lustre/lod/lod_qos.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,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright  2009 Sun Microsystems, Inc. All rights reserved
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 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/lod/lod_qos.c
  33  *
  34  */
  35
  36 #define DEBUG_SUBSYSTEM S_LOV
  37
  38 #include <libcfs/libcfs.h>
  39 #include <obd_class.h>
  40 #include <obd_lov.h>
  41 #include <lustre/lustre_idl.h>
  42 #include "lod_internal.h"
  43
  44 /*
  45  * force QoS policy (not RR) to be used for testing purposes
  46  */
  47 #define FORCE_QOS_
  48
  49 #define D_QOS   D_OTHER
  50
  51 #if 0
  52 #define QOS_DEBUG(fmt, ...)     CDEBUG(D_OTHER, fmt, ## __VA_ARGS__)
  53 #define QOS_CONSOLE(fmt, ...)   LCONSOLE(D_OTHER, fmt, ## __VA_ARGS__)
  54 #else
  55 #define QOS_DEBUG(fmt, ...)
  56 #define QOS_CONSOLE(fmt, ...)
  57 #endif
  58
  59 #define TGT_BAVAIL(i) (OST_TGT(lod,i)->ltd_statfs.os_bavail * \
  60                        OST_TGT(lod,i)->ltd_statfs.os_bsize)
  61
  62 int qos_add_tgt(struct lod_device *lod, struct lod_tgt_desc *ost_desc)
  63 {
  64         struct lov_qos_oss *oss = NULL, *temposs;
  65         struct obd_export  *exp = ost_desc->ltd_exp;
  66         int                 rc = 0, found = 0;
  67         cfs_list_t         *list;
  68         ENTRY;
  69
  70         down_write(&lod->lod_qos.lq_rw_sem);
  71         /*
  72          * a bit hacky approach to learn NID of corresponding connection
  73          * but there is no official API to access information like this
  74          * with OSD API.
  75          */
  76         cfs_list_for_each_entry(oss, &lod->lod_qos.lq_oss_list, lqo_oss_list) {
  77                 if (obd_uuid_equals(&oss->lqo_uuid,
  78                                     &exp->exp_connection->c_remote_uuid)) {
  79                         found++;
  80                         break;
  81                 }
  82         }
  83
  84         if (!found) {
  85                 OBD_ALLOC_PTR(oss);
  86                 if (!oss)
  87                         GOTO(out, rc = -ENOMEM);
  88                 memcpy(&oss->lqo_uuid, &exp->exp_connection->c_remote_uuid,
  89                        sizeof(oss->lqo_uuid));
  90         } else {
  91                 /* Assume we have to move this one */
  92                 cfs_list_del(&oss->lqo_oss_list);
  93         }
  94
  95         oss->lqo_ost_count++;
  96         ost_desc->ltd_qos.ltq_oss = oss;
  97
  98         CDEBUG(D_QOS, "add tgt %s to OSS %s (%d OSTs)\n",
  99                obd_uuid2str(&ost_desc->ltd_uuid), obd_uuid2str(&oss->lqo_uuid),
 100                oss->lqo_ost_count);
 101
 102         /* Add sorted by # of OSTs.  Find the first entry that we're
 103            bigger than... */
 104         list = &lod->lod_qos.lq_oss_list;
 105         cfs_list_for_each_entry(temposs, list, lqo_oss_list) {
 106                 if (oss->lqo_ost_count > temposs->lqo_ost_count)
 107                         break;
 108         }
 109         /* ...and add before it.  If we're the first or smallest, temposs
 110            points to the list head, and we add to the end. */
 111         cfs_list_add_tail(&oss->lqo_oss_list, &temposs->lqo_oss_list);
 112
 113         lod->lod_qos.lq_dirty = 1;
 114         lod->lod_qos.lq_rr.lqr_dirty = 1;
 115
 116 out:
 117         up_write(&lod->lod_qos.lq_rw_sem);
 118         RETURN(rc);
 119 }
 120
 121 int qos_del_tgt(struct lod_device *lod, struct lod_tgt_desc *ost_desc)
 122 {
 123         struct lov_qos_oss *oss;
 124         int                 rc = 0;
 125         ENTRY;
 126
 127         down_write(&lod->lod_qos.lq_rw_sem);
 128         oss = ost_desc->ltd_qos.ltq_oss;
 129         if (!oss)
 130                 GOTO(out, rc = -ENOENT);
 131
 132         oss->lqo_ost_count--;
 133         if (oss->lqo_ost_count == 0) {
 134                 CDEBUG(D_QOS, "removing OSS %s\n",
 135                        obd_uuid2str(&oss->lqo_uuid));
 136                 cfs_list_del(&oss->lqo_oss_list);
 137                 ost_desc->ltd_qos.ltq_oss = NULL;
 138                 OBD_FREE_PTR(oss);
 139         }
 140
 141         lod->lod_qos.lq_dirty = 1;
 142         lod->lod_qos.lq_rr.lqr_dirty = 1;
 143 out:
 144         up_write(&lod->lod_qos.lq_rw_sem);
 145         RETURN(rc);
 146 }
 147
 148 static int lod_statfs_and_check(const struct lu_env *env, struct lod_device *d,
 149                                 int index, struct obd_statfs *sfs)
 150 {
 151         struct lod_tgt_desc *ost;
 152         int                  rc;
 153
 154         LASSERT(d);
 155         ost = OST_TGT(d,index);
 156         LASSERT(ost);
 157
 158         rc = dt_statfs(env, ost->ltd_ost, sfs);
 159         if (rc && rc != -ENOTCONN)
 160                 CERROR("%s: statfs: rc = %d\n", lod2obd(d)->obd_name, rc);
 161
 162         /* check whether device has changed state (active, inactive) */
 163         if (rc != 0 && ost->ltd_active) {
 164                 /* turned inactive? */
 165                 spin_lock(&d->lod_desc_lock);
 166                 if (ost->ltd_active) {
 167                         ost->ltd_active = 0;
 168                         LASSERT(d->lod_desc.ld_active_tgt_count > 0);
 169                         d->lod_desc.ld_active_tgt_count--;
 170                         d->lod_qos.lq_dirty = 1;
 171                         d->lod_qos.lq_rr.lqr_dirty = 1;
 172                         CDEBUG(D_CONFIG, "%s: turns inactive\n",
 173                                ost->ltd_exp->exp_obd->obd_name);
 174                 }
 175                 spin_unlock(&d->lod_desc_lock);
 176         } else if (rc == 0 && ost->ltd_active == 0) {
 177                 /* turned active? */
 178                 LASSERTF(d->lod_desc.ld_active_tgt_count < d->lod_ostnr,
 179                          "active tgt count %d, ost nr %d\n",
 180                          d->lod_desc.ld_active_tgt_count, d->lod_ostnr);
 181                 spin_lock(&d->lod_desc_lock);
 182                 if (ost->ltd_active == 0) {
 183                         ost->ltd_active = 1;
 184                         d->lod_desc.ld_active_tgt_count++;
 185                         d->lod_qos.lq_dirty = 1;
 186                         d->lod_qos.lq_rr.lqr_dirty = 1;
 187                         CDEBUG(D_CONFIG, "%s: turns active\n",
 188                                ost->ltd_exp->exp_obd->obd_name);
 189                 }
 190                 spin_unlock(&d->lod_desc_lock);
 191         }
 192
 193         RETURN(rc);
 194 }
 195
 196 static void lod_qos_statfs_update(const struct lu_env *env,
 197                                   struct lod_device *lod)
 198 {
 199         struct obd_device *obd = lod2obd(lod);
 200         struct ost_pool   *osts = &(lod->lod_pool_info);
 201         int                i, idx, rc = 0;
 202         __u64              max_age, avail;
 203         ENTRY;
 204
 205         max_age = cfs_time_shift_64(-2 * lod->lod_desc.ld_qos_maxage);
 206
 207         if (cfs_time_beforeq_64(max_age, obd->obd_osfs_age))
 208                 /* statfs data are quite recent, don't need to refresh it */
 209                 RETURN_EXIT;
 210
 211         down_write(&lod->lod_qos.lq_rw_sem);
