr_out = fdi.answ; 2893 linux_list_len = list_xattr_out->size; 2894 if (linux_list_len == 0) { 2895 if (ap->a_size != NULL) 2896 *ap->a_size = linux_list_len; 2897 goto out; 2898 } 2899 2900 /* 2901 * Retrieve Linux / FUSE compatible list values. 2902 */ 2903 fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred); 2904 list_xattr_in = fdi.indata; 2905 list_xattr_in->size = linux_list_len; 2906 2907 err = fdisp_wait_answ(&fdi); 2908 if (err == ERANGE) { 2909 /* 2910 * Race detected. The attribute list must've grown since the 2911 * first FUSE_LISTXATTR call. Start over. Go all the way back 2912 * to userland so we can process signals, if necessary, before 2913 * restarting. 2914 */ 2915 err = ERESTART; 2916 goto out; 2917 } else if (err != 0) 2918 goto out; 2919 2920 linux_list = fdi.answ; 2921 /* FUSE doesn't allow the server to return more data than requested */ 2922 if (fdi.iosize > linux_list_len) { 2923 struct fuse_data *data = fuse_get_mpdata(mp); 2924 2925 fuse_warn(data, FSESS_WARN_LSEXTATTR_LONG, 2926 "server returned " 2927 "more extended attribute data than requested; " 2928 "should've returned ERANGE instead."); 2929 } else { 2930 /* But returning less data is fine */ 2931 linux_list_len = fdi.iosize; 2932 } 2933 2934 /* 2935 * Retrieve the BSD compatible list values. 2936 * The Linux / FUSE attribute list format isn't the same 2937 * as FreeBSD's format. So we need to transform it into 2938 * FreeBSD's format before giving it to the user. 2939 */ 2940 bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK); 2941 err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len, 2942 bsd_list, &bsd_list_len); 2943 if (err != 0) 2944 goto out; 2945 2946 if (ap->a_size != NULL) 2947 *ap->a_size = bsd_list_len; 2948 2949 if (uio != NULL) 2950 err = uiomove(bsd_list, bsd_list_len, uio); 2951 2952 out: 2953 free(bsd_list, M_TEMP); 2954 fdisp_destroy(&fdi); 2955 return (err); 2956 } 2957 2958 /* 2959 struct vop_deallocate_args { 2960 struct vop_generic_args a_gen; 2961 struct vnode *a_vp; 2962 off_t *a_offset; 2963 off_t *a_len; 2964 int a_flags; 2965 int a_ioflag; 2966 struct ucred *a_cred; 2967 }; 2968 */ 2969 static int 2970 fuse_vnop_deallocate(struct vop_deallocate_args *ap) 2971 { 2972 struct vnode *vp = ap->a_vp; 2973 struct mount *mp = vnode_mount(vp); 2974 struct fuse_filehandle *fufh; 2975 struct fuse_dispatcher fdi; 2976 struct fuse_fallocate_in *ffi; 2977 struct ucred *cred = ap->a_cred; 2978 pid_t pid = curthread->td_proc->p_pid; 2979 off_t *len = ap->a_len; 2980 off_t *offset = ap->a_offset; 2981 int ioflag = ap->a_ioflag; 2982 off_t filesize; 2983 int err; 2984 bool closefufh = false; 2985 2986 if (fuse_isdeadfs(vp)) 2987 return (ENXIO); 2988 2989 if (vfs_isrdonly(mp)) 2990 return (EROFS); 2991 2992 if (fsess_not_impl(mp, FUSE_FALLOCATE)) 2993 goto fallback; 2994 2995 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); 2996 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) { 2997 /* 2998 * nfsd will do I/O without first doing VOP_OPEN. We 2999 * must implicitly open the file here 3000 */ 3001 err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred); 3002 closefufh = true; 3003 } 3004 if (err) 3005 return (err); 3006 3007 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE); 3008 3009 err = fuse_vnode_size(vp, &filesize, cred, curthread); 3010 if (err) 3011 goto out; 3012 fuse_inval_buf_range(vp, filesize, *offset, *offset + *len); 3013 3014 fdisp_init(&fdi, sizeof(*ffi)); 3015 fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred); 3016 ffi = fdi.indata; 3017 ffi->fh = fufh->fh_id; 3018 ffi->offset = *offset; 3019 ffi->length = *len; 3020 /* 3021 * FreeBSD's fspacectl is equivalent to Linux's fallocate with 3022 * mode == FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE 3023 */ 3024 ffi->mode = FUSE_FALLOC_FL_PUNCH_HOLE | FUSE_FALLOC_FL_KEEP_SIZE; 3025 err = fdisp_wait_answ(&fdi); 3026 3027 if (err == ENOSYS) { 3028 fdisp_destroy(&fdi); 3029 fsess_set_notimpl(mp, FUSE_FALLOCATE); 3030 goto fallback; 3031 } else if (err == EOPNOTSUPP) { 3032 /* 3033 * The file system server does not support FUSE_FALLOCATE with 3034 * the supplied mode for this particular file. 3035 */ 3036 fdisp_destroy(&fdi); 3037 goto fallback; 3038 } else if (!err) { 3039 /* 3040 * Clip the returned offset to EoF. Do it here rather than 3041 * before FUSE_FALLOCATE just in case the kernel's cached file 3042 * size is out of date. Unfortunately, FUSE does not return 3043 * any information about filesize from that operation. 