]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - monitor.c
5 #include <sys/select.h>
8 static char *array_states
[] = {
9 "clear", "inactive", "suspended", "readonly", "read-auto",
10 "clean", "active", "write-pending", "active-idle", NULL
};
11 static char *sync_actions
[] = {
12 "idle", "reshape", "resync", "recover", "check", "repair", NULL
15 static int write_attr(char *attr
, int fd
)
17 return write(fd
, attr
, strlen(attr
));
20 static void add_fd(fd_set
*fds
, int *maxfd
, int fd
)
29 static int read_attr(char *buf
, int len
, int fd
)
38 n
= read(fd
, buf
, len
- 1);
50 static int get_sync_pos(struct active_array
*a
)
55 n
= read_attr(buf
, 30, a
->sync_pos_fd
);
59 if (strncmp(buf
, "max", 3) == 0) {
60 a
->sync_pos
= ~(unsigned long long)0;
63 a
->sync_pos
= strtoull(buf
, NULL
, 10);
67 static int get_resync_start(struct active_array
*a
)
72 n
= read_attr(buf
, 30, a
->resync_start_fd
);
76 a
->resync_start
= strtoull(buf
, NULL
, 10);
81 static int attr_match(const char *attr
, const char *str
)
83 /* See if attr, read from a sysfs file, matches
84 * str. They must either be the same, or attr can
85 * have a trailing newline or comma
87 while (*attr
&& *str
&& *attr
== *str
) {
92 if (*str
|| (*attr
&& *attr
!= ',' && *attr
!= '\n'))
97 static int match_word(const char *word
, char **list
)
100 for (n
=0; list
[n
]; n
++)
101 if (attr_match(word
, list
[n
]))
106 static enum array_state
read_state(int fd
)
109 int n
= read_attr(buf
, 20, fd
);
113 return (enum array_state
) match_word(buf
, array_states
);
116 static enum sync_action
read_action( int fd
)
119 int n
= read_attr(buf
, 20, fd
);
123 return (enum sync_action
) match_word(buf
, sync_actions
);
126 int read_dev_state(int fd
)
129 int n
= read_attr(buf
, 60, fd
);
138 if (attr_match(cp
, "faulty"))
140 if (attr_match(cp
, "in_sync"))
142 if (attr_match(cp
, "write_mostly"))
143 rv
|= DS_WRITE_MOSTLY
;
144 if (attr_match(cp
, "spare"))
146 if (attr_match(cp
, "blocked"))
148 cp
= strchr(cp
, ',');
156 /* Monitor a set of active md arrays - all of which share the
157 * same metadata - and respond to events that require
160 * New arrays are detected by another thread which allocates
161 * required memory and attaches the data structure to our list.
165 * This is detected by array_state going to 'clear' or 'inactive'.
166 * while we thought it was active.
167 * Response is to mark metadata as clean and 'clear' the array(??)
169 * array_state if 'write-pending'
170 * We mark metadata as 'dirty' then set array to 'active'.
172 * Either ignore, or mark clean, then mark metadata as clean.
175 * detected by rd-N/state reporting "faulty"
176 * mark device as 'failed' in metadata, let the kernel release the
177 * device by writing '-blocked' to rd/state, and finally write 'remove' to
181 * sync_action was 'resync' and becomes 'idle' and resync_start becomes
183 * Notify metadata that sync is complete.
184 * "Deal with Degraded"
187 * sync_action changes from 'recover' to 'idle'
188 * Check each device state and mark metadata if 'faulty' or 'in_sync'.
189 * "Deal with Degraded"
191 * deal with degraded array
192 * We only do this when first noticing the array is degraded.
193 * This can be when we first see the array, when sync completes or
194 * when recovery completes.
196 * Check if number of failed devices suggests recovery is needed, and
198 * Ask metadata for a spare device
199 * Add device as not in_sync and give a role
204 * This only happens on finding a new array... mdadm will have set
205 * 'resync_start' to the correct value. If 'resync_start' indicates that an
206 * resync needs to occur set the array to the 'active' state rather than the
207 * initial read-auto state.
211 * We wait for a change (poll/select) on array_state, sync_action, and
212 * each rd-X/state file.
213 * When we get any change, we check everything. So read each state file,
214 * then decide what to do.
216 * The core action is to write new metadata to all devices in the array.
217 * This is done at most once on any wakeup.
218 * After that we might:
219 * - update the array_state
220 * - set the role of some devices.
