2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neilb@suse.de>
31 * Policy module for mdadm.
32 * A policy statement about a device lists a set of values for each
33 * of a set of names. Each value can have a metadata type as context.
36 * action - the actions that can be taken on hot-plug
37 * domain - the domain(s) that the device is part of
39 * Policy information is extracted from various sources, but
40 * particularly from a set of policy rules in mdadm.conf
43 static void pol_new(struct dev_policy
**pol
, char *name
, const char *val
,
46 struct dev_policy
*n
= malloc(sizeof(*n
));
47 const char *real_metadata
= NULL
;
53 /* We need to normalise the metadata name */
55 for (i
= 0; superlist
[i
] ; i
++)
56 if (strcmp(metadata
, superlist
[i
]->name
) == 0) {
57 real_metadata
= superlist
[i
]->name
;
61 if (strcmp(metadata
, "1") == 0 ||
62 strcmp(metadata
, "1.0") == 0 ||
63 strcmp(metadata
, "1.1") == 0 ||
64 strcmp(metadata
, "1.2") == 0)
65 real_metadata
= super1
.name
;
68 static const char *prev
= NULL
;
69 if (prev
!= metadata
) {
70 fprintf(stderr
, Name
": metadata=%s unrecognised - ignoring rule\n",
74 real_metadata
= "unknown";
78 n
->metadata
= real_metadata
;
83 static int pol_lesseq(struct dev_policy
*a
, struct dev_policy
*b
)
87 if (a
->name
< b
->name
)
89 if (a
->name
> b
->name
)
92 cmp
= strcmp(a
->value
, b
->value
);
98 return (a
->metadata
<= b
->metadata
);
101 static void pol_sort(struct dev_policy
**pol
)
103 /* sort policy list in *pol by name/metadata/value
107 struct dev_policy
*pl
[2];
112 struct dev_policy
**plp
[2], *p
[2];
114 struct dev_policy nul
= { NULL
, NULL
, NULL
, NULL
};
115 struct dev_policy
*prev
= &nul
;
118 /* p[] are the two lists that we are merging.
119 * plp[] are the ends of the two lists we create
121 * 'curr' is which of plp[] that we are currently
123 * 'next' is which if p[] we will take the next
125 * 'prev' is that last value, which was placed in
133 /* take least of p[0] and p[1]
134 * if it is larger than prev, add to
135 * plp[curr], else swap curr then add
137 while (p
[0] || p
[1]) {
138 if (p
[next
] == NULL
||
139 (p
[1-next
] != NULL
&&
140 !(pol_lesseq(prev
, p
[1-next
])
141 ^pol_lesseq(prev
, p
[next
])
142 ^pol_lesseq(p
[next
], p
[1-next
])))
146 if (!pol_lesseq(prev
, p
[next
]))
149 *plp
[curr
] = prev
= p
[next
];
150 plp
[curr
] = &p
[next
]->next
;
151 p
[next
] = p
[next
]->next
;
155 } while (pl
[0] && pl
[1]);
162 static void pol_dedup(struct dev_policy
*pol
)
164 /* This is a sorted list - remove duplicates. */
165 while (pol
&& pol
->next
) {
166 if (pol_lesseq(pol
->next
, pol
)) {
167 struct dev_policy
*tmp
= pol
->next
;
168 pol
->next
= tmp
->next
;
176 * pol_find finds the first entry in the policy
177 * list to match name.
178 * If it returns non-NULL there is at least one
179 * value, but how many can only be found by
180 * iterating through the list.
