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: stable/10/sys/kern/vfs_vnops.c 338605 2018-09-12 05:03:30Z gordon $"); 45 46#include <sys/param.h> 47#include <sys/systm.h> 48#include <sys/disk.h> 49#include <sys/fcntl.h> 50#include <sys/file.h> 51#include <sys/kdb.h> 52#include <sys/stat.h> 53#include <sys/priv.h> 54#include <sys/proc.h> 55#include <sys/limits.h> 56#include <sys/lock.h> 57#include <sys/mount.h> 58#include <sys/mutex.h> 59#include <sys/namei.h> 60#include <sys/vnode.h> 61#include <sys/bio.h> 62#include <sys/buf.h> 63#include <sys/filio.h> 64#include <sys/resourcevar.h> 65#include <sys/rwlock.h> 66#include <sys/sx.h> 67#include <sys/sysctl.h> 68#include <sys/ttycom.h> 69#include <sys/conf.h> 70#include <sys/syslog.h> 71#include <sys/unistd.h> 72 73#include <security/audit/audit.h> 74#include <security/mac/mac_framework.h> 75 76#include <vm/vm.h> 77#include <vm/vm_extern.h> 78#include <vm/pmap.h> 79#include <vm/vm_map.h> 80#include <vm/vm_object.h> 81#include <vm/vm_page.h> 82 83static fo_rdwr_t vn_read; 84static fo_rdwr_t vn_write; 85static fo_rdwr_t vn_io_fault; 86static fo_truncate_t vn_truncate; 87static fo_ioctl_t vn_ioctl; 88static fo_poll_t vn_poll; 89static fo_kqfilter_t vn_kqfilter; 90static fo_stat_t vn_statfile; 91static fo_close_t vn_closefile; 92 93struct fileops vnops = { 94 .fo_read = vn_io_fault, 95 .fo_write = vn_io_fault, 96 .fo_truncate = vn_truncate, 97 .fo_ioctl = vn_ioctl, 98 .fo_poll = vn_poll, 99 .fo_kqfilter = vn_kqfilter, 100 .fo_stat = vn_statfile, 101 .fo_close = vn_closefile, 102 .fo_chmod = vn_chmod, 103 .fo_chown = vn_chown, 104 .fo_sendfile = vn_sendfile, 105 .fo_seek = vn_seek, 106 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 107}; 108 109static const int io_hold_cnt = 16; 110static int vn_io_fault_enable = 1; 111SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW, 112 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance"); 113static int vn_io_fault_prefault = 0; 114SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW, 115 &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting"); 116static u_long vn_io_faults_cnt; 117SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD, 118 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers"); 119 120/* 121 * Returns true if vn_io_fault mode of handling the i/o request should 122 * be used. 123 */ 124static bool 125do_vn_io_fault(struct vnode *vp, struct uio *uio) 126{ 127 struct mount *mp; 128 129 return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG && 130 (mp = vp->v_mount) != NULL && 131 (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable); 132} 133 134/* 135 * Structure used to pass arguments to vn_io_fault1(), to do either 136 * file- or vnode-based I/O calls. 137 */ 138struct vn_io_fault_args { 139 enum { 140 VN_IO_FAULT_FOP, 141 VN_IO_FAULT_VOP 142 } kind; 143 struct ucred *cred; 144 int flags; 145 union { 146 struct fop_args_tag { 147 struct file *fp; 148 fo_rdwr_t *doio; 149 } fop_args; 150 struct vop_args_tag { 151 struct vnode *vp; 152 } vop_args; 153 } args; 154}; 155 156static int vn_io_fault1(struct vnode *vp, struct uio *uio, 157 struct vn_io_fault_args *args, struct thread *td); 158 159int 160vn_open(ndp, flagp, cmode, fp) 161 struct nameidata *ndp; 162 int *flagp, cmode; 163 struct file *fp; 164{ 165 struct thread *td = ndp->ni_cnd.cn_thread; 166 167 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp)); 168} 169 170/* 171 * Common code for vnode open operations via a name lookup. 172 * Lookup the vnode and invoke VOP_CREATE if needed. 173 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine. 174 * 175 * Note that this does NOT free nameidata for the successful case, 176 * due to the NDINIT being done elsewhere. 177 */ 178int 179vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, 180 struct ucred *cred, struct file *fp) 181{ 182 struct vnode *vp; 183 struct mount *mp; 184 struct thread *td = ndp->ni_cnd.cn_thread; 185 struct vattr vat; 186 struct vattr *vap = &vat; 187 int fmode, error; 188 189restart: 190 fmode = *flagp; 191 if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT | 192 O_EXCL | O_DIRECTORY)) 193 return (EINVAL); 194 else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) { 195 ndp->ni_cnd.cn_nameiop = CREATE; 196 /* 197 * Set NOCACHE to avoid flushing the cache when 198 * rolling in many files at once. 199 */ 200 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE; 201 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) 202 ndp->ni_cnd.cn_flags |= FOLLOW; 203 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 204 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 205 if (vn_open_flags & VN_OPEN_NOCAPCHECK) 206 ndp->ni_cnd.cn_flags |= NOCAPCHECK; 207 bwillwrite(); 208 if ((error = namei(ndp)) != 0) 209 return (error); 210 if (ndp->ni_vp == NULL) { 211 VATTR_NULL(vap); 212 vap->va_type = VREG; 213 vap->va_mode = cmode; 214 if (fmode & O_EXCL) 215 vap->va_vaflags |= VA_EXCLUSIVE; 216 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) { 217 NDFREE(ndp, NDF_ONLY_PNBUF); 218 vput(ndp->ni_dvp); 219 if ((error = vn_start_write(NULL, &mp, 220 V_XSLEEP | PCATCH)) != 0) 221 return (error); 222 goto restart; 223 } 224 if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0) 225 ndp->ni_cnd.cn_flags |= MAKEENTRY; 226#ifdef MAC 227 error = mac_vnode_check_create(cred, ndp->ni_dvp, 228 &ndp->ni_cnd, vap); 229 if (error == 0) 230#endif 231 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 232 &ndp->ni_cnd, vap); 233 vput(ndp->ni_dvp); 234 vn_finished_write(mp); 235 if (error) { 236 NDFREE(ndp, NDF_ONLY_PNBUF); 237 return (error); 238 } 239 fmode &= ~O_TRUNC; 240 vp = ndp->ni_vp; 241 } else { 242 if (ndp->ni_dvp == ndp->ni_vp) 243 vrele(ndp->ni_dvp); 244 else 245 vput(ndp->ni_dvp); 246 ndp->ni_dvp = NULL; 247 vp = ndp->ni_vp; 248 if (fmode & O_EXCL) { 249 error = EEXIST; 250 goto bad; 251 } 252 fmode &= ~O_CREAT; 253 } 254 } else { 255 ndp->ni_cnd.cn_nameiop = LOOKUP; 256 ndp->ni_cnd.cn_flags = ISOPEN | 257 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF; 258 if (!(fmode & FWRITE)) 259 ndp->ni_cnd.cn_flags |= LOCKSHARED; 260 if (!(vn_open_flags & VN_OPEN_NOAUDIT)) 261 ndp->ni_cnd.cn_flags |= AUDITVNODE1; 262 if (vn_open_flags & VN_OPEN_NOCAPCHECK) 263 ndp->ni_cnd.cn_flags |= NOCAPCHECK; 264 if ((error = namei(ndp)) != 0) 265 return (error); 266 vp = ndp->ni_vp; 267 } 268 error = vn_open_vnode(vp, fmode, cred, td, fp); 269 if (error) 270 goto bad; 271 *flagp = fmode; 272 return (0); 273bad: 274 NDFREE(ndp, NDF_ONLY_PNBUF); 275 vput(vp); 276 *flagp = fmode; 277 ndp->ni_vp = NULL; 278 return (error); 279} 280 281/* 282 * Common code for vnode open operations once a vnode is located. 283 * Check permissions, and call the VOP_OPEN routine. 284 */ 285int 286vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, 287 struct thread *td, struct file *fp) 288{ 289 struct mount *mp; 290 accmode_t accmode; 291 struct flock lf; 292 int error, have_flock, lock_flags, type; 293 294 if (vp->v_type == VLNK) 295 return (EMLINK); 296 if (vp->v_type == VSOCK) 297 return (EOPNOTSUPP); 298 if (vp->v_type != VDIR && fmode & O_DIRECTORY) 299 return (ENOTDIR); 300 accmode = 0; 301 if (fmode & (FWRITE | O_TRUNC)) { 302 if (vp->v_type == VDIR) 303 return (EISDIR); 304 accmode |= VWRITE; 305 } 306 if (fmode & FREAD) 307 accmode |= VREAD; 308 if (fmode & FEXEC) 309 accmode |= VEXEC; 310 if ((fmode & O_APPEND) && (fmode & FWRITE)) 311 accmode |= VAPPEND; 312#ifdef MAC 313 error = mac_vnode_check_open(cred, vp, accmode); 314 if (error) 315 return (error); 316#endif 317 if ((fmode & O_CREAT) == 0) { 318 if (accmode & VWRITE) { 319 error = vn_writechk(vp); 320 if (error) 321 return (error); 322 } 323 if (accmode) { 324 error = VOP_ACCESS(vp, accmode, cred, td); 325 if (error) 326 return (error); 327 } 328 } 329 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) 330 vn_lock(vp, LK_UPGRADE | LK_RETRY); 331 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0) 332 return (error); 333 334 if (fmode & (O_EXLOCK | O_SHLOCK)) { 335 KASSERT(fp != NULL, ("open with flock requires fp")); 336 lock_flags = VOP_ISLOCKED(vp); 337 VOP_UNLOCK(vp, 0); 338 lf.l_whence = SEEK_SET; 339 lf.l_start = 0; 340 lf.l_len = 0; 341 if (fmode & O_EXLOCK) 342 lf.l_type = F_WRLCK; 343 else 344 lf.l_type = F_RDLCK; 345 type = F_FLOCK; 346 if ((fmode & FNONBLOCK) == 0) 347 type |= F_WAIT; 348 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type); 349 have_flock = (error == 0); 350 vn_lock(vp, lock_flags | LK_RETRY); 351 if (error == 0 && vp->v_iflag & VI_DOOMED) 352 error = ENOENT; 353 /* 354 * Another thread might have used this vnode as an 355 * executable while the vnode lock was dropped. 