5 * Data types for the Ceph distributed object storage layer RADOS
6 * (Reliable Autonomic Distributed Object Store).
15 unsigned char fsid[16];
18 static inline int ceph_fsid_compare(const struct ceph_fsid *a,
19 const struct ceph_fsid *b)
21 return memcmp(a, b, sizeof(*a));
27 typedef __le64 ceph_snapid_t;
28 #define CEPH_SNAPDIR ((__u64)(-1)) /* reserved for hidden .snap dir */
29 #define CEPH_NOSNAP ((__u64)(-2)) /* "head", "live" revision */
30 #define CEPH_MAXSNAP ((__u64)(-3)) /* largest valid snapid */
32 struct ceph_timespec {
35 } __attribute__ ((packed));
39 * object layout - how objects are mapped into PGs
41 #define CEPH_OBJECT_LAYOUT_HASH 1
42 #define CEPH_OBJECT_LAYOUT_LINEAR 2
43 #define CEPH_OBJECT_LAYOUT_HASHINO 3
46 * pg layout -- how PGs are mapped onto (sets of) OSDs
48 #define CEPH_PG_LAYOUT_CRUSH 0
49 #define CEPH_PG_LAYOUT_HASH 1
50 #define CEPH_PG_LAYOUT_LINEAR 2
51 #define CEPH_PG_LAYOUT_HYBRID 3
53 #define CEPH_PG_MAX_SIZE 16 /* max # osds in a single pg */
57 * we encode this into one __le64.
60 __le16 preferred; /* preferred primary osd */
61 __le16 ps; /* placement seed */
62 __le32 pool; /* object pool */
63 } __attribute__ ((packed));
68 * NOTE: These map 1:1 on to the pg_pool_t::TYPE_* values. They are
69 * duplicated here only for CrushCompiler's benefit.
71 #define CEPH_PG_TYPE_REPLICATED 1
72 /* #define CEPH_PG_TYPE_RAID4 2 never implemented */
73 #define CEPH_PG_TYPE_ERASURE 3
76 * stable_mod func is used to control number of placement groups.
77 * similar to straight-up modulo, but produces a stable mapping as b
78 * increases over time. b is the number of bins, and bmask is the
79 * containing power of 2 minus 1.
81 * b <= bmask and bmask=(2**n)-1
82 * e.g., b=12 -> bmask=15, b=123 -> bmask=127
84 static inline int ceph_stable_mod(int x, int b, int bmask)
89 return x & (bmask >> 1);
93 * object layout - how a given object should be stored.
95 struct ceph_object_layout {
96 struct ceph_pg ol_pgid; /* raw pg, with _full_ ps precision. */
97 __le32 ol_stripe_unit; /* for per-object parity, if any */
98 } __attribute__ ((packed));
101 * compound epoch+version, used by storage layer to serialize mutations
103 struct ceph_eversion {
106 } __attribute__ ((packed));
113 #define CEPH_OSD_EXISTS (1<<0)
114 #define CEPH_OSD_UP (1<<1)
115 #define CEPH_OSD_AUTOOUT (1<<2) /* osd was automatically marked out */
116 #define CEPH_OSD_NEW (1<<3) /* osd is new, never marked in */
117 #define CEPH_OSD_FULL (1<<4) /* osd is at or above full threshold */
118 #define CEPH_OSD_NEARFULL (1<<5) /* osd is at or above nearfull threshold */
119 #define CEPH_OSD_BACKFILLFULL (1<<6) /* osd is at or above backfillfull threshold */
120 #define CEPH_OSD_DESTROYED (1<<7) /* osd has been destroyed */
121 #define CEPH_OSD_NOUP (1<<8) /* osd can not be marked up */
122 #define CEPH_OSD_NODOWN (1<<9) /* osd can not be marked down */
123 #define CEPH_OSD_NOIN (1<<10) /* osd can not be marked in */
124 #define CEPH_OSD_NOOUT (1<<11) /* osd can not be marked out */
126 extern const char *ceph_osd_state_name(int s);
