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 (c) 2007, 2010, Oracle and/or its affiliates. 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> /* for LASSERT, LPUX64, etc */
96 /* Defn's shared with user-space. */
97 #include <lustre/lustre_user.h>
102 /* FOO_REQUEST_PORTAL is for incoming requests on the FOO
103 * FOO_REPLY_PORTAL is for incoming replies on the FOO
104 * FOO_BULK_PORTAL is for incoming bulk on the FOO
107 #define CONNMGR_REQUEST_PORTAL 1
108 #define CONNMGR_REPLY_PORTAL 2
109 //#define OSC_REQUEST_PORTAL 3
110 #define OSC_REPLY_PORTAL 4
111 //#define OSC_BULK_PORTAL 5
112 #define OST_IO_PORTAL 6
113 #define OST_CREATE_PORTAL 7
114 #define OST_BULK_PORTAL 8
115 //#define MDC_REQUEST_PORTAL 9
116 #define MDC_REPLY_PORTAL 10
117 //#define MDC_BULK_PORTAL 11
118 #define MDS_REQUEST_PORTAL 12
119 //#define MDS_REPLY_PORTAL 13
120 #define MDS_BULK_PORTAL 14
121 #define LDLM_CB_REQUEST_PORTAL 15
122 #define LDLM_CB_REPLY_PORTAL 16
123 #define LDLM_CANCEL_REQUEST_PORTAL 17
124 #define LDLM_CANCEL_REPLY_PORTAL 18
125 //#define PTLBD_REQUEST_PORTAL 19
126 //#define PTLBD_REPLY_PORTAL 20
127 //#define PTLBD_BULK_PORTAL 21
128 #define MDS_SETATTR_PORTAL 22
129 #define MDS_READPAGE_PORTAL 23
130 #define MDS_MDS_PORTAL 24
132 #define MGC_REPLY_PORTAL 25
133 #define MGS_REQUEST_PORTAL 26
134 #define MGS_REPLY_PORTAL 27
135 #define OST_REQUEST_PORTAL 28
136 #define FLD_REQUEST_PORTAL 29
137 #define SEQ_METADATA_PORTAL 30
138 #define SEQ_DATA_PORTAL 31
139 #define SEQ_CONTROLLER_PORTAL 32
141 /* Portal 63 is reserved for the Cray Inc DVS - nic@cray.com, roe@cray.com, n8851@cray.com */
144 #define PTL_RPC_MSG_REQUEST 4711
145 #define PTL_RPC_MSG_ERR 4712
146 #define PTL_RPC_MSG_REPLY 4713
148 /* DON'T use swabbed values of MAGIC as magic! */
149 #define LUSTRE_MSG_MAGIC_V1 0x0BD00BD0
150 #define LUSTRE_MSG_MAGIC_V2 0x0BD00BD3
152 #define LUSTRE_MSG_MAGIC_V1_SWABBED 0xD00BD00B
153 #define LUSTRE_MSG_MAGIC_V2_SWABBED 0xD30BD00B
155 #define LUSTRE_MSG_MAGIC LUSTRE_MSG_MAGIC_V2
157 #define PTLRPC_MSG_VERSION 0x00000003
158 #define LUSTRE_VERSION_MASK 0xffff0000
159 #define LUSTRE_OBD_VERSION 0x00010000
160 #define LUSTRE_MDS_VERSION 0x00020000
161 #define LUSTRE_OST_VERSION 0x00030000
162 #define LUSTRE_DLM_VERSION 0x00040000
163 #define LUSTRE_LOG_VERSION 0x00050000
164 #define LUSTRE_MGS_VERSION 0x00060000
166 typedef __u32 mdsno_t;
167 typedef __u64 seqno_t;
168 typedef __u64 obd_id;
169 typedef __u64 obd_seq;
170 typedef __s64 obd_time;
171 typedef __u64 obd_size;
172 typedef __u64 obd_off;
173 typedef __u64 obd_blocks;
174 typedef __u64 obd_valid;
175 typedef __u32 obd_blksize;
176 typedef __u32 obd_mode;
177 typedef __u32 obd_uid;
178 typedef __u32 obd_gid;
179 typedef __u32 obd_flag;
180 typedef __u32 obd_count;
183 * Describes a range of sequence, lsr_start is included but lsr_end is
185 * Same structure is used in fld module where lsr_index field holds mdt id
189 #define LU_SEQ_RANGE_MDT 0x0
190 #define LU_SEQ_RANGE_OST 0x1
192 struct lu_seq_range {
200 * returns width of given range \a r
203 static inline __u64 range_space(const struct lu_seq_range *range)
205 return range->lsr_end - range->lsr_start;
209 * initialize range to zero
212 static inline void range_init(struct lu_seq_range *range)
214 range->lsr_start = range->lsr_end = range->lsr_index = 0;
218 * check if given seq id \a s is within given range \a r
221 static inline int range_within(const struct lu_seq_range *range,
224 return s >= range->lsr_start && s < range->lsr_end;
227 static inline int range_is_sane(const struct lu_seq_range *range)
229 return (range->lsr_end >= range->lsr_start);
232 static inline int range_is_zero(const struct lu_seq_range *range)
234 return (range->lsr_start == 0 && range->lsr_end == 0);
237 static inline int range_is_exhausted(const struct lu_seq_range *range)
240 return range_space(range) == 0;
243 /* return 0 if two range have the same location */
244 static inline int range_compare_loc(const struct lu_seq_range *r1,
245 const struct lu_seq_range *r2)
247 return r1->lsr_index != r2->lsr_index ||
248 r1->lsr_flags != r2->lsr_flags;
251 #define DRANGE "[%#16.16"LPF64"x-%#16.16"LPF64"x):%x:%x"
253 #define PRANGE(range) \
254 (range)->lsr_start, \
256 (range)->lsr_index, \
259 /** \defgroup lu_fid lu_fid
263 * Flags for lustre_mdt_attrs::lma_compat and lustre_mdt_attrs::lma_incompat.
266 LMAC_HSM = 0x00000001,
267 LMAC_SOM = 0x00000002,
271 * Masks for all features that should be supported by a Lustre version to
272 * access a specific file.
273 * This information is stored in lustre_mdt_attrs::lma_incompat.
275 * NOTE: No incompat feature should be added before bug #17670 is landed.
277 #define LMA_INCOMPAT_SUPP 0x0
280 * Following struct for MDT attributes, that will be kept inode's EA.
281 * Introduced in 2.0 release (please see b15993, for details)
283 struct lustre_mdt_attrs {
285 * Bitfield for supported data in this structure. From enum lma_compat.
286 * lma_self_fid and lma_flags are always available.
290 * Per-file incompat feature list. Lustre version should support all
291 * flags set in this field. The supported feature mask is available in
295 /** FID of this inode */
296 struct lu_fid lma_self_fid;
297 /** mdt/ost type, others */
299 /* IO Epoch SOM attributes belongs to */
301 /** total file size in objects */
303 /** total fs blocks in objects */
304 __u64 lma_som_blocks;
305 /** mds mount id the size is valid for */
306 __u64 lma_som_mountid;
310 * Fill \a lma with its first content.
311 * Only fid is stored.
313 static inline void lustre_lma_init(struct lustre_mdt_attrs *lma,
314 const struct lu_fid *fid)
317 lma->lma_incompat = 0;
318 memcpy(&lma->lma_self_fid, fid, sizeof(*fid));
320 lma->lma_ioepoch = 0;
321 lma->lma_som_size = 0;
322 lma->lma_som_blocks = 0;
323 lma->lma_som_mountid = 0;
325 /* If a field is added in struct lustre_mdt_attrs, zero it explicitly
326 * and change the test below. */
327 LASSERT(sizeof(*lma) ==
328 (offsetof(struct lustre_mdt_attrs, lma_som_mountid) +
329 sizeof(lma->lma_som_mountid)));
332 extern void lustre_swab_lu_fid(struct lu_fid *fid);
335 * Swab, if needed, lustre_mdt_attr struct to on-disk format.
336 * Otherwise, do not touch it.
338 static inline void lustre_lma_swab(struct lustre_mdt_attrs *lma)
340 /* Use LUSTRE_MSG_MAGIC to detect local endianess. */
341 if (LUSTRE_MSG_MAGIC != cpu_to_le32(LUSTRE_MSG_MAGIC)) {
342 __swab32s(&lma->lma_compat);
343 __swab32s(&lma->lma_incompat);
344 lustre_swab_lu_fid(&lma->lma_self_fid);
345 __swab64s(&lma->lma_flags);
346 __swab64s(&lma->lma_ioepoch);
347 __swab64s(&lma->lma_som_size);
348 __swab64s(&lma->lma_som_blocks);
349 __swab64s(&lma->lma_som_mountid);
353 /* This is the maximum number of MDTs allowed in CMD testing until such
354 * a time that FID-on-OST is implemented. This is due to the limitations
355 * of packing non-0-MDT numbers into the FID SEQ namespace. Once FID-on-OST
356 * is implemented this limit will be virtually unlimited. */
357 #define MAX_MDT_COUNT 8
364 /** initial fid id value */
365 LUSTRE_FID_INIT_OID = 1UL
368 /** returns fid object sequence */
369 static inline __u64 fid_seq(const struct lu_fid *fid)
374 /** returns fid object id */
375 static inline __u32 fid_oid(const struct lu_fid *fid)
380 /** returns fid object version */
381 static inline __u32 fid_ver(const struct lu_fid *fid)
386 static inline void fid_zero(struct lu_fid *fid)
388 memset(fid, 0, sizeof(*fid));
391 static inline obd_id fid_ver_oid(const struct lu_fid *fid)
393 return ((__u64)fid_ver(fid) << 32 | fid_oid(fid));
397 * Note that reserved SEQ numbers below 12 will conflict with ldiskfs
398 * inodes in the IGIF namespace, so these reserved SEQ numbers can be
399 * used for other purposes and not risk collisions with existing inodes.
401 * Different FID Format
402 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs#NEW.0
405 FID_SEQ_OST_MDT0 = 0,
408 FID_SEQ_OST_MDT1 = 3,
409 FID_SEQ_OST_MAX = 9, /* Max MDT count before OST_on_FID */
412 FID_SEQ_IGIF_MAX = 0x0ffffffffULL,
413 FID_SEQ_IDIF = 0x100000000ULL,
414 FID_SEQ_IDIF_MAX = 0x1ffffffffULL,
415 /* Normal FID sequence starts from this value, i.e. 1<<33 */
416 FID_SEQ_START = 0x200000000ULL,
417 FID_SEQ_LOCAL_FILE = 0x200000001ULL,
418 FID_SEQ_DOT_LUSTRE = 0x200000002ULL,
419 FID_SEQ_NORMAL = 0x200000400ULL
422 #define OBIF_OID_MAX_BITS 32
423 #define OBIF_MAX_OID (1ULL << OBIF_OID_MAX_BITS)
424 #define OBIF_OID_MASK ((1ULL << OBIF_OID_MAX_BITS) - 1)
425 #define IDIF_OID_MAX_BITS 48
426 #define IDIF_MAX_OID (1ULL << IDIF_OID_MAX_BITS)
427 #define IDIF_OID_MASK ((1ULL << IDIF_OID_MAX_BITS) - 1)
430 static inline int fid_seq_is_mdt0(obd_seq seq)
432 return (seq == FID_SEQ_OST_MDT0);
435 static inline int fid_seq_is_cmd(const __u64 seq)
437 return (seq >= FID_SEQ_OST_MDT1 && seq <= FID_SEQ_OST_MAX);
440 static inline int fid_seq_is_mdt(const __u64 seq)
442 return seq == FID_SEQ_OST_MDT0 ||
443 (seq >= FID_SEQ_OST_MDT1 && seq <= FID_SEQ_OST_MAX);
446 static inline int fid_seq_is_rsvd(const __u64 seq)
448 return seq <= FID_SEQ_RSVD;
451 static inline int fid_is_mdt0(const struct lu_fid *fid)
453 return fid_seq_is_mdt0(fid_seq(fid));
457 * Check if a fid is igif or not.
458 * \param fid the fid to be tested.
459 * \return true if the fid is a igif; otherwise false.
461 static inline int fid_seq_is_igif(const __u64 seq)
463 return seq >= FID_SEQ_IGIF && seq <= FID_SEQ_IGIF_MAX;
466 static inline int fid_is_igif(const struct lu_fid *fid)
468 return fid_seq_is_igif(fid_seq(fid));
472 * Check if a fid is idif or not.
473 * \param fid the fid to be tested.
474 * \return true if the fid is a idif; otherwise false.
