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 1.44 2006/08/12 00:26:20 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 /* locked vnode is returned */ 189 } else { 190 if (fmode & O_EXCL) { 191 error = EEXIST; 192 } else { 193 error = cache_vget(ncp, cred, 194 LK_EXCLUSIVE, &vp); 195 } 196 if (error) 197 return (error); 198 fmode &= ~O_CREAT; 199 } 200 } else { 201 error = cache_vget(ncp, cred, LK_EXCLUSIVE, &vp); 202 if (error) 203 return (error); 204 } 205 206 /* 207 * We have a locked vnode and ncp now. Note that the ncp will 208 * be cleaned up by the caller if nd->nl_ncp is left intact. 209 */ 210 if (vp->v_type == VLNK) { 211 error = EMLINK; 212 goto bad; 213 } 214 if (vp->v_type == VSOCK) { 215 error = EOPNOTSUPP; 216 goto bad; 217 } 218 if ((fmode & O_CREAT) == 0) { 219 mode = 0; 220 if (fmode & (FWRITE | O_TRUNC)) { 221 if (vp->v_type == VDIR) { 222 error = EISDIR; 223 goto bad; 224 } 225 error = vn_writechk(vp); 226 if (error) { 227 /* 228 * Special stale handling, re-resolve the 229 * vnode. 230 */ 231 if (error == ESTALE) { 232 vput(vp); 233 vp = NULL; 234 cache_setunresolved(ncp); 235 error = cache_resolve(ncp, cred); 236 if (error == 0) 237 goto again; 238 } 239 goto bad; 240 } 241 mode |= VWRITE; 242 } 243 if (fmode & FREAD) 244 mode |= VREAD; 245 if (mode) { 246 error = VOP_ACCESS(vp, mode, cred); 247 if (error) { 248 /* 249 * Special stale handling, re-resolve the 250 * vnode. 251 */ 252 if (error == ESTALE) { 253 vput(vp); 254 vp = NULL; 255 cache_setunresolved(ncp); 256 error = cache_resolve(ncp, cred); 257 if (error == 0) 258 goto again; 259 } 260 goto bad; 261 } 262 } 263 } 264 if (fmode & O_TRUNC) { 265 vn_unlock(vp); /* XXX */ 266 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 267 VATTR_NULL(vap); 268 vap->va_size = 0; 269 error = VOP_SETATTR(vp, vap, cred); 270 if (error) 271 goto bad; 272 } 273 274 /* 275 * Setup the fp so VOP_OPEN can override it. No descriptor has been 276 * associated with the fp yet so we own it clean. f_ncp inherits 277 * nl_ncp . 278 */ 279 if (fp) { 280 if (vp->v_type == VDIR) { 281 fp->f_ncp = nd->nl_ncp; 282 nd->nl_ncp = NULL; 283 cache_unlock(fp->f_ncp); 284 } 285 } 286 287 /* 288 * Get rid of nl_ncp. vn_open does not return it (it returns the 289 * vnode or the file pointer). Note: we can't leave nl_ncp locked 290 * through the VOP_OPEN anyway since the VOP_OPEN may block, e.g. 291 * on /dev/ttyd0 292 */ 293 if (nd->nl_ncp) { 294 cache_put(nd->nl_ncp); 295 nd->nl_ncp = NULL; 296 } 297 298 error = VOP_OPEN(vp, fmode, cred, fp); 299 if (error) { 300 /* 301 * setting f_ops to &badfileops will prevent the descriptor 302 * code from trying to close and release the vnode, since 303 * the open failed we do not want to call close. 304 */ 305 if (fp) { 306 fp->f_data = NULL; 307 fp->f_ops = &badfileops; 308 } 309 goto bad; 310 } 311 312 #if 0 313 /* 314 * Assert that VREG files have been setup for vmio. 315 */ 316 KASSERT(vp->v_type != VREG || vp->v_object != NULL, 317 ("vn_open: regular file was not VMIO enabled!")); 318 #endif 319 320 /* 321 * Return the vnode. XXX needs some cleaning up. The vnode is 322 * only returned in the fp == NULL case. 323 */ 324 if (fp == NULL) { 325 nd->nl_open_vp = vp; 326 nd->nl_vp_fmode = fmode; 327 if ((nd->nl_flags & NLC_LOCKVP) == 0) 328 vn_unlock(vp); 329 } else { 330 vput(vp); 331 } 332 return (0); 333 bad: 334 if (vp) 335 vput(vp); 336 return (error); 337 } 338 339 /* 340 * Check for write permissions on the specified vnode. 