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 1.43 2006/08/08 03:52:40 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); 59 static int vn_ioctl (struct file *fp, u_long com, caddr_t data, 60 struct ucred *cred); 61 static int vn_read (struct file *fp, struct uio *uio, 62 struct ucred *cred, int flags); 63 static int svn_read (struct file *fp, struct uio *uio, 64 struct ucred *cred, int flags); 65 static int vn_poll (struct file *fp, int events, struct ucred *cred); 66 static int vn_kqfilter (struct file *fp, struct knote *kn); 67 static int vn_statfile (struct file *fp, struct stat *sb, struct ucred *cred); 68 static int vn_write (struct file *fp, struct uio *uio, 69 struct ucred *cred, int flags); 70 static int svn_write (struct file *fp, struct uio *uio, 71 struct ucred *cred, int flags); 72 73 struct fileops vnode_fileops = { 74 .fo_read = vn_read, 75 .fo_write = vn_write, 76 .fo_ioctl = vn_ioctl, 77 .fo_poll = vn_poll, 78 .fo_kqfilter = vn_kqfilter, 79 .fo_stat = vn_statfile, 80 .fo_close = vn_closefile, 81 .fo_shutdown = nofo_shutdown 82 }; 83 84 struct fileops specvnode_fileops = { 85 .fo_read = svn_read, 86 .fo_write = svn_write, 87 .fo_ioctl = vn_ioctl, 88 .fo_poll = vn_poll, 89 .fo_kqfilter = vn_kqfilter, 90 .fo_stat = vn_statfile, 91 .fo_close = vn_closefile, 92 .fo_shutdown = nofo_shutdown 93 }; 94 95 /* 96 * Shortcut the device read/write. This avoids a lot of vnode junk. 97 * Basically the specfs vnops for read and write take the locked vnode, 98 * unlock it (because we can't hold the vnode locked while reading or writing 99 * a device which may block indefinitely), issues the device operation, then 100 * relock the vnode before returning, plus other junk. This bypasses all 101 * of that and just does the device operation. 102 */ 103 void 104 vn_setspecops(struct file *fp) 105 { 106 if (vfs_fastdev && fp->f_ops == &vnode_fileops) { 107 fp->f_ops = &specvnode_fileops; 108 } 109 } 110 111 /* 112 * Common code for vnode open operations. Check permissions, and call 113 * the VOP_NOPEN or VOP_NCREATE routine. 114 * 115 * The caller is responsible for setting up nd with nlookup_init() and 116 * for cleaning it up with nlookup_done(), whether we return an error 117 * or not. 118 * 119 * On success nd->nl_open_vp will hold a referenced and, if requested, 120 * locked vnode. A locked vnode is requested via NLC_LOCKVP. If fp 121 * is non-NULL the vnode will be installed in the file pointer. 122 * 123 * NOTE: The vnode is referenced just once on return whether or not it 124 * is also installed in the file pointer. 125 */ 126 int 127 vn_open(struct nlookupdata *nd, struct file *fp, int fmode, int cmode) 128 { 129 struct vnode *vp; 130 struct ucred *cred = nd->nl_cred; 131 struct vattr vat; 132 struct vattr *vap = &vat; 133 struct namecache *ncp; 134 int mode, error; 135 136 /* 137 * Lookup the path and create or obtain the vnode. After a 138 * successful lookup a locked nd->nl_ncp will be returned. 139 * 140 * The result of this section should be a locked vnode. 141 * 142 * XXX with only a little work we should be able to avoid locking 143 * the vnode if FWRITE, O_CREAT, and O_TRUNC are *not* set. 144 */ 145 if (fmode & O_CREAT) { 146 /* 147 * CONDITIONAL CREATE FILE CASE 148 * 149 * Setting NLC_CREATE causes a negative hit to store 150 * the negative hit ncp and not return an error. Then 151 * nc_error or nc_vp may be checked to see if the ncp 152 * represents a negative hit. NLC_CREATE also requires 153 * write permission on the governing directory or EPERM 154 * is returned. 