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/counter.h> 67#include <sys/errno.h> 68#include <sys/kernel.h> 69#include <sys/conf.h> 70#include <sys/uio.h> 71#include <sys/malloc.h> 72#include <sys/queue.h> 73#include <sys/lock.h> 74#include <sys/sx.h> 75#include <sys/mutex.h> 76#include <sys/proc.h> 77#include <sys/mount.h> 78#include <sys/sdt.h> 79#include <sys/vnode.h> 80#include <sys/signalvar.h> 81#include <sys/syscallsubr.h> 82#include <sys/sysctl.h> 83#include <vm/uma.h> 84 85#include "fuse.h" 86#include "fuse_node.h" 87#include "fuse_ipc.h" 88#include "fuse_internal.h" 89 90SDT_PROVIDER_DECLARE(fusefs); 91/* 92 * Fuse trace probe: 93 * arg0: verbosity. Higher numbers give more verbose messages 94 * arg1: Textual message 95 */ 96SDT_PROBE_DEFINE2(fusefs, , ipc, trace, "int", "char*"); 97 98static void fdisp_make_pid(struct fuse_dispatcher *fdip, enum fuse_opcode op, 99 struct fuse_data *data, uint64_t nid, pid_t pid, struct ucred *cred); 100static void fuse_interrupt_send(struct fuse_ticket *otick, int err); 101static struct fuse_ticket *fticket_alloc(struct fuse_data *data); 102static void fticket_refresh(struct fuse_ticket *ftick); 103static inline void fticket_reset(struct fuse_ticket *ftick); 104static void fticket_destroy(struct fuse_ticket *ftick); 105static int fticket_wait_answer(struct fuse_ticket *ftick); 106static inline int 107fticket_aw_pull_uio(struct fuse_ticket *ftick, 108 struct uio *uio); 109 110static int fuse_body_audit(struct fuse_ticket *ftick, size_t blen); 111 112static fuse_handler_t fuse_standard_handler; 113 114static counter_u64_t fuse_ticket_count; 115SYSCTL_COUNTER_U64(_vfs_fusefs_stats, OID_AUTO, ticket_count, CTLFLAG_RD, 116 &fuse_ticket_count, "Number of allocated tickets"); 117 118static long fuse_iov_permanent_bufsize = 1 << 19; 119 120SYSCTL_LONG(_vfs_fusefs, OID_AUTO, iov_permanent_bufsize, CTLFLAG_RW, 121 &fuse_iov_permanent_bufsize, 0, 122 "limit for permanently stored buffer size for fuse_iovs"); 123static int fuse_iov_credit = 16; 124 125SYSCTL_INT(_vfs_fusefs, OID_AUTO, iov_credit, CTLFLAG_RW, 126 &fuse_iov_credit, 0, 127 "how many times is an oversized fuse_iov tolerated"); 128 129MALLOC_DEFINE(M_FUSEMSG, "fuse_msgbuf", "fuse message buffer"); 130static uma_zone_t ticket_zone; 131 132/* 133 * TODO: figure out how to timeout INTERRUPT requests, because the daemon may 134 * leagally never respond 135 */ 136static int 137fuse_interrupt_callback(struct fuse_ticket *tick, struct uio *uio) 138{ 139 struct fuse_ticket *otick, *x_tick; 140 struct fuse_interrupt_in *fii; 141 struct fuse_data *data = tick->tk_data; 142 bool found = false; 143 144 fii = (struct fuse_interrupt_in*)((char*)tick->tk_ms_fiov.base + 145 sizeof(struct fuse_in_header)); 146 147 fuse_lck_mtx_lock(data->aw_mtx); 148 TAILQ_FOREACH_SAFE(otick, &data->aw_head, tk_aw_link, x_tick) { 149 if (otick->tk_unique == fii->unique) { 150 found = true; 151 break; 152 } 153 } 154 fuse_lck_mtx_unlock(data->aw_mtx); 155 156 if (!found) { 157 /* Original is already complete. Just return */ 158 return 0; 159 } 160 161 /* Clear the original ticket's interrupt association */ 162 otick->irq_unique = 0; 163 164 if (tick->tk_aw_ohead.error == ENOSYS) { 165 fsess_set_notimpl(data->mp, FUSE_INTERRUPT); 166 return 0; 167 } else if (tick->tk_aw_ohead.error == EAGAIN) { 168 /* 169 * There are two reasons we might get this: 170 * 1) the daemon received the INTERRUPT request before the 171 * original, or 172 * 2) the daemon received the INTERRUPT request after it 173 * completed the original request. 