341 * Prototype text segments cannot be written. 342 */ 343 int 344 vn_writechk(vp) 345 struct vnode *vp; 346 { 347 348 /* 349 * If there's shared text associated with 350 * the vnode, try to free it up once. If 351 * we fail, we can't allow writing. 352 */ 353 if (vp->v_flag & VTEXT) 354 return (ETXTBSY); 355 return (0); 356 } 357 358 /* 359 * Vnode close call 360 */ 361 int 362 vn_close(struct vnode *vp, int flags) 363 { 364 int error; 365 366 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) == 0) { 367 error = VOP_CLOSE(vp, flags); 368 vn_unlock(vp); 369 } 370 vrele(vp); 371 return (error); 372 } 373 374 static __inline 375 int 376 sequential_heuristic(struct uio *uio, struct file *fp) 377 { 378 /* 379 * Sequential heuristic - detect sequential operation 380 */ 381 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 382 uio->uio_offset == fp->f_nextoff) { 383 int tmpseq = fp->f_seqcount; 384 /* 385 * XXX we assume that the filesystem block size is 386 * the default. Not true, but still gives us a pretty 387 * good indicator of how sequential the read operations 388 * are. 389 */ 390 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE; 391 if (tmpseq > IO_SEQMAX) 392 tmpseq = IO_SEQMAX; 393 fp->f_seqcount = tmpseq; 394 return(fp->f_seqcount << IO_SEQSHIFT); 395 } 396 397 /* 398 * Not sequential, quick draw-down of seqcount 399 */ 400 if (fp->f_seqcount > 1) 401 fp->f_seqcount = 1; 402 else 403 fp->f_seqcount = 0; 404 return(0); 405 } 406 407 /* 408 * Package up an I/O request on a vnode into a uio and do it. 409 */ 410 int 411 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 412 off_t offset, enum uio_seg segflg, int ioflg, 413 struct ucred *cred, int *aresid) 414 { 415 struct uio auio; 416 struct iovec aiov; 417 int error; 418 419 if ((ioflg & IO_NODELOCKED) == 0) 420 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 421 auio.uio_iov = &aiov; 422 auio.uio_iovcnt = 1; 423 aiov.iov_base = base; 424 aiov.iov_len = len; 425 auio.uio_resid = len; 426 auio.uio_offset = offset; 427 auio.uio_segflg = segflg; 428 auio.uio_rw = rw; 429 auio.uio_td = curthread; 430 if (rw == UIO_READ) { 431 error = VOP_READ(vp, &auio, ioflg, cred); 432 } else { 433 error = VOP_WRITE(vp, &auio, ioflg, cred); 434 } 435 if (aresid) 436 *aresid = auio.uio_resid; 437 else 438 if (auio.uio_resid && error == 0) 439 error = EIO; 440 if ((ioflg & IO_NODELOCKED) == 0) 441 vn_unlock(vp); 442 return (error); 443 } 444 445 /* 446 * Package up an I/O request on a vnode into a uio and do it. The I/O 447 * request is split up into smaller chunks and we try to avoid saturating 448 * the buffer cache while potentially holding a vnode locked, so we 449 * check bwillwrite() before calling vn_rdwr(). We also call uio_yield() 450 * to give other processes a chance to lock the vnode (either other processes 451 * core'ing the same binary, or unrelated processes scanning the directory). 452 */ 453 int 454 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 455 off_t offset, enum uio_seg segflg, int ioflg, 456 struct ucred *cred, int *aresid) 457 { 458 int error = 0; 459 460 do { 461 int chunk; 462 463 /* 464 * Force `offset' to a multiple of MAXBSIZE except possibly 465 * for the first chunk, so that filesystems only need to 466 * write full blocks except possibly for the first and last 467 * chunks. 