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  *
  31  * lustre/lov/lov_pack.c
  32  *
  33  * (Un)packing of OST/MDS requests
  34  *
  35  * Author: Andreas Dilger <adilger@clusterfs.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_LOV
  39
  40 #include <lustre_net.h>
  41 #include <lustre_swab.h>
  42 #include <obd.h>
  43 #include <obd_class.h>
  44 #include <obd_support.h>
  45
  46 #include "lov_cl_internal.h"
  47 #include "lov_internal.h"
  48
  49 void lov_dump_lmm_common(int level, void *lmmp)
  50 {
  51         struct lov_mds_md *lmm = lmmp;
  52         struct ost_id oi;
  53
  54         lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
  55         CDEBUG_LIMIT(level, "objid "DOSTID", magic 0x%08x, pattern %#x\n",
  56                      POSTID(&oi), le32_to_cpu(lmm->lmm_magic),
  57                      le32_to_cpu(lmm->lmm_pattern));
  58         CDEBUG_LIMIT(level, "stripe_size %u, stripe_count %u, layout_gen %u\n",
  59                      le32_to_cpu(lmm->lmm_stripe_size),
  60                      le16_to_cpu(lmm->lmm_stripe_count),
  61                      le16_to_cpu(lmm->lmm_layout_gen));
  62 }
  63
  64 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
  65                                  int stripe_count)
  66 {
  67         int i;
  68
  69         if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
  70                 CDEBUG_LIMIT(level,
  71                              "bad stripe_count %u > max_stripe_count %u\n",
  72                              stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
  73                 return;
  74         }
  75
  76         for (i = 0; i < stripe_count; ++i, ++lod) {
  77                 struct ost_id oi;
  78
  79                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
  80                 CDEBUG_LIMIT(level, "stripe %u idx %u subobj "DOSTID"\n", i,
  81                              le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
  82         }
  83 }
  84
  85 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
  86 {
  87         lov_dump_lmm_common(level, lmm);
  88         lov_dump_lmm_objects(level, lmm->lmm_objects,
  89                              le16_to_cpu(lmm->lmm_stripe_count));
  90 }
  91
  92 /**
  93  * Pack LOV striping metadata for disk storage format (in little
  94  * endian byte order).
  95  *
  96  * This follows the getxattr() conventions. If \a buf_size is zero
  97  * then return the size needed. If \a buf_size is too small then
  98  * return -ERANGE. Otherwise return the size of the result.
  99  */
 100 static ssize_t lov_lsm_pack_v1v3(const struct lov_stripe_md *lsm, void *buf,
 101                                  size_t buf_size)
 102 {
 103         struct lov_mds_md_v1 *lmmv1 = buf;
 104         struct lov_mds_md_v3 *lmmv3 = buf;
 105         struct lov_ost_data_v1 *lmm_objects;
 106         size_t lmm_size;
 107         unsigned int i;
 108
 109         ENTRY;
 110
 111         lmm_size = lov_mds_md_size(lsm->lsm_entries[0]->lsme_stripe_count,
 112                                    lsm->lsm_magic);
 113         if (buf_size == 0)
 114                 RETURN(lmm_size);
 115
 116         if (buf_size < lmm_size)
 117                 RETURN(-ERANGE);
 118
 119         /*
 120          * lmmv1 and lmmv3 point to the same struct and have the
 121          * same first fields
 122          */
 123         lmmv1->lmm_magic = cpu_to_le32(lsm->lsm_magic);
 124         lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
 125         lmmv1->lmm_stripe_size = cpu_to_le32(
 126                                 lsm->lsm_entries[0]->lsme_stripe_size);
 127         lmmv1->lmm_stripe_count = cpu_to_le16(
 128                                 lsm->lsm_entries[0]->lsme_stripe_count);
 129         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_entries[0]->lsme_pattern);
 130         lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
 131
 132         if (lsm->lsm_magic == LOV_MAGIC_V3) {
 133                 BUILD_BUG_ON(sizeof(lsm->lsm_entries[0]->lsme_pool_name) !=
