fs/ceph/rados.h

   1 #ifndef __RADOS_H
   2 #define __RADOS_H
   3
   4 /*
   5  * Data types for the Ceph distributed object storage layer RADOS
   6  * (Reliable Autonomic Distributed Object Store).
   7  */
   8
   9 #include "msgr.h"
  10
  11 /*
  12  * osdmap encoding versions
  13  */
  14 #define CEPH_OSDMAP_INC_VERSION 4
  15 #define CEPH_OSDMAP_VERSION     4
  16
  17 /*
  18  * fs id
  19  */
  20 struct ceph_fsid {
  21         unsigned char fsid[16];
  22 };
  23
  24 static inline int ceph_fsid_compare(const struct ceph_fsid *a,
  25                                     const struct ceph_fsid *b)
  26 {
  27         return memcmp(a, b, sizeof(*a));
  28 }
  29
  30 /*
  31  * ino, object, etc.
  32  */
  33 typedef __le64 ceph_snapid_t;
  34 #define CEPH_SNAPDIR ((__u64)(-1))  /* reserved for hidden .snap dir */
  35 #define CEPH_NOSNAP  ((__u64)(-2))  /* "head", "live" revision */
  36 #define CEPH_MAXSNAP ((__u64)(-3))  /* largest valid snapid */
  37
  38 struct ceph_timespec {
  39         __le32 tv_sec;
  40         __le32 tv_nsec;
  41 } __attribute__ ((packed));
  42
  43
  44 /*
  45  * object layout - how objects are mapped into PGs
  46  */
  47 #define CEPH_OBJECT_LAYOUT_HASH     1
  48 #define CEPH_OBJECT_LAYOUT_LINEAR   2
  49 #define CEPH_OBJECT_LAYOUT_HASHINO  3
  50
  51 /*
  52  * pg layout -- how PGs are mapped onto (sets of) OSDs
  53  */
  54 #define CEPH_PG_LAYOUT_CRUSH  0
  55 #define CEPH_PG_LAYOUT_HASH   1
  56 #define CEPH_PG_LAYOUT_LINEAR 2
  57 #define CEPH_PG_LAYOUT_HYBRID 3
  58
  59
  60 /*
  61  * placement group.
  62  * we encode this into one __le64.
  63  */
  64 struct ceph_pg {
  65         __le16 preferred; /* preferred primary osd */
  66         __le16 ps;        /* placement seed */
  67         __le32 pool;      /* object pool */
  68 } __attribute__ ((packed));
  69
  70 /*
  71  * pg_pool is a set of pgs storing a pool of objects
  72  *
  73  *  pg_num -- base number of pseudorandomly placed pgs
  74  *
  75  *  pgp_num -- effective number when calculating pg placement.  this
  76  * is used for pg_num increases.  new pgs result in data being "split"
  77  * into new pgs.  for this to proceed smoothly, new pgs are intiially
  78  * colocated with their parents; that is, pgp_num doesn't increase
  79  * until the new pgs have successfully split.  only _then_ are the new
  80  * pgs placed independently.
  81  *
  82  *  lpg_num -- localized pg count (per device).  replicas are randomly
  83  * selected.
  84  *
  85  *  lpgp_num -- as above.
  86  */
  87 #define CEPH_PG_TYPE_REP     1
  88 #define CEPH_PG_TYPE_RAID4   2
  89 #define CEPH_PG_POOL_VERSION 2
  90 struct ceph_pg_pool {
  91         __u8 type;                /* CEPH_PG_TYPE_* */
  92         __u8 size;                /* number of osds in each pg */
  93         __u8 crush_ruleset;       /* crush placement rule */
  94         __u8 object_hash;         /* hash mapping object name to ps */
  95         __le32 pg_num, pgp_num;   /* number of pg's */
  96         __le32 lpg_num, lpgp_num; /* number of localized pg's */
  97         __le32 last_change;       /* most recent epoch changed */
  98         __le64 snap_seq;          /* seq for per-pool snapshot */
  99         __le32 snap_epoch;        /* epoch of last snap */
 100         __le32 num_snaps;
 101         __le32 num_removed_snap_intervals;
 102         __le64 uid;
 103 } __attribute__ ((packed));
 104
 105 /*
 106  * stable_mod func is used to control number of placement groups.