468 */ 469 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 470 471 if (chunk > len) 472 chunk = len; 473 if (rw != UIO_READ && vp->v_type == VREG) 474 bwillwrite(); 475 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 476 ioflg, cred, aresid); 477 len -= chunk; /* aresid calc already includes length */ 478 if (error) 479 break; 480 offset += chunk; 481 base += chunk; 482 uio_yield(); 483 } while (len); 484 if (aresid) 485 *aresid += len; 486 return (error); 487 } 488 489 /* 490 * MPALMOSTSAFE - acquires mplock 491 */ 492 static int 493 vn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 494 { 495 struct vnode *vp; 496 int error, ioflag; 497 498 get_mplock(); 499 KASSERT(uio->uio_td == curthread, 500 ("uio_td %p is not td %p", uio->uio_td, curthread)); 501 vp = (struct vnode *)fp->f_data; 502 503 ioflag = 0; 504 if (flags & O_FBLOCKING) { 505 /* ioflag &= ~IO_NDELAY; */ 506 } else if (flags & O_FNONBLOCKING) { 507 ioflag |= IO_NDELAY; 508 } else if (fp->f_flag & FNONBLOCK) { 509 ioflag |= IO_NDELAY; 510 } 511 if (flags & O_FBUFFERED) { 512 /* ioflag &= ~IO_DIRECT; */ 513 } else if (flags & O_FUNBUFFERED) { 514 ioflag |= IO_DIRECT; 515 } else if (fp->f_flag & O_DIRECT) { 516 ioflag |= IO_DIRECT; 517 } 518 vn_lock(vp, LK_SHARED | LK_RETRY); 519 if ((flags & O_FOFFSET) == 0) 520 uio->uio_offset = fp->f_offset; 521 522 ioflag |= sequential_heuristic(uio, fp); 523 524 error = VOP_READ(vp, uio, ioflag, cred); 525 if ((flags & O_FOFFSET) == 0) 526 fp->f_offset = uio->uio_offset; 527 fp->f_nextoff = uio->uio_offset; 528 vn_unlock(vp); 529 rel_mplock(); 530 return (error); 531 } 532 533 /* 534 * Device-optimized file table vnode read routine. 535 * 536 * This bypasses the VOP table and talks directly to the device. Most 537 * filesystems just route to specfs and can make this optimization. 538 * 539 * MPALMOSTSAFE - acquires mplock 540 */ 541 static int 542 svn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 543 { 544 struct vnode *vp; 545 int ioflag; 546 int error; 547 dev_t dev; 548 549 get_mplock(); 550 KASSERT(uio->uio_td == curthread, 551 ("uio_td %p is not td %p", uio->uio_td, curthread)); 552 553 vp = (struct vnode *)fp->f_data; 554 if (vp == NULL || vp->v_type == VBAD) { 555 error = EBADF; 556 goto done; 557 } 558 559 if ((dev = vp->v_rdev) == NULL) { 560 error = EBADF; 561 goto done; 562 } 563 reference_dev(dev); 564 565 if (uio->uio_resid == 0) { 566 error = 0; 567 goto done; 568 } 569 if ((flags & O_FOFFSET) == 0) 570 uio->uio_offset = fp->f_offset; 571 572 ioflag = 0; 573 if (flags & O_FBLOCKING) { 574 /* ioflag &= ~IO_NDELAY; */ 575 } else if (flags & O_FNONBLOCKING) { 576 ioflag |= IO_NDELAY; 577 } else if (fp->f_flag & FNONBLOCK) { 578 ioflag |= IO_NDELAY; 579 } 580 if (flags & O_FBUFFERED) { 581 /* ioflag &= ~IO_DIRECT; */ 582 } else if (flags & O_FUNBUFFERED) { 583 ioflag |= IO_DIRECT; 584 } else if (fp->f_flag & O_DIRECT) { 585 ioflag |= IO_DIRECT; 586 } 587 ioflag |= sequential_heuristic(uio, fp); 588 589 error = dev_dread(dev, uio, ioflag); 590 591 release_dev(dev); 592 if ((flags & O_FOFFSET) == 0) 593 fp->f_offset = uio->uio_offset; 594 fp->f_nextoff = uio->uio_offset; 595 done: 596 rel_mplock(); 597 return (error); 598 } 599 600 /* 601 * MPALMOSTSAFE - acquires mplock 602 */ 603 static int 604 vn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 605 { 606 struct vnode *vp; 607 int error, ioflag; 608 609 get_mplock(); 610 KASSERT(uio->uio_td == curthread, 611 ("uio_procp %p is not p %p", uio->uio_td, curthread)); 612 vp = (struct vnode *)fp->f_data; 613 if (vp->v_type == VREG) 614 bwillwrite(); 615 vp = (struct vnode *)fp->f_data; /* XXX needed? */ 616 617 