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, 2013, 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_
95 #include <libcfs/libcfs.h> /* for LPUX64, etc */
98 /* Defn's shared with user-space. */
99 #include <lustre/lustre_user.h>
101 #include <lustre/lustre_errno.h>
106 /* FOO_REQUEST_PORTAL is for incoming requests on the FOO
107 * FOO_REPLY_PORTAL is for incoming replies on the FOO
108 * FOO_BULK_PORTAL is for incoming bulk on the FOO
111 #define CONNMGR_REQUEST_PORTAL 1
112 #define CONNMGR_REPLY_PORTAL 2
113 //#define OSC_REQUEST_PORTAL 3
114 #define OSC_REPLY_PORTAL 4
115 //#define OSC_BULK_PORTAL 5
116 #define OST_IO_PORTAL 6
117 #define OST_CREATE_PORTAL 7
118 #define OST_BULK_PORTAL 8
119 //#define MDC_REQUEST_PORTAL 9
120 #define MDC_REPLY_PORTAL 10
121 //#define MDC_BULK_PORTAL 11
122 #define MDS_REQUEST_PORTAL 12
123 //#define MDS_REPLY_PORTAL 13
124 #define MDS_BULK_PORTAL 14
125 #define LDLM_CB_REQUEST_PORTAL 15
126 #define LDLM_CB_REPLY_PORTAL 16
127 #define LDLM_CANCEL_REQUEST_PORTAL 17
128 #define LDLM_CANCEL_REPLY_PORTAL 18
129 //#define PTLBD_REQUEST_PORTAL 19
130 //#define PTLBD_REPLY_PORTAL 20
131 //#define PTLBD_BULK_PORTAL 21
132 #define MDS_SETATTR_PORTAL 22
133 #define MDS_READPAGE_PORTAL 23
134 #define OUT_PORTAL 24
135 #define MGC_REPLY_PORTAL 25
136 #define MGS_REQUEST_PORTAL 26
137 #define MGS_REPLY_PORTAL 27
138 #define OST_REQUEST_PORTAL 28
139 #define FLD_REQUEST_PORTAL 29
140 #define SEQ_METADATA_PORTAL 30
141 #define SEQ_DATA_PORTAL 31
142 #define SEQ_CONTROLLER_PORTAL 32
143 #define MGS_BULK_PORTAL 33
144 #define OST_IDX_PORTAL 34
146 /* Portal 63 is reserved for the Cray Inc DVS - nic@cray.com, roe@cray.com, n8851@cray.com */
149 #define PTL_RPC_MSG_REQUEST 4711
150 #define PTL_RPC_MSG_ERR 4712
151 #define PTL_RPC_MSG_REPLY 4713
153 /* DON'T use swabbed values of MAGIC as magic! */
154 #define LUSTRE_MSG_MAGIC_V1 0x0BD00BD0
155 #define LUSTRE_MSG_MAGIC_V2 0x0BD00BD3
157 #define LUSTRE_MSG_MAGIC_V1_SWABBED 0xD00BD00B
158 #define LUSTRE_MSG_MAGIC_V2_SWABBED 0xD30BD00B
160 #define LUSTRE_MSG_MAGIC LUSTRE_MSG_MAGIC_V2
162 #define PTLRPC_MSG_VERSION 0x00000003
163 #define LUSTRE_VERSION_MASK 0xffff0000
164 #define LUSTRE_OBD_VERSION 0x00010000
165 #define LUSTRE_MDS_VERSION 0x00020000
166 #define LUSTRE_OST_VERSION 0x00030000
167 #define LUSTRE_DLM_VERSION 0x00040000
168 #define LUSTRE_LOG_VERSION 0x00050000
169 #define LUSTRE_MGS_VERSION 0x00060000
171 typedef __u32 mdsno_t;
172 typedef __u64 seqno_t;
173 typedef __u64 obd_id;
174 typedef __u64 obd_seq;
175 typedef __s64 obd_time;
176 typedef __u64 obd_size;
177 typedef __u64 obd_off;
178 typedef __u64 obd_blocks;
179 typedef __u64 obd_valid;
180 typedef __u32 obd_blksize;
181 typedef __u32 obd_mode;
182 typedef __u32 obd_uid;
183 typedef __u32 obd_gid;
184 typedef __u32 obd_flag;
185 typedef __u32 obd_count;
188 * Describes a range of sequence, lsr_start is included but lsr_end is
190 * Same structure is used in fld module where lsr_index field holds mdt id
193 struct lu_seq_range {
200 struct lu_seq_range_array {
203 struct lu_seq_range lsra_lsr[0];
206 #define LU_SEQ_RANGE_MDT 0x0
207 #define LU_SEQ_RANGE_OST 0x1
208 #define LU_SEQ_RANGE_ANY 0x3
210 #define LU_SEQ_RANGE_MASK 0x3
212 static inline unsigned fld_range_type(const struct lu_seq_range *range)
214 return range->lsr_flags & LU_SEQ_RANGE_MASK;
217 static inline int fld_range_is_ost(const struct lu_seq_range *range)
219 return fld_range_type(range) == LU_SEQ_RANGE_OST;
222 static inline int fld_range_is_mdt(const struct lu_seq_range *range)
224 return fld_range_type(range) == LU_SEQ_RANGE_MDT;
228 * This all range is only being used when fld client sends fld query request,
229 * but it does not know whether the seq is MDT or OST, so it will send req
230 * with ALL type, which means either seq type gotten from lookup can be
233 static inline unsigned fld_range_is_any(const struct lu_seq_range *range)
235 return fld_range_type(range) == LU_SEQ_RANGE_ANY;
238 static inline void fld_range_set_type(struct lu_seq_range *range,
241 range->lsr_flags |= flags;
244 static inline void fld_range_set_mdt(struct lu_seq_range *range)
246 fld_range_set_type(range, LU_SEQ_RANGE_MDT);
249 static inline void fld_range_set_ost(struct lu_seq_range *range)
251 fld_range_set_type(range, LU_SEQ_RANGE_OST);
254 static inline void fld_range_set_any(struct lu_seq_range *range)
256 fld_range_set_type(range, LU_SEQ_RANGE_ANY);
260 * returns width of given range \a r
263 static inline __u64 range_space(const struct lu_seq_range *range)
265 return range->lsr_end - range->lsr_start;
269 * initialize range to zero
272 static inline void range_init(struct lu_seq_range *range)
274 memset(range, 0, sizeof(*range));
278 * check if given seq id \a s is within given range \a r
281 static inline int range_within(const struct lu_seq_range *range,
284 return s >= range->lsr_start && s < range->lsr_end;
287 static inline int range_is_sane(const struct lu_seq_range *range)
289 return (range->lsr_end >= range->lsr_start);
292 static inline int range_is_zero(const struct lu_seq_range *range)
294 return (range->lsr_start == 0 && range->lsr_end == 0);
297 static inline int range_is_exhausted(const struct lu_seq_range *range)
300 return range_space(range) == 0;
303 /* return 0 if two range have the same location */
304 static inline int range_compare_loc(const struct lu_seq_range *r1,
305 const struct lu_seq_range *r2)
307 return r1->lsr_index != r2->lsr_index ||
308 r1->lsr_flags != r2->lsr_flags;
311 #define DRANGE "[%#16.16"LPF64"x-%#16.16"LPF64"x):%x:%s"
313 #define PRANGE(range) \
314 (range)->lsr_start, \
316 (range)->lsr_index, \
317 fld_range_is_mdt(range) ? "mdt" : "ost"
320 /** \defgroup lu_fid lu_fid
324 * Flags for lustre_mdt_attrs::lma_compat and lustre_mdt_attrs::lma_incompat.
325 * Deprecated since HSM and SOM attributes are now stored in separate on-disk
329 LMAC_HSM = 0x00000001,
330 LMAC_SOM = 0x00000002,
331 LMAC_NOT_IN_OI = 0x00000004, /* the object does NOT need OI mapping */
332 LMAC_FID_ON_OST = 0x00000008, /* For OST-object, its OI mapping is
333 * under /O/<seq>/d<x>. */
337 * Masks for all features that should be supported by a Lustre version to
338 * access a specific file.
339 * This information is stored in lustre_mdt_attrs::lma_incompat.
342 LMAI_RELEASED = 0x00000001, /* file is released */
343 LMAI_AGENT = 0x00000002, /* agent inode */
344 LMAI_REMOTE_PARENT = 0x00000004, /* the parent of the object
345 is on the remote MDT */
347 #define LMA_INCOMPAT_SUPP (LMAI_AGENT | LMAI_REMOTE_PARENT)
349 extern void lustre_lma_swab(struct lustre_mdt_attrs *lma);
350 extern void lustre_lma_init(struct lustre_mdt_attrs *lma,
351 const struct lu_fid *fid,
352 __u32 compat, __u32 incompat);
354 * SOM on-disk attributes stored in a separate xattr.
357 /** Bitfield for supported data in this structure. For future use. */
360 /** Incompat feature list. The supported feature mask is availabe in
361 * SOM_INCOMPAT_SUPP */
364 /** IO Epoch SOM attributes belongs to */
366 /** total file size in objects */
368 /** total fs blocks in objects */
370 /** mds mount id the size is valid for */
373 extern void lustre_som_swab(struct som_attrs *attrs);
375 #define SOM_INCOMPAT_SUPP 0x0
378 * HSM on-disk attributes stored in a separate xattr.
381 /** Bitfield for supported data in this structure. For future use. */
384 /** HSM flags, see hsm_flags enum below */
386 /** backend archive id associated with the file */
388 /** version associated with the last archiving, if any */
391 extern void lustre_hsm_swab(struct hsm_attrs *attrs);
397 /** LASTID file has zero OID */
398 LUSTRE_FID_LASTID_OID = 0UL,
399 /** initial fid id value */
400 LUSTRE_FID_INIT_OID = 1UL
403 /** returns fid object sequence */
404 static inline __u64 fid_seq(const struct lu_fid *fid)
409 /** returns fid object id */
410 static inline __u32 fid_oid(const struct lu_fid *fid)
415 /** returns fid object version */
416 static inline __u32 fid_ver(const struct lu_fid *fid)
421 static inline void fid_zero(struct lu_fid *fid)
423 memset(fid, 0, sizeof(*fid));
426 static inline obd_id fid_ver_oid(const struct lu_fid *fid)
428 return ((__u64)fid_ver(fid) << 32 | fid_oid(fid));
432 * Note that reserved SEQ numbers below 12 will conflict with ldiskfs
433 * inodes in the IGIF namespace, so these reserved SEQ numbers can be
434 * used for other purposes and not risk collisions with existing inodes.
436 * Different FID Format
437 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs#NEW.0
440 FID_SEQ_OST_MDT0 = 0,
441 FID_SEQ_LLOG = 1, /* unnamed llogs */
443 FID_SEQ_OST_MDT1 = 3,
444 FID_SEQ_OST_MAX = 9, /* Max MDT count before OST_on_FID */
445 FID_SEQ_LLOG_NAME = 10, /* named llogs */
448 FID_SEQ_IGIF_MAX = 0x0ffffffffULL,
449 FID_SEQ_IDIF = 0x100000000ULL,
450 FID_SEQ_IDIF_MAX = 0x1ffffffffULL,
451 /* Normal FID sequence starts from this value, i.e. 1<<33 */
452 FID_SEQ_START = 0x200000000ULL,
453 /* sequence for local pre-defined FIDs listed in local_oid */
454 FID_SEQ_LOCAL_FILE = 0x200000001ULL,
455 FID_SEQ_DOT_LUSTRE = 0x200000002ULL,
456 /* sequence is used for local named objects FIDs generated
457 * by local_object_storage library */
458 FID_SEQ_LOCAL_NAME = 0x200000003ULL,
459 /* Because current FLD will only cache the fid sequence, instead
460 * of oid on the client side, if the FID needs to be exposed to
461 * clients sides, it needs to make sure all of fids under one
462 * sequence will be located in one MDT. */
463 FID_SEQ_SPECIAL = 0x200000004ULL,
464 FID_SEQ_QUOTA = 0x200000005ULL,
465 FID_SEQ_QUOTA_GLB = 0x200000006ULL,
466 FID_SEQ_ROOT = 0x200000007ULL, /* Located on MDT0 */
467 FID_SEQ_LAYOUT_RBTREE = 0x200000008ULL,
468 FID_SEQ_NORMAL = 0x200000400ULL,
469 FID_SEQ_LOV_DEFAULT = 0xffffffffffffffffULL
472 #define OBIF_OID_MAX_BITS 32
473 #define OBIF_MAX_OID (1ULL << OBIF_OID_MAX_BITS)
474 #define OBIF_OID_MASK ((1ULL << OBIF_OID_MAX_BITS) - 1)
475 #define IDIF_OID_MAX_BITS 48
476 #define IDIF_MAX_OID (1ULL << IDIF_OID_MAX_BITS)
477 #define IDIF_OID_MASK ((1ULL << IDIF_OID_MAX_BITS) - 1)
479 /** OID for FID_SEQ_SPECIAL */
481 /* Big Filesystem Lock to serialize rename operations */
482 FID_OID_SPECIAL_BFL = 1UL,
485 /** OID for FID_SEQ_DOT_LUSTRE */
486 enum dot_lustre_oid {
487 FID_OID_DOT_LUSTRE = 1UL,
488 FID_OID_DOT_LUSTRE_OBF = 2UL,
491 static inline int fid_seq_is_mdt0(obd_seq seq)
493 return (seq == FID_SEQ_OST_MDT0);
496 static inline int fid_seq_is_mdt(const __u64 seq)
498 return seq == FID_SEQ_OST_MDT0 || seq >= FID_SEQ_NORMAL;
501 static inline int fid_seq_is_echo(obd_seq seq)
503 return (seq == FID_SEQ_ECHO);
506 static inline int fid_is_echo(const struct lu_fid *fid)
508 return fid_seq_is_echo(fid_seq(fid));
511 static inline int fid_seq_is_llog(obd_seq seq)
513 return (seq == FID_SEQ_LLOG);
516 static inline int fid_is_llog(const struct lu_fid *fid)
518 /* file with OID == 0 is not llog but contains last oid */
519 return fid_seq_is_llog(fid_seq(fid)) && fid_oid(fid) > 0;
522 static inline int fid_seq_is_rsvd(const __u64 seq)
524 return (seq > FID_SEQ_OST_MDT0 && seq <= FID_SEQ_RSVD);
527 static inline int fid_seq_is_special(const __u64 seq)
529 return seq == FID_SEQ_SPECIAL;
532 static inline int fid_seq_is_local_file(const __u64 seq)
534 return seq == FID_SEQ_LOCAL_FILE ||
535 seq == FID_SEQ_LOCAL_NAME;
538 static inline int fid_seq_is_root(const __u64 seq)
540 return seq == FID_SEQ_ROOT;
543 static inline int fid_seq_is_dot(const __u64 seq)
545 return seq == FID_SEQ_DOT_LUSTRE;
548 static inline int fid_seq_is_default(const __u64 seq)
550 return seq == FID_SEQ_LOV_DEFAULT;
553 static inline int fid_is_mdt0(const struct lu_fid *fid)
555 return fid_seq_is_mdt0(fid_seq(fid));
558 static inline void lu_root_fid(struct lu_fid *fid)
560 fid->f_seq = FID_SEQ_ROOT;
566 * Check if a fid is igif or not.
567 * \param fid the fid to be tested.
568 * \return true if the fid is a igif; otherwise false.
570 static inline int fid_seq_is_igif(const __u64 seq)
572 return seq >= FID_SEQ_IGIF && seq <= FID_SEQ_IGIF_MAX;
575 static inline int fid_is_igif(const struct lu_fid *fid)
577 return fid_seq_is_igif(fid_seq(fid));
581 * Check if a fid is idif or not.
582 * \param fid the fid to be tested.
583 * \return true if the fid is a idif; otherwise false.
