4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/include/lustre/lustre_idl.h
34 * Lustre wire protocol definitions.
37 /** \defgroup lustreidl lustreidl
39 * Lustre wire protocol definitions.
41 * ALL structs passing over the wire should be declared here. Structs
42 * that are used in interfaces with userspace should go in lustre_user.h.
44 * All structs being declared here should be built from simple fixed-size
45 * types (__u8, __u16, __u32, __u64) or be built from other types or
46 * structs also declared in this file. Similarly, all flags and magic
47 * values in those structs should also be declared here. This ensures
48 * that the Lustre wire protocol is not influenced by external dependencies.
50 * The only other acceptable items in this file are VERY SIMPLE accessor
51 * functions to avoid callers grubbing inside the structures. Nothing that
52 * depends on external functions or definitions should be in here.
54 * Structs must be properly aligned to put 64-bit values on an 8-byte
55 * boundary. Any structs being added here must also be added to
56 * utils/wirecheck.c and "make newwiretest" run to regenerate the
57 * utils/wiretest.c sources. This allows us to verify that wire structs
58 * have the proper alignment/size on all architectures.
60 * DO NOT CHANGE any of the structs, flags, values declared here and used
61 * in released Lustre versions. Some structs may have padding fields that
62 * can be used. Some structs might allow addition at the end (verify this
63 * in the code to ensure that new/old clients that see this larger struct
64 * do not fail, otherwise you need to implement protocol compatibility).
69 #ifndef _LUSTRE_IDL_H_
70 #define _LUSTRE_IDL_H_
72 #include <asm/byteorder.h>
73 #include <linux/types.h>
75 #include <lnet/types.h>
76 #include <lustre/lustre_user.h> /* Defn's shared with user-space. */
77 #include <lustre_ver.h>
82 /* FOO_REQUEST_PORTAL is for incoming requests on the FOO
83 * FOO_REPLY_PORTAL is for incoming replies on the FOO
84 * FOO_BULK_PORTAL is for incoming bulk on the FOO
87 #define CONNMGR_REQUEST_PORTAL 1
88 #define CONNMGR_REPLY_PORTAL 2
89 //#define OSC_REQUEST_PORTAL 3
90 #define OSC_REPLY_PORTAL 4
91 //#define OSC_BULK_PORTAL 5
92 #define OST_IO_PORTAL 6
93 #define OST_CREATE_PORTAL 7
94 #define OST_BULK_PORTAL 8
95 //#define MDC_REQUEST_PORTAL 9
96 #define MDC_REPLY_PORTAL 10
97 //#define MDC_BULK_PORTAL 11
98 #define MDS_REQUEST_PORTAL 12
99 //#define MDS_REPLY_PORTAL 13
100 #define MDS_BULK_PORTAL 14
101 #define LDLM_CB_REQUEST_PORTAL 15
102 #define LDLM_CB_REPLY_PORTAL 16
103 #define LDLM_CANCEL_REQUEST_PORTAL 17
104 #define LDLM_CANCEL_REPLY_PORTAL 18
105 //#define PTLBD_REQUEST_PORTAL 19
106 //#define PTLBD_REPLY_PORTAL 20
107 //#define PTLBD_BULK_PORTAL 21
108 #define MDS_SETATTR_PORTAL 22
109 #define MDS_READPAGE_PORTAL 23
110 #define OUT_PORTAL 24
111 #define MGC_REPLY_PORTAL 25
112 #define MGS_REQUEST_PORTAL 26
113 #define MGS_REPLY_PORTAL 27
114 #define OST_REQUEST_PORTAL 28
115 #define FLD_REQUEST_PORTAL 29
116 #define SEQ_METADATA_PORTAL 30
117 #define SEQ_DATA_PORTAL 31
118 #define SEQ_CONTROLLER_PORTAL 32
119 #define MGS_BULK_PORTAL 33
121 /* Portal 63 is reserved for the Cray Inc DVS - nic@cray.com, roe@cray.com, n8851@cray.com */
124 #define PTL_RPC_MSG_REQUEST 4711
125 #define PTL_RPC_MSG_ERR 4712
126 #define PTL_RPC_MSG_REPLY 4713
128 /* DON'T use swabbed values of MAGIC as magic! */
129 #define LUSTRE_MSG_MAGIC_V2 0x0BD00BD3
130 #define LUSTRE_MSG_MAGIC_V2_SWABBED 0xD30BD00B
132 #define LUSTRE_MSG_MAGIC LUSTRE_MSG_MAGIC_V2
134 #define PTLRPC_MSG_VERSION 0x00000003
135 #define LUSTRE_VERSION_MASK 0xffff0000
136 #define LUSTRE_OBD_VERSION 0x00010000
137 #define LUSTRE_MDS_VERSION 0x00020000
138 #define LUSTRE_OST_VERSION 0x00030000
139 #define LUSTRE_DLM_VERSION 0x00040000
140 #define LUSTRE_LOG_VERSION 0x00050000
141 #define LUSTRE_MGS_VERSION 0x00060000
144 * Describes a range of sequence, lsr_start is included but lsr_end is
146 * Same structure is used in fld module where lsr_index field holds mdt id
149 struct lu_seq_range {
156 struct lu_seq_range_array {
159 struct lu_seq_range lsra_lsr[0];
162 #define LU_SEQ_RANGE_MDT 0x0
163 #define LU_SEQ_RANGE_OST 0x1
164 #define LU_SEQ_RANGE_ANY 0x3
166 #define LU_SEQ_RANGE_MASK 0x3
168 /** \defgroup lu_fid lu_fid
172 * Flags for lustre_mdt_attrs::lma_compat and lustre_mdt_attrs::lma_incompat.
173 * Deprecated since HSM and SOM attributes are now stored in separate on-disk
177 LMAC_HSM = 0x00000001,
178 /* LMAC_SOM = 0x00000002, obsolete since 2.8.0 */
179 LMAC_NOT_IN_OI = 0x00000004, /* the object does NOT need OI mapping */
180 LMAC_FID_ON_OST = 0x00000008, /* For OST-object, its OI mapping is
181 * under /O/<seq>/d<x>. */
185 * Masks for all features that should be supported by a Lustre version to
186 * access a specific file.
187 * This information is stored in lustre_mdt_attrs::lma_incompat.
190 LMAI_RELEASED = 0x00000001, /* file is released */
191 LMAI_AGENT = 0x00000002, /* agent inode */
192 LMAI_REMOTE_PARENT = 0x00000004, /* the parent of the object
193 is on the remote MDT */
194 LMAI_STRIPED = 0x00000008, /* striped directory inode */
195 LMAI_ORPHAN = 0x00000010, /* inode is orphan */
196 LMA_INCOMPAT_SUPP = (LMAI_AGENT | LMAI_REMOTE_PARENT | \
197 LMAI_STRIPED | LMAI_ORPHAN)
200 extern void lustre_lma_swab(struct lustre_mdt_attrs *lma);
201 extern void lustre_lma_init(struct lustre_mdt_attrs *lma,
202 const struct lu_fid *fid,
203 __u32 compat, __u32 incompat);
205 /* copytool uses a 32b bitmask field to encode archive-Ids during register
207 * archive num = 0 => all
208 * archive num from 1 to 32
210 #define LL_HSM_MAX_ARCHIVE (sizeof(__u32) * 8)
213 * HSM on-disk attributes stored in a separate xattr.
216 /** Bitfield for supported data in this structure. For future use. */
219 /** HSM flags, see hsm_flags enum below */
221 /** backend archive id associated with the file */
223 /** version associated with the last archiving, if any */
226 extern void lustre_hsm_swab(struct hsm_attrs *attrs);
232 /** LASTID file has zero OID */
233 LUSTRE_FID_LASTID_OID = 0UL,
234 /** initial fid id value */
235 LUSTRE_FID_INIT_OID = 1UL
239 * Note that reserved SEQ numbers below 12 will conflict with ldiskfs
240 * inodes in the IGIF namespace, so these reserved SEQ numbers can be
241 * used for other purposes and not risk collisions with existing inodes.
243 * Different FID Format
244 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs#NEW.0
247 FID_SEQ_OST_MDT0 = 0,
248 FID_SEQ_LLOG = 1, /* unnamed llogs */
250 FID_SEQ_UNUSED_START = 3,
251 FID_SEQ_UNUSED_END = 9,
252 FID_SEQ_LLOG_NAME = 10, /* named llogs */
255 FID_SEQ_IGIF_MAX = 0x0ffffffffULL,
256 FID_SEQ_IDIF = 0x100000000ULL,
257 FID_SEQ_IDIF_MAX = 0x1ffffffffULL,
258 /* Normal FID sequence starts from this value, i.e. 1<<33 */
259 FID_SEQ_START = 0x200000000ULL,
260 /* sequence for local pre-defined FIDs listed in local_oid */
261 FID_SEQ_LOCAL_FILE = 0x200000001ULL,
262 FID_SEQ_DOT_LUSTRE = 0x200000002ULL,
263 /* sequence is used for local named objects FIDs generated
264 * by local_object_storage library */
265 FID_SEQ_LOCAL_NAME = 0x200000003ULL,
266 /* Because current FLD will only cache the fid sequence, instead
267 * of oid on the client side, if the FID needs to be exposed to
268 * clients sides, it needs to make sure all of fids under one
269 * sequence will be located in one MDT. */
270 FID_SEQ_SPECIAL = 0x200000004ULL,
271 FID_SEQ_QUOTA = 0x200000005ULL,
272 FID_SEQ_QUOTA_GLB = 0x200000006ULL,
273 FID_SEQ_ROOT = 0x200000007ULL, /* Located on MDT0 */
274 FID_SEQ_LAYOUT_RBTREE = 0x200000008ULL,
275 /* sequence is used for update logs of cross-MDT operation */
276 FID_SEQ_UPDATE_LOG = 0x200000009ULL,
277 /* Sequence is used for the directory under which update logs
279 FID_SEQ_UPDATE_LOG_DIR = 0x20000000aULL,
280 FID_SEQ_NORMAL = 0x200000400ULL,
281 FID_SEQ_LOV_DEFAULT = 0xffffffffffffffffULL
284 #define OBIF_OID_MAX_BITS 32
285 #define OBIF_MAX_OID (1ULL << OBIF_OID_MAX_BITS)
286 #define OBIF_OID_MASK ((1ULL << OBIF_OID_MAX_BITS) - 1)
287 #define IDIF_OID_MAX_BITS 48
288 #define IDIF_MAX_OID (1ULL << IDIF_OID_MAX_BITS)
289 #define IDIF_OID_MASK ((1ULL << IDIF_OID_MAX_BITS) - 1)
291 /** OID for FID_SEQ_SPECIAL */
293 /* Big Filesystem Lock to serialize rename operations */
294 FID_OID_SPECIAL_BFL = 1UL,
297 /** OID for FID_SEQ_DOT_LUSTRE */
298 enum dot_lustre_oid {
299 FID_OID_DOT_LUSTRE = 1UL,
300 FID_OID_DOT_LUSTRE_OBF = 2UL,
301 FID_OID_DOT_LUSTRE_LPF = 3UL,
304 /** OID for FID_SEQ_ROOT */
307 FID_OID_ECHO_ROOT = 2UL,
310 struct lu_orphan_rec {
311 /* The MDT-object's FID referenced by the orphan OST-object */
312 struct lu_fid lor_fid;
317 struct lu_orphan_ent {
318 /* The orphan OST-object's FID */
319 struct lu_fid loe_key;
320 struct lu_orphan_rec loe_rec;
325 /** \defgroup lu_dir lu_dir
329 * Enumeration of possible directory entry attributes.
331 * Attributes follow directory entry header in the order they appear in this
334 enum lu_dirent_attrs {
337 LUDA_64BITHASH = 0x0004,
339 /* The following attrs are used for MDT internal only,
340 * not visible to client */
342 /* Something in the record is unknown, to be verified in further. */
343 LUDA_UNKNOWN = 0x0400,
344 /* Ignore this record, go to next directly. */
345 LUDA_IGNORE = 0x0800,
346 /* The system is upgraded, has beed or to be repaired (dryrun). */
347 LUDA_UPGRADE = 0x1000,
348 /* The dirent has been repaired, or to be repaired (dryrun). */
349 LUDA_REPAIR = 0x2000,
350 /* Only check but not repair the dirent inconsistency */
351 LUDA_VERIFY_DRYRUN = 0x4000,
352 /* Verify the dirent consistency */
353 LUDA_VERIFY = 0x8000,
356 #define LU_DIRENT_ATTRS_MASK 0xff00
359 * Layout of readdir pages, as transmitted on wire.
362 /** valid if LUDA_FID is set. */
363 struct lu_fid lde_fid;
364 /** a unique entry identifier: a hash or an offset. */
366 /** total record length, including all attributes. */
370 /** optional variable size attributes following this entry.
371 * taken from enum lu_dirent_attrs.
374 /** name is followed by the attributes indicated in ->ldp_attrs, in
375 * their natural order. After the last attribute, padding bytes are
376 * added to make ->lde_reclen a multiple of 8.
382 * Definitions of optional directory entry attributes formats.
384 * Individual attributes do not have their length encoded in a generic way. It
385 * is assumed that consumer of an attribute knows its format. This means that
386 * it is impossible to skip over an unknown attribute, except by skipping over all
387 * remaining attributes (by using ->lde_reclen), which is not too
388 * constraining, because new server versions will append new attributes at
389 * the end of an entry.
393 * Fid directory attribute: a fid of an object referenced by the entry. This
394 * will be almost always requested by the client and supplied by the server.
396 * Aligned to 8 bytes.
398 /* To have compatibility with 1.8, lets have fid in lu_dirent struct. */
403 * Aligned to 2 bytes.
410 __u64 ldp_hash_start;
414 struct lu_dirent ldp_entries[0];
417 enum lu_dirpage_flags {
419 * dirpage contains no entry.
423 * last entry's lde_hash equals ldp_hash_end.
428 static inline struct lu_dirent *lu_dirent_start(struct lu_dirpage *dp)
430 if (__le32_to_cpu(dp->ldp_flags) & LDF_EMPTY)
433 return dp->ldp_entries;
436 static inline struct lu_dirent *lu_dirent_next(struct lu_dirent *ent)
438 struct lu_dirent *next;
440 if (__le16_to_cpu(ent->lde_reclen) != 0)
441 next = ((void *)ent) + __le16_to_cpu(ent->lde_reclen);
448 static inline size_t lu_dirent_calc_size(size_t namelen, __u16 attr)
452 if (attr & LUDA_TYPE) {
453 const size_t align = sizeof(struct luda_type) - 1;
454 size = (sizeof(struct lu_dirent) + namelen + align) & ~align;
455 size += sizeof(struct luda_type);
457 size = sizeof(struct lu_dirent) + namelen;
459 return (size + 7) & ~7;
462 #define MDS_DIR_END_OFF 0xfffffffffffffffeULL
465 * MDS_READPAGE page size
467 * This is the directory page size packed in MDS_READPAGE RPC.
468 * It's different than PAGE_SIZE because the client needs to
469 * access the struct lu_dirpage header packed at the beginning of
470 * the "page" and without this there isn't any way to know find the
471 * lu_dirpage header is if client and server PAGE_SIZE differ.
