1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 /* 29 * utility routines for the /dev fs 30 */ 31 32 #include <sys/types.h> 33 #include <sys/param.h> 34 #include <sys/t_lock.h> 35 #include <sys/systm.h> 36 #include <sys/sysmacros.h> 37 #include <sys/user.h> 38 #include <sys/time.h> 39 #include <sys/vfs.h> 40 #include <sys/vnode.h> 41 #include <sys/file.h> 42 #include <sys/fcntl.h> 43 #include <sys/flock.h> 44 #include <sys/kmem.h> 45 #include <sys/uio.h> 46 #include <sys/errno.h> 47 #include <sys/stat.h> 48 #include <sys/cred.h> 49 #include <sys/dirent.h> 50 #include <sys/pathname.h> 51 #include <sys/cmn_err.h> 52 #include <sys/debug.h> 53 #include <sys/mode.h> 54 #include <sys/policy.h> 55 #include <fs/fs_subr.h> 56 #include <sys/mount.h> 57 #include <sys/fs/snode.h> 58 #include <sys/fs/dv_node.h> 59 #include <sys/fs/sdev_impl.h> 60 #include <sys/fs/sdev_node.h> 61 #include <sys/sunndi.h> 62 #include <sys/sunmdi.h> 63 #include <sys/conf.h> 64 #include <sys/proc.h> 65 #include <sys/user.h> 66 #include <sys/modctl.h> 67 68 #ifdef DEBUG 69 int sdev_debug = 0x00000001; 70 int sdev_debug_cache_flags = 0; 71 #endif 72 73 /* 74 * globals 75 */ 76 /* prototype memory vattrs */ 77 vattr_t sdev_vattr_dir = { 78 AT_TYPE|AT_MODE|AT_UID|AT_GID, /* va_mask */ 79 VDIR, /* va_type */ 80 SDEV_DIRMODE_DEFAULT, /* va_mode */ 81 SDEV_UID_DEFAULT, /* va_uid */ 82 SDEV_GID_DEFAULT, /* va_gid */ 83 0, /* va_fsid */ 84 0, /* va_nodeid */ 85 0, /* va_nlink */ 86 0, /* va_size */ 87 0, /* va_atime */ 88 0, /* va_mtime */ 89 0, /* va_ctime */ 90 0, /* va_rdev */ 91 0, /* va_blksize */ 92 0, /* va_nblocks */ 93 0 /* va_vcode */ 94 }; 95 96 vattr_t sdev_vattr_lnk = { 97 AT_TYPE|AT_MODE, /* va_mask */ 98 VLNK, /* va_type */ 99 SDEV_LNKMODE_DEFAULT, /* va_mode */ 100 SDEV_UID_DEFAULT, /* va_uid */ 101 SDEV_GID_DEFAULT, /* va_gid */ 102 0, /* va_fsid */ 103 0, /* va_nodeid */ 104 0, /* va_nlink */ 105 0, /* va_size */ 106 0, /* va_atime */ 107 0, /* va_mtime */ 108 0, /* va_ctime */ 109 0, /* va_rdev */ 110 0, /* va_blksize */ 111 0, /* va_nblocks */ 112 0 /* va_vcode */ 113 }; 114 115 vattr_t sdev_vattr_blk = { 116 AT_TYPE|AT_MODE|AT_UID|AT_GID, /* va_mask */ 117 VBLK, /* va_type */ 118 S_IFBLK | SDEV_DEVMODE_DEFAULT, /* va_mode */ 119 SDEV_UID_DEFAULT, /* va_uid */ 120 SDEV_GID_DEFAULT, /* va_gid */ 121 0, /* va_fsid */ 122 0, /* va_nodeid */ 123 0, /* va_nlink */ 124 0, /* va_size */ 125 0, /* va_atime */ 126 0, /* va_mtime */ 127 0, /* va_ctime */ 128 0, /* va_rdev */ 129 0, /* va_blksize */ 130 0, /* va_nblocks */ 131 0 /* va_vcode */ 132 }; 133 134 vattr_t sdev_vattr_chr = { 135 AT_TYPE|AT_MODE|AT_UID|AT_GID, /* va_mask */ 136 VCHR, /* va_type */ 137 S_IFCHR | SDEV_DEVMODE_DEFAULT, /* va_mode */ 138 SDEV_UID_DEFAULT, /* va_uid */ 139 SDEV_GID_DEFAULT, /* va_gid */ 140 0, /* va_fsid */ 141 0, /* va_nodeid */ 142 0, /* va_nlink */ 143 0, /* va_size */ 144 0, /* va_atime */ 145 0, /* va_mtime */ 146 0, /* va_ctime */ 147 0, /* va_rdev */ 148 0, /* va_blksize */ 149 0, /* va_nblocks */ 150 0 /* va_vcode */ 151 }; 152 153 kmem_cache_t *sdev_node_cache; /* sdev_node cache */ 154 int devtype; /* fstype */ 155 156 struct devname_ops *devname_ns_ops; /* default name service directory ops */ 157 kmutex_t devname_nsmaps_lock; /* protect devname_nsmaps */ 158 159 /* static */ 160 static struct devname_nsmap *devname_nsmaps = NULL; 161 /* contents from /etc/dev/devname_master */ 162 static int devname_nsmaps_invalidated = 0; /* "devfsadm -m" has run */ 163 164 static struct vnodeops *sdev_get_vop(struct sdev_node *); 165 static void sdev_set_no_nocache(struct sdev_node *); 166 static int sdev_get_moduleops(struct sdev_node *); 167 static void sdev_handle_alloc(struct sdev_node *); 168 static fs_operation_def_t *sdev_merge_vtab(const fs_operation_def_t []); 169 static void sdev_free_vtab(fs_operation_def_t *); 170 171 static void 172 sdev_prof_free(struct sdev_node *dv) 173 { 174 ASSERT(!SDEV_IS_GLOBAL(dv)); 175 if (dv->sdev_prof.dev_name) 176 nvlist_free(dv->sdev_prof.dev_name); 177 if (dv->sdev_prof.dev_map) 178 nvlist_free(dv->sdev_prof.dev_map); 179 if (dv->sdev_prof.dev_symlink) 180 nvlist_free(dv->sdev_prof.dev_symlink); 181 if (dv->sdev_prof.dev_glob_incdir) 182 nvlist_free(dv->sdev_prof.dev_glob_incdir); 183 if (dv->sdev_prof.dev_glob_excdir) 184 nvlist_free(dv->sdev_prof.dev_glob_excdir); 185 bzero(&dv->sdev_prof, sizeof (dv->sdev_prof)); 186 } 187 188 /* 189 * sdev_node cache constructor 190 */ 191 /*ARGSUSED1*/ 192 static int 193 i_sdev_node_ctor(void *buf, void *cfarg, int flag) 194 { 195 struct sdev_node *dv = (struct sdev_node *)buf; 196 struct vnode *vp; 197 198 ASSERT(flag == KM_SLEEP); 199 200 bzero(buf, sizeof (struct sdev_node)); 201 rw_init(&dv->sdev_contents, NULL, RW_DEFAULT, NULL); 202 dv->sdev_vnode = vn_alloc(KM_SLEEP); 203 vp = SDEVTOV(dv); 204 vp->v_data = (caddr_t)dv; 205 return (0); 206 } 207 208 /* sdev_node destructor for kmem cache */ 209 /*ARGSUSED1*/ 210 static void 211 i_sdev_node_dtor(void *buf, void *arg) 212 { 213 struct sdev_node *dv = (struct sdev_node *)buf; 214 struct vnode *vp = SDEVTOV(dv); 215 216 rw_destroy(&dv->sdev_contents); 217 vn_free(vp); 218 } 219 220 /* initialize sdev_node cache */ 221 void 222 sdev_node_cache_init() 223 { 224 int flags = 0; 225 226 #ifdef DEBUG 227 flags = sdev_debug_cache_flags; 228 if (flags) 229 sdcmn_err(("cache debug flags 0x%x\n", flags)); 230 #endif /* DEBUG */ 231 232 ASSERT(sdev_node_cache == NULL); 233 sdev_node_cache = kmem_cache_create("sdev_node_cache", 234 sizeof (struct sdev_node), 0, i_sdev_node_ctor, i_sdev_node_dtor, 235 NULL, NULL, NULL, flags); 236 } 237 238 /* destroy sdev_node cache */ 239 void 240 sdev_node_cache_fini() 241 { 242 ASSERT(sdev_node_cache != NULL); 243 kmem_cache_destroy(sdev_node_cache); 244 sdev_node_cache = NULL; 245 } 246 247 void 248 sdev_set_nodestate(struct sdev_node *dv, sdev_node_state_t state) 249 { 250 ASSERT(dv); 251 ASSERT(RW_WRITE_HELD(&dv->sdev_contents)); 252 dv->sdev_state = state; 253 } 254 255 static void 256 sdev_attrinit(struct sdev_node *dv, vattr_t *vap) 257 { 258 timestruc_t now; 259 260 ASSERT(vap); 261 262 dv->sdev_attr = kmem_zalloc(sizeof (struct vattr), KM_SLEEP); 263 *dv->sdev_attr = *vap; 264 265 dv->sdev_attr->va_mode = MAKEIMODE(vap->va_type, vap->va_mode); 266 267 gethrestime(&now); 268 dv->sdev_attr->va_atime = now; 269 dv->sdev_attr->va_mtime = now; 270 dv->sdev_attr->va_ctime = now; 271 } 272 273 /* alloc and initialize a sdev_node */ 274 int 275 sdev_nodeinit(struct sdev_node *ddv, char *nm, struct sdev_node **newdv, 276 vattr_t *vap) 277 { 278 struct sdev_node *dv = NULL; 279 struct vnode *vp; 280 size_t nmlen, len; 281 devname_handle_t *dhl; 282 283 nmlen = strlen(nm) + 1; 284 if (nmlen > MAXNAMELEN) { 285 sdcmn_err9(("sdev_nodeinit: node name %s" 286 " too long\n", nm)); 287 *newdv = NULL; 288 return (ENAMETOOLONG); 289 } 290 291 dv = kmem_cache_alloc(sdev_node_cache, KM_SLEEP); 292 293 dv->sdev_name = kmem_alloc(nmlen, KM_SLEEP); 294 bcopy(nm, dv->sdev_name, nmlen); 295 dv->sdev_namelen = nmlen - 1; /* '\0' not included */ 296 len = strlen(ddv->sdev_path) + strlen(nm) + 2; 297 dv->sdev_path = kmem_alloc(len, KM_SLEEP); 298 (void) snprintf(dv->sdev_path, len, "%s/%s", ddv->sdev_path, nm); 299 /* overwritten for VLNK nodes */ 300 dv->sdev_symlink = NULL; 301 302 vp = SDEVTOV(dv); 303 vn_reinit(vp); 304 vp->v_vfsp = SDEVTOV(ddv)->v_vfsp; 305 if (vap) 306 vp->v_type = vap->va_type; 307 308 /* 309 * initialized to the parent's vnodeops. 310 * maybe overwriten for a VDIR 311 */ 312 vn_setops(vp, vn_getops(SDEVTOV(ddv))); 313 vn_exists(vp); 314 315 dv->sdev_dotdot = NULL; 316 dv->sdev_dot = NULL; 317 dv->sdev_next = NULL; 318 dv->sdev_attrvp = NULL; 319 if (vap) { 320 sdev_attrinit(dv, vap); 321 } else { 322 dv->sdev_attr = NULL; 323 } 324 325 dv->sdev_ino = sdev_mkino(dv); 326 dv->sdev_nlink = 0; /* updated on insert */ 327 dv->sdev_flags = ddv->sdev_flags; /* inherit from the parent first */ 328 dv->sdev_flags |= SDEV_BUILD; 329 mutex_init(&dv->sdev_lookup_lock, NULL, MUTEX_DEFAULT, NULL); 330 cv_init(&dv->sdev_lookup_cv, NULL, CV_DEFAULT, NULL); 331 if (SDEV_IS_GLOBAL(ddv)) { 332 dv->sdev_flags |= SDEV_GLOBAL; 333 dv->sdev_mapinfo = NULL; 334 dhl = &(dv->sdev_handle); 335 dhl->dh_data = dv; 336 dhl->dh_spec = DEVNAME_NS_NONE; 337 dhl->dh_args = NULL; 338 sdev_set_no_nocache(dv); 339 dv->sdev_gdir_gen = 0; 340 } else { 341 dv->sdev_flags &= ~SDEV_GLOBAL; 342 dv->sdev_origin = NULL; /* set later */ 343 bzero(&dv->sdev_prof, sizeof (dv->sdev_prof)); 344 dv->sdev_ldir_gen = 0; 345 dv->sdev_devtree_gen = 0; 346 } 347 348 rw_enter(&dv->sdev_contents, RW_WRITER); 349 sdev_set_nodestate(dv, SDEV_INIT); 350 rw_exit(&dv->sdev_contents); 351 *newdv = dv; 352 353 return (0); 354 } 355 356 /* 357 * transition a sdev_node into SDEV_READY state 358 */ 359 int 360 sdev_nodeready(struct sdev_node *dv, struct vattr *vap, struct vnode *avp, 361 void *args, struct cred *cred) 362 { 363 int error = 0; 364 struct vnode *vp = SDEVTOV(dv); 365 vtype_t type; 366 367 ASSERT(dv && (dv->sdev_state != SDEV_READY) && vap); 368 369 type = vap->va_type; 370 vp->v_type = type; 371 vp->v_rdev = vap->va_rdev; 372 rw_enter(&dv->sdev_contents, RW_WRITER); 373 if (type == VDIR) { 374 dv->sdev_nlink = 2; 375 dv->sdev_flags &= ~SDEV_PERSIST; 376 dv->sdev_flags &= ~SDEV_DYNAMIC; 377 vn_setops(vp, sdev_get_vop(dv)); /* from internal vtab */ 378 error = sdev_get_moduleops(dv); /* from plug-in module */ 379 ASSERT(dv->sdev_dotdot); 380 ASSERT(SDEVTOV(dv->sdev_dotdot)->v_type == VDIR); 381 vp->v_rdev = SDEVTOV(dv->sdev_dotdot)->v_rdev; 382 } else if (type == VLNK) { 383 ASSERT(args); 384 dv->sdev_nlink = 1; 385 dv->sdev_symlink = i_ddi_strdup((char *)args, KM_SLEEP); 386 } else { 387 dv->sdev_nlink = 1; 388 } 389 390 if (!(SDEV_IS_GLOBAL(dv))) { 391 dv->sdev_origin = (struct sdev_node *)args; 392 dv->sdev_flags &= ~SDEV_PERSIST; 393 } 394 395 /* 396 * shadow node is created here OR 397 * if failed (indicated by dv->sdev_attrvp == NULL), 398 * created later in sdev_setattr 399 */ 400 if (avp) { 401 dv->sdev_attrvp = avp; 402 } else { 403 if (dv->sdev_attr == NULL) 404 sdev_attrinit(dv, vap); 405 else 406 *dv->sdev_attr = *vap; 407 408 if ((SDEV_IS_PERSIST(dv) && (dv->sdev_attrvp == NULL)) || 409 ((SDEVTOV(dv)->v_type == VDIR) && 410 (dv->sdev_attrvp == NULL))) 411 error = sdev_shadow_node(dv, cred); 412 } 413 414 /* transition to READY state */ 415 sdev_set_nodestate(dv, SDEV_READY); 416 sdev_nc_node_exists(dv); 417 rw_exit(&dv->sdev_contents); 418 return (error); 419 } 420 421 /* 422 * setting ZOMBIE state 423 */ 424 static int 425 sdev_nodezombied(struct sdev_node *dv) 426 { 427 rw_enter(&dv->sdev_contents, RW_WRITER); 428 sdev_set_nodestate(dv, SDEV_ZOMBIE); 429 rw_exit(&dv->sdev_contents); 430 return (0); 431 } 432 433 /* 434 * Build the VROOT sdev_node. 435 */ 436 /*ARGSUSED*/ 437 struct sdev_node * 438 sdev_mkroot(struct vfs *vfsp, dev_t devdev, struct vnode *mvp, 439 struct vnode *avp, struct cred *cred) 440 { 441 struct sdev_node *dv; 442 struct vnode *vp; 443 char devdir[] = "/dev"; 444 445 ASSERT(sdev_node_cache != NULL); 446 ASSERT(avp); 447 dv = kmem_cache_alloc(sdev_node_cache, KM_SLEEP); 448 vp = SDEVTOV(dv); 449 vn_reinit(vp); 450 vp->v_flag |= VROOT; 451 vp->v_vfsp = vfsp; 452 vp->v_type = VDIR; 453 vp->v_rdev = devdev; 454 vn_setops(vp, sdev_vnodeops); /* apply the default vnodeops at /dev */ 455 vn_exists(vp); 456 457 if (vfsp->vfs_mntpt) 458 dv->sdev_name = i_ddi_strdup( 459 (char *)refstr_value(vfsp->vfs_mntpt), KM_SLEEP); 460 else 461 /* vfs_mountdev1 set mount point later */ 462 dv->sdev_name = i_ddi_strdup("/dev", KM_SLEEP); 463 dv->sdev_namelen = strlen(dv->sdev_name); /* '\0' not included */ 464 dv->sdev_path = i_ddi_strdup(devdir, KM_SLEEP); 465 dv->sdev_ino = SDEV_ROOTINO; 466 dv->sdev_nlink = 2; /* name + . (no sdev_insert) */ 467 dv->sdev_dotdot = dv; /* .. == self */ 468 dv->sdev_attrvp = avp; 469 dv->sdev_attr = NULL; 470 mutex_init(&dv->sdev_lookup_lock, NULL, MUTEX_DEFAULT, NULL); 471 cv_init(&dv->sdev_lookup_cv, NULL, CV_DEFAULT, NULL); 472 if (strcmp(dv->sdev_name, "/dev") == 0) { 473 mutex_init(&devname_nsmaps_lock, NULL, MUTEX_DEFAULT, NULL); 474 dv->sdev_mapinfo = NULL; 475 dv->sdev_flags = SDEV_BUILD|SDEV_GLOBAL|SDEV_PERSIST; 476 bzero(&dv->sdev_handle, sizeof (dv->sdev_handle)); 477 dv->sdev_gdir_gen = 0; 478 } else { 479 dv->sdev_flags = SDEV_BUILD; 480 dv->sdev_flags &= ~SDEV_PERSIST; 481 bzero(&dv->sdev_prof, sizeof (dv->sdev_prof)); 482 dv->sdev_ldir_gen = 0; 483 dv->sdev_devtree_gen = 0; 484 } 485 486 rw_enter(&dv->sdev_contents, RW_WRITER); 487 sdev_set_nodestate(dv, SDEV_READY); 488 rw_exit(&dv->sdev_contents); 489 sdev_nc_node_exists(dv); 490 return (dv); 491 } 492 493 /* 494 * 1. load the module 495 * 2. modload invokes sdev_module_register, which in turn sets 496 * the dv->sdev_mapinfo->dir_ops 497 * 498 * note: locking order: 499 * dv->sdev_contents -> map->dir_lock 500 */ 501 static int 502 sdev_get_moduleops(struct sdev_node *dv) 503 { 504 int error = 0; 505 struct devname_nsmap *map = NULL; 506 char *module; 507 char *path; 508 int load = 1; 509 510 ASSERT(SDEVTOV(dv)->v_type == VDIR); 511 512 if (devname_nsmaps == NULL) 513 return (0); 514 515 if (!sdev_nsmaps_loaded() && !sdev_nsmaps_reloaded()) 516 return (0); 517 518 519 path = dv->sdev_path; 520 if ((map = sdev_get_nsmap_by_dir(path, 0))) { 521 rw_enter(&map->dir_lock, RW_READER); 522 if (map->dir_invalid) { 523 if (map->dir_module && map->dir_newmodule && 524 (strcmp(map->dir_module, 525 map->dir_newmodule) == 0)) { 526 load = 0; 527 } 528 sdev_replace_nsmap(map, map->dir_newmodule, 529 map->dir_newmap); 530 } 531 532 module = map->dir_module; 533 if (module && load) { 534 sdcmn_err6(("sdev_get_moduleops: " 535 "load module %s", module)); 536 rw_exit(&map->dir_lock); 537 error = modload("devname", module); 538 sdcmn_err6(("sdev_get_moduleops: error %d\n", error)); 539 if (error < 0) { 540 return (-1); 541 } 542 } else if (module == NULL) { 543 /* 544 * loading the module ops for name services 545 */ 546 if (devname_ns_ops == NULL) { 547 sdcmn_err6(( 548 "sdev_get_moduleops: modload default\n")); 549 error = modload("devname", DEVNAME_NSCONFIG); 550 sdcmn_err6(( 551 "sdev_get_moduleops: error %d\n", error)); 552 if (error < 0) { 553 return (-1); 554 } 555 } 556 557 if (!rw_tryupgrade(&map->dir_lock)) { 558 rw_exit(&map->dir_lock); 559 rw_enter(&map->dir_lock, RW_WRITER); 560 } 561 ASSERT(devname_ns_ops); 562 map->dir_ops = devname_ns_ops; 563 rw_exit(&map->dir_lock); 564 } 565 } 566 567 dv->sdev_mapinfo = map; 568 return (0); 569 } 570 571 /* directory dependent vop table */ 572 struct sdev_vop_table { 573 char *vt_name; /* subdirectory name */ 574 const fs_operation_def_t *vt_service; /* vnodeops table */ 575 struct vnodeops *vt_vops; /* constructed vop */ 576 struct vnodeops **vt_global_vops; /* global container for vop */ 577 int (*vt_vtor)(struct sdev_node *); /* validate sdev_node */ 578 int vt_flags; 579 }; 580 581 /* 582 * A nice improvement would be to provide a plug-in mechanism 583 * for this table instead of a const table. 584 */ 585 static struct sdev_vop_table vtab[] = 586 { 587 { "pts", devpts_vnodeops_tbl, NULL, &devpts_vnodeops, devpts_validate, 588 SDEV_DYNAMIC | SDEV_VTOR }, 589 590 { "zcons", NULL, NULL, NULL, NULL, SDEV_NO_NCACHE }, 591 592 { NULL, NULL, NULL, NULL, NULL, 0} 593 }; 594 595 596 /* 597 * sets a directory's vnodeops if the directory is in the vtab; 598 */ 599 static struct vnodeops * 600 sdev_get_vop(struct sdev_node *dv) 601 { 602 int i; 603 char *path; 604 605 path = dv->sdev_path; 606 ASSERT(path); 607 608 /* gets the relative path to /dev/ */ 609 path += 5; 610 611 /* gets the vtab entry if matches */ 612 for (i = 0; vtab[i].vt_name; i++) { 613 if (strcmp(vtab[i].vt_name, path) != 0) 614 continue; 615 dv->sdev_flags |= vtab[i].vt_flags; 616 617 if (vtab[i].vt_vops) { 618 if (vtab[i].vt_global_vops) 619 *(vtab[i].vt_global_vops) = vtab[i].vt_vops; 620 return (vtab[i].vt_vops); 621 } 622 623 if (vtab[i].vt_service) { 624 fs_operation_def_t *templ; 625 templ = sdev_merge_vtab(vtab[i].vt_service); 626 if (vn_make_ops(vtab[i].vt_name, 627 (const fs_operation_def_t *)templ, 628 &vtab[i].vt_vops) != 0) { 629 cmn_err(CE_PANIC, "%s: malformed vnode ops\n", 630 vtab[i].vt_name); 631 /*NOTREACHED*/ 632 } 633 if (vtab[i].vt_global_vops) { 634 *(vtab[i].vt_global_vops) = vtab[i].vt_vops; 635 } 636 sdev_free_vtab(templ); 637 return (vtab[i].vt_vops); 638 } 639 return (sdev_vnodeops); 640 } 641 642 /* child inherits the persistence of the parent */ 643 if (SDEV_IS_PERSIST(dv->sdev_dotdot)) 644 dv->sdev_flags |= SDEV_PERSIST; 645 646 return (sdev_vnodeops); 647 } 648 649 static void 650 sdev_set_no_nocache(struct sdev_node *dv) 651 { 652 int i; 653 char *path; 654 655 ASSERT(dv->sdev_path); 656 path = dv->sdev_path + strlen("/dev/"); 657 658 for (i = 0; vtab[i].vt_name; i++) { 659 if (strcmp(vtab[i].vt_name, path) == 0) { 660 if (vtab[i].vt_flags & SDEV_NO_NCACHE) 661 dv->sdev_flags |= SDEV_NO_NCACHE; 662 break; 663 } 664 } 665 } 666 667 void * 668 sdev_get_vtor(struct sdev_node *dv) 669 { 670 int i; 671 672 for (i = 0; vtab[i].vt_name; i++) { 673 if (strcmp(vtab[i].vt_name, dv->sdev_name) != 0) 674 continue; 675 return ((void *)vtab[i].vt_vtor); 676 } 677 return (NULL); 678 } 679 680 /* 681 * Build the base root inode 682 */ 683 ino_t 684 sdev_mkino(struct sdev_node *dv) 685 { 686 ino_t ino; 687 688 /* 689 * for now, follow the lead of tmpfs here 690 * need to someday understand the requirements here 691 */ 692 ino = (ino_t)(uint32_t)((uintptr_t)dv >> 3); 693 ino += SDEV_ROOTINO + 1; 694 695 return (ino); 696 } 697 698 static int 699 sdev_getlink(struct vnode *linkvp, char **link) 700 { 701 int err; 702 char *buf; 703 struct uio uio = {0}; 704 struct iovec iov = {0}; 705 706 if (linkvp == NULL) 707 return (ENOENT); 708 ASSERT(linkvp->v_type == VLNK); 709 710 buf = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 711 iov.iov_base = buf; 712 iov.iov_len = MAXPATHLEN; 713 uio.uio_iov = &iov; 714 uio.uio_iovcnt = 1; 715 uio.uio_resid = MAXPATHLEN; 716 uio.uio_segflg = UIO_SYSSPACE; 717 uio.uio_llimit = MAXOFFSET_T; 718 719 err = VOP_READLINK(linkvp, &uio, kcred); 720 if (err) { 721 cmn_err(CE_WARN, "readlink %s failed in dev\n", buf); 722 kmem_free(buf, MAXPATHLEN); 723 return (ENOENT); 724 } 725 726 /* mission complete */ 727 *link = i_ddi_strdup(buf, KM_SLEEP); 728 kmem_free(buf, MAXPATHLEN); 729 return (0); 730 } 731 732 /* 733 * A convenient wrapper to get the devfs node vnode for a device 734 * minor functionality: readlink() of a /dev symlink 735 * Place the link into dv->sdev_symlink 736 */ 737 static int 738 sdev_follow_link(struct sdev_node *dv) 739 { 740 int err; 741 struct vnode *linkvp; 742 char *link = NULL; 743 744 linkvp = SDEVTOV(dv); 745 if (linkvp == NULL) 746 return (ENOENT); 747 ASSERT(linkvp->v_type == VLNK); 748 err = sdev_getlink(linkvp, &link); 749 if (err) { 750 (void) sdev_nodezombied(dv); 751 dv->sdev_symlink = NULL; 752 return (ENOENT); 753 } 754 755 ASSERT(link != NULL); 756 dv->sdev_symlink = link; 757 return (0); 758 } 759 760 static int 761 sdev_node_check(struct sdev_node *dv, struct vattr *nvap, void *nargs) 762 { 763 vtype_t otype = SDEVTOV(dv)->v_type; 764 765 /* 766 * existing sdev_node has a different type. 767 */ 768 if (otype != nvap->va_type) { 769 sdcmn_err9(("sdev_node_check: existing node " 770 " %s type %d does not match new node type %d\n", 771 dv->sdev_name, otype, nvap->va_type)); 772 return (EEXIST); 773 } 774 775 /* 776 * For a symlink, the target should be the same. 777 */ 778 if (otype == VLNK) { 779 ASSERT(nargs != NULL); 780 ASSERT(dv->sdev_symlink != NULL); 781 if (strcmp(dv->sdev_symlink, (char *)nargs) != 0) { 782 sdcmn_err9(("sdev_node_check: existing node " 783 " %s has different symlink %s as new node " 784 " %s\n", dv->sdev_name, dv->sdev_symlink, 785 (char *)nargs)); 786 return (EEXIST); 787 } 788 } 789 790 return (0); 791 } 792 793 /* 794 * sdev_mknode - a wrapper for sdev_nodeinit(), sdev_nodeready() 795 * 796 * arguments: 797 * - ddv (parent) 798 * - nm (child name) 799 * - newdv (sdev_node for nm is returned here) 800 * - vap (vattr for the node to be created, va_type should be set. 801 * the defaults should be used if unknown) 802 * - cred 803 * - args 804 * . tnm (for VLNK) 805 * . global sdev_node (for !SDEV_GLOBAL) 806 * - state: SDEV_INIT, SDEV_READY 807 * 808 * only ddv, nm, newddv, vap, cred are required for sdev_mknode(SDEV_INIT) 809 * 810 * NOTE: directory contents writers lock needs to be held before 811 * calling this routine. 812 */ 813 int 814 sdev_mknode(struct sdev_node *ddv, char *nm, struct sdev_node **newdv, 815 struct vattr *vap, struct vnode *avp, void *args, struct cred *cred, 816 sdev_node_state_t state) 817 { 818 int error = 0; 819 sdev_node_state_t node_state; 820 struct sdev_node *dv = NULL; 821 822 ASSERT(state != SDEV_ZOMBIE); 823 ASSERT(RW_WRITE_HELD(&ddv->sdev_contents)); 824 825 if (*newdv) { 826 dv = *newdv; 827 } else { 828 /* allocate and initialize a sdev_node */ 829 if (ddv->sdev_state == SDEV_ZOMBIE) { 830 sdcmn_err9(("sdev_mknode: parent %s ZOMBIEd\n", 831 ddv->sdev_path)); 832 return (ENOENT); 833 } 834 835 error = sdev_nodeinit(ddv, nm, &dv, vap); 836 if (error != 0) { 837 sdcmn_err9(("sdev_mknode: error %d," 838 " name %s can not be initialized\n", 839 error, nm)); 840 return (ENOENT); 841 } 842 ASSERT(dv); 843 844 /* insert into the directory cache */ 845 error = sdev_cache_update(ddv, &dv, nm, SDEV_CACHE_ADD); 846 if (error) { 847 sdcmn_err9(("sdev_mknode: node %s can not" 848 " be added into directory cache\n", nm)); 849 return (ENOENT); 850 } 851 } 852 853 ASSERT(dv); 854 node_state = dv->sdev_state; 855 ASSERT(node_state != SDEV_ZOMBIE); 856 857 if (state == SDEV_READY) { 858 switch (node_state) { 859 case SDEV_INIT: 860 error = sdev_nodeready(dv, vap, avp, args, cred); 861 /* 862 * masking the errors with ENOENT 863 */ 864 if (error) { 865 sdcmn_err9(("sdev_mknode: node %s can NOT" 866 " be transitioned into READY state, " 867 "error %d\n", nm, error)); 868 error = ENOENT; 869 } 870 break; 871 case SDEV_READY: 872 /* 873 * Do some sanity checking to make sure 874 * the existing sdev_node is what has been 875 * asked for. 876 */ 877 error = sdev_node_check(dv, vap, args); 878 break; 879 default: 880 break; 881 } 882 } 883 884 if (!error) { 885 *newdv = dv; 886 ASSERT((*newdv)->sdev_state != SDEV_ZOMBIE); 887 } else { 888 SDEV_SIMPLE_RELE(dv); 889 *newdv = NULL; 890 } 891 892 return (error); 893 } 894 895 /* 896 * convenient wrapper to change vp's ATIME, CTIME and ATIME 897 */ 898 void 899 sdev_update_timestamps(struct vnode *vp, cred_t *cred, uint_t mask) 900 { 901 struct vattr attr; 902 timestruc_t now; 903 int err; 904 905 ASSERT(vp); 906 gethrestime(&now); 907 if (mask & AT_CTIME) 908 attr.va_ctime = now; 909 if (mask & AT_MTIME) 910 attr.va_mtime = now; 911 if (mask & AT_ATIME) 912 attr.va_atime = now; 913 914 attr.va_mask = (mask & AT_TIMES); 915 err = VOP_SETATTR(vp, &attr, 0, cred, NULL); 916 if (err && (err != EROFS)) { 917 sdcmn_err(("update timestamps error %d\n", err)); 918 } 919 } 920 921 /* 922 * the backing store vnode is released here 923 */ 924 /*ARGSUSED1*/ 925 void 926 sdev_nodedestroy(struct sdev_node *dv, uint_t flags) 927 { 928 /* no references */ 929 ASSERT(dv->sdev_nlink == 0); 930 931 if (dv->sdev_attrvp != NULLVP) { 932 VN_RELE(dv->sdev_attrvp); 933 /* 934 * reset the attrvp so that no more 935 * references can be made on this already 936 * vn_rele() vnode 937 */ 938 dv->sdev_attrvp = NULLVP; 939 } 940 941 if (dv->sdev_attr != NULL) { 942 kmem_free(dv->sdev_attr, sizeof (struct vattr)); 943 dv->sdev_attr = NULL; 944 } 945 946 if (dv->sdev_name != NULL) { 947 