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 * Copyright (c) 2012 Konstantin Belousov <kib@FreeBSD.org> 11 * Copyright (c) 2013, 2014 The FreeBSD Foundation 12 * 13 * Portions of this software were developed by Konstantin Belousov 14 * under sponsorship from the FreeBSD Foundation. 15 * 16 * Redistribution and use in source and binary forms, with or without 17 * modification, are permitted provided that the following conditions 18 * are met: 19 * 1. Redistributions of source code must retain the above copyright 20 * notice, this list of conditions and the following disclaimer. 21 * 2. Redistributions in binary form must reproduce the above copyright 22 * notice, this list of conditions and the following disclaimer in the 23 * documentation and/or other materials provided with the distribution. 24 * 4. Neither the name of the University nor the names of its contributors 25 * may be used to endorse or promote products derived from this software 26 * without specific prior written permission. 27 * 28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 38 * SUCH DAMAGE. 39 * 40 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94 41 */ 42 43 #include <sys/cdefs.h> 44 __FBSDID("$FreeBSD$"); 45 46 #include "opt_hwpmc_hooks.h" 47 48 #include <sys/param.h> 49 #include <sys/systm.h> 50 #include <sys/disk.h> 51 #include <sys/fail.h> 52 #include <sys/fcntl.h> 53 #include <sys/file.h> 54 #include <sys/kdb.h> 55 #include <sys/stat.h> 56 #include <sys/priv.h> 57 #include <sys/proc.h> 58 #include <sys/limits.h> 59 #include <sys/lock.h> 60 #include <sys/mman.h> 61 #include <sys/mount.h> 62 #include <sys/mutex.h> 63 #include <sys/namei.h> 64 #include <sys/vnode.h> 65 #include <sys/bio.h> 66 #include <sys/buf.h> 67 #include <sys/filio.h> 68 #include <sys/resourcevar.h> 69 #include <sys/rwlock.h> 70 #include <sys/sx.h> 71 #include <sys/sysctl.h> 72 #include <sys/ttycom.h> 73 #include <sys/conf.h> 74 #include <sys/syslog.h> 75 #include <sys/unistd.h> 76 #include <sys/user.h> 77 78 #include <security/audit/audit.h> 79 #include <security/mac/mac_framework.h> 80 81 #include <vm/vm.h> 82 #include <vm/vm_extern.h> 83 #include <vm/pmap.h> 84 #include <vm/vm_map.h> 85 #include <vm/vm_object.h> 86 #include <vm/vm_page.h> 87 #include <vm/vnode_pager.h> 88 89 #ifdef HWPMC_HOOKS 90 #include <sys/pmckern.h> 91 #endif 92 93 static fo_rdwr_t vn_read; 94 static fo_rdwr_t vn_write; 95 static fo_rdwr_t vn_io_fault; 96 static fo_truncate_t vn_truncate; 97 static fo_ioctl_t vn_ioctl; 98 static fo_poll_t vn_poll; 99 static fo_kqfilter_t vn_kqfilter; 100 static fo_stat_t vn_statfile; 101 static fo_close_t vn_closefile; 102 static fo_mmap_t vn_mmap; 103 104 struct fileops vnops = { 105 .fo_read = vn_io_fault, 106 .fo_write = vn_io_fault, 107 .fo_truncate = vn_truncate, 108 .fo_ioctl = vn_ioctl, 109 .fo_poll = vn_poll, 110 .fo_kqfilter = vn_kqfilter, 111 .fo_stat = vn_statfile, 112 .fo_close = vn_closefile, 113 .fo_chmod = vn_chmod, 114 .fo_chown = vn_chown, 115 .fo_sendfile = vn_sendfile, 116 .fo_seek = vn_seek, 117 .fo_fill_kinfo = vn_fill_kinfo, 118 .fo_mmap = vn_mmap, 119 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 120 }; 121 122 static const int io_hold_cnt = 16; 123 static int vn_io_fault_enable = 1; 124 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW, 125 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance"); 126 static int vn_io_fault_prefault = 0; 127 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW, 128 &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting"); 129 static u_long vn_io_faults_cnt; 130 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD, 131 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers"); 132 133 /* 134 * Returns true if vn_io_fault mode of handling the i/o request should 135 * be used. 136 */ 137 static bool 138 do_vn_io_fault(struct vnode *vp, struct uio *uio) 139 { 140 struct mount *mp; 141 142 return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG && 143 (mp = vp->v_mount) != NULL && 144 (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable); 145 } 146 147 /* 148 * Structure used to pass arguments to vn_io_fault1(), to do either 149 * file- or vnode-based I/O calls. 150 */ 151 struct vn_io_fault_args { 152 enum { 153 VN_IO_FAULT_FOP, 154 VN_IO_FAULT_VOP 155 } kind; 156 struct ucred *cred; 157 int flags; 158 union { 159 struct fop_args_tag { 160 struct file *fp; 161 fo_rdwr_t *doio; 162 } fop_args; 163 struct vop_args_tag { 164 struct vnode *vp; 165 } vop_args; 166 } args; 167 }; 168 169 static int vn_io_fault1(struct vnode *vp, struct uio *uio, 170 struct vn_io_fault_args *args, struct thread *td); 171 172 int 173 vn_open(ndp, flagp, cmode, fp) 174 struct nameidata *ndp; 175 int *flagp, cmode; 176 struct file *fp; 177 { 178 struct thread *td = ndp->ni_cnd.cn_thread; 179 180 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp)); 181 } 182 183 /* 184 * Common code for vnode open operations via a name lookup. 185 * Lookup the vnode and invoke VOP_CREATE if needed. 186 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine. 187 * 188 * Note that this does NOT free nameidata for the successful case, 189 * due to the NDINIT being done elsewhere. 190 */ 191 int 192 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, 193 struct ucred *cred, struct file *fp) 194 { 195 struct vnode *vp; 196 struct mount *mp; 197 struct thread *td = ndp->ni_cnd.cn_thread; 198 struct vattr vat; 199 struct vattr *vap = &vat; 200 int fmode, error; 201 202 restart: 203 fmode = *flagp; 204 if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT | 205 O_EXCL | O_DIRECTORY)) 206 return (EINVAL); 207 else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) { 208 ndp->ni_cnd.cn_nameiop = CREATE; 209 /* 210 * Set NOCACHE to avoid flushing the cache when 211 * rolling in many files at once. 212 */ 213 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE; 214 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 215 ndp->ni_cnd.cn_flags |= FOLLOW; 216 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 217 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 218 if (vn_open_flags & VN_OPEN_NOCAPCHECK) 219 ndp->ni_cnd.cn_flags |= NOCAPCHECK; 220 bwillwrite(); 221 if ((error = namei(ndp)) != 0) 222 return (error); 223 if (ndp->ni_vp == NULL) { 224 VATTR_NULL(vap); 225 vap->va_type = VREG; 226 vap->va_mode = cmode; 227 if (fmode & O_EXCL) 228 vap->va_vaflags |= VA_EXCLUSIVE; 229 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) { 230 NDFREE(ndp, NDF_ONLY_PNBUF); 231 vput(ndp->ni_dvp); 232 if ((error = vn_start_write(NULL, &mp, 233 V_XSLEEP | PCATCH)) != 0) 234 return (error); 235 goto restart; 236 } 237 if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0) 238 ndp->ni_cnd.cn_flags |= MAKEENTRY; 239 #ifdef MAC 240 error = mac_vnode_check_create(cred, ndp->ni_dvp, 241 &ndp->ni_cnd, vap); 242 if (error == 0) 243 #endif 244 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 245 &ndp->ni_cnd, vap); 246 vput(ndp->ni_dvp); 247 vn_finished_write(mp); 248 if (error) { 249 NDFREE(ndp, NDF_ONLY_PNBUF); 250 return (error); 251 } 252 fmode &= ~O_TRUNC; 253 vp = ndp->ni_vp; 254 } else { 255 if (ndp->ni_dvp == ndp->ni_vp) 256 vrele(ndp->ni_dvp); 257 else 258 vput(ndp->ni_dvp); 259 ndp->ni_dvp = NULL; 260 vp = ndp->ni_vp; 261 if (fmode & O_EXCL) { 262 error = EEXIST; 263 goto bad; 264 } 265 fmode &= ~O_CREAT; 266 } 267 } else { 268 ndp->ni_cnd.cn_nameiop = LOOKUP; 269 ndp->ni_cnd.cn_flags = ISOPEN | 270 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF; 271 if (!(fmode & FWRITE)) 272 ndp->ni_cnd.cn_flags |= LOCKSHARED; 273 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 274 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 275 if (vn_open_flags & VN_OPEN_NOCAPCHECK) 276 ndp->ni_cnd.cn_flags |= NOCAPCHECK; 277 if ((error = namei(ndp)) != 0) 278 return (error); 279 vp = ndp->ni_vp; 280 } 281 error = vn_open_vnode(vp, fmode, cred, td, fp); 282 if (error) 283 goto bad; 284 *flagp = fmode; 285 return (0); 286 bad: 287 NDFREE(ndp, NDF_ONLY_PNBUF); 288 vput(vp); 289 *flagp = fmode; 290 ndp->ni_vp = NULL; 291 return (error); 292 } 293 294 /* 295 * Common code for vnode open operations once a vnode is located. 296 * Check permissions, and call the VOP_OPEN routine. 