 212         if (cfs_time_beforeq_64(max_age, obd->obd_osfs_age))
 213                 GOTO(out, rc = 0);
 214
 215         for (i = 0; i < osts->op_count; i++) {
 216                 idx = osts->op_array[i];
 217                 avail = OST_TGT(lod,idx)->ltd_statfs.os_bavail;
 218                 rc = lod_statfs_and_check(env, lod, idx,
 219                                           &OST_TGT(lod,idx)->ltd_statfs);
 220                 if (rc)
 221                         break;
 222                 if (OST_TGT(lod,idx)->ltd_statfs.os_bavail != avail)
 223                         /* recalculate weigths */
 224                         lod->lod_qos.lq_dirty = 1;
 225         }
 226         obd->obd_osfs_age = cfs_time_current_64();
 227
 228 out:
 229         up_write(&lod->lod_qos.lq_rw_sem);
 230         EXIT;
 231 }
 232
 233 /* Recalculate per-object penalties for OSSs and OSTs,
 234    depends on size of each ost in an oss */
 235 static int lod_qos_calc_ppo(struct lod_device *lod)
 236 {
 237         struct lov_qos_oss *oss;
 238         __u64               ba_max, ba_min, temp;
 239         __u32               num_active;
 240         int                 rc, i, prio_wide;
 241         time_t              now, age;
 242         ENTRY;
 243
 244         if (!lod->lod_qos.lq_dirty)
 245                 GOTO(out, rc = 0);
 246
 247         num_active = lod->lod_desc.ld_active_tgt_count - 1;
 248         if (num_active < 1)
 249                 GOTO(out, rc = -EAGAIN);
 250
 251         /* find bavail on each OSS */
 252         cfs_list_for_each_entry(oss, &lod->lod_qos.lq_oss_list, lqo_oss_list)
 253                                 oss->lqo_bavail = 0;
 254         lod->lod_qos.lq_active_oss_count = 0;
 255
 256         /*
 257          * How badly user wants to select OSTs "widely" (not recently chosen
 258          * and not on recent OSS's).  As opposed to "freely" (free space
 259          * avail.) 0-256
 260          */
 261         prio_wide = 256 - lod->lod_qos.lq_prio_free;
 262
 263         ba_min = (__u64)(-1);
 264         ba_max = 0;
 265         now = cfs_time_current_sec();
 266         /* Calculate OST penalty per object
 267          * (lod ref taken in lod_qos_prep_create()) */
 268         cfs_foreach_bit(lod->lod_ost_bitmap, i) {
 269                 LASSERT(OST_TGT(lod,i));
 270                 temp = TGT_BAVAIL(i);
 271                 if (!temp)
 272                         continue;
 273                 ba_min = min(temp, ba_min);
 274                 ba_max = max(temp, ba_max);
 275
 276                 /* Count the number of usable OSS's */
 277                 if (OST_TGT(lod,i)->ltd_qos.ltq_oss->lqo_bavail == 0)
 278                         lod->lod_qos.lq_active_oss_count++;
 279                 OST_TGT(lod,i)->ltd_qos.ltq_oss->lqo_bavail += temp;
 280
 281                 /* per-OST penalty is prio * TGT_bavail / (num_ost - 1) / 2 */
 282                 temp >>= 1;
 283                 lov_do_div64(temp, num_active);
 284                 OST_TGT(lod,i)->ltd_qos.ltq_penalty_per_obj =
 285                         (temp * prio_wide) >> 8;
 286
 287                 age = (now - OST_TGT(lod,i)->ltd_qos.ltq_used) >> 3;
 288                 if (lod->lod_qos.lq_reset ||
 289                     age > 32 * lod->lod_desc.ld_qos_maxage)
 290                         OST_TGT(lod,i)->ltd_qos.ltq_penalty = 0;
 291                 else if (age > lod->lod_desc.ld_qos_maxage)
 292                         /* Decay the penalty by half for every 8x the update
 293                          * interval that the device has been idle.  That gives
 294                          * lots of time for the statfs information to be
 295                          * updated (which the penalty is only a proxy for),
 296                          * and avoids penalizing OSS/OSTs under light load. */
 297                         OST_TGT(lod,i)->ltd_qos.ltq_penalty >>=
 298                                 (age / lod->lod_desc.ld_qos_maxage);
 299         }
 300
 301         num_active = lod->lod_qos.lq_active_oss_count - 1;
 302         if (num_active < 1) {
 303                 /* If there's only 1 OSS, we can't penalize it, so instead
 304                    we have to double the OST penalty */
 305                 num_active = 1;
 306                 cfs_foreach_bit(lod->lod_ost_bitmap, i)
 307                         OST_TGT(lod,i)->ltd_qos.ltq_penalty_per_obj <<= 1;
 308         }
 309
 310         /* Per-OSS penalty is prio * oss_avail / oss_osts / (num_oss - 1) / 2 */
 311         cfs_list_for_each_entry(oss, &lod->lod_qos.lq_oss_list, lqo_oss_list) {
 312                 temp = oss->lqo_bavail >> 1;
 313                 lov_do_div64(temp, oss->lqo_ost_count * num_active);
 314                 oss->lqo_penalty_per_obj = (temp * prio_wide) >> 8;
 315
 316                 age = (now - oss->lqo_used) >> 3;
 317                 if (lod->lod_qos.lq_reset ||
 318                     age > 32 * lod->lod_desc.ld_qos_maxage)
 319                         oss->lqo_penalty = 0;
 320                 else if (age > lod->lod_desc.ld_qos_maxage)
 321                         /* Decay the penalty by half for every 8x the update
 322                          * interval that the device has been idle.  That gives
 323                          * lots of time for the statfs information to be
 324                          * updated (which the penalty is only a proxy for),
 325                          * and avoids penalizing OSS/OSTs under light load. */
 326                         oss->lqo_penalty >>= age / lod->lod_desc.ld_qos_maxage;
 327         }
 328
 329         lod->lod_qos.lq_dirty = 0;
 330         lod->lod_qos.lq_reset = 0;
 331
 332         /* If each ost has almost same free space,
 333          * do rr allocation for better creation performance */
 334         lod->lod_qos.lq_same_space = 0;
 335         if ((ba_max * (256 - lod->lod_qos.lq_threshold_rr)) >> 8 < ba_min) {
 336                 lod->lod_qos.lq_same_space = 1;
 337                 /* Reset weights for the next time we enter qos mode */
 338                 lod->lod_qos.lq_reset = 1;
 339         }
 340         rc = 0;
 341
 342 out:
 343 #ifndef FORCE_QOS
 344         if (!rc && lod->lod_qos.lq_same_space)
 345                 RETURN(-EAGAIN);
 346 #endif
 347         RETURN(rc);
 348 }
 349
 350 static int lod_qos_calc_weight(struct lod_device *lod, int i)
 351 {
 352         __u64 temp, temp2;
 353
 354         /* Final ost weight = TGT_BAVAIL - ost_penalty - oss_penalty */
 355         temp = TGT_BAVAIL(i);
 356         temp2 = OST_TGT(lod,i)->ltd_qos.ltq_penalty +
 357                 OST_TGT(lod,i)->ltd_qos.ltq_oss->lqo_penalty;
 358         if (temp < temp2)
 359                 OST_TGT(lod,i)->ltd_qos.ltq_weight = 0;
 360         else
 361                 OST_TGT(lod,i)->ltd_qos.ltq_weight = temp - temp2;
 362         return 0;
 363 }
 364
 365 /* We just used this index for a stripe; adjust everyone's weights */
 366 static int lod_qos_used(struct lod_device *lod, struct ost_pool *osts,
 367                         __u32 index, __u64 *total_wt)
 368 {
 369         struct lod_tgt_desc *ost;
 370         struct lov_qos_oss  *oss;
 371         int j;
 372         ENTRY;
