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, 2012, 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, and the
56 * prototypes of the swabber functions for each struct. Nothing that
57 * depends on external functions or definitions should be in here.
59 * Structs must be properly aligned to put 64-bit values on an 8-byte
60 * boundary. Any structs being added here must also be added to
61 * utils/wirecheck.c and "make newwiretest" run to regenerate the
62 * utils/wiretest.c sources. This allows us to verify that wire structs
63 * have the proper alignment/size on all architectures.
65 * DO NOT CHANGE any of the structs, flags, values declared here and used
66 * in released Lustre versions. Some structs may have padding fields that
67 * can be used. Some structs might allow addition at the end (verify this
68 * in the code to ensure that new/old clients that see this larger struct
69 * do not fail, otherwise you need to implement protocol compatibility).
71 * We assume all nodes are either little-endian or big-endian, and we
72 * always send messages in the sender's native format. The receiver
73 * detects the message format by checking the 'magic' field of the message
74 * (see lustre_msg_swabbed() below).
76 * Each wire type has corresponding 'lustre_swab_xxxtypexxx()' routines,
77 * implemented either here, inline (trivial implementations) or in
78 * ptlrpc/pack_generic.c. These 'swabbers' convert the type from "other"
79 * endian, in-place in the message buffer.
81 * A swabber takes a single pointer argument. The caller must already have
82 * verified that the length of the message buffer >= sizeof (type).
84 * For variable length types, a second 'lustre_swab_v_xxxtypexxx()' routine
85 * may be defined that swabs just the variable part, after the caller has
86 * verified that the message buffer is large enough.
91 #ifndef _LUSTRE_IDL_H_
92 #define _LUSTRE_IDL_H_
94 #if !defined(LASSERT) && !defined(LPU64)
95 #include <libcfs/libcfs.h> /* for LASSERT, LPUX64, etc */
98 /* Defn's shared with user-space. */
99 #include <lustre/lustre_user.h>
104 /* FOO_REQUEST_PORTAL is for incoming requests on the FOO
105 * FOO_REPLY_PORTAL is for incoming replies on the FOO
106 * FOO_BULK_PORTAL is for incoming bulk on the FOO
109 #define CONNMGR_REQUEST_PORTAL 1
110 #define CONNMGR_REPLY_PORTAL 2
111 //#define OSC_REQUEST_PORTAL 3
112 #define OSC_REPLY_PORTAL 4
113 //#define OSC_BULK_PORTAL 5
114 #define OST_IO_PORTAL 6
115 #define OST_CREATE_PORTAL 7
116 #define OST_BULK_PORTAL 8
117 //#define MDC_REQUEST_PORTAL 9
118 #define MDC_REPLY_PORTAL 10
119 //#define MDC_BULK_PORTAL 11
120 #define MDS_REQUEST_PORTAL 12
121 //#define MDS_REPLY_PORTAL 13
122 #define MDS_BULK_PORTAL 14
123 #define LDLM_CB_REQUEST_PORTAL 15
124 #define LDLM_CB_REPLY_PORTAL 16
125 #define LDLM_CANCEL_REQUEST_PORTAL 17
126 #define LDLM_CANCEL_REPLY_PORTAL 18
127 //#define PTLBD_REQUEST_PORTAL 19
128 //#define PTLBD_REPLY_PORTAL 20
129 //#define PTLBD_BULK_PORTAL 21
130 #define MDS_SETATTR_PORTAL 22
131 #define MDS_READPAGE_PORTAL 23
132 #define MDS_MDS_PORTAL 24
134 #define MGC_REPLY_PORTAL 25
135 #define MGS_REQUEST_PORTAL 26
136 #define MGS_REPLY_PORTAL 27
137 #define OST_REQUEST_PORTAL 28
138 #define FLD_REQUEST_PORTAL 29
139 #define SEQ_METADATA_PORTAL 30
140 #define SEQ_DATA_PORTAL 31
141 #define SEQ_CONTROLLER_PORTAL 32
142 #define MGS_BULK_PORTAL 33
144 /* Portal 63 is reserved for the Cray Inc DVS - nic@cray.com, roe@cray.com, n8851@cray.com */
147 #define PTL_RPC_MSG_REQUEST 4711
148 #define PTL_RPC_MSG_ERR 4712
149 #define PTL_RPC_MSG_REPLY 4713
151 /* DON'T use swabbed values of MAGIC as magic! */
152 #define LUSTRE_MSG_MAGIC_V1 0x0BD00BD0
153 #define LUSTRE_MSG_MAGIC_V2 0x0BD00BD3
155 #define LUSTRE_MSG_MAGIC_V1_SWABBED 0xD00BD00B
156 #define LUSTRE_MSG_MAGIC_V2_SWABBED 0xD30BD00B
158 #define LUSTRE_MSG_MAGIC LUSTRE_MSG_MAGIC_V2
160 #define PTLRPC_MSG_VERSION 0x00000003
161 #define LUSTRE_VERSION_MASK 0xffff0000
162 #define LUSTRE_OBD_VERSION 0x00010000
163 #define LUSTRE_MDS_VERSION 0x00020000
164 #define LUSTRE_OST_VERSION 0x00030000
165 #define LUSTRE_DLM_VERSION 0x00040000
166 #define LUSTRE_LOG_VERSION 0x00050000
167 #define LUSTRE_MGS_VERSION 0x00060000
169 typedef __u32 mdsno_t;
170 typedef __u64 seqno_t;
171 typedef __u64 obd_id;
172 typedef __u64 obd_seq;
173 typedef __s64 obd_time;
174 typedef __u64 obd_size;
175 typedef __u64 obd_off;
176 typedef __u64 obd_blocks;
177 typedef __u64 obd_valid;
178 typedef __u32 obd_blksize;
179 typedef __u32 obd_mode;
180 typedef __u32 obd_uid;
181 typedef __u32 obd_gid;
182 typedef __u32 obd_flag;
183 typedef __u32 obd_count;
186 * Describes a range of sequence, lsr_start is included but lsr_end is
188 * Same structure is used in fld module where lsr_index field holds mdt id
192 #define LU_SEQ_RANGE_MDT 0x0
193 #define LU_SEQ_RANGE_OST 0x1
195 struct lu_seq_range {
203 * returns width of given range \a r
206 static inline __u64 range_space(const struct lu_seq_range *range)
208 return range->lsr_end - range->lsr_start;
212 * initialize range to zero
215 static inline void range_init(struct lu_seq_range *range)
217 range->lsr_start = range->lsr_end = range->lsr_index = 0;
221 * check if given seq id \a s is within given range \a r
224 static inline int range_within(const struct lu_seq_range *range,
227 return s >= range->lsr_start && s < range->lsr_end;
230 static inline int range_is_sane(const struct lu_seq_range *range)
232 return (range->lsr_end >= range->lsr_start);
235 static inline int range_is_zero(const struct lu_seq_range *range)
237 return (range->lsr_start == 0 && range->lsr_end == 0);
240 static inline int range_is_exhausted(const struct lu_seq_range *range)
243 return range_space(range) == 0;
246 /* return 0 if two range have the same location */
247 static inline int range_compare_loc(const struct lu_seq_range *r1,
248 const struct lu_seq_range *r2)
250 return r1->lsr_index != r2->lsr_index ||
251 r1->lsr_flags != r2->lsr_flags;
254 #define DRANGE "[%#16.16"LPF64"x-%#16.16"LPF64"x):%x:%s"
256 #define PRANGE(range) \
257 (range)->lsr_start, \
259 (range)->lsr_index, \
260 (range)->lsr_flags == LU_SEQ_RANGE_MDT ? "mdt" : "ost"
263 /** \defgroup lu_fid lu_fid
267 * Flags for lustre_mdt_attrs::lma_compat and lustre_mdt_attrs::lma_incompat.
268 * Deprecated since HSM and SOM attributes are now stored in separate on-disk
272 LMAC_HSM = 0x00000001,
273 LMAC_SOM = 0x00000002,
277 * Masks for all features that should be supported by a Lustre version to
278 * access a specific file.
279 * This information is stored in lustre_mdt_attrs::lma_incompat.
282 LMAI_RELEASED = 0x0000001, /* file is released */
284 #define LMA_INCOMPAT_SUPP 0x0
287 * Following struct for MDT attributes, that will be kept inode's EA.
288 * Introduced in 2.0 release (please see b15993, for details)
290 struct lustre_mdt_attrs {
292 * Bitfield for supported data in this structure. From enum lma_compat.
293 * lma_self_fid and lma_flags are always available.
297 * Per-file incompat feature list. Lustre version should support all
298 * flags set in this field. The supported feature mask is available in
302 /** FID of this inode */
303 struct lu_fid lma_self_fid;
304 /** mdt/ost type, others */
309 * Prior to 2.4, the LMA structure also included SOM attributes which has since
310 * been moved to a dedicated xattr
312 #define LMA_OLD_SIZE (sizeof(struct lustre_mdt_attrs) + 4 * sizeof(__u64))
314 extern void lustre_lma_swab(struct lustre_mdt_attrs *lma);
315 extern void lustre_lma_init(struct lustre_mdt_attrs *lma,
316 const struct lu_fid *fid);
318 * SOM on-disk attributes stored in a separate xattr.
321 /** Bitfield for supported data in this structure. For future use. */
324 /** Incompat feature list. The supported feature mask is availabe in
325 * SOM_INCOMPAT_SUPP */
328 /** IO Epoch SOM attributes belongs to */
330 /** total file size in objects */
332 /** total fs blocks in objects */
334 /** mds mount id the size is valid for */
337 extern void lustre_som_swab(struct som_attrs *attrs);
339 #define SOM_INCOMPAT_SUPP 0x0
342 * HSM on-disk attributes stored in a separate xattr.
345 /** Bitfield for supported data in this structure. For future use. */
348 /** HSM flags, see hsm_flags enum below */
350 /** backend archive id associated with the file */
352 /** version associated with the last archiving, if any */
355 extern void lustre_hsm_swab(struct hsm_attrs *attrs);
362 static inline void ostid_cpu_to_le(struct ost_id *src_oi,
363 struct ost_id *dst_oi)
365 dst_oi->oi_id = cpu_to_le64(src_oi->oi_id);
366 dst_oi->oi_seq = cpu_to_le64(src_oi->oi_seq);
369 static inline void ostid_le_to_cpu(struct ost_id *src_oi,
370 struct ost_id *dst_oi)
372 dst_oi->oi_id = le64_to_cpu(src_oi->oi_id);
373 dst_oi->oi_seq = le64_to_cpu(src_oi->oi_seq);
376 extern void lustre_swab_ost_id(struct ost_id *oid);
382 /** initial fid id value */
383 LUSTRE_FID_INIT_OID = 1UL
386 /** returns fid object sequence */
387 static inline __u64 fid_seq(const struct lu_fid *fid)
392 /** returns fid object id */
393 static inline __u32 fid_oid(const struct lu_fid *fid)
398 /** returns fid object version */
399 static inline __u32 fid_ver(const struct lu_fid *fid)
404 static inline void fid_zero(struct lu_fid *fid)
406 memset(fid, 0, sizeof(*fid));
409 static inline obd_id fid_ver_oid(const struct lu_fid *fid)
411 return ((__u64)fid_ver(fid) << 32 | fid_oid(fid));
415 * Note that reserved SEQ numbers below 12 will conflict with ldiskfs
416 * inodes in the IGIF namespace, so these reserved SEQ numbers can be
417 * used for other purposes and not risk collisions with existing inodes.
419 * Different FID Format
420 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs#NEW.0
423 FID_SEQ_OST_MDT0 = 0,
426 FID_SEQ_OST_MDT1 = 3,
427 FID_SEQ_OST_MAX = 9, /* Max MDT count before OST_on_FID */
430 FID_SEQ_IGIF_MAX = 0x0ffffffffULL,
431 FID_SEQ_IDIF = 0x100000000ULL,
432 FID_SEQ_IDIF_MAX = 0x1ffffffffULL,
433 /* Normal FID sequence starts from this value, i.e. 1<<33 */
434 FID_SEQ_START = 0x200000000ULL,
435 /* sequence for local pre-defined FIDs listed in local_oid */
436 FID_SEQ_LOCAL_FILE = 0x200000001ULL,
437 FID_SEQ_DOT_LUSTRE = 0x200000002ULL,
438 /* sequence is used for local named objects FIDs generated
439 * by local_object_storage library */
440 FID_SEQ_LOCAL_NAME = 0x200000003ULL,
441 /* Because current FLD will only cache the fid sequence, instead
442 * of oid on the client side, if the FID needs to be exposed to
443 * clients sides, it needs to make sure all of fids under one
444 * sequence will be located in one MDT. */
445 FID_SEQ_SPECIAL = 0x200000004ULL,
446 FID_SEQ_QUOTA = 0x200000005ULL,
447 FID_SEQ_QUOTA_GLB = 0x200000006ULL,
448 FID_SEQ_ROOT = 0x200000007ULL, /* Located on MDT0 */
449 FID_SEQ_NORMAL = 0x200000400ULL,
450 FID_SEQ_LOV_DEFAULT = 0xffffffffffffffffULL
453 #define OBIF_OID_MAX_BITS 32
454 #define OBIF_MAX_OID (1ULL << OBIF_OID_MAX_BITS)
455 #define OBIF_OID_MASK ((1ULL << OBIF_OID_MAX_BITS) - 1)
456 #define IDIF_OID_MAX_BITS 48
457 #define IDIF_MAX_OID (1ULL << IDIF_OID_MAX_BITS)
458 #define IDIF_OID_MASK ((1ULL << IDIF_OID_MAX_BITS) - 1)
460 /** OID for FID_SEQ_SPECIAL */
462 /* Big Filesystem Lock to serialize rename operations */
463 FID_OID_SPECIAL_BFL = 1UL,
466 /** OID for FID_SEQ_DOT_LUSTRE */
467 enum dot_lustre_oid {
468 FID_OID_DOT_LUSTRE = 1UL,
469 FID_OID_DOT_LUSTRE_OBF = 2UL,
472 static inline int fid_seq_is_mdt0(obd_seq seq)
474 return (seq == FID_SEQ_OST_MDT0);
477 static inline int fid_seq_is_mdt(const __u64 seq)
479 return seq == FID_SEQ_OST_MDT0 || seq >= FID_SEQ_NORMAL;
482 static inline int fid_seq_is_echo(obd_seq seq)
484 return (seq == FID_SEQ_ECHO);
487 static inline int fid_is_echo(const struct lu_fid *fid)
489 return fid_seq_is_echo(fid_seq(fid));
492 static inline int fid_seq_is_llog(obd_seq seq)
494 return (seq == FID_SEQ_LLOG);
497 static inline int fid_is_llog(const struct lu_fid *fid)
499 return fid_seq_is_llog(fid_seq(fid));
502 static inline int fid_seq_is_rsvd(const __u64 seq)
504 return (seq > FID_SEQ_OST_MDT0 && seq <= FID_SEQ_RSVD);
507 static inline int fid_seq_is_special(const __u64 seq)
509 return seq == FID_SEQ_SPECIAL;
512 static inline int fid_seq_is_local_file(const __u64 seq)
514 return seq == FID_SEQ_LOCAL_FILE;
517 static inline int fid_seq_is_root(const __u64 seq)
519 return seq == FID_SEQ_ROOT;
522 static inline int fid_seq_is_dot(const __u64 seq)
524 return seq == FID_SEQ_DOT_LUSTRE;
527 static inline int fid_is_mdt0(const struct lu_fid *fid)
529 return fid_seq_is_mdt0(fid_seq(fid));
532 static inline void lu_root_fid(struct lu_fid *fid)
534 fid->f_seq = FID_SEQ_ROOT;
540 * Check if a fid is igif or not.
541 * \param fid the fid to be tested.
542 * \return true if the fid is a igif; otherwise false.
544 static inline int fid_seq_is_igif(const __u64 seq)
546 return seq >= FID_SEQ_IGIF && seq <= FID_SEQ_IGIF_MAX;
549 static inline int fid_is_igif(const struct lu_fid *fid)
551 return fid_seq_is_igif(fid_seq(fid));
555 * Check if a fid is idif or not.
