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
191 struct lu_seq_range {
198 #define LU_SEQ_RANGE_MDT 0x0
199 #define LU_SEQ_RANGE_OST 0x1
200 #define LU_SEQ_RANGE_ANY 0x3
202 #define LU_SEQ_RANGE_MASK 0x3
204 static inline unsigned fld_range_type(const struct lu_seq_range *range)
206 return range->lsr_flags & LU_SEQ_RANGE_MASK;
209 static inline int fld_range_is_ost(const struct lu_seq_range *range)
211 return fld_range_type(range) == LU_SEQ_RANGE_OST;
214 static inline int fld_range_is_mdt(const struct lu_seq_range *range)
216 return fld_range_type(range) == LU_SEQ_RANGE_MDT;
220 * This all range is only being used when fld client sends fld query request,
221 * but it does not know whether the seq is MDT or OST, so it will send req
222 * with ALL type, which means either seq type gotten from lookup can be
225 static inline unsigned fld_range_is_any(const struct lu_seq_range *range)
227 return fld_range_type(range) == LU_SEQ_RANGE_ANY;
230 static inline void fld_range_set_type(struct lu_seq_range *range,
233 LASSERT(!(flags & ~LU_SEQ_RANGE_MASK));
234 range->lsr_flags |= flags;
237 static inline void fld_range_set_mdt(struct lu_seq_range *range)
239 fld_range_set_type(range, LU_SEQ_RANGE_MDT);
242 static inline void fld_range_set_ost(struct lu_seq_range *range)
244 fld_range_set_type(range, LU_SEQ_RANGE_OST);
247 static inline void fld_range_set_any(struct lu_seq_range *range)
249 fld_range_set_type(range, LU_SEQ_RANGE_ANY);
253 * returns width of given range \a r
256 static inline __u64 range_space(const struct lu_seq_range *range)
258 return range->lsr_end - range->lsr_start;
262 * initialize range to zero
265 static inline void range_init(struct lu_seq_range *range)
267 range->lsr_start = range->lsr_end = range->lsr_index = 0;
271 * check if given seq id \a s is within given range \a r
274 static inline int range_within(const struct lu_seq_range *range,
277 return s >= range->lsr_start && s < range->lsr_end;
280 static inline int range_is_sane(const struct lu_seq_range *range)
282 return (range->lsr_end >= range->lsr_start);
285 static inline int range_is_zero(const struct lu_seq_range *range)
287 return (range->lsr_start == 0 && range->lsr_end == 0);
290 static inline int range_is_exhausted(const struct lu_seq_range *range)
293 return range_space(range) == 0;
296 /* return 0 if two range have the same location */
297 static inline int range_compare_loc(const struct lu_seq_range *r1,
298 const struct lu_seq_range *r2)
300 return r1->lsr_index != r2->lsr_index ||
301 r1->lsr_flags != r2->lsr_flags;
304 #define DRANGE "[%#16.16"LPF64"x-%#16.16"LPF64"x):%x:%s"
306 #define PRANGE(range) \
307 (range)->lsr_start, \
309 (range)->lsr_index, \
310 fld_range_is_mdt(range) ? "mdt" : "ost"
313 /** \defgroup lu_fid lu_fid
317 * Flags for lustre_mdt_attrs::lma_compat and lustre_mdt_attrs::lma_incompat.
318 * Deprecated since HSM and SOM attributes are now stored in separate on-disk
322 LMAC_HSM = 0x00000001,
323 LMAC_SOM = 0x00000002,
327 * Masks for all features that should be supported by a Lustre version to
328 * access a specific file.
329 * This information is stored in lustre_mdt_attrs::lma_incompat.
332 LMAI_RELEASED = 0x0000001, /* file is released */
333 LMAI_AGENT = 0x00000002, /* agent inode */
334 LMAI_REMOTE_PARENT = 0x00000004, /* the parent of the object
335 is on the remote MDT */
337 #define LMA_INCOMPAT_SUPP (LMAI_AGENT | LMAI_REMOTE_PARENT)
339 extern void lustre_lma_swab(struct lustre_mdt_attrs *lma);
340 extern void lustre_lma_init(struct lustre_mdt_attrs *lma,
341 const struct lu_fid *fid, __u32 incompat);
343 * SOM on-disk attributes stored in a separate xattr.
346 /** Bitfield for supported data in this structure. For future use. */
349 /** Incompat feature list. The supported feature mask is availabe in
350 * SOM_INCOMPAT_SUPP */
353 /** IO Epoch SOM attributes belongs to */
355 /** total file size in objects */
357 /** total fs blocks in objects */
359 /** mds mount id the size is valid for */
362 extern void lustre_som_swab(struct som_attrs *attrs);
364 #define SOM_INCOMPAT_SUPP 0x0
367 * HSM on-disk attributes stored in a separate xattr.
370 /** Bitfield for supported data in this structure. For future use. */
373 /** HSM flags, see hsm_flags enum below */
375 /** backend archive id associated with the file */
377 /** version associated with the last archiving, if any */
380 extern void lustre_hsm_swab(struct hsm_attrs *attrs);
386 /** initial fid id value */
387 LUSTRE_FID_INIT_OID = 1UL
390 /** returns fid object sequence */
391 static inline __u64 fid_seq(const struct lu_fid *fid)
396 /** returns fid object id */
397 static inline __u32 fid_oid(const struct lu_fid *fid)
402 /** returns fid object version */
403 static inline __u32 fid_ver(const struct lu_fid *fid)
408 static inline void fid_zero(struct lu_fid *fid)
410 memset(fid, 0, sizeof(*fid));
413 static inline obd_id fid_ver_oid(const struct lu_fid *fid)
415 return ((__u64)fid_ver(fid) << 32 | fid_oid(fid));
419 * Note that reserved SEQ numbers below 12 will conflict with ldiskfs
420 * inodes in the IGIF namespace, so these reserved SEQ numbers can be
421 * used for other purposes and not risk collisions with existing inodes.
423 * Different FID Format
424 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs#NEW.0
427 FID_SEQ_OST_MDT0 = 0,
430 FID_SEQ_OST_MDT1 = 3,
431 FID_SEQ_OST_MAX = 9, /* Max MDT count before OST_on_FID */
434 FID_SEQ_IGIF_MAX = 0x0ffffffffULL,
435 FID_SEQ_IDIF = 0x100000000ULL,
436 FID_SEQ_IDIF_MAX = 0x1ffffffffULL,
437 /* Normal FID sequence starts from this value, i.e. 1<<33 */
438 FID_SEQ_START = 0x200000000ULL,
439 /* sequence for local pre-defined FIDs listed in local_oid */
440 FID_SEQ_LOCAL_FILE = 0x200000001ULL,
441 FID_SEQ_DOT_LUSTRE = 0x200000002ULL,
442 /* sequence is used for local named objects FIDs generated
443 * by local_object_storage library */
444 FID_SEQ_LOCAL_NAME = 0x200000003ULL,
445 /* Because current FLD will only cache the fid sequence, instead
446 * of oid on the client side, if the FID needs to be exposed to
447 * clients sides, it needs to make sure all of fids under one
448 * sequence will be located in one MDT. */
449 FID_SEQ_SPECIAL = 0x200000004ULL,
450 FID_SEQ_QUOTA = 0x200000005ULL,
451 FID_SEQ_QUOTA_GLB = 0x200000006ULL,
452 FID_SEQ_ROOT = 0x200000007ULL, /* Located on MDT0 */
453 FID_SEQ_NORMAL = 0x200000400ULL,
454 FID_SEQ_LOV_DEFAULT = 0xffffffffffffffffULL
457 #define OBIF_OID_MAX_BITS 32
458 #define OBIF_MAX_OID (1ULL << OBIF_OID_MAX_BITS)
459 #define OBIF_OID_MASK ((1ULL << OBIF_OID_MAX_BITS) - 1)
460 #define IDIF_OID_MAX_BITS 48
461 #define IDIF_MAX_OID (1ULL << IDIF_OID_MAX_BITS)
462 #define IDIF_OID_MASK ((1ULL << IDIF_OID_MAX_BITS) - 1)
464 /** OID for FID_SEQ_SPECIAL */
466 /* Big Filesystem Lock to serialize rename operations */
467 FID_OID_SPECIAL_BFL = 1UL,
470 /** OID for FID_SEQ_DOT_LUSTRE */
471 enum dot_lustre_oid {
472 FID_OID_DOT_LUSTRE = 1UL,
473 FID_OID_DOT_LUSTRE_OBF = 2UL,
476 static inline int fid_seq_is_mdt0(obd_seq seq)
478 return (seq == FID_SEQ_OST_MDT0);
481 static inline int fid_seq_is_mdt(const __u64 seq)
483 return seq == FID_SEQ_OST_MDT0 || seq >= FID_SEQ_NORMAL;
486 static inline int fid_seq_is_echo(obd_seq seq)
488 return (seq == FID_SEQ_ECHO);
491 static inline int fid_is_echo(const struct lu_fid *fid)
493 return fid_seq_is_echo(fid_seq(fid));
496 static inline int fid_seq_is_llog(obd_seq seq)
498 return (seq == FID_SEQ_LLOG);
501 static inline int fid_is_llog(const struct lu_fid *fid)
503 return fid_seq_is_llog(fid_seq(fid));
506 static inline int fid_seq_is_rsvd(const __u64 seq)
508 return (seq > FID_SEQ_OST_MDT0 && seq <= FID_SEQ_RSVD);
511 static inline int fid_seq_is_special(const __u64 seq)
513 return seq == FID_SEQ_SPECIAL;
516 static inline int fid_seq_is_local_file(const __u64 seq)
518 return seq == FID_SEQ_LOCAL_FILE ||
519 seq == FID_SEQ_LOCAL_NAME;
522 static inline int fid_seq_is_root(const __u64 seq)
524 return seq == FID_SEQ_ROOT;
527 static inline int fid_seq_is_dot(const __u64 seq)
529 return seq == FID_SEQ_DOT_LUSTRE;
532 static inline int fid_seq_is_default(const __u64 seq)
534 return seq == FID_SEQ_LOV_DEFAULT;
537 static inline int fid_is_mdt0(const struct lu_fid *fid)
539 return fid_seq_is_mdt0(fid_seq(fid));
542 static inline void lu_root_fid(struct lu_fid *fid)
544 fid->f_seq = FID_SEQ_ROOT;
550 * Check if a fid is igif or not.
551 * \param fid the fid to be tested.
552 * \return true if the fid is a igif; otherwise false.
554 static inline int fid_seq_is_igif(const __u64 seq)
556 return seq >= FID_SEQ_IGIF && seq <= FID_SEQ_IGIF_MAX;
559 static inline int fid_is_igif(const struct lu_fid *fid)
561 return fid_seq_is_igif(fid_seq(fid));
565 * Check if a fid is idif or not.
566 * \param fid the fid to be tested.
567 * \return true if the fid is a idif; otherwise false.
569 static inline int fid_seq_is_idif(const __u64 seq)
571 return seq >= FID_SEQ_IDIF && seq <= FID_SEQ_IDIF_MAX;
574 static inline int fid_is_idif(const struct lu_fid *fid)
576 return fid_seq_is_idif(fid_seq(fid));
579 static inline int fid_is_local_file(const struct lu_fid *fid)
581 return fid_seq_is_local_file(fid_seq(fid));
584 static inline int fid_seq_is_norm(const __u64 seq)
586 return (seq >= FID_SEQ_NORMAL);
589 static inline int fid_is_norm(const struct lu_fid *fid)
591 return fid_seq_is_norm(fid_seq(fid));
594 /* convert an OST objid into an IDIF FID SEQ number */
595 static inline obd_seq fid_idif_seq(obd_id id, __u32 ost_idx)
597 return FID_SEQ_IDIF | (ost_idx << 16) | ((id >> 32) & 0xffff);
600 /* convert a packed IDIF FID into an OST objid */
601 static inline obd_id fid_idif_id(obd_seq seq, __u32 oid, __u32 ver)
603 return ((__u64)ver << 48) | ((seq & 0xffff) << 32) | oid;
606 /* extract ost index from IDIF FID */
607 static inline __u32 fid_idif_ost_idx(const struct lu_fid *fid)
609 LASSERT(fid_is_idif(fid));
610 return (fid_seq(fid) >> 16) & 0xffff;
613 /* extract OST sequence (group) from a wire ost_id (id/seq) pair */
614 static inline obd_seq ostid_seq(const struct ost_id *ostid)
616 if (fid_seq_is_mdt0(ostid->oi.oi_seq))
617 return FID_SEQ_OST_MDT0;
619 if (fid_seq_is_default(ostid->oi.oi_seq))
620 return FID_SEQ_LOV_DEFAULT;
622 if (fid_is_idif(&ostid->oi_fid))
623 return FID_SEQ_OST_MDT0;
625 return fid_seq(&ostid->oi_fid);
628 /* extract OST objid from a wire ost_id (id/seq) pair */
629 static inline obd_id ostid_id(const struct ost_id *ostid)
631 if (fid_seq_is_mdt0(ostid_seq(ostid)))
632 return ostid->oi.oi_id & IDIF_OID_MASK;
634 if (fid_is_idif(&ostid->oi_fid))
635 return fid_idif_id(fid_seq(&ostid->oi_fid),
636 fid_oid(&ostid->oi_fid), 0);
638 return fid_oid(&ostid->oi_fid);
641 static inline void ostid_set_seq(struct ost_id *oi, __u64 seq)
643 if (fid_seq_is_mdt0(seq) || fid_seq_is_default(seq)) {
646 oi->oi_fid.f_seq = seq;
647 /* Note: if f_oid + f_ver is zero, we need init it
648 * to be 1, otherwise, ostid_seq will treat this
649 * as old ostid (oi_seq == 0) */
650 if (oi->oi_fid.f_oid == 0 && oi->oi_fid.f_ver == 0)
651 oi->oi_fid.f_oid = LUSTRE_FID_INIT_OID;
655 static inline void ostid_set_seq_mdt0(struct ost_id *oi)
657 ostid_set_seq(oi, FID_SEQ_OST_MDT0);
660 static inline void ostid_set_seq_echo(struct ost_id *oi)
662 ostid_set_seq(oi, FID_SEQ_ECHO);
665 static inline void ostid_set_seq_llog(struct ost_id *oi)
667 ostid_set_seq(oi, FID_SEQ_LLOG);
671 * Note: we need check oi_seq to decide where to set oi_id,
672 * so oi_seq should always be set ahead of oi_id.
