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 OID for local objects */
93 /** \see osd_oi_index_create */
94 OSD_OI_FID_16_OID = 2UL,
95 /** \see fld_mod_init */
97 /** \see fid_mod_init */
98 FID_SEQ_CTL_OID = 4UL,
99 FID_SEQ_SRV_OID = 5UL,
100 /** \see mdd_mod_init */
101 MDD_ROOT_INDEX_OID = 6UL,
102 MDD_ORPHAN_OID = 7UL,
103 MDD_LOV_OBJ_OID = 8UL,
104 MDD_CAPA_KEYS_OID = 9UL,
105 MDD_OBJECTS_OID = 10UL,
106 /** \see mdt_mod_init */
107 MDT_LAST_RECV_OID = 11UL,
108 /** \see osd_mod_init */
109 OSD_REM_OBJ_DIR_OID = 12UL,
112 static inline void lu_local_obj_fid(struct lu_fid *fid, __u32 oid)
114 fid->f_seq = FID_SEQ_LOCAL_FILE;
121 LUSTRE_SEQ_CONTROLLER
129 struct lu_server_seq;
131 /* Client sequence manager interface. */
132 struct lu_client_seq {
133 /* Sequence-controller export. */
134 struct obd_export *lcs_exp;
135 cfs_semaphore_t lcs_sem;
138 * Range of allowed for allocation sequeces. When using lu_client_seq on
139 * clients, this contains meta-sequence range. And for servers this
140 * contains super-sequence range.
142 struct lu_seq_range lcs_space;
144 /* Seq related proc */
145 cfs_proc_dir_entry_t *lcs_proc_dir;
147 /* This holds last allocated fid in last obtained seq */
148 struct lu_fid lcs_fid;
150 /* LUSTRE_SEQ_METADATA or LUSTRE_SEQ_DATA */
151 enum lu_cli_type lcs_type;
154 * Service uuid, passed from MDT + seq name to form unique seq name to
155 * use it with procfs.
160 * Sequence width, that is how many objects may be allocated in one
161 * sequence. Default value for it is LUSTRE_SEQ_MAX_WIDTH.
165 /* Seq-server for direct talking */
166 struct lu_server_seq *lcs_srv;
168 /* wait queue for fid allocation and update indicator */
169 cfs_waitq_t lcs_waitq;
173 /* server sequence manager interface */
174 struct lu_server_seq {
175 /* Available sequences space */
176 struct lu_seq_range lss_space;
178 /* keeps highwater in lsr_end for seq allocation algorithm */
179 struct lu_seq_range lss_lowater_set;
180 struct lu_seq_range lss_hiwater_set;
183 * Device for server side seq manager needs (saving sequences to backing
186 struct dt_device *lss_dev;
188 /* /seq file object device */
189 struct dt_object *lss_obj;
191 /* Seq related proc */
192 cfs_proc_dir_entry_t *lss_proc_dir;
194 /* LUSTRE_SEQ_SERVER or LUSTRE_SEQ_CONTROLLER */
195 enum lu_mgr_type lss_type;
197 /* Client interafce to request controller */
198 struct lu_client_seq *lss_cli;
200 /* Semaphore for protecting allocation */
201 cfs_semaphore_t lss_sem;
204 * Service uuid, passed from MDT + seq name to form unique seq name to
205 * use it with procfs.
210 * Allocation chunks for super and meta sequences. Default values are
211 * LUSTRE_SEQ_SUPER_WIDTH and LUSTRE_SEQ_META_WIDTH.
216 * minimum lss_alloc_set size that should be allocated from
221 /* sync is needed for update operation */
224 * Pointer to site object, required to access site fld.
226 struct md_site *lss_site;
229 int seq_query(struct com_thread_info *info);
232 int seq_server_init(struct lu_server_seq *seq,
233 struct dt_device *dev,
235 enum lu_mgr_type type,
237 const struct lu_env *env);
239 void seq_server_fini(struct lu_server_seq *seq,
240 const struct lu_env *env);
242 int seq_server_alloc_super(struct lu_server_seq *seq,
243 struct lu_seq_range *out,
244 const struct lu_env *env);
246 int seq_server_alloc_meta(struct lu_server_seq *seq,
247 struct lu_seq_range *out,
248 const struct lu_env *env);
250 int seq_server_set_cli(struct lu_server_seq *seq,
251 struct lu_client_seq *cli,
252 const struct lu_env *env);
255 int seq_client_init(struct lu_client_seq *seq,
256 struct obd_export *exp,
257 enum lu_cli_type type,
259 struct lu_server_seq *srv);
261 void seq_client_fini(struct lu_client_seq *seq);
263 void seq_client_flush(struct lu_client_seq *seq);
265 int seq_client_alloc_fid(const struct lu_env *env, struct lu_client_seq *seq,
267 int seq_client_get_seq(const struct lu_env *env, struct lu_client_seq *seq,
270 /* Fids common stuff */
271 int fid_is_local(const struct lu_env *env,
272 struct lu_site *site, const struct lu_fid *fid);
276 struct ldlm_namespace;
279 LUSTRE_RES_ID_SEQ_OFF = 0,
280 LUSTRE_RES_ID_OID_OFF = 1,
281 LUSTRE_RES_ID_VER_OFF = 2,
282 LUSTRE_RES_ID_HSH_OFF = 3
286 * Build (DLM) resource name from fid.
