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, 128 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 129 &fuse_data_cache_mode, 0, sysctl_fuse_cache_mode, "I", 130 "Zero: disable caching of FUSE file data; One: write-through caching " 131 "(default); Two: write-back caching (generally unsafe)"); 132 133 static int 134 sysctl_fuse_cache_mode(SYSCTL_HANDLER_ARGS) 135 { 136 int val, error; 137 138 val = *(int *)arg1; 139 error = sysctl_handle_int(oidp, &val, 0, req); 140 if (error || !req->newptr) 141 return (error); 142 143 switch (val) { 144 case FUSE_CACHE_UC: 145 case FUSE_CACHE_WT: 146 case FUSE_CACHE_WB: 147 *(int *)arg1 = val; 148 break; 149 default: 150 return (EDOM); 151 } 152 return (0); 153 } 154 155 static void 156 fuse_vnode_init(struct vnode *vp, struct fuse_vnode_data *fvdat, 157 uint64_t nodeid, enum vtype vtyp) 158 { 159 fvdat->nid = nodeid; 160 LIST_INIT(&fvdat->handles); 161 vattr_null(&fvdat->cached_attrs); 162 if (nodeid == FUSE_ROOT_ID) { 163 vp->v_vflag |= VV_ROOT; 164 } 165 vp->v_type = vtyp; 166 vp->v_data = fvdat; 167 cluster_init_vn(&fvdat->clusterw); 168 169 counter_u64_add(fuse_node_count, 1); 170 } 171 172 void 173 fuse_vnode_destroy(struct vnode *vp) 174 { 175 struct fuse_vnode_data *fvdat = vp->v_data; 176 177 vp->v_data = NULL; 178 KASSERT(LIST_EMPTY(&fvdat->handles), 179 ("Destroying fuse vnode with open files!")); 180 free(fvdat, M_FUSEVN); 181 182 counter_u64_add(fuse_node_count, -1); 183 } 184 185 int 186 fuse_vnode_cmp(struct vnode *vp, void *nidp) 187 { 188 return (VTOI(vp) != *((uint64_t *)nidp)); 189 } 190 191 SDT_PROBE_DEFINE3(fusefs, , node, stale_vnode, "struct vnode*", "enum vtype", 192 "uint64_t"); 193 static int 194 fuse_vnode_alloc(struct mount *mp, 195 struct thread *td, 196 uint64_t nodeid, 197 enum vtype vtyp, 198 struct vnode **vpp) 199 { 200 struct fuse_data *data; 201 struct fuse_vnode_data *fvdat; 202 struct vnode *vp2; 203 int err = 0; 204 205 data = fuse_get_mpdata(mp); 206 if (vtyp == VNON) { 207 return EINVAL; 208 } 209 *vpp = NULL; 210 err = vfs_hash_get(mp, fuse_vnode_hash(nodeid), LK_EXCLUSIVE, td, vpp, 211 fuse_vnode_cmp, &nodeid); 212 if (err) 213 return (err); 214 215 if (*vpp) { 216 if ((*vpp)->v_type != vtyp) { 217 /* 218 * STALE vnode! This probably indicates a buggy 219 * server, but it could also be the result of a race 220 * between FUSE_LOOKUP and another client's 221 * FUSE_UNLINK/FUSE_CREATE 222 */ 223 SDT_PROBE3(fusefs, , node, stale_vnode, *vpp, vtyp, 224 nodeid); 225 fuse_internal_vnode_disappear(*vpp); 226 lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL); 227 *vpp = NULL; 228 return (EAGAIN); 229 } 230 MPASS((*vpp)->v_data != NULL); 231 MPASS(VTOFUD(*vpp)->nid == nodeid); 232 SDT_PROBE2(fusefs, , node, trace, 1, "vnode taken from hash"); 233 return (0); 234 } 235 fvdat = malloc(sizeof(*fvdat), M_FUSEVN, M_WAITOK | M_ZERO); 236 switch (vtyp) { 237 case VFIFO: 238 err = getnewvnode("fuse", mp, &fuse_fifoops, vpp); 239 break; 240 default: 241 err = getnewvnode("fuse", mp, &fuse_vnops, vpp); 242 break; 243 } 244 if (err) { 245 free(fvdat, M_FUSEVN); 246 return (err); 247 } 248 lockmgr((*vpp)->v_vnlock, LK_EXCLUSIVE, NULL); 249 fuse_vnode_init(*vpp, fvdat, nodeid, vtyp); 250 err = insmntque(*vpp, mp); 251 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc"); 252 if (err) { 253 lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL); 254 free(fvdat, M_FUSEVN); 255 *vpp = NULL; 256 return (err); 257 } 258 /* Disallow