556 * \param fid the fid to be tested.
557 * \return true if the fid is a idif; otherwise false.
559 static inline int fid_seq_is_idif(const __u64 seq)
561 return seq >= FID_SEQ_IDIF && seq <= FID_SEQ_IDIF_MAX;
564 static inline int fid_is_idif(const struct lu_fid *fid)
566 return fid_seq_is_idif(fid_seq(fid));
569 static inline int fid_is_local_file(const struct lu_fid *fid)
571 return fid_seq_is_local_file(fid_seq(fid));
574 static inline int fid_seq_is_norm(const __u64 seq)
576 return (seq >= FID_SEQ_NORMAL);
579 static inline int fid_is_norm(const struct lu_fid *fid)
581 return fid_seq_is_norm(fid_seq(fid));
584 /* convert an OST objid into an IDIF FID SEQ number */
585 static inline obd_seq fid_idif_seq(obd_id id, __u32 ost_idx)
587 return FID_SEQ_IDIF | (ost_idx << 16) | ((id >> 32) & 0xffff);
590 /* convert a packed IDIF FID into an OST objid */
591 static inline obd_id fid_idif_id(obd_seq seq, __u32 oid, __u32 ver)
593 return ((__u64)ver << 48) | ((seq & 0xffff) << 32) | oid;
596 /* extract ost index from IDIF FID */
597 static inline __u32 fid_idif_ost_idx(const struct lu_fid *fid)
599 LASSERT(fid_is_idif(fid));
600 return (fid_seq(fid) >> 16) & 0xffff;
603 /* unpack an ostid (id/seq) from a wire/disk structure into an IDIF FID */
604 static inline void ostid_idif_unpack(struct ost_id *ostid,
605 struct lu_fid *fid, __u32 ost_idx)
607 fid->f_seq = fid_idif_seq(ostid->oi_id, ost_idx);
608 fid->f_oid = ostid->oi_id; /* truncate to 32 bits by assignment */
609 fid->f_ver = ostid->oi_id >> 48; /* in theory, not currently used */
612 /* unpack an ostid (id/seq) from a wire/disk structure into a non-IDIF FID */
613 static inline void ostid_fid_unpack(struct ost_id *ostid, struct lu_fid *fid)
615 fid->f_seq = ostid->oi_seq;
616 fid->f_oid = ostid->oi_id; /* truncate to 32 bits by assignment */
617 fid->f_ver = ostid->oi_id >> 32; /* in theory, not currently used */
620 /* Unpack an OST object id/seq (group) into a FID. This is needed for
621 * converting all obdo, lmm, lsm, etc. 64-bit id/seq pairs into proper
622 * FIDs. Note that if an id/seq is already in FID/IDIF format it will
623 * be passed through unchanged. Only legacy OST objects in "group 0"
624 * will be mapped into the IDIF namespace so that they can fit into the
625 * struct lu_fid fields without loss. For reference see:
626 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs
628 static inline int fid_ostid_unpack(struct lu_fid *fid, struct ost_id *ostid,
631 if (ost_idx > 0xffff) {
632 CERROR("bad ost_idx, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
633 ostid->oi_seq, ostid->oi_id, ost_idx);
637 if (fid_seq_is_mdt0(ostid->oi_seq)) {
638 /* This is a "legacy" (old 1.x/2.early) OST object in "group 0"
639 * that we map into the IDIF namespace. It allows up to 2^48
640 * objects per OST, as this is the object namespace that has
641 * been in production for years. This can handle create rates
642 * of 1M objects/s/OST for 9 years, or combinations thereof. */
643 if (ostid->oi_id >= IDIF_MAX_OID) {
644 CERROR("bad MDT0 id, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
645 ostid->oi_seq, ostid->oi_id, ost_idx);
648 ostid_idif_unpack(ostid, fid, ost_idx);
650 } else if (fid_seq_is_rsvd(ostid->oi_seq)) {
651 /* These are legacy OST objects for LLOG/ECHO and CMD testing.
652 * We only support 2^32 objects in these groups, and cannot
653 * uniquely identify them in the system (i.e. they are the
654 * duplicated on all OSTs), but this is not strictly required
655 * for the old object protocol, which has a separate ost_idx. */
656 if (ostid->oi_id >= 0xffffffffULL) {
657 CERROR("bad RSVD id, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
658 ostid->oi_seq, ostid->oi_id, ost_idx);
661 ostid_fid_unpack(ostid, fid);
663 } else if (unlikely(fid_seq_is_igif(ostid->oi_seq))) {
664 /* This is an MDT inode number, which should never collide with
665 * proper OST object IDs, and is probably a broken filesystem */
666 CERROR("bad IGIF, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
667 ostid->oi_seq, ostid->oi_id, ost_idx);
670 } else /* if (fid_seq_is_idif(seq) || fid_seq_is_norm(seq)) */ {
671 /* This is either an IDIF object, which identifies objects across
672 * all OSTs, or a regular FID. The IDIF namespace maps legacy
673 * OST objects into the FID namespace. In both cases, we just
674 * pass the FID through, no conversion needed. */
675 ostid_fid_unpack(ostid, fid);
681 /* pack an IDIF FID into an ostid (id/seq) for the wire/disk */
682 static inline void ostid_idif_pack(const struct lu_fid *fid,
683 struct ost_id *ostid)
685 ostid->oi_seq = FID_SEQ_OST_MDT0;
686 ostid->oi_id = fid_idif_id(fid->f_seq, fid->f_oid, fid->f_ver);
689 /* pack a non-IDIF FID into an ostid (id/seq) for the wire/disk */
690 static inline void ostid_normal_fid_pack(const struct lu_fid *fid,
691 struct ost_id *ostid)
693 ostid->oi_seq = fid_seq(fid);
694 ostid->oi_id = fid_ver_oid(fid);
697 /* pack any OST FID into an ostid (id/seq) for the wire/disk */
698 static inline int fid_ostid_pack(const struct lu_fid *fid,
699 struct ost_id *ostid)
701 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
702 CERROR("bad IGIF, "DFID"\n", PFID(fid));
706 if (fid_is_idif(fid))
707 ostid_idif_pack(fid, ostid);
709 ostid_normal_fid_pack(fid, ostid);
714 /* extract OST sequence (group) from a wire ost_id (id/seq) pair */
715 static inline obd_seq ostid_seq(struct ost_id *ostid)
717 if (unlikely(fid_seq_is_igif(ostid->oi_seq)))
718 CWARN("bad IGIF, oi_seq: "LPU64" oi_id: "LPX64"\n",
719 ostid->oi_seq, ostid->oi_id);
721 if (unlikely(fid_seq_is_idif(ostid->oi_seq)))
722 return FID_SEQ_OST_MDT0;
724 return ostid->oi_seq;
727 /* extract OST objid from a wire ost_id (id/seq) pair */
728 static inline obd_id ostid_id(struct ost_id *ostid)
730 if (ostid->oi_seq == FID_SEQ_OST_MDT0)
731 return ostid->oi_id & IDIF_OID_MASK;
733 if (fid_seq_is_rsvd(ostid->oi_seq))
734 return ostid->oi_id & OBIF_OID_MASK;
736 if (fid_seq_is_idif(ostid->oi_seq))
737 return fid_idif_id(ostid->oi_seq, ostid->oi_id, 0);
742 /* Check whether the fid is for LAST_ID */
743 static inline int fid_is_last_id(const struct lu_fid *fid)
745 return (fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid)) &&
750 * Get inode number from a igif.
751 * \param fid a igif to get inode number from.
752 * \return inode number for the igif.
754 static inline ino_t lu_igif_ino(const struct lu_fid *fid)
760 * Get inode generation from a igif.
761 * \param fid a igif to get inode generation from.
762 * \return inode generation for the igif.
764 static inline __u32 lu_igif_gen(const struct lu_fid *fid)
770 * Build igif from the inode number/generation.
772 static inline void lu_igif_build(struct lu_fid *fid, __u32 ino, __u32 gen)
780 * Fids are transmitted across network (in the sender byte-ordering),
781 * and stored on disk in big-endian order.
783 static inline void fid_cpu_to_le(struct lu_fid *dst, const struct lu_fid *src)
785 /* check that all fields are converted */
786 CLASSERT(sizeof *src ==
787 sizeof fid_seq(src) +
788 sizeof fid_oid(src) + sizeof fid_ver(src));
789 dst->f_seq = cpu_to_le64(fid_seq(src));
790 dst->f_oid = cpu_to_le32(fid_oid(src));
791 dst->f_ver = cpu_to_le32(fid_ver(src));
794 static inline void fid_le_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
796 /* check that all fields are converted */
797 CLASSERT(sizeof *src ==
798 sizeof fid_seq(src) +
799 sizeof fid_oid(src) + sizeof fid_ver(src));
800 dst->f_seq = le64_to_cpu(fid_seq(src));
801 dst->f_oid = le32_to_cpu(fid_oid(src));
802 dst->f_ver = le32_to_cpu(fid_ver(src));
805 static inline void fid_cpu_to_be(struct lu_fid *dst, const struct lu_fid *src)
807 /* check that all fields are converted */
808 CLASSERT(sizeof *src ==
809 sizeof fid_seq(src) +
810 sizeof fid_oid(src) + sizeof fid_ver(src));
811 dst->f_seq = cpu_to_be64(fid_seq(src));
812 dst->f_oid = cpu_to_be32(fid_oid(src));
813 dst->f_ver = cpu_to_be32(fid_ver(src));
816 static inline void fid_be_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
818 /* check that all fields are converted */
819 CLASSERT(sizeof *src ==
820 sizeof fid_seq(src) +
821 sizeof fid_oid(src) + sizeof fid_ver(src));
822 dst->f_seq = be64_to_cpu(fid_seq(src));
823 dst->f_oid = be32_to_cpu(fid_oid(src));
824 dst->f_ver = be32_to_cpu(fid_ver(src));
827 static inline int fid_is_sane(const struct lu_fid *fid)
829 return fid != NULL &&
830 ((fid_seq(fid) >= FID_SEQ_START && fid_ver(fid) == 0) ||
831 fid_is_igif(fid) || fid_is_idif(fid) ||
832 fid_seq_is_rsvd(fid_seq(fid)));
835 static inline int fid_is_zero(const struct lu_fid *fid)
837 return fid_seq(fid) == 0 && fid_oid(fid) == 0;
840 extern void lustre_swab_lu_fid(struct lu_fid *fid);
841 extern void lustre_swab_lu_seq_range(struct lu_seq_range *range);
843 static inline int lu_fid_eq(const struct lu_fid *f0, const struct lu_fid *f1)
845 /* Check that there is no alignment padding. */
846 CLASSERT(sizeof *f0 ==
847 sizeof f0->f_seq + sizeof f0->f_oid + sizeof f0->f_ver);
848 return memcmp(f0, f1, sizeof *f0) == 0;
851 #define __diff_normalize(val0, val1) \
853 typeof(val0) __val0 = (val0); \
854 typeof(val1) __val1 = (val1); \
856 (__val0 == __val1 ? 0 : __val0 > __val1 ? +1 : -1); \
859 static inline int lu_fid_cmp(const struct lu_fid *f0,
860 const struct lu_fid *f1)
863 __diff_normalize(fid_seq(f0), fid_seq(f1)) ?:
864 __diff_normalize(fid_oid(f0), fid_oid(f1)) ?:
865 __diff_normalize(fid_ver(f0), fid_ver(f1));
870 /** \defgroup lu_dir lu_dir
874 * Enumeration of possible directory entry attributes.
876 * Attributes follow directory entry header in the order they appear in this
879 enum lu_dirent_attrs {
882 LUDA_64BITHASH = 0x0004,
884 /* The following attrs are used for MDT interanl only,
885 * not visible to client */
887 /* Verify the dirent consistency */
888 LUDA_VERIFY = 0x8000,
889 /* Only check but not repair the dirent inconsistency */
890 LUDA_VERIFY_DRYRUN = 0x4000,
891 /* The dirent has been repaired, or to be repaired (dryrun). */
892 LUDA_REPAIR = 0x2000,
893 /* The system is upgraded, has beed or to be repaired (dryrun). */
894 LUDA_UPGRADE = 0x1000,
895 /* Ignore this record, go to next directly. */
896 LUDA_IGNORE = 0x0800,
899 #define LU_DIRENT_ATTRS_MASK 0xf800
902 * Layout of readdir pages, as transmitted on wire.
905 /** valid if LUDA_FID is set. */
906 struct lu_fid lde_fid;
907 /** a unique entry identifier: a hash or an offset. */
909 /** total record length, including all attributes. */
913 /** optional variable size attributes following this entry.
914 * taken from enum lu_dirent_attrs.
917 /** name is followed by the attributes indicated in ->ldp_attrs, in
918 * their natural order. After the last attribute, padding bytes are
919 * added to make ->lde_reclen a multiple of 8.
925 * Definitions of optional directory entry attributes formats.
927 * Individual attributes do not have their length encoded in a generic way. It
928 * is assumed that consumer of an attribute knows its format. This means that
929 * it is impossible to skip over an unknown attribute, except by skipping over all
930 * remaining attributes (by using ->lde_reclen), which is not too
931 * constraining, because new server versions will append new attributes at
932 * the end of an entry.
936 * Fid directory attribute: a fid of an object referenced by the entry. This
937 * will be almost always requested by the client and supplied by the server.
939 * Aligned to 8 bytes.
941 /* To have compatibility with 1.8, lets have fid in lu_dirent struct. */
946 * Aligned to 2 bytes.
953 __u64 ldp_hash_start;
957 struct lu_dirent ldp_entries[0];
960 enum lu_dirpage_flags {
962 * dirpage contains no entry.
966 * last entry's lde_hash equals ldp_hash_end.
971 static inline struct lu_dirent *lu_dirent_start(struct lu_dirpage *dp)
973 if (le32_to_cpu(dp->ldp_flags) & LDF_EMPTY)
976 return dp->ldp_entries;
979 static inline struct lu_dirent *lu_dirent_next(struct lu_dirent *ent)
981 struct lu_dirent *next;
983 if (le16_to_cpu(ent->lde_reclen) != 0)
984 next = ((void *)ent) + le16_to_cpu(ent->lde_reclen);
991 static inline int lu_dirent_calc_size(int namelen, __u16 attr)
995 if (attr & LUDA_TYPE) {
996 const unsigned align = sizeof(struct luda_type) - 1;
997 size = (sizeof(struct lu_dirent) + namelen + align) & ~align;
998 size += sizeof(struct luda_type);
1000 size = sizeof(struct lu_dirent) + namelen;
1002 return (size + 7) & ~7;
1005 static inline int lu_dirent_size(struct lu_dirent *ent)
1007 if (le16_to_cpu(ent->lde_reclen) == 0) {
1008 return lu_dirent_calc_size(le16_to_cpu(ent->lde_namelen),
1009 le32_to_cpu(ent->lde_attrs));
1011 return le16_to_cpu(ent->lde_reclen);
1014 #define MDS_DIR_END_OFF 0xfffffffffffffffeULL
1017 * MDS_READPAGE page size
1019 * This is the directory page size packed in MDS_READPAGE RPC.
1020 * It's different than CFS_PAGE_SIZE because the client needs to
1021 * access the struct lu_dirpage header packed at the beginning of
1022 * the "page" and without this there isn't any way to know find the
1023 * lu_dirpage header is if client and server CFS_PAGE_SIZE differ.
