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 * (error is 0 on entry / retry) 166 */ 167 again: 168 /* 169 * Checks for (likely) filesystem-modifying cases and allows 170 * the filesystem to stall the front-end. 171 */ 172 if ((fmode & (FWRITE | O_TRUNC)) || 173 ((fmode & O_CREAT) && nd->nl_nch.ncp->nc_vp == NULL)) { 174 error = ncp_writechk(&nd->nl_nch); 175 if (error) 176 return error; 177 } 178 179 if (fmode & O_CREAT) { 180 if (nd->nl_nch.ncp->nc_vp == NULL) { 181 VATTR_NULL(vap); 182 vap->va_type = VREG; 183 vap->va_mode = cmode; 184 vap->va_fuseflags = fmode; /* FUSE */ 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 if (nd->nl_flags & NLC_SHAREDLOCK) { 206 error = cache_vget(&nd->nl_nch, cred, LK_SHARED, &vp); 207 } else { 208 error = cache_vget(&nd->nl_nch, cred, 209 LK_EXCLUSIVE, &vp); 210 } 211 if (error) 212 return (error); 213 } 214 215 /* 216 * We have a locked vnode and ncp now. Note that the ncp will 217 * be cleaned up by the caller if nd->nl_nch is left intact. 218 */ 219 if (vp->v_type == VLNK) { 220 error = EMLINK; 221 goto bad; 222 } 223 if (vp->v_type == VSOCK) { 224 error = EOPNOTSUPP; 225 goto bad; 226 } 227 if (vp->v_type != VDIR && (fmode & O_DIRECTORY)) { 228 error = ENOTDIR; 229 goto bad; 230 } 231 if ((fmode & O_CREAT) == 0) { 232 if (fmode & (FWRITE | O_TRUNC)) { 233 if (vp->v_type == VDIR) { 234 error = EISDIR; 235 goto bad; 236 } 237 238 /* 239 * Additional checks on vnode (does not substitute 240 * for ncp_writechk()). 241 */ 242 error = vn_writechk(vp); 243 if (error) { 244 /* 245 * Special stale handling, re-resolve the 246 * vnode. 247 */ 248 if (error == ESTALE) { 249 vput(vp); 250 vp = NULL; 251 if (nd->nl_flags & NLC_SHAREDLOCK) { 252 cache_unlock(&nd->nl_nch); 253 cache_lock(&nd->nl_nch); 254 } 255 cache_setunresolved(&nd->nl_nch); 256 error = cache_resolve(&nd->nl_nch, 257 cred); 258 if (error == 0) 259 goto again; 260 } 261 goto bad; 262 } 263 } 264 } 265 if (fmode & O_TRUNC) { 266 vn_unlock(vp); /* XXX */ 267 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 268 osize = vp->v_filesize; 269 VATTR_NULL(vap); 270 vap->va_size = 0; 271 error = VOP_SETATTR_FP(vp, vap, cred, fp); 272 if (error) 273 goto bad; 274 error = VOP_GETATTR(vp, vap); 275 if (error) 276 goto bad; 277 mp = vq_vptomp(vp); 278 VFS_ACCOUNT(mp, vap->va_uid, vap->va_gid, -osize); 279 } 280 281 /* 282 * Set or clear VNSWAPCACHE on the vp based on nd->nl_nch.ncp->nc_flag. 283 * These particular bits a tracked all the way from the root. 284 * 285 * NOTE: Might not work properly on NFS servers due to the 286 * disconnected namecache. 287 */ 288 flags = nd->nl_nch.ncp->nc_flag; 289 if ((flags & (NCF_UF_CACHE | NCF_UF_PCACHE)) && 290 (flags & (NCF_SF_NOCACHE | NCF_SF_PNOCACHE)) == 0) { 291 vsetflags(vp, VSWAPCACHE); 292 } else { 293 vclrflags(vp, VSWAPCACHE); 294 } 295 296 /* 297 * Setup the fp so VOP_OPEN can override it. No descriptor has been 298 * associated with the fp yet so we own it clean. 