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/rwlock.h> 59 #include <sys/sx.h> 60 #include <sys/sysctl.h> 61 #include <sys/ttycom.h> 62 #include <sys/conf.h> 63 #include <sys/syslog.h> 64 #include <sys/unistd.h> 65 66 #include <security/audit/audit.h> 67 #include <security/mac/mac_framework.h> 68 69 #include <vm/vm.h> 70 #include <vm/vm_extern.h> 71 #include <vm/pmap.h> 72 #include <vm/vm_map.h> 73 #include <vm/vm_object.h> 74 #include <vm/vm_page.h> 75 76 static fo_rdwr_t vn_read; 77 static fo_rdwr_t vn_write; 78 static fo_rdwr_t vn_io_fault; 79 static fo_truncate_t vn_truncate; 80 static fo_ioctl_t vn_ioctl; 81 static fo_poll_t vn_poll; 82 static fo_kqfilter_t vn_kqfilter; 83 static fo_stat_t vn_statfile; 84 static fo_close_t vn_closefile; 85 86 struct fileops vnops = { 87 .fo_read = vn_io_fault, 88 .fo_write = vn_io_fault, 89 .fo_truncate = vn_truncate, 90 .fo_ioctl = vn_ioctl, 91 .fo_poll = vn_poll, 92 .fo_kqfilter = vn_kqfilter, 93 .fo_stat = vn_statfile, 94 .fo_close = vn_closefile, 95 .fo_chmod = vn_chmod, 96 .fo_chown = vn_chown, 97 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 98 }; 99 100 int 101 vn_open(ndp, flagp, cmode, fp) 102 struct nameidata *ndp; 103 int *flagp, cmode; 104 struct file *fp; 105 { 106 struct thread *td = ndp->ni_cnd.cn_thread; 107 108 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp)); 109 } 110 111 /* 112 * Common code for vnode open operations via a name lookup. 113 * Lookup the vnode and invoke VOP_CREATE if needed. 114 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine. 115 * 116 * Note that this does NOT free nameidata for the successful case, 117 * due to the NDINIT being done elsewhere. 118 */ 119 int 120 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, 121 struct ucred *cred, struct file *fp) 122 { 123 struct vnode *vp; 124 struct mount *mp; 125 struct thread *td = ndp->ni_cnd.cn_thread; 126 struct vattr vat; 127 struct vattr *vap = &vat; 128 int fmode, error; 129 130 restart: 131 fmode = *flagp; 132 if (fmode & O_CREAT) { 133 ndp->ni_cnd.cn_nameiop = CREATE; 134 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF; 135 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 136 ndp->ni_cnd.cn_flags |= FOLLOW; 137 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 138 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 139 if (vn_open_flags & VN_OPEN_NOCAPCHECK) 140 ndp->ni_cnd.cn_flags |= NOCAPCHECK; 141 bwillwrite(); 142 if ((error = namei(ndp)) != 0) 143 return (error); 144 if (ndp->ni_vp == NULL) { 145 VATTR_NULL(vap); 146 vap->va_type = VREG; 147 vap->va_mode = cmode; 148 if (fmode & O_EXCL) 149 vap->va_vaflags |= VA_EXCLUSIVE; 150 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) { 151 NDFREE(ndp, NDF_ONLY_PNBUF); 152 vput(ndp->ni_dvp); 153 if ((error = vn_start_write(NULL, &mp, 154 V_XSLEEP | PCATCH)) != 0) 155 return (error); 156 goto restart; 157 } 158 #ifdef MAC 159 error = mac_vnode_check_create(cred, ndp->ni_dvp, 160 &ndp->ni_cnd, vap); 161 if (error == 0) 162 #endif 163 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 164 &ndp->ni_cnd, vap); 165 vput(ndp->ni_dvp); 166 vn_finished_write(mp); 167 if (error) { 168 NDFREE(ndp, NDF_ONLY_PNBUF); 169 return (error); 170 } 171 fmode &= ~O_TRUNC; 172 vp = ndp->ni_vp; 173 } else { 174 if (ndp->ni_dvp == ndp->ni_vp) 175 vrele(ndp->ni_dvp); 176 else 177 vput(ndp->ni_dvp); 178 ndp->ni_dvp = NULL; 179 vp = ndp->ni_vp; 180 if (fmode & O_EXCL) { 181 error = EEXIST; 182 goto bad; 183 } 184 fmode &= ~O_CREAT; 185 } 186 } else { 187 ndp->ni_cnd.cn_nameiop = LOOKUP; 188 ndp->ni_cnd.cn_flags = ISOPEN | 189 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF; 190 if (!(fmode & FWRITE)) 191 ndp->ni_cnd.cn_flags |= LOCKSHARED; 192 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 193 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 194 if (vn_open_flags & VN_OPEN_NOCAPCHECK) 195 ndp->ni_cnd.cn_flags |= NOCAPCHECK; 196 if ((error = namei(ndp)) != 0) 197 return (error); 198 vp = ndp->ni_vp; 199 } 200 error = vn_open_vnode(vp, fmode, cred, td, fp); 201 if (error) 202 goto bad; 203 *flagp = fmode; 204 return (0); 205 bad: 206 NDFREE(ndp, NDF_ONLY_PNBUF); 207 vput(vp); 208 *flagp = fmode; 209 ndp->ni_vp = NULL; 210 return (error); 211 } 212 213 /* 214 * Common code for vnode open operations once a vnode is located. 215 * Check permissions, and call the VOP_OPEN routine. 216 */ 217 int 218 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, 219 struct thread *td, struct file *fp) 220 { 221 struct mount *mp; 222 accmode_t accmode; 223 struct flock lf; 224 int error, have_flock, lock_flags, type; 225 226 if (vp->v_type == VLNK) 227 return (EMLINK); 228 if (vp->v_type == VSOCK) 229 return (EOPNOTSUPP); 230 if (vp->v_type != VDIR && fmode & O_DIRECTORY) 231 return (ENOTDIR); 232 accmode = 0; 233 if (fmode & (FWRITE | O_TRUNC)) { 234 if (vp->v_type == VDIR) 235 return (EISDIR); 236 accmode |= VWRITE; 237 } 238 if (fmode & FREAD) 239 accmode |= VREAD; 240 if (fmode & FEXEC) 241 accmode |= VEXEC; 242 if ((fmode & O_APPEND) && (fmode & FWRITE)) 243 accmode |= VAPPEND; 244 #ifdef MAC 245 error = mac_vnode_check_open(cred, vp, accmode); 246 if (error) 247 return (error); 248 #endif 249 if ((fmode & O_CREAT) == 0) { 250 if (accmode & VWRITE) { 251 error = vn_writechk(vp); 252 if (error) 253 return (error); 254 } 255 if (accmode) { 256 error = VOP_ACCESS(vp, accmode, cred, td); 257 if (error) 258 return (error); 259 } 260 } 261 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0) 262 return (error); 263 264 if (fmode & (O_EXLOCK | O_SHLOCK)) { 265 KASSERT(fp != NULL, ("open with flock requires fp")); 266 lock_flags = VOP_ISLOCKED(vp); 267 VOP_UNLOCK(vp, 0); 268 lf.l_whence = SEEK_SET; 269 lf.l_start = 0; 270 lf.l_len = 0; 271 if (fmode & O_EXLOCK) 272 lf.l_type = F_WRLCK; 273 else 274 lf.l_type = F_RDLCK; 275 type = F_FLOCK; 276 if ((fmode & FNONBLOCK) == 0) 277 type |= F_WAIT; 278 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type); 279 have_flock = (error == 0); 280 vn_lock(vp, lock_flags | LK_RETRY); 281 if (error == 0 && vp->v_iflag & VI_DOOMED) 282 error = ENOENT; 283 /* 284 * Another thread might have used this vnode as an 285 * executable while the vnode lock was dropped. 286 * Ensure the vnode is still able to be opened for 287 * writing after the lock has been obtained. 