drivers/staging/lustre/lustre/fld/fld_request.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   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.
   9  *
  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).
  15  *
  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
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2015, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/fld/fld_request.c
  33  *
  34  * FLD (Fids Location Database)
  35  *
  36  * Author: Yury Umanets <umka@clusterfs.com>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_FLD
  40
  41 #include <linux/libcfs/libcfs.h>
  42 #include <linux/module.h>
  43 #include <asm/div64.h>
  44
  45 #include <obd.h>
  46 #include <obd_class.h>
  47 #include <uapi/linux/lustre/lustre_ver.h>
  48 #include <obd_support.h>
  49 #include <lprocfs_status.h>
  50
  51 #include <lustre_req_layout.h>
  52 #include <lustre_fld.h>
  53 #include <lustre_mdc.h>
  54 #include "fld_internal.h"
  55
  56 static int fld_rrb_hash(struct lu_client_fld *fld, u64 seq)
  57 {
  58         LASSERT(fld->lcf_count > 0);
  59         return do_div(seq, fld->lcf_count);
  60 }
  61
  62 static struct lu_fld_target *
  63 fld_rrb_scan(struct lu_client_fld *fld, u64 seq)
  64 {
  65         struct lu_fld_target *target;
  66         int hash;
  67
  68         /* Because almost all of special sequence located in MDT0,
  69          * it should go to index 0 directly, instead of calculating
  70          * hash again, and also if other MDTs is not being connected,
  71          * the fld lookup requests(for seq on MDT0) should not be
  72          * blocked because of other MDTs
  73          */
  74         if (fid_seq_is_norm(seq))
  75                 hash = fld_rrb_hash(fld, seq);
  76         else
  77                 hash = 0;
  78
  79 again:
  80         list_for_each_entry(target, &fld->lcf_targets, ft_chain) {
  81                 if (target->ft_idx == hash)
  82                         return target;
  83         }
  84
  85         if (hash != 0) {
  86                 /* It is possible the remote target(MDT) are not connected to
  87                  * with client yet, so we will refer this to MDT0, which should
  88                  * be connected during mount
  89                  */
  90                 hash = 0;
  91                 goto again;
  92         }
  93
  94         CERROR("%s: Can't find target by hash %d (seq %#llx). Targets (%d):\n",
  95                fld->lcf_name, hash, seq, fld->lcf_count);
  96
  97         list_for_each_entry(target, &fld->lcf_targets, ft_chain) {
  98                 const char *srv_name = target->ft_srv ?
  99                         target->ft_srv->lsf_name : "<null>";
 100                 const char *exp_name = target->ft_exp ?
 101                         (char *)target->ft_exp->exp_obd->obd_uuid.uuid :
 102                         "<null>";
 103
 104                 CERROR("  exp: 0x%p (%s), srv: 0x%p (%s), idx: %llu\n",
 105                        target->ft_exp, exp_name, target->ft_srv,
 106                        srv_name, target->ft_idx);
 107         }
 108
 109         /*
 110          * If target is not found, there is logical error anyway, so here is
 111          * LBUG() to catch this situation.
 112          */
 113         LBUG();
 114         return NULL;
 115 }
 116
 117 struct lu_fld_hash fld_hash[] = {
 118         {
 119                 .fh_name = "RRB",
 120                 .fh_hash_func = fld_rrb_hash,
 121                 .fh_scan_func = fld_rrb_scan
 122         },
 123         {
 124                 NULL,
 125         }
 126 };
 127
 128 static struct lu_fld_target *
 129 fld_client_get_target(struct lu_client_fld *fld, u64 seq)
 130 {
 131         struct lu_fld_target *target;
 132
 133         LASSERT(fld->lcf_hash);
 134
 135         spin_lock(&fld->lcf_lock);
 136         target = fld->lcf_hash->fh_scan_func(fld, seq);
 137         spin_unlock(&fld->lcf_lock);
 138
 139         if (target) {
 140                 CDEBUG(D_INFO, "%s: Found target (idx %llu) by seq %#llx\n",
 141                        fld->lcf_name, target->ft_idx, seq);
 142         }
 143
 144         return target;
 145 }
 146
 147 /*
 148  * Add export to FLD. This is usually done by CMM and LMV as they are main users
 149  * of FLD module.
