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 **fpp, int fmode, int cmode) 99 { 100 struct file *fp = fpp ? *fpp : NULL; 101 struct vnode *vp; 102 struct ucred *cred = nd->nl_cred; 103 struct vattr vat; 104 struct vattr *vap = &vat; 105 int error; 106 int vpexcl; 107 u_int flags; 108 uint64_t osize; 109 struct mount *mp; 110 111 /* 112 * Certain combinations are illegal 113 */ 114 if ((fmode & (FWRITE | O_TRUNC)) == O_TRUNC) 115 return(EACCES); 116 117 /* 118 * Lookup the path and create or obtain the vnode. After a 119 * successful lookup a locked nd->nl_nch will be returned. 120 * 121 * The result of this section should be a locked vnode. 122 * 123 * XXX with only a little work we should be able to avoid locking 124 * the vnode if FWRITE, O_CREAT, and O_TRUNC are *not* set. 125 */ 126 nd->nl_flags |= NLC_OPEN; 127 if (fmode & O_APPEND) 128 nd->nl_flags |= NLC_APPEND; 129 if (fmode & O_TRUNC) 130 nd->nl_flags |= NLC_TRUNCATE; 131 if (fmode & FREAD) 132 nd->nl_flags |= NLC_READ; 133 if (fmode & FWRITE) 134 nd->nl_flags |= NLC_WRITE; 135 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 136 nd->nl_flags |= NLC_FOLLOW; 137 138 if (fmode & O_CREAT) { 139 /* 140 * CONDITIONAL CREATE FILE CASE 141 * 142 * Setting NLC_CREATE causes a negative hit to store 143 * the negative hit ncp and not return an error. Then 144 * nc_error or nc_vp may be checked to see if the ncp 145 * represents a negative hit. NLC_CREATE also requires 146 * write permission on the governing directory or EPERM 147 * is returned. 148 * 149 * If the file exists but is missing write permission, 150 * nlookup() returns EACCES. This has to be handled specially 151 * when combined with O_EXCL. 152 */ 153 nd->nl_flags |= NLC_CREATE; 154 nd->nl_flags |= NLC_REFDVP; 155 bwillinode(1); 156 error = nlookup(nd); 157 if (error == EACCES && nd->nl_nch.ncp->nc_vp != NULL && 158 (fmode & O_EXCL)) { 159 error = EEXIST; 160 } 161 162 /* 163 * If no error and nd->nl_dvp is NULL, the nlookup represents 164 * a mount-point or cross-mount situation. e.g. 165 * open("/var/cache", O_CREAT), where /var/cache is a 166 * mount point or a null-mount point. 167 */ 168 if (error == 0 && nd->nl_dvp == NULL) 169 error = EINVAL; 170 } else { 171 /* 172 * NORMAL OPEN FILE CASE 173 */ 174 error = nlookup(nd); 175 } 176 177 if (error) 178 return (error); 179 180 /* 181 * split case to allow us to re-resolve and retry the ncp in case 182 * we get ESTALE. 183 * 184 * (error is 0 on entry / retry) 185 */ 186 again: 187 /* 188 * Checks for (likely) filesystem-modifying cases and allows 189 * the filesystem to stall the front-end. 190 */ 191 if ((fmode & (FWRITE | O_TRUNC)) || 192 ((fmode & O_CREAT) && nd->nl_nch.ncp->nc_vp == NULL)) { 193 error = ncp_writechk(&nd->nl_nch); 194 if (error) 195 return error; 196 } 197 198 vpexcl = 1; 199 if (fmode & O_CREAT) { 200 if (nd->nl_nch.ncp->nc_vp == NULL) { 201 VATTR_NULL(vap); 202 vap->va_type = VREG; 203 vap->va_mode = cmode; 204 vap->va_fuseflags = fmode; /* FUSE */ 205 if (fmode & O_EXCL) 206 vap->va_vaflags |= VA_EXCLUSIVE; 207 error = VOP_NCREATE(&nd->nl_nch, nd->nl_dvp, &vp, 208 nd->nl_cred, vap); 209 if (error) 210 return (error); 211 fmode &= ~O_TRUNC; 212 /* locked vnode is returned */ 213 } else { 214 if (fmode & O_EXCL) { 215 error = EEXIST; 216 } else { 217 error = cache_vget(&nd->nl_nch, cred, 218 LK_EXCLUSIVE, &vp); 219 } 220 if (error) 221 return (error); 