128 /* osd weights. fixed point value: 0x10000 == 1.0 ("in"), 0 == "out" */
129 #define CEPH_OSD_IN 0x10000
130 #define CEPH_OSD_OUT 0
132 #define CEPH_OSD_MAX_PRIMARY_AFFINITY 0x10000
133 #define CEPH_OSD_DEFAULT_PRIMARY_AFFINITY 0x10000
139 #define CEPH_OSDMAP_NEARFULL (1<<0) /* sync writes (near ENOSPC) */
140 #define CEPH_OSDMAP_FULL (1<<1) /* no data writes (ENOSPC) */
141 #define CEPH_OSDMAP_PAUSERD (1<<2) /* pause all reads */
142 #define CEPH_OSDMAP_PAUSEWR (1<<3) /* pause all writes */
143 #define CEPH_OSDMAP_PAUSEREC (1<<4) /* pause recovery */
144 #define CEPH_OSDMAP_NOUP (1<<5) /* block osd boot */
145 #define CEPH_OSDMAP_NODOWN (1<<6) /* block osd mark-down/failure */
146 #define CEPH_OSDMAP_NOOUT (1<<7) /* block osd auto mark-out */
147 #define CEPH_OSDMAP_NOIN (1<<8) /* block osd auto mark-in */
148 #define CEPH_OSDMAP_NOBACKFILL (1<<9) /* block osd backfill */
149 #define CEPH_OSDMAP_NORECOVER (1<<10) /* block osd recovery and backfill */
150 #define CEPH_OSDMAP_NOSCRUB (1<<11) /* block periodic scrub */
151 #define CEPH_OSDMAP_NODEEP_SCRUB (1<<12) /* block periodic deep-scrub */
152 #define CEPH_OSDMAP_NOTIERAGENT (1<<13) /* disable tiering agent */
153 #define CEPH_OSDMAP_NOREBALANCE (1<<14) /* block osd backfill unless pg is degraded */
154 #define CEPH_OSDMAP_SORTBITWISE (1<<15) /* use bitwise hobject_t sort */
155 #define CEPH_OSDMAP_REQUIRE_JEWEL (1<<16) /* require jewel for booting osds */
156 #define CEPH_OSDMAP_REQUIRE_KRAKEN (1<<17) /* require kraken for booting osds */
157 #define CEPH_OSDMAP_REQUIRE_LUMINOUS (1<<18) /* require l for booting osds */
158 #define CEPH_OSDMAP_RECOVERY_DELETES (1<<19) /* deletes performed during recovery instead of peering */
159 #define CEPH_OSDMAP_PURGED_SNAPDIRS (1<<20) /* osds have converted snapsets */
161 /* these are hidden in 'ceph status' view */
162 #define CEPH_OSDMAP_SEMIHIDDEN_FLAGS (CEPH_OSDMAP_REQUIRE_JEWEL| \
163 CEPH_OSDMAP_REQUIRE_KRAKEN | \
164 CEPH_OSDMAP_REQUIRE_LUMINOUS | \
165 CEPH_OSDMAP_RECOVERY_DELETES | \
166 CEPH_OSDMAP_SORTBITWISE | \
167 CEPH_OSDMAP_PURGED_SNAPDIRS)
168 #define CEPH_OSDMAP_LEGACY_REQUIRE_FLAGS (CEPH_OSDMAP_REQUIRE_JEWEL | \
169 CEPH_OSDMAP_REQUIRE_KRAKEN | \
170 CEPH_OSDMAP_REQUIRE_LUMINOUS)
173 * major ceph release numbers
175 #define CEPH_RELEASE_ARGONAUT 1
176 #define CEPH_RELEASE_BOBTAIL 2
177 #define CEPH_RELEASE_CUTTLEFISH 3
178 #define CEPH_RELEASE_DUMPLING 4
179 #define CEPH_RELEASE_EMPEROR 5
180 #define CEPH_RELEASE_FIREFLY 6
181 #define CEPH_RELEASE_GIANT 7
182 #define CEPH_RELEASE_HAMMER 8
183 #define CEPH_RELEASE_INFERNALIS 9
184 #define CEPH_RELEASE_JEWEL 10
185 #define CEPH_RELEASE_KRAKEN 11
186 #define CEPH_RELEASE_LUMINOUS 12
187 #define CEPH_RELEASE_MIMIC 13
188 #define CEPH_RELEASE_MAX 14 /* highest + 1 */
190 extern const char *ceph_release_name(int r);
191 extern int ceph_release_from_name(const char *s);
192 extern uint64_t ceph_release_features(int r);
193 extern int ceph_release_from_features(uint64_t features);
196 * The error code to return when an OSD can't handle a write
197 * because it is too large.
