1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Copyright (c) 2012 Konstantin Belousov <kib@FreeBSD.org> 13 * Copyright (c) 2013, 2014 The FreeBSD Foundation 14 * 15 * Portions of this software were developed by Konstantin Belousov 16 * under sponsorship from the FreeBSD Foundation. 17 * 18 * Redistribution and use in source and binary forms, with or without 19 * modification, are permitted provided that the following conditions 20 * are met: 21 * 1. Redistributions of source code must retain the above copyright 22 * notice, this list of conditions and the following disclaimer. 23 * 2. Redistributions in binary form must reproduce the above copyright 24 * notice, this list of conditions and the following disclaimer in the 25 * documentation and/or other materials provided with the distribution. 26 * 3. Neither the name of the University nor the names of its contributors 27 * may be used to endorse or promote products derived from this software 28 * without specific prior written permission. 29 * 30 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 31 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 32 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 33 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 34 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 35 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 36 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 37 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 38 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 39 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 40 * SUCH DAMAGE. 41 * 42 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94 43 */ 44 45 #include <sys/cdefs.h> 46 __FBSDID("$FreeBSD$"); 47 48 #include "opt_hwpmc_hooks.h" 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/disk.h> 53 #include <sys/fail.h> 54 #include <sys/fcntl.h> 55 #include <sys/file.h> 56 #include <sys/kdb.h> 57 #include <sys/stat.h> 58 #include <sys/priv.h> 59 #include <sys/proc.h> 60 #include <sys/limits.h> 61 #include <sys/lock.h> 62 #include <sys/mman.h> 63 #include <sys/mount.h> 64 #include <sys/mutex.h> 65 #include <sys/namei.h> 66 #include <sys/vnode.h> 67 #include <sys/bio.h> 68 #include <sys/buf.h> 69 #include <sys/filio.h> 70 #include <sys/resourcevar.h> 71 #include <sys/rwlock.h> 72 #include <sys/sx.h> 73 #include <sys/sysctl.h> 74 #include <sys/ttycom.h> 75 #include <sys/conf.h> 76 #include <sys/syslog.h> 77 #include <sys/unistd.h> 78 #include <sys/user.h> 79 80 #include <security/audit/audit.h> 81 #include <security/mac/mac_framework.h> 82 83 #include <vm/vm.h> 84 #include <vm/vm_extern.h> 85 #include <vm/pmap.h> 86 #include <vm/vm_map.h> 87 #include <vm/vm_object.h> 88 #include <vm/vm_page.h> 89 #include <vm/vnode_pager.h> 90 91 #ifdef HWPMC_HOOKS 92 #include <sys/pmckern.h> 93 #endif 94 95 static fo_rdwr_t vn_read; 96 static fo_rdwr_t vn_write; 97 static fo_rdwr_t vn_io_fault; 98 static fo_truncate_t vn_truncate; 99 static fo_ioctl_t vn_ioctl; 100 static fo_poll_t vn_poll; 101 static fo_kqfilter_t vn_kqfilter; 102 static fo_stat_t vn_statfile; 103 static fo_close_t vn_closefile; 104 static fo_mmap_t vn_mmap; 105 106 struct fileops vnops = { 107 .fo_read = vn_io_fault, 108 .fo_write = vn_io_fault, 109 .fo_truncate = vn_truncate, 110 .fo_ioctl = vn_ioctl, 111 .fo_poll = vn_poll, 112 .fo_kqfilter = vn_kqfilter, 113 .fo_stat = vn_statfile, 114 .fo_close = vn_closefile, 115 .fo_chmod = vn_chmod, 116 .fo_chown = vn_chown, 117 .fo_sendfile = vn_sendfile, 118 .fo_seek = vn_seek, 119 .fo_fill_kinfo = vn_fill_kinfo, 120 .fo_mmap = vn_mmap, 121 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 122 }; 123 124 static const int io_hold_cnt = 16; 125 static int vn_io_fault_enable = 1; 126 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW, 127 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance"); 128 static int vn_io_fault_prefault = 0; 129 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW, 130 &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting"); 131 static u_long vn_io_faults_cnt; 132 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD, 133 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers"); 134 135 /* 136 * Returns true if vn_io_fault mode of handling the i/o request should 137 * be used. 138 */ 139 static bool 140 do_vn_io_fault(struct vnode *vp, struct uio *uio) 141 { 142 struct mount *mp; 143 144 return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG && 145 (mp = vp->v_mount) != NULL && 146 (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable); 147 } 148 149 /* 150 * Structure used to pass arguments to vn_io_fault1(), to do either 151 * file- or vnode-based I/O calls. 152 */ 153 struct vn_io_fault_args { 154 enum { 155 VN_IO_FAULT_FOP, 156 VN_IO_FAULT_VOP 157 } kind; 158 struct ucred *cred; 159 int flags; 160 union { 161 struct fop_args_tag { 162 struct file *fp; 163 fo_rdwr_t *doio; 164 } fop_args; 165 struct vop_args_tag { 166 struct vnode *vp; 167 } vop_args; 168 } args; 169 }; 170 171 static int vn_io_fault1(struct vnode *vp, struct uio *uio, 172 struct vn_io_fault_args *args, struct thread *td); 173 174 int 175 vn_open(struct nameidata *ndp, int *flagp, int cmode, struct file *fp) 176 { 177 struct thread *td = ndp->ni_cnd.cn_thread; 178 179 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp)); 180 } 181 182 /* 183 * Common code for vnode open operations via a name lookup. 184 * Lookup the vnode and invoke VOP_CREATE if needed. 185 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine. 186 * 187 * Note that this does NOT free nameidata for the successful case, 188 * due to the NDINIT being done elsewhere. 189 */ 190 int 191 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, 192 struct ucred *cred, struct file *fp) 193 { 194 struct vnode *vp; 195 struct mount *mp; 196 struct thread *td = ndp->ni_cnd.cn_thread; 197 struct vattr vat; 198 struct vattr *vap = &vat; 199 int fmode, error; 200 201 restart: 202 fmode = *flagp; 203 if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT | 204 O_EXCL | O_DIRECTORY)) 205 return (EINVAL); 206 else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) { 207 ndp->ni_cnd.cn_nameiop = CREATE; 208 /* 209 * Set NOCACHE to avoid flushing the cache when 210 * rolling in many files at once. 211 */ 212 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE; 213 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 214 ndp->ni_cnd.cn_flags |= FOLLOW; 215 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 216 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 217 if (vn_open_flags & VN_OPEN_NOCAPCHECK) 218 ndp->ni_cnd.cn_flags |= NOCAPCHECK; 219 bwillwrite(); 220 if ((error = namei(ndp)) != 0) 221 return (error); 222 if (ndp->ni_vp == NULL) { 223 VATTR_NULL(vap); 224 vap->va_type = VREG; 225 vap->va_mode = cmode; 226 if (fmode & O_EXCL) 227 vap->va_vaflags |= VA_EXCLUSIVE; 228 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) { 229 NDFREE(ndp, NDF_ONLY_PNBUF); 230 vput(ndp->ni_dvp); 231 if ((error = vn_start_write(NULL, &mp, 232 V_XSLEEP | PCATCH)) != 0) 233 return (error); 234 goto restart; 235 } 236 if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0) 237 ndp->ni_cnd.cn_flags |= MAKEENTRY; 238 #ifdef MAC 239 error = mac_vnode_check_create(cred, ndp->ni_dvp, 240 &ndp->ni_cnd, vap); 241 if (error == 0) 242 #endif 243 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 244 &ndp->ni_cnd, vap); 245 vput(ndp->ni_dvp); 246 vn_finished_write(mp); 247 if (error) { 248 NDFREE(ndp, NDF_ONLY_PNBUF); 249 return (error); 250 } 251 fmode &= ~O_TRUNC; 252 vp = ndp->ni_vp; 253 } else { 254 if (ndp->ni_dvp == ndp->ni_vp) 255 vrele(ndp->ni_dvp); 256 else 257 vput(ndp->ni_dvp); 258 ndp->ni_dvp = NULL; 259 vp = ndp->ni_vp; 260 if (fmode & O_EXCL) { 261 error = EEXIST; 262 goto bad; 263 } 264 fmode &= ~O_CREAT; 265 } 266 } else { 267 ndp->ni_cnd.cn_nameiop = LOOKUP; 268 ndp->ni_cnd.cn_flags = ISOPEN | 269 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF; 270 if (!(fmode & FWRITE)) 271 ndp->ni_cnd.cn_flags |= LOCKSHARED; 272 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 273 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 274 if (vn_open_flags & VN_OPEN_NOCAPCHECK) 275 ndp->ni_cnd.cn_flags |= NOCAPCHECK; 276 if ((error = namei(ndp)) != 0) 277 return (error); 278 vp = ndp->ni_vp; 279 } 280 error = vn_open_vnode(vp, fmode, cred, td, fp); 281 if (error) 282 goto bad; 283 *flagp = fmode; 284 return (0); 285 bad: 286 NDFREE(ndp, NDF_ONLY_PNBUF); 287 vput(vp); 288 *flagp = fmode; 289 ndp->ni_vp = NULL; 290 return (error); 291 } 292 293 /* 294 * Common code for vnode open operations once a vnode is located. 