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 int vpexcl; 106 u_int flags; 107 uint64_t osize; 108 struct mount *mp; 109 110 /* 111 * Certain combinations are illegal 112 */ 113 if ((fmode & (FWRITE | O_TRUNC)) == O_TRUNC) 114 return(EACCES); 115 116 /* 117 * Lookup the path and create or obtain the vnode. After a 118 * successful lookup a locked nd->nl_nch will be returned. 119 * 120 * The result of this section should be a locked vnode. 121 * 122 * XXX with only a little work we should be able to avoid locking 123 * the vnode if FWRITE, O_CREAT, and O_TRUNC are *not* set. 124 */ 125 nd->nl_flags |= NLC_OPEN; 126 if (fmode & O_APPEND) 127 nd->nl_flags |= NLC_APPEND; 128 if (fmode & O_TRUNC) 129 nd->nl_flags |= NLC_TRUNCATE; 130 if (fmode & FREAD) 131 nd->nl_flags |= NLC_READ; 132 if (fmode & FWRITE) 133 nd->nl_flags |= NLC_WRITE; 134 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 135 nd->nl_flags |= NLC_FOLLOW; 136 137 if (fmode & O_CREAT) { 138 /* 139 * CONDITIONAL CREATE FILE CASE 140 * 141 * Setting NLC_CREATE causes a negative hit to store 142 * the negative hit ncp and not return an error. Then 143 * nc_error or nc_vp may be checked to see if the ncp 144 * represents a negative hit. NLC_CREATE also requires 145 * write permission on the governing directory or EPERM 146 * is returned. 147 */ 148 nd->nl_flags |= NLC_CREATE; 149 nd->nl_flags |= NLC_REFDVP; 150 bwillinode(1); 151 error = nlookup(nd); 152 } else { 153 /* 154 * NORMAL OPEN FILE CASE 155 */ 156 error = nlookup(nd); 157 } 158 159 if (error) 160 return (error); 161 162 /* 163 * split case to allow us to re-resolve and retry the ncp in case 164 * we get ESTALE. 165 * 166 * (error is 0 on entry / retry) 167 */ 168 again: 169 /* 170 * Checks for (likely) filesystem-modifying cases and allows 171 * the filesystem to stall the front-end. 172 */ 173 if ((fmode & (FWRITE | O_TRUNC)) || 174 ((fmode & O_CREAT) && nd->nl_nch.ncp->nc_vp == NULL)) { 175 error = ncp_writechk(&nd->nl_nch); 176 if (error) 177 return error; 178 } 179 180 vpexcl = 1; 181 if (fmode & O_CREAT) { 182 if (nd->nl_nch.ncp->nc_vp == NULL) { 183 VATTR_NULL(vap); 184 vap->va_type = VREG; 185 vap->va_mode = cmode; 186 vap->va_fuseflags = fmode; /* FUSE */ 187 if (fmode & O_EXCL) 188 vap->va_vaflags |= VA_EXCLUSIVE; 189 error = VOP_NCREATE(&nd->nl_nch, nd->nl_dvp, &vp, 190 nd->nl_cred, vap); 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 /* 208 * In most other cases a shared lock on the vnode is 209 * sufficient. However, the O_RDWR case needs an 210 * exclusive lock if the vnode is executable. The 211 * NLC_EXCLLOCK_IFEXEC and NCF_NOTX flags help resolve 212 * this. 213 * 214 * NOTE: If NCF_NOTX is not set, we do not know the 215 * the state of the 'x' bits and have to get 216 * an exclusive lock for the EXCLLOCK_IFEXEC case. 217 */ 218 if ((nd->nl_flags & NLC_SHAREDLOCK) && 219 ((nd->nl_flags & NLC_EXCLLOCK_IFEXEC) == 0 || 220 nd->nl_nch.ncp->nc_flag & NCF_NOTX)) { 221 error = cache_vget(&nd->nl_nch, cred, LK_SHARED, &vp); 222 vpexcl = 0; 223 } else { 224 error = cache_vget(&nd->nl_nch, cred, 225 LK_EXCLUSIVE, &vp); 226 } 227 if (error) 228 return (error); 229 } 230 231 /* 232 * We have a locked vnode and ncp now. Note that the ncp will 233 * be cleaned up by the caller if nd->nl_nch is left intact. 