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 * 4. 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 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/fcntl.h> 43 #include <sys/file.h> 44 #include <sys/kdb.h> 45 #include <sys/stat.h> 46 #include <sys/priv.h> 47 #include <sys/proc.h> 48 #include <sys/limits.h> 49 #include <sys/lock.h> 50 #include <sys/mount.h> 51 #include <sys/mutex.h> 52 #include <sys/namei.h> 53 #include <sys/vnode.h> 54 #include <sys/bio.h> 55 #include <sys/buf.h> 56 #include <sys/filio.h> 57 #include <sys/resourcevar.h> 58 #include <sys/sx.h> 59 #include <sys/ttycom.h> 60 #include <sys/conf.h> 61 #include <sys/syslog.h> 62 #include <sys/unistd.h> 63 64 #include <security/mac/mac_framework.h> 65 66 static fo_rdwr_t vn_read; 67 static fo_rdwr_t vn_write; 68 static fo_truncate_t vn_truncate; 69 static fo_ioctl_t vn_ioctl; 70 static fo_poll_t vn_poll; 71 static fo_kqfilter_t vn_kqfilter; 72 static fo_stat_t vn_statfile; 73 static fo_close_t vn_closefile; 74 75 struct fileops vnops = { 76 .fo_read = vn_read, 77 .fo_write = vn_write, 78 .fo_truncate = vn_truncate, 79 .fo_ioctl = vn_ioctl, 80 .fo_poll = vn_poll, 81 .fo_kqfilter = vn_kqfilter, 82 .fo_stat = vn_statfile, 83 .fo_close = vn_closefile, 84 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 85 }; 86 87 int 88 vn_open(ndp, flagp, cmode, fp) 89 struct nameidata *ndp; 90 int *flagp, cmode; 91 struct file *fp; 92 { 93 struct thread *td = ndp->ni_cnd.cn_thread; 94 95 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp)); 96 } 97 98 /* 99 * Common code for vnode open operations. 100 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine. 101 * 102 * Note that this does NOT free nameidata for the successful case, 103 * due to the NDINIT being done elsewhere. 104 */ 105 int 106 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, 107 struct ucred *cred, struct file *fp) 108 { 109 struct vnode *vp; 110 struct mount *mp; 111 struct thread *td = ndp->ni_cnd.cn_thread; 112 struct vattr vat; 113 struct vattr *vap = &vat; 114 int fmode, error; 115 accmode_t accmode; 116 int vfslocked, mpsafe; 117 118 mpsafe = ndp->ni_cnd.cn_flags & MPSAFE; 119 restart: 120 vfslocked = 0; 121 fmode = *flagp; 122 if (fmode & O_CREAT) { 123 ndp->ni_cnd.cn_nameiop = CREATE; 124 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | 125 MPSAFE; 126 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 127 ndp->ni_cnd.cn_flags |= FOLLOW; 128 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 129 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 130 bwillwrite(); 131 if ((error = namei(ndp)) != 0) 132 return (error); 133 vfslocked = NDHASGIANT(ndp); 134 if (!mpsafe) 135 ndp->ni_cnd.cn_flags &= ~MPSAFE; 136 if (ndp->ni_vp == NULL) { 137 VATTR_NULL(vap); 138 vap->va_type = VREG; 139 vap->va_mode = cmode; 140 if (fmode & O_EXCL) 141 vap->va_vaflags |= VA_EXCLUSIVE; 142 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) { 143 NDFREE(ndp, NDF_ONLY_PNBUF); 144 vput(ndp->ni_dvp); 145 VFS_UNLOCK_GIANT(vfslocked); 146 if ((error = vn_start_write(NULL, &mp, 147 V_XSLEEP | PCATCH)) != 0) 148 return (error); 149 goto restart; 150 } 151 #ifdef MAC 152 error = mac_vnode_check_create(cred, ndp->ni_dvp, 153 &ndp->ni_cnd, vap); 154 if (error == 0) 155 #endif 156 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 157 &ndp->ni_cnd, vap); 158 vput(ndp->ni_dvp); 159 vn_finished_write(mp); 160 if (error) { 161 VFS_UNLOCK_GIANT(vfslocked); 162 NDFREE(ndp, NDF_ONLY_PNBUF); 163 return (error); 164 } 165 fmode &= ~O_TRUNC; 166 vp = ndp->ni_vp; 167 } else { 168 if (ndp->ni_dvp == ndp->ni_vp) 169 vrele(ndp->ni_dvp); 170 else 171 vput(ndp->ni_dvp); 172 ndp->ni_dvp = NULL; 173 vp = ndp->ni_vp; 174 if (fmode & O_EXCL) { 175 error = EEXIST; 176 goto bad; 177 } 178 fmode &= ~O_CREAT; 179 } 180 } else { 181 ndp->ni_cnd.cn_nameiop = LOOKUP; 182 ndp->ni_cnd.cn_flags = ISOPEN | 183 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | 184 LOCKLEAF | MPSAFE; 185 if (!