lustre/llite/dir.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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/dir.c
  37  *
  38  * Directory code for lustre client.
  39  */
  40
  41 #include <linux/fs.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <linux/version.h>
  45 #include <asm/uaccess.h>
  46 #include <linux/buffer_head.h>   // for wait_on_buffer
  47 #include <linux/pagevec.h>
  48
  49 #define DEBUG_SUBSYSTEM S_LLITE
  50
  51 #include <lustre/lustre_idl.h>
  52 #include <obd_support.h>
  53 #include <obd_class.h>
  54 #include <lustre_ioctl.h>
  55 #include <lustre_lib.h>
  56 #include <lustre_dlm.h>
  57 #include <lustre_fid.h>
  58 #include "llite_internal.h"
  59
  60 /*
  61  * (new) readdir implementation overview.
  62  *
  63  * Original lustre readdir implementation cached exact copy of raw directory
  64  * pages on the client. These pages were indexed in client page cache by
  65  * logical offset in the directory file. This design, while very simple and
  66  * intuitive had some inherent problems:
  67  *
  68  *     . it implies that byte offset to the directory entry serves as a
  69  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  70  *     ext3/htree directory entries may move due to splits, and more
  71  *     importantly,
  72  *
  73  *     . it is incompatible with the design of split directories for cmd3,
  74  *     that assumes that names are distributed across nodes based on their
  75  *     hash, and so readdir should be done in hash order.
  76  *
  77  * New readdir implementation does readdir in hash order, and uses hash of a
  78  * file name as a telldir/seekdir cookie. This led to number of complications:
  79  *
  80  *     . hash is not unique, so it cannot be used to index cached directory
  81  *     pages on the client (note, that it requires a whole pageful of hash
  82  *     collided entries to cause two pages to have identical hashes);
  83  *
  84  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  85  *     entry cookie. ext3/htree has the same problem and lustre implementation
  86  *     mimics their solution: seekdir(hash) positions directory at the first
  87  *     entry with the given hash.
  88  *
  89  * Client side.
  90  *
  91  * 0. caching
  92  *
  93  * Client caches directory pages using hash of the first entry as an index. As
  94  * noted above hash is not unique, so this solution doesn't work as is:
  95  * special processing is needed for "page hash chains" (i.e., sequences of
  96  * pages filled with entries all having the same hash value).
  97  *
  98  * First, such chains have to be detected. To this end, server returns to the
  99  * client the hash of the first entry on the page next to one returned. When
 100  * client detects that this hash is the same as hash of the first entry on the
 101  * returned page, page hash collision has to be handled. Pages in the
 102  * hash chain, except first one, are termed "overflow pages".
 103  *
 104  * Solution to index uniqueness problem is to not cache overflow
 105  * pages. Instead, when page hash collision is detected, all overflow pages
 106  * from emerging chain are immediately requested from the server and placed in
 107  * a special data structure (struct ll_dir_chain). This data structure is used
 108  * by ll_readdir() to process entries from overflow pages. When readdir
 109  * invocation finishes, overflow pages are discarded. If page hash collision
 110  * chain weren't completely processed, next call to readdir will again detect
 111  * page hash collision, again read overflow pages in, process next portion of
 112  * entries and again discard the pages. This is not as wasteful as it looks,
 113  * because, given reasonable hash, page hash collisions are extremely rare.
 114  *
 115  * 1. directory positioning
 116  *
 117  * When seekdir(hash) is called, original
 118  *
 119  *
 120  *
 121  *
 122  *
 123  *
 124  *
 125  *
 126  * Server.
 127  *
 128  * identification of and access to overflow pages
 129  *
 130  * page format
 131  *
 132  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 133  * a header lu_dirpage which describes the start/end hash, and whether this
 134  * page is empty (contains no dir entry) or hash collide with next page.
 135  * After client receives reply, several pages will be integrated into dir page
 136  * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 137  * lu_dirpage for this integrated page will be adjusted. See
 138  * mdc_adjust_dirpages().