476 static inline int fid_seq_is_idif(const __u64 seq)
478 return seq >= FID_SEQ_IDIF && seq <= FID_SEQ_IDIF_MAX;
481 static inline int fid_is_idif(const struct lu_fid *fid)
483 return fid_seq_is_idif(fid_seq(fid));
491 static inline int fid_seq_is_norm(const __u64 seq)
493 return (seq >= FID_SEQ_NORMAL);
496 static inline int fid_is_norm(const struct lu_fid *fid)
498 return fid_seq_is_norm(fid_seq(fid));
501 /* convert an OST objid into an IDIF FID SEQ number */
502 static inline obd_seq fid_idif_seq(obd_id id, __u32 ost_idx)
504 return FID_SEQ_IDIF | (ost_idx << 16) | ((id >> 32) & 0xffff);
507 /* convert a packed IDIF FID into an OST objid */
508 static inline obd_id fid_idif_id(obd_seq seq, __u32 oid, __u32 ver)
510 return ((__u64)ver << 48) | ((seq & 0xffff) << 32) | oid;
513 /* unpack an ostid (id/seq) from a wire/disk structure into an IDIF FID */
514 static inline void ostid_idif_unpack(struct ost_id *ostid,
515 struct lu_fid *fid, __u32 ost_idx)
517 fid->f_seq = fid_idif_seq(ostid->oi_id, ost_idx);
518 fid->f_oid = ostid->oi_id; /* truncate to 32 bits by assignment */
519 fid->f_ver = ostid->oi_id >> 48; /* in theory, not currently used */
522 /* unpack an ostid (id/seq) from a wire/disk structure into a non-IDIF FID */
523 static inline void ostid_fid_unpack(struct ost_id *ostid, struct lu_fid *fid)
525 fid->f_seq = ostid->oi_seq;
526 fid->f_oid = ostid->oi_id; /* truncate to 32 bits by assignment */
527 fid->f_ver = ostid->oi_id >> 32; /* in theory, not currently used */
530 /* Unpack an OST object id/seq (group) into a FID. This is needed for
531 * converting all obdo, lmm, lsm, etc. 64-bit id/seq pairs into proper
532 * FIDs. Note that if an id/seq is already in FID/IDIF format it will
533 * be passed through unchanged. Only legacy OST objects in "group 0"
534 * will be mapped into the IDIF namespace so that they can fit into the
535 * struct lu_fid fields without loss. For reference see:
536 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs
538 static inline int fid_ostid_unpack(struct lu_fid *fid, struct ost_id *ostid,
541 if (ost_idx > 0xffff) {
542 CERROR("bad ost_idx, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
543 ostid->oi_seq, ostid->oi_id, ost_idx);
547 if (fid_seq_is_mdt0(ostid->oi_seq)) {
548 /* This is a "legacy" (old 1.x/2.early) OST object in "group 0"
549 * that we map into the IDIF namespace. It allows up to 2^48
550 * objects per OST, as this is the object namespace that has
551 * been in production for years. This can handle create rates
552 * of 1M objects/s/OST for 9 years, or combinations thereof. */
553 if (ostid->oi_id >= IDIF_MAX_OID) {
554 CERROR("bad MDT0 id, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
555 ostid->oi_seq, ostid->oi_id, ost_idx);
558 ostid_idif_unpack(ostid, fid, ost_idx);
560 } else if (fid_seq_is_rsvd(ostid->oi_seq)) {
561 /* These are legacy OST objects for LLOG/ECHO and CMD testing.
562 * We only support 2^32 objects in these groups, and cannot
563 * uniquely identify them in the system (i.e. they are the
564 * duplicated on all OSTs), but this is not strictly required
565 * for the old object protocol, which has a separate ost_idx. */
566 if (ostid->oi_id >= 0xffffffffULL) {
567 CERROR("bad RSVD id, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
568 ostid->oi_seq, ostid->oi_id, ost_idx);
571 ostid_fid_unpack(ostid, fid);
573 } else if (unlikely(fid_seq_is_igif(ostid->oi_seq))) {
574 /* This is an MDT inode number, which should never collide with
575 * proper OST object IDs, and is probably a broken filesystem */
576 CERROR("bad IGIF, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
577 ostid->oi_seq, ostid->oi_id, ost_idx);
580 } else /* if (fid_seq_is_idif(seq) || fid_seq_is_norm(seq)) */ {
581 /* This is either an IDIF object, which identifies objects across
582 * all OSTs, or a regular FID. The IDIF namespace maps legacy
583 * OST objects into the FID namespace. In both cases, we just
584 * pass the FID through, no conversion needed. */
585 ostid_fid_unpack(ostid, fid);
591 /* pack an IDIF FID into an ostid (id/seq) for the wire/disk */
592 static inline void ostid_idif_pack(struct lu_fid *fid, struct ost_id *ostid)
594 ostid->oi_seq = FID_SEQ_OST_MDT0;
595 ostid->oi_id = fid_idif_id(fid->f_seq, fid->f_oid, fid->f_ver);
598 /* pack a non-IDIF FID into an ostid (id/seq) for the wire/disk */
599 static inline void ostid_fid_pack(struct lu_fid *fid, struct ost_id *ostid)
601 ostid->oi_seq = fid_seq(fid);
602 ostid->oi_id = fid_ver_oid(fid);
605 /* pack any OST FID into an ostid (id/seq) for the wire/disk */
606 static inline int fid_ostid_pack(struct lu_fid *fid, struct ost_id *ostid)
608 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
609 CERROR("bad IGIF, "DFID"\n", PFID(fid));
613 if (fid_is_idif(fid))
614 ostid_idif_pack(fid, ostid);
616 ostid_fid_pack(fid, ostid);
621 /* extract OST sequence (group) from a wire ost_id (id/seq) pair */
622 static inline obd_seq ostid_seq(struct ost_id *ostid)
624 if (unlikely(fid_seq_is_igif(ostid->oi_seq)))
625 CWARN("bad IGIF, oi_seq: "LPU64" oi_id: "LPX64"\n",
626 ostid->oi_seq, ostid->oi_id);
628 if (unlikely(fid_seq_is_idif(ostid->oi_seq)))
629 return FID_SEQ_OST_MDT0;
631 return ostid->oi_seq;
634 /* extract OST objid from a wire ost_id (id/seq) pair */
635 static inline obd_id ostid_id(struct ost_id *ostid)
637 if (ostid->oi_seq == FID_SEQ_OST_MDT0)
638 return ostid->oi_id & IDIF_OID_MASK;
640 if (fid_seq_is_rsvd(ostid->oi_seq))
641 return ostid->oi_id & OBIF_OID_MASK;
643 if (fid_seq_is_idif(ostid->oi_seq))
644 return fid_idif_id(ostid->oi_seq, ostid->oi_id, 0);
650 * Get inode number from a igif.
651 * \param fid a igif to get inode number from.
652 * \return inode number for the igif.
654 static inline ino_t lu_igif_ino(const struct lu_fid *fid)
660 * Build igif from the inode number/generation.
662 #define LU_IGIF_BUILD(fid, ino, gen) \
668 static inline void lu_igif_build(struct lu_fid *fid, __u32 ino, __u32 gen)
670 LU_IGIF_BUILD(fid, ino, gen);
671 LASSERT(fid_is_igif(fid));
675 * Get inode generation from a igif.
676 * \param fid a igif to get inode generation from.
677 * \return inode generation for the igif.
679 static inline __u32 lu_igif_gen(const struct lu_fid *fid)
685 * Fids are transmitted across network (in the sender byte-ordering),
686 * and stored on disk in big-endian order.
688 static inline void fid_cpu_to_le(struct lu_fid *dst, const struct lu_fid *src)
690 /* check that all fields are converted */
691 CLASSERT(sizeof *src ==
692 sizeof fid_seq(src) +
693 sizeof fid_oid(src) + sizeof fid_ver(src));
694 LASSERTF(fid_is_igif(src) || fid_ver(src) == 0, DFID"\n", PFID(src));
695 dst->f_seq = cpu_to_le64(fid_seq(src));
696 dst->f_oid = cpu_to_le32(fid_oid(src));
697 dst->f_ver = cpu_to_le32(fid_ver(src));
700 static inline void fid_le_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
702 /* check that all fields are converted */
703 CLASSERT(sizeof *src ==
704 sizeof fid_seq(src) +
705 sizeof fid_oid(src) + sizeof fid_ver(src));
706 dst->f_seq = le64_to_cpu(fid_seq(src));
707 dst->f_oid = le32_to_cpu(fid_oid(src));
708 dst->f_ver = le32_to_cpu(fid_ver(src));
709 LASSERTF(fid_is_igif(dst) || fid_ver(dst) == 0, DFID"\n", PFID(dst));
712 static inline void fid_cpu_to_be(struct lu_fid *dst, const struct lu_fid *src)
714 /* check that all fields are converted */
715 CLASSERT(sizeof *src ==
716 sizeof fid_seq(src) +
717 sizeof fid_oid(src) + sizeof fid_ver(src));
718 LASSERTF(fid_is_igif(src) || fid_ver(src) == 0, DFID"\n", PFID(src));
719 dst->f_seq = cpu_to_be64(fid_seq(src));
720 dst->f_oid = cpu_to_be32(fid_oid(src));
721 dst->f_ver = cpu_to_be32(fid_ver(src));
724 static inline void fid_be_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
726 /* check that all fields are converted */
727 CLASSERT(sizeof *src ==
728 sizeof fid_seq(src) +
729 sizeof fid_oid(src) + sizeof fid_ver(src));
730 dst->f_seq = be64_to_cpu(fid_seq(src));
731 dst->f_oid = be32_to_cpu(fid_oid(src));
732 dst->f_ver = be32_to_cpu(fid_ver(src));
733 LASSERTF(fid_is_igif(dst) || fid_ver(dst) == 0, DFID"\n", PFID(dst));
736 static inline int fid_is_sane(const struct lu_fid *fid)
740 ((fid_seq(fid) >= FID_SEQ_START && fid_oid(fid) != 0
741 && fid_ver(fid) == 0) ||
745 static inline int fid_is_zero(const struct lu_fid *fid)
747 return fid_seq(fid) == 0 && fid_oid(fid) == 0;
750 extern void lustre_swab_lu_fid(struct lu_fid *fid);
751 extern void lustre_swab_lu_seq_range(struct lu_seq_range *range);
753 static inline int lu_fid_eq(const struct lu_fid *f0,
754 const struct lu_fid *f1)
756 /* Check that there is no alignment padding. */
757 CLASSERT(sizeof *f0 ==
758 sizeof f0->f_seq + sizeof f0->f_oid + sizeof f0->f_ver);
759 LASSERTF(fid_is_igif(f0) || fid_ver(f0) == 0, DFID, PFID(f0));
760 LASSERTF(fid_is_igif(f1) || fid_ver(f1) == 0, DFID, PFID(f1));
761 return memcmp(f0, f1, sizeof *f0) == 0;
764 #define __diff_normalize(val0, val1) \
766 typeof(val0) __val0 = (val0); \
767 typeof(val1) __val1 = (val1); \
769 (__val0 == __val1 ? 0 : __val0 > __val1 ? +1 : -1); \
772 static inline int lu_fid_cmp(const struct lu_fid *f0,
773 const struct lu_fid *f1)
776 __diff_normalize(fid_seq(f0), fid_seq(f1)) ?:
777 __diff_normalize(fid_oid(f0), fid_oid(f1)) ?:
778 __diff_normalize(fid_ver(f0), fid_ver(f1));
783 /** \defgroup lu_dir lu_dir
787 * Enumeration of possible directory entry attributes.
789 * Attributes follow directory entry header in the order they appear in this
792 enum lu_dirent_attrs {
798 * Layout of readdir pages, as transmitted on wire.
801 /** valid if LUDA_FID is set. */
802 struct lu_fid lde_fid;
803 /** a unique entry identifier: a hash or an offset. */
805 /** total record length, including all attributes. */
809 /** optional variable size attributes following this entry.
810 * taken from enum lu_dirent_attrs.
813 /** name is followed by the attributes indicated in ->ldp_attrs, in
814 * their natural order. After the last attribute, padding bytes are
815 * added to make ->lde_reclen a multiple of 8.
821 * Definitions of optional directory entry attributes formats.
823 * Individual attributes do not have their length encoded in a generic way. It
824 * is assumed that consumer of an attribute knows its format. This means that
825 * it is impossible to skip over an unknown attribute, except by skipping over all
826 * remaining attributes (by using ->lde_reclen), which is not too
827 * constraining, because new server versions will append new attributes at
828 * the end of an entry.
832 * Fid directory attribute: a fid of an object referenced by the entry. This
833 * will be almost always requested by the client and supplied by the server.
835 * Aligned to 8 bytes.
837 /* To have compatibility with 1.8, lets have fid in lu_dirent struct. */
842 * Aligned to 2 bytes.
