1 /* 2 * ceph_fs.h - Ceph constants and data types to share between kernel and 3 * user space. 4 * 5 * Most types in this file are defined as little-endian, and are 6 * primarily intended to describe data structures that pass over the 7 * wire or that are stored on disk. 8 * 9 * LGPL2 10 */ 11 12 #ifndef CEPH_FS_H 13 #define CEPH_FS_H 14 15 #include <linux/ceph/msgr.h> 16 #include <linux/ceph/rados.h> 17 18 /* 19 * subprotocol versions. when specific messages types or high-level 20 * protocols change, bump the affected components. we keep rev 21 * internal cluster protocols separately from the public, 22 * client-facing protocol. 23 */ 24 #define CEPH_OSDC_PROTOCOL 24 /* server/client */ 25 #define CEPH_MDSC_PROTOCOL 32 /* server/client */ 26 #define CEPH_MONC_PROTOCOL 15 /* server/client */ 27 28 29 #define CEPH_INO_ROOT 1 30 #define CEPH_INO_CEPH 2 /* hidden .ceph dir */ 31 #define CEPH_INO_DOTDOT 3 /* used by ceph fuse for parent (..) */ 32 33 /* arbitrary limit on max # of monitors (cluster of 3 is typical) */ 34 #define CEPH_MAX_MON 31 35 36 /* 37 * legacy ceph_file_layoute 38 */ 39 struct ceph_file_layout_legacy { 40 /* file -> object mapping */ 41 __le32 fl_stripe_unit; /* stripe unit, in bytes. must be multiple 42 of page size. */ 43 __le32 fl_stripe_count; /* over this many objects */ 44 __le32 fl_object_size; /* until objects are this big, then move to 45 new objects */ 46 __le32 fl_cas_hash; /* UNUSED. 0 = none; 1 = sha256 */ 47 48 /* pg -> disk layout */ 49 __le32 fl_object_stripe_unit; /* UNUSED. for per-object parity, if any */ 50 51 /* object -> pg layout */ 52 __le32 fl_unused; /* unused; used to be preferred primary for pg (-1 for none) */ 53 __le32 fl_pg_pool; /* namespace, crush ruleset, rep level */ 54 } __attribute__ ((packed)); 55 56 struct ceph_string; 57 /* 58 * ceph_file_layout - describe data layout for a file/inode 59 */ 60 struct ceph_file_layout { 61 /* file -> object mapping */ 62 u32 stripe_unit; /* stripe unit, in bytes */ 63 u32 stripe_count; /* over this many objects */ 64 u32 object_size; /* until objects are this big */ 65 s64 pool_id; /* rados pool id */ 66 struct ceph_string __rcu *pool_ns; /* rados pool namespace */ 67 }; 68 69 extern int ceph_file_layout_is_valid(const struct ceph_file_layout *layout); 70 extern void ceph_file_layout_from_legacy(struct ceph_file_layout *fl, 71 struct ceph_file_layout_legacy *legacy); 72 extern void ceph_file_layout_to_legacy(struct ceph_file_layout *fl, 73 struct ceph_file_layout_legacy *legacy); 74 75 #define CEPH_MIN_STRIPE_UNIT 65536 76 77 struct ceph_dir_layout { 78 __u8 dl_dir_hash; /* see ceph_hash.h for ids */ 79 __u8 dl_unused1; 80 __u16 dl_unused2; 81 __u32 dl_unused3; 82 } __attribute__ ((packed)); 83 84 /* crypto algorithms */ 85 #define CEPH_CRYPTO_NONE 0x0 86 #define CEPH_CRYPTO_AES 0x1 87 88 #define CEPH_AES_IV "cephsageyudagreg" 89 90 /* security/authentication protocols */ 91 #define CEPH_AUTH_UNKNOWN 0x0 92 #define CEPH_AUTH_NONE 0x1 93 #define CEPH_AUTH_CEPHX 0x2 94 95 #define CEPH_AUTH_UID_DEFAULT ((__u64) -1) 96 97 98 /********************************************* 99 * message layer 100 */ 101 102 /* 103 * message types 104 */ 105 106 /* misc */ 107 #define CEPH_MSG_SHUTDOWN 1 108 #define CEPH_MSG_PING 2 109 110 /* client <-> monitor */ 111 #define CEPH_MSG_MON_MAP 4 112 #define CEPH_MSG_MON_GET_MAP 5 113 #define CEPH_MSG_STATFS 13 114 #define CEPH_MSG_STATFS_REPLY 14 115 #define CEPH_MSG_MON_SUBSCRIBE 15 116 #define CEPH_MSG_MON_SUBSCRIBE_ACK 16 117 #define CEPH_MSG_AUTH 17 118 #define CEPH_MSG_AUTH_REPLY 18 119 #define CEPH_MSG_MON_GET_VERSION 19 120 #define CEPH_MSG_MON_GET_VERSION_REPLY 20 121 122 /* client <-> mds */ 123 #define CEPH_MSG_MDS_MAP 21 124 #define CEPH_MSG_FS_MAP_USER 103 125 126 #define CEPH_MSG_CLIENT_SESSION 22 127 #define CEPH_MSG_CLIENT_RECONNECT 23 128 129 #define CEPH_MSG_CLIENT_REQUEST 24 130 #define CEPH_MSG_CLIENT_REQUEST_FORWARD 25 131 #define CEPH_MSG_CLIENT_REPLY 26 132 #define CEPH_MSG_CLIENT_CAPS 0x310 133 #define CEPH_MSG_CLIENT_LEASE 0x311 134 #define CEPH_MSG_CLIENT_SNAP 0x312 135 #define CEPH_MSG_CLIENT_CAPRELEASE 0x313 136 137 /* pool ops */ 138 #define CEPH_MSG_POOLOP_REPLY 48 139 #define CEPH_MSG_POOLOP 49 140 141 /* mon commands */ 142 #define CEPH_MSG_MON_COMMAND 50 143 #define CEPH_MSG_MON_COMMAND_ACK 51 144 145 /* osd */ 146 #define CEPH_MSG_OSD_MAP 41 147 #define CEPH_MSG_OSD_OP 42 148 #define CEPH_MSG_OSD_OPREPLY 43 149 #define CEPH_MSG_WATCH_NOTIFY 44 150 #define CEPH_MSG_OSD_BACKOFF 61 151 152 153 /* watch-notify operations */ 154 enum { 155 CEPH_WATCH_EVENT_NOTIFY = 1, /* notifying watcher */ 156 CEPH_WATCH_EVENT_NOTIFY_COMPLETE = 2, /* notifier notified when done */ 157 CEPH_WATCH_EVENT_DISCONNECT = 3, /* we were disconnected */ 158 }; 159 160 161 struct ceph_mon_request_header { 162 __le64 have_version; 163 __le16 session_mon; 164 __le64 session_mon_tid; 165 } __attribute__ ((packed)); 166 167 struct ceph_mon_statfs { 168 struct ceph_mon_request_header monhdr; 169 struct ceph_fsid fsid; 170 __u8 contains_data_pool; 171 __le64 data_pool; 172 } __attribute__ ((packed)); 173 174 struct ceph_statfs { 175 __le64 kb, kb_used, kb_avail; 176 __le64 num_objects; 177 } __attribute__ ((packed)); 178 179 struct ceph_mon_statfs_reply { 180 struct ceph_fsid fsid; 181 __le64 version; 182 struct ceph_statfs st; 183 } __attribute__ ((packed)); 184 185 struct ceph_mon_command { 186 struct ceph_mon_request_header monhdr; 187 struct ceph_fsid fsid; 188 __le32 num_strs; /* always 1 */ 189 __le32 str_len; 190 char str[]; 191 } __attribute__ ((packed)); 192 193 struct ceph_osd_getmap { 194 struct ceph_mon_request_header monhdr; 195 struct ceph_fsid fsid; 196 __le32 start; 197 } __attribute__ ((packed)); 198 199 struct ceph_mds_getmap { 200 struct ceph_mon_request_header monhdr; 201 struct ceph_fsid fsid; 202 } __attribute__ ((packed)); 203 204 struct ceph_client_mount { 205 struct ceph_mon_request_header monhdr; 206 } __attribute__ ((packed)); 207 208 #define CEPH_SUBSCRIBE_ONETIME 1 /* i want only 1 update after have */ 209 210 struct ceph_mon_subscribe_item { 211 __le64 start; 212 __u8 flags; 213 } __attribute__ ((packed)); 214 215 struct ceph_mon_subscribe_ack { 216 __le32 duration; /* seconds */ 217 struct ceph_fsid fsid; 218 } __attribute__ ((packed)); 219 220 #define CEPH_FS_CLUSTER_ID_NONE -1 221 222 /* 223 * mdsmap flags 224 */ 225 #define CEPH_MDSMAP_DOWN (1<<0) /* cluster deliberately down */ 226 227 /* 228 * mds states 229 * > 0 -> in 230 * <= 0 -> out 231 */ 232 #define CEPH_MDS_STATE_DNE 0 /* down, does not exist. */ 233 #define CEPH_MDS_STATE_STOPPED -1 /* down, once existed, but no subtrees. 