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 == ENXIO) { 399 /* Note: ENXIO means "no more hole/data regions until EOF" */ 400 fsess_set_impl(mp, FUSE_LSEEK); 401 } else if (err == 0) { 402 fsess_set_impl(mp, FUSE_LSEEK); 403 flso = fdi.answ; 404 *offp = flso->offset; 405 } 406 fdisp_destroy(&fdi); 407 408 return (err); 409 } 410 411 /* 412 struct vnop_access_args { 413 struct vnode *a_vp; 414 #if VOP_ACCESS_TAKES_ACCMODE_T 415 accmode_t a_accmode; 416 #else 417 int a_mode; 418 #endif 419 struct ucred *a_cred; 420 struct thread *a_td; 421 }; 422 */ 423 static int 424 fuse_vnop_access(struct vop_access_args *ap) 425 { 426 struct vnode *vp = ap->a_vp; 427 int accmode = ap->a_accmode; 428 struct ucred *cred = ap->a_cred; 429 430 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp)); 431 432 int err; 433 434 if (fuse_isdeadfs(vp)) { 435 if (vnode_isvroot(vp)) { 436 return 0; 437 } 438 return ENXIO; 439 } 440 if (!(data->dataflags & FSESS_INITED)) { 441 if (vnode_isvroot(vp)) { 442 if (priv_check_cred(cred, PRIV_VFS_ADMIN) || 443 (fuse_match_cred(data->daemoncred, cred) == 0)) { 444 return 0; 445 } 446 } 447 return EBADF; 448 } 449 if (vnode_islnk(vp)) { 450 return 0; 451 } 452 453 err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred); 454 return err; 455 } 456 457 /* 458 * struct vop_advlock_args { 459 * struct vop_generic_args a_gen; 460 * struct vnode *a_vp; 461 * void *a_id; 462 * int a_op; 463 * struct flock *a_fl; 464 * int a_flags; 465 * } 466 */ 467 static int 468 fuse_vnop_advlock(struct vop_advlock_args *ap) 469 { 470 struct vnode *vp = ap->a_vp; 471 struct flock *fl = ap->a_fl; 472 struct thread *td = curthread; 473 struct ucred *cred = td->td_ucred; 474 pid_t pid = td->td_proc->p_pid; 475 struct fuse_filehandle *fufh; 476 struct fuse_dispatcher fdi; 477 struct fuse_lk_in *fli; 478 struct fuse_lk_out *flo; 479 struct vattr vattr; 480 enum fuse_opcode op; 481 off_t size, start; 482 int dataflags, err; 483 int flags = ap->a_flags; 484 485 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags; 486 487 if (fuse_isdeadfs(vp)) { 488 return ENXIO; 489 } 490 491 switch(ap->a_op) { 492 case F_GETLK: 493 op = FUSE_GETLK; 494 break; 495 case F_SETLK: 496 if (flags & F_WAIT) 497 op = FUSE_SETLKW; 498 else 499 op = FUSE_SETLK; 500 break; 501 case F_UNLCK: 502 op = FUSE_SETLK; 503 break; 504 default: 505 return EINVAL; 506 } 507 508 if (!(dataflags & FSESS_POSIX_LOCKS)) 509 return vop_stdadvlock(ap); 510 /* FUSE doesn't properly support flock until protocol 7.17 */ 511 if (flags & F_FLOCK) 512 return vop_stdadvlock(ap); 513 514 vn_lock(vp, LK_SHARED | LK_RETRY); 515 516 switch (fl->l_whence) { 517 case SEEK_SET: 518 case SEEK_CUR: 519 /* 520 * Caller is responsible for adding any necessary offset 521 * when SEEK_CUR is used. 522 */ 523 start = fl->l_start; 524 break; 525 526 case SEEK_END: 527 err = fuse_internal_getattr(vp, &vattr, cred, td); 528 if (err) 529 goto out; 530 size = vattr.va_size; 531 if (size > OFF_MAX || 532 (fl->l_start > 0 && size > OFF_MAX - fl->l_start)) { 533 err = EOVERFLOW; 534 goto out; 535 } 536 start = size + fl->l_start; 537 break; 538 539 default: 540 return (EINVAL); 541 } 542 543 err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid); 544 if (err) 545 goto out; 546 547 fdisp_init(&fdi, sizeof(*fli)); 548 549 fdisp_make_vp(&fdi, op, vp, td, cred); 550 fli = fdi.indata; 551 fli->fh = fufh->fh_id; 552 fli->owner = td->td_proc->p_pid; 553 fli->lk.start = start; 554 if (fl->l_len != 0) 555 fli->lk.end = start + fl->l_len - 1; 556 else 557 fli->lk.end = INT64_MAX; 558 fli->lk.type = fl->l_type; 559 fli->lk.pid = td->td_proc->p_pid; 560 561 err = fdisp_wait_answ(&fdi); 562 fdisp_destroy(&fdi); 563 564 if (err == 0 && op == FUSE_GETLK) { 565 flo = fdi.answ; 566 fl->l_type = flo->lk.type; 567 fl->l_whence = SEEK_SET; 568 if (flo->lk.type != F_UNLCK) { 569 fl->l_pid = flo->lk.pid; 570 fl->l_start = flo->lk.start; 571 if (flo->lk.end == INT64_MAX) 572 fl->l_len = 0; 573 else 574 fl->l_len = flo->lk.end - flo->lk.start + 1; 575 fl->l_start = flo->lk.start; 576 } 577 } 578 579 out: 580 VOP_UNLOCK(vp); 581 return err; 582 } 583 584 static int 585 fuse_vnop_allocate(struct vop_allocate_args *ap) 586 { 587 struct vnode *vp = ap->a_vp; 588 off_t *len = ap->a_len; 589 off_t *offset = ap->a_offset; 590 struct ucred *cred = ap->a_cred; 591 struct fuse_filehandle *fufh; 592 struct mount *mp = vnode_mount(vp); 593 struct fuse_dispatcher fdi; 594 struct fuse_fallocate_in *ffi; 595 struct uio io; 596 pid_t pid = curthread->td_proc->p_pid; 597 struct fuse_vnode_data *fvdat = VTOFUD(vp); 598 off_t filesize; 599 int err; 600 601 if (fuse_isdeadfs(vp)) 602 return (ENXIO); 603 604 switch (vp->v_type) { 605 case VFIFO: 606 return (ESPIPE); 607 case VLNK: 608 case VREG: 609 if (vfs_isrdonly(mp)) 610 return (EROFS); 611 break; 612 default: 613 return (ENODEV); 614 } 615 616 if (vfs_isrdonly(mp)) 617 return (EROFS); 618 619 if (fsess_not_impl(mp, FUSE_FALLOCATE)) 620 return (EINVAL); 621 622 io.uio_offset = *offset; 623 io.uio_resid = *len; 624 err = vn_rlimit_fsize(vp, &io, curthread); 625 if (err) 626 return (err); 627 628 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); 629 if (err) 630 return (err); 631 632 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE); 633 634 err = fuse_vnode_size(vp, &filesize, cred, curthread); 635 if (err) 636 return (err); 637 fuse_inval_buf_range(vp, filesize, *offset, *offset + *len); 638 639 fdisp_init(&fdi, sizeof(*ffi)); 640 fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred); 641 ffi = fdi.indata; 642 ffi->fh = fufh->fh_id; 643 ffi->offset = *offset; 644 ffi->length = *len; 645 ffi->mode = 0; 646 err = fdisp_wait_answ(&fdi); 647 648 if (err == ENOSYS) { 649 fsess_set_notimpl(mp, FUSE_FALLOCATE); 650 err = EINVAL; 651 } else if (err == EOPNOTSUPP) { 652 /* 653 * The file system server does not support FUSE_FALLOCATE with 654 * the supplied mode for this particular file. 655 */ 656 err = EINVAL; 657 } else if (!err) { 658 *offset += *len; 659 *len = 0; 660 fuse_vnode_undirty_cached_timestamps(vp, false); 661 fuse_internal_clear_suid_on_write(vp, cred, curthread); 662 if (*offset > fvdat->cached_attrs.va_size) { 663 fuse_vnode_setsize(vp, *offset, false); 664 getnanouptime(&fvdat->last_local_modify); 665 } 666 } 667 668 fdisp_destroy(&fdi); 669 return (err); 670 } 671 672 /* { 673 struct vnode *a_vp; 674 daddr_t a_bn; 675 struct bufobj **a_bop; 676 daddr_t *a_bnp; 677 int *a_runp; 678 int *a_runb; 679 } */ 680 static int 681 fuse_vnop_bmap(struct vop_bmap_args *ap) 682 { 683 struct vnode *vp = ap->a_vp; 684 struct bufobj **bo = ap->a_bop; 685 struct thread *td = curthread; 686 struct mount *mp; 687 struct fuse_dispatcher fdi; 688 struct fuse_bmap_in *fbi; 689 struct fuse_bmap_out *fbo; 690 struct fuse_data *data; 691 struct fuse_vnode_data *fvdat = VTOFUD(vp); 692 uint64_t biosize; 693 off_t fsize; 694 daddr_t lbn = ap->a_bn; 695 daddr_t *pbn = ap->a_bnp; 696 int *runp = ap->a_runp; 697 int *runb = ap->a_runb; 698 int error = 0; 699 int maxrun; 700 701 if (fuse_isdeadfs(vp)) { 702 return ENXIO; 703 } 704 705 mp = vnode_mount(vp); 706 data = fuse_get_mpdata(mp); 707 biosize = fuse_iosize(vp); 708 maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1, 709 data->max_readahead_blocks); 710 711 if (bo != NULL) 712 *bo = &vp->v_bufobj; 713 714 /* 715 * The FUSE_BMAP operation does not include the runp and runb 716 * variables, so we must guess. Report nonzero contiguous runs so 717 * cluster_read will combine adjacent reads. It's worthwhile to reduce 718 * upcalls even if we don't know the true physical layout of the file. 719 * 720 * FUSE file systems may opt out of read clustering in two ways: 721 * * mounting with -onoclusterr 722 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT 723 */ 724 if (runb != NULL) 725 *runb = MIN(lbn, maxrun); 726 if (runp != NULL && maxrun == 0) 727 *runp = 0; 728 else if (runp != NULL) { 729 /* 730 * If the file's size is cached, use that value to calculate 731 * runp, even if the cache is expired. runp is only advisory, 732 * and the risk of getting it wrong is not worth the cost of 733 * another upcall. 