234 */ 235 if (vp->v_type == VLNK) { 236 error = EMLINK; 237 goto bad; 238 } 239 if (vp->v_type == VSOCK) { 240 error = EOPNOTSUPP; 241 goto bad; 242 } 243 if (vp->v_type != VDIR && (fmode & O_DIRECTORY)) { 244 error = ENOTDIR; 245 goto bad; 246 } 247 if ((fmode & O_CREAT) == 0) { 248 if (fmode & (FWRITE | O_TRUNC)) { 249 if (vp->v_type == VDIR) { 250 error = EISDIR; 251 goto bad; 252 } 253 254 /* 255 * Additional checks on vnode (does not substitute 256 * for ncp_writechk()). 257 */ 258 error = vn_writechk(vp); 259 if (error) { 260 /* 261 * Special stale handling, re-resolve the 262 * vnode. 263 */ 264 if (error == ESTALE) { 265 vput(vp); 266 vp = NULL; 267 if (vpexcl == 0) { 268 cache_unlock(&nd->nl_nch); 269 cache_lock(&nd->nl_nch); 270 } 271 cache_setunresolved(&nd->nl_nch); 272 error = cache_resolve(&nd->nl_nch, 273 cred); 274 if (error == 0) 275 goto again; 276 } 277 goto bad; 278 } 279 } 280 } 281 if (fmode & O_TRUNC) { 282 vn_unlock(vp); /* XXX */ 283 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 284 osize = vp->v_filesize; 285 VATTR_NULL(vap); 286 vap->va_size = 0; 287 error = VOP_SETATTR_FP(vp, vap, cred, fp); 288 if (error) 289 goto bad; 290 error = VOP_GETATTR(vp, vap); 291 if (error) 292 goto bad; 293 mp = vq_vptomp(vp); 294 VFS_ACCOUNT(mp, vap->va_uid, vap->va_gid, -osize); 295 } 296 297 /* 298 * Set or clear VNSWAPCACHE on the vp based on nd->nl_nch.ncp->nc_flag. 299 * These particular bits a tracked all the way from the root. 300 * 301 * NOTE: Might not work properly on NFS servers due to the 302 * disconnected namecache. 303 */ 304 flags = nd->nl_nch.ncp->nc_flag; 305 if ((flags & (NCF_UF_CACHE | NCF_UF_PCACHE)) && 306 (flags & (NCF_SF_NOCACHE | NCF_SF_PNOCACHE)) == 0) { 307 vsetflags(vp, VSWAPCACHE); 308 } else { 309 vclrflags(vp, VSWAPCACHE); 310 } 311 312 /* 313 * Setup the fp so VOP_OPEN can override it. No descriptor has been 314 * associated with the fp yet so we own it clean. 315 * 316 * f_nchandle inherits nl_nch. This used to be necessary only for 317 * directories but now we do it unconditionally so f*() ops 318 * such as fchmod() can access the actual namespace that was 319 * used to open the file. 320 */ 321 if (fp) { 322 if (nd->nl_flags & NLC_APPENDONLY) 323 fmode |= FAPPENDONLY; 324 fp->f_nchandle = nd->nl_nch; 325 cache_zero(&nd->nl_nch); 326 cache_unlock(&fp->f_nchandle); 327 } 328 329 /* 330 * Get rid of nl_nch. vn_open does not return it (it returns the 331 * vnode or the file pointer). Note: we can't leave nl_nch locked 332 * through the VOP_OPEN anyway since the VOP_OPEN may block, e.g. 333 * on /dev/ttyd0 334 */ 335 if (nd->nl_nch.ncp) 336 cache_put(&nd->nl_nch); 337 338 error = VOP_OPEN(vp, fmode, cred, fp); 339 if (error) { 340 /* 341 * setting f_ops to &badfileops will prevent the descriptor 342 * code from trying to close and release the vnode, since 343 * the open failed we do not want to call close. 