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/caps.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) && !nd->nl_dir_error) 159 { 160 error = EEXIST; 161 } 162 163 /* 164 * If no error and nd->nl_dvp is NULL, the nlookup represents 165 * a mount-point or cross-mount situation. e.g. 166 * open("/var/cache", O_CREAT), where /var/cache is a 167 * mount point or a null-mount point. 168 */ 169 if (error == 0 && nd->nl_dvp == NULL) 170 error = EINVAL; 171 } else { 172 /* 173 * NORMAL OPEN FILE CASE 174 */ 175 error = nlookup(nd); 176 } 177 178 if (error) 179 return (error); 180 181 /* 182 * split case to allow us to re-resolve and retry the ncp in case 183 * we get ESTALE. 184 * 185 * (error is 0 on entry / retry) 186 */ 187 again: 188 /* 189 * Checks for (likely) filesystem-modifying cases and allows 190 * the filesystem to stall the front-end. 191 */ 192 if ((fmode & (FWRITE | O_TRUNC)) || 193 ((fmode & O_CREAT) && nd->nl_nch.ncp->nc_vp == NULL)) { 194 error = ncp_writechk(&nd->nl_nch); 195 if (error) 196 return error; 197 } 198 199 vpexcl = 1; 200 if (fmode & O_CREAT) { 201 if (nd->nl_nch.ncp->nc_vp == NULL) { 202 VATTR_NULL(vap); 203 vap->va_type = VREG; 204 vap->va_mode = cmode; 205 vap->va_fuseflags = fmode; /* FUSE */ 206 if (fmode & O_EXCL) 207 vap->va_vaflags |= VA_EXCLUSIVE; 208 error = VOP_NCREATE(&nd->nl_nch, nd->nl_dvp, &vp, 209 nd->nl_cred, vap); 210 if (error) 211 return (error); 212 fmode &= ~O_TRUNC; 213 /* locked vnode is returned */ 214 } else { 215 if (fmode & O_EXCL) { 216 error = EEXIST; 217 } else { 218 error = cache_vget(&nd->nl_nch, cred, 219 LK_EXCLUSIVE, &vp); 220 } 221 if (error) 222 return (error); 223 fmode &= ~O_CREAT; 224 } 225 } else { 226 /* 227 * In most other cases a shared lock on the vnode is 228 * sufficient. However, the O_RDWR case needs an 229 * exclusive lock if the vnode is executable. The 230 * NLC_EXCLLOCK_IFEXEC and NCF_NOTX flags help resolve 231 * this. 232 * 233 * NOTE: If NCF_NOTX is not set, we do not know the 234 * the state of the 'x' bits and have to get 235 * an exclusive lock for the EXCLLOCK_IFEXEC case. 236 */ 237 if ((nd->nl_flags & NLC_SHAREDLOCK) && 238 ((nd->nl_flags & NLC_EXCLLOCK_IFEXEC) == 0 || 239 nd->nl_nch.ncp->nc_flag & NCF_NOTX)) { 240 error = cache_vget(&nd->nl_nch, cred, LK_SHARED, &vp); 241 vpexcl = 0; 242 } else { 243 error = cache_vget(&nd->nl_nch, cred, 244 LK_EXCLUSIVE, &vp); 245 } 246 if (error) 247 return (error); 248 } 249 250 /* 251 * We have a locked vnode and ncp now. Note that the ncp will 252 * be cleaned up by the caller if nd->nl_nch is left intact. 253 */ 254 if (vp->v_type == VLNK) { 255 error = EMLINK; 256 goto bad; 257 } 258 if (vp->v_type == VSOCK) { 259 error = EOPNOTSUPP; 260 goto bad; 261 } 262 if (vp->v_type != VDIR && (fmode & O_DIRECTORY)) { 263 error = ENOTDIR; 264 goto bad; 265 } 266 if ((fmode & O_CREAT) == 0) { 267 if (fmode & (FWRITE | O_TRUNC)) { 268 if (vp->v_type == VDIR) { 269 error = EISDIR; 270 goto bad; 271 } 272 273 /* 274 * Additional checks on vnode (does not substitute 275 * for ncp_writechk()). 276 */ 277 error = vn_writechk(vp); 278 if (error) { 279 /* 280 * Special stale handling, re-resolve the 281 * vnode. 