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/namei.h> 50 #include <sys/dirent.h> 51 #include <sys/malloc.h> 52 #include <sys/stat.h> 53 #include <sys/reg.h> 54 #include <vm/vm_pager.h> 55 #include <vm/vm_zone.h> 56 #include <vm/vm_object.h> 57 #include <sys/filio.h> 58 #include <sys/ttycom.h> 59 #include <sys/tty.h> 60 #include <sys/diskslice.h> 61 #include <sys/sysctl.h> 62 #include <sys/devfs.h> 63 #include <sys/pioctl.h> 64 #include <vfs/fifofs/fifo.h> 65 66 #include <machine/limits.h> 67 68 #include <sys/buf2.h> 69 #include <sys/sysref2.h> 70 #include <sys/mplock2.h> 71 #include <vm/vm_page2.h> 72 73 #ifndef SPEC_CHAIN_DEBUG 74 #define SPEC_CHAIN_DEBUG 0 75 #endif 76 77 MALLOC_DECLARE(M_DEVFS); 78 #define DEVFS_BADOP (void *)devfs_vop_badop 79 80 static int devfs_vop_badop(struct vop_generic_args *); 81 static int devfs_vop_access(struct vop_access_args *); 82 static int devfs_vop_inactive(struct vop_inactive_args *); 83 static int devfs_vop_reclaim(struct vop_reclaim_args *); 84 static int devfs_vop_readdir(struct vop_readdir_args *); 85 static int devfs_vop_getattr(struct vop_getattr_args *); 86 static int devfs_vop_setattr(struct vop_setattr_args *); 87 static int devfs_vop_readlink(struct vop_readlink_args *); 88 static int devfs_vop_print(struct vop_print_args *); 89 90 static int devfs_vop_nresolve(struct vop_nresolve_args *); 91 static int devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *); 92 static int devfs_vop_nmkdir(struct vop_nmkdir_args *); 93 static int devfs_vop_nsymlink(struct vop_nsymlink_args *); 94 static int devfs_vop_nrmdir(struct vop_nrmdir_args *); 95 static int devfs_vop_nremove(struct vop_nremove_args *); 96 97 static int devfs_spec_open(struct vop_open_args *); 98 static int devfs_spec_close(struct vop_close_args *); 99 static int devfs_spec_fsync(struct vop_fsync_args *); 100 101 static int devfs_spec_read(struct vop_read_args *); 102 static int devfs_spec_write(struct vop_write_args *); 103 static int devfs_spec_ioctl(struct vop_ioctl_args *); 104 static int devfs_spec_kqfilter(struct vop_kqfilter_args *); 105 static int devfs_spec_strategy(struct vop_strategy_args *); 106 static void devfs_spec_strategy_done(struct bio *); 107 static int devfs_spec_freeblks(struct vop_freeblks_args *); 108 static int devfs_spec_bmap(struct vop_bmap_args *); 109 static int devfs_spec_advlock(struct vop_advlock_args *); 110 static void devfs_spec_getpages_iodone(struct bio *); 111 static int devfs_spec_getpages(struct vop_getpages_args *); 112 113 static int devfs_fo_close(struct file *); 114 static int devfs_fo_read(struct file *, struct uio *, struct ucred *, int); 115 static int devfs_fo_write(struct file *, struct uio *, struct ucred *, int); 116 static int devfs_fo_stat(struct file *, struct stat *, struct ucred *); 117 static int devfs_fo_kqfilter(struct file *, struct knote *); 118 static int devfs_fo_ioctl(struct file *, u_long, caddr_t, 119 struct ucred *, struct sysmsg *); 120 static __inline int sequential_heuristic(struct uio *, struct file *); 121 122 extern struct lock devfs_lock; 123 124 /* 125 * devfs vnode operations for regular files. All vnode ops are MPSAFE. 126 */ 127 struct vop_ops devfs_vnode_norm_vops = { 128 .vop_default = vop_defaultop, 129 .vop_access = devfs_vop_access, 130 .vop_advlock = DEVFS_BADOP, 131 .vop_bmap = DEVFS_BADOP, 132 .vop_close = vop_stdclose, 133 .vop_getattr = devfs_vop_getattr, 134 .vop_inactive = devfs_vop_inactive, 135 .vop_ncreate = DEVFS_BADOP, 136 .vop_nresolve = devfs_vop_nresolve, 137 .vop_nlookupdotdot = devfs_vop_nlookupdotdot, 138 .vop_nlink = DEVFS_BADOP, 139 .vop_nmkdir = devfs_vop_nmkdir, 140 .vop_nmknod = DEVFS_BADOP, 141 .vop_nremove = devfs_vop_nremove, 142 .vop_nrename = DEVFS_BADOP, 143 .vop_nrmdir = devfs_vop_nrmdir, 144 .vop_nsymlink = devfs_vop_nsymlink, 145 .vop_open = vop_stdopen, 146 .vop_pathconf = vop_stdpathconf, 147 .vop_print = devfs_vop_print, 148 .vop_read = DEVFS_BADOP, 149 .vop_readdir = devfs_vop_readdir, 150 .vop_readlink = devfs_vop_readlink, 151 .vop_reclaim = devfs_vop_reclaim, 152 .vop_setattr = devfs_vop_setattr, 153 .vop_write = DEVFS_BADOP, 154 .vop_ioctl = DEVFS_BADOP 155 }; 156 157 /* 158 * devfs vnode operations for character devices. All vnode ops are MPSAFE. 159 */ 160 struct vop_ops devfs_vnode_dev_vops = { 161 .vop_default = vop_defaultop, 162 .vop_access = devfs_vop_access, 163 .vop_advlock = devfs_spec_advlock, 164 .vop_bmap = devfs_spec_bmap, 165 .vop_close = devfs_spec_close, 166 .vop_freeblks = devfs_spec_freeblks, 167 .vop_fsync = devfs_spec_fsync, 168 .vop_getattr = devfs_vop_getattr, 169 .vop_getpages = devfs_spec_getpages, 170 .vop_inactive = devfs_vop_inactive, 171 .vop_open = devfs_spec_open, 172 .vop_pathconf = vop_stdpathconf, 173 .vop_print = devfs_vop_print, 174 .vop_kqfilter = devfs_spec_kqfilter, 175 .vop_read = devfs_spec_read, 176 .vop_readdir = DEVFS_BADOP, 177 .vop_readlink = DEVFS_BADOP, 178 .vop_reclaim = devfs_vop_reclaim, 179 .vop_setattr = devfs_vop_setattr, 180 .vop_strategy = devfs_spec_strategy, 181 .vop_write = devfs_spec_write, 182 .vop_ioctl = devfs_spec_ioctl 183 }; 184 185 /* 186 * devfs file pointer operations. All fileops are MPSAFE. 187 */ 188 struct vop_ops *devfs_vnode_dev_vops_p = &devfs_vnode_dev_vops; 189 190 struct fileops devfs_dev_fileops = { 191 .fo_read = devfs_fo_read, 192 .fo_write = devfs_fo_write, 193 .fo_ioctl = devfs_fo_ioctl, 194 .fo_kqfilter = devfs_fo_kqfilter, 195 .fo_stat = devfs_fo_stat, 196 .fo_close = devfs_fo_close, 197 .fo_shutdown = nofo_shutdown 198 }; 199 200 /* 201 * These two functions are possibly temporary hacks for devices (aka 202 * the pty code) which want to control the node attributes themselves. 203 * 204 * XXX we may ultimately desire to simply remove the uid/gid/mode 205 * from the node entirely. 206 * 207 * MPSAFE - sorta. Theoretically the overwrite can compete since they 208 * are loading from the same fields. 209 */ 210 static __inline void 211 node_sync_dev_get(struct devfs_node *node) 212 { 213 cdev_t dev; 214 215 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) { 216 node->uid = dev->si_uid; 217 node->gid = dev->si_gid; 218 node->mode = dev->si_perms; 219 } 220 } 221 222 static __inline void 223 node_sync_dev_set(struct devfs_node *node) 224 { 225 cdev_t dev; 226 227 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) { 228 dev->si_uid = node->uid; 229 dev->si_gid = node->gid; 230 dev->si_perms = node->mode; 231 } 232 } 233 234 /* 235 * generic entry point for unsupported operations 236 */ 237 static int 238 devfs_vop_badop(struct vop_generic_args *ap) 239 { 240 return (EIO); 241 } 242 243 244 static int 245 devfs_vop_access(struct vop_access_args *ap) 246 { 247 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 248 int error; 249 250 if (!devfs_node_is_accessible(node)) 251 return ENOENT; 252 node_sync_dev_get(node); 253 error = vop_helper_access(ap, node->uid, node->gid, 254 node->mode, node->flags); 255 256 return error; 257 } 258 259 260 static int 261 devfs_vop_inactive(struct vop_inactive_args *ap) 262 { 263 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 264 265 if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0) 266 vrecycle(ap->a_vp); 267 return 0; 268 } 269 270 271 static int 272 devfs_vop_reclaim(struct vop_reclaim_args *ap) 273 { 274 struct devfs_node *node; 275 struct vnode *vp; 276 int locked; 277 278 /* 279 * Check if it is locked already. if not, we acquire the devfs lock 280 */ 281 if ((lockstatus(&devfs_lock, curthread)) != LK_EXCLUSIVE) { 282 lockmgr(&devfs_lock, LK_EXCLUSIVE); 283 locked = 1; 284 } else { 285 locked = 0; 286 } 287 288 /* 289 * Get rid of the devfs_node if it is no longer linked into the 290 * topology. 