async reads for fifos because UFS does. I don't know why */ 259 if (data->dataflags & FSESS_ASYNC_READ && vtyp != VFIFO) 260 VN_LOCK_ASHARE(*vpp); 261 262 err = vfs_hash_insert(*vpp, fuse_vnode_hash(nodeid), LK_EXCLUSIVE, 263 td, &vp2, fuse_vnode_cmp, &nodeid); 264 if (err) { 265 lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL); 266 free(fvdat, M_FUSEVN); 267 *vpp = NULL; 268 return (err); 269 } 270 if (vp2 != NULL) { 271 *vpp = vp2; 272 return (0); 273 } 274 275 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc"); 276 277 return (0); 278 } 279 280 int 281 fuse_vnode_get(struct mount *mp, 282 struct fuse_entry_out *feo, 283 uint64_t nodeid, 284 struct vnode *dvp, 285 struct vnode **vpp, 286 struct componentname *cnp, 287 enum vtype vtyp) 288 { 289 struct thread *td = curthread; 290 /* 291 * feo should only be NULL for the root directory, which (when libfuse 292 * is used) always has generation 0 293 */ 294 uint64_t generation = feo ? feo->generation : 0; 295 int err = 0; 296 297 err = fuse_vnode_alloc(mp, td, nodeid, vtyp, vpp); 298 if (err) { 299 return err; 300 } 301 if (dvp != NULL) { 302 MPASS(cnp && (cnp->cn_flags & ISDOTDOT) == 0); 303 MPASS(cnp && 304 !(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.')); 305 fuse_vnode_setparent(*vpp, dvp); 306 } 307 if (dvp != NULL && cnp != NULL && (cnp->cn_flags & MAKEENTRY) != 0 && 308 feo != NULL && 309 (feo->entry_valid != 0 || feo->entry_valid_nsec != 0)) { 310 struct timespec timeout; 311 312 ASSERT_VOP_LOCKED(*vpp, "fuse_vnode_get"); 313 ASSERT_VOP_LOCKED(dvp, "fuse_vnode_get"); 314 315 fuse_validity_2_timespec(feo, &timeout); 316 cache_enter_time(dvp, *vpp, cnp, &timeout, NULL); 317 } 318 319 VTOFUD(*vpp)->generation = generation; 320 /* 321 * In userland, libfuse uses cached lookups for dot and dotdot entries, 322 * thus it does not really bump the nlookup counter for forget. 323 * Follow the same semantic and avoid the bump in order to keep 324 * nlookup counters consistent. 325 */ 326 if (cnp == NULL || ((cnp->cn_flags & ISDOTDOT) == 0 && 327 (cnp->cn_namelen != 1 || cnp->cn_nameptr[0] != '.'))) 328 VTOFUD(*vpp)->nlookup++; 329 330 return 0; 331 } 332 333 /* 334 * Called for every fusefs vnode open to initialize the vnode (not 335 * fuse_filehandle) for use 336 */ 337 void 338 fuse_vnode_open(struct vnode *vp, int32_t fuse_open_flags, struct thread *td) 339 { 340 if (vnode_vtype(vp) == VREG) 341 vnode_create_vobject(vp, 0, td); 342 } 343 344 int 345 fuse_vnode_savesize(struct vnode *vp, struct ucred *cred, pid_t pid) 346 { 347 struct fuse_vnode_data *fvdat = VTOFUD(vp); 348 struct thread *td = curthread; 349 struct fuse_filehandle *fufh = NULL; 350 struct fuse_dispatcher fdi; 351 struct fuse_setattr_in *fsai; 352 int err = 0; 353 354 ASSERT_VOP_ELOCKED(vp, "fuse_io_extend"); 355 356 if (fuse_isdeadfs(vp)) { 357 return EBADF; 358 } 359 if (vnode_vtype(vp) == VDIR) { 360 return EISDIR; 361 } 362 if (vfs_isrdonly(vnode_mount(vp))) { 363 return EROFS; 364 } 365 if (cred == NULL) { 366 cred = td->td_ucred; 367 } 368 fdisp_init(&fdi, sizeof(*fsai)); 369 fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred); 370 fsai = fdi.indata; 371 fsai->valid = 0; 372 373 /* Truncate to a new value. */ 374 MPASS((fvdat->flag & FN_SIZECHANGE) != 0); 375 fsai->size = fvdat->cached_attrs.va_size; 376 fsai->valid |= FATTR_SIZE; 377 378 fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); 379 if (fufh) { 380 fsai->fh = fufh->fh_id; 381 fsai->valid |= FATTR_FH; 382 } 383 err = fdisp_wait_answ(&fdi); 384 fdisp_destroy(&fdi); 385 if (err == 0) 386 fvdat->flag &= ~FN_SIZECHANGE; 387 388 return err; 389 } 390 391 /* 392 * Adjust the vnode's size to a new value. 