585 static inline int fid_seq_is_idif(const __u64 seq)
587 return seq >= FID_SEQ_IDIF && seq <= FID_SEQ_IDIF_MAX;
590 static inline int fid_is_idif(const struct lu_fid *fid)
592 return fid_seq_is_idif(fid_seq(fid));
595 static inline int fid_is_local_file(const struct lu_fid *fid)
597 return fid_seq_is_local_file(fid_seq(fid));
600 static inline int fid_seq_is_norm(const __u64 seq)
602 return (seq >= FID_SEQ_NORMAL);
605 static inline int fid_is_norm(const struct lu_fid *fid)
607 return fid_seq_is_norm(fid_seq(fid));
610 static inline int fid_is_layout_rbtree(const struct lu_fid *fid)
612 return fid_seq(fid) == FID_SEQ_LAYOUT_RBTREE;
615 /* convert an OST objid into an IDIF FID SEQ number */
616 static inline obd_seq fid_idif_seq(obd_id id, __u32 ost_idx)
618 return FID_SEQ_IDIF | (ost_idx << 16) | ((id >> 32) & 0xffff);
621 /* convert a packed IDIF FID into an OST objid */
622 static inline obd_id fid_idif_id(obd_seq seq, __u32 oid, __u32 ver)
624 return ((__u64)ver << 48) | ((seq & 0xffff) << 32) | oid;
627 static inline __u32 idif_ost_idx(obd_seq seq)
629 return (seq >> 16) & 0xffff;
632 /* extract ost index from IDIF FID */
633 static inline __u32 fid_idif_ost_idx(const struct lu_fid *fid)
635 return idif_ost_idx(fid_seq(fid));
638 /* extract OST sequence (group) from a wire ost_id (id/seq) pair */
639 static inline obd_seq ostid_seq(const struct ost_id *ostid)
641 if (fid_seq_is_mdt0(ostid->oi.oi_seq))
642 return FID_SEQ_OST_MDT0;
644 if (unlikely(fid_seq_is_default(ostid->oi.oi_seq)))
645 return FID_SEQ_LOV_DEFAULT;
647 if (fid_is_idif(&ostid->oi_fid))
648 return FID_SEQ_OST_MDT0;
650 return fid_seq(&ostid->oi_fid);
653 /* extract OST objid from a wire ost_id (id/seq) pair */
654 static inline obd_id ostid_id(const struct ost_id *ostid)
656 if (fid_seq_is_mdt0(ostid->oi.oi_seq))
657 return ostid->oi.oi_id & IDIF_OID_MASK;
659 if (unlikely(fid_seq_is_default(ostid->oi.oi_seq)))
660 return ostid->oi.oi_id;
662 if (fid_is_idif(&ostid->oi_fid))
663 return fid_idif_id(fid_seq(&ostid->oi_fid),
664 fid_oid(&ostid->oi_fid), 0);
666 return fid_oid(&ostid->oi_fid);
669 static inline void ostid_set_seq(struct ost_id *oi, __u64 seq)
671 if (fid_seq_is_mdt0(seq) || fid_seq_is_default(seq)) {
674 oi->oi_fid.f_seq = seq;
675 /* Note: if f_oid + f_ver is zero, we need init it
676 * to be 1, otherwise, ostid_seq will treat this
677 * as old ostid (oi_seq == 0) */
678 if (oi->oi_fid.f_oid == 0 && oi->oi_fid.f_ver == 0)
679 oi->oi_fid.f_oid = LUSTRE_FID_INIT_OID;
683 static inline void ostid_set_seq_mdt0(struct ost_id *oi)
685 ostid_set_seq(oi, FID_SEQ_OST_MDT0);
688 static inline void ostid_set_seq_echo(struct ost_id *oi)
690 ostid_set_seq(oi, FID_SEQ_ECHO);
693 static inline void ostid_set_seq_llog(struct ost_id *oi)
695 ostid_set_seq(oi, FID_SEQ_LLOG);
699 * Note: we need check oi_seq to decide where to set oi_id,
700 * so oi_seq should always be set ahead of oi_id.
702 static inline void ostid_set_id(struct ost_id *oi, __u64 oid)
704 if (fid_seq_is_mdt0(oi->oi.oi_seq)) {
705 if (oid >= IDIF_MAX_OID) {
706 CERROR("Bad "LPU64" to set "DOSTID"\n",
711 } else if (fid_is_idif(&oi->oi_fid)) {
712 if (oid >= IDIF_MAX_OID) {
713 CERROR("Bad "LPU64" to set "DOSTID"\n",
717 oi->oi_fid.f_seq = fid_idif_seq(oid,
718 fid_idif_ost_idx(&oi->oi_fid));
719 oi->oi_fid.f_oid = oid;
720 oi->oi_fid.f_ver = oid >> 48;
722 if (oid > OBIF_MAX_OID) {
723 CERROR("Bad "LPU64" to set "DOSTID"\n",
727 oi->oi_fid.f_oid = oid;
731 static inline int fid_set_id(struct lu_fid *fid, __u64 oid)
733 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
734 CERROR("bad IGIF, "DFID"\n", PFID(fid));
738 if (fid_is_idif(fid)) {
739 if (oid >= IDIF_MAX_OID) {
740 CERROR("Bad "LPU64" to set "DFID"\n",
744 fid->f_seq = fid_idif_seq(oid, fid_idif_ost_idx(fid));
746 fid->f_ver = oid >> 48;
748 if (oid > OBIF_MAX_OID) {
749 CERROR("Bad "LPU64" to set "DFID"\n",
759 * Unpack an OST object id/seq (group) into a FID. This is needed for
760 * converting all obdo, lmm, lsm, etc. 64-bit id/seq pairs into proper
761 * FIDs. Note that if an id/seq is already in FID/IDIF format it will
762 * be passed through unchanged. Only legacy OST objects in "group 0"
763 * will be mapped into the IDIF namespace so that they can fit into the
764 * struct lu_fid fields without loss. For reference see:
765 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs
767 static inline int ostid_to_fid(struct lu_fid *fid, const struct ost_id *ostid,
770 obd_seq seq = ostid_seq(ostid);
772 if (ost_idx > 0xffff) {
773 CERROR("bad ost_idx, "DOSTID" ost_idx:%u\n", POSTID(ostid),
778 if (fid_seq_is_mdt0(seq)) {
779 obd_id oid = ostid_id(ostid);
781 /* This is a "legacy" (old 1.x/2.early) OST object in "group 0"
782 * that we map into the IDIF namespace. It allows up to 2^48
783 * objects per OST, as this is the object namespace that has
784 * been in production for years. This can handle create rates
785 * of 1M objects/s/OST for 9 years, or combinations thereof. */
786 if (oid >= IDIF_MAX_OID) {
787 CERROR("bad MDT0 id, "DOSTID" ost_idx:%u\n",
788 POSTID(ostid), ost_idx);
791 fid->f_seq = fid_idif_seq(oid, ost_idx);
792 /* truncate to 32 bits by assignment */
794 /* in theory, not currently used */
795 fid->f_ver = oid >> 48;
796 } else if (likely(!fid_seq_is_default(seq)))
797 /* if (fid_seq_is_idif(seq) || fid_seq_is_norm(seq)) */ {
798 /* This is either an IDIF object, which identifies objects across
799 * all OSTs, or a regular FID. The IDIF namespace maps legacy
800 * OST objects into the FID namespace. In both cases, we just
801 * pass the FID through, no conversion needed. */
802 if (ostid->oi_fid.f_ver != 0) {
803 CERROR("bad MDT0 id, "DOSTID" ost_idx:%u\n",
804 POSTID(ostid), ost_idx);
807 *fid = ostid->oi_fid;
813 /* pack any OST FID into an ostid (id/seq) for the wire/disk */
814 static inline int fid_to_ostid(const struct lu_fid *fid, struct ost_id *ostid)
816 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
817 CERROR("bad IGIF, "DFID"\n", PFID(fid));
821 if (fid_is_idif(fid)) {
822 ostid_set_seq_mdt0(ostid);
823 ostid_set_id(ostid, fid_idif_id(fid_seq(fid), fid_oid(fid),
826 ostid->oi_fid = *fid;
832 /* Check whether the fid is for LAST_ID */
833 static inline int fid_is_last_id(const struct lu_fid *fid)
835 return (fid_oid(fid) == 0);
839 * Get inode number from a igif.
840 * \param fid a igif to get inode number from.
841 * \return inode number for the igif.
843 static inline ino_t lu_igif_ino(const struct lu_fid *fid)
848 extern void lustre_swab_ost_id(struct ost_id *oid);
851 * Get inode generation from a igif.
852 * \param fid a igif to get inode generation from.
853 * \return inode generation for the igif.
855 static inline __u32 lu_igif_gen(const struct lu_fid *fid)
861 * Build igif from the inode number/generation.
863 static inline void lu_igif_build(struct lu_fid *fid, __u32 ino, __u32 gen)
871 * Fids are transmitted across network (in the sender byte-ordering),
872 * and stored on disk in big-endian order.
874 static inline void fid_cpu_to_le(struct lu_fid *dst, const struct lu_fid *src)
876 dst->f_seq = cpu_to_le64(fid_seq(src));
877 dst->f_oid = cpu_to_le32(fid_oid(src));
878 dst->f_ver = cpu_to_le32(fid_ver(src));
881 static inline void fid_le_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
883 dst->f_seq = le64_to_cpu(fid_seq(src));
884 dst->f_oid = le32_to_cpu(fid_oid(src));
885 dst->f_ver = le32_to_cpu(fid_ver(src));
888 static inline void fid_cpu_to_be(struct lu_fid *dst, const struct lu_fid *src)
890 dst->f_seq = cpu_to_be64(fid_seq(src));
891 dst->f_oid = cpu_to_be32(fid_oid(src));
892 dst->f_ver = cpu_to_be32(fid_ver(src));
895 static inline void fid_be_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
897 dst->f_seq = be64_to_cpu(fid_seq(src));
898 dst->f_oid = be32_to_cpu(fid_oid(src));
899 dst->f_ver = be32_to_cpu(fid_ver(src));
902 static inline int fid_is_sane(const struct lu_fid *fid)
904 return fid != NULL &&
905 ((fid_seq(fid) >= FID_SEQ_START && fid_ver(fid) == 0) ||
906 fid_is_igif(fid) || fid_is_idif(fid) ||
907 fid_seq_is_rsvd(fid_seq(fid)));
910 static inline int fid_is_zero(const struct lu_fid *fid)
912 return fid_seq(fid) == 0 && fid_oid(fid) == 0;
915 extern void lustre_swab_lu_fid(struct lu_fid *fid);
916 extern void lustre_swab_lu_seq_range(struct lu_seq_range *range);
918 static inline int lu_fid_eq(const struct lu_fid *f0, const struct lu_fid *f1)
920 return memcmp(f0, f1, sizeof *f0) == 0;
923 #define __diff_normalize(val0, val1) \
925 typeof(val0) __val0 = (val0); \
926 typeof(val1) __val1 = (val1); \
928 (__val0 == __val1 ? 0 : __val0 > __val1 ? +1 : -1); \
931 static inline int lu_fid_cmp(const struct lu_fid *f0,
932 const struct lu_fid *f1)
935 __diff_normalize(fid_seq(f0), fid_seq(f1)) ?:
936 __diff_normalize(fid_oid(f0), fid_oid(f1)) ?:
937 __diff_normalize(fid_ver(f0), fid_ver(f1));
940 static inline void ostid_cpu_to_le(const struct ost_id *src_oi,
941 struct ost_id *dst_oi)
943 if (fid_seq_is_mdt0(ostid_seq(src_oi))) {
944 dst_oi->oi.oi_id = cpu_to_le64(src_oi->oi.oi_id);
945 dst_oi->oi.oi_seq = cpu_to_le64(src_oi->oi.oi_seq);
947 fid_cpu_to_le(&dst_oi->oi_fid, &src_oi->oi_fid);
951 static inline void ostid_le_to_cpu(const struct ost_id *src_oi,
952 struct ost_id *dst_oi)
954 if (fid_seq_is_mdt0(ostid_seq(src_oi))) {
955 dst_oi->oi.oi_id = le64_to_cpu(src_oi->oi.oi_id);
956 dst_oi->oi.oi_seq = le64_to_cpu(src_oi->oi.oi_seq);
958 fid_le_to_cpu(&dst_oi->oi_fid, &src_oi->oi_fid);
962 struct lu_orphan_rec {
963 /* The MDT-object's FID referenced by the orphan OST-object */
964 struct lu_fid lor_fid;
969 struct lu_orphan_ent {
970 /* The orphan OST-object's FID */
971 struct lu_fid loe_key;
972 struct lu_orphan_rec loe_rec;
974 void lustre_swab_orphan_ent(struct lu_orphan_ent *ent);
978 /** \defgroup lu_dir lu_dir
982 * Enumeration of possible directory entry attributes.
984 * Attributes follow directory entry header in the order they appear in this
987 enum lu_dirent_attrs {
990 LUDA_64BITHASH = 0x0004,
992 /* The following attrs are used for MDT interanl only,
993 * not visible to client */
995 /* Verify the dirent consistency */
996 LUDA_VERIFY = 0x8000,
997 /* Only check but not repair the dirent inconsistency */
998 LUDA_VERIFY_DRYRUN = 0x4000,
999 /* The dirent has been repaired, or to be repaired (dryrun). */
1000 LUDA_REPAIR = 0x2000,
1001 /* The system is upgraded, has beed or to be repaired (dryrun). */
1002 LUDA_UPGRADE = 0x1000,
1003 /* Ignore this record, go to next directly. */
1004 LUDA_IGNORE = 0x0800,
1007 #define LU_DIRENT_ATTRS_MASK 0xf800
1010 * Layout of readdir pages, as transmitted on wire.
1013 /** valid if LUDA_FID is set. */
1014 struct lu_fid lde_fid;
1015 /** a unique entry identifier: a hash or an offset. */
1017 /** total record length, including all attributes. */
1021 /** optional variable size attributes following this entry.
1022 * taken from enum lu_dirent_attrs.
1025 /** name is followed by the attributes indicated in ->ldp_attrs, in
1026 * their natural order. After the last attribute, padding bytes are
1027 * added to make ->lde_reclen a multiple of 8.
1033 * Definitions of optional directory entry attributes formats.
1035 * Individual attributes do not have their length encoded in a generic way. It
1036 * is assumed that consumer of an attribute knows its format. This means that
1037 * it is impossible to skip over an unknown attribute, except by skipping over all
1038 * remaining attributes (by using ->lde_reclen), which is not too
1039 * constraining, because new server versions will append new attributes at
1040 * the end of an entry.
1044 * Fid directory attribute: a fid of an object referenced by the entry. This
1045 * will be almost always requested by the client and supplied by the server.
1047 * Aligned to 8 bytes.
1049 /* To have compatibility with 1.8, lets have fid in lu_dirent struct. */
1054 * Aligned to 2 bytes.
1061 __u64 ldp_hash_start;
1065 struct lu_dirent ldp_entries[0];
1068 enum lu_dirpage_flags {
1070 * dirpage contains no entry.
1074 * last entry's lde_hash equals ldp_hash_end.
1076 LDF_COLLIDE = 1 << 1
1079 static inline struct lu_dirent *lu_dirent_start(struct lu_dirpage *dp)
1081 if (le32_to_cpu(dp->ldp_flags) & LDF_EMPTY)
1084 return dp->ldp_entries;
1087 static inline struct lu_dirent *lu_dirent_next(struct lu_dirent *ent)
1089 struct lu_dirent *next;
1091 if (le16_to_cpu(ent->lde_reclen) != 0)
1092 next = ((void *)ent) + le16_to_cpu(ent->lde_reclen);
1099 static inline int lu_dirent_calc_size(int namelen, __u16 attr)
1103 if (attr & LUDA_TYPE) {
1104 const unsigned align = sizeof(struct luda_type) - 1;
1105 size = (sizeof(struct lu_dirent) + namelen + align) & ~align;
1106 size += sizeof(struct luda_type);
1108 size = sizeof(struct lu_dirent) + namelen;
1110 return (size + 7) & ~7;
1113 static inline int lu_dirent_size(struct lu_dirent *ent)
1115 if (le16_to_cpu(ent->lde_reclen) == 0) {
1116 return lu_dirent_calc_size(le16_to_cpu(ent->lde_namelen),
1117 le32_to_cpu(ent->lde_attrs));
1119 return le16_to_cpu(ent->lde_reclen);
1122 #define MDS_DIR_END_OFF 0xfffffffffffffffeULL
1125 * MDS_READPAGE page size
1127 * This is the directory page size packed in MDS_READPAGE RPC.
1128 * It's different than PAGE_CACHE_SIZE because the client needs to
1129 * access the struct lu_dirpage header packed at the beginning of
1130 * the "page" and without this there isn't any way to know find the
1131 * lu_dirpage header is if client and server PAGE_CACHE_SIZE differ.