234 empty log. */ 235 #define CEPH_MDS_STATE_BOOT -4 /* up, boot announcement. */ 236 #define CEPH_MDS_STATE_STANDBY -5 /* up, idle. waiting for assignment. */ 237 #define CEPH_MDS_STATE_CREATING -6 /* up, creating MDS instance. */ 238 #define CEPH_MDS_STATE_STARTING -7 /* up, starting previously stopped mds */ 239 #define CEPH_MDS_STATE_STANDBY_REPLAY -8 /* up, tailing active node's journal */ 240 #define CEPH_MDS_STATE_REPLAYONCE -9 /* up, replaying an active node's journal */ 241 242 #define CEPH_MDS_STATE_REPLAY 8 /* up, replaying journal. */ 243 #define CEPH_MDS_STATE_RESOLVE 9 /* up, disambiguating distributed 244 operations (import, rename, etc.) */ 245 #define CEPH_MDS_STATE_RECONNECT 10 /* up, reconnect to clients */ 246 #define CEPH_MDS_STATE_REJOIN 11 /* up, rejoining distributed cache */ 247 #define CEPH_MDS_STATE_CLIENTREPLAY 12 /* up, replaying client operations */ 248 #define CEPH_MDS_STATE_ACTIVE 13 /* up, active */ 249 #define CEPH_MDS_STATE_STOPPING 14 /* up, but exporting metadata */ 250 251 extern const char *ceph_mds_state_name(int s); 252 253 254 /* 255 * metadata lock types. 256 * - these are bitmasks.. we can compose them 257 * - they also define the lock ordering by the MDS 258 * - a few of these are internal to the mds 259 */ 260 #define CEPH_LOCK_DVERSION 1 261 #define CEPH_LOCK_DN 2 262 #define CEPH_LOCK_ISNAP 16 263 #define CEPH_LOCK_IVERSION 32 /* mds internal */ 264 #define CEPH_LOCK_IFILE 64 265 #define CEPH_LOCK_IAUTH 128 266 #define CEPH_LOCK_ILINK 256 267 #define CEPH_LOCK_IDFT 512 /* dir frag tree */ 268 #define CEPH_LOCK_INEST 1024 /* mds internal */ 269 #define CEPH_LOCK_IXATTR 2048 270 #define CEPH_LOCK_IFLOCK 4096 /* advisory file locks */ 271 #define CEPH_LOCK_INO 8192 /* immutable inode bits; not a lock */ 272 #define CEPH_LOCK_IPOLICY 16384 /* policy lock on dirs. MDS internal */ 273 274 /* client_session ops */ 275 enum { 276 CEPH_SESSION_REQUEST_OPEN, 277 CEPH_SESSION_OPEN, 278 CEPH_SESSION_REQUEST_CLOSE, 279 CEPH_SESSION_CLOSE, 280 CEPH_SESSION_REQUEST_RENEWCAPS, 281 CEPH_SESSION_RENEWCAPS, 282 CEPH_SESSION_STALE, 283 CEPH_SESSION_RECALL_STATE, 284 CEPH_SESSION_FLUSHMSG, 285 CEPH_SESSION_FLUSHMSG_ACK, 286 CEPH_SESSION_FORCE_RO, 287 CEPH_SESSION_REJECT, 288 }; 289 290 extern const char *ceph_session_op_name(int op); 291 292 struct ceph_mds_session_head { 293 __le32 op; 294 __le64 seq; 295 struct ceph_timespec stamp; 296 __le32 max_caps, max_leases; 297 } __attribute__ ((packed)); 298 299 /* client_request */ 300 /* 301 * metadata ops. 302 * & 0x001000 -> write op 303 * & 0x010000 -> follow symlink (e.g. stat(), not lstat()). 304 & & 0x100000 -> use weird ino/path trace 305 */ 306 #define CEPH_MDS_OP_WRITE 0x001000 307 enum { 308 CEPH_MDS_OP_LOOKUP = 0x00100, 309 CEPH_MDS_OP_GETATTR = 0x00101, 310 CEPH_MDS_OP_LOOKUPHASH = 0x00102, 311 CEPH_MDS_OP_LOOKUPPARENT = 0x00103, 312 CEPH_MDS_OP_LOOKUPINO = 0x00104, 313 CEPH_MDS_OP_LOOKUPNAME = 0x00105, 314 315 CEPH_MDS_OP_SETXATTR = 0x01105, 316 CEPH_MDS_OP_RMXATTR = 0x01106, 317 CEPH_MDS_OP_SETLAYOUT = 0x01107, 318 CEPH_MDS_OP_SETATTR = 0x01108, 319 CEPH_MDS_OP_SETFILELOCK= 0x01109, 320 CEPH_MDS_OP_GETFILELOCK= 0x00110, 321 CEPH_MDS_OP_SETDIRLAYOUT=0x0110a, 322 