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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/include/lustre/lustre_idl.h
38 * Lustre wire protocol definitions.
41 /** \defgroup lustreidl lustreidl
43 * Lustre wire protocol definitions.
45 * ALL structs passing over the wire should be declared here. Structs
46 * that are used in interfaces with userspace should go in lustre_user.h.
48 * All structs being declared here should be built from simple fixed-size
49 * types (__u8, __u16, __u32, __u64) or be built from other types or
50 * structs also declared in this file. Similarly, all flags and magic
51 * values in those structs should also be declared here. This ensures
52 * that the Lustre wire protocol is not influenced by external dependencies.
54 * The only other acceptable items in this file are VERY SIMPLE accessor
55 * functions to avoid callers grubbing inside the structures. Nothing that
56 * depends on external functions or definitions should be in here.
58 * Structs must be properly aligned to put 64-bit values on an 8-byte
59 * boundary. Any structs being added here must also be added to
60 * utils/wirecheck.c and "make newwiretest" run to regenerate the
61 * utils/wiretest.c sources. This allows us to verify that wire structs
62 * have the proper alignment/size on all architectures.
64 * DO NOT CHANGE any of the structs, flags, values declared here and used
65 * in released Lustre versions. Some structs may have padding fields that
66 * can be used. Some structs might allow addition at the end (verify this
67 * in the code to ensure that new/old clients that see this larger struct
68 * do not fail, otherwise you need to implement protocol compatibility).
73 #ifndef _LUSTRE_IDL_H_
74 #define _LUSTRE_IDL_H_
76 #include <libcfs/libcfs.h>
77 #include <lnet/types.h>
78 #include <lustre/lustre_user.h> /* Defn's shared with user-space. */
79 #include <lustre/lustre_errno.h>
80 #include <lustre_ver.h>
85 /* FOO_REQUEST_PORTAL is for incoming requests on the FOO
86 * FOO_REPLY_PORTAL is for incoming replies on the FOO
87 * FOO_BULK_PORTAL is for incoming bulk on the FOO
90 #define CONNMGR_REQUEST_PORTAL 1
91 #define CONNMGR_REPLY_PORTAL 2
92 //#define OSC_REQUEST_PORTAL 3
93 #define OSC_REPLY_PORTAL 4
94 //#define OSC_BULK_PORTAL 5
95 #define OST_IO_PORTAL 6
96 #define OST_CREATE_PORTAL 7
97 #define OST_BULK_PORTAL 8
98 //#define MDC_REQUEST_PORTAL 9
99 #define MDC_REPLY_PORTAL 10
100 //#define MDC_BULK_PORTAL 11
101 #define MDS_REQUEST_PORTAL 12
102 //#define MDS_REPLY_PORTAL 13
103 #define MDS_BULK_PORTAL 14
104 #define LDLM_CB_REQUEST_PORTAL 15
105 #define LDLM_CB_REPLY_PORTAL 16
106 #define LDLM_CANCEL_REQUEST_PORTAL 17
107 #define LDLM_CANCEL_REPLY_PORTAL 18
108 //#define PTLBD_REQUEST_PORTAL 19
109 //#define PTLBD_REPLY_PORTAL 20
110 //#define PTLBD_BULK_PORTAL 21
111 #define MDS_SETATTR_PORTAL 22
112 #define MDS_READPAGE_PORTAL 23
113 #define OUT_PORTAL 24
114 #define MGC_REPLY_PORTAL 25
115 #define MGS_REQUEST_PORTAL 26
116 #define MGS_REPLY_PORTAL 27
117 #define OST_REQUEST_PORTAL 28
118 #define FLD_REQUEST_PORTAL 29
119 #define SEQ_METADATA_PORTAL 30
120 #define SEQ_DATA_PORTAL 31
121 #define SEQ_CONTROLLER_PORTAL 32
122 #define MGS_BULK_PORTAL 33
124 /* Portal 63 is reserved for the Cray Inc DVS - nic@cray.com, roe@cray.com, n8851@cray.com */
127 #define PTL_RPC_MSG_REQUEST 4711
128 #define PTL_RPC_MSG_ERR 4712
129 #define PTL_RPC_MSG_REPLY 4713
131 /* DON'T use swabbed values of MAGIC as magic! */
132 #define LUSTRE_MSG_MAGIC_V2 0x0BD00BD3
133 #define LUSTRE_MSG_MAGIC_V2_SWABBED 0xD30BD00B
135 #define LUSTRE_MSG_MAGIC LUSTRE_MSG_MAGIC_V2
137 #define PTLRPC_MSG_VERSION 0x00000003
138 #define LUSTRE_VERSION_MASK 0xffff0000
139 #define LUSTRE_OBD_VERSION 0x00010000
140 #define LUSTRE_MDS_VERSION 0x00020000
141 #define LUSTRE_OST_VERSION 0x00030000
142 #define LUSTRE_DLM_VERSION 0x00040000
143 #define LUSTRE_LOG_VERSION 0x00050000
144 #define LUSTRE_MGS_VERSION 0x00060000
147 * Describes a range of sequence, lsr_start is included but lsr_end is
149 * Same structure is used in fld module where lsr_index field holds mdt id
152 struct lu_seq_range {
159 struct lu_seq_range_array {
162 struct lu_seq_range lsra_lsr[0];
165 #define LU_SEQ_RANGE_MDT 0x0
166 #define LU_SEQ_RANGE_OST 0x1
167 #define LU_SEQ_RANGE_ANY 0x3
169 #define LU_SEQ_RANGE_MASK 0x3
171 /** \defgroup lu_fid lu_fid
175 * Flags for lustre_mdt_attrs::lma_compat and lustre_mdt_attrs::lma_incompat.
176 * Deprecated since HSM and SOM attributes are now stored in separate on-disk
180 LMAC_HSM = 0x00000001,
181 /* LMAC_SOM = 0x00000002, obsolete since 2.8.0 */
182 LMAC_NOT_IN_OI = 0x00000004, /* the object does NOT need OI mapping */
183 LMAC_FID_ON_OST = 0x00000008, /* For OST-object, its OI mapping is
184 * under /O/<seq>/d<x>. */
188 * Masks for all features that should be supported by a Lustre version to
189 * access a specific file.
190 * This information is stored in lustre_mdt_attrs::lma_incompat.
193 LMAI_RELEASED = 0x00000001, /* file is released */
194 LMAI_AGENT = 0x00000002, /* agent inode */
195 LMAI_REMOTE_PARENT = 0x00000004, /* the parent of the object
196 is on the remote MDT */
197 LMAI_STRIPED = 0x00000008, /* striped directory inode */
199 #define LMA_INCOMPAT_SUPP (LMAI_AGENT | LMAI_REMOTE_PARENT | LMAI_STRIPED)
201 extern void lustre_lma_swab(struct lustre_mdt_attrs *lma);
202 extern void lustre_lma_init(struct lustre_mdt_attrs *lma,
203 const struct lu_fid *fid,
204 __u32 compat, __u32 incompat);
206 /* copytool uses a 32b bitmask field to encode archive-Ids during register
208 * archive num = 0 => all
209 * archive num from 1 to 32
211 #define LL_HSM_MAX_ARCHIVE (sizeof(__u32) * 8)
214 * HSM on-disk attributes stored in a separate xattr.
217 /** Bitfield for supported data in this structure. For future use. */
220 /** HSM flags, see hsm_flags enum below */
222 /** backend archive id associated with the file */
224 /** version associated with the last archiving, if any */
227 extern void lustre_hsm_swab(struct hsm_attrs *attrs);
233 /** LASTID file has zero OID */
234 LUSTRE_FID_LASTID_OID = 0UL,
235 /** initial fid id value */
236 LUSTRE_FID_INIT_OID = 1UL
239 /** returns fid object sequence */
240 static inline __u64 fid_seq(const struct lu_fid *fid)
245 /** returns fid object id */
246 static inline __u32 fid_oid(const struct lu_fid *fid)
251 /** returns fid object version */
252 static inline __u32 fid_ver(const struct lu_fid *fid)
257 static inline void fid_zero(struct lu_fid *fid)
259 memset(fid, 0, sizeof(*fid));
262 static inline __u64 fid_ver_oid(const struct lu_fid *fid)
264 return ((__u64)fid_ver(fid) << 32 | fid_oid(fid));
268 * Note that reserved SEQ numbers below 12 will conflict with ldiskfs
269 * inodes in the IGIF namespace, so these reserved SEQ numbers can be
270 * used for other purposes and not risk collisions with existing inodes.
272 * Different FID Format
273 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs#NEW.0
276 FID_SEQ_OST_MDT0 = 0,
277 FID_SEQ_LLOG = 1, /* unnamed llogs */
279 FID_SEQ_UNUSED_START = 3,
280 FID_SEQ_UNUSED_END = 9,
281 FID_SEQ_LLOG_NAME = 10, /* named llogs */
284 FID_SEQ_IGIF_MAX = 0x0ffffffffULL,
285 FID_SEQ_IDIF = 0x100000000ULL,
286 FID_SEQ_IDIF_MAX = 0x1ffffffffULL,
287 /* Normal FID sequence starts from this value, i.e. 1<<33 */
288 FID_SEQ_START = 0x200000000ULL,
289 /* sequence for local pre-defined FIDs listed in local_oid */
290 FID_SEQ_LOCAL_FILE = 0x200000001ULL,
291 FID_SEQ_DOT_LUSTRE = 0x200000002ULL,
292 /* sequence is used for local named objects FIDs generated
293 * by local_object_storage library */
294 FID_SEQ_LOCAL_NAME = 0x200000003ULL,
295 /* Because current FLD will only cache the fid sequence, instead
296 * of oid on the client side, if the FID needs to be exposed to
297 * clients sides, it needs to make sure all of fids under one
298 * sequence will be located in one MDT. */
299 FID_SEQ_SPECIAL = 0x200000004ULL,
300 FID_SEQ_QUOTA = 0x200000005ULL,
301 FID_SEQ_QUOTA_GLB = 0x200000006ULL,
302 FID_SEQ_ROOT = 0x200000007ULL, /* Located on MDT0 */
303 FID_SEQ_LAYOUT_RBTREE = 0x200000008ULL,
304 /* sequence is used for update logs of cross-MDT operation */
305 FID_SEQ_UPDATE_LOG = 0x200000009ULL,
306 /* Sequence is used for the directory under which update logs
308 FID_SEQ_UPDATE_LOG_DIR = 0x20000000aULL,
309 FID_SEQ_NORMAL = 0x200000400ULL,
310 FID_SEQ_LOV_DEFAULT = 0xffffffffffffffffULL
313 #define OBIF_OID_MAX_BITS 32
314 #define OBIF_MAX_OID (1ULL << OBIF_OID_MAX_BITS)
315 #define OBIF_OID_MASK ((1ULL << OBIF_OID_MAX_BITS) - 1)
316 #define IDIF_OID_MAX_BITS 48
317 #define IDIF_MAX_OID (1ULL << IDIF_OID_MAX_BITS)
318 #define IDIF_OID_MASK ((1ULL << IDIF_OID_MAX_BITS) - 1)
320 /** OID for FID_SEQ_SPECIAL */
322 /* Big Filesystem Lock to serialize rename operations */
323 FID_OID_SPECIAL_BFL = 1UL,
326 /** OID for FID_SEQ_DOT_LUSTRE */
327 enum dot_lustre_oid {
328 FID_OID_DOT_LUSTRE = 1UL,
329 FID_OID_DOT_LUSTRE_OBF = 2UL,
330 FID_OID_DOT_LUSTRE_LPF = 3UL,
333 /** OID for FID_SEQ_ROOT */
336 FID_OID_ECHO_ROOT = 2UL,
339 static inline bool fid_seq_is_mdt0(__u64 seq)
341 return seq == FID_SEQ_OST_MDT0;
344 static inline bool fid_seq_is_mdt(__u64 seq)
346 return seq == FID_SEQ_OST_MDT0 || seq >= FID_SEQ_NORMAL;
349 static inline bool fid_seq_is_echo(__u64 seq)
351 return seq == FID_SEQ_ECHO;
354 static inline bool fid_is_echo(const struct lu_fid *fid)
356 return fid_seq_is_echo(fid_seq(fid));
359 static inline bool fid_seq_is_llog(__u64 seq)
361 return seq == FID_SEQ_LLOG;
364 static inline bool fid_is_llog(const struct lu_fid *fid)
366 /* file with OID == 0 is not llog but contains last oid */
367 return fid_seq_is_llog(fid_seq(fid)) && fid_oid(fid) > 0;
370 static inline bool fid_seq_is_rsvd(__u64 seq)
372 return seq > FID_SEQ_OST_MDT0 && seq <= FID_SEQ_RSVD;
375 static inline bool fid_seq_is_special(__u64 seq)
377 return seq == FID_SEQ_SPECIAL;
380 static inline bool fid_seq_is_local_file(__u64 seq)
382 return seq == FID_SEQ_LOCAL_FILE ||
383 seq == FID_SEQ_LOCAL_NAME;
386 static inline bool fid_seq_is_root(__u64 seq)
388 return seq == FID_SEQ_ROOT;
391 static inline bool fid_seq_is_dot(__u64 seq)
393 return seq == FID_SEQ_DOT_LUSTRE;
396 static inline bool fid_seq_is_default(__u64 seq)
398 return seq == FID_SEQ_LOV_DEFAULT;
401 static inline bool fid_is_mdt0(const struct lu_fid *fid)
403 return fid_seq_is_mdt0(fid_seq(fid));
406 static inline void lu_root_fid(struct lu_fid *fid)
408 fid->f_seq = FID_SEQ_ROOT;
409 fid->f_oid = FID_OID_ROOT;
413 static inline void lu_echo_root_fid(struct lu_fid *fid)
415 fid->f_seq = FID_SEQ_ROOT;
416 fid->f_oid = FID_OID_ECHO_ROOT;
420 static inline void lu_update_log_fid(struct lu_fid *fid, __u32 index)
422 fid->f_seq = FID_SEQ_UPDATE_LOG;
427 static inline void lu_update_log_dir_fid(struct lu_fid *fid, __u32 index)
429 fid->f_seq = FID_SEQ_UPDATE_LOG_DIR;
435 * Check if a fid is igif or not.
436 * \param fid the fid to be tested.
437 * \return true if the fid is an igif; otherwise false.
439 static inline bool fid_seq_is_igif(__u64 seq)
441 return seq >= FID_SEQ_IGIF && seq <= FID_SEQ_IGIF_MAX;
444 static inline bool fid_is_igif(const struct lu_fid *fid)
446 return fid_seq_is_igif(fid_seq(fid));
450 * Check if a fid is idif or not.
451 * \param fid the fid to be tested.
452 * \return true if the fid is an idif; otherwise false.
454 static inline bool fid_seq_is_idif(__u64 seq)
456 return seq >= FID_SEQ_IDIF && seq <= FID_SEQ_IDIF_MAX;
459 static inline bool fid_is_idif(const struct lu_fid *fid)
461 return fid_seq_is_idif(fid_seq(fid));
464 static inline bool fid_is_local_file(const struct lu_fid *fid)
466 return fid_seq_is_local_file(fid_seq(fid));
469 static inline bool fid_seq_is_norm(__u64 seq)
471 return (seq >= FID_SEQ_NORMAL);
474 static inline bool fid_is_norm(const struct lu_fid *fid)
476 return fid_seq_is_norm(fid_seq(fid));
479 static inline int fid_is_layout_rbtree(const struct lu_fid *fid)
481 return fid_seq(fid) == FID_SEQ_LAYOUT_RBTREE;
484 static inline bool fid_seq_is_update_log(__u64 seq)
486 return seq == FID_SEQ_UPDATE_LOG;
489 static inline bool fid_is_update_log(const struct lu_fid *fid)
491 return fid_seq_is_update_log(fid_seq(fid));
494 static inline bool fid_seq_is_update_log_dir(__u64 seq)
496 return seq == FID_SEQ_UPDATE_LOG_DIR;
499 static inline bool fid_is_update_log_dir(const struct lu_fid *fid)
501 return fid_seq_is_update_log_dir(fid_seq(fid));
504 /* convert an OST objid into an IDIF FID SEQ number */
505 static inline __u64 fid_idif_seq(__u64 id, __u32 ost_idx)
507 return FID_SEQ_IDIF | (ost_idx << 16) | ((id >> 32) & 0xffff);
510 /* convert a packed IDIF FID into an OST objid */
511 static inline __u64 fid_idif_id(__u64 seq, __u32 oid, __u32 ver)
513 return ((__u64)ver << 48) | ((seq & 0xffff) << 32) | oid;
516 static inline __u32 idif_ost_idx(__u64 seq)
518 return (seq >> 16) & 0xffff;
521 /* extract ost index from IDIF FID */
522 static inline __u32 fid_idif_ost_idx(const struct lu_fid *fid)
524 return idif_ost_idx(fid_seq(fid));
527 /* extract OST sequence (group) from a wire ost_id (id/seq) pair */
528 static inline __u64 ostid_seq(const struct ost_id *ostid)
530 if (fid_seq_is_mdt0(ostid->oi.oi_seq))
531 return FID_SEQ_OST_MDT0;
533 if (unlikely(fid_seq_is_default(ostid->oi.oi_seq)))
534 return FID_SEQ_LOV_DEFAULT;
536 if (fid_is_idif(&ostid->oi_fid))
537 return FID_SEQ_OST_MDT0;
539 return fid_seq(&ostid->oi_fid);
542 /* extract OST objid from a wire ost_id (id/seq) pair */
543 static inline __u64 ostid_id(const struct ost_id *ostid)
545 if (fid_seq_is_mdt0(ostid->oi.oi_seq))
546 return ostid->oi.oi_id & IDIF_OID_MASK;
548 if (unlikely(fid_seq_is_default(ostid->oi.oi_seq)))
549 return ostid->oi.oi_id;
551 if (fid_is_idif(&ostid->oi_fid))
552 return fid_idif_id(fid_seq(&ostid->oi_fid),
553 fid_oid(&ostid->oi_fid), 0);
555 return fid_oid(&ostid->oi_fid);
558 static inline void ostid_set_seq(struct ost_id *oi, __u64 seq)
560 if (fid_seq_is_mdt0(seq) || fid_seq_is_default(seq)) {
563 oi->oi_fid.f_seq = seq;
564 /* Note: if f_oid + f_ver is zero, we need init it
565 * to be 1, otherwise, ostid_seq will treat this
566 * as old ostid (oi_seq == 0) */
567 if (oi->oi_fid.f_oid == 0 && oi->oi_fid.f_ver == 0)
568 oi->oi_fid.f_oid = LUSTRE_FID_INIT_OID;
572 static inline void ostid_set_seq_mdt0(struct ost_id *oi)
574 ostid_set_seq(oi, FID_SEQ_OST_MDT0);
577 static inline void ostid_set_seq_echo(struct ost_id *oi)
579 ostid_set_seq(oi, FID_SEQ_ECHO);
582 static inline void ostid_set_seq_llog(struct ost_id *oi)
584 ostid_set_seq(oi, FID_SEQ_LLOG);
588 * Note: we need check oi_seq to decide where to set oi_id,
589 * so oi_seq should always be set ahead of oi_id.
