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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/fld/fld_index.c
34 * Author: WangDi <wangdi@clusterfs.com>
35 * Author: Yury Umanets <umka@clusterfs.com>
38 #define DEBUG_SUBSYSTEM S_FLD
40 #include <libcfs/libcfs.h>
41 #include <linux/module.h>
42 #include <obd_support.h>
43 #include <dt_object.h>
44 #include <lustre_fid.h>
45 #include <lustre_fld.h>
46 #include "fld_internal.h"
48 static const char fld_index_name[] = "fld";
50 static const struct lu_seq_range IGIF_FLD_RANGE = {
51 .lsr_start = FID_SEQ_IGIF,
52 .lsr_end = FID_SEQ_IGIF_MAX + 1,
54 .lsr_flags = LU_SEQ_RANGE_MDT
57 static const struct lu_seq_range DOT_LUSTRE_FLD_RANGE = {
58 .lsr_start = FID_SEQ_DOT_LUSTRE,
59 .lsr_end = FID_SEQ_DOT_LUSTRE + 1,
61 .lsr_flags = LU_SEQ_RANGE_MDT
64 static const struct lu_seq_range ROOT_FLD_RANGE = {
65 .lsr_start = FID_SEQ_ROOT,
66 .lsr_end = FID_SEQ_ROOT + 1,
68 .lsr_flags = LU_SEQ_RANGE_MDT
71 static const struct dt_index_features fld_index_features = {
72 .dif_flags = DT_IND_UPDATE,
73 .dif_keysize_min = sizeof(u64),
74 .dif_keysize_max = sizeof(u64),
75 .dif_recsize_min = sizeof(struct lu_seq_range),
76 .dif_recsize_max = sizeof(struct lu_seq_range),
80 extern struct lu_context_key fld_thread_key;
82 int fld_declare_index_create(const struct lu_env *env,
83 struct lu_server_fld *fld,
84 const struct lu_seq_range *new_range,
87 struct lu_seq_range *tmp;
88 struct lu_seq_range *range;
89 struct fld_thread_info *info;
94 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
95 range = &info->fti_lrange;
96 tmp = &info->fti_irange;
97 memset(range, 0, sizeof(*range));
99 rc = fld_index_lookup(env, fld, new_range->lsr_start, range);
101 /* In case of duplicate entry, the location must be same */
102 LASSERT((lu_seq_range_compare_loc(new_range, range) == 0));
103 GOTO(out, rc = -EEXIST);
107 CERROR("%s: lookup range "DRANGE" error: rc = %d\n",
108 fld->lsf_name, PRANGE(range), rc);
113 * Check for merge case, since the fld entry can only be increamental,
114 * so we will only check whether it can be merged from the left.
116 if (new_range->lsr_start == range->lsr_end && range->lsr_end != 0 &&
117 lu_seq_range_compare_loc(new_range, range) == 0) {
118 range_cpu_to_be(tmp, range);
119 rc = dt_declare_delete(env, fld->lsf_obj,
120 (struct dt_key *)&tmp->lsr_start, th);
122 CERROR("%s: declare record "DRANGE" failed: rc = %d\n",
123 fld->lsf_name, PRANGE(range), rc);
127 tmp->lsr_start = range->lsr_start;
132 range_cpu_to_be(tmp, tmp);
133 rc = dt_declare_insert(env, fld->lsf_obj, (struct dt_rec *)tmp,
134 (struct dt_key *)&tmp->lsr_start, th);
140 * insert range in fld store.
142 * \param range range to be inserted
143 * \param th transaction for this operation as it could compound
149 * The whole fld index insertion is protected by seq->lss_mutex (see
150 * seq_server_alloc_super), i.e. only one thread will access fldb each
151 * time, so we do not need worry the fld file and cache will being
152 * changed between declare and create.
153 * Because the fld entry can only be increamental, so we will only check
154 * whether it can be merged from the left.