 107  * similar to straight-up modulo, but produces a stable mapping as b
 108  * increases over time.  b is the number of bins, and bmask is the
 109  * containing power of 2 minus 1.
 110  *
 111  * b <= bmask and bmask=(2**n)-1
 112  * e.g., b=12 -> bmask=15, b=123 -> bmask=127
 113  */
 114 static inline int ceph_stable_mod(int x, int b, int bmask)
 115 {
 116         if ((x & bmask) < b)
 117                 return x & bmask;
 118         else
 119                 return x & (bmask >> 1);
 120 }
 121
 122 /*
 123  * object layout - how a given object should be stored.
 124  */
 125 struct ceph_object_layout {
 126         struct ceph_pg ol_pgid;   /* raw pg, with _full_ ps precision. */
 127         __le32 ol_stripe_unit;    /* for per-object parity, if any */
 128 } __attribute__ ((packed));
 129
 130 /*
 131  * compound epoch+version, used by storage layer to serialize mutations
 132  */
 133 struct ceph_eversion {
 134         __le32 epoch;
 135         __le64 version;
 136 } __attribute__ ((packed));
 137
 138 /*
 139  * osd map bits
 140  */
 141
 142 /* status bits */
 143 #define CEPH_OSD_EXISTS 1
 144 #define CEPH_OSD_UP     2
 145
 146 /* osd weights.  fixed point value: 0x10000 == 1.0 ("in"), 0 == "out" */
 147 #define CEPH_OSD_IN  0x10000
 148 #define CEPH_OSD_OUT 0
 149
 150
 151 /*
 152  * osd map flag bits
 153  */
 154 #define CEPH_OSDMAP_NEARFULL (1<<0)  /* sync writes (near ENOSPC) */
 155 #define CEPH_OSDMAP_FULL     (1<<1)  /* no data writes (ENOSPC) */
 156 #define CEPH_OSDMAP_PAUSERD  (1<<2)  /* pause all reads */
 157 #define CEPH_OSDMAP_PAUSEWR  (1<<3)  /* pause all writes */
 158 #define CEPH_OSDMAP_PAUSEREC (1<<4)  /* pause recovery */
 159
 160 /*
 161  * osd ops
 162  */
 163 #define CEPH_OSD_OP_MODE       0xf000
 164 #define CEPH_OSD_OP_MODE_RD    0x1000
 165 #define CEPH_OSD_OP_MODE_WR    0x2000
 166 #define CEPH_OSD_OP_MODE_RMW   0x3000
 167 #define CEPH_OSD_OP_MODE_SUB   0x4000
 168
 169 #define CEPH_OSD_OP_TYPE       0x0f00
 170 #define CEPH_OSD_OP_TYPE_LOCK  0x0100
 171 #define CEPH_OSD_OP_TYPE_DATA  0x0200
 172 #define CEPH_OSD_OP_TYPE_ATTR  0x0300
 173 #define CEPH_OSD_OP_TYPE_EXEC  0x0400
 174 #define CEPH_OSD_OP_TYPE_PG    0x0500
 175
 176 enum {
 177         /** data **/
 178         /* read */
 179         CEPH_OSD_OP_READ      = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_DATA | 1,
 180         CEPH_OSD_OP_STAT      = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_DATA | 2,
 181
 182         /* fancy read */
 183         CEPH_OSD_OP_MASKTRUNC = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_DATA | 4,
 184
 185         /* write */
 186         CEPH_OSD_OP_WRITE     = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 1,
 187         CEPH_OSD_OP_WRITEFULL = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 2,
 188         CEPH_OSD_OP_TRUNCATE  = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 3,
 189         CEPH_OSD_OP_ZERO      = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 4,
 190         CEPH_OSD_OP_DELETE    = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 5,
 191
 192         /* fancy write */
 193         CEPH_OSD_OP_APPEND    = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 6,
 194         CEPH_OSD_OP_STARTSYNC = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 7,
 195         CEPH_OSD_OP_SETTRUNC  = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 8,
 196         CEPH_OSD_OP_TRIMTRUNC = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 9,
 197
 198         CEPH_OSD_OP_TMAPUP  = CEPH_OSD_OP_MODE_RMW | CEPH_OSD_OP_TYPE_DATA | 10,
