1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/include/lustre/lustre_idl.h
38 * Lustre wire protocol definitions.
41 /** \defgroup lustreidl lustreidl
43 * Lustre wire protocol definitions.
45 * ALL structs passing over the wire should be declared here. Structs
46 * that are used in interfaces with userspace should go in lustre_user.h.
48 * All structs being declared here should be built from simple fixed-size
49 * types (__u8, __u16, __u32, __u64) or be built from other types or
50 * structs also declared in this file. Similarly, all flags and magic
51 * values in those structs should also be declared here. This ensures
52 * that the Lustre wire protocol is not influenced by external dependencies.
54 * The only other acceptable items in this file are VERY SIMPLE accessor
55 * functions to avoid callers grubbing inside the structures, and the
56 * prototypes of the swabber functions for each struct. Nothing that
57 * depends on external functions or definitions should be in here.
59 * Structs must be properly aligned to put 64-bit values on an 8-byte
60 * boundary. Any structs being added here must also be added to
61 * utils/wirecheck.c and "make newwiretest" run to regenerate the
62 * utils/wiretest.c sources. This allows us to verify that wire structs
63 * have the proper alignment/size on all architectures.
65 * DO NOT CHANGE any of the structs, flags, values declared here and used
66 * in released Lustre versions. Some structs may have padding fields that
67 * can be used. Some structs might allow addition at the end (verify this
68 * in the code to ensure that new/old clients that see this larger struct
69 * do not fail, otherwise you need to implement protocol compatibility).
71 * We assume all nodes are either little-endian or big-endian, and we
72 * always send messages in the sender's native format. The receiver
73 * detects the message format by checking the 'magic' field of the message
74 * (see lustre_msg_swabbed() below).
76 * Each wire type has corresponding 'lustre_swab_xxxtypexxx()' routines,
77 * implemented either here, inline (trivial implementations) or in
78 * ptlrpc/pack_generic.c. These 'swabbers' convert the type from "other"
79 * endian, in-place in the message buffer.
81 * A swabber takes a single pointer argument. The caller must already have
82 * verified that the length of the message buffer >= sizeof (type).
84 * For variable length types, a second 'lustre_swab_v_xxxtypexxx()' routine
85 * may be defined that swabs just the variable part, after the caller has
86 * verified that the message buffer is large enough.
91 #ifndef _LUSTRE_IDL_H_
92 #define _LUSTRE_IDL_H_
94 #include <libcfs/libcfs.h>
96 /* Defn's shared with user-space. */
97 #include <lustre/lustre_user.h>
98 #include <lustre/ll_fiemap.h>
103 /* FOO_REQUEST_PORTAL is for incoming requests on the FOO
104 * FOO_REPLY_PORTAL is for incoming replies on the FOO
105 * FOO_BULK_PORTAL is for incoming bulk on the FOO
108 #define CONNMGR_REQUEST_PORTAL 1
109 #define CONNMGR_REPLY_PORTAL 2
110 //#define OSC_REQUEST_PORTAL 3
111 #define OSC_REPLY_PORTAL 4
112 //#define OSC_BULK_PORTAL 5
113 #define OST_IO_PORTAL 6
114 #define OST_CREATE_PORTAL 7
115 #define OST_BULK_PORTAL 8
116 //#define MDC_REQUEST_PORTAL 9
117 #define MDC_REPLY_PORTAL 10
118 //#define MDC_BULK_PORTAL 11
119 #define MDS_REQUEST_PORTAL 12
120 //#define MDS_REPLY_PORTAL 13
121 #define MDS_BULK_PORTAL 14
122 #define LDLM_CB_REQUEST_PORTAL 15
123 #define LDLM_CB_REPLY_PORTAL 16
124 #define LDLM_CANCEL_REQUEST_PORTAL 17
125 #define LDLM_CANCEL_REPLY_PORTAL 18
126 //#define PTLBD_REQUEST_PORTAL 19
127 //#define PTLBD_REPLY_PORTAL 20
128 //#define PTLBD_BULK_PORTAL 21
129 #define MDS_SETATTR_PORTAL 22
130 #define MDS_READPAGE_PORTAL 23
131 #define MDS_MDS_PORTAL 24
133 #define MGC_REPLY_PORTAL 25
134 #define MGS_REQUEST_PORTAL 26
135 #define MGS_REPLY_PORTAL 27
136 #define OST_REQUEST_PORTAL 28
137 #define FLD_REQUEST_PORTAL 29
138 #define SEQ_METADATA_PORTAL 30
139 #define SEQ_DATA_PORTAL 31
140 #define SEQ_CONTROLLER_PORTAL 32
145 #define SVC_RUNNING 8
146 #define SVC_STOPPING 16
147 #define SVC_STOPPED 32
150 #define PTL_RPC_MSG_REQUEST 4711
151 #define PTL_RPC_MSG_ERR 4712
152 #define PTL_RPC_MSG_REPLY 4713
154 /* DON'T use swabbed values of MAGIC as magic! */
155 #define LUSTRE_MSG_MAGIC_V1 0x0BD00BD0
156 #define LUSTRE_MSG_MAGIC_V2 0x0BD00BD3
158 #define LUSTRE_MSG_MAGIC_V1_SWABBED 0xD00BD00B
159 #define LUSTRE_MSG_MAGIC_V2_SWABBED 0xD30BD00B
161 #define LUSTRE_MSG_MAGIC LUSTRE_MSG_MAGIC_V2
163 #define PTLRPC_MSG_VERSION 0x00000003
164 #define LUSTRE_VERSION_MASK 0xffff0000
165 #define LUSTRE_OBD_VERSION 0x00010000
166 #define LUSTRE_MDS_VERSION 0x00020000
167 #define LUSTRE_OST_VERSION 0x00030000
168 #define LUSTRE_DLM_VERSION 0x00040000
169 #define LUSTRE_LOG_VERSION 0x00050000
170 #define LUSTRE_MGS_VERSION 0x00060000
172 typedef __u32 mdsno_t;
173 typedef __u64 seqno_t;
176 * Describes a range of sequence, lsr_start is included but lsr_end is
178 * Same structure is used in fld module where lsr_mdt field holds mdt id
182 struct lu_seq_range {
190 * returns width of given range \a r
193 static inline __u64 range_space(const struct lu_seq_range *range)
195 return range->lsr_end - range->lsr_start;
199 * initialize range to zero
202 static inline void range_init(struct lu_seq_range *range)
204 range->lsr_start = range->lsr_end = range->lsr_mdt = 0;
208 * check if given seq id \a s is within given range \a r
211 static inline int range_within(const struct lu_seq_range *range,
214 return s >= range->lsr_start && s < range->lsr_end;
218 * allocate \a w units of sequence from range \a from.
220 static inline void range_alloc(struct lu_seq_range *to,
221 struct lu_seq_range *from,
224 to->lsr_start = from->lsr_start;
225 to->lsr_end = from->lsr_start + width;
226 from->lsr_start += width;
229 static inline int range_is_sane(const struct lu_seq_range *range)
231 return (range->lsr_end >= range->lsr_start);
234 static inline int range_is_zero(const struct lu_seq_range *range)
236 return (range->lsr_start == 0 && range->lsr_end == 0);
239 static inline int range_is_exhausted(const struct lu_seq_range *range)
242 return range_space(range) == 0;
245 #define DRANGE "[%#16.16"LPF64"x-%#16.16"LPF64"x):%x"
247 #define PRANGE(range) \
248 (range)->lsr_start, \
252 /** \defgroup lu_fid lu_fid
258 * Fid is a cluster-wide unique identifier of a file or an object
259 * (stripe). Fids are never reused. Fids are transmitted across network (in
260 * the sender byte-ordering), and stored on disk in a packed form (struct
261 * lu_fid_pack) in a big-endian order.
265 * fid sequence. Sequence is a unit of migration: all files (objects)
266 * with fids from a given sequence are stored on the same
269 * Lustre should support 2 ^ 64 objects, thus even if one
270 * sequence has one object we will never reach this value.
273 /** fid number within sequence. */
276 * fid version, used to distinguish different versions (in the sense
277 * of snapshots, etc.) of the same file system object. Not currently
284 * Following struct for MDT attributes, that will be kept inode's EA.
285 * Introduced in 2.0 release (please see b15993, for details)
287 struct lustre_mdt_attrs {
288 /** FID of this inode */
289 struct lu_fid lma_self_fid;
290 /** SOM state, mdt/ost type, others */
292 /** total sectors in objects */
293 __u64 lma_som_sectors;
301 /** initial fid id value */
302 LUSTRE_FID_INIT_OID = 1UL
305 /** returns fid object sequence */
306 static inline __u64 fid_seq(const struct lu_fid *fid)
311 /** returns fid object id */
312 static inline __u32 fid_oid(const struct lu_fid *fid)
317 /** returns fid object version */
318 static inline __u32 fid_ver(const struct lu_fid *fid)
323 static inline void fid_zero(struct lu_fid *fid)
325 memset(fid, 0, sizeof(*fid));
328 /* Normal FID sequence starts from this value, i.e. 1<<33 */
329 #define FID_SEQ_START 0x200000000ULL
331 /* IDIF sequence starts from this value, i.e. 1<<32 */
332 #define IDIF_SEQ_START 0x100000000ULL
335 * Check if a fid is igif or not.
336 * \param fid the fid to be tested.
337 * \return true if the fid is a igif; otherwise false.
339 static inline int fid_is_igif(const struct lu_fid *fid)
341 return fid_seq(fid) > 0 && fid_seq(fid) < IDIF_SEQ_START;
345 * Check if a fid is idif or not.
346 * \param fid the fid to be tested.
347 * \return true if the fid is a idif; otherwise false.
349 static inline int fid_is_idif(const struct lu_fid *fid)
351 return fid_seq(fid) >= IDIF_SEQ_START && fid_seq(fid) < FID_SEQ_START;
355 * Get inode number from a igif.
356 * \param fid a igif to get inode number from.
357 * \return inode number for the igif.
359 static inline ino_t lu_igif_ino(const struct lu_fid *fid)
365 * Get inode generation from a igif.
366 * \param fid a igif to get inode generation from.
367 * \return inode generation for the igif.
369 static inline __u32 lu_igif_gen(const struct lu_fid *fid)
374 #define DFID "["LPX64":0x%x:0x%x]"
375 #define SFID "0x%llx:0x%x:0x%x"
382 static inline void fid_cpu_to_le(struct lu_fid *dst, const struct lu_fid *src)
384 /* check that all fields are converted */
385 CLASSERT(sizeof *src ==
386 sizeof fid_seq(src) +
387 sizeof fid_oid(src) + sizeof fid_ver(src));
388 LASSERTF(fid_is_igif(src) || fid_ver(src) == 0, DFID"\n", PFID(src));
389 dst->f_seq = cpu_to_le64(fid_seq(src));
390 dst->f_oid = cpu_to_le32(fid_oid(src));
391 dst->f_ver = cpu_to_le32(fid_ver(src));
394 static inline void fid_le_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
396 /* check that all fields are converted */
397 CLASSERT(sizeof *src ==
398 sizeof fid_seq(src) +
399 sizeof fid_oid(src) + sizeof fid_ver(src));
400 dst->f_seq = le64_to_cpu(fid_seq(src));
401 dst->f_oid = le32_to_cpu(fid_oid(src));
402 dst->f_ver = le32_to_cpu(fid_ver(src));
403 LASSERTF(fid_is_igif(dst) || fid_ver(dst) == 0, DFID"\n", PFID(dst));
406 static inline void fid_cpu_to_be(struct lu_fid *dst, const struct lu_fid *src)
408 /* check that all fields are converted */
409 CLASSERT(sizeof *src ==
410 sizeof fid_seq(src) +
411 sizeof fid_oid(src) + sizeof fid_ver(src));
412 LASSERTF(fid_is_igif(src) || fid_ver(src) == 0, DFID"\n", PFID(src));
413 dst->f_seq = cpu_to_be64(fid_seq(src));
414 dst->f_oid = cpu_to_be32(fid_oid(src));
415 dst->f_ver = cpu_to_be32(fid_ver(src));
418 static inline void fid_be_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
420 /* check that all fields are converted */
421 CLASSERT(sizeof *src ==
422 sizeof fid_seq(src) +
423 sizeof fid_oid(src) + sizeof fid_ver(src));
424 dst->f_seq = be64_to_cpu(fid_seq(src));
425 dst->f_oid = be32_to_cpu(fid_oid(src));
426 dst->f_ver = be32_to_cpu(fid_ver(src));
427 LASSERTF(fid_is_igif(dst) || fid_ver(dst) == 0, DFID"\n", PFID(dst));
431 * Storage representation for fids.
