1 /* 2 * (MPSAFE) 3 * 4 * Copyright (c) 2009 The DragonFly Project. All rights reserved. 5 * 6 * This code is derived from software contributed to The DragonFly Project 7 * by Alex Hornung <ahornung@gmail.com> 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in 17 * the documentation and/or other materials provided with the 18 * distribution. 19 * 3. Neither the name of The DragonFly Project nor the names of its 20 * contributors may be used to endorse or promote products derived 21 * from this software without specific, prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 24 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 26 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 27 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 28 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 29 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 30 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 31 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 32 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 33 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 */ 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/time.h> 39 #include <sys/kernel.h> 40 #include <sys/lock.h> 41 #include <sys/fcntl.h> 42 #include <sys/proc.h> 43 #include <sys/priv.h> 44 #include <sys/signalvar.h> 45 #include <sys/vnode.h> 46 #include <sys/uio.h> 47 #include <sys/mount.h> 48 #include <sys/file.h> 49 #include <sys/fcntl.h> 50 #include <sys/namei.h> 51 #include <sys/dirent.h> 52 #include <sys/malloc.h> 53 #include <sys/stat.h> 54 #include <sys/reg.h> 55 #include <vm/vm_pager.h> 56 #include <vm/vm_zone.h> 57 #include <vm/vm_object.h> 58 #include <sys/filio.h> 59 #include <sys/ttycom.h> 60 #include <sys/tty.h> 61 #include <sys/diskslice.h> 62 #include <sys/sysctl.h> 63 #include <sys/devfs.h> 64 #include <sys/pioctl.h> 65 #include <vfs/fifofs/fifo.h> 66 67 #include <machine/limits.h> 68 69 #include <sys/buf2.h> 70 #include <sys/sysref2.h> 71 #include <sys/mplock2.h> 72 #include <vm/vm_page2.h> 73 74 MALLOC_DECLARE(M_DEVFS); 75 #define DEVFS_BADOP (void *)devfs_vop_badop 76 77 static int devfs_vop_badop(struct vop_generic_args *); 78 static int devfs_vop_access(struct vop_access_args *); 79 static int devfs_vop_inactive(struct vop_inactive_args *); 80 static int devfs_vop_reclaim(struct vop_reclaim_args *); 81 static int devfs_vop_readdir(struct vop_readdir_args *); 82 static int devfs_vop_getattr(struct vop_getattr_args *); 83 static int devfs_vop_setattr(struct vop_setattr_args *); 84 static int devfs_vop_readlink(struct vop_readlink_args *); 85 static int devfs_vop_print(struct vop_print_args *); 86 87 static int devfs_vop_nresolve(struct vop_nresolve_args *); 88 static int devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *); 89 static int devfs_vop_nmkdir(struct vop_nmkdir_args *); 90 static int devfs_vop_nsymlink(struct vop_nsymlink_args *); 91 static int devfs_vop_nrmdir(struct vop_nrmdir_args *); 92 static int devfs_vop_nremove(struct vop_nremove_args *); 93 94 static int devfs_spec_open(struct vop_open_args *); 95 static int devfs_spec_close(struct vop_close_args *); 96 static int devfs_spec_fsync(struct vop_fsync_args *); 97 98 static int devfs_spec_read(struct vop_read_args *); 99 static int devfs_spec_write(struct vop_write_args *); 100 static int devfs_spec_ioctl(struct vop_ioctl_args *); 101 static int devfs_spec_kqfilter(struct vop_kqfilter_args *); 102 static int devfs_spec_strategy(struct vop_strategy_args *); 103 static void devfs_spec_strategy_done(struct bio *); 104 static int devfs_spec_freeblks(struct vop_freeblks_args *); 105 static int devfs_spec_bmap(struct vop_bmap_args *); 106 static int devfs_spec_advlock(struct vop_advlock_args *); 107 static void devfs_spec_getpages_iodone(struct bio *); 108 static int devfs_spec_getpages(struct vop_getpages_args *); 109 110 static int devfs_fo_close(struct file *); 111 static int devfs_fo_read(struct file *, struct uio *, struct ucred *, int); 112 static int devfs_fo_write(struct file *, struct uio *, struct ucred *, int); 113 static int devfs_fo_stat(struct file *, struct stat *, struct ucred *); 114 static int devfs_fo_kqfilter(struct file *, struct knote *); 115 static int devfs_fo_ioctl(struct file *, u_long, caddr_t, 116 struct ucred *, struct sysmsg *); 117 static __inline int sequential_heuristic(struct uio *, struct file *); 118 119 extern struct lock devfs_lock; 120 121 /* 122 * devfs vnode operations for regular files. All vnode ops are MPSAFE. 123 */ 124 struct vop_ops devfs_vnode_norm_vops = { 125 .vop_default = vop_defaultop, 126 .vop_access = devfs_vop_access, 127 .vop_advlock = DEVFS_BADOP, 128 .vop_bmap = DEVFS_BADOP, 129 .vop_close = vop_stdclose, 130 .vop_getattr = devfs_vop_getattr, 131 .vop_inactive = devfs_vop_inactive, 132 .vop_ncreate = DEVFS_BADOP, 133 .vop_nresolve = devfs_vop_nresolve, 134 .vop_nlookupdotdot = devfs_vop_nlookupdotdot, 135 .vop_nlink = DEVFS_BADOP, 136 .vop_nmkdir = devfs_vop_nmkdir, 137 .vop_nmknod = DEVFS_BADOP, 138 .vop_nremove = devfs_vop_nremove, 139 .vop_nrename = DEVFS_BADOP, 140 .vop_nrmdir = devfs_vop_nrmdir, 141 .vop_nsymlink = devfs_vop_nsymlink, 142 .vop_open = vop_stdopen, 143 .vop_pathconf = vop_stdpathconf, 144 .vop_print = devfs_vop_print, 145 .vop_read = DEVFS_BADOP, 146 .vop_readdir = devfs_vop_readdir, 147 .vop_readlink = devfs_vop_readlink, 148 .vop_reclaim = devfs_vop_reclaim, 149 .vop_setattr = devfs_vop_setattr, 150 .vop_write = DEVFS_BADOP, 151 .vop_ioctl = DEVFS_BADOP 152 }; 153 154 /* 155 * devfs vnode operations for character devices. All vnode ops are MPSAFE. 156 */ 157 struct vop_ops devfs_vnode_dev_vops = { 158 .vop_default = vop_defaultop, 159 .vop_access = devfs_vop_access, 160 .vop_advlock = devfs_spec_advlock, 161 .vop_bmap = devfs_spec_bmap, 162 .vop_close = devfs_spec_close, 163 .vop_freeblks = devfs_spec_freeblks, 164 .vop_fsync = devfs_spec_fsync, 165 .vop_getattr = devfs_vop_getattr, 166 .vop_getpages = devfs_spec_getpages, 167 .vop_inactive = devfs_vop_inactive, 168 .vop_open = devfs_spec_open, 169 .vop_pathconf = vop_stdpathconf, 170 .vop_print = devfs_vop_print, 171 .vop_kqfilter = devfs_spec_kqfilter, 172 .vop_read = devfs_spec_read, 173 .vop_readdir = DEVFS_BADOP, 174 .vop_readlink = DEVFS_BADOP, 175 .vop_reclaim = devfs_vop_reclaim, 176 .vop_setattr = devfs_vop_setattr, 177 .vop_strategy = devfs_spec_strategy, 178 .vop_write = devfs_spec_write, 179 .vop_ioctl = devfs_spec_ioctl 180 }; 181 182 /* 183 * devfs file pointer operations. All fileops are MPSAFE. 184 */ 185 struct vop_ops *devfs_vnode_dev_vops_p = &devfs_vnode_dev_vops; 186 187 struct fileops devfs_dev_fileops = { 188 .fo_read = devfs_fo_read, 189 .fo_write = devfs_fo_write, 190 .fo_ioctl = devfs_fo_ioctl, 191 .fo_kqfilter = devfs_fo_kqfilter, 192 .fo_stat = devfs_fo_stat, 193 .fo_close = devfs_fo_close, 194 .fo_shutdown = nofo_shutdown 195 }; 196 197 /* 198 * These two functions are possibly temporary hacks for devices (aka 199 * the pty code) which want to control the node attributes themselves. 