ioflag = IO_UNIT; 618 if (vp->v_type == VREG && 619 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 620 ioflag |= IO_APPEND; 621 } 622 623 if (flags & O_FBLOCKING) { 624 /* ioflag &= ~IO_NDELAY; */ 625 } else if (flags & O_FNONBLOCKING) { 626 ioflag |= IO_NDELAY; 627 } else if (fp->f_flag & FNONBLOCK) { 628 ioflag |= IO_NDELAY; 629 } 630 if (flags & O_FBUFFERED) { 631 /* ioflag &= ~IO_DIRECT; */ 632 } else if (flags & O_FUNBUFFERED) { 633 ioflag |= IO_DIRECT; 634 } else if (fp->f_flag & O_DIRECT) { 635 ioflag |= IO_DIRECT; 636 } 637 if (flags & O_FASYNCWRITE) { 638 /* ioflag &= ~IO_SYNC; */ 639 } else if (flags & O_FSYNCWRITE) { 640 ioflag |= IO_SYNC; 641 } else if (fp->f_flag & O_FSYNC) { 642 ioflag |= IO_SYNC; 643 } 644 645 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 646 ioflag |= IO_SYNC; 647 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 648 if ((flags & O_FOFFSET) == 0) 649 uio->uio_offset = fp->f_offset; 650 ioflag |= sequential_heuristic(uio, fp); 651 error = VOP_WRITE(vp, uio, ioflag, cred); 652 if ((flags & O_FOFFSET) == 0) 653 fp->f_offset = uio->uio_offset; 654 fp->f_nextoff = uio->uio_offset; 655 vn_unlock(vp); 656 rel_mplock(); 657 return (error); 658 } 659 660 /* 661 * Device-optimized file table vnode write routine. 662 * 663 * This bypasses the VOP table and talks directly to the device. Most 664 * filesystems just route to specfs and can make this optimization. 665 * 666 * MPALMOSTSAFE - acquires mplock 667 */ 668 static int 669 svn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 670 { 671 struct vnode *vp; 672 int ioflag; 673 int error; 674 dev_t dev; 675 676 get_mplock(); 677 KASSERT(uio->uio_td == curthread, 678 ("uio_procp %p is not p %p", uio->uio_td, curthread)); 679 680 vp = (struct vnode *)fp->f_data; 681 if (vp == NULL || vp->v_type == VBAD) { 682 error = EBADF; 683 goto done; 684 } 685 if (vp->v_type == VREG) 686 bwillwrite(); 687 vp = (struct vnode *)fp->f_data; /* XXX needed? */ 688 689 if ((dev = vp->v_rdev) == NULL) { 690 error = EBADF; 691 goto done; 692 } 693 reference_dev(dev); 694 695 if ((flags & O_FOFFSET) == 0) 696 uio->uio_offset = fp->f_offset; 697 698 ioflag = IO_UNIT; 699 if (vp->v_type == VREG && 700 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 701 ioflag |= IO_APPEND; 702 } 703 704 if (flags & O_FBLOCKING) { 705 /* ioflag &= ~IO_NDELAY; */ 706 } else if (flags & O_FNONBLOCKING) { 707 ioflag |= IO_NDELAY; 708 } else if (fp->f_flag & FNONBLOCK) { 709 ioflag |= IO_NDELAY; 710 } 711 if (flags & O_FBUFFERED) { 712 /* ioflag &= ~IO_DIRECT; */ 713 } else if (flags & O_FUNBUFFERED) { 714 ioflag |= IO_DIRECT; 715 } else if (fp->f_flag & O_DIRECT) { 716 ioflag |= IO_DIRECT; 717 } 718 if (flags & O_FASYNCWRITE) { 719 /* ioflag &= ~IO_SYNC; */ 720 } else if (flags & O_FSYNCWRITE) { 721 ioflag |= IO_SYNC; 722 } else if (fp->f_flag & O_FSYNC) { 723 ioflag |= IO_SYNC; 724 } 725 726 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 727 ioflag |= IO_SYNC; 728 ioflag |= sequential_heuristic(uio, fp); 729 730 error = dev_dwrite(dev, uio, ioflag); 731 732 release_dev(dev); 733 if ((flags & O_FOFFSET) == 0) 734 fp->f_offset = uio->uio_offset; 735 fp->f_nextoff = uio->uio_offset; 736 done: 737 rel_mplock(); 738 return (error); 739 } 740 741 /* 742 * MPALMOSTSAFE - acquires mplock 743 */ 744 static int 745 vn_statfile(struct file *fp, struct stat *sb, struct ucred *cred) 746 { 747 struct vnode *vp; 748 int error; 749 750 get_mplock(); 751 vp = (struct vnode *)fp->f_data; 