| 1 | /* |
| 2 | * (MPSAFE) |
| 3 | * |
| 4 | * Copyright (c) 2009 The DragonFly Project. All rights reserved. |
| 5 | * |
| 6 | * This code is derived from software contributed to The DragonFly Project |
| 7 | * by Alex Hornung <ahornung@gmail.com> |
| 8 | * |
| 9 | * Redistribution and use in source and binary forms, with or without |
| 10 | * modification, are permitted provided that the following conditions |
| 11 | * are met: |
| 12 | * |
| 13 | * 1. Redistributions of source code must retain the above copyright |
| 14 | * notice, this list of conditions and the following disclaimer. |
| 15 | * 2. Redistributions in binary form must reproduce the above copyright |
| 16 | * notice, this list of conditions and the following disclaimer in |
| 17 | * the documentation and/or other materials provided with the |
| 18 | * distribution. |
| 19 | * 3. Neither the name of The DragonFly Project nor the names of its |
| 20 | * contributors may be used to endorse or promote products derived |
| 21 | * from this software without specific, prior written permission. |
| 22 | * |
| 23 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 24 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 25 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
| 26 | * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
| 27 | * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
| 28 | * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, |
| 29 | * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
| 30 | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED |
| 31 | * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, |
| 32 | * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT |
| 33 | * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 34 | * SUCH DAMAGE. |
| 35 | */ |
| 36 | #include <sys/param.h> |
| 37 | #include <sys/systm.h> |
| 38 | #include <sys/time.h> |
| 39 | #include <sys/kernel.h> |
| 40 | #include <sys/lock.h> |
| 41 | #include <sys/fcntl.h> |
| 42 | #include <sys/proc.h> |
| 43 | #include <sys/priv.h> |
| 44 | #include <sys/signalvar.h> |
| 45 | #include <sys/vnode.h> |
| 46 | #include <sys/uio.h> |
| 47 | #include <sys/mount.h> |
| 48 | #include <sys/file.h> |
| 49 | #include <sys/fcntl.h> |
| 50 | #include <sys/namei.h> |
| 51 | #include <sys/dirent.h> |
| 52 | #include <sys/malloc.h> |
| 53 | #include <sys/stat.h> |
| 54 | #include <sys/reg.h> |
| 55 | #include <vm/vm_pager.h> |
| 56 | #include <vm/vm_zone.h> |
| 57 | #include <vm/vm_object.h> |
| 58 | #include <sys/filio.h> |
| 59 | #include <sys/ttycom.h> |
| 60 | #include <sys/tty.h> |
| 61 | #include <sys/diskslice.h> |
| 62 | #include <sys/sysctl.h> |
| 63 | #include <sys/devfs.h> |
| 64 | #include <sys/pioctl.h> |
| 65 | #include <vfs/fifofs/fifo.h> |
| 66 | |
| 67 | #include <machine/limits.h> |
| 68 | |
| 69 | #include <sys/buf2.h> |
| 70 | #include <sys/sysref2.h> |
| 71 | #include <sys/mplock2.h> |
| 72 | #include <vm/vm_page2.h> |
| 73 | |
| 74 | MALLOC_DECLARE(M_DEVFS); |
| 75 | #define DEVFS_BADOP (void *)devfs_vop_badop |
| 76 | |
| 77 | static int devfs_vop_badop(struct vop_generic_args *); |
| 78 | static int devfs_vop_access(struct vop_access_args *); |
| 79 | static int devfs_vop_inactive(struct vop_inactive_args *); |
| 80 | static int devfs_vop_reclaim(struct vop_reclaim_args *); |
| 81 | static int devfs_vop_readdir(struct vop_readdir_args *); |
| 82 | static int devfs_vop_getattr(struct vop_getattr_args *); |
| 83 | static int devfs_vop_setattr(struct vop_setattr_args *); |
| 84 | static int devfs_vop_readlink(struct vop_readlink_args *); |
| 85 | static int devfs_vop_print(struct vop_print_args *); |
| 86 | |
| 87 | static int devfs_vop_nresolve(struct vop_nresolve_args *); |
| 88 | static int devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *); |
| 89 | static int devfs_vop_nmkdir(struct vop_nmkdir_args *); |
| 90 | static int devfs_vop_nsymlink(struct vop_nsymlink_args *); |
| 91 | static int devfs_vop_nrmdir(struct vop_nrmdir_args *); |
| 92 | static int devfs_vop_nremove(struct vop_nremove_args *); |
| 93 | |
| 94 | static int devfs_spec_open(struct vop_open_args *); |
| 95 | static int devfs_spec_close(struct vop_close_args *); |
| 96 | static int devfs_spec_fsync(struct vop_fsync_args *); |
| 97 | |
| 98 | static int devfs_spec_read(struct vop_read_args *); |
| 99 | static int devfs_spec_write(struct vop_write_args *); |
| 100 | static int devfs_spec_ioctl(struct vop_ioctl_args *); |
| 101 | static int devfs_spec_kqfilter(struct vop_kqfilter_args *); |
| 102 | static int devfs_spec_strategy(struct vop_strategy_args *); |
| 103 | static void devfs_spec_strategy_done(struct bio *); |
| 104 | static int devfs_spec_freeblks(struct vop_freeblks_args *); |
| 105 | static int devfs_spec_bmap(struct vop_bmap_args *); |
| 106 | static int devfs_spec_advlock(struct vop_advlock_args *); |
| 107 | static void devfs_spec_getpages_iodone(struct bio *); |
| 108 | static int devfs_spec_getpages(struct vop_getpages_args *); |
| 109 | |
| 110 | static int devfs_fo_close(struct file *); |
| 111 | static int devfs_fo_read(struct file *, struct uio *, struct ucred *, int); |
| 112 | static int devfs_fo_write(struct file *, struct uio *, struct ucred *, int); |
| 113 | static int devfs_fo_stat(struct file *, struct stat *, struct ucred *); |
| 114 | static int devfs_fo_kqfilter(struct file *, struct knote *); |
| 115 | static int devfs_fo_ioctl(struct file *, u_long, caddr_t, |
| 116 | struct ucred *, struct sysmsg *); |
| 117 | static __inline int sequential_heuristic(struct uio *, struct file *); |
| 118 | |
| 119 | extern struct lock devfs_lock; |
| 120 | |
| 121 | /* |
| 122 | * devfs vnode operations for regular files. All vnode ops are MPSAFE. |
| 123 | */ |
| 124 | struct vop_ops devfs_vnode_norm_vops = { |
| 125 | .vop_default = vop_defaultop, |
| 126 | .vop_access = devfs_vop_access, |
| 127 | .vop_advlock = DEVFS_BADOP, |
| 128 | .vop_bmap = DEVFS_BADOP, |
| 129 | .vop_close = vop_stdclose, |
| 130 | .vop_getattr = devfs_vop_getattr, |
| 131 | .vop_inactive = devfs_vop_inactive, |
| 132 | .vop_ncreate = DEVFS_BADOP, |
| 133 | .vop_nresolve = devfs_vop_nresolve, |
| 134 | .vop_nlookupdotdot = devfs_vop_nlookupdotdot, |
| 135 | .vop_nlink = DEVFS_BADOP, |
| 136 | .vop_nmkdir = devfs_vop_nmkdir, |
| 137 | .vop_nmknod = DEVFS_BADOP, |
| 138 | .vop_nremove = devfs_vop_nremove, |
| 139 | .vop_nrename = DEVFS_BADOP, |
| 140 | .vop_nrmdir = devfs_vop_nrmdir, |
| 141 | .vop_nsymlink = devfs_vop_nsymlink, |
| 142 | .vop_open = vop_stdopen, |
| 143 | .vop_pathconf = vop_stdpathconf, |
| 144 | .vop_print = devfs_vop_print, |
| 145 | .vop_read = DEVFS_BADOP, |
| 146 | .vop_readdir = devfs_vop_readdir, |
| 147 | .vop_readlink = devfs_vop_readlink, |
| 148 | .vop_reclaim = devfs_vop_reclaim, |
| 149 | .vop_setattr = devfs_vop_setattr, |
| 150 | .vop_write = DEVFS_BADOP, |
| 151 | .vop_ioctl = DEVFS_BADOP |
| 152 | }; |
| 153 | |
| 154 | /* |
| 155 | * devfs vnode operations for character devices. All vnode ops are MPSAFE. |
| 156 | */ |
| 157 | struct vop_ops devfs_vnode_dev_vops = { |
| 158 | .vop_default = vop_defaultop, |
| 159 | .vop_access = devfs_vop_access, |
| 160 | .vop_advlock = devfs_spec_advlock, |
| 161 | .vop_bmap = devfs_spec_bmap, |
| 162 | .vop_close = devfs_spec_close, |
| 163 | .vop_freeblks = devfs_spec_freeblks, |
| 164 | .vop_fsync = devfs_spec_fsync, |
| 165 | .vop_getattr = devfs_vop_getattr, |
| 166 | .vop_getpages = devfs_spec_getpages, |
| 167 | .vop_inactive = devfs_vop_inactive, |
| 168 | .vop_open = devfs_spec_open, |
| 169 | .vop_pathconf = vop_stdpathconf, |
| 170 | .vop_print = devfs_vop_print, |
| 171 | .vop_kqfilter = devfs_spec_kqfilter, |
| 172 | .vop_read = devfs_spec_read, |
| 173 | .vop_readdir = DEVFS_BADOP, |
| 174 | .vop_readlink = DEVFS_BADOP, |
| 175 | .vop_reclaim = devfs_vop_reclaim, |
| 176 | .vop_setattr = devfs_vop_setattr, |
| 177 | .vop_strategy = devfs_spec_strategy, |
| 178 | .vop_write = devfs_spec_write, |
| 179 | .vop_ioctl = devfs_spec_ioctl |
| 180 | }; |
| 181 | |
| 182 | /* |
| 183 | * devfs file pointer operations. All fileops are MPSAFE. |
| 184 | */ |
| 185 | struct vop_ops *devfs_vnode_dev_vops_p = &devfs_vnode_dev_vops; |
| 186 | |
| 187 | struct fileops devfs_dev_fileops = { |
| 188 | .fo_read = devfs_fo_read, |
| 189 | .fo_write = devfs_fo_write, |
| 190 | .fo_ioctl = devfs_fo_ioctl, |
| 191 | .fo_kqfilter = devfs_fo_kqfilter, |
| 192 | .fo_stat = devfs_fo_stat, |
| 193 | .fo_close = devfs_fo_close, |
| 194 | .fo_shutdown = nofo_shutdown |
| 195 | }; |
| 196 | |
| 197 | /* |
| 198 | * These two functions are possibly temporary hacks for devices (aka |