299 * 300 * f_nchandle inherits nl_nch. This used to be necessary only for 301 * directories but now we do it unconditionally so f*() ops 302 * such as fchmod() can access the actual namespace that was 303 * used to open the file. 304 */ 305 if (fp) { 306 if (nd->nl_flags & NLC_APPENDONLY) 307 fmode |= FAPPENDONLY; 308 fp->f_nchandle = nd->nl_nch; 309 cache_zero(&nd->nl_nch); 310 cache_unlock(&fp->f_nchandle); 311 } 312 313 /* 314 * Get rid of nl_nch. vn_open does not return it (it returns the 315 * vnode or the file pointer). Note: we can't leave nl_nch locked 316 * through the VOP_OPEN anyway since the VOP_OPEN may block, e.g. 317 * on /dev/ttyd0 318 */ 319 if (nd->nl_nch.ncp) 320 cache_put(&nd->nl_nch); 321 322 error = VOP_OPEN(vp, fmode, cred, fp); 323 if (error) { 324 /* 325 * setting f_ops to &badfileops will prevent the descriptor 326 * code from trying to close and release the vnode, since 327 * the open failed we do not want to call close. 328 */ 329 if (fp) { 330 fp->f_data = NULL; 331 fp->f_ops = &badfileops; 332 } 333 goto bad; 334 } 335 336 #if 0 337 /* 338 * Assert that VREG files have been setup for vmio. 339 */ 340 KASSERT(vp->v_type != VREG || vp->v_object != NULL, 341 ("vn_open: regular file was not VMIO enabled!")); 342 #endif 343 344 /* 345 * Return the vnode. XXX needs some cleaning up. The vnode is 346 * only returned in the fp == NULL case. 347 */ 348 if (fp == NULL) { 349 nd->nl_open_vp = vp; 350 nd->nl_vp_fmode = fmode; 351 if ((nd->nl_flags & NLC_LOCKVP) == 0) 352 vn_unlock(vp); 353 } else { 354 vput(vp); 355 } 356 return (0); 357 bad: 358 if (vp) 359 vput(vp); 360 return (error); 361 } 362 363 int 364 vn_opendisk(const char *devname, int fmode, struct vnode **vpp) 365 { 366 struct vnode *vp; 367 int error; 368 369 if (strncmp(devname, "/dev/", 5) == 0) 370 devname += 5; 371 if ((vp = getsynthvnode(devname)) == NULL) { 372 error = ENODEV; 373 } else { 374 error = VOP_OPEN(vp, fmode, proc0.p_ucred, NULL); 375 vn_unlock(vp); 376 if (error) { 377 vrele(vp); 378 vp = NULL; 379 } 380 } 381 *vpp = vp; 382 return (error); 383 } 384 385 /* 386 * Checks for special conditions on the vnode which might prevent writing 387 * after the vnode has (likely) been locked. The vnode might or might not 388 * be locked as of this call, but will be at least referenced. 389 * 390 * Also re-checks the mount RDONLY flag that ncp_writechk() checked prior 391 * to the vnode being locked. 392 */ 393 int 394 vn_writechk(struct vnode *vp) 395 { 396 /* 397 * If there's shared text associated with 398 * the vnode, try to free it up once. If 399 * we fail, we can't allow writing. 400 */ 401 if (vp->v_flag & VTEXT) 402 return (ETXTBSY); 403 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_RDONLY)) 404 return (EROFS); 405 return 0; 406 } 407 408 /* 409 * Check whether the underlying mount is read-only. The mount point 410 * referenced by the namecache may be different from the mount point 411 * used by the underlying vnode in the case of NULLFS, so a separate 412 * check is needed. 413 * 414 * Must be called PRIOR to any vnodes being locked. 415 */ 416 int 417 ncp_writechk(struct nchandle *nch) 418 { 419 struct mount *mp; 420 421 if ((mp = nch->mount) != NULL) { 422 if (mp->mnt_flag & MNT_RDONLY) 423 return (EROFS); 424 if (mp->mnt_op->vfs_modifying != vfs_stdmodifying) 425 VFS_MODIFYING(mp); 426 } 427 return(0); 428 } 429 430 /* 431 * Vnode close call 432 * 433 * MPSAFE 434 */ 435 int 436 vn_close(struct vnode *vp, int flags, struct file *fp) 437 { 438 int error; 439 440 error = vn_lock(vp, LK_SHARED | LK_RETRY | LK_FAILRECLAIM); 441 if (error == 0) { 442 error = VOP_CLOSE(vp, flags, fp); 443 vn_unlock(vp); 444 } 445 vrele(vp); 446 return (error); 447 } 448 449 /* 450 * Sequential heuristic. 