356 * Ensure the vnode is still able to be opened for 357 * writing after the lock has been obtained. 358 */ 359 if (error == 0 && accmode & VWRITE) 360 error = vn_writechk(vp); 361 if (error) { 362 VOP_UNLOCK(vp, 0); 363 if (have_flock) { 364 lf.l_whence = SEEK_SET; 365 lf.l_start = 0; 366 lf.l_len = 0; 367 lf.l_type = F_UNLCK; 368 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, 369 F_FLOCK); 370 } 371 vn_start_write(vp, &mp, V_WAIT); 372 vn_lock(vp, lock_flags | LK_RETRY); 373 (void)VOP_CLOSE(vp, fmode, cred, td); 374 vn_finished_write(mp); 375 /* Prevent second close from fdrop()->vn_close(). */ 376 if (fp != NULL) 377 fp->f_ops= &badfileops; 378 return (error); 379 } 380 fp->f_flag |= FHASLOCK; 381 } 382 if (fmode & FWRITE) { 383 VOP_ADD_WRITECOUNT(vp, 1); 384 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", 385 __func__, vp, vp->v_writecount); 386 } 387 ASSERT_VOP_LOCKED(vp, "vn_open_vnode"); 388 return (0); 389} 390 391/* 392 * Check for write permissions on the specified vnode. 393 * Prototype text segments cannot be written. 394 */ 395int 396vn_writechk(vp) 397 register struct vnode *vp; 398{ 399 400 ASSERT_VOP_LOCKED(vp, "vn_writechk"); 401 /* 402 * If there's shared text associated with 403 * the vnode, try to free it up once. If 404 * we fail, we can't allow writing. 405 */ 406 if (VOP_IS_TEXT(vp)) 407 return (ETXTBSY); 408 409 return (0); 410} 411 412/* 413 * Vnode close call 414 */ 415static int 416vn_close1(struct vnode *vp, int flags, struct ucred *file_cred, 417 struct thread *td, bool keep_ref) 418{ 419 struct mount *mp; 420 int error, lock_flags; 421 422 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 && 423 MNT_EXTENDED_SHARED(vp->v_mount)) 424 lock_flags = LK_SHARED; 425 else 426 lock_flags = LK_EXCLUSIVE; 427 428 vn_start_write(vp, &mp, V_WAIT); 429 vn_lock(vp, lock_flags | LK_RETRY); 430 if (flags & FWRITE) { 431 VNASSERT(vp->v_writecount > 0, vp, 432 ("vn_close: negative writecount")); 433 VOP_ADD_WRITECOUNT(vp, -1); 434 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", 435 __func__, vp, vp->v_writecount); 436 } 437 error = VOP_CLOSE(vp, flags, file_cred, td); 438 if (keep_ref) 439 VOP_UNLOCK(vp, 0); 440 else 441 vput(vp); 442 vn_finished_write(mp); 443 return (error); 444} 445 446int 447vn_close(struct vnode *vp, int flags, struct ucred *file_cred, 448 struct thread *td) 449{ 450 451 return (vn_close1(vp, flags, file_cred, td, false)); 452} 453 454/* 455 * Heuristic to detect sequential operation. 456 */ 457static int 458sequential_heuristic(struct uio *uio, struct file *fp) 459{ 460 461 ASSERT_VOP_LOCKED(fp->f_vnode, __func__); 462 if (fp->f_flag & FRDAHEAD) 463 return (fp->f_seqcount << IO_SEQSHIFT); 464 465 /* 466 * Offset 0 is handled specially. open() sets f_seqcount to 1 so 467 * that the first I/O is normally considered to be slightly 468 * sequential. Seeking to offset 0 doesn't change sequentiality 469 * unless previous seeks have reduced f_seqcount to 0, in which 470 * case offset 0 is not special. 471 */ 472 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 473 uio->uio_offset == fp->f_nextoff) { 474 /* 475 * f_seqcount is in units of fixed-size blocks so that it 476 * depends mainly on the amount of sequential I/O and not 477 * much on the number of sequential I/O's. The fixed size 478 * of 16384 is hard-coded here since it is (not quite) just 479 * a magic size that works well here. This size is more 480 * closely related to the best I/O size for real disks than 481 * to any block size used by software. 482 */ 483 fp->f_seqcount += howmany(uio->uio_resid, 16384); 484 if (fp->f_seqcount > IO_SEQMAX) 485 fp->f_seqcount = IO_SEQMAX; 486 return (fp->f_seqcount << IO_SEQSHIFT); 487 } 488 489 /* Not sequential. Quickly draw-down sequentiality. */ 490 if (fp->f_seqcount > 1) 491 fp->f_seqcount = 1; 492 else 493 fp->f_seqcount = 0; 494 return (0); 495} 496 497/* 498 * Package up an I/O request on a vnode into a uio and do it. 499 */ 500int 501vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, 502 enum uio_seg segflg, int ioflg, struct ucred *active_cred, 503 struct ucred *file_cred, ssize_t *aresid, struct thread *td) 504{ 505 struct uio auio; 506 struct iovec aiov; 507 struct mount *mp; 508 struct ucred *cred; 509 void *rl_cookie; 510 struct vn_io_fault_args args; 511 int error, lock_flags; 512 513 if (offset < 0 && vp->v_type != VCHR) 514 return (EINVAL); 515 auio.uio_iov = &aiov; 516 auio.uio_iovcnt = 1; 517 aiov.iov_base = base; 518 aiov.iov_len = len; 519 auio.uio_resid = len; 520 auio.uio_offset = offset; 521 auio.uio_segflg = segflg; 522 auio.uio_rw = rw; 523 auio.uio_td = td; 524 error = 0; 525 526 if ((ioflg & IO_NODELOCKED) == 0) { 527 if ((ioflg & IO_RANGELOCKED) == 0) { 528 if (rw == UIO_READ) { 529 rl_cookie = vn_rangelock_rlock(vp, offset, 530 offset + len); 531 } else { 532 rl_cookie = vn_rangelock_wlock(vp, offset, 533 offset + len); 534 } 535 } else 536 rl_cookie = NULL; 537 mp = NULL; 538 if (rw == UIO_WRITE) { 539 if (vp->v_type != VCHR && 540 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) 541 != 0) 542 goto out; 543 if (MNT_SHARED_WRITES(mp) || 544 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) 545 lock_flags = LK_SHARED; 546 else 547 lock_flags = LK_EXCLUSIVE; 548 } else 549 lock_flags = LK_SHARED; 550 vn_lock(vp, lock_flags | LK_RETRY); 551 } else 552 rl_cookie = NULL; 553 554 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 555#ifdef MAC 556 if ((ioflg & IO_NOMACCHECK) == 0) { 557 if (rw == UIO_READ) 558 error = mac_vnode_check_read(active_cred, file_cred, 559 vp); 560 else 561 error = mac_vnode_check_write(active_cred, file_cred, 562 vp); 563 } 564#endif 565 if (error == 0) { 566 if (file_cred != NULL) 567 cred = file_cred; 568 else 569 cred = active_cred; 570 if (do_vn_io_fault(vp, &auio)) { 571 args.kind = VN_IO_FAULT_VOP; 572 args.cred = cred; 573 args.flags = ioflg; 574 args.args.vop_args.vp = vp; 575 error = vn_io_fault1(vp, &auio, &args, td); 576 } else if (rw == UIO_READ) { 577 error = VOP_READ(vp, &auio, ioflg, cred); 578 } else /* if (rw == UIO_WRITE) */ { 579 error = VOP_WRITE(vp, &auio, ioflg, cred); 580 } 581 } 582 if (aresid) 583 *aresid = auio.uio_resid; 584 else 585 if (auio.uio_resid && error == 0) 586 error = EIO; 587 if ((ioflg & IO_NODELOCKED) == 0) { 588 VOP_UNLOCK(vp, 0); 589 if (mp != NULL) 590 vn_finished_write(mp); 591 } 592 out: 593 if (rl_cookie != NULL) 594 vn_rangelock_unlock(vp, rl_cookie); 595 return (error); 596} 597 598/* 599 * Package up an I/O request on a vnode into a uio and do it. The I/O 600 * request is split up into smaller chunks and we try to avoid saturating 601 * the buffer cache while potentially holding a vnode locked, so we 602 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield() 603 * to give other processes a chance to lock the vnode (either other processes 604 * core'ing the same binary, or unrelated processes scanning the directory). 