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(ndp, flagp, cmode, fp) 176 struct nameidata *ndp; 177 int *flagp, cmode; 178 struct file *fp; 179 { 180 struct thread *td = ndp->ni_cnd.cn_thread; 181 182 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp)); 183 } 184 185 /* 186 * Common code for vnode open operations via a name lookup. 187 * Lookup the vnode and invoke VOP_CREATE if needed. 188 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine. 189 * 190 * Note that this does NOT free nameidata for the successful case, 191 * due to the NDINIT being done elsewhere. 192 */ 193 int 194 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, 195 struct ucred *cred, struct file *fp) 196 { 197 struct vnode *vp; 198 struct mount *mp; 199 struct thread *td = ndp->ni_cnd.cn_thread; 200 struct vattr vat; 201 struct vattr *vap = &vat; 202 int fmode, error; 203 204 restart: 205 fmode = *flagp; 206 if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT | 207 O_EXCL | O_DIRECTORY)) 208 return (EINVAL); 209 else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) { 210 ndp->ni_cnd.cn_nameiop = CREATE; 211 /* 212 * Set NOCACHE to avoid flushing the cache when 213 * rolling in many files at once. 214 */ 215 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE; 216 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 217 ndp->ni_cnd.cn_flags |= FOLLOW; 218 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 219 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 220 if (vn_open_flags & VN_OPEN_NOCAPCHECK) 221 ndp->ni_cnd.cn_flags |= NOCAPCHECK; 222 bwillwrite(); 223 if ((error = namei(ndp)) != 0) 224 return (error); 225 if (ndp->ni_vp == NULL) { 226 VATTR_NULL(vap); 227 vap->va_type = VREG; 228 vap->va_mode = cmode; 229 if (fmode & O_EXCL) 230 vap->va_vaflags |= VA_EXCLUSIVE; 231 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) { 232 NDFREE(ndp, NDF_ONLY_PNBUF); 233 vput(ndp->ni_dvp); 234 if ((error = vn_start_write(NULL, &mp, 235 V_XSLEEP | PCATCH)) != 0) 236 return (error); 237 goto restart; 238 } 239 if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0) 240 ndp->ni_cnd.cn_flags |= MAKEENTRY; 241 #ifdef MAC 242 error = mac_vnode_check_create(cred, ndp->ni_dvp, 243 &ndp->ni_cnd, vap); 244 if (error == 0) 245 #endif 246 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 247 &ndp->ni_cnd, vap); 248 vput(ndp->ni_dvp); 249 vn_finished_write(mp); 250 if (error) { 251 NDFREE(ndp, NDF_ONLY_PNBUF); 252 return (error); 253 } 254 fmode &= ~O_TRUNC; 255 vp = ndp->ni_vp; 256 } else { 257 if (ndp->ni_dvp == ndp->ni_vp) 258 vrele(ndp->ni_dvp); 259 else 260 vput(ndp->ni_dvp); 261 ndp->ni_dvp = NULL; 262 vp = ndp->ni_vp; 263 if (fmode & O_EXCL) { 264 error = EEXIST; 265 goto bad; 266 } 267 fmode &= ~O_CREAT; 268 } 269 } else { 270 ndp->ni_cnd.cn_nameiop = LOOKUP; 271 ndp->ni_cnd.cn_flags = ISOPEN | 272 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF; 273 if (!(fmode & FWRITE)) 274 ndp->ni_cnd.cn_flags |= LOCKSHARED; 275 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 276 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 277 if (vn_open_flags & VN_OPEN_NOCAPCHECK) 278 ndp->ni_cnd.cn_flags |= NOCAPCHECK; 279 if ((error = namei(ndp)) != 0) 280 return (error); 281 vp = ndp->ni_vp; 282 } 283 error = vn_open_vnode(vp, fmode, cred, td, fp); 284 if (error) 285 goto bad; 286 *flagp = fmode; 287 return (0); 288 bad: 289 NDFREE(ndp, NDF_ONLY_PNBUF); 290 vput(vp); 291 *flagp = fmode; 292 ndp->ni_vp = NULL; 293 return (error); 294 } 295 296 /* 297 * Common code for vnode open operations once a vnode is located. 298 * Check permissions, and call the VOP_OPEN routine. 299 */ 300 int 301 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, 302 struct thread *td, struct file *fp) 303 { 304 accmode_t accmode; 305 struct flock lf; 306 int error, lock_flags, type; 307 308 if (vp->v_type == VLNK) 309 return (EMLINK); 310 if (vp->v_type == VSOCK) 311 return (EOPNOTSUPP); 312 if (vp->v_type != VDIR && fmode & O_DIRECTORY) 313 return (ENOTDIR); 314 accmode = 0; 315 if (fmode & (FWRITE | O_TRUNC)) { 316 if (vp->v_type == VDIR) 317 return (EISDIR); 318 accmode |= VWRITE; 319 } 320 if (fmode & FREAD) 321 accmode |= VREAD; 322 if (fmode & FEXEC) 323 accmode |= VEXEC; 324 if ((fmode & O_APPEND) && (fmode & FWRITE)) 325 accmode |= VAPPEND; 326 #ifdef MAC 327 if (fmode & O_CREAT) 328 accmode |= VCREAT; 329 if (fmode & O_VERIFY) 330 accmode |= VVERIFY; 331 error = mac_vnode_check_open(cred, vp, accmode); 332 if (error) 333 return (error); 334 335 accmode &= ~(VCREAT | VVERIFY); 336 #endif 337 if ((fmode & O_CREAT) == 0) { 338 if (accmode & VWRITE) { 339 error = vn_writechk(vp); 340 if (error) 341 return (error); 342 } 343 if (accmode) { 344 error = VOP_ACCESS(vp, accmode, cred, td); 345 if (error) 346 return (error); 347 } 348 } 349 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) 350 vn_lock(vp, LK_UPGRADE | LK_RETRY); 351 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0) 352 return (error); 353 354 while ((fmode & (O_EXLOCK | O_SHLOCK)) != 0) { 355 KASSERT(fp != NULL, ("open with flock requires fp")); 356 if (fp->f_type != DTYPE_NONE && fp->f_type != DTYPE_VNODE) { 357 error = EOPNOTSUPP; 358 break; 359 } 360 lock_flags = VOP_ISLOCKED(vp); 361 VOP_UNLOCK(vp, 0); 362 lf.l_whence = SEEK_SET; 363 lf.l_start = 0; 364 lf.l_len = 0; 365 if (fmode & O_EXLOCK) 366 lf.l_type = F_WRLCK; 367 else 368 lf.l_type = F_RDLCK; 369 type = F_FLOCK; 370 if ((fmode & FNONBLOCK) == 0) 371 type |= F_WAIT; 372 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type); 373 if (error == 0) 374 fp->f_flag |= FHASLOCK; 375 vn_lock(vp, lock_flags | LK_RETRY); 376 if (error != 0) 377 break; 378 if ((vp->v_iflag & VI_DOOMED) != 0) { 379 error = ENOENT; 380 break; 381 } 382 383 /* 384 * Another thread might have used this vnode as an 385 * executable while the vnode lock was dropped. 386 * Ensure the vnode is still able to be opened for 387 * writing after the lock has been obtained. 388 */ 389 if ((accmode & VWRITE) != 0) 390 error = vn_writechk(vp); 391 break; 392 } 393 394 if (error != 0) { 395 fp->f_flag |= FOPENFAILED; 396 fp->f_vnode = vp; 397 if (fp->f_ops == &badfileops) { 398 fp->f_type = DTYPE_VNODE; 399 fp->f_ops = &vnops; 400 } 401 vref(vp); 402 } else if ((fmode & FWRITE) != 0) { 403 VOP_ADD_WRITECOUNT(vp, 1); 404 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", 405 __func__, vp, vp->v_writecount); 406 } 407 ASSERT_VOP_LOCKED(vp, "vn_open_vnode"); 408 return (error); 409 } 410 411 /* 412 * Check for write permissions on the specified vnode. 413 * Prototype text segments cannot be written. 414 */ 415 int 416 vn_writechk(struct vnode *vp) 417 { 418 419 ASSERT_VOP_LOCKED(vp, "vn_writechk"); 420 /* 421 * If there's shared text associated with 422 * the vnode, try to free it up once. If 423 * we fail, we can't allow writing. 424 */ 425 if (VOP_IS_TEXT(vp)) 426 return (ETXTBSY); 427 428 return (0); 429 } 430 431 /* 432 * Vnode close call 433 */ 434 static int 435 vn_close1(struct vnode *vp, int flags, struct ucred *file_cred, 436 struct thread *td, bool keep_ref) 437 { 438 struct mount *mp; 439 int error, lock_flags; 440 441 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 && 442 MNT_EXTENDED_SHARED(vp->v_mount)) 443 lock_flags = LK_SHARED; 444 else 445 lock_flags = LK_EXCLUSIVE; 446 447 vn_start_write(vp, &mp, V_WAIT); 448 vn_lock(vp, lock_flags | LK_RETRY); 449 AUDIT_ARG_VNODE1(vp); 450 if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) { 451 VNASSERT(vp->v_writecount > 0, vp, 452 ("vn_close: negative writecount")); 453 VOP_ADD_WRITECOUNT(vp, -1); 454 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", 455 __func__, vp, vp->v_writecount); 456 } 457 error = VOP_CLOSE(vp, flags, file_cred, td); 458 if (keep_ref) 459 VOP_UNLOCK(vp, 0); 460 else 461 vput(vp); 462 vn_finished_write(mp); 463 return (error); 464 } 465 466 int 467 vn_close(struct vnode *vp, int flags, struct ucred *file_cred, 468 struct thread *td) 469 { 470 471 return (vn_close1(vp, flags, file_cred, td, false)); 472 } 473 474 /* 475 * Heuristic to detect sequential operation. 