xref: /freebsd/sys/geom/raid/tr_raid5.c (revision 06c3fb2749bda94cb5201f81ffdb8fa6c3161b2e)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2012 Alexander Motin <mav@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/param.h>
30 #include <sys/bio.h>
31 #include <sys/endian.h>
32 #include <sys/kernel.h>
33 #include <sys/kobj.h>
34 #include <sys/limits.h>
35 #include <sys/lock.h>
36 #include <sys/malloc.h>
37 #include <sys/mutex.h>
38 #include <sys/sysctl.h>
39 #include <sys/systm.h>
40 #include <geom/geom.h>
41 #include "geom/raid/g_raid.h"
42 #include "g_raid_tr_if.h"
43 
44 static MALLOC_DEFINE(M_TR_RAID5, "tr_raid5_data", "GEOM_RAID RAID5 data");
45 
46 #define TR_RAID5_NONE 0
47 #define TR_RAID5_REBUILD 1
48 #define TR_RAID5_RESYNC 2
49 
50 #define TR_RAID5_F_DOING_SOME	0x1
51 #define TR_RAID5_F_LOCKED	0x2
52 #define TR_RAID5_F_ABORT	0x4
53 
54 struct g_raid_tr_raid5_object {
55 	struct g_raid_tr_object	 trso_base;
56 	int			 trso_starting;
57 	int			 trso_stopping;
58 	int			 trso_type;
59 	int			 trso_recover_slabs; /* slabs before rest */
60 	int			 trso_fair_io;
61 	int			 trso_meta_update;
62 	int			 trso_flags;
63 	struct g_raid_subdisk	*trso_failed_sd; /* like per volume */
64 	void			*trso_buffer;	 /* Buffer space */
65 	struct bio		 trso_bio;
66 };
67 
68 static g_raid_tr_taste_t g_raid_tr_taste_raid5;
69 static g_raid_tr_event_t g_raid_tr_event_raid5;
70 static g_raid_tr_start_t g_raid_tr_start_raid5;
71 static g_raid_tr_stop_t g_raid_tr_stop_raid5;
72 static g_raid_tr_iostart_t g_raid_tr_iostart_raid5;
73 static g_raid_tr_iodone_t g_raid_tr_iodone_raid5;
74 static g_raid_tr_kerneldump_t g_raid_tr_kerneldump_raid5;
75 static g_raid_tr_locked_t g_raid_tr_locked_raid5;
76 static g_raid_tr_free_t g_raid_tr_free_raid5;
77 
78 static kobj_method_t g_raid_tr_raid5_methods[] = {
79 	KOBJMETHOD(g_raid_tr_taste,	g_raid_tr_taste_raid5),
80 	KOBJMETHOD(g_raid_tr_event,	g_raid_tr_event_raid5),
81 	KOBJMETHOD(g_raid_tr_start,	g_raid_tr_start_raid5),
82 	KOBJMETHOD(g_raid_tr_stop,	g_raid_tr_stop_raid5),
83 	KOBJMETHOD(g_raid_tr_iostart,	g_raid_tr_iostart_raid5),
84 	KOBJMETHOD(g_raid_tr_iodone,	g_raid_tr_iodone_raid5),
85 	KOBJMETHOD(g_raid_tr_kerneldump, g_raid_tr_kerneldump_raid5),
86 	KOBJMETHOD(g_raid_tr_locked,	g_raid_tr_locked_raid5),
87 	KOBJMETHOD(g_raid_tr_free,	g_raid_tr_free_raid5),
88 	{ 0, 0 }
89 };
90 
91 static struct g_raid_tr_class g_raid_tr_raid5_class = {
92 	"RAID5",
93 	g_raid_tr_raid5_methods,
94 	sizeof(struct g_raid_tr_raid5_object),
95 	.trc_enable = 1,
96 	.trc_priority = 100
97 };
98 
99 static int
100 g_raid_tr_taste_raid5(struct g_raid_tr_object *tr, struct g_raid_volume *vol)
101 {
102 	struct g_raid_tr_raid5_object *trs;
103 	u_int qual;
104 
105 	trs = (struct g_raid_tr_raid5_object *)tr;
106 	qual = tr->tro_volume->v_raid_level_qualifier;
107 	if (tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID4 &&
108 	    (qual == G_RAID_VOLUME_RLQ_R4P0 ||
109 	     qual == G_RAID_VOLUME_RLQ_R4PN)) {
110 		/* RAID4 */
111 	} else if ((tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5 ||