 139  *
 140  */
 141 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
 142                              __u64 offset, struct ll_dir_chain *chain)
 143 {
 144         struct md_callback      cb_op;
 145         struct page             *page;
 146         int                     rc;
 147
 148         cb_op.md_blocking_ast = ll_md_blocking_ast;
 149         rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
 150         if (rc != 0)
 151                 return ERR_PTR(rc);
 152
 153         return page;
 154 }
 155
 156 void ll_release_page(struct inode *inode, struct page *page,
 157                      bool remove)
 158 {
 159         kunmap(page);
 160
 161         /* Always remove the page for striped dir, because the page is
 162          * built from temporarily in LMV layer */
 163         if (inode != NULL && S_ISDIR(inode->i_mode) &&
 164             ll_i2info(inode)->lli_lsm_md != NULL) {
 165                 __free_page(page);
 166                 return;
 167         }
 168
 169         if (remove) {
 170                 lock_page(page);
 171                 if (likely(page->mapping != NULL))
 172                         truncate_complete_page(page->mapping, page);
 173                 unlock_page(page);
 174         }
 175         page_cache_release(page);
 176 }
 177
 178 #ifdef HAVE_DIR_CONTEXT
 179 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 180                 struct dir_context *ctx)
 181 {
 182 #else
 183 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 184                 void *cookie, filldir_t filldir)
 185 {
 186 #endif
 187         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 188         __u64                 pos        = *ppos;
 189         bool                  is_api32 = ll_need_32bit_api(sbi);
 190         bool                  is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 191         struct page          *page;
 192         struct ll_dir_chain   chain;
 193         bool                  done = false;
 194         int                   rc = 0;
 195         ENTRY;
 196
 197         ll_dir_chain_init(&chain);
 198
 199         page = ll_get_dir_page(inode, op_data, pos, &chain);
 200
 201         while (rc == 0 && !done) {
 202                 struct lu_dirpage *dp;
 203                 struct lu_dirent  *ent;
 204                 __u64 hash;
 205                 __u64 next;
 206
 207                 if (IS_ERR(page)) {
 208                         rc = PTR_ERR(page);
 209                         break;
 210                 }
 211
 212                 hash = MDS_DIR_END_OFF;
 213                 dp = page_address(page);
 214                 for (ent = lu_dirent_start(dp); ent != NULL && !done;
 215                      ent = lu_dirent_next(ent)) {
 216                         __u16          type;
 217                         int            namelen;
 218                         struct lu_fid  fid;
 219                         __u64          lhash;
 220                         __u64          ino;
 221
 222                         hash = le64_to_cpu(ent->lde_hash);
 223                         if (hash < pos)
 224                                 /*
 225                                  * Skip until we find target hash
 226                                  * value.
 227                                  */
 228                                 continue;
 229
 230                         namelen = le16_to_cpu(ent->lde_namelen);
 231                         if (namelen == 0)
 232                                 /*
 233                                  * Skip dummy record.
 234                                  */
 235                                 continue;
 236
 237                         if (is_api32 && is_hash64)
 238                                 lhash = hash >> 32;
 239                         else
 240                                 lhash = hash;
 241                         fid_le_to_cpu(&fid, &ent->lde_fid);
 242                         ino = cl_fid_build_ino(&fid, is_api32);
 243                         type = ll_dirent_type_get(ent);
 244                         /* For 'll_nfs_get_name_filldir()', it will try
 245                          * to access the 'ent' through its 'lde_name',
 246                          * so the parameter 'name' for 'filldir()' must
 247                          * be part of the 'ent'. */
 248 #ifdef HAVE_DIR_CONTEXT
 249                         ctx->pos = lhash;
 250                         done = !dir_emit(ctx, ent->lde_name, namelen, ino,
 251                                          type);
 252 #else
 253                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 254                                        ino, type);
 255 #endif
 256                 }
 257
 258                 if (done) {
 259                         pos = hash;
 260                         ll_release_page(inode, page, false);
 261                         break;
 262                 }
 263
 264                 next = le64_to_cpu(dp->ldp_hash_end);
 265                 pos = next;
 266                 if (pos == MDS_DIR_END_OFF) {
 267                         /*
 268                          * End of directory reached.
 269                          */
 270                         done = 1;
 271                         ll_release_page(inode, page, false);
 272                 } else {
 273                         /*
 274                          * Normal case: continue to the next
 275                          * page.
 276                          */
 277                         ll_release_page(inode, page,
 278                                         le32_to_cpu(dp->ldp_flags) &
 279                                         LDF_COLLIDE);
 280                         next = pos;
 281                         page = ll_get_dir_page(inode, op_data, pos,
 282                                                &chain);
 283                 }
 284         }
 285 #ifdef HAVE_DIR_CONTEXT
 286         ctx->pos = pos;
 287 #else
 288         *ppos = pos;
 289 #endif
 290         ll_dir_chain_fini(&chain);
 291         RETURN(rc);
 292 }
 293
 294 #ifdef HAVE_DIR_CONTEXT
 295 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 296 #else
 297 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 298 #endif
 299 {
 300         struct inode            *inode  = filp->f_dentry->d_inode;
 301         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 302         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 303         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 304         int                     api32   = ll_need_32bit_api(sbi);
 305         struct md_op_data       *op_data;
 306         __u64                   pos;
 307         int                     rc;
 308         ENTRY;
 309
 310         if (lfd != NULL)
 311                 pos = lfd->lfd_pos;
 312         else
 313                 pos = 0;
 314
 315         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 316                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 317                inode, (unsigned long)pos, i_size_read(inode), api32);
 318
 319         if (pos == MDS_DIR_END_OFF)
 320                 /*
 321                  * end-of-file.