1025 #define LU_PAGE_SHIFT 12
1026 #define LU_PAGE_SIZE (1UL << LU_PAGE_SHIFT)
1027 #define LU_PAGE_MASK (~(LU_PAGE_SIZE - 1))
1029 #define LU_PAGE_COUNT (1 << (CFS_PAGE_SHIFT - LU_PAGE_SHIFT))
1033 struct lustre_handle {
1036 #define DEAD_HANDLE_MAGIC 0xdeadbeefcafebabeULL
1038 static inline int lustre_handle_is_used(struct lustre_handle *lh)
1040 return lh->cookie != 0ull;
1043 static inline int lustre_handle_equal(const struct lustre_handle *lh1,
1044 const struct lustre_handle *lh2)
1046 return lh1->cookie == lh2->cookie;
1049 static inline void lustre_handle_copy(struct lustre_handle *tgt,
1050 struct lustre_handle *src)
1052 tgt->cookie = src->cookie;
1055 /* flags for lm_flags */
1056 #define MSGHDR_AT_SUPPORT 0x1
1057 #define MSGHDR_CKSUM_INCOMPAT18 0x2
1059 #define lustre_msg lustre_msg_v2
1060 /* we depend on this structure to be 8-byte aligned */
1061 /* this type is only endian-adjusted in lustre_unpack_msg() */
1062 struct lustre_msg_v2 {
1071 __u32 lm_buflens[0];
1074 /* without gss, ptlrpc_body is put at the first buffer. */
1075 #define PTLRPC_NUM_VERSIONS 4
1076 #define JOBSTATS_JOBID_SIZE 32 /* 32 bytes string */
1077 struct ptlrpc_body_v3 {
1078 struct lustre_handle pb_handle;
1085 __u64 pb_last_committed;
1090 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1091 __u32 pb_service_time; /* for rep, actual service time */
1094 /* VBR: pre-versions */
1095 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1096 /* padding for future needs */
1097 __u64 pb_padding[4];
1098 char pb_jobid[JOBSTATS_JOBID_SIZE];
1100 #define ptlrpc_body ptlrpc_body_v3
1102 struct ptlrpc_body_v2 {
1103 struct lustre_handle pb_handle;
1110 __u64 pb_last_committed;
1115 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1116 __u32 pb_service_time; /* for rep, actual service time, also used for
1117 net_latency of req */
1120 /* VBR: pre-versions */
1121 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1122 /* padding for future needs */
1123 __u64 pb_padding[4];
1126 extern void lustre_swab_ptlrpc_body(struct ptlrpc_body *pb);
1128 /* message body offset for lustre_msg_v2 */
1129 /* ptlrpc body offset in all request/reply messages */
1130 #define MSG_PTLRPC_BODY_OFF 0
1132 /* normal request/reply message record offset */
1133 #define REQ_REC_OFF 1
1134 #define REPLY_REC_OFF 1
1136 /* ldlm request message body offset */
1137 #define DLM_LOCKREQ_OFF 1 /* lockreq offset */
1138 #define DLM_REQ_REC_OFF 2 /* normal dlm request record offset */
1140 /* ldlm intent lock message body offset */
1141 #define DLM_INTENT_IT_OFF 2 /* intent lock it offset */
1142 #define DLM_INTENT_REC_OFF 3 /* intent lock record offset */
1144 /* ldlm reply message body offset */
1145 #define DLM_LOCKREPLY_OFF 1 /* lockrep offset */
1146 #define DLM_REPLY_REC_OFF 2 /* reply record offset */
1148 /** only use in req->rq_{req,rep}_swab_mask */
1149 #define MSG_PTLRPC_HEADER_OFF 31
1151 /* Flags that are operation-specific go in the top 16 bits. */
1152 #define MSG_OP_FLAG_MASK 0xffff0000
1153 #define MSG_OP_FLAG_SHIFT 16
1155 /* Flags that apply to all requests are in the bottom 16 bits */
1156 #define MSG_GEN_FLAG_MASK 0x0000ffff
1157 #define MSG_LAST_REPLAY 0x0001
1158 #define MSG_RESENT 0x0002
1159 #define MSG_REPLAY 0x0004
1160 /* #define MSG_AT_SUPPORT 0x0008
1161 * This was used in early prototypes of adaptive timeouts, and while there
1162 * shouldn't be any users of that code there also isn't a need for using this
1163 * bits. Defer usage until at least 1.10 to avoid potential conflict. */
1164 #define MSG_DELAY_REPLAY 0x0010
1165 #define MSG_VERSION_REPLAY 0x0020
1166 #define MSG_REQ_REPLAY_DONE 0x0040
1167 #define MSG_LOCK_REPLAY_DONE 0x0080
1170 * Flags for all connect opcodes (MDS_CONNECT, OST_CONNECT)
1173 #define MSG_CONNECT_RECOVERING 0x00000001
1174 #define MSG_CONNECT_RECONNECT 0x00000002
1175 #define MSG_CONNECT_REPLAYABLE 0x00000004
1176 //#define MSG_CONNECT_PEER 0x8
1177 #define MSG_CONNECT_LIBCLIENT 0x00000010
1178 #define MSG_CONNECT_INITIAL 0x00000020
1179 #define MSG_CONNECT_ASYNC 0x00000040
1180 #define MSG_CONNECT_NEXT_VER 0x00000080 /* use next version of lustre_msg */
1181 #define MSG_CONNECT_TRANSNO 0x00000100 /* report transno */
1184 #define OBD_CONNECT_RDONLY 0x1ULL /*client has read-only access*/
1185 #define OBD_CONNECT_INDEX 0x2ULL /*connect specific LOV idx */
1186 #define OBD_CONNECT_MDS 0x4ULL /*connect from MDT to OST */
1187 #define OBD_CONNECT_GRANT 0x8ULL /*OSC gets grant at connect */
1188 #define OBD_CONNECT_SRVLOCK 0x10ULL /*server takes locks for cli */
1189 #define OBD_CONNECT_VERSION 0x20ULL /*Lustre versions in ocd */
1190 #define OBD_CONNECT_REQPORTAL 0x40ULL /*Separate non-IO req portal */
1191 #define OBD_CONNECT_ACL 0x80ULL /*access control lists */
1192 #define OBD_CONNECT_XATTR 0x100ULL /*client use extended attr */
1193 #define OBD_CONNECT_CROW 0x200ULL /*MDS+OST create obj on write*/
1194 #define OBD_CONNECT_TRUNCLOCK 0x400ULL /*locks on server for punch */
1195 #define OBD_CONNECT_TRANSNO 0x800ULL /*replay sends init transno */
1196 #define OBD_CONNECT_IBITS 0x1000ULL /*support for inodebits locks*/
1197 #define OBD_CONNECT_JOIN 0x2000ULL /*files can be concatenated.
1198 *We do not support JOIN FILE
1199 *anymore, reserve this flags
1200 *just for preventing such bit
1202 #define OBD_CONNECT_ATTRFID 0x4000ULL /*Server can GetAttr By Fid*/
1203 #define OBD_CONNECT_NODEVOH 0x8000ULL /*No open hndl on specl nodes*/
1204 #define OBD_CONNECT_RMT_CLIENT 0x10000ULL /*Remote client */
1205 #define OBD_CONNECT_RMT_CLIENT_FORCE 0x20000ULL /*Remote client by force */
1206 #define OBD_CONNECT_BRW_SIZE 0x40000ULL /*Max bytes per rpc */
1207 #define OBD_CONNECT_QUOTA64 0x80000ULL /*Not used since 2.4 */
1208 #define OBD_CONNECT_MDS_CAPA 0x100000ULL /*MDS capability */
1209 #define OBD_CONNECT_OSS_CAPA 0x200000ULL /*OSS capability */
1210 #define OBD_CONNECT_CANCELSET 0x400000ULL /*Early batched cancels. */
1211 #define OBD_CONNECT_SOM 0x800000ULL /*Size on MDS */
1212 #define OBD_CONNECT_AT 0x1000000ULL /*client uses AT */
1213 #define OBD_CONNECT_LRU_RESIZE 0x2000000ULL /*LRU resize feature. */
1214 #define OBD_CONNECT_MDS_MDS 0x4000000ULL /*MDS-MDS connection */
1215 #define OBD_CONNECT_REAL 0x8000000ULL /*real connection */
1216 #define OBD_CONNECT_CHANGE_QS 0x10000000ULL /*Not used since 2.4 */
1217 #define OBD_CONNECT_CKSUM 0x20000000ULL /*support several cksum algos*/
1218 #define OBD_CONNECT_FID 0x40000000ULL /*FID is supported by server */
1219 #define OBD_CONNECT_VBR 0x80000000ULL /*version based recovery */
1220 #define OBD_CONNECT_LOV_V3 0x100000000ULL /*client supports LOV v3 EA */
1221 #define OBD_CONNECT_GRANT_SHRINK 0x200000000ULL /* support grant shrink */
1222 #define OBD_CONNECT_SKIP_ORPHAN 0x400000000ULL /* don't reuse orphan objids */
1223 #define OBD_CONNECT_MAX_EASIZE 0x800000000ULL /* preserved for large EA */
1224 #define OBD_CONNECT_FULL20 0x1000000000ULL /* it is 2.0 client */
1225 #define OBD_CONNECT_LAYOUTLOCK 0x2000000000ULL /* client uses layout lock */
1226 #define OBD_CONNECT_64BITHASH 0x4000000000ULL /* client supports 64-bits
1228 #define OBD_CONNECT_MAXBYTES 0x8000000000ULL /* max stripe size */
1229 #define OBD_CONNECT_IMP_RECOV 0x10000000000ULL /* imp recovery support */
1230 #define OBD_CONNECT_JOBSTATS 0x20000000000ULL /* jobid in ptlrpc_body */
1231 #define OBD_CONNECT_UMASK 0x40000000000ULL /* create uses client umask */
1232 #define OBD_CONNECT_EINPROGRESS 0x80000000000ULL /* client handles -EINPROGRESS
1233 * RPC error properly */
1234 #define OBD_CONNECT_GRANT_PARAM 0x100000000000ULL/* extra grant params used for
1235 * finer space reservation */
1236 #define OBD_CONNECT_FLOCK_OWNER 0x200000000000ULL /* for the fixed 1.8
1237 * policy and 2.x server */
1238 #define OBD_CONNECT_LVB_TYPE 0x400000000000ULL /* variable type of LVB */
1239 #define OBD_CONNECT_NANOSEC_TIME 0x800000000000ULL /* nanosecond timestamps */
1240 #define OBD_CONNECT_LIGHTWEIGHT 0x1000000000000ULL/* lightweight connection */
1241 #define OBD_CONNECT_SHORTIO 0x2000000000000ULL/* short io */
1242 #define OBD_CONNECT_PINGLESS 0x4000000000000ULL/* pings not required */
1244 * Please DO NOT add flag values here before first ensuring that this same
1245 * flag value is not in use on some other branch. Please clear any such
1246 * changes with senior engineers before starting to use a new flag. Then,
1247 * submit a small patch against EVERY branch that ONLY adds the new flag,
1248 * updates obd_connect_names[] for lprocfs_rd_connect_flags(), adds the
1249 * flag to check_obd_connect_data(), and updates wiretests accordingly, so it
1250 * can be approved and landed easily to reserve the flag for future use. */
1252 /* The MNE_SWAB flag is overloading the MDS_MDS bit only for the MGS
1253 * connection. It is a temporary bug fix for Imperative Recovery interop
1254 * between 2.2 and 2.3 x86/ppc nodes, and can be removed when interop for
1255 * 2.2 clients/servers is no longer needed. LU-1252/LU-1644. */
1256 #define OBD_CONNECT_MNE_SWAB OBD_CONNECT_MDS_MDS
1258 #define OCD_HAS_FLAG(ocd, flg) \
1259 (!!((ocd)->ocd_connect_flags & OBD_CONNECT_##flg))
1262 #ifdef HAVE_LRU_RESIZE_SUPPORT
1263 #define LRU_RESIZE_CONNECT_FLAG OBD_CONNECT_LRU_RESIZE
1265 #define LRU_RESIZE_CONNECT_FLAG 0
1268 #define MDT_CONNECT_SUPPORTED (OBD_CONNECT_RDONLY | OBD_CONNECT_VERSION | \
1269 OBD_CONNECT_ACL | OBD_CONNECT_XATTR | \
1270 OBD_CONNECT_IBITS | \
1271 OBD_CONNECT_NODEVOH | OBD_CONNECT_ATTRFID | \
1272 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1273 OBD_CONNECT_RMT_CLIENT | \
1274 OBD_CONNECT_RMT_CLIENT_FORCE | \
1275 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_MDS_CAPA | \
1276 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_MDS_MDS | \
1277 OBD_CONNECT_FID | LRU_RESIZE_CONNECT_FLAG | \
1278 OBD_CONNECT_VBR | OBD_CONNECT_LOV_V3 | \
1279 OBD_CONNECT_SOM | OBD_CONNECT_FULL20 | \
1280 OBD_CONNECT_64BITHASH | OBD_CONNECT_JOBSTATS | \
1281 OBD_CONNECT_EINPROGRESS | \
1282 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_UMASK | \
1283 OBD_CONNECT_LVB_TYPE | OBD_CONNECT_LAYOUTLOCK |\
1284 OBD_CONNECT_PINGLESS)
1285 #define OST_CONNECT_SUPPORTED (OBD_CONNECT_SRVLOCK | OBD_CONNECT_GRANT | \
1286 OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | \
1287 OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_INDEX | \
1288 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_OSS_CAPA | \
1289 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1290 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_CKSUM | \
1291 OBD_CONNECT_RMT_CLIENT | \
1292 OBD_CONNECT_RMT_CLIENT_FORCE | OBD_CONNECT_VBR | \
1293 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN | \
1294 OBD_CONNECT_GRANT_SHRINK | OBD_CONNECT_FULL20 | \
1295 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES | \
1296 OBD_CONNECT_MAX_EASIZE | \
1297 OBD_CONNECT_EINPROGRESS | \
1298 OBD_CONNECT_JOBSTATS | \
1299 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LVB_TYPE|\
1300 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_FID | \
1301 OBD_CONNECT_PINGLESS)
1302 #define ECHO_CONNECT_SUPPORTED (0)
1303 #define MGS_CONNECT_SUPPORTED (OBD_CONNECT_VERSION | OBD_CONNECT_AT | \
1304 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV | \
1305 OBD_CONNECT_MNE_SWAB | OBD_CONNECT_PINGLESS)
1307 /* Features required for this version of the client to work with server */
1308 #define CLIENT_CONNECT_MDT_REQD (OBD_CONNECT_IBITS | OBD_CONNECT_FID | \
1311 #define OBD_OCD_VERSION(major,minor,patch,fix) (((major)<<24) + ((minor)<<16) +\
1312 ((patch)<<8) + (fix))
1313 #define OBD_OCD_VERSION_MAJOR(version) ((int)((version)>>24)&255)
1314 #define OBD_OCD_VERSION_MINOR(version) ((int)((version)>>16)&255)
1315 #define OBD_OCD_VERSION_PATCH(version) ((int)((version)>>8)&255)
1316 #define OBD_OCD_VERSION_FIX(version) ((int)(version)&255)
1318 /* This structure is used for both request and reply.