282 */ 283 if (error == ESTALE) { 284 u_int dummy_gen = 0; 285 286 vput(vp); 287 vp = NULL; 288 if (vpexcl == 0) { 289 cache_unlock(&nd->nl_nch); 290 cache_lock(&nd->nl_nch); 291 } 292 cache_setunresolved(&nd->nl_nch); 293 error = cache_resolve(&nd->nl_nch, 294 &dummy_gen, 295 cred); 296 if (error == 0) 297 goto again; 298 } 299 goto bad; 300 } 301 } 302 } 303 if (fmode & O_TRUNC) { 304 vn_unlock(vp); /* XXX */ 305 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 306 osize = vp->v_filesize; 307 VATTR_NULL(vap); 308 vap->va_size = 0; 309 error = VOP_SETATTR_FP(vp, vap, cred, fp); 310 if (error) 311 goto bad; 312 error = VOP_GETATTR(vp, vap); 313 if (error) 314 goto bad; 315 mp = vq_vptomp(vp); 316 VFS_ACCOUNT(mp, vap->va_uid, vap->va_gid, -osize); 317 } 318 319 /* 320 * Set or clear VNSWAPCACHE on the vp based on nd->nl_nch.ncp->nc_flag. 321 * These particular bits a tracked all the way from the root. 322 * 323 * NOTE: Might not work properly on NFS servers due to the 324 * disconnected namecache. 325 */ 326 flags = nd->nl_nch.ncp->nc_flag; 327 if ((flags & (NCF_UF_CACHE | NCF_UF_PCACHE)) && 328 (flags & (NCF_SF_NOCACHE | NCF_SF_PNOCACHE)) == 0) { 329 vsetflags(vp, VSWAPCACHE); 330 } else { 331 vclrflags(vp, VSWAPCACHE); 332 } 333 334 /* 335 * Setup the fp so VOP_OPEN can override it. No descriptor has been 336 * associated with the fp yet so we own it clean. 337 * 338 * f_nchandle inherits nl_nch. This used to be necessary only for 339 * directories but now we do it unconditionally so f*() ops 340 * such as fchmod() can access the actual namespace that was 341 * used to open the file. 342 */ 343 if (fp) { 344 if (nd->nl_flags & NLC_APPENDONLY) 345 fmode |= FAPPENDONLY; 346 fp->f_nchandle = nd->nl_nch; 347 cache_zero(&nd->nl_nch); 348 cache_unlock(&fp->f_nchandle); 349 } 350 351 /* 352 * Get rid of nl_nch. vn_open does not return it (it returns the 353 * vnode or the file pointer). 354 * 355 * NOTE: We can't leave nl_nch locked through the VOP_OPEN anyway 356 * since the VOP_OPEN may block, e.g. on /dev/ttyd0 357 * 358 * NOTE: The VOP_OPEN() can replace the *fpp we supply with its own 359 * (it will fdrop/fhold), and can also set the *fpp up however 360 * it wants, not necessarily using DTYPE_VNODE. 361 */ 362 if (nd->nl_nch.ncp) 363 cache_put(&nd->nl_nch); 364 365 error = VOP_OPEN(vp, fmode, cred, fpp); 366 fp = fpp ? *fpp : NULL; 367 368 if (error) { 369 /* 370 * setting f_ops to &badfileops will prevent the descriptor 371 * code from trying to close and release the vnode, since 372 * the open failed we do not want to call close. 373 */ 374 if (fp) { 375 fp->f_data = NULL; 376 fp->f_ops = &badfileops; 377 } 378 goto bad; 379 } 380 381 #if 0 382 /* 383 * Assert that VREG files have been setup for vmio. 384 */ 385 KASSERT(vp->v_type != VREG || vp->v_object != NULL, 386 ("vn_open: regular file was not VMIO enabled!")); 387 #endif 388 389 /* 390 * Return the vnode. XXX needs some cleaning up. The vnode is 391 * only returned in the fp == NULL case. 392 * 393 * NOTE: vnode stored in fp may be different 394 */ 395 if (fp == NULL) { 396 nd->nl_open_vp = vp; 397 nd->nl_vp_fmode = fmode; 398 if ((nd->nl_flags & NLC_LOCKVP) == 0) 399 vn_unlock(vp); 400 } else { 