kmem_free(dv->sdev_name, dv->sdev_namelen + 1); 948 dv->sdev_name = NULL; 949 } 950 951 if (dv->sdev_symlink != NULL) { 952 kmem_free(dv->sdev_symlink, strlen(dv->sdev_symlink) + 1); 953 dv->sdev_symlink = NULL; 954 } 955 956 if (dv->sdev_path) { 957 kmem_free(dv->sdev_path, strlen(dv->sdev_path) + 1); 958 dv->sdev_path = NULL; 959 } 960 961 if (!SDEV_IS_GLOBAL(dv)) 962 sdev_prof_free(dv); 963 964 mutex_destroy(&dv->sdev_lookup_lock); 965 cv_destroy(&dv->sdev_lookup_cv); 966 967 /* return node to initial state as per constructor */ 968 (void) memset((void *)&dv->sdev_instance_data, 0, 969 sizeof (dv->sdev_instance_data)); 970 vn_invalid(SDEVTOV(dv)); 971 kmem_cache_free(sdev_node_cache, dv); 972 } 973 974 /* 975 * DIRECTORY CACHE lookup 976 */ 977 struct sdev_node * 978 sdev_findbyname(struct sdev_node *ddv, char *nm) 979 { 980 struct sdev_node *dv; 981 size_t nmlen = strlen(nm); 982 983 ASSERT(RW_LOCK_HELD(&ddv->sdev_contents)); 984 for (dv = ddv->sdev_dot; dv; dv = dv->sdev_next) { 985 if (dv->sdev_namelen != nmlen) { 986 continue; 987 } 988 989 /* 990 * Can't lookup stale nodes 991 */ 992 if (dv->sdev_flags & SDEV_STALE) { 993 sdcmn_err9(( 994 "sdev_findbyname: skipped stale node: %s\n", 995 dv->sdev_name)); 996 continue; 997 } 998 999 if (strcmp(dv->sdev_name, nm) == 0) { 1000 SDEV_HOLD(dv); 1001 return (dv); 1002 } 1003 } 1004 return (NULL); 1005 } 1006 1007 /* 1008 * Inserts a new sdev_node in a parent directory 1009 */ 1010 void 1011 sdev_direnter(struct sdev_node *ddv, struct sdev_node *dv) 1012 { 1013 ASSERT(RW_WRITE_HELD(&ddv->sdev_contents)); 1014 ASSERT(SDEVTOV(ddv)->v_type == VDIR); 1015 ASSERT(ddv->sdev_nlink >= 2); 1016 ASSERT(dv->sdev_nlink == 0); 1017 1018 dv->sdev_dotdot = ddv; 1019 dv->sdev_next = ddv->sdev_dot; 1020 ddv->sdev_dot = dv; 1021 ddv->sdev_nlink++; 1022 } 1023 1024 /* 1025 * The following check is needed because while sdev_nodes are linked 1026 * in SDEV_INIT state, they have their link counts incremented only 1027 * in SDEV_READY state. 1028 */ 1029 static void 1030 decr_link(struct sdev_node *dv) 1031 { 1032 if (dv->sdev_state != SDEV_INIT) 1033 dv->sdev_nlink--; 1034 else 1035 ASSERT(dv->sdev_nlink == 0); 1036 } 1037 1038 /* 1039 * Delete an existing dv from directory cache 1040 * 1041 * In the case of a node is still held by non-zero reference count, 1042 * the node is put into ZOMBIE state. Once the reference count 1043 * reaches "0", the node is unlinked and destroyed, 1044 * in sdev_inactive(). 1045 */ 1046 static int 1047 sdev_dirdelete(struct sdev_node *ddv, struct sdev_node *dv) 1048 { 1049 struct sdev_node *idv; 1050 struct sdev_node *prev = NULL; 1051 struct vnode *vp; 1052 1053 ASSERT(RW_WRITE_HELD(&ddv->sdev_contents)); 1054 1055 vp = SDEVTOV(dv); 1056 mutex_enter(&vp->v_lock); 1057 1058 /* dv is held still */ 1059 if (vp->v_count > 1) { 1060 rw_enter(&dv->sdev_contents, RW_WRITER); 1061 if (dv->sdev_state == SDEV_READY) { 1062 sdcmn_err9(( 1063 "sdev_delete: node %s busy with count %d\n", 1064 dv->sdev_name, vp->v_count)); 1065 dv->sdev_state = SDEV_ZOMBIE; 1066 } 1067 rw_exit(&dv->sdev_contents); 1068 --vp->v_count; 1069 mutex_exit(&vp->v_lock); 1070 return (EBUSY); 1071 } 1072 ASSERT(vp->v_count == 1); 1073 1074 /* unlink from the memory cache */ 1075 ddv->sdev_nlink--; /* .. to above */ 1076 if (vp->v_type == VDIR) { 1077 decr_link(dv); /* . to self */ 1078 } 1079 1080 for (idv = ddv->sdev_dot; idv && idv != dv; 1081 prev = idv, idv = idv->sdev_next) 1082 ; 1083 ASSERT(idv == dv); /* node to be deleted must exist */ 1084 if (prev == NULL) 1085 ddv->sdev_dot = dv->sdev_next; 1086 else 1087 prev->sdev_next = dv->sdev_next; 1088 dv->sdev_next = NULL; 1089 decr_link(dv); /* name, back to zero */ 1090 vp->v_count--; 1091 mutex_exit(&vp->v_lock); 1092 1093 /* destroy the node */ 1094 sdev_nodedestroy(dv, 0); 1095 return (0); 1096 } 1097 1098 /* 1099 * check if the source is in the path of the target 1100 * 1101 * source and target are different 1102 */ 1103 /*ARGSUSED2*/ 1104 static int 1105 sdev_checkpath(struct sdev_node *sdv, struct sdev_node *tdv, struct cred *cred) 1106 { 1107 int error = 0; 1108 struct sdev_node *dotdot, *dir; 1109 1110 dotdot = tdv->sdev_dotdot; 1111 ASSERT(dotdot); 1112 1113 /* fs root */ 1114 if (dotdot == tdv) { 1115 return (0); 1116 } 1117 1118 for (;;) { 1119 /* 1120 * avoid error cases like 1121 * mv a a/b 1122 * mv a a/b/c 1123 * etc. 1124 */ 1125 if (dotdot == sdv) { 1126 error = EINVAL; 1127 break; 1128 } 1129 1130 dir = dotdot; 1131 dotdot = dir->sdev_dotdot; 1132 1133 /* done checking because root is reached */ 1134 if (dir == dotdot) { 1135 break; 1136 } 1137 } 1138 return (error); 1139 } 1140 1141 int 1142 sdev_rnmnode(struct sdev_node *oddv, struct sdev_node *odv, 1143 struct sdev_node *nddv, struct sdev_node **ndvp, char *nnm, 1144 struct cred *cred) 1145 { 1146 int error = 0; 1147 struct vnode *ovp = SDEVTOV(odv); 1148 struct vnode *nvp; 1149 struct vattr vattr; 1150 int doingdir = (ovp->v_type == VDIR); 1151 char *link = NULL; 1152 int samedir = (oddv == nddv) ? 1 : 0; 1153 int bkstore = 0; 1154 struct sdev_node *idv = NULL; 1155 struct sdev_node *ndv = NULL; 1156 timestruc_t now; 1157 1158 vattr.va_mask = AT_MODE|AT_UID|AT_GID; 1159 error = VOP_GETATTR(ovp, &vattr, 0, cred); 1160 if (error) 1161 return (error); 1162 1163 if (!samedir) 1164 rw_enter(&oddv->sdev_contents, RW_WRITER); 1165 rw_enter(&nddv->sdev_contents, RW_WRITER); 1166 1167 /* 1168 * the source may have been deleted by another thread before 1169 * we gets here. 1170 */ 1171 if (odv->sdev_state != SDEV_READY) { 1172 error = ENOENT; 1173 goto err_out; 1174 } 1175 1176 if (doingdir && (odv == nddv)) { 1177 error = EINVAL; 1178 goto err_out; 1179 } 1180 1181 /* 1182 * If renaming a directory, and the parents are different (".." must be 1183 * changed) then the source dir must not be in the dir hierarchy above 1184 * the target since it would orphan everything below the source dir. 1185 */ 1186 if (doingdir && (oddv != nddv)) { 1187 error = sdev_checkpath(odv, nddv, cred); 1188 if (error) 1189 goto err_out; 1190 } 1191 1192 /* destination existing */ 1193 if (*ndvp) { 1194 nvp = SDEVTOV(*ndvp); 1195 ASSERT(nvp); 1196 1197 /* handling renaming to itself */ 1198 if (odv == *ndvp) { 1199 error = 0; 1200 goto err_out; 1201 } 1202 1203 if (nvp->v_type == VDIR) { 1204 if (!doingdir) { 1205 error = EISDIR; 1206 goto err_out; 1207 } 1208 1209 if (vn_vfswlock(nvp)) { 1210 error = EBUSY; 1211 goto err_out; 1212 } 1213 1214 if (vn_mountedvfs(nvp) != NULL) { 1215 vn_vfsunlock(nvp); 1216 error = EBUSY; 1217 goto err_out; 1218 } 1219 1220 /* in case dir1 exists in dir2 and "mv dir1 dir2" */ 1221 if ((*ndvp)->sdev_nlink > 2) { 1222 vn_vfsunlock(nvp); 1223 error = EEXIST; 1224 goto err_out; 1225 } 1226 vn_vfsunlock(nvp); 1227 1228 (void) sdev_dirdelete(nddv, *ndvp); 1229 *ndvp = NULL; 1230 error = VOP_RMDIR(nddv->sdev_attrvp, nnm, 1231 nddv->sdev_attrvp, cred); 1232 if (error) 1233 goto err_out; 1234 } else { 1235 if (doingdir) { 1236 error = ENOTDIR; 1237 goto err_out; 1238 } 1239 1240 if (SDEV_IS_PERSIST((*ndvp))) { 1241 bkstore = 1; 1242 } 1243 1244 /* 1245 * get rid of the node from the directory cache 1246 * note, in case EBUSY is returned, the ZOMBIE 1247 * node is taken care in sdev_mknode. 1248 */ 1249 (void) sdev_dirdelete(nddv, *ndvp); 1250 *ndvp = NULL; 1251 if (bkstore) { 1252 error = VOP_REMOVE(nddv->sdev_attrvp, 1253 nnm, cred); 1254 if (error) 1255 goto err_out; 1256 } 1257 } 1258 } 1259 1260 /* fix the source for a symlink */ 1261 if (vattr.va_type == VLNK) { 1262 if (odv->sdev_symlink == NULL) { 1263 error = sdev_follow_link(odv); 1264 if (error) { 1265 error = ENOENT; 1266 goto err_out; 1267 } 1268 } 1269 ASSERT(odv->sdev_symlink); 1270 link = i_ddi_strdup(odv->sdev_symlink, KM_SLEEP); 1271 } 1272 1273 /* 1274 * make a fresh node from the source attrs 1275 */ 1276 ASSERT(RW_WRITE_HELD(&nddv->sdev_contents)); 1277 error = sdev_mknode(nddv, nnm, ndvp, &vattr, 1278 NULL, (void *)link, cred, SDEV_READY); 1279 1280 if (link) 1281 kmem_free(link, strlen(link) + 1); 1282 1283 if (error) 1284 goto err_out; 1285 ASSERT(*ndvp); 1286 ASSERT((*ndvp)->sdev_state == SDEV_READY); 1287 1288 /* move dir contents */ 1289 if (doingdir) { 1290 for (idv = odv->sdev_dot; idv; idv = idv->sdev_next) { 1291 error = sdev_rnmnode(odv, idv, 1292 (struct sdev_node *)(*ndvp), &ndv, 1293 idv->sdev_name, cred); 1294 1295 if (error) 1296 goto err_out; 1297 ndv = NULL; 1298 } 1299 1300 } 1301 1302 if ((*ndvp)->sdev_attrvp) { 1303 sdev_update_timestamps((*ndvp)->sdev_attrvp, kcred, 1304 AT_CTIME|AT_ATIME); 1305 } else { 1306 ASSERT((*ndvp)->sdev_attr); 1307 gethrestime(&now); 1308 (*ndvp)->sdev_attr->va_ctime = now; 1309 (*ndvp)->sdev_attr->va_atime = now; 1310 } 1311 1312 if (nddv->sdev_attrvp) { 1313 sdev_update_timestamps(nddv->sdev_attrvp, kcred, 1314 AT_MTIME|AT_ATIME); 1315 } else { 1316 ASSERT(nddv->sdev_attr); 1317 gethrestime(&now); 1318 nddv->sdev_attr->va_mtime = now; 1319 nddv->sdev_attr->va_atime = now; 1320 } 1321 rw_exit(&nddv->sdev_contents); 1322 if (!samedir) 1323 rw_exit(&oddv->sdev_contents); 1324 1325 SDEV_RELE(*ndvp); 1326 return (error); 1327 1328 err_out: 1329 rw_exit(&nddv->sdev_contents); 1330 if (!samedir) 1331 rw_exit(&oddv->sdev_contents); 1332 return (error); 1333 } 1334 1335 /* 1336 * Merge sdev_node specific information into an attribute structure. 1337 * 1338 * note: sdev_node is not locked here 1339 */ 1340 void 1341 sdev_vattr_merge(struct sdev_node *dv, struct vattr *vap) 1342 { 1343 struct vnode *vp = SDEVTOV(dv); 1344 1345 vap->va_nlink = dv->sdev_nlink; 1346 vap->va_nodeid = dv->sdev_ino; 1347 vap->va_fsid = SDEVTOV(dv->sdev_dotdot)->v_rdev; 1348 vap->va_type = vp->v_type; 1349 1350 if (vp->v_type == VDIR) { 1351 vap->va_rdev = 0; 1352 vap->va_fsid = vp->v_rdev; 1353 } else if (vp->v_type == VLNK) { 1354 vap->va_rdev = 0; 1355 vap->va_mode &= ~S_IFMT; 1356 vap->va_mode |= S_IFLNK; 1357 } else if ((vp->v_type == VCHR) || (vp->v_type == VBLK)) { 1358 vap->va_rdev = vp->v_rdev; 1359 vap->va_mode &= ~S_IFMT; 1360 if (vap->va_type == VCHR) 1361 vap->va_mode |= S_IFCHR; 1362 else 1363 vap->va_mode |= S_IFBLK; 1364 } else { 1365 vap->va_rdev = 0; 1366 } 1367 } 1368 1369 static struct vattr * 1370 sdev_getdefault_attr(enum vtype type) 1371 { 1372 if (type == VDIR) 1373 return (&sdev_vattr_dir); 1374 else if (type == VCHR) 1375 return (&sdev_vattr_chr); 1376 else if (type == VBLK) 1377 return (&sdev_vattr_blk); 1378 else if (type == VLNK) 1379 return (&sdev_vattr_lnk); 1380 else 1381 return (NULL); 1382 } 1383 int 1384 sdev_to_vp(struct sdev_node *dv, struct vnode **vpp) 1385 { 1386 int rv = 0; 1387 struct vnode *vp = SDEVTOV(dv); 1388 1389 switch (vp->v_type) { 1390 case VCHR: 1391 case VBLK: 1392 /* 1393 * If vnode is a device, return special vnode instead 1394 * (though it knows all about -us- via sp->s_realvp) 1395 */ 1396 *vpp = specvp(vp, vp->v_rdev, vp->v_type, kcred); 1397 VN_RELE(vp); 1398 if (*vpp == NULLVP) 1399 rv = ENOSYS; 1400 break; 1401 default: /* most types are returned as is */ 1402 *vpp = vp; 1403 break; 1404 } 1405 return (rv); 1406 } 1407 1408 /* 1409 * loopback into sdev_lookup() 1410 */ 1411 static struct vnode * 1412 devname_find_by_devpath(char *devpath, struct vattr *vattr) 1413 { 1414 int error = 0; 1415 struct vnode *vp; 1416 1417 error = lookupname(devpath, UIO_SYSSPACE, NO_FOLLOW, NULLVPP, &vp); 1418 if (error) { 1419 return (NULL); 1420 } 1421 1422 if (vattr) 1423 (void) VOP_GETATTR(vp, vattr, 0, kcred); 1424 return (vp); 1425 } 1426 1427 /* 1428 * the junction between devname and devfs 1429 */ 1430 static struct vnode * 1431 devname_configure_by_path(char *physpath, struct vattr *vattr) 1432 { 1433 int error = 0; 1434 struct vnode *vp; 1435 1436 ASSERT(strncmp(physpath, "/devices/", sizeof ("/devices/") - 1) 1437 == 0); 1438 1439 error = devfs_lookupname(physpath + sizeof ("/devices/") - 1, 1440 NULLVPP, &vp); 1441 if (error != 0) { 1442 if (error == ENODEV) { 1443 cmn_err(CE_CONT, "%s: not found (line %d)\n", 1444 physpath, __LINE__); 1445 } 1446 1447 return (NULL); 1448 } 1449 1450 if (vattr) 1451 (void) VOP_GETATTR(vp, vattr, 0, kcred); 1452 return (vp); 1453 } 1454 1455 /* 1456 * junction between devname and root file system, e.g. ufs 1457 */ 1458 int 1459 devname_backstore_lookup(struct sdev_node *ddv, char *nm, struct vnode **rvp) 1460 { 1461 struct vnode *rdvp = ddv->sdev_attrvp; 1462 int rval = 0; 1463 1464 ASSERT(rdvp); 1465 1466 rval = VOP_LOOKUP(rdvp, nm, rvp, NULL, 0, NULL, kcred); 1467 return (rval); 1468 } 1469 1470 static int 1471 sdev_filldir_from_store(struct sdev_node *ddv, int dlen, struct cred *cred) 1472 { 1473 struct sdev_node *dv = NULL; 1474 char *nm; 1475 struct vnode *dirvp; 1476 int error; 1477 vnode_t *vp; 1478 int eof; 1479 struct iovec iov; 1480 struct uio uio; 1481 struct dirent64 *dp; 1482 dirent64_t *dbuf; 1483 size_t dbuflen; 1484 struct vattr vattr; 1485 char *link = NULL; 1486 1487 if (ddv->sdev_attrvp == NULL) 1488 return (0); 1489 if (!