591 static inline void ostid_set_id(struct ost_id *oi, __u64 oid)
593 if (fid_seq_is_mdt0(oi->oi.oi_seq)) {
594 if (oid >= IDIF_MAX_OID) {
595 CERROR("Too large OID %#llx to set MDT0 "DOSTID"\n",
596 (unsigned long long)oid, POSTID(oi));
600 } else if (fid_is_idif(&oi->oi_fid)) {
601 if (oid >= IDIF_MAX_OID) {
602 CERROR("Too large OID %#llx to set IDIF "DOSTID"\n",
603 (unsigned long long)oid, POSTID(oi));
606 oi->oi_fid.f_seq = fid_idif_seq(oid,
607 fid_idif_ost_idx(&oi->oi_fid));
608 oi->oi_fid.f_oid = oid;
609 oi->oi_fid.f_ver = oid >> 48;
611 if (oid > OBIF_MAX_OID) {
612 CERROR("Too large oid %#llx to set REG "DOSTID"\n",
613 (unsigned long long)oid, POSTID(oi));
616 oi->oi_fid.f_oid = oid;
620 static inline int fid_set_id(struct lu_fid *fid, __u64 oid)
622 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
623 CERROR("bad IGIF, "DFID"\n", PFID(fid));
627 if (fid_is_idif(fid)) {
628 if (oid >= IDIF_MAX_OID) {
629 CERROR("Too large OID %#llx to set IDIF "DFID"\n",
630 (unsigned long long)oid, PFID(fid));
633 fid->f_seq = fid_idif_seq(oid, fid_idif_ost_idx(fid));
635 fid->f_ver = oid >> 48;
637 if (oid > OBIF_MAX_OID) {
638 CERROR("Too large OID %#llx to set REG "DFID"\n",
639 (unsigned long long)oid, PFID(fid));
648 * Unpack an OST object id/seq (group) into a FID. This is needed for
649 * converting all obdo, lmm, lsm, etc. 64-bit id/seq pairs into proper
650 * FIDs. Note that if an id/seq is already in FID/IDIF format it will
651 * be passed through unchanged. Only legacy OST objects in "group 0"
652 * will be mapped into the IDIF namespace so that they can fit into the
653 * struct lu_fid fields without loss. For reference see:
654 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs
656 static inline int ostid_to_fid(struct lu_fid *fid, const struct ost_id *ostid,
659 __u64 seq = ostid_seq(ostid);
661 if (ost_idx > 0xffff) {
662 CERROR("bad ost_idx, "DOSTID" ost_idx:%u\n", POSTID(ostid),
667 if (fid_seq_is_mdt0(seq)) {
668 __u64 oid = ostid_id(ostid);
670 /* This is a "legacy" (old 1.x/2.early) OST object in "group 0"
671 * that we map into the IDIF namespace. It allows up to 2^48
672 * objects per OST, as this is the object namespace that has
673 * been in production for years. This can handle create rates
674 * of 1M objects/s/OST for 9 years, or combinations thereof. */
675 if (oid >= IDIF_MAX_OID) {
676 CERROR("bad MDT0 id(1), "DOSTID" ost_idx:%u\n",
677 POSTID(ostid), ost_idx);
680 fid->f_seq = fid_idif_seq(oid, ost_idx);
681 /* truncate to 32 bits by assignment */
683 /* in theory, not currently used */
684 fid->f_ver = oid >> 48;
685 } else if (likely(!fid_seq_is_default(seq)))
686 /* if (fid_seq_is_idif(seq) || fid_seq_is_norm(seq)) */ {
687 /* This is either an IDIF object, which identifies objects
688 * across all OSTs, or a regular FID. The IDIF namespace maps
689 * legacy OST objects into the FID namespace. In both cases,
690 * we just pass the FID through, no conversion needed. */
691 if (ostid->oi_fid.f_ver != 0) {
692 CERROR("bad MDT0 id(2), "DOSTID" ost_idx:%u\n",
693 POSTID(ostid), ost_idx);
696 *fid = ostid->oi_fid;
702 /* pack any OST FID into an ostid (id/seq) for the wire/disk */
703 static inline int fid_to_ostid(const struct lu_fid *fid, struct ost_id *ostid)
705 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
706 CERROR("bad IGIF, "DFID"\n", PFID(fid));
710 if (fid_is_idif(fid)) {
711 ostid_set_seq_mdt0(ostid);
712 ostid_set_id(ostid, fid_idif_id(fid_seq(fid), fid_oid(fid),
715 ostid->oi_fid = *fid;
721 /* Check whether the fid is for LAST_ID */
722 static inline bool fid_is_last_id(const struct lu_fid *fid)
724 return fid_oid(fid) == 0 && fid_seq(fid) != FID_SEQ_UPDATE_LOG &&
725 fid_seq(fid) != FID_SEQ_UPDATE_LOG_DIR;
729 * Get inode number from an igif.
730 * \param fid an igif to get inode number from.
731 * \return inode number for the igif.
733 static inline ino_t lu_igif_ino(const struct lu_fid *fid)
739 * Get inode generation from an igif.
740 * \param fid an igif to get inode generation from.
741 * \return inode generation for the igif.
743 static inline __u32 lu_igif_gen(const struct lu_fid *fid)
749 * Build igif from the inode number/generation.
751 static inline void lu_igif_build(struct lu_fid *fid, __u32 ino, __u32 gen)
759 * Fids are transmitted across network (in the sender byte-ordering),
760 * and stored on disk in big-endian order.
762 static inline void fid_cpu_to_le(struct lu_fid *dst, const struct lu_fid *src)
764 dst->f_seq = cpu_to_le64(fid_seq(src));
765 dst->f_oid = cpu_to_le32(fid_oid(src));
766 dst->f_ver = cpu_to_le32(fid_ver(src));
769 static inline void fid_le_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
771 dst->f_seq = le64_to_cpu(fid_seq(src));
772 dst->f_oid = le32_to_cpu(fid_oid(src));
773 dst->f_ver = le32_to_cpu(fid_ver(src));
776 static inline void fid_cpu_to_be(struct lu_fid *dst, const struct lu_fid *src)
778 dst->f_seq = cpu_to_be64(fid_seq(src));
779 dst->f_oid = cpu_to_be32(fid_oid(src));
780 dst->f_ver = cpu_to_be32(fid_ver(src));
783 static inline void fid_be_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
785 dst->f_seq = be64_to_cpu(fid_seq(src));
786 dst->f_oid = be32_to_cpu(fid_oid(src));
787 dst->f_ver = be32_to_cpu(fid_ver(src));
790 static inline bool fid_is_sane(const struct lu_fid *fid)
792 return fid != NULL &&
793 ((fid_seq(fid) >= FID_SEQ_START && fid_ver(fid) == 0) ||
794 fid_is_igif(fid) || fid_is_idif(fid) ||
795 fid_seq_is_rsvd(fid_seq(fid)));
798 static inline bool lu_fid_eq(const struct lu_fid *f0, const struct lu_fid *f1)
800 return memcmp(f0, f1, sizeof *f0) == 0;
803 #define __diff_normalize(val0, val1) \
805 typeof(val0) __val0 = (val0); \
806 typeof(val1) __val1 = (val1); \
808 (__val0 == __val1 ? 0 : __val0 > __val1 ? +1 : -1); \
811 static inline int lu_fid_cmp(const struct lu_fid *f0,
812 const struct lu_fid *f1)
815 __diff_normalize(fid_seq(f0), fid_seq(f1)) ?:
816 __diff_normalize(fid_oid(f0), fid_oid(f1)) ?:
817 __diff_normalize(fid_ver(f0), fid_ver(f1));
820 static inline void ostid_cpu_to_le(const struct ost_id *src_oi,
821 struct ost_id *dst_oi)
823 if (fid_seq_is_mdt0(src_oi->oi.oi_seq)) {
824 dst_oi->oi.oi_id = cpu_to_le64(src_oi->oi.oi_id);
825 dst_oi->oi.oi_seq = cpu_to_le64(src_oi->oi.oi_seq);
827 fid_cpu_to_le(&dst_oi->oi_fid, &src_oi->oi_fid);
831 static inline void ostid_le_to_cpu(const struct ost_id *src_oi,
832 struct ost_id *dst_oi)
834 if (fid_seq_is_mdt0(src_oi->oi.oi_seq)) {
835 dst_oi->oi.oi_id = le64_to_cpu(src_oi->oi.oi_id);
836 dst_oi->oi.oi_seq = le64_to_cpu(src_oi->oi.oi_seq);
838 fid_le_to_cpu(&dst_oi->oi_fid, &src_oi->oi_fid);
842 struct lu_orphan_rec {
843 /* The MDT-object's FID referenced by the orphan OST-object */
844 struct lu_fid lor_fid;
849 struct lu_orphan_ent {
850 /* The orphan OST-object's FID */
851 struct lu_fid loe_key;
852 struct lu_orphan_rec loe_rec;
857 /** \defgroup lu_dir lu_dir
861 * Enumeration of possible directory entry attributes.
863 * Attributes follow directory entry header in the order they appear in this
866 enum lu_dirent_attrs {
869 LUDA_64BITHASH = 0x0004,
871 /* The following attrs are used for MDT internal only,
872 * not visible to client */
874 /* Something in the record is unknown, to be verified in further. */
875 LUDA_UNKNOWN = 0x0400,
876 /* Ignore this record, go to next directly. */
877 LUDA_IGNORE = 0x0800,
878 /* The system is upgraded, has beed or to be repaired (dryrun). */
879 LUDA_UPGRADE = 0x1000,
880 /* The dirent has been repaired, or to be repaired (dryrun). */
881 LUDA_REPAIR = 0x2000,
882 /* Only check but not repair the dirent inconsistency */
883 LUDA_VERIFY_DRYRUN = 0x4000,
884 /* Verify the dirent consistency */
885 LUDA_VERIFY = 0x8000,
888 #define LU_DIRENT_ATTRS_MASK 0xff00
891 * Layout of readdir pages, as transmitted on wire.
894 /** valid if LUDA_FID is set. */
895 struct lu_fid lde_fid;
896 /** a unique entry identifier: a hash or an offset. */
898 /** total record length, including all attributes. */
902 /** optional variable size attributes following this entry.
903 * taken from enum lu_dirent_attrs.
906 /** name is followed by the attributes indicated in ->ldp_attrs, in
907 * their natural order. After the last attribute, padding bytes are
908 * added to make ->lde_reclen a multiple of 8.
914 * Definitions of optional directory entry attributes formats.
916 * Individual attributes do not have their length encoded in a generic way. It
917 * is assumed that consumer of an attribute knows its format. This means that
918 * it is impossible to skip over an unknown attribute, except by skipping over all
919 * remaining attributes (by using ->lde_reclen), which is not too
920 * constraining, because new server versions will append new attributes at
921 * the end of an entry.
925 * Fid directory attribute: a fid of an object referenced by the entry. This
926 * will be almost always requested by the client and supplied by the server.
928 * Aligned to 8 bytes.
930 /* To have compatibility with 1.8, lets have fid in lu_dirent struct. */
935 * Aligned to 2 bytes.
942 __u64 ldp_hash_start;
946 struct lu_dirent ldp_entries[0];
949 enum lu_dirpage_flags {
951 * dirpage contains no entry.
955 * last entry's lde_hash equals ldp_hash_end.
960 static inline struct lu_dirent *lu_dirent_start(struct lu_dirpage *dp)
962 if (le32_to_cpu(dp->ldp_flags) & LDF_EMPTY)
965 return dp->ldp_entries;
968 static inline struct lu_dirent *lu_dirent_next(struct lu_dirent *ent)
970 struct lu_dirent *next;
972 if (le16_to_cpu(ent->lde_reclen) != 0)
973 next = ((void *)ent) + le16_to_cpu(ent->lde_reclen);
980 static inline size_t lu_dirent_calc_size(size_t namelen, __u16 attr)
984 if (attr & LUDA_TYPE) {
985 const size_t align = sizeof(struct luda_type) - 1;
986 size = (sizeof(struct lu_dirent) + namelen + align) & ~align;
987 size += sizeof(struct luda_type);
989 size = sizeof(struct lu_dirent) + namelen;
991 return (size + 7) & ~7;
994 #define MDS_DIR_END_OFF 0xfffffffffffffffeULL
997 * MDS_READPAGE page size
999 * This is the directory page size packed in MDS_READPAGE RPC.
1000 * It's different than PAGE_CACHE_SIZE because the client needs to
1001 * access the struct lu_dirpage header packed at the beginning of
1002 * the "page" and without this there isn't any way to know find the
1003 * lu_dirpage header is if client and server PAGE_CACHE_SIZE differ.
1005 #define LU_PAGE_SHIFT 12
1006 #define LU_PAGE_SIZE (1UL << LU_PAGE_SHIFT)
1007 #define LU_PAGE_MASK (~(LU_PAGE_SIZE - 1))
1009 #define LU_PAGE_COUNT (1 << (PAGE_CACHE_SHIFT - LU_PAGE_SHIFT))
1013 struct lustre_handle {
1016 #define DEAD_HANDLE_MAGIC 0xdeadbeefcafebabeULL
1018 static inline bool lustre_handle_is_used(const struct lustre_handle *lh)
1020 return lh->cookie != 0;
1023 static inline bool lustre_handle_equal(const struct lustre_handle *lh1,
1024 const struct lustre_handle *lh2)
1026 return lh1->cookie == lh2->cookie;
1029 static inline void lustre_handle_copy(struct lustre_handle *tgt,
1030 const struct lustre_handle *src)
1032 tgt->cookie = src->cookie;
1035 struct lustre_handle_array {
1037 struct lustre_handle handles[0];
1040 /* flags for lm_flags */
1041 #define MSGHDR_AT_SUPPORT 0x1
1042 #define MSGHDR_CKSUM_INCOMPAT18 0x2
1044 #define lustre_msg lustre_msg_v2
1045 /* we depend on this structure to be 8-byte aligned */
1046 /* this type is only endian-adjusted in lustre_unpack_msg() */
1047 struct lustre_msg_v2 {
1056 __u32 lm_buflens[0];
1059 /* without gss, ptlrpc_body is put at the first buffer. */
1060 #define PTLRPC_NUM_VERSIONS 4
1061 struct ptlrpc_body_v3 {
1062 struct lustre_handle pb_handle;
1067 __u64 pb_last_xid; /* highest replied XID without lower unreplied XID */
1068 __u16 pb_tag; /* virtual slot idx for multiple modifying RPCs */
1071 __u64 pb_last_committed;
1076 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1077 __u32 pb_service_time; /* for rep, actual service time */
1080 /* VBR: pre-versions */
1081 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1082 __u64 pb_mbits; /**< match bits for bulk request */
1083 /* padding for future needs */
1084 __u64 pb_padding64_0;
1085 __u64 pb_padding64_1;
1086 __u64 pb_padding64_2;
1087 char pb_jobid[LUSTRE_JOBID_SIZE];
1089 #define ptlrpc_body ptlrpc_body_v3
1091 struct ptlrpc_body_v2 {
1092 struct lustre_handle pb_handle;
1097 __u64 pb_last_xid; /* highest replied XID without lower unreplied XID */
1098 __u16 pb_tag; /* virtual slot idx for multiple modifying RPCs */
1101 __u64 pb_last_committed;
1106 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1107 __u32 pb_service_time; /* for rep, actual service time, also used for
1108 net_latency of req */
1111 /* VBR: pre-versions */
1112 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1113 __u64 pb_mbits; /**< unused in V2 */
1114 /* padding for future needs */
1115 __u64 pb_padding64_0;
1116 __u64 pb_padding64_1;
1117 __u64 pb_padding64_2;
1120 /* message body offset for lustre_msg_v2 */
1121 /* ptlrpc body offset in all request/reply messages */
1122 #define MSG_PTLRPC_BODY_OFF 0
1124 /* normal request/reply message record offset */
1125 #define REQ_REC_OFF 1
1126 #define REPLY_REC_OFF 1
1128 /* ldlm request message body offset */
1129 #define DLM_LOCKREQ_OFF 1 /* lockreq offset */
1130 #define DLM_REQ_REC_OFF 2 /* normal dlm request record offset */
1132 /* ldlm intent lock message body offset */
1133 #define DLM_INTENT_IT_OFF 2 /* intent lock it offset */
1134 #define DLM_INTENT_REC_OFF 3 /* intent lock record offset */
1136 /* ldlm reply message body offset */
1137 #define DLM_LOCKREPLY_OFF 1 /* lockrep offset */
1138 #define DLM_REPLY_REC_OFF 2 /* reply record offset */
1140 /** only use in req->rq_{req,rep}_swab_mask */
1141 #define MSG_PTLRPC_HEADER_OFF 31
1143 /* Flags that are operation-specific go in the top 16 bits. */
1144 #define MSG_OP_FLAG_MASK 0xffff0000
1145 #define MSG_OP_FLAG_SHIFT 16
1147 /* Flags that apply to all requests are in the bottom 16 bits */
1148 #define MSG_GEN_FLAG_MASK 0x0000ffff
1149 #define MSG_LAST_REPLAY 0x0001
1150 #define MSG_RESENT 0x0002
1151 #define MSG_REPLAY 0x0004
1152 /* #define MSG_AT_SUPPORT 0x0008
1153 * This was used in early prototypes of adaptive timeouts, and while there
1154 * shouldn't be any users of that code there also isn't a need for using this
1155 * bits. Defer usage until at least 1.10 to avoid potential conflict. */
1156 #define MSG_DELAY_REPLAY 0x0010
1157 #define MSG_VERSION_REPLAY 0x0020
1158 #define MSG_REQ_REPLAY_DONE 0x0040
1159 #define MSG_LOCK_REPLAY_DONE 0x0080
1162 * Flags for all connect opcodes (MDS_CONNECT, OST_CONNECT)
1165 #define MSG_CONNECT_RECOVERING 0x00000001
1166 #define MSG_CONNECT_RECONNECT 0x00000002
1167 #define MSG_CONNECT_REPLAYABLE 0x00000004
1168 //#define MSG_CONNECT_PEER 0x8
1169 #define MSG_CONNECT_LIBCLIENT 0x00000010
1170 #define MSG_CONNECT_INITIAL 0x00000020
1171 #define MSG_CONNECT_ASYNC 0x00000040
1172 #define MSG_CONNECT_NEXT_VER 0x00000080 /* use next version of lustre_msg */
1173 #define MSG_CONNECT_TRANSNO 0x00000100 /* report transno */
1176 #define OBD_CONNECT_RDONLY 0x1ULL /*client has read-only access*/
1177 #define OBD_CONNECT_INDEX 0x2ULL /*connect specific LOV idx */
1178 #define OBD_CONNECT_MDS 0x4ULL /*connect from MDT to OST */
1179 #define OBD_CONNECT_GRANT 0x8ULL /*OSC gets grant at connect */
1180 #define OBD_CONNECT_SRVLOCK 0x10ULL /*server takes locks for cli */
1181 #define OBD_CONNECT_VERSION 0x20ULL /*Lustre versions in ocd */
1182 #define OBD_CONNECT_REQPORTAL 0x40ULL /*Separate non-IO req portal */
1183 #define OBD_CONNECT_ACL 0x80ULL /*access control lists */
1184 #define OBD_CONNECT_XATTR 0x100ULL /*client use extended attr */
1185 #define OBD_CONNECT_CROW 0x200ULL /*MDS+OST create obj on write*/
1186 #define OBD_CONNECT_TRUNCLOCK 0x400ULL /*locks on server for punch */
1187 #define OBD_CONNECT_TRANSNO 0x800ULL /*replay sends init transno */
1188 #define OBD_CONNECT_IBITS 0x1000ULL /*support for inodebits locks*/
1189 #define OBD_CONNECT_JOIN 0x2000ULL /*files can be concatenated.
