lustre/lov/lov_pack.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) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, 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/lov/lov_pack.c
  33  *
  34  * (Un)packing of OST/MDS requests
  35  *
  36  * Author: Andreas Dilger <adilger@clusterfs.com>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_LOV
  40
  41 #include <lustre_net.h>
  42 #include <lustre_swab.h>
  43 #include <obd.h>
  44 #include <obd_class.h>
  45 #include <obd_support.h>
  46
  47 #include "lov_cl_internal.h"
  48 #include "lov_internal.h"
  49
  50 void lov_dump_lmm_common(int level, void *lmmp)
  51 {
  52         struct lov_mds_md *lmm = lmmp;
  53         struct ost_id oi;
  54
  55         lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
  56         CDEBUG_LIMIT(level, "objid "DOSTID", magic 0x%08x, pattern %#x\n",
  57                      POSTID(&oi), le32_to_cpu(lmm->lmm_magic),
  58                      le32_to_cpu(lmm->lmm_pattern));
  59         CDEBUG_LIMIT(level, "stripe_size %u, stripe_count %u, layout_gen %u\n",
  60                      le32_to_cpu(lmm->lmm_stripe_size),
  61                      le16_to_cpu(lmm->lmm_stripe_count),
  62                      le16_to_cpu(lmm->lmm_layout_gen));
  63 }
  64
  65 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
  66                                  int stripe_count)
  67 {
  68         int i;
  69
  70         if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
  71                 CDEBUG_LIMIT(level,
  72                              "bad stripe_count %u > max_stripe_count %u\n",
  73                              stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
  74                 return;
  75         }
  76
  77         for (i = 0; i < stripe_count; ++i, ++lod) {
  78                 struct ost_id oi;
  79
  80                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
  81                 CDEBUG_LIMIT(level, "stripe %u idx %u subobj "DOSTID"\n", i,
  82                              le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
  83         }
  84 }
  85
  86 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
  87 {
  88         lov_dump_lmm_common(level, lmm);
  89         lov_dump_lmm_objects(level, lmm->lmm_objects,
  90                              le16_to_cpu(lmm->lmm_stripe_count));
  91 }
  92
  93 void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm)
  94 {
  95         lov_dump_lmm_common(level, lmm);
  96         CDEBUG_LIMIT(level, "pool_name "LOV_POOLNAMEF"\n", lmm->lmm_pool_name);
  97         lov_dump_lmm_objects(level, lmm->lmm_objects,
  98                              le16_to_cpu(lmm->lmm_stripe_count));
  99 }
 100
 101 void lov_dump_lmm(int level, void *lmm)
 102 {
 103         int magic;
 104
 105         magic = le32_to_cpu(((struct lov_mds_md *)lmm)->lmm_magic);
 106         switch (magic) {
 107         case LOV_MAGIC_V1:
 108                 lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)lmm);
 109                 break;
 110         case LOV_MAGIC_V3:
 111                 lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)lmm);
 112                 break;
 113         default:
 114                 CDEBUG_LIMIT(level, "unrecognized lmm_magic %x, assuming %x\n",
 115                              magic, LOV_MAGIC_V1);
 116                 lov_dump_lmm_common(level, lmm);
 117                 break;
 118         }
 119 }
 120
 121 /**
 122  * Pack LOV striping metadata for disk storage format (in little
 123  * endian byte order).
 124  *
 125  * This follows the getxattr() conventions. If \a buf_size is zero
 126  * then return the size needed. If \a buf_size is too small then
 127  * return -ERANGE. Otherwise return the size of the result.