(fmode & FWRITE)) 186 ndp->ni_cnd.cn_flags |= LOCKSHARED; 187 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 188 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 189 if ((error = namei(ndp)) != 0) 190 return (error); 191 if (!mpsafe) 192 ndp->ni_cnd.cn_flags &= ~MPSAFE; 193 vfslocked = NDHASGIANT(ndp); 194 vp = ndp->ni_vp; 195 } 196 if (vp->v_type == VLNK) { 197 error = EMLINK; 198 goto bad; 199 } 200 if (vp->v_type == VSOCK) { 201 error = EOPNOTSUPP; 202 goto bad; 203 } 204 if (vp->v_type != VDIR && fmode & O_DIRECTORY) { 205 error = ENOTDIR; 206 goto bad; 207 } 208 accmode = 0; 209 if (fmode & (FWRITE | O_TRUNC)) { 210 if (vp->v_type == VDIR) { 211 error = EISDIR; 212 goto bad; 213 } 214 accmode |= VWRITE; 215 } 216 if (fmode & FREAD) 217 accmode |= VREAD; 218 if (fmode & FEXEC) 219 accmode |= VEXEC; 220 if ((fmode & O_APPEND) && (fmode & FWRITE)) 221 accmode |= VAPPEND; 222 #ifdef MAC 223 error = mac_vnode_check_open(cred, vp, accmode); 224 if (error) 225 goto bad; 226 #endif 227 if ((fmode & O_CREAT) == 0) { 228 if (accmode & VWRITE) { 229 error = vn_writechk(vp); 230 if (error) 231 goto bad; 232 } 233 if (accmode) { 234 error = VOP_ACCESS(vp, accmode, cred, td); 235 if (error) 236 goto bad; 237 } 238 } 239 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0) 240 goto bad; 241 242 if (fmode & FWRITE) 243 vp->v_writecount++; 244 *flagp = fmode; 245 ASSERT_VOP_LOCKED(vp, "vn_open_cred"); 246 if (!mpsafe) 247 VFS_UNLOCK_GIANT(vfslocked); 248 return (0); 249 bad: 250 NDFREE(ndp, NDF_ONLY_PNBUF); 251 vput(vp); 252 VFS_UNLOCK_GIANT(vfslocked); 253 *flagp = fmode; 254 ndp->ni_vp = NULL; 255 return (error); 256 } 257 258 /* 259 * Check for write permissions on the specified vnode. 260 * Prototype text segments cannot be written. 261 */ 262 int 263 vn_writechk(vp) 264 register struct vnode *vp; 265 { 266 267 ASSERT_VOP_LOCKED(vp, "vn_writechk"); 268 /* 269 * If there's shared text associated with 270 * the vnode, try to free it up once. If 271 * we fail, we can't allow writing. 272 */ 273 if (vp->v_vflag & VV_TEXT) 274 return (ETXTBSY); 275 276 return (0); 277 } 278 279 /* 280 * Vnode close call 281 */ 282 int 283 vn_close(vp, flags, file_cred, td) 284 register struct vnode *vp; 285 int flags; 286 struct ucred *file_cred; 287 struct thread *td; 288 { 289 struct mount *mp; 290 int error, lock_flags; 291 292 if (!(flags & FWRITE) && vp->v_mount != NULL && 293 vp->v_mount->mnt_kern_flag & MNTK_EXTENDED_SHARED) 294 lock_flags = LK_SHARED; 295 else 296 lock_flags = LK_EXCLUSIVE; 297 298 VFS_ASSERT_GIANT(vp->v_mount); 299 300 vn_start_write(vp, &mp, V_WAIT); 301 vn_lock(vp, lock_flags | LK_RETRY); 302 if (flags & FWRITE) { 303 VNASSERT(vp->v_writecount > 0, vp, 304 ("vn_close: negative writecount")); 305 vp->v_writecount--; 306 } 307 error = VOP_CLOSE(vp, flags, file_cred, td); 308 vput(vp); 309 vn_finished_write(mp); 310 return (error); 311 } 312 313 /* 314 * Heuristic to detect sequential operation. 315 */ 316 static int 317 sequential_heuristic(struct uio *uio, struct file *fp) 318 { 319 320 if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD) 321 return (fp->f_seqcount << IO_SEQSHIFT); 322 323 /* 324 * Offset 0 is handled specially. open() sets f_seqcount to 1 so 325 * that the first I/O is normally considered to be slightly 326 * sequential. Seeking to offset 0 doesn't change sequentiality 327 * unless previous seeks have reduced f_seqcount to 0, in which 328 * case offset 0 is not special. 329 */ 330 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 331 uio->uio_offset == fp->f_nextoff) { 332 /* 333 * f_seqcount is in units of fixed-size blocks so that it 334 * depends mainly on the amount of sequential I/O and not 335 * much on the number of sequential I/O's. The fixed size 336 * of 16384 is hard-coded here since it is (not quite) just 337 * a magic size that works well here. This size is more 338 * closely related to the best I/O size for real disks than 339 * to any block size used by software. 