1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2007-2009 Google Inc. and Amit Singh 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions are 9 * met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above 14 * copyright notice, this list of conditions and the following disclaimer 15 * in the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Google Inc. nor the names of its 18 * contributors may be used to endorse or promote products derived from 19 * this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Copyright (C) 2005 Csaba Henk. 34 * All rights reserved. 35 * 36 * Copyright (c) 2019 The FreeBSD Foundation 37 * 38 * Portions of this software were developed by BFF Storage Systems, LLC under 39 * sponsorship from the FreeBSD Foundation. 40 * 41 * Redistribution and use in source and binary forms, with or without 42 * modification, are permitted provided that the following conditions 43 * are met: 44 * 1. Redistributions of source code must retain the above copyright 45 * notice, this list of conditions and the following disclaimer. 46 * 2. Redistributions in binary form must reproduce the above copyright 47 * notice, this list of conditions and the following disclaimer in the 48 * documentation and/or other materials provided with the distribution. 49 * 50 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND 51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 53 * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE 54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 60 * SUCH DAMAGE. 61 */ 62 63 #include <sys/cdefs.h> 64 __FBSDID("$FreeBSD$"); 65 66 #include <sys/types.h> 67 #include <sys/systm.h> 68 #include <sys/counter.h> 69 #include <sys/module.h> 70 #include <sys/errno.h> 71 #include <sys/param.h> 72 #include <sys/kernel.h> 73 #include <sys/conf.h> 74 #include <sys/uio.h> 75 #include <sys/malloc.h> 76 #include <sys/queue.h> 77 #include <sys/lock.h> 78 #include <sys/sx.h> 79 #include <sys/mutex.h> 80 #include <sys/proc.h> 81 #include <sys/vnode.h> 82 #include <sys/namei.h> 83 #include <sys/mount.h> 84 #include <sys/sysctl.h> 85 #include <sys/fcntl.h> 86 #include <sys/priv.h> 87 #include <sys/buf.h> 88 #include <security/mac/mac_framework.h> 89 #include <vm/vm.h> 90 #include <vm/vm_extern.h> 91 92 #include "fuse.h" 93 #include "fuse_node.h" 94 #include "fuse_internal.h" 95 #include "fuse_io.h" 96 #include "fuse_ipc.h" 97 98 SDT_PROVIDER_DECLARE(fusefs); 99 /* 100 * Fuse trace probe: 101 * arg0: verbosity. Higher numbers give more verbose messages 102 * arg1: Textual message 103 */ 104 SDT_PROBE_DEFINE2(fusefs, , node, trace, "int", "char*"); 105 106 MALLOC_DEFINE(M_FUSEVN, "fuse_vnode", "fuse vnode private data"); 107 108 static int sysctl_fuse_cache_mode(SYSCTL_HANDLER_ARGS); 109 110 static counter_u64_t fuse_node_count; 111 112 SYSCTL_COUNTER_U64(_vfs_fusefs_stats, OID_AUTO, node_count, CTLFLAG_RD, 113 &fuse_node_count, "Count of FUSE vnodes"); 114 115 int fuse_data_cache_mode = FUSE_CACHE_WT; 116 117 /* 118 * DEPRECATED 119 * This sysctl is no longer needed as of fuse protocol 7.23. Individual 120 * servers can select the cache behavior they need for each mountpoint: 121 * - writethrough: the default 122 * - writeback: set FUSE_WRITEBACK_CACHE in fuse_init_out.flags 123 * - uncached: set FOPEN_DIRECT_IO for every file 124 * The sysctl is retained primarily for use by jails supporting older FUSE 125 * protocols. It may be removed entirely once FreeBSD 11.3 and 12.0 are EOL. 126 */ 127 SYSCTL_PROC(_vfs_fusefs, OID_AUTO, data_cache_mode, CTLTYPE_INT|CTLFLAG_RW, 128 &fuse_data_cache_mode, 0, sysctl_fuse_cache_mode, "I", 129 "Zero: disable caching of FUSE file data; One: write-through caching " 130 "(default); Two: write-back caching (generally unsafe)"); 131 132 static int 133 sysctl_fuse_cache_mode(SYSCTL_HANDLER_ARGS) 134 { 135 int val, error; 136 137 val = *(int *)arg1; 138 error = sysctl_handle_int(oidp, &val, 0, req); 139 if (error || !req->newptr) 140 return (error); 141 142 switch (val) { 143 case FUSE_CACHE_UC: 144 case FUSE_CACHE_WT: 145 case FUSE_CACHE_WB: 146 *(int *)arg1 = val; 147 break; 148 default: 149 return (EDOM); 150 } 151 return (0); 152 } 153 154 static void 155 fuse_vnode_init(struct vnode *vp, struct fuse_vnode_data *fvdat, 156 uint64_t nodeid, enum vtype vtyp) 157 { 158 fvdat->nid = nodeid; 159 LIST_INIT(&fvdat->handles); 160 vattr_null(&fvdat->cached_attrs); 161 if (nodeid == FUSE_ROOT_ID) { 162 vp->v_vflag |= VV_ROOT; 163 } 164 vp->v_type = vtyp; 165 vp->v_data = fvdat; 166 167 counter_u64_add(fuse_node_count, 1); 168 } 169 170 void 171 fuse_vnode_destroy(struct vnode *vp) 172 { 173 struct fuse_vnode_data *fvdat = vp->v_data; 174 175 vp->v_data = NULL; 176 KASSERT(LIST_EMPTY(&fvdat->handles), 177 ("Destroying fuse vnode with open files!")); 178 free(fvdat, M_FUSEVN); 179 180 counter_u64_add(fuse_node_count, -1); 181 } 182 183 int 184 fuse_vnode_cmp(struct vnode *vp, void *nidp) 185 { 186 return (VTOI(vp) != *((uint64_t *)nidp)); 187 } 188 189 SDT_PROBE_DEFINE3(fusefs, , node, stale_vnode, "struct vnode*", "enum vtype", 190 "uint64_t"); 191 static int 192 fuse_vnode_alloc(struct mount *mp, 193 struct thread *td, 194 uint64_t nodeid, 195 enum vtype vtyp, 196 struct vnode **vpp) 197 { 198 struct fuse_data *data; 199 struct fuse_vnode_data *fvdat; 200 struct vnode *vp2; 201 int err = 0; 202 203 data = fuse_get_mpdata(mp); 204 if (vtyp == VNON) { 205 return EINVAL; 206 } 207 *vpp = NULL; 208 err = vfs_hash_get(mp, fuse_vnode_hash(nodeid), LK_EXCLUSIVE, td, vpp, 209 fuse_vnode_cmp, &nodeid); 210 if (err) 211 return (err); 212 213 if (*vpp) { 214 if ((*vpp)->v_type != vtyp) { 215 /* 216 * STALE vnode! This probably indicates a buggy 217 * server, but it could also be the result of a race 218 * between FUSE_LOOKUP and another client's 219 * FUSE_UNLINK/FUSE_CREATE 220 */ 221 SDT_PROBE3(fusefs, , node, stale_vnode, *vpp, vtyp, 222 nodeid); 223 fuse_internal_vnode_disappear(*vpp); 224 lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL); 225 *vpp = NULL; 226 return (EAGAIN); 227 } 228 MPASS((*vpp)->v_data != NULL); 229 MPASS(VTOFUD(*vpp)->nid == nodeid); 230 SDT_PROBE2(fusefs, , node, trace, 1, "vnode taken from hash"); 231 return (0); 232 } 233 fvdat = malloc(sizeof(*fvdat), M_FUSEVN, M_WAITOK | M_ZERO); 234 switch (vtyp) { 235 case VFIFO: 236 err = getnewvnode("fuse", mp, &fuse_fifoops, vpp); 237 break; 238 default: 239 err = getnewvnode("fuse", mp, &fuse_vnops, vpp); 240 break; 241 } 242 if (err) { 243 free(fvdat, M_FUSEVN); 244 return (err); 245 } 246 lockmgr((*vpp)->v_vnlock, LK_EXCLUSIVE, NULL); 247 fuse_vnode_init(*vpp, fvdat, nodeid, vtyp); 248 err = insmntque(*vpp, mp); 249 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc"); 250 if (err) { 251 lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL); 252 free(fvdat, M_FUSEVN); 253 *vpp = NULL; 254 return (err); 255 } 256 /* Disallow async reads for fifos because UFS does. I don't know why */ 257 if (data->dataflags & FSESS_ASYNC_READ && vtyp != VFIFO) 258 VN_LOCK_ASHARE(*vpp); 259 260 err = vfs_hash_insert(*vpp, fuse_vnode_hash(nodeid), LK_EXCLUSIVE, 261 td, &vp2, fuse_vnode_cmp, &nodeid); 262 if (err) { 263 lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL); 264 free(fvdat, M_FUSEVN); 265 *vpp = NULL; 266 return (err); 267 } 268 if (vp2 != NULL) { 269 *vpp = vp2; 270 return (0); 271 } 272 273 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc"); 274 275 return (0); 276 } 277 278 int 279 fuse_vnode_get(struct mount *mp, 280 struct fuse_entry_out *feo, 281 uint64_t nodeid, 282 struct vnode *dvp, 283 struct vnode **vpp, 284 struct componentname *cnp, 285 enum vtype vtyp) 286 { 287 struct thread *td = (cnp != NULL ? cnp->cn_thread : curthread); 288 /* 289 * feo should only be NULL for the root directory, which (when libfuse 290 * is used) always has generation 0 291 */ 292 uint64_t generation = feo ? feo->generation : 0; 293 int err = 0; 294 295 err = fuse_vnode_alloc(mp, td, nodeid, vtyp, vpp); 296 if (err) { 297 return err; 298 } 299 if (dvp != NULL) { 300 MPASS(cnp && (cnp->cn_flags & ISDOTDOT) == 0); 301 MPASS(cnp && 302 !(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.')); 303 fuse_vnode_setparent(*vpp, dvp); 304 } 305 if (dvp != NULL && cnp != NULL && (cnp->cn_flags & MAKEENTRY) != 0 && 306 feo != NULL && 307 (feo->entry_valid != 0 || feo->entry_valid_nsec != 0)) { 308 struct timespec timeout; 309 310 ASSERT_VOP_LOCKED(*vpp, "fuse_vnode_get"); 311 ASSERT_VOP_LOCKED(dvp, "fuse_vnode_get"); 312 313 fuse_validity_2_timespec(feo, &timeout); 314 cache_enter_time(dvp, *vpp, cnp, &timeout, NULL); 315 } 316 317 VTOFUD(*vpp)->generation = generation; 318 /* 319 * In userland, libfuse uses cached lookups for dot and dotdot entries, 320 * thus it does not really bump the nlookup counter for forget. 321 * Follow the same semantic and avoid the bump in order to keep 322 * nlookup counters consistent. 323 */ 324 if (cnp == NULL || ((cnp->cn_flags & ISDOTDOT) == 0 && 325 (cnp->cn_namelen != 1 || cnp->cn_nameptr[0] != '.'))) 326 VTOFUD(*vpp)->nlookup++; 327 328 return 0; 329 } 330 331 /* 332 * Called for every fusefs vnode open to initialize the vnode (not 333 * fuse_filehandle) for use 334 */ 335 void 336 fuse_vnode_open(struct vnode *vp, int32_t fuse_open_flags, struct thread *td) 337 { 338 if (vnode_vtype(vp) == VREG) 339 vnode_create_vobject(vp, 0, td); 340 } 341 342 int 343 fuse_vnode_savesize(struct vnode *vp, struct ucred *cred, pid_t pid) 344 { 345 struct fuse_vnode_data *fvdat = VTOFUD(vp); 346 struct thread *td = curthread; 347 struct fuse_filehandle *fufh = NULL; 348 struct fuse_dispatcher fdi; 349 struct fuse_setattr_in *fsai; 350 int err = 0; 351 352 ASSERT_VOP_ELOCKED(vp, "fuse_io_extend"); 353 354 if (fuse_isdeadfs(vp)) { 355 return EBADF; 356 } 357 if (vnode_vtype(vp) == VDIR) { 358 return EISDIR; 359 } 360 if (vfs_isrdonly(vnode_mount(vp))) { 361 return EROFS; 362 } 363 if (cred == NULL) { 364 cred = td->td_ucred; 365 } 366 