sys/sys/journal.h

   1 /*
   2  * Copyright (c) 2004 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   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  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/sys/journal.h,v 1.13 2007/05/09 00:53:35 dillon Exp $
  35  */
  36
  37 #ifndef _SYS_JOURNAL_H_
  38 #define _SYS_JOURNAL_H_
  39
  40 #ifndef _SYS_TYPES_H_
  41 #include <sys/types.h>
  42 #endif
  43 #ifndef _SYS_TIME_H_
  44 #include <sys/time.h>
  45 #endif
  46
  47 /*
  48  * Physical file format (binary)
  49  *
  50  * All raw records are 128-bit aligned, but all record sizes are actual.
  51  * This means that any scanning code must 16-byte-align the recsize field
  52  * when calculating skips.  The top level raw record has a header and a
  53  * trailer to allow both forwards and backwards scanning of the journal.
  54  * The alignment requirement allows the worker thread FIFO reservation
  55  * API to operate efficiently, amoung other things.
  56  *
  57  * Logical data stream records are usually no larger then the journal's
  58  * in-memory FIFO, since the journal's transactional APIs return contiguous
  59  * blocks of buffer space and since logical stream records are used to avoid
  60  * stalls when concurrent blocking operations are being written to the journal.
  61  * Programs can depend on a logical stream record being a 'reasonable' size.
  62  *
  63  * Multiple logical data streams may operate concurrently in the journal,
  64  * reflecting the fact that the system may be executing multiple blocking
  65  * operations on the filesystem all at the same time.  These logical data
  66  * streams are short-lived transactional entities which use a 13 bit id
  67  * plus a transaction start bit, end bit, and abort bit.
  68  *
  69  * Stream identifiers in the 0x00-0xFF range are special and not used for
  70  * normal transactional commands.
  71  *
  72  * Stream id 0x00 indicates that no other streams should be active at that
  73  * point in the journal, which helps the journaling code detect corruption.
  74  *
  75  * Stream id 0x01 is used for pad.  Pads are used to align data on convenient
  76  * boundaries and to deal with dead space.
  77  *
  78  * Stream id 0x02 indicates a discontinuity in the streamed data and typically
  79  * contains information relating to the reason for the discontinuity.
  80  * JTYPE_ASSOCIATE and JTYPE_DISASSOCIATE are usually emplaced in stream 0x02.
  81  *
  82  * Stream id 0x03 may be used to annotate the journal with text comments
  83  * via mountctl commands.  This can be extremely useful to note situations
  84  * that may help with later recovery or audit operations.
  85  *
  86  * Stream id 0x04-0x7F are reserved by DragonFly for future protocol expansion.
  87  *
  88  * Stream id 0x80-0xFF may be used for third-party protocol expansion.
  89  *
  90  * Stream id's 0x0100-0x1FFF typically represent short-lived transactions
  91  * (i.e. an id may be reused once the previous use has completed).  The
  92  * journaling system runs through these id's sequentially which means that
  93  * the journaling code can handle up to 8192-256 = 7936 simultanious
  94  * transactions at any given moment.
  95  *
  96  * The sequence number field is context-sensitive.  It is typically used by
  97  * a journaling stream to provide an incrementing counter and/or timestamp
  98  * so recovery utilities can determine if any data is missing.
  99  *
 100  * The check word in the trailer may be used to provide an integrity check
 101  * on the journaled data.  A value of 0 always means that no check word
 102  * has been calculated.
 103  *
 104  * The journal_rawrecbeg structure MUST be a multiple of 16 bytes.
 105  * The journal_rawrecend structure MUST be a multiple of 8 bytes.
 106  *
 107  * NOTE: PAD RECORD SPECIAL CASE.  Pad records can be 16 bytes and have the
 108  * rawrecend structure overlayed on the sequence number field of the
 109  * rawrecbeg structure.  This is necessary because stream records are
 110  * 16 byte aligned, not 24 byte aligned, and dead space is not allowed.
 111  * So the pad record must fit into any dead space.  THEREFORE, THE TRANSID
 112  * FIELD FOR A PAD RECORD MUST BE IGNORED.
