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/diskslice.h> 60 #include <sys/devfs.h> 61 #include <sys/pioctl.h> 62 63 #include <machine/limits.h> 64 #include <vm/vm_page2.h> 65 #include <sys/buf2.h> 66 #include <sys/sysref2.h> 67 68 MALLOC_DECLARE(M_DEVFS); 69 #define DEVFS_BADOP (void *)devfs_badop 70 71 static int devfs_badop(struct vop_generic_args *); 72 static int devfs_access(struct vop_access_args *); 73 static int devfs_inactive(struct vop_inactive_args *); 74 static int devfs_reclaim(struct vop_reclaim_args *); 75 static int devfs_readdir(struct vop_readdir_args *); 76 static int devfs_getattr(struct vop_getattr_args *); 77 static int devfs_setattr(struct vop_setattr_args *); 78 static int devfs_readlink(struct vop_readlink_args *); 79 static int devfs_print(struct vop_print_args *); 80 81 static int devfs_nresolve(struct vop_nresolve_args *); 82 static int devfs_nlookupdotdot(struct vop_nlookupdotdot_args *); 83 static int devfs_nsymlink(struct vop_nsymlink_args *); 84 static int devfs_nremove(struct vop_nremove_args *); 85 86 static int devfs_spec_open(struct vop_open_args *); 87 static int devfs_spec_close(struct vop_close_args *); 88 static int devfs_spec_fsync(struct vop_fsync_args *); 89 90 static int devfs_spec_read(struct vop_read_args *); 91 static int devfs_spec_write(struct vop_write_args *); 92 static int devfs_spec_ioctl(struct vop_ioctl_args *); 93 static int devfs_spec_poll(struct vop_poll_args *); 94 static int devfs_spec_kqfilter(struct vop_kqfilter_args *); 95 static int devfs_spec_strategy(struct vop_strategy_args *); 96 static void devfs_spec_strategy_done(struct bio *); 97 static int devfs_spec_freeblks(struct vop_freeblks_args *); 98 static int devfs_spec_bmap(struct vop_bmap_args *); 99 static int devfs_spec_advlock(struct vop_advlock_args *); 100 static void devfs_spec_getpages_iodone(struct bio *); 101 static int devfs_spec_getpages(struct vop_getpages_args *); 102 103 104 static int devfs_specf_close(struct file *); 105 static int devfs_specf_read(struct file *, struct uio *, struct ucred *, int); 106 static int devfs_specf_write(struct file *, struct uio *, struct ucred *, int); 107 static int devfs_specf_stat(struct file *, struct stat *, struct ucred *); 108 static int devfs_specf_kqfilter(struct file *, struct knote *); 109 static int devfs_specf_poll(struct file *, int, struct ucred *); 110 static int devfs_specf_ioctl(struct file *, u_long, caddr_t, 111 struct ucred *, struct sysmsg *); 112 static __inline int sequential_heuristic(struct uio *, struct file *); 113 114 extern struct lock devfs_lock; 115 116 /* 117 * devfs vnode operations for regular files 118 */ 119 struct vop_ops devfs_vnode_norm_vops = { 120 .vop_default = vop_defaultop, 121 .vop_access = devfs_access, 122 .vop_advlock = DEVFS_BADOP, 123 .vop_bmap = DEVFS_BADOP, 124 .vop_close = vop_stdclose, 125 .vop_getattr = devfs_getattr, 126 .vop_inactive = devfs_inactive, 127 .vop_ncreate = DEVFS_BADOP, 128 .vop_nresolve = devfs_nresolve, 129 .vop_nlookupdotdot = devfs_nlookupdotdot, 130 .vop_nlink = DEVFS_BADOP, 131 .vop_nmkdir = DEVFS_BADOP, 132 .vop_nmknod = DEVFS_BADOP, 133 .vop_nremove = devfs_nremove, 134 .vop_nrename = DEVFS_BADOP, 135 .vop_nrmdir = DEVFS_BADOP, 136 .vop_nsymlink = devfs_nsymlink, 137 .vop_open = vop_stdopen, 138 .vop_pathconf = vop_stdpathconf, 139 .vop_print = devfs_print, 140 .vop_read = DEVFS_BADOP, 141 .vop_readdir = devfs_readdir, 142 .vop_readlink = devfs_readlink, 143 .vop_reclaim = devfs_reclaim, 144 .vop_setattr = devfs_setattr, 145 .vop_write = DEVFS_BADOP, 146 .vop_ioctl = DEVFS_BADOP 147 }; 148 149 /* 150 * devfs vnode operations for character devices 151 */ 152 struct vop_ops devfs_vnode_dev_vops = { 153 .vop_default = vop_defaultop, 154 .vop_access = devfs_access, 155 .vop_advlock = devfs_spec_advlock, 156 .vop_bmap = devfs_spec_bmap, 157 .vop_close = devfs_spec_close, 158 .vop_freeblks = devfs_spec_freeblks, 159 .vop_fsync = devfs_spec_fsync, 160 .vop_getattr = devfs_getattr, 161 .vop_getpages = devfs_spec_getpages, 162 .vop_inactive = devfs_inactive, 163 .vop_open = devfs_spec_open, 164 .vop_pathconf = vop_stdpathconf, 165 .vop_print = devfs_print, 166 .vop_poll = devfs_spec_poll, 167 .vop_kqfilter = devfs_spec_kqfilter, 168 .vop_read = devfs_spec_read, 169 .vop_readdir = DEVFS_BADOP, 170 .vop_readlink = DEVFS_BADOP, 171 .vop_reclaim = devfs_reclaim, 172 .vop_setattr = devfs_setattr, 173 .vop_strategy = devfs_spec_strategy, 174 .vop_write = devfs_spec_write, 175 .vop_ioctl = devfs_spec_ioctl 176 }; 177 178 struct vop_ops *devfs_vnode_dev_vops_p = &devfs_vnode_dev_vops; 179 180 struct fileops devfs_dev_fileops = { 181 .fo_read = devfs_specf_read, 182 .fo_write = devfs_specf_write, 183 .fo_ioctl = devfs_specf_ioctl, 184 .fo_poll = devfs_specf_poll, 185 .fo_kqfilter = devfs_specf_kqfilter, 186 .fo_stat = devfs_specf_stat, 187 .fo_close = devfs_specf_close, 188 .fo_shutdown = nofo_shutdown 189 }; 190 191 /* 192 * These two functions are possibly temporary hacks for 193 * devices (aka the pty code) which want to control the 194 * node attributes themselves. 195 * 196 * XXX we may ultimately desire to simply remove the uid/gid/mode 197 * from the node entirely. 198 */ 199 static __inline void 200 node_sync_dev_get(struct devfs_node *node) 201 { 202 cdev_t dev; 203 204 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) { 205 node->uid = dev->si_uid; 206 node->gid = dev->si_gid; 207 node->mode = dev->si_perms; 208 } 209 } 210 211 static __inline void 212 node_sync_dev_set(struct devfs_node *node) 213 { 214 cdev_t dev; 215 216 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) { 217 dev->si_uid = node->uid; 218 dev->si_gid = node->gid; 219 dev->si_perms = node->mode; 220 } 221 } 222 223 /* 224 * generic entry point for unsupported operations 225 */ 226 static int 227 devfs_badop(struct vop_generic_args *ap) 228 { 229 return (EIO); 230 } 231 232 233 static int 234 devfs_access(struct vop_access_args *ap) 235 { 236 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 237 int error; 238 239 if (!devfs_node_is_accessible(node)) 240 return ENOENT; 241 node_sync_dev_get(node); 242 error = vop_helper_access(ap, node->uid, node->gid, 243 node->mode, node->flags); 244 245 return error; 246 } 247 248 249 static int 250 devfs_inactive(struct vop_inactive_args *ap) 251 { 252 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 253 254 if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0) 255 vrecycle(ap->a_vp); 256 return 0; 257 } 258 259 260 static int 261 devfs_reclaim(struct vop_reclaim_args *ap) 262 { 263 struct devfs_node *node; 264 struct vnode *vp; 265 int locked; 266 267 /* 268 * Check if it is locked already. if not, we acquire the devfs lock 269 */ 270 if (!(lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE) { 271 lockmgr(&devfs_lock, LK_EXCLUSIVE); 272 locked = 1; 273 } else { 274 locked = 0; 275 } 276 277 /* 278 * Get rid of the devfs_node if it is no longer linked into the 279 * topology. 280 */ 281 vp = ap->a_vp; 282 if ((node = DEVFS_NODE(vp)) != NULL) { 283 node->v_node = NULL; 284 if ((node->flags & DEVFS_NODE_LINKED) == 0) 285 devfs_freep(node); 286 } 287 288 if (locked) 289 lockmgr(&devfs_lock, LK_RELEASE); 290 291 /* 292 * v_rdev needs to be properly released using v_release_rdev 293 * Make sure v_data is NULL as well. 294 */ 295 vp->v_data = NULL; 296 v_release_rdev(vp); 297 return 0; 298 } 299 300 301 static int 302 devfs_readdir(struct vop_readdir_args *ap) 303 { 304 struct devfs_node *dnode = DEVFS_NODE(ap->a_vp); 305 struct devfs_node *node; 306 int cookie_index; 307 int ncookies; 308 int error2; 309 int error; 310 int r; 311 off_t *cookies; 312 off_t saveoff; 313 314 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n"); 315 316 if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX) 317 return (EINVAL); 318 if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0) 319 return (error); 320 321 if (!devfs_node_is_accessible(dnode)) { 322 vn_unlock(ap->a_vp); 323 return ENOENT; 324 } 325 326 lockmgr(&devfs_lock, LK_EXCLUSIVE); 327 328 saveoff = ap->a_uio->uio_offset; 329 330 if (ap->a_ncookies) { 331 ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */ 332 if (ncookies > 256) 333 ncookies = 256; 334 cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK); 335 cookie_index = 0; 336 } else { 337 ncookies = -1; 338 cookies = NULL; 339 cookie_index = 0; 340 } 341 342 nanotime(&dnode->atime); 343 344 if (saveoff == 0) { 345 r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino, 346 DT_DIR, 1, "."); 347 if (r) 348 goto done; 349 if (cookies) 350 cookies[cookie_index] = saveoff; 351 saveoff++; 352 cookie_index++; 353 if (cookie_index == ncookies) 354 goto done; 355 } 356 357 if (saveoff == 1) { 358 if (dnode->parent) { 359 r = vop_write_dirent(&error, ap->a_uio, 360 dnode->parent->d_dir.d_ino, 361 DT_DIR, 2, ".."); 362 } else { 363 r = vop_write_dirent(&error, ap->a_uio, 364 dnode->d_dir.d_ino, 365 DT_DIR, 2, ".."); 366 } 367 if (r) 368 goto done; 369 if (cookies) 370 cookies[cookie_index] = saveoff; 371 saveoff++; 372 cookie_index++; 373 if (cookie_index == ncookies) 374 goto done; 375 } 376 377 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { 378 if ((node->flags & DEVFS_HIDDEN) || 379 (node->flags & DEVFS_INVISIBLE)) { 380 continue; 381 } 382 383 /* 384 * If the node type is a valid devfs alias, then we make sure that the 385 * target isn't hidden. If it is, we don't show the link in the 386 * directory listing. 387 */ 388 if ((node->node_type == Plink) && (node->link_target != NULL) && 389 (node->link_target->flags & DEVFS_HIDDEN)) 390 continue; 391 392 if (node->cookie < saveoff) 393 continue; 394 395 saveoff = node->cookie; 396 397 error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino, 398 node->d_dir.d_type, 399 node->d_dir.d_namlen, 400 node->d_dir.d_name); 401 402 if (error2) 403 break; 404 405 saveoff++; 406 407 if (cookies) 408 cookies[cookie_index] = node->cookie; 409 ++cookie_index; 410 if (cookie_index == ncookies) 411 break; 412 } 413 414 done: 415 lockmgr(&devfs_lock, LK_RELEASE); 416 vn_unlock(ap->a_vp); 417 418 ap->a_uio->uio_offset = saveoff; 419 if (error && cookie_index == 0) { 420 if (cookies) { 421 kfree(cookies, M_TEMP); 422 *ap->a_ncookies = 0; 423 *ap->a_cookies = NULL; 424 } 425 } else { 426 if (cookies) { 427 *ap->a_ncookies = cookie_index; 428 *ap->a_cookies = cookies; 429 } 430 } 431 return (error); 432 } 433 434 435 static int 436 devfs_nresolve(struct vop_nresolve_args *ap) 437 { 438 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); 439 struct devfs_node *node, *found = NULL; 440 struct namecache *ncp; 441 struct vnode *vp = NULL; 442 int error = 0; 443 int len; 444 int depth; 445 446 ncp = ap->a_nch->ncp; 447 len = ncp->nc_nlen; 448 449 if (!devfs_node_is_accessible(dnode)) 450 return ENOENT; 451 452 lockmgr(&devfs_lock, LK_EXCLUSIVE); 453 454 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) { 455 error = ENOENT; 456 cache_setvp(ap->a_nch, NULL); 457 goto out; 458 } 459 460 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { 461 if (len == node->d_dir.d_namlen) { 462 if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) { 463 found = node; 464 break; 465 } 466 } 467 } 468 469 if (found) { 470 depth = 0; 471 while ((found->node_type == Plink) && (found->link_target)) { 472 if (depth >= 8) { 473 devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8"); 474 break; 475 } 476 477 found = found->link_target; 478 ++depth; 479 } 480 481 if (!(found->flags & DEVFS_HIDDEN)) 482 devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found); 483 } 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 struct partinfo pinfo; 528 int error = 0; 529 530 #if 0 531 if (!devfs_node_is_accessible(node)) 532 return ENOENT; 533 #endif 534 node_sync_dev_get(node); 535 536 lockmgr(&devfs_lock, LK_EXCLUSIVE); 537 538 /* start by zeroing out the attributes */ 539 VATTR_NULL(vap); 540 541 /* next do all the common fields */ 542 vap->va_type = ap->a_vp->v_type; 543 vap->va_mode = node->mode; 544 vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ; 545 vap->va_flags = 0; /* XXX: what should this be? */ 546 vap->va_blocksize = DEV_BSIZE; 547 vap->va_bytes = vap->va_size = sizeof(struct devfs_node); 548 549 vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0]; 550 551 vap->va_atime = node->atime; 552 vap->va_mtime = node->mtime; 553 vap->va_ctime = node->ctime; 554 555 vap->va_nlink = 1; /* number of references to file */ 556 557 vap->va_uid = node->uid; 558 vap->va_gid = node->gid; 559 560 vap->va_rmajor = 0; 561 vap->va_rminor = 0; 562 563 if ((node->node_type == Pdev) && node->d_dev) { 564 reference_dev(node->d_dev); 565 vap->va_rminor = node->d_dev->si_uminor; 566 release_dev(node->d_dev); 567 } 568 569 /* For a softlink the va_size is the length of the softlink */ 570 if (node->symlink_name != 0) { 571 vap->va_bytes = vap->va_size = node->symlink_namelen; 572 } 573 574 /* 575 * For a disk-type device, va_size is the size of the underlying 576 * device, so that lseek() works properly. 