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