199 #define OSD_WRITETOOBIG EMSGSIZE
204 * WARNING: do not use these op codes directly. Use the helpers
205 * defined below instead. In certain cases, op code behavior was
206 * redefined, resulting in special-cases in the helpers.
208 #define CEPH_OSD_OP_MODE 0xf000
209 #define CEPH_OSD_OP_MODE_RD 0x1000
210 #define CEPH_OSD_OP_MODE_WR 0x2000
211 #define CEPH_OSD_OP_MODE_RMW 0x3000
212 #define CEPH_OSD_OP_MODE_SUB 0x4000
213 #define CEPH_OSD_OP_MODE_CACHE 0x8000
215 #define CEPH_OSD_OP_TYPE 0x0f00
216 #define CEPH_OSD_OP_TYPE_DATA 0x0200
217 #define CEPH_OSD_OP_TYPE_ATTR 0x0300
218 #define CEPH_OSD_OP_TYPE_EXEC 0x0400
219 #define CEPH_OSD_OP_TYPE_PG 0x0500
220 // LEAVE UNUSED 0x0600 used to be multiobject ops
222 #define __CEPH_OSD_OP1(mode, nr) \
223 (CEPH_OSD_OP_MODE_##mode | (nr))
225 #define __CEPH_OSD_OP(mode, type, nr) \
226 (CEPH_OSD_OP_MODE_##mode | CEPH_OSD_OP_TYPE_##type | (nr))
228 #define __CEPH_FORALL_OSD_OPS(f) \
231 f(READ, __CEPH_OSD_OP(RD, DATA, 1), "read") \
232 f(STAT, __CEPH_OSD_OP(RD, DATA, 2), "stat") \
233 f(MAPEXT, __CEPH_OSD_OP(RD, DATA, 3), "mapext") \
234 f(CHECKSUM, __CEPH_OSD_OP(RD, DATA, 31), "checksum") \
237 f(MASKTRUNC, __CEPH_OSD_OP(RD, DATA, 4), "masktrunc") \
238 f(SPARSE_READ, __CEPH_OSD_OP(RD, DATA, 5), "sparse-read") \
240 f(NOTIFY, __CEPH_OSD_OP(RD, DATA, 6), "notify") \
241 f(NOTIFY_ACK, __CEPH_OSD_OP(RD, DATA, 7), "notify-ack") \
244 f(ASSERT_VER, __CEPH_OSD_OP(RD, DATA, 8), "assert-version") \
246 f(LIST_WATCHERS, __CEPH_OSD_OP(RD, DATA, 9), "list-watchers") \
248 f(LIST_SNAPS, __CEPH_OSD_OP(RD, DATA, 10), "list-snaps") \
251 f(SYNC_READ, __CEPH_OSD_OP(RD, DATA, 11), "sync_read") \
254 f(WRITE, __CEPH_OSD_OP(WR, DATA, 1), "write") \
255 f(WRITEFULL, __CEPH_OSD_OP(WR, DATA, 2), "writefull") \
256 f(TRUNCATE, __CEPH_OSD_OP(WR, DATA, 3), "truncate") \
257 f(ZERO, __CEPH_OSD_OP(WR, DATA, 4), "zero") \
258 f(DELETE, __CEPH_OSD_OP(WR, DATA, 5), "delete") \
261 f(APPEND, __CEPH_OSD_OP(WR, DATA, 6), "append") \
262 f(STARTSYNC, __CEPH_OSD_OP(WR, DATA, 7), "startsync") \
263 f(SETTRUNC, __CEPH_OSD_OP(WR, DATA, 8), "settrunc") \
264 f(TRIMTRUNC, __CEPH_OSD_OP(WR, DATA, 9), "trimtrunc") \
266 f(TMAPUP, __CEPH_OSD_OP(RMW, DATA, 10), "tmapup") \
267 f(TMAPPUT, __CEPH_OSD_OP(WR, DATA, 11), "tmapput") \
268 f(TMAPGET, __CEPH_OSD_OP(RD, DATA, 12), "tmapget") \
270 f(CREATE, __CEPH_OSD_OP(WR, DATA, 13), "create") \