734 */ 735 if (fvdat->cached_attrs.va_size != VNOVAL) 736 fsize = fvdat->cached_attrs.va_size; 737 else 738 error = fuse_vnode_size(vp, &fsize, td->td_ucred, td); 739 if (error == 0) 740 *runp = MIN(MAX(0, fsize / (off_t)biosize - lbn - 1), 741 maxrun); 742 else 743 *runp = 0; 744 } 745 746 if (fsess_maybe_impl(mp, FUSE_BMAP)) { 747 fdisp_init(&fdi, sizeof(*fbi)); 748 fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred); 749 fbi = fdi.indata; 750 fbi->block = lbn; 751 fbi->blocksize = biosize; 752 error = fdisp_wait_answ(&fdi); 753 if (error == ENOSYS) { 754 fdisp_destroy(&fdi); 755 fsess_set_notimpl(mp, FUSE_BMAP); 756 error = 0; 757 } else { 758 fbo = fdi.answ; 759 if (error == 0 && pbn != NULL) 760 *pbn = fbo->block; 761 fdisp_destroy(&fdi); 762 return error; 763 } 764 } 765 766 /* If the daemon doesn't support BMAP, make up a sensible default */ 767 if (pbn != NULL) 768 *pbn = lbn * btodb(biosize); 769 return (error); 770 } 771 772 /* 773 struct vop_close_args { 774 struct vnode *a_vp; 775 int a_fflag; 776 struct ucred *a_cred; 777 struct thread *a_td; 778 }; 779 */ 780 static int 781 fuse_vnop_close(struct vop_close_args *ap) 782 { 783 struct vnode *vp = ap->a_vp; 784 struct mount *mp = vnode_mount(vp); 785 struct ucred *cred = ap->a_cred; 786 int fflag = ap->a_fflag; 787 struct thread *td = ap->a_td; 788 pid_t pid = td->td_proc->p_pid; 789 struct fuse_vnode_data *fvdat = VTOFUD(vp); 790 int err = 0; 791 792 if (fuse_isdeadfs(vp)) 793 return 0; 794 if (vnode_isdir(vp)) 795 return 0; 796 if (fflag & IO_NDELAY) 797 return 0; 798 799 err = fuse_flush(vp, cred, pid, fflag); 800 if (err == 0 && (fvdat->flag & FN_ATIMECHANGE) && !vfs_isrdonly(mp)) { 801 struct vattr vap; 802 struct fuse_data *data; 803 int dataflags; 804 int access_e = 0; 805 806 data = fuse_get_mpdata(mp); 807 dataflags = data->dataflags; 808 if (dataflags & FSESS_DEFAULT_PERMISSIONS) { 809 struct vattr va; 810 811 fuse_internal_getattr(vp, &va, cred, td); 812 access_e = vaccess(vp->v_type, va.va_mode, va.va_uid, 813 va.va_gid, VWRITE, cred); 814 } 815 if (access_e == 0) { 816 VATTR_NULL(&vap); 817 vap.va_atime = fvdat->cached_attrs.va_atime; 818 /* 819 * Ignore errors setting when setting atime. That 820 * should not cause close(2) to fail. 821 */ 822 fuse_internal_setattr(vp, &vap, td, NULL); 823 } 824 } 825 /* TODO: close the file handle, if we're sure it's no longer used */ 826 if ((fvdat->flag & FN_SIZECHANGE) != 0) { 827 fuse_vnode_savesize(vp, cred, td->td_proc->p_pid); 828 } 829 return err; 830 } 831 832 /* 833 struct vop_copy_file_range_args { 834 struct vop_generic_args a_gen; 835 struct vnode *a_invp; 836 off_t *a_inoffp; 837 struct vnode *a_outvp; 838 off_t *a_outoffp; 839 size_t *a_lenp; 840 unsigned int a_flags; 841 struct ucred *a_incred; 842 struct ucred *a_outcred; 843 struct thread *a_fsizetd; 844 } 845 */ 846 static int 847 fuse_vnop_copy_file_range(struct vop_copy_file_range_args *ap) 848 { 849 struct vnode *invp = ap->a_invp; 850 struct vnode *outvp = ap->a_outvp; 851 struct mount *mp = vnode_mount(invp); 852 struct fuse_vnode_data *outfvdat = VTOFUD(outvp); 853 struct fuse_dispatcher fdi; 854 struct fuse_filehandle *infufh, *outfufh; 855 struct fuse_copy_file_range_in *fcfri; 856 struct ucred *incred = ap->a_incred; 857 struct ucred *outcred = ap->a_outcred; 858 struct fuse_write_out *fwo; 859 struct thread *td; 860 struct uio io; 861 off_t outfilesize; 862 ssize_t r = 0; 863 pid_t pid; 864 int err; 865 866 err = ENOSYS; 867 if (mp == NULL || mp != vnode_mount(outvp)) 868 goto fallback; 869 870 if (incred->cr_uid != outcred->cr_uid) 871 goto fallback; 872 873 if (incred->cr_groups[0] != outcred->cr_groups[0]) 874 goto fallback; 875 876 /* Caller busied mp, mnt_data can be safely accessed. */ 877 if (fsess_not_impl(mp, FUSE_COPY_FILE_RANGE)) 878 goto fallback; 879 880 if (ap->a_fsizetd == NULL) 881 td = curthread; 882 else 883 td = ap->a_fsizetd; 884 pid = td->td_proc->p_pid; 885 886 vn_lock_pair(invp, false, LK_SHARED, outvp, false, LK_EXCLUSIVE); 887 if (invp->v_data == NULL || outvp->v_data == NULL) { 888 err = EBADF; 889 goto unlock; 890 } 891 892 err = fuse_filehandle_getrw(invp, FREAD, &infufh, incred, pid); 893 if (err) 894 goto unlock; 895 896 err = fuse_filehandle_getrw(outvp, FWRITE, &outfufh, outcred, pid); 897 if (err) 898 goto unlock; 899 900 io.uio_resid = *ap->a_lenp; 901 if (ap->a_fsizetd) { 902 io.uio_offset = *ap->a_outoffp; 903 err = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd); 904 if (err != 0) 905 goto unlock; 906 } 907 908 err = fuse_vnode_size(outvp, &outfilesize, outcred, curthread); 909 if (err) 910 goto unlock; 911 912 vnode_pager_clean_sync(invp); 913 err = fuse_inval_buf_range(outvp, outfilesize, *ap->a_outoffp, 914 *ap->a_outoffp + io.uio_resid); 915 if (err) 916 goto unlock; 917 918 fdisp_init(&fdi, sizeof(*fcfri)); 919 fdisp_make_vp(&fdi, FUSE_COPY_FILE_RANGE, invp, td, incred); 920 fcfri = fdi.indata; 921 fcfri->fh_in = infufh->fh_id; 922 fcfri->off_in = *ap->a_inoffp; 923 fcfri->nodeid_out = VTOI(outvp); 924 fcfri->fh_out = outfufh->fh_id; 925 fcfri->off_out = *ap->a_outoffp; 926 fcfri->len = io.uio_resid; 927 fcfri->flags = 0; 928 929 err = fdisp_wait_answ(&fdi); 930 if (err == 0) { 931 fwo = fdi.answ; 932 *ap->a_lenp = fwo->size; 933 *ap->a_inoffp += fwo->size; 934 *ap->a_outoffp += fwo->size; 935 fuse_internal_clear_suid_on_write(outvp, outcred, td); 936 if (*ap->a_outoffp > outfvdat->cached_attrs.va_size) { 937 fuse_vnode_setsize(outvp, *ap->a_outoffp, false); 938 getnanouptime(&outfvdat->last_local_modify); 939 } 940 fuse_vnode_update(invp, FN_ATIMECHANGE); 941 fuse_vnode_update(outvp, FN_MTIMECHANGE | FN_CTIMECHANGE); 942 } 943 fdisp_destroy(&fdi); 944 945 unlock: 946 if (invp != outvp) 947 VOP_UNLOCK(invp); 948 VOP_UNLOCK(outvp); 949 950 if (err == ENOSYS) 951 fsess_set_notimpl(mp, FUSE_COPY_FILE_RANGE); 952 fallback: 953 954 /* 955 * No need to call vn_rlimit_fsizex_res before return, since the uio is 956 * local. 957 */ 958 return (err); 959 } 960 961 static void 962 fdisp_make_mknod_for_fallback( 963 struct fuse_dispatcher *fdip, 964 struct componentname *cnp, 965 struct vnode *dvp, 966 uint64_t parentnid, 967 struct thread *td, 968 struct ucred *cred, 969 mode_t mode, 970 enum fuse_opcode *op) 971 { 972 struct fuse_mknod_in *fmni; 973 974 fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1); 975 *op = FUSE_MKNOD; 976 fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred); 977 fmni = fdip->indata; 978 fmni->mode = mode; 979 fmni->rdev = 0; 980 memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr, 981 cnp->cn_namelen); 982 ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0'; 983 } 984 /* 985 struct vnop_create_args { 986 struct vnode *a_dvp; 987 struct vnode **a_vpp; 988 struct componentname *a_cnp; 989 struct vattr *a_vap; 990 }; 991 */ 992 static int 993 fuse_vnop_create(struct vop_create_args *ap) 994 { 995 struct vnode *dvp = ap->a_dvp; 996 struct vnode **vpp = ap->a_vpp; 997 struct componentname *cnp = ap->a_cnp; 998 struct vattr *vap = ap->a_vap; 999 struct thread *td = curthread; 1000 struct ucred *cred = cnp->cn_cred; 1001 1002 struct fuse_data *data; 1003 struct fuse_create_in *fci; 1004 struct fuse_entry_out *feo; 1005 struct fuse_open_out *foo; 1006 struct fuse_dispatcher fdi, fdi2; 1007 struct fuse_dispatcher *fdip = &fdi; 1008 struct fuse_dispatcher *fdip2 = NULL; 1009 1010 int err; 1011 1012 struct mount *mp = vnode_mount(dvp); 1013 data = fuse_get_mpdata(mp); 1014 uint64_t parentnid = VTOFUD(dvp)->nid; 1015 mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode); 1016 enum fuse_opcode op; 1017 int flags; 1018 1019 if (fuse_isdeadfs(dvp)) 1020 return ENXIO; 1021 1022 /* FUSE expects sockets to be created with FUSE_MKNOD */ 1023 if (vap->va_type == VSOCK) 1024 return fuse_internal_mknod(dvp, vpp, cnp, vap); 1025 1026 /* 1027 * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a 1028 * writable mode makes sense, and we might as well include readability 1029 * too. 1030 */ 1031 flags = O_RDWR; 1032 1033 bzero(&fdi, sizeof(fdi)); 1034 1035 if (vap->va_type != VREG) 1036 return (EINVAL); 1037 1038 if (fsess_not_impl(mp, FUSE_CREATE) || vap->va_type == VSOCK) { 1039 /* Fallback to FUSE_MKNOD/FUSE_OPEN */ 1040 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td, 1041 cred, mode, &op); 1042 } else { 1043 /* Use FUSE_CREATE */ 1044 size_t insize; 1045 1046 op = FUSE_CREATE; 1047 fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1); 1048 fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred); 1049 fci = fdip->indata; 1050 fci->mode = mode; 1051 fci->flags = O_CREAT | flags; 1052 if (fuse_libabi_geq(data, 7, 12)) { 1053 insize = sizeof(*fci); 1054 fci->umask = td->td_proc->p_pd->pd_cmask; 1055 } else { 1056 insize = sizeof(struct fuse_open_in); 1057 } 1058 1059 memcpy((char *)fdip->indata + insize, cnp->cn_nameptr, 1060 cnp->cn_namelen); 1061 ((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0'; 1062 } 1063 1064 err = fdisp_wait_answ(fdip); 1065 1066 if (err) { 1067 if (err == ENOSYS && op == FUSE_CREATE) { 1068 fsess_set_notimpl(mp, FUSE_CREATE); 1069 fdisp_destroy(fdip); 1070 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, 1071 parentnid, td, cred, mode, &op); 1072 err = fdisp_wait_answ(fdip); 1073 } 1074 if (err) 1075 goto out; 1076 } 1077 1078 feo = fdip->answ; 1079 1080 if ((err = fuse_internal_checkentry(feo, vap->va_type))) { 1081 goto out; 1082 } 1083 1084 if (op == FUSE_CREATE) { 1085 if (fuse_libabi_geq(data, 7, 9)) 1086 foo = (struct fuse_open_out*)(feo + 1); 1087 else 1088 foo = (struct fuse_open_out*)((char*)feo + 1089 FUSE_COMPAT_ENTRY_OUT_SIZE); 1090 } else { 1091 /* Issue a separate FUSE_OPEN */ 1092 struct fuse_open_in *foi; 1093 1094 fdip2 = &fdi2; 1095 fdisp_init(fdip2, sizeof(*foi)); 1096 fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td, 1097 cred); 1098 foi = fdip2->indata; 1099 foi->flags = flags; 1100 err = fdisp_wait_answ(fdip2); 1101 if (err) 1102 goto out; 1103 foo = fdip2->answ; 1104 } 1105 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type); 1106 if (err) { 1107 struct fuse_release_in *fri; 1108 uint64_t nodeid = feo->nodeid; 1109 uint64_t fh_id = foo->fh; 1110 1111 fdisp_destroy(fdip); 1112 fdisp_init(fdip, sizeof(*fri)); 1113 fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred); 1114 fri = fdip->indata; 1115 fri->fh = fh_id; 1116 fri->flags = flags; 1117 fuse_insert_callback(fdip->tick, fuse_internal_forget_callback); 1118 fuse_insert_message(fdip->tick, false); 1119 goto out; 1120 } 1121 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create"); 1122 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid, 1123 feo->attr_valid_nsec, NULL, true); 1124 1125 fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo); 1126 