849 __u64 ldp_hash_start;
853 struct lu_dirent ldp_entries[0];
856 enum lu_dirpage_flags {
860 static inline struct lu_dirent *lu_dirent_start(struct lu_dirpage *dp)
862 if (le32_to_cpu(dp->ldp_flags) & LDF_EMPTY)
865 return dp->ldp_entries;
868 static inline struct lu_dirent *lu_dirent_next(struct lu_dirent *ent)
870 struct lu_dirent *next;
872 if (le16_to_cpu(ent->lde_reclen) != 0)
873 next = ((void *)ent) + le16_to_cpu(ent->lde_reclen);
880 static inline int lu_dirent_calc_size(int namelen, __u16 attr)
884 if (attr & LUDA_TYPE) {
885 const unsigned align = sizeof(struct luda_type) - 1;
886 size = (sizeof(struct lu_dirent) + namelen + align) & ~align;
887 size += sizeof(struct luda_type);
889 size = sizeof(struct lu_dirent) + namelen;
891 return (size + 7) & ~7;
894 static inline int lu_dirent_size(struct lu_dirent *ent)
896 if (le16_to_cpu(ent->lde_reclen) == 0) {
897 return lu_dirent_calc_size(le16_to_cpu(ent->lde_namelen),
898 le32_to_cpu(ent->lde_attrs));
900 return le16_to_cpu(ent->lde_reclen);
903 #define DIR_END_OFF 0xfffffffffffffffeULL
907 struct lustre_handle {
910 #define DEAD_HANDLE_MAGIC 0xdeadbeefcafebabeULL
912 static inline int lustre_handle_is_used(struct lustre_handle *lh)
914 return lh->cookie != 0ull;
917 static inline int lustre_handle_equal(const struct lustre_handle *lh1,
918 const struct lustre_handle *lh2)
920 return lh1->cookie == lh2->cookie;
923 static inline void lustre_handle_copy(struct lustre_handle *tgt,
924 struct lustre_handle *src)
926 tgt->cookie = src->cookie;
929 /* flags for lm_flags */
930 #define MSGHDR_AT_SUPPORT 0x1
931 #define MSGHDR_CKSUM_INCOMPAT18 0x2
933 #define lustre_msg lustre_msg_v2
934 /* we depend on this structure to be 8-byte aligned */
935 /* this type is only endian-adjusted in lustre_unpack_msg() */
936 struct lustre_msg_v2 {
948 /* without gss, ptlrpc_body is put at the first buffer. */
949 #define PTLRPC_NUM_VERSIONS 4
951 struct lustre_handle pb_handle;
958 __u64 pb_last_committed;
963 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
964 __u32 pb_service_time; /* for rep, actual service time */
967 /* VBR: pre-versions */
968 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
969 /* padding for future needs */
973 extern void lustre_swab_ptlrpc_body(struct ptlrpc_body *pb);
975 /* message body offset for lustre_msg_v2 */
976 /* ptlrpc body offset in all request/reply messages */
977 #define MSG_PTLRPC_BODY_OFF 0
979 /* normal request/reply message record offset */
980 #define REQ_REC_OFF 1
981 #define REPLY_REC_OFF 1
983 /* ldlm request message body offset */
984 #define DLM_LOCKREQ_OFF 1 /* lockreq offset */
985 #define DLM_REQ_REC_OFF 2 /* normal dlm request record offset */
987 /* ldlm intent lock message body offset */
988 #define DLM_INTENT_IT_OFF 2 /* intent lock it offset */
989 #define DLM_INTENT_REC_OFF 3 /* intent lock record offset */
991 /* ldlm reply message body offset */
992 #define DLM_LOCKREPLY_OFF 1 /* lockrep offset */
993 #define DLM_REPLY_REC_OFF 2 /* reply record offset */
995 /** only use in req->rq_{req,rep}_swab_mask */
996 #define MSG_PTLRPC_HEADER_OFF 31
998 /* Flags that are operation-specific go in the top 16 bits. */
999 #define MSG_OP_FLAG_MASK 0xffff0000
1000 #define MSG_OP_FLAG_SHIFT 16
1002 /* Flags that apply to all requests are in the bottom 16 bits */
1003 #define MSG_GEN_FLAG_MASK 0x0000ffff
1004 #define MSG_LAST_REPLAY 0x0001
1005 #define MSG_RESENT 0x0002
1006 #define MSG_REPLAY 0x0004
1007 /* #define MSG_AT_SUPPORT 0x0008
1008 * This was used in early prototypes of adaptive timeouts, and while there
1009 * shouldn't be any users of that code there also isn't a need for using this
1010 * bits. Defer usage until at least 1.10 to avoid potential conflict. */
1011 #define MSG_DELAY_REPLAY 0x0010
1012 #define MSG_VERSION_REPLAY 0x0020
1013 #define MSG_REQ_REPLAY_DONE 0x0040
1014 #define MSG_LOCK_REPLAY_DONE 0x0080
1017 * Flags for all connect opcodes (MDS_CONNECT, OST_CONNECT)
1020 #define MSG_CONNECT_RECOVERING 0x00000001
1021 #define MSG_CONNECT_RECONNECT 0x00000002
1022 #define MSG_CONNECT_REPLAYABLE 0x00000004
1023 //#define MSG_CONNECT_PEER 0x8
1024 #define MSG_CONNECT_LIBCLIENT 0x00000010
1025 #define MSG_CONNECT_INITIAL 0x00000020
1026 #define MSG_CONNECT_ASYNC 0x00000040
1027 #define MSG_CONNECT_NEXT_VER 0x00000080 /* use next version of lustre_msg */
1028 #define MSG_CONNECT_TRANSNO 0x00000100 /* report transno */
1031 #define OBD_CONNECT_RDONLY 0x1ULL /*client has read-only access*/
1032 #define OBD_CONNECT_INDEX 0x2ULL /*connect specific LOV idx */
1033 #define OBD_CONNECT_MDS 0x4ULL /*connect from MDT to OST */
1034 #define OBD_CONNECT_GRANT 0x8ULL /*OSC gets grant at connect */
1035 #define OBD_CONNECT_SRVLOCK 0x10ULL /*server takes locks for cli */
1036 #define OBD_CONNECT_VERSION 0x20ULL /*Lustre versions in ocd */
1037 #define OBD_CONNECT_REQPORTAL 0x40ULL /*Separate non-IO req portal */
1038 #define OBD_CONNECT_ACL 0x80ULL /*access control lists */
1039 #define OBD_CONNECT_XATTR 0x100ULL /*client use extended attr */
1040 #define OBD_CONNECT_CROW 0x200ULL /*MDS+OST create obj on write*/
1041 #define OBD_CONNECT_TRUNCLOCK 0x400ULL /*locks on server for punch */
1042 #define OBD_CONNECT_TRANSNO 0x800ULL /*replay sends init transno */
1043 #define OBD_CONNECT_IBITS 0x1000ULL /*support for inodebits locks*/
1044 #define OBD_CONNECT_JOIN 0x2000ULL /*files can be concatenated.
1045 *We do not support JOIN FILE
1046 *anymore, reserve this flags
1047 *just for preventing such bit
1049 #define OBD_CONNECT_ATTRFID 0x4000ULL /*Server can GetAttr By Fid*/
1050 #define OBD_CONNECT_NODEVOH 0x8000ULL /*No open hndl on specl nodes*/
1051 #define OBD_CONNECT_RMT_CLIENT 0x10000ULL /*Remote client */
1052 #define OBD_CONNECT_RMT_CLIENT_FORCE 0x20000ULL /*Remote client by force */
1053 #define OBD_CONNECT_BRW_SIZE 0x40000ULL /*Max bytes per rpc */
1054 #define OBD_CONNECT_QUOTA64 0x80000ULL /*64bit qunit_data.qd_count */
1055 #define OBD_CONNECT_MDS_CAPA 0x100000ULL /*MDS capability */
1056 #define OBD_CONNECT_OSS_CAPA 0x200000ULL /*OSS capability */
1057 #define OBD_CONNECT_CANCELSET 0x400000ULL /*Early batched cancels. */
1058 #define OBD_CONNECT_SOM 0x800000ULL /*Size on MDS */
1059 #define OBD_CONNECT_AT 0x1000000ULL /*client uses AT */
1060 #define OBD_CONNECT_LRU_RESIZE 0x2000000ULL /*LRU resize feature. */
1061 #define OBD_CONNECT_MDS_MDS 0x4000000ULL /*MDS-MDS connection */
1062 #define OBD_CONNECT_REAL 0x8000000ULL /*real connection */
1063 #define OBD_CONNECT_CHANGE_QS 0x10000000ULL /*shrink/enlarge qunit */
1064 #define OBD_CONNECT_CKSUM 0x20000000ULL /*support several cksum algos*/
1065 #define OBD_CONNECT_FID 0x40000000ULL /*FID is supported by server */
1066 #define OBD_CONNECT_VBR 0x80000000ULL /*version based recovery */
1067 #define OBD_CONNECT_LOV_V3 0x100000000ULL /*client supports LOV v3 EA */
1068 #define OBD_CONNECT_GRANT_SHRINK 0x200000000ULL /* support grant shrink */
1069 #define OBD_CONNECT_SKIP_ORPHAN 0x400000000ULL /* don't reuse orphan objids */
1070 #define OBD_CONNECT_MAX_EASIZE 0x800000000ULL /* preserved for large EA */
1071 #define OBD_CONNECT_FULL20 0x1000000000ULL /* it is 2.0 client */
1072 #define OBD_CONNECT_LAYOUTLOCK 0x2000000000ULL /* client supports layout lock */
1073 /* also update obd_connect_names[] for lprocfs_rd_connect_flags()
1074 * and lustre/utils/wirecheck.c */
1076 #ifdef HAVE_LRU_RESIZE_SUPPORT
1077 #define LRU_RESIZE_CONNECT_FLAG OBD_CONNECT_LRU_RESIZE
1079 #define LRU_RESIZE_CONNECT_FLAG 0
1082 #define MDT_CONNECT_SUPPORTED (OBD_CONNECT_RDONLY | OBD_CONNECT_VERSION | \
1083 OBD_CONNECT_ACL | OBD_CONNECT_XATTR | \
1084 OBD_CONNECT_IBITS | OBD_CONNECT_JOIN | \
1085 OBD_CONNECT_NODEVOH | OBD_CONNECT_ATTRFID | \
1086 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1087 OBD_CONNECT_RMT_CLIENT | \
1088 OBD_CONNECT_RMT_CLIENT_FORCE | \
1089 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA | \
1090 OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID | \
1091 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_VBR | \
1092 OBD_CONNECT_LOV_V3 | OBD_CONNECT_SOM | \
1094 #define OST_CONNECT_SUPPORTED (OBD_CONNECT_SRVLOCK | OBD_CONNECT_GRANT | \
1095 OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | \
1096 OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_INDEX | \
1097 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_QUOTA64 | \
1098 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1099 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_CKSUM | \
1100 OBD_CONNECT_CHANGE_QS | \
1101 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_RMT_CLIENT | \
1102 OBD_CONNECT_RMT_CLIENT_FORCE | OBD_CONNECT_VBR | \
1103 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN | \
1104 OBD_CONNECT_GRANT_SHRINK | OBD_CONNECT_FULL20)
1105 #define ECHO_CONNECT_SUPPORTED (0)
1106 #define MGS_CONNECT_SUPPORTED (OBD_CONNECT_VERSION | OBD_CONNECT_AT | \
1109 #define OBD_OCD_VERSION(major,minor,patch,fix) (((major)<<24) + ((minor)<<16) +\
1110 ((patch)<<8) + (fix))
1111 #define OBD_OCD_VERSION_MAJOR(version) ((int)((version)>>24)&255)
1112 #define OBD_OCD_VERSION_MINOR(version) ((int)((version)>>16)&255)
1113 #define OBD_OCD_VERSION_PATCH(version) ((int)((version)>>8)&255)
1114 #define OBD_OCD_VERSION_FIX(version) ((int)(version)&255)
1116 /* This structure is used for both request and reply.
1118 * If we eventually have separate connect data for different types, which we
1119 * almost certainly will, then perhaps we stick a union in here. */
1120 struct obd_connect_data {
1121 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1122 __u32 ocd_version; /* lustre release version number */
1123 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1124 __u32 ocd_index; /* LOV index to connect to */
1125 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
1126 __u64 ocd_ibits_known; /* inode bits this client understands */
1127 __u32 ocd_nllu; /* non-local-lustre-user */
1128 __u32 ocd_nllg; /* non-local-lustre-group */
1129 __u64 ocd_transno; /* first transno from client to be replayed */
1130 __u32 ocd_group; /* MDS group on OST */
1131 __u32 ocd_cksum_types; /* supported checksum algorithms */
1132 __u64 padding1; /* also fix lustre_swab_connect */
1133 __u64 padding2; /* also fix lustre_swab_connect */
1136 extern void lustre_swab_connect(struct obd_connect_data *ocd);
1139 * Supported checksum algorithms. Up to 32 checksum types are supported.
1140 * (32-bit mask stored in obd_connect_data::ocd_cksum_types)
1141 * Please update DECLARE_CKSUM_NAME/OBD_CKSUM_ALL in obd.h when adding a new
1142 * algorithm and also the OBD_FL_CKSUM* flags.