222 fmode &= ~O_CREAT; 223 } 224 } else { 225 /* 226 * In most other cases a shared lock on the vnode is 227 * sufficient. However, the O_RDWR case needs an 228 * exclusive lock if the vnode is executable. The 229 * NLC_EXCLLOCK_IFEXEC and NCF_NOTX flags help resolve 230 * this. 231 * 232 * NOTE: If NCF_NOTX is not set, we do not know the 233 * the state of the 'x' bits and have to get 234 * an exclusive lock for the EXCLLOCK_IFEXEC case. 235 */ 236 if ((nd->nl_flags & NLC_SHAREDLOCK) && 237 ((nd->nl_flags & NLC_EXCLLOCK_IFEXEC) == 0 || 238 nd->nl_nch.ncp->nc_flag & NCF_NOTX)) { 239 error = cache_vget(&nd->nl_nch, cred, LK_SHARED, &vp); 240 vpexcl = 0; 241 } else { 242 error = cache_vget(&nd->nl_nch, cred, 243 LK_EXCLUSIVE, &vp); 244 } 245 if (error) 246 return (error); 247 } 248 249 /* 250 * We have a locked vnode and ncp now. Note that the ncp will 251 * be cleaned up by the caller if nd->nl_nch is left intact. 252 */ 253 if (vp->v_type == VLNK) { 254 error = EMLINK; 255 goto bad; 256 } 257 if (vp->v_type == VSOCK) { 258 error = EOPNOTSUPP; 259 goto bad; 260 } 261 if (vp->v_type != VDIR && (fmode & O_DIRECTORY)) { 262 error = ENOTDIR; 263 goto bad; 264 } 265 if ((fmode & O_CREAT) == 0) { 266 if (fmode & (FWRITE | O_TRUNC)) { 267 if (vp->v_type == VDIR) { 268 error = EISDIR; 269 goto bad; 270 } 271 272 /* 273 * Additional checks on vnode (does not substitute 274 * for ncp_writechk()). 275 */ 276 error = vn_writechk(vp); 277 if (error) { 278 /* 279 * Special stale handling, re-resolve the 280 * vnode. 281 */ 282 if (error == ESTALE) { 283 vput(vp); 284 vp = NULL; 285 if (vpexcl == 0) { 286 cache_unlock(&nd->nl_nch); 287 cache_lock(&nd->nl_nch); 288 } 289 cache_setunresolved(&nd->nl_nch); 290 error = cache_resolve(&nd->nl_nch, 291 cred); 292 if (error == 0) 293 goto again; 294 } 295 goto bad; 296 } 297 } 298 } 299 if (fmode & O_TRUNC) { 300 vn_unlock(vp); /* XXX */ 301 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 302 osize = vp->v_filesize; 303 VATTR_NULL(vap); 304 vap->va_size = 0; 305 error = VOP_SETATTR_FP(vp, vap, cred, fp); 306 if (error) 307 goto bad; 308 error = VOP_GETATTR(vp, vap); 309 if (error) 310 goto bad; 311 mp = vq_vptomp(vp); 312 VFS_ACCOUNT(mp, vap->va_uid, vap->va_gid, -osize); 313 } 314 315 /* 316 * Set or clear VNSWAPCACHE on the vp based on nd->nl_nch.ncp->nc_flag. 317 * These particular bits a tracked all the way from the root. 318 * 319 * NOTE: Might not work properly on NFS servers due to the 320 * disconnected namecache. 321 */ 322 flags = nd->nl_nch.ncp->nc_flag; 323 if ((flags & (NCF_UF_CACHE | NCF_UF_PCACHE)) && 324 (flags & (NCF_SF_NOCACHE | NCF_SF_PNOCACHE)) == 0) { 325 vsetflags(vp, VSWAPCACHE); 326 } else { 327 vclrflags(vp, VSWAPCACHE); 328 } 329 330 /* 331 * Setup the fp so VOP_OPEN can override it. No descriptor has been 332 * associated with the fp yet so we own it clean. 333 * 334 * f_nchandle inherits nl_nch. This used to be necessary only for 335 * directories but now we do it unconditionally so f*() ops 336 * such as fchmod() can access the actual namespace that was 337 * used to open the file. 338 */ 339 if (fp) { 340 if (nd->nl_flags & NLC_APPENDONLY) 341 fmode |= FAPPENDONLY; 342 fp->f_nchandle = nd->nl_nch; 343 cache_zero(&nd->nl_nch); 344 cache_unlock(&fp->f_nchandle); 345 } 346 347 /* 348 * Get rid of nl_nch. vn_open does not return it (it returns the 349 * vnode or the file pointer). 