1 /*- 2 * Copyright (c) 2000,2004 3 * Poul-Henning Kamp. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Neither the name of the University nor the names of its contributors 11 * may be used to endorse or promote products derived from this software 12 * without specific prior written permission. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 * 26 * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vfsops.c 1.36 27 * 28 * $FreeBSD$ 29 */ 30 31 #include "opt_mac.h" 32 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/conf.h> 36 #include <sys/dirent.h> 37 #include <sys/kernel.h> 38 #include <sys/limits.h> 39 #include <sys/lock.h> 40 #include <sys/malloc.h> 41 #include <sys/proc.h> 42 #include <sys/sx.h> 43 #include <sys/sysctl.h> 44 #include <sys/vnode.h> 45 46 #include <sys/kdb.h> 47 48 #include <fs/devfs/devfs.h> 49 #include <fs/devfs/devfs_int.h> 50 51 #include <security/mac/mac_framework.h> 52 53 /* 54 * The one true (but secret) list of active devices in the system. 55 * Locked by dev_lock()/devmtx 56 */ 57 struct cdev_priv_list cdevp_list = TAILQ_HEAD_INITIALIZER(cdevp_list); 58 59 struct unrhdr *devfs_inos; 60 61 62 static MALLOC_DEFINE(M_DEVFS2, "DEVFS2", "DEVFS data 2"); 63 static MALLOC_DEFINE(M_DEVFS3, "DEVFS3", "DEVFS data 3"); 64 static MALLOC_DEFINE(M_CDEVP, "DEVFS1", "DEVFS cdev_priv storage"); 65 66 static SYSCTL_NODE(_vfs, OID_AUTO, devfs, CTLFLAG_RW, 0, "DEVFS filesystem"); 67 68 static unsigned devfs_generation; 69 SYSCTL_UINT(_vfs_devfs, OID_AUTO, generation, CTLFLAG_RD, 70 &devfs_generation, 0, "DEVFS generation number"); 71 72 unsigned devfs_rule_depth = 1; 73 SYSCTL_UINT(_vfs_devfs, OID_AUTO, rule_depth, CTLFLAG_RW, 74 &devfs_rule_depth, 0, "Max depth of ruleset include"); 75 76 /* 77 * Helper sysctl for devname(3). We're given a struct cdev * and return 78 * the name, if any, registered by the device driver. 79 */ 80 static int 81 sysctl_devname(SYSCTL_HANDLER_ARGS) 82 { 83 int error; 84 dev_t ud; 85 struct cdev_priv *cdp; 86 87 error = SYSCTL_IN(req, &ud, sizeof (ud)); 88 if (error) 89 return (error); 90 if (ud == NODEV) 91 return(EINVAL); 92 /* 93 ud ^ devfs_random(); 94 */ 95 dev_lock(); 96 TAILQ_FOREACH(cdp, &cdevp_list, cdp_list) 97 if (cdp->cdp_inode == ud) 98 break; 99 dev_unlock(); 100 if (cdp == NULL) 101 return(ENOENT); 102 return(SYSCTL_OUT(req, cdp->cdp_c.si_name, strlen(cdp->cdp_c.si_name) + 1)); 103 return (error); 104 } 105 106 SYSCTL_PROC(_kern, OID_AUTO, devname, 107 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_ANYBODY|CTLFLAG_MPSAFE, 108 NULL, 0, sysctl_devname, "", "devname(3) handler"); 109 110 SYSCTL_INT(_debug_sizeof, OID_AUTO, cdev, CTLFLAG_RD, 111 0, sizeof(struct cdev), "sizeof(struct cdev)"); 112 113 SYSCTL_INT(_debug_sizeof, OID_AUTO, cdev_priv, CTLFLAG_RD, 