 134                                     sizeof(lmmv3->lmm_pool_name));
 135                 strlcpy(lmmv3->lmm_pool_name,
 136                         lsm->lsm_entries[0]->lsme_pool_name,
 137                         sizeof(lmmv3->lmm_pool_name));
 138                 lmm_objects = lmmv3->lmm_objects;
 139         } else {
 140                 lmm_objects = lmmv1->lmm_objects;
 141         }
 142
 143         if (lsm->lsm_is_released)
 144                 RETURN(lmm_size);
 145
 146         for (i = 0; i < lsm->lsm_entries[0]->lsme_stripe_count; i++) {
 147                 struct lov_oinfo *loi = lsm->lsm_entries[0]->lsme_oinfo[i];
 148
 149                 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
 150                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
 151                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
 152         }
 153
 154         RETURN(lmm_size);
 155 }
 156
 157 static ssize_t lov_lsm_pack_foreign(const struct lov_stripe_md *lsm, void *buf,
 158                                     size_t buf_size)
 159 {
 160         struct lov_foreign_md *lfm = buf;
 161         size_t lfm_size;
 162
 163         lfm_size = lsm->lsm_foreign_size;
 164
 165         if (buf_size == 0)
 166                 RETURN(lfm_size);
 167
 168         /* if buffer too small return ERANGE but copy the size the
 169          * caller has requested anyway. This may be useful to get
 170          * only the header without the need to alloc the full size
 171          */
 172         if (buf_size < lfm_size) {
 173                 memcpy(lfm, lsm_foreign(lsm), buf_size);
 174                 RETURN(-ERANGE);
 175         }
 176
 177         /* full foreign LOV is already avail in its cache
 178          * no need to translate format fields to little-endian
 179          */
 180         memcpy(lfm, lsm_foreign(lsm), lsm->lsm_foreign_size);
 181
 182         RETURN(lfm_size);
 183 }
 184
 185 unsigned int lov_lsme_pack_foreign(struct lov_stripe_md_entry *lsme, void *lmm)
 186 {
 187         struct lov_foreign_md *lfm = (struct lov_foreign_md *)lmm;
 188
 189         lfm->lfm_magic = cpu_to_le32(lsme->lsme_magic);
 190         lfm->lfm_length = cpu_to_le32(lsme->lsme_length);
 191         lfm->lfm_type = cpu_to_le32(lsme->lsme_type);
 192         lfm->lfm_flags = cpu_to_le32(lsme->lsme_foreign_flags);
 193
 194         /* TODO: support for foreign layout other than HSM, i.e. DAOS. */
 195         if (lov_hsm_type_supported(lsme->lsme_type))
 196                 lov_foreign_hsm_to_le(lfm, &lsme->lsme_hsm);
 197
 198         return lov_foreign_md_size(lsme->lsme_length);
 199 }
 200
 201 unsigned int lov_lsme_pack_v1v3(struct lov_stripe_md_entry *lsme,
 202                                 struct lov_mds_md *lmm)
 203 {
 204         struct lov_ost_data_v1 *lmm_objects;
 205         __u16 stripe_count;
 206         unsigned int i;
 207
 208         lmm->lmm_magic = cpu_to_le32(lsme->lsme_magic);
 209         /* lmm->lmm_oi not set */
 210         lmm->lmm_pattern = cpu_to_le32(lsme->lsme_pattern);
 211         lmm->lmm_stripe_size = cpu_to_le32(lsme->lsme_stripe_size);
 212         lmm->lmm_stripe_count = cpu_to_le16(lsme->lsme_stripe_count);
 213         lmm->lmm_layout_gen = cpu_to_le16(lsme->lsme_layout_gen);
 214
 215         if (lsme->lsme_magic == LOV_MAGIC_V3) {
 216                 struct lov_mds_md_v3 *lmmv3 = (struct lov_mds_md_v3 *)lmm;
 217
 218                 strlcpy(lmmv3->lmm_pool_name, lsme->lsme_pool_name,
 219                         sizeof(lmmv3->lmm_pool_name));
 220                 lmm_objects = lmmv3->lmm_objects;
 221         } else {
 222                 lmm_objects = ((struct lov_mds_md_v1 *)lmm)->lmm_objects;
 223         }
 224
 225         if (lsme_inited(lsme) && !(lsme->lsme_pattern & LOV_PATTERN_F_RELEASED))
 226                 stripe_count = lsme->lsme_stripe_count;
 227         else
 228                 stripe_count = 0;
 229
 230         for (i = 0; i < stripe_count; i++) {
 231                 struct lov_oinfo *loi = lsme->lsme_oinfo[i];
 232
 233                 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
