blob: 8fcc023056c7642787f33b1aeb35c93ce26492b0 [file] [log] [blame]
Sage Weil0dee3c22009-10-06 11:31:06 -07001#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/*
Sage Weil02f90c62010-02-04 16:18:10 -080012 * osdmap encoding versions
13 */
Sage Weil2844a762010-04-09 15:46:42 -070014#define CEPH_OSDMAP_INC_VERSION 5
15#define CEPH_OSDMAP_INC_VERSION_EXT 5
16#define CEPH_OSDMAP_VERSION 5
17#define CEPH_OSDMAP_VERSION_EXT 5
Sage Weil02f90c62010-02-04 16:18:10 -080018
19/*
Sage Weil0dee3c22009-10-06 11:31:06 -070020 * fs id
21 */
22struct ceph_fsid {
23 unsigned char fsid[16];
24};
25
26static 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 */
35typedef __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
40struct 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
Sage Weild85b7052010-05-10 10:24:48 -070061#define CEPH_PG_MAX_SIZE 16 /* max # osds in a single pg */
Sage Weil0dee3c22009-10-06 11:31:06 -070062
63/*
64 * placement group.
65 * we encode this into one __le64.
66 */
Sage Weil51042122009-11-04 11:39:12 -080067struct ceph_pg {
68 __le16 preferred; /* preferred primary osd */
69 __le16 ps; /* placement seed */
70 __le32 pool; /* object pool */
Sage Weil0dee3c22009-10-06 11:31:06 -070071} __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
Sage Weil02f90c62010-02-04 16:18:10 -080092#define CEPH_PG_POOL_VERSION 2
Sage Weil0dee3c22009-10-06 11:31:06 -070093struct 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 */
Sage Weil1654dd02009-11-06 21:55:25 -080097 __u8 object_hash; /* hash mapping object name to ps */
Sage Weil0dee3c22009-10-06 11:31:06 -070098 __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;
Sage Weilca9d93a2010-05-12 14:48:20 -0700104 __le32 num_removed_snap_intervals; /* if non-empty, NO per-pool snaps */
105 __le64 auid; /* who owns the pg */
Sage Weil0dee3c22009-10-06 11:31:06 -0700106} __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 */
117static 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 */
128struct ceph_object_layout {
Sage Weil51042122009-11-04 11:39:12 -0800129 struct ceph_pg ol_pgid; /* raw pg, with _full_ ps precision. */
Sage Weil0dee3c22009-10-06 11:31:06 -0700130 __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 */
136struct 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
Sage Weil0dee3c22009-10-06 11:31:06 -0700171
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
179enum {
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
185 /* fancy read */
186 CEPH_OSD_OP_MASKTRUNC = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_DATA | 4,
187
188 /* write */
189 CEPH_OSD_OP_WRITE = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 1,
190 CEPH_OSD_OP_WRITEFULL = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 2,
191 CEPH_OSD_OP_TRUNCATE = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 3,
192 CEPH_OSD_OP_ZERO = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 4,
193 CEPH_OSD_OP_DELETE = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 5,
194
195 /* fancy write */
196 CEPH_OSD_OP_APPEND = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 6,
197 CEPH_OSD_OP_STARTSYNC = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 7,
198 CEPH_OSD_OP_SETTRUNC = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 8,
199 CEPH_OSD_OP_TRIMTRUNC = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 9,
