1 #ifndef CEPH_RADOS_H
2 #define CEPH_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 5
15 #define CEPH_OSDMAP_INC_VERSION_EXT 6
16 #define CEPH_OSDMAP_VERSION 5
17 #define CEPH_OSDMAP_VERSION_EXT 6
18
19 /*
20 * fs id
21 */
22 struct ceph_fsid {
23 unsigned char fsid[16];
24 };
25
ceph_fsid_compare(const struct ceph_fsid * a,const struct ceph_fsid * b)26 static inline int ceph_fsid_compare(const struct ceph_fsid *a,
27 const struct ceph_fsid *b)
28 {
29 return memcmp(a, b, sizeof(*a));
30 }
31
32 /*
33 * ino, object, etc.
34 */
35 typedef __le64 ceph_snapid_t;
36 #define CEPH_SNAPDIR ((__u64)(-1)) /* reserved for hidden .snap dir */
37 #define CEPH_NOSNAP ((__u64)(-2)) /* "head", "live" revision */
38 #define CEPH_MAXSNAP ((__u64)(-3)) /* largest valid snapid */
39
40 struct ceph_timespec {
41 __le32 tv_sec;
42 __le32 tv_nsec;
43 } __attribute__ ((packed));
44
45
46 /*
47 * object layout - how objects are mapped into PGs
48 */
49 #define CEPH_OBJECT_LAYOUT_HASH 1
50 #define CEPH_OBJECT_LAYOUT_LINEAR 2
51 #define CEPH_OBJECT_LAYOUT_HASHINO 3
52
53 /*
54 * pg layout -- how PGs are mapped onto (sets of) OSDs
55 */
56 #define CEPH_PG_LAYOUT_CRUSH 0
57 #define CEPH_PG_LAYOUT_HASH 1
58 #define CEPH_PG_LAYOUT_LINEAR 2
59 #define CEPH_PG_LAYOUT_HYBRID 3
60
61 #define CEPH_PG_MAX_SIZE 16 /* max # osds in a single pg */
62
63 /*
64 * placement group.
65 * we encode this into one __le64.
66 */
67 struct ceph_pg {
68 __le16 preferred; /* preferred primary osd */
69 __le16 ps; /* placement seed */
70 __le32 pool; /* object pool */
71 } __attribute__ ((packed));
72
73 /*
74 * pg_pool is a set of pgs storing a pool of objects
75 *
76 * pg_num -- base number of pseudorandomly placed pgs
77 *
78 * pgp_num -- effective number when calculating pg placement. this
79 * is used for pg_num increases. new pgs result in data being "split"
80 * into new pgs. for this to proceed smoothly, new pgs are intiially
81 * colocated with their parents; that is, pgp_num doesn't increase
82 * until the new pgs have successfully split. only _then_ are the new
83 * pgs placed independently.
84 *
85 * lpg_num -- localized pg count (per device). replicas are randomly
86 * selected.
87 *
88 * lpgp_num -- as above.
89 */
90 #define CEPH_PG_TYPE_REP 1
91 #define CEPH_PG_TYPE_RAID4 2
92 #define CEPH_PG_POOL_VERSION 2
93 struct ceph_pg_pool {
94 __u8 type; /* CEPH_PG_TYPE_* */
95 __u8 size; /* number of osds in each pg */
96 __u8 crush_ruleset; /* crush placement rule */
97 __u8 object_hash; /* hash mapping object name to ps */
98 __le32 pg_num, pgp_num; /* number of pg's */
99 __le32 lpg_num, lpgp_num; /* number of localized pg's */
100 __le32 last_change; /* most recent epoch changed */
101 __le64 snap_seq; /* seq for per-pool snapshot */
102 __le32 snap_epoch; /* epoch of last snap */
103 __le32 num_snaps;
104 __le32 num_removed_snap_intervals; /* if non-empty, NO per-pool snaps */
105 __le64 auid; /* who owns the pg */
106 } __attribute__ ((packed));
107
108 /*
109 * stable_mod func is used to control number of placement groups.
110 * similar to straight-up modulo, but produces a stable mapping as b
111 * increases over time. b is the number of bins, and bmask is the
112 * containing power of 2 minus 1.
113 *
114 * b <= bmask and bmask=(2**n)-1
115 * e.g., b=12 -> bmask=15, b=123 -> bmask=127
116 */
ceph_stable_mod(int x,int b,int bmask)117 static inline int ceph_stable_mod(int x, int b, int bmask)
118 {
119 if ((x & bmask) < b)
120 return x & bmask;
121 else
122 return x & (bmask >> 1);
123 }
124
125 /*
126 * object layout - how a given object should be stored.