200 * 201 * XXX we may ultimately desire to simply remove the uid/gid/mode 202 * from the node entirely. 203 * 204 * MPSAFE - sorta. Theoretically the overwrite can compete since they 205 * are loading from the same fields. 206 */ 207 static __inline void 208 node_sync_dev_get(struct devfs_node *node) 209 { 210 cdev_t dev; 211 212 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) { 213 node->uid = dev->si_uid; 214 node->gid = dev->si_gid; 215 node->mode = dev->si_perms; 216 } 217 } 218 219 static __inline void 220 node_sync_dev_set(struct devfs_node *node) 221 { 222 cdev_t dev; 223 224 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) { 225 dev->si_uid = node->uid; 226 dev->si_gid = node->gid; 227 dev->si_perms = node->mode; 228 } 229 } 230 231 /* 232 * generic entry point for unsupported operations 233 */ 234 static int 235 devfs_vop_badop(struct vop_generic_args *ap) 236 { 237 return (EIO); 238 } 239 240 241 static int 242 devfs_vop_access(struct vop_access_args *ap) 243 { 244 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 245 int error; 246 247 if (!devfs_node_is_accessible(node)) 248 return ENOENT; 249 node_sync_dev_get(node); 250 error = vop_helper_access(ap, node->uid, node->gid, 251 node->mode, node->flags); 252 253 return error; 254 } 255 256 257 static int 258 devfs_vop_inactive(struct vop_inactive_args *ap) 259 { 260 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 261 262 if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0) 263 vrecycle(ap->a_vp); 264 return 0; 265 } 266 267 268 static int 269 devfs_vop_reclaim(struct vop_reclaim_args *ap) 270 { 271 struct devfs_node *node; 272 struct vnode *vp; 273 int locked; 274 275 /* 276 * Check if it is locked already. if not, we acquire the devfs lock 277 */ 278 if (!(lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE) { 279 lockmgr(&devfs_lock, LK_EXCLUSIVE); 280 locked = 1; 281 } else { 282 locked = 0; 283 } 284 285 /* 286 * Get rid of the devfs_node if it is no longer linked into the 287 * topology. 288 */ 289 vp = ap->a_vp; 290 if ((node = DEVFS_NODE(vp)) != NULL) { 291 node->v_node = NULL; 292 if ((node->flags & DEVFS_NODE_LINKED) == 0) 293 devfs_freep(node); 294 } 295 296 if (locked) 297 lockmgr(&devfs_lock, LK_RELEASE); 298 299 /* 300 * v_rdev needs to be properly released using v_release_rdev 301 * Make sure v_data is NULL as well. 302 */ 303 vp->v_data = NULL; 304 v_release_rdev(vp); 305 return 0; 306 } 307 308 309 static int 310 devfs_vop_readdir(struct vop_readdir_args *ap) 311 { 312 struct devfs_node *dnode = DEVFS_NODE(ap->a_vp); 313 struct devfs_node *node; 314 int cookie_index; 315 int ncookies; 316 int error2; 317 int error; 318 int r; 319 off_t *cookies; 320 off_t saveoff; 321 322 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n"); 323 324 if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX) 325 return (EINVAL); 326 if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0) 327 return (error); 328 329 if (!devfs_node_is_accessible(dnode)) { 330 vn_unlock(ap->a_vp); 331 return ENOENT; 332 } 333 334 lockmgr(&devfs_lock, LK_EXCLUSIVE); 335 336 saveoff = ap->a_uio->uio_offset; 337 338 if (ap->a_ncookies) { 339 ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */ 340 if (ncookies > 256) 341 ncookies = 256; 342 cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK); 343 cookie_index = 0; 344 } else { 345 ncookies = -1; 346 cookies = NULL; 347 cookie_index = 0; 348 } 349 350 nanotime(&dnode->atime); 351 352 if (saveoff == 0) { 353 r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino, 354 DT_DIR, 1, "."); 355 if (r) 356 goto done; 357 if (cookies) 358 cookies[cookie_index] = saveoff; 359 saveoff++; 360 cookie_index++; 361 if (cookie_index == ncookies) 362 goto done; 363 } 364 365 if (saveoff == 1) { 366 if (dnode->parent) { 367 r = vop_write_dirent(&error, ap->a_uio, 368 dnode->parent->d_dir.d_ino, 369 DT_DIR, 2, ".."); 370 } else { 371 r = vop_write_dirent(&error, ap->a_uio, 372 dnode->d_dir.d_ino, 373 DT_DIR, 2, ".."); 374 } 375 if (r) 376 goto done; 377 if (cookies) 378 cookies[cookie_index] = saveoff; 379 saveoff++; 380 cookie_index++; 381 if (cookie_index == ncookies) 382 goto done; 383 } 384 385 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { 386 if ((node->flags & DEVFS_HIDDEN) || 387 (node->flags & DEVFS_INVISIBLE)) { 388 continue; 389 } 390 391 /* 392 * If the node type is a valid devfs alias, then we make 393 * sure that the target isn't hidden. If it is, we don't 394 * show the link in the directory listing. 395 */ 396 if ((node->node_type == Plink) && (node->link_target != NULL) && 397 (node->link_target->flags & DEVFS_HIDDEN)) 398 continue; 399 400 if (node->cookie < saveoff) 401 continue; 402 403 saveoff = node->cookie; 404 405 error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino, 406 node->d_dir.d_type, 407 node->d_dir.d_namlen, 408 node->d_dir.d_name); 409 410 if (error2) 411 break; 412 413 saveoff++; 414 415 if (cookies) 416 cookies[cookie_index] = node->cookie; 417 ++cookie_index; 418 if (cookie_index == ncookies) 419 break; 420 } 421 422 done: 423 lockmgr(&devfs_lock, LK_RELEASE); 424 vn_unlock(ap->a_vp); 425 426 ap->a_uio->uio_offset = saveoff; 427 if (error && cookie_index == 0) { 428 if (cookies) { 429 kfree(cookies, M_TEMP); 430 *ap->a_ncookies = 0; 431 *ap->a_cookies = NULL; 432 } 433 } else { 434 if (cookies) { 435 *ap->a_ncookies = cookie_index; 436 *ap->a_cookies = cookies; 437 } 438 } 439 return (error); 440 } 441 442 443 static int 444 devfs_vop_nresolve(struct vop_nresolve_args *ap) 445 { 446 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); 447 struct devfs_node *node, *found = NULL; 448 struct namecache *ncp; 449 struct vnode *vp = NULL; 450 int error = 0; 451 int len; 452 int depth; 453 454 ncp = ap->a_nch->ncp; 455 len = ncp->nc_nlen; 456 457 if (!devfs_node_is_accessible(dnode)) 458 return ENOENT; 459 460 lockmgr(&devfs_lock, LK_EXCLUSIVE); 461 462 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) { 463 error = ENOENT; 464 cache_setvp(ap->a_nch, NULL); 465 goto out; 466 } 467 468 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { 469 if (len == node->d_dir.d_namlen) { 470 if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) { 471 found = node; 472 break; 473 } 474 } 475 } 476 477 if (found) { 478 depth = 0; 479 while ((found->node_type == Plink) && (found->link_target)) { 480 if (depth >= 8) { 481 devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8"); 482 break; 483 } 484 485 found = found->link_target; 486 ++depth; 487 } 488 489 if (!