 113  *
 114  * NOTE: ENDIAN HANDLING.  Data records can be in little or big endian form.
 115  * The receiver detects the state by observing the 'begmagic' field.  Each
 116  * direction in a full-duplex connection can be operating with different
 117  * endianess.  Checksum data is always calculated on the raw record (including
 118  * dead space) in a byte-stream fashion, and then converted to the transmit
 119  * endianess like everything else.  If the receiver's endianess is different
 120  * it must convert it back to host normal form to compare it against the
 121  * calculated checksum.
 122  */
 123 struct journal_rawrecbeg {
 124         u_int16_t begmagic;     /* recovery scan, endianess detection */
 125         u_int16_t streamid;     /* start/stop bits and stream identifier */
 126         int32_t recsize;        /* stream data block (incls beg & end) */
 127         int64_t transid;        /* sequence number or transaction id */
 128         /* ADDITIONAL DATA */
 129 };
 130
 131 struct journal_rawrecend {
 132         u_int16_t endmagic;     /* recovery scan, endianess detection */
 133         u_int16_t check;        /* check word or 0 */
 134         int32_t recsize;        /* same as rawrecbeg->recsize, for rev scan */
 135 };
 136
 137 struct journal_ackrecord {
 138         struct journal_rawrecbeg        rbeg;
 139         int32_t                         filler0;
 140         int32_t                         filler1;
 141         struct journal_rawrecend        rend;
 142 };
 143
 144 /*
 145  * Constants for stream record magic numbers.    The incomplete magic
 146  * number code is used internally by the memory FIFO reservation API
 147  * and worker thread, allowing a block of space in the journaling
 148  * stream (aka a stream block) to be reserved and then populated without
 149  * stalling other threads doing their own reservation and population.
 150  */
 151 #define JREC_BEGMAGIC           0x1234
 152 #define JREC_ENDMAGIC           0xCDEF
 153 #define JREC_INCOMPLETEMAGIC    0xFFFF
 154
 155 /*
 156  * Stream ids are 14 bits.  The top 2 bits specify when a new logical
 157  * stream is being created or an existing logical stream is being terminated.
 158  * A single raw stream record will set both the BEGIN and END bits if the
 159  * entire transaction is encapsulated in a single stream record.
 160  */
 161 #define JREC_STREAMCTL_MASK     0xE000
 162 #define JREC_STREAMCTL_BEGIN    0x8000  /* start a new logical stream */
 163 #define JREC_STREAMCTL_END      0x4000  /* terminate a logical stream */
 164 #define JREC_STREAMCTL_ABORTED  0x2000
 165
 166 #define JREC_STREAMID_MASK      0x1FFF
 167 #define JREC_STREAMID_SYNCPT    (JREC_STREAMCTL_BEGIN|JREC_STREAMCTL_END|0x0000)
 168 #define JREC_STREAMID_PAD       (JREC_STREAMCTL_BEGIN|JREC_STREAMCTL_END|0x0001)
 169 #define JREC_STREAMID_DISCONT   0x0002  /* discontinuity */
 170 #define JREC_STREAMID_ANNOTATE  0x0003  /* annotation */
 171 #define JREC_STREAMID_ACK       0x0004  /* acknowledgement */
 172 #define JREC_STREAMID_RESTART   0x0005  /* disctoninuity - journal restart */
 173                                 /* 0x0006-0x007F reserved by DragonFly */
 174                                 /* 0x0080-0x00FF for third party use */
 175 #define JREC_STREAMID_JMIN      0x0100  /* lowest allowed general id */
 176 #define JREC_STREAMID_JMAX      0x2000  /* (one past the highest allowed id) */
 177
 178 #define JREC_DEFAULTSIZE        64      /* reasonable initial reservation */
 179 #define JREC_MINRECSIZE         16      /* (after alignment) */
 180 #define JREC_MAXRECSIZE         (128*1024*1024)
 181
 182 /*
 183  * Each logical journaling stream typically represents a transaction...
 184  * that is, a VFS operation.  The VFS operation is written out using
 185  * sub-records and may contain multiple, possibly nested sub-transactions.