182 struct dev_policy
*pol_find(struct dev_policy
*pol
, char *name
)
184 while (pol
&& pol
->name
< name
)
187 if (!pol
|| pol
->name
!= name
)
192 static char *disk_path(struct mdinfo
*disk
)
197 char symlink
[PATH_MAX
] = "/dev/disk/by-path/";
200 by_path
= opendir(symlink
);
203 prefix_len
= strlen(symlink
);
205 while ((ent
= readdir(by_path
)) != NULL
) {
206 if (ent
->d_type
!= DT_LNK
)
208 strncpy(symlink
+ prefix_len
,
210 sizeof(symlink
) - prefix_len
);
211 if (stat(symlink
, &stb
) < 0)
213 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
)
215 if (stb
.st_rdev
!= makedev(disk
->disk
.major
, disk
->disk
.minor
))
218 return strdup(ent
->d_name
);
224 char type_part
[] = "part";
225 char type_disk
[] = "disk";
226 static char *disk_type(struct mdinfo
*disk
)
230 sprintf(buf
, "/sys/dev/block/%d:%d/partition",
231 disk
->disk
.major
, disk
->disk
.minor
);
232 if (stat(buf
, &stb
) == 0)
238 static int pol_match(struct rule
*rule
, char *path
, char *type
)
240 /* check if this rule matches on path and type */
241 int pathok
= 0; /* 0 == no path, 1 == match, -1 == no match yet */
245 if (rule
->name
== rule_path
) {
248 if (fnmatch(rule
->value
, path
, 0) == 0)
251 if (rule
->name
== rule_type
) {
254 if (strcmp(rule
->value
, type
) == 0)
259 return pathok
>= 0 && typeok
>= 0;
262 static void pol_merge(struct dev_policy
**pol
, struct rule
*rule
)
264 /* copy any name assignments from rule into pol */
266 char *metadata
= NULL
;
267 for (r
= rule
; r
; r
= r
->next
)
268 if (r
->name
== pol_metadata
)
271 for (r
= rule
; r
; r
= r
->next
)
272 if (r
->name
== pol_act
||
273 r
->name
== pol_domain
)
274 pol_new(pol
, r
->name
, r
->value
, metadata
);
277 static int path_has_part(char *path
, char **part
)
279 /* check if path ends with "-partNN" and
280 * if it does, place a pointer to "-pathNN"
283 int l
= strlen(path
);
284 while (l
> 1 && isdigit(path
[l
-1]))
286 if (l
< 5 || strncmp(path
+l
-5, "-part", 5) != 0)
292 static void pol_merge_part(struct dev_policy
**pol
, struct rule
*rule
, char *part
)
294 /* copy any name assignments from rule into pol, appending
295 * -part to any domain. The string with -part appended is
296 * stored with the rule so it has a lifetime to match
300 char *metadata
= NULL
;
301 for (r
= rule
; r
; r
= r
->next
)
302 if (r
->name
== pol_metadata
)
305 for (r
= rule
; r
; r
= r
->next
) {
306 if (r
->name
== pol_act
)
307 pol_new(pol
, r
->name
, r
->value
, metadata
);
308 else if (r
->name
== pol_domain
) {
313 len
= strlen(r
->value
);
314 for (dom
= dl_next(r
->dups
); dom
!= r
->dups
;
316 if (strcmp(dom
+len
+1, part
)== 0)
318 if (dom
== r
->dups
) {
319 char *newdom
= dl_strndup(
320 r
->value
, len
+ 1 + strlen(part
));
321 strcat(strcat(newdom
, "-"), part
);
322 dl_add(r
->dups
, newdom
);
325 pol_new(pol
, r
->name
, dom
, metadata
);
330 static struct pol_rule
*config_rules
= NULL
;
331 static struct pol_rule
**config_rules_end
= NULL
;
332 static int config_rules_has_path
= 0;
335 * most policy comes from a set policy rules that are
336 * read from the config file.
337 * path_policy() gathers policy information for the
338 * disk described in the given a 'path' and a 'type'.
340 struct dev_policy
*path_policy(char *path
, char *type
)
342 struct pol_rule
*rules
;
343 struct dev_policy
*pol
= NULL
;
349 rules
= config_rules
;
353 if (rules
->type
== rule_policy
)
354 if (pol_match(rules
->rule
, path
, type
))
355 pol_merge(&pol
, rules
->rule
);
356 if (rules
->type
== rule_part
&& strcmp(type
, type_part
) == 0)
357 if (path_has_part(path
, &part
)) {
359 if (pol_match(rules
->rule
, path
, type_disk
))
360 pol_merge_part(&pol
, rules
->rule
, part
+1);
366 /* Now add any metadata-specific internal knowledge
369 for (i
=0; superlist
[i
]; i
++)
370 if (superlist
[i
]->get_disk_controller_domain
) {
372 superlist
[i
]->get_disk_controller_domain(path
);
374 pol_new(&pol
, pol_domain
, d
, superlist
[i
]->name
);
382 void pol_add(struct dev_policy
**pol
,
383 char *name
, char *val
,
386 pol_new(pol
, name
, val
, metadata
);
393 * disk_policy() gathers policy information for the
394 * disk described in the given mdinfo (disk.{major,minor}).
