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
145 /* Portal 63 is reserved for the Cray Inc DVS - nic@cray.com, roe@cray.com, n8851@cray.com */
148 #define PTL_RPC_MSG_REQUEST 4711
149 #define PTL_RPC_MSG_ERR 4712
150 #define PTL_RPC_MSG_REPLY 4713
152 /* DON'T use swabbed values of MAGIC as magic! */
153 #define LUSTRE_MSG_MAGIC_V1 0x0BD00BD0
154 #define LUSTRE_MSG_MAGIC_V2 0x0BD00BD3
156 #define LUSTRE_MSG_MAGIC_V1_SWABBED 0xD00BD00B
157 #define LUSTRE_MSG_MAGIC_V2_SWABBED 0xD30BD00B
159 #define LUSTRE_MSG_MAGIC LUSTRE_MSG_MAGIC_V2
161 #define PTLRPC_MSG_VERSION 0x00000003
162 #define LUSTRE_VERSION_MASK 0xffff0000
163 #define LUSTRE_OBD_VERSION 0x00010000
164 #define LUSTRE_MDS_VERSION 0x00020000
165 #define LUSTRE_OST_VERSION 0x00030000
166 #define LUSTRE_DLM_VERSION 0x00040000
167 #define LUSTRE_LOG_VERSION 0x00050000
168 #define LUSTRE_MGS_VERSION 0x00060000
170 typedef __u32 mdsno_t;
171 typedef __u64 seqno_t;
172 typedef __u64 obd_id;
173 typedef __u64 obd_seq;
174 typedef __s64 obd_time;
175 typedef __u64 obd_size;
176 typedef __u64 obd_off;
177 typedef __u64 obd_blocks;
178 typedef __u64 obd_valid;
179 typedef __u32 obd_blksize;
180 typedef __u32 obd_mode;
181 typedef __u32 obd_uid;
182 typedef __u32 obd_gid;
183 typedef __u32 obd_flag;
184 typedef __u32 obd_count;
187 * Describes a range of sequence, lsr_start is included but lsr_end is
189 * Same structure is used in fld module where lsr_index field holds mdt id
192 struct lu_seq_range {
199 struct lu_seq_range_array {
202 struct lu_seq_range lsra_lsr[0];
205 #define LU_SEQ_RANGE_MDT 0x0
206 #define LU_SEQ_RANGE_OST 0x1
207 #define LU_SEQ_RANGE_ANY 0x3
209 #define LU_SEQ_RANGE_MASK 0x3
211 static inline unsigned fld_range_type(const struct lu_seq_range *range)
213 return range->lsr_flags & LU_SEQ_RANGE_MASK;
216 static inline int fld_range_is_ost(const struct lu_seq_range *range)
218 return fld_range_type(range) == LU_SEQ_RANGE_OST;
221 static inline int fld_range_is_mdt(const struct lu_seq_range *range)
223 return fld_range_type(range) == LU_SEQ_RANGE_MDT;
227 * This all range is only being used when fld client sends fld query request,
228 * but it does not know whether the seq is MDT or OST, so it will send req
229 * with ALL type, which means either seq type gotten from lookup can be
232 static inline unsigned fld_range_is_any(const struct lu_seq_range *range)
234 return fld_range_type(range) == LU_SEQ_RANGE_ANY;
237 static inline void fld_range_set_type(struct lu_seq_range *range,
240 range->lsr_flags |= flags;
243 static inline void fld_range_set_mdt(struct lu_seq_range *range)
245 fld_range_set_type(range, LU_SEQ_RANGE_MDT);
248 static inline void fld_range_set_ost(struct lu_seq_range *range)
250 fld_range_set_type(range, LU_SEQ_RANGE_OST);
253 static inline void fld_range_set_any(struct lu_seq_range *range)
255 fld_range_set_type(range, LU_SEQ_RANGE_ANY);
259 * returns width of given range \a r
262 static inline __u64 range_space(const struct lu_seq_range *range)
264 return range->lsr_end - range->lsr_start;
268 * initialize range to zero
271 static inline void range_init(struct lu_seq_range *range)
273 memset(range, 0, sizeof(*range));
277 * check if given seq id \a s is within given range \a r
280 static inline int range_within(const struct lu_seq_range *range,
283 return s >= range->lsr_start && s < range->lsr_end;
286 static inline int range_is_sane(const struct lu_seq_range *range)
288 return (range->lsr_end >= range->lsr_start);
291 static inline int range_is_zero(const struct lu_seq_range *range)
293 return (range->lsr_start == 0 && range->lsr_end == 0);
296 static inline int range_is_exhausted(const struct lu_seq_range *range)
299 return range_space(range) == 0;
302 /* return 0 if two range have the same location */
303 static inline int range_compare_loc(const struct lu_seq_range *r1,
304 const struct lu_seq_range *r2)
306 return r1->lsr_index != r2->lsr_index ||
307 r1->lsr_flags != r2->lsr_flags;
310 #define DRANGE "[%#16.16"LPF64"x-%#16.16"LPF64"x):%x:%s"
312 #define PRANGE(range) \
313 (range)->lsr_start, \
315 (range)->lsr_index, \
316 fld_range_is_mdt(range) ? "mdt" : "ost"
319 /** \defgroup lu_fid lu_fid
323 * Flags for lustre_mdt_attrs::lma_compat and lustre_mdt_attrs::lma_incompat.
324 * Deprecated since HSM and SOM attributes are now stored in separate on-disk
328 LMAC_HSM = 0x00000001,
329 LMAC_SOM = 0x00000002,
330 LMAC_NOT_IN_OI = 0x00000004, /* the object does NOT need OI mapping */
331 LMAC_FID_ON_OST = 0x00000008, /* For OST-object, its OI mapping is
332 * under /O/<seq>/d<x>. */
336 * Masks for all features that should be supported by a Lustre version to
337 * access a specific file.
338 * This information is stored in lustre_mdt_attrs::lma_incompat.
341 LMAI_RELEASED = 0x00000001, /* file is released */
342 LMAI_AGENT = 0x00000002, /* agent inode */
343 LMAI_REMOTE_PARENT = 0x00000004, /* the parent of the object
344 is on the remote MDT */
346 #define LMA_INCOMPAT_SUPP (LMAI_AGENT | LMAI_REMOTE_PARENT)
348 extern void lustre_lma_swab(struct lustre_mdt_attrs *lma);
349 extern void lustre_lma_init(struct lustre_mdt_attrs *lma,
350 const struct lu_fid *fid,
351 __u32 compat, __u32 incompat);
353 * SOM on-disk attributes stored in a separate xattr.
356 /** Bitfield for supported data in this structure. For future use. */
359 /** Incompat feature list. The supported feature mask is availabe in
360 * SOM_INCOMPAT_SUPP */
363 /** IO Epoch SOM attributes belongs to */
365 /** total file size in objects */
367 /** total fs blocks in objects */
369 /** mds mount id the size is valid for */
372 extern void lustre_som_swab(struct som_attrs *attrs);
374 #define SOM_INCOMPAT_SUPP 0x0
377 * HSM on-disk attributes stored in a separate xattr.
380 /** Bitfield for supported data in this structure. For future use. */
383 /** HSM flags, see hsm_flags enum below */
385 /** backend archive id associated with the file */
387 /** version associated with the last archiving, if any */
390 extern void lustre_hsm_swab(struct hsm_attrs *attrs);
396 /** LASTID file has zero OID */
397 LUSTRE_FID_LASTID_OID = 0UL,
398 /** initial fid id value */
399 LUSTRE_FID_INIT_OID = 1UL
402 /** returns fid object sequence */
403 static inline __u64 fid_seq(const struct lu_fid *fid)
408 /** returns fid object id */
409 static inline __u32 fid_oid(const struct lu_fid *fid)
414 /** returns fid object version */
415 static inline __u32 fid_ver(const struct lu_fid *fid)
420 static inline void fid_zero(struct lu_fid *fid)
422 memset(fid, 0, sizeof(*fid));
425 static inline obd_id fid_ver_oid(const struct lu_fid *fid)
427 return ((__u64)fid_ver(fid) << 32 | fid_oid(fid));
431 * Note that reserved SEQ numbers below 12 will conflict with ldiskfs
432 * inodes in the IGIF namespace, so these reserved SEQ numbers can be
433 * used for other purposes and not risk collisions with existing inodes.
435 * Different FID Format
436 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs#NEW.0
439 FID_SEQ_OST_MDT0 = 0,
440 FID_SEQ_LLOG = 1, /* unnamed llogs */
442 FID_SEQ_OST_MDT1 = 3,
443 FID_SEQ_OST_MAX = 9, /* Max MDT count before OST_on_FID */
444 FID_SEQ_LLOG_NAME = 10, /* named llogs */
447 FID_SEQ_IGIF_MAX = 0x0ffffffffULL,
448 FID_SEQ_IDIF = 0x100000000ULL,
449 FID_SEQ_IDIF_MAX = 0x1ffffffffULL,
450 /* Normal FID sequence starts from this value, i.e. 1<<33 */
451 FID_SEQ_START = 0x200000000ULL,
452 /* sequence for local pre-defined FIDs listed in local_oid */
453 FID_SEQ_LOCAL_FILE = 0x200000001ULL,
454 FID_SEQ_DOT_LUSTRE = 0x200000002ULL,
455 /* sequence is used for local named objects FIDs generated
456 * by local_object_storage library */
457 FID_SEQ_LOCAL_NAME = 0x200000003ULL,
458 /* Because current FLD will only cache the fid sequence, instead
459 * of oid on the client side, if the FID needs to be exposed to
460 * clients sides, it needs to make sure all of fids under one
461 * sequence will be located in one MDT. */
462 FID_SEQ_SPECIAL = 0x200000004ULL,
463 FID_SEQ_QUOTA = 0x200000005ULL,
464 FID_SEQ_QUOTA_GLB = 0x200000006ULL,
465 FID_SEQ_ROOT = 0x200000007ULL, /* Located on MDT0 */
466 FID_SEQ_NORMAL = 0x200000400ULL,
467 FID_SEQ_LOV_DEFAULT = 0xffffffffffffffffULL
470 #define OBIF_OID_MAX_BITS 32
471 #define OBIF_MAX_OID (1ULL << OBIF_OID_MAX_BITS)
472 #define OBIF_OID_MASK ((1ULL << OBIF_OID_MAX_BITS) - 1)
473 #define IDIF_OID_MAX_BITS 48
474 #define IDIF_MAX_OID (1ULL << IDIF_OID_MAX_BITS)
475 #define IDIF_OID_MASK ((1ULL << IDIF_OID_MAX_BITS) - 1)
477 /** OID for FID_SEQ_SPECIAL */
479 /* Big Filesystem Lock to serialize rename operations */
480 FID_OID_SPECIAL_BFL = 1UL,
483 /** OID for FID_SEQ_DOT_LUSTRE */
484 enum dot_lustre_oid {
485 FID_OID_DOT_LUSTRE = 1UL,
486 FID_OID_DOT_LUSTRE_OBF = 2UL,
489 static inline int fid_seq_is_mdt0(obd_seq seq)
491 return (seq == FID_SEQ_OST_MDT0);
494 static inline int fid_seq_is_mdt(const __u64 seq)
496 return seq == FID_SEQ_OST_MDT0 || seq >= FID_SEQ_NORMAL;
499 static inline int fid_seq_is_echo(obd_seq seq)
501 return (seq == FID_SEQ_ECHO);
504 static inline int fid_is_echo(const struct lu_fid *fid)
506 return fid_seq_is_echo(fid_seq(fid));
509 static inline int fid_seq_is_llog(obd_seq seq)
511 return (seq == FID_SEQ_LLOG);
514 static inline int fid_is_llog(const struct lu_fid *fid)
516 /* file with OID == 0 is not llog but contains last oid */
517 return fid_seq_is_llog(fid_seq(fid)) && fid_oid(fid) > 0;
520 static inline int fid_seq_is_rsvd(const __u64 seq)
522 return (seq > FID_SEQ_OST_MDT0 && seq <= FID_SEQ_RSVD);
525 static inline int fid_seq_is_special(const __u64 seq)
527 return seq == FID_SEQ_SPECIAL;
530 static inline int fid_seq_is_local_file(const __u64 seq)
532 return seq == FID_SEQ_LOCAL_FILE ||
533 seq == FID_SEQ_LOCAL_NAME;
536 static inline int fid_seq_is_root(const __u64 seq)
538 return seq == FID_SEQ_ROOT;
541 static inline int fid_seq_is_dot(const __u64 seq)
543 return seq == FID_SEQ_DOT_LUSTRE;
546 static inline int fid_seq_is_default(const __u64 seq)
548 return seq == FID_SEQ_LOV_DEFAULT;
551 static inline int fid_is_mdt0(const struct lu_fid *fid)
553 return fid_seq_is_mdt0(fid_seq(fid));
556 static inline void lu_root_fid(struct lu_fid *fid)
558 fid->f_seq = FID_SEQ_ROOT;
564 * Check if a fid is igif or not.
565 * \param fid the fid to be tested.
566 * \return true if the fid is a igif; otherwise false.
568 static inline int fid_seq_is_igif(const __u64 seq)
570 return seq >= FID_SEQ_IGIF && seq <= FID_SEQ_IGIF_MAX;
573 static inline int fid_is_igif(const struct lu_fid *fid)
575 return fid_seq_is_igif(fid_seq(fid));
579 * Check if a fid is idif or not.
580 * \param fid the fid to be tested.
581 * \return true if the fid is a idif; otherwise false.
583 static inline int fid_seq_is_idif(const __u64 seq)
585 return seq >= FID_SEQ_IDIF && seq <= FID_SEQ_IDIF_MAX;
588 static inline int fid_is_idif(const struct lu_fid *fid)
590 return fid_seq_is_idif(fid_seq(fid));
593 static inline int fid_is_local_file(const struct lu_fid *fid)
595 return fid_seq_is_local_file(fid_seq(fid));
598 static inline int fid_seq_is_norm(const __u64 seq)
600 return (seq >= FID_SEQ_NORMAL);
603 static inline int fid_is_norm(const struct lu_fid *fid)
605 return fid_seq_is_norm(fid_seq(fid));
608 /* convert an OST objid into an IDIF FID SEQ number */
609 static inline obd_seq fid_idif_seq(obd_id id, __u32 ost_idx)
611 return FID_SEQ_IDIF | (ost_idx << 16) | ((id >> 32) & 0xffff);
614 /* convert a packed IDIF FID into an OST objid */
615 static inline obd_id fid_idif_id(obd_seq seq, __u32 oid, __u32 ver)
617 return ((__u64)ver << 48) | ((seq & 0xffff) << 32) | oid;
620 static inline __u32 idif_ost_idx(obd_seq seq)
622 return (seq >> 16) & 0xffff;
625 /* extract ost index from IDIF FID */
626 static inline __u32 fid_idif_ost_idx(const struct lu_fid *fid)
628 return idif_ost_idx(fid_seq(fid));
631 /* extract OST sequence (group) from a wire ost_id (id/seq) pair */
632 static inline obd_seq ostid_seq(const struct ost_id *ostid)
634 if (fid_seq_is_mdt0(ostid->oi.oi_seq))
635 return FID_SEQ_OST_MDT0;
637 if (fid_seq_is_default(ostid->oi.oi_seq))
638 return FID_SEQ_LOV_DEFAULT;
640 if (fid_is_idif(&ostid->oi_fid))
641 return FID_SEQ_OST_MDT0;
643 return fid_seq(&ostid->oi_fid);
646 /* extract OST objid from a wire ost_id (id/seq) pair */
647 static inline obd_id ostid_id(const struct ost_id *ostid)
649 if (fid_seq_is_mdt0(ostid_seq(ostid)))
650 return ostid->oi.oi_id & IDIF_OID_MASK;
652 if (fid_is_idif(&ostid->oi_fid))
653 return fid_idif_id(fid_seq(&ostid->oi_fid),
654 fid_oid(&ostid->oi_fid), 0);
656 return fid_oid(&ostid->oi_fid);
659 static inline void ostid_set_seq(struct ost_id *oi, __u64 seq)
661 if (fid_seq_is_mdt0(seq) || fid_seq_is_default(seq)) {
664 oi->oi_fid.f_seq = seq;
665 /* Note: if f_oid + f_ver is zero, we need init it
666 * to be 1, otherwise, ostid_seq will treat this
667 * as old ostid (oi_seq == 0) */
668 if (oi->oi_fid.f_oid == 0 && oi->oi_fid.f_ver == 0)
669 oi->oi_fid.f_oid = LUSTRE_FID_INIT_OID;
673 static inline void ostid_set_seq_mdt0(struct ost_id *oi)
675 ostid_set_seq(oi, FID_SEQ_OST_MDT0);
678 static inline void ostid_set_seq_echo(struct ost_id *oi)
680 ostid_set_seq(oi, FID_SEQ_ECHO);
683 static inline void ostid_set_seq_llog(struct ost_id *oi)
685 ostid_set_seq(oi, FID_SEQ_LLOG);
689 * Note: we need check oi_seq to decide where to set oi_id,
690 * so oi_seq should always be set ahead of oi_id.