344 */ 345 if (fp) { 346 fp->f_data = NULL; 347 fp->f_ops = &badfileops; 348 } 349 goto bad; 350 } 351 352 #if 0 353 /* 354 * Assert that VREG files have been setup for vmio. 355 */ 356 KASSERT(vp->v_type != VREG || vp->v_object != NULL, 357 ("vn_open: regular file was not VMIO enabled!")); 358 #endif 359 360 /* 361 * Return the vnode. XXX needs some cleaning up. The vnode is 362 * only returned in the fp == NULL case. 363 */ 364 if (fp == NULL) { 365 nd->nl_open_vp = vp; 366 nd->nl_vp_fmode = fmode; 367 if ((nd->nl_flags & NLC_LOCKVP) == 0) 368 vn_unlock(vp); 369 } else { 370 vput(vp); 371 } 372 return (0); 373 bad: 374 if (vp) 375 vput(vp); 376 return (error); 377 } 378 379 int 380 vn_opendisk(const char *devname, int fmode, struct vnode **vpp) 381 { 382 struct vnode *vp; 383 int error; 384 385 if (strncmp(devname, "/dev/", 5) == 0) 386 devname += 5; 387 if ((vp = getsynthvnode(devname)) == NULL) { 388 error = ENODEV; 389 } else { 390 error = VOP_OPEN(vp, fmode, proc0.p_ucred, NULL); 391 vn_unlock(vp); 392 if (error) { 393 vrele(vp); 394 vp = NULL; 395 } 396 } 397 *vpp = vp; 398 return (error); 399 } 400 401 /* 402 * Checks for special conditions on the vnode which might prevent writing 403 * after the vnode has (likely) been locked. The vnode might or might not 404 * be locked as of this call, but will be at least referenced. 405 * 406 * Also re-checks the mount RDONLY flag that ncp_writechk() checked prior 407 * to the vnode being locked. 408 */ 409 int 410 vn_writechk(struct vnode *vp) 411 { 412 /* 413 * If there's shared text associated with 414 * the vnode, try to free it up once. If 415 * we fail, we can't allow writing. 416 */ 417 if (vp->v_flag & VTEXT) 418 return (ETXTBSY); 419 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_RDONLY)) 420 return (EROFS); 421 return 0; 422 } 423 424 /* 425 * Check whether the underlying mount is read-only. The mount point 426 * referenced by the namecache may be different from the mount point 427 * used by the underlying vnode in the case of NULLFS, so a separate 428 * check is needed. 429 * 430 * Must be called PRIOR to any vnodes being locked. 431 */ 432 int 433 ncp_writechk(struct nchandle *nch) 434 { 435 struct mount *mp; 436 437 if ((mp = nch->mount) != NULL) { 438 if (mp->mnt_flag & MNT_RDONLY) 439 return (EROFS); 440 if (mp->mnt_op->vfs_modifying != vfs_stdmodifying) 441 VFS_MODIFYING(mp); 442 } 443 return(0); 444 } 445 446 /* 447 * Vnode close call 448 * 449 * MPSAFE 450 */ 451 int 452 vn_close(struct vnode *vp, int flags, struct file *fp) 453 { 454 int error; 455 456 error = vn_lock(vp, LK_SHARED | LK_RETRY | LK_FAILRECLAIM); 457 if (error == 0) { 458 error = VOP_CLOSE(vp, flags, fp); 459 vn_unlock(vp); 460 } 461 vrele(vp); 462 return (error); 463 } 464 465 /* 466 * Sequential heuristic. 467 * 468 * MPSAFE (f_seqcount and f_nextoff are allowed to race) 469 */ 470 static __inline 471 int 472 sequential_heuristic(struct uio *uio, struct file *fp) 473 { 474 /* 475 * Sequential heuristic - detect sequential operation 476 * 477 * NOTE: SMP: We allow f_seqcount updates to race. 478 */ 479 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 480 uio->uio_offset == fp->f_nextoff) { 481 int tmpseq = fp->f_seqcount; 482 483 tmpseq += (uio->uio_resid + MAXBSIZE - 1) / MAXBSIZE; 484 if (tmpseq > IO_SEQMAX) 485 tmpseq = IO_SEQMAX; 486 fp->f_seqcount = tmpseq; 487 return(fp->f_seqcount << IO_SEQSHIFT); 488 } 489 490 /* 491 * Not sequential, quick draw-down of seqcount 492 * 493 * NOTE: SMP: We allow f_seqcount updates to race. 494 */ 495 if (fp->f_seqcount > 1) 496 fp->f_seqcount = 1; 497 else 498 fp->f_seqcount = 0; 499 return(0); 500 } 501 502 /* 503 * get - lock and return the f_offset field. 