221 * - request a sync_action
225 static int read_and_act(struct active_array
*a
)
231 a
->next_state
= bad_word
;
232 a
->next_action
= bad_action
;
234 a
->curr_state
= read_state(a
->info
.state_fd
);
235 a
->curr_action
= read_action(a
->action_fd
);
236 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
238 if (mdi
->state_fd
> 0)
239 mdi
->curr_state
= read_dev_state(mdi
->state_fd
);
242 if (a
->curr_state
<= inactive
&&
243 a
->prev_state
> inactive
) {
244 /* array has been stopped */
246 a
->container
->ss
->mark_clean(a
, a
->sync_pos
);
247 a
->next_state
= clear
;
250 if (a
->curr_state
== write_pending
) {
251 a
->container
->ss
->mark_dirty(a
);
252 a
->next_state
= active
;
254 if (a
->curr_state
== active_idle
) {
255 /* Set array to 'clean' FIRST, then
256 * a->ss->mark_clean(a);
257 * just ignore for now.
261 if (a
->curr_state
== readonly
) {
262 /* Well, I'm ready to handle things, so
263 * read-auto is OK. FIXME what if we really want
267 if (a
->resync_start
== ~0ULL)
268 a
->next_state
= read_auto
; /* array is clean */
270 a
->container
->ss
->mark_dirty(a
);
271 a
->next_state
= active
;
275 if (a
->curr_action
== idle
&&
276 a
->prev_action
== resync
) {
277 /* check resync_start to see if it is 'max' */
279 a
->container
->ss
->mark_sync(a
, a
->resync_start
);
283 if (a
->curr_action
== idle
&&
284 a
->prev_action
== recover
) {
285 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
286 a
->container
->ss
->set_disk(a
, mdi
->disk
.raid_disk
,
288 if (! (mdi
->curr_state
& DS_INSYNC
))
294 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
295 if (mdi
->curr_state
& DS_FAULTY
) {
296 a
->container
->ss
->set_disk(a
, mdi
->disk
.raid_disk
,
299 mdi
->next_state
= DS_REMOVE
;
303 if (check_degraded
) {
307 a
->container
->ss
->sync_metadata(a
);
309 /* Effect state changes in the array */
310 if (a
->next_state
!= bad_word
)
311 write_attr(array_states
[a
->next_state
], a
->info
.state_fd
);
312 if (a
->next_action
!= bad_action
)
313 write_attr(sync_actions
[a
->next_action
], a
->action_fd
);
314 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
315 if (mdi
->next_state
== DS_REMOVE
&& mdi
->state_fd
> 0) {
318 write_attr("-blocked", mdi
->state_fd
);
319 /* the kernel may not be able to immediately remove the
320 * disk, we can simply wait until the next event to try
323 remove_err
= write_attr("remove", mdi
->state_fd
);
325 close(mdi
->state_fd
);
329 if (mdi
->next_state
& DS_INSYNC
)
330 write_attr("+in_sync", mdi
->state_fd
);
333 /* move curr_ to prev_ */
334 a
->prev_state
= a
->curr_state
;
336 a
->prev_action
= a
->curr_action
;
338 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
339 mdi
->prev_state
= mdi
->curr_state
;
349 static int wait_and_act(struct active_array
*aa
, int pfd
, int nowait
)
353 struct active_array
*a
;
358 add_fd(&rfds
, &maxfd
, pfd
);
359 for (a
= aa
; a
; a
= a
->next
) {
362 /* once an array has been deactivated only the manager
363 * thread can make us care about it again
368 add_fd(&rfds
, &maxfd
, a
->info
.state_fd
);
369 add_fd(&rfds
, &maxfd
, a
->action_fd
);
370 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
371 add_fd(&rfds
, &maxfd
, mdi
->state_fd
);
375 rv
= select(maxfd
+1, &rfds
, NULL
, NULL
, NULL
);
380 if (FD_ISSET(pfd
, &rfds
)) {
383 ; // FIXME read from the pipe
387 for (a
= aa
; a
; a
= a
->next
) {
388 if (a
->replaces
&& !discard_this
) {
389 struct active_array
**ap
;
390 for (ap
= &a
->next
; *ap
&& *ap
!= a
->replaces
;
395 discard_this
= a
->replaces
;
399 rv
+= read_and_act(a
);
404 void do_monitor(struct supertype
*container
)
409 rv
= wait_and_act(container
->arrays
, container
->pipe
[0], first
);