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 depth; 444 445 ncp = ap->a_nch->ncp; 446 len = ncp->nc_nlen; 447 448 if (!devfs_node_is_accessible(dnode)) 449 return ENOENT; 450 451 lockmgr(&devfs_lock, LK_EXCLUSIVE); 452 453 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) { 454 error = ENOENT; 455 cache_setvp(ap->a_nch, NULL); 456 goto out; 457 } 458 459 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { 460 if (len == node->d_dir.d_namlen) { 461 if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) { 462 found = node; 463 break; 464 } 465 } 466 } 467 468 if (found) { 469 depth = 0; 470 while ((found->node_type == Plink) && (found->link_target)) { 471 if (depth >= 8) { 472 devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8"); 473 break; 474 } 475 476 found = found->link_target; 477 ++depth; 478 } 479 480 if (!(found->flags & DEVFS_HIDDEN)) 481 devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found); 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 * This checks if the disk device is going to be opened for writing. 826 * It will be only allowed in the cases where securelevel permits it 827 * and it's not mounted R/W. 828 */ 829 if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) && 830 (ap->a_cred != FSCRED)) { 831 832 /* Very secure mode. No open for writing allowed */ 833 if (securelevel >= 2) 834 return EPERM; 835 836 /* 837 * If it is mounted R/W, do not allow to open for writing. 838 * In the case it's mounted read-only but securelevel 839 * is >= 1, then do not allow opening for writing either. 840 */ 841 if (vfs_mountedon(vp)) { 842 if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY)) 843 return EBUSY; 844 else if (securelevel >= 1) 845 return EPERM; 846 } 847 } 848 849 if (dev_dflags(dev) & D_TTY) { 850 if (dev->si_tty) { 851 struct tty *tp; 852 tp = dev->si_tty; 853 if (!tp->t_stop) { 854 devfs_debug(DEVFS_DEBUG_DEBUG, 855 "devfs: no t_stop\n"); 856 tp->t_stop = nottystop; 857 } 858 } 859 } 860 861 862 if (vn_isdisk(vp, NULL)) { 863 if (!dev->si_bsize_phys) 864 dev->si_bsize_phys = DEV_BSIZE; 865 vinitvmio(vp, IDX_TO_OFF(INT_MAX)); 866 } 867 868 vop_stdopen(ap); 869 #if 0 870 if (node) 871 nanotime(&node->atime); 872 #endif 873 874 if (orig_vp) 875 vn_unlock(vp); 876 877 /* Ugly pty magic, to make pty devices appear once they are opened */ 878 if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY) 879 node->flags &= ~DEVFS_INVISIBLE; 880 881 if (ap->a_fp) { 882 ap->a_fp->f_type = DTYPE_VNODE; 883 ap->a_fp->f_flag = ap->a_mode & FMASK; 884 ap->a_fp->f_ops = &devfs_dev_fileops; 885 ap->a_fp->f_data = vp; 886 } 887 888 return 0; 889 } 890 891 892 static int 893 devfs_spec_close(struct vop_close_args *ap) 894 { 895 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 896 struct proc *p = curproc; 897 struct vnode *vp = ap->a_vp; 898 cdev_t dev = vp->v_rdev; 899 int error = 0; 900 int needrelock; 901 902 devfs_debug(DEVFS_DEBUG_DEBUG, 903 "devfs_spec_close() called on %s! \n", 904 dev->si_name); 905 906 /* 907 * A couple of hacks for devices and tty devices. The 908 * vnode ref count cannot be used to figure out the 909 * last close, but we can use v_opencount now that 910 * revoke works properly. 911 * 912 * Detect the last close on a controlling terminal and clear 913 * the session (half-close). 914 */ 915 if (dev) 916 reference_dev(dev); 917 918 if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) { 919 p->p_session->s_ttyvp = NULL; 920 vrele(vp); 921 } 922 923 /* 924 * Vnodes can be opened and closed multiple times. Do not really 925 * close the device unless (1) it is being closed forcibly, 926 * (2) the device wants to track closes, or (3) this is the last 927 * vnode doing its last close on the device. 928 * 929 * XXX the VXLOCK (force close) case can leave vnodes referencing 930 * a closed device. This might not occur now that our revoke is 931 * fixed. 