1133 #define LU_PAGE_SHIFT 12
1134 #define LU_PAGE_SIZE (1UL << LU_PAGE_SHIFT)
1135 #define LU_PAGE_MASK (~(LU_PAGE_SIZE - 1))
1137 #define LU_PAGE_COUNT (1 << (PAGE_CACHE_SHIFT - LU_PAGE_SHIFT))
1141 struct lustre_handle {
1144 #define DEAD_HANDLE_MAGIC 0xdeadbeefcafebabeULL
1146 static inline int lustre_handle_is_used(struct lustre_handle *lh)
1148 return lh->cookie != 0ull;
1151 static inline int lustre_handle_equal(const struct lustre_handle *lh1,
1152 const struct lustre_handle *lh2)
1154 return lh1->cookie == lh2->cookie;
1157 static inline void lustre_handle_copy(struct lustre_handle *tgt,
1158 struct lustre_handle *src)
1160 tgt->cookie = src->cookie;
1163 /* flags for lm_flags */
1164 #define MSGHDR_AT_SUPPORT 0x1
1165 #define MSGHDR_CKSUM_INCOMPAT18 0x2
1167 #define lustre_msg lustre_msg_v2
1168 /* we depend on this structure to be 8-byte aligned */
1169 /* this type is only endian-adjusted in lustre_unpack_msg() */
1170 struct lustre_msg_v2 {
1179 __u32 lm_buflens[0];
1182 /* without gss, ptlrpc_body is put at the first buffer. */
1183 #define PTLRPC_NUM_VERSIONS 4
1184 #define JOBSTATS_JOBID_SIZE 32 /* 32 bytes string */
1185 struct ptlrpc_body_v3 {
1186 struct lustre_handle pb_handle;
1193 __u64 pb_last_committed;
1198 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1199 __u32 pb_service_time; /* for rep, actual service time */
1202 /* VBR: pre-versions */
1203 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1204 /* padding for future needs */
1205 __u64 pb_padding[4];
1206 char pb_jobid[JOBSTATS_JOBID_SIZE];
1208 #define ptlrpc_body ptlrpc_body_v3
1210 struct ptlrpc_body_v2 {
1211 struct lustre_handle pb_handle;
1218 __u64 pb_last_committed;
1223 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1224 __u32 pb_service_time; /* for rep, actual service time, also used for
1225 net_latency of req */
1228 /* VBR: pre-versions */
1229 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1230 /* padding for future needs */
1231 __u64 pb_padding[4];
1234 extern void lustre_swab_ptlrpc_body(struct ptlrpc_body *pb);
1236 /* message body offset for lustre_msg_v2 */
1237 /* ptlrpc body offset in all request/reply messages */
1238 #define MSG_PTLRPC_BODY_OFF 0
1240 /* normal request/reply message record offset */
1241 #define REQ_REC_OFF 1
1242 #define REPLY_REC_OFF 1
1244 /* ldlm request message body offset */
1245 #define DLM_LOCKREQ_OFF 1 /* lockreq offset */
1246 #define DLM_REQ_REC_OFF 2 /* normal dlm request record offset */
1248 /* ldlm intent lock message body offset */
1249 #define DLM_INTENT_IT_OFF 2 /* intent lock it offset */
1250 #define DLM_INTENT_REC_OFF 3 /* intent lock record offset */
1252 /* ldlm reply message body offset */
1253 #define DLM_LOCKREPLY_OFF 1 /* lockrep offset */
1254 #define DLM_REPLY_REC_OFF 2 /* reply record offset */
1256 /** only use in req->rq_{req,rep}_swab_mask */
1257 #define MSG_PTLRPC_HEADER_OFF 31
1259 /* Flags that are operation-specific go in the top 16 bits. */
1260 #define MSG_OP_FLAG_MASK 0xffff0000
1261 #define MSG_OP_FLAG_SHIFT 16
1263 /* Flags that apply to all requests are in the bottom 16 bits */
1264 #define MSG_GEN_FLAG_MASK 0x0000ffff
1265 #define MSG_LAST_REPLAY 0x0001
1266 #define MSG_RESENT 0x0002
1267 #define MSG_REPLAY 0x0004
1268 /* #define MSG_AT_SUPPORT 0x0008
1269 * This was used in early prototypes of adaptive timeouts, and while there
1270 * shouldn't be any users of that code there also isn't a need for using this
1271 * bits. Defer usage until at least 1.10 to avoid potential conflict. */
1272 #define MSG_DELAY_REPLAY 0x0010
1273 #define MSG_VERSION_REPLAY 0x0020
1274 #define MSG_REQ_REPLAY_DONE 0x0040
1275 #define MSG_LOCK_REPLAY_DONE 0x0080
1278 * Flags for all connect opcodes (MDS_CONNECT, OST_CONNECT)
1281 #define MSG_CONNECT_RECOVERING 0x00000001
1282 #define MSG_CONNECT_RECONNECT 0x00000002
1283 #define MSG_CONNECT_REPLAYABLE 0x00000004
1284 //#define MSG_CONNECT_PEER 0x8
1285 #define MSG_CONNECT_LIBCLIENT 0x00000010
1286 #define MSG_CONNECT_INITIAL 0x00000020
1287 #define MSG_CONNECT_ASYNC 0x00000040
1288 #define MSG_CONNECT_NEXT_VER 0x00000080 /* use next version of lustre_msg */
1289 #define MSG_CONNECT_TRANSNO 0x00000100 /* report transno */
1292 #define OBD_CONNECT_RDONLY 0x1ULL /*client has read-only access*/
1293 #define OBD_CONNECT_INDEX 0x2ULL /*connect specific LOV idx */
1294 #define OBD_CONNECT_MDS 0x4ULL /*connect from MDT to OST */
1295 #define OBD_CONNECT_GRANT 0x8ULL /*OSC gets grant at connect */
1296 #define OBD_CONNECT_SRVLOCK 0x10ULL /*server takes locks for cli */
1297 #define OBD_CONNECT_VERSION 0x20ULL /*Lustre versions in ocd */
1298 #define OBD_CONNECT_REQPORTAL 0x40ULL /*Separate non-IO req portal */
1299 #define OBD_CONNECT_ACL 0x80ULL /*access control lists */
1300 #define OBD_CONNECT_XATTR 0x100ULL /*client use extended attr */
1301 #define OBD_CONNECT_CROW 0x200ULL /*MDS+OST create obj on write*/
1302 #define OBD_CONNECT_TRUNCLOCK 0x400ULL /*locks on server for punch */
1303 #define OBD_CONNECT_TRANSNO 0x800ULL /*replay sends init transno */
1304 #define OBD_CONNECT_IBITS 0x1000ULL /*support for inodebits locks*/
1305 #define OBD_CONNECT_JOIN 0x2000ULL /*files can be concatenated.
1306 *We do not support JOIN FILE
1307 *anymore, reserve this flags
1308 *just for preventing such bit
1310 #define OBD_CONNECT_ATTRFID 0x4000ULL /*Server can GetAttr By Fid*/
1311 #define OBD_CONNECT_NODEVOH 0x8000ULL /*No open hndl on specl nodes*/
1312 #define OBD_CONNECT_RMT_CLIENT 0x10000ULL /*Remote client */
1313 #define OBD_CONNECT_RMT_CLIENT_FORCE 0x20000ULL /*Remote client by force */
1314 #define OBD_CONNECT_BRW_SIZE 0x40000ULL /*Max bytes per rpc */
1315 #define OBD_CONNECT_QUOTA64 0x80000ULL /*Not used since 2.4 */
1316 #define OBD_CONNECT_MDS_CAPA 0x100000ULL /*MDS capability */
1317 #define OBD_CONNECT_OSS_CAPA 0x200000ULL /*OSS capability */
1318 #define OBD_CONNECT_CANCELSET 0x400000ULL /*Early batched cancels. */
1319 #define OBD_CONNECT_SOM 0x800000ULL /*Size on MDS */
1320 #define OBD_CONNECT_AT 0x1000000ULL /*client uses AT */
1321 #define OBD_CONNECT_LRU_RESIZE 0x2000000ULL /*LRU resize feature. */
1322 #define OBD_CONNECT_MDS_MDS 0x4000000ULL /*MDS-MDS connection */
1323 #define OBD_CONNECT_REAL 0x8000000ULL /*real connection */
1324 #define OBD_CONNECT_CHANGE_QS 0x10000000ULL /*Not used since 2.4 */
1325 #define OBD_CONNECT_CKSUM 0x20000000ULL /*support several cksum algos*/
1326 #define OBD_CONNECT_FID 0x40000000ULL /*FID is supported by server */
1327 #define OBD_CONNECT_VBR 0x80000000ULL /*version based recovery */
1328 #define OBD_CONNECT_LOV_V3 0x100000000ULL /*client supports LOV v3 EA */
1329 #define OBD_CONNECT_GRANT_SHRINK 0x200000000ULL /* support grant shrink */
1330 #define OBD_CONNECT_SKIP_ORPHAN 0x400000000ULL /* don't reuse orphan objids */
1331 #define OBD_CONNECT_MAX_EASIZE 0x800000000ULL /* preserved for large EA */
1332 #define OBD_CONNECT_FULL20 0x1000000000ULL /* it is 2.0 client */
1333 #define OBD_CONNECT_LAYOUTLOCK 0x2000000000ULL /* client uses layout lock */
1334 #define OBD_CONNECT_64BITHASH 0x4000000000ULL /* client supports 64-bits
1336 #define OBD_CONNECT_MAXBYTES 0x8000000000ULL /* max stripe size */
1337 #define OBD_CONNECT_IMP_RECOV 0x10000000000ULL /* imp recovery support */
1338 #define OBD_CONNECT_JOBSTATS 0x20000000000ULL /* jobid in ptlrpc_body */
1339 #define OBD_CONNECT_UMASK 0x40000000000ULL /* create uses client umask */
1340 #define OBD_CONNECT_EINPROGRESS 0x80000000000ULL /* client handles -EINPROGRESS
1341 * RPC error properly */
1342 #define OBD_CONNECT_GRANT_PARAM 0x100000000000ULL/* extra grant params used for
1343 * finer space reservation */
1344 #define OBD_CONNECT_FLOCK_OWNER 0x200000000000ULL /* for the fixed 1.8
1345 * policy and 2.x server */
1346 #define OBD_CONNECT_LVB_TYPE 0x400000000000ULL /* variable type of LVB */
1347 #define OBD_CONNECT_NANOSEC_TIME 0x800000000000ULL /* nanosecond timestamps */
1348 #define OBD_CONNECT_LIGHTWEIGHT 0x1000000000000ULL/* lightweight connection */
1349 #define OBD_CONNECT_SHORTIO 0x2000000000000ULL/* short io */
1350 #define OBD_CONNECT_PINGLESS 0x4000000000000ULL/* pings not required */
1351 #define OBD_CONNECT_FLOCK_DEAD 0x8000000000000ULL/* improved flock deadlock detection */
1352 #define OBD_CONNECT_DISP_STRIPE 0x10000000000000ULL/* create stripe disposition*/
1353 #define OBD_CONNECT_OPEN_BY_FID 0x20000000000000ULL /* open by fid won't pack
1355 #define OBD_CONNECT_LFSCK 0x40000000000000ULL/* support online LFSCK */
1358 * Please DO NOT add flag values here before first ensuring that this same
1359 * flag value is not in use on some other branch. Please clear any such
1360 * changes with senior engineers before starting to use a new flag. Then,
1361 * submit a small patch against EVERY branch that ONLY adds the new flag,
1362 * updates obd_connect_names[] for lprocfs_rd_connect_flags(), adds the
1363 * flag to check_obd_connect_data(), and updates wiretests accordingly, so it
1364 * can be approved and landed easily to reserve the flag for future use. */
1366 /* The MNE_SWAB flag is overloading the MDS_MDS bit only for the MGS
1367 * connection. It is a temporary bug fix for Imperative Recovery interop
1368 * between 2.2 and 2.3 x86/ppc nodes, and can be removed when interop for
1369 * 2.2 clients/servers is no longer needed. LU-1252/LU-1644. */
1370 #define OBD_CONNECT_MNE_SWAB OBD_CONNECT_MDS_MDS
1372 #define OCD_HAS_FLAG(ocd, flg) \
1373 (!!((ocd)->ocd_connect_flags & OBD_CONNECT_##flg))
1376 #ifdef HAVE_LRU_RESIZE_SUPPORT
1377 #define LRU_RESIZE_CONNECT_FLAG OBD_CONNECT_LRU_RESIZE
1379 #define LRU_RESIZE_CONNECT_FLAG 0
1382 #define MDT_CONNECT_SUPPORTED (OBD_CONNECT_RDONLY | OBD_CONNECT_VERSION | \
1383 OBD_CONNECT_ACL | OBD_CONNECT_XATTR | \
1384 OBD_CONNECT_IBITS | \
1385 OBD_CONNECT_NODEVOH | OBD_CONNECT_ATTRFID | \
1386 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1387 OBD_CONNECT_RMT_CLIENT | \
1388 OBD_CONNECT_RMT_CLIENT_FORCE | \
1389 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_MDS_CAPA | \
1390 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_MDS_MDS | \
1391 OBD_CONNECT_FID | LRU_RESIZE_CONNECT_FLAG | \
1392 OBD_CONNECT_VBR | OBD_CONNECT_LOV_V3 | \
1393 OBD_CONNECT_SOM | OBD_CONNECT_FULL20 | \
1394 OBD_CONNECT_64BITHASH | OBD_CONNECT_JOBSTATS | \
1395 OBD_CONNECT_EINPROGRESS | \
1396 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_UMASK | \
1397 OBD_CONNECT_LVB_TYPE | OBD_CONNECT_LAYOUTLOCK |\
1398 OBD_CONNECT_PINGLESS | OBD_CONNECT_MAX_EASIZE |\
1399 OBD_CONNECT_FLOCK_DEAD | \
1400 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK)
1402 #define OST_CONNECT_SUPPORTED (OBD_CONNECT_SRVLOCK | OBD_CONNECT_GRANT | \
1403 OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | \
1404 OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_INDEX | \
1405 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_OSS_CAPA | \
1406 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1407 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_CKSUM | \
1408 OBD_CONNECT_RMT_CLIENT | \
1409 OBD_CONNECT_RMT_CLIENT_FORCE | OBD_CONNECT_VBR | \
1410 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN | \
1411 OBD_CONNECT_GRANT_SHRINK | OBD_CONNECT_FULL20 | \
1412 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES | \
1413 OBD_CONNECT_MAX_EASIZE | \
1414 OBD_CONNECT_EINPROGRESS | \
1415 OBD_CONNECT_JOBSTATS | \
1416 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LVB_TYPE|\
1417 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_FID | \
1418 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK)
1419 #define ECHO_CONNECT_SUPPORTED (0)
1420 #define MGS_CONNECT_SUPPORTED (OBD_CONNECT_VERSION | OBD_CONNECT_AT | \
1421 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV | \
1422 OBD_CONNECT_MNE_SWAB | OBD_CONNECT_PINGLESS)
1424 /* Features required for this version of the client to work with server */
1425 #define CLIENT_CONNECT_MDT_REQD (OBD_CONNECT_IBITS | OBD_CONNECT_FID | \
1428 #define OBD_OCD_VERSION(major,minor,patch,fix) (((major)<<24) + ((minor)<<16) +\
1429 ((patch)<<8) + (fix))
1430 #define OBD_OCD_VERSION_MAJOR(version) ((int)((version)>>24)&255)
1431 #define OBD_OCD_VERSION_MINOR(version) ((int)((version)>>16)&255)
1432 #define OBD_OCD_VERSION_PATCH(version) ((int)((version)>>8)&255)
1433 #define OBD_OCD_VERSION_FIX(version) ((int)(version)&255)
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_v1 {
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, must be 2^n */
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; /* also fix lustre_swab_connect */
1455 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1458 struct obd_connect_data {
1459 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1460 __u32 ocd_version; /* lustre release version number */
1461 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1462 __u32 ocd_index; /* LOV index to connect to */
1463 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
1464 __u64 ocd_ibits_known; /* inode bits this client understands */
1465 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1466 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1467 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1468 __u32 ocd_unused; /* also fix lustre_swab_connect */
1469 __u64 ocd_transno; /* first transno from client to be replayed */
1470 __u32 ocd_group; /* MDS group on OST */
1471 __u32 ocd_cksum_types; /* supported checksum algorithms */
1472 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1473 __u32 ocd_instance; /* instance # of this target */
1474 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1475 /* Fields after ocd_maxbytes are only accessible by the receiver
1476 * if the corresponding flag in ocd_connect_flags is set. Accessing
1477 * any field after ocd_maxbytes on the receiver without a valid flag
1478 * may result in out-of-bound memory access and kernel oops. */
1479 __u64 padding1; /* added 2.1.0. also fix lustre_swab_connect */
1480 __u64 padding2; /* added 2.1.0. also fix lustre_swab_connect */
1481 __u64 padding3; /* added 2.1.0. also fix lustre_swab_connect */
1482 __u64 padding4; /* added 2.1.0. also fix lustre_swab_connect */
1483 __u64 padding5; /* added 2.1.0. also fix lustre_swab_connect */
1484 __u64 padding6; /* added 2.1.0. also fix lustre_swab_connect */
1485 __u64 padding7; /* added 2.1.0. also fix lustre_swab_connect */
1486 __u64 padding8; /* added 2.1.0. also fix lustre_swab_connect */
1487 __u64 padding9; /* added 2.1.0. also fix lustre_swab_connect */
1488 __u64 paddingA; /* added 2.1.0. also fix lustre_swab_connect */
1489 __u64 paddingB; /* added 2.1.0. also fix lustre_swab_connect */
1490 __u64 paddingC; /* added 2.1.0. also fix lustre_swab_connect */
1491 __u64 paddingD; /* added 2.1.0. also fix lustre_swab_connect */
1492 __u64 paddingE; /* added 2.1.0. also fix lustre_swab_connect */
1493 __u64 paddingF; /* added 2.1.0. also fix lustre_swab_connect */
1496 * Please DO NOT use any fields here before first ensuring that this same
1497 * field is not in use on some other branch. Please clear any such changes
1498 * with senior engineers before starting to use a new field. Then, submit
1499 * a small patch against EVERY branch that ONLY adds the new field along with
1500 * the matching OBD_CONNECT flag, so that can be approved and landed easily to
1501 * reserve the flag for future use. */
1504 extern void lustre_swab_connect(struct obd_connect_data *ocd);
1507 * Supported checksum algorithms. Up to 32 checksum types are supported.
1508 * (32-bit mask stored in obd_connect_data::ocd_cksum_types)
1509 * Please update DECLARE_CKSUM_NAME/OBD_CKSUM_ALL in obd.h when adding a new
1510 * algorithm and also the OBD_FL_CKSUM* flags.
