g_raid3.h revision 133808
1149871Sscottl/*-
2149871Sscottl * Copyright (c) 2004 Pawel Jakub Dawidek <pjd@FreeBSD.org>
3136849Sscottl * All rights reserved.
4136849Sscottl *
5136849Sscottl * Redistribution and use in source and binary forms, with or without
6136849Sscottl * modification, are permitted provided that the following conditions
7136849Sscottl * are met:
8136849Sscottl * 1. Redistributions of source code must retain the above copyright
9136849Sscottl *    notice, this list of conditions and the following disclaimer.
10136849Sscottl * 2. Redistributions in binary form must reproduce the above copyright
11136849Sscottl *    notice, this list of conditions and the following disclaimer in the
12136849Sscottl *    documentation and/or other materials provided with the distribution.
13136849Sscottl *
14136849Sscottl * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
15136849Sscottl * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16136849Sscottl * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17136849Sscottl * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
18136849Sscottl * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19136849Sscottl * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20136849Sscottl * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21136849Sscottl * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22136849Sscottl * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23136849Sscottl * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24136849Sscottl * SUCH DAMAGE.
25136849Sscottl *
26136849Sscottl * $FreeBSD: head/sys/geom/raid3/g_raid3.h 133808 2004-08-16 06:23:14Z pjd $
27136849Sscottl */
28136849Sscottl
29136849Sscottl#ifndef	_G_RAID3_H_
30149871Sscottl#define	_G_RAID3_H_
31149871Sscottl
32136849Sscottl#include <sys/endian.h>
33149871Sscottl#include <sys/md5.h>
34149871Sscottl
35136849Sscottl#define	G_RAID3_CLASS_NAME	"RAID3"
36149871Sscottl
37149871Sscottl#define	G_RAID3_MAGIC		"GEOM::RAID3"
38149871Sscottl#define	G_RAID3_VERSION		0
39136849Sscottl
40149871Sscottl#define	G_RAID3_DISK_FLAG_DIRTY		0x0000000000000001ULL
41149871Sscottl#define	G_RAID3_DISK_FLAG_SYNCHRONIZING	0x0000000000000002ULL
42149871Sscottl#define	G_RAID3_DISK_FLAG_FORCE_SYNC	0x0000000000000004ULL
43149871Sscottl#define	G_RAID3_DISK_FLAG_HARDCODED	0x0000000000000008ULL
44149871Sscottl#define	G_RAID3_DISK_FLAG_MASK		(G_RAID3_DISK_FLAG_DIRTY |	\
45149871Sscottl					 G_RAID3_DISK_FLAG_SYNCHRONIZING | \
46149871Sscottl					 G_RAID3_DISK_FLAG_FORCE_SYNC)
47149871Sscottl
48149871Sscottl#define	G_RAID3_DEVICE_FLAG_NOAUTOSYNC	0x0000000000000001ULL
49149871Sscottl#define	G_RAID3_DEVICE_FLAG_MASK	(G_RAID3_DEVICE_FLAG_NOAUTOSYNC)
50149871Sscottl
51149871Sscottl#ifdef _KERNEL
52136849Sscottlextern u_int g_raid3_debug;
53136849Sscottl
54136849Sscottl#define	G_RAID3_DEBUG(lvl, ...)	do {					\
55136849Sscottl	if (g_raid3_debug >= (lvl)) {					\
56136849Sscottl		printf("GEOM_RAID3");					\
57149871Sscottl		if (g_raid3_debug > 0)					\