127 */
128 struct ceph_object_layout {
129 struct ceph_pg ol_pgid; /* raw pg, with _full_ ps precision. */
130 __le32 ol_stripe_unit; /* for per-object parity, if any */
131 } __attribute__ ((packed));
132
133 /*
134 * compound epoch+version, used by storage layer to serialize mutations
135 */
136 struct ceph_eversion {
137 __le32 epoch;
138 __le64 version;
139 } __attribute__ ((packed));
140
141 /*
142 * osd map bits
143 */
144
145 /* status bits */
146 #define CEPH_OSD_EXISTS 1
147 #define CEPH_OSD_UP 2
148
149 /* osd weights. fixed point value: 0x10000 == 1.0 ("in"), 0 == "out" */
150 #define CEPH_OSD_IN 0x10000
151 #define CEPH_OSD_OUT 0
152
153
154 /*
155 * osd map flag bits
156 */
157 #define CEPH_OSDMAP_NEARFULL (1<<0) /* sync writes (near ENOSPC) */
158 #define CEPH_OSDMAP_FULL (1<<1) /* no data writes (ENOSPC) */
159 #define CEPH_OSDMAP_PAUSERD (1<<2) /* pause all reads */
160 #define CEPH_OSDMAP_PAUSEWR (1<<3) /* pause all writes */
161 #define CEPH_OSDMAP_PAUSEREC (1<<4) /* pause recovery */
162
163 /*
164 * osd ops
165 */
166 #define CEPH_OSD_OP_MODE 0xf000
167 #define CEPH_OSD_OP_MODE_RD 0x1000
168 #define CEPH_OSD_OP_MODE_WR 0x2000
169 #define CEPH_OSD_OP_MODE_RMW 0x3000
170 #define CEPH_OSD_OP_MODE_SUB 0x4000
171
172 #define CEPH_OSD_OP_TYPE 0x0f00
173 #define CEPH_OSD_OP_TYPE_LOCK 0x0100
174 #define CEPH_OSD_OP_TYPE_DATA 0x0200
175 #define CEPH_OSD_OP_TYPE_ATTR 0x0300
176 #define CEPH_OSD_OP_TYPE_EXEC 0x0400
177 #define CEPH_OSD_OP_TYPE_PG 0x0500
178
179 enum {
180 /** data **/
181 /* read */
182 CEPH_OSD_OP_READ = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_DATA | 1,
183 CEPH_OSD_OP_STAT = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_DATA | 2,
184 CEPH_OSD_OP_MAPEXT = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_DATA | 3,
185
186 /* fancy read */
187 CEPH_OSD_OP_MASKTRUNC = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_DATA | 4,
188 CEPH_OSD_OP_SPARSE_READ = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_DATA | 5,
189
190 CEPH_OSD_OP_NOTIFY = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_DATA | 6,
191 CEPH_OSD_OP_NOTIFY_ACK = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_DATA | 7,
192
193 /* versioning */
194 CEPH_OSD_OP_ASSERT_VER = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_DATA | 8,
195
196 /* write */
197 CEPH_OSD_OP_WRITE = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 1,
198 CEPH_OSD_OP_WRITEFULL = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 2,
199 CEPH_OSD_OP_TRUNCATE = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 3,
200 CEPH_OSD_OP_ZERO = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 4,
201 CEPH_OSD_OP_DELETE = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 5,
202
203 /* fancy write */
204 CEPH_OSD_OP_APPEND = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 6,
205 CEPH_OSD_OP_STARTSYNC = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 7,
206 CEPH_OSD_OP_SETTRUNC = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 8,
207 CEPH_OSD_OP_TRIMTRUNC = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 9,
208
209 CEPH_OSD_OP_TMAPUP = CEPH_OSD_OP_MODE_RMW | CEPH_OSD_OP_TYPE_DATA | 10,
210 CEPH_OSD_OP_TMAPPUT = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 11,
211 CEPH_OSD_OP_TMAPGET = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_DATA | 12,
212
213 CEPH_OSD_OP_CREATE = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 13,
214 CEPH_OSD_OP_ROLLBACK= CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 14,
215
216 CEPH_OSD_OP_WATCH = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 15,
