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 <libcfs/libcfs.h>
76 #include <lnet/types.h>
77 #include <lustre/lustre_user.h> /* Defn's shared with user-space. */
78 #include <lustre/lustre_errno.h>
79 #include <lustre_ver.h>
84 /* FOO_REQUEST_PORTAL is for incoming requests on the FOO
85 * FOO_REPLY_PORTAL is for incoming replies on the FOO
86 * FOO_BULK_PORTAL is for incoming bulk on the FOO
89 #define CONNMGR_REQUEST_PORTAL 1
90 #define CONNMGR_REPLY_PORTAL 2
91 //#define OSC_REQUEST_PORTAL 3
92 #define OSC_REPLY_PORTAL 4
93 //#define OSC_BULK_PORTAL 5
94 #define OST_IO_PORTAL 6
95 #define OST_CREATE_PORTAL 7
96 #define OST_BULK_PORTAL 8
97 //#define MDC_REQUEST_PORTAL 9
98 #define MDC_REPLY_PORTAL 10
99 //#define MDC_BULK_PORTAL 11
100 #define MDS_REQUEST_PORTAL 12
101 //#define MDS_REPLY_PORTAL 13
102 #define MDS_BULK_PORTAL 14
103 #define LDLM_CB_REQUEST_PORTAL 15
104 #define LDLM_CB_REPLY_PORTAL 16
105 #define LDLM_CANCEL_REQUEST_PORTAL 17
106 #define LDLM_CANCEL_REPLY_PORTAL 18
107 //#define PTLBD_REQUEST_PORTAL 19
108 //#define PTLBD_REPLY_PORTAL 20
109 //#define PTLBD_BULK_PORTAL 21
110 #define MDS_SETATTR_PORTAL 22
111 #define MDS_READPAGE_PORTAL 23
112 #define OUT_PORTAL 24
113 #define MGC_REPLY_PORTAL 25
114 #define MGS_REQUEST_PORTAL 26
115 #define MGS_REPLY_PORTAL 27
116 #define OST_REQUEST_PORTAL 28
117 #define FLD_REQUEST_PORTAL 29
118 #define SEQ_METADATA_PORTAL 30
119 #define SEQ_DATA_PORTAL 31
120 #define SEQ_CONTROLLER_PORTAL 32
121 #define MGS_BULK_PORTAL 33
123 /* Portal 63 is reserved for the Cray Inc DVS - nic@cray.com, roe@cray.com, n8851@cray.com */
126 #define PTL_RPC_MSG_REQUEST 4711
127 #define PTL_RPC_MSG_ERR 4712
128 #define PTL_RPC_MSG_REPLY 4713
130 /* DON'T use swabbed values of MAGIC as magic! */
131 #define LUSTRE_MSG_MAGIC_V2 0x0BD00BD3
132 #define LUSTRE_MSG_MAGIC_V2_SWABBED 0xD30BD00B
134 #define LUSTRE_MSG_MAGIC LUSTRE_MSG_MAGIC_V2
136 #define PTLRPC_MSG_VERSION 0x00000003
137 #define LUSTRE_VERSION_MASK 0xffff0000
138 #define LUSTRE_OBD_VERSION 0x00010000
139 #define LUSTRE_MDS_VERSION 0x00020000
140 #define LUSTRE_OST_VERSION 0x00030000
141 #define LUSTRE_DLM_VERSION 0x00040000
142 #define LUSTRE_LOG_VERSION 0x00050000
143 #define LUSTRE_MGS_VERSION 0x00060000
146 * Describes a range of sequence, lsr_start is included but lsr_end is
148 * Same structure is used in fld module where lsr_index field holds mdt id
151 struct lu_seq_range {
158 struct lu_seq_range_array {
161 struct lu_seq_range lsra_lsr[0];
164 #define LU_SEQ_RANGE_MDT 0x0
165 #define LU_SEQ_RANGE_OST 0x1
166 #define LU_SEQ_RANGE_ANY 0x3
168 #define LU_SEQ_RANGE_MASK 0x3
170 /** \defgroup lu_fid lu_fid
174 * Flags for lustre_mdt_attrs::lma_compat and lustre_mdt_attrs::lma_incompat.
175 * Deprecated since HSM and SOM attributes are now stored in separate on-disk
179 LMAC_HSM = 0x00000001,
180 /* LMAC_SOM = 0x00000002, obsolete since 2.8.0 */
181 LMAC_NOT_IN_OI = 0x00000004, /* the object does NOT need OI mapping */
182 LMAC_FID_ON_OST = 0x00000008, /* For OST-object, its OI mapping is
183 * under /O/<seq>/d<x>. */
187 * Masks for all features that should be supported by a Lustre version to
188 * access a specific file.
189 * This information is stored in lustre_mdt_attrs::lma_incompat.
192 LMAI_RELEASED = 0x00000001, /* file is released */
193 LMAI_AGENT = 0x00000002, /* agent inode */
194 LMAI_REMOTE_PARENT = 0x00000004, /* the parent of the object
195 is on the remote MDT */
196 LMAI_STRIPED = 0x00000008, /* striped directory inode */
197 LMAI_ORPHAN = 0x00000010, /* inode is orphan */
198 LMA_INCOMPAT_SUPP = (LMAI_AGENT | LMAI_REMOTE_PARENT | \
199 LMAI_STRIPED | LMAI_ORPHAN)
202 extern void lustre_lma_swab(struct lustre_mdt_attrs *lma);
203 extern void lustre_lma_init(struct lustre_mdt_attrs *lma,
204 const struct lu_fid *fid,
205 __u32 compat, __u32 incompat);
207 /* copytool uses a 32b bitmask field to encode archive-Ids during register
209 * archive num = 0 => all
210 * archive num from 1 to 32
212 #define LL_HSM_MAX_ARCHIVE (sizeof(__u32) * 8)
215 * HSM on-disk attributes stored in a separate xattr.
218 /** Bitfield for supported data in this structure. For future use. */
221 /** HSM flags, see hsm_flags enum below */
223 /** backend archive id associated with the file */
225 /** version associated with the last archiving, if any */
228 extern void lustre_hsm_swab(struct hsm_attrs *attrs);
234 /** LASTID file has zero OID */
235 LUSTRE_FID_LASTID_OID = 0UL,
236 /** initial fid id value */
237 LUSTRE_FID_INIT_OID = 1UL
240 /** returns fid object sequence */
241 static inline __u64 fid_seq(const struct lu_fid *fid)
246 /** returns fid object id */
247 static inline __u32 fid_oid(const struct lu_fid *fid)
252 /** returns fid object version */
253 static inline __u32 fid_ver(const struct lu_fid *fid)
258 static inline void fid_zero(struct lu_fid *fid)
260 memset(fid, 0, sizeof(*fid));
263 static inline __u64 fid_ver_oid(const struct lu_fid *fid)
265 return ((__u64)fid_ver(fid) << 32 | fid_oid(fid));
269 * Note that reserved SEQ numbers below 12 will conflict with ldiskfs
270 * inodes in the IGIF namespace, so these reserved SEQ numbers can be
271 * used for other purposes and not risk collisions with existing inodes.
273 * Different FID Format
274 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs#NEW.0
277 FID_SEQ_OST_MDT0 = 0,
278 FID_SEQ_LLOG = 1, /* unnamed llogs */
280 FID_SEQ_UNUSED_START = 3,
281 FID_SEQ_UNUSED_END = 9,
282 FID_SEQ_LLOG_NAME = 10, /* named llogs */
285 FID_SEQ_IGIF_MAX = 0x0ffffffffULL,
286 FID_SEQ_IDIF = 0x100000000ULL,
287 FID_SEQ_IDIF_MAX = 0x1ffffffffULL,
288 /* Normal FID sequence starts from this value, i.e. 1<<33 */
289 FID_SEQ_START = 0x200000000ULL,
290 /* sequence for local pre-defined FIDs listed in local_oid */
291 FID_SEQ_LOCAL_FILE = 0x200000001ULL,
292 FID_SEQ_DOT_LUSTRE = 0x200000002ULL,
293 /* sequence is used for local named objects FIDs generated
294 * by local_object_storage library */
295 FID_SEQ_LOCAL_NAME = 0x200000003ULL,
296 /* Because current FLD will only cache the fid sequence, instead
297 * of oid on the client side, if the FID needs to be exposed to
298 * clients sides, it needs to make sure all of fids under one
299 * sequence will be located in one MDT. */
300 FID_SEQ_SPECIAL = 0x200000004ULL,
301 FID_SEQ_QUOTA = 0x200000005ULL,
302 FID_SEQ_QUOTA_GLB = 0x200000006ULL,
303 FID_SEQ_ROOT = 0x200000007ULL, /* Located on MDT0 */
304 FID_SEQ_LAYOUT_RBTREE = 0x200000008ULL,
305 /* sequence is used for update logs of cross-MDT operation */
306 FID_SEQ_UPDATE_LOG = 0x200000009ULL,
307 /* Sequence is used for the directory under which update logs
309 FID_SEQ_UPDATE_LOG_DIR = 0x20000000aULL,
310 FID_SEQ_NORMAL = 0x200000400ULL,
311 FID_SEQ_LOV_DEFAULT = 0xffffffffffffffffULL
314 #define OBIF_OID_MAX_BITS 32
315 #define OBIF_MAX_OID (1ULL << OBIF_OID_MAX_BITS)
316 #define OBIF_OID_MASK ((1ULL << OBIF_OID_MAX_BITS) - 1)
317 #define IDIF_OID_MAX_BITS 48
318 #define IDIF_MAX_OID (1ULL << IDIF_OID_MAX_BITS)
319 #define IDIF_OID_MASK ((1ULL << IDIF_OID_MAX_BITS) - 1)
321 /** OID for FID_SEQ_SPECIAL */
323 /* Big Filesystem Lock to serialize rename operations */
324 FID_OID_SPECIAL_BFL = 1UL,
327 /** OID for FID_SEQ_DOT_LUSTRE */
328 enum dot_lustre_oid {
329 FID_OID_DOT_LUSTRE = 1UL,
330 FID_OID_DOT_LUSTRE_OBF = 2UL,
331 FID_OID_DOT_LUSTRE_LPF = 3UL,
334 /** OID for FID_SEQ_ROOT */
337 FID_OID_ECHO_ROOT = 2UL,
340 static inline bool fid_seq_is_mdt0(__u64 seq)
342 return seq == FID_SEQ_OST_MDT0;
345 static inline bool fid_seq_is_mdt(__u64 seq)
347 return seq == FID_SEQ_OST_MDT0 || seq >= FID_SEQ_NORMAL;
350 static inline bool fid_seq_is_echo(__u64 seq)
352 return seq == FID_SEQ_ECHO;
355 static inline bool fid_is_echo(const struct lu_fid *fid)
357 return fid_seq_is_echo(fid_seq(fid));
360 static inline bool fid_seq_is_llog(__u64 seq)
362 return seq == FID_SEQ_LLOG;
365 static inline bool fid_is_llog(const struct lu_fid *fid)
367 /* file with OID == 0 is not llog but contains last oid */
368 return fid_seq_is_llog(fid_seq(fid)) && fid_oid(fid) > 0;
371 static inline bool fid_seq_is_rsvd(__u64 seq)
373 return seq > FID_SEQ_OST_MDT0 && seq <= FID_SEQ_RSVD;
376 static inline bool fid_seq_is_special(__u64 seq)
378 return seq == FID_SEQ_SPECIAL;
381 static inline bool fid_seq_is_local_file(__u64 seq)
383 return seq == FID_SEQ_LOCAL_FILE ||
384 seq == FID_SEQ_LOCAL_NAME;
387 static inline bool fid_seq_is_root(__u64 seq)
389 return seq == FID_SEQ_ROOT;
392 static inline bool fid_seq_is_dot(__u64 seq)
394 return seq == FID_SEQ_DOT_LUSTRE;
397 static inline bool fid_seq_is_default(__u64 seq)
399 return seq == FID_SEQ_LOV_DEFAULT;
402 static inline bool fid_is_mdt0(const struct lu_fid *fid)
404 return fid_seq_is_mdt0(fid_seq(fid));
407 static inline void lu_root_fid(struct lu_fid *fid)
409 fid->f_seq = FID_SEQ_ROOT;
410 fid->f_oid = FID_OID_ROOT;
414 static inline void lu_echo_root_fid(struct lu_fid *fid)
416 fid->f_seq = FID_SEQ_ROOT;
417 fid->f_oid = FID_OID_ECHO_ROOT;
421 static inline void lu_update_log_fid(struct lu_fid *fid, __u32 index)
423 fid->f_seq = FID_SEQ_UPDATE_LOG;
428 static inline void lu_update_log_dir_fid(struct lu_fid *fid, __u32 index)
430 fid->f_seq = FID_SEQ_UPDATE_LOG_DIR;
436 * Check if a fid is igif or not.
437 * \param fid the fid to be tested.
438 * \return true if the fid is an igif; otherwise false.
440 static inline bool fid_seq_is_igif(__u64 seq)
442 return seq >= FID_SEQ_IGIF && seq <= FID_SEQ_IGIF_MAX;
445 static inline bool fid_is_igif(const struct lu_fid *fid)
447 return fid_seq_is_igif(fid_seq(fid));
451 * Check if a fid is idif or not.
452 * \param fid the fid to be tested.
453 * \return true if the fid is an idif; otherwise false.
455 static inline bool fid_seq_is_idif(__u64 seq)
457 return seq >= FID_SEQ_IDIF && seq <= FID_SEQ_IDIF_MAX;
460 static inline bool fid_is_idif(const struct lu_fid *fid)
462 return fid_seq_is_idif(fid_seq(fid));
465 static inline bool fid_is_local_file(const struct lu_fid *fid)
467 return fid_seq_is_local_file(fid_seq(fid));
470 static inline bool fid_seq_is_norm(__u64 seq)
472 return (seq >= FID_SEQ_NORMAL);
475 static inline bool fid_is_norm(const struct lu_fid *fid)
477 return fid_seq_is_norm(fid_seq(fid));
480 static inline int fid_is_layout_rbtree(const struct lu_fid *fid)
482 return fid_seq(fid) == FID_SEQ_LAYOUT_RBTREE;
485 static inline bool fid_seq_is_update_log(__u64 seq)
487 return seq == FID_SEQ_UPDATE_LOG;
490 static inline bool fid_is_update_log(const struct lu_fid *fid)
492 return fid_seq_is_update_log(fid_seq(fid));
495 static inline bool fid_seq_is_update_log_dir(__u64 seq)
497 return seq == FID_SEQ_UPDATE_LOG_DIR;
500 static inline bool fid_is_update_log_dir(const struct lu_fid *fid)
502 return fid_seq_is_update_log_dir(fid_seq(fid));
505 /* convert an OST objid into an IDIF FID SEQ number */
506 static inline __u64 fid_idif_seq(__u64 id, __u32 ost_idx)
508 return FID_SEQ_IDIF | (ost_idx << 16) | ((id >> 32) & 0xffff);
511 /* convert a packed IDIF FID into an OST objid */
512 static inline __u64 fid_idif_id(__u64 seq, __u32 oid, __u32 ver)
514 return ((__u64)ver << 48) | ((seq & 0xffff) << 32) | oid;
517 static inline __u32 idif_ost_idx(__u64 seq)
519 return (seq >> 16) & 0xffff;
522 /* extract ost index from IDIF FID */
523 static inline __u32 fid_idif_ost_idx(const struct lu_fid *fid)
525 return idif_ost_idx(fid_seq(fid));
528 /* extract OST sequence (group) from a wire ost_id (id/seq) pair */
529 static inline __u64 ostid_seq(const struct ost_id *ostid)
531 if (fid_seq_is_mdt0(ostid->oi.oi_seq))
532 return FID_SEQ_OST_MDT0;
534 if (unlikely(fid_seq_is_default(ostid->oi.oi_seq)))
535 return FID_SEQ_LOV_DEFAULT;
537 if (fid_is_idif(&ostid->oi_fid))
538 return FID_SEQ_OST_MDT0;
540 return fid_seq(&ostid->oi_fid);
543 /* extract OST objid from a wire ost_id (id/seq) pair */
544 static inline __u64 ostid_id(const struct ost_id *ostid)
546 if (fid_seq_is_mdt0(ostid->oi.oi_seq))
547 return ostid->oi.oi_id & IDIF_OID_MASK;
549 if (unlikely(fid_seq_is_default(ostid->oi.oi_seq)))
550 return ostid->oi.oi_id;
552 if (fid_is_idif(&ostid->oi_fid))
553 return fid_idif_id(fid_seq(&ostid->oi_fid),
554 fid_oid(&ostid->oi_fid), 0);
556 return fid_oid(&ostid->oi_fid);
559 static inline void ostid_set_seq(struct ost_id *oi, __u64 seq)
561 if (fid_seq_is_mdt0(seq) || fid_seq_is_default(seq)) {
564 oi->oi_fid.f_seq = seq;
565 /* Note: if f_oid + f_ver is zero, we need init it
566 * to be 1, otherwise, ostid_seq will treat this
567 * as old ostid (oi_seq == 0) */
568 if (oi->oi_fid.f_oid == 0 && oi->oi_fid.f_ver == 0)
569 oi->oi_fid.f_oid = LUSTRE_FID_INIT_OID;
573 static inline void ostid_set_seq_mdt0(struct ost_id *oi)
575 ostid_set_seq(oi, FID_SEQ_OST_MDT0);
578 static inline void ostid_set_seq_echo(struct ost_id *oi)
580 ostid_set_seq(oi, FID_SEQ_ECHO);
583 static inline void ostid_set_seq_llog(struct ost_id *oi)
585 ostid_set_seq(oi, FID_SEQ_LLOG);
589 * Note: we need check oi_seq to decide where to set oi_id,
590 * so oi_seq should always be set ahead of oi_id.