| 199 | * the pty code) which want to control the node attributes themselves. |
| 200 | * |
| 201 | * XXX we may ultimately desire to simply remove the uid/gid/mode |
| 202 | * from the node entirely. |
| 203 | * |
| 204 | * MPSAFE - sorta. Theoretically the overwrite can compete since they |
| 205 | * are loading from the same fields. |
| 206 | */ |
| 207 | static __inline void |
| 208 | node_sync_dev_get(struct devfs_node *node) |
| 209 | { |
| 210 | cdev_t dev; |
| 211 | |
| 212 | if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) { |
| 213 | node->uid = dev->si_uid; |
| 214 | node->gid = dev->si_gid; |
| 215 | node->mode = dev->si_perms; |
| 216 | } |
| 217 | } |
| 218 | |
| 219 | static __inline void |
| 220 | node_sync_dev_set(struct devfs_node *node) |
| 221 | { |
| 222 | cdev_t dev; |
| 223 | |
| 224 | if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) { |
| 225 | dev->si_uid = node->uid; |
| 226 | dev->si_gid = node->gid; |
| 227 | dev->si_perms = node->mode; |
| 228 | } |
| 229 | } |
| 230 | |
| 231 | /* |
| 232 | * generic entry point for unsupported operations |
| 233 | */ |
| 234 | static int |
| 235 | devfs_vop_badop(struct vop_generic_args *ap) |
| 236 | { |
| 237 | return (EIO); |
| 238 | } |
| 239 | |
| 240 | |
| 241 | static int |
| 242 | devfs_vop_access(struct vop_access_args *ap) |
| 243 | { |
| 244 | struct devfs_node *node = DEVFS_NODE(ap->a_vp); |
| 245 | int error; |
| 246 | |
| 247 | if (!devfs_node_is_accessible(node)) |
| 248 | return ENOENT; |
| 249 | node_sync_dev_get(node); |
| 250 | error = vop_helper_access(ap, node->uid, node->gid, |
| 251 | node->mode, node->flags); |
| 252 | |
| 253 | return error; |
| 254 | } |
| 255 | |
| 256 | |
| 257 | static int |
| 258 | devfs_vop_inactive(struct vop_inactive_args *ap) |
| 259 | { |
| 260 | struct devfs_node *node = DEVFS_NODE(ap->a_vp); |
| 261 | |
| 262 | if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0) |
| 263 | vrecycle(ap->a_vp); |
| 264 | return 0; |
| 265 | } |
| 266 | |
| 267 | |
| 268 | static int |
| 269 | devfs_vop_reclaim(struct vop_reclaim_args *ap) |
| 270 | { |
| 271 | struct devfs_node *node; |
| 272 | struct vnode *vp; |
| 273 | int locked; |
| 274 | |
| 275 | /* |
| 276 | * Check if it is locked already. if not, we acquire the devfs lock |
| 277 | */ |
| 278 | if (!(lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE) { |
| 279 | lockmgr(&devfs_lock, LK_EXCLUSIVE); |
| 280 | locked = 1; |
| 281 | } else { |
| 282 | locked = 0; |
| 283 | } |
| 284 | |
| 285 | /* |
| 286 | * Get rid of the devfs_node if it is no longer linked into the |
| 287 | * topology. |
| 288 | */ |
| 289 | vp = ap->a_vp; |
| 290 | if ((node = DEVFS_NODE(vp)) != NULL) { |
| 291 | node->v_node = NULL; |
| 292 | if ((node->flags & DEVFS_NODE_LINKED) == 0) |
| 293 | devfs_freep(node); |
| 294 | } |
| 295 | |
| 296 | if (locked) |
| 297 | lockmgr(&devfs_lock, LK_RELEASE); |
| 298 | |
| 299 | /* |
| 300 | * v_rdev needs to be properly released using v_release_rdev |
| 301 | * Make sure v_data is NULL as well. |
| 302 | */ |
| 303 | vp->v_data = NULL; |
| 304 | v_release_rdev(vp); |
| 305 | return 0; |
| 306 | } |
| 307 | |
| 308 | |
| 309 | static int |
| 310 | devfs_vop_readdir(struct vop_readdir_args *ap) |
| 311 | { |
| 312 | struct devfs_node *dnode = DEVFS_NODE(ap->a_vp); |
| 313 | struct devfs_node *node; |
| 314 | int cookie_index; |
| 315 | int ncookies; |
| 316 | int error2; |
| 317 | int error; |
| 318 | int r; |
| 319 | off_t *cookies; |
| 320 | off_t saveoff; |
| 321 | |
| 322 | devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n"); |
| 323 | |
| 324 | if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX) |
| 325 | return (EINVAL); |
| 326 | if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0) |
| 327 | return (error); |
| 328 | |
| 329 | if (!devfs_node_is_accessible(dnode)) { |
| 330 | vn_unlock(ap->a_vp); |
| 331 | return ENOENT; |
| 332 | } |
| 333 | |
| 334 | lockmgr(&devfs_lock, LK_EXCLUSIVE); |
| 335 | |
| 336 | saveoff = ap->a_uio->uio_offset; |
| 337 | |
| 338 | if (ap->a_ncookies) { |
| 339 | ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */ |
| 340 | if (ncookies > 256) |
| 341 | ncookies = 256; |
| 342 | cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK); |
| 343 | cookie_index = 0; |
| 344 | } else { |
| 345 | ncookies = -1; |
| 346 | cookies = NULL; |
| 347 | cookie_index = 0; |
| 348 | } |
| 349 | |
| 350 | nanotime(&dnode->atime); |
| 351 | |
| 352 | if (saveoff == 0) { |
| 353 | r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino, |
| 354 | DT_DIR, 1, "."); |
| 355 | if (r) |
| 356 | goto done; |
| 357 | if (cookies) |
| 358 | cookies[cookie_index] = saveoff; |
| 359 | saveoff++; |
| 360 | cookie_index++; |
| 361 | if (cookie_index == ncookies) |
| 362 | goto done; |
| 363 | } |
| 364 | |
| 365 | if (saveoff == 1) { |
| 366 | if (dnode->parent) { |
| 367 | r = vop_write_dirent(&error, ap->a_uio, |
| 368 | dnode->parent->d_dir.d_ino, |
| 369 | DT_DIR, 2, ".."); |
| 370 | } else { |
| 371 | r = vop_write_dirent(&error, ap->a_uio, |
| 372 | dnode->d_dir.d_ino, |
| 373 | DT_DIR, 2, ".."); |
| 374 | } |
| 375 | if (r) |
| 376 | goto done; |
| 377 | if (cookies) |
| 378 | cookies[cookie_index] = saveoff; |
| 379 | saveoff++; |
| 380 | cookie_index++; |
| 381 | if (cookie_index == ncookies) |
| 382 | goto done; |
| 383 | } |
| 384 | |
| 385 | TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { |
| 386 | if ((node->flags & DEVFS_HIDDEN) || |
| 387 | (node->flags & DEVFS_INVISIBLE)) { |
| 388 | continue; |
| 389 | } |
| 390 | |
| 391 | /* |
| 392 | * If the node type is a valid devfs alias, then we make |
| 393 | * sure that the target isn't hidden. If it is, we don't |
| 394 | * show the link in the directory listing. |
| 395 | */ |
| 396 | if ((node->node_type == Nlink) && (node->link_target != NULL) && |
| 397 | (node->link_target->flags & DEVFS_HIDDEN)) |
| 398 | continue; |
| 399 | |
| 400 | if (node->cookie < saveoff) |
| 401 | continue; |
| 402 | |
| 403 | saveoff = node->cookie; |
| 404 | |
| 405 | error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino, |
| 406 | node->d_dir.d_type, |
| 407 | node->d_dir.d_namlen, |
| 408 | node->d_dir.d_name); |
| 409 | |
| 410 | if (error2) |
| 411 | break; |
| 412 | |
| 413 | saveoff++; |
| 414 | |
| 415 | if (cookies) |
| 416 | cookies[cookie_index] = node->cookie; |
| 417 | ++cookie_index; |
| 418 | if (cookie_index == ncookies) |
| 419 | break; |
| 420 | } |
| 421 | |
| 422 | done: |
| 423 | lockmgr(&devfs_lock, LK_RELEASE); |
| 424 | vn_unlock(ap->a_vp); |
| 425 | |
| 426 | ap->a_uio->uio_offset = saveoff; |
| 427 | if (error && cookie_index == 0) { |
| 428 | if (cookies) { |
| 429 | kfree(cookies, M_TEMP); |
| 430 | *ap->a_ncookies = 0; |
| 431 | *ap->a_cookies = NULL; |
| 432 | } |
| 433 | } else { |
| 434 | if (cookies) { |
| 435 | *ap->a_ncookies = cookie_index; |
| 436 | *ap->a_cookies = cookies; |
| 437 | } |
| 438 | } |
| 439 | return (error); |
| 440 | } |
| 441 | |
| 442 | |
| 443 | static int |
| 444 | devfs_vop_nresolve(struct vop_nresolve_args *ap) |
| 445 | { |
| 446 | struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); |
| 447 | struct devfs_node *node, *found = NULL; |
| 448 | struct namecache *ncp; |
| 449 | struct vnode *vp = NULL; |
| 450 | int error = 0; |
| 451 | int len; |
| 452 | int depth; |
| 453 | |
| 454 | ncp = ap->a_nch->ncp; |
| 455 | len = ncp->nc_nlen; |
| 456 | |
| 457 | if (!devfs_node_is_accessible(dnode)) |
| 458 | return ENOENT; |
| 459 | |
| 460 | lockmgr(&devfs_lock, LK_EXCLUSIVE); |
| 461 | |
| 462 | if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) { |
| 463 | error = ENOENT; |
| 464 | cache_setvp(ap->a_nch, NULL); |
| 465 | goto out; |
| 466 | } |
| 467 | |
| 468 | TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { |
| 469 | if (len == node->d_dir.d_namlen) { |
| 470 | if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) { |
| 471 | found = node; |
| 472 | break; |
| 473 | } |
| 474 | } |
| 475 | } |
| 476 | |
| 477 | if (found) { |
| 478 | depth = 0; |
| 479 | while ((found->node_type == Nlink) && (found->link_target)) { |
| 480 | if (depth >= 8) { |
| 481 | devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8"); |
| 482 | break; |
| 483 | } |
| 484 | |
| 485 | found = found->link_target; |
| 486 | ++depth; |
| 487 | } |
| 488 | |
| 489 | if (!(found->flags & DEVFS_HIDDEN)) |
| 490 | devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found); |
| 491 | } |
| 492 | |
| 493 | if (vp == NULL) { |
| 494 | error = ENOENT; |
| 495 | cache_setvp(ap->a_nch, NULL); |
| 496 | goto out; |
| 497 | |
| 498 | } |
| 499 | KKASSERT(vp); |
| 500 | vn_unlock(vp); |
| 501 | cache_setvp(ap->a_nch, vp); |