692 static inline void ostid_set_id(struct ost_id *oi, __u64 oid)
694 if (fid_seq_is_mdt0(ostid_seq(oi))) {
695 if (oid >= IDIF_MAX_OID) {
696 CERROR("Bad "LPU64" to set "DOSTID"\n",
702 if (oid > OBIF_MAX_OID) {
703 CERROR("Bad "LPU64" to set "DOSTID"\n",
707 oi->oi_fid.f_oid = oid;
711 static inline void ostid_inc_id(struct ost_id *oi)
713 if (fid_seq_is_mdt0(ostid_seq(oi))) {
714 if (unlikely(ostid_id(oi) + 1 > IDIF_MAX_OID)) {
715 CERROR("Bad inc "DOSTID"\n", POSTID(oi));
724 static inline void ostid_dec_id(struct ost_id *oi)
726 if (fid_seq_is_mdt0(ostid_seq(oi)))
733 * Unpack an OST object id/seq (group) into a FID. This is needed for
734 * converting all obdo, lmm, lsm, etc. 64-bit id/seq pairs into proper
735 * FIDs. Note that if an id/seq is already in FID/IDIF format it will
736 * be passed through unchanged. Only legacy OST objects in "group 0"
737 * will be mapped into the IDIF namespace so that they can fit into the
738 * struct lu_fid fields without loss. For reference see:
739 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs
741 static inline int ostid_to_fid(struct lu_fid *fid, struct ost_id *ostid,
744 if (ost_idx > 0xffff) {
745 CERROR("bad ost_idx, "DOSTID" ost_idx:%u\n", POSTID(ostid),
750 if (fid_seq_is_mdt0(ostid_seq(ostid))) {
751 /* This is a "legacy" (old 1.x/2.early) OST object in "group 0"
752 * that we map into the IDIF namespace. It allows up to 2^48
753 * objects per OST, as this is the object namespace that has
754 * been in production for years. This can handle create rates
755 * of 1M objects/s/OST for 9 years, or combinations thereof. */
756 if (ostid_id(ostid) >= IDIF_MAX_OID) {
757 CERROR("bad MDT0 id, "DOSTID" ost_idx:%u\n",
758 POSTID(ostid), ost_idx);
761 fid->f_seq = fid_idif_seq(ostid_id(ostid), ost_idx);
762 /* truncate to 32 bits by assignment */
763 fid->f_oid = ostid_id(ostid);
764 /* in theory, not currently used */
765 fid->f_ver = ostid_id(ostid) >> 48;
766 } else /* if (fid_seq_is_idif(seq) || fid_seq_is_norm(seq)) */ {
767 /* This is either an IDIF object, which identifies objects across
768 * all OSTs, or a regular FID. The IDIF namespace maps legacy
769 * OST objects into the FID namespace. In both cases, we just
770 * pass the FID through, no conversion needed. */
771 if (ostid->oi_fid.f_ver != 0) {
772 CERROR("bad MDT0 id, "DOSTID" ost_idx:%u\n",
773 POSTID(ostid), ost_idx);
776 *fid = ostid->oi_fid;
782 /* pack any OST FID into an ostid (id/seq) for the wire/disk */
783 static inline int fid_to_ostid(const struct lu_fid *fid, struct ost_id *ostid)
785 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
786 CERROR("bad IGIF, "DFID"\n", PFID(fid));
790 if (fid_is_idif(fid)) {
791 ostid_set_seq_mdt0(ostid);
792 ostid_set_id(ostid, fid_idif_id(fid_seq(fid), fid_oid(fid),
795 ostid->oi_fid = *fid;
801 /* Check whether the fid is for LAST_ID */
802 static inline int fid_is_last_id(const struct lu_fid *fid)
804 return (fid_oid(fid) == 0);
808 * Get inode number from a igif.
809 * \param fid a igif to get inode number from.
810 * \return inode number for the igif.
812 static inline ino_t lu_igif_ino(const struct lu_fid *fid)
817 extern void lustre_swab_ost_id(struct ost_id *oid);
820 * Get inode generation from a igif.
821 * \param fid a igif to get inode generation from.
822 * \return inode generation for the igif.
824 static inline __u32 lu_igif_gen(const struct lu_fid *fid)
830 * Build igif from the inode number/generation.
832 static inline void lu_igif_build(struct lu_fid *fid, __u32 ino, __u32 gen)
840 * Fids are transmitted across network (in the sender byte-ordering),
841 * and stored on disk in big-endian order.
843 static inline void fid_cpu_to_le(struct lu_fid *dst, const struct lu_fid *src)
845 dst->f_seq = cpu_to_le64(fid_seq(src));
846 dst->f_oid = cpu_to_le32(fid_oid(src));
847 dst->f_ver = cpu_to_le32(fid_ver(src));
850 static inline void fid_le_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
852 dst->f_seq = le64_to_cpu(fid_seq(src));
853 dst->f_oid = le32_to_cpu(fid_oid(src));
854 dst->f_ver = le32_to_cpu(fid_ver(src));
857 static inline void fid_cpu_to_be(struct lu_fid *dst, const struct lu_fid *src)
859 dst->f_seq = cpu_to_be64(fid_seq(src));
860 dst->f_oid = cpu_to_be32(fid_oid(src));
861 dst->f_ver = cpu_to_be32(fid_ver(src));
864 static inline void fid_be_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
866 dst->f_seq = be64_to_cpu(fid_seq(src));
867 dst->f_oid = be32_to_cpu(fid_oid(src));
868 dst->f_ver = be32_to_cpu(fid_ver(src));
871 static inline int fid_is_sane(const struct lu_fid *fid)
873 return fid != NULL &&
874 ((fid_seq(fid) >= FID_SEQ_START && fid_ver(fid) == 0) ||
875 fid_is_igif(fid) || fid_is_idif(fid) ||
876 fid_seq_is_rsvd(fid_seq(fid)));
879 static inline int fid_is_zero(const struct lu_fid *fid)
881 return fid_seq(fid) == 0 && fid_oid(fid) == 0;
884 extern void lustre_swab_lu_fid(struct lu_fid *fid);
885 extern void lustre_swab_lu_seq_range(struct lu_seq_range *range);
887 static inline int lu_fid_eq(const struct lu_fid *f0, const struct lu_fid *f1)
889 return memcmp(f0, f1, sizeof *f0) == 0;
892 #define __diff_normalize(val0, val1) \
894 typeof(val0) __val0 = (val0); \
895 typeof(val1) __val1 = (val1); \
897 (__val0 == __val1 ? 0 : __val0 > __val1 ? +1 : -1); \
900 static inline int lu_fid_cmp(const struct lu_fid *f0,
901 const struct lu_fid *f1)
904 __diff_normalize(fid_seq(f0), fid_seq(f1)) ?:
905 __diff_normalize(fid_oid(f0), fid_oid(f1)) ?:
906 __diff_normalize(fid_ver(f0), fid_ver(f1));
909 static inline void ostid_cpu_to_le(const struct ost_id *src_oi,
910 struct ost_id *dst_oi)
912 if (fid_seq_is_mdt0(ostid_seq(src_oi))) {
913 dst_oi->oi.oi_id = cpu_to_le64(src_oi->oi.oi_id);
914 dst_oi->oi.oi_seq = cpu_to_le64(src_oi->oi.oi_seq);
916 fid_cpu_to_le(&dst_oi->oi_fid, &src_oi->oi_fid);
920 static inline void ostid_le_to_cpu(const struct ost_id *src_oi,
921 struct ost_id *dst_oi)
923 if (fid_seq_is_mdt0(ostid_seq(src_oi))) {
924 dst_oi->oi.oi_id = le64_to_cpu(src_oi->oi.oi_id);
925 dst_oi->oi.oi_seq = le64_to_cpu(src_oi->oi.oi_seq);
927 fid_le_to_cpu(&dst_oi->oi_fid, &src_oi->oi_fid);
933 /** \defgroup lu_dir lu_dir
937 * Enumeration of possible directory entry attributes.
939 * Attributes follow directory entry header in the order they appear in this
942 enum lu_dirent_attrs {
945 LUDA_64BITHASH = 0x0004,
947 /* The following attrs are used for MDT interanl only,
948 * not visible to client */
950 /* Verify the dirent consistency */
951 LUDA_VERIFY = 0x8000,
952 /* Only check but not repair the dirent inconsistency */
953 LUDA_VERIFY_DRYRUN = 0x4000,
954 /* The dirent has been repaired, or to be repaired (dryrun). */
955 LUDA_REPAIR = 0x2000,
956 /* The system is upgraded, has beed or to be repaired (dryrun). */
957 LUDA_UPGRADE = 0x1000,
958 /* Ignore this record, go to next directly. */
959 LUDA_IGNORE = 0x0800,
962 #define LU_DIRENT_ATTRS_MASK 0xf800
965 * Layout of readdir pages, as transmitted on wire.
968 /** valid if LUDA_FID is set. */
969 struct lu_fid lde_fid;
970 /** a unique entry identifier: a hash or an offset. */
972 /** total record length, including all attributes. */
976 /** optional variable size attributes following this entry.
977 * taken from enum lu_dirent_attrs.
980 /** name is followed by the attributes indicated in ->ldp_attrs, in
981 * their natural order. After the last attribute, padding bytes are
982 * added to make ->lde_reclen a multiple of 8.
988 * Definitions of optional directory entry attributes formats.
990 * Individual attributes do not have their length encoded in a generic way. It
991 * is assumed that consumer of an attribute knows its format. This means that
992 * it is impossible to skip over an unknown attribute, except by skipping over all
993 * remaining attributes (by using ->lde_reclen), which is not too
994 * constraining, because new server versions will append new attributes at
995 * the end of an entry.
999 * Fid directory attribute: a fid of an object referenced by the entry. This
1000 * will be almost always requested by the client and supplied by the server.
1002 * Aligned to 8 bytes.
1004 /* To have compatibility with 1.8, lets have fid in lu_dirent struct. */
1009 * Aligned to 2 bytes.
1016 __u64 ldp_hash_start;
1020 struct lu_dirent ldp_entries[0];
1023 enum lu_dirpage_flags {
1025 * dirpage contains no entry.
1029 * last entry's lde_hash equals ldp_hash_end.
1031 LDF_COLLIDE = 1 << 1
1034 static inline struct lu_dirent *lu_dirent_start(struct lu_dirpage *dp)
1036 if (le32_to_cpu(dp->ldp_flags) & LDF_EMPTY)
1039 return dp->ldp_entries;
1042 static inline struct lu_dirent *lu_dirent_next(struct lu_dirent *ent)
1044 struct lu_dirent *next;
1046 if (le16_to_cpu(ent->lde_reclen) != 0)
1047 next = ((void *)ent) + le16_to_cpu(ent->lde_reclen);
1054 static inline int lu_dirent_calc_size(int namelen, __u16 attr)
1058 if (attr & LUDA_TYPE) {
1059 const unsigned align = sizeof(struct luda_type) - 1;
1060 size = (sizeof(struct lu_dirent) + namelen + align) & ~align;
1061 size += sizeof(struct luda_type);
1063 size = sizeof(struct lu_dirent) + namelen;
1065 return (size + 7) & ~7;
1068 static inline int lu_dirent_size(struct lu_dirent *ent)
1070 if (le16_to_cpu(ent->lde_reclen) == 0) {
1071 return lu_dirent_calc_size(le16_to_cpu(ent->lde_namelen),
1072 le32_to_cpu(ent->lde_attrs));
1074 return le16_to_cpu(ent->lde_reclen);
1077 #define MDS_DIR_END_OFF 0xfffffffffffffffeULL
1080 * MDS_READPAGE page size
1082 * This is the directory page size packed in MDS_READPAGE RPC.
1083 * It's different than PAGE_CACHE_SIZE because the client needs to
1084 * access the struct lu_dirpage header packed at the beginning of
1085 * the "page" and without this there isn't any way to know find the
1086 * lu_dirpage header is if client and server PAGE_CACHE_SIZE differ.
1088 #define LU_PAGE_SHIFT 12
1089 #define LU_PAGE_SIZE (1UL << LU_PAGE_SHIFT)
1090 #define LU_PAGE_MASK (~(LU_PAGE_SIZE - 1))
1092 #define LU_PAGE_COUNT (1 << (PAGE_CACHE_SHIFT - LU_PAGE_SHIFT))
1096 struct lustre_handle {
1099 #define DEAD_HANDLE_MAGIC 0xdeadbeefcafebabeULL
1101 static inline int lustre_handle_is_used(struct lustre_handle *lh)
1103 return lh->cookie != 0ull;
1106 static inline int lustre_handle_equal(const struct lustre_handle *lh1,
1107 const struct lustre_handle *lh2)
1109 return lh1->cookie == lh2->cookie;
1112 static inline void lustre_handle_copy(struct lustre_handle *tgt,
1113 struct lustre_handle *src)
1115 tgt->cookie = src->cookie;
1118 /* flags for lm_flags */
1119 #define MSGHDR_AT_SUPPORT 0x1
1120 #define MSGHDR_CKSUM_INCOMPAT18 0x2
1122 #define lustre_msg lustre_msg_v2
1123 /* we depend on this structure to be 8-byte aligned */
1124 /* this type is only endian-adjusted in lustre_unpack_msg() */
1125 struct lustre_msg_v2 {
1134 __u32 lm_buflens[0];
1137 /* without gss, ptlrpc_body is put at the first buffer. */
1138 #define PTLRPC_NUM_VERSIONS 4
1139 #define JOBSTATS_JOBID_SIZE 32 /* 32 bytes string */
1140 struct ptlrpc_body_v3 {
1141 struct lustre_handle pb_handle;
1148 __u64 pb_last_committed;
1153 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1154 __u32 pb_service_time; /* for rep, actual service time */
1157 /* VBR: pre-versions */
1158 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1159 /* padding for future needs */
1160 __u64 pb_padding[4];
1161 char pb_jobid[JOBSTATS_JOBID_SIZE];
1163 #define ptlrpc_body ptlrpc_body_v3
1165 struct ptlrpc_body_v2 {
1166 struct lustre_handle pb_handle;
1173 __u64 pb_last_committed;
1178 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1179 __u32 pb_service_time; /* for rep, actual service time, also used for
1180 net_latency of req */
1183 /* VBR: pre-versions */
1184 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1185 /* padding for future needs */
1186 __u64 pb_padding[4];
1189 extern void lustre_swab_ptlrpc_body(struct ptlrpc_body *pb);
1191 /* message body offset for lustre_msg_v2 */
1192 /* ptlrpc body offset in all request/reply messages */
1193 #define MSG_PTLRPC_BODY_OFF 0
1195 /* normal request/reply message record offset */
1196 #define REQ_REC_OFF 1
1197 #define REPLY_REC_OFF 1
1199 /* ldlm request message body offset */
1200 #define DLM_LOCKREQ_OFF 1 /* lockreq offset */
1201 #define DLM_REQ_REC_OFF 2 /* normal dlm request record offset */
1203 /* ldlm intent lock message body offset */
1204 #define DLM_INTENT_IT_OFF 2 /* intent lock it offset */
1205 #define DLM_INTENT_REC_OFF 3 /* intent lock record offset */
1207 /* ldlm reply message body offset */
1208 #define DLM_LOCKREPLY_OFF 1 /* lockrep offset */
1209 #define DLM_REPLY_REC_OFF 2 /* reply record offset */
1211 /** only use in req->rq_{req,rep}_swab_mask */
1212 #define MSG_PTLRPC_HEADER_OFF 31
1214 /* Flags that are operation-specific go in the top 16 bits. */
1215 #define MSG_OP_FLAG_MASK 0xffff0000
1216 #define MSG_OP_FLAG_SHIFT 16
1218 /* Flags that apply to all requests are in the bottom 16 bits */
1219 #define MSG_GEN_FLAG_MASK 0x0000ffff
1220 #define MSG_LAST_REPLAY 0x0001
1221 #define MSG_RESENT 0x0002
1222 #define MSG_REPLAY 0x0004
1223 /* #define MSG_AT_SUPPORT 0x0008
1224 * This was used in early prototypes of adaptive timeouts, and while there
1225 * shouldn't be any users of that code there also isn't a need for using this
1226 * bits. Defer usage until at least 1.10 to avoid potential conflict. */
1227 #define MSG_DELAY_REPLAY 0x0010
1228 #define MSG_VERSION_REPLAY 0x0020
1229 #define MSG_REQ_REPLAY_DONE 0x0040
1230 #define MSG_LOCK_REPLAY_DONE 0x0080
1233 * Flags for all connect opcodes (MDS_CONNECT, OST_CONNECT)
1236 #define MSG_CONNECT_RECOVERING 0x00000001
1237 #define MSG_CONNECT_RECONNECT 0x00000002
1238 #define MSG_CONNECT_REPLAYABLE 0x00000004
1239 //#define MSG_CONNECT_PEER 0x8
1240 #define MSG_CONNECT_LIBCLIENT 0x00000010
1241 #define MSG_CONNECT_INITIAL 0x00000020
1242 #define MSG_CONNECT_ASYNC 0x00000040
1243 #define MSG_CONNECT_NEXT_VER 0x00000080 /* use next version of lustre_msg */
1244 #define MSG_CONNECT_TRANSNO 0x00000100 /* report transno */
1247 #define OBD_CONNECT_RDONLY 0x1ULL /*client has read-only access*/
1248 #define OBD_CONNECT_INDEX 0x2ULL /*connect specific LOV idx */
1249 #define OBD_CONNECT_MDS 0x4ULL /*connect from MDT to OST */
1250 #define OBD_CONNECT_GRANT 0x8ULL /*OSC gets grant at connect */
1251 #define OBD_CONNECT_SRVLOCK 0x10ULL /*server takes locks for cli */
1252 #define OBD_CONNECT_VERSION 0x20ULL /*Lustre versions in ocd */
1253 #define OBD_CONNECT_REQPORTAL 0x40ULL /*Separate non-IO req portal */
1254 #define OBD_CONNECT_ACL 0x80ULL /*access control lists */
1255 #define OBD_CONNECT_XATTR 0x100ULL /*client use extended attr */
1256 #define OBD_CONNECT_CROW 0x200ULL /*MDS+OST create obj on write*/
1257 #define OBD_CONNECT_TRUNCLOCK 0x400ULL /*locks on server for punch */
1258 #define OBD_CONNECT_TRANSNO 0x800ULL /*replay sends init transno */
1259 #define OBD_CONNECT_IBITS 0x1000ULL /*support for inodebits locks*/
1260 #define OBD_CONNECT_JOIN 0x2000ULL /*files can be concatenated.