fuse_vnode_open(*vpp, foo->open_flags, td); 1127 /* 1128 * Purge the parent's attribute cache because the daemon should've 1129 * updated its mtime and ctime 1130 */ 1131 fuse_vnode_clear_attr_cache(dvp); 1132 cache_purge_negative(dvp); 1133 1134 out: 1135 if (fdip2) 1136 fdisp_destroy(fdip2); 1137 fdisp_destroy(fdip); 1138 return err; 1139 } 1140 1141 /* 1142 struct vnop_fdatasync_args { 1143 struct vop_generic_args a_gen; 1144 struct vnode * a_vp; 1145 struct thread * a_td; 1146 }; 1147 */ 1148 static int 1149 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap) 1150 { 1151 struct vnode *vp = ap->a_vp; 1152 struct thread *td = ap->a_td; 1153 int waitfor = MNT_WAIT; 1154 1155 int err = 0; 1156 1157 if (fuse_isdeadfs(vp)) { 1158 return 0; 1159 } 1160 if ((err = vop_stdfdatasync_buf(ap))) 1161 return err; 1162 1163 return fuse_internal_fsync(vp, td, waitfor, true); 1164 } 1165 1166 /* 1167 struct vnop_fsync_args { 1168 struct vop_generic_args a_gen; 1169 struct vnode * a_vp; 1170 int a_waitfor; 1171 struct thread * a_td; 1172 }; 1173 */ 1174 static int 1175 fuse_vnop_fsync(struct vop_fsync_args *ap) 1176 { 1177 struct vnode *vp = ap->a_vp; 1178 struct thread *td = ap->a_td; 1179 int waitfor = ap->a_waitfor; 1180 int err = 0; 1181 1182 if (fuse_isdeadfs(vp)) { 1183 return 0; 1184 } 1185 if ((err = vop_stdfsync(ap))) 1186 return err; 1187 1188 return fuse_internal_fsync(vp, td, waitfor, false); 1189 } 1190 1191 /* 1192 struct vnop_getattr_args { 1193 struct vnode *a_vp; 1194 struct vattr *a_vap; 1195 struct ucred *a_cred; 1196 struct thread *a_td; 1197 }; 1198 */ 1199 static int 1200 fuse_vnop_getattr(struct vop_getattr_args *ap) 1201 { 1202 struct vnode *vp = ap->a_vp; 1203 struct vattr *vap = ap->a_vap; 1204 struct ucred *cred = ap->a_cred; 1205 struct thread *td = curthread; 1206 1207 int err = 0; 1208 int dataflags; 1209 1210 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags; 1211 1212 /* Note that we are not bailing out on a dead file system just yet. */ 1213 1214 if (!(dataflags & FSESS_INITED)) { 1215 if (!vnode_isvroot(vp)) { 1216 fdata_set_dead(fuse_get_mpdata(vnode_mount(vp))); 1217 err = ENOTCONN; 1218 return err; 1219 } else { 1220 goto fake; 1221 } 1222 } 1223 err = fuse_internal_getattr(vp, vap, cred, td); 1224 if (err == ENOTCONN && vnode_isvroot(vp)) { 1225 /* see comment in fuse_vfsop_statfs() */ 1226 goto fake; 1227 } else { 1228 return err; 1229 } 1230 1231 fake: 1232 bzero(vap, sizeof(*vap)); 1233 vap->va_type = vnode_vtype(vp); 1234 1235 return 0; 1236 } 1237 1238 /* 1239 struct vnop_inactive_args { 1240 struct vnode *a_vp; 1241 }; 1242 */ 1243 static int 1244 fuse_vnop_inactive(struct vop_inactive_args *ap) 1245 { 1246 struct vnode *vp = ap->a_vp; 1247 struct thread *td = curthread; 1248 1249 struct fuse_vnode_data *fvdat = VTOFUD(vp); 1250 struct fuse_filehandle *fufh, *fufh_tmp; 1251 1252 int need_flush = 1; 1253 1254 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) { 1255 if (need_flush && vp->v_type == VREG) { 1256 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) { 1257 fuse_vnode_savesize(vp, NULL, 0); 1258 } 1259 if ((fvdat->flag & FN_REVOKED) != 0) 1260 fuse_io_invalbuf(vp, td); 1261 else 1262 fuse_io_flushbuf(vp, MNT_WAIT, td); 1263 need_flush = 0; 1264 } 1265 fuse_filehandle_close(vp, fufh, td, NULL); 1266 } 1267 1268 if ((fvdat->flag & FN_REVOKED) != 0) 1269 vrecycle(vp); 1270 1271 return 0; 1272 } 1273 1274 /* 1275 struct vnop_ioctl_args { 1276 struct vnode *a_vp; 1277 u_long a_command; 1278 caddr_t a_data; 1279 int a_fflag; 1280 struct ucred *a_cred; 1281 struct thread *a_td; 1282 }; 1283 */ 1284 static int 1285 fuse_vnop_ioctl(struct vop_ioctl_args *ap) 1286 { 1287 struct vnode *vp = ap->a_vp; 1288 struct mount *mp = vnode_mount(vp); 1289 struct ucred *cred = ap->a_cred; 1290 off_t *offp; 1291 pid_t pid = ap->a_td->td_proc->p_pid; 1292 int err; 1293 1294 switch (ap->a_command) { 1295 case FIOSEEKDATA: 1296 case FIOSEEKHOLE: 1297 /* Call FUSE_LSEEK, if we can, or fall back to vop_stdioctl */ 1298 if (fsess_maybe_impl(mp, FUSE_LSEEK)) { 1299 int whence; 1300 1301 offp = ap->a_data; 1302 if (ap->a_command == FIOSEEKDATA) 1303 whence = SEEK_DATA; 1304 else 1305 whence = SEEK_HOLE; 1306 1307 vn_lock(vp, LK_SHARED | LK_RETRY); 1308 err = fuse_vnop_do_lseek(vp, ap->a_td, cred, pid, offp, 1309 whence); 1310 VOP_UNLOCK(vp); 1311 } 1312 if (fsess_not_impl(mp, FUSE_LSEEK)) 1313 err = vop_stdioctl(ap); 1314 break; 1315 default: 1316 /* TODO: implement FUSE_IOCTL */ 1317 err = ENOTTY; 1318 break; 1319 } 1320 return (err); 1321 } 1322 1323 1324 /* 1325 struct vnop_link_args { 1326 struct vnode *a_tdvp; 1327 struct vnode *a_vp; 1328 struct componentname *a_cnp; 1329 }; 1330 */ 1331 static int 1332 fuse_vnop_link(struct vop_link_args *ap) 1333 { 1334 struct vnode *vp = ap->a_vp; 1335 struct vnode *tdvp = ap->a_tdvp; 1336 struct componentname *cnp = ap->a_cnp; 1337 1338 struct vattr *vap = VTOVA(vp); 1339 1340 struct fuse_dispatcher fdi; 1341 struct fuse_entry_out *feo; 1342 struct fuse_link_in fli; 1343 1344 int err; 1345 1346 if (fuse_isdeadfs(vp)) { 1347 return ENXIO; 1348 } 1349 if (vnode_mount(tdvp) != vnode_mount(vp)) { 1350 return EXDEV; 1351 } 1352 1353 /* 1354 * This is a seatbelt check to protect naive userspace filesystems from 1355 * themselves and the limitations of the FUSE IPC protocol. If a 1356 * filesystem does not allow attribute caching, assume it is capable of 1357 * validating that nlink does not overflow. 1358 */ 1359 if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX) 1360 return EMLINK; 1361 fli.oldnodeid = VTOI(vp); 1362 1363 fdisp_init(&fdi, 0); 1364 fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp, 1365 FUSE_LINK, &fli, sizeof(fli), &fdi); 1366 if ((err = fdisp_wait_answ(&fdi))) { 1367 goto out; 1368 } 1369 feo = fdi.answ; 1370 1371 if (fli.oldnodeid != feo->nodeid) { 1372 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp)); 1373 fuse_warn(data, FSESS_WARN_ILLEGAL_INODE, 1374 "Assigned wrong inode for a hard link."); 1375 fuse_vnode_clear_attr_cache(vp); 1376 fuse_vnode_clear_attr_cache(tdvp); 1377 err = EIO; 1378 goto out; 1379 } 1380 1381 err = fuse_internal_checkentry(feo, vnode_vtype(vp)); 1382 if (!err) { 1383 /* 1384 * Purge the parent's attribute cache because the daemon 1385 * should've updated its mtime and ctime 1386 */ 1387 fuse_vnode_clear_attr_cache(tdvp); 1388 fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid, 1389 feo->attr_valid_nsec, NULL, true); 1390 } 1391 out: 1392 fdisp_destroy(&fdi); 1393 return err; 1394 } 1395 1396 struct fuse_lookup_alloc_arg { 1397 struct fuse_entry_out *feo; 1398 struct componentname *cnp; 1399 uint64_t nid; 1400 __enum_uint8(vtype) vtyp; 1401 }; 1402 1403 /* Callback for vn_get_ino */ 1404 static int 1405 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp) 1406 { 1407 struct fuse_lookup_alloc_arg *flaa = arg; 1408 1409 return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp, 1410 flaa->vtyp); 1411 } 1412 1413 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup, 1414 "int", "struct timespec*", "struct timespec*"); 1415 /* 1416 struct vnop_lookup_args { 1417 struct vnodeop_desc *a_desc; 1418 struct vnode *a_dvp; 1419 struct vnode **a_vpp; 1420 struct componentname *a_cnp; 1421 }; 1422 */ 1423 int 1424 fuse_vnop_lookup(struct vop_lookup_args *ap) 1425 { 1426 struct vnode *dvp = ap->a_dvp; 1427 struct vnode **vpp = ap->a_vpp; 1428 struct componentname *cnp = ap->a_cnp; 1429 struct thread *td = curthread; 1430 struct ucred *cred = cnp->cn_cred; 1431 struct timespec now; 1432 1433 int nameiop = cnp->cn_nameiop; 1434 int flags = cnp->cn_flags; 1435 int islastcn = flags & ISLASTCN; 1436 struct mount *mp = vnode_mount(dvp); 1437 struct fuse_data *data = fuse_get_mpdata(mp); 1438 int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS; 1439 bool is_dot; 1440 1441 int err = 0; 1442 int lookup_err = 0; 1443 struct vnode *vp = NULL; 1444 1445 struct fuse_dispatcher fdi; 1446 bool did_lookup = false; 1447 struct fuse_entry_out *feo = NULL; 1448 __enum_uint8(vtype) vtyp; /* vnode type of target */ 1449 1450 uint64_t nid; 1451 1452 if (fuse_isdeadfs(dvp)) { 1453 *vpp = NULL; 1454 return ENXIO; 1455 } 1456 if (!vnode_isdir(dvp)) 1457 return ENOTDIR; 1458 1459 if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP)) 1460 return EROFS; 1461 1462 if ((cnp->cn_flags & NOEXECCHECK) != 0) 1463 cnp->cn_flags &= ~NOEXECCHECK; 1464 else if ((err = fuse_internal_access(dvp, VEXEC, td, cred))) 1465 return err; 1466 1467 is_dot = cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.'; 1468 if ((flags & ISDOTDOT) && !(data->dataflags & FSESS_EXPORT_SUPPORT)) 1469 { 1470 if (!(VTOFUD(dvp)->flag & FN_PARENT_NID)) { 1471 /* 1472 * Since the file system doesn't support ".." lookups, 1473 * we have no way to find this entry. 