433 * Variable size, first byte contains the length of the whole record.
437 char fp_area[sizeof(struct lu_fid)];
440 void fid_pack(struct lu_fid_pack *pack, const struct lu_fid *fid,
441 struct lu_fid *befider);
442 int fid_unpack(const struct lu_fid_pack *pack, struct lu_fid *fid);
444 static inline int fid_is_sane(const struct lu_fid *fid)
448 ((fid_seq(fid) >= FID_SEQ_START && fid_oid(fid) != 0
449 && fid_ver(fid) == 0) ||
453 static inline int fid_is_zero(const struct lu_fid *fid)
455 return fid_seq(fid) == 0 && fid_oid(fid) == 0;
458 extern void lustre_swab_lu_fid(struct lu_fid *fid);
459 extern void lustre_swab_lu_seq_range(struct lu_seq_range *range);
461 static inline int lu_fid_eq(const struct lu_fid *f0,
462 const struct lu_fid *f1)
464 /* Check that there is no alignment padding. */
465 CLASSERT(sizeof *f0 ==
466 sizeof f0->f_seq + sizeof f0->f_oid + sizeof f0->f_ver);
467 LASSERTF(fid_is_igif(f0) || fid_ver(f0) == 0, DFID, PFID(f0));
468 LASSERTF(fid_is_igif(f1) || fid_ver(f1) == 0, DFID, PFID(f1));
469 return memcmp(f0, f1, sizeof *f0) == 0;
472 #define __diff_normalize(val0, val1) \
474 typeof(val0) __val0 = (val0); \
475 typeof(val1) __val1 = (val1); \
477 (__val0 == __val1 ? 0 : __val0 > __val1 ? +1 : -1); \
480 static inline int lu_fid_cmp(const struct lu_fid *f0,
481 const struct lu_fid *f1)
484 __diff_normalize(fid_seq(f0), fid_seq(f1)) ?:
485 __diff_normalize(fid_oid(f0), fid_oid(f1)) ?:
486 __diff_normalize(fid_ver(f0), fid_ver(f1));
491 /** \defgroup lu_dir lu_dir
495 * Enumeration of possible directory entry attributes.
497 * Attributes follow directory entry header in the order they appear in this
500 enum lu_dirent_attrs {
506 * Layout of readdir pages, as transmitted on wire.
509 /** valid if LUDA_FID is set. */
510 struct lu_fid lde_fid;
511 /** a unique entry identifier: a hash or an offset. */
513 /** total record length, including all attributes. */
517 /** optional variable size attributes following this entry.
518 * taken from enum lu_dirent_attrs.
521 /** name is followed by the attributes indicated in ->ldp_attrs, in
522 * their natural order. After the last attribute, padding bytes are
523 * added to make ->lde_reclen a multiple of 8.
529 * Definitions of optional directory entry attributes formats.
531 * Individual attributes do not have their length encoded in a generic way. It
532 * is assumed that consumer of an attribute knows its format. This means that
533 * it is impossible to skip over an unknown attribute, except by skipping over all
534 * remaining attributes (by using ->lde_reclen), which is not too
535 * constraining, because new server versions will append new attributes at
536 * the end of an entry.
540 * Fid directory attribute: a fid of an object referenced by the entry. This
541 * will be almost always requested by the client and supplied by the server.
543 * Aligned to 8 bytes.
545 /* To have compatibility with 1.8, lets have fid in lu_dirent struct. */
550 * Aligned to 2 bytes.
557 __u64 ldp_hash_start;
561 struct lu_dirent ldp_entries[0];
564 enum lu_dirpage_flags {
568 static inline struct lu_dirent *lu_dirent_start(struct lu_dirpage *dp)
570 if (le32_to_cpu(dp->ldp_flags) & LDF_EMPTY)
573 return dp->ldp_entries;
576 static inline struct lu_dirent *lu_dirent_next(struct lu_dirent *ent)
578 struct lu_dirent *next;
580 if (le16_to_cpu(ent->lde_reclen) != 0)
581 next = ((void *)ent) + le16_to_cpu(ent->lde_reclen);
588 static inline int lu_dirent_calc_size(int namelen, __u16 attr)
592 if (attr & LUDA_TYPE) {
593 const unsigned align = sizeof(struct luda_type) - 1;
594 size = (sizeof(struct lu_dirent) + namelen + align) & ~align;
595 size += sizeof(struct luda_type);
597 size = sizeof(struct lu_dirent) + namelen;
599 return (size + 7) & ~7;
602 static inline int lu_dirent_size(struct lu_dirent *ent)
604 if (le16_to_cpu(ent->lde_reclen) == 0) {
605 return lu_dirent_calc_size(le16_to_cpu(ent->lde_namelen),
606 le32_to_cpu(ent->lde_attrs));
608 return le16_to_cpu(ent->lde_reclen);
611 #define DIR_END_OFF 0xfffffffffffffffeULL
615 struct lustre_handle {
618 #define DEAD_HANDLE_MAGIC 0xdeadbeefcafebabeULL
620 static inline int lustre_handle_is_used(struct lustre_handle *lh)
622 return lh->cookie != 0ull;
625 static inline int lustre_handle_equal(struct lustre_handle *lh1,
626 struct lustre_handle *lh2)
628 return lh1->cookie == lh2->cookie;
631 static inline void lustre_handle_copy(struct lustre_handle *tgt,
632 struct lustre_handle *src)
634 tgt->cookie = src->cookie;
637 /* flags for lm_flags */
638 #define MSGHDR_AT_SUPPORT 0x1
640 #define lustre_msg lustre_msg_v2
641 /* we depend on this structure to be 8-byte aligned */
642 /* this type is only endian-adjusted in lustre_unpack_msg() */
643 struct lustre_msg_v2 {
655 /* without gss, ptlrpc_body is put at the first buffer. */
656 #define PTLRPC_NUM_VERSIONS 4
658 struct lustre_handle pb_handle;
665 __u64 pb_last_committed;
670 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
671 __u32 pb_service_time; /* for rep, actual service time */
674 /* VBR: pre-versions */
675 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
676 /* padding for future needs */
680 extern void lustre_swab_ptlrpc_body(struct ptlrpc_body *pb);
682 /* message body offset for lustre_msg_v2 */
683 /* ptlrpc body offset in all request/reply messages */
684 #define MSG_PTLRPC_BODY_OFF 0
686 /* normal request/reply message record offset */
687 #define REQ_REC_OFF 1
688 #define REPLY_REC_OFF 1
690 /* ldlm request message body offset */
691 #define DLM_LOCKREQ_OFF 1 /* lockreq offset */
692 #define DLM_REQ_REC_OFF 2 /* normal dlm request record offset */
694 /* ldlm intent lock message body offset */
695 #define DLM_INTENT_IT_OFF 2 /* intent lock it offset */
696 #define DLM_INTENT_REC_OFF 3 /* intent lock record offset */
698 /* ldlm reply message body offset */
699 #define DLM_LOCKREPLY_OFF 1 /* lockrep offset */
700 #define DLM_REPLY_REC_OFF 2 /* reply record offset */
702 /* Flags that are operation-specific go in the top 16 bits. */
703 #define MSG_OP_FLAG_MASK 0xffff0000
704 #define MSG_OP_FLAG_SHIFT 16
706 /* Flags that apply to all requests are in the bottom 16 bits */
707 #define MSG_GEN_FLAG_MASK 0x0000ffff
708 #define MSG_LAST_REPLAY 0x0001
709 #define MSG_RESENT 0x0002
710 #define MSG_REPLAY 0x0004
711 /* #define MSG_AT_SUPPORT 0x0008
712 * This was used in early prototypes of adaptive timeouts, and while there
713 * shouldn't be any users of that code there also isn't a need for using this
714 * bits. Defer usage until at least 1.10 to avoid potential conflict. */
715 #define MSG_DELAY_REPLAY 0x0010
716 #define MSG_VERSION_REPLAY 0x0020
717 #define MSG_REQ_REPLAY_DONE 0x0040
718 #define MSG_LOCK_REPLAY_DONE 0x0080
721 * Flags for all connect opcodes (MDS_CONNECT, OST_CONNECT)
724 #define MSG_CONNECT_RECOVERING 0x00000001
725 #define MSG_CONNECT_RECONNECT 0x00000002
726 #define MSG_CONNECT_REPLAYABLE 0x00000004
727 //#define MSG_CONNECT_PEER 0x8
728 #define MSG_CONNECT_LIBCLIENT 0x00000010
729 #define MSG_CONNECT_INITIAL 0x00000020
730 #define MSG_CONNECT_ASYNC 0x00000040
731 #define MSG_CONNECT_NEXT_VER 0x00000080 /* use next version of lustre_msg */
732 #define MSG_CONNECT_TRANSNO 0x00000100 /* report transno */
735 #define OBD_CONNECT_RDONLY 0x1ULL /*client has read-only access*/
736 #define OBD_CONNECT_INDEX 0x2ULL /*connect specific LOV idx */
737 #define OBD_CONNECT_MDS 0x4ULL /*connect from MDT to OST */
738 #define OBD_CONNECT_GRANT 0x8ULL /*OSC gets grant at connect */
739 #define OBD_CONNECT_SRVLOCK 0x10ULL /*server takes locks for cli */
740 #define OBD_CONNECT_VERSION 0x20ULL /*Lustre versions in ocd */
741 #define OBD_CONNECT_REQPORTAL 0x40ULL /*Separate non-IO req portal */
742 #define OBD_CONNECT_ACL 0x80ULL /*access control lists */
743 #define OBD_CONNECT_XATTR 0x100ULL /*client use extended attr */
744 #define OBD_CONNECT_CROW 0x200ULL /*MDS+OST create obj on write*/
745 #define OBD_CONNECT_TRUNCLOCK 0x400ULL /*locks on server for punch */
746 #define OBD_CONNECT_TRANSNO 0x800ULL /*replay sends init transno */
747 #define OBD_CONNECT_IBITS 0x1000ULL /*support for inodebits locks*/
748 #define OBD_CONNECT_JOIN 0x2000ULL /*files can be concatenated */
749 #define OBD_CONNECT_ATTRFID 0x4000ULL /*Server can GetAttr By Fid*/
750 #define OBD_CONNECT_NODEVOH 0x8000ULL /*No open hndl on specl nodes*/
751 #define OBD_CONNECT_RMT_CLIENT 0x10000ULL /*Remote client */
752 #define OBD_CONNECT_RMT_CLIENT_FORCE 0x20000ULL /*Remote client by force */
753 #define OBD_CONNECT_BRW_SIZE 0x40000ULL /*Max bytes per rpc */
754 #define OBD_CONNECT_QUOTA64 0x80000ULL /*64bit qunit_data.qd_count */
755 #define OBD_CONNECT_MDS_CAPA 0x100000ULL /*MDS capability */
756 #define OBD_CONNECT_OSS_CAPA 0x200000ULL /*OSS capability */
757 #define OBD_CONNECT_CANCELSET 0x400000ULL /*Early batched cancels. */
758 #define OBD_CONNECT_SOM 0x800000ULL /*Size on MDS */
759 #define OBD_CONNECT_AT 0x1000000ULL /*client uses AT */
760 #define OBD_CONNECT_LRU_RESIZE 0x2000000ULL /*LRU resize feature. */
761 #define OBD_CONNECT_MDS_MDS 0x4000000ULL /*MDS-MDS connection */
762 #define OBD_CONNECT_REAL 0x8000000ULL /*real connection */
763 #define OBD_CONNECT_CHANGE_QS 0x10000000ULL /*shrink/enlarge qunit */
764 #define OBD_CONNECT_CKSUM 0x20000000ULL /*support several cksum algos*/
765 #define OBD_CONNECT_FID 0x40000000ULL /*FID is supported by server */
766 #define OBD_CONNECT_VBR 0x80000000ULL /*version based recovery */
767 #define OBD_CONNECT_LOV_V3 0x100000000ULL /*client supports LOV v3 EA */
768 #define OBD_CONNECT_GRANT_SHRINK 0x200000000ULL /* support grant shrink */
769 #define OBD_CONNECT_SKIP_ORPHAN 0x400000000ULL /* don't reuse orphan objids */
770 /* also update obd_connect_names[] for lprocfs_rd_connect_flags()
771 * and lustre/utils/wirecheck.c */
773 #ifdef HAVE_LRU_RESIZE_SUPPORT
774 #define LRU_RESIZE_CONNECT_FLAG OBD_CONNECT_LRU_RESIZE
776 #define LRU_RESIZE_CONNECT_FLAG 0
779 #define MDT_CONNECT_SUPPORTED (OBD_CONNECT_RDONLY | OBD_CONNECT_VERSION | \
780 OBD_CONNECT_ACL | OBD_CONNECT_XATTR | \
781 OBD_CONNECT_IBITS | OBD_CONNECT_JOIN | \
782 OBD_CONNECT_NODEVOH |/* OBD_CONNECT_ATTRFID |*/\
783 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
784 OBD_CONNECT_RMT_CLIENT | \
785 OBD_CONNECT_RMT_CLIENT_FORCE | \
786 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA | \
787 OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID | \
788 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_VBR | \
790 #define OST_CONNECT_SUPPORTED (OBD_CONNECT_SRVLOCK | OBD_CONNECT_GRANT | \
791 OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | \
792 OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_INDEX | \
793 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_QUOTA64 | \
794 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
795 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_CKSUM | \
796 OBD_CONNECT_CHANGE_QS | \
797 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_RMT_CLIENT | \
798 OBD_CONNECT_RMT_CLIENT_FORCE | OBD_CONNECT_VBR | \
799 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN | \
800 OBD_CONNECT_GRANT_SHRINK)
801 #define ECHO_CONNECT_SUPPORTED (0)
802 #define MGS_CONNECT_SUPPORTED (OBD_CONNECT_VERSION | OBD_CONNECT_AT)
804 #define OBD_OCD_VERSION(major,minor,patch,fix) (((major)<<24) + ((minor)<<16) +\
805 ((patch)<<8) + (fix))
806 #define OBD_OCD_VERSION_MAJOR(version) ((int)((version)>>24)&255)
807 #define OBD_OCD_VERSION_MINOR(version) ((int)((version)>>16)&255)
808 #define OBD_OCD_VERSION_PATCH(version) ((int)((version)>>8)&255)
809 #define OBD_OCD_VERSION_FIX(version) ((int)(version)&255)
811 /* This structure is used for both request and reply.
