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
142 /* Portal 63 is reserved for the Cray Inc DVS - nic@cray.com, roe@cray.com, n8851@cray.com */
147 #define SVC_RUNNING 8
148 #define SVC_STOPPING 16
149 #define SVC_STOPPED 32
152 #define PTL_RPC_MSG_REQUEST 4711
153 #define PTL_RPC_MSG_ERR 4712
154 #define PTL_RPC_MSG_REPLY 4713
156 /* DON'T use swabbed values of MAGIC as magic! */
157 #define LUSTRE_MSG_MAGIC_V1 0x0BD00BD0
158 #define LUSTRE_MSG_MAGIC_V2 0x0BD00BD3
160 #define LUSTRE_MSG_MAGIC_V1_SWABBED 0xD00BD00B
161 #define LUSTRE_MSG_MAGIC_V2_SWABBED 0xD30BD00B
163 #define LUSTRE_MSG_MAGIC LUSTRE_MSG_MAGIC_V2
165 #define PTLRPC_MSG_VERSION 0x00000003
166 #define LUSTRE_VERSION_MASK 0xffff0000
167 #define LUSTRE_OBD_VERSION 0x00010000
168 #define LUSTRE_MDS_VERSION 0x00020000
169 #define LUSTRE_OST_VERSION 0x00030000
170 #define LUSTRE_DLM_VERSION 0x00040000
171 #define LUSTRE_LOG_VERSION 0x00050000
172 #define LUSTRE_MGS_VERSION 0x00060000
174 typedef __u32 mdsno_t;
175 typedef __u64 seqno_t;
178 * Describes a range of sequence, lsr_start is included but lsr_end is
180 * Same structure is used in fld module where lsr_mdt field holds mdt id
184 struct lu_seq_range {
192 * returns width of given range \a r
195 static inline __u64 range_space(const struct lu_seq_range *range)
197 return range->lsr_end - range->lsr_start;
201 * initialize range to zero
204 static inline void range_init(struct lu_seq_range *range)
206 range->lsr_start = range->lsr_end = range->lsr_mdt = 0;
210 * check if given seq id \a s is within given range \a r
213 static inline int range_within(const struct lu_seq_range *range,
216 return s >= range->lsr_start && s < range->lsr_end;
219 static inline int range_is_sane(const struct lu_seq_range *range)
221 return (range->lsr_end >= range->lsr_start);
224 static inline int range_is_zero(const struct lu_seq_range *range)
226 return (range->lsr_start == 0 && range->lsr_end == 0);
229 static inline int range_is_exhausted(const struct lu_seq_range *range)
232 return range_space(range) == 0;
235 #define DRANGE "[%#16.16"LPF64"x-%#16.16"LPF64"x):%x"
237 #define PRANGE(range) \
238 (range)->lsr_start, \
242 /** \defgroup lu_fid lu_fid
246 * Flags for lustre_mdt_attrs::lma_compat and lustre_mdt_attrs::lma_incompat.
249 LMAC_HSM = 0x00000001,
250 LMAC_SOM = 0x00000002,
254 * Masks for all features that should be supported by a Lustre version to
255 * access a specific file.
256 * This information is stored in lustre_mdt_attrs::lma_incompat.
258 * NOTE: No incompat feature should be added before bug #17670 is landed.
260 #define LMA_INCOMPAT_SUPP 0x0
263 * Following struct for MDT attributes, that will be kept inode's EA.
264 * Introduced in 2.0 release (please see b15993, for details)
266 struct lustre_mdt_attrs {
268 * Bitfield for supported data in this structure. From enum lma_compat.
269 * lma_self_fid and lma_flags are always available.
273 * Per-file incompat feature list. Lustre version should support all
274 * flags set in this field. The supported feature mask is available in
278 /** FID of this inode */
279 struct lu_fid lma_self_fid;
280 /** mdt/ost type, others */
282 /* IO Epoch SOM attributes belongs to */
284 /** total file size in objects */
286 /** total fs blocks in objects */
287 __u64 lma_som_blocks;
288 /** mds mount id the size is valid for */
289 __u64 lma_som_mountid;
293 * Fill \a lma with its first content.
294 * Only fid is stored.
296 static inline void lustre_lma_init(struct lustre_mdt_attrs *lma,
297 const struct lu_fid *fid)
300 lma->lma_incompat = 0;
301 memcpy(&lma->lma_self_fid, fid, sizeof(*fid));
303 lma->lma_ioepoch = 0;
304 lma->lma_som_size = 0;
305 lma->lma_som_blocks = 0;
306 lma->lma_som_mountid = 0;
308 /* If a field is added in struct lustre_mdt_attrs, zero it explicitly
309 * and change the test below. */
310 LASSERT(sizeof(*lma) ==
311 (offsetof(struct lustre_mdt_attrs, lma_som_mountid) +
312 sizeof(lma->lma_som_mountid)));
315 extern void lustre_swab_lu_fid(struct lu_fid *fid);
318 * Swab, if needed, lustre_mdt_attr struct to on-disk format.
319 * Otherwise, do not touch it.
321 static inline void lustre_lma_swab(struct lustre_mdt_attrs *lma)
323 /* Use LUSTRE_MSG_MAGIC to detect local endianess. */
324 if (LUSTRE_MSG_MAGIC != cpu_to_le32(LUSTRE_MSG_MAGIC)) {
325 __swab32s(&lma->lma_compat);
326 __swab32s(&lma->lma_incompat);
327 lustre_swab_lu_fid(&lma->lma_self_fid);
328 __swab64s(&lma->lma_flags);
329 __swab64s(&lma->lma_ioepoch);
330 __swab64s(&lma->lma_som_size);
331 __swab64s(&lma->lma_som_blocks);
332 __swab64s(&lma->lma_som_mountid);
341 /** initial fid id value */
342 LUSTRE_FID_INIT_OID = 1UL
345 /** returns fid object sequence */
346 static inline __u64 fid_seq(const struct lu_fid *fid)
351 /** returns fid object id */
352 static inline __u32 fid_oid(const struct lu_fid *fid)
357 /** returns fid object version */
358 static inline __u32 fid_ver(const struct lu_fid *fid)
363 static inline void fid_zero(struct lu_fid *fid)
365 memset(fid, 0, sizeof(*fid));
368 /* Normal FID sequence starts from this value, i.e. 1<<33 */
369 #define FID_SEQ_START 0x200000000ULL
371 /* IDIF sequence starts from this value, i.e. 1<<32 */
372 #define IDIF_SEQ_START 0x100000000ULL
375 * Check if a fid is igif or not.
376 * \param fid the fid to be tested.
377 * \return true if the fid is a igif; otherwise false.
379 static inline int fid_is_igif(const struct lu_fid *fid)
381 return fid_seq(fid) > 0 && fid_seq(fid) < IDIF_SEQ_START;
385 * Check if a fid is idif or not.
386 * \param fid the fid to be tested.
387 * \return true if the fid is a idif; otherwise false.
389 static inline int fid_is_idif(const struct lu_fid *fid)
391 return fid_seq(fid) >= IDIF_SEQ_START && fid_seq(fid) < FID_SEQ_START;
395 * Get inode number from a igif.
396 * \param fid a igif to get inode number from.
397 * \return inode number for the igif.
399 static inline ino_t lu_igif_ino(const struct lu_fid *fid)
405 * Get inode generation from a igif.
406 * \param fid a igif to get inode generation from.
407 * \return inode generation for the igif.
409 static inline __u32 lu_igif_gen(const struct lu_fid *fid)
415 * Fids are transmitted across network (in the sender byte-ordering),
416 * and stored on disk in big-endian order.
418 static inline void fid_cpu_to_le(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 LASSERTF(fid_is_igif(src) || fid_ver(src) == 0, DFID"\n", PFID(src));
425 dst->f_seq = cpu_to_le64(fid_seq(src));
426 dst->f_oid = cpu_to_le32(fid_oid(src));
427 dst->f_ver = cpu_to_le32(fid_ver(src));
430 static inline void fid_le_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
432 /* check that all fields are converted */
433 CLASSERT(sizeof *src ==
434 sizeof fid_seq(src) +
435 sizeof fid_oid(src) + sizeof fid_ver(src));
436 dst->f_seq = le64_to_cpu(fid_seq(src));
437 dst->f_oid = le32_to_cpu(fid_oid(src));
438 dst->f_ver = le32_to_cpu(fid_ver(src));
439 LASSERTF(fid_is_igif(dst) || fid_ver(dst) == 0, DFID"\n", PFID(dst));
442 static inline void fid_cpu_to_be(struct lu_fid *dst, const struct lu_fid *src)
444 /* check that all fields are converted */
445 CLASSERT(sizeof *src ==
446 sizeof fid_seq(src) +
447 sizeof fid_oid(src) + sizeof fid_ver(src));
448 LASSERTF(fid_is_igif(src) || fid_ver(src) == 0, DFID"\n", PFID(src));
449 dst->f_seq = cpu_to_be64(fid_seq(src));
450 dst->f_oid = cpu_to_be32(fid_oid(src));
451 dst->f_ver = cpu_to_be32(fid_ver(src));
454 static inline void fid_be_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
456 /* check that all fields are converted */
457 CLASSERT(sizeof *src ==
458 sizeof fid_seq(src) +
459 sizeof fid_oid(src) + sizeof fid_ver(src));
460 dst->f_seq = be64_to_cpu(fid_seq(src));
461 dst->f_oid = be32_to_cpu(fid_oid(src));
462 dst->f_ver = be32_to_cpu(fid_ver(src));
463 LASSERTF(fid_is_igif(dst) || fid_ver(dst) == 0, DFID"\n", PFID(dst));
466 static inline int fid_is_sane(const struct lu_fid *fid)
470 ((fid_seq(fid) >= FID_SEQ_START && fid_oid(fid) != 0
471 && fid_ver(fid) == 0) ||
475 static inline int fid_is_zero(const struct lu_fid *fid)
477 return fid_seq(fid) == 0 && fid_oid(fid) == 0;
480 extern void lustre_swab_lu_fid(struct lu_fid *fid);
481 extern void lustre_swab_lu_seq_range(struct lu_seq_range *range);
483 static inline int lu_fid_eq(const struct lu_fid *f0,
484 const struct lu_fid *f1)
486 /* Check that there is no alignment padding. */
487 CLASSERT(sizeof *f0 ==
488 sizeof f0->f_seq + sizeof f0->f_oid + sizeof f0->f_ver);
489 LASSERTF(fid_is_igif(f0) || fid_ver(f0) == 0, DFID, PFID(f0));
490 LASSERTF(fid_is_igif(f1) || fid_ver(f1) == 0, DFID, PFID(f1));
491 return memcmp(f0, f1, sizeof *f0) == 0;
494 #define __diff_normalize(val0, val1) \
496 typeof(val0) __val0 = (val0); \
497 typeof(val1) __val1 = (val1); \
499 (__val0 == __val1 ? 0 : __val0 > __val1 ? +1 : -1); \
502 static inline int lu_fid_cmp(const struct lu_fid *f0,
503 const struct lu_fid *f1)
506 __diff_normalize(fid_seq(f0), fid_seq(f1)) ?:
507 __diff_normalize(fid_oid(f0), fid_oid(f1)) ?:
508 __diff_normalize(fid_ver(f0), fid_ver(f1));
513 /** \defgroup lu_dir lu_dir
517 * Enumeration of possible directory entry attributes.
