1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Copyright (c) 2012 Konstantin Belousov <kib@FreeBSD.org> 13 * Copyright (c) 2013, 2014 The FreeBSD Foundation 14 * 15 * Portions of this software were developed by Konstantin Belousov 16 * under sponsorship from the FreeBSD Foundation. 17 * 18 * Redistribution and use in source and binary forms, with or without 19 * modification, are permitted provided that the following conditions 20 * are met: 21 * 1. Redistributions of source code must retain the above copyright 22 * notice, this list of conditions and the following disclaimer. 23 * 2. Redistributions in binary form must reproduce the above copyright 24 * notice, this list of conditions and the following disclaimer in the 25 * documentation and/or other materials provided with the distribution. 26 * 3. Neither the name of the University nor the names of its contributors 27 * may be used to endorse or promote products derived from this software 28 * without specific prior written permission. 29 * 30 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 31 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 32 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 33 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 34 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 35 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 36 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 37 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 38 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 39 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 40 * SUCH DAMAGE. 41 * 42 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94 43 */ 44 45 #include <sys/cdefs.h> 46 __FBSDID("$FreeBSD$"); 47 48 #include "opt_hwpmc_hooks.h" 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/disk.h> 53 #include <sys/fail.h> 54 #include <sys/fcntl.h> 55 #include <sys/file.h> 56 #include <sys/kdb.h> 57 #include <sys/stat.h> 58 #include <sys/priv.h> 59 #include <sys/proc.h> 60 #include <sys/limits.h> 61 #include <sys/lock.h> 62 #include <sys/mman.h> 63 #include <sys/mount.h> 64 #include <sys/mutex.h> 65 #include <sys/namei.h> 66 #include <sys/vnode.h> 67 #include <sys/bio.h> 68 #include <sys/buf.h> 69 #include <sys/filio.h> 70 #include <sys/resourcevar.h> 71 #include <sys/rwlock.h> 72 #include <sys/sx.h> 73 #include <sys/sysctl.h> 74 #include <sys/ttycom.h> 75 #include <sys/conf.h> 76 #include <sys/syslog.h> 77 #include <sys/unistd.h> 78 #include <sys/user.h> 79 80 #include <security/audit/audit.h> 81 #include <security/mac/mac_framework.h> 82 83 #include <vm/vm.h> 84 #include <vm/vm_extern.h> 85 #include <vm/pmap.h> 86 #include <vm/vm_map.h> 87 #include <vm/vm_object.h> 88 #include <vm/vm_page.h> 89 #include <vm/vnode_pager.h> 90 91 #ifdef HWPMC_HOOKS 92 #include <sys/pmckern.h> 93 #endif 94 95 static fo_rdwr_t vn_read; 96 static fo_rdwr_t vn_write; 97 static fo_rdwr_t vn_io_fault; 98 static fo_truncate_t vn_truncate; 99 static fo_ioctl_t vn_ioctl; 100 static fo_poll_t vn_poll; 101 static fo_kqfilter_t vn_kqfilter; 102 static fo_stat_t vn_statfile; 103 static fo_close_t vn_closefile; 104 static fo_mmap_t vn_mmap; 105 106 struct fileops vnops = { 107 .fo_read = vn_io_fault, 108 .fo_write = vn_io_fault, 109 .fo_truncate = vn_truncate, 110 .fo_ioctl = vn_ioctl, 111 .fo_poll = vn_poll, 112 .fo_kqfilter = vn_kqfilter, 113 .fo_stat = vn_statfile, 114 .fo_close = vn_closefile, 115 .fo_chmod = vn_chmod, 116 .fo_chown = vn_chown, 117 .fo_sendfile = vn_sendfile, 118 .fo_seek = vn_seek, 119 .fo_fill_kinfo = vn_fill_kinfo, 120 .fo_mmap = vn_mmap, 121 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 122 }; 123 124 static const int io_hold_cnt = 16; 125 static int vn_io_fault_enable = 1; 126 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW, 127 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance"); 128 static int vn_io_fault_prefault = 0; 129 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW, 130 &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting"); 131 static u_long vn_io_faults_cnt; 132 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD, 133 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers"); 134 135 /* 136 * Returns true if vn_io_fault mode of handling the i/o request should 137 * be used. 138 */ 139 static bool 140 do_vn_io_fault(struct vnode *vp, struct uio *uio) 141 { 142 struct mount *mp; 143 144 return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG && 145 (mp = vp->v_mount) != NULL && 146 (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable); 147 } 148 149 /* 150 * Structure used to pass arguments to vn_io_fault1(), to do either 151 * file- or vnode-based I/O calls. 152 */ 153 struct vn_io_fault_args { 154 enum { 155 VN_IO_FAULT_FOP, 156 VN_IO_FAULT_VOP 157 } kind; 158 struct ucred *cred; 159 int flags; 160 union { 161 struct fop_args_tag { 162 struct file *fp; 163 fo_rdwr_t *doio; 164 } fop_args; 165 struct vop_args_tag { 166 struct vnode *vp; 167 } vop_args; 168 } args; 169 }; 170 171 static int vn_io_fault1(struct vnode *vp, struct uio *uio, 172 struct vn_io_fault_args *args, struct thread *td); 173 174 int 175 vn_open(struct nameidata *ndp, int *flagp, int cmode, struct file *fp) 176 { 177 struct thread *td = ndp->ni_cnd.cn_thread; 178 179 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp)); 180 } 181 182 /* 183 * Common code for vnode open operations via a name lookup. 184 * Lookup the vnode and invoke VOP_CREATE if needed. 185 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine. 186 * 187 * Note that this does NOT free nameidata for the successful case, 188 * due to the NDINIT being done elsewhere. 189 */ 190 int 191 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, 192 struct ucred *cred, struct file *fp) 193 { 194 struct vnode *vp; 195 struct mount *mp; 196 struct thread *td = ndp->ni_cnd.cn_thread; 197 struct vattr vat; 198 struct vattr *vap = &vat; 199 int fmode, error; 200 201 restart: 202 fmode = *flagp; 203 if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT | 204 O_EXCL | O_DIRECTORY)) 205 return (EINVAL); 206 else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) { 207 ndp->ni_cnd.cn_nameiop = CREATE; 208 /* 209 * Set NOCACHE to avoid flushing the cache when 210 * rolling in many files at once. 211 */ 212 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE; 213 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 214 ndp->ni_cnd.cn_flags |= FOLLOW; 215 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 216 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 217 if (vn_open_flags & VN_OPEN_NOCAPCHECK) 218 ndp->ni_cnd.cn_flags |= NOCAPCHECK; 219 bwillwrite(); 220 if ((error = namei(ndp)) != 0) 221 return (error); 222 if (ndp->ni_vp == NULL) { 223 VATTR_NULL(vap); 224 vap->va_type = VREG; 225 vap->va_mode = cmode; 226 if (fmode & O_EXCL) 227 vap->va_vaflags |= VA_EXCLUSIVE; 228 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) { 229 NDFREE(ndp, NDF_ONLY_PNBUF); 230 vput(ndp->ni_dvp); 231 if ((error = vn_start_write(NULL, &mp, 232 V_XSLEEP | PCATCH)) != 0) 233 return (error); 234 goto restart; 235 } 236 if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0) 237 ndp->ni_cnd.cn_flags |= MAKEENTRY; 238 #ifdef MAC 239 error = mac_vnode_check_create(cred, ndp->ni_dvp, 240 &ndp->ni_cnd, vap); 241 if (error == 0) 242 #endif 243 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 244 &ndp->ni_cnd, vap); 245 vput(ndp->ni_dvp); 246 vn_finished_write(mp); 247 if (error) { 248 NDFREE(ndp, NDF_ONLY_PNBUF); 249 return (error); 250 } 251 fmode &= ~O_TRUNC; 252 vp = ndp->ni_vp; 253 } else { 254 if (ndp->ni_dvp == ndp->ni_vp) 255 vrele(ndp->ni_dvp); 256 else 257 vput(ndp->ni_dvp); 258 ndp->ni_dvp = NULL; 259 vp = ndp->ni_vp; 260 if (fmode & O_EXCL) { 261 error = EEXIST; 262 goto bad; 263 } 264 fmode &= ~O_CREAT; 265 } 266 } else { 267 ndp->ni_cnd.cn_nameiop = LOOKUP; 268 ndp->ni_cnd.cn_flags = ISOPEN | 269 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF; 270 if (!(fmode & FWRITE)) 271 ndp->ni_cnd.cn_flags |= LOCKSHARED; 272 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 273 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 274 if (vn_open_flags & VN_OPEN_NOCAPCHECK) 275 ndp->ni_cnd.cn_flags |= NOCAPCHECK; 276 if ((error = namei(ndp)) != 0) 277 return (error); 278 vp = ndp->ni_vp; 279 } 280 error = vn_open_vnode(vp, fmode, cred, td, fp); 281 if (error) 282 goto bad; 283 *flagp = fmode; 284 return (0); 285 bad: 286 NDFREE(ndp, NDF_ONLY_PNBUF); 287 vput(vp); 288 *flagp = fmode; 289 ndp->ni_vp = NULL; 290 return (error); 291 } 292 293 /* 294 * Common code for vnode open operations once a vnode is located. 