932 */ 933 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n"); 934 if (dev && ((vp->v_flag & VRECLAIMED) || 935 (dev_dflags(dev) & D_TRACKCLOSE) || 936 (vp->v_opencount == 1))) { 937 /* 938 * Unlock around dev_dclose() 939 */ 940 needrelock = 0; 941 if (vn_islocked(vp)) { 942 needrelock = 1; 943 vn_unlock(vp); 944 } 945 error = dev_dclose(dev, ap->a_fflag, S_IFCHR); 946 947 /* 948 * Ugly pty magic, to make pty devices disappear again once 949 * they are closed 950 */ 951 if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY) 952 node->flags |= DEVFS_INVISIBLE; 953 954 if (needrelock) 955 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 956 } else { 957 error = 0; 958 } 959 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n"); 960 961 /* 962 * Track the actual opens and closes on the vnode. The last close 963 * disassociates the rdev. If the rdev is already disassociated or 964 * the opencount is already 0, the vnode might have been revoked 965 * and no further opencount tracking occurs. 966 */ 967 if (dev) 968 release_dev(dev); 969 if (vp->v_opencount > 0) 970 vop_stdclose(ap); 971 return(error); 972 973 } 974 975 976 static int 977 devfs_specf_close(struct file *fp) 978 { 979 struct vnode *vp = (struct vnode *)fp->f_data; 980 int error; 981 982 get_mplock(); 983 fp->f_ops = &badfileops; 984 error = vn_close(vp, fp->f_flag); 985 rel_mplock(); 986 987 return (error); 988 } 989 990 991 /* 992 * Device-optimized file table vnode read routine. 993 * 994 * This bypasses the VOP table and talks directly to the device. Most 995 * filesystems just route to specfs and can make this optimization. 996 * 997 * MPALMOSTSAFE - acquires mplock 998 */ 999 static int 1000 devfs_specf_read(struct file *fp, struct uio *uio, 1001 struct ucred *cred, int flags) 1002 { 1003 struct devfs_node *node; 1004 struct vnode *vp; 1005 int ioflag; 1006 int error; 1007 cdev_t dev; 1008 1009 get_mplock(); 1010 KASSERT(uio->uio_td == curthread, 1011 ("uio_td %p is not td %p", uio->uio_td, curthread)); 1012 1013 vp = (struct vnode *)fp->f_data; 1014 if (vp == NULL || vp->v_type == VBAD) { 1015 error = EBADF; 1016 goto done; 1017 } 1018 node = DEVFS_NODE(vp); 1019 1020 if ((dev = vp->v_rdev) == NULL) { 1021 error = EBADF; 1022 goto done; 1023 } 1024 1025 reference_dev(dev); 1026 1027 if (uio->uio_resid == 0) { 1028 error = 0; 1029 goto done; 1030 } 1031 if ((flags & O_FOFFSET) == 0) 1032 uio->uio_offset = fp->f_offset; 1033 1034 ioflag = 0; 1035 if (flags & O_FBLOCKING) { 1036 /* ioflag &= ~IO_NDELAY; */ 1037 } else if (flags & O_FNONBLOCKING) { 1038 ioflag |= IO_NDELAY; 1039 } else if (fp->f_flag & FNONBLOCK) { 1040 ioflag |= IO_NDELAY; 1041 } 1042 if (flags & O_FBUFFERED) { 1043 /* ioflag &= ~IO_DIRECT; */ 1044 } else if (flags & O_FUNBUFFERED) { 1045 ioflag |= IO_DIRECT; 1046 } else if (fp->f_flag & O_DIRECT) { 1047 ioflag |= IO_DIRECT; 1048 } 1049 ioflag |= sequential_heuristic(uio, fp); 1050 1051 error = dev_dread(dev, uio, ioflag); 1052 1053 release_dev(dev); 1054 if (node) 1055 nanotime(&node->atime); 1056 if ((flags & O_FOFFSET) == 0) 1057 fp->f_offset = uio->uio_offset; 1058 fp->f_nextoff = uio->uio_offset; 1059 done: 1060 rel_mplock(); 1061 return (error); 1062 } 1063 1064 1065 static int 1066 devfs_specf_write(struct file *fp, struct uio *uio, 1067 struct ucred *cred, int flags) 1068 { 1069 struct devfs_node *node; 1070 struct vnode *vp; 1071 int ioflag; 1072 int error; 1073 cdev_t dev; 1074 1075 get_mplock(); 1076 KASSERT(uio->uio_td == curthread, 1077 ("uio_td %p is not p %p", uio->uio_td, curthread)); 1078 1079 vp = (struct vnode *)fp->f_data; 1080 if (vp == NULL || vp->v_type == VBAD) { 1081 error = EBADF; 1082 goto done; 1083 } 1084 node = DEVFS_NODE(vp); 1085 if (vp->v_type == VREG) 1086 bwillwrite(uio->uio_resid); 1087 vp = (struct vnode *)fp->f_data; 1088 1089 if ((dev = vp->v_rdev) == NULL) { 1090 error = EBADF; 1091 goto done; 1092 } 1093 reference_dev(dev); 1094 1095 if ((flags & O_FOFFSET) == 0) 1096 uio->uio_offset = fp->f_offset; 1097 1098 ioflag = IO_UNIT; 1099 if (vp->v_type == VREG && 1100 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 1101 ioflag |= IO_APPEND; 1102 } 1103 1104 if (flags & O_FBLOCKING) { 1105 /* ioflag &= ~IO_NDELAY; */ 1106 } else if (flags & O_FNONBLOCKING) { 1107 ioflag |= IO_NDELAY; 1108 } else if (fp->f_flag & FNONBLOCK) { 