(found->flags & DEVFS_HIDDEN)) 490 devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found); 491 } 492 493 if (vp == NULL) { 494 error = ENOENT; 495 cache_setvp(ap->a_nch, NULL); 496 goto out; 497 498 } 499 KKASSERT(vp); 500 vn_unlock(vp); 501 cache_setvp(ap->a_nch, vp); 502 vrele(vp); 503 out: 504 lockmgr(&devfs_lock, LK_RELEASE); 505 506 return error; 507 } 508 509 510 static int 511 devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap) 512 { 513 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); 514 515 *ap->a_vpp = NULL; 516 if (!devfs_node_is_accessible(dnode)) 517 return ENOENT; 518 519 lockmgr(&devfs_lock, LK_EXCLUSIVE); 520 if (dnode->parent != NULL) { 521 devfs_allocv(ap->a_vpp, dnode->parent); 522 vn_unlock(*ap->a_vpp); 523 } 524 lockmgr(&devfs_lock, LK_RELEASE); 525 526 return ((*ap->a_vpp == NULL) ? ENOENT : 0); 527 } 528 529 530 static int 531 devfs_vop_getattr(struct vop_getattr_args *ap) 532 { 533 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 534 struct vattr *vap = ap->a_vap; 535 struct partinfo pinfo; 536 int error = 0; 537 538 #if 0 539 if (!devfs_node_is_accessible(node)) 540 return ENOENT; 541 #endif 542 node_sync_dev_get(node); 543 544 lockmgr(&devfs_lock, LK_EXCLUSIVE); 545 546 /* start by zeroing out the attributes */ 547 VATTR_NULL(vap); 548 549 /* next do all the common fields */ 550 vap->va_type = ap->a_vp->v_type; 551 vap->va_mode = node->mode; 552 vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ; 553 vap->va_flags = 0; 554 vap->va_blocksize = DEV_BSIZE; 555 vap->va_bytes = vap->va_size = 0; 556 557 vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0]; 558 559 vap->va_atime = node->atime; 560 vap->va_mtime = node->mtime; 561 vap->va_ctime = node->ctime; 562 563 vap->va_nlink = 1; /* number of references to file */ 564 565 vap->va_uid = node->uid; 566 vap->va_gid = node->gid; 567 568 vap->va_rmajor = 0; 569 vap->va_rminor = 0; 570 571 if ((node->node_type == Pdev) && node->d_dev) { 572 reference_dev(node->d_dev); 573 vap->va_rminor = node->d_dev->si_uminor; 574 release_dev(node->d_dev); 575 } 576 577 /* For a softlink the va_size is the length of the softlink */ 578 if (node->symlink_name != 0) { 579 vap->va_bytes = vap->va_size = node->symlink_namelen; 580 } 581 582 /* 583 * For a disk-type device, va_size is the size of the underlying 584 * device, so that lseek() works properly. 585 */ 586 if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) { 587 bzero(&pinfo, sizeof(pinfo)); 588 error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo, 589 0, proc0.p_ucred, NULL); 590 if ((error == 0) && (pinfo.media_blksize != 0)) { 591 vap->va_size = pinfo.media_size; 592 } else { 593 vap->va_size = 0; 594 error = 0; 595 } 596 } 597 598 lockmgr(&devfs_lock, LK_RELEASE); 599 600 return (error); 601 } 602 603 604 static int 605 devfs_vop_setattr(struct vop_setattr_args *ap) 606 { 607 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 608 struct vattr *vap; 609 int error = 0; 610 611 if (!devfs_node_is_accessible(node)) 612 return ENOENT; 613 node_sync_dev_get(node); 614 615 lockmgr(&devfs_lock, LK_EXCLUSIVE); 616 617 vap = ap->a_vap; 618 619 if (vap->va_uid != (uid_t)VNOVAL) { 620 if ((ap->a_cred->cr_uid != node->uid) && 621 (!groupmember(node->gid, ap->a_cred))) { 622 error = priv_check(curthread, PRIV_VFS_CHOWN); 623 if (error) 624 goto out; 625 } 626 node->uid = vap->va_uid; 627 } 628 629 if (vap->va_gid != (uid_t)VNOVAL) { 630 if ((ap->a_cred->cr_uid != node->uid) && 631 (!groupmember(node->gid, ap->a_cred))) { 632 error = priv_check(curthread, PRIV_VFS_CHOWN); 633 if (error) 634 goto out; 635 } 636 node->gid = vap->va_gid; 637 } 638 639 if (vap->va_mode != (mode_t)VNOVAL) { 640 if (ap->a_cred->cr_uid != node->uid) { 641 error = priv_check(curthread, PRIV_VFS_ADMIN); 642 if (error) 643 goto out; 644 } 645 node->mode = vap->va_mode; 646 } 647 648 out: 649 node_sync_dev_set(node); 650 nanotime(&node->ctime); 651 lockmgr(&devfs_lock, LK_RELEASE); 652 653 return error; 654 } 655 656 657 static int 658 devfs_vop_readlink(struct vop_readlink_args *ap) 659 { 660 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 661 int ret; 662 663 if (!devfs_node_is_accessible(node)) 664 return ENOENT; 665 666 lockmgr(&devfs_lock, LK_EXCLUSIVE); 667 ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio); 668 lockmgr(&devfs_lock, LK_RELEASE); 669 670 return ret; 671 } 672 673 674 static int 675 devfs_vop_print(struct vop_print_args *ap) 676 { 677 return (0); 678 } 679 680 static int 681 devfs_vop_nmkdir(struct vop_nmkdir_args *ap) 682 { 683 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); 684 struct devfs_node *node; 685 686 if (!devfs_node_is_accessible(dnode)) 687 return ENOENT; 688 689 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) 690 goto out; 691 692 lockmgr(&devfs_lock, LK_EXCLUSIVE); 693 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Pdir, 694 ap->a_nch->ncp->nc_name, dnode, NULL); 695 696 if (*ap->a_vpp) { 697 node = DEVFS_NODE(*ap->a_vpp); 698 node->flags |= DEVFS_USER_CREATED; 699 cache_setunresolved(ap->a_nch); 700 cache_setvp(ap->a_nch, *ap->a_vpp); 701 } 702 lockmgr(&devfs_lock, LK_RELEASE); 703 out: 704 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0); 705 } 706 707 static int 708 devfs_vop_nsymlink(struct vop_nsymlink_args *ap) 709 { 710 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); 711 struct devfs_node *node; 712 size_t targetlen; 713 714 if (!devfs_node_is_accessible(dnode)) 715 return ENOENT; 716 717 ap->a_vap->va_type = VLNK; 718 719 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) 720 goto out; 721 722 lockmgr(&devfs_lock, LK_EXCLUSIVE); 723 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Plink, 724 ap->a_nch->ncp->nc_name, dnode, NULL); 725 726 targetlen = strlen(ap->a_target); 727 if (*ap->a_vpp) { 728 node = DEVFS_NODE(*ap->a_vpp); 729 node->flags |= DEVFS_USER_CREATED; 730 node->symlink_namelen = targetlen; 731 node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK); 732 memcpy(node->symlink_name, ap->a_target, targetlen); 733 node->symlink_name[targetlen] = '\0'; 734 cache_setunresolved(ap->a_nch); 735 cache_setvp(ap->a_nch, *ap->a_vpp); 736 } 737 lockmgr(&devfs_lock, LK_RELEASE); 738 out: 739 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0); 740 } 741 742 static int 743 devfs_vop_nrmdir(struct vop_nrmdir_args *ap) 744 { 745 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); 746 struct devfs_node *node; 747 struct namecache *ncp; 748 int error = ENOENT; 749 750 ncp = ap->a_nch->ncp; 751 752 if (!devfs_node_is_accessible(dnode)) 753 return ENOENT; 754 755 lockmgr(&devfs_lock, LK_EXCLUSIVE); 756 757 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) 758 goto out; 759 760 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { 761 if (ncp->nc_nlen != node->d_dir.d_namlen) 762 continue; 763 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen)) 764 continue; 765 766 /* 767 * only allow removal of user created dirs 768 */ 769 if ((node->flags & DEVFS_USER_CREATED) == 0) { 770 error = EPERM; 771 goto out; 772 } else if (node->node_type != Pdir) { 773 error = ENOTDIR; 774 goto out; 775 } else if (node->nchildren > 2) { 776 error = ENOTEMPTY; 777 goto out; 778 } else { 779 if (node->v_node) 780 cache_inval_vp(node->v_node, CINV_DESTROY); 781 devfs_unlinkp(node); 782 error = 0; 783 break; 784 } 785 } 786 787 cache_setunresolved(ap->a_nch); 788 cache_setvp(ap->a_nch, NULL); 789 790 out: 791 lockmgr(&devfs_lock, LK_RELEASE); 792 return error; 793 } 794 795 static int 796 devfs_vop_nremove(struct vop_nremove_args *ap) 797 { 798 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); 799 struct devfs_node *node; 800 struct namecache *ncp; 801 int error = ENOENT; 802 803 ncp = ap->a_nch->ncp; 804 805 if (!devfs_node_is_accessible(dnode)) 806 return ENOENT; 807 808 lockmgr(&devfs_lock, LK_EXCLUSIVE); 809 810 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) 811 goto out; 812 813 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { 814 if (ncp->nc_nlen != node->d_dir.d_namlen) 815 continue; 816 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen)) 817 continue; 818 819 /* 820 * only allow removal of user created stuff (e.g. symlinks) 821 */ 822 if ((node->flags & DEVFS_USER_CREATED) == 0) { 823 error = EPERM; 824 goto out; 825 } else if (node->node_type == Pdir) { 826 error = EISDIR; 827 goto out; 828 } else { 829 if (node->v_node) 830 cache_inval_vp(node->v_node, CINV_DESTROY); 831 devfs_unlinkp(node); 832 error = 0; 833 break; 834 } 835 } 836 837 cache_setunresolved(ap->a_nch); 838 cache_setvp(ap->a_nch, NULL); 839 840 out: 841 lockmgr(&devfs_lock, LK_RELEASE); 842 return error; 843 } 844 845 846 static int 847 devfs_spec_open(struct vop_open_args *ap) 848 { 849 struct vnode *vp = ap->a_vp; 850 struct vnode *orig_vp = NULL; 851 struct devfs_node *node = DEVFS_NODE(vp); 852 struct devfs_node *newnode; 853 cdev_t dev, ndev = NULL; 854 int error = 0; 855 856 if (node) { 857 if (node->d_dev == NULL) 858 return ENXIO; 859 if (!devfs_node_is_accessible(node)) 860 return ENOENT; 861 } 862 863 if ((dev = vp->v_rdev) == NULL) 864 return ENXIO; 865 866 if (node && ap->a_fp) { 867 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n"); 868 lockmgr(&devfs_lock, LK_EXCLUSIVE); 869 870 ndev = devfs_clone(dev, node->d_dir.d_name, node->d_dir.d_namlen, 871 ap->a_mode, ap->a_cred); 872 if (ndev != NULL) { 873 newnode = devfs_create_device_node( 874 DEVFS_MNTDATA(vp->v_mount)->root_node, 875 ndev, NULL, NULL); 876 /* XXX: possibly destroy device if this happens */ 877 878 if (newnode != NULL) { 879 dev = ndev; 880 devfs_link_dev(dev); 881 882 devfs_debug(DEVFS_DEBUG_DEBUG, 883 "parent here is: %s, node is: |%s|\n", 884 ((node->parent->node_type == Proot) ? 885 "ROOT!" : node->parent->d_dir.d_name), 886 newnode->d_dir.d_name); 887 devfs_debug(DEVFS_DEBUG_DEBUG, 888 "test: %s\n", 889 ((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name); 890 891 /* 892 * orig_vp is set to the original vp if we cloned. 893 */ 894 /* node->flags |= DEVFS_CLONED; */ 895 devfs_allocv(&vp, newnode); 896 orig_vp = ap->a_vp; 897 ap->a_vp = vp; 898 } 899 } 900 lockmgr(&devfs_lock, LK_RELEASE); 901 } 902 903 devfs_debug(DEVFS_DEBUG_DEBUG, 904 "devfs_spec_open() called on %s! \n", 905 dev->si_name); 906 907 /* 908 * Make this field valid before any I/O in ->d_open 909 */ 910 if (!dev->si_iosize_max) 911 dev->si_iosize_max = DFLTPHYS; 912 913 if (dev_dflags(dev) & D_TTY) 914 vsetflags(vp, VISTTY); 915 916 /* 917 * Open underlying device 918 */ 919 vn_unlock(vp); 920 error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred); 921 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 922 923 /* 924 * Clean up any cloned vp if we error out. 925 */ 926 if (error) { 927 if (orig_vp) { 928 vput(vp); 929 ap->a_vp = orig_vp; 930 /* orig_vp = NULL; */ 931 } 932 return error; 933 } 934 935 /* 936 * This checks if the disk device is going to be opened for writing. 937 * It will be only allowed in the cases where securelevel permits it 938 * and it's not mounted R/W. 939 */ 940 if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) && 941 (ap->a_cred != FSCRED)) { 942 943 /* Very secure mode. No open for writing allowed */ 944 if (securelevel >= 2) 945 return EPERM; 946 947 /* 948 * If it is mounted R/W, do not allow to open for writing. 949 * In the case it's mounted read-only but securelevel 950 * is >= 1, then do not allow opening for writing either. 951 */ 952 if (vfs_mountedon(vp)) { 953 if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY)) 954 return EBUSY; 955 else if (securelevel >= 1) 956 return EPERM; 957 } 958 } 959 960 if (dev_dflags(dev) & D_TTY) { 961 if (dev->si_tty) { 962 struct tty *tp; 963 tp = dev->si_tty; 964 if (!tp->t_stop) { 965 devfs_debug(DEVFS_DEBUG_DEBUG, 966 "devfs: no t_stop\n"); 967 tp->t_stop = nottystop; 968 } 969 } 970 } 971 972 973 if (vn_isdisk(vp, NULL)) { 974 if (!dev->si_bsize_phys) 975 dev->si_bsize_phys = DEV_BSIZE; 976 vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1); 977 } 978 979 vop_stdopen(ap); 980 #if 0 981 if (node) 982 nanotime(&node->atime); 983 #endif 984 985 /* 986 * If we replaced the vp the vop_stdopen() call will have loaded 987 * it into fp->f_data and vref()d the vp, giving us two refs. So 988 * instead of just unlocking it here we have to vput() it. 989 */ 990 if (orig_vp) 991 vput(vp); 992 993 /* Ugly pty magic, to make pty devices appear once they are opened */ 994 if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY) 995 node->flags &= ~DEVFS_INVISIBLE; 996 997 if (ap->a_fp) { 998 KKASSERT(ap->a_fp->f_type == DTYPE_VNODE); 999 KKASSERT(ap->a_fp->f_flag == (ap->a_mode & FMASK)); 1000 ap->a_fp->f_ops = &devfs_dev_fileops; 1001 KKASSERT(ap->a_fp->f_data == (void *)vp); 1002 } 1003 1004 return 0; 1005 } 1006 1007 1008 static int 1009 devfs_spec_close(struct vop_close_args *ap) 1010 { 1011 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 1012 struct proc *p = curproc; 1013 struct vnode *vp = ap->a_vp; 1014 cdev_t dev = vp->v_rdev; 1015 int error = 0; 1016 int needrelock; 1017 1018 if (dev) 1019 devfs_debug(DEVFS_DEBUG_DEBUG, 1020 "devfs_spec_close() called on %s! \n", 1021 dev->si_name); 1022 else 1023 devfs_debug(DEVFS_DEBUG_DEBUG, 1024 "devfs_spec_close() called, null vode!\n"); 1025 1026 /* 1027 * A couple of hacks for devices and tty devices. The 1028 * vnode ref count cannot be used to figure out the 1029 * last close, but we can use v_opencount now that 1030 * revoke works properly. 1031 * 1032 * Detect the last close on a controlling terminal and clear 1033 * the session (half-close). 1034 */ 1035 if (dev) 1036 reference_dev(dev); 1037 1038 if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) { 1039 p->p_session->s_ttyvp = NULL; 1040 vrele(vp); 1041 } 1042 1043 /* 1044 * Vnodes can be opened and closed multiple times. Do not really 1045 * close the device unless (1) it is being closed forcibly, 1046 * (2) the device wants to track closes, or (3) this is the last 1047 * vnode doing its last close on the device. 