1513 OBD_CKSUM_CRC32 = 0x00000001,
1514 OBD_CKSUM_ADLER = 0x00000002,
1515 OBD_CKSUM_CRC32C= 0x00000004,
1519 * OST requests: OBDO & OBD request records
1524 OST_REPLY = 0, /* reply ? */
1540 OST_QUOTACHECK = 18,
1542 OST_QUOTA_ADJUST_QUNIT = 20, /* not used since 2.4 */
1545 #define OST_FIRST_OPC OST_REPLY
1548 OBD_FL_INLINEDATA = 0x00000001,
1549 OBD_FL_OBDMDEXISTS = 0x00000002,
1550 OBD_FL_DELORPHAN = 0x00000004, /* if set in o_flags delete orphans */
1551 OBD_FL_NORPC = 0x00000008, /* set in o_flags do in OSC not OST */
1552 OBD_FL_IDONLY = 0x00000010, /* set in o_flags only adjust obj id*/
1553 OBD_FL_RECREATE_OBJS= 0x00000020, /* recreate missing obj */
1554 OBD_FL_DEBUG_CHECK = 0x00000040, /* echo client/server debug check */
1555 OBD_FL_NO_USRQUOTA = 0x00000100, /* the object's owner is over quota */
1556 OBD_FL_NO_GRPQUOTA = 0x00000200, /* the object's group is over quota */
1557 OBD_FL_CREATE_CROW = 0x00000400, /* object should be create on write */
1558 OBD_FL_SRVLOCK = 0x00000800, /* delegate DLM locking to server */
1559 OBD_FL_CKSUM_CRC32 = 0x00001000, /* CRC32 checksum type */
1560 OBD_FL_CKSUM_ADLER = 0x00002000, /* ADLER checksum type */
1561 OBD_FL_CKSUM_CRC32C = 0x00004000, /* CRC32C checksum type */
1562 OBD_FL_CKSUM_RSVD2 = 0x00008000, /* for future cksum types */
1563 OBD_FL_CKSUM_RSVD3 = 0x00010000, /* for future cksum types */
1564 OBD_FL_SHRINK_GRANT = 0x00020000, /* object shrink the grant */
1565 OBD_FL_MMAP = 0x00040000, /* object is mmapped on the client.
1566 * XXX: obsoleted - reserved for old
1567 * clients prior than 2.2 */
1568 OBD_FL_RECOV_RESEND = 0x00080000, /* recoverable resent */
1569 OBD_FL_NOSPC_BLK = 0x00100000, /* no more block space on OST */
1570 OBD_FL_FLUSH = 0x00200000, /* flush pages on the OST */
1571 OBD_FL_SHORT_IO = 0x00400000, /* short io request */
1573 /* Note that while these checksum values are currently separate bits,
1574 * in 2.x we can actually allow all values from 1-31 if we wanted. */
1575 OBD_FL_CKSUM_ALL = OBD_FL_CKSUM_CRC32 | OBD_FL_CKSUM_ADLER |
1576 OBD_FL_CKSUM_CRC32C,
1578 /* mask for local-only flag, which won't be sent over network */
1579 OBD_FL_LOCAL_MASK = 0xF0000000,
1582 #define LOV_MAGIC_V1 0x0BD10BD0
1583 #define LOV_MAGIC LOV_MAGIC_V1
1584 #define LOV_MAGIC_JOIN_V1 0x0BD20BD0
1585 #define LOV_MAGIC_V3 0x0BD30BD0
1588 * magic for fully defined striping
1589 * the idea is that we should have different magics for striping "hints"
1590 * (struct lov_user_md_v[13]) and defined ready-to-use striping (struct
1591 * lov_mds_md_v[13]). at the moment the magics are used in wire protocol,
1592 * we can't just change it w/o long way preparation, but we still need a
1593 * mechanism to allow LOD to differentiate hint versus ready striping.
1594 * so, at the moment we do a trick: MDT knows what to expect from request
1595 * depending on the case (replay uses ready striping, non-replay req uses
1596 * hints), so MDT replaces magic with appropriate one and now LOD can
1597 * easily understand what's inside -bzzz
1599 #define LOV_MAGIC_V1_DEF 0x0CD10BD0
1600 #define LOV_MAGIC_V3_DEF 0x0CD30BD0
1602 #define LOV_PATTERN_RAID0 0x001 /* stripes are used round-robin */
1603 #define LOV_PATTERN_RAID1 0x002 /* stripes are mirrors of each other */
1604 #define LOV_PATTERN_FIRST 0x100 /* first stripe is not in round-robin */
1605 #define LOV_PATTERN_CMOBD 0x200
1607 #define LOV_PATTERN_F_MASK 0xffff0000
1608 #define LOV_PATTERN_F_RELEASED 0x80000000 /* HSM released file */
1610 #define lov_pattern(pattern) (pattern & ~LOV_PATTERN_F_MASK)
1611 #define lov_pattern_flags(pattern) (pattern & LOV_PATTERN_F_MASK)
1613 #define lov_ost_data lov_ost_data_v1
1614 struct lov_ost_data_v1 { /* per-stripe data structure (little-endian)*/
1615 struct ost_id l_ost_oi; /* OST object ID */
1616 __u32 l_ost_gen; /* generation of this l_ost_idx */
1617 __u32 l_ost_idx; /* OST index in LOV (lov_tgt_desc->tgts) */
1620 #define lov_mds_md lov_mds_md_v1
1621 struct lov_mds_md_v1 { /* LOV EA mds/wire data (little-endian) */
1622 __u32 lmm_magic; /* magic number = LOV_MAGIC_V1 */
1623 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1624 struct ost_id lmm_oi; /* LOV object ID */
1625 __u32 lmm_stripe_size; /* size of stripe in bytes */
1626 /* lmm_stripe_count used to be __u32 */
1627 __u16 lmm_stripe_count; /* num stripes in use for this object */
1628 __u16 lmm_layout_gen; /* layout generation number */
1629 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1633 * Sigh, because pre-2.4 uses
1634 * struct lov_mds_md_v1 {
1636 * __u64 lmm_object_id;
1637 * __u64 lmm_object_seq;
1640 * to identify the LOV(MDT) object, and lmm_object_seq will
1641 * be normal_fid, which make it hard to combine these conversion
1642 * to ostid_to FID. so we will do lmm_oi/fid conversion separately
1644 * We can tell the lmm_oi by this way,
1645 * 1.8: lmm_object_id = {inode}, lmm_object_gr = 0
1646 * 2.1: lmm_object_id = {oid < 128k}, lmm_object_seq = FID_SEQ_NORMAL
1647 * 2.4: lmm_oi.f_seq = FID_SEQ_NORMAL, lmm_oi.f_oid = {oid < 128k},
1650 * But currently lmm_oi/lsm_oi does not have any "real" usages,
1651 * except for printing some information, and the user can always
1652 * get the real FID from LMA, besides this multiple case check might
1653 * make swab more complicate. So we will keep using id/seq for lmm_oi.
1656 static inline void fid_to_lmm_oi(const struct lu_fid *fid,
1659 oi->oi.oi_id = fid_oid(fid);
1660 oi->oi.oi_seq = fid_seq(fid);
1663 static inline void lmm_oi_set_seq(struct ost_id *oi, __u64 seq)
1665 oi->oi.oi_seq = seq;
1668 static inline void lmm_oi_set_id(struct ost_id *oi, __u64 oid)
1673 static inline __u64 lmm_oi_id(struct ost_id *oi)
1675 return oi->oi.oi_id;
1678 static inline __u64 lmm_oi_seq(struct ost_id *oi)
1680 return oi->oi.oi_seq;
1683 static inline void lmm_oi_le_to_cpu(struct ost_id *dst_oi,
1684 struct ost_id *src_oi)
1686 dst_oi->oi.oi_id = le64_to_cpu(src_oi->oi.oi_id);
1687 dst_oi->oi.oi_seq = le64_to_cpu(src_oi->oi.oi_seq);
1690 static inline void lmm_oi_cpu_to_le(struct ost_id *dst_oi,
1691 struct ost_id *src_oi)
1693 dst_oi->oi.oi_id = cpu_to_le64(src_oi->oi.oi_id);
1694 dst_oi->oi.oi_seq = cpu_to_le64(src_oi->oi.oi_seq);
1697 /* extern void lustre_swab_lov_mds_md(struct lov_mds_md *llm); */
1699 #define MAX_MD_SIZE (sizeof(struct lov_mds_md) + 4 * sizeof(struct lov_ost_data))
1700 #define MIN_MD_SIZE (sizeof(struct lov_mds_md) + 1 * sizeof(struct lov_ost_data))
1702 #define XATTR_NAME_ACL_ACCESS "system.posix_acl_access"
1703 #define XATTR_NAME_ACL_DEFAULT "system.posix_acl_default"
1704 #define XATTR_USER_PREFIX "user."
1705 #define XATTR_TRUSTED_PREFIX "trusted."
1706 #define XATTR_SECURITY_PREFIX "security."
1707 #define XATTR_LUSTRE_PREFIX "lustre."
1709 #define XATTR_NAME_LOV "trusted.lov"
1710 #define XATTR_NAME_LMA "trusted.lma"
1711 #define XATTR_NAME_LMV "trusted.lmv"
1712 #define XATTR_NAME_DEFAULT_LMV "trusted.dmv"
1713 #define XATTR_NAME_LINK "trusted.link"
1714 #define XATTR_NAME_FID "trusted.fid"
1715 #define XATTR_NAME_VERSION "trusted.version"
1716 #define XATTR_NAME_SOM "trusted.som"
1717 #define XATTR_NAME_HSM "trusted.hsm"
1718 #define XATTR_NAME_LFSCK_NAMESPACE "trusted.lfsck_namespace"
1720 struct lov_mds_md_v3 { /* LOV EA mds/wire data (little-endian) */
1721 __u32 lmm_magic; /* magic number = LOV_MAGIC_V3 */
1722 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1723 struct ost_id lmm_oi; /* LOV object ID */
1724 __u32 lmm_stripe_size; /* size of stripe in bytes */
1725 /* lmm_stripe_count used to be __u32 */
1726 __u16 lmm_stripe_count; /* num stripes in use for this object */
1727 __u16 lmm_layout_gen; /* layout generation number */
1728 char lmm_pool_name[LOV_MAXPOOLNAME]; /* must be 32bit aligned */
1729 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1732 static inline __u32 lov_mds_md_size(__u16 stripes, __u32 lmm_magic)
1734 if (lmm_magic == LOV_MAGIC_V3)
1735 return sizeof(struct lov_mds_md_v3) +
1736 stripes * sizeof(struct lov_ost_data_v1);
1738 return sizeof(struct lov_mds_md_v1) +
1739 stripes * sizeof(struct lov_ost_data_v1);
1743 #define OBD_MD_FLID (0x00000001ULL) /* object ID */
1744 #define OBD_MD_FLATIME (0x00000002ULL) /* access time */
1745 #define OBD_MD_FLMTIME (0x00000004ULL) /* data modification time */
1746 #define OBD_MD_FLCTIME (0x00000008ULL) /* change time */
1747 #define OBD_MD_FLSIZE (0x00000010ULL) /* size */
1748 #define OBD_MD_FLBLOCKS (0x00000020ULL) /* allocated blocks count */
1749 #define OBD_MD_FLBLKSZ (0x00000040ULL) /* block size */
1750 #define OBD_MD_FLMODE (0x00000080ULL) /* access bits (mode & ~S_IFMT) */
1751 #define OBD_MD_FLTYPE (0x00000100ULL) /* object type (mode & S_IFMT) */
1752 #define OBD_MD_FLUID (0x00000200ULL) /* user ID */
1753 #define OBD_MD_FLGID (0x00000400ULL) /* group ID */
1754 #define OBD_MD_FLFLAGS (0x00000800ULL) /* flags word */
1755 #define OBD_MD_FLNLINK (0x00002000ULL) /* link count */
1756 #define OBD_MD_FLGENER (0x00004000ULL) /* generation number */
1757 /*#define OBD_MD_FLINLINE (0x00008000ULL) inline data. used until 1.6.5 */
1758 #define OBD_MD_FLRDEV (0x00010000ULL) /* device number */
1759 #define OBD_MD_FLEASIZE (0x00020000ULL) /* extended attribute data */
1760 #define OBD_MD_LINKNAME (0x00040000ULL) /* symbolic link target */
1761 #define OBD_MD_FLHANDLE (0x00080000ULL) /* file/lock handle */
1762 #define OBD_MD_FLCKSUM (0x00100000ULL) /* bulk data checksum */
1763 #define OBD_MD_FLQOS (0x00200000ULL) /* quality of service stats */
1764 /*#define OBD_MD_FLOSCOPQ (0x00400000ULL) osc opaque data, never used */
1765 #define OBD_MD_FLCOOKIE (0x00800000ULL) /* log cancellation cookie */
1766 #define OBD_MD_FLGROUP (0x01000000ULL) /* group */
1767 #define OBD_MD_FLFID (0x02000000ULL) /* ->ost write inline fid */
1768 #define OBD_MD_FLEPOCH (0x04000000ULL) /* ->ost write with ioepoch */
1769 /* ->mds if epoch opens or closes */
1770 #define OBD_MD_FLGRANT (0x08000000ULL) /* ost preallocation space grant */
1771 #define OBD_MD_FLDIREA (0x10000000ULL) /* dir's extended attribute data */
1772 #define OBD_MD_FLUSRQUOTA (0x20000000ULL) /* over quota flags sent from ost */
1773 #define OBD_MD_FLGRPQUOTA (0x40000000ULL) /* over quota flags sent from ost */
1774 #define OBD_MD_FLMODEASIZE (0x80000000ULL) /* EA size will be changed */
1776 #define OBD_MD_MDS (0x0000000100000000ULL) /* where an inode lives on */
1777 #define OBD_MD_REINT (0x0000000200000000ULL) /* reintegrate oa */
1778 #define OBD_MD_MEA (0x0000000400000000ULL) /* CMD split EA */
1779 #define OBD_MD_TSTATE (0x0000000800000000ULL) /* transient state field */
1781 #define OBD_MD_FLXATTR (0x0000001000000000ULL) /* xattr */
1782 #define OBD_MD_FLXATTRLS (0x0000002000000000ULL) /* xattr list */
1783 #define OBD_MD_FLXATTRRM (0x0000004000000000ULL) /* xattr remove */
1784 #define OBD_MD_FLACL (0x0000008000000000ULL) /* ACL */
1785 #define OBD_MD_FLRMTPERM (0x0000010000000000ULL) /* remote permission */
1786 #define OBD_MD_FLMDSCAPA (0x0000020000000000ULL) /* MDS capability */
1787 #define OBD_MD_FLOSSCAPA (0x0000040000000000ULL) /* OSS capability */
1788 #define OBD_MD_FLCKSPLIT (0x0000080000000000ULL) /* Check split on server */
1789 #define OBD_MD_FLCROSSREF (0x0000100000000000ULL) /* Cross-ref case */
1790 #define OBD_MD_FLGETATTRLOCK (0x0000200000000000ULL) /* Get IOEpoch attributes
1791 * under lock; for xattr
1792 * requests means the
1793 * client holds the lock */
1794 #define OBD_MD_FLOBJCOUNT (0x0000400000000000ULL) /* for multiple destroy */
1796 #define OBD_MD_FLRMTLSETFACL (0x0001000000000000ULL) /* lfs lsetfacl case */
1797 #define OBD_MD_FLRMTLGETFACL (0x0002000000000000ULL) /* lfs lgetfacl case */
1798 #define OBD_MD_FLRMTRSETFACL (0x0004000000000000ULL) /* lfs rsetfacl case */
1799 #define OBD_MD_FLRMTRGETFACL (0x0008000000000000ULL) /* lfs rgetfacl case */
1801 #define OBD_MD_FLDATAVERSION (0x0010000000000000ULL) /* iversion sum */
1802 #define OBD_MD_FLRELEASED (0x0020000000000000ULL) /* file released */
1804 #define OBD_MD_DEFAULT_MEA (0x0040000000000000ULL) /* default MEA */
1806 #define OBD_MD_FLGETATTR (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLMTIME | \
1807 OBD_MD_FLCTIME | OBD_MD_FLSIZE | OBD_MD_FLBLKSZ | \
1808 OBD_MD_FLMODE | OBD_MD_FLTYPE | OBD_MD_FLUID | \
1809 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLNLINK | \
1810 OBD_MD_FLGENER | OBD_MD_FLRDEV | OBD_MD_FLGROUP)
1812 #define OBD_MD_FLXATTRALL (OBD_MD_FLXATTR | OBD_MD_FLXATTRLS)
1814 /* don't forget obdo_fid which is way down at the bottom so it can
1815 * come after the definition of llog_cookie */
1819 HSS_CLEARMASK = 0x02,
1820 HSS_ARCHIVE_ID = 0x04,
1823 struct hsm_state_set {
1825 __u32 hss_archive_id;
1827 __u64 hss_clearmask;
1830 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
1831 extern void lustre_swab_hsm_state_set(struct hsm_state_set *hss);
1833 extern void lustre_swab_obd_statfs (struct obd_statfs *os);
1835 /* ost_body.data values for OST_BRW */
1837 #define OBD_BRW_READ 0x01
1838 #define OBD_BRW_WRITE 0x02
1839 #define OBD_BRW_RWMASK (OBD_BRW_READ | OBD_BRW_WRITE)
1840 #define OBD_BRW_SYNC 0x08 /* this page is a part of synchronous
1841 * transfer and is not accounted in
1843 #define OBD_BRW_CHECK 0x10
1844 #define OBD_BRW_FROM_GRANT 0x20 /* the osc manages this under llite */
1845 #define OBD_BRW_GRANTED 0x40 /* the ost manages this */
1846 #define OBD_BRW_NOCACHE 0x80 /* this page is a part of non-cached IO */
1847 #define OBD_BRW_NOQUOTA 0x100
1848 #define OBD_BRW_SRVLOCK 0x200 /* Client holds no lock over this page */
1849 #define OBD_BRW_ASYNC 0x400 /* Server may delay commit to disk */
1850 #define OBD_BRW_MEMALLOC 0x800 /* Client runs in the "kswapd" context */
1851 #define OBD_BRW_OVER_USRQUOTA 0x1000 /* Running out of user quota */
1852 #define OBD_BRW_OVER_GRPQUOTA 0x2000 /* Running out of group quota */
1853 #define OBD_BRW_SOFT_SYNC 0x4000 /* This flag notifies the server
1854 * that the client is running low on
1855 * space for unstable pages; asking
