1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/include/lustre/lustre_idl.h
40 * Lustre wire protocol definitions.
43 /** \defgroup lustreidl lustreidl
45 * Lustre wire protocol definitions.
47 * ALL structs passing over the wire should be declared here. Structs
48 * that are used in interfaces with userspace should go in lustre_user.h.
50 * All structs being declared here should be built from simple fixed-size
51 * types (__u8, __u16, __u32, __u64) or be built from other types or
52 * structs also declared in this file. Similarly, all flags and magic
53 * values in those structs should also be declared here. This ensures
54 * that the Lustre wire protocol is not influenced by external dependencies.
56 * The only other acceptable items in this file are VERY SIMPLE accessor
57 * functions to avoid callers grubbing inside the structures, and the
58 * prototypes of the swabber functions for each struct. Nothing that
59 * depends on external functions or definitions should be in here.
61 * Structs must be properly aligned to put 64-bit values on an 8-byte
62 * boundary. Any structs being added here must also be added to
63 * utils/wirecheck.c and "make newwiretest" run to regenerate the
64 * utils/wiretest.c sources. This allows us to verify that wire structs
65 * have the proper alignment/size on all architectures.
67 * DO NOT CHANGE any of the structs, flags, values declared here and used
68 * in released Lustre versions. Some structs may have padding fields that
69 * can be used. Some structs might allow addition at the end (verify this
70 * in the code to ensure that new/old clients that see this larger struct
71 * do not fail, otherwise you need to implement protocol compatibility).
73 * We assume all nodes are either little-endian or big-endian, and we
74 * always send messages in the sender's native format. The receiver
75 * detects the message format by checking the 'magic' field of the message
76 * (see lustre_msg_swabbed() below).
78 * Each wire type has corresponding 'lustre_swab_xxxtypexxx()' routines,
79 * implemented either here, inline (trivial implementations) or in
80 * ptlrpc/pack_generic.c. These 'swabbers' convert the type from "other"
81 * endian, in-place in the message buffer.
83 * A swabber takes a single pointer argument. The caller must already have
84 * verified that the length of the message buffer >= sizeof (type).
86 * For variable length types, a second 'lustre_swab_v_xxxtypexxx()' routine
87 * may be defined that swabs just the variable part, after the caller has
88 * verified that the message buffer is large enough.
93 #ifndef _LUSTRE_IDL_H_
94 #define _LUSTRE_IDL_H_
96 #if !defined(LASSERT) && !defined(LPU64)
97 #include <libcfs/libcfs.h> /* for LASSERT, LPUX64, etc */
100 /* Defn's shared with user-space. */
101 #include <lustre/lustre_user.h>
106 /* FOO_REQUEST_PORTAL is for incoming requests on the FOO
107 * FOO_REPLY_PORTAL is for incoming replies on the FOO
108 * FOO_BULK_PORTAL is for incoming bulk on the FOO
111 #define CONNMGR_REQUEST_PORTAL 1
112 #define CONNMGR_REPLY_PORTAL 2
113 //#define OSC_REQUEST_PORTAL 3
114 #define OSC_REPLY_PORTAL 4
115 //#define OSC_BULK_PORTAL 5
116 #define OST_IO_PORTAL 6
117 #define OST_CREATE_PORTAL 7
118 #define OST_BULK_PORTAL 8
119 //#define MDC_REQUEST_PORTAL 9
120 #define MDC_REPLY_PORTAL 10
121 //#define MDC_BULK_PORTAL 11
122 #define MDS_REQUEST_PORTAL 12
123 //#define MDS_REPLY_PORTAL 13
124 #define MDS_BULK_PORTAL 14
125 #define LDLM_CB_REQUEST_PORTAL 15
126 #define LDLM_CB_REPLY_PORTAL 16
127 #define LDLM_CANCEL_REQUEST_PORTAL 17
128 #define LDLM_CANCEL_REPLY_PORTAL 18
129 //#define PTLBD_REQUEST_PORTAL 19
130 //#define PTLBD_REPLY_PORTAL 20
131 //#define PTLBD_BULK_PORTAL 21
132 #define MDS_SETATTR_PORTAL 22
133 #define MDS_READPAGE_PORTAL 23
134 #define MDS_MDS_PORTAL 24
136 #define MGC_REPLY_PORTAL 25
137 #define MGS_REQUEST_PORTAL 26
138 #define MGS_REPLY_PORTAL 27
139 #define OST_REQUEST_PORTAL 28
140 #define FLD_REQUEST_PORTAL 29
141 #define SEQ_METADATA_PORTAL 30
142 #define SEQ_DATA_PORTAL 31
143 #define SEQ_CONTROLLER_PORTAL 32
144 #define MGS_BULK_PORTAL 33
146 /* Portal 63 is reserved for the Cray Inc DVS - nic@cray.com, roe@cray.com, n8851@cray.com */
149 #define PTL_RPC_MSG_REQUEST 4711
150 #define PTL_RPC_MSG_ERR 4712
151 #define PTL_RPC_MSG_REPLY 4713
153 /* DON'T use swabbed values of MAGIC as magic! */
154 #define LUSTRE_MSG_MAGIC_V1 0x0BD00BD0
155 #define LUSTRE_MSG_MAGIC_V2 0x0BD00BD3
157 #define LUSTRE_MSG_MAGIC_V1_SWABBED 0xD00BD00B
158 #define LUSTRE_MSG_MAGIC_V2_SWABBED 0xD30BD00B
160 #define LUSTRE_MSG_MAGIC LUSTRE_MSG_MAGIC_V2
162 #define PTLRPC_MSG_VERSION 0x00000003
163 #define LUSTRE_VERSION_MASK 0xffff0000
164 #define LUSTRE_OBD_VERSION 0x00010000
165 #define LUSTRE_MDS_VERSION 0x00020000
166 #define LUSTRE_OST_VERSION 0x00030000
167 #define LUSTRE_DLM_VERSION 0x00040000
168 #define LUSTRE_LOG_VERSION 0x00050000
169 #define LUSTRE_MGS_VERSION 0x00060000
171 typedef __u32 mdsno_t;
172 typedef __u64 seqno_t;
173 typedef __u64 obd_id;
174 typedef __u64 obd_seq;
175 typedef __s64 obd_time;
176 typedef __u64 obd_size;
177 typedef __u64 obd_off;
178 typedef __u64 obd_blocks;
179 typedef __u64 obd_valid;
180 typedef __u32 obd_blksize;
181 typedef __u32 obd_mode;
182 typedef __u32 obd_uid;
183 typedef __u32 obd_gid;
184 typedef __u32 obd_flag;
185 typedef __u32 obd_count;
188 * Describes a range of sequence, lsr_start is included but lsr_end is
190 * Same structure is used in fld module where lsr_index field holds mdt id
194 #define LU_SEQ_RANGE_MDT 0x0
195 #define LU_SEQ_RANGE_OST 0x1
197 struct lu_seq_range {
205 * returns width of given range \a r
208 static inline __u64 range_space(const struct lu_seq_range *range)
210 return range->lsr_end - range->lsr_start;
214 * initialize range to zero
217 static inline void range_init(struct lu_seq_range *range)
219 range->lsr_start = range->lsr_end = range->lsr_index = 0;
223 * check if given seq id \a s is within given range \a r
226 static inline int range_within(const struct lu_seq_range *range,
229 return s >= range->lsr_start && s < range->lsr_end;
232 static inline int range_is_sane(const struct lu_seq_range *range)
234 return (range->lsr_end >= range->lsr_start);
237 static inline int range_is_zero(const struct lu_seq_range *range)
239 return (range->lsr_start == 0 && range->lsr_end == 0);
242 static inline int range_is_exhausted(const struct lu_seq_range *range)
245 return range_space(range) == 0;
248 /* return 0 if two range have the same location */
249 static inline int range_compare_loc(const struct lu_seq_range *r1,
250 const struct lu_seq_range *r2)
252 return r1->lsr_index != r2->lsr_index ||
253 r1->lsr_flags != r2->lsr_flags;
256 #define DRANGE "[%#16.16"LPF64"x-%#16.16"LPF64"x):%x:%x"
258 #define PRANGE(range) \
259 (range)->lsr_start, \
261 (range)->lsr_index, \
264 /** \defgroup lu_fid lu_fid
268 * Flags for lustre_mdt_attrs::lma_compat and lustre_mdt_attrs::lma_incompat.
271 LMAC_HSM = 0x00000001,
272 LMAC_SOM = 0x00000002,
276 * Masks for all features that should be supported by a Lustre version to
277 * access a specific file.
278 * This information is stored in lustre_mdt_attrs::lma_incompat.
280 * NOTE: No incompat feature should be added before bug #17670 is landed.
282 #define LMA_INCOMPAT_SUPP 0x0
285 * Following struct for MDT attributes, that will be kept inode's EA.
286 * Introduced in 2.0 release (please see b15993, for details)
288 struct lustre_mdt_attrs {
290 * Bitfield for supported data in this structure. From enum lma_compat.
291 * lma_self_fid and lma_flags are always available.
295 * Per-file incompat feature list. Lustre version should support all
296 * flags set in this field. The supported feature mask is available in
300 /** FID of this inode */
301 struct lu_fid lma_self_fid;
302 /** mdt/ost type, others */
304 /* IO Epoch SOM attributes belongs to */
306 /** total file size in objects */
308 /** total fs blocks in objects */
309 __u64 lma_som_blocks;
310 /** mds mount id the size is valid for */
311 __u64 lma_som_mountid;
315 * Fill \a lma with its first content.
316 * Only fid is stored.
318 static inline void lustre_lma_init(struct lustre_mdt_attrs *lma,
319 const struct lu_fid *fid)
322 lma->lma_incompat = 0;
323 memcpy(&lma->lma_self_fid, fid, sizeof(*fid));
325 lma->lma_ioepoch = 0;
326 lma->lma_som_size = 0;
327 lma->lma_som_blocks = 0;
328 lma->lma_som_mountid = 0;
330 /* If a field is added in struct lustre_mdt_attrs, zero it explicitly
331 * and change the test below. */
332 LASSERT(sizeof(*lma) ==
333 (offsetof(struct lustre_mdt_attrs, lma_som_mountid) +
334 sizeof(lma->lma_som_mountid)));
337 extern void lustre_swab_lu_fid(struct lu_fid *fid);
340 * Swab, if needed, lustre_mdt_attr struct to on-disk format.
341 * Otherwise, do not touch it.
343 static inline void lustre_lma_swab(struct lustre_mdt_attrs *lma)
345 /* Use LUSTRE_MSG_MAGIC to detect local endianess. */
346 if (LUSTRE_MSG_MAGIC != cpu_to_le32(LUSTRE_MSG_MAGIC)) {
347 __swab32s(&lma->lma_compat);
348 __swab32s(&lma->lma_incompat);
349 lustre_swab_lu_fid(&lma->lma_self_fid);
350 __swab64s(&lma->lma_flags);
351 __swab64s(&lma->lma_ioepoch);
352 __swab64s(&lma->lma_som_size);
353 __swab64s(&lma->lma_som_blocks);
354 __swab64s(&lma->lma_som_mountid);
358 /* This is the maximum number of MDTs allowed in CMD testing until such
359 * a time that FID-on-OST is implemented. This is due to the limitations
360 * of packing non-0-MDT numbers into the FID SEQ namespace. Once FID-on-OST
361 * is implemented this limit will be virtually unlimited. */
362 #define MAX_MDT_COUNT 8
369 /** initial fid id value */
370 LUSTRE_FID_INIT_OID = 1UL
373 /** returns fid object sequence */
374 static inline __u64 fid_seq(const struct lu_fid *fid)
379 /** returns fid object id */
380 static inline __u32 fid_oid(const struct lu_fid *fid)
385 /** returns fid object version */
386 static inline __u32 fid_ver(const struct lu_fid *fid)
391 static inline void fid_zero(struct lu_fid *fid)
393 memset(fid, 0, sizeof(*fid));
396 static inline obd_id fid_ver_oid(const struct lu_fid *fid)
398 return ((__u64)fid_ver(fid) << 32 | fid_oid(fid));
402 * Note that reserved SEQ numbers below 12 will conflict with ldiskfs
403 * inodes in the IGIF namespace, so these reserved SEQ numbers can be
404 * used for other purposes and not risk collisions with existing inodes.
406 * Different FID Format
407 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs#NEW.0
410 FID_SEQ_OST_MDT0 = 0,
413 FID_SEQ_OST_MDT1 = 3,
414 FID_SEQ_OST_MAX = 9, /* Max MDT count before OST_on_FID */
417 FID_SEQ_IGIF_MAX = 0x0ffffffffULL,
418 FID_SEQ_IDIF = 0x100000000ULL,
419 FID_SEQ_IDIF_MAX = 0x1ffffffffULL,
420 /* Normal FID sequence starts from this value, i.e. 1<<33 */
421 FID_SEQ_START = 0x200000000ULL,
422 FID_SEQ_LOCAL_FILE = 0x200000001ULL,
423 FID_SEQ_DOT_LUSTRE = 0x200000002ULL,
424 FID_SEQ_NORMAL = 0x200000400ULL,
425 FID_SEQ_LOV_DEFAULT= 0xffffffffffffffffULL
428 #define OBIF_OID_MAX_BITS 32
429 #define OBIF_MAX_OID (1ULL << OBIF_OID_MAX_BITS)
430 #define OBIF_OID_MASK ((1ULL << OBIF_OID_MAX_BITS) - 1)
431 #define IDIF_OID_MAX_BITS 48
432 #define IDIF_MAX_OID (1ULL << IDIF_OID_MAX_BITS)
433 #define IDIF_OID_MASK ((1ULL << IDIF_OID_MAX_BITS) - 1)
436 static inline int fid_seq_is_mdt0(obd_seq seq)
438 return (seq == FID_SEQ_OST_MDT0);
441 static inline int fid_seq_is_cmd(const __u64 seq)
443 return (seq >= FID_SEQ_OST_MDT1 && seq <= FID_SEQ_OST_MAX);
446 static inline int fid_seq_is_mdt(const __u64 seq)
448 return seq == FID_SEQ_OST_MDT0 ||
449 (seq >= FID_SEQ_OST_MDT1 && seq <= FID_SEQ_OST_MAX);
452 static inline int fid_seq_is_rsvd(const __u64 seq)
454 return seq <= FID_SEQ_RSVD;
457 static inline int fid_is_mdt0(const struct lu_fid *fid)
459 return fid_seq_is_mdt0(fid_seq(fid));
463 * Check if a fid is igif or not.
