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 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94 39 * $FreeBSD: src/sys/kern/vfs_vnops.c,v 1.87.2.13 2002/12/29 18:19:53 dillon Exp $ 40 * $DragonFly: src/sys/kern/vfs_vnops.c,v 1.30 2005/07/13 01:38:50 dillon Exp $ 41 */ 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/fcntl.h> 46 #include <sys/file.h> 47 #include <sys/stat.h> 48 #include <sys/proc.h> 49 #include <sys/mount.h> 50 #include <sys/nlookup.h> 51 #include <sys/vnode.h> 52 #include <sys/buf.h> 53 #include <sys/filio.h> 54 #include <sys/ttycom.h> 55 #include <sys/conf.h> 56 #include <sys/syslog.h> 57 58 static int vn_closefile (struct file *fp, struct thread *td); 59 static int vn_ioctl (struct file *fp, u_long com, caddr_t data, 60 struct thread *td); 61 static int vn_read (struct file *fp, struct uio *uio, 62 struct ucred *cred, int flags, struct thread *td); 63 static int svn_read (struct file *fp, struct uio *uio, 64 struct ucred *cred, int flags, struct thread *td); 65 static int vn_poll (struct file *fp, int events, struct ucred *cred, 66 struct thread *td); 67 static int vn_kqfilter (struct file *fp, struct knote *kn); 68 static int vn_statfile (struct file *fp, struct stat *sb, struct thread *td); 69 static int vn_write (struct file *fp, struct uio *uio, 70 struct ucred *cred, int flags, struct thread *td); 71 static int svn_write (struct file *fp, struct uio *uio, 72 struct ucred *cred, int flags, struct thread *td); 73 74 struct fileops vnode_fileops = { 75 NULL, /* port */ 76 NULL, /* clone */ 77 vn_read, vn_write, vn_ioctl, vn_poll, vn_kqfilter, 78 vn_statfile, vn_closefile, nofo_shutdown 79 }; 80 81 struct fileops specvnode_fileops = { 82 NULL, /* port */ 83 NULL, /* clone */ 84 svn_read, svn_write, vn_ioctl, vn_poll, vn_kqfilter, 85 vn_statfile, vn_closefile, nofo_shutdown 86 }; 87 88 /* 89 * Shortcut the device read/write. This avoids a lot of vnode junk. 90 * Basically the specfs vnops for read and write take the locked vnode, 91 * unlock it (because we can't hold the vnode locked while reading or writing 92 * a device which may block indefinitely), issues the device operation, then 93 * relock the vnode before returning, plus other junk. This bypasses all 94 * of that and just does the device operation. 95 */ 96 void 97 vn_setspecops(struct file *fp) 98 { 99 if (vfs_fastdev && fp->f_ops == &vnode_fileops) { 100 fp->f_ops = &specvnode_fileops; 101 } 102 } 103 104 /* 105 * Common code for vnode open operations. Check permissions, and call 106 * the VOP_NOPEN or VOP_NCREATE routine. 107 * 108 * The caller is responsible for setting up nd with nlookup_init() and 109 * for cleaning it up with nlookup_done(), whether we return an error 110 * or not. 111 * 112 * On success nd->nl_open_vp will hold a referenced and, if requested, 113 * locked vnode. A locked vnode is requested via NLC_LOCKVP. If fp 114 * is non-NULL the vnode will be installed in the file pointer. 115 * 116 * NOTE: The vnode is referenced just once on return whether or not it 117 * is also installed in the file pointer. 