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 /* sequence for local pre-defined FIDs listed in local_oid */
422 FID_SEQ_LOCAL_FILE = 0x200000001ULL,
423 FID_SEQ_DOT_LUSTRE = 0x200000002ULL,
424 /* XXX 0x200000003ULL is reserved for FID_SEQ_LLOG_OBJ */
425 /* sequence is used for local named objects FIDs generated
426 * by local_object_storage library */
427 FID_SEQ_SPECIAL = 0x200000004ULL,
428 FID_SEQ_QUOTA = 0x200000005ULL,
429 FID_SEQ_QUOTA_GLB = 0x200000006ULL,
430 FID_SEQ_LOCAL_NAME = 0x200000007ULL,
431 FID_SEQ_NORMAL = 0x200000400ULL,
432 FID_SEQ_LOV_DEFAULT= 0xffffffffffffffffULL
435 #define OBIF_OID_MAX_BITS 32
436 #define OBIF_MAX_OID (1ULL << OBIF_OID_MAX_BITS)
437 #define OBIF_OID_MASK ((1ULL << OBIF_OID_MAX_BITS) - 1)
438 #define IDIF_OID_MAX_BITS 48
439 #define IDIF_MAX_OID (1ULL << IDIF_OID_MAX_BITS)
440 #define IDIF_OID_MASK ((1ULL << IDIF_OID_MAX_BITS) - 1)
442 /** OID for FID_SEQ_SPECIAL */
444 /* Big Filesystem Lock to serialize rename operations */
445 FID_OID_SPECIAL_BFL = 1UL,
448 /** OID for FID_SEQ_DOT_LUSTRE */
449 enum dot_lustre_oid {
450 FID_OID_DOT_LUSTRE = 1UL,
451 FID_OID_DOT_LUSTRE_OBF = 2UL,
454 static inline int fid_seq_is_mdt0(obd_seq seq)
456 return (seq == FID_SEQ_OST_MDT0);
459 static inline int fid_seq_is_cmd(const __u64 seq)
461 return (seq >= FID_SEQ_OST_MDT1 && seq <= FID_SEQ_OST_MAX);
464 static inline int fid_seq_is_mdt(const __u64 seq)
466 return seq == FID_SEQ_OST_MDT0 ||
467 (seq >= FID_SEQ_OST_MDT1 && seq <= FID_SEQ_OST_MAX);
470 static inline int fid_seq_is_rsvd(const __u64 seq)
472 return (seq > FID_SEQ_OST_MDT0 && seq <= FID_SEQ_RSVD);
475 static inline int fid_is_mdt0(const struct lu_fid *fid)
477 return fid_seq_is_mdt0(fid_seq(fid));
481 * Check if a fid is igif or not.
482 * \param fid the fid to be tested.
483 * \return true if the fid is a igif; otherwise false.
485 static inline int fid_seq_is_igif(const __u64 seq)
487 return seq >= FID_SEQ_IGIF && seq <= FID_SEQ_IGIF_MAX;
490 static inline int fid_is_igif(const struct lu_fid *fid)
492 return fid_seq_is_igif(fid_seq(fid));
496 * Check if a fid is idif or not.
497 * \param fid the fid to be tested.
498 * \return true if the fid is a idif; otherwise false.
500 static inline int fid_seq_is_idif(const __u64 seq)
502 return seq >= FID_SEQ_IDIF && seq <= FID_SEQ_IDIF_MAX;
505 static inline int fid_is_idif(const struct lu_fid *fid)
507 return fid_seq_is_idif(fid_seq(fid));
515 static inline int fid_seq_is_norm(const __u64 seq)
517 return (seq >= FID_SEQ_NORMAL);
520 static inline int fid_is_norm(const struct lu_fid *fid)
522 return fid_seq_is_norm(fid_seq(fid));
525 /* convert an OST objid into an IDIF FID SEQ number */
526 static inline obd_seq fid_idif_seq(obd_id id, __u32 ost_idx)
528 return FID_SEQ_IDIF | (ost_idx << 16) | ((id >> 32) & 0xffff);
531 /* convert a packed IDIF FID into an OST objid */
532 static inline obd_id fid_idif_id(obd_seq seq, __u32 oid, __u32 ver)
534 return ((__u64)ver << 48) | ((seq & 0xffff) << 32) | oid;
537 /* unpack an ostid (id/seq) from a wire/disk structure into an IDIF FID */
538 static inline void ostid_idif_unpack(struct ost_id *ostid,
539 struct lu_fid *fid, __u32 ost_idx)
541 fid->f_seq = fid_idif_seq(ostid->oi_id, ost_idx);
542 fid->f_oid = ostid->oi_id; /* truncate to 32 bits by assignment */
543 fid->f_ver = ostid->oi_id >> 48; /* in theory, not currently used */
546 /* unpack an ostid (id/seq) from a wire/disk structure into a non-IDIF FID */
547 static inline void ostid_fid_unpack(struct ost_id *ostid, struct lu_fid *fid)
549 fid->f_seq = ostid->oi_seq;
550 fid->f_oid = ostid->oi_id; /* truncate to 32 bits by assignment */
551 fid->f_ver = ostid->oi_id >> 32; /* in theory, not currently used */
554 /* Unpack an OST object id/seq (group) into a FID. This is needed for
555 * converting all obdo, lmm, lsm, etc. 64-bit id/seq pairs into proper
556 * FIDs. Note that if an id/seq is already in FID/IDIF format it will
557 * be passed through unchanged. Only legacy OST objects in "group 0"
558 * will be mapped into the IDIF namespace so that they can fit into the
559 * struct lu_fid fields without loss. For reference see:
560 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs
562 static inline int fid_ostid_unpack(struct lu_fid *fid, struct ost_id *ostid,
565 if (ost_idx > 0xffff) {
566 CERROR("bad ost_idx, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
567 ostid->oi_seq, ostid->oi_id, ost_idx);
571 if (fid_seq_is_mdt0(ostid->oi_seq)) {
572 /* This is a "legacy" (old 1.x/2.early) OST object in "group 0"
573 * that we map into the IDIF namespace. It allows up to 2^48
574 * objects per OST, as this is the object namespace that has
575 * been in production for years. This can handle create rates
576 * of 1M objects/s/OST for 9 years, or combinations thereof. */
577 if (ostid->oi_id >= IDIF_MAX_OID) {
578 CERROR("bad MDT0 id, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
579 ostid->oi_seq, ostid->oi_id, ost_idx);
582 ostid_idif_unpack(ostid, fid, ost_idx);
584 } else if (fid_seq_is_rsvd(ostid->oi_seq)) {
585 /* These are legacy OST objects for LLOG/ECHO and CMD testing.
586 * We only support 2^32 objects in these groups, and cannot
587 * uniquely identify them in the system (i.e. they are the
588 * duplicated on all OSTs), but this is not strictly required
589 * for the old object protocol, which has a separate ost_idx. */
590 if (ostid->oi_id >= 0xffffffffULL) {
591 CERROR("bad RSVD id, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
592 ostid->oi_seq, ostid->oi_id, ost_idx);
595 ostid_fid_unpack(ostid, fid);
597 } else if (unlikely(fid_seq_is_igif(ostid->oi_seq))) {
598 /* This is an MDT inode number, which should never collide with
599 * proper OST object IDs, and is probably a broken filesystem */
600 CERROR("bad IGIF, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
601 ostid->oi_seq, ostid->oi_id, ost_idx);
604 } else /* if (fid_seq_is_idif(seq) || fid_seq_is_norm(seq)) */ {
605 /* This is either an IDIF object, which identifies objects across
606 * all OSTs, or a regular FID. The IDIF namespace maps legacy
607 * OST objects into the FID namespace. In both cases, we just
608 * pass the FID through, no conversion needed. */
609 ostid_fid_unpack(ostid, fid);
615 /* pack an IDIF FID into an ostid (id/seq) for the wire/disk */
616 static inline void ostid_idif_pack(const struct lu_fid *fid,
617 struct ost_id *ostid)
619 ostid->oi_seq = FID_SEQ_OST_MDT0;
620 ostid->oi_id = fid_idif_id(fid->f_seq, fid->f_oid, fid->f_ver);
623 /* pack a non-IDIF FID into an ostid (id/seq) for the wire/disk */
624 static inline void ostid_fid_pack(const struct lu_fid *fid,
625 struct ost_id *ostid)
627 ostid->oi_seq = fid_seq(fid);
628 ostid->oi_id = fid_ver_oid(fid);
631 /* pack any OST FID into an ostid (id/seq) for the wire/disk */
632 static inline int fid_ostid_pack(const struct lu_fid *fid,
633 struct ost_id *ostid)
635 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
636 CERROR("bad IGIF, "DFID"\n", PFID(fid));
640 if (fid_is_idif(fid))
641 ostid_idif_pack(fid, ostid);
643 ostid_fid_pack(fid, ostid);
648 /* extract OST sequence (group) from a wire ost_id (id/seq) pair */
649 static inline obd_seq ostid_seq(struct ost_id *ostid)
651 if (unlikely(fid_seq_is_igif(ostid->oi_seq)))
652 CWARN("bad IGIF, oi_seq: "LPU64" oi_id: "LPX64"\n",
653 ostid->oi_seq, ostid->oi_id);
655 if (unlikely(fid_seq_is_idif(ostid->oi_seq)))
656 return FID_SEQ_OST_MDT0;
658 return ostid->oi_seq;
661 /* extract OST objid from a wire ost_id (id/seq) pair */
662 static inline obd_id ostid_id(struct ost_id *ostid)
664 if (ostid->oi_seq == FID_SEQ_OST_MDT0)
665 return ostid->oi_id & IDIF_OID_MASK;
667 if (fid_seq_is_rsvd(ostid->oi_seq))
668 return ostid->oi_id & OBIF_OID_MASK;
670 if (fid_seq_is_idif(ostid->oi_seq))
671 return fid_idif_id(ostid->oi_seq, ostid->oi_id, 0);
677 * Get inode number from a igif.
678 * \param fid a igif to get inode number from.
679 * \return inode number for the igif.
681 static inline ino_t lu_igif_ino(const struct lu_fid *fid)
687 * Build igif from the inode number/generation.
689 #define LU_IGIF_BUILD(fid, ino, gen) \
695 static inline void lu_igif_build(struct lu_fid *fid, __u32 ino, __u32 gen)
697 LU_IGIF_BUILD(fid, ino, gen);
698 LASSERT(fid_is_igif(fid));
702 * Get inode generation from a igif.
703 * \param fid a igif to get inode generation from.
704 * \return inode generation for the igif.
706 static inline __u32 lu_igif_gen(const struct lu_fid *fid)
712 * Fids are transmitted across network (in the sender byte-ordering),
713 * and stored on disk in big-endian order.
715 static inline void fid_cpu_to_le(struct lu_fid *dst, const struct lu_fid *src)
717 /* check that all fields are converted */
718 CLASSERT(sizeof *src ==
719 sizeof fid_seq(src) +
720 sizeof fid_oid(src) + sizeof fid_ver(src));
721 LASSERTF(fid_is_igif(src) || fid_ver(src) == 0, DFID"\n", PFID(src));
722 dst->f_seq = cpu_to_le64(fid_seq(src));
723 dst->f_oid = cpu_to_le32(fid_oid(src));
724 dst->f_ver = cpu_to_le32(fid_ver(src));
727 static inline void fid_le_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
729 /* check that all fields are converted */
730 CLASSERT(sizeof *src ==
731 sizeof fid_seq(src) +
732 sizeof fid_oid(src) + sizeof fid_ver(src));
733 dst->f_seq = le64_to_cpu(fid_seq(src));
734 dst->f_oid = le32_to_cpu(fid_oid(src));
735 dst->f_ver = le32_to_cpu(fid_ver(src));
736 LASSERTF(fid_is_igif(dst) || fid_ver(dst) == 0, DFID"\n", PFID(dst));
739 static inline void fid_cpu_to_be(struct lu_fid *dst, const struct lu_fid *src)
741 /* check that all fields are converted */
742 CLASSERT(sizeof *src ==
743 sizeof fid_seq(src) +
744 sizeof fid_oid(src) + sizeof fid_ver(src));
745 LASSERTF(fid_is_igif(src) || fid_ver(src) == 0, DFID"\n", PFID(src));
746 dst->f_seq = cpu_to_be64(fid_seq(src));
747 dst->f_oid = cpu_to_be32(fid_oid(src));
748 dst->f_ver = cpu_to_be32(fid_ver(src));
751 static inline void fid_be_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
753 /* check that all fields are converted */
754 CLASSERT(sizeof *src ==
755 sizeof fid_seq(src) +
756 sizeof fid_oid(src) + sizeof fid_ver(src));
757 dst->f_seq = be64_to_cpu(fid_seq(src));
758 dst->f_oid = be32_to_cpu(fid_oid(src));
759 dst->f_ver = be32_to_cpu(fid_ver(src));
760 LASSERTF(fid_is_igif(dst) || fid_ver(dst) == 0, DFID"\n", PFID(dst));
763 static inline int fid_is_sane(const struct lu_fid *fid)
767 ((fid_seq(fid) >= FID_SEQ_START && fid_oid(fid) != 0
768 && fid_ver(fid) == 0) ||
769 fid_is_igif(fid) || fid_seq_is_rsvd(fid_seq(fid)));
772 static inline int fid_is_zero(const struct lu_fid *fid)
774 return fid_seq(fid) == 0 && fid_oid(fid) == 0;
777 extern void lustre_swab_lu_fid(struct lu_fid *fid);
778 extern void lustre_swab_lu_seq_range(struct lu_seq_range *range);
780 static inline int lu_fid_eq(const struct lu_fid *f0,
781 const struct lu_fid *f1)
783 /* Check that there is no alignment padding. */
784 CLASSERT(sizeof *f0 ==
785 sizeof f0->f_seq + sizeof f0->f_oid + sizeof f0->f_ver);
786 LASSERTF((fid_is_igif(f0) || fid_is_idif(f0)) ||
787 fid_ver(f0) == 0, DFID, PFID(f0));
788 LASSERTF((fid_is_igif(f1) || fid_is_idif(f1)) ||
789 fid_ver(f1) == 0, DFID, PFID(f1));
790 return memcmp(f0, f1, sizeof *f0) == 0;
793 #define __diff_normalize(val0, val1) \
795 typeof(val0) __val0 = (val0); \
796 typeof(val1) __val1 = (val1); \
798 (__val0 == __val1 ? 0 : __val0 > __val1 ? +1 : -1); \
801 static inline int lu_fid_cmp(const struct lu_fid *f0,
802 const struct lu_fid *f1)
805 __diff_normalize(fid_seq(f0), fid_seq(f1)) ?:
806 __diff_normalize(fid_oid(f0), fid_oid(f1)) ?:
807 __diff_normalize(fid_ver(f0), fid_ver(f1));
812 /** \defgroup lu_dir lu_dir
816 * Enumeration of possible directory entry attributes.
