xref: /freebsd/sys/geom/raid/tr_raid5.c (revision e796cc77c586c2955b2f3940dbf4991b31e8d289)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/bio.h>
34 #include <sys/endian.h>
35 #include <sys/kernel.h>
36 #include <sys/kobj.h>
37 #include <sys/limits.h>
38 #include <sys/lock.h>
39 #include <sys/malloc.h>
40 #include <sys/mutex.h>
41 #include <sys/sysctl.h>
42 #include <sys/systm.h>
43 #include <geom/geom.h>
44 #include "geom/raid/g_raid.h"
45 #include "g_raid_tr_if.h"
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_enable = 1,
99 	.trc_priority = 100
100 };
101 
102 static int
103 g_raid_tr_taste_raid5(struct g_raid_tr_object *tr, struct g_raid_volume *vol)
104 {
105 	struct g_raid_tr_raid5_object *trs;
106 	u_int qual;
107 
108 	trs = (struct g_raid_tr_raid5_object *)tr;
109 	qual = tr->tro_volume->v_raid_level_qualifier;
110 	if (tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID4 &&
111 	    (qual == G_RAID_VOLUME_RLQ_R4P0 ||
112 	     qual == G_RAID_VOLUME_RLQ_R4PN)) {
113 		/* RAID4 */
114 	} else if ((tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5 ||
115 	     tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5E ||
116 	     tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5EE ||
117 	     tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID5R ||
118 	     tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAID6 ||
119 	     tr->tro_volume->v_raid_level == G_RAID_VOLUME_RL_RAIDMDF) &&
120 	    (qual == G_RAID_VOLUME_RLQ_R5RA ||
121 	     qual == G_RAID_VOLUME_RLQ_R5RS ||
122 	     qual == G_RAID_VOLUME_RLQ_R5LA ||
123 	     qual == G_RAID_VOLUME_RLQ_R5LS)) {
124 		/* RAID5/5E/5EE/5R/6/MDF */
125 	} else
126 		return (G_RAID_TR_TASTE_FAIL);
127 	trs->trso_starting = 1;
128 	return (G_RAID_TR_TASTE_SUCCEED);
129 }
130 
131 static int
132 g_raid_tr_update_state_raid5(struct g_raid_volume *vol,
133     struct g_raid_subdisk *sd)
134 {
135 	struct g_raid_tr_raid5_object *trs;
136 	struct g_raid_softc *sc;
137 	u_int s;
138 	int na, ns, nu;
139 
140 	sc = vol->v_softc;
141 	trs = (struct g_raid_tr_raid5_object *)vol->v_tr;
142 	if (trs->trso_stopping &&
143 	    (trs->trso_flags & TR_RAID5_F_DOING_SOME) == 0)
144 		s = G_RAID_VOLUME_S_STOPPED;
145 	else if (trs->trso_starting)
146 		s = G_RAID_VOLUME_S_STARTING;
147 	else {
148 		na = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_ACTIVE);
149 		ns = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_STALE) +
150 		     g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_RESYNC);
151 		nu = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_UNINITIALIZED);
152 		if (na == vol->v_disks_count)
153 			s = G_RAID_VOLUME_S_OPTIMAL;
154 		else if (na + ns == vol->v_disks_count ||
155 		    na + ns + nu == vol->v_disks_count /* XXX: Temporary. */)
156 			s = G_RAID_VOLUME_S_SUBOPTIMAL;
157 		else if (na == vol->v_disks_count - 1 ||
158 		    na + ns + nu == vol->v_disks_count)
159 			s = G_RAID_VOLUME_S_DEGRADED;
160 		else
161 			s = G_RAID_VOLUME_S_BROKEN;
162 	}
163 	if (s != vol->v_state) {
164 		g_raid_event_send(vol, G_RAID_VOLUME_S_ALIVE(s) ?