297 */ 298 int 299 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, 300 struct thread *td, struct file *fp) 301 { 302 accmode_t accmode; 303 struct flock lf; 304 int error, lock_flags, type; 305 306 if (vp->v_type == VLNK) 307 return (EMLINK); 308 if (vp->v_type == VSOCK) 309 return (EOPNOTSUPP); 310 if (vp->v_type != VDIR && fmode & O_DIRECTORY) 311 return (ENOTDIR); 312 accmode = 0; 313 if (fmode & (FWRITE | O_TRUNC)) { 314 if (vp->v_type == VDIR) 315 return (EISDIR); 316 accmode |= VWRITE; 317 } 318 if (fmode & FREAD) 319 accmode |= VREAD; 320 if (fmode & FEXEC) 321 accmode |= VEXEC; 322 if ((fmode & O_APPEND) && (fmode & FWRITE)) 323 accmode |= VAPPEND; 324 #ifdef MAC 325 if (fmode & O_CREAT) 326 accmode |= VCREAT; 327 if (fmode & O_VERIFY) 328 accmode |= VVERIFY; 329 error = mac_vnode_check_open(cred, vp, accmode); 330 if (error) 331 return (error); 332 333 accmode &= ~(VCREAT | VVERIFY); 334 #endif 335 if ((fmode & O_CREAT) == 0) { 336 if (accmode & VWRITE) { 337 error = vn_writechk(vp); 338 if (error) 339 return (error); 340 } 341 if (accmode) { 342 error = VOP_ACCESS(vp, accmode, cred, td); 343 if (error) 344 return (error); 345 } 346 } 347 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) 348 vn_lock(vp, LK_UPGRADE | LK_RETRY); 349 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0) 350 return (error); 351 352 if (fmode & (O_EXLOCK | O_SHLOCK)) { 353 KASSERT(fp != NULL, ("open with flock requires fp")); 354 lock_flags = VOP_ISLOCKED(vp); 355 VOP_UNLOCK(vp, 0); 356 lf.l_whence = SEEK_SET; 357 lf.l_start = 0; 358 lf.l_len = 0; 359 if (fmode & O_EXLOCK) 360 lf.l_type = F_WRLCK; 361 else 362 lf.l_type = F_RDLCK; 363 type = F_FLOCK; 364 if ((fmode & FNONBLOCK) == 0) 365 type |= F_WAIT; 366 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type); 367 if (error == 0) 368 fp->f_flag |= FHASLOCK; 369 vn_lock(vp, lock_flags | LK_RETRY); 370 if (error == 0 && vp->v_iflag & VI_DOOMED) 371 error = ENOENT; 372 373 /* 374 * Another thread might have used this vnode as an 375 * executable while the vnode lock was dropped. 376 * Ensure the vnode is still able to be opened for 377 * writing after the lock has been obtained. 378 */ 379 if (error == 0 && accmode & VWRITE) 380 error = vn_writechk(vp); 381 382 if (error != 0) { 383 fp->f_flag |= FOPENFAILED; 384 fp->f_vnode = vp; 385 if (fp->f_ops == &badfileops) { 386 fp->f_type = DTYPE_VNODE; 387 fp->f_ops = &vnops; 388 } 389 vref(vp); 390 } 391 } 392 if (error == 0 && fmode & FWRITE) { 393 VOP_ADD_WRITECOUNT(vp, 1); 394 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", 395 __func__, vp, vp->v_writecount); 396 } 397 ASSERT_VOP_LOCKED(vp, "vn_open_vnode"); 398 return (error); 399 } 400 401 /* 402 * Check for write permissions on the specified vnode. 403 * Prototype text segments cannot be written. 404 */ 405 int 406 vn_writechk(vp) 407 register struct vnode *vp; 408 { 409 410 ASSERT_VOP_LOCKED(vp, "vn_writechk"); 411 /* 412 * If there's shared text associated with 413 * the vnode, try to free it up once. If 414 * we fail, we can't allow writing. 415 */ 416 if (VOP_IS_TEXT(vp)) 417 return (ETXTBSY); 418 419 return (0); 420 } 421 422 /* 423 * Vnode close call 424 */ 425 int 426 vn_close(vp, flags, file_cred, td) 427 register struct vnode *vp; 428 int flags; 429 struct ucred *file_cred; 430 struct thread *td; 431 { 432 struct mount *mp; 433 int error, lock_flags; 434 435 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 && 436 MNT_EXTENDED_SHARED(vp->v_mount)) 437 lock_flags = LK_SHARED; 438 else 439 lock_flags = LK_EXCLUSIVE; 440 441 vn_start_write(vp, &mp, V_WAIT); 442 vn_lock(vp, lock_flags | LK_RETRY); 443 AUDIT_ARG_VNODE1(vp); 444 if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) { 445 VNASSERT(vp->v_writecount > 0, vp, 446 ("vn_close: negative writecount")); 447 VOP_ADD_WRITECOUNT(vp, -1); 448 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", 449 __func__, vp, vp->v_writecount); 450 } 451 error = VOP_CLOSE(vp, flags, file_cred, td); 452 vput(vp); 453 vn_finished_write(mp); 454 return (error); 455 } 456 457 /* 458 * Heuristic to detect sequential operation. 459 */ 460 static int 461 sequential_heuristic(struct uio *uio, struct file *fp) 462 { 463 464 ASSERT_VOP_LOCKED(fp->f_vnode, __func__); 465 if (fp->f_flag & FRDAHEAD) 466 return (fp->f_seqcount << IO_SEQSHIFT); 467 468 /* 469 * Offset 0 is handled specially. open() sets f_seqcount to 1 so 470 * that the first I/O is normally considered to be slightly 471 * sequential. Seeking to offset 0 doesn't change sequentiality 472 * unless previous seeks have reduced f_seqcount to 0, in which 473 * case offset 0 is not special. 474 */ 475 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 476 uio->uio_offset == fp->f_nextoff) { 477 /* 478 * f_seqcount is in units of fixed-size blocks so that it 479 * depends mainly on the amount of sequential I/O and not 480 * much on the number of sequential I/O's. The fixed size 481 * of 16384 is hard-coded here since it is (not quite) just 482 * a magic size that works well here. This size is more 483 * closely related to the best I/O size for real disks than 484 * to any block size used by software. 485 */ 486 fp->f_seqcount += howmany(uio->uio_resid, 16384); 487 if (fp->f_seqcount > IO_SEQMAX) 488 fp->f_seqcount = IO_SEQMAX; 489 return (fp->f_seqcount << IO_SEQSHIFT); 490 } 491 492 /* Not sequential. Quickly draw-down sequentiality. */ 493 if (fp->f_seqcount > 1) 494 fp->f_seqcount = 1; 495 else 496 fp->f_seqcount = 0; 497 return (0); 498 } 499 500 /* 501 * Package up an I/O request on a vnode into a uio and do it. 502 */ 503 int 504 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, 505 enum uio_seg segflg, int ioflg, struct ucred *active_cred, 506 struct ucred *file_cred, ssize_t *aresid, struct thread *td) 507 { 508 struct uio auio; 509 struct iovec aiov; 510 struct mount *mp; 511 struct ucred *cred; 512 void *rl_cookie; 513 struct vn_io_fault_args args; 514 int error, lock_flags; 515 516 auio.uio_iov = &aiov; 517 auio.uio_iovcnt = 1; 518 aiov.iov_base = base; 519 aiov.iov_len = len; 520 auio.uio_resid = len; 521 auio.uio_offset = offset; 522 auio.uio_segflg = segflg; 523 auio.uio_rw = rw; 524 auio.uio_td = td; 525 error = 0; 526 527 if ((ioflg & IO_NODELOCKED) == 0) { 528 if ((ioflg & IO_RANGELOCKED) == 0) { 529 if (rw == UIO_READ) { 530 rl_cookie = vn_rangelock_rlock(vp, offset, 531 offset + len); 532 } else { 533 rl_cookie = vn_rangelock_wlock(vp, offset, 534 offset + len); 535 } 536 } else 537 rl_cookie = NULL; 538 mp = NULL; 539 if (rw == UIO_WRITE) { 540 if (vp->v_type != VCHR && 541 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) 542 != 0) 543 goto out; 544 if (MNT_SHARED_WRITES(mp) || 545 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) 546 lock_flags = LK_SHARED; 547 else 548 lock_flags = LK_EXCLUSIVE; 549 } else 550 lock_flags = LK_SHARED; 551 vn_lock(vp, lock_flags | LK_RETRY); 552 } else 553 rl_cookie = NULL; 554 555 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 556 #ifdef MAC 557 if ((ioflg & IO_NOMACCHECK) == 0) { 558 if (rw == UIO_READ) 559 error = mac_vnode_check_read(active_cred, file_cred, 560 vp); 561 else 562 error = mac_vnode_check_write(active_cred, file_cred, 563 vp); 564 } 565 #endif 566 if (error == 0) { 567 if (file_cred != NULL) 568 cred = file_cred; 569 else 570 cred = active_cred; 571 if (do_vn_io_fault(vp, &auio)) { 572 args.kind = VN_IO_FAULT_VOP; 573 args.cred = cred; 574 args.flags = ioflg; 575 args.args.vop_args.vp = vp; 576 error = vn_io_fault1(vp, &auio, &args, td); 577 } else if (rw == UIO_READ) { 578 error = VOP_READ(vp, &auio, ioflg, cred); 579 } else /* if (rw == UIO_WRITE) */ { 580 error = VOP_WRITE(vp, &auio, ioflg, cred); 581 } 582 } 583 if (aresid) 584 *aresid = auio.uio_resid; 585 else 586 if (auio.uio_resid && error == 0) 587 error = EIO; 588 if ((ioflg & IO_NODELOCKED) == 0) { 589 VOP_UNLOCK(vp, 0); 590 if (mp != NULL) 591 vn_finished_write(mp); 592 } 593 out: 594 if (rl_cookie != NULL) 595 vn_rangelock_unlock(vp, rl_cookie); 596 return (error); 597 } 598 599 /* 600 * Package up an I/O request on a vnode into a uio and do it. The I/O 601 * request is split up into smaller chunks and we try to avoid saturating 602 * the buffer cache while potentially holding a vnode locked, so we 603 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield() 604 * to give other processes a chance to lock the vnode (either other processes 605 * core'ing the same binary, or unrelated processes scanning the directory). 