396 struct dev_policy
*disk_policy(struct mdinfo
*disk
)
399 char *type
= disk_type(disk
);
400 struct dev_policy
*pol
= NULL
;
404 if (config_rules_has_path
)
405 path
= disk_path(disk
);
409 pol
= path_policy(path
, type
);
415 struct dev_policy
*devnum_policy(int dev
)
418 disk
.disk
.major
= major(dev
);
419 disk
.disk
.minor
= minor(dev
);
420 return disk_policy(&disk
);
424 * process policy rules read from config file.
427 char rule_path
[] = "path";
428 char rule_type
[] = "type";
430 char rule_policy
[] = "policy";
431 char rule_part
[] = "part-policy";
433 char pol_metadata
[] = "metadata";
434 char pol_act
[] = "action";
435 char pol_domain
[] = "domain";
436 char pol_auto
[] = "auto";
438 static int try_rule(char *w
, char *name
, struct rule
**rp
)
441 int len
= strlen(name
);
442 if (strncmp(w
, name
, len
) != 0 ||
445 r
= malloc(sizeof(*r
));
448 r
->value
= strdup(w
+len
+1);
454 void policyline(char *line
, char *type
)
459 if (config_rules_end
== NULL
)
460 config_rules_end
= &config_rules
;
462 pr
= malloc(sizeof(*pr
));
465 for (w
= dl_next(line
); w
!= line
; w
= dl_next(w
)) {
466 if (try_rule(w
, rule_path
, &pr
->rule
))
467 config_rules_has_path
= 1;
468 else if (! try_rule(w
, rule_type
, &pr
->rule
) &&
469 ! try_rule(w
, pol_metadata
, &pr
->rule
) &&
470 ! try_rule(w
, pol_act
, &pr
->rule
) &&
471 ! try_rule(w
, pol_domain
, &pr
->rule
) &&
472 ! try_rule(w
, pol_auto
, &pr
->rule
))
473 fprintf(stderr
, Name
": policy rule %s unrecognised and ignored\n",
476 pr
->next
= config_rules
;
480 void policy_add(char *type
, ...)
486 pr
= malloc(sizeof(*pr
));
491 while ((name
= va_arg(ap
, char*)) != NULL
) {
494 val
= va_arg(ap
, char*);
495 r
= malloc(sizeof(*r
));
498 r
->value
= strdup(val
);
502 pr
->next
= config_rules
;
506 void policy_free(void)
508 while (config_rules
) {
509 struct pol_rule
*pr
= config_rules
;
512 config_rules
= config_rules
->next
;
514 for (r
= pr
->rule
; r
; ) {
515 struct rule
*next
= r
->next
;
524 config_rules_end
= NULL
;
525 config_rules_has_path
= 0;
528 void dev_policy_free(struct dev_policy
*p
)
530 struct dev_policy
*t
;
538 static enum policy_action
map_act(const char *act
)
540 if (strcmp(act
, "include") == 0)
542 if (strcmp(act
, "re-add") == 0)
544 if (strcmp(act
, "spare") == 0)
546 if (strcmp(act
, "spare-same-slot") == 0)
547 return act_spare_same_slot
;
548 if (strcmp(act
, "force-spare") == 0)
549 return act_force_spare
;
553 static enum policy_action
policy_action(struct dev_policy
*plist
, const char *metadata
)
555 enum policy_action rv
= act_default
;
556 struct dev_policy
*p
;
558 plist
= pol_find(plist
, pol_act
);
559 pol_for_each(p
, plist
, metadata
) {
560 enum policy_action a
= map_act(p
->value
);
567 int policy_action_allows(struct dev_policy
*plist
, const char *metadata
, enum policy_action want
)
569 enum policy_action act
= policy_action(plist
, metadata
);
573 return (act
>= want
);
576 int disk_action_allows(struct mdinfo
*disk
, const char *metadata
, enum policy_action want
)
578 struct dev_policy
*pol
= disk_policy(disk
);
579 int rv
= policy_action_allows(pol
, metadata
, want
);
581 dev_policy_free(pol
);
587 * Any device can have a list of domains asserted by different policy
589 * An array also has a list of domains comprising all the domains of
590 * all the devices in an array.