476 */ 477 static int 478 sequential_heuristic(struct uio *uio, struct file *fp) 479 { 480 481 ASSERT_VOP_LOCKED(fp->f_vnode, __func__); 482 if (fp->f_flag & FRDAHEAD) 483 return (fp->f_seqcount << IO_SEQSHIFT); 484 485 /* 486 * Offset 0 is handled specially. open() sets f_seqcount to 1 so 487 * that the first I/O is normally considered to be slightly 488 * sequential. Seeking to offset 0 doesn't change sequentiality 489 * unless previous seeks have reduced f_seqcount to 0, in which 490 * case offset 0 is not special. 491 */ 492 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 493 uio->uio_offset == fp->f_nextoff) { 494 /* 495 * f_seqcount is in units of fixed-size blocks so that it 496 * depends mainly on the amount of sequential I/O and not 497 * much on the number of sequential I/O's. The fixed size 498 * of 16384 is hard-coded here since it is (not quite) just 499 * a magic size that works well here. This size is more 500 * closely related to the best I/O size for real disks than 501 * to any block size used by software. 502 */ 503 fp->f_seqcount += howmany(uio->uio_resid, 16384); 504 if (fp->f_seqcount > IO_SEQMAX) 505 fp->f_seqcount = IO_SEQMAX; 506 return (fp->f_seqcount << IO_SEQSHIFT); 507 } 508 509 /* Not sequential. Quickly draw-down sequentiality. */ 510 if (fp->f_seqcount > 1) 511 fp->f_seqcount = 1; 512 else 513 fp->f_seqcount = 0; 514 return (0); 515 } 516 517 /* 518 * Package up an I/O request on a vnode into a uio and do it. 519 */ 520 int 521 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, 522 enum uio_seg segflg, int ioflg, struct ucred *active_cred, 523 struct ucred *file_cred, ssize_t *aresid, struct thread *td) 524 { 525 struct uio auio; 526 struct iovec aiov; 527 struct mount *mp; 528 struct ucred *cred; 529 void *rl_cookie; 530 struct vn_io_fault_args args; 531 int error, lock_flags; 532 533 auio.uio_iov = &aiov; 534 auio.uio_iovcnt = 1; 535 aiov.iov_base = base; 536 aiov.iov_len = len; 537 auio.uio_resid = len; 538 auio.uio_offset = offset; 539 auio.uio_segflg = segflg; 540 auio.uio_rw = rw; 541 auio.uio_td = td; 542 error = 0; 543 544 if ((ioflg & IO_NODELOCKED) == 0) { 545 if ((ioflg & IO_RANGELOCKED) == 0) { 546 if (rw == UIO_READ) { 547 rl_cookie = vn_rangelock_rlock(vp, offset, 548 offset + len); 549 } else { 550 rl_cookie = vn_rangelock_wlock(vp, offset, 551 offset + len); 552 } 553 } else 554 rl_cookie = NULL; 555 mp = NULL; 556 if (rw == UIO_WRITE) { 557 if (vp->v_type != VCHR && 558 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) 559 != 0) 560 goto out; 561 if (MNT_SHARED_WRITES(mp) || 562 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) 563 lock_flags = LK_SHARED; 564 else 565 lock_flags = LK_EXCLUSIVE; 566 } else 567 lock_flags = LK_SHARED; 568 vn_lock(vp, lock_flags | LK_RETRY); 569 } else 570 rl_cookie = NULL; 571 572 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 573 #ifdef MAC 574 if ((ioflg & IO_NOMACCHECK) == 0) { 575 if (rw == UIO_READ) 576 error = mac_vnode_check_read(active_cred, file_cred, 577 vp); 578 else 579 error = mac_vnode_check_write(active_cred, file_cred, 580 vp); 581 } 582 #endif 583 if (error == 0) { 584 if (file_cred != NULL) 585 cred = file_cred; 586 else 587 cred = active_cred; 588 if (do_vn_io_fault(vp, &auio)) { 589 args.kind = VN_IO_FAULT_VOP; 590 args.cred = cred; 591 args.flags = ioflg; 592 args.args.vop_args.vp = vp; 593 error = vn_io_fault1(vp, &auio, &args, td); 594 } else if (rw == UIO_READ) { 595 error = VOP_READ(vp, &auio, ioflg, cred); 596 } else /* if (rw == UIO_WRITE) */ { 597 error = VOP_WRITE(vp, &auio, ioflg, cred); 598 } 599 } 600 if (aresid) 601 *aresid = auio.uio_resid; 602 else 603 if (auio.uio_resid && error == 0) 604 error = EIO; 605 if ((ioflg & IO_NODELOCKED) == 0) { 606 VOP_UNLOCK(vp, 0); 607 if (mp != NULL) 608 vn_finished_write(mp); 609 } 610 out: 611 if (rl_cookie != NULL) 612 vn_rangelock_unlock(vp, rl_cookie); 613 return (error); 614 } 615 616 /* 617 * Package up an I/O request on a vnode into a uio and do it. The I/O 618 * request is split up into smaller chunks and we try to avoid saturating 619 * the buffer cache while potentially holding a vnode locked, so we 620 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield() 621 * to give other processes a chance to lock the vnode (either other processes 622 * core'ing the same binary, or unrelated processes scanning the directory). 623 */ 624 int 625 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred, 626 file_cred, aresid, td) 627 enum uio_rw rw; 628 struct vnode *vp; 629 void *base; 630 size_t len; 631 off_t offset; 632 enum uio_seg segflg; 633 int ioflg; 634 struct ucred *active_cred; 635 struct ucred *file_cred; 636 size_t *aresid; 637 struct thread *td; 638 { 639 int error = 0; 640 ssize_t iaresid; 641 642 do { 643 int chunk; 644 645 /* 646 * Force `offset' to a multiple of MAXBSIZE except possibly 647 * for the first chunk, so that filesystems only need to 648 * write full blocks except possibly for the first and last 649 * chunks. 650 */ 651 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 652 653 if (chunk > len) 654 chunk = len; 655 if (rw != UIO_READ && vp->v_type == VREG) 656 bwillwrite(); 657 iaresid = 0; 658 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 659 ioflg, active_cred, file_cred, &iaresid, td); 660 len -= chunk; /* aresid calc already includes length */ 661 if (error) 662 break; 663 offset += chunk; 664 base = (char *)base + chunk; 665 kern_yield(PRI_USER); 666 } while (len); 667 if (aresid) 668 *aresid = len + iaresid; 669 return (error); 670 } 671 672 off_t 673 foffset_lock(struct file *fp, int flags) 674 { 675 struct mtx *mtxp; 676 off_t res; 677 678 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); 679 680 #if OFF_MAX <= LONG_MAX 681 /* 682 * Caller only wants the current f_offset value. Assume that 683 * the long and shorter integer types reads are atomic. 684 */ 685 if ((flags & FOF_NOLOCK) != 0) 686 return (fp->f_offset); 687 #endif 688 689 /* 690 * According to McKusick the vn lock was protecting f_offset here. 691 * It is now protected by the FOFFSET_LOCKED flag. 692 */ 693 mtxp = mtx_pool_find(mtxpool_sleep, fp); 694 mtx_lock(mtxp); 695 if ((flags & FOF_NOLOCK) == 0) { 696 while (fp->f_vnread_flags & FOFFSET_LOCKED) { 697 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING; 698 msleep(&fp->f_vnread_flags, mtxp, PUSER -1, 699 "vofflock", 0); 700 } 701 fp->f_vnread_flags |= FOFFSET_LOCKED; 702 } 703 res = fp->f_offset; 704 mtx_unlock(mtxp); 705 return (res); 706 } 707 708 void 709 foffset_unlock(struct file *fp, off_t val, int flags) 710 { 711 struct mtx *mtxp; 712 713 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); 714 715 #if OFF_MAX <= LONG_MAX 716 if ((flags & FOF_NOLOCK) != 0) { 717 if ((flags & FOF_NOUPDATE) == 0) 718 fp->f_offset = val; 719 if ((flags & FOF_NEXTOFF) != 0) 720 fp->f_nextoff = val; 721 return; 722 } 723 #endif 724 725 mtxp = mtx_pool_find(mtxpool_sleep, fp); 726 mtx_lock(mtxp); 727 if ((flags & FOF_NOUPDATE) == 0) 728 fp->f_offset = val; 729 if ((flags & FOF_NEXTOFF) != 0) 730 fp->f_nextoff = val; 731 if ((flags & FOF_NOLOCK) == 0) { 732 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0, 733 ("Lost FOFFSET_LOCKED")); 734 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING) 735 wakeup(&fp->f_vnread_flags); 736 fp->f_vnread_flags = 0; 737 } 738 mtx_unlock(mtxp); 739 } 740 741 void 742 foffset_lock_uio(struct file *fp, struct uio *uio, int flags) 743 { 744 745 if ((flags & FOF_OFFSET) == 0) 746 uio->uio_offset = foffset_lock(fp, flags); 747 } 748 749 void 750 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags) 751 { 752 753 if ((flags & FOF_OFFSET) == 0) 754 foffset_unlock(fp, uio->uio_offset, flags); 755 } 756 757 static int 758 get_advice(struct file *fp, struct uio *uio) 759 { 760 struct mtx *mtxp; 761 int ret; 762 763 ret = POSIX_FADV_NORMAL; 764 if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG) 765 return (ret); 766 767 mtxp = mtx_pool_find(mtxpool_sleep, fp); 768 mtx_lock(mtxp); 769 if (fp->f_advice != NULL && 770 uio->uio_offset >= fp->f_advice->fa_start && 771 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end) 772 ret = fp->f_advice->fa_advice; 773 mtx_unlock(mtxp); 774 return (ret); 775 } 776 777 /* 778 * File table vnode read routine. 