401 vput(vp); 402 } 403 return (0); 404 bad: 405 if (vp) 406 vput(vp); 407 return (error); 408 } 409 410 int 411 vn_opendisk(const char *devname, int fmode, struct vnode **vpp) 412 { 413 struct vnode *vp; 414 int error; 415 416 if (strncmp(devname, "/dev/", 5) == 0) 417 devname += 5; 418 if ((vp = getsynthvnode(devname)) == NULL) { 419 error = ENODEV; 420 } else { 421 error = VOP_OPEN(vp, fmode, proc0.p_ucred, NULL); 422 vn_unlock(vp); 423 if (error) { 424 vrele(vp); 425 vp = NULL; 426 } 427 } 428 *vpp = vp; 429 return (error); 430 } 431 432 /* 433 * Checks for special conditions on the vnode which might prevent writing 434 * after the vnode has (likely) been locked. The vnode might or might not 435 * be locked as of this call, but will be at least referenced. 436 * 437 * Also re-checks the mount RDONLY flag that ncp_writechk() checked prior 438 * to the vnode being locked. 439 */ 440 int 441 vn_writechk(struct vnode *vp) 442 { 443 /* 444 * If there's shared text associated with 445 * the vnode, try to free it up once. If 446 * we fail, we can't allow writing. 447 */ 448 if (vp->v_flag & VTEXT) 449 return (ETXTBSY); 450 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_RDONLY)) 451 return (EROFS); 452 return 0; 453 } 454 455 /* 456 * Check whether the underlying mount is read-only. The mount point 457 * referenced by the namecache may be different from the mount point 458 * used by the underlying vnode in the case of NULLFS, so a separate 459 * check is needed. 460 * 461 * Must be called PRIOR to any vnodes being locked. 462 */ 463 int 464 ncp_writechk(struct nchandle *nch) 465 { 466 struct mount *mp; 467 468 if ((mp = nch->mount) != NULL) { 469 if (mp->mnt_flag & MNT_RDONLY) 470 return (EROFS); 471 if (mp->mnt_op->vfs_modifying != vfs_stdmodifying) 472 VFS_MODIFYING(mp); 473 } 474 return(0); 475 } 476 477 /* 478 * Vnode close call 479 * 480 * MPSAFE 481 */ 482 int 483 vn_close(struct vnode *vp, int flags, struct file *fp) 484 { 485 int error; 486 487 error = vn_lock(vp, LK_SHARED | LK_RETRY | LK_FAILRECLAIM); 488 if (error == 0) { 489 error = VOP_CLOSE(vp, flags, fp); 490 vn_unlock(vp); 491 } 492 vrele(vp); 493 return (error); 494 } 495 496 /* 497 * Sequential heuristic. 498 * 499 * MPSAFE (f_seqcount and f_nextoff are allowed to race) 500 */ 501 static __inline 502 int 503 sequential_heuristic(struct uio *uio, struct file *fp) 504 { 505 /* 506 * Sequential heuristic - detect sequential operation 507 * 508 * NOTE: SMP: We allow f_seqcount updates to race. 509 */ 510 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 511 uio->uio_offset == fp->f_nextoff) { 512 int tmpseq = fp->f_seqcount; 513 514 tmpseq += howmany(uio->uio_resid, MAXBSIZE); 515 if (tmpseq > IO_SEQMAX) 516 tmpseq = IO_SEQMAX; 517 fp->f_seqcount = tmpseq; 518 return(fp->f_seqcount << IO_SEQSHIFT); 519 } 520 521 /* 522 * Not sequential, quick draw-down of seqcount 523 * 524 * NOTE: SMP: We allow f_seqcount updates to race. 525 */ 526 if (fp->f_seqcount > 1) 527 fp->f_seqcount = 1; 528 else 529 fp->f_seqcount = 0; 530 return(0); 531 } 532 533 /* 534 * get - lock and return the f_offset field. 535 * set - set and unlock the f_offset field. 