605 */ 606int 607vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred, 608 file_cred, aresid, td) 609 enum uio_rw rw; 610 struct vnode *vp; 611 void *base; 612 size_t len; 613 off_t offset; 614 enum uio_seg segflg; 615 int ioflg; 616 struct ucred *active_cred; 617 struct ucred *file_cred; 618 size_t *aresid; 619 struct thread *td; 620{ 621 int error = 0; 622 ssize_t iaresid; 623 624 do { 625 int chunk; 626 627 /* 628 * Force `offset' to a multiple of MAXBSIZE except possibly 629 * for the first chunk, so that filesystems only need to 630 * write full blocks except possibly for the first and last 631 * chunks. 632 */ 633 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; 634 635 if (chunk > len) 636 chunk = len; 637 if (rw != UIO_READ && vp->v_type == VREG) 638 bwillwrite(); 639 iaresid = 0; 640 error = vn_rdwr(rw, vp, base, chunk, offset, segflg, 641 ioflg, active_cred, file_cred, &iaresid, td); 642 len -= chunk; /* aresid calc already includes length */ 643 if (error) 644 break; 645 offset += chunk; 646 base = (char *)base + chunk; 647 kern_yield(PRI_USER); 648 } while (len); 649 if (aresid) 650 *aresid = len + iaresid; 651 return (error); 652} 653 654off_t 655foffset_lock(struct file *fp, int flags) 656{ 657 struct mtx *mtxp; 658 off_t res; 659 660 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); 661 662#if OFF_MAX <= LONG_MAX 663 /* 664 * Caller only wants the current f_offset value. Assume that 665 * the long and shorter integer types reads are atomic. 666 */ 667 if ((flags & FOF_NOLOCK) != 0) 668 return (fp->f_offset); 669#endif 670 671 /* 672 * According to McKusick the vn lock was protecting f_offset here. 673 * It is now protected by the FOFFSET_LOCKED flag. 674 */ 675 mtxp = mtx_pool_find(mtxpool_sleep, fp); 676 mtx_lock(mtxp); 677 if ((flags & FOF_NOLOCK) == 0) { 678 while (fp->f_vnread_flags & FOFFSET_LOCKED) { 679 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING; 680 msleep(&fp->f_vnread_flags, mtxp, PUSER -1, 681 "vofflock", 0); 682 } 683 fp->f_vnread_flags |= FOFFSET_LOCKED; 684 } 685 res = fp->f_offset; 686 mtx_unlock(mtxp); 687 return (res); 688} 689 690void 691foffset_unlock(struct file *fp, off_t val, int flags) 692{ 693 struct mtx *mtxp; 694 695 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); 696 697#if OFF_MAX <= LONG_MAX 698 if ((flags & FOF_NOLOCK) != 0) { 699 if ((flags & FOF_NOUPDATE) == 0) 700 fp->f_offset = val; 701 if ((flags & FOF_NEXTOFF) != 0) 702 fp->f_nextoff = val; 703 return; 704 } 705#endif 706 707 mtxp = mtx_pool_find(mtxpool_sleep, fp); 708 mtx_lock(mtxp); 709 if ((flags & FOF_NOUPDATE) == 0) 710 fp->f_offset = val; 711 if ((flags & FOF_NEXTOFF) != 0) 712 fp->f_nextoff = val; 713 if ((flags & FOF_NOLOCK) == 0) { 714 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0, 715 ("Lost FOFFSET_LOCKED")); 716 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING) 717 wakeup(&fp->f_vnread_flags); 718 fp->f_vnread_flags = 0; 719 } 720 mtx_unlock(mtxp); 721} 722 723void 724foffset_lock_uio(struct file *fp, struct uio *uio, int flags) 725{ 726 727 if ((flags & FOF_OFFSET) == 0) 728 uio->uio_offset = foffset_lock(fp, flags); 729} 730 731void 732foffset_unlock_uio(struct file *fp, struct uio *uio, int flags) 733{ 734 735 if ((flags & FOF_OFFSET) == 0) 736 foffset_unlock(fp, uio->uio_offset, flags); 737} 738 739static int 740get_advice(struct file *fp, struct uio *uio) 741{ 742 struct mtx *mtxp; 743 int ret; 744 745 ret = POSIX_FADV_NORMAL; 746 if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG) 747 return (ret); 748 749 mtxp = mtx_pool_find(mtxpool_sleep, fp); 750 mtx_lock(mtxp); 751 if (fp->f_advice != NULL && 752 uio->uio_offset >= fp->f_advice->fa_start && 753 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end) 754 ret = fp->f_advice->fa_advice; 755 mtx_unlock(mtxp); 756 return (ret); 757} 758 759/* 760 * File table vnode read routine. 761 */ 762static int 763vn_read(fp, uio, active_cred, flags, td) 764 struct file *fp; 765 struct uio *uio; 766 struct ucred *active_cred; 767 int flags; 768 struct thread *td; 769{ 770 struct vnode *vp; 771 struct mtx *mtxp; 772 int error, ioflag; 773 int advice; 774 off_t offset, start, end; 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 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 offset != uio->uio_offset) { 809 /* 810 * Use POSIX_FADV_DONTNEED to flush clean pages and 811 * buffers for the backing file after a 812 * POSIX_FADV_NOREUSE read(2). To optimize the common 813 * case of using POSIX_FADV_NOREUSE with sequential 814 * access, track the previous implicit DONTNEED 815 * request and grow this request to include the 816 * current read(2) in addition to the previous 817 * DONTNEED. With purely sequential access this will 818 * cause the DONTNEED requests to continously grow to 819 * cover all of the previously read regions of the 820 * file. This allows filesystem blocks that are 821 * accessed by multiple calls to read(2) to be flushed 822 * once the last read(2) finishes. 823 */ 824 start = offset; 825 end = uio->uio_offset - 1; 826 mtxp = mtx_pool_find(mtxpool_sleep, fp); 827 mtx_lock(mtxp); 828 if (fp->f_advice != NULL && 829 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) { 830 if (start != 0 && fp->f_advice->fa_prevend + 1 == start) 831 start = fp->f_advice->fa_prevstart; 832 else if (fp->f_advice->fa_prevstart != 0 && 833 fp->f_advice->fa_prevstart == end + 1) 834 end = fp->f_advice->fa_prevend; 835 fp->f_advice->fa_prevstart = start; 836 fp->f_advice->fa_prevend = end; 837 } 838 mtx_unlock(mtxp); 839 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED); 840 } 841 return (error); 842} 843 844/* 845 * File table vnode write routine. 846 */ 847static int 848vn_write(fp, uio, active_cred, flags, td) 849 struct file *fp; 850 struct uio *uio; 851 struct ucred *active_cred; 852 int flags; 853 struct thread *td; 854{ 855 struct vnode *vp; 856 struct mount *mp; 857 struct mtx *mtxp; 858 int error, ioflag, lock_flags; 859 int advice; 860 off_t offset, start, end; 861 862 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", 863 uio->uio_td, td)); 864 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); 865 vp = fp->f_vnode; 866 if (vp->v_type == VREG) 867 bwillwrite(); 868 ioflag = IO_UNIT; 869 if (vp->v_type == VREG && (fp->f_flag & O_APPEND)) 870 ioflag |= IO_APPEND; 871 if (fp->f_flag & FNONBLOCK) 872 ioflag |= IO_NDELAY; 873 if (fp->f_flag & O_DIRECT) 874 ioflag |= IO_DIRECT; 875 if ((fp->f_flag & O_FSYNC) || 876 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) 877 ioflag |= IO_SYNC; 878 mp = NULL; 879 if (vp->v_type != VCHR && 880 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) 881 goto unlock; 882 883 advice = get_advice(fp, uio); 884 885 if (MNT_SHARED_WRITES(mp) || 886 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) { 887 lock_flags = LK_SHARED; 888 } else { 889 lock_flags = LK_EXCLUSIVE; 890 } 891 892 vn_lock(vp, lock_flags | LK_RETRY); 893 switch (advice) { 894 case POSIX_FADV_NORMAL: 895 case POSIX_FADV_SEQUENTIAL: 896 case POSIX_FADV_NOREUSE: 897 ioflag |= sequential_heuristic(uio, fp); 898 break; 899 case POSIX_FADV_RANDOM: 900 /* XXX: Is this correct? */ 901 break; 902 } 903 offset = uio->uio_offset; 904 905#ifdef MAC 906 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 907 if (error == 0) 908#endif 909 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred); 910 fp->f_nextoff = uio->uio_offset; 911 VOP_UNLOCK(vp, 0); 912 if (vp->v_type != VCHR) 913 vn_finished_write(mp); 914 if (error == 0 && advice == POSIX_FADV_NOREUSE && 915 offset != uio->uio_offset) { 916 /* 917 * Use POSIX_FADV_DONTNEED to flush clean pages and 918 * buffers for the backing file after a 919 * POSIX_FADV_NOREUSE write(2). To optimize the 920 * common case of using POSIX_FADV_NOREUSE with 921 * sequential access, track the previous implicit 922 * DONTNEED request and grow this request to include 923 * the current write(2) in addition to the previous 924 * DONTNEED. With purely sequential access this will 925 * cause the DONTNEED requests to continously grow to 926 * cover all of the previously written regions of the 927 * file. 928 * 929 * Note that the blocks just written are almost 930 * certainly still dirty, so this only works when 931 * VOP_ADVISE() calls from subsequent writes push out 932 * the data written by this write(2) once the backing 933 * buffers are clean. However, as compared to forcing 934 * IO_DIRECT, this gives much saner behavior. Write 935 * clustering is still allowed, and clean pages are 936 * merely moved to the cache page queue rather than 937 * outright thrown away. This means a subsequent 938 * read(2) can still avoid hitting the disk if the 939 * pages have not been reclaimed. 