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, 2014, 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 #include <libcfs/libcfs.h> /* for LPUX64, etc */
95 #include <lnet/types.h>
96 #include <lustre/lustre_user.h> /* Defn's shared with user-space. */
97 #include <lustre/lustre_errno.h>
98 #include <lustre_ver.h>
103 /* FOO_REQUEST_PORTAL is for incoming requests on the FOO
104 * FOO_REPLY_PORTAL is for incoming replies on the FOO
105 * FOO_BULK_PORTAL is for incoming bulk on the FOO
108 #define CONNMGR_REQUEST_PORTAL 1
109 #define CONNMGR_REPLY_PORTAL 2
110 //#define OSC_REQUEST_PORTAL 3
111 #define OSC_REPLY_PORTAL 4
112 //#define OSC_BULK_PORTAL 5
113 #define OST_IO_PORTAL 6
114 #define OST_CREATE_PORTAL 7
115 #define OST_BULK_PORTAL 8
116 //#define MDC_REQUEST_PORTAL 9
117 #define MDC_REPLY_PORTAL 10
118 //#define MDC_BULK_PORTAL 11
119 #define MDS_REQUEST_PORTAL 12
120 //#define MDS_REPLY_PORTAL 13
121 #define MDS_BULK_PORTAL 14
122 #define LDLM_CB_REQUEST_PORTAL 15
123 #define LDLM_CB_REPLY_PORTAL 16
124 #define LDLM_CANCEL_REQUEST_PORTAL 17
125 #define LDLM_CANCEL_REPLY_PORTAL 18
126 //#define PTLBD_REQUEST_PORTAL 19
127 //#define PTLBD_REPLY_PORTAL 20
128 //#define PTLBD_BULK_PORTAL 21
129 #define MDS_SETATTR_PORTAL 22
130 #define MDS_READPAGE_PORTAL 23
131 #define OUT_PORTAL 24
132 #define MGC_REPLY_PORTAL 25
133 #define MGS_REQUEST_PORTAL 26
134 #define MGS_REPLY_PORTAL 27
135 #define OST_REQUEST_PORTAL 28
136 #define FLD_REQUEST_PORTAL 29
137 #define SEQ_METADATA_PORTAL 30
138 #define SEQ_DATA_PORTAL 31
139 #define SEQ_CONTROLLER_PORTAL 32
140 #define MGS_BULK_PORTAL 33
142 /* Portal 63 is reserved for the Cray Inc DVS - nic@cray.com, roe@cray.com, n8851@cray.com */
145 #define PTL_RPC_MSG_REQUEST 4711
146 #define PTL_RPC_MSG_ERR 4712
147 #define PTL_RPC_MSG_REPLY 4713
149 /* DON'T use swabbed values of MAGIC as magic! */
150 #define LUSTRE_MSG_MAGIC_V1 0x0BD00BD0
151 #define LUSTRE_MSG_MAGIC_V2 0x0BD00BD3
153 #define LUSTRE_MSG_MAGIC_V1_SWABBED 0xD00BD00B
154 #define LUSTRE_MSG_MAGIC_V2_SWABBED 0xD30BD00B
156 #define LUSTRE_MSG_MAGIC LUSTRE_MSG_MAGIC_V2
158 #define PTLRPC_MSG_VERSION 0x00000003
159 #define LUSTRE_VERSION_MASK 0xffff0000
160 #define LUSTRE_OBD_VERSION 0x00010000
161 #define LUSTRE_MDS_VERSION 0x00020000
162 #define LUSTRE_OST_VERSION 0x00030000
163 #define LUSTRE_DLM_VERSION 0x00040000
164 #define LUSTRE_LOG_VERSION 0x00050000
165 #define LUSTRE_MGS_VERSION 0x00060000
167 /* TODO: All obd_* typedefs will be removed in last patch in series */
168 typedef __u64 obd_id;
169 typedef __u64 obd_seq;
170 typedef __s64 obd_time;
171 typedef __u64 obd_size;
172 typedef __u64 obd_off;
173 typedef __u64 obd_blocks;
174 typedef __u64 obd_valid;
175 typedef __u32 obd_blksize;
176 typedef __u32 obd_mode;
177 typedef __u32 obd_uid;
178 typedef __u32 obd_gid;
179 typedef __u32 obd_flag;
180 typedef __u32 obd_count;
183 * Describes a range of sequence, lsr_start is included but lsr_end is
185 * Same structure is used in fld module where lsr_index field holds mdt id
188 struct lu_seq_range {
195 struct lu_seq_range_array {
198 struct lu_seq_range lsra_lsr[0];
201 #define LU_SEQ_RANGE_MDT 0x0
202 #define LU_SEQ_RANGE_OST 0x1
203 #define LU_SEQ_RANGE_ANY 0x3
205 #define LU_SEQ_RANGE_MASK 0x3
207 static inline unsigned fld_range_type(const struct lu_seq_range *range)
209 return range->lsr_flags & LU_SEQ_RANGE_MASK;
212 static inline bool fld_range_is_ost(const struct lu_seq_range *range)
214 return fld_range_type(range) == LU_SEQ_RANGE_OST;
217 static inline bool fld_range_is_mdt(const struct lu_seq_range *range)
219 return fld_range_type(range) == LU_SEQ_RANGE_MDT;
223 * This all range is only being used when fld client sends fld query request,
224 * but it does not know whether the seq is MDT or OST, so it will send req
225 * with ALL type, which means either seq type gotten from lookup can be
228 static inline unsigned fld_range_is_any(const struct lu_seq_range *range)
230 return fld_range_type(range) == LU_SEQ_RANGE_ANY;
233 static inline void fld_range_set_type(struct lu_seq_range *range,
236 range->lsr_flags |= flags;
239 static inline void fld_range_set_mdt(struct lu_seq_range *range)
241 fld_range_set_type(range, LU_SEQ_RANGE_MDT);
244 static inline void fld_range_set_ost(struct lu_seq_range *range)
246 fld_range_set_type(range, LU_SEQ_RANGE_OST);
249 static inline void fld_range_set_any(struct lu_seq_range *range)
251 fld_range_set_type(range, LU_SEQ_RANGE_ANY);
255 * returns width of given range \a r
258 static inline __u64 range_space(const struct lu_seq_range *range)
260 return range->lsr_end - range->lsr_start;
264 * initialize range to zero
267 static inline void range_init(struct lu_seq_range *range)
269 memset(range, 0, sizeof(*range));
273 * check if given seq id \a s is within given range \a r
276 static inline bool range_within(const struct lu_seq_range *range,
279 return s >= range->lsr_start && s < range->lsr_end;
282 static inline bool range_is_sane(const struct lu_seq_range *range)
284 return range->lsr_end >= range->lsr_start;
287 static inline bool range_is_zero(const struct lu_seq_range *range)
289 return range->lsr_start == 0 && range->lsr_end == 0;
292 static inline bool range_is_exhausted(const struct lu_seq_range *range)
294 return range_space(range) == 0;
297 /* return 0 if two range have the same location */
298 static inline int range_compare_loc(const struct lu_seq_range *r1,
299 const struct lu_seq_range *r2)
301 return r1->lsr_index != r2->lsr_index ||
302 r1->lsr_flags != r2->lsr_flags;
305 #define DRANGE "[%#16.16"LPF64"x-%#16.16"LPF64"x):%x:%s"
307 #define PRANGE(range) \
308 (range)->lsr_start, \
310 (range)->lsr_index, \
311 fld_range_is_mdt(range) ? "mdt" : "ost"
314 /** \defgroup lu_fid lu_fid
318 * Flags for lustre_mdt_attrs::lma_compat and lustre_mdt_attrs::lma_incompat.
319 * Deprecated since HSM and SOM attributes are now stored in separate on-disk
323 LMAC_HSM = 0x00000001,
324 LMAC_SOM = 0x00000002,
325 LMAC_NOT_IN_OI = 0x00000004, /* the object does NOT need OI mapping */
326 LMAC_FID_ON_OST = 0x00000008, /* For OST-object, its OI mapping is
327 * under /O/<seq>/d<x>. */
331 * Masks for all features that should be supported by a Lustre version to
332 * access a specific file.
333 * This information is stored in lustre_mdt_attrs::lma_incompat.
336 LMAI_RELEASED = 0x00000001, /* file is released */
337 LMAI_AGENT = 0x00000002, /* agent inode */
338 LMAI_REMOTE_PARENT = 0x00000004, /* the parent of the object
339 is on the remote MDT */
340 LMAI_STRIPED = 0x00000008, /* striped directory inode */
342 #define LMA_INCOMPAT_SUPP (LMAI_AGENT | LMAI_REMOTE_PARENT | LMAI_STRIPED)
344 extern void lustre_lma_swab(struct lustre_mdt_attrs *lma);
345 extern void lustre_lma_init(struct lustre_mdt_attrs *lma,
346 const struct lu_fid *fid,
347 __u32 compat, __u32 incompat);
349 * SOM on-disk attributes stored in a separate xattr.
352 /** Bitfield for supported data in this structure. For future use. */
355 /** Incompat feature list. The supported feature mask is availabe in
356 * SOM_INCOMPAT_SUPP */
359 /** IO Epoch SOM attributes belongs to */
361 /** total file size in objects */
363 /** total fs blocks in objects */
365 /** mds mount id the size is valid for */
368 extern void lustre_som_swab(struct som_attrs *attrs);
370 #define SOM_INCOMPAT_SUPP 0x0
372 /* copytool uses a 32b bitmask field to encode archive-Ids during register
374 * archive num = 0 => all
375 * archive num from 1 to 32
377 #define LL_HSM_MAX_ARCHIVE (sizeof(__u32) * 8)
380 * HSM on-disk attributes stored in a separate xattr.
383 /** Bitfield for supported data in this structure. For future use. */
386 /** HSM flags, see hsm_flags enum below */
388 /** backend archive id associated with the file */
390 /** version associated with the last archiving, if any */
393 extern void lustre_hsm_swab(struct hsm_attrs *attrs);
399 /** LASTID file has zero OID */
400 LUSTRE_FID_LASTID_OID = 0UL,
401 /** initial fid id value */
402 LUSTRE_FID_INIT_OID = 1UL
405 /** returns fid object sequence */
406 static inline __u64 fid_seq(const struct lu_fid *fid)
411 /** returns fid object id */
412 static inline __u32 fid_oid(const struct lu_fid *fid)
417 /** returns fid object version */
418 static inline __u32 fid_ver(const struct lu_fid *fid)
423 static inline void fid_zero(struct lu_fid *fid)
425 memset(fid, 0, sizeof(*fid));
428 static inline __u64 fid_ver_oid(const struct lu_fid *fid)
430 return ((__u64)fid_ver(fid) << 32 | fid_oid(fid));
434 * Note that reserved SEQ numbers below 12 will conflict with ldiskfs
435 * inodes in the IGIF namespace, so these reserved SEQ numbers can be
436 * used for other purposes and not risk collisions with existing inodes.
438 * Different FID Format
439 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs#NEW.0
442 FID_SEQ_OST_MDT0 = 0,
443 FID_SEQ_LLOG = 1, /* unnamed llogs */
445 FID_SEQ_OST_MDT1 = 3,
446 FID_SEQ_OST_MAX = 9, /* Max MDT count before OST_on_FID */
447 FID_SEQ_LLOG_NAME = 10, /* named llogs */
450 FID_SEQ_IGIF_MAX = 0x0ffffffffULL,
451 FID_SEQ_IDIF = 0x100000000ULL,
452 FID_SEQ_IDIF_MAX = 0x1ffffffffULL,
453 /* Normal FID sequence starts from this value, i.e. 1<<33 */
454 FID_SEQ_START = 0x200000000ULL,
455 /* sequence for local pre-defined FIDs listed in local_oid */
456 FID_SEQ_LOCAL_FILE = 0x200000001ULL,
457 FID_SEQ_DOT_LUSTRE = 0x200000002ULL,
458 /* sequence is used for local named objects FIDs generated
459 * by local_object_storage library */
460 FID_SEQ_LOCAL_NAME = 0x200000003ULL,
461 /* Because current FLD will only cache the fid sequence, instead
462 * of oid on the client side, if the FID needs to be exposed to
463 * clients sides, it needs to make sure all of fids under one
464 * sequence will be located in one MDT. */
465 FID_SEQ_SPECIAL = 0x200000004ULL,
466 FID_SEQ_QUOTA = 0x200000005ULL,
467 FID_SEQ_QUOTA_GLB = 0x200000006ULL,
468 FID_SEQ_ROOT = 0x200000007ULL, /* Located on MDT0 */
469 FID_SEQ_LAYOUT_RBTREE = 0x200000008ULL,
470 FID_SEQ_NORMAL = 0x200000400ULL,
471 FID_SEQ_LOV_DEFAULT = 0xffffffffffffffffULL
474 #define OBIF_OID_MAX_BITS 32
475 #define OBIF_MAX_OID (1ULL << OBIF_OID_MAX_BITS)
476 #define OBIF_OID_MASK ((1ULL << OBIF_OID_MAX_BITS) - 1)
477 #define IDIF_OID_MAX_BITS 48
478 #define IDIF_MAX_OID (1ULL << IDIF_OID_MAX_BITS)
479 #define IDIF_OID_MASK ((1ULL << IDIF_OID_MAX_BITS) - 1)
481 /** OID for FID_SEQ_SPECIAL */
483 /* Big Filesystem Lock to serialize rename operations */
484 FID_OID_SPECIAL_BFL = 1UL,
487 /** OID for FID_SEQ_DOT_LUSTRE */
488 enum dot_lustre_oid {
489 FID_OID_DOT_LUSTRE = 1UL,
490 FID_OID_DOT_LUSTRE_OBF = 2UL,
491 FID_OID_DOT_LUSTRE_LPF = 3UL,
494 /** OID for FID_SEQ_ROOT */
497 FID_OID_ECHO_ROOT = 2UL,
500 static inline bool fid_seq_is_mdt0(__u64 seq)
502 return seq == FID_SEQ_OST_MDT0;
505 static inline bool fid_seq_is_mdt(__u64 seq)
507 return seq == FID_SEQ_OST_MDT0 || seq >= FID_SEQ_NORMAL;
510 static inline bool fid_seq_is_echo(__u64 seq)
512 return seq == FID_SEQ_ECHO;
515 static inline bool fid_is_echo(const struct lu_fid *fid)
517 return fid_seq_is_echo(fid_seq(fid));
520 static inline bool fid_seq_is_llog(__u64 seq)
522 return seq == FID_SEQ_LLOG;
525 static inline bool fid_is_llog(const struct lu_fid *fid)
527 /* file with OID == 0 is not llog but contains last oid */
528 return fid_seq_is_llog(fid_seq(fid)) && fid_oid(fid) > 0;
531 static inline bool fid_seq_is_rsvd(__u64 seq)
533 return seq > FID_SEQ_OST_MDT0 && seq <= FID_SEQ_RSVD;
536 static inline bool fid_seq_is_special(__u64 seq)
538 return seq == FID_SEQ_SPECIAL;
541 static inline bool fid_seq_is_local_file(__u64 seq)
543 return seq == FID_SEQ_LOCAL_FILE ||
544 seq == FID_SEQ_LOCAL_NAME;
547 static inline bool fid_seq_is_root(__u64 seq)
549 return seq == FID_SEQ_ROOT;
552 static inline bool fid_seq_is_dot(__u64 seq)
554 return seq == FID_SEQ_DOT_LUSTRE;
557 static inline bool fid_seq_is_default(__u64 seq)
559 return seq == FID_SEQ_LOV_DEFAULT;
562 static inline bool fid_is_mdt0(const struct lu_fid *fid)
564 return fid_seq_is_mdt0(fid_seq(fid));
567 static inline void lu_root_fid(struct lu_fid *fid)
569 fid->f_seq = FID_SEQ_ROOT;
570 fid->f_oid = FID_OID_ROOT;
574 static inline void lu_echo_root_fid(struct lu_fid *fid)
576 fid->f_seq = FID_SEQ_ROOT;
577 fid->f_oid = FID_OID_ECHO_ROOT;
582 * Check if a fid is igif or not.
583 * \param fid the fid to be tested.
584 * \return true if the fid is a igif; otherwise false.
586 static inline bool fid_seq_is_igif(__u64 seq)
588 return seq >= FID_SEQ_IGIF && seq <= FID_SEQ_IGIF_MAX;
591 static inline bool fid_is_igif(const struct lu_fid *fid)
593 return fid_seq_is_igif(fid_seq(fid));
597 * Check if a fid is idif or not.
598 * \param fid the fid to be tested.
599 * \return true if the fid is a idif; otherwise false.
601 static inline bool fid_seq_is_idif(__u64 seq)
603 return seq >= FID_SEQ_IDIF && seq <= FID_SEQ_IDIF_MAX;
606 static inline bool fid_is_idif(const struct lu_fid *fid)
608 return fid_seq_is_idif(fid_seq(fid));
611 static inline bool fid_is_local_file(const struct lu_fid *fid)
613 return fid_seq_is_local_file(fid_seq(fid));
616 static inline bool fid_seq_is_norm(__u64 seq)
618 return (seq >= FID_SEQ_NORMAL);
621 static inline bool fid_is_norm(const struct lu_fid *fid)
623 return fid_seq_is_norm(fid_seq(fid));
626 static inline int fid_is_layout_rbtree(const struct lu_fid *fid)
628 return fid_seq(fid) == FID_SEQ_LAYOUT_RBTREE;
631 /* convert an OST objid into an IDIF FID SEQ number */
632 static inline __u64 fid_idif_seq(__u64 id, __u32 ost_idx)
634 return FID_SEQ_IDIF | (ost_idx << 16) | ((id >> 32) & 0xffff);
637 /* convert a packed IDIF FID into an OST objid */
638 static inline __u64 fid_idif_id(__u64 seq, __u32 oid, __u32 ver)
640 return ((__u64)ver << 48) | ((seq & 0xffff) << 32) | oid;
643 static inline __u32 idif_ost_idx(__u64 seq)
645 return (seq >> 16) & 0xffff;
648 /* extract ost index from IDIF FID */
649 static inline __u32 fid_idif_ost_idx(const struct lu_fid *fid)
651 return idif_ost_idx(fid_seq(fid));
654 /* extract OST sequence (group) from a wire ost_id (id/seq) pair */
655 static inline __u64 ostid_seq(const struct ost_id *ostid)
657 if (fid_seq_is_mdt0(ostid->oi.oi_seq))
658 return FID_SEQ_OST_MDT0;
660 if (unlikely(fid_seq_is_default(ostid->oi.oi_seq)))
661 return FID_SEQ_LOV_DEFAULT;
663 if (fid_is_idif(&ostid->oi_fid))
664 return FID_SEQ_OST_MDT0;
666 return fid_seq(&ostid->oi_fid);
669 /* extract OST objid from a wire ost_id (id/seq) pair */
670 static inline __u64 ostid_id(const struct ost_id *ostid)
672 if (fid_seq_is_mdt0(ostid->oi.oi_seq))
673 return ostid->oi.oi_id & IDIF_OID_MASK;
675 if (unlikely(fid_seq_is_default(ostid->oi.oi_seq)))
676 return ostid->oi.oi_id;
678 if (fid_is_idif(&ostid->oi_fid))
679 return fid_idif_id(fid_seq(&ostid->oi_fid),
680 fid_oid(&ostid->oi_fid), 0);
682 return fid_oid(&ostid->oi_fid);
685 static inline void ostid_set_seq(struct ost_id *oi, __u64 seq)
687 if (fid_seq_is_mdt0(seq) || fid_seq_is_default(seq)) {
690 oi->oi_fid.f_seq = seq;
691 /* Note: if f_oid + f_ver is zero, we need init it
692 * to be 1, otherwise, ostid_seq will treat this
693 * as old ostid (oi_seq == 0) */
694 if (oi->oi_fid.f_oid == 0 && oi->oi_fid.f_ver == 0)
695 oi->oi_fid.f_oid = LUSTRE_FID_INIT_OID;
699 static inline void ostid_set_seq_mdt0(struct ost_id *oi)
701 ostid_set_seq(oi, FID_SEQ_OST_MDT0);
704 static inline void ostid_set_seq_echo(struct ost_id *oi)
706 ostid_set_seq(oi, FID_SEQ_ECHO);
709 static inline void ostid_set_seq_llog(struct ost_id *oi)
711 ostid_set_seq(oi, FID_SEQ_LLOG);
715 * Note: we need check oi_seq to decide where to set oi_id,
716 * so oi_seq should always be set ahead of oi_id.