813 * If we eventually have separate connect data for different types, which we
814 * almost certainly will, then perhaps we stick a union in here. */
815 struct obd_connect_data {
816 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
817 __u32 ocd_version; /* lustre release version number */
818 __u32 ocd_grant; /* initial cache grant amount (bytes) */
819 __u32 ocd_index; /* LOV index to connect to */
820 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
821 __u64 ocd_ibits_known; /* inode bits this client understands */
822 __u32 ocd_nllu; /* non-local-lustre-user */
823 __u32 ocd_nllg; /* non-local-lustre-group */
824 __u64 ocd_transno; /* first transno from client to be replayed */
825 __u32 ocd_group; /* MDS group on OST */
826 __u32 ocd_cksum_types; /* supported checksum algorithms */
827 __u64 padding1; /* also fix lustre_swab_connect */
828 __u64 padding2; /* also fix lustre_swab_connect */
831 extern void lustre_swab_connect(struct obd_connect_data *ocd);
834 * Supported checksum algorithms. Up to 32 checksum types are supported.
835 * (32-bit mask stored in obd_connect_data::ocd_cksum_types)
836 * Please update DECLARE_CKSUM_NAME/OBD_CKSUM_ALL in obd.h when adding a new
837 * algorithm and also the OBD_FL_CKSUM* flags.
840 OBD_CKSUM_CRC32 = 0x00000001,
841 OBD_CKSUM_ADLER = 0x00000002,
845 * OST requests: OBDO & OBD request records
850 OST_REPLY = 0, /* reply ? */
868 OST_QUOTA_ADJUST_QUNIT = 20,
871 #define OST_FIRST_OPC OST_REPLY
873 typedef __u64 obd_id;
874 typedef __u64 obd_gr;
875 typedef __u64 obd_time;
876 typedef __u64 obd_size;
877 typedef __u64 obd_off;
878 typedef __u64 obd_blocks;
879 typedef __u64 obd_valid;
880 typedef __u32 obd_blksize;
881 typedef __u32 obd_mode;
882 typedef __u32 obd_uid;
883 typedef __u32 obd_gid;
884 typedef __u32 obd_flag;
885 typedef __u32 obd_count;
887 #define OBD_FL_INLINEDATA (0x00000001)
888 #define OBD_FL_OBDMDEXISTS (0x00000002)
889 #define OBD_FL_DELORPHAN (0x00000004) /* if set in o_flags delete orphans */
890 #define OBD_FL_NORPC (0x00000008) /* set in o_flags do in OSC not OST */
891 #define OBD_FL_IDONLY (0x00000010) /* set in o_flags only adjust obj id*/
892 #define OBD_FL_RECREATE_OBJS (0x00000020) /* recreate missing obj */
893 #define OBD_FL_DEBUG_CHECK (0x00000040) /* echo client/server debug check */
894 #define OBD_FL_NO_USRQUOTA (0x00000100) /* the object's owner is over quota */
895 #define OBD_FL_NO_GRPQUOTA (0x00000200) /* the object's group is over quota */
896 #define OBD_FL_CREATE_CROW (0x00000400) /* object should be create on write */
898 #define OBD_FL_TRUNCLOCK (0x00000800) /* delegate DLM locking during punch */
899 #define OBD_FL_CKSUM_CRC32 (0x00001000) /* CRC32 checksum type */
900 #define OBD_FL_CKSUM_ADLER (0x00002000) /* ADLER checksum type */
901 #define OBD_FL_CKSUM_RESV1 (0x00004000) /* reserved for future checksum type */
902 #define OBD_FL_CKSUM_RESV2 (0x00008000) /* reserved for future checksum type */
903 #define OBD_FL_CKSUM_RESV3 (0x00010000) /* reserved for future checksum type */
904 #define OBD_FL_SHRINK_GRANT (0x00020000) /* object shrink the grant */
906 #define OBD_FL_CKSUM_ALL (OBD_FL_CKSUM_CRC32 | OBD_FL_CKSUM_ADLER)
908 #define LOV_MAGIC_V1 0x0BD10BD0
909 #define LOV_MAGIC LOV_MAGIC_V1
910 #define LOV_MAGIC_JOIN 0x0BD20BD0
911 #define LOV_MAGIC_V3 0x0BD30BD0
913 #define LOV_PATTERN_RAID0 0x001 /* stripes are used round-robin */
914 #define LOV_PATTERN_RAID1 0x002 /* stripes are mirrors of each other */
915 #define LOV_PATTERN_FIRST 0x100 /* first stripe is not in round-robin */
916 #define LOV_PATTERN_CMOBD 0x200
918 #define LOV_OBJECT_GROUP_DEFAULT ~0ULL
919 #define LOV_OBJECT_GROUP_CLEAR 0ULL
921 #define lov_ost_data lov_ost_data_v1
922 struct lov_ost_data_v1 { /* per-stripe data structure (little-endian)*/
923 __u64 l_object_id; /* OST object ID */
924 __u64 l_object_gr; /* OST object group (creating MDS number) */
925 __u32 l_ost_gen; /* generation of this l_ost_idx */
926 __u32 l_ost_idx; /* OST index in LOV (lov_tgt_desc->tgts) */
929 #define lov_mds_md lov_mds_md_v1
930 struct lov_mds_md_v1 { /* LOV EA mds/wire data (little-endian) */
931 __u32 lmm_magic; /* magic number = LOV_MAGIC_V1 */
932 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
933 __u64 lmm_object_id; /* LOV object ID */
934 __u64 lmm_object_gr; /* LOV object group */
935 __u32 lmm_stripe_size; /* size of stripe in bytes */
936 __u32 lmm_stripe_count; /* num stripes in use for this object */
937 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
940 /* extern void lustre_swab_lov_mds_md(struct lov_mds_md *llm); */
942 #define MAX_MD_SIZE (sizeof(struct lov_mds_md) + 4 * sizeof(struct lov_ost_data))
943 #define MIN_MD_SIZE (sizeof(struct lov_mds_md) + 1 * sizeof(struct lov_ost_data))
945 #define XATTR_NAME_ACL_ACCESS "system.posix_acl_access"
946 #define XATTR_NAME_ACL_DEFAULT "system.posix_acl_default"
947 #define XATTR_USER_PREFIX "user."
948 #define XATTR_TRUSTED_PREFIX "trusted."
949 #define XATTR_SECURITY_PREFIX "security."
950 #define XATTR_LUSTRE_PREFIX "lustre."