291 */ 292 vp = ap->a_vp; 293 if ((node = DEVFS_NODE(vp)) != NULL) { 294 node->v_node = NULL; 295 if ((node->flags & DEVFS_NODE_LINKED) == 0) 296 devfs_freep(node); 297 } 298 299 if (locked) 300 lockmgr(&devfs_lock, LK_RELEASE); 301 302 /* 303 * v_rdev needs to be properly released using v_release_rdev 304 * Make sure v_data is NULL as well. 305 */ 306 vp->v_data = NULL; 307 v_release_rdev(vp); 308 return 0; 309 } 310 311 312 static int 313 devfs_vop_readdir(struct vop_readdir_args *ap) 314 { 315 struct devfs_node *dnode = DEVFS_NODE(ap->a_vp); 316 struct devfs_node *node; 317 int cookie_index; 318 int ncookies; 319 int error2; 320 int error; 321 int r; 322 off_t *cookies; 323 off_t saveoff; 324 325 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n"); 326 327 if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX) 328 return (EINVAL); 329 error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM); 330 if (error) 331 return (error); 332 333 if (!devfs_node_is_accessible(dnode)) { 334 vn_unlock(ap->a_vp); 335 return ENOENT; 336 } 337 338 lockmgr(&devfs_lock, LK_EXCLUSIVE); 339 340 saveoff = ap->a_uio->uio_offset; 341 342 if (ap->a_ncookies) { 343 ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */ 344 if (ncookies > 256) 345 ncookies = 256; 346 cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK); 347 cookie_index = 0; 348 } else { 349 ncookies = -1; 350 cookies = NULL; 351 cookie_index = 0; 352 } 353 354 nanotime(&dnode->atime); 355 356 if (saveoff == 0) { 357 r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino, 358 DT_DIR, 1, "."); 359 if (r) 360 goto done; 361 if (cookies) 362 cookies[cookie_index] = saveoff; 363 saveoff++; 364 cookie_index++; 365 if (cookie_index == ncookies) 366 goto done; 367 } 368 369 if (saveoff == 1) { 370 if (dnode->parent) { 371 r = vop_write_dirent(&error, ap->a_uio, 372 dnode->parent->d_dir.d_ino, 373 DT_DIR, 2, ".."); 374 } else { 375 r = vop_write_dirent(&error, ap->a_uio, 376 dnode->d_dir.d_ino, 377 DT_DIR, 2, ".."); 378 } 379 if (r) 380 goto done; 381 if (cookies) 382 cookies[cookie_index] = saveoff; 383 saveoff++; 384 cookie_index++; 385 if (cookie_index == ncookies) 386 goto done; 387 } 388 389 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { 390 if ((node->flags & DEVFS_HIDDEN) || 391 (node->flags & DEVFS_INVISIBLE)) { 392 continue; 393 } 394 395 /* 396 * If the node type is a valid devfs alias, then we make 397 * sure that the target isn't hidden. If it is, we don't 398 * show the link in the directory listing. 399 */ 400 if ((node->node_type == Nlink) && (node->link_target != NULL) && 401 (node->link_target->flags & DEVFS_HIDDEN)) 402 continue; 403 404 if (node->cookie < saveoff) 405 continue; 406 407 saveoff = node->cookie; 408 409 error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino, 410 node->d_dir.d_type, 411 node->d_dir.d_namlen, 412 node->d_dir.d_name); 413 414 if (error2) 415 break; 416 417 saveoff++; 418 419 if (cookies) 420 cookies[cookie_index] = node->cookie; 421 ++cookie_index; 422 if (cookie_index == ncookies) 423 break; 424 } 425 426 done: 427 lockmgr(&devfs_lock, LK_RELEASE); 428 vn_unlock(ap->a_vp); 429 430 ap->a_uio->uio_offset = saveoff; 431 if (error && cookie_index == 0) { 432 if (cookies) { 433 kfree(cookies, M_TEMP); 434 *ap->a_ncookies = 0; 435 *ap->a_cookies = NULL; 436 } 437 } else { 438 if (cookies) { 439 *ap->a_ncookies = cookie_index; 440 *ap->a_cookies = cookies; 441 } 442 } 443 return (error); 444 } 445 446 447 static int 448 devfs_vop_nresolve(struct vop_nresolve_args *ap) 449 { 450 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); 451 struct devfs_node *node, *found = NULL; 452 struct namecache *ncp; 453 struct vnode *vp = NULL; 454 int error = 0; 455 int len; 456 int depth; 457 458 ncp = ap->a_nch->ncp; 459 len = ncp->nc_nlen; 460 461 if (!devfs_node_is_accessible(dnode)) 462 return ENOENT; 463 464 lockmgr(&devfs_lock, LK_EXCLUSIVE); 465 466 if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) { 467 error = ENOENT; 468 cache_setvp(ap->a_nch, NULL); 469 goto out; 470 } 471 472 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { 473 if (len == node->d_dir.d_namlen) { 474 if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) { 475 found = node; 476 break; 477 } 478 } 479 } 480 481 if (found) { 482 depth = 0; 483 while ((found->node_type == Nlink) && (found->link_target)) { 484 if (depth >= 8) { 485 devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8"); 486 break; 487 } 488 489 found = found->link_target; 490 ++depth; 491 } 492 493 if (!(found->flags & DEVFS_HIDDEN)) 494 devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found); 495 } 496 497 if (vp == NULL) { 498 error = ENOENT; 499 cache_setvp(ap->a_nch, NULL); 500 goto out; 501 502 } 503 KKASSERT(vp); 504 vn_unlock(vp); 505 cache_setvp(ap->a_nch, vp); 506 vrele(vp); 507 out: 508 lockmgr(&devfs_lock, LK_RELEASE); 509 510 return error; 511 } 512 513 514 static int 515 devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap) 516 { 517 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); 518 519 *ap->a_vpp = NULL; 520 if (!devfs_node_is_accessible(dnode)) 521 return ENOENT; 522 523 lockmgr(&devfs_lock, LK_EXCLUSIVE); 524 if (dnode->parent != NULL) { 525 devfs_allocv(ap->a_vpp, dnode->parent); 526 vn_unlock(*ap->a_vpp); 527 } 528 lockmgr(&devfs_lock, LK_RELEASE); 529 530 return ((*ap->a_vpp == NULL) ? ENOENT : 0); 531 } 532 533 534 static int 535 devfs_vop_getattr(struct vop_getattr_args *ap) 536 { 537 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 538 struct vattr *vap = ap->a_vap; 539 struct partinfo pinfo; 540 int error = 0; 541 542 #if 0 543 if (!devfs_node_is_accessible(node)) 544 return ENOENT; 545 #endif 546 node_sync_dev_get(node); 547 548 lockmgr(&devfs_lock, LK_EXCLUSIVE); 549 550 /* start by zeroing out the attributes */ 551 VATTR_NULL(vap); 552 553 /* next do all the common fields */ 554 vap->va_type = ap->a_vp->v_type; 555 vap->va_mode = node->mode; 556 vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ; 557 vap->va_flags = 0; 558 vap->va_blocksize = DEV_BSIZE; 559 vap->va_bytes = vap->va_size = 0; 560 561 vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0]; 562 563 vap->va_atime = node->atime; 564 vap->va_mtime = node->mtime; 565 vap->va_ctime = node->ctime; 566 567 vap->va_nlink = 1; /* number of references to file */ 568 569 vap->va_uid = node->uid; 570 vap->va_gid = node->gid; 571 572 vap->va_rmajor = 0; 573 vap->va_rminor = 0; 574 575 if ((node->node_type == Ndev) && node->d_dev) { 576 reference_dev(node->d_dev); 577 vap->va_rminor = node->d_dev->si_uminor; 578 release_dev(node->d_dev); 579 } 580 581 /* For a softlink the va_size is the length of the softlink */ 582 if (node->symlink_name != 0) { 583 vap->va_bytes = vap->va_size = node->symlink_namelen; 584 } 585 586 /* 587 * For a disk-type device, va_size is the size of the underlying 588 * device, so that lseek() works properly. 