1048 * 1049 * XXX the VXLOCK (force close) case can leave vnodes referencing 1050 * a closed device. This might not occur now that our revoke is 1051 * fixed. 1052 */ 1053 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n"); 1054 if (dev && ((vp->v_flag & VRECLAIMED) || 1055 (dev_dflags(dev) & D_TRACKCLOSE) || 1056 (vp->v_opencount == 1))) { 1057 /* 1058 * Unlock around dev_dclose() 1059 */ 1060 needrelock = 0; 1061 if (vn_islocked(vp)) { 1062 needrelock = 1; 1063 vn_unlock(vp); 1064 } 1065 error = dev_dclose(dev, ap->a_fflag, S_IFCHR); 1066 1067 /* 1068 * Ugly pty magic, to make pty devices disappear again once 1069 * they are closed 1070 */ 1071 if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY) 1072 node->flags |= DEVFS_INVISIBLE; 1073 1074 if (needrelock) 1075 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1076 } else { 1077 error = 0; 1078 } 1079 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n"); 1080 1081 /* 1082 * Track the actual opens and closes on the vnode. The last close 1083 * disassociates the rdev. If the rdev is already disassociated or 1084 * the opencount is already 0, the vnode might have been revoked 1085 * and no further opencount tracking occurs. 1086 */ 1087 if (dev) 1088 release_dev(dev); 1089 if (vp->v_opencount > 0) 1090 vop_stdclose(ap); 1091 return(error); 1092 1093 } 1094 1095 1096 static int 1097 devfs_fo_close(struct file *fp) 1098 { 1099 struct vnode *vp = (struct vnode *)fp->f_data; 1100 int error; 1101 1102 fp->f_ops = &badfileops; 1103 error = vn_close(vp, fp->f_flag); 1104 1105 return (error); 1106 } 1107 1108 1109 /* 1110 * Device-optimized file table vnode read routine. 1111 * 1112 * This bypasses the VOP table and talks directly to the device. Most 1113 * filesystems just route to specfs and can make this optimization. 1114 * 1115 * MPALMOSTSAFE - acquires mplock 1116 */ 1117 static int 1118 devfs_fo_read(struct file *fp, struct uio *uio, 1119 struct ucred *cred, int flags) 1120 { 1121 struct devfs_node *node; 1122 struct vnode *vp; 1123 int ioflag; 1124 int error; 1125 cdev_t dev; 1126 1127 KASSERT(uio->uio_td == curthread, 1128 ("uio_td %p is not td %p", uio->uio_td, curthread)); 1129 1130 if (uio->uio_resid == 0) 1131 return 0; 1132 1133 vp = (struct vnode *)fp->f_data; 1134 if (vp == NULL || vp->v_type == VBAD) 1135 return EBADF; 1136 1137 node = DEVFS_NODE(vp); 1138 1139 if ((dev = vp->v_rdev) == NULL) 1140 return EBADF; 1141 1142 reference_dev(dev); 1143 1144 if ((flags & O_FOFFSET) == 0) 1145 uio->uio_offset = fp->f_offset; 1146 1147 ioflag = 0; 1148 if (flags & O_FBLOCKING) { 1149 /* ioflag &= ~IO_NDELAY; */ 1150 } else if (flags & O_FNONBLOCKING) { 1151 ioflag |= IO_NDELAY; 1152 } else if (fp->f_flag & FNONBLOCK) { 1153 ioflag |= IO_NDELAY; 1154 } 1155 if (flags & O_FBUFFERED) { 1156 /* ioflag &= ~IO_DIRECT; */ 1157 } else if (flags & O_FUNBUFFERED) { 1158 ioflag |= IO_DIRECT; 1159 } else if (fp->f_flag & O_DIRECT) { 1160 ioflag |= IO_DIRECT; 1161 } 1162 ioflag |= sequential_heuristic(uio, fp); 1163 1164 error = dev_dread(dev, uio, ioflag); 1165 1166 release_dev(dev); 1167 if (node) 1168 nanotime(&node->atime); 1169 if ((flags & O_FOFFSET) == 0) 1170 fp->f_offset = uio->uio_offset; 1171 fp->f_nextoff = uio->uio_offset; 1172 1173 return (error); 1174 } 1175 1176 1177 static int 1178 devfs_fo_write(struct file *fp, struct uio *uio, 1179 struct ucred *cred, int flags) 1180 { 1181 struct devfs_node *node; 1182 struct vnode *vp; 1183 int ioflag; 1184 int error; 1185 cdev_t dev; 1186 1187 KASSERT(uio->uio_td == curthread, 1188 ("uio_td %p is not p %p", uio->uio_td, curthread)); 1189 1190 vp = (struct vnode *)fp->f_data; 1191 if (vp == NULL || vp->v_type == VBAD) 1192 return EBADF; 1193 1194 node = DEVFS_NODE(vp); 1195 1196 if (vp->v_type == VREG) 1197 bwillwrite(uio->uio_resid); 1198 1199 vp = (struct vnode *)fp->f_data; 1200 1201 if ((dev = vp->v_rdev) == NULL) 1202 return EBADF; 1203 1204 reference_dev(dev); 1205 1206 if ((flags & O_FOFFSET) == 0) 1207 uio->uio_offset = fp->f_offset; 1208 1209 ioflag = IO_UNIT; 1210 if (vp->v_type == VREG && 1211 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 1212 ioflag |= IO_APPEND; 1213 } 1214 1215 if (flags & O_FBLOCKING) { 1216 /* ioflag &= ~IO_NDELAY; */ 1217 } else if (flags & O_FNONBLOCKING) { 1218 ioflag |= IO_NDELAY; 1219 } else if (fp->f_flag & FNONBLOCK) { 1220 ioflag |= IO_NDELAY; 1221 } 1222 if (flags & O_FBUFFERED) { 1223 /* ioflag &= ~IO_DIRECT; */ 1224 } else if (flags & O_FUNBUFFERED) { 1225 ioflag |= IO_DIRECT; 1226 } else if (fp->f_flag & O_DIRECT) { 1227 ioflag |= IO_DIRECT; 1228 } 1229 if (flags & O_FASYNCWRITE) { 1230 /* ioflag &= ~IO_SYNC; */ 1231 } else if (flags & O_FSYNCWRITE) { 1232 ioflag |= IO_SYNC; 1233 } else if (fp->f_flag & O_FSYNC) { 1234 ioflag |= IO_SYNC; 1235 } 1236 1237 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 1238 ioflag |= IO_SYNC; 1239 ioflag |= sequential_heuristic(uio, fp); 1240 1241 error = dev_dwrite(dev, uio, ioflag); 1242 1243 release_dev(dev); 1244 if (node) { 1245 nanotime(&node->atime); 1246 nanotime(&node->mtime); 1247 } 1248 1249 if ((flags & O_FOFFSET) == 0) 1250 fp->f_offset = uio->uio_offset; 1251 fp->f_nextoff = uio->uio_offset; 1252 1253 return (error); 1254 } 1255 1256 1257 static int 1258 devfs_fo_stat(struct file *fp, struct stat *sb, struct ucred *cred) 1259 { 1260 struct vnode *vp; 1261 struct vattr vattr; 1262 struct vattr *vap; 1263 u_short mode; 1264 cdev_t dev; 1265 int error; 1266 1267 vp = (struct vnode *)fp->f_data; 1268 if (vp == NULL || vp->v_type == VBAD) 1269 return EBADF; 1270 1271 error = vn_stat(vp, sb, cred); 1272 if (error) 1273 return (error); 1274 1275 vap = &vattr; 1276 error = VOP_GETATTR(vp, vap); 1277 if (error) 1278 return (error); 1279 1280 /* 1281 * Zero the spare stat fields 1282 */ 1283 sb->st_lspare = 0; 1284 sb->st_qspare1 = 0; 1285 sb->st_qspare2 = 0; 1286 1287 /* 1288 * Copy from vattr table ... or not in case it's a cloned device 1289 */ 1290 if (vap->va_fsid != VNOVAL) 1291 sb->st_dev = vap->va_fsid; 1292 else 1293 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 1294 1295 sb->st_ino = vap->va_fileid; 1296 1297 mode = vap->va_mode; 1298 mode |= S_IFCHR; 1299 sb->st_mode = mode; 1300 1301 if (vap->va_nlink > (nlink_t)-1) 1302 sb->st_nlink = (nlink_t)-1; 1303 else 1304 sb->st_nlink = vap->va_nlink; 1305 1306 sb->st_uid = vap->va_uid; 1307 sb->st_gid = vap->va_gid; 1308 sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev); 1309 sb->st_size = vap->va_bytes; 1310 sb->st_atimespec = vap->va_atime; 1311 sb->st_mtimespec = vap->va_mtime; 1312 sb->st_ctimespec = vap->va_ctime; 1313 1314 /* 1315 * A VCHR and VBLK device may track the last access and last modified 1316 * time independantly of the filesystem. This is particularly true 1317 * because device read and write calls may bypass the filesystem. 