1261 *We do not support JOIN FILE
1262 *anymore, reserve this flags
1263 *just for preventing such bit
1265 #define OBD_CONNECT_ATTRFID 0x4000ULL /*Server can GetAttr By Fid*/
1266 #define OBD_CONNECT_NODEVOH 0x8000ULL /*No open hndl on specl nodes*/
1267 #define OBD_CONNECT_RMT_CLIENT 0x10000ULL /*Remote client */
1268 #define OBD_CONNECT_RMT_CLIENT_FORCE 0x20000ULL /*Remote client by force */
1269 #define OBD_CONNECT_BRW_SIZE 0x40000ULL /*Max bytes per rpc */
1270 #define OBD_CONNECT_QUOTA64 0x80000ULL /*Not used since 2.4 */
1271 #define OBD_CONNECT_MDS_CAPA 0x100000ULL /*MDS capability */
1272 #define OBD_CONNECT_OSS_CAPA 0x200000ULL /*OSS capability */
1273 #define OBD_CONNECT_CANCELSET 0x400000ULL /*Early batched cancels. */
1274 #define OBD_CONNECT_SOM 0x800000ULL /*Size on MDS */
1275 #define OBD_CONNECT_AT 0x1000000ULL /*client uses AT */
1276 #define OBD_CONNECT_LRU_RESIZE 0x2000000ULL /*LRU resize feature. */
1277 #define OBD_CONNECT_MDS_MDS 0x4000000ULL /*MDS-MDS connection */
1278 #define OBD_CONNECT_REAL 0x8000000ULL /*real connection */
1279 #define OBD_CONNECT_CHANGE_QS 0x10000000ULL /*Not used since 2.4 */
1280 #define OBD_CONNECT_CKSUM 0x20000000ULL /*support several cksum algos*/
1281 #define OBD_CONNECT_FID 0x40000000ULL /*FID is supported by server */
1282 #define OBD_CONNECT_VBR 0x80000000ULL /*version based recovery */
1283 #define OBD_CONNECT_LOV_V3 0x100000000ULL /*client supports LOV v3 EA */
1284 #define OBD_CONNECT_GRANT_SHRINK 0x200000000ULL /* support grant shrink */
1285 #define OBD_CONNECT_SKIP_ORPHAN 0x400000000ULL /* don't reuse orphan objids */
1286 #define OBD_CONNECT_MAX_EASIZE 0x800000000ULL /* preserved for large EA */
1287 #define OBD_CONNECT_FULL20 0x1000000000ULL /* it is 2.0 client */
1288 #define OBD_CONNECT_LAYOUTLOCK 0x2000000000ULL /* client uses layout lock */
1289 #define OBD_CONNECT_64BITHASH 0x4000000000ULL /* client supports 64-bits
1291 #define OBD_CONNECT_MAXBYTES 0x8000000000ULL /* max stripe size */
1292 #define OBD_CONNECT_IMP_RECOV 0x10000000000ULL /* imp recovery support */
1293 #define OBD_CONNECT_JOBSTATS 0x20000000000ULL /* jobid in ptlrpc_body */
1294 #define OBD_CONNECT_UMASK 0x40000000000ULL /* create uses client umask */
1295 #define OBD_CONNECT_EINPROGRESS 0x80000000000ULL /* client handles -EINPROGRESS
1296 * RPC error properly */
1297 #define OBD_CONNECT_GRANT_PARAM 0x100000000000ULL/* extra grant params used for
1298 * finer space reservation */
1299 #define OBD_CONNECT_FLOCK_OWNER 0x200000000000ULL /* for the fixed 1.8
1300 * policy and 2.x server */
1301 #define OBD_CONNECT_LVB_TYPE 0x400000000000ULL /* variable type of LVB */
1302 #define OBD_CONNECT_NANOSEC_TIME 0x800000000000ULL /* nanosecond timestamps */
1303 #define OBD_CONNECT_LIGHTWEIGHT 0x1000000000000ULL/* lightweight connection */
1304 #define OBD_CONNECT_SHORTIO 0x2000000000000ULL/* short io */
1305 #define OBD_CONNECT_PINGLESS 0x4000000000000ULL/* pings not required */
1306 #define OBD_CONNECT_FLOCK_DEAD 0x8000000000000ULL/* improved flock deadlock detection */
1307 #define OBD_CONNECT_DISP_STRIPE 0x10000000000000ULL/* create stripe disposition*/
1308 #define OBD_CONNECT_OPEN_BY_FID 0x20000000000000ULL /* open by fid won't pack
1312 * Please DO NOT add flag values here before first ensuring that this same
1313 * flag value is not in use on some other branch. Please clear any such
1314 * changes with senior engineers before starting to use a new flag. Then,
1315 * submit a small patch against EVERY branch that ONLY adds the new flag,
1316 * updates obd_connect_names[] for lprocfs_rd_connect_flags(), adds the
1317 * flag to check_obd_connect_data(), and updates wiretests accordingly, so it
1318 * can be approved and landed easily to reserve the flag for future use. */
1320 /* The MNE_SWAB flag is overloading the MDS_MDS bit only for the MGS
1321 * connection. It is a temporary bug fix for Imperative Recovery interop
1322 * between 2.2 and 2.3 x86/ppc nodes, and can be removed when interop for
1323 * 2.2 clients/servers is no longer needed. LU-1252/LU-1644. */
1324 #define OBD_CONNECT_MNE_SWAB OBD_CONNECT_MDS_MDS
1326 #define OCD_HAS_FLAG(ocd, flg) \
1327 (!!((ocd)->ocd_connect_flags & OBD_CONNECT_##flg))
1330 #ifdef HAVE_LRU_RESIZE_SUPPORT
1331 #define LRU_RESIZE_CONNECT_FLAG OBD_CONNECT_LRU_RESIZE
1333 #define LRU_RESIZE_CONNECT_FLAG 0
1336 #define MDT_CONNECT_SUPPORTED (OBD_CONNECT_RDONLY | OBD_CONNECT_VERSION | \
1337 OBD_CONNECT_ACL | OBD_CONNECT_XATTR | \
1338 OBD_CONNECT_IBITS | \
1339 OBD_CONNECT_NODEVOH | OBD_CONNECT_ATTRFID | \
1340 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1341 OBD_CONNECT_RMT_CLIENT | \
1342 OBD_CONNECT_RMT_CLIENT_FORCE | \
1343 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_MDS_CAPA | \
1344 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_MDS_MDS | \
1345 OBD_CONNECT_FID | LRU_RESIZE_CONNECT_FLAG | \
1346 OBD_CONNECT_VBR | OBD_CONNECT_LOV_V3 | \
1347 OBD_CONNECT_SOM | OBD_CONNECT_FULL20 | \
1348 OBD_CONNECT_64BITHASH | OBD_CONNECT_JOBSTATS | \
1349 OBD_CONNECT_EINPROGRESS | \
1350 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_UMASK | \
1351 OBD_CONNECT_LVB_TYPE | OBD_CONNECT_LAYOUTLOCK |\
1352 OBD_CONNECT_PINGLESS | OBD_CONNECT_MAX_EASIZE |\
1353 OBD_CONNECT_FLOCK_DEAD | \
1354 OBD_CONNECT_DISP_STRIPE)
1356 #define OST_CONNECT_SUPPORTED (OBD_CONNECT_SRVLOCK | OBD_CONNECT_GRANT | \
1357 OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | \
1358 OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_INDEX | \
1359 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_OSS_CAPA | \
1360 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1361 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_CKSUM | \
1362 OBD_CONNECT_RMT_CLIENT | \
1363 OBD_CONNECT_RMT_CLIENT_FORCE | OBD_CONNECT_VBR | \
1364 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN | \
1365 OBD_CONNECT_GRANT_SHRINK | OBD_CONNECT_FULL20 | \
1366 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES | \
1367 OBD_CONNECT_MAX_EASIZE | \
1368 OBD_CONNECT_EINPROGRESS | \
1369 OBD_CONNECT_JOBSTATS | \
1370 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LVB_TYPE|\
1371 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_FID | \
1372 OBD_CONNECT_PINGLESS)
1373 #define ECHO_CONNECT_SUPPORTED (0)
1374 #define MGS_CONNECT_SUPPORTED (OBD_CONNECT_VERSION | OBD_CONNECT_AT | \
1375 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV | \
1376 OBD_CONNECT_MNE_SWAB | OBD_CONNECT_PINGLESS)
1378 /* Features required for this version of the client to work with server */
1379 #define CLIENT_CONNECT_MDT_REQD (OBD_CONNECT_IBITS | OBD_CONNECT_FID | \
1382 #define OBD_OCD_VERSION(major,minor,patch,fix) (((major)<<24) + ((minor)<<16) +\
1383 ((patch)<<8) + (fix))
1384 #define OBD_OCD_VERSION_MAJOR(version) ((int)((version)>>24)&255)
1385 #define OBD_OCD_VERSION_MINOR(version) ((int)((version)>>16)&255)
1386 #define OBD_OCD_VERSION_PATCH(version) ((int)((version)>>8)&255)
1387 #define OBD_OCD_VERSION_FIX(version) ((int)(version)&255)
1389 /* This structure is used for both request and reply.
1391 * If we eventually have separate connect data for different types, which we
1392 * almost certainly will, then perhaps we stick a union in here. */
1393 struct obd_connect_data_v1 {
1394 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1395 __u32 ocd_version; /* lustre release version number */
1396 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1397 __u32 ocd_index; /* LOV index to connect to */
1398 __u32 ocd_brw_size; /* Maximum BRW size in bytes, must be 2^n */
1399 __u64 ocd_ibits_known; /* inode bits this client understands */
1400 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1401 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1402 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1403 __u32 ocd_unused; /* also fix lustre_swab_connect */
1404 __u64 ocd_transno; /* first transno from client to be replayed */
1405 __u32 ocd_group; /* MDS group on OST */
1406 __u32 ocd_cksum_types; /* supported checksum algorithms */
1407 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1408 __u32 ocd_instance; /* also fix lustre_swab_connect */
1409 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1412 struct obd_connect_data {
1413 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1414 __u32 ocd_version; /* lustre release version number */
1415 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1416 __u32 ocd_index; /* LOV index to connect to */
1417 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
1418 __u64 ocd_ibits_known; /* inode bits this client understands */
1419 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1420 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1421 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1422 __u32 ocd_unused; /* also fix lustre_swab_connect */
1423 __u64 ocd_transno; /* first transno from client to be replayed */
1424 __u32 ocd_group; /* MDS group on OST */
1425 __u32 ocd_cksum_types; /* supported checksum algorithms */
1426 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1427 __u32 ocd_instance; /* instance # of this target */
1428 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1429 /* Fields after ocd_maxbytes are only accessible by the receiver
1430 * if the corresponding flag in ocd_connect_flags is set. Accessing
1431 * any field after ocd_maxbytes on the receiver without a valid flag
1432 * may result in out-of-bound memory access and kernel oops. */
1433 __u64 padding1; /* added 2.1.0. also fix lustre_swab_connect */
1434 __u64 padding2; /* added 2.1.0. also fix lustre_swab_connect */
1435 __u64 padding3; /* added 2.1.0. also fix lustre_swab_connect */
1436 __u64 padding4; /* added 2.1.0. also fix lustre_swab_connect */
1437 __u64 padding5; /* added 2.1.0. also fix lustre_swab_connect */
1438 __u64 padding6; /* added 2.1.0. also fix lustre_swab_connect */
1439 __u64 padding7; /* added 2.1.0. also fix lustre_swab_connect */
1440 __u64 padding8; /* added 2.1.0. also fix lustre_swab_connect */
1441 __u64 padding9; /* added 2.1.0. also fix lustre_swab_connect */
1442 __u64 paddingA; /* added 2.1.0. also fix lustre_swab_connect */
1443 __u64 paddingB; /* added 2.1.0. also fix lustre_swab_connect */
1444 __u64 paddingC; /* added 2.1.0. also fix lustre_swab_connect */
1445 __u64 paddingD; /* added 2.1.0. also fix lustre_swab_connect */
1446 __u64 paddingE; /* added 2.1.0. also fix lustre_swab_connect */
1447 __u64 paddingF; /* added 2.1.0. also fix lustre_swab_connect */
1450 * Please DO NOT use any fields here before first ensuring that this same
1451 * field is not in use on some other branch. Please clear any such changes
1452 * with senior engineers before starting to use a new field. Then, submit
1453 * a small patch against EVERY branch that ONLY adds the new field along with
1454 * the matching OBD_CONNECT flag, so that can be approved and landed easily to
1455 * reserve the flag for future use. */
1458 extern void lustre_swab_connect(struct obd_connect_data *ocd);
1461 * Supported checksum algorithms. Up to 32 checksum types are supported.
1462 * (32-bit mask stored in obd_connect_data::ocd_cksum_types)
1463 * Please update DECLARE_CKSUM_NAME/OBD_CKSUM_ALL in obd.h when adding a new
1464 * algorithm and also the OBD_FL_CKSUM* flags.