779 */ 780 static int 781 vn_read(fp, uio, active_cred, flags, td) 782 struct file *fp; 783 struct uio *uio; 784 struct ucred *active_cred; 785 int flags; 786 struct thread *td; 787 { 788 struct vnode *vp; 789 off_t orig_offset; 790 int error, ioflag; 791 int advice; 792 793 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", 794 uio->uio_td, td)); 795 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); 796 vp = fp->f_vnode; 797 ioflag = 0; 798 if (fp->f_flag & FNONBLOCK) 799 ioflag |= IO_NDELAY; 800 if (fp->f_flag & O_DIRECT) 801 ioflag |= IO_DIRECT; 802 advice = get_advice(fp, uio); 803 vn_lock(vp, LK_SHARED | LK_RETRY); 804 805 switch (advice) { 806 case POSIX_FADV_NORMAL: 807 case POSIX_FADV_SEQUENTIAL: 808 case POSIX_FADV_NOREUSE: 809 ioflag |= sequential_heuristic(uio, fp); 810 break; 811 case POSIX_FADV_RANDOM: 812 /* Disable read-ahead for random I/O. */ 813 break; 814 } 815 orig_offset = uio->uio_offset; 816 817 #ifdef MAC 818 error = mac_vnode_check_read(active_cred, fp->f_cred, vp); 819 if (error == 0) 820 #endif 821 error = VOP_READ(vp, uio, ioflag, fp->f_cred); 822 fp->f_nextoff = uio->uio_offset; 823 VOP_UNLOCK(vp, 0); 824 if (error == 0 && advice == POSIX_FADV_NOREUSE && 825 orig_offset != uio->uio_offset) 826 /* 827 * Use POSIX_FADV_DONTNEED to flush pages and buffers 828 * for the backing file after a POSIX_FADV_NOREUSE 829 * read(2). 830 */ 831 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1, 832 POSIX_FADV_DONTNEED); 833 return (error); 834 } 835 836 /* 837 * File table vnode write routine. 838 */ 839 static int 840 vn_write(fp, uio, active_cred, flags, td) 841 struct file *fp; 842 struct uio *uio; 843 struct ucred *active_cred; 844 int flags; 845 struct thread *td; 846 { 847 struct vnode *vp; 848 struct mount *mp; 849 off_t orig_offset; 850 int error, ioflag, lock_flags; 851 int advice; 852 853 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", 854 uio->uio_td, td)); 855 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); 856 vp = fp->f_vnode; 857 if (vp->v_type == VREG) 858 bwillwrite(); 859 ioflag = IO_UNIT; 860 if (vp->v_type == VREG && (fp->f_flag & O_APPEND)) 861 ioflag |= IO_APPEND; 862 if (fp->f_flag & FNONBLOCK) 863 ioflag |= IO_NDELAY; 864 if (fp->f_flag & O_DIRECT) 865 ioflag |= IO_DIRECT; 866 if ((fp->f_flag & O_FSYNC) || 867 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) 868 ioflag |= IO_SYNC; 869 mp = NULL; 870 if (vp->v_type != VCHR && 871 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) 872 goto unlock; 873 874 advice = get_advice(fp, uio); 875 876 if (MNT_SHARED_WRITES(mp) || 877 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) { 878 lock_flags = LK_SHARED; 879 } else { 880 lock_flags = LK_EXCLUSIVE; 881 } 882 883 vn_lock(vp, lock_flags | LK_RETRY); 884 switch (advice) { 885 case POSIX_FADV_NORMAL: 886 case POSIX_FADV_SEQUENTIAL: 887 case POSIX_FADV_NOREUSE: 888 ioflag |= sequential_heuristic(uio, fp); 889 break; 890 case POSIX_FADV_RANDOM: 891 /* XXX: Is this correct? */ 892 break; 893 } 894 orig_offset = uio->uio_offset; 895 896 #ifdef MAC 897 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 898 if (error == 0) 899 #endif 900 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred); 901 fp->f_nextoff = uio->uio_offset; 902 VOP_UNLOCK(vp, 0); 903 if (vp->v_type != VCHR) 904 vn_finished_write(mp); 905 if (error == 0 && advice == POSIX_FADV_NOREUSE && 906 orig_offset != uio->uio_offset) 907 /* 908 * Use POSIX_FADV_DONTNEED to flush pages and buffers 909 * for the backing file after a POSIX_FADV_NOREUSE 910 * write(2). 911 */ 912 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1, 913 POSIX_FADV_DONTNEED); 914 unlock: 915 return (error); 916 } 917 918 /* 919 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to 920 * prevent the following deadlock: 921 * 922 * Assume that the thread A reads from the vnode vp1 into userspace 923 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is 924 * currently not resident, then system ends up with the call chain 925 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] -> 926 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2) 927 * which establishes lock order vp1->vn_lock, then vp2->vn_lock. 928 * If, at the same time, thread B reads from vnode vp2 into buffer buf2 929 * backed by the pages of vnode vp1, and some page in buf2 is not 930 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock. 931 * 932 * To prevent the lock order reversal and deadlock, vn_io_fault() does 933 * not allow page faults to happen during VOP_READ() or VOP_WRITE(). 934 * Instead, it first tries to do the whole range i/o with pagefaults 935 * disabled. If all pages in the i/o buffer are resident and mapped, 936 * VOP will succeed (ignoring the genuine filesystem errors). 937 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do 938 * i/o in chunks, with all pages in the chunk prefaulted and held 939 * using vm_fault_quick_hold_pages(). 940 * 941 * Filesystems using this deadlock avoidance scheme should use the 942 * array of the held pages from uio, saved in the curthread->td_ma, 943 * instead of doing uiomove(). A helper function 944 * vn_io_fault_uiomove() converts uiomove request into 945 * uiomove_fromphys() over td_ma array. 946 * 947 * Since vnode locks do not cover the whole i/o anymore, rangelocks 948 * make the current i/o request atomic with respect to other i/os and 949 * truncations. 950 */ 951 952 /* 953 * Decode vn_io_fault_args and perform the corresponding i/o. 954 */ 955 static int 956 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio, 957 struct thread *td) 958 { 959 int error, save; 960 961 error = 0; 962 save = vm_fault_disable_pagefaults(); 963 switch (args->kind) { 964 case VN_IO_FAULT_FOP: 965 error = (args->args.fop_args.doio)(args->args.fop_args.fp, 966 uio, args->cred, args->flags, td); 967 break; 968 case VN_IO_FAULT_VOP: 969 if (uio->uio_rw == UIO_READ) { 970 error = VOP_READ(args->args.vop_args.vp, uio, 971 args->flags, args->cred); 972 } else if (uio->uio_rw == UIO_WRITE) { 973 error = VOP_WRITE(args->args.vop_args.vp, uio, 974 args->flags, args->cred); 975 } 976 break; 977 default: 978 panic("vn_io_fault_doio: unknown kind of io %d %d", 979 args->kind, uio->uio_rw); 980 } 981 vm_fault_enable_pagefaults(save); 982 return (error); 983 } 984 985 static int 986 vn_io_fault_touch(char *base, const struct uio *uio) 987 { 988 int r; 989 990 r = fubyte(base); 991 if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1)) 992 return (EFAULT); 993 return (0); 994 } 995 996 static int 997 vn_io_fault_prefault_user(const struct uio *uio) 998 { 999 char *base; 1000 const struct iovec *iov; 1001 size_t len; 1002 ssize_t resid; 1003 int error, i; 1004 1005 KASSERT(uio->uio_segflg == UIO_USERSPACE, 1006 ("vn_io_fault_prefault userspace")); 1007 1008 error = i = 0; 1009 iov = uio->uio_iov; 1010 resid = uio->uio_resid; 1011 base = iov->iov_base; 1012 len = iov->iov_len; 1013 while (resid > 0) { 1014 error = vn_io_fault_touch(base, uio); 1015 if (error != 0) 1016 break; 1017 if (len < PAGE_SIZE) { 1018 if (len != 0) { 1019 error = vn_io_fault_touch(base + len - 1, uio); 1020 if (error != 0) 1021 break; 1022 resid -= len; 1023 } 1024 if (++i >= uio->uio_iovcnt) 1025 break; 1026 iov = uio->uio_iov + i; 1027 base = iov->iov_base; 1028 len = iov->iov_len; 1029 } else { 1030 len -= PAGE_SIZE; 1031 base += PAGE_SIZE; 1032 resid -= PAGE_SIZE; 1033 } 1034 } 1035 return (error); 1036 } 1037 1038 /* 1039 * Common code for vn_io_fault(), agnostic to the kind of i/o request. 1040 * Uses vn_io_fault_doio() to make the call to an actual i/o function. 1041 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request 1042 * into args and call vn_io_fault1() to handle faults during the user 1043 * mode buffer accesses. 