818 * Attributes follow directory entry header in the order they appear in this
821 enum lu_dirent_attrs {
824 LUDA_64BITHASH = 0x0004,
828 * Layout of readdir pages, as transmitted on wire.
831 /** valid if LUDA_FID is set. */
832 struct lu_fid lde_fid;
833 /** a unique entry identifier: a hash or an offset. */
835 /** total record length, including all attributes. */
839 /** optional variable size attributes following this entry.
840 * taken from enum lu_dirent_attrs.
843 /** name is followed by the attributes indicated in ->ldp_attrs, in
844 * their natural order. After the last attribute, padding bytes are
845 * added to make ->lde_reclen a multiple of 8.
851 * Definitions of optional directory entry attributes formats.
853 * Individual attributes do not have their length encoded in a generic way. It
854 * is assumed that consumer of an attribute knows its format. This means that
855 * it is impossible to skip over an unknown attribute, except by skipping over all
856 * remaining attributes (by using ->lde_reclen), which is not too
857 * constraining, because new server versions will append new attributes at
858 * the end of an entry.
862 * Fid directory attribute: a fid of an object referenced by the entry. This
863 * will be almost always requested by the client and supplied by the server.
865 * Aligned to 8 bytes.
867 /* To have compatibility with 1.8, lets have fid in lu_dirent struct. */
872 * Aligned to 2 bytes.
879 __u64 ldp_hash_start;
883 struct lu_dirent ldp_entries[0];
886 enum lu_dirpage_flags {
888 * dirpage contains no entry.
892 * last entry's lde_hash equals ldp_hash_end.
897 static inline struct lu_dirent *lu_dirent_start(struct lu_dirpage *dp)
899 if (le32_to_cpu(dp->ldp_flags) & LDF_EMPTY)
902 return dp->ldp_entries;
905 static inline struct lu_dirent *lu_dirent_next(struct lu_dirent *ent)
907 struct lu_dirent *next;
909 if (le16_to_cpu(ent->lde_reclen) != 0)
910 next = ((void *)ent) + le16_to_cpu(ent->lde_reclen);
917 static inline int lu_dirent_calc_size(int namelen, __u16 attr)
921 if (attr & LUDA_TYPE) {
922 const unsigned align = sizeof(struct luda_type) - 1;
923 size = (sizeof(struct lu_dirent) + namelen + align) & ~align;
924 size += sizeof(struct luda_type);
926 size = sizeof(struct lu_dirent) + namelen;
928 return (size + 7) & ~7;
931 static inline int lu_dirent_size(struct lu_dirent *ent)
933 if (le16_to_cpu(ent->lde_reclen) == 0) {
934 return lu_dirent_calc_size(le16_to_cpu(ent->lde_namelen),
935 le32_to_cpu(ent->lde_attrs));
937 return le16_to_cpu(ent->lde_reclen);
940 #define MDS_DIR_END_OFF 0xfffffffffffffffeULL
943 * MDS_READPAGE page size
945 * This is the directory page size packed in MDS_READPAGE RPC.
946 * It's different than CFS_PAGE_SIZE because the client needs to
947 * access the struct lu_dirpage header packed at the beginning of
948 * the "page" and without this there isn't any way to know find the
949 * lu_dirpage header is if client and server CFS_PAGE_SIZE differ.
951 #define LU_PAGE_SHIFT 12
952 #define LU_PAGE_SIZE (1UL << LU_PAGE_SHIFT)
953 #define LU_PAGE_MASK (~(LU_PAGE_SIZE - 1))
955 #define LU_PAGE_COUNT 1 << (CFS_PAGE_SHIFT - LU_PAGE_SHIFT)
959 struct lustre_handle {
962 #define DEAD_HANDLE_MAGIC 0xdeadbeefcafebabeULL
964 static inline int lustre_handle_is_used(struct lustre_handle *lh)
966 return lh->cookie != 0ull;
969 static inline int lustre_handle_equal(const struct lustre_handle *lh1,
970 const struct lustre_handle *lh2)
972 return lh1->cookie == lh2->cookie;
975 static inline void lustre_handle_copy(struct lustre_handle *tgt,
976 struct lustre_handle *src)
978 tgt->cookie = src->cookie;
981 /* flags for lm_flags */
982 #define MSGHDR_AT_SUPPORT 0x1
983 #define MSGHDR_CKSUM_INCOMPAT18 0x2
985 #define lustre_msg lustre_msg_v2
986 /* we depend on this structure to be 8-byte aligned */
987 /* this type is only endian-adjusted in lustre_unpack_msg() */
988 struct lustre_msg_v2 {
1000 /* without gss, ptlrpc_body is put at the first buffer. */
1001 #define PTLRPC_NUM_VERSIONS 4
1002 #define JOBSTATS_JOBID_SIZE 32 /* 32 bytes string */
1003 struct ptlrpc_body_v3 {
1004 struct lustre_handle pb_handle;
1011 __u64 pb_last_committed;
1016 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1017 __u32 pb_service_time; /* for rep, actual service time */
1020 /* VBR: pre-versions */
1021 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1022 /* padding for future needs */
1023 __u64 pb_padding[4];
1024 char pb_jobid[JOBSTATS_JOBID_SIZE];
1026 #define ptlrpc_body ptlrpc_body_v3
1028 struct ptlrpc_body_v2 {
1029 struct lustre_handle pb_handle;
1036 __u64 pb_last_committed;
1041 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1042 __u32 pb_service_time; /* for rep, actual service time, also used for
1043 net_latency of req */
1046 /* VBR: pre-versions */
1047 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1048 /* padding for future needs */
1049 __u64 pb_padding[4];
1052 extern void lustre_swab_ptlrpc_body(struct ptlrpc_body *pb);
1054 /* message body offset for lustre_msg_v2 */
1055 /* ptlrpc body offset in all request/reply messages */
1056 #define MSG_PTLRPC_BODY_OFF 0
1058 /* normal request/reply message record offset */
1059 #define REQ_REC_OFF 1
1060 #define REPLY_REC_OFF 1
1062 /* ldlm request message body offset */
1063 #define DLM_LOCKREQ_OFF 1 /* lockreq offset */
1064 #define DLM_REQ_REC_OFF 2 /* normal dlm request record offset */
1066 /* ldlm intent lock message body offset */
1067 #define DLM_INTENT_IT_OFF 2 /* intent lock it offset */
1068 #define DLM_INTENT_REC_OFF 3 /* intent lock record offset */
1070 /* ldlm reply message body offset */
1071 #define DLM_LOCKREPLY_OFF 1 /* lockrep offset */
1072 #define DLM_REPLY_REC_OFF 2 /* reply record offset */
1074 /** only use in req->rq_{req,rep}_swab_mask */
1075 #define MSG_PTLRPC_HEADER_OFF 31
1077 /* Flags that are operation-specific go in the top 16 bits. */
1078 #define MSG_OP_FLAG_MASK 0xffff0000
1079 #define MSG_OP_FLAG_SHIFT 16
1081 /* Flags that apply to all requests are in the bottom 16 bits */
1082 #define MSG_GEN_FLAG_MASK 0x0000ffff
1083 #define MSG_LAST_REPLAY 0x0001
1084 #define MSG_RESENT 0x0002
1085 #define MSG_REPLAY 0x0004
1086 /* #define MSG_AT_SUPPORT 0x0008
1087 * This was used in early prototypes of adaptive timeouts, and while there
1088 * shouldn't be any users of that code there also isn't a need for using this
1089 * bits. Defer usage until at least 1.10 to avoid potential conflict. */
1090 #define MSG_DELAY_REPLAY 0x0010
1091 #define MSG_VERSION_REPLAY 0x0020
1092 #define MSG_REQ_REPLAY_DONE 0x0040
1093 #define MSG_LOCK_REPLAY_DONE 0x0080
1096 * Flags for all connect opcodes (MDS_CONNECT, OST_CONNECT)
1099 #define MSG_CONNECT_RECOVERING 0x00000001
1100 #define MSG_CONNECT_RECONNECT 0x00000002
1101 #define MSG_CONNECT_REPLAYABLE 0x00000004
1102 //#define MSG_CONNECT_PEER 0x8
1103 #define MSG_CONNECT_LIBCLIENT 0x00000010
1104 #define MSG_CONNECT_INITIAL 0x00000020
1105 #define MSG_CONNECT_ASYNC 0x00000040
1106 #define MSG_CONNECT_NEXT_VER 0x00000080 /* use next version of lustre_msg */
1107 #define MSG_CONNECT_TRANSNO 0x00000100 /* report transno */
1110 #define OBD_CONNECT_RDONLY 0x1ULL /*client has read-only access*/
1111 #define OBD_CONNECT_INDEX 0x2ULL /*connect specific LOV idx */
1112 #define OBD_CONNECT_MDS 0x4ULL /*connect from MDT to OST */
1113 #define OBD_CONNECT_GRANT 0x8ULL /*OSC gets grant at connect */
1114 #define OBD_CONNECT_SRVLOCK 0x10ULL /*server takes locks for cli */
1115 #define OBD_CONNECT_VERSION 0x20ULL /*Lustre versions in ocd */
1116 #define OBD_CONNECT_REQPORTAL 0x40ULL /*Separate non-IO req portal */
1117 #define OBD_CONNECT_ACL 0x80ULL /*access control lists */
1118 #define OBD_CONNECT_XATTR 0x100ULL /*client use extended attr */
1119 #define OBD_CONNECT_CROW 0x200ULL /*MDS+OST create obj on write*/
1120 #define OBD_CONNECT_TRUNCLOCK 0x400ULL /*locks on server for punch */
1121 #define OBD_CONNECT_TRANSNO 0x800ULL /*replay sends init transno */
1122 #define OBD_CONNECT_IBITS 0x1000ULL /*support for inodebits locks*/
1123 #define OBD_CONNECT_JOIN 0x2000ULL /*files can be concatenated.
