2 * Copyright (c) 2002 Dima Dorfman.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * DEVFS ruleset implementation.
32 * A note on terminology: To "run" a rule on a dirent is to take the
33 * prescribed action; to "apply" a rule is to check whether it matches
34 * a dirent and run if if it does.
36 * A note on locking: Only foreign entry points (non-static functions)
37 * should deal with locking. Everything else assumes we already hold
38 * the required kind of lock.
40 * A note on namespace: devfs_rules_* are the non-static functions for
41 * the entire "ruleset" subsystem, devfs_rule_* are the static
42 * functions that operate on rules, and devfs_ruleset_* are the static
43 * functions that operate on rulesets. The line between the last two
44 * isn't always clear, but the guideline is still useful.
46 * A note on "special" identifiers: Ruleset 0 is the NULL, or empty,
47 * ruleset; it cannot be deleted or changed in any way. This may be
48 * assumed inside the code; e.g., a ruleset of 0 may be interpeted to
49 * mean "no ruleset". The interpretation of rule 0 is
50 * command-dependent, but in no case is there a real rule with number
53 * A note on errno codes: To make it easier for the userland to tell
54 * what went wrong, we sometimes use errno codes that are not entirely
55 * appropriate for the error but that would be less ambiguous than the
56 * appropriate "generic" code. For example, when we can't find a
57 * ruleset, we return ESRCH instead of ENOENT (except in
58 * DEVFSIO_{R,S}GETNEXT, where a nonexistent ruleset means "end of
59 * list", and the userland expects ENOENT to be this indicator); this
60 * way, when an operation fails, it's clear that what couldn't be
61 * found is a ruleset and not a rule (well, it's clear to those who
62 * know the convention).
65 #include <sys/param.h>
66 #include <sys/systm.h>
68 #include <sys/kernel.h>
69 #include <sys/malloc.h>
71 #include <sys/dirent.h>
72 #include <sys/ioccom.h>
76 #include <fs/devfs/devfs.h>
77 #include <fs/devfs/devfs_int.h>
80 * Kernel version of devfs_rule.
83 TAILQ_ENTRY(devfs_krule) dk_list;
84 struct devfs_ruleset *dk_ruleset;
85 struct devfs_rule dk_rule;
88 TAILQ_HEAD(rulehead, devfs_krule);
89 static MALLOC_DEFINE(M_DEVFSRULE, "DEVFS_RULE", "DEVFS rule storage");
92 * Structure to describe a ruleset.
94 struct devfs_ruleset {
95 TAILQ_ENTRY(devfs_ruleset) ds_list;
96 struct rulehead ds_rules;
97 devfs_rsnum ds_number;
101 static devfs_rid devfs_rid_input(devfs_rid rid, struct devfs_mount *dm);
103 static void devfs_rule_applyde_recursive(struct devfs_krule *dk,
104 struct devfs_dirent *de);
105 static void devfs_rule_applydm(struct devfs_krule *dk, struct devfs_mount *dm);
106 static int devfs_rule_autonumber(struct devfs_ruleset *ds, devfs_rnum *rnp);
107 static struct devfs_krule *devfs_rule_byid(devfs_rid rid);
108 static int devfs_rule_delete(struct devfs_krule *dkp);
109 static struct cdev *devfs_rule_getdev(struct devfs_dirent *de);
110 static int devfs_rule_input(struct devfs_rule *dr, struct devfs_mount *dm);
111 static int devfs_rule_insert(struct devfs_rule *dr);
112 static int devfs_rule_match(struct devfs_krule *dk, struct devfs_dirent *de);
113 static int devfs_rule_matchpath(struct devfs_krule *dk,
114 struct devfs_dirent *de);
115 static void devfs_rule_run(struct devfs_krule *dk, struct devfs_dirent *de, unsigned depth);
117 static void devfs_ruleset_applyde(struct devfs_ruleset *ds,
118 struct devfs_dirent *de, unsigned depth);
119 static void devfs_ruleset_applydm(struct devfs_ruleset *ds,
120 struct devfs_mount *dm);
121 static struct devfs_ruleset *devfs_ruleset_bynum(devfs_rsnum rsnum);
122 static struct devfs_ruleset *devfs_ruleset_create(devfs_rsnum rsnum);
123 static void devfs_ruleset_reap(struct devfs_ruleset *dsp);
124 static int devfs_ruleset_use(devfs_rsnum rsnum, struct devfs_mount *dm);
126 static struct sx sx_rules;
127 SX_SYSINIT(sx_rules, &sx_rules, "DEVFS ruleset lock");
129 static TAILQ_HEAD(, devfs_ruleset) devfs_rulesets =
130 TAILQ_HEAD_INITIALIZER(devfs_rulesets);
133 * Called to apply the proper rules for 'de' before it can be
134 * exposed to the userland. This should be called with an exclusive
135 * lock on dm in case we need to run anything.