1190 *We do not support JOIN FILE
1191 *anymore, reserve this flags
1192 *just for preventing such bit
1194 #define OBD_CONNECT_ATTRFID 0x4000ULL /*Server can GetAttr By Fid*/
1195 #define OBD_CONNECT_NODEVOH 0x8000ULL /*No open hndl on specl nodes*/
1196 #define OBD_CONNECT_RMT_CLIENT 0x10000ULL /*Remote client */
1197 #define OBD_CONNECT_RMT_CLIENT_FORCE 0x20000ULL /*Remote client by force */
1198 #define OBD_CONNECT_BRW_SIZE 0x40000ULL /*Max bytes per rpc */
1199 #define OBD_CONNECT_QUOTA64 0x80000ULL /*Not used since 2.4 */
1200 #define OBD_CONNECT_MDS_CAPA 0x100000ULL /*MDS capability */
1201 #define OBD_CONNECT_OSS_CAPA 0x200000ULL /*OSS capability */
1202 #define OBD_CONNECT_CANCELSET 0x400000ULL /*Early batched cancels. */
1203 #define OBD_CONNECT_SOM 0x800000ULL /*Size on MDS */
1204 #define OBD_CONNECT_AT 0x1000000ULL /*client uses AT */
1205 #define OBD_CONNECT_LRU_RESIZE 0x2000000ULL /*LRU resize feature. */
1206 #define OBD_CONNECT_MDS_MDS 0x4000000ULL /*MDS-MDS connection */
1207 #define OBD_CONNECT_REAL 0x8000000ULL /*real connection */
1208 #define OBD_CONNECT_CHANGE_QS 0x10000000ULL /*Not used since 2.4 */
1209 #define OBD_CONNECT_CKSUM 0x20000000ULL /*support several cksum algos*/
1210 #define OBD_CONNECT_FID 0x40000000ULL /*FID is supported by server */
1211 #define OBD_CONNECT_VBR 0x80000000ULL /*version based recovery */
1212 #define OBD_CONNECT_LOV_V3 0x100000000ULL /*client supports LOV v3 EA */
1213 #define OBD_CONNECT_GRANT_SHRINK 0x200000000ULL /* support grant shrink */
1214 #define OBD_CONNECT_SKIP_ORPHAN 0x400000000ULL /* don't reuse orphan objids */
1215 #define OBD_CONNECT_MAX_EASIZE 0x800000000ULL /* preserved for large EA */
1216 #define OBD_CONNECT_FULL20 0x1000000000ULL /* it is 2.0 client */
1217 #define OBD_CONNECT_LAYOUTLOCK 0x2000000000ULL /* client uses layout lock */
1218 #define OBD_CONNECT_64BITHASH 0x4000000000ULL /* client supports 64-bits
1220 #define OBD_CONNECT_MAXBYTES 0x8000000000ULL /* max stripe size */
1221 #define OBD_CONNECT_IMP_RECOV 0x10000000000ULL /* imp recovery support */
1222 #define OBD_CONNECT_JOBSTATS 0x20000000000ULL /* jobid in ptlrpc_body */
1223 #define OBD_CONNECT_UMASK 0x40000000000ULL /* create uses client umask */
1224 #define OBD_CONNECT_EINPROGRESS 0x80000000000ULL /* client handles -EINPROGRESS
1225 * RPC error properly */
1226 #define OBD_CONNECT_GRANT_PARAM 0x100000000000ULL/* extra grant params used for
1227 * finer space reservation */
1228 #define OBD_CONNECT_FLOCK_OWNER 0x200000000000ULL /* for the fixed 1.8
1229 * policy and 2.x server */
1230 #define OBD_CONNECT_LVB_TYPE 0x400000000000ULL /* variable type of LVB */
1231 #define OBD_CONNECT_NANOSEC_TIME 0x800000000000ULL /* nanosecond timestamps */
1232 #define OBD_CONNECT_LIGHTWEIGHT 0x1000000000000ULL/* lightweight connection */
1233 #define OBD_CONNECT_SHORTIO 0x2000000000000ULL/* short io */
1234 #define OBD_CONNECT_PINGLESS 0x4000000000000ULL/* pings not required */
1235 #define OBD_CONNECT_FLOCK_DEAD 0x8000000000000ULL/* improved flock deadlock detection */
1236 #define OBD_CONNECT_DISP_STRIPE 0x10000000000000ULL/* create stripe disposition*/
1237 #define OBD_CONNECT_OPEN_BY_FID 0x20000000000000ULL /* open by fid won't pack
1239 #define OBD_CONNECT_LFSCK 0x40000000000000ULL/* support online LFSCK */
1240 #define OBD_CONNECT_UNLINK_CLOSE 0x100000000000000ULL/* close file in unlink */
1241 #define OBD_CONNECT_MULTIMODRPCS 0x200000000000000ULL /* support multiple modify
1243 #define OBD_CONNECT_DIR_STRIPE 0x400000000000000ULL /* striped DNE dir */
1244 #define OBD_CONNECT_SUBTREE 0x800000000000000ULL /* fileset mount */
1245 #define OBD_CONNECT_LOCK_AHEAD 0x1000000000000000ULL /* lock ahead */
1246 /** bulk matchbits is sent within ptlrpc_body */
1247 #define OBD_CONNECT_BULK_MBITS 0x2000000000000000ULL
1248 #define OBD_CONNECT_OBDOPACK 0x4000000000000000ULL /* compact OUT obdo */
1249 #define OBD_CONNECT_FLAGS2 0x8000000000000000ULL /* second flags word */
1251 * Please DO NOT add flag values here before first ensuring that this same
1252 * flag value is not in use on some other branch. Please clear any such
1253 * changes with senior engineers before starting to use a new flag. Then,
1254 * submit a small patch against EVERY branch that ONLY adds the new flag,
1255 * updates obd_connect_names[] for lprocfs_rd_connect_flags(), adds the
1256 * flag to check_obd_connect_data(), and updates wiretests accordingly, so it
1257 * can be approved and landed easily to reserve the flag for future use. */
1259 /* The MNE_SWAB flag is overloading the MDS_MDS bit only for the MGS
1260 * connection. It is a temporary bug fix for Imperative Recovery interop
1261 * between 2.2 and 2.3 x86/ppc nodes, and can be removed when interop for
1262 * 2.2 clients/servers is no longer needed. LU-1252/LU-1644. */
1263 #define OBD_CONNECT_MNE_SWAB OBD_CONNECT_MDS_MDS
1265 #define OCD_HAS_FLAG(ocd, flg) \
1266 (!!((ocd)->ocd_connect_flags & OBD_CONNECT_##flg))
1269 #ifdef HAVE_LRU_RESIZE_SUPPORT
1270 #define LRU_RESIZE_CONNECT_FLAG OBD_CONNECT_LRU_RESIZE
1272 #define LRU_RESIZE_CONNECT_FLAG 0
1275 #define MDT_CONNECT_SUPPORTED (OBD_CONNECT_RDONLY | OBD_CONNECT_VERSION | \
1276 OBD_CONNECT_ACL | OBD_CONNECT_XATTR | \
1277 OBD_CONNECT_IBITS | \
1278 OBD_CONNECT_NODEVOH | OBD_CONNECT_ATTRFID | \
1279 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1280 OBD_CONNECT_RMT_CLIENT | \
1281 OBD_CONNECT_RMT_CLIENT_FORCE | \
1282 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_MDS_MDS | \
1283 OBD_CONNECT_FID | LRU_RESIZE_CONNECT_FLAG | \
1284 OBD_CONNECT_VBR | OBD_CONNECT_LOV_V3 | \
1285 OBD_CONNECT_FULL20 | \
1286 OBD_CONNECT_64BITHASH | OBD_CONNECT_JOBSTATS | \
1287 OBD_CONNECT_EINPROGRESS | \
1288 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_UMASK | \
1289 OBD_CONNECT_LVB_TYPE | OBD_CONNECT_LAYOUTLOCK |\
1290 OBD_CONNECT_PINGLESS | OBD_CONNECT_MAX_EASIZE |\
1291 OBD_CONNECT_FLOCK_DEAD | \
1292 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK | \
1293 OBD_CONNECT_OPEN_BY_FID | \
1294 OBD_CONNECT_DIR_STRIPE | \
1295 OBD_CONNECT_BULK_MBITS | \
1296 OBD_CONNECT_MULTIMODRPCS)
1298 #define OST_CONNECT_SUPPORTED (OBD_CONNECT_SRVLOCK | OBD_CONNECT_GRANT | \
1299 OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | \
1300 OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_INDEX | \
1301 OBD_CONNECT_BRW_SIZE | \
1302 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1303 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_CKSUM | \
1304 OBD_CONNECT_RMT_CLIENT | \
1305 OBD_CONNECT_RMT_CLIENT_FORCE | OBD_CONNECT_VBR | \
1306 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN | \
1307 OBD_CONNECT_GRANT_SHRINK | OBD_CONNECT_FULL20 | \
1308 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES | \
1309 OBD_CONNECT_MAX_EASIZE | \
1310 OBD_CONNECT_EINPROGRESS | \
1311 OBD_CONNECT_JOBSTATS | \
1312 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LVB_TYPE|\
1313 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_FID | \
1314 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK | \
1315 OBD_CONNECT_BULK_MBITS)
1316 #define ECHO_CONNECT_SUPPORTED (0)
1317 #define MGS_CONNECT_SUPPORTED (OBD_CONNECT_VERSION | OBD_CONNECT_AT | \
1318 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV | \
1319 OBD_CONNECT_MNE_SWAB | OBD_CONNECT_PINGLESS |\
1320 OBD_CONNECT_BULK_MBITS)
1322 /* Features required for this version of the client to work with server */
1323 #define CLIENT_CONNECT_MDT_REQD (OBD_CONNECT_IBITS | OBD_CONNECT_FID | \
1326 /* This structure is used for both request and reply.
1328 * If we eventually have separate connect data for different types, which we
1329 * almost certainly will, then perhaps we stick a union in here. */
1330 struct obd_connect_data {
1331 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1332 __u32 ocd_version; /* lustre release version number */
1333 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1334 __u32 ocd_index; /* LOV index to connect to */
1335 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
1336 __u64 ocd_ibits_known; /* inode bits this client understands */
1337 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1338 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1339 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1340 __u32 ocd_unused; /* also fix lustre_swab_connect */
1341 __u64 ocd_transno; /* first transno from client to be replayed */
1342 __u32 ocd_group; /* MDS group on OST */
1343 __u32 ocd_cksum_types; /* supported checksum algorithms */
1344 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1345 __u32 ocd_instance; /* instance # of this target */
1346 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1347 /* Fields after ocd_maxbytes are only accessible by the receiver
1348 * if the corresponding flag in ocd_connect_flags is set. Accessing
1349 * any field after ocd_maxbytes on the receiver without a valid flag
1350 * may result in out-of-bound memory access and kernel oops. */
1351 __u16 ocd_maxmodrpcs; /* Maximum modify RPCs in parallel */
1352 __u16 padding0; /* added 2.1.0. also fix lustre_swab_connect */
1353 __u32 padding1; /* added 2.1.0. also fix lustre_swab_connect */
1354 __u64 ocd_connect_flags2;
1355 __u64 padding3; /* added 2.1.0. also fix lustre_swab_connect */
1356 __u64 padding4; /* added 2.1.0. also fix lustre_swab_connect */
1357 __u64 padding5; /* added 2.1.0. also fix lustre_swab_connect */
1358 __u64 padding6; /* added 2.1.0. also fix lustre_swab_connect */
1359 __u64 padding7; /* added 2.1.0. also fix lustre_swab_connect */
1360 __u64 padding8; /* added 2.1.0. also fix lustre_swab_connect */
1361 __u64 padding9; /* added 2.1.0. also fix lustre_swab_connect */
1362 __u64 paddingA; /* added 2.1.0. also fix lustre_swab_connect */
1363 __u64 paddingB; /* added 2.1.0. also fix lustre_swab_connect */
1364 __u64 paddingC; /* added 2.1.0. also fix lustre_swab_connect */
1365 __u64 paddingD; /* added 2.1.0. also fix lustre_swab_connect */
1366 __u64 paddingE; /* added 2.1.0. also fix lustre_swab_connect */
1367 __u64 paddingF; /* added 2.1.0. also fix lustre_swab_connect */
1370 * Please DO NOT use any fields here before first ensuring that this same
1371 * field is not in use on some other branch. Please clear any such changes
1372 * with senior engineers before starting to use a new field. Then, submit
1373 * a small patch against EVERY branch that ONLY adds the new field along with
1374 * the matching OBD_CONNECT flag, so that can be approved and landed easily to
1375 * reserve the flag for future use. */
1378 * Supported checksum algorithms. Up to 32 checksum types are supported.
1379 * (32-bit mask stored in obd_connect_data::ocd_cksum_types)
1380 * Please update DECLARE_CKSUM_NAME/OBD_CKSUM_ALL in obd.h when adding a new
1381 * algorithm and also the OBD_FL_CKSUM* flags.
1384 OBD_CKSUM_CRC32 = 0x00000001,
1385 OBD_CKSUM_ADLER = 0x00000002,
1386 OBD_CKSUM_CRC32C= 0x00000004,
1390 * OST requests: OBDO & OBD request records
1395 OST_REPLY = 0, /* reply ? */
1411 OST_QUOTACHECK = 18, /* not used since 2.4 */
1413 OST_QUOTA_ADJUST_QUNIT = 20, /* not used since 2.4 */
1416 #define OST_FIRST_OPC OST_REPLY
1419 OBD_FL_INLINEDATA = 0x00000001,
1420 OBD_FL_OBDMDEXISTS = 0x00000002,
1421 OBD_FL_DELORPHAN = 0x00000004, /* if set in o_flags delete orphans */
1422 OBD_FL_NORPC = 0x00000008, /* set in o_flags do in OSC not OST */
1423 OBD_FL_IDONLY = 0x00000010, /* set in o_flags only adjust obj id*/
1424 OBD_FL_RECREATE_OBJS= 0x00000020, /* recreate missing obj */
1425 OBD_FL_DEBUG_CHECK = 0x00000040, /* echo client/server debug check */
1426 OBD_FL_NO_USRQUOTA = 0x00000100, /* the object's owner is over quota */
1427 OBD_FL_NO_GRPQUOTA = 0x00000200, /* the object's group is over quota */
1428 OBD_FL_CREATE_CROW = 0x00000400, /* object should be create on write */
1429 OBD_FL_SRVLOCK = 0x00000800, /* delegate DLM locking to server */
1430 OBD_FL_CKSUM_CRC32 = 0x00001000, /* CRC32 checksum type */
1431 OBD_FL_CKSUM_ADLER = 0x00002000, /* ADLER checksum type */
1432 OBD_FL_CKSUM_CRC32C = 0x00004000, /* CRC32C checksum type */
1433 OBD_FL_CKSUM_RSVD2 = 0x00008000, /* for future cksum types */
1434 OBD_FL_CKSUM_RSVD3 = 0x00010000, /* for future cksum types */
1435 OBD_FL_SHRINK_GRANT = 0x00020000, /* object shrink the grant */
1436 OBD_FL_MMAP = 0x00040000, /* object is mmapped on the client.
1437 * XXX: obsoleted - reserved for old
1438 * clients prior than 2.2 */
1439 OBD_FL_RECOV_RESEND = 0x00080000, /* recoverable resent */
1440 OBD_FL_NOSPC_BLK = 0x00100000, /* no more block space on OST */
1441 OBD_FL_FLUSH = 0x00200000, /* flush pages on the OST */
1442 OBD_FL_SHORT_IO = 0x00400000, /* short io request */
1444 /* Note that while these checksum values are currently separate bits,
1445 * in 2.x we can actually allow all values from 1-31 if we wanted. */
1446 OBD_FL_CKSUM_ALL = OBD_FL_CKSUM_CRC32 | OBD_FL_CKSUM_ADLER |
1447 OBD_FL_CKSUM_CRC32C,
1449 /* mask for local-only flag, which won't be sent over network */
1450 OBD_FL_LOCAL_MASK = 0xF0000000,
1454 * All LOV EA magics should have the same postfix, if some new version
1455 * Lustre instroduces new LOV EA magic, then when down-grade to an old
1456 * Lustre, even though the old version system does not recognizes such
1457 * new magic, it still can distinguish the corrupted cases by checking
1458 * the magic's postfix.
