1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/include/lustre_fid.h
38 * Author: Yury Umanets <umka@clusterfs.com>
52 #include <libcfs/libcfs.h>
53 #include <lustre/lustre_idl.h>
54 #include <lustre_req_layout.h>
55 #include <lustre_mdt.h>
61 /* Whole sequences space range and zero range definitions */
62 extern const struct lu_seq_range LUSTRE_SEQ_SPACE_RANGE;
63 extern const struct lu_seq_range LUSTRE_SEQ_ZERO_RANGE;
64 extern const struct lu_fid LUSTRE_BFL_FID;
65 extern const struct lu_fid LU_OBF_FID;
66 extern const struct lu_fid LU_DOT_LUSTRE_FID;
70 * This is how may FIDs may be allocated in one sequence(128k)
72 LUSTRE_SEQ_MAX_WIDTH = 0x0000000000020000ULL,
75 * How many sequences to allocate to a client at once.
77 LUSTRE_SEQ_META_WIDTH = 0x0000000000000001ULL,
80 * seq allocation pool size.
82 LUSTRE_SEQ_BATCH_WIDTH = LUSTRE_SEQ_META_WIDTH * 1000,
85 * This is how many sequences may be in one super-sequence allocated to
88 LUSTRE_SEQ_SUPER_WIDTH = ((1ULL << 30ULL) * LUSTRE_SEQ_META_WIDTH)
91 /** special fid seq: used for local object create. */
92 #define FID_SEQ_LOCAL_FILE (FID_SEQ_START + 1)
94 /** special fid seq: used for .lustre objects. */
95 #define LU_DOT_LUSTRE_SEQ (FID_SEQ_START + 0x02ULL)
97 /* Note that reserved SEQ numbers below 12 will conflict with ldiskfs
98 * inodes in the IGIF namespace, so these reserved SEQ numbers must be
99 * used sparingly until ldiskfs-based MDT backends and/or IGIF FIDs
100 * have been completely removed. */
102 /** fid sequence for distributed fs objects */
103 #define FID_SEQ_DISTRIBUTED_START (FID_SEQ_START + 0x400ULL)
105 /** special OID for local objects */
107 /** \see osd_oi_index_create */
108 OSD_OI_FID_16_OID = 2UL,
109 /** \see fld_mod_init */
111 /** \see fid_mod_init */
112 FID_SEQ_CTL_OID = 4UL,
113 FID_SEQ_SRV_OID = 5UL,
114 /** \see mdd_mod_init */
115 MDD_ROOT_INDEX_OID = 6UL,
116 MDD_ORPHAN_OID = 7UL,
117 MDD_LOV_OBJ_OID = 8UL,
118 MDD_CAPA_KEYS_OID = 9UL,
119 MDD_OBJECTS_OID = 10UL,
120 /** \see mdt_mod_init */
121 MDT_LAST_RECV_OID = 11UL,
122 /** \see osd_mod_init */
123 OSD_REM_OBJ_DIR_OID = 12UL,
126 static inline void lu_local_obj_fid(struct lu_fid *fid, __u32 oid)
128 fid->f_seq = FID_SEQ_LOCAL_FILE;
135 LUSTRE_SEQ_CONTROLLER
143 struct lu_server_seq;
145 /* Client sequence manager interface. */
146 struct lu_client_seq {
147 /* Sequence-controller export. */
148 struct obd_export *lcs_exp;
149 cfs_semaphore_t lcs_sem;
152 * Range of allowed for allocation sequeces. When using lu_client_seq on
153 * clients, this contains meta-sequence range. And for servers this
154 * contains super-sequence range.
156 struct lu_seq_range lcs_space;
158 /* Seq related proc */
159 cfs_proc_dir_entry_t *lcs_proc_dir;
161 /* This holds last allocated fid in last obtained seq */
162 struct lu_fid lcs_fid;
164 /* LUSTRE_SEQ_METADATA or LUSTRE_SEQ_DATA */
165 enum lu_cli_type lcs_type;
168 * Service uuid, passed from MDT + seq name to form unique seq name to
169 * use it with procfs.