 128  */
 129 ssize_t lov_lsm_pack_v1v3(const struct lov_stripe_md *lsm, void *buf,
 130                           size_t buf_size)
 131 {
 132         struct lov_mds_md_v1 *lmmv1 = buf;
 133         struct lov_mds_md_v3 *lmmv3 = buf;
 134         struct lov_ost_data_v1 *lmm_objects;
 135         size_t lmm_size;
 136         unsigned int i;
 137
 138         ENTRY;
 139
 140         lmm_size = lov_mds_md_size(lsm->lsm_entries[0]->lsme_stripe_count,
 141                                    lsm->lsm_magic);
 142         if (buf_size == 0)
 143                 RETURN(lmm_size);
 144
 145         if (buf_size < lmm_size)
 146                 RETURN(-ERANGE);
 147
 148         /*
 149          * lmmv1 and lmmv3 point to the same struct and have the
 150          * same first fields
 151          */
 152         lmmv1->lmm_magic = cpu_to_le32(lsm->lsm_magic);
 153         lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
 154         lmmv1->lmm_stripe_size = cpu_to_le32(
 155                                 lsm->lsm_entries[0]->lsme_stripe_size);
 156         lmmv1->lmm_stripe_count = cpu_to_le16(
 157                                 lsm->lsm_entries[0]->lsme_stripe_count);
 158         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_entries[0]->lsme_pattern);
 159         lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
 160
 161         if (lsm->lsm_magic == LOV_MAGIC_V3) {
 162                 CLASSERT(sizeof(lsm->lsm_entries[0]->lsme_pool_name) ==
 163                          sizeof(lmmv3->lmm_pool_name));
 164                 strlcpy(lmmv3->lmm_pool_name,
 165                         lsm->lsm_entries[0]->lsme_pool_name,
 166                         sizeof(lmmv3->lmm_pool_name));
 167                 lmm_objects = lmmv3->lmm_objects;
 168         } else {
 169                 lmm_objects = lmmv1->lmm_objects;
 170         }
 171
 172         if (lsm->lsm_is_released)
 173                 RETURN(lmm_size);
 174
 175         for (i = 0; i < lsm->lsm_entries[0]->lsme_stripe_count; i++) {
 176                 struct lov_oinfo *loi = lsm->lsm_entries[0]->lsme_oinfo[i];
 177
 178                 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
 179                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
 180                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
 181         }
 182
 183         RETURN(lmm_size);
 184 }
 185
 186 ssize_t lov_lsm_pack(const struct lov_stripe_md *lsm, void *buf,
 187                      size_t buf_size)
 188 {
 189         struct lov_comp_md_v1 *lcmv1 = buf;
 190         struct lov_comp_md_entry_v1 *lcme;
 191         struct lov_ost_data_v1 *lmm_objects;
 192         size_t lmm_size;
 193         unsigned int entry;
 194         unsigned int offset;
 195         unsigned int size;
 196         unsigned int i;
 197
 198         ENTRY;
 199
 200         if (lsm->lsm_magic == LOV_MAGIC_V1 || lsm->lsm_magic == LOV_MAGIC_V3)
 201                 return lov_lsm_pack_v1v3(lsm, buf, buf_size);
 202
 203         lmm_size = lov_comp_md_size(lsm);
 204         if (buf_size == 0)
 205                 RETURN(lmm_size);
 206
 207         if (buf_size < lmm_size)
 208                 RETURN(-ERANGE);
 209
 210         lcmv1->lcm_magic = cpu_to_le32(lsm->lsm_magic);
 211         lcmv1->lcm_size = cpu_to_le32(lmm_size);
 212         lcmv1->lcm_layout_gen = cpu_to_le32(lsm->lsm_layout_gen);
 213         lcmv1->lcm_flags = cpu_to_le16(lsm->lsm_flags);
 214         lcmv1->lcm_mirror_count = cpu_to_le16(lsm->lsm_mirror_count);
 215         lcmv1->lcm_entry_count = cpu_to_le16(lsm->lsm_entry_count);
 216
 217         offset = sizeof(*lcmv1) + sizeof(*lcme) * lsm->lsm_entry_count;
 218
 219         for (entry = 0; entry < lsm->lsm_entry_count; entry++) {
 220                 struct lov_stripe_md_entry *lsme;
 221                 struct lov_mds_md *lmm;
 222                 __u16 stripe_count;
 223
 224                 lsme = lsm->lsm_entries[entry];
 225                 lcme = &lcmv1->lcm_entries[entry];
 226
 227                 lcme->lcme_id = cpu_to_le32(lsme->lsme_id);
 228                 lcme->lcme_flags = cpu_to_le32(lsme->lsme_flags);
 229                 if (lsme->lsme_flags & LCME_FL_NOSYNC)
 230                         lcme->lcme_timestamp =
 231                                 cpu_to_le64(lsme->lsme_timestamp);
 232                 lcme->lcme_extent.e_start =
 233                         cpu_to_le64(lsme->lsme_extent.e_start);
 234                 lcme->lcme_extent.e_end =
 235                         cpu_to_le64(lsme->lsme_extent.e_end);
 236                 lcme->lcme_offset = cpu_to_le32(offset);
 237
 238                 lmm = (struct lov_mds_md *)((char *)lcmv1 + offset);
