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 1.54 2007/11/02 19:52:25 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 vnode *dvp; 131 struct ucred *cred = nd->nl_cred; 132 struct vattr vat; 133 struct vattr *vap = &vat; 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_nch 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); 170 171 /* 172 * split case to allow us to re-resolve and retry the ncp in case 173 * we get ESTALE. 174 */ 175 again: 176 if (fmode & O_CREAT) { 177 if (nd->nl_nch.ncp->nc_vp == NULL) { 178 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 179 return (error); 180 if ((dvp = nd->nl_nch.ncp->nc_parent->nc_vp) == NULL) 181 return (EPERM); 182 /* vhold(dvp); - dvp can't go away */ 183 VATTR_NULL(vap); 184 vap->va_type = VREG; 185 vap->va_mode = cmode; 186 if (fmode & O_EXCL) 187 vap->va_vaflags |= VA_EXCLUSIVE; 188 error = VOP_NCREATE(&nd->nl_nch, dvp, &vp, 189 nd->nl_cred, vap); 190 /* vdrop(dvp); */ 191 if (error) 192 return (error); 193 fmode &= ~O_TRUNC; 194 /* locked vnode is returned */ 195 } else { 196 if (fmode & O_EXCL) { 197 error = EEXIST; 198 } else { 199 error = cache_vget(&nd->nl_nch, cred, 200 LK_EXCLUSIVE, &vp); 201 } 202 if (error) 203 return (error); 204 fmode &= ~O_CREAT; 205 } 206 } else { 207 error = cache_vget(&nd->nl_nch, cred, LK_EXCLUSIVE, &vp); 208 if (error) 209 return (error); 210 } 211 212 /* 213 * We have a locked vnode and ncp now. Note that the ncp will 214 * be cleaned up by the caller if nd->nl_nch is left intact. 215 */ 216 if (vp->v_type == VLNK) { 217 error = EMLINK; 218 goto bad; 219 } 220 if (vp->v_type == VSOCK) { 221 error = EOPNOTSUPP; 222 goto bad; 223 } 224 if ((fmode & O_CREAT) == 0) { 225 mode = 0; 226 if (fmode & (FWRITE | O_TRUNC)) { 227 if (vp->v_type == VDIR) { 228 error = EISDIR; 229 goto bad; 230 } 231 error = vn_writechk(vp, &nd->nl_nch); 232 if (error) { 233 /* 234 * Special stale handling, re-resolve the 235 * vnode. 236 */ 237 if (error == ESTALE) { 238 vput(vp); 239 vp = NULL; 240 cache_setunresolved(&nd->nl_nch); 241 error = cache_resolve(&nd->nl_nch, cred); 242 if (error == 0) 243 goto again; 244 } 245 goto bad; 246 } 247 mode |= VWRITE; 248 } 249 if (fmode & FREAD) 250 mode |= VREAD; 251 if (mode) { 252 error = VOP_ACCESS(vp, mode, cred); 253 if (error) { 254 /* 255 * Special stale handling, re-resolve the 256 * vnode. 257 */ 258 if (error == ESTALE) { 259 vput(vp); 260 vp = NULL; 261 cache_setunresolved(&nd->nl_nch); 262 error = cache_resolve(&nd->nl_nch, cred); 263 if (error == 0) 264 goto again; 265 } 266 goto bad; 267 } 268 } 269 } 270 if (fmode & O_TRUNC) { 271 vn_unlock(vp); /* XXX */ 272 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 273 VATTR_NULL(vap); 274 vap->va_size = 0; 275 error = VOP_SETATTR(vp, vap, cred); 276 if (error) 277 goto bad; 278 } 279 280 /* 281 * Setup the fp so VOP_OPEN can override it. No descriptor has been 282 * associated with the fp yet so we own it clean. 283 * 284 * f_nchandle inherits nl_nch. This used to be necessary only for 285 * directories but now we do it unconditionally so f*() ops 286 * such as fchmod() can access the actual namespace that was 287 * used to open the file. 288 */ 289 if (fp) { 290 fp->f_nchandle = nd->nl_nch; 291 cache_zero(&nd->nl_nch); 292 cache_unlock(&fp->f_nchandle); 293 } 294 295 /* 296 * Get rid of nl_nch. vn_open does not return it (it returns the 297 * vnode or the file pointer). Note: we can't leave nl_nch locked 298 * through the VOP_OPEN anyway since the VOP_OPEN may block, e.g. 299 * on /dev/ttyd0 300 */ 301 if (nd->nl_nch.ncp) 302 cache_put(&nd->nl_nch); 303 304 error = VOP_OPEN(vp, fmode, cred, fp); 305 if (error) { 306 /* 307 * setting f_ops to &badfileops will prevent the descriptor 308 * code from trying to close and release the vnode, since 309 * the open failed we do not want to call close. 