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 "../../include/linux/libcfs/libcfs.h"
  42 #include <linux/module.h>
  43 #include <asm/div64.h>
  44
  45 #include "../include/obd.h"
  46 #include "../include/obd_class.h"
  47 #include "../include/lustre_ver.h"
  48 #include "../include/obd_support.h"
  49 #include "../include/lprocfs_status.h"
  50
  51 #include "../include/lustre_req_layout.h"
  52 #include "../include/lustre_fld.h"
  53 #include "../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         if (fld->lcf_flags != LUSTRE_FLD_INIT) {
 163                 CERROR("%s: Attempt to add target %s (idx %llu) on fly - skip it\n",
 164                        fld->lcf_name, name, tar->ft_idx);
 165                 return 0;
 166         }
 167         CDEBUG(D_INFO, "%s: Adding target %s (idx %llu)\n",
 168                fld->lcf_name, name, tar->ft_idx);
 169
 170         target = kzalloc(sizeof(*target), GFP_NOFS);
 171         if (!target)
 172                 return -ENOMEM;
 173
 174         spin_lock(&fld->lcf_lock);
 175         list_for_each_entry(tmp, &fld->lcf_targets, ft_chain) {
 176                 if (tmp->ft_idx == tar->ft_idx) {
 177                         spin_unlock(&fld->lcf_lock);
 178                         kfree(target);
 179                         CERROR("Target %s exists in FLD and known as %s:#%llu\n",
 180                                name, fld_target_name(tmp), tmp->ft_idx);
 181                         return -EEXIST;
 182                 }
 183         }
 184
 185         target->ft_exp = tar->ft_exp;
 186         if (target->ft_exp)
 187                 class_export_get(target->ft_exp);
 188         target->ft_srv = tar->ft_srv;
 189         target->ft_idx = tar->ft_idx;
 190
 191         list_add_tail(&target->ft_chain, &fld->lcf_targets);
 192
 193         fld->lcf_count++;
 194         spin_unlock(&fld->lcf_lock);
 195
 196         return 0;
 197 }
 198 EXPORT_SYMBOL(fld_client_add_target);
 199
 200 /* Remove export from FLD */
 201 int fld_client_del_target(struct lu_client_fld *fld, __u64 idx)
 202 {
 203         struct lu_fld_target *target, *tmp;
 204
 205         spin_lock(&fld->lcf_lock);
 206         list_for_each_entry_safe(target, tmp, &fld->lcf_targets, ft_chain) {
 207                 if (target->ft_idx == idx) {
 208                         fld->lcf_count--;
 209                         list_del(&target->ft_chain);
 210                         spin_unlock(&fld->lcf_lock);
 211
 212                         if (target->ft_exp)
 213                                 class_export_put(target->ft_exp);
 214
 215                         kfree(target);
 216                         return 0;
 217                 }
 218         }
 219         spin_unlock(&fld->lcf_lock);
 220         return -ENOENT;
 221 }
 222
 223 static struct dentry *fld_debugfs_dir;
 224
 225 static int fld_client_debugfs_init(struct lu_client_fld *fld)
 226 {
 227         int rc;
 228
 229         fld->lcf_debugfs_entry = ldebugfs_register(fld->lcf_name,
 230                                                    fld_debugfs_dir,
 231                                                    NULL, NULL);
 232
 233         if (IS_ERR_OR_NULL(fld->lcf_debugfs_entry)) {
 234                 CERROR("%s: LdebugFS failed in fld-init\n", fld->lcf_name);
 235                 rc = fld->lcf_debugfs_entry ? PTR_ERR(fld->lcf_debugfs_entry)
 236                                             : -ENOMEM;
 237                 fld->lcf_debugfs_entry = NULL;
 238                 return rc;
 239         }
 240
 241         rc = ldebugfs_add_vars(fld->lcf_debugfs_entry,
 242                                fld_client_debugfs_list, fld);
 243         if (rc) {
 244                 CERROR("%s: Can't init FLD debufs, rc %d\n", fld->lcf_name, rc);
 245                 goto out_cleanup;
 246         }
 247
 248         return 0;
 249
 250 out_cleanup:
