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, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/fld/fld_index.c
38 * Author: WangDi <wangdi@clusterfs.com>
39 * Author: Yury Umanets <umka@clusterfs.com>
42 #define DEBUG_SUBSYSTEM S_FLD
44 #include <libcfs/libcfs.h>
45 #include <linux/module.h>
46 #include <obd_support.h>
47 #include <dt_object.h>
48 #include <lustre_fid.h>
49 #include <lustre_fld.h>
50 #include "fld_internal.h"
52 static const char fld_index_name[] = "fld";
54 static const struct lu_seq_range IGIF_FLD_RANGE = {
55 .lsr_start = FID_SEQ_IGIF,
56 .lsr_end = FID_SEQ_IGIF_MAX + 1,
58 .lsr_flags = LU_SEQ_RANGE_MDT
61 static const struct lu_seq_range DOT_LUSTRE_FLD_RANGE = {
62 .lsr_start = FID_SEQ_DOT_LUSTRE,
63 .lsr_end = FID_SEQ_DOT_LUSTRE + 1,
65 .lsr_flags = LU_SEQ_RANGE_MDT
68 static const struct lu_seq_range ROOT_FLD_RANGE = {
69 .lsr_start = FID_SEQ_ROOT,
70 .lsr_end = FID_SEQ_ROOT + 1,
72 .lsr_flags = LU_SEQ_RANGE_MDT
75 static const struct dt_index_features fld_index_features = {
76 .dif_flags = DT_IND_UPDATE,
77 .dif_keysize_min = sizeof(u64),
78 .dif_keysize_max = sizeof(u64),
79 .dif_recsize_min = sizeof(struct lu_seq_range),
80 .dif_recsize_max = sizeof(struct lu_seq_range),
84 extern struct lu_context_key fld_thread_key;
86 int fld_declare_index_create(const struct lu_env *env,
87 struct lu_server_fld *fld,
88 const struct lu_seq_range *new_range,
91 struct lu_seq_range *tmp;
92 struct lu_seq_range *range;
93 struct fld_thread_info *info;
98 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
99 range = &info->fti_lrange;
100 tmp = &info->fti_irange;
101 memset(range, 0, sizeof(*range));
103 rc = fld_index_lookup(env, fld, new_range->lsr_start, range);
105 /* In case of duplicate entry, the location must be same */
106 LASSERT((range_compare_loc(new_range, range) == 0));
107 GOTO(out, rc = -EEXIST);
111 CERROR("%s: lookup range "DRANGE" error: rc = %d\n",
112 fld->lsf_name, PRANGE(range), rc);
116 /* Check for merge case, since the fld entry can only be increamental,
117 * so we will only check whether it can be merged from the left. */
118 if (new_range->lsr_start == range->lsr_end && range->lsr_end != 0 &&
119 range_compare_loc(new_range, range) == 0) {
120 range_cpu_to_be(tmp, range);
121 rc = dt_declare_delete(env, fld->lsf_obj,
122 (struct dt_key *)&tmp->lsr_start, th);
124 CERROR("%s: declare record "DRANGE" failed: rc = %d\n",
125 fld->lsf_name, PRANGE(range), rc);
129 tmp->lsr_start = range->lsr_start;
134 range_cpu_to_be(tmp, tmp);
135 rc = dt_declare_insert(env, fld->lsf_obj, (struct dt_rec *)tmp,
136 (struct dt_key *)&tmp->lsr_start, th);
142 * insert range in fld store.
144 * \param range range to be inserted
145 * \param th transaction for this operation as it could compound
151 * The whole fld index insertion is protected by seq->lss_mutex (see
152 * seq_server_alloc_super), i.e. only one thread will access fldb each
153 * time, so we do not need worry the fld file and cache will being
154 * changed between declare and create.
155 * Because the fld entry can only be increamental, so we will only check
156 * whether it can be merged from the left.