952 #define XATTR_NAME_LOV "trusted.lov"
953 #define XATTR_NAME_LMA "trusted.lma"
954 #define XATTR_NAME_LMV "trusted.lmv"
955 #define XATTR_NAME_LINK "trusted.link"
958 struct lov_mds_md_v3 { /* LOV EA mds/wire data (little-endian) */
959 __u32 lmm_magic; /* magic number = LOV_MAGIC_V3 */
960 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
961 __u64 lmm_object_id; /* LOV object ID */
962 __u64 lmm_object_gr; /* LOV object group */
963 __u32 lmm_stripe_size; /* size of stripe in bytes */
964 __u32 lmm_stripe_count; /* num stripes in use for this object */
965 char lmm_pool_name[LOV_MAXPOOLNAME]; /* must be 32bit aligned */
966 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
970 #define OBD_MD_FLID (0x00000001ULL) /* object ID */
971 #define OBD_MD_FLATIME (0x00000002ULL) /* access time */
972 #define OBD_MD_FLMTIME (0x00000004ULL) /* data modification time */
973 #define OBD_MD_FLCTIME (0x00000008ULL) /* change time */
974 #define OBD_MD_FLSIZE (0x00000010ULL) /* size */
975 #define OBD_MD_FLBLOCKS (0x00000020ULL) /* allocated blocks count */
976 #define OBD_MD_FLBLKSZ (0x00000040ULL) /* block size */
977 #define OBD_MD_FLMODE (0x00000080ULL) /* access bits (mode & ~S_IFMT) */
978 #define OBD_MD_FLTYPE (0x00000100ULL) /* object type (mode & S_IFMT) */
979 #define OBD_MD_FLUID (0x00000200ULL) /* user ID */
980 #define OBD_MD_FLGID (0x00000400ULL) /* group ID */
981 #define OBD_MD_FLFLAGS (0x00000800ULL) /* flags word */
982 #define OBD_MD_FLNLINK (0x00002000ULL) /* link count */
983 #define OBD_MD_FLGENER (0x00004000ULL) /* generation number */
984 /*#define OBD_MD_FLINLINE (0x00008000ULL) inline data. used until 1.6.5 */
985 #define OBD_MD_FLRDEV (0x00010000ULL) /* device number */
986 #define OBD_MD_FLEASIZE (0x00020000ULL) /* extended attribute data */
987 #define OBD_MD_LINKNAME (0x00040000ULL) /* symbolic link target */
988 #define OBD_MD_FLHANDLE (0x00080000ULL) /* file/lock handle */
989 #define OBD_MD_FLCKSUM (0x00100000ULL) /* bulk data checksum */
990 #define OBD_MD_FLQOS (0x00200000ULL) /* quality of service stats */
991 /*#define OBD_MD_FLOSCOPQ (0x00400000ULL) osc opaque data, never used */
992 #define OBD_MD_FLCOOKIE (0x00800000ULL) /* log cancellation cookie */
993 #define OBD_MD_FLGROUP (0x01000000ULL) /* group */
994 #define OBD_MD_FLFID (0x02000000ULL) /* ->ost write inline fid */
995 #define OBD_MD_FLEPOCH (0x04000000ULL) /* ->ost write easize is epoch */
996 /* ->mds if epoch opens or closes */
997 #define OBD_MD_FLGRANT (0x08000000ULL) /* ost preallocation space grant */
998 #define OBD_MD_FLDIREA (0x10000000ULL) /* dir's extended attribute data */
999 #define OBD_MD_FLUSRQUOTA (0x20000000ULL) /* over quota flags sent from ost */
1000 #define OBD_MD_FLGRPQUOTA (0x40000000ULL) /* over quota flags sent from ost */
1001 #define OBD_MD_FLMODEASIZE (0x80000000ULL) /* EA size will be changed */
1003 #define OBD_MD_MDS (0x0000000100000000ULL) /* where an inode lives on */
1004 #define OBD_MD_REINT (0x0000000200000000ULL) /* reintegrate oa */
1005 #define OBD_MD_MEA (0x0000000400000000ULL) /* CMD split EA */
1007 #define OBD_MD_FLXATTR (0x0000001000000000ULL) /* xattr */
1008 #define OBD_MD_FLXATTRLS (0x0000002000000000ULL) /* xattr list */
1009 #define OBD_MD_FLXATTRRM (0x0000004000000000ULL) /* xattr remove */
1010 #define OBD_MD_FLACL (0x0000008000000000ULL) /* ACL */
1011 #define OBD_MD_FLRMTPERM (0x0000010000000000ULL) /* remote permission */
1012 #define OBD_MD_FLMDSCAPA (0x0000020000000000ULL) /* MDS capability */
1013 #define OBD_MD_FLOSSCAPA (0x0000040000000000ULL) /* OSS capability */
1014 #define OBD_MD_FLCKSPLIT (0x0000080000000000ULL) /* Check split on server */
1015 #define OBD_MD_FLCROSSREF (0x0000100000000000ULL) /* Cross-ref case */
1017 #define OBD_FL_TRUNC (0x0000200000000000ULL) /* for filter_truncate */
1019 #define OBD_MD_FLRMTLSETFACL (0x0001000000000000ULL) /* lfs lsetfacl case */
1020 #define OBD_MD_FLRMTLGETFACL (0x0002000000000000ULL) /* lfs lgetfacl case */
1021 #define OBD_MD_FLRMTRSETFACL (0x0004000000000000ULL) /* lfs rsetfacl case */
1022 #define OBD_MD_FLRMTRGETFACL (0x0008000000000000ULL) /* lfs rgetfacl case */
1024 #define OBD_MD_FLGETATTR (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLMTIME | \
1025 OBD_MD_FLCTIME | OBD_MD_FLSIZE | OBD_MD_FLBLKSZ | \
1026 OBD_MD_FLMODE | OBD_MD_FLTYPE | OBD_MD_FLUID | \
1027 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLNLINK | \
1028 OBD_MD_FLGENER | OBD_MD_FLRDEV | OBD_MD_FLGROUP)
1030 /* don't forget obdo_fid which is way down at the bottom so it can
1031 * come after the definition of llog_cookie */
1033 enum obd_statfs_state {
1034 OS_STATE_DEGRADED = 0x00000001, /**< RAID degraded/rebuilding */
1035 OS_STATE_READONLY = 0x00000002, /**< filesystem is read-only */
1036 OS_STATE_RDONLY_1 = 0x00000004, /**< obsolete 1.6, was EROFS=30 */
1037 OS_STATE_RDONLY_2 = 0x00000008, /**< obsolete 1.6, was EROFS=30 */
1038 OS_STATE_RDONLY_3 = 0x00000010, /**< obsolete 1.6, was EROFS=30 */
1052 __u32 os_state; /**< obd_statfs_state OS_STATE_* flag */
1064 extern void lustre_swab_obd_statfs (struct obd_statfs *os);
1065 #define OBD_STATFS_NODELAY 0x0001 /* requests should be send without delay
1066 * and resends for avoid deadlocks */
1067 #define OBD_STATFS_FROM_CACHE 0x0002 /* the statfs callback should not update
1069 #define OBD_STATFS_PTLRPCD 0x0004 /* requests will be sent via ptlrpcd
1070 * instead of a specific set. This
1071 * means that we cannot rely on the set
1072 * interpret routine to be called.
1073 * lov_statfs_fini() must thus be called
1074 * by the request interpret routine */
1076 /* ost_body.data values for OST_BRW */
1078 #define OBD_BRW_READ 0x01
1079 #define OBD_BRW_WRITE 0x02
1080 #define OBD_BRW_RWMASK (OBD_BRW_READ | OBD_BRW_WRITE)
1081 #define OBD_BRW_SYNC 0x08 /* this page is a part of synchronous
1082 * transfer and is not accounted in
1084 #define OBD_BRW_CHECK 0x10
1085 #define OBD_BRW_FROM_GRANT 0x20 /* the osc manages this under llite */
1086 #define OBD_BRW_GRANTED 0x40 /* the ost manages this */
1087 #define OBD_BRW_NOCACHE 0x80 /* this page is a part of non-cached IO */
1088 #define OBD_BRW_NOQUOTA 0x100
1089 #define OBD_BRW_SRVLOCK 0x200 /* Client holds no lock over this page */
1091 #define OBD_OBJECT_EOF 0xffffffffffffffffULL
1093 #define OST_MIN_PRECREATE 32
1094 #define OST_MAX_PRECREATE 20000
1103 extern void lustre_swab_obd_ioobj (struct obd_ioobj *ioo);
1105 /* multiple of 8 bytes => can array */
1106 struct niobuf_remote {
1112 extern void lustre_swab_niobuf_remote (struct niobuf_remote *nbr);
1114 /* lock value block communicated between the filter and llite */
1116 /* OST_LVB_ERR_INIT is needed because the return code in rc is
1117 * negative, i.e. because ((MASK + rc) & MASK) != MASK. */
1118 #define OST_LVB_ERR_INIT 0xffbadbad80000000ULL
1119 #define OST_LVB_ERR_MASK 0xffbadbad00000000ULL
1120 #define OST_LVB_IS_ERR(blocks) \
1121 ((blocks & OST_LVB_ERR_MASK) == OST_LVB_ERR_MASK)
1122 #define OST_LVB_SET_ERR(blocks, rc) \
1123 do { blocks = OST_LVB_ERR_INIT + rc; } while (0)
1124 #define OST_LVB_GET_ERR(blocks) (int)(blocks - OST_LVB_ERR_INIT)
1134 extern void lustre_swab_ost_lvb(struct ost_lvb *);
1143 MDS_GETATTR_NAME = 34,
1148 MDS_DISCONNECT = 39,
1154 MDS_DONE_WRITING = 45,
1156 MDS_QUOTACHECK = 47,
1159 MDS_SETXATTR = 50, /* obsolete, now it's MDS_REINT op */
1165 #define MDS_FIRST_OPC MDS_GETATTR
1182 } mds_reint_t, mdt_reint_t;
1184 extern void lustre_swab_generic_32s (__u32 *val);
1186 /* the disposition of the intent outlines what was executed */
1187 #define DISP_IT_EXECD 0x00000001
1188 #define DISP_LOOKUP_EXECD 0x00000002
1189 #define DISP_LOOKUP_NEG 0x00000004
1190 #define DISP_LOOKUP_POS 0x00000008
1191 #define DISP_OPEN_CREATE 0x00000010
1192 #define DISP_OPEN_OPEN 0x00000020
1193 #define DISP_ENQ_COMPLETE 0x00400000
1194 #define DISP_ENQ_OPEN_REF 0x00800000
1195 #define DISP_ENQ_CREATE_REF 0x01000000
1196 #define DISP_OPEN_LOCK 0x02000000
1198 /* INODE LOCK PARTS */
1199 #define MDS_INODELOCK_LOOKUP 0x000001 /* dentry, mode, owner, group */
1200 #define MDS_INODELOCK_UPDATE 0x000002 /* size, links, timestamps */
1201 #define MDS_INODELOCK_OPEN 0x000004 /* For opened files */
1203 /* Do not forget to increase MDS_INODELOCK_MAXSHIFT when adding new bits */
1204 #define MDS_INODELOCK_MAXSHIFT 2
1205 /* This FULL lock is useful to take on unlink sort of operations */
1206 #define MDS_INODELOCK_FULL ((1<<(MDS_INODELOCK_MAXSHIFT+1))-1)
1209 __u64 id; /* holds object id */
1210 __u32 generation; /* holds object generation */
1212 __u32 f_type; /* holds object type or stripe idx when passing it to
1213 * OST for saving into EA. */
1216 extern void lustre_swab_ll_fid (struct ll_fid *fid);
1218 #define MDS_STATUS_CONN 1
1219 #define MDS_STATUS_LOV 2
1221 struct mds_status_req {
1226 extern void lustre_swab_mds_status_req (struct mds_status_req *r);
1228 /* mdt_thread_info.mti_flags. */
1230 /* The flag indicates Size-on-MDS attributes are changed. */
1231 MF_SOM_CHANGE = (1 << 0),
1232 /* Flags indicates an epoch opens or closes. */
1233 MF_EPOCH_OPEN = (1 << 1),
1234 MF_EPOCH_CLOSE = (1 << 2),
1235 MF_MDC_CANCEL_FID1 = (1 << 3),
1236 MF_MDC_CANCEL_FID2 = (1 << 4),
1237 MF_MDC_CANCEL_FID3 = (1 << 5),
1238 MF_MDC_CANCEL_FID4 = (1 << 6),
1241 #define MF_SOM_LOCAL_FLAGS (MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID2 | \
1242 MF_MDC_CANCEL_FID3 | MF_MDC_CANCEL_FID4)
1244 #define MDS_BFLAG_UNCOMMITTED_WRITES 0x1
1245 #define MDS_BFLAG_EXT_FLAGS 0x80000000 /* == EXT3_RESERVED_FL */
1247 /* these should be identical to their EXT3_*_FL counterparts, and are
1248 * redefined here only to avoid dragging in ext3_fs.h */
1249 #define MDS_SYNC_FL 0x00000008 /* Synchronous updates */
1250 #define MDS_IMMUTABLE_FL 0x00000010 /* Immutable file */
1251 #define MDS_APPEND_FL 0x00000020 /* writes to file may only append */
1252 #define MDS_NOATIME_FL 0x00000080 /* do not update atime */
1253 #define MDS_DIRSYNC_FL 0x00010000 /* dirsync behaviour (dir only) */
1256 /* If MDS_BFLAG_IOC_FLAGS is set it means we requested EXT3_*_FL inode flags
1257 * and we need to decode these into local S_* flags in the inode. Otherwise
1258 * we pass flags straight through (see bug 9486). */
1259 static inline int ll_ext_to_inode_flags(int flags)
1261 return (flags & MDS_BFLAG_EXT_FLAGS) ?