323 CEPH_MDS_OP_MKNOD = 0x01201, 324 CEPH_MDS_OP_LINK = 0x01202, 325 CEPH_MDS_OP_UNLINK = 0x01203, 326 CEPH_MDS_OP_RENAME = 0x01204, 327 CEPH_MDS_OP_MKDIR = 0x01220, 328 CEPH_MDS_OP_RMDIR = 0x01221, 329 CEPH_MDS_OP_SYMLINK = 0x01222, 330 331 CEPH_MDS_OP_CREATE = 0x01301, 332 CEPH_MDS_OP_OPEN = 0x00302, 333 CEPH_MDS_OP_READDIR = 0x00305, 334 335 CEPH_MDS_OP_LOOKUPSNAP = 0x00400, 336 CEPH_MDS_OP_MKSNAP = 0x01400, 337 CEPH_MDS_OP_RMSNAP = 0x01401, 338 CEPH_MDS_OP_LSSNAP = 0x00402, 339 CEPH_MDS_OP_RENAMESNAP = 0x01403, 340 }; 341 342 extern const char *ceph_mds_op_name(int op); 343 344 345 #define CEPH_SETATTR_MODE 1 346 #define CEPH_SETATTR_UID 2 347 #define CEPH_SETATTR_GID 4 348 #define CEPH_SETATTR_MTIME 8 349 #define CEPH_SETATTR_ATIME 16 350 #define CEPH_SETATTR_SIZE 32 351 #define CEPH_SETATTR_CTIME 64 352 353 /* 354 * Ceph setxattr request flags. 355 */ 356 #define CEPH_XATTR_CREATE (1 << 0) 357 #define CEPH_XATTR_REPLACE (1 << 1) 358 #define CEPH_XATTR_REMOVE (1 << 31) 359 360 /* 361 * readdir request flags; 362 */ 363 #define CEPH_READDIR_REPLY_BITFLAGS (1<<0) 364 365 /* 366 * readdir reply flags. 367 */ 368 #define CEPH_READDIR_FRAG_END (1<<0) 369 #define CEPH_READDIR_FRAG_COMPLETE (1<<8) 370 #define CEPH_READDIR_HASH_ORDER (1<<9) 371 #define CEPH_READDIR_OFFSET_HASH (1<<10) 372 373 /* 374 * open request flags 375 */ 376 #define CEPH_O_RDONLY 00000000 377 #define CEPH_O_WRONLY 00000001 378 #define CEPH_O_RDWR 00000002 379 #define CEPH_O_CREAT 00000100 380 #define CEPH_O_EXCL 00000200 381 #define CEPH_O_TRUNC 00001000 382 #define CEPH_O_DIRECTORY 00200000 383 #define CEPH_O_NOFOLLOW 00400000 384 385 union ceph_mds_request_args { 386 struct { 387 __le32 mask; /* CEPH_CAP_* */ 388 } __attribute__ ((packed)) getattr; 389 struct { 390 __le32 mode; 391 __le32 uid; 392 __le32 gid; 393 struct ceph_timespec mtime; 394 struct ceph_timespec atime; 395 __le64 size, old_size; /* old_size needed by truncate */ 396 __le32 mask; /* CEPH_SETATTR_* */ 397 } __attribute__ ((packed)) setattr; 398 struct { 399 __le32 frag; /* which dir fragment */ 400 __le32 max_entries; /* how many dentries to grab */ 401 __le32 max_bytes; 402 __le16 flags; 403 __le32 offset_hash; 404 } __attribute__ ((packed)) readdir; 405 struct { 406 __le32 mode; 407 __le32 rdev; 408 } __attribute__ ((packed)) mknod; 409 struct { 410 __le32 mode; 411 } __attribute__ ((packed)) mkdir; 412 struct { 413 __le32 flags; 414 __le32 mode; 415 __le32 stripe_unit; /* layout for newly created file */ 416 __le32 stripe_count; /* ... */ 417 __le32 object_size; 418 __le32 file_replication; 419 __le32 mask; /* CEPH_CAP_* */ 420 __le32 old_size; 421 } __attribute__ ((packed)) open; 422 struct { 423 __le32 flags; 424 } __attribute__ ((packed)) setxattr; 425 struct { 426 struct ceph_file_layout_legacy layout; 427 } __attribute__ ((packed)) setlayout; 428 struct { 429 __u8 rule; /* currently fcntl or flock */ 430 __u8 type; /* shared, exclusive, remove*/ 431 __le64 owner; /* owner of the lock */ 432 __le64 pid; /* process id requesting the lock */ 433 __le64 start; /* initial location to lock */ 434 __le64 length; /* num bytes to lock from start */ 435 __u8 wait; /* will caller wait for lock to become available? */ 436 } __attribute__ ((packed)) filelock_change; 437 } __attribute__ ((packed)); 