295 * Check permissions, and call the VOP_OPEN routine. 296 */ 297 int 298 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, 299 struct thread *td, struct file *fp) 300 { 301 accmode_t accmode; 302 struct flock lf; 303 int error, lock_flags, type; 304 305 if (vp->v_type == VLNK) 306 return (EMLINK); 307 if (vp->v_type == VSOCK) 308 return (EOPNOTSUPP); 309 if (vp->v_type != VDIR && fmode & O_DIRECTORY) 310 return (ENOTDIR); 311 accmode = 0; 312 if (fmode & (FWRITE | O_TRUNC)) { 313 if (vp->v_type == VDIR) 314 return (EISDIR); 315 accmode |= VWRITE; 316 } 317 if (fmode & FREAD) 318 accmode |= VREAD; 319 if (fmode & FEXEC) 320 accmode |= VEXEC; 321 if ((fmode & O_APPEND) && (fmode & FWRITE)) 322 accmode |= VAPPEND; 323 #ifdef MAC 324 if (fmode & O_CREAT) 325 accmode |= VCREAT; 326 if (fmode & O_VERIFY) 327 accmode |= VVERIFY; 328 error = mac_vnode_check_open(cred, vp, accmode); 329 if (error) 330 return (error); 331 332 accmode &= ~(VCREAT | VVERIFY); 333 #endif 334 if ((fmode & O_CREAT) == 0) { 335 if (accmode & VWRITE) { 336 error = vn_writechk(vp); 337 if (error) 338 return (error); 339 } 340 if (accmode) { 341 error = VOP_ACCESS(vp, accmode, cred, td); 342 if (error) 343 return (error); 344 } 345 } 346 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) 347 vn_lock(vp, LK_UPGRADE | LK_RETRY); 348 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0) 349 return (error); 350 351 while ((fmode & (O_EXLOCK | O_SHLOCK)) != 0) { 352 KASSERT(fp != NULL, ("open with flock requires fp")); 353 if (fp->f_type != DTYPE_NONE && fp->f_type != DTYPE_VNODE) { 354 error = EOPNOTSUPP; 355 break; 356 } 357 lock_flags = VOP_ISLOCKED(vp); 358 VOP_UNLOCK(vp, 0); 359 lf.l_whence = SEEK_SET; 360 lf.l_start = 0; 361 lf.l_len = 0; 362 if (fmode & O_EXLOCK) 363 lf.l_type = F_WRLCK; 364 else 365 lf.l_type = F_RDLCK; 366 type = F_FLOCK; 367 if ((fmode & FNONBLOCK) == 0) 368 type |= F_WAIT; 369 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type); 370 if (error == 0) 371 fp->f_flag |= FHASLOCK; 372 vn_lock(vp, lock_flags | LK_RETRY); 373 if (error != 0) 374 break; 375 if ((vp->v_iflag & VI_DOOMED) != 0) { 376 error = ENOENT; 377 break; 378 } 379 380 /* 381 * Another thread might have used this vnode as an 382 * executable while the vnode lock was dropped. 383 * Ensure the vnode is still able to be opened for 384 * writing after the lock has been obtained. 385 */ 386 if ((accmode & VWRITE) != 0) 387 error = vn_writechk(vp); 388 break; 389 } 390 391 if (error != 0) { 392 fp->f_flag |= FOPENFAILED; 393 fp->f_vnode = vp; 394 if (fp->f_ops == &badfileops) { 395 fp->f_type = DTYPE_VNODE; 396 fp->f_ops = &vnops; 397 } 398 vref(vp); 399 } else if ((fmode & FWRITE) != 0) { 400 VOP_ADD_WRITECOUNT(vp, 1); 401 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", 402 __func__, vp, vp->v_writecount); 403 } 404 ASSERT_VOP_LOCKED(vp, "vn_open_vnode"); 405 return (error); 406 } 407 408 /* 409 * Check for write permissions on the specified vnode. 410 * Prototype text segments cannot be written. 411 */ 412 int 413 vn_writechk(struct vnode *vp) 414 { 415 416 ASSERT_VOP_LOCKED(vp, "vn_writechk"); 417 /* 418 * If there's shared text associated with 419 * the vnode, try to free it up once. If 420 * we fail, we can't allow writing. 421 */ 422 if (VOP_IS_TEXT(vp)) 423 return (ETXTBSY); 424 425 return (0); 426 } 427 428 /* 429 * Vnode close call 430 */ 431 static int 432 vn_close1(struct vnode *vp, int flags, struct ucred *file_cred, 433 struct thread *td, bool keep_ref) 434 { 435 struct mount *mp; 436 int error, lock_flags; 437 438 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 && 439 MNT_EXTENDED_SHARED(vp->v_mount)) 440 lock_flags = LK_SHARED; 441 else 442 lock_flags = LK_EXCLUSIVE; 443 444 vn_start_write(vp, &mp, V_WAIT); 445 vn_lock(vp, lock_flags | LK_RETRY); 446 AUDIT_ARG_VNODE1(vp); 447 if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) { 448 VNASSERT(vp->v_writecount > 0, vp, 449 ("vn_close: negative writecount")); 450 VOP_ADD_WRITECOUNT(vp, -1); 451 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", 452 __func__, vp, vp->v_writecount); 453 } 454 error = VOP_CLOSE(vp, flags, file_cred, td); 455 if (keep_ref) 456 VOP_UNLOCK(vp, 0); 457 else 458 vput(vp); 459 vn_finished_write(mp); 460 return (error); 461 } 462 463 int 464 vn_close(struct vnode *vp, int flags, struct ucred *file_cred, 465 struct thread *td) 466 { 467 468 return (vn_close1(vp, flags, file_cred, td, false)); 469 } 470 471 /* 472 * Heuristic to detect sequential operation. 473 */ 474 static int 475 sequential_heuristic(struct uio *uio, struct file *fp) 476 { 477 478 ASSERT_VOP_LOCKED(fp->f_vnode, __func__); 479 if (fp->f_flag & FRDAHEAD) 480 return (fp->f_seqcount << IO_SEQSHIFT); 481 482 /* 483 * Offset 0 is handled specially. open() sets f_seqcount to 1 so 484 * that the first I/O is normally considered to be slightly 485 * sequential. Seeking to offset 0 doesn't change sequentiality 486 * unless previous seeks have reduced f_seqcount to 0, in which 487 * case offset 0 is not special. 488 */ 489 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 490 uio->uio_offset == fp->f_nextoff) { 491 /* 492 * f_seqcount is in units of fixed-size blocks so that it 493 * depends mainly on the amount of sequential I/O and not 494 * much on the number of sequential I/O's. The fixed size 495 * of 16384 is hard-coded here since it is (not quite) just 496 * a magic size that works well here. This size is more 497 * closely related to the best I/O size for real disks than 498 * to any block size used by software. 499 */ 500 fp->f_seqcount += howmany(uio->uio_resid, 16384); 501 if (fp->f_seqcount > IO_SEQMAX) 502 fp->f_seqcount = IO_SEQMAX; 503 return (fp->f_seqcount << IO_SEQSHIFT); 504 } 505 506 /* Not sequential. Quickly draw-down sequentiality. */ 507 if (fp->f_seqcount > 1) 508 fp->f_seqcount = 1; 509 else 510 fp->f_seqcount = 0; 511 return (0); 512 } 513 514 /* 515 * Package up an I/O request on a vnode into a uio and do it. 516 */ 517 int 518 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, 519 enum uio_seg segflg, int ioflg, struct ucred *active_cred, 520 struct ucred *file_cred, ssize_t *aresid, struct thread *td) 521 { 522 struct uio auio; 523 struct iovec aiov; 524 struct mount *mp; 525 struct ucred *cred; 526 void *rl_cookie; 527 struct vn_io_fault_args args; 528 int error, lock_flags; 529 530 auio.uio_iov = &aiov; 531 auio.uio_iovcnt = 1; 532 aiov.iov_base = base; 533 aiov.iov_len = len; 534 auio.uio_resid = len; 535 auio.uio_offset = offset; 536 auio.uio_segflg = segflg; 537 auio.uio_rw = rw; 538 auio.uio_td = td; 539 error = 0; 540 541 if ((ioflg & IO_NODELOCKED) == 0) { 542 if ((ioflg & IO_RANGELOCKED) == 0) { 543 if (rw == UIO_READ) { 544 rl_cookie = vn_rangelock_rlock(vp, offset, 545 offset + len); 546 } else { 547 rl_cookie = vn_rangelock_wlock(vp, offset, 548 offset + len); 549 } 550 } else 551 rl_cookie = NULL; 552 mp = NULL; 553 if (rw == UIO_WRITE) { 554 if (vp->v_type != VCHR && 555 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) 556 != 0) 557 goto out; 558 if (MNT_SHARED_WRITES(mp) || 559 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) 560 lock_flags = LK_SHARED; 561 else 562 lock_flags = LK_EXCLUSIVE; 563 } else 564 lock_flags = LK_SHARED; 565 vn_lock(vp, lock_flags | LK_RETRY); 566 } else 567 rl_cookie = NULL; 568 569 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 570 #ifdef MAC 571 if ((ioflg & IO_NOMACCHECK) == 0) { 572 if (rw == UIO_READ) 573 error = mac_vnode_check_read(active_cred, file_cred, 574 vp); 575 else 576 error = mac_vnode_check_write(active_cred, file_cred, 577 vp); 578 } 579 #endif 580 if (error == 0) { 581 if (file_cred != NULL) 582 cred = file_cred; 583 else 584 cred = active_cred; 585 if (do_vn_io_fault(vp, &auio)) { 586 args.kind = VN_IO_FAULT_VOP; 587 args.cred = cred; 588 args.flags = ioflg; 589 args.args.vop_args.vp = vp; 590 error = vn_io_fault1(vp, &auio, &args, td); 591 } else if (rw == UIO_READ) { 592 error = VOP_READ(vp, &auio, ioflg, cred); 593 } else /* if (rw == UIO_WRITE) */ { 594 error = VOP_WRITE(vp, &auio, ioflg, cred); 595 } 596 } 597 if (aresid) 598 *aresid = auio.uio_resid; 599 else 600 if (auio.uio_resid && error == 0) 601 error = EIO; 602 if ((ioflg & IO_NODELOCKED) == 0) { 603 VOP_UNLOCK(vp, 0); 604 if (mp != NULL) 605 vn_finished_write(mp); 606 } 607 out: 608 if (rl_cookie != NULL) 609 vn_rangelock_unlock(vp, rl_cookie); 610 return (error); 611 } 612 613 /* 614 * Package up an I/O request on a vnode into a uio and do it. The I/O 615 * request is split up into smaller chunks and we try to avoid saturating 616 * the buffer cache while potentially holding a vnode locked, so we 617 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield() 618 * to give other processes a chance to lock the vnode (either other processes 619 * core'ing the same binary, or unrelated processes scanning the directory). 