1044 */ 1045 static int 1046 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, 1047 struct thread *td) 1048 { 1049 vm_page_t ma[io_hold_cnt + 2]; 1050 struct uio *uio_clone, short_uio; 1051 struct iovec short_iovec[1]; 1052 vm_page_t *prev_td_ma; 1053 vm_prot_t prot; 1054 vm_offset_t addr, end; 1055 size_t len, resid; 1056 ssize_t adv; 1057 int error, cnt, saveheld, prev_td_ma_cnt; 1058 1059 if (vn_io_fault_prefault) { 1060 error = vn_io_fault_prefault_user(uio); 1061 if (error != 0) 1062 return (error); /* Or ignore ? */ 1063 } 1064 1065 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ; 1066 1067 /* 1068 * The UFS follows IO_UNIT directive and replays back both 1069 * uio_offset and uio_resid if an error is encountered during the 1070 * operation. But, since the iovec may be already advanced, 1071 * uio is still in an inconsistent state. 1072 * 1073 * Cache a copy of the original uio, which is advanced to the redo 1074 * point using UIO_NOCOPY below. 1075 */ 1076 uio_clone = cloneuio(uio); 1077 resid = uio->uio_resid; 1078 1079 short_uio.uio_segflg = UIO_USERSPACE; 1080 short_uio.uio_rw = uio->uio_rw; 1081 short_uio.uio_td = uio->uio_td; 1082 1083 error = vn_io_fault_doio(args, uio, td); 1084 if (error != EFAULT) 1085 goto out; 1086 1087 atomic_add_long(&vn_io_faults_cnt, 1); 1088 uio_clone->uio_segflg = UIO_NOCOPY; 1089 uiomove(NULL, resid - uio->uio_resid, uio_clone); 1090 uio_clone->uio_segflg = uio->uio_segflg; 1091 1092 saveheld = curthread_pflags_set(TDP_UIOHELD); 1093 prev_td_ma = td->td_ma; 1094 prev_td_ma_cnt = td->td_ma_cnt; 1095 1096 while (uio_clone->uio_resid != 0) { 1097 len = uio_clone->uio_iov->iov_len; 1098 if (len == 0) { 1099 KASSERT(uio_clone->uio_iovcnt >= 1, 1100 ("iovcnt underflow")); 1101 uio_clone->uio_iov++; 1102 uio_clone->uio_iovcnt--; 1103 continue; 1104 } 1105 if (len > io_hold_cnt * PAGE_SIZE) 1106 len = io_hold_cnt * PAGE_SIZE; 1107 addr = (uintptr_t)uio_clone->uio_iov->iov_base; 1108 end = round_page(addr + len); 1109 if (end < addr) { 1110 error = EFAULT; 1111 break; 1112 } 1113 cnt = atop(end - trunc_page(addr)); 1114 /* 1115 * A perfectly misaligned address and length could cause 1116 * both the start and the end of the chunk to use partial 1117 * page. +2 accounts for such a situation. 1118 */ 1119 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map, 1120 addr, len, prot, ma, io_hold_cnt + 2); 1121 if (cnt == -1) { 1122 error = EFAULT; 1123 break; 1124 } 1125 short_uio.uio_iov = &short_iovec[0]; 1126 short_iovec[0].iov_base = (void *)addr; 1127 short_uio.uio_iovcnt = 1; 1128 short_uio.uio_resid = short_iovec[0].iov_len = len; 1129 short_uio.uio_offset = uio_clone->uio_offset; 1130 td->td_ma = ma; 1131 td->td_ma_cnt = cnt; 1132 1133 error = vn_io_fault_doio(args, &short_uio, td); 1134 vm_page_unhold_pages(ma, cnt); 1135 adv = len - short_uio.uio_resid; 1136 1137 uio_clone->uio_iov->iov_base = 1138 (char *)uio_clone->uio_iov->iov_base + adv; 1139 uio_clone->uio_iov->iov_len -= adv; 1140 uio_clone->uio_resid -= adv; 1141 uio_clone->uio_offset += adv; 1142 1143 uio->uio_resid -= adv; 1144 uio->uio_offset += adv; 1145 1146 if (error != 0 || adv == 0) 1147 break; 1148 } 1149 td->td_ma = prev_td_ma; 1150 td->td_ma_cnt = prev_td_ma_cnt; 1151 curthread_pflags_restore(saveheld); 1152 out: 1153 free(uio_clone, M_IOV); 1154 return (error); 1155 } 1156 1157 static int 1158 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred, 1159 int flags, struct thread *td) 1160 { 1161 fo_rdwr_t *doio; 1162 struct vnode *vp; 1163 void *rl_cookie; 1164 struct vn_io_fault_args args; 1165 int error; 1166 1167 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write; 1168 vp = fp->f_vnode; 1169 foffset_lock_uio(fp, uio, flags); 1170 if (do_vn_io_fault(vp, uio)) { 1171 args.kind = VN_IO_FAULT_FOP; 1172 args.args.fop_args.fp = fp; 1173 args.args.fop_args.doio = doio; 1174 args.cred = active_cred; 1175 args.flags = flags | FOF_OFFSET; 1176 if (uio->uio_rw == UIO_READ) { 1177 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset, 1178 uio->uio_offset + uio->uio_resid); 1179 } else if ((fp->f_flag & O_APPEND) != 0 || 1180 (flags & FOF_OFFSET) == 0) { 1181 /* For appenders, punt and lock the whole range. */ 1182 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); 1183 } else { 1184 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset, 1185 uio->uio_offset + uio->uio_resid); 1186 } 1187 error = vn_io_fault1(vp, uio, &args, td); 1188 vn_rangelock_unlock(vp, rl_cookie); 1189 } else { 1190 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td); 1191 } 1192 foffset_unlock_uio(fp, uio, flags); 1193 return (error); 1194 } 1195 1196 /* 1197 * Helper function to perform the requested uiomove operation using 1198 * the held pages for io->uio_iov[0].iov_base buffer instead of 1199 * copyin/copyout. Access to the pages with uiomove_fromphys() 1200 * instead of iov_base prevents page faults that could occur due to 1201 * pmap_collect() invalidating the mapping created by 1202 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or 1203 * object cleanup revoking the write access from page mappings. 1204 * 1205 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove() 1206 * instead of plain uiomove(). 1207 */ 1208 int 1209 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio) 1210 { 1211 struct uio transp_uio; 1212 struct iovec transp_iov[1]; 1213 struct thread *td; 1214 size_t adv; 1215 int error, pgadv; 1216 1217 td = curthread; 1218 if ((td->td_pflags & TDP_UIOHELD) == 0 || 1219 uio->uio_segflg != UIO_USERSPACE) 1220 return (uiomove(data, xfersize, uio)); 1221 1222 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); 1223 transp_iov[0].iov_base = data; 1224 transp_uio.uio_iov = &transp_iov[0]; 1225 transp_uio.uio_iovcnt = 1; 1226 if (xfersize > uio->uio_resid) 1227 xfersize = uio->uio_resid; 1228 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize; 1229 transp_uio.uio_offset = 0; 1230 transp_uio.uio_segflg = UIO_SYSSPACE; 1231 /* 1232 * Since transp_iov points to data, and td_ma page array 1233 * corresponds to original uio->uio_iov, we need to invert the 1234 * direction of the i/o operation as passed to 1235 * uiomove_fromphys(). 1236 */ 1237 switch (uio->uio_rw) { 1238 case UIO_WRITE: 1239 transp_uio.uio_rw = UIO_READ; 1240 break; 1241 case UIO_READ: 1242 transp_uio.uio_rw = UIO_WRITE; 1243 break; 1244 } 1245 transp_uio.uio_td = uio->uio_td; 1246 error = uiomove_fromphys(td->td_ma, 1247 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK, 1248 xfersize, &transp_uio); 1249 adv = xfersize - transp_uio.uio_resid; 1250 pgadv = 1251 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) - 1252 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT); 1253 td->td_ma += pgadv; 1254 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, 1255 pgadv)); 1256 td->td_ma_cnt -= pgadv; 1257 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv; 1258 uio->uio_iov->iov_len -= adv; 1259 uio->uio_resid -= adv; 1260 uio->uio_offset += adv; 1261 return (error); 1262 } 1263 1264 int 1265 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize, 1266 struct uio *uio) 1267 { 1268 struct thread *td; 1269 vm_offset_t iov_base; 1270 int cnt, pgadv; 1271 1272 td = curthread; 1273 if ((td->td_pflags & TDP_UIOHELD) == 0 || 1274 uio->uio_segflg != UIO_USERSPACE) 1275 return (uiomove_fromphys(ma, offset, xfersize, uio)); 1276 1277 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); 1278 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize; 1279 iov_base = (vm_offset_t)uio->uio_iov->iov_base; 1280 switch (uio->uio_rw) { 1281 case UIO_WRITE: 1282 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma, 1283 offset, cnt); 1284 break; 1285 case UIO_READ: 1286 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK, 1287 cnt); 1288 break; 1289 } 1290 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT); 1291 td->td_ma += pgadv; 1292 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, 1293 pgadv)); 1294 td->td_ma_cnt -= pgadv; 1295 uio->uio_iov->iov_base = (char *)(iov_base + cnt); 1296 uio->uio_iov->iov_len -= cnt; 1297 uio->uio_resid -= cnt; 1298 uio->uio_offset += cnt; 1299 return (0); 1300 } 1301 1302 1303 /* 1304 * File table truncate routine. 