1460 #define LOV_MAGIC_MAGIC 0x0BD0
1461 #define LOV_MAGIC_MASK 0xFFFF
1463 #define LOV_MAGIC_V1 (0x0BD10000 | LOV_MAGIC_MAGIC)
1464 #define LOV_MAGIC_JOIN_V1 (0x0BD20000 | LOV_MAGIC_MAGIC)
1465 #define LOV_MAGIC_V3 (0x0BD30000 | LOV_MAGIC_MAGIC)
1466 #define LOV_MAGIC_MIGRATE (0x0BD40000 | LOV_MAGIC_MAGIC)
1467 /* reserved for specifying OSTs */
1468 #define LOV_MAGIC_SPECIFIC (0x0BD50000 | LOV_MAGIC_MAGIC)
1469 #define LOV_MAGIC LOV_MAGIC_V1
1472 * magic for fully defined striping
1473 * the idea is that we should have different magics for striping "hints"
1474 * (struct lov_user_md_v[13]) and defined ready-to-use striping (struct
1475 * lov_mds_md_v[13]). at the moment the magics are used in wire protocol,
1476 * we can't just change it w/o long way preparation, but we still need a
1477 * mechanism to allow LOD to differentiate hint versus ready striping.
1478 * so, at the moment we do a trick: MDT knows what to expect from request
1479 * depending on the case (replay uses ready striping, non-replay req uses
1480 * hints), so MDT replaces magic with appropriate one and now LOD can
1481 * easily understand what's inside -bzzz
1483 #define LOV_MAGIC_V1_DEF 0x0CD10BD0
1484 #define LOV_MAGIC_V3_DEF 0x0CD30BD0
1486 #define lov_pattern(pattern) (pattern & ~LOV_PATTERN_F_MASK)
1487 #define lov_pattern_flags(pattern) (pattern & LOV_PATTERN_F_MASK)
1489 #define lov_ost_data lov_ost_data_v1
1490 struct lov_ost_data_v1 { /* per-stripe data structure (little-endian)*/
1491 struct ost_id l_ost_oi; /* OST object ID */
1492 __u32 l_ost_gen; /* generation of this l_ost_idx */
1493 __u32 l_ost_idx; /* OST index in LOV (lov_tgt_desc->tgts) */
1496 #define lov_mds_md lov_mds_md_v1
1497 struct lov_mds_md_v1 { /* LOV EA mds/wire data (little-endian) */
1498 __u32 lmm_magic; /* magic number = LOV_MAGIC_V1 */
1499 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1500 struct ost_id lmm_oi; /* LOV object ID */
1501 __u32 lmm_stripe_size; /* size of stripe in bytes */
1502 /* lmm_stripe_count used to be __u32 */
1503 __u16 lmm_stripe_count; /* num stripes in use for this object */
1504 __u16 lmm_layout_gen; /* layout generation number */
1505 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1509 * Sigh, because pre-2.4 uses
1510 * struct lov_mds_md_v1 {
1512 * __u64 lmm_object_id;
1513 * __u64 lmm_object_seq;
1516 * to identify the LOV(MDT) object, and lmm_object_seq will
1517 * be normal_fid, which make it hard to combine these conversion
1518 * to ostid_to FID. so we will do lmm_oi/fid conversion separately
1520 * We can tell the lmm_oi by this way,
1521 * 1.8: lmm_object_id = {inode}, lmm_object_gr = 0
1522 * 2.1: lmm_object_id = {oid < 128k}, lmm_object_seq = FID_SEQ_NORMAL
1523 * 2.4: lmm_oi.f_seq = FID_SEQ_NORMAL, lmm_oi.f_oid = {oid < 128k},
1526 * But currently lmm_oi/lsm_oi does not have any "real" usages,
1527 * except for printing some information, and the user can always
1528 * get the real FID from LMA, besides this multiple case check might
1529 * make swab more complicate. So we will keep using id/seq for lmm_oi.
1532 static inline void fid_to_lmm_oi(const struct lu_fid *fid,
1535 oi->oi.oi_id = fid_oid(fid);
1536 oi->oi.oi_seq = fid_seq(fid);
1539 static inline void lmm_oi_set_seq(struct ost_id *oi, __u64 seq)
1541 oi->oi.oi_seq = seq;
1544 static inline void lmm_oi_set_id(struct ost_id *oi, __u64 oid)
1549 static inline __u64 lmm_oi_id(const struct ost_id *oi)
1551 return oi->oi.oi_id;
1554 static inline __u64 lmm_oi_seq(const struct ost_id *oi)
1556 return oi->oi.oi_seq;
1559 static inline void lmm_oi_le_to_cpu(struct ost_id *dst_oi,
1560 const struct ost_id *src_oi)
1562 dst_oi->oi.oi_id = le64_to_cpu(src_oi->oi.oi_id);
1563 dst_oi->oi.oi_seq = le64_to_cpu(src_oi->oi.oi_seq);
1566 static inline void lmm_oi_cpu_to_le(struct ost_id *dst_oi,
1567 const struct ost_id *src_oi)
1569 dst_oi->oi.oi_id = cpu_to_le64(src_oi->oi.oi_id);
1570 dst_oi->oi.oi_seq = cpu_to_le64(src_oi->oi.oi_seq);
1573 #define MAX_MD_SIZE (sizeof(struct lov_mds_md) + 4 * sizeof(struct lov_ost_data))
1574 #define MIN_MD_SIZE (sizeof(struct lov_mds_md) + 1 * sizeof(struct lov_ost_data))
1576 /* This is the default MDT reply size allocated, should the striping be bigger,
1577 * it will be reallocated in mdt_fix_reply.
1578 * 100 stripes is a bit less than 2.5k of data */
1579 #define DEF_REP_MD_SIZE (sizeof(struct lov_mds_md) + \
1580 100 * sizeof(struct lov_ost_data))
1582 #define XATTR_NAME_ACL_ACCESS "system.posix_acl_access"
1583 #define XATTR_NAME_ACL_DEFAULT "system.posix_acl_default"
1584 #define XATTR_USER_PREFIX "user."
1585 #define XATTR_TRUSTED_PREFIX "trusted."
1586 #define XATTR_SECURITY_PREFIX "security."
1588 #define XATTR_NAME_LOV "trusted.lov"
1589 #define XATTR_NAME_LMA "trusted.lma"
1590 #define XATTR_NAME_LMV "trusted.lmv"
1591 #define XATTR_NAME_DEFAULT_LMV "trusted.dmv"
1592 #define XATTR_NAME_LINK "trusted.link"
1593 #define XATTR_NAME_FID "trusted.fid"
1594 #define XATTR_NAME_VERSION "trusted.version"
1595 #define XATTR_NAME_SOM "trusted.som"
1596 #define XATTR_NAME_HSM "trusted.hsm"
1597 #define XATTR_NAME_LFSCK_BITMAP "trusted.lfsck_bitmap"
1598 #define XATTR_NAME_DUMMY "trusted.dummy"
1600 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1601 # define XATTR_NAME_LFSCK_NAMESPACE_OLD "trusted.lfsck_namespace"
1604 #define XATTR_NAME_LFSCK_NAMESPACE "trusted.lfsck_ns"
1605 #define XATTR_NAME_MAX_LEN 32 /* increase this, if there is longer name. */
1607 struct lov_mds_md_v3 { /* LOV EA mds/wire data (little-endian) */
1608 __u32 lmm_magic; /* magic number = LOV_MAGIC_V3 */
1609 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1610 struct ost_id lmm_oi; /* LOV object ID */
1611 __u32 lmm_stripe_size; /* size of stripe in bytes */
1612 /* lmm_stripe_count used to be __u32 */
1613 __u16 lmm_stripe_count; /* num stripes in use for this object */
1614 __u16 lmm_layout_gen; /* layout generation number */
1615 char lmm_pool_name[LOV_MAXPOOLNAME + 1]; /* must be 32bit aligned */
1616 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1619 static inline __u32 lov_mds_md_size(__u16 stripes, __u32 lmm_magic)
1621 if (lmm_magic == LOV_MAGIC_V3)
1622 return sizeof(struct lov_mds_md_v3) +
1623 stripes * sizeof(struct lov_ost_data_v1);
1625 return sizeof(struct lov_mds_md_v1) +
1626 stripes * sizeof(struct lov_ost_data_v1);
1630 lov_mds_md_max_stripe_count(size_t buf_size, __u32 lmm_magic)
1632 switch (lmm_magic) {
1633 case LOV_MAGIC_V1: {
1634 struct lov_mds_md_v1 lmm;
1636 if (buf_size < sizeof(lmm))
1639 return (buf_size - sizeof(lmm)) / sizeof(lmm.lmm_objects[0]);
1641 case LOV_MAGIC_V3: {
1642 struct lov_mds_md_v3 lmm;
1644 if (buf_size < sizeof(lmm))
1647 return (buf_size - sizeof(lmm)) / sizeof(lmm.lmm_objects[0]);
1654 #define OBD_MD_FLID (0x00000001ULL) /* object ID */
1655 #define OBD_MD_FLATIME (0x00000002ULL) /* access time */
1656 #define OBD_MD_FLMTIME (0x00000004ULL) /* data modification time */
1657 #define OBD_MD_FLCTIME (0x00000008ULL) /* change time */
1658 #define OBD_MD_FLSIZE (0x00000010ULL) /* size */
1659 #define OBD_MD_FLBLOCKS (0x00000020ULL) /* allocated blocks count */
1660 #define OBD_MD_FLBLKSZ (0x00000040ULL) /* block size */
1661 #define OBD_MD_FLMODE (0x00000080ULL) /* access bits (mode & ~S_IFMT) */
1662 #define OBD_MD_FLTYPE (0x00000100ULL) /* object type (mode & S_IFMT) */
1663 #define OBD_MD_FLUID (0x00000200ULL) /* user ID */
1664 #define OBD_MD_FLGID (0x00000400ULL) /* group ID */
1665 #define OBD_MD_FLFLAGS (0x00000800ULL) /* flags word */
1666 #define OBD_MD_FLNLINK (0x00002000ULL) /* link count */
1667 #define OBD_MD_FLGENER (0x00004000ULL) /* generation number */
1668 /*#define OBD_MD_FLINLINE (0x00008000ULL) inline data. used until 1.6.5 */
1669 #define OBD_MD_FLRDEV (0x00010000ULL) /* device number */
1670 #define OBD_MD_FLEASIZE (0x00020000ULL) /* extended attribute data */
1671 #define OBD_MD_LINKNAME (0x00040000ULL) /* symbolic link target */
1672 #define OBD_MD_FLHANDLE (0x00080000ULL) /* file/lock handle */
1673 #define OBD_MD_FLCKSUM (0x00100000ULL) /* bulk data checksum */
1674 #define OBD_MD_FLQOS (0x00200000ULL) /* quality of service stats */
1675 /*#define OBD_MD_FLOSCOPQ (0x00400000ULL) osc opaque data, never used */
1676 /* OBD_MD_FLCOOKIE (0x00800000ULL) obsolete in 2.8 */
1677 #define OBD_MD_FLGROUP (0x01000000ULL) /* group */
1678 #define OBD_MD_FLFID (0x02000000ULL) /* ->ost write inline fid */
1679 #define OBD_MD_FLEPOCH (0x04000000ULL) /* ->ost write with ioepoch */
1680 /* ->mds if epoch opens or closes */
1681 #define OBD_MD_FLGRANT (0x08000000ULL) /* ost preallocation space grant */
1682 #define OBD_MD_FLDIREA (0x10000000ULL) /* dir's extended attribute data */
1683 #define OBD_MD_FLUSRQUOTA (0x20000000ULL) /* over quota flags sent from ost */
1684 #define OBD_MD_FLGRPQUOTA (0x40000000ULL) /* over quota flags sent from ost */
1685 #define OBD_MD_FLMODEASIZE (0x80000000ULL) /* EA size will be changed */
1687 #define OBD_MD_MDS (0x0000000100000000ULL) /* where an inode lives on */
1688 #define OBD_MD_REINT (0x0000000200000000ULL) /* reintegrate oa */
1689 #define OBD_MD_MEA (0x0000000400000000ULL) /* CMD split EA */
1690 #define OBD_MD_TSTATE (0x0000000800000000ULL) /* transient state field */
1692 #define OBD_MD_FLXATTR (0x0000001000000000ULL) /* xattr */
1693 #define OBD_MD_FLXATTRLS (0x0000002000000000ULL) /* xattr list */
1694 #define OBD_MD_FLXATTRRM (0x0000004000000000ULL) /* xattr remove */
1695 #define OBD_MD_FLACL (0x0000008000000000ULL) /* ACL */
1696 #define OBD_MD_FLRMTPERM (0x0000010000000000ULL) /* remote permission */
1697 #define OBD_MD_FLMDSCAPA (0x0000020000000000ULL) /* MDS capability */
1698 #define OBD_MD_FLOSSCAPA (0x0000040000000000ULL) /* OSS capability */
1699 #define OBD_MD_FLCKSPLIT (0x0000080000000000ULL) /* Check split on server */
1700 #define OBD_MD_FLCROSSREF (0x0000100000000000ULL) /* Cross-ref case */
1701 #define OBD_MD_FLGETATTRLOCK (0x0000200000000000ULL) /* Get IOEpoch attributes
1702 * under lock; for xattr
1703 * requests means the
1704 * client holds the lock */
1705 #define OBD_MD_FLOBJCOUNT (0x0000400000000000ULL) /* for multiple destroy */
1707 #define OBD_MD_FLRMTLSETFACL (0x0001000000000000ULL) /* lfs lsetfacl case */
1708 #define OBD_MD_FLRMTLGETFACL (0x0002000000000000ULL) /* lfs lgetfacl case */
1709 #define OBD_MD_FLRMTRSETFACL (0x0004000000000000ULL) /* lfs rsetfacl case */
1710 #define OBD_MD_FLRMTRGETFACL (0x0008000000000000ULL) /* lfs rgetfacl case */
1712 #define OBD_MD_FLDATAVERSION (0x0010000000000000ULL) /* iversion sum */
1713 #define OBD_MD_CLOSE_INTENT_EXECED (0x0020000000000000ULL) /* close intent
1716 #define OBD_MD_DEFAULT_MEA (0x0040000000000000ULL) /* default MEA */
1718 #define OBD_MD_FLGETATTR (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLMTIME | \
1719 OBD_MD_FLCTIME | OBD_MD_FLSIZE | OBD_MD_FLBLKSZ | \
1720 OBD_MD_FLMODE | OBD_MD_FLTYPE | OBD_MD_FLUID | \
1721 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLNLINK | \
1722 OBD_MD_FLGENER | OBD_MD_FLRDEV | OBD_MD_FLGROUP)
1724 #define OBD_MD_FLXATTRALL (OBD_MD_FLXATTR | OBD_MD_FLXATTRLS)
1726 /* don't forget obdo_fid which is way down at the bottom so it can
1727 * come after the definition of llog_cookie */
1731 HSS_CLEARMASK = 0x02,
1732 HSS_ARCHIVE_ID = 0x04,
1735 struct hsm_state_set {
1737 __u32 hss_archive_id;
1739 __u64 hss_clearmask;
1742 /* ost_body.data values for OST_BRW */
1744 #define OBD_BRW_READ 0x01
1745 #define OBD_BRW_WRITE 0x02
1746 #define OBD_BRW_RWMASK (OBD_BRW_READ | OBD_BRW_WRITE)
1747 #define OBD_BRW_SYNC 0x08 /* this page is a part of synchronous
1748 * transfer and is not accounted in
1750 #define OBD_BRW_CHECK 0x10
1751 #define OBD_BRW_FROM_GRANT 0x20 /* the osc manages this under llite */
1752 #define OBD_BRW_GRANTED 0x40 /* the ost manages this */
1753 #define OBD_BRW_NOCACHE 0x80 /* this page is a part of non-cached IO */
1754 #define OBD_BRW_NOQUOTA 0x100
1755 #define OBD_BRW_SRVLOCK 0x200 /* Client holds no lock over this page */
1756 #define OBD_BRW_ASYNC 0x400 /* Server may delay commit to disk */
1757 #define OBD_BRW_MEMALLOC 0x800 /* Client runs in the "kswapd" context */
1758 #define OBD_BRW_OVER_USRQUOTA 0x1000 /* Running out of user quota */
1759 #define OBD_BRW_OVER_GRPQUOTA 0x2000 /* Running out of group quota */
1760 #define OBD_BRW_SOFT_SYNC 0x4000 /* This flag notifies the server
1761 * that the client is running low on
1762 * space for unstable pages; asking
1763 * it to sync quickly */
1765 #define OBD_OBJECT_EOF LUSTRE_EOF
1767 #define OST_MIN_PRECREATE 32
1768 #define OST_MAX_PRECREATE 20000
1771 struct ost_id ioo_oid; /* object ID, if multi-obj BRW */
1772 __u32 ioo_max_brw; /* low 16 bits were o_mode before 2.4,
1773 * now (PTLRPC_BULK_OPS_COUNT - 1) in
1774 * high 16 bits in 2.4 and later */
1775 __u32 ioo_bufcnt; /* number of niobufs for this object */
1778 #define IOOBJ_MAX_BRW_BITS 16
1779 #define IOOBJ_TYPE_MASK ((1U << IOOBJ_MAX_BRW_BITS) - 1)
1780 #define ioobj_max_brw_get(ioo) (((ioo)->ioo_max_brw >> IOOBJ_MAX_BRW_BITS) + 1)
1781 #define ioobj_max_brw_set(ioo, num) \
1782 do { (ioo)->ioo_max_brw = ((num) - 1) << IOOBJ_MAX_BRW_BITS; } while (0)
1784 /* multiple of 8 bytes => can array */
1785 struct niobuf_remote {
1791 /* lock value block communicated between the filter and llite */
1793 /* OST_LVB_ERR_INIT is needed because the return code in rc is
1794 * negative, i.e. because ((MASK + rc) & MASK) != MASK. */
1795 #define OST_LVB_ERR_INIT 0xffbadbad80000000ULL
1796 #define OST_LVB_ERR_MASK 0xffbadbad00000000ULL
1797 #define OST_LVB_IS_ERR(blocks) \
1798 ((blocks & OST_LVB_ERR_MASK) == OST_LVB_ERR_MASK)
1799 #define OST_LVB_SET_ERR(blocks, rc) \
1800 do { blocks = OST_LVB_ERR_INIT + rc; } while (0)
1801 #define OST_LVB_GET_ERR(blocks) (int)(blocks - OST_LVB_ERR_INIT)