350 * 351 * NOTE: We can't leave nl_nch locked through the VOP_OPEN anyway 352 * since the VOP_OPEN may block, e.g. on /dev/ttyd0 353 * 354 * NOTE: The VOP_OPEN() can replace the *fpp we supply with its own 355 * (it will fdrop/fhold), and can also set the *fpp up however 356 * it wants, not necessarily using DTYPE_VNODE. 357 */ 358 if (nd->nl_nch.ncp) 359 cache_put(&nd->nl_nch); 360 361 error = VOP_OPEN(vp, fmode, cred, fpp); 362 fp = fpp ? *fpp : NULL; 363 364 if (error) { 365 /* 366 * setting f_ops to &badfileops will prevent the descriptor 367 * code from trying to close and release the vnode, since 368 * the open failed we do not want to call close. 369 */ 370 if (fp) { 371 fp->f_data = NULL; 372 fp->f_ops = &badfileops; 373 } 374 goto bad; 375 } 376 377 #if 0 378 /* 379 * Assert that VREG files have been setup for vmio. 380 */ 381 KASSERT(vp->v_type != VREG || vp->v_object != NULL, 382 ("vn_open: regular file was not VMIO enabled!")); 383 #endif 384 385 /* 386 * Return the vnode. XXX needs some cleaning up. The vnode is 387 * only returned in the fp == NULL case. 388 * 389 * NOTE: vnode stored in fp may be different 390 */ 391 if (fp == NULL) { 392 nd->nl_open_vp = vp; 393 nd->nl_vp_fmode = fmode; 394 if ((nd->nl_flags & NLC_LOCKVP) == 0) 395 vn_unlock(vp); 396 } else { 397 vput(vp); 398 } 399 return (0); 400 bad: 401 if (vp) 402 vput(vp); 403 return (error); 404 } 405 406 int 407 vn_opendisk(const char *devname, int fmode, struct vnode **vpp) 408 { 409 struct vnode *vp; 410 int error; 411 412 if (strncmp(devname, "/dev/", 5) == 0) 413 devname += 5; 414 if ((vp = getsynthvnode(devname)) == NULL) { 415 error = ENODEV; 416 } else { 417 error = VOP_OPEN(vp, fmode, proc0.p_ucred, NULL); 418 vn_unlock(vp); 419 if (error) { 420 vrele(vp); 421 vp = NULL; 422 } 423 } 424 *vpp = vp; 425 return (error); 426 } 427 428 /* 429 * Checks for special conditions on the vnode which might prevent writing 430 * after the vnode has (likely) been locked. The vnode might or might not 431 * be locked as of this call, but will be at least referenced. 432 * 433 * Also re-checks the mount RDONLY flag that ncp_writechk() checked prior 434 * to the vnode being locked. 435 */ 436 int 437 vn_writechk(struct vnode *vp) 438 { 439 /* 440 * If there's shared text associated with 441 * the vnode, try to free it up once. If 442 * we fail, we can't allow writing. 443 */ 444 if (vp->v_flag & VTEXT) 445 return (ETXTBSY); 446 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_RDONLY)) 447 return (EROFS); 448 return 0; 449 } 450 451 /* 452 * Check whether the underlying mount is read-only. The mount point 453 * referenced by the namecache may be different from the mount point 454 * used by the underlying vnode in the case of NULLFS, so a separate 455 * check is needed. 456 * 457 * Must be called PRIOR to any vnodes being locked. 458 */ 459 int 460 ncp_writechk(struct nchandle *nch) 461 { 462 struct mount *mp; 463 464 if ((mp = nch->mount) != NULL) { 465 if (mp->mnt_flag & MNT_RDONLY) 466 return (EROFS); 467 if (mp->mnt_op->vfs_modifying != vfs_stdmodifying) 468 VFS_MODIFYING(mp); 469 } 470 return(0); 471 } 472 473 /* 474 * Vnode close call 475 * 476 * MPSAFE 477 */ 478 int 479 vn_close(struct vnode *vp, int flags, struct file *fp) 480 { 481 int error; 482 483 error = vn_lock(vp, LK_SHARED | LK_RETRY | LK_FAILRECLAIM); 484 if (error == 0) { 485 error = VOP_CLOSE(vp, flags, fp); 486 vn_unlock(vp); 487 } 488 vrele(vp); 489 return (error); 490 } 491 492 /* 493 * Sequential heuristic. 