589 */ 590 if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) { 591 bzero(&pinfo, sizeof(pinfo)); 592 error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo, 593 0, proc0.p_ucred, NULL); 594 if ((error == 0) && (pinfo.media_blksize != 0)) { 595 vap->va_size = pinfo.media_size; 596 } else { 597 vap->va_size = 0; 598 error = 0; 599 } 600 } 601 602 lockmgr(&devfs_lock, LK_RELEASE); 603 604 return (error); 605 } 606 607 608 static int 609 devfs_vop_setattr(struct vop_setattr_args *ap) 610 { 611 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 612 struct vattr *vap; 613 uid_t cur_uid; 614 gid_t cur_gid; 615 mode_t cur_mode; 616 int error = 0; 617 618 if (!devfs_node_is_accessible(node)) 619 return ENOENT; 620 node_sync_dev_get(node); 621 622 lockmgr(&devfs_lock, LK_EXCLUSIVE); 623 624 vap = ap->a_vap; 625 626 if ((vap->va_uid != (uid_t)VNOVAL) || (vap->va_gid != (gid_t)VNOVAL)) { 627 cur_uid = node->uid; 628 cur_gid = node->gid; 629 cur_mode = node->mode; 630 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid, 631 ap->a_cred, &cur_uid, &cur_gid, &cur_mode); 632 if (error) 633 goto out; 634 635 if (node->uid != cur_uid || node->gid != cur_gid) { 636 node->uid = cur_uid; 637 node->gid = cur_gid; 638 node->mode = cur_mode; 639 } 640 } 641 642 if (vap->va_mode != (mode_t)VNOVAL) { 643 cur_mode = node->mode; 644 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred, 645 node->uid, node->gid, &cur_mode); 646 if (error == 0 && node->mode != cur_mode) { 647 node->mode = cur_mode; 648 } 649 } 650 651 out: 652 node_sync_dev_set(node); 653 nanotime(&node->ctime); 654 lockmgr(&devfs_lock, LK_RELEASE); 655 656 return error; 657 } 658 659 660 static int 661 devfs_vop_readlink(struct vop_readlink_args *ap) 662 { 663 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 664 int ret; 665 666 if (!devfs_node_is_accessible(node)) 667 return ENOENT; 668 669 lockmgr(&devfs_lock, LK_EXCLUSIVE); 670 ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio); 671 lockmgr(&devfs_lock, LK_RELEASE); 672 673 return ret; 674 } 675 676 677 static int 678 devfs_vop_print(struct vop_print_args *ap) 679 { 680 return (0); 681 } 682 683 static int 684 devfs_vop_nmkdir(struct vop_nmkdir_args *ap) 685 { 686 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); 687 struct devfs_node *node; 688 689 if (!devfs_node_is_accessible(dnode)) 690 return ENOENT; 691 692 if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) 693 goto out; 694 695 lockmgr(&devfs_lock, LK_EXCLUSIVE); 696 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Ndir, 697 ap->a_nch->ncp->nc_name, dnode, NULL); 698 699 if (*ap->a_vpp) { 700 node = DEVFS_NODE(*ap->a_vpp); 701 node->flags |= DEVFS_USER_CREATED; 702 cache_setunresolved(ap->a_nch); 703 cache_setvp(ap->a_nch, *ap->a_vpp); 704 } 705 lockmgr(&devfs_lock, LK_RELEASE); 706 out: 707 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0); 708 } 709 710 static int 711 devfs_vop_nsymlink(struct vop_nsymlink_args *ap) 712 { 713 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); 714 struct devfs_node *node; 715 size_t targetlen; 716 717 if (!devfs_node_is_accessible(dnode)) 718 return ENOENT; 719 720 ap->a_vap->va_type = VLNK; 721 722 if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) 723 goto out; 724 725 lockmgr(&devfs_lock, LK_EXCLUSIVE); 726 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Nlink, 727 ap->a_nch->ncp->nc_name, dnode, NULL); 728 729 targetlen = strlen(ap->a_target); 730 if (*ap->a_vpp) { 731 node = DEVFS_NODE(*ap->a_vpp); 732 node->flags |= DEVFS_USER_CREATED; 733 node->symlink_namelen = targetlen; 734 node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK); 735 memcpy(node->symlink_name, ap->a_target, targetlen); 736 node->symlink_name[targetlen] = '\0'; 737 cache_setunresolved(ap->a_nch); 738 cache_setvp(ap->a_nch, *ap->a_vpp); 739 } 740 lockmgr(&devfs_lock, LK_RELEASE); 741 out: 742 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0); 743 } 744 745 static int 746 devfs_vop_nrmdir(struct vop_nrmdir_args *ap) 747 { 748 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); 749 struct devfs_node *node; 750 struct namecache *ncp; 751 int error = ENOENT; 752 753 ncp = ap->a_nch->ncp; 754 755 if (!devfs_node_is_accessible(dnode)) 756 return ENOENT; 757 758 lockmgr(&devfs_lock, LK_EXCLUSIVE); 759 760 if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) 761 goto out; 762 763 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { 764 if (ncp->nc_nlen != node->d_dir.d_namlen) 765 continue; 766 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen)) 767 continue; 768 769 /* 770 * only allow removal of user created dirs 771 */ 772 if ((node->flags & DEVFS_USER_CREATED) == 0) { 773 error = EPERM; 774 goto out; 775 } else if (node->node_type != Ndir) { 776 error = ENOTDIR; 777 goto out; 778 } else if (node->nchildren > 2) { 779 error = ENOTEMPTY; 780 goto out; 781 } else { 782 if (node->v_node) 783 cache_inval_vp(node->v_node, CINV_DESTROY); 784 devfs_unlinkp(node); 785 error = 0; 786 break; 787 } 788 } 789 790 cache_unlink(ap->a_nch); 791 out: 792 lockmgr(&devfs_lock, LK_RELEASE); 793 return error; 794 } 795 796 static int 797 devfs_vop_nremove(struct vop_nremove_args *ap) 798 { 799 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); 800 struct devfs_node *node; 801 struct namecache *ncp; 802 int error = ENOENT; 803 804 ncp = ap->a_nch->ncp; 805 806 if (!devfs_node_is_accessible(dnode)) 807 return ENOENT; 808 809 lockmgr(&devfs_lock, LK_EXCLUSIVE); 810 811 if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) 812 goto out; 813 814 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { 815 if (ncp->nc_nlen != node->d_dir.d_namlen) 816 continue; 817 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen)) 818 continue; 819 820 /* 821 * only allow removal of user created stuff (e.g. symlinks) 822 */ 823 if ((node->flags & DEVFS_USER_CREATED) == 0) { 824 error = EPERM; 825 goto out; 826 } else