 322                  */
 323                 GOTO(out, rc = 0);
 324
 325         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 326                                      LUSTRE_OPC_ANY, inode);
 327         if (IS_ERR(op_data))
 328                 GOTO(out, rc = PTR_ERR(op_data));
 329
 330         if (unlikely(op_data->op_mea1 != NULL)) {
 331                 /* This is only needed for striped dir to fill ..,
 332                  * see lmv_read_entry */
 333                 if (filp->f_dentry->d_parent != NULL &&
 334                     filp->f_dentry->d_parent->d_inode != NULL) {
 335                         __u64 ibits = MDS_INODELOCK_UPDATE;
 336                         struct inode *parent =
 337                                 filp->f_dentry->d_parent->d_inode;
 338
 339                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 340                                 op_data->op_fid3 = *ll_inode2fid(parent);
 341                 }
 342
 343                 /* If it can not find in cache, do lookup .. on the master
 344                  * object */
 345                 if (fid_is_zero(&op_data->op_fid3)) {
 346                         rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
 347                         if (rc != 0) {
 348                                 ll_finish_md_op_data(op_data);
 349                                 RETURN(rc);
 350                         }
 351                 }
 352         }
 353         op_data->op_max_pages = sbi->ll_md_brw_pages;
 354 #ifdef HAVE_DIR_CONTEXT
 355         ctx->pos = pos;
 356         rc = ll_dir_read(inode, &pos, op_data, ctx);
 357         pos = ctx->pos;
 358 #else
 359         rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
 360 #endif
 361         if (lfd != NULL)
 362                 lfd->lfd_pos = pos;
 363
 364         if (pos == MDS_DIR_END_OFF) {
 365                 if (api32)
 366                         pos = LL_DIR_END_OFF_32BIT;
 367                 else
 368                         pos = LL_DIR_END_OFF;
 369         } else {
 370                 if (api32 && hash64)
 371                         pos = pos >> 32;
 372         }
 373 #ifdef HAVE_DIR_CONTEXT
 374         ctx->pos = pos;
 375 #else
 376         filp->f_pos = pos;
 377 #endif
 378         ll_finish_md_op_data(op_data);
 379         filp->f_version = inode->i_version;
 380
 381 out:
 382         if (!rc)
 383                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 384
 385         RETURN(rc);
 386 }
 387
 388 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 389 {
 390         struct mgs_send_param *msp;
 391         int rc = 0;
 392
 393         OBD_ALLOC_PTR(msp);
 394         if (!msp)
 395                 return -ENOMEM;
 396
 397         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 398         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 399                                 sizeof(struct mgs_send_param), msp, NULL);
 400         if (rc)
 401                 CERROR("Failed to set parameter: %d\n", rc);
 402         OBD_FREE_PTR(msp);
 403
 404         return rc;
 405 }
 406
 407 /**
 408  * Create striped directory with specified stripe(@lump)
 409  *
 410  * param[in]parent      the parent of the directory.
 411  * param[in]lump        the specified stripes.
 412  * param[in]dirname     the name of the directory.
 413  * param[in]mode        the specified mode of the directory.
 414  *
 415  * retval               =0 if striped directory is being created successfully.
 416  *                      <0 if the creation is failed.
 417  */
 418 static int ll_dir_setdirstripe(struct inode *parent, struct lmv_user_md *lump,
 419                                const char *dirname, umode_t mode)
 420 {
 421         struct ptlrpc_request *request = NULL;
 422         struct md_op_data *op_data;
 423         struct ll_sb_info *sbi = ll_i2sbi(parent);
 424         int err;
 425         ENTRY;
 426
 427         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 428                 RETURN(-EINVAL);
 429
 430         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
 431                "stripe_offset %d, stripe_count: %u\n",
 432                PFID(ll_inode2fid(parent)), parent, dirname,
 433                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 434
 435         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 436                 lustre_swab_lmv_user_md(lump);
 437
 438         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 439                 mode &= ~current_umask();
 440         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 441         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 442                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 443                                      lump);
 444         if (IS_ERR(op_data))
 445                 GOTO(err_exit, err = PTR_ERR(op_data));
 446
 447         op_data->op_cli_flags |= CLI_SET_MEA;
 448         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 449                         from_kuid(&init_user_ns, current_fsuid()),
 450                         from_kgid(&init_user_ns, current_fsgid()),
 451                         cfs_curproc_cap_pack(), 0, &request);
 452         ll_finish_md_op_data(op_data);
 453         if (err)
 454                 GOTO(err_exit, err);
 455 err_exit:
 456         ptlrpc_req_finished(request);
 457         return err;
 458 }
 459
 460 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 461                      int set_default)
 462 {
 463         struct ll_sb_info *sbi = ll_i2sbi(inode);
 464         struct md_op_data *op_data;
 465         struct ptlrpc_request *req = NULL;
 466         int rc = 0;
 467         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 468         struct obd_device *mgc = lsi->lsi_mgc;
 469         int lum_size;
 470         ENTRY;
 471
 472         if (lump != NULL) {
 473                 /*
 474                  * This is coming from userspace, so should be in
 475                  * local endian.  But the MDS would like it in little
 476                  * endian, so we swab it before we send it.