118 */ 119 int 120 vn_open(struct nlookupdata *nd, struct file *fp, int fmode, int cmode) 121 { 122 struct vnode *vp; 123 struct thread *td = nd->nl_td; 124 struct ucred *cred = nd->nl_cred; 125 struct vattr vat; 126 struct vattr *vap = &vat; 127 struct namecache *ncp; 128 int mode, error; 129 130 /* 131 * Lookup the path and create or obtain the vnode. After a 132 * successful lookup a locked nd->nl_ncp will be returned. 133 * 134 * The result of this section should be a locked vnode. 135 * 136 * XXX with only a little work we should be able to avoid locking 137 * the vnode if FWRITE, O_CREAT, and O_TRUNC are *not* set. 138 */ 139 if (fmode & O_CREAT) { 140 /* 141 * CONDITIONAL CREATE FILE CASE 142 * 143 * Setting NLC_CREATE causes a negative hit to store 144 * the negative hit ncp and not return an error. Then 145 * nc_error or nc_vp may be checked to see if the ncp 146 * represents a negative hit. NLC_CREATE also requires 147 * write permission on the governing directory or EPERM 148 * is returned. 149 */ 150 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 151 nd->nl_flags |= NLC_FOLLOW; 152 nd->nl_flags |= NLC_CREATE; 153 bwillwrite(); 154 error = nlookup(nd); 155 } else { 156 /* 157 * NORMAL OPEN FILE CASE 158 */ 159 error = nlookup(nd); 160 } 161 162 if (error) 163 return (error); 164 ncp = nd->nl_ncp; 165 166 /* 167 * split case to allow us to re-resolve and retry the ncp in case 168 * we get ESTALE. 169 */ 170 again: 171 if (fmode & O_CREAT) { 172 if (ncp->nc_vp == NULL) { 173 VATTR_NULL(vap); 174 vap->va_type = VREG; 175 vap->va_mode = cmode; 176 if (fmode & O_EXCL) 177 vap->va_vaflags |= VA_EXCLUSIVE; 178 error = VOP_NCREATE(ncp, &vp, nd->nl_cred, vap); 179 if (error) 180 return (error); 181 fmode &= ~O_TRUNC; 182 ASSERT_VOP_LOCKED(vp, "create"); 183 /* locked vnode is returned */ 184 } else { 185 if (fmode & O_EXCL) { 186 error = EEXIST; 187 } else { 188 error = cache_vget(ncp, cred, 189 LK_EXCLUSIVE, &vp); 190 } 191 if (error) 192 return (error); 193 fmode &= ~O_CREAT; 194 } 195 } else { 196 error = cache_vget(ncp, cred, LK_EXCLUSIVE, &vp); 197 if (error) 198 return (error); 199 } 200 201 /* 202 * We have a locked vnode and ncp now. Note that the ncp will 203 * be cleaned up by the caller if nd->nl_ncp is left intact. 204 */ 205 if (vp->v_type == VLNK) { 206 error = EMLINK; 207 goto bad; 208 } 209 if (vp->v_type == VSOCK) { 210 error = EOPNOTSUPP; 211 goto bad; 212 } 213 if ((fmode & O_CREAT) == 0) { 214 mode = 0; 215 if (fmode & (FWRITE | O_TRUNC)) { 216 if (vp->v_type == VDIR) { 217 error = EISDIR; 218 goto bad; 219 } 220 error = vn_writechk(vp); 221 if (error) { 222 /* 223 * Special stale handling, re-resolve the 224 * vnode. 225 */ 226 if (error == ESTALE) { 227 vput(vp); 228 vp = NULL; 229 cache_setunresolved(ncp); 230 error = cache_resolve(ncp, cred); 231 if (error == 0) 232 goto again; 233 } 234 goto bad; 235 } 236 mode |= VWRITE; 237 } 238 if (fmode & FREAD) 239 mode |= VREAD; 240 if (mode) { 241 error = VOP_ACCESS(vp, mode, cred, td); 242 if (error) { 243 /* 244 * Special stale handling, re-resolve the 245 * vnode. 