 239                 lmm->lmm_magic = cpu_to_le32(lsme->lsme_magic);
 240                 /* lmm->lmm_oi not set */
 241                 lmm->lmm_pattern = cpu_to_le32(lsme->lsme_pattern);
 242                 lmm->lmm_stripe_size = cpu_to_le32(lsme->lsme_stripe_size);
 243                 lmm->lmm_stripe_count = cpu_to_le16(lsme->lsme_stripe_count);
 244                 lmm->lmm_layout_gen = cpu_to_le16(lsme->lsme_layout_gen);
 245
 246                 if (lsme->lsme_magic == LOV_MAGIC_V3) {
 247                         struct lov_mds_md_v3 *lmmv3 =
 248                                                 (struct lov_mds_md_v3 *)lmm;
 249
 250                         strlcpy(lmmv3->lmm_pool_name, lsme->lsme_pool_name,
 251                                 sizeof(lmmv3->lmm_pool_name));
 252                         lmm_objects = lmmv3->lmm_objects;
 253                 } else {
 254                         lmm_objects =
 255                                 ((struct lov_mds_md_v1 *)lmm)->lmm_objects;
 256                 }
 257
 258                 if (lsme_inited(lsme) &&
 259                     !(lsme->lsme_pattern & LOV_PATTERN_F_RELEASED))
 260                         stripe_count = lsme->lsme_stripe_count;
 261                 else
 262                         stripe_count = 0;
 263
 264                 for (i = 0; i < stripe_count; i++) {
 265                         struct lov_oinfo *loi = lsme->lsme_oinfo[i];
 266
 267                         ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
 268                         lmm_objects[i].l_ost_gen =
 269                                         cpu_to_le32(loi->loi_ost_gen);
 270                         lmm_objects[i].l_ost_idx =
 271                                         cpu_to_le32(loi->loi_ost_idx);
 272                 }
 273
 274                 size = lov_mds_md_size(stripe_count, lsme->lsme_magic);
 275                 lcme->lcme_size = cpu_to_le32(size);
 276                 offset += size;
 277         } /* for each layout component */
 278
 279         RETURN(lmm_size);
 280 }
 281
 282 /* Find the max stripecount we should use */
 283 __u16 lov_get_stripe_count(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
 284 {
 285         __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
 286
 287         if (!stripe_count)
 288                 stripe_count = lov->desc.ld_default_stripe_count;
 289         if (stripe_count > lov->desc.ld_active_tgt_count)
 290                 stripe_count = lov->desc.ld_active_tgt_count;
 291         if (!stripe_count)
 292                 stripe_count = 1;
 293
 294         /*
 295          * stripe count is based on whether ldiskfs can handle
 296          * larger EA sizes
 297          */
 298         if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
 299             lov->lov_ocd.ocd_max_easize)
 300                 max_stripes = lov_mds_md_max_stripe_count(
 301                         lov->lov_ocd.ocd_max_easize, magic);
 302
 303         if (stripe_count > max_stripes)
 304                 stripe_count = max_stripes;
 305
 306         return stripe_count;
 307 }
 308
 309 int lov_free_memmd(struct lov_stripe_md **lsmp)
 310 {
 311         struct lov_stripe_md *lsm = *lsmp;
 312         int refc;
 313
 314         *lsmp = NULL;
 315         refc = atomic_dec_return(&lsm->lsm_refc);
 316         LASSERT(refc >= 0);
 317         if (refc == 0)
 318                 lsm_free(lsm);
 319
 320         return refc;
 321 }
 322
 323 /*
 324  * Unpack LOV object metadata from disk storage.  It is packed in LE byte
 325  * order and is opaque to the networking layer.
 326  */
 327 struct lov_stripe_md *lov_unpackmd(struct lov_obd *lov, void *buf,
 328                                    size_t buf_size)
 329 {
 330         const struct lsm_operations *op;
 331         struct lov_stripe_md *lsm;
 332         u32 magic;
 333
 334         ENTRY;
 335
 336         if (buf_size < sizeof(magic))
 337                 RETURN(ERR_PTR(-EINVAL));
 338
 339         magic = le32_to_cpu(*(u32 *)buf);
 340         op = lsm_op_find(magic);
 341         if (!op)
 342                 RETURN(ERR_PTR(-EINVAL));
 343
 344         lsm = op->lsm_unpackmd(lov, buf, buf_size);
 345
 346         RETURN(lsm);
 347 }
 348
 349 /*
 350  * Retrieve object striping information.
 351  *
 352  * @lump is a pointer to an in-core struct with lmm_ost_count indicating
 353  * the maximum number of OST indices which will fit in the user buffer.
 354  * lmm_magic must be LOV_USER_MAGIC.
 355  *
 356  * If @size > 0, User specified limited buffer size, usually the buffer is from
 357  * ll_lov_setstripe(), and the buffer can only hold basic layout template info.