200
201 CEPH_OSD_OP_TMAPUP = CEPH_OSD_OP_MODE_RMW | CEPH_OSD_OP_TYPE_DATA | 10,
202 CEPH_OSD_OP_TMAPPUT = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 11,
203 CEPH_OSD_OP_TMAPGET = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_DATA | 12,
204
205 CEPH_OSD_OP_CREATE = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_DATA | 13,
206
207 /** attrs **/
208 /* read */
209 CEPH_OSD_OP_GETXATTR = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_ATTR | 1,
210 CEPH_OSD_OP_GETXATTRS = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_ATTR | 2,
Sage Weilca9d93a2010-05-12 14:48:20 -0700211 CEPH_OSD_OP_CMPXATTR = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_ATTR | 3,
Sage Weil0dee3c22009-10-06 11:31:06 -0700212
213 /* write */
214 CEPH_OSD_OP_SETXATTR = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_ATTR | 1,
215 CEPH_OSD_OP_SETXATTRS = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_ATTR | 2,
216 CEPH_OSD_OP_RESETXATTRS = CEPH_OSD_OP_MODE_WR|CEPH_OSD_OP_TYPE_ATTR | 3,
217 CEPH_OSD_OP_RMXATTR = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_ATTR | 4,
218
219 /** subop **/
220 CEPH_OSD_OP_PULL = CEPH_OSD_OP_MODE_SUB | 1,
221 CEPH_OSD_OP_PUSH = CEPH_OSD_OP_MODE_SUB | 2,
222 CEPH_OSD_OP_BALANCEREADS = CEPH_OSD_OP_MODE_SUB | 3,
223 CEPH_OSD_OP_UNBALANCEREADS = CEPH_OSD_OP_MODE_SUB | 4,
224 CEPH_OSD_OP_SCRUB = CEPH_OSD_OP_MODE_SUB | 5,
225
226 /** lock **/
227 CEPH_OSD_OP_WRLOCK = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 1,
228 CEPH_OSD_OP_WRUNLOCK = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 2,
229 CEPH_OSD_OP_RDLOCK = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 3,
230 CEPH_OSD_OP_RDUNLOCK = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 4,
231 CEPH_OSD_OP_UPLOCK = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 5,
232 CEPH_OSD_OP_DNLOCK = CEPH_OSD_OP_MODE_WR | CEPH_OSD_OP_TYPE_LOCK | 6,
233
234 /** exec **/
235 CEPH_OSD_OP_CALL = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_EXEC | 1,
236
237 /** pg **/
238 CEPH_OSD_OP_PGLS = CEPH_OSD_OP_MODE_RD | CEPH_OSD_OP_TYPE_PG | 1,
239};
240
241static inline int ceph_osd_op_type_lock(int op)
242{
243 return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_LOCK;
244}
245static inline int ceph_osd_op_type_data(int op)
246{
247 return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_DATA;
248}
249static inline int ceph_osd_op_type_attr(int op)
250{
251 return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_ATTR;
252}
253static inline int ceph_osd_op_type_exec(int op)
254{
255 return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_EXEC;
256}
257static inline int ceph_osd_op_type_pg(int op)
258{
259 return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_PG;
260}
261
262static inline int ceph_osd_op_mode_subop(int op)
263{
264 return (op & CEPH_OSD_OP_MODE) == CEPH_OSD_OP_MODE_SUB;
265}
266static inline int ceph_osd_op_mode_read(int op)
267{
268 return (op & CEPH_OSD_OP_MODE) == CEPH_OSD_OP_MODE_RD;
269}
270static inline int ceph_osd_op_mode_modify(int op)
271{
272 return (op & CEPH_OSD_OP_MODE) == CEPH_OSD_OP_MODE_WR;
273}
274
275#define CEPH_OSD_TMAP_HDR 'h'
276#define CEPH_OSD_TMAP_SET 's'
277#define CEPH_OSD_TMAP_RM 'r'
278
279extern const char *ceph_osd_op_name(int op);
280
281
282/*
283 * osd op flags
284 *
285 * An op may be READ, WRITE, or READ|WRITE.
