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