494 * 495 * MPSAFE (f_seqcount and f_nextoff are allowed to race) 496 */ 497 static __inline 498 int 499 sequential_heuristic(struct uio *uio, struct file *fp) 500 { 501 /* 502 * Sequential heuristic - detect sequential operation 503 * 504 * NOTE: SMP: We allow f_seqcount updates to race. 505 */ 506 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 507 uio->uio_offset == fp->f_nextoff) { 508 int tmpseq = fp->f_seqcount; 509 510 tmpseq += howmany(uio->uio_resid, MAXBSIZE); 511 if (tmpseq > IO_SEQMAX) 512 tmpseq = IO_SEQMAX; 513 fp->f_seqcount = tmpseq; 514 return(fp->f_seqcount << IO_SEQSHIFT); 515 } 516 517 /* 518 * Not sequential, quick draw-down of seqcount 519 * 520 * NOTE: SMP: We allow f_seqcount updates to race. 521 */ 522 if (fp->f_seqcount > 1) 523 fp->f_seqcount = 1; 524 else 525 fp->f_seqcount = 0; 526 return(0); 527 } 528 529 /* 530 * get - lock and return the f_offset field. 531 * set - set and unlock the f_offset field. 532 * 533 * These routines serve the dual purpose of serializing access to the 534 * f_offset field (at least on x86) and guaranteeing operational integrity 535 * when multiple read()ers and write()ers are present on the same fp. 536 * 537 * MPSAFE 538 */ 539 static __inline off_t 540 vn_get_fpf_offset(struct file *fp) 541 { 542 u_int flags; 543 u_int nflags; 544 545 /* 546 * Shortcut critical path. 547 */ 548 flags = fp->f_flag & ~FOFFSETLOCK; 549 if (atomic_cmpset_int(&fp->f_flag, flags, flags | FOFFSETLOCK)) 550 return(fp->f_offset); 551 552 /* 553 * The hard way 554 */ 555 for (;;) { 556 flags = fp->f_flag; 557 if (flags & FOFFSETLOCK) { 558 nflags = flags | FOFFSETWAKE; 559 tsleep_interlock(&fp->f_flag, 0); 560 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) 561 tsleep(&fp->f_flag, PINTERLOCKED, "fpoff", 0); 562 } else { 563 nflags = flags | FOFFSETLOCK; 564 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) 565 break; 566 } 567 } 568 return(fp->f_offset); 569 } 570 571 /* 572 * MPSAFE 573 */ 574 static __inline void 575 vn_set_fpf_offset(struct file *fp, off_t offset) 576 { 577 u_int flags; 578 u_int nflags; 579 580 /* 581 * We hold the lock so we can set the offset without interference. 582 */ 583 fp->f_offset = offset; 584 585 /* 586 * Normal release is already a reasonably critical path. 587 */ 588 for (;;) { 589 flags = fp->f_flag; 590 nflags = flags & ~(FOFFSETLOCK | FOFFSETWAKE); 591 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) { 592 if (flags & FOFFSETWAKE) 593 wakeup(&fp->f_flag); 594 break; 595 } 596 } 597 } 598 599 /* 600 * MPSAFE 601 */ 602 static __inline off_t 603 vn_poll_fpf_offset(struct file *fp) 604 { 605 #if defined(__x86_64__) 606 return(fp->f_offset); 607 #else 608 off_t off = vn_get_fpf_offset(fp); 609 vn_set_fpf_offset(fp, off); 610 return(off); 611 #endif 612 } 613 614 /* 615 * Package up an I/O request on a vnode into a uio and do it. 616 * 617 * MPSAFE 618 */ 619 int 620 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 621 off_t offset, enum uio_seg segflg, int ioflg, 622 struct ucred *cred, int *aresid) 623 { 624 struct uio auio; 625 struct iovec aiov; 626 int error; 627 628 if ((ioflg & IO_NODELOCKED) == 0) 629 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 630 auio.uio_iov = &aiov; 631 auio.uio_iovcnt = 1; 632 aiov.iov_base = base; 633 aiov.iov_len = len; 634 