451 * 452 * MPSAFE (f_seqcount and f_nextoff are allowed to race) 453 */ 454 static __inline 455 int 456 sequential_heuristic(struct uio *uio, struct file *fp) 457 { 458 /* 459 * Sequential heuristic - detect sequential operation 460 * 461 * NOTE: SMP: We allow f_seqcount updates to race. 462 */ 463 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 464 uio->uio_offset == fp->f_nextoff) { 465 int tmpseq = fp->f_seqcount; 466 467 tmpseq += (uio->uio_resid + MAXBSIZE - 1) / MAXBSIZE; 468 if (tmpseq > IO_SEQMAX) 469 tmpseq = IO_SEQMAX; 470 fp->f_seqcount = tmpseq; 471 return(fp->f_seqcount << IO_SEQSHIFT); 472 } 473 474 /* 475 * Not sequential, quick draw-down of seqcount 476 * 477 * NOTE: SMP: We allow f_seqcount updates to race. 478 */ 479 if (fp->f_seqcount > 1) 480 fp->f_seqcount = 1; 481 else 482 fp->f_seqcount = 0; 483 return(0); 484 } 485 486 /* 487 * get - lock and return the f_offset field. 488 * set - set and unlock the f_offset field. 489 * 490 * These routines serve the dual purpose of serializing access to the 491 * f_offset field (at least on x86) and guaranteeing operational integrity 492 * when multiple read()ers and write()ers are present on the same fp. 493 * 494 * MPSAFE 495 */ 496 static __inline off_t 497 vn_get_fpf_offset(struct file *fp) 498 { 499 u_int flags; 500 u_int nflags; 501 502 /* 503 * Shortcut critical path. 504 */ 505 flags = fp->f_flag & ~FOFFSETLOCK; 506 if (atomic_cmpset_int(&fp->f_flag, flags, flags | FOFFSETLOCK)) 507 return(fp->f_offset); 508 509 /* 510 * The hard way 511 */ 512 for (;;) { 513 flags = fp->f_flag; 514 if (flags & FOFFSETLOCK) { 515 nflags = flags | FOFFSETWAKE; 516 tsleep_interlock(&fp->f_flag, 0); 517 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) 518 tsleep(&fp->f_flag, PINTERLOCKED, "fpoff", 0); 519 } else { 520 nflags = flags | FOFFSETLOCK; 521 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) 522 break; 523 } 524 } 525 return(fp->f_offset); 526 } 527 528 /* 529 * MPSAFE 530 */ 531 static __inline void 532 vn_set_fpf_offset(struct file *fp, off_t offset) 533 { 534 u_int flags; 535 u_int nflags; 536 537 /* 538 * We hold the lock so we can set the offset without interference. 539 */ 540 fp->f_offset = offset; 541 542 /* 543 * Normal release is already a reasonably critical path. 544 */ 545 for (;;) { 546 flags = fp->f_flag; 547 nflags = flags & ~(FOFFSETLOCK | FOFFSETWAKE); 548 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) { 549 if (flags & FOFFSETWAKE) 550 wakeup(&fp->f_flag); 551 break; 552 } 553 } 554 } 555 556 /* 557 * MPSAFE 558 */ 559 static __inline off_t 560 vn_poll_fpf_offset(struct file *fp) 561 { 562 #if defined(__x86_64__) 563 return(fp->f_offset); 564 #else 565 off_t off = vn_get_fpf_offset(fp); 566 vn_set_fpf_offset(fp, off); 567 return(off); 568 #endif 569 } 570 571 /* 572 * Package up an I/O request on a vnode into a uio and do it. 573 * 574 * MPSAFE 575 */ 576 int 577 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 578 off_t offset, enum uio_seg segflg, int ioflg, 579 struct ucred *cred, int *aresid) 580 { 581 struct uio auio; 582 struct iovec aiov; 583 int error; 584 585 if ((ioflg & IO_NODELOCKED) == 0) 586 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 587 auio.uio_iov = &aiov; 588 auio.uio_iovcnt = 1; 589 aiov.iov_base = base; 590 aiov.iov_len = len; 591 auio.uio_resid = len; 592 auio.uio_offset = offset; 593 auio.uio_segflg = segflg; 594 auio.uio_rw = rw; 595 auio.uio_td = curthread; 596 if (rw == UIO_READ) { 597 error = VOP_READ(vp, &auio, ioflg, cred); 598 } else { 599 error = VOP_WRITE(vp, &auio, ioflg, cred); 600 } 601 if (aresid) 602 *aresid = auio.uio_resid; 603 else 604 if (auio.uio_resid && error == 0) 605 error = EIO; 606 if ((ioflg & IO_NODELOCKED) == 0) 607 vn_unlock(vp); 608 return (error); 609 } 610 611 /* 612 * Package up an I/O request on a vnode into a uio and do it. The I/O 613 * request is split up into smaller chunks and we try to avoid saturating 614 * the buffer cache while potentially holding a vnode locked, so we 615 * check bwillwrite() before calling vn_rdwr(). We also call lwkt_user_yield() 616 * to give other processes a chance to lock the vnode (either other processes 617 * core'ing the same binary, or unrelated processes scanning the directory). 618 * 619 * MPSAFE 620 */ 621 int 622 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 623 off_t offset, enum uio_seg segflg, int ioflg, 624 struct ucred *cred, int *aresid) 625 { 626 int error = 0; 627 628 do { 629 int chunk; 630 631 /* 632 * Force `offset' to a multiple of MAXBSIZE except possibly 633 * for the first chunk, so that filesystems only need to 634 * write full blocks except possibly for the first and last 635 * chunks. 636 */ 637 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 638 639 if (chunk > len) 640 chunk = len; 641 if (vp->v_type == VREG && (ioflg & IO_RECURSE) == 0) { 642 switch(rw) { 643 case UIO_READ: 644 bwillread(chunk); 645 break; 646 case UIO_WRITE: 647 bwillwrite(chunk); 648 break; 649 } 650 } 651 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 652 ioflg, cred, aresid); 653 len -= chunk; /* aresid calc already includes length */ 654 if (error) 655 break; 656 offset += chunk; 657 base += chunk; 658 lwkt_user_yield(); 659 } while (len); 660 if (aresid) 661 *aresid += len; 662 return (error); 663 } 664 665 /* 666 * File pointers can no longer get ripped up by revoke so 667 * we don't need to lock access to the vp. 668 * 669 * f_offset updates are not guaranteed against multiple readers 670 */ 671 static int 672 vn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 673 { 674 struct vnode *vp; 675 int error, ioflag; 676 677 KASSERT(uio->uio_td == curthread, 678 ("uio_td %p is not td %p", uio->uio_td, curthread)); 679 vp = (struct vnode *)fp->f_data; 680 681 ioflag = 0; 682 if (flags & O_FBLOCKING) { 683 /* ioflag &= ~IO_NDELAY; */ 684 } else if (flags & O_FNONBLOCKING) { 685 ioflag |= IO_NDELAY; 686 } else if (fp->f_flag & FNONBLOCK) { 687 ioflag |= IO_NDELAY; 688 } 689 if (fp->f_flag & O_DIRECT) { 690 ioflag |= IO_DIRECT; 691 } 692 if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0) 693 uio->uio_offset = vn_get_fpf_offset(fp); 694 vn_lock(vp, LK_SHARED | LK_RETRY); 695 ioflag |= sequential_heuristic(uio, fp); 696 697 error = VOP_READ_FP(vp, uio, ioflag, cred, fp); 698 fp->f_nextoff = uio->uio_offset; 699 vn_unlock(vp); 700 if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0) 701 vn_set_fpf_offset(fp, uio->uio_offset); 702 return (error); 703 } 704 705 /* 706 * MPSAFE 707 */ 708 static int 709 