1109 ioflag |= IO_NDELAY; 1110 } 1111 if (flags & O_FBUFFERED) { 1112 /* ioflag &= ~IO_DIRECT; */ 1113 } else if (flags & O_FUNBUFFERED) { 1114 ioflag |= IO_DIRECT; 1115 } else if (fp->f_flag & O_DIRECT) { 1116 ioflag |= IO_DIRECT; 1117 } 1118 if (flags & O_FASYNCWRITE) { 1119 /* ioflag &= ~IO_SYNC; */ 1120 } else if (flags & O_FSYNCWRITE) { 1121 ioflag |= IO_SYNC; 1122 } else if (fp->f_flag & O_FSYNC) { 1123 ioflag |= IO_SYNC; 1124 } 1125 1126 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 1127 ioflag |= IO_SYNC; 1128 ioflag |= sequential_heuristic(uio, fp); 1129 1130 error = dev_dwrite(dev, uio, ioflag); 1131 1132 release_dev(dev); 1133 if (node) { 1134 nanotime(&node->atime); 1135 nanotime(&node->mtime); 1136 } 1137 1138 if ((flags & O_FOFFSET) == 0) 1139 fp->f_offset = uio->uio_offset; 1140 fp->f_nextoff = uio->uio_offset; 1141 done: 1142 rel_mplock(); 1143 return (error); 1144 } 1145 1146 1147 static int 1148 devfs_specf_stat(struct file *fp, struct stat *sb, struct ucred *cred) 1149 { 1150 struct vnode *vp; 1151 int error; 1152 1153 get_mplock(); 1154 vp = (struct vnode *)fp->f_data; 1155 error = vn_stat(vp, sb, cred); 1156 if (error) { 1157 rel_mplock(); 1158 return (error); 1159 } 1160 1161 struct vattr vattr; 1162 struct vattr *vap; 1163 u_short mode; 1164 cdev_t dev; 1165 1166 vap = &vattr; 1167 error = VOP_GETATTR(vp, vap); 1168 if (error) { 1169 rel_mplock(); 1170 return (error); 1171 } 1172 1173 /* 1174 * Zero the spare stat fields 1175 */ 1176 sb->st_lspare = 0; 1177 sb->st_qspare = 0; 1178 1179 /* 1180 * Copy from vattr table ... or not in case it's a cloned device 1181 */ 1182 if (vap->va_fsid != VNOVAL) 1183 sb->st_dev = vap->va_fsid; 1184 else 1185 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 1186 1187 sb->st_ino = vap->va_fileid; 1188 1189 mode = vap->va_mode; 1190 mode |= S_IFCHR; 1191 sb->st_mode = mode; 1192 1193 if (vap->va_nlink > (nlink_t)-1) 1194 sb->st_nlink = (nlink_t)-1; 1195 else 1196 sb->st_nlink = vap->va_nlink; 1197 sb->st_uid = vap->va_uid; 1198 sb->st_gid = vap->va_gid; 1199 sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev); 1200 sb->st_size = vap->va_size; 1201 sb->st_atimespec = vap->va_atime; 1202 sb->st_mtimespec = vap->va_mtime; 1203 sb->st_ctimespec = vap->va_ctime; 1204 1205 /* 1206 * A VCHR and VBLK device may track the last access and last modified 1207 * time independantly of the filesystem. This is particularly true 1208 * because device read and write calls may bypass the filesystem. 1209 */ 1210 if (vp->v_type == VCHR || vp->v_type == VBLK) { 1211 dev = vp->v_rdev; 1212 if (dev != NULL) { 1213 if (dev->si_lastread) { 1214 sb->st_atimespec.tv_sec = dev->si_lastread; 1215 sb->st_atimespec.tv_nsec = 0; 1216 } 1217 if (dev->si_lastwrite) { 1218 sb->st_atimespec.tv_sec = dev->si_lastwrite; 1219 sb->st_atimespec.tv_nsec = 0; 1220 } 1221 } 1222 } 1223 1224 /* 1225 * According to www.opengroup.org, the meaning of st_blksize is 1226 * "a filesystem-specific preferred I/O block size for this 1227 * object. In some filesystem types, this may vary from file 1228 * to file" 1229 * Default to PAGE_SIZE after much discussion. 1230 */ 1231 1232 sb->st_blksize = PAGE_SIZE; 1233 1234 sb->st_flags = vap->va_flags; 1235 1236 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0); 1237 if (error) 1238 sb->st_gen = 0; 1239 else 1240 sb->st_gen = (u_int32_t)vap->va_gen; 1241 1242 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 1243 sb->st_fsmid = vap->va_fsmid; 1244 1245 rel_mplock(); 1246 return (0); 1247 } 1248 1249 1250 static int 1251 devfs_specf_kqfilter(struct file *fp, struct knote *kn) 1252 { 1253 struct vnode *vp; 1254 int error; 1255 cdev_t dev; 1256 1257 get_mplock(); 1258 1259 vp = (struct vnode *)fp->f_data; 1260 if (vp == NULL || vp->v_type == VBAD) { 1261 error = EBADF; 1262 goto done; 1263 } 1264 if ((dev = vp->v_rdev) == NULL) { 1265 error = EBADF; 1266 goto done; 1267 } 1268 reference_dev(dev); 1269 1270 error = dev_dkqfilter(dev, kn); 1271 1272 release_dev(dev); 1273 1274 done: 1275 rel_mplock(); 1276 return (error); 1277 } 1278 1279 1280 static int 1281 devfs_specf_poll(struct