473 #define LU_PAGE_SHIFT 12
474 #define LU_PAGE_SIZE (1UL << LU_PAGE_SHIFT)
475 #define LU_PAGE_MASK (~(LU_PAGE_SIZE - 1))
477 #define LU_PAGE_COUNT (1 << (PAGE_SHIFT - LU_PAGE_SHIFT))
481 struct lustre_handle {
484 #define DEAD_HANDLE_MAGIC 0xdeadbeefcafebabeULL
486 static inline bool lustre_handle_is_used(const struct lustre_handle *lh)
488 return lh->cookie != 0;
491 static inline bool lustre_handle_equal(const struct lustre_handle *lh1,
492 const struct lustre_handle *lh2)
494 return lh1->cookie == lh2->cookie;
497 static inline void lustre_handle_copy(struct lustre_handle *tgt,
498 const struct lustre_handle *src)
500 tgt->cookie = src->cookie;
503 struct lustre_handle_array {
505 struct lustre_handle handles[0];
508 /* flags for lm_flags */
509 #define MSGHDR_AT_SUPPORT 0x1
510 #define MSGHDR_CKSUM_INCOMPAT18 0x2
512 #define lustre_msg lustre_msg_v2
513 /* we depend on this structure to be 8-byte aligned */
514 /* this type is only endian-adjusted in lustre_unpack_msg() */
515 struct lustre_msg_v2 {
527 /* without gss, ptlrpc_body is put at the first buffer. */
528 #define PTLRPC_NUM_VERSIONS 4
529 struct ptlrpc_body_v3 {
530 struct lustre_handle pb_handle;
535 __u64 pb_last_xid; /* highest replied XID without lower unreplied XID */
536 __u16 pb_tag; /* virtual slot idx for multiple modifying RPCs */
539 __u64 pb_last_committed;
544 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
545 __u32 pb_service_time; /* for rep, actual service time */
548 /* VBR: pre-versions */
549 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
550 __u64 pb_mbits; /**< match bits for bulk request */
551 /* padding for future needs */
552 __u64 pb_padding64_0;
553 __u64 pb_padding64_1;
554 __u64 pb_padding64_2;
555 char pb_jobid[LUSTRE_JOBID_SIZE];
557 #define ptlrpc_body ptlrpc_body_v3
559 struct ptlrpc_body_v2 {
560 struct lustre_handle pb_handle;
565 __u64 pb_last_xid; /* highest replied XID without lower unreplied XID */
566 __u16 pb_tag; /* virtual slot idx for multiple modifying RPCs */
569 __u64 pb_last_committed;
574 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
575 __u32 pb_service_time; /* for rep, actual service time, also used for
576 net_latency of req */
579 /* VBR: pre-versions */
580 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
581 __u64 pb_mbits; /**< unused in V2 */
582 /* padding for future needs */
583 __u64 pb_padding64_0;
584 __u64 pb_padding64_1;
585 __u64 pb_padding64_2;
588 /* message body offset for lustre_msg_v2 */
589 /* ptlrpc body offset in all request/reply messages */
590 #define MSG_PTLRPC_BODY_OFF 0
592 /* normal request/reply message record offset */
593 #define REQ_REC_OFF 1
594 #define REPLY_REC_OFF 1
596 /* ldlm request message body offset */
597 #define DLM_LOCKREQ_OFF 1 /* lockreq offset */
598 #define DLM_REQ_REC_OFF 2 /* normal dlm request record offset */
600 /* ldlm intent lock message body offset */
601 #define DLM_INTENT_IT_OFF 2 /* intent lock it offset */
602 #define DLM_INTENT_REC_OFF 3 /* intent lock record offset */
604 /* ldlm reply message body offset */
605 #define DLM_LOCKREPLY_OFF 1 /* lockrep offset */
606 #define DLM_REPLY_REC_OFF 2 /* reply record offset */
608 /** only use in req->rq_{req,rep}_swab_mask */
609 #define MSG_PTLRPC_HEADER_OFF 31
611 /* Flags that are operation-specific go in the top 16 bits. */
612 #define MSG_OP_FLAG_MASK 0xffff0000
613 #define MSG_OP_FLAG_SHIFT 16
615 /* Flags that apply to all requests are in the bottom 16 bits */
616 #define MSG_GEN_FLAG_MASK 0x0000ffff
617 #define MSG_LAST_REPLAY 0x0001
618 #define MSG_RESENT 0x0002
619 #define MSG_REPLAY 0x0004
620 /* #define MSG_AT_SUPPORT 0x0008
621 * This was used in early prototypes of adaptive timeouts, and while there
622 * shouldn't be any users of that code there also isn't a need for using this
623 * bits. Defer usage until at least 1.10 to avoid potential conflict. */
624 #define MSG_DELAY_REPLAY 0x0010
625 #define MSG_VERSION_REPLAY 0x0020
626 #define MSG_REQ_REPLAY_DONE 0x0040
627 #define MSG_LOCK_REPLAY_DONE 0x0080
630 * Flags for all connect opcodes (MDS_CONNECT, OST_CONNECT)
633 #define MSG_CONNECT_RECOVERING 0x00000001
634 #define MSG_CONNECT_RECONNECT 0x00000002
635 #define MSG_CONNECT_REPLAYABLE 0x00000004
636 //#define MSG_CONNECT_PEER 0x8
637 #define MSG_CONNECT_LIBCLIENT 0x00000010
638 #define MSG_CONNECT_INITIAL 0x00000020
639 #define MSG_CONNECT_ASYNC 0x00000040
640 #define MSG_CONNECT_NEXT_VER 0x00000080 /* use next version of lustre_msg */
641 #define MSG_CONNECT_TRANSNO 0x00000100 /* report transno */
644 #define OBD_CONNECT_RDONLY 0x1ULL /*client has read-only access*/
645 #define OBD_CONNECT_INDEX 0x2ULL /*connect specific LOV idx */
646 #define OBD_CONNECT_MDS 0x4ULL /*connect from MDT to OST */
647 #define OBD_CONNECT_GRANT 0x8ULL /*OSC gets grant at connect */
648 #define OBD_CONNECT_SRVLOCK 0x10ULL /*server takes locks for cli */
649 #define OBD_CONNECT_VERSION 0x20ULL /*Lustre versions in ocd */
650 #define OBD_CONNECT_REQPORTAL 0x40ULL /*Separate non-IO req portal */
651 #define OBD_CONNECT_ACL 0x80ULL /*access control lists */
652 #define OBD_CONNECT_XATTR 0x100ULL /*client use extended attr */
653 #define OBD_CONNECT_CROW 0x200ULL /*MDS+OST create obj on write*/
654 #define OBD_CONNECT_TRUNCLOCK 0x400ULL /*locks on server for punch */
655 #define OBD_CONNECT_TRANSNO 0x800ULL /*replay sends init transno */
656 #define OBD_CONNECT_IBITS 0x1000ULL /*support for inodebits locks*/
657 #define OBD_CONNECT_BARRIER 0x2000ULL /* write barrier */
658 #define OBD_CONNECT_ATTRFID 0x4000ULL /*Server can GetAttr By Fid*/
659 #define OBD_CONNECT_NODEVOH 0x8000ULL /*No open hndl on specl nodes*/
660 #define OBD_CONNECT_RMT_CLIENT 0x10000ULL /* Remote client, never used
661 * in production. Removed in
662 * 2.9. Keep this flag to
665 #define OBD_CONNECT_RMT_CLIENT_FORCE 0x20000ULL /* Remote client by force,
666 * never used in production.
667 * Removed in 2.9. Keep this
668 * flag to avoid reusing.
670 #define OBD_CONNECT_BRW_SIZE 0x40000ULL /*Max bytes per rpc */
671 #define OBD_CONNECT_QUOTA64 0x80000ULL /*Not used since 2.4 */
672 #define OBD_CONNECT_MDS_CAPA 0x100000ULL /*MDS capability */
673 #define OBD_CONNECT_OSS_CAPA 0x200000ULL /*OSS capability */
674 #define OBD_CONNECT_CANCELSET 0x400000ULL /*Early batched cancels. */
675 #define OBD_CONNECT_SOM 0x800000ULL /*Size on MDS */
676 #define OBD_CONNECT_AT 0x1000000ULL /*client uses AT */
677 #define OBD_CONNECT_LRU_RESIZE 0x2000000ULL /*LRU resize feature. */
678 #define OBD_CONNECT_MDS_MDS 0x4000000ULL /*MDS-MDS connection */
679 #define OBD_CONNECT_REAL 0x8000000ULL /* obsolete since 2.8 */
680 #define OBD_CONNECT_CHANGE_QS 0x10000000ULL /*Not used since 2.4 */
681 #define OBD_CONNECT_CKSUM 0x20000000ULL /*support several cksum algos*/
682 #define OBD_CONNECT_FID 0x40000000ULL /*FID is supported by server */
683 #define OBD_CONNECT_VBR 0x80000000ULL /*version based recovery */
684 #define OBD_CONNECT_LOV_V3 0x100000000ULL /*client supports LOV v3 EA */
685 #define OBD_CONNECT_GRANT_SHRINK 0x200000000ULL /* support grant shrink */
686 #define OBD_CONNECT_SKIP_ORPHAN 0x400000000ULL /* don't reuse orphan objids */
687 #define OBD_CONNECT_MAX_EASIZE 0x800000000ULL /* preserved for large EA */
688 #define OBD_CONNECT_FULL20 0x1000000000ULL /* it is 2.0 client */
689 #define OBD_CONNECT_LAYOUTLOCK 0x2000000000ULL /* client uses layout lock */
690 #define OBD_CONNECT_64BITHASH 0x4000000000ULL /* client supports 64-bits
692 #define OBD_CONNECT_MAXBYTES 0x8000000000ULL /* max stripe size */
693 #define OBD_CONNECT_IMP_RECOV 0x10000000000ULL /* imp recovery support */
694 #define OBD_CONNECT_JOBSTATS 0x20000000000ULL /* jobid in ptlrpc_body */
695 #define OBD_CONNECT_UMASK 0x40000000000ULL /* create uses client umask */
696 #define OBD_CONNECT_EINPROGRESS 0x80000000000ULL /* client handles -EINPROGRESS
697 * RPC error properly */
698 #define OBD_CONNECT_GRANT_PARAM 0x100000000000ULL/* extra grant params used for
699 * finer space reservation */
700 #define OBD_CONNECT_FLOCK_OWNER 0x200000000000ULL /* for the fixed 1.8
701 * policy and 2.x server */
702 #define OBD_CONNECT_LVB_TYPE 0x400000000000ULL /* variable type of LVB */
703 #define OBD_CONNECT_NANOSEC_TIME 0x800000000000ULL /* nanosecond timestamps */
704 #define OBD_CONNECT_LIGHTWEIGHT 0x1000000000000ULL/* lightweight connection */
705 #define OBD_CONNECT_SHORTIO 0x2000000000000ULL/* short io */
706 #define OBD_CONNECT_PINGLESS 0x4000000000000ULL/* pings not required */
707 #define OBD_CONNECT_FLOCK_DEAD 0x8000000000000ULL/* improved flock deadlock detection */
708 #define OBD_CONNECT_DISP_STRIPE 0x10000000000000ULL/* create stripe disposition*/
709 #define OBD_CONNECT_OPEN_BY_FID 0x20000000000000ULL /* open by fid won't pack
711 #define OBD_CONNECT_LFSCK 0x40000000000000ULL/* support online LFSCK */
712 #define OBD_CONNECT_UNLINK_CLOSE 0x100000000000000ULL/* close file in unlink */
713 #define OBD_CONNECT_MULTIMODRPCS 0x200000000000000ULL /* support multiple modify
715 #define OBD_CONNECT_DIR_STRIPE 0x400000000000000ULL /* striped DNE dir */
716 #define OBD_CONNECT_SUBTREE 0x800000000000000ULL /* fileset mount */
717 #define OBD_CONNECT_LOCK_AHEAD 0x1000000000000000ULL /* lock ahead */
718 /** bulk matchbits is sent within ptlrpc_body */
719 #define OBD_CONNECT_BULK_MBITS 0x2000000000000000ULL
720 #define OBD_CONNECT_OBDOPACK 0x4000000000000000ULL /* compact OUT obdo */
721 #define OBD_CONNECT_FLAGS2 0x8000000000000000ULL /* second flags word */
722 /* ocd_connect_flags2 flags */
723 #define OBD_CONNECT2_FILE_SECCTX 0x1ULL /* set file security context at create */
726 * Please DO NOT add flag values here before first ensuring that this same
727 * flag value is not in use on some other branch. Please clear any such
728 * changes with senior engineers before starting to use a new flag. Then,
729 * submit a small patch against EVERY branch that ONLY adds the new flag,
730 * updates obd_connect_names[] for lprocfs_rd_connect_flags(), adds the
731 * flag to check_obd_connect_data(), and updates wiretests accordingly, so it
732 * can be approved and landed easily to reserve the flag for future use. */
734 /* The MNE_SWAB flag is overloading the MDS_MDS bit only for the MGS
735 * connection. It is a temporary bug fix for Imperative Recovery interop
736 * between 2.2 and 2.3 x86/ppc nodes, and can be removed when interop for
737 * 2.2 clients/servers is no longer needed. LU-1252/LU-1644. */
738 #define OBD_CONNECT_MNE_SWAB OBD_CONNECT_MDS_MDS
740 #define OCD_HAS_FLAG(ocd, flg) \
741 (!!((ocd)->ocd_connect_flags & OBD_CONNECT_##flg))
744 #ifdef HAVE_LRU_RESIZE_SUPPORT
745 #define LRU_RESIZE_CONNECT_FLAG OBD_CONNECT_LRU_RESIZE
747 #define LRU_RESIZE_CONNECT_FLAG 0
750 #define MDT_CONNECT_SUPPORTED (OBD_CONNECT_RDONLY | OBD_CONNECT_VERSION | \
751 OBD_CONNECT_ACL | OBD_CONNECT_XATTR | \
752 OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH | \
753 OBD_CONNECT_ATTRFID | OBD_CONNECT_CANCELSET | \
754 OBD_CONNECT_AT | OBD_CONNECT_BRW_SIZE | \
755 OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID | \
756 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_VBR | \
757 OBD_CONNECT_LOV_V3 | OBD_CONNECT_FULL20 | \
758 OBD_CONNECT_64BITHASH | OBD_CONNECT_JOBSTATS | \
759 OBD_CONNECT_EINPROGRESS | \
760 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_UMASK | \
761 OBD_CONNECT_LVB_TYPE | OBD_CONNECT_LAYOUTLOCK |\
762 OBD_CONNECT_PINGLESS | OBD_CONNECT_MAX_EASIZE |\
763 OBD_CONNECT_FLOCK_DEAD | \
764 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK | \
765 OBD_CONNECT_OPEN_BY_FID | \
766 OBD_CONNECT_DIR_STRIPE | \
767 OBD_CONNECT_BULK_MBITS | \
768 OBD_CONNECT_MULTIMODRPCS | \
769 OBD_CONNECT_SUBTREE | \
772 #define MDT_CONNECT_SUPPORTED2 OBD_CONNECT2_FILE_SECCTX
774 #define OST_CONNECT_SUPPORTED (OBD_CONNECT_SRVLOCK | OBD_CONNECT_GRANT | \
775 OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | \
776 OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_INDEX | \
777 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_CANCELSET | \
778 OBD_CONNECT_AT | LRU_RESIZE_CONNECT_FLAG | \
779 OBD_CONNECT_CKSUM | OBD_CONNECT_VBR | \
780 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN | \
781 OBD_CONNECT_GRANT_SHRINK | OBD_CONNECT_FULL20 |\
782 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES | \
783 OBD_CONNECT_MAX_EASIZE | \
784 OBD_CONNECT_EINPROGRESS | \
785 OBD_CONNECT_JOBSTATS | \
786 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LVB_TYPE|\
787 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_FID | \
788 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK | \
789 OBD_CONNECT_BULK_MBITS | \
790 OBD_CONNECT_GRANT_PARAM)
791 #define OST_CONNECT_SUPPORTED2 0
793 #define ECHO_CONNECT_SUPPORTED 0
794 #define ECHO_CONNECT_SUPPORTED2 0
796 #define MGS_CONNECT_SUPPORTED (OBD_CONNECT_VERSION | OBD_CONNECT_AT | \
797 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV | \
798 OBD_CONNECT_MNE_SWAB | OBD_CONNECT_PINGLESS |\
799 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_BARRIER)
801 #define MGS_CONNECT_SUPPORTED2 0
803 /* Features required for this version of the client to work with server */
804 #define CLIENT_CONNECT_MDT_REQD (OBD_CONNECT_IBITS | OBD_CONNECT_FID | \
807 /* This structure is used for both request and reply.
