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. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94 35 * $FreeBSD: src/sys/kern/vfs_vnops.c,v 1.87.2.13 2002/12/29 18:19:53 dillon Exp $ 36 */ 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/uio.h> 41 #include <sys/fcntl.h> 42 #include <sys/file.h> 43 #include <sys/stat.h> 44 #include <sys/proc.h> 45 #include <sys/priv.h> 46 #include <sys/mount.h> 47 #include <sys/nlookup.h> 48 #include <sys/vnode.h> 49 #include <sys/buf.h> 50 #include <sys/filio.h> 51 #include <sys/ttycom.h> 52 #include <sys/conf.h> 53 #include <sys/sysctl.h> 54 #include <sys/syslog.h> 55 56 #include <sys/mplock2.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, struct sysmsg *msg); 61 static int vn_read (struct file *fp, struct uio *uio, 62 struct ucred *cred, int flags); 63 static int vn_kqfilter (struct file *fp, struct knote *kn); 64 static int vn_statfile (struct file *fp, struct stat *sb, struct ucred *cred); 65 static int vn_write (struct file *fp, struct uio *uio, 66 struct ucred *cred, int flags); 67 68 struct fileops vnode_fileops = { 69 .fo_read = vn_read, 70 .fo_write = vn_write, 71 .fo_ioctl = vn_ioctl, 72 .fo_kqfilter = vn_kqfilter, 73 .fo_stat = vn_statfile, 74 .fo_close = vn_closefile, 75 .fo_shutdown = nofo_shutdown 76 }; 77 78 /* 79 * Common code for vnode open operations. Check permissions, and call 80 * the VOP_NOPEN or VOP_NCREATE routine. 81 * 82 * The caller is responsible for setting up nd with nlookup_init() and 83 * for cleaning it up with nlookup_done(), whether we return an error 84 * or not. 85 * 86 * On success nd->nl_open_vp will hold a referenced and, if requested, 87 * locked vnode. A locked vnode is requested via NLC_LOCKVP. If fp 88 * is non-NULL the vnode will be installed in the file pointer. 89 * 90 * NOTE: If the caller wishes the namecache entry to be operated with 91 * a shared lock it must use NLC_SHAREDLOCK. If NLC_LOCKVP is set 92 * then the vnode lock will also be shared. 93 * 94 * NOTE: The vnode is referenced just once on return whether or not it 95 * is also installed in the file pointer. 96 */ 97 int 98 vn_open(struct nlookupdata *nd, struct file *fp, int fmode, int cmode) 99 { 100 struct vnode *vp; 101 struct ucred *cred = nd->nl_cred; 102 struct vattr vat; 103 struct vattr *vap = &vat; 104 int error; 105 u_int flags; 106 uint64_t osize; 107 struct mount *mp; 108 109 /* 110 * Certain combinations are illegal 111 */ 112 if ((fmode & (FWRITE | O_TRUNC)) == O_TRUNC) 113 return(EACCES); 114 115 /* 116 * Lookup the path and create or obtain the vnode. After a 117 * successful lookup a locked nd->nl_nch will be returned. 118 * 119 * The result of this section should be a locked vnode. 120 * 121 * XXX with only a little work we should be able to avoid locking 122 * the vnode if FWRITE, O_CREAT, and O_TRUNC are *not* set. 123 */ 124 nd->nl_flags |= NLC_OPEN; 125 if (fmode & O_APPEND) 126 nd->nl_flags |= NLC_APPEND; 127 if (fmode & O_TRUNC) 128 nd->nl_flags |= NLC_TRUNCATE; 129 if (fmode & FREAD) 130 nd->nl_flags |= NLC_READ; 131 if (fmode & FWRITE) 132 nd->nl_flags |= NLC_WRITE; 133 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 134 nd->nl_flags |= NLC_FOLLOW; 135 136 if (fmode & O_CREAT) { 137 /* 138 * CONDITIONAL CREATE FILE CASE 139 * 140 * Setting NLC_CREATE causes a negative hit to store 141 * the negative hit ncp and not return an error. Then 142 * nc_error or nc_vp may be checked to see if the ncp 143 * represents a negative hit. NLC_CREATE also requires 144 * write permission on the governing directory or EPERM 145 * is returned. 