536 * 537 * These routines serve the dual purpose of serializing access to the 538 * f_offset field (at least on x86) and guaranteeing operational integrity 539 * when multiple read()ers and write()ers are present on the same fp. 540 * 541 * MPSAFE 542 */ 543 static __inline off_t 544 vn_get_fpf_offset(struct file *fp) 545 { 546 u_int flags; 547 u_int nflags; 548 549 /* 550 * Shortcut critical path. 551 */ 552 flags = fp->f_flag & ~FOFFSETLOCK; 553 if (atomic_cmpset_int(&fp->f_flag, flags, flags | FOFFSETLOCK)) 554 return(fp->f_offset); 555 556 /* 557 * The hard way 558 */ 559 for (;;) { 560 flags = fp->f_flag; 561 if (flags & FOFFSETLOCK) { 562 nflags = flags | FOFFSETWAKE; 563 tsleep_interlock(&fp->f_flag, 0); 564 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) 565 tsleep(&fp->f_flag, PINTERLOCKED, "fpoff", 0); 566 } else { 567 nflags = flags | FOFFSETLOCK; 568 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) 569 break; 570 } 571 } 572 return(fp->f_offset); 573 } 574 575 /* 576 * MPSAFE 577 */ 578 static __inline void 579 vn_set_fpf_offset(struct file *fp, off_t offset) 580 { 581 u_int flags; 582 u_int nflags; 583 584 /* 585 * We hold the lock so we can set the offset without interference. 586 */ 587 fp->f_offset = offset; 588 589 /* 590 * Normal release is already a reasonably critical path. 591 */ 592 for (;;) { 593 flags = fp->f_flag; 594 nflags = flags & ~(FOFFSETLOCK | FOFFSETWAKE); 595 if (atomic_cmpset_int(&fp->f_flag, flags, nflags)) { 596 if (flags & FOFFSETWAKE) 597 wakeup(&fp->f_flag); 598 break; 599 } 600 } 601 } 602 603 /* 604 * MPSAFE 605 */ 606 static __inline off_t 607 vn_poll_fpf_offset(struct file *fp) 608 { 609 #if defined(__x86_64__) 610 return(fp->f_offset); 611 #else 612 off_t off = vn_get_fpf_offset(fp); 613 vn_set_fpf_offset(fp, off); 614 return(off); 615 #endif 616 } 617 618 /* 619 * Package up an I/O request on a vnode into a uio and do it. 620 * 621 * MPSAFE 622 */ 623 int 624 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 625 off_t offset, enum uio_seg segflg, int ioflg, 626 struct ucred *cred, int *aresid) 627 { 628 struct uio auio; 629 struct iovec aiov; 630 int error; 631 632 if ((ioflg & IO_NODELOCKED) == 0) 633 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 634 auio.uio_iov = &aiov; 635 auio.uio_iovcnt = 1; 636 aiov.iov_base = base; 637 aiov.iov_len = len; 638 auio.uio_resid = len; 639 auio.uio_offset = offset; 640 auio.uio_segflg = segflg; 641 auio.uio_rw = rw; 642 auio.uio_td = curthread; 643 if (rw == UIO_READ) { 644 error = VOP_READ(vp, &auio, ioflg, cred); 645 } else { 646 error = VOP_WRITE(vp, &auio, ioflg, cred); 647 } 648 if (aresid) 649 *aresid = auio.uio_resid; 650 else 651 if (auio.uio_resid && error == 0) 652 error = EIO; 653 if ((ioflg & IO_NODELOCKED) == 0) 654 vn_unlock(vp); 655 return (error); 656 } 657 658 /* 659 * Package up an I/O request on a vnode into a uio and do it. The I/O 660 * request is split up into smaller chunks and we try to avoid saturating 661 * the buffer cache while potentially holding a vnode locked, so we 662 * check bwillwrite() before calling vn_rdwr(). We also call lwkt_user_yield() 663 * to give other processes a chance to lock the vnode (either other processes 664 * core'ing the same binary, or unrelated processes scanning the directory). 