sys/geom/journal/g_journal.h

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2005-2006 Pawel Jakub Dawidek <pjd@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  * $FreeBSD$
  29  */
  30
  31 #ifndef _G_JOURNAL_H_
  32 #define _G_JOURNAL_H_
  33
  34 #include <sys/endian.h>
  35 #include <sys/md5.h>
  36 #ifdef _KERNEL
  37 #include <sys/bio.h>
  38 #endif
  39
  40 #define G_JOURNAL_CLASS_NAME    "JOURNAL"
  41
  42 #define G_JOURNAL_MAGIC         "GEOM::JOURNAL"
  43 /*
  44  * Version history:
  45  * 0 - Initial version number.
  46  */
  47 #define G_JOURNAL_VERSION       0
  48
  49 #ifdef _KERNEL
  50 extern int g_journal_debug;
  51
  52 #define GJ_DEBUG(lvl, ...) \
  53     _GEOM_DEBUG("GEOM_JOURNAL", g_journal_debug, (lvl), NULL, __VA_ARGS__)
  54 #define GJ_LOGREQ(lvl, bp, ...) \
  55     _GEOM_DEBUG("GEOM_JOURNAL", g_journal_debug, (lvl), (bp), __VA_ARGS__)
  56
  57 #define JEMPTY(sc)      ((sc)->sc_journal_offset -                      \
  58                          (sc)->sc_jprovider->sectorsize ==              \
  59                          (sc)->sc_active.jj_offset &&                   \
  60                          (sc)->sc_current_count == 0)
  61
  62 #define GJ_BIO_REGULAR          0x00
  63 #define GJ_BIO_READ             0x01
  64 #define GJ_BIO_JOURNAL          0x02
  65 #define GJ_BIO_COPY             0x03
  66 #define GJ_BIO_MASK             0x0f
  67
  68 #if 0
  69 #define GJF_BIO_DONT_FREE       0x10
  70 #define GJF_BIO_MASK            0xf0
  71 #endif
  72
  73 #define GJF_DEVICE_HARDCODED            0x0001
  74 #define GJF_DEVICE_DESTROY              0x0010
  75 #define GJF_DEVICE_SWITCH               0x0020
  76 #define GJF_DEVICE_BEFORE_SWITCH        0x0040
  77 #define GJF_DEVICE_CLEAN                0x0080
  78 #define GJF_DEVICE_CHECKSUM             0x0100
  79
  80 #define GJ_HARD_LIMIT           64
  81
  82 /*
  83  * We keep pointers to journaled data in bio structure and because we
  84  * need to store two off_t values (offset in data provider and offset in
  85  * journal), we have to borrow bio_completed field for this.
  86  */
  87 #define bio_joffset     bio_completed
  88 /*
  89  * Use bio_caller1 field as a pointer in queue.
  90  */
  91 #define bio_next        bio_caller1
  92
  93 /*
  94  * There are two such structures maintained inside each journaled device.
  95  * One describes active part of the journal, were recent requests are stored.
  96  * The second describes the last consistent part of the journal with requests
  97  * that are copied to the destination provider.