112 	     tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5E ||
113 	     tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5EE ||
114 	     tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5R ||
115 	     tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID6 ||
116 	     tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAIDMDF) &&
117 	    (qual == G_RAID_VOLUME_RLQ_R5RA ||
118 	     qual == G_RAID_VOLUME_RLQ_R5RS ||
119 	     qual == G_RAID_VOLUME_RLQ_R5LA ||
120 	     qual == G_RAID_VOLUME_RLQ_R5LS)) {
121 		/* RAID5/5E/5EE/5R/6/MDF */
122 	} else
123 		return (G_RAID_TR_TASTE_FAIL);
124 	trs->trso_starting = 1;
125 	return (G_RAID_TR_TASTE_SUCCEED);
126 }
127 
128 static int
129 g_raid_tr_update_state_raid5(struct g_raid_volume *vol,
130     struct g_raid_subdisk *sd)
131 {
132 	struct g_raid_tr_raid5_object *trs;
133 	struct g_raid_softc *sc;
134 	u_int s;
135 	int na, ns, nu;
136 
137 	sc = vol->v_softc;
138 	trs = (struct g_raid_tr_raid5_object *)vol->v_tr;
139 	if (trs->trso_stopping &&
140 	    (trs->trso_flags & TR_RAID5_F_DOING_SOME) == 0)
141 		s = G_RAID_VOLUME_S_STOPPED;
142 	else if (trs->trso_starting)
143 		s = G_RAID_VOLUME_S_STARTING;
144 	else {
145 		na = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_ACTIVE);
146 		ns = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_STALE) +
147 		     g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_RESYNC);
148 		nu = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_UNINITIALIZED);
149 		if (na == vol->v_disks_count)
150 			s = G_RAID_VOLUME_S_OPTIMAL;
151 		else if (na + ns == vol->v_disks_count ||
152 		    na + ns + nu == vol->v_disks_count /* XXX: Temporary. */)
153 			s = G_RAID_VOLUME_S_SUBOPTIMAL;
154 		else if (na == vol->v_disks_count - 1 ||
155 		    na + ns + nu == vol->v_disks_count)
156 			s = G_RAID_VOLUME_S_DEGRADED;
157 		else
158 			s = G_RAID_VOLUME_S_BROKEN;
159 	}
160 	if (s != vol->v_state) {
161 		g_raid_event_send(vol, G_RAID_VOLUME_S_ALIVE(s) ?
162 		    G_RAID_VOLUME_E_UP : G_RAID_VOLUME_E_DOWN,
163 		    G_RAID_EVENT_VOLUME);
164 		g_raid_change_volume_state(vol, s);
165 		if (!trs->trso_starting && !trs->trso_stopping)
166 			g_raid_write_metadata(sc, vol, NULL, NULL);
167 	}
168 	return (0);
169 }
170 
171 static int
172 g_raid_tr_event_raid5(struct g_raid_tr_object *tr,
173     struct g_raid_subdisk *sd, u_int event)
174 {
175 
176 	g_raid_tr_update_state_raid5(tr->tro_volume, sd);
177 	return (0);
178 }
179 
180 static int
181 g_raid_tr_start_raid5(struct g_raid_tr_object *tr)
182 {
183 	struct g_raid_tr_raid5_object *trs;
184 	struct g_raid_volume *vol;
185 
186 	trs = (struct g_raid_tr_raid5_object *)tr;
187 	trs->trso_starting = 0;
188 	vol = tr->tro_volume;
189 	vol->v_read_only = 1;
190 	g_raid_tr_update_state_raid5(vol, NULL);
191 	return (0);
192 }
193 
194 static int
195 g_raid_tr_stop_raid5(struct g_raid_tr_object *tr)
196 {
197 	struct g_raid_tr_raid5_object *trs;
198 	struct g_raid_volume *vol;
199 
200 	trs = (struct g_raid_tr_raid5_object *)tr;