665 * 666 * MPSAFE 667 */ 668 int 669 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, caddr_t base, int len, 670 off_t offset, enum uio_seg segflg, int ioflg, 671 struct ucred *cred, int *aresid) 672 { 673 int error = 0; 674 675 do { 676 int chunk; 677 678 /* 679 * Force `offset' to a multiple of MAXBSIZE except possibly 680 * for the first chunk, so that filesystems only need to 681 * write full blocks except possibly for the first and last 682 * chunks. 683 */ 684 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 685 686 if (chunk > len) 687 chunk = len; 688 if (vp->v_type == VREG && (ioflg & IO_RECURSE) == 0) { 689 switch(rw) { 690 case UIO_READ: 691 bwillread(chunk); 692 break; 693 case UIO_WRITE: 694 bwillwrite(chunk); 695 break; 696 } 697 } 698 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 699 ioflg, cred, aresid); 700 len -= chunk; /* aresid calc already includes length */ 701 if (error) 702 break; 703 offset += chunk; 704 base += chunk; 705 lwkt_user_yield(); 706 } while (len); 707 if (aresid) 708 *aresid += len; 709 return (error); 710 } 711 712 /* 713 * File pointers can no longer get ripped up by revoke so 714 * we don't need to lock access to the vp. 715 * 716 * f_offset updates are not guaranteed against multiple readers 717 */ 718 static int 719 vn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 720 { 721 struct vnode *vp; 722 int error, ioflag; 723 724 KASSERT(uio->uio_td == curthread, 725 ("uio_td %p is not td %p", uio->uio_td, curthread)); 726 vp = (struct vnode *)fp->f_data; 727 728 ioflag = 0; 729 if (flags & O_FBLOCKING) { 730 /* ioflag &= ~IO_NDELAY; */ 731 } else if (flags & O_FNONBLOCKING) { 732 ioflag |= IO_NDELAY; 733 } else if (fp->f_flag & FNONBLOCK) { 734 ioflag |= IO_NDELAY; 735 } 736 if (fp->f_flag & O_DIRECT) { 737 ioflag |= IO_DIRECT; 738 } 739 if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0) 740 uio->uio_offset = vn_get_fpf_offset(fp); 741 vn_lock(vp, LK_SHARED | LK_RETRY); 742 ioflag |= sequential_heuristic(uio, fp); 743 744 error = VOP_READ_FP(vp, uio, ioflag, cred, fp); 745 fp->f_nextoff = uio->uio_offset; 746 vn_unlock(vp); 747 if ((flags & O_FOFFSET) == 0 && (vp->v_flag & VNOTSEEKABLE) == 0) 748 vn_set_fpf_offset(fp, uio->uio_offset); 749 return (error); 750 } 751 752 /* 753 * MPSAFE 754 */ 755 static int 756 vn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 757 { 758 struct vnode *vp; 759 int error, ioflag; 760 761 KASSERT(uio->uio_td == curthread, 762 ("uio_td %p is not p %p", uio->uio_td, curthread)); 763 vp = (struct vnode *)fp->f_data; 764 765 ioflag = IO_UNIT; 766 if (vp->v_type == VREG && 767 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 768 ioflag |= IO_APPEND; 769 } 770 771 if (flags & O_FBLOCKING) { 772 /* ioflag &= ~IO_NDELAY; */ 773 } else if (flags & O_FNONBLOCKING) { 774 ioflag |= IO_NDELAY; 775 } else if (fp->f_flag & FNONBLOCK) { 776 ioflag |= IO_NDELAY; 777 } 778 if (fp->f_flag & O_DIRECT) { 779 ioflag |= IO_DIRECT; 780 } 781 if (flags & O_FASYNCWRITE) { 782 /* ioflag &= ~IO_SYNC; */ 783 } else if (flags & O_FSYNCWRITE) { 784 ioflag |= IO_SYNC; 785 } else if (fp->f_flag & O_FSYNC) { 