288 */ 289 if (error == 0 && accmode & VWRITE) 290 error = vn_writechk(vp); 291 if (error) { 292 VOP_UNLOCK(vp, 0); 293 if (have_flock) { 294 lf.l_whence = SEEK_SET; 295 lf.l_start = 0; 296 lf.l_len = 0; 297 lf.l_type = F_UNLCK; 298 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, 299 F_FLOCK); 300 } 301 vn_start_write(vp, &mp, V_WAIT); 302 vn_lock(vp, lock_flags | LK_RETRY); 303 (void)VOP_CLOSE(vp, fmode, cred, td); 304 vn_finished_write(mp); 305 return (error); 306 } 307 fp->f_flag |= FHASLOCK; 308 } 309 if (fmode & FWRITE) { 310 VOP_ADD_WRITECOUNT(vp, 1); 311 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", 312 __func__, vp, vp->v_writecount); 313 } 314 ASSERT_VOP_LOCKED(vp, "vn_open_vnode"); 315 return (0); 316 } 317 318 /* 319 * Check for write permissions on the specified vnode. 320 * Prototype text segments cannot be written. 321 */ 322 int 323 vn_writechk(vp) 324 register struct vnode *vp; 325 { 326 327 ASSERT_VOP_LOCKED(vp, "vn_writechk"); 328 /* 329 * If there's shared text associated with 330 * the vnode, try to free it up once. If 331 * we fail, we can't allow writing. 332 */ 333 if (VOP_IS_TEXT(vp)) 334 return (ETXTBSY); 335 336 return (0); 337 } 338 339 /* 340 * Vnode close call 341 */ 342 int 343 vn_close(vp, flags, file_cred, td) 344 register struct vnode *vp; 345 int flags; 346 struct ucred *file_cred; 347 struct thread *td; 348 { 349 struct mount *mp; 350 int error, lock_flags; 351 352 if (!(flags & FWRITE) && vp->v_mount != NULL && 353 vp->v_mount->mnt_kern_flag & MNTK_EXTENDED_SHARED) 354 lock_flags = LK_SHARED; 355 else 356 lock_flags = LK_EXCLUSIVE; 357 358 vn_start_write(vp, &mp, V_WAIT); 359 vn_lock(vp, lock_flags | LK_RETRY); 360 if (flags & FWRITE) { 361 VNASSERT(vp->v_writecount > 0, vp, 362 ("vn_close: negative writecount")); 363 VOP_ADD_WRITECOUNT(vp, -1); 364 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", 365 __func__, vp, vp->v_writecount); 366 } 367 error = VOP_CLOSE(vp, flags, file_cred, td); 368 vput(vp); 369 vn_finished_write(mp); 370 return (error); 371 } 372 373 /* 374 * Heuristic to detect sequential operation. 375 */ 376 static int 377 sequential_heuristic(struct uio *uio, struct file *fp) 378 { 379 380 if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD) 381 return (fp->f_seqcount << IO_SEQSHIFT); 382 383 /* 384 * Offset 0 is handled specially. open() sets f_seqcount to 1 so 385 * that the first I/O is normally considered to be slightly 386 * sequential. Seeking to offset 0 doesn't change sequentiality 387 * unless previous seeks have reduced f_seqcount to 0, in which 388 * case offset 0 is not special. 389 */ 390 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 391 uio->uio_offset == fp->f_nextoff) { 392 /* 393 * f_seqcount is in units of fixed-size blocks so that it 394 * depends mainly on the amount of sequential I/O and not 395 * much on the number of sequential I/O's. The fixed size 396 * of 16384 is hard-coded here since it is (not quite) just 397 * a magic size that works well here. This size is more 398 * closely related to the best I/O size for real disks than 399 * to any block size used by software. 400 */ 401 fp->f_seqcount += howmany(uio->uio_resid, 16384); 402 if (fp->f_seqcount > IO_SEQMAX) 403 fp->f_seqcount = IO_SEQMAX; 404 return (fp->f_seqcount << IO_SEQSHIFT); 405 } 406 407 /* Not sequential. Quickly draw-down sequentiality. */ 408 if (fp->f_seqcount > 1) 409 fp->f_seqcount = 1; 410 else 411 fp->f_seqcount = 0; 412 return (0); 413 } 414 415 /* 416 * Package up an I/O request on a vnode into a uio and do it. 417 */ 418 int 419 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, 420 enum uio_seg segflg, int ioflg, struct ucred *active_cred, 421 struct ucred *file_cred, ssize_t *aresid, struct thread *td) 422 { 423 struct uio auio; 424 struct iovec aiov; 425 struct mount *mp; 426 struct ucred *cred; 427 void *rl_cookie; 428 int error, lock_flags; 429 430 auio.uio_iov = &aiov; 431 auio.uio_iovcnt = 1; 432 aiov.iov_base = base; 433 aiov.iov_len = len; 434 auio.uio_resid = len; 435 auio.uio_offset = offset; 436 auio.uio_segflg = segflg; 437 auio.uio_rw = rw; 438 auio.uio_td = td; 439 error = 0; 440 441 if ((ioflg & IO_NODELOCKED) == 0) { 442 if (rw == UIO_READ) { 443 rl_cookie = vn_rangelock_rlock(vp, offset, 444 offset + len); 445 } else { 446 rl_cookie = vn_rangelock_wlock(vp, offset, 447 offset + len); 448 } 449 mp = NULL; 450 if (rw == UIO_WRITE) { 451 if (vp->v_type != VCHR && 452 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) 453 != 0) 454 goto out; 455 if (MNT_SHARED_WRITES(mp) || 456 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) 457 lock_flags = LK_SHARED; 458 else 459 lock_flags = LK_EXCLUSIVE; 460 } else 461 lock_flags = LK_SHARED; 462 vn_lock(vp, lock_flags | LK_RETRY); 463 } else 464 rl_cookie = NULL; 465 466 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 467 #ifdef MAC 468 if ((ioflg & IO_NOMACCHECK) == 0) { 469 if (rw == UIO_READ) 470 error = mac_vnode_check_read(active_cred, file_cred, 471 vp); 472 else 473 error = mac_vnode_check_write(active_cred, file_cred, 474 vp); 475 } 476 #endif 477 if (error == 0) { 478 if (file_cred != NULL) 479 cred = file_cred; 480 else 481 cred = active_cred; 482 if (rw == UIO_READ) 483 error = VOP_READ(vp, &auio, ioflg, cred); 484 else 485 error = VOP_WRITE(vp, &auio, ioflg, cred); 486 } 487 if (aresid) 488 *aresid = auio.uio_resid; 489 else 490 if (auio.uio_resid && error == 0) 491 error = EIO; 492 if ((ioflg & IO_NODELOCKED) == 0) { 493 VOP_UNLOCK(vp, 0); 494 if (mp != NULL) 495 vn_finished_write(mp); 496 } 497 out: 498 if (rl_cookie != NULL) 499 vn_rangelock_unlock(vp, rl_cookie); 500 return (error); 501 } 502 503 /* 504 * Package up an I/O request on a vnode into a uio and do it. The I/O 505 * request is split up into smaller chunks and we try to avoid saturating 506 * the buffer cache while potentially holding a vnode locked, so we 507 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield() 508 * to give other processes a chance to lock the vnode (either other processes 509 * core'ing the same binary, or unrelated processes scanning the directory). 510 */ 511 int 512 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred, 513 file_cred, aresid, td) 514 enum uio_rw rw; 515 struct vnode *vp; 516 void *base; 517 size_t len; 518 off_t offset; 519 enum uio_seg segflg; 520 int ioflg; 521 struct ucred *active_cred; 522 struct ucred *file_cred; 523 size_t *aresid; 524 struct thread *td; 525 { 526 int error = 0; 527 ssize_t iaresid; 528 529 do { 530 int chunk; 531 532 /* 533 * Force `offset' to a multiple of MAXBSIZE except possibly 534 * for the first chunk, so that filesystems only need to 535 * write full blocks except possibly for the first and last 536 * chunks. 