1 /*- 2 * Copyright (c) 1982, 1986, 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/fcntl.h> 43 #include <sys/file.h> 44 #include <sys/kdb.h> 45 #include <sys/stat.h> 46 #include <sys/priv.h> 47 #include <sys/proc.h> 48 #include <sys/limits.h> 49 #include <sys/lock.h> 50 #include <sys/mount.h> 51 #include <sys/mutex.h> 52 #include <sys/namei.h> 53 #include <sys/vnode.h> 54 #include <sys/bio.h> 55 #include <sys/buf.h> 56 #include <sys/filio.h> 57 #include <sys/resourcevar.h> 58 #include <sys/sx.h> 59 #include <sys/sysctl.h> 60 #include <sys/ttycom.h> 61 #include <sys/conf.h> 62 #include <sys/syslog.h> 63 #include <sys/unistd.h> 64 65 #include <security/audit/audit.h> 66 #include <security/mac/mac_framework.h> 67 68 #include <vm/vm.h> 69 #include <vm/vm_extern.h> 70 #include <vm/pmap.h> 71 #include <vm/vm_map.h> 72 #include <vm/vm_object.h> 73 #include <vm/vm_page.h> 74 75 static fo_rdwr_t vn_read; 76 static fo_rdwr_t vn_write; 77 static fo_rdwr_t vn_io_fault; 78 static fo_truncate_t vn_truncate; 79 static fo_ioctl_t vn_ioctl; 80 static fo_poll_t vn_poll; 81 static fo_kqfilter_t vn_kqfilter; 82 static fo_stat_t vn_statfile; 83 static fo_close_t vn_closefile; 84 85 struct fileops vnops = { 86 .fo_read = vn_io_fault, 87 .fo_write = vn_io_fault, 88 .fo_truncate = vn_truncate, 89 .fo_ioctl = vn_ioctl, 90 .fo_poll = vn_poll, 91 .fo_kqfilter = vn_kqfilter, 92 .fo_stat = vn_statfile, 93 .fo_close = vn_closefile, 94 .fo_chmod = vn_chmod, 95 .fo_chown = vn_chown, 96 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 97 }; 98 99 int 100 vn_open(ndp, flagp, cmode, fp) 101 struct nameidata *ndp; 102 int *flagp, cmode; 103 struct file *fp; 104 { 105 struct thread *td = ndp->ni_cnd.cn_thread; 106 107 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp)); 108 } 109 110 /* 111 * Common code for vnode open operations via a name lookup. 112 * Lookup the vnode and invoke VOP_CREATE if needed. 113 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine. 114 * 115 * Note that this does NOT free nameidata for the successful case, 116 * due to the NDINIT being done elsewhere. 117 */ 118 int 119 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, 120 struct ucred *cred, struct file *fp) 121 { 122 struct vnode *vp; 123 struct mount *mp; 124 struct thread *td = ndp->ni_cnd.cn_thread; 125 struct vattr vat; 126 struct vattr *vap = &vat; 127 int fmode, error; 128 int vfslocked, mpsafe; 129 130 mpsafe = ndp->ni_cnd.cn_flags & MPSAFE; 131 restart: 132 vfslocked = 0; 133 fmode = *flagp; 134 if (fmode & O_CREAT) { 135 ndp->ni_cnd.cn_nameiop = CREATE; 136 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | 137 MPSAFE; 138 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 139 ndp->ni_cnd.cn_flags |= FOLLOW; 140 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 141 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 142 bwillwrite(); 143 if ((error = namei(ndp)) != 0) 144 return (error); 145 vfslocked = NDHASGIANT(ndp); 146 if (!mpsafe) 147 ndp->ni_cnd.cn_flags &= ~MPSAFE; 148 if (ndp->ni_vp == NULL) { 149 VATTR_NULL(vap); 150 vap->va_type = VREG; 151 vap->va_mode = cmode; 152 if (fmode & O_EXCL) 153 vap->va_vaflags |= VA_EXCLUSIVE; 154 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) { 155 NDFREE(ndp, NDF_ONLY_PNBUF); 156 vput(ndp->ni_dvp); 157 VFS_UNLOCK_GIANT(vfslocked); 158 if ((error = vn_start_write(NULL, &mp, 159 V_XSLEEP | PCATCH)) != 0) 160 return (error); 161 goto restart; 162 } 163 #ifdef MAC 164 error = mac_vnode_check_create(cred, ndp->ni_dvp, 165 &ndp->ni_cnd, vap); 166 if (error == 0) 167 #endif 168 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 169 &ndp->ni_cnd, vap); 170 vput(ndp->ni_dvp); 171 vn_finished_write(mp); 172 if (error) { 173 VFS_UNLOCK_GIANT(vfslocked); 174 NDFREE(ndp, NDF_ONLY_PNBUF); 175 return (error); 176 } 177 fmode &= ~O_TRUNC; 178 vp = ndp->ni_vp; 179 } else { 180 if (ndp->ni_dvp == ndp->ni_vp) 181 vrele(ndp->ni_dvp); 182 else 183 vput(ndp->ni_dvp); 184 ndp->ni_dvp = NULL; 185 vp = ndp->ni_vp; 186 if (fmode & O_EXCL) { 187 error = EEXIST; 188 goto bad; 189 } 190 fmode &= ~O_CREAT; 191 } 192 } else { 193 ndp->ni_cnd.cn_nameiop = LOOKUP; 194 ndp->ni_cnd.cn_flags = ISOPEN | 195 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | 196 LOCKLEAF | MPSAFE; 197 if (!(fmode & FWRITE)) 198 ndp->ni_cnd.cn_flags |= LOCKSHARED; 199 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 200 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 201 if ((error = namei(ndp)) != 0) 202 return (error); 203 if (!mpsafe) 204 ndp->ni_cnd.cn_flags &= ~MPSAFE; 205 vfslocked = NDHASGIANT(ndp); 206 vp = ndp->ni_vp; 207 } 208 error = vn_open_vnode(vp, fmode, cred, td, fp); 209 if (error) 210 goto bad; 211 *flagp = fmode; 212 if (!mpsafe) 213 VFS_UNLOCK_GIANT(vfslocked); 214 return (0); 215 bad: 216 NDFREE(ndp, NDF_ONLY_PNBUF); 217 vput(vp); 218 VFS_UNLOCK_GIANT(vfslocked); 219 *flagp = fmode; 220 ndp->ni_vp = NULL; 221 return (error); 222 } 223 224 /* 225 * Common code for vnode open operations once a vnode is located. 226 * Check permissions, and call the VOP_OPEN routine. 227 */ 228 int 229 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, 230 struct thread *td, struct file *fp) 231 { 232 accmode_t accmode; 233 int error; 234 235 VFS_ASSERT_GIANT(vp->v_mount); 236 if (vp->v_type == VLNK) 237 return (EMLINK); 238 if (vp->v_type == VSOCK) 239 return (EOPNOTSUPP); 240 if (vp->v_type != VDIR && fmode & O_DIRECTORY) 241 return (ENOTDIR); 242 accmode = 0; 243 if (fmode & (FWRITE | O_TRUNC)) { 244 if (vp->v_type == VDIR) 245 return (EISDIR); 246 accmode |= VWRITE; 247 } 248 if (fmode & FREAD) 249 accmode |= VREAD; 250 if (fmode & FEXEC) 251 accmode |= VEXEC; 252 if ((fmode & O_APPEND) && (fmode & FWRITE)) 253 accmode |= VAPPEND; 254 #ifdef MAC 255 error = mac_vnode_check_open(cred, vp, accmode); 256 if (error) 257 return (error); 258 #endif 259 if ((fmode & O_CREAT) == 0) { 260 if (accmode & VWRITE) { 261 error = vn_writechk(vp); 262 if (error) 263 return (error); 264 } 265 if (accmode) { 266 error = VOP_ACCESS(vp, accmode, cred, td); 267 if (error) 268 return (error); 269 } 270 } 271 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0) 272 return (error); 273 274 if (fmode & FWRITE) { 275 vp->v_writecount++; 276 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", 277 __func__, vp, vp->v_writecount); 278 } 279 ASSERT_VOP_LOCKED(vp, "vn_open_vnode"); 280 return (0); 281 } 282 283 /* 284 * Check for write permissions on the specified vnode. 285 * Prototype text segments cannot be written. 