295 * Check permissions, and call the VOP_OPEN routine. 296 */ 297 int 298 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, 299 struct thread *td, struct file *fp) 300 { 301 accmode_t accmode; 302 struct flock lf; 303 int error, lock_flags, type; 304 305 if (vp->v_type == VLNK) 306 return (EMLINK); 307 if (vp->v_type == VSOCK) 308 return (EOPNOTSUPP); 309 if (vp->v_type != VDIR && fmode & O_DIRECTORY) 310 return (ENOTDIR); 311 accmode = 0; 312 if (fmode & (FWRITE | O_TRUNC)) { 313 if (vp->v_type == VDIR) 314 return (EISDIR); 315 accmode |= VWRITE; 316 } 317 if (fmode & FREAD) 318 accmode |= VREAD; 319 if (fmode & FEXEC) 320 accmode |= VEXEC; 321 if ((fmode & O_APPEND) && (fmode & FWRITE)) 322 accmode |= VAPPEND; 323 #ifdef MAC 324 if (fmode & O_CREAT) 325 accmode |= VCREAT; 326 if (fmode & O_VERIFY) 327 accmode |= VVERIFY; 328 error = mac_vnode_check_open(cred, vp, accmode); 329 if (error) 330 return (error); 331 332 accmode &= ~(VCREAT | VVERIFY); 333 #endif 334 if ((fmode & O_CREAT) == 0) { 335 if (accmode & VWRITE) { 336 error = vn_writechk(vp); 337 if (error) 338 return (error); 339 } 340 if (accmode) { 341 error = VOP_ACCESS(vp, accmode, cred, td); 342 if (error) 343 return (error); 344 } 345 } 346 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) 347 vn_lock(vp, LK_UPGRADE | LK_RETRY); 348 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0) 349 return (error); 350 351 while ((fmode & (O_EXLOCK | O_SHLOCK)) != 0) { 352 KASSERT(fp != NULL, ("open with flock requires fp")); 353 if (fp->f_type != DTYPE_NONE && fp->f_type != DTYPE_VNODE) { 354 error = EOPNOTSUPP; 355 break; 356 } 357 lock_flags = VOP_ISLOCKED(vp); 358 VOP_UNLOCK(vp, 0); 359 lf.l_whence = SEEK_SET; 360 lf.l_start = 0; 361 lf.l_len = 0; 362 if (fmode & O_EXLOCK) 363 lf.l_type = F_WRLCK; 364 else 365 lf.l_type = F_RDLCK; 366 type = F_FLOCK; 367 if ((fmode & FNONBLOCK) == 0) 368 type |= F_WAIT; 369 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type); 370 if (error == 0) 371 fp->f_flag |= FHASLOCK; 372 vn_lock(vp, lock_flags | LK_RETRY); 373 if (error != 0) 374 break; 375 if ((vp->v_iflag & VI_DOOMED) != 0) { 376 error = ENOENT; 377 break; 378 } 379 380 /* 381 * Another thread might have used this vnode as an 382 * executable while the vnode lock was dropped. 383 * Ensure the vnode is still able to be opened for 384 * writing after the lock has been obtained. 385 */ 386 if ((accmode & VWRITE) != 0) 387 error = vn_writechk(vp); 388 break; 389 } 390 391 if (error != 0) { 392 fp->f_flag |= FOPENFAILED; 393 fp->f_vnode = vp; 394 if (fp->f_ops == &badfileops) { 395 fp->f_type = DTYPE_VNODE; 396 fp->f_ops = &vnops; 397 } 398 vref(vp); 399 } else if ((fmode & FWRITE) != 0) { 400 VOP_ADD_WRITECOUNT(vp, 1); 401 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", 402 __func__, vp, vp->v_writecount); 403 } 404 ASSERT_VOP_LOCKED(vp, "vn_open_vnode"); 405 return (error); 406 } 407 408 /* 409 * Check for write permissions on the specified vnode. 410 * Prototype text segments cannot be written. 411 */ 412 int 413 vn_writechk(struct vnode *vp) 414 { 415 416 ASSERT_VOP_LOCKED(vp, "vn_writechk"); 417 /* 418 * If there's shared text associated with 419 * the vnode, try to free it up once. If 420 * we fail, we can't allow writing. 421 */ 422 if (VOP_IS_TEXT(vp)) 423 return (ETXTBSY); 424 425 return (0); 426 } 427 428 /* 429 * Vnode close call 430 */ 431 static int 432 vn_close1(struct vnode *vp, int flags, struct ucred *file_cred, 433 struct thread *td, bool keep_ref) 434 { 435 struct mount *mp; 436 int error, lock_flags; 437 438 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 && 439 MNT_EXTENDED_SHARED(vp->v_mount)) 440 lock_flags = LK_SHARED; 441 else 442 lock_flags = LK_EXCLUSIVE; 443 444 vn_start_write(vp, &mp, V_WAIT); 445 vn_lock(vp, lock_flags | LK_RETRY); 446 AUDIT_ARG_VNODE1(vp); 447 if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) { 448 VNASSERT(vp->v_writecount > 0, vp, 449 ("vn_close: negative writecount")); 450 VOP_ADD_WRITECOUNT(vp, -1); 451 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", 452 __func__, vp, vp->v_writecount); 453 } 454 error = VOP_CLOSE(vp, flags, file_cred, td); 455 if (keep_ref) 456 VOP_UNLOCK(vp, 0); 457 else 458 vput(vp); 459 vn_finished_write(mp); 460 return (error); 461 } 462 463 int 464 vn_close(struct vnode *vp, int flags, struct ucred *file_cred, 465 struct thread *td) 466 { 467 468 return (vn_close1(vp, flags, file_cred, td, false)); 469 } 470 471 /* 472 * Heuristic to detect sequential operation. 473 */ 474 static int 475 sequential_heuristic(struct uio *uio, struct file *fp) 476 { 477 478 ASSERT_VOP_LOCKED(fp->f_vnode, __func__); 479 if (fp->f_flag & FRDAHEAD) 480 return (fp->f_seqcount << IO_SEQSHIFT); 481 482 /* 483 * Offset 0 is handled specially. open() sets f_seqcount to 1 so 484 * that the first I/O is normally considered to be slightly 485 * sequential. Seeking to offset 0 doesn't change sequentiality 486 * unless previous seeks have reduced f_seqcount to 0, in which 487 * case offset 0 is not special. 488 */ 489 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 490 uio->uio_offset == fp->f_nextoff) { 491 /* 492 * f_seqcount is in units of fixed-size blocks so that it 493 * depends mainly on the amount of sequential I/O and not 494 * much on the number of sequential I/O's. The fixed size 495 * of 16384 is hard-coded here since it is (not quite) just 496 * a magic size that works well here. This size is more 497 * closely related to the best I/O size for real disks than 498 * to any block size used by software. 499 */ 500 fp->f_seqcount += howmany(uio->uio_resid, 16384); 501 if (fp->f_seqcount > IO_SEQMAX) 502 fp->f_seqcount = IO_SEQMAX; 503 return (fp->f_seqcount << IO_SEQSHIFT); 504 } 505 506 /* Not sequential. Quickly draw-down sequentiality. */ 507 if (fp->f_seqcount > 1) 508 fp->f_seqcount = 1; 509 else 510 fp->f_seqcount = 0; 511 return (0); 512 } 513 514 /* 515 * Package up an I/O request on a vnode into a uio and do it. 516 */ 517 int 518 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, 519 enum uio_seg segflg, int ioflg, struct ucred *active_cred, 520 struct ucred *file_cred, ssize_t *aresid, struct thread *td) 521 { 522 struct uio auio; 523 struct iovec aiov; 524 struct mount *mp; 525 struct ucred *cred; 526 void *rl_cookie; 527 struct vn_io_fault_args args; 528 int error, lock_flags; 529 530 if (offset < 0 && vp->v_type != VCHR) 531 return (EINVAL); 532 auio.uio_iov = &aiov; 533 auio.uio_iovcnt = 1; 534 aiov.iov_base = base; 535 aiov.iov_len = len; 536 auio.uio_resid = len; 537 auio.uio_offset = offset; 538 auio.uio_segflg = segflg; 539 auio.uio_rw = rw; 540 auio.uio_td = td; 541 error = 0; 542 543 if ((ioflg & IO_NODELOCKED) == 0) { 544 if ((ioflg & IO_RANGELOCKED) == 0) { 545 if (rw == UIO_READ) { 546 rl_cookie = vn_rangelock_rlock(vp, offset, 547 offset + len); 548 } else { 549 rl_cookie = vn_rangelock_wlock(vp, offset, 550 offset + len); 551 } 552 } else 553 rl_cookie = NULL; 554 mp = NULL; 555 if (rw == UIO_WRITE) { 556 if (vp->v_type != VCHR && 557 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) 558 != 0) 559 goto out; 560 if (MNT_SHARED_WRITES(mp) || 561 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) 562 lock_flags = LK_SHARED; 563 else 564 lock_flags = LK_EXCLUSIVE; 565 } else 566 lock_flags = LK_SHARED; 567 vn_lock(vp, lock_flags | LK_RETRY); 568 } else 569 rl_cookie = NULL; 570 571 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 572 #ifdef MAC 573 if ((ioflg & IO_NOMACCHECK) == 0) { 574 if (rw == UIO_READ) 575 error = mac_vnode_check_read(active_cred, file_cred, 576 vp); 577 else 578 error = mac_vnode_check_write(active_cred, file_cred, 579 vp); 580 } 581 #endif 582 if (error == 0) { 583 if (file_cred != NULL) 584 cred = file_cred; 585 else 586 cred = active_cred; 587 if (do_vn_io_fault(vp, &auio)) { 588 args.kind = VN_IO_FAULT_VOP; 589 args.cred = cred; 590 args.flags = ioflg; 591 args.args.vop_args.vp = vp; 592 error = vn_io_fault1(vp, &auio, &args, td); 593 } else if (rw == UIO_READ) { 594 error = VOP_READ(vp, &auio, ioflg, cred); 595 } else /* if (rw == UIO_WRITE) */ { 596 error = VOP_WRITE(vp, &auio, ioflg, cred); 597 } 598 } 599 if (aresid) 600 *aresid = auio.uio_resid; 601 else 602 if (auio.uio_resid && error == 0) 603 error = EIO; 604 if ((ioflg & IO_NODELOCKED) == 0) { 605 VOP_UNLOCK(vp, 0); 606 if (mp != NULL) 607 vn_finished_write(mp); 608 } 609 out: 610 if (rl_cookie != NULL) 611 vn_rangelock_unlock(vp, rl_cookie); 612 return (error); 613 } 614 615 /* 616 * Package up an I/O request on a vnode into a uio and do it. The I/O 617 * request is split up into smaller chunks and we try to avoid saturating 618 * the buffer cache while potentially holding a vnode locked, so we 619 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield() 620 * to give other processes a chance to lock the vnode (either other processes 621 * core'ing the same binary, or unrelated processes scanning the directory). 