165 		    G_RAID_VOLUME_E_UP : G_RAID_VOLUME_E_DOWN,
166 		    G_RAID_EVENT_VOLUME);
167 		g_raid_change_volume_state(vol, s);
168 		if (!trs->trso_starting && !trs->trso_stopping)
169 			g_raid_write_metadata(sc, vol, NULL, NULL);
170 	}
171 	return (0);
172 }
173 
174 static int
175 g_raid_tr_event_raid5(struct g_raid_tr_object *tr,
176     struct g_raid_subdisk *sd, u_int event)
177 {
178 
179 	g_raid_tr_update_state_raid5(tr->tro_volume, sd);
180 	return (0);
181 }
182 
183 static int
184 g_raid_tr_start_raid5(struct g_raid_tr_object *tr)
185 {
186 	struct g_raid_tr_raid5_object *trs;
187 	struct g_raid_volume *vol;
188 
189 	trs = (struct g_raid_tr_raid5_object *)tr;
190 	trs->trso_starting = 0;
191 	vol = tr->tro_volume;
192 	vol->v_read_only = 1;
193 	g_raid_tr_update_state_raid5(vol, NULL);
194 	return (0);
195 }
196 
197 static int
198 g_raid_tr_stop_raid5(struct g_raid_tr_object *tr)
199 {
200 	struct g_raid_tr_raid5_object *trs;
201 	struct g_raid_volume *vol;
202 
203 	trs = (struct g_raid_tr_raid5_object *)tr;
204 	vol = tr->tro_volume;
205 	trs->trso_starting = 0;
206 	trs->trso_stopping = 1;
207 	g_raid_tr_update_state_raid5(vol, NULL);
208 	return (0);
209 }
210 
211 static void
212 g_raid_tr_iostart_raid5_read(struct g_raid_tr_object *tr, struct bio *bp)
213 {
214 	struct g_raid_volume *vol;
215 	struct g_raid_subdisk *sd;
216 	struct bio_queue_head queue;
217 	struct bio *cbp;
218 	char *addr;
219 	off_t offset, start, length, nstripe, remain;
220 	int no, pno, ddisks, pdisks, protate, pleft;
221 	u_int strip_size, lvl, qual;
222 
223 	vol = tr->tro_volume;
224 	addr = bp->bio_data;
225 	strip_size = vol->v_strip_size;
226 	lvl = tr->tro_volume->v_raid_level;
227 	qual = tr->tro_volume->v_raid_level_qualifier;
228 	protate = tr->tro_volume->v_rotate_parity;
229 
230 	/* Stripe number. */
231 	nstripe = bp->bio_offset / strip_size;
232 	/* Start position in stripe. */
233 	start = bp->bio_offset % strip_size;
234 	/* Number of data and parity disks. */
235 	if (lvl == G_RAID_VOLUME_RL_RAIDMDF)
236 		pdisks = tr->tro_volume->v_mdf_pdisks;
237 	else if (lvl == G_RAID_VOLUME_RL_RAID5EE ||
238 	    lvl == G_RAID_VOLUME_RL_RAID6)
239 		pdisks = 2;
240 	else
241 		pdisks = 1;
242 	ddisks = vol->v_disks_count - pdisks;
243 	/* Parity disk number. */
244 	if (lvl == G_RAID_VOLUME_RL_RAID4) {
245 		if (qual == 0)		/* P0 */
246 			pno = 0;
247 		else			/* PN */
248 			pno = ddisks;
249 		pleft = -1;
250 	} else {
251 		pno = (nstripe / (ddisks * protate)) % vol->v_disks_count;
252 		pleft = protate - (nstripe / ddisks) % protate;
253 		if (qual >= 2) {	/* PN/Left */
254 			pno = ddisks - pno;
255 			if (pno < 0)
256 				pno += vol->v_disks_count;
257 		}
258 	}
259 	/* Data disk number. */
260 	no = nstripe % ddisks;
261 	if (lvl == G_RAID_VOLUME_RL_RAID4) {
262 		if (qual == 0)
263 			no += pdisks;
264 	} else if (qual & 1) {	/* Continuation/Symmetric */
265 		no = (pno + pdisks + no) % vol->v_disks_count;
266 	} else if (no >= pno)	/* Restart/Asymmetric */
267 		no += pdisks;
268 	else
269 		no += imax(0, pno + pdisks - vol->v_disks_count);
270 	/* Stripe start position in disk. */
271 	offset = (nstripe / ddisks) * strip_size;
272 	/* Length of data to operate. */
273 	remain = bp->bio_length;
274 
275 	bioq_init(&queue);
276 	do {
277 		length = MIN(strip_size - start, remain);
278 		cbp = g_clone_bio(bp);
279 		if (cbp == NULL)
280 			goto failure;
281 		cbp->bio_offset = offset + start;
282 		cbp->bio_data = addr;
283 		cbp->bio_length = length;
284 		cbp->bio_caller1 = &vol->v_subdisks[no];