606 */ 607 int 608 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred, 609 file_cred, aresid, td) 610 enum uio_rw rw; 611 struct vnode *vp; 612 void *base; 613 size_t len; 614 off_t offset; 615 enum uio_seg segflg; 616 int ioflg; 617 struct ucred *active_cred; 618 struct ucred *file_cred; 619 size_t *aresid; 620 struct thread *td; 621 { 622 int error = 0; 623 ssize_t iaresid; 624 625 do { 626 int chunk; 627 628 /* 629 * Force `offset' to a multiple of MAXBSIZE except possibly 630 * for the first chunk, so that filesystems only need to 631 * write full blocks except possibly for the first and last 632 * chunks. 633 */ 634 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 635 636 if (chunk > len) 637 chunk = len; 638 if (rw != UIO_READ && vp->v_type == VREG) 639 bwillwrite(); 640 iaresid = 0; 641 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 642 ioflg, active_cred, file_cred, &iaresid, td); 643 len -= chunk; /* aresid calc already includes length */ 644 if (error) 645 break; 646 offset += chunk; 647 base = (char *)base + chunk; 648 kern_yield(PRI_USER); 649 } while (len); 650 if (aresid) 651 *aresid = len + iaresid; 652 return (error); 653 } 654 655 off_t 656 foffset_lock(struct file *fp, int flags) 657 { 658 struct mtx *mtxp; 659 off_t res; 660 661 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); 662 663 #if OFF_MAX <= LONG_MAX 664 /* 665 * Caller only wants the current f_offset value. Assume that 666 * the long and shorter integer types reads are atomic. 667 */ 668 if ((flags & FOF_NOLOCK) != 0) 669 return (fp->f_offset); 670 #endif 671 672 /* 673 * According to McKusick the vn lock was protecting f_offset here. 674 * It is now protected by the FOFFSET_LOCKED flag. 675 */ 676 mtxp = mtx_pool_find(mtxpool_sleep, fp); 677 mtx_lock(mtxp); 678 if ((flags & FOF_NOLOCK) == 0) { 679 while (fp->f_vnread_flags & FOFFSET_LOCKED) { 680 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING; 681 msleep(&fp->f_vnread_flags, mtxp, PUSER -1, 682 "vofflock", 0); 683 } 684 fp->f_vnread_flags |= FOFFSET_LOCKED; 685 } 686 res = fp->f_offset; 687 mtx_unlock(mtxp); 688 return (res); 689 } 690 691 void 692 foffset_unlock(struct file *fp, off_t val, int flags) 693 { 694 struct mtx *mtxp; 695 696 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); 697 698 #if OFF_MAX <= LONG_MAX 699 if ((flags & FOF_NOLOCK) != 0) { 700 if ((flags & FOF_NOUPDATE) == 0) 701 fp->f_offset = val; 702 if ((flags & FOF_NEXTOFF) != 0) 703 fp->f_nextoff = val; 704 return; 705 } 706 #endif 707 708 mtxp = mtx_pool_find(mtxpool_sleep, fp); 709 mtx_lock(mtxp); 710 if ((flags & FOF_NOUPDATE) == 0) 711 fp->f_offset = val; 712 if ((flags & FOF_NEXTOFF) != 0) 713 fp->f_nextoff = val; 714 if ((flags & FOF_NOLOCK) == 0) { 715 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0, 716 ("Lost FOFFSET_LOCKED")); 717 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING) 718 wakeup(&fp->f_vnread_flags); 719 fp->f_vnread_flags = 0; 720 } 721 mtx_unlock(mtxp); 722 } 723 724 void 725 foffset_lock_uio(struct file *fp, struct uio *uio, int flags) 726 { 727 728 if ((flags & FOF_OFFSET) == 0) 729 uio->uio_offset = foffset_lock(fp, flags); 730 } 731 732 void 733 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags) 734 { 735 736 if ((flags & FOF_OFFSET) == 0) 737 foffset_unlock(fp, uio->uio_offset, flags); 738 } 739 740 static int 741 get_advice(struct file *fp, struct uio *uio) 742 { 743 struct mtx *mtxp; 744 int ret; 745 746 ret = POSIX_FADV_NORMAL; 747 if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG) 748 return (ret); 749 750 mtxp = mtx_pool_find(mtxpool_sleep, fp); 751 mtx_lock(mtxp); 752 if (fp->f_advice != NULL && 753 uio->uio_offset >= fp->f_advice->fa_start && 754 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end) 755 ret = fp->f_advice->fa_advice; 756 mtx_unlock(mtxp); 757 return (ret); 758 } 759 760 /* 761 * File table vnode read routine. 762 */ 763 static int 764 vn_read(fp, uio, active_cred, flags, td) 765 struct file *fp; 766 struct uio *uio; 767 struct ucred *active_cred; 768 int flags; 769 struct thread *td; 770 { 771 struct vnode *vp; 772 off_t orig_offset; 773 int error, ioflag; 774 int advice; 775 776 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", 777 uio->uio_td, td)); 778 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); 779 vp = fp->f_vnode; 780 ioflag = 0; 781 if (fp->f_flag & FNONBLOCK) 782 ioflag |= IO_NDELAY; 783 if (fp->f_flag & O_DIRECT) 784 ioflag |= IO_DIRECT; 785 advice = get_advice(fp, uio); 786 vn_lock(vp, LK_SHARED | LK_RETRY); 787 788 switch (advice) { 789 case POSIX_FADV_NORMAL: 790 case POSIX_FADV_SEQUENTIAL: 791 case POSIX_FADV_NOREUSE: 792 ioflag |= sequential_heuristic(uio, fp); 793 break; 794 case POSIX_FADV_RANDOM: 795 /* Disable read-ahead for random I/O. */ 796 break; 797 } 798 orig_offset = uio->uio_offset; 799 800 #ifdef MAC 801 error = mac_vnode_check_read(active_cred, fp->f_cred, vp); 802 if (error == 0) 803 #endif 804 error = VOP_READ(vp, uio, ioflag, fp->f_cred); 805 fp->f_nextoff = uio->uio_offset; 806 VOP_UNLOCK(vp, 0); 807 if (error == 0 && advice == POSIX_FADV_NOREUSE && 808 orig_offset != uio->uio_offset) 809 /* 810 * Use POSIX_FADV_DONTNEED to flush pages and buffers 811 * for the backing file after a POSIX_FADV_NOREUSE 812 * read(2). 813 */ 814 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1, 815 POSIX_FADV_DONTNEED); 816 return (error); 817 } 818 819 /* 820 * File table vnode write routine. 821 */ 822 static int 823 vn_write(fp, uio, active_cred, flags, td) 824 struct file *fp; 825 struct uio *uio; 826 struct ucred *active_cred; 827 int flags; 828 struct thread *td; 829 { 830 struct vnode *vp; 831 struct mount *mp; 832 off_t orig_offset; 833 int error, ioflag, lock_flags; 834 int advice; 835 836 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", 837 uio->uio_td, td)); 838 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); 839 vp = fp->f_vnode; 840 if (vp->v_type == VREG) 841 bwillwrite(); 842 ioflag = IO_UNIT; 843 if (vp->v_type == VREG && (fp->f_flag & O_APPEND)) 844 ioflag |= IO_APPEND; 845 if (fp->f_flag & FNONBLOCK) 846 ioflag |= IO_NDELAY; 847 if (fp->f_flag & O_DIRECT) 848 ioflag |= IO_DIRECT; 849 if ((fp->f_flag & O_FSYNC) || 850 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) 851 ioflag |= IO_SYNC; 852 mp = NULL; 853 if (vp->v_type != VCHR && 854 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) 855 goto unlock; 856 857 advice = get_advice(fp, uio); 858 859 if (MNT_SHARED_WRITES(mp) || 860 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) { 861 lock_flags = LK_SHARED; 862 } else { 863 lock_flags = LK_EXCLUSIVE; 864 } 865 866 vn_lock(vp, lock_flags | LK_RETRY); 867 switch (advice) { 868 case POSIX_FADV_NORMAL: 869 case POSIX_FADV_SEQUENTIAL: 870 case POSIX_FADV_NOREUSE: 871 ioflag |= sequential_heuristic(uio, fp); 872 break; 873 case POSIX_FADV_RANDOM: 874 /* XXX: Is this correct? */ 875 break; 876 } 877 orig_offset = uio->uio_offset; 878 879 #ifdef MAC 880 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 881 if (error == 0) 882 #endif 883 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred); 884 fp->f_nextoff = uio->uio_offset; 885 VOP_UNLOCK(vp, 0); 886 if (vp->v_type != VCHR) 887 vn_finished_write(mp); 888 if (error == 0 && advice == POSIX_FADV_NOREUSE && 889 orig_offset != uio->uio_offset) 890 /* 891 * Use POSIX_FADV_DONTNEED to flush pages and buffers 892 * for the backing file after a POSIX_FADV_NOREUSE 893 * write(2). 894 */ 895 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1, 896 POSIX_FADV_DONTNEED); 897 unlock: 898 return (error); 899 } 900 901 /* 902 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to 903 * prevent the following deadlock: 904 * 905 * Assume that the thread A reads from the vnode vp1 into userspace 906 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is 907 * currently not resident, then system ends up with the call chain 908 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] -> 909 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2) 910 * which establishes lock order vp1->vn_lock, then vp2->vn_lock. 911 * If, at the same time, thread B reads from vnode vp2 into buffer buf2 912 * backed by the pages of vnode vp1, and some page in buf2 is not 913 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock. 914 * 915 * To prevent the lock order reversal and deadlock, vn_io_fault() does 916 * not allow page faults to happen during VOP_READ() or VOP_WRITE(). 917 * Instead, it first tries to do the whole range i/o with pagefaults 918 * disabled. If all pages in the i/o buffer are resident and mapped, 919 * VOP will succeed (ignoring the genuine filesystem errors). 