620 */ 621 int 622 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, void *base, size_t len, 623 off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, 624 struct ucred *file_cred, size_t *aresid, struct thread *td) 625 { 626 int error = 0; 627 ssize_t iaresid; 628 629 do { 630 int chunk; 631 632 /* 633 * Force `offset' to a multiple of MAXBSIZE except possibly 634 * for the first chunk, so that filesystems only need to 635 * write full blocks except possibly for the first and last 636 * chunks. 637 */ 638 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 639 640 if (chunk > len) 641 chunk = len; 642 if (rw != UIO_READ && vp->v_type == VREG) 643 bwillwrite(); 644 iaresid = 0; 645 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 646 ioflg, active_cred, file_cred, &iaresid, td); 647 len -= chunk; /* aresid calc already includes length */ 648 if (error) 649 break; 650 offset += chunk; 651 base = (char *)base + chunk; 652 kern_yield(PRI_USER); 653 } while (len); 654 if (aresid) 655 *aresid = len + iaresid; 656 return (error); 657 } 658 659 off_t 660 foffset_lock(struct file *fp, int flags) 661 { 662 struct mtx *mtxp; 663 off_t res; 664 665 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); 666 667 #if OFF_MAX <= LONG_MAX 668 /* 669 * Caller only wants the current f_offset value. Assume that 670 * the long and shorter integer types reads are atomic. 671 */ 672 if ((flags & FOF_NOLOCK) != 0) 673 return (fp->f_offset); 674 #endif 675 676 /* 677 * According to McKusick the vn lock was protecting f_offset here. 678 * It is now protected by the FOFFSET_LOCKED flag. 679 */ 680 mtxp = mtx_pool_find(mtxpool_sleep, fp); 681 mtx_lock(mtxp); 682 if ((flags & FOF_NOLOCK) == 0) { 683 while (fp->f_vnread_flags & FOFFSET_LOCKED) { 684 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING; 685 msleep(&fp->f_vnread_flags, mtxp, PUSER -1, 686 "vofflock", 0); 687 } 688 fp->f_vnread_flags |= FOFFSET_LOCKED; 689 } 690 res = fp->f_offset; 691 mtx_unlock(mtxp); 692 return (res); 693 } 694 695 void 696 foffset_unlock(struct file *fp, off_t val, int flags) 697 { 698 struct mtx *mtxp; 699 700 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); 701 702 #if OFF_MAX <= LONG_MAX 703 if ((flags & FOF_NOLOCK) != 0) { 704 if ((flags & FOF_NOUPDATE) == 0) 705 fp->f_offset = val; 706 if ((flags & FOF_NEXTOFF) != 0) 707 fp->f_nextoff = val; 708 return; 709 } 710 #endif 711 712 mtxp = mtx_pool_find(mtxpool_sleep, fp); 713 mtx_lock(mtxp); 714 if ((flags & FOF_NOUPDATE) == 0) 715 fp->f_offset = val; 716 if ((flags & FOF_NEXTOFF) != 0) 717 fp->f_nextoff = val; 718 if ((flags & FOF_NOLOCK) == 0) { 719 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0, 720 ("Lost FOFFSET_LOCKED")); 721 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING) 722 wakeup(&fp->f_vnread_flags); 723 fp->f_vnread_flags = 0; 724 } 725 mtx_unlock(mtxp); 726 } 727 728 void 729 foffset_lock_uio(struct file *fp, struct uio *uio, int flags) 730 { 731 732 if ((flags & FOF_OFFSET) == 0) 733 uio->uio_offset = foffset_lock(fp, flags); 734 } 735 736 void 737 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags) 738 { 739 740 if ((flags & FOF_OFFSET) == 0) 741 foffset_unlock(fp, uio->uio_offset, flags); 742 } 743 744 static int 745 get_advice(struct file *fp, struct uio *uio) 746 { 747 struct mtx *mtxp; 748 int ret; 749 750 ret = POSIX_FADV_NORMAL; 751 if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG) 752 return (ret); 753 754 mtxp = mtx_pool_find(mtxpool_sleep, fp); 755 mtx_lock(mtxp); 756 if (fp->f_advice != NULL && 757 uio->uio_offset >= fp->f_advice->fa_start && 758 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end) 759 ret = fp->f_advice->fa_advice; 760 mtx_unlock(mtxp); 761 return (ret); 762 } 763 764 /* 765 * File table vnode read routine. 766 */ 767 static int 768 vn_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 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(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, 824 struct thread *td) 825 { 826 struct vnode *vp; 827 struct mount *mp; 828 off_t orig_offset; 829 int error, ioflag, lock_flags; 830 int advice; 831 832 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", 833 uio->uio_td, td)); 834 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); 835 vp = fp->f_vnode; 836 if (vp->v_type == VREG) 837 bwillwrite(); 838 ioflag = IO_UNIT; 839 if (vp->v_type == VREG && (fp->f_flag & O_APPEND)) 840 ioflag |= IO_APPEND; 841 if (fp->f_flag & FNONBLOCK) 842 ioflag |= IO_NDELAY; 843 if (fp->f_flag & O_DIRECT) 844 ioflag |= IO_DIRECT; 845 if ((fp->f_flag & O_FSYNC) || 846 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) 847 ioflag |= IO_SYNC; 848 mp = NULL; 849 if (vp->v_type != VCHR && 850 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) 851 goto unlock; 852 853 advice = get_advice(fp, uio); 854 855 if (MNT_SHARED_WRITES(mp) || 856 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) { 857 lock_flags = LK_SHARED; 858 } else { 859 lock_flags = LK_EXCLUSIVE; 860 } 861 862 vn_lock(vp, lock_flags | LK_RETRY); 863 switch (advice) { 864 case POSIX_FADV_NORMAL: 865 case POSIX_FADV_SEQUENTIAL: 866 case POSIX_FADV_NOREUSE: 867 ioflag |= sequential_heuristic(uio, fp); 868 break; 869 case POSIX_FADV_RANDOM: 870 /* XXX: Is this correct? */ 871 break; 872 } 873 orig_offset = uio->uio_offset; 874 875 #ifdef MAC 876 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 877 if (error == 0) 878 #endif 879 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred); 880 fp->f_nextoff = uio->uio_offset; 881 VOP_UNLOCK(vp, 0); 882 if (vp->v_type != VCHR) 883 vn_finished_write(mp); 884 if (error == 0 && advice == POSIX_FADV_NOREUSE && 885 orig_offset != uio->uio_offset) 886 /* 887 * Use POSIX_FADV_DONTNEED to flush pages and buffers 888 * for the backing file after a POSIX_FADV_NOREUSE 889 * write(2). 890 */ 891 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1, 892 POSIX_FADV_DONTNEED); 893 unlock: 894 return (error); 895 } 896 897 /* 898 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to 899 * prevent the following deadlock: 900 * 901 * Assume that the thread A reads from the vnode vp1 into userspace 902 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is 903 * currently not resident, then system ends up with the call chain 904 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] -> 905 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2) 906 * which establishes lock order vp1->vn_lock, then vp2->vn_lock. 907 * If, at the same time, thread B reads from vnode vp2 into buffer buf2 908 * backed by the pages of vnode vp1, and some page in buf2 is not 909 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock. 910 * 911 * To prevent the lock order reversal and deadlock, vn_io_fault() does 912 * not allow page faults to happen during VOP_READ() or VOP_WRITE(). 913 * Instead, it first tries to do the whole range i/o with pagefaults 914 * disabled. If all pages in the i/o buffer are resident and mapped, 915 * VOP will succeed (ignoring the genuine filesystem errors). 916 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do 917 * i/o in chunks, with all pages in the chunk prefaulted and held 918 * using vm_fault_quick_hold_pages(). 919 * 920 * Filesystems using this deadlock avoidance scheme should use the 921 * array of the held pages from uio, saved in the curthread->td_ma, 922 * instead of doing uiomove(). A helper function 923 * vn_io_fault_uiomove() converts uiomove request into 924 * uiomove_fromphys() over td_ma array. 925 * 926 * Since vnode locks do not cover the whole i/o anymore, rangelocks 927 * make the current i/o request atomic with respect to other i/os and 928 * truncations. 