504 * set - set and unlock the f_offset field. 505 * 506 * These routines serve the dual purpose of serializing access to the 507 * f_offset field (at least on x86) and guaranteeing operational integrity 508 * when multiple read()ers and write()ers are present on the same fp. 509 * 510 * MPSAFE 511 */ 512 static __inline off_t 513 vn_get_fpf_offset(struct file *fp) 514 { 515 u_int flags; 516 u_int nflags; 517 518 /* 519 * Shortcut critical path. 520 */ 521 flags = fp->f_flag & ~FOFFSETLOCK; 522 if (atomic_cmpset_int(&fp->f_flag, flags, flags | FOFFSETLOCK)) 523 return(fp->f_offset); 524 525 /* 526 * The hard way 527 */ 528 for (;;) { 529 flags = fp->f_flag; 530 if (flags & FOFFSETLOCK) { 531 nflags = flags | FOFFSETWAKE; 532 tsleep_interlock(&fp->f_flag, 0); 533 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) 534 tsleep(&fp->f_flag, PINTERLOCKED, "fpoff", 0); 535 } else { 536 nflags = flags | FOFFSETLOCK; 537 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) 538 break; 539 } 540 } 541 return(fp->f_offset); 542 } 543 544 /* 545 * MPSAFE 546 */ 547 static __inline void 548 vn_set_fpf_offset(struct file *fp, off_t offset) 549 { 550 u_int flags; 551 u_int nflags; 552 553 /* 554 * We hold the lock so we can set the offset without interference. 555 */ 556 fp->f_offset = offset; 557 558 /* 559 * Normal release is already a reasonably critical path. 560 */ 561 for (;;) { 562 flags = fp->f_flag; 563 nflags = flags & ~(FOFFSETLOCK | FOFFSETWAKE); 564 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) { 565 if (flags & FOFFSETWAKE) 566 wakeup(&fp->f_flag); 567 break; 568 } 569 } 570 } 571 572 /* 573 * MPSAFE 574 */ 575 static __inline off_t 576 vn_poll_fpf_offset(struct file *fp) 577 { 578 #if defined(__x86_64__) 579 return(fp->f_offset); 580 #else 581 off_t off = vn_get_fpf_offset(fp); 582 vn_set_fpf_offset(fp, off); 583 return(off); 584 #endif 585 } 586 587 /* 588 * Package up an I/O request on a vnode into a uio and do it. 589 * 590 * MPSAFE 591 */ 592 int 593 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 594 off_t offset, enum uio_seg segflg, int ioflg, 595 struct ucred *cred, int *aresid) 596 { 597 struct uio auio; 598 struct iovec aiov; 599 int error; 600 601 if ((ioflg & IO_NODELOCKED) == 0) 602 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 603 auio.uio_iov = &aiov; 604 auio.uio_iovcnt = 1; 605 aiov.iov_base = base; 606 aiov.iov_len = len; 607 auio.uio_resid = len; 608 auio.uio_offset = offset; 609 auio.uio_segflg = segflg; 610 auio.uio_rw = rw; 611 auio.uio_td = curthread; 612 if (rw == UIO_READ) { 613 error = VOP_READ(vp, &auio, ioflg, cred); 614 } else { 615 error = VOP_WRITE(vp, &auio, ioflg, cred); 616 } 617 if (aresid) 618 *aresid = auio.uio_resid; 619 else 620 if (auio.uio_resid && error == 0) 621 error = EIO; 622 if ((ioflg & IO_NODELOCKED) == 0) 623 vn_unlock(vp); 624 return (error); 625 } 626 627 /* 628 * Package up an I/O request on a vnode into a uio and do it. The I/O 629 * request is split up into smaller chunks and we try to avoid saturating 630 * the buffer cache while potentially holding a vnode locked, so we 631 * check bwillwrite() before calling vn_rdwr(). We also call lwkt_user_yield() 632 * to give other processes a chance to lock the vnode (either other processes 633 * core'ing the same binary, or unrelated processes scanning the directory). 