 477                  */
 478                 switch (lump->lmm_magic) {
 479                 case LOV_USER_MAGIC_V1: {
 480                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 481                                 lustre_swab_lov_user_md_v1(lump);
 482                         lum_size = sizeof(struct lov_user_md_v1);
 483                         break;
 484                 }
 485                 case LOV_USER_MAGIC_V3: {
 486                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 487                                 lustre_swab_lov_user_md_v3(
 488                                         (struct lov_user_md_v3 *)lump);
 489                         lum_size = sizeof(struct lov_user_md_v3);
 490                         break;
 491                 }
 492                 case LMV_USER_MAGIC: {
 493                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 494                                 lustre_swab_lmv_user_md(
 495                                         (struct lmv_user_md *)lump);
 496                         lum_size = sizeof(struct lmv_user_md);
 497                         break;
 498                 }
 499                 default: {
 500                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 501                                         " %#08x != %#08x nor %#08x\n",
 502                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 503                                         LOV_USER_MAGIC_V3);
 504                         RETURN(-EINVAL);
 505                 }
 506                 }
 507         } else {
 508                 lum_size = sizeof(struct lov_user_md_v1);
 509         }
 510
 511         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 512                                      LUSTRE_OPC_ANY, NULL);
 513         if (IS_ERR(op_data))
 514                 RETURN(PTR_ERR(op_data));
 515
 516         /* swabbing is done in lov_setstripe() on server side */
 517         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
 518         ll_finish_md_op_data(op_data);
 519         ptlrpc_req_finished(req);
 520         if (rc) {
 521                 if (rc != -EPERM && rc != -EACCES)
 522                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 523         }
 524
 525         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 526          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 527          need the make the distiction between the 2 versions */
 528         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 529                 char *param = NULL;
 530                 char *buf;
 531
 532                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 533                 if (param == NULL)
 534                         GOTO(end, rc = -ENOMEM);
 535
 536                 buf = param;
 537                 /* Get fsname and assume devname to be -MDT0000. */
 538                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 539                 strcat(buf, "-MDT0000.lov");
 540                 buf += strlen(buf);
 541
 542                 /* Set root stripesize */
 543                 sprintf(buf, ".stripesize=%u",
 544                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 545                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 546                 if (rc)
 547                         GOTO(end, rc);
 548
 549                 /* Set root stripecount */
 550                 sprintf(buf, ".stripecount=%hd",
 551                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 552                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 553                 if (rc)
 554                         GOTO(end, rc);
 555
 556                 /* Set root stripeoffset */
 557                 sprintf(buf, ".stripeoffset=%hd",
 558                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 559                         (typeof(lump->lmm_stripe_offset))(-1));
 560                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 561
 562 end:
 563                 if (param != NULL)
 564                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 565         }
 566         RETURN(rc);
 567 }
 568
 569 /**
 570  * This function will be used to get default LOV/LMV/Default LMV
 571  * @valid will be used to indicate which stripe it will retrieve
 572  *      OBD_MD_MEA              LMV stripe EA
 573  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 574  *      otherwise               Default LOV EA.