592 static inline void ostid_set_id(struct ost_id *oi, __u64 oid)
594 if (fid_seq_is_mdt0(oi->oi.oi_seq)) {
595 if (oid >= IDIF_MAX_OID) {
596 CERROR("Too large OID %#llx to set MDT0 "DOSTID"\n",
597 (unsigned long long)oid, POSTID(oi));
601 } else if (fid_is_idif(&oi->oi_fid)) {
602 if (oid >= IDIF_MAX_OID) {
603 CERROR("Too large OID %#llx to set IDIF "DOSTID"\n",
604 (unsigned long long)oid, POSTID(oi));
607 oi->oi_fid.f_seq = fid_idif_seq(oid,
608 fid_idif_ost_idx(&oi->oi_fid));
609 oi->oi_fid.f_oid = oid;
610 oi->oi_fid.f_ver = oid >> 48;
612 if (oid > OBIF_MAX_OID) {
613 CERROR("Too large oid %#llx to set REG "DOSTID"\n",
614 (unsigned long long)oid, POSTID(oi));
617 oi->oi_fid.f_oid = oid;
621 static inline int fid_set_id(struct lu_fid *fid, __u64 oid)
623 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
624 CERROR("bad IGIF, "DFID"\n", PFID(fid));
628 if (fid_is_idif(fid)) {
629 if (oid >= IDIF_MAX_OID) {
630 CERROR("Too large OID %#llx to set IDIF "DFID"\n",
631 (unsigned long long)oid, PFID(fid));
634 fid->f_seq = fid_idif_seq(oid, fid_idif_ost_idx(fid));
636 fid->f_ver = oid >> 48;
638 if (oid > OBIF_MAX_OID) {
639 CERROR("Too large OID %#llx to set REG "DFID"\n",
640 (unsigned long long)oid, PFID(fid));
649 * Unpack an OST object id/seq (group) into a FID. This is needed for
650 * converting all obdo, lmm, lsm, etc. 64-bit id/seq pairs into proper
651 * FIDs. Note that if an id/seq is already in FID/IDIF format it will
652 * be passed through unchanged. Only legacy OST objects in "group 0"
653 * will be mapped into the IDIF namespace so that they can fit into the
654 * struct lu_fid fields without loss. For reference see:
655 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs
657 static inline int ostid_to_fid(struct lu_fid *fid, const struct ost_id *ostid,
660 __u64 seq = ostid_seq(ostid);
662 if (ost_idx > 0xffff) {
663 CERROR("bad ost_idx, "DOSTID" ost_idx:%u\n", POSTID(ostid),
668 if (fid_seq_is_mdt0(seq)) {
669 __u64 oid = ostid_id(ostid);
671 /* This is a "legacy" (old 1.x/2.early) OST object in "group 0"
672 * that we map into the IDIF namespace. It allows up to 2^48
673 * objects per OST, as this is the object namespace that has
674 * been in production for years. This can handle create rates
675 * of 1M objects/s/OST for 9 years, or combinations thereof. */
676 if (oid >= IDIF_MAX_OID) {
677 CERROR("bad MDT0 id(1), "DOSTID" ost_idx:%u\n",
678 POSTID(ostid), ost_idx);
681 fid->f_seq = fid_idif_seq(oid, ost_idx);
682 /* truncate to 32 bits by assignment */
684 /* in theory, not currently used */
685 fid->f_ver = oid >> 48;
686 } else if (likely(!fid_seq_is_default(seq)))
687 /* if (fid_seq_is_idif(seq) || fid_seq_is_norm(seq)) */ {
688 /* This is either an IDIF object, which identifies objects
689 * across all OSTs, or a regular FID. The IDIF namespace maps
690 * legacy OST objects into the FID namespace. In both cases,
691 * we just pass the FID through, no conversion needed. */
692 if (ostid->oi_fid.f_ver != 0) {
693 CERROR("bad MDT0 id(2), "DOSTID" ost_idx:%u\n",
694 POSTID(ostid), ost_idx);
697 *fid = ostid->oi_fid;
703 /* pack any OST FID into an ostid (id/seq) for the wire/disk */
704 static inline int fid_to_ostid(const struct lu_fid *fid, struct ost_id *ostid)
706 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
707 CERROR("bad IGIF, "DFID"\n", PFID(fid));
711 if (fid_is_idif(fid)) {
712 ostid_set_seq_mdt0(ostid);
713 ostid_set_id(ostid, fid_idif_id(fid_seq(fid), fid_oid(fid),
716 ostid->oi_fid = *fid;
722 /* Check whether the fid is for LAST_ID */
723 static inline bool fid_is_last_id(const struct lu_fid *fid)
725 return fid_oid(fid) == 0 && fid_seq(fid) != FID_SEQ_UPDATE_LOG &&
726 fid_seq(fid) != FID_SEQ_UPDATE_LOG_DIR;
730 * Get inode number from an igif.
731 * \param fid an igif to get inode number from.
732 * \return inode number for the igif.
734 static inline ino_t lu_igif_ino(const struct lu_fid *fid)
740 * Get inode generation from an igif.
741 * \param fid an igif to get inode generation from.
742 * \return inode generation for the igif.
744 static inline __u32 lu_igif_gen(const struct lu_fid *fid)
750 * Build igif from the inode number/generation.
752 static inline void lu_igif_build(struct lu_fid *fid, __u32 ino, __u32 gen)
760 * Fids are transmitted across network (in the sender byte-ordering),
761 * and stored on disk in big-endian order.
763 static inline void fid_cpu_to_le(struct lu_fid *dst, const struct lu_fid *src)
765 dst->f_seq = __cpu_to_le64(fid_seq(src));
766 dst->f_oid = __cpu_to_le32(fid_oid(src));
767 dst->f_ver = __cpu_to_le32(fid_ver(src));
770 static inline void fid_le_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
772 dst->f_seq = __le64_to_cpu(fid_seq(src));
773 dst->f_oid = __le32_to_cpu(fid_oid(src));
774 dst->f_ver = __le32_to_cpu(fid_ver(src));
777 static inline void fid_cpu_to_be(struct lu_fid *dst, const struct lu_fid *src)
779 dst->f_seq = __cpu_to_be64(fid_seq(src));
780 dst->f_oid = __cpu_to_be32(fid_oid(src));
781 dst->f_ver = __cpu_to_be32(fid_ver(src));
784 static inline void fid_be_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
786 dst->f_seq = __be64_to_cpu(fid_seq(src));
787 dst->f_oid = __be32_to_cpu(fid_oid(src));
788 dst->f_ver = __be32_to_cpu(fid_ver(src));
791 static inline bool fid_is_sane(const struct lu_fid *fid)
793 return fid != NULL &&
794 ((fid_seq(fid) >= FID_SEQ_START && fid_ver(fid) == 0) ||
795 fid_is_igif(fid) || fid_is_idif(fid) ||
796 fid_seq_is_rsvd(fid_seq(fid)));
799 static inline bool lu_fid_eq(const struct lu_fid *f0, const struct lu_fid *f1)
801 return memcmp(f0, f1, sizeof *f0) == 0;
804 #define __diff_normalize(val0, val1) \
806 typeof(val0) __val0 = (val0); \
807 typeof(val1) __val1 = (val1); \
809 (__val0 == __val1 ? 0 : __val0 > __val1 ? +1 : -1); \
812 static inline int lu_fid_cmp(const struct lu_fid *f0,
813 const struct lu_fid *f1)
816 __diff_normalize(fid_seq(f0), fid_seq(f1)) ?:
817 __diff_normalize(fid_oid(f0), fid_oid(f1)) ?:
818 __diff_normalize(fid_ver(f0), fid_ver(f1));
821 static inline void ostid_cpu_to_le(const struct ost_id *src_oi,
822 struct ost_id *dst_oi)
824 if (fid_seq_is_mdt0(src_oi->oi.oi_seq)) {
825 dst_oi->oi.oi_id = __cpu_to_le64(src_oi->oi.oi_id);
826 dst_oi->oi.oi_seq = __cpu_to_le64(src_oi->oi.oi_seq);
828 fid_cpu_to_le(&dst_oi->oi_fid, &src_oi->oi_fid);
832 static inline void ostid_le_to_cpu(const struct ost_id *src_oi,
833 struct ost_id *dst_oi)
835 if (fid_seq_is_mdt0(src_oi->oi.oi_seq)) {
836 dst_oi->oi.oi_id = __le64_to_cpu(src_oi->oi.oi_id);
837 dst_oi->oi.oi_seq = __le64_to_cpu(src_oi->oi.oi_seq);
839 fid_le_to_cpu(&dst_oi->oi_fid, &src_oi->oi_fid);
843 struct lu_orphan_rec {
844 /* The MDT-object's FID referenced by the orphan OST-object */
845 struct lu_fid lor_fid;
850 struct lu_orphan_ent {
851 /* The orphan OST-object's FID */
852 struct lu_fid loe_key;
853 struct lu_orphan_rec loe_rec;
858 /** \defgroup lu_dir lu_dir
862 * Enumeration of possible directory entry attributes.
864 * Attributes follow directory entry header in the order they appear in this
867 enum lu_dirent_attrs {
870 LUDA_64BITHASH = 0x0004,
872 /* The following attrs are used for MDT internal only,
873 * not visible to client */
875 /* Something in the record is unknown, to be verified in further. */
876 LUDA_UNKNOWN = 0x0400,
877 /* Ignore this record, go to next directly. */
878 LUDA_IGNORE = 0x0800,
879 /* The system is upgraded, has beed or to be repaired (dryrun). */
880 LUDA_UPGRADE = 0x1000,
881 /* The dirent has been repaired, or to be repaired (dryrun). */
882 LUDA_REPAIR = 0x2000,
883 /* Only check but not repair the dirent inconsistency */
884 LUDA_VERIFY_DRYRUN = 0x4000,
885 /* Verify the dirent consistency */
886 LUDA_VERIFY = 0x8000,
889 #define LU_DIRENT_ATTRS_MASK 0xff00
892 * Layout of readdir pages, as transmitted on wire.
895 /** valid if LUDA_FID is set. */
896 struct lu_fid lde_fid;
897 /** a unique entry identifier: a hash or an offset. */
899 /** total record length, including all attributes. */
903 /** optional variable size attributes following this entry.
904 * taken from enum lu_dirent_attrs.
907 /** name is followed by the attributes indicated in ->ldp_attrs, in
908 * their natural order. After the last attribute, padding bytes are
909 * added to make ->lde_reclen a multiple of 8.
915 * Definitions of optional directory entry attributes formats.
917 * Individual attributes do not have their length encoded in a generic way. It
918 * is assumed that consumer of an attribute knows its format. This means that
919 * it is impossible to skip over an unknown attribute, except by skipping over all
920 * remaining attributes (by using ->lde_reclen), which is not too
921 * constraining, because new server versions will append new attributes at
922 * the end of an entry.
926 * Fid directory attribute: a fid of an object referenced by the entry. This
927 * will be almost always requested by the client and supplied by the server.
929 * Aligned to 8 bytes.
931 /* To have compatibility with 1.8, lets have fid in lu_dirent struct. */
936 * Aligned to 2 bytes.
943 __u64 ldp_hash_start;
947 struct lu_dirent ldp_entries[0];
950 enum lu_dirpage_flags {
952 * dirpage contains no entry.
956 * last entry's lde_hash equals ldp_hash_end.
961 static inline struct lu_dirent *lu_dirent_start(struct lu_dirpage *dp)
963 if (__le32_to_cpu(dp->ldp_flags) & LDF_EMPTY)
966 return dp->ldp_entries;
969 static inline struct lu_dirent *lu_dirent_next(struct lu_dirent *ent)
971 struct lu_dirent *next;
973 if (__le16_to_cpu(ent->lde_reclen) != 0)
974 next = ((void *)ent) + __le16_to_cpu(ent->lde_reclen);
981 static inline size_t lu_dirent_calc_size(size_t namelen, __u16 attr)
985 if (attr & LUDA_TYPE) {
986 const size_t align = sizeof(struct luda_type) - 1;
987 size = (sizeof(struct lu_dirent) + namelen + align) & ~align;
988 size += sizeof(struct luda_type);
990 size = sizeof(struct lu_dirent) + namelen;
992 return (size + 7) & ~7;
995 #define MDS_DIR_END_OFF 0xfffffffffffffffeULL
998 * MDS_READPAGE page size
1000 * This is the directory page size packed in MDS_READPAGE RPC.
1001 * It's different than PAGE_SIZE because the client needs to
1002 * access the struct lu_dirpage header packed at the beginning of
1003 * the "page" and without this there isn't any way to know find the
1004 * lu_dirpage header is if client and server PAGE_SIZE differ.
1006 #define LU_PAGE_SHIFT 12
1007 #define LU_PAGE_SIZE (1UL << LU_PAGE_SHIFT)
1008 #define LU_PAGE_MASK (~(LU_PAGE_SIZE - 1))
1010 #define LU_PAGE_COUNT (1 << (PAGE_SHIFT - LU_PAGE_SHIFT))
1014 struct lustre_handle {
1017 #define DEAD_HANDLE_MAGIC 0xdeadbeefcafebabeULL
1019 static inline bool lustre_handle_is_used(const struct lustre_handle *lh)
1021 return lh->cookie != 0;
1024 static inline bool lustre_handle_equal(const struct lustre_handle *lh1,
1025 const struct lustre_handle *lh2)
1027 return lh1->cookie == lh2->cookie;
1030 static inline void lustre_handle_copy(struct lustre_handle *tgt,
1031 const struct lustre_handle *src)
1033 tgt->cookie = src->cookie;
1036 struct lustre_handle_array {
1038 struct lustre_handle handles[0];
1041 /* flags for lm_flags */
1042 #define MSGHDR_AT_SUPPORT 0x1
1043 #define MSGHDR_CKSUM_INCOMPAT18 0x2
1045 #define lustre_msg lustre_msg_v2
1046 /* we depend on this structure to be 8-byte aligned */
1047 /* this type is only endian-adjusted in lustre_unpack_msg() */
1048 struct lustre_msg_v2 {
1057 __u32 lm_buflens[0];
1060 /* without gss, ptlrpc_body is put at the first buffer. */
1061 #define PTLRPC_NUM_VERSIONS 4
1062 struct ptlrpc_body_v3 {
1063 struct lustre_handle pb_handle;
1068 __u64 pb_last_xid; /* highest replied XID without lower unreplied XID */
1069 __u16 pb_tag; /* virtual slot idx for multiple modifying RPCs */
1072 __u64 pb_last_committed;
1077 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1078 __u32 pb_service_time; /* for rep, actual service time */
1081 /* VBR: pre-versions */
1082 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1083 __u64 pb_mbits; /**< match bits for bulk request */
1084 /* padding for future needs */
1085 __u64 pb_padding64_0;
1086 __u64 pb_padding64_1;
1087 __u64 pb_padding64_2;
1088 char pb_jobid[LUSTRE_JOBID_SIZE];
1090 #define ptlrpc_body ptlrpc_body_v3
1092 struct ptlrpc_body_v2 {
1093 struct lustre_handle pb_handle;
1098 __u64 pb_last_xid; /* highest replied XID without lower unreplied XID */
1099 __u16 pb_tag; /* virtual slot idx for multiple modifying RPCs */
1102 __u64 pb_last_committed;
1107 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1108 __u32 pb_service_time; /* for rep, actual service time, also used for
1109 net_latency of req */
1112 /* VBR: pre-versions */
1113 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1114 __u64 pb_mbits; /**< unused in V2 */
1115 /* padding for future needs */
1116 __u64 pb_padding64_0;
1117 __u64 pb_padding64_1;
1118 __u64 pb_padding64_2;
1121 /* message body offset for lustre_msg_v2 */
1122 /* ptlrpc body offset in all request/reply messages */
1123 #define MSG_PTLRPC_BODY_OFF 0
1125 /* normal request/reply message record offset */
1126 #define REQ_REC_OFF 1
1127 #define REPLY_REC_OFF 1
1129 /* ldlm request message body offset */
1130 #define DLM_LOCKREQ_OFF 1 /* lockreq offset */
1131 #define DLM_REQ_REC_OFF 2 /* normal dlm request record offset */
1133 /* ldlm intent lock message body offset */
1134 #define DLM_INTENT_IT_OFF 2 /* intent lock it offset */
1135 #define DLM_INTENT_REC_OFF 3 /* intent lock record offset */
1137 /* ldlm reply message body offset */
1138 #define DLM_LOCKREPLY_OFF 1 /* lockrep offset */
1139 #define DLM_REPLY_REC_OFF 2 /* reply record offset */
1141 /** only use in req->rq_{req,rep}_swab_mask */
1142 #define MSG_PTLRPC_HEADER_OFF 31
1144 /* Flags that are operation-specific go in the top 16 bits. */
1145 #define MSG_OP_FLAG_MASK 0xffff0000
1146 #define MSG_OP_FLAG_SHIFT 16
1148 /* Flags that apply to all requests are in the bottom 16 bits */
1149 #define MSG_GEN_FLAG_MASK 0x0000ffff
1150 #define MSG_LAST_REPLAY 0x0001
1151 #define MSG_RESENT 0x0002
1152 #define MSG_REPLAY 0x0004
1153 /* #define MSG_AT_SUPPORT 0x0008
1154 * This was used in early prototypes of adaptive timeouts, and while there
1155 * shouldn't be any users of that code there also isn't a need for using this
1156 * bits. Defer usage until at least 1.10 to avoid potential conflict. */
1157 #define MSG_DELAY_REPLAY 0x0010
1158 #define MSG_VERSION_REPLAY 0x0020
1159 #define MSG_REQ_REPLAY_DONE 0x0040
1160 #define MSG_LOCK_REPLAY_DONE 0x0080
1163 * Flags for all connect opcodes (MDS_CONNECT, OST_CONNECT)
1166 #define MSG_CONNECT_RECOVERING 0x00000001
1167 #define MSG_CONNECT_RECONNECT 0x00000002
1168 #define MSG_CONNECT_REPLAYABLE 0x00000004
1169 //#define MSG_CONNECT_PEER 0x8
1170 #define MSG_CONNECT_LIBCLIENT 0x00000010
1171 #define MSG_CONNECT_INITIAL 0x00000020
1172 #define MSG_CONNECT_ASYNC 0x00000040
1173 #define MSG_CONNECT_NEXT_VER 0x00000080 /* use next version of lustre_msg */
1174 #define MSG_CONNECT_TRANSNO 0x00000100 /* report transno */
1177 #define OBD_CONNECT_RDONLY 0x1ULL /*client has read-only access*/
1178 #define OBD_CONNECT_INDEX 0x2ULL /*connect specific LOV idx */
1179 #define OBD_CONNECT_MDS 0x4ULL /*connect from MDT to OST */
1180 #define OBD_CONNECT_GRANT 0x8ULL /*OSC gets grant at connect */
1181 #define OBD_CONNECT_SRVLOCK 0x10ULL /*server takes locks for cli */
1182 #define OBD_CONNECT_VERSION 0x20ULL /*Lustre versions in ocd */
1183 #define OBD_CONNECT_REQPORTAL 0x40ULL /*Separate non-IO req portal */
1184 #define OBD_CONNECT_ACL 0x80ULL /*access control lists */
1185 #define OBD_CONNECT_XATTR 0x100ULL /*client use extended attr */
1186 #define OBD_CONNECT_CROW 0x200ULL /*MDS+OST create obj on write*/
1187 #define OBD_CONNECT_TRUNCLOCK 0x400ULL /*locks on server for punch */
1188 #define OBD_CONNECT_TRANSNO 0x800ULL /*replay sends init transno */
1189 #define OBD_CONNECT_IBITS 0x1000ULL /*support for inodebits locks*/
1190 #define OBD_CONNECT_JOIN 0x2000ULL /*files can be concatenated.