 150  */
 151 int fld_client_add_target(struct lu_client_fld *fld,
 152                           struct lu_fld_target *tar)
 153 {
 154         const char *name;
 155         struct lu_fld_target *target, *tmp;
 156
 157         LASSERT(tar);
 158         name = fld_target_name(tar);
 159         LASSERT(name);
 160         LASSERT(tar->ft_srv || tar->ft_exp);
 161
 162         CDEBUG(D_INFO, "%s: Adding target %s (idx %llu)\n",
 163                fld->lcf_name, name, tar->ft_idx);
 164
 165         target = kzalloc(sizeof(*target), GFP_NOFS);
 166         if (!target)
 167                 return -ENOMEM;
 168
 169         spin_lock(&fld->lcf_lock);
 170         list_for_each_entry(tmp, &fld->lcf_targets, ft_chain) {
 171                 if (tmp->ft_idx == tar->ft_idx) {
 172                         spin_unlock(&fld->lcf_lock);
 173                         kfree(target);
 174                         CERROR("Target %s exists in FLD and known as %s:#%llu\n",
 175                                name, fld_target_name(tmp), tmp->ft_idx);
 176                         return -EEXIST;
 177                 }
 178         }
 179
 180         target->ft_exp = tar->ft_exp;
 181         if (target->ft_exp)
 182                 class_export_get(target->ft_exp);
 183         target->ft_srv = tar->ft_srv;
 184         target->ft_idx = tar->ft_idx;
 185
 186         list_add_tail(&target->ft_chain, &fld->lcf_targets);
 187
 188         fld->lcf_count++;
 189         spin_unlock(&fld->lcf_lock);
 190
 191         return 0;
 192 }
 193 EXPORT_SYMBOL(fld_client_add_target);
 194
 195 /* Remove export from FLD */
 196 int fld_client_del_target(struct lu_client_fld *fld, __u64 idx)
 197 {
 198         struct lu_fld_target *target, *tmp;
 199
 200         spin_lock(&fld->lcf_lock);
 201         list_for_each_entry_safe(target, tmp, &fld->lcf_targets, ft_chain) {
 202                 if (target->ft_idx == idx) {
 203                         fld->lcf_count--;
 204                         list_del(&target->ft_chain);
 205                         spin_unlock(&fld->lcf_lock);
 206
 207                         if (target->ft_exp)
 208                                 class_export_put(target->ft_exp);
 209
 210                         kfree(target);
 211                         return 0;
 212                 }
 213         }
 214         spin_unlock(&fld->lcf_lock);
 215         return -ENOENT;
 216 }
 217
 218 static struct dentry *fld_debugfs_dir;
 219
 220 static int fld_client_debugfs_init(struct lu_client_fld *fld)
 221 {
 222         int rc;
 223
 224         fld->lcf_debugfs_entry = ldebugfs_register(fld->lcf_name,
 225                                                    fld_debugfs_dir,
 226                                                    NULL, NULL);
 227
 228         if (IS_ERR_OR_NULL(fld->lcf_debugfs_entry)) {
 229                 CERROR("%s: LdebugFS failed in fld-init\n", fld->lcf_name);
 230                 rc = fld->lcf_debugfs_entry ? PTR_ERR(fld->lcf_debugfs_entry)
 231                                             : -ENOMEM;
 232                 fld->lcf_debugfs_entry = NULL;
 233                 return rc;
 234         }
 235
 236         rc = ldebugfs_add_vars(fld->lcf_debugfs_entry,
 237                                fld_client_debugfs_list, fld);
 238         if (rc) {
 239                 CERROR("%s: Can't init FLD debufs, rc %d\n", fld->lcf_name, rc);
 240                 goto out_cleanup;
 241         }
 242
 243         return 0;
 244
 245 out_cleanup:
 246         fld_client_debugfs_fini(fld);
 247         return rc;
 248 }
 249
 250 void fld_client_debugfs_fini(struct lu_client_fld *fld)
 251 {
 252         if (!IS_ERR_OR_NULL(fld->lcf_debugfs_entry))
 253                 ldebugfs_remove(&fld->lcf_debugfs_entry);