1320 * If we eventually have separate connect data for different types, which we
1321 * almost certainly will, then perhaps we stick a union in here. */
1322 struct obd_connect_data_v1 {
1323 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1324 __u32 ocd_version; /* lustre release version number */
1325 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1326 __u32 ocd_index; /* LOV index to connect to */
1327 __u32 ocd_brw_size; /* Maximum BRW size in bytes, must be 2^n */
1328 __u64 ocd_ibits_known; /* inode bits this client understands */
1329 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1330 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1331 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1332 __u32 ocd_unused; /* also fix lustre_swab_connect */
1333 __u64 ocd_transno; /* first transno from client to be replayed */
1334 __u32 ocd_group; /* MDS group on OST */
1335 __u32 ocd_cksum_types; /* supported checksum algorithms */
1336 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1337 __u32 ocd_instance; /* also fix lustre_swab_connect */
1338 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1341 struct obd_connect_data {
1342 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1343 __u32 ocd_version; /* lustre release version number */
1344 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1345 __u32 ocd_index; /* LOV index to connect to */
1346 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
1347 __u64 ocd_ibits_known; /* inode bits this client understands */
1348 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1349 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1350 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1351 __u32 ocd_unused; /* also fix lustre_swab_connect */
1352 __u64 ocd_transno; /* first transno from client to be replayed */
1353 __u32 ocd_group; /* MDS group on OST */
1354 __u32 ocd_cksum_types; /* supported checksum algorithms */
1355 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1356 __u32 ocd_instance; /* instance # of this target */
1357 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1358 /* Fields after ocd_maxbytes are only accessible by the receiver
1359 * if the corresponding flag in ocd_connect_flags is set. Accessing
1360 * any field after ocd_maxbytes on the receiver without a valid flag
1361 * may result in out-of-bound memory access and kernel oops. */
1362 __u64 padding1; /* added 2.1.0. also fix lustre_swab_connect */
1363 __u64 padding2; /* added 2.1.0. also fix lustre_swab_connect */
1364 __u64 padding3; /* added 2.1.0. also fix lustre_swab_connect */
1365 __u64 padding4; /* added 2.1.0. also fix lustre_swab_connect */
1366 __u64 padding5; /* added 2.1.0. also fix lustre_swab_connect */
1367 __u64 padding6; /* added 2.1.0. also fix lustre_swab_connect */
1368 __u64 padding7; /* added 2.1.0. also fix lustre_swab_connect */
1369 __u64 padding8; /* added 2.1.0. also fix lustre_swab_connect */
1370 __u64 padding9; /* added 2.1.0. also fix lustre_swab_connect */
1371 __u64 paddingA; /* added 2.1.0. also fix lustre_swab_connect */
1372 __u64 paddingB; /* added 2.1.0. also fix lustre_swab_connect */
1373 __u64 paddingC; /* added 2.1.0. also fix lustre_swab_connect */
1374 __u64 paddingD; /* added 2.1.0. also fix lustre_swab_connect */
1375 __u64 paddingE; /* added 2.1.0. also fix lustre_swab_connect */
1376 __u64 paddingF; /* added 2.1.0. also fix lustre_swab_connect */
1379 * Please DO NOT use any fields here before first ensuring that this same
1380 * field is not in use on some other branch. Please clear any such changes
1381 * with senior engineers before starting to use a new field. Then, submit
1382 * a small patch against EVERY branch that ONLY adds the new field along with
1383 * the matching OBD_CONNECT flag, so that can be approved and landed easily to
1384 * reserve the flag for future use. */
1387 extern void lustre_swab_connect(struct obd_connect_data *ocd);
1390 * Supported checksum algorithms. Up to 32 checksum types are supported.
1391 * (32-bit mask stored in obd_connect_data::ocd_cksum_types)
1392 * Please update DECLARE_CKSUM_NAME/OBD_CKSUM_ALL in obd.h when adding a new
1393 * algorithm and also the OBD_FL_CKSUM* flags.
1396 OBD_CKSUM_CRC32 = 0x00000001,
1397 OBD_CKSUM_ADLER = 0x00000002,
1398 OBD_CKSUM_CRC32C= 0x00000004,
1402 * OST requests: OBDO & OBD request records
1407 OST_REPLY = 0, /* reply ? */
1423 OST_QUOTACHECK = 18,
1425 OST_QUOTA_ADJUST_QUNIT = 20, /* not used since 2.4 */
1428 #define OST_FIRST_OPC OST_REPLY
1431 OBD_FL_INLINEDATA = 0x00000001,
1432 OBD_FL_OBDMDEXISTS = 0x00000002,
1433 OBD_FL_DELORPHAN = 0x00000004, /* if set in o_flags delete orphans */
1434 OBD_FL_NORPC = 0x00000008, /* set in o_flags do in OSC not OST */
1435 OBD_FL_IDONLY = 0x00000010, /* set in o_flags only adjust obj id*/
1436 OBD_FL_RECREATE_OBJS= 0x00000020, /* recreate missing obj */
1437 OBD_FL_DEBUG_CHECK = 0x00000040, /* echo client/server debug check */
1438 OBD_FL_NO_USRQUOTA = 0x00000100, /* the object's owner is over quota */
1439 OBD_FL_NO_GRPQUOTA = 0x00000200, /* the object's group is over quota */
1440 OBD_FL_CREATE_CROW = 0x00000400, /* object should be create on write */
1441 OBD_FL_SRVLOCK = 0x00000800, /* delegate DLM locking to server */
1442 OBD_FL_CKSUM_CRC32 = 0x00001000, /* CRC32 checksum type */
1443 OBD_FL_CKSUM_ADLER = 0x00002000, /* ADLER checksum type */
1444 OBD_FL_CKSUM_CRC32C = 0x00004000, /* CRC32C checksum type */
1445 OBD_FL_CKSUM_RSVD2 = 0x00008000, /* for future cksum types */
1446 OBD_FL_CKSUM_RSVD3 = 0x00010000, /* for future cksum types */
1447 OBD_FL_SHRINK_GRANT = 0x00020000, /* object shrink the grant */
1448 OBD_FL_MMAP = 0x00040000, /* object is mmapped on the client.
1449 * XXX: obsoleted - reserved for old
1450 * clients prior than 2.2 */
1451 OBD_FL_RECOV_RESEND = 0x00080000, /* recoverable resent */
1452 OBD_FL_NOSPC_BLK = 0x00100000, /* no more block space on OST */
1454 /* Note that while these checksum values are currently separate bits,
1455 * in 2.x we can actually allow all values from 1-31 if we wanted. */
1456 OBD_FL_CKSUM_ALL = OBD_FL_CKSUM_CRC32 | OBD_FL_CKSUM_ADLER |
1457 OBD_FL_CKSUM_CRC32C,
1459 /* mask for local-only flag, which won't be sent over network */
1460 OBD_FL_LOCAL_MASK = 0xF0000000,
1463 #define LOV_MAGIC_V1 0x0BD10BD0
1464 #define LOV_MAGIC LOV_MAGIC_V1
1465 #define LOV_MAGIC_JOIN_V1 0x0BD20BD0
1466 #define LOV_MAGIC_V3 0x0BD30BD0
1469 * magic for fully defined striping
1470 * the idea is that we should have different magics for striping "hints"
1471 * (struct lov_user_md_v[13]) and defined ready-to-use striping (struct
1472 * lov_mds_md_v[13]). at the moment the magics are used in wire protocol,
1473 * we can't just change it w/o long way preparation, but we still need a
1474 * mechanism to allow LOD to differentiate hint versus ready striping.
1475 * so, at the moment we do a trick: MDT knows what to expect from request
1476 * depending on the case (replay uses ready striping, non-replay req uses
1477 * hints), so MDT replaces magic with appropriate one and now LOD can
1478 * easily understand what's inside -bzzz
1480 #define LOV_MAGIC_V1_DEF 0x0CD10BD0
1481 #define LOV_MAGIC_V3_DEF 0x0CD30BD0
1483 #define LOV_PATTERN_RAID0 0x001 /* stripes are used round-robin */
1484 #define LOV_PATTERN_RAID1 0x002 /* stripes are mirrors of each other */
1485 #define LOV_PATTERN_FIRST 0x100 /* first stripe is not in round-robin */
1486 #define LOV_PATTERN_CMOBD 0x200
1488 #define lov_ost_data lov_ost_data_v1
1489 struct lov_ost_data_v1 { /* per-stripe data structure (little-endian)*/
1490 __u64 l_object_id; /* OST object ID */
1491 __u64 l_object_seq; /* OST object seq number */
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 __u64 lmm_object_id; /* LOV object ID */
1501 __u64 lmm_object_seq; /* LOV object seq number */
1502 __u32 lmm_stripe_size; /* size of stripe in bytes */
1503 /* lmm_stripe_count used to be __u32 */
1504 __u16 lmm_stripe_count; /* num stripes in use for this object */
1505 __u16 lmm_layout_gen; /* layout generation number */
1506 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1509 /* extern void lustre_swab_lov_mds_md(struct lov_mds_md *llm); */
1511 #define MAX_MD_SIZE (sizeof(struct lov_mds_md) + 4 * sizeof(struct lov_ost_data))
1512 #define MIN_MD_SIZE (sizeof(struct lov_mds_md) + 1 * sizeof(struct lov_ost_data))
1514 #define XATTR_NAME_ACL_ACCESS "system.posix_acl_access"
1515 #define XATTR_NAME_ACL_DEFAULT "system.posix_acl_default"
1516 #define XATTR_USER_PREFIX "user."
1517 #define XATTR_TRUSTED_PREFIX "trusted."
1518 #define XATTR_SECURITY_PREFIX "security."
1519 #define XATTR_LUSTRE_PREFIX "lustre."
1521 #define XATTR_NAME_LOV "trusted.lov"
1522 #define XATTR_NAME_LMA "trusted.lma"
1523 #define XATTR_NAME_LMV "trusted.lmv"
1524 #define XATTR_NAME_LINK "trusted.link"
1525 #define XATTR_NAME_FID "trusted.fid"
1526 #define XATTR_NAME_VERSION "trusted.version"
1527 #define XATTR_NAME_SOM "trusted.som"
1528 #define XATTR_NAME_HSM "trusted.hsm"
1529 #define XATTR_NAME_LFSCK_NAMESPACE "trusted.lfsck_namespace"
1532 struct lov_mds_md_v3 { /* LOV EA mds/wire data (little-endian) */
1533 __u32 lmm_magic; /* magic number = LOV_MAGIC_V3 */
1534 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1535 __u64 lmm_object_id; /* LOV object ID */
1536 __u64 lmm_object_seq; /* LOV object seq number */
1537 __u32 lmm_stripe_size; /* size of stripe in bytes */
1538 /* lmm_stripe_count used to be __u32 */
1539 __u16 lmm_stripe_count; /* num stripes in use for this object */
1540 __u16 lmm_layout_gen; /* layout generation number */
1541 char lmm_pool_name[LOV_MAXPOOLNAME]; /* must be 32bit aligned */
1542 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1546 #define OBD_MD_FLID (0x00000001ULL) /* object ID */
1547 #define OBD_MD_FLATIME (0x00000002ULL) /* access time */
1548 #define OBD_MD_FLMTIME (0x00000004ULL) /* data modification time */
1549 #define OBD_MD_FLCTIME (0x00000008ULL) /* change time */
1550 #define OBD_MD_FLSIZE (0x00000010ULL) /* size */
1551 #define OBD_MD_FLBLOCKS (0x00000020ULL) /* allocated blocks count */
1552 #define OBD_MD_FLBLKSZ (0x00000040ULL) /* block size */
1553 #define OBD_MD_FLMODE (0x00000080ULL) /* access bits (mode & ~S_IFMT) */
1554 #define OBD_MD_FLTYPE (0x00000100ULL) /* object type (mode & S_IFMT) */
1555 #define OBD_MD_FLUID (0x00000200ULL) /* user ID */
1556 #define OBD_MD_FLGID (0x00000400ULL) /* group ID */
1557 #define OBD_MD_FLFLAGS (0x00000800ULL) /* flags word */
1558 #define OBD_MD_FLNLINK (0x00002000ULL) /* link count */
1559 #define OBD_MD_FLGENER (0x00004000ULL) /* generation number */
1560 /*#define OBD_MD_FLINLINE (0x00008000ULL) inline data. used until 1.6.5 */
1561 #define OBD_MD_FLRDEV (0x00010000ULL) /* device number */
1562 #define OBD_MD_FLEASIZE (0x00020000ULL) /* extended attribute data */
1563 #define OBD_MD_LINKNAME (0x00040000ULL) /* symbolic link target */
1564 #define OBD_MD_FLHANDLE (0x00080000ULL) /* file/lock handle */
1565 #define OBD_MD_FLCKSUM (0x00100000ULL) /* bulk data checksum */
1566 #define OBD_MD_FLQOS (0x00200000ULL) /* quality of service stats */
1567 /*#define OBD_MD_FLOSCOPQ (0x00400000ULL) osc opaque data, never used */
1568 #define OBD_MD_FLCOOKIE (0x00800000ULL) /* log cancellation cookie */
1569 #define OBD_MD_FLGROUP (0x01000000ULL) /* group */
1570 #define OBD_MD_FLFID (0x02000000ULL) /* ->ost write inline fid */
1571 #define OBD_MD_FLEPOCH (0x04000000ULL) /* ->ost write with ioepoch */
1572 /* ->mds if epoch opens or closes */
1573 #define OBD_MD_FLGRANT (0x08000000ULL) /* ost preallocation space grant */
1574 #define OBD_MD_FLDIREA (0x10000000ULL) /* dir's extended attribute data */
1575 #define OBD_MD_FLUSRQUOTA (0x20000000ULL) /* over quota flags sent from ost */
1576 #define OBD_MD_FLGRPQUOTA (0x40000000ULL) /* over quota flags sent from ost */
1577 #define OBD_MD_FLMODEASIZE (0x80000000ULL) /* EA size will be changed */
1579 #define OBD_MD_MDS (0x0000000100000000ULL) /* where an inode lives on */
1580 #define OBD_MD_REINT (0x0000000200000000ULL) /* reintegrate oa */
1581 #define OBD_MD_MEA (0x0000000400000000ULL) /* CMD split EA */
1583 /* OBD_MD_MDTIDX is used to get MDT index, but it is never been used overwire,
1584 * and it is already obsolete since 2.3 */
1585 /* #define OBD_MD_MDTIDX (0x0000000800000000ULL) */
1587 #define OBD_MD_FLXATTR (0x0000001000000000ULL) /* xattr */
1588 #define OBD_MD_FLXATTRLS (0x0000002000000000ULL) /* xattr list */
1589 #define OBD_MD_FLXATTRRM (0x0000004000000000ULL) /* xattr remove */
1590 #define OBD_MD_FLACL (0x0000008000000000ULL) /* ACL */
1591 #define OBD_MD_FLRMTPERM (0x0000010000000000ULL) /* remote permission */
1592 #define OBD_MD_FLMDSCAPA (0x0000020000000000ULL) /* MDS capability */
1593 #define OBD_MD_FLOSSCAPA (0x0000040000000000ULL) /* OSS capability */
1594 #define OBD_MD_FLCKSPLIT (0x0000080000000000ULL) /* Check split on server */
1595 #define OBD_MD_FLCROSSREF (0x0000100000000000ULL) /* Cross-ref case */