file *fp, int events, struct ucred *cred) 1282 { 1283 struct devfs_node *node; 1284 struct vnode *vp; 1285 int error; 1286 cdev_t dev; 1287 1288 get_mplock(); 1289 1290 vp = (struct vnode *)fp->f_data; 1291 if (vp == NULL || vp->v_type == VBAD) { 1292 error = EBADF; 1293 goto done; 1294 } 1295 node = DEVFS_NODE(vp); 1296 1297 if ((dev = vp->v_rdev) == NULL) { 1298 error = EBADF; 1299 goto done; 1300 } 1301 reference_dev(dev); 1302 error = dev_dpoll(dev, events); 1303 1304 release_dev(dev); 1305 1306 #if 0 1307 if (node) 1308 nanotime(&node->atime); 1309 #endif 1310 done: 1311 rel_mplock(); 1312 return (error); 1313 } 1314 1315 1316 /* 1317 * MPALMOSTSAFE - acquires mplock 1318 */ 1319 static int 1320 devfs_specf_ioctl(struct file *fp, u_long com, caddr_t data, 1321 struct ucred *ucred, struct sysmsg *msg) 1322 { 1323 struct devfs_node *node; 1324 struct vnode *vp; 1325 struct vnode *ovp; 1326 cdev_t dev; 1327 int error; 1328 struct fiodname_args *name_args; 1329 size_t namlen; 1330 const char *name; 1331 1332 get_mplock(); 1333 vp = ((struct vnode *)fp->f_data); 1334 if ((dev = vp->v_rdev) == NULL) { 1335 error = EBADF; /* device was revoked */ 1336 goto out; 1337 } 1338 1339 node = DEVFS_NODE(vp); 1340 1341 devfs_debug(DEVFS_DEBUG_DEBUG, 1342 "devfs_specf_ioctl() called! for dev %s\n", 1343 dev->si_name); 1344 1345 if (com == FIODTYPE) { 1346 *(int *)data = dev_dflags(dev) & D_TYPEMASK; 1347 error = 0; 1348 goto out; 1349 } else if (com == FIODNAME) { 1350 name_args = (struct fiodname_args *)data; 1351 name = dev->si_name; 1352 namlen = strlen(name) + 1; 1353 1354 devfs_debug(DEVFS_DEBUG_DEBUG, 1355 "ioctl, got: FIODNAME for %s\n", name); 1356 1357 if (namlen <= name_args->len) 1358 error = copyout(dev->si_name, name_args->name, namlen); 1359 else 1360 error = EINVAL; 1361 1362 devfs_debug(DEVFS_DEBUG_DEBUG, 1363 "ioctl stuff: error: %d\n", error); 1364 goto out; 1365 } 1366 reference_dev(dev); 1367 error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg); 1368 release_dev(dev); 1369 #if 0 1370 if (node) { 1371 nanotime(&node->atime); 1372 nanotime(&node->mtime); 1373 } 1374 #endif 1375 1376 if (com == TIOCSCTTY) { 1377 devfs_debug(DEVFS_DEBUG_DEBUG, 1378 "devfs_specf_ioctl: got TIOCSCTTY on %s\n", 1379 dev->si_name); 1380 } 1381 if (error == 0 && com == TIOCSCTTY) { 1382 struct proc *p = curthread->td_proc; 1383 struct session *sess; 1384 1385 devfs_debug(DEVFS_DEBUG_DEBUG, 1386 "devfs_specf_ioctl: dealing with TIOCSCTTY on %s\n", 1387 dev->si_name); 1388 if (p == NULL) { 1389 error = ENOTTY; 1390 goto out; 1391 } 1392 sess = p->p_session; 1393 1394 /* 1395 * Do nothing if reassigning same control tty 1396 */ 1397 if (sess->s_ttyvp == vp) { 1398 error = 0; 1399 goto out; 1400 } 1401 1402 /* 1403 * Get rid of reference to old control tty 1404 */ 1405 ovp = sess->s_ttyvp; 1406 vref(vp); 1407 sess->s_ttyvp = vp; 1408 if (ovp) 1409 vrele(ovp); 1410 } 1411 1412 out: 1413 rel_mplock(); 1414 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_specf_ioctl() finished! \n"); 1415 return (error); 1416 } 1417 1418 1419 static int 1420 devfs_spec_fsync(struct vop_fsync_args *ap) 1421 { 1422 struct vnode *vp = ap->a_vp; 1423 int error; 1424 1425 if (!vn_isdisk(vp, NULL)) 1426 return (0); 1427 1428 /* 1429 * Flush all dirty buffers associated with a block device. 1430 */ 1431 error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL); 1432 return (error); 1433 } 1434 1435 static int 1436 devfs_spec_read(struct vop_read_args *ap) 1437 { 1438 struct devfs_node *node; 1439 struct vnode *vp; 1440 struct uio *uio; 1441 cdev_t dev; 1442 int error; 1443 1444 vp = ap->a_vp; 1445 dev = vp->v_rdev; 1446 uio = ap->a_uio; 1447 node = DEVFS_NODE(vp); 1448 1449 if (dev == NULL) /* device was revoked */ 1450 return (EBADF); 1451 if (uio->uio_resid == 0) 1452 return (0); 1453 1454 vn_unlock(vp); 1455 error = dev_dread(dev, uio, ap->a_ioflag); 1456 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1457 1458 if (node) 1459 nanotime(&node->atime); 1460 1461 return (error); 1462 } 1463 1464 /* 1465 * Vnode op for write 1466 * 1467 