438 439 #define CEPH_MDS_FLAG_REPLAY 1 /* this is a replayed op */ 440 #define CEPH_MDS_FLAG_WANT_DENTRY 2 /* want dentry in reply */ 441 442 struct ceph_mds_request_head { 443 __le64 oldest_client_tid; 444 __le32 mdsmap_epoch; /* on client */ 445 __le32 flags; /* CEPH_MDS_FLAG_* */ 446 __u8 num_retry, num_fwd; /* count retry, fwd attempts */ 447 __le16 num_releases; /* # include cap/lease release records */ 448 __le32 op; /* mds op code */ 449 __le32 caller_uid, caller_gid; 450 __le64 ino; /* use this ino for openc, mkdir, mknod, 451 etc. (if replaying) */ 452 union ceph_mds_request_args args; 453 } __attribute__ ((packed)); 454 455 /* cap/lease release record */ 456 struct ceph_mds_request_release { 457 __le64 ino, cap_id; /* ino and unique cap id */ 458 __le32 caps, wanted; /* new issued, wanted */ 459 __le32 seq, issue_seq, mseq; 460 __le32 dname_seq; /* if releasing a dentry lease, a */ 461 __le32 dname_len; /* string follows. */ 462 } __attribute__ ((packed)); 463 464 /* client reply */ 465 struct ceph_mds_reply_head { 466 __le32 op; 467 __le32 result; 468 __le32 mdsmap_epoch; 469 __u8 safe; /* true if committed to disk */ 470 __u8 is_dentry, is_target; /* true if dentry, target inode records 471 are included with reply */ 472 } __attribute__ ((packed)); 473 474 /* one for each node split */ 475 struct ceph_frag_tree_split { 476 __le32 frag; /* this frag splits... */ 477 __le32 by; /* ...by this many bits */ 478 } __attribute__ ((packed)); 479 480 struct ceph_frag_tree_head { 481 __le32 nsplits; /* num ceph_frag_tree_split records */ 482 struct ceph_frag_tree_split splits[]; 483 } __attribute__ ((packed)); 484 485 /* capability issue, for bundling with mds reply */ 486 struct ceph_mds_reply_cap { 487 __le32 caps, wanted; /* caps issued, wanted */ 488 __le64 cap_id; 489 __le32 seq, mseq; 490 __le64 realm; /* snap realm */ 491 __u8 flags; /* CEPH_CAP_FLAG_* */ 492 } __attribute__ ((packed)); 493 494 #define CEPH_CAP_FLAG_AUTH (1 << 0) /* cap is issued by auth mds */ 495 #define CEPH_CAP_FLAG_RELEASE (1 << 1) /* release the cap */ 496 497 /* inode record, for bundling with mds reply */ 498 struct ceph_mds_reply_inode { 499 __le64 ino; 500 __le64 snapid; 501 __le32 rdev; 502 __le64 version; /* inode version */ 503 __le64 xattr_version; /* version for xattr blob */ 504 struct ceph_mds_reply_cap cap; /* caps issued for this inode */ 505 struct ceph_file_layout_legacy layout; 506 struct ceph_timespec ctime, mtime, atime; 507 __le32 time_warp_seq; 508 __le64 size, max_size, truncate_size; 509 __le32 truncate_seq; 510 __le32 mode, uid, gid; 511 __le32 nlink; 512 __le64 files, subdirs, rbytes, rfiles, rsubdirs; /* dir stats */ 513 struct ceph_timespec rctime; 514 struct ceph_frag_tree_head fragtree; /* (must be at end of struct) */ 515 } __attribute__ ((packed)); 516 /* followed by frag array, symlink string, dir layout, xattr blob */ 517 518 /* reply_lease follows dname, and reply_inode */ 519 struct ceph_mds_reply_lease { 520 __le16 mask; /* lease type(s) */ 521 __le32 duration_ms; /* lease duration */ 522 __le32 seq; 523 } __attribute__ ((packed)); 524 525 struct ceph_mds_reply_dirfrag { 526 __le32 frag; /* fragment */ 527 __le32 auth; /* auth mds, if this is a delegation point */ 528 __le32 ndist; /* number of mds' this