940 * 941 * This does make POSIX_FADV_NOREUSE largely useless 942 * with non-sequential access. However, sequential 943 * access is the more common use case and the flag is 944 * merely advisory. 945 */ 946 start = offset; 947 end = uio->uio_offset - 1; 948 mtxp = mtx_pool_find(mtxpool_sleep, fp); 949 mtx_lock(mtxp); 950 if (fp->f_advice != NULL && 951 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) { 952 if (start != 0 && fp->f_advice->fa_prevend + 1 == start) 953 start = fp->f_advice->fa_prevstart; 954 else if (fp->f_advice->fa_prevstart != 0 && 955 fp->f_advice->fa_prevstart == end + 1) 956 end = fp->f_advice->fa_prevend; 957 fp->f_advice->fa_prevstart = start; 958 fp->f_advice->fa_prevend = end; 959 } 960 mtx_unlock(mtxp); 961 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED); 962 } 963 964unlock: 965 return (error); 966} 967 968/* 969 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to 970 * prevent the following deadlock: 971 * 972 * Assume that the thread A reads from the vnode vp1 into userspace 973 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is 974 * currently not resident, then system ends up with the call chain 975 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] -> 976 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2) 977 * which establishes lock order vp1->vn_lock, then vp2->vn_lock. 978 * If, at the same time, thread B reads from vnode vp2 into buffer buf2 979 * backed by the pages of vnode vp1, and some page in buf2 is not 980 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock. 981 * 982 * To prevent the lock order reversal and deadlock, vn_io_fault() does 983 * not allow page faults to happen during VOP_READ() or VOP_WRITE(). 984 * Instead, it first tries to do the whole range i/o with pagefaults 985 * disabled. If all pages in the i/o buffer are resident and mapped, 986 * VOP will succeed (ignoring the genuine filesystem errors). 987 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do 988 * i/o in chunks, with all pages in the chunk prefaulted and held 989 * using vm_fault_quick_hold_pages(). 990 * 991 * Filesystems using this deadlock avoidance scheme should use the 992 * array of the held pages from uio, saved in the curthread->td_ma, 993 * instead of doing uiomove(). A helper function 994 * vn_io_fault_uiomove() converts uiomove request into 995 * uiomove_fromphys() over td_ma array. 996 * 997 * Since vnode locks do not cover the whole i/o anymore, rangelocks 998 * make the current i/o request atomic with respect to other i/os and 999 * truncations. 1000 */ 1001 1002/* 1003 * Decode vn_io_fault_args and perform the corresponding i/o. 1004 */ 1005static int 1006vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio, 1007 struct thread *td) 1008{ 1009 1010 switch (args->kind) { 1011 case VN_IO_FAULT_FOP: 1012 return ((args->args.fop_args.doio)(args->args.fop_args.fp, 1013 uio, args->cred, args->flags, td)); 1014 case VN_IO_FAULT_VOP: 1015 if (uio->uio_rw == UIO_READ) { 1016 return (VOP_READ(args->args.vop_args.vp, uio, 1017 args->flags, args->cred)); 1018 } else if (uio->uio_rw == UIO_WRITE) { 1019 return (VOP_WRITE(args->args.vop_args.vp, uio, 1020 args->flags, args->cred)); 1021 } 1022 break; 1023 } 1024 panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind, 1025 uio->uio_rw); 1026} 1027 1028static int 1029vn_io_fault_touch(char *base, const struct uio *uio) 1030{ 1031 int r; 1032 1033 r = fubyte(base); 1034 if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1)) 1035 return (EFAULT); 1036 return (0); 1037} 1038 1039static int 1040vn_io_fault_prefault_user(const struct uio *uio) 1041{ 1042 char *base; 1043 const struct iovec *iov; 1044 size_t len; 1045 ssize_t resid; 1046 int error, i; 1047 1048 KASSERT(uio->uio_segflg == UIO_USERSPACE, 1049 ("vn_io_fault_prefault userspace")); 1050 1051 error = i = 0; 1052 iov = uio->uio_iov; 1053 resid = uio->uio_resid; 1054 base = iov->iov_base; 1055 len = iov->iov_len; 1056 while (resid > 0) { 1057 error = vn_io_fault_touch(base, uio); 1058 if (error != 0) 1059 break; 1060 if (len < PAGE_SIZE) { 1061 if (len != 0) { 1062 error = vn_io_fault_touch(base + len - 1, uio); 1063 if (error != 0) 1064 break; 1065 resid -= len; 1066 } 1067 if (++i >= uio->uio_iovcnt) 1068 break; 1069 iov = uio->uio_iov + i; 1070 base = iov->iov_base; 1071 len = iov->iov_len; 1072 } else { 1073 len -= PAGE_SIZE; 1074 base += PAGE_SIZE; 1075 resid -= PAGE_SIZE; 1076 } 1077 } 1078 return (error); 1079} 1080 1081/* 1082 * Common code for vn_io_fault(), agnostic to the kind of i/o request. 1083 * Uses vn_io_fault_doio() to make the call to an actual i/o function. 1084 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request 1085 * into args and call vn_io_fault1() to handle faults during the user 1086 * mode buffer accesses. 1087 */ 1088static int 1089vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, 1090 struct thread *td) 1091{ 1092 vm_page_t ma[io_hold_cnt + 2]; 1093 struct uio *uio_clone, short_uio; 1094 struct iovec short_iovec[1]; 1095 vm_page_t *prev_td_ma; 1096 vm_prot_t prot; 1097 vm_offset_t addr, end; 1098 size_t len, resid; 1099 ssize_t adv; 1100 int error, cnt, save, saveheld, prev_td_ma_cnt; 1101 1102 if (vn_io_fault_prefault) { 1103 error = vn_io_fault_prefault_user(uio); 1104 if (error != 0) 1105 return (error); /* Or ignore ? */ 1106 } 1107 1108 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ; 1109 1110 /* 1111 * The UFS follows IO_UNIT directive and replays back both 1112 * uio_offset and uio_resid if an error is encountered during the 1113 * operation. But, since the iovec may be already advanced, 1114 * uio is still in an inconsistent state. 1115 * 1116 * Cache a copy of the original uio, which is advanced to the redo 1117 * point using UIO_NOCOPY below. 1118 */ 1119 uio_clone = cloneuio(uio); 1120 resid = uio->uio_resid; 1121 1122 short_uio.uio_segflg = UIO_USERSPACE; 1123 short_uio.uio_rw = uio->uio_rw; 1124 short_uio.uio_td = uio->uio_td; 1125 1126 save = vm_fault_disable_pagefaults(); 1127 error = vn_io_fault_doio(args, uio, td); 1128 if (error != EFAULT) 1129 goto out; 1130 1131 atomic_add_long(&vn_io_faults_cnt, 1); 1132 uio_clone->uio_segflg = UIO_NOCOPY; 1133 uiomove(NULL, resid - uio->uio_resid, uio_clone); 1134 uio_clone->uio_segflg = uio->uio_segflg; 1135 1136 saveheld = curthread_pflags_set(TDP_UIOHELD); 1137 prev_td_ma = td->td_ma; 1138 prev_td_ma_cnt = td->td_ma_cnt; 1139 1140 while (uio_clone->uio_resid != 0) { 1141 len = uio_clone->uio_iov->iov_len; 1142 if (len == 0) { 1143 KASSERT(uio_clone->uio_iovcnt >= 1, 1144 ("iovcnt underflow")); 1145 uio_clone->uio_iov++; 1146 uio_clone->uio_iovcnt--; 1147 continue; 1148 } 1149 if (len > io_hold_cnt * PAGE_SIZE) 1150 len = io_hold_cnt * PAGE_SIZE; 1151 addr = (uintptr_t)uio_clone->uio_iov->iov_base; 1152 end = round_page(addr + len); 1153 if (end < addr) { 1154 error = EFAULT; 1155 break; 1156 } 1157 cnt = atop(end - trunc_page(addr)); 1158 /* 1159 * A perfectly misaligned address and length could cause 1160 * both the start and the end of the chunk to use partial 1161 * page. +2 accounts for such a situation. 