786 ioflag |= IO_SYNC; 787 } 788 789 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 790 ioflag |= IO_SYNC; 791 if ((flags & O_FOFFSET) == 0) 792 uio->uio_offset = vn_get_fpf_offset(fp); 793 if (vp->v_mount) 794 VFS_MODIFYING(vp->v_mount); 795 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 796 ioflag |= sequential_heuristic(uio, fp); 797 error = VOP_WRITE_FP(vp, uio, ioflag, cred, fp); 798 fp->f_nextoff = uio->uio_offset; 799 vn_unlock(vp); 800 if ((flags & O_FOFFSET) == 0) 801 vn_set_fpf_offset(fp, uio->uio_offset); 802 return (error); 803 } 804 805 /* 806 * MPSAFE 807 */ 808 static int 809 vn_statfile(struct file *fp, struct stat *sb, struct ucred *cred) 810 { 811 struct vnode *vp; 812 int error; 813 814 vp = (struct vnode *)fp->f_data; 815 error = vn_stat(vp, sb, cred); 816 return (error); 817 } 818 819 /* 820 * MPSAFE 821 */ 822 int 823 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *cred) 824 { 825 struct vattr vattr; 826 struct vattr *vap; 827 int error; 828 u_short mode; 829 cdev_t dev; 830 831 /* 832 * vp already has a ref and is validated, can call unlocked. 833 */ 834 vap = &vattr; 835 error = VOP_GETATTR(vp, vap); 836 if (error) 837 return (error); 838 839 /* 840 * Zero the spare stat fields 841 */ 842 sb->st_lspare = 0; 843 sb->st_qspare2 = 0; 844 845 /* 846 * Copy from vattr table 847 */ 848 if (vap->va_fsid != VNOVAL) 849 sb->st_dev = vap->va_fsid; 850 else 851 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 852 sb->st_ino = vap->va_fileid; 853 mode = vap->va_mode; 854 switch (vap->va_type) { 855 case VREG: 856 mode |= S_IFREG; 857 break; 858 case VDATABASE: 859 mode |= S_IFDB; 860 break; 861 case VDIR: 862 mode |= S_IFDIR; 863 break; 864 case VBLK: 865 mode |= S_IFBLK; 866 break; 867 case VCHR: 868 mode |= S_IFCHR; 869 break; 870 case VLNK: 871 mode |= S_IFLNK; 872 /* This is a cosmetic change, symlinks do not have a mode. */ 873 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) 874 sb->st_mode &= ~ACCESSPERMS; /* 0000 */ 875 else 876 sb->st_mode |= ACCESSPERMS; /* 0777 */ 877 break; 878 case VSOCK: 879 mode |= S_IFSOCK; 880 break; 881 case VFIFO: 882 mode |= S_IFIFO; 883 break; 884 default: 885 return (EBADF); 886 } 887 sb->st_mode = mode; 888 if (vap->va_nlink > (nlink_t)-1) 889 sb->st_nlink = (nlink_t)-1; 890 else 891 sb->st_nlink = vap->va_nlink; 892 sb->st_uid = vap->va_uid; 893 sb->st_gid = vap->va_gid; 894 sb->st_rdev = devid_from_dev(vp->v_rdev); 895 sb->st_size = vap->va_size; 896 sb->st_atimespec = vap->va_atime; 897 sb->st_mtimespec = vap->va_mtime; 898 sb->st_ctimespec = vap->va_ctime; 899 900 /* 901 * A VCHR and VBLK device may track the last access and last modified 902 * time independantly of the filesystem. This is particularly true 903 * because device read and write calls may bypass the filesystem. 904 */ 905 if (vp->v_type == VCHR || vp->v_type == VBLK) { 906 dev = vp->v_rdev; 907 if (dev != NULL) { 908 if (dev->si_lastread) { 909 sb->st_atimespec.tv_sec = time_second + 910 (dev->si_lastread - 911 time_uptime); 912 sb->st_atimespec.tv_nsec = 0; 913 } 914 if (dev->si_lastwrite) { 915 sb->st_mtimespec.tv_sec = time_second + 916 (dev->si_lastwrite - 917 time_uptime); 918 sb->st_mtimespec.tv_nsec = 0; 919 } 920 } 921 } 922 923 /* 924 * According to www.opengroup.org, the meaning of st_blksize is 925 * "a filesystem-specific preferred I/O block size for this 926 * object. In some filesystem types, this may vary from file 927 * to file" 928 * Default to PAGE_SIZE after much discussion. 