1124 *We do not support JOIN FILE
1125 *anymore, reserve this flags
1126 *just for preventing such bit
1128 #define OBD_CONNECT_ATTRFID 0x4000ULL /*Server can GetAttr By Fid*/
1129 #define OBD_CONNECT_NODEVOH 0x8000ULL /*No open hndl on specl nodes*/
1130 #define OBD_CONNECT_RMT_CLIENT 0x10000ULL /*Remote client */
1131 #define OBD_CONNECT_RMT_CLIENT_FORCE 0x20000ULL /*Remote client by force */
1132 #define OBD_CONNECT_BRW_SIZE 0x40000ULL /*Max bytes per rpc */
1133 #define OBD_CONNECT_QUOTA64 0x80000ULL /*64bit qunit_data.qd_count */
1134 #define OBD_CONNECT_MDS_CAPA 0x100000ULL /*MDS capability */
1135 #define OBD_CONNECT_OSS_CAPA 0x200000ULL /*OSS capability */
1136 #define OBD_CONNECT_CANCELSET 0x400000ULL /*Early batched cancels. */
1137 #define OBD_CONNECT_SOM 0x800000ULL /*Size on MDS */
1138 #define OBD_CONNECT_AT 0x1000000ULL /*client uses AT */
1139 #define OBD_CONNECT_LRU_RESIZE 0x2000000ULL /*LRU resize feature. */
1140 #define OBD_CONNECT_MDS_MDS 0x4000000ULL /*MDS-MDS connection */
1141 #define OBD_CONNECT_REAL 0x8000000ULL /*real connection */
1142 #define OBD_CONNECT_CHANGE_QS 0x10000000ULL /*shrink/enlarge qunit */
1143 #define OBD_CONNECT_CKSUM 0x20000000ULL /*support several cksum algos*/
1144 #define OBD_CONNECT_FID 0x40000000ULL /*FID is supported by server */
1145 #define OBD_CONNECT_VBR 0x80000000ULL /*version based recovery */
1146 #define OBD_CONNECT_LOV_V3 0x100000000ULL /*client supports LOV v3 EA */
1147 #define OBD_CONNECT_GRANT_SHRINK 0x200000000ULL /* support grant shrink */
1148 #define OBD_CONNECT_SKIP_ORPHAN 0x400000000ULL /* don't reuse orphan objids */
1149 #define OBD_CONNECT_MAX_EASIZE 0x800000000ULL /* preserved for large EA */
1150 #define OBD_CONNECT_FULL20 0x1000000000ULL /* it is 2.0 client */
1151 #define OBD_CONNECT_LAYOUTLOCK 0x2000000000ULL /* client uses layout lock */
1152 #define OBD_CONNECT_64BITHASH 0x4000000000ULL /* client supports 64-bits
1154 #define OBD_CONNECT_MAXBYTES 0x8000000000ULL /* max stripe size */
1155 #define OBD_CONNECT_IMP_RECOV 0x10000000000ULL /* imp recovery support */
1156 #define OBD_CONNECT_JOBSTATS 0x20000000000ULL /* jobid in ptlrpc_body */
1157 #define OBD_CONNECT_UMASK 0x40000000000ULL /* create uses client umask */
1158 #define OBD_CONNECT_EINPROGRESS 0x80000000000ULL /* client handles -EINPROGRESS
1159 * RPC error properly */
1160 #define OBD_CONNECT_GRANT_PARAM 0x100000000000ULL/* extra grant params used for
1161 * finer space reservation */
1162 #define OBD_CONNECT_NANOSEC_TIME 0x200000000000ULL /* nanosecond timestamps */
1163 #define OBD_CONNECT_LVB_TYPE 0x400000000000ULL /* variable type of LVB */
1164 #define OBD_CONNECT_LIGHTWEIGHT 0x1000000000000ULL/* lightweight connection */
1166 * Please DO NOT add flag values here before first ensuring that this same
1167 * flag value is not in use on some other branch. Please clear any such
1168 * changes with senior engineers before starting to use a new flag. Then,
1169 * submit a small patch against EVERY branch that ONLY adds the new flag
1170 * and updates obd_connect_names[] for lprocfs_rd_connect_flags(), so it
1171 * can be approved and landed easily to reserve the flag for future use. */
1173 /* The MNE_SWAB flag is overloading the MDS_MDS bit only for the MGS
1174 * connection. It is a temporary bug fix for Imperative Recovery interop
1175 * between 2.2 and 2.3 x86/ppc nodes, and can be removed when interop for
1176 * 2.2 clients/servers is no longer needed. LU-1252/LU-1644. */
1177 #define OBD_CONNECT_MNE_SWAB OBD_CONNECT_MDS_MDS
1179 #define OCD_HAS_FLAG(ocd, flg) \
1180 (!!((ocd)->ocd_connect_flags & OBD_CONNECT_##flg))
1183 #ifdef HAVE_LRU_RESIZE_SUPPORT
1184 #define LRU_RESIZE_CONNECT_FLAG OBD_CONNECT_LRU_RESIZE
1186 #define LRU_RESIZE_CONNECT_FLAG 0
1189 #define MDT_CONNECT_SUPPORTED (OBD_CONNECT_RDONLY | OBD_CONNECT_VERSION | \
1190 OBD_CONNECT_ACL | OBD_CONNECT_XATTR | \
1191 OBD_CONNECT_IBITS | \
1192 OBD_CONNECT_NODEVOH | OBD_CONNECT_ATTRFID | \
1193 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1194 OBD_CONNECT_RMT_CLIENT | \
1195 OBD_CONNECT_RMT_CLIENT_FORCE | \
1196 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_MDS_CAPA | \
1197 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_MDS_MDS | \
1198 OBD_CONNECT_FID | LRU_RESIZE_CONNECT_FLAG | \
1199 OBD_CONNECT_VBR | OBD_CONNECT_LOV_V3 | \
1200 OBD_CONNECT_SOM | OBD_CONNECT_FULL20 | \
1201 OBD_CONNECT_64BITHASH | \
1202 OBD_CONNECT_EINPROGRESS | OBD_CONNECT_JOBSTATS)
1203 #define OST_CONNECT_SUPPORTED (OBD_CONNECT_SRVLOCK | OBD_CONNECT_GRANT | \
1204 OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | \
1205 OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_INDEX | \
1206 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_QUOTA64 | \
1207 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1208 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_CKSUM | \
1209 OBD_CONNECT_CHANGE_QS | \
1210 OBD_CONNECT_OSS_CAPA | \
1211 OBD_CONNECT_RMT_CLIENT | \
1212 OBD_CONNECT_RMT_CLIENT_FORCE | OBD_CONNECT_VBR | \
1213 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN | \
1214 OBD_CONNECT_GRANT_SHRINK | OBD_CONNECT_FULL20 | \
1215 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES | \
1216 OBD_CONNECT_MAX_EASIZE | \
1217 OBD_CONNECT_EINPROGRESS | OBD_CONNECT_JOBSTATS)
1218 #define ECHO_CONNECT_SUPPORTED (0)
1219 #define MGS_CONNECT_SUPPORTED (OBD_CONNECT_VERSION | OBD_CONNECT_AT | \
1220 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV | \
1221 OBD_CONNECT_MNE_SWAB)
1223 /* Features required for this version of the client to work with server */
1224 #define CLIENT_CONNECT_MDT_REQD (OBD_CONNECT_IBITS | OBD_CONNECT_FID | \
1227 #define OBD_OCD_VERSION(major,minor,patch,fix) (((major)<<24) + ((minor)<<16) +\
1228 ((patch)<<8) + (fix))
1229 #define OBD_OCD_VERSION_MAJOR(version) ((int)((version)>>24)&255)
1230 #define OBD_OCD_VERSION_MINOR(version) ((int)((version)>>16)&255)
1231 #define OBD_OCD_VERSION_PATCH(version) ((int)((version)>>8)&255)
1232 #define OBD_OCD_VERSION_FIX(version) ((int)(version)&255)
1234 /* This structure is used for both request and reply.
1236 * If we eventually have separate connect data for different types, which we
1237 * almost certainly will, then perhaps we stick a union in here. */
1238 struct obd_connect_data_v1 {
1239 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1240 __u32 ocd_version; /* lustre release version number */
1241 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1242 __u32 ocd_index; /* LOV index to connect to */
1243 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
1244 __u64 ocd_ibits_known; /* inode bits this client understands */
1245 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1246 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1247 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1248 __u32 ocd_unused; /* also fix lustre_swab_connect */
1249 __u64 ocd_transno; /* first transno from client to be replayed */
1250 __u32 ocd_group; /* MDS group on OST */
1251 __u32 ocd_cksum_types; /* supported checksum algorithms */
1252 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1253 __u32 ocd_instance; /* also fix lustre_swab_connect */
1254 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1257 struct obd_connect_data {
1258 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1259 __u32 ocd_version; /* lustre release version number */
1260 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1261 __u32 ocd_index; /* LOV index to connect to */
1262 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
1263 __u64 ocd_ibits_known; /* inode bits this client understands */
1264 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1265 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1266 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1267 __u32 ocd_unused; /* also fix lustre_swab_connect */
1268 __u64 ocd_transno; /* first transno from client to be replayed */
1269 __u32 ocd_group; /* MDS group on OST */
1270 __u32 ocd_cksum_types; /* supported checksum algorithms */
1271 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1272 __u32 ocd_instance; /* instance # of this target */
1273 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1274 /* Fields after ocd_maxbytes are only accessible by the receiver
1275 * if the corresponding flag in ocd_connect_flags is set. Accessing
1276 * any field after ocd_maxbytes on the receiver without a valid flag
1277 * may result in out-of-bound memory access and kernel oops. */
1278 __u64 padding1; /* added 2.1.0. also fix lustre_swab_connect */
1279 __u64 padding2; /* added 2.1.0. also fix lustre_swab_connect */
1280 __u64 padding3; /* added 2.1.0. also fix lustre_swab_connect */
1281 __u64 padding4; /* added 2.1.0. also fix lustre_swab_connect */
1282 __u64 padding5; /* added 2.1.0. also fix lustre_swab_connect */
1283 __u64 padding6; /* added 2.1.0. also fix lustre_swab_connect */
1284 __u64 padding7; /* added 2.1.0. also fix lustre_swab_connect */
1285 __u64 padding8; /* added 2.1.0. also fix lustre_swab_connect */
1286 __u64 padding9; /* added 2.1.0. also fix lustre_swab_connect */
1287 __u64 paddingA; /* added 2.1.0. also fix lustre_swab_connect */
1288 __u64 paddingB; /* added 2.1.0. also fix lustre_swab_connect */
1289 __u64 paddingC; /* added 2.1.0. also fix lustre_swab_connect */
1290 __u64 paddingD; /* added 2.1.0. also fix lustre_swab_connect */
1291 __u64 paddingE; /* added 2.1.0. also fix lustre_swab_connect */
1292 __u64 paddingF; /* added 2.1.0. also fix lustre_swab_connect */
1295 * Please DO NOT use any fields here before first ensuring that this same
1296 * field is not in use on some other branch. Please clear any such changes
1297 * with senior engineers before starting to use a new field. Then, submit
1298 * a small patch against EVERY branch that ONLY adds the new field along with
1299 * the matching OBD_CONNECT flag, so that can be approved and landed easily to
1300 * reserve the flag for future use. */
1303 extern void lustre_swab_connect(struct obd_connect_data *ocd);
1306 * Supported checksum algorithms. Up to 32 checksum types are supported.
1307 * (32-bit mask stored in obd_connect_data::ocd_cksum_types)
1308 * Please update DECLARE_CKSUM_NAME/OBD_CKSUM_ALL in obd.h when adding a new
1309 * algorithm and also the OBD_FL_CKSUM* flags.