146 */ 147 nd->nl_flags |= NLC_CREATE; 148 nd->nl_flags |= NLC_REFDVP; 149 bwillinode(1); 150 error = nlookup(nd); 151 } else { 152 /* 153 * NORMAL OPEN FILE CASE 154 */ 155 error = nlookup(nd); 156 } 157 158 if (error) 159 return (error); 160 161 /* 162 * split case to allow us to re-resolve and retry the ncp in case 163 * we get ESTALE. 164 */ 165 again: 166 if (fmode & O_CREAT) { 167 if (nd->nl_nch.ncp->nc_vp == NULL) { 168 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 169 return (error); 170 VATTR_NULL(vap); 171 vap->va_type = VREG; 172 vap->va_mode = cmode; 173 vap->va_fuseflags = fmode; /* FUSE */ 174 if (fmode & O_EXCL) 175 vap->va_vaflags |= VA_EXCLUSIVE; 176 error = VOP_NCREATE(&nd->nl_nch, nd->nl_dvp, &vp, 177 nd->nl_cred, vap); 178 if (error) 179 return (error); 180 fmode &= ~O_TRUNC; 181 /* locked vnode is returned */ 182 } else { 183 if (fmode & O_EXCL) { 184 error = EEXIST; 185 } else { 186 error = cache_vget(&nd->nl_nch, cred, 187 LK_EXCLUSIVE, &vp); 188 } 189 if (error) 190 return (error); 191 fmode &= ~O_CREAT; 192 } 193 } else { 194 if (nd->nl_flags & NLC_SHAREDLOCK) { 195 error = cache_vget(&nd->nl_nch, cred, LK_SHARED, &vp); 196 } else { 197 error = cache_vget(&nd->nl_nch, cred, 198 LK_EXCLUSIVE, &vp); 199 } 200 if (error) 201 return (error); 202 } 203 204 /* 205 * We have a locked vnode and ncp now. Note that the ncp will 206 * be cleaned up by the caller if nd->nl_nch is left intact. 207 */ 208 if (vp->v_type == VLNK) { 209 error = EMLINK; 210 goto bad; 211 } 212 if (vp->v_type == VSOCK) { 213 error = EOPNOTSUPP; 214 goto bad; 215 } 216 if (vp->v_type != VDIR && (fmode & O_DIRECTORY)) { 217 error = ENOTDIR; 218 goto bad; 219 } 220 if ((fmode & O_CREAT) == 0) { 221 if (fmode & (FWRITE | O_TRUNC)) { 222 if (vp->v_type == VDIR) { 223 error = EISDIR; 224 goto bad; 225 } 226 error = vn_writechk(vp, &nd->nl_nch); 227 if (error) { 228 /* 229 * Special stale handling, re-resolve the 230 * vnode. 231 */ 232 if (error == ESTALE) { 233 vput(vp); 234 vp = NULL; 235 if (nd->nl_flags & NLC_SHAREDLOCK) { 236 cache_unlock(&nd->nl_nch); 237 cache_lock(&nd->nl_nch); 238 } 239 cache_setunresolved(&nd->nl_nch); 240 error = cache_resolve(&nd->nl_nch, 241 cred); 242 if (error == 0) 243 goto again; 244 } 245 goto bad; 246 } 247 } 248 } 249 if (fmode & O_TRUNC) { 250 vn_unlock(vp); /* XXX */ 251 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 252 osize = vp->v_filesize; 253 VATTR_NULL(vap); 254 vap->va_size = 0; 255 error = VOP_SETATTR_FP(vp, vap, cred, fp); 256 if (error) 257 goto bad; 258 error = VOP_GETATTR(vp, vap); 259 if (error) 260 goto bad; 261 mp = vq_vptomp(vp); 262 VFS_ACCOUNT(mp, vap->va_uid, vap->va_gid, -osize); 263 } 264 265 /* 266 * Set or clear VNSWAPCACHE on the vp based on nd->nl_nch.ncp->nc_flag. 267 * These particular bits a tracked all the way from the root. 268 * 269 * NOTE: Might not work properly on NFS servers due to the 270 * disconnected namecache. 