1162 */ 1163 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map, 1164 addr, len, prot, ma, io_hold_cnt + 2); 1165 if (cnt == -1) { 1166 error = EFAULT; 1167 break; 1168 } 1169 short_uio.uio_iov = &short_iovec[0]; 1170 short_iovec[0].iov_base = (void *)addr; 1171 short_uio.uio_iovcnt = 1; 1172 short_uio.uio_resid = short_iovec[0].iov_len = len; 1173 short_uio.uio_offset = uio_clone->uio_offset; 1174 td->td_ma = ma; 1175 td->td_ma_cnt = cnt; 1176 1177 error = vn_io_fault_doio(args, &short_uio, td); 1178 vm_page_unhold_pages(ma, cnt); 1179 adv = len - short_uio.uio_resid; 1180 1181 uio_clone->uio_iov->iov_base = 1182 (char *)uio_clone->uio_iov->iov_base + adv; 1183 uio_clone->uio_iov->iov_len -= adv; 1184 uio_clone->uio_resid -= adv; 1185 uio_clone->uio_offset += adv; 1186 1187 uio->uio_resid -= adv; 1188 uio->uio_offset += adv; 1189 1190 if (error != 0 || adv == 0) 1191 break; 1192 } 1193 td->td_ma = prev_td_ma; 1194 td->td_ma_cnt = prev_td_ma_cnt; 1195 curthread_pflags_restore(saveheld); 1196out: 1197 vm_fault_enable_pagefaults(save); 1198 free(uio_clone, M_IOV); 1199 return (error); 1200} 1201 1202static int 1203vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred, 1204 int flags, struct thread *td) 1205{ 1206 fo_rdwr_t *doio; 1207 struct vnode *vp; 1208 void *rl_cookie; 1209 struct vn_io_fault_args args; 1210 int error; 1211 1212 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write; 1213 vp = fp->f_vnode; 1214 foffset_lock_uio(fp, uio, flags); 1215 if (do_vn_io_fault(vp, uio)) { 1216 args.kind = VN_IO_FAULT_FOP; 1217 args.args.fop_args.fp = fp; 1218 args.args.fop_args.doio = doio; 1219 args.cred = active_cred; 1220 args.flags = flags | FOF_OFFSET; 1221 if (uio->uio_rw == UIO_READ) { 1222 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset, 1223 uio->uio_offset + uio->uio_resid); 1224 } else if ((fp->f_flag & O_APPEND) != 0 || 1225 (flags & FOF_OFFSET) == 0) { 1226 /* For appenders, punt and lock the whole range. */ 1227 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); 1228 } else { 1229 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset, 1230 uio->uio_offset + uio->uio_resid); 1231 } 1232 error = vn_io_fault1(vp, uio, &args, td); 1233 vn_rangelock_unlock(vp, rl_cookie); 1234 } else { 1235 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td); 1236 } 1237 foffset_unlock_uio(fp, uio, flags); 1238 return (error); 1239} 1240 1241/* 1242 * Helper function to perform the requested uiomove operation using 1243 * the held pages for io->uio_iov[0].iov_base buffer instead of 1244 * copyin/copyout. Access to the pages with uiomove_fromphys() 1245 * instead of iov_base prevents page faults that could occur due to 1246 * pmap_collect() invalidating the mapping created by 1247 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or 1248 * object cleanup revoking the write access from page mappings. 1249 * 1250 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove() 1251 * instead of plain uiomove(). 1252 */ 1253int 1254vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio) 1255{ 1256 struct uio transp_uio; 1257 struct iovec transp_iov[1]; 1258 struct thread *td; 1259 size_t adv; 1260 int error, pgadv; 1261 1262 td = curthread; 1263 if ((td->td_pflags & TDP_UIOHELD) == 0 || 1264 uio->uio_segflg != UIO_USERSPACE) 1265 return (uiomove(data, xfersize, uio)); 1266 1267 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); 1268 transp_iov[0].iov_base = data; 1269 transp_uio.uio_iov = &transp_iov[0]; 1270 transp_uio.uio_iovcnt = 1; 1271 if (xfersize > uio->uio_resid) 1272 xfersize = uio->uio_resid; 1273 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize; 1274 transp_uio.uio_offset = 0; 1275 transp_uio.uio_segflg = UIO_SYSSPACE; 1276 /* 1277 * Since transp_iov points to data, and td_ma page array 1278 * corresponds to original uio->uio_iov, we need to invert the 1279 * direction of the i/o operation as passed to 1280 * uiomove_fromphys(). 1281 */ 1282 switch (uio->uio_rw) { 1283 case UIO_WRITE: 1284 transp_uio.uio_rw = UIO_READ; 1285 break; 1286 case UIO_READ: 1287 transp_uio.uio_rw = UIO_WRITE; 1288 break; 1289 } 1290 transp_uio.uio_td = uio->uio_td; 1291 error = uiomove_fromphys(td->td_ma, 1292 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK, 1293 xfersize, &transp_uio); 1294 adv = xfersize - transp_uio.uio_resid; 1295 pgadv = 1296 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) - 1297 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT); 1298 td->td_ma += pgadv; 1299 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, 1300 pgadv)); 1301 td->td_ma_cnt -= pgadv; 1302 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv; 1303 uio->uio_iov->iov_len -= adv; 1304 uio->uio_resid -= adv; 1305 uio->uio_offset += adv; 1306 return (error); 1307} 1308 1309int 1310vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize, 1311 struct uio *uio) 1312{ 1313 struct thread *td; 1314 vm_offset_t iov_base; 1315 int cnt, pgadv; 1316 1317 td = curthread; 1318 if ((td->td_pflags & TDP_UIOHELD) == 0 || 1319 uio->uio_segflg != UIO_USERSPACE) 1320 return (uiomove_fromphys(ma, offset, xfersize, uio)); 1321 1322 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); 1323 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize; 1324 iov_base = (vm_offset_t)uio->uio_iov->iov_base; 1325 switch (uio->uio_rw) { 1326 case UIO_WRITE: 1327 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma, 1328 offset, cnt); 1329 break; 1330 case UIO_READ: 1331 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK, 1332 cnt); 1333 break; 1334 } 1335 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT); 1336 td->td_ma += pgadv; 1337 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, 1338 pgadv)); 1339 td->td_ma_cnt -= pgadv; 1340 uio->uio_iov->iov_base = (char *)(iov_base + cnt); 1341 uio->uio_iov->iov_len -= cnt; 1342 uio->uio_resid -= cnt; 1343 uio->uio_offset += cnt; 1344 return (0); 1345} 1346 1347 1348/* 1349 * File table truncate routine. 1350 */ 1351static int 1352vn_truncate(struct file *fp, off_t length, struct ucred *active_cred, 1353 struct thread *td) 1354{ 1355 struct vattr vattr; 1356 struct mount *mp; 1357 struct vnode *vp; 1358 void *rl_cookie; 1359 int error; 1360 1361 vp = fp->f_vnode; 1362 1363 /* 1364 * Lock the whole range for truncation. Otherwise split i/o 1365 * might happen partly before and partly after the truncation. 1366 */ 1367 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); 1368 error = vn_start_write(vp, &mp, V_WAIT | PCATCH); 1369 if (error) 1370 goto out1; 1371 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1372 if (vp->v_type == VDIR) { 1373 error = EISDIR; 1374 goto out; 1375 } 1376#ifdef MAC 1377 error = mac_vnode_check_write(active_cred, fp->f_cred, vp); 1378 if (error) 1379 goto out; 1380#endif 1381 error = vn_writechk(vp); 1382 if (error == 0) { 1383 VATTR_NULL(&vattr); 1384 vattr.va_size = length; 1385 if ((fp->f_flag & O_FSYNC) != 0) 1386 vattr.va_vaflags |= VA_SYNC; 1387 error = VOP_SETATTR(vp, &vattr, fp->f_cred); 1388 } 1389out: 1390 VOP_UNLOCK(vp, 0); 1391 vn_finished_write(mp); 1392out1: 1393 vn_rangelock_unlock(vp, rl_cookie); 1394 return (error); 1395} 1396 1397/* 1398 * File table vnode stat routine. 