1312 OBD_CKSUM_CRC32 = 0x00000001,
1313 OBD_CKSUM_ADLER = 0x00000002,
1314 OBD_CKSUM_CRC32C= 0x00000004,
1318 * OST requests: OBDO & OBD request records
1323 OST_REPLY = 0, /* reply ? */
1339 OST_QUOTACHECK = 18,
1341 OST_QUOTA_ADJUST_QUNIT = 20,
1344 #define OST_FIRST_OPC OST_REPLY
1347 OBD_FL_INLINEDATA = 0x00000001,
1348 OBD_FL_OBDMDEXISTS = 0x00000002,
1349 OBD_FL_DELORPHAN = 0x00000004, /* if set in o_flags delete orphans */
1350 OBD_FL_NORPC = 0x00000008, /* set in o_flags do in OSC not OST */
1351 OBD_FL_IDONLY = 0x00000010, /* set in o_flags only adjust obj id*/
1352 OBD_FL_RECREATE_OBJS= 0x00000020, /* recreate missing obj */
1353 OBD_FL_DEBUG_CHECK = 0x00000040, /* echo client/server debug check */
1354 OBD_FL_NO_USRQUOTA = 0x00000100, /* the object's owner is over quota */
1355 OBD_FL_NO_GRPQUOTA = 0x00000200, /* the object's group is over quota */
1356 OBD_FL_CREATE_CROW = 0x00000400, /* object should be create on write */
1357 OBD_FL_SRVLOCK = 0x00000800, /* delegate DLM locking to server */
1358 OBD_FL_CKSUM_CRC32 = 0x00001000, /* CRC32 checksum type */
1359 OBD_FL_CKSUM_ADLER = 0x00002000, /* ADLER checksum type */
1360 OBD_FL_CKSUM_CRC32C = 0x00004000, /* CRC32C checksum type */
1361 OBD_FL_CKSUM_RSVD2 = 0x00008000, /* for future cksum types */
1362 OBD_FL_CKSUM_RSVD3 = 0x00010000, /* for future cksum types */
1363 OBD_FL_SHRINK_GRANT = 0x00020000, /* object shrink the grant */
1364 OBD_FL_MMAP = 0x00040000, /* object is mmapped on the client.
1365 * XXX: obsoleted - reserved for old
1366 * clients prior than 2.2 */
1367 OBD_FL_RECOV_RESEND = 0x00080000, /* recoverable resent */
1368 OBD_FL_NOSPC_BLK = 0x00100000, /* no more block space on OST */
1370 /* Note that while these checksum values are currently separate bits,
1371 * in 2.x we can actually allow all values from 1-31 if we wanted. */
1372 OBD_FL_CKSUM_ALL = OBD_FL_CKSUM_CRC32 | OBD_FL_CKSUM_ADLER |
1373 OBD_FL_CKSUM_CRC32C,
1375 /* mask for local-only flag, which won't be sent over network */
1376 OBD_FL_LOCAL_MASK = 0xF0000000,
1379 #define LOV_MAGIC_V1 0x0BD10BD0
1380 #define LOV_MAGIC LOV_MAGIC_V1
1381 #define LOV_MAGIC_JOIN_V1 0x0BD20BD0
1382 #define LOV_MAGIC_V3 0x0BD30BD0
1384 #define LOV_PATTERN_RAID0 0x001 /* stripes are used round-robin */
1385 #define LOV_PATTERN_RAID1 0x002 /* stripes are mirrors of each other */
1386 #define LOV_PATTERN_FIRST 0x100 /* first stripe is not in round-robin */
1387 #define LOV_PATTERN_CMOBD 0x200
1389 #define lov_ost_data lov_ost_data_v1
1390 struct lov_ost_data_v1 { /* per-stripe data structure (little-endian)*/
1391 __u64 l_object_id; /* OST object ID */
1392 __u64 l_object_seq; /* OST object seq number */
1393 __u32 l_ost_gen; /* generation of this l_ost_idx */
1394 __u32 l_ost_idx; /* OST index in LOV (lov_tgt_desc->tgts) */
1397 #define lov_mds_md lov_mds_md_v1
1398 struct lov_mds_md_v1 { /* LOV EA mds/wire data (little-endian) */
1399 __u32 lmm_magic; /* magic number = LOV_MAGIC_V1 */
1400 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1401 __u64 lmm_object_id; /* LOV object ID */
1402 __u64 lmm_object_seq; /* LOV object seq number */
1403 __u32 lmm_stripe_size; /* size of stripe in bytes */
1404 /* lmm_stripe_count used to be __u32 */
1405 __u16 lmm_stripe_count; /* num stripes in use for this object */
1406 __u16 lmm_layout_gen; /* layout generation number */
1407 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1410 /* extern void lustre_swab_lov_mds_md(struct lov_mds_md *llm); */
1412 #define MAX_MD_SIZE (sizeof(struct lov_mds_md) + 4 * sizeof(struct lov_ost_data))
1413 #define MIN_MD_SIZE (sizeof(struct lov_mds_md) + 1 * sizeof(struct lov_ost_data))
1415 #define XATTR_NAME_ACL_ACCESS "system.posix_acl_access"
1416 #define XATTR_NAME_ACL_DEFAULT "system.posix_acl_default"
1417 #define XATTR_USER_PREFIX "user."
1418 #define XATTR_TRUSTED_PREFIX "trusted."
1419 #define XATTR_SECURITY_PREFIX "security."
1420 #define XATTR_LUSTRE_PREFIX "lustre."
1422 #define XATTR_NAME_LOV "trusted.lov"
1423 #define XATTR_NAME_LMA "trusted.lma"
1424 #define XATTR_NAME_LMV "trusted.lmv"
1425 #define XATTR_NAME_LINK "trusted.link"
1426 #define XATTR_NAME_FID "trusted.fid"
1427 #define XATTR_NAME_VERSION "trusted.version"
1430 struct lov_mds_md_v3 { /* LOV EA mds/wire data (little-endian) */
1431 __u32 lmm_magic; /* magic number = LOV_MAGIC_V3 */
1432 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1433 __u64 lmm_object_id; /* LOV object ID */
1434 __u64 lmm_object_seq; /* LOV object seq number */
1435 __u32 lmm_stripe_size; /* size of stripe in bytes */
1436 /* lmm_stripe_count used to be __u32 */
1437 __u16 lmm_stripe_count; /* num stripes in use for this object */
1438 __u16 lmm_layout_gen; /* layout generation number */
1439 char lmm_pool_name[LOV_MAXPOOLNAME]; /* must be 32bit aligned */
1440 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1444 #define OBD_MD_FLID (0x00000001ULL) /* object ID */
1445 #define OBD_MD_FLATIME (0x00000002ULL) /* access time */
1446 #define OBD_MD_FLMTIME (0x00000004ULL) /* data modification time */
1447 #define OBD_MD_FLCTIME (0x00000008ULL) /* change time */
1448 #define OBD_MD_FLSIZE (0x00000010ULL) /* size */
1449 #define OBD_MD_FLBLOCKS (0x00000020ULL) /* allocated blocks count */
1450 #define OBD_MD_FLBLKSZ (0x00000040ULL) /* block size */
1451 #define OBD_MD_FLMODE (0x00000080ULL) /* access bits (mode & ~S_IFMT) */
1452 #define OBD_MD_FLTYPE (0x00000100ULL) /* object type (mode & S_IFMT) */
1453 #define OBD_MD_FLUID (0x00000200ULL) /* user ID */
1454 #define OBD_MD_FLGID (0x00000400ULL) /* group ID */
1455 #define OBD_MD_FLFLAGS (0x00000800ULL) /* flags word */
1456 #define OBD_MD_FLNLINK (0x00002000ULL) /* link count */
1457 #define OBD_MD_FLGENER (0x00004000ULL) /* generation number */
1458 /*#define OBD_MD_FLINLINE (0x00008000ULL) inline data. used until 1.6.5 */
1459 #define OBD_MD_FLRDEV (0x00010000ULL) /* device number */
1460 #define OBD_MD_FLEASIZE (0x00020000ULL) /* extended attribute data */
1461 #define OBD_MD_LINKNAME (0x00040000ULL) /* symbolic link target */
1462 #define OBD_MD_FLHANDLE (0x00080000ULL) /* file/lock handle */
1463 #define OBD_MD_FLCKSUM (0x00100000ULL) /* bulk data checksum */
1464 #define OBD_MD_FLQOS (0x00200000ULL) /* quality of service stats */
1465 /*#define OBD_MD_FLOSCOPQ (0x00400000ULL) osc opaque data, never used */
1466 #define OBD_MD_FLCOOKIE (0x00800000ULL) /* log cancellation cookie */
1467 #define OBD_MD_FLGROUP (0x01000000ULL) /* group */
1468 #define OBD_MD_FLFID (0x02000000ULL) /* ->ost write inline fid */
1469 #define OBD_MD_FLEPOCH (0x04000000ULL) /* ->ost write with ioepoch */
1470 /* ->mds if epoch opens or closes */
1471 #define OBD_MD_FLGRANT (0x08000000ULL) /* ost preallocation space grant */
1472 #define OBD_MD_FLDIREA (0x10000000ULL) /* dir's extended attribute data */
1473 #define OBD_MD_FLUSRQUOTA (0x20000000ULL) /* over quota flags sent from ost */
1474 #define OBD_MD_FLGRPQUOTA (0x40000000ULL) /* over quota flags sent from ost */
1475 #define OBD_MD_FLMODEASIZE (0x80000000ULL) /* EA size will be changed */
1477 #define OBD_MD_MDS (0x0000000100000000ULL) /* where an inode lives on */
1478 #define OBD_MD_REINT (0x0000000200000000ULL) /* reintegrate oa */
1479 #define OBD_MD_MEA (0x0000000400000000ULL) /* CMD split EA */
1480 #define OBD_MD_MDTIDX (0x0000000800000000ULL) /* Get MDT index */
1482 #define OBD_MD_FLXATTR (0x0000001000000000ULL) /* xattr */
1483 #define OBD_MD_FLXATTRLS (0x0000002000000000ULL) /* xattr list */
1484 #define OBD_MD_FLXATTRRM (0x0000004000000000ULL) /* xattr remove */
1485 #define OBD_MD_FLACL (0x0000008000000000ULL) /* ACL */
1486 #define OBD_MD_FLRMTPERM (0x0000010000000000ULL) /* remote permission */
1487 #define OBD_MD_FLMDSCAPA (0x0000020000000000ULL) /* MDS capability */
1488 #define OBD_MD_FLOSSCAPA (0x0000040000000000ULL) /* OSS capability */
1489 #define OBD_MD_FLCKSPLIT (0x0000080000000000ULL) /* Check split on server */
1490 #define OBD_MD_FLCROSSREF (0x0000100000000000ULL) /* Cross-ref case */
1491 #define OBD_MD_FLGETATTRLOCK (0x0000200000000000ULL) /* Get IOEpoch attributes
1493 #define OBD_MD_FLOBJCOUNT (0x0000400000000000ULL) /* for multiple destroy */
1495 #define OBD_MD_FLRMTLSETFACL (0x0001000000000000ULL) /* lfs lsetfacl case */
1496 #define OBD_MD_FLRMTLGETFACL (0x0002000000000000ULL) /* lfs lgetfacl case */
1497 #define OBD_MD_FLRMTRSETFACL (0x0004000000000000ULL) /* lfs rsetfacl case */
1498 #define OBD_MD_FLRMTRGETFACL (0x0008000000000000ULL) /* lfs rgetfacl case */
1500 #define OBD_MD_FLDATAVERSION (0x0010000000000000ULL) /* iversion sum */
1502 #define OBD_MD_FLGETATTR (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLMTIME | \
1503 OBD_MD_FLCTIME | OBD_MD_FLSIZE | OBD_MD_FLBLKSZ | \
1504 OBD_MD_FLMODE | OBD_MD_FLTYPE | OBD_MD_FLUID | \
1505 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLNLINK | \
1506 OBD_MD_FLGENER | OBD_MD_FLRDEV | OBD_MD_FLGROUP)
1508 /* don't forget obdo_fid which is way down at the bottom so it can
1509 * come after the definition of llog_cookie */
1512 extern void lustre_swab_obd_statfs (struct obd_statfs *os);
1514 /* ost_body.data values for OST_BRW */
1516 #define OBD_BRW_READ 0x01
1517 #define OBD_BRW_WRITE 0x02
1518 #define OBD_BRW_RWMASK (OBD_BRW_READ | OBD_BRW_WRITE)
1519 #define OBD_BRW_SYNC 0x08 /* this page is a part of synchronous
1520 * transfer and is not accounted in
1522 #define OBD_BRW_CHECK 0x10
1523 #define OBD_BRW_FROM_GRANT 0x20 /* the osc manages this under llite */
1524 #define OBD_BRW_GRANTED 0x40 /* the ost manages this */
1525 #define OBD_BRW_NOCACHE 0x80 /* this page is a part of non-cached IO */
1526 #define OBD_BRW_NOQUOTA 0x100
1527 #define OBD_BRW_SRVLOCK 0x200 /* Client holds no lock over this page */
1528 #define OBD_BRW_ASYNC 0x400 /* Server may delay commit to disk */
1529 #define OBD_BRW_MEMALLOC 0x800 /* Client runs in the "kswapd" context */
1530 #define OBD_BRW_OVER_USRQUOTA 0x1000 /* Running out of user quota */
1531 #define OBD_BRW_OVER_GRPQUOTA 0x2000 /* Running out of group quota */
1533 #define OBD_OBJECT_EOF 0xffffffffffffffffULL
1535 #define OST_MIN_PRECREATE 32
1536 #define OST_MAX_PRECREATE 20000
1545 extern void lustre_swab_obd_ioobj (struct obd_ioobj *ioo);
1547 /* multiple of 8 bytes => can array */
1548 struct niobuf_remote {
1554 extern void lustre_swab_niobuf_remote (struct niobuf_remote *nbr);
1556 /* lock value block communicated between the filter and llite */
1558 /* OST_LVB_ERR_INIT is needed because the return code in rc is
1559 * negative, i.e. because ((MASK + rc) & MASK) != MASK. */
1560 #define OST_LVB_ERR_INIT 0xffbadbad80000000ULL
1561 #define OST_LVB_ERR_MASK 0xffbadbad00000000ULL
1562 #define OST_LVB_IS_ERR(blocks) \
1563 ((blocks & OST_LVB_ERR_MASK) == OST_LVB_ERR_MASK)
1564 #define OST_LVB_SET_ERR(blocks, rc) \
1565 do { blocks = OST_LVB_ERR_INIT + rc; } while (0)
1566 #define OST_LVB_GET_ERR(blocks) (int)(blocks - OST_LVB_ERR_INIT)
1583 MDS_GETATTR_NAME = 34,
1588 MDS_DISCONNECT = 39,
1594 MDS_DONE_WRITING = 45,
1596 MDS_QUOTACHECK = 47,
1599 MDS_SETXATTR = 50, /* obsolete, now it's MDS_REINT op */
1606 #define MDS_FIRST_OPC MDS_GETATTR
1623 } mds_reint_t, mdt_reint_t;
1625 extern void lustre_swab_generic_32s (__u32 *val);
1627 /* the disposition of the intent outlines what was executed */
1628 #define DISP_IT_EXECD 0x00000001
1629 #define DISP_LOOKUP_EXECD 0x00000002
1630 #define DISP_LOOKUP_NEG 0x00000004
1631 #define DISP_LOOKUP_POS 0x00000008
1632 #define DISP_OPEN_CREATE 0x00000010
1633 #define DISP_OPEN_OPEN 0x00000020
1634 #define DISP_ENQ_COMPLETE 0x00400000
1635 #define DISP_ENQ_OPEN_REF 0x00800000
1636 #define DISP_ENQ_CREATE_REF 0x01000000
1637 #define DISP_OPEN_LOCK 0x02000000
1639 /* INODE LOCK PARTS */
1640 #define MDS_INODELOCK_LOOKUP 0x000001 /* dentry, mode, owner, group */
1641 #define MDS_INODELOCK_UPDATE 0x000002 /* size, links, timestamps */
1642 #define MDS_INODELOCK_OPEN 0x000004 /* For opened files */
1643 #define MDS_INODELOCK_LAYOUT 0x000008 /* for layout */
1645 /* Do not forget to increase MDS_INODELOCK_MAXSHIFT when adding new bits
1646 * XXX: MDS_INODELOCK_MAXSHIFT should be increased to 3 once the layout lock is
1648 #define MDS_INODELOCK_MAXSHIFT 2
1649 /* This FULL lock is useful to take on unlink sort of operations */
1650 #define MDS_INODELOCK_FULL ((1<<(MDS_INODELOCK_MAXSHIFT+1))-1)
1652 extern void lustre_swab_ll_fid (struct ll_fid *fid);
1654 /* NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
1655 * but was moved into name[1] along with the OID to avoid consuming the
1656 * name[2,3] fields that need to be used for the quota id (also a FID). */
1658 LUSTRE_RES_ID_SEQ_OFF = 0,
1659 LUSTRE_RES_ID_VER_OID_OFF = 1,
1660 LUSTRE_RES_ID_WAS_VER_OFF = 2, /* see note above */
1661 LUSTRE_RES_ID_HSH_OFF = 3
1664 #define MDS_STATUS_CONN 1
1665 #define MDS_STATUS_LOV 2
1667 /* mdt_thread_info.mti_flags. */
1669 /* The flag indicates Size-on-MDS attributes are changed. */
1670 MF_SOM_CHANGE = (1 << 0),
1671 /* Flags indicates an epoch opens or closes. */
1672 MF_EPOCH_OPEN = (1 << 1),
1673 MF_EPOCH_CLOSE = (1 << 2),
1674 MF_MDC_CANCEL_FID1 = (1 << 3),
1675 MF_MDC_CANCEL_FID2 = (1 << 4),
1676 MF_MDC_CANCEL_FID3 = (1 << 5),
1677 MF_MDC_CANCEL_FID4 = (1 << 6),
1678 /* There is a pending attribute update. */
1679 MF_SOM_AU = (1 << 7),
1680 /* Cancel OST locks while getattr OST attributes. */
1681 MF_GETATTR_LOCK = (1 << 8),
1684 #define MF_SOM_LOCAL_FLAGS (MF_SOM_CHANGE | MF_EPOCH_OPEN | MF_EPOCH_CLOSE)
1686 #define LUSTRE_BFLAG_UNCOMMITTED_WRITES 0x1
1688 /* these should be identical to their EXT4_*_FL counterparts, they are
1689 * redefined here only to avoid dragging in fs/ext4/ext4.h */
1690 #define LUSTRE_SYNC_FL 0x00000008 /* Synchronous updates */
1691 #define LUSTRE_IMMUTABLE_FL 0x00000010 /* Immutable file */
1692 #define LUSTRE_APPEND_FL 0x00000020 /* writes to file may only append */
1693 #define LUSTRE_NOATIME_FL 0x00000080 /* do not update atime */
1694 #define LUSTRE_DIRSYNC_FL 0x00010000 /* dirsync behaviour (dir only) */
1697 /* Convert wire LUSTRE_*_FL to corresponding client local VFS S_* values
1698 * for the client inode i_flags. The LUSTRE_*_FL are the Lustre wire
1699 * protocol equivalents of LDISKFS_*_FL values stored on disk, while
1700 * the S_* flags are kernel-internal values that change between kernel
1701 * versions. These flags are set/cleared via FSFILT_IOC_{GET,SET}_FLAGS.