674 static inline void ostid_set_id(struct ost_id *oi, __u64 oid)
676 if (fid_seq_is_mdt0(ostid_seq(oi))) {
677 if (oid >= IDIF_MAX_OID) {
678 CERROR("Bad "LPU64" to set "DOSTID"\n",
684 if (oid > OBIF_MAX_OID) {
685 CERROR("Bad "LPU64" to set "DOSTID"\n",
689 oi->oi_fid.f_oid = oid;
693 static inline void ostid_inc_id(struct ost_id *oi)
695 if (fid_seq_is_mdt0(ostid_seq(oi))) {
696 if (unlikely(ostid_id(oi) + 1 > IDIF_MAX_OID)) {
697 CERROR("Bad inc "DOSTID"\n", POSTID(oi));
706 static inline void ostid_dec_id(struct ost_id *oi)
708 if (fid_seq_is_mdt0(ostid_seq(oi)))
715 * Unpack an OST object id/seq (group) into a FID. This is needed for
716 * converting all obdo, lmm, lsm, etc. 64-bit id/seq pairs into proper
717 * FIDs. Note that if an id/seq is already in FID/IDIF format it will
718 * be passed through unchanged. Only legacy OST objects in "group 0"
719 * will be mapped into the IDIF namespace so that they can fit into the
720 * struct lu_fid fields without loss. For reference see:
721 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs
723 static inline int ostid_to_fid(struct lu_fid *fid, struct ost_id *ostid,
726 if (ost_idx > 0xffff) {
727 CERROR("bad ost_idx, "DOSTID" ost_idx:%u\n", POSTID(ostid),
732 if (fid_seq_is_mdt0(ostid_seq(ostid))) {
733 /* This is a "legacy" (old 1.x/2.early) OST object in "group 0"
734 * that we map into the IDIF namespace. It allows up to 2^48
735 * objects per OST, as this is the object namespace that has
736 * been in production for years. This can handle create rates
737 * of 1M objects/s/OST for 9 years, or combinations thereof. */
738 if (ostid_id(ostid) >= IDIF_MAX_OID) {
739 CERROR("bad MDT0 id, "DOSTID" ost_idx:%u\n",
740 POSTID(ostid), ost_idx);
743 fid->f_seq = fid_idif_seq(ostid_id(ostid), ost_idx);
744 /* truncate to 32 bits by assignment */
745 fid->f_oid = ostid_id(ostid);
746 /* in theory, not currently used */
747 fid->f_ver = ostid_id(ostid) >> 48;
748 } else /* if (fid_seq_is_idif(seq) || fid_seq_is_norm(seq)) */ {
749 /* This is either an IDIF object, which identifies objects across
750 * all OSTs, or a regular FID. The IDIF namespace maps legacy
751 * OST objects into the FID namespace. In both cases, we just
752 * pass the FID through, no conversion needed. */
753 if (ostid->oi_fid.f_ver != 0) {
754 CERROR("bad MDT0 id, "DOSTID" ost_idx:%u\n",
755 POSTID(ostid), ost_idx);
758 *fid = ostid->oi_fid;
764 /* pack any OST FID into an ostid (id/seq) for the wire/disk */
765 static inline int fid_to_ostid(const struct lu_fid *fid, struct ost_id *ostid)
767 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
768 CERROR("bad IGIF, "DFID"\n", PFID(fid));
772 if (fid_is_idif(fid)) {
773 ostid_set_seq_mdt0(ostid);
774 ostid_set_id(ostid, fid_idif_id(fid_seq(fid), fid_oid(fid),
777 ostid->oi_fid = *fid;
783 /* Check whether the fid is for LAST_ID */
784 static inline int fid_is_last_id(const struct lu_fid *fid)
786 return (fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid)) &&
791 * Get inode number from a igif.
792 * \param fid a igif to get inode number from.
793 * \return inode number for the igif.
795 static inline ino_t lu_igif_ino(const struct lu_fid *fid)
800 extern void lustre_swab_ost_id(struct ost_id *oid);
803 * Get inode generation from a igif.
804 * \param fid a igif to get inode generation from.
805 * \return inode generation for the igif.
807 static inline __u32 lu_igif_gen(const struct lu_fid *fid)
813 * Build igif from the inode number/generation.
815 static inline void lu_igif_build(struct lu_fid *fid, __u32 ino, __u32 gen)
823 * Fids are transmitted across network (in the sender byte-ordering),
824 * and stored on disk in big-endian order.
826 static inline void fid_cpu_to_le(struct lu_fid *dst, const struct lu_fid *src)
828 /* check that all fields are converted */
829 CLASSERT(sizeof *src ==
830 sizeof fid_seq(src) +
831 sizeof fid_oid(src) + sizeof fid_ver(src));
832 dst->f_seq = cpu_to_le64(fid_seq(src));
833 dst->f_oid = cpu_to_le32(fid_oid(src));
834 dst->f_ver = cpu_to_le32(fid_ver(src));
837 static inline void fid_le_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
839 /* check that all fields are converted */
840 CLASSERT(sizeof *src ==
841 sizeof fid_seq(src) +
842 sizeof fid_oid(src) + sizeof fid_ver(src));
843 dst->f_seq = le64_to_cpu(fid_seq(src));
844 dst->f_oid = le32_to_cpu(fid_oid(src));
845 dst->f_ver = le32_to_cpu(fid_ver(src));
848 static inline void fid_cpu_to_be(struct lu_fid *dst, const struct lu_fid *src)
850 /* check that all fields are converted */
851 CLASSERT(sizeof *src ==
852 sizeof fid_seq(src) +
853 sizeof fid_oid(src) + sizeof fid_ver(src));
854 dst->f_seq = cpu_to_be64(fid_seq(src));
855 dst->f_oid = cpu_to_be32(fid_oid(src));
856 dst->f_ver = cpu_to_be32(fid_ver(src));
859 static inline void fid_be_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
861 /* check that all fields are converted */
862 CLASSERT(sizeof *src ==
863 sizeof fid_seq(src) +
864 sizeof fid_oid(src) + sizeof fid_ver(src));
865 dst->f_seq = be64_to_cpu(fid_seq(src));
866 dst->f_oid = be32_to_cpu(fid_oid(src));
867 dst->f_ver = be32_to_cpu(fid_ver(src));
870 static inline int fid_is_sane(const struct lu_fid *fid)
872 return fid != NULL &&
873 ((fid_seq(fid) >= FID_SEQ_START && fid_ver(fid) == 0) ||
874 fid_is_igif(fid) || fid_is_idif(fid) ||
875 fid_seq_is_rsvd(fid_seq(fid)));
878 static inline int fid_is_zero(const struct lu_fid *fid)
880 return fid_seq(fid) == 0 && fid_oid(fid) == 0;
883 extern void lustre_swab_lu_fid(struct lu_fid *fid);
884 extern void lustre_swab_lu_seq_range(struct lu_seq_range *range);
886 static inline int lu_fid_eq(const struct lu_fid *f0, const struct lu_fid *f1)
888 /* Check that there is no alignment padding. */
889 CLASSERT(sizeof *f0 ==
890 sizeof f0->f_seq + sizeof f0->f_oid + sizeof f0->f_ver);
891 return memcmp(f0, f1, sizeof *f0) == 0;
894 #define __diff_normalize(val0, val1) \
896 typeof(val0) __val0 = (val0); \
897 typeof(val1) __val1 = (val1); \
899 (__val0 == __val1 ? 0 : __val0 > __val1 ? +1 : -1); \
902 static inline int lu_fid_cmp(const struct lu_fid *f0,
903 const struct lu_fid *f1)
906 __diff_normalize(fid_seq(f0), fid_seq(f1)) ?:
907 __diff_normalize(fid_oid(f0), fid_oid(f1)) ?:
908 __diff_normalize(fid_ver(f0), fid_ver(f1));
911 static inline void ostid_cpu_to_le(struct ost_id *src_oi,
912 struct ost_id *dst_oi)
914 if (fid_seq_is_mdt0(ostid_seq(src_oi))) {
915 dst_oi->oi.oi_id = cpu_to_le64(src_oi->oi.oi_id);
916 dst_oi->oi.oi_seq = cpu_to_le64(src_oi->oi.oi_seq);
918 fid_cpu_to_le(&dst_oi->oi_fid, &src_oi->oi_fid);
922 static inline void ostid_le_to_cpu(struct ost_id *src_oi,
923 struct ost_id *dst_oi)
925 if (fid_seq_is_mdt0(ostid_seq(src_oi))) {
926 dst_oi->oi.oi_id = le64_to_cpu(src_oi->oi.oi_id);
927 dst_oi->oi.oi_seq = le64_to_cpu(src_oi->oi.oi_seq);
929 fid_le_to_cpu(&dst_oi->oi_fid, &src_oi->oi_fid);
935 /** \defgroup lu_dir lu_dir
939 * Enumeration of possible directory entry attributes.
941 * Attributes follow directory entry header in the order they appear in this
944 enum lu_dirent_attrs {
947 LUDA_64BITHASH = 0x0004,
949 /* The following attrs are used for MDT interanl only,
950 * not visible to client */
952 /* Verify the dirent consistency */
953 LUDA_VERIFY = 0x8000,
954 /* Only check but not repair the dirent inconsistency */
955 LUDA_VERIFY_DRYRUN = 0x4000,
956 /* The dirent has been repaired, or to be repaired (dryrun). */
957 LUDA_REPAIR = 0x2000,
958 /* The system is upgraded, has beed or to be repaired (dryrun). */
959 LUDA_UPGRADE = 0x1000,
960 /* Ignore this record, go to next directly. */
961 LUDA_IGNORE = 0x0800,
964 #define LU_DIRENT_ATTRS_MASK 0xf800
967 * Layout of readdir pages, as transmitted on wire.
970 /** valid if LUDA_FID is set. */
971 struct lu_fid lde_fid;
972 /** a unique entry identifier: a hash or an offset. */
974 /** total record length, including all attributes. */
978 /** optional variable size attributes following this entry.
979 * taken from enum lu_dirent_attrs.
982 /** name is followed by the attributes indicated in ->ldp_attrs, in
983 * their natural order. After the last attribute, padding bytes are
984 * added to make ->lde_reclen a multiple of 8.
990 * Definitions of optional directory entry attributes formats.
992 * Individual attributes do not have their length encoded in a generic way. It
993 * is assumed that consumer of an attribute knows its format. This means that
994 * it is impossible to skip over an unknown attribute, except by skipping over all
995 * remaining attributes (by using ->lde_reclen), which is not too
996 * constraining, because new server versions will append new attributes at
997 * the end of an entry.
1001 * Fid directory attribute: a fid of an object referenced by the entry. This
1002 * will be almost always requested by the client and supplied by the server.
1004 * Aligned to 8 bytes.
1006 /* To have compatibility with 1.8, lets have fid in lu_dirent struct. */
1011 * Aligned to 2 bytes.
1018 __u64 ldp_hash_start;
1022 struct lu_dirent ldp_entries[0];
1025 enum lu_dirpage_flags {
1027 * dirpage contains no entry.
1031 * last entry's lde_hash equals ldp_hash_end.
1033 LDF_COLLIDE = 1 << 1
1036 static inline struct lu_dirent *lu_dirent_start(struct lu_dirpage *dp)
1038 if (le32_to_cpu(dp->ldp_flags) & LDF_EMPTY)
1041 return dp->ldp_entries;
1044 static inline struct lu_dirent *lu_dirent_next(struct lu_dirent *ent)
1046 struct lu_dirent *next;
1048 if (le16_to_cpu(ent->lde_reclen) != 0)
1049 next = ((void *)ent) + le16_to_cpu(ent->lde_reclen);
1056 static inline int lu_dirent_calc_size(int namelen, __u16 attr)
1060 if (attr & LUDA_TYPE) {
1061 const unsigned align = sizeof(struct luda_type) - 1;
1062 size = (sizeof(struct lu_dirent) + namelen + align) & ~align;
1063 size += sizeof(struct luda_type);
1065 size = sizeof(struct lu_dirent) + namelen;
1067 return (size + 7) & ~7;
1070 static inline int lu_dirent_size(struct lu_dirent *ent)
1072 if (le16_to_cpu(ent->lde_reclen) == 0) {
1073 return lu_dirent_calc_size(le16_to_cpu(ent->lde_namelen),
1074 le32_to_cpu(ent->lde_attrs));
1076 return le16_to_cpu(ent->lde_reclen);
1079 #define MDS_DIR_END_OFF 0xfffffffffffffffeULL
1082 * MDS_READPAGE page size
1084 * This is the directory page size packed in MDS_READPAGE RPC.
1085 * It's different than CFS_PAGE_SIZE because the client needs to
1086 * access the struct lu_dirpage header packed at the beginning of
1087 * the "page" and without this there isn't any way to know find the
1088 * lu_dirpage header is if client and server CFS_PAGE_SIZE differ.
1090 #define LU_PAGE_SHIFT 12
1091 #define LU_PAGE_SIZE (1UL << LU_PAGE_SHIFT)
1092 #define LU_PAGE_MASK (~(LU_PAGE_SIZE - 1))
1094 #define LU_PAGE_COUNT (1 << (CFS_PAGE_SHIFT - LU_PAGE_SHIFT))
1098 struct lustre_handle {
1101 #define DEAD_HANDLE_MAGIC 0xdeadbeefcafebabeULL
1103 static inline int lustre_handle_is_used(struct lustre_handle *lh)
1105 return lh->cookie != 0ull;
1108 static inline int lustre_handle_equal(const struct lustre_handle *lh1,
1109 const struct lustre_handle *lh2)
1111 return lh1->cookie == lh2->cookie;
1114 static inline void lustre_handle_copy(struct lustre_handle *tgt,
1115 struct lustre_handle *src)
1117 tgt->cookie = src->cookie;
1120 /* flags for lm_flags */
1121 #define MSGHDR_AT_SUPPORT 0x1
1122 #define MSGHDR_CKSUM_INCOMPAT18 0x2
1124 #define lustre_msg lustre_msg_v2
1125 /* we depend on this structure to be 8-byte aligned */
1126 /* this type is only endian-adjusted in lustre_unpack_msg() */
1127 struct lustre_msg_v2 {
1136 __u32 lm_buflens[0];
1139 /* without gss, ptlrpc_body is put at the first buffer. */
1140 #define PTLRPC_NUM_VERSIONS 4
1141 #define JOBSTATS_JOBID_SIZE 32 /* 32 bytes string */
1142 struct ptlrpc_body_v3 {
1143 struct lustre_handle pb_handle;
1150 __u64 pb_last_committed;
1155 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1156 __u32 pb_service_time; /* for rep, actual service time */
1159 /* VBR: pre-versions */
1160 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1161 /* padding for future needs */
1162 __u64 pb_padding[4];
1163 char pb_jobid[JOBSTATS_JOBID_SIZE];
1165 #define ptlrpc_body ptlrpc_body_v3
1167 struct ptlrpc_body_v2 {
1168 struct lustre_handle pb_handle;
1175 __u64 pb_last_committed;
1180 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1181 __u32 pb_service_time; /* for rep, actual service time, also used for
1182 net_latency of req */
1185 /* VBR: pre-versions */
1186 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1187 /* padding for future needs */
1188 __u64 pb_padding[4];
1191 extern void lustre_swab_ptlrpc_body(struct ptlrpc_body *pb);
1193 /* message body offset for lustre_msg_v2 */
1194 /* ptlrpc body offset in all request/reply messages */
1195 #define MSG_PTLRPC_BODY_OFF 0
1197 /* normal request/reply message record offset */
1198 #define REQ_REC_OFF 1
1199 #define REPLY_REC_OFF 1
1201 /* ldlm request message body offset */
1202 #define DLM_LOCKREQ_OFF 1 /* lockreq offset */
1203 #define DLM_REQ_REC_OFF 2 /* normal dlm request record offset */
1205 /* ldlm intent lock message body offset */
1206 #define DLM_INTENT_IT_OFF 2 /* intent lock it offset */
1207 #define DLM_INTENT_REC_OFF 3 /* intent lock record offset */
1209 /* ldlm reply message body offset */
1210 #define DLM_LOCKREPLY_OFF 1 /* lockrep offset */
1211 #define DLM_REPLY_REC_OFF 2 /* reply record offset */
1213 /** only use in req->rq_{req,rep}_swab_mask */
1214 #define MSG_PTLRPC_HEADER_OFF 31
1216 /* Flags that are operation-specific go in the top 16 bits. */
1217 #define MSG_OP_FLAG_MASK 0xffff0000
1218 #define MSG_OP_FLAG_SHIFT 16
1220 /* Flags that apply to all requests are in the bottom 16 bits */
1221 #define MSG_GEN_FLAG_MASK 0x0000ffff
1222 #define MSG_LAST_REPLAY 0x0001
1223 #define MSG_RESENT 0x0002
1224 #define MSG_REPLAY 0x0004
1225 /* #define MSG_AT_SUPPORT 0x0008
1226 * This was used in early prototypes of adaptive timeouts, and while there
1227 * shouldn't be any users of that code there also isn't a need for using this
1228 * bits. Defer usage until at least 1.10 to avoid potential conflict. */
1229 #define MSG_DELAY_REPLAY 0x0010
1230 #define MSG_VERSION_REPLAY 0x0020
1231 #define MSG_REQ_REPLAY_DONE 0x0040
1232 #define MSG_LOCK_REPLAY_DONE 0x0080
1235 * Flags for all connect opcodes (MDS_CONNECT, OST_CONNECT)
1238 #define MSG_CONNECT_RECOVERING 0x00000001
1239 #define MSG_CONNECT_RECONNECT 0x00000002
1240 #define MSG_CONNECT_REPLAYABLE 0x00000004
1241 //#define MSG_CONNECT_PEER 0x8
1242 #define MSG_CONNECT_LIBCLIENT 0x00000010
1243 #define MSG_CONNECT_INITIAL 0x00000020
1244 #define MSG_CONNECT_ASYNC 0x00000040
1245 #define MSG_CONNECT_NEXT_VER 0x00000080 /* use next version of lustre_msg */
1246 #define MSG_CONNECT_TRANSNO 0x00000100 /* report transno */
1249 #define OBD_CONNECT_RDONLY 0x1ULL /*client has read-only access*/
1250 #define OBD_CONNECT_INDEX 0x2ULL /*connect specific LOV idx */
1251 #define OBD_CONNECT_MDS 0x4ULL /*connect from MDT to OST */
1252 #define OBD_CONNECT_GRANT 0x8ULL /*OSC gets grant at connect */
1253 #define OBD_CONNECT_SRVLOCK 0x10ULL /*server takes locks for cli */
1254 #define OBD_CONNECT_VERSION 0x20ULL /*Lustre versions in ocd */
1255 #define OBD_CONNECT_REQPORTAL 0x40ULL /*Separate non-IO req portal */
1256 #define OBD_CONNECT_ACL 0x80ULL /*access control lists */
1257 #define OBD_CONNECT_XATTR 0x100ULL /*client use extended attr */
1258 #define OBD_CONNECT_CROW 0x200ULL /*MDS+OST create obj on write*/
1259 #define OBD_CONNECT_TRUNCLOCK 0x400ULL /*locks on server for punch */
1260 #define OBD_CONNECT_TRANSNO 0x800ULL /*replay sends init transno */
1261 #define OBD_CONNECT_IBITS 0x1000ULL /*support for inodebits locks*/
1262 #define OBD_CONNECT_JOIN 0x2000ULL /*files can be concatenated.