286 */ 287 int 288 vn_writechk(vp) 289 register struct vnode *vp; 290 { 291 292 ASSERT_VOP_LOCKED(vp, "vn_writechk"); 293 /* 294 * If there's shared text associated with 295 * the vnode, try to free it up once. If 296 * we fail, we can't allow writing. 297 */ 298 if (vp->v_vflag & VV_TEXT) 299 return (ETXTBSY); 300 301 return (0); 302 } 303 304 /* 305 * Vnode close call 306 */ 307 int 308 vn_close(vp, flags, file_cred, td) 309 register struct vnode *vp; 310 int flags; 311 struct ucred *file_cred; 312 struct thread *td; 313 { 314 struct mount *mp; 315 int error, lock_flags; 316 317 if (!(flags & FWRITE) && vp->v_mount != NULL && 318 vp->v_mount->mnt_kern_flag & MNTK_EXTENDED_SHARED) 319 lock_flags = LK_SHARED; 320 else 321 lock_flags = LK_EXCLUSIVE; 322 323 VFS_ASSERT_GIANT(vp->v_mount); 324 325 vn_start_write(vp, &mp, V_WAIT); 326 vn_lock(vp, lock_flags | LK_RETRY); 327 if (flags & FWRITE) { 328 VNASSERT(vp->v_writecount > 0, vp, 329 ("vn_close: negative writecount")); 330 vp->v_writecount--; 331 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", 332 __func__, vp, vp->v_writecount); 333 } 334 error = VOP_CLOSE(vp, flags, file_cred, td); 335 vput(vp); 336 vn_finished_write(mp); 337 return (error); 338 } 339 340 /* 341 * Heuristic to detect sequential operation. 342 */ 343 static int 344 sequential_heuristic(struct uio *uio, struct file *fp) 345 { 346 347 if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD) 348 return (fp->f_seqcount << IO_SEQSHIFT); 349 350 /* 351 * Offset 0 is handled specially. open() sets f_seqcount to 1 so 352 * that the first I/O is normally considered to be slightly 353 * sequential. Seeking to offset 0 doesn't change sequentiality 354 * unless previous seeks have reduced f_seqcount to 0, in which 355 * case offset 0 is not special. 356 */ 357 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 358 uio->uio_offset == fp->f_nextoff) { 359 /* 360 * f_seqcount is in units of fixed-size blocks so that it 361 * depends mainly on the amount of sequential I/O and not 362 * much on the number of sequential I/O's. The fixed size 363 * of 16384 is hard-coded here since it is (not quite) just 364 * a magic size that works well here. This size is more 365 * closely related to the best I/O size for real disks than 366 * to any block size used by software. 367 */ 368 fp->f_seqcount += howmany(uio->uio_resid, 16384); 369 if (fp->f_seqcount > IO_SEQMAX) 370 fp->f_seqcount = IO_SEQMAX; 371 return (fp->f_seqcount << IO_SEQSHIFT); 372 } 373 374 /* Not sequential. Quickly draw-down sequentiality. */ 375 if (fp->f_seqcount > 1) 376 fp->f_seqcount = 1; 377 else 378 fp->f_seqcount = 0; 379 return (0); 380 } 381 382 /* 383 * Package up an I/O request on a vnode into a uio and do it. 384 */ 385 int 386 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, 387 enum uio_seg segflg, int ioflg, struct ucred *active_cred, 388 struct ucred *file_cred, ssize_t *aresid, struct thread *td) 389 { 390 struct uio auio; 391 struct iovec aiov; 392 struct mount *mp; 393 struct ucred *cred; 394 void *rl_cookie; 395 int error, lock_flags; 396 397 VFS_ASSERT_GIANT(vp->v_mount); 398 399 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; 409 410 if ((ioflg & IO_NODELOCKED) == 0) { 411 if (rw == UIO_READ) { 412 rl_cookie = vn_rangelock_rlock(vp, offset, 413 offset + len); 414 } else { 415 rl_cookie = vn_rangelock_wlock(vp, offset, 416 offset + len); 417 } 418 mp = NULL; 419 if (rw == UIO_WRITE) { 420 if (vp->v_type != VCHR && 421 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) 422 != 0) 423 goto out; 424 if (MNT_SHARED_WRITES(mp) || 425 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) 426 lock_flags = LK_SHARED; 427 else 428 lock_flags = LK_EXCLUSIVE; 429 } else 430 lock_flags = LK_SHARED; 431 vn_lock(vp, lock_flags | LK_RETRY); 432 } else 433 rl_cookie = NULL; 434 435 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 436 #ifdef MAC 437 if ((ioflg & IO_NOMACCHECK) == 0) { 438 if (rw == UIO_READ) 439 error = mac_vnode_check_read(active_cred, file_cred, 440 vp); 441 else 442 error = mac_vnode_check_write(active_cred, file_cred, 443 vp); 444 } 445 #endif 446 if (error == 0) { 447 if (file_cred != NULL) 448 cred = file_cred; 449 else 450 cred = active_cred; 451 if (rw == UIO_READ) 452 error = VOP_READ(vp, &auio, ioflg, cred); 453 else 454 error = VOP_WRITE(vp, &auio, ioflg, cred); 455 } 456 if (aresid) 457 *aresid = auio.uio_resid; 458 else 459 if (auio.uio_resid && error == 0) 460 error = EIO; 461 if ((ioflg & IO_NODELOCKED) == 0) { 462 VOP_UNLOCK(vp, 0); 463 if (mp != NULL) 464 vn_finished_write(mp); 465 } 466 out: 467 if (rl_cookie != NULL) 468 vn_rangelock_unlock(vp, rl_cookie); 469 return (error); 470 } 471 472 /* 473 * Package up an I/O request on a vnode into a uio and do it. The I/O 474 * request is split up into smaller chunks and we try to avoid saturating 475 * the buffer cache while potentially holding a vnode locked, so we 476 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield() 477 * to give other processes a chance to lock the vnode (either other processes 478 * core'ing the same binary, or unrelated processes scanning the directory). 479 */ 480 int 481 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred, 482 file_cred, aresid, td) 483 enum uio_rw rw; 484 struct vnode *vp; 485 void *base; 486 size_t len; 487 off_t offset; 488 enum uio_seg segflg; 489 int ioflg; 490 struct ucred *active_cred; 491 struct ucred *file_cred; 492 size_t *aresid; 493 struct thread *td; 494 { 495 int error = 0; 496 ssize_t iaresid; 497 498 VFS_ASSERT_GIANT(vp->v_mount); 499 500 do { 501 int chunk; 502 503 /* 504 * Force `offset' to a multiple of MAXBSIZE except possibly 505 * for the first chunk, so that filesystems only need to 506 * write full blocks except possibly for the first and last 507 * chunks. 508 */ 509 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 510 511 if (chunk > len) 512 chunk = len; 513 if (rw != UIO_READ && vp->v_type == VREG) 514 bwillwrite(); 515 iaresid = 0; 516 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 517 ioflg, active_cred, file_cred, &iaresid, td); 518 len -= chunk; /* aresid calc already includes length */ 519 if (error) 520 break; 521 offset += chunk; 522 base = (char *)base + chunk; 523 kern_yield(PRI_USER); 524 } while (len); 525 if (aresid) 526 *aresid = len + iaresid; 527 return (error); 528 } 529 530 off_t 531 foffset_lock(struct file *fp, int flags) 532 { 533 struct mtx *mtxp; 534 off_t res; 535 536 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); 537 538 #if OFF_MAX <= LONG_MAX 539 /* 540 * Caller only wants the current f_offset value. Assume that 541 * the long and shorter integer types reads are atomic. 