577 */ 578 if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) { 579 bzero(&pinfo, sizeof(pinfo)); 580 error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo, 581 0, proc0.p_ucred, NULL); 582 if ((error == 0) && (pinfo.media_blksize != 0)) { 583 vap->va_size = pinfo.media_size; 584 } else { 585 vap->va_size = 0; 586 error = 0; 587 } 588 } 589 590 lockmgr(&devfs_lock, LK_RELEASE); 591 592 return (error); 593 } 594 595 596 static int 597 devfs_setattr(struct vop_setattr_args *ap) 598 { 599 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 600 struct vattr *vap; 601 int error = 0; 602 603 if (!devfs_node_is_accessible(node)) 604 return ENOENT; 605 node_sync_dev_get(node); 606 607 lockmgr(&devfs_lock, LK_EXCLUSIVE); 608 609 vap = ap->a_vap; 610 611 if (vap->va_uid != (uid_t)VNOVAL) { 612 if ((ap->a_cred->cr_uid != node->uid) && 613 (!groupmember(node->gid, ap->a_cred))) { 614 error = priv_check(curthread, PRIV_VFS_CHOWN); 615 if (error) 616 goto out; 617 } 618 node->uid = vap->va_uid; 619 } 620 621 if (vap->va_gid != (uid_t)VNOVAL) { 622 if ((ap->a_cred->cr_uid != node->uid) && 623 (!groupmember(node->gid, ap->a_cred))) { 624 error = priv_check(curthread, PRIV_VFS_CHOWN); 625 if (error) 626 goto out; 627 } 628 node->gid = vap->va_gid; 629 } 630 631 if (vap->va_mode != (mode_t)VNOVAL) { 632 if (ap->a_cred->cr_uid != node->uid) { 633 error = priv_check(curthread, PRIV_VFS_ADMIN); 634 if (error) 635 goto out; 636 } 637 node->mode = vap->va_mode; 638 } 639 640 out: 641 node_sync_dev_set(node); 642 nanotime(&node->ctime); 643 lockmgr(&devfs_lock, LK_RELEASE); 644 645 return error; 646 } 647 648 649 static int 650 devfs_readlink(struct vop_readlink_args *ap) 651 { 652 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 653 int ret; 654 655 if (!devfs_node_is_accessible(node)) 656 return ENOENT; 657 658 lockmgr(&devfs_lock, LK_EXCLUSIVE); 659 ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio); 660 lockmgr(&devfs_lock, LK_RELEASE); 661 662 return ret; 663 } 664 665 666 static int 667 devfs_print(struct vop_print_args *ap) 668 { 669 return (0); 670 } 671 672 673 static int 674 devfs_nsymlink(struct vop_nsymlink_args *ap) 675 { 676 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); 677 struct devfs_node *node; 678 size_t targetlen; 679 680 if (!devfs_node_is_accessible(dnode)) 681 return ENOENT; 682 683 ap->a_vap->va_type = VLNK; 684 685 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) 686 goto out; 687 688 lockmgr(&devfs_lock, LK_EXCLUSIVE); 689 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Plink, 690 ap->a_nch->ncp->nc_name, dnode, NULL); 691 692 targetlen = strlen(ap->a_target); 693 if (*ap->a_vpp) { 694 node = DEVFS_NODE(*ap->a_vpp); 695 node->flags |= DEVFS_USER_CREATED; 696 node->symlink_namelen = targetlen; 697 node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK); 698 memcpy(node->symlink_name, ap->a_target, targetlen); 699 node->symlink_name[targetlen] = '\0'; 700 cache_setunresolved(ap->a_nch); 701 cache_setvp(ap->a_nch, *ap->a_vpp); 702 } 703 lockmgr(&devfs_lock, LK_RELEASE); 704 out: 705 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0); 706 } 707 708 709 static int 710 devfs_nremove(struct vop_nremove_args *ap) 711 { 712 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); 713 struct devfs_node *node; 714 struct namecache *ncp; 715 int error = ENOENT; 716 717 ncp = ap->a_nch->ncp; 718 719 if (!devfs_node_is_accessible(dnode)) 720 return ENOENT; 721 722 lockmgr(&devfs_lock, LK_EXCLUSIVE); 723 724 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) 725 goto out; 726 727 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { 728 if (ncp->nc_nlen != node->d_dir.d_namlen) 729 continue; 730 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen)) 731 continue; 732 733 /* 734 * only allow removal of user created stuff (e.g. symlinks) 735 */ 736 if ((node->flags & DEVFS_USER_CREATED) == 0) { 737 error = EPERM; 738 goto out; 739 } else { 740 if (node->v_node) 741 cache_inval_vp(node->v_node, CINV_DESTROY); 742 devfs_unlinkp(node); 743 error = 0; 744 break; 745 } 746 } 747 748 cache_setunresolved(ap->a_nch); 749 cache_setvp(ap->a_nch, NULL); 750 751 out: 752 lockmgr(&devfs_lock, LK_RELEASE); 753 return error; 754 } 755 756 757 static int 758 devfs_spec_open(struct vop_open_args *ap) 759 { 760 struct vnode *vp = ap->a_vp; 761 struct vnode *orig_vp = NULL; 762 struct devfs_node *node = DEVFS_NODE(vp); 763 struct devfs_node *newnode; 764 cdev_t dev, ndev = NULL; 765 int error = 0; 766 767 if (node) { 768 if (node->d_dev == NULL) 769 return ENXIO; 770 if (!devfs_node_is_accessible(node)) 771 return ENOENT; 772 } 773 774 if ((dev = vp->v_rdev) == NULL) 775 return ENXIO; 776 777 if (node && ap->a_fp) { 778 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n"); 779 lockmgr(&devfs_lock, LK_EXCLUSIVE); 780 781 ndev = devfs_clone(dev, node->d_dir.d_name, node->d_dir.d_namlen, 782 ap->a_mode, ap->a_cred); 783 if (ndev != NULL) { 784 newnode = devfs_create_device_node( 785 DEVFS_MNTDATA(vp->v_mount)->root_node, 786 ndev, NULL, NULL); 787 /* XXX: possibly destroy device if this happens */ 788 789 if (newnode != NULL) { 790 dev = ndev; 791 devfs_link_dev(dev); 792 793 devfs_debug(DEVFS_DEBUG_DEBUG, 794 "parent here is: %s, node is: |%s|\n", 795 ((node->parent->node_type == Proot) ? 796 "ROOT!" : node->parent->d_dir.d_name), 797 newnode->d_dir.d_name); 798 devfs_debug(DEVFS_DEBUG_DEBUG, 799 "test: %s\n", 800 ((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name); 801 802 /* 803 * orig_vp is set to the original vp if we cloned. 804 */ 805 /* node->flags |= DEVFS_CLONED; */ 806 devfs_allocv(&vp, newnode); 807 orig_vp = ap->a_vp; 808 ap->a_vp = vp; 809 } 810 } 811 lockmgr(&devfs_lock, LK_RELEASE); 812 } 813 814 devfs_debug(DEVFS_DEBUG_DEBUG, 815 "devfs_spec_open() called on %s! \n", 816 dev->si_name); 817 818 /* 819 * Make this field valid before any I/O in ->d_open 820 */ 821 if (!dev->si_iosize_max) 822 dev->si_iosize_max = DFLTPHYS; 823 824 if (dev_dflags(dev) & D_TTY) 825 vp->v_flag |= VISTTY; 826 827 vn_unlock(vp); 828 error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred); 829 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 830 831 /* 832 * Clean up any cloned vp if we error out. 833 */ 834 if (error) { 835 if (orig_vp) { 836 vput(vp); 837 ap->a_vp = orig_vp; 838 /* orig_vp = NULL; */ 839 } 840 return error; 841 } 842 843 /* 844 * This checks if the disk device is going to be opened for writing. 