271 */ 272 flags = nd->nl_nch.ncp->nc_flag; 273 if ((flags & (NCF_UF_CACHE | NCF_UF_PCACHE)) && 274 (flags & (NCF_SF_NOCACHE | NCF_SF_PNOCACHE)) == 0) { 275 vsetflags(vp, VSWAPCACHE); 276 } else { 277 vclrflags(vp, VSWAPCACHE); 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 if (nd->nl_flags & NLC_APPENDONLY) 291 fmode |= FAPPENDONLY; 292 fp->f_nchandle = nd->nl_nch; 293 cache_zero(&nd->nl_nch); 294 cache_unlock(&fp->f_nchandle); 295 } 296 297 /* 298 * Get rid of nl_nch. vn_open does not return it (it returns the 299 * vnode or the file pointer). Note: we can't leave nl_nch locked 300 * through the VOP_OPEN anyway since the VOP_OPEN may block, e.g. 301 * on /dev/ttyd0 302 */ 303 if (nd->nl_nch.ncp) 304 cache_put(&nd->nl_nch); 305 306 error = VOP_OPEN(vp, fmode, cred, fp); 307 if (error) { 308 /* 309 * setting f_ops to &badfileops will prevent the descriptor 310 * code from trying to close and release the vnode, since 311 * the open failed we do not want to call close. 312 */ 313 if (fp) { 314 fp->f_data = NULL; 315 fp->f_ops = &badfileops; 316 } 317 goto bad; 318 } 319 320 #if 0 321 /* 322 * Assert that VREG files have been setup for vmio. 323 */ 324 KASSERT(vp->v_type != VREG || vp->v_object != NULL, 325 ("vn_open: regular file was not VMIO enabled!")); 326 #endif 327 328 /* 329 * Return the vnode. XXX needs some cleaning up. The vnode is 330 * only returned in the fp == NULL case. 331 */ 332 if (fp == NULL) { 333 nd->nl_open_vp = vp; 334 nd->nl_vp_fmode = fmode; 335 if ((nd->nl_flags & NLC_LOCKVP) == 0) 336 vn_unlock(vp); 337 } else { 338 vput(vp); 339 } 340 return (0); 341 bad: 342 if (vp) 343 vput(vp); 344 return (error); 345 } 346 347 int 348 vn_opendisk(const char *devname, int fmode, struct vnode **vpp) 349 { 350 struct vnode *vp; 351 int error; 352 353 if (strncmp(devname, "/dev/", 5) == 0) 354 devname += 5; 355 if ((vp = getsynthvnode(devname)) == NULL) { 356 error = ENODEV; 357 } else { 358 error = VOP_OPEN(vp, fmode, proc0.p_ucred, NULL); 359 vn_unlock(vp); 360 if (error) { 361 vrele(vp); 362 vp = NULL; 363 } 364 } 365 *vpp = vp; 366 return (error); 367 } 368 369 /* 370 * Check for write permissions on the specified vnode. nch may be NULL. 371 */ 372 int 373 vn_writechk(struct vnode *vp, struct nchandle *nch) 374 { 375 /* 376 * If there's shared text associated with 377 * the vnode, try to free it up once. If 378 * we fail, we can't allow writing. 379 */ 380 if (vp->v_flag & VTEXT) 381 return (ETXTBSY); 382 383 /* 384 * If the vnode represents a regular file, check the mount 385 * point via the nch. This may be a different mount point 386 * then the one embedded in the vnode (e.g. nullfs). 387 * 388 * We can still write to non-regular files (e.g. devices) 389 * via read-only mounts. 390 */ 391 if (nch && nch->ncp && vp->v_type == VREG) 392 return (ncp_writechk(nch)); 393 return (0); 394 } 395 396 /* 397 * Check whether the underlying mount is read-only. The mount point 398 * referenced by the namecache may be different from the mount point 399 * used by the underlying vnode in the case of NULLFS, so a separate 400 * check is needed. 401 */ 402 int 403 ncp_writechk(struct nchandle *nch) 404 { 405 struct mount *mp; 406 407 if ((mp = nch->mount) != NULL) { 408 if (mp->mnt_flag & MNT_RDONLY) 409 return (EROFS); 410 if (mp->mnt_op->vfs_modifying != vfs_stdmodifying) 411 VFS_MODIFYING(mp); 412 } 413 return(0); 414 } 415 416 /* 417 * Vnode close call 418 * 419 * MPSAFE 420 */ 421 int 422 vn_close(struct vnode *vp, int flags, struct file *fp) 423 { 424 int error; 425 426 error = vn_lock(vp, LK_SHARED | LK_RETRY | LK_FAILRECLAIM); 427 if (error == 0) { 428 error = VOP_CLOSE(vp, flags, fp); 429 vn_unlock(vp); 430 } 431 vrele(vp); 432 return (error); 433 } 434 435 /* 436 * Sequential heuristic. 