542 */ 543 if ((flags & FOF_NOLOCK) != 0) 544 return (fp->f_offset); 545 #endif 546 547 /* 548 * According to McKusick the vn lock was protecting f_offset here. 549 * It is now protected by the FOFFSET_LOCKED flag. 550 */ 551 mtxp = mtx_pool_find(mtxpool_sleep, fp); 552 mtx_lock(mtxp); 553 if ((flags & FOF_NOLOCK) == 0) { 554 while (fp->f_vnread_flags & FOFFSET_LOCKED) { 555 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING; 556 msleep(&fp->f_vnread_flags, mtxp, PUSER -1, 557 "vofflock", 0); 558 } 559 fp->f_vnread_flags |= FOFFSET_LOCKED; 560 } 561 res = fp->f_offset; 562 mtx_unlock(mtxp); 563 return (res); 564 } 565 566 void 567 foffset_unlock(struct file *fp, off_t val, int flags) 568 { 569 struct mtx *mtxp; 570 571 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); 572 573 #if OFF_MAX <= LONG_MAX 574 if ((flags & FOF_NOLOCK) != 0) { 575 if ((flags & FOF_NOUPDATE) == 0) 576 fp->f_offset = val; 577 if ((flags & FOF_NEXTOFF) != 0) 578 fp->f_nextoff = val; 579 return; 580 } 581 #endif 582 583 mtxp = mtx_pool_find(mtxpool_sleep, fp); 584 mtx_lock(mtxp); 585 if ((flags & FOF_NOUPDATE) == 0) 586 fp->f_offset = val; 587 if ((flags & FOF_NEXTOFF) != 0) 588 fp->f_nextoff = val; 589 if ((flags & FOF_NOLOCK) == 0) { 590 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0, 591 ("Lost FOFFSET_LOCKED")); 592 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING) 593 wakeup(&fp->f_vnread_flags); 594 fp->f_vnread_flags = 0; 595 } 596 mtx_unlock(mtxp); 597 } 598 599 void 600 foffset_lock_uio(struct file *fp, struct uio *uio, int flags) 601 { 602 603 if ((flags & FOF_OFFSET) == 0) 604 uio->uio_offset = foffset_lock(fp, flags); 605 } 606 607 void 608 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags) 609 { 610 611 if ((flags & FOF_OFFSET) == 0) 612 foffset_unlock(fp, uio->uio_offset, flags); 613 } 614 615 static int 616 get_advice(struct file *fp, struct uio *uio) 617 { 618 struct mtx *mtxp; 619 int ret; 620 621 ret = POSIX_FADV_NORMAL; 622 if (fp->f_advice == NULL) 623 return (ret); 624 625 mtxp = mtx_pool_find(mtxpool_sleep, fp); 626 mtx_lock(mtxp); 627 if (uio->uio_offset >= fp->f_advice->fa_start && 628 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end) 629 ret = fp->f_advice->fa_advice; 630 mtx_unlock(mtxp); 631 return (ret); 632 } 633 634 /* 635 * File table vnode read routine. 636 */ 637 static int 638 vn_read(fp, uio, active_cred, flags, td) 639 struct file *fp; 640 struct uio *uio; 641 struct ucred *active_cred; 642 int flags; 643 struct thread *td; 644 { 645 struct vnode *vp; 646 struct mtx *mtxp; 647 int error, ioflag; 648 int advice, vfslocked; 649 off_t offset, start, end; 650 651 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", 652 uio->uio_td, td)); 653 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); 654 vp = fp->f_vnode; 655 ioflag = 0; 656 if (fp->f_flag & FNONBLOCK) 657 ioflag |= IO_NDELAY; 658 if (fp->f_flag & O_DIRECT) 659 ioflag |= IO_DIRECT; 660 advice = get_advice(fp, uio); 661 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 662 vn_lock(vp, LK_SHARED | LK_RETRY); 663 664 switch (advice) { 665 case POSIX_FADV_NORMAL: 666 case POSIX_FADV_SEQUENTIAL: 667 case POSIX_FADV_NOREUSE: 668 ioflag |= sequential_heuristic(uio, fp); 669 break; 670 case POSIX_FADV_RANDOM: 671 /* Disable read-ahead for random I/O. */ 672 break; 673 } 674 offset = uio->uio_offset; 675 676 #ifdef MAC 677 error = mac_vnode_check_read(active_cred, fp->f_cred, vp); 678 if (error == 0) 679 #endif 680 error = VOP_READ(vp, uio, ioflag, fp->f_cred); 681 fp->f_nextoff = uio->uio_offset; 682 VOP_UNLOCK(vp, 0); 683 if (error == 0 && advice == POSIX_FADV_NOREUSE && 684 offset != uio->uio_offset) { 685 /* 686 * Use POSIX_FADV_DONTNEED to flush clean pages and 687 * buffers for the backing file after a 688 * POSIX_FADV_NOREUSE read(2). To optimize the common 689 * case of using POSIX_FADV_NOREUSE with sequential 690 * access, track the previous implicit DONTNEED 691 * request and grow this request to include the 692 * current read(2) in addition to the previous 693 * DONTNEED. With purely sequential access this will 694 * cause the DONTNEED requests to continously grow to 695 * cover all of the previously read regions of the 696 * file. This allows filesystem blocks that are 697 * accessed by multiple calls to read(2) to be flushed 698 * once the last read(2) finishes. 699 */ 700 start = offset; 701 end = uio->uio_offset - 1; 702 mtxp = mtx_pool_find(mtxpool_sleep, fp); 703 mtx_lock(mtxp); 704 if (fp->f_advice != NULL && 705 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) { 706 if (start != 0 && fp->f_advice->fa_prevend + 1 == start) 707 start = fp->f_advice->fa_prevstart; 708 else if (fp->f_advice->fa_prevstart != 0 && 709 fp->f_advice->fa_prevstart == end + 1) 710 end = fp->f_advice->fa_prevend; 711 fp->f_advice->fa_prevstart = start; 712 fp->f_advice->fa_prevend = end; 713 } 714 mtx_unlock(mtxp); 715 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED); 716 } 717 VFS_UNLOCK_GIANT(vfslocked); 718 return (error); 719 } 720 721 /* 722 * File table vnode write routine. 723 */ 724 static int 725 vn_write(fp, uio, active_cred, flags, td) 726 struct file *fp; 727 struct uio *uio; 728 struct ucred *active_cred; 729 int flags; 730 struct thread *td; 731 { 732 struct vnode *vp; 733 struct mount *mp; 734 struct mtx *mtxp; 735 int error, ioflag, lock_flags; 736 int advice, vfslocked; 737 off_t offset, start, end; 738 739 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", 740 uio->uio_td, td)); 741 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); 742 vp = fp->f_vnode; 743 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 744 if (vp->v_type == VREG) 745 bwillwrite(); 746 ioflag = IO_UNIT; 747 if (vp->v_type == VREG && (fp->f_flag & O_APPEND)) 748 ioflag |= IO_APPEND; 749 if (fp->f_flag & FNONBLOCK) 750 ioflag |= IO_NDELAY; 751 if (fp->f_flag & O_DIRECT) 752 ioflag |= IO_DIRECT; 753 if ((fp->f_flag & O_FSYNC) || 754 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) 755 ioflag |= IO_SYNC; 756 mp = NULL; 757 if (vp->v_type != VCHR && 758 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) 759 goto unlock; 760 761 advice = get_advice(fp, uio); 762 763 if ((MNT_SHARED_WRITES(mp) || 764 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) && 765 (flags & FOF_OFFSET) != 0) { 766 lock_flags = LK_SHARED; 767 } else { 768 lock_flags = LK_EXCLUSIVE; 769 } 770 771 vn_lock(vp, lock_flags | LK_RETRY); 772 switch (advice) { 