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