1263 *We do not support JOIN FILE
1264 *anymore, reserve this flags
1265 *just for preventing such bit
1267 #define OBD_CONNECT_ATTRFID 0x4000ULL /*Server can GetAttr By Fid*/
1268 #define OBD_CONNECT_NODEVOH 0x8000ULL /*No open hndl on specl nodes*/
1269 #define OBD_CONNECT_RMT_CLIENT 0x10000ULL /*Remote client */
1270 #define OBD_CONNECT_RMT_CLIENT_FORCE 0x20000ULL /*Remote client by force */
1271 #define OBD_CONNECT_BRW_SIZE 0x40000ULL /*Max bytes per rpc */
1272 #define OBD_CONNECT_QUOTA64 0x80000ULL /*Not used since 2.4 */
1273 #define OBD_CONNECT_MDS_CAPA 0x100000ULL /*MDS capability */
1274 #define OBD_CONNECT_OSS_CAPA 0x200000ULL /*OSS capability */
1275 #define OBD_CONNECT_CANCELSET 0x400000ULL /*Early batched cancels. */
1276 #define OBD_CONNECT_SOM 0x800000ULL /*Size on MDS */
1277 #define OBD_CONNECT_AT 0x1000000ULL /*client uses AT */
1278 #define OBD_CONNECT_LRU_RESIZE 0x2000000ULL /*LRU resize feature. */
1279 #define OBD_CONNECT_MDS_MDS 0x4000000ULL /*MDS-MDS connection */
1280 #define OBD_CONNECT_REAL 0x8000000ULL /*real connection */
1281 #define OBD_CONNECT_CHANGE_QS 0x10000000ULL /*Not used since 2.4 */
1282 #define OBD_CONNECT_CKSUM 0x20000000ULL /*support several cksum algos*/
1283 #define OBD_CONNECT_FID 0x40000000ULL /*FID is supported by server */
1284 #define OBD_CONNECT_VBR 0x80000000ULL /*version based recovery */
1285 #define OBD_CONNECT_LOV_V3 0x100000000ULL /*client supports LOV v3 EA */
1286 #define OBD_CONNECT_GRANT_SHRINK 0x200000000ULL /* support grant shrink */
1287 #define OBD_CONNECT_SKIP_ORPHAN 0x400000000ULL /* don't reuse orphan objids */
1288 #define OBD_CONNECT_MAX_EASIZE 0x800000000ULL /* preserved for large EA */
1289 #define OBD_CONNECT_FULL20 0x1000000000ULL /* it is 2.0 client */
1290 #define OBD_CONNECT_LAYOUTLOCK 0x2000000000ULL /* client uses layout lock */
1291 #define OBD_CONNECT_64BITHASH 0x4000000000ULL /* client supports 64-bits
1293 #define OBD_CONNECT_MAXBYTES 0x8000000000ULL /* max stripe size */
1294 #define OBD_CONNECT_IMP_RECOV 0x10000000000ULL /* imp recovery support */
1295 #define OBD_CONNECT_JOBSTATS 0x20000000000ULL /* jobid in ptlrpc_body */
1296 #define OBD_CONNECT_UMASK 0x40000000000ULL /* create uses client umask */
1297 #define OBD_CONNECT_EINPROGRESS 0x80000000000ULL /* client handles -EINPROGRESS
1298 * RPC error properly */
1299 #define OBD_CONNECT_GRANT_PARAM 0x100000000000ULL/* extra grant params used for
1300 * finer space reservation */
1301 #define OBD_CONNECT_FLOCK_OWNER 0x200000000000ULL /* for the fixed 1.8
1302 * policy and 2.x server */
1303 #define OBD_CONNECT_LVB_TYPE 0x400000000000ULL /* variable type of LVB */
1304 #define OBD_CONNECT_NANOSEC_TIME 0x800000000000ULL /* nanosecond timestamps */
1305 #define OBD_CONNECT_LIGHTWEIGHT 0x1000000000000ULL/* lightweight connection */
1306 #define OBD_CONNECT_SHORTIO 0x2000000000000ULL/* short io */
1307 #define OBD_CONNECT_PINGLESS 0x4000000000000ULL/* pings not required */
1309 * Please DO NOT add flag values here before first ensuring that this same
1310 * flag value is not in use on some other branch. Please clear any such
1311 * changes with senior engineers before starting to use a new flag. Then,
1312 * submit a small patch against EVERY branch that ONLY adds the new flag,
1313 * updates obd_connect_names[] for lprocfs_rd_connect_flags(), adds the
1314 * flag to check_obd_connect_data(), and updates wiretests accordingly, so it
1315 * can be approved and landed easily to reserve the flag for future use. */
1317 /* The MNE_SWAB flag is overloading the MDS_MDS bit only for the MGS
1318 * connection. It is a temporary bug fix for Imperative Recovery interop
1319 * between 2.2 and 2.3 x86/ppc nodes, and can be removed when interop for
1320 * 2.2 clients/servers is no longer needed. LU-1252/LU-1644. */
1321 #define OBD_CONNECT_MNE_SWAB OBD_CONNECT_MDS_MDS
1323 #define OCD_HAS_FLAG(ocd, flg) \
1324 (!!((ocd)->ocd_connect_flags & OBD_CONNECT_##flg))
1327 #ifdef HAVE_LRU_RESIZE_SUPPORT
1328 #define LRU_RESIZE_CONNECT_FLAG OBD_CONNECT_LRU_RESIZE
1330 #define LRU_RESIZE_CONNECT_FLAG 0
1333 #define MDT_CONNECT_SUPPORTED (OBD_CONNECT_RDONLY | OBD_CONNECT_VERSION | \
1334 OBD_CONNECT_ACL | OBD_CONNECT_XATTR | \
1335 OBD_CONNECT_IBITS | \
1336 OBD_CONNECT_NODEVOH | OBD_CONNECT_ATTRFID | \
1337 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1338 OBD_CONNECT_RMT_CLIENT | \
1339 OBD_CONNECT_RMT_CLIENT_FORCE | \
1340 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_MDS_CAPA | \
1341 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_MDS_MDS | \
1342 OBD_CONNECT_FID | LRU_RESIZE_CONNECT_FLAG | \
1343 OBD_CONNECT_VBR | OBD_CONNECT_LOV_V3 | \
1344 OBD_CONNECT_SOM | OBD_CONNECT_FULL20 | \
1345 OBD_CONNECT_64BITHASH | OBD_CONNECT_JOBSTATS | \
1346 OBD_CONNECT_EINPROGRESS | \
1347 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_UMASK | \
1348 OBD_CONNECT_LVB_TYPE | OBD_CONNECT_LAYOUTLOCK |\
1349 OBD_CONNECT_PINGLESS)
1350 #define OST_CONNECT_SUPPORTED (OBD_CONNECT_SRVLOCK | OBD_CONNECT_GRANT | \
1351 OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | \
1352 OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_INDEX | \
1353 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_OSS_CAPA | \
1354 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1355 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_CKSUM | \
1356 OBD_CONNECT_RMT_CLIENT | \
1357 OBD_CONNECT_RMT_CLIENT_FORCE | OBD_CONNECT_VBR | \
1358 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN | \
1359 OBD_CONNECT_GRANT_SHRINK | OBD_CONNECT_FULL20 | \
1360 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES | \
1361 OBD_CONNECT_MAX_EASIZE | \
1362 OBD_CONNECT_EINPROGRESS | \
1363 OBD_CONNECT_JOBSTATS | \
1364 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LVB_TYPE|\
1365 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_FID | \
1366 OBD_CONNECT_PINGLESS)
1367 #define ECHO_CONNECT_SUPPORTED (0)
1368 #define MGS_CONNECT_SUPPORTED (OBD_CONNECT_VERSION | OBD_CONNECT_AT | \
1369 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV | \
1370 OBD_CONNECT_MNE_SWAB | OBD_CONNECT_PINGLESS)
1372 /* Features required for this version of the client to work with server */
1373 #define CLIENT_CONNECT_MDT_REQD (OBD_CONNECT_IBITS | OBD_CONNECT_FID | \
1376 #define OBD_OCD_VERSION(major,minor,patch,fix) (((major)<<24) + ((minor)<<16) +\
1377 ((patch)<<8) + (fix))
1378 #define OBD_OCD_VERSION_MAJOR(version) ((int)((version)>>24)&255)
1379 #define OBD_OCD_VERSION_MINOR(version) ((int)((version)>>16)&255)
1380 #define OBD_OCD_VERSION_PATCH(version) ((int)((version)>>8)&255)
1381 #define OBD_OCD_VERSION_FIX(version) ((int)(version)&255)
1383 /* This structure is used for both request and reply.
1385 * If we eventually have separate connect data for different types, which we
1386 * almost certainly will, then perhaps we stick a union in here. */
1387 struct obd_connect_data_v1 {
1388 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1389 __u32 ocd_version; /* lustre release version number */
1390 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1391 __u32 ocd_index; /* LOV index to connect to */
1392 __u32 ocd_brw_size; /* Maximum BRW size in bytes, must be 2^n */
1393 __u64 ocd_ibits_known; /* inode bits this client understands */
1394 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1395 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1396 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1397 __u32 ocd_unused; /* also fix lustre_swab_connect */
1398 __u64 ocd_transno; /* first transno from client to be replayed */
1399 __u32 ocd_group; /* MDS group on OST */
1400 __u32 ocd_cksum_types; /* supported checksum algorithms */
1401 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1402 __u32 ocd_instance; /* also fix lustre_swab_connect */
1403 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1406 struct obd_connect_data {
1407 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1408 __u32 ocd_version; /* lustre release version number */
1409 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1410 __u32 ocd_index; /* LOV index to connect to */
1411 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
1412 __u64 ocd_ibits_known; /* inode bits this client understands */
1413 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1414 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1415 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1416 __u32 ocd_unused; /* also fix lustre_swab_connect */
1417 __u64 ocd_transno; /* first transno from client to be replayed */
1418 __u32 ocd_group; /* MDS group on OST */
1419 __u32 ocd_cksum_types; /* supported checksum algorithms */
1420 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1421 __u32 ocd_instance; /* instance # of this target */
1422 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1423 /* Fields after ocd_maxbytes are only accessible by the receiver
1424 * if the corresponding flag in ocd_connect_flags is set. Accessing
1425 * any field after ocd_maxbytes on the receiver without a valid flag
1426 * may result in out-of-bound memory access and kernel oops. */
1427 __u64 padding1; /* added 2.1.0. also fix lustre_swab_connect */
1428 __u64 padding2; /* added 2.1.0. also fix lustre_swab_connect */
1429 __u64 padding3; /* added 2.1.0. also fix lustre_swab_connect */
1430 __u64 padding4; /* added 2.1.0. also fix lustre_swab_connect */
1431 __u64 padding5; /* added 2.1.0. also fix lustre_swab_connect */
1432 __u64 padding6; /* added 2.1.0. also fix lustre_swab_connect */
1433 __u64 padding7; /* added 2.1.0. also fix lustre_swab_connect */
1434 __u64 padding8; /* added 2.1.0. also fix lustre_swab_connect */
1435 __u64 padding9; /* added 2.1.0. also fix lustre_swab_connect */
1436 __u64 paddingA; /* added 2.1.0. also fix lustre_swab_connect */
1437 __u64 paddingB; /* added 2.1.0. also fix lustre_swab_connect */
1438 __u64 paddingC; /* added 2.1.0. also fix lustre_swab_connect */
1439 __u64 paddingD; /* added 2.1.0. also fix lustre_swab_connect */
1440 __u64 paddingE; /* added 2.1.0. also fix lustre_swab_connect */
1441 __u64 paddingF; /* added 2.1.0. also fix lustre_swab_connect */
1444 * Please DO NOT use any fields here before first ensuring that this same
1445 * field is not in use on some other branch. Please clear any such changes
1446 * with senior engineers before starting to use a new field. Then, submit
1447 * a small patch against EVERY branch that ONLY adds the new field along with
1448 * the matching OBD_CONNECT flag, so that can be approved and landed easily to
1449 * reserve the flag for future use. */
1452 extern void lustre_swab_connect(struct obd_connect_data *ocd);
1455 * Supported checksum algorithms. Up to 32 checksum types are supported.
1456 * (32-bit mask stored in obd_connect_data::ocd_cksum_types)
1457 * Please update DECLARE_CKSUM_NAME/OBD_CKSUM_ALL in obd.h when adding a new
1458 * algorithm and also the OBD_FL_CKSUM* flags.
