xref: /freebsd/sys/geom/raid/tr_raid5.c (revision c243e4902be8df1e643c76b5f18b68bb77cc5268)
1 /*-
2  * Copyright (c) 2012 Alexander Motin <mav@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
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.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS 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 AUTHORS 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
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include <sys/param.h>
31 #include <sys/bio.h>
32 #include <sys/endian.h>
33 #include <sys/kernel.h>
34 #include <sys/kobj.h>
35 #include <sys/limits.h>
36 #include <sys/lock.h>
37 #include <sys/malloc.h>
38 #include <sys/mutex.h>
39 #include <sys/sysctl.h>
40 #include <sys/systm.h>
41 #include <geom/geom.h>
42 #include "geom/raid/g_raid.h"
43 #include "g_raid_tr_if.h"
44 
45 SYSCTL_DECL(_kern_geom_raid);
46 
47 static MALLOC_DEFINE(M_TR_RAID5, "tr_raid5_data", "GEOM_RAID RAID5 data");
48 
49 #define TR_RAID5_NONE 0
50 #define TR_RAID5_REBUILD 1
51 #define TR_RAID5_RESYNC 2
52 
53 #define TR_RAID5_F_DOING_SOME	0x1
54 #define TR_RAID5_F_LOCKED	0x2
55 #define TR_RAID5_F_ABORT	0x4
56 
57 struct g_raid_tr_raid5_object {
58 	struct g_raid_tr_object	 trso_base;
59 	int			 trso_starting;
60 	int			 trso_stopping;
61 	int			 trso_type;
62 	int			 trso_recover_slabs; /* slabs before rest */
63 	int			 trso_fair_io;
64 	int			 trso_meta_update;
65 	int			 trso_flags;
66 	struct g_raid_subdisk	*trso_failed_sd; /* like per volume */
67 	void			*trso_buffer;	 /* Buffer space */
68 	struct bio		 trso_bio;
69 };
70 
71 static g_raid_tr_taste_t g_raid_tr_taste_raid5;
72 static g_raid_tr_event_t g_raid_tr_event_raid5;
73 static g_raid_tr_start_t g_raid_tr_start_raid5;
74 static g_raid_tr_stop_t g_raid_tr_stop_raid5;
75 static g_raid_tr_iostart_t g_raid_tr_iostart_raid5;
76 static g_raid_tr_iodone_t g_raid_tr_iodone_raid5;
77 static g_raid_tr_kerneldump_t g_raid_tr_kerneldump_raid5;
78 static g_raid_tr_locked_t g_raid_tr_locked_raid5;
79 static g_raid_tr_free_t g_raid_tr_free_raid5;
80 
81 static kobj_method_t g_raid_tr_raid5_methods[] = {
82 	KOBJMETHOD(g_raid_tr_taste,	g_raid_tr_taste_raid5),
83 	KOBJMETHOD(g_raid_tr_event,	g_raid_tr_event_raid5),
84 	KOBJMETHOD(g_raid_tr_start,	g_raid_tr_start_raid5),
85 	KOBJMETHOD(g_raid_tr_stop,	g_raid_tr_stop_raid5),
86 	KOBJMETHOD(g_raid_tr_iostart,	g_raid_tr_iostart_raid5),
87 	KOBJMETHOD(g_raid_tr_iodone,	g_raid_tr_iodone_raid5),
88 	KOBJMETHOD(g_raid_tr_kerneldump, g_raid_tr_kerneldump_raid5),
89 	KOBJMETHOD(g_raid_tr_locked,	g_raid_tr_locked_raid5),
90 	KOBJMETHOD(g_raid_tr_free,	g_raid_tr_free_raid5),
91 	{ 0, 0 }
92 };
93 
94 static struct g_raid_tr_class g_raid_tr_raid5_class = {
95 	"RAID5",
96 	g_raid_tr_raid5_methods,
97 	sizeof(struct g_raid_tr_raid5_object),
98 	.trc_priority = 100
99 };
100 
101 static int
102 g_raid_tr_taste_raid5(struct g_raid_tr_object *tr, struct g_raid_volume *vol)
103 {
104 	struct g_raid_tr_raid5_object *trs;
105 	u_int qual;
106 
107 	trs = (struct g_raid_tr_raid5_object *)tr;
108 	qual = tr->tro_volume->v_raid_level_qualifier;