519 * Attributes follow directory entry header in the order they appear in this
522 enum lu_dirent_attrs {
528 * Layout of readdir pages, as transmitted on wire.
531 /** valid if LUDA_FID is set. */
532 struct lu_fid lde_fid;
533 /** a unique entry identifier: a hash or an offset. */
535 /** total record length, including all attributes. */
539 /** optional variable size attributes following this entry.
540 * taken from enum lu_dirent_attrs.
543 /** name is followed by the attributes indicated in ->ldp_attrs, in
544 * their natural order. After the last attribute, padding bytes are
545 * added to make ->lde_reclen a multiple of 8.
551 * Definitions of optional directory entry attributes formats.
553 * Individual attributes do not have their length encoded in a generic way. It
554 * is assumed that consumer of an attribute knows its format. This means that
555 * it is impossible to skip over an unknown attribute, except by skipping over all
556 * remaining attributes (by using ->lde_reclen), which is not too
557 * constraining, because new server versions will append new attributes at
558 * the end of an entry.
562 * Fid directory attribute: a fid of an object referenced by the entry. This
563 * will be almost always requested by the client and supplied by the server.
565 * Aligned to 8 bytes.
567 /* To have compatibility with 1.8, lets have fid in lu_dirent struct. */
572 * Aligned to 2 bytes.
579 __u64 ldp_hash_start;
583 struct lu_dirent ldp_entries[0];
586 enum lu_dirpage_flags {
590 static inline struct lu_dirent *lu_dirent_start(struct lu_dirpage *dp)
592 if (le32_to_cpu(dp->ldp_flags) & LDF_EMPTY)
595 return dp->ldp_entries;
598 static inline struct lu_dirent *lu_dirent_next(struct lu_dirent *ent)
600 struct lu_dirent *next;
602 if (le16_to_cpu(ent->lde_reclen) != 0)
603 next = ((void *)ent) + le16_to_cpu(ent->lde_reclen);
610 static inline int lu_dirent_calc_size(int namelen, __u16 attr)
614 if (attr & LUDA_TYPE) {
615 const unsigned align = sizeof(struct luda_type) - 1;
616 size = (sizeof(struct lu_dirent) + namelen + align) & ~align;
617 size += sizeof(struct luda_type);
619 size = sizeof(struct lu_dirent) + namelen;
621 return (size + 7) & ~7;
624 static inline int lu_dirent_size(struct lu_dirent *ent)
626 if (le16_to_cpu(ent->lde_reclen) == 0) {
627 return lu_dirent_calc_size(le16_to_cpu(ent->lde_namelen),
628 le32_to_cpu(ent->lde_attrs));
630 return le16_to_cpu(ent->lde_reclen);
633 #define DIR_END_OFF 0xfffffffffffffffeULL
637 struct lustre_handle {
640 #define DEAD_HANDLE_MAGIC 0xdeadbeefcafebabeULL
642 static inline int lustre_handle_is_used(struct lustre_handle *lh)
644 return lh->cookie != 0ull;
647 static inline int lustre_handle_equal(struct lustre_handle *lh1,
648 struct lustre_handle *lh2)
650 return lh1->cookie == lh2->cookie;
653 static inline void lustre_handle_copy(struct lustre_handle *tgt,
654 struct lustre_handle *src)
656 tgt->cookie = src->cookie;
659 /* flags for lm_flags */
660 #define MSGHDR_AT_SUPPORT 0x1
662 #define lustre_msg lustre_msg_v2
663 /* we depend on this structure to be 8-byte aligned */
664 /* this type is only endian-adjusted in lustre_unpack_msg() */
665 struct lustre_msg_v2 {
677 /* without gss, ptlrpc_body is put at the first buffer. */
678 #define PTLRPC_NUM_VERSIONS 4
680 struct lustre_handle pb_handle;
687 __u64 pb_last_committed;
692 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
693 __u32 pb_service_time; /* for rep, actual service time */
696 /* VBR: pre-versions */
697 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
698 /* padding for future needs */
702 extern void lustre_swab_ptlrpc_body(struct ptlrpc_body *pb);
704 /* message body offset for lustre_msg_v2 */
705 /* ptlrpc body offset in all request/reply messages */
706 #define MSG_PTLRPC_BODY_OFF 0
708 /* normal request/reply message record offset */
709 #define REQ_REC_OFF 1
710 #define REPLY_REC_OFF 1
712 /* ldlm request message body offset */
713 #define DLM_LOCKREQ_OFF 1 /* lockreq offset */
714 #define DLM_REQ_REC_OFF 2 /* normal dlm request record offset */
716 /* ldlm intent lock message body offset */
717 #define DLM_INTENT_IT_OFF 2 /* intent lock it offset */
718 #define DLM_INTENT_REC_OFF 3 /* intent lock record offset */
720 /* ldlm reply message body offset */
721 #define DLM_LOCKREPLY_OFF 1 /* lockrep offset */
722 #define DLM_REPLY_REC_OFF 2 /* reply record offset */
724 /** only use in req->rq_{req,rep}_swab_mask */
725 #define MSG_PTLRPC_HEADER_OFF 31
727 /* Flags that are operation-specific go in the top 16 bits. */
728 #define MSG_OP_FLAG_MASK 0xffff0000
729 #define MSG_OP_FLAG_SHIFT 16
731 /* Flags that apply to all requests are in the bottom 16 bits */
732 #define MSG_GEN_FLAG_MASK 0x0000ffff
733 #define MSG_LAST_REPLAY 0x0001
734 #define MSG_RESENT 0x0002
735 #define MSG_REPLAY 0x0004
736 /* #define MSG_AT_SUPPORT 0x0008
737 * This was used in early prototypes of adaptive timeouts, and while there
738 * shouldn't be any users of that code there also isn't a need for using this
739 * bits. Defer usage until at least 1.10 to avoid potential conflict. */
740 #define MSG_DELAY_REPLAY 0x0010
741 #define MSG_VERSION_REPLAY 0x0020
742 #define MSG_REQ_REPLAY_DONE 0x0040
743 #define MSG_LOCK_REPLAY_DONE 0x0080
746 * Flags for all connect opcodes (MDS_CONNECT, OST_CONNECT)
749 #define MSG_CONNECT_RECOVERING 0x00000001
750 #define MSG_CONNECT_RECONNECT 0x00000002
751 #define MSG_CONNECT_REPLAYABLE 0x00000004
752 //#define MSG_CONNECT_PEER 0x8
753 #define MSG_CONNECT_LIBCLIENT 0x00000010
754 #define MSG_CONNECT_INITIAL 0x00000020
755 #define MSG_CONNECT_ASYNC 0x00000040
756 #define MSG_CONNECT_NEXT_VER 0x00000080 /* use next version of lustre_msg */
757 #define MSG_CONNECT_TRANSNO 0x00000100 /* report transno */
760 #define OBD_CONNECT_RDONLY 0x1ULL /*client has read-only access*/
761 #define OBD_CONNECT_INDEX 0x2ULL /*connect specific LOV idx */
762 #define OBD_CONNECT_MDS 0x4ULL /*connect from MDT to OST */
763 #define OBD_CONNECT_GRANT 0x8ULL /*OSC gets grant at connect */
764 #define OBD_CONNECT_SRVLOCK 0x10ULL /*server takes locks for cli */
765 #define OBD_CONNECT_VERSION 0x20ULL /*Lustre versions in ocd */
766 #define OBD_CONNECT_REQPORTAL 0x40ULL /*Separate non-IO req portal */
767 #define OBD_CONNECT_ACL 0x80ULL /*access control lists */
768 #define OBD_CONNECT_XATTR 0x100ULL /*client use extended attr */
769 #define OBD_CONNECT_CROW 0x200ULL /*MDS+OST create obj on write*/
770 #define OBD_CONNECT_TRUNCLOCK 0x400ULL /*locks on server for punch */
771 #define OBD_CONNECT_TRANSNO 0x800ULL /*replay sends init transno */
772 #define OBD_CONNECT_IBITS 0x1000ULL /*support for inodebits locks*/
773 #define OBD_CONNECT_JOIN 0x2000ULL /*files can be concatenated.