1474 */ 1475 return ESTALE; 1476 } 1477 nid = VTOFUD(dvp)->parent_nid; 1478 if (nid == 0) 1479 return ENOENT; 1480 /* .. is obviously a directory */ 1481 vtyp = VDIR; 1482 } else if (is_dot) { 1483 nid = VTOI(dvp); 1484 /* . is obviously a directory */ 1485 vtyp = VDIR; 1486 } else { 1487 struct timespec timeout; 1488 int ncpticks; /* here to accommodate for API contract */ 1489 1490 err = cache_lookup(dvp, vpp, cnp, &timeout, &ncpticks); 1491 getnanouptime(&now); 1492 SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now); 1493 switch (err) { 1494 case -1: /* positive match */ 1495 if (timespeccmp(&timeout, &now, >)) { 1496 counter_u64_add(fuse_lookup_cache_hits, 1); 1497 } else { 1498 /* Cache timeout */ 1499 counter_u64_add(fuse_lookup_cache_misses, 1); 1500 bintime_clear( 1501 &VTOFUD(*vpp)->entry_cache_timeout); 1502 cache_purge(*vpp); 1503 if (dvp != *vpp) 1504 vput(*vpp); 1505 else 1506 vrele(*vpp); 1507 *vpp = NULL; 1508 break; 1509 } 1510 return 0; 1511 1512 case 0: /* no match in cache */ 1513 counter_u64_add(fuse_lookup_cache_misses, 1); 1514 break; 1515 1516 case ENOENT: /* negative match */ 1517 if (timespeccmp(&timeout, &now, <=)) { 1518 /* Cache timeout */ 1519 cache_purge_negative(dvp); 1520 break; 1521 } 1522 /* fall through */ 1523 default: 1524 return err; 1525 } 1526 1527 fdisp_init(&fdi, cnp->cn_namelen + 1); 1528 fdisp_make(&fdi, FUSE_LOOKUP, mp, VTOI(dvp), td, cred); 1529 1530 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen); 1531 ((char *)fdi.indata)[cnp->cn_namelen] = '\0'; 1532 lookup_err = fdisp_wait_answ(&fdi); 1533 did_lookup = true; 1534 1535 if (!lookup_err) { 1536 /* lookup call succeeded */ 1537 feo = (struct fuse_entry_out *)fdi.answ; 1538 nid = feo->nodeid; 1539 if (nid == 0) { 1540 /* zero nodeid means ENOENT and cache it */ 1541 struct timespec timeout; 1542 1543 fdi.answ_stat = ENOENT; 1544 lookup_err = ENOENT; 1545 if (cnp->cn_flags & MAKEENTRY) { 1546 fuse_validity_2_timespec(feo, &timeout); 1547 /* Use the same entry_time for .. as for 1548 * the file itself. That doesn't honor 1549 * exactly what the fuse server tells 1550 * us, but to do otherwise would require 1551 * another cache lookup at this point. 1552 */ 1553 struct timespec *dtsp = NULL; 1554 cache_enter_time(dvp, *vpp, cnp, 1555 &timeout, dtsp); 1556 } 1557 } 1558 vtyp = IFTOVT(feo->attr.mode); 1559 } 1560 if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) { 1561 fdisp_destroy(&fdi); 1562 return lookup_err; 1563 } 1564 } 1565 /* lookup_err, if non-zero, must be ENOENT at this point */ 1566 1567 if (lookup_err) { 1568 /* Entry not found */ 1569 if ((nameiop == CREATE || nameiop == RENAME) && islastcn) { 1570 if (default_permissions) 1571 err = fuse_internal_access(dvp, VWRITE, td, 1572 cred); 1573 else 1574 err = 0; 1575 if (!err) { 1576 err = EJUSTRETURN; 1577 } 1578 } else { 1579 err = ENOENT; 1580 } 1581 } else { 1582 /* Entry was found */ 1583 if (flags & ISDOTDOT) { 1584 struct fuse_lookup_alloc_arg flaa; 1585 1586 flaa.nid = nid; 1587 flaa.feo = feo; 1588 flaa.cnp = cnp; 1589 flaa.vtyp = vtyp; 1590 err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0, 1591 &vp); 1592 *vpp = vp; 1593 } else if (nid == VTOI(dvp)) { 1594 if (is_dot) { 1595 vref(dvp); 1596 *vpp = dvp; 1597 } else { 1598 fuse_warn(fuse_get_mpdata(mp), 1599 FSESS_WARN_ILLEGAL_INODE, 1600 "Assigned same inode to both parent and " 1601 "child."); 1602 err = EIO; 1603 } 1604 1605 } else { 1606 struct fuse_vnode_data *fvdat; 1607 1608 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp, 1609 &vp, cnp, vtyp); 1610 if (err) 1611 goto out; 1612 *vpp = vp; 1613 fvdat = VTOFUD(vp); 1614 1615 MPASS(feo != NULL); 1616 if (timespeccmp(&now, &fvdat->last_local_modify, >)) { 1617 /* 1618 * Attributes from the server are definitely 1619 * newer than the last attributes we sent to 1620 * the server, so cache them. 1621 */ 1622 fuse_internal_cache_attrs(*vpp, &feo->attr, 1623 feo->attr_valid, feo->attr_valid_nsec, 1624 NULL, true); 1625 } 1626 fuse_validity_2_bintime(feo->entry_valid, 1627 feo->entry_valid_nsec, 1628 &fvdat->entry_cache_timeout); 1629 1630 if ((nameiop == DELETE || nameiop == RENAME) && 1631 islastcn && default_permissions) 1632 { 1633 struct vattr dvattr; 1634 1635 err = fuse_internal_access(dvp, VWRITE, td, 1636 cred); 1637 if (err != 0) 1638 goto out; 1639 /* 1640 * if the parent's sticky bit is set, check 1641 * whether we're allowed to remove the file. 1642 * Need to figure out the vnode locking to make 1643 * this work. 1644 */ 1645 fuse_internal_getattr(dvp, &dvattr, cred, td); 1646 if ((dvattr.va_mode & S_ISTXT) && 1647 fuse_internal_access(dvp, VADMIN, td, 1648 cred) && 1649 fuse_internal_access(*vpp, VADMIN, td, 1650 cred)) { 1651 err = EPERM; 1652 goto out; 1653 } 1654 } 1655 } 1656 } 1657 out: 1658 if (err) { 1659 if (vp != NULL && dvp != vp) 1660 vput(vp); 1661 else if (vp != NULL) 1662 vrele(vp); 1663 *vpp = NULL; 1664 } 1665 if (did_lookup) 1666 fdisp_destroy(&fdi); 1667 1668 return err; 1669 } 1670 1671 /* 1672 struct vnop_mkdir_args { 1673 struct vnode *a_dvp; 1674 struct vnode **a_vpp; 1675 struct componentname *a_cnp; 1676 struct vattr *a_vap; 1677 }; 1678 */ 1679 static int 1680 fuse_vnop_mkdir(struct vop_mkdir_args *ap) 1681 { 1682 struct vnode *dvp = ap->a_dvp; 1683 struct vnode **vpp = ap->a_vpp; 1684 struct componentname *cnp = ap->a_cnp; 1685 struct vattr *vap = ap->a_vap; 1686 1687 struct fuse_mkdir_in fmdi; 1688 1689 if (fuse_isdeadfs(dvp)) { 1690 return ENXIO; 1691 } 1692 fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode); 1693 fmdi.umask = curthread->td_proc->p_pd->pd_cmask; 1694 1695 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi, 1696 sizeof(fmdi), VDIR)); 1697 } 1698 1699 /* 1700 struct vnop_mknod_args { 1701 struct vnode *a_dvp; 1702 struct vnode **a_vpp; 1703 struct componentname *a_cnp; 1704 struct vattr *a_vap; 1705 }; 1706 */ 1707 static int 1708 fuse_vnop_mknod(struct vop_mknod_args *ap) 1709 { 1710 1711 struct vnode *dvp = ap->a_dvp; 1712 struct vnode **vpp = ap->a_vpp; 1713 struct componentname *cnp = ap->a_cnp; 1714 struct vattr *vap = ap->a_vap; 1715 1716 if (fuse_isdeadfs(dvp)) 1717 return ENXIO; 1718 1719 return fuse_internal_mknod(dvp, vpp, cnp, vap); 1720 } 1721 1722 /* 1723 struct vop_open_args { 1724 struct vnode *a_vp; 1725 int a_mode; 1726 struct ucred *a_cred; 1727 struct thread *a_td; 1728 int a_fdidx; / struct file *a_fp; 1729 }; 1730 */ 1731 static int 1732 fuse_vnop_open(struct vop_open_args *ap) 1733 { 1734 struct vnode *vp = ap->a_vp; 1735 int a_mode = ap->a_mode; 1736 struct thread *td = ap->a_td; 1737 struct ucred *cred = ap->a_cred; 1738 pid_t pid = td->td_proc->p_pid; 1739 1740 if (fuse_isdeadfs(vp)) 1741 return ENXIO; 1742 if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO) 1743 return (EOPNOTSUPP); 1744 if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0) 1745 return EINVAL; 1746 1747 if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) { 1748 fuse_vnode_open(vp, 0, td); 1749 return 0; 1750 } 1751 1752 return fuse_filehandle_open(vp, a_mode, NULL, td, cred); 1753 } 1754 1755 static int 1756 fuse_vnop_pathconf(struct vop_pathconf_args *ap) 1757 { 1758 struct vnode *vp = ap->a_vp; 1759 struct mount *mp; 1760 struct fuse_filehandle *fufh; 1761 int err; 1762 bool closefufh = false; 1763 1764 switch (ap->a_name) { 1765 case _PC_FILESIZEBITS: 1766 *ap->a_retval = 64; 1767 return (0); 1768 case _PC_NAME_MAX: 1769 *ap->a_retval = NAME_MAX; 1770 return (0); 1771 case _PC_LINK_MAX: 1772 *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX); 1773 return (0); 1774 case _PC_SYMLINK_MAX: 1775 *ap->a_retval = MAXPATHLEN; 1776 return (0); 1777 case _PC_NO_TRUNC: 1778 *ap->a_retval = 1; 1779 return (0); 1780 case _PC_MIN_HOLE_SIZE: 1781 /* 1782 * The FUSE protocol provides no mechanism for a server to 1783 * report _PC_MIN_HOLE_SIZE. It's a protocol bug. Instead, 1784 * return EINVAL if the server does not support FUSE_LSEEK, or 1785 * 1 if it does. 1786 */ 1787 mp = vnode_mount(vp); 1788 if (!fsess_is_impl(mp, FUSE_LSEEK) && 1789 !fsess_not_impl(mp, FUSE_LSEEK)) { 1790 off_t offset = 0; 1791 1792 /* 1793 * Issue a FUSE_LSEEK to find out if it's supported. 1794 * Use SEEK_DATA instead of SEEK_HOLE, because the 1795 * latter generally requires sequential scans of file 1796 * metadata, which can be slow. 1797 */ 1798 err = fuse_vnop_do_lseek(vp, curthread, 1799 curthread->td_ucred, curthread->td_proc->p_pid, 1800 &offset, SEEK_DATA); 1801 if (err == EBADF) { 1802 /* 1803 * pathconf() doesn't necessarily open the 1804 * file. So we may need to do it here. 1805 */ 1806 err = fuse_filehandle_open(vp, FREAD, &fufh, 1807 curthread, curthread->td_ucred); 1808 if (err == 0) { 1809 closefufh = true; 1810 err = fuse_vnop_do_lseek(vp, curthread, 1811 curthread->td_ucred, 1812 curthread->td_proc->p_pid, &offset, 1813 SEEK_DATA); 1814 } 1815 if (closefufh) 1816 fuse_filehandle_close(vp, fufh, 1817 curthread, curthread->td_ucred); 1818 } 1819 1820 } 1821 1822 if (fsess_is_impl(mp, FUSE_LSEEK)) { 1823 *ap->a_retval = 1; 1824 return (0); 1825 } else if (fsess_not_impl(mp, FUSE_LSEEK)) { 1826 /* FUSE_LSEEK is not implemented */ 1827 return (EINVAL); 1828 } else { 1829 return (err); 1830 } 1831 default: 1832 return (vop_stdpathconf(ap)); 1833 } 1834 } 1835 1836 SDT_PROBE_DEFINE3(fusefs, , vnops, filehandles_closed, "struct vnode*", 1837 "struct uio*", "struct ucred*"); 1838 /* 1839 struct vnop_read_args { 1840 struct vnode *a_vp; 1841 struct uio *a_uio; 1842 int a_ioflag; 1843 struct ucred *a_cred; 1844 }; 1845 */ 1846 static int 1847 fuse_vnop_read(struct vop_read_args *ap) 1848 { 1849 struct vnode *vp = ap->a_vp; 1850 struct uio *uio = ap->a_uio; 1851 int ioflag = ap->a_ioflag; 1852 struct ucred *cred = ap->a_cred; 1853 pid_t pid = curthread->td_proc->p_pid; 1854 struct fuse_filehandle *fufh; 1855 int err; 1856 bool closefufh = false, directio; 1857 1858 MPASS(vp->v_type == VREG || vp->v_type == VDIR); 1859 1860 if (fuse_isdeadfs(vp)) { 1861 return ENXIO; 1862 } 1863 1864 if (VTOFUD(vp)->flag & FN_DIRECTIO) { 1865 ioflag |= IO_DIRECT; 1866 } 1867 1868 err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid); 1869 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) { 1870 /* 1871 * nfsd will do I/O without first doing VOP_OPEN. We 1872 * must implicitly open the file here 1873 */ 1874 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred); 1875 closefufh = true; 1876 } 1877 if (err) { 1878 SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred); 1879 return err; 1880 } 1881 1882 /* 1883 * Ideally, when the daemon asks for direct io at open time, the 1884 * standard file flag should be set according to this, so that would 1885 * just change the default mode, which later on could be changed via 1886 * fcntl(2). 