 186  * multiple sub-transactions occur when a VFS operation cannot be represented
 187  * by a single command.  This is typically the case when a journal is
 188  * configured to be reversable because UNDO sequences almost always have to
 189  * be specified in such cases.  For example, if you ftruncate() a file the
 190  * journal might have to write out a sequence of WRITE records representing
 191  * the lost data, otherwise the journal would not be reversable.
 192  * Sub-transactions within a particular stream do not have their own sequence
 193  * number field and thus may not be parallelized (the protocol is already
 194  * complex enough!).
 195  *
 196  * In order to support streaming operation with a limited buffer the recsize
 197  * field is allowed to be 0 for subrecords with the JMASK_NESTED bit set.
 198  * If this case occurs a scanner can determine that the recursion has ended
 199  * by detecting a nested subrecord with the JMASK_LAST bit set.  A scanner
 200  * may also set the field to the proper value after the fact to make later
 201  * operations more efficient.
 202  *
 203  * Note that this bit must be properly set even if the recsize field is
 204  * non-zero.  The recsize must always be properly specified for 'leaf'
 205  * subrecords, however in order to allow subsystems to potentially allocate
 206  * more data space then they use the protocol allows any 'dead' space to be
 207  * filled with JLEAF_PAD records.
 208  *
 209  * The recsize field may indicate data well past the size of the current
 210  * raw stream record.  That is, the scanner may have to glue together
 211  * multiple stream records with the same stream id to fully decode the
 212  * embedded subrecords.  In particular, a subrecord could very well represent
 213  * hundreds of megabytes of data (e.g. if a program were to do a
 214  * multi-megabyte write()) and be split up across thousands of raw streaming
 215  * records, possibly interlaced with other unrelated streams from other
 216  * unrelated processes.
 217  *
 218  * If a large sub-transaction is aborted the logical stream may be
 219  * terminated without writing out all the expected data.  When this occurs
 220  * the stream's ending record must also have the JREC_STREAMCTL_ABORTED bit
 221  * set.  However, scanners should still be robust enough to detect such
 222  * overflows even if the aborted bit is not set and consider them data
 223  * corruption.
 224  *
 225  * Aborts may also occur in the normal course of operations, especially once
 226  * the journaling API is integrated into the cache coherency API.  A normal
 227  * abort is issued by emplacing a JLEAF_ABORT record within the transaction
 228  * being aborted.  Such records must be the last record in the sub-transaction,
 229  * so JLEAF_LAST is also usually set.  In a transaction with many
 230  * sub-transactions only those sub-transactions with an abort record are
 231  * aborted, the rest remain valid.  Abort records are considered S.O.P. for
 232  * two reasons:  First, limited memory buffer space may make it impossible
 233  * to delete the portion of the stream being aborted (the data may have
 234  * already been sent to the target).  Second, the journaling code will
 235  * eventually be used to support a cache coherency layer which may have to
 236  * abort operations as part of the cache coherency protocol.  Note that
 237  * subrecord aborts are different from stream record aborts.  Stream record
 238  * aborts are considered to be extrodinary situations while subrecord aborts
 239  * are S.O.P.