174 * In the first case we should re-send the INTERRUPT. In the 175 * second, we should ignore it. 176 */ 177 /* Resend */ 178 fuse_interrupt_send(otick, EINTR); 179 return 0; 180 } else { 181 /* Illegal FUSE_INTERRUPT response */ 182 return EINVAL; 183 } 184} 185 186/* Interrupt the operation otick. Return err as its error code */ 187void 188fuse_interrupt_send(struct fuse_ticket *otick, int err) 189{ 190 struct fuse_dispatcher fdi; 191 struct fuse_interrupt_in *fii; 192 struct fuse_in_header *ftick_hdr; 193 struct fuse_data *data = otick->tk_data; 194 struct fuse_ticket *tick, *xtick; 195 struct ucred reused_creds; 196 gid_t reused_groups[1]; 197 198 if (otick->irq_unique == 0) { 199 /* 200 * If the daemon hasn't yet received otick, then we can answer 201 * it ourselves and return. 202 */ 203 fuse_lck_mtx_lock(data->ms_mtx); 204 STAILQ_FOREACH_SAFE(tick, &otick->tk_data->ms_head, tk_ms_link, 205 xtick) { 206 if (tick == otick) { 207 STAILQ_REMOVE(&otick->tk_data->ms_head, tick, 208 fuse_ticket, tk_ms_link); 209 otick->tk_data->ms_count--; 210 otick->tk_ms_link.stqe_next = NULL; 211 fuse_lck_mtx_unlock(data->ms_mtx); 212 213 fuse_lck_mtx_lock(otick->tk_aw_mtx); 214 if (!fticket_answered(otick)) { 215 fticket_set_answered(otick); 216 otick->tk_aw_errno = err; 217 wakeup(otick); 218 } 219 fuse_lck_mtx_unlock(otick->tk_aw_mtx); 220 221 fuse_ticket_drop(tick); 222 return; 223 } 224 } 225 fuse_lck_mtx_unlock(data->ms_mtx); 226 227 /* 228 * If the fuse daemon doesn't support interrupts, then there's 229 * nothing more that we can do 230 */ 231 if (fsess_not_impl(data->mp, FUSE_INTERRUPT)) 232 return; 233 234 /* 235 * If the fuse daemon has already received otick, then we must 236 * send FUSE_INTERRUPT. 237 */ 238 ftick_hdr = fticket_in_header(otick); 239 reused_creds.cr_uid = ftick_hdr->uid; 240 reused_groups[0] = ftick_hdr->gid; 241 reused_creds.cr_groups = reused_groups; 242 fdisp_init(&fdi, sizeof(*fii)); 243 fdisp_make_pid(&fdi, FUSE_INTERRUPT, data, ftick_hdr->nodeid, 244 ftick_hdr->pid, &reused_creds); 245 246 fii = fdi.indata; 247 fii->unique = otick->tk_unique; 248 fuse_insert_callback(fdi.tick, fuse_interrupt_callback); 249 250 otick->irq_unique = fdi.tick->tk_unique; 251 /* Interrupt ops should be delivered ASAP */ 252 fuse_insert_message(fdi.tick, true); 253 fdisp_destroy(&fdi); 254 } else { 255 /* This ticket has already been interrupted */ 256 } 257} 258 259void 260fiov_init(struct fuse_iov *fiov, size_t size) 261{ 262 uint32_t msize = FU_AT_LEAST(size); 263 264 fiov->len = 0; 265 266 fiov->base = malloc(msize, M_FUSEMSG, M_WAITOK | M_ZERO); 267 268 fiov->allocated_size = msize; 269 fiov->credit = fuse_iov_credit; 270} 271 272void 273fiov_teardown(struct fuse_iov *fiov) 274{ 275 MPASS(fiov->base != NULL); 276 free(fiov->base, M_FUSEMSG); 277} 278 279void 280fiov_adjust(struct fuse_iov *fiov, size_t size) 281{ 282 if (fiov->allocated_size < size || 283 (fuse_iov_permanent_bufsize >= 0 && 284 fiov->allocated_size - size > fuse_iov_permanent_bufsize && 285 --fiov->credit < 0)) { 286 fiov->base = realloc(fiov->base, FU_AT_LEAST(size), M_FUSEMSG, 287 M_WAITOK | M_ZERO); 288 if (!fiov->base) { 289 panic("FUSE: realloc failed"); 290 } 291 fiov->allocated_size = FU_AT_LEAST(size); 292 fiov->credit = fuse_iov_credit; 293 /* Clear data buffer after reallocation */ 294 bzero(fiov->base, size); 