464 * \param fid the fid to be tested.
465 * \return true if the fid is a igif; otherwise false.
467 static inline int fid_seq_is_igif(const __u64 seq)
469 return seq >= FID_SEQ_IGIF && seq <= FID_SEQ_IGIF_MAX;
472 static inline int fid_is_igif(const struct lu_fid *fid)
474 return fid_seq_is_igif(fid_seq(fid));
478 * Check if a fid is idif or not.
479 * \param fid the fid to be tested.
480 * \return true if the fid is a idif; otherwise false.
482 static inline int fid_seq_is_idif(const __u64 seq)
484 return seq >= FID_SEQ_IDIF && seq <= FID_SEQ_IDIF_MAX;
487 static inline int fid_is_idif(const struct lu_fid *fid)
489 return fid_seq_is_idif(fid_seq(fid));
497 static inline int fid_seq_is_norm(const __u64 seq)
499 return (seq >= FID_SEQ_NORMAL);
502 static inline int fid_is_norm(const struct lu_fid *fid)
504 return fid_seq_is_norm(fid_seq(fid));
507 /* convert an OST objid into an IDIF FID SEQ number */
508 static inline obd_seq fid_idif_seq(obd_id id, __u32 ost_idx)
510 return FID_SEQ_IDIF | (ost_idx << 16) | ((id >> 32) & 0xffff);
513 /* convert a packed IDIF FID into an OST objid */
514 static inline obd_id fid_idif_id(obd_seq seq, __u32 oid, __u32 ver)
516 return ((__u64)ver << 48) | ((seq & 0xffff) << 32) | oid;
519 /* unpack an ostid (id/seq) from a wire/disk structure into an IDIF FID */
520 static inline void ostid_idif_unpack(struct ost_id *ostid,
521 struct lu_fid *fid, __u32 ost_idx)
523 fid->f_seq = fid_idif_seq(ostid->oi_id, ost_idx);
524 fid->f_oid = ostid->oi_id; /* truncate to 32 bits by assignment */
525 fid->f_ver = ostid->oi_id >> 48; /* in theory, not currently used */
528 /* unpack an ostid (id/seq) from a wire/disk structure into a non-IDIF FID */
529 static inline void ostid_fid_unpack(struct ost_id *ostid, struct lu_fid *fid)
531 fid->f_seq = ostid->oi_seq;
532 fid->f_oid = ostid->oi_id; /* truncate to 32 bits by assignment */
533 fid->f_ver = ostid->oi_id >> 32; /* in theory, not currently used */
536 /* Unpack an OST object id/seq (group) into a FID. This is needed for
537 * converting all obdo, lmm, lsm, etc. 64-bit id/seq pairs into proper
538 * FIDs. Note that if an id/seq is already in FID/IDIF format it will
539 * be passed through unchanged. Only legacy OST objects in "group 0"
540 * will be mapped into the IDIF namespace so that they can fit into the
541 * struct lu_fid fields without loss. For reference see:
542 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs
544 static inline int fid_ostid_unpack(struct lu_fid *fid, struct ost_id *ostid,
547 if (ost_idx > 0xffff) {
548 CERROR("bad ost_idx, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
549 ostid->oi_seq, ostid->oi_id, ost_idx);
553 if (fid_seq_is_mdt0(ostid->oi_seq)) {
554 /* This is a "legacy" (old 1.x/2.early) OST object in "group 0"
555 * that we map into the IDIF namespace. It allows up to 2^48
556 * objects per OST, as this is the object namespace that has
557 * been in production for years. This can handle create rates
558 * of 1M objects/s/OST for 9 years, or combinations thereof. */
559 if (ostid->oi_id >= IDIF_MAX_OID) {
560 CERROR("bad MDT0 id, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
561 ostid->oi_seq, ostid->oi_id, ost_idx);
564 ostid_idif_unpack(ostid, fid, ost_idx);
566 } else if (fid_seq_is_rsvd(ostid->oi_seq)) {
567 /* These are legacy OST objects for LLOG/ECHO and CMD testing.
568 * We only support 2^32 objects in these groups, and cannot
569 * uniquely identify them in the system (i.e. they are the
570 * duplicated on all OSTs), but this is not strictly required
571 * for the old object protocol, which has a separate ost_idx. */
572 if (ostid->oi_id >= 0xffffffffULL) {
573 CERROR("bad RSVD id, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
574 ostid->oi_seq, ostid->oi_id, ost_idx);
577 ostid_fid_unpack(ostid, fid);
579 } else if (unlikely(fid_seq_is_igif(ostid->oi_seq))) {
580 /* This is an MDT inode number, which should never collide with
581 * proper OST object IDs, and is probably a broken filesystem */
582 CERROR("bad IGIF, seq:"LPU64" id:"LPU64" ost_idx:%u\n",
583 ostid->oi_seq, ostid->oi_id, ost_idx);
586 } else /* if (fid_seq_is_idif(seq) || fid_seq_is_norm(seq)) */ {
587 /* This is either an IDIF object, which identifies objects across
588 * all OSTs, or a regular FID. The IDIF namespace maps legacy
589 * OST objects into the FID namespace. In both cases, we just
590 * pass the FID through, no conversion needed. */
591 ostid_fid_unpack(ostid, fid);
597 /* pack an IDIF FID into an ostid (id/seq) for the wire/disk */
598 static inline void ostid_idif_pack(struct lu_fid *fid, struct ost_id *ostid)
600 ostid->oi_seq = FID_SEQ_OST_MDT0;
601 ostid->oi_id = fid_idif_id(fid->f_seq, fid->f_oid, fid->f_ver);
604 /* pack a non-IDIF FID into an ostid (id/seq) for the wire/disk */
605 static inline void ostid_fid_pack(struct lu_fid *fid, struct ost_id *ostid)
607 ostid->oi_seq = fid_seq(fid);
608 ostid->oi_id = fid_ver_oid(fid);
611 /* pack any OST FID into an ostid (id/seq) for the wire/disk */
612 static inline int fid_ostid_pack(struct lu_fid *fid, struct ost_id *ostid)
614 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
615 CERROR("bad IGIF, "DFID"\n", PFID(fid));
619 if (fid_is_idif(fid))
620 ostid_idif_pack(fid, ostid);
622 ostid_fid_pack(fid, ostid);
627 /* extract OST sequence (group) from a wire ost_id (id/seq) pair */
628 static inline obd_seq ostid_seq(struct ost_id *ostid)
630 if (unlikely(fid_seq_is_igif(ostid->oi_seq)))
631 CWARN("bad IGIF, oi_seq: "LPU64" oi_id: "LPX64"\n",
632 ostid->oi_seq, ostid->oi_id);
634 if (unlikely(fid_seq_is_idif(ostid->oi_seq)))
635 return FID_SEQ_OST_MDT0;
637 return ostid->oi_seq;
640 /* extract OST objid from a wire ost_id (id/seq) pair */
641 static inline obd_id ostid_id(struct ost_id *ostid)
643 if (ostid->oi_seq == FID_SEQ_OST_MDT0)
644 return ostid->oi_id & IDIF_OID_MASK;
646 if (fid_seq_is_rsvd(ostid->oi_seq))
647 return ostid->oi_id & OBIF_OID_MASK;
649 if (fid_seq_is_idif(ostid->oi_seq))
650 return fid_idif_id(ostid->oi_seq, ostid->oi_id, 0);
656 * Get inode number from a igif.
657 * \param fid a igif to get inode number from.
658 * \return inode number for the igif.
660 static inline ino_t lu_igif_ino(const struct lu_fid *fid)
666 * Build igif from the inode number/generation.
668 #define LU_IGIF_BUILD(fid, ino, gen) \
674 static inline void lu_igif_build(struct lu_fid *fid, __u32 ino, __u32 gen)
676 LU_IGIF_BUILD(fid, ino, gen);
677 LASSERT(fid_is_igif(fid));
681 * Get inode generation from a igif.
682 * \param fid a igif to get inode generation from.
683 * \return inode generation for the igif.
685 static inline __u32 lu_igif_gen(const struct lu_fid *fid)
691 * Fids are transmitted across network (in the sender byte-ordering),
692 * and stored on disk in big-endian order.
694 static inline void fid_cpu_to_le(struct lu_fid *dst, const struct lu_fid *src)
696 /* check that all fields are converted */
697 CLASSERT(sizeof *src ==
698 sizeof fid_seq(src) +
699 sizeof fid_oid(src) + sizeof fid_ver(src));
700 LASSERTF(fid_is_igif(src) || fid_ver(src) == 0, DFID"\n", PFID(src));
701 dst->f_seq = cpu_to_le64(fid_seq(src));
702 dst->f_oid = cpu_to_le32(fid_oid(src));
703 dst->f_ver = cpu_to_le32(fid_ver(src));
706 static inline void fid_le_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
708 /* check that all fields are converted */
709 CLASSERT(sizeof *src ==
710 sizeof fid_seq(src) +
711 sizeof fid_oid(src) + sizeof fid_ver(src));
712 dst->f_seq = le64_to_cpu(fid_seq(src));
713 dst->f_oid = le32_to_cpu(fid_oid(src));
714 dst->f_ver = le32_to_cpu(fid_ver(src));
715 LASSERTF(fid_is_igif(dst) || fid_ver(dst) == 0, DFID"\n", PFID(dst));
718 static inline void fid_cpu_to_be(struct lu_fid *dst, const struct lu_fid *src)
720 /* check that all fields are converted */
721 CLASSERT(sizeof *src ==
722 sizeof fid_seq(src) +
723 sizeof fid_oid(src) + sizeof fid_ver(src));
724 LASSERTF(fid_is_igif(src) || fid_ver(src) == 0, DFID"\n", PFID(src));
725 dst->f_seq = cpu_to_be64(fid_seq(src));
726 dst->f_oid = cpu_to_be32(fid_oid(src));
727 dst->f_ver = cpu_to_be32(fid_ver(src));
730 static inline void fid_be_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
732 /* check that all fields are converted */
733 CLASSERT(sizeof *src ==
734 sizeof fid_seq(src) +
735 sizeof fid_oid(src) + sizeof fid_ver(src));
736 dst->f_seq = be64_to_cpu(fid_seq(src));
737 dst->f_oid = be32_to_cpu(fid_oid(src));
738 dst->f_ver = be32_to_cpu(fid_ver(src));
739 LASSERTF(fid_is_igif(dst) || fid_ver(dst) == 0, DFID"\n", PFID(dst));
742 static inline int fid_is_sane(const struct lu_fid *fid)
746 ((fid_seq(fid) >= FID_SEQ_START && fid_oid(fid) != 0
747 && fid_ver(fid) == 0) ||
751 static inline int fid_is_zero(const struct lu_fid *fid)
753 return fid_seq(fid) == 0 && fid_oid(fid) == 0;
756 extern void lustre_swab_lu_fid(struct lu_fid *fid);
757 extern void lustre_swab_lu_seq_range(struct lu_seq_range *range);
759 static inline int lu_fid_eq(const struct lu_fid *f0,
760 const struct lu_fid *f1)
762 /* Check that there is no alignment padding. */
763 CLASSERT(sizeof *f0 ==
764 sizeof f0->f_seq + sizeof f0->f_oid + sizeof f0->f_ver);
765 LASSERTF(fid_is_igif(f0) || fid_ver(f0) == 0, DFID, PFID(f0));
766 LASSERTF(fid_is_igif(f1) || fid_ver(f1) == 0, DFID, PFID(f1));
767 return memcmp(f0, f1, sizeof *f0) == 0;
770 #define __diff_normalize(val0, val1) \
772 typeof(val0) __val0 = (val0); \
773 typeof(val1) __val1 = (val1); \
775 (__val0 == __val1 ? 0 : __val0 > __val1 ? +1 : -1); \
778 static inline int lu_fid_cmp(const struct lu_fid *f0,
779 const struct lu_fid *f1)
782 __diff_normalize(fid_seq(f0), fid_seq(f1)) ?:
783 __diff_normalize(fid_oid(f0), fid_oid(f1)) ?:
784 __diff_normalize(fid_ver(f0), fid_ver(f1));
789 /** \defgroup lu_dir lu_dir
793 * Enumeration of possible directory entry attributes.
795 * Attributes follow directory entry header in the order they appear in this
798 enum lu_dirent_attrs {
801 LUDA_64BITHASH = 0x0004,
805 * Layout of readdir pages, as transmitted on wire.
808 /** valid if LUDA_FID is set. */
809 struct lu_fid lde_fid;
810 /** a unique entry identifier: a hash or an offset. */
812 /** total record length, including all attributes. */
816 /** optional variable size attributes following this entry.
817 * taken from enum lu_dirent_attrs.
820 /** name is followed by the attributes indicated in ->ldp_attrs, in
821 * their natural order. After the last attribute, padding bytes are
822 * added to make ->lde_reclen a multiple of 8.
828 * Definitions of optional directory entry attributes formats.
830 * Individual attributes do not have their length encoded in a generic way. It
831 * is assumed that consumer of an attribute knows its format. This means that
832 * it is impossible to skip over an unknown attribute, except by skipping over all
833 * remaining attributes (by using ->lde_reclen), which is not too
834 * constraining, because new server versions will append new attributes at
835 * the end of an entry.
839 * Fid directory attribute: a fid of an object referenced by the entry. This
840 * will be almost always requested by the client and supplied by the server.
842 * Aligned to 8 bytes.
844 /* To have compatibility with 1.8, lets have fid in lu_dirent struct. */
849 * Aligned to 2 bytes.
856 __u64 ldp_hash_start;
860 struct lu_dirent ldp_entries[0];
863 enum lu_dirpage_flags {
865 * dirpage contains no entry.
869 * last entry's lde_hash equals ldp_hash_end.