591 * Where an array has a spare-group, that becomes an addition domain for
592 * every device in the array and thus for the array.
594 * We keep the list of domains in a sorted linked list
595 * As dev policies are already sorted, this is fairly easy to manage.
598 static struct domainlist
**domain_merge_one(struct domainlist
**domp
,
601 /* merge a domain name into a sorted list and return the
602 * location of the insertion or match
604 struct domainlist
*dom
= *domp
;
606 while (dom
&& strcmp(dom
->dom
, domain
) < 0) {
610 if (dom
== NULL
|| strcmp(dom
->dom
, domain
) != 0) {
611 dom
= malloc(sizeof(*dom
));
620 void dump_policy(struct dev_policy
*policy
)
623 dprintf("policy: %p name: %s value: %s metadata: %s\n",
628 policy
= policy
->next
;
633 void domain_merge(struct domainlist
**domp
, struct dev_policy
*pollist
,
634 const char *metadata
)
636 /* Add to 'domp' all the domains in pol that apply to 'metadata'
637 * which are not already in domp
639 struct dev_policy
*pol
;
640 pollist
= pol_find(pollist
, pol_domain
);
641 pol_for_each(pol
, pollist
, metadata
)
642 domain_merge_one(domp
, pol
->value
);
645 int domain_test(struct domainlist
*dom
, struct dev_policy
*pol
,
646 const char *metadata
)
648 /* Check that all domains in pol (for metadata) are also in
649 * dom. Both lists are sorted.
650 * If pol has no domains, we don't really know about this device
651 * so we allow caller to choose:
653 * 0: has domains, not all match
654 * 1: has domains, all match
657 struct dev_policy
*p
;
659 pol
= pol_find(pol
, pol_domain
);
660 pol_for_each(p
, pol
, metadata
) {
662 while (dom
&& strcmp(dom
->dom
, p
->value
) < 0)
664 if (!dom
|| strcmp(dom
->dom
, p
->value
) != 0)
670 void domainlist_add_dev(struct domainlist
**dom
, int devnum
, const char *metadata
)
672 struct dev_policy
*pol
= devnum_policy(devnum
);
673 domain_merge(dom
, pol
, metadata
);
674 dev_policy_free(pol
);
677 struct domainlist
*domain_from_array(struct mdinfo
*mdi
, const char *metadata
)
679 struct domainlist
*domlist
= NULL
;
681 for (mdi
= mdi
->devs
; mdi
; mdi
= mdi
->next
)
682 domainlist_add_dev(&domlist
, makedev(mdi
->disk
.major
,
689 void domain_add(struct domainlist
**domp
, char *domain
)
691 domain_merge_one(domp
, domain
);
695 void domain_free(struct domainlist
*dl
)
698 struct domainlist
*head
= dl
;
706 * Some policy decisions are guided by knowledge of which
707 * array previously owned the device at a given physical location (path).
708 * When removing a device from an array we might record the array against
709 * the path, and when finding a new device, we might look for which
710 * array previously used that path.
712 * The 'array' is described by a map_ent, and the path by a the disk in an
713 * mdinfo, or a string.