1305 */ 1306 static int 1307 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred, 1308 struct thread *td) 1309 { 1310 struct vattr vattr; 1311 struct mount *mp; 1312 struct vnode *vp; 1313 void *rl_cookie; 1314 int error; 1315 1316 vp = fp->f_vnode; 1317 1318 /* 1319 * Lock the whole range for truncation. Otherwise split i/o 1320 * might happen partly before and partly after the truncation. 1321 */ 1322 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); 1323 error = vn_start_write(vp, &mp, V_WAIT | PCATCH); 1324 if (error) 1325 goto out1; 1326 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1327 AUDIT_ARG_VNODE1(vp); 1328 if (vp->v_type == VDIR) { 1329 error = EISDIR; 1330 goto out; 1331 } 1332 #ifdef MAC 1333 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 1334 if (error) 1335 goto out; 1336 #endif 1337 error = vn_writechk(vp); 1338 if (error == 0) { 1339 VATTR_NULL(&vattr); 1340 vattr.va_size = length; 1341 if ((fp->f_flag & O_FSYNC) != 0) 1342 vattr.va_vaflags |= VA_SYNC; 1343 error = VOP_SETATTR(vp, &vattr, fp->f_cred); 1344 } 1345 out: 1346 VOP_UNLOCK(vp, 0); 1347 vn_finished_write(mp); 1348 out1: 1349 vn_rangelock_unlock(vp, rl_cookie); 1350 return (error); 1351 } 1352 1353 /* 1354 * File table vnode stat routine. 1355 */ 1356 static int 1357 vn_statfile(fp, sb, active_cred, td) 1358 struct file *fp; 1359 struct stat *sb; 1360 struct ucred *active_cred; 1361 struct thread *td; 1362 { 1363 struct vnode *vp = fp->f_vnode; 1364 int error; 1365 1366 vn_lock(vp, LK_SHARED | LK_RETRY); 1367 error = vn_stat(vp, sb, active_cred, fp->f_cred, td); 1368 VOP_UNLOCK(vp, 0); 1369 1370 return (error); 1371 } 1372 1373 /* 1374 * Stat a vnode; implementation for the stat syscall 1375 */ 1376 int 1377 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *active_cred, 1378 struct ucred *file_cred, struct thread *td) 1379 { 1380 struct vattr vattr; 1381 struct vattr *vap; 1382 int error; 1383 u_short mode; 1384 1385 AUDIT_ARG_VNODE1(vp); 1386 #ifdef MAC 1387 error = mac_vnode_check_stat(active_cred, file_cred, vp); 1388 if (error) 1389 return (error); 1390 #endif 1391 1392 vap = &vattr; 1393 1394 /* 1395 * Initialize defaults for new and unusual fields, so that file 1396 * systems which don't support these fields don't need to know 1397 * about them. 1398 */ 1399 vap->va_birthtime.tv_sec = -1; 1400 vap->va_birthtime.tv_nsec = 0; 1401 vap->va_fsid = VNOVAL; 1402 vap->va_rdev = NODEV; 1403 1404 error = VOP_GETATTR(vp, vap, active_cred); 1405 if (error) 1406 return (error); 1407 1408 /* 1409 * Zero the spare stat fields 1410 */ 1411 bzero(sb, sizeof *sb); 1412 1413 /* 1414 * Copy from vattr table 1415 */ 1416 if (vap->va_fsid != VNOVAL) 1417 sb->st_dev = vap->va_fsid; 1418 else 1419 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 1420 sb->st_ino = vap->va_fileid; 1421 mode = vap->va_mode; 1422 switch (vap->va_type) { 1423 case VREG: 1424 mode |= S_IFREG; 1425 break; 1426 case VDIR: 1427 mode |= S_IFDIR; 1428 break; 1429 case VBLK: 1430 mode |= S_IFBLK; 1431 break; 1432 case VCHR: 1433 mode |= S_IFCHR; 1434 break; 1435 case VLNK: 1436 mode |= S_IFLNK; 1437 break; 1438 case VSOCK: 1439 mode |= S_IFSOCK; 1440 break; 1441 case VFIFO: 1442 mode |= S_IFIFO; 1443 break; 1444 default: 1445 return (EBADF); 1446 } 1447 sb->st_mode = mode; 1448 sb->st_nlink = vap->va_nlink; 1449 sb->st_uid = vap->va_uid; 1450 sb->st_gid = vap->va_gid; 1451 sb->st_rdev = vap->va_rdev; 1452 if (vap->va_size > OFF_MAX) 1453 return (EOVERFLOW); 1454 sb->st_size = vap->va_size; 1455 sb->st_atim = vap->va_atime; 1456 sb->st_mtim = vap->va_mtime; 1457 sb->st_ctim = vap->va_ctime; 1458 sb->st_birthtim = vap->va_birthtime; 1459 1460 /* 1461 * According to www.opengroup.org, the meaning of st_blksize is 1462 * "a filesystem-specific preferred I/O block size for this 1463 * object. In some filesystem types, this may vary from file 1464 * to file" 1465 * Use miminum/default of PAGE_SIZE (e.g. for VCHR). 1466 */ 1467 1468 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize); 1469 1470 sb->st_flags = vap->va_flags; 1471 if (priv_check(td, PRIV_VFS_GENERATION)) 1472 sb->st_gen = 0; 1473 else 1474 sb->st_gen = vap->va_gen; 1475 1476 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 1477 return (0); 1478 } 1479 1480 /* 1481 * File table vnode ioctl routine. 1482 */ 1483 static int 1484 vn_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 1485 struct thread *td) 1486 { 1487 struct vattr vattr; 1488 struct vnode *vp; 1489 int error; 1490 1491 vp = fp->f_vnode; 1492 switch (vp->v_type) { 1493 case VDIR: 1494 case VREG: 1495 switch (com) { 1496 case FIONREAD: 1497 vn_lock(vp, LK_SHARED | LK_RETRY); 1498 error = VOP_GETATTR(vp, &vattr, active_cred); 1499 VOP_UNLOCK(vp, 0); 1500 if (error == 0) 1501 *(int *)data = vattr.va_size - fp->f_offset; 1502 return (error); 1503 case FIONBIO: 1504 case FIOASYNC: 1505 return (0); 1506 default: 1507 return (VOP_IOCTL(vp, com, data, fp->f_flag, 1508 active_cred, td)); 1509 } 1510 break; 1511 case VCHR: 1512 return (VOP_IOCTL(vp, com, data, fp->f_flag, 1513 active_cred, td)); 1514 default: 1515 return (ENOTTY); 1516 } 1517 } 1518 1519 /* 1520 * File table vnode poll routine. 1521 */ 1522 static int 1523 vn_poll(struct file *fp, int events, struct ucred *active_cred, 1524 struct thread *td) 1525 { 1526 struct vnode *vp; 1527 int error; 1528 1529 vp = fp->f_vnode; 1530 #ifdef MAC 1531 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1532 AUDIT_ARG_VNODE1(vp); 1533 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp); 1534 VOP_UNLOCK(vp, 0); 1535 if (!error) 1536 #endif 1537 1538 error = VOP_POLL(vp, events, fp->f_cred, td); 1539 return (error); 1540 } 1541 1542 /* 1543 * Acquire the requested lock and then check for validity. LK_RETRY 1544 * permits vn_lock to return doomed vnodes. 1545 */ 1546 int 1547 _vn_lock(struct vnode *vp, int flags, char *file, int line) 1548 { 1549 int error; 1550 1551 VNASSERT((flags & LK_TYPE_MASK) != 0, vp, 1552 ("vn_lock: no locktype")); 1553 VNASSERT(vp->v_holdcnt != 0, vp, ("vn_lock: zero hold count")); 1554 retry: 1555 error = VOP_LOCK1(vp, flags, file, line); 1556 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */ 1557 KASSERT((flags & LK_RETRY) == 0 || error == 0, 1558 ("vn_lock: error %d incompatible with flags %#x", error, flags)); 1559 1560 if ((flags & LK_RETRY) == 0) { 1561 if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) { 1562 VOP_UNLOCK(vp, 0); 1563 error = ENOENT; 1564 } 1565 } else if (error != 0) 1566 goto retry; 1567 return (error); 1568 } 1569 1570 /* 1571 * File table vnode close routine. 1572 */ 1573 static int 1574 vn_closefile(struct file *fp, struct thread *td) 1575 { 1576 struct vnode *vp; 1577 struct flock lf; 1578 int error; 1579 bool ref; 1580 1581 vp = fp->f_vnode; 1582 fp->f_ops = &badfileops; 1583 ref= (fp->f_flag & FHASLOCK) != 0 && fp->f_type == DTYPE_VNODE; 1584 1585 error = vn_close1(vp, fp->f_flag, fp->f_cred, td, ref); 1586 1587 if (__predict_false(ref)) { 1588 lf.l_whence = SEEK_SET; 1589 lf.l_start = 0; 1590 lf.l_len = 0; 1591 lf.l_type = F_UNLCK; 1592 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); 1593 vrele(vp); 1594 } 1595 return (error); 1596 } 1597 1598 static bool 1599 vn_suspendable(struct mount *mp) 1600 { 1601 1602 return (mp->mnt_op->vfs_susp_clean != NULL); 1603 } 1604 1605 /* 1606 * Preparing to start a filesystem write operation. If the operation is 1607 * permitted, then we bump the count of operations in progress and 1608 * proceed. If a suspend request is in progress, we wait until the 1609 * suspension is over, and then proceed. 1610 */ 1611 static int 1612 vn_start_write_locked(struct mount *mp, int flags) 1613 { 1614 int error, mflags; 1615 1616 mtx_assert(MNT_MTX(mp), MA_OWNED); 1617 error = 0; 1618 1619 /* 1620 * Check on status of suspension. 1621 */ 1622 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 || 1623 mp->mnt_susp_owner != curthread) { 1624 mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? 