58136849Sscottl			printf("[%u]", lvl);				\
59136849Sscottl		printf(": ");						\
60149871Sscottl		printf(__VA_ARGS__);					\
61149871Sscottl		printf("\n");						\
62136849Sscottl	}								\
63136849Sscottl} while (0)
64136849Sscottl#define	G_RAID3_LOGREQ(lvl, bp, ...)	do {				\
65136849Sscottl	if (g_raid3_debug >= (lvl)) {					\
66149871Sscottl		printf("GEOM_RAID3");					\
67149871Sscottl		if (g_raid3_debug > 0)					\
68136849Sscottl			printf("[%u]", lvl);				\
69136849Sscottl		printf(": ");						\
70136849Sscottl		printf(__VA_ARGS__);					\
71136849Sscottl		printf(" ");						\
72136849Sscottl		g_print_bio(bp);					\
73149871Sscottl		printf("\n");						\
74136849Sscottl	}								\
75136849Sscottl} while (0)
76149871Sscottl
77149871Sscottl#define	G_RAID3_MAX_IO_SIZE	(DFLTPHYS * 2)
78136849Sscottl
79149871Sscottl#define	G_RAID3_BIO_CFLAG_REGULAR	0x01
80136849Sscottl#define	G_RAID3_BIO_CFLAG_SYNC		0x02
81136849Sscottl#define	G_RAID3_BIO_CFLAG_PARITY	0x04
82149871Sscottl#define	G_RAID3_BIO_CFLAG_NODISK	0x08
83149871Sscottl#define	G_RAID3_BIO_CFLAG_REGSYNC	0x10
84136849Sscottl
85149871Sscottl#define	G_RAID3_BIO_PFLAG_DEGRADED	0x01
86136849Sscottl#define	G_RAID3_BIO_PFLAG_NOPARITY	0x02
87136849Sscottl
88149871Sscottl/*
89149871Sscottl * Informations needed for synchronization.
90136849Sscottl */
91149871Sscottlstruct g_raid3_disk_sync {
92136849Sscottl	struct g_consumer *ds_consumer;	/* Consumer connected to our device. */
93136849Sscottl	off_t		 ds_offset;	/* Offset of next request to send. */
94136849Sscottl	off_t		 ds_offset_done; /* Offset of already synchronized
95136849Sscottl					   region. */
96136849Sscottl	u_int		 ds_syncid;	/* Disk's synchronization ID. */
97136849Sscottl	u_char		*ds_data;
98136849Sscottl};
99136849Sscottl
100136849Sscottl/*
101136849Sscottl * Informations needed for synchronization.
102136849Sscottl */
103136849Sscottlstruct g_raid3_device_sync {
104136849Sscottl	struct g_geom	*ds_geom;	/* Synchronization geom. */
105136849Sscottl};
106149871Sscottl
107149871Sscottl#define	G_RAID3_DISK_STATE_NODISK		0
108136849Sscottl#define	G_RAID3_DISK_STATE_NONE			1
109136849Sscottl#define	G_RAID3_DISK_STATE_NEW			2
110136849Sscottl#define	G_RAID3_DISK_STATE_ACTIVE		3
111136849Sscottl#define	G_RAID3_DISK_STATE_STALE		4
112149871Sscottl#define	G_RAID3_DISK_STATE_SYNCHRONIZING	5
113149871Sscottl#define	G_RAID3_DISK_STATE_DISCONNECTED		6
114136849Sscottl#define	G_RAID3_DISK_STATE_DESTROY		7
115136849Sscottlstruct g_raid3_disk {
116136849Sscottl	u_int		 d_no;		/* Disk number. */
117136849Sscottl	struct g_consumer *d_consumer;	/* Consumer. */
118149871Sscottl	struct g_raid3_softc *d_softc;	/* Back-pointer to softc. */
119149871Sscottl	int		 d_state;	/* Disk state. */
120136849Sscottl	uint64_t	 d_flags;	/* Additional flags. */
121136849Sscottl	struct g_raid3_disk_sync d_sync; /* Sync information. */
122136849Sscottl	LIST_ENTRY(g_raid3_disk) d_next;
123136849Sscottl};
124149871Sscottl#define	d_name	d_consumer->provider->name
125149871Sscottl
126136849Sscottl#define	G_RAID3_EVENT_DONTWAIT	0x1