752 error = vn_stat(vp, sb, cred); 753 rel_mplock(); 754 return (error); 755 } 756 757 int 758 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *cred) 759 { 760 struct vattr vattr; 761 struct vattr *vap; 762 int error; 763 u_short mode; 764 dev_t dev; 765 766 vap = &vattr; 767 error = VOP_GETATTR(vp, vap); 768 if (error) 769 return (error); 770 771 /* 772 * Zero the spare stat fields 773 */ 774 sb->st_lspare = 0; 775 sb->st_qspare = 0; 776 777 /* 778 * Copy from vattr table 779 */ 780 if (vap->va_fsid != VNOVAL) 781 sb->st_dev = vap->va_fsid; 782 else 783 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 784 sb->st_ino = vap->va_fileid; 785 mode = vap->va_mode; 786 switch (vap->va_type) { 787 case VREG: 788 mode |= S_IFREG; 789 break; 790 case VDIR: 791 mode |= S_IFDIR; 792 break; 793 case VBLK: 794 mode |= S_IFBLK; 795 break; 796 case VCHR: 797 mode |= S_IFCHR; 798 break; 799 case VLNK: 800 mode |= S_IFLNK; 801 /* This is a cosmetic change, symlinks do not have a mode. */ 802 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) 803 sb->st_mode &= ~ACCESSPERMS; /* 0000 */ 804 else 805 sb->st_mode |= ACCESSPERMS; /* 0777 */ 806 break; 807 case VSOCK: 808 mode |= S_IFSOCK; 809 break; 810 case VFIFO: 811 mode |= S_IFIFO; 812 break; 813 default: 814 return (EBADF); 815 }; 816 sb->st_mode = mode; 817 sb->st_nlink = vap->va_nlink; 818 sb->st_uid = vap->va_uid; 819 sb->st_gid = vap->va_gid; 820 sb->st_rdev = vap->va_rdev; 821 sb->st_size = vap->va_size; 822 sb->st_atimespec = vap->va_atime; 823 sb->st_mtimespec = vap->va_mtime; 824 sb->st_ctimespec = vap->va_ctime; 825 826 /* 827 * A VCHR and VBLK device may track the last access and last modified 828 * time independantly of the filesystem. This is particularly true 829 * because device read and write calls may bypass the filesystem. 830 */ 831 if (vp->v_type == VCHR || vp->v_type == VBLK) { 832 if ((dev = vp->v_rdev) != NULL) { 833 if (dev->si_lastread) { 834 sb->st_atimespec.tv_sec = dev->si_lastread; 835 sb->st_atimespec.tv_nsec = 0; 836 } 837 if (dev->si_lastwrite) { 838 sb->st_atimespec.tv_sec = dev->si_lastwrite; 839 sb->st_atimespec.tv_nsec = 0; 840 } 841 } 842 } 843 844 /* 845 * According to www.opengroup.org, the meaning of st_blksize is 846 * "a filesystem-specific preferred I/O block size for this 847 * object. In some filesystem types, this may vary from file 848 * to file" 849 * Default to PAGE_SIZE after much discussion. 850 */ 851 852 if (vap->va_type == VREG) { 853 sb->st_blksize = vap->va_blocksize; 854 } else if (vn_isdisk(vp, NULL)) { 855 /* 856 * XXX this is broken. If the device is not yet open (aka 857 * stat() call, aka v_rdev == NULL), how are we supposed 858 * to get a valid block size out of it? 859 */ 860 dev_t dev; 861 862 if ((dev = vp->v_rdev) == NULL) 863 dev = udev2dev(vp->v_udev, vp->v_type == VBLK); 864 sb->st_blksize = dev->si_bsize_best; 865 if (sb->st_blksize < dev->si_bsize_phys) 866 sb->st_blksize = dev->si_bsize_phys; 867 if (sb->st_blksize < BLKDEV_IOSIZE) 868 sb->st_blksize = BLKDEV_IOSIZE; 869 } else { 870 sb->st_blksize = PAGE_SIZE; 871 } 872 873 sb->st_flags = vap->va_flags; 874 if (suser_cred(cred, 0)) 875 sb->st_gen = 0; 876 else 877 sb->st_gen = vap->va_gen; 878 879 #if (S_BLKSIZE == 512) 880 /* Optimize this case */ 881 sb->st_blocks = vap->va_bytes >> 9; 882 #else 883 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 884 #endif 885 sb->st_fsmid = vap->va_fsmid; 886 return (0); 887 } 888 889 /* 890 * MPALMOSTSAFE - acquires