4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
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 #if !defined(LASSERT) && !defined(LPU64)
95 #include <libcfs/libcfs.h> /* for LASSERT, LPUX64, etc */
98 /* Defn's shared with user-space. */
99 #include <lustre/lustre_user.h>
104 /* FOO_REQUEST_PORTAL is for incoming requests on the FOO
105 * FOO_REPLY_PORTAL is for incoming replies on the FOO
106 * FOO_BULK_PORTAL is for incoming bulk on the FOO
109 #define CONNMGR_REQUEST_PORTAL 1
110 #define CONNMGR_REPLY_PORTAL 2
111 //#define OSC_REQUEST_PORTAL 3
112 #define OSC_REPLY_PORTAL 4
113 //#define OSC_BULK_PORTAL 5
114 #define OST_IO_PORTAL 6
115 #define OST_CREATE_PORTAL 7
116 #define OST_BULK_PORTAL 8
117 //#define MDC_REQUEST_PORTAL 9
118 #define MDC_REPLY_PORTAL 10
119 //#define MDC_BULK_PORTAL 11
120 #define MDS_REQUEST_PORTAL 12
121 //#define MDS_REPLY_PORTAL 13
122 #define MDS_BULK_PORTAL 14
123 #define LDLM_CB_REQUEST_PORTAL 15
124 #define LDLM_CB_REPLY_PORTAL 16
125 #define LDLM_CANCEL_REQUEST_PORTAL 17
126 #define LDLM_CANCEL_REPLY_PORTAL 18
127 //#define PTLBD_REQUEST_PORTAL 19
128 //#define PTLBD_REPLY_PORTAL 20
129 //#define PTLBD_BULK_PORTAL 21
130 #define MDS_SETATTR_PORTAL 22
131 #define MDS_READPAGE_PORTAL 23
132 #define MDS_MDS_PORTAL 24
134 #define MGC_REPLY_PORTAL 25
135 #define MGS_REQUEST_PORTAL 26
136 #define MGS_REPLY_PORTAL 27
137 #define OST_REQUEST_PORTAL 28
138 #define FLD_REQUEST_PORTAL 29
139 #define SEQ_METADATA_PORTAL 30
140 #define SEQ_DATA_PORTAL 31
141 #define SEQ_CONTROLLER_PORTAL 32
142 #define MGS_BULK_PORTAL 33
144 /* Portal 63 is reserved for the Cray Inc DVS - nic@cray.com, roe@cray.com, n8851@cray.com */
147 #define PTL_RPC_MSG_REQUEST 4711
148 #define PTL_RPC_MSG_ERR 4712
149 #define PTL_RPC_MSG_REPLY 4713
151 /* DON'T use swabbed values of MAGIC as magic! */
152 #define LUSTRE_MSG_MAGIC_V1 0x0BD00BD0
153 #define LUSTRE_MSG_MAGIC_V2 0x0BD00BD3
155 #define LUSTRE_MSG_MAGIC_V1_SWABBED 0xD00BD00B
156 #define LUSTRE_MSG_MAGIC_V2_SWABBED 0xD30BD00B
158 #define LUSTRE_MSG_MAGIC LUSTRE_MSG_MAGIC_V2
160 #define PTLRPC_MSG_VERSION 0x00000003
161 #define LUSTRE_VERSION_MASK 0xffff0000
162 #define LUSTRE_OBD_VERSION 0x00010000
163 #define LUSTRE_MDS_VERSION 0x00020000
164 #define LUSTRE_OST_VERSION 0x00030000
165 #define LUSTRE_DLM_VERSION 0x00040000
166 #define LUSTRE_LOG_VERSION 0x00050000
167 #define LUSTRE_MGS_VERSION 0x00060000
169 typedef __u32 mdsno_t;
170 typedef __u64 seqno_t;
171 typedef __u64 obd_id;
172 typedef __u64 obd_seq;
173 typedef __s64 obd_time;
174 typedef __u64 obd_size;
175 typedef __u64 obd_off;
176 typedef __u64 obd_blocks;
177 typedef __u64 obd_valid;
178 typedef __u32 obd_blksize;
179 typedef __u32 obd_mode;
180 typedef __u32 obd_uid;
181 typedef __u32 obd_gid;
182 typedef __u32 obd_flag;
183 typedef __u32 obd_count;
186 * Describes a range of sequence, lsr_start is included but lsr_end is
188 * Same structure is used in fld module where lsr_index field holds mdt id
192 #define LU_SEQ_RANGE_MDT 0x0
193 #define LU_SEQ_RANGE_OST 0x1
195 struct lu_seq_range {
203 * returns width of given range \a r
206 static inline __u64 range_space(const struct lu_seq_range *range)
208 return range->lsr_end - range->lsr_start;
212 * initialize range to zero
215 static inline void range_init(struct lu_seq_range *range)
217 range->lsr_start = range->lsr_end = range->lsr_index = 0;
221 * check if given seq id \a s is within given range \a r
224 static inline int range_within(const struct lu_seq_range *range,
227 return s >= range->lsr_start && s < range->lsr_end;
230 static inline int range_is_sane(const struct lu_seq_range *range)
232 return (range->lsr_end >= range->lsr_start);
235 static inline int range_is_zero(const struct lu_seq_range *range)
237 return (range->lsr_start == 0 && range->lsr_end == 0);
240 static inline int range_is_exhausted(const struct lu_seq_range *range)
243 return range_space(range) == 0;
246 /* return 0 if two range have the same location */
247 static inline int range_compare_loc(const struct lu_seq_range *r1,
248 const struct lu_seq_range *r2)
250 return r1->lsr_index != r2->lsr_index ||
251 r1->lsr_flags != r2->lsr_flags;
254 #define DRANGE "[%#16.16"LPF64"x-%#16.16"LPF64"x):%x:%s"
256 #define PRANGE(range) \
257 (range)->lsr_start, \
259 (range)->lsr_index, \
260 (range)->lsr_flags == LU_SEQ_RANGE_MDT ? "mdt" : "ost"
263 /** \defgroup lu_fid lu_fid
267 * Flags for lustre_mdt_attrs::lma_compat and lustre_mdt_attrs::lma_incompat.
270 LMAC_HSM = 0x00000001,
271 LMAC_SOM = 0x00000002,
275 * Masks for all features that should be supported by a Lustre version to
276 * access a specific file.
277 * This information is stored in lustre_mdt_attrs::lma_incompat.
279 * NOTE: No incompat feature should be added before bug #17670 is landed.
281 #define LMA_INCOMPAT_SUPP 0x0
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 {
289 * Bitfield for supported data in this structure. From enum lma_compat.
290 * lma_self_fid and lma_flags are always available.
294 * Per-file incompat feature list. Lustre version should support all
295 * flags set in this field. The supported feature mask is available in
299 /** FID of this inode */
300 struct lu_fid lma_self_fid;
301 /** mdt/ost type, others */
303 /* IO Epoch SOM attributes belongs to */
305 /** total file size in objects */
307 /** total fs blocks in objects */
308 __u64 lma_som_blocks;
309 /** mds mount id the size is valid for */
310 __u64 lma_som_mountid;
314 * Fill \a lma with its first content.
315 * Only fid is stored.
317 static inline void lustre_lma_init(struct lustre_mdt_attrs *lma,
318 const struct lu_fid *fid)
321 lma->lma_incompat = 0;
322 memcpy(&lma->lma_self_fid, fid, sizeof(*fid));
324 lma->lma_ioepoch = 0;
325 lma->lma_som_size = 0;
326 lma->lma_som_blocks = 0;
327 lma->lma_som_mountid = 0;
329 /* If a field is added in struct lustre_mdt_attrs, zero it explicitly
330 * and change the test below. */
331 LASSERT(sizeof(*lma) ==
332 (offsetof(struct lustre_mdt_attrs, lma_som_mountid) +
333 sizeof(lma->lma_som_mountid)));
336 extern void lustre_swab_lu_fid(struct lu_fid *fid);
339 * Swab, if needed, lustre_mdt_attr struct to on-disk format.
340 * Otherwise, do not touch it.
342 static inline void lustre_lma_swab(struct lustre_mdt_attrs *lma)
344 /* Use LUSTRE_MSG_MAGIC to detect local endianess. */
345 if (LUSTRE_MSG_MAGIC != cpu_to_le32(LUSTRE_MSG_MAGIC)) {
346 __swab32s(&lma->lma_compat);
347 __swab32s(&lma->lma_incompat);
348 lustre_swab_lu_fid(&lma->lma_self_fid);
349 __swab64s(&lma->lma_flags);
350 __swab64s(&lma->lma_ioepoch);
351 __swab64s(&lma->lma_som_size);
352 __swab64s(&lma->lma_som_blocks);
353 __swab64s(&lma->lma_som_mountid);
357 /* This is the maximum number of MDTs allowed in CMD testing until such
358 * a time that FID-on-OST is implemented. This is due to the limitations
359 * of packing non-0-MDT numbers into the FID SEQ namespace. Once FID-on-OST
360 * is implemented this limit will be virtually unlimited. */
361 #define MAX_MDT_COUNT 8
368 /** initial fid id value */
369 LUSTRE_FID_INIT_OID = 1UL
372 /** returns fid object sequence */
373 static inline __u64 fid_seq(const struct lu_fid *fid)
378 /** returns fid object id */
379 static inline __u32 fid_oid(const struct lu_fid *fid)
384 /** returns fid object version */
385 static inline __u32 fid_ver(const struct lu_fid *fid)
390 static inline void fid_zero(struct lu_fid *fid)
392 memset(fid, 0, sizeof(*fid));
395 static inline obd_id fid_ver_oid(const struct lu_fid *fid)
397 return ((__u64)fid_ver(fid) << 32 | fid_oid(fid));
401 * Note that reserved SEQ numbers below 12 will conflict with ldiskfs
402 * inodes in the IGIF namespace, so these reserved SEQ numbers can be
403 * used for other purposes and not risk collisions with existing inodes.
405 * Different FID Format
406 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs#NEW.0
409 FID_SEQ_OST_MDT0 = 0,
412 FID_SEQ_OST_MDT1 = 3,
413 FID_SEQ_OST_MAX = 9, /* Max MDT count before OST_on_FID */
416 FID_SEQ_IGIF_MAX = 0x0ffffffffULL,
417 FID_SEQ_IDIF = 0x100000000ULL,
418 FID_SEQ_IDIF_MAX = 0x1ffffffffULL,
419 /* Normal FID sequence starts from this value, i.e. 1<<33 */
420 FID_SEQ_START = 0x200000000ULL,
421 FID_SEQ_LOCAL_FILE = 0x200000001ULL,
422 FID_SEQ_DOT_LUSTRE = 0x200000002ULL,
423 /* XXX 0x200000003ULL is reserved for FID_SEQ_LLOG_OBJ */
424 FID_SEQ_SPECIAL = 0x200000004ULL,
425 FID_SEQ_QUOTA = 0x200000005ULL,
426 FID_SEQ_QUOTA_GLB = 0x200000006ULL,
427 FID_SEQ_NORMAL = 0x200000400ULL,
428 FID_SEQ_LOV_DEFAULT= 0xffffffffffffffffULL
431 #define OBIF_OID_MAX_BITS 32
432 #define OBIF_MAX_OID (1ULL << OBIF_OID_MAX_BITS)
433 #define OBIF_OID_MASK ((1ULL << OBIF_OID_MAX_BITS) - 1)
434 #define IDIF_OID_MAX_BITS 48
435 #define IDIF_MAX_OID (1ULL << IDIF_OID_MAX_BITS)
436 #define IDIF_OID_MASK ((1ULL << IDIF_OID_MAX_BITS) - 1)
438 /** OID for FID_SEQ_SPECIAL */
440 /* Big Filesystem Lock to serialize rename operations */
441 FID_OID_SPECIAL_BFL = 1UL,
444 /** OID for FID_SEQ_DOT_LUSTRE */
445 enum dot_lustre_oid {
446 FID_OID_DOT_LUSTRE = 1UL,
447 FID_OID_DOT_LUSTRE_OBF = 2UL,
450 static inline int fid_seq_is_mdt0(obd_seq seq)
452 return (seq == FID_SEQ_OST_MDT0);
455 static inline int fid_seq_is_cmd(const __u64 seq)
457 return (seq >= FID_SEQ_OST_MDT1 && seq <= FID_SEQ_OST_MAX);
460 static inline int fid_seq_is_mdt(const __u64 seq)
462 return seq == FID_SEQ_OST_MDT0 ||
463 (seq >= FID_SEQ_OST_MDT1 && seq <= FID_SEQ_OST_MAX);
466 static inline int fid_seq_is_rsvd(const __u64 seq)
468 return (seq > FID_SEQ_OST_MDT0 && seq <= FID_SEQ_RSVD);
471 static inline int fid_is_mdt0(const struct lu_fid *fid)
473 return fid_seq_is_mdt0(fid_seq(fid));
477 * Check if a fid is igif or not.
478 * \param fid the fid to be tested.
479 * \return true if the fid is a igif; otherwise false.
481 static inline int fid_seq_is_igif(const __u64 seq)
483 return seq >= FID_SEQ_IGIF && seq <= FID_SEQ_IGIF_MAX;
486 static inline int fid_is_igif(const struct lu_fid *fid)
488 return fid_seq_is_igif(fid_seq(fid));
492 * Check if a fid is idif or not.
493 * \param fid the fid to be tested.
494 * \return true if the fid is a idif; otherwise false.