437 * 438 * MPSAFE (f_seqcount and f_nextoff are allowed to race) 439 */ 440 static __inline 441 int 442 sequential_heuristic(struct uio *uio, struct file *fp) 443 { 444 /* 445 * Sequential heuristic - detect sequential operation 446 * 447 * NOTE: SMP: We allow f_seqcount updates to race. 448 */ 449 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 450 uio->uio_offset == fp->f_nextoff) { 451 int tmpseq = fp->f_seqcount; 452 453 tmpseq += (uio->uio_resid + MAXBSIZE - 1) / MAXBSIZE; 454 if (tmpseq > IO_SEQMAX) 455 tmpseq = IO_SEQMAX; 456 fp->f_seqcount = tmpseq; 457 return(fp->f_seqcount << IO_SEQSHIFT); 458 } 459 460 /* 461 * Not sequential, quick draw-down of seqcount 462 * 463 * NOTE: SMP: We allow f_seqcount updates to race. 464 */ 465 if (fp->f_seqcount > 1) 466 fp->f_seqcount = 1; 467 else 468 fp->f_seqcount = 0; 469 return(0); 470 } 471 472 /* 473 * get - lock and return the f_offset field. 474 * set - set and unlock the f_offset field. 475 * 476 * These routines serve the dual purpose of serializing access to the 477 * f_offset field (at least on x86) and guaranteeing operational integrity 478 * when multiple read()ers and write()ers are present on the same fp. 479 * 480 * MPSAFE 481 */ 482 static __inline off_t 483 vn_get_fpf_offset(struct file *fp) 484 { 485 u_int flags; 486 u_int nflags; 487 488 /* 489 * Shortcut critical path. 490 */ 491 flags = fp->f_flag & ~FOFFSETLOCK; 492 if (atomic_cmpset_int(&fp->f_flag, flags, flags | FOFFSETLOCK)) 493 return(fp->f_offset); 494 495 /* 496 * The hard way 497 */ 498 for (;;) { 499 flags = fp->f_flag; 500 if (flags & FOFFSETLOCK) { 501 nflags = flags | FOFFSETWAKE; 502 tsleep_interlock(&fp->f_flag, 0); 503 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) 504 tsleep(&fp->f_flag, PINTERLOCKED, "fpoff", 0); 505 } else { 506 nflags = flags | FOFFSETLOCK; 507 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) 508 break; 509 } 510 } 511 return(fp->f_offset); 512 } 513 514 /* 515 * MPSAFE 516 */ 517 static __inline void 518 vn_set_fpf_offset(struct file *fp, off_t offset) 519 { 520 u_int flags; 521 u_int nflags; 522 523 /* 524 * We hold the lock so we can set the offset without interference. 525 */ 526 fp->f_offset = offset; 527 528 /* 529 * Normal release is already a reasonably critical path. 530 */ 531 for (;;) { 532 flags = fp->f_flag; 533 nflags = flags & ~(FOFFSETLOCK | FOFFSETWAKE); 534 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) { 535 if (flags & FOFFSETWAKE) 536 wakeup(&fp->f_flag); 537 break; 538 } 539 } 540 } 541 542 /* 543 * MPSAFE 544 */ 545 static __inline off_t 546 vn_poll_fpf_offset(struct file *fp) 547 { 548 #if defined(__x86_64__) 549 return(fp->f_offset); 550 #else 551 off_t off = vn_get_fpf_offset(fp); 552 vn_set_fpf_offset(fp, off); 553 return(off); 554 #endif 555 } 556 557 /* 558 * Package up an I/O request on a vnode into a uio and do it. 559 * 560 * MPSAFE 561 */ 562 int 563 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 564 off_t offset, enum uio_seg segflg, int ioflg, 565 struct ucred *cred, int *aresid) 566 { 567 struct uio auio; 568 struct iovec aiov; 569 int error; 570 571 if ((ioflg & IO_NODELOCKED) == 0) 572 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 573 auio.uio_iov = &aiov; 574 auio.uio_iovcnt = 1; 575 aiov.iov_base = base; 576 aiov.iov_len = len; 577 auio.uio_resid = len; 578 auio.uio_offset = offset; 579 auio.uio_segflg = segflg; 580 auio.uio_rw = rw; 581 auio.uio_td = curthread; 582 if (rw == UIO_READ) { 583 error = VOP_READ(vp, &auio, ioflg, cred); 584 } else { 585 error = VOP_WRITE(vp, &auio, ioflg, cred); 586 } 587 if (aresid) 588 *aresid = auio.uio_resid; 589 else 590 if (auio.uio_resid && error == 0) 591 error = EIO; 592 if ((ioflg & IO_NODELOCKED) == 0) 593 vn_unlock(vp); 594 return (error); 595 } 596 597 /* 598 * Package up an I/O request on a vnode into a uio and do it. The I/O 599 * request is split up into smaller chunks and we try to avoid saturating 600 * the buffer cache while potentially holding a vnode locked, so we 601 * check bwillwrite() before calling vn_rdwr(). We also call lwkt_user_yield() 602 * to give other processes a chance to lock the vnode (either other processes 603 * core'ing the same binary, or unrelated processes scanning the directory). 604 * 605 * MPSAFE 606 */ 607 int 608 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 609 off_t offset, enum uio_seg segflg, int ioflg, 610 struct ucred *cred, int *aresid) 611 { 612 int error = 0; 613 614 do { 615 int chunk; 616 617 /* 618 * Force `offset' to a multiple of MAXBSIZE except possibly 619 * for the first chunk, so that filesystems only need to 620 * write full blocks except possibly for the first and last 621 * chunks. 622 */ 623 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 624 625 if (chunk > len) 626 chunk = len; 627 if (vp->v_type == VREG && (ioflg & IO_RECURSE) == 0) { 628 switch(rw) { 629 case UIO_READ: 630 bwillread(chunk); 631 break; 632 case UIO_WRITE: 633 bwillwrite(chunk); 634 break; 635 } 636 } 637 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 638 ioflg, cred, aresid); 639 len -= chunk; /* aresid calc already includes length */ 640 if (error) 641 break; 642 offset += chunk; 643 base += chunk; 644 lwkt_user_yield(); 645 } while (len); 646 if (aresid) 647 *aresid += len; 648 return (error); 649 } 650 651 /* 652 * File pointers can no longer get ripped up by revoke so 653 * we don't need to lock access to the vp. 654 * 655 * f_offset updates are not guaranteed against multiple readers 656 */ 657 static int 658 vn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 659 { 660 struct vnode *vp; 661 int error, ioflag; 662 663 KASSERT(uio->uio_td == curthread, 664 ("uio_td %p is not td %p", uio->uio_td, curthread)); 665 vp = (struct vnode *)fp->f_data; 666 667 ioflag = 0; 668 if (flags & O_FBLOCKING) { 669 /* ioflag &= ~IO_NDELAY; */ 670 } else if (flags & O_FNONBLOCKING) { 671 ioflag |= IO_NDELAY; 672 } else if (fp->f_flag & FNONBLOCK) { 673 ioflag |= IO_NDELAY; 674 } 675 if (fp->f_flag & O_DIRECT) { 676 ioflag |= IO_DIRECT; 677 } 678 if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0) 679 uio->uio_offset = vn_get_fpf_offset(fp); 680 vn_lock(vp, LK_SHARED | LK_RETRY); 681 ioflag |= sequential_heuristic(uio, fp); 682 683 error = VOP_READ_FP(vp, uio, ioflag, cred, fp); 684 fp->f_nextoff = uio->uio_offset; 685 vn_unlock(vp); 686 if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0) 687 vn_set_fpf_offset(fp, uio->uio_offset); 688 return (error); 689 } 690 691 /* 692 * MPSAFE 693 */ 694 static int 695 vn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 696 { 697 struct vnode *vp; 698 int error, ioflag; 699 700 KASSERT(uio->uio_td == curthread, 701 ("uio_td %p is not p %p", uio->uio_td, curthread)); 702 vp = (struct vnode *)fp->f_data; 703 704 ioflag = IO_UNIT; 705 if (vp->v_type == VREG && 706 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 707 ioflag |= IO_APPEND; 708 } 709 710 if (flags & O_FBLOCKING) { 711 /* ioflag &= ~IO_NDELAY; */ 712 } else if (flags & O_FNONBLOCKING) { 713 ioflag |= IO_NDELAY; 714 } else if (fp->f_flag & FNONBLOCK) { 715 ioflag |= IO_NDELAY; 716 } 717 if (fp->f_flag & O_DIRECT) { 718 ioflag |= IO_DIRECT; 719 } 720 if (flags & O_FASYNCWRITE) { 721 /* ioflag &= ~IO_SYNC; */ 722 } else if (flags & O_FSYNCWRITE) { 723 ioflag |= IO_SYNC; 724 } else if (fp->f_flag & O_FSYNC) { 725 ioflag |= IO_SYNC; 726 } 727 728 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 729 ioflag |= IO_SYNC; 730 if ((flags & O_FOFFSET) == 0) 731 uio->uio_offset = vn_get_fpf_offset(fp); 732 if (vp->v_mount) 733 VFS_MODIFYING(vp->v_mount); 734 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 735 ioflag |= sequential_heuristic(uio, fp); 736 error = VOP_WRITE_FP(vp, uio, ioflag, cred, fp); 737 fp->f_nextoff = uio->uio_offset; 738 vn_unlock(vp); 739 if ((flags & O_FOFFSET) == 0) 740 vn_set_fpf_offset(fp, uio->uio_offset); 741 return (error); 742 } 743 744 /* 745 * MPSAFE 746 */ 747 static int 748 vn_statfile(struct file *fp, struct stat *sb, struct ucred *cred) 749 { 750 struct vnode *vp; 751 int error; 752 753 vp = (struct vnode *)fp->f_data; 754 error = vn_stat(vp, sb, cred); 755 return (error); 756 } 757 758 /* 759 * MPSAFE 760 */ 761 int 762 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *cred) 763 { 764 struct vattr vattr; 765 struct vattr *vap; 766 int error; 767 u_short mode; 768 cdev_t dev; 769 770 vap = &vattr; 771 error = VOP_GETATTR(vp, vap); 772 if (error) 773 return (error); 774 775 /* 776 * Zero the spare stat fields 777 */ 778 sb->st_lspare = 0; 779 sb->st_qspare2 = 0; 780 781 /* 782 * Copy from vattr table 783 */ 784 if (vap->va_fsid != VNOVAL) 785 sb->st_dev = vap->va_fsid; 786 else 787 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 788 sb->st_ino = vap->va_fileid; 789 mode = vap->va_mode; 790 switch (vap->va_type) { 791 case VREG: 792 mode |= S_IFREG; 793 break; 794 case VDATABASE: 795 mode |= S_IFDB; 796 break; 797 case VDIR: 798 mode |= S_IFDIR; 799 break; 800 case VBLK: 801 mode |= S_IFBLK; 802 break; 803 case VCHR: 804 mode |= S_IFCHR; 805 break; 806 case VLNK: 807 mode |= S_IFLNK; 808 /* This is a cosmetic change, symlinks do not have a mode. */ 809 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) 810 sb->st_mode &= ~ACCESSPERMS; /* 0000 */ 811 else 812 sb->st_mode |= ACCESSPERMS; /* 0777 */ 813 break; 814 case VSOCK: 815 mode |= S_IFSOCK; 816 break; 817 case VFIFO: 818 mode |= S_IFIFO; 819 break; 820 default: 821 return (EBADF); 822 } 823 sb->st_mode = mode; 824 if (vap->va_nlink > (nlink_t)-1) 825 sb->st_nlink = (nlink_t)-1; 826 else 827 sb->st_nlink = vap->va_nlink; 828 sb->st_uid = vap->va_uid; 829 sb->st_gid = vap->va_gid; 830 sb->st_rdev = dev2udev(vp->v_rdev); 831 sb->st_size = vap->va_size; 832 sb->st_atimespec = vap->va_atime; 833 sb->st_mtimespec = vap->va_mtime; 834 sb->st_ctimespec = vap->va_ctime; 835 836 /* 837 * A VCHR and VBLK device may track the last access and last modified 838 * time independantly of the filesystem. This is particularly true 839 * because device read and write calls may bypass the filesystem. 