1702 * See b=16526 for a full history. */
1703 static inline int ll_ext_to_inode_flags(int flags)
1705 return (((flags & LUSTRE_SYNC_FL) ? S_SYNC : 0) |
1706 ((flags & LUSTRE_NOATIME_FL) ? S_NOATIME : 0) |
1707 ((flags & LUSTRE_APPEND_FL) ? S_APPEND : 0) |
1708 #if defined(S_DIRSYNC)
1709 ((flags & LUSTRE_DIRSYNC_FL) ? S_DIRSYNC : 0) |
1711 ((flags & LUSTRE_IMMUTABLE_FL) ? S_IMMUTABLE : 0));
1714 static inline int ll_inode_to_ext_flags(int iflags)
1716 return (((iflags & S_SYNC) ? LUSTRE_SYNC_FL : 0) |
1717 ((iflags & S_NOATIME) ? LUSTRE_NOATIME_FL : 0) |
1718 ((iflags & S_APPEND) ? LUSTRE_APPEND_FL : 0) |
1719 #if defined(S_DIRSYNC)
1720 ((iflags & S_DIRSYNC) ? LUSTRE_DIRSYNC_FL : 0) |
1722 ((iflags & S_IMMUTABLE) ? LUSTRE_IMMUTABLE_FL : 0));
1729 struct lustre_handle handle;
1731 __u64 size; /* Offset, in the case of MDS_READPAGE */
1735 __u64 blocks; /* XID, in the case of MDS_READPAGE */
1744 __u32 flags; /* from vfs for pin/unpin, LUSTRE_BFLAG close */
1746 __u32 nlink; /* #bytes to read in the case of MDS_READPAGE */
1752 __u32 max_cookiesize;
1753 __u32 uid_h; /* high 32-bits of uid, for FUID */
1754 __u32 gid_h; /* high 32-bits of gid, for FUID */
1755 __u32 padding_5; /* also fix lustre_swab_mdt_body */
1763 extern void lustre_swab_mdt_body (struct mdt_body *b);
1765 struct mdt_ioepoch {
1766 struct lustre_handle handle;
1772 extern void lustre_swab_mdt_ioepoch (struct mdt_ioepoch *b);
1774 /* The lquota_id structure is an union of all the possible identifier types that
1775 * can be used with quota, this includes:
1778 * - a FID which can be used for per-directory quota in the future */
1780 struct lu_fid qid_fid; /* FID for per-directory quota */
1781 __u64 qid_uid; /* user identifier */
1782 __u64 qid_gid; /* group identifier */
1785 #define Q_QUOTACHECK 0x800100
1786 #define Q_INITQUOTA 0x800101 /* init slave limits */
1787 #define Q_GETOINFO 0x800102 /* get obd quota info */
1788 #define Q_GETOQUOTA 0x800103 /* get obd quotas */
1789 #define Q_FINVALIDATE 0x800104 /* invalidate operational quotas */
1791 #define Q_TYPEMATCH(id, type) \
1792 ((id) == (type) || (id) == UGQUOTA)
1794 #define Q_TYPESET(oqc, type) Q_TYPEMATCH((oqc)->qc_type, type)
1796 #define Q_GETOCMD(oqc) \
1797 ((oqc)->qc_cmd == Q_GETOINFO || (oqc)->qc_cmd == Q_GETOQUOTA)
1799 #define QCTL_COPY(out, in) \
1801 Q_COPY(out, in, qc_cmd); \
1802 Q_COPY(out, in, qc_type); \
1803 Q_COPY(out, in, qc_id); \
1804 Q_COPY(out, in, qc_stat); \
1805 Q_COPY(out, in, qc_dqinfo); \
1806 Q_COPY(out, in, qc_dqblk); \
1809 struct obd_quotactl {
1814 struct obd_dqinfo qc_dqinfo;
1815 struct obd_dqblk qc_dqblk;
1818 extern void lustre_swab_obd_quotactl(struct obd_quotactl *q);
1820 struct quota_adjust_qunit {
1827 extern void lustre_swab_quota_adjust_qunit(struct quota_adjust_qunit *q);
1830 * Space accounting support
1831 * Format of an accounting record, providing disk usage information for a given
1834 struct lquota_acct_rec { /* 16 bytes */
1835 __u64 bspace; /* current space in use */
1836 __u64 ispace; /* current # inodes in use */
1840 * Global quota index support
1841 * Format of a global record, providing global quota settings for a given quota
1844 struct lquota_glb_rec { /* 32 bytes */
1845 __u64 qbr_hardlimit; /* quota hard limit, in #inodes or kbytes */
1846 __u64 qbr_softlimit; /* quota soft limit, in #inodes or kbytes */
1847 __u64 qbr_time; /* grace time, in seconds */
1848 __u64 qbr_granted; /* how much is granted to slaves, in #inodes or
1853 * Slave index support
1854 * Format of a slave record, recording how much space is granted to a given
1857 struct lquota_slv_rec { /* 8 bytes */
1858 __u64 qsr_granted; /* space granted to the slave for the key=ID,
1859 * in #inodes or kbytes */
1862 /* flags is shared among quota structures */
1863 #define LQUOTA_FLAGS_GRP 1UL /* 0 is user, 1 is group */
1864 #define LQUOTA_FLAGS_BLK 2UL /* 0 is inode, 1 is block */
1865 #define LQUOTA_FLAGS_ADJBLK 4UL /* adjust the block qunit size */
1866 #define LQUOTA_FLAGS_ADJINO 8UL /* adjust the inode qunit size */
1867 #define LQUOTA_FLAGS_CHG_QS 16UL /* indicate whether it has capability of
1868 * OBD_CONNECT_CHANGE_QS */
1869 #define LQUOTA_FLAGS_RECOVERY 32UL /* recovery is going on a uid/gid */
1870 #define LQUOTA_FLAGS_SETQUOTA 64UL /* being setquota on a uid/gid */
1872 /* flags is specific for quota_adjust_qunit */
1873 #define LQUOTA_QAQ_CREATE_LQS (1UL << 31) /* when it is set, need create lqs */
1875 /* the status of lqs_flags in struct lustre_qunit_size */
1876 #define LQUOTA_QUNIT_FLAGS (LQUOTA_FLAGS_GRP | LQUOTA_FLAGS_BLK)
1878 #define QAQ_IS_GRP(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_GRP)
1879 #define QAQ_IS_ADJBLK(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_ADJBLK)
1880 #define QAQ_IS_ADJINO(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_ADJINO)
1881 #define QAQ_IS_CREATE_LQS(qaq) ((qaq)->qaq_flags & LQUOTA_QAQ_CREATE_LQS)
1883 #define QAQ_SET_GRP(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_GRP)
1884 #define QAQ_SET_ADJBLK(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_ADJBLK)
1885 #define QAQ_SET_ADJINO(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_ADJINO)
1886 #define QAQ_SET_CREATE_LQS(qaq) ((qaq)->qaq_flags |= LQUOTA_QAQ_CREATE_LQS)
1888 /* permissions for md_perm.mp_perm */
1890 CFS_SETUID_PERM = 0x01,
1891 CFS_SETGID_PERM = 0x02,
1892 CFS_SETGRP_PERM = 0x04,
1893 CFS_RMTACL_PERM = 0x08,
1894 CFS_RMTOWN_PERM = 0x10
1897 /* inode access permission for remote user, the inode info are omitted,
1898 * for client knows them. */
1899 struct mdt_remote_perm {
1906 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
1910 extern void lustre_swab_mdt_remote_perm(struct mdt_remote_perm *p);
1912 struct mdt_rec_setattr {
1922 __u32 sa_padding_1_h;
1923 struct lu_fid sa_fid;
1932 __u32 sa_attr_flags;
1940 extern void lustre_swab_mdt_rec_setattr (struct mdt_rec_setattr *sa);
1943 * Attribute flags used in mdt_rec_setattr::sa_valid.
1944 * The kernel's #defines for ATTR_* should not be used over the network
1945 * since the client and MDS may run different kernels (see bug 13828)
1946 * Therefore, we should only use MDS_ATTR_* attributes for sa_valid.