1191 *We do not support JOIN FILE
1192 *anymore, reserve this flags
1193 *just for preventing such bit
1195 #define OBD_CONNECT_ATTRFID 0x4000ULL /*Server can GetAttr By Fid*/
1196 #define OBD_CONNECT_NODEVOH 0x8000ULL /*No open hndl on specl nodes*/
1197 #define OBD_CONNECT_RMT_CLIENT 0x10000ULL /* Remote client, never used
1198 * in production. Removed in
1199 * 2.9. Keep this flag to
1202 #define OBD_CONNECT_RMT_CLIENT_FORCE 0x20000ULL /* Remote client by force,
1203 * never used in production.
1204 * Removed in 2.9. Keep this
1205 * flag to avoid reusing.
1207 #define OBD_CONNECT_BRW_SIZE 0x40000ULL /*Max bytes per rpc */
1208 #define OBD_CONNECT_QUOTA64 0x80000ULL /*Not used since 2.4 */
1209 #define OBD_CONNECT_MDS_CAPA 0x100000ULL /*MDS capability */
1210 #define OBD_CONNECT_OSS_CAPA 0x200000ULL /*OSS capability */
1211 #define OBD_CONNECT_CANCELSET 0x400000ULL /*Early batched cancels. */
1212 #define OBD_CONNECT_SOM 0x800000ULL /*Size on MDS */
1213 #define OBD_CONNECT_AT 0x1000000ULL /*client uses AT */
1214 #define OBD_CONNECT_LRU_RESIZE 0x2000000ULL /*LRU resize feature. */
1215 #define OBD_CONNECT_MDS_MDS 0x4000000ULL /*MDS-MDS connection */
1216 #define OBD_CONNECT_REAL 0x8000000ULL /* obsolete since 2.8 */
1217 #define OBD_CONNECT_CHANGE_QS 0x10000000ULL /*Not used since 2.4 */
1218 #define OBD_CONNECT_CKSUM 0x20000000ULL /*support several cksum algos*/
1219 #define OBD_CONNECT_FID 0x40000000ULL /*FID is supported by server */
1220 #define OBD_CONNECT_VBR 0x80000000ULL /*version based recovery */
1221 #define OBD_CONNECT_LOV_V3 0x100000000ULL /*client supports LOV v3 EA */
1222 #define OBD_CONNECT_GRANT_SHRINK 0x200000000ULL /* support grant shrink */
1223 #define OBD_CONNECT_SKIP_ORPHAN 0x400000000ULL /* don't reuse orphan objids */
1224 #define OBD_CONNECT_MAX_EASIZE 0x800000000ULL /* preserved for large EA */
1225 #define OBD_CONNECT_FULL20 0x1000000000ULL /* it is 2.0 client */
1226 #define OBD_CONNECT_LAYOUTLOCK 0x2000000000ULL /* client uses layout lock */
1227 #define OBD_CONNECT_64BITHASH 0x4000000000ULL /* client supports 64-bits
1229 #define OBD_CONNECT_MAXBYTES 0x8000000000ULL /* max stripe size */
1230 #define OBD_CONNECT_IMP_RECOV 0x10000000000ULL /* imp recovery support */
1231 #define OBD_CONNECT_JOBSTATS 0x20000000000ULL /* jobid in ptlrpc_body */
1232 #define OBD_CONNECT_UMASK 0x40000000000ULL /* create uses client umask */
1233 #define OBD_CONNECT_EINPROGRESS 0x80000000000ULL /* client handles -EINPROGRESS
1234 * RPC error properly */
1235 #define OBD_CONNECT_GRANT_PARAM 0x100000000000ULL/* extra grant params used for
1236 * finer space reservation */
1237 #define OBD_CONNECT_FLOCK_OWNER 0x200000000000ULL /* for the fixed 1.8
1238 * policy and 2.x server */
1239 #define OBD_CONNECT_LVB_TYPE 0x400000000000ULL /* variable type of LVB */
1240 #define OBD_CONNECT_NANOSEC_TIME 0x800000000000ULL /* nanosecond timestamps */
1241 #define OBD_CONNECT_LIGHTWEIGHT 0x1000000000000ULL/* lightweight connection */
1242 #define OBD_CONNECT_SHORTIO 0x2000000000000ULL/* short io */
1243 #define OBD_CONNECT_PINGLESS 0x4000000000000ULL/* pings not required */
1244 #define OBD_CONNECT_FLOCK_DEAD 0x8000000000000ULL/* improved flock deadlock detection */
1245 #define OBD_CONNECT_DISP_STRIPE 0x10000000000000ULL/* create stripe disposition*/
1246 #define OBD_CONNECT_OPEN_BY_FID 0x20000000000000ULL /* open by fid won't pack
1248 #define OBD_CONNECT_LFSCK 0x40000000000000ULL/* support online LFSCK */
1249 #define OBD_CONNECT_UNLINK_CLOSE 0x100000000000000ULL/* close file in unlink */
1250 #define OBD_CONNECT_MULTIMODRPCS 0x200000000000000ULL /* support multiple modify
1252 #define OBD_CONNECT_DIR_STRIPE 0x400000000000000ULL /* striped DNE dir */
1253 #define OBD_CONNECT_SUBTREE 0x800000000000000ULL /* fileset mount */
1254 #define OBD_CONNECT_LOCK_AHEAD 0x1000000000000000ULL /* lock ahead */
1255 /** bulk matchbits is sent within ptlrpc_body */
1256 #define OBD_CONNECT_BULK_MBITS 0x2000000000000000ULL
1257 #define OBD_CONNECT_OBDOPACK 0x4000000000000000ULL /* compact OUT obdo */
1258 #define OBD_CONNECT_FLAGS2 0x8000000000000000ULL /* second flags word */
1259 /* ocd_connect_flags2 flags */
1260 #define OBD_CONNECT2_FILE_SECCTX 0x1ULL /* set file security context at create */
1263 * Please DO NOT add flag values here before first ensuring that this same
1264 * flag value is not in use on some other branch. Please clear any such
1265 * changes with senior engineers before starting to use a new flag. Then,
1266 * submit a small patch against EVERY branch that ONLY adds the new flag,
1267 * updates obd_connect_names[] for lprocfs_rd_connect_flags(), adds the
1268 * flag to check_obd_connect_data(), and updates wiretests accordingly, so it
1269 * can be approved and landed easily to reserve the flag for future use. */
1271 /* The MNE_SWAB flag is overloading the MDS_MDS bit only for the MGS
1272 * connection. It is a temporary bug fix for Imperative Recovery interop
1273 * between 2.2 and 2.3 x86/ppc nodes, and can be removed when interop for
1274 * 2.2 clients/servers is no longer needed. LU-1252/LU-1644. */
1275 #define OBD_CONNECT_MNE_SWAB OBD_CONNECT_MDS_MDS
1277 #define OCD_HAS_FLAG(ocd, flg) \
1278 (!!((ocd)->ocd_connect_flags & OBD_CONNECT_##flg))
1281 #ifdef HAVE_LRU_RESIZE_SUPPORT
1282 #define LRU_RESIZE_CONNECT_FLAG OBD_CONNECT_LRU_RESIZE
1284 #define LRU_RESIZE_CONNECT_FLAG 0
1287 #define MDT_CONNECT_SUPPORTED (OBD_CONNECT_RDONLY | OBD_CONNECT_VERSION | \
1288 OBD_CONNECT_ACL | OBD_CONNECT_XATTR | \
1289 OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH | \
1290 OBD_CONNECT_ATTRFID | OBD_CONNECT_CANCELSET | \
1291 OBD_CONNECT_AT | OBD_CONNECT_BRW_SIZE | \
1292 OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID | \
1293 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_VBR | \
1294 OBD_CONNECT_LOV_V3 | OBD_CONNECT_FULL20 | \
1295 OBD_CONNECT_64BITHASH | OBD_CONNECT_JOBSTATS | \
1296 OBD_CONNECT_EINPROGRESS | \
1297 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_UMASK | \
1298 OBD_CONNECT_LVB_TYPE | OBD_CONNECT_LAYOUTLOCK |\
1299 OBD_CONNECT_PINGLESS | OBD_CONNECT_MAX_EASIZE |\
1300 OBD_CONNECT_FLOCK_DEAD | \
1301 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK | \
1302 OBD_CONNECT_OPEN_BY_FID | \
1303 OBD_CONNECT_DIR_STRIPE | \
1304 OBD_CONNECT_BULK_MBITS | \
1305 OBD_CONNECT_MULTIMODRPCS | \
1306 OBD_CONNECT_SUBTREE | \
1309 #define MDT_CONNECT_SUPPORTED2 OBD_CONNECT2_FILE_SECCTX
1311 #define OST_CONNECT_SUPPORTED (OBD_CONNECT_SRVLOCK | OBD_CONNECT_GRANT | \
1312 OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | \
1313 OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_INDEX | \
1314 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_CANCELSET | \
1315 OBD_CONNECT_AT | LRU_RESIZE_CONNECT_FLAG | \
1316 OBD_CONNECT_CKSUM | OBD_CONNECT_VBR | \
1317 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN | \
1318 OBD_CONNECT_GRANT_SHRINK | OBD_CONNECT_FULL20 |\
1319 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES | \
1320 OBD_CONNECT_MAX_EASIZE | \
1321 OBD_CONNECT_EINPROGRESS | \
1322 OBD_CONNECT_JOBSTATS | \
1323 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LVB_TYPE|\
1324 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_FID | \
1325 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK | \
1326 OBD_CONNECT_BULK_MBITS | \
1327 OBD_CONNECT_GRANT_PARAM)
1328 #define OST_CONNECT_SUPPORTED2 0
1330 #define ECHO_CONNECT_SUPPORTED 0
1331 #define ECHO_CONNECT_SUPPORTED2 0
1333 #define MGS_CONNECT_SUPPORTED (OBD_CONNECT_VERSION | OBD_CONNECT_AT | \
1334 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV | \
1335 OBD_CONNECT_MNE_SWAB | OBD_CONNECT_PINGLESS |\
1336 OBD_CONNECT_BULK_MBITS)
1338 #define MGS_CONNECT_SUPPORTED2 0
1340 /* Features required for this version of the client to work with server */
1341 #define CLIENT_CONNECT_MDT_REQD (OBD_CONNECT_IBITS | OBD_CONNECT_FID | \
1344 /* This structure is used for both request and reply.
1346 * If we eventually have separate connect data for different types, which we
1347 * almost certainly will, then perhaps we stick a union in here. */
1348 struct obd_connect_data {
1349 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1350 __u32 ocd_version; /* lustre release version number */
1351 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1352 __u32 ocd_index; /* LOV index to connect to */
1353 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
1354 __u64 ocd_ibits_known; /* inode bits this client understands */
1355 __u8 ocd_grant_blkbits; /* log2 of the backend filesystem blocksize */
1356 __u8 ocd_grant_inobits; /* log2 of the per-inode space consumption */
1357 __u16 ocd_grant_tax_kb; /* extent insertion overhead, in 1K blocks */
1358 __u32 ocd_grant_max_blks;/* maximum number of blocks per extent */
1359 __u64 ocd_transno; /* first transno from client to be replayed */
1360 __u32 ocd_group; /* MDS group on OST */
1361 __u32 ocd_cksum_types; /* supported checksum algorithms */
1362 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1363 __u32 ocd_instance; /* instance # of this target */
1364 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1365 /* Fields after ocd_maxbytes are only accessible by the receiver
1366 * if the corresponding flag in ocd_connect_flags is set. Accessing
1367 * any field after ocd_maxbytes on the receiver without a valid flag
1368 * may result in out-of-bound memory access and kernel oops. */
1369 __u16 ocd_maxmodrpcs; /* Maximum modify RPCs in parallel */
1370 __u16 padding0; /* added 2.1.0. also fix lustre_swab_connect */
1371 __u32 padding1; /* added 2.1.0. also fix lustre_swab_connect */
1372 __u64 ocd_connect_flags2;
1373 __u64 padding3; /* added 2.1.0. also fix lustre_swab_connect */
1374 __u64 padding4; /* added 2.1.0. also fix lustre_swab_connect */
1375 __u64 padding5; /* added 2.1.0. also fix lustre_swab_connect */
1376 __u64 padding6; /* added 2.1.0. also fix lustre_swab_connect */
1377 __u64 padding7; /* added 2.1.0. also fix lustre_swab_connect */
1378 __u64 padding8; /* added 2.1.0. also fix lustre_swab_connect */
1379 __u64 padding9; /* added 2.1.0. also fix lustre_swab_connect */
1380 __u64 paddingA; /* added 2.1.0. also fix lustre_swab_connect */
1381 __u64 paddingB; /* added 2.1.0. also fix lustre_swab_connect */
1382 __u64 paddingC; /* added 2.1.0. also fix lustre_swab_connect */
1383 __u64 paddingD; /* added 2.1.0. also fix lustre_swab_connect */
1384 __u64 paddingE; /* added 2.1.0. also fix lustre_swab_connect */
1385 __u64 paddingF; /* added 2.1.0. also fix lustre_swab_connect */
1388 * Please DO NOT use any fields here before first ensuring that this same
1389 * field is not in use on some other branch. Please clear any such changes
1390 * with senior engineers before starting to use a new field. Then, submit
1391 * a small patch against EVERY branch that ONLY adds the new field along with
1392 * the matching OBD_CONNECT flag, so that can be approved and landed easily to
1393 * reserve the flag for future use. */
1396 * Supported checksum algorithms. Up to 32 checksum types are supported.
1397 * (32-bit mask stored in obd_connect_data::ocd_cksum_types)
1398 * Please update DECLARE_CKSUM_NAME/OBD_CKSUM_ALL in obd.h when adding a new
1399 * algorithm and also the OBD_FL_CKSUM* flags.
1402 OBD_CKSUM_CRC32 = 0x00000001,
1403 OBD_CKSUM_ADLER = 0x00000002,
1404 OBD_CKSUM_CRC32C= 0x00000004,
1408 * OST requests: OBDO & OBD request records
1413 OST_REPLY = 0, /* reply ? */
1429 OST_QUOTACHECK = 18, /* not used since 2.4 */
1431 OST_QUOTA_ADJUST_QUNIT = 20, /* not used since 2.4 */
1433 OST_LAST_OPC /* must be < 33 to avoid MDS_GETATTR */
1435 #define OST_FIRST_OPC OST_REPLY
1438 OBD_FL_INLINEDATA = 0x00000001,
1439 OBD_FL_OBDMDEXISTS = 0x00000002,
1440 OBD_FL_DELORPHAN = 0x00000004, /* if set in o_flags delete orphans */
1441 OBD_FL_NORPC = 0x00000008, /* set in o_flags do in OSC not OST */
1442 OBD_FL_IDONLY = 0x00000010, /* set in o_flags only adjust obj id*/
1443 OBD_FL_RECREATE_OBJS= 0x00000020, /* recreate missing obj */
1444 OBD_FL_DEBUG_CHECK = 0x00000040, /* echo client/server debug check */
1445 OBD_FL_NO_USRQUOTA = 0x00000100, /* the object's owner is over quota */
1446 OBD_FL_NO_GRPQUOTA = 0x00000200, /* the object's group is over quota */
1447 OBD_FL_CREATE_CROW = 0x00000400, /* object should be create on write */
1448 OBD_FL_SRVLOCK = 0x00000800, /* delegate DLM locking to server */
1449 OBD_FL_CKSUM_CRC32 = 0x00001000, /* CRC32 checksum type */
1450 OBD_FL_CKSUM_ADLER = 0x00002000, /* ADLER checksum type */
1451 OBD_FL_CKSUM_CRC32C = 0x00004000, /* CRC32C checksum type */
1452 OBD_FL_CKSUM_RSVD2 = 0x00008000, /* for future cksum types */
1453 OBD_FL_CKSUM_RSVD3 = 0x00010000, /* for future cksum types */
1454 OBD_FL_SHRINK_GRANT = 0x00020000, /* object shrink the grant */
1455 OBD_FL_MMAP = 0x00040000, /* object is mmapped on the client.
1456 * XXX: obsoleted - reserved for old
1457 * clients prior than 2.2 */
1458 OBD_FL_RECOV_RESEND = 0x00080000, /* recoverable resent */
1459 OBD_FL_NOSPC_BLK = 0x00100000, /* no more block space on OST */
1460 OBD_FL_FLUSH = 0x00200000, /* flush pages on the OST */
1461 OBD_FL_SHORT_IO = 0x00400000, /* short io request */
1463 /* Note that while these checksum values are currently separate bits,
1464 * in 2.x we can actually allow all values from 1-31 if we wanted. */
1465 OBD_FL_CKSUM_ALL = OBD_FL_CKSUM_CRC32 | OBD_FL_CKSUM_ADLER |
1466 OBD_FL_CKSUM_CRC32C,
1468 /* mask for local-only flag, which won't be sent over network */
1469 OBD_FL_LOCAL_MASK = 0xF0000000,
1473 * All LOV EA magics should have the same postfix, if some new version
1474 * Lustre instroduces new LOV EA magic, then when down-grade to an old
1475 * Lustre, even though the old version system does not recognizes such
1476 * new magic, it still can distinguish the corrupted cases by checking
1477 * the magic's postfix.