340 */ 341 fp->f_seqcount += howmany(uio->uio_resid, 16384); 342 if (fp->f_seqcount > IO_SEQMAX) 343 fp->f_seqcount = IO_SEQMAX; 344 return (fp->f_seqcount << IO_SEQSHIFT); 345 } 346 347 /* Not sequential. Quickly draw-down sequentiality. */ 348 if (fp->f_seqcount > 1) 349 fp->f_seqcount = 1; 350 else 351 fp->f_seqcount = 0; 352 return (0); 353 } 354 355 /* 356 * Package up an I/O request on a vnode into a uio and do it. 357 */ 358 int 359 vn_rdwr(rw, vp, base, len, offset, segflg, ioflg, active_cred, file_cred, 360 aresid, td) 361 enum uio_rw rw; 362 struct vnode *vp; 363 void *base; 364 int len; 365 off_t offset; 366 enum uio_seg segflg; 367 int ioflg; 368 struct ucred *active_cred; 369 struct ucred *file_cred; 370 int *aresid; 371 struct thread *td; 372 { 373 struct uio auio; 374 struct iovec aiov; 375 struct mount *mp; 376 struct ucred *cred; 377 int error, lock_flags; 378 379 VFS_ASSERT_GIANT(vp->v_mount); 380 381 if ((ioflg & IO_NODELOCKED) == 0) { 382 mp = NULL; 383 if (rw == UIO_WRITE) { 384 if (vp->v_type != VCHR && 385 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) 386 != 0) 387 return (error); 388 if (MNT_SHARED_WRITES(mp) || 389 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) { 390 lock_flags = LK_SHARED; 391 } else { 392 lock_flags = LK_EXCLUSIVE; 393 } 394 vn_lock(vp, lock_flags | LK_RETRY); 395 } else 396 vn_lock(vp, LK_SHARED | LK_RETRY); 397 398 } 399 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 400 auio.uio_iov = &aiov; 401 auio.uio_iovcnt = 1; 402 aiov.iov_base = base; 403 aiov.iov_len = len; 404 auio.uio_resid = len; 405 auio.uio_offset = offset; 406 auio.uio_segflg = segflg; 407 auio.uio_rw = rw; 408 auio.uio_td = td; 409 error = 0; 410 #ifdef MAC 411 if ((ioflg & IO_NOMACCHECK) == 0) { 412 if (rw == UIO_READ) 413 error = mac_vnode_check_read(active_cred, file_cred, 414 vp); 415 else 416 error = mac_vnode_check_write(active_cred, file_cred, 417 vp); 418 } 419 #endif 420 if (error == 0) { 421 if (file_cred) 422 cred = file_cred; 423 else 424 cred = active_cred; 425 if (rw == UIO_READ) 426 error = VOP_READ(vp, &auio, ioflg, cred); 427 else 428 error = VOP_WRITE(vp, &auio, ioflg, cred); 429 } 430 if (aresid) 431 *aresid = auio.uio_resid; 432 else 433 if (auio.uio_resid && error == 0) 434 error = EIO; 435 if ((ioflg & IO_NODELOCKED) == 0) { 436 if (rw == UIO_WRITE && vp->v_type != VCHR) 437 vn_finished_write(mp); 438 VOP_UNLOCK(vp, 0); 439 } 440 return (error); 441 } 442 443 /* 444 * Package up an I/O request on a vnode into a uio and do it. The I/O 445 * request is split up into smaller chunks and we try to avoid saturating 446 * the buffer cache while potentially holding a vnode locked, so we 447 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield() 448 * to give other processes a chance to lock the vnode (either other processes 449 * core'ing the same binary, or unrelated processes scanning the directory). 450 */ 451 int 452 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred, 453 file_cred, aresid, td) 454 enum uio_rw rw; 455 struct vnode *vp; 456 void *base; 457 size_t len; 458 off_t offset; 459 enum uio_seg segflg; 460 int ioflg; 461 struct ucred *active_cred; 462 struct ucred *file_cred; 463 size_t *aresid; 464 struct thread *td; 465 { 466 int error = 0; 467 int iaresid; 468 469 VFS_ASSERT_GIANT(vp->v_mount); 470 471 do { 472 int chunk; 473 474 /* 475 * Force `offset' to a multiple of MAXBSIZE except possibly 476 * for the first chunk, so that filesystems only need to 477 * write full blocks except possibly for the first and last 478 * chunks. 479 */ 480 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 481 482 if (chunk > len) 483 chunk = len; 484 if (rw != UIO_READ && vp->v_type == VREG) 485 bwillwrite(); 486 iaresid = 0; 487 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 488 ioflg, active_cred, file_cred, &iaresid, td); 489 len -= chunk; /* aresid calc already includes length */ 490 if (error) 491 break; 492 offset += chunk; 493 base = (char *)base + chunk; 494 kern_yield(curthread->td_user_pri); 495 } while (len); 496 if (aresid) 497 *aresid = len + iaresid; 498 return (error); 499 } 500 501 /* 502 * File table vnode read routine. 