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 1.49 2006/10/27 04:56:31 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 int mode, error; 134 135 /* 136 * Lookup the path and create or obtain the vnode. After a 137 * successful lookup a locked nd->nl_nch will be returned. 138 * 139 * The result of this section should be a locked vnode. 140 * 141 * XXX with only a little work we should be able to avoid locking 142 * the vnode if FWRITE, O_CREAT, and O_TRUNC are *not* set. 143 */ 144 if (fmode & O_CREAT) { 145 /* 146 * CONDITIONAL CREATE FILE CASE 147 * 148 * Setting NLC_CREATE causes a negative hit to store 149 * the negative hit ncp and not return an error. Then 150 * nc_error or nc_vp may be checked to see if the ncp 151 * represents a negative hit. NLC_CREATE also requires 152 * write permission on the governing directory or EPERM 153 * is returned. 154 */ 155 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 156 nd->nl_flags |= NLC_FOLLOW; 157 nd->nl_flags |= NLC_CREATE; 158 bwillwrite(); 159 error = nlookup(nd); 160 } else { 161 /* 162 * NORMAL OPEN FILE CASE 163 */ 164 error = nlookup(nd); 165 } 166 167 if (error) 168 return (error); 169 170 /* 171 * split case to allow us to re-resolve and retry the ncp in case 172 * we get ESTALE. 173 */ 174 again: 175 if (fmode & O_CREAT) { 176 if (nd->nl_nch.ncp->nc_vp == NULL) { 177 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 178 return (error); 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(&nd->nl_nch, &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(&nd->nl_nch, cred, 194 LK_EXCLUSIVE, &vp); 195 } 196 if (error) 197 return (error); 198 fmode &= ~O_CREAT; 199 } 200 } else { 201 error = cache_vget(&nd->nl_nch, 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_nch 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, &nd->nl_nch); 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(&nd->nl_nch); 235 error = cache_resolve(&nd->nl_nch, 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(&nd->nl_nch); 256 error = cache_resolve(&nd->nl_nch, 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. 277 * 278 * f_nchandle inherits nl_nch. This used to be necessary only for 279 * directories but now we do it unconditionally so f*() ops 280 * such as fchmod() can access the actual namespace that was 281 * used to open the file. 282 */ 283 if (fp) { 284 fp->f_nchandle = nd->nl_nch; 285 cache_zero(&nd->nl_nch); 286 cache_unlock(&fp->f_nchandle); 287 } 288 289 /* 290 * Get rid of nl_nch. vn_open does not return it (it returns the 291 * vnode or the file pointer). Note: we can't leave nl_nch locked 292 * through the VOP_OPEN anyway since the VOP_OPEN may block, e.g. 293 * on /dev/ttyd0 294 */ 295 if (nd->nl_nch.ncp) 296 cache_put(&nd->nl_nch); 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. nch may be NULL. 341 */ 342 int 343 vn_writechk(struct vnode *vp, struct nchandle *nch) 344 { 345 /* 346 * If there's shared text associated with 347 * the vnode, try to free it up once. If 348 * we fail, we can't allow writing. 349 */ 350 if (vp->v_flag & VTEXT) 351 return (ETXTBSY); 352 353 /* 354 * If the vnode represents a regular file, check the mount 355 * point via the nch. This may be a different mount point 356 * then the one embedded in the vnode (e.g. nullfs). 357 * 358 * We can still write to non-regular files (e.g. devices) 359 * via read-only mounts. 