920 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do 921 * i/o in chunks, with all pages in the chunk prefaulted and held 922 * using vm_fault_quick_hold_pages(). 923 * 924 * Filesystems using this deadlock avoidance scheme should use the 925 * array of the held pages from uio, saved in the curthread->td_ma, 926 * instead of doing uiomove(). A helper function 927 * vn_io_fault_uiomove() converts uiomove request into 928 * uiomove_fromphys() over td_ma array. 929 * 930 * Since vnode locks do not cover the whole i/o anymore, rangelocks 931 * make the current i/o request atomic with respect to other i/os and 932 * truncations. 933 */ 934 935 /* 936 * Decode vn_io_fault_args and perform the corresponding i/o. 937 */ 938 static int 939 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio, 940 struct thread *td) 941 { 942 943 switch (args->kind) { 944 case VN_IO_FAULT_FOP: 945 return ((args->args.fop_args.doio)(args->args.fop_args.fp, 946 uio, args->cred, args->flags, td)); 947 case VN_IO_FAULT_VOP: 948 if (uio->uio_rw == UIO_READ) { 949 return (VOP_READ(args->args.vop_args.vp, uio, 950 args->flags, args->cred)); 951 } else if (uio->uio_rw == UIO_WRITE) { 952 return (VOP_WRITE(args->args.vop_args.vp, uio, 953 args->flags, args->cred)); 954 } 955 break; 956 } 957 panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind, 958 uio->uio_rw); 959 } 960 961 static int 962 vn_io_fault_touch(char *base, const struct uio *uio) 963 { 964 int r; 965 966 r = fubyte(base); 967 if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1)) 968 return (EFAULT); 969 return (0); 970 } 971 972 static int 973 vn_io_fault_prefault_user(const struct uio *uio) 974 { 975 char *base; 976 const struct iovec *iov; 977 size_t len; 978 ssize_t resid; 979 int error, i; 980 981 KASSERT(uio->uio_segflg == UIO_USERSPACE, 982 ("vn_io_fault_prefault userspace")); 983 984 error = i = 0; 985 iov = uio->uio_iov; 986 resid = uio->uio_resid; 987 base = iov->iov_base; 988 len = iov->iov_len; 989 while (resid > 0) { 990 error = vn_io_fault_touch(base, uio); 991 if (error != 0) 992 break; 993 if (len < PAGE_SIZE) { 994 if (len != 0) { 995 error = vn_io_fault_touch(base + len - 1, uio); 996 if (error != 0) 997 break; 998 resid -= len; 999 } 1000 if (++i >= uio->uio_iovcnt) 1001 break; 1002 iov = uio->uio_iov + i; 1003 base = iov->iov_base; 1004 len = iov->iov_len; 1005 } else { 1006 len -= PAGE_SIZE; 1007 base += PAGE_SIZE; 1008 resid -= PAGE_SIZE; 1009 } 1010 } 1011 return (error); 1012 } 1013 1014 /* 1015 * Common code for vn_io_fault(), agnostic to the kind of i/o request. 1016 * Uses vn_io_fault_doio() to make the call to an actual i/o function. 1017 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request 1018 * into args and call vn_io_fault1() to handle faults during the user 1019 * mode buffer accesses. 1020 */ 1021 static int 1022 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, 1023 struct thread *td) 1024 { 1025 vm_page_t ma[io_hold_cnt + 2]; 1026 struct uio *uio_clone, short_uio; 1027 struct iovec short_iovec[1]; 1028 vm_page_t *prev_td_ma; 1029 vm_prot_t prot; 1030 vm_offset_t addr, end; 1031 size_t len, resid; 1032 ssize_t adv; 1033 int error, cnt, save, saveheld, prev_td_ma_cnt; 1034 1035 if (vn_io_fault_prefault) { 1036 error = vn_io_fault_prefault_user(uio); 1037 if (error != 0) 1038 return (error); /* Or ignore ? */ 1039 } 1040 1041 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ; 1042 1043 /* 1044 * The UFS follows IO_UNIT directive and replays back both 1045 * uio_offset and uio_resid if an error is encountered during the 1046 * operation. But, since the iovec may be already advanced, 1047 * uio is still in an inconsistent state. 1048 * 1049 * Cache a copy of the original uio, which is advanced to the redo 1050 * point using UIO_NOCOPY below. 1051 */ 1052 uio_clone = cloneuio(uio); 1053 resid = uio->uio_resid; 1054 1055 short_uio.uio_segflg = UIO_USERSPACE; 1056 short_uio.uio_rw = uio->uio_rw; 1057 short_uio.uio_td = uio->uio_td; 1058 1059 save = vm_fault_disable_pagefaults(); 1060 error = vn_io_fault_doio(args, uio, td); 1061 if (error != EFAULT) 1062 goto out; 1063 1064 atomic_add_long(&vn_io_faults_cnt, 1); 1065 uio_clone->uio_segflg = UIO_NOCOPY; 1066 uiomove(NULL, resid - uio->uio_resid, uio_clone); 1067 uio_clone->uio_segflg = uio->uio_segflg; 1068 1069 saveheld = curthread_pflags_set(TDP_UIOHELD); 1070 prev_td_ma = td->td_ma; 1071 prev_td_ma_cnt = td->td_ma_cnt; 1072 1073 while (uio_clone->uio_resid != 0) { 1074 len = uio_clone->uio_iov->iov_len; 1075 if (len == 0) { 1076 KASSERT(uio_clone->uio_iovcnt >= 1, 1077 ("iovcnt underflow")); 1078 uio_clone->uio_iov++; 1079 uio_clone->uio_iovcnt--; 1080 continue; 1081 } 1082 if (len > io_hold_cnt * PAGE_SIZE) 1083 len = io_hold_cnt * PAGE_SIZE; 1084 addr = (uintptr_t)uio_clone->uio_iov->iov_base; 1085 end = round_page(addr + len); 1086 if (end < addr) { 1087 error = EFAULT; 1088 break; 1089 } 1090 cnt = atop(end - trunc_page(addr)); 1091 /* 1092 * A perfectly misaligned address and length could cause 1093 * both the start and the end of the chunk to use partial 1094 * page. +2 accounts for such a situation. 1095 */ 1096 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map, 1097 addr, len, prot, ma, io_hold_cnt + 2); 1098 if (cnt == -1) { 1099 error = EFAULT; 1100 break; 1101 } 1102 short_uio.uio_iov = &short_iovec[0]; 1103 short_iovec[0].iov_base = (void *)addr; 1104 short_uio.uio_iovcnt = 1; 1105 short_uio.uio_resid = short_iovec[0].iov_len = len; 1106 short_uio.uio_offset = uio_clone->uio_offset; 1107 td->td_ma = ma; 1108 td->td_ma_cnt = cnt; 1109 1110 error = vn_io_fault_doio(args, &short_uio, td); 1111 vm_page_unhold_pages(ma, cnt); 1112 adv = len - short_uio.uio_resid; 1113 1114 uio_clone->uio_iov->iov_base = 1115 (char *)uio_clone->uio_iov->iov_base + adv; 1116 uio_clone->uio_iov->iov_len -= adv; 1117 uio_clone->uio_resid -= adv; 1118 uio_clone->uio_offset += adv; 1119 1120 uio->uio_resid -= adv; 1121 uio->uio_offset += adv; 1122 1123 if (error != 0 || adv == 0) 1124 break; 1125 } 1126 td->td_ma = prev_td_ma; 1127 td->td_ma_cnt = prev_td_ma_cnt; 1128 curthread_pflags_restore(saveheld); 1129 out: 1130 vm_fault_enable_pagefaults(save); 1131 free(uio_clone, M_IOV); 1132 return (error); 1133 } 1134 1135 static int 1136 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred, 1137 int flags, struct thread *td) 1138 { 1139 fo_rdwr_t *doio; 1140 struct vnode *vp; 1141 void *rl_cookie; 1142 struct vn_io_fault_args args; 1143 int error; 1144 1145 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write; 1146 vp = fp->f_vnode; 1147 foffset_lock_uio(fp, uio, flags); 1148 if (do_vn_io_fault(vp, uio)) { 1149 args.kind = VN_IO_FAULT_FOP; 1150 args.args.fop_args.fp = fp; 1151 args.args.fop_args.doio = doio; 1152 args.cred = active_cred; 1153 args.flags = flags | FOF_OFFSET; 1154 if (uio->uio_rw == UIO_READ) { 1155 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset, 1156 uio->uio_offset + uio->uio_resid); 1157 } else if ((fp->f_flag & O_APPEND) != 0 || 1158 (flags & FOF_OFFSET) == 0) { 1159 /* For appenders, punt and lock the whole range. */ 1160 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); 1161 } else { 1162 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset, 1163 uio->uio_offset + uio->uio_resid); 1164 } 1165 error = vn_io_fault1(vp, uio, &args, td); 1166 vn_rangelock_unlock(vp, rl_cookie); 1167 } else { 1168 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td); 1169 } 1170 foffset_unlock_uio(fp, uio, flags); 1171 return (error); 1172 } 1173 1174 /* 1175 * Helper function to perform the requested uiomove operation using 1176 * the held pages for io->uio_iov[0].iov_base buffer instead of 1177 * copyin/copyout. Access to the pages with uiomove_fromphys() 1178 * instead of iov_base prevents page faults that could occur due to 1179 * pmap_collect() invalidating the mapping created by 1180 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or 1181 * object cleanup revoking the write access from page mappings. 1182 * 1183 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove() 1184 * instead of plain uiomove(). 