| 502 | vrele(vp); |
| 503 | out: |
| 504 | lockmgr(&devfs_lock, LK_RELEASE); |
| 505 | |
| 506 | return error; |
| 507 | } |
| 508 | |
| 509 | |
| 510 | static int |
| 511 | devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap) |
| 512 | { |
| 513 | struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); |
| 514 | |
| 515 | *ap->a_vpp = NULL; |
| 516 | if (!devfs_node_is_accessible(dnode)) |
| 517 | return ENOENT; |
| 518 | |
| 519 | lockmgr(&devfs_lock, LK_EXCLUSIVE); |
| 520 | if (dnode->parent != NULL) { |
| 521 | devfs_allocv(ap->a_vpp, dnode->parent); |
| 522 | vn_unlock(*ap->a_vpp); |
| 523 | } |
| 524 | lockmgr(&devfs_lock, LK_RELEASE); |
| 525 | |
| 526 | return ((*ap->a_vpp == NULL) ? ENOENT : 0); |
| 527 | } |
| 528 | |
| 529 | |
| 530 | static int |
| 531 | devfs_vop_getattr(struct vop_getattr_args *ap) |
| 532 | { |
| 533 | struct devfs_node *node = DEVFS_NODE(ap->a_vp); |
| 534 | struct vattr *vap = ap->a_vap; |
| 535 | struct partinfo pinfo; |
| 536 | int error = 0; |
| 537 | |
| 538 | #if 0 |
| 539 | if (!devfs_node_is_accessible(node)) |
| 540 | return ENOENT; |
| 541 | #endif |
| 542 | node_sync_dev_get(node); |
| 543 | |
| 544 | lockmgr(&devfs_lock, LK_EXCLUSIVE); |
| 545 | |
| 546 | /* start by zeroing out the attributes */ |
| 547 | VATTR_NULL(vap); |
| 548 | |
| 549 | /* next do all the common fields */ |
| 550 | vap->va_type = ap->a_vp->v_type; |
| 551 | vap->va_mode = node->mode; |
| 552 | vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ; |
| 553 | vap->va_flags = 0; |
| 554 | vap->va_blocksize = DEV_BSIZE; |
| 555 | vap->va_bytes = vap->va_size = 0; |
| 556 | |
| 557 | vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0]; |
| 558 | |
| 559 | vap->va_atime = node->atime; |
| 560 | vap->va_mtime = node->mtime; |
| 561 | vap->va_ctime = node->ctime; |
| 562 | |
| 563 | vap->va_nlink = 1; /* number of references to file */ |
| 564 | |
| 565 | vap->va_uid = node->uid; |
| 566 | vap->va_gid = node->gid; |
| 567 | |
| 568 | vap->va_rmajor = 0; |
| 569 | vap->va_rminor = 0; |
| 570 | |
| 571 | if ((node->node_type == Ndev) && node->d_dev) { |
| 572 | reference_dev(node->d_dev); |
| 573 | vap->va_rminor = node->d_dev->si_uminor; |
| 574 | release_dev(node->d_dev); |
| 575 | } |
| 576 | |
| 577 | /* For a softlink the va_size is the length of the softlink */ |
| 578 | if (node->symlink_name != 0) { |
| 579 | vap->va_bytes = vap->va_size = node->symlink_namelen; |
| 580 | } |
| 581 | |
| 582 | /* |
| 583 | * For a disk-type device, va_size is the size of the underlying |
| 584 | * device, so that lseek() works properly. |
| 585 | */ |
| 586 | if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) { |
| 587 | bzero(&pinfo, sizeof(pinfo)); |
| 588 | error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo, |
| 589 | 0, proc0.p_ucred, NULL); |
| 590 | if ((error == 0) && (pinfo.media_blksize != 0)) { |
| 591 | vap->va_size = pinfo.media_size; |
| 592 | } else { |
| 593 | vap->va_size = 0; |
| 594 | error = 0; |
| 595 | } |
| 596 | } |
| 597 | |
| 598 | lockmgr(&devfs_lock, LK_RELEASE); |
| 599 | |
| 600 | return (error); |
| 601 | } |
| 602 | |
| 603 | |
| 604 | static int |
| 605 | devfs_vop_setattr(struct vop_setattr_args *ap) |
| 606 | { |
| 607 | struct devfs_node *node = DEVFS_NODE(ap->a_vp); |
| 608 | struct vattr *vap; |
| 609 | uid_t cur_uid; |
| 610 | gid_t cur_gid; |
| 611 | mode_t cur_mode; |
| 612 | int error = 0; |
| 613 | |
| 614 | if (!devfs_node_is_accessible(node)) |
| 615 | return ENOENT; |
| 616 | node_sync_dev_get(node); |
| 617 | |
| 618 | lockmgr(&devfs_lock, LK_EXCLUSIVE); |
| 619 | |
| 620 | vap = ap->a_vap; |
| 621 | |
| 622 | if ((vap->va_uid != (uid_t)VNOVAL) || (vap->va_gid != (gid_t)VNOVAL)) { |
| 623 | cur_uid = node->uid; |
| 624 | cur_gid = node->gid; |
| 625 | cur_mode = node->mode; |
| 626 | error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid, |
| 627 | ap->a_cred, &cur_uid, &cur_gid, &cur_mode); |
| 628 | if (error) |
| 629 | goto out; |
| 630 | |
| 631 | if (node->uid != cur_uid || node->gid != cur_gid) { |
| 632 | node->uid = cur_uid; |
| 633 | node->gid = cur_gid; |
| 634 | node->mode = cur_mode; |
| 635 | } |
| 636 | } |
| 637 | |
| 638 | if (vap->va_mode != (mode_t)VNOVAL) { |
| 639 | cur_mode = node->mode; |
| 640 | error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred, |
| 641 | node->uid, node->gid, &cur_mode); |
| 642 | if (error == 0 && node->mode != cur_mode) { |
| 643 | node->mode = cur_mode; |
| 644 | } |
| 645 | } |
| 646 | |
| 647 | out: |
| 648 | node_sync_dev_set(node); |
| 649 | nanotime(&node->ctime); |
| 650 | lockmgr(&devfs_lock, LK_RELEASE); |
| 651 | |
| 652 | return error; |
| 653 | } |
| 654 | |
| 655 | |
| 656 | static int |
| 657 | devfs_vop_readlink(struct vop_readlink_args *ap) |
| 658 | { |
| 659 | struct devfs_node *node = DEVFS_NODE(ap->a_vp); |
| 660 | int ret; |
| 661 | |
| 662 | if (!devfs_node_is_accessible(node)) |
| 663 | return ENOENT; |
| 664 | |
| 665 | lockmgr(&devfs_lock, LK_EXCLUSIVE); |
| 666 | ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio); |
| 667 | lockmgr(&devfs_lock, LK_RELEASE); |
| 668 | |
| 669 | return ret; |
| 670 | } |
| 671 | |
| 672 | |
| 673 | static int |
| 674 | devfs_vop_print(struct vop_print_args *ap) |
| 675 | { |
| 676 | return (0); |
| 677 | } |
| 678 | |
| 679 | static int |
| 680 | devfs_vop_nmkdir(struct vop_nmkdir_args *ap) |
| 681 | { |
| 682 | struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); |
| 683 | struct devfs_node *node; |
| 684 | |
| 685 | if (!devfs_node_is_accessible(dnode)) |
| 686 | return ENOENT; |
| 687 | |
| 688 | if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) |
| 689 | goto out; |
| 690 | |
| 691 | lockmgr(&devfs_lock, LK_EXCLUSIVE); |
| 692 | devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Ndir, |
| 693 | ap->a_nch->ncp->nc_name, dnode, NULL); |
| 694 | |
| 695 | if (*ap->a_vpp) { |
| 696 | node = DEVFS_NODE(*ap->a_vpp); |
| 697 | node->flags |= DEVFS_USER_CREATED; |
| 698 | cache_setunresolved(ap->a_nch); |
| 699 | cache_setvp(ap->a_nch, *ap->a_vpp); |
| 700 | } |
| 701 | lockmgr(&devfs_lock, LK_RELEASE); |
| 702 | out: |
| 703 | return ((*ap->a_vpp == NULL) ? ENOTDIR : 0); |
| 704 | } |
| 705 | |
| 706 | static int |
| 707 | devfs_vop_nsymlink(struct vop_nsymlink_args *ap) |
| 708 | { |
| 709 | struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); |
| 710 | struct devfs_node *node; |
| 711 | size_t targetlen; |
| 712 | |
| 713 | if (!devfs_node_is_accessible(dnode)) |
| 714 | return ENOENT; |
| 715 | |
| 716 | ap->a_vap->va_type = VLNK; |
| 717 | |
| 718 | if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) |
| 719 | goto out; |
| 720 | |
| 721 | lockmgr(&devfs_lock, LK_EXCLUSIVE); |
| 722 | devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Nlink, |
| 723 | ap->a_nch->ncp->nc_name, dnode, NULL); |
| 724 | |
| 725 | targetlen = strlen(ap->a_target); |
| 726 | if (*ap->a_vpp) { |
| 727 | node = DEVFS_NODE(*ap->a_vpp); |
| 728 | node->flags |= DEVFS_USER_CREATED; |
| 729 | node->symlink_namelen = targetlen; |
| 730 | node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK); |
| 731 | memcpy(node->symlink_name, ap->a_target, targetlen); |
| 732 | node->symlink_name[targetlen] = '\0'; |
| 733 | cache_setunresolved(ap->a_nch); |
| 734 | cache_setvp(ap->a_nch, *ap->a_vpp); |
| 735 | } |
| 736 | lockmgr(&devfs_lock, LK_RELEASE); |
| 737 | out: |
| 738 | return ((*ap->a_vpp == NULL) ? ENOTDIR : 0); |
| 739 | } |
| 740 | |
| 741 | static int |
| 742 | devfs_vop_nrmdir(struct vop_nrmdir_args *ap) |
| 743 | { |
| 744 | struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); |
| 745 | struct devfs_node *node; |
| 746 | struct namecache *ncp; |
| 747 | int error = ENOENT; |
| 748 | |
| 749 | ncp = ap->a_nch->ncp; |
| 750 | |
| 751 | if (!devfs_node_is_accessible(dnode)) |
| 752 | return ENOENT; |
| 753 | |
| 754 | lockmgr(&devfs_lock, LK_EXCLUSIVE); |
| 755 | |
| 756 | if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) |
| 757 | goto out; |
| 758 | |
| 759 | TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { |
| 760 | if (ncp->nc_nlen != node->d_dir.d_namlen) |
| 761 | continue; |
| 762 | if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen)) |
| 763 | continue; |
| 764 | |
| 765 | /* |
| 766 | * only allow removal of user created dirs |
| 767 | */ |
| 768 | if ((node->flags & DEVFS_USER_CREATED) == 0) { |
| 769 | error = EPERM; |
| 770 | goto out; |
| 771 | } else if (node->node_type != Ndir) { |
| 772 | error = ENOTDIR; |
| 773 | goto out; |
| 774 | } else if (node->nchildren > 2) { |
| 775 | error = ENOTEMPTY; |
| 776 | goto out; |
| 777 | } else { |
| 778 | if (node->v_node) |