840 */ 841 if (vp->v_type == VCHR || vp->v_type == VBLK) { 842 dev = vp->v_rdev; 843 if (dev != NULL) { 844 if (dev->si_lastread) { 845 sb->st_atimespec.tv_sec = time_second + 846 (time_uptime - 847 dev->si_lastread); 848 sb->st_atimespec.tv_nsec = 0; 849 } 850 if (dev->si_lastwrite) { 851 sb->st_atimespec.tv_sec = time_second + 852 (time_uptime - 853 dev->si_lastwrite); 854 sb->st_atimespec.tv_nsec = 0; 855 } 856 } 857 } 858 859 /* 860 * According to www.opengroup.org, the meaning of st_blksize is 861 * "a filesystem-specific preferred I/O block size for this 862 * object. In some filesystem types, this may vary from file 863 * to file" 864 * Default to PAGE_SIZE after much discussion. 865 */ 866 867 if (vap->va_type == VREG) { 868 sb->st_blksize = vap->va_blocksize; 869 } else if (vn_isdisk(vp, NULL)) { 870 /* 871 * XXX this is broken. If the device is not yet open (aka 872 * stat() call, aka v_rdev == NULL), how are we supposed 873 * to get a valid block size out of it? 874 */ 875 dev = vp->v_rdev; 876 877 sb->st_blksize = dev->si_bsize_best; 878 if (sb->st_blksize < dev->si_bsize_phys) 879 sb->st_blksize = dev->si_bsize_phys; 880 if (sb->st_blksize < BLKDEV_IOSIZE) 881 sb->st_blksize = BLKDEV_IOSIZE; 882 } else { 883 sb->st_blksize = PAGE_SIZE; 884 } 885 886 sb->st_flags = vap->va_flags; 887 888 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0); 889 if (error) 890 sb->st_gen = 0; 891 else 892 sb->st_gen = (u_int32_t)vap->va_gen; 893 894 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 895 896 /* 897 * This is for ABI compatibility <= 5.7 (for ABI change made in 898 * 5.7 master). 899 */ 900 sb->__old_st_blksize = sb->st_blksize; 901 902 return (0); 903 } 904 905 /* 906 * MPALMOSTSAFE - acquires mplock 907 */ 908 static int 909 vn_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *ucred, 910 struct sysmsg *msg) 911 { 912 struct vnode *vp = ((struct vnode *)fp->f_data); 913 struct vnode *ovp; 914 struct vattr vattr; 915 int error; 916 off_t size; 917 918 switch (vp->v_type) { 919 case VREG: 920 case VDIR: 921 if (com == FIONREAD) { 922 error = VOP_GETATTR(vp, &vattr); 923 if (error) 924 break; 925 size = vattr.va_size; 926 if ((vp->v_flag & VNOTSEEKABLE) == 0) 927 size -= vn_poll_fpf_offset(fp); 928 if (size > 0x7FFFFFFF) 929 size = 0x7FFFFFFF; 930 *(int *)data = size; 931 error = 0; 932 break; 933 } 934 if (com == FIOASYNC) { /* XXX */ 935 error = 0; /* XXX */ 936 break; 937 } 938 /* fall into ... */ 939 default: 940 #if 0 941 return (ENOTTY); 942 #endif 943 case VFIFO: 944 case VCHR: 945 case VBLK: 946 if (com == FIODTYPE) { 947 if (vp->v_type != VCHR && vp->v_type != VBLK) { 948 error = ENOTTY; 949 break; 950 } 951 *(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK; 952 error = 0; 953 break; 954 } 955 error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred, msg); 956 if (error == 0 && com == TIOCSCTTY) { 957 struct proc *p = curthread->td_proc; 958 struct session *sess; 959 960 if (p == NULL) { 961 error = ENOTTY; 962 break; 963 } 964 965 get_mplock(); 966 sess = p->p_session; 967 /* Do nothing if reassigning same control tty */ 968 if (sess->s_ttyvp == vp) { 969 error = 0; 970 rel_mplock(); 971 break; 972 } 973 974 /* Get rid of reference to old control tty */ 975 ovp = sess->s_ttyvp; 976 vref(vp); 977 sess->s_ttyvp = vp; 978 if (ovp) 979 vrele(ovp); 980 rel_mplock(); 981 } 982 break; 983 } 984 return (error); 985 } 986 987 /* 988 * Obtain the requested vnode lock 989 * 990 * LK_RETRY Automatically retry on timeout 991 * LK_FAILRECLAIM Fail if the vnode is being reclaimed 992 * 993 * Failures will occur if the vnode is undergoing recyclement, but not 994 * all callers expect that the function will fail so the caller must pass 995 * LK_FAILOK if it wants to process an error code. 996 * 997 * Errors can occur for other reasons if you pass in other LK_ flags, 998 * regardless of whether you pass in LK_FAILRECLAIM 999 */ 1000 int 1001 vn_lock(struct vnode *vp, int flags) 1002 { 1003 int error; 1004 1005 do { 1006 error = lockmgr(&vp->v_lock, flags); 1007 if (error == 0) 1008 break; 1009 } while (flags & LK_RETRY); 1010 1011 /* 1012 * Because we (had better!) have a ref on the vnode, once it 1013 * goes to VRECLAIMED state it will not be recycled until all 1014 * refs go away. So we can just check the flag. 1015 */ 1016 if (error == 0 && (vp->v_flag & VRECLAIMED)) { 1017 if (flags & LK_FAILRECLAIM) { 1018 lockmgr(&vp->v_lock, LK_RELEASE); 1019 error = ENOENT; 1020 } 1021 } 1022 return (error); 1023 } 1024 1025 #ifdef DEBUG_VN_UNLOCK 1026 1027 void 1028 debug_vn_unlock(struct vnode *vp, const char *filename, int line) 1029 { 1030 kprintf("vn_unlock from %s:%d\n", filename, line); 1031 lockmgr(&vp->v_lock, LK_RELEASE); 1032 } 1033 1034 #else 1035 1036 void 1037 vn_unlock(struct vnode *vp) 1038 { 1039 lockmgr(&vp->v_lock, LK_RELEASE); 1040 } 1041 1042 #endif 1043 1044 /* 1045 * MPSAFE 1046 */ 1047 int 1048 vn_islocked(struct vnode *vp) 1049 { 1050 return (lockstatus(&vp->v_lock, curthread)); 1051 } 1052 1053 /* 1054 * Return the lock status of a vnode and unlock the vnode 1055 * if we owned the lock. This is not a boolean, if the 1056 * caller cares what the lock status is the caller must 1057 * check the various possible values. 1058 * 1059 * This only unlocks exclusive locks held by the caller, 1060 * it will NOT unlock shared locks (there is no way to 1061 * tell who the shared lock belongs to). 1062 * 1063 * MPSAFE 1064 */ 1065 int 1066 vn_islocked_unlock(struct vnode *vp) 1067 { 1068 int vpls; 1069 1070 vpls = lockstatus(&vp->v_lock, curthread); 1071 if (vpls == LK_EXCLUSIVE) 1072 lockmgr(&vp->v_lock, LK_RELEASE); 1073 return(vpls); 1074 } 1075 1076 /* 1077 * Restore a vnode lock that we previously released via 1078 * vn_islocked_unlock(). This is a NOP if we did not 1079 * own the original lock. 1080 * 1081 * MPSAFE 1082 */ 1083 void 1084 vn_islocked_relock(struct vnode *vp, int vpls) 1085 { 1086 int error; 1087 1088 if (vpls == LK_EXCLUSIVE) 1089 error = lockmgr(&vp->v_lock, vpls); 1090 } 1091 1092 /* 1093 * MPSAFE 1094 */ 1095 static int 1096 vn_closefile(struct file *fp) 1097 { 1098 int error; 1099 1100 fp->f_ops = &badfileops; 1101 error = vn_close(((struct vnode *)fp->f_data), fp->f_flag, fp); 1102 return (error); 1103 } 1104 1105 /* 1106 * MPSAFE 1107 */ 1108 static int 1109 vn_kqfilter(struct file *fp, struct knote *kn) 1110 { 1111 int error; 1112 1113 error = VOP_KQFILTER(((struct vnode *)fp->f_data), kn); 1114 return (error); 1115 } 1116