503 */ 504 static int 505 vn_read(fp, uio, active_cred, flags, td) 506 struct file *fp; 507 struct uio *uio; 508 struct ucred *active_cred; 509 int flags; 510 struct thread *td; 511 { 512 struct vnode *vp; 513 int error, ioflag; 514 struct mtx *mtxp; 515 int vfslocked; 516 517 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", 518 uio->uio_td, td)); 519 mtxp = NULL; 520 vp = fp->f_vnode; 521 ioflag = 0; 522 if (fp->f_flag & FNONBLOCK) 523 ioflag |= IO_NDELAY; 524 if (fp->f_flag & O_DIRECT) 525 ioflag |= IO_DIRECT; 526 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 527 /* 528 * According to McKusick the vn lock was protecting f_offset here. 529 * It is now protected by the FOFFSET_LOCKED flag. 530 */ 531 if ((flags & FOF_OFFSET) == 0) { 532 mtxp = mtx_pool_find(mtxpool_sleep, fp); 533 mtx_lock(mtxp); 534 while(fp->f_vnread_flags & FOFFSET_LOCKED) { 535 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING; 536 msleep(&fp->f_vnread_flags, mtxp, PUSER -1, 537 "vnread offlock", 0); 538 } 539 fp->f_vnread_flags |= FOFFSET_LOCKED; 540 mtx_unlock(mtxp); 541 vn_lock(vp, LK_SHARED | LK_RETRY); 542 uio->uio_offset = fp->f_offset; 543 } else 544 vn_lock(vp, LK_SHARED | LK_RETRY); 545 546 ioflag |= sequential_heuristic(uio, fp); 547 548 #ifdef MAC 549 error = mac_vnode_check_read(active_cred, fp->f_cred, vp); 550 if (error == 0) 551 #endif 552 error = VOP_READ(vp, uio, ioflag, fp->f_cred); 553 if ((flags & FOF_OFFSET) == 0) { 554 fp->f_offset = uio->uio_offset; 555 mtx_lock(mtxp); 556 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING) 557 wakeup(&fp->f_vnread_flags); 558 fp->f_vnread_flags = 0; 559 mtx_unlock(mtxp); 560 } 561 fp->f_nextoff = uio->uio_offset; 562 VOP_UNLOCK(vp, 0); 563 VFS_UNLOCK_GIANT(vfslocked); 564 return (error); 565 } 566 567 /* 568 * File table vnode write routine. 569 */ 570 static int 571 vn_write(fp, uio, active_cred, flags, td) 572 struct file *fp; 573 struct uio *uio; 574 struct ucred *active_cred; 575 int flags; 576 struct thread *td; 577 { 578 struct vnode *vp; 579 struct mount *mp; 580 int error, ioflag, lock_flags; 581 int vfslocked; 582 583 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", 584 uio->uio_td, td)); 585 vp = fp->f_vnode; 586 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 587 if (vp->v_type == VREG) 588 bwillwrite(); 589 ioflag = IO_UNIT; 590 if (vp->v_type == VREG && (fp->f_flag & O_APPEND)) 591 ioflag |= IO_APPEND; 592 if (fp->f_flag & FNONBLOCK) 593 ioflag |= IO_NDELAY; 594 if (fp->f_flag & O_DIRECT) 595 ioflag |= IO_DIRECT; 596 if ((fp->f_flag & O_FSYNC) || 597 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) 598 ioflag |= IO_SYNC; 599 mp = NULL; 600 if (vp->v_type != VCHR && 601 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) 602 goto unlock; 603 604 if ((MNT_SHARED_WRITES(mp) || 605 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) && 606 (flags & FOF_OFFSET) != 0) { 607 lock_flags = LK_SHARED; 608 } else { 609 lock_flags = LK_EXCLUSIVE; 610 } 611 612 vn_lock(vp, lock_flags | LK_RETRY); 613 if ((flags & FOF_OFFSET) == 0) 614 uio->uio_offset = fp->f_offset; 615 ioflag |= sequential_heuristic(uio, fp); 616 #ifdef MAC 617 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 618 if (error == 0) 619 #endif 620 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred); 621 if ((flags & FOF_OFFSET) == 0) 622 fp->f_offset = uio->uio_offset; 623 fp->f_nextoff = uio->uio_offset; 624 VOP_UNLOCK(vp, 0); 625 if (vp->v_type != VCHR) 626 vn_finished_write(mp); 627 unlock: 628 VFS_UNLOCK_GIANT(vfslocked); 629 return (error); 630 } 631 632 /* 633 * File table truncate routine. 634 */ 635 static int 636 vn_truncate(fp, length, active_cred, td) 637 struct file *fp; 638 off_t length; 639 struct ucred *active_cred; 640 struct thread *td; 641 { 642 struct vattr vattr; 643 struct mount *mp; 644 struct vnode *vp; 645 int vfslocked; 646 int error; 647 648 vp = fp->f_vnode; 649 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 650 error = vn_start_write(vp, &mp, V_WAIT | PCATCH); 651 if (error) { 652 VFS_UNLOCK_GIANT(vfslocked); 653 return (error); 654 } 655 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 656 if (vp->v_type == VDIR) { 657 error = EISDIR; 658 goto out; 659 } 660 #ifdef MAC 661 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 662 if (error) 663 goto out; 664 #endif 665 error = vn_writechk(vp); 666 if (error == 0) { 667 VATTR_NULL(&vattr); 668 vattr.va_size = length; 669 error = VOP_SETATTR(vp, &vattr, fp->f_cred); 670 } 671 out: 672 VOP_UNLOCK(vp, 0); 673 vn_finished_write(mp); 674 VFS_UNLOCK_GIANT(vfslocked); 675 return (error); 676 } 677 678 /* 679 * File table vnode stat routine. 