 575  * Each time, it can only retrieve 1 stripe EA
 576  **/
 577 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 578                      struct ptlrpc_request **request, u64 valid)
 579 {
 580         struct ll_sb_info *sbi = ll_i2sbi(inode);
 581         struct mdt_body   *body;
 582         struct lov_mds_md *lmm = NULL;
 583         struct ptlrpc_request *req = NULL;
 584         int rc, lmm_size;
 585         struct md_op_data *op_data;
 586         ENTRY;
 587
 588         rc = ll_get_default_mdsize(sbi, &lmm_size);
 589         if (rc)
 590                 RETURN(rc);
 591
 592         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 593                                      0, lmm_size, LUSTRE_OPC_ANY,
 594                                      NULL);
 595         if (IS_ERR(op_data))
 596                 RETURN(PTR_ERR(op_data));
 597
 598         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 599         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 600         ll_finish_md_op_data(op_data);
 601         if (rc < 0) {
 602                 CDEBUG(D_INFO, "md_getattr failed on inode "
 603                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 604                 GOTO(out, rc);
 605         }
 606
 607         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 608         LASSERT(body != NULL);
 609
 610         lmm_size = body->mbo_eadatasize;
 611
 612         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 613             lmm_size == 0) {
 614                 GOTO(out, rc = -ENODATA);
 615         }
 616
 617         lmm = req_capsule_server_sized_get(&req->rq_pill,
 618                                            &RMF_MDT_MD, lmm_size);
 619         LASSERT(lmm != NULL);
 620
 621         /*
 622          * This is coming from the MDS, so is probably in
 623          * little endian.  We convert it to host endian before
 624          * passing it to userspace.
 625          */
 626         /* We don't swab objects for directories */
 627         switch (le32_to_cpu(lmm->lmm_magic)) {
 628         case LOV_MAGIC_V1:
 629                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 630                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 631                 break;
 632         case LOV_MAGIC_V3:
 633                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 634                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 635                 break;
 636         case LMV_MAGIC_V1:
 637                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 638                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 639                 break;
 640         case LMV_USER_MAGIC:
 641                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 642                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 643                 break;
 644         default:
 645                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 646                 rc = -EPROTO;
 647         }
 648 out:
 649         *plmm = lmm;
 650         *plmm_size = lmm_size;
 651         *request = req;
 652         return rc;
 653 }
 654
 655 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 656 {
 657         struct md_op_data       *op_data;
 658         int                     rc;
 659         int                     mdt_index;
 660         ENTRY;
 661
 662         OBD_ALLOC_PTR(op_data);
 663         if (op_data == NULL)
 664                 RETURN(-ENOMEM);
 665
 666         op_data->op_flags |= MF_GET_MDT_IDX;
 667         op_data->op_fid1 = *fid;
 668         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 669         mdt_index = op_data->op_mds;
 670         OBD_FREE_PTR(op_data);
 671         if (rc < 0)
 672                 RETURN(rc);
 673
 674         RETURN(mdt_index);
 675 }
 676
 677 /*
 678  *  Get MDT index for the inode.
 679  */
 680 int ll_get_mdt_idx(struct inode *inode)
 681 {
 682         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 683 }
 684
 685 /**
 686  * Generic handler to do any pre-copy work.
 687  *
 688  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 689  * first information for it that real work has started.
 690  *
 691  * Moreover, for a ARCHIVE request, it will sample the file data version and
 692  * store it in \a copy.
 693  *
 694  * \return 0 on success.
 695  */
 696 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 697 {
 698         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 699         struct hsm_progress_kernel       hpk;
 700         int                              rc;
 701         ENTRY;
 702
 703         /* Forge a hsm_progress based on data from copy. */
 704         hpk.hpk_fid = copy->hc_hai.hai_fid;
 705         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 706         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 707         hpk.hpk_extent.length = 0;
 708         hpk.hpk_flags = 0;
 709         hpk.hpk_errval = 0;
 710         hpk.hpk_data_version = 0;
 711
 712
 713         /* For archive request, we need to read the current file version. */
 714         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 715                 struct inode    *inode;
 716                 __u64            data_version = 0;
 717
 718                 /* Get inode for this fid */
 719                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 720                 if (IS_ERR(inode)) {
 721                         hpk.hpk_flags |= HP_FLAG_RETRY;
 722                         /* hpk_errval is >= 0 */
 723                         hpk.hpk_errval = -PTR_ERR(inode);
 724                         GOTO(progress, rc = PTR_ERR(inode));
 725                 }
 726
 727                 /* Read current file data version */
 728                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 729                 iput(inode);
 730                 if (rc != 0) {
 731                         CDEBUG(D_HSM, "Could not read file data version of "
 732                                       DFID" (rc = %d). Archive request ("
 733                                       LPX64") could not be done.\n",
 734                                       PFID(&copy->hc_hai.hai_fid), rc,
 735                                       copy->hc_hai.hai_cookie);
 736                         hpk.hpk_flags |= HP_FLAG_RETRY;
 737                         /* hpk_errval must be >= 0 */
 738                         hpk.hpk_errval = -rc;
 739                         GOTO(progress, rc);
 740                 }
 741
 742                 /* Store it the hsm_copy for later copytool use.
 743                  * Always modified even if no lsm. */
 744                 copy->hc_data_version = data_version;
 745         }
 746
 747 progress:
 748         /* On error, the request should be considered as completed */
 749         if (hpk.hpk_errval > 0)
 750                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 751         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 752                            &hpk, NULL);
 753
 754         RETURN(rc);
 755 }
 756
 757 /**
 758  * Generic handler to do any post-copy work.