845 * It will be only allowed in the cases where securelevel permits it 846 * and it's not mounted R/W. 847 */ 848 if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) && 849 (ap->a_cred != FSCRED)) { 850 851 /* Very secure mode. No open for writing allowed */ 852 if (securelevel >= 2) 853 return EPERM; 854 855 /* 856 * If it is mounted R/W, do not allow to open for writing. 857 * In the case it's mounted read-only but securelevel 858 * is >= 1, then do not allow opening for writing either. 859 */ 860 if (vfs_mountedon(vp)) { 861 if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY)) 862 return EBUSY; 863 else if (securelevel >= 1) 864 return EPERM; 865 } 866 } 867 868 if (dev_dflags(dev) & D_TTY) { 869 if (dev->si_tty) { 870 struct tty *tp; 871 tp = dev->si_tty; 872 if (!tp->t_stop) { 873 devfs_debug(DEVFS_DEBUG_DEBUG, 874 "devfs: no t_stop\n"); 875 tp->t_stop = nottystop; 876 } 877 } 878 } 879 880 881 if (vn_isdisk(vp, NULL)) { 882 if (!dev->si_bsize_phys) 883 dev->si_bsize_phys = DEV_BSIZE; 884 vinitvmio(vp, IDX_TO_OFF(INT_MAX)); 885 } 886 887 vop_stdopen(ap); 888 #if 0 889 if (node) 890 nanotime(&node->atime); 891 #endif 892 893 if (orig_vp) 894 vn_unlock(vp); 895 896 /* Ugly pty magic, to make pty devices appear once they are opened */ 897 if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY) 898 node->flags &= ~DEVFS_INVISIBLE; 899 900 if (ap->a_fp) { 901 ap->a_fp->f_type = DTYPE_VNODE; 902 ap->a_fp->f_flag = ap->a_mode & FMASK; 903 ap->a_fp->f_ops = &devfs_dev_fileops; 904 ap->a_fp->f_data = vp; 905 } 906 907 return 0; 908 } 909 910 911 static int 912 devfs_spec_close(struct vop_close_args *ap) 913 { 914 struct devfs_node *node = DEVFS_NODE(ap->a_vp); 915 struct proc *p = curproc; 916 struct vnode *vp = ap->a_vp; 917 cdev_t dev = vp->v_rdev; 918 int error = 0; 919 int needrelock; 920 921 devfs_debug(DEVFS_DEBUG_DEBUG, 922 "devfs_spec_close() called on %s! \n", 923 dev->si_name); 924 925 /* 926 * A couple of hacks for devices and tty devices. The 927 * vnode ref count cannot be used to figure out the 928 * last close, but we can use v_opencount now that 929 * revoke works properly. 930 * 931 * Detect the last close on a controlling terminal and clear 932 * the session (half-close). 933 */ 934 if (dev) 935 reference_dev(dev); 936 937 if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) { 938 p->p_session->s_ttyvp = NULL; 939 vrele(vp); 940 } 941 942 /* 943 * Vnodes can be opened and closed multiple times. Do not really 944 * close the device unless (1) it is being closed forcibly, 945 * (2) the device wants to track closes, or (3) this is the last 946 * vnode doing its last close on the device. 947 * 948 * XXX the VXLOCK (force close) case can leave vnodes referencing 949 * a closed device. This might not occur now that our revoke is 950 * fixed. 951 */ 952 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n"); 953 if (dev && ((vp->v_flag & VRECLAIMED) || 954 (dev_dflags(dev) & D_TRACKCLOSE) || 955 (vp->v_opencount == 1))) { 956 /* 957 * Unlock around dev_dclose() 958 */ 959 needrelock = 0; 960 if (vn_islocked(vp)) { 961 needrelock = 1; 962 vn_unlock(vp); 963 } 964 error = dev_dclose(dev, ap->a_fflag, S_IFCHR); 965 966 /* 967 * Ugly pty magic, to make pty devices disappear again once 968 * they are closed 969 */ 970 if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY) 971 node->flags |= DEVFS_INVISIBLE; 972 973 if (needrelock) 974 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 975 } else { 976 error = 0; 977 } 978 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n"); 979 980 /* 981 * Track the actual opens and closes on the vnode. The last close 982 * disassociates the rdev. If the rdev is already disassociated or 983 * the opencount is already 0, the vnode might have been revoked 984 * and no further opencount tracking occurs. 985 */ 986 if (dev) 987 release_dev(dev); 988 if (vp->v_opencount > 0) 989 vop_stdclose(ap); 990 return(error); 991 992 } 993 994 995 static int 996 devfs_specf_close(struct file *fp) 997 { 998 struct vnode *vp = (struct vnode *)fp->f_data; 999 int error; 1000 1001 get_mplock(); 1002 fp->f_ops = &badfileops; 1003 error = vn_close(vp, fp->f_flag); 1004 rel_mplock(); 1005 1006 return (error); 1007 } 1008 1009 1010 /* 1011 * Device-optimized file table vnode read routine. 1012 * 1013 * This bypasses the VOP table and talks directly to the device. Most 1014 * filesystems just route to specfs and can make this optimization. 1015 * 1016 * MPALMOSTSAFE - acquires mplock 1017 */ 1018 static int 1019 devfs_specf_read(struct file *fp, struct uio *uio, 1020 struct ucred *cred, int flags) 1021 { 1022 struct devfs_node *node; 1023 struct vnode *vp; 1024 int ioflag; 1025 int error; 1026 cdev_t dev; 1027 1028 get_mplock(); 1029 KASSERT(uio->uio_td == curthread, 1030 ("uio_td %p is not td %p", uio->uio_td, curthread)); 1031 1032 vp = (struct vnode *)fp->f_data; 1033 if (vp == NULL || vp->v_type == VBAD) { 1034 error = EBADF; 1035 goto done; 1036 } 1037 node = DEVFS_NODE(vp); 1038 1039 if ((dev = vp->v_rdev) == NULL) { 1040 error = EBADF; 1041 goto done; 1042 } 1043 1044 reference_dev(dev); 1045 1046 if (uio->uio_resid == 0) { 1047 error = 0; 1048 goto done; 1049 } 1050 if ((flags & O_FOFFSET) == 0) 1051 uio->uio_offset = fp->f_offset; 1052 1053 ioflag = 0; 1054 if (flags & O_FBLOCKING) { 1055 /* ioflag &= ~IO_NDELAY; */ 1056 } else if (flags & O_FNONBLOCKING) { 1057 ioflag |= IO_NDELAY; 1058 } else if (fp->f_flag & FNONBLOCK) { 1059 ioflag |= IO_NDELAY; 1060 } 1061 if (flags & O_FBUFFERED) { 1062 /* ioflag &= ~IO_DIRECT; */ 1063 } else if (flags & O_FUNBUFFERED) { 1064 ioflag |= IO_DIRECT; 1065 } else if (fp->f_flag & O_DIRECT) { 1066 ioflag |= IO_DIRECT; 1067 } 1068 ioflag |= sequential_heuristic(uio, fp); 1069 1070 error = dev_dread(dev, uio, ioflag); 1071 1072 release_dev(dev); 1073 if (node) 1074 nanotime(&node->atime); 1075 if ((flags & O_FOFFSET) == 0) 1076 fp->f_offset = uio->uio_offset; 1077 fp->f_nextoff = uio->uio_offset; 1078 done: 1079 rel_mplock(); 1080 return (error); 1081 } 1082 1083 1084 static int 1085 devfs_specf_write(struct file *fp, struct uio *uio, 1086 struct ucred *cred, int flags) 1087 { 1088 struct devfs_node *node; 1089 struct vnode *vp; 1090 int ioflag; 1091 int error; 1092 cdev_t dev; 1093 1094 get_mplock(); 1095 KASSERT(uio->uio_td == curthread, 1096 ("uio_td %p is not p %p", uio->uio_td, curthread)); 1097 1098 vp = (struct vnode *)fp->f_data; 1099 if (vp == NULL || vp->v_type == VBAD) { 1100 error = EBADF; 1101 goto done; 1102 } 1103 node = DEVFS_NODE(vp); 1104 if (vp->v_type == VREG) 1105 bwillwrite(uio->uio_resid); 1106 vp = (struct vnode *)fp->f_data; 1107 1108 if ((dev = vp->v_rdev) == NULL) { 1109 error = EBADF; 1110 goto done; 1111 } 1112 reference_dev(dev); 1113 1114 if ((flags & O_FOFFSET) == 0) 1115 uio->uio_offset = fp->f_offset; 1116 1117 ioflag = IO_UNIT; 1118 if (vp->v_type == VREG && 1119 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { 1120 ioflag |= IO_APPEND; 1121 } 1122 1123 if (flags & O_FBLOCKING) { 1124 /* ioflag &= ~IO_NDELAY; */ 1125 } else if (flags & O_FNONBLOCKING) { 1126 ioflag |= IO_NDELAY; 1127 } else if (fp->f_flag & FNONBLOCK) { 1128 ioflag |= IO_NDELAY; 1129 } 1130 if (flags & O_FBUFFERED) { 1131 /* ioflag &= ~IO_DIRECT; */ 1132 } else if (flags & O_FUNBUFFERED) { 1133 ioflag |= IO_DIRECT; 1134 } else if (fp->f_flag & O_DIRECT) { 1135 ioflag |= IO_DIRECT; 1136 } 1137 if (flags & O_FASYNCWRITE) { 1138 /* ioflag &= ~IO_SYNC; */ 1139 } else if (flags & O_FSYNCWRITE) { 1140 ioflag |= IO_SYNC; 1141 } else if (fp->f_flag & O_FSYNC) { 1142 ioflag |= IO_SYNC; 1143 } 1144 1145 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) 1146 ioflag |= IO_SYNC; 1147 ioflag |= sequential_heuristic(uio, fp); 1148 1149 error = dev_dwrite(dev, uio, ioflag); 1150 1151 release_dev(dev); 1152 if (node) { 1153 nanotime(&node->atime); 1154 nanotime(&node->mtime); 1155 } 1156 1157 if ((flags & O_FOFFSET) == 0) 1158 fp->f_offset = uio->uio_offset; 1159 fp->f_nextoff = uio->uio_offset; 1160 done: 1161 rel_mplock(); 1162 return (error); 1163 } 1164 1165 1166 static int 1167 devfs_specf_stat(struct file *fp, struct stat *sb, struct ucred *cred) 1168 { 1169 struct vnode *vp; 1170 int error; 1171 1172 get_mplock(); 1173 vp = (struct vnode *)fp->f_data; 1174 error = vn_stat(vp, sb, cred); 1175 if (error) { 1176 rel_mplock(); 1177 return (error); 1178 } 1179 1180 struct vattr vattr; 1181 struct vattr *vap; 1182 u_short mode; 1183 cdev_t dev; 1184 1185 vap = &vattr; 1186 error = VOP_GETATTR(vp, vap); 1187 if (error) { 1188 rel_mplock(); 1189 return (error); 1190 } 1191 1192 /* 1193 * Zero the spare stat fields 1194 */ 1195 sb->st_lspare = 0; 1196 sb->st_qspare = 0; 1197 1198 /* 1199 * Copy from vattr table ... or not in case it's a cloned device 1200 */ 1201 if (vap->va_fsid != VNOVAL) 1202 sb->st_dev = vap->va_fsid; 1203 else 1204 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 1205 1206 sb->st_ino = vap->va_fileid; 1207 1208 mode = vap->va_mode; 1209 mode |= S_IFCHR; 1210 sb->st_mode = mode; 1211 1212 if (vap->va_nlink > (nlink_t)-1) 1213 sb->st_nlink = (nlink_t)-1; 1214 else 1215 sb->st_nlink = vap->va_nlink; 1216 sb->st_uid = vap->va_uid; 1217 sb->st_gid = vap->va_gid; 1218 sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev); 1219 sb->st_size = vap->va_bytes; 1220 sb->st_atimespec = vap->va_atime; 1221 sb->st_mtimespec = vap->va_mtime; 1222 sb->st_ctimespec = vap->va_ctime; 1223 1224 /* 1225 * A VCHR and VBLK device may track the last access and last modified 1226 * time independantly of the filesystem. This is particularly true 1227 * because device read and write calls may bypass the filesystem. 1228 */ 1229 if (vp->v_type == VCHR || vp->v_type == VBLK) { 1230 dev = vp->v_rdev; 1231 if (dev != NULL) { 1232 if (dev->si_lastread) { 1233 sb->st_atimespec.tv_sec = dev->si_lastread; 1234 sb->st_atimespec.tv_nsec = 0; 1235 } 1236 if (dev->si_lastwrite) { 1237 sb->st_atimespec.tv_sec = dev->si_lastwrite; 1238 sb->st_atimespec.tv_nsec = 0; 1239 } 1240 } 1241 } 1242 1243 /* 1244 * According to www.opengroup.org, the meaning of st_blksize is 1245 * "a filesystem-specific preferred I/O block size for this 1246 * object. In some filesystem types, this may vary from file 1247 * to file" 1248 * Default to PAGE_SIZE after much discussion. 1249 */ 1250 1251 sb->st_blksize = PAGE_SIZE; 1252 1253 sb->st_flags = vap->va_flags; 1254 1255 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0); 1256 if (error) 1257 sb->st_gen = 0; 1258 else 1259 sb->st_gen = (u_int32_t)vap->va_gen; 1260 1261 sb->st_blocks = vap->va_bytes / S_BLKSIZE; 1262 sb->st_fsmid = vap->va_fsmid; 1263 1264 rel_mplock(); 1265 return (0); 1266 } 1267 1268 1269 static int 1270 devfs_specf_kqfilter(struct file *fp, struct knote *kn) 1271 { 1272 struct vnode *vp; 1273 int error; 1274 cdev_t dev; 1275 1276 get_mplock(); 1277 1278 vp = (struct vnode *)fp->f_data; 1279 if (vp == NULL || vp->v_type == VBAD) { 1280 error = EBADF; 1281 goto done; 1282 } 1283 if ((dev = vp->v_rdev) == NULL) { 1284 error = EBADF; 1285 goto done; 1286 } 1287 reference_dev(dev); 1288 1289 error = dev_dkqfilter(dev, kn); 1290 1291 release_dev(dev); 1292 1293 done: 1294 rel_mplock(); 1295 return (error); 1296 } 1297 1298 1299 static int 1300 devfs_specf_poll(struct file *fp, int events, struct ucred *cred) 1301 { 1302 struct devfs_node *node; 1303 struct vnode *vp; 1304 int error; 1305 cdev_t dev; 1306 1307 get_mplock(); 1308 1309 vp = (struct vnode *)fp->f_data; 1310 if (vp == NULL || vp->v_type == VBAD) { 1311 error = EBADF; 1312 goto done; 1313 } 1314 node = DEVFS_NODE(vp); 1315 1316 if ((dev = vp->v_rdev) == NULL) { 1317 error = EBADF; 1318 goto done; 1319 } 1320 reference_dev(dev); 1321 error = dev_dpoll(dev, events); 1322 1323 release_dev(dev); 1324 1325 #if 0 1326 if (node) 1327 nanotime(&node->atime); 1328 #endif 1329 done: 1330 rel_mplock(); 1331 return (error); 1332 } 1333 1334 1335 /* 1336 * MPALMOSTSAFE - acquires mplock 1337 */ 1338 static int 1339 devfs_specf_ioctl(struct file *fp, u_long com, caddr_t data, 1340 struct ucred *ucred, struct sysmsg *msg) 1341 { 1342 struct devfs_node *node; 1343 struct vnode *vp; 1344 struct vnode *ovp; 1345 cdev_t dev; 1346 int error; 1347 struct fiodname_args *name_args; 1348 size_t namlen; 1349 const char *name; 1350 1351 get_mplock(); 1352 vp = ((struct vnode *)fp->f_data); 1353 if ((dev = vp->v_rdev) == NULL) { 1354 error = EBADF; /* device was revoked */ 1355 goto out; 1356 } 1357 1358 node = DEVFS_NODE(vp); 1359 1360 devfs_debug(DEVFS_DEBUG_DEBUG, 1361 "devfs_specf_ioctl() called! for dev %s\n", 1362 dev->si_name); 1363 1364 if (com == FIODTYPE) { 1365 *(int *)data = dev_dflags(dev) & D_TYPEMASK; 1366 error = 0; 1367 goto out; 1368 } else if (com == FIODNAME) { 1369 name_args = (struct fiodname_args *)data; 1370 name = dev->si_name; 1371 namlen = strlen(name) + 1; 1372 1373 devfs_debug(DEVFS_DEBUG_DEBUG, 1374 "ioctl, got: FIODNAME for %s\n", name); 1375 1376 if (namlen <= name_args->len) 1377 error = copyout(dev->si_name, name_args->name, namlen); 1378 else 1379 error = EINVAL; 1380 1381 devfs_debug(DEVFS_DEBUG_DEBUG, 1382 "ioctl stuff: error: %d\n", error); 1383 goto out; 1384 } 1385 reference_dev(dev); 1386 error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg); 1387 release_dev(dev); 1388 #if 0 1389 if (node) { 1390 nanotime(&node->atime); 1391 nanotime(&node->mtime); 1392 } 1393 #endif 1394 1395 if (com == TIOCSCTTY) { 1396 devfs_debug(DEVFS_DEBUG_DEBUG, 1397 "devfs_specf_ioctl: got TIOCSCTTY on %s\n", 1398 dev->si_name); 1399 } 1400 if (error == 0 && com == TIOCSCTTY) { 1401 struct proc *p = curthread->td_proc; 1402 struct session *sess; 1403 1404 devfs_debug(DEVFS_DEBUG_DEBUG, 1405 "devfs_specf_ioctl: dealing with TIOCSCTTY on %s\n", 1406 dev->si_name); 1407 if (p == NULL) { 1408 error = ENOTTY; 1409 goto out; 1410 } 1411 sess = p->p_session; 1412 1413 /* 1414 * Do nothing if reassigning same control tty 1415 */ 1416 if (sess->s_ttyvp == vp) { 1417 error = 0; 1418 goto out; 1419 } 1420 1421 /* 1422 * Get rid of reference to old control tty 1423 */ 1424 ovp = sess->s_ttyvp; 1425 vref(vp); 1426 sess->s_ttyvp = vp; 1427 if (ovp) 1428 vrele(ovp); 1429 } 1430 1431 out: 1432 rel_mplock(); 1433 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_specf_ioctl() finished! \n"); 1434 return (error); 1435 } 1436 1437 1438 static int 1439 devfs_spec_fsync(struct vop_fsync_args *ap) 1440 { 1441 struct vnode *vp = ap->a_vp; 1442 int error; 1443 1444 if (!vn_isdisk(vp, NULL)) 1445 return (0); 1446 1447 /* 1448 * Flush all dirty buffers associated with a block device. 1449 */ 1450 error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL); 1451 return (error); 1452 } 1453 1454 static int 1455 devfs_spec_read(struct vop_read_args *ap) 1456 { 1457 struct devfs_node *node; 1458 struct vnode *vp; 1459 struct uio *uio; 1460 cdev_t dev; 1461 int error; 1462 1463 vp = ap->a_vp; 1464 dev = vp->v_rdev; 1465 uio = ap->a_uio; 1466 node = DEVFS_NODE(vp); 1467 1468 if (dev == NULL) /* device was revoked */ 1469 return (EBADF); 1470 if (uio->uio_resid == 0) 1471 return (0); 1472 1473 vn_unlock(vp); 1474 error = dev_dread(dev, uio, ap->a_ioflag); 1475 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1476 1477 if (node) 1478 nanotime(&node->atime); 1479 1480 return (error); 1481 } 1482 1483 /* 1484 * Vnode op for write 1485 * 1486 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, 1487 * struct ucred *a_cred) 1488 */ 1489 static int 1490 devfs_spec_write(struct vop_write_args *ap) 1491 { 1492 struct devfs_node *node; 1493 struct vnode *vp; 1494 struct uio *uio; 1495 cdev_t dev; 1496 int error; 1497 1498 vp = ap->a_vp; 1499 dev = vp->v_rdev; 1500 uio = ap->a_uio; 1501 node = DEVFS_NODE(vp); 1502 1503 KKASSERT(uio->uio_segflg != UIO_NOCOPY); 1504 1505 if (dev == NULL) /* device was revoked */ 1506 return (EBADF); 1507 1508 vn_unlock(vp); 1509 error = dev_dwrite(dev, uio, ap->a_ioflag); 1510 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1511 1512 if (node) { 1513 nanotime(&node->atime); 1514 nanotime(&node->mtime); 1515 } 1516 1517 return (error); 1518 } 1519 1520 /* 1521 * Device ioctl operation. 1522 * 1523 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data, 1524 * int a_fflag, struct ucred *a_cred, struct sysmsg *msg) 1525 */ 1526 static int 1527 devfs_spec_ioctl(struct vop_ioctl_args *ap) 1528 { 1529 struct vnode *vp = ap->a_vp; 1530 struct devfs_node *node; 1531 cdev_t dev; 1532 1533 if ((dev = vp->v_rdev) == NULL) 1534 return (EBADF); /* device was revoked */ 1535 node = DEVFS_NODE(vp); 1536 1537 #if 0 1538 if (node) { 1539 nanotime(&node->atime); 1540 nanotime(&node->mtime); 1541 } 1542 #endif 1543 1544 return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag, 1545 ap->a_cred, ap->a_sysmsg)); 1546 } 1547 1548 /* 1549 * spec_poll(struct vnode *a_vp, int a_events, struct ucred *a_cred) 1550 */ 1551 /* ARGSUSED */ 1552 static int 1553 devfs_spec_poll(struct vop_poll_args *ap) 1554 { 1555 struct vnode *vp = ap->a_vp; 1556 struct devfs_node *node; 1557 cdev_t dev; 1558 1559 if ((dev = vp->v_rdev) == NULL) 1560 return (EBADF); /* device was revoked */ 1561 node = DEVFS_NODE(vp); 1562 1563 #if 0 1564 if (node) 1565 nanotime(&node->atime); 1566 #endif 1567 1568 return (dev_dpoll(dev, ap->a_events)); 1569 } 1570 1571 /* 1572 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn) 1573 */ 1574 /* ARGSUSED */ 1575 static int 1576 devfs_spec_kqfilter(struct vop_kqfilter_args *ap) 1577 { 1578 struct vnode *vp = ap->a_vp; 1579 struct devfs_node *node; 1580 cdev_t dev; 1581 1582 if ((dev = vp->v_rdev) == NULL) 1583 return (EBADF); /* device was revoked */ 1584 node = DEVFS_NODE(vp); 1585 1586 #if 0 1587 if (node) 1588 nanotime(&node->atime); 1589 #endif 1590 1591 return (dev_dkqfilter(dev, ap->a_kn)); 1592 } 1593 1594 /* 1595 * Convert a vnode strategy call into a device strategy call. Vnode strategy 1596 * calls are not limited to device DMA limits so we have to deal with the 1597 * case. 1598 * 1599 * spec_strategy(struct vnode *a_vp, struct bio *a_bio) 1600 */ 1601 static int 1602 devfs_spec_strategy(struct vop_strategy_args *ap) 1603 { 1604 struct bio *bio = ap->a_bio; 1605 struct buf *bp = bio->bio_buf; 1606 struct buf *nbp; 1607 struct vnode *vp; 1608 struct mount *mp; 1609 int chunksize; 1610 int maxiosize; 1611 1612 if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL) 1613 buf_start(bp); 1614 1615 /* 1616 * Collect statistics on synchronous and asynchronous read 1617 * and write counts for disks that have associated filesystems. 1618 */ 1619 vp = ap->a_vp; 1620 KKASSERT(vp->v_rdev != NULL); /* XXX */ 1621 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) { 1622 if (bp->b_cmd == BUF_CMD_READ) { 1623 if (bp->b_flags & BIO_SYNC) 1624 mp->mnt_stat.f_syncreads++; 1625 else 1626 mp->mnt_stat.f_asyncreads++; 1627 } else { 1628 if (bp->b_flags & BIO_SYNC) 1629 mp->mnt_stat.f_syncwrites++; 1630 else 1631 mp->mnt_stat.f_asyncwrites++; 1632 } 1633 } 1634 1635 /* 1636 * Device iosize limitations only apply to read and write. Shortcut 1637 * the I/O if it fits. 1638 */ 1639 if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) { 1640 devfs_debug(DEVFS_DEBUG_DEBUG, 1641 "%s: si_iosize_max not set!