1948 #define MDS_ATTR_MODE 0x1ULL /* = 1 */
1949 #define MDS_ATTR_UID 0x2ULL /* = 2 */
1950 #define MDS_ATTR_GID 0x4ULL /* = 4 */
1951 #define MDS_ATTR_SIZE 0x8ULL /* = 8 */
1952 #define MDS_ATTR_ATIME 0x10ULL /* = 16 */
1953 #define MDS_ATTR_MTIME 0x20ULL /* = 32 */
1954 #define MDS_ATTR_CTIME 0x40ULL /* = 64 */
1955 #define MDS_ATTR_ATIME_SET 0x80ULL /* = 128 */
1956 #define MDS_ATTR_MTIME_SET 0x100ULL /* = 256 */
1957 #define MDS_ATTR_FORCE 0x200ULL /* = 512, Not a change, but a change it */
1958 #define MDS_ATTR_ATTR_FLAG 0x400ULL /* = 1024 */
1959 #define MDS_ATTR_KILL_SUID 0x800ULL /* = 2048 */
1960 #define MDS_ATTR_KILL_SGID 0x1000ULL /* = 4096 */
1961 #define MDS_ATTR_CTIME_SET 0x2000ULL /* = 8192 */
1962 #define MDS_ATTR_FROM_OPEN 0x4000ULL /* = 16384, called from open path, ie O_TRUNC */
1963 #define MDS_ATTR_BLOCKS 0x8000ULL /* = 32768 */
1966 #define FMODE_READ 00000001
1967 #define FMODE_WRITE 00000002
1970 #define MDS_FMODE_CLOSED 00000000
1971 #define MDS_FMODE_EXEC 00000004
1972 /* IO Epoch is opened on a closed file. */
1973 #define MDS_FMODE_EPOCH 01000000
1974 /* IO Epoch is opened on a file truncate. */
1975 #define MDS_FMODE_TRUNC 02000000
1976 /* Size-on-MDS Attribute Update is pending. */
1977 #define MDS_FMODE_SOM 04000000
1979 #define MDS_OPEN_CREATED 00000010
1980 #define MDS_OPEN_CROSS 00000020
1982 #define MDS_OPEN_CREAT 00000100
1983 #define MDS_OPEN_EXCL 00000200
1984 #define MDS_OPEN_TRUNC 00001000
1985 #define MDS_OPEN_APPEND 00002000
1986 #define MDS_OPEN_SYNC 00010000
1987 #define MDS_OPEN_DIRECTORY 00200000
1989 #define MDS_OPEN_BY_FID 040000000 /* open_by_fid for known object */
1990 #define MDS_OPEN_DELAY_CREATE 0100000000 /* delay initial object create */
1991 #define MDS_OPEN_OWNEROVERRIDE 0200000000 /* NFSD rw-reopen ro file for owner */
1992 #define MDS_OPEN_JOIN_FILE 0400000000 /* open for join file.
1993 * We do not support JOIN FILE
1994 * anymore, reserve this flags
1995 * just for preventing such bit
1997 #define MDS_CREATE_RMT_ACL 01000000000 /* indicate create on remote server
1998 * with default ACL */
1999 #define MDS_CREATE_SLAVE_OBJ 02000000000 /* indicate create slave object
2000 * actually, this is for create, not
2001 * conflict with other open flags */
2002 #define MDS_OPEN_LOCK 04000000000 /* This open requires open lock */
2003 #define MDS_OPEN_HAS_EA 010000000000 /* specify object create pattern */
2004 #define MDS_OPEN_HAS_OBJS 020000000000 /* Just set the EA the obj exist */
2005 #define MDS_OPEN_NORESTORE 0100000000000ULL /* Do not restore file at open */
2006 #define MDS_OPEN_NEWSTRIPE 0200000000000ULL /* New stripe needed (restripe or
2009 /* permission for create non-directory file */
2010 #define MAY_CREATE (1 << 7)
2011 /* permission for create directory file */
2012 #define MAY_LINK (1 << 8)
2013 /* permission for delete from the directory */
2014 #define MAY_UNLINK (1 << 9)
2015 /* source's permission for rename */
2016 #define MAY_RENAME_SRC (1 << 10)
2017 /* target's permission for rename */
2018 #define MAY_RENAME_TAR (1 << 11)
2019 /* part (parent's) VTX permission check */
2020 #define MAY_VTX_PART (1 << 12)
2021 /* full VTX permission check */
2022 #define MAY_VTX_FULL (1 << 13)
2023 /* lfs rgetfacl permission check */
2024 #define MAY_RGETFACL (1 << 14)
2027 MDS_CHECK_SPLIT = 1 << 0,
2028 MDS_CROSS_REF = 1 << 1,
2029 MDS_VTX_BYPASS = 1 << 2,
2030 MDS_PERM_BYPASS = 1 << 3,
2032 MDS_QUOTA_IGNORE = 1 << 5,
2033 MDS_CLOSE_CLEANUP = 1 << 6,
2034 MDS_KEEP_ORPHAN = 1 << 7,
2035 MDS_RECOV_OPEN = 1 << 8,
2036 MDS_UNLINK_DESTROY = 1 << 9, /* Destory ost object in mdd_unlink */
2039 /* instance of mdt_reint_rec */
2040 struct mdt_rec_create {
2048 __u32 cr_suppgid1_h;
2050 __u32 cr_suppgid2_h;
2051 struct lu_fid cr_fid1;
2052 struct lu_fid cr_fid2;
2053 struct lustre_handle cr_old_handle; /* handle in case of open replay */
2057 __u64 cr_padding_1; /* rr_blocks */
2060 /* use of helpers set/get_mrc_cr_flags() is needed to access
2061 * 64 bits cr_flags [cr_flags_l, cr_flags_h], this is done to
2062 * extend cr_flags size without breaking 1.8 compat */
2063 __u32 cr_flags_l; /* for use with open, low 32 bits */
2064 __u32 cr_flags_h; /* for use with open, high 32 bits */
2065 __u32 cr_padding_3; /* rr_padding_3 */
2066 __u32 cr_padding_4; /* rr_padding_4 */
2069 static inline void set_mrc_cr_flags(struct mdt_rec_create *mrc, __u64 flags)
2071 mrc->cr_flags_l = (__u32)(flags & 0xFFFFFFFFUll);
2072 mrc->cr_flags_h = (__u32)(flags >> 32);
2075 static inline __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
2077 return ((__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32));
2080 /* instance of mdt_reint_rec */
2081 struct mdt_rec_link {
2089 __u32 lk_suppgid1_h;
2091 __u32 lk_suppgid2_h;
2092 struct lu_fid lk_fid1;
2093 struct lu_fid lk_fid2;
2095 __u64 lk_padding_1; /* rr_atime */
2096 __u64 lk_padding_2; /* rr_ctime */
2097 __u64 lk_padding_3; /* rr_size */
2098 __u64 lk_padding_4; /* rr_blocks */
2100 __u32 lk_padding_5; /* rr_mode */
2101 __u32 lk_padding_6; /* rr_flags */
2102 __u32 lk_padding_7; /* rr_padding_2 */
2103 __u32 lk_padding_8; /* rr_padding_3 */
2104 __u32 lk_padding_9; /* rr_padding_4 */
2107 /* instance of mdt_reint_rec */
2108 struct mdt_rec_unlink {
2116 __u32 ul_suppgid1_h;
2118 __u32 ul_suppgid2_h;
2119 struct lu_fid ul_fid1;
2120 struct lu_fid ul_fid2;
2122 __u64 ul_padding_2; /* rr_atime */
2123 __u64 ul_padding_3; /* rr_ctime */
2124 __u64 ul_padding_4; /* rr_size */
2125 __u64 ul_padding_5; /* rr_blocks */
2128 __u32 ul_padding_6; /* rr_flags */
2129 __u32 ul_padding_7; /* rr_padding_2 */
2130 __u32 ul_padding_8; /* rr_padding_3 */
2131 __u32 ul_padding_9; /* rr_padding_4 */
2134 /* instance of mdt_reint_rec */
2135 struct mdt_rec_rename {
2143 __u32 rn_suppgid1_h;
2145 __u32 rn_suppgid2_h;
2146 struct lu_fid rn_fid1;
2147 struct lu_fid rn_fid2;
2149 __u64 rn_padding_1; /* rr_atime */
2150 __u64 rn_padding_2; /* rr_ctime */
2151 __u64 rn_padding_3; /* rr_size */
2152 __u64 rn_padding_4; /* rr_blocks */
2153 __u32 rn_bias; /* some operation flags */
2154 __u32 rn_mode; /* cross-ref rename has mode */
2155 __u32 rn_padding_5; /* rr_flags */
2156 __u32 rn_padding_6; /* rr_padding_2 */
2157 __u32 rn_padding_7; /* rr_padding_3 */
2158 __u32 rn_padding_8; /* rr_padding_4 */
2161 /* instance of mdt_reint_rec */
2162 struct mdt_rec_setxattr {
2170 __u32 sx_suppgid1_h;
2172 __u32 sx_suppgid2_h;
2173 struct lu_fid sx_fid;
2174 __u64 sx_padding_1; /* These three are rr_fid2 */
2179 __u64 sx_padding_5; /* rr_ctime */
2180 __u64 sx_padding_6; /* rr_size */
2181 __u64 sx_padding_7; /* rr_blocks */
2184 __u32 sx_padding_8; /* rr_flags */
2185 __u32 sx_padding_9; /* rr_padding_2 */
2186 __u32 sx_padding_10; /* rr_padding_3 */
2187 __u32 sx_padding_11; /* rr_padding_4 */
2191 * mdt_rec_reint is the template for all mdt_reint_xxx structures.
2192 * Do NOT change the size of various members, otherwise the value
2193 * will be broken in lustre_swab_mdt_rec_reint().
2195 * If you add new members in other mdt_reint_xxx structres and need to use the
2196 * rr_padding_x fields, then update lustre_swab_mdt_rec_reint() also.
2198 struct mdt_rec_reint {
2206 __u32 rr_suppgid1_h;
2208 __u32 rr_suppgid2_h;
2209 struct lu_fid rr_fid1;
2210 struct lu_fid rr_fid2;
2219 __u32 rr_padding_2; /* also fix lustre_swab_mdt_rec_reint */
2220 __u32 rr_padding_3; /* also fix lustre_swab_mdt_rec_reint */
2221 __u32 rr_padding_4; /* also fix lustre_swab_mdt_rec_reint */
2224 extern void lustre_swab_mdt_rec_reint(struct mdt_rec_reint *rr);
2227 __u32 ld_tgt_count; /* how many MDS's */
2228 __u32 ld_active_tgt_count; /* how many active */
2229 __u32 ld_default_stripe_count; /* how many objects are used */
2230 __u32 ld_pattern; /* default MEA_MAGIC_* */
2231 __u64 ld_default_hash_size;
2232 __u64 ld_padding_1; /* also fix lustre_swab_lmv_desc */
2233 __u32 ld_padding_2; /* also fix lustre_swab_lmv_desc */
2234 __u32 ld_qos_maxage; /* in second */
2235 __u32 ld_padding_3; /* also fix lustre_swab_lmv_desc */
2236 __u32 ld_padding_4; /* also fix lustre_swab_lmv_desc */
2237 struct obd_uuid ld_uuid;
2240 extern void lustre_swab_lmv_desc (struct lmv_desc *ld);
2242 /* TODO: lmv_stripe_md should contain mds capabilities for all slave fids */
2243 struct lmv_stripe_md {
2248 char mea_pool_name[LOV_MAXPOOLNAME];
2249 struct lu_fid mea_ids[0];
2252 extern void lustre_swab_lmv_stripe_md(struct lmv_stripe_md *mea);
2254 /* lmv structures */
2255 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
2256 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
2257 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
2259 #define MAX_HASH_SIZE_32 0x7fffffffUL
2260 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
2261 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
2266 FLD_FIRST_OPC = FLD_QUERY
2272 SEQ_FIRST_OPC = SEQ_QUERY
2276 SEQ_ALLOC_SUPER = 0,
2281 * LOV data structures
2284 #define LOV_MAX_UUID_BUFFER_SIZE 8192
2285 /* The size of the buffer the lov/mdc reserves for the
2286 * array of UUIDs returned by the MDS. With the current
2287 * protocol, this will limit the max number of OSTs per LOV */
2289 #define LOV_DESC_MAGIC 0xB0CCDE5C
2291 /* LOV settings descriptor (should only contain static info) */
2293 __u32 ld_tgt_count; /* how many OBD's */
2294 __u32 ld_active_tgt_count; /* how many active */
2295 __u32 ld_default_stripe_count; /* how many objects are used */
2296 __u32 ld_pattern; /* default PATTERN_RAID0 */
2297 __u64 ld_default_stripe_size; /* in bytes */
2298 __u64 ld_default_stripe_offset; /* in bytes */
2299 __u32 ld_padding_0; /* unused */
2300 __u32 ld_qos_maxage; /* in second */
2301 __u32 ld_padding_1; /* also fix lustre_swab_lov_desc */
2302 __u32 ld_padding_2; /* also fix lustre_swab_lov_desc */
2303 struct obd_uuid ld_uuid;
2306 #define ld_magic ld_active_tgt_count /* for swabbing from llogs */
2308 extern void lustre_swab_lov_desc (struct lov_desc *ld);
2313 /* opcodes -- MUST be distinct from OST/MDS opcodes */
2318 LDLM_BL_CALLBACK = 104,
2319 LDLM_CP_CALLBACK = 105,
2320 LDLM_GL_CALLBACK = 106,
2321 LDLM_SET_INFO = 107,
2324 #define LDLM_FIRST_OPC LDLM_ENQUEUE
2326 #define RES_NAME_SIZE 4
2327 struct ldlm_res_id {
2328 __u64 name[RES_NAME_SIZE];
2331 extern void lustre_swab_ldlm_res_id (struct ldlm_res_id *id);
2333 static inline int ldlm_res_eq(const struct ldlm_res_id *res0,
2334 const struct ldlm_res_id *res1)
2336 return !memcmp(res0, res1, sizeof(*res0));
2353 #define LCK_MODE_NUM 8
2363 #define LDLM_MIN_TYPE LDLM_PLAIN
2365 struct ldlm_extent {
2371 static inline int ldlm_extent_overlap(struct ldlm_extent *ex1,
2372 struct ldlm_extent *ex2)
2374 return (ex1->start <= ex2->end) && (ex2->start <= ex1->end);
2377 /* check if @ex1 contains @ex2 */
2378 static inline int ldlm_extent_contain(struct ldlm_extent *ex1,
2379 struct ldlm_extent *ex2)
2381 return (ex1->start <= ex2->start) && (ex1->end >= ex2->end);
2384 struct ldlm_inodebits {
2388 struct ldlm_flock_wire {
2396 /* it's important that the fields of the ldlm_extent structure match
2397 * the first fields of the ldlm_flock structure because there is only
2398 * one ldlm_swab routine to process the ldlm_policy_data_t union. if
2399 * this ever changes we will need to swab the union differently based
2400 * on the resource type. */
2403 struct ldlm_extent l_extent;
2404 struct ldlm_flock_wire l_flock;
2405 struct ldlm_inodebits l_inodebits;
2406 } ldlm_wire_policy_data_t;
2408 extern void lustre_swab_ldlm_policy_data (ldlm_wire_policy_data_t *d);
2410 /* Similarly to ldlm_wire_policy_data_t, there is one common swabber for all
2411 * LVB types. As a result, any new LVB structure must match the fields of the
2412 * ost_lvb structure. */
2413 union ldlm_wire_lvb {
2414 struct ost_lvb l_ost;
2417 extern void lustre_swab_lvb(union ldlm_wire_lvb *);
2419 struct ldlm_intent {
2423 extern void lustre_swab_ldlm_intent (struct ldlm_intent *i);
2425 struct ldlm_resource_desc {
2426 ldlm_type_t lr_type;
2427 __u32 lr_padding; /* also fix lustre_swab_ldlm_resource_desc */
2428 struct ldlm_res_id lr_name;
2431 extern void lustre_swab_ldlm_resource_desc (struct ldlm_resource_desc *r);
2433 struct ldlm_lock_desc {
2434 struct ldlm_resource_desc l_resource;
2435 ldlm_mode_t l_req_mode;
2436 ldlm_mode_t l_granted_mode;
2437 ldlm_wire_policy_data_t l_policy_data;
2440 extern void lustre_swab_ldlm_lock_desc (struct ldlm_lock_desc *l);
2442 #define LDLM_LOCKREQ_HANDLES 2
2443 #define LDLM_ENQUEUE_CANCEL_OFF 1
2445 struct ldlm_request {
2448 struct ldlm_lock_desc lock_desc;
2449 struct lustre_handle lock_handle[LDLM_LOCKREQ_HANDLES];
2452 extern void lustre_swab_ldlm_request (struct ldlm_request *rq);
2454 /* If LDLM_ENQUEUE, 1 slot is already occupied, 1 is available.