1185 */ 1186 int 1187 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio) 1188 { 1189 struct uio transp_uio; 1190 struct iovec transp_iov[1]; 1191 struct thread *td; 1192 size_t adv; 1193 int error, pgadv; 1194 1195 td = curthread; 1196 if ((td->td_pflags & TDP_UIOHELD) == 0 || 1197 uio->uio_segflg != UIO_USERSPACE) 1198 return (uiomove(data, xfersize, uio)); 1199 1200 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); 1201 transp_iov[0].iov_base = data; 1202 transp_uio.uio_iov = &transp_iov[0]; 1203 transp_uio.uio_iovcnt = 1; 1204 if (xfersize > uio->uio_resid) 1205 xfersize = uio->uio_resid; 1206 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize; 1207 transp_uio.uio_offset = 0; 1208 transp_uio.uio_segflg = UIO_SYSSPACE; 1209 /* 1210 * Since transp_iov points to data, and td_ma page array 1211 * corresponds to original uio->uio_iov, we need to invert the 1212 * direction of the i/o operation as passed to 1213 * uiomove_fromphys(). 1214 */ 1215 switch (uio->uio_rw) { 1216 case UIO_WRITE: 1217 transp_uio.uio_rw = UIO_READ; 1218 break; 1219 case UIO_READ: 1220 transp_uio.uio_rw = UIO_WRITE; 1221 break; 1222 } 1223 transp_uio.uio_td = uio->uio_td; 1224 error = uiomove_fromphys(td->td_ma, 1225 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK, 1226 xfersize, &transp_uio); 1227 adv = xfersize - transp_uio.uio_resid; 1228 pgadv = 1229 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) - 1230 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT); 1231 td->td_ma += pgadv; 1232 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, 1233 pgadv)); 1234 td->td_ma_cnt -= pgadv; 1235 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv; 1236 uio->uio_iov->iov_len -= adv; 1237 uio->uio_resid -= adv; 1238 uio->uio_offset += adv; 1239 return (error); 1240 } 1241 1242 int 1243 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize, 1244 struct uio *uio) 1245 { 1246 struct thread *td; 1247 vm_offset_t iov_base; 1248 int cnt, pgadv; 1249 1250 td = curthread; 1251 if ((td->td_pflags & TDP_UIOHELD) == 0 || 1252 uio->uio_segflg != UIO_USERSPACE) 1253 return (uiomove_fromphys(ma, offset, xfersize, uio)); 1254 1255 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); 1256 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize; 1257 iov_base = (vm_offset_t)uio->uio_iov->iov_base; 1258 switch (uio->uio_rw) { 1259 case UIO_WRITE: 1260 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma, 1261 offset, cnt); 1262 break; 1263 case UIO_READ: 1264 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK, 1265 cnt); 1266 break; 1267 } 1268 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT); 1269 td->td_ma += pgadv; 1270 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, 1271 pgadv)); 1272 td->td_ma_cnt -= pgadv; 1273 uio->uio_iov->iov_base = (char *)(iov_base + cnt); 1274 uio->uio_iov->iov_len -= cnt; 1275 uio->uio_resid -= cnt; 1276 uio->uio_offset += cnt; 1277 return (0); 1278 } 1279 1280 1281 /* 1282 * File table truncate routine. 1283 */ 1284 static int 1285 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred, 1286 struct thread *td) 1287 { 1288 struct vattr vattr; 1289 struct mount *mp; 1290 struct vnode *vp; 1291 void *rl_cookie; 1292 int error; 1293 1294 vp = fp->f_vnode; 1295 1296 /* 1297 * Lock the whole range for truncation. Otherwise split i/o 1298 * might happen partly before and partly after the truncation. 1299 */ 1300 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); 1301 error = vn_start_write(vp, &mp, V_WAIT | PCATCH); 1302 if (error) 1303 goto out1; 1304 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1305 if (vp->v_type == VDIR) { 1306 error = EISDIR; 1307 goto out; 1308 } 1309 #ifdef MAC 1310 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 1311 if (error) 1312 goto out; 1313 #endif 1314 error = vn_writechk(vp); 1315 if (error == 0) { 1316 VATTR_NULL(&vattr); 1317 vattr.va_size = length; 1318 if ((fp->f_flag & O_FSYNC) != 0) 1319 vattr.va_vaflags |= VA_SYNC; 1320 error = VOP_SETATTR(vp, &vattr, fp->f_cred); 1321 } 1322 out: 1323 VOP_UNLOCK(vp, 0); 1324 vn_finished_write(mp); 1325 out1: 1326 vn_rangelock_unlock(vp, rl_cookie); 1327 return (error); 1328 } 1329 1330 /* 1331 * File table vnode stat routine. 1332 */ 1333 static int 1334 vn_statfile(fp, sb, active_cred, td) 1335 struct file *fp; 1336 struct stat *sb; 1337 struct ucred *active_cred; 1338 struct thread *td; 1339 { 1340 struct vnode *vp = fp->f_vnode; 1341 int error; 1342 1343 vn_lock(vp, LK_SHARED | LK_RETRY); 1344 error = vn_stat(vp, sb, active_cred, fp->f_cred, td); 1345 VOP_UNLOCK(vp, 0); 1346 1347 return (error); 1348 } 1349 1350 /* 1351 * Stat a vnode; implementation for the stat syscall 1352 */ 1353 int 1354 vn_stat(vp, sb, active_cred, file_cred, td) 1355 struct vnode *vp; 1356 register struct stat *sb; 1357 struct ucred *active_cred; 1358 struct ucred *file_cred; 1359 struct thread *td; 1360 { 1361 struct vattr vattr; 1362 register struct vattr *vap; 1363 int error; 1364 u_short mode; 1365 1366 AUDIT_ARG_VNODE1(vp); 1367 #ifdef MAC 1368 error = mac_vnode_check_stat(active_cred, file_cred, vp); 1369 if (error) 1370 return (error); 1371 #endif 1372 1373 vap = &vattr; 1374 1375 /* 1376 * Initialize defaults for new and unusual fields, so that file 1377 * systems which don't support these fields don't need to know 1378 * about them. 1379 */ 1380 vap->va_birthtime.tv_sec = -1; 1381 vap->va_birthtime.tv_nsec = 0; 1382 vap->va_fsid = VNOVAL; 1383 vap->va_rdev = NODEV; 1384 1385 error = VOP_GETATTR(vp, vap, active_cred); 1386 if (error) 1387 return (error); 1388 1389 /* 1390 * Zero the spare stat fields 1391 */ 1392 bzero(sb, sizeof *sb); 1393 1394 /* 1395 * Copy from vattr table 1396 */ 1397 if (vap->va_fsid != VNOVAL) 1398 sb->st_dev = vap->va_fsid; 1399 else 1400 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 1401 sb->st_ino = vap->va_fileid; 1402 mode = vap->va_mode; 1403 switch (vap->va_type) { 1404 case VREG: 1405 mode |= S_IFREG; 1406 break; 1407 case VDIR: 1408 mode |= S_IFDIR; 1409 break; 1410 case VBLK: 1411 mode |= S_IFBLK; 1412 break; 1413 case VCHR: 1414 mode |= S_IFCHR; 1415 break; 1416 case VLNK: 1417 mode |= S_IFLNK; 1418 break; 1419 case VSOCK: 1420 mode |= S_IFSOCK; 1421 break; 1422 case VFIFO: 1423 mode |= S_IFIFO; 1424 break; 1425 default: 1426 return (EBADF); 1427 } 1428 sb->st_mode = mode; 1429 sb->st_nlink = vap->va_nlink; 1430 sb->st_uid = vap->va_uid; 1431 sb->st_gid = vap->va_gid; 1432 sb->st_rdev = vap->va_rdev; 1433 if (vap->va_size > OFF_MAX) 1434 return (EOVERFLOW); 1435 sb->st_size = vap->va_size; 1436 sb->st_atim = vap->va_atime; 1437 sb->st_mtim = vap->va_mtime; 1438 sb->st_ctim = vap->va_ctime; 1439 sb->st_birthtim = vap->va_birthtime; 1440 1441 /* 1442 * According to www.opengroup.org, the meaning of st_blksize is 1443 * "a filesystem-specific preferred I/O block size for this 1444 * object. In some filesystem types, this may vary from file 1445 * to file" 1446 * Use miminum/default of PAGE_SIZE (e.g. for VCHR). 1447 */ 1448 1449 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize); 1450 1451 sb->st_flags = vap->va_flags; 1452 if (priv_check(td, PRIV_VFS_GENERATION)) 1453 sb->st_gen = 0; 1454 else 1455 sb->st_gen = vap->va_gen; 1456 1457 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 1458 return (0); 1459 } 1460 1461 /* 1462 * File table vnode ioctl routine. 1463 */ 1464 static int 1465 vn_ioctl(fp, com, data, active_cred, td) 1466 struct file *fp; 1467 u_long com; 1468 void *data; 1469 struct ucred *active_cred; 1470 struct thread *td; 1471 { 1472 struct vattr vattr; 1473 struct vnode *vp; 1474 int error; 1475 1476 vp = fp->f_vnode; 1477 switch (vp->v_type) { 1478 case VDIR: 1479 case VREG: 1480 switch (com) { 1481 case FIONREAD: 1482 vn_lock(vp, LK_SHARED | LK_RETRY); 1483 error = VOP_GETATTR(vp, &vattr, active_cred); 1484 VOP_UNLOCK(vp, 0); 1485 if (error == 0) 1486 *(int *)data = vattr.va_size - fp->f_offset; 1487 return (error); 1488 case FIONBIO: 1489 case FIOASYNC: 1490 return (0); 1491 default: 1492 return (VOP_IOCTL(vp, com, data, fp->f_flag, 1493 active_cred, td)); 1494 } 1495 break; 1496 case VCHR: 1497 return (VOP_IOCTL(vp, com, data, fp->f_flag, 1498 active_cred, td)); 1499 default: 1500 return (ENOTTY); 1501 } 1502 } 1503 1504 /* 1505 * File table vnode poll routine. 1506 */ 1507 static int 1508 vn_poll(fp, events, active_cred, td) 1509 struct file *fp; 1510 int events; 1511 struct ucred *active_cred; 1512 struct thread *td; 1513 { 1514 struct vnode *vp; 1515 int error; 1516 1517 vp = fp->f_vnode; 1518 #ifdef MAC 1519 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1520 AUDIT_ARG_VNODE1(vp); 1521 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp); 1522 VOP_UNLOCK(vp, 0); 1523 if (!error) 1524 #endif 1525 1526 error = VOP_POLL(vp, events, fp->f_cred, td); 1527 return (error); 1528 } 1529 1530 /* 1531 * Acquire the requested lock and then check for validity. LK_RETRY 1532 * permits vn_lock to return doomed vnodes. 