246 */ 247 if (error == ESTALE) { 248 vput(vp); 249 vp = NULL; 250 cache_setunresolved(ncp); 251 error = cache_resolve(ncp, cred); 252 if (error == 0) 253 goto again; 254 } 255 goto bad; 256 } 257 } 258 } 259 if (fmode & O_TRUNC) { 260 VOP_UNLOCK(vp, 0, td); /* XXX */ 261 VOP_LEASE(vp, td, cred, LEASE_WRITE); 262 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); /* XXX */ 263 VATTR_NULL(vap); 264 vap->va_size = 0; 265 error = VOP_SETATTR(vp, vap, cred, td); 266 if (error) 267 goto bad; 268 } 269 270 /* 271 * Setup the fp so VOP_OPEN can override it. No descriptor has been 272 * associated with the fp yet so we own it clean. f_data will inherit 273 * our vp reference as long as we do not shift f_ops to &badfileops. 274 * f_ncp inherits nl_ncp . 275 */ 276 if (fp) { 277 fp->f_data = (caddr_t)vp; 278 fp->f_flag = fmode & FMASK; 279 fp->f_ops = &vnode_fileops; 280 fp->f_type = (vp->v_type == VFIFO ? DTYPE_FIFO : DTYPE_VNODE); 281 if (vp->v_type == VDIR) { 282 fp->f_ncp = nd->nl_ncp; 283 nd->nl_ncp = NULL; 284 cache_unlock(fp->f_ncp); 285 } 286 } 287 288 /* 289 * Get rid of nl_ncp. vn_open does not return it (it returns the 290 * vnode or the file pointer). Note: we can't leave nl_ncp locked 291 * through the VOP_OPEN anyway since the VOP_OPEN may block, e.g. 292 * on /dev/ttyd0 293 */ 294 if (nd->nl_ncp) { 295 cache_put(nd->nl_ncp); 296 nd->nl_ncp = NULL; 297 } 298 299 error = VOP_OPEN(vp, fmode, cred, fp, td); 300 if (error) { 301 /* 302 * setting f_ops to &badfileops will prevent the descriptor 303 * code from trying to close and release the vnode, since 304 * the open failed we do not want to call close. 305 */ 306 if (fp) { 307 fp->f_data = NULL; 308 fp->f_ops = &badfileops; 309 } 310 goto bad; 311 } 312 if (fmode & FWRITE) 313 vp->v_writecount++; 314 315 /* 316 * Make sure that a VM object is created for VMIO support. If this 317 * fails we have to be sure to match VOP_CLOSE's with VOP_OPEN's. 318 * Cleanup the fp so we can just vput() the vp in 'bad'. 319 */ 320 if (vn_canvmio(vp) == TRUE) { 321 if ((error = vfs_object_create(vp, td)) != 0) { 322 if (fp) { 323 fp->f_data = NULL; 324 fp->f_ops = &badfileops; 325 } 326 VOP_CLOSE(vp, fmode, td); 327 goto bad; 328 } 329 } 330 331 /* 332 * Return the vnode. XXX needs some cleaning up. The vnode is 333 * only returned in the fp == NULL case, otherwise the vnode ref 334 * is inherited by the fp and we unconditionally unlock it. 335 */ 336 if (fp == NULL) { 337 nd->nl_open_vp = vp; 338 nd->nl_vp_fmode = fmode; 339 if ((nd->nl_flags & NLC_LOCKVP) == 0) 340 VOP_UNLOCK(vp, 0, td); 341 } else { 342 VOP_UNLOCK(vp, 0, td); 343 } 344 return (0); 345 bad: 346 if (vp) 347 vput(vp); 348 return (error); 349 } 350 351 /* 352 * Check for write permissions on the specified vnode. 353 * Prototype text segments cannot be written. 354 */ 355 int 356 vn_writechk(vp) 357 struct vnode *vp; 358 { 359 360 /* 361 * If there's shared text associated with 362 * the vnode, try to free it up once. If 363 * we fail, we can't allow writing. 