1856 * it to sync quickly */
1858 #define OBD_OBJECT_EOF 0xffffffffffffffffULL
1860 #define OST_MIN_PRECREATE 32
1861 #define OST_MAX_PRECREATE 20000
1864 struct ost_id ioo_oid; /* object ID, if multi-obj BRW */
1865 __u32 ioo_max_brw; /* low 16 bits were o_mode before 2.4,
1866 * now (PTLRPC_BULK_OPS_COUNT - 1) in
1867 * high 16 bits in 2.4 and later */
1868 __u32 ioo_bufcnt; /* number of niobufs for this object */
1871 #define IOOBJ_MAX_BRW_BITS 16
1872 #define IOOBJ_TYPE_MASK ((1U << IOOBJ_MAX_BRW_BITS) - 1)
1873 #define ioobj_max_brw_get(ioo) (((ioo)->ioo_max_brw >> IOOBJ_MAX_BRW_BITS) + 1)
1874 #define ioobj_max_brw_set(ioo, num) \
1875 do { (ioo)->ioo_max_brw = ((num) - 1) << IOOBJ_MAX_BRW_BITS; } while (0)
1877 extern void lustre_swab_obd_ioobj (struct obd_ioobj *ioo);
1879 /* multiple of 8 bytes => can array */
1880 struct niobuf_remote {
1886 extern void lustre_swab_niobuf_remote (struct niobuf_remote *nbr);
1888 /* lock value block communicated between the filter and llite */
1890 /* OST_LVB_ERR_INIT is needed because the return code in rc is
1891 * negative, i.e. because ((MASK + rc) & MASK) != MASK. */
1892 #define OST_LVB_ERR_INIT 0xffbadbad80000000ULL
1893 #define OST_LVB_ERR_MASK 0xffbadbad00000000ULL
1894 #define OST_LVB_IS_ERR(blocks) \
1895 ((blocks & OST_LVB_ERR_MASK) == OST_LVB_ERR_MASK)
1896 #define OST_LVB_SET_ERR(blocks, rc) \
1897 do { blocks = OST_LVB_ERR_INIT + rc; } while (0)
1898 #define OST_LVB_GET_ERR(blocks) (int)(blocks - OST_LVB_ERR_INIT)
1908 extern void lustre_swab_ost_lvb_v1(struct ost_lvb_v1 *lvb);
1922 extern void lustre_swab_ost_lvb(struct ost_lvb *lvb);
1925 * lquota data structures
1928 #ifndef QUOTABLOCK_BITS
1929 #define QUOTABLOCK_BITS 10
1932 #ifndef QUOTABLOCK_SIZE
1933 #define QUOTABLOCK_SIZE (1 << QUOTABLOCK_BITS)
1937 #define toqb(x) (((x) + QUOTABLOCK_SIZE - 1) >> QUOTABLOCK_BITS)
1940 /* The lquota_id structure is an union of all the possible identifier types that
1941 * can be used with quota, this includes:
1944 * - a FID which can be used for per-directory quota in the future */
1946 struct lu_fid qid_fid; /* FID for per-directory quota */
1947 __u64 qid_uid; /* user identifier */
1948 __u64 qid_gid; /* group identifier */
1951 /* quotactl management */
1952 struct obd_quotactl {
1954 __u32 qc_type; /* see Q_* flag below */
1957 struct obd_dqinfo qc_dqinfo;
1958 struct obd_dqblk qc_dqblk;
1961 extern void lustre_swab_obd_quotactl(struct obd_quotactl *q);
1963 #define Q_QUOTACHECK 0x800100 /* deprecated as of 2.4 */
1964 #define Q_INITQUOTA 0x800101 /* deprecated as of 2.4 */
1965 #define Q_GETOINFO 0x800102 /* get obd quota info */
1966 #define Q_GETOQUOTA 0x800103 /* get obd quotas */
1967 #define Q_FINVALIDATE 0x800104 /* deprecated as of 2.4 */
1969 #define Q_COPY(out, in, member) (out)->member = (in)->member
1971 #define QCTL_COPY(out, in) \
1973 Q_COPY(out, in, qc_cmd); \
1974 Q_COPY(out, in, qc_type); \
1975 Q_COPY(out, in, qc_id); \
1976 Q_COPY(out, in, qc_stat); \
1977 Q_COPY(out, in, qc_dqinfo); \
1978 Q_COPY(out, in, qc_dqblk); \
1981 /* Body of quota request used for quota acquire/release RPCs between quota
1982 * master (aka QMT) and slaves (ak QSD). */
1984 struct lu_fid qb_fid; /* FID of global index packing the pool ID
1985 * and type (data or metadata) as well as
1986 * the quota type (user or group). */
1987 union lquota_id qb_id; /* uid or gid or directory FID */
1988 __u32 qb_flags; /* see below */
1990 __u64 qb_count; /* acquire/release count (kbytes/inodes) */
1991 __u64 qb_usage; /* current slave usage (kbytes/inodes) */
1992 __u64 qb_slv_ver; /* slave index file version */
1993 struct lustre_handle qb_lockh; /* per-ID lock handle */
1994 struct lustre_handle qb_glb_lockh; /* global lock handle */
1995 __u64 qb_padding1[4];
1998 /* When the quota_body is used in the reply of quota global intent
1999 * lock (IT_QUOTA_CONN) reply, qb_fid contains slave index file FID. */
2000 #define qb_slv_fid qb_fid
2001 /* qb_usage is the current qunit (in kbytes/inodes) when quota_body is used in
2003 #define qb_qunit qb_usage
2005 #define QUOTA_DQACQ_FL_ACQ 0x1 /* acquire quota */
2006 #define QUOTA_DQACQ_FL_PREACQ 0x2 /* pre-acquire */
2007 #define QUOTA_DQACQ_FL_REL 0x4 /* release quota */
2008 #define QUOTA_DQACQ_FL_REPORT 0x8 /* report usage */
2010 extern void lustre_swab_quota_body(struct quota_body *b);
2012 /* Quota types currently supported */
2014 LQUOTA_TYPE_USR = 0x00, /* maps to USRQUOTA */
2015 LQUOTA_TYPE_GRP = 0x01, /* maps to GRPQUOTA */
2019 /* There are 2 different resource types on which a quota limit can be enforced:
2020 * - inodes on the MDTs
2021 * - blocks on the OSTs */
2023 LQUOTA_RES_MD = 0x01, /* skip 0 to avoid null oid in FID */
2024 LQUOTA_RES_DT = 0x02,
2026 LQUOTA_FIRST_RES = LQUOTA_RES_MD
2028 #define LQUOTA_NR_RES (LQUOTA_LAST_RES - LQUOTA_FIRST_RES + 1)
2031 * Space accounting support
2032 * Format of an accounting record, providing disk usage information for a given
2035 struct lquota_acct_rec { /* 16 bytes */
2036 __u64 bspace; /* current space in use */
2037 __u64 ispace; /* current # inodes in use */
2041 * Global quota index support
2042 * Format of a global record, providing global quota settings for a given quota
2045 struct lquota_glb_rec { /* 32 bytes */
2046 __u64 qbr_hardlimit; /* quota hard limit, in #inodes or kbytes */
2047 __u64 qbr_softlimit; /* quota soft limit, in #inodes or kbytes */
2048 __u64 qbr_time; /* grace time, in seconds */
2049 __u64 qbr_granted; /* how much is granted to slaves, in #inodes or
2054 * Slave index support
2055 * Format of a slave record, recording how much space is granted to a given
2058 struct lquota_slv_rec { /* 8 bytes */
2059 __u64 qsr_granted; /* space granted to the slave for the key=ID,
2060 * in #inodes or kbytes */
2063 /* Data structures associated with the quota locks */
2065 /* Glimpse descriptor used for the index & per-ID quota locks */
2066 struct ldlm_gl_lquota_desc {
2067 union lquota_id gl_id; /* quota ID subject to the glimpse */
2068 __u64 gl_flags; /* see LQUOTA_FL* below */
2069 __u64 gl_ver; /* new index version */
2070 __u64 gl_hardlimit; /* new hardlimit or qunit value */
2071 __u64 gl_softlimit; /* new softlimit */
2075 #define gl_qunit gl_hardlimit /* current qunit value used when
2076 * glimpsing per-ID quota locks */
2078 /* quota glimpse flags */
2079 #define LQUOTA_FL_EDQUOT 0x1 /* user/group out of quota space on QMT */
2081 /* LVB used with quota (global and per-ID) locks */
2083 __u64 lvb_flags; /* see LQUOTA_FL* above */
2084 __u64 lvb_id_may_rel; /* space that might be released later */
2085 __u64 lvb_id_rel; /* space released by the slave for this ID */
2086 __u64 lvb_id_qunit; /* current qunit value */
2090 extern void lustre_swab_lquota_lvb(struct lquota_lvb *lvb);
2092 /* LVB used with global quota lock */
2093 #define lvb_glb_ver lvb_id_may_rel /* current version of the global index */
2101 #define QUOTA_FIRST_OPC QUOTA_DQACQ
2110 MDS_GETATTR_NAME = 34,
2115 MDS_DISCONNECT = 39,
2121 MDS_DONE_WRITING = 45,
2123 MDS_QUOTACHECK = 47,
2126 MDS_SETXATTR = 50, /* obsolete, now it's MDS_REINT op */
2130 MDS_HSM_STATE_GET = 54,
2131 MDS_HSM_STATE_SET = 55,
2132 MDS_HSM_ACTION = 56,
2133 MDS_HSM_PROGRESS = 57,
2134 MDS_HSM_REQUEST = 58,
2135 MDS_HSM_CT_REGISTER = 59,
2136 MDS_HSM_CT_UNREGISTER = 60,
2137 MDS_SWAP_LAYOUTS = 61,
2141 #define MDS_FIRST_OPC MDS_GETATTR
2144 /* opcodes for object update */
2150 #define UPDATE_FIRST_OPC UPDATE_OBJ
2167 } mds_reint_t, mdt_reint_t;
2169 extern void lustre_swab_generic_32s (__u32 *val);
2171 /* the disposition of the intent outlines what was executed */
2172 #define DISP_IT_EXECD 0x00000001
2173 #define DISP_LOOKUP_EXECD 0x00000002
2174 #define DISP_LOOKUP_NEG 0x00000004
2175 #define DISP_LOOKUP_POS 0x00000008
2176 #define DISP_OPEN_CREATE 0x00000010
2177 #define DISP_OPEN_OPEN 0x00000020
2178 #define DISP_ENQ_COMPLETE 0x00400000 /* obsolete and unused */
2179 #define DISP_ENQ_OPEN_REF 0x00800000
2180 #define DISP_ENQ_CREATE_REF 0x01000000
2181 #define DISP_OPEN_LOCK 0x02000000
2182 #define DISP_OPEN_LEASE 0x04000000
2183 #define DISP_OPEN_STRIPE 0x08000000
2185 /* INODE LOCK PARTS */
2186 #define MDS_INODELOCK_LOOKUP 0x000001 /* For namespace, dentry etc, and also
2187 * was used to protect permission (mode,
2188 * owner, group etc) before 2.4. */
2189 #define MDS_INODELOCK_UPDATE 0x000002 /* size, links, timestamps */
2190 #define MDS_INODELOCK_OPEN 0x000004 /* For opened files */
2191 #define MDS_INODELOCK_LAYOUT 0x000008 /* for layout */
2193 /* The PERM bit is added int 2.4, and it is used to protect permission(mode,
2194 * owner, group, acl etc), so to separate the permission from LOOKUP lock.
2195 * Because for remote directories(in DNE), these locks will be granted by
2196 * different MDTs(different ldlm namespace).
2198 * For local directory, MDT will always grant UPDATE_LOCK|PERM_LOCK together.
2199 * For Remote directory, the master MDT, where the remote directory is, will
2200 * grant UPDATE_LOCK|PERM_LOCK, and the remote MDT, where the name entry is,
2201 * will grant LOOKUP_LOCK. */
2202 #define MDS_INODELOCK_PERM 0x000010
2203 #define MDS_INODELOCK_XATTR 0x000020 /* extended attributes */
2205 #define MDS_INODELOCK_MAXSHIFT 5
2206 /* This FULL lock is useful to take on unlink sort of operations */
2207 #define MDS_INODELOCK_FULL ((1<<(MDS_INODELOCK_MAXSHIFT+1))-1)
2209 extern void lustre_swab_ll_fid (struct ll_fid *fid);
2211 /* NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
2212 * but was moved into name[1] along with the OID to avoid consuming the
2213 * name[2,3] fields that need to be used for the quota id (also a FID). */
2215 LUSTRE_RES_ID_SEQ_OFF = 0,
2216 LUSTRE_RES_ID_VER_OID_OFF = 1,
2217 LUSTRE_RES_ID_WAS_VER_OFF = 2, /* see note above */
2218 LUSTRE_RES_ID_QUOTA_SEQ_OFF = 2,
2219 LUSTRE_RES_ID_QUOTA_VER_OID_OFF = 3,
2220 LUSTRE_RES_ID_HSH_OFF = 3
2223 #define MDS_STATUS_CONN 1
2224 #define MDS_STATUS_LOV 2
2226 /* mdt_thread_info.mti_flags. */
2228 /* The flag indicates Size-on-MDS attributes are changed. */
2229 MF_SOM_CHANGE = (1 << 0),
2230 /* Flags indicates an epoch opens or closes. */
2231 MF_EPOCH_OPEN = (1 << 1),
2232 MF_EPOCH_CLOSE = (1 << 2),
2233 MF_MDC_CANCEL_FID1 = (1 << 3),
2234 MF_MDC_CANCEL_FID2 = (1 << 4),
2235 MF_MDC_CANCEL_FID3 = (1 << 5),
2236 MF_MDC_CANCEL_FID4 = (1 << 6),
2237 /* There is a pending attribute update. */
2238 MF_SOM_AU = (1 << 7),
2239 /* Cancel OST locks while getattr OST attributes. */
2240 MF_GETATTR_LOCK = (1 << 8),
2241 MF_GET_MDT_IDX = (1 << 9),
2244 #define MF_SOM_LOCAL_FLAGS (MF_SOM_CHANGE | MF_EPOCH_OPEN | MF_EPOCH_CLOSE)
2246 #define LUSTRE_BFLAG_UNCOMMITTED_WRITES 0x1
2248 /* these should be identical to their EXT4_*_FL counterparts, they are
2249 * redefined here only to avoid dragging in fs/ext4/ext4.h */
2250 #define LUSTRE_SYNC_FL 0x00000008 /* Synchronous updates */
2251 #define LUSTRE_IMMUTABLE_FL 0x00000010 /* Immutable file */
2252 #define LUSTRE_APPEND_FL 0x00000020 /* writes to file may only append */
2253 #define LUSTRE_NOATIME_FL 0x00000080 /* do not update atime */
2254 #define LUSTRE_DIRSYNC_FL 0x00010000 /* dirsync behaviour (dir only) */
2257 /* Convert wire LUSTRE_*_FL to corresponding client local VFS S_* values
2258 * for the client inode i_flags. The LUSTRE_*_FL are the Lustre wire
2259 * protocol equivalents of LDISKFS_*_FL values stored on disk, while
2260 * the S_* flags are kernel-internal values that change between kernel
2261 * versions. These flags are set/cleared via FSFILT_IOC_{GET,SET}_FLAGS.
2262 * See b=16526 for a full history. */
2263 static inline int ll_ext_to_inode_flags(int flags)
2265 return (((flags & LUSTRE_SYNC_FL) ? S_SYNC : 0) |
2266 ((flags & LUSTRE_NOATIME_FL) ? S_NOATIME : 0) |
2267 ((flags & LUSTRE_APPEND_FL) ? S_APPEND : 0) |
2268 #if defined(S_DIRSYNC)
2269 ((flags & LUSTRE_DIRSYNC_FL) ? S_DIRSYNC : 0) |
2271 ((flags & LUSTRE_IMMUTABLE_FL) ? S_IMMUTABLE : 0));
2274 static inline int ll_inode_to_ext_flags(int iflags)
2276 return (((iflags & S_SYNC) ? LUSTRE_SYNC_FL : 0) |
2277 ((iflags & S_NOATIME) ? LUSTRE_NOATIME_FL : 0) |
2278 ((iflags & S_APPEND) ? LUSTRE_APPEND_FL : 0) |
2279 #if defined(S_DIRSYNC)
2280 ((iflags & S_DIRSYNC) ? LUSTRE_DIRSYNC_FL : 0) |
2282 ((iflags & S_IMMUTABLE) ? LUSTRE_IMMUTABLE_FL : 0));
2286 /* 64 possible states */
2287 enum md_transient_state {
2288 MS_RESTORE = (1 << 0), /* restore is running */
2294 struct lustre_handle handle;
2296 __u64 size; /* Offset, in the case of MDS_READPAGE */
2300 __u64 blocks; /* XID, in the case of MDS_READPAGE */
2302 __u64 t_state; /* transient file state defined in
2303 * enum md_transient_state
2304 * was "ino" until 2.4.0 */
2311 __u32 flags; /* from vfs for pin/unpin, LUSTRE_BFLAG close */
2313 __u32 nlink; /* #bytes to read in the case of MDS_READPAGE */
2314 __u32 unused2; /* was "generation" until 2.4.0 */
2319 __u32 max_cookiesize;
2320 __u32 uid_h; /* high 32-bits of uid, for FUID */
2321 __u32 gid_h; /* high 32-bits of gid, for FUID */
2322 __u32 padding_5; /* also fix lustre_swab_mdt_body */
2330 extern void lustre_swab_mdt_body (struct mdt_body *b);
2332 struct mdt_ioepoch {
2333 struct lustre_handle handle;
2339 extern void lustre_swab_mdt_ioepoch (struct mdt_ioepoch *b);
2341 /* permissions for md_perm.mp_perm */
2343 CFS_SETUID_PERM = 0x01,
2344 CFS_SETGID_PERM = 0x02,
2345 CFS_SETGRP_PERM = 0x04,
2346 CFS_RMTACL_PERM = 0x08,
2347 CFS_RMTOWN_PERM = 0x10
2350 /* inode access permission for remote user, the inode info are omitted,
2351 * for client knows them. */
2352 struct mdt_remote_perm {
2359 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
2363 extern void lustre_swab_mdt_remote_perm(struct mdt_remote_perm *p);
2365 struct mdt_rec_setattr {
2375 __u32 sa_padding_1_h;
2376 struct lu_fid sa_fid;
2385 __u32 sa_attr_flags;
2387 __u32 sa_bias; /* some operation flags */
2393 extern void lustre_swab_mdt_rec_setattr (struct mdt_rec_setattr *sa);
2396 * Attribute flags used in mdt_rec_setattr::sa_valid.
