?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
9075 Improve ZFS pool import/load process and corrupted pool recovery
[unleashed.git] / usr / src / uts / common / fs / devfs / devfs_subr.c
blob3407d31a3abd582c1169c3ff78336d2e0f681ec3
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 (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
23 */
24
25 /*
26 * miscellaneous routines for the devfs
27 */
28
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/t_lock.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/user.h>
35 #include <sys/time.h>
36 #include <sys/vfs.h>
37 #include <sys/vnode.h>
38 #include <sys/file.h>
39 #include <sys/fcntl.h>
40 #include <sys/flock.h>
41 #include <sys/kmem.h>
42 #include <sys/uio.h>
43 #include <sys/errno.h>
44 #include <sys/stat.h>
45 #include <sys/cred.h>
46 #include <sys/dirent.h>
47 #include <sys/pathname.h>
48 #include <sys/cmn_err.h>
49 #include <sys/debug.h>
50 #include <sys/modctl.h>
51 #include <fs/fs_subr.h>
52 #include <sys/fs/dv_node.h>
53 #include <sys/fs/snode.h>
54 #include <sys/sunndi.h>
55 #include <sys/sunmdi.h>
56 #include <sys/conf.h>
57
58 #ifdef DEBUG
59 int devfs_debug = 0x0;
60 #endif
61
62 const char dvnm[] = "devfs";
63 kmem_cache_t *dv_node_cache; /* dv_node cache */
64
65 /*
66 * The devfs_clean_key is taken during a devfs_clean operation: it is used to
67 * prevent unnecessary code execution and for detection of potential deadlocks.
68 */
69 uint_t devfs_clean_key;
70
71 struct dv_node *dvroot;
72
73 /* prototype memory vattrs */
74 vattr_t dv_vattr_dir = {
75 AT_TYPE|AT_MODE|AT_UID|AT_GID, /* va_mask */
76 VDIR, /* va_type */
77 DV_DIRMODE_DEFAULT, /* va_mode */
78 DV_UID_DEFAULT, /* va_uid */
79 DV_GID_DEFAULT, /* va_gid */
80 0, /* va_fsid; */
81 0, /* va_nodeid; */
82 0, /* va_nlink; */
83 0, /* va_size; */
84 0, /* va_atime; */
85 0, /* va_mtime; */
86 0, /* va_ctime; */
87 0, /* va_rdev; */
88 0, /* va_blksize; */
89 0, /* va_nblocks; */
90 0, /* va_seq; */
91 };
92
93 vattr_t dv_vattr_file = {
94 AT_TYPE|AT_MODE|AT_SIZE|AT_UID|AT_GID|AT_RDEV, /* va_mask */
95 0, /* va_type */
96 DV_DEVMODE_DEFAULT, /* va_mode */
97 DV_UID_DEFAULT, /* va_uid */
98 DV_GID_DEFAULT, /* va_gid */
99 0, /* va_fsid; */
100 0, /* va_nodeid; */
101 0, /* va_nlink; */
102 0, /* va_size; */
103 0, /* va_atime; */
104 0, /* va_mtime; */
105 0, /* va_ctime; */
106 0, /* va_rdev; */
107 0, /* va_blksize; */
108 0, /* va_nblocks; */
109 0, /* va_seq; */
110 };
111
112 vattr_t dv_vattr_priv = {
113 AT_TYPE|AT_MODE|AT_SIZE|AT_UID|AT_GID|AT_RDEV, /* va_mask */
114 0, /* va_type */
115 DV_DEVMODE_PRIV, /* va_mode */
116 DV_UID_DEFAULT, /* va_uid */
117 DV_GID_DEFAULT, /* va_gid */
118 0, /* va_fsid; */
119 0, /* va_nodeid; */
120 0, /* va_nlink; */
121 0, /* va_size; */
122 0, /* va_atime; */
123 0, /* va_mtime; */
124 0, /* va_ctime; */
125 0, /* va_rdev; */
126 0, /* va_blksize; */
127 0, /* va_nblocks; */
128 0, /* va_seq; */
129 };
130
131 extern dev_info_t *clone_dip;
132 extern major_t clone_major;
133 extern struct dev_ops *ddi_hold_driver(major_t);
134
135 /* dv_node node constructor for kmem cache */
136 static int
137 i_dv_node_ctor(void *buf, void *cfarg, int flag)
138 {
139 _NOTE(ARGUNUSED(cfarg, flag))
140 struct dv_node *dv = (struct dv_node *)buf;
141 struct vnode *vp;
142
143 bzero(buf, sizeof (struct dv_node));
144 vp = dv->dv_vnode = vn_alloc(flag);
145 if (vp == NULL) {
146 return (-1);
147 }
148 vp->v_data = dv;
149 rw_init(&dv->dv_contents, NULL, RW_DEFAULT, NULL);
150 return (0);
151 }
152
153 /* dv_node node destructor for kmem cache */
154 static void
155 i_dv_node_dtor(void *buf, void *arg)
156 {
157 _NOTE(ARGUNUSED(arg))
158 struct dv_node *dv = (struct dv_node *)buf;
159 struct vnode *vp = DVTOV(dv);
160
161 rw_destroy(&dv->dv_contents);
162 vn_invalid(vp);
163 vn_free(vp);
164 }
165
166
167 /* initialize dv_node node cache */
168 void
169 dv_node_cache_init()
170 {
171 ASSERT(dv_node_cache == NULL);
172 dv_node_cache = kmem_cache_create("dv_node_cache",
173 sizeof (struct dv_node), 0, i_dv_node_ctor, i_dv_node_dtor,
174 NULL, NULL, NULL, 0);
175
176 tsd_create(&devfs_clean_key, NULL);
177 }
178
179 /* destroy dv_node node cache */
180 void
181 dv_node_cache_fini()
182 {
183 ASSERT(dv_node_cache != NULL);
184 kmem_cache_destroy(dv_node_cache);
185 dv_node_cache = NULL;
186
187 tsd_destroy(&devfs_clean_key);
188 }
189
190 /*
191 * dv_mkino - Generate a unique inode number for devfs nodes.
192 *
193 * Although ino_t is 64 bits, the inode number is truncated to 32 bits for 32
194 * bit non-LARGEFILE applications. This means that there is a requirement to
195 * maintain the inode number as a 32 bit value or applications will have
196 * stat(2) calls fail with EOVERFLOW. We form a 32 bit inode number from the
197 * dev_t. but if the minor number is larger than L_MAXMIN32 we fold extra minor
198 *
199 * To generate inode numbers for directories, we assume that we will never use
200 * more than half the major space - this allows for ~8190 drivers. We use this
201 * upper major number space to allocate inode numbers for directories by
202 * encoding the major and instance into this space.
203 *
204 * We also skew the result so that inode 2 is reserved for the root of the file
205 * system.
206 *
207 * As part of the future support for 64-bit dev_t APIs, the upper minor bits
208 * should be folded into the high inode bits by adding the following code
209 * after "ino |= 1":
210 *
211 * #if (L_BITSMINOR32 != L_BITSMINOR)
212 * |* fold overflow minor bits into high bits of inode number *|
213 * ino |= ((ino_t)(minor >> L_BITSMINOR32)) << L_BITSMINOR;
214 * #endif |* (L_BITSMINOR32 != L_BITSMINOR) *|
215 *
216 * This way only applications that use devices that overflow their minor
217 * space will have an application level impact.
218 */
219 static ino_t
220 dv_mkino(dev_info_t *devi, vtype_t typ, dev_t dev)
221 {
222 major_t major;
223 minor_t minor;
224 ino_t ino;
225 static int warn;
226
227 if (typ == VDIR) {
228 major = ((L_MAXMAJ32 + 1) >> 1) + DEVI(devi)->devi_major;
229 minor = ddi_get_instance(devi);
230
231 /* makedevice32 in high half of major number space */
232 ino = (ino_t)((major << L_BITSMINOR32) | (minor & L_MAXMIN32));
233
234 major = DEVI(devi)->devi_major;
235 } else {
236 major = getmajor(dev);
237 minor = getminor(dev);
238
239 /* makedevice32 */
240 ino = (ino_t)((major << L_BITSMINOR32) | (minor & L_MAXMIN32));
241
242 /* make ino for VCHR different than VBLK */
243 ino <<= 1;
244 if (typ == VCHR)
245 ino |= 1;
246 }
247
248 ino += DV_ROOTINO + 1; /* skew */
249
250 /*
251 * diagnose things a little early because adding the skew to a large
252 * minor number could roll over the major.
253 */
254 if ((major >= (L_MAXMAJ32 >> 1)) && (warn == 0)) {
255 warn = 1;
256 cmn_err(CE_WARN, "%s: inode numbers are not unique", dvnm);
257 }
258
259 return (ino);
260 }
261
262 /*
263 * Compare two nodes lexographically to balance avl tree
264 */
265 static int
266 dv_compare_nodes(const struct dv_node *dv1, const struct dv_node *dv2)
267 {
268 int rv;
269
270 if ((rv = strcmp(dv1->dv_name, dv2->dv_name)) == 0)
271 return (0);
272 return ((rv < 0) ? -1 : 1);
273 }
274
275 /*
276 * dv_mkroot
277 *
278 * Build the first VDIR dv_node.