809 * If we eventually have separate connect data for different types, which we
810 * almost certainly will, then perhaps we stick a union in here. */
811 struct obd_connect_data {
812 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
813 __u32 ocd_version; /* lustre release version number */
814 __u32 ocd_grant; /* initial cache grant amount (bytes) */
815 __u32 ocd_index; /* LOV index to connect to */
816 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
817 __u64 ocd_ibits_known; /* inode bits this client understands */
818 __u8 ocd_grant_blkbits; /* log2 of the backend filesystem blocksize */
819 __u8 ocd_grant_inobits; /* log2 of the per-inode space consumption */
820 __u16 ocd_grant_tax_kb; /* extent insertion overhead, in 1K blocks */
821 __u32 ocd_grant_max_blks;/* maximum number of blocks per extent */
822 __u64 ocd_transno; /* first transno from client to be replayed */
823 __u32 ocd_group; /* MDS group on OST */
824 __u32 ocd_cksum_types; /* supported checksum algorithms */
825 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
826 __u32 ocd_instance; /* instance # of this target */
827 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
828 /* Fields after ocd_maxbytes are only accessible by the receiver
829 * if the corresponding flag in ocd_connect_flags is set. Accessing
830 * any field after ocd_maxbytes on the receiver without a valid flag
831 * may result in out-of-bound memory access and kernel oops. */
832 __u16 ocd_maxmodrpcs; /* Maximum modify RPCs in parallel */
833 __u16 padding0; /* added 2.1.0. also fix lustre_swab_connect */
834 __u32 padding1; /* added 2.1.0. also fix lustre_swab_connect */
835 __u64 ocd_connect_flags2;
836 __u64 padding3; /* added 2.1.0. also fix lustre_swab_connect */
837 __u64 padding4; /* added 2.1.0. also fix lustre_swab_connect */
838 __u64 padding5; /* added 2.1.0. also fix lustre_swab_connect */
839 __u64 padding6; /* added 2.1.0. also fix lustre_swab_connect */
840 __u64 padding7; /* added 2.1.0. also fix lustre_swab_connect */
841 __u64 padding8; /* added 2.1.0. also fix lustre_swab_connect */
842 __u64 padding9; /* added 2.1.0. also fix lustre_swab_connect */
843 __u64 paddingA; /* added 2.1.0. also fix lustre_swab_connect */
844 __u64 paddingB; /* added 2.1.0. also fix lustre_swab_connect */
845 __u64 paddingC; /* added 2.1.0. also fix lustre_swab_connect */
846 __u64 paddingD; /* added 2.1.0. also fix lustre_swab_connect */
847 __u64 paddingE; /* added 2.1.0. also fix lustre_swab_connect */
848 __u64 paddingF; /* added 2.1.0. also fix lustre_swab_connect */
851 * Please DO NOT use any fields here before first ensuring that this same
852 * field is not in use on some other branch. Please clear any such changes
853 * with senior engineers before starting to use a new field. Then, submit
854 * a small patch against EVERY branch that ONLY adds the new field along with
855 * the matching OBD_CONNECT flag, so that can be approved and landed easily to
856 * reserve the flag for future use. */
859 * Supported checksum algorithms. Up to 32 checksum types are supported.
860 * (32-bit mask stored in obd_connect_data::ocd_cksum_types)
861 * Please update DECLARE_CKSUM_NAME/OBD_CKSUM_ALL in obd.h when adding a new
862 * algorithm and also the OBD_FL_CKSUM* flags.
864 typedef enum cksum_types {
865 OBD_CKSUM_CRC32 = 0x00000001,
866 OBD_CKSUM_ADLER = 0x00000002,
867 OBD_CKSUM_CRC32C= 0x00000004,
871 * OST requests: OBDO & OBD request records
876 OST_REPLY = 0, /* reply ? */
892 OST_QUOTACHECK = 18, /* not used since 2.4 */
894 OST_QUOTA_ADJUST_QUNIT = 20, /* not used since 2.4 */
896 OST_LAST_OPC /* must be < 33 to avoid MDS_GETATTR */
898 #define OST_FIRST_OPC OST_REPLY
901 OBD_FL_INLINEDATA = 0x00000001,
902 OBD_FL_OBDMDEXISTS = 0x00000002,
903 OBD_FL_DELORPHAN = 0x00000004, /* if set in o_flags delete orphans */
904 OBD_FL_NORPC = 0x00000008, /* set in o_flags do in OSC not OST */
905 OBD_FL_IDONLY = 0x00000010, /* set in o_flags only adjust obj id*/
906 OBD_FL_RECREATE_OBJS= 0x00000020, /* recreate missing obj */
907 OBD_FL_DEBUG_CHECK = 0x00000040, /* echo client/server debug check */
908 OBD_FL_NO_USRQUOTA = 0x00000100, /* the object's owner is over quota */
909 OBD_FL_NO_GRPQUOTA = 0x00000200, /* the object's group is over quota */
910 OBD_FL_CREATE_CROW = 0x00000400, /* object should be create on write */
911 OBD_FL_SRVLOCK = 0x00000800, /* delegate DLM locking to server */
912 OBD_FL_CKSUM_CRC32 = 0x00001000, /* CRC32 checksum type */
913 OBD_FL_CKSUM_ADLER = 0x00002000, /* ADLER checksum type */
914 OBD_FL_CKSUM_CRC32C = 0x00004000, /* CRC32C checksum type */
915 OBD_FL_CKSUM_RSVD2 = 0x00008000, /* for future cksum types */
916 OBD_FL_CKSUM_RSVD3 = 0x00010000, /* for future cksum types */
917 OBD_FL_SHRINK_GRANT = 0x00020000, /* object shrink the grant */
918 OBD_FL_MMAP = 0x00040000, /* object is mmapped on the client.
919 * XXX: obsoleted - reserved for old
920 * clients prior than 2.2 */
921 OBD_FL_RECOV_RESEND = 0x00080000, /* recoverable resent */
922 OBD_FL_NOSPC_BLK = 0x00100000, /* no more block space on OST */
923 OBD_FL_FLUSH = 0x00200000, /* flush pages on the OST */
924 OBD_FL_SHORT_IO = 0x00400000, /* short io request */
926 /* Note that while these checksum values are currently separate bits,
927 * in 2.x we can actually allow all values from 1-31 if we wanted. */
928 OBD_FL_CKSUM_ALL = OBD_FL_CKSUM_CRC32 | OBD_FL_CKSUM_ADLER |
931 /* mask for local-only flag, which won't be sent over network */
932 OBD_FL_LOCAL_MASK = 0xF0000000,
936 * All LOV EA magics should have the same postfix, if some new version
937 * Lustre instroduces new LOV EA magic, then when down-grade to an old
938 * Lustre, even though the old version system does not recognizes such
939 * new magic, it still can distinguish the corrupted cases by checking
940 * the magic's postfix.
942 #define LOV_MAGIC_MAGIC 0x0BD0
943 #define LOV_MAGIC_MASK 0xFFFF
945 #define LOV_MAGIC_V1 (0x0BD10000 | LOV_MAGIC_MAGIC)
946 #define LOV_MAGIC_JOIN_V1 (0x0BD20000 | LOV_MAGIC_MAGIC)
947 #define LOV_MAGIC_V3 (0x0BD30000 | LOV_MAGIC_MAGIC)
948 #define LOV_MAGIC_MIGRATE (0x0BD40000 | LOV_MAGIC_MAGIC)
949 /* reserved for specifying OSTs */
950 #define LOV_MAGIC_SPECIFIC (0x0BD50000 | LOV_MAGIC_MAGIC)
951 #define LOV_MAGIC LOV_MAGIC_V1
952 #define LOV_MAGIC_COMP_V1 (0x0BD60000 | LOV_MAGIC_MAGIC)
955 * magic for fully defined striping
956 * the idea is that we should have different magics for striping "hints"
957 * (struct lov_user_md_v[13]) and defined ready-to-use striping (struct
958 * lov_mds_md_v[13]). at the moment the magics are used in wire protocol,
959 * we can't just change it w/o long way preparation, but we still need a
960 * mechanism to allow LOD to differentiate hint versus ready striping.
961 * so, at the moment we do a trick: MDT knows what to expect from request
962 * depending on the case (replay uses ready striping, non-replay req uses
963 * hints), so MDT replaces magic with appropriate one and now LOD can
964 * easily understand what's inside -bzzz
966 * those *_DEF magics are only used on server side internally, they
967 * won't be put on wire or disk.
969 #define LOV_MAGIC_DEF 0x10000000
970 #define LOV_MAGIC_V1_DEF (LOV_MAGIC_DEF | LOV_MAGIC_V1)
971 #define LOV_MAGIC_V3_DEF (LOV_MAGIC_DEF | LOV_MAGIC_V3)
972 #define LOV_MAGIC_COMP_V1_DEF (LOV_MAGIC_DEF | LOV_MAGIC_COMP_V1)
974 #define lov_pattern(pattern) (pattern & ~LOV_PATTERN_F_MASK)
975 #define lov_pattern_flags(pattern) (pattern & LOV_PATTERN_F_MASK)
977 #define lov_ost_data lov_ost_data_v1
978 struct lov_ost_data_v1 { /* per-stripe data structure (little-endian)*/
979 struct ost_id l_ost_oi; /* OST object ID */
980 __u32 l_ost_gen; /* generation of this l_ost_idx */
981 __u32 l_ost_idx; /* OST index in LOV (lov_tgt_desc->tgts) */
984 #define lov_mds_md lov_mds_md_v1
985 struct lov_mds_md_v1 { /* LOV EA mds/wire data (little-endian) */
986 __u32 lmm_magic; /* magic number = LOV_MAGIC_V1 */
987 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
988 struct ost_id lmm_oi; /* LOV object ID */
989 __u32 lmm_stripe_size; /* size of stripe in bytes */
990 /* lmm_stripe_count used to be __u32 */
991 __u16 lmm_stripe_count; /* num stripes in use for this object */
992 __u16 lmm_layout_gen; /* layout generation number */
993 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
996 #define MAX_MD_SIZE (sizeof(struct lov_mds_md) + 4 * sizeof(struct lov_ost_data))
997 #define MIN_MD_SIZE (sizeof(struct lov_mds_md) + 1 * sizeof(struct lov_ost_data))
999 /* This is the default MDT reply size allocated, should the striping be bigger,
1000 * it will be reallocated in mdt_fix_reply.
1001 * 100 stripes is a bit less than 2.5k of data */
1002 #define DEF_REP_MD_SIZE (sizeof(struct lov_mds_md) + \
1003 100 * sizeof(struct lov_ost_data))
1005 #define XATTR_NAME_ACL_ACCESS "system.posix_acl_access"
1006 #define XATTR_NAME_ACL_DEFAULT "system.posix_acl_default"
1007 #define XATTR_USER_PREFIX "user."
1008 #define XATTR_TRUSTED_PREFIX "trusted."
1009 #define XATTR_SECURITY_PREFIX "security."