874 static inline struct lu_dirent *lu_dirent_start(struct lu_dirpage *dp)
876 if (le32_to_cpu(dp->ldp_flags) & LDF_EMPTY)
879 return dp->ldp_entries;
882 static inline struct lu_dirent *lu_dirent_next(struct lu_dirent *ent)
884 struct lu_dirent *next;
886 if (le16_to_cpu(ent->lde_reclen) != 0)
887 next = ((void *)ent) + le16_to_cpu(ent->lde_reclen);
894 static inline int lu_dirent_calc_size(int namelen, __u16 attr)
898 if (attr & LUDA_TYPE) {
899 const unsigned align = sizeof(struct luda_type) - 1;
900 size = (sizeof(struct lu_dirent) + namelen + align) & ~align;
901 size += sizeof(struct luda_type);
903 size = sizeof(struct lu_dirent) + namelen;
905 return (size + 7) & ~7;
908 static inline int lu_dirent_size(struct lu_dirent *ent)
910 if (le16_to_cpu(ent->lde_reclen) == 0) {
911 return lu_dirent_calc_size(le16_to_cpu(ent->lde_namelen),
912 le32_to_cpu(ent->lde_attrs));
914 return le16_to_cpu(ent->lde_reclen);
917 #define MDS_DIR_END_OFF 0xfffffffffffffffeULL
920 * MDS_READPAGE page size
922 * This is the directory page size packed in MDS_READPAGE RPC.
923 * It's different than CFS_PAGE_SIZE because the client needs to
924 * access the struct lu_dirpage header packed at the beginning of
925 * the "page" and without this there isn't any way to know find the
926 * lu_dirpage header is if client and server CFS_PAGE_SIZE differ.
928 #define LU_PAGE_SHIFT 12
929 #define LU_PAGE_SIZE (1UL << LU_PAGE_SHIFT)
930 #define LU_PAGE_MASK (~(LU_PAGE_SIZE - 1))
932 #define LU_PAGE_COUNT 1 << (CFS_PAGE_SHIFT - LU_PAGE_SHIFT)
936 struct lustre_handle {
939 #define DEAD_HANDLE_MAGIC 0xdeadbeefcafebabeULL
941 static inline int lustre_handle_is_used(struct lustre_handle *lh)
943 return lh->cookie != 0ull;
946 static inline int lustre_handle_equal(const struct lustre_handle *lh1,
947 const struct lustre_handle *lh2)
949 return lh1->cookie == lh2->cookie;
952 static inline void lustre_handle_copy(struct lustre_handle *tgt,
953 struct lustre_handle *src)
955 tgt->cookie = src->cookie;
958 /* flags for lm_flags */
959 #define MSGHDR_AT_SUPPORT 0x1
960 #define MSGHDR_CKSUM_INCOMPAT18 0x2
962 #define lustre_msg lustre_msg_v2
963 /* we depend on this structure to be 8-byte aligned */
964 /* this type is only endian-adjusted in lustre_unpack_msg() */
965 struct lustre_msg_v2 {
977 /* without gss, ptlrpc_body is put at the first buffer. */
978 #define PTLRPC_NUM_VERSIONS 4
980 struct lustre_handle pb_handle;
987 __u64 pb_last_committed;
992 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
993 __u32 pb_service_time; /* for rep, actual service time, also used for
994 net_latency of req */
997 /* VBR: pre-versions */
998 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
999 /* padding for future needs */
1000 __u64 pb_padding[4];
1003 extern void lustre_swab_ptlrpc_body(struct ptlrpc_body *pb);
1005 /* message body offset for lustre_msg_v2 */
1006 /* ptlrpc body offset in all request/reply messages */
1007 #define MSG_PTLRPC_BODY_OFF 0
1009 /* normal request/reply message record offset */
1010 #define REQ_REC_OFF 1
1011 #define REPLY_REC_OFF 1
1013 /* ldlm request message body offset */
1014 #define DLM_LOCKREQ_OFF 1 /* lockreq offset */
1015 #define DLM_REQ_REC_OFF 2 /* normal dlm request record offset */
1017 /* ldlm intent lock message body offset */
1018 #define DLM_INTENT_IT_OFF 2 /* intent lock it offset */
1019 #define DLM_INTENT_REC_OFF 3 /* intent lock record offset */
1021 /* ldlm reply message body offset */
1022 #define DLM_LOCKREPLY_OFF 1 /* lockrep offset */
1023 #define DLM_REPLY_REC_OFF 2 /* reply record offset */
1025 /** only use in req->rq_{req,rep}_swab_mask */
1026 #define MSG_PTLRPC_HEADER_OFF 31
1028 /* Flags that are operation-specific go in the top 16 bits. */
1029 #define MSG_OP_FLAG_MASK 0xffff0000
1030 #define MSG_OP_FLAG_SHIFT 16
1032 /* Flags that apply to all requests are in the bottom 16 bits */
1033 #define MSG_GEN_FLAG_MASK 0x0000ffff
1034 #define MSG_LAST_REPLAY 0x0001
1035 #define MSG_RESENT 0x0002
1036 #define MSG_REPLAY 0x0004
1037 /* #define MSG_AT_SUPPORT 0x0008
1038 * This was used in early prototypes of adaptive timeouts, and while there
1039 * shouldn't be any users of that code there also isn't a need for using this
1040 * bits. Defer usage until at least 1.10 to avoid potential conflict. */
1041 #define MSG_DELAY_REPLAY 0x0010
1042 #define MSG_VERSION_REPLAY 0x0020
1043 #define MSG_REQ_REPLAY_DONE 0x0040
1044 #define MSG_LOCK_REPLAY_DONE 0x0080
1047 * Flags for all connect opcodes (MDS_CONNECT, OST_CONNECT)
1050 #define MSG_CONNECT_RECOVERING 0x00000001
1051 #define MSG_CONNECT_RECONNECT 0x00000002
1052 #define MSG_CONNECT_REPLAYABLE 0x00000004
1053 //#define MSG_CONNECT_PEER 0x8
1054 #define MSG_CONNECT_LIBCLIENT 0x00000010
1055 #define MSG_CONNECT_INITIAL 0x00000020
1056 #define MSG_CONNECT_ASYNC 0x00000040
1057 #define MSG_CONNECT_NEXT_VER 0x00000080 /* use next version of lustre_msg */
1058 #define MSG_CONNECT_TRANSNO 0x00000100 /* report transno */
1061 #define OBD_CONNECT_RDONLY 0x1ULL /*client has read-only access*/
1062 #define OBD_CONNECT_INDEX 0x2ULL /*connect specific LOV idx */
1063 #define OBD_CONNECT_MDS 0x4ULL /*connect from MDT to OST */
1064 #define OBD_CONNECT_GRANT 0x8ULL /*OSC gets grant at connect */
1065 #define OBD_CONNECT_SRVLOCK 0x10ULL /*server takes locks for cli */
1066 #define OBD_CONNECT_VERSION 0x20ULL /*Lustre versions in ocd */
1067 #define OBD_CONNECT_REQPORTAL 0x40ULL /*Separate non-IO req portal */
1068 #define OBD_CONNECT_ACL 0x80ULL /*access control lists */
1069 #define OBD_CONNECT_XATTR 0x100ULL /*client use extended attr */
1070 #define OBD_CONNECT_CROW 0x200ULL /*MDS+OST create obj on write*/
1071 #define OBD_CONNECT_TRUNCLOCK 0x400ULL /*locks on server for punch */
1072 #define OBD_CONNECT_TRANSNO 0x800ULL /*replay sends init transno */
1073 #define OBD_CONNECT_IBITS 0x1000ULL /*support for inodebits locks*/
1074 #define OBD_CONNECT_JOIN 0x2000ULL /*files can be concatenated.
1075 *We do not support JOIN FILE
1076 *anymore, reserve this flags
1077 *just for preventing such bit
1079 #define OBD_CONNECT_ATTRFID 0x4000ULL /*Server can GetAttr By Fid*/
1080 #define OBD_CONNECT_NODEVOH 0x8000ULL /*No open hndl on specl nodes*/
1081 #define OBD_CONNECT_RMT_CLIENT 0x10000ULL /*Remote client */
1082 #define OBD_CONNECT_RMT_CLIENT_FORCE 0x20000ULL /*Remote client by force */
1083 #define OBD_CONNECT_BRW_SIZE 0x40000ULL /*Max bytes per rpc */
1084 #define OBD_CONNECT_QUOTA64 0x80000ULL /*64bit qunit_data.qd_count */
1085 #define OBD_CONNECT_MDS_CAPA 0x100000ULL /*MDS capability */
1086 #define OBD_CONNECT_OSS_CAPA 0x200000ULL /*OSS capability */
1087 #define OBD_CONNECT_CANCELSET 0x400000ULL /*Early batched cancels. */
1088 #define OBD_CONNECT_SOM 0x800000ULL /*Size on MDS */
1089 #define OBD_CONNECT_AT 0x1000000ULL /*client uses AT */
1090 #define OBD_CONNECT_LRU_RESIZE 0x2000000ULL /*LRU resize feature. */
1091 #define OBD_CONNECT_MDS_MDS 0x4000000ULL /*MDS-MDS connection */
1092 #define OBD_CONNECT_REAL 0x8000000ULL /*real connection */
1093 #define OBD_CONNECT_CHANGE_QS 0x10000000ULL /*shrink/enlarge qunit */
1094 #define OBD_CONNECT_CKSUM 0x20000000ULL /*support several cksum algos*/
1095 #define OBD_CONNECT_FID 0x40000000ULL /*FID is supported by server */
1096 #define OBD_CONNECT_VBR 0x80000000ULL /*version based recovery */
1097 #define OBD_CONNECT_LOV_V3 0x100000000ULL /*client supports LOV v3 EA */
1098 #define OBD_CONNECT_GRANT_SHRINK 0x200000000ULL /* support grant shrink */
1099 #define OBD_CONNECT_SKIP_ORPHAN 0x400000000ULL /* don't reuse orphan objids */
1100 #define OBD_CONNECT_MAX_EASIZE 0x800000000ULL /* preserved for large EA */
1101 #define OBD_CONNECT_FULL20 0x1000000000ULL /* it is 2.0 client */
1102 #define OBD_CONNECT_LAYOUTLOCK 0x2000000000ULL /* client supports layout lock */
1103 #define OBD_CONNECT_64BITHASH 0x4000000000ULL /* client supports 64-bits
1105 #define OBD_CONNECT_MAXBYTES 0x8000000000ULL /* max stripe size */
1106 #define OBD_CONNECT_IMP_RECOV 0x10000000000ULL /* imp recovery support */
1107 #define OBD_CONNECT_JOBSTATS 0x20000000000ULL /* jobid in ptlrpc_body */
1108 #define OBD_CONNECT_GRANT_PARAM 0x40000000000ULL /* additional grant parameters
1109 * are passed at connect time
1110 * to have finer grant space
1112 #define OBD_CONNECT_EINPROGRESS 0x80000000000ULL /* client can handle the
1113 * -EINPROGRESS error for write
1116 #define OCD_HAS_FLAG(ocd, flg) \
1117 (!!((ocd)->ocd_connect_flags & OBD_CONNECT_##flg))
1119 /* also update obd_connect_names[] for lprocfs_rd_connect_flags()
1120 * and lustre/utils/wirecheck.c */
1122 #ifdef HAVE_LRU_RESIZE_SUPPORT
1123 #define LRU_RESIZE_CONNECT_FLAG OBD_CONNECT_LRU_RESIZE
1125 #define LRU_RESIZE_CONNECT_FLAG 0
1128 #define MDT_CONNECT_SUPPORTED (OBD_CONNECT_RDONLY | OBD_CONNECT_VERSION | \
1129 OBD_CONNECT_ACL | OBD_CONNECT_XATTR | \
1130 OBD_CONNECT_IBITS | \
1131 OBD_CONNECT_NODEVOH | OBD_CONNECT_ATTRFID | \
1132 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1133 OBD_CONNECT_RMT_CLIENT | \
1134 OBD_CONNECT_RMT_CLIENT_FORCE | \
1135 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_MDS_CAPA | \
1136 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_MDS_MDS | \
1137 OBD_CONNECT_FID | LRU_RESIZE_CONNECT_FLAG | \
1138 OBD_CONNECT_VBR | OBD_CONNECT_LOV_V3 | \
1139 OBD_CONNECT_SOM | OBD_CONNECT_FULL20 | \
1140 OBD_CONNECT_64BITHASH)
1141 #define OST_CONNECT_SUPPORTED (OBD_CONNECT_SRVLOCK | OBD_CONNECT_GRANT | \
1142 OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | \
1143 OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_INDEX | \
1144 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_QUOTA64 | \
1145 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1146 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_CKSUM | \
1147 OBD_CONNECT_CHANGE_QS | \
1148 OBD_CONNECT_OSS_CAPA | \
1149 OBD_CONNECT_RMT_CLIENT | \
1150 OBD_CONNECT_RMT_CLIENT_FORCE | OBD_CONNECT_VBR | \
1151 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN | \
1152 OBD_CONNECT_GRANT_SHRINK | OBD_CONNECT_FULL20 | \
1153 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES | \
1154 OBD_CONNECT_MAX_EASIZE)
1155 #define ECHO_CONNECT_SUPPORTED (0)
1156 #define MGS_CONNECT_SUPPORTED (OBD_CONNECT_VERSION | OBD_CONNECT_AT | \
1157 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV)
1159 /* Features required for this version of the client to work with server */
1160 #define CLIENT_CONNECT_MDT_REQD (OBD_CONNECT_IBITS | OBD_CONNECT_FID | \
1163 #define OBD_OCD_VERSION(major,minor,patch,fix) (((major)<<24) + ((minor)<<16) +\
1164 ((patch)<<8) + (fix))
1165 #define OBD_OCD_VERSION_MAJOR(version) ((int)((version)>>24)&255)
1166 #define OBD_OCD_VERSION_MINOR(version) ((int)((version)>>16)&255)
1167 #define OBD_OCD_VERSION_PATCH(version) ((int)((version)>>8)&255)
1168 #define OBD_OCD_VERSION_FIX(version) ((int)(version)&255)
1170 /* This structure is used for both request and reply.