138 devfs_rules_apply(struct devfs_mount *dm, struct devfs_dirent *de)
140 struct devfs_ruleset *ds;
142 sx_assert(&dm->dm_lock, SX_XLOCKED);
144 if (dm->dm_ruleset == 0)
147 ds = devfs_ruleset_bynum(dm->dm_ruleset);
148 KASSERT(ds != NULL, ("mount-point has NULL ruleset"));
149 devfs_ruleset_applyde(ds, de, devfs_rule_depth);
150 sx_sunlock(&sx_rules);
154 * Rule subsystem ioctl hook.
157 devfs_rules_ioctl(struct devfs_mount *dm, u_long cmd, caddr_t data, struct thread *td)
159 struct devfs_ruleset *ds;
160 struct devfs_krule *dk;
161 struct devfs_rule *dr;
167 sx_assert(&dm->dm_lock, SX_XLOCKED);
170 * XXX: This returns an error regardless of whether we actually
171 * support the cmd or not.
173 * We could make this privileges finer grained if desired.
175 error = priv_check(td, PRIV_DEVFS_RULE);
183 dr = (struct devfs_rule *)data;
184 error = devfs_rule_input(dr, dm);
187 dk = devfs_rule_byid(dr->dr_id);
192 if (rid2rsn(dr->dr_id) == 0) {
196 error = devfs_rule_insert(dr);
199 dr = (struct devfs_rule *)data;
200 error = devfs_rule_input(dr, dm);
205 * This is one of many possible hackish
206 * implementations. The primary contender is an
207 * implementation where the rule we read in is
208 * temporarily inserted into some ruleset, perhaps
209 * with a hypothetical DRO_NOAUTO flag so that it
210 * doesn't get used where it isn't intended, and
211 * applied in the normal way. This can be done in the
212 * userland (DEVFSIO_ADD, DEVFSIO_APPLYID,
213 * DEVFSIO_DEL) or in the kernel; either way it breaks
214 * some corner case assumptions in other parts of the
215 * code (not that this implementation doesn't do
218 if (dr->dr_iacts & DRA_INCSET &&
219 devfs_ruleset_bynum(dr->dr_incset) == NULL) {
223 dk = malloc(sizeof(*dk), M_TEMP, M_WAITOK | M_ZERO);
224 memcpy(&dk->dk_rule, dr, sizeof(*dr));
225 devfs_rule_applydm(dk, dm);
228 case DEVFSIO_RAPPLYID:
229 rid = *(devfs_rid *)data;
230 rid = devfs_rid_input(rid, dm);
231 dk = devfs_rule_byid(rid);
236 devfs_rule_applydm(dk, dm);
239 rid = *(devfs_rid *)data;
240 rid = devfs_rid_input(rid, dm);
241 dk = devfs_rule_byid(rid);
247 error = devfs_rule_delete(dk);
249 case DEVFSIO_RGETNEXT:
250 dr = (struct devfs_rule *)data;
251 error = devfs_rule_input(dr, dm);
255 * We can't use devfs_rule_byid() here since that
256 * requires the rule specified to exist, but we want
257 * getnext(N) to work whether there is a rule N or not
258 * (specifically, getnext(0) must work, but we should
259 * never have a rule 0 since the add command
260 * interprets 0 to mean "auto-number").
262 ds = devfs_ruleset_bynum(rid2rsn(dr->dr_id));
267 rnum = rid2rn(dr->dr_id);
268 TAILQ_FOREACH(dk, &ds->ds_rules, dk_list) {
269 if (rid2rn(dk->dk_rule.dr_id) > rnum)
276 memcpy(dr, &dk->dk_rule, sizeof(*dr));
279 rsnum = *(devfs_rsnum *)data;
280 error = devfs_ruleset_use(rsnum, dm);
283 rsnum = *(devfs_rsnum *)data;
284 rsnum = rid2rsn(devfs_rid_input(mkrid(rsnum, 0), dm));
285 ds = devfs_ruleset_bynum(rsnum);
290 devfs_ruleset_applydm(ds, dm);
292 case DEVFSIO_SGETNEXT:
293 rsnum = *(devfs_rsnum *)data;
294 TAILQ_FOREACH(ds, &devfs_rulesets, ds_list) {
295 if (ds->ds_number > rsnum)
302 *(devfs_rsnum *)data = ds->ds_number;
309 sx_xunlock(&sx_rules);
314 * Adjust the rule identifier to use the ruleset of dm if one isn't
315 * explicitly specified.