1145 OBD_CKSUM_CRC32 = 0x00000001,
1146 OBD_CKSUM_ADLER = 0x00000002,
1150 * OST requests: OBDO & OBD request records
1155 OST_REPLY = 0, /* reply ? */
1171 OST_QUOTACHECK = 18,
1173 OST_QUOTA_ADJUST_QUNIT = 20,
1176 #define OST_FIRST_OPC OST_REPLY
1179 OBD_FL_INLINEDATA = 0x00000001,
1180 OBD_FL_OBDMDEXISTS = 0x00000002,
1181 OBD_FL_DELORPHAN = 0x00000004, /* if set in o_flags delete orphans */
1182 OBD_FL_NORPC = 0x00000008, /* set in o_flags do in OSC not OST */
1183 OBD_FL_IDONLY = 0x00000010, /* set in o_flags only adjust obj id*/
1184 OBD_FL_RECREATE_OBJS= 0x00000020, /* recreate missing obj */
1185 OBD_FL_DEBUG_CHECK = 0x00000040, /* echo client/server debug check */
1186 OBD_FL_NO_USRQUOTA = 0x00000100, /* the object's owner is over quota */
1187 OBD_FL_NO_GRPQUOTA = 0x00000200, /* the object's group is over quota */
1188 OBD_FL_CREATE_CROW = 0x00000400, /* object should be create on write */
1189 OBD_FL_SRVLOCK = 0x00000800, /* delegate DLM locking to server */
1190 OBD_FL_CKSUM_CRC32 = 0x00001000, /* CRC32 checksum type */
1191 OBD_FL_CKSUM_ADLER = 0x00002000, /* ADLER checksum type */
1192 OBD_FL_CKSUM_RSVD1 = 0x00004000, /* for future cksum types */
1193 OBD_FL_CKSUM_RSVD2 = 0x00008000, /* for future cksum types */
1194 OBD_FL_CKSUM_RSVD3 = 0x00010000, /* for future cksum types */
1195 OBD_FL_SHRINK_GRANT = 0x00020000, /* object shrink the grant */
1196 OBD_FL_MMAP = 0x00040000, /* object is mmapped on the client */
1197 OBD_FL_RECOV_RESEND = 0x00080000, /* recoverable resent */
1199 OBD_FL_CKSUM_ALL = OBD_FL_CKSUM_CRC32 | OBD_FL_CKSUM_ADLER,
1201 /* mask for local-only flag, which won't be sent over network */
1202 OBD_FL_LOCAL_MASK = 0xF0000000,
1205 #define LOV_MAGIC_V1 0x0BD10BD0
1206 #define LOV_MAGIC LOV_MAGIC_V1
1207 #define LOV_MAGIC_JOIN_V1 0x0BD20BD0
1208 #define LOV_MAGIC_V3 0x0BD30BD0
1210 #define LOV_PATTERN_RAID0 0x001 /* stripes are used round-robin */
1211 #define LOV_PATTERN_RAID1 0x002 /* stripes are mirrors of each other */
1212 #define LOV_PATTERN_FIRST 0x100 /* first stripe is not in round-robin */
1213 #define LOV_PATTERN_CMOBD 0x200
1215 #define LOV_OBJECT_GROUP_DEFAULT ~0ULL
1216 #define LOV_OBJECT_GROUP_CLEAR 0ULL
1218 #define lov_ost_data lov_ost_data_v1
1219 struct lov_ost_data_v1 { /* per-stripe data structure (little-endian)*/
1220 __u64 l_object_id; /* OST object ID */
1221 __u64 l_object_seq; /* OST object seq number */
1222 __u32 l_ost_gen; /* generation of this l_ost_idx */
1223 __u32 l_ost_idx; /* OST index in LOV (lov_tgt_desc->tgts) */
1226 #define lov_mds_md lov_mds_md_v1
1227 struct lov_mds_md_v1 { /* LOV EA mds/wire data (little-endian) */
1228 __u32 lmm_magic; /* magic number = LOV_MAGIC_V1 */
1229 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1230 __u64 lmm_object_id; /* LOV object ID */
1231 __u64 lmm_object_seq; /* LOV object seq number */
1232 __u32 lmm_stripe_size; /* size of stripe in bytes */
1233 __u32 lmm_stripe_count; /* num stripes in use for this object */
1234 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1237 /* extern void lustre_swab_lov_mds_md(struct lov_mds_md *llm); */
1239 #define MAX_MD_SIZE (sizeof(struct lov_mds_md) + 4 * sizeof(struct lov_ost_data))
1240 #define MIN_MD_SIZE (sizeof(struct lov_mds_md) + 1 * sizeof(struct lov_ost_data))
1242 #define XATTR_NAME_ACL_ACCESS "system.posix_acl_access"
1243 #define XATTR_NAME_ACL_DEFAULT "system.posix_acl_default"
1244 #define XATTR_USER_PREFIX "user."
1245 #define XATTR_TRUSTED_PREFIX "trusted."
1246 #define XATTR_SECURITY_PREFIX "security."
1247 #define XATTR_LUSTRE_PREFIX "lustre."
1249 #define XATTR_NAME_LOV "trusted.lov"
1250 #define XATTR_NAME_LMA "trusted.lma"
1251 #define XATTR_NAME_LMV "trusted.lmv"
1252 #define XATTR_NAME_LINK "trusted.link"
1255 struct lov_mds_md_v3 { /* LOV EA mds/wire data (little-endian) */
1256 __u32 lmm_magic; /* magic number = LOV_MAGIC_V3 */
1257 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1258 __u64 lmm_object_id; /* LOV object ID */
1259 __u64 lmm_object_seq; /* LOV object seq number */
1260 __u32 lmm_stripe_size; /* size of stripe in bytes */
1261 __u32 lmm_stripe_count; /* num stripes in use for this object */
1262 char lmm_pool_name[LOV_MAXPOOLNAME]; /* must be 32bit aligned */
1263 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1267 #define OBD_MD_FLID (0x00000001ULL) /* object ID */
1268 #define OBD_MD_FLATIME (0x00000002ULL) /* access time */
1269 #define OBD_MD_FLMTIME (0x00000004ULL) /* data modification time */
1270 #define OBD_MD_FLCTIME (0x00000008ULL) /* change time */
1271 #define OBD_MD_FLSIZE (0x00000010ULL) /* size */
1272 #define OBD_MD_FLBLOCKS (0x00000020ULL) /* allocated blocks count */
1273 #define OBD_MD_FLBLKSZ (0x00000040ULL) /* block size */
1274 #define OBD_MD_FLMODE (0x00000080ULL) /* access bits (mode & ~S_IFMT) */
1275 #define OBD_MD_FLTYPE (0x00000100ULL) /* object type (mode & S_IFMT) */
1276 #define OBD_MD_FLUID (0x00000200ULL) /* user ID */
1277 #define OBD_MD_FLGID (0x00000400ULL) /* group ID */
1278 #define OBD_MD_FLFLAGS (0x00000800ULL) /* flags word */
1279 #define OBD_MD_FLNLINK (0x00002000ULL) /* link count */
1280 #define OBD_MD_FLGENER (0x00004000ULL) /* generation number */
1281 /*#define OBD_MD_FLINLINE (0x00008000ULL) inline data. used until 1.6.5 */
1282 #define OBD_MD_FLRDEV (0x00010000ULL) /* device number */
1283 #define OBD_MD_FLEASIZE (0x00020000ULL) /* extended attribute data */
1284 #define OBD_MD_LINKNAME (0x00040000ULL) /* symbolic link target */
1285 #define OBD_MD_FLHANDLE (0x00080000ULL) /* file/lock handle */
1286 #define OBD_MD_FLCKSUM (0x00100000ULL) /* bulk data checksum */
1287 #define OBD_MD_FLQOS (0x00200000ULL) /* quality of service stats */
1288 /*#define OBD_MD_FLOSCOPQ (0x00400000ULL) osc opaque data, never used */
1289 #define OBD_MD_FLCOOKIE (0x00800000ULL) /* log cancellation cookie */
1290 #define OBD_MD_FLGROUP (0x01000000ULL) /* group */
1291 #define OBD_MD_FLFID (0x02000000ULL) /* ->ost write inline fid */
1292 #define OBD_MD_FLEPOCH (0x04000000ULL) /* ->ost write with ioepoch */
1293 /* ->mds if epoch opens or closes */
1294 #define OBD_MD_FLGRANT (0x08000000ULL) /* ost preallocation space grant */
1295 #define OBD_MD_FLDIREA (0x10000000ULL) /* dir's extended attribute data */
1296 #define OBD_MD_FLUSRQUOTA (0x20000000ULL) /* over quota flags sent from ost */
1297 #define OBD_MD_FLGRPQUOTA (0x40000000ULL) /* over quota flags sent from ost */
1298 #define OBD_MD_FLMODEASIZE (0x80000000ULL) /* EA size will be changed */
1300 #define OBD_MD_MDS (0x0000000100000000ULL) /* where an inode lives on */
1301 #define OBD_MD_REINT (0x0000000200000000ULL) /* reintegrate oa */
1302 #define OBD_MD_MEA (0x0000000400000000ULL) /* CMD split EA */
1303 #define OBD_MD_MDTIDX (0x0000000800000000ULL) /* Get MDT index */
1305 #define OBD_MD_FLXATTR (0x0000001000000000ULL) /* xattr */
1306 #define OBD_MD_FLXATTRLS (0x0000002000000000ULL) /* xattr list */
1307 #define OBD_MD_FLXATTRRM (0x0000004000000000ULL) /* xattr remove */
1308 #define OBD_MD_FLACL (0x0000008000000000ULL) /* ACL */
1309 #define OBD_MD_FLRMTPERM (0x0000010000000000ULL) /* remote permission */
1310 #define OBD_MD_FLMDSCAPA (0x0000020000000000ULL) /* MDS capability */
1311 #define OBD_MD_FLOSSCAPA (0x0000040000000000ULL) /* OSS capability */
1312 #define OBD_MD_FLCKSPLIT (0x0000080000000000ULL) /* Check split on server */
1313 #define OBD_MD_FLCROSSREF (0x0000100000000000ULL) /* Cross-ref case */
1314 #define OBD_MD_FLGETATTRLOCK (0x0000200000000000ULL) /* Get IOEpoch attributes
1316 #define OBD_FL_TRUNC (0x0000200000000000ULL) /* for filter_truncate */
1318 #define OBD_MD_FLRMTLSETFACL (0x0001000000000000ULL) /* lfs lsetfacl case */
1319 #define OBD_MD_FLRMTLGETFACL (0x0002000000000000ULL) /* lfs lgetfacl case */
1320 #define OBD_MD_FLRMTRSETFACL (0x0004000000000000ULL) /* lfs rsetfacl case */
1321 #define OBD_MD_FLRMTRGETFACL (0x0008000000000000ULL) /* lfs rgetfacl case */
1323 #define OBD_MD_FLGETATTR (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLMTIME | \
1324 OBD_MD_FLCTIME | OBD_MD_FLSIZE | OBD_MD_FLBLKSZ | \
1325 OBD_MD_FLMODE | OBD_MD_FLTYPE | OBD_MD_FLUID | \
1326 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLNLINK | \
1327 OBD_MD_FLGENER | OBD_MD_FLRDEV | OBD_MD_FLGROUP)
1329 /* don't forget obdo_fid which is way down at the bottom so it can
1330 * come after the definition of llog_cookie */
1333 extern void lustre_swab_obd_statfs (struct obd_statfs *os);
1334 #define OBD_STATFS_NODELAY 0x0001 /* requests should be send without delay
1335 * and resends for avoid deadlocks */
1336 #define OBD_STATFS_FROM_CACHE 0x0002 /* the statfs callback should not update
1338 #define OBD_STATFS_PTLRPCD 0x0004 /* requests will be sent via ptlrpcd
1339 * instead of a specific set. This
1340 * means that we cannot rely on the set
1341 * interpret routine to be called.