718 static inline void ostid_set_id(struct ost_id *oi, __u64 oid)
720 if (fid_seq_is_mdt0(oi->oi.oi_seq)) {
721 if (oid >= IDIF_MAX_OID) {
722 CERROR("Bad "LPU64" to set "DOSTID"\n",
727 } else if (fid_is_idif(&oi->oi_fid)) {
728 if (oid >= IDIF_MAX_OID) {
729 CERROR("Bad "LPU64" to set "DOSTID"\n",
733 oi->oi_fid.f_seq = fid_idif_seq(oid,
734 fid_idif_ost_idx(&oi->oi_fid));
735 oi->oi_fid.f_oid = oid;
736 oi->oi_fid.f_ver = oid >> 48;
738 if (oid > OBIF_MAX_OID) {
739 CERROR("Bad "LPU64" to set "DOSTID"\n",
743 oi->oi_fid.f_oid = oid;
747 static inline int fid_set_id(struct lu_fid *fid, __u64 oid)
749 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
750 CERROR("bad IGIF, "DFID"\n", PFID(fid));
754 if (fid_is_idif(fid)) {
755 if (oid >= IDIF_MAX_OID) {
756 CERROR("Bad "LPU64" to set "DFID"\n",
760 fid->f_seq = fid_idif_seq(oid, fid_idif_ost_idx(fid));
762 fid->f_ver = oid >> 48;
764 if (oid > OBIF_MAX_OID) {
765 CERROR("Bad "LPU64" to set "DFID"\n",
775 * Unpack an OST object id/seq (group) into a FID. This is needed for
776 * converting all obdo, lmm, lsm, etc. 64-bit id/seq pairs into proper
777 * FIDs. Note that if an id/seq is already in FID/IDIF format it will
778 * be passed through unchanged. Only legacy OST objects in "group 0"
779 * will be mapped into the IDIF namespace so that they can fit into the
780 * struct lu_fid fields without loss. For reference see:
781 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs
783 static inline int ostid_to_fid(struct lu_fid *fid, const struct ost_id *ostid,
786 __u64 seq = ostid_seq(ostid);
788 if (ost_idx > 0xffff) {
789 CERROR("bad ost_idx, "DOSTID" ost_idx:%u\n", POSTID(ostid),
794 if (fid_seq_is_mdt0(seq)) {
795 __u64 oid = ostid_id(ostid);
797 /* This is a "legacy" (old 1.x/2.early) OST object in "group 0"
798 * that we map into the IDIF namespace. It allows up to 2^48
799 * objects per OST, as this is the object namespace that has
800 * been in production for years. This can handle create rates
801 * of 1M objects/s/OST for 9 years, or combinations thereof. */
802 if (oid >= IDIF_MAX_OID) {
803 CERROR("bad MDT0 id(1), "DOSTID" ost_idx:%u\n",
804 POSTID(ostid), ost_idx);
807 fid->f_seq = fid_idif_seq(oid, ost_idx);
808 /* truncate to 32 bits by assignment */
810 /* in theory, not currently used */
811 fid->f_ver = oid >> 48;
812 } else if (likely(!fid_seq_is_default(seq)))
813 /* if (fid_seq_is_idif(seq) || fid_seq_is_norm(seq)) */ {
814 /* This is either an IDIF object, which identifies objects across
815 * all OSTs, or a regular FID. The IDIF namespace maps legacy
816 * OST objects into the FID namespace. In both cases, we just
817 * pass the FID through, no conversion needed. */
818 if (ostid->oi_fid.f_ver != 0) {
819 CERROR("bad MDT0 id(2), "DOSTID" ost_idx:%u\n",
820 POSTID(ostid), ost_idx);
823 *fid = ostid->oi_fid;
829 /* pack any OST FID into an ostid (id/seq) for the wire/disk */
830 static inline int fid_to_ostid(const struct lu_fid *fid, struct ost_id *ostid)
832 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
833 CERROR("bad IGIF, "DFID"\n", PFID(fid));
837 if (fid_is_idif(fid)) {
838 ostid_set_seq_mdt0(ostid);
839 ostid_set_id(ostid, fid_idif_id(fid_seq(fid), fid_oid(fid),
842 ostid->oi_fid = *fid;
848 /* Check whether the fid is for LAST_ID */
849 static inline bool fid_is_last_id(const struct lu_fid *fid)
851 return fid_oid(fid) == 0;
855 * Get inode number from a igif.
856 * \param fid a igif to get inode number from.
857 * \return inode number for the igif.
859 static inline ino_t lu_igif_ino(const struct lu_fid *fid)
864 extern void lustre_swab_ost_id(struct ost_id *oid);
867 * Get inode generation from a igif.
868 * \param fid a igif to get inode generation from.
869 * \return inode generation for the igif.
871 static inline __u32 lu_igif_gen(const struct lu_fid *fid)
877 * Build igif from the inode number/generation.
879 static inline void lu_igif_build(struct lu_fid *fid, __u32 ino, __u32 gen)
887 * Fids are transmitted across network (in the sender byte-ordering),
888 * and stored on disk in big-endian order.
890 static inline void fid_cpu_to_le(struct lu_fid *dst, const struct lu_fid *src)
892 dst->f_seq = cpu_to_le64(fid_seq(src));
893 dst->f_oid = cpu_to_le32(fid_oid(src));
894 dst->f_ver = cpu_to_le32(fid_ver(src));
897 static inline void fid_le_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
899 dst->f_seq = le64_to_cpu(fid_seq(src));
900 dst->f_oid = le32_to_cpu(fid_oid(src));
901 dst->f_ver = le32_to_cpu(fid_ver(src));
904 static inline void fid_cpu_to_be(struct lu_fid *dst, const struct lu_fid *src)
906 dst->f_seq = cpu_to_be64(fid_seq(src));
907 dst->f_oid = cpu_to_be32(fid_oid(src));
908 dst->f_ver = cpu_to_be32(fid_ver(src));
911 static inline void fid_be_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
913 dst->f_seq = be64_to_cpu(fid_seq(src));
914 dst->f_oid = be32_to_cpu(fid_oid(src));
915 dst->f_ver = be32_to_cpu(fid_ver(src));
918 static inline bool fid_is_sane(const struct lu_fid *fid)
920 return fid != NULL &&
921 ((fid_seq(fid) >= FID_SEQ_START && fid_ver(fid) == 0) ||
922 fid_is_igif(fid) || fid_is_idif(fid) ||
923 fid_seq_is_rsvd(fid_seq(fid)));
926 extern void lustre_swab_lu_fid(struct lu_fid *fid);
927 extern void lustre_swab_lu_seq_range(struct lu_seq_range *range);
929 static inline bool lu_fid_eq(const struct lu_fid *f0, const struct lu_fid *f1)
931 return memcmp(f0, f1, sizeof *f0) == 0;
934 #define __diff_normalize(val0, val1) \
936 typeof(val0) __val0 = (val0); \
937 typeof(val1) __val1 = (val1); \
939 (__val0 == __val1 ? 0 : __val0 > __val1 ? +1 : -1); \
942 static inline int lu_fid_cmp(const struct lu_fid *f0,
943 const struct lu_fid *f1)
946 __diff_normalize(fid_seq(f0), fid_seq(f1)) ?:
947 __diff_normalize(fid_oid(f0), fid_oid(f1)) ?:
948 __diff_normalize(fid_ver(f0), fid_ver(f1));
951 static inline void ostid_cpu_to_le(const struct ost_id *src_oi,
952 struct ost_id *dst_oi)
954 if (fid_seq_is_mdt0(src_oi->oi.oi_seq)) {
955 dst_oi->oi.oi_id = cpu_to_le64(src_oi->oi.oi_id);
956 dst_oi->oi.oi_seq = cpu_to_le64(src_oi->oi.oi_seq);
958 fid_cpu_to_le(&dst_oi->oi_fid, &src_oi->oi_fid);
962 static inline void ostid_le_to_cpu(const struct ost_id *src_oi,
963 struct ost_id *dst_oi)
965 if (fid_seq_is_mdt0(src_oi->oi.oi_seq)) {
966 dst_oi->oi.oi_id = le64_to_cpu(src_oi->oi.oi_id);
967 dst_oi->oi.oi_seq = le64_to_cpu(src_oi->oi.oi_seq);
969 fid_le_to_cpu(&dst_oi->oi_fid, &src_oi->oi_fid);
973 struct lu_orphan_rec {
974 /* The MDT-object's FID referenced by the orphan OST-object */
975 struct lu_fid lor_fid;
980 struct lu_orphan_ent {
981 /* The orphan OST-object's FID */
982 struct lu_fid loe_key;
983 struct lu_orphan_rec loe_rec;
985 void lustre_swab_orphan_ent(struct lu_orphan_ent *ent);
989 /** \defgroup lu_dir lu_dir
993 * Enumeration of possible directory entry attributes.
995 * Attributes follow directory entry header in the order they appear in this
998 enum lu_dirent_attrs {
1001 LUDA_64BITHASH = 0x0004,
1003 /* The following attrs are used for MDT interanl only,
1004 * not visible to client */
1006 /* Verify the dirent consistency */
1007 LUDA_VERIFY = 0x8000,
1008 /* Only check but not repair the dirent inconsistency */
1009 LUDA_VERIFY_DRYRUN = 0x4000,
1010 /* The dirent has been repaired, or to be repaired (dryrun). */
1011 LUDA_REPAIR = 0x2000,
1012 /* The system is upgraded, has beed or to be repaired (dryrun). */
1013 LUDA_UPGRADE = 0x1000,
1014 /* Ignore this record, go to next directly. */
1015 LUDA_IGNORE = 0x0800,
1018 #define LU_DIRENT_ATTRS_MASK 0xf800
1021 * Layout of readdir pages, as transmitted on wire.
1024 /** valid if LUDA_FID is set. */
1025 struct lu_fid lde_fid;
1026 /** a unique entry identifier: a hash or an offset. */
1028 /** total record length, including all attributes. */
1032 /** optional variable size attributes following this entry.
1033 * taken from enum lu_dirent_attrs.
1036 /** name is followed by the attributes indicated in ->ldp_attrs, in
1037 * their natural order. After the last attribute, padding bytes are
1038 * added to make ->lde_reclen a multiple of 8.
1044 * Definitions of optional directory entry attributes formats.
1046 * Individual attributes do not have their length encoded in a generic way. It
1047 * is assumed that consumer of an attribute knows its format. This means that
1048 * it is impossible to skip over an unknown attribute, except by skipping over all
1049 * remaining attributes (by using ->lde_reclen), which is not too
1050 * constraining, because new server versions will append new attributes at
1051 * the end of an entry.
1055 * Fid directory attribute: a fid of an object referenced by the entry. This
1056 * will be almost always requested by the client and supplied by the server.
1058 * Aligned to 8 bytes.
1060 /* To have compatibility with 1.8, lets have fid in lu_dirent struct. */
1065 * Aligned to 2 bytes.
1072 __u64 ldp_hash_start;
1076 struct lu_dirent ldp_entries[0];
1079 enum lu_dirpage_flags {
1081 * dirpage contains no entry.
1085 * last entry's lde_hash equals ldp_hash_end.
1087 LDF_COLLIDE = 1 << 1
1090 static inline struct lu_dirent *lu_dirent_start(struct lu_dirpage *dp)
1092 if (le32_to_cpu(dp->ldp_flags) & LDF_EMPTY)
1095 return dp->ldp_entries;
1098 static inline struct lu_dirent *lu_dirent_next(struct lu_dirent *ent)
1100 struct lu_dirent *next;
1102 if (le16_to_cpu(ent->lde_reclen) != 0)
1103 next = ((void *)ent) + le16_to_cpu(ent->lde_reclen);
1110 static inline size_t lu_dirent_calc_size(size_t namelen, __u16 attr)
1114 if (attr & LUDA_TYPE) {
1115 const size_t align = sizeof(struct luda_type) - 1;
1116 size = (sizeof(struct lu_dirent) + namelen + align) & ~align;
1117 size += sizeof(struct luda_type);
1119 size = sizeof(struct lu_dirent) + namelen;
1121 return (size + 7) & ~7;
1124 #define MDS_DIR_END_OFF 0xfffffffffffffffeULL
1127 * MDS_READPAGE page size
1129 * This is the directory page size packed in MDS_READPAGE RPC.
1130 * It's different than PAGE_CACHE_SIZE because the client needs to
1131 * access the struct lu_dirpage header packed at the beginning of
1132 * the "page" and without this there isn't any way to know find the
1133 * lu_dirpage header is if client and server PAGE_CACHE_SIZE differ.
1135 #define LU_PAGE_SHIFT 12
1136 #define LU_PAGE_SIZE (1UL << LU_PAGE_SHIFT)
1137 #define LU_PAGE_MASK (~(LU_PAGE_SIZE - 1))
1139 #define LU_PAGE_COUNT (1 << (PAGE_CACHE_SHIFT - LU_PAGE_SHIFT))
1143 struct lustre_handle {
1146 #define DEAD_HANDLE_MAGIC 0xdeadbeefcafebabeULL
1148 static inline bool lustre_handle_is_used(const struct lustre_handle *lh)
1150 return lh->cookie != 0;
1153 static inline bool lustre_handle_equal(const struct lustre_handle *lh1,
1154 const struct lustre_handle *lh2)
1156 return lh1->cookie == lh2->cookie;
1159 static inline void lustre_handle_copy(struct lustre_handle *tgt,
1160 const struct lustre_handle *src)
1162 tgt->cookie = src->cookie;
1165 /* flags for lm_flags */
1166 #define MSGHDR_AT_SUPPORT 0x1
1167 #define MSGHDR_CKSUM_INCOMPAT18 0x2
1169 #define lustre_msg lustre_msg_v2
1170 /* we depend on this structure to be 8-byte aligned */
1171 /* this type is only endian-adjusted in lustre_unpack_msg() */
1172 struct lustre_msg_v2 {
1181 __u32 lm_buflens[0];
1184 /* without gss, ptlrpc_body is put at the first buffer. */
1185 #define PTLRPC_NUM_VERSIONS 4
1186 struct ptlrpc_body_v3 {
1187 struct lustre_handle pb_handle;
1194 __u64 pb_last_committed;
1199 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1200 __u32 pb_service_time; /* for rep, actual service time */
1203 /* VBR: pre-versions */
1204 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1205 /* padding for future needs */
1206 __u64 pb_padding[4];
1207 char pb_jobid[LUSTRE_JOBID_SIZE];
1209 #define ptlrpc_body ptlrpc_body_v3
1211 struct ptlrpc_body_v2 {
1212 struct lustre_handle pb_handle;
1219 __u64 pb_last_committed;
1224 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1225 __u32 pb_service_time; /* for rep, actual service time, also used for
1226 net_latency of req */
1229 /* VBR: pre-versions */
1230 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1231 /* padding for future needs */
1232 __u64 pb_padding[4];
1235 extern void lustre_swab_ptlrpc_body(struct ptlrpc_body *pb);
1237 /* message body offset for lustre_msg_v2 */
1238 /* ptlrpc body offset in all request/reply messages */
1239 #define MSG_PTLRPC_BODY_OFF 0
1241 /* normal request/reply message record offset */
1242 #define REQ_REC_OFF 1
1243 #define REPLY_REC_OFF 1
1245 /* ldlm request message body offset */
1246 #define DLM_LOCKREQ_OFF 1 /* lockreq offset */
1247 #define DLM_REQ_REC_OFF 2 /* normal dlm request record offset */
1249 /* ldlm intent lock message body offset */
1250 #define DLM_INTENT_IT_OFF 2 /* intent lock it offset */
1251 #define DLM_INTENT_REC_OFF 3 /* intent lock record offset */
1253 /* ldlm reply message body offset */
1254 #define DLM_LOCKREPLY_OFF 1 /* lockrep offset */
1255 #define DLM_REPLY_REC_OFF 2 /* reply record offset */
1257 /** only use in req->rq_{req,rep}_swab_mask */
1258 #define MSG_PTLRPC_HEADER_OFF 31
1260 /* Flags that are operation-specific go in the top 16 bits. */
1261 #define MSG_OP_FLAG_MASK 0xffff0000
1262 #define MSG_OP_FLAG_SHIFT 16
1264 /* Flags that apply to all requests are in the bottom 16 bits */
1265 #define MSG_GEN_FLAG_MASK 0x0000ffff
1266 #define MSG_LAST_REPLAY 0x0001
1267 #define MSG_RESENT 0x0002
1268 #define MSG_REPLAY 0x0004
1269 /* #define MSG_AT_SUPPORT 0x0008
1270 * This was used in early prototypes of adaptive timeouts, and while there
1271 * shouldn't be any users of that code there also isn't a need for using this
1272 * bits. Defer usage until at least 1.10 to avoid potential conflict. */
1273 #define MSG_DELAY_REPLAY 0x0010
1274 #define MSG_VERSION_REPLAY 0x0020
1275 #define MSG_REQ_REPLAY_DONE 0x0040
1276 #define MSG_LOCK_REPLAY_DONE 0x0080
1279 * Flags for all connect opcodes (MDS_CONNECT, OST_CONNECT)
1282 #define MSG_CONNECT_RECOVERING 0x00000001
1283 #define MSG_CONNECT_RECONNECT 0x00000002
1284 #define MSG_CONNECT_REPLAYABLE 0x00000004
1285 //#define MSG_CONNECT_PEER 0x8
1286 #define MSG_CONNECT_LIBCLIENT 0x00000010
1287 #define MSG_CONNECT_INITIAL 0x00000020
1288 #define MSG_CONNECT_ASYNC 0x00000040
1289 #define MSG_CONNECT_NEXT_VER 0x00000080 /* use next version of lustre_msg */
1290 #define MSG_CONNECT_TRANSNO 0x00000100 /* report transno */
1293 #define OBD_CONNECT_RDONLY 0x1ULL /*client has read-only access*/
1294 #define OBD_CONNECT_INDEX 0x2ULL /*connect specific LOV idx */
1295 #define OBD_CONNECT_MDS 0x4ULL /*connect from MDT to OST */
1296 #define OBD_CONNECT_GRANT 0x8ULL /*OSC gets grant at connect */
1297 #define OBD_CONNECT_SRVLOCK 0x10ULL /*server takes locks for cli */
1298 #define OBD_CONNECT_VERSION 0x20ULL /*Lustre versions in ocd */
1299 #define OBD_CONNECT_REQPORTAL 0x40ULL /*Separate non-IO req portal */
1300 #define OBD_CONNECT_ACL 0x80ULL /*access control lists */
1301 #define OBD_CONNECT_XATTR 0x100ULL /*client use extended attr */
1302 #define OBD_CONNECT_CROW 0x200ULL /*MDS+OST create obj on write*/
1303 #define OBD_CONNECT_TRUNCLOCK 0x400ULL /*locks on server for punch */
1304 #define OBD_CONNECT_TRANSNO 0x800ULL /*replay sends init transno */
1305 #define OBD_CONNECT_IBITS 0x1000ULL /*support for inodebits locks*/
1306 #define OBD_CONNECT_JOIN 0x2000ULL /*files can be concatenated.