155 */ 156 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 157 nd->nl_flags |= NLC_FOLLOW; 158 nd->nl_flags |= NLC_CREATE; 159 bwillwrite(); 160 error = nlookup(nd); 161 } else { 162 /* 163 * NORMAL OPEN FILE CASE 164 */ 165 error = nlookup(nd); 166 } 167 168 if (error) 169 return (error); 170 ncp = nd->nl_ncp; 171 172 /* 173 * split case to allow us to re-resolve and retry the ncp in case 174 * we get ESTALE. 175 */ 176 again: 177 if (fmode & O_CREAT) { 178 if (ncp->nc_vp == NULL) { 179 VATTR_NULL(vap); 180 vap->va_type = VREG; 181 vap->va_mode = cmode; 182 if (fmode & O_EXCL) 183 vap->va_vaflags |= VA_EXCLUSIVE; 184 error = VOP_NCREATE(ncp, &vp, nd->nl_cred, vap); 185 if (error) 186 return (error); 187 fmode &= ~O_TRUNC; 188 ASSERT_VOP_LOCKED(vp, "create"); 189 /* locked vnode is returned */ 190 } else { 191 if (fmode & O_EXCL) { 192 error = EEXIST; 193 } else { 194 error = cache_vget(ncp, cred, 195 LK_EXCLUSIVE, &vp); 196 } 197 if (error) 198 return (error); 199 fmode &= ~O_CREAT; 200 } 201 } else { 202 error = cache_vget(ncp, cred, LK_EXCLUSIVE, &vp); 203 if (error) 204 return (error); 205 } 206 207 /* 208 * We have a locked vnode and ncp now. Note that the ncp will 209 * be cleaned up by the caller if nd->nl_ncp is left intact. 210 */ 211 if (vp->v_type == VLNK) { 212 error = EMLINK; 213 goto bad; 214 } 215 if (vp->v_type == VSOCK) { 216 error = EOPNOTSUPP; 217 goto bad; 218 } 219 if ((fmode & O_CREAT) == 0) { 220 mode = 0; 221 if (fmode & (FWRITE | O_TRUNC)) { 222 if (vp->v_type == VDIR) { 223 error = EISDIR; 224 goto bad; 225 } 226 error = vn_writechk(vp); 227 if (error) { 228 /* 229 * Special stale handling, re-resolve the 230 * vnode. 231 */ 232 if (error == ESTALE) { 233 vput(vp); 234 vp = NULL; 235 cache_setunresolved(ncp); 236 error = cache_resolve(ncp, cred); 237 if (error == 0) 238 goto again; 239 } 240 goto bad; 241 } 242 mode |= VWRITE; 243 } 244 if (fmode & FREAD) 245 mode |= VREAD; 246 if (mode) { 247 error = VOP_ACCESS(vp, mode, cred); 248 if (error) { 249 /* 250 * Special stale handling, re-resolve the 251 * vnode. 252 */ 253 if (error == ESTALE) { 254 vput(vp); 255 vp = NULL; 256 cache_setunresolved(ncp); 257 error = cache_resolve(ncp, cred); 258 if (error == 0) 259 goto again; 260 } 261 goto bad; 262 } 263 } 264 } 265 if (fmode & O_TRUNC) { 266 VOP_UNLOCK(vp, 0); /* XXX */ 267 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 268 VATTR_NULL(vap); 269 vap->va_size = 0; 270 error = VOP_SETATTR(vp, vap, cred); 271 if (error) 272 goto bad; 273 } 274 275 /* 276 * Setup the fp so VOP_OPEN can override it. No descriptor has been 277 * associated with the fp yet so we own it clean. f_ncp inherits 278 * nl_ncp . 279 */ 280 if (fp) { 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); 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 } 312 313 #if 0 314 /* 315 * Assert that VREG files have been setup for vmio. 316 */ 317 KASSERT(vp->v_type != VREG || vp->v_object != NULL, 318 ("vn_open: regular file was not VMIO enabled!")); 319 #endif 320 321 /* 322 * Return the vnode. XXX needs some cleaning up. The vnode is 323 * only returned in the fp == NULL case. 324 */ 325 if (fp == NULL) { 326 nd->nl_open_vp = vp; 327 nd->nl_vp_fmode = fmode; 328 if ((nd->nl_flags & NLC_LOCKVP) == 0) 329 VOP_UNLOCK(vp, 0); 330 } else { 331 vput(vp); 332 } 333 return (0); 334 bad: 335 if (vp) 336 vput(vp); 337 return (error); 338 } 339 340 /* 341 * Check for write permissions on the specified vnode. 