1824 * lquota data structures
1827 #ifndef QUOTABLOCK_BITS
1828 # define QUOTABLOCK_BITS LUSTRE_QUOTABLOCK_BITS
1831 #ifndef QUOTABLOCK_SIZE
1832 # define QUOTABLOCK_SIZE LUSTRE_QUOTABLOCK_SIZE
1836 # define toqb lustre_stoqb
1839 /* The lquota_id structure is an union of all the possible identifier types that
1840 * can be used with quota, this includes:
1843 * - a FID which can be used for per-directory quota in the future */
1845 struct lu_fid qid_fid; /* FID for per-directory quota */
1846 __u64 qid_uid; /* user identifier */
1847 __u64 qid_gid; /* group identifier */
1850 /* quotactl management */
1851 struct obd_quotactl {
1853 __u32 qc_type; /* see Q_* flag below */
1856 struct obd_dqinfo qc_dqinfo;
1857 struct obd_dqblk qc_dqblk;
1860 #define Q_COPY(out, in, member) (out)->member = (in)->member
1862 #define QCTL_COPY(out, in) \
1864 Q_COPY(out, in, qc_cmd); \
1865 Q_COPY(out, in, qc_type); \
1866 Q_COPY(out, in, qc_id); \
1867 Q_COPY(out, in, qc_stat); \
1868 Q_COPY(out, in, qc_dqinfo); \
1869 Q_COPY(out, in, qc_dqblk); \
1872 /* Body of quota request used for quota acquire/release RPCs between quota
1873 * master (aka QMT) and slaves (ak QSD). */
1875 struct lu_fid qb_fid; /* FID of global index packing the pool ID
1876 * and type (data or metadata) as well as
1877 * the quota type (user or group). */
1878 union lquota_id qb_id; /* uid or gid or directory FID */
1879 __u32 qb_flags; /* see below */
1881 __u64 qb_count; /* acquire/release count (kbytes/inodes) */
1882 __u64 qb_usage; /* current slave usage (kbytes/inodes) */
1883 __u64 qb_slv_ver; /* slave index file version */
1884 struct lustre_handle qb_lockh; /* per-ID lock handle */
1885 struct lustre_handle qb_glb_lockh; /* global lock handle */
1886 __u64 qb_padding1[4];
1889 /* When the quota_body is used in the reply of quota global intent
1890 * lock (IT_QUOTA_CONN) reply, qb_fid contains slave index file FID. */
1891 #define qb_slv_fid qb_fid
1892 /* qb_usage is the current qunit (in kbytes/inodes) when quota_body is used in
1894 #define qb_qunit qb_usage
1896 #define QUOTA_DQACQ_FL_ACQ 0x1 /* acquire quota */
1897 #define QUOTA_DQACQ_FL_PREACQ 0x2 /* pre-acquire */
1898 #define QUOTA_DQACQ_FL_REL 0x4 /* release quota */
1899 #define QUOTA_DQACQ_FL_REPORT 0x8 /* report usage */
1901 /* Quota types currently supported */
1903 LQUOTA_TYPE_USR = 0x00, /* maps to USRQUOTA */
1904 LQUOTA_TYPE_GRP = 0x01, /* maps to GRPQUOTA */
1908 /* There are 2 different resource types on which a quota limit can be enforced:
1909 * - inodes on the MDTs
1910 * - blocks on the OSTs */
1912 LQUOTA_RES_MD = 0x01, /* skip 0 to avoid null oid in FID */
1913 LQUOTA_RES_DT = 0x02,
1915 LQUOTA_FIRST_RES = LQUOTA_RES_MD
1917 #define LQUOTA_NR_RES (LQUOTA_LAST_RES - LQUOTA_FIRST_RES + 1)
1920 * Space accounting support
1921 * Format of an accounting record, providing disk usage information for a given
1924 struct lquota_acct_rec { /* 16 bytes */
1925 __u64 bspace; /* current space in use */
1926 __u64 ispace; /* current # inodes in use */
1930 * Global quota index support
1931 * Format of a global record, providing global quota settings for a given quota
1934 struct lquota_glb_rec { /* 32 bytes */
1935 __u64 qbr_hardlimit; /* quota hard limit, in #inodes or kbytes */
1936 __u64 qbr_softlimit; /* quota soft limit, in #inodes or kbytes */
1937 __u64 qbr_time; /* grace time, in seconds */
1938 __u64 qbr_granted; /* how much is granted to slaves, in #inodes or
1943 * Slave index support
1944 * Format of a slave record, recording how much space is granted to a given
1947 struct lquota_slv_rec { /* 8 bytes */
1948 __u64 qsr_granted; /* space granted to the slave for the key=ID,
1949 * in #inodes or kbytes */
1952 /* Data structures associated with the quota locks */
1954 /* Glimpse descriptor used for the index & per-ID quota locks */
1955 struct ldlm_gl_lquota_desc {
1956 union lquota_id gl_id; /* quota ID subject to the glimpse */
1957 __u64 gl_flags; /* see LQUOTA_FL* below */
1958 __u64 gl_ver; /* new index version */
1959 __u64 gl_hardlimit; /* new hardlimit or qunit value */
1960 __u64 gl_softlimit; /* new softlimit */
1964 #define gl_qunit gl_hardlimit /* current qunit value used when
1965 * glimpsing per-ID quota locks */
1967 /* quota glimpse flags */
1968 #define LQUOTA_FL_EDQUOT 0x1 /* user/group out of quota space on QMT */
1970 /* LVB used with quota (global and per-ID) locks */
1972 __u64 lvb_flags; /* see LQUOTA_FL* above */
1973 __u64 lvb_id_may_rel; /* space that might be released later */
1974 __u64 lvb_id_rel; /* space released by the slave for this ID */
1975 __u64 lvb_id_qunit; /* current qunit value */
1979 /* LVB used with global quota lock */
1980 #define lvb_glb_ver lvb_id_may_rel /* current version of the global index */
1988 #define QUOTA_FIRST_OPC QUOTA_DQACQ
1997 MDS_GETATTR_NAME = 34,
2002 MDS_DISCONNECT = 39,
2005 MDS_PIN = 42, /* obsolete, never used in a release */
2006 MDS_UNPIN = 43, /* obsolete, never used in a release */
2008 MDS_DONE_WRITING = 45, /* obsolete since 2.8.0 */
2010 MDS_QUOTACHECK = 47, /* not used since 2.4 */
2013 MDS_SETXATTR = 50, /* obsolete, now it's MDS_REINT op */
2015 MDS_IS_SUBDIR = 52, /* obsolete, never used in a release */
2017 MDS_HSM_STATE_GET = 54,
2018 MDS_HSM_STATE_SET = 55,
2019 MDS_HSM_ACTION = 56,
2020 MDS_HSM_PROGRESS = 57,
2021 MDS_HSM_REQUEST = 58,
2022 MDS_HSM_CT_REGISTER = 59,
2023 MDS_HSM_CT_UNREGISTER = 60,
2024 MDS_SWAP_LAYOUTS = 61,
2028 #define MDS_FIRST_OPC MDS_GETATTR
2031 /* opcodes for object update */
2037 #define OUT_UPDATE_FIRST_OPC OUT_UPDATE
2054 } mds_reint_t, mdt_reint_t;
2056 /* the disposition of the intent outlines what was executed */
2057 #define DISP_IT_EXECD 0x00000001
2058 #define DISP_LOOKUP_EXECD 0x00000002
2059 #define DISP_LOOKUP_NEG 0x00000004
2060 #define DISP_LOOKUP_POS 0x00000008
2061 #define DISP_OPEN_CREATE 0x00000010
2062 #define DISP_OPEN_OPEN 0x00000020
2063 #define DISP_ENQ_COMPLETE 0x00400000 /* obsolete and unused */
2064 #define DISP_ENQ_OPEN_REF 0x00800000
2065 #define DISP_ENQ_CREATE_REF 0x01000000
2066 #define DISP_OPEN_LOCK 0x02000000
2067 #define DISP_OPEN_LEASE 0x04000000
2068 #define DISP_OPEN_STRIPE 0x08000000
2069 #define DISP_OPEN_DENY 0x10000000
2071 /* INODE LOCK PARTS */
2072 #define MDS_INODELOCK_LOOKUP 0x000001 /* For namespace, dentry etc, and also
2073 * was used to protect permission (mode,
2074 * owner, group etc) before 2.4. */
2075 #define MDS_INODELOCK_UPDATE 0x000002 /* size, links, timestamps */
2076 #define MDS_INODELOCK_OPEN 0x000004 /* For opened files */
2077 #define MDS_INODELOCK_LAYOUT 0x000008 /* for layout */
2079 /* The PERM bit is added int 2.4, and it is used to protect permission(mode,
2080 * owner, group, acl etc), so to separate the permission from LOOKUP lock.
2081 * Because for remote directories(in DNE), these locks will be granted by
2082 * different MDTs(different ldlm namespace).
2084 * For local directory, MDT will always grant UPDATE_LOCK|PERM_LOCK together.
2085 * For Remote directory, the master MDT, where the remote directory is, will
2086 * grant UPDATE_LOCK|PERM_LOCK, and the remote MDT, where the name entry is,
2087 * will grant LOOKUP_LOCK. */
2088 #define MDS_INODELOCK_PERM 0x000010
2089 #define MDS_INODELOCK_XATTR 0x000020 /* extended attributes */
2091 #define MDS_INODELOCK_MAXSHIFT 5
2092 /* This FULL lock is useful to take on unlink sort of operations */
2093 #define MDS_INODELOCK_FULL ((1<<(MDS_INODELOCK_MAXSHIFT+1))-1)
2095 /* NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
2096 * but was moved into name[1] along with the OID to avoid consuming the
2097 * name[2,3] fields that need to be used for the quota id (also a FID). */
2099 LUSTRE_RES_ID_SEQ_OFF = 0,
2100 LUSTRE_RES_ID_VER_OID_OFF = 1,
2101 LUSTRE_RES_ID_WAS_VER_OFF = 2, /* see note above */
2102 LUSTRE_RES_ID_QUOTA_SEQ_OFF = 2,
2103 LUSTRE_RES_ID_QUOTA_VER_OID_OFF = 3,
2104 LUSTRE_RES_ID_HSH_OFF = 3
2107 #define MDS_STATUS_CONN 1
2108 #define MDS_STATUS_LOV 2
2110 /* these should be identical to their EXT4_*_FL counterparts, they are
2111 * redefined here only to avoid dragging in fs/ext4/ext4.h */
2112 #define LUSTRE_SYNC_FL 0x00000008 /* Synchronous updates */
2113 #define LUSTRE_IMMUTABLE_FL 0x00000010 /* Immutable file */
2114 #define LUSTRE_APPEND_FL 0x00000020 /* writes to file may only append */
2115 #define LUSTRE_NODUMP_FL 0x00000040 /* do not dump file */
2116 #define LUSTRE_NOATIME_FL 0x00000080 /* do not update atime */
2117 #define LUSTRE_INDEX_FL 0x00001000 /* hash-indexed directory */
2118 #define LUSTRE_DIRSYNC_FL 0x00010000 /* dirsync behaviour (dir only) */
2119 #define LUSTRE_TOPDIR_FL 0x00020000 /* Top of directory hierarchies*/
2120 #define LUSTRE_DIRECTIO_FL 0x00100000 /* Use direct i/o */
2121 #define LUSTRE_INLINE_DATA_FL 0x10000000 /* Inode has inline data. */
2124 /* Convert wire LUSTRE_*_FL to corresponding client local VFS S_* values
2125 * for the client inode i_flags. The LUSTRE_*_FL are the Lustre wire
2126 * protocol equivalents of LDISKFS_*_FL values stored on disk, while
2127 * the S_* flags are kernel-internal values that change between kernel
2128 * versions. These flags are set/cleared via FSFILT_IOC_{GET,SET}_FLAGS.
2129 * See b=16526 for a full history. */
2130 static inline int ll_ext_to_inode_flags(int flags)
2132 return (((flags & LUSTRE_SYNC_FL) ? S_SYNC : 0) |
2133 ((flags & LUSTRE_NOATIME_FL) ? S_NOATIME : 0) |
2134 ((flags & LUSTRE_APPEND_FL) ? S_APPEND : 0) |
2135 #if defined(S_DIRSYNC)
2136 ((flags & LUSTRE_DIRSYNC_FL) ? S_DIRSYNC : 0) |
2138 ((flags & LUSTRE_IMMUTABLE_FL) ? S_IMMUTABLE : 0));
2141 static inline int ll_inode_to_ext_flags(int iflags)
2143 return (((iflags & S_SYNC) ? LUSTRE_SYNC_FL : 0) |
2144 ((iflags & S_NOATIME) ? LUSTRE_NOATIME_FL : 0) |
2145 ((iflags & S_APPEND) ? LUSTRE_APPEND_FL : 0) |
2146 #if defined(S_DIRSYNC)
2147 ((iflags & S_DIRSYNC) ? LUSTRE_DIRSYNC_FL : 0) |
2149 ((iflags & S_IMMUTABLE) ? LUSTRE_IMMUTABLE_FL : 0));
2153 /* 64 possible states */
2154 enum md_transient_state {
2155 MS_RESTORE = (1 << 0), /* restore is running */
2159 struct lu_fid mbo_fid1;
2160 struct lu_fid mbo_fid2;
2161 struct lustre_handle mbo_handle;
2163 __u64 mbo_size; /* Offset, in the case of MDS_READPAGE */
2167 __u64 mbo_blocks; /* XID, in the case of MDS_READPAGE */
2169 __u64 mbo_t_state; /* transient file state defined in
2170 * enum md_transient_state
2171 * was "ino" until 2.4.0 */
2174 __u32 mbo_capability;
2178 __u32 mbo_flags; /* LUSTRE_*_FL file attributes */
2180 __u32 mbo_nlink; /* #bytes to read in the case of MDS_READPAGE */
2181 __u32 mbo_unused2; /* was "generation" until 2.4.0 */
2183 __u32 mbo_eadatasize;
2185 __u32 mbo_max_mdsize;
2186 __u32 mbo_unused3; /* was max_cookiesize until 2.8 */
2187 __u32 mbo_uid_h; /* high 32-bits of uid, for FUID */
2188 __u32 mbo_gid_h; /* high 32-bits of gid, for FUID */
2189 __u32 mbo_padding_5; /* also fix lustre_swab_mdt_body */
2190 __u64 mbo_padding_6;
2191 __u64 mbo_padding_7;
2192 __u64 mbo_padding_8;
2193 __u64 mbo_padding_9;
2194 __u64 mbo_padding_10;
2197 struct mdt_ioepoch {
2198 struct lustre_handle mio_handle;
2199 __u64 mio_unused1; /* was ioepoch */
2200 __u32 mio_unused2; /* was flags */
2204 /* permissions for md_perm.mp_perm */
2206 CFS_SETUID_PERM = 0x01,
2207 CFS_SETGID_PERM = 0x02,
2208 CFS_SETGRP_PERM = 0x04,
2209 CFS_RMTACL_PERM = 0x08,
2210 CFS_RMTOWN_PERM = 0x10
2213 /* inode access permission for remote user, the inode info are omitted,
2214 * for client knows them. */
2215 struct mdt_remote_perm {
2222 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
2226 struct mdt_rec_setattr {
2236 __u32 sa_padding_1_h;
2237 struct lu_fid sa_fid;
2246 __u32 sa_attr_flags;
2248 __u32 sa_bias; /* some operation flags */
2255 * Attribute flags used in mdt_rec_setattr::sa_valid.
2256 * The kernel's #defines for ATTR_* should not be used over the network
2257 * since the client and MDS may run different kernels (see bug 13828)
2258 * Therefore, we should only use MDS_ATTR_* attributes for sa_valid.
2260 #define MDS_ATTR_MODE 0x1ULL /* = 1 */
2261 #define MDS_ATTR_UID 0x2ULL /* = 2 */
2262 #define MDS_ATTR_GID 0x4ULL /* = 4 */
2263 #define MDS_ATTR_SIZE 0x8ULL /* = 8 */
2264 #define MDS_ATTR_ATIME 0x10ULL /* = 16 */
2265 #define MDS_ATTR_MTIME 0x20ULL /* = 32 */
2266 #define MDS_ATTR_CTIME 0x40ULL /* = 64 */
2267 #define MDS_ATTR_ATIME_SET 0x80ULL /* = 128 */
2268 #define MDS_ATTR_MTIME_SET 0x100ULL /* = 256 */
2269 #define MDS_ATTR_FORCE 0x200ULL /* = 512, Not a change, but a change it */
2270 #define MDS_ATTR_ATTR_FLAG 0x400ULL /* = 1024 */
2271 #define MDS_ATTR_KILL_SUID 0x800ULL /* = 2048 */
2272 #define MDS_ATTR_KILL_SGID 0x1000ULL /* = 4096 */
2273 #define MDS_ATTR_CTIME_SET 0x2000ULL /* = 8192 */
2274 #define MDS_ATTR_FROM_OPEN 0x4000ULL /* = 16384, called from open path, ie O_TRUNC */
2275 #define MDS_ATTR_BLOCKS 0x8000ULL /* = 32768 */
2278 #define FMODE_READ 00000001
2279 #define FMODE_WRITE 00000002
2282 #define MDS_FMODE_CLOSED 00000000
2283 #define MDS_FMODE_EXEC 00000004
2284 /* MDS_FMODE_EPOCH 01000000 obsolete since 2.8.0 */
2285 /* MDS_FMODE_TRUNC 02000000 obsolete since 2.8.0 */
2286 /* MDS_FMODE_SOM 04000000 obsolete since 2.8.0 */
2288 #define MDS_OPEN_CREATED 00000010
2289 #define MDS_OPEN_CROSS 00000020
2291 #define MDS_OPEN_CREAT 00000100
2292 #define MDS_OPEN_EXCL 00000200
2293 #define MDS_OPEN_TRUNC 00001000
2294 #define MDS_OPEN_APPEND 00002000
2295 #define MDS_OPEN_SYNC 00010000
2296 #define MDS_OPEN_DIRECTORY 00200000
2298 #define MDS_OPEN_BY_FID 040000000 /* open_by_fid for known object */
2299 #define MDS_OPEN_DELAY_CREATE 0100000000 /* delay initial object create */
2300 #define MDS_OPEN_OWNEROVERRIDE 0200000000 /* NFSD rw-reopen ro file for owner */
2301 #define MDS_OPEN_JOIN_FILE 0400000000 /* open for join file.