288 static inline struct ldlm_res_id *
289 fid_build_reg_res_name(const struct lu_fid *f,
290 struct ldlm_res_id *name)
292 memset(name, 0, sizeof *name);
293 name->name[LUSTRE_RES_ID_SEQ_OFF] = fid_seq(f);
294 name->name[LUSTRE_RES_ID_OID_OFF] = fid_oid(f);
295 name->name[LUSTRE_RES_ID_VER_OFF] = fid_ver(f);
300 * Return true if resource is for object identified by fid.
302 static inline int fid_res_name_eq(const struct lu_fid *f,
303 const struct ldlm_res_id *name)
306 name->name[LUSTRE_RES_ID_SEQ_OFF] == fid_seq(f) &&
307 name->name[LUSTRE_RES_ID_OID_OFF] == fid_oid(f) &&
308 name->name[LUSTRE_RES_ID_VER_OFF] == fid_ver(f);
312 static inline struct ldlm_res_id *
313 fid_build_pdo_res_name(const struct lu_fid *f,
315 struct ldlm_res_id *name)
317 fid_build_reg_res_name(f, name);
318 name->name[LUSTRE_RES_ID_HSH_OFF] = hash;
324 * Flatten 128-bit FID values into a 64-bit value for use as an inode number.
325 * For non-IGIF FIDs this starts just over 2^32, and continues without
326 * conflict until 2^64, at which point we wrap the high 24 bits of the SEQ
327 * into the range where there may not be many OID values in use, to minimize
328 * the risk of conflict.
330 * Suppose LUSTRE_SEQ_MAX_WIDTH less than (1 << 24) which is currently true,
331 * the time between re-used inode numbers is very long - 2^40 SEQ numbers,
332 * or about 2^40 client mounts, if clients create less than 2^24 files/mount.
334 static inline __u64 fid_flatten(const struct lu_fid *fid)
339 if (fid_is_igif(fid)) {
340 ino = lu_igif_ino(fid);
346 ino = (seq << 24) + ((seq >> 24) & 0xffffff0000ULL) + fid_oid(fid);
348 RETURN(ino ? ino : fid_oid(fid));
351 static inline __u32 fid_hash(const struct lu_fid *f, int bits)
353 /* all objects with same id and different versions will belong to same
354 * collisions list. */
355 return cfs_hash_long(fid_flatten(f), bits);
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;
372 /* Map the high bits of the OID into higher bits of the inode number so
373 * that inodes generated at about the same time have a reduced chance
374 * of collisions. This will give a period of 2^12 = 1024 unique clients
375 * (from SEQ) and up to min(LUSTRE_SEQ_MAX_WIDTH, 2^20) = 128k objects
376 * (from OID), or up to 128M inodes without collisions for new files. */
377 ino = ((seq & 0x000fffffULL) << 12) + ((seq >> 8) & 0xfffff000) +
378 (seq >> (64 - (40-8)) & 0xffffff00) +
379 (fid_oid(fid) & 0xff000fff) + ((fid_oid(fid) & 0x00fff000) << 8);
381 RETURN(ino ? ino : fid_oid(fid));
384 #define LUSTRE_SEQ_SRV_NAME "seq_srv"
385 #define LUSTRE_SEQ_CTL_NAME "seq_ctl"
387 /* Range common stuff */
388 static inline void range_cpu_to_le(struct lu_seq_range *dst, const struct lu_seq_range *src)
390 dst->lsr_start = cpu_to_le64(src->lsr_start);
391 dst->lsr_end = cpu_to_le64(src->lsr_end);
392 dst->lsr_index = cpu_to_le32(src->lsr_index);
393 dst->lsr_flags = cpu_to_le32(src->lsr_flags);
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_index = le32_to_cpu(src->lsr_index);
401 dst->lsr_flags = le32_to_cpu(src->lsr_flags);
404 static inline void range_cpu_to_be(struct lu_seq_range *dst, const struct lu_seq_range *src)
406 dst->lsr_start = cpu_to_be64(src->lsr_start);
407 dst->lsr_end = cpu_to_be64(src->lsr_end);
408 dst->lsr_index = cpu_to_be32(src->lsr_index);
409 dst->lsr_flags = cpu_to_be32(src->lsr_flags);
412 static inline void range_be_to_cpu(struct lu_seq_range *dst, const struct lu_seq_range *src)
414 dst->lsr_start = be64_to_cpu(src->lsr_start);
415 dst->lsr_end = be64_to_cpu(src->lsr_end);
416 dst->lsr_index = be32_to_cpu(src->lsr_index);
417 dst->lsr_flags = be32_to_cpu(src->lsr_flags);
422 #endif /* __LINUX_FID_H */