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