317 * Note that after this operation, rid2rsn(rid) might still be 0, and
318 * that's okay; ruleset 0 is a valid ruleset, but when it's read in
319 * from the userland, it means "current ruleset for this mount-point".
322 devfs_rid_input(devfs_rid rid, struct devfs_mount *dm)
325 if (rid2rsn(rid) == 0)
326 return (mkrid(dm->dm_ruleset, rid2rn(rid)));
332 * Apply dk to de and everything under de.
334 * XXX: This method needs a function call for every nested
335 * subdirectory in a devfs mount. If we plan to have many of these,
336 * we might eventually run out of kernel stack space.
337 * XXX: a linear search could be done through the cdev list instead.
340 devfs_rule_applyde_recursive(struct devfs_krule *dk, struct devfs_dirent *de)
342 struct devfs_dirent *de2;
344 TAILQ_FOREACH(de2, &de->de_dlist, de_list)
345 devfs_rule_applyde_recursive(dk, de2);
346 devfs_rule_run(dk, de, devfs_rule_depth);
350 * Apply dk to all entires in dm.
353 devfs_rule_applydm(struct devfs_krule *dk, struct devfs_mount *dm)
356 devfs_rule_applyde_recursive(dk, dm->dm_rootdir);
360 * Automatically select a number for a new rule in ds, and write the
364 devfs_rule_autonumber(struct devfs_ruleset *ds, devfs_rnum *rnump)
366 struct devfs_krule *dk;
368 /* Find the last rule. */
369 dk = TAILQ_LAST(&ds->ds_rules, rulehead);
373 *rnump = rid2rn(dk->dk_rule.dr_id) + 100;
374 /* Detect overflow. */
375 if (*rnump < rid2rn(dk->dk_rule.dr_id))
378 KASSERT(devfs_rule_byid(mkrid(ds->ds_number, *rnump)) == NULL,
379 ("autonumbering resulted in an already existing rule"));
384 * Find a krule by id.
386 static struct devfs_krule *
387 devfs_rule_byid(devfs_rid rid)
389 struct devfs_ruleset *ds;
390 struct devfs_krule *dk;
394 ds = devfs_ruleset_bynum(rid2rsn(rid));
397 TAILQ_FOREACH(dk, &ds->ds_rules, dk_list) {
398 if (rid2rn(dk->dk_rule.dr_id) == rn)
400 else if (rid2rn(dk->dk_rule.dr_id) > rn)
407 * Remove dkp from any lists it may be on and remove memory associated
411 devfs_rule_delete(struct devfs_krule *dk)
413 struct devfs_ruleset *ds;
415 if (dk->dk_rule.dr_iacts & DRA_INCSET) {
416 ds = devfs_ruleset_bynum(dk->dk_rule.dr_incset);
417 KASSERT(ds != NULL, ("DRA_INCSET but bad dr_incset"));
419 devfs_ruleset_reap(ds);
422 TAILQ_REMOVE(&ds->ds_rules, dk, dk_list);
423 devfs_ruleset_reap(ds);
424 free(dk, M_DEVFSRULE);
429 * Get a struct cdev *corresponding to de so we can try to match rules based
430 * on it. If this routine returns NULL, there is no struct cdev *associated
431 * with the dirent (symlinks and directories don't have dev_ts), and
432 * the caller should assume that any critera dependent on a dev_t
436 devfs_rule_getdev(struct devfs_dirent *de)
439 if (de->de_cdp == NULL)
441 if (de->de_cdp->cdp_flags & CDP_ACTIVE)
442 return (&de->de_cdp->cdp_c);
448 * Do what we need to do to a rule that we just loaded from the
449 * userland. In particular, we need to check the magic, and adjust
450 * the ruleset appropriate if desired.