929 */ 930 931 if (vap->va_type == VREG) { 932 sb->st_blksize = vap->va_blocksize; 933 } else if (vn_isdisk(vp, NULL)) { 934 /* 935 * XXX this is broken. If the device is not yet open (aka 936 * stat() call, aka v_rdev == NULL), how are we supposed 937 * to get a valid block size out of it? 938 */ 939 dev = vp->v_rdev; 940 941 sb->st_blksize = dev->si_bsize_best; 942 if (sb->st_blksize < dev->si_bsize_phys) 943 sb->st_blksize = dev->si_bsize_phys; 944 if (sb->st_blksize < BLKDEV_IOSIZE) 945 sb->st_blksize = BLKDEV_IOSIZE; 946 } else { 947 sb->st_blksize = PAGE_SIZE; 948 } 949 950 sb->st_flags = vap->va_flags; 951 952 error = caps_priv_check(cred, SYSCAP_NOVFS_GENERATION); 953 if (error) 954 sb->st_gen = 0; 955 else 956 sb->st_gen = (u_int32_t)vap->va_gen; 957 958 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 959 960 /* 961 * This is for ABI compatibility <= 5.7 (for ABI change made in 962 * 5.7 master). 963 */ 964 sb->__old_st_blksize = sb->st_blksize; 965 966 return (0); 967 } 968 969 /* 970 * MPALMOSTSAFE - acquires mplock 971 */ 972 static int 973 vn_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *ucred, 974 struct sysmsg *msg) 975 { 976 struct vnode *vp = ((struct vnode *)fp->f_data); 977 struct vnode *ovp; 978 struct vattr vattr; 979 int error; 980 off_t size; 981 982 switch (vp->v_type) { 983 case VREG: 984 case VDIR: 985 if (com == FIONREAD) { 986 error = VOP_GETATTR(vp, &vattr); 987 if (error) 988 break; 989 size = vattr.va_size; 990 if ((vp->v_flag & VNOTSEEKABLE) == 0) 991 size -= vn_poll_fpf_offset(fp); 992 if (size > 0x7FFFFFFF) 993 size = 0x7FFFFFFF; 994 *(int *)data = size; 995 error = 0; 996 break; 997 } 998 if (com == FIOASYNC) { /* XXX */ 999 error = 0; /* XXX */ 1000 break; 1001 } 1002 /* fall into ... */ 1003 default: 1004 #if 0 1005 return (ENOTTY); 1006 #endif 1007 case VFIFO: 1008 case VCHR: 1009 case VBLK: 1010 if (com == FIODTYPE) { 1011 if (vp->v_type != VCHR && vp->v_type != VBLK) { 1012 error = ENOTTY; 1013 break; 1014 } 1015 *(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK; 1016 error = 0; 1017 break; 1018 } 1019 error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred, msg); 1020 if (error == 0 && com == TIOCSCTTY) { 1021 struct proc *p = curthread->td_proc; 1022 struct session *sess; 1023 1024 if (p == NULL) { 1025 error = ENOTTY; 1026 break; 1027 } 1028 1029 get_mplock(); 1030 sess = p->p_session; 1031 /* Do nothing if reassigning same control tty */ 1032 if (sess->s_ttyvp == vp) { 1033 error = 0; 1034 rel_mplock(); 1035 break; 1036 } 1037 1038 /* Get rid of reference to old control tty */ 1039 ovp = sess->s_ttyvp; 1040 vref(vp); 1041 sess->s_ttyvp = vp; 1042 if (ovp) 1043 vrele(ovp); 1044 rel_mplock(); 1045 } 1046 break; 1047 } 1048 return (error); 1049 } 1050 1051 /* 1052 * Obtain the requested vnode lock 1053 * 1054 * LK_RETRY Automatically retry on timeout 1055 * LK_FAILRECLAIM Fail if