364 */ 365 if (vp->v_flag & VTEXT) 366 return (ETXTBSY); 367 return (0); 368 } 369 370 /* 371 * Vnode close call 372 */ 373 int 374 vn_close(struct vnode *vp, int flags, struct thread *td) 375 { 376 int error; 377 378 if (flags & FWRITE) 379 vp->v_writecount--; 380 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td)) == 0) { 381 error = VOP_CLOSE(vp, flags, td); 382 VOP_UNLOCK(vp, 0, td); 383 } 384 vrele(vp); 385 return (error); 386 } 387 388 static __inline 389 int 390 sequential_heuristic(struct uio *uio, struct file *fp) 391 { 392 /* 393 * Sequential heuristic - detect sequential operation 394 */ 395 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 396 uio->uio_offset == fp->f_nextoff) { 397 int tmpseq = fp->f_seqcount; 398 /* 399 * XXX we assume that the filesystem block size is 400 * the default. Not true, but still gives us a pretty 401 * good indicator of how sequential the read operations 402 * are. 403 */ 404 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE; 405 if (tmpseq > IO_SEQMAX) 406 tmpseq = IO_SEQMAX; 407 fp->f_seqcount = tmpseq; 408 return(fp->f_seqcount << IO_SEQSHIFT); 409 } 410 411 /* 412 * Not sequential, quick draw-down of seqcount 413 */ 414 if (fp->f_seqcount > 1) 415 fp->f_seqcount = 1; 416 else 417 fp->f_seqcount = 0; 418 return(0); 419 } 420 421 /* 422 * Package up an I/O request on a vnode into a uio and do it. 423 * 424 * We are going to assume the caller has done the appropriate 425 * VOP_LEASE() call before calling vn_rdwr() 426 */ 427 int 428 vn_rdwr(rw, vp, base, len, offset, segflg, ioflg, cred, aresid, td) 429 enum uio_rw rw; 430 struct vnode *vp; 431 caddr_t base; 432 int len; 433 off_t offset; 434 enum uio_seg segflg; 435 int ioflg; 436 struct ucred *cred; 437 int *aresid; 438 struct thread *td; 439 { 440 struct uio auio; 441 struct iovec aiov; 442 int error; 443 444 if ((ioflg & IO_NODELOCKED) == 0) 445 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 446 auio.uio_iov = &aiov; 447 auio.uio_iovcnt = 1; 448 aiov.iov_base = base; 449 aiov.iov_len = len; 450 auio.uio_resid = len; 451 auio.uio_offset = offset; 452 auio.uio_segflg = segflg; 453 auio.uio_rw = rw; 454 auio.uio_td = td; 455 if (rw == UIO_READ) { 456 error = VOP_READ(vp, &auio, ioflg, cred); 457 } else { 458 error = VOP_WRITE(vp, &auio, ioflg, cred); 459 } 460 if (aresid) 461 *aresid = auio.uio_resid; 462 else 463 if (auio.uio_resid && error == 0) 464 error = EIO; 465 if ((ioflg & IO_NODELOCKED) == 0) 466 VOP_UNLOCK(vp, 0, td); 467 return (error); 468 } 469 470 /* 471 * Package up an I/O request on a vnode into a uio and do it. The I/O 472 * request is split up into smaller chunks and we try to avoid saturating 473 * the buffer cache while potentially holding a vnode locked, so we 474 * check bwillwrite() before calling vn_rdwr(). We also call uio_yield() 475 * to give other processes a chance to lock the vnode (either other processes 476 * core'ing the same binary, or unrelated processes scanning the directory). 