537 */ 538 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 539 540 if (chunk > len) 541 chunk = len; 542 if (rw != UIO_READ && vp->v_type == VREG) 543 bwillwrite(); 544 iaresid = 0; 545 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 546 ioflg, active_cred, file_cred, &iaresid, td); 547 len -= chunk; /* aresid calc already includes length */ 548 if (error) 549 break; 550 offset += chunk; 551 base = (char *)base + chunk; 552 kern_yield(PRI_USER); 553 } while (len); 554 if (aresid) 555 *aresid = len + iaresid; 556 return (error); 557 } 558 559 off_t 560 foffset_lock(struct file *fp, int flags) 561 { 562 struct mtx *mtxp; 563 off_t res; 564 565 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); 566 567 #if OFF_MAX <= LONG_MAX 568 /* 569 * Caller only wants the current f_offset value. Assume that 570 * the long and shorter integer types reads are atomic. 571 */ 572 if ((flags & FOF_NOLOCK) != 0) 573 return (fp->f_offset); 574 #endif 575 576 /* 577 * According to McKusick the vn lock was protecting f_offset here. 578 * It is now protected by the FOFFSET_LOCKED flag. 579 */ 580 mtxp = mtx_pool_find(mtxpool_sleep, fp); 581 mtx_lock(mtxp); 582 if ((flags & FOF_NOLOCK) == 0) { 583 while (fp->f_vnread_flags & FOFFSET_LOCKED) { 584 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING; 585 msleep(&fp->f_vnread_flags, mtxp, PUSER -1, 586 "vofflock", 0); 587 } 588 fp->f_vnread_flags |= FOFFSET_LOCKED; 589 } 590 res = fp->f_offset; 591 mtx_unlock(mtxp); 592 return (res); 593 } 594 595 void 596 foffset_unlock(struct file *fp, off_t val, int flags) 597 { 598 struct mtx *mtxp; 599 600 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); 601 602 #if OFF_MAX <= LONG_MAX 603 if ((flags & FOF_NOLOCK) != 0) { 604 if ((flags & FOF_NOUPDATE) == 0) 605 fp->f_offset = val; 606 if ((flags & FOF_NEXTOFF) != 0) 607 fp->f_nextoff = val; 608 return; 609 } 610 #endif 611 612 mtxp = mtx_pool_find(mtxpool_sleep, fp); 613 mtx_lock(mtxp); 614 if ((flags & FOF_NOUPDATE) == 0) 615 fp->f_offset = val; 616 if ((flags & FOF_NEXTOFF) != 0) 617 fp->f_nextoff = val; 618 if ((flags & FOF_NOLOCK) == 0) { 619 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0, 620 ("Lost FOFFSET_LOCKED")); 621 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING) 622 wakeup(&fp->f_vnread_flags); 623 fp->f_vnread_flags = 0; 624 } 625 mtx_unlock(mtxp); 626 } 627 628 void 629 foffset_lock_uio(struct file *fp, struct uio *uio, int flags) 630 { 631 632 if ((flags & FOF_OFFSET) == 0) 633 uio->uio_offset = foffset_lock(fp, flags); 634 } 635 636 void 637 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags) 638 { 639 640 if ((flags & FOF_OFFSET) == 0) 641 foffset_unlock(fp, uio->uio_offset, flags); 642 } 643 644 static int 645 get_advice(struct file *fp, struct uio *uio) 646 { 647 struct mtx *mtxp; 648 int ret; 649 650 ret = POSIX_FADV_NORMAL; 651 if (fp->f_advice == NULL) 652 return (ret); 653 654 mtxp = mtx_pool_find(mtxpool_sleep, fp); 655 mtx_lock(mtxp); 656 if (uio->uio_offset >= fp->f_advice->fa_start && 657 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end) 658 ret = fp->f_advice->fa_advice; 659 mtx_unlock(mtxp); 660 return (ret); 661 } 662 663 /* 664 * File table vnode read routine. 665 */ 666 static int 667 vn_read(fp, uio, active_cred, flags, td) 668 struct file *fp; 669 struct uio *uio; 670 struct ucred *active_cred; 671 int flags; 672 struct thread *td; 673 { 674 struct vnode *vp; 675 struct mtx *mtxp; 676 int error, ioflag; 677 int advice; 678 off_t offset, start, end; 679 680 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", 681 uio->uio_td, td)); 682 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); 683 vp = fp->f_vnode; 684 ioflag = 0; 685 if (fp->f_flag & FNONBLOCK) 686 ioflag |= IO_NDELAY; 687 if (fp->f_flag & O_DIRECT) 688 ioflag |= IO_DIRECT; 689 advice = get_advice(fp, uio); 690 vn_lock(vp, LK_SHARED | LK_RETRY); 691 692 switch (advice) { 693 case POSIX_FADV_NORMAL: 694 case POSIX_FADV_SEQUENTIAL: 695 case POSIX_FADV_NOREUSE: 696 ioflag |= sequential_heuristic(uio, fp); 697 break; 698 case POSIX_FADV_RANDOM: 699 /* Disable read-ahead for random I/O. */ 700 break; 701 } 702 offset = uio->uio_offset; 703 704 #ifdef MAC 705 error = mac_vnode_check_read(active_cred, fp->f_cred, vp); 706 if (error == 0) 707 #endif 708 error = VOP_READ(vp, uio, ioflag, fp->f_cred); 709 fp->f_nextoff = uio->uio_offset; 710 VOP_UNLOCK(vp, 0); 711 if (error == 0 && advice == POSIX_FADV_NOREUSE && 712 offset != uio->uio_offset) { 713 /* 714 * Use POSIX_FADV_DONTNEED to flush clean pages and 715 * buffers for the backing file after a 716 * POSIX_FADV_NOREUSE read(2). To optimize the common 717 * case of using POSIX_FADV_NOREUSE with sequential 718 * access, track the previous implicit DONTNEED 719 * request and grow this request to include the 720 * current read(2) in addition to the previous 721 * DONTNEED. With purely sequential access this will 722 * cause the DONTNEED requests to continously grow to 723 * cover all of the previously read regions of the 724 * file. This allows filesystem blocks that are 725 * accessed by multiple calls to read(2) to be flushed 726 * once the last read(2) finishes. 727 */ 728 start = offset; 729 end = uio->uio_offset - 1; 730 mtxp = mtx_pool_find(mtxpool_sleep, fp); 731 mtx_lock(mtxp); 732 if (fp->f_advice != NULL && 733 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) { 734 if (start != 0 && fp->f_advice->fa_prevend + 1 == start) 735 start = fp->f_advice->fa_prevstart; 736 else if (fp->f_advice->fa_prevstart != 0 && 737 fp->f_advice->fa_prevstart == end + 1) 738 end = fp->f_advice->fa_prevend; 739 fp->f_advice->fa_prevstart = start; 740 fp->f_advice->fa_prevend = end; 741 } 742 mtx_unlock(mtxp); 743 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED); 744 } 745 return (error); 746 } 747 748 /* 749 * File table vnode write routine. 