1479 #define LOV_MAGIC_MAGIC 0x0BD0
1480 #define LOV_MAGIC_MASK 0xFFFF
1482 #define LOV_MAGIC_V1 (0x0BD10000 | LOV_MAGIC_MAGIC)
1483 #define LOV_MAGIC_JOIN_V1 (0x0BD20000 | LOV_MAGIC_MAGIC)
1484 #define LOV_MAGIC_V3 (0x0BD30000 | LOV_MAGIC_MAGIC)
1485 #define LOV_MAGIC_MIGRATE (0x0BD40000 | LOV_MAGIC_MAGIC)
1486 /* reserved for specifying OSTs */
1487 #define LOV_MAGIC_SPECIFIC (0x0BD50000 | LOV_MAGIC_MAGIC)
1488 #define LOV_MAGIC LOV_MAGIC_V1
1491 * magic for fully defined striping
1492 * the idea is that we should have different magics for striping "hints"
1493 * (struct lov_user_md_v[13]) and defined ready-to-use striping (struct
1494 * lov_mds_md_v[13]). at the moment the magics are used in wire protocol,
1495 * we can't just change it w/o long way preparation, but we still need a
1496 * mechanism to allow LOD to differentiate hint versus ready striping.
1497 * so, at the moment we do a trick: MDT knows what to expect from request
1498 * depending on the case (replay uses ready striping, non-replay req uses
1499 * hints), so MDT replaces magic with appropriate one and now LOD can
1500 * easily understand what's inside -bzzz
1502 #define LOV_MAGIC_V1_DEF 0x0CD10BD0
1503 #define LOV_MAGIC_V3_DEF 0x0CD30BD0
1505 #define lov_pattern(pattern) (pattern & ~LOV_PATTERN_F_MASK)
1506 #define lov_pattern_flags(pattern) (pattern & LOV_PATTERN_F_MASK)
1508 #define lov_ost_data lov_ost_data_v1
1509 struct lov_ost_data_v1 { /* per-stripe data structure (little-endian)*/
1510 struct ost_id l_ost_oi; /* OST object ID */
1511 __u32 l_ost_gen; /* generation of this l_ost_idx */
1512 __u32 l_ost_idx; /* OST index in LOV (lov_tgt_desc->tgts) */
1515 #define lov_mds_md lov_mds_md_v1
1516 struct lov_mds_md_v1 { /* LOV EA mds/wire data (little-endian) */
1517 __u32 lmm_magic; /* magic number = LOV_MAGIC_V1 */
1518 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1519 struct ost_id lmm_oi; /* LOV object ID */
1520 __u32 lmm_stripe_size; /* size of stripe in bytes */
1521 /* lmm_stripe_count used to be __u32 */
1522 __u16 lmm_stripe_count; /* num stripes in use for this object */
1523 __u16 lmm_layout_gen; /* layout generation number */
1524 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1528 * Sigh, because pre-2.4 uses
1529 * struct lov_mds_md_v1 {
1531 * __u64 lmm_object_id;
1532 * __u64 lmm_object_seq;
1535 * to identify the LOV(MDT) object, and lmm_object_seq will
1536 * be normal_fid, which make it hard to combine these conversion
1537 * to ostid_to FID. so we will do lmm_oi/fid conversion separately
1539 * We can tell the lmm_oi by this way,
1540 * 1.8: lmm_object_id = {inode}, lmm_object_gr = 0
1541 * 2.1: lmm_object_id = {oid < 128k}, lmm_object_seq = FID_SEQ_NORMAL
1542 * 2.4: lmm_oi.f_seq = FID_SEQ_NORMAL, lmm_oi.f_oid = {oid < 128k},
1545 * But currently lmm_oi/lsm_oi does not have any "real" usages,
1546 * except for printing some information, and the user can always
1547 * get the real FID from LMA, besides this multiple case check might
1548 * make swab more complicate. So we will keep using id/seq for lmm_oi.
1551 static inline void fid_to_lmm_oi(const struct lu_fid *fid,
1554 oi->oi.oi_id = fid_oid(fid);
1555 oi->oi.oi_seq = fid_seq(fid);
1558 static inline void lmm_oi_set_seq(struct ost_id *oi, __u64 seq)
1560 oi->oi.oi_seq = seq;
1563 static inline void lmm_oi_set_id(struct ost_id *oi, __u64 oid)
1568 static inline __u64 lmm_oi_id(const struct ost_id *oi)
1570 return oi->oi.oi_id;
1573 static inline __u64 lmm_oi_seq(const struct ost_id *oi)
1575 return oi->oi.oi_seq;
1578 static inline void lmm_oi_le_to_cpu(struct ost_id *dst_oi,
1579 const struct ost_id *src_oi)
1581 dst_oi->oi.oi_id = __le64_to_cpu(src_oi->oi.oi_id);
1582 dst_oi->oi.oi_seq = __le64_to_cpu(src_oi->oi.oi_seq);
1585 static inline void lmm_oi_cpu_to_le(struct ost_id *dst_oi,
1586 const struct ost_id *src_oi)
1588 dst_oi->oi.oi_id = __cpu_to_le64(src_oi->oi.oi_id);
1589 dst_oi->oi.oi_seq = __cpu_to_le64(src_oi->oi.oi_seq);
1592 #define MAX_MD_SIZE (sizeof(struct lov_mds_md) + 4 * sizeof(struct lov_ost_data))
1593 #define MIN_MD_SIZE (sizeof(struct lov_mds_md) + 1 * sizeof(struct lov_ost_data))
1595 /* This is the default MDT reply size allocated, should the striping be bigger,
1596 * it will be reallocated in mdt_fix_reply.
1597 * 100 stripes is a bit less than 2.5k of data */
1598 #define DEF_REP_MD_SIZE (sizeof(struct lov_mds_md) + \
1599 100 * sizeof(struct lov_ost_data))
1601 #define XATTR_NAME_ACL_ACCESS "system.posix_acl_access"
1602 #define XATTR_NAME_ACL_DEFAULT "system.posix_acl_default"
1603 #define XATTR_USER_PREFIX "user."
1604 #define XATTR_TRUSTED_PREFIX "trusted."
1605 #define XATTR_SECURITY_PREFIX "security."
1607 #define XATTR_NAME_LOV "trusted.lov"
1608 #define XATTR_NAME_LMA "trusted.lma"
1609 #define XATTR_NAME_LMV "trusted.lmv"
1610 #define XATTR_NAME_DEFAULT_LMV "trusted.dmv"
1611 #define XATTR_NAME_LINK "trusted.link"
1612 #define XATTR_NAME_FID "trusted.fid"
1613 #define XATTR_NAME_VERSION "trusted.version"
1614 #define XATTR_NAME_SOM "trusted.som"
1615 #define XATTR_NAME_HSM "trusted.hsm"
1616 #define XATTR_NAME_LFSCK_BITMAP "trusted.lfsck_bitmap"
1617 #define XATTR_NAME_DUMMY "trusted.dummy"
1619 #define XATTR_NAME_LFSCK_NAMESPACE "trusted.lfsck_ns"
1620 #define XATTR_NAME_MAX_LEN 32 /* increase this, if there is longer name. */
1622 struct lov_mds_md_v3 { /* LOV EA mds/wire data (little-endian) */
1623 __u32 lmm_magic; /* magic number = LOV_MAGIC_V3 */
1624 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1625 struct ost_id lmm_oi; /* LOV object ID */
1626 __u32 lmm_stripe_size; /* size of stripe in bytes */
1627 /* lmm_stripe_count used to be __u32 */
1628 __u16 lmm_stripe_count; /* num stripes in use for this object */
1629 __u16 lmm_layout_gen; /* layout generation number */
1630 char lmm_pool_name[LOV_MAXPOOLNAME + 1]; /* must be 32bit aligned */
1631 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1634 static inline __u32 lov_mds_md_size(__u16 stripes, __u32 lmm_magic)
1636 if (lmm_magic == LOV_MAGIC_V3)
1637 return sizeof(struct lov_mds_md_v3) +
1638 stripes * sizeof(struct lov_ost_data_v1);
1640 return sizeof(struct lov_mds_md_v1) +
1641 stripes * sizeof(struct lov_ost_data_v1);
1645 lov_mds_md_max_stripe_count(size_t buf_size, __u32 lmm_magic)
1647 switch (lmm_magic) {
1648 case LOV_MAGIC_V1: {
1649 struct lov_mds_md_v1 lmm;
1651 if (buf_size < sizeof(lmm))
1654 return (buf_size - sizeof(lmm)) / sizeof(lmm.lmm_objects[0]);
1656 case LOV_MAGIC_V3: {
1657 struct lov_mds_md_v3 lmm;
1659 if (buf_size < sizeof(lmm))
1662 return (buf_size - sizeof(lmm)) / sizeof(lmm.lmm_objects[0]);
1669 #define OBD_MD_FLID (0x00000001ULL) /* object ID */
1670 #define OBD_MD_FLATIME (0x00000002ULL) /* access time */
1671 #define OBD_MD_FLMTIME (0x00000004ULL) /* data modification time */
1672 #define OBD_MD_FLCTIME (0x00000008ULL) /* change time */
1673 #define OBD_MD_FLSIZE (0x00000010ULL) /* size */
1674 #define OBD_MD_FLBLOCKS (0x00000020ULL) /* allocated blocks count */
1675 #define OBD_MD_FLBLKSZ (0x00000040ULL) /* block size */
1676 #define OBD_MD_FLMODE (0x00000080ULL) /* access bits (mode & ~S_IFMT) */
1677 #define OBD_MD_FLTYPE (0x00000100ULL) /* object type (mode & S_IFMT) */
1678 #define OBD_MD_FLUID (0x00000200ULL) /* user ID */
1679 #define OBD_MD_FLGID (0x00000400ULL) /* group ID */
1680 #define OBD_MD_FLFLAGS (0x00000800ULL) /* flags word */
1681 #define OBD_MD_FLNLINK (0x00002000ULL) /* link count */
1682 #define OBD_MD_FLGENER (0x00004000ULL) /* generation number */
1683 /*#define OBD_MD_FLINLINE (0x00008000ULL) inline data. used until 1.6.5 */
1684 #define OBD_MD_FLRDEV (0x00010000ULL) /* device number */
1685 #define OBD_MD_FLEASIZE (0x00020000ULL) /* extended attribute data */
1686 #define OBD_MD_LINKNAME (0x00040000ULL) /* symbolic link target */
1687 #define OBD_MD_FLHANDLE (0x00080000ULL) /* file/lock handle */
1688 #define OBD_MD_FLCKSUM (0x00100000ULL) /* bulk data checksum */
1689 #define OBD_MD_FLQOS (0x00200000ULL) /* quality of service stats */
1690 /*#define OBD_MD_FLOSCOPQ (0x00400000ULL) osc opaque data, never used */
1691 /* OBD_MD_FLCOOKIE (0x00800000ULL) obsolete in 2.8 */
1692 #define OBD_MD_FLGROUP (0x01000000ULL) /* group */
1693 #define OBD_MD_FLFID (0x02000000ULL) /* ->ost write inline fid */
1694 #define OBD_MD_FLEPOCH (0x04000000ULL) /* ->ost write with ioepoch */
1695 /* ->mds if epoch opens or closes */
1696 #define OBD_MD_FLGRANT (0x08000000ULL) /* ost preallocation space grant */
1697 #define OBD_MD_FLDIREA (0x10000000ULL) /* dir's extended attribute data */
1698 #define OBD_MD_FLUSRQUOTA (0x20000000ULL) /* over quota flags sent from ost */
1699 #define OBD_MD_FLGRPQUOTA (0x40000000ULL) /* over quota flags sent from ost */
1700 #define OBD_MD_FLMODEASIZE (0x80000000ULL) /* EA size will be changed */
1702 #define OBD_MD_MDS (0x0000000100000000ULL) /* where an inode lives on */
1703 #define OBD_MD_REINT (0x0000000200000000ULL) /* reintegrate oa */
1704 #define OBD_MD_MEA (0x0000000400000000ULL) /* CMD split EA */
1705 #define OBD_MD_TSTATE (0x0000000800000000ULL) /* transient state field */
1707 #define OBD_MD_FLXATTR (0x0000001000000000ULL) /* xattr */
1708 #define OBD_MD_FLXATTRLS (0x0000002000000000ULL) /* xattr list */
1709 #define OBD_MD_FLXATTRRM (0x0000004000000000ULL) /* xattr remove */
1710 #define OBD_MD_FLACL (0x0000008000000000ULL) /* ACL */
1711 /* OBD_MD_FLRMTPERM (0x0000010000000000ULL) remote perm, obsolete */
1712 #define OBD_MD_FLMDSCAPA (0x0000020000000000ULL) /* MDS capability */
1713 #define OBD_MD_FLOSSCAPA (0x0000040000000000ULL) /* OSS capability */
1714 #define OBD_MD_FLCKSPLIT (0x0000080000000000ULL) /* Check split on server */
1715 #define OBD_MD_FLCROSSREF (0x0000100000000000ULL) /* Cross-ref case */
1716 #define OBD_MD_FLGETATTRLOCK (0x0000200000000000ULL) /* Get IOEpoch attributes
1717 * under lock; for xattr
1718 * requests means the
1719 * client holds the lock */
1720 #define OBD_MD_FLOBJCOUNT (0x0000400000000000ULL) /* for multiple destroy */
1722 /* OBD_MD_FLRMTLSETFACL (0x0001000000000000ULL) lfs lsetfacl, obsolete */
1723 /* OBD_MD_FLRMTLGETFACL (0x0002000000000000ULL) lfs lgetfacl, obsolete */
1724 /* OBD_MD_FLRMTRSETFACL (0x0004000000000000ULL) lfs rsetfacl, obsolete */
1725 /* OBD_MD_FLRMTRGETFACL (0x0008000000000000ULL) lfs rgetfacl, obsolete */
1727 #define OBD_MD_FLDATAVERSION (0x0010000000000000ULL) /* iversion sum */
1728 #define OBD_MD_CLOSE_INTENT_EXECED (0x0020000000000000ULL) /* close intent
1731 #define OBD_MD_DEFAULT_MEA (0x0040000000000000ULL) /* default MEA */
1733 #define OBD_MD_FLGETATTR (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLMTIME | \
1734 OBD_MD_FLCTIME | OBD_MD_FLSIZE | OBD_MD_FLBLKSZ | \
1735 OBD_MD_FLMODE | OBD_MD_FLTYPE | OBD_MD_FLUID | \
1736 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLNLINK | \
1737 OBD_MD_FLGENER | OBD_MD_FLRDEV | OBD_MD_FLGROUP)
1739 #define OBD_MD_FLXATTRALL (OBD_MD_FLXATTR | OBD_MD_FLXATTRLS)
1741 /* don't forget obdo_fid which is way down at the bottom so it can
1742 * come after the definition of llog_cookie */
1746 HSS_CLEARMASK = 0x02,
1747 HSS_ARCHIVE_ID = 0x04,
1750 struct hsm_state_set {
1752 __u32 hss_archive_id;
1754 __u64 hss_clearmask;
1757 /* ost_body.data values for OST_BRW */
1759 #define OBD_BRW_READ 0x01
1760 #define OBD_BRW_WRITE 0x02
1761 #define OBD_BRW_RWMASK (OBD_BRW_READ | OBD_BRW_WRITE)
1762 #define OBD_BRW_SYNC 0x08 /* this page is a part of synchronous
1763 * transfer and is not accounted in
1765 #define OBD_BRW_CHECK 0x10
1766 #define OBD_BRW_FROM_GRANT 0x20 /* the osc manages this under llite */
1767 #define OBD_BRW_GRANTED 0x40 /* the ost manages this */
1768 #define OBD_BRW_NOCACHE 0x80 /* this page is a part of non-cached IO */
1769 #define OBD_BRW_NOQUOTA 0x100
1770 #define OBD_BRW_SRVLOCK 0x200 /* Client holds no lock over this page */
1771 #define OBD_BRW_ASYNC 0x400 /* Server may delay commit to disk */
1772 #define OBD_BRW_MEMALLOC 0x800 /* Client runs in the "kswapd" context */
1773 #define OBD_BRW_OVER_USRQUOTA 0x1000 /* Running out of user quota */
1774 #define OBD_BRW_OVER_GRPQUOTA 0x2000 /* Running out of group quota */
1775 #define OBD_BRW_SOFT_SYNC 0x4000 /* This flag notifies the server
1776 * that the client is running low on
1777 * space for unstable pages; asking
1778 * it to sync quickly */
1780 #define OBD_OBJECT_EOF LUSTRE_EOF
1782 #define OST_MIN_PRECREATE 32
1783 #define OST_MAX_PRECREATE 20000
1786 struct ost_id ioo_oid; /* object ID, if multi-obj BRW */
1787 __u32 ioo_max_brw; /* low 16 bits were o_mode before 2.4,
1788 * now (PTLRPC_BULK_OPS_COUNT - 1) in
1789 * high 16 bits in 2.4 and later */
1790 __u32 ioo_bufcnt; /* number of niobufs for this object */
1793 /* NOTE: IOOBJ_MAX_BRW_BITS defines the _offset_ of the max_brw field in
1794 * ioo_max_brw, NOT the maximum number of bits in PTLRPC_BULK_OPS_BITS.