453 devfs_rule_input(struct devfs_rule *dr, struct devfs_mount *dm)
456 if (dr->dr_magic != DEVFS_MAGIC)
457 return (ERPCMISMATCH);
458 dr->dr_id = devfs_rid_input(dr->dr_id, dm);
463 * Import dr into the appropriate place in the kernel (i.e., make a
464 * krule). The value of dr is copied, so the pointer may be destroyed
465 * after this call completes.
468 devfs_rule_insert(struct devfs_rule *dr)
470 struct devfs_ruleset *ds, *dsi;
471 struct devfs_krule *k1;
472 struct devfs_krule *dk;
478 * This stuff seems out of place here, but we want to do it as
479 * soon as possible so that if it fails, we don't have to roll
480 * back any changes we already made (e.g., ruleset creation).
482 if (dr->dr_iacts & DRA_INCSET) {
483 dsi = devfs_ruleset_bynum(dr->dr_incset);
489 rsnum = rid2rsn(dr->dr_id);
490 KASSERT(rsnum != 0, ("Inserting into ruleset zero"));
492 ds = devfs_ruleset_bynum(rsnum);
494 ds = devfs_ruleset_create(rsnum);
495 dkrn = rid2rn(dr->dr_id);
497 error = devfs_rule_autonumber(ds, &dkrn);
499 devfs_ruleset_reap(ds);
504 dk = malloc(sizeof(*dk), M_DEVFSRULE, M_WAITOK | M_ZERO);
508 /* XXX: Inspect dr? */
509 memcpy(&dk->dk_rule, dr, sizeof(*dr));
510 dk->dk_rule.dr_id = mkrid(rid2rsn(dk->dk_rule.dr_id), dkrn);
512 TAILQ_FOREACH(k1, &ds->ds_rules, dk_list) {
513 if (rid2rn(k1->dk_rule.dr_id) > dkrn) {
514 TAILQ_INSERT_BEFORE(k1, dk, dk_list);
519 TAILQ_INSERT_TAIL(&ds->ds_rules, dk, dk_list);
524 * Determine whether dk matches de. Returns 1 if dk should be run on
528 devfs_rule_match(struct devfs_krule *dk, struct devfs_dirent *de)
530 struct devfs_rule *dr = &dk->dk_rule;
535 dev = devfs_rule_getdev(de);
537 * At this point, if dev is NULL, we should assume that any
538 * criteria that depend on it don't match. We should *not*
539 * just ignore them (i.e., act like they weren't specified),
540 * since that makes a rule that only has criteria dependent on
541 * the struct cdev *match all symlinks and directories.
543 * Note also that the following tests are somewhat reversed:
544 * They're actually testing to see whether the condition does
545 * *not* match, since the default is to assume the rule should
546 * be run (such as if there are no conditions).
548 if (dr->dr_icond & DRC_DSWFLAGS) {
551 dsw = dev_refthread(dev, &ref);
554 if ((dsw->d_flags & dr->dr_dswflags) == 0) {
555 dev_relthread(dev, ref);
558 dev_relthread(dev, ref);
560 if (dr->dr_icond & DRC_PATHPTRN)
561 if (!devfs_rule_matchpath(dk, de))
568 * Determine whether dk matches de on account of dr_pathptrn.
571 devfs_rule_matchpath(struct devfs_krule *dk, struct devfs_dirent *de)
573 struct devfs_rule *dr = &dk->dk_rule;
577 dev = devfs_rule_getdev(de);
579 pname = dev->si_name;
580 else if (de->de_dirent->d_type == DT_LNK ||
581 de->de_dirent->d_type == DT_DIR)
582 pname = de->de_dirent->d_name;
585 KASSERT(pname != NULL, ("devfs_rule_matchpath: NULL pname"));
587 return (fnmatch(dr->dr_pathptrn, pname, 0) == 0);
594 devfs_rule_run(struct devfs_krule *dk, struct devfs_dirent *de, unsigned depth)
596 struct devfs_rule *dr = &dk->dk_rule;
597 struct devfs_ruleset *ds;
599 if (!devfs_rule_match(dk, de))
601 if (dr->dr_iacts & DRA_BACTS) {
602 if (dr->dr_bacts & DRB_HIDE)
603 de->de_flags |= DE_WHITEOUT;
604 if (dr->dr_bacts & DRB_UNHIDE)
605 de->de_flags &= ~DE_WHITEOUT;
607 if (dr->dr_iacts & DRA_UID)
608 de->de_uid = dr->dr_uid;
609 if (dr->dr_iacts & DRA_GID)
610 de->de_gid = dr->dr_gid;
611 if (dr->dr_iacts & DRA_MODE)
612 de->de_mode = dr->dr_mode;
613 if (dr->dr_iacts & DRA_INCSET) {
615 * XXX: we should tell the user if the depth is exceeded here
616 * XXX: but it is not obvious how to. A return value will
617 * XXX: not work as this is called when devices are created
618 * XXX: long time after the rules were instantiated.