279 */
280 struct dv_node *
281 dv_mkroot(struct vfs *vfsp, dev_t devfsdev)
282 {
283 struct dv_node *dv;
284 struct vnode *vp;
285
286 ASSERT(ddi_root_node() != NULL);
287 ASSERT(dv_node_cache != NULL);
288
289 dcmn_err3(("dv_mkroot\n"));
290 dv = kmem_cache_alloc(dv_node_cache, KM_SLEEP);
291 vp = DVTOV(dv);
292 vn_reinit(vp);
293 vp->v_flag = VROOT;
294 vp->v_vfsp = vfsp;
295 vp->v_type = VDIR;
296 vp->v_rdev = devfsdev;
297 vn_setops(vp, dv_vnodeops);
298 vn_exists(vp);
299
300 dvroot = dv;
301
302 dv->dv_name = NULL; /* not needed */
303 dv->dv_namelen = 0;
304
305 dv->dv_devi = ddi_root_node();
306
307 dv->dv_ino = DV_ROOTINO;
308 dv->dv_nlink = 2; /* name + . (no dv_insert) */
309 dv->dv_dotdot = dv; /* .. == self */
310 dv->dv_attrvp = NULLVP;
311 dv->dv_attr = NULL;
312 dv->dv_flags = DV_BUILD;
313 dv->dv_priv = NULL;
314 dv->dv_busy = 0;
315 dv->dv_dflt_mode = 0;
316
317 avl_create(&dv->dv_entries,
318 (int (*)(const void *, const void *))dv_compare_nodes,
319 sizeof (struct dv_node), offsetof(struct dv_node, dv_avllink));
320
321 return (dv);
322 }
323
324 /*
325 * dv_mkdir
326 *
327 * Given an probed or attached nexus node, create a VDIR dv_node.
328 * No dv_attrvp is created at this point.
329 */
330 struct dv_node *
331 dv_mkdir(struct dv_node *ddv, dev_info_t *devi, char *nm)
332 {
333 struct dv_node *dv;
334 struct vnode *vp;
335 size_t nmlen;
336
337 ASSERT((devi));
338 dcmn_err4(("dv_mkdir: %s\n", nm));
339
340 dv = kmem_cache_alloc(dv_node_cache, KM_SLEEP);
341 nmlen = strlen(nm) + 1;
342 dv->dv_name = kmem_alloc(nmlen, KM_SLEEP);
343 bcopy(nm, dv->dv_name, nmlen);
344 dv->dv_namelen = nmlen - 1; /* '\0' not included */
345
346 vp = DVTOV(dv);
347 vn_reinit(vp);
348 vp->v_flag = 0;
349 vp->v_vfsp = DVTOV(ddv)->v_vfsp;
350 vp->v_type = VDIR;
351 vp->v_rdev = DVTOV(ddv)->v_rdev;
352 vn_setops(vp, vn_getops(DVTOV(ddv)));
353 vn_exists(vp);
354
355 dv->dv_devi = devi;
356 ndi_hold_devi(devi);
357
358 dv->dv_ino = dv_mkino(devi, VDIR, NODEV);
359 dv->dv_nlink = 0; /* updated on insert */
360 dv->dv_dotdot = ddv;
361 dv->dv_attrvp = NULLVP;
362 dv->dv_attr = NULL;
363 dv->dv_flags = DV_BUILD;
364 dv->dv_priv = NULL;
365 dv->dv_busy = 0;
366 dv->dv_dflt_mode = 0;
367
368 avl_create(&dv->dv_entries,
369 (int (*)(const void *, const void *))dv_compare_nodes,
370 sizeof (struct dv_node), offsetof(struct dv_node, dv_avllink));
371
372 return (dv);
373 }
374
375 /*
376 * dv_mknod
377 *
378 * Given a minor node, create a VCHR or VBLK dv_node.
379 * No dv_attrvp is created at this point.
380 */
381 static struct dv_node *
382 dv_mknod(struct dv_node *ddv, dev_info_t *devi, char *nm,
383 struct ddi_minor_data *dmd)
384 {
385 struct dv_node *dv;
386 struct vnode *vp;
387 size_t nmlen;
388
389 dcmn_err4(("dv_mknod: %s\n", nm));
390
391 dv = kmem_cache_alloc(dv_node_cache, KM_SLEEP);
392 nmlen = strlen(nm) + 1;
393 dv->dv_name = kmem_alloc(nmlen, KM_SLEEP);
394 bcopy(nm, dv->dv_name, nmlen);
395 dv->dv_namelen = nmlen - 1; /* no '\0' */
396
397 vp = DVTOV(dv);
398 vn_reinit(vp);
399 vp->v_flag = 0;
400 vp->v_vfsp = DVTOV(ddv)->v_vfsp;
401 vp->v_type = dmd->ddm_spec_type == S_IFCHR ? VCHR : VBLK;
402 vp->v_rdev = dmd->ddm_dev;
403 vn_setops(vp, vn_getops(DVTOV(ddv)));
404 vn_exists(vp);
405
406 /* increment dev_ref with devi_lock held */
407 ASSERT(DEVI_BUSY_OWNED(devi));
408 mutex_enter(&DEVI(devi)->devi_lock);
409 dv->dv_devi = devi;
410 DEVI(devi)->devi_ref++; /* ndi_hold_devi(dip) */
411 mutex_exit(&DEVI(devi)->devi_lock);
412
413 dv->dv_ino = dv_mkino(devi, vp->v_type, vp->v_rdev);
414 dv->dv_nlink = 0; /* updated on insert */
415 dv->dv_dotdot = ddv;
416 dv->dv_attrvp = NULLVP;
417 dv->dv_attr = NULL;
418 dv->dv_flags = 0;
419
420 if (dmd->type == DDM_INTERNAL_PATH)
421 dv->dv_flags |= DV_INTERNAL;
422 if (dmd->ddm_flags & DM_NO_FSPERM)
423 dv->dv_flags |= DV_NO_FSPERM;
424
425 dv->dv_priv = dmd->ddm_node_priv;
426 if (dv->dv_priv)
427 dphold(dv->dv_priv);
428
429 /*
430 * Minors created with ddi_create_priv_minor_node can specify
431 * a default mode permission other than the devfs default.
432 */
433 if (dv->dv_priv || dv->dv_flags & DV_NO_FSPERM) {
434 dcmn_err5(("%s: dv_mknod default priv mode 0%o\n",
435 dv->dv_name, dmd->ddm_priv_mode));
436 dv->dv_flags |= DV_DFLT_MODE;
437 dv->dv_dflt_mode = dmd->ddm_priv_mode & S_IAMB;
438 }
439
440 return (dv);
441 }
442
443 /*
444 * dv_destroy
445 *
446 * Destroy what we created in dv_mkdir or dv_mknod.
447 * In the case of a *referenced* directory, do nothing.
448 */
449 void
450 dv_destroy(struct dv_node *dv, uint_t flags)
451 {
452 vnode_t *vp = DVTOV(dv);
453 ASSERT(dv->dv_nlink == 0); /* no references */
454
455 dcmn_err4(("dv_destroy: %s\n", dv->dv_name));
456
457 /*
458 * We may be asked to unlink referenced directories.
459 * In this case, there is nothing to be done.
460 * The eventual memory free will be done in
461 * devfs_inactive.