 199         CEPH_OSD_OP_TMAPPUT = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 11,
 200         CEPH_OSD_OP_TMAPGET = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_DATA | 12,
 201
 202         CEPH_OSD_OP_CREATE  = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 13,
 203
 204         /** attrs **/
 205         /* read */
 206         CEPH_OSD_OP_GETXATTR  = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_ATTR | 1,
 207         CEPH_OSD_OP_GETXATTRS = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_ATTR | 2,
 208
 209         /* write */
 210         CEPH_OSD_OP_SETXATTR  = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_ATTR | 1,
 211         CEPH_OSD_OP_SETXATTRS = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_ATTR | 2,
 212         CEPH_OSD_OP_RESETXATTRS = CEPH_OSD_OP_MODE_WR|CEPH_OSD_OP_TYPE_ATTR | 3,
 213         CEPH_OSD_OP_RMXATTR   = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_ATTR | 4,
 214
 215         /** subop **/
 216         CEPH_OSD_OP_PULL           = CEPH_OSD_OP_MODE_SUB | 1,
 217         CEPH_OSD_OP_PUSH           = CEPH_OSD_OP_MODE_SUB | 2,
 218         CEPH_OSD_OP_BALANCEREADS   = CEPH_OSD_OP_MODE_SUB | 3,
 219         CEPH_OSD_OP_UNBALANCEREADS = CEPH_OSD_OP_MODE_SUB | 4,
 220         CEPH_OSD_OP_SCRUB          = CEPH_OSD_OP_MODE_SUB | 5,
 221
 222         /** lock **/
 223         CEPH_OSD_OP_WRLOCK    = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 1,
 224         CEPH_OSD_OP_WRUNLOCK  = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 2,
 225         CEPH_OSD_OP_RDLOCK    = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 3,
 226         CEPH_OSD_OP_RDUNLOCK  = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 4,
 227         CEPH_OSD_OP_UPLOCK    = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 5,
 228         CEPH_OSD_OP_DNLOCK    = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 6,
 229
 230         /** exec **/
 231         CEPH_OSD_OP_CALL    = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_EXEC | 1,
 232
 233         /** pg **/
 234         CEPH_OSD_OP_PGLS      = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_PG | 1,
 235 };
 236
 237 static inline int ceph_osd_op_type_lock(int op)
 238 {
 239         return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_LOCK;
 240 }
 241 static inline int ceph_osd_op_type_data(int op)
 242 {
 243         return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_DATA;
 244 }
 245 static inline int ceph_osd_op_type_attr(int op)
 246 {
 247         return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_ATTR;
 248 }
 249 static inline int ceph_osd_op_type_exec(int op)
 250 {
 251         return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_EXEC;
 252 }
 253 static inline int ceph_osd_op_type_pg(int op)
 254 {
 255         return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_PG;
 256 }
 257
 258 static inline int ceph_osd_op_mode_subop(int op)
 259 {
 260         return (op & CEPH_OSD_OP_MODE) == CEPH_OSD_OP_MODE_SUB;
 261 }
 262 static inline int ceph_osd_op_mode_read(int op)
 263 {
 264         return (op & CEPH_OSD_OP_MODE) == CEPH_OSD_OP_MODE_RD;
 265 }
 266 static inline int ceph_osd_op_mode_modify(int op)
 267 {
 268         return (op & CEPH_OSD_OP_MODE) == CEPH_OSD_OP_MODE_WR;
 269 }
 270
 271 #define CEPH_OSD_TMAP_HDR 'h'
 272 #define CEPH_OSD_TMAP_SET 's'
 273 #define CEPH_OSD_TMAP_RM  'r'
 274
 275 extern const char *ceph_osd_op_name(int op);
 276
 277
 278 /*
 279  * osd op flags
 280  *
 281  * An op may be READ, WRITE, or READ|WRITE.