109 	if (tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID4 &&
110 	    qual >= 0 && qual <= 1) {
111 		/* RAID4 */
112 	} else if ((tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5 ||
113 	     tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5E ||
114 	     tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5EE ||
115 	     tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5R ||
116 	     tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID6 ||
117 	     tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAIDMDF) &&
118 	    qual >= 0 && qual <= 3) {
119 		/* RAID5/5E/5EE/5R/6/MDF */
120 	} else
121 		return (G_RAID_TR_TASTE_FAIL);
122 	trs->trso_starting = 1;
123 	return (G_RAID_TR_TASTE_SUCCEED);
124 }
125 
126 static int
127 g_raid_tr_update_state_raid5(struct g_raid_volume *vol,
128     struct g_raid_subdisk *sd)
129 {
130 	struct g_raid_tr_raid5_object *trs;
131 	struct g_raid_softc *sc;
132 	u_int s;
133 	int na, ns, nu;
134 
135 	sc = vol->v_softc;
136 	trs = (struct g_raid_tr_raid5_object *)vol->v_tr;
137 	if (trs->trso_stopping &&
138 	    (trs->trso_flags & TR_RAID5_F_DOING_SOME) == 0)
139 		s = G_RAID_VOLUME_S_STOPPED;
140 	else if (trs->trso_starting)
141 		s = G_RAID_VOLUME_S_STARTING;
142 	else {
143 		na = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_ACTIVE);
144 		ns = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_STALE) +
145 		     g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_RESYNC);
146 		nu = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_UNINITIALIZED);
147 		if (na == vol->v_disks_count)
148 			s = G_RAID_VOLUME_S_OPTIMAL;
149 		else if (na + ns == vol->v_disks_count ||
150 		    na + ns + nu == vol->v_disks_count /* XXX: Temporary. */)
151 			s = G_RAID_VOLUME_S_SUBOPTIMAL;
152 		else if (na == vol->v_disks_count - 1 ||
153 		    na + ns + nu == vol->v_disks_count)
154 			s = G_RAID_VOLUME_S_DEGRADED;
155 		else
156 			s = G_RAID_VOLUME_S_BROKEN;
157 	}
158 	if (s != vol->v_state) {
159 		g_raid_event_send(vol, G_RAID_VOLUME_S_ALIVE(s) ?
160 		    G_RAID_VOLUME_E_UP : G_RAID_VOLUME_E_DOWN,
161 		    G_RAID_EVENT_VOLUME);
162 		g_raid_change_volume_state(vol, s);
163 		if (!trs->trso_starting && !trs->trso_stopping)
164 			g_raid_write_metadata(sc, vol, NULL, NULL);
165 	}
166 	return (0);
167 }
168 
169 static int
170 g_raid_tr_event_raid5(struct g_raid_tr_object *tr,
171     struct g_raid_subdisk *sd, u_int event)
172 {
173 
174 	g_raid_tr_update_state_raid5(tr->tro_volume, sd);
175 	return (0);
176 }
177 
178 static int
179 g_raid_tr_start_raid5(struct g_raid_tr_object *tr)
180 {
181 	struct g_raid_tr_raid5_object *trs;
182 	struct g_raid_volume *vol;
183 
184 	trs = (struct g_raid_tr_raid5_object *)tr;
185 	vol = tr->tro_volume;
186 	trs->trso_starting = 0;
187 	g_raid_tr_update_state_raid5(vol, NULL);
188 	return (0);
189 }
190 
191 static int
192 g_raid_tr_stop_raid5(struct g_raid_tr_object *tr)
193 {
194 	struct g_raid_tr_raid5_object *trs;
195 	struct g_raid_volume *vol;
196 
197 	trs = (struct g_raid_tr_raid5_object *)tr;
198 	vol = tr->tro_volume;
199 	trs->trso_starting = 0;