1307 *We do not support JOIN FILE
1308 *anymore, reserve this flags
1309 *just for preventing such bit
1311 #define OBD_CONNECT_ATTRFID 0x4000ULL /*Server can GetAttr By Fid*/
1312 #define OBD_CONNECT_NODEVOH 0x8000ULL /*No open hndl on specl nodes*/
1313 #define OBD_CONNECT_RMT_CLIENT 0x10000ULL /*Remote client */
1314 #define OBD_CONNECT_RMT_CLIENT_FORCE 0x20000ULL /*Remote client by force */
1315 #define OBD_CONNECT_BRW_SIZE 0x40000ULL /*Max bytes per rpc */
1316 #define OBD_CONNECT_QUOTA64 0x80000ULL /*Not used since 2.4 */
1317 #define OBD_CONNECT_MDS_CAPA 0x100000ULL /*MDS capability */
1318 #define OBD_CONNECT_OSS_CAPA 0x200000ULL /*OSS capability */
1319 #define OBD_CONNECT_CANCELSET 0x400000ULL /*Early batched cancels. */
1320 #define OBD_CONNECT_SOM 0x800000ULL /*Size on MDS */
1321 #define OBD_CONNECT_AT 0x1000000ULL /*client uses AT */
1322 #define OBD_CONNECT_LRU_RESIZE 0x2000000ULL /*LRU resize feature. */
1323 #define OBD_CONNECT_MDS_MDS 0x4000000ULL /*MDS-MDS connection */
1324 #define OBD_CONNECT_REAL 0x8000000ULL /*real connection */
1325 #define OBD_CONNECT_CHANGE_QS 0x10000000ULL /*Not used since 2.4 */
1326 #define OBD_CONNECT_CKSUM 0x20000000ULL /*support several cksum algos*/
1327 #define OBD_CONNECT_FID 0x40000000ULL /*FID is supported by server */
1328 #define OBD_CONNECT_VBR 0x80000000ULL /*version based recovery */
1329 #define OBD_CONNECT_LOV_V3 0x100000000ULL /*client supports LOV v3 EA */
1330 #define OBD_CONNECT_GRANT_SHRINK 0x200000000ULL /* support grant shrink */
1331 #define OBD_CONNECT_SKIP_ORPHAN 0x400000000ULL /* don't reuse orphan objids */
1332 #define OBD_CONNECT_MAX_EASIZE 0x800000000ULL /* preserved for large EA */
1333 #define OBD_CONNECT_FULL20 0x1000000000ULL /* it is 2.0 client */
1334 #define OBD_CONNECT_LAYOUTLOCK 0x2000000000ULL /* client uses layout lock */
1335 #define OBD_CONNECT_64BITHASH 0x4000000000ULL /* client supports 64-bits
1337 #define OBD_CONNECT_MAXBYTES 0x8000000000ULL /* max stripe size */
1338 #define OBD_CONNECT_IMP_RECOV 0x10000000000ULL /* imp recovery support */
1339 #define OBD_CONNECT_JOBSTATS 0x20000000000ULL /* jobid in ptlrpc_body */
1340 #define OBD_CONNECT_UMASK 0x40000000000ULL /* create uses client umask */
1341 #define OBD_CONNECT_EINPROGRESS 0x80000000000ULL /* client handles -EINPROGRESS
1342 * RPC error properly */
1343 #define OBD_CONNECT_GRANT_PARAM 0x100000000000ULL/* extra grant params used for
1344 * finer space reservation */
1345 #define OBD_CONNECT_FLOCK_OWNER 0x200000000000ULL /* for the fixed 1.8
1346 * policy and 2.x server */
1347 #define OBD_CONNECT_LVB_TYPE 0x400000000000ULL /* variable type of LVB */
1348 #define OBD_CONNECT_NANOSEC_TIME 0x800000000000ULL /* nanosecond timestamps */
1349 #define OBD_CONNECT_LIGHTWEIGHT 0x1000000000000ULL/* lightweight connection */
1350 #define OBD_CONNECT_SHORTIO 0x2000000000000ULL/* short io */
1351 #define OBD_CONNECT_PINGLESS 0x4000000000000ULL/* pings not required */
1352 #define OBD_CONNECT_FLOCK_DEAD 0x8000000000000ULL/* improved flock deadlock detection */
1353 #define OBD_CONNECT_DISP_STRIPE 0x10000000000000ULL/* create stripe disposition*/
1354 #define OBD_CONNECT_OPEN_BY_FID 0x20000000000000ULL /* open by fid won't pack
1356 #define OBD_CONNECT_LFSCK 0x40000000000000ULL/* support online LFSCK */
1357 #define OBD_CONNECT_UNLINK_CLOSE 0x100000000000000ULL/* close file in unlink */
1358 #define OBD_CONNECT_MULTIMODRPCS 0x200000000000000ULL /* support multiple modify
1360 #define OBD_CONNECT_DIR_STRIPE 0x400000000000000ULL /* striped DNE dir */
1363 * Please DO NOT add flag values here before first ensuring that this same
1364 * flag value is not in use on some other branch. Please clear any such
1365 * changes with senior engineers before starting to use a new flag. Then,
1366 * submit a small patch against EVERY branch that ONLY adds the new flag,
1367 * updates obd_connect_names[] for lprocfs_rd_connect_flags(), adds the
1368 * flag to check_obd_connect_data(), and updates wiretests accordingly, so it
1369 * can be approved and landed easily to reserve the flag for future use. */
1371 /* The MNE_SWAB flag is overloading the MDS_MDS bit only for the MGS
1372 * connection. It is a temporary bug fix for Imperative Recovery interop
1373 * between 2.2 and 2.3 x86/ppc nodes, and can be removed when interop for
1374 * 2.2 clients/servers is no longer needed. LU-1252/LU-1644. */
1375 #define OBD_CONNECT_MNE_SWAB OBD_CONNECT_MDS_MDS
1377 #define OCD_HAS_FLAG(ocd, flg) \
1378 (!!((ocd)->ocd_connect_flags & OBD_CONNECT_##flg))
1381 #ifdef HAVE_LRU_RESIZE_SUPPORT
1382 #define LRU_RESIZE_CONNECT_FLAG OBD_CONNECT_LRU_RESIZE
1384 #define LRU_RESIZE_CONNECT_FLAG 0
1387 #define MDT_CONNECT_SUPPORTED (OBD_CONNECT_RDONLY | OBD_CONNECT_VERSION | \
1388 OBD_CONNECT_ACL | OBD_CONNECT_XATTR | \
1389 OBD_CONNECT_IBITS | \
1390 OBD_CONNECT_NODEVOH | OBD_CONNECT_ATTRFID | \
1391 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1392 OBD_CONNECT_RMT_CLIENT | \
1393 OBD_CONNECT_RMT_CLIENT_FORCE | \
1394 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_MDS_CAPA | \
1395 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_MDS_MDS | \
1396 OBD_CONNECT_FID | LRU_RESIZE_CONNECT_FLAG | \
1397 OBD_CONNECT_VBR | OBD_CONNECT_LOV_V3 | \
1398 OBD_CONNECT_SOM | OBD_CONNECT_FULL20 | \
1399 OBD_CONNECT_64BITHASH | OBD_CONNECT_JOBSTATS | \
1400 OBD_CONNECT_EINPROGRESS | \
1401 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_UMASK | \
1402 OBD_CONNECT_LVB_TYPE | OBD_CONNECT_LAYOUTLOCK |\
1403 OBD_CONNECT_PINGLESS | OBD_CONNECT_MAX_EASIZE |\
1404 OBD_CONNECT_FLOCK_DEAD | \
1405 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK | \
1406 OBD_CONNECT_OPEN_BY_FID | \
1407 OBD_CONNECT_DIR_STRIPE)
1409 #define OST_CONNECT_SUPPORTED (OBD_CONNECT_SRVLOCK | OBD_CONNECT_GRANT | \
1410 OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | \
1411 OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_INDEX | \
1412 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_OSS_CAPA | \
1413 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1414 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_CKSUM | \
1415 OBD_CONNECT_RMT_CLIENT | \
1416 OBD_CONNECT_RMT_CLIENT_FORCE | OBD_CONNECT_VBR | \
1417 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN | \
1418 OBD_CONNECT_GRANT_SHRINK | OBD_CONNECT_FULL20 | \
1419 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES | \
1420 OBD_CONNECT_MAX_EASIZE | \
1421 OBD_CONNECT_EINPROGRESS | \
1422 OBD_CONNECT_JOBSTATS | \
1423 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LVB_TYPE|\
1424 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_FID | \
1425 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK)
1426 #define ECHO_CONNECT_SUPPORTED (0)
1427 #define MGS_CONNECT_SUPPORTED (OBD_CONNECT_VERSION | OBD_CONNECT_AT | \
1428 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV | \
1429 OBD_CONNECT_MNE_SWAB | OBD_CONNECT_PINGLESS)
1431 /* Features required for this version of the client to work with server */
1432 #define CLIENT_CONNECT_MDT_REQD (OBD_CONNECT_IBITS | OBD_CONNECT_FID | \
1435 /* This structure is used for both request and reply.
1437 * If we eventually have separate connect data for different types, which we
1438 * almost certainly will, then perhaps we stick a union in here. */
1439 struct obd_connect_data {
1440 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1441 __u32 ocd_version; /* lustre release version number */
1442 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1443 __u32 ocd_index; /* LOV index to connect to */
1444 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
1445 __u64 ocd_ibits_known; /* inode bits this client understands */
1446 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1447 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1448 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1449 __u32 ocd_unused; /* also fix lustre_swab_connect */
1450 __u64 ocd_transno; /* first transno from client to be replayed */
1451 __u32 ocd_group; /* MDS group on OST */
1452 __u32 ocd_cksum_types; /* supported checksum algorithms */
1453 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1454 __u32 ocd_instance; /* instance # of this target */
1455 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1456 /* Fields after ocd_maxbytes are only accessible by the receiver
1457 * if the corresponding flag in ocd_connect_flags is set. Accessing
1458 * any field after ocd_maxbytes on the receiver without a valid flag
1459 * may result in out-of-bound memory access and kernel oops. */
1460 __u16 ocd_maxmodrpcs; /* Maximum modify RPCs in parallel */
1461 __u16 padding0; /* added 2.1.0. also fix lustre_swab_connect */
1462 __u32 padding1; /* added 2.1.0. also fix lustre_swab_connect */
1463 __u64 padding2; /* added 2.1.0. also fix lustre_swab_connect */
1464 __u64 padding3; /* added 2.1.0. also fix lustre_swab_connect */
1465 __u64 padding4; /* added 2.1.0. also fix lustre_swab_connect */
1466 __u64 padding5; /* added 2.1.0. also fix lustre_swab_connect */
1467 __u64 padding6; /* added 2.1.0. also fix lustre_swab_connect */
1468 __u64 padding7; /* added 2.1.0. also fix lustre_swab_connect */
1469 __u64 padding8; /* added 2.1.0. also fix lustre_swab_connect */
1470 __u64 padding9; /* added 2.1.0. also fix lustre_swab_connect */
1471 __u64 paddingA; /* added 2.1.0. also fix lustre_swab_connect */
1472 __u64 paddingB; /* added 2.1.0. also fix lustre_swab_connect */
1473 __u64 paddingC; /* added 2.1.0. also fix lustre_swab_connect */
1474 __u64 paddingD; /* added 2.1.0. also fix lustre_swab_connect */
1475 __u64 paddingE; /* added 2.1.0. also fix lustre_swab_connect */
1476 __u64 paddingF; /* added 2.1.0. also fix lustre_swab_connect */
1479 * Please DO NOT use any fields here before first ensuring that this same
1480 * field is not in use on some other branch. Please clear any such changes
1481 * with senior engineers before starting to use a new field. Then, submit
1482 * a small patch against EVERY branch that ONLY adds the new field along with
1483 * the matching OBD_CONNECT flag, so that can be approved and landed easily to
1484 * reserve the flag for future use. */
1487 extern void lustre_swab_connect(struct obd_connect_data *ocd);
1490 * Supported checksum algorithms. Up to 32 checksum types are supported.
1491 * (32-bit mask stored in obd_connect_data::ocd_cksum_types)
1492 * Please update DECLARE_CKSUM_NAME/OBD_CKSUM_ALL in obd.h when adding a new
1493 * algorithm and also the OBD_FL_CKSUM* flags.
1496 OBD_CKSUM_CRC32 = 0x00000001,
1497 OBD_CKSUM_ADLER = 0x00000002,
1498 OBD_CKSUM_CRC32C= 0x00000004,
1502 * OST requests: OBDO & OBD request records
1507 OST_REPLY = 0, /* reply ? */
1523 OST_QUOTACHECK = 18,
1525 OST_QUOTA_ADJUST_QUNIT = 20, /* not used since 2.4 */
1528 #define OST_FIRST_OPC OST_REPLY
1531 OBD_FL_INLINEDATA = 0x00000001,
1532 OBD_FL_OBDMDEXISTS = 0x00000002,
1533 OBD_FL_DELORPHAN = 0x00000004, /* if set in o_flags delete orphans */
1534 OBD_FL_NORPC = 0x00000008, /* set in o_flags do in OSC not OST */
1535 OBD_FL_IDONLY = 0x00000010, /* set in o_flags only adjust obj id*/
1536 OBD_FL_RECREATE_OBJS= 0x00000020, /* recreate missing obj */
1537 OBD_FL_DEBUG_CHECK = 0x00000040, /* echo client/server debug check */
1538 OBD_FL_NO_USRQUOTA = 0x00000100, /* the object's owner is over quota */
1539 OBD_FL_NO_GRPQUOTA = 0x00000200, /* the object's group is over quota */
1540 OBD_FL_CREATE_CROW = 0x00000400, /* object should be create on write */
1541 OBD_FL_SRVLOCK = 0x00000800, /* delegate DLM locking to server */
1542 OBD_FL_CKSUM_CRC32 = 0x00001000, /* CRC32 checksum type */
1543 OBD_FL_CKSUM_ADLER = 0x00002000, /* ADLER checksum type */
1544 OBD_FL_CKSUM_CRC32C = 0x00004000, /* CRC32C checksum type */
1545 OBD_FL_CKSUM_RSVD2 = 0x00008000, /* for future cksum types */
1546 OBD_FL_CKSUM_RSVD3 = 0x00010000, /* for future cksum types */
1547 OBD_FL_SHRINK_GRANT = 0x00020000, /* object shrink the grant */
1548 OBD_FL_MMAP = 0x00040000, /* object is mmapped on the client.
1549 * XXX: obsoleted - reserved for old
1550 * clients prior than 2.2 */
1551 OBD_FL_RECOV_RESEND = 0x00080000, /* recoverable resent */
1552 OBD_FL_NOSPC_BLK = 0x00100000, /* no more block space on OST */
1553 OBD_FL_FLUSH = 0x00200000, /* flush pages on the OST */
1554 OBD_FL_SHORT_IO = 0x00400000, /* short io request */
1556 /* Note that while these checksum values are currently separate bits,
1557 * in 2.x we can actually allow all values from 1-31 if we wanted. */
1558 OBD_FL_CKSUM_ALL = OBD_FL_CKSUM_CRC32 | OBD_FL_CKSUM_ADLER |
1559 OBD_FL_CKSUM_CRC32C,
1561 /* mask for local-only flag, which won't be sent over network */
1562 OBD_FL_LOCAL_MASK = 0xF0000000,
1566 * All LOV EA magics should have the same postfix, if some new version
1567 * Lustre instroduces new LOV EA magic, then when down-grade to an old
1568 * Lustre, even though the old version system does not recognizes such
1569 * new magic, it still can distinguish the corrupted cases by checking
1570 * the magic's postfix.