2302 * We do not support JOIN FILE
2303 * anymore, reserve this flags
2304 * just for preventing such bit
2307 #define MDS_OPEN_LOCK 04000000000 /* This open requires open lock */
2308 #define MDS_OPEN_HAS_EA 010000000000 /* specify object create pattern */
2309 #define MDS_OPEN_HAS_OBJS 020000000000 /* Just set the EA the obj exist */
2310 #define MDS_OPEN_NORESTORE 0100000000000ULL /* Do not restore file at open */
2311 #define MDS_OPEN_NEWSTRIPE 0200000000000ULL /* New stripe needed (restripe or
2313 #define MDS_OPEN_VOLATILE 0400000000000ULL /* File is volatile = created
2315 #define MDS_OPEN_LEASE 01000000000000ULL /* Open the file and grant lease
2316 * delegation, succeed if it's not
2317 * being opened with conflict mode.
2319 #define MDS_OPEN_RELEASE 02000000000000ULL /* Open the file for HSM release */
2321 /* lustre internal open flags, which should not be set from user space */
2322 #define MDS_OPEN_FL_INTERNAL (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS | \
2323 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK | \
2324 MDS_OPEN_BY_FID | MDS_OPEN_LEASE | \
2328 MDS_CHECK_SPLIT = 1 << 0,
2329 MDS_CROSS_REF = 1 << 1,
2330 MDS_VTX_BYPASS = 1 << 2,
2331 MDS_PERM_BYPASS = 1 << 3,
2332 /* MDS_SOM = 1 << 4, obsolete since 2.8.0 */
2333 MDS_QUOTA_IGNORE = 1 << 5,
2334 /* Was MDS_CLOSE_CLEANUP (1 << 6), No more used */
2335 MDS_KEEP_ORPHAN = 1 << 7,
2336 MDS_RECOV_OPEN = 1 << 8,
2337 MDS_DATA_MODIFIED = 1 << 9,
2338 MDS_CREATE_VOLATILE = 1 << 10,
2339 MDS_OWNEROVERRIDE = 1 << 11,
2340 MDS_HSM_RELEASE = 1 << 12,
2341 MDS_RENAME_MIGRATE = 1 << 13,
2342 MDS_CLOSE_LAYOUT_SWAP = 1 << 14,
2345 /* instance of mdt_reint_rec */
2346 struct mdt_rec_create {
2354 __u32 cr_suppgid1_h;
2356 __u32 cr_suppgid2_h;
2357 struct lu_fid cr_fid1;
2358 struct lu_fid cr_fid2;
2359 struct lustre_handle cr_old_handle; /* handle in case of open replay */
2363 __u64 cr_padding_1; /* rr_blocks */
2366 /* use of helpers set/get_mrc_cr_flags() is needed to access
2367 * 64 bits cr_flags [cr_flags_l, cr_flags_h], this is done to
2368 * extend cr_flags size without breaking 1.8 compat */
2369 __u32 cr_flags_l; /* for use with open, low 32 bits */
2370 __u32 cr_flags_h; /* for use with open, high 32 bits */
2371 __u32 cr_umask; /* umask for create */
2372 __u32 cr_padding_4; /* rr_padding_4 */
2375 static inline void set_mrc_cr_flags(struct mdt_rec_create *mrc, __u64 flags)
2377 mrc->cr_flags_l = (__u32)(flags & 0xFFFFFFFFUll);
2378 mrc->cr_flags_h = (__u32)(flags >> 32);
2381 static inline __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
2383 return ((__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32));
2386 /* instance of mdt_reint_rec */
2387 struct mdt_rec_link {
2395 __u32 lk_suppgid1_h;
2397 __u32 lk_suppgid2_h;
2398 struct lu_fid lk_fid1;
2399 struct lu_fid lk_fid2;
2401 __u64 lk_padding_1; /* rr_atime */
2402 __u64 lk_padding_2; /* rr_ctime */
2403 __u64 lk_padding_3; /* rr_size */
2404 __u64 lk_padding_4; /* rr_blocks */
2406 __u32 lk_padding_5; /* rr_mode */
2407 __u32 lk_padding_6; /* rr_flags */
2408 __u32 lk_padding_7; /* rr_padding_2 */
2409 __u32 lk_padding_8; /* rr_padding_3 */
2410 __u32 lk_padding_9; /* rr_padding_4 */
2413 /* instance of mdt_reint_rec */
2414 struct mdt_rec_unlink {
2422 __u32 ul_suppgid1_h;
2424 __u32 ul_suppgid2_h;
2425 struct lu_fid ul_fid1;
2426 struct lu_fid ul_fid2;
2428 __u64 ul_padding_2; /* rr_atime */
2429 __u64 ul_padding_3; /* rr_ctime */
2430 __u64 ul_padding_4; /* rr_size */
2431 __u64 ul_padding_5; /* rr_blocks */
2434 __u32 ul_padding_6; /* rr_flags */
2435 __u32 ul_padding_7; /* rr_padding_2 */
2436 __u32 ul_padding_8; /* rr_padding_3 */
2437 __u32 ul_padding_9; /* rr_padding_4 */
2440 /* instance of mdt_reint_rec */
2441 struct mdt_rec_rename {
2449 __u32 rn_suppgid1_h;
2451 __u32 rn_suppgid2_h;
2452 struct lu_fid rn_fid1;
2453 struct lu_fid rn_fid2;
2455 __u64 rn_padding_1; /* rr_atime */
2456 __u64 rn_padding_2; /* rr_ctime */
2457 __u64 rn_padding_3; /* rr_size */
2458 __u64 rn_padding_4; /* rr_blocks */
2459 __u32 rn_bias; /* some operation flags */
2460 __u32 rn_mode; /* cross-ref rename has mode */
2461 __u32 rn_padding_5; /* rr_flags */
2462 __u32 rn_padding_6; /* rr_padding_2 */
2463 __u32 rn_padding_7; /* rr_padding_3 */
2464 __u32 rn_padding_8; /* rr_padding_4 */
2467 /* instance of mdt_reint_rec */
2468 struct mdt_rec_setxattr {
2476 __u32 sx_suppgid1_h;
2478 __u32 sx_suppgid2_h;
2479 struct lu_fid sx_fid;
2480 __u64 sx_padding_1; /* These three are rr_fid2 */
2485 __u64 sx_padding_5; /* rr_ctime */
2486 __u64 sx_padding_6; /* rr_size */
2487 __u64 sx_padding_7; /* rr_blocks */
2490 __u32 sx_padding_8; /* rr_flags */
2491 __u32 sx_padding_9; /* rr_padding_2 */
2492 __u32 sx_padding_10; /* rr_padding_3 */
2493 __u32 sx_padding_11; /* rr_padding_4 */
2497 * mdt_rec_reint is the template for all mdt_reint_xxx structures.
2498 * Do NOT change the size of various members, otherwise the value
2499 * will be broken in lustre_swab_mdt_rec_reint().
2501 * If you add new members in other mdt_reint_xxx structres and need to use the
2502 * rr_padding_x fields, then update lustre_swab_mdt_rec_reint() also.
2504 struct mdt_rec_reint {
2512 __u32 rr_suppgid1_h;
2514 __u32 rr_suppgid2_h;
2515 struct lu_fid rr_fid1;
2516 struct lu_fid rr_fid2;
2527 __u32 rr_padding_4; /* also fix lustre_swab_mdt_rec_reint */
2530 /* lmv structures */
2532 __u32 ld_tgt_count; /* how many MDS's */
2533 __u32 ld_active_tgt_count; /* how many active */
2534 __u32 ld_default_stripe_count; /* how many objects are used */
2535 __u32 ld_pattern; /* default hash pattern */
2536 __u64 ld_default_hash_size;
2537 __u64 ld_padding_1; /* also fix lustre_swab_lmv_desc */
2538 __u32 ld_padding_2; /* also fix lustre_swab_lmv_desc */
2539 __u32 ld_qos_maxage; /* in second */
2540 __u32 ld_padding_3; /* also fix lustre_swab_lmv_desc */
2541 __u32 ld_padding_4; /* also fix lustre_swab_lmv_desc */
2542 struct obd_uuid ld_uuid;
2545 /* LMV layout EA, and it will be stored both in master and slave object */
2546 struct lmv_mds_md_v1 {
2548 __u32 lmv_stripe_count;
2549 __u32 lmv_master_mdt_index; /* On master object, it is master
2550 * MDT index, on slave object, it
2551 * is stripe index of the slave obj */
2552 __u32 lmv_hash_type; /* dir stripe policy, i.e. indicate
2553 * which hash function to be used,
2554 * Note: only lower 16 bits is being
2555 * used for now. Higher 16 bits will
2556 * be used to mark the object status,
2557 * for example migrating or dead. */
2558 __u32 lmv_layout_version; /* Used for directory restriping */
2562 char lmv_pool_name[LOV_MAXPOOLNAME + 1]; /* pool name */
2563 struct lu_fid lmv_stripe_fids[0]; /* FIDs for each stripe */
2566 #define LMV_MAGIC_V1 0x0CD20CD0 /* normal stripe lmv magic */
2567 #define LMV_MAGIC LMV_MAGIC_V1
2569 /* #define LMV_USER_MAGIC 0x0CD30CD0 */
2570 #define LMV_MAGIC_STRIPE 0x0CD40CD0 /* magic for dir sub_stripe */
2572 /* Right now only the lower part(0-16bits) of lmv_hash_type is being used,
2573 * and the higher part will be the flag to indicate the status of object,
2574 * for example the object is being migrated. And the hash function
2575 * might be interpreted differently with different flags. */
2576 #define LMV_HASH_TYPE_MASK 0x0000ffff
2578 #define LMV_HASH_FLAG_MIGRATION 0x80000000
2579 #define LMV_HASH_FLAG_DEAD 0x40000000
2580 #define LMV_HASH_FLAG_BAD_TYPE 0x20000000
2582 /* The striped directory has ever lost its master LMV EA, then LFSCK
2583 * re-generated it. This flag is used to indicate such case. It is an
2585 #define LMV_HASH_FLAG_LOST_LMV 0x10000000
2588 * The FNV-1a hash algorithm is as follows:
2589 * hash = FNV_offset_basis
2590 * for each octet_of_data to be hashed
2591 * hash = hash XOR octet_of_data
2592 * hash = hash × FNV_prime
2594 * http://en.wikipedia.org/wiki/Fowler–Noll–Vo_hash_function#FNV-1a_hash
2596 * http://www.isthe.com/chongo/tech/comp/fnv/index.html#FNV-reference-source
2597 * FNV_prime is 2^40 + 2^8 + 0xb3 = 0x100000001b3ULL
2599 #define LUSTRE_FNV_1A_64_PRIME 0x100000001b3ULL
2600 #define LUSTRE_FNV_1A_64_OFFSET_BIAS 0xcbf29ce484222325ULL
2601 static inline __u64 lustre_hash_fnv_1a_64(const void *buf, size_t size)
2603 __u64 hash = LUSTRE_FNV_1A_64_OFFSET_BIAS;
2604 const unsigned char *p = buf;
2607 for (i = 0; i < size; i++) {
2609 hash *= LUSTRE_FNV_1A_64_PRIME;
2617 struct lmv_mds_md_v1 lmv_md_v1;
2618 struct lmv_user_md lmv_user_md;
2621 static inline int lmv_mds_md_size(int stripe_count, unsigned int lmm_magic)
2623 switch (lmm_magic) {
2625 struct lmv_mds_md_v1 *lmm1;
2627 return sizeof(*lmm1) + stripe_count *
2628 sizeof(lmm1->lmv_stripe_fids[0]);
2635 static inline int lmv_mds_md_stripe_count_get(const union lmv_mds_md *lmm)
2637 switch (le32_to_cpu(lmm->lmv_magic)) {
2639 return le32_to_cpu(lmm->lmv_md_v1.lmv_stripe_count);
2640 case LMV_USER_MAGIC:
2641 return le32_to_cpu(lmm->lmv_user_md.lum_stripe_count);
2647 static inline int lmv_mds_md_stripe_count_set(union lmv_mds_md *lmm,
2648 unsigned int stripe_count)
2650 switch (le32_to_cpu(lmm->lmv_magic)) {
2652 lmm->lmv_md_v1.lmv_stripe_count = cpu_to_le32(stripe_count);
2654 case LMV_USER_MAGIC:
2655 lmm->lmv_user_md.lum_stripe_count = cpu_to_le32(stripe_count);
2667 FLD_FIRST_OPC = FLD_QUERY
2673 SEQ_FIRST_OPC = SEQ_QUERY
2677 SEQ_ALLOC_SUPER = 0,
2689 LFSCK_NOTIFY = 1101,
2692 LFSCK_FIRST_OPC = LFSCK_NOTIFY
2696 * LOV data structures
2699 #define LOV_MAX_UUID_BUFFER_SIZE 8192
2700 /* The size of the buffer the lov/mdc reserves for the
2701 * array of UUIDs returned by the MDS. With the current
2702 * protocol, this will limit the max number of OSTs per LOV */
2704 #define LOV_DESC_MAGIC 0xB0CCDE5C
2705 #define LOV_DESC_QOS_MAXAGE_DEFAULT 5 /* Seconds */
2706 #define LOV_DESC_STRIPE_SIZE_DEFAULT (1 << LNET_MTU_BITS)
2708 /* LOV settings descriptor (should only contain static info) */
2710 __u32 ld_tgt_count; /* how many OBD's */
2711 __u32 ld_active_tgt_count; /* how many active */
2712 __u32 ld_default_stripe_count; /* how many objects are used */
2713 __u32 ld_pattern; /* default PATTERN_RAID0 */
2714 __u64 ld_default_stripe_size; /* in bytes */
2715 __u64 ld_default_stripe_offset; /* in bytes */
2716 __u32 ld_padding_0; /* unused */
2717 __u32 ld_qos_maxage; /* in second */
2718 __u32 ld_padding_1; /* also fix lustre_swab_lov_desc */
2719 __u32 ld_padding_2; /* also fix lustre_swab_lov_desc */
2720 struct obd_uuid ld_uuid;
2723 #define ld_magic ld_active_tgt_count /* for swabbing from llogs */
2728 /* opcodes -- MUST be distinct from OST/MDS opcodes */
2733 LDLM_BL_CALLBACK = 104,
2734 LDLM_CP_CALLBACK = 105,
2735 LDLM_GL_CALLBACK = 106,
2736 LDLM_SET_INFO = 107,
2739 #define LDLM_FIRST_OPC LDLM_ENQUEUE
2741 #define RES_NAME_SIZE 4
2742 struct ldlm_res_id {
2743 __u64 name[RES_NAME_SIZE];
2746 #define DLDLMRES "[%#llx:%#llx:%#llx].%#llx"
2747 #define PLDLMRES(res) (unsigned long long)(res)->lr_name.name[0], \
2748 (unsigned long long)(res)->lr_name.name[1], \
2749 (unsigned long long)(res)->lr_name.name[2], \
2750 (unsigned long long)(res)->lr_name.name[3]
2752 static inline bool ldlm_res_eq(const struct ldlm_res_id *res0,
2753 const struct ldlm_res_id *res1)
2755 return memcmp(res0, res1, sizeof(*res0)) == 0;
2759 typedef enum ldlm_mode {
2772 #define LCK_MODE_NUM 8
2774 typedef enum ldlm_type {
2782 #define LDLM_MIN_TYPE LDLM_PLAIN
2784 struct ldlm_extent {
2790 static inline int ldlm_extent_overlap(const struct ldlm_extent *ex1,
2791 const struct ldlm_extent *ex2)
2793 return ex1->start <= ex2->end && ex2->start <= ex1->end;
2796 /* check if @ex1 contains @ex2 */
2797 static inline int ldlm_extent_contain(const struct ldlm_extent *ex1,
2798 const struct ldlm_extent *ex2)
2800 return ex1->start <= ex2->start && ex1->end >= ex2->end;
2803 struct ldlm_inodebits {
2807 struct ldlm_flock_wire {
2815 /* it's important that the fields of the ldlm_extent structure match
2816 * the first fields of the ldlm_flock structure because there is only
2817 * one ldlm_swab routine to process the ldlm_policy_data_t union. if
2818 * this ever changes we will need to swab the union differently based
2819 * on the resource type. */
2821 typedef union ldlm_wire_policy_data {
2822 struct ldlm_extent l_extent;
2823 struct ldlm_flock_wire l_flock;
2824 struct ldlm_inodebits l_inodebits;
2825 } ldlm_wire_policy_data_t;
2827 union ldlm_gl_desc {
2828 struct ldlm_gl_lquota_desc lquota_desc;
2831 enum ldlm_intent_flags {
2832 IT_OPEN = 0x00000001,
2833 IT_CREAT = 0x00000002,
2834 IT_OPEN_CREAT = 0x00000003,
2835 IT_READDIR = 0x00000004,
2836 IT_GETATTR = 0x00000008,
2837 IT_LOOKUP = 0x00000010,
2838 IT_UNLINK = 0x00000020,
2839 IT_TRUNC = 0x00000040,
2840 IT_GETXATTR = 0x00000080,
2841 IT_EXEC = 0x00000100,
2842 IT_PIN = 0x00000200,
2843 IT_LAYOUT = 0x00000400,
2844 IT_QUOTA_DQACQ = 0x00000800,
2845 IT_QUOTA_CONN = 0x00001000,
2846 IT_SETXATTR = 0x00002000,
2849 struct ldlm_intent {
2853 struct ldlm_resource_desc {
2854 enum ldlm_type lr_type;
2855 __u32 lr_pad; /* also fix lustre_swab_ldlm_resource_desc */
2856 struct ldlm_res_id lr_name;
2859 struct ldlm_lock_desc {
2860 struct ldlm_resource_desc l_resource;
2861 enum ldlm_mode l_req_mode;
2862 enum ldlm_mode l_granted_mode;
2863 union ldlm_wire_policy_data l_policy_data;
2866 #define LDLM_LOCKREQ_HANDLES 2
2867 #define LDLM_ENQUEUE_CANCEL_OFF 1
2869 struct ldlm_request {
2872 struct ldlm_lock_desc lock_desc;
2873 struct lustre_handle lock_handle[LDLM_LOCKREQ_HANDLES];
2876 /* If LDLM_ENQUEUE, 1 slot is already occupied, 1 is available.