286 */
287enum {
288 CEPH_OSD_FLAG_ACK = 1, /* want (or is) "ack" ack */
289 CEPH_OSD_FLAG_ONNVRAM = 2, /* want (or is) "onnvram" ack */
290 CEPH_OSD_FLAG_ONDISK = 4, /* want (or is) "ondisk" ack */
291 CEPH_OSD_FLAG_RETRY = 8, /* resend attempt */
292 CEPH_OSD_FLAG_READ = 16, /* op may read */
293 CEPH_OSD_FLAG_WRITE = 32, /* op may write */
294 CEPH_OSD_FLAG_ORDERSNAP = 64, /* EOLDSNAP if snapc is out of order */
295 CEPH_OSD_FLAG_PEERSTAT = 128, /* msg includes osd_peer_stat */
296 CEPH_OSD_FLAG_BALANCE_READS = 256,
297 CEPH_OSD_FLAG_PARALLELEXEC = 512, /* execute op in parallel */
298 CEPH_OSD_FLAG_PGOP = 1024, /* pg op, no object */
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800299 CEPH_OSD_FLAG_EXEC = 2048, /* op may exec */
Sage Weil0dee3c22009-10-06 11:31:06 -0700300};
301
302enum {
303 CEPH_OSD_OP_FLAG_EXCL = 1, /* EXCL object create */
304};
305
306#define EOLDSNAPC ERESTART /* ORDERSNAP flag set; writer has old snapc*/
307#define EBLACKLISTED ESHUTDOWN /* blacklisted */
308
Sage Weilca9d93a2010-05-12 14:48:20 -0700309/* xattr comparison */
310enum {
311 CEPH_OSD_CMPXATTR_OP_NOP = 0,
312 CEPH_OSD_CMPXATTR_OP_EQ = 1,
313 CEPH_OSD_CMPXATTR_OP_NE = 2,
314 CEPH_OSD_CMPXATTR_OP_GT = 3,
315 CEPH_OSD_CMPXATTR_OP_GTE = 4,
316 CEPH_OSD_CMPXATTR_OP_LT = 5,
317 CEPH_OSD_CMPXATTR_OP_LTE = 6
318};
319
320enum {
321 CEPH_OSD_CMPXATTR_MODE_STRING = 1,
322 CEPH_OSD_CMPXATTR_MODE_U64 = 2
323};
324
Sage Weil0dee3c22009-10-06 11:31:06 -0700325/*
326 * an individual object operation. each may be accompanied by some data
327 * payload
328 */
329struct ceph_osd_op {
330 __le16 op; /* CEPH_OSD_OP_* */
331 __le32 flags; /* CEPH_OSD_FLAG_* */
332 union {
333 struct {
334 __le64 offset, length;
Yehuda Sadeh0c948992010-02-01 16:10:45 -0800335 __le64 truncate_size;
336 __le32 truncate_seq;
Sage Weil0dee3c22009-10-06 11:31:06 -0700337 } __attribute__ ((packed)) extent;
338 struct {
339 __le32 name_len;
340 __le32 value_len;
Sage Weilca9d93a2010-05-12 14:48:20 -0700341 __u8 cmp_op; /* CEPH_OSD_CMPXATTR_OP_* */
342 __u8 cmp_mode; /* CEPH_OSD_CMPXATTR_MODE_* */
Sage Weil0dee3c22009-10-06 11:31:06 -0700343 } __attribute__ ((packed)) xattr;
344 struct {
Sage Weil0dee3c22009-10-06 11:31:06 -0700345 __u8 class_len;
346 __u8 method_len;
347 __u8 argc;
348 __le32 indata_len;
349 } __attribute__ ((packed)) cls;
350 struct {
351 __le64 cookie, count;
352 } __attribute__ ((packed)) pgls;
353 };
354 __le32 payload_len;
355} __attribute__ ((packed));
356
357/*
358 * osd request message header. each request may include multiple
359 * ceph_osd_op object operations.
360 */
361struct ceph_osd_request_head {
Sage Weil0dee3c22009-10-06 11:31:06 -0700362 __le32 client_inc; /* client incarnation */
363 struct ceph_object_layout layout; /* pgid */
364 __le32 osdmap_epoch; /* client's osdmap epoch */
365
366 __le32 flags;
367
368 struct ceph_timespec mtime; /* for mutations only */
369 struct ceph_eversion reassert_version; /* if we are replaying op */
370
371 __le32 object_len; /* length of object name */
372
373 __le64 snapid; /* snapid to read */
374 __le64 snap_seq; /* writer's snap context */
375 __le32 num_snaps;
376
377 __le16 num_ops;
378 struct ceph_osd_op ops[]; /* followed by ops[], obj, ticket, snaps */
379} __attribute__ ((packed));
380
381struct ceph_osd_reply_head {
Sage Weil0dee3c22009-10-06 11:31:06 -0700382 __le32 client_inc; /* client incarnation */
383 __le32 flags;
384 struct ceph_object_layout layout;
385 __le32 osdmap_epoch;
386 struct ceph_eversion reassert_version; /* for replaying uncommitted */
387
388 __le32 result; /* result code */
389
390 __le32 object_len; /* length of object name */
391 __le32 num_ops;
392 struct ceph_osd_op ops[0]; /* ops[], object */
393} __attribute__ ((packed));
394
395
396#endif