1533 */ 1534 int 1535 _vn_lock(struct vnode *vp, int flags, char *file, int line) 1536 { 1537 int error; 1538 1539 VNASSERT((flags & LK_TYPE_MASK) != 0, vp, 1540 ("vn_lock called with no locktype.")); 1541 do { 1542 #ifdef DEBUG_VFS_LOCKS 1543 KASSERT(vp->v_holdcnt != 0, 1544 ("vn_lock %p: zero hold count", vp)); 1545 #endif 1546 error = VOP_LOCK1(vp, flags, file, line); 1547 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */ 1548 KASSERT((flags & LK_RETRY) == 0 || error == 0, 1549 ("LK_RETRY set with incompatible flags (0x%x) or an error occurred (%d)", 1550 flags, error)); 1551 /* 1552 * Callers specify LK_RETRY if they wish to get dead vnodes. 1553 * If RETRY is not set, we return ENOENT instead. 1554 */ 1555 if (error == 0 && vp->v_iflag & VI_DOOMED && 1556 (flags & LK_RETRY) == 0) { 1557 VOP_UNLOCK(vp, 0); 1558 error = ENOENT; 1559 break; 1560 } 1561 } while (flags & LK_RETRY && error != 0); 1562 return (error); 1563 } 1564 1565 /* 1566 * File table vnode close routine. 1567 */ 1568 static int 1569 vn_closefile(fp, td) 1570 struct file *fp; 1571 struct thread *td; 1572 { 1573 struct vnode *vp; 1574 struct flock lf; 1575 int error; 1576 1577 vp = fp->f_vnode; 1578 fp->f_ops = &badfileops; 1579 1580 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) 1581 vref(vp); 1582 1583 error = vn_close(vp, fp->f_flag, fp->f_cred, td); 1584 1585 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) { 1586 lf.l_whence = SEEK_SET; 1587 lf.l_start = 0; 1588 lf.l_len = 0; 1589 lf.l_type = F_UNLCK; 1590 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); 1591 vrele(vp); 1592 } 1593 return (error); 1594 } 1595 1596 static bool 1597 vn_suspendable(struct mount *mp) 1598 { 1599 1600 return (mp->mnt_op->vfs_susp_clean != NULL); 1601 } 1602 1603 /* 1604 * Preparing to start a filesystem write operation. If the operation is 1605 * permitted, then we bump the count of operations in progress and 1606 * proceed. If a suspend request is in progress, we wait until the 1607 * suspension is over, and then proceed. 1608 */ 1609 static int 1610 vn_start_write_locked(struct mount *mp, int flags) 1611 { 1612 int error, mflags; 1613 1614 mtx_assert(MNT_MTX(mp), MA_OWNED); 1615 error = 0; 1616 1617 /* 1618 * Check on status of suspension. 1619 */ 1620 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 || 1621 mp->mnt_susp_owner != curthread) { 1622 mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? 1623 (flags & PCATCH) : 0) | (PUSER - 1); 1624 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 1625 if (flags & V_NOWAIT) { 1626 error = EWOULDBLOCK; 1627 goto unlock; 1628 } 1629 error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags, 1630 "suspfs", 0); 1631 if (error) 1632 goto unlock; 1633 } 1634 } 1635 if (flags & V_XSLEEP) 1636 goto unlock; 1637 mp->mnt_writeopcount++; 1638 unlock: 1639 if (error != 0 || (flags & V_XSLEEP) != 0) 1640 MNT_REL(mp); 1641 MNT_IUNLOCK(mp); 1642 return (error); 1643 } 1644 1645 int 1646 vn_start_write(struct vnode *vp, struct mount **mpp, int flags) 1647 { 1648 struct mount *mp; 1649 int error; 1650 1651 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), 1652 ("V_MNTREF requires mp")); 1653 1654 error = 0; 1655 /* 1656 * If a vnode is provided, get and return the mount point that 1657 * to which it will write. 1658 */ 1659 if (vp != NULL) { 1660 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 1661 *mpp = NULL; 1662 if (error != EOPNOTSUPP) 1663 return (error); 1664 return (0); 1665 } 1666 } 1667 if ((mp = *mpp) == NULL) 1668 return (0); 1669 1670 if (!vn_suspendable(mp)) { 1671 if (vp != NULL || (flags & V_MNTREF) != 0) 1672 vfs_rel(mp); 1673 return (0); 1674 } 1675 1676 /* 1677 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 1678 * a vfs_ref(). 1679 * As long as a vnode is not provided we need to acquire a 1680 * refcount for the provided mountpoint too, in order to 1681 * emulate a vfs_ref(). 1682 */ 1683 MNT_ILOCK(mp); 1684 if (vp == NULL && (flags & V_MNTREF) == 0) 1685 MNT_REF(mp); 1686 1687 return (vn_start_write_locked(mp, flags)); 1688 } 1689 1690 /* 1691 * Secondary suspension. Used by operations such as vop_inactive 1692 * routines that are needed by the higher level functions. These 1693 * are allowed to proceed until all the higher level functions have 1694 * completed (indicated by mnt_writeopcount dropping to zero). At that 1695 * time, these operations are halted until the suspension is over. 1696 */ 1697 int 1698 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags) 1699 { 1700 struct mount *mp; 1701 int error; 1702 1703 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), 1704 ("V_MNTREF requires mp")); 1705 1706 retry: 1707 if (vp != NULL) { 1708 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 1709 *mpp = NULL; 1710 if (error != EOPNOTSUPP) 1711 return (error); 1712 return (0); 1713 } 1714 } 1715 /* 1716 * If we are not suspended or have not yet reached suspended 1717 * mode, then let the operation proceed. 1718 */ 1719 if ((mp = *mpp) == NULL) 1720 return (0); 1721 1722 if (!vn_suspendable(mp)) { 1723 if (vp != NULL || (flags & V_MNTREF) != 0) 1724 vfs_rel(mp); 1725 return (0); 1726 } 1727 1728 /* 1729 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 1730 * a vfs_ref(). 1731 * As long as a vnode is not provided we need to acquire a 1732 * refcount for the provided mountpoint too, in order to 1733 * emulate a vfs_ref(). 1734 */ 1735 MNT_ILOCK(mp); 1736 if (vp == NULL && (flags & V_MNTREF) == 0) 1737 MNT_REF(mp); 1738 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) { 1739 mp->mnt_secondary_writes++; 1740 mp->mnt_secondary_accwrites++; 1741 MNT_IUNLOCK(mp); 1742 return (0); 1743 } 1744 if (flags & V_NOWAIT) { 1745 MNT_REL(mp); 1746 MNT_IUNLOCK(mp); 1747 return (EWOULDBLOCK); 1748 } 1749 /* 1750 * Wait for the suspension to finish. 1751 */ 1752 error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP | 1753 ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0), 1754 "suspfs", 0); 1755 vfs_rel(mp); 1756 if (error == 0) 1757 goto retry; 1758 return (error); 1759 } 1760 1761 /* 1762 * Filesystem write operation has completed. If we are suspending and this 1763 * operation is the last one, notify the suspender that the suspension is 1764 * now in effect. 1765 */ 1766 void 1767 vn_finished_write(mp) 1768 struct mount *mp; 1769 { 1770 if (mp == NULL || !vn_suspendable(mp)) 1771 return; 1772 MNT_ILOCK(mp); 1773 MNT_REL(mp); 1774 mp->mnt_writeopcount--; 1775 if (mp->mnt_writeopcount < 0) 1776 panic("vn_finished_write: neg cnt"); 1777 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1778 mp->mnt_writeopcount <= 0) 1779 wakeup(&mp->mnt_writeopcount); 1780 MNT_IUNLOCK(mp); 1781 } 1782 1783 1784 /* 1785 * Filesystem secondary write operation has completed. If we are 1786 * suspending and this operation is the last one, notify the suspender 1787 * that the suspension is now in effect. 1788 */ 1789 void 1790 vn_finished_secondary_write(mp) 1791 struct mount *mp; 1792 { 1793 if (mp == NULL || !vn_suspendable(mp)) 1794 return; 1795 MNT_ILOCK(mp); 1796 MNT_REL(mp); 1797 mp->mnt_secondary_writes--; 1798 if (mp->mnt_secondary_writes < 0) 1799 panic("vn_finished_secondary_write: neg cnt"); 1800 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1801 mp->mnt_secondary_writes <= 0) 1802 wakeup(&mp->mnt_secondary_writes); 1803 MNT_IUNLOCK(mp); 1804 } 1805 1806 1807 1808 /* 1809 * Request a filesystem to suspend write operations. 1810 */ 1811 int 1812 vfs_write_suspend(struct mount *mp, int flags) 1813 { 1814 int error; 1815 1816 MPASS(vn_suspendable(mp)); 1817 1818 MNT_ILOCK(mp); 1819 if (mp->mnt_susp_owner == curthread) { 1820 MNT_IUNLOCK(mp); 1821 return (EALREADY); 1822 } 1823 while (mp->mnt_kern_flag & MNTK_SUSPEND) 1824 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0); 1825 1826 /* 1827 * Unmount holds a write reference on the mount point. If we 1828 * own busy reference and drain for writers, we deadlock with 1829 * the reference draining in the unmount path. Callers of 1830 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if 1831 * vfs_busy() reference is owned and caller is not in the 1832 * unmount context. 