477 */ 478 int 479 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, cred, aresid, td) 480 enum uio_rw rw; 481 struct vnode *vp; 482 caddr_t base; 483 int len; 484 off_t offset; 485 enum uio_seg segflg; 486 int ioflg; 487 struct ucred *cred; 488 int *aresid; 489 struct thread *td; 490 { 491 int error = 0; 492 493 do { 494 int chunk; 495 496 /* 497 * Force `offset' to a multiple of MAXBSIZE except possibly 498 * for the first chunk, so that filesystems only need to 499 * write full blocks except possibly for the first and last 500 * chunks. 501 */ 502 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 503 504 if (chunk > len) 505 chunk = len; 506 if (rw != UIO_READ && vp->v_type == VREG) 507 bwillwrite(); 508 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 509 ioflg, cred, aresid, td); 510 len -= chunk; /* aresid calc already includes length */ 511 if (error) 512 break; 513 offset += chunk; 514 base += chunk; 515 uio_yield(); 516 } while (len); 517 if (aresid) 518 *aresid += len; 519 return (error); 520 } 521 522 /* 523 * File table vnode read routine. 524 */ 525 static int 526 vn_read(fp, uio, cred, flags, td) 527 struct file *fp; 528 struct uio *uio; 529 struct ucred *cred; 530 struct thread *td; 531 int flags; 532 { 533 struct vnode *vp; 534 int error, ioflag; 535 536 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); 537 vp = (struct vnode *)fp->f_data; 538 ioflag = 0; 539 if (fp->f_flag & FNONBLOCK) 540 ioflag |= IO_NDELAY; 541 if (fp->f_flag & O_DIRECT) 542 ioflag |= IO_DIRECT; 543 VOP_LEASE(vp, td, cred, LEASE_READ); 544 vn_lock(vp, LK_SHARED | LK_NOPAUSE | LK_RETRY, td); 545 if ((flags & FOF_OFFSET) == 0) 546 uio->uio_offset = fp->f_offset; 547 548 ioflag |= sequential_heuristic(uio, fp); 549 550 error = VOP_READ(vp, uio, ioflag, cred); 551 if ((flags & FOF_OFFSET) == 0) 552 fp->f_offset = uio->uio_offset; 553 fp->f_nextoff = uio->uio_offset; 554 VOP_UNLOCK(vp, 0, td); 555 return (error); 556 } 557 558 /* 559 * Device-optimized file table vnode read routine. 560 * 561 * This bypasses the VOP table and talks directly to the device. Most 562 * filesystems just route to specfs and can make this optimization. 563 */ 564 static int 565 svn_read(fp, uio, cred, flags, td) 566 struct file *fp; 567 struct uio *uio; 568 struct ucred *cred; 569 struct thread *td; 570 int flags; 571 { 572 struct vnode *vp; 573 int ioflag; 574 int error; 575 dev_t dev; 576 577 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); 578 579 vp = (struct vnode *)fp->f_data; 580 if (vp == NULL || vp->v_type == VBAD) 581 return (EBADF); 582 583 if ((dev = vp->v_rdev) == NULL) 584 return (EBADF); 585 reference_dev(dev); 586 587 if (uio->uio_resid == 0) 588 return (0); 589 if ((flags & FOF_OFFSET) == 0) 590 uio->uio_offset = fp->f_offset; 591 592 ioflag = 0; 593 if (fp->f_flag & FNONBLOCK) 594 ioflag |= IO_NDELAY; 595 if (fp->f_flag & O_DIRECT) 596 ioflag |= IO_DIRECT; 597 ioflag |= sequential_heuristic(uio, fp); 598 599 error = dev_dread(dev, uio, ioflag); 600 601 release_dev(dev); 602 if ((flags & FOF_OFFSET) == 0) 603 fp->f_offset = uio->uio_offset; 604 fp->f_nextoff = uio->uio_offset; 605 return (error); 606 } 607 608 /* 609 * File table vnode write routine. 