496 static inline int fid_seq_is_idif(const __u64 seq)
498 return seq >= FID_SEQ_IDIF && seq <= FID_SEQ_IDIF_MAX;
501 static inline int fid_is_idif(const struct lu_fid *fid)
503 return fid_seq_is_idif(fid_seq(fid));
511 static inline int fid_seq_is_norm(const __u64 seq)
513 return (seq >= FID_SEQ_NORMAL);
516 static inline int fid_is_norm(const struct lu_fid *fid)
518 return fid_seq_is_norm(fid_seq(fid));
521 /* convert an OST objid into an IDIF FID SEQ number */
522 static inline obd_seq fid_idif_seq(obd_id id, __u32 ost_idx)
524 return FID_SEQ_IDIF | (ost_idx << 16) | ((id >> 32) & 0xffff);
527 /* convert a packed IDIF FID into an OST objid */
528 static inline obd_id fid_idif_id(obd_seq seq, __u32 oid, __u32 ver)
530 return ((__u64)ver << 48) | ((seq & 0xffff) << 32) | oid;
533 /* unpack an ostid (id/seq) from a wire/disk structure into an IDIF FID */
534 static inline void ostid_idif_unpack(struct ost_id *ostid,
535 struct lu_fid *fid, __u32 ost_idx)
537 fid->f_seq = fid_idif_seq(ostid->oi_id, ost_idx);
538 fid->f_oid = ostid->oi_id; /* truncate to 32 bits by assignment */
539 fid->f_ver = ostid->oi_id >> 48; /* in theory, not currently used */
542 /* unpack an ostid (id/seq) from a wire/disk structure into a non-IDIF FID */
543 static inline void ostid_fid_unpack(struct ost_id *ostid, struct lu_fid *fid)
545 fid->f_seq = ostid->oi_seq;
546 fid->f_oid = ostid->oi_id; /* truncate to 32 bits by assignment */
547 fid->f_ver = ostid->oi_id >> 32; /* in theory, not currently used */
550 /* Unpack an OST object id/seq (group) into a FID. This is needed for
551 * converting all obdo, lmm, lsm, etc. 64-bit id/seq pairs into proper
552 * FIDs. Note that if an id/seq is already in FID/IDIF format it will
553 * be passed through unchanged. Only legacy OST objects in "group 0"
554 * will be mapped into the IDIF namespace so that they can fit into the
555 * struct lu_fid fields without loss. For reference see:
556 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs
558 static inline int fid_ostid_unpack(struct lu_fid *fid, struct ost_id *ostid,
561 if (ost_idx > 0xffff) {
562 CERROR("bad ost_idx, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
563 ostid->oi_seq, ostid->oi_id, ost_idx);
567 if (fid_seq_is_mdt0(ostid->oi_seq)) {
568 /* This is a "legacy" (old 1.x/2.early) OST object in "group 0"
569 * that we map into the IDIF namespace. It allows up to 2^48
570 * objects per OST, as this is the object namespace that has
571 * been in production for years. This can handle create rates
572 * of 1M objects/s/OST for 9 years, or combinations thereof. */
573 if (ostid->oi_id >= IDIF_MAX_OID) {
574 CERROR("bad MDT0 id, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
575 ostid->oi_seq, ostid->oi_id, ost_idx);
578 ostid_idif_unpack(ostid, fid, ost_idx);
580 } else if (fid_seq_is_rsvd(ostid->oi_seq)) {
581 /* These are legacy OST objects for LLOG/ECHO and CMD testing.
582 * We only support 2^32 objects in these groups, and cannot
583 * uniquely identify them in the system (i.e. they are the
584 * duplicated on all OSTs), but this is not strictly required
585 * for the old object protocol, which has a separate ost_idx. */
586 if (ostid->oi_id >= 0xffffffffULL) {
587 CERROR("bad RSVD id, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
588 ostid->oi_seq, ostid->oi_id, ost_idx);
591 ostid_fid_unpack(ostid, fid);
593 } else if (unlikely(fid_seq_is_igif(ostid->oi_seq))) {
594 /* This is an MDT inode number, which should never collide with
595 * proper OST object IDs, and is probably a broken filesystem */
596 CERROR("bad IGIF, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
597 ostid->oi_seq, ostid->oi_id, ost_idx);
600 } else /* if (fid_seq_is_idif(seq) || fid_seq_is_norm(seq)) */ {
601 /* This is either an IDIF object, which identifies objects across
602 * all OSTs, or a regular FID. The IDIF namespace maps legacy
603 * OST objects into the FID namespace. In both cases, we just
604 * pass the FID through, no conversion needed. */
605 ostid_fid_unpack(ostid, fid);
611 /* pack an IDIF FID into an ostid (id/seq) for the wire/disk */
612 static inline void ostid_idif_pack(const struct lu_fid *fid,
613 struct ost_id *ostid)
615 ostid->oi_seq = FID_SEQ_OST_MDT0;
616 ostid->oi_id = fid_idif_id(fid->f_seq, fid->f_oid, fid->f_ver);
619 /* pack a non-IDIF FID into an ostid (id/seq) for the wire/disk */
620 static inline void ostid_fid_pack(const struct lu_fid *fid,
621 struct ost_id *ostid)
623 ostid->oi_seq = fid_seq(fid);
624 ostid->oi_id = fid_ver_oid(fid);
627 /* pack any OST FID into an ostid (id/seq) for the wire/disk */
628 static inline int fid_ostid_pack(const struct lu_fid *fid,
629 struct ost_id *ostid)
631 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
632 CERROR("bad IGIF, "DFID"\n", PFID(fid));
636 if (fid_is_idif(fid))
637 ostid_idif_pack(fid, ostid);
639 ostid_fid_pack(fid, ostid);
644 /* extract OST sequence (group) from a wire ost_id (id/seq) pair */
645 static inline obd_seq ostid_seq(struct ost_id *ostid)
647 if (unlikely(fid_seq_is_igif(ostid->oi_seq)))
648 CWARN("bad IGIF, oi_seq: "LPU64" oi_id: "LPX64"\n",
649 ostid->oi_seq, ostid->oi_id);
651 if (unlikely(fid_seq_is_idif(ostid->oi_seq)))
652 return FID_SEQ_OST_MDT0;
654 return ostid->oi_seq;
657 /* extract OST objid from a wire ost_id (id/seq) pair */
658 static inline obd_id ostid_id(struct ost_id *ostid)
660 if (ostid->oi_seq == FID_SEQ_OST_MDT0)
661 return ostid->oi_id & IDIF_OID_MASK;
663 if (fid_seq_is_rsvd(ostid->oi_seq))
664 return ostid->oi_id & OBIF_OID_MASK;
666 if (fid_seq_is_idif(ostid->oi_seq))
667 return fid_idif_id(ostid->oi_seq, ostid->oi_id, 0);
673 * Get inode number from a igif.
674 * \param fid a igif to get inode number from.
675 * \return inode number for the igif.
677 static inline ino_t lu_igif_ino(const struct lu_fid *fid)
683 * Build igif from the inode number/generation.
685 #define LU_IGIF_BUILD(fid, ino, gen) \
691 static inline void lu_igif_build(struct lu_fid *fid, __u32 ino, __u32 gen)
693 LU_IGIF_BUILD(fid, ino, gen);
694 LASSERT(fid_is_igif(fid));
698 * Get inode generation from a igif.
699 * \param fid a igif to get inode generation from.
700 * \return inode generation for the igif.
702 static inline __u32 lu_igif_gen(const struct lu_fid *fid)
708 * Fids are transmitted across network (in the sender byte-ordering),
709 * and stored on disk in big-endian order.
711 static inline void fid_cpu_to_le(struct lu_fid *dst, const struct lu_fid *src)
713 /* check that all fields are converted */
714 CLASSERT(sizeof *src ==
715 sizeof fid_seq(src) +
716 sizeof fid_oid(src) + sizeof fid_ver(src));
717 LASSERTF(fid_is_igif(src) || fid_ver(src) == 0, DFID"\n", PFID(src));
718 dst->f_seq = cpu_to_le64(fid_seq(src));
719 dst->f_oid = cpu_to_le32(fid_oid(src));
720 dst->f_ver = cpu_to_le32(fid_ver(src));
723 static inline void fid_le_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
725 /* check that all fields are converted */
726 CLASSERT(sizeof *src ==
727 sizeof fid_seq(src) +
728 sizeof fid_oid(src) + sizeof fid_ver(src));
729 dst->f_seq = le64_to_cpu(fid_seq(src));
730 dst->f_oid = le32_to_cpu(fid_oid(src));
731 dst->f_ver = le32_to_cpu(fid_ver(src));
732 LASSERTF(fid_is_igif(dst) || fid_ver(dst) == 0, DFID"\n", PFID(dst));
735 static inline void fid_cpu_to_be(struct lu_fid *dst, const struct lu_fid *src)
737 /* check that all fields are converted */
738 CLASSERT(sizeof *src ==
739 sizeof fid_seq(src) +
740 sizeof fid_oid(src) + sizeof fid_ver(src));
741 LASSERTF(fid_is_igif(src) || fid_ver(src) == 0, DFID"\n", PFID(src));
742 dst->f_seq = cpu_to_be64(fid_seq(src));
743 dst->f_oid = cpu_to_be32(fid_oid(src));
744 dst->f_ver = cpu_to_be32(fid_ver(src));
747 static inline void fid_be_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
749 /* check that all fields are converted */
750 CLASSERT(sizeof *src ==
751 sizeof fid_seq(src) +
752 sizeof fid_oid(src) + sizeof fid_ver(src));
753 dst->f_seq = be64_to_cpu(fid_seq(src));
754 dst->f_oid = be32_to_cpu(fid_oid(src));
755 dst->f_ver = be32_to_cpu(fid_ver(src));
756 LASSERTF(fid_is_igif(dst) || fid_ver(dst) == 0, DFID"\n", PFID(dst));
759 static inline int fid_is_sane(const struct lu_fid *fid)
763 ((fid_seq(fid) >= FID_SEQ_START && fid_oid(fid) != 0
764 && fid_ver(fid) == 0) ||
765 fid_is_igif(fid) || fid_seq_is_rsvd(fid_seq(fid)));
768 static inline int fid_is_zero(const struct lu_fid *fid)
770 return fid_seq(fid) == 0 && fid_oid(fid) == 0;
773 extern void lustre_swab_lu_fid(struct lu_fid *fid);
774 extern void lustre_swab_lu_seq_range(struct lu_seq_range *range);
776 static inline int lu_fid_eq(const struct lu_fid *f0,
777 const struct lu_fid *f1)
779 /* Check that there is no alignment padding. */
780 CLASSERT(sizeof *f0 ==
781 sizeof f0->f_seq + sizeof f0->f_oid + sizeof f0->f_ver);
782 LASSERTF((fid_is_igif(f0) || fid_is_idif(f0)) ||
783 fid_ver(f0) == 0, DFID, PFID(f0));
784 LASSERTF((fid_is_igif(f1) || fid_is_idif(f1)) ||
785 fid_ver(f1) == 0, DFID, PFID(f1));
786 return memcmp(f0, f1, sizeof *f0) == 0;
789 #define __diff_normalize(val0, val1) \
791 typeof(val0) __val0 = (val0); \
792 typeof(val1) __val1 = (val1); \
794 (__val0 == __val1 ? 0 : __val0 > __val1 ? +1 : -1); \
797 static inline int lu_fid_cmp(const struct lu_fid *f0,
798 const struct lu_fid *f1)
801 __diff_normalize(fid_seq(f0), fid_seq(f1)) ?:
802 __diff_normalize(fid_oid(f0), fid_oid(f1)) ?:
803 __diff_normalize(fid_ver(f0), fid_ver(f1));
808 /** \defgroup lu_dir lu_dir
812 * Enumeration of possible directory entry attributes.
814 * Attributes follow directory entry header in the order they appear in this
817 enum lu_dirent_attrs {
820 LUDA_64BITHASH = 0x0004,
824 * Layout of readdir pages, as transmitted on wire.
827 /** valid if LUDA_FID is set. */
828 struct lu_fid lde_fid;
829 /** a unique entry identifier: a hash or an offset. */
831 /** total record length, including all attributes. */
835 /** optional variable size attributes following this entry.
836 * taken from enum lu_dirent_attrs.
839 /** name is followed by the attributes indicated in ->ldp_attrs, in
840 * their natural order. After the last attribute, padding bytes are
841 * added to make ->lde_reclen a multiple of 8.
847 * Definitions of optional directory entry attributes formats.
849 * Individual attributes do not have their length encoded in a generic way. It
850 * is assumed that consumer of an attribute knows its format. This means that
851 * it is impossible to skip over an unknown attribute, except by skipping over all
852 * remaining attributes (by using ->lde_reclen), which is not too
853 * constraining, because new server versions will append new attributes at
854 * the end of an entry.
858 * Fid directory attribute: a fid of an object referenced by the entry. This
859 * will be almost always requested by the client and supplied by the server.
861 * Aligned to 8 bytes.
863 /* To have compatibility with 1.8, lets have fid in lu_dirent struct. */
868 * Aligned to 2 bytes.
875 __u64 ldp_hash_start;
879 struct lu_dirent ldp_entries[0];
882 enum lu_dirpage_flags {
884 * dirpage contains no entry.
888 * last entry's lde_hash equals ldp_hash_end.
893 static inline struct lu_dirent *lu_dirent_start(struct lu_dirpage *dp)
895 if (le32_to_cpu(dp->ldp_flags) & LDF_EMPTY)
898 return dp->ldp_entries;
901 static inline struct lu_dirent *lu_dirent_next(struct lu_dirent *ent)
903 struct lu_dirent *next;
905 if (le16_to_cpu(ent->lde_reclen) != 0)
906 next = ((void *)ent) + le16_to_cpu(ent->lde_reclen);
913 static inline int lu_dirent_calc_size(int namelen, __u16 attr)
917 if (attr & LUDA_TYPE) {
918 const unsigned align = sizeof(struct luda_type) - 1;
919 size = (sizeof(struct lu_dirent) + namelen + align) & ~align;
920 size += sizeof(struct luda_type);
922 size = sizeof(struct lu_dirent) + namelen;
924 return (size + 7) & ~7;
927 static inline int lu_dirent_size(struct lu_dirent *ent)
929 if (le16_to_cpu(ent->lde_reclen) == 0) {
930 return lu_dirent_calc_size(le16_to_cpu(ent->lde_namelen),
931 le32_to_cpu(ent->lde_attrs));
933 return le16_to_cpu(ent->lde_reclen);
936 #define MDS_DIR_END_OFF 0xfffffffffffffffeULL
939 * MDS_READPAGE page size
941 * This is the directory page size packed in MDS_READPAGE RPC.
942 * It's different than CFS_PAGE_SIZE because the client needs to
943 * access the struct lu_dirpage header packed at the beginning of
944 * the "page" and without this there isn't any way to know find the
945 * lu_dirpage header is if client and server CFS_PAGE_SIZE differ.
