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 1.58 2008/06/28 17:59:49 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/priv.h> 50 #include <sys/mount.h> 51 #include <sys/nlookup.h> 52 #include <sys/vnode.h> 53 #include <sys/buf.h> 54 #include <sys/filio.h> 55 #include <sys/ttycom.h> 56 #include <sys/conf.h> 57 #include <sys/sysctl.h> 58 #include <sys/syslog.h> 59 60 #include <sys/thread2.h> 61 62 static int vn_closefile (struct file *fp); 63 static int vn_ioctl (struct file *fp, u_long com, caddr_t data, 64 struct ucred *cred, struct sysmsg *msg); 65 static int vn_read (struct file *fp, struct uio *uio, 66 struct ucred *cred, int flags); 67 static int vn_poll (struct file *fp, int events, struct ucred *cred); 68 static int vn_kqfilter (struct file *fp, struct knote *kn); 69 static int vn_statfile (struct file *fp, struct stat *sb, struct ucred *cred); 70 static int vn_write (struct file *fp, struct uio *uio, 71 struct ucred *cred, int flags); 72 73 #ifdef SMP 74 static int read_mpsafe = 0; 75 SYSCTL_INT(_vfs, OID_AUTO, read_mpsafe, CTLFLAG_RW, &read_mpsafe, 0, ""); 76 static int write_mpsafe = 0; 77 SYSCTL_INT(_vfs, OID_AUTO, write_mpsafe, CTLFLAG_RW, &write_mpsafe, 0, ""); 78 static int getattr_mpsafe = 0; 79 SYSCTL_INT(_vfs, OID_AUTO, getattr_mpsafe, CTLFLAG_RW, &getattr_mpsafe, 0, ""); 80 #else 81 #define read_mpsafe 0 82 #define write_mpsafe 0 83 #define getattr_mpsafe 0 84 #endif 85 86 struct fileops vnode_fileops = { 87 .fo_read = vn_read, 88 .fo_write = vn_write, 89 .fo_ioctl = vn_ioctl, 90 .fo_poll = vn_poll, 91 .fo_kqfilter = vn_kqfilter, 92 .fo_stat = vn_statfile, 93 .fo_close = vn_closefile, 94 .fo_shutdown = nofo_shutdown 95 }; 96 97 /* 98 * Common code for vnode open operations. Check permissions, and call 99 * the VOP_NOPEN or VOP_NCREATE routine. 100 * 101 * The caller is responsible for setting up nd with nlookup_init() and 102 * for cleaning it up with nlookup_done(), whether we return an error 103 * or not. 104 * 105 * On success nd->nl_open_vp will hold a referenced and, if requested, 106 * locked vnode. A locked vnode is requested via NLC_LOCKVP. If fp 107 * is non-NULL the vnode will be installed in the file pointer. 108 * 109 * NOTE: The vnode is referenced just once on return whether or not it 110 * is also installed in the file pointer. 111 */ 112 int 113 vn_open(struct nlookupdata *nd, struct file *fp, int fmode, int cmode) 114 { 115 struct vnode *vp; 116 struct ucred *cred = nd->nl_cred; 117 struct vattr vat; 118 struct vattr *vap = &vat; 119 int error; 120 121 /* 122 * Certain combinations are illegal 123 */ 124 if ((fmode & (FWRITE | O_TRUNC)) == O_TRUNC) 125 return(EACCES); 126 127 /* 128 * Lookup the path and create or obtain the vnode. After a 129 * successful lookup a locked nd->nl_nch will be returned. 130 * 131 * The result of this section should be a locked vnode. 