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, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/include/lustre/lustre_idl.h
38 * Lustre wire protocol definitions.
41 /** \defgroup lustreidl lustreidl
43 * Lustre wire protocol definitions.
45 * ALL structs passing over the wire should be declared here. Structs
46 * that are used in interfaces with userspace should go in lustre_user.h.
48 * All structs being declared here should be built from simple fixed-size
49 * types (__u8, __u16, __u32, __u64) or be built from other types or
50 * structs also declared in this file. Similarly, all flags and magic
51 * values in those structs should also be declared here. This ensures
52 * that the Lustre wire protocol is not influenced by external dependencies.
54 * The only other acceptable items in this file are VERY SIMPLE accessor
55 * functions to avoid callers grubbing inside the structures, and the
56 * prototypes of the swabber functions for each struct. Nothing that
57 * depends on external functions or definitions should be in here.
59 * Structs must be properly aligned to put 64-bit values on an 8-byte
60 * boundary. Any structs being added here must also be added to
61 * utils/wirecheck.c and "make newwiretest" run to regenerate the
62 * utils/wiretest.c sources. This allows us to verify that wire structs
63 * have the proper alignment/size on all architectures.
65 * DO NOT CHANGE any of the structs, flags, values declared here and used
66 * in released Lustre versions. Some structs may have padding fields that
67 * can be used. Some structs might allow addition at the end (verify this
68 * in the code to ensure that new/old clients that see this larger struct
69 * do not fail, otherwise you need to implement protocol compatibility).
71 * We assume all nodes are either little-endian or big-endian, and we
72 * always send messages in the sender's native format. The receiver
73 * detects the message format by checking the 'magic' field of the message
74 * (see lustre_msg_swabbed() below).
76 * Each wire type has corresponding 'lustre_swab_xxxtypexxx()' routines,
77 * implemented either here, inline (trivial implementations) or in
78 * ptlrpc/pack_generic.c. These 'swabbers' convert the type from "other"
79 * endian, in-place in the message buffer.
81 * A swabber takes a single pointer argument. The caller must already have
82 * verified that the length of the message buffer >= sizeof (type).
84 * For variable length types, a second 'lustre_swab_v_xxxtypexxx()' routine
85 * may be defined that swabs just the variable part, after the caller has
86 * verified that the message buffer is large enough.
91 #ifndef _LUSTRE_IDL_H_
92 #define _LUSTRE_IDL_H_
94 #include <libcfs/libcfs.h>
95 #include <lnet/types.h>
96 #include <lustre/lustre_user.h> /* Defn's shared with user-space. */
97 #include <lustre/lustre_errno.h>
98 #include <lustre_ver.h>
103 /* FOO_REQUEST_PORTAL is for incoming requests on the FOO
104 * FOO_REPLY_PORTAL is for incoming replies on the FOO
105 * FOO_BULK_PORTAL is for incoming bulk on the FOO
108 #define CONNMGR_REQUEST_PORTAL 1
109 #define CONNMGR_REPLY_PORTAL 2
110 //#define OSC_REQUEST_PORTAL 3
111 #define OSC_REPLY_PORTAL 4
112 //#define OSC_BULK_PORTAL 5
113 #define OST_IO_PORTAL 6
114 #define OST_CREATE_PORTAL 7
115 #define OST_BULK_PORTAL 8
116 //#define MDC_REQUEST_PORTAL 9
117 #define MDC_REPLY_PORTAL 10
118 //#define MDC_BULK_PORTAL 11
119 #define MDS_REQUEST_PORTAL 12
120 //#define MDS_REPLY_PORTAL 13
121 #define MDS_BULK_PORTAL 14
122 #define LDLM_CB_REQUEST_PORTAL 15
123 #define LDLM_CB_REPLY_PORTAL 16
124 #define LDLM_CANCEL_REQUEST_PORTAL 17
125 #define LDLM_CANCEL_REPLY_PORTAL 18
126 //#define PTLBD_REQUEST_PORTAL 19
127 //#define PTLBD_REPLY_PORTAL 20
128 //#define PTLBD_BULK_PORTAL 21
129 #define MDS_SETATTR_PORTAL 22
130 #define MDS_READPAGE_PORTAL 23
131 #define OUT_PORTAL 24
132 #define MGC_REPLY_PORTAL 25
133 #define MGS_REQUEST_PORTAL 26
134 #define MGS_REPLY_PORTAL 27
135 #define OST_REQUEST_PORTAL 28
136 #define FLD_REQUEST_PORTAL 29
137 #define SEQ_METADATA_PORTAL 30
138 #define SEQ_DATA_PORTAL 31
139 #define SEQ_CONTROLLER_PORTAL 32
140 #define MGS_BULK_PORTAL 33
142 /* Portal 63 is reserved for the Cray Inc DVS - nic@cray.com, roe@cray.com, n8851@cray.com */
145 #define PTL_RPC_MSG_REQUEST 4711
146 #define PTL_RPC_MSG_ERR 4712
147 #define PTL_RPC_MSG_REPLY 4713
149 /* DON'T use swabbed values of MAGIC as magic! */
150 #define LUSTRE_MSG_MAGIC_V2 0x0BD00BD3
151 #define LUSTRE_MSG_MAGIC_V2_SWABBED 0xD30BD00B
153 #define LUSTRE_MSG_MAGIC LUSTRE_MSG_MAGIC_V2
155 #define PTLRPC_MSG_VERSION 0x00000003
156 #define LUSTRE_VERSION_MASK 0xffff0000
157 #define LUSTRE_OBD_VERSION 0x00010000
158 #define LUSTRE_MDS_VERSION 0x00020000
159 #define LUSTRE_OST_VERSION 0x00030000
160 #define LUSTRE_DLM_VERSION 0x00040000
161 #define LUSTRE_LOG_VERSION 0x00050000
162 #define LUSTRE_MGS_VERSION 0x00060000
165 * Describes a range of sequence, lsr_start is included but lsr_end is
167 * Same structure is used in fld module where lsr_index field holds mdt id
170 struct lu_seq_range {
177 struct lu_seq_range_array {
180 struct lu_seq_range lsra_lsr[0];
183 #define LU_SEQ_RANGE_MDT 0x0
184 #define LU_SEQ_RANGE_OST 0x1
185 #define LU_SEQ_RANGE_ANY 0x3
187 #define LU_SEQ_RANGE_MASK 0x3
189 /** \defgroup lu_fid lu_fid
193 * Flags for lustre_mdt_attrs::lma_compat and lustre_mdt_attrs::lma_incompat.
194 * Deprecated since HSM and SOM attributes are now stored in separate on-disk
198 LMAC_HSM = 0x00000001,
199 /* LMAC_SOM = 0x00000002, obsolete since 2.8.0 */
200 LMAC_NOT_IN_OI = 0x00000004, /* the object does NOT need OI mapping */
201 LMAC_FID_ON_OST = 0x00000008, /* For OST-object, its OI mapping is
202 * under /O/<seq>/d<x>. */
206 * Masks for all features that should be supported by a Lustre version to
207 * access a specific file.
208 * This information is stored in lustre_mdt_attrs::lma_incompat.
211 LMAI_RELEASED = 0x00000001, /* file is released */
212 LMAI_AGENT = 0x00000002, /* agent inode */
213 LMAI_REMOTE_PARENT = 0x00000004, /* the parent of the object
214 is on the remote MDT */
215 LMAI_STRIPED = 0x00000008, /* striped directory inode */
217 #define LMA_INCOMPAT_SUPP (LMAI_AGENT | LMAI_REMOTE_PARENT | LMAI_STRIPED)
219 extern void lustre_lma_swab(struct lustre_mdt_attrs *lma);
220 extern void lustre_lma_init(struct lustre_mdt_attrs *lma,
221 const struct lu_fid *fid,
222 __u32 compat, __u32 incompat);
224 /* copytool uses a 32b bitmask field to encode archive-Ids during register
226 * archive num = 0 => all
227 * archive num from 1 to 32
229 #define LL_HSM_MAX_ARCHIVE (sizeof(__u32) * 8)
232 * HSM on-disk attributes stored in a separate xattr.
235 /** Bitfield for supported data in this structure. For future use. */
238 /** HSM flags, see hsm_flags enum below */
240 /** backend archive id associated with the file */
242 /** version associated with the last archiving, if any */
245 extern void lustre_hsm_swab(struct hsm_attrs *attrs);
251 /** LASTID file has zero OID */
252 LUSTRE_FID_LASTID_OID = 0UL,
253 /** initial fid id value */
254 LUSTRE_FID_INIT_OID = 1UL
257 /** returns fid object sequence */
258 static inline __u64 fid_seq(const struct lu_fid *fid)
263 /** returns fid object id */
264 static inline __u32 fid_oid(const struct lu_fid *fid)
269 /** returns fid object version */
270 static inline __u32 fid_ver(const struct lu_fid *fid)
275 static inline void fid_zero(struct lu_fid *fid)
277 memset(fid, 0, sizeof(*fid));
280 static inline __u64 fid_ver_oid(const struct lu_fid *fid)
282 return ((__u64)fid_ver(fid) << 32 | fid_oid(fid));
286 * Note that reserved SEQ numbers below 12 will conflict with ldiskfs
287 * inodes in the IGIF namespace, so these reserved SEQ numbers can be
288 * used for other purposes and not risk collisions with existing inodes.
290 * Different FID Format
291 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs#NEW.0
294 FID_SEQ_OST_MDT0 = 0,
295 FID_SEQ_LLOG = 1, /* unnamed llogs */
297 FID_SEQ_UNUSED_START = 3,
298 FID_SEQ_UNUSED_END = 9,
299 FID_SEQ_LLOG_NAME = 10, /* named llogs */
302 FID_SEQ_IGIF_MAX = 0x0ffffffffULL,
303 FID_SEQ_IDIF = 0x100000000ULL,
304 FID_SEQ_IDIF_MAX = 0x1ffffffffULL,
305 /* Normal FID sequence starts from this value, i.e. 1<<33 */
306 FID_SEQ_START = 0x200000000ULL,
307 /* sequence for local pre-defined FIDs listed in local_oid */
308 FID_SEQ_LOCAL_FILE = 0x200000001ULL,
309 FID_SEQ_DOT_LUSTRE = 0x200000002ULL,
310 /* sequence is used for local named objects FIDs generated
311 * by local_object_storage library */
312 FID_SEQ_LOCAL_NAME = 0x200000003ULL,
313 /* Because current FLD will only cache the fid sequence, instead
314 * of oid on the client side, if the FID needs to be exposed to
315 * clients sides, it needs to make sure all of fids under one
316 * sequence will be located in one MDT. */
317 FID_SEQ_SPECIAL = 0x200000004ULL,
318 FID_SEQ_QUOTA = 0x200000005ULL,
319 FID_SEQ_QUOTA_GLB = 0x200000006ULL,
320 FID_SEQ_ROOT = 0x200000007ULL, /* Located on MDT0 */
321 FID_SEQ_LAYOUT_RBTREE = 0x200000008ULL,
322 /* sequence is used for update logs of cross-MDT operation */
323 FID_SEQ_UPDATE_LOG = 0x200000009ULL,
324 /* Sequence is used for the directory under which update logs
326 FID_SEQ_UPDATE_LOG_DIR = 0x20000000aULL,
327 FID_SEQ_NORMAL = 0x200000400ULL,
328 FID_SEQ_LOV_DEFAULT = 0xffffffffffffffffULL
331 #define OBIF_OID_MAX_BITS 32
332 #define OBIF_MAX_OID (1ULL << OBIF_OID_MAX_BITS)
333 #define OBIF_OID_MASK ((1ULL << OBIF_OID_MAX_BITS) - 1)
334 #define IDIF_OID_MAX_BITS 48
335 #define IDIF_MAX_OID (1ULL << IDIF_OID_MAX_BITS)
336 #define IDIF_OID_MASK ((1ULL << IDIF_OID_MAX_BITS) - 1)
338 /** OID for FID_SEQ_SPECIAL */
340 /* Big Filesystem Lock to serialize rename operations */
341 FID_OID_SPECIAL_BFL = 1UL,
344 /** OID for FID_SEQ_DOT_LUSTRE */
345 enum dot_lustre_oid {
346 FID_OID_DOT_LUSTRE = 1UL,
347 FID_OID_DOT_LUSTRE_OBF = 2UL,
348 FID_OID_DOT_LUSTRE_LPF = 3UL,
351 /** OID for FID_SEQ_ROOT */
354 FID_OID_ECHO_ROOT = 2UL,
357 static inline bool fid_seq_is_mdt0(__u64 seq)
359 return seq == FID_SEQ_OST_MDT0;
362 static inline bool fid_seq_is_mdt(__u64 seq)
364 return seq == FID_SEQ_OST_MDT0 || seq >= FID_SEQ_NORMAL;
367 static inline bool fid_seq_is_echo(__u64 seq)
369 return seq == FID_SEQ_ECHO;
372 static inline bool fid_is_echo(const struct lu_fid *fid)
374 return fid_seq_is_echo(fid_seq(fid));
377 static inline bool fid_seq_is_llog(__u64 seq)
379 return seq == FID_SEQ_LLOG;
382 static inline bool fid_is_llog(const struct lu_fid *fid)
384 /* file with OID == 0 is not llog but contains last oid */
385 return fid_seq_is_llog(fid_seq(fid)) && fid_oid(fid) > 0;
388 static inline bool fid_seq_is_rsvd(__u64 seq)
390 return seq > FID_SEQ_OST_MDT0 && seq <= FID_SEQ_RSVD;
393 static inline bool fid_seq_is_special(__u64 seq)
395 return seq == FID_SEQ_SPECIAL;
398 static inline bool fid_seq_is_local_file(__u64 seq)
400 return seq == FID_SEQ_LOCAL_FILE ||
401 seq == FID_SEQ_LOCAL_NAME;
404 static inline bool fid_seq_is_root(__u64 seq)
406 return seq == FID_SEQ_ROOT;
409 static inline bool fid_seq_is_dot(__u64 seq)
411 return seq == FID_SEQ_DOT_LUSTRE;
414 static inline bool fid_seq_is_default(__u64 seq)
416 return seq == FID_SEQ_LOV_DEFAULT;
419 static inline bool fid_is_mdt0(const struct lu_fid *fid)
421 return fid_seq_is_mdt0(fid_seq(fid));
424 static inline void lu_root_fid(struct lu_fid *fid)
426 fid->f_seq = FID_SEQ_ROOT;
427 fid->f_oid = FID_OID_ROOT;
431 static inline void lu_echo_root_fid(struct lu_fid *fid)
433 fid->f_seq = FID_SEQ_ROOT;
434 fid->f_oid = FID_OID_ECHO_ROOT;
438 static inline void lu_update_log_fid(struct lu_fid *fid, __u32 index)
440 fid->f_seq = FID_SEQ_UPDATE_LOG;
445 static inline void lu_update_log_dir_fid(struct lu_fid *fid, __u32 index)
447 fid->f_seq = FID_SEQ_UPDATE_LOG_DIR;
453 * Check if a fid is igif or not.
454 * \param fid the fid to be tested.
455 * \return true if the fid is an igif; otherwise false.
457 static inline bool fid_seq_is_igif(__u64 seq)
459 return seq >= FID_SEQ_IGIF && seq <= FID_SEQ_IGIF_MAX;
462 static inline bool fid_is_igif(const struct lu_fid *fid)
464 return fid_seq_is_igif(fid_seq(fid));
468 * Check if a fid is idif or not.
469 * \param fid the fid to be tested.
470 * \return true if the fid is an idif; otherwise false.
472 static inline bool fid_seq_is_idif(__u64 seq)
474 return seq >= FID_SEQ_IDIF && seq <= FID_SEQ_IDIF_MAX;
477 static inline bool fid_is_idif(const struct lu_fid *fid)
479 return fid_seq_is_idif(fid_seq(fid));
482 static inline bool fid_is_local_file(const struct lu_fid *fid)
484 return fid_seq_is_local_file(fid_seq(fid));
487 static inline bool fid_seq_is_norm(__u64 seq)
489 return (seq >= FID_SEQ_NORMAL);
492 static inline bool fid_is_norm(const struct lu_fid *fid)
494 return fid_seq_is_norm(fid_seq(fid));
497 static inline int fid_is_layout_rbtree(const struct lu_fid *fid)
499 return fid_seq(fid) == FID_SEQ_LAYOUT_RBTREE;
502 static inline bool fid_seq_is_update_log(__u64 seq)
504 return seq == FID_SEQ_UPDATE_LOG;
507 static inline bool fid_is_update_log(const struct lu_fid *fid)
509 return fid_seq_is_update_log(fid_seq(fid));
512 static inline bool fid_seq_is_update_log_dir(__u64 seq)
514 return seq == FID_SEQ_UPDATE_LOG_DIR;
517 static inline bool fid_is_update_log_dir(const struct lu_fid *fid)
519 return fid_seq_is_update_log_dir(fid_seq(fid));
522 /* convert an OST objid into an IDIF FID SEQ number */
523 static inline __u64 fid_idif_seq(__u64 id, __u32 ost_idx)
525 return FID_SEQ_IDIF | (ost_idx << 16) | ((id >> 32) & 0xffff);
528 /* convert a packed IDIF FID into an OST objid */
529 static inline __u64 fid_idif_id(__u64 seq, __u32 oid, __u32 ver)
531 return ((__u64)ver << 48) | ((seq & 0xffff) << 32) | oid;
534 static inline __u32 idif_ost_idx(__u64 seq)
536 return (seq >> 16) & 0xffff;
539 /* extract ost index from IDIF FID */
540 static inline __u32 fid_idif_ost_idx(const struct lu_fid *fid)
542 return idif_ost_idx(fid_seq(fid));
545 /* extract OST sequence (group) from a wire ost_id (id/seq) pair */
546 static inline __u64 ostid_seq(const struct ost_id *ostid)
548 if (fid_seq_is_mdt0(ostid->oi.oi_seq))
549 return FID_SEQ_OST_MDT0;
551 if (unlikely(fid_seq_is_default(ostid->oi.oi_seq)))
552 return FID_SEQ_LOV_DEFAULT;
554 if (fid_is_idif(&ostid->oi_fid))
555 return FID_SEQ_OST_MDT0;
557 return fid_seq(&ostid->oi_fid);
560 /* extract OST objid from a wire ost_id (id/seq) pair */
561 static inline __u64 ostid_id(const struct ost_id *ostid)
563 if (fid_seq_is_mdt0(ostid->oi.oi_seq))
564 return ostid->oi.oi_id & IDIF_OID_MASK;
566 if (unlikely(fid_seq_is_default(ostid->oi.oi_seq)))
567 return ostid->oi.oi_id;
569 if (fid_is_idif(&ostid->oi_fid))
570 return fid_idif_id(fid_seq(&ostid->oi_fid),
571 fid_oid(&ostid->oi_fid), 0);
573 return fid_oid(&ostid->oi_fid);
576 static inline void ostid_set_seq(struct ost_id *oi, __u64 seq)
578 if (fid_seq_is_mdt0(seq) || fid_seq_is_default(seq)) {
581 oi->oi_fid.f_seq = seq;
582 /* Note: if f_oid + f_ver is zero, we need init it
583 * to be 1, otherwise, ostid_seq will treat this
584 * as old ostid (oi_seq == 0) */
585 if (oi->oi_fid.f_oid == 0 && oi->oi_fid.f_ver == 0)
586 oi->oi_fid.f_oid = LUSTRE_FID_INIT_OID;
590 static inline void ostid_set_seq_mdt0(struct ost_id *oi)
592 ostid_set_seq(oi, FID_SEQ_OST_MDT0);
595 static inline void ostid_set_seq_echo(struct ost_id *oi)
597 ostid_set_seq(oi, FID_SEQ_ECHO);
600 static inline void ostid_set_seq_llog(struct ost_id *oi)
602 ostid_set_seq(oi, FID_SEQ_LLOG);
606 * Note: we need check oi_seq to decide where to set oi_id,
607 * so oi_seq should always be set ahead of oi_id.