947 #define LU_PAGE_SHIFT 12
948 #define LU_PAGE_SIZE (1UL << LU_PAGE_SHIFT)
949 #define LU_PAGE_MASK (~(LU_PAGE_SIZE - 1))
951 #define LU_PAGE_COUNT 1 << (CFS_PAGE_SHIFT - LU_PAGE_SHIFT)
955 struct lustre_handle {
958 #define DEAD_HANDLE_MAGIC 0xdeadbeefcafebabeULL
960 static inline int lustre_handle_is_used(struct lustre_handle *lh)
962 return lh->cookie != 0ull;
965 static inline int lustre_handle_equal(const struct lustre_handle *lh1,
966 const struct lustre_handle *lh2)
968 return lh1->cookie == lh2->cookie;
971 static inline void lustre_handle_copy(struct lustre_handle *tgt,
972 struct lustre_handle *src)
974 tgt->cookie = src->cookie;
977 /* flags for lm_flags */
978 #define MSGHDR_AT_SUPPORT 0x1
979 #define MSGHDR_CKSUM_INCOMPAT18 0x2
981 #define lustre_msg lustre_msg_v2
982 /* we depend on this structure to be 8-byte aligned */
983 /* this type is only endian-adjusted in lustre_unpack_msg() */
984 struct lustre_msg_v2 {
996 /* without gss, ptlrpc_body is put at the first buffer. */
997 #define PTLRPC_NUM_VERSIONS 4
998 #define JOBSTATS_JOBID_SIZE 32 /* 32 bytes string */
999 struct ptlrpc_body_v3 {
1000 struct lustre_handle pb_handle;
1007 __u64 pb_last_committed;
1012 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1013 __u32 pb_service_time; /* for rep, actual service time */
1016 /* VBR: pre-versions */
1017 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1018 /* padding for future needs */
1019 __u64 pb_padding[4];
1020 char pb_jobid[JOBSTATS_JOBID_SIZE];
1022 #define ptlrpc_body ptlrpc_body_v3
1024 struct ptlrpc_body_v2 {
1025 struct lustre_handle pb_handle;
1032 __u64 pb_last_committed;
1037 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1038 __u32 pb_service_time; /* for rep, actual service time, also used for
1039 net_latency of req */
1042 /* VBR: pre-versions */
1043 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1044 /* padding for future needs */
1045 __u64 pb_padding[4];
1048 extern void lustre_swab_ptlrpc_body(struct ptlrpc_body *pb);
1050 /* message body offset for lustre_msg_v2 */
1051 /* ptlrpc body offset in all request/reply messages */
1052 #define MSG_PTLRPC_BODY_OFF 0
1054 /* normal request/reply message record offset */
1055 #define REQ_REC_OFF 1
1056 #define REPLY_REC_OFF 1
1058 /* ldlm request message body offset */
1059 #define DLM_LOCKREQ_OFF 1 /* lockreq offset */
1060 #define DLM_REQ_REC_OFF 2 /* normal dlm request record offset */
1062 /* ldlm intent lock message body offset */
1063 #define DLM_INTENT_IT_OFF 2 /* intent lock it offset */
1064 #define DLM_INTENT_REC_OFF 3 /* intent lock record offset */
1066 /* ldlm reply message body offset */
1067 #define DLM_LOCKREPLY_OFF 1 /* lockrep offset */
1068 #define DLM_REPLY_REC_OFF 2 /* reply record offset */
1070 /** only use in req->rq_{req,rep}_swab_mask */
1071 #define MSG_PTLRPC_HEADER_OFF 31
1073 /* Flags that are operation-specific go in the top 16 bits. */
1074 #define MSG_OP_FLAG_MASK 0xffff0000
1075 #define MSG_OP_FLAG_SHIFT 16
1077 /* Flags that apply to all requests are in the bottom 16 bits */
1078 #define MSG_GEN_FLAG_MASK 0x0000ffff
1079 #define MSG_LAST_REPLAY 0x0001
1080 #define MSG_RESENT 0x0002
1081 #define MSG_REPLAY 0x0004
1082 /* #define MSG_AT_SUPPORT 0x0008
1083 * This was used in early prototypes of adaptive timeouts, and while there
1084 * shouldn't be any users of that code there also isn't a need for using this
1085 * bits. Defer usage until at least 1.10 to avoid potential conflict. */
1086 #define MSG_DELAY_REPLAY 0x0010
1087 #define MSG_VERSION_REPLAY 0x0020
1088 #define MSG_REQ_REPLAY_DONE 0x0040
1089 #define MSG_LOCK_REPLAY_DONE 0x0080
1092 * Flags for all connect opcodes (MDS_CONNECT, OST_CONNECT)
1095 #define MSG_CONNECT_RECOVERING 0x00000001
1096 #define MSG_CONNECT_RECONNECT 0x00000002
1097 #define MSG_CONNECT_REPLAYABLE 0x00000004
1098 //#define MSG_CONNECT_PEER 0x8
1099 #define MSG_CONNECT_LIBCLIENT 0x00000010
1100 #define MSG_CONNECT_INITIAL 0x00000020
1101 #define MSG_CONNECT_ASYNC 0x00000040
1102 #define MSG_CONNECT_NEXT_VER 0x00000080 /* use next version of lustre_msg */
1103 #define MSG_CONNECT_TRANSNO 0x00000100 /* report transno */
1106 #define OBD_CONNECT_RDONLY 0x1ULL /*client has read-only access*/
1107 #define OBD_CONNECT_INDEX 0x2ULL /*connect specific LOV idx */
1108 #define OBD_CONNECT_MDS 0x4ULL /*connect from MDT to OST */
1109 #define OBD_CONNECT_GRANT 0x8ULL /*OSC gets grant at connect */
1110 #define OBD_CONNECT_SRVLOCK 0x10ULL /*server takes locks for cli */
1111 #define OBD_CONNECT_VERSION 0x20ULL /*Lustre versions in ocd */
1112 #define OBD_CONNECT_REQPORTAL 0x40ULL /*Separate non-IO req portal */
1113 #define OBD_CONNECT_ACL 0x80ULL /*access control lists */
1114 #define OBD_CONNECT_XATTR 0x100ULL /*client use extended attr */
1115 #define OBD_CONNECT_CROW 0x200ULL /*MDS+OST create obj on write*/
1116 #define OBD_CONNECT_TRUNCLOCK 0x400ULL /*locks on server for punch */
1117 #define OBD_CONNECT_TRANSNO 0x800ULL /*replay sends init transno */
1118 #define OBD_CONNECT_IBITS 0x1000ULL /*support for inodebits locks*/
1119 #define OBD_CONNECT_JOIN 0x2000ULL /*files can be concatenated.
1120 *We do not support JOIN FILE
1121 *anymore, reserve this flags
1122 *just for preventing such bit
1124 #define OBD_CONNECT_ATTRFID 0x4000ULL /*Server can GetAttr By Fid*/
1125 #define OBD_CONNECT_NODEVOH 0x8000ULL /*No open hndl on specl nodes*/
1126 #define OBD_CONNECT_RMT_CLIENT 0x10000ULL /*Remote client */
1127 #define OBD_CONNECT_RMT_CLIENT_FORCE 0x20000ULL /*Remote client by force */
1128 #define OBD_CONNECT_BRW_SIZE 0x40000ULL /*Max bytes per rpc */
1129 #define OBD_CONNECT_QUOTA64 0x80000ULL /*64bit qunit_data.qd_count */
1130 #define OBD_CONNECT_MDS_CAPA 0x100000ULL /*MDS capability */
1131 #define OBD_CONNECT_OSS_CAPA 0x200000ULL /*OSS capability */
1132 #define OBD_CONNECT_CANCELSET 0x400000ULL /*Early batched cancels. */
1133 #define OBD_CONNECT_SOM 0x800000ULL /*Size on MDS */
1134 #define OBD_CONNECT_AT 0x1000000ULL /*client uses AT */
1135 #define OBD_CONNECT_LRU_RESIZE 0x2000000ULL /*LRU resize feature. */
1136 #define OBD_CONNECT_MDS_MDS 0x4000000ULL /*MDS-MDS connection */
1137 #define OBD_CONNECT_REAL 0x8000000ULL /*real connection */
1138 #define OBD_CONNECT_CHANGE_QS 0x10000000ULL /*shrink/enlarge qunit */
1139 #define OBD_CONNECT_CKSUM 0x20000000ULL /*support several cksum algos*/
1140 #define OBD_CONNECT_FID 0x40000000ULL /*FID is supported by server */
1141 #define OBD_CONNECT_VBR 0x80000000ULL /*version based recovery */
1142 #define OBD_CONNECT_LOV_V3 0x100000000ULL /*client supports LOV v3 EA */
1143 #define OBD_CONNECT_GRANT_SHRINK 0x200000000ULL /* support grant shrink */
1144 #define OBD_CONNECT_SKIP_ORPHAN 0x400000000ULL /* don't reuse orphan objids */
1145 #define OBD_CONNECT_MAX_EASIZE 0x800000000ULL /* preserved for large EA */
1146 #define OBD_CONNECT_FULL20 0x1000000000ULL /* it is 2.0 client */
1147 #define OBD_CONNECT_LAYOUTLOCK 0x2000000000ULL /* client uses layout lock */
1148 #define OBD_CONNECT_64BITHASH 0x4000000000ULL /* client supports 64-bits
1150 #define OBD_CONNECT_MAXBYTES 0x8000000000ULL /* max stripe size */
1151 #define OBD_CONNECT_IMP_RECOV 0x10000000000ULL /* imp recovery support */
1152 #define OBD_CONNECT_JOBSTATS 0x20000000000ULL /* jobid in ptlrpc_body */
1153 #define OBD_CONNECT_UMASK 0x40000000000ULL /* create uses client umask */
1154 #define OBD_CONNECT_EINPROGRESS 0x80000000000ULL /* client handles -EINPROGRESS
1155 * RPC error properly */
1156 #define OBD_CONNECT_GRANT_PARAM 0x100000000000ULL/* extra grant params used for
1157 * finer space reservation */
1158 #define OBD_CONNECT_NANOSEC_TIME 0x200000000000ULL /* nanosecond timestamps */
1159 #define OBD_CONNECT_LVB_TYPE 0x400000000000ULL /* variable type of LVB */
1161 * Please DO NOT add flag values here before first ensuring that this same
1162 * flag value is not in use on some other branch. Please clear any such
1163 * changes with senior engineers before starting to use a new flag. Then,
1164 * submit a small patch against EVERY branch that ONLY adds the new flag
1165 * and updates obd_connect_names[] for lprocfs_rd_connect_flags(), so it
1166 * can be approved and landed easily to reserve the flag for future use. */
1168 /* The MNE_SWAB flag is overloading the MDS_MDS bit only for the MGS
1169 * connection. It is a temporary bug fix for Imperative Recovery interop
1170 * between 2.2 and 2.3 x86/ppc nodes, and can be removed when interop for
1171 * 2.2 clients/servers is no longer needed. LU-1252/LU-1644. */
1172 #define OBD_CONNECT_MNE_SWAB OBD_CONNECT_MDS_MDS
1174 #define OCD_HAS_FLAG(ocd, flg) \
1175 (!!((ocd)->ocd_connect_flags & OBD_CONNECT_##flg))
1178 #ifdef HAVE_LRU_RESIZE_SUPPORT
1179 #define LRU_RESIZE_CONNECT_FLAG OBD_CONNECT_LRU_RESIZE
1181 #define LRU_RESIZE_CONNECT_FLAG 0
1184 #define MDT_CONNECT_SUPPORTED (OBD_CONNECT_RDONLY | OBD_CONNECT_VERSION | \
1185 OBD_CONNECT_ACL | OBD_CONNECT_XATTR | \
1186 OBD_CONNECT_IBITS | \
1187 OBD_CONNECT_NODEVOH | OBD_CONNECT_ATTRFID | \
1188 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1189 OBD_CONNECT_RMT_CLIENT | \
1190 OBD_CONNECT_RMT_CLIENT_FORCE | \
1191 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_MDS_CAPA | \
1192 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_MDS_MDS | \
1193 OBD_CONNECT_FID | LRU_RESIZE_CONNECT_FLAG | \
1194 OBD_CONNECT_VBR | OBD_CONNECT_LOV_V3 | \
1195 OBD_CONNECT_SOM | OBD_CONNECT_FULL20 | \
1196 OBD_CONNECT_64BITHASH | \
1197 OBD_CONNECT_EINPROGRESS | OBD_CONNECT_JOBSTATS)
1198 #define OST_CONNECT_SUPPORTED (OBD_CONNECT_SRVLOCK | OBD_CONNECT_GRANT | \
1199 OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | \
1200 OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_INDEX | \
1201 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_QUOTA64 | \
1202 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1203 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_CKSUM | \
1204 OBD_CONNECT_CHANGE_QS | \
1205 OBD_CONNECT_OSS_CAPA | \
1206 OBD_CONNECT_RMT_CLIENT | \
1207 OBD_CONNECT_RMT_CLIENT_FORCE | OBD_CONNECT_VBR | \
1208 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN | \
1209 OBD_CONNECT_GRANT_SHRINK | OBD_CONNECT_FULL20 | \
1210 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES | \
1211 OBD_CONNECT_MAX_EASIZE | \
1212 OBD_CONNECT_EINPROGRESS | OBD_CONNECT_JOBSTATS)
1213 #define ECHO_CONNECT_SUPPORTED (0)
1214 #define MGS_CONNECT_SUPPORTED (OBD_CONNECT_VERSION | OBD_CONNECT_AT | \
1215 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV | \
1216 OBD_CONNECT_MNE_SWAB)
1218 /* Features required for this version of the client to work with server */
1219 #define CLIENT_CONNECT_MDT_REQD (OBD_CONNECT_IBITS | OBD_CONNECT_FID | \
1222 #define OBD_OCD_VERSION(major,minor,patch,fix) (((major)<<24) + ((minor)<<16) +\
1223 ((patch)<<8) + (fix))
1224 #define OBD_OCD_VERSION_MAJOR(version) ((int)((version)>>24)&255)
1225 #define OBD_OCD_VERSION_MINOR(version) ((int)((version)>>16)&255)
1226 #define OBD_OCD_VERSION_PATCH(version) ((int)((version)>>8)&255)
1227 #define OBD_OCD_VERSION_FIX(version) ((int)(version)&255)
1229 /* This structure is used for both request and reply.