2877 * Otherwise, 2 are available. */
2878 #define ldlm_request_bufsize(count,type) \
2880 int _avail = LDLM_LOCKREQ_HANDLES; \
2881 _avail -= (type == LDLM_ENQUEUE ? LDLM_ENQUEUE_CANCEL_OFF : 0); \
2882 sizeof(struct ldlm_request) + \
2883 (count > _avail ? count - _avail : 0) * \
2884 sizeof(struct lustre_handle); \
2889 __u32 lock_padding; /* also fix lustre_swab_ldlm_reply */
2890 struct ldlm_lock_desc lock_desc;
2891 struct lustre_handle lock_handle;
2892 __u64 lock_policy_res1;
2893 __u64 lock_policy_res2;
2896 #define ldlm_flags_to_wire(flags) ((__u32)(flags))
2897 #define ldlm_flags_from_wire(flags) ((__u64)(flags))
2900 * Opcodes for mountconf (mgs and mgc)
2905 MGS_EXCEPTION, /* node died, etc. */
2906 MGS_TARGET_REG, /* whenever target starts up */
2912 #define MGS_FIRST_OPC MGS_CONNECT
2914 #define MGS_PARAM_MAXLEN 1024
2915 #define KEY_SET_INFO "set_info"
2917 struct mgs_send_param {
2918 char mgs_param[MGS_PARAM_MAXLEN];
2921 /* We pass this info to the MGS so it can write config logs */
2922 #define MTI_NAME_MAXLEN 64
2923 #define MTI_PARAM_MAXLEN 4096
2924 #define MTI_NIDS_MAX 32
2925 struct mgs_target_info {
2926 __u32 mti_lustre_ver;
2927 __u32 mti_stripe_index;
2928 __u32 mti_config_ver;
2930 __u32 mti_nid_count;
2931 __u32 mti_instance; /* Running instance of target */
2932 char mti_fsname[MTI_NAME_MAXLEN];
2933 char mti_svname[MTI_NAME_MAXLEN];
2934 char mti_uuid[sizeof(struct obd_uuid)];
2935 __u64 mti_nids[MTI_NIDS_MAX]; /* host nids (lnet_nid_t)*/
2936 char mti_params[MTI_PARAM_MAXLEN];
2939 struct mgs_nidtbl_entry {
2940 __u64 mne_version; /* table version of this entry */
2941 __u32 mne_instance; /* target instance # */
2942 __u32 mne_index; /* target index */
2943 __u32 mne_length; /* length of this entry - by bytes */
2944 __u8 mne_type; /* target type LDD_F_SV_TYPE_OST/MDT */
2945 __u8 mne_nid_type; /* type of nid(mbz). for ipv6. */
2946 __u8 mne_nid_size; /* size of each NID, by bytes */
2947 __u8 mne_nid_count; /* # of NIDs in buffer */
2949 lnet_nid_t nids[0]; /* variable size buffer for NIDs. */
2953 struct mgs_config_body {
2954 char mcb_name[MTI_NAME_MAXLEN]; /* logname */
2955 __u64 mcb_offset; /* next index of config log to request */
2956 __u16 mcb_type; /* type of log: CONFIG_T_[CONFIG|RECOVER] */
2958 __u8 mcb_bits; /* bits unit size of config log */
2959 __u32 mcb_units; /* # of units for bulk transfer */
2962 struct mgs_config_res {
2963 __u64 mcr_offset; /* index of last config log */
2964 __u64 mcr_size; /* size of the log */
2967 /* Config marker flags (in config log) */
2968 #define CM_START 0x01
2970 #define CM_SKIP 0x04
2971 #define CM_UPGRADE146 0x08
2972 #define CM_EXCLUDE 0x10
2973 #define CM_START_SKIP (CM_START | CM_SKIP)
2976 __u32 cm_step; /* aka config version */
2978 __u32 cm_vers; /* lustre release version number */
2979 __u32 cm_padding; /* 64 bit align */
2980 __s64 cm_createtime; /*when this record was first created */
2981 __s64 cm_canceltime; /*when this record is no longer valid*/
2982 char cm_tgtname[MTI_NAME_MAXLEN];
2983 char cm_comment[MTI_NAME_MAXLEN];
2987 * Opcodes for multiple servers.
2993 OBD_QC_CALLBACK, /* not used since 2.4 */
2997 #define OBD_FIRST_OPC OBD_PING
3000 * llog contexts indices.
3002 * There is compatibility problem with indexes below, they are not
3003 * continuous and must keep their numbers for compatibility needs.
3004 * See LU-5218 for details.
3007 LLOG_CONFIG_ORIG_CTXT = 0,
3008 LLOG_CONFIG_REPL_CTXT = 1,
3009 LLOG_MDS_OST_ORIG_CTXT = 2,
3010 LLOG_MDS_OST_REPL_CTXT = 3, /* kept just to avoid re-assignment */
3011 LLOG_SIZE_ORIG_CTXT = 4,
3012 LLOG_SIZE_REPL_CTXT = 5,
3013 LLOG_TEST_ORIG_CTXT = 8,
3014 LLOG_TEST_REPL_CTXT = 9, /* kept just to avoid re-assignment */
3015 LLOG_CHANGELOG_ORIG_CTXT = 12, /**< changelog generation on mdd */
3016 LLOG_CHANGELOG_REPL_CTXT = 13, /**< changelog access on clients */
3017 /* for multiple changelog consumers */
3018 LLOG_CHANGELOG_USER_ORIG_CTXT = 14,
3019 LLOG_AGENT_ORIG_CTXT = 15, /**< agent requests generation on cdt */
3020 LLOG_UPDATELOG_ORIG_CTXT = 16, /* update log */
3021 LLOG_UPDATELOG_REPL_CTXT = 17, /* update log */
3025 /** Identifier for a single log object */
3027 struct ost_id lgl_oi;
3029 } __attribute__((packed));
3031 /** Records written to the CATALOGS list */
3032 #define CATLIST "CATALOGS"
3034 struct llog_logid lci_logid;
3038 } __attribute__((packed));
3040 /* Log data record types - there is no specific reason that these need to
3041 * be related to the RPC opcodes, but no reason not to (may be handy later?)
3043 #define LLOG_OP_MAGIC 0x10600000
3044 #define LLOG_OP_MASK 0xfff00000
3047 LLOG_PAD_MAGIC = LLOG_OP_MAGIC | 0x00000,
3048 OST_SZ_REC = LLOG_OP_MAGIC | 0x00f00,
3049 /* OST_RAID1_REC = LLOG_OP_MAGIC | 0x01000, never used */
3050 MDS_UNLINK_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) |
3051 REINT_UNLINK, /* obsolete after 2.5.0 */
3052 MDS_UNLINK64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
3054 /* MDS_SETATTR_REC = LLOG_OP_MAGIC | 0x12401, obsolete 1.8.0 */
3055 MDS_SETATTR64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
3057 OBD_CFG_REC = LLOG_OP_MAGIC | 0x20000,
3058 /* PTL_CFG_REC = LLOG_OP_MAGIC | 0x30000, obsolete 1.4.0 */
3059 LLOG_GEN_REC = LLOG_OP_MAGIC | 0x40000,
3060 /* LLOG_JOIN_REC = LLOG_OP_MAGIC | 0x50000, obsolete 1.8.0 */
3061 CHANGELOG_REC = LLOG_OP_MAGIC | 0x60000,
3062 CHANGELOG_USER_REC = LLOG_OP_MAGIC | 0x70000,
3063 HSM_AGENT_REC = LLOG_OP_MAGIC | 0x80000,
3064 UPDATE_REC = LLOG_OP_MAGIC | 0xa0000,
3065 LLOG_HDR_MAGIC = LLOG_OP_MAGIC | 0x45539,
3066 LLOG_LOGID_MAGIC = LLOG_OP_MAGIC | 0x4553b,
3069 #define LLOG_REC_HDR_NEEDS_SWABBING(r) \
3070 (((r)->lrh_type & __swab32(LLOG_OP_MASK)) == __swab32(LLOG_OP_MAGIC))
3072 /** Log record header - stored in little endian order.
3073 * Each record must start with this struct, end with a llog_rec_tail,
3074 * and be a multiple of 256 bits in size.
3076 struct llog_rec_hdr {
3083 struct llog_rec_tail {
3088 /* Where data follow just after header */
3089 #define REC_DATA(ptr) \
3090 ((void *)((char *)ptr + sizeof(struct llog_rec_hdr)))
3092 #define REC_DATA_LEN(rec) \
3093 (rec->lrh_len - sizeof(struct llog_rec_hdr) - \
3094 sizeof(struct llog_rec_tail))
3096 static inline void *rec_tail(struct llog_rec_hdr *rec)
3098 return (void *)((char *)rec + rec->lrh_len -
3099 sizeof(struct llog_rec_tail));
3102 struct llog_logid_rec {
3103 struct llog_rec_hdr lid_hdr;
3104 struct llog_logid lid_id;
3108 struct llog_rec_tail lid_tail;
3109 } __attribute__((packed));
3111 struct llog_unlink_rec {
3112 struct llog_rec_hdr lur_hdr;
3116 struct llog_rec_tail lur_tail;
3117 } __attribute__((packed));
3119 struct llog_unlink64_rec {
3120 struct llog_rec_hdr lur_hdr;
3121 struct lu_fid lur_fid;
3122 __u32 lur_count; /* to destroy the lost precreated */
3126 struct llog_rec_tail lur_tail;
3127 } __attribute__((packed));
3129 struct llog_setattr64_rec {
3130 struct llog_rec_hdr lsr_hdr;
3131 struct ost_id lsr_oi;
3137 struct llog_rec_tail lsr_tail;
3138 } __attribute__((packed));
3140 struct llog_size_change_rec {
3141 struct llog_rec_hdr lsc_hdr;
3142 struct ll_fid lsc_fid;
3147 struct llog_rec_tail lsc_tail;
3148 } __attribute__((packed));
3150 #define CHANGELOG_MAGIC 0xca103000
3152 /** \a changelog_rec_type's that can't be masked */
3153 #define CHANGELOG_MINMASK (1 << CL_MARK)
3154 /** bits covering all \a changelog_rec_type's */
3155 #define CHANGELOG_ALLMASK 0XFFFFFFFF
3156 /** default \a changelog_rec_type mask. Allow all of them, except
3157 * CL_ATIME since it can really be time consuming, and not necessary
3158 * under normal use. */
3159 #define CHANGELOG_DEFMASK (CHANGELOG_ALLMASK & ~(1 << CL_ATIME))
3161 /* changelog llog name, needed by client replicators */
3162 #define CHANGELOG_CATALOG "changelog_catalog"
3164 struct changelog_setinfo {
3167 } __attribute__((packed));
3169 /** changelog record */
3170 struct llog_changelog_rec {
3171 struct llog_rec_hdr cr_hdr;
3172 struct changelog_rec cr; /**< Variable length field */
3173 struct llog_rec_tail cr_do_not_use; /**< for_sizeof_only */
3174 } __attribute__((packed));
3176 #define CHANGELOG_USER_PREFIX "cl"
3178 struct llog_changelog_user_rec {
3179 struct llog_rec_hdr cur_hdr;
3183 struct llog_rec_tail cur_tail;
3184 } __attribute__((packed));
3186 enum agent_req_status {
3194 static inline const char *agent_req_status2name(enum agent_req_status ars)
3212 static inline bool agent_req_in_final_state(enum agent_req_status ars)
3214 return ((ars == ARS_SUCCEED) || (ars == ARS_FAILED) ||
3215 (ars == ARS_CANCELED));
3218 struct llog_agent_req_rec {
3219 struct llog_rec_hdr arr_hdr; /**< record header */
3220 __u32 arr_status; /**< status of the request */
3222 * agent_req_status */
3223 __u32 arr_archive_id; /**< backend archive number */
3224 __u64 arr_flags; /**< req flags */
3225 __u64 arr_compound_id; /**< compound cookie */
3226 __u64 arr_req_create; /**< req. creation time */
3227 __u64 arr_req_change; /**< req. status change time */
3228 struct hsm_action_item arr_hai; /**< req. to the agent */
3229 struct llog_rec_tail arr_tail; /**< record tail for_sizezof_only */
3230 } __attribute__((packed));
3232 /* Old llog gen for compatibility */
3236 } __attribute__((packed));
3238 struct llog_gen_rec {
3239 struct llog_rec_hdr lgr_hdr;
3240 struct llog_gen lgr_gen;
3244 struct llog_rec_tail lgr_tail;
3247 /* flags for the logs */
3249 LLOG_F_ZAP_WHEN_EMPTY = 0x1,
3250 LLOG_F_IS_CAT = 0x2,
3251 LLOG_F_IS_PLAIN = 0x4,
3252 LLOG_F_EXT_JOBID = 0x8,
3253 LLOG_F_IS_FIXSIZE = 0x10,
3255 /* Note: Flags covered by LLOG_F_EXT_MASK will be inherited from
3256 * catlog to plain log, so do not add LLOG_F_IS_FIXSIZE here,
3257 * because the catlog record is usually fixed size, but its plain
3258 * log record can be variable */
3259 LLOG_F_EXT_MASK = LLOG_F_EXT_JOBID,
3262 /* On-disk header structure of each log object, stored in little endian order */
3263 #define LLOG_MIN_CHUNK_SIZE 8192
3264 #define LLOG_HEADER_SIZE (96) /* sizeof (llog_log_hdr) + sizeof(llh_tail)
3265 * - sizeof(llh_bitmap) */
3266 #define LLOG_BITMAP_BYTES (LLOG_MIN_CHUNK_SIZE - LLOG_HEADER_SIZE)
3267 #define LLOG_MIN_REC_SIZE (24) /* round(llog_rec_hdr + llog_rec_tail) */
3269 struct llog_log_hdr {
3270 struct llog_rec_hdr llh_hdr;
3271 __s64 llh_timestamp;
3273 __u32 llh_bitmap_offset;
3276 /* for a catalog the first/oldest and still in-use plain slot is just
3277 * next to it. It will serve as the upper limit after Catalog has
3280 struct obd_uuid llh_tgtuuid;
3281 __u32 llh_reserved[LLOG_HEADER_SIZE/sizeof(__u32)-23];
3282 /* These fields must always be at the end of the llog_log_hdr.
3283 * Note: llh_bitmap size is variable because llog chunk size could be
3284 * bigger than LLOG_MIN_CHUNK_SIZE, i.e. sizeof(llog_log_hdr) > 8192
3285 * bytes, and the real size is stored in llh_hdr.lrh_len, which means
3286 * llh_tail should only be refered by LLOG_HDR_TAIL().
3287 * But this structure is also used by client/server llog interface
3288 * (see llog_client.c), it will be kept in its original way to avoid
3289 * compatiblity issue. */
3290 __u32 llh_bitmap[LLOG_BITMAP_BYTES / sizeof(__u32)];
3291 struct llog_rec_tail llh_tail;
3292 } __attribute__((packed));
3293 #undef LLOG_HEADER_SIZE
3294 #undef LLOG_BITMAP_BYTES
3296 #define LLOG_HDR_BITMAP_SIZE(llh) (__u32)((llh->llh_hdr.lrh_len - \
3297 llh->llh_bitmap_offset - \
3298 sizeof(llh->llh_tail)) * 8)
3299 #define LLOG_HDR_BITMAP(llh) (__u32 *)((char *)(llh) + \
3300 (llh)->llh_bitmap_offset)
3301 #define LLOG_HDR_TAIL(llh) ((struct llog_rec_tail *)((char *)llh + \
3302 llh->llh_hdr.lrh_len - \
3303 sizeof(llh->llh_tail)))
3305 /** log cookies are used to reference a specific log file and a record therein */
3306 struct llog_cookie {
3307 struct llog_logid lgc_lgl;
3311 } __attribute__((packed));
3313 /** llog protocol */
3314 enum llogd_rpc_ops {
3315 LLOG_ORIGIN_HANDLE_CREATE = 501,
3316 LLOG_ORIGIN_HANDLE_NEXT_BLOCK = 502,
3317 LLOG_ORIGIN_HANDLE_READ_HEADER = 503,
3318 LLOG_ORIGIN_HANDLE_WRITE_REC = 504,
3319 LLOG_ORIGIN_HANDLE_CLOSE = 505,
3320 LLOG_ORIGIN_CONNECT = 506,
3321 LLOG_CATINFO = 507, /* deprecated */
3322 LLOG_ORIGIN_HANDLE_PREV_BLOCK = 508,
3323 LLOG_ORIGIN_HANDLE_DESTROY = 509, /* for destroy llog object*/
3325 LLOG_FIRST_OPC = LLOG_ORIGIN_HANDLE_CREATE
3329 struct llog_logid lgd_logid;
3331 __u32 lgd_llh_flags;
3333 __u32 lgd_saved_index;
3335 __u64 lgd_cur_offset;
3336 } __attribute__((packed));
3338 struct llogd_conn_body {
3339 struct llog_gen lgdc_gen;
3340 struct llog_logid lgdc_logid;
3341 __u32 lgdc_ctxt_idx;
3342 } __attribute__((packed));
3344 /* Note: 64-bit types are 64-bit aligned in structure */
3346 __u64 o_valid; /* hot fields in this obdo */
3349 __u64 o_size; /* o_size-o_blocks == ost_lvb */
3353 __u64 o_blocks; /* brw: cli sent cached bytes */
3356 /* 32-bit fields start here: keep an even number of them via padding */
3357 __u32 o_blksize; /* optimal IO blocksize */
3358 __u32 o_mode; /* brw: cli sent cache remain */
3362 __u32 o_nlink; /* brw: checksum */
3364 __u32 o_misc; /* brw: o_dropped */
3366 __u64 o_ioepoch; /* epoch in ost writes */
3367 __u32 o_stripe_idx; /* holds stripe idx */
3369 struct lustre_handle o_handle; /* brw: lock handle to prolong
3371 struct llog_cookie o_lcookie; /* destroy: unlink cookie from
3372 * MDS, obsolete in 2.8, reused
3377 __u64 o_data_version; /* getattr: sum of iversion for
3379 * brw: grant space consumed on
3380 * the client for the write */
3386 #define o_dirty o_blocks
3387 #define o_undirty o_mode
3388 #define o_dropped o_misc
3389 #define o_cksum o_nlink
3390 #define o_grant_used o_data_version
3392 struct lfsck_request {
3405 __u16 lr_async_windows;
3407 struct lu_fid lr_fid;
3408 struct lu_fid lr_fid2;
3409 struct lu_fid lr_fid3;
3414 struct lfsck_reply {
3421 LE_LASTID_REBUILDING = 1,
3422 LE_LASTID_REBUILT = 2,
3428 LE_FID_ACCESSED = 8,
3430 LE_CONDITIONAL_DESTROY = 10,
3431 LE_PAIRS_VERIFY = 11,
3432 LE_SKIP_NLINK_DECLARE = 13,
3434 LE_SET_LMV_MASTER = 15,
3435 LE_SET_LMV_SLAVE = 16,
3438 enum lfsck_event_flags {
3439 LEF_TO_OST = 0x00000001,
3440 LEF_FROM_OST = 0x00000002,
3441 LEF_SET_LMV_HASH = 0x00000004,
3442 LEF_SET_LMV_ALL = 0x00000008,
3443 LEF_RECHECK_NAME_HASH = 0x00000010,
3446 static inline void lustre_set_wire_obdo(const struct obd_connect_data *ocd,
3448 const struct obdo *lobdo)
3451 wobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
3455 if (unlikely(!(ocd->ocd_connect_flags & OBD_CONNECT_FID)) &&
3456 fid_seq_is_echo(ostid_seq(&lobdo->o_oi))) {
3457 /* Currently OBD_FL_OSTID will only be used when 2.4 echo
3458 * client communicate with pre-2.4 server */
3459 wobdo->o_oi.oi.oi_id = fid_oid(&lobdo->o_oi.oi_fid);
3460 wobdo->o_oi.oi.oi_seq = fid_seq(&lobdo->o_oi.oi_fid);
3464 static inline void lustre_get_wire_obdo(const struct obd_connect_data *ocd,
3466 const struct obdo *wobdo)
3468 __u32 local_flags = 0;
3470 if (lobdo->o_valid & OBD_MD_FLFLAGS)
3471 local_flags = lobdo->o_flags & OBD_FL_LOCAL_MASK;
3474 if (local_flags != 0) {
3475 lobdo->o_valid |= OBD_MD_FLFLAGS;
3476 lobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
3477 lobdo->o_flags |= local_flags;
3482 if (unlikely(!(ocd->ocd_connect_flags & OBD_CONNECT_FID)) &&
3483 fid_seq_is_echo(wobdo->o_oi.oi.oi_seq)) {
3485 lobdo->o_oi.oi_fid.f_seq = wobdo->o_oi.oi.oi_seq;
3486 lobdo->o_oi.oi_fid.f_oid = wobdo->o_oi.oi.oi_id;
3487 lobdo->o_oi.oi_fid.f_ver = 0;
3491 /* request structure for OST's */
3496 /* Key for FIEMAP to be used in get_info calls */
3497 struct ll_fiemap_info_key {
3499 struct obdo lfik_oa;
3500 struct fiemap lfik_fiemap;
3503 void lustre_print_user_md(unsigned int level, struct lov_user_md *lum,
3506 /* Functions for dumping PTLRPC fields */
3507 void dump_rniobuf(struct niobuf_remote *rnb);
3508 void dump_ioo(struct obd_ioobj *nb);
3509 void dump_obdo(struct obdo *oa);
3510 void dump_ost_body(struct ost_body *ob);
3511 void dump_rcs(__u32 *rc);
3513 #define IDX_INFO_MAGIC 0x3D37CC37
3515 /* Index file transfer through the network. The server serializes the index into
3516 * a byte stream which is sent to the client via a bulk transfer */
3520 /* reply: see idx_info_flags below */
3523 /* request & reply: number of lu_idxpage (to be) transferred */
3527 /* request: requested attributes passed down to the iterator API */
3530 /* request & reply: index file identifier (FID) */
3531 struct lu_fid ii_fid;
3533 /* reply: version of the index file before starting to walk the index.