342 * Prototype text segments cannot be written. 343 */ 344 int 345 vn_writechk(vp) 346 struct vnode *vp; 347 { 348 349 /* 350 * If there's shared text associated with 351 * the vnode, try to free it up once. If 352 * we fail, we can't allow writing. 353 */ 354 if (vp->v_flag & VTEXT) 355 return (ETXTBSY); 356 return (0); 357 } 358 359 /* 360 * Vnode close call 361 */ 362 int 363 vn_close(struct vnode *vp, int flags) 364 { 365 int error; 366 367 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) == 0) { 368 error = VOP_CLOSE(vp, flags); 369 VOP_UNLOCK(vp, 0); 370 } 371 vrele(vp); 372 return (error); 373 } 374 375 static __inline 376 int 377 sequential_heuristic(struct uio *uio, struct file *fp) 378 { 379 /* 380 * Sequential heuristic - detect sequential operation 381 */ 382 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 383 uio->uio_offset == fp->f_nextoff) { 384 int tmpseq = fp->f_seqcount; 385 /* 386 * XXX we assume that the filesystem block size is 387 * the default. Not true, but still gives us a pretty 388 * good indicator of how sequential the read operations 389 * are. 390 */ 391 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE; 392 if (tmpseq > IO_SEQMAX) 393 tmpseq = IO_SEQMAX; 394 fp->f_seqcount = tmpseq; 395 return(fp->f_seqcount << IO_SEQSHIFT); 396 } 397 398 /* 399 * Not sequential, quick draw-down of seqcount 400 */ 401 if (fp->f_seqcount > 1) 402 fp->f_seqcount = 1; 403 else 404 fp->f_seqcount = 0; 405 return(0); 406 } 407 408 /* 409 * Package up an I/O request on a vnode into a uio and do it. 410 */ 411 int 412 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 413 off_t offset, enum uio_seg segflg, int ioflg, 414 struct ucred *cred, int *aresid) 415 { 416 struct uio auio; 417 struct iovec aiov; 418 int error; 419 420 if ((ioflg & IO_NODELOCKED) == 0) 421 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 422 auio.uio_iov = &aiov; 423 auio.uio_iovcnt = 1; 424 aiov.iov_base = base; 425 aiov.iov_len = len; 426 auio.uio_resid = len; 427 auio.uio_offset = offset; 428 auio.uio_segflg = segflg; 429 auio.uio_rw = rw; 430 auio.uio_td = curthread; 431 if (rw == UIO_READ) { 432 error = VOP_READ(vp, &auio, ioflg, cred); 433 } else { 434 error = VOP_WRITE(vp, &auio, ioflg, cred); 435 } 436 if (aresid) 437 *aresid = auio.uio_resid; 438 else 439 if (auio.uio_resid && error == 0) 440 error = EIO; 441 if ((ioflg & IO_NODELOCKED) == 0) 442 VOP_UNLOCK(vp, 0); 443 return (error); 444 } 445 446 /* 447 * Package up an I/O request on a vnode into a uio and do it. The I/O 448 * request is split up into smaller chunks and we try to avoid saturating 449 * the buffer cache while potentially holding a vnode locked, so we 450 * check bwillwrite() before calling vn_rdwr(). We also call uio_yield() 451 * to give other processes a chance to lock the vnode (either other processes 452 * core'ing the same binary, or unrelated processes scanning the directory). 453 */ 454 int 455 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 456 off_t offset, enum uio_seg segflg, int ioflg, 457 struct ucred *cred, int *aresid) 458 { 459 int error = 0; 460 461 do { 462 int chunk; 463 464 /* 465 * Force `offset' to a multiple of MAXBSIZE except possibly 466 * for the first chunk, so that filesystems only need to 467 * write full blocks except possibly for the first and last 468 * chunks. 