1342 * lov_statfs_fini() must thus be called
1343 * by the request interpret routine */
1345 /* ost_body.data values for OST_BRW */
1347 #define OBD_BRW_READ 0x01
1348 #define OBD_BRW_WRITE 0x02
1349 #define OBD_BRW_RWMASK (OBD_BRW_READ | OBD_BRW_WRITE)
1350 #define OBD_BRW_SYNC 0x08 /* this page is a part of synchronous
1351 * transfer and is not accounted in
1353 #define OBD_BRW_CHECK 0x10
1354 #define OBD_BRW_FROM_GRANT 0x20 /* the osc manages this under llite */
1355 #define OBD_BRW_GRANTED 0x40 /* the ost manages this */
1356 #define OBD_BRW_NOCACHE 0x80 /* this page is a part of non-cached IO */
1357 #define OBD_BRW_NOQUOTA 0x100
1358 #define OBD_BRW_SRVLOCK 0x200 /* Client holds no lock over this page */
1359 #define OBD_BRW_ASYNC 0x400 /* Server may delay commit to disk */
1360 #define OBD_BRW_MEMALLOC 0x800 /* Client runs in the "kswapd" context */
1362 #define OBD_OBJECT_EOF 0xffffffffffffffffULL
1364 #define OST_MIN_PRECREATE 32
1365 #define OST_MAX_PRECREATE 20000
1374 extern void lustre_swab_obd_ioobj (struct obd_ioobj *ioo);
1376 /* multiple of 8 bytes => can array */
1377 struct niobuf_remote {
1383 extern void lustre_swab_niobuf_remote (struct niobuf_remote *nbr);
1385 /* lock value block communicated between the filter and llite */
1387 /* OST_LVB_ERR_INIT is needed because the return code in rc is
1388 * negative, i.e. because ((MASK + rc) & MASK) != MASK. */
1389 #define OST_LVB_ERR_INIT 0xffbadbad80000000ULL
1390 #define OST_LVB_ERR_MASK 0xffbadbad00000000ULL
1391 #define OST_LVB_IS_ERR(blocks) \
1392 ((blocks & OST_LVB_ERR_MASK) == OST_LVB_ERR_MASK)
1393 #define OST_LVB_SET_ERR(blocks, rc) \
1394 do { blocks = OST_LVB_ERR_INIT + rc; } while (0)
1395 #define OST_LVB_GET_ERR(blocks) (int)(blocks - OST_LVB_ERR_INIT)
1405 extern void lustre_swab_ost_lvb(struct ost_lvb *);
1414 MDS_GETATTR_NAME = 34,
1419 MDS_DISCONNECT = 39,
1425 MDS_DONE_WRITING = 45,
1427 MDS_QUOTACHECK = 47,
1430 MDS_SETXATTR = 50, /* obsolete, now it's MDS_REINT op */
1437 #define MDS_FIRST_OPC MDS_GETATTR
1454 } mds_reint_t, mdt_reint_t;
1456 extern void lustre_swab_generic_32s (__u32 *val);
1458 /* the disposition of the intent outlines what was executed */
1459 #define DISP_IT_EXECD 0x00000001
1460 #define DISP_LOOKUP_EXECD 0x00000002
1461 #define DISP_LOOKUP_NEG 0x00000004
1462 #define DISP_LOOKUP_POS 0x00000008
1463 #define DISP_OPEN_CREATE 0x00000010
1464 #define DISP_OPEN_OPEN 0x00000020
1465 #define DISP_ENQ_COMPLETE 0x00400000
1466 #define DISP_ENQ_OPEN_REF 0x00800000
1467 #define DISP_ENQ_CREATE_REF 0x01000000
1468 #define DISP_OPEN_LOCK 0x02000000
1470 /* INODE LOCK PARTS */
1471 #define MDS_INODELOCK_LOOKUP 0x000001 /* dentry, mode, owner, group */
1472 #define MDS_INODELOCK_UPDATE 0x000002 /* size, links, timestamps */
1473 #define MDS_INODELOCK_OPEN 0x000004 /* For opened files */
1475 /* Do not forget to increase MDS_INODELOCK_MAXSHIFT when adding new bits */
1476 #define MDS_INODELOCK_MAXSHIFT 2
1477 /* This FULL lock is useful to take on unlink sort of operations */
1478 #define MDS_INODELOCK_FULL ((1<<(MDS_INODELOCK_MAXSHIFT+1))-1)
1480 extern void lustre_swab_ll_fid (struct ll_fid *fid);
1482 #define MDS_STATUS_CONN 1
1483 #define MDS_STATUS_LOV 2
1485 struct mds_status_req {
1490 extern void lustre_swab_mds_status_req (struct mds_status_req *r);
1492 /* mdt_thread_info.mti_flags. */
1494 /* The flag indicates Size-on-MDS attributes are changed. */
1495 MF_SOM_CHANGE = (1 << 0),
1496 /* Flags indicates an epoch opens or closes. */
1497 MF_EPOCH_OPEN = (1 << 1),
1498 MF_EPOCH_CLOSE = (1 << 2),
1499 MF_MDC_CANCEL_FID1 = (1 << 3),
1500 MF_MDC_CANCEL_FID2 = (1 << 4),
1501 MF_MDC_CANCEL_FID3 = (1 << 5),
1502 MF_MDC_CANCEL_FID4 = (1 << 6),
1503 /* There is a pending attribute update. */
1504 MF_SOM_AU = (1 << 7),
1505 /* Cancel OST locks while getattr OST attributes. */
1506 MF_GETATTR_LOCK = (1 << 8),
1509 #define MF_SOM_LOCAL_FLAGS (MF_SOM_CHANGE | MF_EPOCH_OPEN | MF_EPOCH_CLOSE)
1511 #define MDS_BFLAG_UNCOMMITTED_WRITES 0x1
1513 /* these should be identical to their EXT3_*_FL counterparts, and are
1514 * redefined here only to avoid dragging in ext3_fs.h */
1515 #define MDS_SYNC_FL 0x00000008 /* Synchronous updates */
1516 #define MDS_IMMUTABLE_FL 0x00000010 /* Immutable file */
1517 #define MDS_APPEND_FL 0x00000020 /* writes to file may only append */
1518 #define MDS_NOATIME_FL 0x00000080 /* do not update atime */
1519 #define MDS_DIRSYNC_FL 0x00010000 /* dirsync behaviour (dir only) */
1522 /* Convert wire MDS_*_FL to corresponding client local VFS S_* values
1523 * for the client inode i_flags. The MDS_*_FL are the Lustre wire
1524 * protocol equivalents of LDISKFS_*_FL values stored on disk, while
1525 * the S_* flags are kernel-internal values that change between kernel
1526 * versions. These flags are set/cleared via FSFILT_IOC_{GET,SET}_FLAGS.
1527 * See b=16526 for a full history. */
1528 static inline int ll_ext_to_inode_flags(int flags)
1530 return (((flags & MDS_SYNC_FL) ? S_SYNC : 0) |
1531 ((flags & MDS_NOATIME_FL) ? S_NOATIME : 0) |
1532 ((flags & MDS_APPEND_FL) ? S_APPEND : 0) |
1533 #if defined(S_DIRSYNC)
1534 ((flags & MDS_DIRSYNC_FL) ? S_DIRSYNC : 0) |
1536 ((flags & MDS_IMMUTABLE_FL) ? S_IMMUTABLE : 0));
1539 static inline int ll_inode_to_ext_flags(int iflags)
1541 return (((iflags & S_SYNC) ? MDS_SYNC_FL : 0) |
1542 ((iflags & S_NOATIME) ? MDS_NOATIME_FL : 0) |
1543 ((iflags & S_APPEND) ? MDS_APPEND_FL : 0) |
1544 #if defined(S_DIRSYNC)
1545 ((iflags & S_DIRSYNC) ? MDS_DIRSYNC_FL : 0) |
1547 ((iflags & S_IMMUTABLE) ? MDS_IMMUTABLE_FL : 0));
1552 * while mds_body is to interact with 1.6, mdt_body is to interact with 2.0.
1553 * both of them should have the same fields layout, because at client side
1554 * one could be dynamically cast to the other.
1556 * mdt_body has large size than mds_body, with unused padding (48 bytes)
1557 * at the end. client always use size of mdt_body to prepare request/reply
1558 * buffers, and actual data could be interepeted as mdt_body or mds_body
1564 struct lustre_handle handle;
1566 __u64 size; /* Offset, in the case of MDS_READPAGE */
1570 __u64 blocks; /* XID, in the case of MDS_READPAGE */
1579 __u32 flags; /* from vfs for pin/unpin, MDS_BFLAG for close */
1581 __u32 nlink; /* #bytes to read in the case of MDS_READPAGE */
1587 __u32 max_cookiesize;
1588 __u32 padding_4; /* also fix lustre_swab_mds_body */
1591 extern void lustre_swab_mds_body (struct mds_body *b);
1596 struct lustre_handle handle;
1598 __u64 size; /* Offset, in the case of MDS_READPAGE */
1602 __u64 blocks; /* XID, in the case of MDS_READPAGE */
1604 __u64 ino; /* for 1.6 compatibility */
1611 __u32 flags; /* from vfs for pin/unpin, MDS_BFLAG for close */
1613 __u32 nlink; /* #bytes to read in the case of MDS_READPAGE */
1614 __u32 generation; /* for 1.6 compatibility */
1619 __u32 max_cookiesize;
1620 __u32 uid_h; /* high 32-bits of uid, for FUID */
1621 __u32 gid_h; /* high 32-bits of gid, for FUID */
1622 __u32 padding_5; /* also fix lustre_swab_mdt_body */
1630 extern void lustre_swab_mdt_body (struct mdt_body *b);
1632 struct mdt_ioepoch {
1633 struct lustre_handle handle;
1639 extern void lustre_swab_mdt_ioepoch (struct mdt_ioepoch *b);
1641 #define Q_QUOTACHECK 0x800100
1642 #define Q_INITQUOTA 0x800101 /* init slave limits */
1643 #define Q_GETOINFO 0x800102 /* get obd quota info */
1644 #define Q_GETOQUOTA 0x800103 /* get obd quotas */
1645 #define Q_FINVALIDATE 0x800104 /* invalidate operational quotas */
1647 #define Q_TYPEMATCH(id, type) \
1648 ((id) == (type) || (id) == UGQUOTA)
1650 #define Q_TYPESET(oqc, type) Q_TYPEMATCH((oqc)->qc_type, type)
1652 #define Q_GETOCMD(oqc) \
1653 ((oqc)->qc_cmd == Q_GETOINFO || (oqc)->qc_cmd == Q_GETOQUOTA)
1655 #define QCTL_COPY(out, in) \
1657 Q_COPY(out, in, qc_cmd); \
1658 Q_COPY(out, in, qc_type); \
1659 Q_COPY(out, in, qc_id); \
1660 Q_COPY(out, in, qc_stat); \
1661 Q_COPY(out, in, qc_dqinfo); \
1662 Q_COPY(out, in, qc_dqblk); \
1665 struct obd_quotactl {
1670 struct obd_dqinfo qc_dqinfo;
1671 struct obd_dqblk qc_dqblk;
1674 extern void lustre_swab_obd_quotactl(struct obd_quotactl *q);
1676 struct quota_adjust_qunit {
1683 extern void lustre_swab_quota_adjust_qunit(struct quota_adjust_qunit *q);
1685 /* flags is shared among quota structures */
1686 #define LQUOTA_FLAGS_GRP 1UL /* 0 is user, 1 is group */
1687 #define LQUOTA_FLAGS_BLK 2UL /* 0 is inode, 1 is block */
1688 #define LQUOTA_FLAGS_ADJBLK 4UL /* adjust the block qunit size */
1689 #define LQUOTA_FLAGS_ADJINO 8UL /* adjust the inode qunit size */
1690 #define LQUOTA_FLAGS_CHG_QS 16UL /* indicate whether it has capability of
1691 * OBD_CONNECT_CHANGE_QS */
1692 #define LQUOTA_FLAGS_RECOVERY 32UL /* recovery is going on a uid/gid */
1693 #define LQUOTA_FLAGS_SETQUOTA 64UL /* being setquota on a uid/gid */
1695 /* flags is specific for quota_adjust_qunit */
1696 #define LQUOTA_QAQ_CREATE_LQS (1 << 31) /* when it is set, need create lqs */
1698 /* the status of lqs_flags in struct lustre_qunit_size */
1699 #define LQUOTA_QUNIT_FLAGS (LQUOTA_FLAGS_GRP | LQUOTA_FLAGS_BLK)
1701 #define QAQ_IS_GRP(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_GRP)
1702 #define QAQ_IS_ADJBLK(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_ADJBLK)
1703 #define QAQ_IS_ADJINO(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_ADJINO)
1704 #define QAQ_IS_CREATE_LQS(qaq) ((qaq)->qaq_flags & LQUOTA_QAQ_CREATE_LQS)
1706 #define QAQ_SET_GRP(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_GRP)
1707 #define QAQ_SET_ADJBLK(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_ADJBLK)
1708 #define QAQ_SET_ADJINO(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_ADJINO)
1709 #define QAQ_SET_CREATE_LQS(qaq) ((qaq)->qaq_flags |= LQUOTA_QAQ_CREATE_LQS)
1711 /* inode access permission for remote user, the inode info are omitted,
1712 * for client knows them. */
1713 struct mds_remote_perm {
1718 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
1721 /* permissions for md_perm.mp_perm */
1723 CFS_SETUID_PERM = 0x01,
1724 CFS_SETGID_PERM = 0x02,
1725 CFS_SETGRP_PERM = 0x04,
1726 CFS_RMTACL_PERM = 0x08,
1727 CFS_RMTOWN_PERM = 0x10
1730 extern void lustre_swab_mds_remote_perm(struct mds_remote_perm *p);
1732 struct mdt_remote_perm {
1739 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
1742 extern void lustre_swab_mdt_remote_perm(struct mdt_remote_perm *p);
1744 struct mdt_rec_setattr {
1754 __u32 sa_padding_1_h;
1755 struct lu_fid sa_fid;
1764 __u32 sa_attr_flags;
1772 extern void lustre_swab_mdt_rec_setattr (struct mdt_rec_setattr *sa);
1775 * Attribute flags used in mdt_rec_setattr::sa_valid.