929 */ 930 931 /* 932 * Decode vn_io_fault_args and perform the corresponding i/o. 933 */ 934 static int 935 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio, 936 struct thread *td) 937 { 938 int error, save; 939 940 error = 0; 941 save = vm_fault_disable_pagefaults(); 942 switch (args->kind) { 943 case VN_IO_FAULT_FOP: 944 error = (args->args.fop_args.doio)(args->args.fop_args.fp, 945 uio, args->cred, args->flags, td); 946 break; 947 case VN_IO_FAULT_VOP: 948 if (uio->uio_rw == UIO_READ) { 949 error = VOP_READ(args->args.vop_args.vp, uio, 950 args->flags, args->cred); 951 } else if (uio->uio_rw == UIO_WRITE) { 952 error = VOP_WRITE(args->args.vop_args.vp, uio, 953 args->flags, args->cred); 954 } 955 break; 956 default: 957 panic("vn_io_fault_doio: unknown kind of io %d %d", 958 args->kind, uio->uio_rw); 959 } 960 vm_fault_enable_pagefaults(save); 961 return (error); 962 } 963 964 static int 965 vn_io_fault_touch(char *base, const struct uio *uio) 966 { 967 int r; 968 969 r = fubyte(base); 970 if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1)) 971 return (EFAULT); 972 return (0); 973 } 974 975 static int 976 vn_io_fault_prefault_user(const struct uio *uio) 977 { 978 char *base; 979 const struct iovec *iov; 980 size_t len; 981 ssize_t resid; 982 int error, i; 983 984 KASSERT(uio->uio_segflg == UIO_USERSPACE, 985 ("vn_io_fault_prefault userspace")); 986 987 error = i = 0; 988 iov = uio->uio_iov; 989 resid = uio->uio_resid; 990 base = iov->iov_base; 991 len = iov->iov_len; 992 while (resid > 0) { 993 error = vn_io_fault_touch(base, uio); 994 if (error != 0) 995 break; 996 if (len < PAGE_SIZE) { 997 if (len != 0) { 998 error = vn_io_fault_touch(base + len - 1, uio); 999 if (error != 0) 1000 break; 1001 resid -= len; 1002 } 1003 if (++i >= uio->uio_iovcnt) 1004 break; 1005 iov = uio->uio_iov + i; 1006 base = iov->iov_base; 1007 len = iov->iov_len; 1008 } else { 1009 len -= PAGE_SIZE; 1010 base += PAGE_SIZE; 1011 resid -= PAGE_SIZE; 1012 } 1013 } 1014 return (error); 1015 } 1016 1017 /* 1018 * Common code for vn_io_fault(), agnostic to the kind of i/o request. 1019 * Uses vn_io_fault_doio() to make the call to an actual i/o function. 1020 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request 1021 * into args and call vn_io_fault1() to handle faults during the user 1022 * mode buffer accesses. 1023 */ 1024 static int 1025 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, 1026 struct thread *td) 1027 { 1028 vm_page_t ma[io_hold_cnt + 2]; 1029 struct uio *uio_clone, short_uio; 1030 struct iovec short_iovec[1]; 1031 vm_page_t *prev_td_ma; 1032 vm_prot_t prot; 1033 vm_offset_t addr, end; 1034 size_t len, resid; 1035 ssize_t adv; 1036 int error, cnt, saveheld, prev_td_ma_cnt; 1037 1038 if (vn_io_fault_prefault) { 1039 error = vn_io_fault_prefault_user(uio); 1040 if (error != 0) 1041 return (error); /* Or ignore ? */ 1042 } 1043 1044 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ; 1045 1046 /* 1047 * The UFS follows IO_UNIT directive and replays back both 1048 * uio_offset and uio_resid if an error is encountered during the 1049 * operation. But, since the iovec may be already advanced, 1050 * uio is still in an inconsistent state. 1051 * 1052 * Cache a copy of the original uio, which is advanced to the redo 1053 * point using UIO_NOCOPY below. 1054 */ 1055 uio_clone = cloneuio(uio); 1056 resid = uio->uio_resid; 1057 1058 short_uio.uio_segflg = UIO_USERSPACE; 1059 short_uio.uio_rw = uio->uio_rw; 1060 short_uio.uio_td = uio->uio_td; 1061 1062 error = vn_io_fault_doio(args, uio, td); 1063 if (error != EFAULT) 1064 goto out; 1065 1066 atomic_add_long(&vn_io_faults_cnt, 1); 1067 uio_clone->uio_segflg = UIO_NOCOPY; 1068 uiomove(NULL, resid - uio->uio_resid, uio_clone); 1069 uio_clone->uio_segflg = uio->uio_segflg; 1070 1071 saveheld = curthread_pflags_set(TDP_UIOHELD); 1072 prev_td_ma = td->td_ma; 1073 prev_td_ma_cnt = td->td_ma_cnt; 1074 1075 while (uio_clone->uio_resid != 0) { 1076 len = uio_clone->uio_iov->iov_len; 1077 if (len == 0) { 1078 KASSERT(uio_clone->uio_iovcnt >= 1, 1079 ("iovcnt underflow")); 1080 uio_clone->uio_iov++; 1081 uio_clone->uio_iovcnt--; 1082 continue; 1083 } 1084 if (len > io_hold_cnt * PAGE_SIZE) 1085 len = io_hold_cnt * PAGE_SIZE; 1086 addr = (uintptr_t)uio_clone->uio_iov->iov_base; 1087 end = round_page(addr + len); 1088 if (end < addr) { 1089 error = EFAULT; 1090 break; 1091 } 1092 cnt = atop(end - trunc_page(addr)); 1093 /* 1094 * A perfectly misaligned address and length could cause 1095 * both the start and the end of the chunk to use partial 1096 * page. +2 accounts for such a situation. 1097 */ 1098 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map, 1099 addr, len, prot, ma, io_hold_cnt + 2); 1100 if (cnt == -1) { 1101 error = EFAULT; 1102 break; 1103 } 1104 short_uio.uio_iov = &short_iovec[0]; 1105 short_iovec[0].iov_base = (void *)addr; 1106 short_uio.uio_iovcnt = 1; 1107 short_uio.uio_resid = short_iovec[0].iov_len = len; 1108 short_uio.uio_offset = uio_clone->uio_offset; 1109 td->td_ma = ma; 1110 td->td_ma_cnt = cnt; 1111 1112 error = vn_io_fault_doio(args, &short_uio, td); 1113 vm_page_unhold_pages(ma, cnt); 1114 adv = len - short_uio.uio_resid; 1115 1116 uio_clone->uio_iov->iov_base = 1117 (char *)uio_clone->uio_iov->iov_base + adv; 1118 uio_clone->uio_iov->iov_len -= adv; 1119 uio_clone->uio_resid -= adv; 1120 uio_clone->uio_offset += adv; 1121 1122 uio->uio_resid -= adv; 1123 uio->uio_offset += adv; 1124 1125 if (error != 0 || adv == 0) 1126 break; 1127 } 1128 td->td_ma = prev_td_ma; 1129 td->td_ma_cnt = prev_td_ma_cnt; 1130 curthread_pflags_restore(saveheld); 1131 out: 1132 free(uio_clone, M_IOV); 1133 return (error); 1134 } 1135 1136 static int 1137 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred, 1138 int flags, struct thread *td) 1139 { 1140 fo_rdwr_t *doio; 1141 struct vnode *vp; 1142 void *rl_cookie; 1143 struct vn_io_fault_args args; 1144 int error; 1145 1146 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write; 1147 vp = fp->f_vnode; 1148 foffset_lock_uio(fp, uio, flags); 1149 if (do_vn_io_fault(vp, uio)) { 1150 args.kind = VN_IO_FAULT_FOP; 1151 args.args.fop_args.fp = fp; 1152 args.args.fop_args.doio = doio; 1153 args.cred = active_cred; 1154 args.flags = flags | FOF_OFFSET; 1155 if (uio->uio_rw == UIO_READ) { 1156 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset, 1157 uio->uio_offset + uio->uio_resid); 1158 } else if ((fp->f_flag & O_APPEND) != 0 || 1159 (flags & FOF_OFFSET) == 0) { 1160 /* For appenders, punt and lock the whole range. */ 1161 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); 1162 } else { 1163 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset, 1164 uio->uio_offset + uio->uio_resid); 1165 } 1166 error = vn_io_fault1(vp, uio, &args, td); 1167 vn_rangelock_unlock(vp, rl_cookie); 1168 } else { 1169 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td); 1170 } 1171 foffset_unlock_uio(fp, uio, flags); 1172 return (error); 1173 } 1174 1175 /* 1176 * Helper function to perform the requested uiomove operation using 1177 * the held pages for io->uio_iov[0].iov_base buffer instead of 1178 * copyin/copyout. Access to the pages with uiomove_fromphys() 1179 * instead of iov_base prevents page faults that could occur due to 1180 * pmap_collect() invalidating the mapping created by 1181 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or 1182 * object cleanup revoking the write access from page mappings. 1183 * 1184 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove() 1185 * instead of plain uiomove(). 1186 */ 1187 int 1188 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio) 1189 { 1190 struct uio transp_uio; 1191 struct iovec transp_iov[1]; 1192 struct thread *td; 1193 size_t adv; 1194 int error, pgadv; 1195 1196 td = curthread; 1197 if ((td->td_pflags & TDP_UIOHELD) == 0 || 1198 uio->uio_segflg != UIO_USERSPACE) 1199 return (uiomove(data, xfersize, uio)); 1200 1201 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); 1202 transp_iov[0].iov_base = data; 1203 transp_uio.uio_iov = &transp_iov[0]; 1204 transp_uio.uio_iovcnt = 1; 1205 if (xfersize > uio->uio_resid) 1206 xfersize = uio->uio_resid; 1207 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize; 1208 transp_uio.uio_offset = 0; 1209 transp_uio.uio_segflg = UIO_SYSSPACE; 1210 /* 1211 * Since transp_iov points to data, and td_ma page array 1212 * corresponds to original uio->uio_iov, we need to invert the 1213 * direction of the i/o operation as passed to 1214 * uiomove_fromphys(). 