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1468 * struct ucred *a_cred) 1469 */ 1470 static int 1471 devfs_spec_write(struct vop_write_args *ap) 1472 { 1473 struct devfs_node *node; 1474 struct vnode *vp; 1475 struct uio *uio; 1476 cdev_t dev; 1477 int error; 1478 1479 vp = ap->a_vp; 1480 dev = vp->v_rdev; 1481 uio = ap->a_uio; 1482 node = DEVFS_NODE(vp); 1483 1484 KKASSERT(uio->uio_segflg != UIO_NOCOPY); 1485 1486 if (dev == NULL) /* device was revoked */ 1487 return (EBADF); 1488 1489 vn_unlock(vp); 1490 error = dev_dwrite(dev, uio, ap->a_ioflag); 1491 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1492 1493 if (node) { 1494 nanotime(&node->atime); 1495 nanotime(&node->mtime); 1496 } 1497 1498 return (error); 1499 } 1500 1501 /* 1502 * Device ioctl operation. 1503 * 1504 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data, 1505 * int a_fflag, struct ucred *a_cred, struct sysmsg *msg) 1506 */ 1507 static int 1508 devfs_spec_ioctl(struct vop_ioctl_args *ap) 1509 { 1510 struct vnode *vp = ap->a_vp; 1511 struct devfs_node *node; 1512 cdev_t dev; 1513 1514 if ((dev = vp->v_rdev) == NULL) 1515 return (EBADF); /* device was revoked */ 1516 node = DEVFS_NODE(vp); 1517 1518 #if 0 1519 if (node) { 1520 nanotime(&node->atime); 1521 nanotime(&node->mtime); 1522 } 1523 #endif 1524 1525 return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag, 1526 ap->a_cred, ap->a_sysmsg)); 1527 } 1528 1529 /* 1530 * spec_poll(struct vnode *a_vp, int a_events, struct ucred *a_cred) 1531 */ 1532 /* ARGSUSED */ 1533 static int 1534 devfs_spec_poll(struct vop_poll_args *ap) 1535 { 1536 struct vnode *vp = ap->a_vp; 1537 struct devfs_node *node; 1538 cdev_t dev; 1539 1540 if ((dev = vp->v_rdev) == NULL) 1541 return (EBADF); /* device was revoked */ 1542 node = DEVFS_NODE(vp); 1543 1544 #if 0 1545 if (node) 1546 nanotime(&node->atime); 1547 #endif 1548 1549 return (dev_dpoll(dev, ap->a_events)); 1550 } 1551 1552 /* 1553 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn) 1554 */ 1555 /* ARGSUSED */ 1556 static int 1557 devfs_spec_kqfilter(struct vop_kqfilter_args *ap) 1558 { 1559 struct vnode *vp = ap->a_vp; 1560 struct devfs_node *node; 1561 cdev_t dev; 1562 1563 if ((dev = vp->v_rdev) == NULL) 1564 return (EBADF); /* device was revoked */ 1565 node = DEVFS_NODE(vp); 1566 1567 #if 0 1568 if (node) 1569 nanotime(&node->atime); 1570 #endif 1571 1572 return (dev_dkqfilter(dev, ap->a_kn)); 1573 } 1574 1575 /* 1576 * Convert a vnode strategy call into a device strategy call. Vnode strategy 1577 * calls are not limited to device DMA limits so we have to deal with the 1578 * case. 1579 * 1580 * spec_strategy(struct vnode *a_vp, struct bio *a_bio) 1581 */ 1582 static int 1583 devfs_spec_strategy(struct vop_strategy_args *ap) 1584 { 1585 struct bio *bio = ap->a_bio; 1586 struct buf *bp = bio->bio_buf; 1587 struct buf *nbp; 1588 struct vnode *vp; 1589 struct mount *mp; 1590 int chunksize; 1591 int maxiosize; 1592 1593 if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL) 1594 buf_start(bp); 1595 1596 /* 1597 * Collect statistics on synchronous and asynchronous read 1598 * and write counts for disks that have associated filesystems. 1599 */ 1600 vp = ap->a_vp; 1601 KKASSERT(vp->v_rdev != NULL); /* XXX */ 1602 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) { 1603 if (bp->b_cmd == BUF_CMD_READ) { 1604 if (bp->b_flags & BIO_SYNC) 1605 mp->mnt_stat.f_syncreads++; 1606 else 1607 mp->mnt_stat.f_asyncreads++; 1608 } else { 1609 if (bp->b_flags & BIO_SYNC) 1610 mp->mnt_stat.f_syncwrites++; 1611 else 1612 mp->mnt_stat.f_asyncwrites++; 1613 } 1614 } 1615 1616 /* 1617 * Device iosize limitations only apply to read and write. Shortcut 1618 * the I/O if it fits. 1619 */ 1620 if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) { 1621 devfs_debug(DEVFS_DEBUG_DEBUG, 1622 "%s: si_iosize_max not set!