 358  */
 359 int lov_getstripe(const struct lu_env *env, struct lov_object *obj,
 360                   struct lov_stripe_md *lsm, struct lov_user_md __user *lump,
 361                   size_t size)
 362 {
 363         /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
 364         struct lov_mds_md *lmmk, *lmm;
 365         struct lov_user_md_v1 lum;
 366         size_t lmmk_size;
 367         ssize_t lmm_size, lum_size = 0;
 368         static bool printed;
 369         int rc = 0;
 370
 371         ENTRY;
 372
 373         if (lsm->lsm_magic != LOV_MAGIC_V1 && lsm->lsm_magic != LOV_MAGIC_V3 &&
 374             lsm->lsm_magic != LOV_MAGIC_COMP_V1) {
 375                 CERROR("bad LSM MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
 376                        lsm->lsm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
 377                 GOTO(out, rc = -EIO);
 378         }
 379
 380         if (!printed) {
 381                 LCONSOLE_WARN("%s: using old ioctl(LL_IOC_LOV_GETSTRIPE) on "
 382                               DFID", use llapi_layout_get_by_path()\n",
 383                               current->comm,
 384                               PFID(&obj->lo_cl.co_lu.lo_header->loh_fid));
 385                 printed = true;
 386         }
 387
 388         lmmk_size = lov_comp_md_size(lsm);
 389
 390         OBD_ALLOC_LARGE(lmmk, lmmk_size);
 391         if (!lmmk)
 392                 GOTO(out, rc = -ENOMEM);
 393
 394         lmm_size = lov_lsm_pack(lsm, lmmk, lmmk_size);
 395         if (lmm_size < 0)
 396                 GOTO(out_free, rc = lmm_size);
 397
 398         if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
 399                 if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
 400                     lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
 401                         lustre_swab_lov_mds_md(lmmk);
 402                         lustre_swab_lov_user_md_objects(
 403                                 (struct lov_user_ost_data *)lmmk->lmm_objects,
 404                                 lmmk->lmm_stripe_count);
 405                 } else if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
 406                         lustre_swab_lov_comp_md_v1(
 407                                         (struct lov_comp_md_v1 *)lmmk);
 408                 }
 409         }
 410
 411         /*
 412          * Legacy appication passes limited buffer, we need to figure out
 413          * the user buffer size by the passed in lmm_stripe_count.
 414          */
 415         if (copy_from_user(&lum, lump, sizeof(struct lov_user_md_v1)))
 416                 GOTO(out_free, rc = -EFAULT);
 417
 418         if (lum.lmm_magic == LOV_USER_MAGIC_V1 ||
 419             lum.lmm_magic == LOV_USER_MAGIC_V3)
 420                 lum_size = lov_user_md_size(lum.lmm_stripe_count,
 421                                             lum.lmm_magic);
 422
 423         if (lum_size != 0) {
 424                 struct lov_mds_md *comp_md = lmmk;
 425
 426                 /*
 427                  * Legacy app (ADIO for instance) treats the layout as V1/V3
 428                  * blindly, we'd return a reasonable V1/V3 for them.
 429                  */
 430                 if (lmmk->lmm_magic == LOV_MAGIC_COMP_V1) {
 431                         struct lov_comp_md_v1 *comp_v1;
 432                         struct cl_object *cl_obj;
 433                         struct cl_attr attr;
 434                         int i;
 435
 436                         attr.cat_size = 0;
 437                         cl_obj = cl_object_top(&obj->lo_cl);
 438                         cl_object_attr_lock(cl_obj);
 439                         cl_object_attr_get(env, cl_obj, &attr);
 440                         cl_object_attr_unlock(cl_obj);
 441
 442                         /*
 443                          * return the last instantiated component if file size
 444                          * is non-zero, otherwise, return the last component.
 445                          */
 446                         comp_v1 = (struct lov_comp_md_v1 *)lmmk;
 447                         i = attr.cat_size == 0 ? comp_v1->lcm_entry_count : 0;
 448                         for (; i < comp_v1->lcm_entry_count; i++) {
 449                                 if (!(comp_v1->lcm_entries[i].lcme_flags &
 450                                                 LCME_FL_INIT))
 451                                         break;
 452                         }
 453                         if (i > 0)
 454                                 i--;
 455                         comp_md = (struct lov_mds_md *)((char *)comp_v1 +
 456                                         comp_v1->lcm_entries[i].lcme_offset);
 457                 }
 458
 459                 lmm = comp_md;
 460                 lmm_size = lum_size;
 461         } else {
 462                 lmm = lmmk;
 463                 lmm_size = lmmk_size;
 464         }
 465         /**
 466          * User specified limited buffer size, usually the buffer is
 467          * from ll_lov_setstripe(), and the buffer can only hold basic
 468          * layout template info.
 469          */
 470         if (size == 0 || size > lmm_size)
 471                 size = lmm_size;
 472         if (copy_to_user(lump, lmm, size))
 473                 GOTO(out_free, rc = -EFAULT);
 474
 475 out_free:
 476         OBD_FREE_LARGE(lmmk, lmmk_size);
 477 out:
 478         RETURN(rc);
 479 }