| 779 | cache_inval_vp(node->v_node, CINV_DESTROY); |
| 780 | devfs_unlinkp(node); |
| 781 | error = 0; |
| 782 | break; |
| 783 | } |
| 784 | } |
| 785 | |
| 786 | cache_unlink(ap->a_nch); |
| 787 | out: |
| 788 | lockmgr(&devfs_lock, LK_RELEASE); |
| 789 | return error; |
| 790 | } |
| 791 | |
| 792 | static int |
| 793 | devfs_vop_nremove(struct vop_nremove_args *ap) |
| 794 | { |
| 795 | struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp); |
| 796 | struct devfs_node *node; |
| 797 | struct namecache *ncp; |
| 798 | int error = ENOENT; |
| 799 | |
| 800 | ncp = ap->a_nch->ncp; |
| 801 | |
| 802 | if (!devfs_node_is_accessible(dnode)) |
| 803 | return ENOENT; |
| 804 | |
| 805 | lockmgr(&devfs_lock, LK_EXCLUSIVE); |
| 806 | |
| 807 | if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) |
| 808 | goto out; |
| 809 | |
| 810 | TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) { |
| 811 | if (ncp->nc_nlen != node->d_dir.d_namlen) |
| 812 | continue; |
| 813 | if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen)) |
| 814 | continue; |
| 815 | |
| 816 | /* |
| 817 | * only allow removal of user created stuff (e.g. symlinks) |
| 818 | */ |
| 819 | if ((node->flags & DEVFS_USER_CREATED) == 0) { |
| 820 | error = EPERM; |
| 821 | goto out; |
| 822 | } else if (node->node_type == Ndir) { |
| 823 | error = EISDIR; |
| 824 | goto out; |
| 825 | } else { |
| 826 | if (node->v_node) |
| 827 | cache_inval_vp(node->v_node, CINV_DESTROY); |
| 828 | devfs_unlinkp(node); |
| 829 | error = 0; |
| 830 | break; |
| 831 | } |
| 832 | } |
| 833 | |
| 834 | cache_unlink(ap->a_nch); |
| 835 | out: |
| 836 | lockmgr(&devfs_lock, LK_RELEASE); |
| 837 | return error; |
| 838 | } |
| 839 | |
| 840 | |
| 841 | static int |
| 842 | devfs_spec_open(struct vop_open_args *ap) |
| 843 | { |
| 844 | struct vnode *vp = ap->a_vp; |
| 845 | struct vnode *orig_vp = NULL; |
| 846 | struct devfs_node *node = DEVFS_NODE(vp); |
| 847 | struct devfs_node *newnode; |
| 848 | cdev_t dev, ndev = NULL; |
| 849 | int error = 0; |
| 850 | |
| 851 | if (node) { |
| 852 | if (node->d_dev == NULL) |
| 853 | return ENXIO; |
| 854 | if (!devfs_node_is_accessible(node)) |
| 855 | return ENOENT; |
| 856 | } |
| 857 | |
| 858 | if ((dev = vp->v_rdev) == NULL) |
| 859 | return ENXIO; |
| 860 | |
| 861 | if (node && ap->a_fp) { |
| 862 | devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n"); |
| 863 | lockmgr(&devfs_lock, LK_EXCLUSIVE); |
| 864 | |
| 865 | ndev = devfs_clone(dev, node->d_dir.d_name, node->d_dir.d_namlen, |
| 866 | ap->a_mode, ap->a_cred); |
| 867 | if (ndev != NULL) { |
| 868 | newnode = devfs_create_device_node( |
| 869 | DEVFS_MNTDATA(vp->v_mount)->root_node, |
| 870 | ndev, NULL, NULL); |
| 871 | /* XXX: possibly destroy device if this happens */ |
| 872 | |
| 873 | if (newnode != NULL) { |
| 874 | dev = ndev; |
| 875 | devfs_link_dev(dev); |
| 876 | |
| 877 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 878 | "parent here is: %s, node is: |%s|\n", |
| 879 | ((node->parent->node_type == Nroot) ? |
| 880 | "ROOT!" : node->parent->d_dir.d_name), |
| 881 | newnode->d_dir.d_name); |
| 882 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 883 | "test: %s\n", |
| 884 | ((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name); |
| 885 | |
| 886 | /* |
| 887 | * orig_vp is set to the original vp if we cloned. |
| 888 | */ |
| 889 | /* node->flags |= DEVFS_CLONED; */ |
| 890 | devfs_allocv(&vp, newnode); |
| 891 | orig_vp = ap->a_vp; |
| 892 | ap->a_vp = vp; |
| 893 | } |
| 894 | } |
| 895 | lockmgr(&devfs_lock, LK_RELEASE); |
| 896 | } |
| 897 | |
| 898 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 899 | "devfs_spec_open() called on %s! \n", |
| 900 | dev->si_name); |
| 901 | |
| 902 | /* |
| 903 | * Make this field valid before any I/O in ->d_open |
| 904 | */ |
| 905 | if (!dev->si_iosize_max) |
| 906 | /* XXX: old DFLTPHYS == 64KB dependency */ |
| 907 | dev->si_iosize_max = min(MAXPHYS,64*1024); |
| 908 | |
| 909 | if (dev_dflags(dev) & D_TTY) |
| 910 | vsetflags(vp, VISTTY); |
| 911 | |
| 912 | /* |
| 913 | * Open underlying device |
| 914 | */ |
| 915 | vn_unlock(vp); |
| 916 | error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred); |
| 917 | vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); |
| 918 | |
| 919 | /* |
| 920 | * Clean up any cloned vp if we error out. |
| 921 | */ |
| 922 | if (error) { |
| 923 | if (orig_vp) { |
| 924 | vput(vp); |
| 925 | ap->a_vp = orig_vp; |
| 926 | /* orig_vp = NULL; */ |
| 927 | } |
| 928 | return error; |
| 929 | } |
| 930 | |
| 931 | /* |
| 932 | * This checks if the disk device is going to be opened for writing. |
| 933 | * It will be only allowed in the cases where securelevel permits it |
| 934 | * and it's not mounted R/W. |
| 935 | */ |
| 936 | if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) && |
| 937 | (ap->a_cred != FSCRED)) { |
| 938 | |
| 939 | /* Very secure mode. No open for writing allowed */ |
| 940 | if (securelevel >= 2) |
| 941 | return EPERM; |
| 942 | |
| 943 | /* |
| 944 | * If it is mounted R/W, do not allow to open for writing. |
| 945 | * In the case it's mounted read-only but securelevel |
| 946 | * is >= 1, then do not allow opening for writing either. |
| 947 | */ |
| 948 | if (vfs_mountedon(vp)) { |
| 949 | if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY)) |
| 950 | return EBUSY; |
| 951 | else if (securelevel >= 1) |
| 952 | return EPERM; |
| 953 | } |
| 954 | } |
| 955 | |
| 956 | if (dev_dflags(dev) & D_TTY) { |
| 957 | if (dev->si_tty) { |
| 958 | struct tty *tp; |
| 959 | tp = dev->si_tty; |
| 960 | if (!tp->t_stop) { |
| 961 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 962 | "devfs: no t_stop\n"); |
| 963 | tp->t_stop = nottystop; |
| 964 | } |
| 965 | } |
| 966 | } |
| 967 | |
| 968 | |
| 969 | if (vn_isdisk(vp, NULL)) { |
| 970 | if (!dev->si_bsize_phys) |
| 971 | dev->si_bsize_phys = DEV_BSIZE; |
| 972 | vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1); |
| 973 | } |
| 974 | |
| 975 | vop_stdopen(ap); |
| 976 | #if 0 |
| 977 | if (node) |
| 978 | nanotime(&node->atime); |
| 979 | #endif |
| 980 | |
| 981 | /* |
| 982 | * If we replaced the vp the vop_stdopen() call will have loaded |
| 983 | * it into fp->f_data and vref()d the vp, giving us two refs. So |
| 984 | * instead of just unlocking it here we have to vput() it. |
| 985 | */ |
| 986 | if (orig_vp) |
| 987 | vput(vp); |
| 988 | |
| 989 | /* Ugly pty magic, to make pty devices appear once they are opened */ |
| 990 | if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY) |
| 991 | node->flags &= ~DEVFS_INVISIBLE; |
| 992 | |
| 993 | if (ap->a_fp) { |
| 994 | KKASSERT(ap->a_fp->f_type == DTYPE_VNODE); |
| 995 | KKASSERT((ap->a_fp->f_flag & FMASK) == (ap->a_mode & FMASK)); |
| 996 | ap->a_fp->f_ops = &devfs_dev_fileops; |
| 997 | KKASSERT(ap->a_fp->f_data == (void *)vp); |
| 998 | } |
| 999 | |
| 1000 | return 0; |
| 1001 | } |
| 1002 | |
| 1003 | |
| 1004 | static int |
| 1005 | devfs_spec_close(struct vop_close_args *ap) |
| 1006 | { |
| 1007 | struct devfs_node *node; |
| 1008 | struct proc *p = curproc; |
| 1009 | struct vnode *vp = ap->a_vp; |
| 1010 | cdev_t dev = vp->v_rdev; |
| 1011 | int error = 0; |
| 1012 | int needrelock; |
| 1013 | |
| 1014 | if (dev) |
| 1015 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 1016 | "devfs_spec_close() called on %s! \n", |
| 1017 | dev->si_name); |
| 1018 | else |
| 1019 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 1020 | "devfs_spec_close() called, null vode!\n"); |
| 1021 | |
| 1022 | /* |
| 1023 | * A couple of hacks for devices and tty devices. The |
| 1024 | * vnode ref count cannot be used to figure out the |
| 1025 | * last close, but we can use v_opencount now that |
| 1026 | * revoke works properly. |
| 1027 | * |
| 1028 | * Detect the last close on a controlling terminal and clear |
| 1029 | * the session (half-close). |
| 1030 | */ |
| 1031 | if (dev) |
| 1032 | reference_dev(dev); |
| 1033 | |
| 1034 | if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) { |
| 1035 | p->p_session->s_ttyvp = NULL; |
| 1036 | vrele(vp); |
| 1037 | } |
| 1038 | |
| 1039 | /* |
| 1040 | * Vnodes can be opened and closed multiple times. Do not really |
| 1041 | * close the device unless (1) it is being closed forcibly, |
| 1042 | * (2) the device wants to track closes, or (3) this is the last |
| 1043 | * vnode doing its last close on the device. |
| 1044 | * |
| 1045 | * XXX the VXLOCK (force close) case can leave vnodes referencing |
| 1046 | * a closed device. This might not occur now that our revoke is |
| 1047 | * fixed. |
| 1048 | */ |
| 1049 | devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n"); |