201 	vol = tr->tro_volume;
202 	trs->trso_starting = 0;
203 	trs->trso_stopping = 1;
204 	g_raid_tr_update_state_raid5(vol, NULL);
205 	return (0);
206 }
207 
208 static void
209 g_raid_tr_iostart_raid5_read(struct g_raid_tr_object *tr, struct bio *bp)
210 {
211 	struct g_raid_volume *vol;
212 	struct g_raid_subdisk *sd;
213 	struct bio_queue_head queue;
214 	struct bio *cbp;
215 	char *addr;
216 	off_t offset, start, length, nstripe, remain;
217 	int no, pno, ddisks, pdisks, protate, pleft;
218 	u_int strip_size, lvl, qual;
219 
220 	vol = tr->tro_volume;
221 	addr = bp->bio_data;
222 	strip_size = vol->v_strip_size;
223 	lvl = tr->tro_volume->v_raid_level;
224 	qual = tr->tro_volume->v_raid_level_qualifier;
225 	protate = tr->tro_volume->v_rotate_parity;
226 
227 	/* Stripe number. */
228 	nstripe = bp->bio_offset / strip_size;
229 	/* Start position in stripe. */
230 	start = bp->bio_offset % strip_size;
231 	/* Number of data and parity disks. */
232 	if (lvl == G_RAID_VOLUME_RL_RAIDMDF)
233 		pdisks = tr->tro_volume->v_mdf_pdisks;
234 	else if (lvl == G_RAID_VOLUME_RL_RAID5EE ||
235 	    lvl == G_RAID_VOLUME_RL_RAID6)
236 		pdisks = 2;
237 	else
238 		pdisks = 1;
239 	ddisks = vol->v_disks_count - pdisks;
240 	/* Parity disk number. */
241 	if (lvl == G_RAID_VOLUME_RL_RAID4) {
242 		if (qual == 0)		/* P0 */
243 			pno = 0;
244 		else			/* PN */
245 			pno = ddisks;
246 		pleft = -1;
247 	} else {
248 		pno = (nstripe / (ddisks * protate)) % vol->v_disks_count;
249 		pleft = protate - (nstripe / ddisks) % protate;
250 		if (qual >= 2) {	/* PN/Left */
251 			pno = ddisks - pno;
252 			if (pno < 0)
253 				pno += vol->v_disks_count;
254 		}
255 	}
256 	/* Data disk number. */
257 	no = nstripe % ddisks;
258 	if (lvl == G_RAID_VOLUME_RL_RAID4) {
259 		if (qual == 0)
260 			no += pdisks;
261 	} else if (qual & 1) {	/* Continuation/Symmetric */
262 		no = (pno + pdisks + no) % vol->v_disks_count;
263 	} else if (no >= pno)	/* Restart/Asymmetric */
264 		no += pdisks;
265 	else
266 		no += imax(0, pno + pdisks - vol->v_disks_count);
267 	/* Stripe start position in disk. */
268 	offset = (nstripe / ddisks) * strip_size;
269 	/* Length of data to operate. */
270 	remain = bp->bio_length;
271 
272 	bioq_init(&queue);
273 	do {
274 		length = MIN(strip_size - start, remain);
275 		cbp = g_clone_bio(bp);
276 		if (cbp == NULL)
277 			goto failure;
278 		cbp->bio_offset = offset + start;
279 		cbp->bio_data = addr;
280 		cbp->bio_length = length;
281 		cbp->bio_caller1 = &vol->v_subdisks[no];
282 		bioq_insert_tail(&queue, cbp);
283 		no++;
284 		if (lvl == G_RAID_VOLUME_RL_RAID4) {
285 			no %= vol->v_disks_count;
286 			if (no == pno)
287 				no = (no + pdisks) % vol->v_disks_count;
288 		} else if (qual & 1) {	/* Continuation/Symmetric */
289 			no %= vol->v_disks_count;
290 			if (no == pno) {
291 				if ((--pleft) <= 0) {
292 					pleft += protate;
293 					if (qual < 2)	/* P0/Right */
294 						pno++;
295 					else		/* PN/Left */
296 						pno += vol->v_disks_count - 1;
297 					pno %= vol->v_disks_count;