2397 * The kernel's #defines for ATTR_* should not be used over the network
2398 * since the client and MDS may run different kernels (see bug 13828)
2399 * Therefore, we should only use MDS_ATTR_* attributes for sa_valid.
2401 #define MDS_ATTR_MODE 0x1ULL /* = 1 */
2402 #define MDS_ATTR_UID 0x2ULL /* = 2 */
2403 #define MDS_ATTR_GID 0x4ULL /* = 4 */
2404 #define MDS_ATTR_SIZE 0x8ULL /* = 8 */
2405 #define MDS_ATTR_ATIME 0x10ULL /* = 16 */
2406 #define MDS_ATTR_MTIME 0x20ULL /* = 32 */
2407 #define MDS_ATTR_CTIME 0x40ULL /* = 64 */
2408 #define MDS_ATTR_ATIME_SET 0x80ULL /* = 128 */
2409 #define MDS_ATTR_MTIME_SET 0x100ULL /* = 256 */
2410 #define MDS_ATTR_FORCE 0x200ULL /* = 512, Not a change, but a change it */
2411 #define MDS_ATTR_ATTR_FLAG 0x400ULL /* = 1024 */
2412 #define MDS_ATTR_KILL_SUID 0x800ULL /* = 2048 */
2413 #define MDS_ATTR_KILL_SGID 0x1000ULL /* = 4096 */
2414 #define MDS_ATTR_CTIME_SET 0x2000ULL /* = 8192 */
2415 #define MDS_ATTR_FROM_OPEN 0x4000ULL /* = 16384, called from open path, ie O_TRUNC */
2416 #define MDS_ATTR_BLOCKS 0x8000ULL /* = 32768 */
2419 #define FMODE_READ 00000001
2420 #define FMODE_WRITE 00000002
2423 #define MDS_FMODE_CLOSED 00000000
2424 #define MDS_FMODE_EXEC 00000004
2425 /* IO Epoch is opened on a closed file. */
2426 #define MDS_FMODE_EPOCH 01000000
2427 /* IO Epoch is opened on a file truncate. */
2428 #define MDS_FMODE_TRUNC 02000000
2429 /* Size-on-MDS Attribute Update is pending. */
2430 #define MDS_FMODE_SOM 04000000
2432 #define MDS_OPEN_CREATED 00000010
2433 #define MDS_OPEN_CROSS 00000020
2435 #define MDS_OPEN_CREAT 00000100
2436 #define MDS_OPEN_EXCL 00000200
2437 #define MDS_OPEN_TRUNC 00001000
2438 #define MDS_OPEN_APPEND 00002000
2439 #define MDS_OPEN_SYNC 00010000
2440 #define MDS_OPEN_DIRECTORY 00200000
2442 #define MDS_OPEN_BY_FID 040000000 /* open_by_fid for known object */
2443 #define MDS_OPEN_DELAY_CREATE 0100000000 /* delay initial object create */
2444 #define MDS_OPEN_OWNEROVERRIDE 0200000000 /* NFSD rw-reopen ro file for owner */
2445 #define MDS_OPEN_JOIN_FILE 0400000000 /* open for join file.
2446 * We do not support JOIN FILE
2447 * anymore, reserve this flags
2448 * just for preventing such bit
2451 #define MDS_OPEN_LOCK 04000000000 /* This open requires open lock */
2452 #define MDS_OPEN_HAS_EA 010000000000 /* specify object create pattern */
2453 #define MDS_OPEN_HAS_OBJS 020000000000 /* Just set the EA the obj exist */
2454 #define MDS_OPEN_NORESTORE 0100000000000ULL /* Do not restore file at open */
2455 #define MDS_OPEN_NEWSTRIPE 0200000000000ULL /* New stripe needed (restripe or
2457 #define MDS_OPEN_VOLATILE 0400000000000ULL /* File is volatile = created
2459 #define MDS_OPEN_LEASE 01000000000000ULL /* Open the file and grant lease
2460 * delegation, succeed if it's not
2461 * being opened with conflict mode.
2463 #define MDS_OPEN_RELEASE 02000000000000ULL /* Open the file for HSM release */
2465 /* permission for create non-directory file */
2466 #define MAY_CREATE (1 << 7)
2467 /* permission for create directory file */
2468 #define MAY_LINK (1 << 8)
2469 /* permission for delete from the directory */
2470 #define MAY_UNLINK (1 << 9)
2471 /* source's permission for rename */
2472 #define MAY_RENAME_SRC (1 << 10)
2473 /* target's permission for rename */
2474 #define MAY_RENAME_TAR (1 << 11)
2475 /* part (parent's) VTX permission check */
2476 #define MAY_VTX_PART (1 << 12)
2477 /* full VTX permission check */
2478 #define MAY_VTX_FULL (1 << 13)
2479 /* lfs rgetfacl permission check */
2480 #define MAY_RGETFACL (1 << 14)
2483 MDS_CHECK_SPLIT = 1 << 0,
2484 MDS_CROSS_REF = 1 << 1,
2485 MDS_VTX_BYPASS = 1 << 2,
2486 MDS_PERM_BYPASS = 1 << 3,
2488 MDS_QUOTA_IGNORE = 1 << 5,
2489 /* Was MDS_CLOSE_CLEANUP (1 << 6), No more used */
2490 MDS_KEEP_ORPHAN = 1 << 7,
2491 MDS_RECOV_OPEN = 1 << 8,
2492 MDS_DATA_MODIFIED = 1 << 9,
2493 MDS_CREATE_VOLATILE = 1 << 10,
2494 MDS_OWNEROVERRIDE = 1 << 11,
2495 MDS_HSM_RELEASE = 1 << 12,
2498 /* instance of mdt_reint_rec */
2499 struct mdt_rec_create {
2507 __u32 cr_suppgid1_h;
2509 __u32 cr_suppgid2_h;
2510 struct lu_fid cr_fid1;
2511 struct lu_fid cr_fid2;
2512 struct lustre_handle cr_old_handle; /* handle in case of open replay */
2516 __u64 cr_padding_1; /* rr_blocks */
2519 /* use of helpers set/get_mrc_cr_flags() is needed to access
2520 * 64 bits cr_flags [cr_flags_l, cr_flags_h], this is done to
2521 * extend cr_flags size without breaking 1.8 compat */
2522 __u32 cr_flags_l; /* for use with open, low 32 bits */
2523 __u32 cr_flags_h; /* for use with open, high 32 bits */
2524 __u32 cr_umask; /* umask for create */
2525 __u32 cr_padding_4; /* rr_padding_4 */
2528 static inline void set_mrc_cr_flags(struct mdt_rec_create *mrc, __u64 flags)
2530 mrc->cr_flags_l = (__u32)(flags & 0xFFFFFFFFUll);
2531 mrc->cr_flags_h = (__u32)(flags >> 32);
2534 static inline __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
2536 return ((__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32));
2539 /* instance of mdt_reint_rec */
2540 struct mdt_rec_link {
2548 __u32 lk_suppgid1_h;
2550 __u32 lk_suppgid2_h;
2551 struct lu_fid lk_fid1;
2552 struct lu_fid lk_fid2;
2554 __u64 lk_padding_1; /* rr_atime */
2555 __u64 lk_padding_2; /* rr_ctime */
2556 __u64 lk_padding_3; /* rr_size */
2557 __u64 lk_padding_4; /* rr_blocks */
2559 __u32 lk_padding_5; /* rr_mode */
2560 __u32 lk_padding_6; /* rr_flags */
2561 __u32 lk_padding_7; /* rr_padding_2 */
2562 __u32 lk_padding_8; /* rr_padding_3 */
2563 __u32 lk_padding_9; /* rr_padding_4 */
2566 /* instance of mdt_reint_rec */
2567 struct mdt_rec_unlink {
2575 __u32 ul_suppgid1_h;
2577 __u32 ul_suppgid2_h;
2578 struct lu_fid ul_fid1;
2579 struct lu_fid ul_fid2;
2581 __u64 ul_padding_2; /* rr_atime */
2582 __u64 ul_padding_3; /* rr_ctime */
2583 __u64 ul_padding_4; /* rr_size */
2584 __u64 ul_padding_5; /* rr_blocks */
2587 __u32 ul_padding_6; /* rr_flags */
2588 __u32 ul_padding_7; /* rr_padding_2 */
2589 __u32 ul_padding_8; /* rr_padding_3 */
2590 __u32 ul_padding_9; /* rr_padding_4 */
2593 /* instance of mdt_reint_rec */
2594 struct mdt_rec_rename {
2602 __u32 rn_suppgid1_h;
2604 __u32 rn_suppgid2_h;
2605 struct lu_fid rn_fid1;
2606 struct lu_fid rn_fid2;
2608 __u64 rn_padding_1; /* rr_atime */
2609 __u64 rn_padding_2; /* rr_ctime */
2610 __u64 rn_padding_3; /* rr_size */
2611 __u64 rn_padding_4; /* rr_blocks */
2612 __u32 rn_bias; /* some operation flags */
2613 __u32 rn_mode; /* cross-ref rename has mode */
2614 __u32 rn_padding_5; /* rr_flags */
2615 __u32 rn_padding_6; /* rr_padding_2 */
2616 __u32 rn_padding_7; /* rr_padding_3 */
2617 __u32 rn_padding_8; /* rr_padding_4 */
2620 /* instance of mdt_reint_rec */
2621 struct mdt_rec_setxattr {
2629 __u32 sx_suppgid1_h;
2631 __u32 sx_suppgid2_h;
2632 struct lu_fid sx_fid;
2633 __u64 sx_padding_1; /* These three are rr_fid2 */
2638 __u64 sx_padding_5; /* rr_ctime */
2639 __u64 sx_padding_6; /* rr_size */
2640 __u64 sx_padding_7; /* rr_blocks */
2643 __u32 sx_padding_8; /* rr_flags */
2644 __u32 sx_padding_9; /* rr_padding_2 */
2645 __u32 sx_padding_10; /* rr_padding_3 */
2646 __u32 sx_padding_11; /* rr_padding_4 */
2650 * mdt_rec_reint is the template for all mdt_reint_xxx structures.
2651 * Do NOT change the size of various members, otherwise the value
2652 * will be broken in lustre_swab_mdt_rec_reint().
2654 * If you add new members in other mdt_reint_xxx structres and need to use the
2655 * rr_padding_x fields, then update lustre_swab_mdt_rec_reint() also.
2657 struct mdt_rec_reint {
2665 __u32 rr_suppgid1_h;
2667 __u32 rr_suppgid2_h;
2668 struct lu_fid rr_fid1;
2669 struct lu_fid rr_fid2;
2680 __u32 rr_padding_4; /* also fix lustre_swab_mdt_rec_reint */
2683 extern void lustre_swab_mdt_rec_reint(struct mdt_rec_reint *rr);
2685 /* lmv structures */
2687 __u32 ld_tgt_count; /* how many MDS's */
2688 __u32 ld_active_tgt_count; /* how many active */
2689 __u32 ld_default_stripe_count; /* how many objects are used */
2690 __u32 ld_pattern; /* default hash pattern */
2691 __u64 ld_default_hash_size;
2692 __u64 ld_padding_1; /* also fix lustre_swab_lmv_desc */
2693 __u32 ld_padding_2; /* also fix lustre_swab_lmv_desc */
2694 __u32 ld_qos_maxage; /* in second */
2695 __u32 ld_padding_3; /* also fix lustre_swab_lmv_desc */
2696 __u32 ld_padding_4; /* also fix lustre_swab_lmv_desc */
2697 struct obd_uuid ld_uuid;
2700 extern void lustre_swab_lmv_desc (struct lmv_desc *ld);
2702 /* lmv structures */
2703 #define LMV_MAGIC_V1 0x0CD10CD0 /* normal stripe lmv magic */
2704 #define LMV_USER_MAGIC 0x0CD20CD0 /* default lmv magic*/
2705 #define LMV_MAGIC LMV_MAGIC_V1
2707 enum lmv_hash_type {
2708 LMV_HASH_TYPE_ALL_CHARS = 1,
2709 LMV_HASH_TYPE_FNV_1A_64 = 2,
2712 #define LMV_HASH_NAME_ALL_CHARS "all_char"
2713 #define LMV_HASH_NAME_FNV_1A_64 "fnv_1a_64"
2716 * The FNV-1a hash algorithm is as follows:
2717 * hash = FNV_offset_basis
2718 * for each octet_of_data to be hashed
2719 * hash = hash XOR octet_of_data
2720 * hash = hash × FNV_prime
2722 * http://en.wikipedia.org/wiki/Fowler–Noll–Vo_hash_function#FNV-1a_hash
2724 * http://www.isthe.com/chongo/tech/comp/fnv/index.html#FNV-reference-source
2725 * FNV_prime is 2^40 + 2^8 + 0xb3 = 0x100000001b3ULL
2727 #define LUSTRE_FNV_1A_64_PRIME 0x100000001b3ULL
2728 #define LUSTRE_FNV_1A_64_OFFSET_BIAS 0xcbf29ce484222325ULL
2729 static inline __u64 lustre_hash_fnv_1a_64(const void *buf, size_t size)
2731 __u64 hash = LUSTRE_FNV_1A_64_OFFSET_BIAS;
2732 const unsigned char *p = buf;
2735 for (i = 0; i < size; i++) {
2737 hash *= LUSTRE_FNV_1A_64_PRIME;
2743 struct lmv_mds_md_v1 {
2745 __u32 lmv_stripe_count; /* stripe count */
2746 __u32 lmv_master_mdt_index; /* master MDT index */
2747 __u32 lmv_hash_type; /* dir stripe policy, i.e. indicate
2748 * which hash function to be used */
2749 __u32 lmv_layout_version; /* Used for directory restriping */
2751 char lmv_pool_name[LOV_MAXPOOLNAME]; /* pool name */
2752 struct lu_fid lmv_stripe_fids[0]; /* FIDs for each stripe */
2757 struct lmv_mds_md_v1 lmv_md_v1;
2758 struct lmv_user_md lmv_user_md;
2761 extern void lustre_swab_lmv_mds_md(union lmv_mds_md *lmm);
2763 static inline int lmv_mds_md_size(int stripe_count, unsigned int lmm_magic)
2765 switch (lmm_magic) {
2766 case LMV_MAGIC_V1: {
2767 struct lmv_mds_md_v1 *lmm1;
2769 return sizeof(*lmm1) + stripe_count *
2770 sizeof(lmm1->lmv_stripe_fids[0]);
2777 static inline int lmv_mds_md_stripe_count_get(const union lmv_mds_md *lmm)
2779 switch (le32_to_cpu(lmm->lmv_magic)) {
2781 return le32_to_cpu(lmm->lmv_md_v1.lmv_stripe_count);
2782 case LMV_USER_MAGIC:
2783 return le32_to_cpu(lmm->lmv_user_md.lum_stripe_count);
2789 static inline int lmv_mds_md_stripe_count_set(union lmv_mds_md *lmm,
2790 unsigned int stripe_count)
2792 switch (le32_to_cpu(lmm->lmv_magic)) {
2794 lmm->lmv_md_v1.lmv_stripe_count = cpu_to_le32(stripe_count);
2796 case LMV_USER_MAGIC:
2797 lmm->lmv_user_md.lum_stripe_count = cpu_to_le32(stripe_count);
2809 FLD_FIRST_OPC = FLD_QUERY
2815 SEQ_FIRST_OPC = SEQ_QUERY
2819 SEQ_ALLOC_SUPER = 0,
2831 LFSCK_NOTIFY = 1101,
2834 LFSCK_FIRST_OPC = LFSCK_NOTIFY
2838 * LOV data structures
2841 #define LOV_MAX_UUID_BUFFER_SIZE 8192
2842 /* The size of the buffer the lov/mdc reserves for the
2843 * array of UUIDs returned by the MDS. With the current
2844 * protocol, this will limit the max number of OSTs per LOV */
2846 #define LOV_DESC_MAGIC 0xB0CCDE5C
2848 /* LOV settings descriptor (should only contain static info) */
2850 __u32 ld_tgt_count; /* how many OBD's */
2851 __u32 ld_active_tgt_count; /* how many active */
2852 __u32 ld_default_stripe_count; /* how many objects are used */
2853 __u32 ld_pattern; /* default PATTERN_RAID0 */
2854 __u64 ld_default_stripe_size; /* in bytes */
2855 __u64 ld_default_stripe_offset; /* in bytes */
2856 __u32 ld_padding_0; /* unused */
2857 __u32 ld_qos_maxage; /* in second */
2858 __u32 ld_padding_1; /* also fix lustre_swab_lov_desc */
2859 __u32 ld_padding_2; /* also fix lustre_swab_lov_desc */
2860 struct obd_uuid ld_uuid;
2863 #define ld_magic ld_active_tgt_count /* for swabbing from llogs */
2865 extern void lustre_swab_lov_desc (struct lov_desc *ld);
2870 /* opcodes -- MUST be distinct from OST/MDS opcodes */
2875 LDLM_BL_CALLBACK = 104,
2876 LDLM_CP_CALLBACK = 105,
2877 LDLM_GL_CALLBACK = 106,
2878 LDLM_SET_INFO = 107,
2881 #define LDLM_FIRST_OPC LDLM_ENQUEUE
2883 #define RES_NAME_SIZE 4
2884 struct ldlm_res_id {
2885 __u64 name[RES_NAME_SIZE];
2888 #define DLDLMRES "["LPX64":"LPX64":"LPX64"]."LPX64i
2889 #define PLDLMRES(res) (res)->lr_name.name[0], (res)->lr_name.name[1], \
2890 (res)->lr_name.name[2], (res)->lr_name.name[3]
2892 extern void lustre_swab_ldlm_res_id (struct ldlm_res_id *id);
2894 static inline int ldlm_res_eq(const struct ldlm_res_id *res0,
2895 const struct ldlm_res_id *res1)
2897 return !memcmp(res0, res1, sizeof(*res0));
2914 #define LCK_MODE_NUM 8
2924 #define LDLM_MIN_TYPE LDLM_PLAIN
2926 struct ldlm_extent {
2932 static inline int ldlm_extent_overlap(struct ldlm_extent *ex1,
2933 struct ldlm_extent *ex2)
2935 return (ex1->start <= ex2->end) && (ex2->start <= ex1->end);
2938 /* check if @ex1 contains @ex2 */
2939 static inline int ldlm_extent_contain(struct ldlm_extent *ex1,
2940 struct ldlm_extent *ex2)
2942 return (ex1->start <= ex2->start) && (ex1->end >= ex2->end);
2945 struct ldlm_inodebits {
2949 struct ldlm_flock_wire {
2957 /* it's important that the fields of the ldlm_extent structure match
2958 * the first fields of the ldlm_flock structure because there is only
2959 * one ldlm_swab routine to process the ldlm_policy_data_t union. if
2960 * this ever changes we will need to swab the union differently based
2961 * on the resource type. */
2964 struct ldlm_extent l_extent;
2965 struct ldlm_flock_wire l_flock;
2966 struct ldlm_inodebits l_inodebits;
2967 } ldlm_wire_policy_data_t;
2969 extern void lustre_swab_ldlm_policy_data (ldlm_wire_policy_data_t *d);
2971 union ldlm_gl_desc {
2972 struct ldlm_gl_lquota_desc lquota_desc;
2975 extern void lustre_swab_gl_desc(union ldlm_gl_desc *);
2977 struct ldlm_intent {
2981 extern void lustre_swab_ldlm_intent (struct ldlm_intent *i);
2983 struct ldlm_resource_desc {
2984 ldlm_type_t lr_type;
2985 __u32 lr_padding; /* also fix lustre_swab_ldlm_resource_desc */
2986 struct ldlm_res_id lr_name;
2989 extern void lustre_swab_ldlm_resource_desc (struct ldlm_resource_desc *r);
2991 struct ldlm_lock_desc {
2992 struct ldlm_resource_desc l_resource;
2993 ldlm_mode_t l_req_mode;
2994 ldlm_mode_t l_granted_mode;
2995 ldlm_wire_policy_data_t l_policy_data;
2998 extern void lustre_swab_ldlm_lock_desc (struct ldlm_lock_desc *l);
3000 #define LDLM_LOCKREQ_HANDLES 2
3001 #define LDLM_ENQUEUE_CANCEL_OFF 1
3003 struct ldlm_request {
3006 struct ldlm_lock_desc lock_desc;
3007 struct lustre_handle lock_handle[LDLM_LOCKREQ_HANDLES];
3010 extern void lustre_swab_ldlm_request (struct ldlm_request *rq);
3012 /* If LDLM_ENQUEUE, 1 slot is already occupied, 1 is available.