295 } else if (size > fiov->len) { 296 /* Clear newly extended portion of data buffer */ 297 bzero((char*)fiov->base + fiov->len, size - fiov->len); 298 } 299 fiov->len = size; 300} 301 302/* Resize the fiov if needed, and clear it's buffer */ 303void 304fiov_refresh(struct fuse_iov *fiov) 305{ 306 fiov_adjust(fiov, 0); 307} 308 309static int 310fticket_ctor(void *mem, int size, void *arg, int flags) 311{ 312 struct fuse_ticket *ftick = mem; 313 struct fuse_data *data = arg; 314 315 FUSE_ASSERT_MS_DONE(ftick); 316 FUSE_ASSERT_AW_DONE(ftick); 317 318 ftick->tk_data = data; 319 ftick->irq_unique = 0; 320 refcount_init(&ftick->tk_refcount, 1); 321 counter_u64_add(fuse_ticket_count, 1); 322 323 fticket_refresh(ftick); 324 325 return 0; 326} 327 328static void 329fticket_dtor(void *mem, int size, void *arg) 330{ 331#ifdef INVARIANTS 332 struct fuse_ticket *ftick = mem; 333#endif 334 335 FUSE_ASSERT_MS_DONE(ftick); 336 FUSE_ASSERT_AW_DONE(ftick); 337 338 counter_u64_add(fuse_ticket_count, -1); 339} 340 341static int 342fticket_init(void *mem, int size, int flags) 343{ 344 struct fuse_ticket *ftick = mem; 345 346 bzero(ftick, sizeof(struct fuse_ticket)); 347 348 fiov_init(&ftick->tk_ms_fiov, sizeof(struct fuse_in_header)); 349 350 mtx_init(&ftick->tk_aw_mtx, "fuse answer delivery mutex", NULL, MTX_DEF); 351 fiov_init(&ftick->tk_aw_fiov, 0); 352 353 return 0; 354} 355 356static void 357fticket_fini(void *mem, int size) 358{ 359 struct fuse_ticket *ftick = mem; 360 361 fiov_teardown(&ftick->tk_ms_fiov); 362 fiov_teardown(&ftick->tk_aw_fiov); 363 mtx_destroy(&ftick->tk_aw_mtx); 364} 365 366static inline struct fuse_ticket * 367fticket_alloc(struct fuse_data *data) 368{ 369 return uma_zalloc_arg(ticket_zone, data, M_WAITOK); 370} 371 372static inline void 373fticket_destroy(struct fuse_ticket *ftick) 374{ 375 return uma_zfree(ticket_zone, ftick); 376} 377 378/* Prepare the ticket to be reused and clear its data buffers */ 379static inline void 380fticket_refresh(struct fuse_ticket *ftick) 381{ 382 fticket_reset(ftick); 383 384 fiov_refresh(&ftick->tk_ms_fiov); 385 fiov_refresh(&ftick->tk_aw_fiov); 386} 387 388/* Prepare the ticket to be reused, but don't clear its data buffers */ 389static inline void 390fticket_reset(struct fuse_ticket *ftick) 391{ 392 struct fuse_data *data = ftick->tk_data; 393 394 FUSE_ASSERT_MS_DONE(ftick); 395 FUSE_ASSERT_AW_DONE(ftick); 396 397 bzero(&ftick->tk_aw_ohead, sizeof(struct fuse_out_header)); 398 399 ftick->tk_aw_errno = 0; 400 ftick->tk_flag = 0; 401 402 /* May be truncated to 32 bits on LP32 arches */ 403 ftick->tk_unique = atomic_fetchadd_long(&data->ticketer, 1); 404 if (ftick->tk_unique == 0) 405 ftick->tk_unique = atomic_fetchadd_long(&data->ticketer, 1); 406} 407 408static int 409fticket_wait_answer(struct fuse_ticket *ftick) 410{ 411 struct thread *td = curthread; 412 sigset_t blockedset, oldset; 413 int err = 0, stops_deferred; 414 struct fuse_data *data = ftick->tk_data; 415 bool interrupted = false; 416 417 if (fsess_maybe_impl(ftick->tk_data->mp, FUSE_INTERRUPT) && 418 data->dataflags & FSESS_INTR) { 419 SIGEMPTYSET(blockedset); 420 } else { 421 /* Block all signals except (implicitly) SIGKILL */ 422 SIGFILLSET(blockedset); 423 } 424 stops_deferred = sigdeferstop(SIGDEFERSTOP_SILENT); 425 kern_sigprocmask(td, SIG_BLOCK, NULL, &oldset, 0); 426 427 fuse_lck_mtx_lock(ftick->tk_aw_mtx); 428 429retry: 430 if (fticket_answered(ftick)) { 431 goto out; 432 } 