114 0, sizeof(struct cdev_priv), "sizeof(struct cdev_priv)"); 115 116 struct cdev * 117 devfs_alloc(void) 118 { 119 struct cdev_priv *cdp; 120 struct cdev *cdev; 121 struct timespec ts; 122 123 cdp = malloc(sizeof *cdp, M_CDEVP, M_USE_RESERVE | M_ZERO | M_WAITOK); 124 125 cdp->cdp_dirents = &cdp->cdp_dirent0; 126 cdp->cdp_dirent0 = NULL; 127 cdp->cdp_maxdirent = 0; 128 129 cdev = &cdp->cdp_c; 130 131 cdev->si_name = cdev->__si_namebuf; 132 LIST_INIT(&cdev->si_children); 133 vfs_timestamp(&ts); 134 cdev->si_atime = cdev->si_mtime = cdev->si_ctime = ts; 135 136 return (cdev); 137 } 138 139 void 140 devfs_free(struct cdev *cdev) 141 { 142 struct cdev_priv *cdp; 143 144 cdp = cdev2priv(cdev); 145 if (cdev->si_cred != NULL) 146 crfree(cdev->si_cred); 147 if (cdp->cdp_inode > 0) 148 free_unr(devfs_inos, cdp->cdp_inode); 149 if (cdp->cdp_maxdirent > 0) 150 free(cdp->cdp_dirents, M_DEVFS2); 151 free(cdp, M_CDEVP); 152 } 153 154 struct devfs_dirent * 155 devfs_find(struct devfs_dirent *dd, const char *name, int namelen) 156 { 157 struct devfs_dirent *de; 158 159 TAILQ_FOREACH(de, &dd->de_dlist, de_list) { 160 if (namelen != de->de_dirent->d_namlen) 161 continue; 162 if (bcmp(name, de->de_dirent->d_name, namelen) != 0) 163 continue; 164 break; 165 } 166 return (de); 167 } 168 169 struct devfs_dirent * 170 devfs_newdirent(char *name, int namelen) 171 { 172 int i; 173 struct devfs_dirent *de; 174 struct dirent d; 175 176 d.d_namlen = namelen; 177 i = sizeof (*de) + GENERIC_DIRSIZ(&d); 178 de = malloc(i, M_DEVFS3, M_WAITOK | M_ZERO); 179 de->de_dirent = (struct dirent *)(de + 1); 180 de->de_dirent->d_namlen = namelen; 181 de->de_dirent->d_reclen = GENERIC_DIRSIZ(&d); 182 bcopy(name, de->de_dirent->d_name, namelen); 183 de->de_dirent->d_name[namelen] = '\0'; 184 vfs_timestamp(&de->de_ctime); 185 de->de_mtime = de->de_atime = de->de_ctime; 186 de->de_links = 1; 187 de->de_holdcnt = 1; 188 #ifdef MAC 189 mac_devfs_init(de); 190 #endif 191 return (de); 192 } 193 194 struct devfs_dirent * 195 devfs_vmkdir(struct devfs_mount *dmp, char *name, int namelen, struct devfs_dirent *dotdot, u_int inode) 196 { 197 struct devfs_dirent *dd; 198 struct devfs_dirent *de; 199 200 /* Create the new directory */ 201 dd = devfs_newdirent(name, namelen); 202 TAILQ_INIT(&dd->de_dlist); 203 dd->de_dirent->d_type = DT_DIR; 204 dd->de_mode = 0555; 205 dd->de_links = 2; 206 dd->de_dir = dd; 207 if (inode != 0) 208 dd->de_inode = inode; 209 else 210 dd->de_inode = alloc_unr(devfs_inos); 211 212 /* Create the "." entry in the new directory */ 213 de = devfs_newdirent(".", 1); 214 de->de_dirent->d_type = DT_DIR; 215 de->de_flags |= DE_DOT; 216 TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list); 217 de->de_dir = dd; 218 219 /* Create the ".." entry in the new directory */ 220 de = devfs_newdirent("..", 2); 221 de->de_dirent->d_type = DT_DIR; 222 de->de_flags |= DE_DOTDOT; 223 TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list); 224 if (dotdot == NULL) { 225 de->de_dir = dd; 226 } else { 227 de->de_dir = dotdot; 228 TAILQ_INSERT_TAIL(&dotdot->de_dlist, dd, de_list); 229 dotdot->de_links++; 230 } 231 232 #ifdef MAC 233 mac_devfs_create_directory(dmp->dm_mount, name, namelen, dd); 234 #endif 235 return (dd); 236 } 237 238 void 239 devfs_dirent_free(struct devfs_dirent *de) 240 { 241 free(de, M_DEVFS3); 242 } 243 244 /* 245 * The caller needs to hold the dm for the duration of the call since 246 * dm->dm_lock may be temporary dropped. 247 */ 248 void 249 devfs_delete(struct devfs_mount *dm, struct devfs_dirent *de, int vp_locked) 250 { 251 struct vnode *vp; 252 253 KASSERT((de->de_flags & DE_DOOMED) == 0, 254 ("devfs_delete doomed dirent")); 255 de->de_flags |= DE_DOOMED; 256 mtx_lock(&devfs_de_interlock); 257 vp = de->de_vnode; 258 if (vp != NULL) { 259 VI_LOCK(vp); 260 mtx_unlock(&devfs_de_interlock); 261 vholdl(vp); 262 sx_unlock(&dm->dm_lock); 263 if (!vp_locked) 264 vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 265 else 266 VI_UNLOCK(vp); 267 vgone(vp); 268 if (!vp_locked) 269 VOP_UNLOCK(vp, 0); 270 vdrop(vp); 271 sx_xlock(&dm->dm_lock); 272 } else 273 mtx_unlock(&devfs_de_interlock); 274 if (de->de_symlink) { 275 free(de->de_symlink, M_DEVFS); 276 de->de_symlink = NULL; 277 } 278 #ifdef MAC 279 mac_devfs_destroy(de); 280 #endif 281 if (de->de_inode > DEVFS_ROOTINO) { 282 free_unr(devfs_inos, de->de_inode); 283 de->de_inode = 0; 284 } 285 if (DEVFS_DE_DROP(de)) 286 devfs_dirent_free(de); 287 } 288 289 /* 290 * Called on unmount. 291 * Recursively removes the entire tree. 292 * The caller needs to hold the dm for the duration of the call. 293 */ 294 295 static void 296 devfs_purge(struct devfs_mount *dm, struct devfs_dirent *dd) 297 { 298 struct devfs_dirent *de; 299 300 sx_assert(&dm->dm_lock, SX_XLOCKED); 301 for (;;) { 302 de = TAILQ_FIRST(&dd->de_dlist); 303 if (de == NULL) 304 break; 305 TAILQ_REMOVE(&dd->de_dlist, de, de_list); 306 if (de->de_flags & (DE_DOT|DE_DOTDOT)) 307 devfs_delete(dm, de, 0); 308 else if (de->de_dirent->d_type == DT_DIR) 309 devfs_purge(dm, de); 310 else 311 devfs_delete(dm, de, 0); 312 } 313 devfs_delete(dm, dd, 0); 314 } 315 316 /* 317 * Each cdev_priv has an array of pointers to devfs_dirent which is indexed 318 * by the mount points dm_idx. 319 * This function extends the array when necessary, taking into account that 320 * the default array is 1 element and not malloc'ed. 321 */ 322 static void 323 devfs_metoo(struct cdev_priv *cdp, struct devfs_mount *dm) 324 { 325 struct devfs_dirent **dep; 326 int siz; 327 328 siz = (dm->dm_idx + 1) * sizeof *dep; 329 dep = malloc(siz, M_DEVFS2, M_WAITOK | M_ZERO); 330 dev_lock(); 331 if (dm->dm_idx <= cdp->cdp_maxdirent) { 332 /* We got raced */ 333 dev_unlock(); 334 free(dep, M_DEVFS2); 335 return; 336 } 337 memcpy(dep, cdp->cdp_dirents, (cdp->cdp_maxdirent + 1) * sizeof *dep); 338 if (cdp->cdp_maxdirent > 0) 339 free(cdp->cdp_dirents, M_DEVFS2); 340 cdp->cdp_dirents = dep; 341 /* 342 * XXX: if malloc told us how much we actually got this could 343 * XXX: be optimized. 344 */ 345 cdp->cdp_maxdirent = dm->dm_idx; 346 dev_unlock(); 347 } 348 349 /* 350 * The caller needs to hold the dm for the duration of the call. 351 */ 352 static int 353 devfs_populate_loop(struct devfs_mount *dm, int cleanup) 354 { 355 struct cdev_priv *cdp; 356 struct devfs_dirent *de; 357 struct devfs_dirent *dd; 358 struct cdev *pdev; 359 int j; 360 char *q, *s; 361 362 sx_assert(&dm->dm_lock, SX_XLOCKED); 363 dev_lock(); 364 TAILQ_FOREACH(cdp, &cdevp_list, cdp_list) { 365 366 KASSERT(cdp->cdp_dirents != NULL, ("NULL cdp_dirents")); 367 368 /* 369 * If we are unmounting, or the device has been destroyed, 370 * clean up our dirent. 371 */ 372 if ((cleanup || !(cdp->cdp_flags & CDP_ACTIVE)) && 373 dm->dm_idx <= cdp->cdp_maxdirent && 374 cdp->cdp_dirents[dm->dm_idx] != NULL) { 375 de = cdp->cdp_dirents[dm->dm_idx]; 376 cdp->cdp_dirents[dm->dm_idx] = NULL; 377 KASSERT(cdp == de->de_cdp, 378 ("%s %d %s %p %p", __func__, __LINE__, 379 cdp->cdp_c.si_name, cdp, de->de_cdp)); 380 KASSERT(de->de_dir != NULL, ("Null de->de_dir")); 381 dev_unlock(); 382 383 TAILQ_REMOVE(&de->de_dir->de_dlist, de, de_list); 384 de->de_cdp = NULL; 385 de->de_inode = 0; 386 devfs_delete(dm, de, 0); 387 dev_lock(); 388 cdp->cdp_inuse--; 389 dev_unlock(); 390 return (1); 391 } 392 /* 393 * GC any lingering devices 394 */ 395 if (!(cdp->cdp_flags & CDP_ACTIVE)) { 396 if (cdp->cdp_inuse > 0) 397 continue; 398 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list); 399 dev_unlock(); 400 dev_rel(&cdp->cdp_c); 401 return (1); 402 } 403 /* 404 * Don't create any new dirents if we are unmounting 405 */ 406 if (cleanup) 407 continue; 408 KASSERT((cdp->cdp_flags & CDP_ACTIVE), ("Bogons, I tell ya'!")); 409 410 if (dm->dm_idx <= cdp->cdp_maxdirent && 411 cdp->cdp_dirents[dm->dm_idx] != NULL) { 412 de = cdp->cdp_dirents[dm->dm_idx]; 413 KASSERT(cdp == de->de_cdp, ("inconsistent cdp")); 414 continue; 415 } 416 417 418 cdp->cdp_inuse++; 419 dev_unlock(); 420 421 if (dm->dm_idx > cdp->cdp_maxdirent) 422 devfs_metoo(cdp, dm); 423 424 dd = dm->dm_rootdir; 425 s = cdp->cdp_c.si_name; 426 for (;;) { 427 for (q = s; *q != '/' && *q != '\0'; q++) 428 continue; 429 if (*q != '/') 430 break; 431 de = devfs_find(dd, s, q - s); 432 if (de == NULL) 433 de = devfs_vmkdir(dm, s, q - s, dd, 0); 434 s = q + 1; 