1172 * If we eventually have separate connect data for different types, which we
1173 * almost certainly will, then perhaps we stick a union in here. */
1174 struct obd_connect_data_v1 {
1175 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1176 __u32 ocd_version; /* lustre release version number */
1177 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1178 __u32 ocd_index; /* LOV index to connect to */
1179 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
1180 __u64 ocd_ibits_known; /* inode bits this client understands */
1181 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1182 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1183 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1184 __u32 ocd_unused; /* also fix lustre_swab_connect */
1185 __u64 ocd_transno; /* first transno from client to be replayed */
1186 __u32 ocd_group; /* MDS group on OST */
1187 __u32 ocd_cksum_types; /* supported checksum algorithms */
1188 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1189 __u32 ocd_instance; /* also fix lustre_swab_connect */
1190 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1193 struct obd_connect_data {
1194 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1195 __u32 ocd_version; /* lustre release version number */
1196 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1197 __u32 ocd_index; /* LOV index to connect to */
1198 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
1199 __u64 ocd_ibits_known; /* inode bits this client understands */
1200 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1201 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1202 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1203 __u32 ocd_unused; /* also fix lustre_swab_connect */
1204 __u64 ocd_transno; /* first transno from client to be replayed */
1205 __u32 ocd_group; /* MDS group on OST */
1206 __u32 ocd_cksum_types; /* supported checksum algorithms */
1207 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1208 __u32 ocd_instance; /* instance # of this target */
1209 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1210 /* Fields after ocd_maxbytes are only accessible by the receiver
1211 * if the corresponding flag in ocd_connect_flags is set. Accessing
1212 * any field after ocd_maxbytes on the receiver without a valid flag
1213 * may result in out-of-bound memory access and kernel oops. */
1214 __u64 padding1; /* added 2.1.0. also fix lustre_swab_connect */
1215 __u64 padding2; /* added 2.1.0. also fix lustre_swab_connect */
1216 __u64 padding3; /* added 2.1.0. also fix lustre_swab_connect */
1217 __u64 padding4; /* added 2.1.0. also fix lustre_swab_connect */
1218 __u64 padding5; /* added 2.1.0. also fix lustre_swab_connect */
1219 __u64 padding6; /* added 2.1.0. also fix lustre_swab_connect */
1220 __u64 padding7; /* added 2.1.0. also fix lustre_swab_connect */
1221 __u64 padding8; /* added 2.1.0. also fix lustre_swab_connect */
1222 __u64 padding9; /* added 2.1.0. also fix lustre_swab_connect */
1223 __u64 paddingA; /* added 2.1.0. also fix lustre_swab_connect */
1224 __u64 paddingB; /* added 2.1.0. also fix lustre_swab_connect */
1225 __u64 paddingC; /* added 2.1.0. also fix lustre_swab_connect */
1226 __u64 paddingD; /* added 2.1.0. also fix lustre_swab_connect */
1227 __u64 paddingE; /* added 2.1.0. also fix lustre_swab_connect */
1228 __u64 paddingF; /* added 2.1.0. also fix lustre_swab_connect */
1232 extern void lustre_swab_connect(struct obd_connect_data *ocd);
1235 * Supported checksum algorithms. Up to 32 checksum types are supported.
1236 * (32-bit mask stored in obd_connect_data::ocd_cksum_types)
1237 * Please update DECLARE_CKSUM_NAME/OBD_CKSUM_ALL in obd.h when adding a new
1238 * algorithm and also the OBD_FL_CKSUM* flags.
1241 OBD_CKSUM_CRC32 = 0x00000001,
1242 OBD_CKSUM_ADLER = 0x00000002,
1243 OBD_CKSUM_CRC32C= 0x00000004,
1247 * OST requests: OBDO & OBD request records
1252 OST_REPLY = 0, /* reply ? */
1268 OST_QUOTACHECK = 18,
1270 OST_QUOTA_ADJUST_QUNIT = 20,
1273 #define OST_FIRST_OPC OST_REPLY
1276 OBD_FL_INLINEDATA = 0x00000001,
1277 OBD_FL_OBDMDEXISTS = 0x00000002,
1278 OBD_FL_DELORPHAN = 0x00000004, /* if set in o_flags delete orphans */
1279 OBD_FL_NORPC = 0x00000008, /* set in o_flags do in OSC not OST */
1280 OBD_FL_IDONLY = 0x00000010, /* set in o_flags only adjust obj id*/
1281 OBD_FL_RECREATE_OBJS= 0x00000020, /* recreate missing obj */
1282 OBD_FL_DEBUG_CHECK = 0x00000040, /* echo client/server debug check */
1283 OBD_FL_NO_USRQUOTA = 0x00000100, /* the object's owner is over quota */
1284 OBD_FL_NO_GRPQUOTA = 0x00000200, /* the object's group is over quota */
1285 OBD_FL_CREATE_CROW = 0x00000400, /* object should be create on write */
1286 OBD_FL_SRVLOCK = 0x00000800, /* delegate DLM locking to server */
1287 OBD_FL_CKSUM_CRC32 = 0x00001000, /* CRC32 checksum type */
1288 OBD_FL_CKSUM_ADLER = 0x00002000, /* ADLER checksum type */
1289 OBD_FL_CKSUM_CRC32C = 0x00004000, /* CRC32C checksum type */
1290 OBD_FL_CKSUM_RSVD2 = 0x00008000, /* for future cksum types */
1291 OBD_FL_CKSUM_RSVD3 = 0x00010000, /* for future cksum types */
1292 OBD_FL_SHRINK_GRANT = 0x00020000, /* object shrink the grant */
1293 OBD_FL_MMAP = 0x00040000, /* object is mmapped on the client.
1294 * XXX: obsoleted - reserved for old
1295 * clients prior than 2.2 */
1296 OBD_FL_RECOV_RESEND = 0x00080000, /* recoverable resent */
1297 OBD_FL_NOSPC_BLK = 0x00100000, /* no more block space on OST */
1299 /* Note that while these checksum values are currently separate bits,
1300 * in 2.x we can actually allow all values from 1-31 if we wanted. */
1301 OBD_FL_CKSUM_ALL = OBD_FL_CKSUM_CRC32 | OBD_FL_CKSUM_ADLER |
1302 OBD_FL_CKSUM_CRC32C,
1304 /* mask for local-only flag, which won't be sent over network */
1305 OBD_FL_LOCAL_MASK = 0xF0000000,
1308 #define LOV_MAGIC_V1 0x0BD10BD0
1309 #define LOV_MAGIC LOV_MAGIC_V1
1310 #define LOV_MAGIC_JOIN_V1 0x0BD20BD0
1311 #define LOV_MAGIC_V3 0x0BD30BD0
1313 #define LOV_PATTERN_RAID0 0x001 /* stripes are used round-robin */
1314 #define LOV_PATTERN_RAID1 0x002 /* stripes are mirrors of each other */
1315 #define LOV_PATTERN_FIRST 0x100 /* first stripe is not in round-robin */
1316 #define LOV_PATTERN_CMOBD 0x200
1318 #define lov_ost_data lov_ost_data_v1
1319 struct lov_ost_data_v1 { /* per-stripe data structure (little-endian)*/
1320 __u64 l_object_id; /* OST object ID */
1321 __u64 l_object_seq; /* OST object seq number */
1322 __u32 l_ost_gen; /* generation of this l_ost_idx */
1323 __u32 l_ost_idx; /* OST index in LOV (lov_tgt_desc->tgts) */
1326 #define lov_mds_md lov_mds_md_v1
1327 struct lov_mds_md_v1 { /* LOV EA mds/wire data (little-endian) */
1328 __u32 lmm_magic; /* magic number = LOV_MAGIC_V1 */
1329 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1330 __u64 lmm_object_id; /* LOV object ID */
1331 __u64 lmm_object_seq; /* LOV object seq number */
1332 __u32 lmm_stripe_size; /* size of stripe in bytes */
1333 /* lmm_stripe_count used to be __u32 */
1334 __u16 lmm_stripe_count; /* num stripes in use for this object */
1335 __u16 lmm_layout_gen; /* layout generation number */
1336 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1339 /* extern void lustre_swab_lov_mds_md(struct lov_mds_md *llm); */
1341 #define MAX_MD_SIZE (sizeof(struct lov_mds_md) + 4 * sizeof(struct lov_ost_data))
1342 #define MIN_MD_SIZE (sizeof(struct lov_mds_md) + 1 * sizeof(struct lov_ost_data))
1344 #define XATTR_NAME_ACL_ACCESS "system.posix_acl_access"
1345 #define XATTR_NAME_ACL_DEFAULT "system.posix_acl_default"
1346 #define XATTR_USER_PREFIX "user."
1347 #define XATTR_TRUSTED_PREFIX "trusted."
1348 #define XATTR_SECURITY_PREFIX "security."
1349 #define XATTR_LUSTRE_PREFIX "lustre."
1351 #define XATTR_NAME_LOV "trusted.lov"
1352 #define XATTR_NAME_LMA "trusted.lma"
1353 #define XATTR_NAME_LMV "trusted.lmv"
1354 #define XATTR_NAME_LINK "trusted.link"
1355 #define XATTR_NAME_VERSION "trusted.version"
1358 struct lov_mds_md_v3 { /* LOV EA mds/wire data (little-endian) */
1359 __u32 lmm_magic; /* magic number = LOV_MAGIC_V3 */
1360 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1361 __u64 lmm_object_id; /* LOV object ID */
1362 __u64 lmm_object_seq; /* LOV object seq number */
1363 __u32 lmm_stripe_size; /* size of stripe in bytes */
1364 /* lmm_stripe_count used to be __u32 */
1365 __u16 lmm_stripe_count; /* num stripes in use for this object */
1366 __u16 lmm_layout_gen; /* layout generation number */
1367 char lmm_pool_name[LOV_MAXPOOLNAME]; /* must be 32bit aligned */
1368 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1372 #define OBD_MD_FLID (0x00000001ULL) /* object ID */
1373 #define OBD_MD_FLATIME (0x00000002ULL) /* access time */
1374 #define OBD_MD_FLMTIME (0x00000004ULL) /* data modification time */
1375 #define OBD_MD_FLCTIME (0x00000008ULL) /* change time */
1376 #define OBD_MD_FLSIZE (0x00000010ULL) /* size */
1377 #define OBD_MD_FLBLOCKS (0x00000020ULL) /* allocated blocks count */
1378 #define OBD_MD_FLBLKSZ (0x00000040ULL) /* block size */
1379 #define OBD_MD_FLMODE (0x00000080ULL) /* access bits (mode & ~S_IFMT) */
1380 #define OBD_MD_FLTYPE (0x00000100ULL) /* object type (mode & S_IFMT) */
1381 #define OBD_MD_FLUID (0x00000200ULL) /* user ID */
1382 #define OBD_MD_FLGID (0x00000400ULL) /* group ID */
1383 #define OBD_MD_FLFLAGS (0x00000800ULL) /* flags word */
1384 #define OBD_MD_FLNLINK (0x00002000ULL) /* link count */
1385 #define OBD_MD_FLGENER (0x00004000ULL) /* generation number */
1386 /*#define OBD_MD_FLINLINE (0x00008000ULL) inline data. used until 1.6.5 */
1387 #define OBD_MD_FLRDEV (0x00010000ULL) /* device number */
1388 #define OBD_MD_FLEASIZE (0x00020000ULL) /* extended attribute data */
1389 #define OBD_MD_LINKNAME (0x00040000ULL) /* symbolic link target */
1390 #define OBD_MD_FLHANDLE (0x00080000ULL) /* file/lock handle */
1391 #define OBD_MD_FLCKSUM (0x00100000ULL) /* bulk data checksum */
1392 #define OBD_MD_FLQOS (0x00200000ULL) /* quality of service stats */
1393 /*#define OBD_MD_FLOSCOPQ (0x00400000ULL) osc opaque data, never used */
1394 #define OBD_MD_FLCOOKIE (0x00800000ULL) /* log cancellation cookie */
1395 #define OBD_MD_FLGROUP (0x01000000ULL) /* group */
1396 #define OBD_MD_FLFID (0x02000000ULL) /* ->ost write inline fid */
1397 #define OBD_MD_FLEPOCH (0x04000000ULL) /* ->ost write with ioepoch */
1398 /* ->mds if epoch opens or closes */
1399 #define OBD_MD_FLGRANT (0x08000000ULL) /* ost preallocation space grant */
1400 #define OBD_MD_FLDIREA (0x10000000ULL) /* dir's extended attribute data */
1401 #define OBD_MD_FLUSRQUOTA (0x20000000ULL) /* over quota flags sent from ost */
1402 #define OBD_MD_FLGRPQUOTA (0x40000000ULL) /* over quota flags sent from ost */
1403 #define OBD_MD_FLMODEASIZE (0x80000000ULL) /* EA size will be changed */
1405 #define OBD_MD_MDS (0x0000000100000000ULL) /* where an inode lives on */
1406 #define OBD_MD_REINT (0x0000000200000000ULL) /* reintegrate oa */
1407 #define OBD_MD_MEA (0x0000000400000000ULL) /* CMD split EA */
1408 #define OBD_MD_MDTIDX (0x0000000800000000ULL) /* Get MDT index */
1410 #define OBD_MD_FLXATTR (0x0000001000000000ULL) /* xattr */
1411 #define OBD_MD_FLXATTRLS (0x0000002000000000ULL) /* xattr list */
1412 #define OBD_MD_FLXATTRRM (0x0000004000000000ULL) /* xattr remove */