1011 #define XATTR_NAME_LOV "trusted.lov"
1012 #define XATTR_NAME_LMA "trusted.lma"
1013 #define XATTR_NAME_LMV "trusted.lmv"
1014 #define XATTR_NAME_DEFAULT_LMV "trusted.dmv"
1015 #define XATTR_NAME_LINK "trusted.link"
1016 #define XATTR_NAME_FID "trusted.fid"
1017 #define XATTR_NAME_VERSION "trusted.version"
1018 #define XATTR_NAME_SOM "trusted.som"
1019 #define XATTR_NAME_HSM "trusted.hsm"
1020 #define XATTR_NAME_LFSCK_BITMAP "trusted.lfsck_bitmap"
1021 #define XATTR_NAME_DUMMY "trusted.dummy"
1023 #define XATTR_NAME_LFSCK_NAMESPACE "trusted.lfsck_ns"
1024 #define XATTR_NAME_MAX_LEN 32 /* increase this, if there is longer name. */
1026 struct lov_mds_md_v3 { /* LOV EA mds/wire data (little-endian) */
1027 __u32 lmm_magic; /* magic number = LOV_MAGIC_V3 */
1028 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1029 struct ost_id lmm_oi; /* LOV object ID */
1030 __u32 lmm_stripe_size; /* size of stripe in bytes */
1031 /* lmm_stripe_count used to be __u32 */
1032 __u16 lmm_stripe_count; /* num stripes in use for this object */
1033 __u16 lmm_layout_gen; /* layout generation number */
1034 char lmm_pool_name[LOV_MAXPOOLNAME + 1]; /* must be 32bit aligned */
1035 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1038 static inline __u32 lov_mds_md_size(__u16 stripes, __u32 lmm_magic)
1040 if (lmm_magic == LOV_MAGIC_V3)
1041 return sizeof(struct lov_mds_md_v3) +
1042 stripes * sizeof(struct lov_ost_data_v1);
1044 return sizeof(struct lov_mds_md_v1) +
1045 stripes * sizeof(struct lov_ost_data_v1);
1049 lov_mds_md_max_stripe_count(size_t buf_size, __u32 lmm_magic)
1051 switch (lmm_magic) {
1052 case LOV_MAGIC_V1: {
1053 struct lov_mds_md_v1 lmm;
1055 if (buf_size < sizeof(lmm))
1058 return (buf_size - sizeof(lmm)) / sizeof(lmm.lmm_objects[0]);
1060 case LOV_MAGIC_V3: {
1061 struct lov_mds_md_v3 lmm;
1063 if (buf_size < sizeof(lmm))
1066 return (buf_size - sizeof(lmm)) / sizeof(lmm.lmm_objects[0]);
1073 #define OBD_MD_FLID (0x00000001ULL) /* object ID */
1074 #define OBD_MD_FLATIME (0x00000002ULL) /* access time */
1075 #define OBD_MD_FLMTIME (0x00000004ULL) /* data modification time */
1076 #define OBD_MD_FLCTIME (0x00000008ULL) /* change time */
1077 #define OBD_MD_FLSIZE (0x00000010ULL) /* size */
1078 #define OBD_MD_FLBLOCKS (0x00000020ULL) /* allocated blocks count */
1079 #define OBD_MD_FLBLKSZ (0x00000040ULL) /* block size */
1080 #define OBD_MD_FLMODE (0x00000080ULL) /* access bits (mode & ~S_IFMT) */
1081 #define OBD_MD_FLTYPE (0x00000100ULL) /* object type (mode & S_IFMT) */
1082 #define OBD_MD_FLUID (0x00000200ULL) /* user ID */
1083 #define OBD_MD_FLGID (0x00000400ULL) /* group ID */
1084 #define OBD_MD_FLFLAGS (0x00000800ULL) /* flags word */
1085 #define OBD_MD_FLNLINK (0x00002000ULL) /* link count */
1086 #define OBD_MD_FLGENER (0x00004000ULL) /* generation number */
1087 /*#define OBD_MD_FLINLINE (0x00008000ULL) inline data. used until 1.6.5 */
1088 #define OBD_MD_FLRDEV (0x00010000ULL) /* device number */
1089 #define OBD_MD_FLEASIZE (0x00020000ULL) /* extended attribute data */
1090 #define OBD_MD_LINKNAME (0x00040000ULL) /* symbolic link target */
1091 #define OBD_MD_FLHANDLE (0x00080000ULL) /* file/lock handle */
1092 #define OBD_MD_FLCKSUM (0x00100000ULL) /* bulk data checksum */
1093 #define OBD_MD_FLQOS (0x00200000ULL) /* quality of service stats */
1094 /*#define OBD_MD_FLOSCOPQ (0x00400000ULL) osc opaque data, never used */
1095 /* OBD_MD_FLCOOKIE (0x00800000ULL) obsolete in 2.8 */
1096 #define OBD_MD_FLGROUP (0x01000000ULL) /* group */
1097 #define OBD_MD_FLFID (0x02000000ULL) /* ->ost write inline fid */
1098 #define OBD_MD_FLEPOCH (0x04000000ULL) /* ->ost write with ioepoch */
1099 /* ->mds if epoch opens or closes */
1100 #define OBD_MD_FLGRANT (0x08000000ULL) /* ost preallocation space grant */
1101 #define OBD_MD_FLDIREA (0x10000000ULL) /* dir's extended attribute data */
1102 #define OBD_MD_FLUSRQUOTA (0x20000000ULL) /* over quota flags sent from ost */
1103 #define OBD_MD_FLGRPQUOTA (0x40000000ULL) /* over quota flags sent from ost */
1104 #define OBD_MD_FLMODEASIZE (0x80000000ULL) /* EA size will be changed */
1106 #define OBD_MD_MDS (0x0000000100000000ULL) /* where an inode lives on */
1107 #define OBD_MD_REINT (0x0000000200000000ULL) /* reintegrate oa */
1108 #define OBD_MD_MEA (0x0000000400000000ULL) /* CMD split EA */
1109 #define OBD_MD_TSTATE (0x0000000800000000ULL) /* transient state field */
1111 #define OBD_MD_FLXATTR (0x0000001000000000ULL) /* xattr */
1112 #define OBD_MD_FLXATTRLS (0x0000002000000000ULL) /* xattr list */
1113 #define OBD_MD_FLXATTRRM (0x0000004000000000ULL) /* xattr remove */
1114 #define OBD_MD_FLACL (0x0000008000000000ULL) /* ACL */
1115 /* OBD_MD_FLRMTPERM (0x0000010000000000ULL) remote perm, obsolete */
1116 #define OBD_MD_FLMDSCAPA (0x0000020000000000ULL) /* MDS capability */
1117 #define OBD_MD_FLOSSCAPA (0x0000040000000000ULL) /* OSS capability */
1118 #define OBD_MD_FLCKSPLIT (0x0000080000000000ULL) /* Check split on server */
1119 #define OBD_MD_FLCROSSREF (0x0000100000000000ULL) /* Cross-ref case */
1120 #define OBD_MD_FLGETATTRLOCK (0x0000200000000000ULL) /* Get IOEpoch attributes
1121 * under lock; for xattr
1122 * requests means the
1123 * client holds the lock */
1124 #define OBD_MD_FLOBJCOUNT (0x0000400000000000ULL) /* for multiple destroy */
1126 #define OBD_MD_FLDATAVERSION (0x0010000000000000ULL) /* iversion sum */
1127 #define OBD_MD_CLOSE_INTENT_EXECED (0x0020000000000000ULL) /* close intent
1130 #define OBD_MD_DEFAULT_MEA (0x0040000000000000ULL) /* default MEA */
1132 #define OBD_MD_FLGETATTR (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLMTIME | \
1133 OBD_MD_FLCTIME | OBD_MD_FLSIZE | OBD_MD_FLBLKSZ | \
1134 OBD_MD_FLMODE | OBD_MD_FLTYPE | OBD_MD_FLUID | \
1135 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLNLINK | \
1136 OBD_MD_FLGENER | OBD_MD_FLRDEV | OBD_MD_FLGROUP)
1138 #define OBD_MD_FLXATTRALL (OBD_MD_FLXATTR | OBD_MD_FLXATTRLS)
1140 /* don't forget obdo_fid which is way down at the bottom so it can
1141 * come after the definition of llog_cookie */
1145 HSS_CLEARMASK = 0x02,
1146 HSS_ARCHIVE_ID = 0x04,
1149 struct hsm_state_set {
1151 __u32 hss_archive_id;
1153 __u64 hss_clearmask;
1156 /* ost_body.data values for OST_BRW */
1158 #define OBD_BRW_READ 0x01
1159 #define OBD_BRW_WRITE 0x02
1160 #define OBD_BRW_RWMASK (OBD_BRW_READ | OBD_BRW_WRITE)
1161 #define OBD_BRW_SYNC 0x08 /* this page is a part of synchronous
1162 * transfer and is not accounted in
1164 #define OBD_BRW_CHECK 0x10
1165 #define OBD_BRW_FROM_GRANT 0x20 /* the osc manages this under llite */
1166 #define OBD_BRW_GRANTED 0x40 /* the ost manages this */
1167 #define OBD_BRW_NOCACHE 0x80 /* this page is a part of non-cached IO */
1168 #define OBD_BRW_NOQUOTA 0x100
1169 #define OBD_BRW_SRVLOCK 0x200 /* Client holds no lock over this page */
1170 #define OBD_BRW_ASYNC 0x400 /* Server may delay commit to disk */
1171 #define OBD_BRW_MEMALLOC 0x800 /* Client runs in the "kswapd" context */
1172 #define OBD_BRW_OVER_USRQUOTA 0x1000 /* Running out of user quota */
1173 #define OBD_BRW_OVER_GRPQUOTA 0x2000 /* Running out of group quota */
1174 #define OBD_BRW_SOFT_SYNC 0x4000 /* This flag notifies the server
1175 * that the client is running low on
1176 * space for unstable pages; asking
1177 * it to sync quickly */
1179 #define OBD_OBJECT_EOF LUSTRE_EOF
1181 #define OST_MIN_PRECREATE 32
1182 #define OST_MAX_PRECREATE 20000
1185 struct ost_id ioo_oid; /* object ID, if multi-obj BRW */
1186 __u32 ioo_max_brw; /* low 16 bits were o_mode before 2.4,
1187 * now (PTLRPC_BULK_OPS_COUNT - 1) in
1188 * high 16 bits in 2.4 and later */
1189 __u32 ioo_bufcnt; /* number of niobufs for this object */
1192 /* NOTE: IOOBJ_MAX_BRW_BITS defines the _offset_ of the max_brw field in
1193 * ioo_max_brw, NOT the maximum number of bits in PTLRPC_BULK_OPS_BITS.
1194 * That said, ioo_max_brw is a 32-bit field so the limit is also 16 bits. */
1195 #define IOOBJ_MAX_BRW_BITS 16
1196 #define ioobj_max_brw_get(ioo) (((ioo)->ioo_max_brw >> IOOBJ_MAX_BRW_BITS) + 1)
1197 #define ioobj_max_brw_set(ioo, num) \
1198 do { (ioo)->ioo_max_brw = ((num) - 1) << IOOBJ_MAX_BRW_BITS; } while (0)
1200 /* multiple of 8 bytes => can array */
1201 struct niobuf_remote {
1207 /* lock value block communicated between the filter and llite */
1209 /* OST_LVB_ERR_INIT is needed because the return code in rc is
1210 * negative, i.e. because ((MASK + rc) & MASK) != MASK. */
1211 #define OST_LVB_ERR_INIT 0xffbadbad80000000ULL
1212 #define OST_LVB_ERR_MASK 0xffbadbad00000000ULL
1213 #define OST_LVB_IS_ERR(blocks) \
1214 ((blocks & OST_LVB_ERR_MASK) == OST_LVB_ERR_MASK)
1215 #define OST_LVB_SET_ERR(blocks, rc) \
1216 do { blocks = OST_LVB_ERR_INIT + rc; } while (0)
1217 #define OST_LVB_GET_ERR(blocks) (int)(blocks - OST_LVB_ERR_INIT)
1240 * lquota data structures
1243 #ifndef QUOTABLOCK_BITS
1244 # define QUOTABLOCK_BITS LUSTRE_QUOTABLOCK_BITS
1247 #ifndef QUOTABLOCK_SIZE
1248 # define QUOTABLOCK_SIZE LUSTRE_QUOTABLOCK_SIZE
1252 # define toqb lustre_stoqb
1255 /* The lquota_id structure is an union of all the possible identifier types that
1256 * can be used with quota, this includes:
1259 * - a FID which can be used for per-directory quota in the future */
1261 struct lu_fid qid_fid; /* FID for per-directory quota */
1262 __u64 qid_uid; /* user identifier */
1263 __u64 qid_gid; /* group identifier */
1266 /* quotactl management */
1267 struct obd_quotactl {
1269 __u32 qc_type; /* see Q_* flag below */
1272 struct obd_dqinfo qc_dqinfo;
1273 struct obd_dqblk qc_dqblk;
1276 #define Q_COPY(out, in, member) (out)->member = (in)->member
1278 #define QCTL_COPY(out, in) \
1280 Q_COPY(out, in, qc_cmd); \
1281 Q_COPY(out, in, qc_type); \
1282 Q_COPY(out, in, qc_id); \
1283 Q_COPY(out, in, qc_stat); \
1284 Q_COPY(out, in, qc_dqinfo); \
1285 Q_COPY(out, in, qc_dqblk); \
1288 /* Body of quota request used for quota acquire/release RPCs between quota
1289 * master (aka QMT) and slaves (ak QSD). */
1291 struct lu_fid qb_fid; /* FID of global index packing the pool ID
1292 * and type (data or metadata) as well as
1293 * the quota type (user or group). */
1294 union lquota_id qb_id; /* uid or gid or directory FID */
1295 __u32 qb_flags; /* see below */
1297 __u64 qb_count; /* acquire/release count (kbytes/inodes) */
1298 __u64 qb_usage; /* current slave usage (kbytes/inodes) */
1299 __u64 qb_slv_ver; /* slave index file version */
1300 struct lustre_handle qb_lockh; /* per-ID lock handle */
1301 struct lustre_handle qb_glb_lockh; /* global lock handle */
1302 __u64 qb_padding1[4];
1305 /* When the quota_body is used in the reply of quota global intent
1306 * lock (IT_QUOTA_CONN) reply, qb_fid contains slave index file FID. */
1307 #define qb_slv_fid qb_fid
1308 /* qb_usage is the current qunit (in kbytes/inodes) when quota_body is used in
1310 #define qb_qunit qb_usage
1312 #define QUOTA_DQACQ_FL_ACQ 0x1 /* acquire quota */
1313 #define QUOTA_DQACQ_FL_PREACQ 0x2 /* pre-acquire */
1314 #define QUOTA_DQACQ_FL_REL 0x4 /* release quota */
1315 #define QUOTA_DQACQ_FL_REPORT 0x8 /* report usage */
1317 /* Quota types currently supported */
1319 LQUOTA_TYPE_USR = 0x00, /* maps to USRQUOTA */
1320 LQUOTA_TYPE_GRP = 0x01, /* maps to GRPQUOTA */
1324 /* There are 2 different resource types on which a quota limit can be enforced:
1325 * - inodes on the MDTs
1326 * - blocks on the OSTs */
1328 LQUOTA_RES_MD = 0x01, /* skip 0 to avoid null oid in FID */
1329 LQUOTA_RES_DT = 0x02,
1331 LQUOTA_FIRST_RES = LQUOTA_RES_MD
1333 #define LQUOTA_NR_RES (LQUOTA_LAST_RES - LQUOTA_FIRST_RES + 1)
1336 * Space accounting support
1337 * Format of an accounting record, providing disk usage information for a given
1340 struct lquota_acct_rec { /* 16 bytes */
1341 __u64 bspace; /* current space in use */
1342 __u64 ispace; /* current # inodes in use */
1346 * Global quota index support
1347 * Format of a global record, providing global quota settings for a given quota
1350 struct lquota_glb_rec { /* 32 bytes */
1351 __u64 qbr_hardlimit; /* quota hard limit, in #inodes or kbytes */
1352 __u64 qbr_softlimit; /* quota soft limit, in #inodes or kbytes */
1353 __u64 qbr_time; /* grace time, in seconds */
1354 __u64 qbr_granted; /* how much is granted to slaves, in #inodes or
1359 * Slave index support
1360 * Format of a slave record, recording how much space is granted to a given
1363 struct lquota_slv_rec { /* 8 bytes */
1364 __u64 qsr_granted; /* space granted to the slave for the key=ID,
1365 * in #inodes or kbytes */
1368 /* Data structures associated with the quota locks */
1370 /* Glimpse descriptor used for the index & per-ID quota locks */
1371 struct ldlm_gl_lquota_desc {
1372 union lquota_id gl_id; /* quota ID subject to the glimpse */
1373 __u64 gl_flags; /* see LQUOTA_FL* below */
1374 __u64 gl_ver; /* new index version */
1375 __u64 gl_hardlimit; /* new hardlimit or qunit value */
1376 __u64 gl_softlimit; /* new softlimit */
1380 #define gl_qunit gl_hardlimit /* current qunit value used when
1381 * glimpsing per-ID quota locks */
1383 /* quota glimpse flags */
1384 #define LQUOTA_FL_EDQUOT 0x1 /* user/group out of quota space on QMT */
1386 /* LVB used with quota (global and per-ID) locks */
1388 __u64 lvb_flags; /* see LQUOTA_FL* above */
1389 __u64 lvb_id_may_rel; /* space that might be released later */
1390 __u64 lvb_id_rel; /* space released by the slave for this ID */
1391 __u64 lvb_id_qunit; /* current qunit value */
1395 /* LVB used with global quota lock */
1396 #define lvb_glb_ver lvb_id_may_rel /* current version of the global index */
1404 #define QUOTA_FIRST_OPC QUOTA_DQACQ
1413 MDS_GETATTR_NAME = 34,
1418 MDS_DISCONNECT = 39,
1421 MDS_PIN = 42, /* obsolete, never used in a release */
1422 MDS_UNPIN = 43, /* obsolete, never used in a release */
1424 MDS_DONE_WRITING = 45, /* obsolete since 2.8.0 */
1426 MDS_QUOTACHECK = 47, /* not used since 2.4 */
1429 MDS_SETXATTR = 50, /* obsolete, now it's MDS_REINT op */
1431 MDS_IS_SUBDIR = 52, /* obsolete, never used in a release */
1433 MDS_HSM_STATE_GET = 54,
1434 MDS_HSM_STATE_SET = 55,
1435 MDS_HSM_ACTION = 56,
1436 MDS_HSM_PROGRESS = 57,
1437 MDS_HSM_REQUEST = 58,
1438 MDS_HSM_CT_REGISTER = 59,
1439 MDS_HSM_CT_UNREGISTER = 60,
1440 MDS_SWAP_LAYOUTS = 61,
1444 #define MDS_FIRST_OPC MDS_GETATTR
1447 /* opcodes for object update */
1453 #define OUT_UPDATE_FIRST_OPC OUT_UPDATE
1470 } mds_reint_t, mdt_reint_t;
1472 /* the disposition of the intent outlines what was executed */
1473 #define DISP_IT_EXECD 0x00000001
1474 #define DISP_LOOKUP_EXECD 0x00000002
1475 #define DISP_LOOKUP_NEG 0x00000004
1476 #define DISP_LOOKUP_POS 0x00000008
1477 #define DISP_OPEN_CREATE 0x00000010
1478 #define DISP_OPEN_OPEN 0x00000020
1479 #define DISP_ENQ_COMPLETE 0x00400000 /* obsolete and unused */
1480 #define DISP_ENQ_OPEN_REF 0x00800000
1481 #define DISP_ENQ_CREATE_REF 0x01000000
1482 #define DISP_OPEN_LOCK 0x02000000
1483 #define DISP_OPEN_LEASE 0x04000000
1484 #define DISP_OPEN_STRIPE 0x08000000
1485 #define DISP_OPEN_DENY 0x10000000
1487 /* INODE LOCK PARTS */
1488 #define MDS_INODELOCK_LOOKUP 0x000001 /* For namespace, dentry etc, and also
1489 * was used to protect permission (mode,
1490 * owner, group etc) before 2.4. */
1491 #define MDS_INODELOCK_UPDATE 0x000002 /* size, links, timestamps */
1492 #define MDS_INODELOCK_OPEN 0x000004 /* For opened files */
1493 #define MDS_INODELOCK_LAYOUT 0x000008 /* for layout */
1495 /* The PERM bit is added int 2.4, and it is used to protect permission(mode,
1496 * owner, group, acl etc), so to separate the permission from LOOKUP lock.