435 dd = de; 436 } 437 438 de = devfs_newdirent(s, q - s); 439 if (cdp->cdp_c.si_flags & SI_ALIAS) { 440 de->de_uid = 0; 441 de->de_gid = 0; 442 de->de_mode = 0755; 443 de->de_dirent->d_type = DT_LNK; 444 pdev = cdp->cdp_c.si_parent; 445 j = strlen(pdev->si_name) + 1; 446 de->de_symlink = malloc(j, M_DEVFS, M_WAITOK); 447 bcopy(pdev->si_name, de->de_symlink, j); 448 } else { 449 de->de_uid = cdp->cdp_c.si_uid; 450 de->de_gid = cdp->cdp_c.si_gid; 451 de->de_mode = cdp->cdp_c.si_mode; 452 de->de_dirent->d_type = DT_CHR; 453 } 454 de->de_inode = cdp->cdp_inode; 455 de->de_cdp = cdp; 456 #ifdef MAC 457 mac_devfs_create_device(cdp->cdp_c.si_cred, dm->dm_mount, 458 &cdp->cdp_c, de); 459 #endif 460 de->de_dir = dd; 461 TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list); 462 devfs_rules_apply(dm, de); 463 dev_lock(); 464 /* XXX: could check that cdp is still active here */ 465 KASSERT(cdp->cdp_dirents[dm->dm_idx] == NULL, 466 ("%s %d\n", __func__, __LINE__)); 467 cdp->cdp_dirents[dm->dm_idx] = de; 468 KASSERT(de->de_cdp != (void *)0xdeadc0de, 469 ("%s %d\n", __func__, __LINE__)); 470 dev_unlock(); 471 return (1); 472 } 473 dev_unlock(); 474 return (0); 475 } 476 477 /* 478 * The caller needs to hold the dm for the duration of the call. 479 */ 480 void 481 devfs_populate(struct devfs_mount *dm) 482 { 483 484 sx_assert(&dm->dm_lock, SX_XLOCKED); 485 if (dm->dm_generation == devfs_generation) 486 return; 487 while (devfs_populate_loop(dm, 0)) 488 continue; 489 dm->dm_generation = devfs_generation; 490 } 491 492 /* 493 * The caller needs to hold the dm for the duration of the call. 494 */ 495 void 496 devfs_cleanup(struct devfs_mount *dm) 497 { 498 499 sx_assert(&dm->dm_lock, SX_XLOCKED); 500 while (devfs_populate_loop(dm, 1)) 501 continue; 502 devfs_purge(dm, dm->dm_rootdir); 503 } 504 505 /* 506 * devfs_create() and devfs_destroy() are called from kern_conf.c and 507 * in both cases the devlock() mutex is held, so no further locking 508 * is necesary and no sleeping allowed. 509 */ 510 511 void 512 devfs_create(struct cdev *dev) 513 { 514 struct cdev_priv *cdp; 515 516 mtx_assert(&devmtx, MA_OWNED); 517 cdp = cdev2priv(dev); 518 cdp->cdp_flags |= CDP_ACTIVE; 519 cdp->cdp_inode = alloc_unrl(devfs_inos); 520 dev_refl(dev); 521 TAILQ_INSERT_TAIL(&cdevp_list, cdp, cdp_list); 522 devfs_generation++; 523 } 524 525 void 526 devfs_destroy(struct cdev *dev) 527 { 528 struct cdev_priv *cdp; 529 530 mtx_assert(&devmtx, MA_OWNED); 531 cdp = cdev2priv(dev); 532 cdp->cdp_flags &= ~CDP_ACTIVE; 533 devfs_generation++; 534 } 535 536 static void 537 devfs_devs_init(void *junk __unused) 538 { 539 540 devfs_inos = new_unrhdr(DEVFS_ROOTINO + 1, INT_MAX, &devmtx); 541 } 542 543 SYSINIT(devfs_devs, SI_SUB_DEVFS, SI_ORDER_FIRST, devfs_devs_init, NULL); 544