622 */ 623 int 624 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, void *base, size_t len, 625 off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, 626 struct ucred *file_cred, size_t *aresid, struct thread *td) 627 { 628 int error = 0; 629 ssize_t iaresid; 630 631 do { 632 int chunk; 633 634 /* 635 * Force `offset' to a multiple of MAXBSIZE except possibly 636 * for the first chunk, so that filesystems only need to 637 * write full blocks except possibly for the first and last 638 * chunks. 639 */ 640 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 641 642 if (chunk > len) 643 chunk = len; 644 if (rw != UIO_READ && vp->v_type == VREG) 645 bwillwrite(); 646 iaresid = 0; 647 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 648 ioflg, active_cred, file_cred, &iaresid, td); 649 len -= chunk; /* aresid calc already includes length */ 650 if (error) 651 break; 652 offset += chunk; 653 base = (char *)base + chunk; 654 kern_yield(PRI_USER); 655 } while (len); 656 if (aresid) 657 *aresid = len + iaresid; 658 return (error); 659 } 660 661 off_t 662 foffset_lock(struct file *fp, int flags) 663 { 664 struct mtx *mtxp; 665 off_t res; 666 667 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); 668 669 #if OFF_MAX <= LONG_MAX 670 /* 671 * Caller only wants the current f_offset value. Assume that 672 * the long and shorter integer types reads are atomic. 673 */ 674 if ((flags & FOF_NOLOCK) != 0) 675 return (fp->f_offset); 676 #endif 677 678 /* 679 * According to McKusick the vn lock was protecting f_offset here. 680 * It is now protected by the FOFFSET_LOCKED flag. 681 */ 682 mtxp = mtx_pool_find(mtxpool_sleep, fp); 683 mtx_lock(mtxp); 684 if ((flags & FOF_NOLOCK) == 0) { 685 while (fp->f_vnread_flags & FOFFSET_LOCKED) { 686 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING; 687 msleep(&fp->f_vnread_flags, mtxp, PUSER -1, 688 "vofflock", 0); 689 } 690 fp->f_vnread_flags |= FOFFSET_LOCKED; 691 } 692 res = fp->f_offset; 693 mtx_unlock(mtxp); 694 return (res); 695 } 696 697 void 698 foffset_unlock(struct file *fp, off_t val, int flags) 699 { 700 struct mtx *mtxp; 701 702 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); 703 704 #if OFF_MAX <= LONG_MAX 705 if ((flags & FOF_NOLOCK) != 0) { 706 if ((flags & FOF_NOUPDATE) == 0) 707 fp->f_offset = val; 708 if ((flags & FOF_NEXTOFF) != 0) 709 fp->f_nextoff = val; 710 return; 711 } 712 #endif 713 714 mtxp = mtx_pool_find(mtxpool_sleep, fp); 715 mtx_lock(mtxp); 716 if ((flags & FOF_NOUPDATE) == 0) 717 fp->f_offset = val; 718 if ((flags & FOF_NEXTOFF) != 0) 719 fp->f_nextoff = val; 720 if ((flags & FOF_NOLOCK) == 0) { 721 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0, 722 ("Lost FOFFSET_LOCKED")); 723 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING) 724 wakeup(&fp->f_vnread_flags); 725 fp->f_vnread_flags = 0; 726 } 727 mtx_unlock(mtxp); 728 } 729 730 void 731 foffset_lock_uio(struct file *fp, struct uio *uio, int flags) 732 { 733 734 if ((flags & FOF_OFFSET) == 0) 735 uio->uio_offset = foffset_lock(fp, flags); 736 } 737 738 void 739 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags) 740 { 741 742 if ((flags & FOF_OFFSET) == 0) 743 foffset_unlock(fp, uio->uio_offset, flags); 744 } 745 746 static int 747 get_advice(struct file *fp, struct uio *uio) 748 { 749 struct mtx *mtxp; 750 int ret; 751 752 ret = POSIX_FADV_NORMAL; 753 if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG) 754 return (ret); 755 756 mtxp = mtx_pool_find(mtxpool_sleep, fp); 757 mtx_lock(mtxp); 758 if (fp->f_advice != NULL && 759 uio->uio_offset >= fp->f_advice->fa_start && 760 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end) 761 ret = fp->f_advice->fa_advice; 762 mtx_unlock(mtxp); 763 return (ret); 764 } 765 766 /* 767 * File table vnode read routine. 768 */ 769 static int 770 vn_read(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, 771 struct thread *td) 772 { 773 struct vnode *vp; 774 off_t orig_offset; 775 int error, ioflag; 776 int advice; 777 778 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", 779 uio->uio_td, td)); 780 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); 781 vp = fp->f_vnode; 782 ioflag = 0; 783 if (fp->f_flag & FNONBLOCK) 784 ioflag |= IO_NDELAY; 785 if (fp->f_flag & O_DIRECT) 786 ioflag |= IO_DIRECT; 787 advice = get_advice(fp, uio); 788 vn_lock(vp, LK_SHARED | LK_RETRY); 789 790 switch (advice) { 791 case POSIX_FADV_NORMAL: 792 case POSIX_FADV_SEQUENTIAL: 793 case POSIX_FADV_NOREUSE: 794 ioflag |= sequential_heuristic(uio, fp); 795 break; 796 case POSIX_FADV_RANDOM: 797 /* Disable read-ahead for random I/O. */ 798 break; 799 } 800 orig_offset = uio->uio_offset; 801 802 #ifdef MAC 803 error = mac_vnode_check_read(active_cred, fp->f_cred, vp); 804 if (error == 0) 805 #endif 806 error = VOP_READ(vp, uio, ioflag, fp->f_cred); 807 fp->f_nextoff = uio->uio_offset; 808 VOP_UNLOCK(vp, 0); 809 if (error == 0 && advice == POSIX_FADV_NOREUSE && 810 orig_offset != uio->uio_offset) 811 /* 812 * Use POSIX_FADV_DONTNEED to flush pages and buffers 813 * for the backing file after a POSIX_FADV_NOREUSE 814 * read(2). 815 */ 816 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1, 817 POSIX_FADV_DONTNEED); 818 return (error); 819 } 820 821 /* 822 * File table vnode write routine. 823 */ 824 static int 825 vn_write(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, 826 struct thread *td) 827 { 828 struct vnode *vp; 829 struct mount *mp; 830 off_t orig_offset; 831 int error, ioflag, lock_flags; 832 int advice; 833 834 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", 835 uio->uio_td, td)); 836 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); 837 vp = fp->f_vnode; 838 if (vp->v_type == VREG) 839 bwillwrite(); 840 ioflag = IO_UNIT; 841 if (vp->v_type == VREG && (fp->f_flag & O_APPEND)) 842 ioflag |= IO_APPEND; 843 if (fp->f_flag & FNONBLOCK) 844 ioflag |= IO_NDELAY; 845 if (fp->f_flag & O_DIRECT) 846 ioflag |= IO_DIRECT; 847 if ((fp->f_flag & O_FSYNC) || 848 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) 849 ioflag |= IO_SYNC; 850 mp = NULL; 851 if (vp->v_type != VCHR && 852 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) 853 goto unlock; 854 855 advice = get_advice(fp, uio); 856 857 if (MNT_SHARED_WRITES(mp) || 858 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) { 859 lock_flags = LK_SHARED; 860 } else { 861 lock_flags = LK_EXCLUSIVE; 862 } 863 864 vn_lock(vp, lock_flags | LK_RETRY); 865 switch (advice) { 866 case POSIX_FADV_NORMAL: 867 case POSIX_FADV_SEQUENTIAL: 868 case POSIX_FADV_NOREUSE: 869 ioflag |= sequential_heuristic(uio, fp); 870 break; 871 case POSIX_FADV_RANDOM: 872 /* XXX: Is this correct? */ 873 break; 874 } 875 orig_offset = uio->uio_offset; 876 877 #ifdef MAC 878 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 879 if (error == 0) 880 #endif 881 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred); 882 fp->f_nextoff = uio->uio_offset; 883 VOP_UNLOCK(vp, 0); 884 if (vp->v_type != VCHR) 885 vn_finished_write(mp); 886 if (error == 0 && advice == POSIX_FADV_NOREUSE && 887 orig_offset != uio->uio_offset) 888 /* 889 * Use POSIX_FADV_DONTNEED to flush pages and buffers 890 * for the backing file after a POSIX_FADV_NOREUSE 891 * write(2). 892 */ 893 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1, 894 POSIX_FADV_DONTNEED); 895 unlock: 896 return (error); 897 } 898 899 /* 900 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to 901 * prevent the following deadlock: 902 * 903 * Assume that the thread A reads from the vnode vp1 into userspace 904 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is 905 * currently not resident, then system ends up with the call chain 906 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] -> 907 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2) 908 * which establishes lock order vp1->vn_lock, then vp2->vn_lock. 909 * If, at the same time, thread B reads from vnode vp2 into buffer buf2 910 * backed by the pages of vnode vp1, and some page in buf2 is not 911 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock. 912 * 913 * To prevent the lock order reversal and deadlock, vn_io_fault() does 914 * not allow page faults to happen during VOP_READ() or VOP_WRITE(). 915 * Instead, it first tries to do the whole range i/o with pagefaults 916 * disabled. If all pages in the i/o buffer are resident and mapped, 917 * VOP will succeed (ignoring the genuine filesystem errors). 