285 		bioq_insert_tail(&queue, cbp);
286 		no++;
287 		if (lvl == G_RAID_VOLUME_RL_RAID4) {
288 			no %= vol->v_disks_count;
289 			if (no == pno)
290 				no = (no + pdisks) % vol->v_disks_count;
291 		} else if (qual & 1) {	/* Continuation/Symmetric */
292 			no %= vol->v_disks_count;
293 			if (no == pno) {
294 				if ((--pleft) <= 0) {
295 					pleft += protate;
296 					if (qual < 2)	/* P0/Right */
297 						pno++;
298 					else		/* PN/Left */
299 						pno += vol->v_disks_count - 1;
300 					pno %= vol->v_disks_count;
301 				}
302 				no = (pno + pdisks) % vol->v_disks_count;
303 				offset += strip_size;
304 			}
305 		} else {		/* Restart/Asymmetric */
306 			if (no == pno)
307 				no += pdisks;
308 			if (no >= vol->v_disks_count) {
309 				no -= vol->v_disks_count;
310 				if ((--pleft) <= 0) {
311 					pleft += protate;
312 					if (qual < 2)	/* P0/Right */
313 						pno++;
314 					else		/* PN/Left */
315 						pno += vol->v_disks_count - 1;
316 					pno %= vol->v_disks_count;
317 				}
318 				if (no == pno)
319 					no += pdisks;
320 				else
321 					no += imax(0, pno + pdisks - vol->v_disks_count);
322 				offset += strip_size;
323 			}
324 		}
325 		remain -= length;
326 		addr += length;
327 		start = 0;
328 	} while (remain > 0);
329 	while ((cbp = bioq_takefirst(&queue)) != NULL) {
330 		sd = cbp->bio_caller1;
331 		cbp->bio_caller1 = NULL;
332 		g_raid_subdisk_iostart(sd, cbp);
333 	}
334 	return;
335 failure:
336 	while ((cbp = bioq_takefirst(&queue)) != NULL)
337 		g_destroy_bio(cbp);
338 	if (bp->bio_error == 0)
339 		bp->bio_error = ENOMEM;
340 	g_raid_iodone(bp, bp->bio_error);
341 }
342 
343 static void
344 g_raid_tr_iostart_raid5(struct g_raid_tr_object *tr, struct bio *bp)
345 {
346 	struct g_raid_volume *vol;
347 
348 	vol = tr->tro_volume;
349 	if (vol->v_state < G_RAID_VOLUME_S_SUBOPTIMAL) {
350 		g_raid_iodone(bp, EIO);
351 		return;
352 	}
353 	switch (bp->bio_cmd) {
354 	case BIO_READ:
355 		g_raid_tr_iostart_raid5_read(tr, bp);
356 		break;
357 	case BIO_WRITE:
358 	case BIO_DELETE:
359 	case BIO_FLUSH:
360 		g_raid_iodone(bp, ENODEV);
361 		break;
362 	default:
363 		KASSERT(1 == 0, ("Invalid command here: %u (volume=%s)",
364 		    bp->bio_cmd, vol->v_name));
365 		break;
366 	}
367 }
368 
369 static void
370 g_raid_tr_iodone_raid5(struct g_raid_tr_object *tr,
371     struct g_raid_subdisk *sd, struct bio *bp)
372 {
373 	struct bio *pbp;
374 
375 	pbp = bp->bio_parent;
376 	if (pbp->bio_error == 0)
377 		pbp->bio_error = bp->bio_error;
378 	pbp->bio_inbed++;
379 	g_destroy_bio(bp);
380 	if (pbp->bio_children == pbp->bio_inbed) {
381 		pbp->bio_completed = pbp->bio_length;
382 		g_raid_iodone(pbp, pbp->bio_error);
383 	}
384 }
385 
386 static int
387 g_raid_tr_kerneldump_raid5(struct g_raid_tr_object *tr,
388     void *virtual, vm_offset_t physical, off_t offset, size_t length)
389 {
390 
391 	return (ENODEV);
392 }
393 
394 static int
395 g_raid_tr_locked_raid5(struct g_raid_tr_object *tr, void *argp)
396 {
397 	struct bio *bp;
398 	struct g_raid_subdisk *sd;
399 
400 	bp = (struct bio *)argp;
401 	sd = (struct g_raid_subdisk *)bp->bio_caller1;
402 	g_raid_subdisk_iostart(sd, bp);
403 
404 	return (0);
405 }
406 
407 static int
408 g_raid_tr_free_raid5(struct g_raid_tr_object *tr)
409 {
410 	struct g_raid_tr_raid5_object *trs;
411 
412 	trs = (struct g_raid_tr_raid5_object *)tr;
413 
414 	if (trs->trso_buffer != NULL) {
415 		free(trs->trso_buffer, M_TR_RAID5);
416 		trs->trso_buffer = NULL;
417 	}
418 	return (0);
419 }
420 
421 G_RAID_TR_DECLARE(raid5, "RAID5");
422