610 */ 611 static int 612 vn_write(fp, uio, cred, flags, td) 613 struct file *fp; 614 struct uio *uio; 615 struct ucred *cred; 616 struct thread *td; 617 int flags; 618 { 619 struct vnode *vp; 620 int error, ioflag; 621 622 KASSERT(uio->uio_td == td, ("uio_procp %p is not p %p", 623 uio->uio_td, td)); 624 vp = (struct vnode *)fp->f_data; 625 if (vp->v_type == VREG) 626 bwillwrite(); 627 vp = (struct vnode *)fp->f_data; /* XXX needed? */ 628 ioflag = IO_UNIT; 629 if (vp->v_type == VREG && (fp->f_flag & O_APPEND)) 630 ioflag |= IO_APPEND; 631 if (fp->f_flag & FNONBLOCK) 632 ioflag |= IO_NDELAY; 633 if (fp->f_flag & O_DIRECT) 634 ioflag |= IO_DIRECT; 635 if ((fp->f_flag & O_FSYNC) || 636 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) 637 ioflag |= IO_SYNC; 638 VOP_LEASE(vp, td, cred, LEASE_WRITE); 639 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 640 if ((flags & FOF_OFFSET) == 0) 641 uio->uio_offset = fp->f_offset; 642 ioflag |= sequential_heuristic(uio, fp); 643 error = VOP_WRITE(vp, uio, ioflag, cred); 644 if ((flags & FOF_OFFSET) == 0) 645 fp->f_offset = uio->uio_offset; 646 fp->f_nextoff = uio->uio_offset; 647 VOP_UNLOCK(vp, 0, td); 648 return (error); 649 } 650 651 /* 652 * Device-optimized file table vnode write routine. 653 * 654 * This bypasses the VOP table and talks directly to the device. Most 655 * filesystems just route to specfs and can make this optimization. 656 */ 657 static int 658 svn_write(fp, uio, cred, flags, td) 659 struct file *fp; 660 struct uio *uio; 661 struct ucred *cred; 662 struct thread *td; 663 int flags; 664 { 665 struct vnode *vp; 666 int ioflag; 667 int error; 668 dev_t dev; 669 670 KASSERT(uio->uio_td == td, ("uio_procp %p is not p %p", 671 uio->uio_td, td)); 672 673 vp = (struct vnode *)fp->f_data; 674 if (vp == NULL || vp->v_type == VBAD) 675 return (EBADF); 676 if (vp->v_type == VREG) 677 bwillwrite(); 678 vp = (struct vnode *)fp->f_data; /* XXX needed? */ 679 680 if ((dev = vp->v_rdev) == NULL) 681 return (EBADF); 682 reference_dev(dev); 683 684 if ((flags & FOF_OFFSET) == 0) 685 uio->uio_offset = fp->f_offset; 686 687 ioflag = IO_UNIT; 688 if (vp->v_type == VREG && (fp->f_flag & O_APPEND)) 689 ioflag |= IO_APPEND; 690 if (fp->f_flag & FNONBLOCK) 691 ioflag |= IO_NDELAY; 692 if (fp->f_flag & O_DIRECT) 693 ioflag |= IO_DIRECT; 694 if ((fp->f_flag & O_FSYNC) || 695 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) 696 ioflag |= IO_SYNC; 697 ioflag |= sequential_heuristic(uio, fp); 698 699 error = dev_dwrite(dev, uio, ioflag); 700 701 release_dev(dev); 702 if ((flags & FOF_OFFSET) == 0) 703 fp->f_offset = uio->uio_offset; 704 fp->f_nextoff = uio->uio_offset; 705 706 return (error); 707 } 708 709 /* 710 * File table vnode stat routine. 