750 */ 751 static int 752 vn_write(fp, uio, active_cred, flags, td) 753 struct file *fp; 754 struct uio *uio; 755 struct ucred *active_cred; 756 int flags; 757 struct thread *td; 758 { 759 struct vnode *vp; 760 struct mount *mp; 761 struct mtx *mtxp; 762 int error, ioflag, lock_flags; 763 int advice; 764 off_t offset, start, end; 765 766 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", 767 uio->uio_td, td)); 768 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); 769 vp = fp->f_vnode; 770 if (vp->v_type == VREG) 771 bwillwrite(); 772 ioflag = IO_UNIT; 773 if (vp->v_type == VREG && (fp->f_flag & O_APPEND)) 774 ioflag |= IO_APPEND; 775 if (fp->f_flag & FNONBLOCK) 776 ioflag |= IO_NDELAY; 777 if (fp->f_flag & O_DIRECT) 778 ioflag |= IO_DIRECT; 779 if ((fp->f_flag & O_FSYNC) || 780 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) 781 ioflag |= IO_SYNC; 782 mp = NULL; 783 if (vp->v_type != VCHR && 784 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) 785 goto unlock; 786 787 advice = get_advice(fp, uio); 788 789 if (MNT_SHARED_WRITES(mp) || 790 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) { 791 lock_flags = LK_SHARED; 792 } else { 793 lock_flags = LK_EXCLUSIVE; 794 } 795 796 vn_lock(vp, lock_flags | LK_RETRY); 797 switch (advice) { 798 case POSIX_FADV_NORMAL: 799 case POSIX_FADV_SEQUENTIAL: 800 case POSIX_FADV_NOREUSE: 801 ioflag |= sequential_heuristic(uio, fp); 802 break; 803 case POSIX_FADV_RANDOM: 804 /* XXX: Is this correct? */ 805 break; 806 } 807 offset = uio->uio_offset; 808 809 #ifdef MAC 810 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 811 if (error == 0) 812 #endif 813 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred); 814 fp->f_nextoff = uio->uio_offset; 815 VOP_UNLOCK(vp, 0); 816 if (vp->v_type != VCHR) 817 vn_finished_write(mp); 818 if (error == 0 && advice == POSIX_FADV_NOREUSE && 819 offset != uio->uio_offset) { 820 /* 821 * Use POSIX_FADV_DONTNEED to flush clean pages and 822 * buffers for the backing file after a 823 * POSIX_FADV_NOREUSE write(2). To optimize the 824 * common case of using POSIX_FADV_NOREUSE with 825 * sequential access, track the previous implicit 826 * DONTNEED request and grow this request to include 827 * the current write(2) in addition to the previous 828 * DONTNEED. With purely sequential access this will 829 * cause the DONTNEED requests to continously grow to 830 * cover all of the previously written regions of the 831 * file. 832 * 833 * Note that the blocks just written are almost 834 * certainly still dirty, so this only works when 835 * VOP_ADVISE() calls from subsequent writes push out 836 * the data written by this write(2) once the backing 837 * buffers are clean. However, as compared to forcing 838 * IO_DIRECT, this gives much saner behavior. Write 839 * clustering is still allowed, and clean pages are 840 * merely moved to the cache page queue rather than 841 * outright thrown away. This means a subsequent 842 * read(2) can still avoid hitting the disk if the 843 * pages have not been reclaimed. 844 * 845 * This does make POSIX_FADV_NOREUSE largely useless 846 * with non-sequential access. However, sequential 847 * access is the more common use case and the flag is 848 * merely advisory. 849 */ 850 start = offset; 851 end = uio->uio_offset - 1; 852 mtxp = mtx_pool_find(mtxpool_sleep, fp); 853 mtx_lock(mtxp); 854 if (fp->f_advice != NULL && 855 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) { 856 if (start != 0 && fp->f_advice->fa_prevend + 1 == start) 857 start = fp->f_advice->fa_prevstart; 858 else if (fp->f_advice->fa_prevstart != 0 && 859 fp->f_advice->fa_prevstart == end + 1) 860 end = fp->f_advice->fa_prevend; 861 fp->f_advice->fa_prevstart = start; 862 fp->f_advice->fa_prevend = end; 863 } 864 mtx_unlock(mtxp); 865 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED); 866 } 867 868 unlock: 869 return (error); 870 } 871 872 static const int io_hold_cnt = 16; 873 static int vn_io_fault_enable = 1; 874 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW, 875 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance"); 876 static u_long vn_io_faults_cnt; 877 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD, 878 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers"); 879 880 /* 881 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to 882 * prevent the following deadlock: 883 * 884 * Assume that the thread A reads from the vnode vp1 into userspace 885 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is 886 * currently not resident, then system ends up with the call chain 887 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] -> 888 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2) 889 * which establishes lock order vp1->vn_lock, then vp2->vn_lock. 890 * If, at the same time, thread B reads from vnode vp2 into buffer buf2 891 * backed by the pages of vnode vp1, and some page in buf2 is not 892 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock. 893 * 894 * To prevent the lock order reversal and deadlock, vn_io_fault() does 895 * not allow page faults to happen during VOP_READ() or VOP_WRITE(). 896 * Instead, it first tries to do the whole range i/o with pagefaults 897 * disabled. If all pages in the i/o buffer are resident and mapped, 898 * VOP will succeed (ignoring the genuine filesystem errors). 899 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do 900 * i/o in chunks, with all pages in the chunk prefaulted and held 901 * using vm_fault_quick_hold_pages(). 902 * 903 * Filesystems using this deadlock avoidance scheme should use the 904 * array of the held pages from uio, saved in the curthread->td_ma, 905 * instead of doing uiomove(). A helper function 906 * vn_io_fault_uiomove() converts uiomove request into 907 * uiomove_fromphys() over td_ma array. 908 * 909 * Since vnode locks do not cover the whole i/o anymore, rangelocks 910 * make the current i/o request atomic with respect to other i/os and 911 * truncations. 912 */ 913 static int 914 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred, 915 int flags, struct thread *td) 916 { 917 vm_page_t ma[io_hold_cnt + 2]; 918 struct uio *uio_clone, short_uio; 919 struct iovec short_iovec[1]; 920 fo_rdwr_t *doio; 921 struct vnode *vp; 922 void *rl_cookie; 923 struct mount *mp; 924 vm_page_t *prev_td_ma; 925 int cnt, error, save, saveheld, prev_td_ma_cnt; 926 vm_offset_t addr, end; 927 vm_prot_t prot; 928 size_t len, resid; 929 ssize_t adv; 930 931 if (uio->uio_rw == UIO_READ) 932 doio = vn_read; 933 else 934 doio = vn_write; 935 vp = fp->f_vnode; 936 foffset_lock_uio(fp, uio, flags); 937 938 if (uio->uio_segflg != UIO_USERSPACE || vp->v_type != VREG || 939 ((mp = vp->v_mount) != NULL && 940 (mp->mnt_kern_flag & MNTK_NO_IOPF) == 0) || 941 !vn_io_fault_enable) { 942 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td); 943 goto out_last; 944 } 945 946 /* 947 * The UFS follows IO_UNIT directive and replays back both 948 * uio_offset and uio_resid if an error is encountered during the 949 * operation. But, since the iovec may be already advanced, 950 * uio is still in an inconsistent state. 951 * 952 * Cache a copy of the original uio, which is advanced to the redo 953 * point using UIO_NOCOPY below. 