156 * Caller must hold fld->lsf_lock
158 int fld_index_create(const struct lu_env *env, struct lu_server_fld *fld,
159 const struct lu_seq_range *new_range, struct thandle *th)
161 struct lu_seq_range *range;
162 struct lu_seq_range *tmp;
163 struct fld_thread_info *info;
166 struct fld_cache_entry *flde;
170 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
172 LASSERT(mutex_is_locked(&fld->lsf_lock));
174 range = &info->fti_lrange;
175 memset(range, 0, sizeof(*range));
176 tmp = &info->fti_irange;
177 rc = fld_index_lookup(env, fld, new_range->lsr_start, range);
179 rc = rc == 0 ? -EEXIST : rc;
183 if (new_range->lsr_start == range->lsr_end && range->lsr_end != 0 &&
184 lu_seq_range_compare_loc(new_range, range) == 0) {
185 range_cpu_to_be(tmp, range);
186 rc = dt_delete(env, fld->lsf_obj,
187 (struct dt_key *)&tmp->lsr_start, th);
191 tmp->lsr_start = range->lsr_start;
197 range_cpu_to_be(tmp, tmp);
198 rc = dt_insert(env, fld->lsf_obj, (struct dt_rec *)tmp,
199 (struct dt_key *)&tmp->lsr_start, th);
201 CERROR("%s: insert range "DRANGE" failed: rc = %d\n",
202 fld->lsf_name, PRANGE(new_range), rc);
206 flde = fld_cache_entry_create(new_range);
208 GOTO(out, rc = PTR_ERR(flde));
210 write_lock(&fld->lsf_cache->fci_lock);
212 fld_cache_delete_nolock(fld->lsf_cache, new_range);
213 rc = fld_cache_insert_nolock(fld->lsf_cache, flde);
214 write_unlock(&fld->lsf_cache->fci_lock);
222 * lookup range for a seq passed. note here we only care about the start/end,
223 * caller should handle the attached location data (flags, index).
225 * \param seq seq for lookup.
226 * \param range result of lookup.
228 * \retval 0 found, \a range is the matched range;
229 * \retval -ENOENT not found, \a range is the left-side range;
230 * \retval -ve other error;
232 int fld_index_lookup(const struct lu_env *env, struct lu_server_fld *fld,
233 u64 seq, struct lu_seq_range *range)
235 struct lu_seq_range *fld_rec;
236 struct fld_thread_info *info;
241 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
242 fld_rec = &info->fti_rec;
244 rc = fld_cache_lookup(fld->lsf_cache, seq, fld_rec);
247 if (lu_seq_range_within(range, seq))
253 CDEBUG(D_INFO, "%s: lookup seq = %#llx range : "DRANGE" rc = %d\n",
254 fld->lsf_name, seq, PRANGE(range), rc);
260 * insert entry in fld store.
262 * \param env relevant lu_env
263 * \param fld fld store
264 * \param range range to be inserted
269 * Caller must hold fld->lsf_lock
272 int fld_insert_entry(const struct lu_env *env,
273 struct lu_server_fld *fld,
274 const struct lu_seq_range *range)
277 struct dt_device *dt = lu2dt_dev(fld->lsf_obj->do_lu.lo_dev);
282 LASSERT(mutex_is_locked(&fld->lsf_lock));
287 th = dt_trans_create(env, dt);
291 rc = fld_declare_index_create(env, fld, range, th);
298 rc = dt_trans_start_local(env, dt, th);
302 rc = fld_index_create(env, fld, range, th);
306 dt_trans_stop(env, dt, th);
309 EXPORT_SYMBOL(fld_insert_entry);
311 static int fld_insert_special_entries(const struct lu_env *env,
312 struct lu_server_fld *fld)
316 rc = fld_insert_entry(env, fld, &IGIF_FLD_RANGE);
320 rc = fld_insert_entry(env, fld, &DOT_LUSTRE_FLD_RANGE);
324 rc = fld_insert_entry(env, fld, &ROOT_FLD_RANGE);
329 int fld_index_init(const struct lu_env *env, struct lu_server_fld *fld,
330 struct dt_device *dt, int type)
332 struct dt_object *dt_obj = NULL;
334 struct lu_attr *attr = NULL;