609 static inline void ostid_set_id(struct ost_id *oi, __u64 oid)
611 if (fid_seq_is_mdt0(oi->oi.oi_seq)) {
612 if (oid >= IDIF_MAX_OID) {
613 CERROR("Too large OID %#llx to set MDT0 "DOSTID"\n",
614 (unsigned long long)oid, POSTID(oi));
618 } else if (fid_is_idif(&oi->oi_fid)) {
619 if (oid >= IDIF_MAX_OID) {
620 CERROR("Too large OID %#llx to set IDIF "DOSTID"\n",
621 (unsigned long long)oid, POSTID(oi));
624 oi->oi_fid.f_seq = fid_idif_seq(oid,
625 fid_idif_ost_idx(&oi->oi_fid));
626 oi->oi_fid.f_oid = oid;
627 oi->oi_fid.f_ver = oid >> 48;
629 if (oid > OBIF_MAX_OID) {
630 CERROR("Too large oid %#llx to set REG "DOSTID"\n",
631 (unsigned long long)oid, POSTID(oi));
634 oi->oi_fid.f_oid = oid;
638 static inline int fid_set_id(struct lu_fid *fid, __u64 oid)
640 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
641 CERROR("bad IGIF, "DFID"\n", PFID(fid));
645 if (fid_is_idif(fid)) {
646 if (oid >= IDIF_MAX_OID) {
647 CERROR("Too large OID %#llx to set IDIF "DFID"\n",
648 (unsigned long long)oid, PFID(fid));
651 fid->f_seq = fid_idif_seq(oid, fid_idif_ost_idx(fid));
653 fid->f_ver = oid >> 48;
655 if (oid > OBIF_MAX_OID) {
656 CERROR("Too large OID %#llx to set REG "DFID"\n",
657 (unsigned long long)oid, PFID(fid));
666 * Unpack an OST object id/seq (group) into a FID. This is needed for
667 * converting all obdo, lmm, lsm, etc. 64-bit id/seq pairs into proper
668 * FIDs. Note that if an id/seq is already in FID/IDIF format it will
669 * be passed through unchanged. Only legacy OST objects in "group 0"
670 * will be mapped into the IDIF namespace so that they can fit into the
671 * struct lu_fid fields without loss. For reference see:
672 * http://arch.lustre.org/index.php?title=Interoperability_fids_zfs
674 static inline int ostid_to_fid(struct lu_fid *fid, const struct ost_id *ostid,
677 __u64 seq = ostid_seq(ostid);
679 if (ost_idx > 0xffff) {
680 CERROR("bad ost_idx, "DOSTID" ost_idx:%u\n", POSTID(ostid),
685 if (fid_seq_is_mdt0(seq)) {
686 __u64 oid = ostid_id(ostid);
688 /* This is a "legacy" (old 1.x/2.early) OST object in "group 0"
689 * that we map into the IDIF namespace. It allows up to 2^48
690 * objects per OST, as this is the object namespace that has
691 * been in production for years. This can handle create rates
692 * of 1M objects/s/OST for 9 years, or combinations thereof. */
693 if (oid >= IDIF_MAX_OID) {
694 CERROR("bad MDT0 id(1), "DOSTID" ost_idx:%u\n",
695 POSTID(ostid), ost_idx);
698 fid->f_seq = fid_idif_seq(oid, ost_idx);
699 /* truncate to 32 bits by assignment */
701 /* in theory, not currently used */
702 fid->f_ver = oid >> 48;
703 } else if (likely(!fid_seq_is_default(seq)))
704 /* if (fid_seq_is_idif(seq) || fid_seq_is_norm(seq)) */ {
705 /* This is either an IDIF object, which identifies objects
706 * across all OSTs, or a regular FID. The IDIF namespace maps
707 * legacy OST objects into the FID namespace. In both cases,
708 * we just pass the FID through, no conversion needed. */
709 if (ostid->oi_fid.f_ver != 0) {
710 CERROR("bad MDT0 id(2), "DOSTID" ost_idx:%u\n",
711 POSTID(ostid), ost_idx);
714 *fid = ostid->oi_fid;
720 /* pack any OST FID into an ostid (id/seq) for the wire/disk */
721 static inline int fid_to_ostid(const struct lu_fid *fid, struct ost_id *ostid)
723 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
724 CERROR("bad IGIF, "DFID"\n", PFID(fid));
728 if (fid_is_idif(fid)) {
729 ostid_set_seq_mdt0(ostid);
730 ostid_set_id(ostid, fid_idif_id(fid_seq(fid), fid_oid(fid),
733 ostid->oi_fid = *fid;
739 /* Check whether the fid is for LAST_ID */
740 static inline bool fid_is_last_id(const struct lu_fid *fid)
742 return fid_oid(fid) == 0 && fid_seq(fid) != FID_SEQ_UPDATE_LOG &&
743 fid_seq(fid) != FID_SEQ_UPDATE_LOG_DIR;
747 * Get inode number from an igif.
748 * \param fid an igif to get inode number from.
749 * \return inode number for the igif.
751 static inline ino_t lu_igif_ino(const struct lu_fid *fid)
756 extern void lustre_swab_ost_id(struct ost_id *oid);
759 * Get inode generation from an igif.
760 * \param fid an igif to get inode generation from.
761 * \return inode generation for the igif.
763 static inline __u32 lu_igif_gen(const struct lu_fid *fid)
769 * Build igif from the inode number/generation.
771 static inline void lu_igif_build(struct lu_fid *fid, __u32 ino, __u32 gen)
779 * Fids are transmitted across network (in the sender byte-ordering),
780 * and stored on disk in big-endian order.
782 static inline void fid_cpu_to_le(struct lu_fid *dst, const struct lu_fid *src)
784 dst->f_seq = cpu_to_le64(fid_seq(src));
785 dst->f_oid = cpu_to_le32(fid_oid(src));
786 dst->f_ver = cpu_to_le32(fid_ver(src));
789 static inline void fid_le_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
791 dst->f_seq = le64_to_cpu(fid_seq(src));
792 dst->f_oid = le32_to_cpu(fid_oid(src));
793 dst->f_ver = le32_to_cpu(fid_ver(src));
796 static inline void fid_cpu_to_be(struct lu_fid *dst, const struct lu_fid *src)
798 dst->f_seq = cpu_to_be64(fid_seq(src));
799 dst->f_oid = cpu_to_be32(fid_oid(src));
800 dst->f_ver = cpu_to_be32(fid_ver(src));
803 static inline void fid_be_to_cpu(struct lu_fid *dst, const struct lu_fid *src)
805 dst->f_seq = be64_to_cpu(fid_seq(src));
806 dst->f_oid = be32_to_cpu(fid_oid(src));
807 dst->f_ver = be32_to_cpu(fid_ver(src));
810 static inline bool fid_is_sane(const struct lu_fid *fid)
812 return fid != NULL &&
813 ((fid_seq(fid) >= FID_SEQ_START && fid_ver(fid) == 0) ||
814 fid_is_igif(fid) || fid_is_idif(fid) ||
815 fid_seq_is_rsvd(fid_seq(fid)));
818 extern void lustre_swab_lu_fid(struct lu_fid *fid);
820 static inline bool lu_fid_eq(const struct lu_fid *f0, const struct lu_fid *f1)
822 return memcmp(f0, f1, sizeof *f0) == 0;
825 #define __diff_normalize(val0, val1) \
827 typeof(val0) __val0 = (val0); \
828 typeof(val1) __val1 = (val1); \
830 (__val0 == __val1 ? 0 : __val0 > __val1 ? +1 : -1); \
833 static inline int lu_fid_cmp(const struct lu_fid *f0,
834 const struct lu_fid *f1)
837 __diff_normalize(fid_seq(f0), fid_seq(f1)) ?:
838 __diff_normalize(fid_oid(f0), fid_oid(f1)) ?:
839 __diff_normalize(fid_ver(f0), fid_ver(f1));
842 static inline void ostid_cpu_to_le(const struct ost_id *src_oi,
843 struct ost_id *dst_oi)
845 if (fid_seq_is_mdt0(src_oi->oi.oi_seq)) {
846 dst_oi->oi.oi_id = cpu_to_le64(src_oi->oi.oi_id);
847 dst_oi->oi.oi_seq = cpu_to_le64(src_oi->oi.oi_seq);
849 fid_cpu_to_le(&dst_oi->oi_fid, &src_oi->oi_fid);
853 static inline void ostid_le_to_cpu(const struct ost_id *src_oi,
854 struct ost_id *dst_oi)
856 if (fid_seq_is_mdt0(src_oi->oi.oi_seq)) {
857 dst_oi->oi.oi_id = le64_to_cpu(src_oi->oi.oi_id);
858 dst_oi->oi.oi_seq = le64_to_cpu(src_oi->oi.oi_seq);
860 fid_le_to_cpu(&dst_oi->oi_fid, &src_oi->oi_fid);
864 struct lu_orphan_rec {
865 /* The MDT-object's FID referenced by the orphan OST-object */
866 struct lu_fid lor_fid;
871 struct lu_orphan_ent {
872 /* The orphan OST-object's FID */
873 struct lu_fid loe_key;
874 struct lu_orphan_rec loe_rec;
876 void lustre_swab_orphan_ent(struct lu_orphan_ent *ent);
880 /** \defgroup lu_dir lu_dir
884 * Enumeration of possible directory entry attributes.
886 * Attributes follow directory entry header in the order they appear in this
889 enum lu_dirent_attrs {
892 LUDA_64BITHASH = 0x0004,
894 /* The following attrs are used for MDT internal only,
895 * not visible to client */
897 /* Something in the record is unknown, to be verified in further. */
898 LUDA_UNKNOWN = 0x0400,
899 /* Ignore this record, go to next directly. */
900 LUDA_IGNORE = 0x0800,
901 /* The system is upgraded, has beed or to be repaired (dryrun). */
902 LUDA_UPGRADE = 0x1000,
903 /* The dirent has been repaired, or to be repaired (dryrun). */
904 LUDA_REPAIR = 0x2000,
905 /* Only check but not repair the dirent inconsistency */
906 LUDA_VERIFY_DRYRUN = 0x4000,
907 /* Verify the dirent consistency */
908 LUDA_VERIFY = 0x8000,
911 #define LU_DIRENT_ATTRS_MASK 0xff00
914 * Layout of readdir pages, as transmitted on wire.
917 /** valid if LUDA_FID is set. */
918 struct lu_fid lde_fid;
919 /** a unique entry identifier: a hash or an offset. */
921 /** total record length, including all attributes. */
925 /** optional variable size attributes following this entry.
926 * taken from enum lu_dirent_attrs.
929 /** name is followed by the attributes indicated in ->ldp_attrs, in
930 * their natural order. After the last attribute, padding bytes are
931 * added to make ->lde_reclen a multiple of 8.
937 * Definitions of optional directory entry attributes formats.
939 * Individual attributes do not have their length encoded in a generic way. It
940 * is assumed that consumer of an attribute knows its format. This means that
941 * it is impossible to skip over an unknown attribute, except by skipping over all
942 * remaining attributes (by using ->lde_reclen), which is not too
943 * constraining, because new server versions will append new attributes at
944 * the end of an entry.
948 * Fid directory attribute: a fid of an object referenced by the entry. This
949 * will be almost always requested by the client and supplied by the server.
951 * Aligned to 8 bytes.
953 /* To have compatibility with 1.8, lets have fid in lu_dirent struct. */
958 * Aligned to 2 bytes.
965 __u64 ldp_hash_start;
969 struct lu_dirent ldp_entries[0];
972 enum lu_dirpage_flags {
974 * dirpage contains no entry.
978 * last entry's lde_hash equals ldp_hash_end.
983 static inline struct lu_dirent *lu_dirent_start(struct lu_dirpage *dp)
985 if (le32_to_cpu(dp->ldp_flags) & LDF_EMPTY)
988 return dp->ldp_entries;
991 static inline struct lu_dirent *lu_dirent_next(struct lu_dirent *ent)
993 struct lu_dirent *next;
995 if (le16_to_cpu(ent->lde_reclen) != 0)
996 next = ((void *)ent) + le16_to_cpu(ent->lde_reclen);
1003 static inline size_t lu_dirent_calc_size(size_t namelen, __u16 attr)
1007 if (attr & LUDA_TYPE) {
1008 const size_t align = sizeof(struct luda_type) - 1;
1009 size = (sizeof(struct lu_dirent) + namelen + align) & ~align;
1010 size += sizeof(struct luda_type);
1012 size = sizeof(struct lu_dirent) + namelen;
1014 return (size + 7) & ~7;
1017 #define MDS_DIR_END_OFF 0xfffffffffffffffeULL
1020 * MDS_READPAGE page size
1022 * This is the directory page size packed in MDS_READPAGE RPC.
1023 * It's different than PAGE_CACHE_SIZE because the client needs to
1024 * access the struct lu_dirpage header packed at the beginning of
1025 * the "page" and without this there isn't any way to know find the
1026 * lu_dirpage header is if client and server PAGE_CACHE_SIZE differ.
1028 #define LU_PAGE_SHIFT 12
1029 #define LU_PAGE_SIZE (1UL << LU_PAGE_SHIFT)
1030 #define LU_PAGE_MASK (~(LU_PAGE_SIZE - 1))
1032 #define LU_PAGE_COUNT (1 << (PAGE_CACHE_SHIFT - LU_PAGE_SHIFT))
1036 struct lustre_handle {
1039 #define DEAD_HANDLE_MAGIC 0xdeadbeefcafebabeULL
1041 static inline bool lustre_handle_is_used(const struct lustre_handle *lh)
1043 return lh->cookie != 0;
1046 static inline bool lustre_handle_equal(const struct lustre_handle *lh1,
1047 const struct lustre_handle *lh2)
1049 return lh1->cookie == lh2->cookie;
1052 static inline void lustre_handle_copy(struct lustre_handle *tgt,
1053 const struct lustre_handle *src)
1055 tgt->cookie = src->cookie;
1058 /* flags for lm_flags */
1059 #define MSGHDR_AT_SUPPORT 0x1
1060 #define MSGHDR_CKSUM_INCOMPAT18 0x2
1062 #define lustre_msg lustre_msg_v2
1063 /* we depend on this structure to be 8-byte aligned */
1064 /* this type is only endian-adjusted in lustre_unpack_msg() */
1065 struct lustre_msg_v2 {
1074 __u32 lm_buflens[0];
1077 /* without gss, ptlrpc_body is put at the first buffer. */
1078 #define PTLRPC_NUM_VERSIONS 4
1079 struct ptlrpc_body_v3 {
1080 struct lustre_handle pb_handle;
1085 __u64 pb_last_xid; /* highest replied XID without lower unreplied XID */
1086 __u16 pb_tag; /* virtual slot idx for multiple modifying RPCs */
1089 __u64 pb_last_committed;
1094 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1095 __u32 pb_service_time; /* for rep, actual service time */
1098 /* VBR: pre-versions */
1099 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1100 __u64 pb_mbits; /**< match bits for bulk request */
1101 /* padding for future needs */
1102 __u64 pb_padding64_0;
1103 __u64 pb_padding64_1;
1104 __u64 pb_padding64_2;
1105 char pb_jobid[LUSTRE_JOBID_SIZE];
1107 #define ptlrpc_body ptlrpc_body_v3
1109 struct ptlrpc_body_v2 {
1110 struct lustre_handle pb_handle;
1115 __u64 pb_last_xid; /* highest replied XID without lower unreplied XID */
1116 __u16 pb_tag; /* virtual slot idx for multiple modifying RPCs */
1119 __u64 pb_last_committed;
1124 __u32 pb_timeout; /* for req, the deadline, for rep, the service est */
1125 __u32 pb_service_time; /* for rep, actual service time, also used for
1126 net_latency of req */
1129 /* VBR: pre-versions */
1130 __u64 pb_pre_versions[PTLRPC_NUM_VERSIONS];
1131 __u64 pb_mbits; /**< unused in V2 */
1132 /* padding for future needs */
1133 __u64 pb_padding64_0;
1134 __u64 pb_padding64_1;
1135 __u64 pb_padding64_2;
1138 extern void lustre_swab_ptlrpc_body(struct ptlrpc_body *pb);
1140 /* message body offset for lustre_msg_v2 */
1141 /* ptlrpc body offset in all request/reply messages */
1142 #define MSG_PTLRPC_BODY_OFF 0
1144 /* normal request/reply message record offset */
1145 #define REQ_REC_OFF 1
1146 #define REPLY_REC_OFF 1
1148 /* ldlm request message body offset */
1149 #define DLM_LOCKREQ_OFF 1 /* lockreq offset */
1150 #define DLM_REQ_REC_OFF 2 /* normal dlm request record offset */
1152 /* ldlm intent lock message body offset */
1153 #define DLM_INTENT_IT_OFF 2 /* intent lock it offset */
1154 #define DLM_INTENT_REC_OFF 3 /* intent lock record offset */
1156 /* ldlm reply message body offset */
1157 #define DLM_LOCKREPLY_OFF 1 /* lockrep offset */
1158 #define DLM_REPLY_REC_OFF 2 /* reply record offset */
1160 /** only use in req->rq_{req,rep}_swab_mask */
1161 #define MSG_PTLRPC_HEADER_OFF 31
1163 /* Flags that are operation-specific go in the top 16 bits. */
1164 #define MSG_OP_FLAG_MASK 0xffff0000
1165 #define MSG_OP_FLAG_SHIFT 16
1167 /* Flags that apply to all requests are in the bottom 16 bits */
1168 #define MSG_GEN_FLAG_MASK 0x0000ffff
1169 #define MSG_LAST_REPLAY 0x0001
1170 #define MSG_RESENT 0x0002
1171 #define MSG_REPLAY 0x0004
1172 /* #define MSG_AT_SUPPORT 0x0008
1173 * This was used in early prototypes of adaptive timeouts, and while there
1174 * shouldn't be any users of that code there also isn't a need for using this
1175 * bits. Defer usage until at least 1.10 to avoid potential conflict. */
1176 #define MSG_DELAY_REPLAY 0x0010
1177 #define MSG_VERSION_REPLAY 0x0020
1178 #define MSG_REQ_REPLAY_DONE 0x0040
1179 #define MSG_LOCK_REPLAY_DONE 0x0080
1182 * Flags for all connect opcodes (MDS_CONNECT, OST_CONNECT)
1185 #define MSG_CONNECT_RECOVERING 0x00000001
1186 #define MSG_CONNECT_RECONNECT 0x00000002
1187 #define MSG_CONNECT_REPLAYABLE 0x00000004
1188 //#define MSG_CONNECT_PEER 0x8
1189 #define MSG_CONNECT_LIBCLIENT 0x00000010
1190 #define MSG_CONNECT_INITIAL 0x00000020
1191 #define MSG_CONNECT_ASYNC 0x00000040
1192 #define MSG_CONNECT_NEXT_VER 0x00000080 /* use next version of lustre_msg */
1193 #define MSG_CONNECT_TRANSNO 0x00000100 /* report transno */
1196 #define OBD_CONNECT_RDONLY 0x1ULL /*client has read-only access*/
1197 #define OBD_CONNECT_INDEX 0x2ULL /*connect specific LOV idx */
1198 #define OBD_CONNECT_MDS 0x4ULL /*connect from MDT to OST */
1199 #define OBD_CONNECT_GRANT 0x8ULL /*OSC gets grant at connect */
1200 #define OBD_CONNECT_SRVLOCK 0x10ULL /*server takes locks for cli */
1201 #define OBD_CONNECT_VERSION 0x20ULL /*Lustre versions in ocd */
1202 #define OBD_CONNECT_REQPORTAL 0x40ULL /*Separate non-IO req portal */
1203 #define OBD_CONNECT_ACL 0x80ULL /*access control lists */
1204 #define OBD_CONNECT_XATTR 0x100ULL /*client use extended attr */
1205 #define OBD_CONNECT_CROW 0x200ULL /*MDS+OST create obj on write*/
1206 #define OBD_CONNECT_TRUNCLOCK 0x400ULL /*locks on server for punch */
1207 #define OBD_CONNECT_TRANSNO 0x800ULL /*replay sends init transno */
1208 #define OBD_CONNECT_IBITS 0x1000ULL /*support for inodebits locks*/
1209 #define OBD_CONNECT_JOIN 0x2000ULL /*files can be concatenated.