\n", 1623 dev_dname(vp->v_rdev)); 1624 maxiosize = MAXPHYS; 1625 } 1626 #if SPEC_CHAIN_DEBUG & 2 1627 maxiosize = 4096; 1628 #endif 1629 if (bp->b_bcount <= maxiosize || 1630 (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) { 1631 dev_dstrategy_chain(vp->v_rdev, bio); 1632 return (0); 1633 } 1634 1635 /* 1636 * Clone the buffer and set up an I/O chain to chunk up the I/O. 1637 */ 1638 nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO); 1639 initbufbio(nbp); 1640 buf_dep_init(nbp); 1641 BUF_LOCKINIT(nbp); 1642 BUF_LOCK(nbp, LK_EXCLUSIVE); 1643 BUF_KERNPROC(nbp); 1644 nbp->b_vp = vp; 1645 nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP); 1646 nbp->b_data = bp->b_data; 1647 nbp->b_bio1.bio_done = devfs_spec_strategy_done; 1648 nbp->b_bio1.bio_offset = bio->bio_offset; 1649 nbp->b_bio1.bio_caller_info1.ptr = bio; 1650 1651 /* 1652 * Start the first transfer 1653 */ 1654 if (vn_isdisk(vp, NULL)) 1655 chunksize = vp->v_rdev->si_bsize_phys; 1656 else 1657 chunksize = DEV_BSIZE; 1658 chunksize = maxiosize / chunksize * chunksize; 1659 #if SPEC_CHAIN_DEBUG & 1 1660 devfs_debug(DEVFS_DEBUG_DEBUG, 1661 "spec_strategy chained I/O chunksize=%d\n", 1662 chunksize); 1663 #endif 1664 nbp->b_cmd = bp->b_cmd; 1665 nbp->b_bcount = chunksize; 1666 nbp->b_bufsize = chunksize; /* used to detect a short I/O */ 1667 nbp->b_bio1.bio_caller_info2.index = chunksize; 1668 1669 #if SPEC_CHAIN_DEBUG & 1 1670 devfs_debug(DEVFS_DEBUG_DEBUG, 1671 "spec_strategy: chain %p offset %d/%d bcount %d\n", 1672 bp, 0, bp->b_bcount, nbp->b_bcount); 1673 #endif 1674 1675 dev_dstrategy(vp->v_rdev, &nbp->b_bio1); 1676 1677 if (DEVFS_NODE(vp)) { 1678 nanotime(&DEVFS_NODE(vp)->atime); 1679 nanotime(&DEVFS_NODE(vp)->mtime); 1680 } 1681 1682 return (0); 1683 } 1684 1685 /* 1686 * Chunked up transfer completion routine - chain transfers until done 1687 */ 1688 static 1689 void 1690 devfs_spec_strategy_done(struct bio *nbio) 1691 { 1692 struct buf *nbp = nbio->bio_buf; 1693 struct bio *bio = nbio->bio_caller_info1.ptr; /* original bio */ 1694 struct buf *bp = bio->bio_buf; /* original bp */ 1695 int chunksize = nbio->bio_caller_info2.index; /* chunking */ 1696 int boffset = nbp->b_data - bp->b_data; 1697 1698 if (nbp->b_flags & B_ERROR) { 1699 /* 1700 * An error terminates the chain, propogate the error back 1701 * to the original bp 1702 */ 1703 bp->b_flags |= B_ERROR; 1704 bp->b_error = nbp->b_error; 1705 bp->b_resid = bp->b_bcount - boffset + 1706 (nbp->b_bcount - nbp->b_resid); 1707 #if SPEC_CHAIN_DEBUG & 1 1708 devfs_debug(DEVFS_DEBUG_DEBUG, 1709 "spec_strategy: chain %p error %d bcount %d/%d\n", 1710 bp, bp->b_error, bp->b_bcount, 1711 bp->b_bcount - bp->b_resid); 1712 #endif 1713 kfree(nbp, M_DEVBUF); 1714 biodone(bio); 1715 } else if (nbp->b_resid) { 1716 /* 1717 * A short read or write terminates the chain 1718 */ 1719 bp->b_error = nbp->b_error; 1720 bp->b_resid = bp->b_bcount - boffset + 1721 (nbp->b_bcount - nbp->b_resid); 1722 #if SPEC_CHAIN_DEBUG & 1 1723 devfs_debug(DEVFS_DEBUG_DEBUG, 1724 "spec_strategy: chain %p short read(1) " 1725 "bcount %d/%d\n", 1726 bp, bp->b_bcount - bp->b_resid, bp->b_bcount); 1727 #endif 1728 kfree(nbp, M_DEVBUF); 1729 biodone(bio); 1730 } else if (nbp->b_bcount != nbp->b_bufsize) { 1731 /* 1732 * A short read or write can also occur by truncating b_bcount 1733 */ 1734 #if SPEC_CHAIN_DEBUG & 1 1735 devfs_debug(DEVFS_DEBUG_DEBUG, 1736 "spec_strategy: chain %p short read(2) " 1737 "bcount %d/%d\n", 1738 bp, nbp->b_bcount + boffset, bp->b_bcount); 1739 #endif 1740 bp->b_error = 0; 1741 bp->b_bcount = nbp->b_bcount + boffset; 1742 bp->b_resid = nbp->b_resid; 1743 kfree(nbp, M_DEVBUF); 1744 biodone(bio); 1745 } else if (nbp->b_bcount + boffset == bp->b_bcount) { 1746 /* 1747 * No more data terminates the chain 1748 */ 1749 #if SPEC_CHAIN_DEBUG & 1 1750 devfs_debug(DEVFS_DEBUG_DEBUG, 1751 "spec_strategy: chain %p finished bcount %d\n", 1752 bp, bp->b_bcount); 1753 #endif 1754 bp->b_error = 0; 1755 bp->b_resid = 0; 1756 kfree(nbp, M_DEVBUF); 1757 biodone(bio); 1758 } else { 1759 /* 1760 * Continue the chain 1761 */ 1762 boffset += nbp->b_bcount; 1763 nbp->b_data = bp->b_data + boffset; 1764 nbp->b_bcount = bp->b_bcount - boffset; 1765 if (nbp->b_bcount > chunksize) 1766 nbp->b_bcount = chunksize; 1767 nbp->b_bio1.bio_done = devfs_spec_strategy_done; 1768 nbp->b_bio1.bio_offset = bio->bio_offset + boffset; 1769 1770 #if SPEC_CHAIN_DEBUG & 1 1771 devfs_debug(DEVFS_DEBUG_DEBUG, 1772 "spec_strategy: chain %p offset %d/%d bcount %d\n", 1773 bp, boffset, bp->b_bcount, nbp->b_bcount); 1774 #endif 1775 1776 dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1); 1777 } 1778 } 1779 1780 /* 1781 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length) 1782 */ 1783 static int 1784 devfs_spec_freeblks(struct vop_freeblks_args *ap) 1785 { 1786 struct buf *bp; 1787 1788 /* 1789 * XXX: This assumes that strategy does the deed right away. 1790 * XXX: this may not be TRTTD. 