711 */ 712 static int 713 vn_statfile(struct file *fp, struct stat *sb, struct thread *td) 714 { 715 struct vnode *vp = (struct vnode *)fp->f_data; 716 717 return vn_stat(vp, sb, td); 718 } 719 720 int 721 vn_stat(struct vnode *vp, struct stat *sb, struct thread *td) 722 { 723 struct vattr vattr; 724 struct vattr *vap; 725 int error; 726 u_short mode; 727 dev_t dev; 728 729 vap = &vattr; 730 error = VOP_GETATTR(vp, vap, td); 731 if (error) 732 return (error); 733 734 /* 735 * Zero the spare stat fields 736 */ 737 sb->st_lspare = 0; 738 sb->st_qspare[0] = 0; 739 sb->st_qspare[1] = 0; 740 741 /* 742 * Copy from vattr table 743 */ 744 if (vap->va_fsid != VNOVAL) 745 sb->st_dev = vap->va_fsid; 746 else 747 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 748 sb->st_ino = vap->va_fileid; 749 mode = vap->va_mode; 750 switch (vap->va_type) { 751 case VREG: 752 mode |= S_IFREG; 753 break; 754 case VDIR: 755 mode |= S_IFDIR; 756 break; 757 case VBLK: 758 mode |= S_IFBLK; 759 break; 760 case VCHR: 761 mode |= S_IFCHR; 762 break; 763 case VLNK: 764 mode |= S_IFLNK; 765 /* This is a cosmetic change, symlinks do not have a mode. */ 766 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) 767 sb->st_mode &= ~ACCESSPERMS; /* 0000 */ 768 else 769 sb->st_mode |= ACCESSPERMS; /* 0777 */ 770 break; 771 case VSOCK: 772 mode |= S_IFSOCK; 773 break; 774 case VFIFO: 775 mode |= S_IFIFO; 776 break; 777 default: 778 return (EBADF); 779 }; 780 sb->st_mode = mode; 781 sb->st_nlink = vap->va_nlink; 782 sb->st_uid = vap->va_uid; 783 sb->st_gid = vap->va_gid; 784 sb->st_rdev = vap->va_rdev; 785 sb->st_size = vap->va_size; 786 sb->st_atimespec = vap->va_atime; 787 sb->st_mtimespec = vap->va_mtime; 788 sb->st_ctimespec = vap->va_ctime; 789 790 /* 791 * A VCHR and VBLK device may track the last access and last modified 792 * time independantly of the filesystem. This is particularly true 793 * because device read and write calls may bypass the filesystem. 794 */ 795 if (vp->v_type == VCHR || vp->v_type == VBLK) { 796 if ((dev = vp->v_rdev) != NULL) { 797 if (dev->si_lastread) { 798 sb->st_atimespec.tv_sec = dev->si_lastread; 799 sb->st_atimespec.tv_nsec = 0; 800 } 801 if (dev->si_lastwrite) { 802 sb->st_atimespec.tv_sec = dev->si_lastwrite; 803 sb->st_atimespec.tv_nsec = 0; 804 } 805 } 806 } 807 808 /* 809 * According to www.opengroup.org, the meaning of st_blksize is 810 * "a filesystem-specific preferred I/O block size for this 811 * object. In some filesystem types, this may vary from file 812 * to file" 813 * Default to PAGE_SIZE after much discussion. 814 */ 815 816 if (vap->va_type == VREG) { 817 sb->st_blksize = vap->va_blocksize; 818 } else if (vn_isdisk(vp, NULL)) { 819 /* 820 * XXX this is broken. If the device is not yet open (aka 821 * stat() call, aka v_rdev == NULL), how are we supposed 822 * to get a valid block size out of it? 823 */ 824 dev_t dev; 825 826 if ((dev = vp->v_rdev) == NULL) 827 dev = udev2dev(vp->v_udev, vp->v_type == VBLK); 828 sb->st_blksize = dev->si_bsize_best; 829 if (sb->st_blksize < dev->si_bsize_phys) 830 sb->st_blksize = dev->si_bsize_phys; 831 if (sb->st_blksize < BLKDEV_IOSIZE) 832 sb->st_blksize = BLKDEV_IOSIZE; 833 } else { 834 sb->st_blksize = PAGE_SIZE; 835 } 836 837 sb->st_flags = vap->va_flags; 838 if (suser(td)) 839 sb->st_gen = 0; 840 else 841 sb->st_gen = vap->va_gen; 842 843 #if (S_BLKSIZE == 512) 844 /* Optimize this case */ 845 sb->st_blocks = vap->va_bytes >> 9; 846 #else 847 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 848 #endif 849 return (0); 850 } 851 852 /* 853 * File table vnode ioctl routine. 