680 */ 681 static int 682 vn_statfile(fp, sb, active_cred, td) 683 struct file *fp; 684 struct stat *sb; 685 struct ucred *active_cred; 686 struct thread *td; 687 { 688 struct vnode *vp = fp->f_vnode; 689 int vfslocked; 690 int error; 691 692 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 693 vn_lock(vp, LK_SHARED | LK_RETRY); 694 error = vn_stat(vp, sb, active_cred, fp->f_cred, td); 695 VOP_UNLOCK(vp, 0); 696 VFS_UNLOCK_GIANT(vfslocked); 697 698 return (error); 699 } 700 701 /* 702 * Stat a vnode; implementation for the stat syscall 703 */ 704 int 705 vn_stat(vp, sb, active_cred, file_cred, td) 706 struct vnode *vp; 707 register struct stat *sb; 708 struct ucred *active_cred; 709 struct ucred *file_cred; 710 struct thread *td; 711 { 712 struct vattr vattr; 713 register struct vattr *vap; 714 int error; 715 u_short mode; 716 717 #ifdef MAC 718 error = mac_vnode_check_stat(active_cred, file_cred, vp); 719 if (error) 720 return (error); 721 #endif 722 723 vap = &vattr; 724 725 /* 726 * Initialize defaults for new and unusual fields, so that file 727 * systems which don't support these fields don't need to know 728 * about them. 729 */ 730 vap->va_birthtime.tv_sec = -1; 731 vap->va_birthtime.tv_nsec = 0; 732 vap->va_fsid = VNOVAL; 733 vap->va_rdev = NODEV; 734 735 error = VOP_GETATTR(vp, vap, active_cred); 736 if (error) 737 return (error); 738 739 /* 740 * Zero the spare stat fields 741 */ 742 bzero(sb, sizeof *sb); 743 744 /* 745 * Copy from vattr table 746 */ 747 if (vap->va_fsid != VNOVAL) 748 sb->st_dev = vap->va_fsid; 749 else 750 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 751 sb->st_ino = vap->va_fileid; 752 mode = vap->va_mode; 753 switch (vap->va_type) { 754 case VREG: 755 mode |= S_IFREG; 756 break; 757 case VDIR: 758 mode |= S_IFDIR; 759 break; 760 case VBLK: 761 mode |= S_IFBLK; 762 break; 763 case VCHR: 764 mode |= S_IFCHR; 765 break; 766 case VLNK: 767 mode |= S_IFLNK; 768 break; 769 case VSOCK: 770 mode |= S_IFSOCK; 771 break; 772 case VFIFO: 773 mode |= S_IFIFO; 774 break; 775 default: 776 return (EBADF); 777 }; 778 sb->st_mode = mode; 779 sb->st_nlink = vap->va_nlink; 780 sb->st_uid = vap->va_uid; 781 sb->st_gid = vap->va_gid; 782 sb->st_rdev = vap->va_rdev; 783 if (vap->va_size > OFF_MAX) 784 return (EOVERFLOW); 785 sb->st_size = vap->va_size; 786 sb->st_atim = vap->va_atime; 787 sb->st_mtim = vap->va_mtime; 788 sb->st_ctim = vap->va_ctime; 789 sb->st_birthtim = vap->va_birthtime; 790 791 /* 792 * According to www.opengroup.org, the meaning of st_blksize is 793 * "a filesystem-specific preferred I/O block size for this 794 * object. In some filesystem types, this may vary from file 795 * to file" 796 * Use miminum/default of PAGE_SIZE (e.g. for VCHR). 797 */ 798 799 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize); 800 801 sb->st_flags = vap->va_flags; 802 if (priv_check(td, PRIV_VFS_GENERATION)) 803 sb->st_gen = 0; 804 else 805 sb->st_gen = vap->va_gen; 806 807 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 808 return (0); 809 } 810 811 /* 812 * File table vnode ioctl routine. 813 */ 814 static int 815 vn_ioctl(fp, com, data, active_cred, td) 816 struct file *fp; 817 u_long com; 818 void *data; 819 struct ucred *active_cred; 820 struct thread *td; 821 { 822 struct vnode *vp = fp->f_vnode; 823 struct vattr vattr; 824 int vfslocked; 825 int error; 826 827 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 828 error = ENOTTY; 829 switch (vp->v_type) { 830 case VREG: 831 case VDIR: 832 if (com == FIONREAD) { 833 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 834 error = VOP_GETATTR(vp, &vattr, active_cred); 835 VOP_UNLOCK(vp, 0); 836 if (!error) 837 *(int *)data = vattr.va_size - fp->f_offset; 838 } 839 if (com == FIONBIO || com == FIOASYNC) /* XXX */ 840 error = 0; 841 else 842 error = VOP_IOCTL(vp, com, data, fp->f_flag, 843 active_cred, td); 844 break; 845 846 default: 847 break; 848 } 849 VFS_UNLOCK_GIANT(vfslocked); 850 return (error); 851 } 852 853 /* 854 * File table vnode poll routine. 855 */ 856 static int 857 vn_poll(fp, events, active_cred, td) 858 struct file *fp; 859 int events; 860 struct ucred *active_cred; 861 struct thread *td; 862 { 863 struct vnode *vp; 864 int vfslocked; 865 int error; 866 867 vp = fp->f_vnode; 868 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 869 #ifdef MAC 870 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 871 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp); 872 VOP_UNLOCK(vp, 0); 873 if (!error) 874 #endif 875 876 error = VOP_POLL(vp, events, fp->f_cred, td); 877 VFS_UNLOCK_GIANT(vfslocked); 878 return (error); 879 } 880 881 /* 882 * Acquire the requested lock and then check for validity. LK_RETRY 883 * permits vn_lock to return doomed vnodes. 