954 */ 955 uio_clone = cloneuio(uio); 956 resid = uio->uio_resid; 957 958 short_uio.uio_segflg = UIO_USERSPACE; 959 short_uio.uio_rw = uio->uio_rw; 960 short_uio.uio_td = uio->uio_td; 961 962 if (uio->uio_rw == UIO_READ) { 963 prot = VM_PROT_WRITE; 964 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset, 965 uio->uio_offset + uio->uio_resid); 966 } else { 967 prot = VM_PROT_READ; 968 if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0) 969 /* For appenders, punt and lock the whole range. */ 970 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); 971 else 972 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset, 973 uio->uio_offset + uio->uio_resid); 974 } 975 976 save = vm_fault_disable_pagefaults(); 977 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td); 978 if (error != EFAULT) 979 goto out; 980 981 atomic_add_long(&vn_io_faults_cnt, 1); 982 uio_clone->uio_segflg = UIO_NOCOPY; 983 uiomove(NULL, resid - uio->uio_resid, uio_clone); 984 uio_clone->uio_segflg = uio->uio_segflg; 985 986 saveheld = curthread_pflags_set(TDP_UIOHELD); 987 prev_td_ma = td->td_ma; 988 prev_td_ma_cnt = td->td_ma_cnt; 989 990 while (uio_clone->uio_resid != 0) { 991 len = uio_clone->uio_iov->iov_len; 992 if (len == 0) { 993 KASSERT(uio_clone->uio_iovcnt >= 1, 994 ("iovcnt underflow")); 995 uio_clone->uio_iov++; 996 uio_clone->uio_iovcnt--; 997 continue; 998 } 999 1000 addr = (vm_offset_t)uio_clone->uio_iov->iov_base; 1001 end = round_page(addr + len); 1002 cnt = howmany(end - trunc_page(addr), PAGE_SIZE); 1003 /* 1004 * A perfectly misaligned address and length could cause 1005 * both the start and the end of the chunk to use partial 1006 * page. +2 accounts for such a situation. 1007 */ 1008 if (cnt > io_hold_cnt + 2) { 1009 len = io_hold_cnt * PAGE_SIZE; 1010 KASSERT(howmany(round_page(addr + len) - 1011 trunc_page(addr), PAGE_SIZE) <= io_hold_cnt + 2, 1012 ("cnt overflow")); 1013 } 1014 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map, 1015 addr, len, prot, ma, io_hold_cnt + 2); 1016 if (cnt == -1) { 1017 error = EFAULT; 1018 break; 1019 } 1020 short_uio.uio_iov = &short_iovec[0]; 1021 short_iovec[0].iov_base = (void *)addr; 1022 short_uio.uio_iovcnt = 1; 1023 short_uio.uio_resid = short_iovec[0].iov_len = len; 1024 short_uio.uio_offset = uio_clone->uio_offset; 1025 td->td_ma = ma; 1026 td->td_ma_cnt = cnt; 1027 1028 error = doio(fp, &short_uio, active_cred, flags | FOF_OFFSET, 1029 td); 1030 vm_page_unhold_pages(ma, cnt); 1031 adv = len - short_uio.uio_resid; 1032 1033 uio_clone->uio_iov->iov_base = 1034 (char *)uio_clone->uio_iov->iov_base + adv; 1035 uio_clone->uio_iov->iov_len -= adv; 1036 uio_clone->uio_resid -= adv; 1037 uio_clone->uio_offset += adv; 1038 1039 uio->uio_resid -= adv; 1040 uio->uio_offset += adv; 1041 1042 if (error != 0 || adv == 0) 1043 break; 1044 } 1045 td->td_ma = prev_td_ma; 1046 td->td_ma_cnt = prev_td_ma_cnt; 1047 curthread_pflags_restore(saveheld); 1048 out: 1049 vm_fault_enable_pagefaults(save); 1050 vn_rangelock_unlock(vp, rl_cookie); 1051 free(uio_clone, M_IOV); 1052 out_last: 1053 foffset_unlock_uio(fp, uio, flags); 1054 return (error); 1055 } 1056 1057 /* 1058 * Helper function to perform the requested uiomove operation using 1059 * the held pages for io->uio_iov[0].iov_base buffer instead of 1060 * copyin/copyout. Access to the pages with uiomove_fromphys() 1061 * instead of iov_base prevents page faults that could occur due to 1062 * pmap_collect() invalidating the mapping created by 1063 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or 1064 * object cleanup revoking the write access from page mappings. 1065 * 1066 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove() 1067 * instead of plain uiomove(). 1068 */ 1069 int 1070 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio) 1071 { 1072 struct uio transp_uio; 1073 struct iovec transp_iov[1]; 1074 struct thread *td; 1075 size_t adv; 1076 int error, pgadv; 1077 1078 td = curthread; 1079 if ((td->td_pflags & TDP_UIOHELD) == 0 || 1080 uio->uio_segflg != UIO_USERSPACE) 1081 return (uiomove(data, xfersize, uio)); 1082 1083 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); 1084 transp_iov[0].iov_base = data; 1085 transp_uio.uio_iov = &transp_iov[0]; 1086 transp_uio.uio_iovcnt = 1; 1087 if (xfersize > uio->uio_resid) 1088 xfersize = uio->uio_resid; 1089 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize; 1090 transp_uio.uio_offset = 0; 1091 transp_uio.uio_segflg = UIO_SYSSPACE; 1092 /* 1093 * Since transp_iov points to data, and td_ma page array 1094 * corresponds to original uio->uio_iov, we need to invert the 1095 * direction of the i/o operation as passed to 1096 * uiomove_fromphys(). 1097 */ 1098 switch (uio->uio_rw) { 1099 case UIO_WRITE: 1100 transp_uio.uio_rw = UIO_READ; 1101 break; 1102 case UIO_READ: 1103 transp_uio.uio_rw = UIO_WRITE; 1104 break; 1105 } 1106 transp_uio.uio_td = uio->uio_td; 1107 error = uiomove_fromphys(td->td_ma, 1108 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK, 1109 xfersize, &transp_uio); 1110 adv = xfersize - transp_uio.uio_resid; 1111 pgadv = 1112 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) - 1113 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT); 1114 td->td_ma += pgadv; 1115 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, 1116 pgadv)); 1117 td->td_ma_cnt -= pgadv; 1118 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv; 1119 uio->uio_iov->iov_len -= adv; 1120 uio->uio_resid -= adv; 1121 uio->uio_offset += adv; 1122 return (error); 1123 } 1124 1125 /* 1126 * File table truncate routine. 1127 */ 1128 static int 1129 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred, 1130 struct thread *td) 1131 { 1132 struct vattr vattr; 1133 struct mount *mp; 1134 struct vnode *vp; 1135 void *rl_cookie; 1136 int error; 1137 1138 vp = fp->f_vnode; 1139 1140 /* 1141 * Lock the whole range for truncation. Otherwise split i/o 1142 * might happen partly before and partly after the truncation. 1143 */ 1144 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); 1145 error = vn_start_write(vp, &mp, V_WAIT | PCATCH); 1146 if (error) 1147 goto out1; 1148 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1149 if (vp->v_type == VDIR) { 1150 error = EISDIR; 1151 goto out; 1152 } 1153 #ifdef MAC 1154 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 1155 if (error) 1156 goto out; 1157 #endif 1158 error = vn_writechk(vp); 1159 if (error == 0) { 1160 VATTR_NULL(&vattr); 1161 vattr.va_size = length; 1162 error = VOP_SETATTR(vp, &vattr, fp->f_cred); 1163 } 1164 out: 1165 VOP_UNLOCK(vp, 0); 1166 vn_finished_write(mp); 1167 out1: 1168 vn_rangelock_unlock(vp, rl_cookie); 1169 return (error); 1170 } 1171 1172 /* 1173 * File table vnode stat routine. 