310 */ 311 if (fp) { 312 fp->f_data = NULL; 313 fp->f_ops = &badfileops; 314 } 315 goto bad; 316 } 317 318 #if 0 319 /* 320 * Assert that VREG files have been setup for vmio. 321 */ 322 KASSERT(vp->v_type != VREG || vp->v_object != NULL, 323 ("vn_open: regular file was not VMIO enabled!")); 324 #endif 325 326 /* 327 * Return the vnode. XXX needs some cleaning up. The vnode is 328 * only returned in the fp == NULL case. 329 */ 330 if (fp == NULL) { 331 nd->nl_open_vp = vp; 332 nd->nl_vp_fmode = fmode; 333 if ((nd->nl_flags & NLC_LOCKVP) == 0) 334 vn_unlock(vp); 335 } else { 336 vput(vp); 337 } 338 return (0); 339 bad: 340 if (vp) 341 vput(vp); 342 return (error); 343 } 344 345 int 346 vn_opendisk(const char *devname, int fmode, struct vnode **vpp) 347 { 348 struct vnode *vp; 349 int error; 350 351 if (strncmp(devname, "/dev/", 5) == 0) 352 devname += 5; 353 if ((vp = getsynthvnode(devname)) == NULL) { 354 error = ENODEV; 355 } else { 356 error = VOP_OPEN(vp, fmode, proc0.p_ucred, NULL); 357 vn_unlock(vp); 358 if (error) { 359 vrele(vp); 360 vp = NULL; 361 } 362 } 363 *vpp = vp; 364 return (error); 365 } 366 367 /* 368 * Check for write permissions on the specified vnode. nch may be NULL. 369 */ 370 int 371 vn_writechk(struct vnode *vp, struct nchandle *nch) 372 { 373 /* 374 * If there's shared text associated with 375 * the vnode, try to free it up once. If 376 * we fail, we can't allow writing. 377 */ 378 if (vp->v_flag & VTEXT) 379 return (ETXTBSY); 380 381 /* 382 * If the vnode represents a regular file, check the mount 383 * point via the nch. This may be a different mount point 384 * then the one embedded in the vnode (e.g. nullfs). 385 * 386 * We can still write to non-regular files (e.g. devices) 387 * via read-only mounts. 388 */ 389 if (nch && nch->ncp && vp->v_type == VREG) 390 return (ncp_writechk(nch)); 391 return (0); 392 } 393 394 /* 395 * Check whether the underlying mount is read-only. The mount point 396 * referenced by the namecache may be different from the mount point 397 * used by the underlying vnode in the case of NULLFS, so a separate 398 * check is needed. 399 */ 400 int 401 ncp_writechk(struct nchandle *nch) 402 { 403 if (nch->mount && (nch->mount->mnt_flag & MNT_RDONLY)) 404 return (EROFS); 405 return(0); 406 } 407 408 /* 409 * Vnode close call 410 */ 411 int 412 vn_close(struct vnode *vp, int flags) 413 { 414 int error; 415 416 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) == 0) { 417 error = VOP_CLOSE(vp, flags); 418 vn_unlock(vp); 419 } 420 vrele(vp); 421 return (error); 422 } 423 424 static __inline 425 int 426 sequential_heuristic(struct uio *uio, struct file *fp) 427 { 428 /* 429 * Sequential heuristic - detect sequential operation 430 */ 431 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 432 uio->uio_offset == fp->f_nextoff) { 433 int tmpseq = fp->f_seqcount; 434 /* 435 * XXX we assume that the filesystem block size is 436 * the default. Not true, but still gives us a pretty 437 * good indicator