433 434 if (fdata_get_dead(data)) { 435 err = ENOTCONN; 436 fticket_set_answered(ftick); 437 goto out; 438 } 439 kern_sigprocmask(td, SIG_BLOCK, &blockedset, NULL, 0); 440 err = msleep(ftick, &ftick->tk_aw_mtx, PCATCH, "fu_ans", 441 data->daemon_timeout * hz); 442 kern_sigprocmask(td, SIG_SETMASK, &oldset, NULL, 0); 443 if (err == EWOULDBLOCK) { 444 SDT_PROBE2(fusefs, , ipc, trace, 3, 445 "fticket_wait_answer: EWOULDBLOCK"); 446#ifdef XXXIP /* die conditionally */ 447 if (!fdata_get_dead(data)) { 448 fdata_set_dead(data); 449 } 450#endif 451 err = ETIMEDOUT; 452 fticket_set_answered(ftick); 453 } else if ((err == EINTR || err == ERESTART)) { 454 /* 455 * Whether we get EINTR or ERESTART depends on whether 456 * SA_RESTART was set by sigaction(2). 457 * 458 * Try to interrupt the operation and wait for an EINTR response 459 * to the original operation. If the file system does not 460 * support FUSE_INTERRUPT, then we'll just wait for it to 461 * complete like normal. If it does support FUSE_INTERRUPT, 462 * then it will either respond EINTR to the original operation, 463 * or EAGAIN to the interrupt. 464 */ 465 sigset_t tmpset; 466 467 SDT_PROBE2(fusefs, , ipc, trace, 4, 468 "fticket_wait_answer: interrupt"); 469 fuse_lck_mtx_unlock(ftick->tk_aw_mtx); 470 fuse_interrupt_send(ftick, err); 471 472 PROC_LOCK(td->td_proc); 473 mtx_lock(&td->td_proc->p_sigacts->ps_mtx); 474 tmpset = td->td_proc->p_siglist; 475 SIGSETOR(tmpset, td->td_siglist); 476 mtx_unlock(&td->td_proc->p_sigacts->ps_mtx); 477 PROC_UNLOCK(td->td_proc); 478 479 fuse_lck_mtx_lock(ftick->tk_aw_mtx); 480 if (!interrupted && !SIGISMEMBER(tmpset, SIGKILL)) { 481 /* 482 * Block all signals while we wait for an interrupt 483 * response. The protocol doesn't discriminate between 484 * different signals. 485 */ 486 SIGFILLSET(blockedset); 487 interrupted = true; 488 goto retry; 489 } else { 490 /* 491 * Return immediately for fatal signals, or if this is 492 * the second interruption. We should only be 493 * interrupted twice if the thread is stopped, for 494 * example during sigexit. 495 */ 496 } 497 } else if (err) { 498 SDT_PROBE2(fusefs, , ipc, trace, 6, 499 "fticket_wait_answer: other error"); 500 } else { 501 SDT_PROBE2(fusefs, , ipc, trace, 7, "fticket_wait_answer: OK"); 502 } 503out: 504 if (!(err || fticket_answered(ftick))) { 505 SDT_PROBE2(fusefs, , ipc, trace, 1, 506 "FUSE: requester was woken up but still no answer"); 507 err = ENXIO; 508 } 509 fuse_lck_mtx_unlock(ftick->tk_aw_mtx); 510 sigallowstop(stops_deferred); 511 512 return err; 513} 514 515static inline 516int 517fticket_aw_pull_uio(struct fuse_ticket *ftick, struct uio *uio) 518{ 519 int err = 0; 520 size_t len = uio_resid(uio); 521 522 if (len) { 523 fiov_adjust(fticket_resp(ftick), len); 524 err = uiomove(fticket_resp(ftick)->base, len, uio); 525 } 526 return err; 527} 528 529int 530fticket_pull(struct fuse_ticket *ftick, struct uio *uio) 531{ 532 int err = 0; 533 534 if (ftick->tk_aw_ohead.error) { 535 return 0; 536 } 537 err = fuse_body_audit(ftick, uio_resid(uio)); 538 if (!err) { 539 err = fticket_aw_pull_uio(ftick, uio); 540 } 541 return err; 542} 543 544struct fuse_data * 545fdata_alloc(struct cdev *fdev, struct ucred *cred) 546{ 547 struct fuse_data *data; 548 549 data = malloc(sizeof(struct fuse_data), M_FUSEMSG, M_WAITOK | M_ZERO); 550 551 data->fdev = fdev; 552 mtx_init(&data->ms_mtx, "fuse message list mutex", NULL, MTX_DEF); 553 STAILQ_INIT(&data->ms_head); 