217
218 /** attrs **/
219 /* read */
220 CEPH_OSD_OP_GETXATTR = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_ATTR | 1,
221 CEPH_OSD_OP_GETXATTRS = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_ATTR | 2,
222 CEPH_OSD_OP_CMPXATTR = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_ATTR | 3,
223
224 /* write */
225 CEPH_OSD_OP_SETXATTR = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_ATTR | 1,
226 CEPH_OSD_OP_SETXATTRS = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_ATTR | 2,
227 CEPH_OSD_OP_RESETXATTRS = CEPH_OSD_OP_MODE_WR|CEPH_OSD_OP_TYPE_ATTR | 3,
228 CEPH_OSD_OP_RMXATTR = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_ATTR | 4,
229
230 /** subop **/
231 CEPH_OSD_OP_PULL = CEPH_OSD_OP_MODE_SUB | 1,
232 CEPH_OSD_OP_PUSH = CEPH_OSD_OP_MODE_SUB | 2,
233 CEPH_OSD_OP_BALANCEREADS = CEPH_OSD_OP_MODE_SUB | 3,
234 CEPH_OSD_OP_UNBALANCEREADS = CEPH_OSD_OP_MODE_SUB | 4,
235 CEPH_OSD_OP_SCRUB = CEPH_OSD_OP_MODE_SUB | 5,
236 CEPH_OSD_OP_SCRUB_RESERVE = CEPH_OSD_OP_MODE_SUB | 6,
237 CEPH_OSD_OP_SCRUB_UNRESERVE = CEPH_OSD_OP_MODE_SUB | 7,
238 CEPH_OSD_OP_SCRUB_STOP = CEPH_OSD_OP_MODE_SUB | 8,
239
240 /** lock **/
241 CEPH_OSD_OP_WRLOCK = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 1,
242 CEPH_OSD_OP_WRUNLOCK = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 2,
243 CEPH_OSD_OP_RDLOCK = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 3,
244 CEPH_OSD_OP_RDUNLOCK = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 4,
245 CEPH_OSD_OP_UPLOCK = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 5,
246 CEPH_OSD_OP_DNLOCK = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 6,
247
248 /** exec **/
249 CEPH_OSD_OP_CALL = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_EXEC | 1,
250
251 /** pg **/
252 CEPH_OSD_OP_PGLS = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_PG | 1,
253 };
254
ceph_osd_op_type_lock(int op)255 static inline int ceph_osd_op_type_lock(int op)
256 {
257 return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_LOCK;
258 }
ceph_osd_op_type_data(int op)259 static inline int ceph_osd_op_type_data(int op)
260 {
261 return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_DATA;
262 }
ceph_osd_op_type_attr(int op)263 static inline int ceph_osd_op_type_attr(int op)
264 {
265 return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_ATTR;
266 }
ceph_osd_op_type_exec(int op)267 static inline int ceph_osd_op_type_exec(int op)
268 {
269 return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_EXEC;
270 }
ceph_osd_op_type_pg(int op)271 static inline int ceph_osd_op_type_pg(int op)
272 {
273 return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_PG;
274 }
275
ceph_osd_op_mode_subop(int op)276 static inline int ceph_osd_op_mode_subop(int op)
277 {
278 return (op & CEPH_OSD_OP_MODE) == CEPH_OSD_OP_MODE_SUB;
279 }
ceph_osd_op_mode_read(int op)280 static inline int ceph_osd_op_mode_read(int op)
281 {
282 return (op & CEPH_OSD_OP_MODE) == CEPH_OSD_OP_MODE_RD;
283 }
ceph_osd_op_mode_modify(int op)284 static inline int ceph_osd_op_mode_modify(int op)
285 {
286 return (op & CEPH_OSD_OP_MODE) == CEPH_OSD_OP_MODE_WR;
287 }
288
289 /*
290 * note that the following tmap stuff is also defined in the ceph librados.h
291 * any modification here needs to be updated there
292 */
293 #define CEPH_OSD_TMAP_HDR 'h'
294 #define CEPH_OSD_TMAP_SET 's'
295 #define CEPH_OSD_TMAP_RM 'r'
296
297 extern const char *ceph_osd_op_name(int op);
298
299
300 /*
301 * osd op flags
302 *
303 * An op may be READ, WRITE, or READ|WRITE.
