xref: /freebsd/sys/geom/vinum/geom_vinum_init.c (revision f4b37ed0f8b307b1f3f0f630ca725d68f1dff30d)
1 /*-
2  * Copyright (c) 2004, 2007 Lukas Ertl
3  * Copyright (c) 2007, 2009 Ulf Lilleengen
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 #include <sys/param.h>
31 #include <sys/bio.h>
32 #include <sys/libkern.h>
33 #include <sys/malloc.h>
34 
35 #include <geom/geom.h>
36 #include <geom/vinum/geom_vinum_var.h>
37 #include <geom/vinum/geom_vinum.h>
38 
39 static int		 gv_sync(struct gv_volume *);
40 static int		 gv_rebuild_plex(struct gv_plex *);
41 static int		 gv_init_plex(struct gv_plex *);
42 static int		 gv_grow_plex(struct gv_plex *);
43 static int		 gv_sync_plex(struct gv_plex *, struct gv_plex *);
44 static struct gv_plex	*gv_find_good_plex(struct gv_volume *);
45 
46 void
47 gv_start_obj(struct g_geom *gp, struct gctl_req *req)
48 {
49 	struct gv_softc *sc;
50 	struct gv_volume *v;
51 	struct gv_plex *p;
52 	int *argc, *initsize;
53 	char *argv, buf[20];
54 	int i, type;
55 
56 	argc = gctl_get_paraml(req, "argc", sizeof(*argc));
57 	initsize = gctl_get_paraml(req, "initsize", sizeof(*initsize));
58 
59 	if (argc == NULL || *argc == 0) {
60 		gctl_error(req, "no arguments given");
61 		return;
62 	}
63 
64 	sc = gp->softc;
65 
66 	for (i = 0; i < *argc; i++) {
67 		snprintf(buf, sizeof(buf), "argv%d", i);
68 		argv = gctl_get_param(req, buf, NULL);
69 		if (argv == NULL)
70 			continue;
71 		type = gv_object_type(sc, argv);
72 		switch (type) {
73 		case GV_TYPE_VOL:
74 			v = gv_find_vol(sc, argv);
75 			if (v != NULL)
76 				gv_post_event(sc, GV_EVENT_START_VOLUME, v,
77 				    NULL, *initsize, 0);
78 			break;
79 
80 		case GV_TYPE_PLEX:
81 			p = gv_find_plex(sc, argv);
82 			if (p != NULL)
83 				gv_post_event(sc, GV_EVENT_START_PLEX, p, NULL,
84 				    *initsize, 0);
85 			break;
86 
87 		case GV_TYPE_SD:
88 		case GV_TYPE_DRIVE:
89 			/* XXX Not implemented, but what is the use? */
90 			gctl_error(req, "unable to start '%s' - not yet supported",
91 			    argv);
92 			return;
93 		default:
94 			gctl_error(req, "unknown object '%s'", argv);
95 			return;
96 		}
97 	}
98 }
99 
100 int
101 gv_start_plex(struct gv_plex *p)
102 {
103 	struct gv_volume *v;
104 	struct gv_plex *up;
105 	struct gv_sd *s;
106 	int error;
107 
108 	KASSERT(p != NULL, ("gv_start_plex: NULL p"));
109 
110 	error = 0;
111 	v = p->vol_sc;
112 
113 	/* RAID5 plexes can either be init, rebuilt or grown. */
114 	if (p->org == GV_PLEX_RAID5) {
115 		if (p->state > GV_PLEX_DEGRADED) {
116 			LIST_FOREACH(s, &p->subdisks, in_plex) {
117 				if (s->flags & GV_SD_GROW) {
118 					error = gv_grow_plex(p);
119 					return (error);
120 				}
121 			}
122 		} else if (p->state == GV_PLEX_DEGRADED) {
123 			error = gv_rebuild_plex(p);
124 		} else
125 			error = gv_init_plex(p);
126 	} else {
127 		/* We want to sync from the other plex if we're down. */
128 		if (p->state == GV_PLEX_DOWN && v->plexcount > 1) {
129 			up = gv_find_good_plex(v);
130 			if (up == NULL) {
131 				G_VINUM_DEBUG(1, "unable to find a good plex");
132 				return (ENXIO);
133 			}
134 			g_topology_lock();
135 			error = gv_access(v->provider, 1, 1, 0);
136 			if (error) {
137 				g_topology_unlock();
138 				G_VINUM_DEBUG(0, "sync from '%s' failed to "
139 				    "access volume: %d", up->name, error);
140 				return (error);
141 			}
142 			g_topology_unlock();
143 			error = gv_sync_plex(p, up);
144 			if (error)
145 				return (error);
146 		/*
147 		 * In case we have a stripe that is up, check whether it can be
148 		 * grown.