1461 OBD_CKSUM_CRC32 = 0x00000001,
1462 OBD_CKSUM_ADLER = 0x00000002,
1463 OBD_CKSUM_CRC32C= 0x00000004,
1467 * OST requests: OBDO & OBD request records
1472 OST_REPLY = 0, /* reply ? */
1488 OST_QUOTACHECK = 18,
1490 OST_QUOTA_ADJUST_QUNIT = 20, /* not used since 2.4 */
1493 #define OST_FIRST_OPC OST_REPLY
1496 OBD_FL_INLINEDATA = 0x00000001,
1497 OBD_FL_OBDMDEXISTS = 0x00000002,
1498 OBD_FL_DELORPHAN = 0x00000004, /* if set in o_flags delete orphans */
1499 OBD_FL_NORPC = 0x00000008, /* set in o_flags do in OSC not OST */
1500 OBD_FL_IDONLY = 0x00000010, /* set in o_flags only adjust obj id*/
1501 OBD_FL_RECREATE_OBJS= 0x00000020, /* recreate missing obj */
1502 OBD_FL_DEBUG_CHECK = 0x00000040, /* echo client/server debug check */
1503 OBD_FL_NO_USRQUOTA = 0x00000100, /* the object's owner is over quota */
1504 OBD_FL_NO_GRPQUOTA = 0x00000200, /* the object's group is over quota */
1505 OBD_FL_CREATE_CROW = 0x00000400, /* object should be create on write */
1506 OBD_FL_SRVLOCK = 0x00000800, /* delegate DLM locking to server */
1507 OBD_FL_CKSUM_CRC32 = 0x00001000, /* CRC32 checksum type */
1508 OBD_FL_CKSUM_ADLER = 0x00002000, /* ADLER checksum type */
1509 OBD_FL_CKSUM_CRC32C = 0x00004000, /* CRC32C checksum type */
1510 OBD_FL_CKSUM_RSVD2 = 0x00008000, /* for future cksum types */
1511 OBD_FL_CKSUM_RSVD3 = 0x00010000, /* for future cksum types */
1512 OBD_FL_SHRINK_GRANT = 0x00020000, /* object shrink the grant */
1513 OBD_FL_MMAP = 0x00040000, /* object is mmapped on the client.
1514 * XXX: obsoleted - reserved for old
1515 * clients prior than 2.2 */
1516 OBD_FL_RECOV_RESEND = 0x00080000, /* recoverable resent */
1517 OBD_FL_NOSPC_BLK = 0x00100000, /* no more block space on OST */
1519 /* Note that while these checksum values are currently separate bits,
1520 * in 2.x we can actually allow all values from 1-31 if we wanted. */
1521 OBD_FL_CKSUM_ALL = OBD_FL_CKSUM_CRC32 | OBD_FL_CKSUM_ADLER |
1522 OBD_FL_CKSUM_CRC32C,
1524 /* mask for local-only flag, which won't be sent over network */
1525 OBD_FL_LOCAL_MASK = 0xF0000000,
1528 #define LOV_MAGIC_V1 0x0BD10BD0
1529 #define LOV_MAGIC LOV_MAGIC_V1
1530 #define LOV_MAGIC_JOIN_V1 0x0BD20BD0
1531 #define LOV_MAGIC_V3 0x0BD30BD0
1534 * magic for fully defined striping
1535 * the idea is that we should have different magics for striping "hints"
1536 * (struct lov_user_md_v[13]) and defined ready-to-use striping (struct
1537 * lov_mds_md_v[13]). at the moment the magics are used in wire protocol,
1538 * we can't just change it w/o long way preparation, but we still need a
1539 * mechanism to allow LOD to differentiate hint versus ready striping.
1540 * so, at the moment we do a trick: MDT knows what to expect from request
1541 * depending on the case (replay uses ready striping, non-replay req uses
1542 * hints), so MDT replaces magic with appropriate one and now LOD can
1543 * easily understand what's inside -bzzz
1545 #define LOV_MAGIC_V1_DEF 0x0CD10BD0
1546 #define LOV_MAGIC_V3_DEF 0x0CD30BD0
1548 #define LOV_PATTERN_RAID0 0x001 /* stripes are used round-robin */
1549 #define LOV_PATTERN_RAID1 0x002 /* stripes are mirrors of each other */
1550 #define LOV_PATTERN_FIRST 0x100 /* first stripe is not in round-robin */
1551 #define LOV_PATTERN_CMOBD 0x200
1553 #define lov_ost_data lov_ost_data_v1
1554 struct lov_ost_data_v1 { /* per-stripe data structure (little-endian)*/
1555 struct ost_id l_ost_oi; /* OST object ID */
1556 __u32 l_ost_gen; /* generation of this l_ost_idx */
1557 __u32 l_ost_idx; /* OST index in LOV (lov_tgt_desc->tgts) */
1560 #define lov_mds_md lov_mds_md_v1
1561 struct lov_mds_md_v1 { /* LOV EA mds/wire data (little-endian) */
1562 __u32 lmm_magic; /* magic number = LOV_MAGIC_V1 */
1563 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1564 struct ost_id lmm_oi; /* LOV object ID */
1565 __u32 lmm_stripe_size; /* size of stripe in bytes */
1566 /* lmm_stripe_count used to be __u32 */
1567 __u16 lmm_stripe_count; /* num stripes in use for this object */
1568 __u16 lmm_layout_gen; /* layout generation number */
1569 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1573 * Sigh, because pre-2.4 uses
1574 * struct lov_mds_md_v1 {
1576 * __u64 lmm_object_id;
1577 * __u64 lmm_object_seq;
1580 * to identify the LOV(MDT) object, and lmm_object_seq will
1581 * be normal_fid, which make it hard to combine these conversion
1582 * to ostid_to FID. so we will do lmm_oi/fid conversion separately
1584 * We can tell the lmm_oi by this way,
1585 * 1.8: lmm_object_id = {inode}, lmm_object_gr = 0
1586 * 2.1: lmm_object_id = {oid < 128k}, lmm_object_seq = FID_SEQ_NORMAL
1587 * 2.4: lmm_oi.f_seq = FID_SEQ_NORMAL, lmm_oi.f_oid = {oid < 128k},
1590 * But currently lmm_oi/lsm_oi does not have any "real" usages,
1591 * except for printing some information, and the user can always
1592 * get the real FID from LMA, besides this multiple case check might
1593 * make swab more complicate. So we will keep using id/seq for lmm_oi.
1596 static inline void fid_to_lmm_oi(const struct lu_fid *fid,
1599 oi->oi.oi_id = fid_oid(fid);
1600 oi->oi.oi_seq = fid_seq(fid);
1603 static inline void lmm_oi_set_seq(struct ost_id *oi, __u64 seq)
1605 oi->oi.oi_seq = seq;
1608 static inline __u64 lmm_oi_id(struct ost_id *oi)
1610 return oi->oi.oi_id;
1613 static inline __u64 lmm_oi_seq(struct ost_id *oi)
1615 return oi->oi.oi_seq;
1618 static inline void lmm_oi_le_to_cpu(struct ost_id *dst_oi,
1619 struct ost_id *src_oi)
1621 dst_oi->oi.oi_id = le64_to_cpu(src_oi->oi.oi_id);
1622 dst_oi->oi.oi_seq = le64_to_cpu(src_oi->oi.oi_seq);
1625 static inline void lmm_oi_cpu_to_le(struct ost_id *dst_oi,
1626 struct ost_id *src_oi)
1628 dst_oi->oi.oi_id = cpu_to_le64(src_oi->oi.oi_id);
1629 dst_oi->oi.oi_seq = cpu_to_le64(src_oi->oi.oi_seq);
1632 /* extern void lustre_swab_lov_mds_md(struct lov_mds_md *llm); */
1634 #define MAX_MD_SIZE (sizeof(struct lov_mds_md) + 4 * sizeof(struct lov_ost_data))
1635 #define MIN_MD_SIZE (sizeof(struct lov_mds_md) + 1 * sizeof(struct lov_ost_data))
1637 #define XATTR_NAME_ACL_ACCESS "system.posix_acl_access"
1638 #define XATTR_NAME_ACL_DEFAULT "system.posix_acl_default"
1639 #define XATTR_USER_PREFIX "user."
1640 #define XATTR_TRUSTED_PREFIX "trusted."
1641 #define XATTR_SECURITY_PREFIX "security."
1642 #define XATTR_LUSTRE_PREFIX "lustre."
1644 #define XATTR_NAME_LOV "trusted.lov"
1645 #define XATTR_NAME_LMA "trusted.lma"
1646 #define XATTR_NAME_LMV "trusted.lmv"
1647 #define XATTR_NAME_LINK "trusted.link"
1648 #define XATTR_NAME_FID "trusted.fid"
1649 #define XATTR_NAME_VERSION "trusted.version"
1650 #define XATTR_NAME_SOM "trusted.som"
1651 #define XATTR_NAME_HSM "trusted.hsm"
1652 #define XATTR_NAME_LFSCK_NAMESPACE "trusted.lfsck_namespace"
1654 struct lov_mds_md_v3 { /* LOV EA mds/wire data (little-endian) */
1655 __u32 lmm_magic; /* magic number = LOV_MAGIC_V3 */
1656 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1657 struct ost_id lmm_oi; /* LOV object ID */
1658 __u32 lmm_stripe_size; /* size of stripe in bytes */
1659 /* lmm_stripe_count used to be __u32 */
1660 __u16 lmm_stripe_count; /* num stripes in use for this object */
1661 __u16 lmm_layout_gen; /* layout generation number */
1662 char lmm_pool_name[LOV_MAXPOOLNAME]; /* must be 32bit aligned */
1663 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1666 #define OBD_MD_FLID (0x00000001ULL) /* object ID */
1667 #define OBD_MD_FLATIME (0x00000002ULL) /* access time */
1668 #define OBD_MD_FLMTIME (0x00000004ULL) /* data modification time */
1669 #define OBD_MD_FLCTIME (0x00000008ULL) /* change time */
1670 #define OBD_MD_FLSIZE (0x00000010ULL) /* size */
1671 #define OBD_MD_FLBLOCKS (0x00000020ULL) /* allocated blocks count */
1672 #define OBD_MD_FLBLKSZ (0x00000040ULL) /* block size */
1673 #define OBD_MD_FLMODE (0x00000080ULL) /* access bits (mode & ~S_IFMT) */
1674 #define OBD_MD_FLTYPE (0x00000100ULL) /* object type (mode & S_IFMT) */
1675 #define OBD_MD_FLUID (0x00000200ULL) /* user ID */
1676 #define OBD_MD_FLGID (0x00000400ULL) /* group ID */
1677 #define OBD_MD_FLFLAGS (0x00000800ULL) /* flags word */
1678 #define OBD_MD_FLNLINK (0x00002000ULL) /* link count */
1679 #define OBD_MD_FLGENER (0x00004000ULL) /* generation number */
1680 /*#define OBD_MD_FLINLINE (0x00008000ULL) inline data. used until 1.6.5 */
1681 #define OBD_MD_FLRDEV (0x00010000ULL) /* device number */
1682 #define OBD_MD_FLEASIZE (0x00020000ULL) /* extended attribute data */
1683 #define OBD_MD_LINKNAME (0x00040000ULL) /* symbolic link target */
1684 #define OBD_MD_FLHANDLE (0x00080000ULL) /* file/lock handle */
1685 #define OBD_MD_FLCKSUM (0x00100000ULL) /* bulk data checksum */
1686 #define OBD_MD_FLQOS (0x00200000ULL) /* quality of service stats */
1687 /*#define OBD_MD_FLOSCOPQ (0x00400000ULL) osc opaque data, never used */
1688 #define OBD_MD_FLCOOKIE (0x00800000ULL) /* log cancellation cookie */
1689 #define OBD_MD_FLGROUP (0x01000000ULL) /* group */
1690 #define OBD_MD_FLFID (0x02000000ULL) /* ->ost write inline fid */
1691 #define OBD_MD_FLEPOCH (0x04000000ULL) /* ->ost write with ioepoch */
1692 /* ->mds if epoch opens or closes */
1693 #define OBD_MD_FLGRANT (0x08000000ULL) /* ost preallocation space grant */
1694 #define OBD_MD_FLDIREA (0x10000000ULL) /* dir's extended attribute data */
1695 #define OBD_MD_FLUSRQUOTA (0x20000000ULL) /* over quota flags sent from ost */
1696 #define OBD_MD_FLGRPQUOTA (0x40000000ULL) /* over quota flags sent from ost */
1697 #define OBD_MD_FLMODEASIZE (0x80000000ULL) /* EA size will be changed */
1699 #define OBD_MD_MDS (0x0000000100000000ULL) /* where an inode lives on */
1700 #define OBD_MD_REINT (0x0000000200000000ULL) /* reintegrate oa */
1701 #define OBD_MD_MEA (0x0000000400000000ULL) /* CMD split EA */
1703 /* OBD_MD_MDTIDX is used to get MDT index, but it is never been used overwire,
1704 * and it is already obsolete since 2.3 */
1705 /* #define OBD_MD_MDTIDX (0x0000000800000000ULL) */
1707 #define OBD_MD_FLXATTR (0x0000001000000000ULL) /* xattr */
1708 #define OBD_MD_FLXATTRLS (0x0000002000000000ULL) /* xattr list */
1709 #define OBD_MD_FLXATTRRM (0x0000004000000000ULL) /* xattr remove */
1710 #define OBD_MD_FLACL (0x0000008000000000ULL) /* ACL */
1711 #define OBD_MD_FLRMTPERM (0x0000010000000000ULL) /* remote permission */
1712 #define OBD_MD_FLMDSCAPA (0x0000020000000000ULL) /* MDS capability */
1713 #define OBD_MD_FLOSSCAPA (0x0000040000000000ULL) /* OSS capability */
1714 #define OBD_MD_FLCKSPLIT (0x0000080000000000ULL) /* Check split on server */
1715 #define OBD_MD_FLCROSSREF (0x0000100000000000ULL) /* Cross-ref case */
1716 #define OBD_MD_FLGETATTRLOCK (0x0000200000000000ULL) /* Get IOEpoch attributes
1718 #define OBD_MD_FLOBJCOUNT (0x0000400000000000ULL) /* for multiple destroy */
1720 #define OBD_MD_FLRMTLSETFACL (0x0001000000000000ULL) /* lfs lsetfacl case */
1721 #define OBD_MD_FLRMTLGETFACL (0x0002000000000000ULL) /* lfs lgetfacl case */
1722 #define OBD_MD_FLRMTRSETFACL (0x0004000000000000ULL) /* lfs rsetfacl case */
1723 #define OBD_MD_FLRMTRGETFACL (0x0008000000000000ULL) /* lfs rgetfacl case */
1725 #define OBD_MD_FLDATAVERSION (0x0010000000000000ULL) /* iversion sum */
1727 #define OBD_MD_FLGETATTR (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLMTIME | \
1728 OBD_MD_FLCTIME | OBD_MD_FLSIZE | OBD_MD_FLBLKSZ | \
1729 OBD_MD_FLMODE | OBD_MD_FLTYPE | OBD_MD_FLUID | \
1730 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLNLINK | \
1731 OBD_MD_FLGENER | OBD_MD_FLRDEV | OBD_MD_FLGROUP)
1733 /* don't forget obdo_fid which is way down at the bottom so it can
1734 * come after the definition of llog_cookie */
1738 HSS_CLEARMASK = 0x02,
1739 HSS_ARCHIVE_ID = 0x04,
1742 struct hsm_state_set {
1744 __u32 hss_archive_id;
1746 __u64 hss_clearmask;
1749 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