393 * 394 * If the new value came from the server, such as from a FUSE_GETATTR 395 * operation, set `from_server` true. But if it came from a local operation, 396 * such as write(2) or truncate(2), set `from_server` false. 397 */ 398 int 399 fuse_vnode_setsize(struct vnode *vp, off_t newsize, bool from_server) 400 { 401 struct fuse_vnode_data *fvdat = VTOFUD(vp); 402 struct vattr *attrs; 403 off_t oldsize; 404 size_t iosize; 405 struct buf *bp = NULL; 406 int err = 0; 407 408 ASSERT_VOP_ELOCKED(vp, "fuse_vnode_setsize"); 409 410 iosize = fuse_iosize(vp); 411 oldsize = fvdat->cached_attrs.va_size; 412 fvdat->cached_attrs.va_size = newsize; 413 if ((attrs = VTOVA(vp)) != NULL) 414 attrs->va_size = newsize; 415 416 if (newsize < oldsize) { 417 daddr_t lbn; 418 419 err = vtruncbuf(vp, newsize, fuse_iosize(vp)); 420 if (err) 421 goto out; 422 if (newsize % iosize == 0) 423 goto out; 424 /* 425 * Zero the contents of the last partial block. 426 * Sure seems like vtruncbuf should do this for us. 427 */ 428 429 lbn = newsize / iosize; 430 bp = getblk(vp, lbn, iosize, PCATCH, 0, 0); 431 if (!bp) { 432 err = EINTR; 433 goto out; 434 } 435 if (!(bp->b_flags & B_CACHE)) 436 goto out; /* Nothing to do */ 437 MPASS(bp->b_flags & B_VMIO); 438 vfs_bio_clrbuf(bp); 439 bp->b_dirtyend = MIN(bp->b_dirtyend, newsize - lbn * iosize); 440 } else if (from_server && newsize > oldsize && oldsize != VNOVAL) { 441 /* 442 * The FUSE server changed the file size behind our back. We 443 * should invalidate the entire cache. 444 */ 445 daddr_t left_lbn, end_lbn; 446 447 left_lbn = oldsize / iosize; 448 end_lbn = howmany(newsize, iosize); 449 v_inval_buf_range(vp, 0, end_lbn, iosize); 450 } 451 out: 452 if (bp) 453 brelse(bp); 454 vnode_pager_setsize(vp, newsize); 455 return err; 456 } 457 458 /* Get the current, possibly dirty, size of the file */ 459 int 460 fuse_vnode_size(struct vnode *vp, off_t *filesize, struct ucred *cred, 461 struct thread *td) 462 { 463 struct fuse_vnode_data *fvdat = VTOFUD(vp); 464 int error = 0; 465 466 if (!(fvdat->flag & FN_SIZECHANGE) && 467 (!fuse_vnode_attr_cache_valid(vp) || 468 fvdat->cached_attrs.va_size == VNOVAL)) 469 error = fuse_internal_do_getattr(vp, NULL, cred, td); 470 471 if (!error) 472 *filesize = fvdat->cached_attrs.va_size; 473 474 return error; 475 } 476 477 void 478 fuse_vnode_undirty_cached_timestamps(struct vnode *vp) 479 { 480 struct fuse_vnode_data *fvdat = VTOFUD(vp); 481 482 fvdat->flag &= ~(FN_MTIMECHANGE | FN_CTIMECHANGE); 483 } 484 485 /* Update a fuse file's cached timestamps */ 486 void 487 fuse_vnode_update(struct vnode *vp, int flags) 488 { 489 struct fuse_vnode_data *fvdat = VTOFUD(vp); 490 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp)); 491 struct timespec ts; 492 493 vfs_timestamp(&ts); 494 495 if (data->time_gran > 1) 496 ts.tv_nsec = rounddown(ts.tv_nsec, data->time_gran); 497 498 if (flags & FN_MTIMECHANGE) 499 fvdat->cached_attrs.va_mtime = ts; 500 if (flags & FN_CTIMECHANGE) 501 fvdat->cached_attrs.va_ctime = ts; 502 503 fvdat->flag |= flags; 504 } 505 506 void 507 fuse_node_init(void) 508 { 509 fuse_node_count = counter_u64_alloc(M_WAITOK); 510 } 511 512 void 513 fuse_node_destroy(void) 514 { 515 counter_u64_free(fuse_node_count); 516 } 517