]>
Commit | Line | Data |
---|---|---|
549e9569 NB |
1 | |
2 | /* | |
3 | * The management thread for monitoring active md arrays. | |
4 | * This thread does things which might block such as memory | |
5 | * allocation. | |
6 | * In particular: | |
7 | * | |
8 | * - Find out about new arrays in this container. | |
9 | * Allocate the data structures and open the files. | |
10 | * | |
11 | * For this we watch /proc/mdstat and find new arrays with | |
12 | * metadata type that confirms sharing. e.g. "md4" | |
13 | * When we find a new array we slip it into the list of | |
14 | * arrays and signal 'monitor' by writing to a pipe. | |
15 | * | |
16 | * - Respond to reshape requests by allocating new data structures | |
17 | * and opening new files. | |
18 | * | |
19 | * These come as a change to raid_disks. We allocate a new | |
20 | * version of the data structures and slip it into the list. | |
21 | * 'monitor' will notice and release the old version. | |
22 | * Changes to level, chunksize, layout.. do not need re-allocation. | |
23 | * Reductions in raid_disks don't really either, but we handle | |
24 | * them the same way for consistency. | |
25 | * | |
26 | * - When a device is added to the container, we add it to the metadata | |
27 | * as a spare. | |
28 | * | |
29 | * - assist with activating spares by opening relevant sysfs file. | |
30 | * | |
31 | * - Pass on metadata updates from external programs such as | |
32 | * mdadm creating a new array. | |
33 | * | |
34 | * This is most-messy. | |
35 | * It might involve adding a new array or changing the status of | |
36 | * a spare, or any reconfig that the kernel doesn't get involved in. | |
37 | * | |
38 | * The required updates are received via a named pipe. There will | |
39 | * be one named pipe for each container. Each message contains a | |
40 | * sync marker: 0x5a5aa5a5, A byte count, and the message. This is | |
41 | * passed to the metadata handler which will interpret and process it. | |
42 | * For 'DDF' messages are internal data blocks with the leading | |
43 | * 'magic number' signifying what sort of data it is. | |
44 | * | |
45 | */ | |
46 | ||
47 | /* | |
48 | * We select on /proc/mdstat and the named pipe. | |
49 | * We create new arrays or updated version of arrays and slip | |
50 | * them into the head of the list, then signal 'monitor' via a pipe write. | |
51 | * 'monitor' will notice and place the old array on a return list. | |
52 | * Metadata updates are placed on a queue just like they arrive | |
53 | * from the named pipe. | |
54 | * | |
55 | * When new arrays are found based on correct metadata string, we | |
56 | * need to identify them with an entry in the metadata. Maybe we require | |
57 | * the metadata to be mdX/NN when NN is the index into an appropriate table. | |
58 | * | |
59 | */ | |
60 | ||
61 | /* | |
62 | * List of tasks: | |
63 | * - Watch for spares to be added to the container, and write updated | |
64 | * metadata to them. | |
65 | * - Watch for new arrays using this container, confirm they match metadata | |
66 | * and if so, start monitoring them | |
67 | * - Watch for spares being added to monitored arrays. This shouldn't | |
68 | * happen, as we should do all the adding. Just remove them. | |
69 | * - Watch for change in raid-disks, chunk-size, etc. Update metadata and | |
70 | * start a reshape. | |
71 | */ | |
72 | #ifndef _GNU_SOURCE | |
73 | #define _GNU_SOURCE | |
74 | #endif | |
75 | #include "mdadm.h" | |
76 | #include "mdmon.h" | |
77 | #include <sys/socket.h> | |
1ed3f387 | 78 | #include <signal.h> |
549e9569 | 79 | |
2a0bb19e DW |
80 | static void close_aa(struct active_array *aa) |
81 | { | |
82 | struct mdinfo *d; | |
83 | ||