158 * Caller must hold fld->lsf_lock
160 int fld_index_create(const struct lu_env *env, struct lu_server_fld *fld,
161 const struct lu_seq_range *new_range, struct thandle *th)
163 struct lu_seq_range *range;
164 struct lu_seq_range *tmp;
165 struct fld_thread_info *info;
168 struct fld_cache_entry *flde;
171 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
173 LASSERT(mutex_is_locked(&fld->lsf_lock));
175 range = &info->fti_lrange;
176 memset(range, 0, sizeof(*range));
177 tmp = &info->fti_irange;
178 rc = fld_index_lookup(env, fld, new_range->lsr_start, range);
180 rc = rc == 0 ? -EEXIST : rc;
184 if (new_range->lsr_start == range->lsr_end && range->lsr_end != 0 &&
185 range_compare_loc(new_range, range) == 0) {
186 range_cpu_to_be(tmp, range);
187 rc = dt_delete(env, fld->lsf_obj,
188 (struct dt_key *)&tmp->lsr_start, th,
193 tmp->lsr_start = range->lsr_start;
199 range_cpu_to_be(tmp, tmp);
200 rc = dt_insert(env, fld->lsf_obj, (struct dt_rec *)tmp,
201 (struct dt_key *)&tmp->lsr_start, th, BYPASS_CAPA, 1);
203 CERROR("%s: insert range "DRANGE" failed: rc = %d\n",
204 fld->lsf_name, PRANGE(new_range), rc);
208 flde = fld_cache_entry_create(new_range);
210 GOTO(out, rc = PTR_ERR(flde));
212 write_lock(&fld->lsf_cache->fci_lock);
214 fld_cache_delete_nolock(fld->lsf_cache, new_range);
215 rc = fld_cache_insert_nolock(fld->lsf_cache, flde);
216 write_unlock(&fld->lsf_cache->fci_lock);
224 * lookup range for a seq passed. note here we only care about the start/end,
225 * caller should handle the attached location data (flags, index).
227 * \param seq seq for lookup.
228 * \param range result of lookup.
230 * \retval 0 found, \a range is the matched range;
231 * \retval -ENOENT not found, \a range is the left-side range;
232 * \retval -ve other error;
234 int fld_index_lookup(const struct lu_env *env, struct lu_server_fld *fld,
235 u64 seq, struct lu_seq_range *range)
237 struct lu_seq_range *fld_rec;
238 struct fld_thread_info *info;
243 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
244 fld_rec = &info->fti_rec;
246 rc = fld_cache_lookup(fld->lsf_cache, seq, fld_rec);
249 if (range_within(range, seq))
255 CDEBUG(D_INFO, "%s: lookup seq = "LPX64" range : "DRANGE" rc = %d\n",
256 fld->lsf_name, seq, PRANGE(range), rc);
262 * insert entry in fld store.
264 * \param env relevant lu_env
265 * \param fld fld store
266 * \param range range to be inserted
271 * Caller must hold fld->lsf_lock
274 int fld_insert_entry(const struct lu_env *env,
275 struct lu_server_fld *fld,
276 const struct lu_seq_range *range)
282 LASSERT(mutex_is_locked(&fld->lsf_lock));
284 th = dt_trans_create(env, lu2dt_dev(fld->lsf_obj->do_lu.lo_dev));
288 rc = fld_declare_index_create(env, fld, range, th);
295 rc = dt_trans_start_local(env, lu2dt_dev(fld->lsf_obj->do_lu.lo_dev),
300 rc = fld_index_create(env, fld, range, th);
304 dt_trans_stop(env, lu2dt_dev(fld->lsf_obj->do_lu.lo_dev), th);
307 EXPORT_SYMBOL(fld_insert_entry);
309 static int fld_insert_special_entries(const struct lu_env *env,
310 struct lu_server_fld *fld)
314 rc = fld_insert_entry(env, fld, &IGIF_FLD_RANGE);
318 rc = fld_insert_entry(env, fld, &DOT_LUSTRE_FLD_RANGE);
322 rc = fld_insert_entry(env, fld, &ROOT_FLD_RANGE);
327 int fld_index_init(const struct lu_env *env, struct lu_server_fld *fld,
328 struct dt_device *dt, int type)
