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