918 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do 919 * i/o in chunks, with all pages in the chunk prefaulted and held 920 * using vm_fault_quick_hold_pages(). 921 * 922 * Filesystems using this deadlock avoidance scheme should use the 923 * array of the held pages from uio, saved in the curthread->td_ma, 924 * instead of doing uiomove(). A helper function 925 * vn_io_fault_uiomove() converts uiomove request into 926 * uiomove_fromphys() over td_ma array. 927 * 928 * Since vnode locks do not cover the whole i/o anymore, rangelocks 929 * make the current i/o request atomic with respect to other i/os and 930 * truncations. 931 */ 932 933 /* 934 * Decode vn_io_fault_args and perform the corresponding i/o. 935 */ 936 static int 937 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio, 938 struct thread *td) 939 { 940 int error, save; 941 942 error = 0; 943 save = vm_fault_disable_pagefaults(); 944 switch (args->kind) { 945 case VN_IO_FAULT_FOP: 946 error = (args->args.fop_args.doio)(args->args.fop_args.fp, 947 uio, args->cred, args->flags, td); 948 break; 949 case VN_IO_FAULT_VOP: 950 if (uio->uio_rw == UIO_READ) { 951 error = VOP_READ(args->args.vop_args.vp, uio, 952 args->flags, args->cred); 953 } else if (uio->uio_rw == UIO_WRITE) { 954 error = VOP_WRITE(args->args.vop_args.vp, uio, 955 args->flags, args->cred); 956 } 957 break; 958 default: 959 panic("vn_io_fault_doio: unknown kind of io %d %d", 960 args->kind, uio->uio_rw); 961 } 962 vm_fault_enable_pagefaults(save); 963 return (error); 964 } 965 966 static int 967 vn_io_fault_touch(char *base, const struct uio *uio) 968 { 969 int r; 970 971 r = fubyte(base); 972 if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1)) 973 return (EFAULT); 974 return (0); 975 } 976 977 static int 978 vn_io_fault_prefault_user(const struct uio *uio) 979 { 980 char *base; 981 const struct iovec *iov; 982 size_t len; 983 ssize_t resid; 984 int error, i; 985 986 KASSERT(uio->uio_segflg == UIO_USERSPACE, 987 ("vn_io_fault_prefault userspace")); 988 989 error = i = 0; 990 iov = uio->uio_iov; 991 resid = uio->uio_resid; 992 base = iov->iov_base; 993 len = iov->iov_len; 994 while (resid > 0) { 995 error = vn_io_fault_touch(base, uio); 996 if (error != 0) 997 break; 998 if (len < PAGE_SIZE) { 999 if (len != 0) { 1000 error = vn_io_fault_touch(base + len - 1, uio); 1001 if (error != 0) 1002 break; 1003 resid -= len; 1004 } 1005 if (++i >= uio->uio_iovcnt) 1006 break; 1007 iov = uio->uio_iov + i; 1008 base = iov->iov_base; 1009 len = iov->iov_len; 1010 } else { 1011 len -= PAGE_SIZE; 1012 base += PAGE_SIZE; 1013 resid -= PAGE_SIZE; 1014 } 1015 } 1016 return (error); 1017 } 1018 1019 /* 1020 * Common code for vn_io_fault(), agnostic to the kind of i/o request. 1021 * Uses vn_io_fault_doio() to make the call to an actual i/o function. 1022 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request 1023 * into args and call vn_io_fault1() to handle faults during the user 1024 * mode buffer accesses. 1025 */ 1026 static int 1027 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, 1028 struct thread *td) 1029 { 1030 vm_page_t ma[io_hold_cnt + 2]; 1031 struct uio *uio_clone, short_uio; 1032 struct iovec short_iovec[1]; 1033 vm_page_t *prev_td_ma; 1034 vm_prot_t prot; 1035 vm_offset_t addr, end; 1036 size_t len, resid; 1037 ssize_t adv; 1038 int error, cnt, saveheld, prev_td_ma_cnt; 1039 1040 if (vn_io_fault_prefault) { 1041 error = vn_io_fault_prefault_user(uio); 1042 if (error != 0) 1043 return (error); /* Or ignore ? */ 1044 } 1045 1046 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ; 1047 1048 /* 1049 * The UFS follows IO_UNIT directive and replays back both 1050 * uio_offset and uio_resid if an error is encountered during the 1051 * operation. But, since the iovec may be already advanced, 1052 * uio is still in an inconsistent state. 1053 * 1054 * Cache a copy of the original uio, which is advanced to the redo 1055 * point using UIO_NOCOPY below. 1056 */ 1057 uio_clone = cloneuio(uio); 1058 resid = uio->uio_resid; 1059 1060 short_uio.uio_segflg = UIO_USERSPACE; 1061 short_uio.uio_rw = uio->uio_rw; 1062 short_uio.uio_td = uio->uio_td; 1063 1064 error = vn_io_fault_doio(args, uio, td); 1065 if (error != EFAULT) 1066 goto out; 1067 1068 atomic_add_long(&vn_io_faults_cnt, 1); 1069 uio_clone->uio_segflg = UIO_NOCOPY; 1070 uiomove(NULL, resid - uio->uio_resid, uio_clone); 1071 uio_clone->uio_segflg = uio->uio_segflg; 1072 1073 saveheld = curthread_pflags_set(TDP_UIOHELD); 1074 prev_td_ma = td->td_ma; 1075 prev_td_ma_cnt = td->td_ma_cnt; 1076 1077 while (uio_clone->uio_resid != 0) { 1078 len = uio_clone->uio_iov->iov_len; 1079 if (len == 0) { 1080 KASSERT(uio_clone->uio_iovcnt >= 1, 1081 ("iovcnt underflow")); 1082 uio_clone->uio_iov++; 1083 uio_clone->uio_iovcnt--; 1084 continue; 1085 } 1086 if (len > io_hold_cnt * PAGE_SIZE) 1087 len = io_hold_cnt * PAGE_SIZE; 1088 addr = (uintptr_t)uio_clone->uio_iov->iov_base; 1089 end = round_page(addr + len); 1090 if (end < addr) { 1091 error = EFAULT; 1092 break; 1093 } 1094 cnt = atop(end - trunc_page(addr)); 1095 /* 1096 * A perfectly misaligned address and length could cause 1097 * both the start and the end of the chunk to use partial 1098 * page. +2 accounts for such a situation. 1099 */ 1100 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map, 1101 addr, len, prot, ma, io_hold_cnt + 2); 1102 if (cnt == -1) { 1103 error = EFAULT; 1104 break; 1105 } 1106 short_uio.uio_iov = &short_iovec[0]; 1107 short_iovec[0].iov_base = (void *)addr; 1108 short_uio.uio_iovcnt = 1; 1109 short_uio.uio_resid = short_iovec[0].iov_len = len; 1110 short_uio.uio_offset = uio_clone->uio_offset; 1111 td->td_ma = ma; 1112 td->td_ma_cnt = cnt; 1113 1114 error = vn_io_fault_doio(args, &short_uio, td); 1115 vm_page_unhold_pages(ma, cnt); 1116 adv = len - short_uio.uio_resid; 1117 1118 uio_clone->uio_iov->iov_base = 1119 (char *)uio_clone->uio_iov->iov_base + adv; 1120 uio_clone->uio_iov->iov_len -= adv; 1121 uio_clone->uio_resid -= adv; 1122 uio_clone->uio_offset += adv; 1123 1124 uio->uio_resid -= adv; 1125 uio->uio_offset += adv; 1126 1127 if (error != 0 || adv == 0) 1128 break; 1129 } 1130 td->td_ma = prev_td_ma; 1131 td->td_ma_cnt = prev_td_ma_cnt; 1132 curthread_pflags_restore(saveheld); 1133 out: 1134 free(uio_clone, M_IOV); 1135 return (error); 1136 } 1137 1138 static int 1139 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred, 1140 int flags, struct thread *td) 1141 { 1142 fo_rdwr_t *doio; 1143 struct vnode *vp; 1144 void *rl_cookie; 1145 struct vn_io_fault_args args; 1146 int error; 1147 1148 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write; 1149 vp = fp->f_vnode; 1150 foffset_lock_uio(fp, uio, flags); 1151 if (do_vn_io_fault(vp, uio)) { 1152 args.kind = VN_IO_FAULT_FOP; 1153 args.args.fop_args.fp = fp; 1154 args.args.fop_args.doio = doio; 1155 args.cred = active_cred; 1156 args.flags = flags | FOF_OFFSET; 1157 if (uio->uio_rw == UIO_READ) { 1158 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset, 1159 uio->uio_offset + uio->uio_resid); 1160 } else if ((fp->f_flag & O_APPEND) != 0 || 1161 (flags & FOF_OFFSET) == 0) { 1162 /* For appenders, punt and lock the whole range. */ 1163 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); 1164 } else { 1165 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset, 1166 uio->uio_offset + uio->uio_resid); 1167 } 1168 error = vn_io_fault1(vp, uio, &args, td); 1169 vn_rangelock_unlock(vp, rl_cookie); 1170 } else { 1171 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td); 1172 } 1173 foffset_unlock_uio(fp, uio, flags); 1174 return (error); 1175 } 1176 1177 /* 1178 * Helper function to perform the requested uiomove operation using 1179 * the held pages for io->uio_iov[0].iov_base buffer instead of 1180 * copyin/copyout. Access to the pages with uiomove_fromphys() 1181 * instead of iov_base prevents page faults that could occur due to 1182 * pmap_collect() invalidating the mapping created by 1183 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or 1184 * object cleanup revoking the write access from page mappings. 1185 * 1186 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove() 1187 * instead of plain uiomove(). 