vn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 710 { 711 struct vnode *vp; 712 int error, ioflag; 713 714 KASSERT(uio->uio_td == curthread, 715 ("uio_td %p is not p %p", uio->uio_td, curthread)); 716 vp = (struct vnode *)fp->f_data; 717 718 ioflag = IO_UNIT; 719 if (vp->v_type == VREG && 720 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 721 ioflag |= IO_APPEND; 722 } 723 724 if (flags & O_FBLOCKING) { 725 /* ioflag &= ~IO_NDELAY; */ 726 } else if (flags & O_FNONBLOCKING) { 727 ioflag |= IO_NDELAY; 728 } else if (fp->f_flag & FNONBLOCK) { 729 ioflag |= IO_NDELAY; 730 } 731 if (fp->f_flag & O_DIRECT) { 732 ioflag |= IO_DIRECT; 733 } 734 if (flags & O_FASYNCWRITE) { 735 /* ioflag &= ~IO_SYNC; */ 736 } else if (flags & O_FSYNCWRITE) { 737 ioflag |= IO_SYNC; 738 } else if (fp->f_flag & O_FSYNC) { 739 ioflag |= IO_SYNC; 740 } 741 742 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 743 ioflag |= IO_SYNC; 744 if ((flags & O_FOFFSET) == 0) 745 uio->uio_offset = vn_get_fpf_offset(fp); 746 if (vp->v_mount) 747 VFS_MODIFYING(vp->v_mount); 748 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 749 ioflag |= sequential_heuristic(uio, fp); 750 error = VOP_WRITE_FP(vp, uio, ioflag, cred, fp); 751 fp->f_nextoff = uio->uio_offset; 752 vn_unlock(vp); 753 if ((flags & O_FOFFSET) == 0) 754 vn_set_fpf_offset(fp, uio->uio_offset); 755 return (error); 756 } 757 758 /* 759 * MPSAFE 760 */ 761 static int 762 vn_statfile(struct file *fp, struct stat *sb, struct ucred *cred) 763 { 764 struct vnode *vp; 765 int error; 766 767 vp = (struct vnode *)fp->f_data; 768 error = vn_stat(vp, sb, cred); 769 return (error); 770 } 771 772 /* 773 * MPSAFE 774 */ 775 int 776 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *cred) 777 { 778 struct vattr vattr; 779 struct vattr *vap; 780 int error; 781 u_short mode; 782 cdev_t dev; 783 784 vap = &vattr; 785 error = VOP_GETATTR(vp, vap); 786 if (error) 787 return (error); 788 789 /* 790 * Zero the spare stat fields 791 */ 792 sb->st_lspare = 0; 793 sb->st_qspare2 = 0; 794 795 /* 796 * Copy from vattr table 797 */ 798 if (vap->va_fsid != VNOVAL) 799 sb->st_dev = vap->va_fsid; 800 else 801 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 802 sb->st_ino = vap->va_fileid; 803 mode = vap->va_mode; 804 switch (vap->va_type) { 805 case VREG: 806 mode |= S_IFREG; 807 break; 808 case VDATABASE: 809 mode |= S_IFDB; 810 break; 811 case VDIR: 812 mode |= S_IFDIR; 813 break; 814 case VBLK: 815 mode |= S_IFBLK; 816 break; 817 case VCHR: 818 mode |= S_IFCHR; 819 break; 820 case VLNK: 821 mode |= S_IFLNK; 822 /* This is a cosmetic change, symlinks do not have a mode. */ 823 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) 824 sb->st_mode &= ~ACCESSPERMS; /* 0000 */ 825 else 826 sb->st_mode |= ACCESSPERMS; /* 0777 */ 827 break; 828 case VSOCK: 829 mode |= S_IFSOCK; 830 break; 831 case VFIFO: 832 mode |= S_IFIFO; 833 break; 834 default: 835 return (EBADF); 836 } 837 sb->st_mode = mode; 838 if (vap->va_nlink > (nlink_t)-1) 839 sb->st_nlink = (nlink_t)-1; 840 else 841 sb->st_nlink = vap->va_nlink; 842 sb->st_uid = vap->va_uid; 843 sb->st_gid = vap->va_gid; 844 sb->st_rdev = devid_from_dev(vp->v_rdev); 845 sb->st_size = vap->va_size; 846 sb->st_atimespec = vap->va_atime; 847 sb->st_mtimespec = vap->va_mtime; 848 sb->st_ctimespec = vap->va_ctime; 849 850 /* 851 * A VCHR and VBLK device may track the last access and last modified 852 * time independantly of the filesystem. This is particularly true 853 * because device read and write calls may bypass the filesystem. 