3044 */ 3045 *offset = MIN(*offset + *len, filesize); 3046 *len = 0; 3047 fuse_vnode_undirty_cached_timestamps(vp, false); 3048 fuse_internal_clear_suid_on_write(vp, cred, curthread); 3049 3050 if (ioflag & IO_SYNC) 3051 err = fuse_internal_fsync(vp, curthread, MNT_WAIT, 3052 false); 3053 } 3054 3055 out: 3056 fdisp_destroy(&fdi); 3057 if (closefufh) 3058 fuse_filehandle_close(vp, fufh, curthread, cred); 3059 3060 return (err); 3061 3062 fallback: 3063 if (closefufh) 3064 fuse_filehandle_close(vp, fufh, curthread, cred); 3065 3066 return (vop_stddeallocate(ap)); 3067 } 3068 3069 /* 3070 struct vop_deleteextattr_args { 3071 struct vop_generic_args a_gen; 3072 struct vnode *a_vp; 3073 int a_attrnamespace; 3074 const char *a_name; 3075 struct ucred *a_cred; 3076 struct thread *a_td; 3077 }; 3078 */ 3079 static int 3080 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap) 3081 { 3082 struct vnode *vp = ap->a_vp; 3083 struct fuse_dispatcher fdi; 3084 struct mount *mp = vnode_mount(vp); 3085 struct thread *td = ap->a_td; 3086 struct ucred *cred = ap->a_cred; 3087 char *prefix; 3088 size_t len; 3089 char *attr_str; 3090 int err; 3091 3092 if (fuse_isdeadfs(vp)) 3093 return (ENXIO); 3094 3095 if (fsess_not_impl(mp, FUSE_REMOVEXATTR)) 3096 return EOPNOTSUPP; 3097 3098 if (vfs_isrdonly(mp)) 3099 return EROFS; 3100 3101 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, 3102 VWRITE); 3103 if (err) 3104 return err; 3105 3106 /* Default to looking for user attributes. */ 3107 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) 3108 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; 3109 else 3110 prefix = EXTATTR_NAMESPACE_USER_STRING; 3111 3112 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 3113 strlen(ap->a_name) + 1; 3114 3115 fdisp_init(&fdi, len); 3116 fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred); 3117 3118 attr_str = fdi.indata; 3119 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator, 3120 ap->a_name); 3121 3122 err = fdisp_wait_answ(&fdi); 3123 if (err == ENOSYS) { 3124 fsess_set_notimpl(mp, FUSE_REMOVEXATTR); 3125 err = EOPNOTSUPP; 3126 } 3127 3128 fdisp_destroy(&fdi); 3129 return (err); 3130 } 3131 3132 /* 3133 struct vnop_print_args { 3134 struct vnode *a_vp; 3135 }; 3136 */ 3137 static int 3138 fuse_vnop_print(struct vop_print_args *ap) 3139 { 3140 struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp); 3141 3142 printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n", 3143 (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid, 3144 (uintmax_t)fvdat->nlookup, 3145 fvdat->flag); 3146 3147 return 0; 3148 } 3149 3150 /* 3151 * Get an NFS filehandle for a FUSE file. 3152 * 3153 * This will only work for FUSE file systems that guarantee the uniqueness of 3154 * nodeid:generation, which most don't. 3155 */ 3156 /* 3157 vop_vptofh { 3158 IN struct vnode *a_vp; 3159 IN struct fid *a_fhp; 3160 }; 3161 */ 3162 static int 3163 fuse_vnop_vptofh(struct vop_vptofh_args *ap) 3164 { 3165 struct vnode *vp = ap->a_vp; 3166 struct fuse_vnode_data *fvdat = VTOFUD(vp); 3167 struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp); 3168 _Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid), 3169 "FUSE fid type is too big"); 3170 struct mount *mp = vnode_mount(vp); 3171 struct fuse_data *data = fuse_get_mpdata(mp); 3172 struct vattr va; 3173 int err; 3174 3175 if (!(data->dataflags & FSESS_EXPORT_SUPPORT)) { 3176 /* NFS requires lookups for "." and ".." */ 3177 SDT_PROBE2(fusefs, , vnops, trace, 1, 3178 "VOP_VPTOFH without FUSE_EXPORT_SUPPORT"); 3179 return EOPNOTSUPP; 3180 } 3181 if ((mp->mnt_flag & MNT_EXPORTED) && 3182 !(data->dataflags & FSESS_NO_OPENDIR_SUPPORT)) 3183 { 3184 /* 3185 * NFS is stateless, so nfsd must reopen a directory on every 3186 * call to VOP_READDIR, passing in the d_off field from the 3187 * final dirent of the previous invocation. But without 3188 * FUSE_NO_OPENDIR_SUPPORT, the FUSE protocol does not 3189 * guarantee that d_off will be valid after a directory is 3190 * closed and reopened. So prohibit exporting FUSE file 3191 * systems that don't set that flag. 3192 * 3193 * But userspace NFS servers don't have this problem. 3194 */ 3195 SDT_PROBE2(fusefs, , vnops, trace, 1, 3196 "VOP_VPTOFH without FUSE_NO_OPENDIR_SUPPORT"); 3197 return EOPNOTSUPP; 3198 } 3199 3200 err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread); 3201 if (err) 3202 return err; 3203 3204 /*ip = VTOI(ap->a_vp);*/ 3205 /*ufhp = (struct ufid *)ap->a_fhp;*/ 3206 fhp->len = sizeof(struct fuse_fid); 3207 fhp->nid = fvdat->nid; 3208 if (fvdat->generation <= UINT32_MAX) 3209 fhp->gen = fvdat->generation; 3210 else 3211 return EOVERFLOW; 3212 return (0); 3213 } 3214