1231 * If we eventually have separate connect data for different types, which we
1232 * almost certainly will, then perhaps we stick a union in here. */
1233 struct obd_connect_data_v1 {
1234 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1235 __u32 ocd_version; /* lustre release version number */
1236 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1237 __u32 ocd_index; /* LOV index to connect to */
1238 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
1239 __u64 ocd_ibits_known; /* inode bits this client understands */
1240 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1241 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1242 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1243 __u32 ocd_unused; /* also fix lustre_swab_connect */
1244 __u64 ocd_transno; /* first transno from client to be replayed */
1245 __u32 ocd_group; /* MDS group on OST */
1246 __u32 ocd_cksum_types; /* supported checksum algorithms */
1247 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1248 __u32 ocd_instance; /* also fix lustre_swab_connect */
1249 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1252 struct obd_connect_data {
1253 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1254 __u32 ocd_version; /* lustre release version number */
1255 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1256 __u32 ocd_index; /* LOV index to connect to */
1257 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
1258 __u64 ocd_ibits_known; /* inode bits this client understands */
1259 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1260 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1261 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1262 __u32 ocd_unused; /* also fix lustre_swab_connect */
1263 __u64 ocd_transno; /* first transno from client to be replayed */
1264 __u32 ocd_group; /* MDS group on OST */
1265 __u32 ocd_cksum_types; /* supported checksum algorithms */
1266 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1267 __u32 ocd_instance; /* instance # of this target */
1268 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1269 /* Fields after ocd_maxbytes are only accessible by the receiver
1270 * if the corresponding flag in ocd_connect_flags is set. Accessing
1271 * any field after ocd_maxbytes on the receiver without a valid flag
1272 * may result in out-of-bound memory access and kernel oops. */
1273 __u64 padding1; /* added 2.1.0. also fix lustre_swab_connect */
1274 __u64 padding2; /* added 2.1.0. also fix lustre_swab_connect */
1275 __u64 padding3; /* added 2.1.0. also fix lustre_swab_connect */
1276 __u64 padding4; /* added 2.1.0. also fix lustre_swab_connect */
1277 __u64 padding5; /* added 2.1.0. also fix lustre_swab_connect */
1278 __u64 padding6; /* added 2.1.0. also fix lustre_swab_connect */
1279 __u64 padding7; /* added 2.1.0. also fix lustre_swab_connect */
1280 __u64 padding8; /* added 2.1.0. also fix lustre_swab_connect */
1281 __u64 padding9; /* added 2.1.0. also fix lustre_swab_connect */
1282 __u64 paddingA; /* added 2.1.0. also fix lustre_swab_connect */
1283 __u64 paddingB; /* added 2.1.0. also fix lustre_swab_connect */
1284 __u64 paddingC; /* added 2.1.0. also fix lustre_swab_connect */
1285 __u64 paddingD; /* added 2.1.0. also fix lustre_swab_connect */
1286 __u64 paddingE; /* added 2.1.0. also fix lustre_swab_connect */
1287 __u64 paddingF; /* added 2.1.0. also fix lustre_swab_connect */
1290 * Please DO NOT use any fields here before first ensuring that this same
1291 * field is not in use on some other branch. Please clear any such changes
1292 * with senior engineers before starting to use a new field. Then, submit
1293 * a small patch against EVERY branch that ONLY adds the new field along with
1294 * the matching OBD_CONNECT flag, so that can be approved and landed easily to
1295 * reserve the flag for future use. */
1298 extern void lustre_swab_connect(struct obd_connect_data *ocd);
1301 * Supported checksum algorithms. Up to 32 checksum types are supported.
1302 * (32-bit mask stored in obd_connect_data::ocd_cksum_types)
1303 * Please update DECLARE_CKSUM_NAME/OBD_CKSUM_ALL in obd.h when adding a new
1304 * algorithm and also the OBD_FL_CKSUM* flags.
1307 OBD_CKSUM_CRC32 = 0x00000001,
1308 OBD_CKSUM_ADLER = 0x00000002,
1309 OBD_CKSUM_CRC32C= 0x00000004,
1313 * OST requests: OBDO & OBD request records
1318 OST_REPLY = 0, /* reply ? */
1334 OST_QUOTACHECK = 18,
1336 OST_QUOTA_ADJUST_QUNIT = 20,
1339 #define OST_FIRST_OPC OST_REPLY
1342 OBD_FL_INLINEDATA = 0x00000001,
1343 OBD_FL_OBDMDEXISTS = 0x00000002,
1344 OBD_FL_DELORPHAN = 0x00000004, /* if set in o_flags delete orphans */
1345 OBD_FL_NORPC = 0x00000008, /* set in o_flags do in OSC not OST */
1346 OBD_FL_IDONLY = 0x00000010, /* set in o_flags only adjust obj id*/
1347 OBD_FL_RECREATE_OBJS= 0x00000020, /* recreate missing obj */
1348 OBD_FL_DEBUG_CHECK = 0x00000040, /* echo client/server debug check */
1349 OBD_FL_NO_USRQUOTA = 0x00000100, /* the object's owner is over quota */
1350 OBD_FL_NO_GRPQUOTA = 0x00000200, /* the object's group is over quota */
1351 OBD_FL_CREATE_CROW = 0x00000400, /* object should be create on write */
1352 OBD_FL_SRVLOCK = 0x00000800, /* delegate DLM locking to server */
1353 OBD_FL_CKSUM_CRC32 = 0x00001000, /* CRC32 checksum type */
1354 OBD_FL_CKSUM_ADLER = 0x00002000, /* ADLER checksum type */
1355 OBD_FL_CKSUM_CRC32C = 0x00004000, /* CRC32C checksum type */
1356 OBD_FL_CKSUM_RSVD2 = 0x00008000, /* for future cksum types */
1357 OBD_FL_CKSUM_RSVD3 = 0x00010000, /* for future cksum types */
1358 OBD_FL_SHRINK_GRANT = 0x00020000, /* object shrink the grant */
1359 OBD_FL_MMAP = 0x00040000, /* object is mmapped on the client.
1360 * XXX: obsoleted - reserved for old
1361 * clients prior than 2.2 */
1362 OBD_FL_RECOV_RESEND = 0x00080000, /* recoverable resent */
1363 OBD_FL_NOSPC_BLK = 0x00100000, /* no more block space on OST */
1365 /* Note that while these checksum values are currently separate bits,
1366 * in 2.x we can actually allow all values from 1-31 if we wanted. */
1367 OBD_FL_CKSUM_ALL = OBD_FL_CKSUM_CRC32 | OBD_FL_CKSUM_ADLER |
1368 OBD_FL_CKSUM_CRC32C,
1370 /* mask for local-only flag, which won't be sent over network */
1371 OBD_FL_LOCAL_MASK = 0xF0000000,
1374 #define LOV_MAGIC_V1 0x0BD10BD0
1375 #define LOV_MAGIC LOV_MAGIC_V1
1376 #define LOV_MAGIC_JOIN_V1 0x0BD20BD0
1377 #define LOV_MAGIC_V3 0x0BD30BD0
1379 #define LOV_PATTERN_RAID0 0x001 /* stripes are used round-robin */
1380 #define LOV_PATTERN_RAID1 0x002 /* stripes are mirrors of each other */
1381 #define LOV_PATTERN_FIRST 0x100 /* first stripe is not in round-robin */
1382 #define LOV_PATTERN_CMOBD 0x200
1384 #define lov_ost_data lov_ost_data_v1
1385 struct lov_ost_data_v1 { /* per-stripe data structure (little-endian)*/
1386 __u64 l_object_id; /* OST object ID */
1387 __u64 l_object_seq; /* OST object seq number */
1388 __u32 l_ost_gen; /* generation of this l_ost_idx */
1389 __u32 l_ost_idx; /* OST index in LOV (lov_tgt_desc->tgts) */
1392 #define lov_mds_md lov_mds_md_v1
1393 struct lov_mds_md_v1 { /* LOV EA mds/wire data (little-endian) */
1394 __u32 lmm_magic; /* magic number = LOV_MAGIC_V1 */
1395 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1396 __u64 lmm_object_id; /* LOV object ID */
1397 __u64 lmm_object_seq; /* LOV object seq number */
1398 __u32 lmm_stripe_size; /* size of stripe in bytes */
1399 /* lmm_stripe_count used to be __u32 */
1400 __u16 lmm_stripe_count; /* num stripes in use for this object */
1401 __u16 lmm_layout_gen; /* layout generation number */
1402 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1405 /* extern void lustre_swab_lov_mds_md(struct lov_mds_md *llm); */
1407 #define MAX_MD_SIZE (sizeof(struct lov_mds_md) + 4 * sizeof(struct lov_ost_data))
1408 #define MIN_MD_SIZE (sizeof(struct lov_mds_md) + 1 * sizeof(struct lov_ost_data))
1410 #define XATTR_NAME_ACL_ACCESS "system.posix_acl_access"
1411 #define XATTR_NAME_ACL_DEFAULT "system.posix_acl_default"
1412 #define XATTR_USER_PREFIX "user."
1413 #define XATTR_TRUSTED_PREFIX "trusted."
1414 #define XATTR_SECURITY_PREFIX "security."
1415 #define XATTR_LUSTRE_PREFIX "lustre."
1417 #define XATTR_NAME_LOV "trusted.lov"
1418 #define XATTR_NAME_LMA "trusted.lma"
1419 #define XATTR_NAME_LMV "trusted.lmv"
1420 #define XATTR_NAME_LINK "trusted.link"
1421 #define XATTR_NAME_FID "trusted.fid"
1422 #define XATTR_NAME_VERSION "trusted.version"
1425 struct lov_mds_md_v3 { /* LOV EA mds/wire data (little-endian) */
1426 __u32 lmm_magic; /* magic number = LOV_MAGIC_V3 */
1427 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1428 __u64 lmm_object_id; /* LOV object ID */
1429 __u64 lmm_object_seq; /* LOV object seq number */
1430 __u32 lmm_stripe_size; /* size of stripe in bytes */
1431 /* lmm_stripe_count used to be __u32 */
1432 __u16 lmm_stripe_count; /* num stripes in use for this object */
1433 __u16 lmm_layout_gen; /* layout generation number */
1434 char lmm_pool_name[LOV_MAXPOOLNAME]; /* must be 32bit aligned */
1435 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1439 #define OBD_MD_FLID (0x00000001ULL) /* object ID */
1440 #define OBD_MD_FLATIME (0x00000002ULL) /* access time */
1441 #define OBD_MD_FLMTIME (0x00000004ULL) /* data modification time */
1442 #define OBD_MD_FLCTIME (0x00000008ULL) /* change time */
1443 #define OBD_MD_FLSIZE (0x00000010ULL) /* size */
1444 #define OBD_MD_FLBLOCKS (0x00000020ULL) /* allocated blocks count */
1445 #define OBD_MD_FLBLKSZ (0x00000040ULL) /* block size */
1446 #define OBD_MD_FLMODE (0x00000080ULL) /* access bits (mode & ~S_IFMT) */
1447 #define OBD_MD_FLTYPE (0x00000100ULL) /* object type (mode & S_IFMT) */
1448 #define OBD_MD_FLUID (0x00000200ULL) /* user ID */
1449 #define OBD_MD_FLGID (0x00000400ULL) /* group ID */
1450 #define OBD_MD_FLFLAGS (0x00000800ULL) /* flags word */
1451 #define OBD_MD_FLNLINK (0x00002000ULL) /* link count */
1452 #define OBD_MD_FLGENER (0x00004000ULL) /* generation number */
1453 /*#define OBD_MD_FLINLINE (0x00008000ULL) inline data. used until 1.6.5 */
1454 #define OBD_MD_FLRDEV (0x00010000ULL) /* device number */
1455 #define OBD_MD_FLEASIZE (0x00020000ULL) /* extended attribute data */
1456 #define OBD_MD_LINKNAME (0x00040000ULL) /* symbolic link target */
1457 #define OBD_MD_FLHANDLE (0x00080000ULL) /* file/lock handle */
1458 #define OBD_MD_FLCKSUM (0x00100000ULL) /* bulk data checksum */
1459 #define OBD_MD_FLQOS (0x00200000ULL) /* quality of service stats */
1460 /*#define OBD_MD_FLOSCOPQ (0x00400000ULL) osc opaque data, never used */
1461 #define OBD_MD_FLCOOKIE (0x00800000ULL) /* log cancellation cookie */
1462 #define OBD_MD_FLGROUP (0x01000000ULL) /* group */
1463 #define OBD_MD_FLFID (0x02000000ULL) /* ->ost write inline fid */
1464 #define OBD_MD_FLEPOCH (0x04000000ULL) /* ->ost write with ioepoch */
1465 /* ->mds if epoch opens or closes */
1466 #define OBD_MD_FLGRANT (0x08000000ULL) /* ost preallocation space grant */
1467 #define OBD_MD_FLDIREA (0x10000000ULL) /* dir's extended attribute data */
1468 #define OBD_MD_FLUSRQUOTA (0x20000000ULL) /* over quota flags sent from ost */
1469 #define OBD_MD_FLGRPQUOTA (0x40000000ULL) /* over quota flags sent from ost */
1470 #define OBD_MD_FLMODEASIZE (0x80000000ULL) /* EA size will be changed */
1472 #define OBD_MD_MDS (0x0000000100000000ULL) /* where an inode lives on */
1473 #define OBD_MD_REINT (0x0000000200000000ULL) /* reintegrate oa */
1474 #define OBD_MD_MEA (0x0000000400000000ULL) /* CMD split EA */
1475 #define OBD_MD_MDTIDX (0x0000000800000000ULL) /* Get MDT index */