if (node->node_type == Ndir) { 827 error = EISDIR; 828 goto out; 829 } else { 830 if (node->v_node) 831 cache_inval_vp(node->v_node, CINV_DESTROY); 832 devfs_unlinkp(node); 833 error = 0; 834 break; 835 } 836 } 837 838 cache_unlink(ap->a_nch); 839 out: 840 lockmgr(&devfs_lock, LK_RELEASE); 841 return error; 842 } 843 844 845 static int 846 devfs_spec_open(struct vop_open_args *ap) 847 { 848 struct vnode *vp = ap->a_vp; 849 struct vnode *orig_vp = NULL; 850 struct devfs_node *node = DEVFS_NODE(vp); 851 struct devfs_node *newnode; 852 cdev_t dev, ndev = NULL; 853 int error = 0; 854 855 if (node) { 856 if (node->d_dev == NULL) 857 return ENXIO; 858 if (!devfs_node_is_accessible(node)) 859 return ENOENT; 860 } 861 862 if ((dev = vp->v_rdev) == NULL) 863 return ENXIO; 864 865 vn_lock(vp, LK_UPGRADE | LK_RETRY); 866 867 if (node && ap->a_fp) { 868 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n"); 869 lockmgr(&devfs_lock, LK_EXCLUSIVE); 870 871 ndev = devfs_clone(dev, node->d_dir.d_name, 872 node->d_dir.d_namlen, 873 ap->a_mode, ap->a_cred); 874 if (ndev != NULL) { 875 newnode = devfs_create_device_node( 876 DEVFS_MNTDATA(vp->v_mount)->root_node, 877 ndev, NULL, NULL); 878 /* XXX: possibly destroy device if this happens */ 879 880 if (newnode != NULL) { 881 dev = ndev; 882 devfs_link_dev(dev); 883 884 devfs_debug(DEVFS_DEBUG_DEBUG, 885 "parent here is: %s, node is: |%s|\n", 886 ((node->parent->node_type == Nroot) ? 887 "ROOT!" : node->parent->d_dir.d_name), 888 newnode->d_dir.d_name); 889 devfs_debug(DEVFS_DEBUG_DEBUG, 890 "test: %s\n", 891 ((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name); 892 893 /* 894 * orig_vp is set to the original vp if we cloned. 895 */ 896 /* node->flags |= DEVFS_CLONED; */ 897 devfs_allocv(&vp, newnode); 898 orig_vp = ap->a_vp; 899 ap->a_vp = vp; 900 } 901 } 902 lockmgr(&devfs_lock, LK_RELEASE); 903 } 904 905 devfs_debug(DEVFS_DEBUG_DEBUG, 906 "devfs_spec_open() called on %s! \n", 907 dev->si_name); 908 909 /* 910 * Make this field valid before any I/O in ->d_open 911 */ 912 if (!dev->si_iosize_max) 913 /* XXX: old DFLTPHYS == 64KB dependency */ 914 dev->si_iosize_max = min(MAXPHYS,64*1024); 915 916 if (dev_dflags(dev) & D_TTY) 917 vsetflags(vp, VISTTY); 918 919 /* 920 * Open underlying device 921 */ 922 vn_unlock(vp); 923 error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred); 924 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 925 926 /* 927 * Clean up any cloned vp if we error out. 928 */ 929 if (error) { 930 if (orig_vp) { 931 vput(vp); 932 ap->a_vp = orig_vp; 933 /* orig_vp = NULL; */ 934 } 935 return error; 936 } 937 938 /* 939 * This checks if the disk device is going to be opened for writing. 940 * It will be only allowed in the cases where securelevel permits it 941 * and it's not mounted R/W. 942 */ 943 if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) && 944 (ap->a_cred != FSCRED)) { 945 946 /* Very secure mode. No open for writing allowed */ 947 if (securelevel >= 2) 948 return EPERM; 949 950 /* 951 * If it is mounted R/W, do not allow to open for writing. 952 * In the case it's mounted read-only but securelevel 953 * is >= 1, then do not allow opening for writing either. 954 */ 955 if (vfs_mountedon(vp)) { 956 if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY)) 957 return EBUSY; 958 else if (securelevel >= 1) 959 return EPERM; 960 } 961 } 962 963 if (dev_dflags(dev) & D_TTY) { 964 if (dev->si_tty) { 965 struct tty *tp; 966 tp = dev->si_tty; 967 if (!tp->t_stop) { 968 devfs_debug(DEVFS_DEBUG_DEBUG, 969 "devfs: no t_stop\n"); 970 tp->t_stop = nottystop; 971 } 972 } 973 } 974 975 976 if (vn_isdisk(vp, NULL)) { 977 if (!dev->si_bsize_phys) 978 dev->si_bsize_phys = DEV_BSIZE; 979 vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1); 980 } 981 982 vop_stdopen(ap); 983 #if 0 984 if (node) 985 nanotime(&node->atime); 986 #endif 987 988 /* 989 * If we replaced the vp the vop_stdopen() call will have loaded 990 * it into fp->f_data and vref()d the vp, giving us two refs. So 991 * instead of just unlocking it here we have to vput() it. 992 */ 993 if (orig_vp) 994 vput(vp); 995 996 /* Ugly pty magic, to make pty devices appear once they are opened */ 997 if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY) 998 node->flags &= ~DEVFS_INVISIBLE; 999 1000 if (ap->a_fp) { 1001 KKASSERT(ap->a_fp->f_type == DTYPE_VNODE); 1002 KKASSERT((ap->a_fp->f_flag & FMASK) == (ap->a_mode & FMASK)); 1003 ap->a_fp->f_ops = &devfs_dev_fileops; 1004 KKASSERT(ap->a_fp->f_data == (void *)vp); 1005 } 1006 1007 return 0; 1008 } 1009 1010 static int 1011 devfs_spec_close(struct vop_close_args *ap) 1012 { 1013 struct devfs_node *node; 1014 struct proc *p = curproc; 1015 struct vnode *vp = ap->a_vp; 1016 cdev_t dev = vp->v_rdev; 1017 int error = 0; 1018 int needrelock; 1019 1020 /* 1021 * We do special tests on the opencount so unfortunately we need 1022 * an exclusive lock. 1023 */ 1024 vn_lock(vp, LK_UPGRADE | LK_RETRY); 1025 1026 if (dev) 1027 devfs_debug(DEVFS_DEBUG_DEBUG, 1028 "devfs_spec_close() called on %s! \n", 1029 dev->si_name); 1030 else 1031 devfs_debug(DEVFS_DEBUG_DEBUG, 1032 "devfs_spec_close() called, null vode!\n"); 1033 1034 /* 1035 * A couple of hacks for devices and tty devices. The 1036 * vnode ref count cannot be used to figure out the 1037 * last close, but we can use v_opencount now that 1038 * revoke works properly. 1039 * 1040 * Detect the last close on a controlling terminal and clear 1041 * the session (half-close). 1042 */ 1043 if (dev) 1044 reference_dev(dev); 1045 1046 if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) { 1047 p->p_session->s_ttyvp = NULL; 1048 vrele(vp); 1049 } 1050 1051 /* 1052 * Vnodes can be opened and closed multiple times. Do not really 1053 * close the device unless (1) it is being closed forcibly, 1054 * (2) the device wants to track closes, or (3) this is the last 1055 * vnode doing its last close on the device. 1056 * 1057 * XXX the VXLOCK (force close) case can leave vnodes referencing 1058 * a closed device. This might not occur now that our revoke is 1059 * fixed. 1060 */ 1061 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n"); 1062 if (dev && ((vp->v_flag & VRECLAIMED) || 1063 (dev_dflags(dev) & D_TRACKCLOSE) || 1064 (vp->v_opencount == 1))) { 1065 /* 1066 * Ugly pty magic, to make pty devices disappear again once 1067 * they are closed. 1068 */ 1069 node = DEVFS_NODE(ap->a_vp); 1070 if (node && (node->flags & DEVFS_PTY)) 1071 node->flags |= DEVFS_INVISIBLE; 1072 1073 /* 1074 * Unlock around dev_dclose(), unless the vnode is 1075 * undergoing a vgone/reclaim (during umount). 1076 */ 1077 needrelock = 0; 1078 if ((vp->v_flag & VRECLAIMED) == 0 && vn_islocked(vp)) { 1079 needrelock = 1; 1080 vn_unlock(vp); 1081 } 1082 1083 /* 1084 * WARNING! If the device destroys itself the devfs node 1085 * can disappear here. 1086 * 1087 * WARNING! vn_lock() will fail if the vp is in a VRECLAIM, 1088 * which can occur during umount. 