1750 extern void lustre_swab_hsm_state_set(struct hsm_state_set *hss);
1752 extern void lustre_swab_obd_statfs (struct obd_statfs *os);
1754 /* ost_body.data values for OST_BRW */
1756 #define OBD_BRW_READ 0x01
1757 #define OBD_BRW_WRITE 0x02
1758 #define OBD_BRW_RWMASK (OBD_BRW_READ | OBD_BRW_WRITE)
1759 #define OBD_BRW_SYNC 0x08 /* this page is a part of synchronous
1760 * transfer and is not accounted in
1762 #define OBD_BRW_CHECK 0x10
1763 #define OBD_BRW_FROM_GRANT 0x20 /* the osc manages this under llite */
1764 #define OBD_BRW_GRANTED 0x40 /* the ost manages this */
1765 #define OBD_BRW_NOCACHE 0x80 /* this page is a part of non-cached IO */
1766 #define OBD_BRW_NOQUOTA 0x100
1767 #define OBD_BRW_SRVLOCK 0x200 /* Client holds no lock over this page */
1768 #define OBD_BRW_ASYNC 0x400 /* Server may delay commit to disk */
1769 #define OBD_BRW_MEMALLOC 0x800 /* Client runs in the "kswapd" context */
1770 #define OBD_BRW_OVER_USRQUOTA 0x1000 /* Running out of user quota */
1771 #define OBD_BRW_OVER_GRPQUOTA 0x2000 /* Running out of group quota */
1773 #define OBD_OBJECT_EOF 0xffffffffffffffffULL
1775 #define OST_MIN_PRECREATE 32
1776 #define OST_MAX_PRECREATE 20000
1779 struct ost_id ioo_oid; /* object ID, if multi-obj BRW */
1780 __u32 ioo_max_brw; /* low 16 bits were o_mode before 2.4,
1781 * now (PTLRPC_BULK_OPS_COUNT - 1) in
1782 * high 16 bits in 2.4 and later */
1783 __u32 ioo_bufcnt; /* number of niobufs for this object */
1786 #define IOOBJ_MAX_BRW_BITS 16
1787 #define IOOBJ_TYPE_MASK ((1U << IOOBJ_MAX_BRW_BITS) - 1)
1788 #define ioobj_max_brw_get(ioo) (((ioo)->ioo_max_brw >> IOOBJ_MAX_BRW_BITS) + 1)
1789 #define ioobj_max_brw_set(ioo, num) \
1790 do { (ioo)->ioo_max_brw = ((num) - 1) << IOOBJ_MAX_BRW_BITS; } while (0)
1792 extern void lustre_swab_obd_ioobj (struct obd_ioobj *ioo);
1794 /* multiple of 8 bytes => can array */
1795 struct niobuf_remote {
1801 extern void lustre_swab_niobuf_remote (struct niobuf_remote *nbr);
1803 /* lock value block communicated between the filter and llite */
1805 /* OST_LVB_ERR_INIT is needed because the return code in rc is
1806 * negative, i.e. because ((MASK + rc) & MASK) != MASK. */
1807 #define OST_LVB_ERR_INIT 0xffbadbad80000000ULL
1808 #define OST_LVB_ERR_MASK 0xffbadbad00000000ULL
1809 #define OST_LVB_IS_ERR(blocks) \
1810 ((blocks & OST_LVB_ERR_MASK) == OST_LVB_ERR_MASK)
1811 #define OST_LVB_SET_ERR(blocks, rc) \
1812 do { blocks = OST_LVB_ERR_INIT + rc; } while (0)
1813 #define OST_LVB_GET_ERR(blocks) (int)(blocks - OST_LVB_ERR_INIT)
1823 extern void lustre_swab_ost_lvb_v1(struct ost_lvb_v1 *lvb);
1837 extern void lustre_swab_ost_lvb(struct ost_lvb *lvb);
1840 * lquota data structures
1843 #ifndef QUOTABLOCK_BITS
1844 #define QUOTABLOCK_BITS 10
1847 #ifndef QUOTABLOCK_SIZE
1848 #define QUOTABLOCK_SIZE (1 << QUOTABLOCK_BITS)
1852 #define toqb(x) (((x) + QUOTABLOCK_SIZE - 1) >> QUOTABLOCK_BITS)
1855 /* The lquota_id structure is an union of all the possible identifier types that
1856 * can be used with quota, this includes:
1859 * - a FID which can be used for per-directory quota in the future */
1861 struct lu_fid qid_fid; /* FID for per-directory quota */
1862 __u64 qid_uid; /* user identifier */
1863 __u64 qid_gid; /* group identifier */
1866 /* quotactl management */
1867 struct obd_quotactl {
1869 __u32 qc_type; /* see Q_* flag below */
1872 struct obd_dqinfo qc_dqinfo;
1873 struct obd_dqblk qc_dqblk;
1876 extern void lustre_swab_obd_quotactl(struct obd_quotactl *q);
1878 #define Q_QUOTACHECK 0x800100 /* deprecated as of 2.4 */
1879 #define Q_INITQUOTA 0x800101 /* deprecated as of 2.4 */
1880 #define Q_GETOINFO 0x800102 /* get obd quota info */
1881 #define Q_GETOQUOTA 0x800103 /* get obd quotas */
1882 #define Q_FINVALIDATE 0x800104 /* deprecated as of 2.4 */
1884 #define Q_COPY(out, in, member) (out)->member = (in)->member
1886 #define QCTL_COPY(out, in) \
1888 Q_COPY(out, in, qc_cmd); \
1889 Q_COPY(out, in, qc_type); \
1890 Q_COPY(out, in, qc_id); \
1891 Q_COPY(out, in, qc_stat); \
1892 Q_COPY(out, in, qc_dqinfo); \
1893 Q_COPY(out, in, qc_dqblk); \
1896 /* Body of quota request used for quota acquire/release RPCs between quota
1897 * master (aka QMT) and slaves (ak QSD). */
1899 struct lu_fid qb_fid; /* FID of global index packing the pool ID
1900 * and type (data or metadata) as well as
1901 * the quota type (user or group). */
1902 union lquota_id qb_id; /* uid or gid or directory FID */
1903 __u32 qb_flags; /* see below */
1905 __u64 qb_count; /* acquire/release count (kbytes/inodes) */
1906 __u64 qb_usage; /* current slave usage (kbytes/inodes) */
1907 __u64 qb_slv_ver; /* slave index file version */
1908 struct lustre_handle qb_lockh; /* per-ID lock handle */
1909 struct lustre_handle qb_glb_lockh; /* global lock handle */
1910 __u64 qb_padding1[4];
1913 /* When the quota_body is used in the reply of quota global intent
1914 * lock (IT_QUOTA_CONN) reply, qb_fid contains slave index file FID. */
1915 #define qb_slv_fid qb_fid
1916 /* qb_usage is the current qunit (in kbytes/inodes) when quota_body is used in
1918 #define qb_qunit qb_usage
1920 #define QUOTA_DQACQ_FL_ACQ 0x1 /* acquire quota */
1921 #define QUOTA_DQACQ_FL_PREACQ 0x2 /* pre-acquire */
1922 #define QUOTA_DQACQ_FL_REL 0x4 /* release quota */
1923 #define QUOTA_DQACQ_FL_REPORT 0x8 /* report usage */
1925 extern void lustre_swab_quota_body(struct quota_body *b);
1927 /* Quota types currently supported */
1929 LQUOTA_TYPE_USR = 0x00, /* maps to USRQUOTA */
1930 LQUOTA_TYPE_GRP = 0x01, /* maps to GRPQUOTA */
1934 /* There are 2 different resource types on which a quota limit can be enforced:
1935 * - inodes on the MDTs
1936 * - blocks on the OSTs */
1938 LQUOTA_RES_MD = 0x01, /* skip 0 to avoid null oid in FID */
1939 LQUOTA_RES_DT = 0x02,
1941 LQUOTA_FIRST_RES = LQUOTA_RES_MD
1943 #define LQUOTA_NR_RES (LQUOTA_LAST_RES - LQUOTA_FIRST_RES + 1)
1946 * Space accounting support
1947 * Format of an accounting record, providing disk usage information for a given
1950 struct lquota_acct_rec { /* 16 bytes */
1951 __u64 bspace; /* current space in use */
1952 __u64 ispace; /* current # inodes in use */
1956 * Global quota index support
1957 * Format of a global record, providing global quota settings for a given quota
1960 struct lquota_glb_rec { /* 32 bytes */
1961 __u64 qbr_hardlimit; /* quota hard limit, in #inodes or kbytes */
1962 __u64 qbr_softlimit; /* quota soft limit, in #inodes or kbytes */
1963 __u64 qbr_time; /* grace time, in seconds */
1964 __u64 qbr_granted; /* how much is granted to slaves, in #inodes or
1969 * Slave index support
1970 * Format of a slave record, recording how much space is granted to a given
1973 struct lquota_slv_rec { /* 8 bytes */
1974 __u64 qsr_granted; /* space granted to the slave for the key=ID,
1975 * in #inodes or kbytes */
1978 /* Data structures associated with the quota locks */
1980 /* Glimpse descriptor used for the index & per-ID quota locks */
1981 struct ldlm_gl_lquota_desc {
1982 union lquota_id gl_id; /* quota ID subject to the glimpse */
1983 __u64 gl_flags; /* see LQUOTA_FL* below */
1984 __u64 gl_ver; /* new index version */
1985 __u64 gl_hardlimit; /* new hardlimit or qunit value */
1986 __u64 gl_softlimit; /* new softlimit */
1990 #define gl_qunit gl_hardlimit /* current qunit value used when
1991 * glimpsing per-ID quota locks */
1993 /* quota glimpse flags */
1994 #define LQUOTA_FL_EDQUOT 0x1 /* user/group out of quota space on QMT */
1996 /* LVB used with quota (global and per-ID) locks */
1998 __u64 lvb_flags; /* see LQUOTA_FL* above */
1999 __u64 lvb_id_may_rel; /* space that might be released later */
2000 __u64 lvb_id_rel; /* space released by the slave for this ID */
2001 __u64 lvb_id_qunit; /* current qunit value */
2005 extern void lustre_swab_lquota_lvb(struct lquota_lvb *lvb);
2007 /* LVB used with global quota lock */
2008 #define lvb_glb_ver lvb_id_may_rel /* current version of the global index */
2016 #define QUOTA_FIRST_OPC QUOTA_DQACQ
2025 MDS_GETATTR_NAME = 34,
2030 MDS_DISCONNECT = 39,
2036 MDS_DONE_WRITING = 45,
2038 MDS_QUOTACHECK = 47,
2041 MDS_SETXATTR = 50, /* obsolete, now it's MDS_REINT op */
2045 MDS_HSM_STATE_GET = 54,
2046 MDS_HSM_STATE_SET = 55,
2047 MDS_HSM_ACTION = 56,
2048 MDS_HSM_PROGRESS = 57,
2049 MDS_HSM_REQUEST = 58,
2050 MDS_HSM_CT_REGISTER = 59,
2051 MDS_HSM_CT_UNREGISTER = 60,
2052 MDS_SWAP_LAYOUTS = 61,
2056 #define MDS_FIRST_OPC MDS_GETATTR
2059 /* opcodes for object update */
2065 #define UPDATE_FIRST_OPC UPDATE_OBJ
2082 } mds_reint_t, mdt_reint_t;
2084 extern void lustre_swab_generic_32s (__u32 *val);
2086 /* the disposition of the intent outlines what was executed */
2087 #define DISP_IT_EXECD 0x00000001
2088 #define DISP_LOOKUP_EXECD 0x00000002
2089 #define DISP_LOOKUP_NEG 0x00000004
2090 #define DISP_LOOKUP_POS 0x00000008
2091 #define DISP_OPEN_CREATE 0x00000010
2092 #define DISP_OPEN_OPEN 0x00000020
2093 #define DISP_ENQ_COMPLETE 0x00400000
2094 #define DISP_ENQ_OPEN_REF 0x00800000
2095 #define DISP_ENQ_CREATE_REF 0x01000000
2096 #define DISP_OPEN_LOCK 0x02000000
2098 /* INODE LOCK PARTS */
2099 #define MDS_INODELOCK_LOOKUP 0x000001 /* dentry, mode, owner, group */
2100 #define MDS_INODELOCK_UPDATE 0x000002 /* size, links, timestamps */
2101 #define MDS_INODELOCK_OPEN 0x000004 /* For opened files */
2102 #define MDS_INODELOCK_LAYOUT 0x000008 /* for layout */
2103 #define MDS_INODELOCK_PERM 0x000010 /* for permission */
2105 #define MDS_INODELOCK_MAXSHIFT 4
2106 /* This FULL lock is useful to take on unlink sort of operations */
2107 #define MDS_INODELOCK_FULL ((1<<(MDS_INODELOCK_MAXSHIFT+1))-1)
2109 extern void lustre_swab_ll_fid (struct ll_fid *fid);
2111 /* NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
2112 * but was moved into name[1] along with the OID to avoid consuming the
2113 * name[2,3] fields that need to be used for the quota id (also a FID). */
2115 LUSTRE_RES_ID_SEQ_OFF = 0,
2116 LUSTRE_RES_ID_VER_OID_OFF = 1,
2117 LUSTRE_RES_ID_WAS_VER_OFF = 2, /* see note above */
2118 LUSTRE_RES_ID_QUOTA_SEQ_OFF = 2,
2119 LUSTRE_RES_ID_QUOTA_VER_OID_OFF = 3,
2120 LUSTRE_RES_ID_HSH_OFF = 3
2123 #define MDS_STATUS_CONN 1
2124 #define MDS_STATUS_LOV 2
2126 /* mdt_thread_info.mti_flags. */
2128 /* The flag indicates Size-on-MDS attributes are changed. */
2129 MF_SOM_CHANGE = (1 << 0),
2130 /* Flags indicates an epoch opens or closes. */
2131 MF_EPOCH_OPEN = (1 << 1),
2132 MF_EPOCH_CLOSE = (1 << 2),
2133 MF_MDC_CANCEL_FID1 = (1 << 3),
2134 MF_MDC_CANCEL_FID2 = (1 << 4),
2135 MF_MDC_CANCEL_FID3 = (1 << 5),
2136 MF_MDC_CANCEL_FID4 = (1 << 6),
2137 /* There is a pending attribute update. */
2138 MF_SOM_AU = (1 << 7),
2139 /* Cancel OST locks while getattr OST attributes. */
2140 MF_GETATTR_LOCK = (1 << 8),
2141 MF_GET_MDT_IDX = (1 << 9),
2144 #define MF_SOM_LOCAL_FLAGS (MF_SOM_CHANGE | MF_EPOCH_OPEN | MF_EPOCH_CLOSE)
2146 #define LUSTRE_BFLAG_UNCOMMITTED_WRITES 0x1
2148 /* these should be identical to their EXT4_*_FL counterparts, they are
2149 * redefined here only to avoid dragging in fs/ext4/ext4.h */
2150 #define LUSTRE_SYNC_FL 0x00000008 /* Synchronous updates */
2151 #define LUSTRE_IMMUTABLE_FL 0x00000010 /* Immutable file */
2152 #define LUSTRE_APPEND_FL 0x00000020 /* writes to file may only append */
2153 #define LUSTRE_NOATIME_FL 0x00000080 /* do not update atime */
2154 #define LUSTRE_DIRSYNC_FL 0x00010000 /* dirsync behaviour (dir only) */
2157 /* Convert wire LUSTRE_*_FL to corresponding client local VFS S_* values
2158 * for the client inode i_flags. The LUSTRE_*_FL are the Lustre wire
2159 * protocol equivalents of LDISKFS_*_FL values stored on disk, while
2160 * the S_* flags are kernel-internal values that change between kernel
2161 * versions. These flags are set/cleared via FSFILT_IOC_{GET,SET}_FLAGS.