1174 */ 1175 static int 1176 vn_statfile(fp, sb, active_cred, td) 1177 struct file *fp; 1178 struct stat *sb; 1179 struct ucred *active_cred; 1180 struct thread *td; 1181 { 1182 struct vnode *vp = fp->f_vnode; 1183 int error; 1184 1185 vn_lock(vp, LK_SHARED | LK_RETRY); 1186 error = vn_stat(vp, sb, active_cred, fp->f_cred, td); 1187 VOP_UNLOCK(vp, 0); 1188 1189 return (error); 1190 } 1191 1192 /* 1193 * Stat a vnode; implementation for the stat syscall 1194 */ 1195 int 1196 vn_stat(vp, sb, active_cred, file_cred, td) 1197 struct vnode *vp; 1198 register struct stat *sb; 1199 struct ucred *active_cred; 1200 struct ucred *file_cred; 1201 struct thread *td; 1202 { 1203 struct vattr vattr; 1204 register struct vattr *vap; 1205 int error; 1206 u_short mode; 1207 1208 #ifdef MAC 1209 error = mac_vnode_check_stat(active_cred, file_cred, vp); 1210 if (error) 1211 return (error); 1212 #endif 1213 1214 vap = &vattr; 1215 1216 /* 1217 * Initialize defaults for new and unusual fields, so that file 1218 * systems which don't support these fields don't need to know 1219 * about them. 1220 */ 1221 vap->va_birthtime.tv_sec = -1; 1222 vap->va_birthtime.tv_nsec = 0; 1223 vap->va_fsid = VNOVAL; 1224 vap->va_rdev = NODEV; 1225 1226 error = VOP_GETATTR(vp, vap, active_cred); 1227 if (error) 1228 return (error); 1229 1230 /* 1231 * Zero the spare stat fields 1232 */ 1233 bzero(sb, sizeof *sb); 1234 1235 /* 1236 * Copy from vattr table 1237 */ 1238 if (vap->va_fsid != VNOVAL) 1239 sb->st_dev = vap->va_fsid; 1240 else 1241 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 1242 sb->st_ino = vap->va_fileid; 1243 mode = vap->va_mode; 1244 switch (vap->va_type) { 1245 case VREG: 1246 mode |= S_IFREG; 1247 break; 1248 case VDIR: 1249 mode |= S_IFDIR; 1250 break; 1251 case VBLK: 1252 mode |= S_IFBLK; 1253 break; 1254 case VCHR: 1255 mode |= S_IFCHR; 1256 break; 1257 case VLNK: 1258 mode |= S_IFLNK; 1259 break; 1260 case VSOCK: 1261 mode |= S_IFSOCK; 1262 break; 1263 case VFIFO: 1264 mode |= S_IFIFO; 1265 break; 1266 default: 1267 return (EBADF); 1268 }; 1269 sb->st_mode = mode; 1270 sb->st_nlink = vap->va_nlink; 1271 sb->st_uid = vap->va_uid; 1272 sb->st_gid = vap->va_gid; 1273 sb->st_rdev = vap->va_rdev; 1274 if (vap->va_size > OFF_MAX) 1275 return (EOVERFLOW); 1276 sb->st_size = vap->va_size; 1277 sb->st_atim = vap->va_atime; 1278 sb->st_mtim = vap->va_mtime; 1279 sb->st_ctim = vap->va_ctime; 1280 sb->st_birthtim = vap->va_birthtime; 1281 1282 /* 1283 * According to www.opengroup.org, the meaning of st_blksize is 1284 * "a filesystem-specific preferred I/O block size for this 1285 * object. In some filesystem types, this may vary from file 1286 * to file" 1287 * Use miminum/default of PAGE_SIZE (e.g. for VCHR). 1288 */ 1289 1290 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize); 1291 1292 sb->st_flags = vap->va_flags; 1293 if (priv_check(td, PRIV_VFS_GENERATION)) 1294 sb->st_gen = 0; 1295 else 1296 sb->st_gen = vap->va_gen; 1297 1298 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 1299 return (0); 1300 } 1301 1302 /* 1303 * File table vnode ioctl routine. 1304 */ 1305 static int 1306 vn_ioctl(fp, com, data, active_cred, td) 1307 struct file *fp; 1308 u_long com; 1309 void *data; 1310 struct ucred *active_cred; 1311 struct thread *td; 1312 { 1313 struct vnode *vp = fp->f_vnode; 1314 struct vattr vattr; 1315 int error; 1316 1317 error = ENOTTY; 1318 switch (vp->v_type) { 1319 case VREG: 1320 case VDIR: 1321 if (com == FIONREAD) { 1322 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1323 error = VOP_GETATTR(vp, &vattr, active_cred); 1324 VOP_UNLOCK(vp, 0); 1325 if (!error) 1326 *(int *)data = vattr.va_size - fp->f_offset; 1327 } 1328 if (com == FIONBIO || com == FIOASYNC) /* XXX */ 1329 error = 0; 1330 else 1331 error = VOP_IOCTL(vp, com, data, fp->f_flag, 1332 active_cred, td); 1333 break; 1334 1335 default: 1336 break; 1337 } 1338 return (error); 1339 } 1340 1341 /* 1342 * File table vnode poll routine. 1343 */ 1344 static int 1345 vn_poll(fp, events, active_cred, td) 1346 struct file *fp; 1347 int events; 1348 struct ucred *active_cred; 1349 struct thread *td; 1350 { 1351 struct vnode *vp; 1352 int error; 1353 1354 vp = fp->f_vnode; 1355 #ifdef MAC 1356 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1357 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp); 1358 VOP_UNLOCK(vp, 0); 1359 if (!error) 1360 #endif 1361 1362 error = VOP_POLL(vp, events, fp->f_cred, td); 1363 return (error); 1364 } 1365 1366 /* 1367 * Acquire the requested lock and then check for validity. LK_RETRY 1368 * permits vn_lock to return doomed vnodes. 1369 */ 1370 int 1371 _vn_lock(struct vnode *vp, int flags, char *file, int line) 1372 { 1373 int error; 1374 1375 VNASSERT((flags & LK_TYPE_MASK) != 0, vp, 1376 ("vn_lock called with no locktype.")); 1377 do { 1378 #ifdef DEBUG_VFS_LOCKS 1379 KASSERT(vp->v_holdcnt != 0, 1380 ("vn_lock %p: zero hold count", vp)); 1381 #endif 1382 error = VOP_LOCK1(vp, flags, file, line); 1383 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */ 1384 KASSERT((flags & LK_RETRY) == 0 || error == 0, 1385 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)", 1386 flags, error)); 1387 /* 1388 * Callers specify LK_RETRY if they wish to get dead vnodes. 1389 * If RETRY is not set, we return ENOENT instead. 1390 */ 1391 if (error == 0 && vp->v_iflag & VI_DOOMED && 1392 (flags & LK_RETRY) == 0) { 1393 VOP_UNLOCK(vp, 0); 1394 error = ENOENT; 1395 break; 1396 } 1397 } while (flags & LK_RETRY && error != 0); 1398 return (error); 1399 } 1400 1401 /* 1402 * File table vnode close routine. 1403 */ 1404 static int 1405 vn_closefile(fp, td) 1406 struct file *fp; 1407 struct thread *td; 1408 { 1409 struct vnode *vp; 1410 struct flock lf; 1411 int error; 1412 1413 vp = fp->f_vnode; 1414 fp->f_ops = &badfileops; 1415 1416 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) 1417 vref(vp); 1418 1419 error = vn_close(vp, fp->f_flag, fp->f_cred, td); 1420 1421 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) { 1422 lf.l_whence = SEEK_SET; 1423 lf.l_start = 0; 1424 lf.l_len = 0; 1425 lf.l_type = F_UNLCK; 1426 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); 1427 vrele(vp); 1428 } 1429 return (error); 1430 } 1431 1432 /* 1433 * Preparing to start a filesystem write operation. If the operation is 1434 * permitted, then we bump the count of operations in progress and 1435 * proceed. If a suspend request is in progress, we wait until the 1436 * suspension is over, and then proceed. 