  98  */
  99 struct g_journal_journal {
 100         struct bio      *jj_queue;      /* Cached journal entries. */
 101         off_t            jj_offset;     /* Journal's start offset. */
 102 };
 103
 104 struct g_journal_softc {
 105         uint32_t         sc_id;
 106         uint8_t          sc_type;
 107         uint8_t          sc_orig_type;
 108         struct g_geom   *sc_geom;
 109         u_int            sc_flags;
 110         struct mtx       sc_mtx;
 111         off_t            sc_mediasize;
 112         u_int            sc_sectorsize;
 113 #define GJ_FLUSH_DATA           0x01
 114 #define GJ_FLUSH_JOURNAL        0x02
 115         u_int            sc_bio_flush;
 116
 117         uint32_t         sc_journal_id;
 118         uint32_t         sc_journal_next_id;
 119         int              sc_journal_copying;
 120         off_t            sc_journal_offset;
 121         off_t            sc_journal_previous_id;
 122
 123         struct bio_queue_head sc_back_queue;
 124         struct bio_queue_head sc_regular_queue;
 125
 126         struct bio_queue_head sc_delayed_queue;
 127         int              sc_delayed_count;
 128
 129         struct bio      *sc_current_queue;
 130         int              sc_current_count;
 131
 132         struct bio      *sc_flush_queue;
 133         int              sc_flush_count;
 134         int              sc_flush_in_progress;
 135
 136         struct bio      *sc_copy_queue;
 137         int              sc_copy_in_progress;
 138
 139         struct g_consumer *sc_dconsumer;
 140         struct g_consumer *sc_jconsumer;
 141
 142         struct g_journal_journal sc_inactive;
 143         struct g_journal_journal sc_active;
 144
 145         off_t            sc_jstart;     /* Journal space start offset. */
 146         off_t            sc_jend;       /* Journal space end offset. */
 147
 148         struct callout   sc_callout;
 149         struct proc     *sc_worker;
 150
 151         struct root_hold_token *sc_rootmount;
 152 };
 153 #define sc_dprovider    sc_dconsumer->provider
 154 #define sc_jprovider    sc_jconsumer->provider
 155 #define sc_name         sc_dprovider->name
 156
 157 #define GJQ_INSERT_HEAD(head, bp)       do {                            \
 158         (bp)->bio_next = (head);                                        \
 159         (head) = (bp);                                                  \
 160 } while (0)
 161 #define GJQ_INSERT_AFTER(head, bp, pbp) do {                            \
 162         if ((pbp) == NULL)                                              \
 163                 GJQ_INSERT_HEAD(head, bp);                              \
 164         else {                                                          \
 165                 (bp)->bio_next = (pbp)->bio_next;                       \
 166                 (pbp)->bio_next = (bp);                                 \
 167         }                                                               \
 168 } while (0)
 169 #define GJQ_LAST(head, bp) do {                                         \
 170         struct bio *_bp;                                                \
 171                                                                         \
 172         if ((head) == NULL) {                                           \
 173                 (bp) = (head);                                          \
 174                 break;                                                  \
 175         }                                                               \
 176         for (_bp = (head); _bp->bio_next != NULL; _bp = _bp->bio_next)  \
 177                 continue;                                               \
 178         (bp) = (_bp);                                                   \
 179 } while (0)
 180 #define GJQ_FIRST(head) (head)
 181 #define GJQ_REMOVE(head, bp)    do {                                    \
 182         struct bio *_bp;                                                \
 183                                                                         \
 184         if ((head) == (bp)) {                                           \
 185                 (head) = (bp)->bio_next;                                \
 186                 (bp)->bio_next = NULL;                                  \
 187                 break;                                                  \