634 * 635 * MPSAFE 636 */ 637 int 638 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 639 off_t offset, enum uio_seg segflg, int ioflg, 640 struct ucred *cred, int *aresid) 641 { 642 int error = 0; 643 644 do { 645 int chunk; 646 647 /* 648 * Force `offset' to a multiple of MAXBSIZE except possibly 649 * for the first chunk, so that filesystems only need to 650 * write full blocks except possibly for the first and last 651 * chunks. 652 */ 653 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 654 655 if (chunk > len) 656 chunk = len; 657 if (vp->v_type == VREG && (ioflg & IO_RECURSE) == 0) { 658 switch(rw) { 659 case UIO_READ: 660 bwillread(chunk); 661 break; 662 case UIO_WRITE: 663 bwillwrite(chunk); 664 break; 665 } 666 } 667 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 668 ioflg, cred, aresid); 669 len -= chunk; /* aresid calc already includes length */ 670 if (error) 671 break; 672 offset += chunk; 673 base += chunk; 674 lwkt_user_yield(); 675 } while (len); 676 if (aresid) 677 *aresid += len; 678 return (error); 679 } 680 681 /* 682 * File pointers can no longer get ripped up by revoke so 683 * we don't need to lock access to the vp. 684 * 685 * f_offset updates are not guaranteed against multiple readers 686 */ 687 static int 688 vn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 689 { 690 struct vnode *vp; 691 int error, ioflag; 692 693 KASSERT(uio->uio_td == curthread, 694 ("uio_td %p is not td %p", uio->uio_td, curthread)); 695 vp = (struct vnode *)fp->f_data; 696 697 ioflag = 0; 698 if (flags & O_FBLOCKING) { 699 /* ioflag &= ~IO_NDELAY; */ 700 } else if (flags & O_FNONBLOCKING) { 701 ioflag |= IO_NDELAY; 702 } else if (fp->f_flag & FNONBLOCK) { 703 ioflag |= IO_NDELAY; 704 } 705 if (fp->f_flag & O_DIRECT) { 706 ioflag |= IO_DIRECT; 707 } 708 if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0) 709 uio->uio_offset = vn_get_fpf_offset(fp); 710 vn_lock(vp, LK_SHARED | LK_RETRY); 711 ioflag |= sequential_heuristic(uio, fp); 712 713 error = VOP_READ_FP(vp, uio, ioflag, cred, fp); 714 fp->f_nextoff = uio->uio_offset; 715 vn_unlock(vp); 716 if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0) 717 vn_set_fpf_offset(fp, uio->uio_offset); 718 return (error); 719 } 720 721 /* 722 * MPSAFE 723 */ 724 static int 725 vn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 726 { 727 struct vnode *vp; 728 int error, ioflag; 729 730 KASSERT(uio->uio_td == curthread, 731 ("uio_td %p is not p %p", uio->uio_td, curthread)); 732 vp = (struct vnode *)fp->f_data; 733 734 ioflag = IO_UNIT; 735 if (vp->v_type == VREG && 736 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 737 ioflag |= IO_APPEND; 738 } 739 740 if (flags & O_FBLOCKING) { 741 /* ioflag &= ~IO_NDELAY; */ 742 } else if (flags & O_FNONBLOCKING) { 743 ioflag |= IO_NDELAY; 744 } else if (fp->f_flag & FNONBLOCK) { 745 ioflag |= IO_NDELAY; 746 } 747 if (fp->f_flag & O_DIRECT) { 748 ioflag |= IO_DIRECT; 749 } 750 if (flags & O_FASYNCWRITE) { 751 /* ioflag &= ~IO_SYNC; */ 752 } else if (flags & O_FSYNCWRITE) { 753 ioflag |= IO_SYNC; 754 } else if (fp->f_flag & O_FSYNC) { 755 ioflag |= IO_SYNC; 756 } 757 758 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 759 ioflag |= IO_SYNC; 760 if ((flags & O_FOFFSET) == 0) 761 uio->uio_offset = vn_get_fpf_offset(fp); 762 if (vp->v_mount) 763 VFS_MODIFYING(vp->v_mount); 764 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 765 ioflag |= sequential_heuristic(uio, fp); 766 error = VOP_WRITE_FP(vp, uio, ioflag, cred, fp); 767 fp->f_nextoff = uio->uio_offset; 768 vn_unlock(vp); 769 if ((flags & O_FOFFSET) == 0) 770 vn_set_fpf_offset(fp, uio->uio_offset); 771 return (error); 772 } 773 774 /* 775 * MPSAFE 776 */ 777 static int 778 vn_statfile(struct file *fp, struct stat *sb, struct ucred *cred) 779 { 780 struct vnode *vp; 781 int error; 782 783 vp = (struct vnode *)fp->f_data; 784 error = vn_stat(vp, sb, cred); 785 return (error); 786 } 787 788 /* 789 * MPSAFE 790 */ 791 int 792 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *cred) 793 { 794 struct vattr vattr; 795 struct vattr *vap; 796 int error; 797 u_short mode; 798 cdev_t dev; 799 800 /* 801 * vp already has a ref and is validated, can call unlocked. 802 */ 803 vap = &vattr; 804 error = VOP_GETATTR(vp, vap); 805 if (error) 806 return (error); 807 808 /* 809 * Zero the spare stat fields 810 */ 811 sb->st_lspare = 0; 812 sb->st_qspare2 = 0; 813 814 /* 815 * Copy from vattr table 816 */ 817 if (vap->va_fsid != VNOVAL) 818 sb->st_dev = vap->va_fsid; 819 else 820 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 821 sb->st_ino = vap->va_fileid; 822 mode = vap->va_mode; 823 switch (vap->va_type) { 824 case VREG: 825 mode |= S_IFREG; 826 break; 827 case VDATABASE: 828 mode |= S_IFDB; 829 break; 830 case VDIR: 831 mode |= S_IFDIR; 832 break; 833 case VBLK: 834 mode |= S_IFBLK; 835 break; 836 case VCHR: 837 mode |= S_IFCHR; 838 break; 839 case VLNK: 840 mode |= S_IFLNK; 841 /* This is a cosmetic change, symlinks do not have a mode. */ 842 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) 843 sb->st_mode &= ~ACCESSPERMS; /* 0000 */ 844 else 845 sb->st_mode |= ACCESSPERMS; /* 0777 */ 846 break; 847 case VSOCK: 848 mode |= S_IFSOCK; 849 break; 850 case VFIFO: 851 mode |= S_IFIFO; 852 break; 853 default: 854 return (EBADF); 855 } 856 sb->st_mode = mode; 857 if (vap->va_nlink > (nlink_t)-1) 858 sb->st_nlink = (nlink_t)-1; 859 else 860 sb->st_nlink = vap->va_nlink; 861 sb->st_uid = vap->va_uid; 862 sb->st_gid = vap->va_gid; 863 sb->st_rdev = devid_from_dev(vp->v_rdev); 864 sb->st_size = vap->va_size; 865 sb->st_atimespec = vap->va_atime; 866 sb->st_mtimespec = vap->va_mtime; 867 sb->st_ctimespec = vap->va_ctime; 868 869 /* 870 * A VCHR and VBLK device may track the last access and last modified 871 * time independantly of the filesystem. This is particularly true 872 * because device read and write calls may bypass the filesystem. 