 759  *
 760  * It will send the last hsm_progress update to coordinator to inform it
 761  * that copy is finished and whether it was successful or not.
 762  *
 763  * Moreover,
 764  * - for ARCHIVE request, it will sample the file data version and compare it
 765  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 766  *   will be considered as failed.
 767  * - for RESTORE request, it will sample the file data version and send it to
 768  *   coordinator which is useful if the file was imported as 'released'.
 769  *
 770  * \return 0 on success.
 771  */
 772 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 773 {
 774         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 775         struct hsm_progress_kernel       hpk;
 776         int                              rc;
 777         ENTRY;
 778
 779         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 780         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 781          * initialized if copy_end was called with copy == NULL.
 782          */
 783
 784         /* Forge a hsm_progress based on data from copy. */
 785         hpk.hpk_fid = copy->hc_hai.hai_fid;
 786         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 787         hpk.hpk_extent = copy->hc_hai.hai_extent;
 788         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 789         hpk.hpk_errval = copy->hc_errval;
 790         hpk.hpk_data_version = 0;
 791
 792         /* For archive request, we need to check the file data was not changed.
 793          *
 794          * For restore request, we need to send the file data version, this is
 795          * useful when the file was created using hsm_import.
 796          */
 797         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 798              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 799             (copy->hc_errval == 0)) {
 800                 struct inode    *inode;
 801                 __u64            data_version = 0;
 802
 803                 /* Get lsm for this fid */
 804                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 805                 if (IS_ERR(inode)) {
 806                         hpk.hpk_flags |= HP_FLAG_RETRY;
 807                         /* hpk_errval must be >= 0 */
 808                         hpk.hpk_errval = -PTR_ERR(inode);
 809                         GOTO(progress, rc = PTR_ERR(inode));
 810                 }
 811
 812                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 813                 iput(inode);
 814                 if (rc) {
 815                         CDEBUG(D_HSM, "Could not read file data version. "
 816                                       "Request could not be confirmed.\n");
 817                         if (hpk.hpk_errval == 0)
 818                                 hpk.hpk_errval = -rc;
 819                         GOTO(progress, rc);
 820                 }
 821
 822                 /* Store it the hsm_copy for later copytool use.
 823                  * Always modified even if no lsm. */
 824                 hpk.hpk_data_version = data_version;
 825
 826                 /* File could have been stripped during archiving, so we need
 827                  * to check anyway. */
 828                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 829                     (copy->hc_data_version != data_version)) {
 830                         CDEBUG(D_HSM, "File data version mismatched. "
 831                               "File content was changed during archiving. "
 832                                DFID", start:"LPX64" current:"LPX64"\n",
 833                                PFID(&copy->hc_hai.hai_fid),
 834                                copy->hc_data_version, data_version);
 835                         /* File was changed, send error to cdt. Do not ask for
 836                          * retry because if a file is modified frequently,
 837                          * the cdt will loop on retried archive requests.
 838                          * The policy engine will ask for a new archive later
 839                          * when the file will not be modified for some tunable
 840                          * time */
 841                         /* we do not notify caller */
 842                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 843                         /* hpk_errval must be >= 0 */
 844                         hpk.hpk_errval = EBUSY;
 845                 }
 846
 847         }
 848
 849 progress:
 850         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 851                            &hpk, NULL);
 852
 853         RETURN(rc);
 854 }
 855
 856
 857 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 858                           const void __user *data, size_t size)
 859 {
 860         void *copy;
 861         int rc;
 862
 863         OBD_ALLOC(copy, size);
 864         if (copy == NULL)
 865                 return -ENOMEM;
 866
 867         if (copy_from_user(copy, data, size)) {
 868                 rc = -EFAULT;
 869                 goto out;
 870         }
 871
 872         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 873 out:
 874         OBD_FREE(copy, size);
 875
 876         return rc;
 877 }
 878
 879 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 880 {
 881         int cmd = qctl->qc_cmd;
 882         int type = qctl->qc_type;
 883         int id = qctl->qc_id;
 884         int valid = qctl->qc_valid;
 885         int rc = 0;
 886         ENTRY;
 887
 888         switch (cmd) {
 889         case LUSTRE_Q_INVALIDATE:
 890         case LUSTRE_Q_FINVALIDATE:
 891         case Q_QUOTAON:
 892         case Q_QUOTAOFF:
 893         case Q_SETQUOTA:
 894         case Q_SETINFO:
 895                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 896                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 897                         RETURN(-EPERM);
 898                 break;
 899         case Q_GETQUOTA:
 900                 if (((type == USRQUOTA &&
 901                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
 902                      (type == GRPQUOTA &&
 903                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
 904                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 905                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 906                         RETURN(-EPERM);
 907                 break;
 908         case Q_GETINFO:
 909                 break;
 910         default:
 911                 CERROR("unsupported quotactl op: %#x\n", cmd);
 912                 RETURN(-ENOTTY);