| 1050 | if (dev && ((vp->v_flag & VRECLAIMED) || |
| 1051 | (dev_dflags(dev) & D_TRACKCLOSE) || |
| 1052 | (vp->v_opencount == 1))) { |
| 1053 | /* |
| 1054 | * Ugly pty magic, to make pty devices disappear again once |
| 1055 | * they are closed. |
| 1056 | */ |
| 1057 | node = DEVFS_NODE(ap->a_vp); |
| 1058 | if (node && (node->flags & DEVFS_PTY)) |
| 1059 | node->flags |= DEVFS_INVISIBLE; |
| 1060 | |
| 1061 | /* |
| 1062 | * Unlock around dev_dclose(), unless the vnode is |
| 1063 | * undergoing a vgone/reclaim (during umount). |
| 1064 | */ |
| 1065 | needrelock = 0; |
| 1066 | if ((vp->v_flag & VRECLAIMED) == 0 && vn_islocked(vp)) { |
| 1067 | needrelock = 1; |
| 1068 | vn_unlock(vp); |
| 1069 | } |
| 1070 | |
| 1071 | /* |
| 1072 | * WARNING! If the device destroys itself the devfs node |
| 1073 | * can disappear here. |
| 1074 | * |
| 1075 | * WARNING! vn_lock() will fail if the vp is in a VRECLAIM, |
| 1076 | * which can occur during umount. |
| 1077 | */ |
| 1078 | error = dev_dclose(dev, ap->a_fflag, S_IFCHR); |
| 1079 | /* node is now stale */ |
| 1080 | |
| 1081 | if (needrelock) { |
| 1082 | if (vn_lock(vp, LK_EXCLUSIVE | LK_RETRY) != 0) { |
| 1083 | panic("devfs_spec_close: vnode %p " |
| 1084 | "unexpectedly could not be relocked", |
| 1085 | vp); |
| 1086 | } |
| 1087 | } |
| 1088 | } else { |
| 1089 | error = 0; |
| 1090 | } |
| 1091 | devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n"); |
| 1092 | |
| 1093 | /* |
| 1094 | * Track the actual opens and closes on the vnode. The last close |
| 1095 | * disassociates the rdev. If the rdev is already disassociated or |
| 1096 | * the opencount is already 0, the vnode might have been revoked |
| 1097 | * and no further opencount tracking occurs. |
| 1098 | */ |
| 1099 | if (dev) |
| 1100 | release_dev(dev); |
| 1101 | if (vp->v_opencount > 0) |
| 1102 | vop_stdclose(ap); |
| 1103 | return(error); |
| 1104 | |
| 1105 | } |
| 1106 | |
| 1107 | |
| 1108 | static int |
| 1109 | devfs_fo_close(struct file *fp) |
| 1110 | { |
| 1111 | struct vnode *vp = (struct vnode *)fp->f_data; |
| 1112 | int error; |
| 1113 | |
| 1114 | fp->f_ops = &badfileops; |
| 1115 | error = vn_close(vp, fp->f_flag); |
| 1116 | |
| 1117 | return (error); |
| 1118 | } |
| 1119 | |
| 1120 | |
| 1121 | /* |
| 1122 | * Device-optimized file table vnode read routine. |
| 1123 | * |
| 1124 | * This bypasses the VOP table and talks directly to the device. Most |
| 1125 | * filesystems just route to specfs and can make this optimization. |
| 1126 | * |
| 1127 | * MPALMOSTSAFE - acquires mplock |
| 1128 | */ |
| 1129 | static int |
| 1130 | devfs_fo_read(struct file *fp, struct uio *uio, |
| 1131 | struct ucred *cred, int flags) |
| 1132 | { |
| 1133 | struct devfs_node *node; |
| 1134 | struct vnode *vp; |
| 1135 | int ioflag; |
| 1136 | int error; |
| 1137 | cdev_t dev; |
| 1138 | |
| 1139 | KASSERT(uio->uio_td == curthread, |
| 1140 | ("uio_td %p is not td %p", uio->uio_td, curthread)); |
| 1141 | |
| 1142 | if (uio->uio_resid == 0) |
| 1143 | return 0; |
| 1144 | |
| 1145 | vp = (struct vnode *)fp->f_data; |
| 1146 | if (vp == NULL || vp->v_type == VBAD) |
| 1147 | return EBADF; |
| 1148 | |
| 1149 | node = DEVFS_NODE(vp); |
| 1150 | |
| 1151 | if ((dev = vp->v_rdev) == NULL) |
| 1152 | return EBADF; |
| 1153 | |
| 1154 | reference_dev(dev); |
| 1155 | |
| 1156 | if ((flags & O_FOFFSET) == 0) |
| 1157 | uio->uio_offset = fp->f_offset; |
| 1158 | |
| 1159 | ioflag = 0; |
| 1160 | if (flags & O_FBLOCKING) { |
| 1161 | /* ioflag &= ~IO_NDELAY; */ |
| 1162 | } else if (flags & O_FNONBLOCKING) { |
| 1163 | ioflag |= IO_NDELAY; |
| 1164 | } else if (fp->f_flag & FNONBLOCK) { |
| 1165 | ioflag |= IO_NDELAY; |
| 1166 | } |
| 1167 | if (flags & O_FBUFFERED) { |
| 1168 | /* ioflag &= ~IO_DIRECT; */ |
| 1169 | } else if (flags & O_FUNBUFFERED) { |
| 1170 | ioflag |= IO_DIRECT; |
| 1171 | } else if (fp->f_flag & O_DIRECT) { |
| 1172 | ioflag |= IO_DIRECT; |
| 1173 | } |
| 1174 | ioflag |= sequential_heuristic(uio, fp); |
| 1175 | |
| 1176 | error = dev_dread(dev, uio, ioflag); |
| 1177 | |
| 1178 | release_dev(dev); |
| 1179 | if (node) |
| 1180 | nanotime(&node->atime); |
| 1181 | if ((flags & O_FOFFSET) == 0) |
| 1182 | fp->f_offset = uio->uio_offset; |
| 1183 | fp->f_nextoff = uio->uio_offset; |
| 1184 | |
| 1185 | return (error); |
| 1186 | } |
| 1187 | |
| 1188 | |
| 1189 | static int |
| 1190 | devfs_fo_write(struct file *fp, struct uio *uio, |
| 1191 | struct ucred *cred, int flags) |
| 1192 | { |
| 1193 | struct devfs_node *node; |
| 1194 | struct vnode *vp; |
| 1195 | int ioflag; |
| 1196 | int error; |
| 1197 | cdev_t dev; |
| 1198 | |
| 1199 | KASSERT(uio->uio_td == curthread, |
| 1200 | ("uio_td %p is not p %p", uio->uio_td, curthread)); |
| 1201 | |
| 1202 | vp = (struct vnode *)fp->f_data; |
| 1203 | if (vp == NULL || vp->v_type == VBAD) |
| 1204 | return EBADF; |
| 1205 | |
| 1206 | node = DEVFS_NODE(vp); |
| 1207 | |
| 1208 | if (vp->v_type == VREG) |
| 1209 | bwillwrite(uio->uio_resid); |
| 1210 | |
| 1211 | vp = (struct vnode *)fp->f_data; |
| 1212 | |
| 1213 | if ((dev = vp->v_rdev) == NULL) |
| 1214 | return EBADF; |
| 1215 | |
| 1216 | reference_dev(dev); |
| 1217 | |
| 1218 | if ((flags & O_FOFFSET) == 0) |
| 1219 | uio->uio_offset = fp->f_offset; |
| 1220 | |
| 1221 | ioflag = IO_UNIT; |
| 1222 | if (vp->v_type == VREG && |
| 1223 | ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) { |
| 1224 | ioflag |= IO_APPEND; |
| 1225 | } |
| 1226 | |
| 1227 | if (flags & O_FBLOCKING) { |
| 1228 | /* ioflag &= ~IO_NDELAY; */ |
| 1229 | } else if (flags & O_FNONBLOCKING) { |
| 1230 | ioflag |= IO_NDELAY; |
| 1231 | } else if (fp->f_flag & FNONBLOCK) { |
| 1232 | ioflag |= IO_NDELAY; |
| 1233 | } |
| 1234 | if (flags & O_FBUFFERED) { |
| 1235 | /* ioflag &= ~IO_DIRECT; */ |
| 1236 | } else if (flags & O_FUNBUFFERED) { |
| 1237 | ioflag |= IO_DIRECT; |
| 1238 | } else if (fp->f_flag & O_DIRECT) { |
| 1239 | ioflag |= IO_DIRECT; |
| 1240 | } |
| 1241 | if (flags & O_FASYNCWRITE) { |
| 1242 | /* ioflag &= ~IO_SYNC; */ |
| 1243 | } else if (flags & O_FSYNCWRITE) { |
| 1244 | ioflag |= IO_SYNC; |
| 1245 | } else if (fp->f_flag & O_FSYNC) { |
| 1246 | ioflag |= IO_SYNC; |
| 1247 | } |
| 1248 | |
| 1249 | if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)) |
| 1250 | ioflag |= IO_SYNC; |
| 1251 | ioflag |= sequential_heuristic(uio, fp); |
| 1252 | |
| 1253 | error = dev_dwrite(dev, uio, ioflag); |
| 1254 | |
| 1255 | release_dev(dev); |
| 1256 | if (node) { |
| 1257 | nanotime(&node->atime); |
| 1258 | nanotime(&node->mtime); |
| 1259 | } |
| 1260 | |
| 1261 | if ((flags & O_FOFFSET) == 0) |
| 1262 | fp->f_offset = uio->uio_offset; |
| 1263 | fp->f_nextoff = uio->uio_offset; |
| 1264 | |
| 1265 | return (error); |
| 1266 | } |
| 1267 | |
| 1268 | |
| 1269 | static int |
| 1270 | devfs_fo_stat(struct file *fp, struct stat *sb, struct ucred *cred) |
| 1271 | { |
| 1272 | struct vnode *vp; |
| 1273 | struct vattr vattr; |
| 1274 | struct vattr *vap; |
| 1275 | u_short mode; |
| 1276 | cdev_t dev; |
| 1277 | int error; |
| 1278 | |
| 1279 | vp = (struct vnode *)fp->f_data; |
| 1280 | if (vp == NULL || vp->v_type == VBAD) |
| 1281 | return EBADF; |
| 1282 | |
| 1283 | error = vn_stat(vp, sb, cred); |
| 1284 | if (error) |
| 1285 | return (error); |
| 1286 | |
| 1287 | vap = &vattr; |
| 1288 | error = VOP_GETATTR(vp, vap); |
| 1289 | if (error) |
| 1290 | return (error); |
| 1291 | |
| 1292 | /* |
| 1293 | * Zero the spare stat fields |
| 1294 | */ |
| 1295 | sb->st_lspare = 0; |
| 1296 | sb->st_qspare1 = 0; |
| 1297 | sb->st_qspare2 = 0; |
| 1298 | |
| 1299 | /* |
| 1300 | * Copy from vattr table ... or not in case it's a cloned device |
| 1301 | */ |
| 1302 | if (vap->va_fsid != VNOVAL) |
| 1303 | sb->st_dev = vap->va_fsid; |
| 1304 | else |
| 1305 | sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; |
| 1306 | |
| 1307 | sb->st_ino = vap->va_fileid; |
| 1308 | |
| 1309 | mode = vap->va_mode; |
| 1310 | mode |= S_IFCHR; |
| 1311 | sb->st_mode = mode; |
| 1312 | |
| 1313 | if (vap->va_nlink > (nlink_t)-1) |
| 1314 | sb->st_nlink = (nlink_t)-1; |
| 1315 | else |
| 1316 | sb->st_nlink = vap->va_nlink; |
| 1317 | |
| 1318 | sb->st_uid = vap->va_uid; |
| 1319 | sb->st_gid = vap->va_gid; |
| 1320 | sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev); |
| 1321 | sb->st_size = vap->va_bytes; |
| 1322 | sb->st_atimespec = vap->va_atime; |
| 1323 | sb->st_mtimespec = vap->va_mtime; |
| 1324 | sb->st_ctimespec = vap->va_ctime; |
| 1325 | |
| 1326 | /* |
| 1327 | * A VCHR and VBLK device may track the last access and last modified |
| 1328 | * time independantly of the filesystem. This is particularly true |