271 f(ROLLBACK, __CEPH_OSD_OP(WR, DATA, 14), "rollback") \
273 f(WATCH, __CEPH_OSD_OP(WR, DATA, 15), "watch") \
276 f(OMAPGETKEYS, __CEPH_OSD_OP(RD, DATA, 17), "omap-get-keys") \
277 f(OMAPGETVALS, __CEPH_OSD_OP(RD, DATA, 18), "omap-get-vals") \
278 f(OMAPGETHEADER, __CEPH_OSD_OP(RD, DATA, 19), "omap-get-header") \
279 f(OMAPGETVALSBYKEYS, __CEPH_OSD_OP(RD, DATA, 20), "omap-get-vals-by-keys") \
280 f(OMAPSETVALS, __CEPH_OSD_OP(WR, DATA, 21), "omap-set-vals") \
281 f(OMAPSETHEADER, __CEPH_OSD_OP(WR, DATA, 22), "omap-set-header") \
282 f(OMAPCLEAR, __CEPH_OSD_OP(WR, DATA, 23), "omap-clear") \
283 f(OMAPRMKEYS, __CEPH_OSD_OP(WR, DATA, 24), "omap-rm-keys") \
284 f(OMAP_CMP, __CEPH_OSD_OP(RD, DATA, 25), "omap-cmp") \
287 f(COPY_FROM, __CEPH_OSD_OP(WR, DATA, 26), "copy-from") \
288 /* was copy-get-classic */ \
289 f(UNDIRTY, __CEPH_OSD_OP(WR, DATA, 28), "undirty") \
290 f(ISDIRTY, __CEPH_OSD_OP(RD, DATA, 29), "isdirty") \
291 f(COPY_GET, __CEPH_OSD_OP(RD, DATA, 30), "copy-get") \
292 f(CACHE_FLUSH, __CEPH_OSD_OP(CACHE, DATA, 31), "cache-flush") \
293 f(CACHE_EVICT, __CEPH_OSD_OP(CACHE, DATA, 32), "cache-evict") \
294 f(CACHE_TRY_FLUSH, __CEPH_OSD_OP(CACHE, DATA, 33), "cache-try-flush") \
296 /* convert tmap to omap */ \
297 f(TMAP2OMAP, __CEPH_OSD_OP(RMW, DATA, 34), "tmap2omap") \
300 f(SETALLOCHINT, __CEPH_OSD_OP(WR, DATA, 35), "set-alloc-hint") \
302 /* cache pin/unpin */ \
303 f(CACHE_PIN, __CEPH_OSD_OP(WR, DATA, 36), "cache-pin") \
304 f(CACHE_UNPIN, __CEPH_OSD_OP(WR, DATA, 37), "cache-unpin") \
307 f(WRITESAME, __CEPH_OSD_OP(WR, DATA, 38), "write-same") \
308 f(CMPEXT, __CEPH_OSD_OP(RD, DATA, 32), "cmpext") \
311 f(SET_REDIRECT, __CEPH_OSD_OP(WR, DATA, 39), "set-redirect") \
315 f(GETXATTR, __CEPH_OSD_OP(RD, ATTR, 1), "getxattr") \
316 f(GETXATTRS, __CEPH_OSD_OP(RD, ATTR, 2), "getxattrs") \
317 f(CMPXATTR, __CEPH_OSD_OP(RD, ATTR, 3), "cmpxattr") \
320 f(SETXATTR, __CEPH_OSD_OP(WR, ATTR, 1), "setxattr") \
321 f(SETXATTRS, __CEPH_OSD_OP(WR, ATTR, 2), "setxattrs") \
322 f(RESETXATTRS, __CEPH_OSD_OP(WR, ATTR, 3), "resetxattrs") \
323 f(RMXATTR, __CEPH_OSD_OP(WR, ATTR, 4), "rmxattr") \
326 f(PULL, __CEPH_OSD_OP1(SUB, 1), "pull") \
327 f(PUSH, __CEPH_OSD_OP1(SUB, 2), "push") \
328 f(BALANCEREADS, __CEPH_OSD_OP1(SUB, 3), "balance-reads") \
329 f(UNBALANCEREADS, __CEPH_OSD_OP1(SUB, 4), "unbalance-reads") \