304 */
305 enum {
306 CEPH_OSD_FLAG_ACK = 1, /* want (or is) "ack" ack */
307 CEPH_OSD_FLAG_ONNVRAM = 2, /* want (or is) "onnvram" ack */
308 CEPH_OSD_FLAG_ONDISK = 4, /* want (or is) "ondisk" ack */
309 CEPH_OSD_FLAG_RETRY = 8, /* resend attempt */
310 CEPH_OSD_FLAG_READ = 16, /* op may read */
311 CEPH_OSD_FLAG_WRITE = 32, /* op may write */
312 CEPH_OSD_FLAG_ORDERSNAP = 64, /* EOLDSNAP if snapc is out of order */
313 CEPH_OSD_FLAG_PEERSTAT = 128, /* msg includes osd_peer_stat */
314 CEPH_OSD_FLAG_BALANCE_READS = 256,
315 CEPH_OSD_FLAG_PARALLELEXEC = 512, /* execute op in parallel */
316 CEPH_OSD_FLAG_PGOP = 1024, /* pg op, no object */
317 CEPH_OSD_FLAG_EXEC = 2048, /* op may exec */
318 CEPH_OSD_FLAG_EXEC_PUBLIC = 4096, /* op may exec (public) */
319 };
320
321 enum {
322 CEPH_OSD_OP_FLAG_EXCL = 1, /* EXCL object create */
323 };
324
325 #define EOLDSNAPC ERESTART /* ORDERSNAP flag set; writer has old snapc*/
326 #define EBLACKLISTED ESHUTDOWN /* blacklisted */
327
328 /* xattr comparison */
329 enum {
330 CEPH_OSD_CMPXATTR_OP_NOP = 0,
331 CEPH_OSD_CMPXATTR_OP_EQ = 1,
332 CEPH_OSD_CMPXATTR_OP_NE = 2,
333 CEPH_OSD_CMPXATTR_OP_GT = 3,
334 CEPH_OSD_CMPXATTR_OP_GTE = 4,
335 CEPH_OSD_CMPXATTR_OP_LT = 5,
336 CEPH_OSD_CMPXATTR_OP_LTE = 6
337 };
338
339 enum {
340 CEPH_OSD_CMPXATTR_MODE_STRING = 1,
341 CEPH_OSD_CMPXATTR_MODE_U64 = 2
342 };
343
344 #define RADOS_NOTIFY_VER 1
345
346 /*
347 * an individual object operation. each may be accompanied by some data
348 * payload
349 */
350 struct ceph_osd_op {
351 __le16 op; /* CEPH_OSD_OP_* */
352 __le32 flags; /* CEPH_OSD_FLAG_* */
353 union {
354 struct {
355 __le64 offset, length;
356 __le64 truncate_size;
357 __le32 truncate_seq;
358 } __attribute__ ((packed)) extent;
359 struct {
360 __le32 name_len;
361 __le32 value_len;
362 __u8 cmp_op; /* CEPH_OSD_CMPXATTR_OP_* */
363 __u8 cmp_mode; /* CEPH_OSD_CMPXATTR_MODE_* */
364 } __attribute__ ((packed)) xattr;
365 struct {
366 __u8 class_len;
367 __u8 method_len;
368 __u8 argc;
369 __le32 indata_len;
370 } __attribute__ ((packed)) cls;
371 struct {
372 __le64 cookie, count;
373 } __attribute__ ((packed)) pgls;
374 struct {
375 __le64 snapid;
376 } __attribute__ ((packed)) snap;
377 struct {
378 __le64 cookie;
379 __le64 ver;
380 __u8 flag; /* 0 = unwatch, 1 = watch */
381 } __attribute__ ((packed)) watch;
382 };
383 __le32 payload_len;
384 } __attribute__ ((packed));
385
386 /*
387 * osd request message header. each request may include multiple
388 * ceph_osd_op object operations.
389 */
390 struct ceph_osd_request_head {
391 __le32 client_inc; /* client incarnation */
392 struct ceph_object_layout layout; /* pgid */
393 __le32 osdmap_epoch; /* client's osdmap epoch */
394
395 __le32 flags;
396
397 struct ceph_timespec mtime; /* for mutations only */
398 struct ceph_eversion reassert_version; /* if we are replaying op */
399
400 __le32 object_len; /* length of object name */
401
402 __le64 snapid; /* snapid to read */
403 __le64 snap_seq; /* writer's snap context */
404 __le32 num_snaps;
405
406 __le16 num_ops;
407 struct ceph_osd_op ops[]; /* followed by ops[], obj, ticket, snaps */
408 } __attribute__ ((packed));
409
410 struct ceph_osd_reply_head {
411 __le32 client_inc; /* client incarnation */
412 __le32 flags;
413 struct ceph_object_layout layout;
414 __le32 osdmap_epoch;
415 struct ceph_eversion reassert_version; /* for replaying uncommitted */
416
417 __le32 result; /* result code */
418
419 __le32 object_len; /* length of object name */
420 __le32 num_ops;
421 struct ceph_osd_op ops[0]; /* ops[], object */
422 } __attribute__ ((packed));
423
424
425
426 #endif
427