1318 */ 1319 if (vp->v_type == VCHR || vp->v_type == VBLK) { 1320 dev = vp->v_rdev; 1321 if (dev != NULL) { 1322 if (dev->si_lastread) { 1323 sb->st_atimespec.tv_sec = dev->si_lastread; 1324 sb->st_atimespec.tv_nsec = 0; 1325 } 1326 if (dev->si_lastwrite) { 1327 sb->st_atimespec.tv_sec = dev->si_lastwrite; 1328 sb->st_atimespec.tv_nsec = 0; 1329 } 1330 } 1331 } 1332 1333 /* 1334 * According to www.opengroup.org, the meaning of st_blksize is 1335 * "a filesystem-specific preferred I/O block size for this 1336 * object. In some filesystem types, this may vary from file 1337 * to file" 1338 * Default to PAGE_SIZE after much discussion. 1339 */ 1340 1341 sb->st_blksize = PAGE_SIZE; 1342 1343 sb->st_flags = vap->va_flags; 1344 1345 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0); 1346 if (error) 1347 sb->st_gen = 0; 1348 else 1349 sb->st_gen = (u_int32_t)vap->va_gen; 1350 1351 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 1352 1353 return (0); 1354 } 1355 1356 1357 static int 1358 devfs_fo_kqfilter(struct file *fp, struct knote *kn) 1359 { 1360 struct vnode *vp; 1361 int error; 1362 cdev_t dev; 1363 1364 vp = (struct vnode *)fp->f_data; 1365 if (vp == NULL || vp->v_type == VBAD) { 1366 error = EBADF; 1367 goto done; 1368 } 1369 if ((dev = vp->v_rdev) == NULL) { 1370 error = EBADF; 1371 goto done; 1372 } 1373 reference_dev(dev); 1374 1375 error = dev_dkqfilter(dev, kn); 1376 1377 release_dev(dev); 1378 1379 done: 1380 return (error); 1381 } 1382 1383 /* 1384 * MPALMOSTSAFE - acquires mplock 1385 */ 1386 static int 1387 devfs_fo_ioctl(struct file *fp, u_long com, caddr_t data, 1388 struct ucred *ucred, struct sysmsg *msg) 1389 { 1390 struct devfs_node *node; 1391 struct vnode *vp; 1392 struct vnode *ovp; 1393 cdev_t dev; 1394 int error; 1395 struct fiodname_args *name_args; 1396 size_t namlen; 1397 const char *name; 1398 1399 vp = ((struct vnode *)fp->f_data); 1400 1401 if ((dev = vp->v_rdev) == NULL) 1402 return EBADF; /* device was revoked */ 1403 1404 reference_dev(dev); 1405 1406 node = DEVFS_NODE(vp); 1407 1408 devfs_debug(DEVFS_DEBUG_DEBUG, 1409 "devfs_fo_ioctl() called! for dev %s\n", 1410 dev->si_name); 1411 1412 if (com == FIODTYPE) { 1413 *(int *)data = dev_dflags(dev) & D_TYPEMASK; 1414 error = 0; 1415 goto out; 1416 } else if (com == FIODNAME) { 1417 name_args = (struct fiodname_args *)data; 1418 name = dev->si_name; 1419 namlen = strlen(name) + 1; 1420 1421 devfs_debug(DEVFS_DEBUG_DEBUG, 1422 "ioctl, got: FIODNAME for %s\n", name); 1423 1424 if (namlen <= name_args->len) 1425 error = copyout(dev->si_name, name_args->name, namlen); 1426 else 1427 error = EINVAL; 1428 1429 devfs_debug(DEVFS_DEBUG_DEBUG, 1430 "ioctl stuff: error: %d\n", error); 1431 goto out; 1432 } 1433 1434 error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg); 1435 1436 #if 0 1437 if (node) { 1438 nanotime(&node->atime); 1439 nanotime(&node->mtime); 1440 } 1441 #endif 1442 if (com == TIOCSCTTY) { 1443 devfs_debug(DEVFS_DEBUG_DEBUG, 1444 "devfs_fo_ioctl: got TIOCSCTTY on %s\n", 1445 dev->si_name); 1446 } 1447 if (error == 0 && com == TIOCSCTTY) { 1448 struct proc *p = curthread->td_proc; 1449 struct session *sess; 1450 1451 devfs_debug(DEVFS_DEBUG_DEBUG, 1452 "devfs_fo_ioctl: dealing with TIOCSCTTY on %s\n", 1453 dev->si_name); 1454 if (p == NULL) { 1455 error = ENOTTY; 1456 goto out; 1457 } 1458 sess = p->p_session; 1459 1460 /* 1461 * Do nothing if reassigning same control tty 1462 */ 1463 if (sess->s_ttyvp == vp) { 1464 error = 0; 1465 goto out; 1466 } 1467 1468 /* 1469 * Get rid of reference to old control tty 1470 */ 1471 ovp = sess->s_ttyvp; 1472 vref(vp); 1473 sess->s_ttyvp = vp; 1474 if (ovp) 1475 vrele(ovp); 1476 } 1477 1478 out: 1479 release_dev(dev); 1480 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_fo_ioctl() finished! \n"); 1481 return (error); 1482 } 1483 1484 1485 static int 1486 devfs_spec_fsync(struct vop_fsync_args *ap) 1487 { 1488 struct vnode *vp = ap->a_vp; 1489 int error; 1490 1491 if (!vn_isdisk(vp, NULL)) 1492 return (0); 1493 1494 /* 1495 * Flush all dirty buffers associated with a block device. 1496 */ 1497 error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL); 1498 return (error); 1499 } 1500 1501 static int 1502 devfs_spec_read(struct vop_read_args *ap) 1503 { 1504 struct devfs_node *node; 1505 struct vnode *vp; 1506 struct uio *uio; 1507 cdev_t dev; 1508 int error; 1509 1510 vp = ap->a_vp; 1511 dev = vp->v_rdev; 1512 uio = ap->a_uio; 1513 node = DEVFS_NODE(vp); 1514 1515 if (dev == NULL) /* device was revoked */ 1516 return (EBADF); 1517 if (uio->uio_resid == 0) 1518 return (0); 1519 1520 vn_unlock(vp); 1521 error = dev_dread(dev, uio, ap->a_ioflag); 1522 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1523 1524 if (node) 1525 nanotime(&node->atime); 1526 1527 return (error); 1528 } 1529 1530 /* 1531 * Vnode op for write 1532 * 1533 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1534 * struct ucred *a_cred) 1535 */ 1536 static int 1537 devfs_spec_write(struct vop_write_args *ap) 1538 { 1539 struct devfs_node *node; 1540 struct vnode *vp; 1541 struct uio *uio; 1542 cdev_t dev; 1543 int error; 1544 1545 vp = ap->a_vp; 1546 dev = vp->v_rdev; 1547 uio = ap->a_uio; 1548 node = DEVFS_NODE(vp); 1549 1550 KKASSERT(uio->uio_segflg != UIO_NOCOPY); 1551 1552 if (dev == NULL) /* device was revoked */ 1553 return (EBADF); 1554 1555 vn_unlock(vp); 1556 error = dev_dwrite(dev, uio, ap->a_ioflag); 1557 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1558 1559 if (node) { 1560 nanotime(&node->atime); 1561 nanotime(&node->mtime); 1562 } 1563 1564 return (error); 1565 } 1566 1567 /* 1568 * Device ioctl operation. 1569 * 1570 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data, 1571 * int a_fflag, struct ucred *a_cred, struct sysmsg *msg) 1572 */ 1573 static int 1574 devfs_spec_ioctl(struct vop_ioctl_args *ap) 1575 { 1576 struct vnode *vp = ap->a_vp; 1577 struct devfs_node *node; 1578 cdev_t dev; 1579 1580 if ((dev = vp->v_rdev) == NULL) 1581 return (EBADF); /* device was revoked */ 1582 node = DEVFS_NODE(vp); 1583 1584 #if 0 1585 if (node) { 1586 nanotime(&node->atime); 1587 nanotime(&node->mtime); 1588 } 1589 #endif 1590 1591 return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag, 1592 ap->a_cred, ap->a_sysmsg)); 1593 } 1594 1595 /* 1596 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn) 1597 */ 1598 /* ARGSUSED */ 1599 static int 1600 devfs_spec_kqfilter(struct vop_kqfilter_args *ap) 1601 { 1602 struct vnode *vp = ap->a_vp; 1603 struct devfs_node *node; 1604 cdev_t dev; 1605 1606 if ((dev = vp->v_rdev) == NULL) 1607 return (EBADF); /* device was revoked (EBADF) */ 1608 node = DEVFS_NODE(vp); 1609 1610 #if 0 1611 if (node) 1612 nanotime(&node->atime); 1613 #endif 1614 1615 return (dev_dkqfilter(dev, ap->a_kn)); 1616 } 1617 1618 /* 1619 * Convert a vnode strategy call into a device strategy call. Vnode strategy 1620 * calls are not limited to device DMA limits so we have to deal with the 1621 * case. 1622 * 1623 * spec_strategy(struct vnode *a_vp, struct bio *a_bio) 1624 */ 1625 static int 1626 devfs_spec_strategy(struct vop_strategy_args *ap) 1627 { 1628 struct bio *bio = ap->a_bio; 1629 struct buf *bp = bio->bio_buf; 1630 struct buf *nbp; 1631 struct vnode *vp; 1632 struct mount *mp; 1633 int chunksize; 1634 int maxiosize; 1635 1636 if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL) 1637 buf_start(bp); 1638 1639 /* 1640 * Collect statistics on synchronous and asynchronous read 1641 * and write counts for disks that have associated filesystems. 