1089 */ 1090 error = dev_dclose(dev, ap->a_fflag, S_IFCHR); 1091 /* node is now stale */ 1092 1093 if (needrelock) { 1094 if (vn_lock(vp, LK_EXCLUSIVE | 1095 LK_RETRY | 1096 LK_FAILRECLAIM) != 0) { 1097 panic("devfs_spec_close: vnode %p " 1098 "unexpectedly could not be relocked", 1099 vp); 1100 } 1101 } 1102 } else { 1103 error = 0; 1104 } 1105 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n"); 1106 1107 /* 1108 * Track the actual opens and closes on the vnode. The last close 1109 * disassociates the rdev. If the rdev is already disassociated or 1110 * the opencount is already 0, the vnode might have been revoked 1111 * and no further opencount tracking occurs. 1112 */ 1113 if (dev) 1114 release_dev(dev); 1115 if (vp->v_opencount > 0) 1116 vop_stdclose(ap); 1117 return(error); 1118 1119 } 1120 1121 1122 static int 1123 devfs_fo_close(struct file *fp) 1124 { 1125 struct vnode *vp = (struct vnode *)fp->f_data; 1126 int error; 1127 1128 fp->f_ops = &badfileops; 1129 error = vn_close(vp, fp->f_flag); 1130 1131 return (error); 1132 } 1133 1134 1135 /* 1136 * Device-optimized file table vnode read routine. 1137 * 1138 * This bypasses the VOP table and talks directly to the device. Most 1139 * filesystems just route to specfs and can make this optimization. 1140 * 1141 * MPALMOSTSAFE - acquires mplock 1142 */ 1143 static int 1144 devfs_fo_read(struct file *fp, struct uio *uio, 1145 struct ucred *cred, int flags) 1146 { 1147 struct devfs_node *node; 1148 struct vnode *vp; 1149 int ioflag; 1150 int error; 1151 cdev_t dev; 1152 1153 KASSERT(uio->uio_td == curthread, 1154 ("uio_td %p is not td %p", uio->uio_td, curthread)); 1155 1156 if (uio->uio_resid == 0) 1157 return 0; 1158 1159 vp = (struct vnode *)fp->f_data; 1160 if (vp == NULL || vp->v_type == VBAD) 1161 return EBADF; 1162 1163 node = DEVFS_NODE(vp); 1164 1165 if ((dev = vp->v_rdev) == NULL) 1166 return EBADF; 1167 1168 reference_dev(dev); 1169 1170 if ((flags & O_FOFFSET) == 0) 1171 uio->uio_offset = fp->f_offset; 1172 1173 ioflag = 0; 1174 if (flags & O_FBLOCKING) { 1175 /* ioflag &= ~IO_NDELAY; */ 1176 } else if (flags & O_FNONBLOCKING) { 1177 ioflag |= IO_NDELAY; 1178 } else if (fp->f_flag & FNONBLOCK) { 1179 ioflag |= IO_NDELAY; 1180 } 1181 if (flags & O_FBUFFERED) { 1182 /* ioflag &= ~IO_DIRECT; */ 1183 } else if (flags & O_FUNBUFFERED) { 1184 ioflag |= IO_DIRECT; 1185 } else if (fp->f_flag & O_DIRECT) { 1186 ioflag |= IO_DIRECT; 1187 } 1188 ioflag |= sequential_heuristic(uio, fp); 1189 1190 error = dev_dread(dev, uio, ioflag); 1191 1192 release_dev(dev); 1193 if (node) 1194 nanotime(&node->atime); 1195 if ((flags & O_FOFFSET) == 0) 1196 fp->f_offset = uio->uio_offset; 1197 fp->f_nextoff = uio->uio_offset; 1198 1199 return (error); 1200 } 1201 1202 1203 static int 1204 devfs_fo_write(struct file *fp, struct uio *uio, 1205 struct ucred *cred, int flags) 1206 { 1207 struct devfs_node *node; 1208 struct vnode *vp; 1209 int ioflag; 1210 int error; 1211 cdev_t dev; 1212 1213 KASSERT(uio->uio_td == curthread, 1214 ("uio_td %p is not p %p", uio->uio_td, curthread)); 1215 1216 vp = (struct vnode *)fp->f_data; 1217 if (vp == NULL || vp->v_type == VBAD) 1218 return EBADF; 1219 1220 node = DEVFS_NODE(vp); 1221 1222 if (vp->v_type == VREG) 1223 bwillwrite(uio->uio_resid); 1224 1225 vp = (struct vnode *)fp->f_data; 1226 1227 if ((dev = vp->v_rdev) == NULL) 1228 return EBADF; 1229 1230 reference_dev(dev); 1231 1232 if ((flags & O_FOFFSET) == 0) 1233 uio->uio_offset = fp->f_offset; 1234 1235 ioflag = IO_UNIT; 1236 if (vp->v_type == VREG && 1237 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 1238 ioflag |= IO_APPEND; 1239 } 1240 1241 if (flags & O_FBLOCKING) { 1242 /* ioflag &= ~IO_NDELAY; */ 1243 } else if (flags & O_FNONBLOCKING) { 1244 ioflag |= IO_NDELAY; 1245 } else if (fp->f_flag & FNONBLOCK) { 1246 ioflag |= IO_NDELAY; 1247 } 1248 if (flags & O_FBUFFERED) { 1249 /* ioflag &= ~IO_DIRECT; */ 1250 } else if (flags & O_FUNBUFFERED) { 1251 ioflag |= IO_DIRECT; 1252 } else if (fp->f_flag & O_DIRECT) { 1253 ioflag |= IO_DIRECT; 1254 } 1255 if (flags & O_FASYNCWRITE) { 1256 /* ioflag &= ~IO_SYNC; */ 1257 } else if (flags & O_FSYNCWRITE) { 1258 ioflag |= IO_SYNC; 1259 } else if (fp->f_flag & O_FSYNC) { 1260 ioflag |= IO_SYNC; 1261 } 1262 1263 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 1264 ioflag |= IO_SYNC; 1265 ioflag |= sequential_heuristic(uio, fp); 1266 1267 error = dev_dwrite(dev, uio, ioflag); 1268 1269 release_dev(dev); 1270 if (node) { 1271 nanotime(&node->atime); 1272 nanotime(&node->mtime); 1273 } 1274 1275 if ((flags & O_FOFFSET) == 0) 1276 fp->f_offset = uio->uio_offset; 1277 fp->f_nextoff = uio->uio_offset; 1278 1279 return (error); 1280 } 1281 1282 1283 static int 1284 devfs_fo_stat(struct file *fp, struct stat *sb, struct ucred *cred) 1285 { 1286 struct vnode *vp; 1287 struct vattr vattr; 1288 struct vattr *vap; 1289 u_short mode; 1290 cdev_t dev; 1291 int error; 1292 1293 vp = (struct vnode *)fp->f_data; 1294 if (vp == NULL || vp->v_type == VBAD) 1295 return EBADF; 1296 1297 error = vn_stat(vp, sb, cred); 1298 if (error) 1299 return (error); 1300 1301 vap = &vattr; 1302 error = VOP_GETATTR(vp, vap); 1303 if (error) 1304 return (error); 1305 1306 /* 1307 * Zero the spare stat fields 1308 */ 1309 sb->st_lspare = 0; 1310 sb->st_qspare1 = 0; 1311 sb->st_qspare2 = 0; 1312 1313 /* 1314 * Copy from vattr table ... or not in case it's a cloned device 1315 */ 1316 if (vap->va_fsid != VNOVAL) 1317 sb->st_dev = vap->va_fsid; 1318 else 1319 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 1320 1321 sb->st_ino = vap->va_fileid; 1322 1323 mode = vap->va_mode; 1324 mode |= S_IFCHR; 1325 sb->st_mode = mode; 1326 1327 if (vap->va_nlink > (nlink_t)-1) 1328 sb->st_nlink = (nlink_t)-1; 1329 else 1330 sb->st_nlink = vap->va_nlink; 1331 1332 sb->st_uid = vap->va_uid; 1333 sb->st_gid = vap->va_gid; 1334 sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev); 1335 sb->st_size = vap->va_bytes; 1336 sb->st_atimespec = vap->va_atime; 1337 sb->st_mtimespec = vap->va_mtime; 1338 sb->st_ctimespec = vap->va_ctime; 1339 1340 /* 1341 * A VCHR and VBLK device may track the last access and last modified 1342 * time independantly of the filesystem. This is particularly true 1343 * because device read and write calls may bypass the filesystem. 1344 */ 1345 if (vp->v_type == VCHR || vp->v_type == VBLK) { 1346 dev = vp->v_rdev; 1347 if (dev != NULL) { 1348 if (dev->si_lastread) { 1349 sb->st_atimespec.tv_sec = time_second + 1350 (time_uptime - 1351 dev->si_lastread); 1352 sb->st_atimespec.tv_nsec = 0; 1353 } 1354 if (dev->si_lastwrite) { 1355 sb->st_atimespec.tv_sec = time_second + 1356 (time_uptime - 1357 dev->si_lastwrite); 1358 sb->st_atimespec.tv_nsec = 0; 1359 } 1360 } 1361 } 1362 1363 /* 1364 * According to www.opengroup.org, the meaning of st_blksize is 1365 * "a filesystem-specific preferred I/O block size for this 1366 * object. In some filesystem types, this may vary from file 1367 * to file" 1368 * Default to PAGE_SIZE after much discussion. 