 373
 374         ost = OST_TGT(lod,index);
 375         LASSERT(ost);
 376
 377         /* Don't allocate on this devuce anymore, until the next alloc_qos */
 378         ost->ltd_qos.ltq_usable = 0;
 379
 380         oss = ost->ltd_qos.ltq_oss;
 381
 382         /* Decay old penalty by half (we're adding max penalty, and don't
 383            want it to run away.) */
 384         ost->ltd_qos.ltq_penalty >>= 1;
 385         oss->lqo_penalty >>= 1;
 386
 387         /* mark the OSS and OST as recently used */
 388         ost->ltd_qos.ltq_used = oss->lqo_used = cfs_time_current_sec();
 389
 390         /* Set max penalties for this OST and OSS */
 391         ost->ltd_qos.ltq_penalty +=
 392                 ost->ltd_qos.ltq_penalty_per_obj * lod->lod_ostnr;
 393         oss->lqo_penalty += oss->lqo_penalty_per_obj *
 394                 lod->lod_qos.lq_active_oss_count;
 395
 396         /* Decrease all OSS penalties */
 397         cfs_list_for_each_entry(oss, &lod->lod_qos.lq_oss_list, lqo_oss_list) {
 398                 if (oss->lqo_penalty < oss->lqo_penalty_per_obj)
 399                         oss->lqo_penalty = 0;
 400                 else
 401                         oss->lqo_penalty -= oss->lqo_penalty_per_obj;
 402         }
 403
 404         *total_wt = 0;
 405         /* Decrease all OST penalties */
 406         for (j = 0; j < osts->op_count; j++) {
 407                 int i;
 408
 409                 i = osts->op_array[j];
 410                 if (!cfs_bitmap_check(lod->lod_ost_bitmap, i))
 411                         continue;
 412
 413                 ost = OST_TGT(lod,i);
 414                 LASSERT(ost);
 415
 416                 if (ost->ltd_qos.ltq_penalty <
 417                                 ost->ltd_qos.ltq_penalty_per_obj)
 418                         ost->ltd_qos.ltq_penalty = 0;
 419                 else
 420                         ost->ltd_qos.ltq_penalty -=
 421                                 ost->ltd_qos.ltq_penalty_per_obj;
 422
 423                 lod_qos_calc_weight(lod, i);
 424
 425                 /* Recalc the total weight of usable osts */
 426                 if (ost->ltd_qos.ltq_usable)
 427                         *total_wt += ost->ltd_qos.ltq_weight;
 428
 429                 QOS_DEBUG("recalc tgt %d usable=%d avail="LPU64
 430                           " ostppo="LPU64" ostp="LPU64" ossppo="LPU64
 431                           " ossp="LPU64" wt="LPU64"\n",
 432                           i, ost->ltd_qos.ltq_usable, TGT_BAVAIL(i) >> 10,
 433                           ost->ltd_qos.ltq_penalty_per_obj >> 10,
 434                           ost->ltd_qos.ltq_penalty >> 10,
 435                           ost->ltd_qos.ltq_oss->lqo_penalty_per_obj >> 10,
 436                           ost->ltd_qos.ltq_oss->lqo_penalty >> 10,
 437                           ost->ltd_qos.ltq_weight >> 10);
 438         }
 439
 440         RETURN(0);
 441 }
 442
 443 #define LOV_QOS_EMPTY ((__u32)-1)
 444 /* compute optimal round-robin order, based on OSTs per OSS */
 445 static int lod_qos_calc_rr(struct lod_device *lod, struct ost_pool *src_pool,
 446                            struct lov_qos_rr *lqr)
 447 {
 448         struct lov_qos_oss  *oss;
 449         struct lod_tgt_desc *ost;
 450         unsigned placed, real_count;
 451         int i, rc;
 452         ENTRY;
 453
 454         if (!lqr->lqr_dirty) {
 455                 LASSERT(lqr->lqr_pool.op_size);
 456                 RETURN(0);
 457         }
 458
 459         /* Do actual allocation. */
 460         down_write(&lod->lod_qos.lq_rw_sem);
 461
 462         /*
 463          * Check again. While we were sleeping on @lq_rw_sem something could
 464          * change.
 465          */
 466         if (!lqr->lqr_dirty) {
 467                 LASSERT(lqr->lqr_pool.op_size);
 468                 up_write(&lod->lod_qos.lq_rw_sem);
 469                 RETURN(0);
 470         }
 471
 472         real_count = src_pool->op_count;
 473
 474         /* Zero the pool array */
 475         /* alloc_rr is holding a read lock on the pool, so nobody is adding/
 476            deleting from the pool. The lq_rw_sem insures that nobody else
 477            is reading. */
 478         lqr->lqr_pool.op_count = real_count;
 479         rc = lod_ost_pool_extend(&lqr->lqr_pool, real_count);
 480         if (rc) {
 481                 up_write(&lod->lod_qos.lq_rw_sem);
 482                 RETURN(rc);
 483         }
 484         for (i = 0; i < lqr->lqr_pool.op_count; i++)
 485                 lqr->lqr_pool.op_array[i] = LOV_QOS_EMPTY;
 486
 487         /* Place all the OSTs from 1 OSS at the same time. */
 488         placed = 0;
 489         cfs_list_for_each_entry(oss, &lod->lod_qos.lq_oss_list, lqo_oss_list) {
 490                 int j = 0;
 491
 492                 for (i = 0; i < lqr->lqr_pool.op_count; i++) {
 493                         int next;
 494
 495                         if (!cfs_bitmap_check(lod->lod_ost_bitmap,
 496                                                 src_pool->op_array[i]))
 497                                 continue;
 498
 499                         ost = OST_TGT(lod,src_pool->op_array[i]);
 500                         LASSERT(ost && ost->ltd_ost);
 501                         if (ost->ltd_qos.ltq_oss != oss)
 502                                 continue;
 503
 504                         /* Evenly space these OSTs across arrayspace */
 505                         next = j * lqr->lqr_pool.op_count / oss->lqo_ost_count;
 506                         while (lqr->lqr_pool.op_array[next] != LOV_QOS_EMPTY)
 507                                 next = (next + 1) % lqr->lqr_pool.op_count;
 508
 509                         lqr->lqr_pool.op_array[next] = src_pool->op_array[i];
 510                         j++;
 511                         placed++;
 512                 }
 513         }
 514
 515         lqr->lqr_dirty = 0;
 516         up_write(&lod->lod_qos.lq_rw_sem);
 517
 518         if (placed != real_count) {
 519                 /* This should never happen */
 520                 LCONSOLE_ERROR_MSG(0x14e, "Failed to place all OSTs in the "
 521                                    "round-robin list (%d of %d).\n",
 522                                    placed, real_count);
 523                 for (i = 0; i < lqr->lqr_pool.op_count; i++) {
 524                         LCONSOLE(D_WARNING, "rr #%d ost idx=%d\n", i,
 525                                  lqr->lqr_pool.op_array[i]);
 526                 }
 527                 lqr->lqr_dirty = 1;
 528                 RETURN(-EAGAIN);
 529         }
 530
 531 #if 0
 532         for (i = 0; i < lqr->lqr_pool.op_count; i++)
 533                 QOS_CONSOLE("rr #%d ost idx=%d\n", i, lqr->lqr_pool.op_array[i]);
 534 #endif
 535
 536         RETURN(0);
 537 }
 538
 539 /**
 540  * A helper function to:
 541  *   create in-core lu object on the specified OSP
 542  *   declare creation of the object
 543  * IMPORTANT: at this stage object is anonymouos - it has no fid assigned
 544  *            this is a workaround till we have natural FIDs on OST
 545  *
 546  *            at this point we want to declare (reserve) object for us as
 547  *            we can't block at execution (when create method is called).