(ddv->sdev_flags & SDEV_BUILD)) 1490 return (0); 1491 1492 dirvp = ddv->sdev_attrvp; 1493 VN_HOLD(dirvp); 1494 dbuf = kmem_zalloc(dlen, KM_SLEEP); 1495 1496 uio.uio_iov = &iov; 1497 uio.uio_iovcnt = 1; 1498 uio.uio_segflg = UIO_SYSSPACE; 1499 uio.uio_fmode = 0; 1500 uio.uio_extflg = UIO_COPY_CACHED; 1501 uio.uio_loffset = 0; 1502 uio.uio_llimit = MAXOFFSET_T; 1503 1504 eof = 0; 1505 error = 0; 1506 while (!error && !eof) { 1507 uio.uio_resid = dlen; 1508 iov.iov_base = (char *)dbuf; 1509 iov.iov_len = dlen; 1510 (void) VOP_RWLOCK(dirvp, V_WRITELOCK_FALSE, NULL); 1511 error = VOP_READDIR(dirvp, &uio, kcred, &eof); 1512 VOP_RWUNLOCK(dirvp, V_WRITELOCK_FALSE, NULL); 1513 1514 dbuflen = dlen - uio.uio_resid; 1515 if (error || dbuflen == 0) 1516 break; 1517 1518 if (!(ddv->sdev_flags & SDEV_BUILD)) { 1519 error = 0; 1520 break; 1521 } 1522 1523 for (dp = dbuf; ((intptr_t)dp < 1524 (intptr_t)dbuf + dbuflen); 1525 dp = (dirent64_t *)((intptr_t)dp + dp->d_reclen)) { 1526 nm = dp->d_name; 1527 1528 if (strcmp(nm, ".") == 0 || 1529 strcmp(nm, "..") == 0) 1530 continue; 1531 1532 vp = NULLVP; 1533 dv = sdev_cache_lookup(ddv, nm); 1534 if (dv) { 1535 if (dv->sdev_state != SDEV_ZOMBIE) { 1536 SDEV_SIMPLE_RELE(dv); 1537 } else { 1538 /* 1539 * A ZOMBIE node may not have been 1540 * cleaned up from the backing store, 1541 * bypass this entry in this case, 1542 * and clean it up from the directory 1543 * cache if this is the last call. 1544 */ 1545 (void) sdev_dirdelete(ddv, dv); 1546 } 1547 continue; 1548 } 1549 1550 /* refill the cache if not already */ 1551 error = devname_backstore_lookup(ddv, nm, &vp); 1552 if (error) 1553 continue; 1554 1555 vattr.va_mask = AT_MODE|AT_UID|AT_GID; 1556 error = VOP_GETATTR(vp, &vattr, 0, cred); 1557 if (error) 1558 continue; 1559 1560 if (vattr.va_type == VLNK) { 1561 error = sdev_getlink(vp, &link); 1562 if (error) { 1563 continue; 1564 } 1565 ASSERT(link != NULL); 1566 } 1567 1568 if (!rw_tryupgrade(&ddv->sdev_contents)) { 1569 rw_exit(&ddv->sdev_contents); 1570 rw_enter(&ddv->sdev_contents, RW_WRITER); 1571 } 1572 error = sdev_mknode(ddv, nm, &dv, &vattr, vp, link, 1573 cred, SDEV_READY); 1574 rw_downgrade(&ddv->sdev_contents); 1575 1576 if (link != NULL) { 1577 kmem_free(link, strlen(link) + 1); 1578 link = NULL; 1579 } 1580 1581 if (!error) { 1582 ASSERT(dv); 1583 ASSERT(dv->sdev_state != SDEV_ZOMBIE); 1584 SDEV_SIMPLE_RELE(dv); 1585 } 1586 vp = NULL; 1587 dv = NULL; 1588 } 1589 } 1590 1591 done: 1592 VN_RELE(dirvp); 1593 kmem_free(dbuf, dlen); 1594 1595 return (error); 1596 } 1597 1598 static int 1599 sdev_filldir_dynamic(struct sdev_node *ddv) 1600 { 1601 int error; 1602 int i; 1603 struct vattr *vap; 1604 char *nm = NULL; 1605 struct sdev_node *dv = NULL; 1606 1607 if (!(ddv->sdev_flags & SDEV_BUILD)) { 1608 return (0); 1609 } 1610 1611 ASSERT(RW_READ_HELD(&ddv->sdev_contents)); 1612 if (!rw_tryupgrade(&ddv->sdev_contents)) { 1613 rw_exit(&ddv->sdev_contents); 1614 rw_enter(&ddv->sdev_contents, RW_WRITER); 1615 } 1616 1617 vap = sdev_getdefault_attr(VDIR); 1618 for (i = 0; vtab[i].vt_name != NULL; i++) { 1619 nm = vtab[i].vt_name; 1620 ASSERT(RW_WRITE_HELD(&ddv->sdev_contents)); 1621 error = sdev_mknode(ddv, nm, &dv, vap, NULL, 1622 NULL, kcred, SDEV_READY); 1623 if (error) 1624 continue; 1625 ASSERT(dv); 1626 ASSERT(dv->sdev_state != SDEV_ZOMBIE); 1627 SDEV_SIMPLE_RELE(dv); 1628 dv = NULL; 1629 } 1630 rw_downgrade(&ddv->sdev_contents); 1631 return (0); 1632 } 1633 1634 /* 1635 * Creating a backing store entry based on sdev_attr. 1636 * This is called either as part of node creation in a persistent directory 1637 * or from setattr/setsecattr to persist access attributes across reboot. 1638 */ 1639 int 1640 sdev_shadow_node(struct sdev_node *dv, struct cred *cred) 1641 { 1642 int error = 0; 1643 struct vnode *dvp = SDEVTOV(dv->sdev_dotdot); 1644 struct vnode *rdvp = VTOSDEV(dvp)->sdev_attrvp; 1645 struct vattr *vap = dv->sdev_attr; 1646 char *nm = dv->sdev_name; 1647 struct vnode *tmpvp, **rvp = &tmpvp, *rrvp = NULL; 1648 1649 ASSERT(dv && dv->sdev_name && rdvp); 1650 ASSERT(RW_WRITE_HELD(&dv->sdev_contents) && dv->sdev_attrvp == NULL); 1651 1652 lookup: 1653 /* try to find it in the backing store */ 1654 error = VOP_LOOKUP(rdvp, nm, rvp, NULL, 0, NULL, cred); 1655 if (error == 0) { 1656 if (VOP_REALVP(*rvp, &rrvp) == 0) { 1657 VN_HOLD(rrvp); 1658 VN_RELE(*rvp); 1659 *rvp = rrvp; 1660 } 1661 1662 kmem_free(dv->sdev_attr, sizeof (vattr_t)); 1663 dv->sdev_attr = NULL; 1664 dv->sdev_attrvp = *rvp; 1665 return (0); 1666 } 1667 1668 /* let's try to persist the node */ 1669 gethrestime(&vap->va_atime); 1670 vap->va_mtime = vap->va_atime; 1671 vap->va_ctime = vap->va_atime; 1672 vap->va_mask |= AT_TYPE|AT_MODE; 1673 switch (vap->va_type) { 1674 case VDIR: 1675 error = VOP_MKDIR(rdvp, nm, vap, rvp, cred); 1676 sdcmn_err9(("sdev_shadow_node: mkdir vp %p error %d\n", 1677 (void *)(*rvp), error)); 1678 break; 1679 case VCHR: 1680 case VBLK: 1681 case VREG: 1682 case VDOOR: 1683 error = VOP_CREATE(rdvp, nm, vap, NONEXCL, VREAD|VWRITE, 1684 rvp, cred, 0); 1685 sdcmn_err9(("sdev_shadow_node: create vp %p, error %d\n", 1686 (void *)(*rvp), error)); 1687 if (!error) 1688 VN_RELE(*rvp); 1689 break; 1690 case VLNK: 1691 ASSERT(dv->sdev_symlink); 1692 error = VOP_SYMLINK(rdvp, nm, vap, dv->sdev_symlink, cred); 1693 sdcmn_err9(("sdev_shadow_node: create symlink error %d\n", 1694 error)); 1695 break; 1696 default: 1697 cmn_err(CE_PANIC, "dev: %s: sdev_shadow_node " 1698 "create\n", nm); 1699 /*NOTREACHED*/ 1700 } 1701 1702 /* go back to lookup to factor out spec node and set attrvp */ 1703 if (error == 0) 1704 goto lookup; 1705 1706 return (error); 1707 } 1708 1709 static int 1710 sdev_cache_add(struct sdev_node *ddv, struct sdev_node **dv, char *nm) 1711 { 1712 int error = 0; 1713 struct sdev_node *dup = NULL; 1714 1715 ASSERT(RW_WRITE_HELD(&ddv->sdev_contents)); 1716 if ((dup = sdev_findbyname(ddv, nm)) == NULL) { 1717 sdev_direnter(ddv, *dv); 1718 } else { 1719 if (dup->sdev_state == SDEV_ZOMBIE) { 1720 error = sdev_dirdelete(ddv, dup); 1721 /* 1722 * The ZOMBIE node is still hanging 1723 * around with more than one reference counts. 1724 * Fail the new node creation so that 1725 * the directory cache won't have 1726 * duplicate entries for the same named node 1727 */ 1728 if (error == EBUSY) { 1729 SDEV_SIMPLE_RELE(*dv); 1730 sdev_nodedestroy(*dv, 0); 1731 *dv = NULL; 1732 return (error); 1733 } 1734 sdev_direnter(ddv, *dv); 1735 } else { 1736 ASSERT((*dv)->sdev_state != SDEV_ZOMBIE); 1737 SDEV_SIMPLE_RELE(*dv); 1738 sdev_nodedestroy(*dv, 0); 1739 *dv = dup; 1740 } 1741 } 1742 1743 return (0); 1744 } 1745 1746 static int 1747 sdev_cache_delete(struct sdev_node *ddv, struct sdev_node **dv) 1748 { 1749 ASSERT(RW_WRITE_HELD(&ddv->sdev_contents)); 1750 return (sdev_dirdelete(ddv, *dv)); 1751 } 1752 1753 /* 1754 * update the in-core directory cache 1755 */ 1756 int 1757 sdev_cache_update(struct sdev_node *ddv, struct sdev_node **dv, char *nm, 1758 sdev_cache_ops_t ops) 1759 { 1760 int error = 0; 1761 1762 ASSERT((SDEV_HELD(*dv))); 1763 1764 ASSERT(RW_WRITE_HELD(&ddv->sdev_contents)); 1765 switch (ops) { 1766 case SDEV_CACHE_ADD: 1767 error = sdev_cache_add(ddv, dv, nm); 1768 break; 1769 case SDEV_CACHE_DELETE: 1770 error = sdev_cache_delete(ddv, dv); 1771 break; 1772 default: 1773 break; 1774 } 1775 1776 return (error); 1777 } 1778 1779 /* 1780 * retrive the named entry from the directory cache 1781 */ 1782 struct sdev_node * 1783 sdev_cache_lookup(struct sdev_node *ddv, char *nm) 1784 { 1785 struct sdev_node *dv = NULL; 1786 1787 ASSERT(RW_LOCK_HELD(&ddv->sdev_contents)); 1788 dv = sdev_findbyname(ddv, nm); 1789 1790 return (dv); 1791 } 1792 1793 /* 1794 * Implicit reconfig for nodes constructed by a link generator 1795 * Start devfsadm if needed, or if devfsadm is in progress, 1796 * prepare to block on devfsadm either completing or 1797 * constructing the desired node. As devfsadmd is global 1798 * in scope, constructing all necessary nodes, we only 1799 * need to initiate it once. 1800 */ 1801 static int 1802 sdev_call_devfsadmd(struct sdev_node *ddv, struct sdev_node *dv, char *nm) 1803 { 1804 int error = 0; 1805 1806 if (DEVNAME_DEVFSADM_IS_RUNNING(devfsadm_state)) { 1807 sdcmn_err6(("lookup: waiting for %s/%s, 0x%x\n", 1808 ddv->sdev_name, nm, devfsadm_state)); 1809 mutex_enter(&dv->sdev_lookup_lock); 1810 SDEV_BLOCK_OTHERS(dv, (SDEV_LOOKUP | SDEV_LGWAITING)); 1811 mutex_exit(&dv->sdev_lookup_lock); 1812 error = 0; 1813 } else if (!DEVNAME_DEVFSADM_HAS_RUN(devfsadm_state)) { 1814 sdcmn_err6(("lookup %s/%s starting devfsadm, 0x%x\n", 1815 ddv->sdev_name, nm, devfsadm_state)); 1816 1817 sdev_devfsadmd_thread(ddv, dv, kcred); 1818 mutex_enter(&dv->sdev_lookup_lock); 1819 SDEV_BLOCK_OTHERS(dv, 1820 (SDEV_LOOKUP | SDEV_LGWAITING)); 1821 mutex_exit(&dv->sdev_lookup_lock); 1822 error = 0; 1823 } else { 1824 error = -1; 1825 } 1826 1827 return (error); 1828 } 1829 1830 static int 1831 sdev_call_modulelookup(struct sdev_node *ddv, struct sdev_node **dvp, char *nm, 1832 int (*fn)(char *, devname_handle_t *, struct cred *), struct cred *cred) 1833 { 1834 struct vnode *rvp = NULL; 1835 int error = 0; 1836 struct vattr *vap; 1837 devname_spec_t spec; 1838 devname_handle_t *hdl; 1839 void *args = NULL; 1840 struct sdev_node *dv = *dvp; 1841 1842 ASSERT(dv && ddv); 1843 hdl = &(dv->sdev_handle); 1844 ASSERT(hdl->dh_data == dv); 1845 mutex_enter(&dv->sdev_lookup_lock); 1846 SDEV_BLOCK_OTHERS(dv, SDEV_LOOKUP); 1847 mutex_exit(&dv->sdev_lookup_lock); 1848 error = (*fn)(nm, hdl, cred); 1849 if (error) { 1850 return (error); 1851 } 1852 1853 spec = hdl->dh_spec; 1854 args = hdl->dh_args; 1855 ASSERT(args); 1856 1857 switch (spec) { 1858 case DEVNAME_NS_PATH: 1859 /* 1860 * symlink of: 1861 * /dev/dir/nm -> /device/... 1862 */ 1863 rvp = devname_configure_by_path((char *)args, NULL); 1864 break; 1865 case DEVNAME_NS_DEV: 1866 /* 1867 * symlink of: 1868 * /dev/dir/nm -> /dev/... 1869 */ 1870 rvp = devname_find_by_devpath((char *)args, NULL); 1871 break; 1872 default: 1873 if (args) 1874 kmem_free((char *)args, strlen(args) + 1); 1875 return (ENOENT); 1876 1877 } 1878 1879 if (rvp == NULL) { 1880 if (args) 1881 kmem_free((char *)args, strlen(args) + 1); 1882 return (ENOENT); 1883 } else { 1884 vap = sdev_getdefault_attr(VLNK); 1885 ASSERT(RW_READ_HELD(&ddv->sdev_contents)); 1886 /* 1887 * Could sdev_mknode return a different dv_node 1888 * once the lock is dropped? 