1625 (flags & PCATCH) : 0) | (PUSER - 1); 1626 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 1627 if (flags & V_NOWAIT) { 1628 error = EWOULDBLOCK; 1629 goto unlock; 1630 } 1631 error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags, 1632 "suspfs", 0); 1633 if (error) 1634 goto unlock; 1635 } 1636 } 1637 if (flags & V_XSLEEP) 1638 goto unlock; 1639 mp->mnt_writeopcount++; 1640 unlock: 1641 if (error != 0 || (flags & V_XSLEEP) != 0) 1642 MNT_REL(mp); 1643 MNT_IUNLOCK(mp); 1644 return (error); 1645 } 1646 1647 int 1648 vn_start_write(struct vnode *vp, struct mount **mpp, int flags) 1649 { 1650 struct mount *mp; 1651 int error; 1652 1653 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), 1654 ("V_MNTREF requires mp")); 1655 1656 error = 0; 1657 /* 1658 * If a vnode is provided, get and return the mount point that 1659 * to which it will write. 1660 */ 1661 if (vp != NULL) { 1662 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 1663 *mpp = NULL; 1664 if (error != EOPNOTSUPP) 1665 return (error); 1666 return (0); 1667 } 1668 } 1669 if ((mp = *mpp) == NULL) 1670 return (0); 1671 1672 if (!vn_suspendable(mp)) { 1673 if (vp != NULL || (flags & V_MNTREF) != 0) 1674 vfs_rel(mp); 1675 return (0); 1676 } 1677 1678 /* 1679 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 1680 * a vfs_ref(). 1681 * As long as a vnode is not provided we need to acquire a 1682 * refcount for the provided mountpoint too, in order to 1683 * emulate a vfs_ref(). 1684 */ 1685 MNT_ILOCK(mp); 1686 if (vp == NULL && (flags & V_MNTREF) == 0) 1687 MNT_REF(mp); 1688 1689 return (vn_start_write_locked(mp, flags)); 1690 } 1691 1692 /* 1693 * Secondary suspension. Used by operations such as vop_inactive 1694 * routines that are needed by the higher level functions. These 1695 * are allowed to proceed until all the higher level functions have 1696 * completed (indicated by mnt_writeopcount dropping to zero). At that 1697 * time, these operations are halted until the suspension is over. 1698 */ 1699 int 1700 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags) 1701 { 1702 struct mount *mp; 1703 int error; 1704 1705 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), 1706 ("V_MNTREF requires mp")); 1707 1708 retry: 1709 if (vp != NULL) { 1710 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 1711 *mpp = NULL; 1712 if (error != EOPNOTSUPP) 1713 return (error); 1714 return (0); 1715 } 1716 } 1717 /* 1718 * If we are not suspended or have not yet reached suspended 1719 * mode, then let the operation proceed. 1720 */ 1721 if ((mp = *mpp) == NULL) 1722 return (0); 1723 1724 if (!vn_suspendable(mp)) { 1725 if (vp != NULL || (flags & V_MNTREF) != 0) 1726 vfs_rel(mp); 1727 return (0); 1728 } 1729 1730 /* 1731 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 1732 * a vfs_ref(). 1733 * As long as a vnode is not provided we need to acquire a 1734 * refcount for the provided mountpoint too, in order to 1735 * emulate a vfs_ref(). 1736 */ 1737 MNT_ILOCK(mp); 1738 if (vp == NULL && (flags & V_MNTREF) == 0) 1739 MNT_REF(mp); 1740 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) { 1741 mp->mnt_secondary_writes++; 1742 mp->mnt_secondary_accwrites++; 1743 MNT_IUNLOCK(mp); 1744 return (0); 1745 } 1746 if (flags & V_NOWAIT) { 1747 MNT_REL(mp); 1748 MNT_IUNLOCK(mp); 1749 return (EWOULDBLOCK); 1750 } 1751 /* 1752 * Wait for the suspension to finish. 1753 */ 1754 error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP | 1755 ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0), 1756 "suspfs", 0); 1757 vfs_rel(mp); 1758 if (error == 0) 1759 goto retry; 1760 return (error); 1761 } 1762 1763 /* 1764 * Filesystem write operation has completed. If we are suspending and this 1765 * operation is the last one, notify the suspender that the suspension is 1766 * now in effect. 1767 */ 1768 void 1769 vn_finished_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_writeopcount--; 1776 if (mp->mnt_writeopcount < 0) 1777 panic("vn_finished_write: neg cnt"); 1778 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1779 mp->mnt_writeopcount <= 0) 1780 wakeup(&mp->mnt_writeopcount); 1781 MNT_IUNLOCK(mp); 1782 } 1783 1784 1785 /* 1786 * Filesystem secondary write operation has completed. If we are 1787 * suspending and this operation is the last one, notify the suspender 1788 * that the suspension is now in effect. 1789 */ 1790 void 1791 vn_finished_secondary_write(struct mount *mp) 1792 { 1793 if (mp == NULL || !vn_suspendable(mp)) 1794 return; 1795 MNT_ILOCK(mp); 1796 MNT_REL(mp); 1797 mp->mnt_secondary_writes--; 1798 if (mp->mnt_secondary_writes < 0) 1799 panic("vn_finished_secondary_write: neg cnt"); 1800 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1801 mp->mnt_secondary_writes <= 0) 1802 wakeup(&mp->mnt_secondary_writes); 1803 MNT_IUNLOCK(mp); 1804 } 1805 1806 1807 1808 /* 1809 * Request a filesystem to suspend write operations. 1810 */ 1811 int 1812 vfs_write_suspend(struct mount *mp, int flags) 1813 { 1814 int error; 1815 1816 MPASS(vn_suspendable(mp)); 1817 1818 MNT_ILOCK(mp); 1819 if (mp->mnt_susp_owner == curthread) { 1820 MNT_IUNLOCK(mp); 1821 return (EALREADY); 1822 } 1823 while (mp->mnt_kern_flag & MNTK_SUSPEND) 1824 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0); 1825 1826 /* 1827 * Unmount holds a write reference on the mount point. If we 1828 * own busy reference and drain for writers, we deadlock with 1829 * the reference draining in the unmount path. Callers of 1830 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if 1831 * vfs_busy() reference is owned and caller is not in the 1832 * unmount context. 1833 */ 1834 if ((flags & VS_SKIP_UNMOUNT) != 0 && 1835 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { 1836 MNT_IUNLOCK(mp); 1837 return (EBUSY); 1838 } 1839 1840 mp->mnt_kern_flag |= MNTK_SUSPEND; 1841 mp->mnt_susp_owner = curthread; 1842 if (mp->mnt_writeopcount > 0) 1843 (void) msleep(&mp->mnt_writeopcount, 1844 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0); 1845 else 1846 MNT_IUNLOCK(mp); 1847 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0) 1848 vfs_write_resume(mp, 0); 1849 return (error); 1850 } 1851 1852 /* 1853 * Request a filesystem to resume write operations. 1854 */ 1855 void 1856 vfs_write_resume(struct mount *mp, int flags) 1857 { 1858 1859 MPASS(vn_suspendable(mp)); 1860 1861 MNT_ILOCK(mp); 1862 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 1863 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner")); 1864 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 | 1865 MNTK_SUSPENDED); 1866 mp->mnt_susp_owner = NULL; 1867 wakeup(&mp->mnt_writeopcount); 1868 wakeup(&mp->mnt_flag); 1869 curthread->td_pflags &= ~TDP_IGNSUSP; 1870 if ((flags & VR_START_WRITE) != 0) { 1871 MNT_REF(mp); 1872 mp->mnt_writeopcount++; 1873 } 1874 MNT_IUNLOCK(mp); 1875 if ((flags & VR_NO_SUSPCLR) == 0) 1876 VFS_SUSP_CLEAN(mp); 1877 } else if ((flags & VR_START_WRITE) != 0) { 1878 MNT_REF(mp); 1879 vn_start_write_locked(mp, 0); 1880 } else { 1881 MNT_IUNLOCK(mp); 1882 } 1883 } 1884 1885 /* 1886 * Helper loop around vfs_write_suspend() for filesystem unmount VFS 1887 * methods. 1888 */ 1889 int 1890 vfs_write_suspend_umnt(struct mount *mp) 1891 { 1892 int error; 1893 1894 MPASS(vn_suspendable(mp)); 1895 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0, 1896 ("vfs_write_suspend_umnt: recursed")); 1897 1898 /* dounmount() already called vn_start_write(). */ 1899 for (;;) { 1900 vn_finished_write(mp); 1901 error = vfs_write_suspend(mp, 0); 1902 if (error != 0) { 1903 vn_start_write(NULL, &mp, V_WAIT); 1904 return (error); 1905 } 1906 MNT_ILOCK(mp); 1907 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0) 1908 break; 1909 MNT_IUNLOCK(mp); 1910 vn_start_write(NULL, &mp, V_WAIT); 1911 } 1912 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); 1913 wakeup(&mp->mnt_flag); 1914 MNT_IUNLOCK(mp); 1915 curthread->td_pflags |= TDP_IGNSUSP; 1916 return (0); 1917 } 1918 1919 /* 1920 * Implement kqueues for files by translating it to vnode operation. 