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