1215 */ 1216 switch (uio->uio_rw) { 1217 case UIO_WRITE: 1218 transp_uio.uio_rw = UIO_READ; 1219 break; 1220 case UIO_READ: 1221 transp_uio.uio_rw = UIO_WRITE; 1222 break; 1223 } 1224 transp_uio.uio_td = uio->uio_td; 1225 error = uiomove_fromphys(td->td_ma, 1226 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK, 1227 xfersize, &transp_uio); 1228 adv = xfersize - transp_uio.uio_resid; 1229 pgadv = 1230 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) - 1231 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT); 1232 td->td_ma += pgadv; 1233 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, 1234 pgadv)); 1235 td->td_ma_cnt -= pgadv; 1236 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv; 1237 uio->uio_iov->iov_len -= adv; 1238 uio->uio_resid -= adv; 1239 uio->uio_offset += adv; 1240 return (error); 1241 } 1242 1243 int 1244 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize, 1245 struct uio *uio) 1246 { 1247 struct thread *td; 1248 vm_offset_t iov_base; 1249 int cnt, pgadv; 1250 1251 td = curthread; 1252 if ((td->td_pflags & TDP_UIOHELD) == 0 || 1253 uio->uio_segflg != UIO_USERSPACE) 1254 return (uiomove_fromphys(ma, offset, xfersize, uio)); 1255 1256 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); 1257 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize; 1258 iov_base = (vm_offset_t)uio->uio_iov->iov_base; 1259 switch (uio->uio_rw) { 1260 case UIO_WRITE: 1261 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma, 1262 offset, cnt); 1263 break; 1264 case UIO_READ: 1265 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK, 1266 cnt); 1267 break; 1268 } 1269 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT); 1270 td->td_ma += pgadv; 1271 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, 1272 pgadv)); 1273 td->td_ma_cnt -= pgadv; 1274 uio->uio_iov->iov_base = (char *)(iov_base + cnt); 1275 uio->uio_iov->iov_len -= cnt; 1276 uio->uio_resid -= cnt; 1277 uio->uio_offset += cnt; 1278 return (0); 1279 } 1280 1281 1282 /* 1283 * File table truncate routine. 1284 */ 1285 static int 1286 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred, 1287 struct thread *td) 1288 { 1289 struct vattr vattr; 1290 struct mount *mp; 1291 struct vnode *vp; 1292 void *rl_cookie; 1293 int error; 1294 1295 vp = fp->f_vnode; 1296 1297 /* 1298 * Lock the whole range for truncation. Otherwise split i/o 1299 * might happen partly before and partly after the truncation. 1300 */ 1301 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); 1302 error = vn_start_write(vp, &mp, V_WAIT | PCATCH); 1303 if (error) 1304 goto out1; 1305 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1306 AUDIT_ARG_VNODE1(vp); 1307 if (vp->v_type == VDIR) { 1308 error = EISDIR; 1309 goto out; 1310 } 1311 #ifdef MAC 1312 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 1313 if (error) 1314 goto out; 1315 #endif 1316 error = vn_writechk(vp); 1317 if (error == 0) { 1318 VATTR_NULL(&vattr); 1319 vattr.va_size = length; 1320 if ((fp->f_flag & O_FSYNC) != 0) 1321 vattr.va_vaflags |= VA_SYNC; 1322 error = VOP_SETATTR(vp, &vattr, fp->f_cred); 1323 } 1324 out: 1325 VOP_UNLOCK(vp, 0); 1326 vn_finished_write(mp); 1327 out1: 1328 vn_rangelock_unlock(vp, rl_cookie); 1329 return (error); 1330 } 1331 1332 /* 1333 * File table vnode stat routine. 1334 */ 1335 static int 1336 vn_statfile(struct file *fp, struct stat *sb, struct ucred *active_cred, 1337 struct thread *td) 1338 { 1339 struct vnode *vp = fp->f_vnode; 1340 int error; 1341 1342 vn_lock(vp, LK_SHARED | LK_RETRY); 1343 error = vn_stat(vp, sb, active_cred, fp->f_cred, td); 1344 VOP_UNLOCK(vp, 0); 1345 1346 return (error); 1347 } 1348 1349 /* 1350 * Stat a vnode; implementation for the stat syscall 1351 */ 1352 int 1353 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *active_cred, 1354 struct ucred *file_cred, struct thread *td) 1355 { 1356 struct vattr vattr; 1357 struct vattr *vap; 1358 int error; 1359 u_short mode; 1360 1361 AUDIT_ARG_VNODE1(vp); 1362 #ifdef MAC 1363 error = mac_vnode_check_stat(active_cred, file_cred, vp); 1364 if (error) 1365 return (error); 1366 #endif 1367 1368 vap = &vattr; 1369 1370 /* 1371 * Initialize defaults for new and unusual fields, so that file 1372 * systems which don't support these fields don't need to know 1373 * about them. 1374 */ 1375 vap->va_birthtime.tv_sec = -1; 1376 vap->va_birthtime.tv_nsec = 0; 1377 vap->va_fsid = VNOVAL; 1378 vap->va_rdev = NODEV; 1379 1380 error = VOP_GETATTR(vp, vap, active_cred); 1381 if (error) 1382 return (error); 1383 1384 /* 1385 * Zero the spare stat fields 1386 */ 1387 bzero(sb, sizeof *sb); 1388 1389 /* 1390 * Copy from vattr table 1391 */ 1392 if (vap->va_fsid != VNOVAL) 1393 sb->st_dev = vap->va_fsid; 1394 else 1395 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 1396 sb->st_ino = vap->va_fileid; 1397 mode = vap->va_mode; 1398 switch (vap->va_type) { 1399 case VREG: 1400 mode |= S_IFREG; 1401 break; 1402 case VDIR: 1403 mode |= S_IFDIR; 1404 break; 1405 case VBLK: 1406 mode |= S_IFBLK; 1407 break; 1408 case VCHR: 1409 mode |= S_IFCHR; 1410 break; 1411 case VLNK: 1412 mode |= S_IFLNK; 1413 break; 1414 case VSOCK: 1415 mode |= S_IFSOCK; 1416 break; 1417 case VFIFO: 1418 mode |= S_IFIFO; 1419 break; 1420 default: 1421 return (EBADF); 1422 } 1423 sb->st_mode = mode; 1424 sb->st_nlink = vap->va_nlink; 1425 sb->st_uid = vap->va_uid; 1426 sb->st_gid = vap->va_gid; 1427 sb->st_rdev = vap->va_rdev; 1428 if (vap->va_size > OFF_MAX) 1429 return (EOVERFLOW); 1430 sb->st_size = vap->va_size; 1431 sb->st_atim = vap->va_atime; 1432 sb->st_mtim = vap->va_mtime; 1433 sb->st_ctim = vap->va_ctime; 1434 sb->st_birthtim = vap->va_birthtime; 1435 1436 /* 1437 * According to www.opengroup.org, the meaning of st_blksize is 1438 * "a filesystem-specific preferred I/O block size for this 1439 * object. In some filesystem types, this may vary from file 1440 * to file" 1441 * Use miminum/default of PAGE_SIZE (e.g. for VCHR). 1442 */ 1443 1444 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize); 1445 1446 sb->st_flags = vap->va_flags; 1447 if (priv_check(td, PRIV_VFS_GENERATION)) 1448 sb->st_gen = 0; 1449 else 1450 sb->st_gen = vap->va_gen; 1451 1452 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 1453 return (0); 1454 } 1455 1456 /* 1457 * File table vnode ioctl routine. 1458 */ 1459 static int 1460 vn_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 1461 struct thread *td) 1462 { 1463 struct vattr vattr; 1464 struct vnode *vp; 1465 int error; 1466 1467 vp = fp->f_vnode; 1468 switch (vp->v_type) { 1469 case VDIR: 1470 case VREG: 1471 switch (com) { 1472 case FIONREAD: 1473 vn_lock(vp, LK_SHARED | LK_RETRY); 1474 error = VOP_GETATTR(vp, &vattr, active_cred); 1475 VOP_UNLOCK(vp, 0); 1476 if (error == 0) 1477 *(int *)data = vattr.va_size - fp->f_offset; 1478 return (error); 1479 case FIONBIO: 1480 case FIOASYNC: 1481 return (0); 1482 default: 1483 return (VOP_IOCTL(vp, com, data, fp->f_flag, 1484 active_cred, td)); 1485 } 1486 break; 1487 case VCHR: 1488 return (VOP_IOCTL(vp, com, data, fp->f_flag, 1489 active_cred, td)); 1490 default: 1491 return (ENOTTY); 1492 } 1493 } 1494 1495 /* 1496 * File table vnode poll routine. 1497 */ 1498 static int 1499 vn_poll(struct file *fp, int events, struct ucred *active_cred, 1500 struct thread *td) 1501 { 1502 struct vnode *vp; 1503 int error; 1504 1505 vp = fp->f_vnode; 1506 #ifdef MAC 1507 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1508 AUDIT_ARG_VNODE1(vp); 1509 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp); 1510 VOP_UNLOCK(vp, 0); 1511 if (!error) 1512 #endif 1513 1514 error = VOP_POLL(vp, events, fp->f_cred, td); 1515 return (error); 1516 } 1517 1518 /* 1519 * Acquire the requested lock and then check for validity. LK_RETRY 1520 * permits vn_lock to return doomed vnodes. 1521 */ 1522 int 1523 _vn_lock(struct vnode *vp, int flags, char *file, int line) 1524 { 1525 int error; 1526 1527 VNASSERT((flags & LK_TYPE_MASK) != 0, vp, 1528 ("vn_lock: no locktype")); 1529 VNASSERT(vp->v_holdcnt != 0, vp, ("vn_lock: zero hold count")); 1530 retry: 1531 error = VOP_LOCK1(vp, flags, file, line); 1532 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */ 1533 KASSERT((flags & LK_RETRY) == 0 || error == 0, 1534 ("vn_lock: error %d incompatible with flags %#x", error, flags)); 1535 1536 if ((flags & LK_RETRY) == 0) { 1537 if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) { 1538 VOP_UNLOCK(vp, 0); 1539 error = ENOENT; 1540 } 1541 } else if (error != 0) 1542 goto retry; 1543 return (error); 1544 } 1545 1546 /* 1547 * File table vnode close routine. 