1477 #define OBD_MD_FLXATTR (0x0000001000000000ULL) /* xattr */
1478 #define OBD_MD_FLXATTRLS (0x0000002000000000ULL) /* xattr list */
1479 #define OBD_MD_FLXATTRRM (0x0000004000000000ULL) /* xattr remove */
1480 #define OBD_MD_FLACL (0x0000008000000000ULL) /* ACL */
1481 #define OBD_MD_FLRMTPERM (0x0000010000000000ULL) /* remote permission */
1482 #define OBD_MD_FLMDSCAPA (0x0000020000000000ULL) /* MDS capability */
1483 #define OBD_MD_FLOSSCAPA (0x0000040000000000ULL) /* OSS capability */
1484 #define OBD_MD_FLCKSPLIT (0x0000080000000000ULL) /* Check split on server */
1485 #define OBD_MD_FLCROSSREF (0x0000100000000000ULL) /* Cross-ref case */
1486 #define OBD_MD_FLGETATTRLOCK (0x0000200000000000ULL) /* Get IOEpoch attributes
1488 #define OBD_MD_FLOBJCOUNT (0x0000400000000000ULL) /* for multiple destroy */
1490 #define OBD_MD_FLRMTLSETFACL (0x0001000000000000ULL) /* lfs lsetfacl case */
1491 #define OBD_MD_FLRMTLGETFACL (0x0002000000000000ULL) /* lfs lgetfacl case */
1492 #define OBD_MD_FLRMTRSETFACL (0x0004000000000000ULL) /* lfs rsetfacl case */
1493 #define OBD_MD_FLRMTRGETFACL (0x0008000000000000ULL) /* lfs rgetfacl case */
1495 #define OBD_MD_FLDATAVERSION (0x0010000000000000ULL) /* iversion sum */
1497 #define OBD_MD_FLGETATTR (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLMTIME | \
1498 OBD_MD_FLCTIME | OBD_MD_FLSIZE | OBD_MD_FLBLKSZ | \
1499 OBD_MD_FLMODE | OBD_MD_FLTYPE | OBD_MD_FLUID | \
1500 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLNLINK | \
1501 OBD_MD_FLGENER | OBD_MD_FLRDEV | OBD_MD_FLGROUP)
1503 /* don't forget obdo_fid which is way down at the bottom so it can
1504 * come after the definition of llog_cookie */
1507 extern void lustre_swab_obd_statfs (struct obd_statfs *os);
1509 /* ost_body.data values for OST_BRW */
1511 #define OBD_BRW_READ 0x01
1512 #define OBD_BRW_WRITE 0x02
1513 #define OBD_BRW_RWMASK (OBD_BRW_READ | OBD_BRW_WRITE)
1514 #define OBD_BRW_SYNC 0x08 /* this page is a part of synchronous
1515 * transfer and is not accounted in
1517 #define OBD_BRW_CHECK 0x10
1518 #define OBD_BRW_FROM_GRANT 0x20 /* the osc manages this under llite */
1519 #define OBD_BRW_GRANTED 0x40 /* the ost manages this */
1520 #define OBD_BRW_NOCACHE 0x80 /* this page is a part of non-cached IO */
1521 #define OBD_BRW_NOQUOTA 0x100
1522 #define OBD_BRW_SRVLOCK 0x200 /* Client holds no lock over this page */
1523 #define OBD_BRW_ASYNC 0x400 /* Server may delay commit to disk */
1524 #define OBD_BRW_MEMALLOC 0x800 /* Client runs in the "kswapd" context */
1526 #define OBD_OBJECT_EOF 0xffffffffffffffffULL
1528 #define OST_MIN_PRECREATE 32
1529 #define OST_MAX_PRECREATE 20000
1538 extern void lustre_swab_obd_ioobj (struct obd_ioobj *ioo);
1540 /* multiple of 8 bytes => can array */
1541 struct niobuf_remote {
1547 extern void lustre_swab_niobuf_remote (struct niobuf_remote *nbr);
1549 /* lock value block communicated between the filter and llite */
1551 /* OST_LVB_ERR_INIT is needed because the return code in rc is
1552 * negative, i.e. because ((MASK + rc) & MASK) != MASK. */
1553 #define OST_LVB_ERR_INIT 0xffbadbad80000000ULL
1554 #define OST_LVB_ERR_MASK 0xffbadbad00000000ULL
1555 #define OST_LVB_IS_ERR(blocks) \
1556 ((blocks & OST_LVB_ERR_MASK) == OST_LVB_ERR_MASK)
1557 #define OST_LVB_SET_ERR(blocks, rc) \
1558 do { blocks = OST_LVB_ERR_INIT + rc; } while (0)
1559 #define OST_LVB_GET_ERR(blocks) (int)(blocks - OST_LVB_ERR_INIT)
1576 MDS_GETATTR_NAME = 34,
1581 MDS_DISCONNECT = 39,
1587 MDS_DONE_WRITING = 45,
1589 MDS_QUOTACHECK = 47,
1592 MDS_SETXATTR = 50, /* obsolete, now it's MDS_REINT op */
1599 #define MDS_FIRST_OPC MDS_GETATTR
1616 } mds_reint_t, mdt_reint_t;
1618 extern void lustre_swab_generic_32s (__u32 *val);
1620 /* the disposition of the intent outlines what was executed */
1621 #define DISP_IT_EXECD 0x00000001
1622 #define DISP_LOOKUP_EXECD 0x00000002
1623 #define DISP_LOOKUP_NEG 0x00000004
1624 #define DISP_LOOKUP_POS 0x00000008
1625 #define DISP_OPEN_CREATE 0x00000010
1626 #define DISP_OPEN_OPEN 0x00000020
1627 #define DISP_ENQ_COMPLETE 0x00400000
1628 #define DISP_ENQ_OPEN_REF 0x00800000
1629 #define DISP_ENQ_CREATE_REF 0x01000000
1630 #define DISP_OPEN_LOCK 0x02000000
1632 /* INODE LOCK PARTS */
1633 #define MDS_INODELOCK_LOOKUP 0x000001 /* dentry, mode, owner, group */
1634 #define MDS_INODELOCK_UPDATE 0x000002 /* size, links, timestamps */
1635 #define MDS_INODELOCK_OPEN 0x000004 /* For opened files */
1636 #define MDS_INODELOCK_LAYOUT 0x000008 /* for layout */
1638 /* Do not forget to increase MDS_INODELOCK_MAXSHIFT when adding new bits
1639 * XXX: MDS_INODELOCK_MAXSHIFT should be increased to 3 once the layout lock is
1641 #define MDS_INODELOCK_MAXSHIFT 2
1642 /* This FULL lock is useful to take on unlink sort of operations */
1643 #define MDS_INODELOCK_FULL ((1<<(MDS_INODELOCK_MAXSHIFT+1))-1)
1645 extern void lustre_swab_ll_fid (struct ll_fid *fid);
1647 /* NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
1648 * but was moved into name[1] along with the OID to avoid consuming the
1649 * name[2,3] fields that need to be used for the quota id (also a FID). */
1651 LUSTRE_RES_ID_SEQ_OFF = 0,
1652 LUSTRE_RES_ID_VER_OID_OFF = 1,
1653 LUSTRE_RES_ID_WAS_VER_OFF = 2, /* see note above */
1654 LUSTRE_RES_ID_HSH_OFF = 3
1657 #define MDS_STATUS_CONN 1
1658 #define MDS_STATUS_LOV 2
1660 /* mdt_thread_info.mti_flags. */
1662 /* The flag indicates Size-on-MDS attributes are changed. */
1663 MF_SOM_CHANGE = (1 << 0),
1664 /* Flags indicates an epoch opens or closes. */
1665 MF_EPOCH_OPEN = (1 << 1),
1666 MF_EPOCH_CLOSE = (1 << 2),
1667 MF_MDC_CANCEL_FID1 = (1 << 3),
1668 MF_MDC_CANCEL_FID2 = (1 << 4),
1669 MF_MDC_CANCEL_FID3 = (1 << 5),
1670 MF_MDC_CANCEL_FID4 = (1 << 6),
1671 /* There is a pending attribute update. */
1672 MF_SOM_AU = (1 << 7),
1673 /* Cancel OST locks while getattr OST attributes. */
1674 MF_GETATTR_LOCK = (1 << 8),
1677 #define MF_SOM_LOCAL_FLAGS (MF_SOM_CHANGE | MF_EPOCH_OPEN | MF_EPOCH_CLOSE)
1679 #define LUSTRE_BFLAG_UNCOMMITTED_WRITES 0x1
1681 /* these should be identical to their EXT4_*_FL counterparts, they are
1682 * redefined here only to avoid dragging in fs/ext4/ext4.h */
1683 #define LUSTRE_SYNC_FL 0x00000008 /* Synchronous updates */
1684 #define LUSTRE_IMMUTABLE_FL 0x00000010 /* Immutable file */
1685 #define LUSTRE_APPEND_FL 0x00000020 /* writes to file may only append */
1686 #define LUSTRE_NOATIME_FL 0x00000080 /* do not update atime */
1687 #define LUSTRE_DIRSYNC_FL 0x00010000 /* dirsync behaviour (dir only) */
1690 /* Convert wire LUSTRE_*_FL to corresponding client local VFS S_* values
1691 * for the client inode i_flags. The LUSTRE_*_FL are the Lustre wire
1692 * protocol equivalents of LDISKFS_*_FL values stored on disk, while
1693 * the S_* flags are kernel-internal values that change between kernel
1694 * versions. These flags are set/cleared via FSFILT_IOC_{GET,SET}_FLAGS.
1695 * See b=16526 for a full history. */
1696 static inline int ll_ext_to_inode_flags(int flags)
1698 return (((flags & LUSTRE_SYNC_FL) ? S_SYNC : 0) |
1699 ((flags & LUSTRE_NOATIME_FL) ? S_NOATIME : 0) |
1700 ((flags & LUSTRE_APPEND_FL) ? S_APPEND : 0) |
1701 #if defined(S_DIRSYNC)
1702 ((flags & LUSTRE_DIRSYNC_FL) ? S_DIRSYNC : 0) |
1704 ((flags & LUSTRE_IMMUTABLE_FL) ? S_IMMUTABLE : 0));
1707 static inline int ll_inode_to_ext_flags(int iflags)
1709 return (((iflags & S_SYNC) ? LUSTRE_SYNC_FL : 0) |
1710 ((iflags & S_NOATIME) ? LUSTRE_NOATIME_FL : 0) |
1711 ((iflags & S_APPEND) ? LUSTRE_APPEND_FL : 0) |
1712 #if defined(S_DIRSYNC)
1713 ((iflags & S_DIRSYNC) ? LUSTRE_DIRSYNC_FL : 0) |
1715 ((iflags & S_IMMUTABLE) ? LUSTRE_IMMUTABLE_FL : 0));
1722 struct lustre_handle handle;
1724 __u64 size; /* Offset, in the case of MDS_READPAGE */
1728 __u64 blocks; /* XID, in the case of MDS_READPAGE */
1737 __u32 flags; /* from vfs for pin/unpin, LUSTRE_BFLAG close */
1739 __u32 nlink; /* #bytes to read in the case of MDS_READPAGE */
1745 __u32 max_cookiesize;
1746 __u32 uid_h; /* high 32-bits of uid, for FUID */
1747 __u32 gid_h; /* high 32-bits of gid, for FUID */
1748 __u32 padding_5; /* also fix lustre_swab_mdt_body */
1756 extern void lustre_swab_mdt_body (struct mdt_body *b);
1758 struct mdt_ioepoch {
1759 struct lustre_handle handle;
1765 extern void lustre_swab_mdt_ioepoch (struct mdt_ioepoch *b);
1767 #define Q_QUOTACHECK 0x800100
1768 #define Q_INITQUOTA 0x800101 /* init slave limits */
1769 #define Q_GETOINFO 0x800102 /* get obd quota info */
1770 #define Q_GETOQUOTA 0x800103 /* get obd quotas */
1771 #define Q_FINVALIDATE 0x800104 /* invalidate operational quotas */
1773 #define Q_TYPEMATCH(id, type) \
1774 ((id) == (type) || (id) == UGQUOTA)
1776 #define Q_TYPESET(oqc, type) Q_TYPEMATCH((oqc)->qc_type, type)
1778 #define Q_GETOCMD(oqc) \
1779 ((oqc)->qc_cmd == Q_GETOINFO || (oqc)->qc_cmd == Q_GETOQUOTA)
1781 #define QCTL_COPY(out, in) \
1783 Q_COPY(out, in, qc_cmd); \
1784 Q_COPY(out, in, qc_type); \
1785 Q_COPY(out, in, qc_id); \
1786 Q_COPY(out, in, qc_stat); \
1787 Q_COPY(out, in, qc_dqinfo); \
1788 Q_COPY(out, in, qc_dqblk); \
1791 struct obd_quotactl {
1796 struct obd_dqinfo qc_dqinfo;
1797 struct obd_dqblk qc_dqblk;
1800 extern void lustre_swab_obd_quotactl(struct obd_quotactl *q);
1802 struct quota_adjust_qunit {
1809 extern void lustre_swab_quota_adjust_qunit(struct quota_adjust_qunit *q);
1811 /* flags is shared among quota structures */
1812 #define LQUOTA_FLAGS_GRP 1UL /* 0 is user, 1 is group */
1813 #define LQUOTA_FLAGS_BLK 2UL /* 0 is inode, 1 is block */
1814 #define LQUOTA_FLAGS_ADJBLK 4UL /* adjust the block qunit size */
1815 #define LQUOTA_FLAGS_ADJINO 8UL /* adjust the inode qunit size */
1816 #define LQUOTA_FLAGS_CHG_QS 16UL /* indicate whether it has capability of
1817 * OBD_CONNECT_CHANGE_QS */
1818 #define LQUOTA_FLAGS_RECOVERY 32UL /* recovery is going on a uid/gid */
1819 #define LQUOTA_FLAGS_SETQUOTA 64UL /* being setquota on a uid/gid */
1821 /* flags is specific for quota_adjust_qunit */
1822 #define LQUOTA_QAQ_CREATE_LQS (1UL << 31) /* when it is set, need create lqs */
1824 /* the status of lqs_flags in struct lustre_qunit_size */
1825 #define LQUOTA_QUNIT_FLAGS (LQUOTA_FLAGS_GRP | LQUOTA_FLAGS_BLK)
1827 #define QAQ_IS_GRP(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_GRP)
1828 #define QAQ_IS_ADJBLK(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_ADJBLK)
1829 #define QAQ_IS_ADJINO(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_ADJINO)
1830 #define QAQ_IS_CREATE_LQS(qaq) ((qaq)->qaq_flags & LQUOTA_QAQ_CREATE_LQS)
1832 #define QAQ_SET_GRP(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_GRP)
1833 #define QAQ_SET_ADJBLK(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_ADJBLK)
1834 #define QAQ_SET_ADJINO(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_ADJINO)
1835 #define QAQ_SET_CREATE_LQS(qaq) ((qaq)->qaq_flags |= LQUOTA_QAQ_CREATE_LQS)
1837 /* permissions for md_perm.mp_perm */
1839 CFS_SETUID_PERM = 0x01,
1840 CFS_SETGID_PERM = 0x02,
1841 CFS_SETGRP_PERM = 0x04,
1842 CFS_RMTACL_PERM = 0x08,
1843 CFS_RMTOWN_PERM = 0x10
1846 /* inode access permission for remote user, the inode info are omitted,
1847 * for client knows them. */
1848 struct mdt_remote_perm {
1855 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
1859 extern void lustre_swab_mdt_remote_perm(struct mdt_remote_perm *p);
1861 struct mdt_rec_setattr {
1871 __u32 sa_padding_1_h;
1872 struct lu_fid sa_fid;
1881 __u32 sa_attr_flags;
1889 extern void lustre_swab_mdt_rec_setattr (struct mdt_rec_setattr *sa);
1892 * Attribute flags used in mdt_rec_setattr::sa_valid.