1889 */ 1890 if (!rw_tryupgrade(&ddv->sdev_contents)) { 1891 rw_exit(&ddv->sdev_contents); 1892 rw_enter(&ddv->sdev_contents, RW_WRITER); 1893 } 1894 error = sdev_mknode(ddv, nm, &dv, vap, NULL, args, cred, 1895 SDEV_READY); 1896 rw_downgrade(&ddv->sdev_contents); 1897 if (error) { 1898 if (args) 1899 kmem_free((char *)args, strlen(args) + 1); 1900 return (error); 1901 } else { 1902 mutex_enter(&dv->sdev_lookup_lock); 1903 SDEV_UNBLOCK_OTHERS(dv, SDEV_LOOKUP); 1904 mutex_exit(&dv->sdev_lookup_lock); 1905 error = 0; 1906 } 1907 } 1908 1909 if (args) 1910 kmem_free((char *)args, strlen(args) + 1); 1911 1912 *dvp = dv; 1913 return (0); 1914 } 1915 1916 /* 1917 * Support for specialized device naming construction mechanisms 1918 */ 1919 static int 1920 sdev_call_dircallback(struct sdev_node *ddv, struct sdev_node **dvp, char *nm, 1921 int (*callback)(struct sdev_node *, char *, void **, struct cred *, 1922 void *, char *), int flags, struct cred *cred) 1923 { 1924 int rv = 0; 1925 char *physpath = NULL; 1926 struct vnode *rvp = NULL; 1927 struct vattr vattr; 1928 struct vattr *vap; 1929 struct sdev_node *dv = *dvp; 1930 1931 mutex_enter(&dv->sdev_lookup_lock); 1932 SDEV_BLOCK_OTHERS(dv, SDEV_LOOKUP); 1933 mutex_exit(&dv->sdev_lookup_lock); 1934 1935 /* for non-devfsadm devices */ 1936 if (flags & SDEV_PATH) { 1937 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 1938 rv = callback(ddv, nm, (void *)&physpath, kcred, NULL, 1939 NULL); 1940 if (rv) { 1941 kmem_free(physpath, MAXPATHLEN); 1942 return (-1); 1943 } 1944 1945 ASSERT(physpath); 1946 rvp = devname_configure_by_path(physpath, NULL); 1947 if (rvp == NULL) { 1948 sdcmn_err3(("devname_configure_by_path: " 1949 "failed for /dev/%s/%s\n", 1950 ddv->sdev_name, nm)); 1951 kmem_free(physpath, MAXPATHLEN); 1952 rv = -1; 1953 } else { 1954 vap = sdev_getdefault_attr(VLNK); 1955 ASSERT(RW_READ_HELD(&ddv->sdev_contents)); 1956 1957 /* 1958 * Sdev_mknode may return back a different sdev_node 1959 * that was created by another thread that 1960 * raced to the directroy cache before this thread. 1961 * 1962 * With current directory cache mechanism 1963 * (linked list with the sdev_node name as 1964 * the entity key), this is a way to make sure 1965 * only one entry exists for the same name 1966 * in the same directory. The outcome is 1967 * the winner wins. 1968 */ 1969 if (!rw_tryupgrade(&ddv->sdev_contents)) { 1970 rw_exit(&ddv->sdev_contents); 1971 rw_enter(&ddv->sdev_contents, RW_WRITER); 1972 } 1973 rv = sdev_mknode(ddv, nm, &dv, vap, NULL, 1974 (void *)physpath, cred, SDEV_READY); 1975 rw_downgrade(&ddv->sdev_contents); 1976 kmem_free(physpath, MAXPATHLEN); 1977 if (rv) { 1978 return (rv); 1979 } else { 1980 mutex_enter(&dv->sdev_lookup_lock); 1981 SDEV_UNBLOCK_OTHERS(dv, SDEV_LOOKUP); 1982 mutex_exit(&dv->sdev_lookup_lock); 1983 return (0); 1984 } 1985 } 1986 } else if (flags & SDEV_VNODE) { 1987 /* 1988 * DBNR has its own way to create the device 1989 * and return a backing store vnode in rvp 1990 */ 1991 ASSERT(callback); 1992 rv = callback(ddv, nm, (void *)&rvp, kcred, NULL, NULL); 1993 if (rv || (rvp == NULL)) { 1994 sdcmn_err3(("devname_lookup_func: SDEV_VNODE " 1995 "callback failed \n")); 1996 return (-1); 1997 } 1998 vap = sdev_getdefault_attr(rvp->v_type); 1999 if (vap == NULL) 2000 return (-1); 2001 2002 ASSERT(RW_READ_HELD(&ddv->sdev_contents)); 2003 if (!rw_tryupgrade(&ddv->sdev_contents)) { 2004 rw_exit(&ddv->sdev_contents); 2005 rw_enter(&ddv->sdev_contents, RW_WRITER); 2006 } 2007 rv = sdev_mknode(ddv, nm, &dv, vap, rvp, NULL, 2008 cred, SDEV_READY); 2009 rw_downgrade(&ddv->sdev_contents); 2010 if (rv) 2011 return (rv); 2012 2013 mutex_enter(&dv->sdev_lookup_lock); 2014 SDEV_UNBLOCK_OTHERS(dv, SDEV_LOOKUP); 2015 mutex_exit(&dv->sdev_lookup_lock); 2016 return (0); 2017 } else if (flags & SDEV_VATTR) { 2018 /* 2019 * /dev/pts 2020 * 2021 * DBNR has its own way to create the device 2022 * "0" is returned upon success. 2023 * 2024 * callback is responsible to set the basic attributes, 2025 * e.g. va_type/va_uid/va_gid/ 2026 * dev_t if VCHR or VBLK/ 2027 */ 2028 ASSERT(callback); 2029 rv = callback(ddv, nm, (void *)&vattr, kcred, NULL, NULL); 2030 if (rv) { 2031 sdcmn_err3(("devname_lookup_func: SDEV_NONE " 2032 "callback failed \n")); 2033 return (-1); 2034 } 2035 2036 ASSERT(RW_READ_HELD(&ddv->sdev_contents)); 2037 if (!rw_tryupgrade(&ddv->sdev_contents)) { 2038 rw_exit(&ddv->sdev_contents); 2039 rw_enter(&ddv->sdev_contents, RW_WRITER); 2040 } 2041 rv = sdev_mknode(ddv, nm, &dv, &vattr, NULL, NULL, 2042 cred, SDEV_READY); 2043 rw_downgrade(&ddv->sdev_contents); 2044 2045 if (rv) 2046 return (rv); 2047 2048 mutex_enter(&dv->sdev_lookup_lock); 2049 SDEV_UNBLOCK_OTHERS(dv, SDEV_LOOKUP); 2050 mutex_exit(&dv->sdev_lookup_lock); 2051 return (0); 2052 } else { 2053 impossible(("lookup: %s/%s by %s not supported (%d)\n", 2054 SDEVTOV(ddv)->v_path, nm, curproc->p_user.u_comm, 2055 __LINE__)); 2056 rv = -1; 2057 } 2058 2059 *dvp = dv; 2060 return (rv); 2061 } 2062 2063 static int 2064 is_devfsadm_thread(char *exec_name) 2065 { 2066 /* 2067 * note: because devfsadmd -> /usr/sbin/devfsadm 2068 * it is safe to use "devfsadm" to capture the lookups 2069 * from devfsadm and its daemon version. 2070 */ 2071 if (strcmp(exec_name, "devfsadm") == 0) 2072 return (1); 2073 return (0); 2074 } 2075 2076 2077 /* 2078 * Lookup Order: 2079 * sdev_node cache; 2080 * backing store (SDEV_PERSIST); 2081 * DBNR: a. dir_ops implemented in the loadable modules; 2082 * b. vnode ops in vtab. 2083 */ 2084 int 2085 devname_lookup_func(struct sdev_node *ddv, char *nm, struct vnode **vpp, 2086 struct cred *cred, int (*callback)(struct sdev_node *, char *, void **, 2087 struct cred *, void *, char *), int flags) 2088 { 2089 int rv = 0, nmlen; 2090 struct vnode *rvp = NULL; 2091 struct sdev_node *dv = NULL; 2092 int retried = 0; 2093 int error = 0; 2094 struct devname_nsmap *map = NULL; 2095 struct devname_ops *dirops = NULL; 2096 int (*fn)(char *, devname_handle_t *, struct cred *) = NULL; 2097 struct vattr vattr; 2098 char *lookup_thread = curproc->p_user.u_comm; 2099 int failed_flags = 0; 2100 int (*vtor)(struct sdev_node *) = NULL; 2101 int state; 2102 int parent_state; 2103 char *link = NULL; 2104 2105 if (SDEVTOV(ddv)->v_type != VDIR) 2106 return (ENOTDIR); 2107 2108 /* 2109 * Empty name or ., return node itself. 2110 */ 2111 nmlen = strlen(nm); 2112 if ((nmlen == 0) || ((nmlen == 1) && (nm[0] == '.'))) { 2113 *vpp = SDEVTOV(ddv); 2114 VN_HOLD(*vpp); 2115 return (0); 2116 } 2117 2118 /* 2119 * .., return the parent directory 2120 */ 2121 if ((nmlen == 2) && (strcmp(nm, "..") == 0)) { 2122 *vpp = SDEVTOV(ddv->sdev_dotdot); 2123 VN_HOLD(*vpp); 2124 return (0); 2125 } 2126 2127 rw_enter(&ddv->sdev_contents, RW_READER); 2128 if (ddv->sdev_flags & SDEV_VTOR) { 2129 vtor = (int (*)(struct sdev_node *))sdev_get_vtor(ddv); 2130 ASSERT(vtor); 2131 } 2132 2133 tryagain: 2134 /* 2135 * (a) directory cache lookup: 2136 */ 2137 ASSERT(RW_READ_HELD(&ddv->sdev_contents)); 2138 parent_state = ddv->sdev_state; 2139 dv = sdev_cache_lookup(ddv, nm); 2140 if (dv) { 2141 state = dv->sdev_state; 2142 switch (state) { 2143 case SDEV_INIT: 2144 if (is_devfsadm_thread(lookup_thread)) 2145 break; 2146 2147 /* ZOMBIED parent won't allow node creation */ 2148 if (parent_state == SDEV_ZOMBIE) { 2149 SD_TRACE_FAILED_LOOKUP(ddv, nm, 2150 retried); 2151 goto nolock_notfound; 2152 } 2153 2154 mutex_enter(&dv->sdev_lookup_lock); 2155 /* compensate the threads started after devfsadm */ 2156 if (DEVNAME_DEVFSADM_IS_RUNNING(devfsadm_state) && 2157 !(SDEV_IS_LOOKUP(dv))) 2158 SDEV_BLOCK_OTHERS(dv, 2159 (SDEV_LOOKUP | SDEV_LGWAITING)); 2160 2161 if (SDEV_IS_LOOKUP(dv)) { 2162 failed_flags |= SLF_REBUILT; 2163 rw_exit(&ddv->sdev_contents); 2164 error = sdev_wait4lookup(dv, SDEV_LOOKUP); 2165 mutex_exit(&dv->sdev_lookup_lock); 2166 rw_enter(&ddv->sdev_contents, RW_READER); 2167 2168 if (error != 0) { 2169 SD_TRACE_FAILED_LOOKUP(ddv, nm, 2170 retried); 2171 goto nolock_notfound; 2172 } 2173 2174 state = dv->sdev_state; 2175 if (state == SDEV_INIT) { 2176 SD_TRACE_FAILED_LOOKUP(ddv, nm, 2177 retried); 2178 goto nolock_notfound; 2179 } else if (state == SDEV_READY) { 2180 goto found; 2181 } else if (state == SDEV_ZOMBIE) { 2182 rw_exit(&ddv->sdev_contents); 2183 SD_TRACE_FAILED_LOOKUP(ddv, nm, 2184 retried); 2185 SDEV_RELE(dv); 2186 goto lookup_failed; 2187 } 2188 } else { 2189 mutex_exit(&dv->sdev_lookup_lock); 2190 } 2191 break; 2192 case SDEV_READY: 2193 goto found; 2194 case SDEV_ZOMBIE: 2195 rw_exit(&ddv->sdev_contents); 2196 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2197 SDEV_RELE(dv); 2198 goto lookup_failed; 2199 default: 2200 rw_exit(&ddv->sdev_contents); 2201 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2202 sdev_lookup_failed(ddv, nm, failed_flags); 2203 *vpp = NULLVP; 2204 return (ENOENT); 2205 } 2206 } 2207 ASSERT(RW_READ_HELD(&ddv->sdev_contents)); 2208 2209 /* 2210 * ZOMBIED parent does not allow new node creation. 2211 * bail out early 2212 */ 2213 if (parent_state == SDEV_ZOMBIE) { 2214 rw_exit(&ddv->sdev_contents); 2215 *vpp = NULL; 2216 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2217 return (ENOENT); 2218 } 2219 2220 /* 2221 * (b0): backing store lookup 2222 * SDEV_PERSIST is default except: 2223 * 1) pts nodes 2224 * 2) non-chmod'ed local nodes 2225 */ 2226 if (SDEV_IS_PERSIST(ddv)) { 2227 error = devname_backstore_lookup(ddv, nm, &rvp); 2228 2229 if (!error) { 2230 sdcmn_err3(("devname_backstore_lookup: " 2231 "found attrvp %p for %s\n", (void *)rvp, nm)); 2232 2233 vattr.va_mask = AT_MODE|AT_UID|AT_GID; 2234 error = VOP_GETATTR(rvp, &vattr, 0, cred); 2235 if (error) { 2236 rw_exit(&ddv->sdev_contents); 2237 if (dv) 2238 SDEV_RELE(dv); 2239 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2240 sdev_lookup_failed(ddv, nm, failed_flags); 2241 *vpp = NULLVP; 2242 return (ENOENT); 2243 } 2244 2245 if (vattr.va_type == VLNK) { 2246 error = sdev_getlink(rvp, &link); 2247 if (error) { 2248 rw_exit(&ddv->sdev_contents); 2249 if (dv) 2250 SDEV_RELE(dv); 2251 SD_TRACE_FAILED_LOOKUP(ddv, nm, 2252 retried); 2253 sdev_lookup_failed(ddv, nm, 2254 failed_flags); 2255 *vpp = NULLVP; 2256 return (ENOENT); 2257 } 2258 ASSERT(link != NULL); 2259 } 2260 2261 if (!rw_tryupgrade(&ddv->sdev_contents)) { 2262 rw_exit(&ddv->sdev_contents); 2263 rw_enter(&ddv->sdev_contents, RW_WRITER); 2264 } 2265 error = sdev_mknode(ddv, nm, &dv, &vattr, 2266 rvp, link, cred, SDEV_READY); 2267 rw_downgrade(&ddv->sdev_contents); 2268 2269 if (link != NULL) { 2270 kmem_free(link, strlen(link) + 1); 2271 link = NULL; 2272 } 2273 2274 if (error) { 2275 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2276 rw_exit(&ddv->sdev_contents); 2277 if (dv) 2278 SDEV_RELE(dv); 2279 goto lookup_failed; 2280 } else { 2281 goto found; 2282 } 2283 } else if (retried) { 2284 rw_exit(&ddv->sdev_contents); 2285 sdcmn_err3(("retry of lookup of %s/%s: failed\n", 2286 ddv->sdev_name, nm)); 2287 if (dv) 2288 SDEV_RELE(dv); 2289 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2290 sdev_lookup_failed(ddv, nm, failed_flags); 2291 *vpp = NULLVP; 2292 return (ENOENT); 2293 } 2294 } 2295 2296 2297 /* first thread that is doing the lookup on this node */ 2298 if (!dv) { 2299 if (!rw_tryupgrade(&ddv->sdev_contents)) { 2300 rw_exit(&ddv->sdev_contents); 2301 rw_enter(&ddv->sdev_contents, RW_WRITER); 2302 } 2303 error = sdev_mknode(ddv, nm, &dv, NULL, NULL, NULL, 2304 cred, SDEV_INIT); 2305 if (!dv) { 2306 rw_exit(&ddv->sdev_contents); 2307 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2308 sdev_lookup_failed(ddv, nm, failed_flags); 2309 *vpp = NULLVP; 2310 return (ENOENT); 2311 } 2312 rw_downgrade(&ddv->sdev_contents); 2313 } 2314 ASSERT(dv); 2315 ASSERT(SDEV_HELD(dv)); 2316 2317 if (SDEV_IS_NO_NCACHE(dv)) { 2318 failed_flags |= SLF_NO_NCACHE; 2319 } 2320 2321 if (SDEV_IS_GLOBAL(ddv)) { 2322 map = sdev_get_map(ddv, 1); 2323 dirops = map ? map->dir_ops : NULL; 2324 fn = dirops ? dirops->devnops_lookup : NULL; 2325 } 2326 2327 /* 2328 * (b1) invoking devfsadm once per life time for devfsadm nodes 2329 */ 2330 if ((fn == NULL) && !callback) { 2331 2332 if (sdev_reconfig_boot || !i_ddi_io_initialized() || 2333 