1795 * That said, ioo_max_brw is a 32-bit field so the limit is also 16 bits. */
1796 #define IOOBJ_MAX_BRW_BITS 16
1797 #define ioobj_max_brw_get(ioo) (((ioo)->ioo_max_brw >> IOOBJ_MAX_BRW_BITS) + 1)
1798 #define ioobj_max_brw_set(ioo, num) \
1799 do { (ioo)->ioo_max_brw = ((num) - 1) << IOOBJ_MAX_BRW_BITS; } while (0)
1801 /* multiple of 8 bytes => can array */
1802 struct niobuf_remote {
1808 /* lock value block communicated between the filter and llite */
1810 /* OST_LVB_ERR_INIT is needed because the return code in rc is
1811 * negative, i.e. because ((MASK + rc) & MASK) != MASK. */
1812 #define OST_LVB_ERR_INIT 0xffbadbad80000000ULL
1813 #define OST_LVB_ERR_MASK 0xffbadbad00000000ULL
1814 #define OST_LVB_IS_ERR(blocks) \
1815 ((blocks & OST_LVB_ERR_MASK) == OST_LVB_ERR_MASK)
1816 #define OST_LVB_SET_ERR(blocks, rc) \
1817 do { blocks = OST_LVB_ERR_INIT + rc; } while (0)
1818 #define OST_LVB_GET_ERR(blocks) (int)(blocks - OST_LVB_ERR_INIT)
1841 * lquota data structures
1844 #ifndef QUOTABLOCK_BITS
1845 # define QUOTABLOCK_BITS LUSTRE_QUOTABLOCK_BITS
1848 #ifndef QUOTABLOCK_SIZE
1849 # define QUOTABLOCK_SIZE LUSTRE_QUOTABLOCK_SIZE
1853 # define toqb lustre_stoqb
1856 /* The lquota_id structure is an union of all the possible identifier types that
1857 * can be used with quota, this includes:
1860 * - a FID which can be used for per-directory quota in the future */
1862 struct lu_fid qid_fid; /* FID for per-directory quota */
1863 __u64 qid_uid; /* user identifier */
1864 __u64 qid_gid; /* group identifier */
1867 /* quotactl management */
1868 struct obd_quotactl {
1870 __u32 qc_type; /* see Q_* flag below */
1873 struct obd_dqinfo qc_dqinfo;
1874 struct obd_dqblk qc_dqblk;
1877 #define Q_COPY(out, in, member) (out)->member = (in)->member
1879 #define QCTL_COPY(out, in) \
1881 Q_COPY(out, in, qc_cmd); \
1882 Q_COPY(out, in, qc_type); \
1883 Q_COPY(out, in, qc_id); \
1884 Q_COPY(out, in, qc_stat); \
1885 Q_COPY(out, in, qc_dqinfo); \
1886 Q_COPY(out, in, qc_dqblk); \
1889 /* Body of quota request used for quota acquire/release RPCs between quota
1890 * master (aka QMT) and slaves (ak QSD). */
1892 struct lu_fid qb_fid; /* FID of global index packing the pool ID
1893 * and type (data or metadata) as well as
1894 * the quota type (user or group). */
1895 union lquota_id qb_id; /* uid or gid or directory FID */
1896 __u32 qb_flags; /* see below */
1898 __u64 qb_count; /* acquire/release count (kbytes/inodes) */
1899 __u64 qb_usage; /* current slave usage (kbytes/inodes) */
1900 __u64 qb_slv_ver; /* slave index file version */
1901 struct lustre_handle qb_lockh; /* per-ID lock handle */
1902 struct lustre_handle qb_glb_lockh; /* global lock handle */
1903 __u64 qb_padding1[4];
1906 /* When the quota_body is used in the reply of quota global intent
1907 * lock (IT_QUOTA_CONN) reply, qb_fid contains slave index file FID. */
1908 #define qb_slv_fid qb_fid
1909 /* qb_usage is the current qunit (in kbytes/inodes) when quota_body is used in
1911 #define qb_qunit qb_usage
1913 #define QUOTA_DQACQ_FL_ACQ 0x1 /* acquire quota */
1914 #define QUOTA_DQACQ_FL_PREACQ 0x2 /* pre-acquire */
1915 #define QUOTA_DQACQ_FL_REL 0x4 /* release quota */
1916 #define QUOTA_DQACQ_FL_REPORT 0x8 /* report usage */
1918 /* Quota types currently supported */
1920 LQUOTA_TYPE_USR = 0x00, /* maps to USRQUOTA */
1921 LQUOTA_TYPE_GRP = 0x01, /* maps to GRPQUOTA */
1925 /* There are 2 different resource types on which a quota limit can be enforced:
1926 * - inodes on the MDTs
1927 * - blocks on the OSTs */
1929 LQUOTA_RES_MD = 0x01, /* skip 0 to avoid null oid in FID */
1930 LQUOTA_RES_DT = 0x02,
1932 LQUOTA_FIRST_RES = LQUOTA_RES_MD
1934 #define LQUOTA_NR_RES (LQUOTA_LAST_RES - LQUOTA_FIRST_RES + 1)
1937 * Space accounting support
1938 * Format of an accounting record, providing disk usage information for a given
1941 struct lquota_acct_rec { /* 16 bytes */
1942 __u64 bspace; /* current space in use */
1943 __u64 ispace; /* current # inodes in use */
1947 * Global quota index support
1948 * Format of a global record, providing global quota settings for a given quota
1951 struct lquota_glb_rec { /* 32 bytes */
1952 __u64 qbr_hardlimit; /* quota hard limit, in #inodes or kbytes */
1953 __u64 qbr_softlimit; /* quota soft limit, in #inodes or kbytes */
1954 __u64 qbr_time; /* grace time, in seconds */
1955 __u64 qbr_granted; /* how much is granted to slaves, in #inodes or
1960 * Slave index support
1961 * Format of a slave record, recording how much space is granted to a given
1964 struct lquota_slv_rec { /* 8 bytes */
1965 __u64 qsr_granted; /* space granted to the slave for the key=ID,
1966 * in #inodes or kbytes */
1969 /* Data structures associated with the quota locks */
1971 /* Glimpse descriptor used for the index & per-ID quota locks */
1972 struct ldlm_gl_lquota_desc {
1973 union lquota_id gl_id; /* quota ID subject to the glimpse */
1974 __u64 gl_flags; /* see LQUOTA_FL* below */
1975 __u64 gl_ver; /* new index version */
1976 __u64 gl_hardlimit; /* new hardlimit or qunit value */
1977 __u64 gl_softlimit; /* new softlimit */
1981 #define gl_qunit gl_hardlimit /* current qunit value used when
1982 * glimpsing per-ID quota locks */
1984 /* quota glimpse flags */
1985 #define LQUOTA_FL_EDQUOT 0x1 /* user/group out of quota space on QMT */
1987 /* LVB used with quota (global and per-ID) locks */
1989 __u64 lvb_flags; /* see LQUOTA_FL* above */
1990 __u64 lvb_id_may_rel; /* space that might be released later */
1991 __u64 lvb_id_rel; /* space released by the slave for this ID */
1992 __u64 lvb_id_qunit; /* current qunit value */
1996 /* LVB used with global quota lock */
1997 #define lvb_glb_ver lvb_id_may_rel /* current version of the global index */
2005 #define QUOTA_FIRST_OPC QUOTA_DQACQ
2014 MDS_GETATTR_NAME = 34,
2019 MDS_DISCONNECT = 39,
2022 MDS_PIN = 42, /* obsolete, never used in a release */
2023 MDS_UNPIN = 43, /* obsolete, never used in a release */
2025 MDS_DONE_WRITING = 45, /* obsolete since 2.8.0 */
2027 MDS_QUOTACHECK = 47, /* not used since 2.4 */
2030 MDS_SETXATTR = 50, /* obsolete, now it's MDS_REINT op */
2032 MDS_IS_SUBDIR = 52, /* obsolete, never used in a release */
2034 MDS_HSM_STATE_GET = 54,
2035 MDS_HSM_STATE_SET = 55,
2036 MDS_HSM_ACTION = 56,
2037 MDS_HSM_PROGRESS = 57,
2038 MDS_HSM_REQUEST = 58,
2039 MDS_HSM_CT_REGISTER = 59,
2040 MDS_HSM_CT_UNREGISTER = 60,
2041 MDS_SWAP_LAYOUTS = 61,
2045 #define MDS_FIRST_OPC MDS_GETATTR
2048 /* opcodes for object update */
2054 #define OUT_UPDATE_FIRST_OPC OUT_UPDATE
2071 } mds_reint_t, mdt_reint_t;
2073 /* the disposition of the intent outlines what was executed */
2074 #define DISP_IT_EXECD 0x00000001
2075 #define DISP_LOOKUP_EXECD 0x00000002
2076 #define DISP_LOOKUP_NEG 0x00000004
2077 #define DISP_LOOKUP_POS 0x00000008
2078 #define DISP_OPEN_CREATE 0x00000010
2079 #define DISP_OPEN_OPEN 0x00000020
2080 #define DISP_ENQ_COMPLETE 0x00400000 /* obsolete and unused */
2081 #define DISP_ENQ_OPEN_REF 0x00800000
2082 #define DISP_ENQ_CREATE_REF 0x01000000
2083 #define DISP_OPEN_LOCK 0x02000000
2084 #define DISP_OPEN_LEASE 0x04000000
2085 #define DISP_OPEN_STRIPE 0x08000000
2086 #define DISP_OPEN_DENY 0x10000000
2088 /* INODE LOCK PARTS */
2089 #define MDS_INODELOCK_LOOKUP 0x000001 /* For namespace, dentry etc, and also
2090 * was used to protect permission (mode,
2091 * owner, group etc) before 2.4. */
2092 #define MDS_INODELOCK_UPDATE 0x000002 /* size, links, timestamps */
2093 #define MDS_INODELOCK_OPEN 0x000004 /* For opened files */
2094 #define MDS_INODELOCK_LAYOUT 0x000008 /* for layout */
2096 /* The PERM bit is added int 2.4, and it is used to protect permission(mode,
2097 * owner, group, acl etc), so to separate the permission from LOOKUP lock.
2098 * Because for remote directories(in DNE), these locks will be granted by
2099 * different MDTs(different ldlm namespace).
2101 * For local directory, MDT will always grant UPDATE_LOCK|PERM_LOCK together.
2102 * For Remote directory, the master MDT, where the remote directory is, will
2103 * grant UPDATE_LOCK|PERM_LOCK, and the remote MDT, where the name entry is,
2104 * will grant LOOKUP_LOCK. */
2105 #define MDS_INODELOCK_PERM 0x000010
2106 #define MDS_INODELOCK_XATTR 0x000020 /* extended attributes */
2108 #define MDS_INODELOCK_MAXSHIFT 5
2109 /* This FULL lock is useful to take on unlink sort of operations */
2110 #define MDS_INODELOCK_FULL ((1<<(MDS_INODELOCK_MAXSHIFT+1))-1)
2112 /* NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
2113 * but was moved into name[1] along with the OID to avoid consuming the
2114 * name[2,3] fields that need to be used for the quota id (also a FID). */
2116 LUSTRE_RES_ID_SEQ_OFF = 0,
2117 LUSTRE_RES_ID_VER_OID_OFF = 1,
2118 LUSTRE_RES_ID_WAS_VER_OFF = 2, /* see note above */
2119 LUSTRE_RES_ID_QUOTA_SEQ_OFF = 2,
2120 LUSTRE_RES_ID_QUOTA_VER_OID_OFF = 3,
2121 LUSTRE_RES_ID_HSH_OFF = 3
2124 #define MDS_STATUS_CONN 1
2125 #define MDS_STATUS_LOV 2
2128 /* these should be identical to their EXT4_*_FL counterparts, they are
2129 * redefined here only to avoid dragging in fs/ext4/ext4.h */
2130 LUSTRE_SYNC_FL = 0x00000008, /* Synchronous updates */
2131 LUSTRE_IMMUTABLE_FL = 0x00000010, /* Immutable file */
2132 LUSTRE_APPEND_FL = 0x00000020, /* writes to file may only append */
2133 LUSTRE_NODUMP_FL = 0x00000040, /* do not dump file */
2134 LUSTRE_NOATIME_FL = 0x00000080, /* do not update atime */
2135 LUSTRE_INDEX_FL = 0x00001000, /* hash-indexed directory */
2136 LUSTRE_DIRSYNC_FL = 0x00010000, /* dirsync behaviour (dir only) */
2137 LUSTRE_TOPDIR_FL = 0x00020000, /* Top of directory hierarchies*/
2138 LUSTRE_DIRECTIO_FL = 0x00100000, /* Use direct i/o */
2139 LUSTRE_INLINE_DATA_FL = 0x10000000, /* Inode has inline data. */
2141 /* These flags will not be identical to any EXT4_*_FL counterparts,
2142 * and only reserved for lustre purpose. Note: these flags might
2143 * be conflict with some of EXT4 flags, so
2144 * 1. these conflict flags needs to be removed when the flag is
2145 * wired by la_flags see osd_attr_get().
2146 * 2. If these flags needs to be stored into inode, they will be
2147 * stored in LMA. see LMAI_XXXX */
2148 LUSTRE_ORPHAN_FL = 0x00002000,
2150 LUSTRE_LMA_FL_MASKS = LUSTRE_ORPHAN_FL,
2154 /* Convert wire LUSTRE_*_FL to corresponding client local VFS S_* values
2155 * for the client inode i_flags. The LUSTRE_*_FL are the Lustre wire
2156 * protocol equivalents of LDISKFS_*_FL values stored on disk, while
2157 * the S_* flags are kernel-internal values that change between kernel
2158 * versions. These flags are set/cleared via FSFILT_IOC_{GET,SET}_FLAGS.
2159 * See b=16526 for a full history. */
2160 static inline int ll_ext_to_inode_flags(int flags)
2162 return (((flags & LUSTRE_SYNC_FL) ? S_SYNC : 0) |
2163 ((flags & LUSTRE_NOATIME_FL) ? S_NOATIME : 0) |
2164 ((flags & LUSTRE_APPEND_FL) ? S_APPEND : 0) |
2165 #if defined(S_DIRSYNC)
2166 ((flags & LUSTRE_DIRSYNC_FL) ? S_DIRSYNC : 0) |
2168 ((flags & LUSTRE_IMMUTABLE_FL) ? S_IMMUTABLE : 0));
2171 static inline int ll_inode_to_ext_flags(int iflags)
2173 return (((iflags & S_SYNC) ? LUSTRE_SYNC_FL : 0) |
2174 ((iflags & S_NOATIME) ? LUSTRE_NOATIME_FL : 0) |
2175 ((iflags & S_APPEND) ? LUSTRE_APPEND_FL : 0) |
2176 #if defined(S_DIRSYNC)
2177 ((iflags & S_DIRSYNC) ? LUSTRE_DIRSYNC_FL : 0) |
2179 ((iflags & S_IMMUTABLE) ? LUSTRE_IMMUTABLE_FL : 0));
2183 /* 64 possible states */
2184 enum md_transient_state {
2185 MS_RESTORE = (1 << 0), /* restore is running */
2189 struct lu_fid mbo_fid1;
2190 struct lu_fid mbo_fid2;
2191 struct lustre_handle mbo_handle;
2193 __u64 mbo_size; /* Offset, in the case of MDS_READPAGE */
2197 __u64 mbo_blocks; /* XID, in the case of MDS_READPAGE */
2199 __u64 mbo_t_state; /* transient file state defined in
2200 * enum md_transient_state
2201 * was "ino" until 2.4.0 */
2204 __u32 mbo_capability;
2208 __u32 mbo_flags; /* LUSTRE_*_FL file attributes */
2210 __u32 mbo_nlink; /* #bytes to read in the case of MDS_READPAGE */
2211 __u32 mbo_unused2; /* was "generation" until 2.4.0 */
2213 __u32 mbo_eadatasize;
2215 __u32 mbo_max_mdsize;
2216 __u32 mbo_unused3; /* was max_cookiesize until 2.8 */
2217 __u32 mbo_uid_h; /* high 32-bits of uid, for FUID */
2218 __u32 mbo_gid_h; /* high 32-bits of gid, for FUID */
2219 __u32 mbo_padding_5; /* also fix lustre_swab_mdt_body */
2220 __u64 mbo_padding_6;
2221 __u64 mbo_padding_7;
2222 __u64 mbo_padding_8;
2223 __u64 mbo_padding_9;
2224 __u64 mbo_padding_10;
2227 struct mdt_ioepoch {
2228 struct lustre_handle mio_handle;
2229 __u64 mio_unused1; /* was ioepoch */
2230 __u32 mio_unused2; /* was flags */
2234 /* permissions for md_perm.mp_perm */
2236 CFS_SETUID_PERM = 0x01,
2237 CFS_SETGID_PERM = 0x02,
2238 CFS_SETGRP_PERM = 0x04,
2241 struct mdt_rec_setattr {
2251 __u32 sa_padding_1_h;
2252 struct lu_fid sa_fid;
2261 __u32 sa_attr_flags;
2263 __u32 sa_bias; /* some operation flags */
2270 * Attribute flags used in mdt_rec_setattr::sa_valid.
2271 * The kernel's #defines for ATTR_* should not be used over the network
2272 * since the client and MDS may run different kernels (see bug 13828)
2273 * Therefore, we should only use MDS_ATTR_* attributes for sa_valid.