127136849Sscottl#define	G_RAID3_EVENT_WAIT	0x2
128136849Sscottl#define	G_RAID3_EVENT_DEVICE	0x4
129#define	G_RAID3_EVENT_DONE	0x8
130struct g_raid3_event {
131	struct g_raid3_disk	*e_disk;
132	int			 e_state;
133	int			 e_flags;
134	int			 e_error;
135	TAILQ_ENTRY(g_raid3_event) e_next;
136};
137
138#define	G_RAID3_DEVICE_FLAG_DESTROY	0x0100000000000000ULL
139#define	G_RAID3_DEVICE_FLAG_WAIT	0x0200000000000000ULL
140
141#define	G_RAID3_DEVICE_STATE_STARTING		0
142#define	G_RAID3_DEVICE_STATE_DEGRADED		1
143#define	G_RAID3_DEVICE_STATE_COMPLETE		2
144
145#define	G_RAID3_BUMP_ON_FIRST_WRITE		1
146#define	G_RAID3_BUMP_IMMEDIATELY		2
147
148struct g_raid3_softc {
149	u_int		sc_state;	/* Device state. */
150	uint64_t	sc_mediasize;	/* Device size. */
151	uint32_t	sc_sectorsize;	/* Sector size. */
152	uint64_t	sc_flags;	/* Additional flags. */
153
154	struct g_geom	*sc_geom;
155	struct g_provider *sc_provider;
156
157	uint32_t	sc_id;		/* Device unique ID. */
158
159	struct bio_queue_head sc_queue;
160	struct mtx	 sc_queue_mtx;
161	struct proc	*sc_worker;
162
163	struct g_raid3_disk *sc_disks;
164	u_int		sc_ndisks;	/* Number of disks. */
165	struct g_raid3_disk *sc_syncdisk;
166
167	uma_zone_t	sc_zone_64k;
168	uma_zone_t	sc_zone_16k;
169	uma_zone_t	sc_zone_4k;
170
171	u_int		sc_syncid;	/* Synchronization ID. */
172	int		sc_bump_syncid;
173	struct g_raid3_device_sync sc_sync;
174
175	TAILQ_HEAD(, g_raid3_event) sc_events;
176	struct mtx	sc_events_mtx;
177
178	struct callout	sc_callout;
179};
180#define	sc_name	sc_geom->name
181
182const char *g_raid3_get_diskname(struct g_raid3_disk *disk);
183u_int g_raid3_ndisks(struct g_raid3_softc *sc, int state);
184int g_raid3_destroy(struct g_raid3_softc *sc, boolean_t force);
185int g_raid3_event_send(void *arg, int state, int flags);
186struct g_raid3_metadata;
187void g_raid3_fill_metadata(struct g_raid3_disk *disk,
188    struct g_raid3_metadata *md);
189int g_raid3_clear_metadata(struct g_raid3_disk *disk);
190void g_raid3_update_metadata(struct g_raid3_disk *disk);
191
192g_ctl_req_t g_raid3_config;
193#endif	/* _KERNEL */
194
195struct g_raid3_metadata {
196	char		md_magic[16];	/* Magic value. */
197	uint32_t	md_version;	/* Version number. */
198	char		md_name[16];	/* Device name. */
199	uint32_t	md_id;		/* Device unique ID. */
200	uint16_t	md_no;		/* Component number. */
201	uint16_t	md_all;		/* Number of disks in device. */
202	uint32_t	md_syncid;	/* Synchronization ID. */
203	uint64_t	md_mediasize;	/* Size of whole device. */
204	uint32_t	md_sectorsize;	/* Sector size. */
205	uint64_t	md_sync_offset;	/* Synchronized offset. */
206	uint64_t	md_mflags;	/* Additional device flags. */
207	uint64_t	md_dflags;	/* Additional disk flags. */
208	char		md_provider[16]; /* Hardcoded provider. */
209	u_char		md_hash[16];	/* MD5 hash. */
210};
211static __inline void
212raid3_metadata_encode(struct g_raid3_metadata *md, u_char *data)
213{
214	MD5_CTX ctx;
215
216	bcopy(md->md_magic, data, 16);
217	le32enc(data + 16, md->md_version);