462 */
463 if (vp->v_count != 0) {
464 ASSERT(vp->v_type == VDIR);
465 ASSERT(flags & DV_CLEAN_FORCE);
466 ASSERT(DV_STALE(dv));
467 return;
468 }
469
470 if (vp->v_type == VDIR) {
471 ASSERT(DV_FIRST_ENTRY(dv) == NULL);
472 avl_destroy(&dv->dv_entries);
473 }
474
475 if (dv->dv_attrvp != NULLVP)
476 VN_RELE(dv->dv_attrvp);
477 if (dv->dv_attr != NULL)
478 kmem_free(dv->dv_attr, sizeof (struct vattr));
479 if (dv->dv_name != NULL)
480 kmem_free(dv->dv_name, dv->dv_namelen + 1);
481 if (dv->dv_devi != NULL) {
482 ndi_rele_devi(dv->dv_devi);
483 }
484 if (dv->dv_priv != NULL) {
485 dpfree(dv->dv_priv);
486 }
487
488 kmem_cache_free(dv_node_cache, dv);
489 }
490
491 /*
492 * Find and hold dv_node by name
493 */
494 static struct dv_node *
495 dv_findbyname(struct dv_node *ddv, char *nm)
496 {
497 struct dv_node *dv;
498 avl_index_t where;
499 struct dv_node dvtmp;
500
501 ASSERT(RW_LOCK_HELD(&ddv->dv_contents));
502 dcmn_err3(("dv_findbyname: %s\n", nm));
503
504 dvtmp.dv_name = nm;
505 dv = avl_find(&ddv->dv_entries, &dvtmp, &where);
506 if (dv) {
507 ASSERT(dv->dv_dotdot == ddv);
508 ASSERT(strcmp(dv->dv_name, nm) == 0);
509 VN_HOLD(DVTOV(dv));
510 return (dv);
511 }
512 return (NULL);
513 }
514
515 /*
516 * Inserts a new dv_node in a parent directory
517 */
518 void
519 dv_insert(struct dv_node *ddv, struct dv_node *dv)
520 {
521 avl_index_t where;
522
523 ASSERT(RW_WRITE_HELD(&ddv->dv_contents));
524 ASSERT(DVTOV(ddv)->v_type == VDIR);
525 ASSERT(ddv->dv_nlink >= 2);
526 ASSERT(dv->dv_nlink == 0);
527
528 dcmn_err3(("dv_insert: %s\n", dv->dv_name));
529
530 dv->dv_dotdot = ddv;
531 if (DVTOV(dv)->v_type == VDIR) {
532 ddv->dv_nlink++; /* .. to containing directory */
533 dv->dv_nlink = 2; /* name + . */
534 } else {
535 dv->dv_nlink = 1; /* name */
536 }
537
538 /* enter node in the avl tree */
539 VERIFY(avl_find(&ddv->dv_entries, dv, &where) == NULL);
540 avl_insert(&ddv->dv_entries, dv, where);
541 }
542
543 /*
544 * Unlink a dv_node from a perent directory
545 */
546 void
547 dv_unlink(struct dv_node *ddv, struct dv_node *dv)
548 {
549 /* verify linkage of arguments */
550 ASSERT(ddv && dv);
551 ASSERT(dv->dv_dotdot == ddv);
552 ASSERT(RW_WRITE_HELD(&ddv->dv_contents));
553 ASSERT(DVTOV(ddv)->v_type == VDIR);
554
555 dcmn_err3(("dv_unlink: %s\n", dv->dv_name));
556
557 if (DVTOV(dv)->v_type == VDIR) {
558 ddv->dv_nlink--; /* .. to containing directory */
559 dv->dv_nlink -= 2; /* name + . */
560 } else {
561 dv->dv_nlink -= 1; /* name */
562 }
563 ASSERT(ddv->dv_nlink >= 2);
564 ASSERT(dv->dv_nlink == 0);
565
566 dv->dv_dotdot = NULL;
567
568 /* remove from avl tree */
569 avl_remove(&ddv->dv_entries, dv);
570 }
571
572 /*
573 * Merge devfs node specific information into an attribute structure.
574 *
575 * NOTE: specfs provides ATIME,MTIME,CTIME,SIZE,BLKSIZE,NBLOCKS on leaf node.
576 */
577 void
578 dv_vattr_merge(struct dv_node *dv, struct vattr *vap)
579 {
580 struct vnode *vp = DVTOV(dv);
581
582 vap->va_nodeid = dv->dv_ino;
583 vap->va_nlink = dv->dv_nlink;
584
585 if (vp->v_type == VDIR) {
586 vap->va_rdev = 0;
587 vap->va_fsid = vp->v_rdev;
588 } else {
589 vap->va_rdev = vp->v_rdev;
590 vap->va_fsid = DVTOV(dv->dv_dotdot)->v_rdev;
591 vap->va_type = vp->v_type;
592 /* don't trust the shadow file type */
593 vap->va_mode &= ~S_IFMT;
594 if (vap->va_type == VCHR)
595 vap->va_mode |= S_IFCHR;
596 else
597 vap->va_mode |= S_IFBLK;
598 }
599 }
600
601 /*
602 * Get default device permission by consulting rules in
603 * privilege specification in minor node and /etc/minor_perm.
604 *
605 * This function is called from the devname filesystem to get default
606 * permissions for a device exported to a non-global zone.
607 */
608 void
609 devfs_get_defattr(struct vnode *vp, struct vattr *vap, int *no_fs_perm)
610 {
611 mperm_t mp;
612 struct dv_node *dv;
613
614 /* If vp isn't a dv_node, return something sensible */
615 if (!vn_matchops(vp, dv_vnodeops)) {
616 if (no_fs_perm)
617 *no_fs_perm = 0;
618 *vap = dv_vattr_file;
619 return;
620 }
621
622 /*
623 * For minors not created by ddi_create_priv_minor_node(),
624 * use devfs defaults.
625 */
626 dv = VTODV(vp);
627 if (vp->v_type == VDIR) {
628 *vap = dv_vattr_dir;
629 } else if (dv->dv_flags & DV_NO_FSPERM) {
630 if (no_fs_perm)
631 *no_fs_perm = 1;
632 *vap = dv_vattr_priv;
633 } else {
634 /*
635 * look up perm bits from minor_perm
636 */
637 *vap = dv_vattr_file;
638 if (dev_minorperm(dv->dv_devi, dv->dv_name, &mp) == 0) {
639 VATTR_MP_MERGE((*vap), mp);
640 dcmn_err5(("%s: minor perm mode 0%o\n",
641 dv->dv_name, vap->va_mode));
642 } else if (dv->dv_flags & DV_DFLT_MODE) {
643 ASSERT((dv->dv_dflt_mode & ~S_IAMB) == 0);
644 vap->va_mode &= ~S_IAMB;
645 vap->va_mode |= dv->dv_dflt_mode;
646 dcmn_err5(("%s: priv mode 0%o\n",
647 dv->dv_name, vap->va_mode));
648 }
649 }
650 }
651
652 /*
653 * dv_shadow_node
654 *
655 * Given a VDIR dv_node, find/create the associated VDIR
656 * node in the shadow attribute filesystem.
657 *
658 * Given a VCHR/VBLK dv_node, find the associated VREG
659 * node in the shadow attribute filesystem. These nodes
660 * are only created to persist non-default attributes.
661 * Lack of such a node implies the default permissions
662 * are sufficient.
663 *
664 * Managing the attribute file entries is slightly tricky (mostly
665 * because we can't intercept VN_HOLD and VN_RELE except on the last
666 * release).
667 *
668 * We assert that if the dv_attrvp pointer is non-NULL, it points
669 * to a singly-held (by us) vnode that represents the shadow entry
670 * in the underlying filesystem. To avoid store-ordering issues,
671 * we assert that the pointer can only be tested under the dv_contents
672 * READERS lock.
673 */
674
675 void
676 dv_shadow_node(
677 struct vnode *dvp, /* devfs parent directory vnode */
678 char *nm, /* name component */
679 struct vnode *vp, /* devfs vnode */
680 struct pathname *pnp, /* the path .. */
681 struct vnode *rdir, /* the root .. */
682 struct cred *cred, /* who's asking? */
683 int flags) /* optionally create shadow node */
684 {
685 struct dv_node *dv; /* dv_node of named directory */
686 struct vnode *rdvp; /* shadow parent directory vnode */
687 struct vnode *rvp; /* shadow vnode */
688 struct vnode *rrvp; /* realvp of shadow vnode */
689 struct vattr vattr;
690 int create_tried;
691 int error;
692
693 ASSERT(vp->v_type == VDIR || vp->v_type == VCHR || vp->v_type == VBLK);
694 dv = VTODV(vp);
695 dcmn_err3(("dv_shadow_node: name %s attr %p\n",
696 nm, (void *)dv->dv_attrvp));
697
698 if ((flags & DV_SHADOW_WRITE_HELD) == 0) {
699 ASSERT(RW_READ_HELD(&dv->dv_contents));
700 if (dv->dv_attrvp != NULLVP)
701 return;
702 if (!rw_tryupgrade(&dv->dv_contents)) {
703 rw_exit(&dv->dv_contents);
704 rw_enter(&dv->dv_contents, RW_WRITER);
705 if (dv->dv_attrvp != NULLVP) {
706 rw_downgrade(&dv->dv_contents);
707 return;
708 }
709 }
710 } else {
711 ASSERT(RW_WRITE_HELD(&dv->dv_contents));
712 if (dv->dv_attrvp != NULLVP)
713 return;
714 }
715
716 ASSERT(RW_WRITE_HELD(&dv->dv_contents) && dv->dv_attrvp == NULL);
717
718 rdvp = VTODV(dvp)->dv_attrvp;
719 create_tried = 0;
720 lookup:
721 if (rdvp && (dv->dv_flags & DV_NO_FSPERM) == 0) {
722 error = VOP_LOOKUP(rdvp, nm, &rvp, pnp, LOOKUP_DIR, rdir, cred,
723 NULL, NULL, NULL);
724
725 /* factor out the snode since we only want the attribute node */
726 if ((error == 0) && (VOP_REALVP(rvp, &rrvp, NULL) == 0)) {
727 VN_HOLD(rrvp);
728 VN_RELE(rvp);
729 rvp = rrvp;
730 }
731 } else
732 error = EROFS; /* no parent, no entry */
733
734 /*
735 * All we want is the permissions (and maybe ACLs and
736 * extended attributes), and we want to perform lookups
737 * by name. Drivers occasionally change their minor
738 * number space. If something changes, there's no
739 * much we can do about it here.