149 		 */
150 		} else if (p->org == GV_PLEX_STRIPED &&
151 		    p->state != GV_PLEX_DOWN) {
152 			LIST_FOREACH(s, &p->subdisks, in_plex) {
153 				if (s->flags & GV_SD_GROW) {
154 					error = gv_grow_plex(p);
155 					break;
156 				}
157 			}
158 		}
159 	}
160 	return (error);
161 }
162 
163 int
164 gv_start_vol(struct gv_volume *v)
165 {
166 	struct gv_plex *p;
167 	int error;
168 
169 	KASSERT(v != NULL, ("gv_start_vol: NULL v"));
170 
171 	error = 0;
172 
173 	if (v->plexcount == 0)
174 		return (ENXIO);
175 
176 	else if (v->plexcount == 1) {
177 		p = LIST_FIRST(&v->plexes);
178 		KASSERT(p != NULL, ("gv_start_vol: NULL p on %s", v->name));
179 		error = gv_start_plex(p);
180 	} else
181 		error = gv_sync(v);
182 
183 	return (error);
184 }
185 
186 /* Sync a plex p from the plex up.  */
187 static int
188 gv_sync_plex(struct gv_plex *p, struct gv_plex *up)
189 {
190 	int error;
191 
192 	KASSERT(p != NULL, ("%s: NULL p", __func__));
193 	KASSERT(up != NULL, ("%s: NULL up", __func__));
194 	if ((p == up) || (p->state == GV_PLEX_UP))
195 		return (0);
196 	if (p->flags & GV_PLEX_SYNCING ||
197 	    p->flags & GV_PLEX_REBUILDING ||
198 	    p->flags & GV_PLEX_GROWING) {
199 		return (EINPROGRESS);
200 	}
201 	p->synced = 0;
202 	p->flags |= GV_PLEX_SYNCING;
203 	G_VINUM_DEBUG(1, "starting sync of plex %s", p->name);
204 	error = gv_sync_request(up, p, p->synced,
205 	    MIN(GV_DFLT_SYNCSIZE, up->size - p->synced),
206 	    BIO_READ, NULL);
207 	if (error) {
208 		G_VINUM_DEBUG(0, "error syncing plex %s", p->name);
209 		return (error);
210 	}
211 	return (0);
212 }
213 
214 /* Return a good plex from volume v. */
215 static struct gv_plex *
216 gv_find_good_plex(struct gv_volume *v)
217 {
218 	struct gv_plex *up;
219 
220 	/* Find the plex that's up. */
221 	up = NULL;
222 	LIST_FOREACH(up, &v->plexes, in_volume) {
223 		if (up->state == GV_PLEX_UP)
224 			break;
225 	}
226 	/* Didn't find a good plex. */
227 	return (up);
228 }
229 
230 static int
231 gv_sync(struct gv_volume *v)
232 {
233 	struct gv_softc *sc;
234 	struct gv_plex *p, *up;
235 	int error;
236 
237 	KASSERT(v != NULL, ("gv_sync: NULL v"));
238 	sc = v->vinumconf;
239 	KASSERT(sc != NULL, ("gv_sync: NULL sc on %s", v->name));
240 
241 
242 	up = gv_find_good_plex(v);
243 	if (up == NULL)
244 		return (ENXIO);
245 	g_topology_lock();
246 	error = gv_access(v->provider, 1, 1, 0);
247 	if (error) {
248 		g_topology_unlock();
249 		G_VINUM_DEBUG(0, "sync from '%s' failed to access volume: %d",
250 		    up->name, error);
251 		return (error);
252 	}
253 	g_topology_unlock();
254 
255 	/* Go through the good plex, and issue BIO's to all other plexes. */
256 	LIST_FOREACH(p, &v->plexes, in_volume) {
257 		error = gv_sync_plex(p, up);
258 		if (error)
259 			break;
260 	}
261 	return (0);
262 }
263 
264 static int
265 gv_rebuild_plex(struct gv_plex *p)
266 {
267 	struct gv_drive *d;
268 	struct gv_sd *s;
269 	int error;
270 
271 	if (p->flags & GV_PLEX_SYNCING ||
272 	    p->flags & GV_PLEX_REBUILDING ||
273 	    p->flags & GV_PLEX_GROWING)
274 		return (EINPROGRESS);
275 	/*
276 	 * Make sure that all subdisks have consumers. We won't allow a rebuild
277 	 * unless every subdisk have one.