1776 * The kernel's #defines for ATTR_* should not be used over the network
1777 * since the client and MDS may run different kernels (see bug 13828)
1778 * Therefore, we should only use MDS_ATTR_* attributes for sa_valid.
1780 #define MDS_ATTR_MODE 0x1ULL /* = 1 */
1781 #define MDS_ATTR_UID 0x2ULL /* = 2 */
1782 #define MDS_ATTR_GID 0x4ULL /* = 4 */
1783 #define MDS_ATTR_SIZE 0x8ULL /* = 8 */
1784 #define MDS_ATTR_ATIME 0x10ULL /* = 16 */
1785 #define MDS_ATTR_MTIME 0x20ULL /* = 32 */
1786 #define MDS_ATTR_CTIME 0x40ULL /* = 64 */
1787 #define MDS_ATTR_ATIME_SET 0x80ULL /* = 128 */
1788 #define MDS_ATTR_MTIME_SET 0x100ULL /* = 256 */
1789 #define MDS_ATTR_FORCE 0x200ULL /* = 512, Not a change, but a change it */
1790 #define MDS_ATTR_ATTR_FLAG 0x400ULL /* = 1024 */
1791 #define MDS_ATTR_KILL_SUID 0x800ULL /* = 2048 */
1792 #define MDS_ATTR_KILL_SGID 0x1000ULL /* = 4096 */
1793 #define MDS_ATTR_CTIME_SET 0x2000ULL /* = 8192 */
1794 #define MDS_ATTR_FROM_OPEN 0x4000ULL /* = 16384, called from open path, ie O_TRUNC */
1795 #define MDS_ATTR_BLOCKS 0x8000ULL /* = 32768 */
1798 #define FMODE_READ 00000001
1799 #define FMODE_WRITE 00000002
1802 /* IO Epoch is opened on a closed file. */
1803 #define FMODE_EPOCH 01000000
1804 /* IO Epoch is opened on a file truncate. */
1805 #define FMODE_TRUNC 02000000
1806 /* Size-on-MDS Attribute Update is pending. */
1807 #define FMODE_SOM 04000000
1808 #define FMODE_CLOSED 0
1810 #define MDS_OPEN_CREATED 00000010
1811 #define MDS_OPEN_CROSS 00000020
1813 #define MDS_FMODE_EXEC 00000004
1814 #define MDS_OPEN_CREAT 00000100
1815 #define MDS_OPEN_EXCL 00000200
1816 #define MDS_OPEN_TRUNC 00001000
1817 #define MDS_OPEN_APPEND 00002000
1818 #define MDS_OPEN_SYNC 00010000
1819 #define MDS_OPEN_DIRECTORY 00200000
1821 #define MDS_OPEN_DELAY_CREATE 0100000000 /* delay initial object create */
1822 #define MDS_OPEN_OWNEROVERRIDE 0200000000 /* NFSD rw-reopen ro file for owner */
1823 #define MDS_OPEN_JOIN_FILE 0400000000 /* open for join file.
1824 * We do not support JOIN FILE
1825 * anymore, reserve this flags
1826 * just for preventing such bit
1828 #define MDS_CREATE_RMT_ACL 01000000000 /* indicate create on remote server
1829 * with default ACL */
1830 #define MDS_CREATE_SLAVE_OBJ 02000000000 /* indicate create slave object
1831 * actually, this is for create, not
1832 * conflict with other open flags */
1833 #define MDS_OPEN_LOCK 04000000000 /* This open requires open lock */
1834 #define MDS_OPEN_HAS_EA 010000000000 /* specify object create pattern */
1835 #define MDS_OPEN_HAS_OBJS 020000000000 /* Just set the EA the obj exist */
1836 #define MDS_OPEN_NORESTORE 0100000000000ULL /* Do not restore file at open */
1837 #define MDS_OPEN_NEWSTRIPE 0200000000000ULL /* New stripe needed (restripe or
1840 /* permission for create non-directory file */
1841 #define MAY_CREATE (1 << 7)
1842 /* permission for create directory file */
1843 #define MAY_LINK (1 << 8)
1844 /* permission for delete from the directory */
1845 #define MAY_UNLINK (1 << 9)
1846 /* source's permission for rename */
1847 #define MAY_RENAME_SRC (1 << 10)
1848 /* target's permission for rename */
1849 #define MAY_RENAME_TAR (1 << 11)
1850 /* part (parent's) VTX permission check */
1851 #define MAY_VTX_PART (1 << 12)
1852 /* full VTX permission check */
1853 #define MAY_VTX_FULL (1 << 13)
1854 /* lfs rgetfacl permission check */
1855 #define MAY_RGETFACL (1 << 14)
1858 MDS_CHECK_SPLIT = 1 << 0,
1859 MDS_CROSS_REF = 1 << 1,
1860 MDS_VTX_BYPASS = 1 << 2,
1861 MDS_PERM_BYPASS = 1 << 3,
1863 MDS_QUOTA_IGNORE = 1 << 5,
1864 MDS_CLOSE_CLEANUP = 1 << 6,
1865 MDS_KEEP_ORPHAN = 1 << 7
1868 /* instance of mdt_reint_rec */
1869 struct mdt_rec_create {
1877 __u32 cr_suppgid1_h;
1879 __u32 cr_suppgid2_h;
1880 struct lu_fid cr_fid1;
1881 struct lu_fid cr_fid2;
1882 struct lustre_handle cr_old_handle; /* handle in case of open replay */
1886 __u64 cr_padding_1; /* rr_blocks */
1889 /* use of helpers set/get_mrc_cr_flags() is needed to access
1890 * 64 bits cr_flags [cr_flags_l, cr_flags_h], this is done to
1891 * extend cr_flags size without breaking 1.8 compat */
1892 __u32 cr_flags_l; /* for use with open, low 32 bits */
1893 __u32 cr_flags_h; /* for use with open, high 32 bits */
1894 __u32 cr_padding_3; /* rr_padding_3 */
1895 __u32 cr_padding_4; /* rr_padding_4 */
1898 static inline void set_mrc_cr_flags(struct mdt_rec_create *mrc, __u64 flags)
1900 mrc->cr_flags_l = (__u32)(flags & 0xFFFFFFFFUll);
1901 mrc->cr_flags_h = (__u32)(flags >> 32);
1904 static inline __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
1906 return ((__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32));
1909 /* instance of mdt_reint_rec */
1910 struct mdt_rec_link {
1918 __u32 lk_suppgid1_h;
1920 __u32 lk_suppgid2_h;
1921 struct lu_fid lk_fid1;
1922 struct lu_fid lk_fid2;
1924 __u64 lk_padding_1; /* rr_atime */
1925 __u64 lk_padding_2; /* rr_ctime */
1926 __u64 lk_padding_3; /* rr_size */
1927 __u64 lk_padding_4; /* rr_blocks */
1929 __u32 lk_padding_5; /* rr_mode */
1930 __u32 lk_padding_6; /* rr_flags */
1931 __u32 lk_padding_7; /* rr_padding_2 */
1932 __u32 lk_padding_8; /* rr_padding_3 */
1933 __u32 lk_padding_9; /* rr_padding_4 */
1936 /* instance of mdt_reint_rec */
1937 struct mdt_rec_unlink {
1945 __u32 ul_suppgid1_h;
1947 __u32 ul_suppgid2_h;
1948 struct lu_fid ul_fid1;
1949 struct lu_fid ul_fid2;
1951 __u64 ul_padding_2; /* rr_atime */
1952 __u64 ul_padding_3; /* rr_ctime */
1953 __u64 ul_padding_4; /* rr_size */
1954 __u64 ul_padding_5; /* rr_blocks */
1957 __u32 ul_padding_6; /* rr_flags */
1958 __u32 ul_padding_7; /* rr_padding_2 */
1959 __u32 ul_padding_8; /* rr_padding_3 */
1960 __u32 ul_padding_9; /* rr_padding_4 */
1963 /* instance of mdt_reint_rec */
1964 struct mdt_rec_rename {
1972 __u32 rn_suppgid1_h;
1974 __u32 rn_suppgid2_h;
1975 struct lu_fid rn_fid1;
1976 struct lu_fid rn_fid2;
1978 __u64 rn_padding_1; /* rr_atime */
1979 __u64 rn_padding_2; /* rr_ctime */
1980 __u64 rn_padding_3; /* rr_size */
1981 __u64 rn_padding_4; /* rr_blocks */
1982 __u32 rn_bias; /* some operation flags */
1983 __u32 rn_mode; /* cross-ref rename has mode */
1984 __u32 rn_padding_5; /* rr_flags */
1985 __u32 rn_padding_6; /* rr_padding_2 */
1986 __u32 rn_padding_7; /* rr_padding_3 */
1987 __u32 rn_padding_8; /* rr_padding_4 */
1990 /* instance of mdt_reint_rec */
1991 struct mdt_rec_setxattr {
1999 __u32 sx_suppgid1_h;
2001 __u32 sx_suppgid2_h;
2002 struct lu_fid sx_fid;
2003 __u64 sx_padding_1; /* These three are rr_fid2 */
2008 __u64 sx_padding_5; /* rr_ctime */
2009 __u64 sx_padding_6; /* rr_size */
2010 __u64 sx_padding_7; /* rr_blocks */
2013 __u32 sx_padding_8; /* rr_flags */
2014 __u32 sx_padding_9; /* rr_padding_2 */
2015 __u32 sx_padding_10; /* rr_padding_3 */
2016 __u32 sx_padding_11; /* rr_padding_4 */
2020 * mdt_rec_reint is the template for all mdt_reint_xxx structures.
2021 * Do NOT change the size of various members, otherwise the value
2022 * will be broken in lustre_swab_mdt_rec_reint().
2024 * If you add new members in other mdt_reint_xxx structres and need to use the
2025 * rr_padding_x fields, then update lustre_swab_mdt_rec_reint() also.