 548  *            otherwise we'd block whole transaction batch
 549  */
 550 static struct dt_object *lod_qos_declare_object_on(const struct lu_env *env,
 551                                                    struct lod_device *d,
 552                                                    int ost_idx,
 553                                                    struct thandle *th)
 554 {
 555         struct lod_tgt_desc *ost;
 556         struct lu_object *o, *n;
 557         struct lu_device *nd;
 558         struct dt_object *dt;
 559         int               rc;
 560         ENTRY;
 561
 562         LASSERT(d);
 563         LASSERT(ost_idx >= 0);
 564         LASSERT(ost_idx < d->lod_osts_size);
 565         ost = OST_TGT(d,ost_idx);
 566         LASSERT(ost);
 567         LASSERT(ost->ltd_ost);
 568
 569         nd = &ost->ltd_ost->dd_lu_dev;
 570
 571         /*
 572          * allocate anonymous object with zero fid, real fid
 573          * will be assigned by OSP within transaction
 574          * XXX: to be fixed with fully-functional OST fids
 575          */
 576         o = lu_object_anon(env, nd, NULL);
 577         if (IS_ERR(o))
 578                 GOTO(out, dt = ERR_PTR(PTR_ERR(o)));
 579
 580         n = lu_object_locate(o->lo_header, nd->ld_type);
 581         if (unlikely(n == NULL)) {
 582                 CERROR("can't find slice\n");
 583                 lu_object_put(env, o);
 584                 GOTO(out, dt = ERR_PTR(-EINVAL));
 585         }
 586
 587         dt = container_of(n, struct dt_object, do_lu);
 588
 589         rc = dt_declare_create(env, dt, NULL, NULL, NULL, th);
 590         if (rc) {
 591                 CDEBUG(D_OTHER, "can't declare creation on #%u: %d\n",
 592                        ost_idx, rc);
 593                 lu_object_put(env, o);
 594                 dt = ERR_PTR(rc);
 595         }
 596
 597 out:
 598         RETURN(dt);
 599 }
 600
 601 static int min_stripe_count(int stripe_cnt, int flags)
 602 {
 603         return (flags & LOV_USES_DEFAULT_STRIPE ?
 604                         stripe_cnt - (stripe_cnt / 4) : stripe_cnt);
 605 }
 606
 607 #define LOV_CREATE_RESEED_MULT 30
 608 #define LOV_CREATE_RESEED_MIN  2000
 609
 610 static int inline lod_qos_dev_is_full(struct obd_statfs *msfs)
 611 {
 612         __u64 used;
 613         int   bs = msfs->os_bsize;
 614
 615         LASSERT(((bs - 1) & bs) == 0);
 616
 617         /* the minimum of 0.1% used blocks and 1GB bytes. */
 618         used = min_t(__u64, (msfs->os_blocks - msfs->os_bfree) >> 10,
 619                         1 << (31 - ffs(bs)));
 620         return (msfs->os_bavail < used);
 621 }
 622
 623 int lod_ea_store_resize(struct lod_thread_info *info, int size);
 624
 625 static inline int lod_qos_ost_in_use_clear(const struct lu_env *env, int stripes)
 626 {
 627         struct lod_thread_info *info = lod_env_info(env);
 628
 629         if (info->lti_ea_store_size < sizeof(int) * stripes)
 630                 lod_ea_store_resize(info, stripes * sizeof(int));
 631         if (info->lti_ea_store_size < sizeof(int) * stripes) {
 632                 CERROR("can't allocate memory for ost-in-use array\n");
 633                 return -ENOMEM;
 634         }
 635         memset(info->lti_ea_store, -1, sizeof(int) * stripes);
 636         return 0;
 637 }
 638
 639 static inline void lod_qos_ost_in_use(const struct lu_env *env, int idx, int ost)
 640 {
 641         struct lod_thread_info *info = lod_env_info(env);
 642         int *osts = info->lti_ea_store;
 643
 644         LASSERT(info->lti_ea_store_size >= idx * sizeof(int));
 645         osts[idx] = ost;
 646 }
 647
 648 static int lod_qos_is_ost_used(const struct lu_env *env, int ost, int stripes)
 649 {
 650         struct lod_thread_info *info = lod_env_info(env);
 651         int *osts = info->lti_ea_store;
 652         int j;
 653
 654         for (j = 0; j < stripes; j++) {
 655                 if (osts[j] == ost)
 656                         return 1;
 657         }
 658         return 0;
 659 }
 660
 661 /* Allocate objects on OSTs with round-robin algorithm */
 662 static int lod_alloc_rr(const struct lu_env *env, struct lod_object *lo,
 663                         struct dt_object **stripe, int flags,
 664                         struct thandle *th)
 665 {
 666         struct lod_device *m = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 667         struct obd_statfs *sfs = &lod_env_info(env)->lti_osfs;
 668         struct pool_desc  *pool = NULL;
 669         struct ost_pool   *osts;
 670         struct lov_qos_rr *lqr;
 671         struct dt_object  *o;
 672         unsigned           array_idx;
 673         int                i, rc;
 674         int                ost_start_idx_temp;
 675         int                speed = 0;
 676         int                stripe_idx = 0;
 677         int                stripe_cnt = lo->ldo_stripenr;
 678         int                stripe_cnt_min = min_stripe_count(stripe_cnt, flags);
 679         __u32              ost_idx;
 680         ENTRY;
 681
 682         if (lo->ldo_pool)
 683                 pool = lod_find_pool(m, lo->ldo_pool);
 684
 685         if (pool != NULL) {
 686                 down_read(&pool_tgt_rw_sem(pool));
 687                 osts = &(pool->pool_obds);
 688                 lqr = &(pool->pool_rr);
 689         } else {
 690                 osts = &(m->lod_pool_info);
 691                 lqr = &(m->lod_qos.lq_rr);
 692         }
 693
 694         rc = lod_qos_calc_rr(m, osts, lqr);
 695         if (rc)
 696                 GOTO(out, rc);
 697
 698         rc = lod_qos_ost_in_use_clear(env, lo->ldo_stripenr);
 699         if (rc)
 700                 GOTO(out, rc);
 701
 702         if (--lqr->lqr_start_count <= 0) {
 703                 lqr->lqr_start_idx = cfs_rand() % osts->op_count;
 704                 lqr->lqr_start_count =
 705                         (LOV_CREATE_RESEED_MIN / max(osts->op_count, 1U) +
 706                          LOV_CREATE_RESEED_MULT) * max(osts->op_count, 1U);
 707         } else if (stripe_cnt_min >= osts->op_count ||
 708                         lqr->lqr_start_idx > osts->op_count) {
 709                 /* If we have allocated from all of the OSTs, slowly
 710                  * precess the next start if the OST/stripe count isn't
 711                  * already doing this for us. */
 712                 lqr->lqr_start_idx %= osts->op_count;
 713                 if (stripe_cnt > 1 && (osts->op_count % stripe_cnt) != 1)
 714                         ++lqr->lqr_offset_idx;
 715         }
 716         down_read(&m->lod_qos.lq_rw_sem);
 717         ost_start_idx_temp = lqr->lqr_start_idx;
 718
 719 repeat_find:
 720         array_idx = (lqr->lqr_start_idx + lqr->lqr_offset_idx) %
 721                         osts->op_count;
 722
 723         QOS_DEBUG("pool '%s' want %d startidx %d startcnt %d offset %d "
 724                   "active %d count %d arrayidx %d\n",