773 case POSIX_FADV_NORMAL: 774 case POSIX_FADV_SEQUENTIAL: 775 case POSIX_FADV_NOREUSE: 776 ioflag |= sequential_heuristic(uio, fp); 777 break; 778 case POSIX_FADV_RANDOM: 779 /* XXX: Is this correct? */ 780 break; 781 } 782 offset = uio->uio_offset; 783 784 #ifdef MAC 785 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 786 if (error == 0) 787 #endif 788 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred); 789 fp->f_nextoff = uio->uio_offset; 790 VOP_UNLOCK(vp, 0); 791 if (vp->v_type != VCHR) 792 vn_finished_write(mp); 793 if (error == 0 && advice == POSIX_FADV_NOREUSE && 794 offset != uio->uio_offset) { 795 /* 796 * Use POSIX_FADV_DONTNEED to flush clean pages and 797 * buffers for the backing file after a 798 * POSIX_FADV_NOREUSE write(2). To optimize the 799 * common case of using POSIX_FADV_NOREUSE with 800 * sequential access, track the previous implicit 801 * DONTNEED request and grow this request to include 802 * the current write(2) in addition to the previous 803 * DONTNEED. With purely sequential access this will 804 * cause the DONTNEED requests to continously grow to 805 * cover all of the previously written regions of the 806 * file. 807 * 808 * Note that the blocks just written are almost 809 * certainly still dirty, so this only works when 810 * VOP_ADVISE() calls from subsequent writes push out 811 * the data written by this write(2) once the backing 812 * buffers are clean. However, as compared to forcing 813 * IO_DIRECT, this gives much saner behavior. Write 814 * clustering is still allowed, and clean pages are 815 * merely moved to the cache page queue rather than 816 * outright thrown away. This means a subsequent 817 * read(2) can still avoid hitting the disk if the 818 * pages have not been reclaimed. 819 * 820 * This does make POSIX_FADV_NOREUSE largely useless 821 * with non-sequential access. However, sequential 822 * access is the more common use case and the flag is 823 * merely advisory. 824 */ 825 start = offset; 826 end = uio->uio_offset - 1; 827 mtxp = mtx_pool_find(mtxpool_sleep, fp); 828 mtx_lock(mtxp); 829 if (fp->f_advice != NULL && 830 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) { 831 if (start != 0 && fp->f_advice->fa_prevend + 1 == start) 832 start = fp->f_advice->fa_prevstart; 833 else if (fp->f_advice->fa_prevstart != 0 && 834 fp->f_advice->fa_prevstart == end + 1) 835 end = fp->f_advice->fa_prevend; 836 fp->f_advice->fa_prevstart = start; 837 fp->f_advice->fa_prevend = end; 838 } 839 mtx_unlock(mtxp); 840 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED); 841 } 842 843 unlock: 844 VFS_UNLOCK_GIANT(vfslocked); 845 return (error); 846 } 847 848 static const int io_hold_cnt = 16; 849 static int vn_io_fault_enable = 1; 850 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW, 851 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance"); 852 static unsigned long vn_io_faults_cnt; 853 SYSCTL_LONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD, 854 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers"); 855 856 /* 857 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to 858 * prevent the following deadlock: 859 * 860 * Assume that the thread A reads from the vnode vp1 into userspace 861 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is 862 * currently not resident, then system ends up with the call chain 863 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] -> 864 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2) 865 * which establishes lock order vp1->vn_lock, then vp2->vn_lock. 866 * If, at the same time, thread B reads from vnode vp2 into buffer buf2 867 * backed by the pages of vnode vp1, and some page in buf2 is not 868 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock. 869 * 870 * To prevent the lock order reversal and deadlock, vn_io_fault() does 871 * not allow page faults to happen during VOP_READ() or VOP_WRITE(). 872 * Instead, it first tries to do the whole range i/o with pagefaults 873 * disabled. If all pages in the i/o buffer are resident and mapped, 874 * VOP will succeed (ignoring the genuine filesystem errors). 875 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do 876 * i/o in chunks, with all pages in the chunk prefaulted and held 877 * using vm_fault_quick_hold_pages(). 878 * 879 * Filesystems using this deadlock avoidance scheme should use the 880 * array of the held pages from uio, saved in the curthread->td_ma, 881 * instead of doing uiomove(). A helper function 882 * vn_io_fault_uiomove() converts uiomove request into 883 * uiomove_fromphys() over td_ma array. 884 * 885 * Since vnode locks do not cover the whole i/o anymore, rangelocks 886 * make the current i/o request atomic with respect to other i/os and 887 * truncations. 888 */ 889 static int 890 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred, 891 int flags, struct thread *td) 892 { 893 vm_page_t ma[io_hold_cnt + 2]; 894 struct uio *uio_clone, short_uio; 895 struct iovec short_iovec[1]; 896 fo_rdwr_t *doio; 897 struct vnode *vp; 898 void *rl_cookie; 899 struct mount *mp; 900 vm_page_t *prev_td_ma; 901 int cnt, error, save, saveheld, prev_td_ma_cnt; 902 vm_offset_t addr, end; 903 vm_prot_t prot; 904 size_t len, resid; 905 ssize_t adv; 906 907 if (uio->uio_rw == UIO_READ) 908 doio = vn_read; 909 else 910 doio = vn_write; 911 vp = fp->f_vnode; 912 foffset_lock_uio(fp, uio, flags); 913 914 if (uio->uio_segflg != UIO_USERSPACE || vp->v_type != VREG || 915 ((mp = vp->v_mount) != NULL && 916 (mp->mnt_kern_flag & MNTK_NO_IOPF) == 0) || 917 !vn_io_fault_enable) { 918 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td); 919 goto out_last; 920 } 921 922 /* 923 * The UFS follows IO_UNIT directive and replays back both 924 * uio_offset and uio_resid if an error is encountered during the 925 * operation. But, since the iovec may be already advanced, 926 * uio is still in an inconsistent state. 927 * 928 * Cache a copy of the original uio, which is advanced to the redo 929 * point using UIO_NOCOPY below. 