1210 *We do not support JOIN FILE
1211 *anymore, reserve this flags
1212 *just for preventing such bit
1214 #define OBD_CONNECT_ATTRFID 0x4000ULL /*Server can GetAttr By Fid*/
1215 #define OBD_CONNECT_NODEVOH 0x8000ULL /*No open hndl on specl nodes*/
1216 #define OBD_CONNECT_RMT_CLIENT 0x10000ULL /*Remote client */
1217 #define OBD_CONNECT_RMT_CLIENT_FORCE 0x20000ULL /*Remote client by force */
1218 #define OBD_CONNECT_BRW_SIZE 0x40000ULL /*Max bytes per rpc */
1219 #define OBD_CONNECT_QUOTA64 0x80000ULL /*Not used since 2.4 */
1220 #define OBD_CONNECT_MDS_CAPA 0x100000ULL /*MDS capability */
1221 #define OBD_CONNECT_OSS_CAPA 0x200000ULL /*OSS capability */
1222 #define OBD_CONNECT_CANCELSET 0x400000ULL /*Early batched cancels. */
1223 #define OBD_CONNECT_SOM 0x800000ULL /*Size on MDS */
1224 #define OBD_CONNECT_AT 0x1000000ULL /*client uses AT */
1225 #define OBD_CONNECT_LRU_RESIZE 0x2000000ULL /*LRU resize feature. */
1226 #define OBD_CONNECT_MDS_MDS 0x4000000ULL /*MDS-MDS connection */
1227 #define OBD_CONNECT_REAL 0x8000000ULL /*real connection */
1228 #define OBD_CONNECT_CHANGE_QS 0x10000000ULL /*Not used since 2.4 */
1229 #define OBD_CONNECT_CKSUM 0x20000000ULL /*support several cksum algos*/
1230 #define OBD_CONNECT_FID 0x40000000ULL /*FID is supported by server */
1231 #define OBD_CONNECT_VBR 0x80000000ULL /*version based recovery */
1232 #define OBD_CONNECT_LOV_V3 0x100000000ULL /*client supports LOV v3 EA */
1233 #define OBD_CONNECT_GRANT_SHRINK 0x200000000ULL /* support grant shrink */
1234 #define OBD_CONNECT_SKIP_ORPHAN 0x400000000ULL /* don't reuse orphan objids */
1235 #define OBD_CONNECT_MAX_EASIZE 0x800000000ULL /* preserved for large EA */
1236 #define OBD_CONNECT_FULL20 0x1000000000ULL /* it is 2.0 client */
1237 #define OBD_CONNECT_LAYOUTLOCK 0x2000000000ULL /* client uses layout lock */
1238 #define OBD_CONNECT_64BITHASH 0x4000000000ULL /* client supports 64-bits
1240 #define OBD_CONNECT_MAXBYTES 0x8000000000ULL /* max stripe size */
1241 #define OBD_CONNECT_IMP_RECOV 0x10000000000ULL /* imp recovery support */
1242 #define OBD_CONNECT_JOBSTATS 0x20000000000ULL /* jobid in ptlrpc_body */
1243 #define OBD_CONNECT_UMASK 0x40000000000ULL /* create uses client umask */
1244 #define OBD_CONNECT_EINPROGRESS 0x80000000000ULL /* client handles -EINPROGRESS
1245 * RPC error properly */
1246 #define OBD_CONNECT_GRANT_PARAM 0x100000000000ULL/* extra grant params used for
1247 * finer space reservation */
1248 #define OBD_CONNECT_FLOCK_OWNER 0x200000000000ULL /* for the fixed 1.8
1249 * policy and 2.x server */
1250 #define OBD_CONNECT_LVB_TYPE 0x400000000000ULL /* variable type of LVB */
1251 #define OBD_CONNECT_NANOSEC_TIME 0x800000000000ULL /* nanosecond timestamps */
1252 #define OBD_CONNECT_LIGHTWEIGHT 0x1000000000000ULL/* lightweight connection */
1253 #define OBD_CONNECT_SHORTIO 0x2000000000000ULL/* short io */
1254 #define OBD_CONNECT_PINGLESS 0x4000000000000ULL/* pings not required */
1255 #define OBD_CONNECT_FLOCK_DEAD 0x8000000000000ULL/* improved flock deadlock detection */
1256 #define OBD_CONNECT_DISP_STRIPE 0x10000000000000ULL/* create stripe disposition*/
1257 #define OBD_CONNECT_OPEN_BY_FID 0x20000000000000ULL /* open by fid won't pack
1259 #define OBD_CONNECT_LFSCK 0x40000000000000ULL/* support online LFSCK */
1260 #define OBD_CONNECT_UNLINK_CLOSE 0x100000000000000ULL/* close file in unlink */
1261 #define OBD_CONNECT_MULTIMODRPCS 0x200000000000000ULL /* support multiple modify
1263 #define OBD_CONNECT_DIR_STRIPE 0x400000000000000ULL /* striped DNE dir */
1264 #define OBD_CONNECT_SUBTREE 0x800000000000000ULL /* fileset mount */
1265 #define OBD_CONNECT_LOCK_AHEAD 0x1000000000000000ULL /* lock ahead */
1266 /** bulk matchbits is sent within ptlrpc_body */
1267 #define OBD_CONNECT_BULK_MBITS 0x2000000000000000ULL
1269 * Please DO NOT add flag values here before first ensuring that this same
1270 * flag value is not in use on some other branch. Please clear any such
1271 * changes with senior engineers before starting to use a new flag. Then,
1272 * submit a small patch against EVERY branch that ONLY adds the new flag,
1273 * updates obd_connect_names[] for lprocfs_rd_connect_flags(), adds the
1274 * flag to check_obd_connect_data(), and updates wiretests accordingly, so it
1275 * can be approved and landed easily to reserve the flag for future use. */
1277 /* The MNE_SWAB flag is overloading the MDS_MDS bit only for the MGS
1278 * connection. It is a temporary bug fix for Imperative Recovery interop
1279 * between 2.2 and 2.3 x86/ppc nodes, and can be removed when interop for
1280 * 2.2 clients/servers is no longer needed. LU-1252/LU-1644. */
1281 #define OBD_CONNECT_MNE_SWAB OBD_CONNECT_MDS_MDS
1283 #define OCD_HAS_FLAG(ocd, flg) \
1284 (!!((ocd)->ocd_connect_flags & OBD_CONNECT_##flg))
1287 #ifdef HAVE_LRU_RESIZE_SUPPORT
1288 #define LRU_RESIZE_CONNECT_FLAG OBD_CONNECT_LRU_RESIZE
1290 #define LRU_RESIZE_CONNECT_FLAG 0
1293 #define MDT_CONNECT_SUPPORTED (OBD_CONNECT_RDONLY | OBD_CONNECT_VERSION | \
1294 OBD_CONNECT_ACL | OBD_CONNECT_XATTR | \
1295 OBD_CONNECT_IBITS | \
1296 OBD_CONNECT_NODEVOH | OBD_CONNECT_ATTRFID | \
1297 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1298 OBD_CONNECT_RMT_CLIENT | \
1299 OBD_CONNECT_RMT_CLIENT_FORCE | \
1300 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_MDS_MDS | \
1301 OBD_CONNECT_FID | LRU_RESIZE_CONNECT_FLAG | \
1302 OBD_CONNECT_VBR | OBD_CONNECT_LOV_V3 | \
1303 OBD_CONNECT_FULL20 | \
1304 OBD_CONNECT_64BITHASH | OBD_CONNECT_JOBSTATS | \
1305 OBD_CONNECT_EINPROGRESS | \
1306 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_UMASK | \
1307 OBD_CONNECT_LVB_TYPE | OBD_CONNECT_LAYOUTLOCK |\
1308 OBD_CONNECT_PINGLESS | OBD_CONNECT_MAX_EASIZE |\
1309 OBD_CONNECT_FLOCK_DEAD | \
1310 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK | \
1311 OBD_CONNECT_OPEN_BY_FID | \
1312 OBD_CONNECT_DIR_STRIPE | \
1313 OBD_CONNECT_BULK_MBITS | \
1314 OBD_CONNECT_MULTIMODRPCS)
1316 #define OST_CONNECT_SUPPORTED (OBD_CONNECT_SRVLOCK | OBD_CONNECT_GRANT | \
1317 OBD_CONNECT_REQPORTAL | OBD_CONNECT_VERSION | \
1318 OBD_CONNECT_TRUNCLOCK | OBD_CONNECT_INDEX | \
1319 OBD_CONNECT_BRW_SIZE | \
1320 OBD_CONNECT_CANCELSET | OBD_CONNECT_AT | \
1321 LRU_RESIZE_CONNECT_FLAG | OBD_CONNECT_CKSUM | \
1322 OBD_CONNECT_RMT_CLIENT | \
1323 OBD_CONNECT_RMT_CLIENT_FORCE | OBD_CONNECT_VBR | \
1324 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN | \
1325 OBD_CONNECT_GRANT_SHRINK | OBD_CONNECT_FULL20 | \
1326 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES | \
1327 OBD_CONNECT_MAX_EASIZE | \
1328 OBD_CONNECT_EINPROGRESS | \
1329 OBD_CONNECT_JOBSTATS | \
1330 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LVB_TYPE|\
1331 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_FID | \
1332 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK | \
1333 OBD_CONNECT_BULK_MBITS)
1334 #define ECHO_CONNECT_SUPPORTED (0)
1335 #define MGS_CONNECT_SUPPORTED (OBD_CONNECT_VERSION | OBD_CONNECT_AT | \
1336 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV | \
1337 OBD_CONNECT_MNE_SWAB | OBD_CONNECT_PINGLESS |\
1338 OBD_CONNECT_BULK_MBITS)
1340 /* Features required for this version of the client to work with server */
1341 #define CLIENT_CONNECT_MDT_REQD (OBD_CONNECT_IBITS | OBD_CONNECT_FID | \
1344 /* This structure is used for both request and reply.
1346 * If we eventually have separate connect data for different types, which we
1347 * almost certainly will, then perhaps we stick a union in here. */
1348 struct obd_connect_data {
1349 __u64 ocd_connect_flags; /* OBD_CONNECT_* per above */
1350 __u32 ocd_version; /* lustre release version number */
1351 __u32 ocd_grant; /* initial cache grant amount (bytes) */
1352 __u32 ocd_index; /* LOV index to connect to */
1353 __u32 ocd_brw_size; /* Maximum BRW size in bytes */
1354 __u64 ocd_ibits_known; /* inode bits this client understands */
1355 __u8 ocd_blocksize; /* log2 of the backend filesystem blocksize */
1356 __u8 ocd_inodespace; /* log2 of the per-inode space consumption */
1357 __u16 ocd_grant_extent; /* per-extent grant overhead, in 1K blocks */
1358 __u32 ocd_unused; /* also fix lustre_swab_connect */
1359 __u64 ocd_transno; /* first transno from client to be replayed */
1360 __u32 ocd_group; /* MDS group on OST */
1361 __u32 ocd_cksum_types; /* supported checksum algorithms */
1362 __u32 ocd_max_easize; /* How big LOV EA can be on MDS */
1363 __u32 ocd_instance; /* instance # of this target */
1364 __u64 ocd_maxbytes; /* Maximum stripe size in bytes */
1365 /* Fields after ocd_maxbytes are only accessible by the receiver
1366 * if the corresponding flag in ocd_connect_flags is set. Accessing
1367 * any field after ocd_maxbytes on the receiver without a valid flag
1368 * may result in out-of-bound memory access and kernel oops. */
1369 __u16 ocd_maxmodrpcs; /* Maximum modify RPCs in parallel */
1370 __u16 padding0; /* added 2.1.0. also fix lustre_swab_connect */
1371 __u32 padding1; /* added 2.1.0. also fix lustre_swab_connect */
1372 __u64 padding2; /* added 2.1.0. also fix lustre_swab_connect */
1373 __u64 padding3; /* added 2.1.0. also fix lustre_swab_connect */
1374 __u64 padding4; /* added 2.1.0. also fix lustre_swab_connect */
1375 __u64 padding5; /* added 2.1.0. also fix lustre_swab_connect */
1376 __u64 padding6; /* added 2.1.0. also fix lustre_swab_connect */
1377 __u64 padding7; /* added 2.1.0. also fix lustre_swab_connect */
1378 __u64 padding8; /* added 2.1.0. also fix lustre_swab_connect */
1379 __u64 padding9; /* added 2.1.0. also fix lustre_swab_connect */
1380 __u64 paddingA; /* added 2.1.0. also fix lustre_swab_connect */
1381 __u64 paddingB; /* added 2.1.0. also fix lustre_swab_connect */
1382 __u64 paddingC; /* added 2.1.0. also fix lustre_swab_connect */
1383 __u64 paddingD; /* added 2.1.0. also fix lustre_swab_connect */
1384 __u64 paddingE; /* added 2.1.0. also fix lustre_swab_connect */
1385 __u64 paddingF; /* added 2.1.0. also fix lustre_swab_connect */
1388 * Please DO NOT use any fields here before first ensuring that this same
1389 * field is not in use on some other branch. Please clear any such changes
1390 * with senior engineers before starting to use a new field. Then, submit
1391 * a small patch against EVERY branch that ONLY adds the new field along with
1392 * the matching OBD_CONNECT flag, so that can be approved and landed easily to
1393 * reserve the flag for future use. */
1396 extern void lustre_swab_connect(struct obd_connect_data *ocd);
1399 * Supported checksum algorithms. Up to 32 checksum types are supported.
1400 * (32-bit mask stored in obd_connect_data::ocd_cksum_types)
1401 * Please update DECLARE_CKSUM_NAME/OBD_CKSUM_ALL in obd.h when adding a new
1402 * algorithm and also the OBD_FL_CKSUM* flags.
1405 OBD_CKSUM_CRC32 = 0x00000001,
1406 OBD_CKSUM_ADLER = 0x00000002,
1407 OBD_CKSUM_CRC32C= 0x00000004,
1411 * OST requests: OBDO & OBD request records
1416 OST_REPLY = 0, /* reply ? */
1432 OST_QUOTACHECK = 18, /* not used since 2.4 */
1434 OST_QUOTA_ADJUST_QUNIT = 20, /* not used since 2.4 */
1437 #define OST_FIRST_OPC OST_REPLY
1440 OBD_FL_INLINEDATA = 0x00000001,
1441 OBD_FL_OBDMDEXISTS = 0x00000002,
1442 OBD_FL_DELORPHAN = 0x00000004, /* if set in o_flags delete orphans */
1443 OBD_FL_NORPC = 0x00000008, /* set in o_flags do in OSC not OST */
1444 OBD_FL_IDONLY = 0x00000010, /* set in o_flags only adjust obj id*/
1445 OBD_FL_RECREATE_OBJS= 0x00000020, /* recreate missing obj */
1446 OBD_FL_DEBUG_CHECK = 0x00000040, /* echo client/server debug check */
1447 OBD_FL_NO_USRQUOTA = 0x00000100, /* the object's owner is over quota */
1448 OBD_FL_NO_GRPQUOTA = 0x00000200, /* the object's group is over quota */
1449 OBD_FL_CREATE_CROW = 0x00000400, /* object should be create on write */
1450 OBD_FL_SRVLOCK = 0x00000800, /* delegate DLM locking to server */
1451 OBD_FL_CKSUM_CRC32 = 0x00001000, /* CRC32 checksum type */
1452 OBD_FL_CKSUM_ADLER = 0x00002000, /* ADLER checksum type */
1453 OBD_FL_CKSUM_CRC32C = 0x00004000, /* CRC32C checksum type */
1454 OBD_FL_CKSUM_RSVD2 = 0x00008000, /* for future cksum types */
1455 OBD_FL_CKSUM_RSVD3 = 0x00010000, /* for future cksum types */
1456 OBD_FL_SHRINK_GRANT = 0x00020000, /* object shrink the grant */
1457 OBD_FL_MMAP = 0x00040000, /* object is mmapped on the client.
1458 * XXX: obsoleted - reserved for old
1459 * clients prior than 2.2 */
1460 OBD_FL_RECOV_RESEND = 0x00080000, /* recoverable resent */
1461 OBD_FL_NOSPC_BLK = 0x00100000, /* no more block space on OST */
1462 OBD_FL_FLUSH = 0x00200000, /* flush pages on the OST */
1463 OBD_FL_SHORT_IO = 0x00400000, /* short io request */
1465 /* Note that while these checksum values are currently separate bits,
1466 * in 2.x we can actually allow all values from 1-31 if we wanted. */
1467 OBD_FL_CKSUM_ALL = OBD_FL_CKSUM_CRC32 | OBD_FL_CKSUM_ADLER |
1468 OBD_FL_CKSUM_CRC32C,
1470 /* mask for local-only flag, which won't be sent over network */
1471 OBD_FL_LOCAL_MASK = 0xF0000000,
1475 * All LOV EA magics should have the same postfix, if some new version
1476 * Lustre instroduces new LOV EA magic, then when down-grade to an old
1477 * Lustre, even though the old version system does not recognizes such
1478 * new magic, it still can distinguish the corrupted cases by checking
1479 * the magic's postfix.
1481 #define LOV_MAGIC_MAGIC 0x0BD0
1482 #define LOV_MAGIC_MASK 0xFFFF
1484 #define LOV_MAGIC_V1 (0x0BD10000 | LOV_MAGIC_MAGIC)
1485 #define LOV_MAGIC_JOIN_V1 (0x0BD20000 | LOV_MAGIC_MAGIC)
1486 #define LOV_MAGIC_V3 (0x0BD30000 | LOV_MAGIC_MAGIC)
1487 #define LOV_MAGIC_MIGRATE (0x0BD40000 | LOV_MAGIC_MAGIC)
1488 /* reserved for specifying OSTs */
1489 #define LOV_MAGIC_SPECIFIC (0x0BD50000 | LOV_MAGIC_MAGIC)
1490 #define LOV_MAGIC LOV_MAGIC_V1
1493 * magic for fully defined striping
1494 * the idea is that we should have different magics for striping "hints"
1495 * (struct lov_user_md_v[13]) and defined ready-to-use striping (struct
1496 * lov_mds_md_v[13]). at the moment the magics are used in wire protocol,
1497 * we can't just change it w/o long way preparation, but we still need a
1498 * mechanism to allow LOD to differentiate hint versus ready striping.