auio.uio_resid = len; 635 auio.uio_offset = offset; 636 auio.uio_segflg = segflg; 637 auio.uio_rw = rw; 638 auio.uio_td = curthread; 639 if (rw == UIO_READ) { 640 error = VOP_READ(vp, &auio, ioflg, cred); 641 } else { 642 error = VOP_WRITE(vp, &auio, ioflg, cred); 643 } 644 if (aresid) 645 *aresid = auio.uio_resid; 646 else 647 if (auio.uio_resid && error == 0) 648 error = EIO; 649 if ((ioflg & IO_NODELOCKED) == 0) 650 vn_unlock(vp); 651 return (error); 652 } 653 654 /* 655 * Package up an I/O request on a vnode into a uio and do it. The I/O 656 * request is split up into smaller chunks and we try to avoid saturating 657 * the buffer cache while potentially holding a vnode locked, so we 658 * check bwillwrite() before calling vn_rdwr(). We also call lwkt_user_yield() 659 * to give other processes a chance to lock the vnode (either other processes 660 * core'ing the same binary, or unrelated processes scanning the directory). 661 * 662 * MPSAFE 663 */ 664 int 665 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 666 off_t offset, enum uio_seg segflg, int ioflg, 667 struct ucred *cred, int *aresid) 668 { 669 int error = 0; 670 671 do { 672 int chunk; 673 674 /* 675 * Force `offset' to a multiple of MAXBSIZE except possibly 676 * for the first chunk, so that filesystems only need to 677 * write full blocks except possibly for the first and last 678 * chunks. 679 */ 680 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 681 682 if (chunk > len) 683 chunk = len; 684 if (vp->v_type == VREG && (ioflg & IO_RECURSE) == 0) { 685 switch(rw) { 686 case UIO_READ: 687 bwillread(chunk); 688 break; 689 case UIO_WRITE: 690 bwillwrite(chunk); 691 break; 692 } 693 } 694 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 695 ioflg, cred, aresid); 696 len -= chunk; /* aresid calc already includes length */ 697 if (error) 698 break; 699 offset += chunk; 700 base += chunk; 701 lwkt_user_yield(); 702 } while (len); 703 if (aresid) 704 *aresid += len; 705 return (error); 706 } 707 708 /* 709 * File pointers can no longer get ripped up by revoke so 710 * we don't need to lock access to the vp. 711 * 712 * f_offset updates are not guaranteed against multiple readers 713 */ 714 static int 715 vn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 716 { 717 struct vnode *vp; 718 int error, ioflag; 719 720 KASSERT(uio->uio_td == curthread, 721 ("uio_td %p is not td %p", uio->uio_td, curthread)); 722 vp = (struct vnode *)fp->f_data; 723 724 ioflag = 0; 725 if (flags & O_FBLOCKING) { 726 /* ioflag &= ~IO_NDELAY; */ 727 } else if (flags & O_FNONBLOCKING) { 728 ioflag |= IO_NDELAY; 729 } else if (fp->f_flag & FNONBLOCK) { 730 ioflag |= IO_NDELAY; 731 } 732 if (fp->f_flag & O_DIRECT) { 733 ioflag |= IO_DIRECT; 734 } 735 if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0) 736 uio->uio_offset = vn_get_fpf_offset(fp); 737 vn_lock(vp, LK_SHARED | LK_RETRY); 738 ioflag |= sequential_heuristic(uio, fp); 739 740 error = VOP_READ_FP(vp, uio, ioflag, cred, fp); 741 fp->f_nextoff = uio->uio_offset; 742 vn_unlock(vp); 743 if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0) 744 vn_set_fpf_offset(fp, uio->uio_offset); 745 return (error); 746 } 747 748 /* 749 * MPSAFE 750 */ 751 static int 752 vn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 753 { 754 struct vnode *vp; 755 int error, ioflag; 756 757 KASSERT(uio->uio_td == curthread, 758 ("uio_td %p is not p %p", uio->uio_td, curthread)); 759 vp = (struct vnode *)fp->f_data; 760 761 ioflag = IO_UNIT; 762 if (vp->v_type == VREG && 763 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 764 ioflag |= IO_APPEND; 765 } 766 767 if (flags & O_FBLOCKING) { 768 /* ioflag &= ~IO_NDELAY; */ 769 } else if (flags & O_FNONBLOCKING) { 770 ioflag |= IO_NDELAY; 771 } else if (fp->f_flag & FNONBLOCK) { 772 ioflag |= IO_NDELAY; 773 } 774 if (fp->f_flag & O_DIRECT) { 775 ioflag |= IO_DIRECT; 776 } 777 if (flags & O_FASYNCWRITE) { 778 /* ioflag &= ~IO_SYNC; */ 779 } else if (flags & O_FSYNCWRITE) { 780 ioflag |= IO_SYNC; 781 } else if (fp->f_flag & O_FSYNC) { 782 ioflag |= IO_SYNC; 783 } 784 785 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 786 ioflag |= IO_SYNC; 787 if ((flags & O_FOFFSET) == 0) 788 uio->uio_offset = vn_get_fpf_offset(fp); 789 if (vp->v_mount) 790 VFS_MODIFYING(vp->v_mount); 791 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 792 ioflag |= sequential_heuristic(uio, fp); 793 error = VOP_WRITE_FP(vp, uio, ioflag, cred, fp); 794 fp->f_nextoff = uio->uio_offset; 795 vn_unlock(vp); 796 if ((flags & O_FOFFSET) == 0) 797 vn_set_fpf_offset(fp, uio->uio_offset); 798 return (error); 799 } 800 801 /* 802 * MPSAFE 803 */ 804 static int 805 vn_statfile(struct file *fp, struct stat *sb, struct ucred *cred) 806 { 807 struct vnode *vp; 808 int error; 809 810 vp = (struct vnode *)fp->f_data; 811 error = vn_stat(vp, sb, cred); 812 return (error); 813 } 814 815 /* 816 * MPSAFE 817 */ 818 int 819 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *cred) 820 { 821 struct vattr vattr; 822 struct vattr *vap; 823 int error; 824 u_short mode; 825 cdev_t dev; 826 827 /* 828 * vp already has a ref and is validated, can call unlocked. 829 */ 830 vap = &vattr; 831 error = VOP_GETATTR(vp, vap); 832 if (error) 833 return (error); 834 835 /* 836 * Zero the spare stat fields 837 */ 838 sb->st_lspare = 0; 839 sb->st_qspare2 = 0; 840 841 /* 842 * Copy from vattr table 843 */ 844 if (vap->va_fsid != VNOVAL) 845 sb->st_dev = vap->va_fsid; 846 else 847 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 848 sb->st_ino = vap->va_fileid; 849 mode = vap->va_mode; 850 switch (vap->va_type) { 851 case VREG: 852 mode |= S_IFREG; 853 break; 854 case VDATABASE: 855 mode |= S_IFDB; 856 break; 857 case VDIR: 858 mode |= S_IFDIR; 859 break; 860 case VBLK: 861 mode |= S_IFBLK; 862 break; 863 case VCHR: 864 mode |= S_IFCHR; 865 break; 866 case VLNK: 867 mode |= S_IFLNK; 868 /* This is a cosmetic change, symlinks do not have a mode. */ 869 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) 870 sb->st_mode &= ~ACCESSPERMS; /* 0000 */ 871 else 872 sb->st_mode |= ACCESSPERMS; /* 0777 */ 873 break; 874 case VSOCK: 875 mode |= S_IFSOCK; 876 break; 877 case VFIFO: 878 mode |= S_IFIFO; 879 break; 880 default: 881 return (EBADF); 882 } 883 sb->st_mode = mode; 884 if (vap->va_nlink > (nlink_t)-1) 885 sb->st_nlink = (nlink_t)-1; 886 else 887 sb->st_nlink = vap->va_nlink; 888 sb->st_uid = vap->va_uid; 889 sb->st_gid = vap->va_gid; 890 sb->st_rdev = devid_from_dev(vp->v_rdev); 891 sb->st_size = vap->va_size; 892 sb->st_atimespec = vap->va_atime; 893 sb->st_mtimespec = vap->va_mtime; 894 sb->st_ctimespec = vap->va_ctime; 895 896 /* 897 * A VCHR and VBLK device may track the last access and last modified 898 * time independantly of the filesystem. This is particularly true 899 * because device read and write calls may bypass the filesystem. 