of how sequential the read operations 438 * are. 439 */ 440 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE; 441 if (tmpseq > IO_SEQMAX) 442 tmpseq = IO_SEQMAX; 443 fp->f_seqcount = tmpseq; 444 return(fp->f_seqcount << IO_SEQSHIFT); 445 } 446 447 /* 448 * Not sequential, quick draw-down of seqcount 449 */ 450 if (fp->f_seqcount > 1) 451 fp->f_seqcount = 1; 452 else 453 fp->f_seqcount = 0; 454 return(0); 455 } 456 457 /* 458 * Package up an I/O request on a vnode into a uio and do it. 459 */ 460 int 461 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 462 off_t offset, enum uio_seg segflg, int ioflg, 463 struct ucred *cred, int *aresid) 464 { 465 struct uio auio; 466 struct iovec aiov; 467 struct ccms_lock ccms_lock; 468 int error; 469 470 if ((ioflg & IO_NODELOCKED) == 0) 471 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 472 auio.uio_iov = &aiov; 473 auio.uio_iovcnt = 1; 474 aiov.iov_base = base; 475 aiov.iov_len = len; 476 auio.uio_resid = len; 477 auio.uio_offset = offset; 478 auio.uio_segflg = segflg; 479 auio.uio_rw = rw; 480 auio.uio_td = curthread; 481 ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, &auio); 482 if (rw == UIO_READ) { 483 error = VOP_READ(vp, &auio, ioflg, cred); 484 } else { 485 error = VOP_WRITE(vp, &auio, ioflg, cred); 486 } 487 ccms_lock_put(&vp->v_ccms, &ccms_lock); 488 if (aresid) 489 *aresid = auio.uio_resid; 490 else 491 if (auio.uio_resid && error == 0) 492 error = EIO; 493 if ((ioflg & IO_NODELOCKED) == 0) 494 vn_unlock(vp); 495 return (error); 496 } 497 498 /* 499 * Package up an I/O request on a vnode into a uio and do it. The I/O 500 * request is split up into smaller chunks and we try to avoid saturating 501 * the buffer cache while potentially holding a vnode locked, so we 502 * check bwillwrite() before calling vn_rdwr(). We also call uio_yield() 503 * to give other processes a chance to lock the vnode (either other processes 504 * core'ing the same binary, or unrelated processes scanning the directory). 505 */ 506 int 507 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 508 off_t offset, enum uio_seg segflg, int ioflg, 509 struct ucred *cred, int *aresid) 510 { 511 int error = 0; 512 513 do { 514 int chunk; 515 516 /* 517 * Force `offset' to a multiple of MAXBSIZE except possibly 518 * for the first chunk, so that filesystems only need to 519 * write full blocks except possibly for the first and last 520 * chunks. 521 */ 522 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 523 524 if (chunk > len) 525 chunk = len; 526 if (rw != UIO_READ && vp->v_type == VREG) 527 bwillwrite(); 528 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 529 ioflg, cred, aresid); 530 len -= chunk; /* aresid calc already includes length */ 531 if (error) 532 break; 533 offset += chunk; 534 base += chunk; 535 uio_yield(); 536 } while (len); 537 if (aresid) 538 *aresid += len; 539 return (error); 540 } 541 542 /* 543 * MPALMOSTSAFE - acquires mplock 544 */ 545 static int 546 vn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 547 { 548 struct ccms_lock ccms_lock; 549 struct vnode *vp; 550 int error, ioflag; 551 552 get_mplock(); 553 KASSERT(uio->uio_td == curthread, 554 ("uio_td %p is not td %p", uio->uio_td, curthread)); 555 vp = (struct vnode *)fp->f_data; 556 557 ioflag = 0; 558 if (flags & O_FBLOCKING) { 559 /* ioflag &= ~IO_NDELAY; */ 560 } else if (flags & O_FNONBLOCKING) { 561 ioflag |= IO_NDELAY; 