3013 * Otherwise, 2 are available. */
3014 #define ldlm_request_bufsize(count,type) \
3016 int _avail = LDLM_LOCKREQ_HANDLES; \
3017 _avail -= (type == LDLM_ENQUEUE ? LDLM_ENQUEUE_CANCEL_OFF : 0); \
3018 sizeof(struct ldlm_request) + \
3019 (count > _avail ? count - _avail : 0) * \
3020 sizeof(struct lustre_handle); \
3025 __u32 lock_padding; /* also fix lustre_swab_ldlm_reply */
3026 struct ldlm_lock_desc lock_desc;
3027 struct lustre_handle lock_handle;
3028 __u64 lock_policy_res1;
3029 __u64 lock_policy_res2;
3032 extern void lustre_swab_ldlm_reply (struct ldlm_reply *r);
3034 #define ldlm_flags_to_wire(flags) ((__u32)(flags))
3035 #define ldlm_flags_from_wire(flags) ((__u64)(flags))
3038 * Opcodes for mountconf (mgs and mgc)
3043 MGS_EXCEPTION, /* node died, etc. */
3044 MGS_TARGET_REG, /* whenever target starts up */
3050 #define MGS_FIRST_OPC MGS_CONNECT
3052 #define MGS_PARAM_MAXLEN 1024
3053 #define KEY_SET_INFO "set_info"
3055 struct mgs_send_param {
3056 char mgs_param[MGS_PARAM_MAXLEN];
3059 /* We pass this info to the MGS so it can write config logs */
3060 #define MTI_NAME_MAXLEN 64
3061 #define MTI_PARAM_MAXLEN 4096
3062 #define MTI_NIDS_MAX 32
3063 struct mgs_target_info {
3064 __u32 mti_lustre_ver;
3065 __u32 mti_stripe_index;
3066 __u32 mti_config_ver;
3068 __u32 mti_nid_count;
3069 __u32 mti_instance; /* Running instance of target */
3070 char mti_fsname[MTI_NAME_MAXLEN];
3071 char mti_svname[MTI_NAME_MAXLEN];
3072 char mti_uuid[sizeof(struct obd_uuid)];
3073 __u64 mti_nids[MTI_NIDS_MAX]; /* host nids (lnet_nid_t)*/
3074 char mti_params[MTI_PARAM_MAXLEN];
3076 extern void lustre_swab_mgs_target_info(struct mgs_target_info *oinfo);
3078 struct mgs_nidtbl_entry {
3079 __u64 mne_version; /* table version of this entry */
3080 __u32 mne_instance; /* target instance # */
3081 __u32 mne_index; /* target index */
3082 __u32 mne_length; /* length of this entry - by bytes */
3083 __u8 mne_type; /* target type LDD_F_SV_TYPE_OST/MDT */
3084 __u8 mne_nid_type; /* type of nid(mbz). for ipv6. */
3085 __u8 mne_nid_size; /* size of each NID, by bytes */
3086 __u8 mne_nid_count; /* # of NIDs in buffer */
3088 lnet_nid_t nids[0]; /* variable size buffer for NIDs. */
3091 extern void lustre_swab_mgs_nidtbl_entry(struct mgs_nidtbl_entry *oinfo);
3093 struct mgs_config_body {
3094 char mcb_name[MTI_NAME_MAXLEN]; /* logname */
3095 __u64 mcb_offset; /* next index of config log to request */
3096 __u16 mcb_type; /* type of log: CONFIG_T_[CONFIG|RECOVER] */
3098 __u8 mcb_bits; /* bits unit size of config log */
3099 __u32 mcb_units; /* # of units for bulk transfer */
3101 extern void lustre_swab_mgs_config_body(struct mgs_config_body *body);
3103 struct mgs_config_res {
3104 __u64 mcr_offset; /* index of last config log */
3105 __u64 mcr_size; /* size of the log */
3107 extern void lustre_swab_mgs_config_res(struct mgs_config_res *body);
3109 /* Config marker flags (in config log) */
3110 #define CM_START 0x01
3112 #define CM_SKIP 0x04
3113 #define CM_UPGRADE146 0x08
3114 #define CM_EXCLUDE 0x10
3115 #define CM_START_SKIP (CM_START | CM_SKIP)
3118 __u32 cm_step; /* aka config version */
3120 __u32 cm_vers; /* lustre release version number */
3121 __u32 cm_padding; /* 64 bit align */
3122 obd_time cm_createtime; /*when this record was first created */
3123 obd_time cm_canceltime; /*when this record is no longer valid*/
3124 char cm_tgtname[MTI_NAME_MAXLEN];
3125 char cm_comment[MTI_NAME_MAXLEN];
3128 extern void lustre_swab_cfg_marker(struct cfg_marker *marker,
3129 int swab, int size);
3132 * Opcodes for multiple servers.
3142 #define OBD_FIRST_OPC OBD_PING
3144 /* catalog of log objects */
3146 /** Identifier for a single log object */
3148 struct ost_id lgl_oi;
3150 } __attribute__((packed));
3152 /** Records written to the CATALOGS list */
3153 #define CATLIST "CATALOGS"
3155 struct llog_logid lci_logid;
3159 } __attribute__((packed));
3161 /* Log data record types - there is no specific reason that these need to
3162 * be related to the RPC opcodes, but no reason not to (may be handy later?)
3164 #define LLOG_OP_MAGIC 0x10600000
3165 #define LLOG_OP_MASK 0xfff00000
3168 LLOG_PAD_MAGIC = LLOG_OP_MAGIC | 0x00000,
3169 OST_SZ_REC = LLOG_OP_MAGIC | 0x00f00,
3170 /* OST_RAID1_REC = LLOG_OP_MAGIC | 0x01000, never used */
3171 MDS_UNLINK_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) |
3172 REINT_UNLINK, /* obsolete after 2.5.0 */
3173 MDS_UNLINK64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
3175 /* MDS_SETATTR_REC = LLOG_OP_MAGIC | 0x12401, obsolete 1.8.0 */
3176 MDS_SETATTR64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
3178 OBD_CFG_REC = LLOG_OP_MAGIC | 0x20000,
3179 /* PTL_CFG_REC = LLOG_OP_MAGIC | 0x30000, obsolete 1.4.0 */
3180 LLOG_GEN_REC = LLOG_OP_MAGIC | 0x40000,
3181 /* LLOG_JOIN_REC = LLOG_OP_MAGIC | 0x50000, obsolete 1.8.0 */
3182 CHANGELOG_REC = LLOG_OP_MAGIC | 0x60000,
3183 CHANGELOG_USER_REC = LLOG_OP_MAGIC | 0x70000,
3184 HSM_AGENT_REC = LLOG_OP_MAGIC | 0x80000,
3185 LLOG_HDR_MAGIC = LLOG_OP_MAGIC | 0x45539,
3186 LLOG_LOGID_MAGIC = LLOG_OP_MAGIC | 0x4553b,
3189 #define LLOG_REC_HDR_NEEDS_SWABBING(r) \
3190 (((r)->lrh_type & __swab32(LLOG_OP_MASK)) == __swab32(LLOG_OP_MAGIC))
3192 /** Log record header - stored in little endian order.
3193 * Each record must start with this struct, end with a llog_rec_tail,
3194 * and be a multiple of 256 bits in size.
3196 struct llog_rec_hdr {
3203 struct llog_rec_tail {
3208 /* Where data follow just after header */
3209 #define REC_DATA(ptr) \
3210 ((void *)((char *)ptr + sizeof(struct llog_rec_hdr)))
3212 #define REC_DATA_LEN(rec) \
3213 (rec->lrh_len - sizeof(struct llog_rec_hdr) - \
3214 sizeof(struct llog_rec_tail))
3216 struct llog_logid_rec {
3217 struct llog_rec_hdr lid_hdr;
3218 struct llog_logid lid_id;
3222 struct llog_rec_tail lid_tail;
3223 } __attribute__((packed));
3225 struct llog_unlink_rec {
3226 struct llog_rec_hdr lur_hdr;
3229 obd_count lur_count;
3230 struct llog_rec_tail lur_tail;
3231 } __attribute__((packed));
3233 struct llog_unlink64_rec {
3234 struct llog_rec_hdr lur_hdr;
3235 struct lu_fid lur_fid;
3236 obd_count lur_count; /* to destroy the lost precreated */
3240 struct llog_rec_tail lur_tail;
3241 } __attribute__((packed));
3243 struct llog_setattr64_rec {
3244 struct llog_rec_hdr lsr_hdr;
3245 struct ost_id lsr_oi;
3251 struct llog_rec_tail lsr_tail;
3252 } __attribute__((packed));
3254 struct llog_size_change_rec {
3255 struct llog_rec_hdr lsc_hdr;
3256 struct ll_fid lsc_fid;
3261 struct llog_rec_tail lsc_tail;
3262 } __attribute__((packed));
3264 #define CHANGELOG_MAGIC 0xca103000
3266 /** \a changelog_rec_type's that can't be masked */
3267 #define CHANGELOG_MINMASK (1 << CL_MARK)
3268 /** bits covering all \a changelog_rec_type's */
3269 #define CHANGELOG_ALLMASK 0XFFFFFFFF
3270 /** default \a changelog_rec_type mask */
3271 #define CHANGELOG_DEFMASK CHANGELOG_ALLMASK & ~(1 << CL_ATIME | 1 << CL_CLOSE)
3273 /* changelog llog name, needed by client replicators */
3274 #define CHANGELOG_CATALOG "changelog_catalog"
3276 struct changelog_setinfo {
3279 } __attribute__((packed));
3281 /** changelog record */
3282 struct llog_changelog_rec {
3283 struct llog_rec_hdr cr_hdr;
3284 struct changelog_rec cr;
3285 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
3286 } __attribute__((packed));
3288 struct llog_changelog_ext_rec {
3289 struct llog_rec_hdr cr_hdr;
3290 struct changelog_ext_rec cr;
3291 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
3292 } __attribute__((packed));
3294 #define CHANGELOG_USER_PREFIX "cl"
3296 struct llog_changelog_user_rec {
3297 struct llog_rec_hdr cur_hdr;
3301 struct llog_rec_tail cur_tail;
3302 } __attribute__((packed));
3304 enum agent_req_status {
3312 static inline char *agent_req_status2name(enum agent_req_status ars)
3330 static inline bool agent_req_in_final_state(enum agent_req_status ars)
3332 return ((ars == ARS_SUCCEED) || (ars == ARS_FAILED) ||
3333 (ars == ARS_CANCELED));
3336 struct llog_agent_req_rec {
3337 struct llog_rec_hdr arr_hdr; /**< record header */
3338 __u32 arr_status; /**< status of the request */
3340 * agent_req_status */
3341 __u32 arr_archive_id; /**< backend archive number */
3342 __u64 arr_flags; /**< req flags */
3343 __u64 arr_compound_id; /**< compound cookie */
3344 __u64 arr_req_create; /**< req. creation time */
3345 __u64 arr_req_change; /**< req. status change time */
3346 struct hsm_action_item arr_hai; /**< req. to the agent */
3347 struct llog_rec_tail arr_tail; /**< record tail for_sizezof_only */
3348 } __attribute__((packed));
3350 /* Old llog gen for compatibility */
3354 } __attribute__((packed));
3356 struct llog_gen_rec {
3357 struct llog_rec_hdr lgr_hdr;
3358 struct llog_gen lgr_gen;
3362 struct llog_rec_tail lgr_tail;
3365 /* On-disk header structure of each log object, stored in little endian order */
3366 #define LLOG_CHUNK_SIZE 8192
3367 #define LLOG_HEADER_SIZE (96)
3368 #define LLOG_BITMAP_BYTES (LLOG_CHUNK_SIZE - LLOG_HEADER_SIZE)
3370 #define LLOG_MIN_REC_SIZE (24) /* round(llog_rec_hdr + llog_rec_tail) */
3372 /* flags for the logs */
3374 LLOG_F_ZAP_WHEN_EMPTY = 0x1,
3375 LLOG_F_IS_CAT = 0x2,
3376 LLOG_F_IS_PLAIN = 0x4,
3379 struct llog_log_hdr {
3380 struct llog_rec_hdr llh_hdr;
3381 obd_time llh_timestamp;
3383 __u32 llh_bitmap_offset;
3387 /* for a catalog the first plain slot is next to it */
3388 struct obd_uuid llh_tgtuuid;
3389 __u32 llh_reserved[LLOG_HEADER_SIZE/sizeof(__u32) - 23];
3390 __u32 llh_bitmap[LLOG_BITMAP_BYTES/sizeof(__u32)];
3391 struct llog_rec_tail llh_tail;
3392 } __attribute__((packed));
3394 #define LLOG_BITMAP_SIZE(llh) (__u32)((llh->llh_hdr.lrh_len - \
3395 llh->llh_bitmap_offset - \
3396 sizeof(llh->llh_tail)) * 8)
3398 /** log cookies are used to reference a specific log file and a record therein */
3399 struct llog_cookie {
3400 struct llog_logid lgc_lgl;
3404 } __attribute__((packed));
3406 /** llog protocol */
3407 enum llogd_rpc_ops {
3408 LLOG_ORIGIN_HANDLE_CREATE = 501,
3409 LLOG_ORIGIN_HANDLE_NEXT_BLOCK = 502,
3410 LLOG_ORIGIN_HANDLE_READ_HEADER = 503,
3411 LLOG_ORIGIN_HANDLE_WRITE_REC = 504,
3412 LLOG_ORIGIN_HANDLE_CLOSE = 505,
3413 LLOG_ORIGIN_CONNECT = 506,
3414 LLOG_CATINFO = 507, /* deprecated */
3415 LLOG_ORIGIN_HANDLE_PREV_BLOCK = 508,
3416 LLOG_ORIGIN_HANDLE_DESTROY = 509, /* for destroy llog object*/
3418 LLOG_FIRST_OPC = LLOG_ORIGIN_HANDLE_CREATE
3422 struct llog_logid lgd_logid;
3424 __u32 lgd_llh_flags;
3426 __u32 lgd_saved_index;
3428 __u64 lgd_cur_offset;
3429 } __attribute__((packed));
3431 struct llogd_conn_body {
3432 struct llog_gen lgdc_gen;
3433 struct llog_logid lgdc_logid;
3434 __u32 lgdc_ctxt_idx;
3435 } __attribute__((packed));
3437 /* Note: 64-bit types are 64-bit aligned in structure */
3439 obd_valid o_valid; /* hot fields in this obdo */
3441 obd_id o_parent_seq;
3442 obd_size o_size; /* o_size-o_blocks == ost_lvb */
3446 obd_blocks o_blocks; /* brw: cli sent cached bytes */
3449 /* 32-bit fields start here: keep an even number of them via padding */
3450 obd_blksize o_blksize; /* optimal IO blocksize */
3451 obd_mode o_mode; /* brw: cli sent cache remain */
3455 obd_count o_nlink; /* brw: checksum */
3456 obd_count o_parent_oid;
3457 obd_count o_misc; /* brw: o_dropped */
3459 __u64 o_ioepoch; /* epoch in ost writes */
3460 __u32 o_stripe_idx; /* holds stripe idx */
3462 struct lustre_handle o_handle; /* brw: lock handle to prolong
3464 struct llog_cookie o_lcookie; /* destroy: unlink cookie from
3469 __u64 o_data_version; /* getattr: sum of iversion for
3471 * brw: grant space consumed on
3472 * the client for the write */
3478 #define o_dirty o_blocks
3479 #define o_undirty o_mode
3480 #define o_dropped o_misc
3481 #define o_cksum o_nlink
3482 #define o_grant_used o_data_version
3484 struct lfsck_request {
3496 __u16 lr_async_windows;
3498 /* lr_fid is used on server-side only, and can be
3499 * reused as others by client in the future. */
3500 struct lu_fid lr_fid;
3505 void lustre_swab_lfsck_request(struct lfsck_request *lr);
3507 struct lfsck_reply {
3513 void lustre_swab_lfsck_reply(struct lfsck_reply *lr);
3515 static inline void lustre_set_wire_obdo(struct obd_connect_data *ocd,