774 *We do not support JOIN FILE
775 *anymore, reserve this flags
776 *just for preventing such bit
778 #define OBD_CONNECT_ATTRFID 0x4000ULL /*Server can GetAttr By Fid*/
779 #define OBD_CONNECT_NODEVOH 0x8000ULL /*No open hndl on specl nodes*/
780 #define OBD_CONNECT_RMT_CLIENT 0x10000ULL /*Remote client */
781 #define OBD_CONNECT_RMT_CLIENT_FORCE 0x20000ULL /*Remote client by force */
782 #define OBD_CONNECT_BRW_SIZE 0x40000ULL /*Max bytes per rpc */
783 #define OBD_CONNECT_QUOTA64 0x80000ULL /*64bit qunit_data.qd_count */
784 #define OBD_CONNECT_MDS_CAPA 0x100000ULL /*MDS capability */
785 #define OBD_CONNECT_OSS_CAPA 0x200000ULL /*OSS capability */
786 #define OBD_CONNECT_CANCELSET 0x400000ULL /*Early batched cancels. */
787 #define OBD_CONNECT_SOM 0x800000ULL /*Size on MDS */
788 #define OBD_CONNECT_AT 0x1000000ULL /*client uses AT */
789 #define OBD_CONNECT_LRU_RESIZE 0x2000000ULL /*LRU resize feature. */
790 #define OBD_CONNECT_MDS_MDS 0x4000000ULL /*MDS-MDS connection */
791 #define OBD_CONNECT_REAL 0x8000000ULL /*real connection */
792 #define OBD_CONNECT_CHANGE_QS 0x10000000ULL /*shrink/enlarge qunit */
793 #define OBD_CONNECT_CKSUM 0x20000000ULL /*support several cksum algos*/
794 #define OBD_CONNECT_FID 0x40000000ULL /*FID is supported by server */
795 #define OBD_CONNECT_VBR 0x80000000ULL /*version based recovery */
796 #define OBD_CONNECT_LOV_V3 0x100000000ULL /*client supports LOV v3 EA */
797 #define OBD_CONNECT_GRANT_SHRINK 0x200000000ULL /* support grant shrink */
798 #define OBD_CONNECT_SKIP_ORPHAN 0x400000000ULL /* don't reuse orphan objids */
799 /* also update obd_connect_names[] for lprocfs_rd_connect_flags()
800 * and lustre/utils/wirecheck.c */
802 #ifdef HAVE_LRU_RESIZE_SUPPORT
803 #define LRU_RESIZE_CONNECT_FLAG OBD_CONNECT_LRU_RESIZE
805 #define LRU_RESIZE_CONNECT_FLAG 0
808 #define MDT_CONNECT_SUPPORTED (OBD_CONNECT_RDONLY | OBD_CONNECT_VERSION | \
809 OBD_CONNECT_ACL | OBD_CONNECT_XATTR | \
810 OBD_CONNECT_IBITS | OBD_CONNECT_JOIN | \
811 OBD_CONNECT_NODEVOH |/* OBD_CONNECT_ATTRFID |*/\
812 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
813 OBD_CONNECT_RMT_CLIENT | \
814 OBD_CONNECT_RMT_CLIENT_FORCE | \
815 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA | \
816 OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID | \
817 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_VBR | \
818 OBD_CONNECT_LOV_V3 | OBD_CONNECT_SOM)
819 #define OST_CONNECT_SUPPORTED (OBD_CONNECT_SRVLOCK | OBD_CONNECT_GRANT | \
820 OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | \
821 OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_INDEX | \
822 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_QUOTA64 | \
823 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
824 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_CKSUM | \
825 OBD_CONNECT_CHANGE_QS | \
826 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_RMT_CLIENT | \
827 OBD_CONNECT_RMT_CLIENT_FORCE | OBD_CONNECT_VBR | \
828 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN | \
829 OBD_CONNECT_GRANT_SHRINK)
830 #define ECHO_CONNECT_SUPPORTED (0)
831 #define MGS_CONNECT_SUPPORTED (OBD_CONNECT_VERSION | OBD_CONNECT_AT)
833 #define OBD_OCD_VERSION(major,minor,patch,fix) (((major)<<24) + ((minor)<<16) +\
834 ((patch)<<8) + (fix))
835 #define OBD_OCD_VERSION_MAJOR(version) ((int)((version)>>24)&255)
836 #define OBD_OCD_VERSION_MINOR(version) ((int)((version)>>16)&255)
837 #define OBD_OCD_VERSION_PATCH(version) ((int)((version)>>8)&255)
838 #define OBD_OCD_VERSION_FIX(version) ((int)(version)&255)
840 /* This structure is used for both request and reply.
842 * If we eventually have separate connect data for different types, which we
843 * almost certainly will, then perhaps we stick a union in here. */
844 struct obd_connect_data {
845 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
846 __u32 ocd_version; /* lustre release version number */
847 __u32 ocd_grant; /* initial cache grant amount (bytes) */
848 __u32 ocd_index; /* LOV index to connect to */
849 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
850 __u64 ocd_ibits_known; /* inode bits this client understands */
851 __u32 ocd_nllu; /* non-local-lustre-user */
852 __u32 ocd_nllg; /* non-local-lustre-group */
853 __u64 ocd_transno; /* first transno from client to be replayed */
854 __u32 ocd_group; /* MDS group on OST */
855 __u32 ocd_cksum_types; /* supported checksum algorithms */
856 __u64 padding1; /* also fix lustre_swab_connect */
857 __u64 padding2; /* also fix lustre_swab_connect */
860 extern void lustre_swab_connect(struct obd_connect_data *ocd);
863 * Supported checksum algorithms. Up to 32 checksum types are supported.
864 * (32-bit mask stored in obd_connect_data::ocd_cksum_types)
865 * Please update DECLARE_CKSUM_NAME/OBD_CKSUM_ALL in obd.h when adding a new
866 * algorithm and also the OBD_FL_CKSUM* flags.
869 OBD_CKSUM_CRC32 = 0x00000001,
870 OBD_CKSUM_ADLER = 0x00000002,
874 * OST requests: OBDO & OBD request records
879 OST_REPLY = 0, /* reply ? */
897 OST_QUOTA_ADJUST_QUNIT = 20,
900 #define OST_FIRST_OPC OST_REPLY
902 typedef __u64 obd_id;
903 typedef __u64 obd_gr;
904 typedef __u64 obd_time;
905 typedef __u64 obd_size;
906 typedef __u64 obd_off;
907 typedef __u64 obd_blocks;
908 typedef __u64 obd_valid;
909 typedef __u32 obd_blksize;
910 typedef __u32 obd_mode;
911 typedef __u32 obd_uid;
912 typedef __u32 obd_gid;
913 typedef __u32 obd_flag;
914 typedef __u32 obd_count;
917 OBD_FL_INLINEDATA = 0x00000001,
918 OBD_FL_OBDMDEXISTS = 0x00000002,
919 OBD_FL_DELORPHAN = 0x00000004, /* if set in o_flags delete orphans */
920 OBD_FL_NORPC = 0x00000008, /* set in o_flags do in OSC not OST */
921 OBD_FL_IDONLY = 0x00000010, /* set in o_flags only adjust obj id*/
922 OBD_FL_RECREATE_OBJS= 0x00000020, /* recreate missing obj */
923 OBD_FL_DEBUG_CHECK = 0x00000040, /* echo client/server debug check */
924 OBD_FL_NO_USRQUOTA = 0x00000100, /* the object's owner is over quota */
925 OBD_FL_NO_GRPQUOTA = 0x00000200, /* the object's group is over quota */
926 OBD_FL_CREATE_CROW = 0x00000400, /* object should be create on write */
927 OBD_FL_SRVLOCK = 0x00000800, /* delegate DLM locking to server */
928 OBD_FL_CKSUM_CRC32 = 0x00001000, /* CRC32 checksum type */
929 OBD_FL_CKSUM_ADLER = 0x00002000, /* ADLER checksum type */
930 OBD_FL_CKSUM_RSVD1 = 0x00004000, /* for future cksum types */
931 OBD_FL_CKSUM_RSVD2 = 0x00008000, /* for future cksum types */
932 OBD_FL_CKSUM_RSVD3 = 0x00010000, /* for future cksum types */
933 OBD_FL_SHRINK_GRANT = 0x00020000, /* object shrink the grant */
935 OBD_FL_CKSUM_ALL = OBD_FL_CKSUM_CRC32 | OBD_FL_CKSUM_ADLER,
937 /* mask for local-only flag, which won't be sent over network */
938 OBD_FL_LOCAL_MASK = 0xF0000000,
939 /* temporary OBDO used by osc_brw_async (see bug 18364) */
940 OBD_FL_TEMPORARY = 0x10000000,
943 #define LOV_MAGIC_V1 0x0BD10BD0
944 #define LOV_MAGIC LOV_MAGIC_V1
945 #define LOV_MAGIC_JOIN_V1 0x0BD20BD0
946 #define LOV_MAGIC_V3 0x0BD30BD0
948 #define LOV_PATTERN_RAID0 0x001 /* stripes are used round-robin */
949 #define LOV_PATTERN_RAID1 0x002 /* stripes are mirrors of each other */
950 #define LOV_PATTERN_FIRST 0x100 /* first stripe is not in round-robin */
951 #define LOV_PATTERN_CMOBD 0x200
953 #define LOV_OBJECT_GROUP_DEFAULT ~0ULL
954 #define LOV_OBJECT_GROUP_CLEAR 0ULL
956 #define lov_ost_data lov_ost_data_v1
957 struct lov_ost_data_v1 { /* per-stripe data structure (little-endian)*/
958 __u64 l_object_id; /* OST object ID */
959 __u64 l_object_gr; /* OST object group (creating MDS number) */
960 __u32 l_ost_gen; /* generation of this l_ost_idx */
961 __u32 l_ost_idx; /* OST index in LOV (lov_tgt_desc->tgts) */
964 #define lov_mds_md lov_mds_md_v1
965 struct lov_mds_md_v1 { /* LOV EA mds/wire data (little-endian) */
966 __u32 lmm_magic; /* magic number = LOV_MAGIC_V1 */
967 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
968 __u64 lmm_object_id; /* LOV object ID */
969 __u64 lmm_object_gr; /* LOV object group */
970 __u32 lmm_stripe_size; /* size of stripe in bytes */
971 __u32 lmm_stripe_count; /* num stripes in use for this object */
972 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
975 /* extern void lustre_swab_lov_mds_md(struct lov_mds_md *llm); */
977 #define MAX_MD_SIZE (sizeof(struct lov_mds_md) + 4 * sizeof(struct lov_ost_data))
978 #define MIN_MD_SIZE (sizeof(struct lov_mds_md) + 1 * sizeof(struct lov_ost_data))
980 #define XATTR_NAME_ACL_ACCESS "system.posix_acl_access"
981 #define XATTR_NAME_ACL_DEFAULT "system.posix_acl_default"
982 #define XATTR_USER_PREFIX "user."
983 #define XATTR_TRUSTED_PREFIX "trusted."
984 #define XATTR_SECURITY_PREFIX "security."
985 #define XATTR_LUSTRE_PREFIX "lustre."