1548 */ 1549 static int 1550 vn_closefile(struct file *fp, struct thread *td) 1551 { 1552 struct vnode *vp; 1553 struct flock lf; 1554 int error; 1555 bool ref; 1556 1557 vp = fp->f_vnode; 1558 fp->f_ops = &badfileops; 1559 ref= (fp->f_flag & FHASLOCK) != 0 && fp->f_type == DTYPE_VNODE; 1560 1561 error = vn_close1(vp, fp->f_flag, fp->f_cred, td, ref); 1562 1563 if (__predict_false(ref)) { 1564 lf.l_whence = SEEK_SET; 1565 lf.l_start = 0; 1566 lf.l_len = 0; 1567 lf.l_type = F_UNLCK; 1568 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); 1569 vrele(vp); 1570 } 1571 return (error); 1572 } 1573 1574 static bool 1575 vn_suspendable(struct mount *mp) 1576 { 1577 1578 return (mp->mnt_op->vfs_susp_clean != NULL); 1579 } 1580 1581 /* 1582 * Preparing to start a filesystem write operation. If the operation is 1583 * permitted, then we bump the count of operations in progress and 1584 * proceed. If a suspend request is in progress, we wait until the 1585 * suspension is over, and then proceed. 1586 */ 1587 static int 1588 vn_start_write_locked(struct mount *mp, int flags) 1589 { 1590 int error, mflags; 1591 1592 mtx_assert(MNT_MTX(mp), MA_OWNED); 1593 error = 0; 1594 1595 /* 1596 * Check on status of suspension. 1597 */ 1598 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 || 1599 mp->mnt_susp_owner != curthread) { 1600 mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? 1601 (flags & PCATCH) : 0) | (PUSER - 1); 1602 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 1603 if (flags & V_NOWAIT) { 1604 error = EWOULDBLOCK; 1605 goto unlock; 1606 } 1607 error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags, 1608 "suspfs", 0); 1609 if (error) 1610 goto unlock; 1611 } 1612 } 1613 if (flags & V_XSLEEP) 1614 goto unlock; 1615 mp->mnt_writeopcount++; 1616 unlock: 1617 if (error != 0 || (flags & V_XSLEEP) != 0) 1618 MNT_REL(mp); 1619 MNT_IUNLOCK(mp); 1620 return (error); 1621 } 1622 1623 int 1624 vn_start_write(struct vnode *vp, struct mount **mpp, int flags) 1625 { 1626 struct mount *mp; 1627 int error; 1628 1629 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), 1630 ("V_MNTREF requires mp")); 1631 1632 error = 0; 1633 /* 1634 * If a vnode is provided, get and return the mount point that 1635 * to which it will write. 1636 */ 1637 if (vp != NULL) { 1638 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 1639 *mpp = NULL; 1640 if (error != EOPNOTSUPP) 1641 return (error); 1642 return (0); 1643 } 1644 } 1645 if ((mp = *mpp) == NULL) 1646 return (0); 1647 1648 if (!vn_suspendable(mp)) { 1649 if (vp != NULL || (flags & V_MNTREF) != 0) 1650 vfs_rel(mp); 1651 return (0); 1652 } 1653 1654 /* 1655 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 1656 * a vfs_ref(). 1657 * As long as a vnode is not provided we need to acquire a 1658 * refcount for the provided mountpoint too, in order to 1659 * emulate a vfs_ref(). 1660 */ 1661 MNT_ILOCK(mp); 1662 if (vp == NULL && (flags & V_MNTREF) == 0) 1663 MNT_REF(mp); 1664 1665 return (vn_start_write_locked(mp, flags)); 1666 } 1667 1668 /* 1669 * Secondary suspension. Used by operations such as vop_inactive 1670 * routines that are needed by the higher level functions. These 1671 * are allowed to proceed until all the higher level functions have 1672 * completed (indicated by mnt_writeopcount dropping to zero). At that 1673 * time, these operations are halted until the suspension is over. 1674 */ 1675 int 1676 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags) 1677 { 1678 struct mount *mp; 1679 int error; 1680 1681 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), 1682 ("V_MNTREF requires mp")); 1683 1684 retry: 1685 if (vp != NULL) { 1686 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 1687 *mpp = NULL; 1688 if (error != EOPNOTSUPP) 1689 return (error); 1690 return (0); 1691 } 1692 } 1693 /* 1694 * If we are not suspended or have not yet reached suspended 1695 * mode, then let the operation proceed. 1696 */ 1697 if ((mp = *mpp) == NULL) 1698 return (0); 1699 1700 if (!vn_suspendable(mp)) { 1701 if (vp != NULL || (flags & V_MNTREF) != 0) 1702 vfs_rel(mp); 1703 return (0); 1704 } 1705 1706 /* 1707 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 1708 * a vfs_ref(). 1709 * As long as a vnode is not provided we need to acquire a 1710 * refcount for the provided mountpoint too, in order to 1711 * emulate a vfs_ref(). 1712 */ 1713 MNT_ILOCK(mp); 1714 if (vp == NULL && (flags & V_MNTREF) == 0) 1715 MNT_REF(mp); 1716 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) { 1717 mp->mnt_secondary_writes++; 1718 mp->mnt_secondary_accwrites++; 1719 MNT_IUNLOCK(mp); 1720 return (0); 1721 } 1722 if (flags & V_NOWAIT) { 1723 MNT_REL(mp); 1724 MNT_IUNLOCK(mp); 1725 return (EWOULDBLOCK); 1726 } 1727 /* 1728 * Wait for the suspension to finish. 1729 */ 1730 error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP | 1731 ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0), 1732 "suspfs", 0); 1733 vfs_rel(mp); 1734 if (error == 0) 1735 goto retry; 1736 return (error); 1737 } 1738 1739 /* 1740 * Filesystem write operation has completed. If we are suspending and this 1741 * operation is the last one, notify the suspender that the suspension is 1742 * now in effect. 1743 */ 1744 void 1745 vn_finished_write(struct mount *mp) 1746 { 1747 if (mp == NULL || !vn_suspendable(mp)) 1748 return; 1749 MNT_ILOCK(mp); 1750 MNT_REL(mp); 1751 mp->mnt_writeopcount--; 1752 if (mp->mnt_writeopcount < 0) 1753 panic("vn_finished_write: neg cnt"); 1754 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1755 mp->mnt_writeopcount <= 0) 1756 wakeup(&mp->mnt_writeopcount); 1757 MNT_IUNLOCK(mp); 1758 } 1759 1760 1761 /* 1762 * Filesystem secondary write operation has completed. If we are 1763 * suspending and this operation is the last one, notify the suspender 1764 * that the suspension is now in effect. 1765 */ 1766 void 1767 vn_finished_secondary_write(struct mount *mp) 1768 { 1769 if (mp == NULL || !vn_suspendable(mp)) 1770 return; 1771 MNT_ILOCK(mp); 1772 MNT_REL(mp); 1773 mp->mnt_secondary_writes--; 1774 if (mp->mnt_secondary_writes < 0) 1775 panic("vn_finished_secondary_write: neg cnt"); 1776 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1777 mp->mnt_secondary_writes <= 0) 1778 wakeup(&mp->mnt_secondary_writes); 1779 MNT_IUNLOCK(mp); 1780 } 1781 1782 1783 1784 /* 1785 * Request a filesystem to suspend write operations. 1786 */ 1787 int 1788 vfs_write_suspend(struct mount *mp, int flags) 1789 { 1790 int error; 1791 1792 MPASS(vn_suspendable(mp)); 1793 1794 MNT_ILOCK(mp); 1795 if (mp->mnt_susp_owner == curthread) { 1796 MNT_IUNLOCK(mp); 1797 return (EALREADY); 1798 } 1799 while (mp->mnt_kern_flag & MNTK_SUSPEND) 1800 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0); 1801 1802 /* 1803 * Unmount holds a write reference on the mount point. If we 1804 * own busy reference and drain for writers, we deadlock with 1805 * the reference draining in the unmount path. Callers of 1806 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if 1807 * vfs_busy() reference is owned and caller is not in the 1808 * unmount context. 1809 */ 1810 if ((flags & VS_SKIP_UNMOUNT) != 0 && 1811 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { 1812 MNT_IUNLOCK(mp); 1813 return (EBUSY); 1814 } 1815 1816 mp->mnt_kern_flag |= MNTK_SUSPEND; 1817 mp->mnt_susp_owner = curthread; 1818 if (mp->mnt_writeopcount > 0) 1819 (void) msleep(&mp->mnt_writeopcount, 1820 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0); 1821 else 1822 MNT_IUNLOCK(mp); 1823 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0) 1824 vfs_write_resume(mp, 0); 1825 return (error); 1826 } 1827 1828 /* 1829 * Request a filesystem to resume write operations. 1830 */ 1831 void 1832 vfs_write_resume(struct mount *mp, int flags) 1833 { 1834 1835 MPASS(vn_suspendable(mp)); 1836 1837 MNT_ILOCK(mp); 1838 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 1839 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner")); 1840 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 | 1841 MNTK_SUSPENDED); 1842 mp->mnt_susp_owner = NULL; 1843 wakeup(&mp->mnt_writeopcount); 1844 wakeup(&mp->mnt_flag); 1845 curthread->td_pflags &= ~TDP_IGNSUSP; 1846 if ((flags & VR_START_WRITE) != 0) { 1847 MNT_REF(mp); 1848 mp->mnt_writeopcount++; 1849 } 1850 MNT_IUNLOCK(mp); 1851 if ((flags & VR_NO_SUSPCLR) == 0) 1852 VFS_SUSP_CLEAN(mp); 1853 } else if ((flags & VR_START_WRITE) != 0) { 1854 MNT_REF(mp); 1855 vn_start_write_locked(mp, 0); 1856 } else { 1857 MNT_IUNLOCK(mp); 1858 } 1859 } 1860 1861 /* 1862 * Helper loop around vfs_write_suspend() for filesystem unmount VFS 1863 * methods. 