\n", 1642 dev_dname(vp->v_rdev)); 1643 maxiosize = MAXPHYS; 1644 } 1645 #if SPEC_CHAIN_DEBUG & 2 1646 maxiosize = 4096; 1647 #endif 1648 if (bp->b_bcount <= maxiosize || 1649 (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) { 1650 dev_dstrategy_chain(vp->v_rdev, bio); 1651 return (0); 1652 } 1653 1654 /* 1655 * Clone the buffer and set up an I/O chain to chunk up the I/O. 1656 */ 1657 nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO); 1658 initbufbio(nbp); 1659 buf_dep_init(nbp); 1660 BUF_LOCKINIT(nbp); 1661 BUF_LOCK(nbp, LK_EXCLUSIVE); 1662 BUF_KERNPROC(nbp); 1663 nbp->b_vp = vp; 1664 nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP); 1665 nbp->b_data = bp->b_data; 1666 nbp->b_bio1.bio_done = devfs_spec_strategy_done; 1667 nbp->b_bio1.bio_offset = bio->bio_offset; 1668 nbp->b_bio1.bio_caller_info1.ptr = bio; 1669 1670 /* 1671 * Start the first transfer 1672 */ 1673 if (vn_isdisk(vp, NULL)) 1674 chunksize = vp->v_rdev->si_bsize_phys; 1675 else 1676 chunksize = DEV_BSIZE; 1677 chunksize = maxiosize / chunksize * chunksize; 1678 #if SPEC_CHAIN_DEBUG & 1 1679 devfs_debug(DEVFS_DEBUG_DEBUG, 1680 "spec_strategy chained I/O chunksize=%d\n", 1681 chunksize); 1682 #endif 1683 nbp->b_cmd = bp->b_cmd; 1684 nbp->b_bcount = chunksize; 1685 nbp->b_bufsize = chunksize; /* used to detect a short I/O */ 1686 nbp->b_bio1.bio_caller_info2.index = chunksize; 1687 1688 #if SPEC_CHAIN_DEBUG & 1 1689 devfs_debug(DEVFS_DEBUG_DEBUG, 1690 "spec_strategy: chain %p offset %d/%d bcount %d\n", 1691 bp, 0, bp->b_bcount, nbp->b_bcount); 1692 #endif 1693 1694 dev_dstrategy(vp->v_rdev, &nbp->b_bio1); 1695 1696 if (DEVFS_NODE(vp)) { 1697 nanotime(&DEVFS_NODE(vp)->atime); 1698 nanotime(&DEVFS_NODE(vp)->mtime); 1699 } 1700 1701 return (0); 1702 } 1703 1704 /* 1705 * Chunked up transfer completion routine - chain transfers until done 1706 */ 1707 static 1708 void 1709 devfs_spec_strategy_done(struct bio *nbio) 1710 { 1711 struct buf *nbp = nbio->bio_buf; 1712 struct bio *bio = nbio->bio_caller_info1.ptr; /* original bio */ 1713 struct buf *bp = bio->bio_buf; /* original bp */ 1714 int chunksize = nbio->bio_caller_info2.index; /* chunking */ 1715 int boffset = nbp->b_data - bp->b_data; 1716 1717 if (nbp->b_flags & B_ERROR) { 1718 /* 1719 * An error terminates the chain, propogate the error back 1720 * to the original bp 1721 */ 1722 bp->b_flags |= B_ERROR; 1723 bp->b_error = nbp->b_error; 1724 bp->b_resid = bp->b_bcount - boffset + 1725 (nbp->b_bcount - nbp->b_resid); 1726 #if SPEC_CHAIN_DEBUG & 1 1727 devfs_debug(DEVFS_DEBUG_DEBUG, 1728 "spec_strategy: chain %p error %d bcount %d/%d\n", 1729 bp, bp->b_error, bp->b_bcount, 1730 bp->b_bcount - bp->b_resid); 1731 #endif 1732 kfree(nbp, M_DEVBUF); 1733 biodone(bio); 1734 } else if (nbp->b_resid) { 1735 /* 1736 * A short read or write terminates the chain 1737 */ 1738 bp->b_error = nbp->b_error; 1739 bp->b_resid = bp->b_bcount - boffset + 1740 (nbp->b_bcount - nbp->b_resid); 1741 #if SPEC_CHAIN_DEBUG & 1 1742 devfs_debug(DEVFS_DEBUG_DEBUG, 1743 "spec_strategy: chain %p short read(1) " 1744 "bcount %d/%d\n", 1745 bp, bp->b_bcount - bp->b_resid, bp->b_bcount); 1746 #endif 1747 kfree(nbp, M_DEVBUF); 1748 biodone(bio); 1749 } else if (nbp->b_bcount != nbp->b_bufsize) { 1750 /* 1751 * A short read or write can also occur by truncating b_bcount 1752 */ 1753 #if SPEC_CHAIN_DEBUG & 1 1754 devfs_debug(DEVFS_DEBUG_DEBUG, 1755 "spec_strategy: chain %p short read(2) " 1756 "bcount %d/%d\n", 1757 bp, nbp->b_bcount + boffset, bp->b_bcount); 1758 #endif 1759 bp->b_error = 0; 1760 bp->b_bcount = nbp->b_bcount + boffset; 1761 bp->b_resid = nbp->b_resid; 1762 kfree(nbp, M_DEVBUF); 1763 biodone(bio); 1764 } else if (nbp->b_bcount + boffset == bp->b_bcount) { 1765 /* 1766 * No more data terminates the chain 1767 */ 1768 #if SPEC_CHAIN_DEBUG & 1 1769 devfs_debug(DEVFS_DEBUG_DEBUG, 1770 "spec_strategy: chain %p finished bcount %d\n", 1771 bp, bp->b_bcount); 1772 #endif 1773 bp->b_error = 0; 1774 bp->b_resid = 0; 1775 kfree(nbp, M_DEVBUF); 1776 biodone(bio); 1777 } else { 1778 /* 1779 * Continue the chain 1780 */ 1781 boffset += nbp->b_bcount; 1782 nbp->b_data = bp->b_data + boffset; 1783 nbp->b_bcount = bp->b_bcount - boffset; 1784 if (nbp->b_bcount > chunksize) 1785 nbp->b_bcount = chunksize; 1786 nbp->b_bio1.bio_done = devfs_spec_strategy_done; 1787 nbp->b_bio1.bio_offset = bio->bio_offset + boffset; 1788 1789 #if SPEC_CHAIN_DEBUG & 1 1790 devfs_debug(DEVFS_DEBUG_DEBUG, 1791 "spec_strategy: chain %p offset %d/%d bcount %d\n", 1792 bp, boffset, bp->b_bcount, nbp->b_bcount); 1793 #endif 1794 1795 dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1); 1796 } 1797 } 1798 1799 /* 1800 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length) 1801 */ 1802 static int 1803 devfs_spec_freeblks(struct vop_freeblks_args *ap) 1804 { 1805 struct buf *bp; 1806 1807 /* 1808 * XXX: This assumes that strategy does the deed right away. 1809 * XXX: this may not be TRTTD. 1810 */ 1811 KKASSERT(ap->a_vp->v_rdev != NULL); 1812 if ((dev_dflags(ap->a_vp->v_rdev) & D_CANFREE) == 0) 1813 return (0); 1814 bp = geteblk(ap->a_length); 1815 bp->b_cmd = BUF_CMD_FREEBLKS; 1816 bp->b_bio1.bio_offset = ap->a_offset; 1817 bp->b_bcount = ap->a_length; 1818 dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1); 1819 return (0); 1820 } 1821 1822 /* 1823 * Implement degenerate case where the block requested is the block 1824 * returned, and assume that the entire device is contiguous in regards 1825 * to the contiguous block range (runp and runb). 1826 * 1827 * spec_bmap(struct vnode *a_vp, off_t a_loffset, 1828 * off_t *a_doffsetp, int *a_runp, int *a_runb) 1829 */ 1830 static int 1831 devfs_spec_bmap(struct vop_bmap_args *ap) 1832 { 1833 if (ap->a_doffsetp != NULL) 1834 *ap->a_doffsetp = ap->a_loffset; 1835 if (ap->a_runp != NULL) 1836 *ap->a_runp = MAXBSIZE; 1837 if (ap->a_runb != NULL) { 1838 if (ap->a_loffset < MAXBSIZE) 1839 *ap->a_runb = (int)ap->a_loffset; 1840 else 1841 *ap->a_runb = MAXBSIZE; 1842 } 1843 return (0); 1844 } 1845 1846 1847 /* 1848 * Special device advisory byte-level locks. 