1188 */ 1189 int 1190 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio) 1191 { 1192 struct uio transp_uio; 1193 struct iovec transp_iov[1]; 1194 struct thread *td; 1195 size_t adv; 1196 int error, pgadv; 1197 1198 td = curthread; 1199 if ((td->td_pflags & TDP_UIOHELD) == 0 || 1200 uio->uio_segflg != UIO_USERSPACE) 1201 return (uiomove(data, xfersize, uio)); 1202 1203 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); 1204 transp_iov[0].iov_base = data; 1205 transp_uio.uio_iov = &transp_iov[0]; 1206 transp_uio.uio_iovcnt = 1; 1207 if (xfersize > uio->uio_resid) 1208 xfersize = uio->uio_resid; 1209 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize; 1210 transp_uio.uio_offset = 0; 1211 transp_uio.uio_segflg = UIO_SYSSPACE; 1212 /* 1213 * Since transp_iov points to data, and td_ma page array 1214 * corresponds to original uio->uio_iov, we need to invert the 1215 * direction of the i/o operation as passed to 1216 * uiomove_fromphys(). 1217 */ 1218 switch (uio->uio_rw) { 1219 case UIO_WRITE: 1220 transp_uio.uio_rw = UIO_READ; 1221 break; 1222 case UIO_READ: 1223 transp_uio.uio_rw = UIO_WRITE; 1224 break; 1225 } 1226 transp_uio.uio_td = uio->uio_td; 1227 error = uiomove_fromphys(td->td_ma, 1228 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK, 1229 xfersize, &transp_uio); 1230 adv = xfersize - transp_uio.uio_resid; 1231 pgadv = 1232 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) - 1233 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT); 1234 td->td_ma += pgadv; 1235 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, 1236 pgadv)); 1237 td->td_ma_cnt -= pgadv; 1238 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv; 1239 uio->uio_iov->iov_len -= adv; 1240 uio->uio_resid -= adv; 1241 uio->uio_offset += adv; 1242 return (error); 1243 } 1244 1245 int 1246 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize, 1247 struct uio *uio) 1248 { 1249 struct thread *td; 1250 vm_offset_t iov_base; 1251 int cnt, pgadv; 1252 1253 td = curthread; 1254 if ((td->td_pflags & TDP_UIOHELD) == 0 || 1255 uio->uio_segflg != UIO_USERSPACE) 1256 return (uiomove_fromphys(ma, offset, xfersize, uio)); 1257 1258 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); 1259 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize; 1260 iov_base = (vm_offset_t)uio->uio_iov->iov_base; 1261 switch (uio->uio_rw) { 1262 case UIO_WRITE: 1263 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma, 1264 offset, cnt); 1265 break; 1266 case UIO_READ: 1267 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK, 1268 cnt); 1269 break; 1270 } 1271 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT); 1272 td->td_ma += pgadv; 1273 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, 1274 pgadv)); 1275 td->td_ma_cnt -= pgadv; 1276 uio->uio_iov->iov_base = (char *)(iov_base + cnt); 1277 uio->uio_iov->iov_len -= cnt; 1278 uio->uio_resid -= cnt; 1279 uio->uio_offset += cnt; 1280 return (0); 1281 } 1282 1283 1284 /* 1285 * File table truncate routine. 1286 */ 1287 static int 1288 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred, 1289 struct thread *td) 1290 { 1291 struct vattr vattr; 1292 struct mount *mp; 1293 struct vnode *vp; 1294 void *rl_cookie; 1295 int error; 1296 1297 vp = fp->f_vnode; 1298 1299 /* 1300 * Lock the whole range for truncation. Otherwise split i/o 1301 * might happen partly before and partly after the truncation. 1302 */ 1303 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); 1304 error = vn_start_write(vp, &mp, V_WAIT | PCATCH); 1305 if (error) 1306 goto out1; 1307 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1308 AUDIT_ARG_VNODE1(vp); 1309 if (vp->v_type == VDIR) { 1310 error = EISDIR; 1311 goto out; 1312 } 1313 #ifdef MAC 1314 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 1315 if (error) 1316 goto out; 1317 #endif 1318 error = vn_writechk(vp); 1319 if (error == 0) { 1320 VATTR_NULL(&vattr); 1321 vattr.va_size = length; 1322 if ((fp->f_flag & O_FSYNC) != 0) 1323 vattr.va_vaflags |= VA_SYNC; 1324 error = VOP_SETATTR(vp, &vattr, fp->f_cred); 1325 } 1326 out: 1327 VOP_UNLOCK(vp, 0); 1328 vn_finished_write(mp); 1329 out1: 1330 vn_rangelock_unlock(vp, rl_cookie); 1331 return (error); 1332 } 1333 1334 /* 1335 * File table vnode stat routine. 1336 */ 1337 static int 1338 vn_statfile(struct file *fp, struct stat *sb, struct ucred *active_cred, 1339 struct thread *td) 1340 { 1341 struct vnode *vp = fp->f_vnode; 1342 int error; 1343 1344 vn_lock(vp, LK_SHARED | LK_RETRY); 1345 error = vn_stat(vp, sb, active_cred, fp->f_cred, td); 1346 VOP_UNLOCK(vp, 0); 1347 1348 return (error); 1349 } 1350 1351 /* 1352 * Stat a vnode; implementation for the stat syscall 1353 */ 1354 int 1355 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *active_cred, 1356 struct ucred *file_cred, struct thread *td) 1357 { 1358 struct vattr vattr; 1359 struct vattr *vap; 1360 int error; 1361 u_short mode; 1362 1363 AUDIT_ARG_VNODE1(vp); 1364 #ifdef MAC 1365 error = mac_vnode_check_stat(active_cred, file_cred, vp); 1366 if (error) 1367 return (error); 1368 #endif 1369 1370 vap = &vattr; 1371 1372 /* 1373 * Initialize defaults for new and unusual fields, so that file 1374 * systems which don't support these fields don't need to know 1375 * about them. 1376 */ 1377 vap->va_birthtime.tv_sec = -1; 1378 vap->va_birthtime.tv_nsec = 0; 1379 vap->va_fsid = VNOVAL; 1380 vap->va_rdev = NODEV; 1381 1382 error = VOP_GETATTR(vp, vap, active_cred); 1383 if (error) 1384 return (error); 1385 1386 /* 1387 * Zero the spare stat fields 1388 */ 1389 bzero(sb, sizeof *sb); 1390 1391 /* 1392 * Copy from vattr table 1393 */ 1394 if (vap->va_fsid != VNOVAL) 1395 sb->st_dev = vap->va_fsid; 1396 else 1397 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 1398 sb->st_ino = vap->va_fileid; 1399 mode = vap->va_mode; 1400 switch (vap->va_type) { 1401 case VREG: 1402 mode |= S_IFREG; 1403 break; 1404 case VDIR: 1405 mode |= S_IFDIR; 1406 break; 1407 case VBLK: 1408 mode |= S_IFBLK; 1409 break; 1410 case VCHR: 1411 mode |= S_IFCHR; 1412 break; 1413 case VLNK: 1414 mode |= S_IFLNK; 1415 break; 1416 case VSOCK: 1417 mode |= S_IFSOCK; 1418 break; 1419 case VFIFO: 1420 mode |= S_IFIFO; 1421 break; 1422 default: 1423 return (EBADF); 1424 } 1425 sb->st_mode = mode; 1426 sb->st_nlink = vap->va_nlink; 1427 sb->st_uid = vap->va_uid; 1428 sb->st_gid = vap->va_gid; 1429 sb->st_rdev = vap->va_rdev; 1430 if (vap->va_size > OFF_MAX) 1431 return (EOVERFLOW); 1432 sb->st_size = vap->va_size; 1433 sb->st_atim = vap->va_atime; 1434 sb->st_mtim = vap->va_mtime; 1435 sb->st_ctim = vap->va_ctime; 1436 sb->st_birthtim = vap->va_birthtime; 1437 1438 /* 1439 * According to www.opengroup.org, the meaning of st_blksize is 1440 * "a filesystem-specific preferred I/O block size for this 1441 * object. In some filesystem types, this may vary from file 1442 * to file" 1443 * Use miminum/default of PAGE_SIZE (e.g. for VCHR). 1444 */ 1445 1446 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize); 1447 1448 sb->st_flags = vap->va_flags; 1449 if (priv_check(td, PRIV_VFS_GENERATION)) 1450 sb->st_gen = 0; 1451 else 1452 sb->st_gen = vap->va_gen; 1453 1454 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 1455 return (0); 1456 } 1457 1458 /* 1459 * File table vnode ioctl routine. 1460 */ 1461 static int 1462 vn_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 1463 struct thread *td) 1464 { 1465 struct vattr vattr; 1466 struct vnode *vp; 1467 int error; 1468 1469 vp = fp->f_vnode; 1470 switch (vp->v_type) { 1471 case VDIR: 1472 case VREG: 1473 switch (com) { 1474 case FIONREAD: 1475 vn_lock(vp, LK_SHARED | LK_RETRY); 1476 error = VOP_GETATTR(vp, &vattr, active_cred); 1477 VOP_UNLOCK(vp, 0); 1478 if (error == 0) 1479 *(int *)data = vattr.va_size - fp->f_offset; 1480 return (error); 1481 case FIONBIO: 1482 case FIOASYNC: 1483 return (0); 1484 default: 1485 return (VOP_IOCTL(vp, com, data, fp->f_flag, 1486 active_cred, td)); 1487 } 1488 break; 1489 case VCHR: 1490 return (VOP_IOCTL(vp, com, data, fp->f_flag, 1491 active_cred, td)); 1492 default: 1493 return (ENOTTY); 1494 } 1495 } 1496 1497 /* 1498 * File table vnode poll routine. 1499 */ 1500 static int 1501 vn_poll(struct file *fp, int events, struct ucred *active_cred, 1502 struct thread *td) 1503 { 1504 struct vnode *vp; 1505 int error; 1506 1507 vp = fp->f_vnode; 1508 #ifdef MAC 1509 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1510 AUDIT_ARG_VNODE1(vp); 1511 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp); 1512 VOP_UNLOCK(vp, 0); 1513 if (!error) 1514 #endif 1515 1516 error = VOP_POLL(vp, events, fp->f_cred, td); 1517 return (error); 1518 } 1519 1520 /* 1521 * Acquire the requested lock and then check for validity. LK_RETRY 1522 * permits vn_lock to return doomed vnodes. 