330 struct dt_object *dt_obj = NULL;
332 struct lu_attr *attr = NULL;
333 struct lu_seq_range *range = NULL;
334 struct fld_thread_info *info;
335 struct dt_object_format dof;
337 const struct dt_it_ops *iops;
342 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
343 LASSERT(info != NULL);
345 lu_local_obj_fid(&fid, FLD_INDEX_OID);
350 memset(attr, 0, sizeof(*attr));
351 attr->la_valid = LA_MODE;
352 attr->la_mode = S_IFREG | 0666;
353 dof.dof_type = DFT_INDEX;
354 dof.u.dof_idx.di_feat = &fld_index_features;
356 dt_obj = dt_locate(env, dt, &fid);
357 if (IS_ERR(dt_obj)) {
358 rc = PTR_ERR(dt_obj);
363 LASSERT(dt_obj != NULL);
364 if (!dt_object_exists(dt_obj)) {
365 lu_object_put(env, &dt_obj->do_lu);
366 dt_obj = dt_find_or_create(env, dt, &fid, &dof, attr);
368 if (IS_ERR(dt_obj)) {
369 rc = PTR_ERR(dt_obj);
370 CERROR("%s: Can't find \"%s\" obj %d\n", fld->lsf_name,
377 fld->lsf_obj = dt_obj;
378 rc = dt_obj->do_ops->do_index_try(env, dt_obj, &fld_index_features);
380 CERROR("%s: File \"%s\" is not an index: rc = %d!\n",
381 fld->lsf_name, fld_index_name, rc);
385 range = &info->fti_rec;
386 /* Load fld entry to cache */
387 iops = &dt_obj->do_index_ops->dio_it;
388 it = iops->init(env, dt_obj, 0, NULL);
390 GOTO(out, rc = PTR_ERR(it));
392 rc = iops->load(env, it, 0);
394 GOTO(out_it_fini, rc);
397 /* Load FLD entry into server cache */
399 rc = iops->rec(env, it, (struct dt_rec *)range, 0);
401 GOTO(out_it_put, rc);
402 LASSERT(range != NULL);
403 range_be_to_cpu(range, range);
404 rc = fld_cache_insert(fld->lsf_cache, range);
406 GOTO(out_it_put, rc);
407 rc = iops->next(env, it);
413 rc = fld_name_to_index(fld->lsf_name, &index);
415 GOTO(out_it_put, rc);
419 if (index == 0 && type == LU_SEQ_RANGE_MDT) {
420 /* Note: fld_insert_entry will detect whether these
421 * special entries already exist inside FLDB */
422 mutex_lock(&fld->lsf_lock);
423 rc = fld_insert_special_entries(env, fld);
424 mutex_unlock(&fld->lsf_lock);
426 CERROR("%s: insert special entries failed!: rc = %d\n",
428 GOTO(out_it_put, rc);
441 lu_object_put(env, &dt_obj->do_lu);
447 void fld_index_fini(const struct lu_env *env, struct lu_server_fld *fld)
450 if (fld->lsf_obj != NULL) {
451 if (!IS_ERR(fld->lsf_obj))
452 lu_object_put(env, &fld->lsf_obj->do_lu);
458 int fld_server_read(const struct lu_env *env, struct lu_server_fld *fld,
459 struct lu_seq_range *range, void *data, int data_len)
461 struct lu_seq_range_array *lsra = data;
462 struct fld_thread_info *info;
463 struct dt_object *dt_obj = fld->lsf_obj;
464 struct lu_seq_range *entry;
466 const struct dt_it_ops *iops;
471 lsra->lsra_count = 0;
472 iops = &dt_obj->do_index_ops->dio_it;
473 it = iops->init(env, dt_obj, 0, NULL);
477 rc = iops->load(env, it, range->lsr_end);
479 GOTO(out_it_fini, rc);
481 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
482 LASSERT(info != NULL);
483 entry = &info->fti_rec;
485 rc = iops->rec(env, it, (struct dt_rec *)entry, 0);
487 GOTO(out_it_put, rc);
489 if (offsetof(typeof(*lsra), lsra_lsr[lsra->lsra_count + 1]) >
491 GOTO(out, rc = -EAGAIN);
493 range_be_to_cpu(entry, entry);
494 if (entry->lsr_index == range->lsr_index &&
495 entry->lsr_flags == range->lsr_flags &&
496 entry->lsr_start > range->lsr_start) {
497 lsra->lsra_lsr[lsra->lsra_count] = *entry;
501 rc = iops->next(env, it);
506 range_array_cpu_to_le(lsra, lsra);