900 */ 901 if (vp->v_type == VCHR || vp->v_type == VBLK) { 902 dev = vp->v_rdev; 903 if (dev != NULL) { 904 if (dev->si_lastread) { 905 sb->st_atimespec.tv_sec = time_second + 906 (dev->si_lastread - 907 time_uptime); 908 sb->st_atimespec.tv_nsec = 0; 909 } 910 if (dev->si_lastwrite) { 911 sb->st_mtimespec.tv_sec = time_second + 912 (dev->si_lastwrite - 913 time_uptime); 914 sb->st_mtimespec.tv_nsec = 0; 915 } 916 } 917 } 918 919 /* 920 * According to www.opengroup.org, the meaning of st_blksize is 921 * "a filesystem-specific preferred I/O block size for this 922 * object. In some filesystem types, this may vary from file 923 * to file" 924 * Default to PAGE_SIZE after much discussion. 925 */ 926 927 if (vap->va_type == VREG) { 928 sb->st_blksize = vap->va_blocksize; 929 } else if (vn_isdisk(vp, NULL)) { 930 /* 931 * XXX this is broken. If the device is not yet open (aka 932 * stat() call, aka v_rdev == NULL), how are we supposed 933 * to get a valid block size out of it? 934 */ 935 dev = vp->v_rdev; 936 937 sb->st_blksize = dev->si_bsize_best; 938 if (sb->st_blksize < dev->si_bsize_phys) 939 sb->st_blksize = dev->si_bsize_phys; 940 if (sb->st_blksize < BLKDEV_IOSIZE) 941 sb->st_blksize = BLKDEV_IOSIZE; 942 } else { 943 sb->st_blksize = PAGE_SIZE; 944 } 945 946 sb->st_flags = vap->va_flags; 947 948 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0); 949 if (error) 950 sb->st_gen = 0; 951 else 952 sb->st_gen = (u_int32_t)vap->va_gen; 953 954 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 955 956 /* 957 * This is for ABI compatibility <= 5.7 (for ABI change made in 958 * 5.7 master). 959 */ 960 sb->__old_st_blksize = sb->st_blksize; 961 962 return (0); 963 } 964 965 /* 966 * MPALMOSTSAFE - acquires mplock 967 */ 968 static int 969 vn_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *ucred, 970 struct sysmsg *msg) 971 { 972 struct vnode *vp = ((struct vnode *)fp->f_data); 973 struct vnode *ovp; 974 struct vattr vattr; 975 int error; 976 off_t size; 977 978 switch (vp->v_type) { 979 case VREG: 980 case VDIR: 981 if (com == FIONREAD) { 982 error = VOP_GETATTR(vp, &vattr); 983 if (error) 984 break; 985 size = vattr.va_size; 986 if ((vp->v_flag & VNOTSEEKABLE) == 0) 987 size -= vn_poll_fpf_offset(fp); 988 if (size > 0x7FFFFFFF) 989 size = 0x7FFFFFFF; 990 *(int *)data = size; 991 error = 0; 992 break; 993 } 994 if (com == FIOASYNC) { /* XXX */ 995 error = 0; /* XXX */ 996 break; 997 } 998 /* fall into ... */ 999 default: 1000 #if 0 1001 return (ENOTTY); 1002 #endif 1003 case VFIFO: 1004 case VCHR: 1005 case VBLK: 1006 if (com == FIODTYPE) { 1007 if (vp->v_type != VCHR && vp->v_type != VBLK) { 1008 error = ENOTTY; 1009 break; 1010 } 1011 *(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK; 1012 error = 0; 1013 break; 1014 } 1015 error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred, msg); 1016 if (error == 0 && com == TIOCSCTTY) { 1017 struct proc *p = curthread->td_proc; 1018 struct session *sess; 1019 1020 if (p == NULL) { 1021 error = ENOTTY; 1022 break; 1023 } 1024 1025 get_mplock(); 1026 sess = p->p_session; 1027 /* Do nothing if reassigning same control tty */ 1028 if (sess->s_ttyvp == vp) { 1029 error = 0; 1030 rel_mplock(); 1031 break; 1032 } 1033 1034 /* Get rid of reference to old control tty */ 1035 ovp = sess->s_ttyvp; 1036 vref(vp); 1037 sess->s_ttyvp = vp; 1038 if (ovp) 1039 vrele(ovp); 