1497 * Because for remote directories(in DNE), these locks will be granted by
1498 * different MDTs(different ldlm namespace).
1500 * For local directory, MDT will always grant UPDATE_LOCK|PERM_LOCK together.
1501 * For Remote directory, the master MDT, where the remote directory is, will
1502 * grant UPDATE_LOCK|PERM_LOCK, and the remote MDT, where the name entry is,
1503 * will grant LOOKUP_LOCK. */
1504 #define MDS_INODELOCK_PERM 0x000010
1505 #define MDS_INODELOCK_XATTR 0x000020 /* extended attributes */
1507 #define MDS_INODELOCK_MAXSHIFT 5
1508 /* This FULL lock is useful to take on unlink sort of operations */
1509 #define MDS_INODELOCK_FULL ((1<<(MDS_INODELOCK_MAXSHIFT+1))-1)
1511 /* NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
1512 * but was moved into name[1] along with the OID to avoid consuming the
1513 * name[2,3] fields that need to be used for the quota id (also a FID). */
1515 LUSTRE_RES_ID_SEQ_OFF = 0,
1516 LUSTRE_RES_ID_VER_OID_OFF = 1,
1517 LUSTRE_RES_ID_WAS_VER_OFF = 2, /* see note above */
1518 LUSTRE_RES_ID_QUOTA_SEQ_OFF = 2,
1519 LUSTRE_RES_ID_QUOTA_VER_OID_OFF = 3,
1520 LUSTRE_RES_ID_HSH_OFF = 3
1523 #define MDS_STATUS_CONN 1
1524 #define MDS_STATUS_LOV 2
1527 /* these should be identical to their EXT4_*_FL counterparts, they are
1528 * redefined here only to avoid dragging in fs/ext4/ext4.h */
1529 LUSTRE_SYNC_FL = 0x00000008, /* Synchronous updates */
1530 LUSTRE_IMMUTABLE_FL = 0x00000010, /* Immutable file */
1531 LUSTRE_APPEND_FL = 0x00000020, /* writes to file may only append */
1532 LUSTRE_NODUMP_FL = 0x00000040, /* do not dump file */
1533 LUSTRE_NOATIME_FL = 0x00000080, /* do not update atime */
1534 LUSTRE_INDEX_FL = 0x00001000, /* hash-indexed directory */
1535 LUSTRE_DIRSYNC_FL = 0x00010000, /* dirsync behaviour (dir only) */
1536 LUSTRE_TOPDIR_FL = 0x00020000, /* Top of directory hierarchies*/
1537 LUSTRE_DIRECTIO_FL = 0x00100000, /* Use direct i/o */
1538 LUSTRE_INLINE_DATA_FL = 0x10000000, /* Inode has inline data. */
1540 /* These flags will not be identical to any EXT4_*_FL counterparts,
1541 * and only reserved for lustre purpose. Note: these flags might
1542 * be conflict with some of EXT4 flags, so
1543 * 1. these conflict flags needs to be removed when the flag is
1544 * wired by la_flags see osd_attr_get().
1545 * 2. If these flags needs to be stored into inode, they will be
1546 * stored in LMA. see LMAI_XXXX */
1547 LUSTRE_ORPHAN_FL = 0x00002000,
1549 LUSTRE_LMA_FL_MASKS = LUSTRE_ORPHAN_FL,
1553 /* Convert wire LUSTRE_*_FL to corresponding client local VFS S_* values
1554 * for the client inode i_flags. The LUSTRE_*_FL are the Lustre wire
1555 * protocol equivalents of LDISKFS_*_FL values stored on disk, while
1556 * the S_* flags are kernel-internal values that change between kernel
1557 * versions. These flags are set/cleared via FSFILT_IOC_{GET,SET}_FLAGS.
1558 * See b=16526 for a full history. */
1559 static inline int ll_ext_to_inode_flags(int flags)
1561 return (((flags & LUSTRE_SYNC_FL) ? S_SYNC : 0) |
1562 ((flags & LUSTRE_NOATIME_FL) ? S_NOATIME : 0) |
1563 ((flags & LUSTRE_APPEND_FL) ? S_APPEND : 0) |
1564 #if defined(S_DIRSYNC)
1565 ((flags & LUSTRE_DIRSYNC_FL) ? S_DIRSYNC : 0) |
1567 ((flags & LUSTRE_IMMUTABLE_FL) ? S_IMMUTABLE : 0));
1570 static inline int ll_inode_to_ext_flags(int iflags)
1572 return (((iflags & S_SYNC) ? LUSTRE_SYNC_FL : 0) |
1573 ((iflags & S_NOATIME) ? LUSTRE_NOATIME_FL : 0) |
1574 ((iflags & S_APPEND) ? LUSTRE_APPEND_FL : 0) |
1575 #if defined(S_DIRSYNC)
1576 ((iflags & S_DIRSYNC) ? LUSTRE_DIRSYNC_FL : 0) |
1578 ((iflags & S_IMMUTABLE) ? LUSTRE_IMMUTABLE_FL : 0));
1582 /* 64 possible states */
1583 enum md_transient_state {
1584 MS_RESTORE = (1 << 0), /* restore is running */
1588 struct lu_fid mbo_fid1;
1589 struct lu_fid mbo_fid2;
1590 struct lustre_handle mbo_handle;
1592 __u64 mbo_size; /* Offset, in the case of MDS_READPAGE */
1596 __u64 mbo_blocks; /* XID, in the case of MDS_READPAGE */
1598 __u64 mbo_t_state; /* transient file state defined in
1599 * enum md_transient_state
1600 * was "ino" until 2.4.0 */
1603 __u32 mbo_capability;
1607 __u32 mbo_flags; /* LUSTRE_*_FL file attributes */
1609 __u32 mbo_nlink; /* #bytes to read in the case of MDS_READPAGE */
1610 __u32 mbo_unused2; /* was "generation" until 2.4.0 */
1612 __u32 mbo_eadatasize;
1614 __u32 mbo_max_mdsize;
1615 __u32 mbo_unused3; /* was max_cookiesize until 2.8 */
1616 __u32 mbo_uid_h; /* high 32-bits of uid, for FUID */
1617 __u32 mbo_gid_h; /* high 32-bits of gid, for FUID */
1618 __u32 mbo_padding_5; /* also fix lustre_swab_mdt_body */
1619 __u64 mbo_padding_6;
1620 __u64 mbo_padding_7;
1621 __u64 mbo_padding_8;
1622 __u64 mbo_padding_9;
1623 __u64 mbo_padding_10;
1626 struct mdt_ioepoch {
1627 struct lustre_handle mio_handle;
1628 __u64 mio_unused1; /* was ioepoch */
1629 __u32 mio_unused2; /* was flags */
1633 /* permissions for md_perm.mp_perm */
1635 CFS_SETUID_PERM = 0x01,
1636 CFS_SETGID_PERM = 0x02,
1637 CFS_SETGRP_PERM = 0x04,
1640 struct mdt_rec_setattr {
1650 __u32 sa_padding_1_h;
1651 struct lu_fid sa_fid;
1660 __u32 sa_attr_flags;
1662 __u32 sa_bias; /* some operation flags */
1669 * Attribute flags used in mdt_rec_setattr::sa_valid.
1670 * The kernel's #defines for ATTR_* should not be used over the network
1671 * since the client and MDS may run different kernels (see bug 13828)
1672 * Therefore, we should only use MDS_ATTR_* attributes for sa_valid.
1674 #define MDS_ATTR_MODE 0x1ULL /* = 1 */
1675 #define MDS_ATTR_UID 0x2ULL /* = 2 */
1676 #define MDS_ATTR_GID 0x4ULL /* = 4 */
1677 #define MDS_ATTR_SIZE 0x8ULL /* = 8 */
1678 #define MDS_ATTR_ATIME 0x10ULL /* = 16 */
1679 #define MDS_ATTR_MTIME 0x20ULL /* = 32 */
1680 #define MDS_ATTR_CTIME 0x40ULL /* = 64 */
1681 #define MDS_ATTR_ATIME_SET 0x80ULL /* = 128 */
1682 #define MDS_ATTR_MTIME_SET 0x100ULL /* = 256 */
1683 #define MDS_ATTR_FORCE 0x200ULL /* = 512, Not a change, but a change it */
1684 #define MDS_ATTR_ATTR_FLAG 0x400ULL /* = 1024 */
1685 #define MDS_ATTR_KILL_SUID 0x800ULL /* = 2048 */
1686 #define MDS_ATTR_KILL_SGID 0x1000ULL /* = 4096 */
1687 #define MDS_ATTR_CTIME_SET 0x2000ULL /* = 8192 */
1688 #define MDS_ATTR_FROM_OPEN 0x4000ULL /* = 16384, called from open path, ie O_TRUNC */
1689 #define MDS_ATTR_BLOCKS 0x8000ULL /* = 32768 */
1692 #define FMODE_READ 00000001
1693 #define FMODE_WRITE 00000002
1696 #define MDS_FMODE_CLOSED 00000000
1697 #define MDS_FMODE_EXEC 00000004
1698 /* MDS_FMODE_EPOCH 01000000 obsolete since 2.8.0 */
1699 /* MDS_FMODE_TRUNC 02000000 obsolete since 2.8.0 */
1700 /* MDS_FMODE_SOM 04000000 obsolete since 2.8.0 */
1702 #define MDS_OPEN_CREATED 00000010
1703 #define MDS_OPEN_CROSS 00000020
1705 #define MDS_OPEN_CREAT 00000100
1706 #define MDS_OPEN_EXCL 00000200
1707 #define MDS_OPEN_TRUNC 00001000
1708 #define MDS_OPEN_APPEND 00002000
1709 #define MDS_OPEN_SYNC 00010000
1710 #define MDS_OPEN_DIRECTORY 00200000
1712 #define MDS_OPEN_BY_FID 040000000 /* open_by_fid for known object */
1713 #define MDS_OPEN_DELAY_CREATE 0100000000 /* delay initial object create */
1714 #define MDS_OPEN_OWNEROVERRIDE 0200000000 /* NFSD rw-reopen ro file for owner */
1715 #define MDS_OPEN_JOIN_FILE 0400000000 /* open for join file.
1716 * We do not support JOIN FILE
1717 * anymore, reserve this flags
1718 * just for preventing such bit
1721 #define MDS_OPEN_LOCK 04000000000 /* This open requires open lock */
1722 #define MDS_OPEN_HAS_EA 010000000000 /* specify object create pattern */
1723 #define MDS_OPEN_HAS_OBJS 020000000000 /* Just set the EA the obj exist */
1724 #define MDS_OPEN_NORESTORE 0100000000000ULL /* Do not restore file at open */
1725 #define MDS_OPEN_NEWSTRIPE 0200000000000ULL /* New stripe needed (restripe or
1727 #define MDS_OPEN_VOLATILE 0400000000000ULL /* File is volatile = created
1729 #define MDS_OPEN_LEASE 01000000000000ULL /* Open the file and grant lease
1730 * delegation, succeed if it's not
1731 * being opened with conflict mode.
1733 #define MDS_OPEN_RELEASE 02000000000000ULL /* Open the file for HSM release */
1735 /* lustre internal open flags, which should not be set from user space */
1736 #define MDS_OPEN_FL_INTERNAL (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS | \
1737 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK | \
1738 MDS_OPEN_BY_FID | MDS_OPEN_LEASE | \
1742 MDS_CHECK_SPLIT = 1 << 0,
1743 MDS_CROSS_REF = 1 << 1,
1744 MDS_VTX_BYPASS = 1 << 2,
1745 MDS_PERM_BYPASS = 1 << 3,
1746 /* MDS_SOM = 1 << 4, obsolete since 2.8.0 */
1747 MDS_QUOTA_IGNORE = 1 << 5,
1748 /* Was MDS_CLOSE_CLEANUP (1 << 6), No more used */
1749 MDS_KEEP_ORPHAN = 1 << 7,
1750 MDS_RECOV_OPEN = 1 << 8,
1751 MDS_DATA_MODIFIED = 1 << 9,
1752 MDS_CREATE_VOLATILE = 1 << 10,
1753 MDS_OWNEROVERRIDE = 1 << 11,
1754 MDS_HSM_RELEASE = 1 << 12,
1755 MDS_RENAME_MIGRATE = 1 << 13,
1756 MDS_CLOSE_LAYOUT_SWAP = 1 << 14,
1759 /* instance of mdt_reint_rec */
1760 struct mdt_rec_create {
1768 __u32 cr_suppgid1_h;
1770 __u32 cr_suppgid2_h;
1771 struct lu_fid cr_fid1;
1772 struct lu_fid cr_fid2;
1773 struct lustre_handle cr_old_handle; /* handle in case of open replay */
1777 __u64 cr_padding_1; /* rr_blocks */
1780 /* use of helpers set/get_mrc_cr_flags() is needed to access
1781 * 64 bits cr_flags [cr_flags_l, cr_flags_h], this is done to
1782 * extend cr_flags size without breaking 1.8 compat */
1783 __u32 cr_flags_l; /* for use with open, low 32 bits */
1784 __u32 cr_flags_h; /* for use with open, high 32 bits */
1785 __u32 cr_umask; /* umask for create */
1786 __u32 cr_padding_4; /* rr_padding_4 */
1789 /* instance of mdt_reint_rec */
1790 struct mdt_rec_link {
1798 __u32 lk_suppgid1_h;
1800 __u32 lk_suppgid2_h;
1801 struct lu_fid lk_fid1;
1802 struct lu_fid lk_fid2;
1804 __u64 lk_padding_1; /* rr_atime */
1805 __u64 lk_padding_2; /* rr_ctime */
1806 __u64 lk_padding_3; /* rr_size */
1807 __u64 lk_padding_4; /* rr_blocks */
1809 __u32 lk_padding_5; /* rr_mode */
1810 __u32 lk_padding_6; /* rr_flags */
1811 __u32 lk_padding_7; /* rr_padding_2 */
1812 __u32 lk_padding_8; /* rr_padding_3 */
1813 __u32 lk_padding_9; /* rr_padding_4 */
1816 /* instance of mdt_reint_rec */
1817 struct mdt_rec_unlink {
1825 __u32 ul_suppgid1_h;
1827 __u32 ul_suppgid2_h;
1828 struct lu_fid ul_fid1;
1829 struct lu_fid ul_fid2;
1831 __u64 ul_padding_2; /* rr_atime */
1832 __u64 ul_padding_3; /* rr_ctime */
1833 __u64 ul_padding_4; /* rr_size */
1834 __u64 ul_padding_5; /* rr_blocks */
1837 __u32 ul_padding_6; /* rr_flags */
1838 __u32 ul_padding_7; /* rr_padding_2 */
1839 __u32 ul_padding_8; /* rr_padding_3 */
1840 __u32 ul_padding_9; /* rr_padding_4 */
1843 /* instance of mdt_reint_rec */
1844 struct mdt_rec_rename {
1852 __u32 rn_suppgid1_h;
1854 __u32 rn_suppgid2_h;
1855 struct lu_fid rn_fid1;
1856 struct lu_fid rn_fid2;
1858 __u64 rn_padding_1; /* rr_atime */
1859 __u64 rn_padding_2; /* rr_ctime */
1860 __u64 rn_padding_3; /* rr_size */
1861 __u64 rn_padding_4; /* rr_blocks */
1862 __u32 rn_bias; /* some operation flags */
1863 __u32 rn_mode; /* cross-ref rename has mode */
1864 __u32 rn_padding_5; /* rr_flags */
1865 __u32 rn_padding_6; /* rr_padding_2 */
1866 __u32 rn_padding_7; /* rr_padding_3 */
1867 __u32 rn_padding_8; /* rr_padding_4 */
1870 /* instance of mdt_reint_rec */
1871 struct mdt_rec_setxattr {
1879 __u32 sx_suppgid1_h;
1881 __u32 sx_suppgid2_h;
1882 struct lu_fid sx_fid;
1883 __u64 sx_padding_1; /* These three are rr_fid2 */
1888 __u64 sx_padding_5; /* rr_ctime */
1889 __u64 sx_padding_6; /* rr_size */
1890 __u64 sx_padding_7; /* rr_blocks */
1893 __u32 sx_padding_8; /* rr_flags */
1894 __u32 sx_padding_9; /* rr_padding_2 */
1895 __u32 sx_padding_10; /* rr_padding_3 */
1896 __u32 sx_padding_11; /* rr_padding_4 */
1900 * mdt_rec_reint is the template for all mdt_reint_xxx structures.