987 #define XATTR_NAME_LOV "trusted.lov"
988 #define XATTR_NAME_LMA "trusted.lma"
989 #define XATTR_NAME_LMV "trusted.lmv"
990 #define XATTR_NAME_LINK "trusted.link"
993 struct lov_mds_md_v3 { /* LOV EA mds/wire data (little-endian) */
994 __u32 lmm_magic; /* magic number = LOV_MAGIC_V3 */
995 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
996 __u64 lmm_object_id; /* LOV object ID */
997 __u64 lmm_object_gr; /* LOV object group */
998 __u32 lmm_stripe_size; /* size of stripe in bytes */
999 __u32 lmm_stripe_count; /* num stripes in use for this object */
1000 char lmm_pool_name[LOV_MAXPOOLNAME]; /* must be 32bit aligned */
1001 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1005 #define OBD_MD_FLID (0x00000001ULL) /* object ID */
1006 #define OBD_MD_FLATIME (0x00000002ULL) /* access time */
1007 #define OBD_MD_FLMTIME (0x00000004ULL) /* data modification time */
1008 #define OBD_MD_FLCTIME (0x00000008ULL) /* change time */
1009 #define OBD_MD_FLSIZE (0x00000010ULL) /* size */
1010 #define OBD_MD_FLBLOCKS (0x00000020ULL) /* allocated blocks count */
1011 #define OBD_MD_FLBLKSZ (0x00000040ULL) /* block size */
1012 #define OBD_MD_FLMODE (0x00000080ULL) /* access bits (mode & ~S_IFMT) */
1013 #define OBD_MD_FLTYPE (0x00000100ULL) /* object type (mode & S_IFMT) */
1014 #define OBD_MD_FLUID (0x00000200ULL) /* user ID */
1015 #define OBD_MD_FLGID (0x00000400ULL) /* group ID */
1016 #define OBD_MD_FLFLAGS (0x00000800ULL) /* flags word */
1017 #define OBD_MD_FLNLINK (0x00002000ULL) /* link count */
1018 #define OBD_MD_FLGENER (0x00004000ULL) /* generation number */
1019 /*#define OBD_MD_FLINLINE (0x00008000ULL) inline data. used until 1.6.5 */
1020 #define OBD_MD_FLRDEV (0x00010000ULL) /* device number */
1021 #define OBD_MD_FLEASIZE (0x00020000ULL) /* extended attribute data */
1022 #define OBD_MD_LINKNAME (0x00040000ULL) /* symbolic link target */
1023 #define OBD_MD_FLHANDLE (0x00080000ULL) /* file/lock handle */
1024 #define OBD_MD_FLCKSUM (0x00100000ULL) /* bulk data checksum */
1025 #define OBD_MD_FLQOS (0x00200000ULL) /* quality of service stats */
1026 /*#define OBD_MD_FLOSCOPQ (0x00400000ULL) osc opaque data, never used */
1027 #define OBD_MD_FLCOOKIE (0x00800000ULL) /* log cancellation cookie */
1028 #define OBD_MD_FLGROUP (0x01000000ULL) /* group */
1029 #define OBD_MD_FLFID (0x02000000ULL) /* ->ost write inline fid */
1030 #define OBD_MD_FLEPOCH (0x04000000ULL) /* ->ost write with ioepoch */
1031 /* ->mds if epoch opens or closes */
1032 #define OBD_MD_FLGRANT (0x08000000ULL) /* ost preallocation space grant */
1033 #define OBD_MD_FLDIREA (0x10000000ULL) /* dir's extended attribute data */
1034 #define OBD_MD_FLUSRQUOTA (0x20000000ULL) /* over quota flags sent from ost */
1035 #define OBD_MD_FLGRPQUOTA (0x40000000ULL) /* over quota flags sent from ost */
1036 #define OBD_MD_FLMODEASIZE (0x80000000ULL) /* EA size will be changed */
1038 #define OBD_MD_MDS (0x0000000100000000ULL) /* where an inode lives on */
1039 #define OBD_MD_REINT (0x0000000200000000ULL) /* reintegrate oa */
1040 #define OBD_MD_MEA (0x0000000400000000ULL) /* CMD split EA */
1041 #define OBD_MD_MDTIDX (0x0000000800000000ULL) /* Get MDT index */
1043 #define OBD_MD_FLXATTR (0x0000001000000000ULL) /* xattr */
1044 #define OBD_MD_FLXATTRLS (0x0000002000000000ULL) /* xattr list */
1045 #define OBD_MD_FLXATTRRM (0x0000004000000000ULL) /* xattr remove */
1046 #define OBD_MD_FLACL (0x0000008000000000ULL) /* ACL */
1047 #define OBD_MD_FLRMTPERM (0x0000010000000000ULL) /* remote permission */
1048 #define OBD_MD_FLMDSCAPA (0x0000020000000000ULL) /* MDS capability */
1049 #define OBD_MD_FLOSSCAPA (0x0000040000000000ULL) /* OSS capability */
1050 #define OBD_MD_FLCKSPLIT (0x0000080000000000ULL) /* Check split on server */
1051 #define OBD_MD_FLCROSSREF (0x0000100000000000ULL) /* Cross-ref case */
1052 #define OBD_MD_FLGETATTRLOCK (0x0000200000000000ULL) /* Get IOEpoch attributes
1054 #define OBD_FL_TRUNC (0x0000200000000000ULL) /* for filter_truncate */
1056 #define OBD_MD_FLRMTLSETFACL (0x0001000000000000ULL) /* lfs lsetfacl case */
1057 #define OBD_MD_FLRMTLGETFACL (0x0002000000000000ULL) /* lfs lgetfacl case */
1058 #define OBD_MD_FLRMTRSETFACL (0x0004000000000000ULL) /* lfs rsetfacl case */
1059 #define OBD_MD_FLRMTRGETFACL (0x0008000000000000ULL) /* lfs rgetfacl case */
1061 #define OBD_MD_FLGETATTR (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLMTIME | \
1062 OBD_MD_FLCTIME | OBD_MD_FLSIZE | OBD_MD_FLBLKSZ | \
1063 OBD_MD_FLMODE | OBD_MD_FLTYPE | OBD_MD_FLUID | \
1064 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLNLINK | \
1065 OBD_MD_FLGENER | OBD_MD_FLRDEV | OBD_MD_FLGROUP)
1067 /* don't forget obdo_fid which is way down at the bottom so it can
1068 * come after the definition of llog_cookie */
1071 extern void lustre_swab_obd_statfs (struct obd_statfs *os);
1072 #define OBD_STATFS_NODELAY 0x0001 /* requests should be send without delay
1073 * and resends for avoid deadlocks */
1074 #define OBD_STATFS_FROM_CACHE 0x0002 /* the statfs callback should not update
1076 #define OBD_STATFS_PTLRPCD 0x0004 /* requests will be sent via ptlrpcd
1077 * instead of a specific set. This
1078 * means that we cannot rely on the set
1079 * interpret routine to be called.
1080 * lov_statfs_fini() must thus be called
1081 * by the request interpret routine */
1083 /* ost_body.data values for OST_BRW */
1085 #define OBD_BRW_READ 0x01
1086 #define OBD_BRW_WRITE 0x02
1087 #define OBD_BRW_RWMASK (OBD_BRW_READ | OBD_BRW_WRITE)
1088 #define OBD_BRW_SYNC 0x08 /* this page is a part of synchronous
1089 * transfer and is not accounted in
1091 #define OBD_BRW_CHECK 0x10
1092 #define OBD_BRW_FROM_GRANT 0x20 /* the osc manages this under llite */
1093 #define OBD_BRW_GRANTED 0x40 /* the ost manages this */
1094 #define OBD_BRW_NOCACHE 0x80 /* this page is a part of non-cached IO */
1095 #define OBD_BRW_NOQUOTA 0x100
1096 #define OBD_BRW_SRVLOCK 0x200 /* Client holds no lock over this page */
1097 #define OBD_BRW_MEMALLOC 0x800 /* Client runs in the "kswapd" context */
1099 #define OBD_OBJECT_EOF 0xffffffffffffffffULL
1101 #define OST_MIN_PRECREATE 32
1102 #define OST_MAX_PRECREATE 20000
1111 extern void lustre_swab_obd_ioobj (struct obd_ioobj *ioo);
1113 /* multiple of 8 bytes => can array */
1114 struct niobuf_remote {
1120 extern void lustre_swab_niobuf_remote (struct niobuf_remote *nbr);
1122 /* lock value block communicated between the filter and llite */
1124 /* OST_LVB_ERR_INIT is needed because the return code in rc is
1125 * negative, i.e. because ((MASK + rc) & MASK) != MASK. */
1126 #define OST_LVB_ERR_INIT 0xffbadbad80000000ULL
1127 #define OST_LVB_ERR_MASK 0xffbadbad00000000ULL
1128 #define OST_LVB_IS_ERR(blocks) \
1129 ((blocks & OST_LVB_ERR_MASK) == OST_LVB_ERR_MASK)
1130 #define OST_LVB_SET_ERR(blocks, rc) \
1131 do { blocks = OST_LVB_ERR_INIT + rc; } while (0)
1132 #define OST_LVB_GET_ERR(blocks) (int)(blocks - OST_LVB_ERR_INIT)
1142 extern void lustre_swab_ost_lvb(struct ost_lvb *);
1151 MDS_GETATTR_NAME = 34,
1156 MDS_DISCONNECT = 39,
1162 MDS_DONE_WRITING = 45,
1164 MDS_QUOTACHECK = 47,
1167 MDS_SETXATTR = 50, /* obsolete, now it's MDS_REINT op */
1174 #define MDS_FIRST_OPC MDS_GETATTR
1191 } mds_reint_t, mdt_reint_t;
1193 extern void lustre_swab_generic_32s (__u32 *val);
1195 /* the disposition of the intent outlines what was executed */
1196 #define DISP_IT_EXECD 0x00000001
1197 #define DISP_LOOKUP_EXECD 0x00000002
1198 #define DISP_LOOKUP_NEG 0x00000004
1199 #define DISP_LOOKUP_POS 0x00000008
1200 #define DISP_OPEN_CREATE 0x00000010
1201 #define DISP_OPEN_OPEN 0x00000020
1202 #define DISP_ENQ_COMPLETE 0x00400000
1203 #define DISP_ENQ_OPEN_REF 0x00800000
1204 #define DISP_ENQ_CREATE_REF 0x01000000
1205 #define DISP_OPEN_LOCK 0x02000000
1207 /* INODE LOCK PARTS */
1208 #define MDS_INODELOCK_LOOKUP 0x000001 /* dentry, mode, owner, group */
1209 #define MDS_INODELOCK_UPDATE 0x000002 /* size, links, timestamps */
1210 #define MDS_INODELOCK_OPEN 0x000004 /* For opened files */
1212 /* Do not forget to increase MDS_INODELOCK_MAXSHIFT when adding new bits */
1213 #define MDS_INODELOCK_MAXSHIFT 2
1214 /* This FULL lock is useful to take on unlink sort of operations */
1215 #define MDS_INODELOCK_FULL ((1<<(MDS_INODELOCK_MAXSHIFT+1))-1)
1217 extern void lustre_swab_ll_fid (struct ll_fid *fid);
1219 #define MDS_STATUS_CONN 1
1220 #define MDS_STATUS_LOV 2
1222 struct mds_status_req {
1227 extern void lustre_swab_mds_status_req (struct mds_status_req *r);
1229 /* mdt_thread_info.mti_flags. */
1231 /* The flag indicates Size-on-MDS attributes are changed. */
1232 MF_SOM_CHANGE = (1 << 0),
1233 /* Flags indicates an epoch opens or closes. */
1234 MF_EPOCH_OPEN = (1 << 1),
1235 MF_EPOCH_CLOSE = (1 << 2),
1236 MF_MDC_CANCEL_FID1 = (1 << 3),
1237 MF_MDC_CANCEL_FID2 = (1 << 4),
1238 MF_MDC_CANCEL_FID3 = (1 << 5),
1239 MF_MDC_CANCEL_FID4 = (1 << 6),
1240 /* There is a pending attribute update. */
1241 MF_SOM_AU = (1 << 7),
1242 /* Cancel OST locks while getattr OST attributes. */
1243 MF_GETATTR_LOCK = (1 << 8),
1246 #define MF_SOM_LOCAL_FLAGS (MF_SOM_CHANGE | MF_EPOCH_OPEN | MF_EPOCH_CLOSE)
1248 #define MDS_BFLAG_UNCOMMITTED_WRITES 0x1
1250 /* these should be identical to their EXT3_*_FL counterparts, and are
1251 * redefined here only to avoid dragging in ext3_fs.h */
1252 #define MDS_SYNC_FL 0x00000008 /* Synchronous updates */
1253 #define MDS_IMMUTABLE_FL 0x00000010 /* Immutable file */
1254 #define MDS_APPEND_FL 0x00000020 /* writes to file may only append */
1255 #define MDS_NOATIME_FL 0x00000080 /* do not update atime */
1256 #define MDS_DIRSYNC_FL 0x00010000 /* dirsync behaviour (dir only) */
1259 /* Convert wire MDS_*_FL to corresponding client local VFS S_* values
1260 * for the client inode i_flags. The MDS_*_FL are the Lustre wire
1261 * protocol equivalents of LDISKFS_*_FL values stored on disk, while
1262 * the S_* flags are kernel-internal values that change between kernel
1263 * versions. These flags are set/cleared via FSFILT_IOC_{GET,SET}_FLAGS.