 725                   lo->ldo_pool ? lo->ldo_pool : "",
 726                   stripe_cnt, lqr->lqr_start_idx, lqr->lqr_start_count,
 727                   lqr->lqr_offset_idx, osts->op_count, osts->op_count,
 728                   array_idx);
 729
 730         for (i = 0; i < osts->op_count && stripe_idx < lo->ldo_stripenr;
 731              i++, array_idx = (array_idx + 1) % osts->op_count) {
 732                 ++lqr->lqr_start_idx;
 733                 ost_idx = lqr->lqr_pool.op_array[array_idx];
 734
 735                 QOS_DEBUG("#%d strt %d act %d strp %d ary %d idx %d\n",
 736                           i, lqr->lqr_start_idx, /* XXX: active*/ 0,
 737                           stripe_idx, array_idx, ost_idx);
 738
 739                 if ((ost_idx == LOV_QOS_EMPTY) ||
 740                     !cfs_bitmap_check(m->lod_ost_bitmap, ost_idx))
 741                         continue;
 742
 743                 /* Fail Check before osc_precreate() is called
 744                    so we can only 'fail' single OSC. */
 745                 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OSC_PRECREATE) && ost_idx == 0)
 746                         continue;
 747
 748                 rc = lod_statfs_and_check(env, m, ost_idx, sfs);
 749                 if (rc) {
 750                         /* this OSP doesn't feel well */
 751                         continue;
 752                 }
 753
 754                 /*
 755                  * skip full devices
 756                  */
 757                 if (lod_qos_dev_is_full(sfs)) {
 758                         QOS_DEBUG("#%d is full\n", ost_idx);
 759                         continue;
 760                 }
 761
 762                 /*
 763                  * We expect number of precreated objects in f_ffree at
 764                  * the first iteration, skip OSPs with no objects ready
 765                  */
 766                 if (sfs->os_fprecreated == 0 && speed == 0) {
 767                         QOS_DEBUG("#%d: precreation is empty\n", ost_idx);
 768                         continue;
 769                 }
 770
 771                 /*
 772                  * try to use another OSP if this one is degraded
 773                  */
 774                 if (sfs->os_state == OS_STATE_DEGRADED && speed < 2) {
 775                         QOS_DEBUG("#%d: degraded\n", ost_idx);
 776                         continue;
 777                 }
 778
 779                 /*
 780                  * do not put >1 objects on a single OST
 781                  */
 782                 if (speed && lod_qos_is_ost_used(env, ost_idx, stripe_idx))
 783                         continue;
 784
 785                 o = lod_qos_declare_object_on(env, m, ost_idx, th);
 786                 if (IS_ERR(o)) {
 787                         CDEBUG(D_OTHER, "can't declare new object on #%u: %d\n",
 788                                ost_idx, (int) PTR_ERR(o));
 789                         rc = PTR_ERR(o);
 790                         continue;
 791                 }
 792
 793                 /*
 794                  * We've successfuly declared (reserved) an object
 795                  */
 796                 lod_qos_ost_in_use(env, stripe_idx, ost_idx);
 797                 stripe[stripe_idx] = o;
 798                 stripe_idx++;
 799
 800         }
 801         if ((speed < 2) && (stripe_idx < stripe_cnt_min)) {
 802                 /* Try again, allowing slower OSCs */
 803                 speed++;
 804                 lqr->lqr_start_idx = ost_start_idx_temp;
 805                 goto repeat_find;
 806         }
 807
 808         up_read(&m->lod_qos.lq_rw_sem);
 809
 810         if (stripe_idx) {
 811                 lo->ldo_stripenr = stripe_idx;
 812                 /* at least one stripe is allocated */
 813                 rc = 0;
 814         } else {
 815                 /* nobody provided us with a single object */
 816                 rc = -ENOSPC;
 817         }
 818
 819 out:
 820         if (pool != NULL) {
 821                 up_read(&pool_tgt_rw_sem(pool));
 822                 /* put back ref got by lod_find_pool() */
 823                 lod_pool_putref(pool);
 824         }
 825
 826         RETURN(rc);
 827 }
 828
 829 /* alloc objects on osts with specific stripe offset */
 830 static int lod_alloc_specific(const struct lu_env *env, struct lod_object *lo,
 831                               struct dt_object **stripe, int flags,
 832                               struct thandle *th)
 833 {
 834         struct lod_device *m = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 835         struct obd_statfs *sfs = &lod_env_info(env)->lti_osfs;
 836         struct dt_object  *o;
 837         unsigned           ost_idx, array_idx, ost_count;
 838         int                i, rc, stripe_num = 0;
 839         int                speed = 0;
 840         struct pool_desc  *pool = NULL;
 841         struct ost_pool   *osts;
 842         ENTRY;
 843
 844         rc = lod_qos_ost_in_use_clear(env, lo->ldo_stripenr);
 845         if (rc)
 846                 GOTO(out, rc);
 847
 848         if (lo->ldo_pool)
 849                 pool = lod_find_pool(m, lo->ldo_pool);
 850
 851         if (pool != NULL) {
 852                 down_read(&pool_tgt_rw_sem(pool));
 853                 osts = &(pool->pool_obds);
 854         } else {
 855                 osts = &(m->lod_pool_info);
 856         }
 857
 858         ost_count = osts->op_count;
 859
 860 repeat_find:
 861         /* search loi_ost_idx in ost array */
 862         array_idx = 0;
 863         for (i = 0; i < ost_count; i++) {
 864                 if (osts->op_array[i] == lo->ldo_def_stripe_offset) {
 865                         array_idx = i;
 866                         break;
 867                 }
 868         }
 869         if (i == ost_count) {
 870                 CERROR("Start index %d not found in pool '%s'\n",
 871                        lo->ldo_def_stripe_offset,
 872                        lo->ldo_pool ? lo->ldo_pool : "");
 873                 GOTO(out, rc = -EINVAL);
 874         }
 875
 876         for (i = 0; i < ost_count;
 877                         i++, array_idx = (array_idx + 1) % ost_count) {
 878                 ost_idx = osts->op_array[array_idx];
 879
 880                 if (!cfs_bitmap_check(m->lod_ost_bitmap, ost_idx))
 881                         continue;
 882
 883                 /* Fail Check before osc_precreate() is called
 884                    so we can only 'fail' single OSC. */
 885                 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OSC_PRECREATE) && ost_idx == 0)
 886                         continue;
 887
 888                 /*
 889                  * do not put >1 objects on a single OST
 890                  */
 891                 if (lod_qos_is_ost_used(env, ost_idx, stripe_num))
 892                         continue;
 893
 894                 /* Drop slow OSCs if we can, but not for requested start idx.