278 	 */
279 	LIST_FOREACH(s, &p->subdisks, in_plex) {
280 		d = s->drive_sc;
281 		if (d == NULL || (d->flags & GV_DRIVE_REFERENCED)) {
282 			G_VINUM_DEBUG(0, "unable to rebuild %s, subdisk(s) have"
283 			    " no drives", p->name);
284 			return (ENXIO);
285 		}
286 	}
287 	p->flags |= GV_PLEX_REBUILDING;
288 	p->synced = 0;
289 
290 	g_topology_assert_not();
291 	g_topology_lock();
292 	error = gv_access(p->vol_sc->provider, 1, 1, 0);
293 	if (error) {
294 		G_VINUM_DEBUG(0, "unable to access provider");
295 		return (0);
296 	}
297 	g_topology_unlock();
298 
299 	gv_parity_request(p, GV_BIO_REBUILD, 0);
300 	return (0);
301 }
302 
303 static int
304 gv_grow_plex(struct gv_plex *p)
305 {
306 	struct gv_volume *v;
307 	struct gv_sd *s;
308 	off_t origsize, origlength;
309 	int error, sdcount;
310 
311 	KASSERT(p != NULL, ("gv_grow_plex: NULL p"));
312 	v = p->vol_sc;
313 	KASSERT(v != NULL, ("gv_grow_plex: NULL v"));
314 
315 	if (p->flags & GV_PLEX_GROWING ||
316 	    p->flags & GV_PLEX_SYNCING ||
317 	    p->flags & GV_PLEX_REBUILDING)
318 		return (EINPROGRESS);
319 	g_topology_lock();
320 	error = gv_access(v->provider, 1, 1, 0);
321 	g_topology_unlock();
322 	if (error) {
323 		G_VINUM_DEBUG(0, "unable to access provider");
324 		return (error);
325 	}
326 
327 	/* XXX: This routine with finding origsize is used two other places as
328 	 * well, so we should create a function for it. */
329 	sdcount = p->sdcount;
330 	LIST_FOREACH(s, &p->subdisks, in_plex) {
331 		if (s->flags & GV_SD_GROW)
332 			sdcount--;
333 	}
334 	s = LIST_FIRST(&p->subdisks);
335 	if (s == NULL) {
336 		G_VINUM_DEBUG(0, "error growing plex without subdisks");
337 		return (GV_ERR_NOTFOUND);
338 	}
339 	p->flags |= GV_PLEX_GROWING;
340 	origsize = (sdcount - 1) * s->size;
341 	origlength = (sdcount - 1) * p->stripesize;
342 	p->synced = 0;
343 	G_VINUM_DEBUG(1, "starting growing of plex %s", p->name);
344 	gv_grow_request(p, 0, MIN(origlength, origsize), BIO_READ, NULL);
345 
346 	return (0);
347 }
348 
349 static int
350 gv_init_plex(struct gv_plex *p)
351 {
352 	struct gv_drive *d;
353 	struct gv_sd *s;
354 	int error;
355 	off_t start;
356 	caddr_t data;
357 
358 	KASSERT(p != NULL, ("gv_init_plex: NULL p"));
359 
360 	LIST_FOREACH(s, &p->subdisks, in_plex) {
361 		if (s->state == GV_SD_INITIALIZING)
362 			return (EINPROGRESS);
363 		gv_set_sd_state(s, GV_SD_INITIALIZING, GV_SETSTATE_FORCE);
364 		s->init_size = GV_DFLT_SYNCSIZE;
365 		start = s->drive_offset + s->initialized;
366 		d = s->drive_sc;
367 		if (d == NULL) {
368 			G_VINUM_DEBUG(0, "subdisk %s has no drive yet", s->name);
369 			break;
370 		}
371 		/*
372 		 * Take the lock here since we need to avoid a race in
373 		 * gv_init_request if the BIO is completed before the lock is
374 		 * released.
375 		 */
376 		g_topology_lock();
377 		error = g_access(d->consumer, 0, 1, 0);
378 		g_topology_unlock();
379 		if (error) {
380 			G_VINUM_DEBUG(0, "error accessing consumer when "
381 			    "initializing %s", s->name);
382 			break;
383 		}
384 		data = g_malloc(s->init_size, M_WAITOK | M_ZERO);
385 		gv_init_request(s, start, data, s->init_size);
386 	}
387 	return (0);
388 }
389