1264 * See b=16526 for a full history. */
1265 static inline int ll_ext_to_inode_flags(int flags)
1267 return (((flags & MDS_SYNC_FL) ? S_SYNC : 0) |
1268 ((flags & MDS_NOATIME_FL) ? S_NOATIME : 0) |
1269 ((flags & MDS_APPEND_FL) ? S_APPEND : 0) |
1270 #if defined(S_DIRSYNC)
1271 ((flags & MDS_DIRSYNC_FL) ? S_DIRSYNC : 0) |
1273 ((flags & MDS_IMMUTABLE_FL) ? S_IMMUTABLE : 0));
1276 static inline int ll_inode_to_ext_flags(int iflags)
1278 return (((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);
1368 struct mdt_ioepoch {
1369 struct lustre_handle handle;
1375 extern void lustre_swab_mdt_ioepoch (struct mdt_ioepoch *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 is shared among quota structures */
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 /* flags is specific for quota_adjust_qunit */
1430 #define LQUOTA_QAQ_CREATE_LQS (1 << 31) /* when it is set, need create lqs */
1432 /* the status of lqs_flags in struct lustre_qunit_size */
1433 #define LQUOTA_QUNIT_FLAGS (LQUOTA_FLAGS_GRP | LQUOTA_FLAGS_BLK)
1435 #define QAQ_IS_GRP(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_GRP)
1436 #define QAQ_IS_ADJBLK(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_ADJBLK)
1437 #define QAQ_IS_ADJINO(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_ADJINO)
1438 #define QAQ_IS_CREATE_LQS(qaq) ((qaq)->qaq_flags & LQUOTA_QAQ_CREATE_LQS)
1440 #define QAQ_SET_GRP(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_GRP)
1441 #define QAQ_SET_ADJBLK(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_ADJBLK)
1442 #define QAQ_SET_ADJINO(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_ADJINO)
1443 #define QAQ_SET_CREATE_LQS(qaq) ((qaq)->qaq_flags |= LQUOTA_QAQ_CREATE_LQS)
1445 /* inode access permission for remote user, the inode info are omitted,
1446 * for client knows them. */
1447 struct mds_remote_perm {
1452 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
1455 /* permissions for md_perm.mp_perm */
1457 CFS_SETUID_PERM = 0x01,
1458 CFS_SETGID_PERM = 0x02,
1459 CFS_SETGRP_PERM = 0x04,
1460 CFS_RMTACL_PERM = 0x08,
1461 CFS_RMTOWN_PERM = 0x10
1464 extern void lustre_swab_mds_remote_perm(struct mds_remote_perm *p);
1466 struct mdt_remote_perm {
1471 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
1474 extern void lustre_swab_mdt_remote_perm(struct mdt_remote_perm *p);
1476 struct mdt_rec_setattr {
1486 __u32 sa_padding_1_h;
1487 struct lu_fid sa_fid;
1496 __u32 sa_attr_flags;
1504 extern void lustre_swab_mdt_rec_setattr (struct mdt_rec_setattr *sa);
1507 * Attribute flags used in mdt_rec_setattr::sa_valid.
1508 * The kernel's #defines for ATTR_* should not be used over the network
1509 * since the client and MDS may run different kernels (see bug 13828)
1510 * Therefore, we should only use MDS_ATTR_* attributes for sa_valid.
1512 #define MDS_ATTR_MODE 0x1ULL /* = 1 */
1513 #define MDS_ATTR_UID 0x2ULL /* = 2 */
1514 #define MDS_ATTR_GID 0x4ULL /* = 4 */
1515 #define MDS_ATTR_SIZE 0x8ULL /* = 8 */
1516 #define MDS_ATTR_ATIME 0x10ULL /* = 16 */
1517 #define MDS_ATTR_MTIME 0x20ULL /* = 32 */
1518 #define MDS_ATTR_CTIME 0x40ULL /* = 64 */
1519 #define MDS_ATTR_ATIME_SET 0x80ULL /* = 128 */
1520 #define MDS_ATTR_MTIME_SET 0x100ULL /* = 256 */
1521 #define MDS_ATTR_FORCE 0x200ULL /* = 512, Not a change, but a change it */
1522 #define MDS_ATTR_ATTR_FLAG 0x400ULL /* = 1024 */
1523 #define MDS_ATTR_KILL_SUID 0x800ULL /* = 2048 */
1524 #define MDS_ATTR_KILL_SGID 0x1000ULL /* = 4096 */
1525 #define MDS_ATTR_CTIME_SET 0x2000ULL /* = 8192 */
1526 #define MDS_ATTR_FROM_OPEN 0x4000ULL /* = 16384, called from open path, ie O_TRUNC */
1527 #define MDS_ATTR_BLOCKS 0x8000ULL /* = 32768 */
1530 #define FMODE_READ 00000001
1531 #define FMODE_WRITE 00000002
1534 /* IO Epoch is opened on a closed file. */
1535 #define FMODE_EPOCH 01000000
1536 /* IO Epoch is opened on a file truncate. */
1537 #define FMODE_TRUNC 02000000
1538 /* Size-on-MDS Attribute Update is pending. */
1539 #define FMODE_SOM 04000000
1540 #define FMODE_CLOSED 0
1542 #define MDS_OPEN_CREATED 00000010
1543 #define MDS_OPEN_CROSS 00000020
1545 #define MDS_FMODE_EXEC 00000004
1546 #define MDS_OPEN_CREAT 00000100
1547 #define MDS_OPEN_EXCL 00000200
1548 #define MDS_OPEN_TRUNC 00001000
1549 #define MDS_OPEN_APPEND 00002000
1550 #define MDS_OPEN_SYNC 00010000
1551 #define MDS_OPEN_DIRECTORY 00200000
1553 #define MDS_OPEN_DELAY_CREATE 0100000000 /* delay initial object create */
1554 #define MDS_OPEN_OWNEROVERRIDE 0200000000 /* NFSD rw-reopen ro file for owner */
1555 #define MDS_OPEN_JOIN_FILE 0400000000 /* open for join file.