84 | for (d = aa->info.devs; d; d = d->next) | |
85 | close(d->state_fd); | |
86 | ||
87 | close(aa->action_fd); | |
88 | close(aa->info.state_fd); | |
89 | close(aa->resync_start_fd); | |
2a0bb19e DW |
90 | } |
91 | ||
549e9569 NB |
92 | static void free_aa(struct active_array *aa) |
93 | { | |
2a0bb19e DW |
94 | /* Note that this doesn't close fds if they are being used |
95 | * by a clone. ->container will be set for a clone | |
549e9569 | 96 | */ |
2a0bb19e DW |
97 | if (!aa->container) |
98 | close_aa(aa); | |
549e9569 NB |
99 | while (aa->info.devs) { |
100 | struct mdinfo *d = aa->info.devs; | |
101 | aa->info.devs = d->next; | |
102 | free(d); | |
103 | } | |
104 | free(aa); | |
105 | } | |
106 | ||
2a0bb19e DW |
107 | static void write_wakeup(struct supertype *c) |
108 | { | |
3e70c845 DW |
109 | static struct md_generic_cmd cmd = { .action = md_action_ping_monitor }; |
110 | int err; | |
111 | ||
112 | active_cmd = &cmd; | |
113 | ||
114 | /* send the monitor thread a pointer to the ping action */ | |
115 | write(c->mgr_pipe[1], &err, 1); | |
116 | read(c->mon_pipe[0], &err, 1); | |
2a0bb19e DW |
117 | } |
118 | ||
1ed3f387 NB |
119 | static void remove_old(void) |
120 | { | |
121 | if (discard_this) { | |
122 | discard_this->next = NULL; | |
123 | free_aa(discard_this); | |
124 | if (pending_discard == discard_this) | |
125 | pending_discard = NULL; | |
126 | discard_this = NULL; | |
127 | } | |
128 | } | |
129 | ||
549e9569 NB |
130 | static void replace_array(struct supertype *container, |
131 | struct active_array *old, | |
132 | struct active_array *new) | |
133 | { | |
134 | /* To replace an array, we add it to the top of the list | |
135 | * marked with ->replaces to point to the original. | |
136 | * 'monitor' will take the original out of the list | |
137 | * and put it on 'discard_this'. We take it from there | |
138 | * and discard it. | |
139 | */ | |
1ed3f387 | 140 | remove_old(); |
549e9569 | 141 | while (pending_discard) { |
1ed3f387 | 142 | write_wakeup(container); |
549e9569 NB |
143 | while (discard_this == NULL) |
144 | sleep(1); | |
1ed3f387 | 145 | remove_old(); |
549e9569 NB |
146 | } |
147 | pending_discard = old; | |
148 | new->replaces = old; | |
149 | new->next = container->arrays; | |
150 | container->arrays = new; | |
2a0bb19e | 151 | write_wakeup(container); |
549e9569 NB |
152 | } |
153 | ||
549e9569 NB |
154 | static void manage_container(struct mdstat_ent *mdstat, |
155 | struct supertype *container) | |
156 | { | |
157 | /* The only thing of interest here is if a new device | |
158 | * has been added to the container. We add it to the | |
159 | * array ignoring any metadata on it. | |
160 | * FIXME should we look for compatible metadata and take hints | |
161 | * about spare assignment.... probably not. | |
162 | * | |
163 | */ | |
164 | if (mdstat->devcnt != container->devcnt) { | |
165 | /* read /sys/block/NAME/md/dev-??/block/dev to find out | |
166 | * what is there, and compare with container->info.devs | |
167 | * To see what is removed and what is added. | |
168 | * These need to be remove from, or added to, the array | |
169 | */ | |
170 | // FIXME | |
171 | container->devcnt = mdstat->devcnt; | |
172 | } | |
173 | } | |
174 | ||
175 | static void manage_member(struct mdstat_ent *mdstat, | |
176 | struct active_array *a) | |
177 | { | |
178 | /* Compare mdstat info with known state of member array. | |
179 | * We do not need to look for device state changes here, that | |
180 | * is dealt with by the monitor. | |
181 | * | |
182 | * We just look for changes which suggest that a reshape is | |
183 | * being requested. | |
184 | * Unfortunately decreases in raid_disks don't show up in | |