SDEV_IS_DYNAMIC(ddv) || SDEV_IS_NO_NCACHE(dv) || 2334 ((moddebug & MODDEBUG_FINI_EBUSY) != 0)) { 2335 ASSERT(SDEV_HELD(dv)); 2336 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2337 goto nolock_notfound; 2338 } 2339 2340 /* 2341 * filter out known non-existent devices recorded 2342 * during initial reconfiguration boot for which 2343 * reconfig should not be done and lookup may 2344 * be short-circuited now. 2345 */ 2346 if (sdev_lookup_filter(ddv, nm)) { 2347 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2348 goto nolock_notfound; 2349 } 2350 2351 /* bypassing devfsadm internal nodes */ 2352 if (is_devfsadm_thread(lookup_thread)) { 2353 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2354 goto nolock_notfound; 2355 } 2356 2357 if (sdev_reconfig_disable) { 2358 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2359 goto nolock_notfound; 2360 } 2361 2362 error = sdev_call_devfsadmd(ddv, dv, nm); 2363 if (error == 0) { 2364 sdcmn_err8(("lookup of %s/%s by %s: reconfig\n", 2365 ddv->sdev_name, nm, curproc->p_user.u_comm)); 2366 if (sdev_reconfig_verbose) { 2367 cmn_err(CE_CONT, 2368 "?lookup of %s/%s by %s: reconfig\n", 2369 ddv->sdev_name, nm, curproc->p_user.u_comm); 2370 } 2371 retried = 1; 2372 failed_flags |= SLF_REBUILT; 2373 ASSERT(dv->sdev_state != SDEV_ZOMBIE); 2374 SDEV_SIMPLE_RELE(dv); 2375 goto tryagain; 2376 } else { 2377 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2378 goto nolock_notfound; 2379 } 2380 } 2381 2382 /* 2383 * (b2) Directory Based Name Resolution (DBNR): 2384 * ddv - parent 2385 * nm - /dev/(ddv->sdev_name)/nm 2386 * 2387 * note: module vnode ops take precedence than the build-in ones 2388 */ 2389 if (fn) { 2390 error = sdev_call_modulelookup(ddv, &dv, nm, fn, cred); 2391 if (error) { 2392 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2393 goto notfound; 2394 } else { 2395 goto found; 2396 } 2397 } else if (callback) { 2398 error = sdev_call_dircallback(ddv, &dv, nm, callback, 2399 flags, cred); 2400 if (error == 0) { 2401 goto found; 2402 } else { 2403 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2404 goto notfound; 2405 } 2406 } 2407 ASSERT(rvp); 2408 2409 found: 2410 ASSERT(!(dv->sdev_flags & SDEV_STALE)); 2411 ASSERT(dv->sdev_state == SDEV_READY); 2412 if (vtor) { 2413 /* 2414 * Check validity of returned node 2415 */ 2416 switch (vtor(dv)) { 2417 case SDEV_VTOR_VALID: 2418 break; 2419 case SDEV_VTOR_INVALID: 2420 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2421 sdcmn_err7(("lookup: destroy invalid " 2422 "node: %s(%p)\n", dv->sdev_name, (void *)dv)); 2423 goto nolock_notfound; 2424 case SDEV_VTOR_SKIP: 2425 sdcmn_err7(("lookup: node not applicable - " 2426 "skipping: %s(%p)\n", dv->sdev_name, (void *)dv)); 2427 rw_exit(&ddv->sdev_contents); 2428 SD_TRACE_FAILED_LOOKUP(ddv, nm, retried); 2429 SDEV_RELE(dv); 2430 goto lookup_failed; 2431 default: 2432 cmn_err(CE_PANIC, 2433 "dev fs: validator failed: %s(%p)\n", 2434 dv->sdev_name, (void *)dv); 2435 break; 2436 /*NOTREACHED*/ 2437 } 2438 } 2439 2440 if ((SDEVTOV(dv)->v_type == VDIR) && SDEV_IS_GLOBAL(dv)) { 2441 rw_enter(&dv->sdev_contents, RW_READER); 2442 (void) sdev_get_map(dv, 1); 2443 rw_exit(&dv->sdev_contents); 2444 } 2445 rw_exit(&ddv->sdev_contents); 2446 rv = sdev_to_vp(dv, vpp); 2447 sdcmn_err3(("devname_lookup_func: returning vp %p v_count %d state %d " 2448 "for nm %s, error %d\n", (void *)*vpp, (*vpp)->v_count, 2449 dv->sdev_state, nm, rv)); 2450 return (rv); 2451 2452 notfound: 2453 mutex_enter(&dv->sdev_lookup_lock); 2454 SDEV_UNBLOCK_OTHERS(dv, SDEV_LOOKUP); 2455 mutex_exit(&dv->sdev_lookup_lock); 2456 nolock_notfound: 2457 /* 2458 * Destroy the node that is created for synchronization purposes. 2459 */ 2460 sdcmn_err3(("devname_lookup_func: %s with state %d\n", 2461 nm, dv->sdev_state)); 2462 ASSERT(RW_READ_HELD(&ddv->sdev_contents)); 2463 if (dv->sdev_state == SDEV_INIT) { 2464 if (!rw_tryupgrade(&ddv->sdev_contents)) { 2465 rw_exit(&ddv->sdev_contents); 2466 rw_enter(&ddv->sdev_contents, RW_WRITER); 2467 } 2468 2469 /* 2470 * Node state may have changed during the lock 2471 * changes. Re-check. 2472 */ 2473 if (dv->sdev_state == SDEV_INIT) { 2474 (void) sdev_dirdelete(ddv, dv); 2475 rw_exit(&ddv->sdev_contents); 2476 sdev_lookup_failed(ddv, nm, failed_flags); 2477 *vpp = NULL; 2478 return (ENOENT); 2479 } 2480 } 2481 2482 rw_exit(&ddv->sdev_contents); 2483 SDEV_RELE(dv); 2484 2485 lookup_failed: 2486 sdev_lookup_failed(ddv, nm, failed_flags); 2487 *vpp = NULL; 2488 return (ENOENT); 2489 } 2490 2491 /* 2492 * Given a directory node, mark all nodes beneath as 2493 * STALE, i.e. nodes that don't exist as far as new 2494 * consumers are concerned 2495 */ 2496 void 2497 sdev_stale(struct sdev_node *ddv) 2498 { 2499 struct sdev_node *dv; 2500 struct vnode *vp; 2501 2502 ASSERT(SDEVTOV(ddv)->v_type == VDIR); 2503 2504 rw_enter(&ddv->sdev_contents, RW_WRITER); 2505 for (dv = ddv->sdev_dot; dv; dv = dv->sdev_next) { 2506 vp = SDEVTOV(dv); 2507 if (vp->v_type == VDIR) 2508 sdev_stale(dv); 2509 2510 sdcmn_err9(("sdev_stale: setting stale %s\n", 2511 dv->sdev_name)); 2512 dv->sdev_flags |= SDEV_STALE; 2513 } 2514 ddv->sdev_flags |= SDEV_BUILD; 2515 rw_exit(&ddv->sdev_contents); 2516 } 2517 2518 /* 2519 * Given a directory node, clean out all the nodes beneath. 2520 * If expr is specified, clean node with names matching expr. 2521 * If SDEV_ENFORCE is specified in flags, busy nodes are made stale, 2522 * so they are excluded from future lookups. 2523 */ 2524 int 2525 sdev_cleandir(struct sdev_node *ddv, char *expr, uint_t flags) 2526 { 2527 int error = 0; 2528 int busy = 0; 2529 struct vnode *vp; 2530 struct sdev_node *dv, *next = NULL; 2531 int bkstore = 0; 2532 int len = 0; 2533 char *bks_name = NULL; 2534 2535 ASSERT(SDEVTOV(ddv)->v_type == VDIR); 2536 2537 /* 2538 * We try our best to destroy all unused sdev_node's 2539 */ 2540 rw_enter(&ddv->sdev_contents, RW_WRITER); 2541 for (dv = ddv->sdev_dot; dv; dv = next) { 2542 next = dv->sdev_next; 2543 vp = SDEVTOV(dv); 2544 2545 if (expr && gmatch(dv->sdev_name, expr) == 0) 2546 continue; 2547 2548 if (vp->v_type == VDIR && 2549 sdev_cleandir(dv, NULL, flags) != 0) { 2550 sdcmn_err9(("sdev_cleandir: dir %s busy\n", 2551 dv->sdev_name)); 2552 busy++; 2553 continue; 2554 } 2555 2556 if (vp->v_count > 0 && (flags & SDEV_ENFORCE) == 0) { 2557 sdcmn_err9(("sdev_cleandir: dir %s busy\n", 2558 dv->sdev_name)); 2559 busy++; 2560 continue; 2561 } 2562 2563 /* 2564 * at this point, either dv is not held or SDEV_ENFORCE 2565 * is specified. In either case, dv needs to be deleted 2566 */ 2567 SDEV_HOLD(dv); 2568 2569 bkstore = SDEV_IS_PERSIST(dv) ? 1 : 0; 2570 if (bkstore && (vp->v_type == VDIR)) 2571 bkstore += 1; 2572 2573 if (bkstore) { 2574 len = strlen(dv->sdev_name) + 1; 2575 bks_name = kmem_alloc(len, KM_SLEEP); 2576 bcopy(dv->sdev_name, bks_name, len); 2577 } 2578 2579 error = sdev_dirdelete(ddv, dv); 2580 2581 if (error == EBUSY) { 2582 sdcmn_err9(("sdev_cleandir: dir busy\n")); 2583 busy++; 2584 } 2585 2586 /* take care the backing store clean up */ 2587 if (bkstore && (error == 0)) { 2588 ASSERT(bks_name); 2589 ASSERT(ddv->sdev_attrvp); 2590 2591 if (bkstore == 1) { 2592 error = VOP_REMOVE(ddv->sdev_attrvp, 2593 bks_name, kcred); 2594 } else if (bkstore == 2) { 2595 error = VOP_RMDIR(ddv->sdev_attrvp, 2596 bks_name, ddv->sdev_attrvp, kcred); 2597 } 2598 2599 /* do not propagate the backing store errors */ 2600 if (error) { 2601 sdcmn_err9(("sdev_cleandir: backing store" 2602 "not cleaned\n")); 2603 error = 0; 2604 } 2605 2606 bkstore = 0; 2607 kmem_free(bks_name, len); 2608 bks_name = NULL; 2609 len = 0; 2610 } 2611 } 2612 2613 ddv->sdev_flags |= SDEV_BUILD; 2614 rw_exit(&ddv->sdev_contents); 2615 2616 if (busy) { 2617 error = EBUSY; 2618 } 2619 2620 return (error); 2621 } 2622 2623 /* 2624 * a convenient wrapper for readdir() funcs 2625 */ 2626 size_t 2627 add_dir_entry(dirent64_t *de, char *nm, size_t size, ino_t ino, offset_t off) 2628 { 2629 size_t reclen = DIRENT64_RECLEN(strlen(nm)); 2630 if (reclen > size) 2631 return (0); 2632 2633 de->d_ino = (ino64_t)ino; 2634 de->d_off = (off64_t)off + 1; 2635 de->d_reclen = (ushort_t)reclen; 2636 (void) strncpy(de->d_name, nm, DIRENT64_NAMELEN(reclen)); 2637 return (reclen); 2638 } 2639 2640 /* 2641 * sdev_mount service routines 2642 */ 2643 int 2644 sdev_copyin_mountargs(struct mounta *uap, struct sdev_mountargs *args) 2645 { 2646 int error; 2647 2648 if (uap->datalen != sizeof (*args)) 2649 return (EINVAL); 2650 2651 if (error = copyin(uap->dataptr, args, sizeof (*args))) { 2652 cmn_err(CE_WARN, "sdev_copyin_mountargs: can not" 2653 "get user data. error %d\n", error); 2654 return (EFAULT); 2655 } 2656 2657 return (0); 2658 } 2659 2660 #ifdef nextdp 2661 #undef nextdp 2662 #endif 2663 #define nextdp(dp) ((struct dirent64 *)((char *)(dp) + (dp)->d_reclen)) 2664 2665 /* 2666 * readdir helper func 2667 */ 2668 int 2669 devname_readdir_func(vnode_t *vp, uio_t *uiop, cred_t *cred, int *eofp, 2670 int flags) 2671 { 2672 struct sdev_node *ddv = VTOSDEV(vp); 2673 struct sdev_node *dv; 2674 dirent64_t *dp; 2675 ulong_t outcount = 0; 2676 size_t namelen; 2677 ulong_t alloc_count; 2678 void *outbuf; 2679 struct iovec *iovp; 2680 int error = 0; 2681 size_t reclen; 2682 offset_t diroff; 2683 offset_t soff; 2684 int this_reclen; 2685 struct devname_nsmap *map = NULL; 2686 struct devname_ops *dirops = NULL; 2687 int (*fn)(devname_handle_t *, struct cred *) = NULL; 2688 int (*vtor)(struct sdev_node *) = NULL; 2689 struct vattr attr; 2690 timestruc_t now; 2691 2692 ASSERT(ddv->sdev_attr || ddv->sdev_attrvp); 2693 ASSERT(RW_READ_HELD(&ddv->sdev_contents)); 2694 2695 if (uiop->uio_loffset >= MAXOFF_T) { 2696 if (eofp) 2697 *eofp = 1; 2698 return (0); 2699 } 2700 2701 if (uiop->uio_iovcnt != 1) 2702 return (EINVAL); 2703 2704 if (vp->v_type != VDIR) 2705 return (ENOTDIR); 2706 2707 if (ddv->sdev_flags & SDEV_VTOR) { 2708 vtor = (int (*)(struct sdev_node *))sdev_get_vtor(ddv); 2709 ASSERT(vtor); 2710 } 2711 2712 if (eofp != NULL) 2713 *eofp = 0; 2714 2715 soff = uiop->uio_offset; 2716 iovp = uiop->uio_iov; 2717 alloc_count = iovp->iov_len; 2718 dp = outbuf = kmem_alloc(alloc_count, KM_SLEEP); 2719 outcount = 0; 2720 2721 if (ddv->sdev_state == SDEV_ZOMBIE) 2722 goto get_cache; 2723 2724 if (SDEV_IS_GLOBAL(ddv)) { 2725 map = sdev_get_map(ddv, 0); 2726 dirops = map ? map->dir_ops : NULL; 2727 fn = dirops ? dirops->devnops_readdir : NULL; 2728 2729 if (map && map->dir_map) { 2730 /* 2731 * load the name mapping rule database 2732 * through invoking devfsadm and symlink 2733 * all the entries in the map 2734 */ 2735 devname_rdr_result_t rdr_result; 2736 int do_thread = 0; 2737 2738 rw_enter(&map->dir_lock, RW_READER); 2739 do_thread = map->dir_maploaded ? 