2455 * Otherwise, 2 are available. */
2456 #define ldlm_request_bufsize(count,type) \
2458 int _avail = LDLM_LOCKREQ_HANDLES; \
2459 _avail -= (type == LDLM_ENQUEUE ? LDLM_ENQUEUE_CANCEL_OFF : 0); \
2460 sizeof(struct ldlm_request) + \
2461 (count > _avail ? count - _avail : 0) * \
2462 sizeof(struct lustre_handle); \
2467 __u32 lock_padding; /* also fix lustre_swab_ldlm_reply */
2468 struct ldlm_lock_desc lock_desc;
2469 struct lustre_handle lock_handle;
2470 __u64 lock_policy_res1;
2471 __u64 lock_policy_res2;
2474 extern void lustre_swab_ldlm_reply (struct ldlm_reply *r);
2477 * Opcodes for mountconf (mgs and mgc)
2482 MGS_EXCEPTION, /* node died, etc. */
2483 MGS_TARGET_REG, /* whenever target starts up */
2489 #define MGS_FIRST_OPC MGS_CONNECT
2491 #define MGS_PARAM_MAXLEN 1024
2492 #define KEY_SET_INFO "set_info"
2494 struct mgs_send_param {
2495 char mgs_param[MGS_PARAM_MAXLEN];
2498 /* We pass this info to the MGS so it can write config logs */
2499 #define MTI_NAME_MAXLEN 64
2500 #define MTI_PARAM_MAXLEN 4096
2501 #define MTI_NIDS_MAX 32
2502 struct mgs_target_info {
2503 __u32 mti_lustre_ver;
2504 __u32 mti_stripe_index;
2505 __u32 mti_config_ver;
2507 __u32 mti_nid_count;
2508 __u32 mti_instance; /* Running instance of target */
2509 char mti_fsname[MTI_NAME_MAXLEN];
2510 char mti_svname[MTI_NAME_MAXLEN];
2511 char mti_uuid[sizeof(struct obd_uuid)];
2512 __u64 mti_nids[MTI_NIDS_MAX]; /* host nids (lnet_nid_t)*/
2513 char mti_params[MTI_PARAM_MAXLEN];
2515 extern void lustre_swab_mgs_target_info(struct mgs_target_info *oinfo);
2517 struct mgs_nidtbl_entry {
2518 __u64 mne_version; /* table version of this entry */
2519 __u32 mne_instance; /* target instance # */
2520 __u32 mne_index; /* target index */
2521 __u32 mne_length; /* length of this entry - by bytes */
2522 __u8 mne_type; /* target type LDD_F_SV_TYPE_OST/MDT */
2523 __u8 mne_nid_type; /* type of nid(mbz). for ipv6. */
2524 __u8 mne_nid_size; /* size of each NID, by bytes */
2525 __u8 mne_nid_count; /* # of NIDs in buffer */
2527 lnet_nid_t nids[0]; /* variable size buffer for NIDs. */
2530 extern void lustre_swab_mgs_nidtbl_entry(struct mgs_nidtbl_entry *oinfo);
2532 struct mgs_config_body {
2533 char mcb_name[MTI_NAME_MAXLEN]; /* logname */
2534 __u64 mcb_offset; /* next index of config log to request */
2535 __u16 mcb_type; /* type of log: CONFIG_T_[CONFIG|RECOVER] */
2537 __u8 mcb_bits; /* bits unit size of config log */
2538 __u32 mcb_units; /* # of units for bulk transfer */
2540 extern void lustre_swab_mgs_config_body(struct mgs_config_body *body);
2542 struct mgs_config_res {
2543 __u64 mcr_offset; /* index of last config log */
2544 __u64 mcr_size; /* size of the log */
2546 extern void lustre_swab_mgs_config_res(struct mgs_config_res *body);
2548 /* Config marker flags (in config log) */
2549 #define CM_START 0x01
2551 #define CM_SKIP 0x04
2552 #define CM_UPGRADE146 0x08
2553 #define CM_EXCLUDE 0x10
2554 #define CM_START_SKIP (CM_START | CM_SKIP)
2557 __u32 cm_step; /* aka config version */
2559 __u32 cm_vers; /* lustre release version number */
2560 __u32 cm_padding; /* 64 bit align */
2561 obd_time cm_createtime; /*when this record was first created */
2562 obd_time cm_canceltime; /*when this record is no longer valid*/
2563 char cm_tgtname[MTI_NAME_MAXLEN];
2564 char cm_comment[MTI_NAME_MAXLEN];
2567 extern void lustre_swab_cfg_marker(struct cfg_marker *marker,
2568 int swab, int size);
2571 * Opcodes for multiple servers.
2580 #define OBD_FIRST_OPC OBD_PING
2582 /* catalog of log objects */
2584 /** Identifier for a single log object */
2589 } __attribute__((packed));
2591 /** Records written to the CATALOGS list */
2592 #define CATLIST "CATALOGS"
2594 struct llog_logid lci_logid;
2598 } __attribute__((packed));
2600 /* Log data record types - there is no specific reason that these need to
2601 * be related to the RPC opcodes, but no reason not to (may be handy later?)
2603 #define LLOG_OP_MAGIC 0x10600000
2604 #define LLOG_OP_MASK 0xfff00000
2607 LLOG_PAD_MAGIC = LLOG_OP_MAGIC | 0x00000,
2608 OST_SZ_REC = LLOG_OP_MAGIC | 0x00f00,
2609 /* OST_RAID1_REC = LLOG_OP_MAGIC | 0x01000, never used */
2610 MDS_UNLINK_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) |
2611 REINT_UNLINK, /* obsolete after 2.5.0 */
2612 MDS_UNLINK64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
2614 /* MDS_SETATTR_REC = LLOG_OP_MAGIC | 0x12401, obsolete 1.8.0 */
2615 MDS_SETATTR64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
2617 OBD_CFG_REC = LLOG_OP_MAGIC | 0x20000,
2618 /* PTL_CFG_REC = LLOG_OP_MAGIC | 0x30000, obsolete 1.4.0 */
2619 LLOG_GEN_REC = LLOG_OP_MAGIC | 0x40000,
2620 /* LLOG_JOIN_REC = LLOG_OP_MAGIC | 0x50000, obsolete 1.8.0 */
2621 CHANGELOG_REC = LLOG_OP_MAGIC | 0x60000,
2622 CHANGELOG_USER_REC = LLOG_OP_MAGIC | 0x70000,
2623 LLOG_HDR_MAGIC = LLOG_OP_MAGIC | 0x45539,
2624 LLOG_LOGID_MAGIC = LLOG_OP_MAGIC | 0x4553b,
2627 #define LLOG_REC_HDR_NEEDS_SWABBING(r) \
2628 (((r)->lrh_type & __swab32(LLOG_OP_MASK)) == __swab32(LLOG_OP_MAGIC))
2630 /** Log record header - stored in little endian order.
2631 * Each record must start with this struct, end with a llog_rec_tail,
2632 * and be a multiple of 256 bits in size.