185 | * mdstat until the reshape completes FIXME. | |
186 | */ | |
187 | // FIXME | |
188 | a->info.array.raid_disks = mdstat->raid_disks; | |
189 | a->info.array.chunk_size = mdstat->chunk_size; | |
190 | // MORE | |
191 | ||
192 | } | |
193 | ||
549e9569 | 194 | static void manage_new(struct mdstat_ent *mdstat, |
2a0bb19e DW |
195 | struct supertype *container, |
196 | struct active_array *victim) | |
549e9569 NB |
197 | { |
198 | /* A new array has appeared in this container. | |
199 | * Hopefully it is already recorded in the metadata. | |
200 | * Check, then create the new array to report it to | |
201 | * the monitor. | |
202 | */ | |
203 | ||
204 | struct active_array *new; | |
205 | struct mdinfo *mdi, *di; | |
cba0191b | 206 | char *inst; |
549e9569 NB |
207 | int i; |
208 | ||
209 | new = malloc(sizeof(*new)); | |
210 | ||
d52690ac NB |
211 | memset(new, 0, sizeof(*new)); |
212 | ||
549e9569 NB |
213 | new->devnum = mdstat->devnum; |
214 | ||
215 | new->prev_state = new->curr_state = new->next_state = inactive; | |
216 | new->prev_action= new->curr_action= new->next_action= idle; | |
217 | ||
218 | new->container = container; | |
219 | ||
cba0191b | 220 | inst = &mdstat->metadata_version[10+strlen(container->devname)+1]; |
549e9569 NB |
221 | |
222 | mdi = sysfs_read(-1, new->devnum, | |
223 | GET_LEVEL|GET_CHUNK|GET_DISKS| | |
224 | GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE); | |
225 | if (!mdi) { | |
226 | /* Eeek. Cannot monitor this array. | |
227 | * Mark it to be ignored by setting container to NULL | |
228 | */ | |
229 | new->container = NULL; | |
2a0bb19e | 230 | replace_array(container, victim, new); |
549e9569 NB |
231 | return; |
232 | } | |
233 | ||
234 | new->info.array = mdi->array; | |
235 | ||
236 | for (i = 0; i < new->info.array.raid_disks; i++) { | |
237 | struct mdinfo *newd = malloc(sizeof(*newd)); | |
238 | ||
239 | for (di = mdi->devs; di; di = di->next) | |
240 | if (i == di->disk.raid_disk) | |
241 | break; | |
242 | ||
243 | if (di) { | |
244 | memcpy(newd, di, sizeof(*newd)); | |
245 | ||
246 | sprintf(newd->sys_name, "rd%d", i); | |
247 | ||
248 | newd->state_fd = sysfs_open(new->devnum, | |
249 | newd->sys_name, | |
250 | "state"); | |
251 | ||
252 | newd->prev_state = read_dev_state(newd->state_fd); | |
253 | newd->curr_state = newd->curr_state; | |
254 | } else { | |
255 | newd->state_fd = -1; | |
256 | } | |
257 | newd->next = new->info.devs; | |
258 | new->info.devs = newd; | |
259 | } | |
260 | new->action_fd = sysfs_open(new->devnum, NULL, "sync_action"); | |
261 | new->info.state_fd = sysfs_open(new->devnum, NULL, "array_state"); | |
c052ba30 | 262 | new->resync_start_fd = sysfs_open(new->devnum, NULL, "resync_start"); |
77402e51 | 263 | new->resync_start = 0; |
549e9569 | 264 | |
4fa5aef9 | 265 | sysfs_free(mdi); |
549e9569 NB |
266 | // finds and compares. |
267 | if (container->ss->open_new(container, new, inst) < 0) { | |
268 | // FIXME close all those files | |
269 | new->container = NULL; | |
2a0bb19e | 270 | replace_array(container, victim, new); |
549e9569 NB |
271 | return; |
272 | } | |
2a0bb19e | 273 | replace_array(container, victim, new); |
549e9569 NB |
274 | return; |
275 | } | |
276 | ||
5d19760d | 277 | void manage(struct mdstat_ent *mdstat, struct supertype *container) |
549e9569 NB |
278 | { |
279 | /* We have just read mdstat and need to compare it with | |
280 | * the known active arrays. | |
281 | * Arrays with the wrong metadata are ignored. | |
282 | */ | |
283 | ||
284 | for ( ; mdstat ; mdstat = mdstat->next) { | |
285 | struct active_array *a; | |
286 | if (mdstat->devnum == container->devnum) { | |
287 | manage_container(mdstat, container); | |
288 | continue; | |