 188         }                                                               \
 189         for (_bp = (head); _bp->bio_next != NULL; _bp = _bp->bio_next) {\
 190                 if (_bp->bio_next == (bp))                              \
 191                         break;                                          \
 192         }                                                               \
 193         KASSERT(_bp->bio_next != NULL, ("NULL bio_next"));              \
 194         KASSERT(_bp->bio_next == (bp), ("bio_next != bp"));             \
 195         _bp->bio_next = (bp)->bio_next;                                 \
 196         (bp)->bio_next = NULL;                                          \
 197 } while (0)
 198 #define GJQ_FOREACH(head, bp)                                           \
 199         for ((bp) = (head); (bp) != NULL; (bp) = (bp)->bio_next)
 200
 201 #define GJ_HEADER_MAGIC "GJHDR"
 202
 203 struct g_journal_header {
 204         char            jh_magic[sizeof(GJ_HEADER_MAGIC)];
 205         uint32_t        jh_journal_id;
 206         uint32_t        jh_journal_next_id;
 207 } __packed;
 208
 209 struct g_journal_entry {
 210         uint64_t        je_joffset;
 211         uint64_t        je_offset;
 212         uint64_t        je_length;
 213 } __packed;
 214
 215 #define GJ_RECORD_HEADER_MAGIC          "GJRHDR"
 216 #define GJ_RECORD_HEADER_NENTRIES       (20)
 217 #define GJ_RECORD_MAX_SIZE(sc)  \
 218         ((sc)->sc_jprovider->sectorsize + GJ_RECORD_HEADER_NENTRIES * MAXPHYS)
 219 #define GJ_VALIDATE_OFFSET(offset, sc)  do {                            \
 220         if ((offset) + GJ_RECORD_MAX_SIZE(sc) >= (sc)->sc_jend) {       \
 221                 (offset) = (sc)->sc_jstart;                             \
 222                 GJ_DEBUG(2, "Starting from the beginning (%s).",                \
 223                     (sc)->sc_name);                                     \
 224         }                                                               \
 225 } while (0)
 226
 227 struct g_journal_record_header {
 228         char            jrh_magic[sizeof(GJ_RECORD_HEADER_MAGIC)];
 229         uint32_t        jrh_journal_id;
 230         uint16_t        jrh_nentries;
 231         u_char          jrh_sum[8];
 232         struct g_journal_entry jrh_entries[GJ_RECORD_HEADER_NENTRIES];
 233 } __packed;
 234
 235 typedef int (g_journal_clean_t)(struct mount *mp);
 236 typedef void (g_journal_dirty_t)(struct g_consumer *cp);
 237
 238 struct g_journal_desc {
 239         const char              *jd_fstype;
 240         g_journal_clean_t       *jd_clean;
 241         g_journal_dirty_t       *jd_dirty;
 242 };
 243
 244 /* Supported file systems. */
 245 extern const struct g_journal_desc g_journal_ufs;
 246
 247 #define GJ_TIMER_START(lvl, bt) do {                                    \
 248         if (g_journal_debug >= (lvl))                                   \
 249                 binuptime(bt);                                          \
 250 } while (0)
 251 #define GJ_TIMER_STOP(lvl, bt, ...)     do {                            \
 252         if (g_journal_debug >= (lvl)) {                                 \
 253                 struct bintime _bt2;                                    \
 254                 struct timeval _tv;                                     \
 255                                                                         \
 256                 binuptime(&_bt2);                                       \
 257                 bintime_sub(&_bt2, bt);                                 \
 258                 bintime2timeval(&_bt2, &_tv);                           \
 259                 printf("GEOM_JOURNAL");                                 \
 260                 if (g_journal_debug > 0)                                \
 261                         printf("[%u]", lvl);                            \
 262                 printf(": ");                                           \
 263                 printf(__VA_ARGS__);                                    \
 264                 printf(": %jd.%06jds\n", (intmax_t)_tv.tv_sec,          \
 265                     (intmax_t)_tv.tv_usec);                             \
 266         }                                                               \
 267 } while (0)
 268 #endif  /* _KERNEL */
 269
 270 #define GJ_TYPE_DATA            0x01
 271 #define GJ_TYPE_JOURNAL         0x02