 913         }
 914
 915         if (valid != QC_GENERAL) {
 916                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 917                         RETURN(-EOPNOTSUPP);
 918
 919                 if (cmd == Q_GETINFO)
 920                         qctl->qc_cmd = Q_GETOINFO;
 921                 else if (cmd == Q_GETQUOTA)
 922                         qctl->qc_cmd = Q_GETOQUOTA;
 923                 else
 924                         RETURN(-EINVAL);
 925
 926                 switch (valid) {
 927                 case QC_MDTIDX:
 928                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 929                                            sizeof(*qctl), qctl, NULL);
 930                         break;
 931                 case QC_OSTIDX:
 932                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 933                                            sizeof(*qctl), qctl, NULL);
 934                         break;
 935                 case QC_UUID:
 936                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 937                                            sizeof(*qctl), qctl, NULL);
 938                         if (rc == -EAGAIN)
 939                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 940                                                    sbi->ll_dt_exp,
 941                                                    sizeof(*qctl), qctl, NULL);
 942                         break;
 943                 default:
 944                         rc = -EINVAL;
 945                         break;
 946                 }
 947
 948                 if (rc)
 949                         RETURN(rc);
 950
 951                 qctl->qc_cmd = cmd;
 952         } else {
 953                 struct obd_quotactl *oqctl;
 954
 955                 OBD_ALLOC_PTR(oqctl);
 956                 if (oqctl == NULL)
 957                         RETURN(-ENOMEM);
 958
 959                 QCTL_COPY(oqctl, qctl);
 960                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
 961                 if (rc) {
 962                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
 963                                 oqctl->qc_cmd = Q_QUOTAOFF;
 964                                 obd_quotactl(sbi->ll_md_exp, oqctl);
 965                         }
 966                         OBD_FREE_PTR(oqctl);
 967                         RETURN(rc);
 968                 }
 969                 /* If QIF_SPACE is not set, client should collect the
 970                  * space usage from OSSs by itself */
 971                 if (cmd == Q_GETQUOTA &&
 972                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
 973                     !oqctl->qc_dqblk.dqb_curspace) {
 974                         struct obd_quotactl *oqctl_tmp;
 975
 976                         OBD_ALLOC_PTR(oqctl_tmp);
 977                         if (oqctl_tmp == NULL)
 978                                 GOTO(out, rc = -ENOMEM);
 979
 980                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
 981                         oqctl_tmp->qc_id = oqctl->qc_id;
 982                         oqctl_tmp->qc_type = oqctl->qc_type;
 983
 984                         /* collect space usage from OSTs */
 985                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 986                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
 987                         if (!rc || rc == -EREMOTEIO) {
 988                                 oqctl->qc_dqblk.dqb_curspace =
 989                                         oqctl_tmp->qc_dqblk.dqb_curspace;
 990                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
 991                         }
 992
 993                         /* collect space & inode usage from MDTs */
 994                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 995                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
 996                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
 997                         if (!rc || rc == -EREMOTEIO) {
 998                                 oqctl->qc_dqblk.dqb_curspace +=
 999                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1000                                 oqctl->qc_dqblk.dqb_curinodes =
1001                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1002                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1003                         } else {
1004                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1005                         }
1006
1007                         OBD_FREE_PTR(oqctl_tmp);
1008                 }
1009 out:
1010                 QCTL_COPY(qctl, oqctl);
1011                 OBD_FREE_PTR(oqctl);
1012         }
1013
1014         RETURN(rc);
1015 }
1016
1017 static char *
1018 ll_getname(const char __user *filename)
1019 {
1020         int ret = 0, len;
1021         char *tmp = __getname();
1022
1023         if (!tmp)
1024                 return ERR_PTR(-ENOMEM);
1025
1026         len = strncpy_from_user(tmp, filename, PATH_MAX);
1027         if (len == 0)
1028                 ret = -ENOENT;
1029         else if (len > PATH_MAX)
1030                 ret = -ENAMETOOLONG;
1031
1032         if (ret) {
1033                 __putname(tmp);
1034                 tmp =  ERR_PTR(ret);
1035         }
1036         return tmp;
1037 }
1038
1039 #define ll_putname(filename) __putname(filename)
1040
1041 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1042 {
1043         struct inode *inode = file->f_dentry->d_inode;
1044         struct ll_sb_info *sbi = ll_i2sbi(inode);
1045         struct obd_ioctl_data *data;
1046         int rc = 0;
1047         ENTRY;
1048
1049         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1050                PFID(ll_inode2fid(inode)), inode, cmd);
1051
1052         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1053         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1054                 return -ENOTTY;
1055
1056         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1057         switch(cmd) {
1058         case FSFILT_IOC_GETFLAGS:
1059         case FSFILT_IOC_SETFLAGS:
1060                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1061         case FSFILT_IOC_GETVERSION_OLD:
1062         case FSFILT_IOC_GETVERSION:
1063                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1064         /* We need to special case any other ioctls we want to handle,
1065          * to send them to the MDS/OST as appropriate and to properly
1066          * network encode the arg field.