0 : 1; 2740 rw_exit(&map->dir_lock); 2741 2742 if (do_thread) { 2743 mutex_enter(&ddv->sdev_lookup_lock); 2744 SDEV_BLOCK_OTHERS(ddv, SDEV_READDIR); 2745 mutex_exit(&ddv->sdev_lookup_lock); 2746 2747 sdev_dispatch_to_nsrdr_thread(ddv, 2748 map->dir_map, &rdr_result); 2749 } 2750 } else if ((sdev_boot_state == SDEV_BOOT_STATE_COMPLETE) && 2751 !sdev_reconfig_boot && (flags & SDEV_BROWSE) && 2752 !SDEV_IS_DYNAMIC(ddv) && !SDEV_IS_NO_NCACHE(ddv) && 2753 ((moddebug & MODDEBUG_FINI_EBUSY) == 0) && 2754 !DEVNAME_DEVFSADM_HAS_RUN(devfsadm_state) && 2755 !DEVNAME_DEVFSADM_IS_RUNNING(devfsadm_state) && 2756 !sdev_reconfig_disable) { 2757 /* 2758 * invoking "devfsadm" to do system device reconfig 2759 */ 2760 mutex_enter(&ddv->sdev_lookup_lock); 2761 SDEV_BLOCK_OTHERS(ddv, 2762 (SDEV_READDIR|SDEV_LGWAITING)); 2763 mutex_exit(&ddv->sdev_lookup_lock); 2764 2765 sdcmn_err8(("readdir of %s by %s: reconfig\n", 2766 ddv->sdev_path, curproc->p_user.u_comm)); 2767 if (sdev_reconfig_verbose) { 2768 cmn_err(CE_CONT, 2769 "?readdir of %s by %s: reconfig\n", 2770 ddv->sdev_path, curproc->p_user.u_comm); 2771 } 2772 2773 sdev_devfsadmd_thread(ddv, NULL, kcred); 2774 } else if (DEVNAME_DEVFSADM_IS_RUNNING(devfsadm_state)) { 2775 /* 2776 * compensate the "ls" started later than "devfsadm" 2777 */ 2778 mutex_enter(&ddv->sdev_lookup_lock); 2779 SDEV_BLOCK_OTHERS(ddv, (SDEV_READDIR|SDEV_LGWAITING)); 2780 mutex_exit(&ddv->sdev_lookup_lock); 2781 } 2782 2783 /* 2784 * release the contents lock so that 2785 * the cache maybe updated by devfsadmd 2786 */ 2787 rw_exit(&ddv->sdev_contents); 2788 mutex_enter(&ddv->sdev_lookup_lock); 2789 if (SDEV_IS_READDIR(ddv)) 2790 (void) sdev_wait4lookup(ddv, SDEV_READDIR); 2791 mutex_exit(&ddv->sdev_lookup_lock); 2792 rw_enter(&ddv->sdev_contents, RW_READER); 2793 2794 sdcmn_err4(("readdir of directory %s by %s\n", 2795 ddv->sdev_name, curproc->p_user.u_comm)); 2796 while (ddv->sdev_flags & SDEV_BUILD) { 2797 if (SDEV_IS_PERSIST(ddv)) { 2798 error = sdev_filldir_from_store(ddv, 2799 alloc_count, cred); 2800 } 2801 2802 /* 2803 * pre-creating the directories 2804 * defined in vtab 2805 */ 2806 if (SDEVTOV(ddv)->v_flag & VROOT) { 2807 error = sdev_filldir_dynamic(ddv); 2808 } 2809 2810 if (!error) 2811 ddv->sdev_flags &= ~SDEV_BUILD; 2812 } 2813 } 2814 2815 get_cache: 2816 /* handle "." and ".." */ 2817 diroff = 0; 2818 if (soff == 0) { 2819 /* first time */ 2820 this_reclen = DIRENT64_RECLEN(1); 2821 if (alloc_count < this_reclen) { 2822 error = EINVAL; 2823 goto done; 2824 } 2825 2826 dp->d_ino = (ino64_t)ddv->sdev_ino; 2827 dp->d_off = (off64_t)1; 2828 dp->d_reclen = (ushort_t)this_reclen; 2829 2830 (void) strncpy(dp->d_name, ".", 2831 DIRENT64_NAMELEN(this_reclen)); 2832 outcount += dp->d_reclen; 2833 dp = nextdp(dp); 2834 } 2835 2836 diroff++; 2837 if (soff <= 1) { 2838 this_reclen = DIRENT64_RECLEN(2); 2839 if (alloc_count < outcount + this_reclen) { 2840 error = EINVAL; 2841 goto done; 2842 } 2843 2844 dp->d_reclen = (ushort_t)this_reclen; 2845 dp->d_ino = (ino64_t)ddv->sdev_dotdot->sdev_ino; 2846 dp->d_off = (off64_t)2; 2847 2848 (void) strncpy(dp->d_name, "..", 2849 DIRENT64_NAMELEN(this_reclen)); 2850 outcount += dp->d_reclen; 2851 2852 dp = nextdp(dp); 2853 } 2854 2855 2856 /* gets the cache */ 2857 diroff++; 2858 for (dv = ddv->sdev_dot; dv; dv = dv->sdev_next, diroff++) { 2859 sdcmn_err3(("sdev_readdir: diroff %lld soff %lld for '%s' \n", 2860 diroff, soff, dv->sdev_name)); 2861 2862 /* bypassing pre-matured nodes */ 2863 if (diroff < soff || (dv->sdev_state != SDEV_READY)) { 2864 sdcmn_err3(("sdev_readdir: pre-mature node " 2865 "%s\n", dv->sdev_name)); 2866 continue; 2867 } 2868 2869 /* don't list stale nodes */ 2870 if (dv->sdev_flags & SDEV_STALE) { 2871 sdcmn_err4(("sdev_readdir: STALE node " 2872 "%s\n", dv->sdev_name)); 2873 continue; 2874 } 2875 2876 /* 2877 * Check validity of node 2878 */ 2879 if (vtor) { 2880 switch (vtor(dv)) { 2881 case SDEV_VTOR_VALID: 2882 break; 2883 case SDEV_VTOR_INVALID: 2884 case SDEV_VTOR_SKIP: 2885 continue; 2886 default: 2887 cmn_err(CE_PANIC, 2888 "dev fs: validator failed: %s(%p)\n", 2889 dv->sdev_name, (void *)dv); 2890 break; 2891 /*NOTREACHED*/ 2892 } 2893 } 2894 2895 /* 2896 * call back into the module for the validity/bookkeeping 2897 * of this entry 2898 */ 2899 if (fn) { 2900 error = (*fn)(&(dv->sdev_handle), cred); 2901 if (error) { 2902 sdcmn_err4(("sdev_readdir: module did not " 2903 "validate %s\n", dv->sdev_name)); 2904 continue; 2905 } 2906 } 2907 2908 namelen = strlen(dv->sdev_name); 2909 reclen = DIRENT64_RECLEN(namelen); 2910 if (outcount + reclen > alloc_count) { 2911 goto full; 2912 } 2913 dp->d_reclen = (ushort_t)reclen; 2914 dp->d_ino = (ino64_t)dv->sdev_ino; 2915 dp->d_off = (off64_t)diroff + 1; 2916 (void) strncpy(dp->d_name, dv->sdev_name, 2917 DIRENT64_NAMELEN(reclen)); 2918 outcount += reclen; 2919 dp = nextdp(dp); 2920 } 2921 2922 full: 2923 sdcmn_err4(("sdev_readdir: moving %lu bytes: " 2924 "diroff %lld, soff %lld, dv %p\n", outcount, diroff, soff, 2925 (void *)dv)); 2926 2927 if (outcount) 2928 error = uiomove(outbuf, outcount, UIO_READ, uiop); 2929 2930 if (!error) { 2931 uiop->uio_offset = diroff; 2932 if (eofp) 2933 *eofp = dv ? 0 : 1; 2934 } 2935 2936 2937 if (ddv->sdev_attrvp) { 2938 gethrestime(&now); 2939 attr.va_ctime = now; 2940 attr.va_atime = now; 2941 attr.va_mask = AT_CTIME|AT_ATIME; 2942 2943 (void) VOP_SETATTR(ddv->sdev_attrvp, &attr, 0, kcred, NULL); 2944 } 2945 done: 2946 kmem_free(outbuf, alloc_count); 2947 return (error); 2948 } 2949 2950 2951 static int 2952 sdev_modctl_lookup(const char *path, vnode_t **r_vp) 2953 { 2954 vnode_t *vp; 2955 vnode_t *cvp; 2956 struct sdev_node *svp; 2957 char *nm; 2958 struct pathname pn; 2959 int error; 2960 int persisted = 0; 2961 2962 if (error = pn_get((char *)path, UIO_SYSSPACE, &pn)) 2963 return (error); 2964 nm = kmem_alloc(MAXNAMELEN, KM_SLEEP); 2965 2966 vp = rootdir; 2967 VN_HOLD(vp); 2968 2969 while (pn_pathleft(&pn)) { 2970 ASSERT(vp->v_type == VDIR); 2971 (void) pn_getcomponent(&pn, nm); 2972 error = VOP_LOOKUP(vp, nm, &cvp, NULL, 0, NULL, kcred); 2973 VN_RELE(vp); 2974 2975 if (error) 2976 break; 2977 2978 /* traverse mount points encountered on our journey */ 2979 if (vn_ismntpt(cvp) && (error = traverse(&cvp)) != 0) { 2980 VN_RELE(cvp); 2981 break; 2982 } 2983 2984 /* 2985 * Direct the operation to the persisting filesystem 2986 * underlying /dev. Bail if we encounter a 2987 * non-persistent dev entity here. 2988 */ 2989 if (cvp->v_vfsp->vfs_fstype == devtype) { 2990 2991 if ((VTOSDEV(cvp)->sdev_flags & SDEV_PERSIST) == 0) { 2992 error = ENOENT; 2993 VN_RELE(cvp); 2994 break; 2995 } 2996 2997 if (VTOSDEV(cvp) == NULL) { 2998 error = ENOENT; 2999 VN_RELE(cvp); 3000 break; 3001 } 3002 svp = VTOSDEV(cvp); 3003 if ((vp = svp->sdev_attrvp) == NULL) { 3004 error = ENOENT; 3005 VN_RELE(cvp); 3006 break; 3007 } 3008 persisted = 1; 3009 VN_HOLD(vp); 3010 VN_RELE(cvp); 3011 cvp = vp; 3012 } 3013 3014 vp = cvp; 3015 pn_skipslash(&pn); 3016 } 3017 3018 kmem_free(nm, MAXNAMELEN); 3019 pn_free(&pn); 3020 3021 if (error) 3022 return (error); 3023 3024 /* 3025 * Only return persisted nodes in the filesystem underlying /dev. 3026 */ 3027 if (!persisted) { 3028 VN_RELE(vp); 3029 return (ENOENT); 3030 } 3031 3032 *r_vp = vp; 3033 return (0); 3034 } 3035 3036 int 3037 sdev_modctl_readdir(const char *dir, char ***dirlistp, 3038 int *npathsp, int *npathsp_alloc) 3039 { 3040 char **pathlist = NULL; 3041 char **newlist = NULL; 3042 int npaths = 0; 3043 int npaths_alloc = 0; 3044 dirent64_t *dbuf = NULL; 3045 int n; 3046 char *s; 3047 int error; 3048 vnode_t *vp; 3049 int eof; 3050 struct iovec iov; 3051 struct uio uio; 3052 struct dirent64 *dp; 3053 size_t dlen; 3054 size_t dbuflen; 3055 int ndirents = 64; 3056 char *nm; 3057 3058 error = sdev_modctl_lookup(dir, &vp); 3059 sdcmn_err11(("modctl readdir: %s by %s: %s\n", 3060 dir, curproc->p_user.u_comm, 3061 (error == 0) ? "ok" : "failed")); 3062 if (error) 3063 return (error); 3064 3065 dlen = ndirents * (sizeof (*dbuf)); 3066 dbuf = kmem_alloc(dlen, KM_SLEEP); 3067 3068 uio.uio_iov = &iov; 3069 uio.uio_iovcnt = 1; 3070 uio.uio_segflg = UIO_SYSSPACE; 3071 uio.uio_fmode = 0; 3072 uio.uio_extflg = UIO_COPY_CACHED; 3073 uio.uio_loffset = 0; 3074 uio.uio_llimit = MAXOFFSET_T; 3075 3076 eof = 0; 3077 error = 0; 3078 while (!error && !eof) { 3079 uio.uio_resid = dlen; 3080 iov.iov_base = (char *)dbuf; 3081 iov.iov_len = dlen; 3082 3083 (void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, NULL); 3084 error = VOP_READDIR(vp, &uio, kcred, &eof); 3085 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL); 3086 3087 dbuflen = dlen - uio.uio_resid; 3088 3089 if (error || dbuflen == 0) 3090 break; 3091 3092 for (dp = dbuf; ((intptr_t)dp < (intptr_t)dbuf + dbuflen); 3093 dp = (dirent64_t *)((intptr_t)dp + dp->d_reclen)) { 3094 3095 nm = dp->d_name; 3096 3097 if (strcmp(nm, ".") == 0 || strcmp(nm, "..") == 0) 3098 continue; 3099 3100 if (npaths == npaths_alloc) { 3101 npaths_alloc += 64; 3102 newlist = (char **) 3103 kmem_zalloc((npaths_alloc + 1) * 3104 sizeof (char *), KM_SLEEP); 3105 if (pathlist) { 3106 bcopy(pathlist, newlist, 3107 npaths * sizeof (char *)); 3108 kmem_free(pathlist, 3109 (npaths + 1) * sizeof (char *)); 3110 } 3111 pathlist = newlist; 3112 } 3113 n = strlen(nm) + 1; 3114 s = kmem_alloc(n, KM_SLEEP); 3115 bcopy(nm, s, n); 3116 pathlist[npaths++] = s; 3117 sdcmn_err11((" %s/%s\n", dir, s)); 3118 } 3119 } 3120 3121 exit: 3122 VN_RELE(vp); 3123 3124 if (dbuf) 3125 kmem_free(dbuf, dlen); 3126 3127 if (error) 3128 return (error); 3129 3130 *dirlistp = pathlist; 3131 *npathsp = npaths; 3132 *npathsp_alloc = npaths_alloc; 3133 3134 return (0); 3135 } 3136 3137 void 3138 sdev_modctl_readdir_free(char **pathlist, int npaths, int npaths_alloc) 3139 { 3140 int i, n; 3141 3142 for (i = 0; i < npaths; i++) { 3143 n = strlen(pathlist[i]) + 1; 3144 kmem_free(pathlist[i], n); 3145 } 3146 3147 kmem_free(pathlist, (npaths_alloc + 1) * sizeof (char *)); 3148 } 3149 3150 int 3151 sdev_modctl_devexists(const char *path) 3152 { 3153 vnode_t *vp; 3154 int error; 3155 3156 error = sdev_modctl_lookup(path, &vp); 3157 sdcmn_err11(("modctl dev exists: %s by %s: %s\n", 3158 path, curproc->p_user.u_comm, 3159 (error == 0) ? "ok" : "failed")); 3160 if (error == 0) 3161 VN_RELE(vp); 3162 3163 return (error); 3164 } 3165 3166 void 3167 sdev_update_newnsmap(struct devname_nsmap *map, char *module, char *mapname) 3168 { 3169 rw_enter(&map->dir_lock, RW_WRITER); 3170 if (module) { 3171 ASSERT(map->dir_newmodule == NULL); 3172 map->dir_newmodule = i_ddi_strdup(module, KM_SLEEP); 3173 } 3174 if (mapname) { 3175 ASSERT(map->dir_newmap == NULL); 3176 map->dir_newmap = i_ddi_strdup(mapname, KM_SLEEP); 3177 } 3178 3179 map->dir_invalid = 1; 3180 rw_exit(&map->dir_lock); 3181 } 3182 3183 void 3184 sdev_replace_nsmap(struct devname_nsmap *map, char *module, char *mapname) 3185 { 3186 char *old_module = NULL; 3187 char *old_map = NULL; 3188 3189 ASSERT(RW_LOCK_HELD(&map->dir_lock)); 3190 if (!rw_tryupgrade(&map->dir_lock)) { 3191 rw_exit(&map->dir_lock); 3192 rw_enter(&map->dir_lock, RW_WRITER); 3193 } 3194 3195 old_module = map->dir_module; 3196 if (module) { 3197 if (old_module && strcmp(old_module, module) != 0) { 3198 kmem_free(old_module, strlen(old_module) + 1); 3199 } 3200 map->dir_module = module; 3201 map->dir_newmodule = NULL; 3202 } 3203 3204 old_map = map->dir_map; 3205 if (mapname) { 3206 if (old_map && strcmp(old_map, mapname) != 0) { 3207 kmem_free(old_map, strlen(old_map) + 1); 3208 } 3209 3210 map->dir_map = mapname; 3211 map->dir_newmap = NULL; 3212 } 3213 map->dir_maploaded = 0; 3214 map->dir_invalid = 0; 3215 rw_downgrade(&map->dir_lock); 3216 } 3217 3218 /* 3219 * dir_name should have at least one attribute, 3220 * dir_module 3221 * or dir_map 3222 * or both 3223 * caller holds the devname_nsmaps_lock 3224 */ 3225 void 3226 sdev_insert_nsmap(char *dir_name, char *dir_module, char *dir_map) 3227 { 3228 struct devname_nsmap *map; 3229 int len = 0; 3230 3231 ASSERT(dir_name); 3232 ASSERT(dir_module || dir_map); 3233 ASSERT(MUTEX_HELD(&devname_nsmaps_lock)); 3234 3235 if (map = sdev_get_nsmap_by_dir(dir_name, 1)) { 3236 sdev_update_newnsmap(map, dir_module, dir_map); 3237 return; 3238 } 3239 3240 map = (struct devname_nsmap *)kmem_zalloc(sizeof (*map), KM_SLEEP); 3241 map->dir_name = i_ddi_strdup(dir_name, KM_SLEEP); 3242 if (dir_module) { 3243 map->dir_module = i_ddi_strdup(dir_module, KM_SLEEP); 3244 } 3245 3246 if (dir_map) { 3247 if (dir_map[0] != '/') { 3248 len = strlen(ETC_DEV_DIR) + strlen(dir_map) + 2; 3249 map->dir_map = kmem_zalloc(len, KM_SLEEP); 3250 (void) snprintf(map->dir_map, len, "%s/%s", ETC_DEV_DIR, 3251 dir_map); 3252 } else { 3253 map->dir_map = i_ddi_strdup(dir_map, KM_SLEEP); 3254 } 3255 } 3256 3257 map->dir_ops = NULL; 3258 map->dir_maploaded = 0; 3259 map->dir_invalid = 0; 3260 rw_init(&map->dir_lock, NULL, RW_DEFAULT, NULL); 3261 3262 map->next = devname_nsmaps; 3263 