873 */ 874 if (vp->v_type == VCHR || vp->v_type == VBLK) { 875 dev = vp->v_rdev; 876 if (dev != NULL) { 877 if (dev->si_lastread) { 878 sb->st_atimespec.tv_sec = time_second + 879 (time_uptime - 880 dev->si_lastread); 881 sb->st_atimespec.tv_nsec = 0; 882 } 883 if (dev->si_lastwrite) { 884 sb->st_atimespec.tv_sec = time_second + 885 (time_uptime - 886 dev->si_lastwrite); 887 sb->st_atimespec.tv_nsec = 0; 888 } 889 } 890 } 891 892 /* 893 * According to www.opengroup.org, the meaning of st_blksize is 894 * "a filesystem-specific preferred I/O block size for this 895 * object. In some filesystem types, this may vary from file 896 * to file" 897 * Default to PAGE_SIZE after much discussion. 898 */ 899 900 if (vap->va_type == VREG) { 901 sb->st_blksize = vap->va_blocksize; 902 } else if (vn_isdisk(vp, NULL)) { 903 /* 904 * XXX this is broken. If the device is not yet open (aka 905 * stat() call, aka v_rdev == NULL), how are we supposed 906 * to get a valid block size out of it? 907 */ 908 dev = vp->v_rdev; 909 910 sb->st_blksize = dev->si_bsize_best; 911 if (sb->st_blksize < dev->si_bsize_phys) 912 sb->st_blksize = dev->si_bsize_phys; 913 if (sb->st_blksize < BLKDEV_IOSIZE) 914 sb->st_blksize = BLKDEV_IOSIZE; 915 } else { 916 sb->st_blksize = PAGE_SIZE; 917 } 918 919 sb->st_flags = vap->va_flags; 920 921 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0); 922 if (error) 923 sb->st_gen = 0; 924 else 925 sb->st_gen = (u_int32_t)vap->va_gen; 926 927 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 928 929 /* 930 * This is for ABI compatibility <= 5.7 (for ABI change made in 931 * 5.7 master). 932 */ 933 sb->__old_st_blksize = sb->st_blksize; 934 935 return (0); 936 } 937 938 /* 939 * MPALMOSTSAFE - acquires mplock 940 */ 941 static int 942 vn_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *ucred, 943 struct sysmsg *msg) 944 { 945 struct vnode *vp = ((struct vnode *)fp->f_data); 946 struct vnode *ovp; 947 struct vattr vattr; 948 int error; 949 off_t size; 950 951 switch (vp->v_type) { 952 case VREG: 953 case VDIR: 954 if (com == FIONREAD) { 955 error = VOP_GETATTR(vp, &vattr); 956 if (error) 957 break; 958 size = vattr.va_size; 959 if ((vp->v_flag & VNOTSEEKABLE) == 0) 960 size -= vn_poll_fpf_offset(fp); 961 if (size > 0x7FFFFFFF) 962 size = 0x7FFFFFFF; 963 *(int *)data = size; 964 error = 0; 965 break; 966 } 967 if (com == FIOASYNC) { /* XXX */ 968 error = 0; /* XXX */ 969 break; 970 } 971 /* fall into ... */ 972 default: 973 #if 0 974 return (ENOTTY); 975 #endif 976 case VFIFO: 977 case VCHR: 978 case VBLK: 979 if (com == FIODTYPE) { 980 if (vp->v_type != VCHR && vp->v_type != VBLK) { 981 error = ENOTTY; 982 break; 983 } 984 *(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK; 985 error = 0; 986 break; 987 } 988 error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred, msg); 989 if (error == 0 && com == TIOCSCTTY) { 990 struct proc *p = curthread->td_proc; 991 struct session *sess; 992 993 if (p == NULL) { 994 error = ENOTTY; 995 break; 996 } 997 998 get_mplock(); 999 sess = p->p_session; 1000 /* Do nothing if reassigning same control tty */ 1001 if (sess->s_ttyvp == vp) { 1002 error = 0; 1003 rel_mplock(); 1004 break; 1005 } 1006 1007 /* Get rid of reference to old control tty */ 1008 ovp = sess->s_ttyvp; 1009 vref(vp); 1010 sess->s_ttyvp = vp; 1011 if (ovp) 1012 vrele(ovp); 1013 rel_mplock(); 1014 } 1015 break; 1016 } 1017 return (error); 1018 } 1019 1020 /* 1021 * Obtain the requested vnode lock 1022 * 1023 * LK_RETRY Automatically retry