2162 * See b=16526 for a full history. */
2163 static inline int ll_ext_to_inode_flags(int flags)
2165 return (((flags & LUSTRE_SYNC_FL) ? S_SYNC : 0) |
2166 ((flags & LUSTRE_NOATIME_FL) ? S_NOATIME : 0) |
2167 ((flags & LUSTRE_APPEND_FL) ? S_APPEND : 0) |
2168 #if defined(S_DIRSYNC)
2169 ((flags & LUSTRE_DIRSYNC_FL) ? S_DIRSYNC : 0) |
2171 ((flags & LUSTRE_IMMUTABLE_FL) ? S_IMMUTABLE : 0));
2174 static inline int ll_inode_to_ext_flags(int iflags)
2176 return (((iflags & S_SYNC) ? LUSTRE_SYNC_FL : 0) |
2177 ((iflags & S_NOATIME) ? LUSTRE_NOATIME_FL : 0) |
2178 ((iflags & S_APPEND) ? LUSTRE_APPEND_FL : 0) |
2179 #if defined(S_DIRSYNC)
2180 ((iflags & S_DIRSYNC) ? LUSTRE_DIRSYNC_FL : 0) |
2182 ((iflags & S_IMMUTABLE) ? LUSTRE_IMMUTABLE_FL : 0));
2189 struct lustre_handle handle;
2191 __u64 size; /* Offset, in the case of MDS_READPAGE */
2195 __u64 blocks; /* XID, in the case of MDS_READPAGE */
2197 __u64 unused1; /* was "ino" until 2.4.0 */
2204 __u32 flags; /* from vfs for pin/unpin, LUSTRE_BFLAG close */
2206 __u32 nlink; /* #bytes to read in the case of MDS_READPAGE */
2207 __u32 unused2; /* was "generation" until 2.4.0 */
2212 __u32 max_cookiesize;
2213 __u32 uid_h; /* high 32-bits of uid, for FUID */
2214 __u32 gid_h; /* high 32-bits of gid, for FUID */
2215 __u32 padding_5; /* also fix lustre_swab_mdt_body */
2223 extern void lustre_swab_mdt_body (struct mdt_body *b);
2225 struct mdt_ioepoch {
2226 struct lustre_handle handle;
2232 extern void lustre_swab_mdt_ioepoch (struct mdt_ioepoch *b);
2234 /* permissions for md_perm.mp_perm */
2236 CFS_SETUID_PERM = 0x01,
2237 CFS_SETGID_PERM = 0x02,
2238 CFS_SETGRP_PERM = 0x04,
2239 CFS_RMTACL_PERM = 0x08,
2240 CFS_RMTOWN_PERM = 0x10
2243 /* inode access permission for remote user, the inode info are omitted,
2244 * for client knows them. */
2245 struct mdt_remote_perm {
2252 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
2256 extern void lustre_swab_mdt_remote_perm(struct mdt_remote_perm *p);
2258 struct mdt_rec_setattr {
2268 __u32 sa_padding_1_h;
2269 struct lu_fid sa_fid;
2278 __u32 sa_attr_flags;
2280 __u32 sa_bias; /* some operation flags */
2286 extern void lustre_swab_mdt_rec_setattr (struct mdt_rec_setattr *sa);
2289 * Attribute flags used in mdt_rec_setattr::sa_valid.
2290 * The kernel's #defines for ATTR_* should not be used over the network
2291 * since the client and MDS may run different kernels (see bug 13828)
2292 * Therefore, we should only use MDS_ATTR_* attributes for sa_valid.
2294 #define MDS_ATTR_MODE 0x1ULL /* = 1 */
2295 #define MDS_ATTR_UID 0x2ULL /* = 2 */
2296 #define MDS_ATTR_GID 0x4ULL /* = 4 */
2297 #define MDS_ATTR_SIZE 0x8ULL /* = 8 */
2298 #define MDS_ATTR_ATIME 0x10ULL /* = 16 */
2299 #define MDS_ATTR_MTIME 0x20ULL /* = 32 */
2300 #define MDS_ATTR_CTIME 0x40ULL /* = 64 */
2301 #define MDS_ATTR_ATIME_SET 0x80ULL /* = 128 */
2302 #define MDS_ATTR_MTIME_SET 0x100ULL /* = 256 */
2303 #define MDS_ATTR_FORCE 0x200ULL /* = 512, Not a change, but a change it */
2304 #define MDS_ATTR_ATTR_FLAG 0x400ULL /* = 1024 */
2305 #define MDS_ATTR_KILL_SUID 0x800ULL /* = 2048 */
2306 #define MDS_ATTR_KILL_SGID 0x1000ULL /* = 4096 */
2307 #define MDS_ATTR_CTIME_SET 0x2000ULL /* = 8192 */
2308 #define MDS_ATTR_FROM_OPEN 0x4000ULL /* = 16384, called from open path, ie O_TRUNC */
2309 #define MDS_ATTR_BLOCKS 0x8000ULL /* = 32768 */
2312 #define FMODE_READ 00000001
2313 #define FMODE_WRITE 00000002
2316 #define MDS_FMODE_CLOSED 00000000
2317 #define MDS_FMODE_EXEC 00000004
2318 /* IO Epoch is opened on a closed file. */
2319 #define MDS_FMODE_EPOCH 01000000
2320 /* IO Epoch is opened on a file truncate. */
2321 #define MDS_FMODE_TRUNC 02000000
2322 /* Size-on-MDS Attribute Update is pending. */
2323 #define MDS_FMODE_SOM 04000000
2325 #define MDS_OPEN_CREATED 00000010
2326 #define MDS_OPEN_CROSS 00000020
2328 #define MDS_OPEN_CREAT 00000100
2329 #define MDS_OPEN_EXCL 00000200
2330 #define MDS_OPEN_TRUNC 00001000
2331 #define MDS_OPEN_APPEND 00002000
2332 #define MDS_OPEN_SYNC 00010000
2333 #define MDS_OPEN_DIRECTORY 00200000
2335 #define MDS_OPEN_BY_FID 040000000 /* open_by_fid for known object */
2336 #define MDS_OPEN_DELAY_CREATE 0100000000 /* delay initial object create */
2337 #define MDS_OPEN_OWNEROVERRIDE 0200000000 /* NFSD rw-reopen ro file for owner */
2338 #define MDS_OPEN_JOIN_FILE 0400000000 /* open for join file.
2339 * We do not support JOIN FILE
2340 * anymore, reserve this flags
2341 * just for preventing such bit
2344 #define MDS_OPEN_LOCK 04000000000 /* This open requires open lock */
2345 #define MDS_OPEN_HAS_EA 010000000000 /* specify object create pattern */
2346 #define MDS_OPEN_HAS_OBJS 020000000000 /* Just set the EA the obj exist */
2347 #define MDS_OPEN_NORESTORE 0100000000000ULL /* Do not restore file at open */
2348 #define MDS_OPEN_NEWSTRIPE 0200000000000ULL /* New stripe needed (restripe or
2350 #define MDS_OPEN_VOLATILE 0400000000000ULL /* File is volatile = created
2353 /* permission for create non-directory file */
2354 #define MAY_CREATE (1 << 7)
2355 /* permission for create directory file */
2356 #define MAY_LINK (1 << 8)
2357 /* permission for delete from the directory */
2358 #define MAY_UNLINK (1 << 9)
2359 /* source's permission for rename */
2360 #define MAY_RENAME_SRC (1 << 10)
2361 /* target's permission for rename */
2362 #define MAY_RENAME_TAR (1 << 11)
2363 /* part (parent's) VTX permission check */
2364 #define MAY_VTX_PART (1 << 12)
2365 /* full VTX permission check */
2366 #define MAY_VTX_FULL (1 << 13)
2367 /* lfs rgetfacl permission check */
2368 #define MAY_RGETFACL (1 << 14)
2371 MDS_CHECK_SPLIT = 1 << 0,
2372 MDS_CROSS_REF = 1 << 1,
2373 MDS_VTX_BYPASS = 1 << 2,
2374 MDS_PERM_BYPASS = 1 << 3,
2376 MDS_QUOTA_IGNORE = 1 << 5,
2377 MDS_CLOSE_CLEANUP = 1 << 6,
2378 MDS_KEEP_ORPHAN = 1 << 7,
2379 MDS_RECOV_OPEN = 1 << 8,
2380 MDS_DATA_MODIFIED = 1 << 9,
2381 MDS_CREATE_VOLATILE = 1 << 10,
2382 MDS_OWNEROVERRIDE = 1 << 11,
2385 /* instance of mdt_reint_rec */
2386 struct mdt_rec_create {
2394 __u32 cr_suppgid1_h;
2396 __u32 cr_suppgid2_h;
2397 struct lu_fid cr_fid1;
2398 struct lu_fid cr_fid2;
2399 struct lustre_handle cr_old_handle; /* handle in case of open replay */
2403 __u64 cr_padding_1; /* rr_blocks */
2406 /* use of helpers set/get_mrc_cr_flags() is needed to access
2407 * 64 bits cr_flags [cr_flags_l, cr_flags_h], this is done to
2408 * extend cr_flags size without breaking 1.8 compat */
2409 __u32 cr_flags_l; /* for use with open, low 32 bits */
2410 __u32 cr_flags_h; /* for use with open, high 32 bits */
2411 __u32 cr_umask; /* umask for create */
2412 __u32 cr_padding_4; /* rr_padding_4 */
2415 static inline void set_mrc_cr_flags(struct mdt_rec_create *mrc, __u64 flags)
2417 mrc->cr_flags_l = (__u32)(flags & 0xFFFFFFFFUll);
2418 mrc->cr_flags_h = (__u32)(flags >> 32);
2421 static inline __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
2423 return ((__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32));
2426 /* instance of mdt_reint_rec */
2427 struct mdt_rec_link {
2435 __u32 lk_suppgid1_h;
2437 __u32 lk_suppgid2_h;
2438 struct lu_fid lk_fid1;
2439 struct lu_fid lk_fid2;
2441 __u64 lk_padding_1; /* rr_atime */
2442 __u64 lk_padding_2; /* rr_ctime */
2443 __u64 lk_padding_3; /* rr_size */
2444 __u64 lk_padding_4; /* rr_blocks */
2446 __u32 lk_padding_5; /* rr_mode */
2447 __u32 lk_padding_6; /* rr_flags */
2448 __u32 lk_padding_7; /* rr_padding_2 */
2449 __u32 lk_padding_8; /* rr_padding_3 */
2450 __u32 lk_padding_9; /* rr_padding_4 */
2453 /* instance of mdt_reint_rec */
2454 struct mdt_rec_unlink {
2462 __u32 ul_suppgid1_h;
2464 __u32 ul_suppgid2_h;
2465 struct lu_fid ul_fid1;
2466 struct lu_fid ul_fid2;
2468 __u64 ul_padding_2; /* rr_atime */
2469 __u64 ul_padding_3; /* rr_ctime */
2470 __u64 ul_padding_4; /* rr_size */
2471 __u64 ul_padding_5; /* rr_blocks */
2474 __u32 ul_padding_6; /* rr_flags */
2475 __u32 ul_padding_7; /* rr_padding_2 */
2476 __u32 ul_padding_8; /* rr_padding_3 */
2477 __u32 ul_padding_9; /* rr_padding_4 */
2480 /* instance of mdt_reint_rec */
2481 struct mdt_rec_rename {
2489 __u32 rn_suppgid1_h;
2491 __u32 rn_suppgid2_h;
2492 struct lu_fid rn_fid1;
2493 struct lu_fid rn_fid2;
2495 __u64 rn_padding_1; /* rr_atime */
2496 __u64 rn_padding_2; /* rr_ctime */
2497 __u64 rn_padding_3; /* rr_size */
2498 __u64 rn_padding_4; /* rr_blocks */
2499 __u32 rn_bias; /* some operation flags */
2500 __u32 rn_mode; /* cross-ref rename has mode */
2501 __u32 rn_padding_5; /* rr_flags */
2502 __u32 rn_padding_6; /* rr_padding_2 */
2503 __u32 rn_padding_7; /* rr_padding_3 */
2504 __u32 rn_padding_8; /* rr_padding_4 */
2507 /* instance of mdt_reint_rec */
2508 struct mdt_rec_setxattr {
2516 __u32 sx_suppgid1_h;
2518 __u32 sx_suppgid2_h;
2519 struct lu_fid sx_fid;
2520 __u64 sx_padding_1; /* These three are rr_fid2 */
2525 __u64 sx_padding_5; /* rr_ctime */
2526 __u64 sx_padding_6; /* rr_size */
2527 __u64 sx_padding_7; /* rr_blocks */
2530 __u32 sx_padding_8; /* rr_flags */
2531 __u32 sx_padding_9; /* rr_padding_2 */
2532 __u32 sx_padding_10; /* rr_padding_3 */
2533 __u32 sx_padding_11; /* rr_padding_4 */
2537 * mdt_rec_reint is the template for all mdt_reint_xxx structures.
2538 * Do NOT change the size of various members, otherwise the value
2539 * will be broken in lustre_swab_mdt_rec_reint().
2541 * If you add new members in other mdt_reint_xxx structres and need to use the
2542 * rr_padding_x fields, then update lustre_swab_mdt_rec_reint() also.
2544 struct mdt_rec_reint {
2552 __u32 rr_suppgid1_h;
2554 __u32 rr_suppgid2_h;
2555 struct lu_fid rr_fid1;
2556 struct lu_fid rr_fid2;
2565 __u32 rr_padding_2; /* also fix lustre_swab_mdt_rec_reint */
2566 __u32 rr_padding_3; /* also fix lustre_swab_mdt_rec_reint */
2567 __u32 rr_padding_4; /* also fix lustre_swab_mdt_rec_reint */
2570 extern void lustre_swab_mdt_rec_reint(struct mdt_rec_reint *rr);
2573 __u32 ld_tgt_count; /* how many MDS's */
2574 __u32 ld_active_tgt_count; /* how many active */
2575 __u32 ld_default_stripe_count; /* how many objects are used */
2576 __u32 ld_pattern; /* default MEA_MAGIC_* */
2577 __u64 ld_default_hash_size;
2578 __u64 ld_padding_1; /* also fix lustre_swab_lmv_desc */
2579 __u32 ld_padding_2; /* also fix lustre_swab_lmv_desc */
2580 __u32 ld_qos_maxage; /* in second */
2581 __u32 ld_padding_3; /* also fix lustre_swab_lmv_desc */
2582 __u32 ld_padding_4; /* also fix lustre_swab_lmv_desc */
2583 struct obd_uuid ld_uuid;
2586 extern void lustre_swab_lmv_desc (struct lmv_desc *ld);
2588 /* TODO: lmv_stripe_md should contain mds capabilities for all slave fids */
2589 struct lmv_stripe_md {
2594 char mea_pool_name[LOV_MAXPOOLNAME];
2595 struct lu_fid mea_ids[0];
2598 extern void lustre_swab_lmv_stripe_md(struct lmv_stripe_md *mea);
2600 /* lmv structures */
2601 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
2602 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
2603 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
2605 #define MAX_HASH_SIZE_32 0x7fffffffUL
2606 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
2607 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
2612 FLD_FIRST_OPC = FLD_QUERY
2618 SEQ_FIRST_OPC = SEQ_QUERY
2622 SEQ_ALLOC_SUPER = 0,
2627 * LOV data structures
2630 #define LOV_MAX_UUID_BUFFER_SIZE 8192
2631 /* The size of the buffer the lov/mdc reserves for the
2632 * array of UUIDs returned by the MDS. With the current
2633 * protocol, this will limit the max number of OSTs per LOV */
2635 #define LOV_DESC_MAGIC 0xB0CCDE5C
2637 /* LOV settings descriptor (should only contain static info) */
2639 __u32 ld_tgt_count; /* how many OBD's */
2640 __u32 ld_active_tgt_count; /* how many active */
2641 __u32 ld_default_stripe_count; /* how many objects are used */
2642 __u32 ld_pattern; /* default PATTERN_RAID0 */
2643 __u64 ld_default_stripe_size; /* in bytes */
2644 __u64 ld_default_stripe_offset; /* in bytes */
2645 __u32 ld_padding_0; /* unused */
2646 __u32 ld_qos_maxage; /* in second */
2647 __u32 ld_padding_1; /* also fix lustre_swab_lov_desc */
2648 __u32 ld_padding_2; /* also fix lustre_swab_lov_desc */
2649 struct obd_uuid ld_uuid;
2652 #define ld_magic ld_active_tgt_count /* for swabbing from llogs */
2654 extern void lustre_swab_lov_desc (struct lov_desc *ld);
2659 /* opcodes -- MUST be distinct from OST/MDS opcodes */
2664 LDLM_BL_CALLBACK = 104,
2665 LDLM_CP_CALLBACK = 105,
2666 LDLM_GL_CALLBACK = 106,
2667 LDLM_SET_INFO = 107,
2670 #define LDLM_FIRST_OPC LDLM_ENQUEUE
2672 #define RES_NAME_SIZE 4
2673 struct ldlm_res_id {
2674 __u64 name[RES_NAME_SIZE];
2677 extern void lustre_swab_ldlm_res_id (struct ldlm_res_id *id);
2679 static inline int ldlm_res_eq(const struct ldlm_res_id *res0,
2680 const struct ldlm_res_id *res1)
2682 return !memcmp(res0, res1, sizeof(*res0));
2699 #define LCK_MODE_NUM 8
2709 #define LDLM_MIN_TYPE LDLM_PLAIN
2711 struct ldlm_extent {
2717 static inline int ldlm_extent_overlap(struct ldlm_extent *ex1,
2718 struct ldlm_extent *ex2)
2720 return (ex1->start <= ex2->end) && (ex2->start <= ex1->end);
2723 /* check if @ex1 contains @ex2 */
2724 static inline int ldlm_extent_contain(struct ldlm_extent *ex1,
2725 struct ldlm_extent *ex2)
2727 return (ex1->start <= ex2->start) && (ex1->end >= ex2->end);
2730 struct ldlm_inodebits {
2734 struct ldlm_flock_wire {
2742 /* it's important that the fields of the ldlm_extent structure match
2743 * the first fields of the ldlm_flock structure because there is only
2744 * one ldlm_swab routine to process the ldlm_policy_data_t union. if
2745 * this ever changes we will need to swab the union differently based
2746 * on the resource type. */
2749 struct ldlm_extent l_extent;
2750 struct ldlm_flock_wire l_flock;
2751 struct ldlm_inodebits l_inodebits;
2752 } ldlm_wire_policy_data_t;
2754 extern void lustre_swab_ldlm_policy_data (ldlm_wire_policy_data_t *d);
2756 union ldlm_gl_desc {
2757 struct ldlm_gl_lquota_desc lquota_desc;
2760 extern void lustre_swab_gl_desc(union ldlm_gl_desc *);
2762 struct ldlm_intent {
2766 extern void lustre_swab_ldlm_intent (struct ldlm_intent *i);
2768 struct ldlm_resource_desc {
2769 ldlm_type_t lr_type;
2770 __u32 lr_padding; /* also fix lustre_swab_ldlm_resource_desc */
2771 struct ldlm_res_id lr_name;
2774 extern void lustre_swab_ldlm_resource_desc (struct ldlm_resource_desc *r);
2776 struct ldlm_lock_desc {
2777 struct ldlm_resource_desc l_resource;
2778 ldlm_mode_t l_req_mode;
2779 ldlm_mode_t l_granted_mode;
2780 ldlm_wire_policy_data_t l_policy_data;
2783 extern void lustre_swab_ldlm_lock_desc (struct ldlm_lock_desc *l);
2785 #define LDLM_LOCKREQ_HANDLES 2
2786 #define LDLM_ENQUEUE_CANCEL_OFF 1
2788 struct ldlm_request {
2791 struct ldlm_lock_desc lock_desc;
2792 struct lustre_handle lock_handle[LDLM_LOCKREQ_HANDLES];
2795 extern void lustre_swab_ldlm_request (struct ldlm_request *rq);
2797 /* If LDLM_ENQUEUE, 1 slot is already occupied, 1 is available.