884 */ 885 int 886 _vn_lock(struct vnode *vp, int flags, char *file, int line) 887 { 888 int error; 889 890 VNASSERT((flags & LK_TYPE_MASK) != 0, vp, 891 ("vn_lock called with no locktype.")); 892 do { 893 #ifdef DEBUG_VFS_LOCKS 894 KASSERT(vp->v_holdcnt != 0, 895 ("vn_lock %p: zero hold count", vp)); 896 #endif 897 error = VOP_LOCK1(vp, flags, file, line); 898 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */ 899 KASSERT((flags & LK_RETRY) == 0 || error == 0, 900 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)", 901 flags, error)); 902 /* 903 * Callers specify LK_RETRY if they wish to get dead vnodes. 904 * If RETRY is not set, we return ENOENT instead. 905 */ 906 if (error == 0 && vp->v_iflag & VI_DOOMED && 907 (flags & LK_RETRY) == 0) { 908 VOP_UNLOCK(vp, 0); 909 error = ENOENT; 910 break; 911 } 912 } while (flags & LK_RETRY && error != 0); 913 return (error); 914 } 915 916 /* 917 * File table vnode close routine. 918 */ 919 static int 920 vn_closefile(fp, td) 921 struct file *fp; 922 struct thread *td; 923 { 924 struct vnode *vp; 925 struct flock lf; 926 int vfslocked; 927 int error; 928 929 vp = fp->f_vnode; 930 931 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 932 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) { 933 lf.l_whence = SEEK_SET; 934 lf.l_start = 0; 935 lf.l_len = 0; 936 lf.l_type = F_UNLCK; 937 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); 938 } 939 940 fp->f_ops = &badfileops; 941 942 error = vn_close(vp, fp->f_flag, fp->f_cred, td); 943 VFS_UNLOCK_GIANT(vfslocked); 944 return (error); 945 } 946 947 /* 948 * Preparing to start a filesystem write operation. If the operation is 949 * permitted, then we bump the count of operations in progress and 950 * proceed. If a suspend request is in progress, we wait until the 951 * suspension is over, and then proceed. 952 */ 953 int 954 vn_start_write(vp, mpp, flags) 955 struct vnode *vp; 956 struct mount **mpp; 957 int flags; 958 { 959 struct mount *mp; 960 int error; 961 962 error = 0; 963 /* 964 * If a vnode is provided, get and return the mount point that 965 * to which it will write. 966 */ 967 if (vp != NULL) { 968 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 969 *mpp = NULL; 970 if (error != EOPNOTSUPP) 971 return (error); 972 return (0); 973 } 974 } 975 if ((mp = *mpp) == NULL) 976 return (0); 977 978 /* 979 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 980 * a vfs_ref(). 981 * As long as a vnode is not provided we need to acquire a 982 * refcount for the provided mountpoint too, in order to 983 * emulate a vfs_ref(). 984 */ 985 MNT_ILOCK(mp); 986 if (vp == NULL) 987 MNT_REF(mp); 988 989 /* 990 * Check on status of suspension. 991 */ 992 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 || 993 mp->mnt_susp_owner != curthread) { 994 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 995 if (flags & V_NOWAIT) { 996 error = EWOULDBLOCK; 997 goto unlock; 998 } 999 error = msleep(&mp->mnt_flag, MNT_MTX(mp), 1000 (PUSER - 1) | (flags & PCATCH), "suspfs", 0); 1001 if (error) 1002 goto unlock; 1003 } 1004 } 1005 if (flags & V_XSLEEP) 1006 goto unlock; 1007 mp->mnt_writeopcount++; 1008 unlock: 1009 if (error != 0 || (flags & V_XSLEEP) != 0) 1010 MNT_REL(mp); 1011 MNT_IUNLOCK(mp); 1012 return (error); 1013 } 1014 1015 /* 1016 * Secondary suspension. Used by operations such as vop_inactive 1017 * routines that are needed by the higher level functions. These 1018 * are allowed to proceed until all the higher level functions have 1019 * completed (indicated by mnt_writeopcount dropping to zero). At that 1020 * time, these operations are halted until the suspension is over. 1021 */ 1022 int 1023 vn_start_secondary_write(vp, mpp, flags) 1024 struct vnode *vp; 1025 struct mount **mpp; 1026 int flags; 1027 { 1028 struct mount *mp; 1029 int error; 1030 1031 retry: 1032 if (vp != NULL) { 1033 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 1034 *mpp = NULL; 1035 if (error != EOPNOTSUPP) 1036 return (error); 1037 return (0); 1038 } 1039 } 1040 /* 1041 * If we are not suspended or have not yet reached suspended 1042 * mode, then let the operation proceed. 