1437 */ 1438 static int 1439 vn_start_write_locked(struct mount *mp, int flags) 1440 { 1441 int error; 1442 1443 mtx_assert(MNT_MTX(mp), MA_OWNED); 1444 error = 0; 1445 1446 /* 1447 * Check on status of suspension. 1448 */ 1449 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 || 1450 mp->mnt_susp_owner != curthread) { 1451 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 1452 if (flags & V_NOWAIT) { 1453 error = EWOULDBLOCK; 1454 goto unlock; 1455 } 1456 error = msleep(&mp->mnt_flag, MNT_MTX(mp), 1457 (PUSER - 1) | (flags & PCATCH), "suspfs", 0); 1458 if (error) 1459 goto unlock; 1460 } 1461 } 1462 if (flags & V_XSLEEP) 1463 goto unlock; 1464 mp->mnt_writeopcount++; 1465 unlock: 1466 if (error != 0 || (flags & V_XSLEEP) != 0) 1467 MNT_REL(mp); 1468 MNT_IUNLOCK(mp); 1469 return (error); 1470 } 1471 1472 int 1473 vn_start_write(vp, mpp, flags) 1474 struct vnode *vp; 1475 struct mount **mpp; 1476 int flags; 1477 { 1478 struct mount *mp; 1479 int error; 1480 1481 error = 0; 1482 /* 1483 * If a vnode is provided, get and return the mount point that 1484 * to which it will write. 1485 */ 1486 if (vp != NULL) { 1487 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 1488 *mpp = NULL; 1489 if (error != EOPNOTSUPP) 1490 return (error); 1491 return (0); 1492 } 1493 } 1494 if ((mp = *mpp) == NULL) 1495 return (0); 1496 1497 /* 1498 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 1499 * a vfs_ref(). 1500 * As long as a vnode is not provided we need to acquire a 1501 * refcount for the provided mountpoint too, in order to 1502 * emulate a vfs_ref(). 1503 */ 1504 MNT_ILOCK(mp); 1505 if (vp == NULL) 1506 MNT_REF(mp); 1507 1508 return (vn_start_write_locked(mp, flags)); 1509 } 1510 1511 /* 1512 * Secondary suspension. Used by operations such as vop_inactive 1513 * routines that are needed by the higher level functions. These 1514 * are allowed to proceed until all the higher level functions have 1515 * completed (indicated by mnt_writeopcount dropping to zero). At that 1516 * time, these operations are halted until the suspension is over. 1517 */ 1518 int 1519 vn_start_secondary_write(vp, mpp, flags) 1520 struct vnode *vp; 1521 struct mount **mpp; 1522 int flags; 1523 { 1524 struct mount *mp; 1525 int error; 1526 1527 retry: 1528 if (vp != NULL) { 1529 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 1530 *mpp = NULL; 1531 if (error != EOPNOTSUPP) 1532 return (error); 1533 return (0); 1534 } 1535 } 1536 /* 1537 * If we are not suspended or have not yet reached suspended 1538 * mode, then let the operation proceed. 1539 */ 1540 if ((mp = *mpp) == NULL) 1541 return (0); 1542 1543 /* 1544 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 1545 * a vfs_ref(). 1546 * As long as a vnode is not provided we need to acquire a 1547 * refcount for the provided mountpoint too, in order to 1548 * emulate a vfs_ref(). 1549 */ 1550 MNT_ILOCK(mp); 1551 if (vp == NULL) 1552 MNT_REF(mp); 1553 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) { 1554 mp->mnt_secondary_writes++; 1555 mp->mnt_secondary_accwrites++; 1556 MNT_IUNLOCK(mp); 1557 return (0); 1558 } 1559 if (flags & V_NOWAIT) { 1560 MNT_REL(mp); 1561 MNT_IUNLOCK(mp); 1562 return (EWOULDBLOCK); 1563 } 1564 /* 1565 * Wait for the suspension to finish. 1566 */ 1567 error = msleep(&mp->mnt_flag, MNT_MTX(mp), 1568 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0); 1569 vfs_rel(mp); 1570 if (error == 0) 1571 goto retry; 1572 return (error); 1573 } 1574 1575 /* 1576 * Filesystem write operation has completed. If we are suspending and this 1577 * operation is the last one, notify the suspender that the suspension is 1578 * now in effect. 1579 */ 1580 void 1581 vn_finished_write(mp) 1582 struct mount *mp; 1583 { 1584 if (mp == NULL) 1585 return; 1586 MNT_ILOCK(mp); 1587 MNT_REL(mp); 1588 mp->mnt_writeopcount--; 1589 if (mp->mnt_writeopcount < 0) 1590 panic("vn_finished_write: neg cnt"); 1591 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1592 mp->mnt_writeopcount <= 0) 1593 wakeup(&mp->mnt_writeopcount); 1594 MNT_IUNLOCK(mp); 1595 } 1596 1597 1598 /* 1599 * Filesystem secondary write operation has completed. If we are 1600 * suspending and this operation is the last one, notify the suspender 1601 * that the suspension is now in effect. 1602 */ 1603 void 1604 vn_finished_secondary_write(mp) 1605 struct mount *mp; 1606 { 1607 if (mp == NULL) 1608 return; 1609 MNT_ILOCK(mp); 1610 MNT_REL(mp); 1611 mp->mnt_secondary_writes--; 1612 if (mp->mnt_secondary_writes < 0) 1613 panic("vn_finished_secondary_write: neg cnt"); 1614 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1615 mp->mnt_secondary_writes <= 0) 1616 wakeup(&mp->mnt_secondary_writes); 1617 MNT_IUNLOCK(mp); 1618 } 1619 1620 1621 1622 /* 1623 * Request a filesystem to suspend write operations. 1624 */ 1625 int 1626 vfs_write_suspend(mp) 1627 struct mount *mp; 1628 { 1629 int error; 1630 1631 MNT_ILOCK(mp); 1632 if (mp->mnt_susp_owner == curthread) { 1633 MNT_IUNLOCK(mp); 1634 return (EALREADY); 1635 } 1636 while (mp->mnt_kern_flag & MNTK_SUSPEND) 1637 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0); 1638 mp->mnt_kern_flag |= MNTK_SUSPEND; 1639 mp->mnt_susp_owner = curthread; 1640 if (mp->mnt_writeopcount > 0) 1641 (void) msleep(&mp->mnt_writeopcount, 1642 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0); 1643 else 1644 MNT_IUNLOCK(mp); 1645 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0) 1646 vfs_write_resume(mp, 0); 1647 return (error); 1648 } 1649 1650 /* 1651 * Request a filesystem to resume write operations. 1652 */ 1653 void 1654 vfs_write_resume(struct mount *mp, int flags) 1655 { 1656 1657 MNT_ILOCK(mp); 1658 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 1659 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner")); 1660 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 | 1661 MNTK_SUSPENDED); 1662 mp->mnt_susp_owner = NULL; 1663 wakeup(&mp->mnt_writeopcount); 1664 wakeup(&mp->mnt_flag); 1665 curthread->td_pflags &= ~TDP_IGNSUSP; 1666 if ((flags & VR_START_WRITE) != 0) { 1667 MNT_REF(mp); 1668 mp->mnt_writeopcount++; 1669 } 1670 MNT_IUNLOCK(mp); 1671 if ((flags & VR_NO_SUSPCLR) == 0) 1672 VFS_SUSP_CLEAN(mp); 1673 } else if ((flags & VR_START_WRITE) != 0) { 1674 MNT_REF(mp); 1675 vn_start_write_locked(mp, 0); 1676 } else { 1677 MNT_IUNLOCK(mp); 1678 } 1679 } 1680 1681 /* 1682 * Implement kqueues for files by translating it to vnode operation. 1683 */ 1684 static int 1685 vn_kqfilter(struct file *fp, struct knote *kn) 1686 { 1687 1688 return (VOP_KQFILTER(fp->f_vnode, kn)); 1689 } 1690 1691 /* 1692 * Simplified in-kernel wrapper calls for extended attribute access. 