1864 */ 1865 int 1866 vfs_write_suspend_umnt(struct mount *mp) 1867 { 1868 int error; 1869 1870 MPASS(vn_suspendable(mp)); 1871 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0, 1872 ("vfs_write_suspend_umnt: recursed")); 1873 1874 /* dounmount() already called vn_start_write(). */ 1875 for (;;) { 1876 vn_finished_write(mp); 1877 error = vfs_write_suspend(mp, 0); 1878 if (error != 0) { 1879 vn_start_write(NULL, &mp, V_WAIT); 1880 return (error); 1881 } 1882 MNT_ILOCK(mp); 1883 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0) 1884 break; 1885 MNT_IUNLOCK(mp); 1886 vn_start_write(NULL, &mp, V_WAIT); 1887 } 1888 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); 1889 wakeup(&mp->mnt_flag); 1890 MNT_IUNLOCK(mp); 1891 curthread->td_pflags |= TDP_IGNSUSP; 1892 return (0); 1893 } 1894 1895 /* 1896 * Implement kqueues for files by translating it to vnode operation. 1897 */ 1898 static int 1899 vn_kqfilter(struct file *fp, struct knote *kn) 1900 { 1901 1902 return (VOP_KQFILTER(fp->f_vnode, kn)); 1903 } 1904 1905 /* 1906 * Simplified in-kernel wrapper calls for extended attribute access. 1907 * Both calls pass in a NULL credential, authorizing as "kernel" access. 1908 * Set IO_NODELOCKED in ioflg if the vnode is already locked. 1909 */ 1910 int 1911 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, 1912 const char *attrname, int *buflen, char *buf, struct thread *td) 1913 { 1914 struct uio auio; 1915 struct iovec iov; 1916 int error; 1917 1918 iov.iov_len = *buflen; 1919 iov.iov_base = buf; 1920 1921 auio.uio_iov = &iov; 1922 auio.uio_iovcnt = 1; 1923 auio.uio_rw = UIO_READ; 1924 auio.uio_segflg = UIO_SYSSPACE; 1925 auio.uio_td = td; 1926 auio.uio_offset = 0; 1927 auio.uio_resid = *buflen; 1928 1929 if ((ioflg & IO_NODELOCKED) == 0) 1930 vn_lock(vp, LK_SHARED | LK_RETRY); 1931 1932 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1933 1934 /* authorize attribute retrieval as kernel */ 1935 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL, 1936 td); 1937 1938 if ((ioflg & IO_NODELOCKED) == 0) 1939 VOP_UNLOCK(vp, 0); 1940 1941 if (error == 0) { 1942 *buflen = *buflen - auio.uio_resid; 1943 } 1944 1945 return (error); 1946 } 1947 1948 /* 1949 * XXX failure mode if partially written? 1950 */ 1951 int 1952 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, 1953 const char *attrname, int buflen, char *buf, struct thread *td) 1954 { 1955 struct uio auio; 1956 struct iovec iov; 1957 struct mount *mp; 1958 int error; 1959 1960 iov.iov_len = buflen; 1961 iov.iov_base = buf; 1962 1963 auio.uio_iov = &iov; 1964 auio.uio_iovcnt = 1; 1965 auio.uio_rw = UIO_WRITE; 1966 auio.uio_segflg = UIO_SYSSPACE; 1967 auio.uio_td = td; 1968 auio.uio_offset = 0; 1969 auio.uio_resid = buflen; 1970 1971 if ((ioflg & IO_NODELOCKED) == 0) { 1972 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 1973 return (error); 1974 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1975 } 1976 1977 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1978 1979 /* authorize attribute setting as kernel */ 1980 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td); 1981 1982 if ((ioflg & IO_NODELOCKED) == 0) { 1983 vn_finished_write(mp); 1984 VOP_UNLOCK(vp, 0); 1985 } 1986 1987 return (error); 1988 } 1989 1990 int 1991 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, 1992 const char *attrname, struct thread *td) 1993 { 1994 struct mount *mp; 1995 int error; 1996 1997 if ((ioflg & IO_NODELOCKED) == 0) { 1998 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 1999 return (error); 2000 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2001 } 2002 2003 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 2004 2005 /* authorize attribute removal as kernel */ 2006 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td); 2007 if (error == EOPNOTSUPP) 2008 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, 2009 NULL, td); 2010 2011 if ((ioflg & IO_NODELOCKED) == 0) { 2012 vn_finished_write(mp); 2013 VOP_UNLOCK(vp, 0); 2014 } 2015 2016 return (error); 2017 } 2018 2019 static int 2020 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags, 2021 struct vnode **rvp) 2022 { 2023 2024 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp)); 2025 } 2026 2027 int 2028 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp) 2029 { 2030 2031 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino, 2032 lkflags, rvp)); 2033 } 2034 2035 int 2036 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg, 2037 int lkflags, struct vnode **rvp) 2038 { 2039 struct mount *mp; 2040 int ltype, error; 2041 2042 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get"); 2043 mp = vp->v_mount; 2044 ltype = VOP_ISLOCKED(vp); 2045 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED, 2046 ("vn_vget_ino: vp not locked")); 2047 error = vfs_busy(mp, MBF_NOWAIT); 2048 if (error != 0) { 2049 vfs_ref(mp); 2050 VOP_UNLOCK(vp, 0); 2051 error = vfs_busy(mp, 0); 2052 vn_lock(vp, ltype | LK_RETRY); 2053 vfs_rel(mp); 2054 if (error != 0) 2055 return (ENOENT); 2056 if (vp->v_iflag & VI_DOOMED) { 2057 vfs_unbusy(mp); 2058 return (ENOENT); 2059 } 2060 } 2061 VOP_UNLOCK(vp, 0); 2062 error = alloc(mp, alloc_arg, lkflags, rvp); 2063 vfs_unbusy(mp); 2064 if (*rvp != vp) 2065 vn_lock(vp, ltype | LK_RETRY); 2066 if (vp->v_iflag & VI_DOOMED) { 2067 if (error == 0) { 2068 if (*rvp == vp) 2069 vunref(vp); 2070 else 2071 vput(*rvp); 2072 } 2073 error = ENOENT; 2074 } 2075 return (error); 2076 } 2077 2078 int 2079 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, 2080 struct thread *td) 2081 { 2082 2083 if (vp->v_type != VREG || td == NULL) 2084 return (0); 2085 if ((uoff_t)uio->uio_offset + uio->uio_resid > 2086 lim_cur(td, RLIMIT_FSIZE)) { 2087 PROC_LOCK(td->td_proc); 2088 kern_psignal(td->td_proc, SIGXFSZ); 2089 PROC_UNLOCK(td->td_proc); 2090 return (EFBIG); 2091 } 2092 return (0); 2093 } 2094 2095 int 2096 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 2097 struct thread *td) 2098 { 2099 struct vnode *vp; 2100 2101 vp = fp->f_vnode; 2102 #ifdef AUDIT 2103 vn_lock(vp, LK_SHARED | LK_RETRY); 2104 AUDIT_ARG_VNODE1(vp); 2105 VOP_UNLOCK(vp, 0); 2106 #endif 2107 return (setfmode(td, active_cred, vp, mode)); 2108 } 2109 2110 int 2111 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 2112 struct thread *td) 2113 { 2114 struct vnode *vp; 2115 2116 vp = fp->f_vnode; 2117 #ifdef AUDIT 2118 vn_lock(vp, LK_SHARED | LK_RETRY); 2119 AUDIT_ARG_VNODE1(vp); 2120 VOP_UNLOCK(vp, 0); 2121 #endif 2122 return (setfown(td, active_cred, vp, uid, gid)); 2123 } 2124 2125 void 2126 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end) 2127 { 2128 vm_object_t object; 2129 2130 if ((object = vp->v_object) == NULL) 2131 return; 2132 VM_OBJECT_WLOCK(object); 2133 vm_object_page_remove(object, start, end, 0); 2134 VM_OBJECT_WUNLOCK(object); 2135 } 2136 2137 int 2138 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred) 2139 { 2140 struct vattr va; 2141 daddr_t bn, bnp; 2142 uint64_t bsize; 2143 off_t noff; 2144 int error; 2145 2146 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA, 2147 ("Wrong command %lu", cmd)); 2148 2149 if (vn_lock(vp, LK_SHARED) != 0) 2150 return (EBADF); 2151 if (vp->v_type != VREG) { 2152 error = ENOTTY; 2153 goto unlock; 2154 } 2155 error = VOP_GETATTR(vp, &va, cred); 2156 if (error != 0) 2157 goto unlock; 2158 noff = *off; 2159 if (noff >= va.va_size) { 2160 error = ENXIO; 2161 goto unlock; 2162 } 2163 bsize = vp->v_mount->mnt_stat.f_iosize; 2164 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) { 2165 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL); 2166 if (error == EOPNOTSUPP) { 2167 error = ENOTTY; 2168 goto unlock; 2169 } 2170 if ((bnp == -1 && cmd == FIOSEEKHOLE) || 2171 (bnp != -1 && cmd == FIOSEEKDATA)) { 2172 noff = bn * bsize; 2173 if (noff < *off) 2174 noff = *off; 2175 goto unlock; 2176 } 2177 } 2178 if (noff > va.va_size) 2179 noff = va.va_size; 2180 /* noff == va.va_size. There is an implicit hole at the end of file. */ 2181 if (cmd == FIOSEEKDATA) 2182 error = ENXIO; 2183 unlock: 2184 VOP_UNLOCK(vp, 0); 2185 if (error == 0) 2186 *off = noff; 2187 return (error); 2188 } 2189 2190 int 2191 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td) 2192 { 2193 struct ucred *cred; 2194 struct vnode *vp; 2195 struct vattr vattr; 2196 off_t foffset, size; 2197 int error, noneg; 2198 2199 cred = td->td_ucred; 2200 vp = fp->f_vnode; 2201 foffset = foffset_lock(fp, 0); 2202 noneg = (vp->v_type != VCHR); 2203 error = 0; 2204 switch (whence) { 2205 case L_INCR: 2206 if (noneg && 2207 (foffset < 0 || 2208 (offset > 0 && foffset > OFF_MAX - offset))) { 2209 error = EOVERFLOW; 2210 break; 2211 } 2212 offset += foffset; 2213 break; 2214 case L_XTND: 2215 vn_lock(vp, LK_SHARED | LK_RETRY); 2216 error = VOP_GETATTR(vp, &vattr, cred); 2217 VOP_UNLOCK(vp, 0); 2218 if (error) 2219 break; 2220 2221 /* 2222 * If the file references a disk device, then fetch 2223 * the media size and use that to determine the ending 2224 * offset. 