1043 */ 1044 if ((mp = *mpp) == NULL) 1045 return (0); 1046 1047 /* 1048 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 1049 * a vfs_ref(). 1050 * As long as a vnode is not provided we need to acquire a 1051 * refcount for the provided mountpoint too, in order to 1052 * emulate a vfs_ref(). 1053 */ 1054 MNT_ILOCK(mp); 1055 if (vp == NULL) 1056 MNT_REF(mp); 1057 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) { 1058 mp->mnt_secondary_writes++; 1059 mp->mnt_secondary_accwrites++; 1060 MNT_IUNLOCK(mp); 1061 return (0); 1062 } 1063 if (flags & V_NOWAIT) { 1064 MNT_REL(mp); 1065 MNT_IUNLOCK(mp); 1066 return (EWOULDBLOCK); 1067 } 1068 /* 1069 * Wait for the suspension to finish. 1070 */ 1071 error = msleep(&mp->mnt_flag, MNT_MTX(mp), 1072 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0); 1073 vfs_rel(mp); 1074 if (error == 0) 1075 goto retry; 1076 return (error); 1077 } 1078 1079 /* 1080 * Filesystem write operation has completed. If we are suspending and this 1081 * operation is the last one, notify the suspender that the suspension is 1082 * now in effect. 1083 */ 1084 void 1085 vn_finished_write(mp) 1086 struct mount *mp; 1087 { 1088 if (mp == NULL) 1089 return; 1090 MNT_ILOCK(mp); 1091 MNT_REL(mp); 1092 mp->mnt_writeopcount--; 1093 if (mp->mnt_writeopcount < 0) 1094 panic("vn_finished_write: neg cnt"); 1095 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1096 mp->mnt_writeopcount <= 0) 1097 wakeup(&mp->mnt_writeopcount); 1098 MNT_IUNLOCK(mp); 1099 } 1100 1101 1102 /* 1103 * Filesystem secondary write operation has completed. If we are 1104 * suspending and this operation is the last one, notify the suspender 1105 * that the suspension is now in effect. 1106 */ 1107 void 1108 vn_finished_secondary_write(mp) 1109 struct mount *mp; 1110 { 1111 if (mp == NULL) 1112 return; 1113 MNT_ILOCK(mp); 1114 MNT_REL(mp); 1115 mp->mnt_secondary_writes--; 1116 if (mp->mnt_secondary_writes < 0) 1117 panic("vn_finished_secondary_write: neg cnt"); 1118 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1119 mp->mnt_secondary_writes <= 0) 1120 wakeup(&mp->mnt_secondary_writes); 1121 MNT_IUNLOCK(mp); 1122 } 1123 1124 1125 1126 /* 1127 * Request a filesystem to suspend write operations. 1128 */ 1129 int 1130 vfs_write_suspend(mp) 1131 struct mount *mp; 1132 { 1133 int error; 1134 1135 MNT_ILOCK(mp); 1136 if (mp->mnt_susp_owner == curthread) { 1137 MNT_IUNLOCK(mp); 1138 return (EALREADY); 1139 } 1140 while (mp->mnt_kern_flag & MNTK_SUSPEND) 1141 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0); 1142 mp->mnt_kern_flag |= MNTK_SUSPEND; 1143 mp->mnt_susp_owner = curthread; 1144 if (mp->mnt_writeopcount > 0) 1145 (void) msleep(&mp->mnt_writeopcount, 1146 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0); 1147 else 1148 MNT_IUNLOCK(mp); 1149 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0) 1150 vfs_write_resume(mp); 1151 return (error); 1152 } 1153 1154 /* 1155 * Request a filesystem to resume write operations. 1156 */ 1157 void 1158 vfs_write_resume(mp) 1159 struct mount *mp; 1160 { 1161 1162 MNT_ILOCK(mp); 1163 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 1164 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner")); 1165 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 | 1166 MNTK_SUSPENDED); 1167 mp->mnt_susp_owner = NULL; 1168 wakeup(&mp->mnt_writeopcount); 1169 wakeup(&mp->mnt_flag); 1170 curthread->td_pflags &= ~TDP_IGNSUSP; 1171 MNT_IUNLOCK(mp); 1172 VFS_SUSP_CLEAN(mp); 1173 } else 1174 MNT_IUNLOCK(mp); 1175 } 1176 1177 /* 1178 * Implement kqueues for files by translating it to vnode operation. 1179 */ 1180 static int 1181 vn_kqfilter(struct file *fp, struct knote *kn) 1182 { 1183 int vfslocked; 1184 int error; 1185 1186 vfslocked = VFS_LOCK_GIANT(fp->f_vnode->v_mount); 1187 error = VOP_KQFILTER(fp->f_vnode, kn); 1188 VFS_UNLOCK_GIANT(vfslocked); 1189 1190 return error; 1191 } 1192 1193 /* 1194 * Simplified in-kernel wrapper calls for extended attribute access. 