854 */ 855 static int 856 vn_ioctl(struct file *fp, u_long com, caddr_t data, struct thread *td) 857 { 858 struct vnode *vp = ((struct vnode *)fp->f_data); 859 struct vnode *ovp; 860 struct ucred *ucred; 861 struct vattr vattr; 862 int error; 863 864 KKASSERT(td->td_proc != NULL); 865 ucred = td->td_proc->p_ucred; 866 867 switch (vp->v_type) { 868 case VREG: 869 case VDIR: 870 if (com == FIONREAD) { 871 error = VOP_GETATTR(vp, &vattr, td); 872 if (error) 873 return (error); 874 *(int *)data = vattr.va_size - fp->f_offset; 875 return (0); 876 } 877 if (com == FIONBIO || com == FIOASYNC) /* XXX */ 878 return (0); /* XXX */ 879 /* fall into ... */ 880 default: 881 #if 0 882 return (ENOTTY); 883 #endif 884 case VFIFO: 885 case VCHR: 886 case VBLK: 887 if (com == FIODTYPE) { 888 if (vp->v_type != VCHR && vp->v_type != VBLK) 889 return (ENOTTY); 890 *(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK; 891 return (0); 892 } 893 error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred, td); 894 if (error == 0 && com == TIOCSCTTY) { 895 struct session *sess = td->td_proc->p_session; 896 897 /* Do nothing if reassigning same control tty */ 898 if (sess->s_ttyvp == vp) 899 return (0); 900 901 /* Get rid of reference to old control tty */ 902 ovp = sess->s_ttyvp; 903 vref(vp); 904 sess->s_ttyvp = vp; 905 if (ovp) 906 vrele(ovp); 907 } 908 return (error); 909 } 910 } 911 912 /* 913 * File table vnode poll routine. 914 */ 915 static int 916 vn_poll(struct file *fp, int events, struct ucred *cred, struct thread *td) 917 { 918 return (VOP_POLL(((struct vnode *)fp->f_data), events, cred, td)); 919 } 920 921 /* 922 * Check that the vnode is still valid, and if so 923 * acquire requested lock. 924 */ 925 int 926 #ifndef DEBUG_LOCKS 927 vn_lock(struct vnode *vp, int flags, struct thread *td) 928 #else 929 debug_vn_lock(struct vnode *vp, int flags, struct thread *td, 930 const char *filename, int line) 931 #endif 932 { 933 int error; 934 935 do { 936 #ifdef DEBUG_LOCKS 937 vp->filename = filename; 938 vp->line = line; 939 #endif 940 error = VOP_LOCK(vp, flags | LK_NOPAUSE, td); 941 if (error == 0) 942 break; 943 } while (flags & LK_RETRY); 944 945 /* 946 * Because we (had better!) have a ref on the vnode, once it 947 * goes to VRECLAIMED state it will not be recycled until all 948 * refs go away. So we can just check the flag. 949 */ 950 if (error == 0 && (vp->v_flag & VRECLAIMED)) { 951 VOP_UNLOCK(vp, 0, td); 952 error = ENOENT; 953 } 954 return (error); 955 } 956 957 /* 958 * File table vnode close routine. 959 */ 960 static int 961 vn_closefile(struct file *fp, struct thread *td) 962 { 963 int err; 964 965 fp->f_ops = &badfileops; 966 err = vn_close(((struct vnode *)fp->f_data), fp->f_flag, td); 967 return(err); 968 } 969 970 static int 971 vn_kqfilter(struct file *fp, struct knote *kn) 972 { 973 974 return (VOP_KQFILTER(((struct vnode *)fp->f_data), kn)); 975 } 976