1040 rel_mplock(); 1041 } 1042 break; 1043 } 1044 return (error); 1045 } 1046 1047 /* 1048 * Obtain the requested vnode lock 1049 * 1050 * LK_RETRY Automatically retry on timeout 1051 * LK_FAILRECLAIM Fail if the vnode is being reclaimed 1052 * 1053 * Failures will occur if the vnode is undergoing recyclement, but not 1054 * all callers expect that the function will fail so the caller must pass 1055 * LK_FAILOK if it wants to process an error code. 1056 * 1057 * Errors can occur for other reasons if you pass in other LK_ flags, 1058 * regardless of whether you pass in LK_FAILRECLAIM 1059 */ 1060 int 1061 vn_lock(struct vnode *vp, int flags) 1062 { 1063 int error; 1064 1065 do { 1066 error = lockmgr(&vp->v_lock, flags); 1067 if (error == 0) 1068 break; 1069 } while (flags & LK_RETRY); 1070 1071 /* 1072 * Because we (had better!) have a ref on the vnode, once it 1073 * goes to VRECLAIMED state it will not be recycled until all 1074 * refs go away. So we can just check the flag. 1075 */ 1076 if (error == 0 && (vp->v_flag & VRECLAIMED)) { 1077 if (flags & LK_FAILRECLAIM) { 1078 lockmgr(&vp->v_lock, LK_RELEASE); 1079 error = ENOENT; 1080 } 1081 } 1082 return (error); 1083 } 1084 1085 int 1086 vn_relock(struct vnode *vp, int flags) 1087 { 1088 int error; 1089 1090 do { 1091 error = lockmgr(&vp->v_lock, flags); 1092 if (error == 0) 1093 break; 1094 } while (flags & LK_RETRY); 1095 1096 return error; 1097 } 1098 1099 #ifdef DEBUG_VN_UNLOCK 1100 1101 void 1102 debug_vn_unlock(struct vnode *vp, const char *filename, int line) 1103 { 1104 kprintf("vn_unlock from %s:%d\n", filename, line); 1105 lockmgr(&vp->v_lock, LK_RELEASE); 1106 } 1107 1108 #else 1109 1110 void 1111 vn_unlock(struct vnode *vp) 1112 { 1113 lockmgr(&vp->v_lock, LK_RELEASE); 1114 } 1115 1116 #endif 1117 1118 /* 1119 * MPSAFE 1120 */ 1121 int 1122 vn_islocked(struct vnode *vp) 1123 { 1124 return (lockstatus(&vp->v_lock, curthread)); 1125 } 1126 1127 /* 1128 * Return the lock status of a vnode and unlock the vnode 1129 * if we owned the lock. This is not a boolean, if the 1130 * caller cares what the lock status is the caller must 1131 * check the various possible values. 1132 * 1133 * This only unlocks exclusive locks held by the caller, 1134 * it will NOT unlock shared locks (there is no way to 1135 * tell who the shared lock belongs to). 1136 * 1137 * MPSAFE 1138 */ 1139 int 1140 vn_islocked_unlock(struct vnode *vp) 1141 { 1142 int vpls; 1143 1144 vpls = lockstatus(&vp->v_lock, curthread); 1145 if (vpls == LK_EXCLUSIVE) 1146 lockmgr(&vp->v_lock, LK_RELEASE); 1147 return(vpls); 1148 } 1149 1150 /* 1151 * Restore a vnode lock that we previously released via 1152 * vn_islocked_unlock(). This is a NOP if we did not 1153 * own the original lock. 1154 * 1155 * MPSAFE 1156 */ 1157 void 1158 vn_islocked_relock(struct vnode *vp, int vpls) 1159 { 1160 int error; 1161 1162 if (vpls == LK_EXCLUSIVE) 1163 error = lockmgr(&vp->v_lock, vpls); 1164 } 1165 1166 /* 1167 * MPSAFE 1168 */ 1169 static int 1170 vn_closefile(struct file *fp) 1171 { 1172 int error; 1173 1174 fp->f_ops = &badfileops; 1175 error = vn_close(((struct vnode *)fp->f_data), fp->f_flag, fp); 1176 return (error); 1177 } 1178 1179 /* 1180 * MPSAFE 1181 */ 1182 static int 1183 vn_kqfilter(struct file *fp, struct knote *kn) 1184 { 1185 int error; 1186 1187 error = VOP_KQFILTER(((struct vnode *)fp->f_data), kn); 1188 return (error); 1189 } 1190