1413 #define OBD_MD_FLACL (0x0000008000000000ULL) /* ACL */
1414 #define OBD_MD_FLRMTPERM (0x0000010000000000ULL) /* remote permission */
1415 #define OBD_MD_FLMDSCAPA (0x0000020000000000ULL) /* MDS capability */
1416 #define OBD_MD_FLOSSCAPA (0x0000040000000000ULL) /* OSS capability */
1417 #define OBD_MD_FLCKSPLIT (0x0000080000000000ULL) /* Check split on server */
1418 #define OBD_MD_FLCROSSREF (0x0000100000000000ULL) /* Cross-ref case */
1419 #define OBD_MD_FLGETATTRLOCK (0x0000200000000000ULL) /* Get IOEpoch attributes
1422 #define OBD_MD_FLRMTLSETFACL (0x0001000000000000ULL) /* lfs lsetfacl case */
1423 #define OBD_MD_FLRMTLGETFACL (0x0002000000000000ULL) /* lfs lgetfacl case */
1424 #define OBD_MD_FLRMTRSETFACL (0x0004000000000000ULL) /* lfs rsetfacl case */
1425 #define OBD_MD_FLRMTRGETFACL (0x0008000000000000ULL) /* lfs rgetfacl case */
1427 #define OBD_MD_FLDATAVERSION (0x0010000000000000ULL) /* iversion sum */
1429 #define OBD_MD_FLGETATTR (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLMTIME | \
1430 OBD_MD_FLCTIME | OBD_MD_FLSIZE | OBD_MD_FLBLKSZ | \
1431 OBD_MD_FLMODE | OBD_MD_FLTYPE | OBD_MD_FLUID | \
1432 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLNLINK | \
1433 OBD_MD_FLGENER | OBD_MD_FLRDEV | OBD_MD_FLGROUP)
1435 /* don't forget obdo_fid which is way down at the bottom so it can
1436 * come after the definition of llog_cookie */
1439 extern void lustre_swab_obd_statfs (struct obd_statfs *os);
1441 /* ost_body.data values for OST_BRW */
1443 #define OBD_BRW_READ 0x01
1444 #define OBD_BRW_WRITE 0x02
1445 #define OBD_BRW_RWMASK (OBD_BRW_READ | OBD_BRW_WRITE)
1446 #define OBD_BRW_SYNC 0x08 /* this page is a part of synchronous
1447 * transfer and is not accounted in
1449 #define OBD_BRW_CHECK 0x10
1450 #define OBD_BRW_FROM_GRANT 0x20 /* the osc manages this under llite */
1451 #define OBD_BRW_GRANTED 0x40 /* the ost manages this */
1452 #define OBD_BRW_NOCACHE 0x80 /* this page is a part of non-cached IO */
1453 #define OBD_BRW_NOQUOTA 0x100
1454 #define OBD_BRW_SRVLOCK 0x200 /* Client holds no lock over this page */
1455 #define OBD_BRW_ASYNC 0x400 /* Server may delay commit to disk */
1456 #define OBD_BRW_MEMALLOC 0x800 /* Client runs in the "kswapd" context */
1458 #define OBD_OBJECT_EOF 0xffffffffffffffffULL
1460 #define OST_MIN_PRECREATE 32
1461 #define OST_MAX_PRECREATE 20000
1470 extern void lustre_swab_obd_ioobj (struct obd_ioobj *ioo);
1472 /* multiple of 8 bytes => can array */
1473 struct niobuf_remote {
1479 extern void lustre_swab_niobuf_remote (struct niobuf_remote *nbr);
1481 /* lock value block communicated between the filter and llite */
1483 /* OST_LVB_ERR_INIT is needed because the return code in rc is
1484 * negative, i.e. because ((MASK + rc) & MASK) != MASK. */
1485 #define OST_LVB_ERR_INIT 0xffbadbad80000000ULL
1486 #define OST_LVB_ERR_MASK 0xffbadbad00000000ULL
1487 #define OST_LVB_IS_ERR(blocks) \
1488 ((blocks & OST_LVB_ERR_MASK) == OST_LVB_ERR_MASK)
1489 #define OST_LVB_SET_ERR(blocks, rc) \
1490 do { blocks = OST_LVB_ERR_INIT + rc; } while (0)
1491 #define OST_LVB_GET_ERR(blocks) (int)(blocks - OST_LVB_ERR_INIT)
1501 extern void lustre_swab_ost_lvb(struct ost_lvb *);
1510 MDS_GETATTR_NAME = 34,
1515 MDS_DISCONNECT = 39,
1521 MDS_DONE_WRITING = 45,
1523 MDS_QUOTACHECK = 47,
1526 MDS_SETXATTR = 50, /* obsolete, now it's MDS_REINT op */
1533 #define MDS_FIRST_OPC MDS_GETATTR
1550 } mds_reint_t, mdt_reint_t;
1552 extern void lustre_swab_generic_32s (__u32 *val);
1554 /* the disposition of the intent outlines what was executed */
1555 #define DISP_IT_EXECD 0x00000001
1556 #define DISP_LOOKUP_EXECD 0x00000002
1557 #define DISP_LOOKUP_NEG 0x00000004
1558 #define DISP_LOOKUP_POS 0x00000008
1559 #define DISP_OPEN_CREATE 0x00000010
1560 #define DISP_OPEN_OPEN 0x00000020
1561 #define DISP_ENQ_COMPLETE 0x00400000
1562 #define DISP_ENQ_OPEN_REF 0x00800000
1563 #define DISP_ENQ_CREATE_REF 0x01000000
1564 #define DISP_OPEN_LOCK 0x02000000
1566 /* INODE LOCK PARTS */
1567 #define MDS_INODELOCK_LOOKUP 0x000001 /* dentry, mode, owner, group */
1568 #define MDS_INODELOCK_UPDATE 0x000002 /* size, links, timestamps */
1569 #define MDS_INODELOCK_OPEN 0x000004 /* For opened files */
1570 #define MDS_INODELOCK_LAYOUT 0x000008 /* for layout */
1572 /* Do not forget to increase MDS_INODELOCK_MAXSHIFT when adding new bits
1573 * XXX: MDS_INODELOCK_MAXSHIFT should be increased to 3 once the layout lock is
1575 #define MDS_INODELOCK_MAXSHIFT 2
1576 /* This FULL lock is useful to take on unlink sort of operations */
1577 #define MDS_INODELOCK_FULL ((1<<(MDS_INODELOCK_MAXSHIFT+1))-1)
1579 extern void lustre_swab_ll_fid (struct ll_fid *fid);
1581 /* NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
1582 * but was moved into name[1] along with the OID to avoid consuming the
1583 * name[2,3] fields that need to be used for the quota id (also a FID). */
1585 LUSTRE_RES_ID_SEQ_OFF = 0,
1586 LUSTRE_RES_ID_VER_OID_OFF = 1,
1587 LUSTRE_RES_ID_WAS_VER_OFF = 2, /* see note above */
1588 LUSTRE_RES_ID_HSH_OFF = 3
1591 #define MDS_STATUS_CONN 1
1592 #define MDS_STATUS_LOV 2
1594 /* mdt_thread_info.mti_flags. */
1596 /* The flag indicates Size-on-MDS attributes are changed. */
1597 MF_SOM_CHANGE = (1 << 0),
1598 /* Flags indicates an epoch opens or closes. */
1599 MF_EPOCH_OPEN = (1 << 1),
1600 MF_EPOCH_CLOSE = (1 << 2),
1601 MF_MDC_CANCEL_FID1 = (1 << 3),
1602 MF_MDC_CANCEL_FID2 = (1 << 4),
1603 MF_MDC_CANCEL_FID3 = (1 << 5),
1604 MF_MDC_CANCEL_FID4 = (1 << 6),
1605 /* There is a pending attribute update. */
1606 MF_SOM_AU = (1 << 7),
1607 /* Cancel OST locks while getattr OST attributes. */
1608 MF_GETATTR_LOCK = (1 << 8),
1611 #define MF_SOM_LOCAL_FLAGS (MF_SOM_CHANGE | MF_EPOCH_OPEN | MF_EPOCH_CLOSE)
1613 #define LUSTRE_BFLAG_UNCOMMITTED_WRITES 0x1
1615 /* these should be identical to their EXT4_*_FL counterparts, they are
1616 * redefined here only to avoid dragging in fs/ext4/ext4.h */
1617 #define LUSTRE_SYNC_FL 0x00000008 /* Synchronous updates */
1618 #define LUSTRE_IMMUTABLE_FL 0x00000010 /* Immutable file */
1619 #define LUSTRE_APPEND_FL 0x00000020 /* writes to file may only append */
1620 #define LUSTRE_NOATIME_FL 0x00000080 /* do not update atime */
1621 #define LUSTRE_DIRSYNC_FL 0x00010000 /* dirsync behaviour (dir only) */
1624 /* Convert wire LUSTRE_*_FL to corresponding client local VFS S_* values
1625 * for the client inode i_flags. The LUSTRE_*_FL are the Lustre wire
1626 * protocol equivalents of LDISKFS_*_FL values stored on disk, while
1627 * the S_* flags are kernel-internal values that change between kernel
1628 * versions. These flags are set/cleared via FSFILT_IOC_{GET,SET}_FLAGS.
1629 * See b=16526 for a full history. */
1630 static inline int ll_ext_to_inode_flags(int flags)
1632 return (((flags & LUSTRE_SYNC_FL) ? S_SYNC : 0) |
1633 ((flags & LUSTRE_NOATIME_FL) ? S_NOATIME : 0) |
1634 ((flags & LUSTRE_APPEND_FL) ? S_APPEND : 0) |
1635 #if defined(S_DIRSYNC)
1636 ((flags & LUSTRE_DIRSYNC_FL) ? S_DIRSYNC : 0) |
1638 ((flags & LUSTRE_IMMUTABLE_FL) ? S_IMMUTABLE : 0));
1641 static inline int ll_inode_to_ext_flags(int iflags)
1643 return (((iflags & S_SYNC) ? LUSTRE_SYNC_FL : 0) |
1644 ((iflags & S_NOATIME) ? LUSTRE_NOATIME_FL : 0) |
1645 ((iflags & S_APPEND) ? LUSTRE_APPEND_FL : 0) |
1646 #if defined(S_DIRSYNC)
1647 ((iflags & S_DIRSYNC) ? LUSTRE_DIRSYNC_FL : 0) |
1649 ((iflags & S_IMMUTABLE) ? LUSTRE_IMMUTABLE_FL : 0));
1656 struct lustre_handle handle;
1658 __u64 size; /* Offset, in the case of MDS_READPAGE */
1662 __u64 blocks; /* XID, in the case of MDS_READPAGE */
1671 __u32 flags; /* from vfs for pin/unpin, LUSTRE_BFLAG close */
1673 __u32 nlink; /* #bytes to read in the case of MDS_READPAGE */
1679 __u32 max_cookiesize;
1680 __u32 uid_h; /* high 32-bits of uid, for FUID */
1681 __u32 gid_h; /* high 32-bits of gid, for FUID */
1682 __u32 padding_5; /* also fix lustre_swab_mdt_body */
1690 extern void lustre_swab_mdt_body (struct mdt_body *b);
1692 struct mdt_ioepoch {
1693 struct lustre_handle handle;
1699 extern void lustre_swab_mdt_ioepoch (struct mdt_ioepoch *b);
1701 #define Q_QUOTACHECK 0x800100
1702 #define Q_INITQUOTA 0x800101 /* init slave limits */
1703 #define Q_GETOINFO 0x800102 /* get obd quota info */
1704 #define Q_GETOQUOTA 0x800103 /* get obd quotas */
1705 #define Q_FINVALIDATE 0x800104 /* invalidate operational quotas */
1707 #define Q_TYPEMATCH(id, type) \
1708 ((id) == (type) || (id) == UGQUOTA)
1710 #define Q_TYPESET(oqc, type) Q_TYPEMATCH((oqc)->qc_type, type)
1712 #define Q_GETOCMD(oqc) \
1713 ((oqc)->qc_cmd == Q_GETOINFO || (oqc)->qc_cmd == Q_GETOQUOTA)
1715 #define QCTL_COPY(out, in) \
1717 Q_COPY(out, in, qc_cmd); \
1718 Q_COPY(out, in, qc_type); \
1719 Q_COPY(out, in, qc_id); \
1720 Q_COPY(out, in, qc_stat); \
1721 Q_COPY(out, in, qc_dqinfo); \
1722 Q_COPY(out, in, qc_dqblk); \
1725 struct obd_quotactl {
1730 struct obd_dqinfo qc_dqinfo;
1731 struct obd_dqblk qc_dqblk;
1734 extern void lustre_swab_obd_quotactl(struct obd_quotactl *q);
1736 struct quota_adjust_qunit {
1743 extern void lustre_swab_quota_adjust_qunit(struct quota_adjust_qunit *q);
1745 /* flags is shared among quota structures */
1746 #define LQUOTA_FLAGS_GRP 1UL /* 0 is user, 1 is group */
1747 #define LQUOTA_FLAGS_BLK 2UL /* 0 is inode, 1 is block */
1748 #define LQUOTA_FLAGS_ADJBLK 4UL /* adjust the block qunit size */
1749 #define LQUOTA_FLAGS_ADJINO 8UL /* adjust the inode qunit size */
1750 #define LQUOTA_FLAGS_CHG_QS 16UL /* indicate whether it has capability of
1751 * OBD_CONNECT_CHANGE_QS */
1752 #define LQUOTA_FLAGS_RECOVERY 32UL /* recovery is going on a uid/gid */
1753 #define LQUOTA_FLAGS_SETQUOTA 64UL /* being setquota on a uid/gid */
1755 /* flags is specific for quota_adjust_qunit */
1756 #define LQUOTA_QAQ_CREATE_LQS (1UL << 31) /* when it is set, need create lqs */
1758 /* the status of lqs_flags in struct lustre_qunit_size */
1759 #define LQUOTA_QUNIT_FLAGS (LQUOTA_FLAGS_GRP | LQUOTA_FLAGS_BLK)
1761 #define QAQ_IS_GRP(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_GRP)
1762 #define QAQ_IS_ADJBLK(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_ADJBLK)
1763 #define QAQ_IS_ADJINO(qaq) ((qaq)->qaq_flags & LQUOTA_FLAGS_ADJINO)
1764 #define QAQ_IS_CREATE_LQS(qaq) ((qaq)->qaq_flags & LQUOTA_QAQ_CREATE_LQS)
1766 #define QAQ_SET_GRP(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_GRP)
1767 #define QAQ_SET_ADJBLK(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_ADJBLK)
1768 #define QAQ_SET_ADJINO(qaq) ((qaq)->qaq_flags |= LQUOTA_FLAGS_ADJINO)
1769 #define QAQ_SET_CREATE_LQS(qaq) ((qaq)->qaq_flags |= LQUOTA_QAQ_CREATE_LQS)
1771 /* permissions for md_perm.mp_perm */
1773 CFS_SETUID_PERM = 0x01,
1774 CFS_SETGID_PERM = 0x02,
1775 CFS_SETGRP_PERM = 0x04,
1776 CFS_RMTACL_PERM = 0x08,
1777 CFS_RMTOWN_PERM = 0x10
1780 /* inode access permission for remote user, the inode info are omitted,
1781 * for client knows them. */
1782 struct mdt_remote_perm {
1789 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
1793 extern void lustre_swab_mdt_remote_perm(struct mdt_remote_perm *p);
1795 struct mdt_rec_setattr {
1805 __u32 sa_padding_1_h;
1806 struct lu_fid sa_fid;
1815 __u32 sa_attr_flags;
1823 extern void lustre_swab_mdt_rec_setattr (struct mdt_rec_setattr *sa);
1826 * Attribute flags used in mdt_rec_setattr::sa_valid.