1572 #define LOV_MAGIC_MAGIC 0x0BD0
1573 #define LOV_MAGIC_MASK 0xFFFF
1575 #define LOV_MAGIC_V1 (0x0BD10000 | LOV_MAGIC_MAGIC)
1576 #define LOV_MAGIC_JOIN_V1 (0x0BD20000 | LOV_MAGIC_MAGIC)
1577 #define LOV_MAGIC_V3 (0x0BD30000 | LOV_MAGIC_MAGIC)
1578 #define LOV_MAGIC_MIGRATE (0x0BD40000 | LOV_MAGIC_MAGIC)
1579 /* reserved for specifying OSTs */
1580 #define LOV_MAGIC_SPECIFIC (0x0BD50000 | LOV_MAGIC_MAGIC)
1581 #define LOV_MAGIC LOV_MAGIC_V1
1584 * magic for fully defined striping
1585 * the idea is that we should have different magics for striping "hints"
1586 * (struct lov_user_md_v[13]) and defined ready-to-use striping (struct
1587 * lov_mds_md_v[13]). at the moment the magics are used in wire protocol,
1588 * we can't just change it w/o long way preparation, but we still need a
1589 * mechanism to allow LOD to differentiate hint versus ready striping.
1590 * so, at the moment we do a trick: MDT knows what to expect from request
1591 * depending on the case (replay uses ready striping, non-replay req uses
1592 * hints), so MDT replaces magic with appropriate one and now LOD can
1593 * easily understand what's inside -bzzz
1595 #define LOV_MAGIC_V1_DEF 0x0CD10BD0
1596 #define LOV_MAGIC_V3_DEF 0x0CD30BD0
1598 #define lov_pattern(pattern) (pattern & ~LOV_PATTERN_F_MASK)
1599 #define lov_pattern_flags(pattern) (pattern & LOV_PATTERN_F_MASK)
1601 #define lov_ost_data lov_ost_data_v1
1602 struct lov_ost_data_v1 { /* per-stripe data structure (little-endian)*/
1603 struct ost_id l_ost_oi; /* OST object ID */
1604 __u32 l_ost_gen; /* generation of this l_ost_idx */
1605 __u32 l_ost_idx; /* OST index in LOV (lov_tgt_desc->tgts) */
1608 #define lov_mds_md lov_mds_md_v1
1609 struct lov_mds_md_v1 { /* LOV EA mds/wire data (little-endian) */
1610 __u32 lmm_magic; /* magic number = LOV_MAGIC_V1 */
1611 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1612 struct ost_id lmm_oi; /* LOV object ID */
1613 __u32 lmm_stripe_size; /* size of stripe in bytes */
1614 /* lmm_stripe_count used to be __u32 */
1615 __u16 lmm_stripe_count; /* num stripes in use for this object */
1616 __u16 lmm_layout_gen; /* layout generation number */
1617 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1621 * Sigh, because pre-2.4 uses
1622 * struct lov_mds_md_v1 {
1624 * __u64 lmm_object_id;
1625 * __u64 lmm_object_seq;
1628 * to identify the LOV(MDT) object, and lmm_object_seq will
1629 * be normal_fid, which make it hard to combine these conversion
1630 * to ostid_to FID. so we will do lmm_oi/fid conversion separately
1632 * We can tell the lmm_oi by this way,
1633 * 1.8: lmm_object_id = {inode}, lmm_object_gr = 0
1634 * 2.1: lmm_object_id = {oid < 128k}, lmm_object_seq = FID_SEQ_NORMAL
1635 * 2.4: lmm_oi.f_seq = FID_SEQ_NORMAL, lmm_oi.f_oid = {oid < 128k},
1638 * But currently lmm_oi/lsm_oi does not have any "real" usages,
1639 * except for printing some information, and the user can always
1640 * get the real FID from LMA, besides this multiple case check might
1641 * make swab more complicate. So we will keep using id/seq for lmm_oi.
1644 static inline void fid_to_lmm_oi(const struct lu_fid *fid,
1647 oi->oi.oi_id = fid_oid(fid);
1648 oi->oi.oi_seq = fid_seq(fid);
1651 static inline void lmm_oi_set_seq(struct ost_id *oi, __u64 seq)
1653 oi->oi.oi_seq = seq;
1656 static inline void lmm_oi_set_id(struct ost_id *oi, __u64 oid)
1661 static inline __u64 lmm_oi_id(const struct ost_id *oi)
1663 return oi->oi.oi_id;
1666 static inline __u64 lmm_oi_seq(const struct ost_id *oi)
1668 return oi->oi.oi_seq;
1671 static inline void lmm_oi_le_to_cpu(struct ost_id *dst_oi,
1672 const struct ost_id *src_oi)
1674 dst_oi->oi.oi_id = le64_to_cpu(src_oi->oi.oi_id);
1675 dst_oi->oi.oi_seq = le64_to_cpu(src_oi->oi.oi_seq);
1678 static inline void lmm_oi_cpu_to_le(struct ost_id *dst_oi,
1679 const struct ost_id *src_oi)
1681 dst_oi->oi.oi_id = cpu_to_le64(src_oi->oi.oi_id);
1682 dst_oi->oi.oi_seq = cpu_to_le64(src_oi->oi.oi_seq);
1685 /* extern void lustre_swab_lov_mds_md(struct lov_mds_md *llm); */
1687 #define MAX_MD_SIZE (sizeof(struct lov_mds_md) + 4 * sizeof(struct lov_ost_data))
1688 #define MIN_MD_SIZE (sizeof(struct lov_mds_md) + 1 * sizeof(struct lov_ost_data))
1690 /* This is the default MDT reply size allocated, should the striping be bigger,
1691 * it will be reallocated in mdt_fix_reply.
1692 * 100 stripes is a bit less than 2.5k of data */
1693 #define DEF_REP_MD_SIZE (sizeof(struct lov_mds_md) + \
1694 100 * sizeof(struct lov_ost_data))
1696 #define XATTR_NAME_ACL_ACCESS "system.posix_acl_access"
1697 #define XATTR_NAME_ACL_DEFAULT "system.posix_acl_default"
1698 #define XATTR_USER_PREFIX "user."
1699 #define XATTR_TRUSTED_PREFIX "trusted."
1700 #define XATTR_SECURITY_PREFIX "security."
1702 #define XATTR_NAME_LOV "trusted.lov"
1703 #define XATTR_NAME_LMA "trusted.lma"
1704 #define XATTR_NAME_LMV "trusted.lmv"
1705 #define XATTR_NAME_DEFAULT_LMV "trusted.dmv"
1706 #define XATTR_NAME_LINK "trusted.link"
1707 #define XATTR_NAME_FID "trusted.fid"
1708 #define XATTR_NAME_VERSION "trusted.version"
1709 #define XATTR_NAME_SOM "trusted.som"
1710 #define XATTR_NAME_HSM "trusted.hsm"
1711 #define XATTR_NAME_LFSCK_BITMAP "trusted.lfsck_bitmap"
1712 #define XATTR_NAME_DUMMY "trusted.dummy"
1714 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1715 # define XATTR_NAME_LFSCK_NAMESPACE_OLD "trusted.lfsck_namespace"
1718 #define XATTR_NAME_LFSCK_NAMESPACE "trusted.lfsck_ns"
1719 #define XATTR_NAME_MAX_LEN 32 /* increase this, if there is longer name. */
1721 struct lov_mds_md_v3 { /* LOV EA mds/wire data (little-endian) */
1722 __u32 lmm_magic; /* magic number = LOV_MAGIC_V3 */
1723 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1724 struct ost_id lmm_oi; /* LOV object ID */
1725 __u32 lmm_stripe_size; /* size of stripe in bytes */
1726 /* lmm_stripe_count used to be __u32 */
1727 __u16 lmm_stripe_count; /* num stripes in use for this object */
1728 __u16 lmm_layout_gen; /* layout generation number */
1729 char lmm_pool_name[LOV_MAXPOOLNAME + 1]; /* must be 32bit aligned */
1730 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1733 static inline __u32 lov_mds_md_size(__u16 stripes, __u32 lmm_magic)
1735 if (lmm_magic == LOV_MAGIC_V3)
1736 return sizeof(struct lov_mds_md_v3) +
1737 stripes * sizeof(struct lov_ost_data_v1);
1739 return sizeof(struct lov_mds_md_v1) +
1740 stripes * sizeof(struct lov_ost_data_v1);
1744 lov_mds_md_max_stripe_count(size_t buf_size, __u32 lmm_magic)
1746 switch (lmm_magic) {
1747 case LOV_MAGIC_V1: {
1748 struct lov_mds_md_v1 lmm;
1750 if (buf_size < sizeof(lmm))
1753 return (buf_size - sizeof(lmm)) / sizeof(lmm.lmm_objects[0]);
1755 case LOV_MAGIC_V3: {
1756 struct lov_mds_md_v3 lmm;
1758 if (buf_size < sizeof(lmm))
1761 return (buf_size - sizeof(lmm)) / sizeof(lmm.lmm_objects[0]);
1768 #define OBD_MD_FLID (0x00000001ULL) /* object ID */
1769 #define OBD_MD_FLATIME (0x00000002ULL) /* access time */
1770 #define OBD_MD_FLMTIME (0x00000004ULL) /* data modification time */
1771 #define OBD_MD_FLCTIME (0x00000008ULL) /* change time */
1772 #define OBD_MD_FLSIZE (0x00000010ULL) /* size */
1773 #define OBD_MD_FLBLOCKS (0x00000020ULL) /* allocated blocks count */
1774 #define OBD_MD_FLBLKSZ (0x00000040ULL) /* block size */
1775 #define OBD_MD_FLMODE (0x00000080ULL) /* access bits (mode & ~S_IFMT) */
1776 #define OBD_MD_FLTYPE (0x00000100ULL) /* object type (mode & S_IFMT) */
1777 #define OBD_MD_FLUID (0x00000200ULL) /* user ID */
1778 #define OBD_MD_FLGID (0x00000400ULL) /* group ID */
1779 #define OBD_MD_FLFLAGS (0x00000800ULL) /* flags word */
1780 #define OBD_MD_FLNLINK (0x00002000ULL) /* link count */
1781 #define OBD_MD_FLGENER (0x00004000ULL) /* generation number */
1782 /*#define OBD_MD_FLINLINE (0x00008000ULL) inline data. used until 1.6.5 */
1783 #define OBD_MD_FLRDEV (0x00010000ULL) /* device number */
1784 #define OBD_MD_FLEASIZE (0x00020000ULL) /* extended attribute data */
1785 #define OBD_MD_LINKNAME (0x00040000ULL) /* symbolic link target */
1786 #define OBD_MD_FLHANDLE (0x00080000ULL) /* file/lock handle */
1787 #define OBD_MD_FLCKSUM (0x00100000ULL) /* bulk data checksum */
1788 #define OBD_MD_FLQOS (0x00200000ULL) /* quality of service stats */
1789 /*#define OBD_MD_FLOSCOPQ (0x00400000ULL) osc opaque data, never used */
1790 #define OBD_MD_FLCOOKIE (0x00800000ULL) /* log cancellation cookie */
1791 #define OBD_MD_FLGROUP (0x01000000ULL) /* group */
1792 #define OBD_MD_FLFID (0x02000000ULL) /* ->ost write inline fid */
1793 #define OBD_MD_FLEPOCH (0x04000000ULL) /* ->ost write with ioepoch */
1794 /* ->mds if epoch opens or closes */
1795 #define OBD_MD_FLGRANT (0x08000000ULL) /* ost preallocation space grant */
1796 #define OBD_MD_FLDIREA (0x10000000ULL) /* dir's extended attribute data */
1797 #define OBD_MD_FLUSRQUOTA (0x20000000ULL) /* over quota flags sent from ost */
1798 #define OBD_MD_FLGRPQUOTA (0x40000000ULL) /* over quota flags sent from ost */
1799 #define OBD_MD_FLMODEASIZE (0x80000000ULL) /* EA size will be changed */
1801 #define OBD_MD_MDS (0x0000000100000000ULL) /* where an inode lives on */
1802 #define OBD_MD_REINT (0x0000000200000000ULL) /* reintegrate oa */
1803 #define OBD_MD_MEA (0x0000000400000000ULL) /* CMD split EA */
1804 #define OBD_MD_TSTATE (0x0000000800000000ULL) /* transient state field */
1806 #define OBD_MD_FLXATTR (0x0000001000000000ULL) /* xattr */
1807 #define OBD_MD_FLXATTRLS (0x0000002000000000ULL) /* xattr list */
1808 #define OBD_MD_FLXATTRRM (0x0000004000000000ULL) /* xattr remove */
1809 #define OBD_MD_FLACL (0x0000008000000000ULL) /* ACL */
1810 #define OBD_MD_FLRMTPERM (0x0000010000000000ULL) /* remote permission */
1811 #define OBD_MD_FLMDSCAPA (0x0000020000000000ULL) /* MDS capability */
1812 #define OBD_MD_FLOSSCAPA (0x0000040000000000ULL) /* OSS capability */
1813 #define OBD_MD_FLCKSPLIT (0x0000080000000000ULL) /* Check split on server */
1814 #define OBD_MD_FLCROSSREF (0x0000100000000000ULL) /* Cross-ref case */
1815 #define OBD_MD_FLGETATTRLOCK (0x0000200000000000ULL) /* Get IOEpoch attributes
1816 * under lock; for xattr
1817 * requests means the
1818 * client holds the lock */
1819 #define OBD_MD_FLOBJCOUNT (0x0000400000000000ULL) /* for multiple destroy */
1821 #define OBD_MD_FLRMTLSETFACL (0x0001000000000000ULL) /* lfs lsetfacl case */
1822 #define OBD_MD_FLRMTLGETFACL (0x0002000000000000ULL) /* lfs lgetfacl case */
1823 #define OBD_MD_FLRMTRSETFACL (0x0004000000000000ULL) /* lfs rsetfacl case */
1824 #define OBD_MD_FLRMTRGETFACL (0x0008000000000000ULL) /* lfs rgetfacl case */
1826 #define OBD_MD_FLDATAVERSION (0x0010000000000000ULL) /* iversion sum */
1827 #define OBD_MD_FLRELEASED (0x0020000000000000ULL) /* file released */
1829 #define OBD_MD_DEFAULT_MEA (0x0040000000000000ULL) /* default MEA */
1831 #define OBD_MD_FLGETATTR (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLMTIME | \
1832 OBD_MD_FLCTIME | OBD_MD_FLSIZE | OBD_MD_FLBLKSZ | \
1833 OBD_MD_FLMODE | OBD_MD_FLTYPE | OBD_MD_FLUID | \
1834 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLNLINK | \
1835 OBD_MD_FLGENER | OBD_MD_FLRDEV | OBD_MD_FLGROUP)
1837 #define OBD_MD_FLXATTRALL (OBD_MD_FLXATTR | OBD_MD_FLXATTRLS)
1839 /* don't forget obdo_fid which is way down at the bottom so it can
1840 * come after the definition of llog_cookie */
1844 HSS_CLEARMASK = 0x02,
1845 HSS_ARCHIVE_ID = 0x04,
1848 struct hsm_state_set {
1850 __u32 hss_archive_id;
1852 __u64 hss_clearmask;
1855 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
1856 extern void lustre_swab_hsm_state_set(struct hsm_state_set *hss);
1858 extern void lustre_swab_obd_statfs (struct obd_statfs *os);
1860 /* ost_body.data values for OST_BRW */
1862 #define OBD_BRW_READ 0x01
1863 #define OBD_BRW_WRITE 0x02
1864 #define OBD_BRW_RWMASK (OBD_BRW_READ | OBD_BRW_WRITE)
1865 #define OBD_BRW_SYNC 0x08 /* this page is a part of synchronous
1866 * transfer and is not accounted in
1868 #define OBD_BRW_CHECK 0x10
1869 #define OBD_BRW_FROM_GRANT 0x20 /* the osc manages this under llite */
1870 #define OBD_BRW_GRANTED 0x40 /* the ost manages this */
1871 #define OBD_BRW_NOCACHE 0x80 /* this page is a part of non-cached IO */
1872 #define OBD_BRW_NOQUOTA 0x100
1873 #define OBD_BRW_SRVLOCK 0x200 /* Client holds no lock over this page */
1874 #define OBD_BRW_ASYNC 0x400 /* Server may delay commit to disk */
1875 #define OBD_BRW_MEMALLOC 0x800 /* Client runs in the "kswapd" context */
1876 #define OBD_BRW_OVER_USRQUOTA 0x1000 /* Running out of user quota */
1877 #define OBD_BRW_OVER_GRPQUOTA 0x2000 /* Running out of group quota */
1878 #define OBD_BRW_SOFT_SYNC 0x4000 /* This flag notifies the server
1879 * that the client is running low on
1880 * space for unstable pages; asking
1881 * it to sync quickly */
1883 #define OBD_OBJECT_EOF LUSTRE_EOF
1885 #define OST_MIN_PRECREATE 32
1886 #define OST_MAX_PRECREATE 20000
1889 struct ost_id ioo_oid; /* object ID, if multi-obj BRW */
1890 __u32 ioo_max_brw; /* low 16 bits were o_mode before 2.4,
1891 * now (PTLRPC_BULK_OPS_COUNT - 1) in
1892 * high 16 bits in 2.4 and later */
1893 __u32 ioo_bufcnt; /* number of niobufs for this object */
1896 #define IOOBJ_MAX_BRW_BITS 16
1897 #define IOOBJ_TYPE_MASK ((1U << IOOBJ_MAX_BRW_BITS) - 1)
1898 #define ioobj_max_brw_get(ioo) (((ioo)->ioo_max_brw >> IOOBJ_MAX_BRW_BITS) + 1)
1899 #define ioobj_max_brw_set(ioo, num) \
1900 do { (ioo)->ioo_max_brw = ((num) - 1) << IOOBJ_MAX_BRW_BITS; } while (0)
1902 extern void lustre_swab_obd_ioobj (struct obd_ioobj *ioo);
1904 /* multiple of 8 bytes => can array */
1905 struct niobuf_remote {
1911 void lustre_swab_niobuf_remote(struct niobuf_remote *nbr);
1913 /* lock value block communicated between the filter and llite */
1915 /* OST_LVB_ERR_INIT is needed because the return code in rc is
1916 * negative, i.e. because ((MASK + rc) & MASK) != MASK. */
1917 #define OST_LVB_ERR_INIT 0xffbadbad80000000ULL
1918 #define OST_LVB_ERR_MASK 0xffbadbad00000000ULL
1919 #define OST_LVB_IS_ERR(blocks) \
1920 ((blocks & OST_LVB_ERR_MASK) == OST_LVB_ERR_MASK)
1921 #define OST_LVB_SET_ERR(blocks, rc) \
1922 do { blocks = OST_LVB_ERR_INIT + rc; } while (0)
1923 #define OST_LVB_GET_ERR(blocks) (int)(blocks - OST_LVB_ERR_INIT)
1933 extern void lustre_swab_ost_lvb_v1(struct ost_lvb_v1 *lvb);
1947 extern void lustre_swab_ost_lvb(struct ost_lvb *lvb);
1950 * lquota data structures
1953 #ifndef QUOTABLOCK_BITS
1954 # define QUOTABLOCK_BITS LUSTRE_QUOTABLOCK_BITS