854 */ 855 if (vp->v_type == VCHR || vp->v_type == VBLK) { 856 dev = vp->v_rdev; 857 if (dev != NULL) { 858 if (dev->si_lastread) { 859 sb->st_atimespec.tv_sec = time_second + 860 (time_uptime - 861 dev->si_lastread); 862 sb->st_atimespec.tv_nsec = 0; 863 } 864 if (dev->si_lastwrite) { 865 sb->st_atimespec.tv_sec = time_second + 866 (time_uptime - 867 dev->si_lastwrite); 868 sb->st_atimespec.tv_nsec = 0; 869 } 870 } 871 } 872 873 /* 874 * According to www.opengroup.org, the meaning of st_blksize is 875 * "a filesystem-specific preferred I/O block size for this 876 * object. In some filesystem types, this may vary from file 877 * to file" 878 * Default to PAGE_SIZE after much discussion. 879 */ 880 881 if (vap->va_type == VREG) { 882 sb->st_blksize = vap->va_blocksize; 883 } else if (vn_isdisk(vp, NULL)) { 884 /* 885 * XXX this is broken. If the device is not yet open (aka 886 * stat() call, aka v_rdev == NULL), how are we supposed 887 * to get a valid block size out of it? 888 */ 889 dev = vp->v_rdev; 890 891 sb->st_blksize = dev->si_bsize_best; 892 if (sb->st_blksize < dev->si_bsize_phys) 893 sb->st_blksize = dev->si_bsize_phys; 894 if (sb->st_blksize < BLKDEV_IOSIZE) 895 sb->st_blksize = BLKDEV_IOSIZE; 896 } else { 897 sb->st_blksize = PAGE_SIZE; 898 } 899 900 sb->st_flags = vap->va_flags; 901 902 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0); 903 if (error) 904 sb->st_gen = 0; 905 else 906 sb->st_gen = (u_int32_t)vap->va_gen; 907 908 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 909 910 /* 911 * This is for ABI compatibility <= 5.7 (for ABI change made in 912 * 5.7 master). 913 */ 914 sb->__old_st_blksize = sb->st_blksize; 915 916 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 struct sysmsg *msg) 925 { 926 struct vnode *vp = ((struct vnode *)fp->f_data); 927 struct vnode *ovp; 928 struct vattr vattr; 929 int error; 930 off_t size; 931 932 switch (vp->v_type) { 933 case VREG: 934 case VDIR: 935 if (com == FIONREAD) { 936 error = VOP_GETATTR(vp, &vattr); 937 if (error) 938 break; 939 size = vattr.va_size; 940 if ((vp->v_flag & VNOTSEEKABLE) == 0) 941 size -= vn_poll_fpf_offset(fp); 942 if (size > 0x7FFFFFFF) 943 size = 0x7FFFFFFF; 944 *(int *)data = size; 945 error = 0; 946 break; 947 } 948 if (com == FIOASYNC) { /* XXX */ 949 error = 0; /* XXX */ 950 break; 951 } 952 /* fall into ... */ 953 default: 954 #if 0 955 return (ENOTTY); 956 #endif 957 case VFIFO: 958 case VCHR: 959 case VBLK: 960 if (com == FIODTYPE) { 961 if (vp->v_type != VCHR && vp->v_type != VBLK) { 962 error = ENOTTY; 963 break; 964 } 965 *(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK; 966 error = 0; 967 break; 968 } 969 error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred, msg); 970 if (error == 0 && com == TIOCSCTTY) { 971 struct proc *p = curthread->td_proc; 972 struct session *sess; 973 974 if (p == NULL) { 975 error = ENOTTY; 976 break; 977 } 978 979 get_mplock(); 980 sess = p->p_session; 981 /* Do nothing if reassigning same control tty */ 982 if (sess->s_ttyvp == vp) { 983 error = 0; 984 rel_mplock(); 985 break; 