619 * XXX: a printf() would probably give too much noise, or
620 * XXX: DoS the machine. I guess a a rate-limited message
624 ds = devfs_ruleset_bynum(dk->dk_rule.dr_incset);
625 KASSERT(ds != NULL, ("DRA_INCSET but bad dr_incset"));
626 devfs_ruleset_applyde(ds, de, depth - 1);
632 * Apply all the rules in ds to de.
635 devfs_ruleset_applyde(struct devfs_ruleset *ds, struct devfs_dirent *de, unsigned depth)
637 struct devfs_krule *dk;
639 TAILQ_FOREACH(dk, &ds->ds_rules, dk_list)
640 devfs_rule_run(dk, de, depth);
644 * Apply all the rules in ds to all the entires in dm.
647 devfs_ruleset_applydm(struct devfs_ruleset *ds, struct devfs_mount *dm)
649 struct devfs_krule *dk;
652 * XXX: Does it matter whether we do
664 * The end result is obviously the same, but does the order
667 TAILQ_FOREACH(dk, &ds->ds_rules, dk_list)
668 devfs_rule_applydm(dk, dm);
672 * Find a ruleset by number.
674 static struct devfs_ruleset *
675 devfs_ruleset_bynum(devfs_rsnum rsnum)
677 struct devfs_ruleset *ds;
679 TAILQ_FOREACH(ds, &devfs_rulesets, ds_list) {
680 if (ds->ds_number == rsnum)
687 * Create a new ruleset.
689 static struct devfs_ruleset *
690 devfs_ruleset_create(devfs_rsnum rsnum)
692 struct devfs_ruleset *s1;
693 struct devfs_ruleset *ds;
695 KASSERT(rsnum != 0, ("creating ruleset zero"));
697 KASSERT(devfs_ruleset_bynum(rsnum) == NULL,
698 ("creating already existent ruleset %d", rsnum));
700 ds = malloc(sizeof(*ds), M_DEVFSRULE, M_WAITOK | M_ZERO);
701 ds->ds_number = rsnum;
702 TAILQ_INIT(&ds->ds_rules);
704 TAILQ_FOREACH(s1, &devfs_rulesets, ds_list) {
705 if (s1->ds_number > rsnum) {
706 TAILQ_INSERT_BEFORE(s1, ds, ds_list);
711 TAILQ_INSERT_TAIL(&devfs_rulesets, ds, ds_list);
716 * Remove a ruleset from the system if it's empty and not used
717 * anywhere. This should be called after every time a rule is deleted
718 * from this ruleset or the reference count is decremented.
721 devfs_ruleset_reap(struct devfs_ruleset *ds)
724 KASSERT(ds->ds_number != 0, ("reaping ruleset zero "));
726 if (!TAILQ_EMPTY(&ds->ds_rules) || ds->ds_refcount != 0)
729 TAILQ_REMOVE(&devfs_rulesets, ds, ds_list);
730 free(ds, M_DEVFSRULE);
734 * Make rsnum the active ruleset for dm.
737 devfs_ruleset_use(devfs_rsnum rsnum, struct devfs_mount *dm)
739 struct devfs_ruleset *cds, *ds;
741 if (dm->dm_ruleset != 0) {
742 cds = devfs_ruleset_bynum(dm->dm_ruleset);
744 devfs_ruleset_reap(cds);
752 ds = devfs_ruleset_bynum(rsnum);
754 ds = devfs_ruleset_create(rsnum);
755 /* These should probably be made atomic somehow. */
757 dm->dm_ruleset = rsnum;
763 devfs_rules_cleanup(struct devfs_mount *dm)
765 struct devfs_ruleset *ds;
767 sx_assert(&dm->dm_lock, SX_XLOCKED);
768 if (dm->dm_ruleset != 0) {
769 ds = devfs_ruleset_bynum(dm->dm_ruleset);
771 devfs_ruleset_reap(ds);