2275 #define MDS_ATTR_MODE 0x1ULL /* = 1 */
2276 #define MDS_ATTR_UID 0x2ULL /* = 2 */
2277 #define MDS_ATTR_GID 0x4ULL /* = 4 */
2278 #define MDS_ATTR_SIZE 0x8ULL /* = 8 */
2279 #define MDS_ATTR_ATIME 0x10ULL /* = 16 */
2280 #define MDS_ATTR_MTIME 0x20ULL /* = 32 */
2281 #define MDS_ATTR_CTIME 0x40ULL /* = 64 */
2282 #define MDS_ATTR_ATIME_SET 0x80ULL /* = 128 */
2283 #define MDS_ATTR_MTIME_SET 0x100ULL /* = 256 */
2284 #define MDS_ATTR_FORCE 0x200ULL /* = 512, Not a change, but a change it */
2285 #define MDS_ATTR_ATTR_FLAG 0x400ULL /* = 1024 */
2286 #define MDS_ATTR_KILL_SUID 0x800ULL /* = 2048 */
2287 #define MDS_ATTR_KILL_SGID 0x1000ULL /* = 4096 */
2288 #define MDS_ATTR_CTIME_SET 0x2000ULL /* = 8192 */
2289 #define MDS_ATTR_FROM_OPEN 0x4000ULL /* = 16384, called from open path, ie O_TRUNC */
2290 #define MDS_ATTR_BLOCKS 0x8000ULL /* = 32768 */
2293 #define FMODE_READ 00000001
2294 #define FMODE_WRITE 00000002
2297 #define MDS_FMODE_CLOSED 00000000
2298 #define MDS_FMODE_EXEC 00000004
2299 /* MDS_FMODE_EPOCH 01000000 obsolete since 2.8.0 */
2300 /* MDS_FMODE_TRUNC 02000000 obsolete since 2.8.0 */
2301 /* MDS_FMODE_SOM 04000000 obsolete since 2.8.0 */
2303 #define MDS_OPEN_CREATED 00000010
2304 #define MDS_OPEN_CROSS 00000020
2306 #define MDS_OPEN_CREAT 00000100
2307 #define MDS_OPEN_EXCL 00000200
2308 #define MDS_OPEN_TRUNC 00001000
2309 #define MDS_OPEN_APPEND 00002000
2310 #define MDS_OPEN_SYNC 00010000
2311 #define MDS_OPEN_DIRECTORY 00200000
2313 #define MDS_OPEN_BY_FID 040000000 /* open_by_fid for known object */
2314 #define MDS_OPEN_DELAY_CREATE 0100000000 /* delay initial object create */
2315 #define MDS_OPEN_OWNEROVERRIDE 0200000000 /* NFSD rw-reopen ro file for owner */
2316 #define MDS_OPEN_JOIN_FILE 0400000000 /* open for join file.
2317 * We do not support JOIN FILE
2318 * anymore, reserve this flags
2319 * just for preventing such bit
2322 #define MDS_OPEN_LOCK 04000000000 /* This open requires open lock */
2323 #define MDS_OPEN_HAS_EA 010000000000 /* specify object create pattern */
2324 #define MDS_OPEN_HAS_OBJS 020000000000 /* Just set the EA the obj exist */
2325 #define MDS_OPEN_NORESTORE 0100000000000ULL /* Do not restore file at open */
2326 #define MDS_OPEN_NEWSTRIPE 0200000000000ULL /* New stripe needed (restripe or
2328 #define MDS_OPEN_VOLATILE 0400000000000ULL /* File is volatile = created
2330 #define MDS_OPEN_LEASE 01000000000000ULL /* Open the file and grant lease
2331 * delegation, succeed if it's not
2332 * being opened with conflict mode.
2334 #define MDS_OPEN_RELEASE 02000000000000ULL /* Open the file for HSM release */
2336 /* lustre internal open flags, which should not be set from user space */
2337 #define MDS_OPEN_FL_INTERNAL (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS | \
2338 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK | \
2339 MDS_OPEN_BY_FID | MDS_OPEN_LEASE | \
2343 MDS_CHECK_SPLIT = 1 << 0,
2344 MDS_CROSS_REF = 1 << 1,
2345 MDS_VTX_BYPASS = 1 << 2,
2346 MDS_PERM_BYPASS = 1 << 3,
2347 /* MDS_SOM = 1 << 4, obsolete since 2.8.0 */
2348 MDS_QUOTA_IGNORE = 1 << 5,
2349 /* Was MDS_CLOSE_CLEANUP (1 << 6), No more used */
2350 MDS_KEEP_ORPHAN = 1 << 7,
2351 MDS_RECOV_OPEN = 1 << 8,
2352 MDS_DATA_MODIFIED = 1 << 9,
2353 MDS_CREATE_VOLATILE = 1 << 10,
2354 MDS_OWNEROVERRIDE = 1 << 11,
2355 MDS_HSM_RELEASE = 1 << 12,
2356 MDS_RENAME_MIGRATE = 1 << 13,
2357 MDS_CLOSE_LAYOUT_SWAP = 1 << 14,
2360 /* instance of mdt_reint_rec */
2361 struct mdt_rec_create {
2369 __u32 cr_suppgid1_h;
2371 __u32 cr_suppgid2_h;
2372 struct lu_fid cr_fid1;
2373 struct lu_fid cr_fid2;
2374 struct lustre_handle cr_old_handle; /* handle in case of open replay */
2378 __u64 cr_padding_1; /* rr_blocks */
2381 /* use of helpers set/get_mrc_cr_flags() is needed to access
2382 * 64 bits cr_flags [cr_flags_l, cr_flags_h], this is done to
2383 * extend cr_flags size without breaking 1.8 compat */
2384 __u32 cr_flags_l; /* for use with open, low 32 bits */
2385 __u32 cr_flags_h; /* for use with open, high 32 bits */
2386 __u32 cr_umask; /* umask for create */
2387 __u32 cr_padding_4; /* rr_padding_4 */
2390 /* instance of mdt_reint_rec */
2391 struct mdt_rec_link {
2399 __u32 lk_suppgid1_h;
2401 __u32 lk_suppgid2_h;
2402 struct lu_fid lk_fid1;
2403 struct lu_fid lk_fid2;
2405 __u64 lk_padding_1; /* rr_atime */
2406 __u64 lk_padding_2; /* rr_ctime */
2407 __u64 lk_padding_3; /* rr_size */
2408 __u64 lk_padding_4; /* rr_blocks */
2410 __u32 lk_padding_5; /* rr_mode */
2411 __u32 lk_padding_6; /* rr_flags */
2412 __u32 lk_padding_7; /* rr_padding_2 */
2413 __u32 lk_padding_8; /* rr_padding_3 */
2414 __u32 lk_padding_9; /* rr_padding_4 */
2417 /* instance of mdt_reint_rec */
2418 struct mdt_rec_unlink {
2426 __u32 ul_suppgid1_h;
2428 __u32 ul_suppgid2_h;
2429 struct lu_fid ul_fid1;
2430 struct lu_fid ul_fid2;
2432 __u64 ul_padding_2; /* rr_atime */
2433 __u64 ul_padding_3; /* rr_ctime */
2434 __u64 ul_padding_4; /* rr_size */
2435 __u64 ul_padding_5; /* rr_blocks */
2438 __u32 ul_padding_6; /* rr_flags */
2439 __u32 ul_padding_7; /* rr_padding_2 */
2440 __u32 ul_padding_8; /* rr_padding_3 */
2441 __u32 ul_padding_9; /* rr_padding_4 */
2444 /* instance of mdt_reint_rec */
2445 struct mdt_rec_rename {
2453 __u32 rn_suppgid1_h;
2455 __u32 rn_suppgid2_h;
2456 struct lu_fid rn_fid1;
2457 struct lu_fid rn_fid2;
2459 __u64 rn_padding_1; /* rr_atime */
2460 __u64 rn_padding_2; /* rr_ctime */
2461 __u64 rn_padding_3; /* rr_size */
2462 __u64 rn_padding_4; /* rr_blocks */
2463 __u32 rn_bias; /* some operation flags */
2464 __u32 rn_mode; /* cross-ref rename has mode */
2465 __u32 rn_padding_5; /* rr_flags */
2466 __u32 rn_padding_6; /* rr_padding_2 */
2467 __u32 rn_padding_7; /* rr_padding_3 */
2468 __u32 rn_padding_8; /* rr_padding_4 */
2471 /* instance of mdt_reint_rec */
2472 struct mdt_rec_setxattr {
2480 __u32 sx_suppgid1_h;
2482 __u32 sx_suppgid2_h;
2483 struct lu_fid sx_fid;
2484 __u64 sx_padding_1; /* These three are rr_fid2 */
2489 __u64 sx_padding_5; /* rr_ctime */
2490 __u64 sx_padding_6; /* rr_size */
2491 __u64 sx_padding_7; /* rr_blocks */
2494 __u32 sx_padding_8; /* rr_flags */
2495 __u32 sx_padding_9; /* rr_padding_2 */
2496 __u32 sx_padding_10; /* rr_padding_3 */
2497 __u32 sx_padding_11; /* rr_padding_4 */
2501 * mdt_rec_reint is the template for all mdt_reint_xxx structures.
2502 * Do NOT change the size of various members, otherwise the value
2503 * will be broken in lustre_swab_mdt_rec_reint().
2505 * If you add new members in other mdt_reint_xxx structres and need to use the
2506 * rr_padding_x fields, then update lustre_swab_mdt_rec_reint() also.
2508 struct mdt_rec_reint {
2516 __u32 rr_suppgid1_h;
2518 __u32 rr_suppgid2_h;
2519 struct lu_fid rr_fid1;
2520 struct lu_fid rr_fid2;
2531 __u32 rr_padding_4; /* also fix lustre_swab_mdt_rec_reint */
2534 /* lmv structures */
2536 __u32 ld_tgt_count; /* how many MDS's */
2537 __u32 ld_active_tgt_count; /* how many active */
2538 __u32 ld_default_stripe_count; /* how many objects are used */
2539 __u32 ld_pattern; /* default hash pattern */
2540 __u64 ld_default_hash_size;
2541 __u64 ld_padding_1; /* also fix lustre_swab_lmv_desc */
2542 __u32 ld_padding_2; /* also fix lustre_swab_lmv_desc */
2543 __u32 ld_qos_maxage; /* in second */
2544 __u32 ld_padding_3; /* also fix lustre_swab_lmv_desc */
2545 __u32 ld_padding_4; /* also fix lustre_swab_lmv_desc */
2546 struct obd_uuid ld_uuid;
2549 /* LMV layout EA, and it will be stored both in master and slave object */
2550 struct lmv_mds_md_v1 {
2552 __u32 lmv_stripe_count;
2553 __u32 lmv_master_mdt_index; /* On master object, it is master
2554 * MDT index, on slave object, it
2555 * is stripe index of the slave obj */
2556 __u32 lmv_hash_type; /* dir stripe policy, i.e. indicate
2557 * which hash function to be used,
2558 * Note: only lower 16 bits is being
2559 * used for now. Higher 16 bits will
2560 * be used to mark the object status,
2561 * for example migrating or dead. */
2562 __u32 lmv_layout_version; /* Used for directory restriping */
2566 char lmv_pool_name[LOV_MAXPOOLNAME + 1]; /* pool name */
2567 struct lu_fid lmv_stripe_fids[0]; /* FIDs for each stripe */
2570 #define LMV_MAGIC_V1 0x0CD20CD0 /* normal stripe lmv magic */
2571 #define LMV_MAGIC LMV_MAGIC_V1
2573 /* #define LMV_USER_MAGIC 0x0CD30CD0 */
2574 #define LMV_MAGIC_STRIPE 0x0CD40CD0 /* magic for dir sub_stripe */
2576 /* Right now only the lower part(0-16bits) of lmv_hash_type is being used,
2577 * and the higher part will be the flag to indicate the status of object,
2578 * for example the object is being migrated. And the hash function
2579 * might be interpreted differently with different flags. */
2580 #define LMV_HASH_TYPE_MASK 0x0000ffff
2582 #define LMV_HASH_FLAG_MIGRATION 0x80000000
2584 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 10, 53, 0)
2585 /* Since lustre 2.8, this flag will not be needed, instead this DEAD
2586 * and orphan flags will be stored in LMA (see LMAI_ORPHAN)
2587 * Keep this flag just for LFSCK, because it still might meet such
2588 * flag when it checks the old FS */
2589 #define LMV_HASH_FLAG_DEAD 0x40000000
2591 #define LMV_HASH_FLAG_BAD_TYPE 0x20000000
2593 /* The striped directory has ever lost its master LMV EA, then LFSCK
2594 * re-generated it. This flag is used to indicate such case. It is an
2596 #define LMV_HASH_FLAG_LOST_LMV 0x10000000
2599 * The FNV-1a hash algorithm is as follows:
2600 * hash = FNV_offset_basis
2601 * for each octet_of_data to be hashed
2602 * hash = hash XOR octet_of_data
2603 * hash = hash × FNV_prime
2605 * http://en.wikipedia.org/wiki/Fowler–Noll–Vo_hash_function#FNV-1a_hash
2607 * http://www.isthe.com/chongo/tech/comp/fnv/index.html#FNV-reference-source
2608 * FNV_prime is 2^40 + 2^8 + 0xb3 = 0x100000001b3ULL
2610 #define LUSTRE_FNV_1A_64_PRIME 0x100000001b3ULL
2611 #define LUSTRE_FNV_1A_64_OFFSET_BIAS 0xcbf29ce484222325ULL
2612 static inline __u64 lustre_hash_fnv_1a_64(const void *buf, size_t size)
2614 __u64 hash = LUSTRE_FNV_1A_64_OFFSET_BIAS;
2615 const unsigned char *p = buf;
2618 for (i = 0; i < size; i++) {
2620 hash *= LUSTRE_FNV_1A_64_PRIME;
2628 struct lmv_mds_md_v1 lmv_md_v1;
2629 struct lmv_user_md lmv_user_md;
2632 static inline int lmv_mds_md_size(int stripe_count, unsigned int lmm_magic)
2634 switch (lmm_magic) {
2636 struct lmv_mds_md_v1 *lmm1;
2638 return sizeof(*lmm1) + stripe_count *
2639 sizeof(lmm1->lmv_stripe_fids[0]);
2646 static inline int lmv_mds_md_stripe_count_get(const union lmv_mds_md *lmm)
2648 switch (__le32_to_cpu(lmm->lmv_magic)) {
2650 return __le32_to_cpu(lmm->lmv_md_v1.lmv_stripe_count);
2651 case LMV_USER_MAGIC:
2652 return __le32_to_cpu(lmm->lmv_user_md.lum_stripe_count);
2658 static inline int lmv_mds_md_hash_type_get(const union lmv_mds_md *lmm)
2660 switch (__le32_to_cpu(lmm->lmv_magic)) {
2662 return __le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type);
2663 case LMV_USER_MAGIC:
2664 return __le32_to_cpu(lmm->lmv_user_md.lum_hash_type);
2674 FLD_FIRST_OPC = FLD_QUERY
2680 SEQ_FIRST_OPC = SEQ_QUERY
2684 SEQ_ALLOC_SUPER = 0,
2696 LFSCK_NOTIFY = 1101,
2699 LFSCK_FIRST_OPC = LFSCK_NOTIFY
2703 * LOV data structures
2706 #define LOV_MAX_UUID_BUFFER_SIZE 8192
2707 /* The size of the buffer the lov/mdc reserves for the
2708 * array of UUIDs returned by the MDS. With the current
2709 * protocol, this will limit the max number of OSTs per LOV */
2711 #define LOV_DESC_MAGIC 0xB0CCDE5C
2712 #define LOV_DESC_QOS_MAXAGE_DEFAULT 5 /* Seconds */
2713 #define LOV_DESC_STRIPE_SIZE_DEFAULT (1 << LNET_MTU_BITS)
2715 /* LOV settings descriptor (should only contain static info) */
2717 __u32 ld_tgt_count; /* how many OBD's */
2718 __u32 ld_active_tgt_count; /* how many active */
2719 __u32 ld_default_stripe_count; /* how many objects are used */
2720 __u32 ld_pattern; /* default PATTERN_RAID0 */
2721 __u64 ld_default_stripe_size; /* in bytes */
2722 __u64 ld_default_stripe_offset; /* in bytes */
2723 __u32 ld_padding_0; /* unused */
2724 __u32 ld_qos_maxage; /* in second */
2725 __u32 ld_padding_1; /* also fix lustre_swab_lov_desc */
2726 __u32 ld_padding_2; /* also fix lustre_swab_lov_desc */
2727 struct obd_uuid ld_uuid;
2730 #define ld_magic ld_active_tgt_count /* for swabbing from llogs */
2735 /* opcodes -- MUST be distinct from OST/MDS opcodes */
2740 LDLM_BL_CALLBACK = 104,
2741 LDLM_CP_CALLBACK = 105,
2742 LDLM_GL_CALLBACK = 106,
2743 LDLM_SET_INFO = 107,
2746 #define LDLM_FIRST_OPC LDLM_ENQUEUE
2748 #define RES_NAME_SIZE 4
2749 struct ldlm_res_id {
2750 __u64 name[RES_NAME_SIZE];