 251         fld_client_debugfs_fini(fld);
 252         return rc;
 253 }
 254
 255 void fld_client_debugfs_fini(struct lu_client_fld *fld)
 256 {
 257         if (!IS_ERR_OR_NULL(fld->lcf_debugfs_entry))
 258                 ldebugfs_remove(&fld->lcf_debugfs_entry);
 259 }
 260 EXPORT_SYMBOL(fld_client_debugfs_fini);
 261
 262 static inline int hash_is_sane(int hash)
 263 {
 264         return (hash >= 0 && hash < ARRAY_SIZE(fld_hash));
 265 }
 266
 267 int fld_client_init(struct lu_client_fld *fld,
 268                     const char *prefix, int hash)
 269 {
 270         int cache_size, cache_threshold;
 271         int rc;
 272
 273         snprintf(fld->lcf_name, sizeof(fld->lcf_name),
 274                  "cli-%s", prefix);
 275
 276         if (!hash_is_sane(hash)) {
 277                 CERROR("%s: Wrong hash function %#x\n",
 278                        fld->lcf_name, hash);
 279                 return -EINVAL;
 280         }
 281
 282         fld->lcf_count = 0;
 283         spin_lock_init(&fld->lcf_lock);
 284         fld->lcf_hash = &fld_hash[hash];
 285         fld->lcf_flags = LUSTRE_FLD_INIT;
 286         INIT_LIST_HEAD(&fld->lcf_targets);
 287
 288         cache_size = FLD_CLIENT_CACHE_SIZE /
 289                 sizeof(struct fld_cache_entry);
 290
 291         cache_threshold = cache_size *
 292                 FLD_CLIENT_CACHE_THRESHOLD / 100;
 293
 294         fld->lcf_cache = fld_cache_init(fld->lcf_name,
 295                                         cache_size, cache_threshold);
 296         if (IS_ERR(fld->lcf_cache)) {
 297                 rc = PTR_ERR(fld->lcf_cache);
 298                 fld->lcf_cache = NULL;
 299                 goto out;
 300         }
 301
 302         rc = fld_client_debugfs_init(fld);
 303         if (rc)
 304                 goto out;
 305 out:
 306         if (rc)
 307                 fld_client_fini(fld);
 308         else
 309                 CDEBUG(D_INFO, "%s: Using \"%s\" hash\n",
 310                        fld->lcf_name, fld->lcf_hash->fh_name);
 311         return rc;
 312 }
 313 EXPORT_SYMBOL(fld_client_init);
 314
 315 void fld_client_fini(struct lu_client_fld *fld)
 316 {
 317         struct lu_fld_target *target, *tmp;
 318
 319         spin_lock(&fld->lcf_lock);
 320         list_for_each_entry_safe(target, tmp, &fld->lcf_targets, ft_chain) {
 321                 fld->lcf_count--;
 322                 list_del(&target->ft_chain);
 323                 if (target->ft_exp)
 324                         class_export_put(target->ft_exp);
 325                 kfree(target);
 326         }
 327         spin_unlock(&fld->lcf_lock);
 328
 329         if (fld->lcf_cache) {
 330                 if (!IS_ERR(fld->lcf_cache))
 331                         fld_cache_fini(fld->lcf_cache);
 332                 fld->lcf_cache = NULL;
 333         }
 334 }
 335 EXPORT_SYMBOL(fld_client_fini);
 336
 337 int fld_client_rpc(struct obd_export *exp,
 338                    struct lu_seq_range *range, __u32 fld_op,
 339                    struct ptlrpc_request **reqp)
 340 {
 341         struct ptlrpc_request *req = NULL;
 342         struct lu_seq_range   *prange;
 343         __u32            *op;
 344         int                 rc = 0;
 345         struct obd_import     *imp;
 346
 347         LASSERT(exp);
 348
 349         imp = class_exp2cliimp(exp);
 350         switch (fld_op) {
 351         case FLD_QUERY:
 352                 req = ptlrpc_request_alloc_pack(imp, &RQF_FLD_QUERY,
 353                                                 LUSTRE_MDS_VERSION, FLD_QUERY);
 354                 if (!req)
 355                         return -ENOMEM;
 356
 357                 /*
 358                  * XXX: only needed when talking to old server(< 2.6), it should
 359                  * be removed when < 2.6 server is not supported
 360                  */
 361                 op = req_capsule_client_get(&req->rq_pill, &RMF_FLD_OPC);