132 * 133 * XXX with only a little work we should be able to avoid locking 134 * the vnode if FWRITE, O_CREAT, and O_TRUNC are *not* set. 135 */ 136 nd->nl_flags |= NLC_OPEN; 137 if (fmode & O_APPEND) 138 nd->nl_flags |= NLC_APPEND; 139 if (fmode & O_TRUNC) 140 nd->nl_flags |= NLC_TRUNCATE; 141 if (fmode & FREAD) 142 nd->nl_flags |= NLC_READ; 143 if (fmode & FWRITE) 144 nd->nl_flags |= NLC_WRITE; 145 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 146 nd->nl_flags |= NLC_FOLLOW; 147 148 if (fmode & O_CREAT) { 149 /* 150 * CONDITIONAL CREATE FILE CASE 151 * 152 * Setting NLC_CREATE causes a negative hit to store 153 * the negative hit ncp and not return an error. Then 154 * nc_error or nc_vp may be checked to see if the ncp 155 * represents a negative hit. NLC_CREATE also requires 156 * write permission on the governing directory or EPERM 157 * is returned. 158 */ 159 nd->nl_flags |= NLC_CREATE; 160 nd->nl_flags |= NLC_REFDVP; 161 bwillinode(1); 162 error = nlookup(nd); 163 } else { 164 /* 165 * NORMAL OPEN FILE CASE 166 */ 167 error = nlookup(nd); 168 } 169 170 if (error) 171 return (error); 172 173 /* 174 * split case to allow us to re-resolve and retry the ncp in case 175 * we get ESTALE. 176 */ 177 again: 178 if (fmode & O_CREAT) { 179 if (nd->nl_nch.ncp->nc_vp == NULL) { 180 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 181 return (error); 182 VATTR_NULL(vap); 183 vap->va_type = VREG; 184 vap->va_mode = cmode; 185 if (fmode & O_EXCL) 186 vap->va_vaflags |= VA_EXCLUSIVE; 187 error = VOP_NCREATE(&nd->nl_nch, nd->nl_dvp, &vp, 188 nd->nl_cred, vap); 189 if (error) 190 return (error); 191 fmode &= ~O_TRUNC; 192 /* locked vnode is returned */ 193 } else { 194 if (fmode & O_EXCL) { 195 error = EEXIST; 196 } else { 197 error = cache_vget(&nd->nl_nch, cred, 198 LK_EXCLUSIVE, &vp); 199 } 200 if (error) 201 return (error); 202 fmode &= ~O_CREAT; 203 } 204 } else { 205 error = cache_vget(&nd->nl_nch, cred, LK_EXCLUSIVE, &vp); 206 if (error) 207 return (error); 208 } 209 210 /* 211 * We have a locked vnode and ncp now. Note that the ncp will 212 * be cleaned up by the caller if nd->nl_nch is left intact. 213 */ 214 if (vp->v_type == VLNK) { 215 error = EMLINK; 216 goto bad; 217 } 218 if (vp->v_type == VSOCK) { 219 error = EOPNOTSUPP; 220 goto bad; 221 } 222 if ((fmode & O_CREAT) == 0) { 223 if (fmode & (FWRITE | O_TRUNC)) { 224 if (vp->v_type == VDIR) { 225 error = EISDIR; 226 goto bad; 227 } 228 error = vn_writechk(vp, &nd->nl_nch); 229 if (error) { 230 /* 231 * Special stale handling, re-resolve the 232 * vnode. 233 */ 234 if (error == ESTALE) { 235 vput(vp); 236 vp = NULL; 237 cache_setunresolved(&nd->nl_nch); 238 error = cache_resolve(&nd->nl_nch, cred); 239 if (error == 0) 240 goto again; 241 } 242 goto bad; 243 } 244 } 245 } 246 if (fmode & O_TRUNC) { 247 vn_unlock(vp); /* XXX */ 248 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 249 VATTR_NULL(vap); 250 vap->va_size = 0; 251 error = VOP_SETATTR(vp, vap, cred); 252 if (error) 253 goto bad; 254 } 255 256 /* 257 * Setup the fp so VOP_OPEN can override it. No descriptor has been 258 * associated with the fp yet so we own it clean. 