930 */ 931 uio_clone = cloneuio(uio); 932 resid = uio->uio_resid; 933 934 short_uio.uio_segflg = UIO_USERSPACE; 935 short_uio.uio_rw = uio->uio_rw; 936 short_uio.uio_td = uio->uio_td; 937 938 if (uio->uio_rw == UIO_READ) { 939 prot = VM_PROT_WRITE; 940 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset, 941 uio->uio_offset + uio->uio_resid); 942 } else { 943 prot = VM_PROT_READ; 944 if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0) 945 /* For appenders, punt and lock the whole range. */ 946 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); 947 else 948 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset, 949 uio->uio_offset + uio->uio_resid); 950 } 951 952 save = vm_fault_disable_pagefaults(); 953 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td); 954 if (error != EFAULT) 955 goto out; 956 957 atomic_add_long(&vn_io_faults_cnt, 1); 958 uio_clone->uio_segflg = UIO_NOCOPY; 959 uiomove(NULL, resid - uio->uio_resid, uio_clone); 960 uio_clone->uio_segflg = uio->uio_segflg; 961 962 saveheld = curthread_pflags_set(TDP_UIOHELD); 963 prev_td_ma = td->td_ma; 964 prev_td_ma_cnt = td->td_ma_cnt; 965 966 while (uio_clone->uio_resid != 0) { 967 len = uio_clone->uio_iov->iov_len; 968 if (len == 0) { 969 KASSERT(uio_clone->uio_iovcnt >= 1, 970 ("iovcnt underflow")); 971 uio_clone->uio_iov++; 972 uio_clone->uio_iovcnt--; 973 continue; 974 } 975 976 addr = (vm_offset_t)uio_clone->uio_iov->iov_base; 977 end = round_page(addr + len); 978 cnt = howmany(end - trunc_page(addr), PAGE_SIZE); 979 /* 980 * A perfectly misaligned address and length could cause 981 * both the start and the end of the chunk to use partial 982 * page. +2 accounts for such a situation. 983 */ 984 if (cnt > io_hold_cnt + 2) { 985 len = io_hold_cnt * PAGE_SIZE; 986 KASSERT(howmany(round_page(addr + len) - 987 trunc_page(addr), PAGE_SIZE) <= io_hold_cnt + 2, 988 ("cnt overflow")); 989 } 990 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map, 991 addr, len, prot, ma, io_hold_cnt + 2); 992 if (cnt == -1) { 993 error = EFAULT; 994 break; 995 } 996 short_uio.uio_iov = &short_iovec[0]; 997 short_iovec[0].iov_base = (void *)addr; 998 short_uio.uio_iovcnt = 1; 999 short_uio.uio_resid = short_iovec[0].iov_len = len; 1000 short_uio.uio_offset = uio_clone->uio_offset; 1001 td->td_ma = ma; 1002 td->td_ma_cnt = cnt; 1003 1004 error = doio(fp, &short_uio, active_cred, flags | FOF_OFFSET, 1005 td); 1006 vm_page_unhold_pages(ma, cnt); 1007 adv = len - short_uio.uio_resid; 1008 1009 uio_clone->uio_iov->iov_base = 1010 (char *)uio_clone->uio_iov->iov_base + adv; 1011 uio_clone->uio_iov->iov_len -= adv; 1012 uio_clone->uio_resid -= adv; 1013 uio_clone->uio_offset += adv; 1014 1015 uio->uio_resid -= adv; 1016 uio->uio_offset += adv; 1017 1018 if (error != 0 || adv == 0) 1019 break; 1020 } 1021 td->td_ma = prev_td_ma; 1022 td->td_ma_cnt = prev_td_ma_cnt; 1023 curthread_pflags_restore(saveheld); 1024 out: 1025 vm_fault_enable_pagefaults(save); 1026 vn_rangelock_unlock(vp, rl_cookie); 1027 free(uio_clone, M_IOV); 1028 out_last: 1029 foffset_unlock_uio(fp, uio, flags); 1030 return (error); 1031 } 1032 1033 /* 1034 * Helper function to perform the requested uiomove operation using 1035 * the held pages for io->uio_iov[0].iov_base buffer instead of 1036 * copyin/copyout. Access to the pages with uiomove_fromphys() 1037 * instead of iov_base prevents page faults that could occur due to 1038 * pmap_collect() invalidating the mapping created by 1039 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or 1040 * object cleanup revoking the write access from page mappings. 1041 * 1042 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove() 1043 * instead of plain uiomove(). 1044 */ 1045 int 1046 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio) 1047 { 1048 struct uio transp_uio; 1049 struct iovec transp_iov[1]; 1050 struct thread *td; 1051 size_t adv; 1052 int error, pgadv; 1053 1054 td = curthread; 1055 if ((td->td_pflags & TDP_UIOHELD) == 0 || 1056 uio->uio_segflg != UIO_USERSPACE) 1057 return (uiomove(data, xfersize, uio)); 1058 1059 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); 1060 transp_iov[0].iov_base = data; 1061 transp_uio.uio_iov = &transp_iov[0]; 1062 transp_uio.uio_iovcnt = 1; 1063 if (xfersize > uio->uio_resid) 1064 xfersize = uio->uio_resid; 1065 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize; 1066 transp_uio.uio_offset = 0; 1067 transp_uio.uio_segflg = UIO_SYSSPACE; 1068 /* 1069 * Since transp_iov points to data, and td_ma page array 1070 * corresponds to original uio->uio_iov, we need to invert the 1071 * direction of the i/o operation as passed to 1072 * uiomove_fromphys(). 1073 */ 1074 switch (uio->uio_rw) { 1075 case UIO_WRITE: 1076 transp_uio.uio_rw = UIO_READ; 1077 break; 1078 case UIO_READ: 1079 transp_uio.uio_rw = UIO_WRITE; 1080 break; 1081 } 1082 transp_uio.uio_td = uio->uio_td; 1083 error = uiomove_fromphys(td->td_ma, 1084 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK, 1085 xfersize, &transp_uio); 1086 adv = xfersize - transp_uio.uio_resid; 1087 pgadv = 1088 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) - 1089 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT); 1090 td->td_ma += pgadv; 1091 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, 1092 pgadv)); 1093 td->td_ma_cnt -= pgadv; 1094 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv; 1095 uio->uio_iov->iov_len -= adv; 1096 uio->uio_resid -= adv; 1097 uio->uio_offset += adv; 1098 return (error); 1099 } 1100 1101 /* 1102 * File table truncate routine. 1103 */ 1104 static int 1105 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred, 1106 struct thread *td) 1107 { 1108 struct vattr vattr; 1109 struct mount *mp; 1110 struct vnode *vp; 1111 void *rl_cookie; 1112 int vfslocked; 1113 int error; 1114 1115 vp = fp->f_vnode; 1116 1117 /* 1118 * Lock the whole range for truncation. Otherwise split i/o 1119 * might happen partly before and partly after the truncation. 1120 */ 1121 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); 1122 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1123 error = vn_start_write(vp, &mp, V_WAIT | PCATCH); 1124 if (error) 1125 goto out1; 1126 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1127 if (vp->v_type == VDIR) { 1128 error = EISDIR; 1129 goto out; 1130 } 1131 #ifdef MAC 1132 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 1133 if (error) 1134 goto out; 1135 #endif 1136 error = vn_writechk(vp); 1137 if (error == 0) { 1138 VATTR_NULL(&vattr); 1139 vattr.va_size = length; 1140 error = VOP_SETATTR(vp, &vattr, fp->f_cred); 1141 } 1142 out: 1143 VOP_UNLOCK(vp, 0); 1144 vn_finished_write(mp); 1145 out1: 1146 VFS_UNLOCK_GIANT(vfslocked); 1147 vn_rangelock_unlock(vp, rl_cookie); 1148 return (error); 1149 } 1150 1151 /* 1152 * File table vnode stat routine. 