2225 */ 2226 if (vattr.va_size == 0 && vp->v_type == VCHR && 2227 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0) 2228 vattr.va_size = size; 2229 if (noneg && 2230 (vattr.va_size > OFF_MAX || 2231 (offset > 0 && vattr.va_size > OFF_MAX - offset))) { 2232 error = EOVERFLOW; 2233 break; 2234 } 2235 offset += vattr.va_size; 2236 break; 2237 case L_SET: 2238 break; 2239 case SEEK_DATA: 2240 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td); 2241 break; 2242 case SEEK_HOLE: 2243 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td); 2244 break; 2245 default: 2246 error = EINVAL; 2247 } 2248 if (error == 0 && noneg && offset < 0) 2249 error = EINVAL; 2250 if (error != 0) 2251 goto drop; 2252 VFS_KNOTE_UNLOCKED(vp, 0); 2253 td->td_uretoff.tdu_off = offset; 2254 drop: 2255 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); 2256 return (error); 2257 } 2258 2259 int 2260 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred, 2261 struct thread *td) 2262 { 2263 int error; 2264 2265 /* 2266 * Grant permission if the caller is the owner of the file, or 2267 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on 2268 * on the file. If the time pointer is null, then write 2269 * permission on the file is also sufficient. 2270 * 2271 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes: 2272 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES 2273 * will be allowed to set the times [..] to the current 2274 * server time. 2275 */ 2276 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td); 2277 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0) 2278 error = VOP_ACCESS(vp, VWRITE, cred, td); 2279 return (error); 2280 } 2281 2282 int 2283 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 2284 { 2285 struct vnode *vp; 2286 int error; 2287 2288 if (fp->f_type == DTYPE_FIFO) 2289 kif->kf_type = KF_TYPE_FIFO; 2290 else 2291 kif->kf_type = KF_TYPE_VNODE; 2292 vp = fp->f_vnode; 2293 vref(vp); 2294 FILEDESC_SUNLOCK(fdp); 2295 error = vn_fill_kinfo_vnode(vp, kif); 2296 vrele(vp); 2297 FILEDESC_SLOCK(fdp); 2298 return (error); 2299 } 2300 2301 static inline void 2302 vn_fill_junk(struct kinfo_file *kif) 2303 { 2304 size_t len, olen; 2305 2306 /* 2307 * Simulate vn_fullpath returning changing values for a given 2308 * vp during e.g. coredump. 2309 */ 2310 len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1; 2311 olen = strlen(kif->kf_path); 2312 if (len < olen) 2313 strcpy(&kif->kf_path[len - 1], "$"); 2314 else 2315 for (; olen < len; olen++) 2316 strcpy(&kif->kf_path[olen], "A"); 2317 } 2318 2319 int 2320 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif) 2321 { 2322 struct vattr va; 2323 char *fullpath, *freepath; 2324 int error; 2325 2326 kif->kf_un.kf_file.kf_file_type = vntype_to_kinfo(vp->v_type); 2327 freepath = NULL; 2328 fullpath = "-"; 2329 error = vn_fullpath(curthread, vp, &fullpath, &freepath); 2330 if (error == 0) { 2331 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 2332 } 2333 if (freepath != NULL) 2334 free(freepath, M_TEMP); 2335 2336 KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path, 2337 vn_fill_junk(kif); 2338 ); 2339 2340 /* 2341 * Retrieve vnode attributes. 2342 */ 2343 va.va_fsid = VNOVAL; 2344 va.va_rdev = NODEV; 2345 vn_lock(vp, LK_SHARED | LK_RETRY); 2346 error = VOP_GETATTR(vp, &va, curthread->td_ucred); 2347 VOP_UNLOCK(vp, 0); 2348 if (error != 0) 2349 return (error); 2350 if (va.va_fsid != VNOVAL) 2351 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; 2352 else 2353 kif->kf_un.kf_file.kf_file_fsid = 2354 vp->v_mount->mnt_stat.f_fsid.val[0]; 2355 kif->kf_un.kf_file.kf_file_fsid_freebsd11 = 2356 kif->kf_un.kf_file.kf_file_fsid; /* truncate */ 2357 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; 2358 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); 2359 kif->kf_un.kf_file.kf_file_size = va.va_size; 2360 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; 2361 kif->kf_un.kf_file.kf_file_rdev_freebsd11 = 2362 kif->kf_un.kf_file.kf_file_rdev; /* truncate */ 2363 return (0); 2364 } 2365 2366 int 2367 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, 2368 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, 2369 struct thread *td) 2370 { 2371 #ifdef HWPMC_HOOKS 2372 struct pmckern_map_in pkm; 2373 #endif 2374 struct mount *mp; 2375 struct vnode *vp; 2376 vm_object_t object; 2377 vm_prot_t maxprot; 2378 boolean_t writecounted; 2379 int error; 2380 2381 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \ 2382 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) 2383 /* 2384 * POSIX shared-memory objects are defined to have 2385 * kernel persistence, and are not defined to support 2386 * read(2)/write(2) -- or even open(2). Thus, we can 2387 * use MAP_ASYNC to trade on-disk coherence for speed. 2388 * The shm_open(3) library routine turns on the FPOSIXSHM 2389 * flag to request this behavior. 2390 */ 2391 if ((fp->f_flag & FPOSIXSHM) != 0) 2392 flags |= MAP_NOSYNC; 2393 #endif 2394 vp = fp->f_vnode; 2395 2396 /* 2397 * Ensure that file and memory protections are 2398 * compatible. Note that we only worry about 2399 * writability if mapping is shared; in this case, 2400 * current and max prot are dictated by the open file. 2401 * XXX use the vnode instead? Problem is: what 2402 * credentials do we use for determination? What if 2403 * proc does a setuid? 2404 */ 2405 mp = vp->v_mount; 2406 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) { 2407 maxprot = VM_PROT_NONE; 2408 if ((prot & VM_PROT_EXECUTE) != 0) 2409 return (EACCES); 2410 } else 2411 maxprot = VM_PROT_EXECUTE; 2412 if ((fp->f_flag & FREAD) != 0) 2413 maxprot |= VM_PROT_READ; 2414 else if ((prot & VM_PROT_READ) != 0) 2415 return (EACCES); 2416 2417 /* 2418 * If we are sharing potential changes via MAP_SHARED and we 2419 * are trying to get write permission although we opened it 2420 * without asking for it, bail out. 2421 */ 2422 if ((flags & MAP_SHARED) != 0) { 2423 if ((fp->f_flag & FWRITE) != 0) 2424 maxprot |= VM_PROT_WRITE; 2425 else if ((prot & VM_PROT_WRITE) != 0) 2426 return (EACCES); 2427 } else { 2428 maxprot |= VM_PROT_WRITE; 2429 cap_maxprot |= VM_PROT_WRITE; 2430 } 2431 maxprot &= cap_maxprot; 2432 2433 /* 2434 * For regular files and shared memory, POSIX requires that 2435 * the value of foff be a legitimate offset within the data 2436 * object. In particular, negative offsets are invalid. 2437 * Blocking negative offsets and overflows here avoids 2438 * possible wraparound or user-level access into reserved 2439 * ranges of the data object later. In contrast, POSIX does 2440 * not dictate how offsets are used by device drivers, so in 2441 * the case of a device mapping a negative offset is passed 2442 * on. 2443 */ 2444 if ( 2445 #ifdef _LP64 2446 size > OFF_MAX || 2447 #endif 2448 foff < 0 || foff > OFF_MAX - size) 2449 return (EINVAL); 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 (PMC_HOOK_INSTALLED(PMC_FN_MMAP)) { 2470 if ((prot & VM_PROT_EXECUTE) != 0 && error == 0) { 2471 pkm.pm_file = vp; 2472 pkm.pm_address = (uintptr_t) *addr; 2473 PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_MMAP, (void *) &pkm); 2474 } 2475 } 2476 #endif 2477 return (error); 2478 } 2479 2480 void 2481 vn_fsid(struct vnode *vp, struct vattr *va) 2482 { 2483 fsid_t *f; 2484 2485 f = &vp->v_mount->mnt_stat.f_fsid; 2486 va->va_fsid = (uint32_t)f->val[1]; 2487 va->va_fsid <<= sizeof(f->val[1]) * NBBY; 2488 va->va_fsid += (uint32_t)f->val[0]; 2489 } 2490