 240  */
 241
 242 struct journal_subrecord {
 243         u_int16_t rectype;      /* 2 control bits, 14 record type bits */
 244         int16_t reserved;       /* future use */
 245         int32_t recsize;        /* record size (mandatory if not NESTED) */
 246         /* ADDITIONAL DATA */
 247 };
 248
 249 #define JMASK_NESTED            0x8000  /* data is a nested recursion */
 250 #define JMASK_LAST              0x4000
 251 #define JMASK_SUBRECORD         0x0400
 252 #define JTYPE_MASK              (~JMASK_LAST)
 253
 254 #define JLEAF_PAD               0x0000
 255 #define JLEAF_ABORT             0x0001
 256 #define JTYPE_ASSOCIATE         0x0002
 257 #define JTYPE_DISASSOCIATE      0x0003
 258 #define JTYPE_UNDO              (JMASK_NESTED|0x0004)
 259 #define JTYPE_AUDIT             (JMASK_NESTED|0x0005)
 260 #define JTYPE_REDO              (JMASK_NESTED|0x0006)
 261
 262 #define JTYPE_SETATTR           (JMASK_NESTED|0x0010)
 263 #define JTYPE_WRITE             (JMASK_NESTED|0x0011)
 264 #define JTYPE_PUTPAGES          (JMASK_NESTED|0x0012)
 265 #define JTYPE_SETACL            (JMASK_NESTED|0x0013)
 266 #define JTYPE_SETEXTATTR        (JMASK_NESTED|0x0014)
 267 #define JTYPE_CREATE            (JMASK_NESTED|0x0015)
 268 #define JTYPE_MKNOD             (JMASK_NESTED|0x0016)
 269 #define JTYPE_LINK              (JMASK_NESTED|0x0017)
 270 #define JTYPE_SYMLINK           (JMASK_NESTED|0x0018)
 271 #define JTYPE_WHITEOUT          (JMASK_NESTED|0x0019)
 272 #define JTYPE_REMOVE            (JMASK_NESTED|0x001A)
 273 #define JTYPE_MKDIR             (JMASK_NESTED|0x001B)
 274 #define JTYPE_RMDIR             (JMASK_NESTED|0x001C)
 275 #define JTYPE_RENAME            (JMASK_NESTED|0x001D)
 276
 277 #define JTYPE_VATTR             (JMASK_NESTED|0x0100)
 278 #define JTYPE_CRED              (JMASK_NESTED|0x0101)
 279
 280 /*
 281  * Low level record types
 282  */
 283 #define JLEAF_FILEDATA          0x0401
 284 #define JLEAF_PATH1             0x0402
 285 #define JLEAF_PATH2             0x0403
 286 #define JLEAF_PATH3             0x0404
 287 #define JLEAF_PATH4             0x0405
 288 #define JLEAF_UID               0x0406
 289 #define JLEAF_GID               0x0407
 290 #define JLEAF_MODES             0x0408
 291 #define JLEAF_FFLAGS            0x0409
 292 #define JLEAF_PID               0x040A
 293 #define JLEAF_PPID              0x040B
 294 #define JLEAF_COMM              0x040C
 295 #define JLEAF_ATTRNAME          0x040D
 296 #define JLEAF_PATH_REF          0x040E
 297 #define JLEAF_RESERVED_0F       0x040F
 298 #define JLEAF_SYMLINKDATA       0x0410
 299 #define JLEAF_SEEKPOS           0x0411
 300 #define JLEAF_INUM              0x0412
 301 #define JLEAF_NLINK             0x0413
 302 #define JLEAF_FSID              0x0414
 303 #define JLEAF_SIZE              0x0415
 304 #define JLEAF_ATIME             0x0416
 305 #define JLEAF_MTIME             0x0417
 306 #define JLEAF_CTIME             0x0418
 307 #define JLEAF_GEN               0x0419
 308 #define JLEAF_FLAGS             0x041A
 309 #define JLEAF_UDEV              0x041B
 310 #define JLEAF_FILEREV           0x041C
 311 #define JLEAF_VTYPE             0x041D
 312 #define JLEAF_ERROR             0x041E
 313 #define JLEAF_UMAJOR            0x041F
 314 #define JLEAF_UMINOR            0x0420
 315
 316 /*
 317  * Low level journal data file structures
 318  *
 319  * NOTE: embedded strings may use the full width of the field and thus
 320  * may not be 0-terminated.
 321  */
 322 struct jleaf_path {
 323         char    path[4];        /* path from base of mount point */
 324         /* path is variable length and 0-terminated */
 325 };
 326
 327 struct jleaf_vattr {
 328         int32_t modes;
 329         int32_t fflags;
 330         struct timespec atime;
 331         struct timespec mtime;
 332         struct timespec ctime;
 333         int64_t inum;
 334 };
 335
 336 struct jleaf_cred {
 337         int32_t uid;
 338         int32_t gid;
 339         int32_t pid;
 340         int32_t flags;          /* suid/sgid and other flags */
 341         char    line[8];        /* ttyname or other session identification */
 342         char    comm[8];        /* simplified command name for reference */
 343 };
 344
 345 struct jleaf_ioinfo {
 346         int64_t offset;
 347 };
 348
 349 #endif