1556 * We do not support JOIN FILE
1557 * anymore, reserve this flags
1558 * just for preventing such bit
1560 #define MDS_CREATE_RMT_ACL 01000000000 /* indicate create on remote server
1561 * with default ACL */
1562 #define MDS_CREATE_SLAVE_OBJ 02000000000 /* indicate create slave object
1563 * actually, this is for create, not
1564 * conflict with other open flags */
1565 #define MDS_OPEN_LOCK 04000000000 /* This open requires open lock */
1566 #define MDS_OPEN_HAS_EA 010000000000 /* specify object create pattern */
1567 #define MDS_OPEN_HAS_OBJS 020000000000 /* Just set the EA the obj exist */
1569 /* permission for create non-directory file */
1570 #define MAY_CREATE (1 << 7)
1571 /* permission for create directory file */
1572 #define MAY_LINK (1 << 8)
1573 /* permission for delete from the directory */
1574 #define MAY_UNLINK (1 << 9)
1575 /* source's permission for rename */
1576 #define MAY_RENAME_SRC (1 << 10)
1577 /* target's permission for rename */
1578 #define MAY_RENAME_TAR (1 << 11)
1579 /* part (parent's) VTX permission check */
1580 #define MAY_VTX_PART (1 << 12)
1581 /* full VTX permission check */
1582 #define MAY_VTX_FULL (1 << 13)
1583 /* lfs rgetfacl permission check */
1584 #define MAY_RGETFACL (1 << 14)
1587 MDS_CHECK_SPLIT = 1 << 0,
1588 MDS_CROSS_REF = 1 << 1,
1589 MDS_VTX_BYPASS = 1 << 2,
1590 MDS_PERM_BYPASS = 1 << 3,
1592 MDS_QUOTA_IGNORE = 1 << 5,
1593 MDS_CLOSE_CLEANUP = 1 << 6
1596 struct mdt_rec_create {
1604 __u32 cr_suppgid1_h;
1606 __u32 cr_suppgid2_h;
1607 struct lu_fid cr_fid1;
1608 struct lu_fid cr_fid2;
1609 struct lustre_handle cr_old_handle; /* handle in case of open replay */
1613 __u64 cr_padding_1; /* pad for 64 bits*/
1616 __u32 cr_flags; /* for use with open */
1622 struct mdt_rec_link {
1630 __u32 lk_suppgid1_h;
1632 __u32 lk_suppgid2_h;
1633 struct lu_fid lk_fid1;
1634 struct lu_fid lk_fid2;
1648 struct mdt_rec_unlink {
1656 __u32 ul_suppgid1_h;
1658 __u32 ul_suppgid2_h;
1659 struct lu_fid ul_fid1;
1660 struct lu_fid ul_fid2;
1674 struct mdt_rec_rename {
1682 __u32 rn_suppgid1_h;
1684 __u32 rn_suppgid2_h;
1685 struct lu_fid rn_fid1;
1686 struct lu_fid rn_fid2;
1692 __u32 rn_bias; /* some operation flags */
1693 __u32 rn_mode; /* cross-ref rename has mode */
1700 struct mdt_rec_setxattr {
1708 __u32 sx_suppgid1_h;
1710 __u32 sx_suppgid2_h;
1711 struct lu_fid sx_fid;
1712 __u64 sx_padding_1; /* These three members are lu_fid size */
1724 __u32 sx_padding_10;
1725 __u32 sx_padding_11;
1728 struct mdt_rec_reint {
1736 __u32 rr_suppgid1_h;
1738 __u32 rr_suppgid2_h;
1739 struct lu_fid rr_fid1;
1740 struct lu_fid rr_fid2;
1748 __u32 rr_padding_1; /* also fix lustre_swab_mdt_rec_reint */
1749 __u32 rr_padding_2; /* also fix lustre_swab_mdt_rec_reint */
1750 __u32 rr_padding_3; /* also fix lustre_swab_mdt_rec_reint */
1751 __u32 rr_padding_4; /* also fix lustre_swab_mdt_rec_reint */
1754 extern void lustre_swab_mdt_rec_reint(struct mdt_rec_reint *rr);
1757 __u32 ld_tgt_count; /* how many MDS's */
1758 __u32 ld_active_tgt_count; /* how many active */
1759 __u32 ld_default_stripe_count; /* how many objects are used */
1760 __u32 ld_pattern; /* default MEA_MAGIC_* */
1761 __u64 ld_default_hash_size;
1762 __u64 ld_padding_1; /* also fix lustre_swab_lmv_desc */
1763 __u32 ld_padding_2; /* also fix lustre_swab_lmv_desc */
1764 __u32 ld_qos_maxage; /* in second */
1765 __u32 ld_padding_3; /* also fix lustre_swab_lmv_desc */
1766 __u32 ld_padding_4; /* also fix lustre_swab_lmv_desc */
1767 struct obd_uuid ld_uuid;
1770 extern void lustre_swab_lmv_desc (struct lmv_desc *ld);
1772 /* TODO: lmv_stripe_md should contain mds capabilities for all slave fids */
1773 struct lmv_stripe_md {
1778 char mea_pool_name[LOV_MAXPOOLNAME];
1779 struct lu_fid mea_ids[0];
1782 extern void lustre_swab_lmv_stripe_md(struct lmv_stripe_md *mea);
1784 /* lmv structures */
1785 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
1786 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
1787 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
1789 #define MAX_HASH_SIZE_32 0x7fffffffUL
1790 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
1791 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
1798 extern void lustre_swab_md_fld (struct md_fld *mf);
1803 FLD_FIRST_OPC = FLD_QUERY
1809 SEQ_FIRST_OPC = SEQ_QUERY
1813 SEQ_ALLOC_SUPER = 0,
1818 * LOV data structures
1821 #define LOV_MIN_STRIPE_BITS 16 /* maximum PAGE_SIZE (ia64), power of 2 */
1822 #define LOV_MIN_STRIPE_SIZE (1<<LOV_MIN_STRIPE_BITS)
1823 #define LOV_MAX_STRIPE_COUNT 160 /* until bug 4424 is fixed */
1824 #define LOV_V1_INSANE_STRIPE_COUNT 65532 /* maximum stripe count bz13933 */
1826 #define LOV_MAX_UUID_BUFFER_SIZE 8192
1827 /* The size of the buffer the lov/mdc reserves for the
1828 * array of UUIDs returned by the MDS. With the current
1829 * protocol, this will limit the max number of OSTs per LOV */
1831 #define LOV_DESC_MAGIC 0xB0CCDE5C
1833 /* LOV settings descriptor (should only contain static info) */
1835 __u32 ld_tgt_count; /* how many OBD's */
1836 __u32 ld_active_tgt_count; /* how many active */
1837 __u32 ld_default_stripe_count; /* how many objects are used */
1838 __u32 ld_pattern; /* default PATTERN_RAID0 */
1839 __u64 ld_default_stripe_size; /* in bytes */
1840 __u64 ld_default_stripe_offset; /* in bytes */
1841 __u32 ld_padding_0; /* unused */
1842 __u32 ld_qos_maxage; /* in second */
1843 __u32 ld_padding_1; /* also fix lustre_swab_lov_desc */
1844 __u32 ld_padding_2; /* also fix lustre_swab_lov_desc */
1845 struct obd_uuid ld_uuid;
1848 #define ld_magic ld_active_tgt_count /* for swabbing from llogs */
1850 extern void lustre_swab_lov_desc (struct lov_desc *ld);
1855 /* opcodes -- MUST be distinct from OST/MDS opcodes */
1860 LDLM_BL_CALLBACK = 104,
1861 LDLM_CP_CALLBACK = 105,
1862 LDLM_GL_CALLBACK = 106,
1863 LDLM_SET_INFO = 107,
1866 #define LDLM_FIRST_OPC LDLM_ENQUEUE
1868 #define RES_NAME_SIZE 4
1869 struct ldlm_res_id {
1870 __u64 name[RES_NAME_SIZE];
1873 extern void lustre_swab_ldlm_res_id (struct ldlm_res_id *id);
1889 #define LCK_MODE_NUM 8
1899 #define LDLM_MIN_TYPE LDLM_PLAIN
1901 struct ldlm_extent {
1907 static inline int ldlm_extent_overlap(struct ldlm_extent *ex1,
1908 struct ldlm_extent *ex2)
1910 return (ex1->start <= ex2->end) && (ex2->start <= ex1->end);
1913 struct ldlm_inodebits {
1920 __u64 blocking_export; /* not actually used over the wire */
1921 __u32 blocking_pid; /* not actually used over the wire */
1925 /* it's important that the fields of the ldlm_extent structure match
1926 * the first fields of the ldlm_flock structure because there is only
1927 * one ldlm_swab routine to process the ldlm_policy_data_t union. if
1928 * this ever changes we will need to swab the union differently based
1929 * on the resource type. */
1932 struct ldlm_extent l_extent;
1933 struct ldlm_flock l_flock;
1934 struct ldlm_inodebits l_inodebits;
1935 } ldlm_policy_data_t;
1937 extern void lustre_swab_ldlm_policy_data (ldlm_policy_data_t *d);
1939 struct ldlm_intent {
1943 extern void lustre_swab_ldlm_intent (struct ldlm_intent *i);
1945 struct ldlm_resource_desc {
1946 ldlm_type_t lr_type;
1947 __u32 lr_padding; /* also fix lustre_swab_ldlm_resource_desc */
1948 struct ldlm_res_id lr_name;
1951 extern void lustre_swab_ldlm_resource_desc (struct ldlm_resource_desc *r);
1953 struct ldlm_lock_desc {
1954 struct ldlm_resource_desc l_resource;
1955 ldlm_mode_t l_req_mode;
1956 ldlm_mode_t l_granted_mode;
1957 ldlm_policy_data_t l_policy_data;
1960 extern void lustre_swab_ldlm_lock_desc (struct ldlm_lock_desc *l);
1962 #define LDLM_LOCKREQ_HANDLES 2
1963 #define LDLM_ENQUEUE_CANCEL_OFF 1
1965 struct ldlm_request {
1968 struct ldlm_lock_desc lock_desc;
1969 struct lustre_handle lock_handle[LDLM_LOCKREQ_HANDLES];
1972 extern void lustre_swab_ldlm_request (struct ldlm_request *rq);
1974 /* If LDLM_ENQUEUE, 1 slot is already occupied, 1 is available.