1957 #ifndef QUOTABLOCK_SIZE
1958 # define QUOTABLOCK_SIZE LUSTRE_QUOTABLOCK_SIZE
1962 # define toqb lustre_stoqb
1965 /* The lquota_id structure is an union of all the possible identifier types that
1966 * can be used with quota, this includes:
1969 * - a FID which can be used for per-directory quota in the future */
1971 struct lu_fid qid_fid; /* FID for per-directory quota */
1972 __u64 qid_uid; /* user identifier */
1973 __u64 qid_gid; /* group identifier */
1976 /* quotactl management */
1977 struct obd_quotactl {
1979 __u32 qc_type; /* see Q_* flag below */
1982 struct obd_dqinfo qc_dqinfo;
1983 struct obd_dqblk qc_dqblk;
1986 extern void lustre_swab_obd_quotactl(struct obd_quotactl *q);
1988 #define Q_COPY(out, in, member) (out)->member = (in)->member
1990 #define QCTL_COPY(out, in) \
1992 Q_COPY(out, in, qc_cmd); \
1993 Q_COPY(out, in, qc_type); \
1994 Q_COPY(out, in, qc_id); \
1995 Q_COPY(out, in, qc_stat); \
1996 Q_COPY(out, in, qc_dqinfo); \
1997 Q_COPY(out, in, qc_dqblk); \
2000 /* Body of quota request used for quota acquire/release RPCs between quota
2001 * master (aka QMT) and slaves (ak QSD). */
2003 struct lu_fid qb_fid; /* FID of global index packing the pool ID
2004 * and type (data or metadata) as well as
2005 * the quota type (user or group). */
2006 union lquota_id qb_id; /* uid or gid or directory FID */
2007 __u32 qb_flags; /* see below */
2009 __u64 qb_count; /* acquire/release count (kbytes/inodes) */
2010 __u64 qb_usage; /* current slave usage (kbytes/inodes) */
2011 __u64 qb_slv_ver; /* slave index file version */
2012 struct lustre_handle qb_lockh; /* per-ID lock handle */
2013 struct lustre_handle qb_glb_lockh; /* global lock handle */
2014 __u64 qb_padding1[4];
2017 /* When the quota_body is used in the reply of quota global intent
2018 * lock (IT_QUOTA_CONN) reply, qb_fid contains slave index file FID. */
2019 #define qb_slv_fid qb_fid
2020 /* qb_usage is the current qunit (in kbytes/inodes) when quota_body is used in
2022 #define qb_qunit qb_usage
2024 #define QUOTA_DQACQ_FL_ACQ 0x1 /* acquire quota */
2025 #define QUOTA_DQACQ_FL_PREACQ 0x2 /* pre-acquire */
2026 #define QUOTA_DQACQ_FL_REL 0x4 /* release quota */
2027 #define QUOTA_DQACQ_FL_REPORT 0x8 /* report usage */
2029 extern void lustre_swab_quota_body(struct quota_body *b);
2031 /* Quota types currently supported */
2033 LQUOTA_TYPE_USR = 0x00, /* maps to USRQUOTA */
2034 LQUOTA_TYPE_GRP = 0x01, /* maps to GRPQUOTA */
2038 /* There are 2 different resource types on which a quota limit can be enforced:
2039 * - inodes on the MDTs
2040 * - blocks on the OSTs */
2042 LQUOTA_RES_MD = 0x01, /* skip 0 to avoid null oid in FID */
2043 LQUOTA_RES_DT = 0x02,
2045 LQUOTA_FIRST_RES = LQUOTA_RES_MD
2047 #define LQUOTA_NR_RES (LQUOTA_LAST_RES - LQUOTA_FIRST_RES + 1)
2050 * Space accounting support
2051 * Format of an accounting record, providing disk usage information for a given
2054 struct lquota_acct_rec { /* 16 bytes */
2055 __u64 bspace; /* current space in use */
2056 __u64 ispace; /* current # inodes in use */
2060 * Global quota index support
2061 * Format of a global record, providing global quota settings for a given quota
2064 struct lquota_glb_rec { /* 32 bytes */
2065 __u64 qbr_hardlimit; /* quota hard limit, in #inodes or kbytes */
2066 __u64 qbr_softlimit; /* quota soft limit, in #inodes or kbytes */
2067 __u64 qbr_time; /* grace time, in seconds */
2068 __u64 qbr_granted; /* how much is granted to slaves, in #inodes or
2073 * Slave index support
2074 * Format of a slave record, recording how much space is granted to a given
2077 struct lquota_slv_rec { /* 8 bytes */
2078 __u64 qsr_granted; /* space granted to the slave for the key=ID,
2079 * in #inodes or kbytes */
2082 /* Data structures associated with the quota locks */
2084 /* Glimpse descriptor used for the index & per-ID quota locks */
2085 struct ldlm_gl_lquota_desc {
2086 union lquota_id gl_id; /* quota ID subject to the glimpse */
2087 __u64 gl_flags; /* see LQUOTA_FL* below */
2088 __u64 gl_ver; /* new index version */
2089 __u64 gl_hardlimit; /* new hardlimit or qunit value */
2090 __u64 gl_softlimit; /* new softlimit */
2094 #define gl_qunit gl_hardlimit /* current qunit value used when
2095 * glimpsing per-ID quota locks */
2097 /* quota glimpse flags */
2098 #define LQUOTA_FL_EDQUOT 0x1 /* user/group out of quota space on QMT */
2100 /* LVB used with quota (global and per-ID) locks */
2102 __u64 lvb_flags; /* see LQUOTA_FL* above */
2103 __u64 lvb_id_may_rel; /* space that might be released later */
2104 __u64 lvb_id_rel; /* space released by the slave for this ID */
2105 __u64 lvb_id_qunit; /* current qunit value */
2109 extern void lustre_swab_lquota_lvb(struct lquota_lvb *lvb);
2111 /* LVB used with global quota lock */
2112 #define lvb_glb_ver lvb_id_may_rel /* current version of the global index */
2120 #define QUOTA_FIRST_OPC QUOTA_DQACQ
2129 MDS_GETATTR_NAME = 34,
2134 MDS_DISCONNECT = 39,
2137 MDS_PIN = 42, /* obsolete, never used in a release */
2138 MDS_UNPIN = 43, /* obsolete, never used in a release */
2140 MDS_DONE_WRITING = 45,
2142 MDS_QUOTACHECK = 47,
2145 MDS_SETXATTR = 50, /* obsolete, now it's MDS_REINT op */
2147 MDS_IS_SUBDIR = 52, /* obsolete, never used in a release */
2149 MDS_HSM_STATE_GET = 54,
2150 MDS_HSM_STATE_SET = 55,
2151 MDS_HSM_ACTION = 56,
2152 MDS_HSM_PROGRESS = 57,
2153 MDS_HSM_REQUEST = 58,
2154 MDS_HSM_CT_REGISTER = 59,
2155 MDS_HSM_CT_UNREGISTER = 60,
2156 MDS_SWAP_LAYOUTS = 61,
2160 #define MDS_FIRST_OPC MDS_GETATTR
2163 /* opcodes for object update */
2169 #define OUT_UPDATE_FIRST_OPC OUT_UPDATE
2186 } mds_reint_t, mdt_reint_t;
2188 extern void lustre_swab_generic_32s (__u32 *val);
2190 /* the disposition of the intent outlines what was executed */
2191 #define DISP_IT_EXECD 0x00000001
2192 #define DISP_LOOKUP_EXECD 0x00000002
2193 #define DISP_LOOKUP_NEG 0x00000004
2194 #define DISP_LOOKUP_POS 0x00000008
2195 #define DISP_OPEN_CREATE 0x00000010
2196 #define DISP_OPEN_OPEN 0x00000020
2197 #define DISP_ENQ_COMPLETE 0x00400000 /* obsolete and unused */
2198 #define DISP_ENQ_OPEN_REF 0x00800000
2199 #define DISP_ENQ_CREATE_REF 0x01000000
2200 #define DISP_OPEN_LOCK 0x02000000
2201 #define DISP_OPEN_LEASE 0x04000000
2202 #define DISP_OPEN_STRIPE 0x08000000
2203 #define DISP_OPEN_DENY 0x10000000
2205 /* INODE LOCK PARTS */
2206 #define MDS_INODELOCK_LOOKUP 0x000001 /* For namespace, dentry etc, and also
2207 * was used to protect permission (mode,
2208 * owner, group etc) before 2.4. */
2209 #define MDS_INODELOCK_UPDATE 0x000002 /* size, links, timestamps */
2210 #define MDS_INODELOCK_OPEN 0x000004 /* For opened files */
2211 #define MDS_INODELOCK_LAYOUT 0x000008 /* for layout */
2213 /* The PERM bit is added int 2.4, and it is used to protect permission(mode,
2214 * owner, group, acl etc), so to separate the permission from LOOKUP lock.
2215 * Because for remote directories(in DNE), these locks will be granted by
2216 * different MDTs(different ldlm namespace).
2218 * For local directory, MDT will always grant UPDATE_LOCK|PERM_LOCK together.
2219 * For Remote directory, the master MDT, where the remote directory is, will
2220 * grant UPDATE_LOCK|PERM_LOCK, and the remote MDT, where the name entry is,
2221 * will grant LOOKUP_LOCK. */
2222 #define MDS_INODELOCK_PERM 0x000010
2223 #define MDS_INODELOCK_XATTR 0x000020 /* extended attributes */
2225 #define MDS_INODELOCK_MAXSHIFT 5
2226 /* This FULL lock is useful to take on unlink sort of operations */
2227 #define MDS_INODELOCK_FULL ((1<<(MDS_INODELOCK_MAXSHIFT+1))-1)
2229 extern void lustre_swab_ll_fid (struct ll_fid *fid);
2231 /* NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
2232 * but was moved into name[1] along with the OID to avoid consuming the
2233 * name[2,3] fields that need to be used for the quota id (also a FID). */
2235 LUSTRE_RES_ID_SEQ_OFF = 0,
2236 LUSTRE_RES_ID_VER_OID_OFF = 1,
2237 LUSTRE_RES_ID_WAS_VER_OFF = 2, /* see note above */
2238 LUSTRE_RES_ID_QUOTA_SEQ_OFF = 2,
2239 LUSTRE_RES_ID_QUOTA_VER_OID_OFF = 3,
2240 LUSTRE_RES_ID_HSH_OFF = 3
2243 #define MDS_STATUS_CONN 1
2244 #define MDS_STATUS_LOV 2
2246 /* mdt_thread_info.mti_flags. */
2248 /* The flag indicates Size-on-MDS attributes are changed. */
2249 MF_SOM_CHANGE = (1 << 0),
2250 /* Flags indicates an epoch opens or closes. */
2251 MF_EPOCH_OPEN = (1 << 1),
2252 MF_EPOCH_CLOSE = (1 << 2),
2253 MF_MDC_CANCEL_FID1 = (1 << 3),
2254 MF_MDC_CANCEL_FID2 = (1 << 4),
2255 MF_MDC_CANCEL_FID3 = (1 << 5),
2256 MF_MDC_CANCEL_FID4 = (1 << 6),
2257 /* There is a pending attribute update. */
2258 MF_SOM_AU = (1 << 7),
2259 /* Cancel OST locks while getattr OST attributes. */
2260 MF_GETATTR_LOCK = (1 << 8),
2261 MF_GET_MDT_IDX = (1 << 9),
2264 #define LUSTRE_BFLAG_UNCOMMITTED_WRITES 0x1
2266 /* these should be identical to their EXT4_*_FL counterparts, they are
2267 * redefined here only to avoid dragging in fs/ext4/ext4.h */
2268 #define LUSTRE_SYNC_FL 0x00000008 /* Synchronous updates */
2269 #define LUSTRE_IMMUTABLE_FL 0x00000010 /* Immutable file */
2270 #define LUSTRE_APPEND_FL 0x00000020 /* writes to file may only append */
2271 #define LUSTRE_NOATIME_FL 0x00000080 /* do not update atime */
2272 #define LUSTRE_DIRSYNC_FL 0x00010000 /* dirsync behaviour (dir only) */
2275 /* Convert wire LUSTRE_*_FL to corresponding client local VFS S_* values
2276 * for the client inode i_flags. The LUSTRE_*_FL are the Lustre wire
2277 * protocol equivalents of LDISKFS_*_FL values stored on disk, while
2278 * the S_* flags are kernel-internal values that change between kernel
2279 * versions. These flags are set/cleared via FSFILT_IOC_{GET,SET}_FLAGS.
2280 * See b=16526 for a full history. */
2281 static inline int ll_ext_to_inode_flags(int flags)
2283 return (((flags & LUSTRE_SYNC_FL) ? S_SYNC : 0) |
2284 ((flags & LUSTRE_NOATIME_FL) ? S_NOATIME : 0) |
2285 ((flags & LUSTRE_APPEND_FL) ? S_APPEND : 0) |
2286 #if defined(S_DIRSYNC)
2287 ((flags & LUSTRE_DIRSYNC_FL) ? S_DIRSYNC : 0) |
2289 ((flags & LUSTRE_IMMUTABLE_FL) ? S_IMMUTABLE : 0));
2292 static inline int ll_inode_to_ext_flags(int iflags)
2294 return (((iflags & S_SYNC) ? LUSTRE_SYNC_FL : 0) |
2295 ((iflags & S_NOATIME) ? LUSTRE_NOATIME_FL : 0) |
2296 ((iflags & S_APPEND) ? LUSTRE_APPEND_FL : 0) |
2297 #if defined(S_DIRSYNC)
2298 ((iflags & S_DIRSYNC) ? LUSTRE_DIRSYNC_FL : 0) |
2300 ((iflags & S_IMMUTABLE) ? LUSTRE_IMMUTABLE_FL : 0));
2304 /* 64 possible states */
2305 enum md_transient_state {
2306 MS_RESTORE = (1 << 0), /* restore is running */
2310 struct lu_fid mbo_fid1;
2311 struct lu_fid mbo_fid2;
2312 struct lustre_handle mbo_handle;
2314 __u64 mbo_size; /* Offset, in the case of MDS_READPAGE */
2318 __u64 mbo_blocks; /* XID, in the case of MDS_READPAGE */
2320 __u64 mbo_t_state; /* transient file state defined in
2321 * enum md_transient_state
2322 * was "ino" until 2.4.0 */
2325 __u32 mbo_capability;
2331 __u32 mbo_nlink; /* #bytes to read in the case of MDS_READPAGE */
2332 __u32 mbo_unused2; /* was "generation" until 2.4.0 */
2334 __u32 mbo_eadatasize;
2336 __u32 mbo_max_mdsize;
2337 __u32 mbo_max_cookiesize;
2338 __u32 mbo_uid_h; /* high 32-bits of uid, for FUID */
2339 __u32 mbo_gid_h; /* high 32-bits of gid, for FUID */
2340 __u32 mbo_padding_5; /* also fix lustre_swab_mdt_body */
2341 __u64 mbo_padding_6;
2342 __u64 mbo_padding_7;
2343 __u64 mbo_padding_8;
2344 __u64 mbo_padding_9;
2345 __u64 mbo_padding_10;
2348 extern void lustre_swab_mdt_body (struct mdt_body *b);
2350 struct mdt_ioepoch {
2351 struct lustre_handle handle;
2357 extern void lustre_swab_mdt_ioepoch (struct mdt_ioepoch *b);
2359 /* permissions for md_perm.mp_perm */
2361 CFS_SETUID_PERM = 0x01,
2362 CFS_SETGID_PERM = 0x02,
2363 CFS_SETGRP_PERM = 0x04,
2364 CFS_RMTACL_PERM = 0x08,
2365 CFS_RMTOWN_PERM = 0x10
2368 /* inode access permission for remote user, the inode info are omitted,
2369 * for client knows them. */
2370 struct mdt_remote_perm {
2377 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
2381 extern void lustre_swab_mdt_remote_perm(struct mdt_remote_perm *p);
2383 struct mdt_rec_setattr {
2393 __u32 sa_padding_1_h;
2394 struct lu_fid sa_fid;
2403 __u32 sa_attr_flags;
2405 __u32 sa_bias; /* some operation flags */
2411 extern void lustre_swab_mdt_rec_setattr (struct mdt_rec_setattr *sa);
2414 * Attribute flags used in mdt_rec_setattr::sa_valid.
2415 * The kernel's #defines for ATTR_* should not be used over the network
2416 * since the client and MDS may run different kernels (see bug 13828)
2417 * Therefore, we should only use MDS_ATTR_* attributes for sa_valid.
2419 #define MDS_ATTR_MODE 0x1ULL /* = 1 */
2420 #define MDS_ATTR_UID 0x2ULL /* = 2 */
2421 #define MDS_ATTR_GID 0x4ULL /* = 4 */
2422 #define MDS_ATTR_SIZE 0x8ULL /* = 8 */
2423 #define MDS_ATTR_ATIME 0x10ULL /* = 16 */
2424 #define MDS_ATTR_MTIME 0x20ULL /* = 32 */
2425 #define MDS_ATTR_CTIME 0x40ULL /* = 64 */
2426 #define MDS_ATTR_ATIME_SET 0x80ULL /* = 128 */
2427 #define MDS_ATTR_MTIME_SET 0x100ULL /* = 256 */
2428 #define MDS_ATTR_FORCE 0x200ULL /* = 512, Not a change, but a change it */
2429 #define MDS_ATTR_ATTR_FLAG 0x400ULL /* = 1024 */
2430 #define MDS_ATTR_KILL_SUID 0x800ULL /* = 2048 */
2431 #define MDS_ATTR_KILL_SGID 0x1000ULL /* = 4096 */
2432 #define MDS_ATTR_CTIME_SET 0x2000ULL /* = 8192 */
2433 #define MDS_ATTR_FROM_OPEN 0x4000ULL /* = 16384, called from open path, ie O_TRUNC */
2434 #define MDS_ATTR_BLOCKS 0x8000ULL /* = 32768 */
2437 #define FMODE_READ 00000001
2438 #define FMODE_WRITE 00000002
2441 #define MDS_FMODE_CLOSED 00000000
2442 #define MDS_FMODE_EXEC 00000004
2443 /* IO Epoch is opened on a closed file. */
2444 #define MDS_FMODE_EPOCH 01000000
2445 /* IO Epoch is opened on a file truncate. */
2446 #define MDS_FMODE_TRUNC 02000000
2447 /* Size-on-MDS Attribute Update is pending. */
2448 #define MDS_FMODE_SOM 04000000
2450 #define MDS_OPEN_CREATED 00000010
2451 #define MDS_OPEN_CROSS 00000020
2453 #define MDS_OPEN_CREAT 00000100
2454 #define MDS_OPEN_EXCL 00000200
2455 #define MDS_OPEN_TRUNC 00001000
2456 #define MDS_OPEN_APPEND 00002000
2457 #define MDS_OPEN_SYNC 00010000
2458 #define MDS_OPEN_DIRECTORY 00200000
2460 #define MDS_OPEN_BY_FID 040000000 /* open_by_fid for known object */
2461 #define MDS_OPEN_DELAY_CREATE 0100000000 /* delay initial object create */
2462 #define MDS_OPEN_OWNEROVERRIDE 0200000000 /* NFSD rw-reopen ro file for owner */
2463 #define MDS_OPEN_JOIN_FILE 0400000000 /* open for join file.