360 */ 361 if (nch && nch->ncp && vp->v_type == VREG) 362 return (ncp_writechk(nch)); 363 return (0); 364 } 365 366 /* 367 * Check whether the underlying mount is read-only. The mount point 368 * referenced by the namecache may be different from the mount point 369 * used by the underlying vnode in the case of NULLFS, so a separate 370 * check is needed. 371 */ 372 int 373 ncp_writechk(struct nchandle *nch) 374 { 375 if (nch->mount && (nch->mount->mnt_flag & MNT_RDONLY)) 376 return (EROFS); 377 return(0); 378 } 379 380 /* 381 * Vnode close call 382 */ 383 int 384 vn_close(struct vnode *vp, int flags) 385 { 386 int error; 387 388 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) == 0) { 389 error = VOP_CLOSE(vp, flags); 390 vn_unlock(vp); 391 } 392 vrele(vp); 393 return (error); 394 } 395 396 static __inline 397 int 398 sequential_heuristic(struct uio *uio, struct file *fp) 399 { 400 /* 401 * Sequential heuristic - detect sequential operation 402 */ 403 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 404 uio->uio_offset == fp->f_nextoff) { 405 int tmpseq = fp->f_seqcount; 406 /* 407 * XXX we assume that the filesystem block size is 408 * the default. Not true, but still gives us a pretty 409 * good indicator of how sequential the read operations 410 * are. 411 */ 412 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE; 413 if (tmpseq > IO_SEQMAX) 414 tmpseq = IO_SEQMAX; 415 fp->f_seqcount = tmpseq; 416 return(fp->f_seqcount << IO_SEQSHIFT); 417 } 418 419 /* 420 * Not sequential, quick draw-down of seqcount 421 */ 422 if (fp->f_seqcount > 1) 423 fp->f_seqcount = 1; 424 else 425 fp->f_seqcount = 0; 426 return(0); 427 } 428 429 /* 430 * Package up an I/O request on a vnode into a uio and do it. 431 */ 432 int 433 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 434 off_t offset, enum uio_seg segflg, int ioflg, 435 struct ucred *cred, int *aresid) 436 { 437 struct uio auio; 438 struct iovec aiov; 439 struct ccms_lock ccms_lock; 440 int error; 441 442 if ((ioflg & IO_NODELOCKED) == 0) 443 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 444 auio.uio_iov = &aiov; 445 auio.uio_iovcnt = 1; 446 aiov.iov_base = base; 447 aiov.iov_len = len; 448 auio.uio_resid = len; 449 auio.uio_offset = offset; 450 auio.uio_segflg = segflg; 451 auio.uio_rw = rw; 452 auio.uio_td = curthread; 453 ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, &auio); 454 if (rw == UIO_READ) { 455 error = VOP_READ(vp, &auio, ioflg, cred); 456 } else { 457 error = VOP_WRITE(vp, &auio, ioflg, cred); 458 } 459 ccms_lock_put(&vp->v_ccms, &ccms_lock); 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 vn_unlock(vp); 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(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 480 off_t offset, enum uio_seg segflg, int ioflg, 481 struct ucred *cred, int *aresid) 482 { 483 int error = 0; 484 485 do { 486 int chunk; 487 488 /* 489 * Force `offset' to a multiple of MAXBSIZE except possibly 490 * for the first chunk, so that filesystems only need to 491 * write full blocks except possibly for the first and last 492 * chunks. 