554 data->ms_count = 0; 555 knlist_init_mtx(&data->ks_rsel.si_note, &data->ms_mtx); 556 mtx_init(&data->aw_mtx, "fuse answer list mutex", NULL, MTX_DEF); 557 TAILQ_INIT(&data->aw_head); 558 data->daemoncred = crhold(cred); 559 data->daemon_timeout = FUSE_DEFAULT_DAEMON_TIMEOUT; 560 sx_init(&data->rename_lock, "fuse rename lock"); 561 data->ref = 1; 562 563 return data; 564} 565 566void 567fdata_trydestroy(struct fuse_data *data) 568{ 569 data->ref--; 570 MPASS(data->ref >= 0); 571 if (data->ref != 0) 572 return; 573 574 /* Driving off stage all that stuff thrown at device... */ 575 sx_destroy(&data->rename_lock); 576 crfree(data->daemoncred); 577 mtx_destroy(&data->aw_mtx); 578 knlist_delete(&data->ks_rsel.si_note, curthread, 0); 579 knlist_destroy(&data->ks_rsel.si_note); 580 mtx_destroy(&data->ms_mtx); 581 582 free(data, M_FUSEMSG); 583} 584 585void 586fdata_set_dead(struct fuse_data *data) 587{ 588 FUSE_LOCK(); 589 if (fdata_get_dead(data)) { 590 FUSE_UNLOCK(); 591 return; 592 } 593 fuse_lck_mtx_lock(data->ms_mtx); 594 data->dataflags |= FSESS_DEAD; 595 wakeup_one(data); 596 selwakeuppri(&data->ks_rsel, PZERO + 1); 597 wakeup(&data->ticketer); 598 fuse_lck_mtx_unlock(data->ms_mtx); 599 FUSE_UNLOCK(); 600} 601 602struct fuse_ticket * 603fuse_ticket_fetch(struct fuse_data *data) 604{ 605 int err = 0; 606 struct fuse_ticket *ftick; 607 608 ftick = fticket_alloc(data); 609 610 if (!(data->dataflags & FSESS_INITED)) { 611 /* Sleep until get answer for INIT message */ 612 FUSE_LOCK(); 613 if (!(data->dataflags & FSESS_INITED) && data->ticketer > 2) { 614 err = msleep(&data->ticketer, &fuse_mtx, PCATCH | PDROP, 615 "fu_ini", 0); 616 if (err) 617 fdata_set_dead(data); 618 } else 619 FUSE_UNLOCK(); 620 } 621 return ftick; 622} 623 624int 625fuse_ticket_drop(struct fuse_ticket *ftick) 626{ 627 int die; 628 629 die = refcount_release(&ftick->tk_refcount); 630 if (die) 631 fticket_destroy(ftick); 632 633 return die; 634} 635 636void 637fuse_insert_callback(struct fuse_ticket *ftick, fuse_handler_t * handler) 638{ 639 if (fdata_get_dead(ftick->tk_data)) { 640 return; 641 } 642 ftick->tk_aw_handler = handler; 643 644 fuse_lck_mtx_lock(ftick->tk_data->aw_mtx); 645 fuse_aw_push(ftick); 646 fuse_lck_mtx_unlock(ftick->tk_data->aw_mtx); 647} 648 649/* 650 * Insert a new upgoing ticket into the message queue 651 * 652 * If urgent is true, insert at the front of the queue. Otherwise, insert in 653 * FIFO order. 654 */ 655void 656fuse_insert_message(struct fuse_ticket *ftick, bool urgent) 657{ 658 if (ftick->tk_flag & FT_DIRTY) { 659 panic("FUSE: ticket reused without being refreshed"); 660 } 661 ftick->tk_flag |= FT_DIRTY; 662 663 if (fdata_get_dead(ftick->tk_data)) { 664 return; 665 } 666 fuse_lck_mtx_lock(ftick->tk_data->ms_mtx); 667 if (urgent) 668 fuse_ms_push_head(ftick); 669 else 670 fuse_ms_push(ftick); 671 wakeup_one(ftick->tk_data); 672 selwakeuppri(&ftick->tk_data->ks_rsel, PZERO + 1); 673 KNOTE_LOCKED(&ftick->tk_data->ks_rsel.si_note, 0); 674 fuse_lck_mtx_unlock(ftick->tk_data->ms_mtx); 675} 676 677static int 678fuse_body_audit(struct fuse_ticket *ftick, size_t blen) 679{ 680 int err = 0; 681 enum fuse_opcode opcode; 682 683 opcode = fticket_opcode(ftick); 684 685 switch (opcode) { 686 case FUSE_BMAP: 687 err = (blen == sizeof(struct fuse_bmap_out)) ? 0 : EINVAL; 688 break; 689 690 case FUSE_LINK: 691 case FUSE_LOOKUP: 692 case FUSE_MKDIR: 693 case FUSE_MKNOD: 694 case FUSE_SYMLINK: 695 if (fuse_libabi_geq(ftick->tk_data, 7, 9)) { 696 err = (blen == sizeof(struct fuse_entry_out)) ? 