1499 * so, at the moment we do a trick: MDT knows what to expect from request
1500 * depending on the case (replay uses ready striping, non-replay req uses
1501 * hints), so MDT replaces magic with appropriate one and now LOD can
1502 * easily understand what's inside -bzzz
1504 #define LOV_MAGIC_V1_DEF 0x0CD10BD0
1505 #define LOV_MAGIC_V3_DEF 0x0CD30BD0
1507 #define lov_pattern(pattern) (pattern & ~LOV_PATTERN_F_MASK)
1508 #define lov_pattern_flags(pattern) (pattern & LOV_PATTERN_F_MASK)
1510 #define lov_ost_data lov_ost_data_v1
1511 struct lov_ost_data_v1 { /* per-stripe data structure (little-endian)*/
1512 struct ost_id l_ost_oi; /* OST object ID */
1513 __u32 l_ost_gen; /* generation of this l_ost_idx */
1514 __u32 l_ost_idx; /* OST index in LOV (lov_tgt_desc->tgts) */
1517 #define lov_mds_md lov_mds_md_v1
1518 struct lov_mds_md_v1 { /* LOV EA mds/wire data (little-endian) */
1519 __u32 lmm_magic; /* magic number = LOV_MAGIC_V1 */
1520 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1521 struct ost_id lmm_oi; /* LOV object ID */
1522 __u32 lmm_stripe_size; /* size of stripe in bytes */
1523 /* lmm_stripe_count used to be __u32 */
1524 __u16 lmm_stripe_count; /* num stripes in use for this object */
1525 __u16 lmm_layout_gen; /* layout generation number */
1526 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1530 * Sigh, because pre-2.4 uses
1531 * struct lov_mds_md_v1 {
1533 * __u64 lmm_object_id;
1534 * __u64 lmm_object_seq;
1537 * to identify the LOV(MDT) object, and lmm_object_seq will
1538 * be normal_fid, which make it hard to combine these conversion
1539 * to ostid_to FID. so we will do lmm_oi/fid conversion separately
1541 * We can tell the lmm_oi by this way,
1542 * 1.8: lmm_object_id = {inode}, lmm_object_gr = 0
1543 * 2.1: lmm_object_id = {oid < 128k}, lmm_object_seq = FID_SEQ_NORMAL
1544 * 2.4: lmm_oi.f_seq = FID_SEQ_NORMAL, lmm_oi.f_oid = {oid < 128k},
1547 * But currently lmm_oi/lsm_oi does not have any "real" usages,
1548 * except for printing some information, and the user can always
1549 * get the real FID from LMA, besides this multiple case check might
1550 * make swab more complicate. So we will keep using id/seq for lmm_oi.
1553 static inline void fid_to_lmm_oi(const struct lu_fid *fid,
1556 oi->oi.oi_id = fid_oid(fid);
1557 oi->oi.oi_seq = fid_seq(fid);
1560 static inline void lmm_oi_set_seq(struct ost_id *oi, __u64 seq)
1562 oi->oi.oi_seq = seq;
1565 static inline void lmm_oi_set_id(struct ost_id *oi, __u64 oid)
1570 static inline __u64 lmm_oi_id(const struct ost_id *oi)
1572 return oi->oi.oi_id;
1575 static inline __u64 lmm_oi_seq(const struct ost_id *oi)
1577 return oi->oi.oi_seq;
1580 static inline void lmm_oi_le_to_cpu(struct ost_id *dst_oi,
1581 const struct ost_id *src_oi)
1583 dst_oi->oi.oi_id = le64_to_cpu(src_oi->oi.oi_id);
1584 dst_oi->oi.oi_seq = le64_to_cpu(src_oi->oi.oi_seq);
1587 static inline void lmm_oi_cpu_to_le(struct ost_id *dst_oi,
1588 const struct ost_id *src_oi)
1590 dst_oi->oi.oi_id = cpu_to_le64(src_oi->oi.oi_id);
1591 dst_oi->oi.oi_seq = cpu_to_le64(src_oi->oi.oi_seq);
1594 /* extern void lustre_swab_lov_mds_md(struct lov_mds_md *llm); */
1596 #define MAX_MD_SIZE (sizeof(struct lov_mds_md) + 4 * sizeof(struct lov_ost_data))
1597 #define MIN_MD_SIZE (sizeof(struct lov_mds_md) + 1 * sizeof(struct lov_ost_data))
1599 /* This is the default MDT reply size allocated, should the striping be bigger,
1600 * it will be reallocated in mdt_fix_reply.
1601 * 100 stripes is a bit less than 2.5k of data */
1602 #define DEF_REP_MD_SIZE (sizeof(struct lov_mds_md) + \
1603 100 * sizeof(struct lov_ost_data))
1605 #define XATTR_NAME_ACL_ACCESS "system.posix_acl_access"
1606 #define XATTR_NAME_ACL_DEFAULT "system.posix_acl_default"
1607 #define XATTR_USER_PREFIX "user."
1608 #define XATTR_TRUSTED_PREFIX "trusted."
1609 #define XATTR_SECURITY_PREFIX "security."
1611 #define XATTR_NAME_LOV "trusted.lov"
1612 #define XATTR_NAME_LMA "trusted.lma"
1613 #define XATTR_NAME_LMV "trusted.lmv"
1614 #define XATTR_NAME_DEFAULT_LMV "trusted.dmv"
1615 #define XATTR_NAME_LINK "trusted.link"
1616 #define XATTR_NAME_FID "trusted.fid"
1617 #define XATTR_NAME_VERSION "trusted.version"
1618 #define XATTR_NAME_SOM "trusted.som"
1619 #define XATTR_NAME_HSM "trusted.hsm"
1620 #define XATTR_NAME_LFSCK_BITMAP "trusted.lfsck_bitmap"
1621 #define XATTR_NAME_DUMMY "trusted.dummy"
1623 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
1624 # define XATTR_NAME_LFSCK_NAMESPACE_OLD "trusted.lfsck_namespace"
1627 #define XATTR_NAME_LFSCK_NAMESPACE "trusted.lfsck_ns"
1628 #define XATTR_NAME_MAX_LEN 32 /* increase this, if there is longer name. */
1630 struct lov_mds_md_v3 { /* LOV EA mds/wire data (little-endian) */
1631 __u32 lmm_magic; /* magic number = LOV_MAGIC_V3 */
1632 __u32 lmm_pattern; /* LOV_PATTERN_RAID0, LOV_PATTERN_RAID1 */
1633 struct ost_id lmm_oi; /* LOV object ID */
1634 __u32 lmm_stripe_size; /* size of stripe in bytes */
1635 /* lmm_stripe_count used to be __u32 */
1636 __u16 lmm_stripe_count; /* num stripes in use for this object */
1637 __u16 lmm_layout_gen; /* layout generation number */
1638 char lmm_pool_name[LOV_MAXPOOLNAME + 1]; /* must be 32bit aligned */
1639 struct lov_ost_data_v1 lmm_objects[0]; /* per-stripe data */
1642 static inline __u32 lov_mds_md_size(__u16 stripes, __u32 lmm_magic)
1644 if (lmm_magic == LOV_MAGIC_V3)
1645 return sizeof(struct lov_mds_md_v3) +
1646 stripes * sizeof(struct lov_ost_data_v1);
1648 return sizeof(struct lov_mds_md_v1) +
1649 stripes * sizeof(struct lov_ost_data_v1);
1653 lov_mds_md_max_stripe_count(size_t buf_size, __u32 lmm_magic)
1655 switch (lmm_magic) {
1656 case LOV_MAGIC_V1: {
1657 struct lov_mds_md_v1 lmm;
1659 if (buf_size < sizeof(lmm))
1662 return (buf_size - sizeof(lmm)) / sizeof(lmm.lmm_objects[0]);
1664 case LOV_MAGIC_V3: {
1665 struct lov_mds_md_v3 lmm;
1667 if (buf_size < sizeof(lmm))
1670 return (buf_size - sizeof(lmm)) / sizeof(lmm.lmm_objects[0]);
1677 #define OBD_MD_FLID (0x00000001ULL) /* object ID */
1678 #define OBD_MD_FLATIME (0x00000002ULL) /* access time */
1679 #define OBD_MD_FLMTIME (0x00000004ULL) /* data modification time */
1680 #define OBD_MD_FLCTIME (0x00000008ULL) /* change time */
1681 #define OBD_MD_FLSIZE (0x00000010ULL) /* size */
1682 #define OBD_MD_FLBLOCKS (0x00000020ULL) /* allocated blocks count */
1683 #define OBD_MD_FLBLKSZ (0x00000040ULL) /* block size */
1684 #define OBD_MD_FLMODE (0x00000080ULL) /* access bits (mode & ~S_IFMT) */
1685 #define OBD_MD_FLTYPE (0x00000100ULL) /* object type (mode & S_IFMT) */
1686 #define OBD_MD_FLUID (0x00000200ULL) /* user ID */
1687 #define OBD_MD_FLGID (0x00000400ULL) /* group ID */
1688 #define OBD_MD_FLFLAGS (0x00000800ULL) /* flags word */
1689 #define OBD_MD_FLNLINK (0x00002000ULL) /* link count */
1690 #define OBD_MD_FLGENER (0x00004000ULL) /* generation number */
1691 /*#define OBD_MD_FLINLINE (0x00008000ULL) inline data. used until 1.6.5 */
1692 #define OBD_MD_FLRDEV (0x00010000ULL) /* device number */
1693 #define OBD_MD_FLEASIZE (0x00020000ULL) /* extended attribute data */
1694 #define OBD_MD_LINKNAME (0x00040000ULL) /* symbolic link target */
1695 #define OBD_MD_FLHANDLE (0x00080000ULL) /* file/lock handle */
1696 #define OBD_MD_FLCKSUM (0x00100000ULL) /* bulk data checksum */
1697 #define OBD_MD_FLQOS (0x00200000ULL) /* quality of service stats */
1698 /*#define OBD_MD_FLOSCOPQ (0x00400000ULL) osc opaque data, never used */
1699 /* OBD_MD_FLCOOKIE (0x00800000ULL) obsolete in 2.8 */
1700 #define OBD_MD_FLGROUP (0x01000000ULL) /* group */
1701 #define OBD_MD_FLFID (0x02000000ULL) /* ->ost write inline fid */
1702 #define OBD_MD_FLEPOCH (0x04000000ULL) /* ->ost write with ioepoch */
1703 /* ->mds if epoch opens or closes */
1704 #define OBD_MD_FLGRANT (0x08000000ULL) /* ost preallocation space grant */
1705 #define OBD_MD_FLDIREA (0x10000000ULL) /* dir's extended attribute data */
1706 #define OBD_MD_FLUSRQUOTA (0x20000000ULL) /* over quota flags sent from ost */
1707 #define OBD_MD_FLGRPQUOTA (0x40000000ULL) /* over quota flags sent from ost */
1708 #define OBD_MD_FLMODEASIZE (0x80000000ULL) /* EA size will be changed */
1710 #define OBD_MD_MDS (0x0000000100000000ULL) /* where an inode lives on */
1711 #define OBD_MD_REINT (0x0000000200000000ULL) /* reintegrate oa */
1712 #define OBD_MD_MEA (0x0000000400000000ULL) /* CMD split EA */
1713 #define OBD_MD_TSTATE (0x0000000800000000ULL) /* transient state field */
1715 #define OBD_MD_FLXATTR (0x0000001000000000ULL) /* xattr */
1716 #define OBD_MD_FLXATTRLS (0x0000002000000000ULL) /* xattr list */
1717 #define OBD_MD_FLXATTRRM (0x0000004000000000ULL) /* xattr remove */
1718 #define OBD_MD_FLACL (0x0000008000000000ULL) /* ACL */
1719 #define OBD_MD_FLRMTPERM (0x0000010000000000ULL) /* remote permission */
1720 #define OBD_MD_FLMDSCAPA (0x0000020000000000ULL) /* MDS capability */
1721 #define OBD_MD_FLOSSCAPA (0x0000040000000000ULL) /* OSS capability */
1722 #define OBD_MD_FLCKSPLIT (0x0000080000000000ULL) /* Check split on server */
1723 #define OBD_MD_FLCROSSREF (0x0000100000000000ULL) /* Cross-ref case */
1724 #define OBD_MD_FLGETATTRLOCK (0x0000200000000000ULL) /* Get IOEpoch attributes
1725 * under lock; for xattr
1726 * requests means the
1727 * client holds the lock */
1728 #define OBD_MD_FLOBJCOUNT (0x0000400000000000ULL) /* for multiple destroy */
1730 #define OBD_MD_FLRMTLSETFACL (0x0001000000000000ULL) /* lfs lsetfacl case */
1731 #define OBD_MD_FLRMTLGETFACL (0x0002000000000000ULL) /* lfs lgetfacl case */
1732 #define OBD_MD_FLRMTRSETFACL (0x0004000000000000ULL) /* lfs rsetfacl case */
1733 #define OBD_MD_FLRMTRGETFACL (0x0008000000000000ULL) /* lfs rgetfacl case */
1735 #define OBD_MD_FLDATAVERSION (0x0010000000000000ULL) /* iversion sum */
1736 #define OBD_MD_CLOSE_INTENT_EXECED (0x0020000000000000ULL) /* close intent
1739 #define OBD_MD_DEFAULT_MEA (0x0040000000000000ULL) /* default MEA */
1741 #define OBD_MD_FLGETATTR (OBD_MD_FLID | OBD_MD_FLATIME | OBD_MD_FLMTIME | \
1742 OBD_MD_FLCTIME | OBD_MD_FLSIZE | OBD_MD_FLBLKSZ | \
1743 OBD_MD_FLMODE | OBD_MD_FLTYPE | OBD_MD_FLUID | \
1744 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLNLINK | \
1745 OBD_MD_FLGENER | OBD_MD_FLRDEV | OBD_MD_FLGROUP)
1747 #define OBD_MD_FLXATTRALL (OBD_MD_FLXATTR | OBD_MD_FLXATTRLS)
1749 /* don't forget obdo_fid which is way down at the bottom so it can
1750 * come after the definition of llog_cookie */
1754 HSS_CLEARMASK = 0x02,
1755 HSS_ARCHIVE_ID = 0x04,
1758 struct hsm_state_set {
1760 __u32 hss_archive_id;
1762 __u64 hss_clearmask;
1765 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
1766 extern void lustre_swab_hsm_state_set(struct hsm_state_set *hss);
1768 extern void lustre_swab_obd_statfs (struct obd_statfs *os);
1770 /* ost_body.data values for OST_BRW */
1772 #define OBD_BRW_READ 0x01
1773 #define OBD_BRW_WRITE 0x02
1774 #define OBD_BRW_RWMASK (OBD_BRW_READ | OBD_BRW_WRITE)
1775 #define OBD_BRW_SYNC 0x08 /* this page is a part of synchronous
1776 * transfer and is not accounted in
1778 #define OBD_BRW_CHECK 0x10
1779 #define OBD_BRW_FROM_GRANT 0x20 /* the osc manages this under llite */
1780 #define OBD_BRW_GRANTED 0x40 /* the ost manages this */
1781 #define OBD_BRW_NOCACHE 0x80 /* this page is a part of non-cached IO */
1782 #define OBD_BRW_NOQUOTA 0x100
1783 #define OBD_BRW_SRVLOCK 0x200 /* Client holds no lock over this page */
1784 #define OBD_BRW_ASYNC 0x400 /* Server may delay commit to disk */
1785 #define OBD_BRW_MEMALLOC 0x800 /* Client runs in the "kswapd" context */
1786 #define OBD_BRW_OVER_USRQUOTA 0x1000 /* Running out of user quota */
1787 #define OBD_BRW_OVER_GRPQUOTA 0x2000 /* Running out of group quota */
1788 #define OBD_BRW_SOFT_SYNC 0x4000 /* This flag notifies the server
1789 * that the client is running low on
1790 * space for unstable pages; asking
1791 * it to sync quickly */
1793 #define OBD_OBJECT_EOF LUSTRE_EOF
1795 #define OST_MIN_PRECREATE 32
1796 #define OST_MAX_PRECREATE 20000
1799 struct ost_id ioo_oid; /* object ID, if multi-obj BRW */
1800 __u32 ioo_max_brw; /* low 16 bits were o_mode before 2.4,
1801 * now (PTLRPC_BULK_OPS_COUNT - 1) in
1802 * high 16 bits in 2.4 and later */
1803 __u32 ioo_bufcnt; /* number of niobufs for this object */
1806 #define IOOBJ_MAX_BRW_BITS 16
1807 #define IOOBJ_TYPE_MASK ((1U << IOOBJ_MAX_BRW_BITS) - 1)
1808 #define ioobj_max_brw_get(ioo) (((ioo)->ioo_max_brw >> IOOBJ_MAX_BRW_BITS) + 1)
1809 #define ioobj_max_brw_set(ioo, num) \
1810 do { (ioo)->ioo_max_brw = ((num) - 1) << IOOBJ_MAX_BRW_BITS; } while (0)
1812 extern void lustre_swab_obd_ioobj (struct obd_ioobj *ioo);
1814 /* multiple of 8 bytes => can array */
1815 struct niobuf_remote {
1821 void lustre_swab_niobuf_remote(struct niobuf_remote *nbr);
1823 /* lock value block communicated between the filter and llite */
1825 /* OST_LVB_ERR_INIT is needed because the return code in rc is
1826 * negative, i.e. because ((MASK + rc) & MASK) != MASK. */
1827 #define OST_LVB_ERR_INIT 0xffbadbad80000000ULL
1828 #define OST_LVB_ERR_MASK 0xffbadbad00000000ULL
1829 #define OST_LVB_IS_ERR(blocks) \
1830 ((blocks & OST_LVB_ERR_MASK) == OST_LVB_ERR_MASK)
1831 #define OST_LVB_SET_ERR(blocks, rc) \
1832 do { blocks = OST_LVB_ERR_INIT + rc; } while (0)
1833 #define OST_LVB_GET_ERR(blocks) (int)(blocks - OST_LVB_ERR_INIT)
1843 extern void lustre_swab_ost_lvb_v1(struct ost_lvb_v1 *lvb);
1857 extern void lustre_swab_ost_lvb(struct ost_lvb *lvb);
1860 * lquota data structures
1863 #ifndef QUOTABLOCK_BITS
1864 # define QUOTABLOCK_BITS LUSTRE_QUOTABLOCK_BITS
1867 #ifndef QUOTABLOCK_SIZE
1868 # define QUOTABLOCK_SIZE LUSTRE_QUOTABLOCK_SIZE
1872 # define toqb lustre_stoqb
1875 /* The lquota_id structure is an union of all the possible identifier types that
1876 * can be used with quota, this includes:
1879 * - a FID which can be used for per-directory quota in the future */
1881 struct lu_fid qid_fid; /* FID for per-directory quota */
1882 __u64 qid_uid; /* user identifier */
1883 __u64 qid_gid; /* group identifier */
1886 /* quotactl management */
1887 struct obd_quotactl {
1889 __u32 qc_type; /* see Q_* flag below */
1892 struct obd_dqinfo qc_dqinfo;
1893 struct obd_dqblk qc_dqblk;
1896 extern void lustre_swab_obd_quotactl(struct obd_quotactl *q);