1596 #define OBD_MD_FLGETATTRLOCK (0x0000200000000000ULL) /* Get IOEpoch attributes
1598 #define OBD_MD_FLOBJCOUNT (0x0000400000000000ULL) /* for multiple destroy */
1600 #define OBD_MD_FLRMTLSETFACL (0x0001000000000000ULL) /* lfs lsetfacl case */
1601 #define OBD_MD_FLRMTLGETFACL (0x0002000000000000ULL) /* lfs lgetfacl case */
1602 #define OBD_MD_FLRMTRSETFACL (0x0004000000000000ULL) /* lfs rsetfacl case */
1603 #define OBD_MD_FLRMTRGETFACL (0x0008000000000000ULL) /* lfs rgetfacl case */
1605 #define OBD_MD_FLDATAVERSION (0x0010000000000000ULL) /* iversion sum */
1607 #define OBD_MD_FLGETATTR (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLMTIME | \
1608 OBD_MD_FLCTIME | OBD_MD_FLSIZE | OBD_MD_FLBLKSZ | \
1609 OBD_MD_FLMODE | OBD_MD_FLTYPE | OBD_MD_FLUID | \
1610 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLNLINK | \
1611 OBD_MD_FLGENER | OBD_MD_FLRDEV | OBD_MD_FLGROUP)
1613 /* don't forget obdo_fid which is way down at the bottom so it can
1614 * come after the definition of llog_cookie */
1618 HSS_CLEARMASK = 0x02,
1619 HSS_ARCHIVE_ID = 0x04,
1622 struct hsm_state_set {
1624 __u32 hss_archive_id;
1626 __u64 hss_clearmask;
1629 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
1630 extern void lustre_swab_hsm_state_set(struct hsm_state_set *hss);
1632 extern void lustre_swab_obd_statfs (struct obd_statfs *os);
1634 /* ost_body.data values for OST_BRW */
1636 #define OBD_BRW_READ 0x01
1637 #define OBD_BRW_WRITE 0x02
1638 #define OBD_BRW_RWMASK (OBD_BRW_READ | OBD_BRW_WRITE)
1639 #define OBD_BRW_SYNC 0x08 /* this page is a part of synchronous
1640 * transfer and is not accounted in
1642 #define OBD_BRW_CHECK 0x10
1643 #define OBD_BRW_FROM_GRANT 0x20 /* the osc manages this under llite */
1644 #define OBD_BRW_GRANTED 0x40 /* the ost manages this */
1645 #define OBD_BRW_NOCACHE 0x80 /* this page is a part of non-cached IO */
1646 #define OBD_BRW_NOQUOTA 0x100
1647 #define OBD_BRW_SRVLOCK 0x200 /* Client holds no lock over this page */
1648 #define OBD_BRW_ASYNC 0x400 /* Server may delay commit to disk */
1649 #define OBD_BRW_MEMALLOC 0x800 /* Client runs in the "kswapd" context */
1650 #define OBD_BRW_OVER_USRQUOTA 0x1000 /* Running out of user quota */
1651 #define OBD_BRW_OVER_GRPQUOTA 0x2000 /* Running out of group quota */
1653 #define OBD_OBJECT_EOF 0xffffffffffffffffULL
1655 #define OST_MIN_PRECREATE 32
1656 #define OST_MAX_PRECREATE 20000
1659 struct ost_id ioo_oid; /* object ID, if multi-obj BRW */
1660 __u32 ioo_max_brw; /* low 16 bits were o_mode before 2.4,
1661 * now (PTLRPC_BULK_OPS_COUNT - 1) in
1662 * high 16 bits in 2.4 and later */
1663 __u32 ioo_bufcnt; /* number of niobufs for this object */
1666 #define IOOBJ_MAX_BRW_BITS 16
1667 #define IOOBJ_TYPE_MASK ((1U << IOOBJ_MAX_BRW_BITS) - 1)
1668 #define ioobj_max_brw_get(ioo) (((ioo)->ioo_max_brw >> IOOBJ_MAX_BRW_BITS) + 1)
1669 #define ioobj_max_brw_set(ioo, num) \
1670 do { (ioo)->ioo_max_brw = ((num) - 1) << IOOBJ_MAX_BRW_BITS; } while (0)
1672 #define ioo_id ioo_oid.oi_id
1673 #define ioo_seq ioo_oid.oi_seq
1675 extern void lustre_swab_obd_ioobj (struct obd_ioobj *ioo);
1677 /* multiple of 8 bytes => can array */
1678 struct niobuf_remote {
1684 extern void lustre_swab_niobuf_remote (struct niobuf_remote *nbr);
1686 /* lock value block communicated between the filter and llite */
1688 /* OST_LVB_ERR_INIT is needed because the return code in rc is
1689 * negative, i.e. because ((MASK + rc) & MASK) != MASK. */
1690 #define OST_LVB_ERR_INIT 0xffbadbad80000000ULL
1691 #define OST_LVB_ERR_MASK 0xffbadbad00000000ULL
1692 #define OST_LVB_IS_ERR(blocks) \
1693 ((blocks & OST_LVB_ERR_MASK) == OST_LVB_ERR_MASK)
1694 #define OST_LVB_SET_ERR(blocks, rc) \
1695 do { blocks = OST_LVB_ERR_INIT + rc; } while (0)
1696 #define OST_LVB_GET_ERR(blocks) (int)(blocks - OST_LVB_ERR_INIT)
1706 extern void lustre_swab_ost_lvb_v1(struct ost_lvb_v1 *lvb);
1720 extern void lustre_swab_ost_lvb(struct ost_lvb *lvb);
1723 * lquota data structures
1726 #ifndef QUOTABLOCK_BITS
1727 #define QUOTABLOCK_BITS 10
1730 #ifndef QUOTABLOCK_SIZE
1731 #define QUOTABLOCK_SIZE (1 << QUOTABLOCK_BITS)
1735 #define toqb(x) (((x) + QUOTABLOCK_SIZE - 1) >> QUOTABLOCK_BITS)
1738 /* The lquota_id structure is an union of all the possible identifier types that
1739 * can be used with quota, this includes:
1742 * - a FID which can be used for per-directory quota in the future */
1744 struct lu_fid qid_fid; /* FID for per-directory quota */
1745 __u64 qid_uid; /* user identifier */
1746 __u64 qid_gid; /* group identifier */
1749 /* quotactl management */
1750 struct obd_quotactl {
1752 __u32 qc_type; /* see Q_* flag below */
1755 struct obd_dqinfo qc_dqinfo;
1756 struct obd_dqblk qc_dqblk;
1759 extern void lustre_swab_obd_quotactl(struct obd_quotactl *q);
1761 #define Q_QUOTACHECK 0x800100 /* deprecated as of 2.4 */
1762 #define Q_INITQUOTA 0x800101 /* deprecated as of 2.4 */
1763 #define Q_GETOINFO 0x800102 /* get obd quota info */
1764 #define Q_GETOQUOTA 0x800103 /* get obd quotas */
1765 #define Q_FINVALIDATE 0x800104 /* deprecated as of 2.4 */
1767 #define Q_COPY(out, in, member) (out)->member = (in)->member
1769 #define QCTL_COPY(out, in) \
1771 Q_COPY(out, in, qc_cmd); \
1772 Q_COPY(out, in, qc_type); \
1773 Q_COPY(out, in, qc_id); \
1774 Q_COPY(out, in, qc_stat); \
1775 Q_COPY(out, in, qc_dqinfo); \
1776 Q_COPY(out, in, qc_dqblk); \
1779 /* Body of quota request used for quota acquire/release RPCs between quota
1780 * master (aka QMT) and slaves (ak QSD). */
1782 struct lu_fid qb_fid; /* FID of global index packing the pool ID
1783 * and type (data or metadata) as well as
1784 * the quota type (user or group). */
1785 union lquota_id qb_id; /* uid or gid or directory FID */
1786 __u32 qb_flags; /* see below */
1788 __u64 qb_count; /* acquire/release count (kbytes/inodes) */
1789 __u64 qb_usage; /* current slave usage (kbytes/inodes) */
1790 __u64 qb_slv_ver; /* slave index file version */
1791 struct lustre_handle qb_lockh; /* per-ID lock handle */
1792 struct lustre_handle qb_glb_lockh; /* global lock handle */
1793 __u64 qb_padding1[4];
1796 /* When the quota_body is used in the reply of quota global intent
1797 * lock (IT_QUOTA_CONN) reply, qb_fid contains slave index file FID. */
1798 #define qb_slv_fid qb_fid
1799 /* qb_usage is the current qunit (in kbytes/inodes) when quota_body is used in
1801 #define qb_qunit qb_usage
1803 #define QUOTA_DQACQ_FL_ACQ 0x1 /* acquire quota */
1804 #define QUOTA_DQACQ_FL_PREACQ 0x2 /* pre-acquire */
1805 #define QUOTA_DQACQ_FL_REL 0x4 /* release quota */
1806 #define QUOTA_DQACQ_FL_REPORT 0x8 /* report usage */
1808 extern void lustre_swab_quota_body(struct quota_body *b);
1810 /* Quota types currently supported */
1812 LQUOTA_TYPE_USR = 0x00, /* maps to USRQUOTA */
1813 LQUOTA_TYPE_GRP = 0x01, /* maps to GRPQUOTA */
1817 /* There are 2 different resource types on which a quota limit can be enforced:
1818 * - inodes on the MDTs
1819 * - blocks on the OSTs */
1821 LQUOTA_RES_MD = 0x01, /* skip 0 to avoid null oid in FID */
1822 LQUOTA_RES_DT = 0x02,
1824 LQUOTA_FIRST_RES = LQUOTA_RES_MD
1826 #define LQUOTA_NR_RES (LQUOTA_LAST_RES - LQUOTA_FIRST_RES + 1)
1829 * Space accounting support
1830 * Format of an accounting record, providing disk usage information for a given
1833 struct lquota_acct_rec { /* 16 bytes */
1834 __u64 bspace; /* current space in use */
1835 __u64 ispace; /* current # inodes in use */
1839 * Global quota index support
1840 * Format of a global record, providing global quota settings for a given quota
1843 struct lquota_glb_rec { /* 32 bytes */
1844 __u64 qbr_hardlimit; /* quota hard limit, in #inodes or kbytes */
1845 __u64 qbr_softlimit; /* quota soft limit, in #inodes or kbytes */
1846 __u64 qbr_time; /* grace time, in seconds */
1847 __u64 qbr_granted; /* how much is granted to slaves, in #inodes or
1852 * Slave index support
1853 * Format of a slave record, recording how much space is granted to a given
1856 struct lquota_slv_rec { /* 8 bytes */
1857 __u64 qsr_granted; /* space granted to the slave for the key=ID,
1858 * in #inodes or kbytes */
1861 /* Data structures associated with the quota locks */
1863 /* Glimpse descriptor used for the index & per-ID quota locks */
1864 struct ldlm_gl_lquota_desc {
1865 union lquota_id gl_id; /* quota ID subject to the glimpse */
1866 __u64 gl_flags; /* see LQUOTA_FL* below */
1867 __u64 gl_ver; /* new index version */
1868 __u64 gl_hardlimit; /* new hardlimit or qunit value */
1869 __u64 gl_softlimit; /* new softlimit */
1873 #define gl_qunit gl_hardlimit /* current qunit value used when
1874 * glimpsing per-ID quota locks */
1876 /* quota glimpse flags */
1877 #define LQUOTA_FL_EDQUOT 0x1 /* user/group out of quota space on QMT */
1879 /* LVB used with quota (global and per-ID) locks */
1881 __u64 lvb_flags; /* see LQUOTA_FL* above */
1882 __u64 lvb_id_may_rel; /* space that might be released later */
1883 __u64 lvb_id_rel; /* space released by the slave for this ID */
1884 __u64 lvb_id_qunit; /* current qunit value */
1888 extern void lustre_swab_lquota_lvb(struct lquota_lvb *lvb);
1890 /* LVB used with global quota lock */
1891 #define lvb_glb_ver lvb_id_may_rel /* current version of the global index */
1899 #define QUOTA_FIRST_OPC QUOTA_DQACQ
1908 MDS_GETATTR_NAME = 34,
1913 MDS_DISCONNECT = 39,
1919 MDS_DONE_WRITING = 45,
1921 MDS_QUOTACHECK = 47,
1924 MDS_SETXATTR = 50, /* obsolete, now it's MDS_REINT op */
1928 MDS_HSM_STATE_GET = 54,
1929 MDS_HSM_STATE_SET = 55,
1930 MDS_HSM_ACTION = 56,
1931 MDS_HSM_PROGRESS = 57,
1932 MDS_HSM_REQUEST = 58,
1933 MDS_HSM_CT_REGISTER = 59,
1934 MDS_HSM_CT_UNREGISTER = 60,
1935 MDS_SWAP_LAYOUTS = 61,
1939 #define MDS_FIRST_OPC MDS_GETATTR
1942 /* opcodes for object update */
1948 #define UPDATE_FIRST_OPC UPDATE_OBJ
1965 } mds_reint_t, mdt_reint_t;
1967 extern void lustre_swab_generic_32s (__u32 *val);
1969 /* the disposition of the intent outlines what was executed */
1970 #define DISP_IT_EXECD 0x00000001
1971 #define DISP_LOOKUP_EXECD 0x00000002
1972 #define DISP_LOOKUP_NEG 0x00000004
1973 #define DISP_LOOKUP_POS 0x00000008
1974 #define DISP_OPEN_CREATE 0x00000010
1975 #define DISP_OPEN_OPEN 0x00000020
1976 #define DISP_ENQ_COMPLETE 0x00400000
1977 #define DISP_ENQ_OPEN_REF 0x00800000
1978 #define DISP_ENQ_CREATE_REF 0x01000000
1979 #define DISP_OPEN_LOCK 0x02000000
1981 /* INODE LOCK PARTS */
1982 #define MDS_INODELOCK_LOOKUP 0x000001 /* dentry, mode, owner, group */
1983 #define MDS_INODELOCK_UPDATE 0x000002 /* size, links, timestamps */
1984 #define MDS_INODELOCK_OPEN 0x000004 /* For opened files */
1985 #define MDS_INODELOCK_LAYOUT 0x000008 /* for layout */
1986 #define MDS_INODELOCK_PERM 0x000010 /* for permission */
1988 #define MDS_INODELOCK_MAXSHIFT 4
1989 /* This FULL lock is useful to take on unlink sort of operations */
1990 #define MDS_INODELOCK_FULL ((1<<(MDS_INODELOCK_MAXSHIFT+1))-1)
1992 extern void lustre_swab_ll_fid (struct ll_fid *fid);
1994 /* NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
1995 * but was moved into name[1] along with the OID to avoid consuming the
1996 * name[2,3] fields that need to be used for the quota id (also a FID). */
1998 LUSTRE_RES_ID_SEQ_OFF = 0,
1999 LUSTRE_RES_ID_VER_OID_OFF = 1,
2000 LUSTRE_RES_ID_WAS_VER_OFF = 2, /* see note above */
2001 LUSTRE_RES_ID_QUOTA_SEQ_OFF = 2,
2002 LUSTRE_RES_ID_QUOTA_VER_OID_OFF = 3,
2003 LUSTRE_RES_ID_HSH_OFF = 3
2006 #define MDS_STATUS_CONN 1
2007 #define MDS_STATUS_LOV 2
2009 /* mdt_thread_info.mti_flags. */
2011 /* The flag indicates Size-on-MDS attributes are changed. */
2012 MF_SOM_CHANGE = (1 << 0),
2013 /* Flags indicates an epoch opens or closes. */
2014 MF_EPOCH_OPEN = (1 << 1),
2015 MF_EPOCH_CLOSE = (1 << 2),
2016 MF_MDC_CANCEL_FID1 = (1 << 3),
2017 MF_MDC_CANCEL_FID2 = (1 << 4),
2018 MF_MDC_CANCEL_FID3 = (1 << 5),
2019 MF_MDC_CANCEL_FID4 = (1 << 6),
2020 /* There is a pending attribute update. */
2021 MF_SOM_AU = (1 << 7),
2022 /* Cancel OST locks while getattr OST attributes. */
2023 MF_GETATTR_LOCK = (1 << 8),
2024 MF_GET_MDT_IDX = (1 << 9),
2027 #define MF_SOM_LOCAL_FLAGS (MF_SOM_CHANGE | MF_EPOCH_OPEN | MF_EPOCH_CLOSE)
2029 #define LUSTRE_BFLAG_UNCOMMITTED_WRITES 0x1
2031 /* these should be identical to their EXT4_*_FL counterparts, they are
2032 * redefined here only to avoid dragging in fs/ext4/ext4.h */
2033 #define LUSTRE_SYNC_FL 0x00000008 /* Synchronous updates */
2034 #define LUSTRE_IMMUTABLE_FL 0x00000010 /* Immutable file */
2035 #define LUSTRE_APPEND_FL 0x00000020 /* writes to file may only append */
2036 #define LUSTRE_NOATIME_FL 0x00000080 /* do not update atime */
2037 #define LUSTRE_DIRSYNC_FL 0x00010000 /* dirsync behaviour (dir only) */
2040 /* Convert wire LUSTRE_*_FL to corresponding client local VFS S_* values
2041 * for the client inode i_flags. The LUSTRE_*_FL are the Lustre wire
2042 * protocol equivalents of LDISKFS_*_FL values stored on disk, while
2043 * the S_* flags are kernel-internal values that change between kernel
2044 * versions. These flags are set/cleared via FSFILT_IOC_{GET,SET}_FLAGS.