1887 * But this doesn't work, the O_DIRECT flag gets cleared at some point 1888 * (don't know where). So to make any use of the Fuse direct_io option, 1889 * we hardwire it into the file's private data (similarly to Linux, 1890 * btw.). 1891 */ 1892 directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp)); 1893 1894 fuse_vnode_update(vp, FN_ATIMECHANGE); 1895 if (directio) { 1896 SDT_PROBE2(fusefs, , vnops, trace, 1, "direct read of vnode"); 1897 err = fuse_read_directbackend(vp, uio, cred, fufh); 1898 } else { 1899 SDT_PROBE2(fusefs, , vnops, trace, 1, "buffered read of vnode"); 1900 err = fuse_read_biobackend(vp, uio, ioflag, cred, fufh, pid); 1901 } 1902 1903 if (closefufh) 1904 fuse_filehandle_close(vp, fufh, curthread, cred); 1905 1906 return (err); 1907 } 1908 1909 /* 1910 struct vnop_readdir_args { 1911 struct vnode *a_vp; 1912 struct uio *a_uio; 1913 struct ucred *a_cred; 1914 int *a_eofflag; 1915 int *a_ncookies; 1916 uint64_t **a_cookies; 1917 }; 1918 */ 1919 static int 1920 fuse_vnop_readdir(struct vop_readdir_args *ap) 1921 { 1922 struct vnode *vp = ap->a_vp; 1923 struct uio *uio = ap->a_uio; 1924 struct ucred *cred = ap->a_cred; 1925 struct fuse_filehandle *fufh = NULL; 1926 struct mount *mp = vnode_mount(vp); 1927 struct fuse_iov cookediov; 1928 int err = 0; 1929 uint64_t *cookies; 1930 ssize_t tresid; 1931 int ncookies; 1932 bool closefufh = false; 1933 pid_t pid = curthread->td_proc->p_pid; 1934 1935 if (ap->a_eofflag) 1936 *ap->a_eofflag = 0; 1937 if (fuse_isdeadfs(vp)) { 1938 return ENXIO; 1939 } 1940 if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */ 1941 (uio_resid(uio) < sizeof(struct dirent))) { 1942 return EINVAL; 1943 } 1944 1945 tresid = uio->uio_resid; 1946 err = fuse_filehandle_get_dir(vp, &fufh, cred, pid); 1947 if (err == EBADF && mp->mnt_flag & MNT_EXPORTED) { 1948 KASSERT(fuse_get_mpdata(mp)->dataflags 1949 & FSESS_NO_OPENDIR_SUPPORT, 1950 ("FUSE file systems that don't set " 1951 "FUSE_NO_OPENDIR_SUPPORT should not be exported")); 1952 /* 1953 * nfsd will do VOP_READDIR without first doing VOP_OPEN. We 1954 * must implicitly open the directory here. 1955 */ 1956 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred); 1957 closefufh = true; 1958 } 1959 if (err) 1960 return (err); 1961 if (ap->a_ncookies != NULL) { 1962 ncookies = uio->uio_resid / 1963 (offsetof(struct dirent, d_name) + 4) + 1; 1964 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK); 1965 *ap->a_ncookies = ncookies; 1966 *ap->a_cookies = cookies; 1967 } else { 1968 ncookies = 0; 1969 cookies = NULL; 1970 } 1971 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1) 1972 fiov_init(&cookediov, DIRCOOKEDSIZE); 1973 1974 err = fuse_internal_readdir(vp, uio, fufh, &cookediov, 1975 &ncookies, cookies); 1976 1977 fiov_teardown(&cookediov); 1978 if (closefufh) 1979 fuse_filehandle_close(vp, fufh, curthread, cred); 1980 1981 if (ap->a_ncookies != NULL) { 1982 if (err == 0) { 1983 *ap->a_ncookies -= ncookies; 1984 } else { 1985 free(*ap->a_cookies, M_TEMP); 1986 *ap->a_ncookies = 0; 1987 *ap->a_cookies = NULL; 1988 } 1989 } 1990 if (err == 0 && tresid == uio->uio_resid) 1991 *ap->a_eofflag = 1; 1992 1993 return err; 1994 } 1995 1996 /* 1997 struct vnop_readlink_args { 1998 struct vnode *a_vp; 1999 struct uio *a_uio; 2000 struct ucred *a_cred; 2001 }; 2002 */ 2003 static int 2004 fuse_vnop_readlink(struct vop_readlink_args *ap) 2005 { 2006 struct vnode *vp = ap->a_vp; 2007 struct uio *uio = ap->a_uio; 2008 struct ucred *cred = ap->a_cred; 2009 2010 struct fuse_dispatcher fdi; 2011 int err; 2012 2013 if (fuse_isdeadfs(vp)) { 2014 return ENXIO; 2015 } 2016 if (!vnode_islnk(vp)) { 2017 return EINVAL; 2018 } 2019 fdisp_init(&fdi, 0); 2020 err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred); 2021 if (err) { 2022 goto out; 2023 } 2024 if (strnlen(fdi.answ, fdi.iosize) + 1 < fdi.iosize) { 2025 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp)); 2026 fuse_warn(data, FSESS_WARN_READLINK_EMBEDDED_NUL, 2027 "Returned an embedded NUL from FUSE_READLINK."); 2028 err = EIO; 2029 goto out; 2030 } 2031 if (((char *)fdi.answ)[0] == '/' && 2032 fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) { 2033 char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname; 2034 2035 err = uiomove(mpth, strlen(mpth), uio); 2036 } 2037 if (!err) { 2038 err = uiomove(fdi.answ, fdi.iosize, uio); 2039 } 2040 out: 2041 fdisp_destroy(&fdi); 2042 return err; 2043 } 2044 2045 /* 2046 struct vnop_reclaim_args { 2047 struct vnode *a_vp; 2048 }; 2049 */ 2050 static int 2051 fuse_vnop_reclaim(struct vop_reclaim_args *ap) 2052 { 2053 struct vnode *vp = ap->a_vp; 2054 struct thread *td = curthread; 2055 struct fuse_vnode_data *fvdat = VTOFUD(vp); 2056 struct fuse_filehandle *fufh, *fufh_tmp; 2057 2058 if (!fvdat) { 2059 panic("FUSE: no vnode data during recycling"); 2060 } 2061 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) { 2062 printf("FUSE: vnode being reclaimed with open fufh " 2063 "(type=%#x)", fufh->fufh_type); 2064 fuse_filehandle_close(vp, fufh, td, NULL); 2065 } 2066 2067 if (VTOI(vp) == 1) { 2068 /* 2069 * Don't send FUSE_FORGET for the root inode, because 2070 * we never send FUSE_LOOKUP for it (see 2071 * fuse_vfsop_root) and we don't want the server to see 2072 * mismatched lookup counts. 2073 */ 2074 struct fuse_data *data; 2075 struct vnode *vroot; 2076 2077 data = fuse_get_mpdata(vnode_mount(vp)); 2078 FUSE_LOCK(); 2079 vroot = data->vroot; 2080 data->vroot = NULL; 2081 FUSE_UNLOCK(); 2082 if (vroot) 2083 vrele(vroot); 2084 } else if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) { 2085 fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp), 2086 fvdat->nlookup); 2087 } 2088 cache_purge(vp); 2089 vfs_hash_remove(vp); 2090 fuse_vnode_destroy(vp); 2091 2092 return 0; 2093 } 2094 2095 /* 2096 struct vnop_remove_args { 2097 struct vnode *a_dvp; 2098 struct vnode *a_vp; 2099 struct componentname *a_cnp; 2100 }; 2101 */ 2102 static int 2103 fuse_vnop_remove(struct vop_remove_args *ap) 2104 { 2105 struct vnode *dvp = ap->a_dvp; 2106 struct vnode *vp = ap->a_vp; 2107 struct componentname *cnp = ap->a_cnp; 2108 2109 int err; 2110 2111 if (fuse_isdeadfs(vp)) { 2112 return ENXIO; 2113 } 2114 if (vnode_isdir(vp)) { 2115 return EPERM; 2116 } 2117 2118 err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK); 2119 2120 return err; 2121 } 2122 2123 /* 2124 struct vnop_rename_args { 2125 struct vnode *a_fdvp; 2126 struct vnode *a_fvp; 2127 struct componentname *a_fcnp; 2128 struct vnode *a_tdvp; 2129 struct vnode *a_tvp; 2130 struct componentname *a_tcnp; 2131 }; 2132 */ 2133 static int 2134 fuse_vnop_rename(struct vop_rename_args *ap) 2135 { 2136 struct vnode *fdvp = ap->a_fdvp; 2137 struct vnode *fvp = ap->a_fvp; 2138 struct componentname *fcnp = ap->a_fcnp; 2139 struct vnode *tdvp = ap->a_tdvp; 2140 struct vnode *tvp = ap->a_tvp; 2141 struct componentname *tcnp = ap->a_tcnp; 2142 struct fuse_data *data; 2143 bool newparent = fdvp != tdvp; 2144 bool isdir = fvp->v_type == VDIR; 2145 int err = 0; 2146 2147 if (fuse_isdeadfs(fdvp)) { 2148 return ENXIO; 2149 } 2150 if (fvp->v_mount != tdvp->v_mount || 2151 (tvp && fvp->v_mount != tvp->v_mount)) { 2152 SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename"); 2153 err = EXDEV; 2154 goto out; 2155 } 2156 cache_purge(fvp); 2157 2158 /* 2159 * FUSE library is expected to check if target directory is not 2160 * under the source directory in the file system tree. 2161 * Linux performs this check at VFS level. 2162 */ 2163 /* 2164 * If source is a directory, and it will get a new parent, user must 2165 * have write permission to it, so ".." can be modified. 