1399 */ 1400static int 1401vn_statfile(fp, sb, active_cred, td) 1402 struct file *fp; 1403 struct stat *sb; 1404 struct ucred *active_cred; 1405 struct thread *td; 1406{ 1407 struct vnode *vp = fp->f_vnode; 1408 int error; 1409 1410 vn_lock(vp, LK_SHARED | LK_RETRY); 1411 error = vn_stat(vp, sb, active_cred, fp->f_cred, td); 1412 VOP_UNLOCK(vp, 0); 1413 1414 return (error); 1415} 1416 1417/* 1418 * Stat a vnode; implementation for the stat syscall 1419 */ 1420int 1421vn_stat(vp, sb, active_cred, file_cred, td) 1422 struct vnode *vp; 1423 register struct stat *sb; 1424 struct ucred *active_cred; 1425 struct ucred *file_cred; 1426 struct thread *td; 1427{ 1428 struct vattr vattr; 1429 register struct vattr *vap; 1430 int error; 1431 u_short mode; 1432 1433#ifdef MAC 1434 error = mac_vnode_check_stat(active_cred, file_cred, vp); 1435 if (error) 1436 return (error); 1437#endif 1438 1439 vap = &vattr; 1440 1441 /* 1442 * Initialize defaults for new and unusual fields, so that file 1443 * systems which don't support these fields don't need to know 1444 * about them. 1445 */ 1446 vap->va_birthtime.tv_sec = -1; 1447 vap->va_birthtime.tv_nsec = 0; 1448 vap->va_fsid = VNOVAL; 1449 vap->va_rdev = NODEV; 1450 1451 error = VOP_GETATTR(vp, vap, active_cred); 1452 if (error) 1453 return (error); 1454 1455 /* 1456 * Zero the spare stat fields 1457 */ 1458 bzero(sb, sizeof *sb); 1459 1460 /* 1461 * Copy from vattr table 1462 */ 1463 if (vap->va_fsid != VNOVAL) 1464 sb->st_dev = vap->va_fsid; 1465 else 1466 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 1467 sb->st_ino = vap->va_fileid; 1468 mode = vap->va_mode; 1469 switch (vap->va_type) { 1470 case VREG: 1471 mode |= S_IFREG; 1472 break; 1473 case VDIR: 1474 mode |= S_IFDIR; 1475 break; 1476 case VBLK: 1477 mode |= S_IFBLK; 1478 break; 1479 case VCHR: 1480 mode |= S_IFCHR; 1481 break; 1482 case VLNK: 1483 mode |= S_IFLNK; 1484 break; 1485 case VSOCK: 1486 mode |= S_IFSOCK; 1487 break; 1488 case VFIFO: 1489 mode |= S_IFIFO; 1490 break; 1491 default: 1492 return (EBADF); 1493 }; 1494 sb->st_mode = mode; 1495 sb->st_nlink = vap->va_nlink; 1496 sb->st_uid = vap->va_uid; 1497 sb->st_gid = vap->va_gid; 1498 sb->st_rdev = vap->va_rdev; 1499 if (vap->va_size > OFF_MAX) 1500 return (EOVERFLOW); 1501 sb->st_size = vap->va_size; 1502 sb->st_atim = vap->va_atime; 1503 sb->st_mtim = vap->va_mtime; 1504 sb->st_ctim = vap->va_ctime; 1505 sb->st_birthtim = vap->va_birthtime; 1506 1507 /* 1508 * According to www.opengroup.org, the meaning of st_blksize is 1509 * "a filesystem-specific preferred I/O block size for this 1510 * object. In some filesystem types, this may vary from file 1511 * to file" 1512 * Use miminum/default of PAGE_SIZE (e.g. for VCHR). 1513 */ 1514 1515 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize); 1516 1517 sb->st_flags = vap->va_flags; 1518 if (priv_check(td, PRIV_VFS_GENERATION)) 1519 sb->st_gen = 0; 1520 else 1521 sb->st_gen = vap->va_gen; 1522 1523 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 1524 return (0); 1525} 1526 1527/* 1528 * File table vnode ioctl routine. 1529 */ 1530static int 1531vn_ioctl(fp, com, data, active_cred, td) 1532 struct file *fp; 1533 u_long com; 1534 void *data; 1535 struct ucred *active_cred; 1536 struct thread *td; 1537{ 1538 struct vattr vattr; 1539 struct vnode *vp; 1540 int error; 1541 1542 vp = fp->f_vnode; 1543 switch (vp->v_type) { 1544 case VDIR: 1545 case VREG: 1546 switch (com) { 1547 case FIONREAD: 1548 vn_lock(vp, LK_SHARED | LK_RETRY); 1549 error = VOP_GETATTR(vp, &vattr, active_cred); 1550 VOP_UNLOCK(vp, 0); 1551 if (error == 0) 1552 *(int *)data = vattr.va_size - fp->f_offset; 1553 return (error); 1554 case FIONBIO: 1555 case FIOASYNC: 1556 return (0); 1557 default: 1558 return (VOP_IOCTL(vp, com, data, fp->f_flag, 1559 active_cred, td)); 1560 } 1561 default: 1562 return (ENOTTY); 1563 } 1564} 1565 1566/* 1567 * File table vnode poll routine. 1568 */ 1569static int 1570vn_poll(fp, events, active_cred, td) 1571 struct file *fp; 1572 int events; 1573 struct ucred *active_cred; 1574 struct thread *td; 1575{ 1576 struct vnode *vp; 1577 int error; 1578 1579 vp = fp->f_vnode; 1580#ifdef MAC 1581 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1582 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp); 1583 VOP_UNLOCK(vp, 0); 1584 if (!error) 1585#endif 1586 1587 error = VOP_POLL(vp, events, fp->f_cred, td); 1588 return (error); 1589} 1590 1591/* 1592 * Acquire the requested lock and then check for validity. LK_RETRY 1593 * permits vn_lock to return doomed vnodes. 1594 */ 1595int 1596_vn_lock(struct vnode *vp, int flags, char *file, int line) 1597{ 1598 int error; 1599 1600 VNASSERT((flags & LK_TYPE_MASK) != 0, vp, 1601 ("vn_lock called with no locktype.")); 1602 do { 1603#ifdef DEBUG_VFS_LOCKS 1604 KASSERT(vp->v_holdcnt != 0, 1605 ("vn_lock %p: zero hold count", vp)); 1606#endif 1607 error = VOP_LOCK1(vp, flags, file, line); 1608 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */ 1609 KASSERT((flags & LK_RETRY) == 0 || error == 0, 1610 ("LK_RETRY set with incompatible flags (0x%x) or an error occurred (%d)", 1611 flags, error)); 1612 /* 1613 * Callers specify LK_RETRY if they wish to get dead vnodes. 1614 * If RETRY is not set, we return ENOENT instead. 1615 */ 1616 if (error == 0 && vp->v_iflag & VI_DOOMED && 1617 (flags & LK_RETRY) == 0) { 1618 VOP_UNLOCK(vp, 0); 1619 error = ENOENT; 1620 break; 1621 } 1622 } while (flags & LK_RETRY && error != 0); 1623 return (error); 1624} 1625 1626/* 1627 * File table vnode close routine. 1628 */ 1629static int 1630vn_closefile(fp, td) 1631 struct file *fp; 1632 struct thread *td; 1633{ 1634 struct vnode *vp; 1635 struct flock lf; 1636 int error; 1637 bool ref; 1638 1639 vp = fp->f_vnode; 1640 fp->f_ops = &badfileops; 1641 ref= (fp->f_flag & FHASLOCK) != 0 && fp->f_type == DTYPE_VNODE; 1642 1643 error = vn_close1(vp, fp->f_flag, fp->f_cred, td, ref); 1644 1645 if (__predict_false(ref)) { 1646 lf.l_whence = SEEK_SET; 1647 lf.l_start = 0; 1648 lf.l_len = 0; 1649 lf.l_type = F_UNLCK; 1650 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); 1651 vrele(vp); 1652 } 1653 return (error); 1654} 1655 1656/* 1657 * Preparing to start a filesystem write operation. If the operation is 1658 * permitted, then we bump the count of operations in progress and 1659 * proceed. If a suspend request is in progress, we wait until the 1660 * suspension is over, and then proceed. 1661 */ 1662static int 1663vn_start_write_locked(struct mount *mp, int flags) 1664{ 1665 int error, mflags; 1666 1667 mtx_assert(MNT_MTX(mp), MA_OWNED); 1668 error = 0; 1669 1670 /* 1671 * Check on status of suspension. 1672 */ 1673 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 || 1674 mp->mnt_susp_owner != curthread) { 1675 mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? 1676 (flags & PCATCH) : 0) | (PUSER - 1); 1677 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 1678 if (flags & V_NOWAIT) { 1679 error = EWOULDBLOCK; 1680 goto unlock; 1681 } 1682 error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags, 1683 "suspfs", 0); 1684 if (error) 1685 goto unlock; 1686 } 1687 } 1688 if (flags & V_XSLEEP) 1689 goto unlock; 1690 mp->mnt_writeopcount++; 1691unlock: 1692 if (error != 0 || (flags & V_XSLEEP) != 0) 1693 MNT_REL(mp); 1694 MNT_IUNLOCK(mp); 1695 return (error); 1696} 1697 1698int 1699vn_start_write(struct vnode *vp, struct mount **mpp, int flags) 1700{ 1701 struct mount *mp; 1702 int error; 1703 1704 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), 1705 ("V_MNTREF requires mp")); 1706 1707 error = 0; 1708 /* 1709 * If a vnode is provided, get and return the mount point that 1710 * to which it will write. 1711 */ 1712 if (vp != NULL) { 1713 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 1714 *mpp = NULL; 1715 if (error != EOPNOTSUPP) 1716 return (error); 1717 return (0); 1718 } 1719 } 1720 if ((mp = *mpp) == NULL) 1721 return (0); 1722 1723 /* 1724 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 1725 * a vfs_ref(). 1726 * As long as a vnode is not provided we need to acquire a 1727 * refcount for the provided mountpoint too, in order to 1728 * emulate a vfs_ref(). 