1195 * Both calls pass in a NULL credential, authorizing as "kernel" access. 1196 * Set IO_NODELOCKED in ioflg if the vnode is already locked. 1197 */ 1198 int 1199 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, 1200 const char *attrname, int *buflen, char *buf, struct thread *td) 1201 { 1202 struct uio auio; 1203 struct iovec iov; 1204 int error; 1205 1206 iov.iov_len = *buflen; 1207 iov.iov_base = buf; 1208 1209 auio.uio_iov = &iov; 1210 auio.uio_iovcnt = 1; 1211 auio.uio_rw = UIO_READ; 1212 auio.uio_segflg = UIO_SYSSPACE; 1213 auio.uio_td = td; 1214 auio.uio_offset = 0; 1215 auio.uio_resid = *buflen; 1216 1217 if ((ioflg & IO_NODELOCKED) == 0) 1218 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1219 1220 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1221 1222 /* authorize attribute retrieval as kernel */ 1223 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL, 1224 td); 1225 1226 if ((ioflg & IO_NODELOCKED) == 0) 1227 VOP_UNLOCK(vp, 0); 1228 1229 if (error == 0) { 1230 *buflen = *buflen - auio.uio_resid; 1231 } 1232 1233 return (error); 1234 } 1235 1236 /* 1237 * XXX failure mode if partially written? 1238 */ 1239 int 1240 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, 1241 const char *attrname, int buflen, char *buf, struct thread *td) 1242 { 1243 struct uio auio; 1244 struct iovec iov; 1245 struct mount *mp; 1246 int error; 1247 1248 iov.iov_len = buflen; 1249 iov.iov_base = buf; 1250 1251 auio.uio_iov = &iov; 1252 auio.uio_iovcnt = 1; 1253 auio.uio_rw = UIO_WRITE; 1254 auio.uio_segflg = UIO_SYSSPACE; 1255 auio.uio_td = td; 1256 auio.uio_offset = 0; 1257 auio.uio_resid = buflen; 1258 1259 if ((ioflg & IO_NODELOCKED) == 0) { 1260 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 1261 return (error); 1262 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1263 } 1264 1265 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1266 1267 /* authorize attribute setting as kernel */ 1268 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td); 1269 1270 if ((ioflg & IO_NODELOCKED) == 0) { 1271 vn_finished_write(mp); 1272 VOP_UNLOCK(vp, 0); 1273 } 1274 1275 return (error); 1276 } 1277 1278 int 1279 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, 1280 const char *attrname, struct thread *td) 1281 { 1282 struct mount *mp; 1283 int error; 1284 1285 if ((ioflg & IO_NODELOCKED) == 0) { 1286 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 1287 return (error); 1288 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1289 } 1290 1291 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1292 1293 /* authorize attribute removal as kernel */ 1294 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td); 1295 if (error == EOPNOTSUPP) 1296 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, 1297 NULL, td); 1298 1299 if ((ioflg & IO_NODELOCKED) == 0) { 1300 vn_finished_write(mp); 1301 VOP_UNLOCK(vp, 0); 1302 } 1303 1304 return (error); 1305 } 1306 1307 int 1308 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp) 1309 { 1310 struct mount *mp; 1311 int ltype, error; 1312 1313 mp = vp->v_mount; 1314 ltype = VOP_ISLOCKED(vp); 1315 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED, 1316 ("vn_vget_ino: vp not locked")); 1317 error = vfs_busy(mp, MBF_NOWAIT); 1318 if (error != 0) { 1319 vfs_ref(mp); 1320 VOP_UNLOCK(vp, 0); 1321 error = vfs_busy(mp, 0); 1322 vn_lock(vp, ltype | LK_RETRY); 1323 vfs_rel(mp); 1324 if (error != 0) 1325 return (ENOENT); 1326 if (vp->v_iflag & VI_DOOMED) { 1327 vfs_unbusy(mp); 1328 return (ENOENT); 1329 } 1330 } 1331 VOP_UNLOCK(vp, 0); 1332 error = VFS_VGET(mp, ino, lkflags, rvp); 1333 vfs_unbusy(mp); 1334 vn_lock(vp, ltype | LK_RETRY); 1335 if (vp->v_iflag & VI_DOOMED) { 1336 if (error == 0) 1337 vput(*rvp); 1338 error = ENOENT; 1339 } 1340 return (error); 1341 } 1342 1343 int 1344 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, 1345 const struct thread *td) 1346 { 1347 1348 if (vp->v_type != VREG || td == NULL) 1349 return (0); 1350 PROC_LOCK(td->td_proc); 1351 if ((uoff_t)uio->uio_offset + uio->uio_resid > 1352 lim_cur(td->td_proc, RLIMIT_FSIZE)) { 1353 psignal(td->td_proc, SIGXFSZ); 1354 PROC_UNLOCK(td->td_proc); 1355 return (EFBIG); 1356 } 1357 PROC_UNLOCK(td->td_proc); 1358 return (0); 1359 } 1360