1898 #define Q_COPY(out, in, member) (out)->member = (in)->member
1900 #define QCTL_COPY(out, in) \
1902 Q_COPY(out, in, qc_cmd); \
1903 Q_COPY(out, in, qc_type); \
1904 Q_COPY(out, in, qc_id); \
1905 Q_COPY(out, in, qc_stat); \
1906 Q_COPY(out, in, qc_dqinfo); \
1907 Q_COPY(out, in, qc_dqblk); \
1910 /* Body of quota request used for quota acquire/release RPCs between quota
1911 * master (aka QMT) and slaves (ak QSD). */
1913 struct lu_fid qb_fid; /* FID of global index packing the pool ID
1914 * and type (data or metadata) as well as
1915 * the quota type (user or group). */
1916 union lquota_id qb_id; /* uid or gid or directory FID */
1917 __u32 qb_flags; /* see below */
1919 __u64 qb_count; /* acquire/release count (kbytes/inodes) */
1920 __u64 qb_usage; /* current slave usage (kbytes/inodes) */
1921 __u64 qb_slv_ver; /* slave index file version */
1922 struct lustre_handle qb_lockh; /* per-ID lock handle */
1923 struct lustre_handle qb_glb_lockh; /* global lock handle */
1924 __u64 qb_padding1[4];
1927 /* When the quota_body is used in the reply of quota global intent
1928 * lock (IT_QUOTA_CONN) reply, qb_fid contains slave index file FID. */
1929 #define qb_slv_fid qb_fid
1930 /* qb_usage is the current qunit (in kbytes/inodes) when quota_body is used in
1932 #define qb_qunit qb_usage
1934 #define QUOTA_DQACQ_FL_ACQ 0x1 /* acquire quota */
1935 #define QUOTA_DQACQ_FL_PREACQ 0x2 /* pre-acquire */
1936 #define QUOTA_DQACQ_FL_REL 0x4 /* release quota */
1937 #define QUOTA_DQACQ_FL_REPORT 0x8 /* report usage */
1939 extern void lustre_swab_quota_body(struct quota_body *b);
1941 /* Quota types currently supported */
1943 LQUOTA_TYPE_USR = 0x00, /* maps to USRQUOTA */
1944 LQUOTA_TYPE_GRP = 0x01, /* maps to GRPQUOTA */
1948 /* There are 2 different resource types on which a quota limit can be enforced:
1949 * - inodes on the MDTs
1950 * - blocks on the OSTs */
1952 LQUOTA_RES_MD = 0x01, /* skip 0 to avoid null oid in FID */
1953 LQUOTA_RES_DT = 0x02,
1955 LQUOTA_FIRST_RES = LQUOTA_RES_MD
1957 #define LQUOTA_NR_RES (LQUOTA_LAST_RES - LQUOTA_FIRST_RES + 1)
1960 * Space accounting support
1961 * Format of an accounting record, providing disk usage information for a given
1964 struct lquota_acct_rec { /* 16 bytes */
1965 __u64 bspace; /* current space in use */
1966 __u64 ispace; /* current # inodes in use */
1970 * Global quota index support
1971 * Format of a global record, providing global quota settings for a given quota
1974 struct lquota_glb_rec { /* 32 bytes */
1975 __u64 qbr_hardlimit; /* quota hard limit, in #inodes or kbytes */
1976 __u64 qbr_softlimit; /* quota soft limit, in #inodes or kbytes */
1977 __u64 qbr_time; /* grace time, in seconds */
1978 __u64 qbr_granted; /* how much is granted to slaves, in #inodes or
1983 * Slave index support
1984 * Format of a slave record, recording how much space is granted to a given
1987 struct lquota_slv_rec { /* 8 bytes */
1988 __u64 qsr_granted; /* space granted to the slave for the key=ID,
1989 * in #inodes or kbytes */
1992 /* Data structures associated with the quota locks */
1994 /* Glimpse descriptor used for the index & per-ID quota locks */
1995 struct ldlm_gl_lquota_desc {
1996 union lquota_id gl_id; /* quota ID subject to the glimpse */
1997 __u64 gl_flags; /* see LQUOTA_FL* below */
1998 __u64 gl_ver; /* new index version */
1999 __u64 gl_hardlimit; /* new hardlimit or qunit value */
2000 __u64 gl_softlimit; /* new softlimit */
2004 #define gl_qunit gl_hardlimit /* current qunit value used when
2005 * glimpsing per-ID quota locks */
2007 /* quota glimpse flags */
2008 #define LQUOTA_FL_EDQUOT 0x1 /* user/group out of quota space on QMT */
2010 /* LVB used with quota (global and per-ID) locks */
2012 __u64 lvb_flags; /* see LQUOTA_FL* above */
2013 __u64 lvb_id_may_rel; /* space that might be released later */
2014 __u64 lvb_id_rel; /* space released by the slave for this ID */
2015 __u64 lvb_id_qunit; /* current qunit value */
2019 extern void lustre_swab_lquota_lvb(struct lquota_lvb *lvb);
2021 /* LVB used with global quota lock */
2022 #define lvb_glb_ver lvb_id_may_rel /* current version of the global index */
2030 #define QUOTA_FIRST_OPC QUOTA_DQACQ
2039 MDS_GETATTR_NAME = 34,
2044 MDS_DISCONNECT = 39,
2047 MDS_PIN = 42, /* obsolete, never used in a release */
2048 MDS_UNPIN = 43, /* obsolete, never used in a release */
2050 MDS_DONE_WRITING = 45, /* obsolete since 2.8.0 */
2052 MDS_QUOTACHECK = 47, /* not used since 2.4 */
2055 MDS_SETXATTR = 50, /* obsolete, now it's MDS_REINT op */
2057 MDS_IS_SUBDIR = 52, /* obsolete, never used in a release */
2059 MDS_HSM_STATE_GET = 54,
2060 MDS_HSM_STATE_SET = 55,
2061 MDS_HSM_ACTION = 56,
2062 MDS_HSM_PROGRESS = 57,
2063 MDS_HSM_REQUEST = 58,
2064 MDS_HSM_CT_REGISTER = 59,
2065 MDS_HSM_CT_UNREGISTER = 60,
2066 MDS_SWAP_LAYOUTS = 61,
2070 #define MDS_FIRST_OPC MDS_GETATTR
2073 /* opcodes for object update */
2079 #define OUT_UPDATE_FIRST_OPC OUT_UPDATE
2096 } mds_reint_t, mdt_reint_t;
2098 extern void lustre_swab_generic_32s (__u32 *val);
2100 /* the disposition of the intent outlines what was executed */
2101 #define DISP_IT_EXECD 0x00000001
2102 #define DISP_LOOKUP_EXECD 0x00000002
2103 #define DISP_LOOKUP_NEG 0x00000004
2104 #define DISP_LOOKUP_POS 0x00000008
2105 #define DISP_OPEN_CREATE 0x00000010
2106 #define DISP_OPEN_OPEN 0x00000020
2107 #define DISP_ENQ_COMPLETE 0x00400000 /* obsolete and unused */
2108 #define DISP_ENQ_OPEN_REF 0x00800000
2109 #define DISP_ENQ_CREATE_REF 0x01000000
2110 #define DISP_OPEN_LOCK 0x02000000
2111 #define DISP_OPEN_LEASE 0x04000000
2112 #define DISP_OPEN_STRIPE 0x08000000
2113 #define DISP_OPEN_DENY 0x10000000
2115 /* INODE LOCK PARTS */
2116 #define MDS_INODELOCK_LOOKUP 0x000001 /* For namespace, dentry etc, and also
2117 * was used to protect permission (mode,
2118 * owner, group etc) before 2.4. */
2119 #define MDS_INODELOCK_UPDATE 0x000002 /* size, links, timestamps */
2120 #define MDS_INODELOCK_OPEN 0x000004 /* For opened files */
2121 #define MDS_INODELOCK_LAYOUT 0x000008 /* for layout */
2123 /* The PERM bit is added int 2.4, and it is used to protect permission(mode,
2124 * owner, group, acl etc), so to separate the permission from LOOKUP lock.
2125 * Because for remote directories(in DNE), these locks will be granted by
2126 * different MDTs(different ldlm namespace).
2128 * For local directory, MDT will always grant UPDATE_LOCK|PERM_LOCK together.
2129 * For Remote directory, the master MDT, where the remote directory is, will
2130 * grant UPDATE_LOCK|PERM_LOCK, and the remote MDT, where the name entry is,
2131 * will grant LOOKUP_LOCK. */
2132 #define MDS_INODELOCK_PERM 0x000010
2133 #define MDS_INODELOCK_XATTR 0x000020 /* extended attributes */
2135 #define MDS_INODELOCK_MAXSHIFT 5
2136 /* This FULL lock is useful to take on unlink sort of operations */
2137 #define MDS_INODELOCK_FULL ((1<<(MDS_INODELOCK_MAXSHIFT+1))-1)
2139 extern void lustre_swab_ll_fid (struct ll_fid *fid);
2141 /* NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
2142 * but was moved into name[1] along with the OID to avoid consuming the
2143 * name[2,3] fields that need to be used for the quota id (also a FID). */
2145 LUSTRE_RES_ID_SEQ_OFF = 0,
2146 LUSTRE_RES_ID_VER_OID_OFF = 1,
2147 LUSTRE_RES_ID_WAS_VER_OFF = 2, /* see note above */
2148 LUSTRE_RES_ID_QUOTA_SEQ_OFF = 2,
2149 LUSTRE_RES_ID_QUOTA_VER_OID_OFF = 3,
2150 LUSTRE_RES_ID_HSH_OFF = 3
2153 #define MDS_STATUS_CONN 1
2154 #define MDS_STATUS_LOV 2
2156 /* these should be identical to their EXT4_*_FL counterparts, they are
2157 * redefined here only to avoid dragging in fs/ext4/ext4.h */
2158 #define LUSTRE_SYNC_FL 0x00000008 /* Synchronous updates */
2159 #define LUSTRE_IMMUTABLE_FL 0x00000010 /* Immutable file */
2160 #define LUSTRE_APPEND_FL 0x00000020 /* writes to file may only append */
2161 #define LUSTRE_NODUMP_FL 0x00000040 /* do not dump file */
2162 #define LUSTRE_NOATIME_FL 0x00000080 /* do not update atime */
2163 #define LUSTRE_INDEX_FL 0x00001000 /* hash-indexed directory */
2164 #define LUSTRE_DIRSYNC_FL 0x00010000 /* dirsync behaviour (dir only) */
2165 #define LUSTRE_TOPDIR_FL 0x00020000 /* Top of directory hierarchies*/
2166 #define LUSTRE_DIRECTIO_FL 0x00100000 /* Use direct i/o */
2167 #define LUSTRE_INLINE_DATA_FL 0x10000000 /* Inode has inline data. */
2170 /* Convert wire LUSTRE_*_FL to corresponding client local VFS S_* values
2171 * for the client inode i_flags. The LUSTRE_*_FL are the Lustre wire
2172 * protocol equivalents of LDISKFS_*_FL values stored on disk, while
2173 * the S_* flags are kernel-internal values that change between kernel
2174 * versions. These flags are set/cleared via FSFILT_IOC_{GET,SET}_FLAGS.
2175 * See b=16526 for a full history. */
2176 static inline int ll_ext_to_inode_flags(int flags)
2178 return (((flags & LUSTRE_SYNC_FL) ? S_SYNC : 0) |
2179 ((flags & LUSTRE_NOATIME_FL) ? S_NOATIME : 0) |
2180 ((flags & LUSTRE_APPEND_FL) ? S_APPEND : 0) |
2181 #if defined(S_DIRSYNC)
2182 ((flags & LUSTRE_DIRSYNC_FL) ? S_DIRSYNC : 0) |
2184 ((flags & LUSTRE_IMMUTABLE_FL) ? S_IMMUTABLE : 0));
2187 static inline int ll_inode_to_ext_flags(int iflags)
2189 return (((iflags & S_SYNC) ? LUSTRE_SYNC_FL : 0) |
2190 ((iflags & S_NOATIME) ? LUSTRE_NOATIME_FL : 0) |
2191 ((iflags & S_APPEND) ? LUSTRE_APPEND_FL : 0) |
2192 #if defined(S_DIRSYNC)
2193 ((iflags & S_DIRSYNC) ? LUSTRE_DIRSYNC_FL : 0) |
2195 ((iflags & S_IMMUTABLE) ? LUSTRE_IMMUTABLE_FL : 0));
2199 /* 64 possible states */
2200 enum md_transient_state {
2201 MS_RESTORE = (1 << 0), /* restore is running */
2205 struct lu_fid mbo_fid1;
2206 struct lu_fid mbo_fid2;
2207 struct lustre_handle mbo_handle;
2209 __u64 mbo_size; /* Offset, in the case of MDS_READPAGE */
2213 __u64 mbo_blocks; /* XID, in the case of MDS_READPAGE */
2215 __u64 mbo_t_state; /* transient file state defined in
2216 * enum md_transient_state
2217 * was "ino" until 2.4.0 */
2220 __u32 mbo_capability;
2224 __u32 mbo_flags; /* LUSTRE_*_FL file attributes */
2226 __u32 mbo_nlink; /* #bytes to read in the case of MDS_READPAGE */
2227 __u32 mbo_unused2; /* was "generation" until 2.4.0 */
2229 __u32 mbo_eadatasize;
2231 __u32 mbo_max_mdsize;
2232 __u32 mbo_unused3; /* was max_cookiesize until 2.8 */
2233 __u32 mbo_uid_h; /* high 32-bits of uid, for FUID */
2234 __u32 mbo_gid_h; /* high 32-bits of gid, for FUID */
2235 __u32 mbo_padding_5; /* also fix lustre_swab_mdt_body */
2236 __u64 mbo_padding_6;
2237 __u64 mbo_padding_7;
2238 __u64 mbo_padding_8;
2239 __u64 mbo_padding_9;
2240 __u64 mbo_padding_10;
2243 extern void lustre_swab_mdt_body (struct mdt_body *b);
2245 struct mdt_ioepoch {
2246 struct lustre_handle mio_handle;
2247 __u64 mio_unused1; /* was ioepoch */
2248 __u32 mio_unused2; /* was flags */
2252 extern void lustre_swab_mdt_ioepoch (struct mdt_ioepoch *b);
2254 /* permissions for md_perm.mp_perm */
2256 CFS_SETUID_PERM = 0x01,
2257 CFS_SETGID_PERM = 0x02,
2258 CFS_SETGRP_PERM = 0x04,
2259 CFS_RMTACL_PERM = 0x08,
2260 CFS_RMTOWN_PERM = 0x10
2263 /* inode access permission for remote user, the inode info are omitted,
2264 * for client knows them. */
2265 struct mdt_remote_perm {
2272 __u32 rp_access_perm; /* MAY_READ/WRITE/EXEC */
2276 extern void lustre_swab_mdt_remote_perm(struct mdt_remote_perm *p);
2278 struct mdt_rec_setattr {
2288 __u32 sa_padding_1_h;
2289 struct lu_fid sa_fid;
2298 __u32 sa_attr_flags;
2300 __u32 sa_bias; /* some operation flags */
2306 extern void lustre_swab_mdt_rec_setattr (struct mdt_rec_setattr *sa);
2309 * Attribute flags used in mdt_rec_setattr::sa_valid.
2310 * The kernel's #defines for ATTR_* should not be used over the network
2311 * since the client and MDS may run different kernels (see bug 13828)
2312 * Therefore, we should only use MDS_ATTR_* attributes for sa_valid.
2314 #define MDS_ATTR_MODE 0x1ULL /* = 1 */
2315 #define MDS_ATTR_UID 0x2ULL /* = 2 */
2316 #define MDS_ATTR_GID 0x4ULL /* = 4 */
2317 #define MDS_ATTR_SIZE 0x8ULL /* = 8 */
2318 #define MDS_ATTR_ATIME 0x10ULL /* = 16 */
2319 #define MDS_ATTR_MTIME 0x20ULL /* = 32 */
2320 #define MDS_ATTR_CTIME 0x40ULL /* = 64 */
2321 #define MDS_ATTR_ATIME_SET 0x80ULL /* = 128 */
2322 #define MDS_ATTR_MTIME_SET 0x100ULL /* = 256 */
2323 #define MDS_ATTR_FORCE 0x200ULL /* = 512, Not a change, but a change it */
2324 #define MDS_ATTR_ATTR_FLAG 0x400ULL /* = 1024 */
2325 #define MDS_ATTR_KILL_SUID 0x800ULL /* = 2048 */
2326 #define MDS_ATTR_KILL_SGID 0x1000ULL /* = 4096 */
2327 #define MDS_ATTR_CTIME_SET 0x2000ULL /* = 8192 */
2328 #define MDS_ATTR_FROM_OPEN 0x4000ULL /* = 16384, called from open path, ie O_TRUNC */
2329 #define MDS_ATTR_BLOCKS 0x8000ULL /* = 32768 */
2332 #define FMODE_READ 00000001
2333 #define FMODE_WRITE 00000002
2336 #define MDS_FMODE_CLOSED 00000000
2337 #define MDS_FMODE_EXEC 00000004
2338 /* MDS_FMODE_EPOCH 01000000 obsolete since 2.8.0 */
2339 /* MDS_FMODE_TRUNC 02000000 obsolete since 2.8.0 */
2340 /* MDS_FMODE_SOM 04000000 obsolete since 2.8.0 */
2342 #define MDS_OPEN_CREATED 00000010
2343 #define MDS_OPEN_CROSS 00000020
2345 #define MDS_OPEN_CREAT 00000100
2346 #define MDS_OPEN_EXCL 00000200
2347 #define MDS_OPEN_TRUNC 00001000
2348 #define MDS_OPEN_APPEND 00002000
2349 #define MDS_OPEN_SYNC 00010000
2350 #define MDS_OPEN_DIRECTORY 00200000
2352 #define MDS_OPEN_BY_FID 040000000 /* open_by_fid for known object */
2353 #define MDS_OPEN_DELAY_CREATE 0100000000 /* delay initial object create */
2354 #define MDS_OPEN_OWNEROVERRIDE 0200000000 /* NFSD rw-reopen ro file for owner */
2355 #define MDS_OPEN_JOIN_FILE 0400000000 /* open for join file.
2356 * We do not support JOIN FILE
2357 * anymore, reserve this flags
2358 * just for preventing such bit
2361 #define MDS_OPEN_LOCK 04000000000 /* This open requires open lock */
2362 #define MDS_OPEN_HAS_EA 010000000000 /* specify object create pattern */
2363 #define MDS_OPEN_HAS_OBJS 020000000000 /* Just set the EA the obj exist */
2364 #define MDS_OPEN_NORESTORE 0100000000000ULL /* Do not restore file at open */
2365 #define MDS_OPEN_NEWSTRIPE 0200000000000ULL /* New stripe needed (restripe or
2367 #define MDS_OPEN_VOLATILE 0400000000000ULL /* File is volatile = created
2369 #define MDS_OPEN_LEASE 01000000000000ULL /* Open the file and grant lease
2370 * delegation, succeed if it's not
2371 * being opened with conflict mode.