 362                 *op = FLD_LOOKUP;
 363
 364                 if (imp->imp_connect_flags_orig & OBD_CONNECT_MDS_MDS)
 365                         req->rq_allow_replay = 1;
 366                 break;
 367         case FLD_READ:
 368                 req = ptlrpc_request_alloc_pack(imp, &RQF_FLD_READ,
 369                                                 LUSTRE_MDS_VERSION, FLD_READ);
 370                 if (!req)
 371                         return -ENOMEM;
 372
 373                 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA,
 374                                      RCL_SERVER, PAGE_SIZE);
 375                 break;
 376         default:
 377                 rc = -EINVAL;
 378                 break;
 379         }
 380         if (rc)
 381                 return rc;
 382
 383         prange = req_capsule_client_get(&req->rq_pill, &RMF_FLD_MDFLD);
 384         *prange = *range;
 385         ptlrpc_request_set_replen(req);
 386         req->rq_request_portal = FLD_REQUEST_PORTAL;
 387         req->rq_reply_portal = MDC_REPLY_PORTAL;
 388         ptlrpc_at_set_req_timeout(req);
 389
 390         obd_get_request_slot(&exp->exp_obd->u.cli);
 391         rc = ptlrpc_queue_wait(req);
 392         obd_put_request_slot(&exp->exp_obd->u.cli);
 393         if (rc)
 394                 goto out_req;
 395
 396         if (fld_op == FLD_QUERY) {
 397                 prange = req_capsule_server_get(&req->rq_pill, &RMF_FLD_MDFLD);
 398                 if (!prange) {
 399                         rc = -EFAULT;
 400                         goto out_req;
 401                 }
 402                 *range = *prange;
 403         }
 404
 405 out_req:
 406         if (rc || !reqp) {
 407                 ptlrpc_req_finished(req);
 408                 req = NULL;
 409         }
 410
 411         if (reqp)
 412                 *reqp = req;
 413
 414         return rc;
 415 }
 416
 417 int fld_client_lookup(struct lu_client_fld *fld, u64 seq, u32 *mds,
 418                       __u32 flags, const struct lu_env *env)
 419 {
 420         struct lu_seq_range res = { 0 };
 421         struct lu_fld_target *target;
 422         int rc;
 423
 424         fld->lcf_flags |= LUSTRE_FLD_RUN;
 425
 426         rc = fld_cache_lookup(fld->lcf_cache, seq, &res);
 427         if (rc == 0) {
 428                 *mds = res.lsr_index;
 429                 return 0;
 430         }
 431
 432         /* Can not find it in the cache */
 433         target = fld_client_get_target(fld, seq);
 434         LASSERT(target);
 435
 436         CDEBUG(D_INFO, "%s: Lookup fld entry (seq: %#llx) on target %s (idx %llu)\n",
 437                fld->lcf_name, seq, fld_target_name(target), target->ft_idx);
 438
 439         res.lsr_start = seq;
 440         fld_range_set_type(&res, flags);
 441         rc = fld_client_rpc(target->ft_exp, &res, FLD_QUERY, NULL);
 442
 443         if (rc == 0) {
 444                 *mds = res.lsr_index;
 445
 446                 fld_cache_insert(fld->lcf_cache, &res);
 447         }
 448         return rc;
 449 }
 450 EXPORT_SYMBOL(fld_client_lookup);
 451
 452 void fld_client_flush(struct lu_client_fld *fld)
 453 {
 454         fld_cache_flush(fld->lcf_cache);
 455 }
 456
 457 static int __init fld_init(void)
 458 {
 459         fld_debugfs_dir = ldebugfs_register(LUSTRE_FLD_NAME,
 460                                             debugfs_lustre_root,
 461                                             NULL, NULL);
 462         return PTR_ERR_OR_ZERO(fld_debugfs_dir);
 463 }
 464
 465 static void __exit fld_exit(void)
 466 {
 467         if (!IS_ERR_OR_NULL(fld_debugfs_dir))
 468                 ldebugfs_remove(&fld_debugfs_dir);
 469 }
 470
 471 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
 472 MODULE_DESCRIPTION("Lustre FID Location Database");
 473 MODULE_VERSION(LUSTRE_VERSION_STRING);
 474 MODULE_LICENSE("GPL");
 475
 476 module_init(fld_init)
 477 module_exit(fld_exit)