1893 * The kernel's #defines for ATTR_* should not be used over the network
1894 * since the client and MDS may run different kernels (see bug 13828)
1895 * Therefore, we should only use MDS_ATTR_* attributes for sa_valid.
1897 #define MDS_ATTR_MODE 0x1ULL /* = 1 */
1898 #define MDS_ATTR_UID 0x2ULL /* = 2 */
1899 #define MDS_ATTR_GID 0x4ULL /* = 4 */
1900 #define MDS_ATTR_SIZE 0x8ULL /* = 8 */
1901 #define MDS_ATTR_ATIME 0x10ULL /* = 16 */
1902 #define MDS_ATTR_MTIME 0x20ULL /* = 32 */
1903 #define MDS_ATTR_CTIME 0x40ULL /* = 64 */
1904 #define MDS_ATTR_ATIME_SET 0x80ULL /* = 128 */
1905 #define MDS_ATTR_MTIME_SET 0x100ULL /* = 256 */
1906 #define MDS_ATTR_FORCE 0x200ULL /* = 512, Not a change, but a change it */
1907 #define MDS_ATTR_ATTR_FLAG 0x400ULL /* = 1024 */
1908 #define MDS_ATTR_KILL_SUID 0x800ULL /* = 2048 */
1909 #define MDS_ATTR_KILL_SGID 0x1000ULL /* = 4096 */
1910 #define MDS_ATTR_CTIME_SET 0x2000ULL /* = 8192 */
1911 #define MDS_ATTR_FROM_OPEN 0x4000ULL /* = 16384, called from open path, ie O_TRUNC */
1912 #define MDS_ATTR_BLOCKS 0x8000ULL /* = 32768 */
1915 #define FMODE_READ 00000001
1916 #define FMODE_WRITE 00000002
1919 #define MDS_FMODE_CLOSED 00000000
1920 #define MDS_FMODE_EXEC 00000004
1921 /* IO Epoch is opened on a closed file. */
1922 #define MDS_FMODE_EPOCH 01000000
1923 /* IO Epoch is opened on a file truncate. */
1924 #define MDS_FMODE_TRUNC 02000000
1925 /* Size-on-MDS Attribute Update is pending. */
1926 #define MDS_FMODE_SOM 04000000
1928 #define MDS_OPEN_CREATED 00000010
1929 #define MDS_OPEN_CROSS 00000020
1931 #define MDS_OPEN_CREAT 00000100
1932 #define MDS_OPEN_EXCL 00000200
1933 #define MDS_OPEN_TRUNC 00001000
1934 #define MDS_OPEN_APPEND 00002000
1935 #define MDS_OPEN_SYNC 00010000
1936 #define MDS_OPEN_DIRECTORY 00200000
1938 #define MDS_OPEN_BY_FID 040000000 /* open_by_fid for known object */
1939 #define MDS_OPEN_DELAY_CREATE 0100000000 /* delay initial object create */
1940 #define MDS_OPEN_OWNEROVERRIDE 0200000000 /* NFSD rw-reopen ro file for owner */
1941 #define MDS_OPEN_JOIN_FILE 0400000000 /* open for join file.
1942 * We do not support JOIN FILE
1943 * anymore, reserve this flags
1944 * just for preventing such bit
1946 #define MDS_CREATE_RMT_ACL 01000000000 /* indicate create on remote server
1947 * with default ACL */
1948 #define MDS_CREATE_SLAVE_OBJ 02000000000 /* indicate create slave object
1949 * actually, this is for create, not
1950 * conflict with other open flags */
1951 #define MDS_OPEN_LOCK 04000000000 /* This open requires open lock */
1952 #define MDS_OPEN_HAS_EA 010000000000 /* specify object create pattern */
1953 #define MDS_OPEN_HAS_OBJS 020000000000 /* Just set the EA the obj exist */
1954 #define MDS_OPEN_NORESTORE 0100000000000ULL /* Do not restore file at open */
1955 #define MDS_OPEN_NEWSTRIPE 0200000000000ULL /* New stripe needed (restripe or
1958 /* permission for create non-directory file */
1959 #define MAY_CREATE (1 << 7)
1960 /* permission for create directory file */
1961 #define MAY_LINK (1 << 8)
1962 /* permission for delete from the directory */
1963 #define MAY_UNLINK (1 << 9)
1964 /* source's permission for rename */
1965 #define MAY_RENAME_SRC (1 << 10)
1966 /* target's permission for rename */
1967 #define MAY_RENAME_TAR (1 << 11)
1968 /* part (parent's) VTX permission check */
1969 #define MAY_VTX_PART (1 << 12)
1970 /* full VTX permission check */
1971 #define MAY_VTX_FULL (1 << 13)
1972 /* lfs rgetfacl permission check */
1973 #define MAY_RGETFACL (1 << 14)
1976 MDS_CHECK_SPLIT = 1 << 0,
1977 MDS_CROSS_REF = 1 << 1,
1978 MDS_VTX_BYPASS = 1 << 2,
1979 MDS_PERM_BYPASS = 1 << 3,
1981 MDS_QUOTA_IGNORE = 1 << 5,
1982 MDS_CLOSE_CLEANUP = 1 << 6,
1983 MDS_KEEP_ORPHAN = 1 << 7,
1984 MDS_RECOV_OPEN = 1 << 8,
1985 MDS_UNLINK_DESTROY = 1 << 9, /* Destory ost object in mdd_unlink */
1988 /* instance of mdt_reint_rec */
1989 struct mdt_rec_create {
1997 __u32 cr_suppgid1_h;
1999 __u32 cr_suppgid2_h;
2000 struct lu_fid cr_fid1;
2001 struct lu_fid cr_fid2;
2002 struct lustre_handle cr_old_handle; /* handle in case of open replay */
2006 __u64 cr_padding_1; /* rr_blocks */
2009 /* use of helpers set/get_mrc_cr_flags() is needed to access
2010 * 64 bits cr_flags [cr_flags_l, cr_flags_h], this is done to
2011 * extend cr_flags size without breaking 1.8 compat */
2012 __u32 cr_flags_l; /* for use with open, low 32 bits */
2013 __u32 cr_flags_h; /* for use with open, high 32 bits */
2014 __u32 cr_padding_3; /* rr_padding_3 */
2015 __u32 cr_padding_4; /* rr_padding_4 */
2018 static inline void set_mrc_cr_flags(struct mdt_rec_create *mrc, __u64 flags)
2020 mrc->cr_flags_l = (__u32)(flags & 0xFFFFFFFFUll);
2021 mrc->cr_flags_h = (__u32)(flags >> 32);
2024 static inline __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
2026 return ((__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32));
2029 /* instance of mdt_reint_rec */
2030 struct mdt_rec_link {
2038 __u32 lk_suppgid1_h;
2040 __u32 lk_suppgid2_h;
2041 struct lu_fid lk_fid1;
2042 struct lu_fid lk_fid2;
2044 __u64 lk_padding_1; /* rr_atime */
2045 __u64 lk_padding_2; /* rr_ctime */
2046 __u64 lk_padding_3; /* rr_size */
2047 __u64 lk_padding_4; /* rr_blocks */
2049 __u32 lk_padding_5; /* rr_mode */
2050 __u32 lk_padding_6; /* rr_flags */
2051 __u32 lk_padding_7; /* rr_padding_2 */
2052 __u32 lk_padding_8; /* rr_padding_3 */
2053 __u32 lk_padding_9; /* rr_padding_4 */
2056 /* instance of mdt_reint_rec */
2057 struct mdt_rec_unlink {
2065 __u32 ul_suppgid1_h;
2067 __u32 ul_suppgid2_h;
2068 struct lu_fid ul_fid1;
2069 struct lu_fid ul_fid2;
2071 __u64 ul_padding_2; /* rr_atime */
2072 __u64 ul_padding_3; /* rr_ctime */
2073 __u64 ul_padding_4; /* rr_size */
2074 __u64 ul_padding_5; /* rr_blocks */
2077 __u32 ul_padding_6; /* rr_flags */
2078 __u32 ul_padding_7; /* rr_padding_2 */
2079 __u32 ul_padding_8; /* rr_padding_3 */
2080 __u32 ul_padding_9; /* rr_padding_4 */
2083 /* instance of mdt_reint_rec */
2084 struct mdt_rec_rename {
2092 __u32 rn_suppgid1_h;
2094 __u32 rn_suppgid2_h;
2095 struct lu_fid rn_fid1;
2096 struct lu_fid rn_fid2;
2098 __u64 rn_padding_1; /* rr_atime */
2099 __u64 rn_padding_2; /* rr_ctime */
2100 __u64 rn_padding_3; /* rr_size */
2101 __u64 rn_padding_4; /* rr_blocks */
2102 __u32 rn_bias; /* some operation flags */
2103 __u32 rn_mode; /* cross-ref rename has mode */
2104 __u32 rn_padding_5; /* rr_flags */
2105 __u32 rn_padding_6; /* rr_padding_2 */
2106 __u32 rn_padding_7; /* rr_padding_3 */
2107 __u32 rn_padding_8; /* rr_padding_4 */
2110 /* instance of mdt_reint_rec */
2111 struct mdt_rec_setxattr {
2119 __u32 sx_suppgid1_h;
2121 __u32 sx_suppgid2_h;
2122 struct lu_fid sx_fid;
2123 __u64 sx_padding_1; /* These three are rr_fid2 */
2128 __u64 sx_padding_5; /* rr_ctime */
2129 __u64 sx_padding_6; /* rr_size */
2130 __u64 sx_padding_7; /* rr_blocks */
2133 __u32 sx_padding_8; /* rr_flags */
2134 __u32 sx_padding_9; /* rr_padding_2 */
2135 __u32 sx_padding_10; /* rr_padding_3 */
2136 __u32 sx_padding_11; /* rr_padding_4 */
2140 * mdt_rec_reint is the template for all mdt_reint_xxx structures.
2141 * Do NOT change the size of various members, otherwise the value
2142 * will be broken in lustre_swab_mdt_rec_reint().
2144 * If you add new members in other mdt_reint_xxx structres and need to use the
2145 * rr_padding_x fields, then update lustre_swab_mdt_rec_reint() also.