335 struct lu_seq_range *range = NULL;
336 struct fld_thread_info *info;
337 struct dt_object_format dof;
339 const struct dt_it_ops *iops;
346 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
347 LASSERT(info != NULL);
349 lu_local_obj_fid(&fid, FLD_INDEX_OID);
354 memset(attr, 0, sizeof(*attr));
355 attr->la_valid = LA_MODE;
356 attr->la_mode = S_IFREG | 0666;
357 dof.dof_type = DFT_INDEX;
358 dof.u.dof_idx.di_feat = &fld_index_features;
360 dt_obj = dt_locate(env, dt, &fid);
361 if (IS_ERR(dt_obj)) {
362 rc = PTR_ERR(dt_obj);
367 LASSERT(dt_obj != NULL);
368 if (!dt_object_exists(dt_obj)) {
369 dt_object_put(env, dt_obj);
370 dt_obj = dt_find_or_create(env, dt, &fid, &dof, attr);
372 if (IS_ERR(dt_obj)) {
373 rc = PTR_ERR(dt_obj);
374 CERROR("%s: Can't find \"%s\" obj %d\n", fld->lsf_name,
381 fld->lsf_obj = dt_obj;
382 rc = dt_obj->do_ops->do_index_try(env, dt_obj, &fld_index_features);
384 CERROR("%s: File \"%s\" is not an index: rc = %d!\n",
385 fld->lsf_name, fld_index_name, rc);
389 range = &info->fti_rec;
390 /* Load fld entry to cache */
391 iops = &dt_obj->do_index_ops->dio_it;
392 it = iops->init(env, dt_obj, 0);
394 GOTO(out, rc = PTR_ERR(it));
396 rc = iops->load(env, it, 0);
400 rc = iops->next(env, it);
403 GOTO(out_it_fini, rc);
406 rc = iops->rec(env, it, (struct dt_rec *)range, 0);
408 GOTO(out_it_put, rc);
410 range_be_to_cpu(range, range);
413 * Newly created ldiskfs IAM indexes may include a
414 * zeroed-out key and record. Ignore it here.
416 if (range->lsr_start < range->lsr_end) {
417 rc = fld_cache_insert(fld->lsf_cache, range);
419 GOTO(out_it_put, rc);
424 rc = iops->next(env, it);
426 GOTO(out_it_fini, rc);
429 if (range_count == 0)
432 rc = fld_name_to_index(fld->lsf_name, &index);
434 GOTO(out_it_put, rc);
438 if (index == 0 && type == LU_SEQ_RANGE_MDT) {
440 * Note: fld_insert_entry will detect whether these
441 * special entries already exist inside FLDB
443 mutex_lock(&fld->lsf_lock);
444 rc = fld_insert_special_entries(env, fld);
445 mutex_unlock(&fld->lsf_lock);
447 CERROR("%s: insert special entries failed!: rc = %d\n",
449 GOTO(out_it_put, rc);
462 dt_object_put(env, dt_obj);
468 void fld_index_fini(const struct lu_env *env, struct lu_server_fld *fld)
472 if (!IS_ERR(fld->lsf_obj))
473 dt_object_put(env, fld->lsf_obj);
479 int fld_server_read(const struct lu_env *env, struct lu_server_fld *fld,
480 struct lu_seq_range *range, void *data, int data_len)
482 struct lu_seq_range_array *lsra = data;
483 struct fld_thread_info *info;
484 struct dt_object *dt_obj = fld->lsf_obj;
485 struct lu_seq_range *entry;
487 const struct dt_it_ops *iops;
492 lsra->lsra_count = 0;
493 iops = &dt_obj->do_index_ops->dio_it;
494 it = iops->init(env, dt_obj, 0);
498 rc = iops->load(env, it, range->lsr_end);
500 GOTO(out_it_fini, rc);
502 info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
503 LASSERT(info != NULL);
504 entry = &info->fti_rec;
506 rc = iops->rec(env, it, (struct dt_rec *)entry, 0);
508 GOTO(out_it_put, rc);
510 if (offsetof(typeof(*lsra), lsra_lsr[lsra->lsra_count + 1]) >
512 GOTO(out, rc = -EAGAIN);
514 range_be_to_cpu(entry, entry);
515 if (entry->lsr_index == range->lsr_index &&
516 entry->lsr_flags == range->lsr_flags &&
517 entry->lsr_start > range->lsr_start) {
518 lsra->lsra_lsr[lsra->lsra_count] = *entry;
522 rc = iops->next(env, it);
527 range_array_cpu_to_le(lsra, lsra);