2798 * Otherwise, 2 are available. */
2799 #define ldlm_request_bufsize(count,type) \
2801 int _avail = LDLM_LOCKREQ_HANDLES; \
2802 _avail -= (type == LDLM_ENQUEUE ? LDLM_ENQUEUE_CANCEL_OFF : 0); \
2803 sizeof(struct ldlm_request) + \
2804 (count > _avail ? count - _avail : 0) * \
2805 sizeof(struct lustre_handle); \
2810 __u32 lock_padding; /* also fix lustre_swab_ldlm_reply */
2811 struct ldlm_lock_desc lock_desc;
2812 struct lustre_handle lock_handle;
2813 __u64 lock_policy_res1;
2814 __u64 lock_policy_res2;
2817 extern void lustre_swab_ldlm_reply (struct ldlm_reply *r);
2819 #define ldlm_flags_to_wire(flags) ((__u32)(flags))
2820 #define ldlm_flags_from_wire(flags) ((__u64)(flags))
2823 * Opcodes for mountconf (mgs and mgc)
2828 MGS_EXCEPTION, /* node died, etc. */
2829 MGS_TARGET_REG, /* whenever target starts up */
2835 #define MGS_FIRST_OPC MGS_CONNECT
2837 #define MGS_PARAM_MAXLEN 1024
2838 #define KEY_SET_INFO "set_info"
2840 struct mgs_send_param {
2841 char mgs_param[MGS_PARAM_MAXLEN];
2844 /* We pass this info to the MGS so it can write config logs */
2845 #define MTI_NAME_MAXLEN 64
2846 #define MTI_PARAM_MAXLEN 4096
2847 #define MTI_NIDS_MAX 32
2848 struct mgs_target_info {
2849 __u32 mti_lustre_ver;
2850 __u32 mti_stripe_index;
2851 __u32 mti_config_ver;
2853 __u32 mti_nid_count;
2854 __u32 mti_instance; /* Running instance of target */
2855 char mti_fsname[MTI_NAME_MAXLEN];
2856 char mti_svname[MTI_NAME_MAXLEN];
2857 char mti_uuid[sizeof(struct obd_uuid)];
2858 __u64 mti_nids[MTI_NIDS_MAX]; /* host nids (lnet_nid_t)*/
2859 char mti_params[MTI_PARAM_MAXLEN];
2861 extern void lustre_swab_mgs_target_info(struct mgs_target_info *oinfo);
2863 struct mgs_nidtbl_entry {
2864 __u64 mne_version; /* table version of this entry */
2865 __u32 mne_instance; /* target instance # */
2866 __u32 mne_index; /* target index */
2867 __u32 mne_length; /* length of this entry - by bytes */
2868 __u8 mne_type; /* target type LDD_F_SV_TYPE_OST/MDT */
2869 __u8 mne_nid_type; /* type of nid(mbz). for ipv6. */
2870 __u8 mne_nid_size; /* size of each NID, by bytes */
2871 __u8 mne_nid_count; /* # of NIDs in buffer */
2873 lnet_nid_t nids[0]; /* variable size buffer for NIDs. */
2876 extern void lustre_swab_mgs_nidtbl_entry(struct mgs_nidtbl_entry *oinfo);
2878 struct mgs_config_body {
2879 char mcb_name[MTI_NAME_MAXLEN]; /* logname */
2880 __u64 mcb_offset; /* next index of config log to request */
2881 __u16 mcb_type; /* type of log: CONFIG_T_[CONFIG|RECOVER] */
2883 __u8 mcb_bits; /* bits unit size of config log */
2884 __u32 mcb_units; /* # of units for bulk transfer */
2886 extern void lustre_swab_mgs_config_body(struct mgs_config_body *body);
2888 struct mgs_config_res {
2889 __u64 mcr_offset; /* index of last config log */
2890 __u64 mcr_size; /* size of the log */
2892 extern void lustre_swab_mgs_config_res(struct mgs_config_res *body);
2894 /* Config marker flags (in config log) */
2895 #define CM_START 0x01
2897 #define CM_SKIP 0x04
2898 #define CM_UPGRADE146 0x08
2899 #define CM_EXCLUDE 0x10
2900 #define CM_START_SKIP (CM_START | CM_SKIP)
2903 __u32 cm_step; /* aka config version */
2905 __u32 cm_vers; /* lustre release version number */
2906 __u32 cm_padding; /* 64 bit align */
2907 obd_time cm_createtime; /*when this record was first created */
2908 obd_time cm_canceltime; /*when this record is no longer valid*/
2909 char cm_tgtname[MTI_NAME_MAXLEN];
2910 char cm_comment[MTI_NAME_MAXLEN];
2913 extern void lustre_swab_cfg_marker(struct cfg_marker *marker,
2914 int swab, int size);
2917 * Opcodes for multiple servers.
2927 #define OBD_FIRST_OPC OBD_PING
2929 /* catalog of log objects */
2931 /** Identifier for a single log object */
2933 struct ost_id lgl_oi;
2935 } __attribute__((packed));
2937 /** Records written to the CATALOGS list */
2938 #define CATLIST "CATALOGS"
2940 struct llog_logid lci_logid;
2944 } __attribute__((packed));
2946 /* Log data record types - there is no specific reason that these need to
2947 * be related to the RPC opcodes, but no reason not to (may be handy later?)
2949 #define LLOG_OP_MAGIC 0x10600000
2950 #define LLOG_OP_MASK 0xfff00000
2953 LLOG_PAD_MAGIC = LLOG_OP_MAGIC | 0x00000,
2954 OST_SZ_REC = LLOG_OP_MAGIC | 0x00f00,
2955 /* OST_RAID1_REC = LLOG_OP_MAGIC | 0x01000, never used */
2956 MDS_UNLINK_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) |
2957 REINT_UNLINK, /* obsolete after 2.5.0 */
2958 MDS_UNLINK64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
2960 /* MDS_SETATTR_REC = LLOG_OP_MAGIC | 0x12401, obsolete 1.8.0 */
2961 MDS_SETATTR64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
2963 OBD_CFG_REC = LLOG_OP_MAGIC | 0x20000,
2964 /* PTL_CFG_REC = LLOG_OP_MAGIC | 0x30000, obsolete 1.4.0 */
2965 LLOG_GEN_REC = LLOG_OP_MAGIC | 0x40000,
2966 /* LLOG_JOIN_REC = LLOG_OP_MAGIC | 0x50000, obsolete 1.8.0 */
2967 CHANGELOG_REC = LLOG_OP_MAGIC | 0x60000,
2968 CHANGELOG_USER_REC = LLOG_OP_MAGIC | 0x70000,
2969 LLOG_HDR_MAGIC = LLOG_OP_MAGIC | 0x45539,
2970 LLOG_LOGID_MAGIC = LLOG_OP_MAGIC | 0x4553b,
2973 #define LLOG_REC_HDR_NEEDS_SWABBING(r) \
2974 (((r)->lrh_type & __swab32(LLOG_OP_MASK)) == __swab32(LLOG_OP_MAGIC))
2976 /** Log record header - stored in little endian order.
2977 * Each record must start with this struct, end with a llog_rec_tail,
2978 * and be a multiple of 256 bits in size.
2980 struct llog_rec_hdr {
2987 struct llog_rec_tail {
2992 /* Where data follow just after header */
2993 #define REC_DATA(ptr) \
2994 ((void *)((char *)ptr + sizeof(struct llog_rec_hdr)))
2996 #define REC_DATA_LEN(rec) \
2997 (rec->lrh_len - sizeof(struct llog_rec_hdr) - \
2998 sizeof(struct llog_rec_tail))
3000 struct llog_logid_rec {
3001 struct llog_rec_hdr lid_hdr;
3002 struct llog_logid lid_id;
3006 struct llog_rec_tail lid_tail;
3007 } __attribute__((packed));
3009 struct llog_unlink_rec {
3010 struct llog_rec_hdr lur_hdr;
3013 obd_count lur_count;
3014 struct llog_rec_tail lur_tail;
3015 } __attribute__((packed));
3017 struct llog_unlink64_rec {
3018 struct llog_rec_hdr lur_hdr;
3019 struct lu_fid lur_fid;
3020 obd_count lur_count; /* to destroy the lost precreated */
3024 struct llog_rec_tail lur_tail;
3025 } __attribute__((packed));
3027 struct llog_setattr64_rec {
3028 struct llog_rec_hdr lsr_hdr;
3029 struct ost_id lsr_oi;
3035 struct llog_rec_tail lsr_tail;
3036 } __attribute__((packed));
3038 struct llog_size_change_rec {
3039 struct llog_rec_hdr lsc_hdr;
3040 struct ll_fid lsc_fid;
3045 struct llog_rec_tail lsc_tail;
3046 } __attribute__((packed));
3048 #define CHANGELOG_MAGIC 0xca103000
3050 /** \a changelog_rec_type's that can't be masked */
3051 #define CHANGELOG_MINMASK (1 << CL_MARK)
3052 /** bits covering all \a changelog_rec_type's */
3053 #define CHANGELOG_ALLMASK 0XFFFFFFFF
3054 /** default \a changelog_rec_type mask */
3055 #define CHANGELOG_DEFMASK CHANGELOG_ALLMASK & ~(1 << CL_ATIME | 1 << CL_CLOSE)
3057 /* changelog llog name, needed by client replicators */
3058 #define CHANGELOG_CATALOG "changelog_catalog"
3060 struct changelog_setinfo {
3063 } __attribute__((packed));
3065 /** changelog record */
3066 struct llog_changelog_rec {
3067 struct llog_rec_hdr cr_hdr;
3068 struct changelog_rec cr;
3069 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
3070 } __attribute__((packed));
3072 struct llog_changelog_ext_rec {
3073 struct llog_rec_hdr cr_hdr;
3074 struct changelog_ext_rec cr;
3075 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
3076 } __attribute__((packed));
3078 #define CHANGELOG_USER_PREFIX "cl"
3080 struct llog_changelog_user_rec {
3081 struct llog_rec_hdr cur_hdr;
3085 struct llog_rec_tail cur_tail;
3086 } __attribute__((packed));
3088 /* Old llog gen for compatibility */
3092 } __attribute__((packed));
3094 struct llog_gen_rec {
3095 struct llog_rec_hdr lgr_hdr;
3096 struct llog_gen lgr_gen;
3100 struct llog_rec_tail lgr_tail;
3103 /* On-disk header structure of each log object, stored in little endian order */
3104 #define LLOG_CHUNK_SIZE 8192
3105 #define LLOG_HEADER_SIZE (96)
3106 #define LLOG_BITMAP_BYTES (LLOG_CHUNK_SIZE - LLOG_HEADER_SIZE)
3108 #define LLOG_MIN_REC_SIZE (24) /* round(llog_rec_hdr + llog_rec_tail) */
3110 /* flags for the logs */
3112 LLOG_F_ZAP_WHEN_EMPTY = 0x1,
3113 LLOG_F_IS_CAT = 0x2,
3114 LLOG_F_IS_PLAIN = 0x4,
3117 struct llog_log_hdr {
3118 struct llog_rec_hdr llh_hdr;
3119 obd_time llh_timestamp;
3121 __u32 llh_bitmap_offset;
3125 /* for a catalog the first plain slot is next to it */
3126 struct obd_uuid llh_tgtuuid;
3127 __u32 llh_reserved[LLOG_HEADER_SIZE/sizeof(__u32) - 23];
3128 __u32 llh_bitmap[LLOG_BITMAP_BYTES/sizeof(__u32)];
3129 struct llog_rec_tail llh_tail;
3130 } __attribute__((packed));
3132 #define LLOG_BITMAP_SIZE(llh) (__u32)((llh->llh_hdr.lrh_len - \
3133 llh->llh_bitmap_offset - \
3134 sizeof(llh->llh_tail)) * 8)
3136 /** log cookies are used to reference a specific log file and a record therein */
3137 struct llog_cookie {
3138 struct llog_logid lgc_lgl;
3142 } __attribute__((packed));
3144 /** llog protocol */
3145 enum llogd_rpc_ops {
3146 LLOG_ORIGIN_HANDLE_CREATE = 501,
3147 LLOG_ORIGIN_HANDLE_NEXT_BLOCK = 502,
3148 LLOG_ORIGIN_HANDLE_READ_HEADER = 503,
3149 LLOG_ORIGIN_HANDLE_WRITE_REC = 504,
3150 LLOG_ORIGIN_HANDLE_CLOSE = 505,
3151 LLOG_ORIGIN_CONNECT = 506,
3152 LLOG_CATINFO = 507, /* deprecated */
3153 LLOG_ORIGIN_HANDLE_PREV_BLOCK = 508,
3154 LLOG_ORIGIN_HANDLE_DESTROY = 509, /* for destroy llog object*/
3156 LLOG_FIRST_OPC = LLOG_ORIGIN_HANDLE_CREATE
3160 struct llog_logid lgd_logid;
3162 __u32 lgd_llh_flags;
3164 __u32 lgd_saved_index;
3166 __u64 lgd_cur_offset;
3167 } __attribute__((packed));
3169 struct llogd_conn_body {
3170 struct llog_gen lgdc_gen;
3171 struct llog_logid lgdc_logid;
3172 __u32 lgdc_ctxt_idx;
3173 } __attribute__((packed));
3175 /* Note: 64-bit types are 64-bit aligned in structure */
3177 obd_valid o_valid; /* hot fields in this obdo */
3179 obd_id o_parent_seq;
3180 obd_size o_size; /* o_size-o_blocks == ost_lvb */
3184 obd_blocks o_blocks; /* brw: cli sent cached bytes */
3187 /* 32-bit fields start here: keep an even number of them via padding */
3188 obd_blksize o_blksize; /* optimal IO blocksize */
3189 obd_mode o_mode; /* brw: cli sent cache remain */
3193 obd_count o_nlink; /* brw: checksum */
3194 obd_count o_parent_oid;
3195 obd_count o_misc; /* brw: o_dropped */
3197 __u64 o_ioepoch; /* epoch in ost writes */
3198 __u32 o_stripe_idx; /* holds stripe idx */
3200 struct lustre_handle o_handle; /* brw: lock handle to prolong
3202 struct llog_cookie o_lcookie; /* destroy: unlink cookie from
3207 __u64 o_data_version; /* getattr: sum of iversion for
3209 * brw: grant space consumed on
3210 * the client for the write */
3216 #define o_dirty o_blocks
3217 #define o_undirty o_mode
3218 #define o_dropped o_misc
3219 #define o_cksum o_nlink
3220 #define o_grant_used o_data_version
3222 static inline void lustre_set_wire_obdo(struct obdo *wobdo, struct obdo *lobdo)
3224 memcpy(wobdo, lobdo, sizeof(*lobdo));
3225 wobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
3228 static inline void lustre_get_wire_obdo(struct obdo *lobdo, struct obdo *wobdo)
3230 obd_flag local_flags = 0;
