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