1369 */ 1370 1371 sb->st_blksize = PAGE_SIZE; 1372 1373 sb->st_flags = vap->va_flags; 1374 1375 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0); 1376 if (error) 1377 sb->st_gen = 0; 1378 else 1379 sb->st_gen = (u_int32_t)vap->va_gen; 1380 1381 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 1382 1383 return (0); 1384 } 1385 1386 1387 static int 1388 devfs_fo_kqfilter(struct file *fp, struct knote *kn) 1389 { 1390 struct vnode *vp; 1391 int error; 1392 cdev_t dev; 1393 1394 vp = (struct vnode *)fp->f_data; 1395 if (vp == NULL || vp->v_type == VBAD) { 1396 error = EBADF; 1397 goto done; 1398 } 1399 if ((dev = vp->v_rdev) == NULL) { 1400 error = EBADF; 1401 goto done; 1402 } 1403 reference_dev(dev); 1404 1405 error = dev_dkqfilter(dev, kn); 1406 1407 release_dev(dev); 1408 1409 done: 1410 return (error); 1411 } 1412 1413 /* 1414 * MPALMOSTSAFE - acquires mplock 1415 */ 1416 static int 1417 devfs_fo_ioctl(struct file *fp, u_long com, caddr_t data, 1418 struct ucred *ucred, struct sysmsg *msg) 1419 { 1420 #if 0 1421 struct devfs_node *node; 1422 #endif 1423 struct vnode *vp; 1424 struct vnode *ovp; 1425 cdev_t dev; 1426 int error; 1427 struct fiodname_args *name_args; 1428 size_t namlen; 1429 const char *name; 1430 1431 vp = ((struct vnode *)fp->f_data); 1432 1433 if ((dev = vp->v_rdev) == NULL) 1434 return EBADF; /* device was revoked */ 1435 1436 reference_dev(dev); 1437 1438 #if 0 1439 node = DEVFS_NODE(vp); 1440 #endif 1441 1442 devfs_debug(DEVFS_DEBUG_DEBUG, 1443 "devfs_fo_ioctl() called! for dev %s\n", 1444 dev->si_name); 1445 1446 if (com == FIODTYPE) { 1447 *(int *)data = dev_dflags(dev) & D_TYPEMASK; 1448 error = 0; 1449 goto out; 1450 } else if (com == FIODNAME) { 1451 name_args = (struct fiodname_args *)data; 1452 name = dev->si_name; 1453 namlen = strlen(name) + 1; 1454 1455 devfs_debug(DEVFS_DEBUG_DEBUG, 1456 "ioctl, got: FIODNAME for %s\n", name); 1457 1458 if (namlen <= name_args->len) 1459 error = copyout(dev->si_name, name_args->name, namlen); 1460 else 1461 error = EINVAL; 1462 1463 devfs_debug(DEVFS_DEBUG_DEBUG, 1464 "ioctl stuff: error: %d\n", error); 1465 goto out; 1466 } 1467 1468 error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg); 1469 1470 #if 0 1471 if (node) { 1472 nanotime(&node->atime); 1473 nanotime(&node->mtime); 1474 } 1475 #endif 1476 if (com == TIOCSCTTY) { 1477 devfs_debug(DEVFS_DEBUG_DEBUG, 1478 "devfs_fo_ioctl: got TIOCSCTTY on %s\n", 1479 dev->si_name); 1480 } 1481 if (error == 0 && com == TIOCSCTTY) { 1482 struct proc *p = curthread->td_proc; 1483 struct session *sess; 1484 1485 devfs_debug(DEVFS_DEBUG_DEBUG, 1486 "devfs_fo_ioctl: dealing with TIOCSCTTY on %s\n", 1487 dev->si_name); 1488 if (p == NULL) { 1489 error = ENOTTY; 1490 goto out; 1491 } 1492 sess = p->p_session; 1493 1494 /* 1495 * Do nothing if reassigning same control tty 1496 */ 1497 if (sess->s_ttyvp == vp) { 1498 error = 0; 1499 goto out; 1500 } 1501 1502 /* 1503 * Get rid of reference to old control tty 1504 */ 1505 ovp = sess->s_ttyvp; 1506 vref(vp); 1507 sess->s_ttyvp = vp; 1508 if (ovp) 1509 vrele(ovp); 1510 } 1511 1512 out: 1513 release_dev(dev); 1514 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_fo_ioctl() finished! \n"); 1515 return (error); 1516 } 1517 1518 1519 static int 1520 devfs_spec_fsync(struct vop_fsync_args *ap) 1521 { 1522 struct vnode *vp = ap->a_vp; 1523 int error; 1524 1525 if (!vn_isdisk(vp, NULL)) 1526 return (0); 1527 1528 /* 1529 * Flush all dirty buffers associated with a block device. 1530 */ 1531 error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL); 1532 return (error); 1533 } 1534 1535 static int 1536 devfs_spec_read(struct vop_read_args *ap) 1537 { 1538 struct devfs_node *node; 1539 struct vnode *vp; 1540 struct uio *uio; 1541 cdev_t dev; 1542 int error; 1543 1544 vp = ap->a_vp; 1545 dev = vp->v_rdev; 1546 uio = ap->a_uio; 1547 node = DEVFS_NODE(vp); 1548 1549 if (dev == NULL) /* device was revoked */ 1550 return (EBADF); 1551 if (uio->uio_resid == 0) 1552 return (0); 1553 1554 vn_unlock(vp); 1555 error = dev_dread(dev, uio, ap->a_ioflag); 1556 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1557 1558 if (node) 1559 nanotime(&node->atime); 1560 1561 return (error); 1562 } 1563 1564 /* 1565 * Vnode op for write 1566 * 1567 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1568 * struct ucred *a_cred) 1569 */ 1570 static int 1571 devfs_spec_write(struct vop_write_args *ap) 1572 { 1573 struct devfs_node *node; 1574 struct vnode *vp; 1575 struct uio *uio; 1576 cdev_t dev; 1577 int error; 1578 1579 vp = ap->a_vp; 1580 dev = vp->v_rdev; 1581 uio = ap->a_uio; 1582 node = DEVFS_NODE(vp); 1583 1584 KKASSERT(uio->uio_segflg != UIO_NOCOPY); 1585 1586 if (dev == NULL) /* device was revoked */ 1587 return (EBADF); 1588 1589 vn_unlock(vp); 1590 error = dev_dwrite(dev, uio, ap->a_ioflag); 1591 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1592 1593 if (node) { 1594 nanotime(&node->atime); 1595 nanotime(&node->mtime); 1596 } 1597 1598 return (error); 1599 } 1600 1601 /* 1602 * Device ioctl operation. 1603 * 1604 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data, 1605 * int a_fflag, struct ucred *a_cred, struct sysmsg *msg) 1606 */ 1607 static int 1608 devfs_spec_ioctl(struct vop_ioctl_args *ap) 1609 { 1610 struct vnode *vp = ap->a_vp; 1611 #if 0 1612 struct devfs_node *node; 1613 #endif 1614 cdev_t dev; 1615 1616 if ((dev = vp->v_rdev) == NULL) 1617 return (EBADF); /* device was revoked */ 1618 #if 0 1619 node = DEVFS_NODE(vp); 1620 1621 if (node) { 1622 nanotime(&node->atime); 1623 nanotime(&node->mtime); 1624 } 1625 #endif 1626 1627 return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag, 1628 ap->a_cred, ap->a_sysmsg)); 1629 } 1630 1631 /* 1632 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn) 1633 */ 1634 /* ARGSUSED */ 1635 static int 1636 devfs_spec_kqfilter(struct vop_kqfilter_args *ap) 1637 { 1638 struct vnode *vp = ap->a_vp; 1639 #if 0 1640 struct devfs_node *node; 1641 #endif 1642 cdev_t dev; 1643 1644 if ((dev = vp->v_rdev) == NULL) 1645 return (EBADF); /* device was revoked (EBADF) */ 1646 #if 0 1647 node = DEVFS_NODE(vp); 1648 1649 if (node) 1650 nanotime(&node->atime); 1651 #endif 1652 1653 return (dev_dkqfilter(dev, ap->a_kn)); 1654 } 1655 1656 /* 1657 * Convert a vnode strategy call into a device strategy call. Vnode strategy 1658 * calls are not limited to device DMA limits so we have to deal with the 1659 * case. 1660 * 1661 * spec_strategy(struct vnode *a_vp, struct bio *a_bio) 1662 */ 1663 static int 1664 devfs_spec_strategy(struct vop_strategy_args *ap) 1665 { 1666 struct bio *bio = ap->a_bio; 1667 struct buf *bp = bio->bio_buf; 1668 struct buf *nbp; 1669 struct vnode *vp; 1670 struct mount *mp; 1671 int chunksize; 1672 int maxiosize; 1673 1674 if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL) 1675 buf_start(bp); 1676 1677 /* 1678 * Collect statistics on synchronous and asynchronous read 1679 * and write counts for disks that have associated filesystems. 