1467 OBD_CKSUM_CRC32 = 0x00000001,
1468 OBD_CKSUM_ADLER = 0x00000002,
1469 OBD_CKSUM_CRC32C= 0x00000004,
1473 * OST requests: OBDO & OBD request records
1478 OST_REPLY = 0, /* reply ? */
1494 OST_QUOTACHECK = 18,
1496 OST_QUOTA_ADJUST_QUNIT = 20, /* not used since 2.4 */
1499 #define OST_FIRST_OPC OST_REPLY
1502 OBD_FL_INLINEDATA = 0x00000001,
1503 OBD_FL_OBDMDEXISTS = 0x00000002,
1504 OBD_FL_DELORPHAN = 0x00000004, /* if set in o_flags delete orphans */
1505 OBD_FL_NORPC = 0x00000008, /* set in o_flags do in OSC not OST */
1506 OBD_FL_IDONLY = 0x00000010, /* set in o_flags only adjust obj id*/
1507 OBD_FL_RECREATE_OBJS= 0x00000020, /* recreate missing obj */
1508 OBD_FL_DEBUG_CHECK = 0x00000040, /* echo client/server debug check */
1509 OBD_FL_NO_USRQUOTA = 0x00000100, /* the object's owner is over quota */
1510 OBD_FL_NO_GRPQUOTA = 0x00000200, /* the object's group is over quota */
1511 OBD_FL_CREATE_CROW = 0x00000400, /* object should be create on write */
1512 OBD_FL_SRVLOCK = 0x00000800, /* delegate DLM locking to server */
1513 OBD_FL_CKSUM_CRC32 = 0x00001000, /* CRC32 checksum type */
1514 OBD_FL_CKSUM_ADLER = 0x00002000, /* ADLER checksum type */
1515 OBD_FL_CKSUM_CRC32C = 0x00004000, /* CRC32C checksum type */
1516 OBD_FL_CKSUM_RSVD2 = 0x00008000, /* for future cksum types */
1517 OBD_FL_CKSUM_RSVD3 = 0x00010000, /* for future cksum types */
1518 OBD_FL_SHRINK_GRANT = 0x00020000, /* object shrink the grant */
1519 OBD_FL_MMAP = 0x00040000, /* object is mmapped on the client.
1520 * XXX: obsoleted - reserved for old
1521 * clients prior than 2.2 */
1522 OBD_FL_RECOV_RESEND = 0x00080000, /* recoverable resent */
1523 OBD_FL_NOSPC_BLK = 0x00100000, /* no more block space on OST */
1524 OBD_FL_FLUSH = 0x00200000, /* flush pages on the OST */
1525 OBD_FL_SHORT_IO = 0x00400000, /* short io request */
1527 /* Note that while these checksum values are currently separate bits,
1528 * in 2.x we can actually allow all values from 1-31 if we wanted. */
1529 OBD_FL_CKSUM_ALL = OBD_FL_CKSUM_CRC32 | OBD_FL_CKSUM_ADLER |
1530 OBD_FL_CKSUM_CRC32C,
1532 /* mask for local-only flag, which won't be sent over network */
1533 OBD_FL_LOCAL_MASK = 0xF0000000,
1536 #define LOV_MAGIC_V1 0x0BD10BD0
1537 #define LOV_MAGIC LOV_MAGIC_V1
1538 #define LOV_MAGIC_JOIN_V1 0x0BD20BD0
1539 #define LOV_MAGIC_V3 0x0BD30BD0
1542 * magic for fully defined striping
1543 * the idea is that we should have different magics for striping "hints"
1544 * (struct lov_user_md_v[13]) and defined ready-to-use striping (struct
1545 * lov_mds_md_v[13]). at the moment the magics are used in wire protocol,
1546 * we can't just change it w/o long way preparation, but we still need a
1547 * mechanism to allow LOD to differentiate hint versus ready striping.
1548 * so, at the moment we do a trick: MDT knows what to expect from request
1549 * depending on the case (replay uses ready striping, non-replay req uses
1550 * hints), so MDT replaces magic with appropriate one and now LOD can
1551 * easily understand what's inside -bzzz
1553 #define LOV_MAGIC_V1_DEF 0x0CD10BD0
1554 #define LOV_MAGIC_V3_DEF 0x0CD30BD0
1556 #define LOV_PATTERN_RAID0 0x001 /* stripes are used round-robin */
1557 #define LOV_PATTERN_RAID1 0x002 /* stripes are mirrors of each other */
1558 #define LOV_PATTERN_FIRST 0x100 /* first stripe is not in round-robin */
1559 #define LOV_PATTERN_CMOBD 0x200
1561 #define LOV_PATTERN_F_MASK 0xffff0000
1562 #define LOV_PATTERN_F_RELEASED 0x80000000 /* HSM released file */
1564 #define lov_pattern(pattern) (pattern & ~LOV_PATTERN_F_MASK)
1565 #define lov_pattern_flags(pattern) (pattern & LOV_PATTERN_F_MASK)
1567 #define lov_ost_data lov_ost_data_v1
1568 struct lov_ost_data_v1 { /* per-stripe data structure (little-endian)*/
1569 struct ost_id l_ost_oi; /* OST object ID */
1570 __u32 l_ost_gen; /* generation of this l_ost_idx */
1571 __u32 l_ost_idx; /* OST index in LOV (lov_tgt_desc->tgts) */
1574 #define lov_mds_md lov_mds_md_v1
1575 struct lov_mds_md_v1 { /* LOV EA mds/wire data (little-endian) */
1576 __u32 lmm_magic; /* magic number = LOV_MAGIC_V1 */
1577 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1578 struct ost_id lmm_oi; /* LOV object ID */
1579 __u32 lmm_stripe_size; /* size of stripe in bytes */
1580 /* lmm_stripe_count used to be __u32 */
1581 __u16 lmm_stripe_count; /* num stripes in use for this object */
1582 __u16 lmm_layout_gen; /* layout generation number */
1583 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1587 * Sigh, because pre-2.4 uses
1588 * struct lov_mds_md_v1 {
1590 * __u64 lmm_object_id;
1591 * __u64 lmm_object_seq;
1594 * to identify the LOV(MDT) object, and lmm_object_seq will
1595 * be normal_fid, which make it hard to combine these conversion
1596 * to ostid_to FID. so we will do lmm_oi/fid conversion separately
1598 * We can tell the lmm_oi by this way,
1599 * 1.8: lmm_object_id = {inode}, lmm_object_gr = 0
1600 * 2.1: lmm_object_id = {oid < 128k}, lmm_object_seq = FID_SEQ_NORMAL
1601 * 2.4: lmm_oi.f_seq = FID_SEQ_NORMAL, lmm_oi.f_oid = {oid < 128k},
1604 * But currently lmm_oi/lsm_oi does not have any "real" usages,
1605 * except for printing some information, and the user can always
1606 * get the real FID from LMA, besides this multiple case check might
1607 * make swab more complicate. So we will keep using id/seq for lmm_oi.
1610 static inline void fid_to_lmm_oi(const struct lu_fid *fid,
1613 oi->oi.oi_id = fid_oid(fid);
1614 oi->oi.oi_seq = fid_seq(fid);
1617 static inline void lmm_oi_set_seq(struct ost_id *oi, __u64 seq)
1619 oi->oi.oi_seq = seq;
1622 static inline __u64 lmm_oi_id(struct ost_id *oi)
1624 return oi->oi.oi_id;
1627 static inline __u64 lmm_oi_seq(struct ost_id *oi)
1629 return oi->oi.oi_seq;
1632 static inline void lmm_oi_le_to_cpu(struct ost_id *dst_oi,
1633 struct ost_id *src_oi)
1635 dst_oi->oi.oi_id = le64_to_cpu(src_oi->oi.oi_id);
1636 dst_oi->oi.oi_seq = le64_to_cpu(src_oi->oi.oi_seq);
1639 static inline void lmm_oi_cpu_to_le(struct ost_id *dst_oi,
1640 struct ost_id *src_oi)
1642 dst_oi->oi.oi_id = cpu_to_le64(src_oi->oi.oi_id);
1643 dst_oi->oi.oi_seq = cpu_to_le64(src_oi->oi.oi_seq);
1646 /* extern void lustre_swab_lov_mds_md(struct lov_mds_md *llm); */
1648 #define MAX_MD_SIZE (sizeof(struct lov_mds_md) + 4 * sizeof(struct lov_ost_data))
1649 #define MIN_MD_SIZE (sizeof(struct lov_mds_md) + 1 * sizeof(struct lov_ost_data))
1651 #define XATTR_NAME_ACL_ACCESS "system.posix_acl_access"
1652 #define XATTR_NAME_ACL_DEFAULT "system.posix_acl_default"
1653 #define XATTR_USER_PREFIX "user."
1654 #define XATTR_TRUSTED_PREFIX "trusted."
1655 #define XATTR_SECURITY_PREFIX "security."
1656 #define XATTR_LUSTRE_PREFIX "lustre."
1658 #define XATTR_NAME_LOV "trusted.lov"
1659 #define XATTR_NAME_LMA "trusted.lma"
1660 #define XATTR_NAME_LMV "trusted.lmv"
1661 #define XATTR_NAME_LINK "trusted.link"
1662 #define XATTR_NAME_FID "trusted.fid"
1663 #define XATTR_NAME_VERSION "trusted.version"
1664 #define XATTR_NAME_SOM "trusted.som"
1665 #define XATTR_NAME_HSM "trusted.hsm"
1666 #define XATTR_NAME_LFSCK_NAMESPACE "trusted.lfsck_namespace"
1668 struct lov_mds_md_v3 { /* LOV EA mds/wire data (little-endian) */
1669 __u32 lmm_magic; /* magic number = LOV_MAGIC_V3 */
1670 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1671 struct ost_id lmm_oi; /* LOV object ID */
1672 __u32 lmm_stripe_size; /* size of stripe in bytes */
1673 /* lmm_stripe_count used to be __u32 */
1674 __u16 lmm_stripe_count; /* num stripes in use for this object */
1675 __u16 lmm_layout_gen; /* layout generation number */
1676 char lmm_pool_name[LOV_MAXPOOLNAME]; /* must be 32bit aligned */
1677 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1680 static inline __u32 lov_mds_md_size(__u16 stripes, __u32 lmm_magic)
1682 if (lmm_magic == LOV_MAGIC_V3)
1683 return sizeof(struct lov_mds_md_v3) +
1684 stripes * sizeof(struct lov_ost_data_v1);
1686 return sizeof(struct lov_mds_md_v1) +
1687 stripes * sizeof(struct lov_ost_data_v1);
1691 #define OBD_MD_FLID (0x00000001ULL) /* object ID */
1692 #define OBD_MD_FLATIME (0x00000002ULL) /* access time */
1693 #define OBD_MD_FLMTIME (0x00000004ULL) /* data modification time */
1694 #define OBD_MD_FLCTIME (0x00000008ULL) /* change time */
1695 #define OBD_MD_FLSIZE (0x00000010ULL) /* size */
1696 #define OBD_MD_FLBLOCKS (0x00000020ULL) /* allocated blocks count */
1697 #define OBD_MD_FLBLKSZ (0x00000040ULL) /* block size */
1698 #define OBD_MD_FLMODE (0x00000080ULL) /* access bits (mode & ~S_IFMT) */
1699 #define OBD_MD_FLTYPE (0x00000100ULL) /* object type (mode & S_IFMT) */
1700 #define OBD_MD_FLUID (0x00000200ULL) /* user ID */
1701 #define OBD_MD_FLGID (0x00000400ULL) /* group ID */
1702 #define OBD_MD_FLFLAGS (0x00000800ULL) /* flags word */
1703 #define OBD_MD_FLNLINK (0x00002000ULL) /* link count */
1704 #define OBD_MD_FLGENER (0x00004000ULL) /* generation number */
1705 /*#define OBD_MD_FLINLINE (0x00008000ULL) inline data. used until 1.6.5 */
1706 #define OBD_MD_FLRDEV (0x00010000ULL) /* device number */
1707 #define OBD_MD_FLEASIZE (0x00020000ULL) /* extended attribute data */
1708 #define OBD_MD_LINKNAME (0x00040000ULL) /* symbolic link target */
1709 #define OBD_MD_FLHANDLE (0x00080000ULL) /* file/lock handle */
1710 #define OBD_MD_FLCKSUM (0x00100000ULL) /* bulk data checksum */
1711 #define OBD_MD_FLQOS (0x00200000ULL) /* quality of service stats */
1712 /*#define OBD_MD_FLOSCOPQ (0x00400000ULL) osc opaque data, never used */
1713 #define OBD_MD_FLCOOKIE (0x00800000ULL) /* log cancellation cookie */
1714 #define OBD_MD_FLGROUP (0x01000000ULL) /* group */
1715 #define OBD_MD_FLFID (0x02000000ULL) /* ->ost write inline fid */
1716 #define OBD_MD_FLEPOCH (0x04000000ULL) /* ->ost write with ioepoch */
1717 /* ->mds if epoch opens or closes */
1718 #define OBD_MD_FLGRANT (0x08000000ULL) /* ost preallocation space grant */
1719 #define OBD_MD_FLDIREA (0x10000000ULL) /* dir's extended attribute data */
1720 #define OBD_MD_FLUSRQUOTA (0x20000000ULL) /* over quota flags sent from ost */
1721 #define OBD_MD_FLGRPQUOTA (0x40000000ULL) /* over quota flags sent from ost */
1722 #define OBD_MD_FLMODEASIZE (0x80000000ULL) /* EA size will be changed */
1724 #define OBD_MD_MDS (0x0000000100000000ULL) /* where an inode lives on */
1725 #define OBD_MD_REINT (0x0000000200000000ULL) /* reintegrate oa */
1726 #define OBD_MD_MEA (0x0000000400000000ULL) /* CMD split EA */
1727 #define OBD_MD_TSTATE (0x0000000800000000ULL) /* transient state field */
1729 #define OBD_MD_FLXATTR (0x0000001000000000ULL) /* xattr */
1730 #define OBD_MD_FLXATTRLS (0x0000002000000000ULL) /* xattr list */
1731 #define OBD_MD_FLXATTRRM (0x0000004000000000ULL) /* xattr remove */
1732 #define OBD_MD_FLACL (0x0000008000000000ULL) /* ACL */
1733 #define OBD_MD_FLRMTPERM (0x0000010000000000ULL) /* remote permission */
1734 #define OBD_MD_FLMDSCAPA (0x0000020000000000ULL) /* MDS capability */
1735 #define OBD_MD_FLOSSCAPA (0x0000040000000000ULL) /* OSS capability */
1736 #define OBD_MD_FLCKSPLIT (0x0000080000000000ULL) /* Check split on server */
1737 #define OBD_MD_FLCROSSREF (0x0000100000000000ULL) /* Cross-ref case */
1738 #define OBD_MD_FLGETATTRLOCK (0x0000200000000000ULL) /* Get IOEpoch attributes
1739 * under lock; for xattr
1740 * requests means the
1741 * client holds the lock */
1742 #define OBD_MD_FLOBJCOUNT (0x0000400000000000ULL) /* for multiple destroy */
1744 #define OBD_MD_FLRMTLSETFACL (0x0001000000000000ULL) /* lfs lsetfacl case */
1745 #define OBD_MD_FLRMTLGETFACL (0x0002000000000000ULL) /* lfs lgetfacl case */
1746 #define OBD_MD_FLRMTRSETFACL (0x0004000000000000ULL) /* lfs rsetfacl case */
1747 #define OBD_MD_FLRMTRGETFACL (0x0008000000000000ULL) /* lfs rgetfacl case */
1749 #define OBD_MD_FLDATAVERSION (0x0010000000000000ULL) /* iversion sum */
1750 #define OBD_MD_FLRELEASED (0x0020000000000000ULL) /* file released */
1752 #define OBD_MD_FLGETATTR (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLMTIME | \
1753 OBD_MD_FLCTIME | OBD_MD_FLSIZE | OBD_MD_FLBLKSZ | \
1754 OBD_MD_FLMODE | OBD_MD_FLTYPE | OBD_MD_FLUID | \
1755 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLNLINK | \
1756 OBD_MD_FLGENER | OBD_MD_FLRDEV | OBD_MD_FLGROUP)
1758 #define OBD_MD_FLXATTRALL (OBD_MD_FLXATTR | OBD_MD_FLXATTRLS)
1760 /* don't forget obdo_fid which is way down at the bottom so it can
1761 * come after the definition of llog_cookie */
1765 HSS_CLEARMASK = 0x02,
1766 HSS_ARCHIVE_ID = 0x04,
1769 struct hsm_state_set {
1771 __u32 hss_archive_id;
1773 __u64 hss_clearmask;
1776 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
1777 extern void lustre_swab_hsm_state_set(struct hsm_state_set *hss);