1833 */ 1834 if ((flags & VS_SKIP_UNMOUNT) != 0 && 1835 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { 1836 MNT_IUNLOCK(mp); 1837 return (EBUSY); 1838 } 1839 1840 mp->mnt_kern_flag |= MNTK_SUSPEND; 1841 mp->mnt_susp_owner = curthread; 1842 if (mp->mnt_writeopcount > 0) 1843 (void) msleep(&mp->mnt_writeopcount, 1844 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0); 1845 else 1846 MNT_IUNLOCK(mp); 1847 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0) 1848 vfs_write_resume(mp, 0); 1849 return (error); 1850 } 1851 1852 /* 1853 * Request a filesystem to resume write operations. 1854 */ 1855 void 1856 vfs_write_resume(struct mount *mp, int flags) 1857 { 1858 1859 MPASS(vn_suspendable(mp)); 1860 1861 MNT_ILOCK(mp); 1862 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 1863 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner")); 1864 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 | 1865 MNTK_SUSPENDED); 1866 mp->mnt_susp_owner = NULL; 1867 wakeup(&mp->mnt_writeopcount); 1868 wakeup(&mp->mnt_flag); 1869 curthread->td_pflags &= ~TDP_IGNSUSP; 1870 if ((flags & VR_START_WRITE) != 0) { 1871 MNT_REF(mp); 1872 mp->mnt_writeopcount++; 1873 } 1874 MNT_IUNLOCK(mp); 1875 if ((flags & VR_NO_SUSPCLR) == 0) 1876 VFS_SUSP_CLEAN(mp); 1877 } else if ((flags & VR_START_WRITE) != 0) { 1878 MNT_REF(mp); 1879 vn_start_write_locked(mp, 0); 1880 } else { 1881 MNT_IUNLOCK(mp); 1882 } 1883 } 1884 1885 /* 1886 * Helper loop around vfs_write_suspend() for filesystem unmount VFS 1887 * methods. 1888 */ 1889 int 1890 vfs_write_suspend_umnt(struct mount *mp) 1891 { 1892 int error; 1893 1894 MPASS(vn_suspendable(mp)); 1895 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0, 1896 ("vfs_write_suspend_umnt: recursed")); 1897 1898 /* dounmount() already called vn_start_write(). */ 1899 for (;;) { 1900 vn_finished_write(mp); 1901 error = vfs_write_suspend(mp, 0); 1902 if (error != 0) { 1903 vn_start_write(NULL, &mp, V_WAIT); 1904 return (error); 1905 } 1906 MNT_ILOCK(mp); 1907 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0) 1908 break; 1909 MNT_IUNLOCK(mp); 1910 vn_start_write(NULL, &mp, V_WAIT); 1911 } 1912 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); 1913 wakeup(&mp->mnt_flag); 1914 MNT_IUNLOCK(mp); 1915 curthread->td_pflags |= TDP_IGNSUSP; 1916 return (0); 1917 } 1918 1919 /* 1920 * Implement kqueues for files by translating it to vnode operation. 1921 */ 1922 static int 1923 vn_kqfilter(struct file *fp, struct knote *kn) 1924 { 1925 1926 return (VOP_KQFILTER(fp->f_vnode, kn)); 1927 } 1928 1929 /* 1930 * Simplified in-kernel wrapper calls for extended attribute access. 1931 * Both calls pass in a NULL credential, authorizing as "kernel" access. 1932 * Set IO_NODELOCKED in ioflg if the vnode is already locked. 1933 */ 1934 int 1935 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, 1936 const char *attrname, int *buflen, char *buf, struct thread *td) 1937 { 1938 struct uio auio; 1939 struct iovec iov; 1940 int error; 1941 1942 iov.iov_len = *buflen; 1943 iov.iov_base = buf; 1944 1945 auio.uio_iov = &iov; 1946 auio.uio_iovcnt = 1; 1947 auio.uio_rw = UIO_READ; 1948 auio.uio_segflg = UIO_SYSSPACE; 1949 auio.uio_td = td; 1950 auio.uio_offset = 0; 1951 auio.uio_resid = *buflen; 1952 1953 if ((ioflg & IO_NODELOCKED) == 0) 1954 vn_lock(vp, LK_SHARED | LK_RETRY); 1955 1956 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1957 1958 /* authorize attribute retrieval as kernel */ 1959 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL, 1960 td); 1961 1962 if ((ioflg & IO_NODELOCKED) == 0) 1963 VOP_UNLOCK(vp, 0); 1964 1965 if (error == 0) { 1966 *buflen = *buflen - auio.uio_resid; 1967 } 1968 1969 return (error); 1970 } 1971 1972 /* 1973 * XXX failure mode if partially written? 1974 */ 1975 int 1976 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, 1977 const char *attrname, int buflen, char *buf, struct thread *td) 1978 { 1979 struct uio auio; 1980 struct iovec iov; 1981 struct mount *mp; 1982 int error; 1983 1984 iov.iov_len = buflen; 1985 iov.iov_base = buf; 1986 1987 auio.uio_iov = &iov; 1988 auio.uio_iovcnt = 1; 1989 auio.uio_rw = UIO_WRITE; 1990 auio.uio_segflg = UIO_SYSSPACE; 1991 auio.uio_td = td; 1992 auio.uio_offset = 0; 1993 auio.uio_resid = buflen; 1994 1995 if ((ioflg & IO_NODELOCKED) == 0) { 1996 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 1997 return (error); 1998 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1999 } 2000 2001 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 2002 2003 /* authorize attribute setting as kernel */ 2004 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td); 2005 2006 if ((ioflg & IO_NODELOCKED) == 0) { 2007 vn_finished_write(mp); 2008 VOP_UNLOCK(vp, 0); 2009 } 2010 2011 return (error); 2012 } 2013 2014 int 2015 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, 2016 const char *attrname, struct thread *td) 2017 { 2018 struct mount *mp; 2019 int error; 2020 2021 if ((ioflg & IO_NODELOCKED) == 0) { 2022 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 2023 return (error); 2024 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2025 } 2026 2027 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 2028 2029 /* authorize attribute removal as kernel */ 2030 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td); 2031 if (error == EOPNOTSUPP) 2032 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, 2033 NULL, td); 2034 2035 if ((ioflg & IO_NODELOCKED) == 0) { 2036 vn_finished_write(mp); 2037 VOP_UNLOCK(vp, 0); 2038 } 2039 2040 return (error); 2041 } 2042 2043 static int 2044 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags, 2045 struct vnode **rvp) 2046 { 2047 2048 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp)); 2049 } 2050 2051 int 2052 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp) 2053 { 2054 2055 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino, 2056 lkflags, rvp)); 2057 } 2058 2059 int 2060 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg, 2061 int lkflags, struct vnode **rvp) 2062 { 2063 struct mount *mp; 2064 int ltype, error; 2065 2066 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get"); 2067 mp = vp->v_mount; 2068 ltype = VOP_ISLOCKED(vp); 2069 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED, 2070 ("vn_vget_ino: vp not locked")); 2071 error = vfs_busy(mp, MBF_NOWAIT); 2072 if (error != 0) { 2073 vfs_ref(mp); 2074 VOP_UNLOCK(vp, 0); 2075 error = vfs_busy(mp, 0); 2076 vn_lock(vp, ltype | LK_RETRY); 2077 vfs_rel(mp); 2078 if (error != 0) 2079 return (ENOENT); 2080 if (vp->v_iflag & VI_DOOMED) { 2081 vfs_unbusy(mp); 2082 return (ENOENT); 2083 } 2084 } 2085 VOP_UNLOCK(vp, 0); 2086 error = alloc(mp, alloc_arg, lkflags, rvp); 2087 vfs_unbusy(mp); 2088 if (*rvp != vp) 2089 vn_lock(vp, ltype | LK_RETRY); 2090 if (vp->v_iflag & VI_DOOMED) { 2091 if (error == 0) { 2092 if (*rvp == vp) 2093 vunref(vp); 2094 else 2095 vput(*rvp); 2096 } 2097 error = ENOENT; 2098 } 2099 return (error); 2100 } 2101 2102 int 2103 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, 2104 struct thread *td) 2105 { 2106 2107 if (vp->v_type != VREG || td == NULL) 2108 return (0); 2109 if ((uoff_t)uio->uio_offset + uio->uio_resid > 2110 lim_cur(td, RLIMIT_FSIZE)) { 2111 PROC_LOCK(td->td_proc); 2112 kern_psignal(td->td_proc, SIGXFSZ); 2113 PROC_UNLOCK(td->td_proc); 2114 return (EFBIG); 2115 } 2116 return (0); 2117 } 2118 2119 int 2120 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 2121 struct thread *td) 2122 { 2123 struct vnode *vp; 2124 2125 vp = fp->f_vnode; 2126 #ifdef AUDIT 2127 vn_lock(vp, LK_SHARED | LK_RETRY); 2128 AUDIT_ARG_VNODE1(vp); 2129 VOP_UNLOCK(vp, 0); 2130 #endif 2131 return (setfmode(td, active_cred, vp, mode)); 2132 } 2133 2134 int 2135 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 2136 struct thread *td) 2137 { 2138 struct vnode *vp; 2139 2140 vp = fp->f_vnode; 2141 #ifdef AUDIT 2142 vn_lock(vp, LK_SHARED | LK_RETRY); 2143 AUDIT_ARG_VNODE1(vp); 2144 VOP_UNLOCK(vp, 0); 2145 #endif 2146 return (setfown(td, active_cred, vp, uid, gid)); 2147 } 2148 2149 void 2150 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end) 2151 { 2152 vm_object_t object; 2153 2154 if ((object = vp->v_object) == NULL) 2155 return; 2156 VM_OBJECT_WLOCK(object); 2157 vm_object_page_remove(object, start, end, 0); 2158 VM_OBJECT_WUNLOCK(object); 2159 } 2160 2161 int 2162 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred) 2163 { 2164 struct vattr va; 2165 daddr_t bn, bnp; 2166 uint64_t bsize; 2167 off_t noff; 2168 int error; 2169 2170 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA, 2171 ("Wrong command %lu", cmd)); 2172 2173 if (vn_lock(vp, LK_SHARED) != 0) 2174 return (EBADF); 2175 if (vp->v_type != VREG) { 2176 error = ENOTTY; 2177 goto unlock; 