1262 (((flags & MDS_SYNC_FL) ? S_SYNC : 0) |
1263 ((flags & MDS_NOATIME_FL) ? S_NOATIME : 0) |
1264 ((flags & MDS_APPEND_FL) ? S_APPEND : 0) |
1265 #if defined(S_DIRSYNC)
1266 ((flags & MDS_DIRSYNC_FL) ? S_DIRSYNC : 0) |
1268 ((flags & MDS_IMMUTABLE_FL) ? S_IMMUTABLE : 0)) :
1269 (flags & ~MDS_BFLAG_EXT_FLAGS);
1272 /* If MDS_BFLAG_EXT_FLAGS is set it means we requested EXT3_*_FL inode flags
1273 * and we pass these straight through. Otherwise we need to convert from
1274 * S_* flags to their EXT3_*_FL equivalents (see bug 9486). */
1275 static inline int ll_inode_to_ext_flags(int oflags, int iflags)
1277 return (oflags & MDS_BFLAG_EXT_FLAGS) ? (oflags & ~MDS_BFLAG_EXT_FLAGS):
1278 (((iflags & S_SYNC) ? MDS_SYNC_FL : 0) |
1279 ((iflags & S_NOATIME) ? MDS_NOATIME_FL : 0) |
1280 ((iflags & S_APPEND) ? MDS_APPEND_FL : 0) |
1281 #if defined(S_DIRSYNC)
1282 ((iflags & S_DIRSYNC) ? MDS_DIRSYNC_FL : 0) |
1284 ((iflags & S_IMMUTABLE) ? MDS_IMMUTABLE_FL : 0));
1289 * while mds_body is to interact with 1.6, mdt_body is to interact with 2.0.
1290 * both of them should have the same fields layout, because at client side
1291 * one could be dynamically cast to the other.
1293 * mdt_body has large size than mds_body, with unused padding (48 bytes)
1294 * at the end. client always use size of mdt_body to prepare request/reply
1295 * buffers, and actual data could be interepeted as mdt_body or mds_body
1301 struct lustre_handle handle;
1303 __u64 size; /* Offset, in the case of MDS_READPAGE */
1307 __u64 blocks; /* XID, in the case of MDS_READPAGE */
1316 __u32 flags; /* from vfs for pin/unpin, MDS_BFLAG for close */
1318 __u32 nlink; /* #bytes to read in the case of MDS_READPAGE */
1324 __u32 max_cookiesize;
1325 __u32 padding_4; /* also fix lustre_swab_mds_body */
1328 extern void lustre_swab_mds_body (struct mds_body *b);
1333 struct lustre_handle handle;
1335 __u64 size; /* Offset, in the case of MDS_READPAGE */
1339 __u64 blocks; /* XID, in the case of MDS_READPAGE */
1341 __u64 ino; /* for 1.6 compatibility */
1348 __u32 flags; /* from vfs for pin/unpin, MDS_BFLAG for close */
1350 __u32 nlink; /* #bytes to read in the case of MDS_READPAGE */
1351 __u32 generation; /* for 1.6 compatibility */
1356 __u32 max_cookiesize;
1357 __u32 padding_4; /* also fix lustre_swab_mdt_body */
1366 extern void lustre_swab_mdt_body (struct mdt_body *b);
1369 struct lustre_handle handle;
1375 extern void lustre_swab_mdt_epoch (struct mdt_epoch *b);
1377 #define Q_QUOTACHECK 0x800100
1378 #define Q_INITQUOTA 0x800101 /* init slave limits */
1379 #define Q_GETOINFO 0x800102 /* get obd quota info */
1380 #define Q_GETOQUOTA 0x800103 /* get obd quotas */
1381 #define Q_FINVALIDATE 0x800104 /* invalidate operational quotas */
1383 #define Q_TYPEMATCH(id, type) \
1384 ((id) == (type) || (id) == UGQUOTA)
1386 #define Q_TYPESET(oqc, type) Q_TYPEMATCH((oqc)->qc_type, type)
1388 #define Q_GETOCMD(oqc) \
1389 ((oqc)->qc_cmd == Q_GETOINFO || (oqc)->qc_cmd == Q_GETOQUOTA)
1391 #define QCTL_COPY(out, in) \
1393 Q_COPY(out, in, qc_cmd); \
1394 Q_COPY(out, in, qc_type); \
1395 Q_COPY(out, in, qc_id); \
1396 Q_COPY(out, in, qc_stat); \
1397 Q_COPY(out, in, qc_dqinfo); \
1398 Q_COPY(out, in, qc_dqblk); \
1401 struct obd_quotactl {
1406 struct obd_dqinfo qc_dqinfo;
1407 struct obd_dqblk qc_dqblk;
1410 extern void lustre_swab_obd_quotactl(struct obd_quotactl *q);
1412 struct quota_adjust_qunit {
1419 extern void lustre_swab_quota_adjust_qunit(struct quota_adjust_qunit *q);
1421 /* flags in qunit_data and quota_adjust_qunit will use macroes below */
1422 #define LQUOTA_FLAGS_GRP 1UL /* 0 is user, 1 is group */
1423 #define LQUOTA_FLAGS_BLK 2UL /* 0 is inode, 1 is block */
1424 #define LQUOTA_FLAGS_ADJBLK 4UL /* adjust the block qunit size */
1425 #define LQUOTA_FLAGS_ADJINO 8UL /* adjust the inode qunit size */
1426 #define LQUOTA_FLAGS_CHG_QS 16UL /* indicate whether it has capability of
1427 * OBD_CONNECT_CHANGE_QS */
1429 /* the status of lqsk_flags in struct lustre_qunit_size_key */
1430 #define LQUOTA_QUNIT_FLAGS (LQUOTA_FLAGS_GRP | LQUOTA_FLAGS_BLK)
1432 #define QAQ_IS_GRP(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_GRP)
1433 #define QAQ_IS_ADJBLK(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_ADJBLK)
1434 #define QAQ_IS_ADJINO(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_ADJINO)
1436 #define QAQ_SET_GRP(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_GRP)
1437 #define QAQ_SET_ADJBLK(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_ADJBLK)
1438 #define QAQ_SET_ADJINO(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_ADJINO)
1440 /* inode access permission for remote user, the inode info are omitted,
1441 * for client knows them. */
1442 struct mds_remote_perm {
1447 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
1450 /* permissions for md_perm.mp_perm */
1452 CFS_SETUID_PERM = 0x01,
1453 CFS_SETGID_PERM = 0x02,
1454 CFS_SETGRP_PERM = 0x04,
1455 CFS_RMTACL_PERM = 0x08,
1456 CFS_RMTOWN_PERM = 0x10
1459 extern void lustre_swab_mds_remote_perm(struct mds_remote_perm *p);
1461 struct mdt_remote_perm {
1466 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
1469 extern void lustre_swab_mdt_remote_perm(struct mdt_remote_perm *p);
1471 struct mds_rec_setattr {
1478 struct ll_fid sa_fid;
1479 __u64 sa_valid; /* MDS_ATTR_* attributes */
1486 __u32 sa_attr_flags;
1487 __u32 sa_padding; /* also fix lustre_swab_mds_rec_setattr */
1490 extern void lustre_swab_mds_rec_setattr (struct mds_rec_setattr *sa);
1492 struct mdt_rec_setattr {
1502 __u32 sa_padding_1_h;
1503 struct lu_fid sa_fid;
1512 __u32 sa_attr_flags;
1520 extern void lustre_swab_mdt_rec_setattr (struct mdt_rec_setattr *sa);
1523 * Attribute flags used in mds_rec_setattr::sa_valid.
1524 * The kernel's #defines for ATTR_* should not be used over the network
1525 * since the client and MDS may run different kernels (see bug 13828)
1526 * Therefore, we should only use MDS_ATTR_* attributes for sa_valid.
1528 #define MDS_ATTR_MODE 0x1ULL /* = 1 */
1529 #define MDS_ATTR_UID 0x2ULL /* = 2 */
1530 #define MDS_ATTR_GID 0x4ULL /* = 4 */
1531 #define MDS_ATTR_SIZE 0x8ULL /* = 8 */
1532 #define MDS_ATTR_ATIME 0x10ULL /* = 16 */
1533 #define MDS_ATTR_MTIME 0x20ULL /* = 32 */
1534 #define MDS_ATTR_CTIME 0x40ULL /* = 64 */
1535 #define MDS_ATTR_ATIME_SET 0x80ULL /* = 128 */
1536 #define MDS_ATTR_MTIME_SET 0x100ULL /* = 256 */
1537 #define MDS_ATTR_FORCE 0x200ULL /* = 512, Not a change, but a change it */
1538 #define MDS_ATTR_ATTR_FLAG 0x400ULL /* = 1024 */
1539 #define MDS_ATTR_KILL_SUID 0x800ULL /* = 2048 */
1540 #define MDS_ATTR_KILL_SGID 0x1000ULL /* = 4096 */
1541 #define MDS_ATTR_CTIME_SET 0x2000ULL /* = 8192 */
1542 #define MDS_ATTR_FROM_OPEN 0x4000ULL /* = 16384, called from open path, ie O_TRUNC */
1543 #define MDS_ATTR_BLOCKS 0x8000ULL /* = 32768 */
1546 #define FMODE_READ 00000001
1547 #define FMODE_WRITE 00000002
1550 #define FMODE_EPOCH 01000000
1551 #define FMODE_EPOCHLCK 02000000
1552 #define FMODE_SOM 04000000
1553 #define FMODE_CLOSED 0
1555 #define MDS_OPEN_CREATED 00000010
1556 #define MDS_OPEN_CROSS 00000020
1558 #define MDS_FMODE_EXEC 00000004
1559 #define MDS_OPEN_CREAT 00000100
1560 #define MDS_OPEN_EXCL 00000200
1561 #define MDS_OPEN_TRUNC 00001000
1562 #define MDS_OPEN_APPEND 00002000
1563 #define MDS_OPEN_SYNC 00010000
1564 #define MDS_OPEN_DIRECTORY 00200000
1566 #define MDS_OPEN_DELAY_CREATE 0100000000 /* delay initial object create */
1567 #define MDS_OPEN_OWNEROVERRIDE 0200000000 /* NFSD rw-reopen ro file for owner */
1568 #define MDS_OPEN_JOIN_FILE 0400000000 /* open for join file*/
1569 #define MDS_CREATE_RMT_ACL 01000000000 /* indicate create on remote server
1570 * with default ACL */
1571 #define MDS_CREATE_SLAVE_OBJ 02000000000 /* indicate create slave object
1572 * actually, this is for create, not
1573 * conflict with other open flags */
1574 #define MDS_OPEN_LOCK 04000000000 /* This open requires open lock */
1575 #define MDS_OPEN_HAS_EA 010000000000 /* specify object create pattern */
1576 #define MDS_OPEN_HAS_OBJS 020000000000 /* Just set the EA the obj exist */
1578 /* permission for create non-directory file */
1579 #define MAY_CREATE (1 << 7)
1580 /* permission for create directory file */
1581 #define MAY_LINK (1 << 8)
1582 /* permission for delete from the directory */
1583 #define MAY_UNLINK (1 << 9)
1584 /* source's permission for rename */
1585 #define MAY_RENAME_SRC (1 << 10)
1586 /* target's permission for rename */
1587 #define MAY_RENAME_TAR (1 << 11)
1588 /* part (parent's) VTX permission check */
1589 #define MAY_VTX_PART (1 << 12)
1590 /* full VTX permission check */
1591 #define MAY_VTX_FULL (1 << 13)
1592 /* lfs rgetfacl permission check */
1593 #define MAY_RGETFACL (1 << 14)