330 f(SCRUB, __CEPH_OSD_OP1(SUB, 5), "scrub") \
331 f(SCRUB_RESERVE, __CEPH_OSD_OP1(SUB, 6), "scrub-reserve") \
332 f(SCRUB_UNRESERVE, __CEPH_OSD_OP1(SUB, 7), "scrub-unreserve") \
333 /* 8 used to be scrub-stop */ \
334 f(SCRUB_MAP, __CEPH_OSD_OP1(SUB, 9), "scrub-map") \
337 /* note: the RD bit here is wrong; see special-case below in helper */ \
338 f(CALL, __CEPH_OSD_OP(RD, EXEC, 1), "call") \
341 f(PGLS, __CEPH_OSD_OP(RD, PG, 1), "pgls") \
342 f(PGLS_FILTER, __CEPH_OSD_OP(RD, PG, 2), "pgls-filter") \
343 f(PG_HITSET_LS, __CEPH_OSD_OP(RD, PG, 3), "pg-hitset-ls") \
344 f(PG_HITSET_GET, __CEPH_OSD_OP(RD, PG, 4), "pg-hitset-get") \
345 f(PGNLS, __CEPH_OSD_OP(RD, PG, 5), "pgnls") \
346 f(PGNLS_FILTER, __CEPH_OSD_OP(RD, PG, 6), "pgnls-filter") \
347 f(SCRUBLS, __CEPH_OSD_OP(RD, PG, 7), "scrubls")
350 #define GENERATE_ENUM_ENTRY(op, opcode, str) CEPH_OSD_OP_##op = (opcode),
351 __CEPH_FORALL_OSD_OPS(GENERATE_ENUM_ENTRY)
352 #undef GENERATE_ENUM_ENTRY
355 static inline int ceph_osd_op_type_data(int op)
357 return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_DATA;
359 static inline int ceph_osd_op_type_attr(int op)
361 return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_ATTR;
363 static inline int ceph_osd_op_type_exec(int op)
365 return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_EXEC;
367 static inline int ceph_osd_op_type_pg(int op)
369 return (op & CEPH_OSD_OP_TYPE) == CEPH_OSD_OP_TYPE_PG;
372 static inline int ceph_osd_op_mode_subop(int op)
374 return (op & CEPH_OSD_OP_MODE) == CEPH_OSD_OP_MODE_SUB;
376 static inline int ceph_osd_op_mode_read(int op)
378 return (op & CEPH_OSD_OP_MODE_RD) &&
379 op != CEPH_OSD_OP_CALL;
381 static inline int ceph_osd_op_mode_modify(int op)
383 return op & CEPH_OSD_OP_MODE_WR;
385 static inline int ceph_osd_op_mode_cache(int op)
387 return op & CEPH_OSD_OP_MODE_CACHE;
389 static inline int ceph_osd_op_uses_extent(int op)
392 case CEPH_OSD_OP_READ:
393 case CEPH_OSD_OP_MAPEXT:
394 case CEPH_OSD_OP_MASKTRUNC:
395 case CEPH_OSD_OP_SPARSE_READ:
396 case CEPH_OSD_OP_SYNC_READ:
397 case CEPH_OSD_OP_WRITE:
398 case CEPH_OSD_OP_WRITEFULL:
399 case CEPH_OSD_OP_TRUNCATE:
400 case CEPH_OSD_OP_ZERO:
401 case CEPH_OSD_OP_APPEND:
402 case CEPH_OSD_OP_TRIMTRUNC:
403 case CEPH_OSD_OP_CMPEXT:
411 * note that the following tmap stuff is also defined in the ceph librados.h