2178 } 2179 error = VOP_GETATTR(vp, &va, cred); 2180 if (error != 0) 2181 goto unlock; 2182 noff = *off; 2183 if (noff >= va.va_size) { 2184 error = ENXIO; 2185 goto unlock; 2186 } 2187 bsize = vp->v_mount->mnt_stat.f_iosize; 2188 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) { 2189 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL); 2190 if (error == EOPNOTSUPP) { 2191 error = ENOTTY; 2192 goto unlock; 2193 } 2194 if ((bnp == -1 && cmd == FIOSEEKHOLE) || 2195 (bnp != -1 && cmd == FIOSEEKDATA)) { 2196 noff = bn * bsize; 2197 if (noff < *off) 2198 noff = *off; 2199 goto unlock; 2200 } 2201 } 2202 if (noff > va.va_size) 2203 noff = va.va_size; 2204 /* noff == va.va_size. There is an implicit hole at the end of file. */ 2205 if (cmd == FIOSEEKDATA) 2206 error = ENXIO; 2207 unlock: 2208 VOP_UNLOCK(vp, 0); 2209 if (error == 0) 2210 *off = noff; 2211 return (error); 2212 } 2213 2214 int 2215 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td) 2216 { 2217 struct ucred *cred; 2218 struct vnode *vp; 2219 struct vattr vattr; 2220 off_t foffset, size; 2221 int error, noneg; 2222 2223 cred = td->td_ucred; 2224 vp = fp->f_vnode; 2225 foffset = foffset_lock(fp, 0); 2226 noneg = (vp->v_type != VCHR); 2227 error = 0; 2228 switch (whence) { 2229 case L_INCR: 2230 if (noneg && 2231 (foffset < 0 || 2232 (offset > 0 && foffset > OFF_MAX - offset))) { 2233 error = EOVERFLOW; 2234 break; 2235 } 2236 offset += foffset; 2237 break; 2238 case L_XTND: 2239 vn_lock(vp, LK_SHARED | LK_RETRY); 2240 error = VOP_GETATTR(vp, &vattr, cred); 2241 VOP_UNLOCK(vp, 0); 2242 if (error) 2243 break; 2244 2245 /* 2246 * If the file references a disk device, then fetch 2247 * the media size and use that to determine the ending 2248 * offset. 2249 */ 2250 if (vattr.va_size == 0 && vp->v_type == VCHR && 2251 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0) 2252 vattr.va_size = size; 2253 if (noneg && 2254 (vattr.va_size > OFF_MAX || 2255 (offset > 0 && vattr.va_size > OFF_MAX - offset))) { 2256 error = EOVERFLOW; 2257 break; 2258 } 2259 offset += vattr.va_size; 2260 break; 2261 case L_SET: 2262 break; 2263 case SEEK_DATA: 2264 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td); 2265 break; 2266 case SEEK_HOLE: 2267 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td); 2268 break; 2269 default: 2270 error = EINVAL; 2271 } 2272 if (error == 0 && noneg && offset < 0) 2273 error = EINVAL; 2274 if (error != 0) 2275 goto drop; 2276 VFS_KNOTE_UNLOCKED(vp, 0); 2277 td->td_uretoff.tdu_off = offset; 2278 drop: 2279 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); 2280 return (error); 2281 } 2282 2283 int 2284 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred, 2285 struct thread *td) 2286 { 2287 int error; 2288 2289 /* 2290 * Grant permission if the caller is the owner of the file, or 2291 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on 2292 * on the file. If the time pointer is null, then write 2293 * permission on the file is also sufficient. 2294 * 2295 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes: 2296 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES 2297 * will be allowed to set the times [..] to the current 2298 * server time. 2299 */ 2300 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td); 2301 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0) 2302 error = VOP_ACCESS(vp, VWRITE, cred, td); 2303 return (error); 2304 } 2305 2306 int 2307 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 2308 { 2309 struct vnode *vp; 2310 int error; 2311 2312 if (fp->f_type == DTYPE_FIFO) 2313 kif->kf_type = KF_TYPE_FIFO; 2314 else 2315 kif->kf_type = KF_TYPE_VNODE; 2316 vp = fp->f_vnode; 2317 vref(vp); 2318 FILEDESC_SUNLOCK(fdp); 2319 error = vn_fill_kinfo_vnode(vp, kif); 2320 vrele(vp); 2321 FILEDESC_SLOCK(fdp); 2322 return (error); 2323 } 2324 2325 static inline void 2326 vn_fill_junk(struct kinfo_file *kif) 2327 { 2328 size_t len, olen; 2329 2330 /* 2331 * Simulate vn_fullpath returning changing values for a given 2332 * vp during e.g. coredump. 2333 */ 2334 len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1; 2335 olen = strlen(kif->kf_path); 2336 if (len < olen) 2337 strcpy(&kif->kf_path[len - 1], "$"); 2338 else 2339 for (; olen < len; olen++) 2340 strcpy(&kif->kf_path[olen], "A"); 2341 } 2342 2343 int 2344 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif) 2345 { 2346 struct vattr va; 2347 char *fullpath, *freepath; 2348 int error; 2349 2350 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type); 2351 freepath = NULL; 2352 fullpath = "-"; 2353 error = vn_fullpath(curthread, vp, &fullpath, &freepath); 2354 if (error == 0) { 2355 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 2356 } 2357 if (freepath != NULL) 2358 free(freepath, M_TEMP); 2359 2360 KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path, 2361 vn_fill_junk(kif); 2362 ); 2363 2364 /* 2365 * Retrieve vnode attributes. 2366 */ 2367 va.va_fsid = VNOVAL; 2368 va.va_rdev = NODEV; 2369 vn_lock(vp, LK_SHARED | LK_RETRY); 2370 error = VOP_GETATTR(vp, &va, curthread->td_ucred); 2371 VOP_UNLOCK(vp, 0); 2372 if (error != 0) 2373 return (error); 2374 if (va.va_fsid != VNOVAL) 2375 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; 2376 else 2377 kif->kf_un.kf_file.kf_file_fsid = 2378 vp->v_mount->mnt_stat.f_fsid.val[0]; 2379 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; 2380 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); 2381 kif->kf_un.kf_file.kf_file_size = va.va_size; 2382 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; 2383 return (0); 2384 } 2385 2386 int 2387 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, 2388 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, 2389 struct thread *td) 2390 { 2391 #ifdef HWPMC_HOOKS 2392 struct pmckern_map_in pkm; 2393 #endif 2394 struct mount *mp; 2395 struct vnode *vp; 2396 vm_object_t object; 2397 vm_prot_t maxprot; 2398 boolean_t writecounted; 2399 int error; 2400 2401 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \ 2402 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) 2403 /* 2404 * POSIX shared-memory objects are defined to have 2405 * kernel persistence, and are not defined to support 2406 * read(2)/write(2) -- or even open(2). Thus, we can 2407 * use MAP_ASYNC to trade on-disk coherence for speed. 2408 * The shm_open(3) library routine turns on the FPOSIXSHM 2409 * flag to request this behavior. 2410 */ 2411 if ((fp->f_flag & FPOSIXSHM) != 0) 2412 flags |= MAP_NOSYNC; 2413 #endif 2414 vp = fp->f_vnode; 2415 2416 /* 2417 * Ensure that file and memory protections are 2418 * compatible. Note that we only worry about 2419 * writability if mapping is shared; in this case, 2420 * current and max prot are dictated by the open file. 2421 * XXX use the vnode instead? Problem is: what 2422 * credentials do we use for determination? What if 2423 * proc does a setuid? 2424 */ 2425 mp = vp->v_mount; 2426 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) 2427 maxprot = VM_PROT_NONE; 2428 else 2429 maxprot = VM_PROT_EXECUTE; 2430 if ((fp->f_flag & FREAD) != 0) 2431 maxprot |= VM_PROT_READ; 2432 else if ((prot & VM_PROT_READ) != 0) 2433 return (EACCES); 2434 2435 /* 2436 * If we are sharing potential changes via MAP_SHARED and we 2437 * are trying to get write permission although we opened it 2438 * without asking for it, bail out. 2439 */ 2440 if ((flags & MAP_SHARED) != 0) { 2441 if ((fp->f_flag & FWRITE) != 0) 2442 maxprot |= VM_PROT_WRITE; 2443 else if ((prot & VM_PROT_WRITE) != 0) 2444 return (EACCES); 2445 } else { 2446 maxprot |= VM_PROT_WRITE; 2447 cap_maxprot |= VM_PROT_WRITE; 2448 } 2449 maxprot &= cap_maxprot; 2450 2451 writecounted = FALSE; 2452 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp, 2453 &foff, &object, &writecounted); 2454 if (error != 0) 2455 return (error); 2456 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 2457 foff, writecounted, td); 2458 if (error != 0) { 2459 /* 2460 * If this mapping was accounted for in the vnode's 2461 * writecount, then undo that now. 2462 */ 2463 if (writecounted) 2464 vnode_pager_release_writecount(object, 0, size); 2465 vm_object_deallocate(object); 2466 } 2467 #ifdef HWPMC_HOOKS 2468 /* Inform hwpmc(4) if an executable is being mapped. */ 2469 if (error == 0 && (prot & VM_PROT_EXECUTE) != 0) { 2470 pkm.pm_file = vp; 2471 pkm.pm_address = (uintptr_t) *addr; 2472 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm); 2473 } 2474 #endif 2475 return (error); 2476 } 2477