2373 #define MDS_OPEN_RELEASE 02000000000000ULL /* Open the file for HSM release */
2375 /* lustre internal open flags, which should not be set from user space */
2376 #define MDS_OPEN_FL_INTERNAL (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS | \
2377 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK | \
2378 MDS_OPEN_BY_FID | MDS_OPEN_LEASE | \
2382 MDS_CHECK_SPLIT = 1 << 0,
2383 MDS_CROSS_REF = 1 << 1,
2384 MDS_VTX_BYPASS = 1 << 2,
2385 MDS_PERM_BYPASS = 1 << 3,
2386 /* MDS_SOM = 1 << 4, obsolete since 2.8.0 */
2387 MDS_QUOTA_IGNORE = 1 << 5,
2388 /* Was MDS_CLOSE_CLEANUP (1 << 6), No more used */
2389 MDS_KEEP_ORPHAN = 1 << 7,
2390 MDS_RECOV_OPEN = 1 << 8,
2391 MDS_DATA_MODIFIED = 1 << 9,
2392 MDS_CREATE_VOLATILE = 1 << 10,
2393 MDS_OWNEROVERRIDE = 1 << 11,
2394 MDS_HSM_RELEASE = 1 << 12,
2395 MDS_RENAME_MIGRATE = 1 << 13,
2396 MDS_CLOSE_LAYOUT_SWAP = 1 << 14,
2399 /* instance of mdt_reint_rec */
2400 struct mdt_rec_create {
2408 __u32 cr_suppgid1_h;
2410 __u32 cr_suppgid2_h;
2411 struct lu_fid cr_fid1;
2412 struct lu_fid cr_fid2;
2413 struct lustre_handle cr_old_handle; /* handle in case of open replay */
2417 __u64 cr_padding_1; /* rr_blocks */
2420 /* use of helpers set/get_mrc_cr_flags() is needed to access
2421 * 64 bits cr_flags [cr_flags_l, cr_flags_h], this is done to
2422 * extend cr_flags size without breaking 1.8 compat */
2423 __u32 cr_flags_l; /* for use with open, low 32 bits */
2424 __u32 cr_flags_h; /* for use with open, high 32 bits */
2425 __u32 cr_umask; /* umask for create */
2426 __u32 cr_padding_4; /* rr_padding_4 */
2429 static inline void set_mrc_cr_flags(struct mdt_rec_create *mrc, __u64 flags)
2431 mrc->cr_flags_l = (__u32)(flags & 0xFFFFFFFFUll);
2432 mrc->cr_flags_h = (__u32)(flags >> 32);
2435 static inline __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
2437 return ((__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32));
2440 /* instance of mdt_reint_rec */
2441 struct mdt_rec_link {
2449 __u32 lk_suppgid1_h;
2451 __u32 lk_suppgid2_h;
2452 struct lu_fid lk_fid1;
2453 struct lu_fid lk_fid2;
2455 __u64 lk_padding_1; /* rr_atime */
2456 __u64 lk_padding_2; /* rr_ctime */
2457 __u64 lk_padding_3; /* rr_size */
2458 __u64 lk_padding_4; /* rr_blocks */
2460 __u32 lk_padding_5; /* rr_mode */
2461 __u32 lk_padding_6; /* rr_flags */
2462 __u32 lk_padding_7; /* rr_padding_2 */
2463 __u32 lk_padding_8; /* rr_padding_3 */
2464 __u32 lk_padding_9; /* rr_padding_4 */
2467 /* instance of mdt_reint_rec */
2468 struct mdt_rec_unlink {
2476 __u32 ul_suppgid1_h;
2478 __u32 ul_suppgid2_h;
2479 struct lu_fid ul_fid1;
2480 struct lu_fid ul_fid2;
2482 __u64 ul_padding_2; /* rr_atime */
2483 __u64 ul_padding_3; /* rr_ctime */
2484 __u64 ul_padding_4; /* rr_size */
2485 __u64 ul_padding_5; /* rr_blocks */
2488 __u32 ul_padding_6; /* rr_flags */
2489 __u32 ul_padding_7; /* rr_padding_2 */
2490 __u32 ul_padding_8; /* rr_padding_3 */
2491 __u32 ul_padding_9; /* rr_padding_4 */
2494 /* instance of mdt_reint_rec */
2495 struct mdt_rec_rename {
2503 __u32 rn_suppgid1_h;
2505 __u32 rn_suppgid2_h;
2506 struct lu_fid rn_fid1;
2507 struct lu_fid rn_fid2;
2509 __u64 rn_padding_1; /* rr_atime */
2510 __u64 rn_padding_2; /* rr_ctime */
2511 __u64 rn_padding_3; /* rr_size */
2512 __u64 rn_padding_4; /* rr_blocks */
2513 __u32 rn_bias; /* some operation flags */
2514 __u32 rn_mode; /* cross-ref rename has mode */
2515 __u32 rn_padding_5; /* rr_flags */
2516 __u32 rn_padding_6; /* rr_padding_2 */
2517 __u32 rn_padding_7; /* rr_padding_3 */
2518 __u32 rn_padding_8; /* rr_padding_4 */
2521 /* instance of mdt_reint_rec */
2522 struct mdt_rec_setxattr {
2530 __u32 sx_suppgid1_h;
2532 __u32 sx_suppgid2_h;
2533 struct lu_fid sx_fid;
2534 __u64 sx_padding_1; /* These three are rr_fid2 */
2539 __u64 sx_padding_5; /* rr_ctime */
2540 __u64 sx_padding_6; /* rr_size */
2541 __u64 sx_padding_7; /* rr_blocks */
2544 __u32 sx_padding_8; /* rr_flags */
2545 __u32 sx_padding_9; /* rr_padding_2 */
2546 __u32 sx_padding_10; /* rr_padding_3 */
2547 __u32 sx_padding_11; /* rr_padding_4 */
2551 * mdt_rec_reint is the template for all mdt_reint_xxx structures.
2552 * Do NOT change the size of various members, otherwise the value
2553 * will be broken in lustre_swab_mdt_rec_reint().
2555 * If you add new members in other mdt_reint_xxx structres and need to use the
2556 * rr_padding_x fields, then update lustre_swab_mdt_rec_reint() also.
2558 struct mdt_rec_reint {
2566 __u32 rr_suppgid1_h;
2568 __u32 rr_suppgid2_h;
2569 struct lu_fid rr_fid1;
2570 struct lu_fid rr_fid2;
2581 __u32 rr_padding_4; /* also fix lustre_swab_mdt_rec_reint */
2584 extern void lustre_swab_mdt_rec_reint(struct mdt_rec_reint *rr);
2586 /* lmv structures */
2588 __u32 ld_tgt_count; /* how many MDS's */
2589 __u32 ld_active_tgt_count; /* how many active */
2590 __u32 ld_default_stripe_count; /* how many objects are used */
2591 __u32 ld_pattern; /* default hash pattern */
2592 __u64 ld_default_hash_size;
2593 __u64 ld_padding_1; /* also fix lustre_swab_lmv_desc */
2594 __u32 ld_padding_2; /* also fix lustre_swab_lmv_desc */
2595 __u32 ld_qos_maxage; /* in second */
2596 __u32 ld_padding_3; /* also fix lustre_swab_lmv_desc */
2597 __u32 ld_padding_4; /* also fix lustre_swab_lmv_desc */
2598 struct obd_uuid ld_uuid;
2601 extern void lustre_swab_lmv_desc (struct lmv_desc *ld);
2603 /* LMV layout EA, and it will be stored both in master and slave object */
2604 struct lmv_mds_md_v1 {
2606 __u32 lmv_stripe_count;
2607 __u32 lmv_master_mdt_index; /* On master object, it is master
2608 * MDT index, on slave object, it
2609 * is stripe index of the slave obj */
2610 __u32 lmv_hash_type; /* dir stripe policy, i.e. indicate
2611 * which hash function to be used,
2612 * Note: only lower 16 bits is being
2613 * used for now. Higher 16 bits will
2614 * be used to mark the object status,
2615 * for example migrating or dead. */
2616 __u32 lmv_layout_version; /* Used for directory restriping */
2620 char lmv_pool_name[LOV_MAXPOOLNAME + 1]; /* pool name */
2621 struct lu_fid lmv_stripe_fids[0]; /* FIDs for each stripe */
2624 #define LMV_MAGIC_V1 0x0CD20CD0 /* normal stripe lmv magic */
2625 #define LMV_MAGIC LMV_MAGIC_V1
2627 /* #define LMV_USER_MAGIC 0x0CD30CD0 */
2628 #define LMV_MAGIC_STRIPE 0x0CD40CD0 /* magic for dir sub_stripe */
2630 /* Right now only the lower part(0-16bits) of lmv_hash_type is being used,
2631 * and the higher part will be the flag to indicate the status of object,
2632 * for example the object is being migrated. And the hash function
2633 * might be interpreted differently with different flags. */
2634 #define LMV_HASH_TYPE_MASK 0x0000ffff
2636 #define LMV_HASH_FLAG_MIGRATION 0x80000000
2637 #define LMV_HASH_FLAG_DEAD 0x40000000
2638 #define LMV_HASH_FLAG_BAD_TYPE 0x20000000
2640 /* The striped directory has ever lost its master LMV EA, then LFSCK
2641 * re-generated it. This flag is used to indicate such case. It is an
2643 #define LMV_HASH_FLAG_LOST_LMV 0x10000000
2646 * The FNV-1a hash algorithm is as follows:
2647 * hash = FNV_offset_basis
2648 * for each octet_of_data to be hashed
2649 * hash = hash XOR octet_of_data
2650 * hash = hash × FNV_prime
2652 * http://en.wikipedia.org/wiki/Fowler–Noll–Vo_hash_function#FNV-1a_hash
2654 * http://www.isthe.com/chongo/tech/comp/fnv/index.html#FNV-reference-source
2655 * FNV_prime is 2^40 + 2^8 + 0xb3 = 0x100000001b3ULL
2657 #define LUSTRE_FNV_1A_64_PRIME 0x100000001b3ULL
2658 #define LUSTRE_FNV_1A_64_OFFSET_BIAS 0xcbf29ce484222325ULL
2659 static inline __u64 lustre_hash_fnv_1a_64(const void *buf, size_t size)
2661 __u64 hash = LUSTRE_FNV_1A_64_OFFSET_BIAS;
2662 const unsigned char *p = buf;
2665 for (i = 0; i < size; i++) {
2667 hash *= LUSTRE_FNV_1A_64_PRIME;
2675 struct lmv_mds_md_v1 lmv_md_v1;
2676 struct lmv_user_md lmv_user_md;
2679 extern void lustre_swab_lmv_mds_md(union lmv_mds_md *lmm);
2681 static inline int lmv_mds_md_size(int stripe_count, unsigned int lmm_magic)
2683 switch (lmm_magic) {
2685 struct lmv_mds_md_v1 *lmm1;
2687 return sizeof(*lmm1) + stripe_count *
2688 sizeof(lmm1->lmv_stripe_fids[0]);
2695 static inline int lmv_mds_md_stripe_count_get(const union lmv_mds_md *lmm)
2697 switch (le32_to_cpu(lmm->lmv_magic)) {
2699 return le32_to_cpu(lmm->lmv_md_v1.lmv_stripe_count);
2700 case LMV_USER_MAGIC:
2701 return le32_to_cpu(lmm->lmv_user_md.lum_stripe_count);
2707 static inline int lmv_mds_md_stripe_count_set(union lmv_mds_md *lmm,
2708 unsigned int stripe_count)
2710 switch (le32_to_cpu(lmm->lmv_magic)) {
2712 lmm->lmv_md_v1.lmv_stripe_count = cpu_to_le32(stripe_count);
2714 case LMV_USER_MAGIC:
2715 lmm->lmv_user_md.lum_stripe_count = cpu_to_le32(stripe_count);
2727 FLD_FIRST_OPC = FLD_QUERY
2733 SEQ_FIRST_OPC = SEQ_QUERY
2737 SEQ_ALLOC_SUPER = 0,
2749 LFSCK_NOTIFY = 1101,
2752 LFSCK_FIRST_OPC = LFSCK_NOTIFY
2756 * LOV data structures
2759 #define LOV_MAX_UUID_BUFFER_SIZE 8192
2760 /* The size of the buffer the lov/mdc reserves for the
2761 * array of UUIDs returned by the MDS. With the current
2762 * protocol, this will limit the max number of OSTs per LOV */
2764 #define LOV_DESC_MAGIC 0xB0CCDE5C
2765 #define LOV_DESC_QOS_MAXAGE_DEFAULT 5 /* Seconds */
2766 #define LOV_DESC_STRIPE_SIZE_DEFAULT (1 << LNET_MTU_BITS)
2768 /* LOV settings descriptor (should only contain static info) */
2770 __u32 ld_tgt_count; /* how many OBD's */
2771 __u32 ld_active_tgt_count; /* how many active */
2772 __u32 ld_default_stripe_count; /* how many objects are used */
2773 __u32 ld_pattern; /* default PATTERN_RAID0 */
2774 __u64 ld_default_stripe_size; /* in bytes */
2775 __u64 ld_default_stripe_offset; /* in bytes */
2776 __u32 ld_padding_0; /* unused */
2777 __u32 ld_qos_maxage; /* in second */
2778 __u32 ld_padding_1; /* also fix lustre_swab_lov_desc */
2779 __u32 ld_padding_2; /* also fix lustre_swab_lov_desc */
2780 struct obd_uuid ld_uuid;
2783 #define ld_magic ld_active_tgt_count /* for swabbing from llogs */
2785 extern void lustre_swab_lov_desc (struct lov_desc *ld);
2790 /* opcodes -- MUST be distinct from OST/MDS opcodes */
2795 LDLM_BL_CALLBACK = 104,
2796 LDLM_CP_CALLBACK = 105,
2797 LDLM_GL_CALLBACK = 106,
2798 LDLM_SET_INFO = 107,
2801 #define LDLM_FIRST_OPC LDLM_ENQUEUE
2803 #define RES_NAME_SIZE 4
2804 struct ldlm_res_id {
2805 __u64 name[RES_NAME_SIZE];
2808 #define DLDLMRES "[%#llx:%#llx:%#llx].%#llx"
2809 #define PLDLMRES(res) (unsigned long long)(res)->lr_name.name[0], \
2810 (unsigned long long)(res)->lr_name.name[1], \
2811 (unsigned long long)(res)->lr_name.name[2], \
2812 (unsigned long long)(res)->lr_name.name[3]
2814 extern void lustre_swab_ldlm_res_id (struct ldlm_res_id *id);
2816 static inline bool ldlm_res_eq(const struct ldlm_res_id *res0,
2817 const struct ldlm_res_id *res1)
2819 return memcmp(res0, res1, sizeof(*res0)) == 0;
2823 typedef enum ldlm_mode {
2836 #define LCK_MODE_NUM 8
2838 typedef enum ldlm_type {
2846 #define LDLM_MIN_TYPE LDLM_PLAIN
2848 struct ldlm_extent {
2854 static inline int ldlm_extent_overlap(const struct ldlm_extent *ex1,
2855 const struct ldlm_extent *ex2)
2857 return ex1->start <= ex2->end && ex2->start <= ex1->end;
2860 /* check if @ex1 contains @ex2 */
2861 static inline int ldlm_extent_contain(const struct ldlm_extent *ex1,
2862 const struct ldlm_extent *ex2)
2864 return ex1->start <= ex2->start && ex1->end >= ex2->end;
2867 struct ldlm_inodebits {
2871 struct ldlm_flock_wire {
2879 /* it's important that the fields of the ldlm_extent structure match
2880 * the first fields of the ldlm_flock structure because there is only
2881 * one ldlm_swab routine to process the ldlm_policy_data_t union. if
2882 * this ever changes we will need to swab the union differently based
2883 * on the resource type. */
2885 typedef union ldlm_wire_policy_data {
2886 struct ldlm_extent l_extent;
2887 struct ldlm_flock_wire l_flock;
2888 struct ldlm_inodebits l_inodebits;
2889 } ldlm_wire_policy_data_t;
2891 extern void lustre_swab_ldlm_policy_data(union ldlm_wire_policy_data *d);
2893 union ldlm_gl_desc {
2894 struct ldlm_gl_lquota_desc lquota_desc;
2897 extern void lustre_swab_gl_desc(union ldlm_gl_desc *);
2899 enum ldlm_intent_flags {
2900 IT_OPEN = 0x00000001,
2901 IT_CREAT = 0x00000002,
2902 IT_OPEN_CREAT = 0x00000003,
2903 IT_READDIR = 0x00000004,
2904 IT_GETATTR = 0x00000008,
2905 IT_LOOKUP = 0x00000010,
2906 IT_UNLINK = 0x00000020,
2907 IT_TRUNC = 0x00000040,
2908 IT_GETXATTR = 0x00000080,
2909 IT_EXEC = 0x00000100,
2910 IT_PIN = 0x00000200,
2911 IT_LAYOUT = 0x00000400,
2912 IT_QUOTA_DQACQ = 0x00000800,
2913 IT_QUOTA_CONN = 0x00001000,
2914 IT_SETXATTR = 0x00002000,
2917 struct ldlm_intent {
2921 extern void lustre_swab_ldlm_intent (struct ldlm_intent *i);
2923 struct ldlm_resource_desc {
2924 enum ldlm_type lr_type;
2925 __u32 lr_pad; /* also fix lustre_swab_ldlm_resource_desc */
2926 struct ldlm_res_id lr_name;
2929 extern void lustre_swab_ldlm_resource_desc (struct ldlm_resource_desc *r);
2931 struct ldlm_lock_desc {
2932 struct ldlm_resource_desc l_resource;
2933 enum ldlm_mode l_req_mode;
2934 enum ldlm_mode l_granted_mode;
2935 union ldlm_wire_policy_data l_policy_data;
2938 extern void lustre_swab_ldlm_lock_desc(struct ldlm_lock_desc *l);
2940 #define LDLM_LOCKREQ_HANDLES 2
2941 #define LDLM_ENQUEUE_CANCEL_OFF 1
2943 struct ldlm_request {
2946 struct ldlm_lock_desc lock_desc;
2947 struct lustre_handle lock_handle[LDLM_LOCKREQ_HANDLES];
2950 extern void lustre_swab_ldlm_request (struct ldlm_request *rq);
2952 /* If LDLM_ENQUEUE, 1 slot is already occupied, 1 is available.