 234                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
 235                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
 236         }
 237
 238         return lov_mds_md_size(stripe_count, lsme->lsme_magic);
 239 }
 240
 241 ssize_t lov_lsm_pack(const struct lov_stripe_md *lsm, void *buf,
 242                      size_t buf_size)
 243 {
 244         struct lov_comp_md_v1 *lcmv1 = buf;
 245         struct lov_comp_md_entry_v1 *lcme;
 246         size_t lmm_size;
 247         unsigned int entry;
 248         unsigned int offset;
 249         unsigned int size;
 250
 251         ENTRY;
 252
 253         if (lsm->lsm_magic == LOV_MAGIC_V1 || lsm->lsm_magic == LOV_MAGIC_V3)
 254                 return lov_lsm_pack_v1v3(lsm, buf, buf_size);
 255
 256         if (lsm->lsm_magic == LOV_MAGIC_FOREIGN)
 257                 return lov_lsm_pack_foreign(lsm, buf, buf_size);
 258
 259         lmm_size = lov_comp_md_size(lsm);
 260         if (buf_size == 0)
 261                 RETURN(lmm_size);
 262
 263         if (buf_size < lmm_size)
 264                 RETURN(-ERANGE);
 265
 266         lcmv1->lcm_magic = cpu_to_le32(lsm->lsm_magic);
 267         lcmv1->lcm_size = cpu_to_le32(lmm_size);
 268         lcmv1->lcm_layout_gen = cpu_to_le32(lsm->lsm_layout_gen);
 269         lcmv1->lcm_flags = cpu_to_le16(lsm->lsm_flags);
 270         lcmv1->lcm_mirror_count = cpu_to_le16(lsm->lsm_mirror_count);
 271         lcmv1->lcm_entry_count = cpu_to_le16(lsm->lsm_entry_count);
 272
 273         offset = sizeof(*lcmv1) + sizeof(*lcme) * lsm->lsm_entry_count;
 274
 275         for (entry = 0; entry < lsm->lsm_entry_count; entry++) {
 276                 struct lov_stripe_md_entry *lsme;
 277                 struct lov_mds_md *lmm;
 278
 279                 lsme = lsm->lsm_entries[entry];
 280                 lcme = &lcmv1->lcm_entries[entry];
 281
 282                 lcme->lcme_id = cpu_to_le32(lsme->lsme_id);
 283                 lcme->lcme_flags = cpu_to_le32(lsme->lsme_flags);
 284                 if (lsme->lsme_flags & LCME_FL_NOSYNC)
 285                         lcme->lcme_timestamp =
 286                                 cpu_to_le64(lsme->lsme_timestamp);
 287                 lcme->lcme_extent.e_start =
 288                         cpu_to_le64(lsme->lsme_extent.e_start);
 289                 lcme->lcme_extent.e_end =
 290                         cpu_to_le64(lsme->lsme_extent.e_end);
 291                 lcme->lcme_offset = cpu_to_le32(offset);
 292
 293                 lmm = (struct lov_mds_md *)((char *)lcmv1 + offset);
 294                 if (lsme->lsme_magic == LOV_MAGIC_FOREIGN)
 295                         size = lov_lsme_pack_foreign(lsme, lmm);
 296                 else
 297                         size = lov_lsme_pack_v1v3(lsme, lmm);
 298                 lcme->lcme_size = cpu_to_le32(size);
 299                 offset += size;
 300         } /* for each layout component */
 301
 302         RETURN(lmm_size);
 303 }
 304
 305 /* Find the max stripecount we should use */
 306 __u16 lov_get_stripe_count(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
 307 {
 308         __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
 309
 310         if (!stripe_count)
 311                 stripe_count = lov->desc.ld_default_stripe_count;
 312         if (stripe_count > lov->desc.ld_active_tgt_count)
 313                 stripe_count = lov->desc.ld_active_tgt_count;
 314         if (!stripe_count)
 315                 stripe_count = 1;
 316
 317         /*
 318          * stripe count is based on whether ldiskfs can handle
 319          * larger EA sizes
 320          */
 321         if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
 322             lov->lov_ocd.ocd_max_easize)
 323                 max_stripes = lov_mds_md_max_stripe_count(
 324                         lov->lov_ocd.ocd_max_easize, magic);
 325
 326         if (stripe_count > max_stripes)
 327                 stripe_count = max_stripes;
 328
 329         return stripe_count;
 330 }
 331
 332 int lov_free_memmd(struct lov_stripe_md **lsmp)
 333 {