493 */ 494 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 495 496 if (chunk > len) 497 chunk = len; 498 if (rw != UIO_READ && vp->v_type == VREG) 499 bwillwrite(); 500 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 501 ioflg, cred, aresid); 502 len -= chunk; /* aresid calc already includes length */ 503 if (error) 504 break; 505 offset += chunk; 506 base += chunk; 507 uio_yield(); 508 } while (len); 509 if (aresid) 510 *aresid += len; 511 return (error); 512 } 513 514 /* 515 * MPALMOSTSAFE - acquires mplock 516 */ 517 static int 518 vn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 519 { 520 struct ccms_lock ccms_lock; 521 struct vnode *vp; 522 int error, ioflag; 523 524 get_mplock(); 525 KASSERT(uio->uio_td == curthread, 526 ("uio_td %p is not td %p", uio->uio_td, curthread)); 527 vp = (struct vnode *)fp->f_data; 528 529 ioflag = 0; 530 if (flags & O_FBLOCKING) { 531 /* ioflag &= ~IO_NDELAY; */ 532 } else if (flags & O_FNONBLOCKING) { 533 ioflag |= IO_NDELAY; 534 } else if (fp->f_flag & FNONBLOCK) { 535 ioflag |= IO_NDELAY; 536 } 537 if (flags & O_FBUFFERED) { 538 /* ioflag &= ~IO_DIRECT; */ 539 } else if (flags & O_FUNBUFFERED) { 540 ioflag |= IO_DIRECT; 541 } else if (fp->f_flag & O_DIRECT) { 542 ioflag |= IO_DIRECT; 543 } 544 vn_lock(vp, LK_SHARED | LK_RETRY); 545 if ((flags & O_FOFFSET) == 0) 546 uio->uio_offset = fp->f_offset; 547 ioflag |= sequential_heuristic(uio, fp); 548 549 ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, uio); 550 error = VOP_READ(vp, uio, ioflag, cred); 551 ccms_lock_put(&vp->v_ccms, &ccms_lock); 552 if ((flags & O_FOFFSET) == 0) 553 fp->f_offset = uio->uio_offset; 554 fp->f_nextoff = uio->uio_offset; 555 vn_unlock(vp); 556 rel_mplock(); 557 return (error); 558 } 559 560 /* 561 * Device-optimized file table vnode read routine. 562 * 563 * This bypasses the VOP table and talks directly to the device. Most 564 * filesystems just route to specfs and can make this optimization. 565 * 566 * MPALMOSTSAFE - acquires mplock 567 */ 568 static int 569 svn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 570 { 571 struct vnode *vp; 572 int ioflag; 573 int error; 574 cdev_t dev; 575 576 get_mplock(); 577 KASSERT(uio->uio_td == curthread, 578 ("uio_td %p is not td %p", uio->uio_td, curthread)); 579 580 vp = (struct vnode *)fp->f_data; 581 if (vp == NULL || vp->v_type == VBAD) { 582 error = EBADF; 583 goto done; 584 } 585 586 if ((dev = vp->v_rdev) == NULL) { 587 error = EBADF; 588 goto done; 589 } 590 reference_dev(dev); 591 592 if (uio->uio_resid == 0) { 593 error = 0; 594 goto done; 595 } 596 if ((flags & O_FOFFSET) == 0) 597 uio->uio_offset = fp->f_offset; 598 599 ioflag = 0; 600 if (flags & O_FBLOCKING) { 601 /* ioflag &= ~IO_NDELAY; */ 602 } else if (flags & O_FNONBLOCKING) { 603 ioflag |= IO_NDELAY; 604 } else if (fp->f_flag & FNONBLOCK) { 605 ioflag |= IO_NDELAY; 606 } 607 if (flags & O_FBUFFERED) { 608 /* ioflag &= ~IO_DIRECT; */ 609 } else if (flags & O_FUNBUFFERED) { 610 ioflag |= IO_DIRECT; 611 } else if (fp->f_flag & O_DIRECT) { 612 ioflag |= IO_DIRECT; 613 } 614 ioflag |= sequential_heuristic(uio, fp); 615 616 error = dev_dread(dev, uio, ioflag); 617 618 release_dev(dev); 619 if ((flags & O_FOFFSET) == 0) 620 fp->f_offset = uio->uio_offset; 621 fp->f_nextoff = uio->uio_offset; 622 done: 623 rel_mplock(); 624 return (error); 625 } 626 627 /* 628 * MPALMOSTSAFE - acquires mplock 629 */ 630 static int 631 vn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 632 { 633 struct ccms_lock ccms_lock; 634 struct vnode *vp; 635 int error, ioflag; 636 637 get_mplock(); 638 KASSERT(uio->uio_td == curthread, 639 ("uio_procp %p is not p %p", uio->uio_td, curthread)); 640 vp = (struct vnode *)fp->f_data; 641 if (vp->v_type == VREG) 642 bwillwrite(); 643 vp = (struct vnode *)fp->f_data; /* XXX needed? */ 644 645 ioflag = IO_UNIT; 646 if (vp->v_type == VREG && 647 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 648 ioflag |= IO_APPEND; 649 } 650 651 if (flags & O_FBLOCKING) { 652 /* ioflag &= ~IO_NDELAY; */ 653 } else if (flags & O_FNONBLOCKING) { 654 ioflag |= IO_NDELAY; 655 } else if (fp->f_flag & FNONBLOCK) { 656 ioflag |= IO_NDELAY; 657 } 658 if (flags & O_FBUFFERED) { 659 /* ioflag &= ~IO_DIRECT; */ 660 } else if (flags & O_FUNBUFFERED) { 661 ioflag |= IO_DIRECT; 662 } else if (fp->f_flag & O_DIRECT) { 663 ioflag |= IO_DIRECT; 664 } 665 if (flags & O_FASYNCWRITE) { 666 /* ioflag &= ~IO_SYNC; */ 667 } else if (flags & O_FSYNCWRITE) { 668 ioflag |= IO_SYNC; 669 } else if (fp->f_flag & O_FSYNC) { 670 ioflag |= IO_SYNC; 671 } 672 673 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 674 ioflag |= IO_SYNC; 675 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 676 if ((flags & O_FOFFSET) == 0) 677 uio->uio_offset = fp->f_offset; 678 ioflag |= sequential_heuristic(uio, fp); 679 ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, uio); 680 error = VOP_WRITE(vp, uio, ioflag, cred); 681 ccms_lock_put(&vp->v_ccms, &ccms_lock); 682 if ((flags & O_FOFFSET) == 0) 683 fp->f_offset = uio->uio_offset; 684 fp->f_nextoff = uio->uio_offset; 685 vn_unlock(vp); 686 rel_mplock(); 687 return (error); 688 } 689 690 /* 691 * Device-optimized file table vnode write routine. 692 * 693 * This bypasses the VOP table and talks directly to the device. Most 694 * filesystems just route to specfs and can make this optimization. 695 * 696 * MPALMOSTSAFE - acquires mplock 697 */ 698 static int 699 svn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 700 { 701 struct vnode *vp; 702 int ioflag; 703 int error; 704 cdev_t dev; 705 706 get_mplock(); 707 KASSERT(uio->uio_td == curthread, 708 ("uio_procp %p is not p %p", uio->uio_td, curthread)); 709 710 vp = (struct vnode *)fp->f_data; 711 if (vp == NULL || vp->v_type == VBAD) { 712 error = EBADF; 713 goto done; 714 } 715 if (vp->v_type == VREG) 716 bwillwrite(); 717 vp = (struct vnode *)fp->f_data; /* XXX needed? */ 718 719 if ((dev = vp->v_rdev) == NULL) { 720 error = EBADF; 721 goto done; 722 } 723 reference_dev(dev); 724 725 if ((flags & O_FOFFSET) == 0) 726 uio->uio_offset = fp->f_offset; 727 728 ioflag = IO_UNIT; 729 if (vp->v_type == VREG && 730 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 731 ioflag |= IO_APPEND; 732 } 733 734 if (flags & O_FBLOCKING) { 735 /* ioflag &= ~IO_NDELAY; */ 736 } else if (flags & O_FNONBLOCKING) { 737 ioflag |= IO_NDELAY; 738 } else if (fp->f_flag & FNONBLOCK) { 739 ioflag |= IO_NDELAY; 740 } 741 if (flags & O_FBUFFERED) { 742 /* ioflag &= ~IO_DIRECT; */ 743 } else if (flags & O_FUNBUFFERED) { 744 ioflag |= IO_DIRECT; 745 } else if (fp->f_flag & O_DIRECT) { 746 ioflag |= IO_DIRECT; 747 } 748 if (flags & O_FASYNCWRITE) { 749 /* ioflag &= ~IO_SYNC; */ 750 } else if (flags & O_FSYNCWRITE) { 751 ioflag |= IO_SYNC; 752 } else if (fp->f_flag & O_FSYNC) { 753 ioflag |= IO_SYNC; 754 } 755 756 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 757 ioflag |= IO_SYNC; 758 ioflag |= sequential_heuristic(uio, fp); 759 760 error = dev_dwrite(dev, uio, ioflag); 761 762 release_dev(dev); 763 if ((flags & O_FOFFSET) == 0) 764 fp->f_offset = uio->uio_offset; 765 fp->f_nextoff = uio->uio_offset; 766 done: 767 rel_mplock(); 768 return (error); 769 } 770 771 /* 772 * MPALMOSTSAFE - acquires mplock 773 */ 774 static int 775 vn_statfile(struct file *fp, struct stat *sb, struct ucred *cred) 776 { 777 struct vnode *vp; 778 int error; 779 780 get_mplock(); 781 vp = (struct vnode *)fp->f_data; 782 error = vn_stat(vp, sb, cred); 