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