2166 */ 2167 data = fuse_get_mpdata(vnode_mount(tdvp)); 2168 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) { 2169 err = fuse_internal_access(fvp, VWRITE, 2170 curthread, tcnp->cn_cred); 2171 if (err) 2172 goto out; 2173 } 2174 sx_xlock(&data->rename_lock); 2175 err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp); 2176 if (err == 0) { 2177 if (tdvp != fdvp) 2178 fuse_vnode_setparent(fvp, tdvp); 2179 if (tvp != NULL) 2180 fuse_vnode_setparent(tvp, NULL); 2181 } 2182 sx_unlock(&data->rename_lock); 2183 2184 if (tvp != NULL && tvp != fvp) { 2185 cache_purge(tvp); 2186 } 2187 if (vnode_isdir(fvp)) { 2188 if (((tvp != NULL) && vnode_isdir(tvp)) || vnode_isdir(fvp)) { 2189 cache_purge(tdvp); 2190 } 2191 cache_purge(fdvp); 2192 } 2193 out: 2194 if (tdvp == tvp) { 2195 vrele(tdvp); 2196 } else { 2197 vput(tdvp); 2198 } 2199 if (tvp != NULL) { 2200 vput(tvp); 2201 } 2202 vrele(fdvp); 2203 vrele(fvp); 2204 2205 return err; 2206 } 2207 2208 /* 2209 struct vnop_rmdir_args { 2210 struct vnode *a_dvp; 2211 struct vnode *a_vp; 2212 struct componentname *a_cnp; 2213 } *ap; 2214 */ 2215 static int 2216 fuse_vnop_rmdir(struct vop_rmdir_args *ap) 2217 { 2218 struct vnode *dvp = ap->a_dvp; 2219 struct vnode *vp = ap->a_vp; 2220 2221 int err; 2222 2223 if (fuse_isdeadfs(vp)) { 2224 return ENXIO; 2225 } 2226 if (VTOFUD(vp) == VTOFUD(dvp)) { 2227 return EINVAL; 2228 } 2229 err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR); 2230 2231 return err; 2232 } 2233 2234 /* 2235 struct vnop_setattr_args { 2236 struct vnode *a_vp; 2237 struct vattr *a_vap; 2238 struct ucred *a_cred; 2239 struct thread *a_td; 2240 }; 2241 */ 2242 static int 2243 fuse_vnop_setattr(struct vop_setattr_args *ap) 2244 { 2245 struct vnode *vp = ap->a_vp; 2246 struct vattr *vap = ap->a_vap; 2247 struct ucred *cred = ap->a_cred; 2248 struct thread *td = curthread; 2249 struct mount *mp; 2250 struct fuse_data *data; 2251 struct vattr old_va; 2252 int dataflags; 2253 int err = 0, err2; 2254 accmode_t accmode = 0; 2255 bool checkperm; 2256 bool drop_suid = false; 2257 2258 mp = vnode_mount(vp); 2259 data = fuse_get_mpdata(mp); 2260 dataflags = data->dataflags; 2261 checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS; 2262 2263 if (fuse_isdeadfs(vp)) { 2264 return ENXIO; 2265 } 2266 2267 if (vap->va_uid != (uid_t)VNOVAL) { 2268 if (checkperm) { 2269 /* Only root may change a file's owner */ 2270 err = priv_check_cred(cred, PRIV_VFS_CHOWN); 2271 if (err) { 2272 /* As a special case, allow the null chown */ 2273 err2 = fuse_internal_getattr(vp, &old_va, cred, 2274 td); 2275 if (err2) 2276 return (err2); 2277 if (vap->va_uid != old_va.va_uid) 2278 return err; 2279 drop_suid = true; 2280 } 2281 } 2282 accmode |= VADMIN; 2283 } 2284 if (vap->va_gid != (gid_t)VNOVAL) { 2285 if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN)) 2286 drop_suid = true; 2287 if (checkperm && !groupmember(vap->va_gid, cred)) { 2288 /* 2289 * Non-root users may only chgrp to one of their own 2290 * groups 2291 */ 2292 err = priv_check_cred(cred, PRIV_VFS_CHOWN); 2293 if (err) { 2294 /* As a special case, allow the null chgrp */ 2295 err2 = fuse_internal_getattr(vp, &old_va, cred, 2296 td); 2297 if (err2) 2298 return (err2); 2299 if (vap->va_gid != old_va.va_gid) 2300 return err; 2301 } 2302 } 2303 accmode |= VADMIN; 2304 } 2305 if (vap->va_size != VNOVAL) { 2306 switch (vp->v_type) { 2307 case VDIR: 2308 return (EISDIR); 2309 case VLNK: 2310 case VREG: 2311 if (vfs_isrdonly(mp)) 2312 return (EROFS); 2313 err = vn_rlimit_trunc(vap->va_size, td); 2314 if (err) 2315 return (err); 2316 break; 2317 default: 2318 /* 2319 * According to POSIX, the result is unspecified 2320 * for file types other than regular files, 2321 * directories and shared memory objects. We 2322 * don't support shared memory objects in the file 2323 * system, and have dubious support for truncating 2324 * symlinks. Just ignore the request in other cases. 2325 */ 2326 return (0); 2327 } 2328 /* Don't set accmode. Permission to trunc is checked upstack */ 2329 } 2330 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 2331 if (vap->va_vaflags & VA_UTIMES_NULL) 2332 accmode |= VWRITE; 2333 else 2334 accmode |= VADMIN; 2335 } 2336 if (drop_suid) { 2337 if (vap->va_mode != (mode_t)VNOVAL) 2338 vap->va_mode &= ~(S_ISUID | S_ISGID); 2339 else { 2340 err = fuse_internal_getattr(vp, &old_va, cred, td); 2341 if (err) 2342 return (err); 2343 vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID); 2344 } 2345 } 2346 if (vap->va_mode != (mode_t)VNOVAL) { 2347 /* Only root may set the sticky bit on non-directories */ 2348 if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT) 2349 && priv_check_cred(cred, PRIV_VFS_STICKYFILE)) 2350 return EFTYPE; 2351 if (checkperm && (vap->va_mode & S_ISGID)) { 2352 err = fuse_internal_getattr(vp, &old_va, cred, td); 2353 if (err) 2354 return (err); 2355 if (!groupmember(old_va.va_gid, cred)) { 2356 err = priv_check_cred(cred, PRIV_VFS_SETGID); 2357 if (err) 2358 return (err); 2359 } 2360 } 2361 accmode |= VADMIN; 2362 } 2363 2364 if (vfs_isrdonly(mp)) 2365 return EROFS; 2366 2367 if (checkperm) { 2368 err = fuse_internal_access(vp, accmode, td, cred); 2369 } else { 2370 err = 0; 2371 } 2372 if (err) 2373 return err; 2374 else 2375 return fuse_internal_setattr(vp, vap, td, cred); 2376 } 2377 2378 /* 2379 struct vnop_strategy_args { 2380 struct vnode *a_vp; 2381 struct buf *a_bp; 2382 }; 2383 */ 2384 static int 2385 fuse_vnop_strategy(struct vop_strategy_args *ap) 2386 { 2387 struct vnode *vp = ap->a_vp; 2388 struct buf *bp = ap->a_bp; 2389 2390 if (!vp || fuse_isdeadfs(vp)) { 2391 bp->b_ioflags |= BIO_ERROR; 2392 bp->b_error = ENXIO; 2393 bufdone(bp); 2394 return 0; 2395 } 2396 2397 /* 2398 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags. 2399 * fuse_io_strategy sets bp's error fields 2400 */ 2401 (void)fuse_io_strategy(vp, bp); 2402 2403 return 0; 2404 } 2405 2406 /* 2407 struct vnop_symlink_args { 2408 struct vnode *a_dvp; 2409 struct vnode **a_vpp; 2410 struct componentname *a_cnp; 2411 struct vattr *a_vap; 2412 char *a_target; 2413 }; 2414 */ 2415 static int 2416 fuse_vnop_symlink(struct vop_symlink_args *ap) 2417 { 2418 struct vnode *dvp = ap->a_dvp; 2419 struct vnode **vpp = ap->a_vpp; 2420 struct componentname *cnp = ap->a_cnp; 2421 const char *target = ap->a_target; 2422 2423 struct fuse_dispatcher fdi; 2424 2425 int err; 2426 size_t len; 2427 2428 if (fuse_isdeadfs(dvp)) { 2429 return ENXIO; 2430 } 2431 /* 2432 * Unlike the other creator type calls, here we have to create a message 2433 * where the name of the new entry comes first, and the data describing 2434 * the entry comes second. 2435 * Hence we can't rely on our handy fuse_internal_newentry() routine, 2436 * but put together the message manually and just call the core part. 2437 */ 2438 2439 len = strlen(target) + 1; 2440 fdisp_init(&fdi, len + cnp->cn_namelen + 1); 2441 fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL); 2442 2443 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen); 2444 ((char *)fdi.indata)[cnp->cn_namelen] = '\0'; 2445 memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len); 2446 2447 err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi); 2448 fdisp_destroy(&fdi); 2449 return err; 2450 } 2451 2452 /* 2453 struct vnop_write_args { 2454 struct vnode *a_vp; 2455 struct uio *a_uio; 2456 int a_ioflag; 2457 struct ucred *a_cred; 2458 }; 2459 */ 2460 static int 2461 fuse_vnop_write(struct vop_write_args *ap) 2462 { 2463 struct vnode *vp = ap->a_vp; 2464 struct uio *uio = ap->a_uio; 2465 int ioflag = ap->a_ioflag; 2466 struct ucred *cred = ap->a_cred; 2467 pid_t pid = curthread->td_proc->p_pid; 2468 struct fuse_filehandle *fufh; 2469 int err; 2470 bool closefufh = false, directio; 2471 2472 MPASS(vp->v_type == VREG || vp->v_type == VDIR); 2473 2474 if (fuse_isdeadfs(vp)) { 2475 return ENXIO; 2476 } 2477 2478 if (VTOFUD(vp)->flag & FN_DIRECTIO) { 2479 ioflag |= IO_DIRECT; 2480 } 2481 2482 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); 2483 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) { 2484 /* 2485 * nfsd will do I/O without first doing VOP_OPEN. We 2486 * must implicitly open the file here 2487 */ 2488 err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred); 2489 closefufh = true; 2490 } 2491 if (err) { 2492 SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred); 2493 return err; 2494 } 2495 2496 /* 2497 * Ideally, when the daemon asks for direct io at open time, the 2498 * standard file flag should be set according to this, so that would 2499 * just change the default mode, which later on could be changed via 2500 * fcntl(2). 2501 * But this doesn't work, the O_DIRECT flag gets cleared at some point 2502 * (don't know where). So to make any use of the Fuse direct_io option, 2503 * we hardwire it into the file's private data (similarly to Linux, 2504 * btw.). 2505 */ 2506 directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp)); 2507 2508 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE); 2509 if (directio) { 2510 off_t start, end, filesize; 2511 bool pages = (ioflag & IO_VMIO) != 0; 2512 2513 SDT_PROBE2(fusefs, , vnops, trace, 1, "direct write of vnode"); 2514 2515 err = fuse_vnode_size(vp, &filesize, cred, curthread); 2516 if (err) 2517 goto out; 2518 2519 start = uio->uio_offset; 2520 end = start + uio->uio_resid; 2521 if (!pages) { 2522 err = fuse_inval_buf_range(vp, filesize, start, 2523 end); 2524 if (err) 2525 goto out; 2526 } 2527 err = fuse_write_directbackend(vp, uio, cred, fufh, 2528 filesize, ioflag, pages); 2529 } else { 2530 SDT_PROBE2(fusefs, , vnops, trace, 1, 2531 "buffered write of vnode"); 2532 if (!fsess_opt_writeback(vnode_mount(vp))) 2533 ioflag |= IO_SYNC; 2534 err = fuse_write_biobackend(vp, uio, cred, fufh, ioflag, pid); 2535 } 2536 fuse_internal_clear_suid_on_write(vp, cred, uio->uio_td); 2537 2538 out: 2539 if (closefufh) 2540 fuse_filehandle_close(vp, fufh, curthread, cred); 2541 2542 return (err); 2543 } 2544 2545 static daddr_t 2546 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off) 2547 { 2548 const int biosize = fuse_iosize(vp); 2549 2550 return (off / biosize); 2551 } 2552 2553 static int 2554 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *blksz) 2555 { 2556 off_t filesize; 2557 int err; 2558 const int biosize = fuse_iosize(vp); 2559 2560 err = fuse_vnode_size(vp, &filesize, NULL, NULL); 2561 if (err) { 2562 /* This will turn into a SIGBUS */ 2563 return (EIO); 2564 } else if ((off_t)lbn * biosize >= filesize) { 2565 *blksz = 0; 2566 } else if ((off_t)(lbn + 1) * biosize > filesize) { 2567 *blksz = filesize - (off_t)lbn *biosize; 2568 } else { 2569 *blksz = biosize; 2570 } 2571 return (0); 2572 } 2573 2574 /* 2575 struct vnop_getpages_args { 2576 struct vnode *a_vp; 2577 vm_page_t *a_m; 2578 int a_count; 2579 int a_reqpage; 2580 }; 2581 */ 2582 static int 2583 fuse_vnop_getpages(struct vop_getpages_args *ap) 2584 { 2585 struct vnode *vp = ap->a_vp; 2586 2587 if (!fsess_opt_mmap(vnode_mount(vp))) { 2588 SDT_PROBE2(fusefs, , vnops, trace, 1, 2589 "called on non-cacheable vnode??