1827 * The kernel's #defines for ATTR_* should not be used over the network
1828 * since the client and MDS may run different kernels (see bug 13828)
1829 * Therefore, we should only use MDS_ATTR_* attributes for sa_valid.
1831 #define MDS_ATTR_MODE 0x1ULL /* = 1 */
1832 #define MDS_ATTR_UID 0x2ULL /* = 2 */
1833 #define MDS_ATTR_GID 0x4ULL /* = 4 */
1834 #define MDS_ATTR_SIZE 0x8ULL /* = 8 */
1835 #define MDS_ATTR_ATIME 0x10ULL /* = 16 */
1836 #define MDS_ATTR_MTIME 0x20ULL /* = 32 */
1837 #define MDS_ATTR_CTIME 0x40ULL /* = 64 */
1838 #define MDS_ATTR_ATIME_SET 0x80ULL /* = 128 */
1839 #define MDS_ATTR_MTIME_SET 0x100ULL /* = 256 */
1840 #define MDS_ATTR_FORCE 0x200ULL /* = 512, Not a change, but a change it */
1841 #define MDS_ATTR_ATTR_FLAG 0x400ULL /* = 1024 */
1842 #define MDS_ATTR_KILL_SUID 0x800ULL /* = 2048 */
1843 #define MDS_ATTR_KILL_SGID 0x1000ULL /* = 4096 */
1844 #define MDS_ATTR_CTIME_SET 0x2000ULL /* = 8192 */
1845 #define MDS_ATTR_FROM_OPEN 0x4000ULL /* = 16384, called from open path, ie O_TRUNC */
1846 #define MDS_ATTR_BLOCKS 0x8000ULL /* = 32768 */
1849 #define FMODE_READ 00000001
1850 #define FMODE_WRITE 00000002
1853 #define MDS_FMODE_CLOSED 00000000
1854 #define MDS_FMODE_EXEC 00000004
1855 /* IO Epoch is opened on a closed file. */
1856 #define MDS_FMODE_EPOCH 01000000
1857 /* IO Epoch is opened on a file truncate. */
1858 #define MDS_FMODE_TRUNC 02000000
1859 /* Size-on-MDS Attribute Update is pending. */
1860 #define MDS_FMODE_SOM 04000000
1862 #define MDS_OPEN_CREATED 00000010
1863 #define MDS_OPEN_CROSS 00000020
1865 #define MDS_OPEN_CREAT 00000100
1866 #define MDS_OPEN_EXCL 00000200
1867 #define MDS_OPEN_TRUNC 00001000
1868 #define MDS_OPEN_APPEND 00002000
1869 #define MDS_OPEN_SYNC 00010000
1870 #define MDS_OPEN_DIRECTORY 00200000
1872 #define MDS_OPEN_DELAY_CREATE 0100000000 /* delay initial object create */
1873 #define MDS_OPEN_OWNEROVERRIDE 0200000000 /* NFSD rw-reopen ro file for owner */
1874 #define MDS_OPEN_JOIN_FILE 0400000000 /* open for join file.
1875 * We do not support JOIN FILE
1876 * anymore, reserve this flags
1877 * just for preventing such bit
1879 #define MDS_CREATE_RMT_ACL 01000000000 /* indicate create on remote server
1880 * with default ACL */
1881 #define MDS_CREATE_SLAVE_OBJ 02000000000 /* indicate create slave object
1882 * actually, this is for create, not
1883 * conflict with other open flags */
1884 #define MDS_OPEN_LOCK 04000000000 /* This open requires open lock */
1885 #define MDS_OPEN_HAS_EA 010000000000 /* specify object create pattern */
1886 #define MDS_OPEN_HAS_OBJS 020000000000 /* Just set the EA the obj exist */
1887 #define MDS_OPEN_NORESTORE 0100000000000ULL /* Do not restore file at open */
1888 #define MDS_OPEN_NEWSTRIPE 0200000000000ULL /* New stripe needed (restripe or
1891 /* permission for create non-directory file */
1892 #define MAY_CREATE (1 << 7)
1893 /* permission for create directory file */
1894 #define MAY_LINK (1 << 8)
1895 /* permission for delete from the directory */
1896 #define MAY_UNLINK (1 << 9)
1897 /* source's permission for rename */
1898 #define MAY_RENAME_SRC (1 << 10)
1899 /* target's permission for rename */
1900 #define MAY_RENAME_TAR (1 << 11)
1901 /* part (parent's) VTX permission check */
1902 #define MAY_VTX_PART (1 << 12)
1903 /* full VTX permission check */
1904 #define MAY_VTX_FULL (1 << 13)
1905 /* lfs rgetfacl permission check */
1906 #define MAY_RGETFACL (1 << 14)
1909 MDS_CHECK_SPLIT = 1 << 0,
1910 MDS_CROSS_REF = 1 << 1,
1911 MDS_VTX_BYPASS = 1 << 2,
1912 MDS_PERM_BYPASS = 1 << 3,
1914 MDS_QUOTA_IGNORE = 1 << 5,
1915 MDS_CLOSE_CLEANUP = 1 << 6,
1916 MDS_KEEP_ORPHAN = 1 << 7,
1917 MDS_RECOV_OPEN = 1 << 8,
1920 /* instance of mdt_reint_rec */
1921 struct mdt_rec_create {
1929 __u32 cr_suppgid1_h;
1931 __u32 cr_suppgid2_h;
1932 struct lu_fid cr_fid1;
1933 struct lu_fid cr_fid2;
1934 struct lustre_handle cr_old_handle; /* handle in case of open replay */
1938 __u64 cr_padding_1; /* rr_blocks */
1941 /* use of helpers set/get_mrc_cr_flags() is needed to access
1942 * 64 bits cr_flags [cr_flags_l, cr_flags_h], this is done to
1943 * extend cr_flags size without breaking 1.8 compat */
1944 __u32 cr_flags_l; /* for use with open, low 32 bits */
1945 __u32 cr_flags_h; /* for use with open, high 32 bits */
1946 __u32 cr_padding_3; /* rr_padding_3 */
1947 __u32 cr_padding_4; /* rr_padding_4 */
1950 static inline void set_mrc_cr_flags(struct mdt_rec_create *mrc, __u64 flags)
1952 mrc->cr_flags_l = (__u32)(flags & 0xFFFFFFFFUll);
1953 mrc->cr_flags_h = (__u32)(flags >> 32);
1956 static inline __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
1958 return ((__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32));
1961 /* instance of mdt_reint_rec */
1962 struct mdt_rec_link {
1970 __u32 lk_suppgid1_h;
1972 __u32 lk_suppgid2_h;
1973 struct lu_fid lk_fid1;
1974 struct lu_fid lk_fid2;
1976 __u64 lk_padding_1; /* rr_atime */
1977 __u64 lk_padding_2; /* rr_ctime */
1978 __u64 lk_padding_3; /* rr_size */
1979 __u64 lk_padding_4; /* rr_blocks */
1981 __u32 lk_padding_5; /* rr_mode */
1982 __u32 lk_padding_6; /* rr_flags */
1983 __u32 lk_padding_7; /* rr_padding_2 */
1984 __u32 lk_padding_8; /* rr_padding_3 */
1985 __u32 lk_padding_9; /* rr_padding_4 */
1988 /* instance of mdt_reint_rec */
1989 struct mdt_rec_unlink {
1997 __u32 ul_suppgid1_h;
1999 __u32 ul_suppgid2_h;
2000 struct lu_fid ul_fid1;
2001 struct lu_fid ul_fid2;
2003 __u64 ul_padding_2; /* rr_atime */
2004 __u64 ul_padding_3; /* rr_ctime */
2005 __u64 ul_padding_4; /* rr_size */
2006 __u64 ul_padding_5; /* rr_blocks */
2009 __u32 ul_padding_6; /* rr_flags */
2010 __u32 ul_padding_7; /* rr_padding_2 */
2011 __u32 ul_padding_8; /* rr_padding_3 */
2012 __u32 ul_padding_9; /* rr_padding_4 */
2015 /* instance of mdt_reint_rec */
2016 struct mdt_rec_rename {
2024 __u32 rn_suppgid1_h;
2026 __u32 rn_suppgid2_h;
2027 struct lu_fid rn_fid1;
2028 struct lu_fid rn_fid2;
2030 __u64 rn_padding_1; /* rr_atime */
2031 __u64 rn_padding_2; /* rr_ctime */
2032 __u64 rn_padding_3; /* rr_size */
2033 __u64 rn_padding_4; /* rr_blocks */
2034 __u32 rn_bias; /* some operation flags */
2035 __u32 rn_mode; /* cross-ref rename has mode */
2036 __u32 rn_padding_5; /* rr_flags */
2037 __u32 rn_padding_6; /* rr_padding_2 */
2038 __u32 rn_padding_7; /* rr_padding_3 */
2039 __u32 rn_padding_8; /* rr_padding_4 */
2042 /* instance of mdt_reint_rec */
2043 struct mdt_rec_setxattr {
2051 __u32 sx_suppgid1_h;
2053 __u32 sx_suppgid2_h;
2054 struct lu_fid sx_fid;
2055 __u64 sx_padding_1; /* These three are rr_fid2 */
2060 __u64 sx_padding_5; /* rr_ctime */
2061 __u64 sx_padding_6; /* rr_size */
2062 __u64 sx_padding_7; /* rr_blocks */
2065 __u32 sx_padding_8; /* rr_flags */
2066 __u32 sx_padding_9; /* rr_padding_2 */
2067 __u32 sx_padding_10; /* rr_padding_3 */
2068 __u32 sx_padding_11; /* rr_padding_4 */
2072 * mdt_rec_reint is the template for all mdt_reint_xxx structures.
2073 * Do NOT change the size of various members, otherwise the value
2074 * will be broken in lustre_swab_mdt_rec_reint().
2076 * If you add new members in other mdt_reint_xxx structres and need to use the
2077 * rr_padding_x fields, then update lustre_swab_mdt_rec_reint() also.
2079 struct mdt_rec_reint {
2087 __u32 rr_suppgid1_h;
2089 __u32 rr_suppgid2_h;
2090 struct lu_fid rr_fid1;
2091 struct lu_fid rr_fid2;
2100 __u32 rr_padding_2; /* also fix lustre_swab_mdt_rec_reint */
2101 __u32 rr_padding_3; /* also fix lustre_swab_mdt_rec_reint */
2102 __u32 rr_padding_4; /* also fix lustre_swab_mdt_rec_reint */
2105 extern void lustre_swab_mdt_rec_reint(struct mdt_rec_reint *rr);
2108 __u32 ld_tgt_count; /* how many MDS's */
2109 __u32 ld_active_tgt_count; /* how many active */
2110 __u32 ld_default_stripe_count; /* how many objects are used */
2111 __u32 ld_pattern; /* default MEA_MAGIC_* */
2112 __u64 ld_default_hash_size;
2113 __u64 ld_padding_1; /* also fix lustre_swab_lmv_desc */
2114 __u32 ld_padding_2; /* also fix lustre_swab_lmv_desc */
2115 __u32 ld_qos_maxage; /* in second */
2116 __u32 ld_padding_3; /* also fix lustre_swab_lmv_desc */
2117 __u32 ld_padding_4; /* also fix lustre_swab_lmv_desc */
2118 struct obd_uuid ld_uuid;
2121 extern void lustre_swab_lmv_desc (struct lmv_desc *ld);
2123 /* TODO: lmv_stripe_md should contain mds capabilities for all slave fids */
2124 struct lmv_stripe_md {
2129 char mea_pool_name[LOV_MAXPOOLNAME];
2130 struct lu_fid mea_ids[0];
2133 extern void lustre_swab_lmv_stripe_md(struct lmv_stripe_md *mea);
2135 /* lmv structures */
2136 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
2137 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
2138 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
2140 #define MAX_HASH_SIZE_32 0x7fffffffUL
2141 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
2142 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
2147 FLD_FIRST_OPC = FLD_QUERY
2153 SEQ_FIRST_OPC = SEQ_QUERY
2157 SEQ_ALLOC_SUPER = 0,
2162 * LOV data structures
2165 #define LOV_MIN_STRIPE_BITS 16 /* maximum PAGE_SIZE (ia64), power of 2 */
2166 #define LOV_MIN_STRIPE_SIZE (1<<LOV_MIN_STRIPE_BITS)
2167 #define LOV_MAX_STRIPE_COUNT_OLD 160
2168 /* This calculation is crafted so that input of 4096 will result in 160
2169 * which in turn is equal to old maximal stripe count.
2170 * XXX: In fact this is too simpified for now, what it also need is to get
2171 * ea_type argument to clearly know how much space each stripe consumes.
2173 * The limit of 12 pages is somewhat arbitrary, but is a reasonably large
2174 * allocation that is sufficient for the current generation of systems.