469 */ 470 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 471 472 if (chunk > len) 473 chunk = len; 474 if (rw != UIO_READ && vp->v_type == VREG) 475 bwillwrite(); 476 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 477 ioflg, cred, aresid); 478 len -= chunk; /* aresid calc already includes length */ 479 if (error) 480 break; 481 offset += chunk; 482 base += chunk; 483 uio_yield(); 484 } while (len); 485 if (aresid) 486 *aresid += len; 487 return (error); 488 } 489 490 /* 491 * MPALMOSTSAFE - acquires mplock 492 */ 493 static int 494 vn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 495 { 496 struct vnode *vp; 497 int error, ioflag; 498 499 get_mplock(); 500 KASSERT(uio->uio_td == curthread, 501 ("uio_td %p is not td %p", uio->uio_td, curthread)); 502 vp = (struct vnode *)fp->f_data; 503 504 ioflag = 0; 505 if (flags & O_FBLOCKING) { 506 /* ioflag &= ~IO_NDELAY; */ 507 } else if (flags & O_FNONBLOCKING) { 508 ioflag |= IO_NDELAY; 509 } else if (fp->f_flag & FNONBLOCK) { 510 ioflag |= IO_NDELAY; 511 } 512 if (flags & O_FBUFFERED) { 513 /* ioflag &= ~IO_DIRECT; */ 514 } else if (flags & O_FUNBUFFERED) { 515 ioflag |= IO_DIRECT; 516 } else if (fp->f_flag & O_DIRECT) { 517 ioflag |= IO_DIRECT; 518 } 519 vn_lock(vp, LK_SHARED | LK_RETRY); 520 if ((flags & O_FOFFSET) == 0) 521 uio->uio_offset = fp->f_offset; 522 523 ioflag |= sequential_heuristic(uio, fp); 524 525 error = VOP_READ(vp, uio, ioflag, cred); 526 if ((flags & O_FOFFSET) == 0) 527 fp->f_offset = uio->uio_offset; 528 fp->f_nextoff = uio->uio_offset; 529 VOP_UNLOCK(vp, 0); 530 rel_mplock(); 531 return (error); 532 } 533 534 /* 535 * Device-optimized file table vnode read routine. 536 * 537 * This bypasses the VOP table and talks directly to the device. Most 538 * filesystems just route to specfs and can make this optimization. 539 * 540 * MPALMOSTSAFE - acquires mplock 541 */ 542 static int 543 svn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 544 { 545 struct vnode *vp; 546 int ioflag; 547 int error; 548 dev_t dev; 549 550 get_mplock(); 551 KASSERT(uio->uio_td == curthread, 552 ("uio_td %p is not td %p", uio->uio_td, curthread)); 553 554 vp = (struct vnode *)fp->f_data; 555 if (vp == NULL || vp->v_type == VBAD) { 556 error = EBADF; 557 goto done; 558 } 559 560 if ((dev = vp->v_rdev) == NULL) { 561 error = EBADF; 562 goto done; 563 } 564 reference_dev(dev); 565 566 if (uio->uio_resid == 0) { 567 error = 0; 568 goto done; 569 } 570 if ((flags & O_FOFFSET) == 0) 571 uio->uio_offset = fp->f_offset; 572 573 ioflag = 0; 574 if (flags & O_FBLOCKING) { 575 /* ioflag &= ~IO_NDELAY; */ 576 } else if (flags & O_FNONBLOCKING) { 577 ioflag |= IO_NDELAY; 578 } else if (fp->f_flag & FNONBLOCK) { 579 ioflag |= IO_NDELAY; 580 } 581 if (flags & O_FBUFFERED) { 582 /* ioflag &= ~IO_DIRECT; */ 583 } else if (flags & O_FUNBUFFERED) { 584 ioflag |= IO_DIRECT; 585 } else if (fp->f_flag & O_DIRECT) { 586 ioflag |= IO_DIRECT; 587 } 588 ioflag |= sequential_heuristic(uio, fp); 589 590 error = dev_dread(dev, uio, ioflag); 591 592 release_dev(dev); 593 if ((flags & O_FOFFSET) == 0) 594 fp->f_offset = uio->uio_offset; 595 fp->f_nextoff = uio->uio_offset; 596 done: 597 rel_mplock(); 598 return (error); 599 } 600 601 /* 602 * MPALMOSTSAFE - acquires mplock 603 */ 604 static int 605 vn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 606 { 607 struct vnode *vp; 608 int error, ioflag; 609 610 get_mplock(); 611 KASSERT(uio->uio_td == curthread, 