2464 * We do not support JOIN FILE
2465 * anymore, reserve this flags
2466 * just for preventing such bit
2469 #define MDS_OPEN_LOCK 04000000000 /* This open requires open lock */
2470 #define MDS_OPEN_HAS_EA 010000000000 /* specify object create pattern */
2471 #define MDS_OPEN_HAS_OBJS 020000000000 /* Just set the EA the obj exist */
2472 #define MDS_OPEN_NORESTORE 0100000000000ULL /* Do not restore file at open */
2473 #define MDS_OPEN_NEWSTRIPE 0200000000000ULL /* New stripe needed (restripe or
2475 #define MDS_OPEN_VOLATILE 0400000000000ULL /* File is volatile = created
2477 #define MDS_OPEN_LEASE 01000000000000ULL /* Open the file and grant lease
2478 * delegation, succeed if it's not
2479 * being opened with conflict mode.
2481 #define MDS_OPEN_RELEASE 02000000000000ULL /* Open the file for HSM release */
2483 /* lustre internal open flags, which should not be set from user space */
2484 #define MDS_OPEN_FL_INTERNAL (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS | \
2485 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK | \
2486 MDS_OPEN_BY_FID | MDS_OPEN_LEASE | \
2489 /* permission for create non-directory file */
2490 #define MAY_CREATE (1 << 7)
2491 /* permission for create directory file */
2492 #define MAY_LINK (1 << 8)
2493 /* permission for delete from the directory */
2494 #define MAY_UNLINK (1 << 9)
2495 /* source's permission for rename */
2496 #define MAY_RENAME_SRC (1 << 10)
2497 /* target's permission for rename */
2498 #define MAY_RENAME_TAR (1 << 11)
2499 /* part (parent's) VTX permission check */
2500 #define MAY_VTX_PART (1 << 12)
2501 /* full VTX permission check */
2502 #define MAY_VTX_FULL (1 << 13)
2503 /* lfs rgetfacl permission check */
2504 #define MAY_RGETFACL (1 << 14)
2507 MDS_CHECK_SPLIT = 1 << 0,
2508 MDS_CROSS_REF = 1 << 1,
2509 MDS_VTX_BYPASS = 1 << 2,
2510 MDS_PERM_BYPASS = 1 << 3,
2512 MDS_QUOTA_IGNORE = 1 << 5,
2513 /* Was MDS_CLOSE_CLEANUP (1 << 6), No more used */
2514 MDS_KEEP_ORPHAN = 1 << 7,
2515 MDS_RECOV_OPEN = 1 << 8,
2516 MDS_DATA_MODIFIED = 1 << 9,
2517 MDS_CREATE_VOLATILE = 1 << 10,
2518 MDS_OWNEROVERRIDE = 1 << 11,
2519 MDS_HSM_RELEASE = 1 << 12,
2520 MDS_RENAME_MIGRATE = 1 << 13,
2523 /* instance of mdt_reint_rec */
2524 struct mdt_rec_create {
2532 __u32 cr_suppgid1_h;
2534 __u32 cr_suppgid2_h;
2535 struct lu_fid cr_fid1;
2536 struct lu_fid cr_fid2;
2537 struct lustre_handle cr_old_handle; /* handle in case of open replay */
2541 __u64 cr_padding_1; /* rr_blocks */
2544 /* use of helpers set/get_mrc_cr_flags() is needed to access
2545 * 64 bits cr_flags [cr_flags_l, cr_flags_h], this is done to
2546 * extend cr_flags size without breaking 1.8 compat */
2547 __u32 cr_flags_l; /* for use with open, low 32 bits */
2548 __u32 cr_flags_h; /* for use with open, high 32 bits */
2549 __u32 cr_umask; /* umask for create */
2550 __u32 cr_padding_4; /* rr_padding_4 */
2553 static inline void set_mrc_cr_flags(struct mdt_rec_create *mrc, __u64 flags)
2555 mrc->cr_flags_l = (__u32)(flags & 0xFFFFFFFFUll);
2556 mrc->cr_flags_h = (__u32)(flags >> 32);
2559 static inline __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
2561 return ((__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32));
2564 /* instance of mdt_reint_rec */
2565 struct mdt_rec_link {
2573 __u32 lk_suppgid1_h;
2575 __u32 lk_suppgid2_h;
2576 struct lu_fid lk_fid1;
2577 struct lu_fid lk_fid2;
2579 __u64 lk_padding_1; /* rr_atime */
2580 __u64 lk_padding_2; /* rr_ctime */
2581 __u64 lk_padding_3; /* rr_size */
2582 __u64 lk_padding_4; /* rr_blocks */
2584 __u32 lk_padding_5; /* rr_mode */
2585 __u32 lk_padding_6; /* rr_flags */
2586 __u32 lk_padding_7; /* rr_padding_2 */
2587 __u32 lk_padding_8; /* rr_padding_3 */
2588 __u32 lk_padding_9; /* rr_padding_4 */
2591 /* instance of mdt_reint_rec */
2592 struct mdt_rec_unlink {
2600 __u32 ul_suppgid1_h;
2602 __u32 ul_suppgid2_h;
2603 struct lu_fid ul_fid1;
2604 struct lu_fid ul_fid2;
2606 __u64 ul_padding_2; /* rr_atime */
2607 __u64 ul_padding_3; /* rr_ctime */
2608 __u64 ul_padding_4; /* rr_size */
2609 __u64 ul_padding_5; /* rr_blocks */
2612 __u32 ul_padding_6; /* rr_flags */
2613 __u32 ul_padding_7; /* rr_padding_2 */
2614 __u32 ul_padding_8; /* rr_padding_3 */
2615 __u32 ul_padding_9; /* rr_padding_4 */
2618 /* instance of mdt_reint_rec */
2619 struct mdt_rec_rename {
2627 __u32 rn_suppgid1_h;
2629 __u32 rn_suppgid2_h;
2630 struct lu_fid rn_fid1;
2631 struct lu_fid rn_fid2;
2633 __u64 rn_padding_1; /* rr_atime */
2634 __u64 rn_padding_2; /* rr_ctime */
2635 __u64 rn_padding_3; /* rr_size */
2636 __u64 rn_padding_4; /* rr_blocks */
2637 __u32 rn_bias; /* some operation flags */
2638 __u32 rn_mode; /* cross-ref rename has mode */
2639 __u32 rn_padding_5; /* rr_flags */
2640 __u32 rn_padding_6; /* rr_padding_2 */
2641 __u32 rn_padding_7; /* rr_padding_3 */
2642 __u32 rn_padding_8; /* rr_padding_4 */
2645 /* instance of mdt_reint_rec */
2646 struct mdt_rec_setxattr {
2654 __u32 sx_suppgid1_h;
2656 __u32 sx_suppgid2_h;
2657 struct lu_fid sx_fid;
2658 __u64 sx_padding_1; /* These three are rr_fid2 */
2663 __u64 sx_padding_5; /* rr_ctime */
2664 __u64 sx_padding_6; /* rr_size */
2665 __u64 sx_padding_7; /* rr_blocks */
2668 __u32 sx_padding_8; /* rr_flags */
2669 __u32 sx_padding_9; /* rr_padding_2 */
2670 __u32 sx_padding_10; /* rr_padding_3 */
2671 __u32 sx_padding_11; /* rr_padding_4 */
2675 * mdt_rec_reint is the template for all mdt_reint_xxx structures.
2676 * Do NOT change the size of various members, otherwise the value
2677 * will be broken in lustre_swab_mdt_rec_reint().
2679 * If you add new members in other mdt_reint_xxx structres and need to use the
2680 * rr_padding_x fields, then update lustre_swab_mdt_rec_reint() also.
2682 struct mdt_rec_reint {
2690 __u32 rr_suppgid1_h;
2692 __u32 rr_suppgid2_h;
2693 struct lu_fid rr_fid1;
2694 struct lu_fid rr_fid2;
2705 __u32 rr_padding_4; /* also fix lustre_swab_mdt_rec_reint */
2708 extern void lustre_swab_mdt_rec_reint(struct mdt_rec_reint *rr);
2710 /* lmv structures */
2712 __u32 ld_tgt_count; /* how many MDS's */
2713 __u32 ld_active_tgt_count; /* how many active */
2714 __u32 ld_default_stripe_count; /* how many objects are used */
2715 __u32 ld_pattern; /* default hash pattern */
2716 __u64 ld_default_hash_size;
2717 __u64 ld_padding_1; /* also fix lustre_swab_lmv_desc */
2718 __u32 ld_padding_2; /* also fix lustre_swab_lmv_desc */
2719 __u32 ld_qos_maxage; /* in second */
2720 __u32 ld_padding_3; /* also fix lustre_swab_lmv_desc */
2721 __u32 ld_padding_4; /* also fix lustre_swab_lmv_desc */
2722 struct obd_uuid ld_uuid;
2725 extern void lustre_swab_lmv_desc (struct lmv_desc *ld);
2727 /* LMV layout EA, and it will be stored both in master and slave object */
2728 struct lmv_mds_md_v1 {
2730 __u32 lmv_stripe_count;
2731 __u32 lmv_master_mdt_index; /* On master object, it is master
2732 * MDT index, on slave object, it
2733 * is stripe index of the slave obj */
2734 __u32 lmv_hash_type; /* dir stripe policy, i.e. indicate
2735 * which hash function to be used,
2736 * Note: only lower 16 bits is being
2737 * used for now. Higher 16 bits will
2738 * be used to mark the object status,
2739 * for example migrating or dead. */
2740 __u32 lmv_layout_version; /* Used for directory restriping */
2744 char lmv_pool_name[LOV_MAXPOOLNAME + 1]; /* pool name */
2745 struct lu_fid lmv_stripe_fids[0]; /* FIDs for each stripe */
2748 #define LMV_MAGIC_V1 0x0CD20CD0 /* normal stripe lmv magic */
2749 #define LMV_MAGIC LMV_MAGIC_V1
2751 /* #define LMV_USER_MAGIC 0x0CD30CD0 */
2752 #define LMV_MAGIC_STRIPE 0x0CD40CD0 /* magic for dir sub_stripe */
2754 /* Right now only the lower part(0-16bits) of lmv_hash_type is being used,
2755 * and the higher part will be the flag to indicate the status of object,
2756 * for example the object is being migrated. And the hash function
2757 * might be interpreted differently with different flags. */
2758 #define LMV_HASH_TYPE_MASK 0x0000ffff
2760 #define LMV_HASH_FLAG_MIGRATION 0x80000000
2761 #define LMV_HASH_FLAG_DEAD 0x40000000
2762 #define LMV_HASH_FLAG_BAD_TYPE 0x20000000
2764 /* The striped directory has ever lost its master LMV EA, then LFSCK
2765 * re-generated it. This flag is used to indicate such case. It is an
2767 #define LMV_HASH_FLAG_LOST_LMV 0x10000000
2770 * The FNV-1a hash algorithm is as follows:
2771 * hash = FNV_offset_basis
2772 * for each octet_of_data to be hashed
2773 * hash = hash XOR octet_of_data
2774 * hash = hash × FNV_prime
2776 * http://en.wikipedia.org/wiki/Fowler–Noll–Vo_hash_function#FNV-1a_hash
2778 * http://www.isthe.com/chongo/tech/comp/fnv/index.html#FNV-reference-source
2779 * FNV_prime is 2^40 + 2^8 + 0xb3 = 0x100000001b3ULL
2781 #define LUSTRE_FNV_1A_64_PRIME 0x100000001b3ULL
2782 #define LUSTRE_FNV_1A_64_OFFSET_BIAS 0xcbf29ce484222325ULL
2783 static inline __u64 lustre_hash_fnv_1a_64(const void *buf, size_t size)
2785 __u64 hash = LUSTRE_FNV_1A_64_OFFSET_BIAS;
2786 const unsigned char *p = buf;
2789 for (i = 0; i < size; i++) {
2791 hash *= LUSTRE_FNV_1A_64_PRIME;
2799 struct lmv_mds_md_v1 lmv_md_v1;
2800 struct lmv_user_md lmv_user_md;
2803 extern void lustre_swab_lmv_mds_md(union lmv_mds_md *lmm);
2805 static inline int lmv_mds_md_size(int stripe_count, unsigned int lmm_magic)
2807 switch (lmm_magic) {
2809 struct lmv_mds_md_v1 *lmm1;
2811 return sizeof(*lmm1) + stripe_count *
2812 sizeof(lmm1->lmv_stripe_fids[0]);
2819 static inline int lmv_mds_md_stripe_count_get(const union lmv_mds_md *lmm)
2821 switch (le32_to_cpu(lmm->lmv_magic)) {
2823 return le32_to_cpu(lmm->lmv_md_v1.lmv_stripe_count);
2824 case LMV_USER_MAGIC:
2825 return le32_to_cpu(lmm->lmv_user_md.lum_stripe_count);
2831 static inline int lmv_mds_md_stripe_count_set(union lmv_mds_md *lmm,
2832 unsigned int stripe_count)
2834 switch (le32_to_cpu(lmm->lmv_magic)) {
2836 lmm->lmv_md_v1.lmv_stripe_count = cpu_to_le32(stripe_count);
2838 case LMV_USER_MAGIC:
2839 lmm->lmv_user_md.lum_stripe_count = cpu_to_le32(stripe_count);
2851 FLD_FIRST_OPC = FLD_QUERY
2857 SEQ_FIRST_OPC = SEQ_QUERY
2861 SEQ_ALLOC_SUPER = 0,
2873 LFSCK_NOTIFY = 1101,
2876 LFSCK_FIRST_OPC = LFSCK_NOTIFY
2880 * LOV data structures
2883 #define LOV_MAX_UUID_BUFFER_SIZE 8192
2884 /* The size of the buffer the lov/mdc reserves for the
2885 * array of UUIDs returned by the MDS. With the current
2886 * protocol, this will limit the max number of OSTs per LOV */
2888 #define LOV_DESC_MAGIC 0xB0CCDE5C
2889 #define LOV_DESC_QOS_MAXAGE_DEFAULT 5 /* Seconds */
2890 #define LOV_DESC_STRIPE_SIZE_DEFAULT (1 << LNET_MTU_BITS)
2892 /* LOV settings descriptor (should only contain static info) */
2894 __u32 ld_tgt_count; /* how many OBD's */
2895 __u32 ld_active_tgt_count; /* how many active */
2896 __u32 ld_default_stripe_count; /* how many objects are used */
2897 __u32 ld_pattern; /* default PATTERN_RAID0 */
2898 __u64 ld_default_stripe_size; /* in bytes */
2899 __u64 ld_default_stripe_offset; /* in bytes */
2900 __u32 ld_padding_0; /* unused */
2901 __u32 ld_qos_maxage; /* in second */
2902 __u32 ld_padding_1; /* also fix lustre_swab_lov_desc */
2903 __u32 ld_padding_2; /* also fix lustre_swab_lov_desc */
2904 struct obd_uuid ld_uuid;
2907 #define ld_magic ld_active_tgt_count /* for swabbing from llogs */
2909 extern void lustre_swab_lov_desc (struct lov_desc *ld);
2914 /* opcodes -- MUST be distinct from OST/MDS opcodes */
2919 LDLM_BL_CALLBACK = 104,
2920 LDLM_CP_CALLBACK = 105,
2921 LDLM_GL_CALLBACK = 106,
2922 LDLM_SET_INFO = 107,
2925 #define LDLM_FIRST_OPC LDLM_ENQUEUE
2927 #define RES_NAME_SIZE 4
2928 struct ldlm_res_id {
2929 __u64 name[RES_NAME_SIZE];
2932 #define DLDLMRES "["LPX64":"LPX64":"LPX64"]."LPX64i
2933 #define PLDLMRES(res) (res)->lr_name.name[0], (res)->lr_name.name[1], \
2934 (res)->lr_name.name[2], (res)->lr_name.name[3]
2936 extern void lustre_swab_ldlm_res_id (struct ldlm_res_id *id);
2938 static inline bool ldlm_res_eq(const struct ldlm_res_id *res0,
2939 const struct ldlm_res_id *res1)
2941 return memcmp(res0, res1, sizeof(*res0)) == 0;
2958 #define LCK_MODE_NUM 8
2968 #define LDLM_MIN_TYPE LDLM_PLAIN
2970 struct ldlm_extent {
2976 static inline int ldlm_extent_overlap(const struct ldlm_extent *ex1,
2977 const struct ldlm_extent *ex2)
2979 return ex1->start <= ex2->end && ex2->start <= ex1->end;
2982 /* check if @ex1 contains @ex2 */
2983 static inline int ldlm_extent_contain(const struct ldlm_extent *ex1,
2984 const struct ldlm_extent *ex2)
2986 return ex1->start <= ex2->start && ex1->end >= ex2->end;
2989 struct ldlm_inodebits {
2993 struct ldlm_flock_wire {
3001 /* it's important that the fields of the ldlm_extent structure match
3002 * the first fields of the ldlm_flock structure because there is only
3003 * one ldlm_swab routine to process the ldlm_policy_data_t union. if
3004 * this ever changes we will need to swab the union differently based
3005 * on the resource type. */
3008 struct ldlm_extent l_extent;
3009 struct ldlm_flock_wire l_flock;
3010 struct ldlm_inodebits l_inodebits;
3011 } ldlm_wire_policy_data_t;
3013 extern void lustre_swab_ldlm_policy_data (ldlm_wire_policy_data_t *d);
3015 union ldlm_gl_desc {
3016 struct ldlm_gl_lquota_desc lquota_desc;
3019 extern void lustre_swab_gl_desc(union ldlm_gl_desc *);
3021 struct ldlm_intent {
3025 extern void lustre_swab_ldlm_intent (struct ldlm_intent *i);
3027 struct ldlm_resource_desc {
3028 ldlm_type_t lr_type;
3029 __u32 lr_padding; /* also fix lustre_swab_ldlm_resource_desc */
3030 struct ldlm_res_id lr_name;
3033 extern void lustre_swab_ldlm_resource_desc (struct ldlm_resource_desc *r);
3035 struct ldlm_lock_desc {
3036 struct ldlm_resource_desc l_resource;
3037 ldlm_mode_t l_req_mode;
3038 ldlm_mode_t l_granted_mode;
3039 ldlm_wire_policy_data_t l_policy_data;
3042 extern void lustre_swab_ldlm_lock_desc (struct ldlm_lock_desc *l);
3044 #define LDLM_LOCKREQ_HANDLES 2
3045 #define LDLM_ENQUEUE_CANCEL_OFF 1
3047 struct ldlm_request {
3050 struct ldlm_lock_desc lock_desc;
3051 struct lustre_handle lock_handle[LDLM_LOCKREQ_HANDLES];
3054 extern void lustre_swab_ldlm_request (struct ldlm_request *rq);
3056 /* If LDLM_ENQUEUE, 1 slot is already occupied, 1 is available.