1849 * 1850 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, 1851 * struct flock *a_fl, int a_flags) 1852 */ 1853 /* ARGSUSED */ 1854 static int 1855 devfs_spec_advlock(struct vop_advlock_args *ap) 1856 { 1857 return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP); 1858 } 1859 1860 static void 1861 devfs_spec_getpages_iodone(struct bio *bio) 1862 { 1863 bio->bio_buf->b_cmd = BUF_CMD_DONE; 1864 wakeup(bio->bio_buf); 1865 } 1866 1867 /* 1868 * spec_getpages() - get pages associated with device vnode. 1869 * 1870 * Note that spec_read and spec_write do not use the buffer cache, so we 1871 * must fully implement getpages here. 1872 */ 1873 static int 1874 devfs_spec_getpages(struct vop_getpages_args *ap) 1875 { 1876 vm_offset_t kva; 1877 int error; 1878 int i, pcount, size; 1879 struct buf *bp; 1880 vm_page_t m; 1881 vm_ooffset_t offset; 1882 int toff, nextoff, nread; 1883 struct vnode *vp = ap->a_vp; 1884 int blksiz; 1885 int gotreqpage; 1886 1887 error = 0; 1888 pcount = round_page(ap->a_count) / PAGE_SIZE; 1889 1890 /* 1891 * Calculate the offset of the transfer and do sanity check. 1892 */ 1893 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset; 1894 1895 /* 1896 * Round up physical size for real devices. We cannot round using 1897 * v_mount's block size data because v_mount has nothing to do with 1898 * the device. i.e. it's usually '/dev'. We need the physical block 1899 * size for the device itself. 1900 * 1901 * We can't use v_rdev->si_mountpoint because it only exists when the 1902 * block device is mounted. However, we can use v_rdev. 1903 */ 1904 if (vn_isdisk(vp, NULL)) 1905 blksiz = vp->v_rdev->si_bsize_phys; 1906 else 1907 blksiz = DEV_BSIZE; 1908 1909 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1); 1910 1911 bp = getpbuf(NULL); 1912 kva = (vm_offset_t)bp->b_data; 1913 1914 /* 1915 * Map the pages to be read into the kva. 1916 */ 1917 pmap_qenter(kva, ap->a_m, pcount); 1918 1919 /* Build a minimal buffer header. */ 1920 bp->b_cmd = BUF_CMD_READ; 1921 bp->b_bcount = size; 1922 bp->b_resid = 0; 1923 bp->b_runningbufspace = size; 1924 if (size) { 1925 runningbufspace += bp->b_runningbufspace; 1926 ++runningbufcount; 1927 } 1928 1929 bp->b_bio1.bio_offset = offset; 1930 bp->b_bio1.bio_done = devfs_spec_getpages_iodone; 1931 1932 mycpu->gd_cnt.v_vnodein++; 1933 mycpu->gd_cnt.v_vnodepgsin += pcount; 1934 1935 /* Do the input. */ 1936 vn_strategy(ap->a_vp, &bp->b_bio1); 1937 1938 crit_enter(); 1939 1940 /* We definitely need to be at splbio here. */ 1941 while (bp->b_cmd != BUF_CMD_DONE) 1942 tsleep(bp, 0, "spread", 0); 1943 1944 crit_exit(); 1945 1946 if (bp->b_flags & B_ERROR) { 1947 if (bp->b_error) 1948 error = bp->b_error; 1949 else 1950 error = EIO; 1951 } 1952 1953 /* 1954 * If EOF is encountered we must zero-extend the result in order 1955 * to ensure that the page does not contain garabge. When no 1956 * error occurs, an early EOF is indicated if b_bcount got truncated. 1957 * b_resid is relative to b_bcount and should be 0, but some devices 1958 * might indicate an EOF with b_resid instead of truncating b_bcount. 1959 */ 1960 nread = bp->b_bcount - bp->b_resid; 1961 if (nread < ap->a_count) 1962 bzero((caddr_t)kva + nread, ap->a_count - nread); 1963 pmap_qremove(kva, pcount); 1964 1965 gotreqpage = 0; 1966 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) { 1967 nextoff = toff + PAGE_SIZE; 1968 m = ap->a_m[i]; 1969 1970 m->flags &= ~PG_ZERO; 1971 1972 /* 1973 * NOTE: vm_page_undirty/clear_dirty etc do not clear the 1974 * pmap modified bit. pmap modified bit should have 1975 * already been cleared. 1976 */ 1977 if (nextoff <= nread) { 1978 m->valid = VM_PAGE_BITS_ALL; 1979 vm_page_undirty(m); 1980 } else if (toff < nread) { 1981 /* 1982 * Since this is a VM request, we have to supply the 1983 * unaligned offset to allow vm_page_set_valid() 1984 * to zero sub-DEV_BSIZE'd portions of the page. 1985 */ 1986 vm_page_set_valid(m, 0, nread - toff); 1987 vm_page_clear_dirty_end_nonincl(m, 0, nread - toff); 1988 } else { 1989 m->valid = 0; 1990 vm_page_undirty(m); 1991 } 1992 1993 if (i != ap->a_reqpage) { 1994 /* 1995 * Just in case someone was asking for this page we 1996 * now tell them that it is ok to use. 1997 */ 1998 if (!error || (m->valid == VM_PAGE_BITS_ALL)) { 1999 if (m->valid) { 2000 if (m->flags & PG_WANTED) { 2001 vm_page_activate(m); 2002 } else { 2003 vm_page_deactivate(m); 2004 } 2005 vm_page_wakeup(m); 2006 } else { 2007 vm_page_free(m); 2008 } 2009 } else { 2010 vm_page_free(m); 2011 } 2012 } else if (m->valid) { 2013 gotreqpage = 1; 2014 /* 2015 * Since this is a VM request, we need to make the 2016 * entire page presentable by zeroing invalid sections. 2017 */ 2018 if (m->valid != VM_PAGE_BITS_ALL) 2019 vm_page_zero_invalid(m, FALSE); 2020 } 2021 } 2022 if (!gotreqpage) { 2023 m = ap->a_m[ap->a_reqpage]; 2024 devfs_debug(DEVFS_DEBUG_WARNING, 2025 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n", 2026 devtoname(vp->v_rdev), error, bp, bp->b_vp); 2027 devfs_debug(DEVFS_DEBUG_WARNING, 2028 " size: %d, resid: %d, a_count: %d, valid: 0x%x\n", 2029 size, bp->b_resid, ap->a_count, m->valid); 2030 devfs_debug(DEVFS_DEBUG_WARNING, 2031 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n", 2032 nread, ap->a_reqpage, (u_long)m->pindex, pcount); 2033 /* 2034 * Free the buffer header back to the swap buffer pool. 2035 */ 2036 relpbuf(bp, NULL); 2037 return VM_PAGER_ERROR; 2038 } 2039 /* 2040 * Free the buffer header back to the swap buffer pool. 2041 */ 2042 relpbuf(bp, NULL); 2043 if (DEVFS_NODE(ap->a_vp)) 2044 nanotime(&DEVFS_NODE(ap->a_vp)->mtime); 2045 return VM_PAGER_OK; 2046 } 2047 2048 static __inline 2049 int 2050 sequential_heuristic(struct uio *uio, struct file *fp) 2051 { 2052 /* 2053 * Sequential heuristic - detect sequential operation 2054 */ 2055 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || 2056 uio->uio_offset == fp->f_nextoff) { 2057 /* 2058 * XXX we assume that the filesystem block size is 2059 * the default. Not true, but still gives us a pretty 2060 * good indicator of how sequential the read operations 2061 * are. 2062 */ 2063 int tmpseq = fp->f_seqcount; 2064 2065 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE; 2066 if (tmpseq > IO_SEQMAX) 2067 tmpseq = IO_SEQMAX; 2068 fp->f_seqcount = tmpseq; 2069 return(fp->f_seqcount << IO_SEQSHIFT); 2070 } 2071 2072 /* 2073 * Not sequential, quick draw-down of seqcount 2074 */ 2075 if (fp->f_seqcount > 1) 2076 fp->f_seqcount = 1; 2077 else 2078 fp->f_seqcount = 0; 2079 return(0); 2080 } 2081