1642 */ 1643 vp = ap->a_vp; 1644 KKASSERT(vp->v_rdev != NULL); /* XXX */ 1645 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) { 1646 if (bp->b_cmd == BUF_CMD_READ) { 1647 if (bp->b_flags & BIO_SYNC) 1648 mp->mnt_stat.f_syncreads++; 1649 else 1650 mp->mnt_stat.f_asyncreads++; 1651 } else { 1652 if (bp->b_flags & BIO_SYNC) 1653 mp->mnt_stat.f_syncwrites++; 1654 else 1655 mp->mnt_stat.f_asyncwrites++; 1656 } 1657 } 1658 1659 /* 1660 * Device iosize limitations only apply to read and write. Shortcut 1661 * the I/O if it fits. 1662 */ 1663 if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) { 1664 devfs_debug(DEVFS_DEBUG_DEBUG, 1665 "%s: si_iosize_max not set!\n", 1666 dev_dname(vp->v_rdev)); 1667 maxiosize = MAXPHYS; 1668 } 1669 #if SPEC_CHAIN_DEBUG & 2 1670 maxiosize = 4096; 1671 #endif 1672 if (bp->b_bcount <= maxiosize || 1673 (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) { 1674 dev_dstrategy_chain(vp->v_rdev, bio); 1675 return (0); 1676 } 1677 1678 /* 1679 * Clone the buffer and set up an I/O chain to chunk up the I/O. 1680 */ 1681 nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO); 1682 initbufbio(nbp); 1683 buf_dep_init(nbp); 1684 BUF_LOCKINIT(nbp); 1685 BUF_LOCK(nbp, LK_EXCLUSIVE); 1686 BUF_KERNPROC(nbp); 1687 nbp->b_vp = vp; 1688 nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP); 1689 nbp->b_data = bp->b_data; 1690 nbp->b_bio1.bio_done = devfs_spec_strategy_done; 1691 nbp->b_bio1.bio_offset = bio->bio_offset; 1692 nbp->b_bio1.bio_caller_info1.ptr = bio; 1693 1694 /* 1695 * Start the first transfer 1696 */ 1697 if (vn_isdisk(vp, NULL)) 1698 chunksize = vp->v_rdev->si_bsize_phys; 1699 else 1700 chunksize = DEV_BSIZE; 1701 chunksize = maxiosize / chunksize * chunksize; 1702 #if SPEC_CHAIN_DEBUG & 1 1703 devfs_debug(DEVFS_DEBUG_DEBUG, 1704 "spec_strategy chained I/O chunksize=%d\n", 1705 chunksize); 1706 #endif 1707 nbp->b_cmd = bp->b_cmd; 1708 nbp->b_bcount = chunksize; 1709 nbp->b_bufsize = chunksize; /* used to detect a short I/O */ 1710 nbp->b_bio1.bio_caller_info2.index = chunksize; 1711 1712 #if SPEC_CHAIN_DEBUG & 1 1713 devfs_debug(DEVFS_DEBUG_DEBUG, 1714 "spec_strategy: chain %p offset %d/%d bcount %d\n", 1715 bp, 0, bp->b_bcount, nbp->b_bcount); 1716 #endif 1717 1718 dev_dstrategy(vp->v_rdev, &nbp->b_bio1); 1719 1720 if (DEVFS_NODE(vp)) { 1721 nanotime(&DEVFS_NODE(vp)->atime); 1722 nanotime(&DEVFS_NODE(vp)->mtime); 1723 } 1724 1725 return (0); 1726 } 1727 1728 /* 1729 * Chunked up transfer completion routine - chain transfers until done 1730 * 1731 * NOTE: MPSAFE callback. 1732 */ 1733 static 1734 void 1735 devfs_spec_strategy_done(struct bio *nbio) 1736 { 1737 struct buf *nbp = nbio->bio_buf; 1738 struct bio *bio = nbio->bio_caller_info1.ptr; /* original bio */ 1739 struct buf *bp = bio->bio_buf; /* original bp */ 1740 int chunksize = nbio->bio_caller_info2.index; /* chunking */ 1741 int boffset = nbp->b_data - bp->b_data; 1742 1743 if (nbp->b_flags & B_ERROR) { 1744 /* 1745 * An error terminates the chain, propogate the error back 1746 * to the original bp 1747 */ 1748 bp->b_flags |= B_ERROR; 1749 bp->b_error = nbp->b_error; 1750 bp->b_resid = bp->b_bcount - boffset + 1751 (nbp->b_bcount - nbp->b_resid); 1752 #if SPEC_CHAIN_DEBUG & 1 1753 devfs_debug(DEVFS_DEBUG_DEBUG, 1754 "spec_strategy: chain %p error %d bcount %d/%d\n", 1755 bp, bp->b_error, bp->b_bcount, 1756 bp->b_bcount - bp->b_resid); 1757 #endif 1758 kfree(nbp, M_DEVBUF); 1759 biodone(bio); 1760 } else if (nbp->b_resid) { 1761 /* 1762 * A short read or write terminates the chain 1763 */ 1764 bp->b_error = nbp->b_error; 1765 bp->b_resid = bp->b_bcount - boffset + 1766 (nbp->b_bcount - nbp->b_resid); 1767 #if SPEC_CHAIN_DEBUG & 1 1768 devfs_debug(DEVFS_DEBUG_DEBUG, 1769 "spec_strategy: chain %p short read(1) " 1770 "bcount %d/%d\n", 1771 bp, bp->b_bcount - bp->b_resid, bp->b_bcount); 1772 #endif 1773 kfree(nbp, M_DEVBUF); 1774 biodone(bio); 1775 } else if (nbp->b_bcount != nbp->b_bufsize) { 1776 /* 1777 * A short read or write can also occur by truncating b_bcount 1778 */ 1779 #if SPEC_CHAIN_DEBUG & 1 1780 devfs_debug(DEVFS_DEBUG_DEBUG, 1781 "spec_strategy: chain %p short read(2) " 1782 "bcount %d/%d\n", 1783 bp, nbp->b_bcount + boffset, bp->b_bcount); 1784 #endif 1785 bp->b_error = 0; 1786 bp->b_bcount = nbp->b_bcount + boffset; 1787 bp->b_resid = nbp->b_resid; 1788 kfree(nbp, M_DEVBUF); 1789 biodone(bio); 1790 } else if (nbp->b_bcount + boffset == bp->b_bcount) { 1791 /* 1792 * No more data terminates the chain 1793 */ 1794 #if SPEC_CHAIN_DEBUG & 1 1795 devfs_debug(DEVFS_DEBUG_DEBUG, 1796 "spec_strategy: chain %p finished bcount %d\n", 1797 bp, bp->b_bcount); 1798 #endif 1799 bp->b_error = 0; 1800 bp->b_resid = 0; 1801 kfree(nbp, M_DEVBUF); 1802 biodone(bio); 1803 } else { 1804 /* 1805 * Continue the chain 1806 */ 1807 boffset += nbp->b_bcount; 1808 nbp->b_data = bp->b_data + boffset; 1809 nbp->b_bcount = bp->b_bcount - boffset; 1810 if (nbp->b_bcount > chunksize) 1811 nbp->b_bcount = chunksize; 1812 nbp->b_bio1.bio_done = devfs_spec_strategy_done; 1813 nbp->b_bio1.bio_offset = bio->bio_offset + boffset; 1814 1815 #if SPEC_CHAIN_DEBUG & 1 1816 devfs_debug(DEVFS_DEBUG_DEBUG, 1817 "spec_strategy: chain %p offset %d/%d bcount %d\n", 1818 bp, boffset, bp->b_bcount, nbp->b_bcount); 1819 #endif 1820 1821 dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1); 1822 } 1823 } 1824 1825 /* 1826 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length) 1827 */ 1828 static int 1829 devfs_spec_freeblks(struct vop_freeblks_args *ap) 1830 { 1831 struct buf *bp; 1832 1833 /* 1834 * XXX: This assumes that strategy does the deed right away. 1835 * XXX: this may not be TRTTD. 1836 */ 1837 KKASSERT(ap->a_vp->v_rdev != NULL); 1838 if ((dev_dflags(ap->a_vp->v_rdev) & D_CANFREE) == 0) 1839 return (0); 1840 bp = geteblk(ap->a_length); 1841 bp->b_cmd = BUF_CMD_FREEBLKS; 1842 bp->b_bio1.bio_offset = ap->a_offset; 1843 bp->b_bcount = ap->a_length; 1844 dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1); 1845 return (0); 1846 } 1847 1848 /* 1849 * Implement degenerate case where the block requested is the block 1850 * returned, and assume that the entire device is contiguous in regards 1851 * to the contiguous block range (runp and runb). 