174 * Sequence width, that is how many objects may be allocated in one
175 * sequence. Default value for it is LUSTRE_SEQ_MAX_WIDTH.
179 /* Seq-server for direct talking */
180 struct lu_server_seq *lcs_srv;
183 /* server sequence manager interface */
184 struct lu_server_seq {
185 /* Available sequences space */
186 struct lu_seq_range lss_space;
188 /* keeps highwater in lsr_end for seq allocation algorithm */
189 struct lu_seq_range lss_lowater_set;
190 struct lu_seq_range lss_hiwater_set;
193 * Device for server side seq manager needs (saving sequences to backing
196 struct dt_device *lss_dev;
198 /* /seq file object device */
199 struct dt_object *lss_obj;
201 /* Seq related proc */
202 cfs_proc_dir_entry_t *lss_proc_dir;
204 /* LUSTRE_SEQ_SERVER or LUSTRE_SEQ_CONTROLLER */
205 enum lu_mgr_type lss_type;
207 /* Client interafce to request controller */
208 struct lu_client_seq *lss_cli;
210 /* Semaphore for protecting allocation */
211 cfs_semaphore_t lss_sem;
214 * Service uuid, passed from MDT + seq name to form unique seq name to
215 * use it with procfs.
220 * Allocation chunks for super and meta sequences. Default values are
221 * LUSTRE_SEQ_SUPER_WIDTH and LUSTRE_SEQ_META_WIDTH.
226 * minimum lss_alloc_set size that should be allocated from
231 /* transaction no of seq update write operation */
232 __u64 lss_set_transno;
234 * Pointer to site object, required to access site fld.
236 struct md_site *lss_site;
239 int seq_query(struct com_thread_info *info);
242 int seq_server_init(struct lu_server_seq *seq,
243 struct dt_device *dev,
245 enum lu_mgr_type type,
247 const struct lu_env *env);
249 void seq_server_fini(struct lu_server_seq *seq,
250 const struct lu_env *env);
252 int seq_server_alloc_super(struct lu_server_seq *seq,
253 struct lu_seq_range *out,
254 const struct lu_env *env);
256 int seq_server_alloc_meta(struct lu_server_seq *seq,
257 struct lu_seq_range *out,
258 const struct lu_env *env);
260 int seq_server_set_cli(struct lu_server_seq *seq,
261 struct lu_client_seq *cli,
262 const struct lu_env *env);
265 int seq_client_init(struct lu_client_seq *seq,
266 struct obd_export *exp,
267 enum lu_cli_type type,
269 struct lu_server_seq *srv);
271 void seq_client_fini(struct lu_client_seq *seq);
273 void seq_client_flush(struct lu_client_seq *seq);
275 int seq_client_alloc_fid(struct lu_client_seq *seq,
278 /* Fids common stuff */
279 int fid_is_local(const struct lu_env *env,
280 struct lu_site *site, const struct lu_fid *fid);
284 struct ldlm_namespace;
287 LUSTRE_RES_ID_SEQ_OFF = 0,
288 LUSTRE_RES_ID_OID_OFF = 1,
289 LUSTRE_RES_ID_VER_OFF = 2,
290 LUSTRE_RES_ID_HSH_OFF = 3
294 * Build (DLM) resource name from fid.
296 static inline struct ldlm_res_id *
297 fid_build_reg_res_name(const struct lu_fid *f,
298 struct ldlm_res_id *name)
300 memset(name, 0, sizeof *name);
301 name->name[LUSTRE_RES_ID_SEQ_OFF] = fid_seq(f);
302 name->name[LUSTRE_RES_ID_OID_OFF] = fid_oid(f);
303 name->name[LUSTRE_RES_ID_VER_OFF] = fid_ver(f);
308 * Return true if resource is for object identified by fid.