1596 MDS_CHECK_SPLIT = 1 << 0,
1597 MDS_CROSS_REF = 1 << 1,
1598 MDS_VTX_BYPASS = 1 << 2,
1599 MDS_PERM_BYPASS = 1 << 3,
1601 MDS_QUOTA_IGNORE = 1 << 5
1604 struct mds_rec_create {
1609 __u32 cr_flags; /* for use with open */
1611 struct ll_fid cr_fid;
1612 struct ll_fid cr_replayfid;
1616 __u32 cr_padding_1; /* also fix lustre_swab_mds_rec_create */
1617 __u32 cr_padding_2; /* also fix lustre_swab_mds_rec_create */
1618 __u32 cr_padding_3; /* also fix lustre_swab_mds_rec_create */
1619 __u32 cr_padding_4; /* also fix lustre_swab_mds_rec_create */
1620 __u32 cr_padding_5; /* also fix lustre_swab_mds_rec_create */
1623 extern void lustre_swab_mds_rec_create (struct mds_rec_create *cr);
1625 struct mdt_rec_create {
1633 __u32 cr_suppgid1_h;
1635 __u32 cr_suppgid2_h;
1636 struct lu_fid cr_fid1;
1637 struct lu_fid cr_fid2;
1638 struct lustre_handle cr_old_handle; /* handle in case of open replay */
1642 __u64 cr_padding_1; /* pad for 64 bits*/
1645 __u32 cr_flags; /* for use with open */
1651 extern void lustre_swab_mdt_rec_create (struct mdt_rec_create *cr);
1653 struct mds_rec_join {
1654 struct ll_fid jr_fid;
1658 extern void lustre_swab_mds_rec_join (struct mds_rec_join *jr);
1660 struct mdt_rec_join {
1661 struct lu_fid jr_fid;
1665 extern void lustre_swab_mdt_rec_join (struct mdt_rec_join *jr);
1667 struct mds_rec_link {
1674 struct ll_fid lk_fid1;
1675 struct ll_fid lk_fid2;
1677 __u32 lk_padding_1; /* also fix lustre_swab_mds_rec_link */
1678 __u32 lk_padding_2; /* also fix lustre_swab_mds_rec_link */
1679 __u32 lk_padding_3; /* also fix lustre_swab_mds_rec_link */
1680 __u32 lk_padding_4; /* also fix lustre_swab_mds_rec_link */
1683 extern void lustre_swab_mds_rec_link (struct mds_rec_link *lk);
1685 struct mdt_rec_link {
1693 __u32 lk_suppgid1_h;
1695 __u32 lk_suppgid2_h;
1696 struct lu_fid lk_fid1;
1697 struct lu_fid lk_fid2;
1711 struct mds_rec_unlink {
1718 struct ll_fid ul_fid1;
1719 struct ll_fid ul_fid2;
1721 __u32 ul_padding_1; /* also fix lustre_swab_mds_rec_unlink */
1722 __u32 ul_padding_2; /* also fix lustre_swab_mds_rec_unlink */
1723 __u32 ul_padding_3; /* also fix lustre_swab_mds_rec_unlink */
1724 __u32 ul_padding_4; /* also fix lustre_swab_mds_rec_unlink */
1727 extern void lustre_swab_mds_rec_unlink (struct mds_rec_unlink *ul);
1729 struct mdt_rec_unlink {
1737 __u32 ul_suppgid1_h;
1739 __u32 ul_suppgid2_h;
1740 struct lu_fid ul_fid1;
1741 struct lu_fid ul_fid2;
1755 struct mds_rec_rename {
1762 struct ll_fid rn_fid1;
1763 struct ll_fid rn_fid2;
1765 __u32 rn_padding_1; /* also fix lustre_swab_mds_rec_rename */
1766 __u32 rn_padding_2; /* also fix lustre_swab_mds_rec_rename */
1767 __u32 rn_padding_3; /* also fix lustre_swab_mds_rec_rename */
1768 __u32 rn_padding_4; /* also fix lustre_swab_mds_rec_rename */
1771 extern void lustre_swab_mds_rec_rename (struct mds_rec_rename *rn);
1773 struct mdt_rec_rename {
1781 __u32 rn_suppgid1_h;
1783 __u32 rn_suppgid2_h;
1784 struct lu_fid rn_fid1;
1785 struct lu_fid rn_fid2;
1791 __u32 rn_bias; /* some operation flags */
1792 __u32 rn_mode; /* cross-ref rename has mode */
1799 struct mdt_rec_setxattr {
1807 __u32 sx_suppgid1_h;
1809 __u32 sx_suppgid2_h;
1810 struct lu_fid sx_fid;
1811 __u64 sx_padding_1; /* These three members are lu_fid size */
1823 __u32 sx_padding_10;
1824 __u32 sx_padding_11;
1827 struct mdt_rec_reint {
1835 __u32 rr_suppgid1_h;
1837 __u32 rr_suppgid2_h;
1838 struct lu_fid rr_fid1;
1839 struct lu_fid rr_fid2;
1847 __u32 rr_padding_1; /* also fix lustre_swab_mdt_rec_reint */
1848 __u32 rr_padding_2; /* also fix lustre_swab_mdt_rec_reint */
1849 __u32 rr_padding_3; /* also fix lustre_swab_mdt_rec_reint */
1850 __u32 rr_padding_4; /* also fix lustre_swab_mdt_rec_reint */
1853 extern void lustre_swab_mdt_rec_reint(struct mdt_rec_reint *rr);
1856 __u32 ld_tgt_count; /* how many MDS's */
1857 __u32 ld_active_tgt_count; /* how many active */
1858 __u32 ld_default_stripe_count; /* how many objects are used */
1859 __u32 ld_pattern; /* default MEA_MAGIC_* */
1860 __u64 ld_default_hash_size;
1861 __u64 ld_padding_1; /* also fix lustre_swab_lmv_desc */
1862 __u32 ld_padding_2; /* also fix lustre_swab_lmv_desc */
1863 __u32 ld_qos_maxage; /* in second */
1864 __u32 ld_padding_3; /* also fix lustre_swab_lmv_desc */
1865 __u32 ld_padding_4; /* also fix lustre_swab_lmv_desc */
1866 struct obd_uuid ld_uuid;
1869 extern void lustre_swab_lmv_desc (struct lmv_desc *ld);
1871 /* TODO: lmv_stripe_md should contain mds capabilities for all slave fids */
1872 struct lmv_stripe_md {
1877 char mea_pool_name[LOV_MAXPOOLNAME];
1878 struct lu_fid mea_ids[0];
1881 extern void lustre_swab_lmv_stripe_md(struct lmv_stripe_md *mea);
1883 /* lmv structures */
1884 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
1885 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
1886 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
1888 #define MAX_HASH_SIZE_32 0x7fffffffUL
1889 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
1890 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
1897 extern void lustre_swab_md_fld (struct md_fld *mf);
1902 FLD_FIRST_OPC = FLD_QUERY
1908 SEQ_FIRST_OPC = SEQ_QUERY
1912 SEQ_ALLOC_SUPER = 0,
1917 * LOV data structures
1920 #define LOV_MIN_STRIPE_BITS 16 /* maximum PAGE_SIZE (ia64), power of 2 */
1921 #define LOV_MIN_STRIPE_SIZE (1<<LOV_MIN_STRIPE_BITS)
1922 #define LOV_MAX_STRIPE_COUNT 160 /* until bug 4424 is fixed */
1923 #define LOV_V1_INSANE_STRIPE_COUNT 65532 /* maximum stripe count bz13933 */
1925 #define LOV_MAX_UUID_BUFFER_SIZE 8192
1926 /* The size of the buffer the lov/mdc reserves for the
1927 * array of UUIDs returned by the MDS. With the current
1928 * protocol, this will limit the max number of OSTs per LOV */
1930 #define LOV_DESC_MAGIC 0xB0CCDE5C
1932 /* LOV settings descriptor (should only contain static info) */
1934 __u32 ld_tgt_count; /* how many OBD's */
1935 __u32 ld_active_tgt_count; /* how many active */
1936 __u32 ld_default_stripe_count; /* how many objects are used */
1937 __u32 ld_pattern; /* default PATTERN_RAID0 */
1938 __u64 ld_default_stripe_size; /* in bytes */
1939 __u64 ld_default_stripe_offset; /* in bytes */
1940 __u32 ld_padding_0; /* unused */
1941 __u32 ld_qos_maxage; /* in second */
1942 __u32 ld_padding_1; /* also fix lustre_swab_lov_desc */
1943 __u32 ld_padding_2; /* also fix lustre_swab_lov_desc */
1944 struct obd_uuid ld_uuid;
1947 #define ld_magic ld_active_tgt_count /* for swabbing from llogs */
1949 extern void lustre_swab_lov_desc (struct lov_desc *ld);
1954 /* opcodes -- MUST be distinct from OST/MDS opcodes */
1959 LDLM_BL_CALLBACK = 104,
1960 LDLM_CP_CALLBACK = 105,
1961 LDLM_GL_CALLBACK = 106,
1964 #define LDLM_FIRST_OPC LDLM_ENQUEUE
1966 #define RES_NAME_SIZE 4
1967 struct ldlm_res_id {
1968 __u64 name[RES_NAME_SIZE];
1971 extern void lustre_swab_ldlm_res_id (struct ldlm_res_id *id);
1987 #define LCK_MODE_NUM 8
1997 #define LDLM_MIN_TYPE LDLM_PLAIN
1999 struct ldlm_extent {
2005 static inline int ldlm_extent_overlap(struct ldlm_extent *ex1,
2006 struct ldlm_extent *ex2)
2008 return (ex1->start <= ex2->end) && (ex2->start <= ex1->end);
2011 struct ldlm_inodebits {
2018 __u64 blocking_export; /* not actually used over the wire */
2019 __u32 blocking_pid; /* not actually used over the wire */
2023 /* it's important that the fields of the ldlm_extent structure match
2024 * the first fields of the ldlm_flock structure because there is only
2025 * one ldlm_swab routine to process the ldlm_policy_data_t union. if
2026 * this ever changes we will need to swab the union differently based
2027 * on the resource type. */
2030 struct ldlm_extent l_extent;
2031 struct ldlm_flock l_flock;
2032 struct ldlm_inodebits l_inodebits;
2033 } ldlm_policy_data_t;
2035 extern void lustre_swab_ldlm_policy_data (ldlm_policy_data_t *d);
2037 struct ldlm_intent {
2041 extern void lustre_swab_ldlm_intent (struct ldlm_intent *i);
2043 struct ldlm_resource_desc {
2044 ldlm_type_t lr_type;
2045 __u32 lr_padding; /* also fix lustre_swab_ldlm_resource_desc */
2046 struct ldlm_res_id lr_name;
2049 extern void lustre_swab_ldlm_resource_desc (struct ldlm_resource_desc *r);
2051 struct ldlm_lock_desc {
2052 struct ldlm_resource_desc l_resource;
2053 ldlm_mode_t l_req_mode;
2054 ldlm_mode_t l_granted_mode;
2055 ldlm_policy_data_t l_policy_data;
2058 extern void lustre_swab_ldlm_lock_desc (struct ldlm_lock_desc *l);
2060 #define LDLM_LOCKREQ_HANDLES 2
2061 #define LDLM_ENQUEUE_CANCEL_OFF 1
2063 struct ldlm_request {
2066 struct ldlm_lock_desc lock_desc;
2067 struct lustre_handle lock_handle[LDLM_LOCKREQ_HANDLES];
2070 extern void lustre_swab_ldlm_request (struct ldlm_request *rq);
2072 /* If LDLM_ENQUEUE, 1 slot is already occupied, 1 is available.