1901 * Do NOT change the size of various members, otherwise the value
1902 * will be broken in lustre_swab_mdt_rec_reint().
1904 * If you add new members in other mdt_reint_xxx structres and need to use the
1905 * rr_padding_x fields, then update lustre_swab_mdt_rec_reint() also.
1907 struct mdt_rec_reint {
1915 __u32 rr_suppgid1_h;
1917 __u32 rr_suppgid2_h;
1918 struct lu_fid rr_fid1;
1919 struct lu_fid rr_fid2;
1930 __u32 rr_padding_4; /* also fix lustre_swab_mdt_rec_reint */
1933 /* lmv structures */
1935 __u32 ld_tgt_count; /* how many MDS's */
1936 __u32 ld_active_tgt_count; /* how many active */
1937 __u32 ld_default_stripe_count; /* how many objects are used */
1938 __u32 ld_pattern; /* default hash pattern */
1939 __u64 ld_default_hash_size;
1940 __u64 ld_padding_1; /* also fix lustre_swab_lmv_desc */
1941 __u32 ld_padding_2; /* also fix lustre_swab_lmv_desc */
1942 __u32 ld_qos_maxage; /* in second */
1943 __u32 ld_padding_3; /* also fix lustre_swab_lmv_desc */
1944 __u32 ld_padding_4; /* also fix lustre_swab_lmv_desc */
1945 struct obd_uuid ld_uuid;
1948 /* LMV layout EA, and it will be stored both in master and slave object */
1949 struct lmv_mds_md_v1 {
1951 __u32 lmv_stripe_count;
1952 __u32 lmv_master_mdt_index; /* On master object, it is master
1953 * MDT index, on slave object, it
1954 * is stripe index of the slave obj */
1955 __u32 lmv_hash_type; /* dir stripe policy, i.e. indicate
1956 * which hash function to be used,
1957 * Note: only lower 16 bits is being
1958 * used for now. Higher 16 bits will
1959 * be used to mark the object status,
1960 * for example migrating or dead. */
1961 __u32 lmv_layout_version; /* Used for directory restriping */
1965 char lmv_pool_name[LOV_MAXPOOLNAME + 1]; /* pool name */
1966 struct lu_fid lmv_stripe_fids[0]; /* FIDs for each stripe */
1969 #define LMV_MAGIC_V1 0x0CD20CD0 /* normal stripe lmv magic */
1970 #define LMV_MAGIC LMV_MAGIC_V1
1972 /* #define LMV_USER_MAGIC 0x0CD30CD0 */
1973 #define LMV_MAGIC_STRIPE 0x0CD40CD0 /* magic for dir sub_stripe */
1975 /* Right now only the lower part(0-16bits) of lmv_hash_type is being used,
1976 * and the higher part will be the flag to indicate the status of object,
1977 * for example the object is being migrated. And the hash function
1978 * might be interpreted differently with different flags. */
1979 #define LMV_HASH_TYPE_MASK 0x0000ffff
1981 #define LMV_HASH_FLAG_MIGRATION 0x80000000
1983 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 10, 53, 0)
1984 /* Since lustre 2.8, this flag will not be needed, instead this DEAD
1985 * and orphan flags will be stored in LMA (see LMAI_ORPHAN)
1986 * Keep this flag just for LFSCK, because it still might meet such
1987 * flag when it checks the old FS */
1988 #define LMV_HASH_FLAG_DEAD 0x40000000
1990 #define LMV_HASH_FLAG_BAD_TYPE 0x20000000
1992 /* The striped directory has ever lost its master LMV EA, then LFSCK
1993 * re-generated it. This flag is used to indicate such case. It is an
1995 #define LMV_HASH_FLAG_LOST_LMV 0x10000000
1998 * The FNV-1a hash algorithm is as follows:
1999 * hash = FNV_offset_basis
2000 * for each octet_of_data to be hashed
2001 * hash = hash XOR octet_of_data
2002 * hash = hash × FNV_prime
2004 * http://en.wikipedia.org/wiki/Fowler–Noll–Vo_hash_function#FNV-1a_hash
2006 * http://www.isthe.com/chongo/tech/comp/fnv/index.html#FNV-reference-source
2007 * FNV_prime is 2^40 + 2^8 + 0xb3 = 0x100000001b3ULL
2009 #define LUSTRE_FNV_1A_64_PRIME 0x100000001b3ULL
2010 #define LUSTRE_FNV_1A_64_OFFSET_BIAS 0xcbf29ce484222325ULL
2011 static inline __u64 lustre_hash_fnv_1a_64(const void *buf, size_t size)
2013 __u64 hash = LUSTRE_FNV_1A_64_OFFSET_BIAS;
2014 const unsigned char *p = buf;
2017 for (i = 0; i < size; i++) {
2019 hash *= LUSTRE_FNV_1A_64_PRIME;
2027 struct lmv_mds_md_v1 lmv_md_v1;
2028 struct lmv_user_md lmv_user_md;
2031 static inline int lmv_mds_md_size(int stripe_count, unsigned int lmm_magic)
2033 switch (lmm_magic) {
2035 struct lmv_mds_md_v1 *lmm1;
2037 return sizeof(*lmm1) + stripe_count *
2038 sizeof(lmm1->lmv_stripe_fids[0]);
2045 static inline int lmv_mds_md_stripe_count_get(const union lmv_mds_md *lmm)
2047 switch (__le32_to_cpu(lmm->lmv_magic)) {
2049 return __le32_to_cpu(lmm->lmv_md_v1.lmv_stripe_count);
2050 case LMV_USER_MAGIC:
2051 return __le32_to_cpu(lmm->lmv_user_md.lum_stripe_count);
2057 static inline int lmv_mds_md_hash_type_get(const union lmv_mds_md *lmm)
2059 switch (__le32_to_cpu(lmm->lmv_magic)) {
2061 return __le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type);
2062 case LMV_USER_MAGIC:
2063 return __le32_to_cpu(lmm->lmv_user_md.lum_hash_type);
2073 FLD_FIRST_OPC = FLD_QUERY
2079 SEQ_FIRST_OPC = SEQ_QUERY
2083 SEQ_ALLOC_SUPER = 0,
2095 LFSCK_NOTIFY = 1101,
2098 LFSCK_FIRST_OPC = LFSCK_NOTIFY
2102 * LOV data structures
2105 #define LOV_MAX_UUID_BUFFER_SIZE 8192
2106 /* The size of the buffer the lov/mdc reserves for the
2107 * array of UUIDs returned by the MDS. With the current
2108 * protocol, this will limit the max number of OSTs per LOV */
2110 #define LOV_DESC_MAGIC 0xB0CCDE5C
2111 #define LOV_DESC_QOS_MAXAGE_DEFAULT 5 /* Seconds */
2112 #define LOV_DESC_STRIPE_SIZE_DEFAULT (1 << LNET_MTU_BITS)
2114 /* LOV settings descriptor (should only contain static info) */
2116 __u32 ld_tgt_count; /* how many OBD's */
2117 __u32 ld_active_tgt_count; /* how many active */
2118 __u32 ld_default_stripe_count; /* how many objects are used */
2119 __u32 ld_pattern; /* default PATTERN_RAID0 */
2120 __u64 ld_default_stripe_size; /* in bytes */
2121 __u64 ld_default_stripe_offset; /* in bytes */
2122 __u32 ld_padding_0; /* unused */
2123 __u32 ld_qos_maxage; /* in second */
2124 __u32 ld_padding_1; /* also fix lustre_swab_lov_desc */
2125 __u32 ld_padding_2; /* also fix lustre_swab_lov_desc */
2126 struct obd_uuid ld_uuid;
2129 #define ld_magic ld_active_tgt_count /* for swabbing from llogs */
2134 /* opcodes -- MUST be distinct from OST/MDS opcodes */
2139 LDLM_BL_CALLBACK = 104,
2140 LDLM_CP_CALLBACK = 105,
2141 LDLM_GL_CALLBACK = 106,
2142 LDLM_SET_INFO = 107,
2145 #define LDLM_FIRST_OPC LDLM_ENQUEUE
2147 #define RES_NAME_SIZE 4
2148 struct ldlm_res_id {
2149 __u64 name[RES_NAME_SIZE];
2152 #define DLDLMRES "[%#llx:%#llx:%#llx].%#llx"
2153 #define PLDLMRES(res) (unsigned long long)(res)->lr_name.name[0], \
2154 (unsigned long long)(res)->lr_name.name[1], \
2155 (unsigned long long)(res)->lr_name.name[2], \
2156 (unsigned long long)(res)->lr_name.name[3]
2159 typedef enum ldlm_mode {
2172 #define LCK_MODE_NUM 8
2174 typedef enum ldlm_type {
2182 #define LDLM_MIN_TYPE LDLM_PLAIN
2184 struct ldlm_extent {
2190 struct ldlm_inodebits {
2194 struct ldlm_flock_wire {
2202 /* it's important that the fields of the ldlm_extent structure match
2203 * the first fields of the ldlm_flock structure because there is only
2204 * one ldlm_swab routine to process the ldlm_policy_data_t union. if
2205 * this ever changes we will need to swab the union differently based
2206 * on the resource type. */
2208 typedef union ldlm_wire_policy_data {
2209 struct ldlm_extent l_extent;
2210 struct ldlm_flock_wire l_flock;
2211 struct ldlm_inodebits l_inodebits;
2212 } ldlm_wire_policy_data_t;
2214 struct barrier_lvb {
2220 struct ldlm_gl_barrier_desc {
2226 union ldlm_gl_desc {
2227 struct ldlm_gl_lquota_desc lquota_desc;
2228 struct ldlm_gl_barrier_desc barrier_desc;
2231 enum ldlm_intent_flags {
2232 IT_OPEN = 0x00000001,
2233 IT_CREAT = 0x00000002,
2234 IT_OPEN_CREAT = 0x00000003,
2235 IT_READDIR = 0x00000004,
2236 IT_GETATTR = 0x00000008,
2237 IT_LOOKUP = 0x00000010,
2238 IT_UNLINK = 0x00000020,
2239 IT_TRUNC = 0x00000040,
2240 IT_GETXATTR = 0x00000080,
2241 IT_EXEC = 0x00000100,
2242 IT_PIN = 0x00000200,
2243 IT_LAYOUT = 0x00000400,
2244 IT_QUOTA_DQACQ = 0x00000800,
2245 IT_QUOTA_CONN = 0x00001000,
2246 IT_SETXATTR = 0x00002000,
2249 struct ldlm_intent {
2253 struct ldlm_resource_desc {
2254 enum ldlm_type lr_type;
2255 __u32 lr_pad; /* also fix lustre_swab_ldlm_resource_desc */
2256 struct ldlm_res_id lr_name;
2259 struct ldlm_lock_desc {
2260 struct ldlm_resource_desc l_resource;
2261 enum ldlm_mode l_req_mode;
2262 enum ldlm_mode l_granted_mode;
2263 union ldlm_wire_policy_data l_policy_data;
2266 #define LDLM_LOCKREQ_HANDLES 2
2267 #define LDLM_ENQUEUE_CANCEL_OFF 1
2269 struct ldlm_request {
2272 struct ldlm_lock_desc lock_desc;
2273 struct lustre_handle lock_handle[LDLM_LOCKREQ_HANDLES];
2278 __u32 lock_padding; /* also fix lustre_swab_ldlm_reply */
2279 struct ldlm_lock_desc lock_desc;
2280 struct lustre_handle lock_handle;
2281 __u64 lock_policy_res1;
2282 __u64 lock_policy_res2;
2285 #define ldlm_flags_to_wire(flags) ((__u32)(flags))
2286 #define ldlm_flags_from_wire(flags) ((__u64)(flags))
2289 * Opcodes for mountconf (mgs and mgc)
2294 MGS_EXCEPTION, /* node died, etc. */
2295 MGS_TARGET_REG, /* whenever target starts up */
2301 #define MGS_FIRST_OPC MGS_CONNECT
2303 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
2304 #define MGS_PARAM_MAXLEN 1024
2305 #define KEY_SET_INFO "set_info"
2307 struct mgs_send_param {
2308 char mgs_param[MGS_PARAM_MAXLEN];
2312 /* We pass this info to the MGS so it can write config logs */
2313 #define MTI_NAME_MAXLEN 64
2314 #define MTI_PARAM_MAXLEN 4096
2315 #define MTI_NIDS_MAX 32
2316 struct mgs_target_info {
2317 __u32 mti_lustre_ver;
2318 __u32 mti_stripe_index;
2319 __u32 mti_config_ver;
2321 __u32 mti_nid_count;
2322 __u32 mti_instance; /* Running instance of target */
2323 char mti_fsname[MTI_NAME_MAXLEN];
2324 char mti_svname[MTI_NAME_MAXLEN];
2325 char mti_uuid[sizeof(struct obd_uuid)];
2326 __u64 mti_nids[MTI_NIDS_MAX]; /* host nids (lnet_nid_t)*/
2327 char mti_params[MTI_PARAM_MAXLEN];
2330 struct mgs_nidtbl_entry {
2331 __u64 mne_version; /* table version of this entry */
2332 __u32 mne_instance; /* target instance # */
2333 __u32 mne_index; /* target index */
2334 __u32 mne_length; /* length of this entry - by bytes */
2335 __u8 mne_type; /* target type LDD_F_SV_TYPE_OST/MDT */
2336 __u8 mne_nid_type; /* type of nid(mbz). for ipv6. */
2337 __u8 mne_nid_size; /* size of each NID, by bytes */
2338 __u8 mne_nid_count; /* # of NIDs in buffer */
2340 lnet_nid_t nids[0]; /* variable size buffer for NIDs. */
2345 CONFIG_T_CONFIG = 0,
2346 CONFIG_T_SPTLRPC = 1,
2347 CONFIG_T_RECOVER = 2,
2348 CONFIG_T_PARAMS = 3,
2349 CONFIG_T_NODEMAP = 4,
2350 CONFIG_T_BARRIER = 5,
2354 struct mgs_config_body {
2355 char mcb_name[MTI_NAME_MAXLEN]; /* logname */
2356 __u64 mcb_offset; /* next index of config log to request */
2357 __u16 mcb_type; /* type of log: CONFIG_T_[CONFIG|RECOVER] */
2358 __u8 mcb_nm_cur_pass;
2359 __u8 mcb_bits; /* bits unit size of config log */
2360 __u32 mcb_units; /* # of units for bulk transfer */
2363 struct mgs_config_res {
2364 __u64 mcr_offset; /* index of last config log */
2366 __u64 mcr_size; /* size of the log */
2367 __u64 mcr_nm_cur_pass; /* current nodemap config pass */
2371 /* Config marker flags (in config log) */
2372 #define CM_START 0x01
2374 #define CM_SKIP 0x04
2375 #define CM_UPGRADE146 0x08
2376 #define CM_EXCLUDE 0x10
2377 #define CM_START_SKIP (CM_START | CM_SKIP)
2380 __u32 cm_step; /* aka config version */
2382 __u32 cm_vers; /* lustre release version number */
2383 __u32 cm_padding; /* 64 bit align */
2384 __s64 cm_createtime; /*when this record was first created */
2385 __s64 cm_canceltime; /*when this record is no longer valid*/
2386 char cm_tgtname[MTI_NAME_MAXLEN];
2387 char cm_comment[MTI_NAME_MAXLEN];
2391 * Opcodes for multiple servers.