2045 * See b=16526 for a full history. */
2046 static inline int ll_ext_to_inode_flags(int flags)
2048 return (((flags & LUSTRE_SYNC_FL) ? S_SYNC : 0) |
2049 ((flags & LUSTRE_NOATIME_FL) ? S_NOATIME : 0) |
2050 ((flags & LUSTRE_APPEND_FL) ? S_APPEND : 0) |
2051 #if defined(S_DIRSYNC)
2052 ((flags & LUSTRE_DIRSYNC_FL) ? S_DIRSYNC : 0) |
2054 ((flags & LUSTRE_IMMUTABLE_FL) ? S_IMMUTABLE : 0));
2057 static inline int ll_inode_to_ext_flags(int iflags)
2059 return (((iflags & S_SYNC) ? LUSTRE_SYNC_FL : 0) |
2060 ((iflags & S_NOATIME) ? LUSTRE_NOATIME_FL : 0) |
2061 ((iflags & S_APPEND) ? LUSTRE_APPEND_FL : 0) |
2062 #if defined(S_DIRSYNC)
2063 ((iflags & S_DIRSYNC) ? LUSTRE_DIRSYNC_FL : 0) |
2065 ((iflags & S_IMMUTABLE) ? LUSTRE_IMMUTABLE_FL : 0));
2072 struct lustre_handle handle;
2074 __u64 size; /* Offset, in the case of MDS_READPAGE */
2078 __u64 blocks; /* XID, in the case of MDS_READPAGE */
2080 __u64 unused1; /* was "ino" until 2.4.0 */
2087 __u32 flags; /* from vfs for pin/unpin, LUSTRE_BFLAG close */
2089 __u32 nlink; /* #bytes to read in the case of MDS_READPAGE */
2090 __u32 unused2; /* was "generation" until 2.4.0 */
2095 __u32 max_cookiesize;
2096 __u32 uid_h; /* high 32-bits of uid, for FUID */
2097 __u32 gid_h; /* high 32-bits of gid, for FUID */
2098 __u32 padding_5; /* also fix lustre_swab_mdt_body */
2106 extern void lustre_swab_mdt_body (struct mdt_body *b);
2108 struct mdt_ioepoch {
2109 struct lustre_handle handle;
2115 extern void lustre_swab_mdt_ioepoch (struct mdt_ioepoch *b);
2117 /* permissions for md_perm.mp_perm */
2119 CFS_SETUID_PERM = 0x01,
2120 CFS_SETGID_PERM = 0x02,
2121 CFS_SETGRP_PERM = 0x04,
2122 CFS_RMTACL_PERM = 0x08,
2123 CFS_RMTOWN_PERM = 0x10
2126 /* inode access permission for remote user, the inode info are omitted,
2127 * for client knows them. */
2128 struct mdt_remote_perm {
2135 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
2139 extern void lustre_swab_mdt_remote_perm(struct mdt_remote_perm *p);
2141 struct mdt_rec_setattr {
2151 __u32 sa_padding_1_h;
2152 struct lu_fid sa_fid;
2161 __u32 sa_attr_flags;
2163 __u32 sa_bias; /* some operation flags */
2169 extern void lustre_swab_mdt_rec_setattr (struct mdt_rec_setattr *sa);
2172 * Attribute flags used in mdt_rec_setattr::sa_valid.
2173 * The kernel's #defines for ATTR_* should not be used over the network
2174 * since the client and MDS may run different kernels (see bug 13828)
2175 * Therefore, we should only use MDS_ATTR_* attributes for sa_valid.
2177 #define MDS_ATTR_MODE 0x1ULL /* = 1 */
2178 #define MDS_ATTR_UID 0x2ULL /* = 2 */
2179 #define MDS_ATTR_GID 0x4ULL /* = 4 */
2180 #define MDS_ATTR_SIZE 0x8ULL /* = 8 */
2181 #define MDS_ATTR_ATIME 0x10ULL /* = 16 */
2182 #define MDS_ATTR_MTIME 0x20ULL /* = 32 */
2183 #define MDS_ATTR_CTIME 0x40ULL /* = 64 */
2184 #define MDS_ATTR_ATIME_SET 0x80ULL /* = 128 */
2185 #define MDS_ATTR_MTIME_SET 0x100ULL /* = 256 */
2186 #define MDS_ATTR_FORCE 0x200ULL /* = 512, Not a change, but a change it */
2187 #define MDS_ATTR_ATTR_FLAG 0x400ULL /* = 1024 */
2188 #define MDS_ATTR_KILL_SUID 0x800ULL /* = 2048 */
2189 #define MDS_ATTR_KILL_SGID 0x1000ULL /* = 4096 */
2190 #define MDS_ATTR_CTIME_SET 0x2000ULL /* = 8192 */
2191 #define MDS_ATTR_FROM_OPEN 0x4000ULL /* = 16384, called from open path, ie O_TRUNC */
2192 #define MDS_ATTR_BLOCKS 0x8000ULL /* = 32768 */
2195 #define FMODE_READ 00000001
2196 #define FMODE_WRITE 00000002
2199 #define MDS_FMODE_CLOSED 00000000
2200 #define MDS_FMODE_EXEC 00000004
2201 /* IO Epoch is opened on a closed file. */
2202 #define MDS_FMODE_EPOCH 01000000
2203 /* IO Epoch is opened on a file truncate. */
2204 #define MDS_FMODE_TRUNC 02000000
2205 /* Size-on-MDS Attribute Update is pending. */
2206 #define MDS_FMODE_SOM 04000000
2208 #define MDS_OPEN_CREATED 00000010
2209 #define MDS_OPEN_CROSS 00000020
2211 #define MDS_OPEN_CREAT 00000100
2212 #define MDS_OPEN_EXCL 00000200
2213 #define MDS_OPEN_TRUNC 00001000
2214 #define MDS_OPEN_APPEND 00002000
2215 #define MDS_OPEN_SYNC 00010000
2216 #define MDS_OPEN_DIRECTORY 00200000
2218 #define MDS_OPEN_BY_FID 040000000 /* open_by_fid for known object */
2219 #define MDS_OPEN_DELAY_CREATE 0100000000 /* delay initial object create */
2220 #define MDS_OPEN_OWNEROVERRIDE 0200000000 /* NFSD rw-reopen ro file for owner */
2221 #define MDS_OPEN_JOIN_FILE 0400000000 /* open for join file.
2222 * We do not support JOIN FILE
2223 * anymore, reserve this flags
2224 * just for preventing such bit
2227 #define MDS_OPEN_LOCK 04000000000 /* This open requires open lock */
2228 #define MDS_OPEN_HAS_EA 010000000000 /* specify object create pattern */
2229 #define MDS_OPEN_HAS_OBJS 020000000000 /* Just set the EA the obj exist */
2230 #define MDS_OPEN_NORESTORE 0100000000000ULL /* Do not restore file at open */
2231 #define MDS_OPEN_NEWSTRIPE 0200000000000ULL /* New stripe needed (restripe or
2233 #define MDS_OPEN_VOLATILE 0400000000000ULL /* File is volatile = created
2236 /* permission for create non-directory file */
2237 #define MAY_CREATE (1 << 7)
2238 /* permission for create directory file */
2239 #define MAY_LINK (1 << 8)
2240 /* permission for delete from the directory */
2241 #define MAY_UNLINK (1 << 9)
2242 /* source's permission for rename */
2243 #define MAY_RENAME_SRC (1 << 10)
2244 /* target's permission for rename */
2245 #define MAY_RENAME_TAR (1 << 11)
2246 /* part (parent's) VTX permission check */
2247 #define MAY_VTX_PART (1 << 12)
2248 /* full VTX permission check */
2249 #define MAY_VTX_FULL (1 << 13)
2250 /* lfs rgetfacl permission check */
2251 #define MAY_RGETFACL (1 << 14)
2254 MDS_CHECK_SPLIT = 1 << 0,
2255 MDS_CROSS_REF = 1 << 1,
2256 MDS_VTX_BYPASS = 1 << 2,
2257 MDS_PERM_BYPASS = 1 << 3,
2259 MDS_QUOTA_IGNORE = 1 << 5,
2260 MDS_CLOSE_CLEANUP = 1 << 6,
2261 MDS_KEEP_ORPHAN = 1 << 7,
2262 MDS_RECOV_OPEN = 1 << 8,
2263 MDS_DATA_MODIFIED = 1 << 9,
2264 MDS_CREATE_VOLATILE = 1 << 10,
2265 MDS_OWNEROVERRIDE = 1 << 11,
2268 /* instance of mdt_reint_rec */
2269 struct mdt_rec_create {
2277 __u32 cr_suppgid1_h;
2279 __u32 cr_suppgid2_h;
2280 struct lu_fid cr_fid1;
2281 struct lu_fid cr_fid2;
2282 struct lustre_handle cr_old_handle; /* handle in case of open replay */
2286 __u64 cr_padding_1; /* rr_blocks */
2289 /* use of helpers set/get_mrc_cr_flags() is needed to access
2290 * 64 bits cr_flags [cr_flags_l, cr_flags_h], this is done to
2291 * extend cr_flags size without breaking 1.8 compat */
2292 __u32 cr_flags_l; /* for use with open, low 32 bits */
2293 __u32 cr_flags_h; /* for use with open, high 32 bits */
2294 __u32 cr_umask; /* umask for create */
2295 __u32 cr_padding_4; /* rr_padding_4 */
2298 static inline void set_mrc_cr_flags(struct mdt_rec_create *mrc, __u64 flags)
2300 mrc->cr_flags_l = (__u32)(flags & 0xFFFFFFFFUll);
2301 mrc->cr_flags_h = (__u32)(flags >> 32);
2304 static inline __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
2306 return ((__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32));
2309 /* instance of mdt_reint_rec */
2310 struct mdt_rec_link {
2318 __u32 lk_suppgid1_h;
2320 __u32 lk_suppgid2_h;
2321 struct lu_fid lk_fid1;
2322 struct lu_fid lk_fid2;
2324 __u64 lk_padding_1; /* rr_atime */
2325 __u64 lk_padding_2; /* rr_ctime */
2326 __u64 lk_padding_3; /* rr_size */
2327 __u64 lk_padding_4; /* rr_blocks */
2329 __u32 lk_padding_5; /* rr_mode */
2330 __u32 lk_padding_6; /* rr_flags */
2331 __u32 lk_padding_7; /* rr_padding_2 */
2332 __u32 lk_padding_8; /* rr_padding_3 */
2333 __u32 lk_padding_9; /* rr_padding_4 */
2336 /* instance of mdt_reint_rec */
2337 struct mdt_rec_unlink {
2345 __u32 ul_suppgid1_h;
2347 __u32 ul_suppgid2_h;
2348 struct lu_fid ul_fid1;
2349 struct lu_fid ul_fid2;
2351 __u64 ul_padding_2; /* rr_atime */
2352 __u64 ul_padding_3; /* rr_ctime */
2353 __u64 ul_padding_4; /* rr_size */
2354 __u64 ul_padding_5; /* rr_blocks */
2357 __u32 ul_padding_6; /* rr_flags */
2358 __u32 ul_padding_7; /* rr_padding_2 */
2359 __u32 ul_padding_8; /* rr_padding_3 */
2360 __u32 ul_padding_9; /* rr_padding_4 */
2363 /* instance of mdt_reint_rec */
2364 struct mdt_rec_rename {
2372 __u32 rn_suppgid1_h;
2374 __u32 rn_suppgid2_h;
2375 struct lu_fid rn_fid1;
2376 struct lu_fid rn_fid2;
2378 __u64 rn_padding_1; /* rr_atime */
2379 __u64 rn_padding_2; /* rr_ctime */
2380 __u64 rn_padding_3; /* rr_size */
2381 __u64 rn_padding_4; /* rr_blocks */
2382 __u32 rn_bias; /* some operation flags */
2383 __u32 rn_mode; /* cross-ref rename has mode */
2384 __u32 rn_padding_5; /* rr_flags */
2385 __u32 rn_padding_6; /* rr_padding_2 */
2386 __u32 rn_padding_7; /* rr_padding_3 */
2387 __u32 rn_padding_8; /* rr_padding_4 */
2390 /* instance of mdt_reint_rec */
2391 struct mdt_rec_setxattr {
2399 __u32 sx_suppgid1_h;
2401 __u32 sx_suppgid2_h;
2402 struct lu_fid sx_fid;
2403 __u64 sx_padding_1; /* These three are rr_fid2 */
2408 __u64 sx_padding_5; /* rr_ctime */
2409 __u64 sx_padding_6; /* rr_size */
2410 __u64 sx_padding_7; /* rr_blocks */
2413 __u32 sx_padding_8; /* rr_flags */
2414 __u32 sx_padding_9; /* rr_padding_2 */
2415 __u32 sx_padding_10; /* rr_padding_3 */
2416 __u32 sx_padding_11; /* rr_padding_4 */
2420 * mdt_rec_reint is the template for all mdt_reint_xxx structures.
2421 * Do NOT change the size of various members, otherwise the value
2422 * will be broken in lustre_swab_mdt_rec_reint().
2424 * If you add new members in other mdt_reint_xxx structres and need to use the
2425 * rr_padding_x fields, then update lustre_swab_mdt_rec_reint() also.