783 rel_mplock(); 784 return (error); 785 } 786 787 int 788 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *cred) 789 { 790 struct vattr vattr; 791 struct vattr *vap; 792 int error; 793 u_short mode; 794 cdev_t dev; 795 796 vap = &vattr; 797 error = VOP_GETATTR(vp, vap); 798 if (error) 799 return (error); 800 801 /* 802 * Zero the spare stat fields 803 */ 804 sb->st_lspare = 0; 805 sb->st_qspare = 0; 806 807 /* 808 * Copy from vattr table 809 */ 810 if (vap->va_fsid != VNOVAL) 811 sb->st_dev = vap->va_fsid; 812 else 813 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 814 sb->st_ino = vap->va_fileid; 815 mode = vap->va_mode; 816 switch (vap->va_type) { 817 case VREG: 818 mode |= S_IFREG; 819 break; 820 case VDIR: 821 mode |= S_IFDIR; 822 break; 823 case VBLK: 824 mode |= S_IFBLK; 825 break; 826 case VCHR: 827 mode |= S_IFCHR; 828 break; 829 case VLNK: 830 mode |= S_IFLNK; 831 /* This is a cosmetic change, symlinks do not have a mode. */ 832 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) 833 sb->st_mode &= ~ACCESSPERMS; /* 0000 */ 834 else 835 sb->st_mode |= ACCESSPERMS; /* 0777 */ 836 break; 837 case VSOCK: 838 mode |= S_IFSOCK; 839 break; 840 case VFIFO: 841 mode |= S_IFIFO; 842 break; 843 default: 844 return (EBADF); 845 }; 846 sb->st_mode = mode; 847 sb->st_nlink = vap->va_nlink; 848 sb->st_uid = vap->va_uid; 849 sb->st_gid = vap->va_gid; 850 sb->st_rdev = vap->va_rdev; 851 sb->st_size = vap->va_size; 852 sb->st_atimespec = vap->va_atime; 853 sb->st_mtimespec = vap->va_mtime; 854 sb->st_ctimespec = vap->va_ctime; 855 856 /* 857 * A VCHR and VBLK device may track the last access and last modified 858 * time independantly of the filesystem. This is particularly true 859 * because device read and write calls may bypass the filesystem. 860 */ 861 if (vp->v_type == VCHR || vp->v_type == VBLK) { 862 if ((dev = vp->v_rdev) != NULL) { 863 if (dev->si_lastread) { 864 sb->st_atimespec.tv_sec = dev->si_lastread; 865 sb->st_atimespec.tv_nsec = 0; 866 } 867 if (dev->si_lastwrite) { 868 sb->st_atimespec.tv_sec = dev->si_lastwrite; 869 sb->st_atimespec.tv_nsec = 0; 870 } 871 } 872 } 873 874 /* 875 * According to www.opengroup.org, the meaning of st_blksize is 876 * "a filesystem-specific preferred I/O block size for this 877 * object. In some filesystem types, this may vary from file 878 * to file" 879 * Default to PAGE_SIZE after much discussion. 880 */ 881 882 if (vap->va_type == VREG) { 883 sb->st_blksize = vap->va_blocksize; 884 } else if (vn_isdisk(vp, NULL)) { 885 /* 886 * XXX this is broken. If the device is not yet open (aka 887 * stat() call, aka v_rdev == NULL), how are we supposed 888 * to get a valid block size out of it? 889 */ 890 cdev_t dev; 891 892 if ((dev = vp->v_rdev) == NULL) 893 dev = udev2dev(vp->v_udev, vp->v_type == VBLK); 894 sb->st_blksize = dev->si_bsize_best; 895 if (sb->st_blksize < dev->si_bsize_phys) 896 sb->st_blksize = dev->si_bsize_phys; 897 if (sb->st_blksize < BLKDEV_IOSIZE) 898 sb->st_blksize = BLKDEV_IOSIZE; 899 } else { 900 sb->st_blksize = PAGE_SIZE; 901 } 902 903 sb->st_flags = vap->va_flags; 904 if (suser_cred(cred, 0)) 905 sb->st_gen = 0; 906 else 907 sb->st_gen = vap->va_gen; 908 909 #if (S_BLKSIZE == 512) 910 /* Optimize this case */ 911 sb->st_blocks = vap->va_bytes >> 9; 912 #else 913 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 914 #endif 915 sb->st_fsmid = vap->va_fsmid; 916 return (0); 917 } 918 919 /* 920 * MPALMOSTSAFE - acquires mplock 921 */ 922 static int 923 