2397 OBD_QC_CALLBACK, /* not used since 2.4 */
2401 #define OBD_FIRST_OPC OBD_PING
2404 * llog contexts indices.
2406 * There is compatibility problem with indexes below, they are not
2407 * continuous and must keep their numbers for compatibility needs.
2408 * See LU-5218 for details.
2411 LLOG_CONFIG_ORIG_CTXT = 0,
2412 LLOG_CONFIG_REPL_CTXT = 1,
2413 LLOG_MDS_OST_ORIG_CTXT = 2,
2414 LLOG_MDS_OST_REPL_CTXT = 3, /* kept just to avoid re-assignment */
2415 LLOG_SIZE_ORIG_CTXT = 4,
2416 LLOG_SIZE_REPL_CTXT = 5,
2417 LLOG_TEST_ORIG_CTXT = 8,
2418 LLOG_TEST_REPL_CTXT = 9, /* kept just to avoid re-assignment */
2419 LLOG_CHANGELOG_ORIG_CTXT = 12, /**< changelog generation on mdd */
2420 LLOG_CHANGELOG_REPL_CTXT = 13, /**< changelog access on clients */
2421 /* for multiple changelog consumers */
2422 LLOG_CHANGELOG_USER_ORIG_CTXT = 14,
2423 LLOG_AGENT_ORIG_CTXT = 15, /**< agent requests generation on cdt */
2424 LLOG_UPDATELOG_ORIG_CTXT = 16, /* update log */
2425 LLOG_UPDATELOG_REPL_CTXT = 17, /* update log */
2429 /** Identifier for a single log object */
2431 struct ost_id lgl_oi;
2433 } __attribute__((packed));
2435 /** Records written to the CATALOGS list */
2436 #define CATLIST "CATALOGS"
2438 struct llog_logid lci_logid;
2442 } __attribute__((packed));
2444 /* Log data record types - there is no specific reason that these need to
2445 * be related to the RPC opcodes, but no reason not to (may be handy later?)
2447 #define LLOG_OP_MAGIC 0x10600000
2448 #define LLOG_OP_MASK 0xfff00000
2451 LLOG_PAD_MAGIC = LLOG_OP_MAGIC | 0x00000,
2452 OST_SZ_REC = LLOG_OP_MAGIC | 0x00f00,
2453 /* OST_RAID1_REC = LLOG_OP_MAGIC | 0x01000, never used */
2454 MDS_UNLINK_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) |
2455 REINT_UNLINK, /* obsolete after 2.5.0 */
2456 MDS_UNLINK64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
2458 /* MDS_SETATTR_REC = LLOG_OP_MAGIC | 0x12401, obsolete 1.8.0 */
2459 MDS_SETATTR64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
2461 OBD_CFG_REC = LLOG_OP_MAGIC | 0x20000,
2462 /* PTL_CFG_REC = LLOG_OP_MAGIC | 0x30000, obsolete 1.4.0 */
2463 LLOG_GEN_REC = LLOG_OP_MAGIC | 0x40000,
2464 /* LLOG_JOIN_REC = LLOG_OP_MAGIC | 0x50000, obsolete 1.8.0 */
2465 CHANGELOG_REC = LLOG_OP_MAGIC | 0x60000,
2466 CHANGELOG_USER_REC = LLOG_OP_MAGIC | 0x70000,
2467 HSM_AGENT_REC = LLOG_OP_MAGIC | 0x80000,
2468 UPDATE_REC = LLOG_OP_MAGIC | 0xa0000,
2469 LLOG_HDR_MAGIC = LLOG_OP_MAGIC | 0x45539,
2470 LLOG_LOGID_MAGIC = LLOG_OP_MAGIC | 0x4553b,
2473 #define LLOG_REC_HDR_NEEDS_SWABBING(r) \
2474 (((r)->lrh_type & __swab32(LLOG_OP_MASK)) == __swab32(LLOG_OP_MAGIC))
2476 /** Log record header - stored in little endian order.
2477 * Each record must start with this struct, end with a llog_rec_tail,
2478 * and be a multiple of 256 bits in size.
2480 struct llog_rec_hdr {
2487 struct llog_rec_tail {
2492 /* Where data follow just after header */
2493 #define REC_DATA(ptr) \
2494 ((void *)((char *)ptr + sizeof(struct llog_rec_hdr)))
2496 #define REC_DATA_LEN(rec) \
2497 (rec->lrh_len - sizeof(struct llog_rec_hdr) - \
2498 sizeof(struct llog_rec_tail))
2500 struct llog_logid_rec {
2501 struct llog_rec_hdr lid_hdr;
2502 struct llog_logid lid_id;
2506 struct llog_rec_tail lid_tail;
2507 } __attribute__((packed));
2509 struct llog_unlink_rec {
2510 struct llog_rec_hdr lur_hdr;
2514 struct llog_rec_tail lur_tail;
2515 } __attribute__((packed));
2517 struct llog_unlink64_rec {
2518 struct llog_rec_hdr lur_hdr;
2519 struct lu_fid lur_fid;
2520 __u32 lur_count; /* to destroy the lost precreated */
2524 struct llog_rec_tail lur_tail;
2525 } __attribute__((packed));
2527 struct llog_setattr64_rec {
2528 struct llog_rec_hdr lsr_hdr;
2529 struct ost_id lsr_oi;
2535 struct llog_rec_tail lsr_tail;
2536 } __attribute__((packed));
2538 struct llog_size_change_rec {
2539 struct llog_rec_hdr lsc_hdr;
2540 struct ll_fid lsc_fid;
2545 struct llog_rec_tail lsc_tail;
2546 } __attribute__((packed));
2548 #define CHANGELOG_MAGIC 0xca103000
2550 /** \a changelog_rec_type's that can't be masked */
2551 #define CHANGELOG_MINMASK (1 << CL_MARK)
2552 /** bits covering all \a changelog_rec_type's */
2553 #define CHANGELOG_ALLMASK 0XFFFFFFFF
2554 /** default \a changelog_rec_type mask. Allow all of them, except
2555 * CL_ATIME since it can really be time consuming, and not necessary
2556 * under normal use. */
2557 #define CHANGELOG_DEFMASK (CHANGELOG_ALLMASK & ~(1 << CL_ATIME))
2559 /* changelog llog name, needed by client replicators */
2560 #define CHANGELOG_CATALOG "changelog_catalog"
2562 struct changelog_setinfo {
2565 } __attribute__((packed));
2567 /** changelog record */
2568 struct llog_changelog_rec {
2569 struct llog_rec_hdr cr_hdr;
2570 struct changelog_rec cr; /**< Variable length field */
2571 struct llog_rec_tail cr_do_not_use; /**< for_sizeof_only */
2572 } __attribute__((packed));
2574 #define CHANGELOG_USER_PREFIX "cl"
2576 struct llog_changelog_user_rec {
2577 struct llog_rec_hdr cur_hdr;
2581 struct llog_rec_tail cur_tail;
2582 } __attribute__((packed));
2584 enum agent_req_status {
2592 static inline const char *agent_req_status2name(enum agent_req_status ars)
2610 struct llog_agent_req_rec {
2611 struct llog_rec_hdr arr_hdr; /**< record header */
2612 __u32 arr_status; /**< status of the request */
2614 * agent_req_status */
2615 __u32 arr_archive_id; /**< backend archive number */
2616 __u64 arr_flags; /**< req flags */
2617 __u64 arr_compound_id; /**< compound cookie */
2618 __u64 arr_req_create; /**< req. creation time */
2619 __u64 arr_req_change; /**< req. status change time */
2620 struct hsm_action_item arr_hai; /**< req. to the agent */
2621 struct llog_rec_tail arr_tail; /**< record tail for_sizezof_only */
2622 } __attribute__((packed));
2624 /* Old llog gen for compatibility */
2628 } __attribute__((packed));
2630 struct llog_gen_rec {
2631 struct llog_rec_hdr lgr_hdr;
2632 struct llog_gen lgr_gen;
2636 struct llog_rec_tail lgr_tail;
2639 /* flags for the logs */
2641 LLOG_F_ZAP_WHEN_EMPTY = 0x1,
2642 LLOG_F_IS_CAT = 0x2,
2643 LLOG_F_IS_PLAIN = 0x4,
2644 LLOG_F_EXT_JOBID = 0x8,
2645 LLOG_F_IS_FIXSIZE = 0x10,
2647 /* Note: Flags covered by LLOG_F_EXT_MASK will be inherited from
2648 * catlog to plain log, so do not add LLOG_F_IS_FIXSIZE here,
2649 * because the catlog record is usually fixed size, but its plain
2650 * log record can be variable */
2651 LLOG_F_EXT_MASK = LLOG_F_EXT_JOBID,
2654 /* On-disk header structure of each log object, stored in little endian order */
2655 #define LLOG_MIN_CHUNK_SIZE 8192
2656 #define LLOG_HEADER_SIZE (96) /* sizeof (llog_log_hdr) + sizeof(llh_tail)
2657 * - sizeof(llh_bitmap) */
2658 #define LLOG_BITMAP_BYTES (LLOG_MIN_CHUNK_SIZE - LLOG_HEADER_SIZE)
2659 #define LLOG_MIN_REC_SIZE (24) /* round(llog_rec_hdr + llog_rec_tail) */
2661 struct llog_log_hdr {
2662 struct llog_rec_hdr llh_hdr;
2663 __s64 llh_timestamp;
2665 __u32 llh_bitmap_offset;
2668 /* for a catalog the first/oldest and still in-use plain slot is just
2669 * next to it. It will serve as the upper limit after Catalog has
2672 struct obd_uuid llh_tgtuuid;
2673 __u32 llh_reserved[LLOG_HEADER_SIZE/sizeof(__u32)-23];
2674 /* These fields must always be at the end of the llog_log_hdr.
2675 * Note: llh_bitmap size is variable because llog chunk size could be
2676 * bigger than LLOG_MIN_CHUNK_SIZE, i.e. sizeof(llog_log_hdr) > 8192
2677 * bytes, and the real size is stored in llh_hdr.lrh_len, which means
2678 * llh_tail should only be refered by LLOG_HDR_TAIL().
2679 * But this structure is also used by client/server llog interface
2680 * (see llog_client.c), it will be kept in its original way to avoid
2681 * compatiblity issue. */
2682 __u32 llh_bitmap[LLOG_BITMAP_BYTES / sizeof(__u32)];
2683 struct llog_rec_tail llh_tail;
2684 } __attribute__((packed));
2685 #undef LLOG_HEADER_SIZE
2686 #undef LLOG_BITMAP_BYTES
2688 #define LLOG_HDR_BITMAP_SIZE(llh) (__u32)((llh->llh_hdr.lrh_len - \
2689 llh->llh_bitmap_offset - \
2690 sizeof(llh->llh_tail)) * 8)
2691 #define LLOG_HDR_BITMAP(llh) (__u32 *)((char *)(llh) + \
2692 (llh)->llh_bitmap_offset)
2693 #define LLOG_HDR_TAIL(llh) ((struct llog_rec_tail *)((char *)llh + \
2694 llh->llh_hdr.lrh_len - \
2695 sizeof(llh->llh_tail)))
2697 /** log cookies are used to reference a specific log file and a record therein */
2698 struct llog_cookie {
2699 struct llog_logid lgc_lgl;
2703 } __attribute__((packed));
2705 /** llog protocol */
2706 enum llogd_rpc_ops {
2707 LLOG_ORIGIN_HANDLE_CREATE = 501,
2708 LLOG_ORIGIN_HANDLE_NEXT_BLOCK = 502,
2709 LLOG_ORIGIN_HANDLE_READ_HEADER = 503,
2710 LLOG_ORIGIN_HANDLE_WRITE_REC = 504,
2711 LLOG_ORIGIN_HANDLE_CLOSE = 505,
2712 LLOG_ORIGIN_CONNECT = 506,
2713 LLOG_CATINFO = 507, /* deprecated */
2714 LLOG_ORIGIN_HANDLE_PREV_BLOCK = 508,
2715 LLOG_ORIGIN_HANDLE_DESTROY = 509, /* for destroy llog object*/
2717 LLOG_FIRST_OPC = LLOG_ORIGIN_HANDLE_CREATE
2721 struct llog_logid lgd_logid;
2723 __u32 lgd_llh_flags;
2725 __u32 lgd_saved_index;
2727 __u64 lgd_cur_offset;
2728 } __attribute__((packed));
2730 struct llogd_conn_body {
2731 struct llog_gen lgdc_gen;
2732 struct llog_logid lgdc_logid;
2733 __u32 lgdc_ctxt_idx;
2734 } __attribute__((packed));
2736 /* Note: 64-bit types are 64-bit aligned in structure */
2738 __u64 o_valid; /* hot fields in this obdo */
2741 __u64 o_size; /* o_size-o_blocks == ost_lvb */
2745 __u64 o_blocks; /* brw: cli sent cached bytes */
2748 /* 32-bit fields start here: keep an even number of them via padding */
2749 __u32 o_blksize; /* optimal IO blocksize */
2750 __u32 o_mode; /* brw: cli sent cache remain */
2754 __u32 o_nlink; /* brw: checksum */
2756 __u32 o_misc; /* brw: o_dropped */
2758 __u64 o_ioepoch; /* epoch in ost writes */
2759 __u32 o_stripe_idx; /* holds stripe idx */
2761 struct lustre_handle o_handle; /* brw: lock handle to prolong
2763 struct llog_cookie o_lcookie; /* destroy: unlink cookie from
2764 * MDS, obsolete in 2.8, reused
2769 __u64 o_data_version; /* getattr: sum of iversion for
2771 * brw: grant space consumed on
2772 * the client for the write */
2778 #define o_dirty o_blocks
2779 #define o_undirty o_mode
2780 #define o_dropped o_misc
2781 #define o_cksum o_nlink
2782 #define o_grant_used o_data_version
2784 struct lfsck_request {
2797 __u16 lr_async_windows;
2799 struct lu_fid lr_fid;
2800 struct lu_fid lr_fid2;
2801 struct lu_fid lr_fid3;
2806 struct lfsck_reply {
2813 LE_LASTID_REBUILDING = 1,
2814 LE_LASTID_REBUILT = 2,
2820 LE_FID_ACCESSED = 8,
2822 LE_CONDITIONAL_DESTROY = 10,
2823 LE_PAIRS_VERIFY = 11,
2824 LE_SET_LMV_MASTER = 15,
2825 LE_SET_LMV_SLAVE = 16,
2828 enum lfsck_event_flags {
2829 LEF_TO_OST = 0x00000001,
2830 LEF_FROM_OST = 0x00000002,
2831 LEF_SET_LMV_HASH = 0x00000004,
2832 LEF_SET_LMV_ALL = 0x00000008,
2833 LEF_RECHECK_NAME_HASH = 0x00000010,
2834 LEF_QUERY_ALL = 0x00000020,
2837 /* request structure for OST's */
2842 /* Key for FIEMAP to be used in get_info calls */
2843 struct ll_fiemap_info_key {
2845 struct obdo lfik_oa;
2846 struct fiemap lfik_fiemap;
2849 #define IDX_INFO_MAGIC 0x3D37CC37
2851 /* Index file transfer through the network. The server serializes the index into
2852 * a byte stream which is sent to the client via a bulk transfer */
2856 /* reply: see idx_info_flags below */
2859 /* request & reply: number of lu_idxpage (to be) transferred */
2863 /* request: requested attributes passed down to the iterator API */
2866 /* request & reply: index file identifier (FID) */
2867 struct lu_fid ii_fid;
2869 /* reply: version of the index file before starting to walk the index.