2073 * Otherwise, 2 are available. */
2074 #define ldlm_request_bufsize(count,type) \
2076 int _avail = LDLM_LOCKREQ_HANDLES; \
2077 _avail -= (type == LDLM_ENQUEUE ? LDLM_ENQUEUE_CANCEL_OFF : 0); \
2078 sizeof(struct ldlm_request) + \
2079 (count > _avail ? count - _avail : 0) * \
2080 sizeof(struct lustre_handle); \
2085 __u32 lock_padding; /* also fix lustre_swab_ldlm_reply */
2086 struct ldlm_lock_desc lock_desc;
2087 struct lustre_handle lock_handle;
2088 __u64 lock_policy_res1;
2089 __u64 lock_policy_res2;
2092 extern void lustre_swab_ldlm_reply (struct ldlm_reply *r);
2095 * Opcodes for mountconf (mgs and mgc)
2100 MGS_EXCEPTION, /* node died, etc. */
2101 MGS_TARGET_REG, /* whenever target starts up */
2106 #define MGS_FIRST_OPC MGS_CONNECT
2108 #define MGS_PARAM_MAXLEN 1024
2109 #define KEY_SET_INFO "set_info"
2111 struct mgs_send_param {
2112 char mgs_param[MGS_PARAM_MAXLEN];
2115 /* We pass this info to the MGS so it can write config logs */
2116 #define MTI_NAME_MAXLEN 64
2117 #define MTI_PARAM_MAXLEN 4096
2118 #define MTI_NIDS_MAX 32
2119 struct mgs_target_info {
2120 __u32 mti_lustre_ver;
2121 __u32 mti_stripe_index;
2122 __u32 mti_config_ver;
2124 __u32 mti_nid_count;
2125 __u32 padding; /* 64 bit align */
2126 char mti_fsname[MTI_NAME_MAXLEN];
2127 char mti_svname[MTI_NAME_MAXLEN];
2128 char mti_uuid[sizeof(struct obd_uuid)];
2129 __u64 mti_nids[MTI_NIDS_MAX]; /* host nids (lnet_nid_t)*/
2130 char mti_params[MTI_PARAM_MAXLEN];
2133 extern void lustre_swab_mgs_target_info(struct mgs_target_info *oinfo);
2135 /* Config marker flags (in config log) */
2136 #define CM_START 0x01
2138 #define CM_SKIP 0x04
2139 #define CM_UPGRADE146 0x08
2140 #define CM_EXCLUDE 0x10
2141 #define CM_START_SKIP (CM_START | CM_SKIP)
2144 __u32 cm_step; /* aka config version */
2146 __u32 cm_vers; /* lustre release version number */
2147 __u32 padding; /* 64 bit align */
2148 __u64 cm_createtime; /*when this record was first created */
2149 __u64 cm_canceltime; /*when this record is no longer valid*/
2150 char cm_tgtname[MTI_NAME_MAXLEN];
2151 char cm_comment[MTI_NAME_MAXLEN];
2154 extern void lustre_swab_cfg_marker(struct cfg_marker *marker,
2155 int swab, int size);
2158 * Opcodes for multiple servers.
2167 #define OBD_FIRST_OPC OBD_PING
2169 /* catalog of log objects */
2171 /** Identifier for a single log object */
2176 } __attribute__((packed));
2178 /** Records written to the CATALOGS list */
2179 #define CATLIST "CATALOGS"
2181 struct llog_logid lci_logid;
2185 } __attribute__((packed));
2187 /** join file lov mds md*/
2188 struct lov_mds_md_join {
2189 struct lov_mds_md lmmj_md;
2190 /*join private info*/
2191 struct llog_logid lmmj_array_id; /*array object id*/
2192 __u32 lmmj_extent_count; /*array extent count*/
2195 /* Log data record types - there is no specific reason that these need to
2196 * be related to the RPC opcodes, but no reason not to (may be handy later?)
2198 #define LLOG_OP_MAGIC 0x10600000
2199 #define LLOG_OP_MASK 0xfff00000
2202 LLOG_PAD_MAGIC = LLOG_OP_MAGIC | 0x00000,
2203 OST_SZ_REC = LLOG_OP_MAGIC | 0x00f00,
2204 OST_RAID1_REC = LLOG_OP_MAGIC | 0x01000,
2205 MDS_UNLINK_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) | REINT_UNLINK,
2206 MDS_SETATTR_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) | REINT_SETATTR,
2207 MDS_SETATTR64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) | REINT_SETATTR,
2208 OBD_CFG_REC = LLOG_OP_MAGIC | 0x20000,
2209 PTL_CFG_REC = LLOG_OP_MAGIC | 0x30000, /* obsolete */
2210 LLOG_GEN_REC = LLOG_OP_MAGIC | 0x40000,
2211 LLOG_JOIN_REC = LLOG_OP_MAGIC | 0x50000,
2212 CHANGELOG_REC = LLOG_OP_MAGIC | 0x60000,
2213 CHANGELOG_USER_REC = LLOG_OP_MAGIC | 0x70000,
2214 LLOG_HDR_MAGIC = LLOG_OP_MAGIC | 0x45539,
2215 LLOG_LOGID_MAGIC = LLOG_OP_MAGIC | 0x4553b,
2219 * for now, continue to support old pad records which have 0 for their
2220 * type but still need to be swabbed for their length
2222 #define LLOG_REC_HDR_NEEDS_SWABBING(r) \
2223 (((r)->lrh_type & __swab32(LLOG_OP_MASK)) == \
2224 __swab32(LLOG_OP_MAGIC) || \
2225 (((r)->lrh_type == 0) && ((r)->lrh_len > LLOG_CHUNK_SIZE)))
2227 /** Log record header - stored in little endian order.
2228 * Each record must start with this struct, end with a llog_rec_tail,
2229 * and be a multiple of 256 bits in size.
2231 struct llog_rec_hdr {
2238 struct llog_rec_tail {
2243 struct llog_logid_rec {
2244 struct llog_rec_hdr lid_hdr;
2245 struct llog_logid lid_id;
2251 struct llog_rec_tail lid_tail;
2252 } __attribute__((packed));
2254 /** MDS extent description
2255 * It is for joined file extent info, each extent info for joined file
2256 * just like (start, end, lmm).
2258 struct mds_extent_desc {
2259 __u64 med_start; /* extent start */
2260 __u64 med_len; /* extent length */
2261 struct lov_mds_md med_lmm; /* extent's lmm */
2264 /** Joined file array extent log record*/
2265 struct llog_array_rec {
2266 struct llog_rec_hdr lmr_hdr;
2267 struct mds_extent_desc lmr_med;
2268 struct llog_rec_tail lmr_tail;
2271 struct llog_create_rec {
2272 struct llog_rec_hdr lcr_hdr;
2273 struct ll_fid lcr_fid;
2277 struct llog_rec_tail lcr_tail;
2278 } __attribute__((packed));
2280 struct llog_orphan_rec {
2281 struct llog_rec_hdr lor_hdr;
2285 struct llog_rec_tail lor_tail;
2286 } __attribute__((packed));
2288 struct llog_unlink_rec {
2289 struct llog_rec_hdr lur_hdr;
2293 struct llog_rec_tail lur_tail;
2294 } __attribute__((packed));
2296 struct llog_setattr_rec {
2297 struct llog_rec_hdr lsr_hdr;
2303 struct llog_rec_tail lsr_tail;
2304 } __attribute__((packed));
2306 struct llog_setattr64_rec {
2307 struct llog_rec_hdr lsr_hdr;
2315 struct llog_rec_tail lsr_tail;
2316 } __attribute__((packed));
2318 struct llog_size_change_rec {
2319 struct llog_rec_hdr lsc_hdr;
2320 struct ll_fid lsc_fid;
2323 struct llog_rec_tail lsc_tail;
2324 } __attribute__((packed));
2326 #define CHANGELOG_MAGIC 0xca103000
2327 /** Changelog record types
2328 * When adding record types, update mdd_lproc.c's changelog_str
2330 enum changelog_rec_type {
2332 CL_CREATE = 1, /* namespace */
2333 CL_MKDIR = 2, /* namespace */
2334 CL_HARDLINK = 3, /* namespace */
2335 CL_SOFTLINK = 4, /* namespace */
2336 CL_MKNOD = 5, /* namespace */
2337 CL_UNLINK = 6, /* namespace */
2338 CL_RMDIR = 7, /* namespace */
2339 CL_RENAME = 8, /* namespace */
2340 CL_EXT = 9, /* namespace extended record (2nd half of rename) */
2341 CL_OPEN = 10, /* not currently used */
2342 CL_CLOSE = 11, /* may be written to log only with mtime change */
2350 /** Changelog entry type names. Must be defined in the same order as the
2351 * \a changelog_rec_type enum.
2353 #define DECLARE_CHANGELOG_NAMES static const char *changelog_str[] = \
2354 {"MARK","CREAT","MKDIR","HLINK","SLINK","MKNOD","UNLNK","RMDIR", \
2355 "RNMFM","RNMTO","OPEN","CLOSE","IOCTL","TRUNC","SATTR","XATTR"}
2357 /** \a changelog_rec_type's that can't be masked */
2358 #define CHANGELOG_MINMASK (1 << CL_MARK)
2359 /** bits covering all \a changelog_rec_type's */
2360 #define CHANGELOG_ALLMASK 0XFFFF
2361 /** default \a changelog_rec_type mask */
2362 #define CHANGELOG_DEFMASK CHANGELOG_ALLMASK
2364 /* per-record flags */
2365 #define CLF_VERSION 0x1000
2366 #define CLF_FLAGMASK 0x0FFF
2367 #define CLF_HSM 0x0001
2369 /* changelog llog name, needed by client replicators */
2370 #define CHANGELOG_CATALOG "changelog_catalog"
2372 struct changelog_setinfo {
2375 } __attribute__((packed));
2377 /** changelog record */
2378 struct llog_changelog_rec {
2379 struct llog_rec_hdr cr_hdr;
2380 __u16 cr_flags; /**< (flags&CLF_FLAGMASK)|CLF_VERSION */
2382 __u32 cr_type; /**< \a changelog_rec_type */
2384 __u64 cr_prev; /**< last index for this target fid */
2387 struct lu_fid cr_tfid; /**< target fid */
2388 __u32 cr_markerflags; /**< CL_MARK flags */
2390 struct lu_fid cr_pfid; /**< parent fid */
2392 char cr_name[0]; /**< last element */
2393 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
2395 } __attribute__((packed));
2397 #define CHANGELOG_USER_PREFIX "cl"
2399 struct llog_changelog_user_rec {
2400 struct llog_rec_hdr cur_hdr;
2404 struct llog_rec_tail cur_tail;
2405 } __attribute__((packed));
2410 } __attribute__((packed));
2412 struct llog_gen_rec {
2413 struct llog_rec_hdr lgr_hdr;
2414 struct llog_gen lgr_gen;
2415 struct llog_rec_tail lgr_tail;
2417 /* On-disk header structure of each log object, stored in little endian order */
2418 #define LLOG_CHUNK_SIZE 8192
2419 #define LLOG_HEADER_SIZE (96)
2420 #define LLOG_BITMAP_BYTES (LLOG_CHUNK_SIZE - LLOG_HEADER_SIZE)
2422 #define LLOG_MIN_REC_SIZE (24) /* round(llog_rec_hdr + llog_rec_tail) */
2424 /* flags for the logs */
2425 #define LLOG_F_ZAP_WHEN_EMPTY 0x1
2426 #define LLOG_F_IS_CAT 0x2
2427 #define LLOG_F_IS_PLAIN 0x4
2429 struct llog_log_hdr {
2430 struct llog_rec_hdr llh_hdr;
2431 __u64 llh_timestamp;
2433 __u32 llh_bitmap_offset;
2437 /* for a catalog the first plain slot is next to it */
2438 struct obd_uuid llh_tgtuuid;
2439 __u32 llh_reserved[LLOG_HEADER_SIZE/sizeof(__u32) - 23];
2440 __u32 llh_bitmap[LLOG_BITMAP_BYTES/sizeof(__u32)];
2441 struct llog_rec_tail llh_tail;
2442 } __attribute__((packed));
2444 #define LLOG_BITMAP_SIZE(llh) ((llh->llh_hdr.lrh_len - \
2445 llh->llh_bitmap_offset - \
2446 sizeof(llh->llh_tail)) * 8)
2448 /** log cookies are used to reference a specific log file and a record therein */
2449 struct llog_cookie {
2450 struct llog_logid lgc_lgl;
2454 } __attribute__((packed));
2456 /** llog protocol */
2457 enum llogd_rpc_ops {