2176 * (max buffer size - lov+rpc header) / sizeof(struct lov_ost_data_v1) */
2177 #define LOV_MAX_STRIPE_COUNT 2000 /* ((12 * 4096 - 256) / 24) */
2178 #define LOV_ALL_STRIPES 0xffff /* only valid for directories */
2179 #define LOV_V1_INSANE_STRIPE_COUNT 65532 /* maximum stripe count bz13933 */
2181 #define LOV_MAX_UUID_BUFFER_SIZE 8192
2182 /* The size of the buffer the lov/mdc reserves for the
2183 * array of UUIDs returned by the MDS. With the current
2184 * protocol, this will limit the max number of OSTs per LOV */
2186 #define LOV_DESC_MAGIC 0xB0CCDE5C
2188 /* LOV settings descriptor (should only contain static info) */
2190 __u32 ld_tgt_count; /* how many OBD's */
2191 __u32 ld_active_tgt_count; /* how many active */
2192 __u32 ld_default_stripe_count; /* how many objects are used */
2193 __u32 ld_pattern; /* default PATTERN_RAID0 */
2194 __u64 ld_default_stripe_size; /* in bytes */
2195 __u64 ld_default_stripe_offset; /* in bytes */
2196 __u32 ld_padding_0; /* unused */
2197 __u32 ld_qos_maxage; /* in second */
2198 __u32 ld_padding_1; /* also fix lustre_swab_lov_desc */
2199 __u32 ld_padding_2; /* also fix lustre_swab_lov_desc */
2200 struct obd_uuid ld_uuid;
2203 #define ld_magic ld_active_tgt_count /* for swabbing from llogs */
2205 extern void lustre_swab_lov_desc (struct lov_desc *ld);
2210 /* opcodes -- MUST be distinct from OST/MDS opcodes */
2215 LDLM_BL_CALLBACK = 104,
2216 LDLM_CP_CALLBACK = 105,
2217 LDLM_GL_CALLBACK = 106,
2218 LDLM_SET_INFO = 107,
2221 #define LDLM_FIRST_OPC LDLM_ENQUEUE
2223 #define RES_NAME_SIZE 4
2224 struct ldlm_res_id {
2225 __u64 name[RES_NAME_SIZE];
2228 extern void lustre_swab_ldlm_res_id (struct ldlm_res_id *id);
2230 static inline int ldlm_res_eq(const struct ldlm_res_id *res0,
2231 const struct ldlm_res_id *res1)
2233 return !memcmp(res0, res1, sizeof(*res0));
2250 #define LCK_MODE_NUM 8
2260 #define LDLM_MIN_TYPE LDLM_PLAIN
2262 struct ldlm_extent {
2268 static inline int ldlm_extent_overlap(struct ldlm_extent *ex1,
2269 struct ldlm_extent *ex2)
2271 return (ex1->start <= ex2->end) && (ex2->start <= ex1->end);
2274 /* check if @ex1 contains @ex2 */
2275 static inline int ldlm_extent_contain(struct ldlm_extent *ex1,
2276 struct ldlm_extent *ex2)
2278 return (ex1->start <= ex2->start) && (ex1->end >= ex2->end);
2281 struct ldlm_inodebits {
2285 struct ldlm_flock_wire {
2293 /* it's important that the fields of the ldlm_extent structure match
2294 * the first fields of the ldlm_flock structure because there is only
2295 * one ldlm_swab routine to process the ldlm_policy_data_t union. if
2296 * this ever changes we will need to swab the union differently based
2297 * on the resource type. */
2300 struct ldlm_extent l_extent;
2301 struct ldlm_flock_wire l_flock;
2302 struct ldlm_inodebits l_inodebits;
2303 } ldlm_wire_policy_data_t;
2305 extern void lustre_swab_ldlm_policy_data (ldlm_wire_policy_data_t *d);
2307 struct ldlm_intent {
2311 extern void lustre_swab_ldlm_intent (struct ldlm_intent *i);
2313 struct ldlm_resource_desc {
2314 ldlm_type_t lr_type;
2315 __u32 lr_padding; /* also fix lustre_swab_ldlm_resource_desc */
2316 struct ldlm_res_id lr_name;
2319 extern void lustre_swab_ldlm_resource_desc (struct ldlm_resource_desc *r);
2321 struct ldlm_lock_desc {
2322 struct ldlm_resource_desc l_resource;
2323 ldlm_mode_t l_req_mode;
2324 ldlm_mode_t l_granted_mode;
2325 ldlm_wire_policy_data_t l_policy_data;
2328 extern void lustre_swab_ldlm_lock_desc (struct ldlm_lock_desc *l);
2330 #define LDLM_LOCKREQ_HANDLES 2
2331 #define LDLM_ENQUEUE_CANCEL_OFF 1
2333 struct ldlm_request {
2336 struct ldlm_lock_desc lock_desc;
2337 struct lustre_handle lock_handle[LDLM_LOCKREQ_HANDLES];
2340 extern void lustre_swab_ldlm_request (struct ldlm_request *rq);
2342 /* If LDLM_ENQUEUE, 1 slot is already occupied, 1 is available.
2343 * Otherwise, 2 are available. */
2344 #define ldlm_request_bufsize(count,type) \
2346 int _avail = LDLM_LOCKREQ_HANDLES; \
2347 _avail -= (type == LDLM_ENQUEUE ? LDLM_ENQUEUE_CANCEL_OFF : 0); \
2348 sizeof(struct ldlm_request) + \
2349 (count > _avail ? count - _avail : 0) * \
2350 sizeof(struct lustre_handle); \
2355 __u32 lock_padding; /* also fix lustre_swab_ldlm_reply */
2356 struct ldlm_lock_desc lock_desc;
2357 struct lustre_handle lock_handle;
2358 __u64 lock_policy_res1;
2359 __u64 lock_policy_res2;
2362 extern void lustre_swab_ldlm_reply (struct ldlm_reply *r);
2365 * Opcodes for mountconf (mgs and mgc)
2370 MGS_EXCEPTION, /* node died, etc. */
2371 MGS_TARGET_REG, /* whenever target starts up */
2377 #define MGS_FIRST_OPC MGS_CONNECT
2379 #define MGS_PARAM_MAXLEN 1024
2380 #define KEY_SET_INFO "set_info"
2382 struct mgs_send_param {
2383 char mgs_param[MGS_PARAM_MAXLEN];
2386 /* We pass this info to the MGS so it can write config logs */
2387 #define MTI_NAME_MAXLEN 64
2388 #define MTI_PARAM_MAXLEN 4096
2389 #define MTI_NIDS_MAX 32
2390 struct mgs_target_info {
2391 __u32 mti_lustre_ver;
2392 __u32 mti_stripe_index;
2393 __u32 mti_config_ver;
2395 __u32 mti_nid_count;
2396 __u32 mti_instance; /* Running instance of target */
2397 char mti_fsname[MTI_NAME_MAXLEN];
2398 char mti_svname[MTI_NAME_MAXLEN];
2399 char mti_uuid[sizeof(struct obd_uuid)];
2400 __u64 mti_nids[MTI_NIDS_MAX]; /* host nids (lnet_nid_t)*/
2401 char mti_params[MTI_PARAM_MAXLEN];
2403 extern void lustre_swab_mgs_target_info(struct mgs_target_info *oinfo);
2405 struct mgs_nidtbl_entry {
2406 __u64 mne_version; /* table version of this entry */
2407 __u32 mne_instance; /* target instance # */
2408 __u32 mne_index; /* target index */
2409 __u32 mne_length; /* length of this entry - by bytes */
2410 __u8 mne_type; /* target type LDD_F_SV_TYPE_OST/MDT */
2411 __u8 mne_nid_type; /* type of nid(mbz). for ipv6. */
2412 __u8 mne_nid_size; /* size of each NID, by bytes */
2413 __u8 mne_nid_count; /* # of NIDs in buffer */
2415 lnet_nid_t nids[0]; /* variable size buffer for NIDs. */
2418 extern void lustre_swab_mgs_nidtbl_entry(struct mgs_nidtbl_entry *oinfo);
2420 struct mgs_config_body {
2421 char mcb_name[MTI_NAME_MAXLEN]; /* logname */
2422 __u64 mcb_offset; /* next index of config log to request */
2423 __u16 mcb_type; /* type of log: CONFIG_T_[CONFIG|RECOVER] */
2425 __u8 mcb_bits; /* bits unit size of config log */
2426 __u32 mcb_units; /* # of units for bulk transfer */
2428 extern void lustre_swab_mgs_config_body(struct mgs_config_body *body);
2430 struct mgs_config_res {
2431 __u64 mcr_offset; /* index of last config log */
2432 __u64 mcr_size; /* size of the log */
2434 extern void lustre_swab_mgs_config_res(struct mgs_config_res *body);
2436 /* Config marker flags (in config log) */
2437 #define CM_START 0x01
2439 #define CM_SKIP 0x04
2440 #define CM_UPGRADE146 0x08
2441 #define CM_EXCLUDE 0x10
2442 #define CM_START_SKIP (CM_START | CM_SKIP)
2445 __u32 cm_step; /* aka config version */
2447 __u32 cm_vers; /* lustre release version number */
2448 __u32 cm_padding; /* 64 bit align */
2449 obd_time cm_createtime; /*when this record was first created */
2450 obd_time cm_canceltime; /*when this record is no longer valid*/
2451 char cm_tgtname[MTI_NAME_MAXLEN];
2452 char cm_comment[MTI_NAME_MAXLEN];
2455 extern void lustre_swab_cfg_marker(struct cfg_marker *marker,
2456 int swab, int size);
2459 * Opcodes for multiple servers.
2468 #define OBD_FIRST_OPC OBD_PING
2470 /* catalog of log objects */
2472 /** Identifier for a single log object */
2477 } __attribute__((packed));
2479 /** Records written to the CATALOGS list */
2480 #define CATLIST "CATALOGS"
2482 struct llog_logid lci_logid;
2486 } __attribute__((packed));
2488 /* Log data record types - there is no specific reason that these need to
2489 * be related to the RPC opcodes, but no reason not to (may be handy later?)
2491 #define LLOG_OP_MAGIC 0x10600000
2492 #define LLOG_OP_MASK 0xfff00000
2495 LLOG_PAD_MAGIC = LLOG_OP_MAGIC | 0x00000,
2496 OST_SZ_REC = LLOG_OP_MAGIC | 0x00f00,
2497 OST_RAID1_REC = LLOG_OP_MAGIC | 0x01000,
2498 MDS_UNLINK_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) | REINT_UNLINK,
2499 MDS_SETATTR_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) | REINT_SETATTR,
2500 MDS_SETATTR64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) | REINT_SETATTR,
2501 OBD_CFG_REC = LLOG_OP_MAGIC | 0x20000,
2502 PTL_CFG_REC = LLOG_OP_MAGIC | 0x30000, /* obsolete */
2503 LLOG_GEN_REC = LLOG_OP_MAGIC | 0x40000,
2504 LLOG_JOIN_REC = LLOG_OP_MAGIC | 0x50000, /* obsolete */
2505 CHANGELOG_REC = LLOG_OP_MAGIC | 0x60000,
2506 CHANGELOG_USER_REC = LLOG_OP_MAGIC | 0x70000,
2507 LLOG_HDR_MAGIC = LLOG_OP_MAGIC | 0x45539,
2508 LLOG_LOGID_MAGIC = LLOG_OP_MAGIC | 0x4553b,
2512 * for now, continue to support old pad records which have 0 for their
2513 * type but still need to be swabbed for their length
2515 #define LLOG_REC_HDR_NEEDS_SWABBING(r) \
2516 (((r)->lrh_type & __swab32(LLOG_OP_MASK)) == \
2517 __swab32(LLOG_OP_MAGIC) || \
2518 (((r)->lrh_type == 0) && ((r)->lrh_len > LLOG_CHUNK_SIZE)))
2520 /** Log record header - stored in little endian order.
2521 * Each record must start with this struct, end with a llog_rec_tail,
2522 * and be a multiple of 256 bits in size.