2753 #define DLDLMRES "[%#llx:%#llx:%#llx].%#llx"
2754 #define PLDLMRES(res) (unsigned long long)(res)->lr_name.name[0], \
2755 (unsigned long long)(res)->lr_name.name[1], \
2756 (unsigned long long)(res)->lr_name.name[2], \
2757 (unsigned long long)(res)->lr_name.name[3]
2760 typedef enum ldlm_mode {
2773 #define LCK_MODE_NUM 8
2775 typedef enum ldlm_type {
2783 #define LDLM_MIN_TYPE LDLM_PLAIN
2785 struct ldlm_extent {
2791 struct ldlm_inodebits {
2795 struct ldlm_flock_wire {
2803 /* it's important that the fields of the ldlm_extent structure match
2804 * the first fields of the ldlm_flock structure because there is only
2805 * one ldlm_swab routine to process the ldlm_policy_data_t union. if
2806 * this ever changes we will need to swab the union differently based
2807 * on the resource type. */
2809 typedef union ldlm_wire_policy_data {
2810 struct ldlm_extent l_extent;
2811 struct ldlm_flock_wire l_flock;
2812 struct ldlm_inodebits l_inodebits;
2813 } ldlm_wire_policy_data_t;
2815 union ldlm_gl_desc {
2816 struct ldlm_gl_lquota_desc lquota_desc;
2819 enum ldlm_intent_flags {
2820 IT_OPEN = 0x00000001,
2821 IT_CREAT = 0x00000002,
2822 IT_OPEN_CREAT = 0x00000003,
2823 IT_READDIR = 0x00000004,
2824 IT_GETATTR = 0x00000008,
2825 IT_LOOKUP = 0x00000010,
2826 IT_UNLINK = 0x00000020,
2827 IT_TRUNC = 0x00000040,
2828 IT_GETXATTR = 0x00000080,
2829 IT_EXEC = 0x00000100,
2830 IT_PIN = 0x00000200,
2831 IT_LAYOUT = 0x00000400,
2832 IT_QUOTA_DQACQ = 0x00000800,
2833 IT_QUOTA_CONN = 0x00001000,
2834 IT_SETXATTR = 0x00002000,
2837 struct ldlm_intent {
2841 struct ldlm_resource_desc {
2842 enum ldlm_type lr_type;
2843 __u32 lr_pad; /* also fix lustre_swab_ldlm_resource_desc */
2844 struct ldlm_res_id lr_name;
2847 struct ldlm_lock_desc {
2848 struct ldlm_resource_desc l_resource;
2849 enum ldlm_mode l_req_mode;
2850 enum ldlm_mode l_granted_mode;
2851 union ldlm_wire_policy_data l_policy_data;
2854 #define LDLM_LOCKREQ_HANDLES 2
2855 #define LDLM_ENQUEUE_CANCEL_OFF 1
2857 struct ldlm_request {
2860 struct ldlm_lock_desc lock_desc;
2861 struct lustre_handle lock_handle[LDLM_LOCKREQ_HANDLES];
2866 __u32 lock_padding; /* also fix lustre_swab_ldlm_reply */
2867 struct ldlm_lock_desc lock_desc;
2868 struct lustre_handle lock_handle;
2869 __u64 lock_policy_res1;
2870 __u64 lock_policy_res2;
2873 #define ldlm_flags_to_wire(flags) ((__u32)(flags))
2874 #define ldlm_flags_from_wire(flags) ((__u64)(flags))
2877 * Opcodes for mountconf (mgs and mgc)
2882 MGS_EXCEPTION, /* node died, etc. */
2883 MGS_TARGET_REG, /* whenever target starts up */
2889 #define MGS_FIRST_OPC MGS_CONNECT
2891 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
2892 #define MGS_PARAM_MAXLEN 1024
2893 #define KEY_SET_INFO "set_info"
2895 struct mgs_send_param {
2896 char mgs_param[MGS_PARAM_MAXLEN];
2900 /* We pass this info to the MGS so it can write config logs */
2901 #define MTI_NAME_MAXLEN 64
2902 #define MTI_PARAM_MAXLEN 4096
2903 #define MTI_NIDS_MAX 32
2904 struct mgs_target_info {
2905 __u32 mti_lustre_ver;
2906 __u32 mti_stripe_index;
2907 __u32 mti_config_ver;
2909 __u32 mti_nid_count;
2910 __u32 mti_instance; /* Running instance of target */
2911 char mti_fsname[MTI_NAME_MAXLEN];
2912 char mti_svname[MTI_NAME_MAXLEN];
2913 char mti_uuid[sizeof(struct obd_uuid)];
2914 __u64 mti_nids[MTI_NIDS_MAX]; /* host nids (lnet_nid_t)*/
2915 char mti_params[MTI_PARAM_MAXLEN];
2918 struct mgs_nidtbl_entry {
2919 __u64 mne_version; /* table version of this entry */
2920 __u32 mne_instance; /* target instance # */
2921 __u32 mne_index; /* target index */
2922 __u32 mne_length; /* length of this entry - by bytes */
2923 __u8 mne_type; /* target type LDD_F_SV_TYPE_OST/MDT */
2924 __u8 mne_nid_type; /* type of nid(mbz). for ipv6. */
2925 __u8 mne_nid_size; /* size of each NID, by bytes */
2926 __u8 mne_nid_count; /* # of NIDs in buffer */
2928 lnet_nid_t nids[0]; /* variable size buffer for NIDs. */
2932 struct mgs_config_body {
2933 char mcb_name[MTI_NAME_MAXLEN]; /* logname */
2934 __u64 mcb_offset; /* next index of config log to request */
2935 __u16 mcb_type; /* type of log: CONFIG_T_[CONFIG|RECOVER] */
2936 __u8 mcb_nm_cur_pass;
2937 __u8 mcb_bits; /* bits unit size of config log */
2938 __u32 mcb_units; /* # of units for bulk transfer */
2941 struct mgs_config_res {
2942 __u64 mcr_offset; /* index of last config log */
2944 __u64 mcr_size; /* size of the log */
2945 __u64 mcr_nm_cur_pass; /* current nodemap config pass */
2949 /* Config marker flags (in config log) */
2950 #define CM_START 0x01
2952 #define CM_SKIP 0x04
2953 #define CM_UPGRADE146 0x08
2954 #define CM_EXCLUDE 0x10
2955 #define CM_START_SKIP (CM_START | CM_SKIP)
2958 __u32 cm_step; /* aka config version */
2960 __u32 cm_vers; /* lustre release version number */
2961 __u32 cm_padding; /* 64 bit align */
2962 __s64 cm_createtime; /*when this record was first created */
2963 __s64 cm_canceltime; /*when this record is no longer valid*/
2964 char cm_tgtname[MTI_NAME_MAXLEN];
2965 char cm_comment[MTI_NAME_MAXLEN];
2969 * Opcodes for multiple servers.
2975 OBD_QC_CALLBACK, /* not used since 2.4 */
2979 #define OBD_FIRST_OPC OBD_PING
2982 * llog contexts indices.
2984 * There is compatibility problem with indexes below, they are not
2985 * continuous and must keep their numbers for compatibility needs.
2986 * See LU-5218 for details.
2989 LLOG_CONFIG_ORIG_CTXT = 0,
2990 LLOG_CONFIG_REPL_CTXT = 1,
2991 LLOG_MDS_OST_ORIG_CTXT = 2,
2992 LLOG_MDS_OST_REPL_CTXT = 3, /* kept just to avoid re-assignment */
2993 LLOG_SIZE_ORIG_CTXT = 4,
2994 LLOG_SIZE_REPL_CTXT = 5,
2995 LLOG_TEST_ORIG_CTXT = 8,
2996 LLOG_TEST_REPL_CTXT = 9, /* kept just to avoid re-assignment */
2997 LLOG_CHANGELOG_ORIG_CTXT = 12, /**< changelog generation on mdd */
2998 LLOG_CHANGELOG_REPL_CTXT = 13, /**< changelog access on clients */
2999 /* for multiple changelog consumers */
3000 LLOG_CHANGELOG_USER_ORIG_CTXT = 14,
3001 LLOG_AGENT_ORIG_CTXT = 15, /**< agent requests generation on cdt */
3002 LLOG_UPDATELOG_ORIG_CTXT = 16, /* update log */
3003 LLOG_UPDATELOG_REPL_CTXT = 17, /* update log */
3007 /** Identifier for a single log object */
3009 struct ost_id lgl_oi;
3011 } __attribute__((packed));
3013 /** Records written to the CATALOGS list */
3014 #define CATLIST "CATALOGS"
3016 struct llog_logid lci_logid;
3020 } __attribute__((packed));
3022 /* Log data record types - there is no specific reason that these need to
3023 * be related to the RPC opcodes, but no reason not to (may be handy later?)
3025 #define LLOG_OP_MAGIC 0x10600000
3026 #define LLOG_OP_MASK 0xfff00000
3029 LLOG_PAD_MAGIC = LLOG_OP_MAGIC | 0x00000,
3030 OST_SZ_REC = LLOG_OP_MAGIC | 0x00f00,
3031 /* OST_RAID1_REC = LLOG_OP_MAGIC | 0x01000, never used */
3032 MDS_UNLINK_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) |
3033 REINT_UNLINK, /* obsolete after 2.5.0 */
3034 MDS_UNLINK64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
3036 /* MDS_SETATTR_REC = LLOG_OP_MAGIC | 0x12401, obsolete 1.8.0 */
3037 MDS_SETATTR64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
3039 OBD_CFG_REC = LLOG_OP_MAGIC | 0x20000,
3040 /* PTL_CFG_REC = LLOG_OP_MAGIC | 0x30000, obsolete 1.4.0 */
3041 LLOG_GEN_REC = LLOG_OP_MAGIC | 0x40000,
3042 /* LLOG_JOIN_REC = LLOG_OP_MAGIC | 0x50000, obsolete 1.8.0 */
3043 CHANGELOG_REC = LLOG_OP_MAGIC | 0x60000,
3044 CHANGELOG_USER_REC = LLOG_OP_MAGIC | 0x70000,
3045 HSM_AGENT_REC = LLOG_OP_MAGIC | 0x80000,
3046 UPDATE_REC = LLOG_OP_MAGIC | 0xa0000,
3047 LLOG_HDR_MAGIC = LLOG_OP_MAGIC | 0x45539,
3048 LLOG_LOGID_MAGIC = LLOG_OP_MAGIC | 0x4553b,
3051 #define LLOG_REC_HDR_NEEDS_SWABBING(r) \
3052 (((r)->lrh_type & __swab32(LLOG_OP_MASK)) == __swab32(LLOG_OP_MAGIC))
3054 /** Log record header - stored in little endian order.
3055 * Each record must start with this struct, end with a llog_rec_tail,
3056 * and be a multiple of 256 bits in size.
3058 struct llog_rec_hdr {
3065 struct llog_rec_tail {
3070 /* Where data follow just after header */
3071 #define REC_DATA(ptr) \
3072 ((void *)((char *)ptr + sizeof(struct llog_rec_hdr)))
3074 #define REC_DATA_LEN(rec) \
3075 (rec->lrh_len - sizeof(struct llog_rec_hdr) - \
3076 sizeof(struct llog_rec_tail))
3078 struct llog_logid_rec {
3079 struct llog_rec_hdr lid_hdr;
3080 struct llog_logid lid_id;
3084 struct llog_rec_tail lid_tail;
3085 } __attribute__((packed));
3087 struct llog_unlink_rec {
3088 struct llog_rec_hdr lur_hdr;
3092 struct llog_rec_tail lur_tail;
3093 } __attribute__((packed));
3095 struct llog_unlink64_rec {
3096 struct llog_rec_hdr lur_hdr;
3097 struct lu_fid lur_fid;
3098 __u32 lur_count; /* to destroy the lost precreated */
3102 struct llog_rec_tail lur_tail;
3103 } __attribute__((packed));
3105 struct llog_setattr64_rec {
3106 struct llog_rec_hdr lsr_hdr;
3107 struct ost_id lsr_oi;
3113 struct llog_rec_tail lsr_tail;
3114 } __attribute__((packed));
3116 struct llog_size_change_rec {
3117 struct llog_rec_hdr lsc_hdr;
3118 struct ll_fid lsc_fid;
3123 struct llog_rec_tail lsc_tail;
3124 } __attribute__((packed));
3126 #define CHANGELOG_MAGIC 0xca103000
3128 /** \a changelog_rec_type's that can't be masked */
3129 #define CHANGELOG_MINMASK (1 << CL_MARK)
3130 /** bits covering all \a changelog_rec_type's */
3131 #define CHANGELOG_ALLMASK 0XFFFFFFFF
3132 /** default \a changelog_rec_type mask. Allow all of them, except
3133 * CL_ATIME since it can really be time consuming, and not necessary
3134 * under normal use. */
3135 #define CHANGELOG_DEFMASK (CHANGELOG_ALLMASK & ~(1 << CL_ATIME))
3137 /* changelog llog name, needed by client replicators */
3138 #define CHANGELOG_CATALOG "changelog_catalog"
3140 struct changelog_setinfo {
3143 } __attribute__((packed));
3145 /** changelog record */
3146 struct llog_changelog_rec {
3147 struct llog_rec_hdr cr_hdr;
3148 struct changelog_rec cr; /**< Variable length field */
3149 struct llog_rec_tail cr_do_not_use; /**< for_sizeof_only */
3150 } __attribute__((packed));
3152 #define CHANGELOG_USER_PREFIX "cl"
3154 struct llog_changelog_user_rec {
3155 struct llog_rec_hdr cur_hdr;
3159 struct llog_rec_tail cur_tail;
3160 } __attribute__((packed));
3162 enum agent_req_status {
3170 static inline const char *agent_req_status2name(enum agent_req_status ars)
3188 struct llog_agent_req_rec {
3189 struct llog_rec_hdr arr_hdr; /**< record header */
3190 __u32 arr_status; /**< status of the request */
3192 * agent_req_status */
3193 __u32 arr_archive_id; /**< backend archive number */
3194 __u64 arr_flags; /**< req flags */
3195 __u64 arr_compound_id; /**< compound cookie */
3196 __u64 arr_req_create; /**< req. creation time */
3197 __u64 arr_req_change; /**< req. status change time */
3198 struct hsm_action_item arr_hai; /**< req. to the agent */
3199 struct llog_rec_tail arr_tail; /**< record tail for_sizezof_only */
3200 } __attribute__((packed));
3202 /* Old llog gen for compatibility */
3206 } __attribute__((packed));
3208 struct llog_gen_rec {
3209 struct llog_rec_hdr lgr_hdr;
3210 struct llog_gen lgr_gen;
3214 struct llog_rec_tail lgr_tail;
3217 /* flags for the logs */
3219 LLOG_F_ZAP_WHEN_EMPTY = 0x1,
3220 LLOG_F_IS_CAT = 0x2,
3221 LLOG_F_IS_PLAIN = 0x4,
3222 LLOG_F_EXT_JOBID = 0x8,
3223 LLOG_F_IS_FIXSIZE = 0x10,
3225 /* Note: Flags covered by LLOG_F_EXT_MASK will be inherited from
3226 * catlog to plain log, so do not add LLOG_F_IS_FIXSIZE here,
3227 * because the catlog record is usually fixed size, but its plain
3228 * log record can be variable */
3229 LLOG_F_EXT_MASK = LLOG_F_EXT_JOBID,
3232 /* On-disk header structure of each log object, stored in little endian order */
3233 #define LLOG_MIN_CHUNK_SIZE 8192
3234 #define LLOG_HEADER_SIZE (96) /* sizeof (llog_log_hdr) + sizeof(llh_tail)
3235 * - sizeof(llh_bitmap) */
3236 #define LLOG_BITMAP_BYTES (LLOG_MIN_CHUNK_SIZE - LLOG_HEADER_SIZE)
3237 #define LLOG_MIN_REC_SIZE (24) /* round(llog_rec_hdr + llog_rec_tail) */
3239 struct llog_log_hdr {
3240 struct llog_rec_hdr llh_hdr;
3241 __s64 llh_timestamp;
3243 __u32 llh_bitmap_offset;
3246 /* for a catalog the first/oldest and still in-use plain slot is just
3247 * next to it. It will serve as the upper limit after Catalog has
3250 struct obd_uuid llh_tgtuuid;
3251 __u32 llh_reserved[LLOG_HEADER_SIZE/sizeof(__u32)-23];
3252 /* These fields must always be at the end of the llog_log_hdr.
3253 * Note: llh_bitmap size is variable because llog chunk size could be
3254 * bigger than LLOG_MIN_CHUNK_SIZE, i.e. sizeof(llog_log_hdr) > 8192
3255 * bytes, and the real size is stored in llh_hdr.lrh_len, which means
3256 * llh_tail should only be refered by LLOG_HDR_TAIL().