\n"); 2590 return (VM_PAGER_ERROR); 2591 } 2592 2593 return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind, 2594 ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz)); 2595 } 2596 2597 static const char extattr_namespace_separator = '.'; 2598 2599 /* 2600 struct vop_getextattr_args { 2601 struct vop_generic_args a_gen; 2602 struct vnode *a_vp; 2603 int a_attrnamespace; 2604 const char *a_name; 2605 struct uio *a_uio; 2606 size_t *a_size; 2607 struct ucred *a_cred; 2608 struct thread *a_td; 2609 }; 2610 */ 2611 static int 2612 fuse_vnop_getextattr(struct vop_getextattr_args *ap) 2613 { 2614 struct vnode *vp = ap->a_vp; 2615 struct uio *uio = ap->a_uio; 2616 struct fuse_dispatcher fdi; 2617 struct fuse_getxattr_in *get_xattr_in; 2618 struct fuse_getxattr_out *get_xattr_out; 2619 struct mount *mp = vnode_mount(vp); 2620 struct thread *td = ap->a_td; 2621 struct ucred *cred = ap->a_cred; 2622 char *prefix; 2623 char *attr_str; 2624 size_t len; 2625 int err; 2626 2627 if (fuse_isdeadfs(vp)) 2628 return (ENXIO); 2629 2630 if (fsess_not_impl(mp, FUSE_GETXATTR)) 2631 return EOPNOTSUPP; 2632 2633 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD); 2634 if (err) 2635 return err; 2636 2637 /* Default to looking for user attributes. */ 2638 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) 2639 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; 2640 else 2641 prefix = EXTATTR_NAMESPACE_USER_STRING; 2642 2643 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 2644 strlen(ap->a_name) + 1; 2645 2646 fdisp_init(&fdi, len + sizeof(*get_xattr_in)); 2647 fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred); 2648 2649 get_xattr_in = fdi.indata; 2650 /* 2651 * Check to see whether we're querying the available size or 2652 * issuing the actual request. If we pass in 0, we get back struct 2653 * fuse_getxattr_out. If we pass in a non-zero size, we get back 2654 * that much data, without the struct fuse_getxattr_out header. 2655 */ 2656 if (uio == NULL) 2657 get_xattr_in->size = 0; 2658 else 2659 get_xattr_in->size = uio->uio_resid; 2660 2661 attr_str = (char *)fdi.indata + sizeof(*get_xattr_in); 2662 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator, 2663 ap->a_name); 2664 2665 err = fdisp_wait_answ(&fdi); 2666 if (err != 0) { 2667 if (err == ENOSYS) { 2668 fsess_set_notimpl(mp, FUSE_GETXATTR); 2669 err = EOPNOTSUPP; 2670 } 2671 goto out; 2672 } 2673 2674 get_xattr_out = fdi.answ; 2675 2676 if (ap->a_size != NULL) 2677 *ap->a_size = get_xattr_out->size; 2678 2679 if (uio != NULL) 2680 err = uiomove(fdi.answ, fdi.iosize, uio); 2681 2682 out: 2683 fdisp_destroy(&fdi); 2684 return (err); 2685 } 2686 2687 /* 2688 struct vop_setextattr_args { 2689 struct vop_generic_args a_gen; 2690 struct vnode *a_vp; 2691 int a_attrnamespace; 2692 const char *a_name; 2693 struct uio *a_uio; 2694 struct ucred *a_cred; 2695 struct thread *a_td; 2696 }; 2697 */ 2698 static int 2699 fuse_vnop_setextattr(struct vop_setextattr_args *ap) 2700 { 2701 struct vnode *vp = ap->a_vp; 2702 struct uio *uio = ap->a_uio; 2703 struct fuse_dispatcher fdi; 2704 struct fuse_setxattr_in *set_xattr_in; 2705 struct mount *mp = vnode_mount(vp); 2706 struct thread *td = ap->a_td; 2707 struct ucred *cred = ap->a_cred; 2708 char *prefix; 2709 size_t len; 2710 char *attr_str; 2711 int err; 2712 2713 if (fuse_isdeadfs(vp)) 2714 return (ENXIO); 2715 2716 if (fsess_not_impl(mp, FUSE_SETXATTR)) 2717 return EOPNOTSUPP; 2718 2719 if (vfs_isrdonly(mp)) 2720 return EROFS; 2721 2722 /* Deleting xattrs must use VOP_DELETEEXTATTR instead */ 2723 if (ap->a_uio == NULL) { 2724 /* 2725 * If we got here as fallback from VOP_DELETEEXTATTR, then 2726 * return EOPNOTSUPP. 2727 */ 2728 if (fsess_not_impl(mp, FUSE_REMOVEXATTR)) 2729 return (EOPNOTSUPP); 2730 else 2731 return (EINVAL); 2732 } 2733 2734 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, 2735 VWRITE); 2736 if (err) 2737 return err; 2738 2739 /* Default to looking for user attributes. */ 2740 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) 2741 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; 2742 else 2743 prefix = EXTATTR_NAMESPACE_USER_STRING; 2744 2745 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 2746 strlen(ap->a_name) + 1; 2747 2748 fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid); 2749 fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred); 2750 2751 set_xattr_in = fdi.indata; 2752 set_xattr_in->size = uio->uio_resid; 2753 2754 attr_str = (char *)fdi.indata + sizeof(*set_xattr_in); 2755 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator, 2756 ap->a_name); 2757 2758 err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len, 2759 uio->uio_resid, uio); 2760 if (err != 0) { 2761 goto out; 2762 } 2763 2764 err = fdisp_wait_answ(&fdi); 2765 2766 if (err == ENOSYS) { 2767 fsess_set_notimpl(mp, FUSE_SETXATTR); 2768 err = EOPNOTSUPP; 2769 } 2770 if (err == ERESTART) { 2771 /* Can't restart after calling uiomove */ 2772 err = EINTR; 2773 } 2774 2775 out: 2776 fdisp_destroy(&fdi); 2777 return (err); 2778 } 2779 2780 /* 2781 * The Linux / FUSE extended attribute list is simply a collection of 2782 * NUL-terminated strings. The FreeBSD extended attribute list is a single 2783 * byte length followed by a non-NUL terminated string. So, this allows 2784 * conversion of the Linux / FUSE format to the FreeBSD format in place. 2785 * Linux attribute names are reported with the namespace as a prefix (e.g. 2786 * "user.attribute_name"), but in FreeBSD they are reported without the 2787 * namespace prefix (e.g. "attribute_name"). So, we're going from: 2788 * 2789 * user.attr_name1\0user.attr_name2\0 2790 * 2791 * to: 2792 * 2793 * <num>attr_name1<num>attr_name2 2794 * 2795 * Where "<num>" is a single byte number of characters in the attribute name. 2796 * 2797 * Args: 2798 * prefix - exattr namespace prefix string 2799 * list, list_len - input list with namespace prefixes 2800 * bsd_list, bsd_list_len - output list compatible with bsd vfs 2801 */ 2802 static int 2803 fuse_xattrlist_convert(char *prefix, const char *list, int list_len, 2804 char *bsd_list, int *bsd_list_len) 2805 { 2806 int len, pos, dist_to_next, prefix_len; 2807 2808 pos = 0; 2809 *bsd_list_len = 0; 2810 prefix_len = strlen(prefix); 2811 2812 while (pos < list_len && list[pos] != '\0') { 2813 dist_to_next = strlen(&list[pos]) + 1; 2814 if (bcmp(&list[pos], prefix, prefix_len) == 0 && 2815 list[pos + prefix_len] == extattr_namespace_separator) { 2816 len = dist_to_next - 2817 (prefix_len + sizeof(extattr_namespace_separator)) - 1; 2818 if (len >= EXTATTR_MAXNAMELEN) 2819 return (ENAMETOOLONG); 2820 2821 bsd_list[*bsd_list_len] = len; 2822 memcpy(&bsd_list[*bsd_list_len + 1], 2823 &list[pos + prefix_len + 2824 sizeof(extattr_namespace_separator)], len); 2825 2826 *bsd_list_len += len + 1; 2827 } 2828 2829 pos += dist_to_next; 2830 } 2831 2832 return (0); 2833 } 2834 2835 /* 2836 * List extended attributes 2837 * 2838 * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which 2839 * has a number of differences compared to its FreeBSD equivalent, 2840 * extattr_list_file: 2841 * 2842 * - FUSE_LISTXATTR returns all extended attributes across all namespaces, 2843 * whereas listxattr(2) only returns attributes for a single namespace 2844 * - FUSE_LISTXATTR prepends each attribute name with "namespace." 2845 * - If the provided buffer is not large enough to hold the result, 2846 * FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to 2847 * return as many results as will fit. 2848 */ 2849 /* 2850 struct vop_listextattr_args { 2851 struct vop_generic_args a_gen; 2852 struct vnode *a_vp; 2853 int a_attrnamespace; 2854 struct uio *a_uio; 2855 size_t *a_size; 2856 struct ucred *a_cred; 2857 struct thread *a_td; 2858 }; 2859 */ 2860 static int 2861 fuse_vnop_listextattr(struct vop_listextattr_args *ap) 2862 { 2863 struct vnode *vp = ap->a_vp; 2864 struct uio *uio = ap->a_uio; 2865 struct fuse_dispatcher fdi; 2866 struct fuse_listxattr_in *list_xattr_in; 2867 struct fuse_listxattr_out *list_xattr_out; 2868 struct mount *mp = vnode_mount(vp); 2869 struct thread *td = ap->a_td; 2870 struct ucred *cred = ap->a_cred; 2871 char *prefix; 2872 char *bsd_list = NULL; 2873 char *linux_list; 2874 int bsd_list_len; 2875 int linux_list_len; 2876 int err; 2877 2878 if (fuse_isdeadfs(vp)) 2879 return (ENXIO); 2880 2881 if (fsess_not_impl(mp, FUSE_LISTXATTR)) 2882 return EOPNOTSUPP; 2883 2884 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD); 2885 if (err) 2886 return err; 2887 2888 /* 2889 * Add space for a NUL and the period separator if enabled. 2890 * Default to looking for user attributes. 