412 * and objclass.h. Any modification here needs to be updated there
414 #define CEPH_OSD_TMAP_HDR 'h'
415 #define CEPH_OSD_TMAP_SET 's'
416 #define CEPH_OSD_TMAP_CREATE 'c' /* create key */
417 #define CEPH_OSD_TMAP_RM 'r'
418 #define CEPH_OSD_TMAP_RMSLOPPY 'R'
420 extern const char *ceph_osd_op_name(int op);
425 * An op may be READ, WRITE, or READ|WRITE.
428 CEPH_OSD_FLAG_ACK = 0x0001, /* want (or is) "ack" ack */
429 CEPH_OSD_FLAG_ONNVRAM = 0x0002, /* want (or is) "onnvram" ack */
430 CEPH_OSD_FLAG_ONDISK = 0x0004, /* want (or is) "ondisk" ack */
431 CEPH_OSD_FLAG_RETRY = 0x0008, /* resend attempt */
432 CEPH_OSD_FLAG_READ = 0x0010, /* op may read */
433 CEPH_OSD_FLAG_WRITE = 0x0020, /* op may write */
434 CEPH_OSD_FLAG_ORDERSNAP = 0x0040, /* EOLDSNAP if snapc is out of order */
435 CEPH_OSD_FLAG_PEERSTAT_OLD = 0x0080, /* DEPRECATED msg includes osd_peer_stat */
436 CEPH_OSD_FLAG_BALANCE_READS = 0x0100,
437 CEPH_OSD_FLAG_PARALLELEXEC = 0x0200, /* execute op in parallel */
438 CEPH_OSD_FLAG_PGOP = 0x0400, /* pg op, no object */
439 CEPH_OSD_FLAG_EXEC = 0x0800, /* op may exec */
440 CEPH_OSD_FLAG_EXEC_PUBLIC = 0x1000, /* DEPRECATED op may exec (public) */
441 CEPH_OSD_FLAG_LOCALIZE_READS = 0x2000, /* read from nearby replica, if any */
442 CEPH_OSD_FLAG_RWORDERED = 0x4000, /* order wrt concurrent reads */
443 CEPH_OSD_FLAG_IGNORE_CACHE = 0x8000, /* ignore cache logic */
444 CEPH_OSD_FLAG_SKIPRWLOCKS = 0x10000, /* skip rw locks */
445 CEPH_OSD_FLAG_IGNORE_OVERLAY =0x20000, /* ignore pool overlay */
446 CEPH_OSD_FLAG_FLUSH = 0x40000, /* this is part of flush */
447 CEPH_OSD_FLAG_MAP_SNAP_CLONE =0x80000, /* map snap direct to clone id
449 CEPH_OSD_FLAG_ENFORCE_SNAPC =0x100000, /* use snapc provided even if
450 pool uses pool snaps */
451 CEPH_OSD_FLAG_REDIRECTED = 0x200000, /* op has been redirected */
452 CEPH_OSD_FLAG_KNOWN_REDIR = 0x400000, /* redirect bit is authoritative */
453 CEPH_OSD_FLAG_FULL_TRY = 0x800000, /* try op despite full flag */
454 CEPH_OSD_FLAG_FULL_FORCE = 0x1000000, /* force op despite full flag */
455 CEPH_OSD_FLAG_IGNORE_REDIRECT = 0x2000000, /* ignore redirection */
459 CEPH_OSD_OP_FLAG_EXCL = 0x1, /* EXCL object create */
460 CEPH_OSD_OP_FLAG_FAILOK = 0x2, /* continue despite failure */
461 CEPH_OSD_OP_FLAG_FADVISE_RANDOM = 0x4, /* the op is random */