2953 * Otherwise, 2 are available. */
2954 #define ldlm_request_bufsize(count,type) \
2956 int _avail = LDLM_LOCKREQ_HANDLES; \
2957 _avail -= (type == LDLM_ENQUEUE ? LDLM_ENQUEUE_CANCEL_OFF : 0); \
2958 sizeof(struct ldlm_request) + \
2959 (count > _avail ? count - _avail : 0) * \
2960 sizeof(struct lustre_handle); \
2965 __u32 lock_padding; /* also fix lustre_swab_ldlm_reply */
2966 struct ldlm_lock_desc lock_desc;
2967 struct lustre_handle lock_handle;
2968 __u64 lock_policy_res1;
2969 __u64 lock_policy_res2;
2972 extern void lustre_swab_ldlm_reply (struct ldlm_reply *r);
2974 #define ldlm_flags_to_wire(flags) ((__u32)(flags))
2975 #define ldlm_flags_from_wire(flags) ((__u64)(flags))
2978 * Opcodes for mountconf (mgs and mgc)
2983 MGS_EXCEPTION, /* node died, etc. */
2984 MGS_TARGET_REG, /* whenever target starts up */
2990 #define MGS_FIRST_OPC MGS_CONNECT
2992 #define MGS_PARAM_MAXLEN 1024
2993 #define KEY_SET_INFO "set_info"
2995 struct mgs_send_param {
2996 char mgs_param[MGS_PARAM_MAXLEN];
2999 /* We pass this info to the MGS so it can write config logs */
3000 #define MTI_NAME_MAXLEN 64
3001 #define MTI_PARAM_MAXLEN 4096
3002 #define MTI_NIDS_MAX 32
3003 struct mgs_target_info {
3004 __u32 mti_lustre_ver;
3005 __u32 mti_stripe_index;
3006 __u32 mti_config_ver;
3008 __u32 mti_nid_count;
3009 __u32 mti_instance; /* Running instance of target */
3010 char mti_fsname[MTI_NAME_MAXLEN];
3011 char mti_svname[MTI_NAME_MAXLEN];
3012 char mti_uuid[sizeof(struct obd_uuid)];
3013 __u64 mti_nids[MTI_NIDS_MAX]; /* host nids (lnet_nid_t)*/
3014 char mti_params[MTI_PARAM_MAXLEN];
3016 extern void lustre_swab_mgs_target_info(struct mgs_target_info *oinfo);
3018 struct mgs_nidtbl_entry {
3019 __u64 mne_version; /* table version of this entry */
3020 __u32 mne_instance; /* target instance # */
3021 __u32 mne_index; /* target index */
3022 __u32 mne_length; /* length of this entry - by bytes */
3023 __u8 mne_type; /* target type LDD_F_SV_TYPE_OST/MDT */
3024 __u8 mne_nid_type; /* type of nid(mbz). for ipv6. */
3025 __u8 mne_nid_size; /* size of each NID, by bytes */
3026 __u8 mne_nid_count; /* # of NIDs in buffer */
3028 lnet_nid_t nids[0]; /* variable size buffer for NIDs. */
3031 extern void lustre_swab_mgs_nidtbl_entry(struct mgs_nidtbl_entry *oinfo);
3033 struct mgs_config_body {
3034 char mcb_name[MTI_NAME_MAXLEN]; /* logname */
3035 __u64 mcb_offset; /* next index of config log to request */
3036 __u16 mcb_type; /* type of log: CONFIG_T_[CONFIG|RECOVER] */
3038 __u8 mcb_bits; /* bits unit size of config log */
3039 __u32 mcb_units; /* # of units for bulk transfer */
3041 extern void lustre_swab_mgs_config_body(struct mgs_config_body *body);
3043 struct mgs_config_res {
3044 __u64 mcr_offset; /* index of last config log */
3045 __u64 mcr_size; /* size of the log */
3047 extern void lustre_swab_mgs_config_res(struct mgs_config_res *body);
3049 /* Config marker flags (in config log) */
3050 #define CM_START 0x01
3052 #define CM_SKIP 0x04
3053 #define CM_UPGRADE146 0x08
3054 #define CM_EXCLUDE 0x10
3055 #define CM_START_SKIP (CM_START | CM_SKIP)
3058 __u32 cm_step; /* aka config version */
3060 __u32 cm_vers; /* lustre release version number */
3061 __u32 cm_padding; /* 64 bit align */
3062 __s64 cm_createtime; /*when this record was first created */
3063 __s64 cm_canceltime; /*when this record is no longer valid*/
3064 char cm_tgtname[MTI_NAME_MAXLEN];
3065 char cm_comment[MTI_NAME_MAXLEN];
3068 extern void lustre_swab_cfg_marker(struct cfg_marker *marker,
3069 int swab, int size);
3072 * Opcodes for multiple servers.
3078 OBD_QC_CALLBACK, /* not used since 2.4 */
3082 #define OBD_FIRST_OPC OBD_PING
3085 * llog contexts indices.
3087 * There is compatibility problem with indexes below, they are not
3088 * continuous and must keep their numbers for compatibility needs.
3089 * See LU-5218 for details.
3092 LLOG_CONFIG_ORIG_CTXT = 0,
3093 LLOG_CONFIG_REPL_CTXT = 1,
3094 LLOG_MDS_OST_ORIG_CTXT = 2,
3095 LLOG_MDS_OST_REPL_CTXT = 3, /* kept just to avoid re-assignment */
3096 LLOG_SIZE_ORIG_CTXT = 4,
3097 LLOG_SIZE_REPL_CTXT = 5,
3098 LLOG_TEST_ORIG_CTXT = 8,
3099 LLOG_TEST_REPL_CTXT = 9, /* kept just to avoid re-assignment */
3100 LLOG_CHANGELOG_ORIG_CTXT = 12, /**< changelog generation on mdd */
3101 LLOG_CHANGELOG_REPL_CTXT = 13, /**< changelog access on clients */
3102 /* for multiple changelog consumers */
3103 LLOG_CHANGELOG_USER_ORIG_CTXT = 14,
3104 LLOG_AGENT_ORIG_CTXT = 15, /**< agent requests generation on cdt */
3105 LLOG_UPDATELOG_ORIG_CTXT = 16, /* update log */
3106 LLOG_UPDATELOG_REPL_CTXT = 17, /* update log */
3110 /** Identifier for a single log object */
3112 struct ost_id lgl_oi;
3114 } __attribute__((packed));
3116 /** Records written to the CATALOGS list */
3117 #define CATLIST "CATALOGS"
3119 struct llog_logid lci_logid;
3123 } __attribute__((packed));
3125 /* Log data record types - there is no specific reason that these need to
3126 * be related to the RPC opcodes, but no reason not to (may be handy later?)
3128 #define LLOG_OP_MAGIC 0x10600000
3129 #define LLOG_OP_MASK 0xfff00000
3132 LLOG_PAD_MAGIC = LLOG_OP_MAGIC | 0x00000,
3133 OST_SZ_REC = LLOG_OP_MAGIC | 0x00f00,
3134 /* OST_RAID1_REC = LLOG_OP_MAGIC | 0x01000, never used */
3135 MDS_UNLINK_REC = LLOG_OP_MAGIC | 0x10000 | (MDS_REINT << 8) |
3136 REINT_UNLINK, /* obsolete after 2.5.0 */
3137 MDS_UNLINK64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
3139 /* MDS_SETATTR_REC = LLOG_OP_MAGIC | 0x12401, obsolete 1.8.0 */
3140 MDS_SETATTR64_REC = LLOG_OP_MAGIC | 0x90000 | (MDS_REINT << 8) |
3142 OBD_CFG_REC = LLOG_OP_MAGIC | 0x20000,
3143 /* PTL_CFG_REC = LLOG_OP_MAGIC | 0x30000, obsolete 1.4.0 */
3144 LLOG_GEN_REC = LLOG_OP_MAGIC | 0x40000,
3145 /* LLOG_JOIN_REC = LLOG_OP_MAGIC | 0x50000, obsolete 1.8.0 */
3146 CHANGELOG_REC = LLOG_OP_MAGIC | 0x60000,
3147 CHANGELOG_USER_REC = LLOG_OP_MAGIC | 0x70000,
3148 HSM_AGENT_REC = LLOG_OP_MAGIC | 0x80000,
3149 UPDATE_REC = LLOG_OP_MAGIC | 0xa0000,
3150 LLOG_HDR_MAGIC = LLOG_OP_MAGIC | 0x45539,
3151 LLOG_LOGID_MAGIC = LLOG_OP_MAGIC | 0x4553b,
3154 #define LLOG_REC_HDR_NEEDS_SWABBING(r) \
3155 (((r)->lrh_type & __swab32(LLOG_OP_MASK)) == __swab32(LLOG_OP_MAGIC))
3157 /** Log record header - stored in little endian order.
3158 * Each record must start with this struct, end with a llog_rec_tail,
3159 * and be a multiple of 256 bits in size.
3161 struct llog_rec_hdr {
3168 struct llog_rec_tail {
3173 /* Where data follow just after header */
3174 #define REC_DATA(ptr) \
3175 ((void *)((char *)ptr + sizeof(struct llog_rec_hdr)))
3177 #define REC_DATA_LEN(rec) \
3178 (rec->lrh_len - sizeof(struct llog_rec_hdr) - \
3179 sizeof(struct llog_rec_tail))
3181 static inline void *rec_tail(struct llog_rec_hdr *rec)
3183 return (void *)((char *)rec + rec->lrh_len -
3184 sizeof(struct llog_rec_tail));
3187 struct llog_logid_rec {
3188 struct llog_rec_hdr lid_hdr;
3189 struct llog_logid lid_id;
3193 struct llog_rec_tail lid_tail;
3194 } __attribute__((packed));
3196 struct llog_unlink_rec {
3197 struct llog_rec_hdr lur_hdr;
3201 struct llog_rec_tail lur_tail;
3202 } __attribute__((packed));
3204 struct llog_unlink64_rec {
3205 struct llog_rec_hdr lur_hdr;
3206 struct lu_fid lur_fid;
3207 __u32 lur_count; /* to destroy the lost precreated */
3211 struct llog_rec_tail lur_tail;
3212 } __attribute__((packed));
3214 struct llog_setattr64_rec {
3215 struct llog_rec_hdr lsr_hdr;
3216 struct ost_id lsr_oi;
3222 struct llog_rec_tail lsr_tail;
3223 } __attribute__((packed));
3225 struct llog_size_change_rec {
3226 struct llog_rec_hdr lsc_hdr;
3227 struct ll_fid lsc_fid;
3232 struct llog_rec_tail lsc_tail;
3233 } __attribute__((packed));
3235 #define CHANGELOG_MAGIC 0xca103000
3237 /** \a changelog_rec_type's that can't be masked */
3238 #define CHANGELOG_MINMASK (1 << CL_MARK)
3239 /** bits covering all \a changelog_rec_type's */
3240 #define CHANGELOG_ALLMASK 0XFFFFFFFF
3241 /** default \a changelog_rec_type mask. Allow all of them, except
3242 * CL_ATIME since it can really be time consuming, and not necessary
3243 * under normal use. */
3244 #define CHANGELOG_DEFMASK (CHANGELOG_ALLMASK & ~(1 << CL_ATIME))
3246 /* changelog llog name, needed by client replicators */
3247 #define CHANGELOG_CATALOG "changelog_catalog"
3249 struct changelog_setinfo {
3252 } __attribute__((packed));
3254 /** changelog record */
3255 struct llog_changelog_rec {
3256 struct llog_rec_hdr cr_hdr;
3257 struct changelog_rec cr; /**< Variable length field */
3258 struct llog_rec_tail cr_do_not_use; /**< for_sizeof_only */
3259 } __attribute__((packed));
3261 #define CHANGELOG_USER_PREFIX "cl"
3263 struct llog_changelog_user_rec {
3264 struct llog_rec_hdr cur_hdr;
3268 struct llog_rec_tail cur_tail;
3269 } __attribute__((packed));
3271 enum agent_req_status {
3279 static inline const char *agent_req_status2name(enum agent_req_status ars)
3297 static inline bool agent_req_in_final_state(enum agent_req_status ars)
3299 return ((ars == ARS_SUCCEED) || (ars == ARS_FAILED) ||
3300 (ars == ARS_CANCELED));
3303 struct llog_agent_req_rec {
3304 struct llog_rec_hdr arr_hdr; /**< record header */
3305 __u32 arr_status; /**< status of the request */
3307 * agent_req_status */
3308 __u32 arr_archive_id; /**< backend archive number */
3309 __u64 arr_flags; /**< req flags */
3310 __u64 arr_compound_id; /**< compound cookie */
3311 __u64 arr_req_create; /**< req. creation time */
3312 __u64 arr_req_change; /**< req. status change time */
3313 struct hsm_action_item arr_hai; /**< req. to the agent */
3314 struct llog_rec_tail arr_tail; /**< record tail for_sizezof_only */
3315 } __attribute__((packed));
3317 /* Old llog gen for compatibility */
3321 } __attribute__((packed));
3323 struct llog_gen_rec {
3324 struct llog_rec_hdr lgr_hdr;
3325 struct llog_gen lgr_gen;
3329 struct llog_rec_tail lgr_tail;
3332 /* flags for the logs */
3334 LLOG_F_ZAP_WHEN_EMPTY = 0x1,
3335 LLOG_F_IS_CAT = 0x2,
3336 LLOG_F_IS_PLAIN = 0x4,
3337 LLOG_F_EXT_JOBID = 0x8,
3338 LLOG_F_IS_FIXSIZE = 0x10,
3340 /* Note: Flags covered by LLOG_F_EXT_MASK will be inherited from
3341 * catlog to plain log, so do not add LLOG_F_IS_FIXSIZE here,
3342 * because the catlog record is usually fixed size, but its plain
3343 * log record can be variable */
3344 LLOG_F_EXT_MASK = LLOG_F_EXT_JOBID,
3347 /* On-disk header structure of each log object, stored in little endian order */
3348 #define LLOG_MIN_CHUNK_SIZE 8192
3349 #define LLOG_HEADER_SIZE (96) /* sizeof (llog_log_hdr) + sizeof(llh_tail)
3350 * - sizeof(llh_bitmap) */
3351 #define LLOG_BITMAP_BYTES (LLOG_MIN_CHUNK_SIZE - LLOG_HEADER_SIZE)
3352 #define LLOG_MIN_REC_SIZE (24) /* round(llog_rec_hdr + llog_rec_tail) */
3354 struct llog_log_hdr {
3355 struct llog_rec_hdr llh_hdr;
3356 __s64 llh_timestamp;
3358 __u32 llh_bitmap_offset;
3361 /* for a catalog the first/oldest and still in-use plain slot is just
3362 * next to it. It will serve as the upper limit after Catalog has
3365 struct obd_uuid llh_tgtuuid;
3366 __u32 llh_reserved[LLOG_HEADER_SIZE/sizeof(__u32)-23];
3367 /* These fields must always be at the end of the llog_log_hdr.
3368 * Note: llh_bitmap size is variable because llog chunk size could be
3369 * bigger than LLOG_MIN_CHUNK_SIZE, i.e. sizeof(llog_log_hdr) > 8192
3370 * bytes, and the real size is stored in llh_hdr.lrh_len, which means
3371 * llh_tail should only be refered by LLOG_HDR_TAIL().
3372 * But this structure is also used by client/server llog interface
3373 * (see llog_client.c), it will be kept in its original way to avoid
3374 * compatiblity issue. */
3375 __u32 llh_bitmap[LLOG_BITMAP_BYTES / sizeof(__u32)];
3376 struct llog_rec_tail llh_tail;
3377 } __attribute__((packed));
3378 #undef LLOG_HEADER_SIZE
3379 #undef LLOG_BITMAP_BYTES
3381 #define LLOG_HDR_BITMAP_SIZE(llh) (__u32)((llh->llh_hdr.lrh_len - \
3382 llh->llh_bitmap_offset - \
3383 sizeof(llh->llh_tail)) * 8)
3384 #define LLOG_HDR_BITMAP(llh) (__u32 *)((char *)(llh) + \
3385 (llh)->llh_bitmap_offset)
3386 #define LLOG_HDR_TAIL(llh) ((struct llog_rec_tail *)((char *)llh + \
3387 llh->llh_hdr.lrh_len - \
3388 sizeof(llh->llh_tail)))
3390 /** log cookies are used to reference a specific log file and a record therein */
3391 struct llog_cookie {
3392 struct llog_logid lgc_lgl;
3396 } __attribute__((packed));
3398 /** llog protocol */
3399 enum llogd_rpc_ops {
3400 LLOG_ORIGIN_HANDLE_CREATE = 501,
3401 LLOG_ORIGIN_HANDLE_NEXT_BLOCK = 502,
3402 LLOG_ORIGIN_HANDLE_READ_HEADER = 503,
3403 LLOG_ORIGIN_HANDLE_WRITE_REC = 504,
3404 LLOG_ORIGIN_HANDLE_CLOSE = 505,
3405 LLOG_ORIGIN_CONNECT = 506,
3406 LLOG_CATINFO = 507, /* deprecated */
3407 LLOG_ORIGIN_HANDLE_PREV_BLOCK = 508,
3408 LLOG_ORIGIN_HANDLE_DESTROY = 509, /* for destroy llog object*/
3410 LLOG_FIRST_OPC = LLOG_ORIGIN_HANDLE_CREATE
3414 struct llog_logid lgd_logid;
3416 __u32 lgd_llh_flags;
3418 __u32 lgd_saved_index;
3420 __u64 lgd_cur_offset;
3421 } __attribute__((packed));
3423 struct llogd_conn_body {
3424 struct llog_gen lgdc_gen;
3425 struct llog_logid lgdc_logid;
3426 __u32 lgdc_ctxt_idx;
3427 } __attribute__((packed));
3429 /* Note: 64-bit types are 64-bit aligned in structure */
3431 __u64 o_valid; /* hot fields in this obdo */
3434 __u64 o_size; /* o_size-o_blocks == ost_lvb */
3438 __u64 o_blocks; /* brw: cli sent cached bytes */
3441 /* 32-bit fields start here: keep an even number of them via padding */
3442 __u32 o_blksize; /* optimal IO blocksize */
3443 __u32 o_mode; /* brw: cli sent cache remain */
3447 __u32 o_nlink; /* brw: checksum */
3449 __u32 o_misc; /* brw: o_dropped */
3451 __u64 o_ioepoch; /* epoch in ost writes */
3452 __u32 o_stripe_idx; /* holds stripe idx */
3454 struct lustre_handle o_handle; /* brw: lock handle to prolong
3456 struct llog_cookie o_lcookie; /* destroy: unlink cookie from
3457 * MDS, obsolete in 2.8, reused
3462 __u64 o_data_version; /* getattr: sum of iversion for
3464 * brw: grant space consumed on
3465 * the client for the write */
3471 #define o_dirty o_blocks
3472 #define o_undirty o_mode
3473 #define o_dropped o_misc
3474 #define o_cksum o_nlink
3475 #define o_grant_used o_data_version
3477 struct lfsck_request {
3490 __u16 lr_async_windows;
3492 struct lu_fid lr_fid;
3493 struct lu_fid lr_fid2;
3494 struct lu_fid lr_fid3;
3499 void lustre_swab_lfsck_request(struct lfsck_request *lr);
3501 struct lfsck_reply {
3507 void lustre_swab_lfsck_reply(struct lfsck_reply *lr);
3510 LE_LASTID_REBUILDING = 1,
3511 LE_LASTID_REBUILT = 2,
3517 LE_FID_ACCESSED = 8,
3519 LE_CONDITIONAL_DESTROY = 10,
3520 LE_PAIRS_VERIFY = 11,
3521 LE_SKIP_NLINK_DECLARE = 13,
3523 LE_SET_LMV_MASTER = 15,
3524 LE_SET_LMV_SLAVE = 16,
3527 enum lfsck_event_flags {
3528 LEF_TO_OST = 0x00000001,
3529 LEF_FROM_OST = 0x00000002,
3530 LEF_SET_LMV_HASH = 0x00000004,
3531 LEF_SET_LMV_ALL = 0x00000008,
3532 LEF_RECHECK_NAME_HASH = 0x00000010,
3535 static inline void lustre_set_wire_obdo(const struct obd_connect_data *ocd,
3537 const struct obdo *lobdo)
3540 wobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
3544 if (unlikely(!(ocd->ocd_connect_flags & OBD_CONNECT_FID)) &&
3545 fid_seq_is_echo(ostid_seq(&lobdo->o_oi))) {
3546 /* Currently OBD_FL_OSTID will only be used when 2.4 echo
3547 * client communicate with pre-2.4 server */
3548 wobdo->o_oi.oi.oi_id = fid_oid(&lobdo->o_oi.oi_fid);
3549 wobdo->o_oi.oi.oi_seq = fid_seq(&lobdo->o_oi.oi_fid);
3553 static inline void lustre_get_wire_obdo(const struct obd_connect_data *ocd,
3555 const struct obdo *wobdo)
3557 __u32 local_flags = 0;
3559 if (lobdo->o_valid & OBD_MD_FLFLAGS)
3560 local_flags = lobdo->o_flags & OBD_FL_LOCAL_MASK;
3563 if (local_flags != 0) {
3564 lobdo->o_valid |= OBD_MD_FLFLAGS;
3565 lobdo->o_flags &= ~OBD_FL_LOCAL_MASK;
3566 lobdo->o_flags |= local_flags;
3571 if (unlikely(!(ocd->ocd_connect_flags & OBD_CONNECT_FID)) &&
3572 fid_seq_is_echo(wobdo->o_oi.oi.oi_seq)) {
3574 lobdo->o_oi.oi_fid.f_seq = wobdo->o_oi.oi.oi_seq;
3575 lobdo->o_oi.oi_fid.f_oid = wobdo->o_oi.oi.oi_id;
3576 lobdo->o_oi.oi_fid.f_ver = 0;
3580 extern void lustre_swab_obdo (struct obdo *o);
3582 /* request structure for OST's */
3587 /* Key for FIEMAP to be used in get_info calls */
3588 struct ll_fiemap_info_key {
3590 struct obdo lfik_oa;
3591 struct fiemap lfik_fiemap;
3594 extern void lustre_swab_ost_body (struct ost_body *b);
3595 extern void lustre_swab_ost_last_id(__u64 *id);
3596 extern void lustre_swab_fiemap(struct fiemap *fiemap);
3598 extern void lustre_swab_lov_user_md_v1(struct lov_user_md_v1 *lum);
3599 extern void lustre_swab_lov_user_md_v3(struct lov_user_md_v3 *lum);
3600 extern void lustre_swab_lov_user_md_objects(struct lov_user_ost_data *lod,
3602 extern void lustre_swab_lov_mds_md(struct lov_mds_md *lmm);
3603 void lustre_print_user_md(unsigned int level, struct lov_user_md *lum,
3607 extern void lustre_swab_llogd_body (struct llogd_body *d);
3608 extern void lustre_swab_llog_hdr (struct llog_log_hdr *h);
3609 extern void lustre_swab_llogd_conn_body (struct llogd_conn_body *d);
3610 extern void lustre_swab_llog_rec(struct llog_rec_hdr *rec);
3611 extern void lustre_swab_llog_id(struct llog_logid *lid);
3614 extern void lustre_swab_lustre_cfg(struct lustre_cfg *lcfg);
3616 /* Functions for dumping PTLRPC fields */
3617 void dump_rniobuf(struct niobuf_remote *rnb);
3618 void dump_ioo(struct obd_ioobj *nb);
3619 void dump_obdo(struct obdo *oa);
3620 void dump_ost_body(struct ost_body *ob);
3621 void dump_rcs(__u32 *rc);
3623 #define IDX_INFO_MAGIC 0x3D37CC37
3625 /* Index file transfer through the network. The server serializes the index into
3626 * a byte stream which is sent to the client via a bulk transfer */
3630 /* reply: see idx_info_flags below */
3633 /* request & reply: number of lu_idxpage (to be) transferred */
3637 /* request: requested attributes passed down to the iterator API */
3640 /* request & reply: index file identifier (FID) */
3641 struct lu_fid ii_fid;
3643 /* reply: version of the index file before starting to walk the index.