562 } else if (fp->f_flag & FNONBLOCK) { 563 ioflag |= IO_NDELAY; 564 } 565 if (flags & O_FBUFFERED) { 566 /* ioflag &= ~IO_DIRECT; */ 567 } else if (flags & O_FUNBUFFERED) { 568 ioflag |= IO_DIRECT; 569 } else if (fp->f_flag & O_DIRECT) { 570 ioflag |= IO_DIRECT; 571 } 572 vn_lock(vp, LK_SHARED | LK_RETRY); 573 if ((flags & O_FOFFSET) == 0) 574 uio->uio_offset = fp->f_offset; 575 ioflag |= sequential_heuristic(uio, fp); 576 577 ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, uio); 578 error = VOP_READ(vp, uio, ioflag, cred); 579 ccms_lock_put(&vp->v_ccms, &ccms_lock); 580 if ((flags & O_FOFFSET) == 0) 581 fp->f_offset = uio->uio_offset; 582 fp->f_nextoff = uio->uio_offset; 583 vn_unlock(vp); 584 rel_mplock(); 585 return (error); 586 } 587 588 /* 589 * Device-optimized file table vnode read routine. 590 * 591 * This bypasses the VOP table and talks directly to the device. Most 592 * filesystems just route to specfs and can make this optimization. 593 * 594 * MPALMOSTSAFE - acquires mplock 595 */ 596 static int 597 svn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 598 { 599 struct vnode *vp; 600 int ioflag; 601 int error; 602 cdev_t dev; 603 604 get_mplock(); 605 KASSERT(uio->uio_td == curthread, 606 ("uio_td %p is not td %p", uio->uio_td, curthread)); 607 608 vp = (struct vnode *)fp->f_data; 609 if (vp == NULL || vp->v_type == VBAD) { 610 error = EBADF; 611 goto done; 612 } 613 614 if ((dev = vp->v_rdev) == NULL) { 615 error = EBADF; 616 goto done; 617 } 618 reference_dev(dev); 619 620 if (uio->uio_resid == 0) { 621 error = 0; 622 goto done; 623 } 624 if ((flags & O_FOFFSET) == 0) 625 uio->uio_offset = fp->f_offset; 626 627 ioflag = 0; 628 if (flags & O_FBLOCKING) { 629 /* ioflag &= ~IO_NDELAY; */ 630 } else if (flags & O_FNONBLOCKING) { 631 ioflag |= IO_NDELAY; 632 } else if (fp->f_flag & FNONBLOCK) { 633 ioflag |= IO_NDELAY; 634 } 635 if (flags & O_FBUFFERED) { 636 /* ioflag &= ~IO_DIRECT; */ 637 } else if (flags & O_FUNBUFFERED) { 638 ioflag |= IO_DIRECT; 639 } else if (fp->f_flag & O_DIRECT) { 640 ioflag |= IO_DIRECT; 641 } 642 ioflag |= sequential_heuristic(uio, fp); 643 644 error = dev_dread(dev, uio, ioflag); 645 646 release_dev(dev); 647 if ((flags & O_FOFFSET) == 0) 648 fp->f_offset = uio->uio_offset; 649 fp->f_nextoff = uio->uio_offset; 650 done: 651 rel_mplock(); 652 return (error); 653 } 654 655 /* 656 * MPALMOSTSAFE - acquires mplock 657 */ 658 static int 659 vn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 660 { 661 struct ccms_lock ccms_lock; 662 struct vnode *vp; 663 int error, ioflag; 664 665 get_mplock(); 666 KASSERT(uio->uio_td == curthread, 667 ("uio_td %p is not p %p", uio->uio_td, curthread)); 668 vp = (struct vnode *)fp->f_data; 669 if (vp->v_type == VREG) 670 bwillwrite(); 671 vp = (struct vnode *)fp->f_data; /* XXX needed? */ 672 673 ioflag = IO_UNIT; 674 if (vp->v_type == VREG && 675 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 676 ioflag |= IO_APPEND; 677 } 678 679 if (flags & O_FBLOCKING) { 680 /* ioflag &= ~IO_NDELAY; */ 681 } else if (flags & O_FNONBLOCKING) { 682 ioflag |= IO_NDELAY; 683 } else if (fp->f_flag & FNONBLOCK) { 684 ioflag |= IO_NDELAY; 685 } 686 if (flags & O_FBUFFERED) { 687 /* ioflag &= ~IO_DIRECT; */ 688 } else if (flags & O_FUNBUFFERED) { 689 ioflag |= IO_DIRECT; 690 } else if (fp->f_flag & O_DIRECT) { 691 ioflag |= IO_DIRECT; 692 } 693 if (flags & O_FASYNCWRITE) { 694 /* ioflag &= ~IO_SYNC; */ 695 } else if (flags & O_FSYNCWRITE) { 696 ioflag |= IO_SYNC; 697 } else if (fp->f_flag & O_FSYNC) { 698 ioflag |= IO_SYNC; 699 } 700 701 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 702 ioflag |= IO_SYNC; 703 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 704 if ((flags & O_FOFFSET) == 0) 705 uio->uio_offset = fp->f_offset; 706 ioflag |= sequential_heuristic(uio, fp); 707 ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, uio); 708 error = VOP_WRITE(vp, uio, ioflag, cred); 709 ccms_lock_put(&vp->v_ccms, &ccms_lock); 710 if ((flags & O_FOFFSET) == 0) 711 fp->f_offset = uio->uio_offset; 712 fp->f_nextoff = uio->uio_offset; 713 vn_unlock(vp); 714 rel_mplock(); 715 return (error); 716 } 717 718 /* 719 * Device-optimized file table vnode write routine. 720 * 721 * This bypasses the VOP table and talks directly to the device. Most 722 * filesystems just route to specfs and can make this optimization. 723 * 724 * MPALMOSTSAFE - acquires mplock 725 */ 726 static int 727 svn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 728 { 729 struct vnode *vp; 730 int ioflag; 731 int error; 732 cdev_t dev; 733 734 get_mplock(); 735 KASSERT(uio->uio_td == curthread, 736 ("uio_td %p is not p %p", uio->uio_td, curthread)); 737 738 vp = (struct vnode *)fp->f_data; 739 if (vp == NULL || vp->v_type == VBAD) { 740 error = EBADF; 741 goto done; 742 } 743 if (vp->v_type == VREG) 744 bwillwrite(); 745 vp = (struct vnode *)fp->f_data; /* XXX needed? */ 746 747 if ((dev = vp->v_rdev) == NULL) { 748 error = EBADF; 749 goto done; 750 } 751 reference_dev(dev); 752 753 if ((flags & O_FOFFSET) == 0) 754 uio->uio_offset = fp->f_offset; 755 756 ioflag = IO_UNIT; 757 if (vp->v_type == VREG && 758 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 759 ioflag |= IO_APPEND; 760 } 761 762 if (flags & O_FBLOCKING) { 763 /* ioflag &= ~IO_NDELAY; */ 764 } else if (flags & O_FNONBLOCKING) { 765 ioflag |= IO_NDELAY; 766 } else if (fp->f_flag & FNONBLOCK) { 767 ioflag |= IO_NDELAY; 768 } 769 if (flags & O_FBUFFERED) { 770 /* ioflag &= ~IO_DIRECT; */ 771 } else if (flags & O_FUNBUFFERED) { 772 ioflag |= IO_DIRECT; 773 } else if (fp->f_flag & O_DIRECT) { 774 ioflag |= IO_DIRECT; 775 } 776 if (flags & O_FASYNCWRITE) { 777 /* ioflag &= ~IO_SYNC; */ 778 } else if (flags & O_FSYNCWRITE) { 779 ioflag |= IO_SYNC; 780 } else if (fp->f_flag & O_FSYNC) { 781 ioflag |= IO_SYNC; 782 } 783 784 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 785 ioflag |= IO_SYNC; 786 ioflag |= sequential_heuristic(uio, fp); 787 788 error = dev_dwrite(dev, uio, ioflag); 789 790 release_dev(dev); 791 if ((flags & O_FOFFSET) == 0) 792 fp->f_offset = uio->uio_offset; 793 fp->f_nextoff = uio->uio_offset; 794 done: 795 rel_mplock(); 796 return (error); 797 } 798 799 /* 800 * MPALMOSTSAFE - acquires mplock 801 */ 802 static int 803 vn_statfile(struct file *fp, struct stat *sb, struct ucred *cred) 804 { 805 struct vnode *vp; 806 int error; 807 808 get_mplock(); 809 vp = (struct vnode *)fp->f_data; 810 error = vn_stat(vp, sb, cred); 811 rel_mplock(); 812 return (error); 813 } 814 815 int 816 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *cred) 817 { 818 struct vattr vattr; 819 struct vattr *vap; 820 int error; 821 u_short mode; 822 cdev_t dev; 823 824 vap = &vattr; 825 error = VOP_GETATTR(vp, vap); 826 if (error) 827 return (error); 828 829 /* 830 * Zero the spare stat fields 831 */ 832 sb->st_lspare = 0; 833 sb->st_qspare = 0; 834 835 /* 836 * Copy from vattr table 837 */ 838 if (vap->va_fsid != VNOVAL) 839 sb->st_dev = vap->va_fsid; 840 else 841 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 842 sb->st_ino = vap->va_fileid; 843 mode = vap->va_mode; 844 switch (vap->va_type) { 845 case VREG: 846 mode |= S_IFREG; 847 break; 848 case VDATABASE: 849 mode |= S_IFDB; 850 break; 851 case VDIR: 852 mode |= S_IFDIR; 853 break; 854 case VBLK: 855 mode |= S_IFBLK; 856 break; 857 case VCHR: 858 mode |= S_IFCHR; 859 break; 860 case VLNK: 861 mode |= S_IFLNK; 862 /* This is a cosmetic change, symlinks do not have a mode. */ 863 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) 864 sb->st_mode &= ~ACCESSPERMS; /* 0000 */ 865 else 866 sb->st_mode |= ACCESSPERMS; /* 0777 */ 867 break; 868 case VSOCK: 869 mode |= S_IFSOCK; 870 break; 871 case VFIFO: 872 mode |= S_IFIFO; 873 break; 874 default: 875 return (EBADF); 876 }; 877 sb->st_mode = mode; 878 if (vap->va_nlink > (nlink_t)-1) 879 sb->st_nlink = (nlink_t)-1; 880 else 881 sb->st_nlink = vap->va_nlink; 882 sb->st_uid = vap->va_uid; 883 sb->st_gid = vap->va_gid; 884 sb->st_rdev = makeudev(vap->va_rmajor, vap->va_rminor); 885 sb->st_size = vap->va_size; 886 sb->st_atimespec = vap->va_atime; 887 sb->st_mtimespec = vap->va_mtime; 888 sb->st_ctimespec = vap->va_ctime; 889 890 /* 891 * A VCHR and VBLK device may track the last access and last modified 892 * time independantly of the filesystem. This is particularly true 893 * because device read and write calls may bypass the filesystem. 894 */ 895 if (vp->v_type == VCHR || vp->v_type == VBLK) { 896 if ((dev = vp->v_rdev) != NULL) { 897 if (dev->si_lastread) { 898 sb->st_atimespec.tv_sec = dev->si_lastread; 899 sb->st_atimespec.tv_nsec = 0; 900 } 901 if (dev->si_lastwrite) { 902 sb->st_atimespec.tv_sec = dev->si_lastwrite; 903 sb->st_atimespec.tv_nsec = 0; 904 } 905 } 906 } 907 908 /* 909 * According to www.opengroup.org, the meaning of st_blksize is 910 * "a filesystem-specific preferred I/O block size for this 911 * object. In some filesystem types, this may vary from file 912 * to file" 913 * Default to PAGE_SIZE after much discussion. 914 */ 915 916 if (vap->va_type == VREG) { 917 sb->st_blksize = vap->va_blocksize; 918 } else if (vn_isdisk(vp, NULL)) { 919 /* 920 * XXX this is broken. If the device is not yet open (aka 921 * stat() call, aka v_rdev == NULL), how are we supposed 922 * to get a valid block size out of it? 923 */ 924 cdev_t dev; 925 926 if ((dev = vp->v_rdev) == NULL) { 927 if (vp->v_type == VCHR) 928 dev = get_dev(vp->v_umajor, vp->v_uminor); 929 } 930 sb->st_blksize = dev->si_bsize_best; 931 if (sb->st_blksize < dev->si_bsize_phys) 932 sb->st_blksize = dev->si_bsize_phys; 933 if (sb->st_blksize < BLKDEV_IOSIZE) 934 sb->st_blksize = BLKDEV_IOSIZE; 935 } else { 936 sb->st_blksize = PAGE_SIZE; 937 } 938 939 sb->st_flags = vap->va_flags; 940 if (suser_cred(cred, 0)) 941 sb->st_gen = 0; 942 else 943 sb->st_gen = (u_int32_t)vap->va_gen; 944 945 #if (S_BLKSIZE == 512) 946 /* Optimize this case */ 947 sb->st_blocks = vap->va_bytes >> 9; 948 #else 949 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 950 #endif 951 sb->st_fsmid = vap->va_fsmid; 952 return (0); 953 } 954 955 /* 956 * MPALMOSTSAFE - acquires mplock 957 */ 958 static int 959 vn_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *ucred) 960 { 961 struct vnode *vp = ((struct vnode *)fp->f_data); 962 struct vnode *ovp; 963 struct vattr vattr; 964 int error; 965 966 get_mplock(); 967 968 switch (vp->v_type) { 969 case VREG: 970 case VDIR: 971 if (com == FIONREAD) { 972 if ((error = VOP_GETATTR(vp, &vattr)) != 0) 973 break; 974 *(int *)data = vattr.va_size - fp->f_offset; 975 error = 0; 976 break; 977 } 978 if (com == FIOASYNC) { /* XXX */ 979 error = 0; /* XXX */ 980 break; 981 } 982 /* fall into ... */ 983 default: 984 #if 0 985 return (ENOTTY); 986 #endif 987 case VFIFO: 988 case VCHR: 989 case VBLK: 990 if (com == FIODTYPE) { 991 if (vp->v_type != VCHR && vp->v_type != VBLK) { 992 error = ENOTTY; 993 break; 994 } 995 *(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK; 996 error = 0; 997 break; 998 } 999 error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred); 1000 if (error == 0 && com == TIOCSCTTY) { 1001 struct proc *p = curthread->td_proc; 1002 struct session *sess; 1003 1004 if (p == NULL) { 1005 error = ENOTTY; 1006 break; 1007 } 1008 1009 sess = p->p_session; 1010 /* Do nothing if reassigning same control tty */ 1011 if (sess->s_ttyvp == vp) { 1012 error = 0; 1013 break; 1014 } 1015 1016 /* Get rid of reference to old control tty */ 1017 ovp = sess->s_ttyvp; 1018 vref(vp); 1019 sess->s_ttyvp = vp; 1020 if (ovp) 1021 vrele(ovp); 1022 } 1023 break; 1024 } 1025 rel_mplock(); 1026 return (error); 1027 } 1028 1029 /* 1030 * MPALMOSTSAFE - acquires mplock 1031 */ 1032 static int 1033 vn_poll(struct file *fp, int events, struct ucred *cred) 1034 { 1035 int error; 1036 1037 get_mplock(); 1038 error = VOP_POLL(((struct vnode *)fp->f_data), events, cred); 1039 rel_mplock(); 1040 return (error); 1041 } 1042 1043 /* 1044 * Check that the vnode is still valid, and if so 1045 * acquire requested lock. 1046 */ 1047 int 1048 #ifndef DEBUG_LOCKS 1049 vn_lock(struct vnode *vp, int flags) 1050 #else 1051 debug_vn_lock(struct vnode *vp, int flags, const char *filename, int line) 1052 #endif 1053 { 1054 int error; 1055 1056 do { 1057 #ifdef DEBUG_LOCKS 1058 vp->filename = filename; 1059 vp->line = line; 1060 error = debuglockmgr(&vp->v_lock, flags, 1061 "vn_lock", filename, line); 1062 #else 1063 error = lockmgr(&vp->v_lock, flags); 1064 #endif 1065 if (error == 0) 1066 break; 1067 } while (flags & LK_RETRY); 1068 1069 /* 1070 * Because we (had better!) have a ref on the vnode, once it 1071 * goes to VRECLAIMED state it will not be recycled until all 1072 * refs go away. So we can just check the flag. 1073 */ 1074 if (error == 0 && (vp->v_flag & VRECLAIMED)) { 1075 lockmgr(&vp->v_lock, LK_RELEASE); 1076 error = ENOENT; 1077 } 1078 return (error); 1079 } 1080 1081 void 1082 vn_unlock(struct vnode *vp) 1083 { 1084 lockmgr(&vp->v_lock, LK_RELEASE); 1085 } 1086 1087 int 1088 vn_islocked(struct vnode *vp) 1089 { 1090 return (lockstatus(&vp->v_lock, curthread)); 1091 } 1092 1093 /* 1094 * MPALMOSTSAFE - acquires mplock 1095 */ 1096 static int 1097 vn_closefile(struct file *fp) 1098 { 1099 int error; 1100 1101 get_mplock(); 1102 fp->f_ops = &badfileops; 1103 error = vn_close(((struct vnode *)fp->f_data), fp->f_flag); 1104 rel_mplock(); 1105 return(error); 1106 } 1107 1108 /* 1109 * MPALMOSTSAFE - acquires mplock 1110 */ 1111 static int 1112 vn_kqfilter(struct file *fp, struct knote *kn) 1113 { 1114 int error; 1115 1116 get_mplock(); 1117 error = VOP_KQFILTER(((struct vnode *)fp->f_data), kn); 1118 rel_mplock(); 1119 return (error); 1120 } 1121