310 static inline int fid_res_name_eq(const struct lu_fid *f,
311 const struct ldlm_res_id *name)
314 name->name[LUSTRE_RES_ID_SEQ_OFF] == fid_seq(f) &&
315 name->name[LUSTRE_RES_ID_OID_OFF] == fid_oid(f) &&
316 name->name[LUSTRE_RES_ID_VER_OFF] == fid_ver(f);
320 static inline struct ldlm_res_id *
321 fid_build_pdo_res_name(const struct lu_fid *f,
323 struct ldlm_res_id *name)
325 fid_build_reg_res_name(f, name);
326 name->name[LUSTRE_RES_ID_HSH_OFF] = hash;
332 * Flatten 128-bit FID values into a 64-bit value for
333 * use as an inode number. For non-IGIF FIDs this
334 * starts just over 2^32, and continues without conflict
335 * until 2^64, at which point we wrap the high 32 bits
336 * of the SEQ into the range where there may not be many
337 * OID values in use, to minimize the risk of conflict.
339 * The time between re-used inode numbers is very long -
340 * 2^32 SEQ numbers, or about 2^32 client mounts. */
341 static inline __u64 fid_flatten(const struct lu_fid *fid)
346 if (fid_is_igif(fid)) {
347 ino = lu_igif_ino(fid);
353 ino = (seq << 24) + ((seq >> (64-8)) & 0xffffff0000ULL) + fid_oid(fid);
355 RETURN(ino ? ino : fid_oid(fid));
359 * map fid to 32 bit value for ino on 32bit systems. */
360 static inline __u32 fid_flatten32(const struct lu_fid *fid)
365 if (fid_is_igif(fid)) {
366 ino = lu_igif_ino(fid);
370 seq = fid_seq(fid) - FID_SEQ_START;
373 map the high bits of the OID into higher bits of the inode number so that
374 inodes generated at about the same time have a reduced chance of collisions.
375 This will give a period of 1024 clients and 128 k = 128M inodes without collisions.
378 ino = ((seq & 0x000fffffULL) << 12) + ((seq >> 8) & 0xfffff000) +
379 (seq >> (64 - (40-8)) & 0xffffff00) +
380 (fid_oid(fid) & 0xff000fff) + ((fid_oid(fid) & 0x00fff000) << 16);
382 RETURN(ino ? ino : fid_oid(fid));
385 #define LUSTRE_SEQ_SRV_NAME "seq_srv"
386 #define LUSTRE_SEQ_CTL_NAME "seq_ctl"
388 /* Range common stuff */
389 static inline void range_cpu_to_le(struct lu_seq_range *dst, const struct lu_seq_range *src)
391 dst->lsr_start = cpu_to_le64(src->lsr_start);
392 dst->lsr_end = cpu_to_le64(src->lsr_end);
393 dst->lsr_mdt = cpu_to_le32(src->lsr_mdt);
396 static inline void range_le_to_cpu(struct lu_seq_range *dst, const struct lu_seq_range *src)
398 dst->lsr_start = le64_to_cpu(src->lsr_start);
399 dst->lsr_end = le64_to_cpu(src->lsr_end);
400 dst->lsr_mdt = le32_to_cpu(src->lsr_mdt);
403 static inline void range_cpu_to_be(struct lu_seq_range *dst, const struct lu_seq_range *src)
405 dst->lsr_start = cpu_to_be64(src->lsr_start);
406 dst->lsr_end = cpu_to_be64(src->lsr_end);
407 dst->lsr_mdt = cpu_to_be32(src->lsr_mdt);
410 static inline void range_be_to_cpu(struct lu_seq_range *dst, const struct lu_seq_range *src)
412 dst->lsr_start = be64_to_cpu(src->lsr_start);
413 dst->lsr_end = be64_to_cpu(src->lsr_end);
414 dst->lsr_mdt = be32_to_cpu(src->lsr_mdt);
419 #endif /* __LINUX_FID_H */