1779 extern void lustre_swab_obd_statfs (struct obd_statfs *os);
1781 /* ost_body.data values for OST_BRW */
1783 #define OBD_BRW_READ 0x01
1784 #define OBD_BRW_WRITE 0x02
1785 #define OBD_BRW_RWMASK (OBD_BRW_READ | OBD_BRW_WRITE)
1786 #define OBD_BRW_SYNC 0x08 /* this page is a part of synchronous
1787 * transfer and is not accounted in
1789 #define OBD_BRW_CHECK 0x10
1790 #define OBD_BRW_FROM_GRANT 0x20 /* the osc manages this under llite */
1791 #define OBD_BRW_GRANTED 0x40 /* the ost manages this */
1792 #define OBD_BRW_NOCACHE 0x80 /* this page is a part of non-cached IO */
1793 #define OBD_BRW_NOQUOTA 0x100
1794 #define OBD_BRW_SRVLOCK 0x200 /* Client holds no lock over this page */
1795 #define OBD_BRW_ASYNC 0x400 /* Server may delay commit to disk */
1796 #define OBD_BRW_MEMALLOC 0x800 /* Client runs in the "kswapd" context */
1797 #define OBD_BRW_OVER_USRQUOTA 0x1000 /* Running out of user quota */
1798 #define OBD_BRW_OVER_GRPQUOTA 0x2000 /* Running out of group quota */
1799 #define OBD_BRW_SOFT_SYNC 0x4000 /* This flag notifies the server
1800 * that the client is running low on
1801 * space for unstable pages; asking
1802 * it to sync quickly */
1804 #define OBD_OBJECT_EOF 0xffffffffffffffffULL
1806 #define OST_MIN_PRECREATE 32
1807 #define OST_MAX_PRECREATE 20000
1810 struct ost_id ioo_oid; /* object ID, if multi-obj BRW */
1811 __u32 ioo_max_brw; /* low 16 bits were o_mode before 2.4,
1812 * now (PTLRPC_BULK_OPS_COUNT - 1) in
1813 * high 16 bits in 2.4 and later */
1814 __u32 ioo_bufcnt; /* number of niobufs for this object */
1817 #define IOOBJ_MAX_BRW_BITS 16
1818 #define IOOBJ_TYPE_MASK ((1U << IOOBJ_MAX_BRW_BITS) - 1)
1819 #define ioobj_max_brw_get(ioo) (((ioo)->ioo_max_brw >> IOOBJ_MAX_BRW_BITS) + 1)
1820 #define ioobj_max_brw_set(ioo, num) \
1821 do { (ioo)->ioo_max_brw = ((num) - 1) << IOOBJ_MAX_BRW_BITS; } while (0)
1823 extern void lustre_swab_obd_ioobj (struct obd_ioobj *ioo);
1825 /* multiple of 8 bytes => can array */
1826 struct niobuf_remote {
1832 extern void lustre_swab_niobuf_remote (struct niobuf_remote *nbr);
1834 /* lock value block communicated between the filter and llite */
1836 /* OST_LVB_ERR_INIT is needed because the return code in rc is
1837 * negative, i.e. because ((MASK + rc) & MASK) != MASK. */
1838 #define OST_LVB_ERR_INIT 0xffbadbad80000000ULL
1839 #define OST_LVB_ERR_MASK 0xffbadbad00000000ULL
1840 #define OST_LVB_IS_ERR(blocks) \
1841 ((blocks & OST_LVB_ERR_MASK) == OST_LVB_ERR_MASK)
1842 #define OST_LVB_SET_ERR(blocks, rc) \
1843 do { blocks = OST_LVB_ERR_INIT + rc; } while (0)
1844 #define OST_LVB_GET_ERR(blocks) (int)(blocks - OST_LVB_ERR_INIT)
1854 extern void lustre_swab_ost_lvb_v1(struct ost_lvb_v1 *lvb);
1868 extern void lustre_swab_ost_lvb(struct ost_lvb *lvb);
1871 * lquota data structures
1874 #ifndef QUOTABLOCK_BITS
1875 #define QUOTABLOCK_BITS 10
1878 #ifndef QUOTABLOCK_SIZE
1879 #define QUOTABLOCK_SIZE (1 << QUOTABLOCK_BITS)
1883 #define toqb(x) (((x) + QUOTABLOCK_SIZE - 1) >> QUOTABLOCK_BITS)
1886 /* The lquota_id structure is an union of all the possible identifier types that
1887 * can be used with quota, this includes:
1890 * - a FID which can be used for per-directory quota in the future */
1892 struct lu_fid qid_fid; /* FID for per-directory quota */
1893 __u64 qid_uid; /* user identifier */
1894 __u64 qid_gid; /* group identifier */
1897 /* quotactl management */
1898 struct obd_quotactl {
1900 __u32 qc_type; /* see Q_* flag below */
1903 struct obd_dqinfo qc_dqinfo;
1904 struct obd_dqblk qc_dqblk;
1907 extern void lustre_swab_obd_quotactl(struct obd_quotactl *q);
1909 #define Q_QUOTACHECK 0x800100 /* deprecated as of 2.4 */
1910 #define Q_INITQUOTA 0x800101 /* deprecated as of 2.4 */
1911 #define Q_GETOINFO 0x800102 /* get obd quota info */
1912 #define Q_GETOQUOTA 0x800103 /* get obd quotas */
1913 #define Q_FINVALIDATE 0x800104 /* deprecated as of 2.4 */
1915 #define Q_COPY(out, in, member) (out)->member = (in)->member
1917 #define QCTL_COPY(out, in) \
1919 Q_COPY(out, in, qc_cmd); \
1920 Q_COPY(out, in, qc_type); \
1921 Q_COPY(out, in, qc_id); \
1922 Q_COPY(out, in, qc_stat); \
1923 Q_COPY(out, in, qc_dqinfo); \
1924 Q_COPY(out, in, qc_dqblk); \
1927 /* Body of quota request used for quota acquire/release RPCs between quota
1928 * master (aka QMT) and slaves (ak QSD). */
1930 struct lu_fid qb_fid; /* FID of global index packing the pool ID
1931 * and type (data or metadata) as well as
1932 * the quota type (user or group). */
1933 union lquota_id qb_id; /* uid or gid or directory FID */
1934 __u32 qb_flags; /* see below */
1936 __u64 qb_count; /* acquire/release count (kbytes/inodes) */
1937 __u64 qb_usage; /* current slave usage (kbytes/inodes) */
1938 __u64 qb_slv_ver; /* slave index file version */
1939 struct lustre_handle qb_lockh; /* per-ID lock handle */
1940 struct lustre_handle qb_glb_lockh; /* global lock handle */
1941 __u64 qb_padding1[4];
1944 /* When the quota_body is used in the reply of quota global intent
1945 * lock (IT_QUOTA_CONN) reply, qb_fid contains slave index file FID. */
1946 #define qb_slv_fid qb_fid
1947 /* qb_usage is the current qunit (in kbytes/inodes) when quota_body is used in
1949 #define qb_qunit qb_usage
1951 #define QUOTA_DQACQ_FL_ACQ 0x1 /* acquire quota */
1952 #define QUOTA_DQACQ_FL_PREACQ 0x2 /* pre-acquire */
1953 #define QUOTA_DQACQ_FL_REL 0x4 /* release quota */
1954 #define QUOTA_DQACQ_FL_REPORT 0x8 /* report usage */
1956 extern void lustre_swab_quota_body(struct quota_body *b);
1958 /* Quota types currently supported */
1960 LQUOTA_TYPE_USR = 0x00, /* maps to USRQUOTA */
1961 LQUOTA_TYPE_GRP = 0x01, /* maps to GRPQUOTA */
1965 /* There are 2 different resource types on which a quota limit can be enforced:
1966 * - inodes on the MDTs
1967 * - blocks on the OSTs */
1969 LQUOTA_RES_MD = 0x01, /* skip 0 to avoid null oid in FID */
1970 LQUOTA_RES_DT = 0x02,
1972 LQUOTA_FIRST_RES = LQUOTA_RES_MD
1974 #define LQUOTA_NR_RES (LQUOTA_LAST_RES - LQUOTA_FIRST_RES + 1)
1977 * Space accounting support
1978 * Format of an accounting record, providing disk usage information for a given
1981 struct lquota_acct_rec { /* 16 bytes */
1982 __u64 bspace; /* current space in use */
1983 __u64 ispace; /* current # inodes in use */
1987 * Global quota index support
1988 * Format of a global record, providing global quota settings for a given quota
1991 struct lquota_glb_rec { /* 32 bytes */
1992 __u64 qbr_hardlimit; /* quota hard limit, in #inodes or kbytes */
1993 __u64 qbr_softlimit; /* quota soft limit, in #inodes or kbytes */
1994 __u64 qbr_time; /* grace time, in seconds */
1995 __u64 qbr_granted; /* how much is granted to slaves, in #inodes or
2000 * Slave index support
2001 * Format of a slave record, recording how much space is granted to a given
2004 struct lquota_slv_rec { /* 8 bytes */
2005 __u64 qsr_granted; /* space granted to the slave for the key=ID,
2006 * in #inodes or kbytes */
2009 /* Data structures associated with the quota locks */
2011 /* Glimpse descriptor used for the index & per-ID quota locks */
2012 struct ldlm_gl_lquota_desc {
2013 union lquota_id gl_id; /* quota ID subject to the glimpse */
2014 __u64 gl_flags; /* see LQUOTA_FL* below */
2015 __u64 gl_ver; /* new index version */
2016 __u64 gl_hardlimit; /* new hardlimit or qunit value */
2017 __u64 gl_softlimit; /* new softlimit */
2021 #define gl_qunit gl_hardlimit /* current qunit value used when
2022 * glimpsing per-ID quota locks */
2024 /* quota glimpse flags */
2025 #define LQUOTA_FL_EDQUOT 0x1 /* user/group out of quota space on QMT */
2027 /* LVB used with quota (global and per-ID) locks */
2029 __u64 lvb_flags; /* see LQUOTA_FL* above */
2030 __u64 lvb_id_may_rel; /* space that might be released later */
2031 __u64 lvb_id_rel; /* space released by the slave for this ID */
2032 __u64 lvb_id_qunit; /* current qunit value */
2036 extern void lustre_swab_lquota_lvb(struct lquota_lvb *lvb);
2038 /* LVB used with global quota lock */
2039 #define lvb_glb_ver lvb_id_may_rel /* current version of the global index */
2047 #define QUOTA_FIRST_OPC QUOTA_DQACQ
2056 MDS_GETATTR_NAME = 34,
2061 MDS_DISCONNECT = 39,
2067 MDS_DONE_WRITING = 45,
2069 MDS_QUOTACHECK = 47,
2072 MDS_SETXATTR = 50, /* obsolete, now it's MDS_REINT op */
2076 MDS_HSM_STATE_GET = 54,
2077 MDS_HSM_STATE_SET = 55,
2078 MDS_HSM_ACTION = 56,
2079 MDS_HSM_PROGRESS = 57,
2080 MDS_HSM_REQUEST = 58,
2081 MDS_HSM_CT_REGISTER = 59,
2082 MDS_HSM_CT_UNREGISTER = 60,
2083 MDS_SWAP_LAYOUTS = 61,
2087 #define MDS_FIRST_OPC MDS_GETATTR
2090 /* opcodes for object update */
2096 #define UPDATE_FIRST_OPC UPDATE_OBJ
2113 } mds_reint_t, mdt_reint_t;
2115 extern void lustre_swab_generic_32s (__u32 *val);
2117 /* the disposition of the intent outlines what was executed */
2118 #define DISP_IT_EXECD 0x00000001
2119 #define DISP_LOOKUP_EXECD 0x00000002
2120 #define DISP_LOOKUP_NEG 0x00000004
2121 #define DISP_LOOKUP_POS 0x00000008
2122 #define DISP_OPEN_CREATE 0x00000010
2123 #define DISP_OPEN_OPEN 0x00000020
2124 #define DISP_ENQ_COMPLETE 0x00400000 /* obsolete and unused */
2125 #define DISP_ENQ_OPEN_REF 0x00800000
2126 #define DISP_ENQ_CREATE_REF 0x01000000
2127 #define DISP_OPEN_LOCK 0x02000000
2128 #define DISP_OPEN_LEASE 0x04000000
2129 #define DISP_OPEN_STRIPE 0x08000000
2131 /* INODE LOCK PARTS */
2132 #define MDS_INODELOCK_LOOKUP 0x000001 /* For namespace, dentry etc, and also
2133 * was used to protect permission (mode,
2134 * owner, group etc) before 2.4. */
2135 #define MDS_INODELOCK_UPDATE 0x000002 /* size, links, timestamps */
2136 #define MDS_INODELOCK_OPEN 0x000004 /* For opened files */
2137 #define MDS_INODELOCK_LAYOUT 0x000008 /* for layout */
2139 /* The PERM bit is added int 2.4, and it is used to protect permission(mode,
2140 * owner, group, acl etc), so to separate the permission from LOOKUP lock.
2141 * Because for remote directories(in DNE), these locks will be granted by
2142 * different MDTs(different ldlm namespace).
2144 * For local directory, MDT will always grant UPDATE_LOCK|PERM_LOCK together.
2145 * For Remote directory, the master MDT, where the remote directory is, will
2146 * grant UPDATE_LOCK|PERM_LOCK, and the remote MDT, where the name entry is,
2147 * will grant LOOKUP_LOCK. */
2148 #define MDS_INODELOCK_PERM 0x000010
2149 #define MDS_INODELOCK_XATTR 0x000020 /* extended attributes */
2151 #define MDS_INODELOCK_MAXSHIFT 5
2152 /* This FULL lock is useful to take on unlink sort of operations */
2153 #define MDS_INODELOCK_FULL ((1<<(MDS_INODELOCK_MAXSHIFT+1))-1)
2155 extern void lustre_swab_ll_fid (struct ll_fid *fid);
2157 /* NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
2158 * but was moved into name[1] along with the OID to avoid consuming the
2159 * name[2,3] fields that need to be used for the quota id (also a FID). */
2161 LUSTRE_RES_ID_SEQ_OFF = 0,
2162 LUSTRE_RES_ID_VER_OID_OFF = 1,
2163 LUSTRE_RES_ID_WAS_VER_OFF = 2, /* see note above */
2164 LUSTRE_RES_ID_QUOTA_SEQ_OFF = 2,
2165 LUSTRE_RES_ID_QUOTA_VER_OID_OFF = 3,
2166 LUSTRE_RES_ID_HSH_OFF = 3
2169 #define MDS_STATUS_CONN 1
2170 #define MDS_STATUS_LOV 2
2172 /* mdt_thread_info.mti_flags. */
2174 /* The flag indicates Size-on-MDS attributes are changed. */
2175 MF_SOM_CHANGE = (1 << 0),
2176 /* Flags indicates an epoch opens or closes. */
2177 MF_EPOCH_OPEN = (1 << 1),
2178 MF_EPOCH_CLOSE = (1 << 2),
2179 MF_MDC_CANCEL_FID1 = (1 << 3),
2180 MF_MDC_CANCEL_FID2 = (1 << 4),
2181 MF_MDC_CANCEL_FID3 = (1 << 5),
2182 MF_MDC_CANCEL_FID4 = (1 << 6),
2183 /* There is a pending attribute update. */
2184 MF_SOM_AU = (1 << 7),
2185 /* Cancel OST locks while getattr OST attributes. */
2186 MF_GETATTR_LOCK = (1 << 8),
2187 MF_GET_MDT_IDX = (1 << 9),
2190 #define MF_SOM_LOCAL_FLAGS (MF_SOM_CHANGE | MF_EPOCH_OPEN | MF_EPOCH_CLOSE)
2192 #define LUSTRE_BFLAG_UNCOMMITTED_WRITES 0x1
2194 /* these should be identical to their EXT4_*_FL counterparts, they are
2195 * redefined here only to avoid dragging in fs/ext4/ext4.h */
2196 #define LUSTRE_SYNC_FL 0x00000008 /* Synchronous updates */
2197 #define LUSTRE_IMMUTABLE_FL 0x00000010 /* Immutable file */
2198 #define LUSTRE_APPEND_FL 0x00000020 /* writes to file may only append */
2199 #define LUSTRE_NOATIME_FL 0x00000080 /* do not update atime */
2200 #define LUSTRE_DIRSYNC_FL 0x00010000 /* dirsync behaviour (dir only) */
2203 /* Convert wire LUSTRE_*_FL to corresponding client local VFS S_* values
2204 * for the client inode i_flags. The LUSTRE_*_FL are the Lustre wire
2205 * protocol equivalents of LDISKFS_*_FL values stored on disk, while
2206 * the S_* flags are kernel-internal values that change between kernel
2207 * versions. These flags are set/cleared via FSFILT_IOC_{GET,SET}_FLAGS.