 282  */
 283 enum {
 284         CEPH_OSD_FLAG_ACK = 1,          /* want (or is) "ack" ack */
 285         CEPH_OSD_FLAG_ONNVRAM = 2,      /* want (or is) "onnvram" ack */
 286         CEPH_OSD_FLAG_ONDISK = 4,       /* want (or is) "ondisk" ack */
 287         CEPH_OSD_FLAG_RETRY = 8,        /* resend attempt */
 288         CEPH_OSD_FLAG_READ = 16,        /* op may read */
 289         CEPH_OSD_FLAG_WRITE = 32,       /* op may write */
 290         CEPH_OSD_FLAG_ORDERSNAP = 64,   /* EOLDSNAP if snapc is out of order */
 291         CEPH_OSD_FLAG_PEERSTAT = 128,   /* msg includes osd_peer_stat */
 292         CEPH_OSD_FLAG_BALANCE_READS = 256,
 293         CEPH_OSD_FLAG_PARALLELEXEC = 512, /* execute op in parallel */
 294         CEPH_OSD_FLAG_PGOP = 1024,      /* pg op, no object */
 295         CEPH_OSD_FLAG_EXEC = 2048,      /* op may exec */
 296 };
 297
 298 enum {
 299         CEPH_OSD_OP_FLAG_EXCL = 1,      /* EXCL object create */
 300 };
 301
 302 #define EOLDSNAPC    ERESTART  /* ORDERSNAP flag set; writer has old snapc*/
 303 #define EBLACKLISTED ESHUTDOWN /* blacklisted */
 304
 305 /*
 306  * an individual object operation.  each may be accompanied by some data
 307  * payload
 308  */
 309 struct ceph_osd_op {
 310         __le16 op;           /* CEPH_OSD_OP_* */
 311         __le32 flags;        /* CEPH_OSD_FLAG_* */
 312         union {
 313                 struct {
 314                         __le64 offset, length;
 315                         __le64 truncate_size;
 316                         __le32 truncate_seq;
 317                 } __attribute__ ((packed)) extent;
 318                 struct {
 319                         __le32 name_len;
 320                         __le32 value_len;
 321                 } __attribute__ ((packed)) xattr;
 322                 struct {
 323                         __u8 class_len;
 324                         __u8 method_len;
 325                         __u8 argc;
 326                         __le32 indata_len;
 327                 } __attribute__ ((packed)) cls;
 328                 struct {
 329                         __le64 cookie, count;
 330                 } __attribute__ ((packed)) pgls;
 331         };
 332         __le32 payload_len;
 333 } __attribute__ ((packed));
 334
 335 /*
 336  * osd request message header.  each request may include multiple
 337  * ceph_osd_op object operations.
 338  */
 339 struct ceph_osd_request_head {
 340         __le32 client_inc;                 /* client incarnation */
 341         struct ceph_object_layout layout;  /* pgid */
 342         __le32 osdmap_epoch;               /* client's osdmap epoch */
 343
 344         __le32 flags;
 345
 346         struct ceph_timespec mtime;        /* for mutations only */
 347         struct ceph_eversion reassert_version; /* if we are replaying op */
 348
 349         __le32 object_len;     /* length of object name */
 350
 351         __le64 snapid;         /* snapid to read */
 352         __le64 snap_seq;       /* writer's snap context */
 353         __le32 num_snaps;
 354
 355         __le16 num_ops;
 356         struct ceph_osd_op ops[];  /* followed by ops[], obj, ticket, snaps */
 357 } __attribute__ ((packed));
 358
 359 struct ceph_osd_reply_head {
 360         __le32 client_inc;                /* client incarnation */
 361         __le32 flags;
 362         struct ceph_object_layout layout;
 363         __le32 osdmap_epoch;
 364         struct ceph_eversion reassert_version; /* for replaying uncommitted */
 365
 366         __le32 result;                    /* result code */
 367
 368         __le32 object_len;                /* length of object name */
 369         __le32 num_ops;
 370         struct ceph_osd_op ops[0];  /* ops[], object */
 371 } __attribute__ ((packed));
 372
 373
 374 #endif