on timeout 1024 * LK_FAILRECLAIM Fail if the vnode is being reclaimed 1025 * 1026 * Failures will occur if the vnode is undergoing recyclement, but not 1027 * all callers expect that the function will fail so the caller must pass 1028 * LK_FAILOK if it wants to process an error code. 1029 * 1030 * Errors can occur for other reasons if you pass in other LK_ flags, 1031 * regardless of whether you pass in LK_FAILRECLAIM 1032 */ 1033 int 1034 vn_lock(struct vnode *vp, int flags) 1035 { 1036 int error; 1037 1038 do { 1039 error = lockmgr(&vp->v_lock, flags); 1040 if (error == 0) 1041 break; 1042 } while (flags & LK_RETRY); 1043 1044 /* 1045 * Because we (had better!) have a ref on the vnode, once it 1046 * goes to VRECLAIMED state it will not be recycled until all 1047 * refs go away. So we can just check the flag. 1048 */ 1049 if (error == 0 && (vp->v_flag & VRECLAIMED)) { 1050 if (flags & LK_FAILRECLAIM) { 1051 lockmgr(&vp->v_lock, LK_RELEASE); 1052 error = ENOENT; 1053 } 1054 } 1055 return (error); 1056 } 1057 1058 int 1059 vn_relock(struct vnode *vp, int flags) 1060 { 1061 int error; 1062 1063 do { 1064 error = lockmgr(&vp->v_lock, flags); 1065 if (error == 0) 1066 break; 1067 } while (flags & LK_RETRY); 1068 1069 return error; 1070 } 1071 1072 #ifdef DEBUG_VN_UNLOCK 1073 1074 void 1075 debug_vn_unlock(struct vnode *vp, const char *filename, int line) 1076 { 1077 kprintf("vn_unlock from %s:%d\n", filename, line); 1078 lockmgr(&vp->v_lock, LK_RELEASE); 1079 } 1080 1081 #else 1082 1083 void 1084 vn_unlock(struct vnode *vp) 1085 { 1086 lockmgr(&vp->v_lock, LK_RELEASE); 1087 } 1088 1089 #endif 1090 1091 /* 1092 * MPSAFE 1093 */ 1094 int 1095 vn_islocked(struct vnode *vp) 1096 { 1097 return (lockstatus(&vp->v_lock, curthread)); 1098 } 1099 1100 /* 1101 * Return the lock status of a vnode and unlock the vnode 1102 * if we owned the lock. This is not a boolean, if the 1103 * caller cares what the lock status is the caller must 1104 * check the various possible values. 1105 * 1106 * This only unlocks exclusive locks held by the caller, 1107 * it will NOT unlock shared locks (there is no way to 1108 * tell who the shared lock belongs to). 1109 * 1110 * MPSAFE 1111 */ 1112 int 1113 vn_islocked_unlock(struct vnode *vp) 1114 { 1115 int vpls; 1116 1117 vpls = lockstatus(&vp->v_lock, curthread); 1118 if (vpls == LK_EXCLUSIVE) 1119 lockmgr(&vp->v_lock, LK_RELEASE); 1120 return(vpls); 1121 } 1122 1123 /* 1124 * Restore a vnode lock that we previously released via 1125 * vn_islocked_unlock(). This is a NOP if we did not 1126 * own the original lock. 1127 * 1128 * MPSAFE 1129 */ 1130 void 1131 vn_islocked_relock(struct vnode *vp, int vpls) 1132 { 1133 int error; 1134 1135 if (vpls == LK_EXCLUSIVE) 1136 error = lockmgr(&vp->v_lock, vpls); 1137 } 1138 1139 /* 1140 * MPSAFE 1141 */ 1142 static int 1143 vn_closefile(struct file *fp) 1144 { 1145 int error; 1146 1147 fp->f_ops = &badfileops; 1148 error = vn_close(((struct vnode *)fp->f_data), fp->f_flag, fp); 1149 return (error); 1150 } 1151 1152 /* 1153 * MPSAFE 1154 */ 1155 static int 1156 vn_kqfilter(struct file *fp, struct knote *kn) 1157 { 1158 int error; 1159 1160 error = VOP_KQFILTER(((struct vnode *)fp->f_data), kn); 1161 return (error); 1162 } 1163