740 */
741
742 /* The shadow node checks out. We are done */
743 if (error == 0) {
744 dv->dv_attrvp = rvp; /* with one hold */
745
746 /*
747 * Determine if we have non-trivial ACLs on this node.
748 * It is not necessary to VOP_RWLOCK since fs_acl_nontrivial
749 * only does VOP_GETSECATTR.
750 */
751 dv->dv_flags &= ~DV_ACL;
752
753 if (fs_acl_nontrivial(rvp, cred))
754 dv->dv_flags |= DV_ACL;
755
756 /*
757 * If we have synced out the memory attributes, free
758 * them and switch back to using the persistent store.
759 */
760 if (rvp && dv->dv_attr) {
761 kmem_free(dv->dv_attr, sizeof (struct vattr));
762 dv->dv_attr = NULL;
763 }
764 if ((flags & DV_SHADOW_WRITE_HELD) == 0)
765 rw_downgrade(&dv->dv_contents);
766 ASSERT(RW_LOCK_HELD(&dv->dv_contents));
767 return;
768 }
769
770 /*
771 * Failed to find attribute in persistent backing store,
772 * get default permission bits.
773 */
774 devfs_get_defattr(vp, &vattr, NULL);
775
776 dv_vattr_merge(dv, &vattr);
777 gethrestime(&vattr.va_atime);
778 vattr.va_mtime = vattr.va_atime;
779 vattr.va_ctime = vattr.va_atime;
780
781 /*
782 * Try to create shadow dir. This is necessary in case
783 * we need to create a shadow leaf node later, when user
784 * executes chmod.
785 */
786 if ((error == ENOENT) && !create_tried) {
787 switch (vp->v_type) {
788 case VDIR:
789 error = VOP_MKDIR(rdvp, nm, &vattr, &rvp, kcred,
790 NULL, 0, NULL);
791 dsysdebug(error, ("vop_mkdir %s %s %d\n",
792 VTODV(dvp)->dv_name, nm, error));
793 create_tried = 1;
794 break;
795
796 case VCHR:
797 case VBLK:
798 /*
799 * Shadow nodes are only created on demand
800 */
801 if (flags & DV_SHADOW_CREATE) {
802 error = VOP_CREATE(rdvp, nm, &vattr, NONEXCL,
803 VREAD|VWRITE, &rvp, kcred, 0, NULL, NULL);
804 dsysdebug(error, ("vop_create %s %s %d\n",
805 VTODV(dvp)->dv_name, nm, error));
806 create_tried = 1;
807 }
808 break;
809
810 default:
811 cmn_err(CE_PANIC, "devfs: %s: create", dvnm);
812 /*NOTREACHED*/
813 }
814
815 if (create_tried &&
816 (error == 0) || (error == EEXIST)) {
817 VN_RELE(rvp);
818 goto lookup;
819 }
820 }
821
822 /* Store attribute in memory */
823 if (dv->dv_attr == NULL) {
824 dv->dv_attr = kmem_alloc(sizeof (struct vattr), KM_SLEEP);
825 *(dv->dv_attr) = vattr;
826 }
827
828 if ((flags & DV_SHADOW_WRITE_HELD) == 0)
829 rw_downgrade(&dv->dv_contents);
830 ASSERT(RW_LOCK_HELD(&dv->dv_contents));
831 }
832
833 /*
834 * Given a devinfo node, and a name, returns the appropriate
835 * minor information for that named node, if it exists.
836 */
837 static int
838 dv_find_leafnode(dev_info_t *devi, char *minor_nm, struct ddi_minor_data *r_mi)
839 {
840 struct ddi_minor_data *dmd;
841
842 ASSERT(i_ddi_devi_attached(devi));
843
844 dcmn_err3(("dv_find_leafnode: %s\n", minor_nm));
845 ASSERT(DEVI_BUSY_OWNED(devi));
846 for (dmd = DEVI(devi)->devi_minor; dmd; dmd = dmd->next) {
847
848 /*
849 * Skip alias nodes and nodes without a name.
850 */
851 if ((dmd->type == DDM_ALIAS) || (dmd->ddm_name == NULL))
852 continue;
853
854 dcmn_err4(("dv_find_leafnode: (%s,%s)\n",
855 minor_nm, dmd->ddm_name));
856 if (strcmp(minor_nm, dmd->ddm_name) == 0) {
857 r_mi->ddm_dev = dmd->ddm_dev;
858 r_mi->ddm_spec_type = dmd->ddm_spec_type;
859 r_mi->type = dmd->type;
860 r_mi->ddm_flags = dmd->ddm_flags;
861 r_mi->ddm_node_priv = dmd->ddm_node_priv;
862 r_mi->ddm_priv_mode = dmd->ddm_priv_mode;
863 if (r_mi->ddm_node_priv)
864 dphold(r_mi->ddm_node_priv);
865 return (0);
866 }
867 }
868
869 dcmn_err3(("dv_find_leafnode: %s: ENOENT\n", minor_nm));
870 return (ENOENT);
871 }
872
873 /*
874 * Special handling for clone node:
875 * Clone minor name is a driver name, the minor number will
876 * be the major number of the driver. There is no minor
877 * node under the clone driver, so we'll manufacture the
878 * dev_t.
879 */
880 static struct dv_node *
881 dv_clone_mknod(struct dv_node *ddv, char *drvname)
882 {
883 major_t major;
884 struct dv_node *dvp;
885 char *devnm;
886 struct ddi_minor_data *dmd;
887
888 /*
889 * Make sure drvname is a STREAMS driver. We load the driver,
890 * but don't attach to any instances. This makes stat(2)
891 * relatively cheap.
892 */
893 major = ddi_name_to_major(drvname);
894 if (major == DDI_MAJOR_T_NONE)
895 return (NULL);
896
897 if (ddi_hold_driver(major) == NULL)
898 return (NULL);
899
900 if (STREAMSTAB(major) == NULL) {
901 ddi_rele_driver(major);
902 return (NULL);
903 }
904
905 ddi_rele_driver(major);
906 devnm = kmem_alloc(MAXNAMELEN, KM_SLEEP);
907 (void) snprintf(devnm, MAXNAMELEN, "clone@0:%s", drvname);
908 dmd = kmem_zalloc(sizeof (*dmd), KM_SLEEP);
909 dmd->ddm_dev = makedevice(clone_major, (minor_t)major);
910 dmd->ddm_spec_type = S_IFCHR;
911 dvp = dv_mknod(ddv, clone_dip, devnm, dmd);
912 kmem_free(dmd, sizeof (*dmd));
913 kmem_free(devnm, MAXNAMELEN);
914 return (dvp);
915 }
916
917 /*
918 * Given the parent directory node, and a name in it, returns the
919 * named dv_node to the caller (as a vnode).
920 *
921 * (We need pnp and rdir for doing shadow lookups; they can be NULL)
922 */
923 int
924 dv_find(struct dv_node *ddv, char *nm, struct vnode **vpp, struct pathname *pnp,
925 struct vnode *rdir, struct cred *cred, uint_t ndi_flags)
926 {
927 extern int isminiroot; /* see modctl.c */
928
929 int circ;
930 int rv = 0, was_busy = 0, nmlen, write_held = 0;
931 struct vnode *vp;
932 struct dv_node *dv, *dup;
933 dev_info_t *pdevi, *devi = NULL;
934 char *mnm;
935 struct ddi_minor_data *dmd;
936
937 dcmn_err3(("dv_find %s\n", nm));
938
939 if (!rw_tryenter(&ddv->dv_contents, RW_READER)) {
940 if (tsd_get(devfs_clean_key))
941 return (EBUSY);
942 rw_enter(&ddv->dv_contents, RW_READER);
943 }
944 start:
945 if (DV_STALE(ddv)) {
946 rw_exit(&ddv->dv_contents);
947 return (ESTALE);
948 }
949
950 /*
951 * Empty name or ., return node itself.
952 */
953 nmlen = strlen(nm);
954 if ((nmlen == 0) || ((nmlen == 1) && (nm[0] == '.'))) {
955 *vpp = DVTOV(ddv);
956 rw_exit(&ddv->dv_contents);
957 VN_HOLD(*vpp);
958 return (0);
959 }
960
961 /*
962 * .., return the parent directory
963 */
964 if ((nmlen == 2) && (strcmp(nm, "..") == 0)) {
965 *vpp = DVTOV(ddv->dv_dotdot);
966 rw_exit(&ddv->dv_contents);
967 VN_HOLD(*vpp);
968 return (0);
969 }
970
971 /*
972 * Fail anything without a valid device name component
973 */
974 if (nm[0] == '@' || nm[0] == ':') {
975 dcmn_err3(("devfs: no driver '%s'\n", nm));
976 rw_exit(&ddv->dv_contents);
977 return (ENOENT);
978 }
979
980 /*
981 * So, now we have to deal with the trickier stuff.