2208 * See b=16526 for a full history. */
2209 static inline int ll_ext_to_inode_flags(int flags)
2211 return (((flags & LUSTRE_SYNC_FL) ? S_SYNC : 0) |
2212 ((flags & LUSTRE_NOATIME_FL) ? S_NOATIME : 0) |
2213 ((flags & LUSTRE_APPEND_FL) ? S_APPEND : 0) |
2214 #if defined(S_DIRSYNC)
2215 ((flags & LUSTRE_DIRSYNC_FL) ? S_DIRSYNC : 0) |
2217 ((flags & LUSTRE_IMMUTABLE_FL) ? S_IMMUTABLE : 0));
2220 static inline int ll_inode_to_ext_flags(int iflags)
2222 return (((iflags & S_SYNC) ? LUSTRE_SYNC_FL : 0) |
2223 ((iflags & S_NOATIME) ? LUSTRE_NOATIME_FL : 0) |
2224 ((iflags & S_APPEND) ? LUSTRE_APPEND_FL : 0) |
2225 #if defined(S_DIRSYNC)
2226 ((iflags & S_DIRSYNC) ? LUSTRE_DIRSYNC_FL : 0) |
2228 ((iflags & S_IMMUTABLE) ? LUSTRE_IMMUTABLE_FL : 0));
2232 /* 64 possible states */
2233 enum md_transient_state {
2234 MS_RESTORE = (1 << 0), /* restore is running */
2240 struct lustre_handle handle;
2242 __u64 size; /* Offset, in the case of MDS_READPAGE */
2246 __u64 blocks; /* XID, in the case of MDS_READPAGE */
2248 __u64 t_state; /* transient file state defined in
2249 * enum md_transient_state
2250 * was "ino" until 2.4.0 */
2257 __u32 flags; /* from vfs for pin/unpin, LUSTRE_BFLAG close */
2259 __u32 nlink; /* #bytes to read in the case of MDS_READPAGE */
2260 __u32 unused2; /* was "generation" until 2.4.0 */
2265 __u32 max_cookiesize;
2266 __u32 uid_h; /* high 32-bits of uid, for FUID */
2267 __u32 gid_h; /* high 32-bits of gid, for FUID */
2268 __u32 padding_5; /* also fix lustre_swab_mdt_body */
2276 extern void lustre_swab_mdt_body (struct mdt_body *b);
2278 struct mdt_ioepoch {
2279 struct lustre_handle handle;
2285 extern void lustre_swab_mdt_ioepoch (struct mdt_ioepoch *b);
2287 /* permissions for md_perm.mp_perm */
2289 CFS_SETUID_PERM = 0x01,
2290 CFS_SETGID_PERM = 0x02,
2291 CFS_SETGRP_PERM = 0x04,
2292 CFS_RMTACL_PERM = 0x08,
2293 CFS_RMTOWN_PERM = 0x10
2296 /* inode access permission for remote user, the inode info are omitted,
2297 * for client knows them. */
2298 struct mdt_remote_perm {
2305 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
2309 extern void lustre_swab_mdt_remote_perm(struct mdt_remote_perm *p);
2311 struct mdt_rec_setattr {
2321 __u32 sa_padding_1_h;
2322 struct lu_fid sa_fid;
2331 __u32 sa_attr_flags;
2333 __u32 sa_bias; /* some operation flags */
2339 extern void lustre_swab_mdt_rec_setattr (struct mdt_rec_setattr *sa);
2342 * Attribute flags used in mdt_rec_setattr::sa_valid.
2343 * The kernel's #defines for ATTR_* should not be used over the network
2344 * since the client and MDS may run different kernels (see bug 13828)
2345 * Therefore, we should only use MDS_ATTR_* attributes for sa_valid.
2347 #define MDS_ATTR_MODE 0x1ULL /* = 1 */
2348 #define MDS_ATTR_UID 0x2ULL /* = 2 */
2349 #define MDS_ATTR_GID 0x4ULL /* = 4 */
2350 #define MDS_ATTR_SIZE 0x8ULL /* = 8 */
2351 #define MDS_ATTR_ATIME 0x10ULL /* = 16 */
2352 #define MDS_ATTR_MTIME 0x20ULL /* = 32 */
2353 #define MDS_ATTR_CTIME 0x40ULL /* = 64 */
2354 #define MDS_ATTR_ATIME_SET 0x80ULL /* = 128 */
2355 #define MDS_ATTR_MTIME_SET 0x100ULL /* = 256 */
2356 #define MDS_ATTR_FORCE 0x200ULL /* = 512, Not a change, but a change it */
2357 #define MDS_ATTR_ATTR_FLAG 0x400ULL /* = 1024 */
2358 #define MDS_ATTR_KILL_SUID 0x800ULL /* = 2048 */
2359 #define MDS_ATTR_KILL_SGID 0x1000ULL /* = 4096 */
2360 #define MDS_ATTR_CTIME_SET 0x2000ULL /* = 8192 */
2361 #define MDS_ATTR_FROM_OPEN 0x4000ULL /* = 16384, called from open path, ie O_TRUNC */
2362 #define MDS_ATTR_BLOCKS 0x8000ULL /* = 32768 */
2365 #define FMODE_READ 00000001
2366 #define FMODE_WRITE 00000002
2369 #define MDS_FMODE_CLOSED 00000000
2370 #define MDS_FMODE_EXEC 00000004
2371 /* IO Epoch is opened on a closed file. */
2372 #define MDS_FMODE_EPOCH 01000000
2373 /* IO Epoch is opened on a file truncate. */
2374 #define MDS_FMODE_TRUNC 02000000
2375 /* Size-on-MDS Attribute Update is pending. */
2376 #define MDS_FMODE_SOM 04000000
2378 #define MDS_OPEN_CREATED 00000010
2379 #define MDS_OPEN_CROSS 00000020
2381 #define MDS_OPEN_CREAT 00000100
2382 #define MDS_OPEN_EXCL 00000200
2383 #define MDS_OPEN_TRUNC 00001000
2384 #define MDS_OPEN_APPEND 00002000
2385 #define MDS_OPEN_SYNC 00010000
2386 #define MDS_OPEN_DIRECTORY 00200000
2388 #define MDS_OPEN_BY_FID 040000000 /* open_by_fid for known object */
2389 #define MDS_OPEN_DELAY_CREATE 0100000000 /* delay initial object create */
2390 #define MDS_OPEN_OWNEROVERRIDE 0200000000 /* NFSD rw-reopen ro file for owner */
2391 #define MDS_OPEN_JOIN_FILE 0400000000 /* open for join file.
2392 * We do not support JOIN FILE
2393 * anymore, reserve this flags
2394 * just for preventing such bit
2397 #define MDS_OPEN_LOCK 04000000000 /* This open requires open lock */
2398 #define MDS_OPEN_HAS_EA 010000000000 /* specify object create pattern */
2399 #define MDS_OPEN_HAS_OBJS 020000000000 /* Just set the EA the obj exist */
2400 #define MDS_OPEN_NORESTORE 0100000000000ULL /* Do not restore file at open */
2401 #define MDS_OPEN_NEWSTRIPE 0200000000000ULL /* New stripe needed (restripe or
2403 #define MDS_OPEN_VOLATILE 0400000000000ULL /* File is volatile = created
2405 #define MDS_OPEN_LEASE 01000000000000ULL /* Open the file and grant lease
2406 * delegation, succeed if it's not
2407 * being opened with conflict mode.
2409 #define MDS_OPEN_RELEASE 02000000000000ULL /* Open the file for HSM release */
2411 /* permission for create non-directory file */
2412 #define MAY_CREATE (1 << 7)
2413 /* permission for create directory file */
2414 #define MAY_LINK (1 << 8)
2415 /* permission for delete from the directory */
2416 #define MAY_UNLINK (1 << 9)
2417 /* source's permission for rename */
2418 #define MAY_RENAME_SRC (1 << 10)
2419 /* target's permission for rename */
2420 #define MAY_RENAME_TAR (1 << 11)
2421 /* part (parent's) VTX permission check */
2422 #define MAY_VTX_PART (1 << 12)
2423 /* full VTX permission check */
2424 #define MAY_VTX_FULL (1 << 13)
2425 /* lfs rgetfacl permission check */
2426 #define MAY_RGETFACL (1 << 14)
2429 MDS_CHECK_SPLIT = 1 << 0,
2430 MDS_CROSS_REF = 1 << 1,
2431 MDS_VTX_BYPASS = 1 << 2,
2432 MDS_PERM_BYPASS = 1 << 3,
2434 MDS_QUOTA_IGNORE = 1 << 5,
2435 /* Was MDS_CLOSE_CLEANUP (1 << 6), No more used */
2436 MDS_KEEP_ORPHAN = 1 << 7,
2437 MDS_RECOV_OPEN = 1 << 8,
2438 MDS_DATA_MODIFIED = 1 << 9,
2439 MDS_CREATE_VOLATILE = 1 << 10,
2440 MDS_OWNEROVERRIDE = 1 << 11,
2441 MDS_HSM_RELEASE = 1 << 12,
2444 /* instance of mdt_reint_rec */
2445 struct mdt_rec_create {
2453 __u32 cr_suppgid1_h;
2455 __u32 cr_suppgid2_h;
2456 struct lu_fid cr_fid1;
2457 struct lu_fid cr_fid2;
2458 struct lustre_handle cr_old_handle; /* handle in case of open replay */
2462 __u64 cr_padding_1; /* rr_blocks */
2465 /* use of helpers set/get_mrc_cr_flags() is needed to access
2466 * 64 bits cr_flags [cr_flags_l, cr_flags_h], this is done to
2467 * extend cr_flags size without breaking 1.8 compat */
2468 __u32 cr_flags_l; /* for use with open, low 32 bits */
2469 __u32 cr_flags_h; /* for use with open, high 32 bits */
2470 __u32 cr_umask; /* umask for create */
2471 __u32 cr_padding_4; /* rr_padding_4 */
2474 static inline void set_mrc_cr_flags(struct mdt_rec_create *mrc, __u64 flags)
2476 mrc->cr_flags_l = (__u32)(flags & 0xFFFFFFFFUll);
2477 mrc->cr_flags_h = (__u32)(flags >> 32);
2480 static inline __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
2482 return ((__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32));
2485 /* instance of mdt_reint_rec */
2486 struct mdt_rec_link {
2494 __u32 lk_suppgid1_h;
2496 __u32 lk_suppgid2_h;
2497 struct lu_fid lk_fid1;
2498 struct lu_fid lk_fid2;
2500 __u64 lk_padding_1; /* rr_atime */
2501 __u64 lk_padding_2; /* rr_ctime */
2502 __u64 lk_padding_3; /* rr_size */
2503 __u64 lk_padding_4; /* rr_blocks */
2505 __u32 lk_padding_5; /* rr_mode */
2506 __u32 lk_padding_6; /* rr_flags */
2507 __u32 lk_padding_7; /* rr_padding_2 */
2508 __u32 lk_padding_8; /* rr_padding_3 */
2509 __u32 lk_padding_9; /* rr_padding_4 */
2512 /* instance of mdt_reint_rec */
2513 struct mdt_rec_unlink {
2521 __u32 ul_suppgid1_h;
2523 __u32 ul_suppgid2_h;
2524 struct lu_fid ul_fid1;
2525 struct lu_fid ul_fid2;
2527 __u64 ul_padding_2; /* rr_atime */
2528 __u64 ul_padding_3; /* rr_ctime */
2529 __u64 ul_padding_4; /* rr_size */
2530 __u64 ul_padding_5; /* rr_blocks */
2533 __u32 ul_padding_6; /* rr_flags */
2534 __u32 ul_padding_7; /* rr_padding_2 */
2535 __u32 ul_padding_8; /* rr_padding_3 */
2536 __u32 ul_padding_9; /* rr_padding_4 */
2539 /* instance of mdt_reint_rec */
2540 struct mdt_rec_rename {
2548 __u32 rn_suppgid1_h;
2550 __u32 rn_suppgid2_h;
2551 struct lu_fid rn_fid1;
2552 struct lu_fid rn_fid2;
2554 __u64 rn_padding_1; /* rr_atime */
2555 __u64 rn_padding_2; /* rr_ctime */
2556 __u64 rn_padding_3; /* rr_size */
2557 __u64 rn_padding_4; /* rr_blocks */
2558 __u32 rn_bias; /* some operation flags */
2559 __u32 rn_mode; /* cross-ref rename has mode */
2560 __u32 rn_padding_5; /* rr_flags */
2561 __u32 rn_padding_6; /* rr_padding_2 */
2562 __u32 rn_padding_7; /* rr_padding_3 */
2563 __u32 rn_padding_8; /* rr_padding_4 */
2566 /* instance of mdt_reint_rec */
2567 struct mdt_rec_setxattr {
2575 __u32 sx_suppgid1_h;
2577 __u32 sx_suppgid2_h;
2578 struct lu_fid sx_fid;
2579 __u64 sx_padding_1; /* These three are rr_fid2 */
2584 __u64 sx_padding_5; /* rr_ctime */
2585 __u64 sx_padding_6; /* rr_size */
2586 __u64 sx_padding_7; /* rr_blocks */
2589 __u32 sx_padding_8; /* rr_flags */
2590 __u32 sx_padding_9; /* rr_padding_2 */
2591 __u32 sx_padding_10; /* rr_padding_3 */
2592 __u32 sx_padding_11; /* rr_padding_4 */
2596 * mdt_rec_reint is the template for all mdt_reint_xxx structures.
2597 * Do NOT change the size of various members, otherwise the value
2598 * will be broken in lustre_swab_mdt_rec_reint().
2600 * If you add new members in other mdt_reint_xxx structres and need to use the
2601 * rr_padding_x fields, then update lustre_swab_mdt_rec_reint() also.
2603 struct mdt_rec_reint {
2611 __u32 rr_suppgid1_h;
2613 __u32 rr_suppgid2_h;
2614 struct lu_fid rr_fid1;
2615 struct lu_fid rr_fid2;
2626 __u32 rr_padding_4; /* also fix lustre_swab_mdt_rec_reint */
2629 extern void lustre_swab_mdt_rec_reint(struct mdt_rec_reint *rr);
2631 /* lmv structures */
2633 __u32 ld_tgt_count; /* how many MDS's */
2634 __u32 ld_active_tgt_count; /* how many active */
2635 __u32 ld_default_stripe_count; /* how many objects are used */
2636 __u32 ld_pattern; /* default MEA_MAGIC_* */
2637 __u64 ld_default_hash_size;
2638 __u64 ld_padding_1; /* also fix lustre_swab_lmv_desc */
2639 __u32 ld_padding_2; /* also fix lustre_swab_lmv_desc */
2640 __u32 ld_qos_maxage; /* in second */
2641 __u32 ld_padding_3; /* also fix lustre_swab_lmv_desc */
2642 __u32 ld_padding_4; /* also fix lustre_swab_lmv_desc */
2643 struct obd_uuid ld_uuid;
2646 extern void lustre_swab_lmv_desc (struct lmv_desc *ld);
2648 /* TODO: lmv_stripe_md should contain mds capabilities for all slave fids */
2649 struct lmv_stripe_md {
2654 char mea_pool_name[LOV_MAXPOOLNAME];
2655 struct lu_fid mea_ids[0];
2658 extern void lustre_swab_lmv_stripe_md(struct lmv_stripe_md *mea);
2660 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
2661 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
2662 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
2664 #define MAX_HASH_SIZE_32 0x7fffffffUL
2665 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
2666 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
2672 FLD_FIRST_OPC = FLD_QUERY
2678 SEQ_FIRST_OPC = SEQ_QUERY
2682 SEQ_ALLOC_SUPER = 0,
2693 * LOV data structures
2696 #define LOV_MAX_UUID_BUFFER_SIZE 8192
2697 /* The size of the buffer the lov/mdc reserves for the
2698 * array of UUIDs returned by the MDS. With the current
2699 * protocol, this will limit the max number of OSTs per LOV */
2701 #define LOV_DESC_MAGIC 0xB0CCDE5C
2703 /* LOV settings descriptor (should only contain static info) */
2705 __u32 ld_tgt_count; /* how many OBD's */
2706 __u32 ld_active_tgt_count; /* how many active */
2707 __u32 ld_default_stripe_count; /* how many objects are used */
2708 __u32 ld_pattern; /* default PATTERN_RAID0 */
2709 __u64 ld_default_stripe_size; /* in bytes */
2710 __u64 ld_default_stripe_offset; /* in bytes */
2711 __u32 ld_padding_0; /* unused */
2712 __u32 ld_qos_maxage; /* in second */
2713 __u32 ld_padding_1; /* also fix lustre_swab_lov_desc */
2714 __u32 ld_padding_2; /* also fix lustre_swab_lov_desc */
2715 struct obd_uuid ld_uuid;
2718 #define ld_magic ld_active_tgt_count /* for swabbing from llogs */
2720 extern void lustre_swab_lov_desc (struct lov_desc *ld);
2725 /* opcodes -- MUST be distinct from OST/MDS opcodes */
2730 LDLM_BL_CALLBACK = 104,
2731 LDLM_CP_CALLBACK = 105,
2732 LDLM_GL_CALLBACK = 106,
2733 LDLM_SET_INFO = 107,
2736 #define LDLM_FIRST_OPC LDLM_ENQUEUE
2738 #define RES_NAME_SIZE 4
2739 struct ldlm_res_id {
2740 __u64 name[RES_NAME_SIZE];
2743 #define DLDLMRES "["LPX64":"LPX64":"LPX64"]."LPX64i
2744 #define PLDLMRES(res) (res)->lr_name.name[0], (res)->lr_name.name[1], \
2745 (res)->lr_name.name[2], (res)->lr_name.name[3]
2747 extern void lustre_swab_ldlm_res_id (struct ldlm_res_id *id);
2749 static inline int ldlm_res_eq(const struct ldlm_res_id *res0,
2750 const struct ldlm_res_id *res1)
2752 return !memcmp(res0, res1, sizeof(*res0));
2769 #define LCK_MODE_NUM 8
2779 #define LDLM_MIN_TYPE LDLM_PLAIN
2781 struct ldlm_extent {
2787 static inline int ldlm_extent_overlap(struct ldlm_extent *ex1,
2788 struct ldlm_extent *ex2)
2790 return (ex1->start <= ex2->end) && (ex2->start <= ex1->end);
2793 /* check if @ex1 contains @ex2 */
2794 static inline int ldlm_extent_contain(struct ldlm_extent *ex1,
2795 struct ldlm_extent *ex2)
2797 return (ex1->start <= ex2->start) && (ex1->end >= ex2->end);
2800 struct ldlm_inodebits {
2804 struct ldlm_flock_wire {
2812 /* it's important that the fields of the ldlm_extent structure match
2813 * the first fields of the ldlm_flock structure because there is only
2814 * one ldlm_swab routine to process the ldlm_policy_data_t union. if
2815 * this ever changes we will need to swab the union differently based
2816 * on the resource type. */
2819 struct ldlm_extent l_extent;
2820 struct ldlm_flock_wire l_flock;
2821 struct ldlm_inodebits l_inodebits;
2822 } ldlm_wire_policy_data_t;
2824 extern void lustre_swab_ldlm_policy_data (ldlm_wire_policy_data_t *d);
2826 union ldlm_gl_desc {
2827 struct ldlm_gl_lquota_desc lquota_desc;
2830 extern void lustre_swab_gl_desc(union ldlm_gl_desc *);
2832 struct ldlm_intent {
2836 extern void lustre_swab_ldlm_intent (struct ldlm_intent *i);
2838 struct ldlm_resource_desc {
2839 ldlm_type_t lr_type;
2840 __u32 lr_padding; /* also fix lustre_swab_ldlm_resource_desc */
2841 struct ldlm_res_id lr_name;
2844 extern void lustre_swab_ldlm_resource_desc (struct ldlm_resource_desc *r);
2846 struct ldlm_lock_desc {
2847 struct ldlm_resource_desc l_resource;
2848 ldlm_mode_t l_req_mode;
2849 ldlm_mode_t l_granted_mode;
2850 ldlm_wire_policy_data_t l_policy_data;
2853 extern void lustre_swab_ldlm_lock_desc (struct ldlm_lock_desc *l);
2855 #define LDLM_LOCKREQ_HANDLES 2
2856 #define LDLM_ENQUEUE_CANCEL_OFF 1
2858 struct ldlm_request {
2861 struct ldlm_lock_desc lock_desc;
2862 struct lustre_handle lock_handle[LDLM_LOCKREQ_HANDLES];
2865 extern void lustre_swab_ldlm_request (struct ldlm_request *rq);
2867 /* If LDLM_ENQUEUE, 1 slot is already occupied, 1 is available.