2147 struct mdt_rec_reint {
2155 __u32 rr_suppgid1_h;
2157 __u32 rr_suppgid2_h;
2158 struct lu_fid rr_fid1;
2159 struct lu_fid rr_fid2;
2168 __u32 rr_padding_2; /* also fix lustre_swab_mdt_rec_reint */
2169 __u32 rr_padding_3; /* also fix lustre_swab_mdt_rec_reint */
2170 __u32 rr_padding_4; /* also fix lustre_swab_mdt_rec_reint */
2173 extern void lustre_swab_mdt_rec_reint(struct mdt_rec_reint *rr);
2176 __u32 ld_tgt_count; /* how many MDS's */
2177 __u32 ld_active_tgt_count; /* how many active */
2178 __u32 ld_default_stripe_count; /* how many objects are used */
2179 __u32 ld_pattern; /* default MEA_MAGIC_* */
2180 __u64 ld_default_hash_size;
2181 __u64 ld_padding_1; /* also fix lustre_swab_lmv_desc */
2182 __u32 ld_padding_2; /* also fix lustre_swab_lmv_desc */
2183 __u32 ld_qos_maxage; /* in second */
2184 __u32 ld_padding_3; /* also fix lustre_swab_lmv_desc */
2185 __u32 ld_padding_4; /* also fix lustre_swab_lmv_desc */
2186 struct obd_uuid ld_uuid;
2189 extern void lustre_swab_lmv_desc (struct lmv_desc *ld);
2191 /* TODO: lmv_stripe_md should contain mds capabilities for all slave fids */
2192 struct lmv_stripe_md {
2197 char mea_pool_name[LOV_MAXPOOLNAME];
2198 struct lu_fid mea_ids[0];
2201 extern void lustre_swab_lmv_stripe_md(struct lmv_stripe_md *mea);
2203 /* lmv structures */
2204 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
2205 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
2206 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
2208 #define MAX_HASH_SIZE_32 0x7fffffffUL
2209 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
2210 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
2215 FLD_FIRST_OPC = FLD_QUERY
2221 SEQ_FIRST_OPC = SEQ_QUERY
2225 SEQ_ALLOC_SUPER = 0,
2230 * LOV data structures
2233 #define LOV_MAX_UUID_BUFFER_SIZE 8192
2234 /* The size of the buffer the lov/mdc reserves for the
2235 * array of UUIDs returned by the MDS. With the current
2236 * protocol, this will limit the max number of OSTs per LOV */
2238 #define LOV_DESC_MAGIC 0xB0CCDE5C
2240 /* LOV settings descriptor (should only contain static info) */
2242 __u32 ld_tgt_count; /* how many OBD's */
2243 __u32 ld_active_tgt_count; /* how many active */
2244 __u32 ld_default_stripe_count; /* how many objects are used */
2245 __u32 ld_pattern; /* default PATTERN_RAID0 */
2246 __u64 ld_default_stripe_size; /* in bytes */
2247 __u64 ld_default_stripe_offset; /* in bytes */
2248 __u32 ld_padding_0; /* unused */
2249 __u32 ld_qos_maxage; /* in second */
2250 __u32 ld_padding_1; /* also fix lustre_swab_lov_desc */
2251 __u32 ld_padding_2; /* also fix lustre_swab_lov_desc */
2252 struct obd_uuid ld_uuid;
2255 #define ld_magic ld_active_tgt_count /* for swabbing from llogs */
2257 extern void lustre_swab_lov_desc (struct lov_desc *ld);
2262 /* opcodes -- MUST be distinct from OST/MDS opcodes */
2267 LDLM_BL_CALLBACK = 104,
2268 LDLM_CP_CALLBACK = 105,
2269 LDLM_GL_CALLBACK = 106,
2270 LDLM_SET_INFO = 107,
2273 #define LDLM_FIRST_OPC LDLM_ENQUEUE
2275 #define RES_NAME_SIZE 4
2276 struct ldlm_res_id {
2277 __u64 name[RES_NAME_SIZE];
2280 extern void lustre_swab_ldlm_res_id (struct ldlm_res_id *id);
2282 static inline int ldlm_res_eq(const struct ldlm_res_id *res0,
2283 const struct ldlm_res_id *res1)
2285 return !memcmp(res0, res1, sizeof(*res0));
2302 #define LCK_MODE_NUM 8
2312 #define LDLM_MIN_TYPE LDLM_PLAIN
2314 struct ldlm_extent {
2320 static inline int ldlm_extent_overlap(struct ldlm_extent *ex1,
2321 struct ldlm_extent *ex2)
2323 return (ex1->start <= ex2->end) && (ex2->start <= ex1->end);
2326 /* check if @ex1 contains @ex2 */
2327 static inline int ldlm_extent_contain(struct ldlm_extent *ex1,
2328 struct ldlm_extent *ex2)
2330 return (ex1->start <= ex2->start) && (ex1->end >= ex2->end);
2333 struct ldlm_inodebits {
2337 struct ldlm_flock_wire {
2345 /* it's important that the fields of the ldlm_extent structure match
2346 * the first fields of the ldlm_flock structure because there is only
2347 * one ldlm_swab routine to process the ldlm_policy_data_t union. if
2348 * this ever changes we will need to swab the union differently based
2349 * on the resource type. */
2352 struct ldlm_extent l_extent;
2353 struct ldlm_flock_wire l_flock;
2354 struct ldlm_inodebits l_inodebits;
2355 } ldlm_wire_policy_data_t;
2357 extern void lustre_swab_ldlm_policy_data (ldlm_wire_policy_data_t *d);
2359 /* Similarly to ldlm_wire_policy_data_t, there is one common swabber for all
2360 * LVB types. As a result, any new LVB structure must match the fields of the
2361 * ost_lvb structure. */
2362 union ldlm_wire_lvb {
2363 struct ost_lvb l_ost;
2366 extern void lustre_swab_lvb(union ldlm_wire_lvb *);
2368 struct ldlm_intent {
2372 extern void lustre_swab_ldlm_intent (struct ldlm_intent *i);
2374 struct ldlm_resource_desc {
2375 ldlm_type_t lr_type;
2376 __u32 lr_padding; /* also fix lustre_swab_ldlm_resource_desc */
2377 struct ldlm_res_id lr_name;
2380 extern void lustre_swab_ldlm_resource_desc (struct ldlm_resource_desc *r);
2382 struct ldlm_lock_desc {
2383 struct ldlm_resource_desc l_resource;
2384 ldlm_mode_t l_req_mode;
2385 ldlm_mode_t l_granted_mode;
2386 ldlm_wire_policy_data_t l_policy_data;
2389 extern void lustre_swab_ldlm_lock_desc (struct ldlm_lock_desc *l);
2391 #define LDLM_LOCKREQ_HANDLES 2
2392 #define LDLM_ENQUEUE_CANCEL_OFF 1
2394 struct ldlm_request {
2397 struct ldlm_lock_desc lock_desc;
2398 struct lustre_handle lock_handle[LDLM_LOCKREQ_HANDLES];
2401 extern void lustre_swab_ldlm_request (struct ldlm_request *rq);
2403 /* If LDLM_ENQUEUE, 1 slot is already occupied, 1 is available.
2404 * Otherwise, 2 are available. */
2405 #define ldlm_request_bufsize(count,type) \
2407 int _avail = LDLM_LOCKREQ_HANDLES; \
2408 _avail -= (type == LDLM_ENQUEUE ? LDLM_ENQUEUE_CANCEL_OFF : 0); \
2409 sizeof(struct ldlm_request) + \
2410 (count > _avail ? count - _avail : 0) * \
2411 sizeof(struct lustre_handle); \
2416 __u32 lock_padding; /* also fix lustre_swab_ldlm_reply */
2417 struct ldlm_lock_desc lock_desc;
2418 struct lustre_handle lock_handle;
2419 __u64 lock_policy_res1;
2420 __u64 lock_policy_res2;
2423 extern void lustre_swab_ldlm_reply (struct ldlm_reply *r);
2426 * Opcodes for mountconf (mgs and mgc)
2431 MGS_EXCEPTION, /* node died, etc. */
2432 MGS_TARGET_REG, /* whenever target starts up */
2438 #define MGS_FIRST_OPC MGS_CONNECT
2440 #define MGS_PARAM_MAXLEN 1024
2441 #define KEY_SET_INFO "set_info"
2443 struct mgs_send_param {
2444 char mgs_param[MGS_PARAM_MAXLEN];
2447 /* We pass this info to the MGS so it can write config logs */
2448 #define MTI_NAME_MAXLEN 64
2449 #define MTI_PARAM_MAXLEN 4096
2450 #define MTI_NIDS_MAX 32
2451 struct mgs_target_info {
2452 __u32 mti_lustre_ver;
2453 __u32 mti_stripe_index;
2454 __u32 mti_config_ver;
2456 __u32 mti_nid_count;
2457 __u32 mti_instance; /* Running instance of target */
2458 char mti_fsname[MTI_NAME_MAXLEN];
2459 char mti_svname[MTI_NAME_MAXLEN];
2460 char mti_uuid[sizeof(struct obd_uuid)];
2461 __u64 mti_nids[MTI_NIDS_MAX]; /* host nids (lnet_nid_t)*/
2462 char mti_params[MTI_PARAM_MAXLEN];
2464 extern void lustre_swab_mgs_target_info(struct mgs_target_info *oinfo);
2466 struct mgs_nidtbl_entry {
2467 __u64 mne_version; /* table version of this entry */
2468 __u32 mne_instance; /* target instance # */
2469 __u32 mne_index; /* target index */
2470 __u32 mne_length; /* length of this entry - by bytes */
2471 __u8 mne_type; /* target type LDD_F_SV_TYPE_OST/MDT */
2472 __u8 mne_nid_type; /* type of nid(mbz). for ipv6. */
2473 __u8 mne_nid_size; /* size of each NID, by bytes */
2474 __u8 mne_nid_count; /* # of NIDs in buffer */
2476 lnet_nid_t nids[0]; /* variable size buffer for NIDs. */
2479 extern void lustre_swab_mgs_nidtbl_entry(struct mgs_nidtbl_entry *oinfo);
2481 struct mgs_config_body {
2482 char mcb_name[MTI_NAME_MAXLEN]; /* logname */
2483 __u64 mcb_offset; /* next index of config log to request */
2484 __u16 mcb_type; /* type of log: CONFIG_T_[CONFIG|RECOVER] */
2486 __u8 mcb_bits; /* bits unit size of config log */
2487 __u32 mcb_units; /* # of units for bulk transfer */
2489 extern void lustre_swab_mgs_config_body(struct mgs_config_body *body);
2491 struct mgs_config_res {
2492 __u64 mcr_offset; /* index of last config log */
2493 __u64 mcr_size; /* size of the log */
2495 extern void lustre_swab_mgs_config_res(struct mgs_config_res *body);
2497 /* Config marker flags (in config log) */
2498 #define CM_START 0x01
2500 #define CM_SKIP 0x04
2501 #define CM_UPGRADE146 0x08
2502 #define CM_EXCLUDE 0x10
2503 #define CM_START_SKIP (CM_START | CM_SKIP)
2506 __u32 cm_step; /* aka config version */
2508 __u32 cm_vers; /* lustre release version number */
2509 __u32 cm_padding; /* 64 bit align */
2510 obd_time cm_createtime; /*when this record was first created */
2511 obd_time cm_canceltime; /*when this record is no longer valid*/
2512 char cm_tgtname[MTI_NAME_MAXLEN];
2513 char cm_comment[MTI_NAME_MAXLEN];
2516 extern void lustre_swab_cfg_marker(struct cfg_marker *marker,
2517 int swab, int size);
2520 * Opcodes for multiple servers.
2529 #define OBD_FIRST_OPC OBD_PING
2531 /* catalog of log objects */
2533 /** Identifier for a single log object */
2538 } __attribute__((packed));
2540 /** Records written to the CATALOGS list */
2541 #define CATLIST "CATALOGS"
2543 struct llog_logid lci_logid;
2547 } __attribute__((packed));
2549 /* Log data record types - there is no specific reason that these need to
2550 * be related to the RPC opcodes, but no reason not to (may be handy later?)
2552 #define LLOG_OP_MAGIC 0x10600000
2553 #define LLOG_OP_MASK 0xfff00000
2556 LLOG_PAD_MAGIC = LLOG_OP_MAGIC | 0x00000,
2557 OST_SZ_REC = LLOG_OP_MAGIC | 0x00f00,
2558 OST_RAID1_REC = LLOG_OP_MAGIC | 0x01000,
2559 MDS_UNLINK_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) | REINT_UNLINK,
2560 MDS_SETATTR_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) | REINT_SETATTR,
2561 MDS_SETATTR64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) | REINT_SETATTR,
2562 OBD_CFG_REC = LLOG_OP_MAGIC | 0x20000,
2563 PTL_CFG_REC = LLOG_OP_MAGIC | 0x30000, /* obsolete */
2564 LLOG_GEN_REC = LLOG_OP_MAGIC | 0x40000,
2565 LLOG_JOIN_REC = LLOG_OP_MAGIC | 0x50000, /* obsolete */
2566 CHANGELOG_REC = LLOG_OP_MAGIC | 0x60000,
2567 CHANGELOG_USER_REC = LLOG_OP_MAGIC | 0x70000,
2568 LLOG_HDR_MAGIC = LLOG_OP_MAGIC | 0x45539,
2569 LLOG_LOGID_MAGIC = LLOG_OP_MAGIC | 0x4553b,
2573 * for now, continue to support old pad records which have 0 for their
2574 * type but still need to be swabbed for their length
2576 #define LLOG_REC_HDR_NEEDS_SWABBING(r) \
2577 (((r)->lrh_type & __swab32(LLOG_OP_MASK)) == \
2578 __swab32(LLOG_OP_MAGIC) || \
2579 (((r)->lrh_type == 0) && ((r)->lrh_len > LLOG_CHUNK_SIZE)))
2581 /** Log record header - stored in little endian order.
2582 * Each record must start with this struct, end with a llog_rec_tail,
2583 * and be a multiple of 256 bits in size.