2027 struct mdt_rec_reint {
2035 __u32 rr_suppgid1_h;
2037 __u32 rr_suppgid2_h;
2038 struct lu_fid rr_fid1;
2039 struct lu_fid rr_fid2;
2048 __u32 rr_padding_2; /* also fix lustre_swab_mdt_rec_reint */
2049 __u32 rr_padding_3; /* also fix lustre_swab_mdt_rec_reint */
2050 __u32 rr_padding_4; /* also fix lustre_swab_mdt_rec_reint */
2053 extern void lustre_swab_mdt_rec_reint(struct mdt_rec_reint *rr);
2056 __u32 ld_tgt_count; /* how many MDS's */
2057 __u32 ld_active_tgt_count; /* how many active */
2058 __u32 ld_default_stripe_count; /* how many objects are used */
2059 __u32 ld_pattern; /* default MEA_MAGIC_* */
2060 __u64 ld_default_hash_size;
2061 __u64 ld_padding_1; /* also fix lustre_swab_lmv_desc */
2062 __u32 ld_padding_2; /* also fix lustre_swab_lmv_desc */
2063 __u32 ld_qos_maxage; /* in second */
2064 __u32 ld_padding_3; /* also fix lustre_swab_lmv_desc */
2065 __u32 ld_padding_4; /* also fix lustre_swab_lmv_desc */
2066 struct obd_uuid ld_uuid;
2069 extern void lustre_swab_lmv_desc (struct lmv_desc *ld);
2071 /* TODO: lmv_stripe_md should contain mds capabilities for all slave fids */
2072 struct lmv_stripe_md {
2077 char mea_pool_name[LOV_MAXPOOLNAME];
2078 struct lu_fid mea_ids[0];
2081 extern void lustre_swab_lmv_stripe_md(struct lmv_stripe_md *mea);
2083 /* lmv structures */
2084 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
2085 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
2086 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
2088 #define MAX_HASH_SIZE_32 0x7fffffffUL
2089 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
2090 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
2097 extern void lustre_swab_md_fld (struct md_fld *mf);
2102 FLD_FIRST_OPC = FLD_QUERY
2108 SEQ_FIRST_OPC = SEQ_QUERY
2112 SEQ_ALLOC_SUPER = 0,
2117 * LOV data structures
2120 #define LOV_MIN_STRIPE_BITS 16 /* maximum PAGE_SIZE (ia64), power of 2 */
2121 #define LOV_MIN_STRIPE_SIZE (1<<LOV_MIN_STRIPE_BITS)
2122 #define LOV_MAX_STRIPE_COUNT 160 /* until bug 4424 is fixed */
2123 #define LOV_V1_INSANE_STRIPE_COUNT 65532 /* maximum stripe count bz13933 */
2125 #define LOV_MAX_UUID_BUFFER_SIZE 8192
2126 /* The size of the buffer the lov/mdc reserves for the
2127 * array of UUIDs returned by the MDS. With the current
2128 * protocol, this will limit the max number of OSTs per LOV */
2130 #define LOV_DESC_MAGIC 0xB0CCDE5C
2132 /* LOV settings descriptor (should only contain static info) */
2134 __u32 ld_tgt_count; /* how many OBD's */
2135 __u32 ld_active_tgt_count; /* how many active */
2136 __u32 ld_default_stripe_count; /* how many objects are used */
2137 __u32 ld_pattern; /* default PATTERN_RAID0 */
2138 __u64 ld_default_stripe_size; /* in bytes */
2139 __u64 ld_default_stripe_offset; /* in bytes */
2140 __u32 ld_padding_0; /* unused */
2141 __u32 ld_qos_maxage; /* in second */
2142 __u32 ld_padding_1; /* also fix lustre_swab_lov_desc */
2143 __u32 ld_padding_2; /* also fix lustre_swab_lov_desc */
2144 struct obd_uuid ld_uuid;
2147 #define ld_magic ld_active_tgt_count /* for swabbing from llogs */
2149 extern void lustre_swab_lov_desc (struct lov_desc *ld);
2154 /* opcodes -- MUST be distinct from OST/MDS opcodes */
2159 LDLM_BL_CALLBACK = 104,
2160 LDLM_CP_CALLBACK = 105,
2161 LDLM_GL_CALLBACK = 106,
2162 LDLM_SET_INFO = 107,
2165 #define LDLM_FIRST_OPC LDLM_ENQUEUE
2167 #define RES_NAME_SIZE 4
2168 struct ldlm_res_id {
2169 __u64 name[RES_NAME_SIZE];
2172 extern void lustre_swab_ldlm_res_id (struct ldlm_res_id *id);
2188 #define LCK_MODE_NUM 8
2198 #define LDLM_MIN_TYPE LDLM_PLAIN
2200 struct ldlm_extent {
2206 static inline int ldlm_extent_overlap(struct ldlm_extent *ex1,
2207 struct ldlm_extent *ex2)
2209 return (ex1->start <= ex2->end) && (ex2->start <= ex1->end);
2212 struct ldlm_inodebits {
2219 __u64 blocking_export; /* not actually used over the wire */
2220 __u32 blocking_pid; /* not actually used over the wire */
2224 /* it's important that the fields of the ldlm_extent structure match
2225 * the first fields of the ldlm_flock structure because there is only
2226 * one ldlm_swab routine to process the ldlm_policy_data_t union. if
2227 * this ever changes we will need to swab the union differently based
2228 * on the resource type. */
2231 struct ldlm_extent l_extent;
2232 struct ldlm_flock l_flock;
2233 struct ldlm_inodebits l_inodebits;
2234 } ldlm_policy_data_t;
2236 extern void lustre_swab_ldlm_policy_data (ldlm_policy_data_t *d);
2238 struct ldlm_intent {
2242 extern void lustre_swab_ldlm_intent (struct ldlm_intent *i);
2244 struct ldlm_resource_desc {
2245 ldlm_type_t lr_type;
2246 __u32 lr_padding; /* also fix lustre_swab_ldlm_resource_desc */
2247 struct ldlm_res_id lr_name;
2250 extern void lustre_swab_ldlm_resource_desc (struct ldlm_resource_desc *r);
2252 struct ldlm_lock_desc {
2253 struct ldlm_resource_desc l_resource;
2254 ldlm_mode_t l_req_mode;
2255 ldlm_mode_t l_granted_mode;
2256 ldlm_policy_data_t l_policy_data;
2259 extern void lustre_swab_ldlm_lock_desc (struct ldlm_lock_desc *l);
2261 #define LDLM_LOCKREQ_HANDLES 2
2262 #define LDLM_ENQUEUE_CANCEL_OFF 1
2264 struct ldlm_request {
2267 struct ldlm_lock_desc lock_desc;
2268 struct lustre_handle lock_handle[LDLM_LOCKREQ_HANDLES];
2271 extern void lustre_swab_ldlm_request (struct ldlm_request *rq);
2273 /* If LDLM_ENQUEUE, 1 slot is already occupied, 1 is available.
2274 * Otherwise, 2 are available. */
2275 #define ldlm_request_bufsize(count,type) \
2277 int _avail = LDLM_LOCKREQ_HANDLES; \
2278 _avail -= (type == LDLM_ENQUEUE ? LDLM_ENQUEUE_CANCEL_OFF : 0); \
2279 sizeof(struct ldlm_request) + \
2280 (count > _avail ? count - _avail : 0) * \
2281 sizeof(struct lustre_handle); \
2286 __u32 lock_padding; /* also fix lustre_swab_ldlm_reply */
2287 struct ldlm_lock_desc lock_desc;
2288 struct lustre_handle lock_handle;
2289 __u64 lock_policy_res1;
2290 __u64 lock_policy_res2;
2293 extern void lustre_swab_ldlm_reply (struct ldlm_reply *r);
2296 * Opcodes for mountconf (mgs and mgc)
2301 MGS_EXCEPTION, /* node died, etc. */
2302 MGS_TARGET_REG, /* whenever target starts up */
2307 #define MGS_FIRST_OPC MGS_CONNECT
2309 #define MGS_PARAM_MAXLEN 1024
2310 #define KEY_SET_INFO "set_info"
2312 struct mgs_send_param {
2313 char mgs_param[MGS_PARAM_MAXLEN];
2316 /* We pass this info to the MGS so it can write config logs */
2317 #define MTI_NAME_MAXLEN 64
2318 #define MTI_PARAM_MAXLEN 4096
2319 #define MTI_NIDS_MAX 32
2320 struct mgs_target_info {
2321 __u32 mti_lustre_ver;
2322 __u32 mti_stripe_index;
2323 __u32 mti_config_ver;
2325 __u32 mti_nid_count;
2326 __u32 padding; /* 64 bit align */
2327 char mti_fsname[MTI_NAME_MAXLEN];
2328 char mti_svname[MTI_NAME_MAXLEN];
2329 char mti_uuid[sizeof(struct obd_uuid)];
2330 __u64 mti_nids[MTI_NIDS_MAX]; /* host nids (lnet_nid_t)*/
2331 char mti_params[MTI_PARAM_MAXLEN];
2334 extern void lustre_swab_mgs_target_info(struct mgs_target_info *oinfo);
2336 /* Config marker flags (in config log) */
2337 #define CM_START 0x01
2339 #define CM_SKIP 0x04
2340 #define CM_UPGRADE146 0x08
2341 #define CM_EXCLUDE 0x10
2342 #define CM_START_SKIP (CM_START | CM_SKIP)
2345 __u32 cm_step; /* aka config version */
2347 __u32 cm_vers; /* lustre release version number */
2348 __u32 padding; /* 64 bit align */
2349 obd_time cm_createtime; /*when this record was first created */
2350 obd_time cm_canceltime; /*when this record is no longer valid*/
2351 char cm_tgtname[MTI_NAME_MAXLEN];
2352 char cm_comment[MTI_NAME_MAXLEN];
2355 extern void lustre_swab_cfg_marker(struct cfg_marker *marker,
2356 int swab, int size);
2359 * Opcodes for multiple servers.
2368 #define OBD_FIRST_OPC OBD_PING
2370 /* catalog of log objects */
2372 /** Identifier for a single log object */
2377 } __attribute__((packed));
2379 /** Records written to the CATALOGS list */
2380 #define CATLIST "CATALOGS"
2382 struct llog_logid lci_logid;
2386 } __attribute__((packed));
2388 /* Log data record types - there is no specific reason that these need to
2389 * be related to the RPC opcodes, but no reason not to (may be handy later?)
2391 #define LLOG_OP_MAGIC 0x10600000
2392 #define LLOG_OP_MASK 0xfff00000
2395 LLOG_PAD_MAGIC = LLOG_OP_MAGIC | 0x00000,
2396 OST_SZ_REC = LLOG_OP_MAGIC | 0x00f00,
2397 OST_RAID1_REC = LLOG_OP_MAGIC | 0x01000,
2398 MDS_UNLINK_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) | REINT_UNLINK,
2399 MDS_SETATTR_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) | REINT_SETATTR,
2400 MDS_SETATTR64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) | REINT_SETATTR,
2401 OBD_CFG_REC = LLOG_OP_MAGIC | 0x20000,
2402 PTL_CFG_REC = LLOG_OP_MAGIC | 0x30000, /* obsolete */
2403 LLOG_GEN_REC = LLOG_OP_MAGIC | 0x40000,
2404 LLOG_JOIN_REC = LLOG_OP_MAGIC | 0x50000, /* obsolete */
2405 CHANGELOG_REC = LLOG_OP_MAGIC | 0x60000,
2406 CHANGELOG_USER_REC = LLOG_OP_MAGIC | 0x70000,
2407 LLOG_HDR_MAGIC = LLOG_OP_MAGIC | 0x45539,
2408 LLOG_LOGID_MAGIC = LLOG_OP_MAGIC | 0x4553b,
2412 * for now, continue to support old pad records which have 0 for their
2413 * type but still need to be swabbed for their length
2415 #define LLOG_REC_HDR_NEEDS_SWABBING(r) \
2416 (((r)->lrh_type & __swab32(LLOG_OP_MASK)) == \
2417 __swab32(LLOG_OP_MAGIC) || \
2418 (((r)->lrh_type == 0) && ((r)->lrh_len > LLOG_CHUNK_SIZE)))
2420 /** Log record header - stored in little endian order.
2421 * Each record must start with this struct, end with a llog_rec_tail,
2422 * and be a multiple of 256 bits in size.