| 1329 | * because device read and write calls may bypass the filesystem. |
| 1330 | */ |
| 1331 | if (vp->v_type == VCHR || vp->v_type == VBLK) { |
| 1332 | dev = vp->v_rdev; |
| 1333 | if (dev != NULL) { |
| 1334 | if (dev->si_lastread) { |
| 1335 | sb->st_atimespec.tv_sec = dev->si_lastread; |
| 1336 | sb->st_atimespec.tv_nsec = 0; |
| 1337 | } |
| 1338 | if (dev->si_lastwrite) { |
| 1339 | sb->st_atimespec.tv_sec = dev->si_lastwrite; |
| 1340 | sb->st_atimespec.tv_nsec = 0; |
| 1341 | } |
| 1342 | } |
| 1343 | } |
| 1344 | |
| 1345 | /* |
| 1346 | * According to www.opengroup.org, the meaning of st_blksize is |
| 1347 | * "a filesystem-specific preferred I/O block size for this |
| 1348 | * object. In some filesystem types, this may vary from file |
| 1349 | * to file" |
| 1350 | * Default to PAGE_SIZE after much discussion. |
| 1351 | */ |
| 1352 | |
| 1353 | sb->st_blksize = PAGE_SIZE; |
| 1354 | |
| 1355 | sb->st_flags = vap->va_flags; |
| 1356 | |
| 1357 | error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0); |
| 1358 | if (error) |
| 1359 | sb->st_gen = 0; |
| 1360 | else |
| 1361 | sb->st_gen = (u_int32_t)vap->va_gen; |
| 1362 | |
| 1363 | sb->st_blocks = vap->va_bytes / S_BLKSIZE; |
| 1364 | |
| 1365 | return (0); |
| 1366 | } |
| 1367 | |
| 1368 | |
| 1369 | static int |
| 1370 | devfs_fo_kqfilter(struct file *fp, struct knote *kn) |
| 1371 | { |
| 1372 | struct vnode *vp; |
| 1373 | int error; |
| 1374 | cdev_t dev; |
| 1375 | |
| 1376 | vp = (struct vnode *)fp->f_data; |
| 1377 | if (vp == NULL || vp->v_type == VBAD) { |
| 1378 | error = EBADF; |
| 1379 | goto done; |
| 1380 | } |
| 1381 | if ((dev = vp->v_rdev) == NULL) { |
| 1382 | error = EBADF; |
| 1383 | goto done; |
| 1384 | } |
| 1385 | reference_dev(dev); |
| 1386 | |
| 1387 | error = dev_dkqfilter(dev, kn); |
| 1388 | |
| 1389 | release_dev(dev); |
| 1390 | |
| 1391 | done: |
| 1392 | return (error); |
| 1393 | } |
| 1394 | |
| 1395 | /* |
| 1396 | * MPALMOSTSAFE - acquires mplock |
| 1397 | */ |
| 1398 | static int |
| 1399 | devfs_fo_ioctl(struct file *fp, u_long com, caddr_t data, |
| 1400 | struct ucred *ucred, struct sysmsg *msg) |
| 1401 | { |
| 1402 | #if 0 |
| 1403 | struct devfs_node *node; |
| 1404 | #endif |
| 1405 | struct vnode *vp; |
| 1406 | struct vnode *ovp; |
| 1407 | cdev_t dev; |
| 1408 | int error; |
| 1409 | struct fiodname_args *name_args; |
| 1410 | size_t namlen; |
| 1411 | const char *name; |
| 1412 | |
| 1413 | vp = ((struct vnode *)fp->f_data); |
| 1414 | |
| 1415 | if ((dev = vp->v_rdev) == NULL) |
| 1416 | return EBADF; /* device was revoked */ |
| 1417 | |
| 1418 | reference_dev(dev); |
| 1419 | |
| 1420 | #if 0 |
| 1421 | node = DEVFS_NODE(vp); |
| 1422 | #endif |
| 1423 | |
| 1424 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 1425 | "devfs_fo_ioctl() called! for dev %s\n", |
| 1426 | dev->si_name); |
| 1427 | |
| 1428 | if (com == FIODTYPE) { |
| 1429 | *(int *)data = dev_dflags(dev) & D_TYPEMASK; |
| 1430 | error = 0; |
| 1431 | goto out; |
| 1432 | } else if (com == FIODNAME) { |
| 1433 | name_args = (struct fiodname_args *)data; |
| 1434 | name = dev->si_name; |
| 1435 | namlen = strlen(name) + 1; |
| 1436 | |
| 1437 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 1438 | "ioctl, got: FIODNAME for %s\n", name); |
| 1439 | |
| 1440 | if (namlen <= name_args->len) |
| 1441 | error = copyout(dev->si_name, name_args->name, namlen); |
| 1442 | else |
| 1443 | error = EINVAL; |
| 1444 | |
| 1445 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 1446 | "ioctl stuff: error: %d\n", error); |
| 1447 | goto out; |
| 1448 | } |
| 1449 | |
| 1450 | error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg); |
| 1451 | |
| 1452 | #if 0 |
| 1453 | if (node) { |
| 1454 | nanotime(&node->atime); |
| 1455 | nanotime(&node->mtime); |
| 1456 | } |
| 1457 | #endif |
| 1458 | if (com == TIOCSCTTY) { |
| 1459 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 1460 | "devfs_fo_ioctl: got TIOCSCTTY on %s\n", |
| 1461 | dev->si_name); |
| 1462 | } |
| 1463 | if (error == 0 && com == TIOCSCTTY) { |
| 1464 | struct proc *p = curthread->td_proc; |
| 1465 | struct session *sess; |
| 1466 | |
| 1467 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 1468 | "devfs_fo_ioctl: dealing with TIOCSCTTY on %s\n", |
| 1469 | dev->si_name); |
| 1470 | if (p == NULL) { |
| 1471 | error = ENOTTY; |
| 1472 | goto out; |
| 1473 | } |
| 1474 | sess = p->p_session; |
| 1475 | |
| 1476 | /* |
| 1477 | * Do nothing if reassigning same control tty |
| 1478 | */ |
| 1479 | if (sess->s_ttyvp == vp) { |
| 1480 | error = 0; |
| 1481 | goto out; |
| 1482 | } |
| 1483 | |
| 1484 | /* |
| 1485 | * Get rid of reference to old control tty |
| 1486 | */ |
| 1487 | ovp = sess->s_ttyvp; |
| 1488 | vref(vp); |
| 1489 | sess->s_ttyvp = vp; |
| 1490 | if (ovp) |
| 1491 | vrele(ovp); |
| 1492 | } |
| 1493 | |
| 1494 | out: |
| 1495 | release_dev(dev); |
| 1496 | devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_fo_ioctl() finished! \n"); |
| 1497 | return (error); |
| 1498 | } |
| 1499 | |
| 1500 | |
| 1501 | static int |
| 1502 | devfs_spec_fsync(struct vop_fsync_args *ap) |
| 1503 | { |
| 1504 | struct vnode *vp = ap->a_vp; |
| 1505 | int error; |
| 1506 | |
| 1507 | if (!vn_isdisk(vp, NULL)) |
| 1508 | return (0); |
| 1509 | |
| 1510 | /* |
| 1511 | * Flush all dirty buffers associated with a block device. |
| 1512 | */ |
| 1513 | error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL); |
| 1514 | return (error); |
| 1515 | } |
| 1516 | |
| 1517 | static int |
| 1518 | devfs_spec_read(struct vop_read_args *ap) |
| 1519 | { |
| 1520 | struct devfs_node *node; |
| 1521 | struct vnode *vp; |
| 1522 | struct uio *uio; |
| 1523 | cdev_t dev; |
| 1524 | int error; |
| 1525 | |
| 1526 | vp = ap->a_vp; |
| 1527 | dev = vp->v_rdev; |
| 1528 | uio = ap->a_uio; |
| 1529 | node = DEVFS_NODE(vp); |
| 1530 | |
| 1531 | if (dev == NULL) /* device was revoked */ |
| 1532 | return (EBADF); |
| 1533 | if (uio->uio_resid == 0) |
| 1534 | return (0); |
| 1535 | |
| 1536 | vn_unlock(vp); |
| 1537 | error = dev_dread(dev, uio, ap->a_ioflag); |
| 1538 | vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); |
| 1539 | |
| 1540 | if (node) |
| 1541 | nanotime(&node->atime); |
| 1542 | |
| 1543 | return (error); |
| 1544 | } |
| 1545 | |
| 1546 | /* |
| 1547 | * Vnode op for write |
| 1548 | * |
| 1549 | * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag, |
| 1550 | * struct ucred *a_cred) |
| 1551 | */ |
| 1552 | static int |
| 1553 | devfs_spec_write(struct vop_write_args *ap) |
| 1554 | { |
| 1555 | struct devfs_node *node; |
| 1556 | struct vnode *vp; |
| 1557 | struct uio *uio; |
| 1558 | cdev_t dev; |
| 1559 | int error; |
| 1560 | |
| 1561 | vp = ap->a_vp; |
| 1562 | dev = vp->v_rdev; |
| 1563 | uio = ap->a_uio; |
| 1564 | node = DEVFS_NODE(vp); |
| 1565 | |
| 1566 | KKASSERT(uio->uio_segflg != UIO_NOCOPY); |
| 1567 | |
| 1568 | if (dev == NULL) /* device was revoked */ |
| 1569 | return (EBADF); |
| 1570 | |
| 1571 | vn_unlock(vp); |
| 1572 | error = dev_dwrite(dev, uio, ap->a_ioflag); |
| 1573 | vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); |
| 1574 | |
| 1575 | if (node) { |
| 1576 | nanotime(&node->atime); |
| 1577 | nanotime(&node->mtime); |
| 1578 | } |
| 1579 | |
| 1580 | return (error); |
| 1581 | } |
| 1582 | |
| 1583 | /* |
| 1584 | * Device ioctl operation. |
| 1585 | * |
| 1586 | * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data, |
| 1587 | * int a_fflag, struct ucred *a_cred, struct sysmsg *msg) |
| 1588 | */ |
| 1589 | static int |
| 1590 | devfs_spec_ioctl(struct vop_ioctl_args *ap) |
| 1591 | { |
| 1592 | struct vnode *vp = ap->a_vp; |
| 1593 | #if 0 |
| 1594 | struct devfs_node *node; |
| 1595 | #endif |
| 1596 | cdev_t dev; |
| 1597 | |
| 1598 | if ((dev = vp->v_rdev) == NULL) |
| 1599 | return (EBADF); /* device was revoked */ |
| 1600 | #if 0 |
| 1601 | node = DEVFS_NODE(vp); |
| 1602 | |
| 1603 | if (node) { |
| 1604 | nanotime(&node->atime); |
| 1605 | nanotime(&node->mtime); |
| 1606 | } |
| 1607 | #endif |
| 1608 | |
| 1609 | return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag, |
| 1610 | ap->a_cred, ap->a_sysmsg)); |
| 1611 | } |
| 1612 | |
| 1613 | /* |
| 1614 | * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn) |
| 1615 | */ |
| 1616 | /* ARGSUSED */ |
| 1617 | static int |
| 1618 | devfs_spec_kqfilter(struct vop_kqfilter_args *ap) |
| 1619 | { |
| 1620 | struct vnode *vp = ap->a_vp; |
| 1621 | #if 0 |
| 1622 | struct devfs_node *node; |
| 1623 | #endif |
| 1624 | cdev_t dev; |
| 1625 | |
| 1626 | if ((dev = vp->v_rdev) == NULL) |