612 ("uio_procp %p is not p %p", uio->uio_td, curthread)); 613 vp = (struct vnode *)fp->f_data; 614 if (vp->v_type == VREG) 615 bwillwrite(); 616 vp = (struct vnode *)fp->f_data; /* XXX needed? */ 617 618 ioflag = IO_UNIT; 619 if (vp->v_type == VREG && 620 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 621 ioflag |= IO_APPEND; 622 } 623 624 if (flags & O_FBLOCKING) { 625 /* ioflag &= ~IO_NDELAY; */ 626 } else if (flags & O_FNONBLOCKING) { 627 ioflag |= IO_NDELAY; 628 } else if (fp->f_flag & FNONBLOCK) { 629 ioflag |= IO_NDELAY; 630 } 631 if (flags & O_FBUFFERED) { 632 /* ioflag &= ~IO_DIRECT; */ 633 } else if (flags & O_FUNBUFFERED) { 634 ioflag |= IO_DIRECT; 635 } else if (fp->f_flag & O_DIRECT) { 636 ioflag |= IO_DIRECT; 637 } 638 if (flags & O_FASYNCWRITE) { 639 /* ioflag &= ~IO_SYNC; */ 640 } else if (flags & O_FSYNCWRITE) { 641 ioflag |= IO_SYNC; 642 } else if (fp->f_flag & O_FSYNC) { 643 ioflag |= IO_SYNC; 644 } 645 646 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 647 ioflag |= IO_SYNC; 648 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 649 if ((flags & O_FOFFSET) == 0) 650 uio->uio_offset = fp->f_offset; 651 ioflag |= sequential_heuristic(uio, fp); 652 error = VOP_WRITE(vp, uio, ioflag, cred); 653 if ((flags & O_FOFFSET) == 0) 654 fp->f_offset = uio->uio_offset; 655 fp->f_nextoff = uio->uio_offset; 656 VOP_UNLOCK(vp, 0); 657 rel_mplock(); 658 return (error); 659 } 660 661 /* 662 * Device-optimized file table vnode write routine. 663 * 664 * This bypasses the VOP table and talks directly to the device. Most 665 * filesystems just route to specfs and can make this optimization. 666 * 667 * MPALMOSTSAFE - acquires mplock 668 */ 669 static int 670 svn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 671 { 672 struct vnode *vp; 673 int ioflag; 674 int error; 675 dev_t dev; 676 677 get_mplock(); 678 KASSERT(uio->uio_td == curthread, 679 ("uio_procp %p is not p %p", uio->uio_td, curthread)); 680 681 vp = (struct vnode *)fp->f_data; 682 if (vp == NULL || vp->v_type == VBAD) { 683 error = EBADF; 684 goto done; 685 } 686 if (vp->v_type == VREG) 687 bwillwrite(); 688 vp = (struct vnode *)fp->f_data; /* XXX needed? */ 689 690 if ((dev = vp->v_rdev) == NULL) { 691 error = EBADF; 692 goto done; 693 } 694 reference_dev(dev); 695 696 if ((flags & O_FOFFSET) == 0) 697 uio->uio_offset = fp->f_offset; 698 699 ioflag = IO_UNIT; 700 if (vp->v_type == VREG && 701 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 702 ioflag |= IO_APPEND; 703 } 704 705 if (flags & O_FBLOCKING) { 706 /* ioflag &= ~IO_NDELAY; */ 707 } else if (flags & O_FNONBLOCKING) { 708 ioflag |= IO_NDELAY; 709 } else if (fp->f_flag & FNONBLOCK) { 710 ioflag |= IO_NDELAY; 711 } 712 if (flags & O_FBUFFERED) { 713 /* ioflag &= ~IO_DIRECT; */ 714 } else if (flags & O_FUNBUFFERED) { 715 ioflag |= IO_DIRECT; 716 } else if (fp->f_flag & O_DIRECT) { 717 ioflag |= IO_DIRECT; 718 } 719 if (flags & O_FASYNCWRITE) { 720 /* ioflag &= ~IO_SYNC; */ 721 } else if (flags & O_FSYNCWRITE) { 722 ioflag |= IO_SYNC; 723 } else if (fp->f_flag & O_FSYNC) { 724 ioflag |= IO_SYNC; 725 } 726 727 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 728 ioflag |= IO_SYNC; 729 ioflag |= sequential_heuristic(uio, fp); 730 731 error = dev_dwrite(dev, uio, ioflag); 732 733 release_dev(dev); 734 if ((flags & O_FOFFSET) == 0) 735 fp->f_offset = uio->uio_offset; 736 fp->f_nextoff = uio->uio_offset; 737 done: 738 rel_mplock(); 739 return (error); 