200 	trs->trso_stopping = 1;
201 	g_raid_tr_update_state_raid5(vol, NULL);
202 	return (0);
203 }
204 
205 static void
206 g_raid_tr_iostart_raid5_read(struct g_raid_tr_object *tr, struct bio *bp)
207 {
208 	struct g_raid_volume *vol;
209 	struct g_raid_subdisk *sd;
210 	struct bio_queue_head queue;
211 	struct bio *cbp;
212 	char *addr;
213 	off_t offset, start, length, nstripe, remain;
214 	int no, pno, ddisks, pdisks, protate, pleft;
215 	u_int strip_size, lvl, qual;
216 
217 	vol = tr->tro_volume;
218 	addr = bp->bio_data;
219 	strip_size = vol->v_strip_size;
220 	lvl = tr->tro_volume->v_raid_level;
221 	qual = tr->tro_volume->v_raid_level_qualifier;
222 	protate = tr->tro_volume->v_rotate_parity;
223 
224 	/* Stripe number. */
225 	nstripe = bp->bio_offset / strip_size;
226 	/* Start position in stripe. */
227 	start = bp->bio_offset % strip_size;
228 	/* Number of data and parity disks. */
229 	if (lvl == G_RAID_VOLUME_RL_RAIDMDF)
230 		pdisks = tr->tro_volume->v_mdf_pdisks;
231 	else if (lvl == G_RAID_VOLUME_RL_RAID5EE ||
232 	    lvl == G_RAID_VOLUME_RL_RAID6)
233 		pdisks = 2;
234 	else
235 		pdisks = 1;
236 	ddisks = vol->v_disks_count - pdisks;
237 	/* Parity disk number. */
238 	if (lvl == G_RAID_VOLUME_RL_RAID4) {
239 		if (qual == 0)		/* P0 */
240 			pno = 0;
241 		else			/* PN */
242 			pno = ddisks;
243 		pleft = -1;
244 	} else {
245 		pno = (nstripe / (ddisks * protate)) % vol->v_disks_count;
246 		pleft = protate - (nstripe / ddisks) % protate;
247 		if (qual >= 2) {	/* PN/Left */
248 			pno = ddisks - pno;
249 			if (pno < 0)
250 				pno += vol->v_disks_count;
251 		}
252 	}
253 	/* Data disk number. */
254 	no = nstripe % ddisks;
255 	if (lvl == G_RAID_VOLUME_RL_RAID4) {
256 		if (qual == 0)
257 			no += pdisks;
258 	} else if (qual & 1) {	/* Continuation/Symmetric */
259 		no = (pno + pdisks + no) % vol->v_disks_count;
260 	} else if (no >= pno)	/* Restart/Asymmetric */
261 		no += pdisks;
262 	else
263 		no += imax(0, pno + pdisks - vol->v_disks_count);
264 	/* Stripe start position in disk. */
265 	offset = (nstripe / ddisks) * strip_size;
266 	/* Length of data to operate. */
267 	remain = bp->bio_length;
268 
269 	bioq_init(&queue);
270 	do {
271 		length = MIN(strip_size - start, remain);
272 		cbp = g_clone_bio(bp);
273 		if (cbp == NULL)
274 			goto failure;
275 		cbp->bio_offset = offset + start;
276 		cbp->bio_data = addr;
277 		cbp->bio_length = length;
278 		cbp->bio_caller1 = &vol->v_subdisks[no];
279 		bioq_insert_tail(&queue, cbp);
280 		no++;
281 		if (lvl == G_RAID_VOLUME_RL_RAID4) {
282 			no %= vol->v_disks_count;
283 			if (no == pno)
284 				no = (no + pdisks) % vol->v_disks_count;
285 		} else if (qual & 1) {	/* Continuation/Symmetric */
286 			no %= vol->v_disks_count;
287 			if (no == pno) {
288 				if ((--pleft) <= 0) {
289 					pleft += protate;
290 					if (qual < 2)	/* P0/Right */
291 						pno++;
292 					else		/* PN/Left */
293 						pno += vol->v_disks_count - 1;
294 					pno %= vol->v_disks_count;
295 				}
296 				no = (pno + pdisks) % vol->v_disks_count;
297 				offset += strip_size;
298 			}