2427 struct mdt_rec_reint {
2435 __u32 rr_suppgid1_h;
2437 __u32 rr_suppgid2_h;
2438 struct lu_fid rr_fid1;
2439 struct lu_fid rr_fid2;
2448 __u32 rr_padding_2; /* also fix lustre_swab_mdt_rec_reint */
2449 __u32 rr_padding_3; /* also fix lustre_swab_mdt_rec_reint */
2450 __u32 rr_padding_4; /* also fix lustre_swab_mdt_rec_reint */
2453 extern void lustre_swab_mdt_rec_reint(struct mdt_rec_reint *rr);
2456 __u32 ld_tgt_count; /* how many MDS's */
2457 __u32 ld_active_tgt_count; /* how many active */
2458 __u32 ld_default_stripe_count; /* how many objects are used */
2459 __u32 ld_pattern; /* default MEA_MAGIC_* */
2460 __u64 ld_default_hash_size;
2461 __u64 ld_padding_1; /* also fix lustre_swab_lmv_desc */
2462 __u32 ld_padding_2; /* also fix lustre_swab_lmv_desc */
2463 __u32 ld_qos_maxage; /* in second */
2464 __u32 ld_padding_3; /* also fix lustre_swab_lmv_desc */
2465 __u32 ld_padding_4; /* also fix lustre_swab_lmv_desc */
2466 struct obd_uuid ld_uuid;
2469 extern void lustre_swab_lmv_desc (struct lmv_desc *ld);
2471 /* TODO: lmv_stripe_md should contain mds capabilities for all slave fids */
2472 struct lmv_stripe_md {
2477 char mea_pool_name[LOV_MAXPOOLNAME];
2478 struct lu_fid mea_ids[0];
2481 extern void lustre_swab_lmv_stripe_md(struct lmv_stripe_md *mea);
2483 /* lmv structures */
2484 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
2485 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
2486 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
2488 #define MAX_HASH_SIZE_32 0x7fffffffUL
2489 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
2490 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
2495 FLD_FIRST_OPC = FLD_QUERY
2501 SEQ_FIRST_OPC = SEQ_QUERY
2505 SEQ_ALLOC_SUPER = 0,
2510 * LOV data structures
2513 #define LOV_MAX_UUID_BUFFER_SIZE 8192
2514 /* The size of the buffer the lov/mdc reserves for the
2515 * array of UUIDs returned by the MDS. With the current
2516 * protocol, this will limit the max number of OSTs per LOV */
2518 #define LOV_DESC_MAGIC 0xB0CCDE5C
2520 /* LOV settings descriptor (should only contain static info) */
2522 __u32 ld_tgt_count; /* how many OBD's */
2523 __u32 ld_active_tgt_count; /* how many active */
2524 __u32 ld_default_stripe_count; /* how many objects are used */
2525 __u32 ld_pattern; /* default PATTERN_RAID0 */
2526 __u64 ld_default_stripe_size; /* in bytes */
2527 __u64 ld_default_stripe_offset; /* in bytes */
2528 __u32 ld_padding_0; /* unused */
2529 __u32 ld_qos_maxage; /* in second */
2530 __u32 ld_padding_1; /* also fix lustre_swab_lov_desc */
2531 __u32 ld_padding_2; /* also fix lustre_swab_lov_desc */
2532 struct obd_uuid ld_uuid;
2535 #define ld_magic ld_active_tgt_count /* for swabbing from llogs */
2537 extern void lustre_swab_lov_desc (struct lov_desc *ld);
2542 /* opcodes -- MUST be distinct from OST/MDS opcodes */
2547 LDLM_BL_CALLBACK = 104,
2548 LDLM_CP_CALLBACK = 105,
2549 LDLM_GL_CALLBACK = 106,
2550 LDLM_SET_INFO = 107,
2553 #define LDLM_FIRST_OPC LDLM_ENQUEUE
2555 #define RES_NAME_SIZE 4
2556 struct ldlm_res_id {
2557 __u64 name[RES_NAME_SIZE];
2560 extern void lustre_swab_ldlm_res_id (struct ldlm_res_id *id);
2562 static inline int ldlm_res_eq(const struct ldlm_res_id *res0,
2563 const struct ldlm_res_id *res1)
2565 return !memcmp(res0, res1, sizeof(*res0));
2582 #define LCK_MODE_NUM 8
2592 #define LDLM_MIN_TYPE LDLM_PLAIN
2594 struct ldlm_extent {
2600 static inline int ldlm_extent_overlap(struct ldlm_extent *ex1,
2601 struct ldlm_extent *ex2)
2603 return (ex1->start <= ex2->end) && (ex2->start <= ex1->end);
2606 /* check if @ex1 contains @ex2 */
2607 static inline int ldlm_extent_contain(struct ldlm_extent *ex1,
2608 struct ldlm_extent *ex2)
2610 return (ex1->start <= ex2->start) && (ex1->end >= ex2->end);
2613 struct ldlm_inodebits {
2617 struct ldlm_flock_wire {
2625 /* it's important that the fields of the ldlm_extent structure match
2626 * the first fields of the ldlm_flock structure because there is only
2627 * one ldlm_swab routine to process the ldlm_policy_data_t union. if
2628 * this ever changes we will need to swab the union differently based
2629 * on the resource type. */
2632 struct ldlm_extent l_extent;
2633 struct ldlm_flock_wire l_flock;
2634 struct ldlm_inodebits l_inodebits;
2635 } ldlm_wire_policy_data_t;
2637 extern void lustre_swab_ldlm_policy_data (ldlm_wire_policy_data_t *d);
2639 union ldlm_gl_desc {
2640 struct ldlm_gl_lquota_desc lquota_desc;
2643 extern void lustre_swab_gl_desc(union ldlm_gl_desc *);
2645 struct ldlm_intent {
2649 extern void lustre_swab_ldlm_intent (struct ldlm_intent *i);
2651 struct ldlm_resource_desc {
2652 ldlm_type_t lr_type;
2653 __u32 lr_padding; /* also fix lustre_swab_ldlm_resource_desc */
2654 struct ldlm_res_id lr_name;
2657 extern void lustre_swab_ldlm_resource_desc (struct ldlm_resource_desc *r);
2659 struct ldlm_lock_desc {
2660 struct ldlm_resource_desc l_resource;
2661 ldlm_mode_t l_req_mode;
2662 ldlm_mode_t l_granted_mode;
2663 ldlm_wire_policy_data_t l_policy_data;
2666 extern void lustre_swab_ldlm_lock_desc (struct ldlm_lock_desc *l);
2668 #define LDLM_LOCKREQ_HANDLES 2
2669 #define LDLM_ENQUEUE_CANCEL_OFF 1
2671 struct ldlm_request {
2674 struct ldlm_lock_desc lock_desc;
2675 struct lustre_handle lock_handle[LDLM_LOCKREQ_HANDLES];
2678 extern void lustre_swab_ldlm_request (struct ldlm_request *rq);
2680 /* If LDLM_ENQUEUE, 1 slot is already occupied, 1 is available.
2681 * Otherwise, 2 are available. */
2682 #define ldlm_request_bufsize(count,type) \
2684 int _avail = LDLM_LOCKREQ_HANDLES; \
2685 _avail -= (type == LDLM_ENQUEUE ? LDLM_ENQUEUE_CANCEL_OFF : 0); \
2686 sizeof(struct ldlm_request) + \
2687 (count > _avail ? count - _avail : 0) * \
2688 sizeof(struct lustre_handle); \
2693 __u32 lock_padding; /* also fix lustre_swab_ldlm_reply */
2694 struct ldlm_lock_desc lock_desc;
2695 struct lustre_handle lock_handle;
2696 __u64 lock_policy_res1;
2697 __u64 lock_policy_res2;
2700 extern void lustre_swab_ldlm_reply (struct ldlm_reply *r);
2702 #define ldlm_flags_to_wire(flags) ((__u32)(flags))
2703 #define ldlm_flags_from_wire(flags) ((__u64)(flags))
2706 * Opcodes for mountconf (mgs and mgc)
2711 MGS_EXCEPTION, /* node died, etc. */
2712 MGS_TARGET_REG, /* whenever target starts up */
2718 #define MGS_FIRST_OPC MGS_CONNECT
2720 #define MGS_PARAM_MAXLEN 1024
2721 #define KEY_SET_INFO "set_info"
2723 struct mgs_send_param {
2724 char mgs_param[MGS_PARAM_MAXLEN];
2727 /* We pass this info to the MGS so it can write config logs */
2728 #define MTI_NAME_MAXLEN 64
2729 #define MTI_PARAM_MAXLEN 4096
2730 #define MTI_NIDS_MAX 32
2731 struct mgs_target_info {
2732 __u32 mti_lustre_ver;
2733 __u32 mti_stripe_index;
2734 __u32 mti_config_ver;
2736 __u32 mti_nid_count;
2737 __u32 mti_instance; /* Running instance of target */
2738 char mti_fsname[MTI_NAME_MAXLEN];
2739 char mti_svname[MTI_NAME_MAXLEN];
2740 char mti_uuid[sizeof(struct obd_uuid)];
2741 __u64 mti_nids[MTI_NIDS_MAX]; /* host nids (lnet_nid_t)*/
2742 char mti_params[MTI_PARAM_MAXLEN];
2744 extern void lustre_swab_mgs_target_info(struct mgs_target_info *oinfo);
2746 struct mgs_nidtbl_entry {
2747 __u64 mne_version; /* table version of this entry */
2748 __u32 mne_instance; /* target instance # */
2749 __u32 mne_index; /* target index */
2750 __u32 mne_length; /* length of this entry - by bytes */
2751 __u8 mne_type; /* target type LDD_F_SV_TYPE_OST/MDT */
2752 __u8 mne_nid_type; /* type of nid(mbz). for ipv6. */
2753 __u8 mne_nid_size; /* size of each NID, by bytes */
2754 __u8 mne_nid_count; /* # of NIDs in buffer */
2756 lnet_nid_t nids[0]; /* variable size buffer for NIDs. */
2759 extern void lustre_swab_mgs_nidtbl_entry(struct mgs_nidtbl_entry *oinfo);
2761 struct mgs_config_body {
2762 char mcb_name[MTI_NAME_MAXLEN]; /* logname */
2763 __u64 mcb_offset; /* next index of config log to request */
2764 __u16 mcb_type; /* type of log: CONFIG_T_[CONFIG|RECOVER] */
2766 __u8 mcb_bits; /* bits unit size of config log */
2767 __u32 mcb_units; /* # of units for bulk transfer */
2769 extern void lustre_swab_mgs_config_body(struct mgs_config_body *body);
2771 struct mgs_config_res {
2772 __u64 mcr_offset; /* index of last config log */
2773 __u64 mcr_size; /* size of the log */
2775 extern void lustre_swab_mgs_config_res(struct mgs_config_res *body);
2777 /* Config marker flags (in config log) */
2778 #define CM_START 0x01
2780 #define CM_SKIP 0x04
2781 #define CM_UPGRADE146 0x08
2782 #define CM_EXCLUDE 0x10
2783 #define CM_START_SKIP (CM_START | CM_SKIP)
2786 __u32 cm_step; /* aka config version */
2788 __u32 cm_vers; /* lustre release version number */
2789 __u32 cm_padding; /* 64 bit align */
2790 obd_time cm_createtime; /*when this record was first created */
2791 obd_time cm_canceltime; /*when this record is no longer valid*/
2792 char cm_tgtname[MTI_NAME_MAXLEN];
2793 char cm_comment[MTI_NAME_MAXLEN];
2796 extern void lustre_swab_cfg_marker(struct cfg_marker *marker,
2797 int swab, int size);
2800 * Opcodes for multiple servers.
2810 #define OBD_FIRST_OPC OBD_PING
2812 /* catalog of log objects */
2814 /** Identifier for a single log object */
2819 } __attribute__((packed));
2821 /** Records written to the CATALOGS list */
2822 #define CATLIST "CATALOGS"
2824 struct llog_logid lci_logid;
2828 } __attribute__((packed));
2830 /* Log data record types - there is no specific reason that these need to
2831 * be related to the RPC opcodes, but no reason not to (may be handy later?)
2833 #define LLOG_OP_MAGIC 0x10600000
2834 #define LLOG_OP_MASK 0xfff00000
2837 LLOG_PAD_MAGIC = LLOG_OP_MAGIC | 0x00000,
2838 OST_SZ_REC = LLOG_OP_MAGIC | 0x00f00,
2839 /* OST_RAID1_REC = LLOG_OP_MAGIC | 0x01000, never used */
2840 MDS_UNLINK_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) |
2841 REINT_UNLINK, /* obsolete after 2.5.0 */
2842 MDS_UNLINK64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
2844 /* MDS_SETATTR_REC = LLOG_OP_MAGIC | 0x12401, obsolete 1.8.0 */
2845 MDS_SETATTR64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
2847 OBD_CFG_REC = LLOG_OP_MAGIC | 0x20000,
2848 /* PTL_CFG_REC = LLOG_OP_MAGIC | 0x30000, obsolete 1.4.0 */
2849 LLOG_GEN_REC = LLOG_OP_MAGIC | 0x40000,
2850 /* LLOG_JOIN_REC = LLOG_OP_MAGIC | 0x50000, obsolete 1.8.0 */
2851 CHANGELOG_REC = LLOG_OP_MAGIC | 0x60000,
2852 CHANGELOG_USER_REC = LLOG_OP_MAGIC | 0x70000,
2853 LLOG_HDR_MAGIC = LLOG_OP_MAGIC | 0x45539,
2854 LLOG_LOGID_MAGIC = LLOG_OP_MAGIC | 0x4553b,
2857 #define LLOG_REC_HDR_NEEDS_SWABBING(r) \
2858 (((r)->lrh_type & __swab32(LLOG_OP_MASK)) == __swab32(LLOG_OP_MAGIC))
2860 /** Log record header - stored in little endian order.
2861 * Each record must start with this struct, end with a llog_rec_tail,
2862 * and be a multiple of 256 bits in size.
2864 struct llog_rec_hdr {
2871 struct llog_rec_tail {
2876 /* Where data follow just after header */
2877 #define REC_DATA(ptr) \
2878 ((void *)((char *)ptr + sizeof(struct llog_rec_hdr)))
2880 #define REC_DATA_LEN(rec) \
2881 (rec->lrh_len - sizeof(struct llog_rec_hdr) - \
2882 sizeof(struct llog_rec_tail))
2884 struct llog_logid_rec {
2885 struct llog_rec_hdr lid_hdr;
2886 struct llog_logid lid_id;
2890 struct llog_rec_tail lid_tail;
2891 } __attribute__((packed));
2893 struct llog_unlink_rec {
2894 struct llog_rec_hdr lur_hdr;
2897 obd_count lur_count;
2898 struct llog_rec_tail lur_tail;
2899 } __attribute__((packed));
2901 struct llog_unlink64_rec {
2902 struct llog_rec_hdr lur_hdr;
2903 struct lu_fid lur_fid;
2904 obd_count lur_count; /* to destroy the lost precreated */
2908 struct llog_rec_tail lur_tail;
2909 } __attribute__((packed));
2911 struct llog_setattr64_rec {
2912 struct llog_rec_hdr lsr_hdr;
2920 struct llog_rec_tail lsr_tail;
2921 } __attribute__((packed));
2923 struct llog_size_change_rec {
2924 struct llog_rec_hdr lsc_hdr;
2925 struct ll_fid lsc_fid;
2930 struct llog_rec_tail lsc_tail;
2931 } __attribute__((packed));
2933 #define CHANGELOG_MAGIC 0xca103000
2935 /** \a changelog_rec_type's that can't be masked */
2936 #define CHANGELOG_MINMASK (1 << CL_MARK)
2937 /** bits covering all \a changelog_rec_type's */
2938 #define CHANGELOG_ALLMASK 0XFFFFFFFF
2939 /** default \a changelog_rec_type mask */
2940 #define CHANGELOG_DEFMASK CHANGELOG_ALLMASK & ~(1 << CL_ATIME | 1 << CL_CLOSE)
2942 /* changelog llog name, needed by client replicators */
2943 #define CHANGELOG_CATALOG "changelog_catalog"
2945 struct changelog_setinfo {
2948 } __attribute__((packed));
2950 /** changelog record */
2951 struct llog_changelog_rec {
2952 struct llog_rec_hdr cr_hdr;
2953 struct changelog_rec cr;
2954 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
2955 } __attribute__((packed));
2957 struct llog_changelog_ext_rec {
2958 struct llog_rec_hdr cr_hdr;
2959 struct changelog_ext_rec cr;
2960 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
2961 } __attribute__((packed));
2963 #define CHANGELOG_USER_PREFIX "cl"
2965 struct llog_changelog_user_rec {
2966 struct llog_rec_hdr cur_hdr;
2970 struct llog_rec_tail cur_tail;
2971 } __attribute__((packed));
2973 /* Old llog gen for compatibility */
2977 } __attribute__((packed));
2979 struct llog_gen_rec {
2980 struct llog_rec_hdr lgr_hdr;
2981 struct llog_gen lgr_gen;
2985 struct llog_rec_tail lgr_tail;
2988 /* On-disk header structure of each log object, stored in little endian order */
2989 #define LLOG_CHUNK_SIZE 8192
2990 #define LLOG_HEADER_SIZE (96)
2991 #define LLOG_BITMAP_BYTES (LLOG_CHUNK_SIZE - LLOG_HEADER_SIZE)
2993 #define LLOG_MIN_REC_SIZE (24) /* round(llog_rec_hdr + llog_rec_tail) */
2995 /* flags for the logs */
2997 LLOG_F_ZAP_WHEN_EMPTY = 0x1,
2998 LLOG_F_IS_CAT = 0x2,
2999 LLOG_F_IS_PLAIN = 0x4,
3002 struct llog_log_hdr {
3003 struct llog_rec_hdr llh_hdr;
3004 obd_time llh_timestamp;
3006 __u32 llh_bitmap_offset;
3010 /* for a catalog the first plain slot is next to it */
3011 struct obd_uuid llh_tgtuuid;
3012 __u32 llh_reserved[LLOG_HEADER_SIZE/sizeof(__u32) - 23];