740 } 741 742 /* 743 * MPALMOSTSAFE - acquires mplock 744 */ 745 static int 746 vn_statfile(struct file *fp, struct stat *sb, struct ucred *cred) 747 { 748 struct vnode *vp; 749 int error; 750 751 get_mplock(); 752 vp = (struct vnode *)fp->f_data; 753 error = vn_stat(vp, sb, cred); 754 rel_mplock(); 755 return (error); 756 } 757 758 int 759 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *cred) 760 { 761 struct vattr vattr; 762 struct vattr *vap; 763 int error; 764 u_short mode; 765 dev_t dev; 766 767 vap = &vattr; 768 error = VOP_GETATTR(vp, vap); 769 if (error) 770 return (error); 771 772 /* 773 * Zero the spare stat fields 774 */ 775 sb->st_lspare = 0; 776 sb->st_qspare = 0; 777 778 /* 779 * Copy from vattr table 780 */ 781 if (vap->va_fsid != VNOVAL) 782 sb->st_dev = vap->va_fsid; 783 else 784 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 785 sb->st_ino = vap->va_fileid; 786 mode = vap->va_mode; 787 switch (vap->va_type) { 788 case VREG: 789 mode |= S_IFREG; 790 break; 791 case VDIR: 792 mode |= S_IFDIR; 793 break; 794 case VBLK: 795 mode |= S_IFBLK; 796 break; 797 case VCHR: 798 mode |= S_IFCHR; 799 break; 800 case VLNK: 801 mode |= S_IFLNK; 802 /* This is a cosmetic change, symlinks do not have a mode. */ 803 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) 804 sb->st_mode &= ~ACCESSPERMS; /* 0000 */ 805 else 806 sb->st_mode |= ACCESSPERMS; /* 0777 */ 807 break; 808 case VSOCK: 809 mode |= S_IFSOCK; 810 break; 811 case VFIFO: 812 mode |= S_IFIFO; 813 break; 814 default: 815 return (EBADF); 816 }; 817 sb->st_mode = mode; 818 sb->st_nlink = vap->va_nlink; 819 sb->st_uid = vap->va_uid; 820 sb->st_gid = vap->va_gid; 821 sb->st_rdev = vap->va_rdev; 822 sb->st_size = vap->va_size; 823 sb->st_atimespec = vap->va_atime; 824 sb->st_mtimespec = vap->va_mtime; 825 sb->st_ctimespec = vap->va_ctime; 826 827 /* 828 * A VCHR and VBLK device may track the last access and last modified 829 * time independantly of the filesystem. This is particularly true 830 * because device read and write calls may bypass the filesystem. 831 */ 832 if (vp->v_type == VCHR || vp->v_type == VBLK) { 833 if ((dev = vp->v_rdev) != NULL) { 834 if (dev->si_lastread) { 835 sb->st_atimespec.tv_sec = dev->si_lastread; 836 sb->st_atimespec.tv_nsec = 0; 837 } 838 if (dev->si_lastwrite) { 839 sb->st_atimespec.tv_sec = dev->si_lastwrite; 840 sb->st_atimespec.tv_nsec = 0; 841 } 842 } 843 } 844 845 /* 846 * According to www.opengroup.org, the meaning of st_blksize is 847 * "a filesystem-specific preferred I/O block size for this 848 * object. In some filesystem types, this may vary from file 849 * to file" 850 * Default to PAGE_SIZE after much discussion. 851 */ 852 853 if (vap->va_type == VREG) { 854 sb->st_blksize = vap->va_blocksize; 855 } else if (vn_isdisk(vp, NULL)) { 856 /* 857 * XXX this is broken. If the device is not yet open (aka 858 * stat() call, aka v_rdev == NULL), how are we supposed 859 * to get a valid block size out of it? 860 */ 861 dev_t dev; 862 863 if ((dev = vp->v_rdev) == NULL) 864 dev = udev2dev(vp->v_udev, vp->v_type == VBLK); 865 sb->st_blksize = dev->si_bsize_best; 866 if (sb->st_blksize < dev->si_bsize_phys) 867 sb->st_blksize = dev->si_bsize_phys; 868 if (sb->st_blksize < BLKDEV_IOSIZE) 869 sb->st_blksize = BLKDEV_IOSIZE; 870 } else { 871 sb->st_blksize = PAGE_SIZE; 872 } 873 874 sb->st_flags = vap->va_flags; 875 if (suser_cred(cred, 0)) 876 sb->st_gen = 0; 877 else 878 sb->st_gen = vap->va_gen; 879 880 #if (S_BLKSIZE == 512) 881 /* Optimize this case */ 882 sb->st_blocks = vap->va_bytes >> 9; 883 #else 884 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 885 #endif 886 sb->st_fsmid = vap->va_fsmid; 887 return (0); 888 } 889 890 /* 891 * MPALMOSTSAFE - acquires mplock 892 */ 893 static int 894 vn_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *ucred) 895 { 896 struct vnode *vp = ((struct vnode *)fp->f_data); 897 struct vnode *ovp; 898 struct vattr vattr; 899 int error; 900 901 get_mplock(); 902 903 switch (vp->v_type) { 904 case VREG: 905 case VDIR: 906 if (com == FIONREAD) { 907 if ((error = VOP_GETATTR(vp, &vattr)) != 0) 908 break; 909 *(int *)data = vattr.va_size - fp->f_offset; 910 error = 0; 911 break; 912 } 913 if (com == FIOASYNC) { /* XXX */ 914 error = 0; /* XXX */ 915 break; 916 } 917 /* fall into ... */ 918 default: 919 #if 0 920 return (ENOTTY); 921 #endif 922 case VFIFO: 923 case VCHR: 924 case VBLK: 925 if (com == FIODTYPE) { 926 if (vp->v_type != VCHR && vp->v_type != VBLK) { 927 error = ENOTTY; 928 break; 929 } 930 *(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK; 931 error = 0; 932 break; 933 } 934 error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred); 935 if (error == 0 && com == TIOCSCTTY) { 936 struct proc *p = curthread->td_proc; 937 struct session *sess; 938 939 if (p == NULL) { 940 error = ENOTTY; 941 break; 942 } 943 944 sess = p->p_session; 945 /* Do nothing if reassigning same control tty */ 946 if (sess->s_ttyvp == vp) { 947 error = 0; 948 break; 949 } 950 951 /* Get rid of reference to old control tty */ 952 ovp = sess->s_ttyvp; 953 vref(vp); 954 sess->s_ttyvp = vp; 955 if (ovp) 956 vrele(ovp); 957 } 958 break; 959 } 960 rel_mplock(); 961 return (error); 962 } 963 964 /* 965 * MPALMOSTSAFE - acquires mplock 966 */ 967 static int 968 vn_poll(struct file *fp, int events, struct ucred *cred) 969 { 970 int error; 971 972 get_mplock(); 973 error = VOP_POLL(((struct vnode *)fp->f_data), events, cred); 974 rel_mplock(); 975 return (error); 976 } 977 978 /* 979 * Check that the vnode is still valid, and if so 980 * acquire requested lock. 981 */ 982 int 983 #ifndef DEBUG_LOCKS 984 vn_lock(struct vnode *vp, int flags) 985 #else 986 debug_vn_lock(struct vnode *vp, int flags, const char *filename, int line) 987 #endif 988 { 989 int error; 990 991 do { 992 #ifdef DEBUG_LOCKS 993 vp->filename = filename; 994 vp->line = line; 995 #endif 996 error = VOP_LOCK(vp, flags); 997 if (error == 0) 998 break; 999 } while (flags & LK_RETRY); 1000 1001 /* 1002 * Because we (had better!) have a ref on the vnode, once it 1003 * goes to VRECLAIMED state it will not be recycled until all 1004 * refs go away. So we can just check the flag. 1005 */ 1006 if (error == 0 && (vp->v_flag & VRECLAIMED)) { 1007 VOP_UNLOCK(vp, 0); 1008 error = ENOENT; 1009 } 1010 return (error); 1011 } 1012 1013 /* 1014 * MPALMOSTSAFE - acquires mplock 1015 */ 1016 static int 1017 vn_closefile(struct file *fp) 1018 { 1019 int error; 1020 1021 get_mplock(); 1022 fp->f_ops = &badfileops; 1023 error = vn_close(((struct vnode *)fp->f_data), fp->f_flag); 1024 rel_mplock(); 1025 return(error); 1026 } 1027 1028 /* 1029 * MPALMOSTSAFE - acquires mplock 1030 */ 1031 static int 1032 vn_kqfilter(struct file *fp, struct knote *kn) 1033 { 1034 int error; 1035 1036 get_mplock(); 1037 error = VOP_KQFILTER(((struct vnode *)fp->f_data), kn); 1038 rel_mplock(); 1039 return (error); 1040 } 1041