1791 */ 1792 KKASSERT(ap->a_vp->v_rdev != NULL); 1793 if ((dev_dflags(ap->a_vp->v_rdev) & D_CANFREE) == 0) 1794 return (0); 1795 bp = geteblk(ap->a_length); 1796 bp->b_cmd = BUF_CMD_FREEBLKS; 1797 bp->b_bio1.bio_offset = ap->a_offset; 1798 bp->b_bcount = ap->a_length; 1799 dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1); 1800 return (0); 1801 } 1802 1803 /* 1804 * Implement degenerate case where the block requested is the block 1805 * returned, and assume that the entire device is contiguous in regards 1806 * to the contiguous block range (runp and runb). 1807 * 1808 * spec_bmap(struct vnode *a_vp, off_t a_loffset, 1809 * off_t *a_doffsetp, int *a_runp, int *a_runb) 1810 */ 1811 static int 1812 devfs_spec_bmap(struct vop_bmap_args *ap) 1813 { 1814 if (ap->a_doffsetp != NULL) 1815 *ap->a_doffsetp = ap->a_loffset; 1816 if (ap->a_runp != NULL) 1817 *ap->a_runp = MAXBSIZE; 1818 if (ap->a_runb != NULL) { 1819 if (ap->a_loffset < MAXBSIZE) 1820 *ap->a_runb = (int)ap->a_loffset; 1821 else 1822 *ap->a_runb = MAXBSIZE; 1823 } 1824 return (0); 1825 } 1826 1827 1828 /* 1829 * Special device advisory byte-level locks. 1830 * 1831 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, 1832 * struct flock *a_fl, int a_flags) 1833 */ 1834 /* ARGSUSED */ 1835 static int 1836 devfs_spec_advlock(struct vop_advlock_args *ap) 1837 { 1838 return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP); 1839 } 1840 1841 static void 1842 devfs_spec_getpages_iodone(struct bio *bio) 1843 { 1844 bio->bio_buf->b_cmd = BUF_CMD_DONE; 1845 wakeup(bio->bio_buf); 1846 } 1847 1848 /* 1849 * spec_getpages() - get pages associated with device vnode. 1850 * 1851 * Note that spec_read and spec_write do not use the buffer cache, so we 1852 * must fully implement getpages here. 1853 */ 1854 static int 1855 devfs_spec_getpages(struct vop_getpages_args *ap) 1856 { 1857 vm_offset_t kva; 1858 int error; 1859 int i, pcount, size; 1860 struct buf *bp; 1861 vm_page_t m; 1862 vm_ooffset_t offset; 1863 int toff, nextoff, nread; 1864 struct vnode *vp = ap->a_vp; 1865 int blksiz; 1866 int gotreqpage; 1867 1868 error = 0; 1869 pcount = round_page(ap->a_count) / PAGE_SIZE; 1870 1871 /* 1872 * Calculate the offset of the transfer and do sanity check. 1873 */ 1874 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset; 1875 1876 /* 1877 * Round up physical size for real devices. We cannot round using 1878 * v_mount's block size data because v_mount has nothing to do with 1879 * the device. i.e. it's usually '/dev'. We need the physical block 1880 * size for the device itself. 1881 * 1882 * We can't use v_rdev->si_mountpoint because it only exists when the 1883 * block device is mounted. However, we can use v_rdev. 1884 */ 1885 if (vn_isdisk(vp, NULL)) 1886 blksiz = vp->v_rdev->si_bsize_phys; 1887 else 1888 blksiz = DEV_BSIZE; 1889 1890 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1); 1891 1892 bp = getpbuf(NULL); 1893 kva = (vm_offset_t)bp->b_data; 1894 1895 /* 1896 * Map the pages to be read into the kva. 1897 */ 1898 pmap_qenter(kva, ap->a_m, pcount); 1899 1900 /* Build a minimal buffer header. */ 1901 bp->b_cmd = BUF_CMD_READ; 1902 bp->b_bcount = size; 1903 bp->b_resid = 0; 1904 bp->b_runningbufspace = size; 1905 if (size) { 1906 runningbufspace += bp->b_runningbufspace; 1907 ++runningbufcount; 1908 } 1909 1910 bp->b_bio1.bio_offset = offset; 1911 bp->b_bio1.bio_done = devfs_spec_getpages_iodone; 1912 1913 mycpu->gd_cnt.v_vnodein++; 1914 mycpu->gd_cnt.v_vnodepgsin += pcount; 1915 1916 /* Do the input. */ 1917 vn_strategy(ap->a_vp, &bp->b_bio1); 1918 1919 crit_enter(); 1920 1921 /* We definitely need to be at splbio here. */ 1922 while (bp->b_cmd != BUF_CMD_DONE) 1923 tsleep(bp, 0, "spread", 0); 1924 1925 crit_exit(); 1926 1927 if (bp->b_flags & B_ERROR) { 1928 if (bp->b_error) 1929 error = bp->b_error; 1930 else 1931 error = EIO; 1932 } 1933 1934 /* 1935 * If EOF is encountered we must zero-extend the result in order 1936 * to ensure that the page does not contain garabge. When no 1937 * error occurs, an early EOF is indicated if b_bcount got truncated. 