1921 */ 1922 static int 1923 vn_kqfilter(struct file *fp, struct knote *kn) 1924 { 1925 1926 return (VOP_KQFILTER(fp->f_vnode, kn)); 1927 } 1928 1929 /* 1930 * Simplified in-kernel wrapper calls for extended attribute access. 1931 * Both calls pass in a NULL credential, authorizing as "kernel" access. 1932 * Set IO_NODELOCKED in ioflg if the vnode is already locked. 1933 */ 1934 int 1935 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, 1936 const char *attrname, int *buflen, char *buf, struct thread *td) 1937 { 1938 struct uio auio; 1939 struct iovec iov; 1940 int error; 1941 1942 iov.iov_len = *buflen; 1943 iov.iov_base = buf; 1944 1945 auio.uio_iov = &iov; 1946 auio.uio_iovcnt = 1; 1947 auio.uio_rw = UIO_READ; 1948 auio.uio_segflg = UIO_SYSSPACE; 1949 auio.uio_td = td; 1950 auio.uio_offset = 0; 1951 auio.uio_resid = *buflen; 1952 1953 if ((ioflg & IO_NODELOCKED) == 0) 1954 vn_lock(vp, LK_SHARED | LK_RETRY); 1955 1956 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1957 1958 /* authorize attribute retrieval as kernel */ 1959 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL, 1960 td); 1961 1962 if ((ioflg & IO_NODELOCKED) == 0) 1963 VOP_UNLOCK(vp, 0); 1964 1965 if (error == 0) { 1966 *buflen = *buflen - auio.uio_resid; 1967 } 1968 1969 return (error); 1970 } 1971 1972 /* 1973 * XXX failure mode if partially written? 1974 */ 1975 int 1976 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, 1977 const char *attrname, int buflen, char *buf, struct thread *td) 1978 { 1979 struct uio auio; 1980 struct iovec iov; 1981 struct mount *mp; 1982 int error; 1983 1984 iov.iov_len = buflen; 1985 iov.iov_base = buf; 1986 1987 auio.uio_iov = &iov; 1988 auio.uio_iovcnt = 1; 1989 auio.uio_rw = UIO_WRITE; 1990 auio.uio_segflg = UIO_SYSSPACE; 1991 auio.uio_td = td; 1992 auio.uio_offset = 0; 1993 auio.uio_resid = buflen; 1994 1995 if ((ioflg & IO_NODELOCKED) == 0) { 1996 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 1997 return (error); 1998 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1999 } 2000 2001 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 2002 2003 /* authorize attribute setting as kernel */ 2004 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td); 2005 2006 if ((ioflg & IO_NODELOCKED) == 0) { 2007 vn_finished_write(mp); 2008 VOP_UNLOCK(vp, 0); 2009 } 2010 2011 return (error); 2012 } 2013 2014 int 2015 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, 2016 const char *attrname, struct thread *td) 2017 { 2018 struct mount *mp; 2019 int error; 2020 2021 if ((ioflg & IO_NODELOCKED) == 0) { 2022 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 2023 return (error); 2024 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2025 } 2026 2027 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 2028 2029 /* authorize attribute removal as kernel */ 2030 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td); 2031 if (error == EOPNOTSUPP) 2032 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, 2033 NULL, td); 2034 2035 if ((ioflg & IO_NODELOCKED) == 0) { 2036 vn_finished_write(mp); 2037 VOP_UNLOCK(vp, 0); 2038 } 2039 2040 return (error); 2041 } 2042 2043 static int 2044 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags, 2045 struct vnode **rvp) 2046 { 2047 2048 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp)); 2049 } 2050 2051 int 2052 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp) 2053 { 2054 2055 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino, 2056 lkflags, rvp)); 2057 } 2058 2059 int 2060 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg, 2061 int lkflags, struct vnode **rvp) 2062 { 2063 struct mount *mp; 2064 int ltype, error; 2065 2066 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get"); 2067 mp = vp->v_mount; 2068 ltype = VOP_ISLOCKED(vp); 2069 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED, 2070 ("vn_vget_ino: vp not locked")); 2071 error = vfs_busy(mp, MBF_NOWAIT); 2072 if (error != 0) { 2073 vfs_ref(mp); 2074 VOP_UNLOCK(vp, 0); 2075 error = vfs_busy(mp, 0); 2076 vn_lock(vp, ltype | LK_RETRY); 2077 vfs_rel(mp); 2078 if (error != 0) 2079 return (ENOENT); 2080 if (vp->v_iflag & VI_DOOMED) { 2081 vfs_unbusy(mp); 2082 return (ENOENT); 2083 } 2084 } 2085 VOP_UNLOCK(vp, 0); 2086 error = alloc(mp, alloc_arg, lkflags, rvp); 2087 vfs_unbusy(mp); 2088 if (*rvp != vp) 2089 vn_lock(vp, ltype | LK_RETRY); 2090 if (vp->v_iflag & VI_DOOMED) { 2091 if (error == 0) { 2092 if (*rvp == vp) 2093 vunref(vp); 2094 else 2095 vput(*rvp); 2096 } 2097 error = ENOENT; 2098 } 2099 return (error); 2100 } 2101 2102 int 2103 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, 2104 struct thread *td) 2105 { 2106 2107 if (vp->v_type != VREG || td == NULL) 2108 return (0); 2109 if ((uoff_t)uio->uio_offset + uio->uio_resid > 2110 lim_cur(td, RLIMIT_FSIZE)) { 2111 PROC_LOCK(td->td_proc); 2112 kern_psignal(td->td_proc, SIGXFSZ); 2113 PROC_UNLOCK(td->td_proc); 2114 return (EFBIG); 2115 } 2116 return (0); 2117 } 2118 2119 int 2120 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 2121 struct thread *td) 2122 { 2123 struct vnode *vp; 2124 2125 vp = fp->f_vnode; 2126 #ifdef AUDIT 2127 vn_lock(vp, LK_SHARED | LK_RETRY); 2128 AUDIT_ARG_VNODE1(vp); 2129 VOP_UNLOCK(vp, 0); 2130 #endif 2131 return (setfmode(td, active_cred, vp, mode)); 2132 } 2133 2134 int 2135 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 2136 struct thread *td) 2137 { 2138 struct vnode *vp; 2139 2140 vp = fp->f_vnode; 2141 #ifdef AUDIT 2142 vn_lock(vp, LK_SHARED | LK_RETRY); 2143 AUDIT_ARG_VNODE1(vp); 2144 VOP_UNLOCK(vp, 0); 2145 #endif 2146 return (setfown(td, active_cred, vp, uid, gid)); 2147 } 2148 2149 void 2150 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end) 2151 { 2152 vm_object_t object; 2153 2154 if ((object = vp->v_object) == NULL) 2155 return; 2156 VM_OBJECT_WLOCK(object); 2157 vm_object_page_remove(object, start, end, 0); 2158 VM_OBJECT_WUNLOCK(object); 2159 } 2160 2161 int 2162 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred) 2163 { 2164 struct vattr va; 2165 daddr_t bn, bnp; 2166 uint64_t bsize; 2167 off_t noff; 2168 int error; 2169 2170 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA, 2171 ("Wrong command %lu", cmd)); 2172 2173 if (vn_lock(vp, LK_SHARED) != 0) 2174 return (EBADF); 2175 if (vp->v_type != VREG) { 2176 error = ENOTTY; 2177 goto unlock; 2178 } 2179 error = VOP_GETATTR(vp, &va, cred); 2180 if (error != 0) 2181 goto unlock; 2182 noff = *off; 2183 if (noff >= va.va_size) { 2184 error = ENXIO; 2185 goto unlock; 2186 } 2187 bsize = vp->v_mount->mnt_stat.f_iosize; 2188 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) { 2189 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL); 2190 if (error == EOPNOTSUPP) { 2191 error = ENOTTY; 2192 goto unlock; 2193 } 2194 if ((bnp == -1 && cmd == FIOSEEKHOLE) || 2195 (bnp != -1 && cmd == FIOSEEKDATA)) { 2196 noff = bn * bsize; 2197 if (noff < *off) 2198 noff = *off; 2199 goto unlock; 2200 } 2201 } 2202 if (noff > va.va_size) 2203 noff = va.va_size; 2204 /* noff == va.va_size. There is an implicit hole at the end of file. */ 2205 if (cmd == FIOSEEKDATA) 2206 error = ENXIO; 2207 unlock: 2208 VOP_UNLOCK(vp, 0); 2209 if (error == 0) 2210 *off = noff; 2211 return (error); 2212 } 2213 2214 int 2215 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td) 2216 { 2217 struct ucred *cred; 2218 struct vnode *vp; 2219 struct vattr vattr; 2220 off_t foffset, size; 2221 int error, noneg; 2222 2223 cred = td->td_ucred; 2224 vp = fp->f_vnode; 2225 foffset = foffset_lock(fp, 0); 2226 noneg = (vp->v_type != VCHR); 2227 error = 0; 2228 switch (whence) { 2229 case L_INCR: 2230 if (noneg && 2231 (foffset < 0 || 2232 (offset > 0 && foffset > OFF_MAX - offset))) { 2233 error = EOVERFLOW; 2234 break; 2235 } 2236 offset += foffset; 2237 break; 2238 case L_XTND: 2239 vn_lock(vp, LK_SHARED | LK_RETRY); 2240 error = VOP_GETATTR(vp, &vattr, cred); 2241 VOP_UNLOCK(vp, 0); 2242 if (error) 2243 break; 2244 2245 /* 2246 * If the file references a disk device, then fetch 2247 * the media size and use that to determine the ending 2248 * offset. 