vn_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *ucred) 924 { 925 struct vnode *vp = ((struct vnode *)fp->f_data); 926 struct vnode *ovp; 927 struct vattr vattr; 928 int error; 929 930 get_mplock(); 931 932 switch (vp->v_type) { 933 case VREG: 934 case VDIR: 935 if (com == FIONREAD) { 936 if ((error = VOP_GETATTR(vp, &vattr)) != 0) 937 break; 938 *(int *)data = vattr.va_size - fp->f_offset; 939 error = 0; 940 break; 941 } 942 if (com == FIOASYNC) { /* XXX */ 943 error = 0; /* XXX */ 944 break; 945 } 946 /* fall into ... */ 947 default: 948 #if 0 949 return (ENOTTY); 950 #endif 951 case VFIFO: 952 case VCHR: 953 case VBLK: 954 if (com == FIODTYPE) { 955 if (vp->v_type != VCHR && vp->v_type != VBLK) { 956 error = ENOTTY; 957 break; 958 } 959 *(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK; 960 error = 0; 961 break; 962 } 963 error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred); 964 if (error == 0 && com == TIOCSCTTY) { 965 struct proc *p = curthread->td_proc; 966 struct session *sess; 967 968 if (p == NULL) { 969 error = ENOTTY; 970 break; 971 } 972 973 sess = p->p_session; 974 /* Do nothing if reassigning same control tty */ 975 if (sess->s_ttyvp == vp) { 976 error = 0; 977 break; 978 } 979 980 /* Get rid of reference to old control tty */ 981 ovp = sess->s_ttyvp; 982 vref(vp); 983 sess->s_ttyvp = vp; 984 if (ovp) 985 vrele(ovp); 986 } 987 break; 988 } 989 rel_mplock(); 990 return (error); 991 } 992 993 /* 994 * MPALMOSTSAFE - acquires mplock 995 */ 996 static int 997 vn_poll(struct file *fp, int events, struct ucred *cred) 998 { 999 int error; 1000 1001 get_mplock(); 1002 error = VOP_POLL(((struct vnode *)fp->f_data), events, cred); 1003 rel_mplock(); 1004 return (error); 1005 } 1006 1007 /* 1008 * Check that the vnode is still valid, and if so 1009 * acquire requested lock. 1010 */ 1011 int 1012 #ifndef DEBUG_LOCKS 1013 vn_lock(struct vnode *vp, int flags) 1014 #else 1015 debug_vn_lock(struct vnode *vp, int flags, const char *filename, int line) 1016 #endif 1017 { 1018 int error; 1019 1020 do { 1021 #ifdef DEBUG_LOCKS 1022 vp->filename = filename; 1023 vp->line = line; 1024 error = debuglockmgr(&vp->v_lock, flags, 1025 "vn_lock", filename, line); 1026 #else 1027 error = lockmgr(&vp->v_lock, flags); 1028 #endif 1029 if (error == 0) 1030 break; 1031 } while (flags & LK_RETRY); 1032 1033 /* 1034 * Because we (had better!) have a ref on the vnode, once it 1035 * goes to VRECLAIMED state it will not be recycled until all 1036 * refs go away. So we can just check the flag. 1037 */ 1038 if (error == 0 && (vp->v_flag & VRECLAIMED)) { 1039 lockmgr(&vp->v_lock, LK_RELEASE); 1040 error = ENOENT; 1041 } 1042 return (error); 1043 } 1044 1045 void 1046 vn_unlock(struct vnode *vp) 1047 { 1048 lockmgr(&vp->v_lock, LK_RELEASE); 1049 } 1050 1051 int 1052 vn_islocked(struct vnode *vp) 1053 { 1054 return (lockstatus(&vp->v_lock, curthread)); 1055 } 1056 1057 /* 1058 * MPALMOSTSAFE - acquires mplock 1059 */ 1060 static int 1061 vn_closefile(struct file *fp) 1062 { 1063 int error; 1064 1065 get_mplock(); 1066 fp->f_ops = &badfileops; 1067 error = vn_close(((struct vnode *)fp->f_data), fp->f_flag); 1068 rel_mplock(); 1069 return(error); 1070 } 1071 1072 /* 1073 * MPALMOSTSAFE - acquires mplock 1074 */ 1075 static int 1076 vn_kqfilter(struct file *fp, struct knote *kn) 1077 { 1078 int error; 1079 1080 get_mplock(); 1081 error = VOP_KQFILTER(((struct vnode *)fp->f_data), kn); 1082 rel_mplock(); 1083 return (error); 1084 } 1085