1729 */ 1730 MNT_ILOCK(mp); 1731 if (vp == NULL && (flags & V_MNTREF) == 0) 1732 MNT_REF(mp); 1733 1734 return (vn_start_write_locked(mp, flags)); 1735} 1736 1737/* 1738 * Secondary suspension. Used by operations such as vop_inactive 1739 * routines that are needed by the higher level functions. These 1740 * are allowed to proceed until all the higher level functions have 1741 * completed (indicated by mnt_writeopcount dropping to zero). At that 1742 * time, these operations are halted until the suspension is over. 1743 */ 1744int 1745vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags) 1746{ 1747 struct mount *mp; 1748 int error; 1749 1750 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), 1751 ("V_MNTREF requires mp")); 1752 1753 retry: 1754 if (vp != NULL) { 1755 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { 1756 *mpp = NULL; 1757 if (error != EOPNOTSUPP) 1758 return (error); 1759 return (0); 1760 } 1761 } 1762 /* 1763 * If we are not suspended or have not yet reached suspended 1764 * mode, then let the operation proceed. 1765 */ 1766 if ((mp = *mpp) == NULL) 1767 return (0); 1768 1769 /* 1770 * VOP_GETWRITEMOUNT() returns with the mp refcount held through 1771 * a vfs_ref(). 1772 * As long as a vnode is not provided we need to acquire a 1773 * refcount for the provided mountpoint too, in order to 1774 * emulate a vfs_ref(). 1775 */ 1776 MNT_ILOCK(mp); 1777 if (vp == NULL && (flags & V_MNTREF) == 0) 1778 MNT_REF(mp); 1779 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) { 1780 mp->mnt_secondary_writes++; 1781 mp->mnt_secondary_accwrites++; 1782 MNT_IUNLOCK(mp); 1783 return (0); 1784 } 1785 if (flags & V_NOWAIT) { 1786 MNT_REL(mp); 1787 MNT_IUNLOCK(mp); 1788 return (EWOULDBLOCK); 1789 } 1790 /* 1791 * Wait for the suspension to finish. 1792 */ 1793 error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP | 1794 ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0), 1795 "suspfs", 0); 1796 vfs_rel(mp); 1797 if (error == 0) 1798 goto retry; 1799 return (error); 1800} 1801 1802/* 1803 * Filesystem write operation has completed. If we are suspending and this 1804 * operation is the last one, notify the suspender that the suspension is 1805 * now in effect. 1806 */ 1807void 1808vn_finished_write(mp) 1809 struct mount *mp; 1810{ 1811 if (mp == NULL) 1812 return; 1813 MNT_ILOCK(mp); 1814 MNT_REL(mp); 1815 mp->mnt_writeopcount--; 1816 if (mp->mnt_writeopcount < 0) 1817 panic("vn_finished_write: neg cnt"); 1818 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1819 mp->mnt_writeopcount <= 0) 1820 wakeup(&mp->mnt_writeopcount); 1821 MNT_IUNLOCK(mp); 1822} 1823 1824 1825/* 1826 * Filesystem secondary write operation has completed. If we are 1827 * suspending and this operation is the last one, notify the suspender 1828 * that the suspension is now in effect. 1829 */ 1830void 1831vn_finished_secondary_write(mp) 1832 struct mount *mp; 1833{ 1834 if (mp == NULL) 1835 return; 1836 MNT_ILOCK(mp); 1837 MNT_REL(mp); 1838 mp->mnt_secondary_writes--; 1839 if (mp->mnt_secondary_writes < 0) 1840 panic("vn_finished_secondary_write: neg cnt"); 1841 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && 1842 mp->mnt_secondary_writes <= 0) 1843 wakeup(&mp->mnt_secondary_writes); 1844 MNT_IUNLOCK(mp); 1845} 1846 1847 1848 1849/* 1850 * Request a filesystem to suspend write operations. 1851 */ 1852int 1853vfs_write_suspend(struct mount *mp, int flags) 1854{ 1855 int error; 1856 1857 MNT_ILOCK(mp); 1858 if (mp->mnt_susp_owner == curthread) { 1859 MNT_IUNLOCK(mp); 1860 return (EALREADY); 1861 } 1862 while (mp->mnt_kern_flag & MNTK_SUSPEND) 1863 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0); 1864 1865 /* 1866 * Unmount holds a write reference on the mount point. If we 1867 * own busy reference and drain for writers, we deadlock with 1868 * the reference draining in the unmount path. Callers of 1869 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if 1870 * vfs_busy() reference is owned and caller is not in the 1871 * unmount context. 1872 */ 1873 if ((flags & VS_SKIP_UNMOUNT) != 0 && 1874 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { 1875 MNT_IUNLOCK(mp); 1876 return (EBUSY); 1877 } 1878 1879 mp->mnt_kern_flag |= MNTK_SUSPEND; 1880 mp->mnt_susp_owner = curthread; 1881 if (mp->mnt_writeopcount > 0) 1882 (void) msleep(&mp->mnt_writeopcount, 1883 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0); 1884 else 1885 MNT_IUNLOCK(mp); 1886 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0) 1887 vfs_write_resume(mp, 0); 1888 return (error); 1889} 1890 1891/* 1892 * Request a filesystem to resume write operations. 1893 */ 1894void 1895vfs_write_resume(struct mount *mp, int flags) 1896{ 1897 1898 MNT_ILOCK(mp); 1899 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 1900 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner")); 1901 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 | 1902 MNTK_SUSPENDED); 1903 mp->mnt_susp_owner = NULL; 1904 wakeup(&mp->mnt_writeopcount); 1905 wakeup(&mp->mnt_flag); 1906 curthread->td_pflags &= ~TDP_IGNSUSP; 1907 if ((flags & VR_START_WRITE) != 0) { 1908 MNT_REF(mp); 1909 mp->mnt_writeopcount++; 1910 } 1911 MNT_IUNLOCK(mp); 1912 if ((flags & VR_NO_SUSPCLR) == 0) 1913 VFS_SUSP_CLEAN(mp); 1914 } else if ((flags & VR_START_WRITE) != 0) { 1915 MNT_REF(mp); 1916 vn_start_write_locked(mp, 0); 1917 } else { 1918 MNT_IUNLOCK(mp); 1919 } 1920} 1921 1922/* 1923 * Helper loop around vfs_write_suspend() for filesystem unmount VFS 1924 * methods. 1925 */ 1926int 1927vfs_write_suspend_umnt(struct mount *mp) 1928{ 1929 int error; 1930 1931 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0, 1932 ("vfs_write_suspend_umnt: recursed")); 1933 1934 /* dounmount() already called vn_start_write(). */ 1935 for (;;) { 1936 vn_finished_write(mp); 1937 error = vfs_write_suspend(mp, 0); 1938 if (error != 0) { 1939 vn_start_write(NULL, &mp, V_WAIT); 1940 return (error); 1941 } 1942 MNT_ILOCK(mp); 1943 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0) 1944 break; 1945 MNT_IUNLOCK(mp); 1946 vn_start_write(NULL, &mp, V_WAIT); 1947 } 1948 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); 1949 wakeup(&mp->mnt_flag); 1950 MNT_IUNLOCK(mp); 1951 curthread->td_pflags |= TDP_IGNSUSP; 1952 return (0); 1953} 1954 1955/* 1956 * Implement kqueues for files by translating it to vnode operation. 1957 */ 1958static int 1959vn_kqfilter(struct file *fp, struct knote *kn) 1960{ 1961 1962 return (VOP_KQFILTER(fp->f_vnode, kn)); 1963} 1964 1965/* 1966 * Simplified in-kernel wrapper calls for extended attribute access. 1967 * Both calls pass in a NULL credential, authorizing as "kernel" access. 1968 * Set IO_NODELOCKED in ioflg if the vnode is already locked. 1969 */ 1970int 1971vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, 1972 const char *attrname, int *buflen, char *buf, struct thread *td) 1973{ 1974 struct uio auio; 1975 struct iovec iov; 1976 int error; 1977 1978 iov.iov_len = *buflen; 1979 iov.iov_base = buf; 1980 1981 auio.uio_iov = &iov; 1982 auio.uio_iovcnt = 1; 1983 auio.uio_rw = UIO_READ; 1984 auio.uio_segflg = UIO_SYSSPACE; 1985 auio.uio_td = td; 1986 auio.uio_offset = 0; 1987 auio.uio_resid = *buflen; 1988 1989 if ((ioflg & IO_NODELOCKED) == 0) 1990 vn_lock(vp, LK_SHARED | LK_RETRY); 1991 1992 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 1993 1994 /* authorize attribute retrieval as kernel */ 1995 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL, 1996 td); 1997 1998 if ((ioflg & IO_NODELOCKED) == 0) 1999 VOP_UNLOCK(vp, 0); 2000 2001 if (error == 0) { 2002 *buflen = *buflen - auio.uio_resid; 2003 } 2004 2005 return (error); 2006} 2007 2008/* 2009 * XXX failure mode if partially written? 