1975 * Otherwise, 2 are available. */
1976 #define ldlm_request_bufsize(count,type) \
1978 int _avail = LDLM_LOCKREQ_HANDLES; \
1979 _avail -= (type == LDLM_ENQUEUE ? LDLM_ENQUEUE_CANCEL_OFF : 0); \
1980 sizeof(struct ldlm_request) + \
1981 (count > _avail ? count - _avail : 0) * \
1982 sizeof(struct lustre_handle); \
1987 __u32 lock_padding; /* also fix lustre_swab_ldlm_reply */
1988 struct ldlm_lock_desc lock_desc;
1989 struct lustre_handle lock_handle;
1990 __u64 lock_policy_res1;
1991 __u64 lock_policy_res2;
1994 extern void lustre_swab_ldlm_reply (struct ldlm_reply *r);
1997 * Opcodes for mountconf (mgs and mgc)
2002 MGS_EXCEPTION, /* node died, etc. */
2003 MGS_TARGET_REG, /* whenever target starts up */
2008 #define MGS_FIRST_OPC MGS_CONNECT
2010 #define MGS_PARAM_MAXLEN 1024
2011 #define KEY_SET_INFO "set_info"
2013 struct mgs_send_param {
2014 char mgs_param[MGS_PARAM_MAXLEN];
2017 /* We pass this info to the MGS so it can write config logs */
2018 #define MTI_NAME_MAXLEN 64
2019 #define MTI_PARAM_MAXLEN 4096
2020 #define MTI_NIDS_MAX 32
2021 struct mgs_target_info {
2022 __u32 mti_lustre_ver;
2023 __u32 mti_stripe_index;
2024 __u32 mti_config_ver;
2026 __u32 mti_nid_count;
2027 __u32 padding; /* 64 bit align */
2028 char mti_fsname[MTI_NAME_MAXLEN];
2029 char mti_svname[MTI_NAME_MAXLEN];
2030 char mti_uuid[sizeof(struct obd_uuid)];
2031 __u64 mti_nids[MTI_NIDS_MAX]; /* host nids (lnet_nid_t)*/
2032 char mti_params[MTI_PARAM_MAXLEN];
2035 extern void lustre_swab_mgs_target_info(struct mgs_target_info *oinfo);
2037 /* Config marker flags (in config log) */
2038 #define CM_START 0x01
2040 #define CM_SKIP 0x04
2041 #define CM_UPGRADE146 0x08
2042 #define CM_EXCLUDE 0x10
2043 #define CM_START_SKIP (CM_START | CM_SKIP)
2046 __u32 cm_step; /* aka config version */
2048 __u32 cm_vers; /* lustre release version number */
2049 __u32 padding; /* 64 bit align */
2050 __u64 cm_createtime; /*when this record was first created */
2051 __u64 cm_canceltime; /*when this record is no longer valid*/
2052 char cm_tgtname[MTI_NAME_MAXLEN];
2053 char cm_comment[MTI_NAME_MAXLEN];
2056 extern void lustre_swab_cfg_marker(struct cfg_marker *marker,
2057 int swab, int size);
2060 * Opcodes for multiple servers.
2069 #define OBD_FIRST_OPC OBD_PING
2071 /* catalog of log objects */
2073 /** Identifier for a single log object */
2078 } __attribute__((packed));
2080 /** Records written to the CATALOGS list */
2081 #define CATLIST "CATALOGS"
2083 struct llog_logid lci_logid;
2087 } __attribute__((packed));
2089 /* Log data record types - there is no specific reason that these need to
2090 * be related to the RPC opcodes, but no reason not to (may be handy later?)
2092 #define LLOG_OP_MAGIC 0x10600000
2093 #define LLOG_OP_MASK 0xfff00000
2096 LLOG_PAD_MAGIC = LLOG_OP_MAGIC | 0x00000,
2097 OST_SZ_REC = LLOG_OP_MAGIC | 0x00f00,
2098 OST_RAID1_REC = LLOG_OP_MAGIC | 0x01000,
2099 MDS_UNLINK_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) | REINT_UNLINK,
2100 MDS_SETATTR_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) | REINT_SETATTR,
2101 MDS_SETATTR64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) | REINT_SETATTR,
2102 OBD_CFG_REC = LLOG_OP_MAGIC | 0x20000,
2103 PTL_CFG_REC = LLOG_OP_MAGIC | 0x30000, /* obsolete */
2104 LLOG_GEN_REC = LLOG_OP_MAGIC | 0x40000,
2105 LLOG_JOIN_REC = LLOG_OP_MAGIC | 0x50000, /* obsolete */
2106 CHANGELOG_REC = LLOG_OP_MAGIC | 0x60000,
2107 CHANGELOG_USER_REC = LLOG_OP_MAGIC | 0x70000,
2108 LLOG_HDR_MAGIC = LLOG_OP_MAGIC | 0x45539,
2109 LLOG_LOGID_MAGIC = LLOG_OP_MAGIC | 0x4553b,
2113 * for now, continue to support old pad records which have 0 for their
2114 * type but still need to be swabbed for their length
2116 #define LLOG_REC_HDR_NEEDS_SWABBING(r) \
2117 (((r)->lrh_type & __swab32(LLOG_OP_MASK)) == \
2118 __swab32(LLOG_OP_MAGIC) || \
2119 (((r)->lrh_type == 0) && ((r)->lrh_len > LLOG_CHUNK_SIZE)))
2121 /** Log record header - stored in little endian order.
2122 * Each record must start with this struct, end with a llog_rec_tail,
2123 * and be a multiple of 256 bits in size.
2125 struct llog_rec_hdr {
2132 struct llog_rec_tail {
2137 struct llog_logid_rec {
2138 struct llog_rec_hdr lid_hdr;
2139 struct llog_logid lid_id;
2145 struct llog_rec_tail lid_tail;
2146 } __attribute__((packed));
2148 struct llog_create_rec {
2149 struct llog_rec_hdr lcr_hdr;
2150 struct ll_fid lcr_fid;
2154 struct llog_rec_tail lcr_tail;
2155 } __attribute__((packed));
2157 struct llog_orphan_rec {
2158 struct llog_rec_hdr lor_hdr;
2162 struct llog_rec_tail lor_tail;
2163 } __attribute__((packed));
2165 struct llog_unlink_rec {
2166 struct llog_rec_hdr lur_hdr;
2169 obd_count lur_count;
2170 struct llog_rec_tail lur_tail;
2171 } __attribute__((packed));
2173 struct llog_setattr_rec {
2174 struct llog_rec_hdr lsr_hdr;
2180 struct llog_rec_tail lsr_tail;
2181 } __attribute__((packed));
2183 struct llog_setattr64_rec {
2184 struct llog_rec_hdr lsr_hdr;
2192 struct llog_rec_tail lsr_tail;
2193 } __attribute__((packed));
2195 struct llog_size_change_rec {
2196 struct llog_rec_hdr lsc_hdr;
2197 struct ll_fid lsc_fid;
2200 struct llog_rec_tail lsc_tail;
2201 } __attribute__((packed));
2203 #define CHANGELOG_MAGIC 0xca103000
2205 /** \a changelog_rec_type's that can't be masked */
2206 #define CHANGELOG_MINMASK (1 << CL_MARK)
2207 /** bits covering all \a changelog_rec_type's */
2208 #define CHANGELOG_ALLMASK 0XFFFFFFFF
2209 /** default \a changelog_rec_type mask */
2210 #define CHANGELOG_DEFMASK CHANGELOG_ALLMASK
2212 /* changelog llog name, needed by client replicators */
2213 #define CHANGELOG_CATALOG "changelog_catalog"
2215 struct changelog_setinfo {
2218 } __attribute__((packed));
2220 /** changelog record */
2221 struct llog_changelog_rec {
2222 struct llog_rec_hdr cr_hdr;
2223 struct changelog_rec cr;
2224 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
2225 } __attribute__((packed));
2227 #define CHANGELOG_USER_PREFIX "cl"
2229 struct llog_changelog_user_rec {
2230 struct llog_rec_hdr cur_hdr;
2234 struct llog_rec_tail cur_tail;
2235 } __attribute__((packed));
2240 } __attribute__((packed));
2242 struct llog_gen_rec {
2243 struct llog_rec_hdr lgr_hdr;
2244 struct llog_gen lgr_gen;
2245 struct llog_rec_tail lgr_tail;
2247 /* On-disk header structure of each log object, stored in little endian order */
2248 #define LLOG_CHUNK_SIZE 8192
2249 #define LLOG_HEADER_SIZE (96)
2250 #define LLOG_BITMAP_BYTES (LLOG_CHUNK_SIZE - LLOG_HEADER_SIZE)
2252 #define LLOG_MIN_REC_SIZE (24) /* round(llog_rec_hdr + llog_rec_tail) */
2254 /* flags for the logs */
2255 #define LLOG_F_ZAP_WHEN_EMPTY 0x1
2256 #define LLOG_F_IS_CAT 0x2
2257 #define LLOG_F_IS_PLAIN 0x4
2259 struct llog_log_hdr {
2260 struct llog_rec_hdr llh_hdr;
2261 __u64 llh_timestamp;
2263 __u32 llh_bitmap_offset;
2267 /* for a catalog the first plain slot is next to it */
2268 struct obd_uuid llh_tgtuuid;
2269 __u32 llh_reserved[LLOG_HEADER_SIZE/sizeof(__u32) - 23];
2270 __u32 llh_bitmap[LLOG_BITMAP_BYTES/sizeof(__u32)];
2271 struct llog_rec_tail llh_tail;
2272 } __attribute__((packed));
2274 #define LLOG_BITMAP_SIZE(llh) ((llh->llh_hdr.lrh_len - \
2275 llh->llh_bitmap_offset - \
2276 sizeof(llh->llh_tail)) * 8)
2278 /** log cookies are used to reference a specific log file and a record therein */
2279 struct llog_cookie {
2280 struct llog_logid lgc_lgl;
2284 } __attribute__((packed));
2286 /** llog protocol */
2287 enum llogd_rpc_ops {
2288 LLOG_ORIGIN_HANDLE_CREATE = 501,
2289 LLOG_ORIGIN_HANDLE_NEXT_BLOCK = 502,
2290 LLOG_ORIGIN_HANDLE_READ_HEADER = 503,
2291 LLOG_ORIGIN_HANDLE_WRITE_REC = 504,
2292 LLOG_ORIGIN_HANDLE_CLOSE = 505,
2293 LLOG_ORIGIN_CONNECT = 506,
2294 LLOG_CATINFO = 507, /* for lfs catinfo */
2295 LLOG_ORIGIN_HANDLE_PREV_BLOCK = 508,
2296 LLOG_ORIGIN_HANDLE_DESTROY = 509, /* for destroy llog object*/
2298 LLOG_FIRST_OPC = LLOG_ORIGIN_HANDLE_CREATE
2302 struct llog_logid lgd_logid;
2304 __u32 lgd_llh_flags;
2306 __u32 lgd_saved_index;
2308 __u64 lgd_cur_offset;
2309 } __attribute__((packed));
2311 struct llogd_conn_body {
2312 struct llog_gen lgdc_gen;
2313 struct llog_logid lgdc_logid;
2314 __u32 lgdc_ctxt_idx;
2315 } __attribute__((packed));
2317 /* Note: 64-bit types are 64-bit aligned in structure */
2319 obd_valid o_valid; /* hot fields in this obdo */
2323 obd_size o_size; /* o_size-o_blocks == ost_lvb */