2891 */ 2892 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) 2893 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; 2894 else 2895 prefix = EXTATTR_NAMESPACE_USER_STRING; 2896 2897 fdisp_init(&fdi, sizeof(*list_xattr_in)); 2898 fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred); 2899 2900 /* 2901 * Retrieve Linux / FUSE compatible list size. 2902 */ 2903 list_xattr_in = fdi.indata; 2904 list_xattr_in->size = 0; 2905 2906 err = fdisp_wait_answ(&fdi); 2907 if (err != 0) { 2908 if (err == ENOSYS) { 2909 fsess_set_notimpl(mp, FUSE_LISTXATTR); 2910 err = EOPNOTSUPP; 2911 } 2912 goto out; 2913 } 2914 2915 list_xattr_out = fdi.answ; 2916 linux_list_len = list_xattr_out->size; 2917 if (linux_list_len == 0) { 2918 if (ap->a_size != NULL) 2919 *ap->a_size = linux_list_len; 2920 goto out; 2921 } 2922 2923 /* 2924 * Retrieve Linux / FUSE compatible list values. 2925 */ 2926 fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred); 2927 list_xattr_in = fdi.indata; 2928 list_xattr_in->size = linux_list_len; 2929 2930 err = fdisp_wait_answ(&fdi); 2931 if (err == ERANGE) { 2932 /* 2933 * Race detected. The attribute list must've grown since the 2934 * first FUSE_LISTXATTR call. Start over. Go all the way back 2935 * to userland so we can process signals, if necessary, before 2936 * restarting. 2937 */ 2938 err = ERESTART; 2939 goto out; 2940 } else if (err != 0) 2941 goto out; 2942 2943 linux_list = fdi.answ; 2944 /* FUSE doesn't allow the server to return more data than requested */ 2945 if (fdi.iosize > linux_list_len) { 2946 struct fuse_data *data = fuse_get_mpdata(mp); 2947 2948 fuse_warn(data, FSESS_WARN_LSEXTATTR_LONG, 2949 "server returned " 2950 "more extended attribute data than requested; " 2951 "should've returned ERANGE instead."); 2952 } else { 2953 /* But returning less data is fine */ 2954 linux_list_len = fdi.iosize; 2955 } 2956 2957 /* 2958 * Retrieve the BSD compatible list values. 2959 * The Linux / FUSE attribute list format isn't the same 2960 * as FreeBSD's format. So we need to transform it into 2961 * FreeBSD's format before giving it to the user. 2962 */ 2963 bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK); 2964 err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len, 2965 bsd_list, &bsd_list_len); 2966 if (err != 0) 2967 goto out; 2968 2969 if (ap->a_size != NULL) 2970 *ap->a_size = bsd_list_len; 2971 2972 if (uio != NULL) 2973 err = uiomove(bsd_list, bsd_list_len, uio); 2974 2975 out: 2976 free(bsd_list, M_TEMP); 2977 fdisp_destroy(&fdi); 2978 return (err); 2979 } 2980 2981 /* 2982 struct vop_deallocate_args { 2983 struct vop_generic_args a_gen; 2984 struct vnode *a_vp; 2985 off_t *a_offset; 2986 off_t *a_len; 2987 int a_flags; 2988 int a_ioflag; 2989 struct ucred *a_cred; 2990 }; 2991 */ 2992 static int 2993 fuse_vnop_deallocate(struct vop_deallocate_args *ap) 2994 { 2995 struct vnode *vp = ap->a_vp; 2996 struct mount *mp = vnode_mount(vp); 2997 struct fuse_filehandle *fufh; 2998 struct fuse_dispatcher fdi; 2999 struct fuse_fallocate_in *ffi; 3000 struct ucred *cred = ap->a_cred; 3001 pid_t pid = curthread->td_proc->p_pid; 3002 off_t *len = ap->a_len; 3003 off_t *offset = ap->a_offset; 3004 int ioflag = ap->a_ioflag; 3005 off_t filesize; 3006 int err; 3007 bool closefufh = false; 3008 3009 if (fuse_isdeadfs(vp)) 3010 return (ENXIO); 3011 3012 if (vfs_isrdonly(mp)) 3013 return (EROFS); 3014 3015 if (fsess_not_impl(mp, FUSE_FALLOCATE)) 3016 goto fallback; 3017 3018 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); 3019 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) { 3020 /* 3021 * nfsd will do I/O without first doing VOP_OPEN. We 3022 * must implicitly open the file here 3023 */ 3024 err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred); 3025 closefufh = true; 3026 } 3027 if (err) 3028 return (err); 3029 3030 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE); 3031 3032 err = fuse_vnode_size(vp, &filesize, cred, curthread); 3033 if (err) 3034 goto out; 3035 fuse_inval_buf_range(vp, filesize, *offset, *offset + *len); 3036 3037 fdisp_init(&fdi, sizeof(*ffi)); 3038 fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred); 3039 ffi = fdi.indata; 3040 ffi->fh = fufh->fh_id; 3041 ffi->offset = *offset; 3042 ffi->length = *len; 3043 /* 3044 * FreeBSD's fspacectl is equivalent to Linux's fallocate with 3045 * mode == FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE 3046 */ 3047 ffi->mode = FUSE_FALLOC_FL_PUNCH_HOLE | FUSE_FALLOC_FL_KEEP_SIZE; 3048 err = fdisp_wait_answ(&fdi); 3049 3050 if (err == ENOSYS) { 3051 fdisp_destroy(&fdi); 3052 fsess_set_notimpl(mp, FUSE_FALLOCATE); 3053 goto fallback; 3054 } else if (err == EOPNOTSUPP) { 3055 /* 3056 * The file system server does not support FUSE_FALLOCATE with 3057 * the supplied mode for this particular file. 3058 */ 3059 fdisp_destroy(&fdi); 3060 goto fallback; 3061 } else if (!err) { 3062 /* 3063 * Clip the returned offset to EoF. Do it here rather than 3064 * before FUSE_FALLOCATE just in case the kernel's cached file 3065 * size is out of date. Unfortunately, FUSE does not return 3066 * any information about filesize from that operation. 3067 */ 3068 *offset = MIN(*offset + *len, filesize); 3069 *len = 0; 3070 fuse_vnode_undirty_cached_timestamps(vp, false); 3071 fuse_internal_clear_suid_on_write(vp, cred, curthread); 3072 3073 if (ioflag & IO_SYNC) 3074 err = fuse_internal_fsync(vp, curthread, MNT_WAIT, 3075 false); 3076 } 3077 3078 out: 3079 fdisp_destroy(&fdi); 3080 if (closefufh) 3081 fuse_filehandle_close(vp, fufh, curthread, cred); 3082 3083 return (err); 3084 3085 fallback: 3086 if (closefufh) 3087 fuse_filehandle_close(vp, fufh, curthread, cred); 3088 3089 return (vop_stddeallocate(ap)); 3090 } 3091 3092 /* 3093 struct vop_deleteextattr_args { 3094 struct vop_generic_args a_gen; 3095 struct vnode *a_vp; 3096 int a_attrnamespace; 3097 const char *a_name; 3098 struct ucred *a_cred; 3099 struct thread *a_td; 3100 }; 3101 */ 3102 static int 3103 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap) 3104 { 3105 struct vnode *vp = ap->a_vp; 3106 struct fuse_dispatcher fdi; 3107 struct mount *mp = vnode_mount(vp); 3108 struct thread *td = ap->a_td; 3109 struct ucred *cred = ap->a_cred; 3110 char *prefix; 3111 size_t len; 3112 char *attr_str; 3113 int err; 3114 3115 if (fuse_isdeadfs(vp)) 3116 return (ENXIO); 3117 3118 if (fsess_not_impl(mp, FUSE_REMOVEXATTR)) 3119 return EOPNOTSUPP; 3120 3121 if (vfs_isrdonly(mp)) 3122 return EROFS; 3123 3124 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, 3125 VWRITE); 3126 if (err) 3127 return err; 3128 3129 /* Default to looking for user attributes. */ 3130 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) 3131 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; 3132 else 3133 prefix = EXTATTR_NAMESPACE_USER_STRING; 3134 3135 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 3136 strlen(ap->a_name) + 1; 3137 3138 fdisp_init(&fdi, len); 3139 fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred); 3140 3141 attr_str = fdi.indata; 3142 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator, 3143 ap->a_name); 3144 3145 err = fdisp_wait_answ(&fdi); 3146 if (err == ENOSYS) { 3147 fsess_set_notimpl(mp, FUSE_REMOVEXATTR); 3148 err = EOPNOTSUPP; 3149 } 3150 3151 fdisp_destroy(&fdi); 3152 return (err); 3153 } 3154 3155 /* 3156 struct vnop_print_args { 3157 struct vnode *a_vp; 3158 }; 3159 */ 3160 static int 3161 fuse_vnop_print(struct vop_print_args *ap) 3162 { 3163 struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp); 3164 3165 printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n", 3166 (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid, 3167 (uintmax_t)fvdat->nlookup, 3168 fvdat->flag); 3169 3170 return 0; 3171 } 3172 3173 /* 3174 * Get an NFS filehandle for a FUSE file. 3175 * 3176 * This will only work for FUSE file systems that guarantee the uniqueness of 3177 * nodeid:generation, which most don't. 3178 */ 3179 /* 3180 vop_vptofh { 3181 IN struct vnode *a_vp; 3182 IN struct fid *a_fhp; 3183 }; 3184 */ 3185 static int 3186 fuse_vnop_vptofh(struct vop_vptofh_args *ap) 3187 { 3188 struct vnode *vp = ap->a_vp; 3189 struct fuse_vnode_data *fvdat = VTOFUD(vp); 3190 struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp); 3191 _Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid), 3192 "FUSE fid type is too big"); 3193 struct mount *mp = vnode_mount(vp); 3194 struct fuse_data *data = fuse_get_mpdata(mp); 3195 struct vattr va; 3196 int err; 3197 3198 if (!(data->dataflags & FSESS_EXPORT_SUPPORT)) { 3199 /* NFS requires lookups for "." and ".." */ 3200 SDT_PROBE2(fusefs, , vnops, trace, 1, 3201 "VOP_VPTOFH without FUSE_EXPORT_SUPPORT"); 3202 return EOPNOTSUPP; 3203 } 3204 if ((mp->mnt_flag & MNT_EXPORTED) && 3205 !(data->dataflags & FSESS_NO_OPENDIR_SUPPORT)) 3206 { 3207 /* 3208 * NFS is stateless, so nfsd must reopen a directory on every 3209 * call to VOP_READDIR, passing in the d_off field from the 3210 * final dirent of the previous invocation. But without 3211 * FUSE_NO_OPENDIR_SUPPORT, the FUSE protocol does not 3212 * guarantee that d_off will be valid after a directory is 3213 * closed and reopened. So prohibit exporting FUSE file 3214 * systems that don't set that flag. 3215 * 3216 * But userspace NFS servers don't have this problem. 3217 */ 3218 SDT_PROBE2(fusefs, , vnops, trace, 1, 3219 "VOP_VPTOFH without FUSE_NO_OPENDIR_SUPPORT"); 3220 return EOPNOTSUPP; 3221 } 3222 3223 err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread); 3224 if (err) 3225 return err; 3226 3227 /*ip = VTOI(ap->a_vp);*/ 3228 /*ufhp = (struct ufid *)ap->a_fhp;*/ 3229 fhp->len = sizeof(struct fuse_fid); 3230 fhp->nid = fvdat->nid; 3231 if (fvdat->generation <= UINT32_MAX) 3232 fhp->gen = fvdat->generation; 3233 else 3234 return EOVERFLOW; 3235 return (0); 3236 } 3237