map->prev = NULL; 3264 if (devname_nsmaps) { 3265 devname_nsmaps->prev = map; 3266 } 3267 devname_nsmaps = map; 3268 } 3269 3270 struct devname_nsmap * 3271 sdev_get_nsmap_by_dir(char *dir_path, int locked) 3272 { 3273 struct devname_nsmap *map = NULL; 3274 3275 if (!locked) 3276 mutex_enter(&devname_nsmaps_lock); 3277 for (map = devname_nsmaps; map; map = map->next) { 3278 sdcmn_err6(("sdev_get_nsmap_by_dir: dir %s\n", map->dir_name)); 3279 if (strcmp(map->dir_name, dir_path) == 0) { 3280 if (!locked) 3281 mutex_exit(&devname_nsmaps_lock); 3282 return (map); 3283 } 3284 } 3285 if (!locked) 3286 mutex_exit(&devname_nsmaps_lock); 3287 return (NULL); 3288 } 3289 3290 struct devname_nsmap * 3291 sdev_get_nsmap_by_module(char *mod_name) 3292 { 3293 struct devname_nsmap *map = NULL; 3294 3295 mutex_enter(&devname_nsmaps_lock); 3296 for (map = devname_nsmaps; map; map = map->next) { 3297 sdcmn_err7(("sdev_get_nsmap_by_module: module %s\n", 3298 map->dir_module)); 3299 if (map->dir_module && strcmp(map->dir_module, mod_name) == 0) { 3300 mutex_exit(&devname_nsmaps_lock); 3301 return (map); 3302 } 3303 } 3304 mutex_exit(&devname_nsmaps_lock); 3305 return (NULL); 3306 } 3307 3308 void 3309 sdev_invalidate_nsmaps() 3310 { 3311 struct devname_nsmap *map = NULL; 3312 3313 ASSERT(MUTEX_HELD(&devname_nsmaps_lock)); 3314 3315 if (devname_nsmaps == NULL) 3316 return; 3317 3318 for (map = devname_nsmaps; map; map = map->next) { 3319 rw_enter(&map->dir_lock, RW_WRITER); 3320 map->dir_invalid = 1; 3321 rw_exit(&map->dir_lock); 3322 } 3323 devname_nsmaps_invalidated = 1; 3324 } 3325 3326 3327 int 3328 sdev_nsmaps_loaded() 3329 { 3330 int ret = 0; 3331 3332 mutex_enter(&devname_nsmaps_lock); 3333 if (devname_nsmaps_loaded) 3334 ret = 1; 3335 3336 mutex_exit(&devname_nsmaps_lock); 3337 return (ret); 3338 } 3339 3340 int 3341 sdev_nsmaps_reloaded() 3342 { 3343 int ret = 0; 3344 3345 mutex_enter(&devname_nsmaps_lock); 3346 if (devname_nsmaps_invalidated) 3347 ret = 1; 3348 3349 mutex_exit(&devname_nsmaps_lock); 3350 return (ret); 3351 } 3352 3353 static void 3354 sdev_free_nsmap(struct devname_nsmap *map) 3355 { 3356 ASSERT(map); 3357 if (map->dir_name) 3358 kmem_free(map->dir_name, strlen(map->dir_name) + 1); 3359 if (map->dir_module) 3360 kmem_free(map->dir_module, strlen(map->dir_module) + 1); 3361 if (map->dir_map) 3362 kmem_free(map->dir_map, strlen(map->dir_map) + 1); 3363 rw_destroy(&map->dir_lock); 3364 kmem_free(map, sizeof (*map)); 3365 } 3366 3367 void 3368 sdev_validate_nsmaps() 3369 { 3370 struct devname_nsmap *map = NULL; 3371 struct devname_nsmap *oldmap = NULL; 3372 3373 ASSERT(MUTEX_HELD(&devname_nsmaps_lock)); 3374 map = devname_nsmaps; 3375 while (map) { 3376 rw_enter(&map->dir_lock, RW_READER); 3377 if ((map->dir_invalid == 1) && (map->dir_newmodule == NULL) && 3378 (map->dir_newmap == NULL)) { 3379 oldmap = map; 3380 rw_exit(&map->dir_lock); 3381 if (map->prev) 3382 map->prev->next = oldmap->next; 3383 if (map == devname_nsmaps) 3384 devname_nsmaps = oldmap->next; 3385 3386 map = oldmap->next; 3387 if (map) 3388 map->prev = oldmap->prev; 3389 sdev_free_nsmap(oldmap); 3390 oldmap = NULL; 3391 } else { 3392 rw_exit(&map->dir_lock); 3393 map = map->next; 3394 } 3395 } 3396 devname_nsmaps_invalidated = 0; 3397 } 3398 3399 static int 3400 sdev_map_is_invalid(struct devname_nsmap *map) 3401 { 3402 int ret = 0; 3403 3404 ASSERT(map); 3405 rw_enter(&map->dir_lock, RW_READER); 3406 if (map->dir_invalid) 3407 ret = 1; 3408 rw_exit(&map->dir_lock); 3409 return (ret); 3410 } 3411 3412 static int 3413 sdev_check_map(struct devname_nsmap *map) 3414 { 3415 struct devname_nsmap *mapp; 3416 3417 mutex_enter(&devname_nsmaps_lock); 3418 if (devname_nsmaps == NULL) { 3419 mutex_exit(&devname_nsmaps_lock); 3420 return (1); 3421 } 3422 3423 for (mapp = devname_nsmaps; mapp; mapp = mapp->next) { 3424 if (mapp == map) { 3425 mutex_exit(&devname_nsmaps_lock); 3426 return (0); 3427 } 3428 } 3429 3430 mutex_exit(&devname_nsmaps_lock); 3431 return (1); 3432 3433 } 3434 3435 struct devname_nsmap * 3436 sdev_get_map(struct sdev_node *dv, int validate) 3437 { 3438 struct devname_nsmap *map; 3439 int error; 3440 3441 ASSERT(RW_READ_HELD(&dv->sdev_contents)); 3442 map = dv->sdev_mapinfo; 3443 if (map && sdev_check_map(map)) { 3444 if (!rw_tryupgrade(&dv->sdev_contents)) { 3445 rw_exit(&dv->sdev_contents); 3446 rw_enter(&dv->sdev_contents, RW_WRITER); 3447 } 3448 dv->sdev_mapinfo = NULL; 3449 rw_downgrade(&dv->sdev_contents); 3450 return (NULL); 3451 } 3452 3453 if (validate && (!map || (map && sdev_map_is_invalid(map)))) { 3454 if (!rw_tryupgrade(&dv->sdev_contents)) { 3455 rw_exit(&dv->sdev_contents); 3456 rw_enter(&dv->sdev_contents, RW_WRITER); 3457 } 3458 error = sdev_get_moduleops(dv); 3459 if (!error) 3460 map = dv->sdev_mapinfo; 3461 rw_downgrade(&dv->sdev_contents); 3462 } 3463 return (map); 3464 } 3465 3466 void 3467 sdev_handle_alloc(struct sdev_node *dv) 3468 { 3469 rw_enter(&dv->sdev_contents, RW_WRITER); 3470 dv->sdev_handle.dh_data = dv; 3471 rw_exit(&dv->sdev_contents); 3472 } 3473 3474 3475 extern int sdev_vnodeops_tbl_size; 3476 3477 /* 3478 * construct a new template with overrides from vtab 3479 */ 3480 static fs_operation_def_t * 3481 sdev_merge_vtab(const fs_operation_def_t tab[]) 3482 { 3483 fs_operation_def_t *new; 3484 const fs_operation_def_t *tab_entry; 3485 3486 /* make a copy of standard vnode ops table */ 3487 new = kmem_alloc(sdev_vnodeops_tbl_size, KM_SLEEP); 3488 bcopy((void *)sdev_vnodeops_tbl, new, sdev_vnodeops_tbl_size); 3489 3490 /* replace the overrides from tab */ 3491 for (tab_entry = tab; tab_entry->name != NULL; tab_entry++) { 3492 fs_operation_def_t *std_entry = new; 3493 while (std_entry->name) { 3494 if (strcmp(tab_entry->name, std_entry->name) == 0) { 3495 std_entry->func = tab_entry->func; 3496 break; 3497 } 3498 std_entry++; 3499 } 3500 if (std_entry->name == NULL) 3501 cmn_err(CE_NOTE, "sdev_merge_vtab: entry %s unused.", 3502 tab_entry->name); 3503 } 3504 3505 return (new); 3506 } 3507 3508 /* free memory allocated by sdev_merge_vtab */ 3509 static void 3510 sdev_free_vtab(fs_operation_def_t *new) 3511 { 3512 kmem_free(new, sdev_vnodeops_tbl_size); 3513 } 3514 3515 void 3516 devname_get_vnode(devname_handle_t *hdl, vnode_t **vpp) 3517 { 3518 struct sdev_node *dv = hdl->dh_data; 3519 3520 ASSERT(dv); 3521 3522 rw_enter(&dv->sdev_contents, RW_READER); 3523 *vpp = SDEVTOV(dv); 3524 rw_exit(&dv->sdev_contents); 3525 } 3526 3527 int 3528 devname_get_path(devname_handle_t *hdl, char **path) 3529 { 3530 struct sdev_node *dv = hdl->dh_data; 3531 3532 ASSERT(dv); 3533 3534 rw_enter(&dv->sdev_contents, RW_READER); 3535 *path = dv->sdev_path; 3536 rw_exit(&dv->sdev_contents); 3537 return (0); 3538 } 3539 3540 int 3541 devname_get_name(devname_handle_t *hdl, char **entry) 3542 { 3543 struct sdev_node *dv = hdl->dh_data; 3544 3545 ASSERT(dv); 3546 rw_enter(&dv->sdev_contents, RW_READER); 3547 *entry = dv->sdev_name; 3548 rw_exit(&dv->sdev_contents); 3549 return (0); 3550 } 3551 3552 void 3553 devname_get_dir_vnode(devname_handle_t *hdl, vnode_t **vpp) 3554 { 3555 struct sdev_node *dv = hdl->dh_data->sdev_dotdot; 3556 3557 ASSERT(dv); 3558 3559 rw_enter(&dv->sdev_contents, RW_READER); 3560 *vpp = SDEVTOV(dv); 3561 rw_exit(&dv->sdev_contents); 3562 } 3563 3564 int 3565 devname_get_dir_path(devname_handle_t *hdl, char **path) 3566 { 3567 struct sdev_node *dv = hdl->dh_data->sdev_dotdot; 3568 3569 ASSERT(dv); 3570 rw_enter(&dv->sdev_contents, RW_READER); 3571 *path = dv->sdev_path; 3572 rw_exit(&dv->sdev_contents); 3573 return (0); 3574 } 3575 3576 int 3577 devname_get_dir_name(devname_handle_t *hdl, char **entry) 3578 { 3579 struct sdev_node *dv = hdl->dh_data->sdev_dotdot; 3580 3581 ASSERT(dv); 3582 rw_enter(&dv->sdev_contents, RW_READER); 3583 *entry = dv->sdev_name; 3584 rw_exit(&dv->sdev_contents); 3585 return (0); 3586 } 3587 3588 int 3589 devname_get_dir_nsmap(devname_handle_t *hdl, struct devname_nsmap **map) 3590 { 3591 struct sdev_node *dv = hdl->dh_data->sdev_dotdot; 3592 3593 ASSERT(dv); 3594 rw_enter(&dv->sdev_contents, RW_READER); 3595 *map = dv->sdev_mapinfo; 3596 rw_exit(&dv->sdev_contents); 3597 return (0); 3598 } 3599 3600 int 3601 devname_get_dir_handle(devname_handle_t *hdl, devname_handle_t **dir_hdl) 3602 { 3603 struct sdev_node *dv = hdl->dh_data->sdev_dotdot; 3604 3605 ASSERT(dv); 3606 rw_enter(&dv->sdev_contents, RW_READER); 3607 *dir_hdl = &(dv->sdev_handle); 3608 rw_exit(&dv->sdev_contents); 3609 return (0); 3610 } 3611 3612 void 3613 devname_set_nodetype(devname_handle_t *hdl, void *args, int spec) 3614 { 3615 struct sdev_node *dv = hdl->dh_data; 3616 3617 ASSERT(dv); 3618 rw_enter(&dv->sdev_contents, RW_WRITER); 3619 hdl->dh_spec = (devname_spec_t)spec; 3620 hdl->dh_args = (void *)i_ddi_strdup((char *)args, KM_SLEEP); 3621 rw_exit(&dv->sdev_contents); 3622 } 3623 3624 /* 3625 * a generic setattr() function 3626 * 3627 * note: flags only supports AT_UID and AT_GID. 3628 * Future enhancements can be done for other types, e.g. AT_MODE 3629 */ 3630 int 3631 devname_setattr_func(struct vnode *vp, struct vattr *vap, int flags, 3632 struct cred *cred, int (*callback)(struct sdev_node *, struct vattr *, 3633 int), int protocol) 3634 { 3635 struct sdev_node *dv = VTOSDEV(vp); 3636 struct sdev_node *parent = dv->sdev_dotdot; 3637 struct vattr *get; 3638 uint_t mask = vap->va_mask; 3639 int error; 3640 3641 /* some sanity checks */ 3642 if (vap->va_mask & AT_NOSET) 3643 return (EINVAL); 3644 3645 if (vap->va_mask & AT_SIZE) { 3646 if (vp->v_type == VDIR) { 3647 return (EISDIR); 3648 } 3649 } 3650 3651 /* no need to set attribute, but do not fail either */ 3652 ASSERT(parent); 3653 rw_enter(&parent->sdev_contents, RW_READER); 3654 if (dv->sdev_state == SDEV_ZOMBIE) { 3655 rw_exit(&parent->sdev_contents); 3656 return (0); 3657 } 3658 3659 /* If backing store exists, just set it. */ 3660 if (dv->sdev_attrvp) { 3661 rw_exit(&parent->sdev_contents); 3662 return (VOP_SETATTR(dv->sdev_attrvp, vap, flags, cred, NULL)); 3663 } 3664 3665 /* 3666 * Otherwise, for nodes with the persistence attribute, create it. 3667 */ 3668 ASSERT(dv->sdev_attr); 3669 if (SDEV_IS_PERSIST(dv) || 3670 ((vap->va_mask & ~AT_TIMES) != 0 && !SDEV_IS_DYNAMIC(dv))) { 3671 sdev_vattr_merge(dv, vap); 3672 rw_enter(&dv->sdev_contents, RW_WRITER); 3673 error = sdev_shadow_node(dv, cred); 3674 rw_exit(&dv->sdev_contents); 3675 rw_exit(&parent->sdev_contents); 3676 3677 if (error) 3678 return (error); 3679 return (VOP_SETATTR(dv->sdev_attrvp, vap, flags, cred, NULL)); 3680 } 3681 3682 3683 /* 3684 * sdev_attr was allocated in sdev_mknode 3685 */ 3686 rw_enter(&dv->sdev_contents, RW_WRITER); 3687 error = secpolicy_vnode_setattr(cred, vp, vap, dv->sdev_attr, 3688 flags, sdev_unlocked_access, dv); 3689 if (error) { 3690 rw_exit(&dv->sdev_contents); 3691 rw_exit(&parent->sdev_contents); 3692 return (error); 3693 } 3694 3695 get = dv->sdev_attr; 3696 if (mask & AT_MODE) { 3697 get->va_mode &= S_IFMT; 3698 get->va_mode |= vap->va_mode & ~S_IFMT; 3699 } 3700 3701 if ((mask & AT_UID) || (mask & AT_GID)) { 3702 if (mask & AT_UID) 3703 get->va_uid = vap->va_uid; 3704 if (mask & AT_GID) 3705 get->va_gid = vap->va_gid; 3706 /* 3707 * a callback must be provided if the protocol is set 3708 */ 3709 if ((protocol & AT_UID) || (protocol & AT_GID)) { 3710 ASSERT(callback); 3711 error = callback(dv, get, protocol); 3712 if (error) { 3713 rw_exit(&dv->sdev_contents); 3714 rw_exit(&parent->sdev_contents); 3715 return (error); 3716 } 3717 } 3718 } 3719 3720 if (mask & AT_ATIME) 3721 get->va_atime = vap->va_atime; 3722 if (mask & AT_MTIME) 3723 get->va_mtime = vap->va_mtime; 3724 if (mask & (AT_MODE | AT_UID | AT_GID | AT_CTIME)) { 3725 gethrestime(&get->va_ctime); 3726 } 3727 3728 sdev_vattr_merge(dv, get); 3729 rw_exit(&dv->sdev_contents); 3730 rw_exit(&parent->sdev_contents); 3731 return (0); 3732 } 3733