1523 */ 1524 int 1525 _vn_lock(struct vnode *vp, int flags, char *file, int line) 1526 { 1527 int error; 1528 1529 VNASSERT((flags & LK_TYPE_MASK) != 0, vp, 1530 ("vn_lock: no locktype")); 1531 VNASSERT(vp->v_holdcnt != 0, vp, ("vn_lock: zero hold count")); 1532 retry: 1533 error = VOP_LOCK1(vp, flags, file, line); 1534 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */ 1535 KASSERT((flags & LK_RETRY) == 0 || error == 0, 1536 ("vn_lock: error %d incompatible with flags %#x", error, flags)); 1537 1538 if ((flags & LK_RETRY) == 0) { 1539 if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) { 1540 VOP_UNLOCK(vp, 0); 1541 error = ENOENT; 1542 } 1543 } else if (error != 0) 1544 goto retry; 1545 return (error); 1546 } 1547 1548 /* 1549 * File table vnode close routine. 1550 */ 1551 static int 1552 vn_closefile(struct file *fp, struct thread *td) 1553 { 1554 struct vnode *vp; 1555 struct flock lf; 1556 int error; 1557 bool ref; 1558 1559 vp = fp->f_vnode; 1560 fp->f_ops = &badfileops; 1561 ref= (fp->f_flag & FHASLOCK) != 0 && fp->f_type == DTYPE_VNODE; 1562 1563 error = vn_close1(vp, fp->f_flag, fp->f_cred, td, ref); 1564 1565 if (__predict_false(ref)) { 1566 lf.l_whence = SEEK_SET; 1567 lf.l_start = 0; 1568 lf.l_len = 0; 1569 lf.l_type = F_UNLCK; 1570 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); 1571 vrele(vp); 1572 } 1573 return (error); 1574 } 1575 1576 static bool 1577 vn_suspendable(struct mount *mp) 1578 { 1579 1580 return (mp->mnt_op->vfs_susp_clean != NULL); 1581 } 1582 1583 /* 1584 * Preparing to start a filesystem write operation. If the operation is 1585 * permitted, then we bump the count of operations in progress and 1586 * proceed. If a suspend request is in progress, we wait until the 1587 * suspension is over, and then proceed. 1588 */ 1589 static int 1590 vn_start_write_locked(struct mount *mp, int flags) 1591 { 1592 int error, mflags; 1593 1594 mtx_assert(MNT_MTX(mp), MA_OWNED); 1595 error = 0; 1596 1597 /* 1598 * Check on status of suspension. 1599 */ 1600 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 || 1601 mp->mnt_susp_owner != curthread) { 1602 mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? 1603 (flags & PCATCH) : 0) | (PUSER - 1); 1604 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 1605 if (flags & V_NOWAIT) { 1606 error = EWOULDBLOCK; 1607 goto unlock; 1608 } 1609 error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags, 1610 "suspfs", 0); 1611 if (error) 1612 goto unlock; 1613 } 1614 } 1615 if (flags & V_XSLEEP) 1616 goto unlock; 1617 mp->mnt_writeopcount++; 1618 unlock: 1619 if (error != 0 || (flags & V_XSLEEP) != 0) 1620 MNT_REL(mp); 1621 MNT_IUNLOCK(mp); 1622 return (error); 1623 } 1624 1625 int 1626 vn_start_write(struct vnode *vp, struct mount **mpp, int flags) 1627 { 1628 struct mount *mp; 1629 int error; 1630 1631 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), 1632 ("V_MNTREF requires mp")); 1633 1634 error = 0; 1635 /* 1636 * If a vnode is provided, get and return the mount point that 1637 * to which it will write. 1638 */ 1639 if (vp != NULL) { 1640 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 1641 *mpp = NULL; 1642 if (error != EOPNOTSUPP) 1643 return (error); 1644 return (0); 1645 } 1646 } 1647 if ((mp = *mpp) == NULL) 1648 return (0); 1649 1650 if (!vn_suspendable(mp)) { 1651 if (vp != NULL || (flags & V_MNTREF) != 0) 1652 vfs_rel(mp); 1653 return (0); 1654 } 1655 1656 /* 1657 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 1658 * a vfs_ref(). 1659 * As long as a vnode is not provided we need to acquire a 1660 * refcount for the provided mountpoint too, in order to 1661 * emulate a vfs_ref(). 1662 */ 1663 MNT_ILOCK(mp); 1664 if (vp == NULL && (flags & V_MNTREF) == 0) 1665 MNT_REF(mp); 1666 1667 return (vn_start_write_locked(mp, flags)); 1668 } 1669 1670 /* 1671 * Secondary suspension. Used by operations such as vop_inactive 1672 * routines that are needed by the higher level functions. These 1673 * are allowed to proceed until all the higher level functions have 1674 * completed (indicated by mnt_writeopcount dropping to zero). At that 1675 * time, these operations are halted until the suspension is over. 1676 */ 1677 int 1678 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags) 1679 { 1680 struct mount *mp; 1681 int error; 1682 1683 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), 1684 ("V_MNTREF requires mp")); 1685 1686 retry: 1687 if (vp != NULL) { 1688 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 1689 *mpp = NULL; 1690 if (error != EOPNOTSUPP) 1691 return (error); 1692 return (0); 1693 } 1694 } 1695 /* 1696 * If we are not suspended or have not yet reached suspended 1697 * mode, then let the operation proceed. 1698 */ 1699 if ((mp = *mpp) == NULL) 1700 return (0); 1701 1702 if (!vn_suspendable(mp)) { 1703 if (vp != NULL || (flags & V_MNTREF) != 0) 1704 vfs_rel(mp); 1705 return (0); 1706 } 1707 1708 /* 1709 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 1710 * a vfs_ref(). 1711 * As long as a vnode is not provided we need to acquire a 1712 * refcount for the provided mountpoint too, in order to 1713 * emulate a vfs_ref(). 1714 */ 1715 MNT_ILOCK(mp); 1716 if (vp == NULL && (flags & V_MNTREF) == 0) 1717 MNT_REF(mp); 1718 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) { 1719 mp->mnt_secondary_writes++; 1720 mp->mnt_secondary_accwrites++; 1721 MNT_IUNLOCK(mp); 1722 return (0); 1723 } 1724 if (flags & V_NOWAIT) { 1725 MNT_REL(mp); 1726 MNT_IUNLOCK(mp); 1727 return (EWOULDBLOCK); 1728 } 1729 /* 1730 * Wait for the suspension to finish. 1731 */ 1732 error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP | 1733 ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0), 1734 "suspfs", 0); 1735 vfs_rel(mp); 1736 if (error == 0) 1737 goto retry; 1738 return (error); 1739 } 1740 1741 /* 1742 * Filesystem write operation has completed. If we are suspending and this 1743 * operation is the last one, notify the suspender that the suspension is 1744 * now in effect. 1745 */ 1746 void 1747 vn_finished_write(struct mount *mp) 1748 { 1749 if (mp == NULL || !vn_suspendable(mp)) 1750 return; 1751 MNT_ILOCK(mp); 1752 MNT_REL(mp); 1753 mp->mnt_writeopcount--; 1754 if (mp->mnt_writeopcount < 0) 1755 panic("vn_finished_write: neg cnt"); 1756 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1757 mp->mnt_writeopcount <= 0) 1758 wakeup(&mp->mnt_writeopcount); 1759 MNT_IUNLOCK(mp); 1760 } 1761 1762 1763 /* 1764 * Filesystem secondary write operation has completed. If we are 1765 * suspending and this operation is the last one, notify the suspender 1766 * that the suspension is now in effect. 1767 */ 1768 void 1769 vn_finished_secondary_write(struct mount *mp) 1770 { 1771 if (mp == NULL || !vn_suspendable(mp)) 1772 return; 1773 MNT_ILOCK(mp); 1774 MNT_REL(mp); 1775 mp->mnt_secondary_writes--; 1776 if (mp->mnt_secondary_writes < 0) 1777 panic("vn_finished_secondary_write: neg cnt"); 1778 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1779 mp->mnt_secondary_writes <= 0) 1780 wakeup(&mp->mnt_secondary_writes); 1781 MNT_IUNLOCK(mp); 1782 } 1783 1784 1785 1786 /* 1787 * Request a filesystem to suspend write operations. 1788 */ 1789 int 1790 vfs_write_suspend(struct mount *mp, int flags) 1791 { 1792 int error; 1793 1794 MPASS(vn_suspendable(mp)); 1795 1796 MNT_ILOCK(mp); 1797 if (mp->mnt_susp_owner == curthread) { 1798 MNT_IUNLOCK(mp); 1799 return (EALREADY); 1800 } 1801 while (mp->mnt_kern_flag & MNTK_SUSPEND) 1802 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0); 1803 1804 /* 1805 * Unmount holds a write reference on the mount point. If we 1806 * own busy reference and drain for writers, we deadlock with 1807 * the reference draining in the unmount path. Callers of 1808 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if 1809 * vfs_busy() reference is owned and caller is not in the 1810 * unmount context. 1811 */ 1812 if ((flags & VS_SKIP_UNMOUNT) != 0 && 1813 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { 1814 MNT_IUNLOCK(mp); 1815 return (EBUSY); 1816 } 1817 1818 mp->mnt_kern_flag |= MNTK_SUSPEND; 1819 mp->mnt_susp_owner = curthread; 1820 if (mp->mnt_writeopcount > 0) 1821 (void) msleep(&mp->mnt_writeopcount, 1822 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0); 1823 else 1824 MNT_IUNLOCK(mp); 1825 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0) 1826 vfs_write_resume(mp, 0); 1827 return (error); 1828 } 1829 1830 /* 1831 * Request a filesystem to resume write operations. 1832 */ 1833 void 1834 vfs_write_resume(struct mount *mp, int flags) 1835 { 1836 1837 MPASS(vn_suspendable(mp)); 1838 1839 MNT_ILOCK(mp); 1840 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 1841 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner")); 1842 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 | 1843 MNTK_SUSPENDED); 1844 mp->mnt_susp_owner = NULL; 1845 wakeup(&mp->mnt_writeopcount); 1846 wakeup(&mp->mnt_flag); 1847 curthread->td_pflags &= ~TDP_IGNSUSP; 1848 if ((flags & VR_START_WRITE) != 0) { 1849 MNT_REF(mp); 1850 mp->mnt_writeopcount++; 1851 } 1852 MNT_IUNLOCK(mp); 1853 if ((flags & VR_NO_SUSPCLR) == 0) 1854 VFS_SUSP_CLEAN(mp); 1855 } else if ((flags & VR_START_WRITE) != 0) { 1856 MNT_REF(mp); 1857 vn_start_write_locked(mp, 0); 1858 } else { 1859 MNT_IUNLOCK(mp); 1860 } 1861 } 1862 1863 /* 1864 * Helper loop around vfs_write_suspend() for filesystem unmount VFS 1865 * methods. 