 895                  *
 896                  * This means "if OSC is slow and it is not the requested
 897                  * start OST, then it can be skipped, otherwise skip it only
 898                  * if it is inactive/recovering/out-of-space." */
 899
 900                 rc = lod_statfs_and_check(env, m, ost_idx, sfs);
 901                 if (rc) {
 902                         /* this OSP doesn't feel well */
 903                         continue;
 904                 }
 905
 906                 /*
 907                  * We expect number of precreated objects in f_ffree at
 908                  * the first iteration, skip OSPs with no objects ready
 909                  * don't apply this logic to OST specified with stripe_offset
 910                  */
 911                 if (i != 0 && sfs->os_fprecreated == 0 && speed == 0)
 912                         continue;
 913
 914                 o = lod_qos_declare_object_on(env, m, ost_idx, th);
 915                 if (IS_ERR(o)) {
 916                         CDEBUG(D_OTHER, "can't declare new object on #%u: %d\n",
 917                                ost_idx, (int) PTR_ERR(o));
 918                         continue;
 919                 }
 920
 921                 /*
 922                  * We've successfuly declared (reserved) an object
 923                  */
 924                 lod_qos_ost_in_use(env, stripe_num, ost_idx);
 925                 stripe[stripe_num] = o;
 926                 stripe_num++;
 927
 928                 /* We have enough stripes */
 929                 if (stripe_num == lo->ldo_stripenr)
 930                         GOTO(out, rc = 0);
 931         }
 932         if (speed < 2) {
 933                 /* Try again, allowing slower OSCs */
 934                 speed++;
 935                 goto repeat_find;
 936         }
 937
 938         /* If we were passed specific striping params, then a failure to
 939          * meet those requirements is an error, since we can't reallocate
 940          * that memory (it might be part of a larger array or something).
 941          *
 942          * We can only get here if lsm_stripe_count was originally > 1.
 943          */
 944         CERROR("can't lstripe objid "DFID": have %d want %u\n",
 945                PFID(lu_object_fid(lod2lu_obj(lo))), stripe_num,
 946                lo->ldo_stripenr);
 947         rc = -EFBIG;
 948 out:
 949         if (pool != NULL) {
 950                 up_read(&pool_tgt_rw_sem(pool));
 951                 /* put back ref got by lod_find_pool() */
 952                 lod_pool_putref(pool);
 953         }
 954
 955         RETURN(rc);
 956 }
 957
 958 static inline int lod_qos_is_usable(struct lod_device *lod)
 959 {
 960 #ifdef FORCE_QOS
 961         /* to be able to debug QoS code */
 962         return 1;
 963 #endif
 964
 965         /* Detect -EAGAIN early, before expensive lock is taken. */
 966         if (!lod->lod_qos.lq_dirty && lod->lod_qos.lq_same_space)
 967                 return 0;
 968
 969         if (lod->lod_desc.ld_active_tgt_count < 2)
 970                 return 0;
 971
 972         return 1;
 973 }
 974
 975 /* Alloc objects on OSTs with optimization based on:
 976    - free space
 977    - network resources (shared OSS's)
 978  */
 979 static int lod_alloc_qos(const struct lu_env *env, struct lod_object *lo,
 980                          struct dt_object **stripe, int flags,
 981                          struct thandle *th)
 982 {
 983         struct lod_device   *m = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 984         struct obd_statfs   *sfs = &lod_env_info(env)->lti_osfs;
 985         struct lod_tgt_desc *ost;
 986         struct dt_object    *o;
 987         __u64                total_weight = 0;
 988         int                  nfound, good_osts, i, rc = 0;
 989         int                  stripe_cnt = lo->ldo_stripenr;
 990         int                  stripe_cnt_min;
 991         struct pool_desc    *pool = NULL;
 992         struct ost_pool    *osts;
 993         ENTRY;
 994
 995         stripe_cnt_min = min_stripe_count(stripe_cnt, flags);
 996         if (stripe_cnt_min < 1)
 997                 RETURN(-EINVAL);
 998
 999         if (lo->ldo_pool)
1000                 pool = lod_find_pool(m, lo->ldo_pool);
1001
1002         if (pool != NULL) {
1003                 down_read(&pool_tgt_rw_sem(pool));
1004                 osts = &(pool->pool_obds);
1005         } else {
1006                 osts = &(m->lod_pool_info);
1007         }
1008
1009         /* Detect -EAGAIN early, before expensive lock is taken. */
1010         if (!lod_qos_is_usable(m))
1011                 GOTO(out_nolock, rc = -EAGAIN);
1012
1013         /* Do actual allocation, use write lock here. */
1014         down_write(&m->lod_qos.lq_rw_sem);
1015
1016         /*
1017          * Check again, while we were sleeping on @lq_rw_sem things could
1018          * change.
1019          */
1020         if (!lod_qos_is_usable(m))
1021                 GOTO(out, rc = -EAGAIN);
1022
1023         rc = lod_qos_calc_ppo(m);
1024         if (rc)
1025                 GOTO(out, rc);
1026
1027         rc = lod_qos_ost_in_use_clear(env, lo->ldo_stripenr);
1028         if (rc)
1029                 GOTO(out, rc);
1030
1031         good_osts = 0;
1032         /* Find all the OSTs that are valid stripe candidates */
1033         for (i = 0; i < osts->op_count; i++) {
1034                 if (!cfs_bitmap_check(m->lod_ost_bitmap, osts->op_array[i]))
1035                         continue;
1036
1037                 rc = lod_statfs_and_check(env, m, osts->op_array[i], sfs);
1038                 if (rc) {
1039                         /* this OSP doesn't feel well */
1040                         continue;
1041                 }
1042
1043                 /*
1044                  * skip full devices
1045                  */
1046                 if (lod_qos_dev_is_full(sfs))
1047                         continue;
1048
1049                 /* Fail Check before osc_precreate() is called
1050                    so we can only 'fail' single OSC. */
1051                 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OSC_PRECREATE) &&
1052                                    osts->op_array[i] == 0)
1053                         continue;
1054
1055                 ost = OST_TGT(m,osts->op_array[i]);
1056                 ost->ltd_qos.ltq_usable = 1;
1057                 lod_qos_calc_weight(m, osts->op_array[i]);
1058                 total_weight += ost->ltd_qos.ltq_weight;
1059
1060                 good_osts++;
1061         }
1062
1063         QOS_DEBUG("found %d good osts\n", good_osts);
1064
1065         if (good_osts < stripe_cnt_min)
1066                 GOTO(out, rc = -EAGAIN);
1067
1068         /* We have enough osts */
1069         if (good_osts < stripe_cnt)
1070                 stripe_cnt = good_osts;
1071
1072         /* Find enough OSTs with weighted random allocation. */
1073         nfound = 0;
1074         while (nfound < stripe_cnt) {
1075                 __u64 rand, cur_weight;
1076
1077                 cur_weight = 0;
1078                 rc = -ENOSPC;
1079
1080                 if (total_weight) {
1081 #if BITS_PER_LONG == 32
1082                         rand = cfs_rand() % (unsigned)total_weight;
1083                         /* If total_weight > 32-bit, first generate the high
1084                          * 32 bits of the random number, then add in the low
1085                          * 32 bits (truncated to the upper limit, if needed) */
1086                         if (total_weight > 0xffffffffULL)
1087                                 rand = (__u64)(cfs_rand() %
1088                                         (unsigned)(total_weight >> 32)) << 32;
1089                         else
1090                                 rand = 0;
1091
1092                         if (rand == (total_weight & 0xffffffff00000000ULL))
1093                                 rand |= cfs_rand() % (unsigned)total_weight;
1094                         else
1095                                 rand |= cfs_rand();
1096
1097 #else
1098                         rand = ((__u64)cfs_rand() << 32 | cfs_rand()) %
1099                                 total_weight;
1100 #endif
1101                 } else {
1102                         rand = 0;
1103                 }
1104
1105                 /* On average, this will hit larger-weighted osts more often.