the vnode is being reclaimed 1056 * 1057 * Failures will occur if the vnode is undergoing recyclement, but not 1058 * all callers expect that the function will fail so the caller must pass 1059 * LK_FAILOK if it wants to process an error code. 1060 * 1061 * Errors can occur for other reasons if you pass in other LK_ flags, 1062 * regardless of whether you pass in LK_FAILRECLAIM 1063 */ 1064 int 1065 vn_lock(struct vnode *vp, int flags) 1066 { 1067 int error; 1068 1069 do { 1070 error = lockmgr(&vp->v_lock, flags); 1071 if (error == 0) 1072 break; 1073 } while (flags & LK_RETRY); 1074 1075 /* 1076 * Because we (had better!) have a ref on the vnode, once it 1077 * goes to VRECLAIMED state it will not be recycled until all 1078 * refs go away. So we can just check the flag. 1079 */ 1080 if (error == 0 && (vp->v_flag & VRECLAIMED)) { 1081 if (flags & LK_FAILRECLAIM) { 1082 lockmgr(&vp->v_lock, LK_RELEASE); 1083 error = ENOENT; 1084 } 1085 } 1086 return (error); 1087 } 1088 1089 int 1090 vn_relock(struct vnode *vp, int flags) 1091 { 1092 int error; 1093 1094 do { 1095 error = lockmgr(&vp->v_lock, flags); 1096 if (error == 0) 1097 break; 1098 } while (flags & LK_RETRY); 1099 1100 return error; 1101 } 1102 1103 #ifdef DEBUG_VN_UNLOCK 1104 1105 void 1106 debug_vn_unlock(struct vnode *vp, const char *filename, int line) 1107 { 1108 kprintf("vn_unlock from %s:%d\n", filename, line); 1109 lockmgr(&vp->v_lock, LK_RELEASE); 1110 } 1111 1112 #else 1113 1114 void 1115 vn_unlock(struct vnode *vp) 1116 { 1117 lockmgr(&vp->v_lock, LK_RELEASE); 1118 } 1119 1120 #endif 1121 1122 /* 1123 * MPSAFE 1124 */ 1125 int 1126 vn_islocked(struct vnode *vp) 1127 { 1128 return (lockstatus(&vp->v_lock, curthread)); 1129 } 1130 1131 /* 1132 * Return the lock status of a vnode and unlock the vnode 1133 * if we owned the lock. This is not a boolean, if the 1134 * caller cares what the lock status is the caller must 1135 * check the various possible values. 1136 * 1137 * This only unlocks exclusive locks held by the caller, 1138 * it will NOT unlock shared locks (there is no way to 1139 * tell who the shared lock belongs to). 1140 * 1141 * MPSAFE 1142 */ 1143 int 1144 vn_islocked_unlock(struct vnode *vp) 1145 { 1146 int vpls; 1147 1148 vpls = lockstatus(&vp->v_lock, curthread); 1149 if (vpls == LK_EXCLUSIVE) 1150 lockmgr(&vp->v_lock, LK_RELEASE); 1151 return(vpls); 1152 } 1153 1154 /* 1155 * Restore a vnode lock that we previously released via 1156 * vn_islocked_unlock(). This is a NOP if we did not 1157 * own the original lock. 1158 * 1159 * MPSAFE 1160 */ 1161 void 1162 vn_islocked_relock(struct vnode *vp, int vpls) 1163 { 1164 int error; 1165 1166 if (vpls == LK_EXCLUSIVE) 1167 error = lockmgr(&vp->v_lock, vpls); 1168 } 1169 1170 /* 1171 * MPSAFE 1172 */ 1173 static int 1174 vn_closefile(struct file *fp) 1175 { 1176 int error; 1177 1178 fp->f_ops = &badfileops; 1179 error = vn_close(((struct vnode *)fp->f_data), fp->f_flag, fp); 1180 return (error); 1181 } 1182 1183 /* 1184 * MPSAFE 1185 */ 1186 static int 1187 vn_kqfilter(struct file *fp, struct knote *kn) 1188 { 1189 int error; 1190 1191 error = VOP_KQFILTER(((struct vnode *)fp->f_data), kn); 1192 return (error); 1193 } 1194