1866 */ 1867 int 1868 vfs_write_suspend_umnt(struct mount *mp) 1869 { 1870 int error; 1871 1872 MPASS(vn_suspendable(mp)); 1873 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0, 1874 ("vfs_write_suspend_umnt: recursed")); 1875 1876 /* dounmount() already called vn_start_write(). */ 1877 for (;;) { 1878 vn_finished_write(mp); 1879 error = vfs_write_suspend(mp, 0); 1880 if (error != 0) { 1881 vn_start_write(NULL, &mp, V_WAIT); 1882 return (error); 1883 } 1884 MNT_ILOCK(mp); 1885 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0) 1886 break; 1887 MNT_IUNLOCK(mp); 1888 vn_start_write(NULL, &mp, V_WAIT); 1889 } 1890 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); 1891 wakeup(&mp->mnt_flag); 1892 MNT_IUNLOCK(mp); 1893 curthread->td_pflags |= TDP_IGNSUSP; 1894 return (0); 1895 } 1896 1897 /* 1898 * Implement kqueues for files by translating it to vnode operation. 1899 */ 1900 static int 1901 vn_kqfilter(struct file *fp, struct knote *kn) 1902 { 1903 1904 return (VOP_KQFILTER(fp->f_vnode, kn)); 1905 } 1906 1907 /* 1908 * Simplified in-kernel wrapper calls for extended attribute access. 1909 * Both calls pass in a NULL credential, authorizing as "kernel" access. 1910 * Set IO_NODELOCKED in ioflg if the vnode is already locked. 1911 */ 1912 int 1913 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, 1914 const char *attrname, int *buflen, char *buf, struct thread *td) 1915 { 1916 struct uio auio; 1917 struct iovec iov; 1918 int error; 1919 1920 iov.iov_len = *buflen; 1921 iov.iov_base = buf; 1922 1923 auio.uio_iov = &iov; 1924 auio.uio_iovcnt = 1; 1925 auio.uio_rw = UIO_READ; 1926 auio.uio_segflg = UIO_SYSSPACE; 1927 auio.uio_td = td; 1928 auio.uio_offset = 0; 1929 auio.uio_resid = *buflen; 1930 1931 if ((ioflg & IO_NODELOCKED) == 0) 1932 vn_lock(vp, LK_SHARED | LK_RETRY); 1933 1934 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1935 1936 /* authorize attribute retrieval as kernel */ 1937 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL, 1938 td); 1939 1940 if ((ioflg & IO_NODELOCKED) == 0) 1941 VOP_UNLOCK(vp, 0); 1942 1943 if (error == 0) { 1944 *buflen = *buflen - auio.uio_resid; 1945 } 1946 1947 return (error); 1948 } 1949 1950 /* 1951 * XXX failure mode if partially written? 1952 */ 1953 int 1954 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, 1955 const char *attrname, int buflen, char *buf, struct thread *td) 1956 { 1957 struct uio auio; 1958 struct iovec iov; 1959 struct mount *mp; 1960 int error; 1961 1962 iov.iov_len = buflen; 1963 iov.iov_base = buf; 1964 1965 auio.uio_iov = &iov; 1966 auio.uio_iovcnt = 1; 1967 auio.uio_rw = UIO_WRITE; 1968 auio.uio_segflg = UIO_SYSSPACE; 1969 auio.uio_td = td; 1970 auio.uio_offset = 0; 1971 auio.uio_resid = buflen; 1972 1973 if ((ioflg & IO_NODELOCKED) == 0) { 1974 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 1975 return (error); 1976 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1977 } 1978 1979 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1980 1981 /* authorize attribute setting as kernel */ 1982 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td); 1983 1984 if ((ioflg & IO_NODELOCKED) == 0) { 1985 vn_finished_write(mp); 1986 VOP_UNLOCK(vp, 0); 1987 } 1988 1989 return (error); 1990 } 1991 1992 int 1993 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, 1994 const char *attrname, struct thread *td) 1995 { 1996 struct mount *mp; 1997 int error; 1998 1999 if ((ioflg & IO_NODELOCKED) == 0) { 2000 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 2001 return (error); 2002 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2003 } 2004 2005 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 2006 2007 /* authorize attribute removal as kernel */ 2008 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td); 2009 if (error == EOPNOTSUPP) 2010 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, 2011 NULL, td); 2012 2013 if ((ioflg & IO_NODELOCKED) == 0) { 2014 vn_finished_write(mp); 2015 VOP_UNLOCK(vp, 0); 2016 } 2017 2018 return (error); 2019 } 2020 2021 static int 2022 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags, 2023 struct vnode **rvp) 2024 { 2025 2026 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp)); 2027 } 2028 2029 int 2030 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp) 2031 { 2032 2033 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino, 2034 lkflags, rvp)); 2035 } 2036 2037 int 2038 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg, 2039 int lkflags, struct vnode **rvp) 2040 { 2041 struct mount *mp; 2042 int ltype, error; 2043 2044 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get"); 2045 mp = vp->v_mount; 2046 ltype = VOP_ISLOCKED(vp); 2047 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED, 2048 ("vn_vget_ino: vp not locked")); 2049 error = vfs_busy(mp, MBF_NOWAIT); 2050 if (error != 0) { 2051 vfs_ref(mp); 2052 VOP_UNLOCK(vp, 0); 2053 error = vfs_busy(mp, 0); 2054 vn_lock(vp, ltype | LK_RETRY); 2055 vfs_rel(mp); 2056 if (error != 0) 2057 return (ENOENT); 2058 if (vp->v_iflag & VI_DOOMED) { 2059 vfs_unbusy(mp); 2060 return (ENOENT); 2061 } 2062 } 2063 VOP_UNLOCK(vp, 0); 2064 error = alloc(mp, alloc_arg, lkflags, rvp); 2065 vfs_unbusy(mp); 2066 if (*rvp != vp) 2067 vn_lock(vp, ltype | LK_RETRY); 2068 if (vp->v_iflag & VI_DOOMED) { 2069 if (error == 0) { 2070 if (*rvp == vp) 2071 vunref(vp); 2072 else 2073 vput(*rvp); 2074 } 2075 error = ENOENT; 2076 } 2077 return (error); 2078 } 2079 2080 int 2081 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, 2082 struct thread *td) 2083 { 2084 2085 if (vp->v_type != VREG || td == NULL) 2086 return (0); 2087 if ((uoff_t)uio->uio_offset + uio->uio_resid > 2088 lim_cur(td, RLIMIT_FSIZE)) { 2089 PROC_LOCK(td->td_proc); 2090 kern_psignal(td->td_proc, SIGXFSZ); 2091 PROC_UNLOCK(td->td_proc); 2092 return (EFBIG); 2093 } 2094 return (0); 2095 } 2096 2097 int 2098 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 2099 struct thread *td) 2100 { 2101 struct vnode *vp; 2102 2103 vp = fp->f_vnode; 2104 #ifdef AUDIT 2105 vn_lock(vp, LK_SHARED | LK_RETRY); 2106 AUDIT_ARG_VNODE1(vp); 2107 VOP_UNLOCK(vp, 0); 2108 #endif 2109 return (setfmode(td, active_cred, vp, mode)); 2110 } 2111 2112 int 2113 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 2114 struct thread *td) 2115 { 2116 struct vnode *vp; 2117 2118 vp = fp->f_vnode; 2119 #ifdef AUDIT 2120 vn_lock(vp, LK_SHARED | LK_RETRY); 2121 AUDIT_ARG_VNODE1(vp); 2122 VOP_UNLOCK(vp, 0); 2123 #endif 2124 return (setfown(td, active_cred, vp, uid, gid)); 2125 } 2126 2127 void 2128 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end) 2129 { 2130 vm_object_t object; 2131 2132 if ((object = vp->v_object) == NULL) 2133 return; 2134 VM_OBJECT_WLOCK(object); 2135 vm_object_page_remove(object, start, end, 0); 2136 VM_OBJECT_WUNLOCK(object); 2137 } 2138 2139 int 2140 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred) 2141 { 2142 struct vattr va; 2143 daddr_t bn, bnp; 2144 uint64_t bsize; 2145 off_t noff; 2146 int error; 2147 2148 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA, 2149 ("Wrong command %lu", cmd)); 2150 2151 if (vn_lock(vp, LK_SHARED) != 0) 2152 return (EBADF); 2153 if (vp->v_type != VREG) { 2154 error = ENOTTY; 2155 goto unlock; 2156 } 2157 error = VOP_GETATTR(vp, &va, cred); 2158 if (error != 0) 2159 goto unlock; 2160 noff = *off; 2161 if (noff >= va.va_size) { 2162 error = ENXIO; 2163 goto unlock; 2164 } 2165 bsize = vp->v_mount->mnt_stat.f_iosize; 2166 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) { 2167 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL); 2168 if (error == EOPNOTSUPP) { 2169 error = ENOTTY; 2170 goto unlock; 2171 } 2172 if ((bnp == -1 && cmd == FIOSEEKHOLE) || 2173 (bnp != -1 && cmd == FIOSEEKDATA)) { 2174 noff = bn * bsize; 2175 if (noff < *off) 2176 noff = *off; 2177 goto unlock; 2178 } 2179 } 2180 if (noff > va.va_size) 2181 noff = va.va_size; 2182 /* noff == va.va_size. There is an implicit hole at the end of file. */ 2183 if (cmd == FIOSEEKDATA) 2184 error = ENXIO; 2185 unlock: 2186 VOP_UNLOCK(vp, 0); 2187 if (error == 0) 2188 *off = noff; 2189 return (error); 2190 } 2191 2192 int 2193 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td) 2194 { 2195 struct ucred *cred; 2196 struct vnode *vp; 2197 struct vattr vattr; 2198 off_t foffset, size; 2199 int error, noneg; 2200 2201 cred = td->td_ucred; 2202 vp = fp->f_vnode; 2203 foffset = foffset_lock(fp, 0); 2204 noneg = (vp->v_type != VCHR); 2205 error = 0; 2206 switch (whence) { 2207 case L_INCR: 2208 if (noneg && 2209 (foffset < 0 || 2210 (offset > 0 && foffset > OFF_MAX - offset))) { 2211 error = EOVERFLOW; 2212 break; 2213 } 2214 offset += foffset; 2215 break; 2216 case L_XTND: 2217 vn_lock(vp, LK_SHARED | LK_RETRY); 2218 error = VOP_GETATTR(vp, &vattr, cred); 2219 VOP_UNLOCK(vp, 0); 2220 if (error) 2221 break; 2222 2223 /* 2224 * If the file references a disk device, then fetch 2225 * the media size and use that to determine the ending 2226 * offset. 