697 0 : EINVAL; 698 } else { 699 err = (blen == FUSE_COMPAT_ENTRY_OUT_SIZE) ? 0 : EINVAL; 700 } 701 break; 702 703 case FUSE_FORGET: 704 panic("FUSE: a handler has been intalled for FUSE_FORGET"); 705 break; 706 707 case FUSE_GETATTR: 708 case FUSE_SETATTR: 709 if (fuse_libabi_geq(ftick->tk_data, 7, 9)) { 710 err = (blen == sizeof(struct fuse_attr_out)) ? 711 0 : EINVAL; 712 } else { 713 err = (blen == FUSE_COMPAT_ATTR_OUT_SIZE) ? 0 : EINVAL; 714 } 715 break; 716 717 case FUSE_READLINK: 718 err = (PAGE_SIZE >= blen) ? 0 : EINVAL; 719 break; 720 721 case FUSE_UNLINK: 722 err = (blen == 0) ? 0 : EINVAL; 723 break; 724 725 case FUSE_RMDIR: 726 err = (blen == 0) ? 0 : EINVAL; 727 break; 728 729 case FUSE_RENAME: 730 err = (blen == 0) ? 0 : EINVAL; 731 break; 732 733 case FUSE_OPEN: 734 err = (blen == sizeof(struct fuse_open_out)) ? 0 : EINVAL; 735 break; 736 737 case FUSE_READ: 738 err = (((struct fuse_read_in *)( 739 (char *)ftick->tk_ms_fiov.base + 740 sizeof(struct fuse_in_header) 741 ))->size >= blen) ? 0 : EINVAL; 742 break; 743 744 case FUSE_WRITE: 745 err = (blen == sizeof(struct fuse_write_out)) ? 0 : EINVAL; 746 break; 747 748 case FUSE_STATFS: 749 if (fuse_libabi_geq(ftick->tk_data, 7, 4)) { 750 err = (blen == sizeof(struct fuse_statfs_out)) ? 751 0 : EINVAL; 752 } else { 753 err = (blen == FUSE_COMPAT_STATFS_SIZE) ? 0 : EINVAL; 754 } 755 break; 756 757 case FUSE_RELEASE: 758 err = (blen == 0) ? 0 : EINVAL; 759 break; 760 761 case FUSE_FSYNC: 762 err = (blen == 0) ? 0 : EINVAL; 763 break; 764 765 case FUSE_SETXATTR: 766 err = (blen == 0) ? 0 : EINVAL; 767 break; 768 769 case FUSE_GETXATTR: 770 case FUSE_LISTXATTR: 771 /* 772 * These can have varying response lengths, and 0 length 773 * isn't necessarily invalid. 774 */ 775 err = 0; 776 break; 777 778 case FUSE_REMOVEXATTR: 779 err = (blen == 0) ? 0 : EINVAL; 780 break; 781 782 case FUSE_FLUSH: 783 err = (blen == 0) ? 0 : EINVAL; 784 break; 785 786 case FUSE_INIT: 787 if (blen == sizeof(struct fuse_init_out) || 788 blen == FUSE_COMPAT_INIT_OUT_SIZE || 789 blen == FUSE_COMPAT_22_INIT_OUT_SIZE) { 790 err = 0; 791 } else { 792 err = EINVAL; 793 } 794 break; 795 796 case FUSE_OPENDIR: 797 err = (blen == sizeof(struct fuse_open_out)) ? 0 : EINVAL; 798 break; 799 800 case FUSE_READDIR: 801 err = (((struct fuse_read_in *)( 802 (char *)ftick->tk_ms_fiov.base + 803 sizeof(struct fuse_in_header) 804 ))->size >= blen) ? 0 : EINVAL; 805 break; 806 807 case FUSE_RELEASEDIR: 808 err = (blen == 0) ? 0 : EINVAL; 809 break; 810 811 case FUSE_FSYNCDIR: 812 err = (blen == 0) ? 0 : EINVAL; 813 break; 814 815 case FUSE_GETLK: 816 err = (blen == sizeof(struct fuse_lk_out)) ? 0 : EINVAL; 817 break; 818 819 case FUSE_SETLK: 820 err = (blen == 0) ? 0 : EINVAL; 821 break; 822 823 case FUSE_SETLKW: 824 err = (blen == 0) ? 0 : EINVAL; 825 break; 826 827 case FUSE_ACCESS: 828 err = (blen == 0) ? 0 : EINVAL; 829 break; 830 831 case FUSE_CREATE: 832 if (fuse_libabi_geq(ftick->tk_data, 7, 9)) { 833 err = (blen == sizeof(struct fuse_entry_out) + 834 sizeof(struct fuse_open_out)) ? 0 : EINVAL; 835 } else { 836 err = (blen == FUSE_COMPAT_ENTRY_OUT_SIZE + 837 sizeof(struct fuse_open_out)) ? 0 : EINVAL; 838 } 839 break; 840 841 case FUSE_DESTROY: 842 err = (blen == 0) ? 0 : EINVAL; 843 break; 844 845 case FUSE_FALLOCATE: 846 err = (blen == 0) ? 