3534 * Please note that the version can be modified at any time during the
3538 /* request: hash to start with:
3539 * reply: hash of the first entry of the first lu_idxpage and hash
3540 * of the entry to read next if any */
3541 __u64 ii_hash_start;
3544 /* reply: size of keys in lu_idxpages, minimal one if II_FL_VARKEY is
3548 /* reply: size of records in lu_idxpages, minimal one if II_FL_VARREC
3557 #define II_END_OFF MDS_DIR_END_OFF /* all entries have been read */
3559 /* List of flags used in idx_info::ii_flags */
3560 enum idx_info_flags {
3561 II_FL_NOHASH = 1 << 0, /* client doesn't care about hash value */
3562 II_FL_VARKEY = 1 << 1, /* keys can be of variable size */
3563 II_FL_VARREC = 1 << 2, /* records can be of variable size */
3564 II_FL_NONUNQ = 1 << 3, /* index supports non-unique keys */
3565 II_FL_NOKEY = 1 << 4, /* client doesn't care about key */
3568 #define LIP_MAGIC 0x8A6D6B6C
3570 /* 4KB (= LU_PAGE_SIZE) container gathering key/record pairs */
3572 /* 16-byte header */
3575 __u16 lip_nr; /* number of entries in the container */
3576 __u64 lip_pad0; /* additional padding for future use */
3578 /* key/record pairs are stored in the remaining 4080 bytes.
3579 * depending upon the flags in idx_info::ii_flags, each key/record
3580 * pair might be preceded by:
3582 * - the key size (II_FL_VARKEY is set)
3583 * - the record size (II_FL_VARREC is set)
3585 * For the time being, we only support fixed-size key & record. */
3586 char lip_entries[0];
3589 #define LIP_HDR_SIZE (offsetof(struct lu_idxpage, lip_entries))
3591 /* Gather all possible type associated with a 4KB container */
3593 struct lu_dirpage lp_dir; /* for MDS_READPAGE */
3594 struct lu_idxpage lp_idx; /* for OBD_IDX_READ */
3595 char lp_array[LU_PAGE_SIZE];
3598 /* security opcodes */
3601 SEC_CTX_INIT_CONT = 802,
3604 SEC_FIRST_OPC = SEC_CTX_INIT
3608 * capa related definitions
3610 #define CAPA_HMAC_MAX_LEN 64
3611 #define CAPA_HMAC_KEY_MAX_LEN 56
3613 /* NB take care when changing the sequence of elements this struct,
3614 * because the offset info is used in find_capa() */
3615 struct lustre_capa {
3616 struct lu_fid lc_fid; /** fid */
3617 __u64 lc_opc; /** operations allowed */
3618 __u64 lc_uid; /** file owner */
3619 __u64 lc_gid; /** file group */
3620 __u32 lc_flags; /** HMAC algorithm & flags */
3621 __u32 lc_keyid; /** key# used for the capability */
3622 __u32 lc_timeout; /** capa timeout value (sec) */
3623 __u32 lc_expiry; /** expiry time (sec) */
3624 __u8 lc_hmac[CAPA_HMAC_MAX_LEN]; /** HMAC */
3625 } __attribute__((packed));
3627 /** lustre_capa::lc_opc */
3629 CAPA_OPC_BODY_WRITE = 1<<0, /**< write object data */
3630 CAPA_OPC_BODY_READ = 1<<1, /**< read object data */
3631 CAPA_OPC_INDEX_LOOKUP = 1<<2, /**< lookup object fid */
3632 CAPA_OPC_INDEX_INSERT = 1<<3, /**< insert object fid */
3633 CAPA_OPC_INDEX_DELETE = 1<<4, /**< delete object fid */
3634 CAPA_OPC_OSS_WRITE = 1<<5, /**< write oss object data */
3635 CAPA_OPC_OSS_READ = 1<<6, /**< read oss object data */
3636 CAPA_OPC_OSS_TRUNC = 1<<7, /**< truncate oss object */
3637 CAPA_OPC_OSS_DESTROY = 1<<8, /**< destroy oss object */
3638 CAPA_OPC_META_WRITE = 1<<9, /**< write object meta data */
3639 CAPA_OPC_META_READ = 1<<10, /**< read object meta data */
3642 #define CAPA_OPC_OSS_RW (CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE)
3643 #define CAPA_OPC_MDS_ONLY \
3644 (CAPA_OPC_BODY_WRITE | CAPA_OPC_BODY_READ | CAPA_OPC_INDEX_LOOKUP | \
3645 CAPA_OPC_INDEX_INSERT | CAPA_OPC_INDEX_DELETE)
3646 #define CAPA_OPC_OSS_ONLY \
3647 (CAPA_OPC_OSS_WRITE | CAPA_OPC_OSS_READ | CAPA_OPC_OSS_TRUNC | \
3648 CAPA_OPC_OSS_DESTROY)
3649 #define CAPA_OPC_MDS_DEFAULT ~CAPA_OPC_OSS_ONLY
3650 #define CAPA_OPC_OSS_DEFAULT ~(CAPA_OPC_MDS_ONLY | CAPA_OPC_OSS_ONLY)
3652 static inline bool lovea_slot_is_dummy(const struct lov_ost_data_v1 *obj)
3654 /* zero area does not care about the bytes-order. */
3655 if (obj->l_ost_oi.oi.oi_id == 0 && obj->l_ost_oi.oi.oi_seq == 0 &&
3656 obj->l_ost_idx == 0 && obj->l_ost_gen == 0)
3662 /* lustre_capa::lc_hmac_alg */
3664 CAPA_HMAC_ALG_SHA1 = 1, /**< sha1 algorithm */
3668 #define CAPA_FL_MASK 0x00ffffff
3669 #define CAPA_HMAC_ALG_MASK 0xff000000
3671 struct lustre_capa_key {
3672 __u64 lk_seq; /**< mds# */
3673 __u32 lk_keyid; /**< key# */
3675 __u8 lk_key[CAPA_HMAC_KEY_MAX_LEN]; /**< key */
3676 } __attribute__((packed));
3678 /** The link ea holds 1 \a link_ea_entry for each hardlink */
3679 #define LINK_EA_MAGIC 0x11EAF1DFUL
3680 struct link_ea_header {
3683 __u64 leh_len; /* total size */
3689 /** Hardlink data is name and parent fid.
3690 * Stored in this crazy struct for maximum packing and endian-neutrality
3692 struct link_ea_entry {
3693 /** __u16 stored big-endian, unaligned */
3694 unsigned char lee_reclen[2];
3695 unsigned char lee_parent_fid[sizeof(struct lu_fid)];
3697 }__attribute__((packed));
3699 /** fid2path request/reply structure */
3700 struct getinfo_fid2path {
3701 struct lu_fid gf_fid;
3706 } __attribute__((packed));
3708 /** path2parent request/reply structures */
3710 struct lu_fid gp_fid; /**< parent FID */
3711 __u32 gp_linkno; /**< hardlink number */
3712 __u32 gp_name_size; /**< size of the name field */
3713 char gp_name[0]; /**< zero-terminated link name */
3714 } __attribute__((packed));
3717 LAYOUT_INTENT_ACCESS = 0,
3718 LAYOUT_INTENT_READ = 1,
3719 LAYOUT_INTENT_WRITE = 2,
3720 LAYOUT_INTENT_GLIMPSE = 3,
3721 LAYOUT_INTENT_TRUNC = 4,
3722 LAYOUT_INTENT_RELEASE = 5,
3723 LAYOUT_INTENT_RESTORE = 6
3726 /* enqueue layout lock with intent */
3727 struct layout_intent {
3728 __u32 li_opc; /* intent operation for enqueue, read, write etc */
3735 * On the wire version of hsm_progress structure.
3737 * Contains the userspace hsm_progress and some internal fields.
3739 struct hsm_progress_kernel {
3740 /* Field taken from struct hsm_progress */
3743 struct hsm_extent hpk_extent;
3745 __u16 hpk_errval; /* positive val */
3747 /* Additional fields */
3748 __u64 hpk_data_version;
3750 } __attribute__((packed));
3753 * OUT_UPDATE RPC Format
3755 * During the cross-ref operation, the Master MDT, which the client send the
3756 * request to, will disassembly the operation into object updates, then OSP
3757 * will send these updates to the remote MDT to be executed.
3759 * An UPDATE_OBJ RPC does a list of updates. Each update belongs to an
3760 * operation and does a type of modification to an object.
3768 * update (ub_count-th)
3770 * ub_count must be less than or equal to UPDATE_PER_RPC_MAX.
3775 * rc [+ buffers] (1st)
3776 * rc [+ buffers] (2st)
3778 * rc [+ buffers] (nr_count-th)
3780 * ur_count must be less than or equal to UPDATE_PER_RPC_MAX and should usually
3781 * be equal to ub_count.
3785 * Type of each update, if adding/deleting update, please also update
3786 * update_opcode in lustre/target/out_lib.c.
3798 OUT_INDEX_LOOKUP = 9,
3799 OUT_INDEX_INSERT = 10,
3800 OUT_INDEX_DELETE = 11,
3810 UPDATE_FL_OST = 0x00000001, /* op from OST (not MDT) */
3811 UPDATE_FL_SYNC = 0x00000002, /* commit before replying */
3812 UPDATE_FL_COMMITTED = 0x00000004, /* op committed globally */
3813 UPDATE_FL_NOLOG = 0x00000008 /* for idempotent updates */
3816 struct object_update_param {
3817 __u16 oup_len; /* length of this parameter */
3823 static inline size_t
3824 object_update_param_size(const struct object_update_param *param)
3826 return cfs_size_round(sizeof(*param) + param->oup_len);
3830 struct object_update {
3831 __u16 ou_type; /* enum update_type */
3832 __u16 ou_params_count; /* update parameters count */
3833 __u32 ou_result_size; /* how many bytes can return */
3834 __u32 ou_flags; /* enum update_flag */
3835 __u32 ou_padding1; /* padding 1 */
3836 __u64 ou_batchid; /* op transno on master */
3837 struct lu_fid ou_fid; /* object to be updated */
3838 struct object_update_param ou_params[0]; /* update params */
3841 #define UPDATE_REQUEST_MAGIC_V1 0xBDDE0001
3842 #define UPDATE_REQUEST_MAGIC_V2 0xBDDE0002
3843 #define UPDATE_REQUEST_MAGIC UPDATE_REQUEST_MAGIC_V2
3844 /* Hold object_updates sending to the remote OUT in single RPC */
3845 struct object_update_request {
3847 __u16 ourq_count; /* number of ourq_updates[] */
3849 struct object_update ourq_updates[0];
3852 #define OUT_UPDATE_HEADER_MAGIC 0xBDDF0001
3853 #define OUT_UPDATE_MAX_INLINE_SIZE 4096
3854 /* Header for updates request between MDTs */
3855 struct out_update_header {
3858 __u32 ouh_inline_length;
3859 __u32 ouh_reply_size;
3860 __u32 ouh_inline_data[0];
3863 struct out_update_buffer {
3868 static inline size_t
3869 object_update_params_size(const struct object_update *update)
3871 const struct object_update_param *param;
3872 size_t total_size = 0;
3875 param = &update->ou_params[0];
3876 for (i = 0; i < update->ou_params_count; i++) {
3877 size_t size = object_update_param_size(param);
3879 param = (struct object_update_param *)((char *)param + size);
3886 static inline size_t
3887 object_update_size(const struct object_update *update)
3889 return offsetof(struct object_update, ou_params[0]) +
3890 object_update_params_size(update);
3893 static inline struct object_update *
3894 object_update_request_get(const struct object_update_request *our,
3895 unsigned int index, size_t *size)
3900 if (index >= our->ourq_count)
3903 ptr = (void *)&our->ourq_updates[0];
3904 for (i = 0; i < index; i++)
3905 ptr += object_update_size(ptr);
3908 *size = object_update_size(ptr);
3914 /* the result of object update */
3915 struct object_update_result {
3922 #define UPDATE_REPLY_MAGIC_V1 0x00BD0001
3923 #define UPDATE_REPLY_MAGIC_V2 0x00BD0002
3924 #define UPDATE_REPLY_MAGIC UPDATE_REPLY_MAGIC_V2
3925 /* Hold object_update_results being replied from the remote OUT. */
3926 struct object_update_reply {
3933 static inline struct object_update_result *
3934 object_update_result_get(const struct object_update_reply *reply,
3935 unsigned int index, size_t *size)
3937 __u16 count = reply->ourp_count;
3944 ptr = (char *)reply +
3945 cfs_size_round(offsetof(struct object_update_reply,
3947 for (i = 0; i < index; i++) {
3948 if (reply->ourp_lens[i] == 0)
3951 ptr += cfs_size_round(reply->ourp_lens[i]);
3955 *size = reply->ourp_lens[index];
3960 /* read update result */
3961 struct out_read_reply {
3968 static inline void orr_cpu_to_le(struct out_read_reply *orr_dst,
3969 const struct out_read_reply *orr_src)
3971 orr_dst->orr_size = cpu_to_le32(orr_src->orr_size);
3972 orr_dst->orr_padding = cpu_to_le32(orr_src->orr_padding);
3973 orr_dst->orr_offset = cpu_to_le64(orr_dst->orr_offset);
3976 static inline void orr_le_to_cpu(struct out_read_reply *orr_dst,
3977 const struct out_read_reply *orr_src)
3979 orr_dst->orr_size = le32_to_cpu(orr_src->orr_size);
3980 orr_dst->orr_padding = le32_to_cpu(orr_src->orr_padding);
3981 orr_dst->orr_offset = le64_to_cpu(orr_dst->orr_offset);
3984 /** layout swap request structure
3985 * fid1 and fid2 are in mdt_body
3987 struct mdc_swap_layouts {
3992 struct lustre_handle cd_handle;
3993 struct lu_fid cd_fid;
3994 __u64 cd_data_version;
3995 __u64 cd_reserved[8];
3998 /* Update llog format */
4000 struct lu_fid uop_fid;
4002 __u16 uop_param_count;
4003 __u16 uop_params_off[0];
4007 struct update_op uops_op[0];
4010 struct update_params {
4011 struct object_update_param up_params[0];
4014 enum update_records_flag {
4015 UPDATE_RECORD_CONTINUE = 1 >> 0,
4018 * This is the update record format used to store the updates in
4019 * disk. All updates of the operation will be stored in ur_ops.
4020 * All of parameters for updates of the operation will be stored
4022 * To save the space of the record, parameters in ur_ops will only
4023 * remember their offset in ur_params, so to avoid storing duplicate
4024 * parameters in ur_params, which can help us save a lot space for
4025 * operation like creating striped directory.
4027 struct update_records {
4028 __u64 ur_master_transno;
4031 /* If the operation includes multiple updates, then ur_index
4032 * means the index of the update inside the whole updates. */
4034 __u32 ur_update_count;
4035 __u32 ur_param_count;
4036 struct update_ops ur_ops;
4037 /* Note ur_ops has a variable size, so comment out
4038 * the following ur_params, in case some use it directly
4039 * update_records->ur_params
4041 * struct update_params ur_params;
4045 struct llog_update_record {
4046 struct llog_rec_hdr lur_hdr;
4047 struct update_records lur_update_rec;
4048 /* Note ur_update_rec has a variable size, so comment out
4049 * the following ur_tail, in case someone use it directly
4051 * struct llog_rec_tail lur_tail;