3257 * But this structure is also used by client/server llog interface
3258 * (see llog_client.c), it will be kept in its original way to avoid
3259 * compatiblity issue. */
3260 __u32 llh_bitmap[LLOG_BITMAP_BYTES / sizeof(__u32)];
3261 struct llog_rec_tail llh_tail;
3262 } __attribute__((packed));
3263 #undef LLOG_HEADER_SIZE
3264 #undef LLOG_BITMAP_BYTES
3266 #define LLOG_HDR_BITMAP_SIZE(llh) (__u32)((llh->llh_hdr.lrh_len - \
3267 llh->llh_bitmap_offset - \
3268 sizeof(llh->llh_tail)) * 8)
3269 #define LLOG_HDR_BITMAP(llh) (__u32 *)((char *)(llh) + \
3270 (llh)->llh_bitmap_offset)
3271 #define LLOG_HDR_TAIL(llh) ((struct llog_rec_tail *)((char *)llh + \
3272 llh->llh_hdr.lrh_len - \
3273 sizeof(llh->llh_tail)))
3275 /** log cookies are used to reference a specific log file and a record therein */
3276 struct llog_cookie {
3277 struct llog_logid lgc_lgl;
3281 } __attribute__((packed));
3283 /** llog protocol */
3284 enum llogd_rpc_ops {
3285 LLOG_ORIGIN_HANDLE_CREATE = 501,
3286 LLOG_ORIGIN_HANDLE_NEXT_BLOCK = 502,
3287 LLOG_ORIGIN_HANDLE_READ_HEADER = 503,
3288 LLOG_ORIGIN_HANDLE_WRITE_REC = 504,
3289 LLOG_ORIGIN_HANDLE_CLOSE = 505,
3290 LLOG_ORIGIN_CONNECT = 506,
3291 LLOG_CATINFO = 507, /* deprecated */
3292 LLOG_ORIGIN_HANDLE_PREV_BLOCK = 508,
3293 LLOG_ORIGIN_HANDLE_DESTROY = 509, /* for destroy llog object*/
3295 LLOG_FIRST_OPC = LLOG_ORIGIN_HANDLE_CREATE
3299 struct llog_logid lgd_logid;
3301 __u32 lgd_llh_flags;
3303 __u32 lgd_saved_index;
3305 __u64 lgd_cur_offset;
3306 } __attribute__((packed));
3308 struct llogd_conn_body {
3309 struct llog_gen lgdc_gen;
3310 struct llog_logid lgdc_logid;
3311 __u32 lgdc_ctxt_idx;
3312 } __attribute__((packed));
3314 /* Note: 64-bit types are 64-bit aligned in structure */
3316 __u64 o_valid; /* hot fields in this obdo */
3319 __u64 o_size; /* o_size-o_blocks == ost_lvb */
3323 __u64 o_blocks; /* brw: cli sent cached bytes */
3326 /* 32-bit fields start here: keep an even number of them via padding */
3327 __u32 o_blksize; /* optimal IO blocksize */
3328 __u32 o_mode; /* brw: cli sent cache remain */
3332 __u32 o_nlink; /* brw: checksum */
3334 __u32 o_misc; /* brw: o_dropped */
3336 __u64 o_ioepoch; /* epoch in ost writes */
3337 __u32 o_stripe_idx; /* holds stripe idx */
3339 struct lustre_handle o_handle; /* brw: lock handle to prolong
3341 struct llog_cookie o_lcookie; /* destroy: unlink cookie from
3342 * MDS, obsolete in 2.8, reused
3347 __u64 o_data_version; /* getattr: sum of iversion for
3349 * brw: grant space consumed on
3350 * the client for the write */
3356 #define o_dirty o_blocks
3357 #define o_undirty o_mode
3358 #define o_dropped o_misc
3359 #define o_cksum o_nlink
3360 #define o_grant_used o_data_version
3362 struct lfsck_request {
3375 __u16 lr_async_windows;
3377 struct lu_fid lr_fid;
3378 struct lu_fid lr_fid2;
3379 struct lu_fid lr_fid3;
3384 struct lfsck_reply {
3391 LE_LASTID_REBUILDING = 1,
3392 LE_LASTID_REBUILT = 2,
3398 LE_FID_ACCESSED = 8,
3400 LE_CONDITIONAL_DESTROY = 10,
3401 LE_PAIRS_VERIFY = 11,
3402 LE_SKIP_NLINK_DECLARE = 13,
3404 LE_SET_LMV_MASTER = 15,
3405 LE_SET_LMV_SLAVE = 16,
3408 enum lfsck_event_flags {
3409 LEF_TO_OST = 0x00000001,
3410 LEF_FROM_OST = 0x00000002,
3411 LEF_SET_LMV_HASH = 0x00000004,
3412 LEF_SET_LMV_ALL = 0x00000008,
3413 LEF_RECHECK_NAME_HASH = 0x00000010,
3414 LEF_QUERY_ALL = 0x00000020,
3417 /* request structure for OST's */
3422 /* Key for FIEMAP to be used in get_info calls */
3423 struct ll_fiemap_info_key {
3425 struct obdo lfik_oa;
3426 struct fiemap lfik_fiemap;
3429 #define IDX_INFO_MAGIC 0x3D37CC37
3431 /* Index file transfer through the network. The server serializes the index into
3432 * a byte stream which is sent to the client via a bulk transfer */
3436 /* reply: see idx_info_flags below */
3439 /* request & reply: number of lu_idxpage (to be) transferred */
3443 /* request: requested attributes passed down to the iterator API */
3446 /* request & reply: index file identifier (FID) */
3447 struct lu_fid ii_fid;
3449 /* reply: version of the index file before starting to walk the index.
3450 * Please note that the version can be modified at any time during the
3454 /* request: hash to start with:
3455 * reply: hash of the first entry of the first lu_idxpage and hash
3456 * of the entry to read next if any */
3457 __u64 ii_hash_start;
3460 /* reply: size of keys in lu_idxpages, minimal one if II_FL_VARKEY is
3464 /* reply: size of records in lu_idxpages, minimal one if II_FL_VARREC
3473 #define II_END_OFF MDS_DIR_END_OFF /* all entries have been read */
3475 /* List of flags used in idx_info::ii_flags */
3476 enum idx_info_flags {
3477 II_FL_NOHASH = 1 << 0, /* client doesn't care about hash value */
3478 II_FL_VARKEY = 1 << 1, /* keys can be of variable size */
3479 II_FL_VARREC = 1 << 2, /* records can be of variable size */
3480 II_FL_NONUNQ = 1 << 3, /* index supports non-unique keys */
3481 II_FL_NOKEY = 1 << 4, /* client doesn't care about key */
3484 #define LIP_MAGIC 0x8A6D6B6C
3486 /* 4KB (= LU_PAGE_SIZE) container gathering key/record pairs */
3488 /* 16-byte header */
3491 __u16 lip_nr; /* number of entries in the container */
3492 __u64 lip_pad0; /* additional padding for future use */
3494 /* key/record pairs are stored in the remaining 4080 bytes.
3495 * depending upon the flags in idx_info::ii_flags, each key/record
3496 * pair might be preceded by:
3498 * - the key size (II_FL_VARKEY is set)
3499 * - the record size (II_FL_VARREC is set)
3501 * For the time being, we only support fixed-size key & record. */
3502 char lip_entries[0];
3505 #define LIP_HDR_SIZE (offsetof(struct lu_idxpage, lip_entries))
3507 /* Gather all possible type associated with a 4KB container */
3509 struct lu_dirpage lp_dir; /* for MDS_READPAGE */
3510 struct lu_idxpage lp_idx; /* for OBD_IDX_READ */
3511 char lp_array[LU_PAGE_SIZE];
3514 /* security opcodes */
3517 SEC_CTX_INIT_CONT = 802,
3520 SEC_FIRST_OPC = SEC_CTX_INIT
3524 * capa related definitions
3526 #define CAPA_HMAC_MAX_LEN 64
3527 #define CAPA_HMAC_KEY_MAX_LEN 56
3529 /* NB take care when changing the sequence of elements this struct,
3530 * because the offset info is used in find_capa() */
3531 struct lustre_capa {
3532 struct lu_fid lc_fid; /** fid */
3533 __u64 lc_opc; /** operations allowed */
3534 __u64 lc_uid; /** file owner */
3535 __u64 lc_gid; /** file group */
3536 __u32 lc_flags; /** HMAC algorithm & flags */
3537 __u32 lc_keyid; /** key# used for the capability */
3538 __u32 lc_timeout; /** capa timeout value (sec) */
3539 __u32 lc_expiry; /** expiry time (sec) */
3540 __u8 lc_hmac[CAPA_HMAC_MAX_LEN]; /** HMAC */
3541 } __attribute__((packed));
3543 /** lustre_capa::lc_opc */
3545 CAPA_OPC_BODY_WRITE = 1<<0, /**< write object data */
3546 CAPA_OPC_BODY_READ = 1<<1, /**< read object data */
3547 CAPA_OPC_INDEX_LOOKUP = 1<<2, /**< lookup object fid */
3548 CAPA_OPC_INDEX_INSERT = 1<<3, /**< insert object fid */
3549 CAPA_OPC_INDEX_DELETE = 1<<4, /**< delete object fid */
3550 CAPA_OPC_OSS_WRITE = 1<<5, /**< write oss object data */
3551 CAPA_OPC_OSS_READ = 1<<6, /**< read oss object data */
3552 CAPA_OPC_OSS_TRUNC = 1<<7, /**< truncate oss object */
3553 CAPA_OPC_OSS_DESTROY = 1<<8, /**< destroy oss object */
3554 CAPA_OPC_META_WRITE = 1<<9, /**< write object meta data */
3555 CAPA_OPC_META_READ = 1<<10, /**< read object meta data */
3558 #define CAPA_OPC_OSS_RW (CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE)
3559 #define CAPA_OPC_MDS_ONLY \
3560 (CAPA_OPC_BODY_WRITE | CAPA_OPC_BODY_READ | CAPA_OPC_INDEX_LOOKUP | \
3561 CAPA_OPC_INDEX_INSERT | CAPA_OPC_INDEX_DELETE)
3562 #define CAPA_OPC_OSS_ONLY \
3563 (CAPA_OPC_OSS_WRITE | CAPA_OPC_OSS_READ | CAPA_OPC_OSS_TRUNC | \
3564 CAPA_OPC_OSS_DESTROY)
3565 #define CAPA_OPC_MDS_DEFAULT ~CAPA_OPC_OSS_ONLY
3566 #define CAPA_OPC_OSS_DEFAULT ~(CAPA_OPC_MDS_ONLY | CAPA_OPC_OSS_ONLY)
3568 /* lustre_capa::lc_hmac_alg */
3570 CAPA_HMAC_ALG_SHA1 = 1, /**< sha1 algorithm */
3574 #define CAPA_FL_MASK 0x00ffffff
3575 #define CAPA_HMAC_ALG_MASK 0xff000000
3577 struct lustre_capa_key {
3578 __u64 lk_seq; /**< mds# */
3579 __u32 lk_keyid; /**< key# */
3581 __u8 lk_key[CAPA_HMAC_KEY_MAX_LEN]; /**< key */
3582 } __attribute__((packed));
3584 /** The link ea holds 1 \a link_ea_entry for each hardlink */
3585 #define LINK_EA_MAGIC 0x11EAF1DFUL
3586 struct link_ea_header {
3589 __u64 leh_len; /* total size */
3590 __u32 leh_overflow_time;
3594 /** Hardlink data is name and parent fid.
3595 * Stored in this crazy struct for maximum packing and endian-neutrality
3597 struct link_ea_entry {
3598 /** __u16 stored big-endian, unaligned */
3599 unsigned char lee_reclen[2];
3600 unsigned char lee_parent_fid[sizeof(struct lu_fid)];
3602 }__attribute__((packed));
3604 /** fid2path request/reply structure */
3605 struct getinfo_fid2path {
3606 struct lu_fid gf_fid;
3612 struct lu_fid gf_root_fid[0];
3614 } __attribute__((packed));
3616 /** path2parent request/reply structures */
3618 struct lu_fid gp_fid; /**< parent FID */
3619 __u32 gp_linkno; /**< hardlink number */
3620 __u32 gp_name_size; /**< size of the name field */
3621 char gp_name[0]; /**< zero-terminated link name */
3622 } __attribute__((packed));
3625 LAYOUT_INTENT_ACCESS = 0,
3626 LAYOUT_INTENT_READ = 1,
3627 LAYOUT_INTENT_WRITE = 2,
3628 LAYOUT_INTENT_GLIMPSE = 3,
3629 LAYOUT_INTENT_TRUNC = 4,
3630 LAYOUT_INTENT_RELEASE = 5,
3631 LAYOUT_INTENT_RESTORE = 6
3634 /* enqueue layout lock with intent */
3635 struct layout_intent {
3636 __u32 li_opc; /* intent operation for enqueue, read, write etc */
3643 * On the wire version of hsm_progress structure.
3645 * Contains the userspace hsm_progress and some internal fields.
3647 struct hsm_progress_kernel {
3648 /* Field taken from struct hsm_progress */
3651 struct hsm_extent hpk_extent;
3653 __u16 hpk_errval; /* positive val */
3655 /* Additional fields */
3656 __u64 hpk_data_version;
3658 } __attribute__((packed));
3661 * OUT_UPDATE RPC Format
3663 * During the cross-ref operation, the Master MDT, which the client send the
3664 * request to, will disassembly the operation into object updates, then OSP
3665 * will send these updates to the remote MDT to be executed.
3667 * An UPDATE_OBJ RPC does a list of updates. Each update belongs to an
3668 * operation and does a type of modification to an object.
3676 * update (ub_count-th)
3678 * ub_count must be less than or equal to UPDATE_PER_RPC_MAX.
3683 * rc [+ buffers] (1st)
3684 * rc [+ buffers] (2st)
3686 * rc [+ buffers] (nr_count-th)
3688 * ur_count must be less than or equal to UPDATE_PER_RPC_MAX and should usually
3689 * be equal to ub_count.
3693 * Type of each update, if adding/deleting update, please also update
3694 * update_opcode in lustre/target/out_lib.c.
3706 OUT_INDEX_LOOKUP = 9,
3707 OUT_INDEX_INSERT = 10,
3708 OUT_INDEX_DELETE = 11,
3718 UPDATE_FL_OST = 0x00000001, /* op from OST (not MDT) */
3719 UPDATE_FL_SYNC = 0x00000002, /* commit before replying */
3720 UPDATE_FL_COMMITTED = 0x00000004, /* op committed globally */
3721 UPDATE_FL_NOLOG = 0x00000008 /* for idempotent updates */
3724 struct object_update_param {
3725 __u16 oup_len; /* length of this parameter */
3732 struct object_update {
3733 __u16 ou_type; /* enum update_type */
3734 __u16 ou_params_count; /* update parameters count */
3735 __u32 ou_result_size; /* how many bytes can return */
3736 __u32 ou_flags; /* enum update_flag */
3737 __u32 ou_padding1; /* padding 1 */
3738 __u64 ou_batchid; /* op transno on master */
3739 struct lu_fid ou_fid; /* object to be updated */
3740 struct object_update_param ou_params[0]; /* update params */
3743 #define UPDATE_REQUEST_MAGIC_V1 0xBDDE0001
3744 #define UPDATE_REQUEST_MAGIC_V2 0xBDDE0002
3745 #define UPDATE_REQUEST_MAGIC UPDATE_REQUEST_MAGIC_V2
3746 /* Hold object_updates sending to the remote OUT in single RPC */
3747 struct object_update_request {
3749 __u16 ourq_count; /* number of ourq_updates[] */
3751 struct object_update ourq_updates[0];
3754 #define OUT_UPDATE_HEADER_MAGIC 0xBDDF0001
3755 #define OUT_UPDATE_MAX_INLINE_SIZE 4096
3756 /* Header for updates request between MDTs */
3757 struct out_update_header {
3760 __u32 ouh_inline_length;
3761 __u32 ouh_reply_size;
3762 __u32 ouh_inline_data[0];
3765 struct out_update_buffer {
3770 /* the result of object update */
3771 struct object_update_result {
3778 #define UPDATE_REPLY_MAGIC_V1 0x00BD0001
3779 #define UPDATE_REPLY_MAGIC_V2 0x00BD0002
3780 #define UPDATE_REPLY_MAGIC UPDATE_REPLY_MAGIC_V2
3781 /* Hold object_update_results being replied from the remote OUT. */
3782 struct object_update_reply {
3789 /* read update result */
3790 struct out_read_reply {
3797 /** layout swap request structure
3798 * fid1 and fid2 are in mdt_body
3800 struct mdc_swap_layouts {
3805 struct lustre_handle cd_handle;
3806 struct lu_fid cd_fid;
3807 __u64 cd_data_version;
3808 __u64 cd_reserved[8];
3811 /* Update llog format */
3813 struct lu_fid uop_fid;
3815 __u16 uop_param_count;
3816 __u16 uop_params_off[0];
3820 struct update_op uops_op[0];
3823 struct update_params {
3824 struct object_update_param up_params[0];
3827 enum update_records_flag {
3828 UPDATE_RECORD_CONTINUE = 1 >> 0,
3831 * This is the update record format used to store the updates in
3832 * disk. All updates of the operation will be stored in ur_ops.
3833 * All of parameters for updates of the operation will be stored
3835 * To save the space of the record, parameters in ur_ops will only
3836 * remember their offset in ur_params, so to avoid storing duplicate
3837 * parameters in ur_params, which can help us save a lot space for
3838 * operation like creating striped directory.
3840 struct update_records {
3841 __u64 ur_master_transno;
3844 /* If the operation includes multiple updates, then ur_index
3845 * means the index of the update inside the whole updates. */
3847 __u32 ur_update_count;
3848 __u32 ur_param_count;
3849 struct update_ops ur_ops;
3850 /* Note ur_ops has a variable size, so comment out
3851 * the following ur_params, in case some use it directly
3852 * update_records->ur_params
3854 * struct update_params ur_params;
3858 struct llog_update_record {
3859 struct llog_rec_hdr lur_hdr;
3860 struct update_records lur_update_rec;
3861 /* Note ur_update_rec has a variable size, so comment out
3862 * the following ur_tail, in case someone use it directly
3864 * struct llog_rec_tail lur_tail;
3868 /* nodemap records, uses 32 byte record length */
3869 #define LUSTRE_NODEMAP_NAME_LENGTH 16
3870 struct nodemap_cluster_rec {
3871 char ncr_name[LUSTRE_NODEMAP_NAME_LENGTH + 1];
3875 __u32 ncr_squash_uid;
3876 __u32 ncr_squash_gid;
3879 /* lnet_nid_t is 8 bytes */
3880 struct nodemap_range_rec {
3881 lnet_nid_t nrr_start_nid;
3882 lnet_nid_t nrr_end_nid;
3887 struct nodemap_id_rec {
3895 struct nodemap_global_rec {
3906 struct nodemap_cluster_rec ncr;
3907 struct nodemap_range_rec nrr;
3908 struct nodemap_id_rec nir;
3909 struct nodemap_global_rec ngr;
3912 /* This is the lu_ladvise struct which goes out on the wire.
3913 * Corresponds to the userspace arg llapi_lu_ladvise.
3914 * value[1-4] are unspecified fields, used differently by different advices */
3916 __u16 lla_advice; /* advice type */
3917 __u16 lla_value1; /* values for different advice types */
3919 __u64 lla_start; /* first byte of extent for advice */
3920 __u64 lla_end; /* last byte of extent for advice */
3925 /* This is the ladvise_hdr which goes on the wire, corresponds to the userspace
3926 * arg llapi_ladvise_hdr.
3927 * value[1-3] are unspecified fields, used differently by different advices */
3928 struct ladvise_hdr {
3929 __u32 lah_magic; /* LADVISE_MAGIC */
3930 __u32 lah_count; /* number of advices */
3931 __u64 lah_flags; /* from enum ladvise_flag */
3932 __u32 lah_value1; /* unused */
3933 __u32 lah_value2; /* unused */
3934 __u64 lah_value3; /* unused */
3935 struct lu_ladvise lah_advise[0]; /* advices in this header */