2424 struct llog_rec_hdr {
2431 struct llog_rec_tail {
2436 struct llog_logid_rec {
2437 struct llog_rec_hdr lid_hdr;
2438 struct llog_logid lid_id;
2444 struct llog_rec_tail lid_tail;
2445 } __attribute__((packed));
2447 struct llog_create_rec {
2448 struct llog_rec_hdr lcr_hdr;
2449 struct ll_fid lcr_fid;
2453 struct llog_rec_tail lcr_tail;
2454 } __attribute__((packed));
2456 struct llog_orphan_rec {
2457 struct llog_rec_hdr lor_hdr;
2461 struct llog_rec_tail lor_tail;
2462 } __attribute__((packed));
2464 struct llog_unlink_rec {
2465 struct llog_rec_hdr lur_hdr;
2468 obd_count lur_count;
2469 struct llog_rec_tail lur_tail;
2470 } __attribute__((packed));
2472 struct llog_setattr_rec {
2473 struct llog_rec_hdr lsr_hdr;
2479 struct llog_rec_tail lsr_tail;
2480 } __attribute__((packed));
2482 struct llog_setattr64_rec {
2483 struct llog_rec_hdr lsr_hdr;
2491 struct llog_rec_tail lsr_tail;
2492 } __attribute__((packed));
2494 struct llog_size_change_rec {
2495 struct llog_rec_hdr lsc_hdr;
2496 struct ll_fid lsc_fid;
2499 struct llog_rec_tail lsc_tail;
2500 } __attribute__((packed));
2502 #define CHANGELOG_MAGIC 0xca103000
2504 /** \a changelog_rec_type's that can't be masked */
2505 #define CHANGELOG_MINMASK (1 << CL_MARK)
2506 /** bits covering all \a changelog_rec_type's */
2507 #define CHANGELOG_ALLMASK 0XFFFFFFFF
2508 /** default \a changelog_rec_type mask */
2509 #define CHANGELOG_DEFMASK CHANGELOG_ALLMASK & ~(1 << CL_ATIME)
2511 /* changelog llog name, needed by client replicators */
2512 #define CHANGELOG_CATALOG "changelog_catalog"
2514 struct changelog_setinfo {
2517 } __attribute__((packed));
2519 /** changelog record */
2520 struct llog_changelog_rec {
2521 struct llog_rec_hdr cr_hdr;
2522 struct changelog_rec cr;
2523 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
2524 } __attribute__((packed));
2526 #define CHANGELOG_USER_PREFIX "cl"
2528 struct llog_changelog_user_rec {
2529 struct llog_rec_hdr cur_hdr;
2533 struct llog_rec_tail cur_tail;
2534 } __attribute__((packed));
2539 } __attribute__((packed));
2541 struct llog_gen_rec {
2542 struct llog_rec_hdr lgr_hdr;
2543 struct llog_gen lgr_gen;
2544 struct llog_rec_tail lgr_tail;
2546 /* On-disk header structure of each log object, stored in little endian order */
2547 #define LLOG_CHUNK_SIZE 8192
2548 #define LLOG_HEADER_SIZE (96)
2549 #define LLOG_BITMAP_BYTES (LLOG_CHUNK_SIZE - LLOG_HEADER_SIZE)
2551 #define LLOG_MIN_REC_SIZE (24) /* round(llog_rec_hdr + llog_rec_tail) */
2553 /* flags for the logs */
2554 #define LLOG_F_ZAP_WHEN_EMPTY 0x1
2555 #define LLOG_F_IS_CAT 0x2
2556 #define LLOG_F_IS_PLAIN 0x4
2558 struct llog_log_hdr {
2559 struct llog_rec_hdr llh_hdr;
2560 obd_time llh_timestamp;
2562 __u32 llh_bitmap_offset;
2566 /* for a catalog the first plain slot is next to it */
2567 struct obd_uuid llh_tgtuuid;
2568 __u32 llh_reserved[LLOG_HEADER_SIZE/sizeof(__u32) - 23];
2569 __u32 llh_bitmap[LLOG_BITMAP_BYTES/sizeof(__u32)];
2570 struct llog_rec_tail llh_tail;
2571 } __attribute__((packed));
2573 #define LLOG_BITMAP_SIZE(llh) ((llh->llh_hdr.lrh_len - \
2574 llh->llh_bitmap_offset - \
2575 sizeof(llh->llh_tail)) * 8)
2577 /** log cookies are used to reference a specific log file and a record therein */
2578 struct llog_cookie {
2579 struct llog_logid lgc_lgl;
2583 } __attribute__((packed));
2585 /** llog protocol */
2586 enum llogd_rpc_ops {
2587 LLOG_ORIGIN_HANDLE_CREATE = 501,
2588 LLOG_ORIGIN_HANDLE_NEXT_BLOCK = 502,
2589 LLOG_ORIGIN_HANDLE_READ_HEADER = 503,
2590 LLOG_ORIGIN_HANDLE_WRITE_REC = 504,
2591 LLOG_ORIGIN_HANDLE_CLOSE = 505,
2592 LLOG_ORIGIN_CONNECT = 506,
2593 LLOG_CATINFO = 507, /* for lfs catinfo */
2594 LLOG_ORIGIN_HANDLE_PREV_BLOCK = 508,
2595 LLOG_ORIGIN_HANDLE_DESTROY = 509, /* for destroy llog object*/
2597 LLOG_FIRST_OPC = LLOG_ORIGIN_HANDLE_CREATE
2601 struct llog_logid lgd_logid;
2603 __u32 lgd_llh_flags;
2605 __u32 lgd_saved_index;
2607 __u64 lgd_cur_offset;
2608 } __attribute__((packed));
2610 struct llogd_conn_body {
2611 struct llog_gen lgdc_gen;
2612 struct llog_logid lgdc_logid;
2613 __u32 lgdc_ctxt_idx;
2614 } __attribute__((packed));
2616 /* Note: 64-bit types are 64-bit aligned in structure */
2618 obd_valid o_valid; /* hot fields in this obdo */
2620 obd_id o_parent_seq;
2621 obd_size o_size; /* o_size-o_blocks == ost_lvb */
2625 obd_blocks o_blocks; /* brw: cli sent cached bytes */
2628 /* 32-bit fields start here: keep an even number of them via padding */
2629 obd_blksize o_blksize; /* optimal IO blocksize */
2630 obd_mode o_mode; /* brw: cli sent cache remain */
2634 obd_count o_nlink; /* brw: checksum */
2635 obd_count o_parent_oid;
2636 obd_count o_misc; /* brw: o_dropped */
2637 __u64 o_ioepoch; /* epoch in ost writes */
2638 __u32 o_stripe_idx; /* holds stripe idx */
2640 struct lustre_handle o_handle; /* brw: lock handle to prolong locks */
2641 struct llog_cookie o_lcookie; /* destroy: unlink cookie from MDS */
2651 #define o_id o_oi.oi_id
2652 #define o_seq o_oi.oi_seq
2653 #define o_dirty o_blocks
2654 #define o_undirty o_mode
2655 #define o_dropped o_misc
2656 #define o_cksum o_nlink
2658 static inline void lustre_set_wire_obdo(struct obdo *wobdo, struct obdo *lobdo)
2660 memcpy(wobdo, lobdo, sizeof(*lobdo));
2661 wobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
2664 static inline void lustre_get_wire_obdo(struct obdo *lobdo, struct obdo *wobdo)
2666 obd_flag local_flags = 0;
2668 if (lobdo->o_valid & OBD_MD_FLFLAGS)
2669 local_flags = lobdo->o_flags & OBD_FL_LOCAL_MASK;
2671 LASSERT(!(wobdo->o_flags & OBD_FL_LOCAL_MASK));
2673 memcpy(lobdo, wobdo, sizeof(*lobdo));
2674 if (local_flags != 0) {
2675 lobdo->o_valid |= OBD_MD_FLFLAGS;
2676 lobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
2677 lobdo->o_flags |= local_flags;
2681 extern void lustre_swab_obdo (struct obdo *o);
2683 /* request structure for OST's */
2688 /* Key for FIEMAP to be used in get_info calls */
2689 struct ll_fiemap_info_key {
2692 struct ll_user_fiemap fiemap;
2695 extern void lustre_swab_ost_body (struct ost_body *b);
2696 extern void lustre_swab_ost_last_id(obd_id *id);
2697 extern void lustre_swab_fiemap(struct ll_user_fiemap *fiemap);
2699 extern void lustre_swab_lov_user_md_v1(struct lov_user_md_v1 *lum);
2700 extern void lustre_swab_lov_user_md_v3(struct lov_user_md_v3 *lum);
2701 extern void lustre_swab_lov_user_md_objects(struct lov_user_ost_data *lod,
2703 extern void lustre_swab_lov_mds_md(struct lov_mds_md *lmm);
2706 extern void lustre_swab_llogd_body (struct llogd_body *d);
2707 extern void lustre_swab_llog_hdr (struct llog_log_hdr *h);
2708 extern void lustre_swab_llogd_conn_body (struct llogd_conn_body *d);
2709 extern void lustre_swab_llog_rec(struct llog_rec_hdr *rec,
2710 struct llog_rec_tail *tail);
2713 extern void lustre_swab_lustre_cfg(struct lustre_cfg *lcfg);
2715 /* Functions for dumping PTLRPC fields */
2716 void dump_rniobuf(struct niobuf_remote *rnb);
2717 void dump_ioo(struct obd_ioobj *nb);
2718 void dump_obdo(struct obdo *oa);
2719 void dump_ost_body(struct ost_body *ob);
2720 void dump_rcs(__u32 *rc);
2722 /* this will be used when OBD_CONNECT_CHANGE_QS is set */
2725 * ID appiles to (uid, gid)
2729 * LQUOTA_FLAGS_* affect the responding bits
2733 * acquire/release count (bytes for block quota)
2737 * when a master returns the reply to a slave, it will
2738 * contain the current corresponding qunit size
2744 #define QDATA_IS_GRP(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_GRP)
2745 #define QDATA_IS_BLK(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_BLK)
2746 #define QDATA_IS_ADJBLK(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_ADJBLK)
2747 #define QDATA_IS_ADJINO(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_ADJINO)
2748 #define QDATA_IS_CHANGE_QS(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_CHG_QS)
2750 #define QDATA_SET_GRP(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_GRP)
2751 #define QDATA_SET_BLK(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_BLK)
2752 #define QDATA_SET_ADJBLK(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_ADJBLK)
2753 #define QDATA_SET_ADJINO(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_ADJINO)
2754 #define QDATA_SET_CHANGE_QS(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_CHG_QS)
2756 #define QDATA_CLR_GRP(qdata) ((qdata)->qd_flags &= ~LQUOTA_FLAGS_GRP)
2757 #define QDATA_CLR_CHANGE_QS(qdata) ((qdata)->qd_flags &= ~LQUOTA_FLAGS_CHG_QS)
2759 extern void lustre_swab_qdata(struct qunit_data *d);
2760 extern struct qunit_data *quota_get_qdata(void *req, int is_req, int is_exp);
2761 extern int quota_copy_qdata(void *request, struct qunit_data *qdata,
2762 int is_req, int is_exp);
2769 #define QUOTA_FIRST_OPC QUOTA_DQACQ
2771 #define QUOTA_REQUEST 1
2772 #define QUOTA_REPLY 0
2773 #define QUOTA_EXPORT 1
2774 #define QUOTA_IMPORT 0
2776 /* quota check function */
2777 #define QUOTA_RET_OK 0 /**< return successfully */
2778 #define QUOTA_RET_NOQUOTA 1 /**< not support quota */
2779 #define QUOTA_RET_NOLIMIT 2 /**< quota limit isn't set */
2780 #define QUOTA_RET_ACQUOTA 4 /**< need to acquire extra quota */
2783 /* security opcodes */
2786 SEC_CTX_INIT_CONT = 802,
2789 SEC_FIRST_OPC = SEC_CTX_INIT
2793 * capa related definitions
2795 #define CAPA_HMAC_MAX_LEN 64
2796 #define CAPA_HMAC_KEY_MAX_LEN 56
2798 /* NB take care when changing the sequence of elements this struct,
2799 * because the offset info is used in find_capa() */
2800 struct lustre_capa {
2801 struct lu_fid lc_fid; /** fid */
2802 __u64 lc_opc; /** operations allowed */
2803 __u64 lc_uid; /** file owner */
2804 __u64 lc_gid; /** file group */
2805 __u32 lc_flags; /** HMAC algorithm & flags */
2806 __u32 lc_keyid; /** key# used for the capability */
2807 __u32 lc_timeout; /** capa timeout value (sec) */
2808 __u32 lc_expiry; /** expiry time (sec) */
2809 __u8 lc_hmac[CAPA_HMAC_MAX_LEN]; /** HMAC */
2810 } __attribute__((packed));
2812 extern void lustre_swab_lustre_capa(struct lustre_capa *c);
2814 /** lustre_capa::lc_opc */
2816 CAPA_OPC_BODY_WRITE = 1<<0, /**< write object data */
2817 CAPA_OPC_BODY_READ = 1<<1, /**< read object data */
2818 CAPA_OPC_INDEX_LOOKUP = 1<<2, /**< lookup object fid */
2819 CAPA_OPC_INDEX_INSERT = 1<<3, /**< insert object fid */
2820 CAPA_OPC_INDEX_DELETE = 1<<4, /**< delete object fid */
2821 CAPA_OPC_OSS_WRITE = 1<<5, /**< write oss object data */
2822 CAPA_OPC_OSS_READ = 1<<6, /**< read oss object data */
2823 CAPA_OPC_OSS_TRUNC = 1<<7, /**< truncate oss object */
2824 CAPA_OPC_OSS_DESTROY = 1<<8, /**< destroy oss object */
2825 CAPA_OPC_META_WRITE = 1<<9, /**< write object meta data */
2826 CAPA_OPC_META_READ = 1<<10, /**< read object meta data */
2829 #define CAPA_OPC_OSS_RW (CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE)
2830 #define CAPA_OPC_MDS_ONLY \
2831 (CAPA_OPC_BODY_WRITE | CAPA_OPC_BODY_READ | CAPA_OPC_INDEX_LOOKUP | \
2832 CAPA_OPC_INDEX_INSERT | CAPA_OPC_INDEX_DELETE)
2833 #define CAPA_OPC_OSS_ONLY \
2834 (CAPA_OPC_OSS_WRITE | CAPA_OPC_OSS_READ | CAPA_OPC_OSS_TRUNC | \
2835 CAPA_OPC_OSS_DESTROY)
2836 #define CAPA_OPC_MDS_DEFAULT ~CAPA_OPC_OSS_ONLY
2837 #define CAPA_OPC_OSS_DEFAULT ~(CAPA_OPC_MDS_ONLY | CAPA_OPC_OSS_ONLY)
2839 /* MDS capability covers object capability for operations of body r/w
2840 * (dir readpage/sendpage), index lookup/insert/delete and meta data r/w,
2841 * while OSS capability only covers object capability for operations of
2842 * oss data(file content) r/w/truncate.
2844 static inline int capa_for_mds(struct lustre_capa *c)
2846 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) != 0;
2849 static inline int capa_for_oss(struct lustre_capa *c)
2851 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) == 0;
2854 /* lustre_capa::lc_hmac_alg */
2856 CAPA_HMAC_ALG_SHA1 = 1, /**< sha1 algorithm */
2860 #define CAPA_FL_MASK 0x00ffffff
2861 #define CAPA_HMAC_ALG_MASK 0xff000000
2863 struct lustre_capa_key {
2864 __u64 lk_seq; /**< mds# */
2865 __u32 lk_keyid; /**< key# */
2867 __u8 lk_key[CAPA_HMAC_KEY_MAX_LEN]; /**< key */
2868 } __attribute__((packed));
2870 extern void lustre_swab_lustre_capa_key(struct lustre_capa_key *k);
2872 /** The link ea holds 1 \a link_ea_entry for each hardlink */
2873 #define LINK_EA_MAGIC 0x11EAF1DFUL
2874 struct link_ea_header {
2877 __u64 leh_len; /* total size */
2883 /** Hardlink data is name and parent fid.
2884 * Stored in this crazy struct for maximum packing and endian-neutrality
2886 struct link_ea_entry {
2887 /** __u16 stored big-endian, unaligned */
2888 unsigned char lee_reclen[2];
2889 unsigned char lee_parent_fid[sizeof(struct lu_fid)];
2891 }__attribute__((packed));
2893 /** fid2path request/reply structure */
2894 struct getinfo_fid2path {
2895 struct lu_fid gf_fid;
2900 } __attribute__((packed));
2902 void lustre_swab_fid2path (struct getinfo_fid2path *gf);