 334         struct lov_stripe_md *lsm = *lsmp;
 335         int refc;
 336
 337         *lsmp = NULL;
 338         refc = atomic_dec_return(&lsm->lsm_refc);
 339         LASSERT(refc >= 0);
 340         if (refc == 0)
 341                 lsm_free(lsm);
 342
 343         return refc;
 344 }
 345
 346 /*
 347  * Unpack LOV object metadata from disk storage.  It is packed in LE byte
 348  * order and is opaque to the networking layer.
 349  */
 350 struct lov_stripe_md *lov_unpackmd(struct lov_obd *lov, void *buf,
 351                                    size_t buf_size)
 352 {
 353         const struct lsm_operations *op;
 354         struct lov_stripe_md *lsm;
 355         u32 magic;
 356
 357         ENTRY;
 358
 359         if (buf_size < sizeof(magic))
 360                 RETURN(ERR_PTR(-EINVAL));
 361
 362         magic = le32_to_cpu(*(u32 *)buf);
 363         op = lsm_op_find(magic);
 364         if (!op)
 365                 RETURN(ERR_PTR(-EINVAL));
 366
 367         lsm = op->lsm_unpackmd(lov, buf, buf_size);
 368
 369         RETURN(lsm);
 370 }
 371
 372 /*
 373  * Retrieve object striping information.
 374  *
 375  * @lump is a pointer to an in-core struct with lmm_ost_count indicating
 376  * the maximum number of OST indices which will fit in the user buffer.
 377  * lmm_magic must be LOV_USER_MAGIC.
 378  *
 379  * If @size > 0, User specified limited buffer size, usually the buffer is from
 380  * ll_lov_setstripe(), and the buffer can only hold basic layout template info.
 381  */
 382 int lov_getstripe(const struct lu_env *env, struct lov_object *obj,
 383                   struct lov_stripe_md *lsm, struct lov_user_md __user *lump,
 384                   size_t size)
 385 {
 386         /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
 387         struct lov_mds_md *lmmk, *lmm;
 388         struct lov_foreign_md *lfm;
 389         struct lov_user_md_v1 lum;
 390         size_t lmmk_size, lum_size = 0;
 391         ssize_t lmm_size;
 392         int rc = 0;
 393
 394         ENTRY;
 395
 396         if (lsm->lsm_magic != LOV_MAGIC_V1 && lsm->lsm_magic != LOV_MAGIC_V3 &&
 397             lsm->lsm_magic != LOV_MAGIC_COMP_V1 &&
 398             lsm->lsm_magic != LOV_MAGIC_FOREIGN) {
 399                 CERROR("bad LSM MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
 400                        lsm->lsm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
 401                 GOTO(out, rc = -EIO);
 402         }
 403
 404         lmmk_size = lov_comp_md_size(lsm);
 405
 406         OBD_ALLOC_LARGE(lmmk, lmmk_size);
 407         if (!lmmk)
 408                 GOTO(out, rc = -ENOMEM);
 409
 410         lmm_size = lov_lsm_pack(lsm, lmmk, lmmk_size);
 411         if (lmm_size < 0)
 412                 GOTO(out_free, rc = lmm_size);
 413
 414         if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
 415                 if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
 416                     lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
 417                         lustre_swab_lov_mds_md(lmmk);
 418                         lustre_swab_lov_user_md_objects(
 419                                 (struct lov_user_ost_data *)lmmk->lmm_objects,
 420                                 lmmk->lmm_stripe_count);
 421                 } else if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
 422                         lustre_swab_lov_comp_md_v1(
 423                                         (struct lov_comp_md_v1 *)lmmk);
 424                 } else if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_FOREIGN)) {
 425                         lfm = (struct lov_foreign_md *)lmmk;
 426                         __swab32s(&lfm->lfm_magic);
 427                         __swab32s(&lfm->lfm_length);
 428                         __swab32s(&lfm->lfm_type);
 429                         __swab32s(&lfm->lfm_flags);
 430                 }
 431         }
 432
 433         /*
 434          * Legacy appication passes limited buffer, we need to figure out
 435          * the user buffer size by the passed in lmm_stripe_count.