3644 * Please note that the version can be modified at any time during the
3648 /* request: hash to start with:
3649 * reply: hash of the first entry of the first lu_idxpage and hash
3650 * of the entry to read next if any */
3651 __u64 ii_hash_start;
3654 /* reply: size of keys in lu_idxpages, minimal one if II_FL_VARKEY is
3658 /* reply: size of records in lu_idxpages, minimal one if II_FL_VARREC
3666 extern void lustre_swab_idx_info(struct idx_info *ii);
3668 #define II_END_OFF MDS_DIR_END_OFF /* all entries have been read */
3670 /* List of flags used in idx_info::ii_flags */
3671 enum idx_info_flags {
3672 II_FL_NOHASH = 1 << 0, /* client doesn't care about hash value */
3673 II_FL_VARKEY = 1 << 1, /* keys can be of variable size */
3674 II_FL_VARREC = 1 << 2, /* records can be of variable size */
3675 II_FL_NONUNQ = 1 << 3, /* index supports non-unique keys */
3676 II_FL_NOKEY = 1 << 4, /* client doesn't care about key */
3679 #define LIP_MAGIC 0x8A6D6B6C
3681 /* 4KB (= LU_PAGE_SIZE) container gathering key/record pairs */
3683 /* 16-byte header */
3686 __u16 lip_nr; /* number of entries in the container */
3687 __u64 lip_pad0; /* additional padding for future use */
3689 /* key/record pairs are stored in the remaining 4080 bytes.
3690 * depending upon the flags in idx_info::ii_flags, each key/record
3691 * pair might be preceded by:
3693 * - the key size (II_FL_VARKEY is set)
3694 * - the record size (II_FL_VARREC is set)
3696 * For the time being, we only support fixed-size key & record. */
3697 char lip_entries[0];
3699 extern void lustre_swab_lip_header(struct lu_idxpage *lip);
3701 #define LIP_HDR_SIZE (offsetof(struct lu_idxpage, lip_entries))
3703 /* Gather all possible type associated with a 4KB container */
3705 struct lu_dirpage lp_dir; /* for MDS_READPAGE */
3706 struct lu_idxpage lp_idx; /* for OBD_IDX_READ */
3707 char lp_array[LU_PAGE_SIZE];
3710 /* security opcodes */
3713 SEC_CTX_INIT_CONT = 802,
3716 SEC_FIRST_OPC = SEC_CTX_INIT
3720 * capa related definitions
3722 #define CAPA_HMAC_MAX_LEN 64
3723 #define CAPA_HMAC_KEY_MAX_LEN 56
3725 /* NB take care when changing the sequence of elements this struct,
3726 * because the offset info is used in find_capa() */
3727 struct lustre_capa {
3728 struct lu_fid lc_fid; /** fid */
3729 __u64 lc_opc; /** operations allowed */
3730 __u64 lc_uid; /** file owner */
3731 __u64 lc_gid; /** file group */
3732 __u32 lc_flags; /** HMAC algorithm & flags */
3733 __u32 lc_keyid; /** key# used for the capability */
3734 __u32 lc_timeout; /** capa timeout value (sec) */
3735 __u32 lc_expiry; /** expiry time (sec) */
3736 __u8 lc_hmac[CAPA_HMAC_MAX_LEN]; /** HMAC */
3737 } __attribute__((packed));
3739 extern void lustre_swab_lustre_capa(struct lustre_capa *c);
3741 /** lustre_capa::lc_opc */
3743 CAPA_OPC_BODY_WRITE = 1<<0, /**< write object data */
3744 CAPA_OPC_BODY_READ = 1<<1, /**< read object data */
3745 CAPA_OPC_INDEX_LOOKUP = 1<<2, /**< lookup object fid */
3746 CAPA_OPC_INDEX_INSERT = 1<<3, /**< insert object fid */
3747 CAPA_OPC_INDEX_DELETE = 1<<4, /**< delete object fid */
3748 CAPA_OPC_OSS_WRITE = 1<<5, /**< write oss object data */
3749 CAPA_OPC_OSS_READ = 1<<6, /**< read oss object data */
3750 CAPA_OPC_OSS_TRUNC = 1<<7, /**< truncate oss object */
3751 CAPA_OPC_OSS_DESTROY = 1<<8, /**< destroy oss object */
3752 CAPA_OPC_META_WRITE = 1<<9, /**< write object meta data */
3753 CAPA_OPC_META_READ = 1<<10, /**< read object meta data */
3756 #define CAPA_OPC_OSS_RW (CAPA_OPC_OSS_READ | CAPA_OPC_OSS_WRITE)
3757 #define CAPA_OPC_MDS_ONLY \
3758 (CAPA_OPC_BODY_WRITE | CAPA_OPC_BODY_READ | CAPA_OPC_INDEX_LOOKUP | \
3759 CAPA_OPC_INDEX_INSERT | CAPA_OPC_INDEX_DELETE)
3760 #define CAPA_OPC_OSS_ONLY \
3761 (CAPA_OPC_OSS_WRITE | CAPA_OPC_OSS_READ | CAPA_OPC_OSS_TRUNC | \
3762 CAPA_OPC_OSS_DESTROY)
3763 #define CAPA_OPC_MDS_DEFAULT ~CAPA_OPC_OSS_ONLY
3764 #define CAPA_OPC_OSS_DEFAULT ~(CAPA_OPC_MDS_ONLY | CAPA_OPC_OSS_ONLY)
3766 static inline bool lovea_slot_is_dummy(const struct lov_ost_data_v1 *obj)
3768 /* zero area does not care about the bytes-order. */
3769 if (obj->l_ost_oi.oi.oi_id == 0 && obj->l_ost_oi.oi.oi_seq == 0 &&
3770 obj->l_ost_idx == 0 && obj->l_ost_gen == 0)
3776 /* lustre_capa::lc_hmac_alg */
3778 CAPA_HMAC_ALG_SHA1 = 1, /**< sha1 algorithm */
3782 #define CAPA_FL_MASK 0x00ffffff
3783 #define CAPA_HMAC_ALG_MASK 0xff000000
3785 struct lustre_capa_key {
3786 __u64 lk_seq; /**< mds# */
3787 __u32 lk_keyid; /**< key# */
3789 __u8 lk_key[CAPA_HMAC_KEY_MAX_LEN]; /**< key */
3790 } __attribute__((packed));
3792 extern void lustre_swab_lustre_capa_key(struct lustre_capa_key *k);
3794 /** The link ea holds 1 \a link_ea_entry for each hardlink */
3795 #define LINK_EA_MAGIC 0x11EAF1DFUL
3796 struct link_ea_header {
3799 __u64 leh_len; /* total size */
3805 /** Hardlink data is name and parent fid.
3806 * Stored in this crazy struct for maximum packing and endian-neutrality
3808 struct link_ea_entry {
3809 /** __u16 stored big-endian, unaligned */
3810 unsigned char lee_reclen[2];
3811 unsigned char lee_parent_fid[sizeof(struct lu_fid)];
3813 }__attribute__((packed));
3815 /** fid2path request/reply structure */
3816 struct getinfo_fid2path {
3817 struct lu_fid gf_fid;
3822 } __attribute__((packed));
3824 void lustre_swab_fid2path (struct getinfo_fid2path *gf);
3826 /** path2parent request/reply structures */
3828 struct lu_fid gp_fid; /**< parent FID */
3829 __u32 gp_linkno; /**< hardlink number */
3830 __u32 gp_name_size; /**< size of the name field */
3831 char gp_name[0]; /**< zero-terminated link name */
3832 } __attribute__((packed));
3835 LAYOUT_INTENT_ACCESS = 0,
3836 LAYOUT_INTENT_READ = 1,
3837 LAYOUT_INTENT_WRITE = 2,
3838 LAYOUT_INTENT_GLIMPSE = 3,
3839 LAYOUT_INTENT_TRUNC = 4,
3840 LAYOUT_INTENT_RELEASE = 5,
3841 LAYOUT_INTENT_RESTORE = 6
3844 /* enqueue layout lock with intent */
3845 struct layout_intent {
3846 __u32 li_opc; /* intent operation for enqueue, read, write etc */
3852 void lustre_swab_layout_intent(struct layout_intent *li);
3855 * On the wire version of hsm_progress structure.
3857 * Contains the userspace hsm_progress and some internal fields.
3859 struct hsm_progress_kernel {
3860 /* Field taken from struct hsm_progress */
3863 struct hsm_extent hpk_extent;
3865 __u16 hpk_errval; /* positive val */
3867 /* Additional fields */
3868 __u64 hpk_data_version;
3870 } __attribute__((packed));
3872 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
3873 extern void lustre_swab_hsm_current_action(struct hsm_current_action *action);
3874 extern void lustre_swab_hsm_progress_kernel(struct hsm_progress_kernel *hpk);
3875 extern void lustre_swab_hsm_user_state(struct hsm_user_state *hus);
3876 extern void lustre_swab_hsm_user_item(struct hsm_user_item *hui);
3877 extern void lustre_swab_hsm_request(struct hsm_request *hr);
3880 * OUT_UPDATE RPC Format
3882 * During the cross-ref operation, the Master MDT, which the client send the
3883 * request to, will disassembly the operation into object updates, then OSP
3884 * will send these updates to the remote MDT to be executed.
3886 * An UPDATE_OBJ RPC does a list of updates. Each update belongs to an
3887 * operation and does a type of modification to an object.
3895 * update (ub_count-th)
3897 * ub_count must be less than or equal to UPDATE_PER_RPC_MAX.
3902 * rc [+ buffers] (1st)
3903 * rc [+ buffers] (2st)
3905 * rc [+ buffers] (nr_count-th)
3907 * ur_count must be less than or equal to UPDATE_PER_RPC_MAX and should usually
3908 * be equal to ub_count.
3912 * Type of each update, if adding/deleting update, please also update
3913 * update_opcode in lustre/target/out_lib.c.
3925 OUT_INDEX_LOOKUP = 9,
3926 OUT_INDEX_INSERT = 10,
3927 OUT_INDEX_DELETE = 11,
3937 UPDATE_FL_OST = 0x00000001, /* op from OST (not MDT) */
3938 UPDATE_FL_SYNC = 0x00000002, /* commit before replying */
3939 UPDATE_FL_COMMITTED = 0x00000004, /* op committed globally */
3940 UPDATE_FL_NOLOG = 0x00000008 /* for idempotent updates */
3943 struct object_update_param {
3944 __u16 oup_len; /* length of this parameter */
3950 static inline size_t
3951 object_update_param_size(const struct object_update_param *param)
3953 return cfs_size_round(sizeof(*param) + param->oup_len);
3957 struct object_update {
3958 __u16 ou_type; /* enum update_type */
3959 __u16 ou_params_count; /* update parameters count */
3960 __u32 ou_result_size; /* how many bytes can return */
3961 __u32 ou_flags; /* enum update_flag */
3962 __u32 ou_padding1; /* padding 1 */
3963 __u64 ou_batchid; /* op transno on master */
3964 struct lu_fid ou_fid; /* object to be updated */
3965 struct object_update_param ou_params[0]; /* update params */
3968 #define UPDATE_REQUEST_MAGIC_V1 0xBDDE0001
3969 #define UPDATE_REQUEST_MAGIC_V2 0xBDDE0002
3970 #define UPDATE_REQUEST_MAGIC UPDATE_REQUEST_MAGIC_V2
3971 /* Hold object_updates sending to the remote OUT in single RPC */
3972 struct object_update_request {
3974 __u16 ourq_count; /* number of ourq_updates[] */
3976 struct object_update ourq_updates[0];
3979 #define OUT_UPDATE_HEADER_MAGIC 0xBDDF0001
3980 #define OUT_UPDATE_MAX_INLINE_SIZE 4096
3981 /* Header for updates request between MDTs */
3982 struct out_update_header {
3985 __u32 ouh_inline_length;
3986 __u32 ouh_reply_size;
3987 __u32 ouh_inline_data[0];
3990 struct out_update_buffer {
3995 void lustre_swab_object_update(struct object_update *ou);
3996 void lustre_swab_object_update_request(struct object_update_request *our);
3997 void lustre_swab_out_update_header(struct out_update_header *ouh);
3998 void lustre_swab_out_update_buffer(struct out_update_buffer *oub);
4000 static inline size_t
4001 object_update_params_size(const struct object_update *update)
4003 const struct object_update_param *param;
4004 size_t total_size = 0;
4007 param = &update->ou_params[0];
4008 for (i = 0; i < update->ou_params_count; i++) {
4009 size_t size = object_update_param_size(param);
4011 param = (struct object_update_param *)((char *)param + size);
4018 static inline size_t
4019 object_update_size(const struct object_update *update)
4021 return offsetof(struct object_update, ou_params[0]) +
4022 object_update_params_size(update);
4025 static inline struct object_update *
4026 object_update_request_get(const struct object_update_request *our,
4027 unsigned int index, size_t *size)
4032 if (index >= our->ourq_count)
4035 ptr = (void *)&our->ourq_updates[0];
4036 for (i = 0; i < index; i++)
4037 ptr += object_update_size(ptr);
4040 *size = object_update_size(ptr);
4046 /* the result of object update */
4047 struct object_update_result {
4054 #define UPDATE_REPLY_MAGIC_V1 0x00BD0001
4055 #define UPDATE_REPLY_MAGIC_V2 0x00BD0002
4056 #define UPDATE_REPLY_MAGIC UPDATE_REPLY_MAGIC_V2
4057 /* Hold object_update_results being replied from the remote OUT. */
4058 struct object_update_reply {
4065 void lustre_swab_object_update_result(struct object_update_result *our);
4066 void lustre_swab_object_update_reply(struct object_update_reply *our);
4068 static inline struct object_update_result *
4069 object_update_result_get(const struct object_update_reply *reply,
4070 unsigned int index, size_t *size)
4072 __u16 count = reply->ourp_count;
4079 ptr = (char *)reply +
4080 cfs_size_round(offsetof(struct object_update_reply,
4082 for (i = 0; i < index; i++) {
4083 if (reply->ourp_lens[i] == 0)
4086 ptr += cfs_size_round(reply->ourp_lens[i]);
4090 *size = reply->ourp_lens[index];
4095 /* read update result */
4096 struct out_read_reply {
4103 static inline void orr_cpu_to_le(struct out_read_reply *orr_dst,
4104 const struct out_read_reply *orr_src)
4106 orr_dst->orr_size = cpu_to_le32(orr_src->orr_size);
4107 orr_dst->orr_padding = cpu_to_le32(orr_src->orr_padding);
4108 orr_dst->orr_offset = cpu_to_le64(orr_dst->orr_offset);
4111 static inline void orr_le_to_cpu(struct out_read_reply *orr_dst,
4112 const struct out_read_reply *orr_src)
4114 orr_dst->orr_size = le32_to_cpu(orr_src->orr_size);
4115 orr_dst->orr_padding = le32_to_cpu(orr_src->orr_padding);
4116 orr_dst->orr_offset = le64_to_cpu(orr_dst->orr_offset);
4119 /** layout swap request structure
4120 * fid1 and fid2 are in mdt_body
4122 struct mdc_swap_layouts {
4126 void lustre_swab_swap_layouts(struct mdc_swap_layouts *msl);
4129 struct lustre_handle cd_handle;
4130 struct lu_fid cd_fid;
4131 __u64 cd_data_version;
4132 __u64 cd_reserved[8];
4135 void lustre_swab_close_data(struct close_data *data);
4138 void lustre_swab_update_ops(struct update_ops *uops, unsigned int op_count);
4140 /* Update llog format */
4142 struct lu_fid uop_fid;
4144 __u16 uop_param_count;
4145 __u16 uop_params_off[0];
4149 struct update_op uops_op[0];
4152 struct update_params {
4153 struct object_update_param up_params[0];
4156 enum update_records_flag {
4157 UPDATE_RECORD_CONTINUE = 1 >> 0,
4160 * This is the update record format used to store the updates in
4161 * disk. All updates of the operation will be stored in ur_ops.
4162 * All of parameters for updates of the operation will be stored
4164 * To save the space of the record, parameters in ur_ops will only
4165 * remember their offset in ur_params, so to avoid storing duplicate
4166 * parameters in ur_params, which can help us save a lot space for
4167 * operation like creating striped directory.
4169 struct update_records {
4170 __u64 ur_master_transno;
4173 /* If the operation includes multiple updates, then ur_index
4174 * means the index of the update inside the whole updates. */
4176 __u32 ur_update_count;
4177 __u32 ur_param_count;
4178 struct update_ops ur_ops;
4179 /* Note ur_ops has a variable size, so comment out
4180 * the following ur_params, in case some use it directly
4181 * update_records->ur_params
4183 * struct update_params ur_params;
4187 struct llog_update_record {
4188 struct llog_rec_hdr lur_hdr;
4189 struct update_records lur_update_rec;
4190 /* Note ur_update_rec has a variable size, so comment out
4191 * the following ur_tail, in case someone use it directly
4193 * struct llog_rec_tail lur_tail;