2010 */ 2011int 2012vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, 2013 const char *attrname, int buflen, char *buf, struct thread *td) 2014{ 2015 struct uio auio; 2016 struct iovec iov; 2017 struct mount *mp; 2018 int error; 2019 2020 iov.iov_len = buflen; 2021 iov.iov_base = buf; 2022 2023 auio.uio_iov = &iov; 2024 auio.uio_iovcnt = 1; 2025 auio.uio_rw = UIO_WRITE; 2026 auio.uio_segflg = UIO_SYSSPACE; 2027 auio.uio_td = td; 2028 auio.uio_offset = 0; 2029 auio.uio_resid = buflen; 2030 2031 if ((ioflg & IO_NODELOCKED) == 0) { 2032 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 2033 return (error); 2034 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2035 } 2036 2037 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 2038 2039 /* authorize attribute setting as kernel */ 2040 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td); 2041 2042 if ((ioflg & IO_NODELOCKED) == 0) { 2043 vn_finished_write(mp); 2044 VOP_UNLOCK(vp, 0); 2045 } 2046 2047 return (error); 2048} 2049 2050int 2051vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, 2052 const char *attrname, struct thread *td) 2053{ 2054 struct mount *mp; 2055 int error; 2056 2057 if ((ioflg & IO_NODELOCKED) == 0) { 2058 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) 2059 return (error); 2060 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 2061 } 2062 2063 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); 2064 2065 /* authorize attribute removal as kernel */ 2066 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td); 2067 if (error == EOPNOTSUPP) 2068 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, 2069 NULL, td); 2070 2071 if ((ioflg & IO_NODELOCKED) == 0) { 2072 vn_finished_write(mp); 2073 VOP_UNLOCK(vp, 0); 2074 } 2075 2076 return (error); 2077} 2078 2079static int 2080vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags, 2081 struct vnode **rvp) 2082{ 2083 2084 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp)); 2085} 2086 2087int 2088vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp) 2089{ 2090 2091 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino, 2092 lkflags, rvp)); 2093} 2094 2095int 2096vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg, 2097 int lkflags, struct vnode **rvp) 2098{ 2099 struct mount *mp; 2100 int ltype, error; 2101 2102 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get"); 2103 mp = vp->v_mount; 2104 ltype = VOP_ISLOCKED(vp); 2105 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED, 2106 ("vn_vget_ino: vp not locked")); 2107 error = vfs_busy(mp, MBF_NOWAIT); 2108 if (error != 0) { 2109 vfs_ref(mp); 2110 VOP_UNLOCK(vp, 0); 2111 error = vfs_busy(mp, 0); 2112 vn_lock(vp, ltype | LK_RETRY); 2113 vfs_rel(mp); 2114 if (error != 0) 2115 return (ENOENT); 2116 if (vp->v_iflag & VI_DOOMED) { 2117 vfs_unbusy(mp); 2118 return (ENOENT); 2119 } 2120 } 2121 VOP_UNLOCK(vp, 0); 2122 error = alloc(mp, alloc_arg, lkflags, rvp); 2123 vfs_unbusy(mp); 2124 if (*rvp != vp) 2125 vn_lock(vp, ltype | LK_RETRY); 2126 if (vp->v_iflag & VI_DOOMED) { 2127 if (error == 0) { 2128 if (*rvp == vp) 2129 vunref(vp); 2130 else 2131 vput(*rvp); 2132 } 2133 error = ENOENT; 2134 } 2135 return (error); 2136} 2137 2138int 2139vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, 2140 const struct thread *td) 2141{ 2142 2143 if (vp->v_type != VREG || td == NULL) 2144 return (0); 2145 PROC_LOCK(td->td_proc); 2146 if ((uoff_t)uio->uio_offset + uio->uio_resid > 2147 lim_cur(td->td_proc, RLIMIT_FSIZE)) { 2148 kern_psignal(td->td_proc, SIGXFSZ); 2149 PROC_UNLOCK(td->td_proc); 2150 return (EFBIG); 2151 } 2152 PROC_UNLOCK(td->td_proc); 2153 return (0); 2154} 2155 2156int 2157vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 2158 struct thread *td) 2159{ 2160 struct vnode *vp; 2161 2162 vp = fp->f_vnode; 2163#ifdef AUDIT 2164 vn_lock(vp, LK_SHARED | LK_RETRY); 2165 AUDIT_ARG_VNODE1(vp); 2166 VOP_UNLOCK(vp, 0); 2167#endif 2168 return (setfmode(td, active_cred, vp, mode)); 2169} 2170 2171int 2172vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 2173 struct thread *td) 2174{ 2175 struct vnode *vp; 2176 2177 vp = fp->f_vnode; 2178#ifdef AUDIT 2179 vn_lock(vp, LK_SHARED | LK_RETRY); 2180 AUDIT_ARG_VNODE1(vp); 2181 VOP_UNLOCK(vp, 0); 2182#endif 2183 return (setfown(td, active_cred, vp, uid, gid)); 2184} 2185 2186void 2187vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end) 2188{ 2189 vm_object_t object; 2190 2191 if ((object = vp->v_object) == NULL) 2192 return; 2193 VM_OBJECT_WLOCK(object); 2194 vm_object_page_remove(object, start, end, 0); 2195 VM_OBJECT_WUNLOCK(object); 2196} 2197 2198int 2199vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred) 2200{ 2201 struct vattr va; 2202 daddr_t bn, bnp; 2203 uint64_t bsize; 2204 off_t noff; 2205 int error; 2206 2207 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA, 2208 ("Wrong command %lu", cmd)); 2209 2210 if (vn_lock(vp, LK_SHARED) != 0) 2211 return (EBADF); 2212 if (vp->v_type != VREG) { 2213 error = ENOTTY; 2214 goto unlock; 2215 } 2216 error = VOP_GETATTR(vp, &va, cred); 2217 if (error != 0) 2218 goto unlock; 2219 noff = *off; 2220 if (noff >= va.va_size) { 2221 error = ENXIO; 2222 goto unlock; 2223 } 2224 bsize = vp->v_mount->mnt_stat.f_iosize; 2225 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) { 2226 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL); 2227 if (error == EOPNOTSUPP) { 2228 error = ENOTTY; 2229 goto unlock; 2230 } 2231 if ((bnp == -1 && cmd == FIOSEEKHOLE) || 2232 (bnp != -1 && cmd == FIOSEEKDATA)) { 2233 noff = bn * bsize; 2234 if (noff < *off) 2235 noff = *off; 2236 goto unlock; 2237 } 2238 } 2239 if (noff > va.va_size) 2240 noff = va.va_size; 2241 /* noff == va.va_size. There is an implicit hole at the end of file. */ 2242 if (cmd == FIOSEEKDATA) 2243 error = ENXIO; 2244unlock: 2245 VOP_UNLOCK(vp, 0); 2246 if (error == 0) 2247 *off = noff; 2248 return (error); 2249} 2250 2251int 2252vn_seek(struct file *fp, off_t offset, int whence, struct thread *td) 2253{ 2254 struct ucred *cred; 2255 struct vnode *vp; 2256 struct vattr vattr; 2257 off_t foffset, size; 2258 int error, noneg; 2259 2260 cred = td->td_ucred; 2261 vp = fp->f_vnode; 2262 foffset = foffset_lock(fp, 0); 2263 noneg = (vp->v_type != VCHR); 2264 error = 0; 2265 switch (whence) { 2266 case L_INCR: 2267 if (noneg && 2268 (foffset < 0 || 2269 (offset > 0 && foffset > OFF_MAX - offset))) { 2270 error = EOVERFLOW; 2271 break; 2272 } 2273 offset += foffset; 2274 break; 2275 case L_XTND: 2276 vn_lock(vp, LK_SHARED | LK_RETRY); 2277 error = VOP_GETATTR(vp, &vattr, cred); 2278 VOP_UNLOCK(vp, 0); 2279 if (error) 2280 break; 2281 2282 /* 2283 * If the file references a disk device, then fetch 2284 * the media size and use that to determine the ending 2285 * offset. 2286 */ 2287 if (vattr.va_size == 0 && vp->v_type == VCHR && 2288 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0) 2289 vattr.va_size = size; 2290 if (noneg && 2291 (vattr.va_size > OFF_MAX || 2292 (offset > 0 && vattr.va_size > OFF_MAX - offset))) { 2293 error = EOVERFLOW; 2294 break; 2295 } 2296 offset += vattr.va_size; 2297 break; 2298 case L_SET: 2299 break; 2300 case SEEK_DATA: 2301 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td); 2302 break; 2303 case SEEK_HOLE: 2304 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td); 2305 break; 2306 default: 2307 error = EINVAL; 2308 } 2309 if (error == 0 && noneg && offset < 0) 2310 error = EINVAL; 2311 if (error != 0) 2312 goto drop; 2313 VFS_KNOTE_UNLOCKED(vp, 0); 2314 *(off_t *)(td->td_retval) = offset; 2315drop: 2316 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); 2317 return (error); 2318} 2319 2320int 2321vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred, 2322 struct thread *td) 2323{ 2324 int error; 2325 2326 /* 2327 * Grant permission if the caller is the owner of the file, or 2328 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on 2329 * on the file. If the time pointer is null, then write 2330 * permission on the file is also sufficient. 2331 * 2332 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes: 2333 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES 2334 * will be allowed to set the times [..] to the current 2335 * server time. 2336 */ 2337 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td); 2338 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0) 2339 error = VOP_ACCESS(vp, VWRITE, cred, td); 2340 return (error); 2341} 2342