299 		} else {		/* Restart/Asymmetric */
300 			if (no == pno)
301 				no += pdisks;
302 			if (no >= vol->v_disks_count) {
303 				no -= vol->v_disks_count;
304 				if ((--pleft) <= 0) {
305 					pleft += protate;
306 					if (qual < 2)	/* P0/Right */
307 						pno++;
308 					else		/* PN/Left */
309 						pno += vol->v_disks_count - 1;
310 					pno %= vol->v_disks_count;
311 				}
312 				if (no == pno)
313 					no += pdisks;
314 				else
315 					no += imax(0, pno + pdisks - vol->v_disks_count);
316 				offset += strip_size;
317 			}
318 		}
319 		remain -= length;
320 		addr += length;
321 		start = 0;
322 	} while (remain > 0);
323 	for (cbp = bioq_first(&queue); cbp != NULL;
324 	    cbp = bioq_first(&queue)) {
325 		bioq_remove(&queue, cbp);
326 		sd = cbp->bio_caller1;
327 		cbp->bio_caller1 = NULL;
328 		g_raid_subdisk_iostart(sd, cbp);
329 	}
330 	return;
331 failure:
332 	for (cbp = bioq_first(&queue); cbp != NULL;
333 	    cbp = bioq_first(&queue)) {
334 		bioq_remove(&queue, cbp);
335 		g_destroy_bio(cbp);
336 	}
337 	if (bp->bio_error == 0)
338 		bp->bio_error = ENOMEM;
339 	g_raid_iodone(bp, bp->bio_error);
340 }
341 
342 static void
343 g_raid_tr_iostart_raid5(struct g_raid_tr_object *tr, struct bio *bp)
344 {
345 	struct g_raid_volume *vol;
346 	struct g_raid_tr_raid5_object *trs;
347 
348 	vol = tr->tro_volume;
349 	trs = (struct g_raid_tr_raid5_object *)tr;
350 	if (vol->v_state < G_RAID_VOLUME_S_SUBOPTIMAL) {
351 		g_raid_iodone(bp, EIO);
352 		return;
353 	}
354 	switch (bp->bio_cmd) {
355 	case BIO_READ:
356 		g_raid_tr_iostart_raid5_read(tr, bp);
357 		break;
358 	case BIO_WRITE:
359 	case BIO_DELETE:
360 	case BIO_FLUSH:
361 		g_raid_iodone(bp, ENODEV);
362 		break;
363 	default:
364 		KASSERT(1 == 0, ("Invalid command here: %u (volume=%s)",
365 		    bp->bio_cmd, vol->v_name));
366 		break;
367 	}
368 }
369 
370 static void
371 g_raid_tr_iodone_raid5(struct g_raid_tr_object *tr,
372     struct g_raid_subdisk *sd, struct bio *bp)
373 {
374 	struct bio *pbp;
375 	int error;
376 
377 	pbp = bp->bio_parent;
378 	pbp->bio_inbed++;
379 	error = bp->bio_error;
380 	g_destroy_bio(bp);
381 	if (pbp->bio_children == pbp->bio_inbed) {
382 		pbp->bio_completed = pbp->bio_length;
383 		g_raid_iodone(pbp, error);
384 	}
385 }
386 
387 static int
388 g_raid_tr_kerneldump_raid5(struct g_raid_tr_object *tr,
389     void *virtual, vm_offset_t physical, off_t offset, size_t length)
390 {
391 
392 	return (ENODEV);
393 }
394 
395 static int
396 g_raid_tr_locked_raid5(struct g_raid_tr_object *tr, void *argp)
397 {
398 	struct bio *bp;
399 	struct g_raid_subdisk *sd;
400 
401 	bp = (struct bio *)argp;
402 	sd = (struct g_raid_subdisk *)bp->bio_caller1;
403 	g_raid_subdisk_iostart(sd, bp);
404 
405 	return (0);
406 }
407 
408 static int
409 g_raid_tr_free_raid5(struct g_raid_tr_object *tr)
410 {
411 	struct g_raid_tr_raid5_object *trs;
412 
413 	trs = (struct g_raid_tr_raid5_object *)tr;
414 
415 	if (trs->trso_buffer != NULL) {
416 		free(trs->trso_buffer, M_TR_RAID5);
417 		trs->trso_buffer = NULL;
418 	}
419 	return (0);
420 }
421 
422 G_RAID_TR_DECLARE(g_raid_tr_raid5);
423