2458 LLOG_ORIGIN_HANDLE_CREATE = 501,
2459 LLOG_ORIGIN_HANDLE_NEXT_BLOCK = 502,
2460 LLOG_ORIGIN_HANDLE_READ_HEADER = 503,
2461 LLOG_ORIGIN_HANDLE_WRITE_REC = 504,
2462 LLOG_ORIGIN_HANDLE_CLOSE = 505,
2463 LLOG_ORIGIN_CONNECT = 506,
2464 LLOG_CATINFO = 507, /* for lfs catinfo */
2465 LLOG_ORIGIN_HANDLE_PREV_BLOCK = 508,
2466 LLOG_ORIGIN_HANDLE_DESTROY = 509, /* for destroy llog object*/
2468 LLOG_FIRST_OPC = LLOG_ORIGIN_HANDLE_CREATE
2472 struct llog_logid lgd_logid;
2474 __u32 lgd_llh_flags;
2476 __u32 lgd_saved_index;
2478 __u64 lgd_cur_offset;
2479 } __attribute__((packed));
2481 struct llogd_conn_body {
2482 struct llog_gen lgdc_gen;
2483 struct llog_logid lgdc_logid;
2484 __u32 lgdc_ctxt_idx;
2485 } __attribute__((packed));
2487 struct lov_user_ost_data_join { /* per-stripe data structure */
2488 __u64 l_extent_start; /* extent start*/
2489 __u64 l_extent_end; /* extent end*/
2490 __u64 l_object_id; /* OST object ID */
2491 __u64 l_object_gr; /* OST object group (creating MDS number) */
2492 __u32 l_ost_gen; /* generation of this OST index */
2493 __u32 l_ost_idx; /* OST index in LOV */
2494 } __attribute__((packed));
2496 struct lov_user_md_join { /* LOV EA user data (host-endian) */
2497 __u32 lmm_magic; /* magic number = LOV_MAGIC_JOIN */
2498 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
2499 __u64 lmm_object_id; /* LOV object ID */
2500 __u64 lmm_object_gr; /* LOV object group */
2501 __u32 lmm_stripe_size; /* size of stripe in bytes */
2502 __u32 lmm_stripe_count; /* num stripes in use for this object */
2503 __u32 lmm_extent_count; /* extent count of lmm*/
2504 __u64 lmm_tree_id; /* mds tree object id */
2505 __u64 lmm_tree_gen; /* mds tree object gen */
2506 struct llog_logid lmm_array_id; /* mds extent desc llog object id */
2507 struct lov_user_ost_data_join lmm_objects[0]; /* per-stripe data */
2508 } __attribute__((packed));
2510 /* Note: 64-bit types are 64-bit aligned in structure */
2512 obd_valid o_valid; /* hot fields in this obdo */
2516 obd_size o_size; /* o_size-o_blocks == ost_lvb */
2520 obd_blocks o_blocks; /* brw: cli sent cached bytes */
2523 /* 32-bit fields start here: keep an even number of them via padding */
2524 obd_blksize o_blksize; /* optimal IO blocksize */
2525 obd_mode o_mode; /* brw: cli sent cache remain */
2529 obd_count o_nlink; /* brw: checksum */
2530 obd_count o_generation;
2531 obd_count o_misc; /* brw: o_dropped */
2532 __u32 o_easize; /* epoch in ost writes */
2534 __u32 o_stripe_idx; /* holds stripe idx */
2536 struct lustre_handle o_handle; /* brw: lock handle to prolong locks */
2537 struct llog_cookie o_lcookie; /* destroy: unlink cookie from MDS */
2546 #define o_dirty o_blocks
2547 #define o_undirty o_mode
2548 #define o_dropped o_misc
2549 #define o_cksum o_nlink
2551 extern void lustre_swab_obdo (struct obdo *o);
2553 /* request structure for OST's */
2558 /* Key for FIEMAP to be used in get_info calls */
2559 struct ll_fiemap_info_key {
2562 struct ll_user_fiemap fiemap;
2565 extern void lustre_swab_ost_body (struct ost_body *b);
2566 extern void lustre_swab_ost_last_id(obd_id *id);
2567 extern void lustre_swab_fiemap(struct ll_user_fiemap *fiemap);
2569 extern void lustre_swab_lov_user_md_v1(struct lov_user_md_v1 *lum);
2570 extern void lustre_swab_lov_user_md_v3(struct lov_user_md_v3 *lum);
2571 extern void lustre_swab_lov_user_md_objects(struct lov_user_ost_data *lod,
2573 extern void lustre_swab_lov_user_md_join(struct lov_user_md_join *lumj);
2574 extern void lustre_swab_lov_mds_md(struct lov_mds_md *lmm);
2577 extern void lustre_swab_llogd_body (struct llogd_body *d);
2578 extern void lustre_swab_llog_hdr (struct llog_log_hdr *h);
2579 extern void lustre_swab_llogd_conn_body (struct llogd_conn_body *d);
2580 extern void lustre_swab_llog_rec(struct llog_rec_hdr *rec,
2581 struct llog_rec_tail *tail);
2584 extern void lustre_swab_lustre_cfg(struct lustre_cfg *lcfg);
2586 /* this will be used when OBD_CONNECT_CHANGE_QS is set */
2589 * ID appiles to (uid, gid)
2593 * LQUOTA_FLAGS_* affect the responding bits
2597 * acquire/release count (bytes for block quota)
2601 * when a master returns the reply to a slave, it will
2602 * contain the current corresponding qunit size
2608 #define QDATA_IS_GRP(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_GRP)
2609 #define QDATA_IS_BLK(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_BLK)
2610 #define QDATA_IS_ADJBLK(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_ADJBLK)
2611 #define QDATA_IS_ADJINO(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_ADJINO)
2612 #define QDATA_IS_CHANGE_QS(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_CHG_QS)
2614 #define QDATA_SET_GRP(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_GRP)
2615 #define QDATA_SET_BLK(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_BLK)
2616 #define QDATA_SET_ADJBLK(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_ADJBLK)
2617 #define QDATA_SET_ADJINO(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_ADJINO)
2618 #define QDATA_SET_CHANGE_QS(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_CHG_QS)
2620 #define QDATA_CLR_GRP(qdata) ((qdata)->qd_flags &= ~LQUOTA_FLAGS_GRP)
2621 #define QDATA_CLR_CHANGE_QS(qdata) ((qdata)->qd_flags &= ~LQUOTA_FLAGS_CHG_QS)
2623 extern void lustre_swab_qdata(struct qunit_data *d);
2624 extern int quota_get_qdata(void*req, struct qunit_data *qdata,
2625 int is_req, int is_exp);
2626 extern int quota_copy_qdata(void *request, struct qunit_data *qdata,
2627 int is_req, int is_exp);
2634 #define QUOTA_FIRST_OPC QUOTA_DQACQ
2636 #define JOIN_FILE_ALIGN 4096
2638 #define QUOTA_REQUEST 1
2639 #define QUOTA_REPLY 0
2640 #define QUOTA_EXPORT 1
2641 #define QUOTA_IMPORT 0
2643 /* quota check function */
2644 #define QUOTA_RET_OK 0 /**< return successfully */
2645 #define QUOTA_RET_NOQUOTA 1 /**< not support quota */
2646 #define QUOTA_RET_NOLIMIT 2 /**< quota limit isn't set */
2647 #define QUOTA_RET_ACQUOTA 4 /**< need to acquire extra quota */
2650 /* security opcodes */
2653 SEC_CTX_INIT_CONT = 802,
2656 SEC_FIRST_OPC = SEC_CTX_INIT
2660 * capa related definitions
2662 #define CAPA_HMAC_MAX_LEN 64
2663 #define CAPA_HMAC_KEY_MAX_LEN 56
2665 /* NB take care when changing the sequence of elements this struct,
2666 * because the offset info is used in find_capa() */
2667 struct lustre_capa {
2668 struct lu_fid lc_fid; /** fid */
2669 __u64 lc_opc; /** operations allowed */
2670 __u64 lc_uid; /** file owner */
2671 __u64 lc_gid; /** file group */
2672 __u32 lc_flags; /** HMAC algorithm & flags */
2673 __u32 lc_keyid; /** key# used for the capability */
2674 __u32 lc_timeout; /** capa timeout value (sec) */
2675 __u32 lc_expiry; /** expiry time (sec) */
2676 __u8 lc_hmac[CAPA_HMAC_MAX_LEN]; /** HMAC */
2677 } __attribute__((packed));
2679 extern void lustre_swab_lustre_capa(struct lustre_capa *c);
2681 /** lustre_capa::lc_opc */
2683 CAPA_OPC_BODY_WRITE = 1<<0, /**< write object data */
2684 CAPA_OPC_BODY_READ = 1<<1, /**< read object data */
2685 CAPA_OPC_INDEX_LOOKUP = 1<<2, /**< lookup object fid */
2686 CAPA_OPC_INDEX_INSERT = 1<<3, /**< insert object fid */
2687 CAPA_OPC_INDEX_DELETE = 1<<4, /**< delete object fid */
2688 CAPA_OPC_OSS_WRITE = 1<<5, /**< write oss object data */
2689 CAPA_OPC_OSS_READ = 1<<6, /**< read oss object data */
2690 CAPA_OPC_OSS_TRUNC = 1<<7, /**< truncate oss object */
2691 CAPA_OPC_OSS_DESTROY = 1<<8, /**< destroy oss object */
2692 CAPA_OPC_META_WRITE = 1<<9, /**< write object meta data */
2693 CAPA_OPC_META_READ = 1<<10, /**< read object meta data */
2696 #define CAPA_OPC_OSS_RW (CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE)
2697 #define CAPA_OPC_MDS_ONLY \
2698 (CAPA_OPC_BODY_WRITE | CAPA_OPC_BODY_READ | CAPA_OPC_INDEX_LOOKUP | \
2699 CAPA_OPC_INDEX_INSERT | CAPA_OPC_INDEX_DELETE)
2700 #define CAPA_OPC_OSS_ONLY \
2701 (CAPA_OPC_OSS_WRITE | CAPA_OPC_OSS_READ | CAPA_OPC_OSS_TRUNC | \
2702 CAPA_OPC_OSS_DESTROY)
2703 #define CAPA_OPC_MDS_DEFAULT ~CAPA_OPC_OSS_ONLY
2704 #define CAPA_OPC_OSS_DEFAULT ~(CAPA_OPC_MDS_ONLY | CAPA_OPC_OSS_ONLY)
2706 /* MDS capability covers object capability for operations of body r/w
2707 * (dir readpage/sendpage), index lookup/insert/delete and meta data r/w,
2708 * while OSS capability only covers object capability for operations of
2709 * oss data(file content) r/w/truncate.
2711 static inline int capa_for_mds(struct lustre_capa *c)
2713 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) != 0;
2716 static inline int capa_for_oss(struct lustre_capa *c)
2718 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) == 0;
2721 /* lustre_capa::lc_hmac_alg */
2723 CAPA_HMAC_ALG_SHA1 = 1, /**< sha1 algorithm */
2727 #define CAPA_FL_MASK 0x00ffffff
2728 #define CAPA_HMAC_ALG_MASK 0xff000000
2730 struct lustre_capa_key {
2731 __u64 lk_mdsid; /**< mds# */
2732 __u32 lk_keyid; /**< key# */
2734 __u8 lk_key[CAPA_HMAC_KEY_MAX_LEN]; /**< key */
2735 } __attribute__((packed));
2737 extern void lustre_swab_lustre_capa_key(struct lustre_capa_key *k);
2739 /** The link ea holds 1 \a link_ea_entry for each hardlink */
2740 #define LINK_EA_MAGIC 0x11EAF1DFUL
2741 struct link_ea_header {
2744 __u64 leh_len; /* total size */
2750 /** Hardlink data is name and parent fid.
2751 * Stored in this crazy struct for maximum packing and endian-neutrality
2753 struct link_ea_entry {
2754 /** __u16 stored big-endian, unaligned */
2756 struct lu_fid_pack lee_parent_fid; /**< variable length */
2757 /** logically after lee_parent_fid; don't use directly */