2868 * Otherwise, 2 are available. */
2869 #define ldlm_request_bufsize(count,type) \
2871 int _avail = LDLM_LOCKREQ_HANDLES; \
2872 _avail -= (type == LDLM_ENQUEUE ? LDLM_ENQUEUE_CANCEL_OFF : 0); \
2873 sizeof(struct ldlm_request) + \
2874 (count > _avail ? count - _avail : 0) * \
2875 sizeof(struct lustre_handle); \
2880 __u32 lock_padding; /* also fix lustre_swab_ldlm_reply */
2881 struct ldlm_lock_desc lock_desc;
2882 struct lustre_handle lock_handle;
2883 __u64 lock_policy_res1;
2884 __u64 lock_policy_res2;
2887 extern void lustre_swab_ldlm_reply (struct ldlm_reply *r);
2889 #define ldlm_flags_to_wire(flags) ((__u32)(flags))
2890 #define ldlm_flags_from_wire(flags) ((__u64)(flags))
2893 * Opcodes for mountconf (mgs and mgc)
2898 MGS_EXCEPTION, /* node died, etc. */
2899 MGS_TARGET_REG, /* whenever target starts up */
2905 #define MGS_FIRST_OPC MGS_CONNECT
2907 #define MGS_PARAM_MAXLEN 1024
2908 #define KEY_SET_INFO "set_info"
2910 struct mgs_send_param {
2911 char mgs_param[MGS_PARAM_MAXLEN];
2914 /* We pass this info to the MGS so it can write config logs */
2915 #define MTI_NAME_MAXLEN 64
2916 #define MTI_PARAM_MAXLEN 4096
2917 #define MTI_NIDS_MAX 32
2918 struct mgs_target_info {
2919 __u32 mti_lustre_ver;
2920 __u32 mti_stripe_index;
2921 __u32 mti_config_ver;
2923 __u32 mti_nid_count;
2924 __u32 mti_instance; /* Running instance of target */
2925 char mti_fsname[MTI_NAME_MAXLEN];
2926 char mti_svname[MTI_NAME_MAXLEN];
2927 char mti_uuid[sizeof(struct obd_uuid)];
2928 __u64 mti_nids[MTI_NIDS_MAX]; /* host nids (lnet_nid_t)*/
2929 char mti_params[MTI_PARAM_MAXLEN];
2931 extern void lustre_swab_mgs_target_info(struct mgs_target_info *oinfo);
2933 struct mgs_nidtbl_entry {
2934 __u64 mne_version; /* table version of this entry */
2935 __u32 mne_instance; /* target instance # */
2936 __u32 mne_index; /* target index */
2937 __u32 mne_length; /* length of this entry - by bytes */
2938 __u8 mne_type; /* target type LDD_F_SV_TYPE_OST/MDT */
2939 __u8 mne_nid_type; /* type of nid(mbz). for ipv6. */
2940 __u8 mne_nid_size; /* size of each NID, by bytes */
2941 __u8 mne_nid_count; /* # of NIDs in buffer */
2943 lnet_nid_t nids[0]; /* variable size buffer for NIDs. */
2946 extern void lustre_swab_mgs_nidtbl_entry(struct mgs_nidtbl_entry *oinfo);
2948 struct mgs_config_body {
2949 char mcb_name[MTI_NAME_MAXLEN]; /* logname */
2950 __u64 mcb_offset; /* next index of config log to request */
2951 __u16 mcb_type; /* type of log: CONFIG_T_[CONFIG|RECOVER] */
2953 __u8 mcb_bits; /* bits unit size of config log */
2954 __u32 mcb_units; /* # of units for bulk transfer */
2956 extern void lustre_swab_mgs_config_body(struct mgs_config_body *body);
2958 struct mgs_config_res {
2959 __u64 mcr_offset; /* index of last config log */
2960 __u64 mcr_size; /* size of the log */
2962 extern void lustre_swab_mgs_config_res(struct mgs_config_res *body);
2964 /* Config marker flags (in config log) */
2965 #define CM_START 0x01
2967 #define CM_SKIP 0x04
2968 #define CM_UPGRADE146 0x08
2969 #define CM_EXCLUDE 0x10
2970 #define CM_START_SKIP (CM_START | CM_SKIP)
2973 __u32 cm_step; /* aka config version */
2975 __u32 cm_vers; /* lustre release version number */
2976 __u32 cm_padding; /* 64 bit align */
2977 obd_time cm_createtime; /*when this record was first created */
2978 obd_time cm_canceltime; /*when this record is no longer valid*/
2979 char cm_tgtname[MTI_NAME_MAXLEN];
2980 char cm_comment[MTI_NAME_MAXLEN];
2983 extern void lustre_swab_cfg_marker(struct cfg_marker *marker,
2984 int swab, int size);
2987 * Opcodes for multiple servers.
2997 #define OBD_FIRST_OPC OBD_PING
2999 /* catalog of log objects */
3001 /** Identifier for a single log object */
3003 struct ost_id lgl_oi;
3005 } __attribute__((packed));
3007 /** Records written to the CATALOGS list */
3008 #define CATLIST "CATALOGS"
3010 struct llog_logid lci_logid;
3014 } __attribute__((packed));
3016 /* Log data record types - there is no specific reason that these need to
3017 * be related to the RPC opcodes, but no reason not to (may be handy later?)
3019 #define LLOG_OP_MAGIC 0x10600000
3020 #define LLOG_OP_MASK 0xfff00000
3023 LLOG_PAD_MAGIC = LLOG_OP_MAGIC | 0x00000,
3024 OST_SZ_REC = LLOG_OP_MAGIC | 0x00f00,
3025 /* OST_RAID1_REC = LLOG_OP_MAGIC | 0x01000, never used */
3026 MDS_UNLINK_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) |
3027 REINT_UNLINK, /* obsolete after 2.5.0 */
3028 MDS_UNLINK64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
3030 /* MDS_SETATTR_REC = LLOG_OP_MAGIC | 0x12401, obsolete 1.8.0 */
3031 MDS_SETATTR64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
3033 OBD_CFG_REC = LLOG_OP_MAGIC | 0x20000,
3034 /* PTL_CFG_REC = LLOG_OP_MAGIC | 0x30000, obsolete 1.4.0 */
3035 LLOG_GEN_REC = LLOG_OP_MAGIC | 0x40000,
3036 /* LLOG_JOIN_REC = LLOG_OP_MAGIC | 0x50000, obsolete 1.8.0 */
3037 CHANGELOG_REC = LLOG_OP_MAGIC | 0x60000,
3038 CHANGELOG_USER_REC = LLOG_OP_MAGIC | 0x70000,
3039 HSM_AGENT_REC = LLOG_OP_MAGIC | 0x80000,
3040 LLOG_HDR_MAGIC = LLOG_OP_MAGIC | 0x45539,
3041 LLOG_LOGID_MAGIC = LLOG_OP_MAGIC | 0x4553b,
3044 #define LLOG_REC_HDR_NEEDS_SWABBING(r) \
3045 (((r)->lrh_type & __swab32(LLOG_OP_MASK)) == __swab32(LLOG_OP_MAGIC))
3047 /** Log record header - stored in little endian order.
3048 * Each record must start with this struct, end with a llog_rec_tail,
3049 * and be a multiple of 256 bits in size.
3051 struct llog_rec_hdr {
3058 struct llog_rec_tail {
3063 /* Where data follow just after header */
3064 #define REC_DATA(ptr) \
3065 ((void *)((char *)ptr + sizeof(struct llog_rec_hdr)))
3067 #define REC_DATA_LEN(rec) \
3068 (rec->lrh_len - sizeof(struct llog_rec_hdr) - \
3069 sizeof(struct llog_rec_tail))
3071 struct llog_logid_rec {
3072 struct llog_rec_hdr lid_hdr;
3073 struct llog_logid lid_id;
3077 struct llog_rec_tail lid_tail;
3078 } __attribute__((packed));
3080 struct llog_unlink_rec {
3081 struct llog_rec_hdr lur_hdr;
3084 obd_count lur_count;
3085 struct llog_rec_tail lur_tail;
3086 } __attribute__((packed));
3088 struct llog_unlink64_rec {
3089 struct llog_rec_hdr lur_hdr;
3090 struct lu_fid lur_fid;
3091 obd_count lur_count; /* to destroy the lost precreated */
3095 struct llog_rec_tail lur_tail;
3096 } __attribute__((packed));
3098 struct llog_setattr64_rec {
3099 struct llog_rec_hdr lsr_hdr;
3100 struct ost_id lsr_oi;
3106 struct llog_rec_tail lsr_tail;
3107 } __attribute__((packed));
3109 struct llog_size_change_rec {
3110 struct llog_rec_hdr lsc_hdr;
3111 struct ll_fid lsc_fid;
3116 struct llog_rec_tail lsc_tail;
3117 } __attribute__((packed));
3119 #define CHANGELOG_MAGIC 0xca103000
3121 /** \a changelog_rec_type's that can't be masked */
3122 #define CHANGELOG_MINMASK (1 << CL_MARK)
3123 /** bits covering all \a changelog_rec_type's */
3124 #define CHANGELOG_ALLMASK 0XFFFFFFFF
3125 /** default \a changelog_rec_type mask */
3126 #define CHANGELOG_DEFMASK CHANGELOG_ALLMASK & ~(1 << CL_ATIME | 1 << CL_CLOSE)
3128 /* changelog llog name, needed by client replicators */
3129 #define CHANGELOG_CATALOG "changelog_catalog"
3131 struct changelog_setinfo {
3134 } __attribute__((packed));
3136 /** changelog record */
3137 struct llog_changelog_rec {
3138 struct llog_rec_hdr cr_hdr;
3139 struct changelog_rec cr;
3140 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
3141 } __attribute__((packed));
3143 struct llog_changelog_ext_rec {
3144 struct llog_rec_hdr cr_hdr;
3145 struct changelog_ext_rec cr;
3146 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
3147 } __attribute__((packed));
3149 #define CHANGELOG_USER_PREFIX "cl"
3151 struct llog_changelog_user_rec {
3152 struct llog_rec_hdr cur_hdr;
3156 struct llog_rec_tail cur_tail;
3157 } __attribute__((packed));
3159 enum agent_req_status {
3167 static inline char *agent_req_status2name(enum agent_req_status ars)
3185 static inline bool agent_req_in_final_state(enum agent_req_status ars)
3187 return ((ars == ARS_SUCCEED) || (ars == ARS_FAILED) ||
3188 (ars == ARS_CANCELED));
3191 struct llog_agent_req_rec {
3192 struct llog_rec_hdr arr_hdr; /**< record header */
3193 __u32 arr_status; /**< status of the request */
3195 * agent_req_status */
3196 __u32 arr_archive_id; /**< backend archive number */
3197 __u64 arr_flags; /**< req flags */
3198 __u64 arr_compound_id; /**< compound cookie */
3199 __u64 arr_req_create; /**< req. creation time */
3200 __u64 arr_req_change; /**< req. status change time */
3201 struct hsm_action_item arr_hai; /**< req. to the agent */
3202 struct llog_rec_tail arr_tail; /**< record tail for_sizezof_only */
3203 } __attribute__((packed));
3205 /* Old llog gen for compatibility */
3209 } __attribute__((packed));
3211 struct llog_gen_rec {
3212 struct llog_rec_hdr lgr_hdr;
3213 struct llog_gen lgr_gen;
3217 struct llog_rec_tail lgr_tail;
3220 /* On-disk header structure of each log object, stored in little endian order */
3221 #define LLOG_CHUNK_SIZE 8192
3222 #define LLOG_HEADER_SIZE (96)
3223 #define LLOG_BITMAP_BYTES (LLOG_CHUNK_SIZE - LLOG_HEADER_SIZE)
3225 #define LLOG_MIN_REC_SIZE (24) /* round(llog_rec_hdr + llog_rec_tail) */
3227 /* flags for the logs */
3229 LLOG_F_ZAP_WHEN_EMPTY = 0x1,
3230 LLOG_F_IS_CAT = 0x2,
3231 LLOG_F_IS_PLAIN = 0x4,
3234 struct llog_log_hdr {
3235 struct llog_rec_hdr llh_hdr;
3236 obd_time llh_timestamp;
3238 __u32 llh_bitmap_offset;
3242 /* for a catalog the first plain slot is next to it */
3243 struct obd_uuid llh_tgtuuid;
3244 __u32 llh_reserved[LLOG_HEADER_SIZE/sizeof(__u32) - 23];
3245 __u32 llh_bitmap[LLOG_BITMAP_BYTES/sizeof(__u32)];
3246 struct llog_rec_tail llh_tail;
3247 } __attribute__((packed));
3249 #define LLOG_BITMAP_SIZE(llh) (__u32)((llh->llh_hdr.lrh_len - \
3250 llh->llh_bitmap_offset - \
3251 sizeof(llh->llh_tail)) * 8)
3253 /** log cookies are used to reference a specific log file and a record therein */
3254 struct llog_cookie {
3255 struct llog_logid lgc_lgl;
3259 } __attribute__((packed));
3261 /** llog protocol */
3262 enum llogd_rpc_ops {
3263 LLOG_ORIGIN_HANDLE_CREATE = 501,
3264 LLOG_ORIGIN_HANDLE_NEXT_BLOCK = 502,
3265 LLOG_ORIGIN_HANDLE_READ_HEADER = 503,
3266 LLOG_ORIGIN_HANDLE_WRITE_REC = 504,
3267 LLOG_ORIGIN_HANDLE_CLOSE = 505,
3268 LLOG_ORIGIN_CONNECT = 506,
3269 LLOG_CATINFO = 507, /* deprecated */
3270 LLOG_ORIGIN_HANDLE_PREV_BLOCK = 508,
3271 LLOG_ORIGIN_HANDLE_DESTROY = 509, /* for destroy llog object*/
3273 LLOG_FIRST_OPC = LLOG_ORIGIN_HANDLE_CREATE
3277 struct llog_logid lgd_logid;
3279 __u32 lgd_llh_flags;
3281 __u32 lgd_saved_index;
3283 __u64 lgd_cur_offset;
3284 } __attribute__((packed));
3286 struct llogd_conn_body {
3287 struct llog_gen lgdc_gen;
3288 struct llog_logid lgdc_logid;
3289 __u32 lgdc_ctxt_idx;
3290 } __attribute__((packed));
3292 /* Note: 64-bit types are 64-bit aligned in structure */
3294 obd_valid o_valid; /* hot fields in this obdo */
3296 obd_id o_parent_seq;
3297 obd_size o_size; /* o_size-o_blocks == ost_lvb */
3301 obd_blocks o_blocks; /* brw: cli sent cached bytes */
3304 /* 32-bit fields start here: keep an even number of them via padding */
3305 obd_blksize o_blksize; /* optimal IO blocksize */
3306 obd_mode o_mode; /* brw: cli sent cache remain */
3310 obd_count o_nlink; /* brw: checksum */
3311 obd_count o_parent_oid;
3312 obd_count o_misc; /* brw: o_dropped */
3314 __u64 o_ioepoch; /* epoch in ost writes */
3315 __u32 o_stripe_idx; /* holds stripe idx */
3317 struct lustre_handle o_handle; /* brw: lock handle to prolong
3319 struct llog_cookie o_lcookie; /* destroy: unlink cookie from
3324 __u64 o_data_version; /* getattr: sum of iversion for
3326 * brw: grant space consumed on
3327 * the client for the write */
3333 #define o_dirty o_blocks
3334 #define o_undirty o_mode
3335 #define o_dropped o_misc
3336 #define o_cksum o_nlink
3337 #define o_grant_used o_data_version
3339 static inline void lustre_set_wire_obdo(struct obd_connect_data *ocd,
3341 const struct obdo *lobdo)
3344 wobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
3348 if (unlikely(!(ocd->ocd_connect_flags & OBD_CONNECT_FID)) &&