fdisp_init(&fdi, sizeof(*fsai)); 367 fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred); 368 fsai = fdi.indata; 369 fsai->valid = 0; 370 371 /* Truncate to a new value. */ 372 MPASS((fvdat->flag & FN_SIZECHANGE) != 0); 373 fsai->size = fvdat->cached_attrs.va_size; 374 fsai->valid |= FATTR_SIZE; 375 376 fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); 377 if (fufh) { 378 fsai->fh = fufh->fh_id; 379 fsai->valid |= FATTR_FH; 380 } 381 err = fdisp_wait_answ(&fdi); 382 fdisp_destroy(&fdi); 383 if (err == 0) 384 fvdat->flag &= ~FN_SIZECHANGE; 385 386 return err; 387 } 388 389 /* 390 * Adjust the vnode's size to a new value, such as that provided by 391 * FUSE_GETATTR. 392 */ 393 int 394 fuse_vnode_setsize(struct vnode *vp, off_t newsize) 395 { 396 struct fuse_vnode_data *fvdat = VTOFUD(vp); 397 struct vattr *attrs; 398 off_t oldsize; 399 size_t iosize; 400 struct buf *bp = NULL; 401 int err = 0; 402 403 ASSERT_VOP_ELOCKED(vp, "fuse_vnode_setsize"); 404 405 iosize = fuse_iosize(vp); 406 oldsize = fvdat->cached_attrs.va_size; 407 fvdat->cached_attrs.va_size = newsize; 408 if ((attrs = VTOVA(vp)) != NULL) 409 attrs->va_size = newsize; 410 411 if (newsize < oldsize) { 412 daddr_t lbn; 413 414 err = vtruncbuf(vp, newsize, fuse_iosize(vp)); 415 if (err) 416 goto out; 417 if (newsize % iosize == 0) 418 goto out; 419 /* 420 * Zero the contents of the last partial block. 421 * Sure seems like vtruncbuf should do this for us. 422 */ 423 424 lbn = newsize / iosize; 425 bp = getblk(vp, lbn, iosize, PCATCH, 0, 0); 426 if (!bp) { 427 err = EINTR; 428 goto out; 429 } 430 if (!(bp->b_flags & B_CACHE)) 431 goto out; /* Nothing to do */ 432 MPASS(bp->b_flags & B_VMIO); 433 vfs_bio_clrbuf(bp); 434 bp->b_dirtyend = MIN(bp->b_dirtyend, newsize - lbn * iosize); 435 } 436 out: 437 if (bp) 438 brelse(bp); 439 vnode_pager_setsize(vp, newsize); 440 return err; 441 } 442 443 /* Get the current, possibly dirty, size of the file */ 444 int 445 fuse_vnode_size(struct vnode *vp, off_t *filesize, struct ucred *cred, 446 struct thread *td) 447 { 448 struct fuse_vnode_data *fvdat = VTOFUD(vp); 449 int error = 0; 450 451 if (!(fvdat->flag & FN_SIZECHANGE) && 452 (VTOVA(vp) == NULL || fvdat->cached_attrs.va_size == VNOVAL)) 453 error = fuse_internal_do_getattr(vp, NULL, cred, td); 454 455 if (!error) 456 *filesize = fvdat->cached_attrs.va_size; 457 458 return error; 459 } 460 461 void 462 fuse_vnode_undirty_cached_timestamps(struct vnode *vp) 463 { 464 struct fuse_vnode_data *fvdat = VTOFUD(vp); 465 466 fvdat->flag &= ~(FN_MTIMECHANGE | FN_CTIMECHANGE); 467 } 468 469 /* Update a fuse file's cached timestamps */ 470 void 471 fuse_vnode_update(struct vnode *vp, int flags) 472 { 473 struct fuse_vnode_data *fvdat = VTOFUD(vp); 474 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp)); 475 struct timespec ts; 476 477 vfs_timestamp(&ts); 478 479 if (data->time_gran > 1) 480 ts.tv_nsec = rounddown(ts.tv_nsec, data->time_gran); 481 482 if (flags & FN_MTIMECHANGE) 483 fvdat->cached_attrs.va_mtime = ts; 484 if (flags & FN_CTIMECHANGE) 485 fvdat->cached_attrs.va_ctime = ts; 486 487 fvdat->flag |= flags; 488 } 489 490 void 491 fuse_node_init(void) 492 { 493 fuse_node_count = counter_u64_alloc(M_WAITOK); 494 } 495 496 void 497 fuse_node_destroy(void) 498 { 499 counter_u64_free(fuse_node_count); 500 } 501