mplock 891 */ 892 static int 893 vn_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *ucred) 894 { 895 struct vnode *vp = ((struct vnode *)fp->f_data); 896 struct vnode *ovp; 897 struct vattr vattr; 898 int error; 899 900 get_mplock(); 901 902 switch (vp->v_type) { 903 case VREG: 904 case VDIR: 905 if (com == FIONREAD) { 906 if ((error = VOP_GETATTR(vp, &vattr)) != 0) 907 break; 908 *(int *)data = vattr.va_size - fp->f_offset; 909 error = 0; 910 break; 911 } 912 if (com == FIOASYNC) { /* XXX */ 913 error = 0; /* XXX */ 914 break; 915 } 916 /* fall into ... */ 917 default: 918 #if 0 919 return (ENOTTY); 920 #endif 921 case VFIFO: 922 case VCHR: 923 case VBLK: 924 if (com == FIODTYPE) { 925 if (vp->v_type != VCHR && vp->v_type != VBLK) { 926 error = ENOTTY; 927 break; 928 } 929 *(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK; 930 error = 0; 931 break; 932 } 933 error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred); 934 if (error == 0 && com == TIOCSCTTY) { 935 struct proc *p = curthread->td_proc; 936 struct session *sess; 937 938 if (p == NULL) { 939 error = ENOTTY; 940 break; 941 } 942 943 sess = p->p_session; 944 /* Do nothing if reassigning same control tty */ 945 if (sess->s_ttyvp == vp) { 946 error = 0; 947 break; 948 } 949 950 /* Get rid of reference to old control tty */ 951 ovp = sess->s_ttyvp; 952 vref(vp); 953 sess->s_ttyvp = vp; 954 if (ovp) 955 vrele(ovp); 956 } 957 break; 958 } 959 rel_mplock(); 960 return (error); 961 } 962 963 /* 964 * MPALMOSTSAFE - acquires mplock 965 */ 966 static int 967 vn_poll(struct file *fp, int events, struct ucred *cred) 968 { 969 int error; 970 971 get_mplock(); 972 error = VOP_POLL(((struct vnode *)fp->f_data), events, cred); 973 rel_mplock(); 974 return (error); 975 } 976 977 /* 978 * Check that the vnode is still valid, and if so 979 * acquire requested lock. 980 */ 981 int 982 #ifndef DEBUG_LOCKS 983 vn_lock(struct vnode *vp, int flags) 984 #else 985 debug_vn_lock(struct vnode *vp, int flags, const char *filename, int line) 986 #endif 987 { 988 int error; 989 990 do { 991 #ifdef DEBUG_LOCKS 992 vp->filename = filename; 993 vp->line = line; 994 error = debuglockmgr(&vp->v_lock, flags, 995 "vn_lock", filename, line); 996 #else 997 error = lockmgr(&vp->v_lock, flags); 998 #endif 999 if (error == 0) 1000 break; 1001 } while (flags & LK_RETRY); 1002 1003 /* 1004 * Because we (had better!) have a ref on the vnode, once it 1005 * goes to VRECLAIMED state it will not be recycled until all 1006 * refs go away. So we can just check the flag. 1007 */ 1008 if (error == 0 && (vp->v_flag & VRECLAIMED)) { 1009 lockmgr(&vp->v_lock, LK_RELEASE); 1010 error = ENOENT; 1011 } 1012 return (error); 1013 } 1014 1015 void 1016 vn_unlock(struct vnode *vp) 1017 { 1018 lockmgr(&vp->v_lock, LK_RELEASE); 1019 } 1020 1021 int 1022 vn_islocked(struct vnode *vp) 1023 { 1024 return (lockstatus(&vp->v_lock, curthread)); 1025 } 1026 1027 /* 1028 * MPALMOSTSAFE - acquires mplock 1029 */ 1030 static int 1031 vn_closefile(struct file *fp) 1032 { 1033 int error; 1034 1035 get_mplock(); 1036 fp->f_ops = &badfileops; 1037 error = vn_close(((struct vnode *)fp->f_data), fp->f_flag); 1038 rel_mplock(); 1039 return(error); 1040 } 1041 1042 /* 1043 * MPALMOSTSAFE - acquires mplock 1044 */ 1045 static int 1046 vn_kqfilter(struct file *fp, struct knote *kn) 1047 { 1048 int error; 1049 1050 get_mplock(); 1051 error = VOP_KQFILTER(((struct vnode *)fp->f_data), kn); 1052 rel_mplock(); 1053 return (error); 1054 } 1055