982 *
983 * (a) search the existing list of dv_nodes on this directory
984 */
985 if ((dv = dv_findbyname(ddv, nm)) != NULL) {
986 founddv:
987 ASSERT(RW_LOCK_HELD(&ddv->dv_contents));
988
989 if (!rw_tryenter(&dv->dv_contents, RW_READER)) {
990 if (tsd_get(devfs_clean_key)) {
991 VN_RELE(DVTOV(dv));
992 rw_exit(&ddv->dv_contents);
993 return (EBUSY);
994 }
995 rw_enter(&dv->dv_contents, RW_READER);
996 }
997
998 vp = DVTOV(dv);
999 if ((dv->dv_attrvp != NULLVP) ||
1000 (vp->v_type != VDIR && dv->dv_attr != NULL)) {
1001 /*
1002 * Common case - we already have attributes
1003 */
1004 rw_exit(&dv->dv_contents);
1005 rw_exit(&ddv->dv_contents);
1006 goto found;
1007 }
1008
1009 /*
1010 * No attribute vp, try and build one.
1011 *
1012 * dv_shadow_node() can briefly drop &dv->dv_contents lock
1013 * if it is unable to upgrade it to a write lock. If the
1014 * current thread has come in through the bottom-up device
1015 * configuration devfs_clean() path, we may deadlock against
1016 * a thread performing top-down device configuration if it
1017 * grabs the contents lock. To avoid this, when we are on the
1018 * devfs_clean() path we attempt to upgrade the dv_contents
1019 * lock before we call dv_shadow_node().
1020 */
1021 if (tsd_get(devfs_clean_key)) {
1022 if (!rw_tryupgrade(&dv->dv_contents)) {
1023 VN_RELE(DVTOV(dv));
1024 rw_exit(&dv->dv_contents);
1025 rw_exit(&ddv->dv_contents);
1026 return (EBUSY);
1027 }
1028
1029 write_held = DV_SHADOW_WRITE_HELD;
1030 }
1031
1032 dv_shadow_node(DVTOV(ddv), nm, vp, pnp, rdir, cred,
1033 write_held);
1034
1035 rw_exit(&dv->dv_contents);
1036 rw_exit(&ddv->dv_contents);
1037 goto found;
1038 }
1039
1040 /*
1041 * (b) Search the child devinfo nodes of our parent directory,
1042 * looking for the named node. If we find it, build a new
1043 * node, then grab the writers lock, search the directory
1044 * if it's still not there, then insert it.
1045 *
1046 * We drop the devfs locks before accessing the device tree.
1047 * Take care to mark the node BUSY so that a forced devfs_clean
1048 * doesn't mark the directory node stale.
1049 *
1050 * Also, check if we are called as part of devfs_clean or
1051 * reset_perm. If so, simply return not found because there
1052 * is nothing to clean.
1053 */
1054 if (tsd_get(devfs_clean_key)) {
1055 rw_exit(&ddv->dv_contents);
1056 return (ENOENT);
1057 }
1058
1059 /*
1060 * We could be either READ or WRITE locked at
1061 * this point. Upgrade if we are read locked.
1062 */
1063 ASSERT(RW_LOCK_HELD(&ddv->dv_contents));
1064 if (rw_read_locked(&ddv->dv_contents) &&
1065 !rw_tryupgrade(&ddv->dv_contents)) {
1066 rw_exit(&ddv->dv_contents);
1067 rw_enter(&ddv->dv_contents, RW_WRITER);
1068 /*
1069 * Things may have changed when we dropped
1070 * the contents lock, so start from top again
1071 */
1072 goto start;
1073 }
1074 ddv->dv_busy++; /* mark busy before dropping lock */
1075 was_busy++;
1076 rw_exit(&ddv->dv_contents);
1077
1078 pdevi = ddv->dv_devi;
1079 ASSERT(pdevi != NULL);
1080
1081 mnm = strchr(nm, ':');
1082 if (mnm)
1083 *mnm = (char)0;
1084
1085 /*
1086 * Configure one nexus child, will call nexus's bus_ops
1087 * If successful, devi is held upon returning.
1088 * Note: devfs lookup should not be configuring grandchildren.
1089 */
1090 ASSERT((ndi_flags & NDI_CONFIG) == 0);
1091
1092 rv = ndi_devi_config_one(pdevi, nm, &devi, ndi_flags | NDI_NO_EVENT);
1093 if (mnm)
1094 *mnm = ':';
1095 if (rv != NDI_SUCCESS) {
1096 rv = ENOENT;
1097 goto notfound;
1098 }
1099
1100 ASSERT(devi);
1101
1102 /* Check if this is a path alias */
1103 if (ddi_aliases_present == B_TRUE && ddi_get_parent(devi) != pdevi) {
1104 char *curr = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1105
1106 (void) ddi_pathname(devi, curr);
1107
1108 vp = NULL;
1109 if (devfs_lookupname(curr, NULL, &vp) == 0 && vp) {
1110 dv = VTODV(vp);
1111 kmem_free(curr, MAXPATHLEN);
1112 goto found;
1113 }
1114 kmem_free(curr, MAXPATHLEN);
1115 }
1116
1117 /*
1118 * If we configured a hidden node, consider it notfound.
1119 */
1120 if (ndi_dev_is_hidden_node(devi)) {
1121 ndi_rele_devi(devi);
1122 rv = ENOENT;
1123 goto notfound;
1124 }
1125
1126 /*
1127 * Don't make vhci clients visible under phci, unless we
1128 * are in miniroot.
1129 */
1130 if (isminiroot == 0 && ddi_get_parent(devi) != pdevi) {
1131 ndi_rele_devi(devi);
1132 rv = ENOENT;
1133 goto notfound;
1134 }
1135
1136 ASSERT(devi && i_ddi_devi_attached(devi));
1137
1138 /*
1139 * Invalidate cache to notice newly created minor nodes.
1140 */
1141 rw_enter(&ddv->dv_contents, RW_WRITER);
1142 ddv->dv_flags |= DV_BUILD;
1143 rw_exit(&ddv->dv_contents);
1144
1145 /*
1146 * mkdir for nexus drivers and leaf nodes as well. If we are racing
1147 * and create a duplicate, the duplicate will be destroyed below.
1148 */
1149 if (mnm == NULL) {
1150 dv = dv_mkdir(ddv, devi, nm);
1151 } else {
1152 /*
1153 * Allocate dmd first to avoid KM_SLEEP with active
1154 * ndi_devi_enter.
1155 */
1156 dmd = kmem_zalloc(sizeof (*dmd), KM_SLEEP);
1157 ndi_devi_enter(devi, &circ);
1158 if (devi == clone_dip) {
1159 /*
1160 * For clone minors, load the driver indicated by
1161 * minor name.
1162 */
1163 dv = dv_clone_mknod(ddv, mnm + 1);
1164 } else {
1165 /*
1166 * Find minor node and make a dv_node
1167 */
1168 if (dv_find_leafnode(devi, mnm + 1, dmd) == 0) {
1169 dv = dv_mknod(ddv, devi, nm, dmd);
1170 if (dmd->ddm_node_priv)
1171 dpfree(dmd->ddm_node_priv);
1172 }
1173 }
1174 ndi_devi_exit(devi, circ);
1175 kmem_free(dmd, sizeof (*dmd));
1176 }
1177 /*
1178 * Release hold from ndi_devi_config_one()
1179 */
1180 ndi_rele_devi(devi);
1181
1182 if (dv == NULL) {
1183 rv = ENOENT;
1184 goto notfound;
1185 }
1186
1187 /*
1188 * We have released the dv_contents lock, need to check
1189 * if another thread already created a duplicate node
1190 */
1191 rw_enter(&ddv->dv_contents, RW_WRITER);
1192 if ((dup = dv_findbyname(ddv, nm)) == NULL) {
1193 dv_insert(ddv, dv);
1194 } else {
1195 /*
1196 * Duplicate found, use the existing node
1197 */
1198 VN_RELE(DVTOV(dv));
1199 dv_destroy(dv, 0);
1200 dv = dup;
1201 }
1202 goto founddv;
1203 /*NOTREACHED*/
1204
1205 found:
1206 /*
1207 * Fail lookup of device that has now become hidden (typically via
1208 * hot removal of open device).
1209 */
1210 if (dv->dv_devi && ndi_dev_is_hidden_node(dv->dv_devi)) {
1211 dcmn_err2(("dv_find: nm %s failed: hidden/removed\n", nm));
1212 VN_RELE(vp);
1213 rv = ENOENT;
1214 goto notfound;
1215 }
1216
1217 /*
1218 * Skip non-kernel lookups of internal nodes.