462 CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL = 0x8, /* the op is sequential */
463 CEPH_OSD_OP_FLAG_FADVISE_WILLNEED = 0x10,/* data will be accessed in the near future */
464 CEPH_OSD_OP_FLAG_FADVISE_DONTNEED = 0x20,/* data will not be accessed in the near future */
465 CEPH_OSD_OP_FLAG_FADVISE_NOCACHE = 0x40, /* data will be accessed only once by this client */
468 #define EOLDSNAPC 85 /* ORDERSNAP flag set; writer has old snapc*/
469 #define EBLACKLISTED 108 /* blacklisted */
471 /* xattr comparison */
473 CEPH_OSD_CMPXATTR_OP_EQ = 1,
474 CEPH_OSD_CMPXATTR_OP_NE = 2,
475 CEPH_OSD_CMPXATTR_OP_GT = 3,
476 CEPH_OSD_CMPXATTR_OP_GTE = 4,
477 CEPH_OSD_CMPXATTR_OP_LT = 5,
478 CEPH_OSD_CMPXATTR_OP_LTE = 6
482 CEPH_OSD_CMPXATTR_MODE_STRING = 1,
483 CEPH_OSD_CMPXATTR_MODE_U64 = 2
487 CEPH_OSD_COPY_FROM_FLAG_FLUSH = 1, /* part of a flush operation */
488 CEPH_OSD_COPY_FROM_FLAG_IGNORE_OVERLAY = 2, /* ignore pool overlay */
489 CEPH_OSD_COPY_FROM_FLAG_IGNORE_CACHE = 4, /* ignore osd cache logic */
490 CEPH_OSD_COPY_FROM_FLAG_MAP_SNAP_CLONE = 8, /* map snap direct to
492 CEPH_OSD_COPY_FROM_FLAG_RWORDERED = 16, /* order with write */
496 CEPH_OSD_TMAP2OMAP_NULLOK = 1,
500 CEPH_OSD_WATCH_OP_UNWATCH = 0,
501 CEPH_OSD_WATCH_OP_LEGACY_WATCH = 1,
502 /* note: use only ODD ids to prevent pre-giant code from
503 interpreting the op as UNWATCH */
504 CEPH_OSD_WATCH_OP_WATCH = 3,
505 CEPH_OSD_WATCH_OP_RECONNECT = 5,
506 CEPH_OSD_WATCH_OP_PING = 7,
510 CEPH_OSD_CHECKSUM_OP_TYPE_XXHASH32 = 0,
511 CEPH_OSD_CHECKSUM_OP_TYPE_XXHASH64 = 1,
512 CEPH_OSD_CHECKSUM_OP_TYPE_CRC32C = 2
515 const char *ceph_osd_watch_op_name(int o);
518 CEPH_OSD_ALLOC_HINT_FLAG_SEQUENTIAL_WRITE = 1,
519 CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_WRITE = 2,
520 CEPH_OSD_ALLOC_HINT_FLAG_SEQUENTIAL_READ = 4,
521 CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_READ = 8,
522 CEPH_OSD_ALLOC_HINT_FLAG_APPEND_ONLY = 16,
523 CEPH_OSD_ALLOC_HINT_FLAG_IMMUTABLE = 32,
524 CEPH_OSD_ALLOC_HINT_FLAG_SHORTLIVED = 64,
525 CEPH_OSD_ALLOC_HINT_FLAG_LONGLIVED = 128,
526 CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE = 256,
527 CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE = 512,
530 const char *ceph_osd_alloc_hint_flag_name(int f);
533 CEPH_OSD_BACKOFF_OP_BLOCK = 1,
534 CEPH_OSD_BACKOFF_OP_ACK_BLOCK = 2,
535 CEPH_OSD_BACKOFF_OP_UNBLOCK = 3,
538 const char *ceph_osd_backoff_op_name(int op);