1852 * 1853 * spec_bmap(struct vnode *a_vp, off_t a_loffset, 1854 * off_t *a_doffsetp, int *a_runp, int *a_runb) 1855 */ 1856 static int 1857 devfs_spec_bmap(struct vop_bmap_args *ap) 1858 { 1859 if (ap->a_doffsetp != NULL) 1860 *ap->a_doffsetp = ap->a_loffset; 1861 if (ap->a_runp != NULL) 1862 *ap->a_runp = MAXBSIZE; 1863 if (ap->a_runb != NULL) { 1864 if (ap->a_loffset < MAXBSIZE) 1865 *ap->a_runb = (int)ap->a_loffset; 1866 else 1867 *ap->a_runb = MAXBSIZE; 1868 } 1869 return (0); 1870 } 1871 1872 1873 /* 1874 * Special device advisory byte-level locks. 1875 * 1876 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, 1877 * struct flock *a_fl, int a_flags) 1878 */ 1879 /* ARGSUSED */ 1880 static int 1881 devfs_spec_advlock(struct vop_advlock_args *ap) 1882 { 1883 return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP); 1884 } 1885 1886 /* 1887 * NOTE: MPSAFE callback. 1888 */ 1889 static void 1890 devfs_spec_getpages_iodone(struct bio *bio) 1891 { 1892 bio->bio_buf->b_cmd = BUF_CMD_DONE; 1893 wakeup(bio->bio_buf); 1894 } 1895 1896 /* 1897 * spec_getpages() - get pages associated with device vnode. 1898 * 1899 * Note that spec_read and spec_write do not use the buffer cache, so we 1900 * must fully implement getpages here. 1901 */ 1902 static int 1903 devfs_spec_getpages(struct vop_getpages_args *ap) 1904 { 1905 vm_offset_t kva; 1906 int error; 1907 int i, pcount, size; 1908 struct buf *bp; 1909 vm_page_t m; 1910 vm_ooffset_t offset; 1911 int toff, nextoff, nread; 1912 struct vnode *vp = ap->a_vp; 1913 int blksiz; 1914 int gotreqpage; 1915 1916 error = 0; 1917 pcount = round_page(ap->a_count) / PAGE_SIZE; 1918 1919 /* 1920 * Calculate the offset of the transfer and do sanity check. 1921 */ 1922 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset; 1923 1924 /* 1925 * Round up physical size for real devices. We cannot round using 1926 * v_mount's block size data because v_mount has nothing to do with 1927 * the device. i.e. it's usually '/dev'. We need the physical block 1928 * size for the device itself. 1929 * 1930 * We can't use v_rdev->si_mountpoint because it only exists when the 1931 * block device is mounted. However, we can use v_rdev. 1932 */ 1933 if (vn_isdisk(vp, NULL)) 1934 blksiz = vp->v_rdev->si_bsize_phys; 1935 else 1936 blksiz = DEV_BSIZE; 1937 1938 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1); 1939 1940 bp = getpbuf_kva(NULL); 1941 kva = (vm_offset_t)bp->b_data; 1942 1943 /* 1944 * Map the pages to be read into the kva. 1945 */ 1946 pmap_qenter(kva, ap->a_m, pcount); 1947 1948 /* Build a minimal buffer header. */ 1949 bp->b_cmd = BUF_CMD_READ; 1950 bp->b_bcount = size; 1951 bp->b_resid = 0; 1952 bsetrunningbufspace(bp, size); 1953 1954 bp->b_bio1.bio_offset = offset; 1955 bp->b_bio1.bio_done = devfs_spec_getpages_iodone; 1956 1957 mycpu->gd_cnt.v_vnodein++; 1958 mycpu->gd_cnt.v_vnodepgsin += pcount; 1959 1960 /* Do the input. */ 1961 vn_strategy(ap->a_vp, &bp->b_bio1); 1962 1963 crit_enter(); 1964 1965 /* We definitely need to be at splbio here. */ 1966 while (bp->b_cmd != BUF_CMD_DONE) 1967 tsleep(bp, 0, "spread", 0); 1968 1969 crit_exit(); 1970 1971 if (bp->b_flags & B_ERROR) { 1972 if (bp->b_error) 1973 error = bp->b_error; 1974 else 1975 error = EIO; 1976 } 1977 1978 /* 1979 * If EOF is encountered we must zero-extend the result in order 1980 * to ensure that the page does not contain garabge. When no 1981 * error occurs, an early EOF is indicated if b_bcount got truncated. 1982 * b_resid is relative to b_bcount and should be 0, but some devices 1983 * might indicate an EOF with b_resid instead of truncating b_bcount. 1984 */ 1985 nread = bp->b_bcount - bp->b_resid; 1986 if (nread < ap->a_count) 1987 bzero((caddr_t)kva + nread, ap->a_count - nread); 1988 pmap_qremove(kva, pcount); 1989 1990 gotreqpage = 0; 1991 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) { 1992 nextoff = toff + PAGE_SIZE; 1993 m = ap->a_m[i]; 1994 1995 m->flags &= ~PG_ZERO; 1996 1997 /* 1998 * NOTE: vm_page_undirty/clear_dirty etc do not clear the 1999 * pmap modified bit. pmap modified bit should have 2000 * already been cleared. 2001 */ 2002 if (nextoff <= nread) { 2003 m->valid = VM_PAGE_BITS_ALL; 2004 vm_page_undirty(m); 2005 } else if (toff < nread) { 2006 /* 2007 * Since this is a VM request, we have to supply the 2008 * unaligned offset to allow vm_page_set_valid() 2009 * to zero sub-DEV_BSIZE'd portions of the page. 2010 */ 2011 vm_page_set_valid(m, 0, nread - toff); 2012 vm_page_clear_dirty_end_nonincl(m, 0, nread - toff); 2013 } else { 2014 m->valid = 0; 2015 vm_page_undirty(m); 2016 } 2017 2018 if (i != ap->a_reqpage) { 2019 /* 2020 * Just in case someone was asking for this page we 2021 * now tell them that it is ok to use. 2022 */ 2023 if (!error || (m->valid == VM_PAGE_BITS_ALL)) { 2024 if (m->valid) { 2025 if (m->flags & PG_WANTED) { 2026 vm_page_activate(m); 2027 } else { 2028 vm_page_deactivate(m); 2029 } 2030 vm_page_wakeup(m); 2031 } else { 2032 vm_page_free(m); 2033 } 2034 } else { 2035 vm_page_free(m); 2036 } 2037 } else if (m->valid) { 2038 gotreqpage = 1; 2039 /* 2040 * Since this is a VM request, we need to make the 2041 * entire page presentable by zeroing invalid sections. 2042 */ 2043 if (m->valid != VM_PAGE_BITS_ALL) 2044 vm_page_zero_invalid(m, FALSE); 2045 } 2046 } 2047 if (!gotreqpage) { 2048 m = ap->a_m[ap->a_reqpage]; 2049 devfs_debug(DEVFS_DEBUG_WARNING, 2050 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n", 2051 devtoname(vp->v_rdev), error, bp, bp->b_vp); 2052 devfs_debug(DEVFS_DEBUG_WARNING, 2053 " size: %d, resid: %d, a_count: %d, valid: 0x%x\n", 2054 size, bp->b_resid, ap->a_count, m->valid); 2055 devfs_debug(DEVFS_DEBUG_WARNING, 2056 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n", 2057 nread, ap->a_reqpage, (u_long)m->pindex, pcount); 2058 /* 2059 * Free the buffer header back to the swap buffer pool. 2060 */ 2061 relpbuf(bp, NULL); 2062 return VM_PAGER_ERROR; 2063 } 2064 /* 2065 * Free the buffer header back to the swap buffer pool. 2066 */ 2067 relpbuf(bp, NULL); 2068 if (DEVFS_NODE(ap->a_vp)) 2069 nanotime(&DEVFS_NODE(ap->a_vp)->mtime); 2070 return VM_PAGER_OK; 2071 } 2072 2073 static __inline 2074 int 2075 sequential_heuristic(struct uio *uio, struct file *fp) 2076 { 2077 /* 2078 * Sequential heuristic - detect sequential operation 2079 */ 2080 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 2081 uio->uio_offset == fp->f_nextoff) { 2082 /* 2083 * XXX we assume that the filesystem block size is 2084 * the default. Not true, but still gives us a pretty 2085 * good indicator of how sequential the read operations 2086 * are. 2087 */ 2088 int tmpseq = fp->f_seqcount; 2089 2090 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE; 2091 if (tmpseq > IO_SEQMAX) 2092 tmpseq = IO_SEQMAX; 2093 fp->f_seqcount = tmpseq; 2094 return(fp->f_seqcount << IO_SEQSHIFT); 2095 } 2096 2097 /* 2098 * Not sequential, quick draw-down of seqcount 2099 */ 2100 if (fp->f_seqcount > 1) 2101 fp->f_seqcount = 1; 2102 else 2103 fp->f_seqcount = 0; 2104 return(0); 2105 } 2106