1219 * This use of kcred to distinguish between user and
1220 * internal kernel lookups is unfortunate. The information
1221 * provided by the seg argument to lookupnameat should
1222 * evolve into a lookup flag for filesystems that need
1223 * this distinction.
1224 */
1225 if ((dv->dv_flags & DV_INTERNAL) && (cred != kcred)) {
1226 dcmn_err2(("dv_find: nm %s failed: internal\n", nm));
1227 VN_RELE(vp);
1228 rv = ENOENT;
1229 goto notfound;
1230 }
1231
1232 dcmn_err2(("dv_find: returning vp for nm %s\n", nm));
1233 if (vp->v_type == VCHR || vp->v_type == VBLK) {
1234 /*
1235 * If vnode is a device, return special vnode instead
1236 * (though it knows all about -us- via sp->s_realvp,
1237 * sp->s_devvp, and sp->s_dip)
1238 */
1239 *vpp = specvp_devfs(vp, vp->v_rdev, vp->v_type, cred,
1240 dv->dv_devi);
1241 VN_RELE(vp);
1242 if (*vpp == NULLVP)
1243 rv = ENOSYS;
1244 } else
1245 *vpp = vp;
1246
1247 notfound:
1248 if (was_busy) {
1249 /*
1250 * Non-zero was_busy tells us that we are not in the
1251 * devfs_clean() path which in turn means that we can afford
1252 * to take the contents lock unconditionally.
1253 */
1254 rw_enter(&ddv->dv_contents, RW_WRITER);
1255 ddv->dv_busy--;
1256 rw_exit(&ddv->dv_contents);
1257 }
1258 return (rv);
1259 }
1260
1261 /*
1262 * The given directory node is out-of-date; that is, it has been
1263 * marked as needing to be rebuilt, possibly because some new devinfo
1264 * node has come into existence, or possibly because this is the first
1265 * time we've been here.
1266 */
1267 void
1268 dv_filldir(struct dv_node *ddv)
1269 {
1270 struct dv_node *dv;
1271 dev_info_t *devi, *pdevi;
1272 struct ddi_minor_data *dmd;
1273 char devnm[MAXNAMELEN];
1274 int circ, ccirc;
1275
1276 ASSERT(DVTOV(ddv)->v_type == VDIR);
1277 ASSERT(RW_WRITE_HELD(&ddv->dv_contents));
1278 ASSERT(ddv->dv_flags & DV_BUILD);
1279
1280 dcmn_err3(("dv_filldir: %s\n", ddv->dv_name));
1281 if (DV_STALE(ddv))
1282 return;
1283 pdevi = ddv->dv_devi;
1284
1285 if (ndi_devi_config(pdevi, NDI_NO_EVENT) != NDI_SUCCESS) {
1286 dcmn_err3(("dv_filldir: config error %s\n", ddv->dv_name));
1287 }
1288
1289 ndi_devi_enter(pdevi, &circ);
1290 for (devi = ddi_get_child(pdevi); devi;
1291 devi = ddi_get_next_sibling(devi)) {
1292 /*
1293 * While we know enough to create a directory at DS_INITIALIZED,
1294 * the directory will be empty until DS_ATTACHED. The existence
1295 * of an empty directory dv_node will cause a devi_ref, which
1296 * has caused problems for existing code paths doing offline/DR
1297 * type operations - making devfs_clean coordination even more
1298 * sensitive and error prone. Given this, the 'continue' below
1299 * is checking for DS_ATTACHED instead of DS_INITIALIZED.
1300 */
1301 if (i_ddi_node_state(devi) < DS_ATTACHED)
1302 continue;
1303
1304 /* skip hidden nodes */
1305 if (ndi_dev_is_hidden_node(devi))
1306 continue;
1307
1308 dcmn_err3(("dv_filldir: node %s\n", ddi_node_name(devi)));
1309
1310 ndi_devi_enter(devi, &ccirc);
1311 for (dmd = DEVI(devi)->devi_minor; dmd; dmd = dmd->next) {
1312 char *addr;
1313
1314 /*
1315 * Skip alias nodes, internal nodes, and nodes
1316 * without a name. We allow DDM_DEFAULT nodes
1317 * to appear in readdir.
1318 */
1319 if ((dmd->type == DDM_ALIAS) ||
1320 (dmd->type == DDM_INTERNAL_PATH) ||
1321 (dmd->ddm_name == NULL))
1322 continue;
1323
1324 addr = ddi_get_name_addr(devi);
1325 if (addr && *addr)
1326 (void) sprintf(devnm, "%s@%s:%s",
1327 ddi_node_name(devi), addr, dmd->ddm_name);
1328 else
1329 (void) sprintf(devnm, "%s:%s",
1330 ddi_node_name(devi), dmd->ddm_name);
1331
1332 if ((dv = dv_findbyname(ddv, devnm)) != NULL) {
1333 /* dv_node already exists */
1334 VN_RELE(DVTOV(dv));
1335 continue;
1336 }
1337
1338 dv = dv_mknod(ddv, devi, devnm, dmd);
1339 dv_insert(ddv, dv);
1340 VN_RELE(DVTOV(dv));
1341 }
1342 ndi_devi_exit(devi, ccirc);
1343
1344 (void) ddi_deviname(devi, devnm);
1345 if ((dv = dv_findbyname(ddv, devnm + 1)) == NULL) {
1346 /* directory doesn't exist */
1347 dv = dv_mkdir(ddv, devi, devnm + 1);
1348 dv_insert(ddv, dv);
1349 }
1350 VN_RELE(DVTOV(dv));
1351 }
1352 ndi_devi_exit(pdevi, circ);
1353
1354 ddv->dv_flags &= ~DV_BUILD;
1355 }
1356
1357 /*
1358 * Given a directory node, clean out all the nodes beneath.
1359 *
1360 * VDIR: Reinvoke to clean them, then delete the directory.
1361 * VCHR, VBLK: Just blow them away.
1362 *
1363 * Mark the directories touched as in need of a rebuild, in case
1364 * we fall over part way through. When DV_CLEAN_FORCE is specified,
1365 * we mark referenced empty directories as stale to facilitate DR.
1366 */
1367 int
1368 dv_cleandir(struct dv_node *ddv, char *devnm, uint_t flags)
1369 {
1370 struct dv_node *dv;
1371 struct dv_node *next;
1372 struct vnode *vp;
1373 int busy = 0;
1374
1375 /*
1376 * We should always be holding the tsd_clean_key here: dv_cleandir()
1377 * will be called as a result of a devfs_clean request and the
1378 * tsd_clean_key will be set in either in devfs_clean() itself or in
1379 * devfs_clean_vhci().
1380 *
1381 * Since we are on the devfs_clean path, we return EBUSY if we cannot
1382 * get the contents lock: if we blocked here we might deadlock against
1383 * a thread performing top-down device configuration.
1384 */
1385 ASSERT(tsd_get(devfs_clean_key));
1386
1387 dcmn_err3(("dv_cleandir: %s\n", ddv->dv_name));
1388
1389 if (!(flags & DV_CLEANDIR_LCK) &&
1390 !rw_tryenter(&ddv->dv_contents, RW_WRITER))
1391 return (EBUSY);
1392
1393 for (dv = DV_FIRST_ENTRY(ddv); dv; dv = next) {
1394 next = DV_NEXT_ENTRY(ddv, dv);
1395
1396 /*
1397 * If devnm is specified, the non-minor portion of the
1398 * name must match devnm.
1399 */
1400 if (devnm &&
1401 (strncmp(devnm, dv->dv_name, strlen(devnm)) ||
1402 (dv->dv_name[strlen(devnm)] != ':' &&
1403 dv->dv_name[strlen(devnm)] != '\0')))
1404 continue;
1405
1406 /* check type of what we are cleaning */
1407 vp = DVTOV(dv);
1408 if (vp->v_type == VDIR) {
1409 /* recurse on directories */
1410 rw_enter(&dv->dv_contents, RW_WRITER);
1411 if (dv_cleandir(dv, NULL,
1412 flags | DV_CLEANDIR_LCK) == EBUSY) {
1413 rw_exit(&dv->dv_contents);
1414 goto set_busy;
1415 }
1416
1417 /* A clean directory is an empty directory... */
1418 ASSERT(dv->dv_nlink == 2);
1419 mutex_enter(&vp->v_lock);
1420 if (vp->v_count > 0) {
1421 /*
1422 * ... but an empty directory can still have
1423 * references to it. If we have dv_busy or
1424 * DV_CLEAN_FORCE is *not* specified then a
1425 * referenced directory is considered busy.