259 * 260 * f_nchandle inherits nl_nch. This used to be necessary only for 261 * directories but now we do it unconditionally so f*() ops 262 * such as fchmod() can access the actual namespace that was 263 * used to open the file. 264 */ 265 if (fp) { 266 if (nd->nl_flags & NLC_APPENDONLY) 267 fmode |= FAPPENDONLY; 268 fp->f_nchandle = nd->nl_nch; 269 cache_zero(&nd->nl_nch); 270 cache_unlock(&fp->f_nchandle); 271 } 272 273 /* 274 * Get rid of nl_nch. vn_open does not return it (it returns the 275 * vnode or the file pointer). Note: we can't leave nl_nch locked 276 * through the VOP_OPEN anyway since the VOP_OPEN may block, e.g. 277 * on /dev/ttyd0 278 */ 279 if (nd->nl_nch.ncp) 280 cache_put(&nd->nl_nch); 281 282 error = VOP_OPEN(vp, fmode, cred, fp); 283 if (error) { 284 /* 285 * setting f_ops to &badfileops will prevent the descriptor 286 * code from trying to close and release the vnode, since 287 * the open failed we do not want to call close. 288 */ 289 if (fp) { 290 fp->f_data = NULL; 291 fp->f_ops = &badfileops; 292 } 293 goto bad; 294 } 295 296 #if 0 297 /* 298 * Assert that VREG files have been setup for vmio. 299 */ 300 KASSERT(vp->v_type != VREG || vp->v_object != NULL, 301 ("vn_open: regular file was not VMIO enabled!")); 302 #endif 303 304 /* 305 * Return the vnode. XXX needs some cleaning up. The vnode is 306 * only returned in the fp == NULL case. 307 */ 308 if (fp == NULL) { 309 nd->nl_open_vp = vp; 310 nd->nl_vp_fmode = fmode; 311 if ((nd->nl_flags & NLC_LOCKVP) == 0) 312 vn_unlock(vp); 313 } else { 314 vput(vp); 315 } 316 return (0); 317 bad: 318 if (vp) 319 vput(vp); 320 return (error); 321 } 322 323 int 324 vn_opendisk(const char *devname, int fmode, struct vnode **vpp) 325 { 326 struct vnode *vp; 327 int error; 328 329 if (strncmp(devname, "/dev/", 5) == 0) 330 devname += 5; 331 if ((vp = getsynthvnode(devname)) == NULL) { 332 error = ENODEV; 333 } else { 334 error = VOP_OPEN(vp, fmode, proc0.p_ucred, NULL); 335 vn_unlock(vp); 336 if (error) { 337 vrele(vp); 338 vp = NULL; 339 } 340 } 341 *vpp = vp; 342 return (error); 343 } 344 345 /* 346 * Check for write permissions on the specified vnode. nch may be NULL. 347 */ 348 int 349 vn_writechk(struct vnode *vp, struct nchandle *nch) 350 { 351 /* 352 * If there's shared text associated with 353 * the vnode, try to free it up once. If 354 * we fail, we can't allow writing. 355 */ 356 if (vp->v_flag & VTEXT) 357 return (ETXTBSY); 358 359 /* 360 * If the vnode represents a regular file, check the mount 361 * point via the nch. This may be a different mount point 362 * then the one embedded in the vnode (e.g. nullfs). 363 * 364 * We can still write to non-regular files (e.g. devices) 365 * via read-only mounts. 366 */ 367 if (nch && nch->ncp && vp->v_type == VREG) 368 return (ncp_writechk(nch)); 369 return (0); 370 } 371 372 /* 373 * Check whether the underlying mount is read-only. The mount point 374 * referenced by the namecache may be different from the mount point 375 * used by the underlying vnode in the case of NULLFS, so a separate 376 * check is needed. 