2585 struct llog_rec_hdr {
2592 struct llog_rec_tail {
2597 struct llog_logid_rec {
2598 struct llog_rec_hdr lid_hdr;
2599 struct llog_logid lid_id;
2605 struct llog_rec_tail lid_tail;
2606 } __attribute__((packed));
2608 struct llog_create_rec {
2609 struct llog_rec_hdr lcr_hdr;
2610 struct ll_fid lcr_fid;
2614 struct llog_rec_tail lcr_tail;
2615 } __attribute__((packed));
2617 struct llog_orphan_rec {
2618 struct llog_rec_hdr lor_hdr;
2622 struct llog_rec_tail lor_tail;
2623 } __attribute__((packed));
2625 struct llog_unlink_rec {
2626 struct llog_rec_hdr lur_hdr;
2629 obd_count lur_count;
2630 struct llog_rec_tail lur_tail;
2631 } __attribute__((packed));
2633 struct llog_setattr_rec {
2634 struct llog_rec_hdr lsr_hdr;
2640 struct llog_rec_tail lsr_tail;
2641 } __attribute__((packed));
2643 struct llog_setattr64_rec {
2644 struct llog_rec_hdr lsr_hdr;
2652 struct llog_rec_tail lsr_tail;
2653 } __attribute__((packed));
2655 struct llog_size_change_rec {
2656 struct llog_rec_hdr lsc_hdr;
2657 struct ll_fid lsc_fid;
2660 struct llog_rec_tail lsc_tail;
2661 } __attribute__((packed));
2663 #define CHANGELOG_MAGIC 0xca103000
2665 /** \a changelog_rec_type's that can't be masked */
2666 #define CHANGELOG_MINMASK (1 << CL_MARK)
2667 /** bits covering all \a changelog_rec_type's */
2668 #define CHANGELOG_ALLMASK 0XFFFFFFFF
2669 /** default \a changelog_rec_type mask */
2670 #define CHANGELOG_DEFMASK CHANGELOG_ALLMASK & ~(1 << CL_ATIME | 1 << CL_CLOSE)
2672 /* changelog llog name, needed by client replicators */
2673 #define CHANGELOG_CATALOG "changelog_catalog"
2675 struct changelog_setinfo {
2678 } __attribute__((packed));
2680 /** changelog record */
2681 struct llog_changelog_rec {
2682 struct llog_rec_hdr cr_hdr;
2683 struct changelog_rec cr;
2684 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
2685 } __attribute__((packed));
2687 struct llog_changelog_ext_rec {
2688 struct llog_rec_hdr cr_hdr;
2689 struct changelog_ext_rec cr;
2690 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
2691 } __attribute__((packed));
2693 #define CHANGELOG_USER_PREFIX "cl"
2695 struct llog_changelog_user_rec {
2696 struct llog_rec_hdr cur_hdr;
2700 struct llog_rec_tail cur_tail;
2701 } __attribute__((packed));
2706 } __attribute__((packed));
2708 struct llog_gen_rec {
2709 struct llog_rec_hdr lgr_hdr;
2710 struct llog_gen lgr_gen;
2711 struct llog_rec_tail lgr_tail;
2713 /* On-disk header structure of each log object, stored in little endian order */
2714 #define LLOG_CHUNK_SIZE 8192
2715 #define LLOG_HEADER_SIZE (96)
2716 #define LLOG_BITMAP_BYTES (LLOG_CHUNK_SIZE - LLOG_HEADER_SIZE)
2718 #define LLOG_MIN_REC_SIZE (24) /* round(llog_rec_hdr + llog_rec_tail) */
2720 /* flags for the logs */
2721 #define LLOG_F_ZAP_WHEN_EMPTY 0x1
2722 #define LLOG_F_IS_CAT 0x2
2723 #define LLOG_F_IS_PLAIN 0x4
2725 struct llog_log_hdr {
2726 struct llog_rec_hdr llh_hdr;
2727 obd_time llh_timestamp;
2729 __u32 llh_bitmap_offset;
2733 /* for a catalog the first plain slot is next to it */
2734 struct obd_uuid llh_tgtuuid;
2735 __u32 llh_reserved[LLOG_HEADER_SIZE/sizeof(__u32) - 23];
2736 __u32 llh_bitmap[LLOG_BITMAP_BYTES/sizeof(__u32)];
2737 struct llog_rec_tail llh_tail;
2738 } __attribute__((packed));
2740 #define LLOG_BITMAP_SIZE(llh) ((llh->llh_hdr.lrh_len - \
2741 llh->llh_bitmap_offset - \
2742 sizeof(llh->llh_tail)) * 8)
2744 /** log cookies are used to reference a specific log file and a record therein */
2745 struct llog_cookie {
2746 struct llog_logid lgc_lgl;
2750 } __attribute__((packed));
2752 /** llog protocol */
2753 enum llogd_rpc_ops {
2754 LLOG_ORIGIN_HANDLE_CREATE = 501,
2755 LLOG_ORIGIN_HANDLE_NEXT_BLOCK = 502,
2756 LLOG_ORIGIN_HANDLE_READ_HEADER = 503,
2757 LLOG_ORIGIN_HANDLE_WRITE_REC = 504,
2758 LLOG_ORIGIN_HANDLE_CLOSE = 505,
2759 LLOG_ORIGIN_CONNECT = 506,
2760 LLOG_CATINFO = 507, /* deprecated */
2761 LLOG_ORIGIN_HANDLE_PREV_BLOCK = 508,
2762 LLOG_ORIGIN_HANDLE_DESTROY = 509, /* for destroy llog object*/
2764 LLOG_FIRST_OPC = LLOG_ORIGIN_HANDLE_CREATE
2768 struct llog_logid lgd_logid;
2770 __u32 lgd_llh_flags;
2772 __u32 lgd_saved_index;
2774 __u64 lgd_cur_offset;
2775 } __attribute__((packed));
2777 struct llogd_conn_body {
2778 struct llog_gen lgdc_gen;
2779 struct llog_logid lgdc_logid;
2780 __u32 lgdc_ctxt_idx;
2781 } __attribute__((packed));
2783 /* Note: 64-bit types are 64-bit aligned in structure */
2785 obd_valid o_valid; /* hot fields in this obdo */
2787 obd_id o_parent_seq;
2788 obd_size o_size; /* o_size-o_blocks == ost_lvb */
2792 obd_blocks o_blocks; /* brw: cli sent cached bytes */
2795 /* 32-bit fields start here: keep an even number of them via padding */
2796 obd_blksize o_blksize; /* optimal IO blocksize */
2797 obd_mode o_mode; /* brw: cli sent cache remain */
2801 obd_count o_nlink; /* brw: checksum */
2802 obd_count o_parent_oid;
2803 obd_count o_misc; /* brw: o_dropped */
2804 __u64 o_ioepoch; /* epoch in ost writes */
2805 __u32 o_stripe_idx; /* holds stripe idx */
2807 struct lustre_handle o_handle; /* brw: lock handle to prolong
2809 struct llog_cookie o_lcookie; /* destroy: unlink cookie from
2814 __u64 o_data_version; /* getattr: sum of iversion for
2816 * brw: grant space consumed on
2817 * the client for the write */
2823 #define o_id o_oi.oi_id
2824 #define o_seq o_oi.oi_seq
2825 #define o_dirty o_blocks
2826 #define o_undirty o_mode
2827 #define o_dropped o_misc
2828 #define o_cksum o_nlink
2829 #define o_grant_used o_data_version
2831 static inline void lustre_set_wire_obdo(struct obdo *wobdo, struct obdo *lobdo)
2833 memcpy(wobdo, lobdo, sizeof(*lobdo));
2834 wobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
2837 static inline void lustre_get_wire_obdo(struct obdo *lobdo, struct obdo *wobdo)
2839 obd_flag local_flags = 0;
2841 if (lobdo->o_valid & OBD_MD_FLFLAGS)
2842 local_flags = lobdo->o_flags & OBD_FL_LOCAL_MASK;
2844 LASSERT(!(wobdo->o_flags & OBD_FL_LOCAL_MASK));
2846 memcpy(lobdo, wobdo, sizeof(*lobdo));
2847 if (local_flags != 0) {
2848 lobdo->o_valid |= OBD_MD_FLFLAGS;
2849 lobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
2850 lobdo->o_flags |= local_flags;
2854 extern void lustre_swab_obdo (struct obdo *o);
2856 /* request structure for OST's */
2861 /* Key for FIEMAP to be used in get_info calls */
2862 struct ll_fiemap_info_key {
2865 struct ll_user_fiemap fiemap;
2868 extern void lustre_swab_ost_body (struct ost_body *b);
2869 extern void lustre_swab_ost_last_id(obd_id *id);
2870 extern void lustre_swab_fiemap(struct ll_user_fiemap *fiemap);
2872 extern void lustre_swab_lov_user_md_v1(struct lov_user_md_v1 *lum);
2873 extern void lustre_swab_lov_user_md_v3(struct lov_user_md_v3 *lum);
2874 extern void lustre_swab_lov_user_md_objects(struct lov_user_ost_data *lod,
2876 extern void lustre_swab_lov_mds_md(struct lov_mds_md *lmm);
2879 extern void lustre_swab_llogd_body (struct llogd_body *d);
2880 extern void lustre_swab_llog_hdr (struct llog_log_hdr *h);
2881 extern void lustre_swab_llogd_conn_body (struct llogd_conn_body *d);
2882 extern void lustre_swab_llog_rec(struct llog_rec_hdr *rec,
2883 struct llog_rec_tail *tail);
2886 extern void lustre_swab_lustre_cfg(struct lustre_cfg *lcfg);
2888 /* Functions for dumping PTLRPC fields */
2889 void dump_rniobuf(struct niobuf_remote *rnb);
2890 void dump_ioo(struct obd_ioobj *nb);
2891 void dump_obdo(struct obdo *oa);
2892 void dump_ost_body(struct ost_body *ob);
2893 void dump_rcs(__u32 *rc);
2895 /* this will be used when OBD_CONNECT_CHANGE_QS is set */
2898 * ID appiles to (uid, gid)
2902 * LQUOTA_FLAGS_* affect the responding bits
2906 * acquire/release count (bytes for block quota)
2910 * when a master returns the reply to a slave, it will
2911 * contain the current corresponding qunit size
2917 #define QDATA_IS_GRP(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_GRP)
2918 #define QDATA_IS_BLK(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_BLK)
2919 #define QDATA_IS_ADJBLK(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_ADJBLK)
2920 #define QDATA_IS_ADJINO(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_ADJINO)
2921 #define QDATA_IS_CHANGE_QS(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_CHG_QS)
2923 #define QDATA_SET_GRP(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_GRP)
2924 #define QDATA_SET_BLK(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_BLK)
2925 #define QDATA_SET_ADJBLK(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_ADJBLK)
2926 #define QDATA_SET_ADJINO(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_ADJINO)
2927 #define QDATA_SET_CHANGE_QS(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_CHG_QS)
2929 #define QDATA_CLR_GRP(qdata) ((qdata)->qd_flags &= ~LQUOTA_FLAGS_GRP)
2930 #define QDATA_CLR_CHANGE_QS(qdata) ((qdata)->qd_flags &= ~LQUOTA_FLAGS_CHG_QS)
2932 extern void lustre_swab_qdata(struct qunit_data *d);
2933 extern struct qunit_data *quota_get_qdata(void *req, int is_req, int is_exp);
2934 extern int quota_copy_qdata(void *request, struct qunit_data *qdata,
2935 int is_req, int is_exp);
2942 #define QUOTA_FIRST_OPC QUOTA_DQACQ
2944 #define QUOTA_REQUEST 1
2945 #define QUOTA_REPLY 0
2946 #define QUOTA_EXPORT 1
2947 #define QUOTA_IMPORT 0
2949 /* quota check function */
2950 #define QUOTA_RET_OK 0 /**< return successfully */
2951 #define QUOTA_RET_NOQUOTA 1 /**< not support quota */
2952 #define QUOTA_RET_NOLIMIT 2 /**< quota limit isn't set */
2953 #define QUOTA_RET_ACQUOTA 4 /**< need to acquire extra quota */
2956 /* security opcodes */
2959 SEC_CTX_INIT_CONT = 802,
2962 SEC_FIRST_OPC = SEC_CTX_INIT
2966 * capa related definitions
2968 #define CAPA_HMAC_MAX_LEN 64
2969 #define CAPA_HMAC_KEY_MAX_LEN 56
2971 /* NB take care when changing the sequence of elements this struct,
2972 * because the offset info is used in find_capa() */
2973 struct lustre_capa {
2974 struct lu_fid lc_fid; /** fid */
2975 __u64 lc_opc; /** operations allowed */
2976 __u64 lc_uid; /** file owner */
2977 __u64 lc_gid; /** file group */
2978 __u32 lc_flags; /** HMAC algorithm & flags */
2979 __u32 lc_keyid; /** key# used for the capability */
2980 __u32 lc_timeout; /** capa timeout value (sec) */
2981 __u32 lc_expiry; /** expiry time (sec) */
2982 __u8 lc_hmac[CAPA_HMAC_MAX_LEN]; /** HMAC */
2983 } __attribute__((packed));
2985 extern void lustre_swab_lustre_capa(struct lustre_capa *c);
2987 /** lustre_capa::lc_opc */
2989 CAPA_OPC_BODY_WRITE = 1<<0, /**< write object data */
2990 CAPA_OPC_BODY_READ = 1<<1, /**< read object data */
2991 CAPA_OPC_INDEX_LOOKUP = 1<<2, /**< lookup object fid */
2992 CAPA_OPC_INDEX_INSERT = 1<<3, /**< insert object fid */
2993 CAPA_OPC_INDEX_DELETE = 1<<4, /**< delete object fid */
2994 CAPA_OPC_OSS_WRITE = 1<<5, /**< write oss object data */
2995 CAPA_OPC_OSS_READ = 1<<6, /**< read oss object data */
2996 CAPA_OPC_OSS_TRUNC = 1<<7, /**< truncate oss object */
2997 CAPA_OPC_OSS_DESTROY = 1<<8, /**< destroy oss object */
2998 CAPA_OPC_META_WRITE = 1<<9, /**< write object meta data */
2999 CAPA_OPC_META_READ = 1<<10, /**< read object meta data */
3002 #define CAPA_OPC_OSS_RW (CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE)
3003 #define CAPA_OPC_MDS_ONLY \
3004 (CAPA_OPC_BODY_WRITE | CAPA_OPC_BODY_READ | CAPA_OPC_INDEX_LOOKUP | \
3005 CAPA_OPC_INDEX_INSERT | CAPA_OPC_INDEX_DELETE)
3006 #define CAPA_OPC_OSS_ONLY \
3007 (CAPA_OPC_OSS_WRITE | CAPA_OPC_OSS_READ | CAPA_OPC_OSS_TRUNC | \
3008 CAPA_OPC_OSS_DESTROY)
3009 #define CAPA_OPC_MDS_DEFAULT ~CAPA_OPC_OSS_ONLY
3010 #define CAPA_OPC_OSS_DEFAULT ~(CAPA_OPC_MDS_ONLY | CAPA_OPC_OSS_ONLY)
3012 /* MDS capability covers object capability for operations of body r/w
3013 * (dir readpage/sendpage), index lookup/insert/delete and meta data r/w,
3014 * while OSS capability only covers object capability for operations of
3015 * oss data(file content) r/w/truncate.
3017 static inline int capa_for_mds(struct lustre_capa *c)
3019 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) != 0;
3022 static inline int capa_for_oss(struct lustre_capa *c)
3024 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) == 0;
3027 /* lustre_capa::lc_hmac_alg */
3029 CAPA_HMAC_ALG_SHA1 = 1, /**< sha1 algorithm */
3033 #define CAPA_FL_MASK 0x00ffffff
3034 #define CAPA_HMAC_ALG_MASK 0xff000000
3036 struct lustre_capa_key {
3037 __u64 lk_seq; /**< mds# */
3038 __u32 lk_keyid; /**< key# */
3040 __u8 lk_key[CAPA_HMAC_KEY_MAX_LEN]; /**< key */
3041 } __attribute__((packed));
3043 extern void lustre_swab_lustre_capa_key(struct lustre_capa_key *k);
3045 /** The link ea holds 1 \a link_ea_entry for each hardlink */
3046 #define LINK_EA_MAGIC 0x11EAF1DFUL
3047 struct link_ea_header {
3050 __u64 leh_len; /* total size */
3056 /** Hardlink data is name and parent fid.
3057 * Stored in this crazy struct for maximum packing and endian-neutrality
3059 struct link_ea_entry {
3060 /** __u16 stored big-endian, unaligned */
3061 unsigned char lee_reclen[2];
3062 unsigned char lee_parent_fid[sizeof(struct lu_fid)];
3064 }__attribute__((packed));
3066 /** fid2path request/reply structure */
3067 struct getinfo_fid2path {
3068 struct lu_fid gf_fid;
3073 } __attribute__((packed));
3075 void lustre_swab_fid2path (struct getinfo_fid2path *gf);