1680 */ 1681 vp = ap->a_vp; 1682 KKASSERT(vp->v_rdev != NULL); /* XXX */ 1683 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) { 1684 if (bp->b_cmd == BUF_CMD_READ) { 1685 if (bp->b_flags & BIO_SYNC) 1686 mp->mnt_stat.f_syncreads++; 1687 else 1688 mp->mnt_stat.f_asyncreads++; 1689 } else { 1690 if (bp->b_flags & BIO_SYNC) 1691 mp->mnt_stat.f_syncwrites++; 1692 else 1693 mp->mnt_stat.f_asyncwrites++; 1694 } 1695 } 1696 1697 /* 1698 * Device iosize limitations only apply to read and write. Shortcut 1699 * the I/O if it fits. 1700 */ 1701 if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) { 1702 devfs_debug(DEVFS_DEBUG_DEBUG, 1703 "%s: si_iosize_max not set!\n", 1704 dev_dname(vp->v_rdev)); 1705 maxiosize = MAXPHYS; 1706 } 1707 #if SPEC_CHAIN_DEBUG & 2 1708 maxiosize = 4096; 1709 #endif 1710 if (bp->b_bcount <= maxiosize || 1711 (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) { 1712 dev_dstrategy_chain(vp->v_rdev, bio); 1713 return (0); 1714 } 1715 1716 /* 1717 * Clone the buffer and set up an I/O chain to chunk up the I/O. 1718 */ 1719 nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO); 1720 initbufbio(nbp); 1721 buf_dep_init(nbp); 1722 BUF_LOCK(nbp, LK_EXCLUSIVE); 1723 BUF_KERNPROC(nbp); 1724 nbp->b_vp = vp; 1725 nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP); 1726 nbp->b_data = bp->b_data; 1727 nbp->b_bio1.bio_done = devfs_spec_strategy_done; 1728 nbp->b_bio1.bio_offset = bio->bio_offset; 1729 nbp->b_bio1.bio_caller_info1.ptr = bio; 1730 1731 /* 1732 * Start the first transfer 1733 */ 1734 if (vn_isdisk(vp, NULL)) 1735 chunksize = vp->v_rdev->si_bsize_phys; 1736 else 1737 chunksize = DEV_BSIZE; 1738 chunksize = maxiosize / chunksize * chunksize; 1739 #if SPEC_CHAIN_DEBUG & 1 1740 devfs_debug(DEVFS_DEBUG_DEBUG, 1741 "spec_strategy chained I/O chunksize=%d\n", 1742 chunksize); 1743 #endif 1744 nbp->b_cmd = bp->b_cmd; 1745 nbp->b_bcount = chunksize; 1746 nbp->b_bufsize = chunksize; /* used to detect a short I/O */ 1747 nbp->b_bio1.bio_caller_info2.index = chunksize; 1748 1749 #if SPEC_CHAIN_DEBUG & 1 1750 devfs_debug(DEVFS_DEBUG_DEBUG, 1751 "spec_strategy: chain %p offset %d/%d bcount %d\n", 1752 bp, 0, bp->b_bcount, nbp->b_bcount); 1753 #endif 1754 1755 dev_dstrategy(vp->v_rdev, &nbp->b_bio1); 1756 1757 if (DEVFS_NODE(vp)) { 1758 nanotime(&DEVFS_NODE(vp)->atime); 1759 nanotime(&DEVFS_NODE(vp)->mtime); 1760 } 1761 1762 return (0); 1763 } 1764 1765 /* 1766 * Chunked up transfer completion routine - chain transfers until done 1767 * 1768 * NOTE: MPSAFE callback. 1769 */ 1770 static 1771 void 1772 devfs_spec_strategy_done(struct bio *nbio) 1773 { 1774 struct buf *nbp = nbio->bio_buf; 1775 struct bio *bio = nbio->bio_caller_info1.ptr; /* original bio */ 1776 struct buf *bp = bio->bio_buf; /* original bp */ 1777 int chunksize = nbio->bio_caller_info2.index; /* chunking */ 1778 int boffset = nbp->b_data - bp->b_data; 1779 1780 if (nbp->b_flags & B_ERROR) { 1781 /* 1782 * An error terminates the chain, propogate the error back 1783 * to the original bp 1784 */ 1785 bp->b_flags |= B_ERROR; 1786 bp->b_error = nbp->b_error; 1787 bp->b_resid = bp->b_bcount - boffset + 1788 (nbp->b_bcount - nbp->b_resid); 1789 #if SPEC_CHAIN_DEBUG & 1 1790 devfs_debug(DEVFS_DEBUG_DEBUG, 1791 "spec_strategy: chain %p error %d bcount %d/%d\n", 1792 bp, bp->b_error, bp->b_bcount, 1793 bp->b_bcount - bp->b_resid); 1794 #endif 1795 } else if (nbp->b_resid) { 1796 /* 1797 * A short read or write terminates the chain 1798 */ 1799 bp->b_error = nbp->b_error; 1800 bp->b_resid = bp->b_bcount - boffset + 1801 (nbp->b_bcount - nbp->b_resid); 1802 #if SPEC_CHAIN_DEBUG & 1 1803 devfs_debug(DEVFS_DEBUG_DEBUG, 1804 "spec_strategy: chain %p short read(1) " 1805 "bcount %d/%d\n", 1806 bp, bp->b_bcount - bp->b_resid, bp->b_bcount); 1807 #endif 1808 } else if (nbp->b_bcount != nbp->b_bufsize) { 1809 /* 1810 * A short read or write can also occur by truncating b_bcount 1811 */ 1812 #if SPEC_CHAIN_DEBUG & 1 1813 devfs_debug(DEVFS_DEBUG_DEBUG, 1814 "spec_strategy: chain %p short read(2) " 1815 "bcount %d/%d\n", 1816 bp, nbp->b_bcount + boffset, bp->b_bcount); 1817 #endif 1818 bp->b_error = 0; 1819 bp->b_bcount = nbp->b_bcount + boffset; 1820 bp->b_resid = nbp->b_resid; 1821 } else if (nbp->b_bcount + boffset == bp->b_bcount) { 1822 /* 1823 * No more data terminates the chain 1824 */ 1825 #if SPEC_CHAIN_DEBUG & 1 1826 devfs_debug(DEVFS_DEBUG_DEBUG, 1827 "spec_strategy: chain %p finished bcount %d\n", 1828 bp, bp->b_bcount); 1829 #endif 1830 bp->b_error = 0; 1831 bp->b_resid = 0; 1832 } else { 1833 /* 1834 * Continue the chain 1835 */ 1836 boffset += nbp->b_bcount; 1837 nbp->b_data = bp->b_data + boffset; 1838 nbp->b_bcount = bp->b_bcount - boffset; 1839 if (nbp->b_bcount > chunksize) 1840 nbp->b_bcount = chunksize; 1841 nbp->b_bio1.bio_done = devfs_spec_strategy_done; 1842 nbp->b_bio1.bio_offset = bio->bio_offset + boffset; 1843 1844 #if SPEC_CHAIN_DEBUG & 1 1845 devfs_debug(DEVFS_DEBUG_DEBUG, 1846 "spec_strategy: chain %p offset %d/%d bcount %d\n", 1847 bp, boffset, bp->b_bcount, nbp->b_bcount); 1848 #endif 1849 1850 dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1); 1851 return; 1852 } 1853 1854 /* 1855 * Fall through to here on termination. biodone(bp) and 1856 * clean up and free nbp. 1857 */ 1858 biodone(bio); 1859 BUF_UNLOCK(nbp); 1860 uninitbufbio(nbp); 1861 kfree(nbp, M_DEVBUF); 1862 } 1863 1864 /* 1865 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length) 1866 */ 1867 static int 1868 devfs_spec_freeblks(struct vop_freeblks_args *ap) 1869 { 1870 struct buf *bp; 1871 1872 /* 1873 * XXX: This assumes that strategy does the deed right away. 1874 * XXX: this may not be TRTTD. 1875 */ 1876 KKASSERT(ap->a_vp->v_rdev != NULL); 1877 if ((ap->a_vp->v_rdev->si_flags & SI_CANFREE) == 0) 1878 return (0); 1879 bp = geteblk(ap->a_length); 1880 bp->b_cmd = BUF_CMD_FREEBLKS; 1881 bp->b_bio1.bio_offset = ap->a_offset; 1882 bp->b_bcount = ap->a_length; 1883 dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1); 1884 return (0); 1885 } 1886 1887 /* 1888 * Implement degenerate case where the block requested is the block 1889 * returned, and assume that the entire device is contiguous in regards 1890 * to the contiguous block range (runp and runb). 