 254 }
 255 EXPORT_SYMBOL(fld_client_debugfs_fini);
 256
 257 static inline int hash_is_sane(int hash)
 258 {
 259         return (hash >= 0 && hash < ARRAY_SIZE(fld_hash));
 260 }
 261
 262 int fld_client_init(struct lu_client_fld *fld,
 263                     const char *prefix, int hash)
 264 {
 265         int cache_size, cache_threshold;
 266         int rc;
 267
 268         snprintf(fld->lcf_name, sizeof(fld->lcf_name),
 269                  "cli-%s", prefix);
 270
 271         if (!hash_is_sane(hash)) {
 272                 CERROR("%s: Wrong hash function %#x\n",
 273                        fld->lcf_name, hash);
 274                 return -EINVAL;
 275         }
 276
 277         fld->lcf_count = 0;
 278         spin_lock_init(&fld->lcf_lock);
 279         fld->lcf_hash = &fld_hash[hash];
 280         INIT_LIST_HEAD(&fld->lcf_targets);
 281
 282         cache_size = FLD_CLIENT_CACHE_SIZE /
 283                 sizeof(struct fld_cache_entry);
 284
 285         cache_threshold = cache_size *
 286                 FLD_CLIENT_CACHE_THRESHOLD / 100;
 287
 288         fld->lcf_cache = fld_cache_init(fld->lcf_name,
 289                                         cache_size, cache_threshold);
 290         if (IS_ERR(fld->lcf_cache)) {
 291                 rc = PTR_ERR(fld->lcf_cache);
 292                 fld->lcf_cache = NULL;
 293                 goto out;
 294         }
 295
 296         rc = fld_client_debugfs_init(fld);
 297         if (rc)
 298                 goto out;
 299 out:
 300         if (rc)
 301                 fld_client_fini(fld);
 302         else
 303                 CDEBUG(D_INFO, "%s: Using \"%s\" hash\n",
 304                        fld->lcf_name, fld->lcf_hash->fh_name);
 305         return rc;
 306 }
 307 EXPORT_SYMBOL(fld_client_init);
 308
 309 void fld_client_fini(struct lu_client_fld *fld)
 310 {
 311         struct lu_fld_target *target, *tmp;
 312
 313         spin_lock(&fld->lcf_lock);
 314         list_for_each_entry_safe(target, tmp, &fld->lcf_targets, ft_chain) {
 315                 fld->lcf_count--;
 316                 list_del(&target->ft_chain);
 317                 if (target->ft_exp)
 318                         class_export_put(target->ft_exp);
 319                 kfree(target);
 320         }
 321         spin_unlock(&fld->lcf_lock);
 322
 323         if (fld->lcf_cache) {
 324                 if (!IS_ERR(fld->lcf_cache))
 325                         fld_cache_fini(fld->lcf_cache);
 326                 fld->lcf_cache = NULL;
 327         }
 328 }
 329 EXPORT_SYMBOL(fld_client_fini);
 330
 331 int fld_client_rpc(struct obd_export *exp,
 332                    struct lu_seq_range *range, __u32 fld_op,
 333                    struct ptlrpc_request **reqp)
 334 {
 335         struct ptlrpc_request *req = NULL;
 336         struct lu_seq_range   *prange;
 337         __u32            *op;
 338         int                 rc = 0;
 339         struct obd_import     *imp;
 340
 341         LASSERT(exp);
 342
 343         imp = class_exp2cliimp(exp);
 344         switch (fld_op) {
 345         case FLD_QUERY:
 346                 req = ptlrpc_request_alloc_pack(imp, &RQF_FLD_QUERY,
 347                                                 LUSTRE_MDS_VERSION, FLD_QUERY);
 348                 if (!req)
 349                         return -ENOMEM;
 350
 351                 /*
 352                  * XXX: only needed when talking to old server(< 2.6), it should
 353                  * be removed when < 2.6 server is not supported
 354                  */
 355                 op = req_capsule_client_get(&req->rq_pill, &RMF_FLD_OPC);
 356                 *op = FLD_LOOKUP;
 357
 358                 if (imp->imp_connect_flags_orig & OBD_CONNECT_MDS_MDS)