3057 * Otherwise, 2 are available. */
3058 #define ldlm_request_bufsize(count,type) \
3060 int _avail = LDLM_LOCKREQ_HANDLES; \
3061 _avail -= (type == LDLM_ENQUEUE ? LDLM_ENQUEUE_CANCEL_OFF : 0); \
3062 sizeof(struct ldlm_request) + \
3063 (count > _avail ? count - _avail : 0) * \
3064 sizeof(struct lustre_handle); \
3069 __u32 lock_padding; /* also fix lustre_swab_ldlm_reply */
3070 struct ldlm_lock_desc lock_desc;
3071 struct lustre_handle lock_handle;
3072 __u64 lock_policy_res1;
3073 __u64 lock_policy_res2;
3076 extern void lustre_swab_ldlm_reply (struct ldlm_reply *r);
3078 #define ldlm_flags_to_wire(flags) ((__u32)(flags))
3079 #define ldlm_flags_from_wire(flags) ((__u64)(flags))
3082 * Opcodes for mountconf (mgs and mgc)
3087 MGS_EXCEPTION, /* node died, etc. */
3088 MGS_TARGET_REG, /* whenever target starts up */
3094 #define MGS_FIRST_OPC MGS_CONNECT
3096 #define MGS_PARAM_MAXLEN 1024
3097 #define KEY_SET_INFO "set_info"
3099 struct mgs_send_param {
3100 char mgs_param[MGS_PARAM_MAXLEN];
3103 /* We pass this info to the MGS so it can write config logs */
3104 #define MTI_NAME_MAXLEN 64
3105 #define MTI_PARAM_MAXLEN 4096
3106 #define MTI_NIDS_MAX 32
3107 struct mgs_target_info {
3108 __u32 mti_lustre_ver;
3109 __u32 mti_stripe_index;
3110 __u32 mti_config_ver;
3112 __u32 mti_nid_count;
3113 __u32 mti_instance; /* Running instance of target */
3114 char mti_fsname[MTI_NAME_MAXLEN];
3115 char mti_svname[MTI_NAME_MAXLEN];
3116 char mti_uuid[sizeof(struct obd_uuid)];
3117 __u64 mti_nids[MTI_NIDS_MAX]; /* host nids (lnet_nid_t)*/
3118 char mti_params[MTI_PARAM_MAXLEN];
3120 extern void lustre_swab_mgs_target_info(struct mgs_target_info *oinfo);
3122 struct mgs_nidtbl_entry {
3123 __u64 mne_version; /* table version of this entry */
3124 __u32 mne_instance; /* target instance # */
3125 __u32 mne_index; /* target index */
3126 __u32 mne_length; /* length of this entry - by bytes */
3127 __u8 mne_type; /* target type LDD_F_SV_TYPE_OST/MDT */
3128 __u8 mne_nid_type; /* type of nid(mbz). for ipv6. */
3129 __u8 mne_nid_size; /* size of each NID, by bytes */
3130 __u8 mne_nid_count; /* # of NIDs in buffer */
3132 lnet_nid_t nids[0]; /* variable size buffer for NIDs. */
3135 extern void lustre_swab_mgs_nidtbl_entry(struct mgs_nidtbl_entry *oinfo);
3137 struct mgs_config_body {
3138 char mcb_name[MTI_NAME_MAXLEN]; /* logname */
3139 __u64 mcb_offset; /* next index of config log to request */
3140 __u16 mcb_type; /* type of log: CONFIG_T_[CONFIG|RECOVER] */
3142 __u8 mcb_bits; /* bits unit size of config log */
3143 __u32 mcb_units; /* # of units for bulk transfer */
3145 extern void lustre_swab_mgs_config_body(struct mgs_config_body *body);
3147 struct mgs_config_res {
3148 __u64 mcr_offset; /* index of last config log */
3149 __u64 mcr_size; /* size of the log */
3151 extern void lustre_swab_mgs_config_res(struct mgs_config_res *body);
3153 /* Config marker flags (in config log) */
3154 #define CM_START 0x01
3156 #define CM_SKIP 0x04
3157 #define CM_UPGRADE146 0x08
3158 #define CM_EXCLUDE 0x10
3159 #define CM_START_SKIP (CM_START | CM_SKIP)
3162 __u32 cm_step; /* aka config version */
3164 __u32 cm_vers; /* lustre release version number */
3165 __u32 cm_padding; /* 64 bit align */
3166 __s64 cm_createtime; /*when this record was first created */
3167 __s64 cm_canceltime; /*when this record is no longer valid*/
3168 char cm_tgtname[MTI_NAME_MAXLEN];
3169 char cm_comment[MTI_NAME_MAXLEN];
3172 extern void lustre_swab_cfg_marker(struct cfg_marker *marker,
3173 int swab, int size);
3176 * Opcodes for multiple servers.
3186 #define OBD_FIRST_OPC OBD_PING
3189 * llog contexts indices.
3191 * There is compatibility problem with indexes below, they are not
3192 * continuous and must keep their numbers for compatibility needs.
3193 * See LU-5218 for details.
3196 LLOG_CONFIG_ORIG_CTXT = 0,
3197 LLOG_CONFIG_REPL_CTXT = 1,
3198 LLOG_MDS_OST_ORIG_CTXT = 2,
3199 LLOG_MDS_OST_REPL_CTXT = 3, /* kept just to avoid re-assignment */
3200 LLOG_SIZE_ORIG_CTXT = 4,
3201 LLOG_SIZE_REPL_CTXT = 5,
3202 LLOG_TEST_ORIG_CTXT = 8,
3203 LLOG_TEST_REPL_CTXT = 9, /* kept just to avoid re-assignment */
3204 LLOG_CHANGELOG_ORIG_CTXT = 12, /**< changelog generation on mdd */
3205 LLOG_CHANGELOG_REPL_CTXT = 13, /**< changelog access on clients */
3206 /* for multiple changelog consumers */
3207 LLOG_CHANGELOG_USER_ORIG_CTXT = 14,
3208 LLOG_AGENT_ORIG_CTXT = 15, /**< agent requests generation on cdt */
3212 /** Identifier for a single log object */
3214 struct ost_id lgl_oi;
3216 } __attribute__((packed));
3218 /** Records written to the CATALOGS list */
3219 #define CATLIST "CATALOGS"
3221 struct llog_logid lci_logid;
3225 } __attribute__((packed));
3227 /* Log data record types - there is no specific reason that these need to
3228 * be related to the RPC opcodes, but no reason not to (may be handy later?)
3230 #define LLOG_OP_MAGIC 0x10600000
3231 #define LLOG_OP_MASK 0xfff00000
3234 LLOG_PAD_MAGIC = LLOG_OP_MAGIC | 0x00000,
3235 OST_SZ_REC = LLOG_OP_MAGIC | 0x00f00,
3236 /* OST_RAID1_REC = LLOG_OP_MAGIC | 0x01000, never used */
3237 MDS_UNLINK_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) |
3238 REINT_UNLINK, /* obsolete after 2.5.0 */
3239 MDS_UNLINK64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
3241 /* MDS_SETATTR_REC = LLOG_OP_MAGIC | 0x12401, obsolete 1.8.0 */
3242 MDS_SETATTR64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
3244 OBD_CFG_REC = LLOG_OP_MAGIC | 0x20000,
3245 /* PTL_CFG_REC = LLOG_OP_MAGIC | 0x30000, obsolete 1.4.0 */
3246 LLOG_GEN_REC = LLOG_OP_MAGIC | 0x40000,
3247 /* LLOG_JOIN_REC = LLOG_OP_MAGIC | 0x50000, obsolete 1.8.0 */
3248 CHANGELOG_REC = LLOG_OP_MAGIC | 0x60000,
3249 CHANGELOG_USER_REC = LLOG_OP_MAGIC | 0x70000,
3250 HSM_AGENT_REC = LLOG_OP_MAGIC | 0x80000,
3251 LLOG_HDR_MAGIC = LLOG_OP_MAGIC | 0x45539,
3252 LLOG_LOGID_MAGIC = LLOG_OP_MAGIC | 0x4553b,
3255 #define LLOG_REC_HDR_NEEDS_SWABBING(r) \
3256 (((r)->lrh_type & __swab32(LLOG_OP_MASK)) == __swab32(LLOG_OP_MAGIC))
3258 /** Log record header - stored in little endian order.
3259 * Each record must start with this struct, end with a llog_rec_tail,
3260 * and be a multiple of 256 bits in size.
3262 struct llog_rec_hdr {
3269 struct llog_rec_tail {
3274 /* Where data follow just after header */
3275 #define REC_DATA(ptr) \
3276 ((void *)((char *)ptr + sizeof(struct llog_rec_hdr)))
3278 #define REC_DATA_LEN(rec) \
3279 (rec->lrh_len - sizeof(struct llog_rec_hdr) - \
3280 sizeof(struct llog_rec_tail))
3282 static inline void *rec_tail(struct llog_rec_hdr *rec)
3284 return (void *)((char *)rec + rec->lrh_len -
3285 sizeof(struct llog_rec_tail));
3288 struct llog_logid_rec {
3289 struct llog_rec_hdr lid_hdr;
3290 struct llog_logid lid_id;
3294 struct llog_rec_tail lid_tail;
3295 } __attribute__((packed));
3297 struct llog_unlink_rec {
3298 struct llog_rec_hdr lur_hdr;
3302 struct llog_rec_tail lur_tail;
3303 } __attribute__((packed));
3305 struct llog_unlink64_rec {
3306 struct llog_rec_hdr lur_hdr;
3307 struct lu_fid lur_fid;
3308 __u32 lur_count; /* to destroy the lost precreated */
3312 struct llog_rec_tail lur_tail;
3313 } __attribute__((packed));
3315 struct llog_setattr64_rec {
3316 struct llog_rec_hdr lsr_hdr;
3317 struct ost_id lsr_oi;
3323 struct llog_rec_tail lsr_tail;
3324 } __attribute__((packed));
3326 struct llog_size_change_rec {
3327 struct llog_rec_hdr lsc_hdr;
3328 struct ll_fid lsc_fid;
3333 struct llog_rec_tail lsc_tail;
3334 } __attribute__((packed));
3336 #define CHANGELOG_MAGIC 0xca103000
3338 /** \a changelog_rec_type's that can't be masked */
3339 #define CHANGELOG_MINMASK (1 << CL_MARK)
3340 /** bits covering all \a changelog_rec_type's */
3341 #define CHANGELOG_ALLMASK 0XFFFFFFFF
3342 /** default \a changelog_rec_type mask. Allow all of them, except
3343 * CL_ATIME since it can really be time consuming, and not necessary
3344 * under normal use. */
3345 #define CHANGELOG_DEFMASK (CHANGELOG_ALLMASK & ~(1 << CL_ATIME))
3347 /* changelog llog name, needed by client replicators */
3348 #define CHANGELOG_CATALOG "changelog_catalog"
3350 struct changelog_setinfo {
3353 } __attribute__((packed));
3355 /** changelog record */
3356 struct llog_changelog_rec {
3357 struct llog_rec_hdr cr_hdr;
3358 struct changelog_rec cr; /**< Variable length field */
3359 struct llog_rec_tail cr_do_not_use; /**< for_sizeof_only */
3360 } __attribute__((packed));
3362 #define CHANGELOG_USER_PREFIX "cl"
3364 struct llog_changelog_user_rec {
3365 struct llog_rec_hdr cur_hdr;
3369 struct llog_rec_tail cur_tail;
3370 } __attribute__((packed));
3372 enum agent_req_status {
3380 static inline const char *agent_req_status2name(enum agent_req_status ars)
3398 static inline bool agent_req_in_final_state(enum agent_req_status ars)
3400 return ((ars == ARS_SUCCEED) || (ars == ARS_FAILED) ||
3401 (ars == ARS_CANCELED));
3404 struct llog_agent_req_rec {
3405 struct llog_rec_hdr arr_hdr; /**< record header */
3406 __u32 arr_status; /**< status of the request */
3408 * agent_req_status */
3409 __u32 arr_archive_id; /**< backend archive number */
3410 __u64 arr_flags; /**< req flags */
3411 __u64 arr_compound_id; /**< compound cookie */
3412 __u64 arr_req_create; /**< req. creation time */
3413 __u64 arr_req_change; /**< req. status change time */
3414 struct hsm_action_item arr_hai; /**< req. to the agent */
3415 struct llog_rec_tail arr_tail; /**< record tail for_sizezof_only */
3416 } __attribute__((packed));
3418 /* Old llog gen for compatibility */
3422 } __attribute__((packed));
3424 struct llog_gen_rec {
3425 struct llog_rec_hdr lgr_hdr;
3426 struct llog_gen lgr_gen;
3430 struct llog_rec_tail lgr_tail;
3433 /* On-disk header structure of each log object, stored in little endian order */
3434 #define LLOG_CHUNK_SIZE 8192
3435 #define LLOG_HEADER_SIZE (96)
3436 #define LLOG_BITMAP_BYTES (LLOG_CHUNK_SIZE - LLOG_HEADER_SIZE)
3438 #define LLOG_MIN_REC_SIZE (24) /* round(llog_rec_hdr + llog_rec_tail) */
3440 /* flags for the logs */
3442 LLOG_F_ZAP_WHEN_EMPTY = 0x1,
3443 LLOG_F_IS_CAT = 0x2,
3444 LLOG_F_IS_PLAIN = 0x4,
3445 LLOG_F_EXT_JOBID = 0x8,
3447 LLOG_F_EXT_MASK = LLOG_F_EXT_JOBID,
3450 struct llog_log_hdr {
3451 struct llog_rec_hdr llh_hdr;
3452 __s64 llh_timestamp;
3454 __u32 llh_bitmap_offset;
3458 /* for a catalog the first plain slot is next to it */
3459 struct obd_uuid llh_tgtuuid;
3460 __u32 llh_reserved[LLOG_HEADER_SIZE/sizeof(__u32) - 23];
3461 __u32 llh_bitmap[LLOG_BITMAP_BYTES/sizeof(__u32)];
3462 struct llog_rec_tail llh_tail;
3463 } __attribute__((packed));
3465 #define LLOG_BITMAP_SIZE(llh) (__u32)((llh->llh_hdr.lrh_len - \
3466 llh->llh_bitmap_offset - \
3467 sizeof(llh->llh_tail)) * 8)
3469 /** log cookies are used to reference a specific log file and a record therein */
3470 struct llog_cookie {
3471 struct llog_logid lgc_lgl;
3475 } __attribute__((packed));
3477 /** llog protocol */
3478 enum llogd_rpc_ops {
3479 LLOG_ORIGIN_HANDLE_CREATE = 501,
3480 LLOG_ORIGIN_HANDLE_NEXT_BLOCK = 502,
3481 LLOG_ORIGIN_HANDLE_READ_HEADER = 503,
3482 LLOG_ORIGIN_HANDLE_WRITE_REC = 504,
3483 LLOG_ORIGIN_HANDLE_CLOSE = 505,
3484 LLOG_ORIGIN_CONNECT = 506,
3485 LLOG_CATINFO = 507, /* deprecated */
3486 LLOG_ORIGIN_HANDLE_PREV_BLOCK = 508,
3487 LLOG_ORIGIN_HANDLE_DESTROY = 509, /* for destroy llog object*/
3489 LLOG_FIRST_OPC = LLOG_ORIGIN_HANDLE_CREATE
3493 struct llog_logid lgd_logid;
3495 __u32 lgd_llh_flags;
3497 __u32 lgd_saved_index;
3499 __u64 lgd_cur_offset;
3500 } __attribute__((packed));
3502 struct llogd_conn_body {
3503 struct llog_gen lgdc_gen;
3504 struct llog_logid lgdc_logid;
3505 __u32 lgdc_ctxt_idx;
3506 } __attribute__((packed));
3508 /* Note: 64-bit types are 64-bit aligned in structure */
3510 __u64 o_valid; /* hot fields in this obdo */
3513 __u64 o_size; /* o_size-o_blocks == ost_lvb */
3517 __u64 o_blocks; /* brw: cli sent cached bytes */
3520 /* 32-bit fields start here: keep an even number of them via padding */
3521 __u32 o_blksize; /* optimal IO blocksize */
3522 __u32 o_mode; /* brw: cli sent cache remain */
3526 __u32 o_nlink; /* brw: checksum */
3528 __u32 o_misc; /* brw: o_dropped */
3530 __u64 o_ioepoch; /* epoch in ost writes */
3531 __u32 o_stripe_idx; /* holds stripe idx */
3533 struct lustre_handle o_handle; /* brw: lock handle to prolong
3535 struct llog_cookie o_lcookie; /* destroy: unlink cookie from
3540 __u64 o_data_version; /* getattr: sum of iversion for
3542 * brw: grant space consumed on
3543 * the client for the write */
3549 #define o_dirty o_blocks
3550 #define o_undirty o_mode
3551 #define o_dropped o_misc
3552 #define o_cksum o_nlink
3553 #define o_grant_used o_data_version
3555 struct lfsck_request {
3568 __u16 lr_async_windows;
3570 struct lu_fid lr_fid;
3571 struct lu_fid lr_fid2;
3572 struct lu_fid lr_fid3;
3577 void lustre_swab_lfsck_request(struct lfsck_request *lr);
3579 struct lfsck_reply {
3585 void lustre_swab_lfsck_reply(struct lfsck_reply *lr);
3588 LE_LASTID_REBUILDING = 1,
3589 LE_LASTID_REBUILT = 2,
3595 LE_FID_ACCESSED = 8,
3597 LE_CONDITIONAL_DESTROY = 10,
3598 LE_PAIRS_VERIFY = 11,
3599 LE_SKIP_NLINK_DECLARE = 13,
3601 LE_SET_LMV_MASTER = 15,
3602 LE_SET_LMV_SLAVE = 16,
3605 enum lfsck_event_flags {
3606 LEF_TO_OST = 0x00000001,
3607 LEF_FROM_OST = 0x00000002,
3608 LEF_SET_LMV_HASH = 0x00000004,
3609 LEF_SET_LMV_ALL = 0x00000008,
3610 LEF_RECHECK_NAME_HASH = 0x00000010,
3613 static inline void lustre_set_wire_obdo(const struct obd_connect_data *ocd,
3615 const struct obdo *lobdo)
3618 wobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
3622 if (unlikely(!(ocd->ocd_connect_flags & OBD_CONNECT_FID)) &&
3623 fid_seq_is_echo(ostid_seq(&lobdo->o_oi))) {
3624 /* Currently OBD_FL_OSTID will only be used when 2.4 echo
3625 * client communicate with pre-2.4 server */
3626 wobdo->o_oi.oi.oi_id = fid_oid(&lobdo->o_oi.oi_fid);
3627 wobdo->o_oi.oi.oi_seq = fid_seq(&lobdo->o_oi.oi_fid);
3631 static inline void lustre_get_wire_obdo(const struct obd_connect_data *ocd,
3633 const struct obdo *wobdo)
3635 __u32 local_flags = 0;
3637 if (lobdo->o_valid & OBD_MD_FLFLAGS)
3638 local_flags = lobdo->o_flags & OBD_FL_LOCAL_MASK;
3641 if (local_flags != 0) {
3642 lobdo->o_valid |= OBD_MD_FLFLAGS;
3643 lobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
3644 lobdo->o_flags |= local_flags;
3649 if (unlikely(!(ocd->ocd_connect_flags & OBD_CONNECT_FID)) &&
3650 fid_seq_is_echo(wobdo->o_oi.oi.oi_seq)) {
3652 lobdo->o_oi.oi_fid.f_seq = wobdo->o_oi.oi.oi_seq;
3653 lobdo->o_oi.oi_fid.f_oid = wobdo->o_oi.oi.oi_id;
3654 lobdo->o_oi.oi_fid.f_ver = 0;
3658 extern void lustre_swab_obdo (struct obdo *o);
3660 /* request structure for OST's */
3665 /* Key for FIEMAP to be used in get_info calls */
3666 struct ll_fiemap_info_key {
3669 struct ll_user_fiemap fiemap;
3672 extern void lustre_swab_ost_body (struct ost_body *b);
3673 extern void lustre_swab_ost_last_id(__u64 *id);
3674 extern void lustre_swab_fiemap(struct ll_user_fiemap *fiemap);
3676 extern void lustre_swab_lov_user_md_v1(struct lov_user_md_v1 *lum);
3677 extern void lustre_swab_lov_user_md_v3(struct lov_user_md_v3 *lum);
3678 extern void lustre_swab_lov_user_md_objects(struct lov_user_ost_data *lod,
3680 extern void lustre_swab_lov_mds_md(struct lov_mds_md *lmm);
3681 void lustre_print_user_md(unsigned int level, struct lov_user_md *lum,
3685 extern void lustre_swab_llogd_body (struct llogd_body *d);
3686 extern void lustre_swab_llog_hdr (struct llog_log_hdr *h);
3687 extern void lustre_swab_llogd_conn_body (struct llogd_conn_body *d);
3688 extern void lustre_swab_llog_rec(struct llog_rec_hdr *rec);
3689 extern void lustre_swab_llog_id(struct llog_logid *lid);
3692 extern void lustre_swab_lustre_cfg(struct lustre_cfg *lcfg);
3694 /* Functions for dumping PTLRPC fields */
3695 void dump_rniobuf(struct niobuf_remote *rnb);
3696 void dump_ioo(struct obd_ioobj *nb);
3697 void dump_obdo(struct obdo *oa);
3698 void dump_ost_body(struct ost_body *ob);
3699 void dump_rcs(__u32 *rc);
3701 #define IDX_INFO_MAGIC 0x3D37CC37
3703 /* Index file transfer through the network. The server serializes the index into
3704 * a byte stream which is sent to the client via a bulk transfer */
3708 /* reply: see idx_info_flags below */
3711 /* request & reply: number of lu_idxpage (to be) transferred */
3715 /* request: requested attributes passed down to the iterator API */
3718 /* request & reply: index file identifier (FID) */
3719 struct lu_fid ii_fid;
3721 /* reply: version of the index file before starting to walk the index.