1426 */
1427 if (dv->dv_busy || !(flags & DV_CLEAN_FORCE)) {
1428 mutex_exit(&vp->v_lock);
1429 rw_exit(&dv->dv_contents);
1430 goto set_busy;
1431 }
1432
1433 /*
1434 * Mark referenced directory stale so that DR
1435 * will succeed even if a shell has
1436 * /devices/xxx as current directory (causing
1437 * VN_HOLD reference to an empty directory).
1438 */
1439 ASSERT(!DV_STALE(dv));
1440 ndi_rele_devi(dv->dv_devi);
1441 dv->dv_devi = NULL; /* mark DV_STALE */
1442 }
1443 } else {
1444 ASSERT((vp->v_type == VCHR) || (vp->v_type == VBLK));
1445 ASSERT(dv->dv_nlink == 1); /* no hard links */
1446 mutex_enter(&vp->v_lock);
1447 if (vp->v_count > 0) {
1448 mutex_exit(&vp->v_lock);
1449 goto set_busy;
1450 }
1451 }
1452
1453 /* unlink from directory */
1454 dv_unlink(ddv, dv);
1455
1456 /* drop locks */
1457 mutex_exit(&vp->v_lock);
1458 if (vp->v_type == VDIR)
1459 rw_exit(&dv->dv_contents);
1460
1461 /* destroy vnode if ref count is zero */
1462 if (vp->v_count == 0)
1463 dv_destroy(dv, flags);
1464
1465 continue;
1466
1467 /*
1468 * If devnm is not NULL we return immediately on busy,
1469 * otherwise we continue destroying unused dv_node's.
1470 */
1471 set_busy: busy++;
1472 if (devnm)
1473 break;
1474 }
1475
1476 /*
1477 * This code may be invoked to inform devfs that a new node has
1478 * been created in the kernel device tree. So we always set
1479 * the DV_BUILD flag to allow the next dv_filldir() to pick
1480 * the new devinfo nodes.
1481 */
1482 ddv->dv_flags |= DV_BUILD;
1483
1484 if (!(flags & DV_CLEANDIR_LCK))
1485 rw_exit(&ddv->dv_contents);
1486
1487 return (busy ? EBUSY : 0);
1488 }
1489
1490 /*
1491 * Walk through the devfs hierarchy, correcting the permissions of
1492 * devices with default permissions that do not match those specified
1493 * by minor perm. This can only be done for all drivers for now.
1494 */
1495 static int
1496 dv_reset_perm_dir(struct dv_node *ddv, uint_t flags)
1497 {
1498 struct dv_node *dv;
1499 struct vnode *vp;
1500 int retval = 0;
1501 struct vattr *attrp;
1502 mperm_t mp;
1503 char *nm;
1504 uid_t old_uid;
1505 gid_t old_gid;
1506 mode_t old_mode;
1507
1508 rw_enter(&ddv->dv_contents, RW_WRITER);
1509 for (dv = DV_FIRST_ENTRY(ddv); dv; dv = DV_NEXT_ENTRY(ddv, dv)) {
1510 int error = 0;
1511 nm = dv->dv_name;
1512
1513 rw_enter(&dv->dv_contents, RW_READER);
1514 vp = DVTOV(dv);
1515 if (vp->v_type == VDIR) {
1516 rw_exit(&dv->dv_contents);
1517 if (dv_reset_perm_dir(dv, flags) != 0) {
1518 error = EBUSY;
1519 }
1520 } else {
1521 ASSERT(vp->v_type == VCHR || vp->v_type == VBLK);
1522
1523 /*
1524 * Check for permissions from minor_perm
1525 * If there are none, we're done
1526 */
1527 rw_exit(&dv->dv_contents);
1528 if (dev_minorperm(dv->dv_devi, nm, &mp) != 0)
1529 continue;
1530
1531 rw_enter(&dv->dv_contents, RW_READER);
1532
1533 /*
1534 * Allow a node's permissions to be altered
1535 * permanently from the defaults by chmod,
1536 * using the shadow node as backing store.
1537 * Otherwise, update node to minor_perm permissions.
1538 */
1539 if (dv->dv_attrvp == NULLVP) {
1540 /*
1541 * No attribute vp, try to find one.
1542 */
1543 dv_shadow_node(DVTOV(ddv), nm, vp,
1544 NULL, NULLVP, kcred, 0);
1545 }
1546 if (dv->dv_attrvp != NULLVP || dv->dv_attr == NULL) {
1547 rw_exit(&dv->dv_contents);
1548 continue;
1549 }
1550
1551 attrp = dv->dv_attr;
1552
1553 if (VATTRP_MP_CMP(attrp, mp) == 0) {
1554 dcmn_err5(("%s: no perm change: "
1555 "%d %d 0%o\n", nm, attrp->va_uid,
1556 attrp->va_gid, attrp->va_mode));
1557 rw_exit(&dv->dv_contents);
1558 continue;
1559 }
1560
1561 old_uid = attrp->va_uid;
1562 old_gid = attrp->va_gid;
1563 old_mode = attrp->va_mode;
1564
1565 VATTRP_MP_MERGE(attrp, mp);
1566 mutex_enter(&vp->v_lock);
1567 if (vp->v_count > 0) {
1568 error = EBUSY;
1569 }
1570 mutex_exit(&vp->v_lock);
1571
1572 dcmn_err5(("%s: perm %d/%d/0%o -> %d/%d/0%o (%d)\n",
1573 nm, old_uid, old_gid, old_mode, attrp->va_uid,
1574 attrp->va_gid, attrp->va_mode, error));
1575
1576 rw_exit(&dv->dv_contents);
1577 }
1578
1579 if (error != 0) {
1580 retval = error;
1581 }
1582 }
1583
1584 ddv->dv_flags |= DV_BUILD;
1585
1586 rw_exit(&ddv->dv_contents);
1587
1588 return (retval);
1589 }
1590
1591 int
1592 devfs_reset_perm(uint_t flags)
1593 {
1594 struct dv_node *dvp;
1595 int rval;
1596
1597 if ((dvp = devfs_dip_to_dvnode(ddi_root_node())) == NULL)
1598 return (0);
1599
1600 VN_HOLD(DVTOV(dvp));
1601 rval = dv_reset_perm_dir(dvp, flags);
1602 VN_RELE(DVTOV(dvp));
1603 return (rval);
1604 }
1605
1606 /*
1607 * Clean up dangling devfs shadow nodes for removed
1608 * drivers so that, in the event the driver is re-added
1609 * to the system, newly created nodes won't incorrectly
1610 * pick up these stale shadow node permissions.
1611 *
1612 * This is accomplished by walking down the pathname
1613 * to the directory, starting at the root's attribute
1614 * node, then removing all minors matching the specified
1615 * node name. Care must be taken to remove all entries
1616 * in a directory before the directory itself, so that
1617 * the clean-up associated with rem_drv'ing a nexus driver
1618 * does not inadvertently result in an inconsistent
1619 * filesystem underlying devfs.