377 */ 378 int 379 ncp_writechk(struct nchandle *nch) 380 { 381 if (nch->mount && (nch->mount->mnt_flag & MNT_RDONLY)) 382 return (EROFS); 383 return(0); 384 } 385 386 /* 387 * Vnode close call 388 */ 389 int 390 vn_close(struct vnode *vp, int flags) 391 { 392 int error; 393 394 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 395 if (error == 0) { 396 error = VOP_CLOSE(vp, flags); 397 vn_unlock(vp); 398 } 399 vrele(vp); 400 return (error); 401 } 402 403 static __inline 404 int 405 sequential_heuristic(struct uio *uio, struct file *fp) 406 { 407 /* 408 * Sequential heuristic - detect sequential operation 409 * 410 * NOTE: SMP: We allow f_seqcount updates to race. 411 */ 412 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 413 uio->uio_offset == fp->f_nextoff) { 414 int tmpseq = fp->f_seqcount; 415 /* 416 * XXX we assume that the filesystem block size is 417 * the default. Not true, but still gives us a pretty 418 * good indicator of how sequential the read operations 419 * are. 420 */ 421 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE; 422 if (tmpseq > IO_SEQMAX) 423 tmpseq = IO_SEQMAX; 424 fp->f_seqcount = tmpseq; 425 return(fp->f_seqcount << IO_SEQSHIFT); 426 } 427 428 /* 429 * Not sequential, quick draw-down of seqcount 430 * 431 * NOTE: SMP: We allow f_seqcount updates to race. 432 */ 433 if (fp->f_seqcount > 1) 434 fp->f_seqcount = 1; 435 else 436 fp->f_seqcount = 0; 437 return(0); 438 } 439 440 /* 441 * get - lock and return the f_offset field. 442 * set - set and unlock the f_offset field. 443 * 444 * These routines serve the dual purpose of serializing access to the 445 * f_offset field (at least on i386) and guaranteeing operational integrity 446 * when multiple read()ers and write()ers are present on the same fp. 447 */ 448 static __inline off_t 449 vn_get_fpf_offset(struct file *fp) 450 { 451 u_int flags; 452 u_int nflags; 453 454 /* 455 * Shortcut critical path. 456 */ 457 flags = fp->f_flag & ~FOFFSETLOCK; 458 if (atomic_cmpset_int(&fp->f_flag, flags, flags | FOFFSETLOCK)) 459 return(fp->f_offset); 460 461 /* 462 * The hard way 463 */ 464 for (;;) { 465 flags = fp->f_flag; 466 if (flags & FOFFSETLOCK) { 467 nflags = flags | FOFFSETWAKE; 468 tsleep_interlock(&fp->f_flag, 0); 469 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) 470 tsleep(&fp->f_flag, PINTERLOCKED, "fpoff", 0); 471 } else { 472 nflags = flags | FOFFSETLOCK; 473 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) 474 break; 475 } 476 } 477 return(fp->f_offset); 478 } 479 480 static __inline void 481 vn_set_fpf_offset(struct file *fp, off_t offset) 482 { 483 u_int flags; 484 u_int nflags; 485 486 /* 487 * We hold the lock so we can set the offset without interference. 488 */ 489 fp->f_offset = offset; 490 491 /* 492 * Normal release is already a reasonably critical path. 493 */ 494 for (;;) { 495 flags = fp->f_flag; 496 nflags = flags & ~(FOFFSETLOCK | FOFFSETWAKE); 497 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) { 498 if (flags & FOFFSETWAKE) 499 wakeup(&fp->f_flag); 500 break; 501 } 502 } 503 } 504 505 static __inline off_t 506 vn_poll_fpf_offset(struct file *fp) 507 { 508 #if defined(__amd64__) || !defined(SMP) 509 return(fp->f_offset); 510 #else 511 off_t off = vn_get_fpf_offset(fp); 512 vn_set_fpf_offset(fp, off); 513 return(off); 514 #endif 515 } 516 517 /* 518 * Package up an I/O request on a vnode into a uio and do it. 519 */ 520 int 521 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 522 off_t offset, enum uio_seg segflg, int ioflg, 523 struct ucred *cred, int *aresid) 524 { 525 struct uio