1153 */ 1154 static int 1155 vn_statfile(fp, sb, active_cred, td) 1156 struct file *fp; 1157 struct stat *sb; 1158 struct ucred *active_cred; 1159 struct thread *td; 1160 { 1161 struct vnode *vp = fp->f_vnode; 1162 int vfslocked; 1163 int error; 1164 1165 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1166 vn_lock(vp, LK_SHARED | LK_RETRY); 1167 error = vn_stat(vp, sb, active_cred, fp->f_cred, td); 1168 VOP_UNLOCK(vp, 0); 1169 VFS_UNLOCK_GIANT(vfslocked); 1170 1171 return (error); 1172 } 1173 1174 /* 1175 * Stat a vnode; implementation for the stat syscall 1176 */ 1177 int 1178 vn_stat(vp, sb, active_cred, file_cred, td) 1179 struct vnode *vp; 1180 register struct stat *sb; 1181 struct ucred *active_cred; 1182 struct ucred *file_cred; 1183 struct thread *td; 1184 { 1185 struct vattr vattr; 1186 register struct vattr *vap; 1187 int error; 1188 u_short mode; 1189 1190 #ifdef MAC 1191 error = mac_vnode_check_stat(active_cred, file_cred, vp); 1192 if (error) 1193 return (error); 1194 #endif 1195 1196 vap = &vattr; 1197 1198 /* 1199 * Initialize defaults for new and unusual fields, so that file 1200 * systems which don't support these fields don't need to know 1201 * about them. 1202 */ 1203 vap->va_birthtime.tv_sec = -1; 1204 vap->va_birthtime.tv_nsec = 0; 1205 vap->va_fsid = VNOVAL; 1206 vap->va_rdev = NODEV; 1207 1208 error = VOP_GETATTR(vp, vap, active_cred); 1209 if (error) 1210 return (error); 1211 1212 /* 1213 * Zero the spare stat fields 1214 */ 1215 bzero(sb, sizeof *sb); 1216 1217 /* 1218 * Copy from vattr table 1219 */ 1220 if (vap->va_fsid != VNOVAL) 1221 sb->st_dev = vap->va_fsid; 1222 else 1223 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 1224 sb->st_ino = vap->va_fileid; 1225 mode = vap->va_mode; 1226 switch (vap->va_type) { 1227 case VREG: 1228 mode |= S_IFREG; 1229 break; 1230 case VDIR: 1231 mode |= S_IFDIR; 1232 break; 1233 case VBLK: 1234 mode |= S_IFBLK; 1235 break; 1236 case VCHR: 1237 mode |= S_IFCHR; 1238 break; 1239 case VLNK: 1240 mode |= S_IFLNK; 1241 break; 1242 case VSOCK: 1243 mode |= S_IFSOCK; 1244 break; 1245 case VFIFO: 1246 mode |= S_IFIFO; 1247 break; 1248 default: 1249 return (EBADF); 1250 }; 1251 sb->st_mode = mode; 1252 sb->st_nlink = vap->va_nlink; 1253 sb->st_uid = vap->va_uid; 1254 sb->st_gid = vap->va_gid; 1255 sb->st_rdev = vap->va_rdev; 1256 if (vap->va_size > OFF_MAX) 1257 return (EOVERFLOW); 1258 sb->st_size = vap->va_size; 1259 sb->st_atim = vap->va_atime; 1260 sb->st_mtim = vap->va_mtime; 1261 sb->st_ctim = vap->va_ctime; 1262 sb->st_birthtim = vap->va_birthtime; 1263 1264 /* 1265 * According to www.opengroup.org, the meaning of st_blksize is 1266 * "a filesystem-specific preferred I/O block size for this 1267 * object. In some filesystem types, this may vary from file 1268 * to file" 1269 * Use miminum/default of PAGE_SIZE (e.g. for VCHR). 1270 */ 1271 1272 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize); 1273 1274 sb->st_flags = vap->va_flags; 1275 if (priv_check(td, PRIV_VFS_GENERATION)) 1276 sb->st_gen = 0; 1277 else 1278 sb->st_gen = vap->va_gen; 1279 1280 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 1281 return (0); 1282 } 1283 1284 /* 1285 * File table vnode ioctl routine. 1286 */ 1287 static int 1288 vn_ioctl(fp, com, data, active_cred, td) 1289 struct file *fp; 1290 u_long com; 1291 void *data; 1292 struct ucred *active_cred; 1293 struct thread *td; 1294 { 1295 struct vnode *vp = fp->f_vnode; 1296 struct vattr vattr; 1297 int vfslocked; 1298 int error; 1299 1300 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1301 error = ENOTTY; 1302 switch (vp->v_type) { 1303 case VREG: 1304 case VDIR: 1305 if (com == FIONREAD) { 1306 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1307 error = VOP_GETATTR(vp, &vattr, active_cred); 1308 VOP_UNLOCK(vp, 0); 1309 if (!error) 1310 *(int *)data = vattr.va_size - fp->f_offset; 1311 } 1312 if (com == FIONBIO || com == FIOASYNC) /* XXX */ 1313 error = 0; 1314 else 1315 error = VOP_IOCTL(vp, com, data, fp->f_flag, 1316 active_cred, td); 1317 break; 1318 1319 default: 1320 break; 1321 } 1322 VFS_UNLOCK_GIANT(vfslocked); 1323 return (error); 1324 } 1325 1326 /* 1327 * File table vnode poll routine. 1328 */ 1329 static int 1330 vn_poll(fp, events, active_cred, td) 1331 struct file *fp; 1332 int events; 1333 struct ucred *active_cred; 1334 struct thread *td; 1335 { 1336 struct vnode *vp; 1337 int vfslocked; 1338 int error; 1339 1340 vp = fp->f_vnode; 1341 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1342 #ifdef MAC 1343 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1344 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp); 1345 VOP_UNLOCK(vp, 0); 1346 if (!error) 1347 #endif 1348 1349 error = VOP_POLL(vp, events, fp->f_cred, td); 1350 VFS_UNLOCK_GIANT(vfslocked); 1351 return (error); 1352 } 1353 1354 /* 1355 * Acquire the requested lock and then check for validity. LK_RETRY 1356 * permits vn_lock to return doomed vnodes. 1357 */ 1358 int 1359 _vn_lock(struct vnode *vp, int flags, char *file, int line) 1360 { 1361 int error; 1362 1363 VNASSERT((flags & LK_TYPE_MASK) != 0, vp, 1364 ("vn_lock called with no locktype.")); 1365 do { 1366 #ifdef DEBUG_VFS_LOCKS 1367 KASSERT(vp->v_holdcnt != 0, 1368 ("vn_lock %p: zero hold count", vp)); 1369 #endif 1370 error = VOP_LOCK1(vp, flags, file, line); 1371 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */ 1372 KASSERT((flags & LK_RETRY) == 0 || error == 0, 1373 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)", 1374 flags, error)); 1375 /* 1376 * Callers specify LK_RETRY if they wish to get dead vnodes. 1377 * If RETRY is not set, we return ENOENT instead. 1378 */ 1379 if (error == 0 && vp->v_iflag & VI_DOOMED && 1380 (flags & LK_RETRY) == 0) { 1381 VOP_UNLOCK(vp, 0); 1382 error = ENOENT; 1383 break; 1384 } 1385 } while (flags & LK_RETRY && error != 0); 1386 return (error); 1387 } 1388 1389 /* 1390 * File table vnode close routine. 1391 */ 1392 static int 1393 vn_closefile(fp, td) 1394 struct file *fp; 1395 struct thread *td; 1396 { 1397 struct vnode *vp; 1398 struct flock lf; 1399 int vfslocked; 1400 int error; 1401 1402 vp = fp->f_vnode; 1403 1404 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1405 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) { 1406 lf.l_whence = SEEK_SET; 1407 lf.l_start = 0; 1408 lf.l_len = 0; 1409 lf.l_type = F_UNLCK; 1410 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); 1411 } 1412 1413 fp->f_ops = &badfileops; 1414 1415 error = vn_close(vp, fp->f_flag, fp->f_cred, td); 1416 VFS_UNLOCK_GIANT(vfslocked); 1417 return (error); 1418 } 1419 1420 /* 1421 * Preparing to start a filesystem write operation. If the operation is 1422 * permitted, then we bump the count of operations in progress and 1423 * proceed. If a suspend request is in progress, we wait until the 1424 * suspension is over, and then proceed. 