 359                         req->rq_allow_replay = 1;
 360                 break;
 361         case FLD_READ:
 362                 req = ptlrpc_request_alloc_pack(imp, &RQF_FLD_READ,
 363                                                 LUSTRE_MDS_VERSION, FLD_READ);
 364                 if (!req)
 365                         return -ENOMEM;
 366
 367                 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA,
 368                                      RCL_SERVER, PAGE_SIZE);
 369                 break;
 370         default:
 371                 rc = -EINVAL;
 372                 break;
 373         }
 374         if (rc)
 375                 return rc;
 376
 377         prange = req_capsule_client_get(&req->rq_pill, &RMF_FLD_MDFLD);
 378         *prange = *range;
 379         ptlrpc_request_set_replen(req);
 380         req->rq_request_portal = FLD_REQUEST_PORTAL;
 381         req->rq_reply_portal = MDC_REPLY_PORTAL;
 382         ptlrpc_at_set_req_timeout(req);
 383
 384         obd_get_request_slot(&exp->exp_obd->u.cli);
 385         rc = ptlrpc_queue_wait(req);
 386         obd_put_request_slot(&exp->exp_obd->u.cli);
 387         if (rc)
 388                 goto out_req;
 389
 390         if (fld_op == FLD_QUERY) {
 391                 prange = req_capsule_server_get(&req->rq_pill, &RMF_FLD_MDFLD);
 392                 if (!prange) {
 393                         rc = -EFAULT;
 394                         goto out_req;
 395                 }
 396                 *range = *prange;
 397         }
 398
 399 out_req:
 400         if (rc || !reqp) {
 401                 ptlrpc_req_finished(req);
 402                 req = NULL;
 403         }
 404
 405         if (reqp)
 406                 *reqp = req;
 407
 408         return rc;
 409 }
 410
 411 int fld_client_lookup(struct lu_client_fld *fld, u64 seq, u32 *mds,
 412                       __u32 flags, const struct lu_env *env)
 413 {
 414         struct lu_seq_range res = { 0 };
 415         struct lu_fld_target *target;
 416         int rc;
 417
 418         rc = fld_cache_lookup(fld->lcf_cache, seq, &res);
 419         if (rc == 0) {
 420                 *mds = res.lsr_index;
 421                 return 0;
 422         }
 423
 424         /* Can not find it in the cache */
 425         target = fld_client_get_target(fld, seq);
 426         LASSERT(target);
 427
 428         CDEBUG(D_INFO, "%s: Lookup fld entry (seq: %#llx) on target %s (idx %llu)\n",
 429                fld->lcf_name, seq, fld_target_name(target), target->ft_idx);
 430
 431         res.lsr_start = seq;
 432         fld_range_set_type(&res, flags);
 433         rc = fld_client_rpc(target->ft_exp, &res, FLD_QUERY, NULL);
 434
 435         if (rc == 0) {
 436                 *mds = res.lsr_index;
 437
 438                 fld_cache_insert(fld->lcf_cache, &res);
 439         }
 440         return rc;
 441 }
 442 EXPORT_SYMBOL(fld_client_lookup);
 443
 444 void fld_client_flush(struct lu_client_fld *fld)
 445 {
 446         fld_cache_flush(fld->lcf_cache);
 447 }
 448
 449 static int __init fld_init(void)
 450 {
 451         fld_debugfs_dir = ldebugfs_register(LUSTRE_FLD_NAME,
 452                                             debugfs_lustre_root,
 453                                             NULL, NULL);
 454         return PTR_ERR_OR_ZERO(fld_debugfs_dir);
 455 }
 456
 457 static void __exit fld_exit(void)
 458 {
 459         if (!IS_ERR_OR_NULL(fld_debugfs_dir))
 460                 ldebugfs_remove(&fld_debugfs_dir);
 461 }
 462
 463 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
 464 MODULE_DESCRIPTION("Lustre FID Location Database");
 465 MODULE_VERSION(LUSTRE_VERSION_STRING);
 466 MODULE_LICENSE("GPL");
 467
 468 module_init(fld_init)
 469 module_exit(fld_exit)