541 * an individual object operation. each may be accompanied by some data
545 __le16 op; /* CEPH_OSD_OP_* */
546 __le32 flags; /* CEPH_OSD_OP_FLAG_* */
549 __le64 offset, length;
550 __le64 truncate_size;
552 } __attribute__ ((packed)) extent;
556 __u8 cmp_op; /* CEPH_OSD_CMPXATTR_OP_* */
557 __u8 cmp_mode; /* CEPH_OSD_CMPXATTR_MODE_* */
558 } __attribute__ ((packed)) xattr;
564 } __attribute__ ((packed)) cls;
567 __le32 start_epoch; /* for the pgls sequence */
568 } __attribute__ ((packed)) pgls;
571 } __attribute__ ((packed)) snap;
574 __le64 ver; /* no longer used */
575 __u8 op; /* CEPH_OSD_WATCH_OP_* */
576 __u32 gen; /* registration generation */
577 __u32 timeout; /* connection timeout */
578 } __attribute__ ((packed)) watch;
581 } __attribute__ ((packed)) notify;
585 } __attribute__ ((packed)) assert_ver;
587 __le64 offset, length;
589 } __attribute__ ((packed)) clonerange;
591 __le64 max; /* max data in reply */
592 } __attribute__ ((packed)) copy_get;
598 * __le32 flags: CEPH_OSD_OP_FLAG_FADVISE_: mean the fadvise flags for dest object
599 * src_fadvise_flags mean the fadvise flags for src object
601 __le32 src_fadvise_flags;
602 } __attribute__ ((packed)) copy_from;
604 struct ceph_timespec stamp;
605 } __attribute__ ((packed)) hit_set_get;
608 } __attribute__ ((packed)) tmap2omap;
610 __le64 expected_object_size;
611 __le64 expected_write_size;
612 __le32 flags; /* CEPH_OSD_OP_ALLOC_HINT_FLAG_* */
613 } __attribute__ ((packed)) alloc_hint;
618 } __attribute__ ((packed)) writesame;
623 __u8 type; /* CEPH_OSD_CHECKSUM_OP_TYPE_* */
624 } __attribute__ ((packed)) checksum;
627 } __attribute__ ((packed));
630 * Check the compatibility of struct ceph_osd_op
631 * (2+4+(2*8+8+4)+4) = (sizeof(ceph_osd_op::op) +
632 * sizeof(ceph_osd_op::flags) +
633 * sizeof(ceph_osd_op::extent) +
634 * sizeof(ceph_osd_op::payload_len))
637 static_assert(sizeof(ceph_osd_op) == (2+4+(2*8+8+4)+4),
638 "sizeof(ceph_osd_op) breaks the compatibility");
641 struct ceph_osd_reply_head {
642 __le32 client_inc; /* client incarnation */
644 struct ceph_object_layout layout;
646 struct ceph_eversion reassert_version; /* for replaying uncommitted */
648 __le32 result; /* result code */
650 __le32 object_len; /* length of object name */
652 struct ceph_osd_op ops[0]; /* ops[], object */
653 } __attribute__ ((packed));