2249 */ 2250 if (vattr.va_size == 0 && vp->v_type == VCHR && 2251 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0) 2252 vattr.va_size = size; 2253 if (noneg && 2254 (vattr.va_size > OFF_MAX || 2255 (offset > 0 && vattr.va_size > OFF_MAX - offset))) { 2256 error = EOVERFLOW; 2257 break; 2258 } 2259 offset += vattr.va_size; 2260 break; 2261 case L_SET: 2262 break; 2263 case SEEK_DATA: 2264 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td); 2265 break; 2266 case SEEK_HOLE: 2267 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td); 2268 break; 2269 default: 2270 error = EINVAL; 2271 } 2272 if (error == 0 && noneg && offset < 0) 2273 error = EINVAL; 2274 if (error != 0) 2275 goto drop; 2276 VFS_KNOTE_UNLOCKED(vp, 0); 2277 td->td_uretoff.tdu_off = offset; 2278 drop: 2279 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); 2280 return (error); 2281 } 2282 2283 int 2284 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred, 2285 struct thread *td) 2286 { 2287 int error; 2288 2289 /* 2290 * Grant permission if the caller is the owner of the file, or 2291 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on 2292 * on the file. If the time pointer is null, then write 2293 * permission on the file is also sufficient. 2294 * 2295 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes: 2296 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES 2297 * will be allowed to set the times [..] to the current 2298 * server time. 2299 */ 2300 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td); 2301 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0) 2302 error = VOP_ACCESS(vp, VWRITE, cred, td); 2303 return (error); 2304 } 2305 2306 int 2307 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 2308 { 2309 struct vnode *vp; 2310 int error; 2311 2312 if (fp->f_type == DTYPE_FIFO) 2313 kif->kf_type = KF_TYPE_FIFO; 2314 else 2315 kif->kf_type = KF_TYPE_VNODE; 2316 vp = fp->f_vnode; 2317 vref(vp); 2318 FILEDESC_SUNLOCK(fdp); 2319 error = vn_fill_kinfo_vnode(vp, kif); 2320 vrele(vp); 2321 FILEDESC_SLOCK(fdp); 2322 return (error); 2323 } 2324 2325 static inline void 2326 vn_fill_junk(struct kinfo_file *kif) 2327 { 2328 size_t len, olen; 2329 2330 /* 2331 * Simulate vn_fullpath returning changing values for a given 2332 * vp during e.g. coredump. 2333 */ 2334 len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1; 2335 olen = strlen(kif->kf_path); 2336 if (len < olen) 2337 strcpy(&kif->kf_path[len - 1], "$"); 2338 else 2339 for (; olen < len; olen++) 2340 strcpy(&kif->kf_path[olen], "A"); 2341 } 2342 2343 int 2344 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif) 2345 { 2346 struct vattr va; 2347 char *fullpath, *freepath; 2348 int error; 2349 2350 kif->kf_un.kf_file.kf_file_type = vntype_to_kinfo(vp->v_type); 2351 freepath = NULL; 2352 fullpath = "-"; 2353 error = vn_fullpath(curthread, vp, &fullpath, &freepath); 2354 if (error == 0) { 2355 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 2356 } 2357 if (freepath != NULL) 2358 free(freepath, M_TEMP); 2359 2360 KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path, 2361 vn_fill_junk(kif); 2362 ); 2363 2364 /* 2365 * Retrieve vnode attributes. 2366 */ 2367 va.va_fsid = VNOVAL; 2368 va.va_rdev = NODEV; 2369 vn_lock(vp, LK_SHARED | LK_RETRY); 2370 error = VOP_GETATTR(vp, &va, curthread->td_ucred); 2371 VOP_UNLOCK(vp, 0); 2372 if (error != 0) 2373 return (error); 2374 if (va.va_fsid != VNOVAL) 2375 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; 2376 else 2377 kif->kf_un.kf_file.kf_file_fsid = 2378 vp->v_mount->mnt_stat.f_fsid.val[0]; 2379 kif->kf_un.kf_file.kf_file_fsid_freebsd11 = 2380 kif->kf_un.kf_file.kf_file_fsid; /* truncate */ 2381 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; 2382 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); 2383 kif->kf_un.kf_file.kf_file_size = va.va_size; 2384 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; 2385 kif->kf_un.kf_file.kf_file_rdev_freebsd11 = 2386 kif->kf_un.kf_file.kf_file_rdev; /* truncate */ 2387 return (0); 2388 } 2389 2390 int 2391 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, 2392 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, 2393 struct thread *td) 2394 { 2395 #ifdef HWPMC_HOOKS 2396 struct pmckern_map_in pkm; 2397 #endif 2398 struct mount *mp; 2399 struct vnode *vp; 2400 vm_object_t object; 2401 vm_prot_t maxprot; 2402 boolean_t writecounted; 2403 int error; 2404 2405 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \ 2406 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) 2407 /* 2408 * POSIX shared-memory objects are defined to have 2409 * kernel persistence, and are not defined to support 2410 * read(2)/write(2) -- or even open(2). Thus, we can 2411 * use MAP_ASYNC to trade on-disk coherence for speed. 2412 * The shm_open(3) library routine turns on the FPOSIXSHM 2413 * flag to request this behavior. 2414 */ 2415 if ((fp->f_flag & FPOSIXSHM) != 0) 2416 flags |= MAP_NOSYNC; 2417 #endif 2418 vp = fp->f_vnode; 2419 2420 /* 2421 * Ensure that file and memory protections are 2422 * compatible. Note that we only worry about 2423 * writability if mapping is shared; in this case, 2424 * current and max prot are dictated by the open file. 2425 * XXX use the vnode instead? Problem is: what 2426 * credentials do we use for determination? What if 2427 * proc does a setuid? 2428 */ 2429 mp = vp->v_mount; 2430 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) { 2431 maxprot = VM_PROT_NONE; 2432 if ((prot & VM_PROT_EXECUTE) != 0) 2433 return (EACCES); 2434 } else 2435 maxprot = VM_PROT_EXECUTE; 2436 if ((fp->f_flag & FREAD) != 0) 2437 maxprot |= VM_PROT_READ; 2438 else if ((prot & VM_PROT_READ) != 0) 2439 return (EACCES); 2440 2441 /* 2442 * If we are sharing potential changes via MAP_SHARED and we 2443 * are trying to get write permission although we opened it 2444 * without asking for it, bail out. 2445 */ 2446 if ((flags & MAP_SHARED) != 0) { 2447 if ((fp->f_flag & FWRITE) != 0) 2448 maxprot |= VM_PROT_WRITE; 2449 else if ((prot & VM_PROT_WRITE) != 0) 2450 return (EACCES); 2451 } else { 2452 maxprot |= VM_PROT_WRITE; 2453 cap_maxprot |= VM_PROT_WRITE; 2454 } 2455 maxprot &= cap_maxprot; 2456 2457 /* 2458 * For regular files and shared memory, POSIX requires that 2459 * the value of foff be a legitimate offset within the data 2460 * object. In particular, negative offsets are invalid. 2461 * Blocking negative offsets and overflows here avoids 2462 * possible wraparound or user-level access into reserved 2463 * ranges of the data object later. In contrast, POSIX does 2464 * not dictate how offsets are used by device drivers, so in 2465 * the case of a device mapping a negative offset is passed 2466 * on. 2467 */ 2468 if ( 2469 #ifdef _LP64 2470 size > OFF_MAX || 2471 #endif 2472 foff < 0 || foff > OFF_MAX - size) 2473 return (EINVAL); 2474 2475 writecounted = FALSE; 2476 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp, 2477 &foff, &object, &writecounted); 2478 if (error != 0) 2479 return (error); 2480 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 2481 foff, writecounted, td); 2482 if (error != 0) { 2483 /* 2484 * If this mapping was accounted for in the vnode's 2485 * writecount, then undo that now. 2486 */ 2487 if (writecounted) 2488 vnode_pager_release_writecount(object, 0, size); 2489 vm_object_deallocate(object); 2490 } 2491 #ifdef HWPMC_HOOKS 2492 /* Inform hwpmc(4) if an executable is being mapped. */ 2493 if (PMC_HOOK_INSTALLED(PMC_FN_MMAP)) { 2494 if ((prot & VM_PROT_EXECUTE) != 0 && error == 0) { 2495 pkm.pm_file = vp; 2496 pkm.pm_address = (uintptr_t) *addr; 2497 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm); 2498 } 2499 } 2500 #endif 2501 return (error); 2502 } 2503 2504 void 2505 vn_fsid(struct vnode *vp, struct vattr *va) 2506 { 2507 fsid_t *f; 2508 2509 f = &vp->v_mount->mnt_stat.f_fsid; 2510 va->va_fsid = (uint32_t)f->val[1]; 2511 va->va_fsid <<= sizeof(f->val[1]) * NBBY; 2512 va->va_fsid += (uint32_t)f->val[0]; 2513 } 2514