298 				}
299 				no = (pno + pdisks) % vol->v_disks_count;
300 				offset += strip_size;
301 			}
302 		} else {		/* Restart/Asymmetric */
303 			if (no == pno)
304 				no += pdisks;
305 			if (no >= vol->v_disks_count) {
306 				no -= vol->v_disks_count;
307 				if ((--pleft) <= 0) {
308 					pleft += protate;
309 					if (qual < 2)	/* P0/Right */
310 						pno++;
311 					else		/* PN/Left */
312 						pno += vol->v_disks_count - 1;
313 					pno %= vol->v_disks_count;
314 				}
315 				if (no == pno)
316 					no += pdisks;
317 				else
318 					no += imax(0, pno + pdisks - vol->v_disks_count);
319 				offset += strip_size;
320 			}
321 		}
322 		remain -= length;
323 		addr += length;
324 		start = 0;
325 	} while (remain > 0);
326 	while ((cbp = bioq_takefirst(&queue)) != NULL) {
327 		sd = cbp->bio_caller1;
328 		cbp->bio_caller1 = NULL;
329 		g_raid_subdisk_iostart(sd, cbp);
330 	}
331 	return;
332 failure:
333 	while ((cbp = bioq_takefirst(&queue)) != NULL)
334 		g_destroy_bio(cbp);
335 	if (bp->bio_error == 0)
336 		bp->bio_error = ENOMEM;
337 	g_raid_iodone(bp, bp->bio_error);
338 }
339 
340 static void
341 g_raid_tr_iostart_raid5(struct g_raid_tr_object *tr, struct bio *bp)
342 {
343 	struct g_raid_volume *vol;
344 
345 	vol = tr->tro_volume;
346 	if (vol->v_state < G_RAID_VOLUME_S_SUBOPTIMAL) {
347 		g_raid_iodone(bp, EIO);
348 		return;
349 	}
350 	switch (bp->bio_cmd) {
351 	case BIO_READ:
352 		g_raid_tr_iostart_raid5_read(tr, bp);
353 		break;
354 	case BIO_WRITE:
355 	case BIO_DELETE:
356 	case BIO_FLUSH:
357 	case BIO_SPEEDUP:
358 		g_raid_iodone(bp, ENODEV);
359 		break;
360 	default:
361 		KASSERT(1 == 0, ("Invalid command here: %u (volume=%s)",
362 		    bp->bio_cmd, vol->v_name));
363 		break;
364 	}
365 }
366 
367 static void
368 g_raid_tr_iodone_raid5(struct g_raid_tr_object *tr,
369     struct g_raid_subdisk *sd, struct bio *bp)
370 {
371 	struct bio *pbp;
372 
373 	pbp = bp->bio_parent;
374 	if (pbp->bio_error == 0)
375 		pbp->bio_error = bp->bio_error;
376 	pbp->bio_inbed++;
377 	g_destroy_bio(bp);
378 	if (pbp->bio_children == pbp->bio_inbed) {
379 		pbp->bio_completed = pbp->bio_length;
380 		g_raid_iodone(pbp, pbp->bio_error);
381 	}
382 }
383 
384 static int
385 g_raid_tr_kerneldump_raid5(struct g_raid_tr_object *tr, void *virtual,
386     off_t offset, size_t length)
387 {
388 
389 	return (ENODEV);
390 }
391 
392 static int
393 g_raid_tr_locked_raid5(struct g_raid_tr_object *tr, void *argp)
394 {
395 	struct bio *bp;
396 	struct g_raid_subdisk *sd;
397 
398 	bp = (struct bio *)argp;
399 	sd = (struct g_raid_subdisk *)bp->bio_caller1;
400 	g_raid_subdisk_iostart(sd, bp);
401 
402 	return (0);
403 }
404 
405 static int
406 g_raid_tr_free_raid5(struct g_raid_tr_object *tr)
407 {
408 	struct g_raid_tr_raid5_object *trs;
409 
410 	trs = (struct g_raid_tr_raid5_object *)tr;
411 
412 	if (trs->trso_buffer != NULL) {
413 		free(trs->trso_buffer, M_TR_RAID5);
414 		trs->trso_buffer = NULL;
415 	}
416 	return (0);
417 }
418 
419 G_RAID_TR_DECLARE(raid5, "RAID5");
420