1938 * b_resid is relative to b_bcount and should be 0, but some devices 1939 * might indicate an EOF with b_resid instead of truncating b_bcount. 1940 */ 1941 nread = bp->b_bcount - bp->b_resid; 1942 if (nread < ap->a_count) 1943 bzero((caddr_t)kva + nread, ap->a_count - nread); 1944 pmap_qremove(kva, pcount); 1945 1946 gotreqpage = 0; 1947 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) { 1948 nextoff = toff + PAGE_SIZE; 1949 m = ap->a_m[i]; 1950 1951 m->flags &= ~PG_ZERO; 1952 1953 /* 1954 * NOTE: vm_page_undirty/clear_dirty etc do not clear the 1955 * pmap modified bit. pmap modified bit should have 1956 * already been cleared. 1957 */ 1958 if (nextoff <= nread) { 1959 m->valid = VM_PAGE_BITS_ALL; 1960 vm_page_undirty(m); 1961 } else if (toff < nread) { 1962 /* 1963 * Since this is a VM request, we have to supply the 1964 * unaligned offset to allow vm_page_set_valid() 1965 * to zero sub-DEV_BSIZE'd portions of the page. 1966 */ 1967 vm_page_set_valid(m, 0, nread - toff); 1968 vm_page_clear_dirty_end_nonincl(m, 0, nread - toff); 1969 } else { 1970 m->valid = 0; 1971 vm_page_undirty(m); 1972 } 1973 1974 if (i != ap->a_reqpage) { 1975 /* 1976 * Just in case someone was asking for this page we 1977 * now tell them that it is ok to use. 1978 */ 1979 if (!error || (m->valid == VM_PAGE_BITS_ALL)) { 1980 if (m->valid) { 1981 if (m->flags & PG_WANTED) { 1982 vm_page_activate(m); 1983 } else { 1984 vm_page_deactivate(m); 1985 } 1986 vm_page_wakeup(m); 1987 } else { 1988 vm_page_free(m); 1989 } 1990 } else { 1991 vm_page_free(m); 1992 } 1993 } else if (m->valid) { 1994 gotreqpage = 1; 1995 /* 1996 * Since this is a VM request, we need to make the 1997 * entire page presentable by zeroing invalid sections. 1998 */ 1999 if (m->valid != VM_PAGE_BITS_ALL) 2000 vm_page_zero_invalid(m, FALSE); 2001 } 2002 } 2003 if (!gotreqpage) { 2004 m = ap->a_m[ap->a_reqpage]; 2005 devfs_debug(DEVFS_DEBUG_WARNING, 2006 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n", 2007 devtoname(vp->v_rdev), error, bp, bp->b_vp); 2008 devfs_debug(DEVFS_DEBUG_WARNING, 2009 " size: %d, resid: %d, a_count: %d, valid: 0x%x\n", 2010 size, bp->b_resid, ap->a_count, m->valid); 2011 devfs_debug(DEVFS_DEBUG_WARNING, 2012 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n", 2013 nread, ap->a_reqpage, (u_long)m->pindex, pcount); 2014 /* 2015 * Free the buffer header back to the swap buffer pool. 2016 */ 2017 relpbuf(bp, NULL); 2018 return VM_PAGER_ERROR; 2019 } 2020 /* 2021 * Free the buffer header back to the swap buffer pool. 2022 */ 2023 relpbuf(bp, NULL); 2024 if (DEVFS_NODE(ap->a_vp)) 2025 nanotime(&DEVFS_NODE(ap->a_vp)->mtime); 2026 return VM_PAGER_OK; 2027 } 2028 2029 static __inline 2030 int 2031 sequential_heuristic(struct uio *uio, struct file *fp) 2032 { 2033 /* 2034 * Sequential heuristic - detect sequential operation 2035 */ 2036 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 2037 uio->uio_offset == fp->f_nextoff) { 2038 /* 2039 * XXX we assume that the filesystem block size is 2040 * the default. Not true, but still gives us a pretty 2041 * good indicator of how sequential the read operations 2042 * are. 2043 */ 2044 int tmpseq = fp->f_seqcount; 2045 2046 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE; 2047 if (tmpseq > IO_SEQMAX) 2048 tmpseq = IO_SEQMAX; 2049 fp->f_seqcount = tmpseq; 2050 return(fp->f_seqcount << IO_SEQSHIFT); 2051 } 2052 2053 /* 2054 * Not sequential, quick draw-down of seqcount 2055 */ 2056 if (fp->f_seqcount > 1) 2057 fp->f_seqcount = 1; 2058 else 2059 fp->f_seqcount = 0; 2060 return(0); 2061 } 2062