| 1627 | return (EBADF); /* device was revoked (EBADF) */ |
| 1628 | #if 0 |
| 1629 | node = DEVFS_NODE(vp); |
| 1630 | |
| 1631 | if (node) |
| 1632 | nanotime(&node->atime); |
| 1633 | #endif |
| 1634 | |
| 1635 | return (dev_dkqfilter(dev, ap->a_kn)); |
| 1636 | } |
| 1637 | |
| 1638 | /* |
| 1639 | * Convert a vnode strategy call into a device strategy call. Vnode strategy |
| 1640 | * calls are not limited to device DMA limits so we have to deal with the |
| 1641 | * case. |
| 1642 | * |
| 1643 | * spec_strategy(struct vnode *a_vp, struct bio *a_bio) |
| 1644 | */ |
| 1645 | static int |
| 1646 | devfs_spec_strategy(struct vop_strategy_args *ap) |
| 1647 | { |
| 1648 | struct bio *bio = ap->a_bio; |
| 1649 | struct buf *bp = bio->bio_buf; |
| 1650 | struct buf *nbp; |
| 1651 | struct vnode *vp; |
| 1652 | struct mount *mp; |
| 1653 | int chunksize; |
| 1654 | int maxiosize; |
| 1655 | |
| 1656 | if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL) |
| 1657 | buf_start(bp); |
| 1658 | |
| 1659 | /* |
| 1660 | * Collect statistics on synchronous and asynchronous read |
| 1661 | * and write counts for disks that have associated filesystems. |
| 1662 | */ |
| 1663 | vp = ap->a_vp; |
| 1664 | KKASSERT(vp->v_rdev != NULL); /* XXX */ |
| 1665 | if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) { |
| 1666 | if (bp->b_cmd == BUF_CMD_READ) { |
| 1667 | if (bp->b_flags & BIO_SYNC) |
| 1668 | mp->mnt_stat.f_syncreads++; |
| 1669 | else |
| 1670 | mp->mnt_stat.f_asyncreads++; |
| 1671 | } else { |
| 1672 | if (bp->b_flags & BIO_SYNC) |
| 1673 | mp->mnt_stat.f_syncwrites++; |
| 1674 | else |
| 1675 | mp->mnt_stat.f_asyncwrites++; |
| 1676 | } |
| 1677 | } |
| 1678 | |
| 1679 | /* |
| 1680 | * Device iosize limitations only apply to read and write. Shortcut |
| 1681 | * the I/O if it fits. |
| 1682 | */ |
| 1683 | if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) { |
| 1684 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 1685 | "%s: si_iosize_max not set!\n", |
| 1686 | dev_dname(vp->v_rdev)); |
| 1687 | maxiosize = MAXPHYS; |
| 1688 | } |
| 1689 | #if SPEC_CHAIN_DEBUG & 2 |
| 1690 | maxiosize = 4096; |
| 1691 | #endif |
| 1692 | if (bp->b_bcount <= maxiosize || |
| 1693 | (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) { |
| 1694 | dev_dstrategy_chain(vp->v_rdev, bio); |
| 1695 | return (0); |
| 1696 | } |
| 1697 | |
| 1698 | /* |
| 1699 | * Clone the buffer and set up an I/O chain to chunk up the I/O. |
| 1700 | */ |
| 1701 | nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO); |
| 1702 | initbufbio(nbp); |
| 1703 | buf_dep_init(nbp); |
| 1704 | BUF_LOCK(nbp, LK_EXCLUSIVE); |
| 1705 | BUF_KERNPROC(nbp); |
| 1706 | nbp->b_vp = vp; |
| 1707 | nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP); |
| 1708 | nbp->b_data = bp->b_data; |
| 1709 | nbp->b_bio1.bio_done = devfs_spec_strategy_done; |
| 1710 | nbp->b_bio1.bio_offset = bio->bio_offset; |
| 1711 | nbp->b_bio1.bio_caller_info1.ptr = bio; |
| 1712 | |
| 1713 | /* |
| 1714 | * Start the first transfer |
| 1715 | */ |
| 1716 | if (vn_isdisk(vp, NULL)) |
| 1717 | chunksize = vp->v_rdev->si_bsize_phys; |
| 1718 | else |
| 1719 | chunksize = DEV_BSIZE; |
| 1720 | chunksize = maxiosize / chunksize * chunksize; |
| 1721 | #if SPEC_CHAIN_DEBUG & 1 |
| 1722 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 1723 | "spec_strategy chained I/O chunksize=%d\n", |
| 1724 | chunksize); |
| 1725 | #endif |
| 1726 | nbp->b_cmd = bp->b_cmd; |
| 1727 | nbp->b_bcount = chunksize; |
| 1728 | nbp->b_bufsize = chunksize; /* used to detect a short I/O */ |
| 1729 | nbp->b_bio1.bio_caller_info2.index = chunksize; |
| 1730 | |
| 1731 | #if SPEC_CHAIN_DEBUG & 1 |
| 1732 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 1733 | "spec_strategy: chain %p offset %d/%d bcount %d\n", |
| 1734 | bp, 0, bp->b_bcount, nbp->b_bcount); |
| 1735 | #endif |
| 1736 | |
| 1737 | dev_dstrategy(vp->v_rdev, &nbp->b_bio1); |
| 1738 | |
| 1739 | if (DEVFS_NODE(vp)) { |
| 1740 | nanotime(&DEVFS_NODE(vp)->atime); |
| 1741 | nanotime(&DEVFS_NODE(vp)->mtime); |
| 1742 | } |
| 1743 | |
| 1744 | return (0); |
| 1745 | } |
| 1746 | |
| 1747 | /* |
| 1748 | * Chunked up transfer completion routine - chain transfers until done |
| 1749 | * |
| 1750 | * NOTE: MPSAFE callback. |
| 1751 | */ |
| 1752 | static |
| 1753 | void |
| 1754 | devfs_spec_strategy_done(struct bio *nbio) |
| 1755 | { |
| 1756 | struct buf *nbp = nbio->bio_buf; |
| 1757 | struct bio *bio = nbio->bio_caller_info1.ptr; /* original bio */ |
| 1758 | struct buf *bp = bio->bio_buf; /* original bp */ |
| 1759 | int chunksize = nbio->bio_caller_info2.index; /* chunking */ |
| 1760 | int boffset = nbp->b_data - bp->b_data; |
| 1761 | |
| 1762 | if (nbp->b_flags & B_ERROR) { |
| 1763 | /* |
| 1764 | * An error terminates the chain, propogate the error back |
| 1765 | * to the original bp |
| 1766 | */ |
| 1767 | bp->b_flags |= B_ERROR; |
| 1768 | bp->b_error = nbp->b_error; |
| 1769 | bp->b_resid = bp->b_bcount - boffset + |
| 1770 | (nbp->b_bcount - nbp->b_resid); |
| 1771 | #if SPEC_CHAIN_DEBUG & 1 |
| 1772 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 1773 | "spec_strategy: chain %p error %d bcount %d/%d\n", |
| 1774 | bp, bp->b_error, bp->b_bcount, |
| 1775 | bp->b_bcount - bp->b_resid); |
| 1776 | #endif |
| 1777 | } else if (nbp->b_resid) { |
| 1778 | /* |
| 1779 | * A short read or write terminates the chain |
| 1780 | */ |
| 1781 | bp->b_error = nbp->b_error; |
| 1782 | bp->b_resid = bp->b_bcount - boffset + |
| 1783 | (nbp->b_bcount - nbp->b_resid); |
| 1784 | #if SPEC_CHAIN_DEBUG & 1 |
| 1785 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 1786 | "spec_strategy: chain %p short read(1) " |
| 1787 | "bcount %d/%d\n", |
| 1788 | bp, bp->b_bcount - bp->b_resid, bp->b_bcount); |
| 1789 | #endif |
| 1790 | } else if (nbp->b_bcount != nbp->b_bufsize) { |
| 1791 | /* |
| 1792 | * A short read or write can also occur by truncating b_bcount |
| 1793 | */ |
| 1794 | #if SPEC_CHAIN_DEBUG & 1 |
| 1795 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 1796 | "spec_strategy: chain %p short read(2) " |
| 1797 | "bcount %d/%d\n", |
| 1798 | bp, nbp->b_bcount + boffset, bp->b_bcount); |
| 1799 | #endif |
| 1800 | bp->b_error = 0; |
| 1801 | bp->b_bcount = nbp->b_bcount + boffset; |
| 1802 | bp->b_resid = nbp->b_resid; |
| 1803 | } else if (nbp->b_bcount + boffset == bp->b_bcount) { |
| 1804 | /* |
| 1805 | * No more data terminates the chain |
| 1806 | */ |
| 1807 | #if SPEC_CHAIN_DEBUG & 1 |
| 1808 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 1809 | "spec_strategy: chain %p finished bcount %d\n", |
| 1810 | bp, bp->b_bcount); |
| 1811 | #endif |
| 1812 | bp->b_error = 0; |
| 1813 | bp->b_resid = 0; |
| 1814 | } else { |
| 1815 | /* |
| 1816 | * Continue the chain |
| 1817 | */ |
| 1818 | boffset += nbp->b_bcount; |
| 1819 | nbp->b_data = bp->b_data + boffset; |
| 1820 | nbp->b_bcount = bp->b_bcount - boffset; |
| 1821 | if (nbp->b_bcount > chunksize) |
| 1822 | nbp->b_bcount = chunksize; |
| 1823 | nbp->b_bio1.bio_done = devfs_spec_strategy_done; |
| 1824 | nbp->b_bio1.bio_offset = bio->bio_offset + boffset; |
| 1825 | |
| 1826 | #if SPEC_CHAIN_DEBUG & 1 |
| 1827 | devfs_debug(DEVFS_DEBUG_DEBUG, |
| 1828 | "spec_strategy: chain %p offset %d/%d bcount %d\n", |
| 1829 | bp, boffset, bp->b_bcount, nbp->b_bcount); |
| 1830 | #endif |
| 1831 | |
| 1832 | dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1); |
| 1833 | return; |
| 1834 | } |
| 1835 | |
| 1836 | /* |
| 1837 | * Fall through to here on termination. biodone(bp) and |
| 1838 | * clean up and free nbp. |
| 1839 | */ |
| 1840 | biodone(bio); |
| 1841 | BUF_UNLOCK(nbp); |
| 1842 | uninitbufbio(nbp); |
| 1843 | kfree(nbp, M_DEVBUF); |
| 1844 | } |
| 1845 | |
| 1846 | /* |
| 1847 | * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length) |
| 1848 | */ |
| 1849 | static int |
| 1850 | devfs_spec_freeblks(struct vop_freeblks_args *ap) |
| 1851 | { |
| 1852 | struct buf *bp; |
| 1853 | |
| 1854 | /* |
| 1855 | * XXX: This assumes that strategy does the deed right away. |
| 1856 | * XXX: this may not be TRTTD. |
| 1857 | */ |
| 1858 | KKASSERT(ap->a_vp->v_rdev != NULL); |
| 1859 | if ((ap->a_vp->v_rdev->si_flags & SI_CANFREE) == 0) |
| 1860 | return (0); |
| 1861 | bp = geteblk(ap->a_length); |
| 1862 | bp->b_cmd = BUF_CMD_FREEBLKS; |
| 1863 | bp->b_bio1.bio_offset = ap->a_offset; |
| 1864 | bp->b_bcount = ap->a_length; |
| 1865 | dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1); |
| 1866 | return (0); |
| 1867 | } |
| 1868 | |
| 1869 | /* |
| 1870 | * Implement degenerate case where the block requested is the block |
| 1871 | * returned, and assume that the entire device is contiguous in regards |