 436          */
 437         if (lsm->lsm_magic != LOV_MAGIC_FOREIGN)
 438                 if (copy_from_user(&lum, lump, sizeof(struct lov_user_md_v1)))
 439                         GOTO(out_free, rc = -EFAULT);
 440
 441         if (lum.lmm_magic == LOV_USER_MAGIC_V1 ||
 442             lum.lmm_magic == LOV_USER_MAGIC_V3)
 443                 lum_size = lov_user_md_size(lum.lmm_stripe_count,
 444                                             lum.lmm_magic);
 445
 446         if (lum_size != 0) {
 447                 struct lov_mds_md *comp_md = lmmk;
 448
 449                 /*
 450                  * Legacy app (ADIO for instance) treats the layout as V1/V3
 451                  * blindly, we'd return a reasonable V1/V3 for them.
 452                  */
 453                 if (lmmk->lmm_magic == LOV_MAGIC_COMP_V1) {
 454                         struct lov_comp_md_v1 *comp_v1;
 455                         struct cl_object *cl_obj;
 456                         struct cl_attr attr;
 457                         int i;
 458
 459                         attr.cat_size = 0;
 460                         cl_obj = cl_object_top(&obj->lo_cl);
 461                         cl_object_attr_lock(cl_obj);
 462                         cl_object_attr_get(env, cl_obj, &attr);
 463                         cl_object_attr_unlock(cl_obj);
 464
 465                         /*
 466                          * return the last instantiated component if file size
 467                          * is non-zero, otherwise, return the last component.
 468                          */
 469                         comp_v1 = (struct lov_comp_md_v1 *)lmmk;
 470                         i = attr.cat_size == 0 ? comp_v1->lcm_entry_count : 0;
 471                         for (; i < comp_v1->lcm_entry_count; i++) {
 472                                 if (!(comp_v1->lcm_entries[i].lcme_flags &
 473                                                 LCME_FL_INIT))
 474                                         break;
 475                         }
 476                         if (i > 0)
 477                                 i--;
 478                         comp_md = (struct lov_mds_md *)((char *)comp_v1 +
 479                                         comp_v1->lcm_entries[i].lcme_offset);
 480                         lum_size = comp_v1->lcm_entries[i].lcme_size;
 481                 }
 482
 483                 lmm = comp_md;
 484                 lmm_size = min(lum_size, lmmk_size);
 485         } else {
 486                 lmm = lmmk;
 487                 lmm_size = lmmk_size;
 488         }
 489
 490         /**
 491          * User specified limited buffer size, usually the buffer is
 492          * from ll_lov_setstripe(), and the buffer can only hold basic
 493          * layout template info.
 494          */
 495         if (size == 0 || size > lmm_size)
 496                 size = lmm_size;
 497         if (copy_to_user(lump, lmm, size))
 498                 GOTO(out_free, rc = -EFAULT);
 499
 500 out_free:
 501         OBD_FREE_LARGE(lmmk, lmmk_size);
 502 out:
 503         RETURN(rc);
 504 }