3517 const struct obdo *lobdo)
3520 wobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
3524 if (unlikely(!(ocd->ocd_connect_flags & OBD_CONNECT_FID)) &&
3525 fid_seq_is_echo(ostid_seq(&lobdo->o_oi))) {
3526 /* Currently OBD_FL_OSTID will only be used when 2.4 echo
3527 * client communicate with pre-2.4 server */
3528 wobdo->o_oi.oi.oi_id = fid_oid(&lobdo->o_oi.oi_fid);
3529 wobdo->o_oi.oi.oi_seq = fid_seq(&lobdo->o_oi.oi_fid);
3533 static inline void lustre_get_wire_obdo(struct obd_connect_data *ocd,
3535 const struct obdo *wobdo)
3537 obd_flag local_flags = 0;
3539 if (lobdo->o_valid & OBD_MD_FLFLAGS)
3540 local_flags = lobdo->o_flags & OBD_FL_LOCAL_MASK;
3543 if (local_flags != 0) {
3544 lobdo->o_valid |= OBD_MD_FLFLAGS;
3545 lobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
3546 lobdo->o_flags |= local_flags;
3551 if (unlikely(!(ocd->ocd_connect_flags & OBD_CONNECT_FID)) &&
3552 fid_seq_is_echo(wobdo->o_oi.oi.oi_seq)) {
3554 lobdo->o_oi.oi_fid.f_seq = wobdo->o_oi.oi.oi_seq;
3555 lobdo->o_oi.oi_fid.f_oid = wobdo->o_oi.oi.oi_id;
3556 lobdo->o_oi.oi_fid.f_ver = 0;
3560 extern void lustre_swab_obdo (struct obdo *o);
3562 /* request structure for OST's */
3567 /* Key for FIEMAP to be used in get_info calls */
3568 struct ll_fiemap_info_key {
3571 struct ll_user_fiemap fiemap;
3574 extern void lustre_swab_ost_body (struct ost_body *b);
3575 extern void lustre_swab_ost_last_id(obd_id *id);
3576 extern void lustre_swab_fiemap(struct ll_user_fiemap *fiemap);
3578 extern void lustre_swab_lov_user_md_v1(struct lov_user_md_v1 *lum);
3579 extern void lustre_swab_lov_user_md_v3(struct lov_user_md_v3 *lum);
3580 extern void lustre_swab_lov_user_md_objects(struct lov_user_ost_data *lod,
3582 extern void lustre_swab_lov_mds_md(struct lov_mds_md *lmm);
3585 extern void lustre_swab_llogd_body (struct llogd_body *d);
3586 extern void lustre_swab_llog_hdr (struct llog_log_hdr *h);
3587 extern void lustre_swab_llogd_conn_body (struct llogd_conn_body *d);
3588 extern void lustre_swab_llog_rec(struct llog_rec_hdr *rec);
3589 extern void lustre_swab_llog_id(struct llog_logid *lid);
3592 extern void lustre_swab_lustre_cfg(struct lustre_cfg *lcfg);
3594 /* Functions for dumping PTLRPC fields */
3595 void dump_rniobuf(struct niobuf_remote *rnb);
3596 void dump_ioo(struct obd_ioobj *nb);
3597 void dump_obdo(struct obdo *oa);
3598 void dump_ost_body(struct ost_body *ob);
3599 void dump_rcs(__u32 *rc);
3601 #define IDX_INFO_MAGIC 0x3D37CC37
3603 /* Index file transfer through the network. The server serializes the index into
3604 * a byte stream which is sent to the client via a bulk transfer */
3608 /* reply: see idx_info_flags below */
3611 /* request & reply: number of lu_idxpage (to be) transferred */
3615 /* request: requested attributes passed down to the iterator API */
3618 /* request & reply: index file identifier (FID) */
3619 struct lu_fid ii_fid;
3621 /* reply: version of the index file before starting to walk the index.
3622 * Please note that the version can be modified at any time during the
3626 /* request: hash to start with:
3627 * reply: hash of the first entry of the first lu_idxpage and hash
3628 * of the entry to read next if any */
3629 __u64 ii_hash_start;
3632 /* reply: size of keys in lu_idxpages, minimal one if II_FL_VARKEY is
3636 /* reply: size of records in lu_idxpages, minimal one if II_FL_VARREC
3644 extern void lustre_swab_idx_info(struct idx_info *ii);
3646 #define II_END_OFF MDS_DIR_END_OFF /* all entries have been read */
3648 /* List of flags used in idx_info::ii_flags */
3649 enum idx_info_flags {
3650 II_FL_NOHASH = 1 << 0, /* client doesn't care about hash value */
3651 II_FL_VARKEY = 1 << 1, /* keys can be of variable size */
3652 II_FL_VARREC = 1 << 2, /* records can be of variable size */
3653 II_FL_NONUNQ = 1 << 3, /* index supports non-unique keys */
3656 #define LIP_MAGIC 0x8A6D6B6C
3658 /* 4KB (= LU_PAGE_SIZE) container gathering key/record pairs */
3660 /* 16-byte header */
3663 __u16 lip_nr; /* number of entries in the container */
3664 __u64 lip_pad0; /* additional padding for future use */
3666 /* key/record pairs are stored in the remaining 4080 bytes.
3667 * depending upon the flags in idx_info::ii_flags, each key/record
3668 * pair might be preceded by:
3670 * - the key size (II_FL_VARKEY is set)
3671 * - the record size (II_FL_VARREC is set)
3673 * For the time being, we only support fixed-size key & record. */
3674 char lip_entries[0];
3676 extern void lustre_swab_lip_header(struct lu_idxpage *lip);
3678 #define LIP_HDR_SIZE (offsetof(struct lu_idxpage, lip_entries))
3680 /* Gather all possible type associated with a 4KB container */
3682 struct lu_dirpage lp_dir; /* for MDS_READPAGE */
3683 struct lu_idxpage lp_idx; /* for OBD_IDX_READ */
3684 char lp_array[LU_PAGE_SIZE];
3687 /* security opcodes */
3690 SEC_CTX_INIT_CONT = 802,
3693 SEC_FIRST_OPC = SEC_CTX_INIT
3697 * capa related definitions
3699 #define CAPA_HMAC_MAX_LEN 64
3700 #define CAPA_HMAC_KEY_MAX_LEN 56
3702 /* NB take care when changing the sequence of elements this struct,
3703 * because the offset info is used in find_capa() */
3704 struct lustre_capa {
3705 struct lu_fid lc_fid; /** fid */
3706 __u64 lc_opc; /** operations allowed */
3707 __u64 lc_uid; /** file owner */
3708 __u64 lc_gid; /** file group */
3709 __u32 lc_flags; /** HMAC algorithm & flags */
3710 __u32 lc_keyid; /** key# used for the capability */
3711 __u32 lc_timeout; /** capa timeout value (sec) */
3712 __u32 lc_expiry; /** expiry time (sec) */
3713 __u8 lc_hmac[CAPA_HMAC_MAX_LEN]; /** HMAC */
3714 } __attribute__((packed));
3716 extern void lustre_swab_lustre_capa(struct lustre_capa *c);
3718 /** lustre_capa::lc_opc */
3720 CAPA_OPC_BODY_WRITE = 1<<0, /**< write object data */
3721 CAPA_OPC_BODY_READ = 1<<1, /**< read object data */
3722 CAPA_OPC_INDEX_LOOKUP = 1<<2, /**< lookup object fid */
3723 CAPA_OPC_INDEX_INSERT = 1<<3, /**< insert object fid */
3724 CAPA_OPC_INDEX_DELETE = 1<<4, /**< delete object fid */
3725 CAPA_OPC_OSS_WRITE = 1<<5, /**< write oss object data */
3726 CAPA_OPC_OSS_READ = 1<<6, /**< read oss object data */
3727 CAPA_OPC_OSS_TRUNC = 1<<7, /**< truncate oss object */
3728 CAPA_OPC_OSS_DESTROY = 1<<8, /**< destroy oss object */
3729 CAPA_OPC_META_WRITE = 1<<9, /**< write object meta data */
3730 CAPA_OPC_META_READ = 1<<10, /**< read object meta data */
3733 #define CAPA_OPC_OSS_RW (CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE)
3734 #define CAPA_OPC_MDS_ONLY \
3735 (CAPA_OPC_BODY_WRITE | CAPA_OPC_BODY_READ | CAPA_OPC_INDEX_LOOKUP | \
3736 CAPA_OPC_INDEX_INSERT | CAPA_OPC_INDEX_DELETE)
3737 #define CAPA_OPC_OSS_ONLY \
3738 (CAPA_OPC_OSS_WRITE | CAPA_OPC_OSS_READ | CAPA_OPC_OSS_TRUNC | \
3739 CAPA_OPC_OSS_DESTROY)
3740 #define CAPA_OPC_MDS_DEFAULT ~CAPA_OPC_OSS_ONLY
3741 #define CAPA_OPC_OSS_DEFAULT ~(CAPA_OPC_MDS_ONLY | CAPA_OPC_OSS_ONLY)
3743 /* MDS capability covers object capability for operations of body r/w
3744 * (dir readpage/sendpage), index lookup/insert/delete and meta data r/w,
3745 * while OSS capability only covers object capability for operations of
3746 * oss data(file content) r/w/truncate.
3748 static inline int capa_for_mds(struct lustre_capa *c)
3750 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) != 0;
3753 static inline int capa_for_oss(struct lustre_capa *c)
3755 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) == 0;
3758 /* lustre_capa::lc_hmac_alg */
3760 CAPA_HMAC_ALG_SHA1 = 1, /**< sha1 algorithm */
3764 #define CAPA_FL_MASK 0x00ffffff
3765 #define CAPA_HMAC_ALG_MASK 0xff000000
3767 struct lustre_capa_key {
3768 __u64 lk_seq; /**< mds# */
3769 __u32 lk_keyid; /**< key# */
3771 __u8 lk_key[CAPA_HMAC_KEY_MAX_LEN]; /**< key */
3772 } __attribute__((packed));
3774 extern void lustre_swab_lustre_capa_key(struct lustre_capa_key *k);
3776 /** The link ea holds 1 \a link_ea_entry for each hardlink */
3777 #define LINK_EA_MAGIC 0x11EAF1DFUL
3778 struct link_ea_header {
3781 __u64 leh_len; /* total size */
3787 /** Hardlink data is name and parent fid.
3788 * Stored in this crazy struct for maximum packing and endian-neutrality
3790 struct link_ea_entry {
3791 /** __u16 stored big-endian, unaligned */
3792 unsigned char lee_reclen[2];
3793 unsigned char lee_parent_fid[sizeof(struct lu_fid)];
3795 }__attribute__((packed));
3797 /** fid2path request/reply structure */
3798 struct getinfo_fid2path {
3799 struct lu_fid gf_fid;
3804 } __attribute__((packed));
3806 void lustre_swab_fid2path (struct getinfo_fid2path *gf);
3809 LAYOUT_INTENT_ACCESS = 0,
3810 LAYOUT_INTENT_READ = 1,
3811 LAYOUT_INTENT_WRITE = 2,
3812 LAYOUT_INTENT_GLIMPSE = 3,
3813 LAYOUT_INTENT_TRUNC = 4,
3814 LAYOUT_INTENT_RELEASE = 5,
3815 LAYOUT_INTENT_RESTORE = 6
3818 /* enqueue layout lock with intent */
3819 struct layout_intent {
3820 __u32 li_opc; /* intent operation for enqueue, read, write etc */
3826 void lustre_swab_layout_intent(struct layout_intent *li);
3829 * On the wire version of hsm_progress structure.
3831 * Contains the userspace hsm_progress and some internal fields.
3833 struct hsm_progress_kernel {
3834 /* Field taken from struct hsm_progress */
3837 struct hsm_extent hpk_extent;
3839 __u16 hpk_errval; /* positive val */
3841 /* Additional fields */
3842 __u64 hpk_data_version;
3844 } __attribute__((packed));
3846 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
3847 extern void lustre_swab_hsm_current_action(struct hsm_current_action *action);
3848 extern void lustre_swab_hsm_progress_kernel(struct hsm_progress_kernel *hpk);
3849 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
3850 extern void lustre_swab_hsm_user_item(struct hsm_user_item *hui);
3851 extern void lustre_swab_hsm_request(struct hsm_request *hr);
3854 * These are object update opcode under UPDATE_OBJ, which is currently
3855 * being used by cross-ref operations between MDT.
3857 * During the cross-ref operation, the Master MDT, which the client send the
3858 * request to, will disassembly the operation into object updates, then OSP
3859 * will send these updates to the remote MDT to be executed.
3861 * Update request format
3862 * magic: UPDATE_BUFFER_MAGIC_V1
3863 * Count: How many updates in the req.
3864 * bufs[0] : following are packets of object.
3866 * type: object_update_op, the op code of update
3867 * fid: The object fid of the update.
3868 * lens/bufs: other parameters of the update.
3870 * type: object_update_op, the op code of update
3871 * fid: The object fid of the update.
3872 * lens/bufs: other parameters of the update.
3874 * update[7]: type: object_update_op, the op code of update
3875 * fid: The object fid of the update.
3876 * lens/bufs: other parameters of the update.
3877 * Current 8 maxim updates per object update request.
3879 *******************************************************************
3880 * update reply format:
3882 * ur_version: UPDATE_REPLY_V1
3883 * ur_count: The count of the reply, which is usually equal
3884 * to the number of updates in the request.
3885 * ur_lens: The reply lengths of each object update.
3887 * replies: 1st update reply [4bytes_ret: other body]
3888 * 2nd update reply [4bytes_ret: other body]
3890 * nth update reply [4bytes_ret: other body]
3892 * For each reply of the update, the format would be
3893 * result(4 bytes):Other stuff
3896 #define UPDATE_MAX_OPS 10
3897 #define UPDATE_BUFFER_MAGIC_V1 0xBDDE0001
3898 #define UPDATE_BUFFER_MAGIC UPDATE_BUFFER_MAGIC_V1
3899 #define UPDATE_BUF_COUNT 8
3900 enum object_update_op {
3909 OBJ_INDEX_LOOKUP = 9,
3910 OBJ_INDEX_INSERT = 10,
3911 OBJ_INDEX_DELETE = 11,
3918 struct lu_fid u_fid;
3919 __u32 u_lens[UPDATE_BUF_COUNT];
3929 #define UPDATE_REPLY_V1 0x00BD0001
3930 struct update_reply {
3936 void lustre_swab_update_buf(struct update_buf *ub);
3937 void lustre_swab_update_reply_buf(struct update_reply *ur);
3939 /** layout swap request structure
3940 * fid1 and fid2 are in mdt_body
3942 struct mdc_swap_layouts {
3946 void lustre_swab_swap_layouts(struct mdc_swap_layouts *msl);
3949 struct lustre_handle cd_handle;
3950 struct lu_fid cd_fid;
3951 __u64 cd_data_version;
3952 __u64 cd_reserved[8];
3955 void lustre_swab_close_data(struct close_data *data);