0 : EINVAL; 847 break; 848 849 case FUSE_LSEEK: 850 err = (blen == sizeof(struct fuse_lseek_out)) ? 0 : EINVAL; 851 break; 852 853 case FUSE_COPY_FILE_RANGE: 854 err = (blen == sizeof(struct fuse_write_out)) ? 0 : EINVAL; 855 break; 856 857 default: 858 panic("FUSE: opcodes out of sync (%d)\n", opcode); 859 } 860 861 return err; 862} 863 864static inline void 865fuse_setup_ihead(struct fuse_in_header *ihead, struct fuse_ticket *ftick, 866 uint64_t nid, enum fuse_opcode op, size_t blen, pid_t pid, 867 struct ucred *cred) 868{ 869 ihead->len = sizeof(*ihead) + blen; 870 ihead->unique = ftick->tk_unique; 871 ihead->nodeid = nid; 872 ihead->opcode = op; 873 874 ihead->pid = pid; 875 ihead->uid = cred->cr_uid; 876 ihead->gid = cred->cr_groups[0]; 877} 878 879/* 880 * fuse_standard_handler just pulls indata and wakes up pretender. 881 * Doesn't try to interpret data, that's left for the pretender. 882 * Though might do a basic size verification before the pull-in takes place 883 */ 884 885static int 886fuse_standard_handler(struct fuse_ticket *ftick, struct uio *uio) 887{ 888 int err = 0; 889 890 err = fticket_pull(ftick, uio); 891 892 fuse_lck_mtx_lock(ftick->tk_aw_mtx); 893 894 if (!fticket_answered(ftick)) { 895 fticket_set_answered(ftick); 896 ftick->tk_aw_errno = err; 897 wakeup(ftick); 898 } 899 fuse_lck_mtx_unlock(ftick->tk_aw_mtx); 900 901 return err; 902} 903 904/* 905 * Reinitialize a dispatcher from a pid and node id, without resizing or 906 * clearing its data buffers 907 */ 908static void 909fdisp_refresh_pid(struct fuse_dispatcher *fdip, enum fuse_opcode op, 910 struct mount *mp, uint64_t nid, pid_t pid, struct ucred *cred) 911{ 912 MPASS(fdip->tick); 913 MPASS2(sizeof(fdip->finh) + fdip->iosize <= fdip->tick->tk_ms_fiov.len, 914 "Must use fdisp_make_pid to increase the size of the fiov"); 915 fticket_reset(fdip->tick); 916 917 FUSE_DIMALLOC(&fdip->tick->tk_ms_fiov, fdip->finh, 918 fdip->indata, fdip->iosize); 919 920 fuse_setup_ihead(fdip->finh, fdip->tick, nid, op, fdip->iosize, pid, 921 cred); 922} 923 924/* Initialize a dispatcher from a pid and node id */ 925static void 926fdisp_make_pid(struct fuse_dispatcher *fdip, enum fuse_opcode op, 927 struct fuse_data *data, uint64_t nid, pid_t pid, struct ucred *cred) 928{ 929 if (fdip->tick) { 930 fticket_refresh(fdip->tick); 931 } else { 932 fdip->tick = fuse_ticket_fetch(data); 933 } 934 935 /* FUSE_DIMALLOC will bzero the fiovs when it enlarges them */ 936 FUSE_DIMALLOC(&fdip->tick->tk_ms_fiov, fdip->finh, 937 fdip->indata, fdip->iosize); 938 939 fuse_setup_ihead(fdip->finh, fdip->tick, nid, op, fdip->iosize, pid, cred); 940} 941 942void 943fdisp_make(struct fuse_dispatcher *fdip, enum fuse_opcode op, struct mount *mp, 944 uint64_t nid, struct thread *td, struct ucred *cred) 945{ 946 struct fuse_data *data = fuse_get_mpdata(mp); 947 RECTIFY_TDCR(td, cred); 948 949 return fdisp_make_pid(fdip, op, data, nid, td->td_proc->p_pid, cred); 950} 951 952void 953fdisp_make_vp(struct fuse_dispatcher *fdip, enum fuse_opcode op, 954 struct vnode *vp, struct thread *td, struct ucred *cred) 955{ 956 struct mount *mp = vnode_mount(vp); 957 struct fuse_data *data = fuse_get_mpdata(mp); 958 959 RECTIFY_TDCR(td, cred); 960 return fdisp_make_pid(fdip, op, data, VTOI(vp), 961 td->td_proc->p_pid, cred); 962} 963 964/* Refresh a fuse_dispatcher so it can be reused, but don't zero its data */ 965void 966fdisp_refresh_vp(struct fuse_dispatcher *fdip, enum fuse_opcode op, 967 struct vnode *vp, struct thread *td, struct ucred *cred) 968{ 969 RECTIFY_TDCR(td, cred); 970 return fdisp_refresh_pid(fdip, op, vnode_mount(vp), VTOI(vp), 971 td->td_proc->p_pid, cred); 972} 973 974SDT_PROBE_DEFINE2(fusefs, , ipc, fdisp_wait_answ_error, "char*", "int"); 975 976int 977fdisp_wait_answ(struct fuse_dispatcher *fdip) 978{ 979 int err = 0; 980 981 fdip->answ_stat = 0; 982 fuse_insert_callback(fdip->tick, fuse_standard_handler); 983 fuse_insert_message(fdip->tick, false); 984 985 if ((err = fticket_wait_answer(fdip->tick))) { 986 fuse_lck_mtx_lock(fdip->tick->tk_aw_mtx); 987 988 if (fticket_answered(fdip->tick)) { 989 /* 990 * Just between noticing the interrupt and getting here, 991 * the standard handler has completed his job. 992 * So we drop the ticket and exit as usual. 993 */ 994 SDT_PROBE2(fusefs, , ipc, fdisp_wait_answ_error, 995 "IPC: interrupted, already answered", err); 996 fuse_lck_mtx_unlock(fdip->tick->tk_aw_mtx); 997 goto out; 998 } else { 999 /* 1000 * So we were faster than the standard handler. 1001 * Then by setting the answered flag we get *him* 1002 * to drop the ticket. 1003 */ 1004 SDT_PROBE2(fusefs, , ipc, fdisp_wait_answ_error, 1005 "IPC: interrupted, setting to answered", err); 1006 fticket_set_answered(fdip->tick); 1007 fuse_lck_mtx_unlock(fdip->tick->tk_aw_mtx); 1008 return err; 1009 } 1010 } 1011 1012 if (fdip->tick->tk_aw_errno == ENOTCONN) { 1013 /* The daemon died while we were waiting for a response */ 1014 err = ENOTCONN; 1015 goto out; 1016 } else if (fdip->tick->tk_aw_errno) { 1017 /* 1018 * There was some sort of communication error with the daemon 1019 * that the client wouldn't understand. 1020 */ 1021 SDT_PROBE2(fusefs, , ipc, fdisp_wait_answ_error, 1022 "IPC: explicit EIO-ing", fdip->tick->tk_aw_errno); 1023 err = EIO; 1024 goto out; 1025 } 1026 if ((err = fdip->tick->tk_aw_ohead.error)) { 1027 SDT_PROBE2(fusefs, , ipc, fdisp_wait_answ_error, 1028 "IPC: setting status", fdip->tick->tk_aw_ohead.error); 1029 /* 1030 * This means a "proper" fuse syscall error. 1031 * We record this value so the caller will 1032 * be able to know it's not a boring messaging 1033 * failure, if she wishes so (and if not, she can 1034 * just simply propagate the return value of this routine). 1035 * [XXX Maybe a bitflag would do the job too, 1036 * if other flags needed, this will be converted thusly.] 1037 */ 1038 fdip->answ_stat = err; 1039 goto out; 1040 } 1041 fdip->answ = fticket_resp(fdip->tick)->base; 1042 fdip->iosize = fticket_resp(fdip->tick)->len; 1043 1044 return 0; 1045 1046out: 1047 return err; 1048} 1049 1050void 1051fuse_ipc_init(void) 1052{ 1053 ticket_zone = uma_zcreate("fuse_ticket", sizeof(struct fuse_ticket), 1054 fticket_ctor, fticket_dtor, fticket_init, fticket_fini, 1055 UMA_ALIGN_PTR, 0); 1056 fuse_ticket_count = counter_u64_alloc(M_WAITOK); 1057} 1058 1059void 1060fuse_ipc_destroy(void) 1061{ 1062 counter_u64_free(fuse_ticket_count); 1063 uma_zdestroy(ticket_zone); 1064} 1065 1066SDT_PROBE_DEFINE3(fusefs,, ipc, warn, "struct fuse_data*", "unsigned", "char*"); 1067void 1068fuse_warn(struct fuse_data *data, unsigned flag, const char *msg) 1069{ 1070 SDT_PROBE3(fusefs, , ipc, warn, data, flag, msg); 1071 if (!(data->dataflags & flag)) { 1072 printf("WARNING: FUSE protocol violation for server mounted at " 1073 "%s: %s " 1074 "This warning will not be repeated.\n", 1075 data->mp->mnt_stat.f_mntonname, msg); 1076 data->dataflags |= flag; 1077 } 1078} 1079