3232 if (lobdo->o_valid & OBD_MD_FLFLAGS)
3233 local_flags = lobdo->o_flags & OBD_FL_LOCAL_MASK;
3235 LASSERT(!(wobdo->o_flags & OBD_FL_LOCAL_MASK));
3237 memcpy(lobdo, wobdo, sizeof(*lobdo));
3238 if (local_flags != 0) {
3239 lobdo->o_valid |= OBD_MD_FLFLAGS;
3240 lobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
3241 lobdo->o_flags |= local_flags;
3245 extern void lustre_swab_obdo (struct obdo *o);
3247 /* request structure for OST's */
3252 /* Key for FIEMAP to be used in get_info calls */
3253 struct ll_fiemap_info_key {
3256 struct ll_user_fiemap fiemap;
3259 extern void lustre_swab_ost_body (struct ost_body *b);
3260 extern void lustre_swab_ost_last_id(obd_id *id);
3261 extern void lustre_swab_fiemap(struct ll_user_fiemap *fiemap);
3263 extern void lustre_swab_lov_user_md_v1(struct lov_user_md_v1 *lum);
3264 extern void lustre_swab_lov_user_md_v3(struct lov_user_md_v3 *lum);
3265 extern void lustre_swab_lov_user_md_objects(struct lov_user_ost_data *lod,
3267 extern void lustre_swab_lov_mds_md(struct lov_mds_md *lmm);
3270 extern void lustre_swab_llogd_body (struct llogd_body *d);
3271 extern void lustre_swab_llog_hdr (struct llog_log_hdr *h);
3272 extern void lustre_swab_llogd_conn_body (struct llogd_conn_body *d);
3273 extern void lustre_swab_llog_rec(struct llog_rec_hdr *rec);
3276 extern void lustre_swab_lustre_cfg(struct lustre_cfg *lcfg);
3278 /* Functions for dumping PTLRPC fields */
3279 void dump_rniobuf(struct niobuf_remote *rnb);
3280 void dump_ioo(struct obd_ioobj *nb);
3281 void dump_obdo(struct obdo *oa);
3282 void dump_ost_body(struct ost_body *ob);
3283 void dump_rcs(__u32 *rc);
3285 #define IDX_INFO_MAGIC 0x3D37CC37
3287 /* Index file transfer through the network. The server serializes the index into
3288 * a byte stream which is sent to the client via a bulk transfer */
3292 /* reply: see idx_info_flags below */
3295 /* request & reply: number of lu_idxpage (to be) transferred */
3299 /* request: requested attributes passed down to the iterator API */
3302 /* request & reply: index file identifier (FID) */
3303 struct lu_fid ii_fid;
3305 /* reply: version of the index file before starting to walk the index.
3306 * Please note that the version can be modified at any time during the
3310 /* request: hash to start with:
3311 * reply: hash of the first entry of the first lu_idxpage and hash
3312 * of the entry to read next if any */
3313 __u64 ii_hash_start;
3316 /* reply: size of keys in lu_idxpages, minimal one if II_FL_VARKEY is
3320 /* reply: size of records in lu_idxpages, minimal one if II_FL_VARREC
3328 extern void lustre_swab_idx_info(struct idx_info *ii);
3330 #define II_END_OFF MDS_DIR_END_OFF /* all entries have been read */
3332 /* List of flags used in idx_info::ii_flags */
3333 enum idx_info_flags {
3334 II_FL_NOHASH = 1 << 0, /* client doesn't care about hash value */
3335 II_FL_VARKEY = 1 << 1, /* keys can be of variable size */
3336 II_FL_VARREC = 1 << 2, /* records can be of variable size */
3337 II_FL_NONUNQ = 1 << 3, /* index supports non-unique keys */
3340 #define LIP_MAGIC 0x8A6D6B6C
3342 /* 4KB (= LU_PAGE_SIZE) container gathering key/record pairs */
3344 /* 16-byte header */
3347 __u16 lip_nr; /* number of entries in the container */
3348 __u64 lip_pad0; /* additional padding for future use */
3350 /* key/record pairs are stored in the remaining 4080 bytes.
3351 * depending upon the flags in idx_info::ii_flags, each key/record
3352 * pair might be preceded by:
3354 * - the key size (II_FL_VARKEY is set)
3355 * - the record size (II_FL_VARREC is set)
3357 * For the time being, we only support fixed-size key & record. */
3358 char lip_entries[0];
3360 extern void lustre_swab_lip_header(struct lu_idxpage *lip);
3362 #define LIP_HDR_SIZE (offsetof(struct lu_idxpage, lip_entries))
3364 /* Gather all possible type associated with a 4KB container */
3366 struct lu_dirpage lp_dir; /* for MDS_READPAGE */
3367 struct lu_idxpage lp_idx; /* for OBD_IDX_READ */
3368 char lp_array[LU_PAGE_SIZE];
3371 /* security opcodes */
3374 SEC_CTX_INIT_CONT = 802,
3377 SEC_FIRST_OPC = SEC_CTX_INIT
3381 * capa related definitions
3383 #define CAPA_HMAC_MAX_LEN 64
3384 #define CAPA_HMAC_KEY_MAX_LEN 56
3386 /* NB take care when changing the sequence of elements this struct,
3387 * because the offset info is used in find_capa() */
3388 struct lustre_capa {
3389 struct lu_fid lc_fid; /** fid */
3390 __u64 lc_opc; /** operations allowed */
3391 __u64 lc_uid; /** file owner */
3392 __u64 lc_gid; /** file group */
3393 __u32 lc_flags; /** HMAC algorithm & flags */
3394 __u32 lc_keyid; /** key# used for the capability */
3395 __u32 lc_timeout; /** capa timeout value (sec) */
3396 __u32 lc_expiry; /** expiry time (sec) */
3397 __u8 lc_hmac[CAPA_HMAC_MAX_LEN]; /** HMAC */
3398 } __attribute__((packed));
3400 extern void lustre_swab_lustre_capa(struct lustre_capa *c);
3402 /** lustre_capa::lc_opc */
3404 CAPA_OPC_BODY_WRITE = 1<<0, /**< write object data */
3405 CAPA_OPC_BODY_READ = 1<<1, /**< read object data */
3406 CAPA_OPC_INDEX_LOOKUP = 1<<2, /**< lookup object fid */
3407 CAPA_OPC_INDEX_INSERT = 1<<3, /**< insert object fid */
3408 CAPA_OPC_INDEX_DELETE = 1<<4, /**< delete object fid */
3409 CAPA_OPC_OSS_WRITE = 1<<5, /**< write oss object data */
3410 CAPA_OPC_OSS_READ = 1<<6, /**< read oss object data */
3411 CAPA_OPC_OSS_TRUNC = 1<<7, /**< truncate oss object */
3412 CAPA_OPC_OSS_DESTROY = 1<<8, /**< destroy oss object */
3413 CAPA_OPC_META_WRITE = 1<<9, /**< write object meta data */
3414 CAPA_OPC_META_READ = 1<<10, /**< read object meta data */
3417 #define CAPA_OPC_OSS_RW (CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE)
3418 #define CAPA_OPC_MDS_ONLY \
3419 (CAPA_OPC_BODY_WRITE | CAPA_OPC_BODY_READ | CAPA_OPC_INDEX_LOOKUP | \
3420 CAPA_OPC_INDEX_INSERT | CAPA_OPC_INDEX_DELETE)
3421 #define CAPA_OPC_OSS_ONLY \
3422 (CAPA_OPC_OSS_WRITE | CAPA_OPC_OSS_READ | CAPA_OPC_OSS_TRUNC | \
3423 CAPA_OPC_OSS_DESTROY)
3424 #define CAPA_OPC_MDS_DEFAULT ~CAPA_OPC_OSS_ONLY
3425 #define CAPA_OPC_OSS_DEFAULT ~(CAPA_OPC_MDS_ONLY | CAPA_OPC_OSS_ONLY)
3427 /* MDS capability covers object capability for operations of body r/w
3428 * (dir readpage/sendpage), index lookup/insert/delete and meta data r/w,
3429 * while OSS capability only covers object capability for operations of
3430 * oss data(file content) r/w/truncate.
3432 static inline int capa_for_mds(struct lustre_capa *c)
3434 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) != 0;
3437 static inline int capa_for_oss(struct lustre_capa *c)
3439 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) == 0;
3442 /* lustre_capa::lc_hmac_alg */
3444 CAPA_HMAC_ALG_SHA1 = 1, /**< sha1 algorithm */
3448 #define CAPA_FL_MASK 0x00ffffff
3449 #define CAPA_HMAC_ALG_MASK 0xff000000
3451 struct lustre_capa_key {
3452 __u64 lk_seq; /**< mds# */
3453 __u32 lk_keyid; /**< key# */
3455 __u8 lk_key[CAPA_HMAC_KEY_MAX_LEN]; /**< key */
3456 } __attribute__((packed));
3458 extern void lustre_swab_lustre_capa_key(struct lustre_capa_key *k);
3460 /** The link ea holds 1 \a link_ea_entry for each hardlink */
3461 #define LINK_EA_MAGIC 0x11EAF1DFUL
3462 struct link_ea_header {
3465 __u64 leh_len; /* total size */
3471 /** Hardlink data is name and parent fid.
3472 * Stored in this crazy struct for maximum packing and endian-neutrality
3474 struct link_ea_entry {
3475 /** __u16 stored big-endian, unaligned */
3476 unsigned char lee_reclen[2];
3477 unsigned char lee_parent_fid[sizeof(struct lu_fid)];
3479 }__attribute__((packed));
3481 /** fid2path request/reply structure */
3482 struct getinfo_fid2path {
3483 struct lu_fid gf_fid;
3488 } __attribute__((packed));
3490 void lustre_swab_fid2path (struct getinfo_fid2path *gf);
3493 LAYOUT_INTENT_ACCESS = 0,
3494 LAYOUT_INTENT_READ = 1,
3495 LAYOUT_INTENT_WRITE = 2,
3496 LAYOUT_INTENT_GLIMPSE = 3,
3497 LAYOUT_INTENT_TRUNC = 4,
3498 LAYOUT_INTENT_RELEASE = 5,
3499 LAYOUT_INTENT_RESTORE = 6
3502 /* enqueue layout lock with intent */
3503 struct layout_intent {
3504 __u32 li_opc; /* intent operation for enqueue, read, write etc */
3510 void lustre_swab_layout_intent(struct layout_intent *li);
3513 * On the wire version of hsm_progress structure.
3515 * Contains the userspace hsm_progress and some internal fields.
3517 struct hsm_progress_kernel {
3518 /* Field taken from struct hsm_progress */
3521 struct hsm_extent hpk_extent;
3523 __u16 hpk_errval; /* positive val */
3525 /* Additional fields */
3526 __u64 hpk_data_version;
3528 } __attribute__((packed));
3530 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
3531 extern void lustre_swab_hsm_current_action(struct hsm_current_action *action);
3532 extern void lustre_swab_hsm_progress_kernel(struct hsm_progress_kernel *hpk);
3533 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
3534 extern void lustre_swab_hsm_user_item(struct hsm_user_item *hui);
3535 extern void lustre_swab_hsm_request(struct hsm_request *hr);
3538 * These are object update opcode under UPDATE_OBJ, which is currently
3539 * being used by cross-ref operations between MDT.
3541 * During the cross-ref operation, the Master MDT, which the client send the
3542 * request to, will disassembly the operation into object updates, then OSP
3543 * will send these updates to the remote MDT to be executed.
3545 * Update request format
3546 * magic: UPDATE_BUFFER_MAGIC_V1
3547 * Count: How many updates in the req.
3548 * bufs[0] : following are packets of object.
3550 * type: object_update_op, the op code of update
3551 * fid: The object fid of the update.
3552 * lens/bufs: other parameters of the update.
3554 * type: object_update_op, the op code of update
3555 * fid: The object fid of the update.
3556 * lens/bufs: other parameters of the update.
3558 * update[7]: type: object_update_op, the op code of update
3559 * fid: The object fid of the update.
3560 * lens/bufs: other parameters of the update.
3561 * Current 8 maxim updates per object update request.
3563 *******************************************************************
3564 * update reply format:
3566 * ur_version: UPDATE_REPLY_V1
3567 * ur_count: The count of the reply, which is usually equal
3568 * to the number of updates in the request.
3569 * ur_lens: The reply lengths of each object update.
3571 * replies: 1st update reply [4bytes_ret: other body]
3572 * 2nd update reply [4bytes_ret: other body]
3574 * nth update reply [4bytes_ret: other body]
3576 * For each reply of the update, the format would be
3577 * result(4 bytes):Other stuff
3580 #define UPDATE_MAX_OPS 10
3581 #define UPDATE_BUFFER_MAGIC_V1 0xBDDE0001
3582 #define UPDATE_BUFFER_MAGIC UPDATE_BUFFER_MAGIC_V1
3583 #define UPDATE_BUF_COUNT 8
3584 enum object_update_op {
3593 OBJ_INDEX_LOOKUP = 9,
3594 OBJ_INDEX_INSERT = 10,
3595 OBJ_INDEX_DELETE = 11,
3602 struct lu_fid u_fid;
3603 __u32 u_lens[UPDATE_BUF_COUNT];
3613 #define UPDATE_REPLY_V1 0x00BD0001
3614 struct update_reply {
3620 void lustre_swab_update_buf(struct update_buf *ub);
3621 void lustre_swab_update_reply_buf(struct update_reply *ur);
3623 /** layout swap request structure
3624 * fid1 and fid2 are in mdt_body
3626 struct mdc_swap_layouts {
3630 void lustre_swab_swap_layouts(struct mdc_swap_layouts *msl);