1891 * 1892 * spec_bmap(struct vnode *a_vp, off_t a_loffset, 1893 * off_t *a_doffsetp, int *a_runp, int *a_runb) 1894 */ 1895 static int 1896 devfs_spec_bmap(struct vop_bmap_args *ap) 1897 { 1898 if (ap->a_doffsetp != NULL) 1899 *ap->a_doffsetp = ap->a_loffset; 1900 if (ap->a_runp != NULL) 1901 *ap->a_runp = MAXBSIZE; 1902 if (ap->a_runb != NULL) { 1903 if (ap->a_loffset < MAXBSIZE) 1904 *ap->a_runb = (int)ap->a_loffset; 1905 else 1906 *ap->a_runb = MAXBSIZE; 1907 } 1908 return (0); 1909 } 1910 1911 1912 /* 1913 * Special device advisory byte-level locks. 1914 * 1915 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, 1916 * struct flock *a_fl, int a_flags) 1917 */ 1918 /* ARGSUSED */ 1919 static int 1920 devfs_spec_advlock(struct vop_advlock_args *ap) 1921 { 1922 return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP); 1923 } 1924 1925 /* 1926 * NOTE: MPSAFE callback. 1927 */ 1928 static void 1929 devfs_spec_getpages_iodone(struct bio *bio) 1930 { 1931 bio->bio_buf->b_cmd = BUF_CMD_DONE; 1932 wakeup(bio->bio_buf); 1933 } 1934 1935 /* 1936 * spec_getpages() - get pages associated with device vnode. 1937 * 1938 * Note that spec_read and spec_write do not use the buffer cache, so we 1939 * must fully implement getpages here. 1940 */ 1941 static int 1942 devfs_spec_getpages(struct vop_getpages_args *ap) 1943 { 1944 vm_offset_t kva; 1945 int error; 1946 int i, pcount, size; 1947 struct buf *bp; 1948 vm_page_t m; 1949 vm_ooffset_t offset; 1950 int toff, nextoff, nread; 1951 struct vnode *vp = ap->a_vp; 1952 int blksiz; 1953 int gotreqpage; 1954 1955 error = 0; 1956 pcount = round_page(ap->a_count) / PAGE_SIZE; 1957 1958 /* 1959 * Calculate the offset of the transfer and do sanity check. 1960 */ 1961 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset; 1962 1963 /* 1964 * Round up physical size for real devices. We cannot round using 1965 * v_mount's block size data because v_mount has nothing to do with 1966 * the device. i.e. it's usually '/dev'. We need the physical block 1967 * size for the device itself. 1968 * 1969 * We can't use v_rdev->si_mountpoint because it only exists when the 1970 * block device is mounted. However, we can use v_rdev. 1971 */ 1972 if (vn_isdisk(vp, NULL)) 1973 blksiz = vp->v_rdev->si_bsize_phys; 1974 else 1975 blksiz = DEV_BSIZE; 1976 1977 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1); 1978 1979 bp = getpbuf_kva(NULL); 1980 kva = (vm_offset_t)bp->b_data; 1981 1982 /* 1983 * Map the pages to be read into the kva. 1984 */ 1985 pmap_qenter(kva, ap->a_m, pcount); 1986 1987 /* Build a minimal buffer header. */ 1988 bp->b_cmd = BUF_CMD_READ; 1989 bp->b_bcount = size; 1990 bp->b_resid = 0; 1991 bsetrunningbufspace(bp, size); 1992 1993 bp->b_bio1.bio_offset = offset; 1994 bp->b_bio1.bio_done = devfs_spec_getpages_iodone; 1995 1996 mycpu->gd_cnt.v_vnodein++; 1997 mycpu->gd_cnt.v_vnodepgsin += pcount; 1998 1999 /* Do the input. */ 2000 vn_strategy(ap->a_vp, &bp->b_bio1); 2001 2002 crit_enter(); 2003 2004 /* We definitely need to be at splbio here. */ 2005 while (bp->b_cmd != BUF_CMD_DONE) 2006 tsleep(bp, 0, "spread", 0); 2007 2008 crit_exit(); 2009 2010 if (bp->b_flags & B_ERROR) { 2011 if (bp->b_error) 2012 error = bp->b_error; 2013 else 2014 error = EIO; 2015 } 2016 2017 /* 2018 * If EOF is encountered we must zero-extend the result in order 2019 * to ensure that the page does not contain garabge. When no 2020 * error occurs, an early EOF is indicated if b_bcount got truncated. 2021 * b_resid is relative to b_bcount and should be 0, but some devices 2022 * might indicate an EOF with b_resid instead of truncating b_bcount. 2023 */ 2024 nread = bp->b_bcount - bp->b_resid; 2025 if (nread < ap->a_count) 2026 bzero((caddr_t)kva + nread, ap->a_count - nread); 2027 pmap_qremove(kva, pcount); 2028 2029 gotreqpage = 0; 2030 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) { 2031 nextoff = toff + PAGE_SIZE; 2032 m = ap->a_m[i]; 2033 2034 m->flags &= ~PG_ZERO; 2035 2036 /* 2037 * NOTE: vm_page_undirty/clear_dirty etc do not clear the 2038 * pmap modified bit. pmap modified bit should have 2039 * already been cleared. 2040 */ 2041 if (nextoff <= nread) { 2042 m->valid = VM_PAGE_BITS_ALL; 2043 vm_page_undirty(m); 2044 } else if (toff < nread) { 2045 /* 2046 * Since this is a VM request, we have to supply the 2047 * unaligned offset to allow vm_page_set_valid() 2048 * to zero sub-DEV_BSIZE'd portions of the page. 2049 */ 2050 vm_page_set_valid(m, 0, nread - toff); 2051 vm_page_clear_dirty_end_nonincl(m, 0, nread - toff); 2052 } else { 2053 m->valid = 0; 2054 vm_page_undirty(m); 2055 } 2056 2057 if (i != ap->a_reqpage) { 2058 /* 2059 * Just in case someone was asking for this page we 2060 * now tell them that it is ok to use. 2061 */ 2062 if (!error || (m->valid == VM_PAGE_BITS_ALL)) { 2063 if (m->valid) { 2064 if (m->flags & PG_REFERENCED) { 2065 vm_page_activate(m); 2066 } else { 2067 vm_page_deactivate(m); 2068 } 2069 vm_page_wakeup(m); 2070 } else { 2071 vm_page_free(m); 2072 } 2073 } else { 2074 vm_page_free(m); 2075 } 2076 } else if (m->valid) { 2077 gotreqpage = 1; 2078 /* 2079 * Since this is a VM request, we need to make the 2080 * entire page presentable by zeroing invalid sections. 2081 */ 2082 if (m->valid != VM_PAGE_BITS_ALL) 2083 vm_page_zero_invalid(m, FALSE); 2084 } 2085 } 2086 if (!gotreqpage) { 2087 m = ap->a_m[ap->a_reqpage]; 2088 devfs_debug(DEVFS_DEBUG_WARNING, 2089 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n", 2090 devtoname(vp->v_rdev), error, bp, bp->b_vp); 2091 devfs_debug(DEVFS_DEBUG_WARNING, 2092 " size: %d, resid: %d, a_count: %d, valid: 0x%x\n", 2093 size, bp->b_resid, ap->a_count, m->valid); 2094 devfs_debug(DEVFS_DEBUG_WARNING, 2095 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n", 2096 nread, ap->a_reqpage, (u_long)m->pindex, pcount); 2097 /* 2098 * Free the buffer header back to the swap buffer pool. 2099 */ 2100 relpbuf(bp, NULL); 2101 return VM_PAGER_ERROR; 2102 } 2103 /* 2104 * Free the buffer header back to the swap buffer pool. 2105 */ 2106 relpbuf(bp, NULL); 2107 if (DEVFS_NODE(ap->a_vp)) 2108 nanotime(&DEVFS_NODE(ap->a_vp)->mtime); 2109 return VM_PAGER_OK; 2110 } 2111 2112 static __inline 2113 int 2114 sequential_heuristic(struct uio *uio, struct file *fp) 2115 { 2116 /* 2117 * Sequential heuristic - detect sequential operation 2118 */ 2119 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 2120 uio->uio_offset == fp->f_nextoff) { 2121 /* 2122 * XXX we assume that the filesystem block size is 2123 * the default. Not true, but still gives us a pretty 2124 * good indicator of how sequential the read operations 2125 * are. 2126 */ 2127 int tmpseq = fp->f_seqcount; 2128 2129 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE; 2130 if (tmpseq > IO_SEQMAX) 2131 tmpseq = IO_SEQMAX; 2132 fp->f_seqcount = tmpseq; 2133 return(fp->f_seqcount << IO_SEQSHIFT); 2134 } 2135 2136 /* 2137 * Not sequential, quick draw-down of seqcount 2138 */ 2139 if (fp->f_seqcount > 1) 2140 fp->f_seqcount = 1; 2141 else 2142 fp->f_seqcount = 0; 2143 return(0); 2144 } 2145