1425 */ 1426 int 1427 vn_start_write(vp, mpp, flags) 1428 struct vnode *vp; 1429 struct mount **mpp; 1430 int flags; 1431 { 1432 struct mount *mp; 1433 int error; 1434 1435 error = 0; 1436 /* 1437 * If a vnode is provided, get and return the mount point that 1438 * to which it will write. 1439 */ 1440 if (vp != NULL) { 1441 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 1442 *mpp = NULL; 1443 if (error != EOPNOTSUPP) 1444 return (error); 1445 return (0); 1446 } 1447 } 1448 if ((mp = *mpp) == NULL) 1449 return (0); 1450 1451 /* 1452 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 1453 * a vfs_ref(). 1454 * As long as a vnode is not provided we need to acquire a 1455 * refcount for the provided mountpoint too, in order to 1456 * emulate a vfs_ref(). 1457 */ 1458 MNT_ILOCK(mp); 1459 if (vp == NULL) 1460 MNT_REF(mp); 1461 1462 /* 1463 * Check on status of suspension. 1464 */ 1465 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 || 1466 mp->mnt_susp_owner != curthread) { 1467 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 1468 if (flags & V_NOWAIT) { 1469 error = EWOULDBLOCK; 1470 goto unlock; 1471 } 1472 error = msleep(&mp->mnt_flag, MNT_MTX(mp), 1473 (PUSER - 1) | (flags & PCATCH), "suspfs", 0); 1474 if (error) 1475 goto unlock; 1476 } 1477 } 1478 if (flags & V_XSLEEP) 1479 goto unlock; 1480 mp->mnt_writeopcount++; 1481 unlock: 1482 if (error != 0 || (flags & V_XSLEEP) != 0) 1483 MNT_REL(mp); 1484 MNT_IUNLOCK(mp); 1485 return (error); 1486 } 1487 1488 /* 1489 * Secondary suspension. Used by operations such as vop_inactive 1490 * routines that are needed by the higher level functions. These 1491 * are allowed to proceed until all the higher level functions have 1492 * completed (indicated by mnt_writeopcount dropping to zero). At that 1493 * time, these operations are halted until the suspension is over. 1494 */ 1495 int 1496 vn_start_secondary_write(vp, mpp, flags) 1497 struct vnode *vp; 1498 struct mount **mpp; 1499 int flags; 1500 { 1501 struct mount *mp; 1502 int error; 1503 1504 retry: 1505 if (vp != NULL) { 1506 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 1507 *mpp = NULL; 1508 if (error != EOPNOTSUPP) 1509 return (error); 1510 return (0); 1511 } 1512 } 1513 /* 1514 * If we are not suspended or have not yet reached suspended 1515 * mode, then let the operation proceed. 1516 */ 1517 if ((mp = *mpp) == NULL) 1518 return (0); 1519 1520 /* 1521 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 1522 * a vfs_ref(). 1523 * As long as a vnode is not provided we need to acquire a 1524 * refcount for the provided mountpoint too, in order to 1525 * emulate a vfs_ref(). 1526 */ 1527 MNT_ILOCK(mp); 1528 if (vp == NULL) 1529 MNT_REF(mp); 1530 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) { 1531 mp->mnt_secondary_writes++; 1532 mp->mnt_secondary_accwrites++; 1533 MNT_IUNLOCK(mp); 1534 return (0); 1535 } 1536 if (flags & V_NOWAIT) { 1537 MNT_REL(mp); 1538 MNT_IUNLOCK(mp); 1539 return (EWOULDBLOCK); 1540 } 1541 /* 1542 * Wait for the suspension to finish. 1543 */ 1544 error = msleep(&mp->mnt_flag, MNT_MTX(mp), 1545 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0); 1546 vfs_rel(mp); 1547 if (error == 0) 1548 goto retry; 1549 return (error); 1550 } 1551 1552 /* 1553 * Filesystem write operation has completed. If we are suspending and this 1554 * operation is the last one, notify the suspender that the suspension is 1555 * now in effect. 1556 */ 1557 void 1558 vn_finished_write(mp) 1559 struct mount *mp; 1560 { 1561 if (mp == NULL) 1562 return; 1563 MNT_ILOCK(mp); 1564 MNT_REL(mp); 1565 mp->mnt_writeopcount--; 1566 if (mp->mnt_writeopcount < 0) 1567 panic("vn_finished_write: neg cnt"); 1568 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1569 mp->mnt_writeopcount <= 0) 1570 wakeup(&mp->mnt_writeopcount); 1571 MNT_IUNLOCK(mp); 1572 } 1573 1574 1575 /* 1576 * Filesystem secondary write operation has completed. If we are 1577 * suspending and this operation is the last one, notify the suspender 1578 * that the suspension is now in effect. 1579 */ 1580 void 1581 vn_finished_secondary_write(mp) 1582 struct mount *mp; 1583 { 1584 if (mp == NULL) 1585 return; 1586 MNT_ILOCK(mp); 1587 MNT_REL(mp); 1588 mp->mnt_secondary_writes--; 1589 if (mp->mnt_secondary_writes < 0) 1590 panic("vn_finished_secondary_write: neg cnt"); 1591 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1592 mp->mnt_secondary_writes <= 0) 1593 wakeup(&mp->mnt_secondary_writes); 1594 MNT_IUNLOCK(mp); 1595 } 1596 1597 1598 1599 /* 1600 * Request a filesystem to suspend write operations. 1601 */ 1602 int 1603 vfs_write_suspend(mp) 1604 struct mount *mp; 1605 { 1606 int error; 1607 1608 MNT_ILOCK(mp); 1609 if (mp->mnt_susp_owner == curthread) { 1610 MNT_IUNLOCK(mp); 1611 return (EALREADY); 1612 } 1613 while (mp->mnt_kern_flag & MNTK_SUSPEND) 1614 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0); 1615 mp->mnt_kern_flag |= MNTK_SUSPEND; 1616 mp->mnt_susp_owner = curthread; 1617 if (mp->mnt_writeopcount > 0) 1618 (void) msleep(&mp->mnt_writeopcount, 1619 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0); 1620 else 1621 MNT_IUNLOCK(mp); 1622 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0) 1623 vfs_write_resume(mp); 1624 return (error); 1625 } 1626 1627 /* 1628 * Request a filesystem to resume write operations. 1629 */ 1630 void 1631 vfs_write_resume(mp) 1632 struct mount *mp; 1633 { 1634 1635 MNT_ILOCK(mp); 1636 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 1637 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner")); 1638 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 | 1639 MNTK_SUSPENDED); 1640 mp->mnt_susp_owner = NULL; 1641 wakeup(&mp->mnt_writeopcount); 1642 wakeup(&mp->mnt_flag); 1643 curthread->td_pflags &= ~TDP_IGNSUSP; 1644 MNT_IUNLOCK(mp); 1645 VFS_SUSP_CLEAN(mp); 1646 } else 1647 MNT_IUNLOCK(mp); 1648 } 1649 1650 /* 1651 * Implement kqueues for files by translating it to vnode operation. 1652 */ 1653 static int 1654 vn_kqfilter(struct file *fp, struct knote *kn) 1655 { 1656 int vfslocked; 1657 int error; 1658 1659 vfslocked = VFS_LOCK_GIANT(fp->f_vnode->v_mount); 1660 error = VOP_KQFILTER(fp->f_vnode, kn); 1661 VFS_UNLOCK_GIANT(vfslocked); 1662 1663 return error; 1664 } 1665 1666 /* 1667 * Simplified in-kernel wrapper calls for extended attribute access. 1668 * Both calls pass in a NULL credential, authorizing as "kernel" access. 1669 * Set IO_NODELOCKED in ioflg if the vnode is already locked. 