is replicated on */ 529 __le32 dist[]; 530 } __attribute__ ((packed)); 531 532 #define CEPH_LOCK_FCNTL 1 533 #define CEPH_LOCK_FLOCK 2 534 #define CEPH_LOCK_FCNTL_INTR 3 535 #define CEPH_LOCK_FLOCK_INTR 4 536 537 538 #define CEPH_LOCK_SHARED 1 539 #define CEPH_LOCK_EXCL 2 540 #define CEPH_LOCK_UNLOCK 4 541 542 struct ceph_filelock { 543 __le64 start;/* file offset to start lock at */ 544 __le64 length; /* num bytes to lock; 0 for all following start */ 545 __le64 client; /* which client holds the lock */ 546 __le64 owner; /* owner the lock */ 547 __le64 pid; /* process id holding the lock on the client */ 548 __u8 type; /* shared lock, exclusive lock, or unlock */ 549 } __attribute__ ((packed)); 550 551 552 /* file access modes */ 553 #define CEPH_FILE_MODE_PIN 0 554 #define CEPH_FILE_MODE_RD 1 555 #define CEPH_FILE_MODE_WR 2 556 #define CEPH_FILE_MODE_RDWR 3 /* RD | WR */ 557 #define CEPH_FILE_MODE_LAZY 4 /* lazy io */ 558 #define CEPH_FILE_MODE_BITS 4 559 560 int ceph_flags_to_mode(int flags); 561 562 #define CEPH_INLINE_NONE ((__u64)-1) 563 564 /* capability bits */ 565 #define CEPH_CAP_PIN 1 /* no specific capabilities beyond the pin */ 566 567 /* generic cap bits */ 568 #define CEPH_CAP_GSHARED 1 /* client can reads */ 569 #define CEPH_CAP_GEXCL 2 /* client can read and update */ 570 #define CEPH_CAP_GCACHE 4 /* (file) client can cache reads */ 571 #define CEPH_CAP_GRD 8 /* (file) client can read */ 572 #define CEPH_CAP_GWR 16 /* (file) client can write */ 573 #define CEPH_CAP_GBUFFER 32 /* (file) client can buffer writes */ 574 #define CEPH_CAP_GWREXTEND 64 /* (file) client can extend EOF */ 575 #define CEPH_CAP_GLAZYIO 128 /* (file) client can perform lazy io */ 576 577 #define CEPH_CAP_SIMPLE_BITS 2 578 #define CEPH_CAP_FILE_BITS 8 579 580 /* per-lock shift */ 581 #define CEPH_CAP_SAUTH 2 582 #define CEPH_CAP_SLINK 4 583 #define CEPH_CAP_SXATTR 6 584 #define CEPH_CAP_SFILE 8 585 #define CEPH_CAP_SFLOCK 20 586 587 #define CEPH_CAP_BITS 22 588 589 /* composed values */ 590 #define CEPH_CAP_AUTH_SHARED (CEPH_CAP_GSHARED << CEPH_CAP_SAUTH) 591 #define CEPH_CAP_AUTH_EXCL (CEPH_CAP_GEXCL << CEPH_CAP_SAUTH) 592 #define CEPH_CAP_LINK_SHARED (CEPH_CAP_GSHARED << CEPH_CAP_SLINK) 593 #define CEPH_CAP_LINK_EXCL (CEPH_CAP_GEXCL << CEPH_CAP_SLINK) 594 #define CEPH_CAP_XATTR_SHARED (CEPH_CAP_GSHARED << CEPH_CAP_SXATTR) 595 #define CEPH_CAP_XATTR_EXCL (CEPH_CAP_GEXCL << CEPH_CAP_SXATTR) 596 #define CEPH_CAP_FILE(x) (x << CEPH_CAP_SFILE) 597 #define CEPH_CAP_FILE_SHARED (CEPH_CAP_GSHARED << CEPH_CAP_SFILE) 598 #define CEPH_CAP_FILE_EXCL (CEPH_CAP_GEXCL << CEPH_CAP_SFILE) 599 #define CEPH_CAP_FILE_CACHE (CEPH_CAP_GCACHE << CEPH_CAP_SFILE) 600 #define CEPH_CAP_FILE_RD (CEPH_CAP_GRD << CEPH_CAP_SFILE) 601 #define CEPH_CAP_FILE_WR (CEPH_CAP_GWR << CEPH_CAP_SFILE) 602 #define CEPH_CAP_FILE_BUFFER (CEPH_CAP_GBUFFER << CEPH_CAP_SFILE) 603 #define CEPH_CAP_FILE_WREXTEND (CEPH_CAP_GWREXTEND << CEPH_CAP_SFILE) 604 #define CEPH_CAP_FILE_LAZYIO (CEPH_CAP_GLAZYIO << CEPH_CAP_SFILE) 605 #define CEPH_CAP_FLOCK_SHARED (CEPH_CAP_GSHARED << CEPH_CAP_SFLOCK) 606 #define CEPH_CAP_FLOCK_EXCL (CEPH_CAP_GEXCL << CEPH_CAP_SFLOCK) 607 608 609 /* cap masks (for getattr) */ 610 #define CEPH_STAT_CAP_INODE CEPH_CAP_PIN 611 #define CEPH_STAT_CAP_TYPE CEPH_CAP_PIN /* mode >> 12 */ 612 #define CEPH_STAT_CAP_SYMLINK CEPH_CAP_PIN 613 #define CEPH_STAT_CAP_UID CEPH_CAP_AUTH_SHARED 614 #define CEPH_STAT_CAP_GID CEPH_CAP_AUTH_SHARED 615 #define CEPH_STAT_CAP_MODE CEPH_CAP_AUTH_SHARED 616 #define CEPH_STAT_CAP_NLINK CEPH_CAP_LINK_SHARED 617 #define CEPH_STAT_CAP_LAYOUT CEPH_CAP_FILE_SHARED 618 #define CEPH_STAT_CAP_MTIME CEPH_CAP_FILE_SHARED 619 #define CEPH_STAT_CAP_SIZE CEPH_CAP_FILE_SHARED 620 #define CEPH_STAT_CAP_ATIME CEPH_CAP_FILE_SHARED /* fixme */ 621 #define CEPH_STAT_CAP_XATTR CEPH_CAP_XATTR_SHARED 622 #define CEPH_STAT_CAP_INODE_ALL (CEPH_CAP_PIN | \ 623 CEPH_CAP_AUTH_SHARED | \ 624 CEPH_CAP_LINK_SHARED | \ 625 CEPH_CAP_FILE_SHARED | \ 626 CEPH_CAP_XATTR_SHARED) 627 #define CEPH_STAT_CAP_INLINE_DATA (CEPH_CAP_FILE_SHARED | \ 628 CEPH_CAP_FILE_RD) 629 630 #define CEPH_CAP_ANY_SHARED (CEPH_CAP_AUTH_SHARED | \ 631 CEPH_CAP_LINK_SHARED | \ 632 CEPH_CAP_XATTR_SHARED | \ 633 CEPH_CAP_FILE_SHARED) 634 #define CEPH_CAP_ANY_RD (CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_RD | \ 635 CEPH_CAP_FILE_CACHE) 636 637 #define CEPH_CAP_ANY_EXCL (CEPH_CAP_AUTH_EXCL | \ 638 CEPH_CAP_LINK_EXCL | \ 639 CEPH_CAP_XATTR_EXCL | \ 640 CEPH_CAP_FILE_EXCL) 641 #define CEPH_CAP_ANY_FILE_RD (CEPH_CAP_FILE_RD | CEPH_CAP_FILE_CACHE | \ 642 CEPH_CAP_FILE_SHARED) 643 #define CEPH_CAP_ANY_FILE_WR (CEPH_CAP_FILE_WR | CEPH_CAP_FILE_BUFFER | \ 644 CEPH_CAP_FILE_EXCL) 645 #define CEPH_CAP_ANY_WR (CEPH_CAP_ANY_EXCL | CEPH_CAP_ANY_FILE_WR) 646 #define CEPH_CAP_ANY (CEPH_CAP_ANY_RD | CEPH_CAP_ANY_EXCL | \ 647 CEPH_CAP_ANY_FILE_WR | CEPH_CAP_FILE_LAZYIO | \ 648 CEPH_CAP_PIN) 649 650 #define CEPH_CAP_LOCKS (CEPH_LOCK_IFILE | CEPH_LOCK_IAUTH | CEPH_LOCK_ILINK | \ 651 CEPH_LOCK_IXATTR) 652 653 int ceph_caps_for_mode(int mode); 654 655 enum { 656 CEPH_CAP_OP_GRANT, /* mds->client grant */ 657 CEPH_CAP_OP_REVOKE, /* mds->client revoke */ 658 CEPH_CAP_OP_TRUNC, /* mds->client trunc notify */ 659 CEPH_CAP_OP_EXPORT, /* mds has exported the cap */ 660 CEPH_CAP_OP_IMPORT, /* mds has imported the cap */ 661 CEPH_CAP_OP_UPDATE, /* client->mds update */ 662 CEPH_CAP_OP_DROP, /* client->mds drop cap bits */ 663 CEPH_CAP_OP_FLUSH, /* client->mds cap writeback */ 664 CEPH_CAP_OP_FLUSH_ACK, /* mds->client flushed */ 665 CEPH_CAP_OP_FLUSHSNAP, /* client->mds flush snapped metadata */ 666 CEPH_CAP_OP_FLUSHSNAP_ACK, /* mds->client flushed snapped metadata */ 667 CEPH_CAP_OP_RELEASE, /* client->mds release (clean) cap */ 668 CEPH_CAP_OP_RENEW, /* client->mds renewal request */ 669 }; 670 671 extern const char *ceph_cap_op_name(int op); 672 673 /* flags field in client cap messages (version >= 10) */ 674 #define CEPH_CLIENT_CAPS_SYNC (1<<0) 675 #define CEPH_CLIENT_CAPS_NO_CAPSNAP (1<<1) 676 #define CEPH_CLIENT_CAPS_PENDING_CAPSNAP (1<<2); 677 678 /* 679 * caps message, used for capability callbacks, acks, requests, etc. 680 */ 681 struct ceph_mds_caps { 682 __le32 op; /* CEPH_CAP_OP_* */ 683 __le64 ino, realm; 684 __le64 cap_id; 685 __le32 seq, issue_seq; 686 __le32 caps, wanted, dirty; /* latest issued/wanted/dirty */ 687 __le32 migrate_seq; 688 __le64 snap_follows; 689 __le32 snap_trace_len; 690 691 /* authlock */ 692 __le32 uid, gid, mode; 693 694 /* linklock */ 695 __le32 nlink; 696 697 /* xattrlock */ 698 __le32 xattr_len; 699 __le64 xattr_version; 700 701 /* filelock */ 702 __le64 size, max_size, truncate_size; 703 __le32 truncate_seq; 704 struct ceph_timespec mtime, atime, ctime; 705 struct ceph_file_layout_legacy layout; 706 __le32 time_warp_seq; 707 } __attribute__ ((packed)); 708 709 struct ceph_mds_cap_peer { 710 __le64 cap_id; 711 __le32 seq; 712 __le32 mseq; 713 __le32 mds; 714 __u8 flags; 715 } __attribute__ ((packed)); 716 717 /* cap release msg head */ 718 struct ceph_mds_cap_release { 719 __le32 num; /* number of cap_items that follow */ 720 } __attribute__ ((packed)); 721 722 struct ceph_mds_cap_item { 723 __le64 ino; 724 __le64 cap_id; 725 __le32 migrate_seq, seq; 726 } __attribute__ ((packed)); 727 728 #define CEPH_MDS_LEASE_REVOKE 1 /* mds -> client */ 729 #define CEPH_MDS_LEASE_RELEASE 2 /* client -> mds */ 730 #define CEPH_MDS_LEASE_RENEW 3 /* client <-> mds */ 731 #define CEPH_MDS_LEASE_REVOKE_ACK 4 /* client -> mds */ 732 733 extern const char *ceph_lease_op_name(int o); 734 735 /* lease msg header */ 736 struct ceph_mds_lease { 737 __u8 action; /* CEPH_MDS_LEASE_* */ 738 __le16 mask; /* which lease */ 739 __le64 ino; 740 __le64 first, last; /* snap range */ 741 __le32 seq; 742 __le32 duration_ms; /* duration of renewal */ 743 } __attribute__ ((packed)); 744 /* followed by a __le32+string for dname */ 745 746 /* client reconnect */ 747 struct ceph_mds_cap_reconnect { 748 __le64 cap_id; 749 __le32 wanted; 750 __le32 issued; 751 __le64 snaprealm; 752 __le64 pathbase; /* base ino for our path to this ino */ 753 __le32 flock_len; /* size of flock state blob, if any */ 754 } __attribute__ ((packed)); 755 /* followed by flock blob */ 756 757 struct ceph_mds_cap_reconnect_v1 { 758 __le64 cap_id; 759 __le32 wanted; 760 __le32 issued; 761 __le64 size; 762 struct ceph_timespec mtime, atime; 763 __le64 snaprealm; 764 __le64 pathbase; /* base ino for our path to this ino */ 765 } __attribute__ ((packed)); 766 767 struct ceph_mds_snaprealm_reconnect { 768 __le64 ino; /* snap realm base */ 769 __le64 seq; /* snap seq for this snap realm */ 770 __le64 parent; /* parent realm */ 771 } __attribute__ ((packed)); 772 773 /* 774 * snaps 775 */ 776 enum { 777 CEPH_SNAP_OP_UPDATE, /* CREATE or DESTROY */ 778 CEPH_SNAP_OP_CREATE, 779 CEPH_SNAP_OP_DESTROY, 780 CEPH_SNAP_OP_SPLIT, 781 }; 782 783 extern const char *ceph_snap_op_name(int o); 784 785 /* snap msg header */ 786 struct ceph_mds_snap_head { 787 __le32 op; /* CEPH_SNAP_OP_* */ 788 __le64 split; /* ino to split off, if any */ 789 __le32 num_split_inos; /* # inos belonging to new child realm */ 790 __le32 num_split_realms; /* # child realms udner new child realm */ 791 __le32 trace_len; /* size of snap trace blob */ 792 } __attribute__ ((packed)); 793 /* followed by split ino list, then split realms, then the trace blob */ 794 795 /* 796 * encode info about a snaprealm, as viewed by a client 797 */ 798 struct ceph_mds_snap_realm { 799 __le64 ino; /* ino */ 800 __le64 created; /* snap: when created */ 801 __le64 parent; /* ino: parent realm */ 802 __le64 parent_since; /* snap: same parent since */ 803 __le64 seq; /* snap: version */ 804 __le32 num_snaps; 805 __le32 num_prior_parent_snaps; 806 } __attribute__ ((packed)); 807 /* followed by my snap list, then prior parent snap list */ 808 809 #endif 810