2870 * Please note that the version can be modified at any time during the
2874 /* request: hash to start with:
2875 * reply: hash of the first entry of the first lu_idxpage and hash
2876 * of the entry to read next if any */
2877 __u64 ii_hash_start;
2880 /* reply: size of keys in lu_idxpages, minimal one if II_FL_VARKEY is
2884 /* reply: size of records in lu_idxpages, minimal one if II_FL_VARREC
2893 #define II_END_OFF MDS_DIR_END_OFF /* all entries have been read */
2895 /* List of flags used in idx_info::ii_flags */
2896 enum idx_info_flags {
2897 II_FL_NOHASH = 1 << 0, /* client doesn't care about hash value */
2898 II_FL_VARKEY = 1 << 1, /* keys can be of variable size */
2899 II_FL_VARREC = 1 << 2, /* records can be of variable size */
2900 II_FL_NONUNQ = 1 << 3, /* index supports non-unique keys */
2901 II_FL_NOKEY = 1 << 4, /* client doesn't care about key */
2904 #define LIP_MAGIC 0x8A6D6B6C
2906 /* 4KB (= LU_PAGE_SIZE) container gathering key/record pairs */
2908 /* 16-byte header */
2911 __u16 lip_nr; /* number of entries in the container */
2912 __u64 lip_pad0; /* additional padding for future use */
2914 /* key/record pairs are stored in the remaining 4080 bytes.
2915 * depending upon the flags in idx_info::ii_flags, each key/record
2916 * pair might be preceded by:
2918 * - the key size (II_FL_VARKEY is set)
2919 * - the record size (II_FL_VARREC is set)
2921 * For the time being, we only support fixed-size key & record. */
2922 char lip_entries[0];
2925 #define LIP_HDR_SIZE (offsetof(struct lu_idxpage, lip_entries))
2927 /* Gather all possible type associated with a 4KB container */
2929 struct lu_dirpage lp_dir; /* for MDS_READPAGE */
2930 struct lu_idxpage lp_idx; /* for OBD_IDX_READ */
2931 char lp_array[LU_PAGE_SIZE];
2934 /* security opcodes */
2937 SEC_CTX_INIT_CONT = 802,
2940 SEC_FIRST_OPC = SEC_CTX_INIT
2944 * capa related definitions
2946 #define CAPA_HMAC_MAX_LEN 64
2947 #define CAPA_HMAC_KEY_MAX_LEN 56
2949 /* NB take care when changing the sequence of elements this struct,
2950 * because the offset info is used in find_capa() */
2951 struct lustre_capa {
2952 struct lu_fid lc_fid; /** fid */
2953 __u64 lc_opc; /** operations allowed */
2954 __u64 lc_uid; /** file owner */
2955 __u64 lc_gid; /** file group */
2956 __u32 lc_flags; /** HMAC algorithm & flags */
2957 __u32 lc_keyid; /** key# used for the capability */
2958 __u32 lc_timeout; /** capa timeout value (sec) */
2959 __u32 lc_expiry; /** expiry time (sec) */
2960 __u8 lc_hmac[CAPA_HMAC_MAX_LEN]; /** HMAC */
2961 } __attribute__((packed));
2963 /** lustre_capa::lc_opc */
2965 CAPA_OPC_BODY_WRITE = 1<<0, /**< write object data */
2966 CAPA_OPC_BODY_READ = 1<<1, /**< read object data */
2967 CAPA_OPC_INDEX_LOOKUP = 1<<2, /**< lookup object fid */
2968 CAPA_OPC_INDEX_INSERT = 1<<3, /**< insert object fid */
2969 CAPA_OPC_INDEX_DELETE = 1<<4, /**< delete object fid */
2970 CAPA_OPC_OSS_WRITE = 1<<5, /**< write oss object data */
2971 CAPA_OPC_OSS_READ = 1<<6, /**< read oss object data */
2972 CAPA_OPC_OSS_TRUNC = 1<<7, /**< truncate oss object */
2973 CAPA_OPC_OSS_DESTROY = 1<<8, /**< destroy oss object */
2974 CAPA_OPC_META_WRITE = 1<<9, /**< write object meta data */
2975 CAPA_OPC_META_READ = 1<<10, /**< read object meta data */
2978 #define CAPA_OPC_OSS_RW (CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE)
2979 #define CAPA_OPC_MDS_ONLY \
2980 (CAPA_OPC_BODY_WRITE | CAPA_OPC_BODY_READ | CAPA_OPC_INDEX_LOOKUP | \
2981 CAPA_OPC_INDEX_INSERT | CAPA_OPC_INDEX_DELETE)
2982 #define CAPA_OPC_OSS_ONLY \
2983 (CAPA_OPC_OSS_WRITE | CAPA_OPC_OSS_READ | CAPA_OPC_OSS_TRUNC | \
2984 CAPA_OPC_OSS_DESTROY)
2985 #define CAPA_OPC_MDS_DEFAULT ~CAPA_OPC_OSS_ONLY
2986 #define CAPA_OPC_OSS_DEFAULT ~(CAPA_OPC_MDS_ONLY | CAPA_OPC_OSS_ONLY)
2988 /* lustre_capa::lc_hmac_alg */
2990 CAPA_HMAC_ALG_SHA1 = 1, /**< sha1 algorithm */
2994 #define CAPA_FL_MASK 0x00ffffff
2995 #define CAPA_HMAC_ALG_MASK 0xff000000
2997 struct lustre_capa_key {
2998 __u64 lk_seq; /**< mds# */
2999 __u32 lk_keyid; /**< key# */
3001 __u8 lk_key[CAPA_HMAC_KEY_MAX_LEN]; /**< key */
3002 } __attribute__((packed));
3004 /** The link ea holds 1 \a link_ea_entry for each hardlink */
3005 #define LINK_EA_MAGIC 0x11EAF1DFUL
3006 struct link_ea_header {
3009 __u64 leh_len; /* total size */
3010 __u32 leh_overflow_time;
3014 /** Hardlink data is name and parent fid.
3015 * Stored in this crazy struct for maximum packing and endian-neutrality
3017 struct link_ea_entry {
3018 /** __u16 stored big-endian, unaligned */
3019 unsigned char lee_reclen[2];
3020 unsigned char lee_parent_fid[sizeof(struct lu_fid)];
3022 }__attribute__((packed));
3024 /** fid2path request/reply structure */
3025 struct getinfo_fid2path {
3026 struct lu_fid gf_fid;
3032 struct lu_fid gf_root_fid[0];
3034 } __attribute__((packed));
3036 /** path2parent request/reply structures */
3038 struct lu_fid gp_fid; /**< parent FID */
3039 __u32 gp_linkno; /**< hardlink number */
3040 __u32 gp_name_size; /**< size of the name field */
3041 char gp_name[0]; /**< zero-terminated link name */
3042 } __attribute__((packed));
3045 LAYOUT_INTENT_ACCESS = 0,
3046 LAYOUT_INTENT_READ = 1,
3047 LAYOUT_INTENT_WRITE = 2,
3048 LAYOUT_INTENT_GLIMPSE = 3,
3049 LAYOUT_INTENT_TRUNC = 4,
3050 LAYOUT_INTENT_RELEASE = 5,
3051 LAYOUT_INTENT_RESTORE = 6
3054 /* enqueue layout lock with intent */
3055 struct layout_intent {
3056 __u32 li_opc; /* intent operation for enqueue, read, write etc */
3063 * On the wire version of hsm_progress structure.
3065 * Contains the userspace hsm_progress and some internal fields.
3067 struct hsm_progress_kernel {
3068 /* Field taken from struct hsm_progress */
3071 struct hsm_extent hpk_extent;
3073 __u16 hpk_errval; /* positive val */
3075 /* Additional fields */
3076 __u64 hpk_data_version;
3078 } __attribute__((packed));
3081 * OUT_UPDATE RPC Format
3083 * During the cross-ref operation, the Master MDT, which the client send the
3084 * request to, will disassembly the operation into object updates, then OSP
3085 * will send these updates to the remote MDT to be executed.
3087 * An UPDATE_OBJ RPC does a list of updates. Each update belongs to an
3088 * operation and does a type of modification to an object.
3096 * update (ub_count-th)
3098 * ub_count must be less than or equal to UPDATE_PER_RPC_MAX.
3103 * rc [+ buffers] (1st)
3104 * rc [+ buffers] (2st)
3106 * rc [+ buffers] (nr_count-th)
3108 * ur_count must be less than or equal to UPDATE_PER_RPC_MAX and should usually
3109 * be equal to ub_count.
3113 * Type of each update, if adding/deleting update, please also update
3114 * update_opcode in lustre/target/out_lib.c.
3126 OUT_INDEX_LOOKUP = 9,
3127 OUT_INDEX_INSERT = 10,
3128 OUT_INDEX_DELETE = 11,
3138 UPDATE_FL_OST = 0x00000001, /* op from OST (not MDT) */
3139 UPDATE_FL_SYNC = 0x00000002, /* commit before replying */
3140 UPDATE_FL_COMMITTED = 0x00000004, /* op committed globally */
3141 UPDATE_FL_NOLOG = 0x00000008 /* for idempotent updates */
3144 struct object_update_param {
3145 __u16 oup_len; /* length of this parameter */
3152 struct object_update {
3153 __u16 ou_type; /* enum update_type */
3154 __u16 ou_params_count; /* update parameters count */
3155 __u32 ou_result_size; /* how many bytes can return */
3156 __u32 ou_flags; /* enum update_flag */
3157 __u32 ou_padding1; /* padding 1 */
3158 __u64 ou_batchid; /* op transno on master */
3159 struct lu_fid ou_fid; /* object to be updated */
3160 struct object_update_param ou_params[0]; /* update params */
3163 #define UPDATE_REQUEST_MAGIC_V1 0xBDDE0001
3164 #define UPDATE_REQUEST_MAGIC_V2 0xBDDE0002
3165 #define UPDATE_REQUEST_MAGIC UPDATE_REQUEST_MAGIC_V2
3166 /* Hold object_updates sending to the remote OUT in single RPC */
3167 struct object_update_request {
3169 __u16 ourq_count; /* number of ourq_updates[] */
3171 struct object_update ourq_updates[0];
3174 #define OUT_UPDATE_HEADER_MAGIC 0xBDDF0001
3175 #define OUT_UPDATE_MAX_INLINE_SIZE 4096
3176 /* Header for updates request between MDTs */
3177 struct out_update_header {
3180 __u32 ouh_inline_length;
3181 __u32 ouh_reply_size;
3182 __u32 ouh_inline_data[0];
3185 struct out_update_buffer {
3190 /* the result of object update */
3191 struct object_update_result {
3198 #define UPDATE_REPLY_MAGIC_V1 0x00BD0001
3199 #define UPDATE_REPLY_MAGIC_V2 0x00BD0002
3200 #define UPDATE_REPLY_MAGIC UPDATE_REPLY_MAGIC_V2
3201 /* Hold object_update_results being replied from the remote OUT. */
3202 struct object_update_reply {
3209 /* read update result */
3210 struct out_read_reply {
3217 /** layout swap request structure
3218 * fid1 and fid2 are in mdt_body
3220 struct mdc_swap_layouts {
3222 } __attribute__((packed));
3225 struct lustre_handle cd_handle;
3226 struct lu_fid cd_fid;
3227 __u64 cd_data_version;
3228 __u64 cd_reserved[8];
3231 /* Update llog format */
3233 struct lu_fid uop_fid;
3235 __u16 uop_param_count;
3236 __u16 uop_params_off[0];
3240 struct update_op uops_op[0];
3243 struct update_params {
3244 struct object_update_param up_params[0];
3247 enum update_records_flag {
3248 UPDATE_RECORD_CONTINUE = 1 >> 0,
3251 * This is the update record format used to store the updates in
3252 * disk. All updates of the operation will be stored in ur_ops.
3253 * All of parameters for updates of the operation will be stored
3255 * To save the space of the record, parameters in ur_ops will only
3256 * remember their offset in ur_params, so to avoid storing duplicate
3257 * parameters in ur_params, which can help us save a lot space for
3258 * operation like creating striped directory.
3260 struct update_records {
3261 __u64 ur_master_transno;
3264 /* If the operation includes multiple updates, then ur_index
3265 * means the index of the update inside the whole updates. */
3267 __u32 ur_update_count;
3268 __u32 ur_param_count;
3269 struct update_ops ur_ops;
3270 /* Note ur_ops has a variable size, so comment out
3271 * the following ur_params, in case some use it directly
3272 * update_records->ur_params
3274 * struct update_params ur_params;
3278 struct llog_update_record {
3279 struct llog_rec_hdr lur_hdr;
3280 struct update_records lur_update_rec;
3281 /* Note ur_update_rec has a variable size, so comment out
3282 * the following ur_tail, in case someone use it directly
3284 * struct llog_rec_tail lur_tail;
3288 /* nodemap records, uses 32 byte record length */
3289 #define LUSTRE_NODEMAP_NAME_LENGTH 16
3290 struct nodemap_cluster_rec {
3291 char ncr_name[LUSTRE_NODEMAP_NAME_LENGTH + 1];
3295 __u32 ncr_squash_uid;
3296 __u32 ncr_squash_gid;
3299 /* lnet_nid_t is 8 bytes */
3300 struct nodemap_range_rec {
3301 lnet_nid_t nrr_start_nid;
3302 lnet_nid_t nrr_end_nid;
3307 struct nodemap_id_rec {
3315 struct nodemap_global_rec {
3326 struct nodemap_cluster_rec ncr;
3327 struct nodemap_range_rec nrr;
3328 struct nodemap_id_rec nir;
3329 struct nodemap_global_rec ngr;
3332 /* This is the lu_ladvise struct which goes out on the wire.
3333 * Corresponds to the userspace arg llapi_lu_ladvise.
3334 * value[1-4] are unspecified fields, used differently by different advices */
3336 __u16 lla_advice; /* advice type */
3337 __u16 lla_value1; /* values for different advice types */
3339 __u64 lla_start; /* first byte of extent for advice */
3340 __u64 lla_end; /* last byte of extent for advice */
3345 /* This is the ladvise_hdr which goes on the wire, corresponds to the userspace
3346 * arg llapi_ladvise_hdr.
3347 * value[1-3] are unspecified fields, used differently by different advices */
3348 struct ladvise_hdr {
3349 __u32 lah_magic; /* LADVISE_MAGIC */
3350 __u32 lah_count; /* number of advices */
3351 __u64 lah_flags; /* from enum ladvise_flag */
3352 __u32 lah_value1; /* unused */
3353 __u32 lah_value2; /* unused */
3354 __u64 lah_value3; /* unused */
3355 struct lu_ladvise lah_advise[0]; /* advices in this header */