1670 */ 1671 int 1672 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, 1673 const char *attrname, int *buflen, char *buf, struct thread *td) 1674 { 1675 struct uio auio; 1676 struct iovec iov; 1677 int error; 1678 1679 iov.iov_len = *buflen; 1680 iov.iov_base = buf; 1681 1682 auio.uio_iov = &iov; 1683 auio.uio_iovcnt = 1; 1684 auio.uio_rw = UIO_READ; 1685 auio.uio_segflg = UIO_SYSSPACE; 1686 auio.uio_td = td; 1687 auio.uio_offset = 0; 1688 auio.uio_resid = *buflen; 1689 1690 if ((ioflg & IO_NODELOCKED) == 0) 1691 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1692 1693 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1694 1695 /* authorize attribute retrieval as kernel */ 1696 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL, 1697 td); 1698 1699 if ((ioflg & IO_NODELOCKED) == 0) 1700 VOP_UNLOCK(vp, 0); 1701 1702 if (error == 0) { 1703 *buflen = *buflen - auio.uio_resid; 1704 } 1705 1706 return (error); 1707 } 1708 1709 /* 1710 * XXX failure mode if partially written? 1711 */ 1712 int 1713 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, 1714 const char *attrname, int buflen, char *buf, struct thread *td) 1715 { 1716 struct uio auio; 1717 struct iovec iov; 1718 struct mount *mp; 1719 int error; 1720 1721 iov.iov_len = buflen; 1722 iov.iov_base = buf; 1723 1724 auio.uio_iov = &iov; 1725 auio.uio_iovcnt = 1; 1726 auio.uio_rw = UIO_WRITE; 1727 auio.uio_segflg = UIO_SYSSPACE; 1728 auio.uio_td = td; 1729 auio.uio_offset = 0; 1730 auio.uio_resid = buflen; 1731 1732 if ((ioflg & IO_NODELOCKED) == 0) { 1733 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 1734 return (error); 1735 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1736 } 1737 1738 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1739 1740 /* authorize attribute setting as kernel */ 1741 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td); 1742 1743 if ((ioflg & IO_NODELOCKED) == 0) { 1744 vn_finished_write(mp); 1745 VOP_UNLOCK(vp, 0); 1746 } 1747 1748 return (error); 1749 } 1750 1751 int 1752 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, 1753 const char *attrname, struct thread *td) 1754 { 1755 struct mount *mp; 1756 int error; 1757 1758 if ((ioflg & IO_NODELOCKED) == 0) { 1759 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 1760 return (error); 1761 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1762 } 1763 1764 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1765 1766 /* authorize attribute removal as kernel */ 1767 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td); 1768 if (error == EOPNOTSUPP) 1769 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, 1770 NULL, td); 1771 1772 if ((ioflg & IO_NODELOCKED) == 0) { 1773 vn_finished_write(mp); 1774 VOP_UNLOCK(vp, 0); 1775 } 1776 1777 return (error); 1778 } 1779 1780 int 1781 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp) 1782 { 1783 struct mount *mp; 1784 int ltype, error; 1785 1786 mp = vp->v_mount; 1787 ltype = VOP_ISLOCKED(vp); 1788 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED, 1789 ("vn_vget_ino: vp not locked")); 1790 error = vfs_busy(mp, MBF_NOWAIT); 1791 if (error != 0) { 1792 vfs_ref(mp); 1793 VOP_UNLOCK(vp, 0); 1794 error = vfs_busy(mp, 0); 1795 vn_lock(vp, ltype | LK_RETRY); 1796 vfs_rel(mp); 1797 if (error != 0) 1798 return (ENOENT); 1799 if (vp->v_iflag & VI_DOOMED) { 1800 vfs_unbusy(mp); 1801 return (ENOENT); 1802 } 1803 } 1804 VOP_UNLOCK(vp, 0); 1805 error = VFS_VGET(mp, ino, lkflags, rvp); 1806 vfs_unbusy(mp); 1807 vn_lock(vp, ltype | LK_RETRY); 1808 if (vp->v_iflag & VI_DOOMED) { 1809 if (error == 0) 1810 vput(*rvp); 1811 error = ENOENT; 1812 } 1813 return (error); 1814 } 1815 1816 int 1817 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, 1818 const struct thread *td) 1819 { 1820 1821 if (vp->v_type != VREG || td == NULL) 1822 return (0); 1823 PROC_LOCK(td->td_proc); 1824 if ((uoff_t)uio->uio_offset + uio->uio_resid > 1825 lim_cur(td->td_proc, RLIMIT_FSIZE)) { 1826 kern_psignal(td->td_proc, SIGXFSZ); 1827 PROC_UNLOCK(td->td_proc); 1828 return (EFBIG); 1829 } 1830 PROC_UNLOCK(td->td_proc); 1831 return (0); 1832 } 1833 1834 int 1835 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 1836 struct thread *td) 1837 { 1838 struct vnode *vp; 1839 int error, vfslocked; 1840 1841 vp = fp->f_vnode; 1842 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1843 #ifdef AUDIT 1844 vn_lock(vp, LK_SHARED | LK_RETRY); 1845 AUDIT_ARG_VNODE1(vp); 1846 VOP_UNLOCK(vp, 0); 1847 #endif 1848 error = setfmode(td, active_cred, vp, mode); 1849 VFS_UNLOCK_GIANT(vfslocked); 1850 return (error); 1851 } 1852 1853 int 1854 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 1855 struct thread *td) 1856 { 1857 struct vnode *vp; 1858 int error, vfslocked; 1859 1860 vp = fp->f_vnode; 1861 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1862 #ifdef AUDIT 1863 vn_lock(vp, LK_SHARED | LK_RETRY); 1864 AUDIT_ARG_VNODE1(vp); 1865 VOP_UNLOCK(vp, 0); 1866 #endif 1867 error = setfown(td, active_cred, vp, uid, gid); 1868 VFS_UNLOCK_GIANT(vfslocked); 1869 return (error); 1870 } 1871 1872 void 1873 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end) 1874 { 1875 vm_object_t object; 1876 1877 if ((object = vp->v_object) == NULL) 1878 return; 1879 VM_OBJECT_LOCK(object); 1880 vm_object_page_remove(object, start, end, 0); 1881 VM_OBJECT_UNLOCK(object); 1882 } 1883 1884 int 1885 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred) 1886 { 1887 struct vattr va; 1888 daddr_t bn, bnp; 1889 uint64_t bsize; 1890 off_t noff; 1891 int error; 1892 1893 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA, 1894 ("Wrong command %lu", cmd)); 1895 1896 if (vn_lock(vp, LK_SHARED) != 0) 1897 return (EBADF); 1898 if (vp->v_type != VREG) { 1899 error = ENOTTY; 1900 goto unlock; 1901 } 1902 error = VOP_GETATTR(vp, &va, cred); 1903 if (error != 0) 1904 goto unlock; 1905 noff = *off; 1906 if (noff >= va.va_size) { 1907 error = ENXIO; 1908 goto unlock; 1909 } 1910 bsize = vp->v_mount->mnt_stat.f_iosize; 1911 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) { 1912 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL); 1913 if (error == EOPNOTSUPP) { 1914 error = ENOTTY; 1915 goto unlock; 1916 } 1917 if ((bnp == -1 && cmd == FIOSEEKHOLE) || 1918 (bnp != -1 && cmd == FIOSEEKDATA)) { 1919 noff = bn * bsize; 1920 if (noff < *off) 1921 noff = *off; 1922 goto unlock; 1923 } 1924 } 1925 if (noff > va.va_size) 1926 noff = va.va_size; 1927 /* noff == va.va_size. There is an implicit hole at the end of file. */ 1928 if (cmd == FIOSEEKDATA) 1929 error = ENXIO; 1930 unlock: 1931 VOP_UNLOCK(vp, 0); 1932 if (error == 0) 1933 *off = noff; 1934 return (error); 1935 } 1936