1693 * Both calls pass in a NULL credential, authorizing as "kernel" access. 1694 * Set IO_NODELOCKED in ioflg if the vnode is already locked. 1695 */ 1696 int 1697 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, 1698 const char *attrname, int *buflen, char *buf, struct thread *td) 1699 { 1700 struct uio auio; 1701 struct iovec iov; 1702 int error; 1703 1704 iov.iov_len = *buflen; 1705 iov.iov_base = buf; 1706 1707 auio.uio_iov = &iov; 1708 auio.uio_iovcnt = 1; 1709 auio.uio_rw = UIO_READ; 1710 auio.uio_segflg = UIO_SYSSPACE; 1711 auio.uio_td = td; 1712 auio.uio_offset = 0; 1713 auio.uio_resid = *buflen; 1714 1715 if ((ioflg & IO_NODELOCKED) == 0) 1716 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1717 1718 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1719 1720 /* authorize attribute retrieval as kernel */ 1721 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL, 1722 td); 1723 1724 if ((ioflg & IO_NODELOCKED) == 0) 1725 VOP_UNLOCK(vp, 0); 1726 1727 if (error == 0) { 1728 *buflen = *buflen - auio.uio_resid; 1729 } 1730 1731 return (error); 1732 } 1733 1734 /* 1735 * XXX failure mode if partially written? 1736 */ 1737 int 1738 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, 1739 const char *attrname, int buflen, char *buf, struct thread *td) 1740 { 1741 struct uio auio; 1742 struct iovec iov; 1743 struct mount *mp; 1744 int error; 1745 1746 iov.iov_len = buflen; 1747 iov.iov_base = buf; 1748 1749 auio.uio_iov = &iov; 1750 auio.uio_iovcnt = 1; 1751 auio.uio_rw = UIO_WRITE; 1752 auio.uio_segflg = UIO_SYSSPACE; 1753 auio.uio_td = td; 1754 auio.uio_offset = 0; 1755 auio.uio_resid = buflen; 1756 1757 if ((ioflg & IO_NODELOCKED) == 0) { 1758 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 1759 return (error); 1760 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1761 } 1762 1763 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1764 1765 /* authorize attribute setting as kernel */ 1766 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td); 1767 1768 if ((ioflg & IO_NODELOCKED) == 0) { 1769 vn_finished_write(mp); 1770 VOP_UNLOCK(vp, 0); 1771 } 1772 1773 return (error); 1774 } 1775 1776 int 1777 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, 1778 const char *attrname, struct thread *td) 1779 { 1780 struct mount *mp; 1781 int error; 1782 1783 if ((ioflg & IO_NODELOCKED) == 0) { 1784 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 1785 return (error); 1786 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1787 } 1788 1789 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1790 1791 /* authorize attribute removal as kernel */ 1792 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td); 1793 if (error == EOPNOTSUPP) 1794 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, 1795 NULL, td); 1796 1797 if ((ioflg & IO_NODELOCKED) == 0) { 1798 vn_finished_write(mp); 1799 VOP_UNLOCK(vp, 0); 1800 } 1801 1802 return (error); 1803 } 1804 1805 int 1806 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp) 1807 { 1808 struct mount *mp; 1809 int ltype, error; 1810 1811 mp = vp->v_mount; 1812 ltype = VOP_ISLOCKED(vp); 1813 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED, 1814 ("vn_vget_ino: vp not locked")); 1815 error = vfs_busy(mp, MBF_NOWAIT); 1816 if (error != 0) { 1817 vfs_ref(mp); 1818 VOP_UNLOCK(vp, 0); 1819 error = vfs_busy(mp, 0); 1820 vn_lock(vp, ltype | LK_RETRY); 1821 vfs_rel(mp); 1822 if (error != 0) 1823 return (ENOENT); 1824 if (vp->v_iflag & VI_DOOMED) { 1825 vfs_unbusy(mp); 1826 return (ENOENT); 1827 } 1828 } 1829 VOP_UNLOCK(vp, 0); 1830 error = VFS_VGET(mp, ino, lkflags, rvp); 1831 vfs_unbusy(mp); 1832 vn_lock(vp, ltype | LK_RETRY); 1833 if (vp->v_iflag & VI_DOOMED) { 1834 if (error == 0) 1835 vput(*rvp); 1836 error = ENOENT; 1837 } 1838 return (error); 1839 } 1840 1841 int 1842 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, 1843 const struct thread *td) 1844 { 1845 1846 if (vp->v_type != VREG || td == NULL) 1847 return (0); 1848 PROC_LOCK(td->td_proc); 1849 if ((uoff_t)uio->uio_offset + uio->uio_resid > 1850 lim_cur(td->td_proc, RLIMIT_FSIZE)) { 1851 kern_psignal(td->td_proc, SIGXFSZ); 1852 PROC_UNLOCK(td->td_proc); 1853 return (EFBIG); 1854 } 1855 PROC_UNLOCK(td->td_proc); 1856 return (0); 1857 } 1858 1859 int 1860 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 1861 struct thread *td) 1862 { 1863 struct vnode *vp; 1864 1865 vp = fp->f_vnode; 1866 #ifdef AUDIT 1867 vn_lock(vp, LK_SHARED | LK_RETRY); 1868 AUDIT_ARG_VNODE1(vp); 1869 VOP_UNLOCK(vp, 0); 1870 #endif 1871 return (setfmode(td, active_cred, vp, mode)); 1872 } 1873 1874 int 1875 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 1876 struct thread *td) 1877 { 1878 struct vnode *vp; 1879 1880 vp = fp->f_vnode; 1881 #ifdef AUDIT 1882 vn_lock(vp, LK_SHARED | LK_RETRY); 1883 AUDIT_ARG_VNODE1(vp); 1884 VOP_UNLOCK(vp, 0); 1885 #endif 1886 return (setfown(td, active_cred, vp, uid, gid)); 1887 } 1888 1889 void 1890 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end) 1891 { 1892 vm_object_t object; 1893 1894 if ((object = vp->v_object) == NULL) 1895 return; 1896 VM_OBJECT_WLOCK(object); 1897 vm_object_page_remove(object, start, end, 0); 1898 VM_OBJECT_WUNLOCK(object); 1899 } 1900 1901 int 1902 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred) 1903 { 1904 struct vattr va; 1905 daddr_t bn, bnp; 1906 uint64_t bsize; 1907 off_t noff; 1908 int error; 1909 1910 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA, 1911 ("Wrong command %lu", cmd)); 1912 1913 if (vn_lock(vp, LK_SHARED) != 0) 1914 return (EBADF); 1915 if (vp->v_type != VREG) { 1916 error = ENOTTY; 1917 goto unlock; 1918 } 1919 error = VOP_GETATTR(vp, &va, cred); 1920 if (error != 0) 1921 goto unlock; 1922 noff = *off; 1923 if (noff >= va.va_size) { 1924 error = ENXIO; 1925 goto unlock; 1926 } 1927 bsize = vp->v_mount->mnt_stat.f_iosize; 1928 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) { 1929 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL); 1930 if (error == EOPNOTSUPP) { 1931 error = ENOTTY; 1932 goto unlock; 1933 } 1934 if ((bnp == -1 && cmd == FIOSEEKHOLE) || 1935 (bnp != -1 && cmd == FIOSEEKDATA)) { 1936 noff = bn * bsize; 1937 if (noff < *off) 1938 noff = *off; 1939 goto unlock; 1940 } 1941 } 1942 if (noff > va.va_size) 1943 noff = va.va_size; 1944 /* noff == va.va_size. There is an implicit hole at the end of file. */ 1945 if (cmd == FIOSEEKDATA) 1946 error = ENXIO; 1947 unlock: 1948 VOP_UNLOCK(vp, 0); 1949 if (error == 0) 1950 *off = noff; 1951 return (error); 1952 } 1953