2524 struct llog_rec_hdr {
2531 struct llog_rec_tail {
2536 struct llog_logid_rec {
2537 struct llog_rec_hdr lid_hdr;
2538 struct llog_logid lid_id;
2544 struct llog_rec_tail lid_tail;
2545 } __attribute__((packed));
2547 struct llog_create_rec {
2548 struct llog_rec_hdr lcr_hdr;
2549 struct ll_fid lcr_fid;
2553 struct llog_rec_tail lcr_tail;
2554 } __attribute__((packed));
2556 struct llog_orphan_rec {
2557 struct llog_rec_hdr lor_hdr;
2561 struct llog_rec_tail lor_tail;
2562 } __attribute__((packed));
2564 struct llog_unlink_rec {
2565 struct llog_rec_hdr lur_hdr;
2568 obd_count lur_count;
2569 struct llog_rec_tail lur_tail;
2570 } __attribute__((packed));
2572 struct llog_setattr_rec {
2573 struct llog_rec_hdr lsr_hdr;
2579 struct llog_rec_tail lsr_tail;
2580 } __attribute__((packed));
2582 struct llog_setattr64_rec {
2583 struct llog_rec_hdr lsr_hdr;
2591 struct llog_rec_tail lsr_tail;
2592 } __attribute__((packed));
2594 struct llog_size_change_rec {
2595 struct llog_rec_hdr lsc_hdr;
2596 struct ll_fid lsc_fid;
2599 struct llog_rec_tail lsc_tail;
2600 } __attribute__((packed));
2602 #define CHANGELOG_MAGIC 0xca103000
2604 /** \a changelog_rec_type's that can't be masked */
2605 #define CHANGELOG_MINMASK (1 << CL_MARK)
2606 /** bits covering all \a changelog_rec_type's */
2607 #define CHANGELOG_ALLMASK 0XFFFFFFFF
2608 /** default \a changelog_rec_type mask */
2609 #define CHANGELOG_DEFMASK CHANGELOG_ALLMASK & ~(1 << CL_ATIME | 1 << CL_CLOSE)
2611 /* changelog llog name, needed by client replicators */
2612 #define CHANGELOG_CATALOG "changelog_catalog"
2614 struct changelog_setinfo {
2617 } __attribute__((packed));
2619 /** changelog record */
2620 struct llog_changelog_rec {
2621 struct llog_rec_hdr cr_hdr;
2622 struct changelog_rec cr;
2623 struct llog_rec_tail cr_tail; /**< for_sizezof_only */
2624 } __attribute__((packed));
2626 #define CHANGELOG_USER_PREFIX "cl"
2628 struct llog_changelog_user_rec {
2629 struct llog_rec_hdr cur_hdr;
2633 struct llog_rec_tail cur_tail;
2634 } __attribute__((packed));
2639 } __attribute__((packed));
2641 struct llog_gen_rec {
2642 struct llog_rec_hdr lgr_hdr;
2643 struct llog_gen lgr_gen;
2644 struct llog_rec_tail lgr_tail;
2646 /* On-disk header structure of each log object, stored in little endian order */
2647 #define LLOG_CHUNK_SIZE 8192
2648 #define LLOG_HEADER_SIZE (96)
2649 #define LLOG_BITMAP_BYTES (LLOG_CHUNK_SIZE - LLOG_HEADER_SIZE)
2651 #define LLOG_MIN_REC_SIZE (24) /* round(llog_rec_hdr + llog_rec_tail) */
2653 /* flags for the logs */
2654 #define LLOG_F_ZAP_WHEN_EMPTY 0x1
2655 #define LLOG_F_IS_CAT 0x2
2656 #define LLOG_F_IS_PLAIN 0x4
2658 struct llog_log_hdr {
2659 struct llog_rec_hdr llh_hdr;
2660 obd_time llh_timestamp;
2662 __u32 llh_bitmap_offset;
2666 /* for a catalog the first plain slot is next to it */
2667 struct obd_uuid llh_tgtuuid;
2668 __u32 llh_reserved[LLOG_HEADER_SIZE/sizeof(__u32) - 23];
2669 __u32 llh_bitmap[LLOG_BITMAP_BYTES/sizeof(__u32)];
2670 struct llog_rec_tail llh_tail;
2671 } __attribute__((packed));
2673 #define LLOG_BITMAP_SIZE(llh) ((llh->llh_hdr.lrh_len - \
2674 llh->llh_bitmap_offset - \
2675 sizeof(llh->llh_tail)) * 8)
2677 /** log cookies are used to reference a specific log file and a record therein */
2678 struct llog_cookie {
2679 struct llog_logid lgc_lgl;
2683 } __attribute__((packed));
2685 /** llog protocol */
2686 enum llogd_rpc_ops {
2687 LLOG_ORIGIN_HANDLE_CREATE = 501,
2688 LLOG_ORIGIN_HANDLE_NEXT_BLOCK = 502,
2689 LLOG_ORIGIN_HANDLE_READ_HEADER = 503,
2690 LLOG_ORIGIN_HANDLE_WRITE_REC = 504,
2691 LLOG_ORIGIN_HANDLE_CLOSE = 505,
2692 LLOG_ORIGIN_CONNECT = 506,
2693 LLOG_CATINFO = 507, /* for lfs catinfo */
2694 LLOG_ORIGIN_HANDLE_PREV_BLOCK = 508,
2695 LLOG_ORIGIN_HANDLE_DESTROY = 509, /* for destroy llog object*/
2697 LLOG_FIRST_OPC = LLOG_ORIGIN_HANDLE_CREATE
2701 struct llog_logid lgd_logid;
2703 __u32 lgd_llh_flags;
2705 __u32 lgd_saved_index;
2707 __u64 lgd_cur_offset;
2708 } __attribute__((packed));
2710 struct llogd_conn_body {
2711 struct llog_gen lgdc_gen;
2712 struct llog_logid lgdc_logid;
2713 __u32 lgdc_ctxt_idx;
2714 } __attribute__((packed));
2716 /* Note: 64-bit types are 64-bit aligned in structure */
2718 obd_valid o_valid; /* hot fields in this obdo */
2720 obd_id o_parent_seq;
2721 obd_size o_size; /* o_size-o_blocks == ost_lvb */
2725 obd_blocks o_blocks; /* brw: cli sent cached bytes */
2728 /* 32-bit fields start here: keep an even number of them via padding */
2729 obd_blksize o_blksize; /* optimal IO blocksize */
2730 obd_mode o_mode; /* brw: cli sent cache remain */
2734 obd_count o_nlink; /* brw: checksum */
2735 obd_count o_parent_oid;
2736 obd_count o_misc; /* brw: o_dropped */
2737 __u64 o_ioepoch; /* epoch in ost writes */
2738 __u32 o_stripe_idx; /* holds stripe idx */
2740 struct lustre_handle o_handle; /* brw: lock handle to prolong
2742 struct llog_cookie o_lcookie; /* destroy: unlink cookie from
2747 __u64 o_data_version; /* getattr: sum of iversion for
2749 * brw: grant space consumed on
2750 * the client for the write */
2756 #define o_id o_oi.oi_id
2757 #define o_seq o_oi.oi_seq
2758 #define o_dirty o_blocks
2759 #define o_undirty o_mode
2760 #define o_dropped o_misc
2761 #define o_cksum o_nlink
2762 #define o_grant_used o_data_version
2764 static inline void lustre_set_wire_obdo(struct obdo *wobdo, struct obdo *lobdo)
2766 memcpy(wobdo, lobdo, sizeof(*lobdo));
2767 wobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
2770 static inline void lustre_get_wire_obdo(struct obdo *lobdo, struct obdo *wobdo)
2772 obd_flag local_flags = 0;
2774 if (lobdo->o_valid & OBD_MD_FLFLAGS)
2775 local_flags = lobdo->o_flags & OBD_FL_LOCAL_MASK;
2777 LASSERT(!(wobdo->o_flags & OBD_FL_LOCAL_MASK));
2779 memcpy(lobdo, wobdo, sizeof(*lobdo));
2780 if (local_flags != 0) {
2781 lobdo->o_valid |= OBD_MD_FLFLAGS;
2782 lobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
2783 lobdo->o_flags |= local_flags;
2787 extern void lustre_swab_obdo (struct obdo *o);
2789 /* request structure for OST's */
2794 /* Key for FIEMAP to be used in get_info calls */
2795 struct ll_fiemap_info_key {
2798 struct ll_user_fiemap fiemap;
2801 extern void lustre_swab_ost_body (struct ost_body *b);
2802 extern void lustre_swab_ost_last_id(obd_id *id);
2803 extern void lustre_swab_fiemap(struct ll_user_fiemap *fiemap);
2805 extern void lustre_swab_lov_user_md_v1(struct lov_user_md_v1 *lum);
2806 extern void lustre_swab_lov_user_md_v3(struct lov_user_md_v3 *lum);
2807 extern void lustre_swab_lov_user_md_objects(struct lov_user_ost_data *lod,
2809 extern void lustre_swab_lov_mds_md(struct lov_mds_md *lmm);
2812 extern void lustre_swab_llogd_body (struct llogd_body *d);
2813 extern void lustre_swab_llog_hdr (struct llog_log_hdr *h);
2814 extern void lustre_swab_llogd_conn_body (struct llogd_conn_body *d);
2815 extern void lustre_swab_llog_rec(struct llog_rec_hdr *rec,
2816 struct llog_rec_tail *tail);
2819 extern void lustre_swab_lustre_cfg(struct lustre_cfg *lcfg);
2821 /* Functions for dumping PTLRPC fields */
2822 void dump_rniobuf(struct niobuf_remote *rnb);
2823 void dump_ioo(struct obd_ioobj *nb);
2824 void dump_obdo(struct obdo *oa);
2825 void dump_ost_body(struct ost_body *ob);
2826 void dump_rcs(__u32 *rc);
2828 /* this will be used when OBD_CONNECT_CHANGE_QS is set */
2831 * ID appiles to (uid, gid)
2835 * LQUOTA_FLAGS_* affect the responding bits
2839 * acquire/release count (bytes for block quota)
2843 * when a master returns the reply to a slave, it will
2844 * contain the current corresponding qunit size
2850 #define QDATA_IS_GRP(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_GRP)
2851 #define QDATA_IS_BLK(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_BLK)
2852 #define QDATA_IS_ADJBLK(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_ADJBLK)
2853 #define QDATA_IS_ADJINO(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_ADJINO)
2854 #define QDATA_IS_CHANGE_QS(qdata) ((qdata)->qd_flags & LQUOTA_FLAGS_CHG_QS)
2856 #define QDATA_SET_GRP(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_GRP)
2857 #define QDATA_SET_BLK(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_BLK)
2858 #define QDATA_SET_ADJBLK(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_ADJBLK)
2859 #define QDATA_SET_ADJINO(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_ADJINO)
2860 #define QDATA_SET_CHANGE_QS(qdata) ((qdata)->qd_flags |= LQUOTA_FLAGS_CHG_QS)
2862 #define QDATA_CLR_GRP(qdata) ((qdata)->qd_flags &= ~LQUOTA_FLAGS_GRP)
2863 #define QDATA_CLR_CHANGE_QS(qdata) ((qdata)->qd_flags &= ~LQUOTA_FLAGS_CHG_QS)
2865 extern void lustre_swab_qdata(struct qunit_data *d);
2866 extern struct qunit_data *quota_get_qdata(void *req, int is_req, int is_exp);
2867 extern int quota_copy_qdata(void *request, struct qunit_data *qdata,
2868 int is_req, int is_exp);
2875 #define QUOTA_FIRST_OPC QUOTA_DQACQ
2877 #define QUOTA_REQUEST 1
2878 #define QUOTA_REPLY 0
2879 #define QUOTA_EXPORT 1
2880 #define QUOTA_IMPORT 0
2882 /* quota check function */
2883 #define QUOTA_RET_OK 0 /**< return successfully */
2884 #define QUOTA_RET_NOQUOTA 1 /**< not support quota */
2885 #define QUOTA_RET_NOLIMIT 2 /**< quota limit isn't set */
2886 #define QUOTA_RET_ACQUOTA 4 /**< need to acquire extra quota */
2889 /* security opcodes */
2892 SEC_CTX_INIT_CONT = 802,
2895 SEC_FIRST_OPC = SEC_CTX_INIT
2899 * capa related definitions
2901 #define CAPA_HMAC_MAX_LEN 64
2902 #define CAPA_HMAC_KEY_MAX_LEN 56
2904 /* NB take care when changing the sequence of elements this struct,
2905 * because the offset info is used in find_capa() */
2906 struct lustre_capa {
2907 struct lu_fid lc_fid; /** fid */
2908 __u64 lc_opc; /** operations allowed */
2909 __u64 lc_uid; /** file owner */
2910 __u64 lc_gid; /** file group */
2911 __u32 lc_flags; /** HMAC algorithm & flags */
2912 __u32 lc_keyid; /** key# used for the capability */
2913 __u32 lc_timeout; /** capa timeout value (sec) */
2914 __u32 lc_expiry; /** expiry time (sec) */
2915 __u8 lc_hmac[CAPA_HMAC_MAX_LEN]; /** HMAC */
2916 } __attribute__((packed));
2918 extern void lustre_swab_lustre_capa(struct lustre_capa *c);
2920 /** lustre_capa::lc_opc */
2922 CAPA_OPC_BODY_WRITE = 1<<0, /**< write object data */
2923 CAPA_OPC_BODY_READ = 1<<1, /**< read object data */
2924 CAPA_OPC_INDEX_LOOKUP = 1<<2, /**< lookup object fid */
2925 CAPA_OPC_INDEX_INSERT = 1<<3, /**< insert object fid */
2926 CAPA_OPC_INDEX_DELETE = 1<<4, /**< delete object fid */
2927 CAPA_OPC_OSS_WRITE = 1<<5, /**< write oss object data */
2928 CAPA_OPC_OSS_READ = 1<<6, /**< read oss object data */
2929 CAPA_OPC_OSS_TRUNC = 1<<7, /**< truncate oss object */
2930 CAPA_OPC_OSS_DESTROY = 1<<8, /**< destroy oss object */
2931 CAPA_OPC_META_WRITE = 1<<9, /**< write object meta data */
2932 CAPA_OPC_META_READ = 1<<10, /**< read object meta data */
2935 #define CAPA_OPC_OSS_RW (CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE)
2936 #define CAPA_OPC_MDS_ONLY \
2937 (CAPA_OPC_BODY_WRITE | CAPA_OPC_BODY_READ | CAPA_OPC_INDEX_LOOKUP | \
2938 CAPA_OPC_INDEX_INSERT | CAPA_OPC_INDEX_DELETE)
2939 #define CAPA_OPC_OSS_ONLY \
2940 (CAPA_OPC_OSS_WRITE | CAPA_OPC_OSS_READ | CAPA_OPC_OSS_TRUNC | \
2941 CAPA_OPC_OSS_DESTROY)
2942 #define CAPA_OPC_MDS_DEFAULT ~CAPA_OPC_OSS_ONLY
2943 #define CAPA_OPC_OSS_DEFAULT ~(CAPA_OPC_MDS_ONLY | CAPA_OPC_OSS_ONLY)
2945 /* MDS capability covers object capability for operations of body r/w
2946 * (dir readpage/sendpage), index lookup/insert/delete and meta data r/w,
2947 * while OSS capability only covers object capability for operations of
2948 * oss data(file content) r/w/truncate.
2950 static inline int capa_for_mds(struct lustre_capa *c)
2952 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) != 0;
2955 static inline int capa_for_oss(struct lustre_capa *c)
2957 return (c->lc_opc & CAPA_OPC_INDEX_LOOKUP) == 0;
2960 /* lustre_capa::lc_hmac_alg */
2962 CAPA_HMAC_ALG_SHA1 = 1, /**< sha1 algorithm */
2966 #define CAPA_FL_MASK 0x00ffffff
2967 #define CAPA_HMAC_ALG_MASK 0xff000000
2969 struct lustre_capa_key {
2970 __u64 lk_seq; /**< mds# */
2971 __u32 lk_keyid; /**< key# */
2973 __u8 lk_key[CAPA_HMAC_KEY_MAX_LEN]; /**< key */
2974 } __attribute__((packed));
2976 extern void lustre_swab_lustre_capa_key(struct lustre_capa_key *k);
2978 /** The link ea holds 1 \a link_ea_entry for each hardlink */
2979 #define LINK_EA_MAGIC 0x11EAF1DFUL
2980 struct link_ea_header {
2983 __u64 leh_len; /* total size */
2989 /** Hardlink data is name and parent fid.
2990 * Stored in this crazy struct for maximum packing and endian-neutrality
2992 struct link_ea_entry {
2993 /** __u16 stored big-endian, unaligned */
2994 unsigned char lee_reclen[2];
2995 unsigned char lee_parent_fid[sizeof(struct lu_fid)];
2997 }__attribute__((packed));
2999 /** fid2path request/reply structure */
3000 struct getinfo_fid2path {
3001 struct lu_fid gf_fid;
3006 } __attribute__((packed));
3008 void lustre_swab_fid2path (struct getinfo_fid2path *gf);