218	bcopy(md->md_name, data + 20, 16);
219	le32enc(data + 36, md->md_id);
220	le16enc(data + 40, md->md_no);
221	le16enc(data + 42, md->md_all);
222	le32enc(data + 44, md->md_syncid);
223	le64enc(data + 48, md->md_mediasize);
224	le32enc(data + 56, md->md_sectorsize);
225	le64enc(data + 60, md->md_sync_offset);
226	le64enc(data + 68, md->md_mflags);
227	le64enc(data + 76, md->md_dflags);
228	bcopy(md->md_provider, data + 84, 16);
229	MD5Init(&ctx);
230	MD5Update(&ctx, data, 100);
231	MD5Final(md->md_hash, &ctx);
232	bcopy(md->md_hash, data + 100, 16);
233}
234static __inline int
235raid3_metadata_decode(const u_char *data, struct g_raid3_metadata *md)
236{
237	MD5_CTX ctx;
238
239	bcopy(data, md->md_magic, 16);
240	md->md_version = le32dec(data + 16);
241	bcopy(data + 20, md->md_name, 16);
242	md->md_id = le32dec(data + 36);
243	md->md_no = le16dec(data + 40);
244	md->md_all = le16dec(data + 42);
245	md->md_syncid = le32dec(data + 44);
246	md->md_mediasize = le64dec(data + 48);
247	md->md_sectorsize = le32dec(data + 56);
248	md->md_sync_offset = le64dec(data + 60);
249	md->md_mflags = le64dec(data + 68);
250	md->md_dflags = le64dec(data + 76);
251	bcopy(data + 84, md->md_provider, 16);
252	bcopy(data + 100, md->md_hash, 16);
253	MD5Init(&ctx);
254	MD5Update(&ctx, data, 100);
255	MD5Final(md->md_hash, &ctx);
256	if (bcmp(md->md_hash, data + 100, 16) != 0)
257		return (EINVAL);
258	return (0);
259}
260
261static __inline void
262raid3_metadata_dump(const struct g_raid3_metadata *md)
263{
264	static const char hex[] = "0123456789abcdef";
265	char hash[16 * 2 + 1];
266	u_int i;
267
268	printf("     magic: %s\n", md->md_magic);
269	printf("   version: %u\n", (u_int)md->md_version);
270	printf("      name: %s\n", md->md_name);
271	printf("        id: %u\n", (u_int)md->md_id);
272	printf("        no: %u\n", (u_int)md->md_no);
273	printf("       all: %u\n", (u_int)md->md_all);
274	printf("    syncid: %u\n", (u_int)md->md_syncid);
275	printf(" mediasize: %jd\n", (intmax_t)md->md_mediasize);
276	printf("sectorsize: %u\n", (u_int)md->md_sectorsize);
277	printf("syncoffset: %jd\n", (intmax_t)md->md_sync_offset);
278	printf("    mflags:");
279	if (md->md_mflags == 0)
280		printf(" NONE");
281	else {
282		if ((md->md_mflags & G_RAID3_DEVICE_FLAG_NOAUTOSYNC) != 0)
283			printf(" NOAUTOSYNC");
284	}
285	printf("\n");
286	printf("    dflags:");
287	if (md->md_dflags == 0)
288		printf(" NONE");
289	else {
290		if ((md->md_dflags & G_RAID3_DISK_FLAG_DIRTY) != 0)
291			printf(" DIRTY");
292		if ((md->md_dflags & G_RAID3_DISK_FLAG_SYNCHRONIZING) != 0)
293			printf(" SYNCHRONIZING");
294		if ((md->md_dflags & G_RAID3_DISK_FLAG_FORCE_SYNC) != 0)
295			printf(" FORCE_SYNC");
296	}
297	printf("\n");
298	printf("hcprovider: %s\n", md->md_provider);
299	bzero(hash, sizeof(hash));
300	for (i = 0; i < 16; i++) {
301		hash[i * 2] = hex[md->md_hash[i] >> 4];
302		hash[i * 2 + 1] = hex[md->md_hash[i] & 0x0f];
303	}
304	printf("  MD5 hash: %s\n", hash);
305}
306#endif	/* !_G_RAID3_H_ */
307