auio; 526 struct iovec aiov; 527 struct ccms_lock ccms_lock; 528 int error; 529 530 if ((ioflg & IO_NODELOCKED) == 0) 531 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 532 auio.uio_iov = &aiov; 533 auio.uio_iovcnt = 1; 534 aiov.iov_base = base; 535 aiov.iov_len = len; 536 auio.uio_resid = len; 537 auio.uio_offset = offset; 538 auio.uio_segflg = segflg; 539 auio.uio_rw = rw; 540 auio.uio_td = curthread; 541 ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, &auio); 542 if (rw == UIO_READ) { 543 error = VOP_READ(vp, &auio, ioflg, cred); 544 } else { 545 error = VOP_WRITE(vp, &auio, ioflg, cred); 546 } 547 ccms_lock_put(&vp->v_ccms, &ccms_lock); 548 if (aresid) 549 *aresid = auio.uio_resid; 550 else 551 if (auio.uio_resid && error == 0) 552 error = EIO; 553 if ((ioflg & IO_NODELOCKED) == 0) 554 vn_unlock(vp); 555 return (error); 556 } 557 558 /* 559 * Package up an I/O request on a vnode into a uio and do it. The I/O 560 * request is split up into smaller chunks and we try to avoid saturating 561 * the buffer cache while potentially holding a vnode locked, so we 562 * check bwillwrite() before calling vn_rdwr(). We also call uio_yield() 563 * to give other processes a chance to lock the vnode (either other processes 564 * core'ing the same binary, or unrelated processes scanning the directory). 565 */ 566 int 567 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 568 off_t offset, enum uio_seg segflg, int ioflg, 569 struct ucred *cred, int *aresid) 570 { 571 int error = 0; 572 573 do { 574 int chunk; 575 576 /* 577 * Force `offset' to a multiple of MAXBSIZE except possibly 578 * for the first chunk, so that filesystems only need to 579 * write full blocks except possibly for the first and last 580 * chunks. 581 */ 582 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 583 584 if (chunk > len) 585 chunk = len; 586 if (vp->v_type == VREG) { 587 switch(rw) { 588 case UIO_READ: 589 bwillread(chunk); 590 break; 591 case UIO_WRITE: 592 bwillwrite(chunk); 593 break; 594 } 595 } 596 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 597 ioflg, cred, aresid); 598 len -= chunk; /* aresid calc already includes length */ 599 if (error) 600 break; 601 offset += chunk; 602 base += chunk; 603 uio_yield(); 604 } while (len); 605 if (aresid) 606 *aresid += len; 607 return (error); 608 } 609 610 /* 611 * MPALMOSTSAFE - acquires mplock 612 * 613 * File pointers can no longer get ripped up by revoke so 614 * we don't need to lock access to the vp. 615 * 616 * f_offset updates are not guaranteed against multiple readers 617 */ 618 static int 619 vn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 620 { 621 struct ccms_lock ccms_lock; 622 struct vnode *vp; 623 int error, ioflag; 624 625 KASSERT(uio->uio_td == curthread, 626 ("uio_td %p is not td %p", uio->uio_td, curthread)); 627 vp = (struct vnode *)fp->f_data; 628 629 ioflag = 0; 630 if (flags & O_FBLOCKING) { 631 /* ioflag &= ~IO_NDELAY; */ 632 } else if (flags & O_FNONBLOCKING) { 633 ioflag |= IO_NDELAY; 634 } else if (fp->f_flag & FNONBLOCK) { 635 ioflag |= IO_NDELAY; 636 } 637 if (flags & O_FBUFFERED) { 638 /* ioflag &= ~IO_DIRECT; */ 639 } else if (flags & O_FUNBUFFERED) { 640 ioflag |= IO_DIRECT; 641 } else if (fp->f_flag & O_DIRECT) { 642 ioflag |= IO_DIRECT; 643 } 644 if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0) 645 uio->uio_offset = vn_get_fpf_offset(fp); 646 vn_lock(vp, LK_SHARED | LK_RETRY); 647 ioflag |= sequential_heuristic(uio, fp); 648 649 ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, uio); 650 if (read_mpsafe && (vp->v_flag & VMP_READ)) { 651 error = VOP_READ(vp, uio, ioflag, cred); 652 } else { 653 get_mplock(); 654 error = VOP_READ(vp, uio, ioflag, cred); 655 rel_mplock(); 656 } 657 ccms_lock_put(&vp->v_ccms, &ccms_lock); 658 fp->f_nextoff = uio->uio_offset; 659 vn_unlock(vp); 660 if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0) 661 vn_set_fpf_offset(fp, uio->uio_offset); 662 return (error); 663 } 664 665 /* 666 * MPALMOSTSAFE - acquires mplock 667 */ 668 static int 669 vn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 670 { 671 struct ccms_lock ccms_lock; 672 struct vnode *vp; 673 int error, ioflag; 674 675 KASSERT(uio->uio_td == curthread, 676 ("uio_td %p is not p %p", uio->uio_td, curthread)); 677 vp = (struct vnode *)fp->f_data; 678 679 ioflag = IO_UNIT; 680 if (vp->v_type == VREG && 681 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 682 ioflag |= IO_APPEND; 683 } 684 685 if (flags & O_FBLOCKING) { 686 /* ioflag &= ~IO_NDELAY; */ 687 } else if (flags & O_FNONBLOCKING) { 688 ioflag |= IO_NDELAY; 689 } else if (fp->f_flag & FNONBLOCK) { 690 ioflag |= IO_NDELAY; 691 } 692 if (flags & O_FBUFFERED) { 693 /* ioflag &= ~IO_DIRECT; */ 694 } else if (flags & O_FUNBUFFERED) { 695 ioflag |= IO_DIRECT; 696 } else if (fp->f_flag & O_DIRECT) { 697 ioflag |= IO_DIRECT; 698 } 699 if (flags & O_FASYNCWRITE) { 700 /* ioflag &= ~IO_SYNC; */ 701 } else if (flags & O_FSYNCWRITE) { 702 ioflag |= IO_SYNC; 703 } else if (fp->f_flag & O_FSYNC) { 704 ioflag |= IO_SYNC; 705 } 706 707 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 708 ioflag |= IO_SYNC; 709 if ((flags & O_FOFFSET) == 0) 710 uio->uio_offset = vn_get_fpf_offset(fp); 711 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 712 ioflag |= sequential_heuristic(uio, fp); 713 ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, uio); 714 if (write_mpsafe && (vp->v_flag & VMP_WRITE)) { 715 error = VOP_WRITE(vp, uio, ioflag, cred); 716 } else { 717 get_mplock(); 718 error = VOP_WRITE(vp, uio, ioflag, cred); 719 rel_mplock(); 720 } 721 ccms_lock_put(&vp->v_ccms, &ccms_lock); 722 fp->f_nextoff = uio->uio_offset; 723 vn_unlock(vp); 724 if ((flags & O_FOFFSET) == 0) 725 vn_set_fpf_offset(fp, uio->uio_offset); 726 return (error); 727 } 728 729 /* 730 * MPSAFE 731 */ 732 static int 733 vn_statfile(struct file *fp, struct stat *sb, struct ucred *cred) 734 { 735 struct vnode *vp; 736 int error; 737 738 vp = (struct vnode *)fp->f_data; 739 error = vn_stat(vp, sb, cred); 740 return (error); 741 } 742 743 /* 744 * MPSAFE (if vnode has VMP_GETATTR) 745 */ 746 int 747 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *cred) 748 { 749 struct vattr vattr; 750 struct vattr *vap; 751 int error; 752 u_short mode; 753 cdev_t dev; 754 755 vap = &vattr; 756 if (getattr_mpsafe && (vp->v_flag & VMP_GETATTR)) { 757 error = VOP_GETATTR(vp, vap); 758 } else { 759 get_mplock(); 760 error = VOP_GETATTR(vp, vap); 761 rel_mplock(); 762 } 763 if (error) 764 return (error); 765 766 /* 767 * Zero the spare stat fields 768 */ 769 sb->st_lspare = 0; 770 sb->st_qspare = 0; 771 772 /* 773 * Copy from vattr table 774 */ 775 if (vap->va_fsid != VNOVAL) 776 sb->st_dev = vap->va_fsid; 777 else 778 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 779 sb->st_ino = vap->va_fileid; 780 mode = vap->va_mode; 781 switch (vap->va_type) { 782 case VREG: 783 mode |= S_IFREG; 784 break; 785 case VDATABASE: 786 mode |= S_IFDB; 787 break; 788 case VDIR: 789 mode |= S_IFDIR; 790 break; 791 case VBLK: 792 mode |= S_IFBLK; 793 break; 794 case VCHR: 795 mode |= S_IFCHR; 796 break; 797 case VLNK: 798 mode |= S_IFLNK; 799 /* This is a cosmetic change, symlinks do not have a mode. */ 800 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) 801 sb->st_mode &= ~ACCESSPERMS; /* 0000 */ 802 else 803 sb->st_mode |= ACCESSPERMS; /* 0777 */ 804 break; 805 case VSOCK: 806 mode |= S_IFSOCK; 807 break; 808 case VFIFO: 809 mode |= S_IFIFO; 810 break; 811 default: 812 return (EBADF); 813 } 814 sb->st_mode = mode; 815 if (vap->va_nlink > (nlink_t)-1) 816 sb->st_nlink = (nlink_t)-1; 817 else 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 = dev2udev(vp->v_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 dev = vp->v_rdev; 834 if (dev != NULL) { 835 if (dev->si_lastread) { 836 sb->st_atimespec.tv_sec = dev->si_lastread; 837 sb->st_atimespec.tv_nsec = 0; 838 } 839 if (dev->si_lastwrite) { 840 sb->st_atimespec.tv_sec = dev->si_lastwrite; 841 sb->st_atimespec.tv_nsec = 0; 842 } 843 } 844 } 845 846 /* 847 * According to www.opengroup.org, the meaning of st_blksize is 848 * "a filesystem-specific preferred I/O block size for this 849 * object. In some filesystem types, this may vary from file 850 * to file" 851 * Default to PAGE_SIZE after much discussion. 852 */ 853 854 if (vap->va_type == VREG) { 855 sb->st_blksize = vap->va_blocksize; 856 } else if (vn_isdisk(vp, NULL)) { 857 /* 858 * XXX this is broken. If the device is not yet open (aka 859 * stat() call, aka v_rdev == NULL), how are we supposed 860 * to get a valid block size out of it? 861 */ 862 dev = vp->v_rdev; 863 864 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; 874 875 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0); 876 if (error) 877 sb->st_gen = 0; 878 else 879 sb->st_gen = (u_int32_t)vap->va_gen; 880 881 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 882 sb->st_fsmid = vap->va_fsmid; 883 return (0); 884 } 885 886 /* 887 * MPALMOSTSAFE - acquires mplock 888 */ 889 static int 890 vn_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *ucred, 891 struct sysmsg *msg) 892 { 893 struct vnode *vp = ((struct vnode *)fp->f_data); 894 struct vnode *ovp; 895 struct vattr vattr; 896 int error; 897 off_t size; 898 899 get_mplock(); 900 901 switch (vp->v_type) { 902 case VREG: 903 case VDIR: 904 if (com == FIONREAD) { 905 error = VOP_GETATTR(vp, &vattr); 906 if (error) 907 break; 908 size = vattr.va_size; 909 if ((vp->v_flag & VNOTSEEKABLE) == 0) 910 size -= vn_poll_fpf_offset(fp); 911 if (size > 0x7FFFFFFF) 912 size = 0x7FFFFFFF; 913 *(int *)data = size; 914 error = 0; 915 break; 916 } 917 if (com == FIOASYNC) { /* XXX */ 918 error = 0; /* XXX */ 919 break; 920 } 921 /* fall into ... */ 922 default: 923 #if 0 924 return (ENOTTY); 925 #endif 926 case VFIFO: 927 case VCHR: 928 case VBLK: 929 if (com == FIODTYPE) { 930 if (vp->v_type != VCHR && vp->v_type != VBLK) { 931 error = ENOTTY; 932 break; 933 } 934 *(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK; 935 error = 0; 936 break; 937 } 938 error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred, msg); 939 if (error == 0 && com == TIOCSCTTY) { 940 struct proc *p = curthread->td_proc; 941 struct session *sess; 942 943 if (p == NULL) { 944 error = ENOTTY; 945 break; 946 } 947 948 sess = p->p_session; 949 /* Do nothing if reassigning same control tty */ 950 if (sess->s_ttyvp == vp) { 951 error = 0; 952 break; 953 } 954 955 /* Get rid of reference to old control tty */ 956 ovp = sess->s_ttyvp; 957 vref(vp); 958 sess->s_ttyvp = vp; 959 if (ovp) 960 vrele(ovp); 961 } 962 break; 963 } 964 rel_mplock(); 965 return (error); 966 } 967 968 /* 969 * MPALMOSTSAFE - acquires mplock 970 */ 971 static int 972 vn_poll(struct file *fp, int events, struct ucred *cred) 973 { 974 int error; 975 976 get_mplock(); 977 error = VOP_POLL(((struct vnode *)fp->f_data), events, cred); 978 rel_mplock(); 979 return (error); 980 } 981 982 /* 983 * Check that the vnode is still valid, and if so 984 * acquire requested lock. 985 */ 986 int 987 #ifndef DEBUG_LOCKS 988 vn_lock(struct vnode *vp, int flags) 989 #else 990 debug_vn_lock(struct vnode *vp, int flags, const char *filename, int line) 991 #endif 992 { 993 int error; 994 995 do { 996 #ifdef DEBUG_LOCKS 997 vp->filename = filename; 998 vp->line = line; 999 error = debuglockmgr(&vp->v_lock, flags, 1000 "vn_lock", filename, line); 1001 #else 1002 error = lockmgr(&vp->v_lock, flags); 1003 #endif 1004 if (error == 0) 1005 break; 1006 } while (flags & LK_RETRY); 1007 1008 /* 1009 * Because we (had better!) have a ref on the vnode, once it 1010 * goes to VRECLAIMED state it will not be recycled until all 1011 * refs go away. So we can just check the flag. 1012 */ 1013 if (error == 0 && (vp->v_flag & VRECLAIMED)) { 1014 lockmgr(&vp->v_lock, LK_RELEASE); 1015 error = ENOENT; 1016 } 1017 return (error); 1018 } 1019 1020 void 1021 vn_unlock(struct vnode *vp) 1022 { 1023 lockmgr(&vp->v_lock, LK_RELEASE); 1024 } 1025 1026 int 1027 vn_islocked(struct vnode *vp) 1028 { 1029 return (lockstatus(&vp->v_lock, curthread)); 1030 } 1031 1032 /* 1033 * MPALMOSTSAFE - acquires mplock 1034 */ 1035 static int 1036 vn_closefile(struct file *fp) 1037 { 1038 int error; 1039 1040 get_mplock(); 1041 fp->f_ops = &badfileops; 1042 error = vn_close(((struct vnode *)fp->f_data), fp->f_flag); 1043 rel_mplock(); 1044 return (error); 1045 } 1046 1047 /* 1048 * MPALMOSTSAFE - acquires mplock 1049 */ 1050 static int 1051 vn_kqfilter(struct file *fp, struct knote *kn) 1052 { 1053 int error; 1054 1055 get_mplock(); 1056 error = VOP_KQFILTER(((struct vnode *)fp->f_data), kn); 1057 rel_mplock(); 1058 return (error); 1059 } 1060