3722 * Please note that the version can be modified at any time during the
3726 /* request: hash to start with:
3727 * reply: hash of the first entry of the first lu_idxpage and hash
3728 * of the entry to read next if any */
3729 __u64 ii_hash_start;
3732 /* reply: size of keys in lu_idxpages, minimal one if II_FL_VARKEY is
3736 /* reply: size of records in lu_idxpages, minimal one if II_FL_VARREC
3744 extern void lustre_swab_idx_info(struct idx_info *ii);
3746 #define II_END_OFF MDS_DIR_END_OFF /* all entries have been read */
3748 /* List of flags used in idx_info::ii_flags */
3749 enum idx_info_flags {
3750 II_FL_NOHASH = 1 << 0, /* client doesn't care about hash value */
3751 II_FL_VARKEY = 1 << 1, /* keys can be of variable size */
3752 II_FL_VARREC = 1 << 2, /* records can be of variable size */
3753 II_FL_NONUNQ = 1 << 3, /* index supports non-unique keys */
3754 II_FL_NOKEY = 1 << 4, /* client doesn't care about key */
3757 #define LIP_MAGIC 0x8A6D6B6C
3759 /* 4KB (= LU_PAGE_SIZE) container gathering key/record pairs */
3761 /* 16-byte header */
3764 __u16 lip_nr; /* number of entries in the container */
3765 __u64 lip_pad0; /* additional padding for future use */
3767 /* key/record pairs are stored in the remaining 4080 bytes.
3768 * depending upon the flags in idx_info::ii_flags, each key/record
3769 * pair might be preceded by:
3771 * - the key size (II_FL_VARKEY is set)
3772 * - the record size (II_FL_VARREC is set)
3774 * For the time being, we only support fixed-size key & record. */
3775 char lip_entries[0];
3777 extern void lustre_swab_lip_header(struct lu_idxpage *lip);
3779 #define LIP_HDR_SIZE (offsetof(struct lu_idxpage, lip_entries))
3781 /* Gather all possible type associated with a 4KB container */
3783 struct lu_dirpage lp_dir; /* for MDS_READPAGE */
3784 struct lu_idxpage lp_idx; /* for OBD_IDX_READ */
3785 char lp_array[LU_PAGE_SIZE];
3788 /* security opcodes */
3791 SEC_CTX_INIT_CONT = 802,
3794 SEC_FIRST_OPC = SEC_CTX_INIT
3798 * capa related definitions
3800 #define CAPA_HMAC_MAX_LEN 64
3801 #define CAPA_HMAC_KEY_MAX_LEN 56
3803 /* NB take care when changing the sequence of elements this struct,
3804 * because the offset info is used in find_capa() */
3805 struct lustre_capa {
3806 struct lu_fid lc_fid; /** fid */
3807 __u64 lc_opc; /** operations allowed */
3808 __u64 lc_uid; /** file owner */
3809 __u64 lc_gid; /** file group */
3810 __u32 lc_flags; /** HMAC algorithm & flags */
3811 __u32 lc_keyid; /** key# used for the capability */
3812 __u32 lc_timeout; /** capa timeout value (sec) */
3813 __u32 lc_expiry; /** expiry time (sec) */
3814 __u8 lc_hmac[CAPA_HMAC_MAX_LEN]; /** HMAC */
3815 } __attribute__((packed));
3817 extern void lustre_swab_lustre_capa(struct lustre_capa *c);
3819 /** lustre_capa::lc_opc */
3821 CAPA_OPC_BODY_WRITE = 1<<0, /**< write object data */
3822 CAPA_OPC_BODY_READ = 1<<1, /**< read object data */
3823 CAPA_OPC_INDEX_LOOKUP = 1<<2, /**< lookup object fid */
3824 CAPA_OPC_INDEX_INSERT = 1<<3, /**< insert object fid */
3825 CAPA_OPC_INDEX_DELETE = 1<<4, /**< delete object fid */
3826 CAPA_OPC_OSS_WRITE = 1<<5, /**< write oss object data */
3827 CAPA_OPC_OSS_READ = 1<<6, /**< read oss object data */
3828 CAPA_OPC_OSS_TRUNC = 1<<7, /**< truncate oss object */
3829 CAPA_OPC_OSS_DESTROY = 1<<8, /**< destroy oss object */
3830 CAPA_OPC_META_WRITE = 1<<9, /**< write object meta data */
3831 CAPA_OPC_META_READ = 1<<10, /**< read object meta data */
3834 #define CAPA_OPC_OSS_RW (CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE)
3835 #define CAPA_OPC_MDS_ONLY \
3836 (CAPA_OPC_BODY_WRITE | CAPA_OPC_BODY_READ | CAPA_OPC_INDEX_LOOKUP | \
3837 CAPA_OPC_INDEX_INSERT | CAPA_OPC_INDEX_DELETE)
3838 #define CAPA_OPC_OSS_ONLY \
3839 (CAPA_OPC_OSS_WRITE | CAPA_OPC_OSS_READ | CAPA_OPC_OSS_TRUNC | \
3840 CAPA_OPC_OSS_DESTROY)
3841 #define CAPA_OPC_MDS_DEFAULT ~CAPA_OPC_OSS_ONLY
3842 #define CAPA_OPC_OSS_DEFAULT ~(CAPA_OPC_MDS_ONLY | CAPA_OPC_OSS_ONLY)
3844 /* MDS capability covers object capability for operations of body r/w
3845 * (dir readpage/sendpage), index lookup/insert/delete and meta data r/w,
3846 * while OSS capability only covers object capability for operations of
3847 * oss data(file content) r/w/truncate.
3849 static inline int capa_for_mds(struct lustre_capa *c)
3851 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) != 0;
3854 static inline int capa_for_oss(struct lustre_capa *c)
3856 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) == 0;
3859 static inline bool lovea_slot_is_dummy(const struct lov_ost_data_v1 *obj)
3861 /* zero area does not care about the bytes-order. */
3862 if (obj->l_ost_oi.oi.oi_id == 0 && obj->l_ost_oi.oi.oi_seq == 0 &&
3863 obj->l_ost_idx == 0 && obj->l_ost_gen == 0)
3869 /* lustre_capa::lc_hmac_alg */
3871 CAPA_HMAC_ALG_SHA1 = 1, /**< sha1 algorithm */
3875 #define CAPA_FL_MASK 0x00ffffff
3876 #define CAPA_HMAC_ALG_MASK 0xff000000
3878 struct lustre_capa_key {
3879 __u64 lk_seq; /**< mds# */
3880 __u32 lk_keyid; /**< key# */
3882 __u8 lk_key[CAPA_HMAC_KEY_MAX_LEN]; /**< key */
3883 } __attribute__((packed));
3885 extern void lustre_swab_lustre_capa_key(struct lustre_capa_key *k);
3887 /** The link ea holds 1 \a link_ea_entry for each hardlink */
3888 #define LINK_EA_MAGIC 0x11EAF1DFUL
3889 struct link_ea_header {
3892 __u64 leh_len; /* total size */
3898 /** Hardlink data is name and parent fid.
3899 * Stored in this crazy struct for maximum packing and endian-neutrality
3901 struct link_ea_entry {
3902 /** __u16 stored big-endian, unaligned */
3903 unsigned char lee_reclen[2];
3904 unsigned char lee_parent_fid[sizeof(struct lu_fid)];
3906 }__attribute__((packed));
3908 /** fid2path request/reply structure */
3909 struct getinfo_fid2path {
3910 struct lu_fid gf_fid;
3915 } __attribute__((packed));
3917 void lustre_swab_fid2path (struct getinfo_fid2path *gf);
3919 /** path2parent request/reply structures */
3921 struct lu_fid gp_fid; /**< parent FID */
3922 __u32 gp_linkno; /**< hardlink number */
3923 __u32 gp_name_size; /**< size of the name field */
3924 char gp_name[0]; /**< zero-terminated link name */
3925 } __attribute__((packed));
3928 LAYOUT_INTENT_ACCESS = 0,
3929 LAYOUT_INTENT_READ = 1,
3930 LAYOUT_INTENT_WRITE = 2,
3931 LAYOUT_INTENT_GLIMPSE = 3,
3932 LAYOUT_INTENT_TRUNC = 4,
3933 LAYOUT_INTENT_RELEASE = 5,
3934 LAYOUT_INTENT_RESTORE = 6
3937 /* enqueue layout lock with intent */
3938 struct layout_intent {
3939 __u32 li_opc; /* intent operation for enqueue, read, write etc */
3945 void lustre_swab_layout_intent(struct layout_intent *li);
3948 * On the wire version of hsm_progress structure.
3950 * Contains the userspace hsm_progress and some internal fields.
3952 struct hsm_progress_kernel {
3953 /* Field taken from struct hsm_progress */
3956 struct hsm_extent hpk_extent;
3958 __u16 hpk_errval; /* positive val */
3960 /* Additional fields */
3961 __u64 hpk_data_version;
3963 } __attribute__((packed));
3965 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
3966 extern void lustre_swab_hsm_current_action(struct hsm_current_action *action);
3967 extern void lustre_swab_hsm_progress_kernel(struct hsm_progress_kernel *hpk);
3968 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
3969 extern void lustre_swab_hsm_user_item(struct hsm_user_item *hui);
3970 extern void lustre_swab_hsm_request(struct hsm_request *hr);
3973 * OUT_UPDATE RPC Format
3975 * During the cross-ref operation, the Master MDT, which the client send the
3976 * request to, will disassembly the operation into object updates, then OSP
3977 * will send these updates to the remote MDT to be executed.
3979 * An UPDATE_OBJ RPC does a list of updates. Each update belongs to an
3980 * operation and does a type of modification to an object.
3988 * update (ub_count-th)
3990 * ub_count must be less than or equal to UPDATE_PER_RPC_MAX.
3995 * rc [+ buffers] (1st)
3996 * rc [+ buffers] (2st)
3998 * rc [+ buffers] (nr_count-th)
4000 * ur_count must be less than or equal to UPDATE_PER_RPC_MAX and should usually
4001 * be equal to ub_count.
4005 * Type of each update
4016 OUT_INDEX_LOOKUP = 9,
4017 OUT_INDEX_INSERT = 10,
4018 OUT_INDEX_DELETE = 11,
4025 UPDATE_FL_OST = 0x00000001, /* op from OST (not MDT) */
4026 UPDATE_FL_SYNC = 0x00000002, /* commit before replying */
4027 UPDATE_FL_COMMITTED = 0x00000004, /* op committed globally */
4028 UPDATE_FL_NOLOG = 0x00000008 /* for idempotent updates */
4031 struct object_update_param {
4032 __u16 oup_len; /* length of this parameter */
4038 static inline size_t
4039 object_update_param_size(const struct object_update_param *param)
4041 return cfs_size_round(sizeof(*param) + param->oup_len);
4045 struct object_update {
4046 __u16 ou_type; /* enum update_type */
4047 __u16 ou_params_count; /* update parameters count */
4048 __u32 ou_master_index; /* master MDT/OST index */
4049 __u32 ou_flags; /* enum update_flag */
4050 __u32 ou_padding1; /* padding 1 */
4051 __u64 ou_batchid; /* op transno on master */
4052 struct lu_fid ou_fid; /* object to be updated */
4053 struct object_update_param ou_params[0]; /* update params */
4056 #define UPDATE_REQUEST_MAGIC_V1 0xBDDE0001
4057 #define UPDATE_REQUEST_MAGIC_V2 0xBDDE0002
4058 #define UPDATE_REQUEST_MAGIC UPDATE_REQUEST_MAGIC_V2
4059 /* Hold object_updates sending to the remote OUT in single RPC */
4060 struct object_update_request {
4062 __u16 ourq_count; /* number of ourq_updates[] */
4064 struct object_update ourq_updates[0];
4067 void lustre_swab_object_update(struct object_update *ou);
4068 void lustre_swab_object_update_request(struct object_update_request *our);
4070 static inline size_t
4071 object_update_size(const struct object_update *update)
4073 const struct object_update_param *param;
4077 size = offsetof(struct object_update, ou_params[0]);
4078 for (i = 0; i < update->ou_params_count; i++) {
4079 param = (struct object_update_param *)((char *)update + size);
4080 size += object_update_param_size(param);
4086 static inline struct object_update *
4087 object_update_request_get(const struct object_update_request *our,
4088 unsigned int index, size_t *size)
4093 if (index >= our->ourq_count)
4096 ptr = (void *)&our->ourq_updates[0];
4097 for (i = 0; i < index; i++)
4098 ptr += object_update_size(ptr);
4101 *size = object_update_size(ptr);
4107 /* the result of object update */
4108 struct object_update_result {
4115 #define UPDATE_REPLY_MAGIC_V1 0x00BD0001
4116 #define UPDATE_REPLY_MAGIC_V2 0x00BD0002
4117 #define UPDATE_REPLY_MAGIC UPDATE_REPLY_MAGIC_V2
4118 /* Hold object_update_results being replied from the remote OUT. */
4119 struct object_update_reply {
4126 void lustre_swab_object_update_result(struct object_update_result *our);
4127 void lustre_swab_object_update_reply(struct object_update_reply *our);
4129 static inline struct object_update_result *
4130 object_update_result_get(const struct object_update_reply *reply,
4131 unsigned int index, size_t *size)
4133 __u16 count = reply->ourp_count;
4140 ptr = (char *)reply +
4141 cfs_size_round(offsetof(struct object_update_reply,
4143 for (i = 0; i < index; i++) {
4144 if (reply->ourp_lens[i] == 0)
4147 ptr += cfs_size_round(reply->ourp_lens[i]);
4151 *size = reply->ourp_lens[index];
4156 /** layout swap request structure
4157 * fid1 and fid2 are in mdt_body
4159 struct mdc_swap_layouts {
4163 void lustre_swab_swap_layouts(struct mdc_swap_layouts *msl);
4166 struct lustre_handle cd_handle;
4167 struct lu_fid cd_fid;
4168 __u64 cd_data_version;
4169 __u64 cd_reserved[8];
4172 void lustre_swab_close_data(struct close_data *data);