2327 obd_blocks o_blocks; /* brw: cli sent cached bytes */
2330 /* 32-bit fields start here: keep an even number of them via padding */
2331 obd_blksize o_blksize; /* optimal IO blocksize */
2332 obd_mode o_mode; /* brw: cli sent cache remain */
2336 obd_count o_nlink; /* brw: checksum */
2337 obd_count o_generation;
2338 obd_count o_misc; /* brw: o_dropped */
2339 __u64 o_ioepoch; /* epoch in ost writes */
2340 __u32 o_stripe_idx; /* holds stripe idx */
2342 struct lustre_handle o_handle; /* brw: lock handle to prolong locks */
2343 struct llog_cookie o_lcookie; /* destroy: unlink cookie from MDS */
2352 #define o_dirty o_blocks
2353 #define o_undirty o_mode
2354 #define o_dropped o_misc
2355 #define o_cksum o_nlink
2357 static inline void lustre_set_wire_obdo(struct obdo *wobdo, struct obdo *lobdo)
2359 memcpy(wobdo, lobdo, sizeof(*lobdo));
2360 wobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
2363 static inline void lustre_get_wire_obdo(struct obdo *lobdo, struct obdo *wobdo)
2365 obd_flag local_flags = lobdo->o_flags & OBD_FL_LOCAL_MASK;
2367 LASSERT(!(wobdo->o_flags & OBD_FL_LOCAL_MASK));
2369 memcpy(lobdo, wobdo, sizeof(*lobdo));
2370 lobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
2371 lobdo->o_flags |= local_flags;
2374 extern void lustre_swab_obdo (struct obdo *o);
2376 /* request structure for OST's */
2381 /* Key for FIEMAP to be used in get_info calls */
2382 struct ll_fiemap_info_key {
2385 struct ll_user_fiemap fiemap;
2388 extern void lustre_swab_ost_body (struct ost_body *b);
2389 extern void lustre_swab_ost_last_id(obd_id *id);
2390 extern void lustre_swab_fiemap(struct ll_user_fiemap *fiemap);
2392 extern void lustre_swab_lov_user_md_v1(struct lov_user_md_v1 *lum);
2393 extern void lustre_swab_lov_user_md_v3(struct lov_user_md_v3 *lum);
2394 extern void lustre_swab_lov_user_md_objects(struct lov_user_ost_data *lod,
2396 extern void lustre_swab_lov_mds_md(struct lov_mds_md *lmm);
2399 extern void lustre_swab_llogd_body (struct llogd_body *d);
2400 extern void lustre_swab_llog_hdr (struct llog_log_hdr *h);
2401 extern void lustre_swab_llogd_conn_body (struct llogd_conn_body *d);
2402 extern void lustre_swab_llog_rec(struct llog_rec_hdr *rec,
2403 struct llog_rec_tail *tail);
2406 extern void lustre_swab_lustre_cfg(struct lustre_cfg *lcfg);
2408 /* Functions for dumping PTLRPC fields */
2409 void dump_rniobuf(struct niobuf_remote *rnb);
2410 void dump_ioo(struct obd_ioobj *nb);
2411 void dump_obdo(struct obdo *oa);
2412 void dump_ost_body(struct ost_body *ob);
2413 void dump_rcs(__u32 *rc);
2415 /* this will be used when OBD_CONNECT_CHANGE_QS is set */
2418 * ID appiles to (uid, gid)
2422 * LQUOTA_FLAGS_* affect the responding bits
2426 * acquire/release count (bytes for block quota)
2430 * when a master returns the reply to a slave, it will
2431 * contain the current corresponding qunit size
2437 #define QDATA_IS_GRP(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_GRP)
2438 #define QDATA_IS_BLK(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_BLK)
2439 #define QDATA_IS_ADJBLK(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_ADJBLK)
2440 #define QDATA_IS_ADJINO(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_ADJINO)
2441 #define QDATA_IS_CHANGE_QS(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_CHG_QS)
2443 #define QDATA_SET_GRP(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_GRP)
2444 #define QDATA_SET_BLK(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_BLK)
2445 #define QDATA_SET_ADJBLK(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_ADJBLK)
2446 #define QDATA_SET_ADJINO(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_ADJINO)
2447 #define QDATA_SET_CHANGE_QS(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_CHG_QS)
2449 #define QDATA_CLR_GRP(qdata) ((qdata)->qd_flags &= ~LQUOTA_FLAGS_GRP)
2450 #define QDATA_CLR_CHANGE_QS(qdata) ((qdata)->qd_flags &= ~LQUOTA_FLAGS_CHG_QS)
2452 extern void lustre_swab_qdata(struct qunit_data *d);
2453 extern struct qunit_data *quota_get_qdata(void *req, int is_req, int is_exp);
2454 extern int quota_copy_qdata(void *request, struct qunit_data *qdata,
2455 int is_req, int is_exp);
2462 #define QUOTA_FIRST_OPC QUOTA_DQACQ
2464 #define QUOTA_REQUEST 1
2465 #define QUOTA_REPLY 0
2466 #define QUOTA_EXPORT 1
2467 #define QUOTA_IMPORT 0
2469 /* quota check function */
2470 #define QUOTA_RET_OK 0 /**< return successfully */
2471 #define QUOTA_RET_NOQUOTA 1 /**< not support quota */
2472 #define QUOTA_RET_NOLIMIT 2 /**< quota limit isn't set */
2473 #define QUOTA_RET_ACQUOTA 4 /**< need to acquire extra quota */
2476 /* security opcodes */
2479 SEC_CTX_INIT_CONT = 802,
2482 SEC_FIRST_OPC = SEC_CTX_INIT
2486 * capa related definitions
2488 #define CAPA_HMAC_MAX_LEN 64
2489 #define CAPA_HMAC_KEY_MAX_LEN 56
2491 /* NB take care when changing the sequence of elements this struct,
2492 * because the offset info is used in find_capa() */
2493 struct lustre_capa {
2494 struct lu_fid lc_fid; /** fid */
2495 __u64 lc_opc; /** operations allowed */
2496 __u64 lc_uid; /** file owner */
2497 __u64 lc_gid; /** file group */
2498 __u32 lc_flags; /** HMAC algorithm & flags */
2499 __u32 lc_keyid; /** key# used for the capability */
2500 __u32 lc_timeout; /** capa timeout value (sec) */
2501 __u32 lc_expiry; /** expiry time (sec) */
2502 __u8 lc_hmac[CAPA_HMAC_MAX_LEN]; /** HMAC */
2503 } __attribute__((packed));
2505 extern void lustre_swab_lustre_capa(struct lustre_capa *c);
2507 /** lustre_capa::lc_opc */
2509 CAPA_OPC_BODY_WRITE = 1<<0, /**< write object data */
2510 CAPA_OPC_BODY_READ = 1<<1, /**< read object data */
2511 CAPA_OPC_INDEX_LOOKUP = 1<<2, /**< lookup object fid */
2512 CAPA_OPC_INDEX_INSERT = 1<<3, /**< insert object fid */
2513 CAPA_OPC_INDEX_DELETE = 1<<4, /**< delete object fid */
2514 CAPA_OPC_OSS_WRITE = 1<<5, /**< write oss object data */
2515 CAPA_OPC_OSS_READ = 1<<6, /**< read oss object data */
2516 CAPA_OPC_OSS_TRUNC = 1<<7, /**< truncate oss object */
2517 CAPA_OPC_OSS_DESTROY = 1<<8, /**< destroy oss object */
2518 CAPA_OPC_META_WRITE = 1<<9, /**< write object meta data */
2519 CAPA_OPC_META_READ = 1<<10, /**< read object meta data */
2522 #define CAPA_OPC_OSS_RW (CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE)
2523 #define CAPA_OPC_MDS_ONLY \
2524 (CAPA_OPC_BODY_WRITE | CAPA_OPC_BODY_READ | CAPA_OPC_INDEX_LOOKUP | \
2525 CAPA_OPC_INDEX_INSERT | CAPA_OPC_INDEX_DELETE)
2526 #define CAPA_OPC_OSS_ONLY \
2527 (CAPA_OPC_OSS_WRITE | CAPA_OPC_OSS_READ | CAPA_OPC_OSS_TRUNC | \
2528 CAPA_OPC_OSS_DESTROY)
2529 #define CAPA_OPC_MDS_DEFAULT ~CAPA_OPC_OSS_ONLY
2530 #define CAPA_OPC_OSS_DEFAULT ~(CAPA_OPC_MDS_ONLY | CAPA_OPC_OSS_ONLY)
2532 /* MDS capability covers object capability for operations of body r/w
2533 * (dir readpage/sendpage), index lookup/insert/delete and meta data r/w,
2534 * while OSS capability only covers object capability for operations of
2535 * oss data(file content) r/w/truncate.
2537 static inline int capa_for_mds(struct lustre_capa *c)
2539 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) != 0;
2542 static inline int capa_for_oss(struct lustre_capa *c)
2544 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) == 0;
2547 /* lustre_capa::lc_hmac_alg */
2549 CAPA_HMAC_ALG_SHA1 = 1, /**< sha1 algorithm */
2553 #define CAPA_FL_MASK 0x00ffffff
2554 #define CAPA_HMAC_ALG_MASK 0xff000000
2556 struct lustre_capa_key {
2557 __u64 lk_mdsid; /**< mds# */
2558 __u32 lk_keyid; /**< key# */
2560 __u8 lk_key[CAPA_HMAC_KEY_MAX_LEN]; /**< key */
2561 } __attribute__((packed));
2563 extern void lustre_swab_lustre_capa_key(struct lustre_capa_key *k);
2565 /** The link ea holds 1 \a link_ea_entry for each hardlink */
2566 #define LINK_EA_MAGIC 0x11EAF1DFUL
2567 struct link_ea_header {
2570 __u64 leh_len; /* total size */
2576 /** Hardlink data is name and parent fid.
2577 * Stored in this crazy struct for maximum packing and endian-neutrality
2579 struct link_ea_entry {
2580 /** __u16 stored big-endian, unaligned */
2581 unsigned char lee_reclen[2];
2582 unsigned char lee_parent_fid[sizeof(struct lu_fid)];
2584 }__attribute__((packed));
2586 /** fid2path request/reply structure */
2587 struct getinfo_fid2path {
2588 struct lu_fid gf_fid;
2593 } __attribute__((packed));
2595 void lustre_swab_fid2path (struct getinfo_fid2path *gf);
2597 extern void lustre_swab_kuch(struct kuc_hdr *);