1620 */
1621
1622 static int
1623 devfs_remdrv_rmdir(vnode_t *dirvp, const char *dir, vnode_t *rvp)
1624 {
1625 int error;
1626 vnode_t *vp;
1627 int eof;
1628 struct iovec iov;
1629 struct uio uio;
1630 struct dirent64 *dp;
1631 dirent64_t *dbuf;
1632 size_t dlen;
1633 size_t dbuflen;
1634 int ndirents = 64;
1635 char *nm;
1636
1637 VN_HOLD(dirvp);
1638
1639 dlen = ndirents * (sizeof (*dbuf));
1640 dbuf = kmem_alloc(dlen, KM_SLEEP);
1641
1642 uio.uio_iov = &iov;
1643 uio.uio_iovcnt = 1;
1644 uio.uio_segflg = UIO_SYSSPACE;
1645 uio.uio_fmode = 0;
1646 uio.uio_extflg = UIO_COPY_CACHED;
1647 uio.uio_loffset = 0;
1648 uio.uio_llimit = MAXOFFSET_T;
1649
1650 eof = 0;
1651 error = 0;
1652 while (!error && !eof) {
1653 uio.uio_resid = dlen;
1654 iov.iov_base = (char *)dbuf;
1655 iov.iov_len = dlen;
1656
1657 (void) VOP_RWLOCK(dirvp, V_WRITELOCK_FALSE, NULL);
1658 error = VOP_READDIR(dirvp, &uio, kcred, &eof, NULL, 0);
1659 VOP_RWUNLOCK(dirvp, V_WRITELOCK_FALSE, NULL);
1660
1661 dbuflen = dlen - uio.uio_resid;
1662
1663 if (error || dbuflen == 0)
1664 break;
1665
1666 for (dp = dbuf; ((intptr_t)dp < (intptr_t)dbuf + dbuflen);
1667 dp = (dirent64_t *)((intptr_t)dp + dp->d_reclen)) {
1668
1669 nm = dp->d_name;
1670
1671 if (strcmp(nm, ".") == 0 || strcmp(nm, "..") == 0)
1672 continue;
1673
1674 error = VOP_LOOKUP(dirvp, nm,
1675 &vp, NULL, 0, NULL, kcred, NULL, NULL, NULL);
1676
1677 dsysdebug(error,
1678 ("rem_drv %s/%s lookup (%d)\n",
1679 dir, nm, error));
1680
1681 if (error)
1682 continue;
1683
1684 ASSERT(vp->v_type == VDIR ||
1685 vp->v_type == VCHR || vp->v_type == VBLK);
1686
1687 if (vp->v_type == VDIR) {
1688 error = devfs_remdrv_rmdir(vp, nm, rvp);
1689 if (error == 0) {
1690 error = VOP_RMDIR(dirvp,
1691 (char *)nm, rvp, kcred, NULL, 0);
1692 dsysdebug(error,
1693 ("rem_drv %s/%s rmdir (%d)\n",
1694 dir, nm, error));
1695 }
1696 } else {
1697 error = VOP_REMOVE(dirvp, (char *)nm, kcred,
1698 NULL, 0);
1699 dsysdebug(error,
1700 ("rem_drv %s/%s remove (%d)\n",
1701 dir, nm, error));
1702 }
1703
1704 VN_RELE(vp);
1705 if (error) {
1706 goto exit;
1707 }
1708 }
1709 }
1710
1711 exit:
1712 VN_RELE(dirvp);
1713 kmem_free(dbuf, dlen);
1714
1715 return (error);
1716 }
1717
1718 int
1719 devfs_remdrv_cleanup(const char *dir, const char *nodename)
1720 {
1721 int error;
1722 vnode_t *vp;
1723 vnode_t *dirvp;
1724 int eof;
1725 struct iovec iov;
1726 struct uio uio;
1727 struct dirent64 *dp;
1728 dirent64_t *dbuf;
1729 size_t dlen;
1730 size_t dbuflen;
1731 int ndirents = 64;
1732 int nodenamelen = strlen(nodename);
1733 char *nm;
1734 struct pathname pn;
1735 vnode_t *rvp; /* root node of the underlying attribute fs */
1736
1737 dcmn_err5(("devfs_remdrv_cleanup: %s %s\n", dir, nodename));
1738
1739 if (error = pn_get((char *)dir, UIO_SYSSPACE, &pn))
1740 return (0);
1741
1742 rvp = dvroot->dv_attrvp;
1743 ASSERT(rvp != NULL);
1744 VN_HOLD(rvp);
1745
1746 pn_skipslash(&pn);
1747 dirvp = rvp;
1748 VN_HOLD(dirvp);
1749
1750 nm = kmem_alloc(MAXNAMELEN, KM_SLEEP);
1751
1752 while (pn_pathleft(&pn)) {
1753 ASSERT(dirvp->v_type == VDIR);
1754 (void) pn_getcomponent(&pn, nm);
1755 ASSERT((strcmp(nm, ".") != 0) && (strcmp(nm, "..") != 0));
1756 error = VOP_LOOKUP(dirvp, nm, &vp, NULL, 0, rvp, kcred,
1757 NULL, NULL, NULL);
1758 if (error) {
1759 dcmn_err5(("remdrv_cleanup %s lookup error %d\n",
1760 nm, error));
1761 VN_RELE(dirvp);
1762 if (dirvp != rvp)
1763 VN_RELE(rvp);
1764 pn_free(&pn);
1765 kmem_free(nm, MAXNAMELEN);
1766 return (0);
1767 }
1768 VN_RELE(dirvp);
1769 dirvp = vp;
1770 pn_skipslash(&pn);
1771 }
1772
1773 ASSERT(dirvp->v_type == VDIR);
1774 if (dirvp != rvp)
1775 VN_RELE(rvp);
1776 pn_free(&pn);
1777 kmem_free(nm, MAXNAMELEN);
1778
1779 dlen = ndirents * (sizeof (*dbuf));
1780 dbuf = kmem_alloc(dlen, KM_SLEEP);
1781
1782 uio.uio_iov = &iov;
1783 uio.uio_iovcnt = 1;
1784 uio.uio_segflg = UIO_SYSSPACE;
1785 uio.uio_fmode = 0;
1786 uio.uio_extflg = UIO_COPY_CACHED;
1787 uio.uio_loffset = 0;
1788 uio.uio_llimit = MAXOFFSET_T;
1789
1790 eof = 0;
1791 error = 0;
1792 while (!error && !eof) {
1793 uio.uio_resid = dlen;
1794 iov.iov_base = (char *)dbuf;
1795 iov.iov_len = dlen;
1796
1797 (void) VOP_RWLOCK(dirvp, V_WRITELOCK_FALSE, NULL);
1798 error = VOP_READDIR(dirvp, &uio, kcred, &eof, NULL, 0);
1799 VOP_RWUNLOCK(dirvp, V_WRITELOCK_FALSE, NULL);
1800
1801 dbuflen = dlen - uio.uio_resid;
1802
1803 if (error || dbuflen == 0)
1804 break;
1805
1806 for (dp = dbuf; ((intptr_t)dp < (intptr_t)dbuf + dbuflen);
1807 dp = (dirent64_t *)((intptr_t)dp + dp->d_reclen)) {
1808
1809 nm = dp->d_name;
1810
1811 if (strcmp(nm, ".") == 0 || strcmp(nm, "..") == 0)
1812 continue;
1813
1814 if (strncmp(nm, nodename, nodenamelen) != 0)
1815 continue;
1816
1817 error = VOP_LOOKUP(dirvp, nm, &vp,
1818 NULL, 0, NULL, kcred, NULL, NULL, NULL);
1819
1820 dsysdebug(error,
1821 ("rem_drv %s/%s lookup (%d)\n",
1822 dir, nm, error));
1823
1824 if (error)
1825 continue;
1826
1827 ASSERT(vp->v_type == VDIR ||
1828 vp->v_type == VCHR || vp->v_type == VBLK);
1829
1830 if (vp->v_type == VDIR) {
1831 error = devfs_remdrv_rmdir(vp, nm, rvp);
1832 if (error == 0) {
1833 error = VOP_RMDIR(dirvp, (char *)nm,
1834 rvp, kcred, NULL, 0);
1835 dsysdebug(error,
1836 ("rem_drv %s/%s rmdir (%d)\n",
1837 dir, nm, error));
1838 }
1839 } else {
1840 error = VOP_REMOVE(dirvp, (char *)nm, kcred,
1841 NULL, 0);
1842 dsysdebug(error,
1843 ("rem_drv %s/%s remove (%d)\n",
1844 dir, nm, error));
1845 }
1846
1847 VN_RELE(vp);
1848 if (error)
1849 goto exit;
1850 }
1851 }
1852
1853 exit:
1854 VN_RELE(dirvp);
1855
1856 kmem_free(dbuf, dlen);
1857
1858 return (0);
1859 }
1860
1861 struct dv_list {
1862 struct dv_node *dv;
1863 struct dv_list *next;
1864 };
1865
1866 void
1867 dv_walk(
1868 struct dv_node *ddv,
1869 char *devnm,
1870 void (*callback)(struct dv_node *, void *),
1871 void *arg)
1872 {
1873 struct vnode *dvp;
1874 struct dv_node *dv;
1875 struct dv_list *head, *tail, *next;
1876 int len;
1877
1878 dcmn_err3(("dv_walk: ddv = %s, devnm = %s\n",
1879 ddv->dv_name, devnm ? devnm : "<null>"));
1880
1881 dvp = DVTOV(ddv);
1882
1883 ASSERT(dvp->v_type == VDIR);
1884
1885 head = tail = next = NULL;
1886
1887 rw_enter(&ddv->dv_contents, RW_READER);
1888 mutex_enter(&dvp->v_lock);
1889 for (dv = DV_FIRST_ENTRY(ddv); dv; dv = DV_NEXT_ENTRY(ddv, dv)) {
1890 /*
1891 * If devnm is not NULL and is not the empty string,
1892 * select only dv_nodes with matching non-minor name
1893 */
1894 if (devnm && (len = strlen(devnm)) &&
1895 (strncmp(devnm, dv->dv_name, len) ||
1896 (dv->dv_name[len] != ':' && dv->dv_name[len] != '\0')))
1897 continue;
1898
1899 callback(dv, arg);
1900
1901 if (DVTOV(dv)->v_type != VDIR)
1902 continue;
1903
1904 next = kmem_zalloc(sizeof (*next), KM_SLEEP);
1905 next->dv = dv;
1906
1907 if (tail)
1908 tail->next = next;
1909 else
1910 head = next;
1911
1912 tail = next;
1913 }
1914
1915 while (head) {
1916 dv_walk(head->dv, NULL, callback, arg);
1917 next = head->next;
1918 kmem_free(head, sizeof (*head));
1919 head = next;
1920 }
1921 rw_exit(&ddv->dv_contents);
1922 mutex_exit(&dvp->v_lock);
1923 }
Unleashed OS