3349 fid_seq_is_echo(ostid_seq(&lobdo->o_oi))) {
3350 /* Currently OBD_FL_OSTID will only be used when 2.4 echo
3351 * client communicate with pre-2.4 server */
3352 wobdo->o_oi.oi.oi_id = fid_oid(&lobdo->o_oi.oi_fid);
3353 wobdo->o_oi.oi.oi_seq = fid_seq(&lobdo->o_oi.oi_fid);
3357 static inline void lustre_get_wire_obdo(struct obd_connect_data *ocd,
3359 const struct obdo *wobdo)
3361 obd_flag local_flags = 0;
3363 if (lobdo->o_valid & OBD_MD_FLFLAGS)
3364 local_flags = lobdo->o_flags & OBD_FL_LOCAL_MASK;
3367 if (local_flags != 0) {
3368 lobdo->o_valid |= OBD_MD_FLFLAGS;
3369 lobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
3370 lobdo->o_flags |= local_flags;
3375 if (unlikely(!(ocd->ocd_connect_flags & OBD_CONNECT_FID)) &&
3376 fid_seq_is_echo(wobdo->o_oi.oi.oi_seq)) {
3378 lobdo->o_oi.oi_fid.f_seq = wobdo->o_oi.oi.oi_seq;
3379 lobdo->o_oi.oi_fid.f_oid = wobdo->o_oi.oi.oi_id;
3380 lobdo->o_oi.oi_fid.f_ver = 0;
3384 extern void lustre_swab_obdo (struct obdo *o);
3386 /* request structure for OST's */
3391 /* Key for FIEMAP to be used in get_info calls */
3392 struct ll_fiemap_info_key {
3395 struct ll_user_fiemap fiemap;
3398 extern void lustre_swab_ost_body (struct ost_body *b);
3399 extern void lustre_swab_ost_last_id(obd_id *id);
3400 extern void lustre_swab_fiemap(struct ll_user_fiemap *fiemap);
3402 extern void lustre_swab_lov_user_md_v1(struct lov_user_md_v1 *lum);
3403 extern void lustre_swab_lov_user_md_v3(struct lov_user_md_v3 *lum);
3404 extern void lustre_swab_lov_user_md_objects(struct lov_user_ost_data *lod,
3406 extern void lustre_swab_lov_mds_md(struct lov_mds_md *lmm);
3409 extern void lustre_swab_llogd_body (struct llogd_body *d);
3410 extern void lustre_swab_llog_hdr (struct llog_log_hdr *h);
3411 extern void lustre_swab_llogd_conn_body (struct llogd_conn_body *d);
3412 extern void lustre_swab_llog_rec(struct llog_rec_hdr *rec);
3413 extern void lustre_swab_llog_id(struct llog_logid *lid);
3416 extern void lustre_swab_lustre_cfg(struct lustre_cfg *lcfg);
3418 /* Functions for dumping PTLRPC fields */
3419 void dump_rniobuf(struct niobuf_remote *rnb);
3420 void dump_ioo(struct obd_ioobj *nb);
3421 void dump_obdo(struct obdo *oa);
3422 void dump_ost_body(struct ost_body *ob);
3423 void dump_rcs(__u32 *rc);
3425 #define IDX_INFO_MAGIC 0x3D37CC37
3427 /* Index file transfer through the network. The server serializes the index into
3428 * a byte stream which is sent to the client via a bulk transfer */
3432 /* reply: see idx_info_flags below */
3435 /* request & reply: number of lu_idxpage (to be) transferred */
3439 /* request: requested attributes passed down to the iterator API */
3442 /* request & reply: index file identifier (FID) */
3443 struct lu_fid ii_fid;
3445 /* reply: version of the index file before starting to walk the index.
3446 * Please note that the version can be modified at any time during the
3450 /* request: hash to start with:
3451 * reply: hash of the first entry of the first lu_idxpage and hash
3452 * of the entry to read next if any */
3453 __u64 ii_hash_start;
3456 /* reply: size of keys in lu_idxpages, minimal one if II_FL_VARKEY is
3460 /* reply: size of records in lu_idxpages, minimal one if II_FL_VARREC
3468 extern void lustre_swab_idx_info(struct idx_info *ii);
3470 #define II_END_OFF MDS_DIR_END_OFF /* all entries have been read */
3472 /* List of flags used in idx_info::ii_flags */
3473 enum idx_info_flags {
3474 II_FL_NOHASH = 1 << 0, /* client doesn't care about hash value */
3475 II_FL_VARKEY = 1 << 1, /* keys can be of variable size */
3476 II_FL_VARREC = 1 << 2, /* records can be of variable size */
3477 II_FL_NONUNQ = 1 << 3, /* index supports non-unique keys */
3480 #define LIP_MAGIC 0x8A6D6B6C
3482 /* 4KB (= LU_PAGE_SIZE) container gathering key/record pairs */
3484 /* 16-byte header */
3487 __u16 lip_nr; /* number of entries in the container */
3488 __u64 lip_pad0; /* additional padding for future use */
3490 /* key/record pairs are stored in the remaining 4080 bytes.
3491 * depending upon the flags in idx_info::ii_flags, each key/record
3492 * pair might be preceded by:
3494 * - the key size (II_FL_VARKEY is set)
3495 * - the record size (II_FL_VARREC is set)
3497 * For the time being, we only support fixed-size key & record. */
3498 char lip_entries[0];
3500 extern void lustre_swab_lip_header(struct lu_idxpage *lip);
3502 #define LIP_HDR_SIZE (offsetof(struct lu_idxpage, lip_entries))
3504 /* Gather all possible type associated with a 4KB container */
3506 struct lu_dirpage lp_dir; /* for MDS_READPAGE */
3507 struct lu_idxpage lp_idx; /* for OBD_IDX_READ */
3508 char lp_array[LU_PAGE_SIZE];
3511 /* security opcodes */
3514 SEC_CTX_INIT_CONT = 802,
3517 SEC_FIRST_OPC = SEC_CTX_INIT
3521 * capa related definitions
3523 #define CAPA_HMAC_MAX_LEN 64
3524 #define CAPA_HMAC_KEY_MAX_LEN 56
3526 /* NB take care when changing the sequence of elements this struct,
3527 * because the offset info is used in find_capa() */
3528 struct lustre_capa {
3529 struct lu_fid lc_fid; /** fid */
3530 __u64 lc_opc; /** operations allowed */
3531 __u64 lc_uid; /** file owner */
3532 __u64 lc_gid; /** file group */
3533 __u32 lc_flags; /** HMAC algorithm & flags */
3534 __u32 lc_keyid; /** key# used for the capability */
3535 __u32 lc_timeout; /** capa timeout value (sec) */
3536 __u32 lc_expiry; /** expiry time (sec) */
3537 __u8 lc_hmac[CAPA_HMAC_MAX_LEN]; /** HMAC */
3538 } __attribute__((packed));
3540 extern void lustre_swab_lustre_capa(struct lustre_capa *c);
3542 /** lustre_capa::lc_opc */
3544 CAPA_OPC_BODY_WRITE = 1<<0, /**< write object data */
3545 CAPA_OPC_BODY_READ = 1<<1, /**< read object data */
3546 CAPA_OPC_INDEX_LOOKUP = 1<<2, /**< lookup object fid */
3547 CAPA_OPC_INDEX_INSERT = 1<<3, /**< insert object fid */
3548 CAPA_OPC_INDEX_DELETE = 1<<4, /**< delete object fid */
3549 CAPA_OPC_OSS_WRITE = 1<<5, /**< write oss object data */
3550 CAPA_OPC_OSS_READ = 1<<6, /**< read oss object data */
3551 CAPA_OPC_OSS_TRUNC = 1<<7, /**< truncate oss object */
3552 CAPA_OPC_OSS_DESTROY = 1<<8, /**< destroy oss object */
3553 CAPA_OPC_META_WRITE = 1<<9, /**< write object meta data */
3554 CAPA_OPC_META_READ = 1<<10, /**< read object meta data */
3557 #define CAPA_OPC_OSS_RW (CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE)
3558 #define CAPA_OPC_MDS_ONLY \
3559 (CAPA_OPC_BODY_WRITE | CAPA_OPC_BODY_READ | CAPA_OPC_INDEX_LOOKUP | \
3560 CAPA_OPC_INDEX_INSERT | CAPA_OPC_INDEX_DELETE)
3561 #define CAPA_OPC_OSS_ONLY \
3562 (CAPA_OPC_OSS_WRITE | CAPA_OPC_OSS_READ | CAPA_OPC_OSS_TRUNC | \
3563 CAPA_OPC_OSS_DESTROY)
3564 #define CAPA_OPC_MDS_DEFAULT ~CAPA_OPC_OSS_ONLY
3565 #define CAPA_OPC_OSS_DEFAULT ~(CAPA_OPC_MDS_ONLY | CAPA_OPC_OSS_ONLY)
3567 /* MDS capability covers object capability for operations of body r/w
3568 * (dir readpage/sendpage), index lookup/insert/delete and meta data r/w,
3569 * while OSS capability only covers object capability for operations of
3570 * oss data(file content) r/w/truncate.
3572 static inline int capa_for_mds(struct lustre_capa *c)
3574 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) != 0;
3577 static inline int capa_for_oss(struct lustre_capa *c)
3579 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) == 0;
3582 /* lustre_capa::lc_hmac_alg */
3584 CAPA_HMAC_ALG_SHA1 = 1, /**< sha1 algorithm */
3588 #define CAPA_FL_MASK 0x00ffffff
3589 #define CAPA_HMAC_ALG_MASK 0xff000000
3591 struct lustre_capa_key {
3592 __u64 lk_seq; /**< mds# */
3593 __u32 lk_keyid; /**< key# */
3595 __u8 lk_key[CAPA_HMAC_KEY_MAX_LEN]; /**< key */
3596 } __attribute__((packed));
3598 extern void lustre_swab_lustre_capa_key(struct lustre_capa_key *k);
3600 /** The link ea holds 1 \a link_ea_entry for each hardlink */
3601 #define LINK_EA_MAGIC 0x11EAF1DFUL
3602 struct link_ea_header {
3605 __u64 leh_len; /* total size */
3611 /** Hardlink data is name and parent fid.
3612 * Stored in this crazy struct for maximum packing and endian-neutrality
3614 struct link_ea_entry {
3615 /** __u16 stored big-endian, unaligned */
3616 unsigned char lee_reclen[2];
3617 unsigned char lee_parent_fid[sizeof(struct lu_fid)];
3619 }__attribute__((packed));
3621 /** fid2path request/reply structure */
3622 struct getinfo_fid2path {
3623 struct lu_fid gf_fid;
3628 } __attribute__((packed));
3630 void lustre_swab_fid2path (struct getinfo_fid2path *gf);
3633 LAYOUT_INTENT_ACCESS = 0,
3634 LAYOUT_INTENT_READ = 1,
3635 LAYOUT_INTENT_WRITE = 2,
3636 LAYOUT_INTENT_GLIMPSE = 3,
3637 LAYOUT_INTENT_TRUNC = 4,
3638 LAYOUT_INTENT_RELEASE = 5,
3639 LAYOUT_INTENT_RESTORE = 6
3642 /* enqueue layout lock with intent */
3643 struct layout_intent {
3644 __u32 li_opc; /* intent operation for enqueue, read, write etc */
3650 void lustre_swab_layout_intent(struct layout_intent *li);
3653 * On the wire version of hsm_progress structure.
3655 * Contains the userspace hsm_progress and some internal fields.
3657 struct hsm_progress_kernel {
3658 /* Field taken from struct hsm_progress */
3661 struct hsm_extent hpk_extent;
3663 __u16 hpk_errval; /* positive val */
3665 /* Additional fields */
3666 __u64 hpk_data_version;
3668 } __attribute__((packed));
3670 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
3671 extern void lustre_swab_hsm_current_action(struct hsm_current_action *action);
3672 extern void lustre_swab_hsm_progress_kernel(struct hsm_progress_kernel *hpk);
3673 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
3674 extern void lustre_swab_hsm_user_item(struct hsm_user_item *hui);
3675 extern void lustre_swab_hsm_request(struct hsm_request *hr);
3678 * These are object update opcode under UPDATE_OBJ, which is currently
3679 * being used by cross-ref operations between MDT.
3681 * During the cross-ref operation, the Master MDT, which the client send the
3682 * request to, will disassembly the operation into object updates, then OSP
3683 * will send these updates to the remote MDT to be executed.
3685 * Update request format
3686 * magic: UPDATE_BUFFER_MAGIC_V1
3687 * Count: How many updates in the req.
3688 * bufs[0] : following are packets of object.
3690 * type: object_update_op, the op code of update
3691 * fid: The object fid of the update.
3692 * lens/bufs: other parameters of the update.
3694 * type: object_update_op, the op code of update
3695 * fid: The object fid of the update.
3696 * lens/bufs: other parameters of the update.
3698 * update[7]: type: object_update_op, the op code of update
3699 * fid: The object fid of the update.
3700 * lens/bufs: other parameters of the update.
3701 * Current 8 maxim updates per object update request.
3703 *******************************************************************
3704 * update reply format:
3706 * ur_version: UPDATE_REPLY_V1
3707 * ur_count: The count of the reply, which is usually equal
3708 * to the number of updates in the request.
3709 * ur_lens: The reply lengths of each object update.
3711 * replies: 1st update reply [4bytes_ret: other body]
3712 * 2nd update reply [4bytes_ret: other body]
3714 * nth update reply [4bytes_ret: other body]
3716 * For each reply of the update, the format would be
3717 * result(4 bytes):Other stuff
3720 #define UPDATE_MAX_OPS 10
3721 #define UPDATE_BUFFER_MAGIC_V1 0xBDDE0001
3722 #define UPDATE_BUFFER_MAGIC UPDATE_BUFFER_MAGIC_V1
3723 #define UPDATE_BUF_COUNT 8
3724 enum object_update_op {
3733 OBJ_INDEX_LOOKUP = 9,
3734 OBJ_INDEX_INSERT = 10,
3735 OBJ_INDEX_DELETE = 11,
3742 struct lu_fid u_fid;
3743 __u32 u_lens[UPDATE_BUF_COUNT];
3753 #define UPDATE_REPLY_V1 0x00BD0001
3754 struct update_reply {
3760 void lustre_swab_update_buf(struct update_buf *ub);
3761 void lustre_swab_update_reply_buf(struct update_reply *ur);
3763 /** layout swap request structure
3764 * fid1 and fid2 are in mdt_body
3766 struct mdc_swap_layouts {
3770 void lustre_swab_swap_layouts(struct mdc_swap_layouts *msl);
3773 struct lustre_handle cd_handle;
3774 struct lu_fid cd_fid;
3775 __u64 cd_data_version;
3776 __u64 cd_reserved[8];
3779 void lustre_swab_close_data(struct close_data *data);