3013 __u32 llh_bitmap[LLOG_BITMAP_BYTES/sizeof(__u32)];
3014 struct llog_rec_tail llh_tail;
3015 } __attribute__((packed));
3017 #define LLOG_BITMAP_SIZE(llh) (__u32)((llh->llh_hdr.lrh_len - \
3018 llh->llh_bitmap_offset - \
3019 sizeof(llh->llh_tail)) * 8)
3021 /** log cookies are used to reference a specific log file and a record therein */
3022 struct llog_cookie {
3023 struct llog_logid lgc_lgl;
3027 } __attribute__((packed));
3029 /** llog protocol */
3030 enum llogd_rpc_ops {
3031 LLOG_ORIGIN_HANDLE_CREATE = 501,
3032 LLOG_ORIGIN_HANDLE_NEXT_BLOCK = 502,
3033 LLOG_ORIGIN_HANDLE_READ_HEADER = 503,
3034 LLOG_ORIGIN_HANDLE_WRITE_REC = 504,
3035 LLOG_ORIGIN_HANDLE_CLOSE = 505,
3036 LLOG_ORIGIN_CONNECT = 506,
3037 LLOG_CATINFO = 507, /* deprecated */
3038 LLOG_ORIGIN_HANDLE_PREV_BLOCK = 508,
3039 LLOG_ORIGIN_HANDLE_DESTROY = 509, /* for destroy llog object*/
3041 LLOG_FIRST_OPC = LLOG_ORIGIN_HANDLE_CREATE
3045 struct llog_logid lgd_logid;
3047 __u32 lgd_llh_flags;
3049 __u32 lgd_saved_index;
3051 __u64 lgd_cur_offset;
3052 } __attribute__((packed));
3054 struct llogd_conn_body {
3055 struct llog_gen lgdc_gen;
3056 struct llog_logid lgdc_logid;
3057 __u32 lgdc_ctxt_idx;
3058 } __attribute__((packed));
3060 /* Note: 64-bit types are 64-bit aligned in structure */
3062 obd_valid o_valid; /* hot fields in this obdo */
3064 obd_id o_parent_seq;
3065 obd_size o_size; /* o_size-o_blocks == ost_lvb */
3069 obd_blocks o_blocks; /* brw: cli sent cached bytes */
3072 /* 32-bit fields start here: keep an even number of them via padding */
3073 obd_blksize o_blksize; /* optimal IO blocksize */
3074 obd_mode o_mode; /* brw: cli sent cache remain */
3078 obd_count o_nlink; /* brw: checksum */
3079 obd_count o_parent_oid;
3080 obd_count o_misc; /* brw: o_dropped */
3082 __u64 o_ioepoch; /* epoch in ost writes */
3083 __u32 o_stripe_idx; /* holds stripe idx */
3085 struct lustre_handle o_handle; /* brw: lock handle to prolong
3087 struct llog_cookie o_lcookie; /* destroy: unlink cookie from
3092 __u64 o_data_version; /* getattr: sum of iversion for
3094 * brw: grant space consumed on
3095 * the client for the write */
3101 #define o_id o_oi.oi_id
3102 #define o_seq o_oi.oi_seq
3103 #define o_dirty o_blocks
3104 #define o_undirty o_mode
3105 #define o_dropped o_misc
3106 #define o_cksum o_nlink
3107 #define o_grant_used o_data_version
3109 static inline void lustre_set_wire_obdo(struct obdo *wobdo, struct obdo *lobdo)
3111 memcpy(wobdo, lobdo, sizeof(*lobdo));
3112 wobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
3115 static inline void lustre_get_wire_obdo(struct obdo *lobdo, struct obdo *wobdo)
3117 obd_flag local_flags = 0;
3119 if (lobdo->o_valid & OBD_MD_FLFLAGS)
3120 local_flags = lobdo->o_flags & OBD_FL_LOCAL_MASK;
3122 LASSERT(!(wobdo->o_flags & OBD_FL_LOCAL_MASK));
3124 memcpy(lobdo, wobdo, sizeof(*lobdo));
3125 if (local_flags != 0) {
3126 lobdo->o_valid |= OBD_MD_FLFLAGS;
3127 lobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
3128 lobdo->o_flags |= local_flags;
3132 extern void lustre_swab_obdo (struct obdo *o);
3134 /* request structure for OST's */
3139 /* Key for FIEMAP to be used in get_info calls */
3140 struct ll_fiemap_info_key {
3143 struct ll_user_fiemap fiemap;
3146 extern void lustre_swab_ost_body (struct ost_body *b);
3147 extern void lustre_swab_ost_last_id(obd_id *id);
3148 extern void lustre_swab_fiemap(struct ll_user_fiemap *fiemap);
3150 extern void lustre_swab_lov_user_md_v1(struct lov_user_md_v1 *lum);
3151 extern void lustre_swab_lov_user_md_v3(struct lov_user_md_v3 *lum);
3152 extern void lustre_swab_lov_user_md_objects(struct lov_user_ost_data *lod,
3154 extern void lustre_swab_lov_mds_md(struct lov_mds_md *lmm);
3157 extern void lustre_swab_llogd_body (struct llogd_body *d);
3158 extern void lustre_swab_llog_hdr (struct llog_log_hdr *h);
3159 extern void lustre_swab_llogd_conn_body (struct llogd_conn_body *d);
3160 extern void lustre_swab_llog_rec(struct llog_rec_hdr *rec);
3163 extern void lustre_swab_lustre_cfg(struct lustre_cfg *lcfg);
3165 /* Functions for dumping PTLRPC fields */
3166 void dump_rniobuf(struct niobuf_remote *rnb);
3167 void dump_ioo(struct obd_ioobj *nb);
3168 void dump_obdo(struct obdo *oa);
3169 void dump_ost_body(struct ost_body *ob);
3170 void dump_rcs(__u32 *rc);
3172 #define IDX_INFO_MAGIC 0x3D37CC37
3174 /* Index file transfer through the network. The server serializes the index into
3175 * a byte stream which is sent to the client via a bulk transfer */
3179 /* reply: see idx_info_flags below */
3182 /* request & reply: number of lu_idxpage (to be) transferred */
3186 /* request: requested attributes passed down to the iterator API */
3189 /* request & reply: index file identifier (FID) */
3190 struct lu_fid ii_fid;
3192 /* reply: version of the index file before starting to walk the index.
3193 * Please note that the version can be modified at any time during the
3197 /* request: hash to start with:
3198 * reply: hash of the first entry of the first lu_idxpage and hash
3199 * of the entry to read next if any */
3200 __u64 ii_hash_start;
3203 /* reply: size of keys in lu_idxpages, minimal one if II_FL_VARKEY is
3207 /* reply: size of records in lu_idxpages, minimal one if II_FL_VARREC
3215 extern void lustre_swab_idx_info(struct idx_info *ii);
3217 #define II_END_OFF MDS_DIR_END_OFF /* all entries have been read */
3219 /* List of flags used in idx_info::ii_flags */
3220 enum idx_info_flags {
3221 II_FL_NOHASH = 1 << 0, /* client doesn't care about hash value */
3222 II_FL_VARKEY = 1 << 1, /* keys can be of variable size */
3223 II_FL_VARREC = 1 << 2, /* records can be of variable size */
3224 II_FL_NONUNQ = 1 << 3, /* index supports non-unique keys */
3227 #define LIP_MAGIC 0x8A6D6B6C
3229 /* 4KB (= LU_PAGE_SIZE) container gathering key/record pairs */
3231 /* 16-byte header */
3234 __u16 lip_nr; /* number of entries in the container */
3235 __u64 lip_pad0; /* additional padding for future use */
3237 /* key/record pairs are stored in the remaining 4080 bytes.
3238 * depending upon the flags in idx_info::ii_flags, each key/record
3239 * pair might be preceded by:
3241 * - the key size (II_FL_VARKEY is set)
3242 * - the record size (II_FL_VARREC is set)
3244 * For the time being, we only support fixed-size key & record. */
3245 char lip_entries[0];
3247 extern void lustre_swab_lip_header(struct lu_idxpage *lip);
3249 #define LIP_HDR_SIZE (offsetof(struct lu_idxpage, lip_entries))
3251 /* Gather all possible type associated with a 4KB container */
3253 struct lu_dirpage lp_dir; /* for MDS_READPAGE */
3254 struct lu_idxpage lp_idx; /* for OBD_IDX_READ */
3255 char lp_array[LU_PAGE_SIZE];
3258 /* security opcodes */
3261 SEC_CTX_INIT_CONT = 802,
3264 SEC_FIRST_OPC = SEC_CTX_INIT
3268 * capa related definitions
3270 #define CAPA_HMAC_MAX_LEN 64
3271 #define CAPA_HMAC_KEY_MAX_LEN 56
3273 /* NB take care when changing the sequence of elements this struct,
3274 * because the offset info is used in find_capa() */
3275 struct lustre_capa {
3276 struct lu_fid lc_fid; /** fid */
3277 __u64 lc_opc; /** operations allowed */
3278 __u64 lc_uid; /** file owner */
3279 __u64 lc_gid; /** file group */
3280 __u32 lc_flags; /** HMAC algorithm & flags */
3281 __u32 lc_keyid; /** key# used for the capability */
3282 __u32 lc_timeout; /** capa timeout value (sec) */
3283 __u32 lc_expiry; /** expiry time (sec) */
3284 __u8 lc_hmac[CAPA_HMAC_MAX_LEN]; /** HMAC */
3285 } __attribute__((packed));
3287 extern void lustre_swab_lustre_capa(struct lustre_capa *c);
3289 /** lustre_capa::lc_opc */
3291 CAPA_OPC_BODY_WRITE = 1<<0, /**< write object data */
3292 CAPA_OPC_BODY_READ = 1<<1, /**< read object data */
3293 CAPA_OPC_INDEX_LOOKUP = 1<<2, /**< lookup object fid */
3294 CAPA_OPC_INDEX_INSERT = 1<<3, /**< insert object fid */
3295 CAPA_OPC_INDEX_DELETE = 1<<4, /**< delete object fid */
3296 CAPA_OPC_OSS_WRITE = 1<<5, /**< write oss object data */
3297 CAPA_OPC_OSS_READ = 1<<6, /**< read oss object data */
3298 CAPA_OPC_OSS_TRUNC = 1<<7, /**< truncate oss object */
3299 CAPA_OPC_OSS_DESTROY = 1<<8, /**< destroy oss object */
3300 CAPA_OPC_META_WRITE = 1<<9, /**< write object meta data */
3301 CAPA_OPC_META_READ = 1<<10, /**< read object meta data */
3304 #define CAPA_OPC_OSS_RW (CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE)
3305 #define CAPA_OPC_MDS_ONLY \
3306 (CAPA_OPC_BODY_WRITE | CAPA_OPC_BODY_READ | CAPA_OPC_INDEX_LOOKUP | \
3307 CAPA_OPC_INDEX_INSERT | CAPA_OPC_INDEX_DELETE)
3308 #define CAPA_OPC_OSS_ONLY \
3309 (CAPA_OPC_OSS_WRITE | CAPA_OPC_OSS_READ | CAPA_OPC_OSS_TRUNC | \
3310 CAPA_OPC_OSS_DESTROY)
3311 #define CAPA_OPC_MDS_DEFAULT ~CAPA_OPC_OSS_ONLY
3312 #define CAPA_OPC_OSS_DEFAULT ~(CAPA_OPC_MDS_ONLY | CAPA_OPC_OSS_ONLY)
3314 /* MDS capability covers object capability for operations of body r/w
3315 * (dir readpage/sendpage), index lookup/insert/delete and meta data r/w,
3316 * while OSS capability only covers object capability for operations of
3317 * oss data(file content) r/w/truncate.
3319 static inline int capa_for_mds(struct lustre_capa *c)
3321 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) != 0;
3324 static inline int capa_for_oss(struct lustre_capa *c)
3326 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) == 0;
3329 /* lustre_capa::lc_hmac_alg */
3331 CAPA_HMAC_ALG_SHA1 = 1, /**< sha1 algorithm */
3335 #define CAPA_FL_MASK 0x00ffffff
3336 #define CAPA_HMAC_ALG_MASK 0xff000000
3338 struct lustre_capa_key {
3339 __u64 lk_seq; /**< mds# */
3340 __u32 lk_keyid; /**< key# */
3342 __u8 lk_key[CAPA_HMAC_KEY_MAX_LEN]; /**< key */
3343 } __attribute__((packed));
3345 extern void lustre_swab_lustre_capa_key(struct lustre_capa_key *k);
3347 /** The link ea holds 1 \a link_ea_entry for each hardlink */
3348 #define LINK_EA_MAGIC 0x11EAF1DFUL
3349 struct link_ea_header {
3352 __u64 leh_len; /* total size */
3358 /** Hardlink data is name and parent fid.
3359 * Stored in this crazy struct for maximum packing and endian-neutrality
3361 struct link_ea_entry {
3362 /** __u16 stored big-endian, unaligned */
3363 unsigned char lee_reclen[2];
3364 unsigned char lee_parent_fid[sizeof(struct lu_fid)];
3366 }__attribute__((packed));
3368 /** fid2path request/reply structure */
3369 struct getinfo_fid2path {
3370 struct lu_fid gf_fid;
3375 } __attribute__((packed));
3377 void lustre_swab_fid2path (struct getinfo_fid2path *gf);
3380 LAYOUT_INTENT_ACCESS = 0,
3381 LAYOUT_INTENT_READ = 1,
3382 LAYOUT_INTENT_WRITE = 2,
3383 LAYOUT_INTENT_GLIMPSE = 3,
3384 LAYOUT_INTENT_TRUNC = 4,
3385 LAYOUT_INTENT_RELEASE = 5,
3386 LAYOUT_INTENT_RESTORE = 6
3389 /* enqueue layout lock with intent */
3390 struct layout_intent {
3391 __u32 li_opc; /* intent operation for enqueue, read, write etc */
3397 void lustre_swab_layout_intent(struct layout_intent *li);
3400 * On the wire version of hsm_progress structure.
3402 * Contains the userspace hsm_progress and some internal fields.
3404 struct hsm_progress_kernel {
3405 /* Field taken from struct hsm_progress */
3408 struct hsm_extent hpk_extent;
3410 __u16 hpk_errval; /* positive val */
3412 /* Additional fields */
3413 __u64 hpk_data_version;
3415 } __attribute__((packed));
3417 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
3418 extern void lustre_swab_hsm_current_action(struct hsm_current_action *action);
3419 extern void lustre_swab_hsm_progress_kernel(struct hsm_progress_kernel *hpk);
3420 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
3421 extern void lustre_swab_hsm_user_item(struct hsm_user_item *hui);
3422 extern void lustre_swab_hsm_request(struct hsm_request *hr);
3425 * These are object update opcode under UPDATE_OBJ, which is currently
3426 * being used by cross-ref operations between MDT.
3428 * During the cross-ref operation, the Master MDT, which the client send the
3429 * request to, will disassembly the operation into object updates, then OSP
3430 * will send these updates to the remote MDT to be executed.
3432 * Update request format
3433 * magic: UPDATE_BUFFER_MAGIC_V1
3434 * Count: How many updates in the req.
3435 * bufs[0] : following are packets of object.
3437 * type: object_update_op, the op code of update
3438 * fid: The object fid of the update.
3439 * lens/bufs: other parameters of the update.
3441 * type: object_update_op, the op code of update
3442 * fid: The object fid of the update.
3443 * lens/bufs: other parameters of the update.
3445 * update[7]: type: object_update_op, the op code of update
3446 * fid: The object fid of the update.
3447 * lens/bufs: other parameters of the update.
3448 * Current 8 maxim updates per object update request.
3450 *******************************************************************
3451 * update reply format:
3453 * ur_version: UPDATE_REPLY_V1
3454 * ur_count: The count of the reply, which is usually equal
3455 * to the number of updates in the request.
3456 * ur_lens: The reply lengths of each object update.
3458 * replies: 1st update reply [4bytes_ret: other body]
3459 * 2nd update reply [4bytes_ret: other body]
3461 * nth update reply [4bytes_ret: other body]
3463 * For each reply of the update, the format would be
3464 * result(4 bytes):Other stuff
3467 #define UPDATE_MAX_OPS 10
3468 #define UPDATE_BUFFER_MAGIC_V1 0xBDDE0001
3469 #define UPDATE_BUFFER_MAGIC UPDATE_BUFFER_MAGIC_V1
3470 #define UPDATE_BUF_COUNT 8
3471 enum object_update_op {
3480 OBJ_INDEX_LOOKUP = 9,
3481 OBJ_INDEX_INSERT = 10,
3482 OBJ_INDEX_DELETE = 11,
3489 struct lu_fid u_fid;
3490 __u32 u_lens[UPDATE_BUF_COUNT];
3500 #define UPDATE_REPLY_V1 0x00BD0001
3501 struct update_reply {
3507 void lustre_swab_update_buf(struct update_buf *ub);
3508 void lustre_swab_update_reply_buf(struct update_reply *ur);
3510 /** layout swap request structure
3511 * fid1 and fid2 are in mdt_body
3513 struct mdc_swap_layouts {
3517 void lustre_swab_swap_layouts(struct mdc_swap_layouts *msl);