1106                    0-weight osts will always get used last (only when rand=0) */
1107                 for (i = 0; i < osts->op_count; i++) {
1108                         int idx = osts->op_array[i];
1109
1110                         if (!cfs_bitmap_check(m->lod_ost_bitmap, idx))
1111                                 continue;
1112
1113                         ost = OST_TGT(m,idx);
1114
1115                         if (!ost->ltd_qos.ltq_usable)
1116                                 continue;
1117
1118                         cur_weight += ost->ltd_qos.ltq_weight;
1119                         QOS_DEBUG("stripe_cnt=%d nfound=%d cur_weight="LPU64
1120                                   " rand="LPU64" total_weight="LPU64"\n",
1121                                   stripe_cnt, nfound, cur_weight, rand,
1122                                   total_weight);
1123
1124                         if (cur_weight < rand)
1125                                 continue;
1126
1127                         QOS_DEBUG("stripe=%d to idx=%d\n", nfound, idx);
1128
1129                         /*
1130                          * do not put >1 objects on a single OST
1131                          */
1132                         if (lod_qos_is_ost_used(env, idx, nfound))
1133                                 continue;
1134                         lod_qos_ost_in_use(env, nfound, idx);
1135
1136                         o = lod_qos_declare_object_on(env, m, idx, th);
1137                         if (IS_ERR(o)) {
1138                                 QOS_DEBUG("can't declare object on #%u: %d\n",
1139                                           idx, (int) PTR_ERR(o));
1140                                 continue;
1141                         }
1142                         stripe[nfound++] = o;
1143                         lod_qos_used(m, osts, idx, &total_weight);
1144                         rc = 0;
1145                         break;
1146                 }
1147
1148                 if (rc) {
1149                         /* no OST found on this iteration, give up */
1150                         break;
1151                 }
1152         }
1153
1154         if (unlikely(nfound != stripe_cnt)) {
1155                 /*
1156                  * when the decision to use weighted algorithm was made
1157                  * we had enough appropriate OSPs, but this state can
1158                  * change anytime (no space on OST, broken connection, etc)
1159                  * so it's possible OSP won't be able to provide us with
1160                  * an object due to just changed state
1161                  */
1162                 LCONSOLE_INFO("wanted %d, found %d\n", stripe_cnt, nfound);
1163                 for (i = 0; i < nfound; i++) {
1164                         LASSERT(stripe[i] != NULL);
1165                         lu_object_put(env, &stripe[i]->do_lu);
1166                         stripe[i] = NULL;
1167                 }
1168
1169                 /* makes sense to rebalance next time */
1170                 m->lod_qos.lq_dirty = 1;
1171                 m->lod_qos.lq_same_space = 0;
1172
1173                 rc = -EAGAIN;
1174         }
1175
1176 out:
1177         up_write(&m->lod_qos.lq_rw_sem);
1178
1179 out_nolock:
1180         if (pool != NULL) {
1181                 up_read(&pool_tgt_rw_sem(pool));
1182                 /* put back ref got by lod_find_pool() */
1183                 lod_pool_putref(pool);
1184         }
1185
1186         RETURN(rc);
1187 }
1188
1189 /* Find the max stripecount we should use */
1190 static __u16 lod_get_stripecnt(struct lod_device *lod, __u32 magic,
1191                                __u16 stripe_count)
1192 {
1193         __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
1194
1195         if (!stripe_count)
1196                 stripe_count = lod->lod_desc.ld_default_stripe_count;
1197         if (stripe_count > lod->lod_desc.ld_active_tgt_count)
1198                 stripe_count = lod->lod_desc.ld_active_tgt_count;
1199         if (!stripe_count)
1200                 stripe_count = 1;
1201
1202         /* stripe count is based on whether OSD can handle larger EA sizes */
1203         if (lod->lod_osd_max_easize > 0)
1204                 max_stripes = lov_mds_md_stripecnt(lod->lod_osd_max_easize,
1205                                                    magic);
1206
1207         return (stripe_count < max_stripes) ? stripe_count : max_stripes;
1208 }
1209
1210 static int lod_use_defined_striping(const struct lu_env *env,
1211                                     struct lod_object *mo,
1212                                     const struct lu_buf *buf)
1213 {
1214         struct lod_device      *d = lu2lod_dev(lod2lu_obj(mo)->lo_dev);
1215         struct lov_mds_md_v1   *v1 = buf->lb_buf;
1216         struct lov_mds_md_v3   *v3 = buf->lb_buf;
1217         struct lov_ost_data_v1 *objs;
1218         __u32                   magic;
1219         int                     rc;
1220         ENTRY;
1221
1222         rc = lod_verify_striping(d, buf, 1);
1223         if (rc)
1224                 RETURN(rc);
1225
1226         magic = le32_to_cpu(v1->lmm_magic);
1227         if (magic == LOV_MAGIC_V1_DEF) {
1228                 objs = &v1->lmm_objects[0];
1229         } else if (magic == LOV_MAGIC_V3_DEF) {
1230                 objs = &v3->lmm_objects[0];
1231                 lod_object_set_pool(mo, v3->lmm_pool_name);
1232         } else {
1233                 GOTO(out, rc = -EINVAL);
1234         }
1235
1236         mo->ldo_stripe_size = le32_to_cpu(v1->lmm_stripe_size);
1237         mo->ldo_stripenr = le16_to_cpu(v1->lmm_stripe_count);
1238         mo->ldo_layout_gen = le16_to_cpu(v1->lmm_layout_gen);
1239         LASSERT(buf->lb_len >= lov_mds_md_size(mo->ldo_stripenr, magic));
1240
1241         rc = lod_initialize_objects(env, mo, objs);
1242
1243 out:
1244         RETURN(rc);
1245 }
1246
1247 static int lod_qos_parse_config(const struct lu_env *env,
1248                                 struct lod_object *lo,
1249                                 const struct lu_buf *buf)
1250 {
1251         struct lod_device     *d = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
1252         struct lov_user_md_v1 *v1 = NULL;
1253         struct lov_user_md_v3 *v3 = NULL;
1254         struct pool_desc      *pool;
1255         __u32                  magic;
1256         int                    rc;
1257         ENTRY;
1258
1259         if (buf == NULL || buf->lb_buf == NULL || buf->lb_len == 0)
1260                 RETURN(0);
1261
1262         v1 = buf->lb_buf;
1263         magic = v1->lmm_magic;
1264
1265         if (magic == __swab32(LOV_USER_MAGIC_V1)) {
1266                 lustre_swab_lov_user_md_v1(v1);
1267                 magic = v1->lmm_magic;
1268         } else if (magic == __swab32(LOV_USER_MAGIC_V3)) {
1269                 v3 = buf->lb_buf;
1270                 lustre_swab_lov_user_md_v3(v3);
1271                 magic = v3->lmm_magic;
1272         }
1273
1274         if (unlikely(magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)) {
1275                 /* try to use as fully defined striping */
1276                 rc = lod_use_defined_striping(env, lo, buf);
1277                 RETURN(rc);