2227 */ 2228 if (vattr.va_size == 0 && vp->v_type == VCHR && 2229 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0) 2230 vattr.va_size = size; 2231 if (noneg && 2232 (vattr.va_size > OFF_MAX || 2233 (offset > 0 && vattr.va_size > OFF_MAX - offset))) { 2234 error = EOVERFLOW; 2235 break; 2236 } 2237 offset += vattr.va_size; 2238 break; 2239 case L_SET: 2240 break; 2241 case SEEK_DATA: 2242 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td); 2243 break; 2244 case SEEK_HOLE: 2245 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td); 2246 break; 2247 default: 2248 error = EINVAL; 2249 } 2250 if (error == 0 && noneg && offset < 0) 2251 error = EINVAL; 2252 if (error != 0) 2253 goto drop; 2254 VFS_KNOTE_UNLOCKED(vp, 0); 2255 td->td_uretoff.tdu_off = offset; 2256 drop: 2257 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); 2258 return (error); 2259 } 2260 2261 int 2262 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred, 2263 struct thread *td) 2264 { 2265 int error; 2266 2267 /* 2268 * Grant permission if the caller is the owner of the file, or 2269 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on 2270 * on the file. If the time pointer is null, then write 2271 * permission on the file is also sufficient. 2272 * 2273 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes: 2274 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES 2275 * will be allowed to set the times [..] to the current 2276 * server time. 2277 */ 2278 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td); 2279 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0) 2280 error = VOP_ACCESS(vp, VWRITE, cred, td); 2281 return (error); 2282 } 2283 2284 int 2285 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 2286 { 2287 struct vnode *vp; 2288 int error; 2289 2290 if (fp->f_type == DTYPE_FIFO) 2291 kif->kf_type = KF_TYPE_FIFO; 2292 else 2293 kif->kf_type = KF_TYPE_VNODE; 2294 vp = fp->f_vnode; 2295 vref(vp); 2296 FILEDESC_SUNLOCK(fdp); 2297 error = vn_fill_kinfo_vnode(vp, kif); 2298 vrele(vp); 2299 FILEDESC_SLOCK(fdp); 2300 return (error); 2301 } 2302 2303 static inline void 2304 vn_fill_junk(struct kinfo_file *kif) 2305 { 2306 size_t len, olen; 2307 2308 /* 2309 * Simulate vn_fullpath returning changing values for a given 2310 * vp during e.g. coredump. 2311 */ 2312 len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1; 2313 olen = strlen(kif->kf_path); 2314 if (len < olen) 2315 strcpy(&kif->kf_path[len - 1], "$"); 2316 else 2317 for (; olen < len; olen++) 2318 strcpy(&kif->kf_path[olen], "A"); 2319 } 2320 2321 int 2322 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif) 2323 { 2324 struct vattr va; 2325 char *fullpath, *freepath; 2326 int error; 2327 2328 kif->kf_un.kf_file.kf_file_type = vntype_to_kinfo(vp->v_type); 2329 freepath = NULL; 2330 fullpath = "-"; 2331 error = vn_fullpath(curthread, vp, &fullpath, &freepath); 2332 if (error == 0) { 2333 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 2334 } 2335 if (freepath != NULL) 2336 free(freepath, M_TEMP); 2337 2338 KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path, 2339 vn_fill_junk(kif); 2340 ); 2341 2342 /* 2343 * Retrieve vnode attributes. 2344 */ 2345 va.va_fsid = VNOVAL; 2346 va.va_rdev = NODEV; 2347 vn_lock(vp, LK_SHARED | LK_RETRY); 2348 error = VOP_GETATTR(vp, &va, curthread->td_ucred); 2349 VOP_UNLOCK(vp, 0); 2350 if (error != 0) 2351 return (error); 2352 if (va.va_fsid != VNOVAL) 2353 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; 2354 else 2355 kif->kf_un.kf_file.kf_file_fsid = 2356 vp->v_mount->mnt_stat.f_fsid.val[0]; 2357 kif->kf_un.kf_file.kf_file_fsid_freebsd11 = 2358 kif->kf_un.kf_file.kf_file_fsid; /* truncate */ 2359 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; 2360 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); 2361 kif->kf_un.kf_file.kf_file_size = va.va_size; 2362 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; 2363 kif->kf_un.kf_file.kf_file_rdev_freebsd11 = 2364 kif->kf_un.kf_file.kf_file_rdev; /* truncate */ 2365 return (0); 2366 } 2367 2368 int 2369 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, 2370 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, 2371 struct thread *td) 2372 { 2373 #ifdef HWPMC_HOOKS 2374 struct pmckern_map_in pkm; 2375 #endif 2376 struct mount *mp; 2377 struct vnode *vp; 2378 vm_object_t object; 2379 vm_prot_t maxprot; 2380 boolean_t writecounted; 2381 int error; 2382 2383 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \ 2384 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) 2385 /* 2386 * POSIX shared-memory objects are defined to have 2387 * kernel persistence, and are not defined to support 2388 * read(2)/write(2) -- or even open(2). Thus, we can 2389 * use MAP_ASYNC to trade on-disk coherence for speed. 2390 * The shm_open(3) library routine turns on the FPOSIXSHM 2391 * flag to request this behavior. 2392 */ 2393 if ((fp->f_flag & FPOSIXSHM) != 0) 2394 flags |= MAP_NOSYNC; 2395 #endif 2396 vp = fp->f_vnode; 2397 2398 /* 2399 * Ensure that file and memory protections are 2400 * compatible. Note that we only worry about 2401 * writability if mapping is shared; in this case, 2402 * current and max prot are dictated by the open file. 2403 * XXX use the vnode instead? Problem is: what 2404 * credentials do we use for determination? What if 2405 * proc does a setuid? 2406 */ 2407 mp = vp->v_mount; 2408 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) { 2409 maxprot = VM_PROT_NONE; 2410 if ((prot & VM_PROT_EXECUTE) != 0) 2411 return (EACCES); 2412 } else 2413 maxprot = VM_PROT_EXECUTE; 2414 if ((fp->f_flag & FREAD) != 0) 2415 maxprot |= VM_PROT_READ; 2416 else if ((prot & VM_PROT_READ) != 0) 2417 return (EACCES); 2418 2419 /* 2420 * If we are sharing potential changes via MAP_SHARED and we 2421 * are trying to get write permission although we opened it 2422 * without asking for it, bail out. 2423 */ 2424 if ((flags & MAP_SHARED) != 0) { 2425 if ((fp->f_flag & FWRITE) != 0) 2426 maxprot |= VM_PROT_WRITE; 2427 else if ((prot & VM_PROT_WRITE) != 0) 2428 return (EACCES); 2429 } else { 2430 maxprot |= VM_PROT_WRITE; 2431 cap_maxprot |= VM_PROT_WRITE; 2432 } 2433 maxprot &= cap_maxprot; 2434 2435 /* 2436 * For regular files and shared memory, POSIX requires that 2437 * the value of foff be a legitimate offset within the data 2438 * object. In particular, negative offsets are invalid. 2439 * Blocking negative offsets and overflows here avoids 2440 * possible wraparound or user-level access into reserved 2441 * ranges of the data object later. In contrast, POSIX does 2442 * not dictate how offsets are used by device drivers, so in 2443 * the case of a device mapping a negative offset is passed 2444 * on. 2445 */ 2446 if ( 2447 #ifdef _LP64 2448 size > OFF_MAX || 2449 #endif 2450 foff < 0 || foff > OFF_MAX - size) 2451 return (EINVAL); 2452 2453 writecounted = FALSE; 2454 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp, 2455 &foff, &object, &writecounted); 2456 if (error != 0) 2457 return (error); 2458 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 2459 foff, writecounted, td); 2460 if (error != 0) { 2461 /* 2462 * If this mapping was accounted for in the vnode's 2463 * writecount, then undo that now. 2464 */ 2465 if (writecounted) 2466 vnode_pager_release_writecount(object, 0, size); 2467 vm_object_deallocate(object); 2468 } 2469 #ifdef HWPMC_HOOKS 2470 /* Inform hwpmc(4) if an executable is being mapped. */ 2471 if (PMC_HOOK_INSTALLED(PMC_FN_MMAP)) { 2472 if ((prot & VM_PROT_EXECUTE) != 0 && error == 0) { 2473 pkm.pm_file = vp; 2474 pkm.pm_address = (uintptr_t) *addr; 2475 PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_MMAP, (void *) &pkm); 2476 } 2477 } 2478 #endif 2479 return (error); 2480 } 2481 2482 void 2483 vn_fsid(struct vnode *vp, struct vattr *va) 2484 { 2485 fsid_t *f; 2486 2487 f = &vp->v_mount->mnt_stat.f_fsid; 2488 va->va_fsid = (uint32_t)f->val[1]; 2489 va->va_fsid <<= sizeof(f->val[1]) * NBBY; 2490 va->va_fsid += (uint32_t)f->val[0]; 2491 } 2492