289 | } | |
290 | if (mdstat->metadata_version == NULL || | |
291 | strncmp(mdstat->metadata_version, "external:/", 10) != 0 || | |
292 | strncmp(mdstat->metadata_version+10, container->devname, | |
293 | strlen(container->devname)) != 0 || | |
294 | mdstat->metadata_version[10+strlen(container->devname)] | |
295 | != '/') | |
296 | /* Not for this array */ | |
297 | continue; | |
298 | /* Looks like a member of this container */ | |
5d19760d | 299 | for (a = container->arrays; a; a = a->next) { |
549e9569 NB |
300 | if (mdstat->devnum == a->devnum) { |
301 | if (a->container) | |
302 | manage_member(mdstat, a); | |
303 | break; | |
304 | } | |
305 | } | |
2a0bb19e DW |
306 | if (a == NULL || !a->container) |
307 | manage_new(mdstat, container, a); | |
549e9569 NB |
308 | } |
309 | } | |
310 | ||
3e70c845 DW |
311 | static int handle_message(struct supertype *container, struct md_message *msg) |
312 | { | |
313 | int err; | |
314 | struct md_generic_cmd *cmd = msg->buf; | |
315 | ||
316 | if (!cmd) | |
317 | return 0; | |
318 | ||
319 | switch (cmd->action) { | |
320 | case md_action_remove_device: | |
321 | ||
322 | /* forward to the monitor */ | |
323 | active_cmd = cmd; | |
324 | write(container->mgr_pipe[1], &err, 1); | |
325 | read(container->mon_pipe[0], &err, 1); | |
326 | return err; | |
327 | ||
328 | default: | |
329 | return -1; | |
330 | } | |
331 | } | |
332 | ||
333 | void read_sock(struct supertype *container) | |
549e9569 NB |
334 | { |
335 | int fd; | |
b109d928 DW |
336 | struct md_message msg; |
337 | int terminate = 0; | |
338 | long fl; | |
339 | int tmo = 3; /* 3 second timeout before hanging up the socket */ | |
549e9569 | 340 | |
3e70c845 | 341 | fd = accept(container->sock, NULL, NULL); |
549e9569 NB |
342 | if (fd < 0) |
343 | return; | |
b109d928 DW |
344 | |
345 | fl = fcntl(fd, F_GETFL, 0); | |
346 | fl |= O_NONBLOCK; | |
347 | fcntl(fd, F_SETFL, fl); | |
348 | ||
349 | do { | |
3e70c845 DW |
350 | int err; |
351 | ||
b109d928 DW |
352 | msg.buf = NULL; |
353 | ||
354 | /* read and validate the message */ | |
355 | if (receive_message(fd, &msg, tmo) == 0) { | |
3e70c845 DW |
356 | err = handle_message(container, &msg); |
357 | if (!err) | |
358 | ack(fd, msg.seq, tmo); | |
359 | else | |
360 | nack(fd, err, tmo); | |
b109d928 DW |
361 | } else { |
362 | terminate = 1; | |
363 | nack(fd, -1, tmo); | |
364 | } | |
365 | ||
366 | if (msg.buf) | |
367 | free(msg.buf); | |
368 | } while (!terminate); | |
369 | ||
549e9569 NB |
370 | close(fd); |
371 | } | |
1ed3f387 NB |
372 | |
373 | static int woke = 0; | |
374 | void wake_me(int sig) | |
375 | { | |
376 | woke = 1; | |
377 | } | |
378 | ||
e0d6609f NB |
379 | int exit_now = 0; |
380 | int manager_ready = 0; | |
549e9569 NB |
381 | void do_manager(struct supertype *container) |
382 | { | |
383 | struct mdstat_ent *mdstat; | |
1ed3f387 NB |
384 | sigset_t block, orig; |
385 | ||
386 | sigemptyset(&block); | |
387 | sigaddset(&block, SIGUSR1); | |
388 | ||
389 | signal(SIGUSR1, wake_me); | |
549e9569 NB |
390 | |
391 | do { | |
1ed3f387 NB |
392 | woke = 0; |
393 | ||
e0d6609f NB |
394 | if (exit_now) |
395 | exit(0); | |
396 | ||
549e9569 NB |
397 | mdstat = mdstat_read(1, 0); |
398 | ||
5d19760d | 399 | manage(mdstat, container); |
549e9569 | 400 | |
3e70c845 | 401 | read_sock(container); |
549e9569 | 402 | |
4fa5aef9 DW |
403 | free_mdstat(mdstat); |
404 | ||
1ed3f387 NB |
405 | remove_old(); |
406 | ||
e0d6609f | 407 | manager_ready = 1; |
1ed3f387 NB |
408 | sigprocmask(SIG_SETMASK, &block, &orig); |
409 | if (woke == 0) | |
410 | mdstat_wait_fd(container->sock, &orig); | |
411 | sigprocmask(SIG_SETMASK, &orig, NULL); | |
549e9569 NB |
412 | } while(1); |
413 | } |