986 } 987 988 /* Get rid of reference to old control tty */ 989 ovp = sess->s_ttyvp; 990 vref(vp); 991 sess->s_ttyvp = vp; 992 if (ovp) 993 vrele(ovp); 994 rel_mplock(); 995 } 996 break; 997 } 998 return (error); 999 } 1000 1001 /* 1002 * Obtain the requested vnode lock 1003 * 1004 * LK_RETRY Automatically retry on timeout 1005 * LK_FAILRECLAIM Fail if the vnode is being reclaimed 1006 * 1007 * Failures will occur if the vnode is undergoing recyclement, but not 1008 * all callers expect that the function will fail so the caller must pass 1009 * LK_FAILOK if it wants to process an error code. 1010 * 1011 * Errors can occur for other reasons if you pass in other LK_ flags, 1012 * regardless of whether you pass in LK_FAILRECLAIM 1013 */ 1014 int 1015 vn_lock(struct vnode *vp, int flags) 1016 { 1017 int error; 1018 1019 do { 1020 error = lockmgr(&vp->v_lock, flags); 1021 if (error == 0) 1022 break; 1023 } while (flags & LK_RETRY); 1024 1025 /* 1026 * Because we (had better!) have a ref on the vnode, once it 1027 * goes to VRECLAIMED state it will not be recycled until all 1028 * refs go away. So we can just check the flag. 1029 */ 1030 if (error == 0 && (vp->v_flag & VRECLAIMED)) { 1031 if (flags & LK_FAILRECLAIM) { 1032 lockmgr(&vp->v_lock, LK_RELEASE); 1033 error = ENOENT; 1034 } 1035 } 1036 return (error); 1037 } 1038 1039 #ifdef DEBUG_VN_UNLOCK 1040 1041 void 1042 debug_vn_unlock(struct vnode *vp, const char *filename, int line) 1043 { 1044 kprintf("vn_unlock from %s:%d\n", filename, line); 1045 lockmgr(&vp->v_lock, LK_RELEASE); 1046 } 1047 1048 #else 1049 1050 void 1051 vn_unlock(struct vnode *vp) 1052 { 1053 lockmgr(&vp->v_lock, LK_RELEASE); 1054 } 1055 1056 #endif 1057 1058 /* 1059 * MPSAFE 1060 */ 1061 int 1062 vn_islocked(struct vnode *vp) 1063 { 1064 return (lockstatus(&vp->v_lock, curthread)); 1065 } 1066 1067 /* 1068 * Return the lock status of a vnode and unlock the vnode 1069 * if we owned the lock. This is not a boolean, if the 1070 * caller cares what the lock status is the caller must 1071 * check the various possible values. 1072 * 1073 * This only unlocks exclusive locks held by the caller, 1074 * it will NOT unlock shared locks (there is no way to 1075 * tell who the shared lock belongs to). 1076 * 1077 * MPSAFE 1078 */ 1079 int 1080 vn_islocked_unlock(struct vnode *vp) 1081 { 1082 int vpls; 1083 1084 vpls = lockstatus(&vp->v_lock, curthread); 1085 if (vpls == LK_EXCLUSIVE) 1086 lockmgr(&vp->v_lock, LK_RELEASE); 1087 return(vpls); 1088 } 1089 1090 /* 1091 * Restore a vnode lock that we previously released via 1092 * vn_islocked_unlock(). This is a NOP if we did not 1093 * own the original lock. 1094 * 1095 * MPSAFE 1096 */ 1097 void 1098 vn_islocked_relock(struct vnode *vp, int vpls) 1099 { 1100 int error; 1101 1102 if (vpls == LK_EXCLUSIVE) 1103 error = lockmgr(&vp->v_lock, vpls); 1104 } 1105 1106 /* 1107 * MPSAFE 1108 */ 1109 static int 1110 vn_closefile(struct file *fp) 1111 { 1112 int error; 1113 1114 fp->f_ops = &badfileops; 1115 error = vn_close(((struct vnode *)fp->f_data), fp->f_flag, fp); 1116 return (error); 1117 } 1118 1119 /* 1120 * MPSAFE 1121 */ 1122 static int 1123 vn_kqfilter(struct file *fp, struct knote *kn) 1124 { 1125 int error; 1126 1127 error = VOP_KQFILTER(((struct vnode *)fp->f_data), kn); 1128 return (error); 1129 } 1130