| 1872 | * to the contiguous block range (runp and runb). |
| 1873 | * |
| 1874 | * spec_bmap(struct vnode *a_vp, off_t a_loffset, |
| 1875 | * off_t *a_doffsetp, int *a_runp, int *a_runb) |
| 1876 | */ |
| 1877 | static int |
| 1878 | devfs_spec_bmap(struct vop_bmap_args *ap) |
| 1879 | { |
| 1880 | if (ap->a_doffsetp != NULL) |
| 1881 | *ap->a_doffsetp = ap->a_loffset; |
| 1882 | if (ap->a_runp != NULL) |
| 1883 | *ap->a_runp = MAXBSIZE; |
| 1884 | if (ap->a_runb != NULL) { |
| 1885 | if (ap->a_loffset < MAXBSIZE) |
| 1886 | *ap->a_runb = (int)ap->a_loffset; |
| 1887 | else |
| 1888 | *ap->a_runb = MAXBSIZE; |
| 1889 | } |
| 1890 | return (0); |
| 1891 | } |
| 1892 | |
| 1893 | |
| 1894 | /* |
| 1895 | * Special device advisory byte-level locks. |
| 1896 | * |
| 1897 | * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, |
| 1898 | * struct flock *a_fl, int a_flags) |
| 1899 | */ |
| 1900 | /* ARGSUSED */ |
| 1901 | static int |
| 1902 | devfs_spec_advlock(struct vop_advlock_args *ap) |
| 1903 | { |
| 1904 | return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP); |
| 1905 | } |
| 1906 | |
| 1907 | /* |
| 1908 | * NOTE: MPSAFE callback. |
| 1909 | */ |
| 1910 | static void |
| 1911 | devfs_spec_getpages_iodone(struct bio *bio) |
| 1912 | { |
| 1913 | bio->bio_buf->b_cmd = BUF_CMD_DONE; |
| 1914 | wakeup(bio->bio_buf); |
| 1915 | } |
| 1916 | |
| 1917 | /* |
| 1918 | * spec_getpages() - get pages associated with device vnode. |
| 1919 | * |
| 1920 | * Note that spec_read and spec_write do not use the buffer cache, so we |
| 1921 | * must fully implement getpages here. |
| 1922 | */ |
| 1923 | static int |
| 1924 | devfs_spec_getpages(struct vop_getpages_args *ap) |
| 1925 | { |
| 1926 | vm_offset_t kva; |
| 1927 | int error; |
| 1928 | int i, pcount, size; |
| 1929 | struct buf *bp; |
| 1930 | vm_page_t m; |
| 1931 | vm_ooffset_t offset; |
| 1932 | int toff, nextoff, nread; |
| 1933 | struct vnode *vp = ap->a_vp; |
| 1934 | int blksiz; |
| 1935 | int gotreqpage; |
| 1936 | |
| 1937 | error = 0; |
| 1938 | pcount = round_page(ap->a_count) / PAGE_SIZE; |
| 1939 | |
| 1940 | /* |
| 1941 | * Calculate the offset of the transfer and do sanity check. |
| 1942 | */ |
| 1943 | offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset; |
| 1944 | |
| 1945 | /* |
| 1946 | * Round up physical size for real devices. We cannot round using |
| 1947 | * v_mount's block size data because v_mount has nothing to do with |
| 1948 | * the device. i.e. it's usually '/dev'. We need the physical block |
| 1949 | * size for the device itself. |
| 1950 | * |
| 1951 | * We can't use v_rdev->si_mountpoint because it only exists when the |
| 1952 | * block device is mounted. However, we can use v_rdev. |
| 1953 | */ |
| 1954 | if (vn_isdisk(vp, NULL)) |
| 1955 | blksiz = vp->v_rdev->si_bsize_phys; |
| 1956 | else |
| 1957 | blksiz = DEV_BSIZE; |
| 1958 | |
| 1959 | size = (ap->a_count + blksiz - 1) & ~(blksiz - 1); |
| 1960 | |
| 1961 | bp = getpbuf_kva(NULL); |
| 1962 | kva = (vm_offset_t)bp->b_data; |
| 1963 | |
| 1964 | /* |
| 1965 | * Map the pages to be read into the kva. |
| 1966 | */ |
| 1967 | pmap_qenter(kva, ap->a_m, pcount); |
| 1968 | |
| 1969 | /* Build a minimal buffer header. */ |
| 1970 | bp->b_cmd = BUF_CMD_READ; |
| 1971 | bp->b_bcount = size; |
| 1972 | bp->b_resid = 0; |
| 1973 | bsetrunningbufspace(bp, size); |
| 1974 | |
| 1975 | bp->b_bio1.bio_offset = offset; |
| 1976 | bp->b_bio1.bio_done = devfs_spec_getpages_iodone; |
| 1977 | |
| 1978 | mycpu->gd_cnt.v_vnodein++; |
| 1979 | mycpu->gd_cnt.v_vnodepgsin += pcount; |
| 1980 | |
| 1981 | /* Do the input. */ |
| 1982 | vn_strategy(ap->a_vp, &bp->b_bio1); |
| 1983 | |
| 1984 | crit_enter(); |
| 1985 | |
| 1986 | /* We definitely need to be at splbio here. */ |
| 1987 | while (bp->b_cmd != BUF_CMD_DONE) |
| 1988 | tsleep(bp, 0, "spread", 0); |
| 1989 | |
| 1990 | crit_exit(); |
| 1991 | |
| 1992 | if (bp->b_flags & B_ERROR) { |
| 1993 | if (bp->b_error) |
| 1994 | error = bp->b_error; |
| 1995 | else |
| 1996 | error = EIO; |
| 1997 | } |
| 1998 | |
| 1999 | /* |
| 2000 | * If EOF is encountered we must zero-extend the result in order |
| 2001 | * to ensure that the page does not contain garabge. When no |
| 2002 | * error occurs, an early EOF is indicated if b_bcount got truncated. |
| 2003 | * b_resid is relative to b_bcount and should be 0, but some devices |
| 2004 | * might indicate an EOF with b_resid instead of truncating b_bcount. |
| 2005 | */ |
| 2006 | nread = bp->b_bcount - bp->b_resid; |
| 2007 | if (nread < ap->a_count) |
| 2008 | bzero((caddr_t)kva + nread, ap->a_count - nread); |
| 2009 | pmap_qremove(kva, pcount); |
| 2010 | |
| 2011 | gotreqpage = 0; |
| 2012 | for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) { |
| 2013 | nextoff = toff + PAGE_SIZE; |
| 2014 | m = ap->a_m[i]; |
| 2015 | |
| 2016 | m->flags &= ~PG_ZERO; |
| 2017 | |
| 2018 | /* |
| 2019 | * NOTE: vm_page_undirty/clear_dirty etc do not clear the |
| 2020 | * pmap modified bit. pmap modified bit should have |
| 2021 | * already been cleared. |
| 2022 | */ |
| 2023 | if (nextoff <= nread) { |
| 2024 | m->valid = VM_PAGE_BITS_ALL; |
| 2025 | vm_page_undirty(m); |
| 2026 | } else if (toff < nread) { |
| 2027 | /* |
| 2028 | * Since this is a VM request, we have to supply the |
| 2029 | * unaligned offset to allow vm_page_set_valid() |
| 2030 | * to zero sub-DEV_BSIZE'd portions of the page. |
| 2031 | */ |
| 2032 | vm_page_set_valid(m, 0, nread - toff); |
| 2033 | vm_page_clear_dirty_end_nonincl(m, 0, nread - toff); |
| 2034 | } else { |
| 2035 | m->valid = 0; |
| 2036 | vm_page_undirty(m); |
| 2037 | } |
| 2038 | |
| 2039 | if (i != ap->a_reqpage) { |
| 2040 | /* |
| 2041 | * Just in case someone was asking for this page we |
| 2042 | * now tell them that it is ok to use. |
| 2043 | */ |
| 2044 | if (!error || (m->valid == VM_PAGE_BITS_ALL)) { |
| 2045 | if (m->valid) { |
| 2046 | if (m->flags & PG_REFERENCED) { |
| 2047 | vm_page_activate(m); |
| 2048 | } else { |
| 2049 | vm_page_deactivate(m); |
| 2050 | } |
| 2051 | vm_page_wakeup(m); |
| 2052 | } else { |
| 2053 | vm_page_free(m); |
| 2054 | } |
| 2055 | } else { |
| 2056 | vm_page_free(m); |
| 2057 | } |
| 2058 | } else if (m->valid) { |
| 2059 | gotreqpage = 1; |
| 2060 | /* |
| 2061 | * Since this is a VM request, we need to make the |
| 2062 | * entire page presentable by zeroing invalid sections. |
| 2063 | */ |
| 2064 | if (m->valid != VM_PAGE_BITS_ALL) |
| 2065 | vm_page_zero_invalid(m, FALSE); |
| 2066 | } |
| 2067 | } |
| 2068 | if (!gotreqpage) { |
| 2069 | m = ap->a_m[ap->a_reqpage]; |
| 2070 | devfs_debug(DEVFS_DEBUG_WARNING, |
| 2071 | "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n", |
| 2072 | devtoname(vp->v_rdev), error, bp, bp->b_vp); |
| 2073 | devfs_debug(DEVFS_DEBUG_WARNING, |
| 2074 | " size: %d, resid: %d, a_count: %d, valid: 0x%x\n", |
| 2075 | size, bp->b_resid, ap->a_count, m->valid); |
| 2076 | devfs_debug(DEVFS_DEBUG_WARNING, |
| 2077 | " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n", |
| 2078 | nread, ap->a_reqpage, (u_long)m->pindex, pcount); |
| 2079 | /* |
| 2080 | * Free the buffer header back to the swap buffer pool. |
| 2081 | */ |
| 2082 | relpbuf(bp, NULL); |
| 2083 | return VM_PAGER_ERROR; |
| 2084 | } |
| 2085 | /* |
| 2086 | * Free the buffer header back to the swap buffer pool. |
| 2087 | */ |
| 2088 | relpbuf(bp, NULL); |
| 2089 | if (DEVFS_NODE(ap->a_vp)) |
| 2090 | nanotime(&DEVFS_NODE(ap->a_vp)->mtime); |
| 2091 | return VM_PAGER_OK; |
| 2092 | } |
| 2093 | |
| 2094 | static __inline |
| 2095 | int |
| 2096 | sequential_heuristic(struct uio *uio, struct file *fp) |
| 2097 | { |
| 2098 | /* |
| 2099 | * Sequential heuristic - detect sequential operation |
| 2100 | */ |
| 2101 | if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || |
| 2102 | uio->uio_offset == fp->f_nextoff) { |
| 2103 | /* |
| 2104 | * XXX we assume that the filesystem block size is |
| 2105 | * the default. Not true, but still gives us a pretty |
| 2106 | * good indicator of how sequential the read operations |
| 2107 | * are. |
| 2108 | */ |
| 2109 | int tmpseq = fp->f_seqcount; |
| 2110 | |
| 2111 | tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE; |
| 2112 | if (tmpseq > IO_SEQMAX) |
| 2113 | tmpseq = IO_SEQMAX; |
| 2114 | fp->f_seqcount = tmpseq; |
| 2115 | return(fp->f_seqcount << IO_SEQSHIFT); |
| 2116 | } |
| 2117 | |
| 2118 | /* |
| 2119 | * Not sequential, quick draw-down of seqcount |
| 2120 | */ |
| 2121 | if (fp->f_seqcount > 1) |
| 2122 | fp->f_seqcount = 1; |
| 2123 | else |
| 2124 | fp->f_seqcount = 0; |
| 2125 | return(0); |
| 2126 | } |