 272 #define GJ_TYPE_COMPLETE        (GJ_TYPE_DATA|GJ_TYPE_JOURNAL)
 273
 274 #define GJ_FLAG_CLEAN           0x01
 275 #define GJ_FLAG_CHECKSUM        0x02
 276
 277 struct g_journal_metadata {
 278         char            md_magic[16];   /* Magic value. */
 279         uint32_t        md_version;     /* Version number. */
 280         uint32_t        md_id;          /* Journal unique ID. */
 281         uint8_t         md_type;        /* Provider type. */
 282         uint64_t        md_jstart;      /* Journal space start offset. */
 283         uint64_t        md_jend;        /* Journal space end offset. */
 284         uint64_t        md_joffset;     /* Last known consistent journal offset. */
 285         uint32_t        md_jid;         /* Last known consistent journal ID. */
 286         uint64_t        md_flags;       /* Journal flags. */
 287         char            md_provider[16]; /* Hardcoded provider. */
 288         uint64_t        md_provsize;    /* Provider's size. */
 289         u_char          md_hash[16];    /* MD5 hash. */
 290 };
 291 static __inline void
 292 journal_metadata_encode(struct g_journal_metadata *md, u_char *data)
 293 {
 294         MD5_CTX ctx;
 295
 296         bcopy(md->md_magic, data, 16);
 297         le32enc(data + 16, md->md_version);
 298         le32enc(data + 20, md->md_id);
 299         *(data + 24) = md->md_type;
 300         le64enc(data + 25, md->md_jstart);
 301         le64enc(data + 33, md->md_jend);
 302         le64enc(data + 41, md->md_joffset);
 303         le32enc(data + 49, md->md_jid);
 304         le64enc(data + 53, md->md_flags);
 305         bcopy(md->md_provider, data + 61, 16);
 306         le64enc(data + 77, md->md_provsize);
 307         MD5Init(&ctx);
 308         MD5Update(&ctx, data, 85);
 309         MD5Final(md->md_hash, &ctx);
 310         bcopy(md->md_hash, data + 85, 16);
 311 }
 312 static __inline int
 313 journal_metadata_decode_v0(const u_char *data, struct g_journal_metadata *md)
 314 {
 315         MD5_CTX ctx;
 316
 317         md->md_id = le32dec(data + 20);
 318         md->md_type = *(data + 24);
 319         md->md_jstart = le64dec(data + 25);
 320         md->md_jend = le64dec(data + 33);
 321         md->md_joffset = le64dec(data + 41);
 322         md->md_jid = le32dec(data + 49);
 323         md->md_flags = le64dec(data + 53);
 324         bcopy(data + 61, md->md_provider, 16);
 325         md->md_provsize = le64dec(data + 77);
 326         MD5Init(&ctx);
 327         MD5Update(&ctx, data, 85);
 328         MD5Final(md->md_hash, &ctx);
 329         if (bcmp(md->md_hash, data + 85, 16) != 0)
 330                 return (EINVAL);
 331         return (0);
 332 }
 333 static __inline int
 334 journal_metadata_decode(const u_char *data, struct g_journal_metadata *md)
 335 {
 336         int error;
 337
 338         bcopy(data, md->md_magic, 16);
 339         md->md_version = le32dec(data + 16);
 340         switch (md->md_version) {
 341         case 0:
 342                 error = journal_metadata_decode_v0(data, md);
 343                 break;
 344         default:
 345                 error = EINVAL;
 346                 break;
 347         }
 348         return (error);
 349 }
 350
 351 static __inline void
 352 journal_metadata_dump(const struct g_journal_metadata *md)
 353 {
 354         static const char hex[] = "0123456789abcdef";
 355         char hash[16 * 2 + 1];
 356         u_int i;
 357
 358         printf("     magic: %s\n", md->md_magic);
 359         printf("   version: %u\n", (u_int)md->md_version);
 360         printf("        id: %u\n", (u_int)md->md_id);
 361         printf("      type: %u\n", (u_int)md->md_type);
 362         printf("     start: %ju\n", (uintmax_t)md->md_jstart);
 363         printf("       end: %ju\n", (uintmax_t)md->md_jend);
 364         printf("   joffset: %ju\n", (uintmax_t)md->md_joffset);
 365         printf("       jid: %u\n", (u_int)md->md_jid);
 366         printf("     flags: %u\n", (u_int)md->md_flags);
 367         printf("hcprovider: %s\n", md->md_provider);
 368         printf("  provsize: %ju\n", (uintmax_t)md->md_provsize);
 369         bzero(hash, sizeof(hash));
 370         for (i = 0; i < 16; i++) {
 371                 hash[i * 2] = hex[md->md_hash[i] >> 4];
 372                 hash[i * 2 + 1] = hex[md->md_hash[i] & 0x0f];
 373         }
 374         printf("  MD5 hash: %s\n", hash);
 375 }
 376 #endif  /* !_G_JOURNAL_H_ */