716 void policy_save_path(char *id_path
, struct map_ent
*array
)
721 if (mkdir(FAILED_SLOTS_DIR
, S_IRWXU
) < 0 && errno
!= EEXIST
) {
722 fprintf(stderr
, Name
": can't create file to save path "
723 "to old disk: %s\n", strerror(errno
));
727 snprintf(path
, PATH_MAX
, FAILED_SLOTS_DIR
"/%s", id_path
);
728 f
= fopen(path
, "w");
730 fprintf(stderr
, Name
": can't create file to"
731 " save path to old disk: %s\n",
736 if (fprintf(f
, "%s %08x:%08x:%08x:%08x\n",
738 array
->uuid
[0], array
->uuid
[1],
739 array
->uuid
[2], array
->uuid
[3]) <= 0)
740 fprintf(stderr
, Name
": Failed to write to "
741 "<id_path> cookie\n");
746 int policy_check_path(struct mdinfo
*disk
, struct map_ent
*array
)
750 char *id_path
= disk_path(disk
);
756 snprintf(path
, PATH_MAX
, FAILED_SLOTS_DIR
"/%s", id_path
);
757 f
= fopen(path
, "r");
761 rv
= fscanf(f
, " %s %x:%x:%x:%x\n",
771 /* invocation of udev rule file */
772 char udev_template_start
[] =
773 "# do not edit this file, it is automatically generated by mdadm\n"
776 /* find rule named rule_type and return its value */
777 char *find_rule(struct rule
*rule
, char *rule_type
)
780 if (rule
->name
== rule_type
)
788 #define UDEV_RULE_FORMAT \
789 "ACTION==\"add\", SUBSYSTEM==\"block\", " \
790 "ENV{DEVTYPE}==\"%s\", ENV{ID_PATH}==\"%s\", " \
791 "RUN+=\"/sbin/mdadm --incremental $env{DEVNAME}\"\n"
793 #define UDEV_RULE_FORMAT_NOTYPE \
794 "ACTION==\"add\", SUBSYSTEM==\"block\", " \
795 "ENV{ID_PATH}==\"%s\", " \
796 "RUN+=\"/sbin/mdadm --incremental $env{DEVNAME}\"\n"
798 /* Write rule in the rule file. Use format from UDEV_RULE_FORMAT */
799 int write_rule(struct rule
*rule
, int fd
, int force_part
)
802 char *pth
= find_rule(rule
, rule_path
);
803 char *typ
= find_rule(rule
, rule_type
);
810 snprintf(line
, sizeof(line
) - 1, UDEV_RULE_FORMAT
, typ
, pth
);
812 snprintf(line
, sizeof(line
) - 1, UDEV_RULE_FORMAT_NOTYPE
, pth
);
813 return write(fd
, line
, strlen(line
)) == (int)strlen(line
);
816 /* Generate single entry in udev rule basing on POLICY line found in config
817 * file. Take only those with paths, only first occurrence if paths are equal
818 * and if actions supports handling of spares (>=act_spare_same_slot)
820 int generate_entries(int fd
)
822 struct pol_rule
*loop
, *dup
;
823 char *loop_value
, *dup_value
;
826 for (loop
= config_rules
; loop
; loop
= loop
->next
) {
827 if (loop
->type
!= rule_policy
&& loop
->type
!= rule_part
)
831 /* only policies with paths and with actions supporting
832 * bare disks are considered */
833 loop_value
= find_rule(loop
->rule
, pol_act
);
834 if (!loop_value
|| map_act(loop_value
) < act_spare_same_slot
)
836 loop_value
= find_rule(loop
->rule
, rule_path
);
839 for (dup
= config_rules
; dup
!= loop
; dup
= dup
->next
) {
840 if (dup
->type
!= rule_policy
&& loop
->type
!= rule_part
)
842 dup_value
= find_rule(dup
->rule
, pol_act
);
843 if (!dup_value
|| map_act(dup_value
) < act_spare_same_slot
)
845 dup_value
= find_rule(dup
->rule
, rule_path
);
848 if (strcmp(loop_value
, dup_value
) == 0) {
854 /* not a dup or first occurrence */
856 if (!write_rule(loop
->rule
, fd
, loop
->type
== rule_part
) )
862 /* Write_rules routine creates dynamic udev rules used to handle
863 * hot-plug events for bare devices (and making them spares)
865 int Write_rules(char *rule_name
)
868 char udev_rule_file
[PATH_MAX
];
871 strcpy(udev_rule_file
, rule_name
);
872 strcat(udev_rule_file
, ".temp");
873 fd
= creat(udev_rule_file
,
874 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IROTH
);
880 /* write static invocation */
881 if (write(fd
, udev_template_start
,
882 sizeof(udev_template_start
) - 1)
883 != (int)sizeof(udev_template_start
)-1)
886 /* iterate, if none created or error occurred, remove file */
887 if (generate_entries(fd
) < 0)
893 rename(udev_rule_file
, rule_name
);
899 unlink(udev_rule_file
);