2634 struct llog_rec_hdr {
2641 struct llog_rec_tail {
2646 struct llog_logid_rec {
2647 struct llog_rec_hdr lid_hdr;
2648 struct llog_logid lid_id;
2652 struct llog_rec_tail lid_tail;
2653 } __attribute__((packed));
2655 struct llog_unlink_rec {
2656 struct llog_rec_hdr lur_hdr;
2659 obd_count lur_count;
2660 struct llog_rec_tail lur_tail;
2661 } __attribute__((packed));
2663 struct llog_unlink64_rec {
2664 struct llog_rec_hdr lur_hdr;
2665 struct lu_fid lur_fid;
2666 obd_count lur_count; /* to destroy the lost precreated */
2670 struct llog_rec_tail lur_tail;
2671 } __attribute__((packed));
2673 struct llog_setattr64_rec {
2674 struct llog_rec_hdr lsr_hdr;
2682 struct llog_rec_tail lsr_tail;
2683 } __attribute__((packed));
2685 struct llog_size_change_rec {
2686 struct llog_rec_hdr lsc_hdr;
2687 struct ll_fid lsc_fid;
2692 struct llog_rec_tail lsc_tail;
2693 } __attribute__((packed));
2695 #define CHANGELOG_MAGIC 0xca103000
2697 /** \a changelog_rec_type's that can't be masked */
2698 #define CHANGELOG_MINMASK (1 << CL_MARK)
2699 /** bits covering all \a changelog_rec_type's */
2700 #define CHANGELOG_ALLMASK 0XFFFFFFFF
2701 /** default \a changelog_rec_type mask */
2702 #define CHANGELOG_DEFMASK CHANGELOG_ALLMASK & ~(1 << CL_ATIME | 1 << CL_CLOSE)
2704 /* changelog llog name, needed by client replicators */
2705 #define CHANGELOG_CATALOG "changelog_catalog"
2707 struct changelog_setinfo {
2710 } __attribute__((packed));
2712 /** changelog record */
2713 struct llog_changelog_rec {
2714 struct llog_rec_hdr cr_hdr;
2715 struct changelog_rec cr;
2716 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
2717 } __attribute__((packed));
2719 struct llog_changelog_ext_rec {
2720 struct llog_rec_hdr cr_hdr;
2721 struct changelog_ext_rec cr;
2722 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
2723 } __attribute__((packed));
2725 #define CHANGELOG_USER_PREFIX "cl"
2727 struct llog_changelog_user_rec {
2728 struct llog_rec_hdr cur_hdr;
2732 struct llog_rec_tail cur_tail;
2733 } __attribute__((packed));
2735 /* Old llog gen for compatibility */
2739 } __attribute__((packed));
2741 struct llog_gen_rec {
2742 struct llog_rec_hdr lgr_hdr;
2743 struct llog_gen lgr_gen;
2747 struct llog_rec_tail lgr_tail;
2750 /* On-disk header structure of each log object, stored in little endian order */
2751 #define LLOG_CHUNK_SIZE 8192
2752 #define LLOG_HEADER_SIZE (96)
2753 #define LLOG_BITMAP_BYTES (LLOG_CHUNK_SIZE - LLOG_HEADER_SIZE)
2755 #define LLOG_MIN_REC_SIZE (24) /* round(llog_rec_hdr + llog_rec_tail) */
2757 /* flags for the logs */
2759 LLOG_F_ZAP_WHEN_EMPTY = 0x1,
2760 LLOG_F_IS_CAT = 0x2,
2761 LLOG_F_IS_PLAIN = 0x4,
2764 struct llog_log_hdr {
2765 struct llog_rec_hdr llh_hdr;
2766 obd_time llh_timestamp;
2768 __u32 llh_bitmap_offset;
2772 /* for a catalog the first plain slot is next to it */
2773 struct obd_uuid llh_tgtuuid;
2774 __u32 llh_reserved[LLOG_HEADER_SIZE/sizeof(__u32) - 23];
2775 __u32 llh_bitmap[LLOG_BITMAP_BYTES/sizeof(__u32)];
2776 struct llog_rec_tail llh_tail;
2777 } __attribute__((packed));
2779 #define LLOG_BITMAP_SIZE(llh) (__u32)((llh->llh_hdr.lrh_len - \
2780 llh->llh_bitmap_offset - \
2781 sizeof(llh->llh_tail)) * 8)
2783 /** log cookies are used to reference a specific log file and a record therein */
2784 struct llog_cookie {
2785 struct llog_logid lgc_lgl;
2789 } __attribute__((packed));
2791 /** llog protocol */
2792 enum llogd_rpc_ops {
2793 LLOG_ORIGIN_HANDLE_CREATE = 501,
2794 LLOG_ORIGIN_HANDLE_NEXT_BLOCK = 502,
2795 LLOG_ORIGIN_HANDLE_READ_HEADER = 503,
2796 LLOG_ORIGIN_HANDLE_WRITE_REC = 504,
2797 LLOG_ORIGIN_HANDLE_CLOSE = 505,
2798 LLOG_ORIGIN_CONNECT = 506,
2799 LLOG_CATINFO = 507, /* deprecated */
2800 LLOG_ORIGIN_HANDLE_PREV_BLOCK = 508,
2801 LLOG_ORIGIN_HANDLE_DESTROY = 509, /* for destroy llog object*/
2803 LLOG_FIRST_OPC = LLOG_ORIGIN_HANDLE_CREATE
2807 struct llog_logid lgd_logid;
2809 __u32 lgd_llh_flags;
2811 __u32 lgd_saved_index;
2813 __u64 lgd_cur_offset;
2814 } __attribute__((packed));
2816 struct llogd_conn_body {
2817 struct llog_gen lgdc_gen;
2818 struct llog_logid lgdc_logid;
2819 __u32 lgdc_ctxt_idx;
2820 } __attribute__((packed));
2822 /* Note: 64-bit types are 64-bit aligned in structure */
2824 obd_valid o_valid; /* hot fields in this obdo */
2826 obd_id o_parent_seq;
2827 obd_size o_size; /* o_size-o_blocks == ost_lvb */
2831 obd_blocks o_blocks; /* brw: cli sent cached bytes */
2834 /* 32-bit fields start here: keep an even number of them via padding */
2835 obd_blksize o_blksize; /* optimal IO blocksize */
2836 obd_mode o_mode; /* brw: cli sent cache remain */
2840 obd_count o_nlink; /* brw: checksum */
2841 obd_count o_parent_oid;
2842 obd_count o_misc; /* brw: o_dropped */
2843 __u64 o_ioepoch; /* epoch in ost writes */
2844 __u32 o_stripe_idx; /* holds stripe idx */
2846 struct lustre_handle o_handle; /* brw: lock handle to prolong
2848 struct llog_cookie o_lcookie; /* destroy: unlink cookie from
2853 __u64 o_data_version; /* getattr: sum of iversion for
2855 * brw: grant space consumed on
2856 * the client for the write */
2862 #define o_id o_oi.oi_id
2863 #define o_seq o_oi.oi_seq
2864 #define o_dirty o_blocks
2865 #define o_undirty o_mode
2866 #define o_dropped o_misc
2867 #define o_cksum o_nlink
2868 #define o_grant_used o_data_version
2870 static inline void lustre_set_wire_obdo(struct obdo *wobdo, struct obdo *lobdo)
2872 memcpy(wobdo, lobdo, sizeof(*lobdo));
2873 wobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
2876 static inline void lustre_get_wire_obdo(struct obdo *lobdo, struct obdo *wobdo)
2878 obd_flag local_flags = 0;
2880 if (lobdo->o_valid & OBD_MD_FLFLAGS)
2881 local_flags = lobdo->o_flags & OBD_FL_LOCAL_MASK;
2883 LASSERT(!(wobdo->o_flags & OBD_FL_LOCAL_MASK));
2885 memcpy(lobdo, wobdo, sizeof(*lobdo));
2886 if (local_flags != 0) {
2887 lobdo->o_valid |= OBD_MD_FLFLAGS;
2888 lobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
2889 lobdo->o_flags |= local_flags;
2893 extern void lustre_swab_obdo (struct obdo *o);
2895 /* request structure for OST's */
2900 /* Key for FIEMAP to be used in get_info calls */
2901 struct ll_fiemap_info_key {
2904 struct ll_user_fiemap fiemap;
2907 extern void lustre_swab_ost_body (struct ost_body *b);
2908 extern void lustre_swab_ost_last_id(obd_id *id);
2909 extern void lustre_swab_fiemap(struct ll_user_fiemap *fiemap);
2911 extern void lustre_swab_lov_user_md_v1(struct lov_user_md_v1 *lum);
2912 extern void lustre_swab_lov_user_md_v3(struct lov_user_md_v3 *lum);
2913 extern void lustre_swab_lov_user_md_objects(struct lov_user_ost_data *lod,
2915 extern void lustre_swab_lov_mds_md(struct lov_mds_md *lmm);
2918 extern void lustre_swab_llogd_body (struct llogd_body *d);
2919 extern void lustre_swab_llog_hdr (struct llog_log_hdr *h);
2920 extern void lustre_swab_llogd_conn_body (struct llogd_conn_body *d);
2921 extern void lustre_swab_llog_rec(struct llog_rec_hdr *rec);
2924 extern void lustre_swab_lustre_cfg(struct lustre_cfg *lcfg);
2926 /* Functions for dumping PTLRPC fields */
2927 void dump_rniobuf(struct niobuf_remote *rnb);
2928 void dump_ioo(struct obd_ioobj *nb);
2929 void dump_obdo(struct obdo *oa);
2930 void dump_ost_body(struct ost_body *ob);
2931 void dump_rcs(__u32 *rc);
2933 /* this will be used when OBD_CONNECT_CHANGE_QS is set */
2936 * ID appiles to (uid, gid)
2940 * LQUOTA_FLAGS_* affect the responding bits
2944 * acquire/release count (bytes for block quota)
2948 * when a master returns the reply to a slave, it will
2949 * contain the current corresponding qunit size
2955 #define QDATA_IS_GRP(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_GRP)
2956 #define QDATA_IS_BLK(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_BLK)
2957 #define QDATA_IS_ADJBLK(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_ADJBLK)
2958 #define QDATA_IS_ADJINO(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_ADJINO)
2959 #define QDATA_IS_CHANGE_QS(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_CHG_QS)
2961 #define QDATA_SET_GRP(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_GRP)
2962 #define QDATA_SET_BLK(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_BLK)
2963 #define QDATA_SET_ADJBLK(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_ADJBLK)
2964 #define QDATA_SET_ADJINO(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_ADJINO)
2965 #define QDATA_SET_CHANGE_QS(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_CHG_QS)
2967 #define QDATA_CLR_GRP(qdata) ((qdata)->qd_flags &= ~LQUOTA_FLAGS_GRP)
2968 #define QDATA_CLR_CHANGE_QS(qdata) ((qdata)->qd_flags &= ~LQUOTA_FLAGS_CHG_QS)
2970 extern void lustre_swab_qdata(struct qunit_data *d);
2971 extern struct qunit_data *quota_get_qdata(void *req, int is_req, int is_exp);
2972 extern int quota_copy_qdata(void *request, struct qunit_data *qdata,
2973 int is_req, int is_exp);
2980 #define QUOTA_FIRST_OPC QUOTA_DQACQ
2982 #define QUOTA_REQUEST 1
2983 #define QUOTA_REPLY 0
2984 #define QUOTA_EXPORT 1
2985 #define QUOTA_IMPORT 0
2987 /* quota check function */
2988 #define QUOTA_RET_OK 0 /**< return successfully */
2989 #define QUOTA_RET_NOQUOTA 1 /**< not support quota */
2990 #define QUOTA_RET_NOLIMIT 2 /**< quota limit isn't set */
2991 #define QUOTA_RET_ACQUOTA 4 /**< need to acquire extra quota */
2994 /* security opcodes */
2997 SEC_CTX_INIT_CONT = 802,
3000 SEC_FIRST_OPC = SEC_CTX_INIT
3004 * capa related definitions
3006 #define CAPA_HMAC_MAX_LEN 64
3007 #define CAPA_HMAC_KEY_MAX_LEN 56
3009 /* NB take care when changing the sequence of elements this struct,
3010 * because the offset info is used in find_capa() */
3011 struct lustre_capa {
3012 struct lu_fid lc_fid; /** fid */
3013 __u64 lc_opc; /** operations allowed */
3014 __u64 lc_uid; /** file owner */
3015 __u64 lc_gid; /** file group */
3016 __u32 lc_flags; /** HMAC algorithm & flags */
3017 __u32 lc_keyid; /** key# used for the capability */
3018 __u32 lc_timeout; /** capa timeout value (sec) */
3019 __u32 lc_expiry; /** expiry time (sec) */
3020 __u8 lc_hmac[CAPA_HMAC_MAX_LEN]; /** HMAC */
3021 } __attribute__((packed));
3023 extern void lustre_swab_lustre_capa(struct lustre_capa *c);
3025 /** lustre_capa::lc_opc */
3027 CAPA_OPC_BODY_WRITE = 1<<0, /**< write object data */
3028 CAPA_OPC_BODY_READ = 1<<1, /**< read object data */
3029 CAPA_OPC_INDEX_LOOKUP = 1<<2, /**< lookup object fid */
3030 CAPA_OPC_INDEX_INSERT = 1<<3, /**< insert object fid */
3031 CAPA_OPC_INDEX_DELETE = 1<<4, /**< delete object fid */
3032 CAPA_OPC_OSS_WRITE = 1<<5, /**< write oss object data */
3033 CAPA_OPC_OSS_READ = 1<<6, /**< read oss object data */
3034 CAPA_OPC_OSS_TRUNC = 1<<7, /**< truncate oss object */
3035 CAPA_OPC_OSS_DESTROY = 1<<8, /**< destroy oss object */
3036 CAPA_OPC_META_WRITE = 1<<9, /**< write object meta data */
3037 CAPA_OPC_META_READ = 1<<10, /**< read object meta data */
3040 #define CAPA_OPC_OSS_RW (CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE)
3041 #define CAPA_OPC_MDS_ONLY \
3042 (CAPA_OPC_BODY_WRITE | CAPA_OPC_BODY_READ | CAPA_OPC_INDEX_LOOKUP | \
3043 CAPA_OPC_INDEX_INSERT | CAPA_OPC_INDEX_DELETE)
3044 #define CAPA_OPC_OSS_ONLY \
3045 (CAPA_OPC_OSS_WRITE | CAPA_OPC_OSS_READ | CAPA_OPC_OSS_TRUNC | \
3046 CAPA_OPC_OSS_DESTROY)
3047 #define CAPA_OPC_MDS_DEFAULT ~CAPA_OPC_OSS_ONLY
3048 #define CAPA_OPC_OSS_DEFAULT ~(CAPA_OPC_MDS_ONLY | CAPA_OPC_OSS_ONLY)
3050 /* MDS capability covers object capability for operations of body r/w
3051 * (dir readpage/sendpage), index lookup/insert/delete and meta data r/w,
3052 * while OSS capability only covers object capability for operations of
3053 * oss data(file content) r/w/truncate.
3055 static inline int capa_for_mds(struct lustre_capa *c)
3057 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) != 0;
3060 static inline int capa_for_oss(struct lustre_capa *c)
3062 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) == 0;
3065 /* lustre_capa::lc_hmac_alg */
3067 CAPA_HMAC_ALG_SHA1 = 1, /**< sha1 algorithm */
3071 #define CAPA_FL_MASK 0x00ffffff
3072 #define CAPA_HMAC_ALG_MASK 0xff000000
3074 struct lustre_capa_key {
3075 __u64 lk_seq; /**< mds# */
3076 __u32 lk_keyid; /**< key# */
3078 __u8 lk_key[CAPA_HMAC_KEY_MAX_LEN]; /**< key */
3079 } __attribute__((packed));
3081 extern void lustre_swab_lustre_capa_key(struct lustre_capa_key *k);
3083 /** The link ea holds 1 \a link_ea_entry for each hardlink */
3084 #define LINK_EA_MAGIC 0x11EAF1DFUL
3085 struct link_ea_header {
3088 __u64 leh_len; /* total size */
3094 /** Hardlink data is name and parent fid.
3095 * Stored in this crazy struct for maximum packing and endian-neutrality
3097 struct link_ea_entry {
3098 /** __u16 stored big-endian, unaligned */
3099 unsigned char lee_reclen[2];
3100 unsigned char lee_parent_fid[sizeof(struct lu_fid)];
3102 }__attribute__((packed));
3104 /** fid2path request/reply structure */
3105 struct getinfo_fid2path {
3106 struct lu_fid gf_fid;
3111 } __attribute__((packed));
3113 void lustre_swab_fid2path (struct getinfo_fid2path *gf);