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