1278         }
1279
1280         if (unlikely(buf->lb_len < sizeof(*v1))) {
1281                 CERROR("wrong size: %u\n", (unsigned) buf->lb_len);
1282                 RETURN(-EINVAL);
1283         }
1284
1285         if (v1->lmm_pattern != 0 && v1->lmm_pattern != LOV_PATTERN_RAID0) {
1286                 CERROR("invalid pattern: %x\n", v1->lmm_pattern);
1287                 RETURN(-EINVAL);
1288         }
1289
1290         if (v1->lmm_stripe_size)
1291                 lo->ldo_stripe_size = v1->lmm_stripe_size;
1292         if (lo->ldo_stripe_size & (LOV_MIN_STRIPE_SIZE - 1))
1293                 lo->ldo_stripe_size = LOV_MIN_STRIPE_SIZE;
1294
1295         if (v1->lmm_stripe_count)
1296                 lo->ldo_stripenr = v1->lmm_stripe_count;
1297
1298         if ((v1->lmm_stripe_offset >= d->lod_desc.ld_tgt_count) &&
1299             (v1->lmm_stripe_offset != (typeof(v1->lmm_stripe_offset))(-1))) {
1300                 CERROR("invalid offset: %x\n", v1->lmm_stripe_offset);
1301                 RETURN(-EINVAL);
1302         }
1303         lo->ldo_def_stripe_offset = v1->lmm_stripe_offset;
1304
1305         CDEBUG(D_OTHER, "lsm: %u size, %u stripes, %u offset\n",
1306                v1->lmm_stripe_size, v1->lmm_stripe_count,
1307                v1->lmm_stripe_offset);
1308
1309         if (v1->lmm_magic == LOV_MAGIC_V3) {
1310                 if (buf->lb_len < sizeof(*v3)) {
1311                         CERROR("wrong size: %u\n", (unsigned) buf->lb_len);
1312                         RETURN(-EINVAL);
1313                 }
1314
1315                 v3 = buf->lb_buf;
1316                 lod_object_set_pool(lo, v3->lmm_pool_name);
1317
1318                 /* In the function below, .hs_keycmp resolves to
1319                  * pool_hashkey_keycmp() */
1320                 /* coverity[overrun-buffer-val] */
1321                 pool = lod_find_pool(d, v3->lmm_pool_name);
1322                 if (pool != NULL) {
1323                         if (lo->ldo_def_stripe_offset !=
1324                             (typeof(v1->lmm_stripe_offset))(-1)) {
1325                                 rc = lo->ldo_def_stripe_offset;
1326                                 rc = lod_check_index_in_pool(rc, pool);
1327                                 if (rc < 0) {
1328                                         lod_pool_putref(pool);
1329                                         CERROR("invalid offset\n");
1330                                         RETURN(-EINVAL);
1331                                 }
1332                         }
1333
1334                         if (lo->ldo_stripenr > pool_tgt_count(pool))
1335                                 lo->ldo_stripenr= pool_tgt_count(pool);
1336
1337                         lod_pool_putref(pool);
1338                 }
1339         } else
1340                 lod_object_set_pool(lo, NULL);
1341
1342         RETURN(0);
1343 }
1344
1345 /*
1346  * buf should be NULL or contain striping settings
1347  */
1348 int lod_qos_prep_create(const struct lu_env *env, struct lod_object *lo,
1349                         struct lu_attr *attr, const struct lu_buf *buf,
1350                         struct thandle *th)
1351 {
1352         struct lod_device      *d = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
1353         struct dt_object      **stripe;
1354         int                     stripe_len;
1355         int                     flag = LOV_USES_ASSIGNED_STRIPE;
1356         int                     i, rc;
1357         ENTRY;
1358
1359         LASSERT(lo);
1360
1361         /* no OST available */
1362         /* XXX: should we be waiting a bit to prevent failures during
1363          * cluster initialization? */
1364         if (d->lod_ostnr == 0)
1365                 GOTO(out, rc = -EIO);
1366
1367         /*
1368          * by this time, the object's ldo_stripenr and ldo_stripe_size
1369          * contain default value for striping: taken from the parent
1370          * or from filesystem defaults
1371          *
1372          * in case the caller is passing lovea with new striping config,
1373          * we may need to parse lovea and apply new configuration
1374          */
1375         rc = lod_qos_parse_config(env, lo, buf);
1376         if (rc)
1377                 GOTO(out, rc);
1378
1379         if (likely(lo->ldo_stripe == NULL)) {
1380                 /*
1381                  * no striping has been created so far
1382                  */
1383                 LASSERT(lo->ldo_stripenr > 0);
1384                 /*
1385                  * statfs and check OST targets now, since ld_active_tgt_count
1386                  * could be changed if some OSTs are [de]activated manually.
1387                  */
1388                 lod_qos_statfs_update(env, d);
1389                 lo->ldo_stripenr = lod_get_stripecnt(d, LOV_MAGIC,
1390                                 lo->ldo_stripenr);
1391
1392                 stripe_len = lo->ldo_stripenr;
1393                 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
1394                 if (stripe == NULL)
1395                         GOTO(out, rc = -ENOMEM);
1396
1397                 lod_getref(&d->lod_ost_descs);
1398                 /* XXX: support for non-0 files w/o objects */
1399                 CDEBUG(D_OTHER, "tgt_count %d stripenr %d\n",
1400                                 d->lod_desc.ld_tgt_count, stripe_len);
1401                 if (lo->ldo_def_stripe_offset >= d->lod_desc.ld_tgt_count) {
1402                         rc = lod_alloc_qos(env, lo, stripe, flag, th);
1403                         if (rc == -EAGAIN)
1404                                 rc = lod_alloc_rr(env, lo, stripe, flag, th);
1405                 } else {
1406                         rc = lod_alloc_specific(env, lo, stripe, flag, th);
1407                 }
1408                 lod_putref(d, &d->lod_ost_descs);
1409
1410                 if (rc < 0) {
1411                         for (i = 0; i < stripe_len; i++)
1412                                 if (stripe[i] != NULL)
1413                                         lu_object_put(env, &stripe[i]->do_lu);
1414
1415                         OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
1416                 } else {
1417                         lo->ldo_stripe = stripe;
1418                         lo->ldo_stripes_allocated = stripe_len;
1419                 }
1420         } else {
1421                 /*
1422                  * lod_qos_parse_config() found supplied buf as a predefined
1423                  * striping (not a hint), so it allocated all the object
1424                  * now we need to create them
1425                  */
1426                 for (i = 0; i < lo->ldo_stripenr; i++) {
1427                         struct dt_object  *o;
1428
1429                         o = lo->ldo_stripe[i];
1430                         LASSERT(o);
1431
1432                         rc = dt_declare_create(env, o, attr, NULL, NULL, th);
1433                         if (rc) {
1434                                 CERROR("can't declare create: %d\n", rc);
1435                                 break;
1436                         }
1437                 }
1438         }
1439
1440 out:
1441         RETURN(rc);
1442 }
1443