1067         case FSFILT_IOC_SETVERSION_OLD:
1068         case FSFILT_IOC_SETVERSION:
1069         */
1070         case LL_IOC_GET_MDTIDX: {
1071                 int mdtidx;
1072
1073                 mdtidx = ll_get_mdt_idx(inode);
1074                 if (mdtidx < 0)
1075                         RETURN(mdtidx);
1076
1077                 if (put_user((int)mdtidx, (int __user *)arg))
1078                         RETURN(-EFAULT);
1079
1080                 return 0;
1081         }
1082         case IOC_MDC_LOOKUP: {
1083                 int namelen, len = 0;
1084                 char *buf = NULL;
1085                 char *filename;
1086
1087                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1088                 if (rc != 0)
1089                         RETURN(rc);
1090                 data = (void *)buf;
1091
1092                 filename = data->ioc_inlbuf1;
1093                 namelen = strlen(filename);
1094                 if (namelen < 1) {
1095                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1096                         GOTO(out_free, rc = -EINVAL);
1097                 }
1098
1099                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL);
1100                 if (rc < 0) {
1101                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1102                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1103                                filename, rc);
1104                         GOTO(out_free, rc);
1105                 }
1106 out_free:
1107                 obd_ioctl_freedata(buf, len);
1108                 return rc;
1109         }
1110         case LL_IOC_LMV_SETSTRIPE: {
1111                 struct lmv_user_md  *lum;
1112                 char            *buf = NULL;
1113                 char            *filename;
1114                 int              namelen = 0;
1115                 int              lumlen = 0;
1116                 umode_t          mode;
1117                 int              len;
1118                 int              rc;
1119
1120                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1121                 if (rc)
1122                         RETURN(rc);
1123
1124                 data = (void *)buf;
1125                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1126                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1127                         GOTO(lmv_out_free, rc = -EINVAL);
1128
1129                 filename = data->ioc_inlbuf1;
1130                 namelen = data->ioc_inllen1;
1131
1132                 if (namelen < 1) {
1133                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1134                         GOTO(lmv_out_free, rc = -EINVAL);
1135                 }
1136                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1137                 lumlen = data->ioc_inllen2;
1138
1139                 if (lum->lum_magic != LMV_USER_MAGIC ||
1140                     lumlen != sizeof(*lum)) {
1141                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1142                                filename, lum->lum_magic, lumlen, -EFAULT);
1143                         GOTO(lmv_out_free, rc = -EINVAL);
1144                 }
1145
1146 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1147                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1148 #else
1149                 mode = data->ioc_type;
1150 #endif
1151                 rc = ll_dir_setdirstripe(inode, lum, filename, mode);
1152 lmv_out_free:
1153                 obd_ioctl_freedata(buf, len);
1154                 RETURN(rc);
1155
1156         }
1157         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1158                 struct lmv_user_md        lum;
1159                 struct lmv_user_md __user *ulump =
1160                                         (struct lmv_user_md __user *)arg;
1161                 int                       rc;
1162
1163                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1164                         RETURN(-EFAULT);
1165
1166                 if (lum.lum_magic != LMV_USER_MAGIC)
1167                         RETURN(-EINVAL);
1168
1169                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1170
1171                 RETURN(rc);
1172         }
1173         case LL_IOC_LOV_SETSTRIPE: {
1174                 struct lov_user_md_v3 lumv3;
1175                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1176                 struct lov_user_md_v1 __user *lumv1p =
1177                         (struct lov_user_md_v1 __user *)arg;
1178                 struct lov_user_md_v3 __user *lumv3p =
1179                         (struct lov_user_md_v3 __user *)arg;
1180
1181                 int set_default = 0;
1182
1183                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1184                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1185                         sizeof(lumv3p->lmm_objects[0]));
1186                 /* first try with v1 which is smaller than v3 */
1187                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1188                         RETURN(-EFAULT);
1189
1190                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1191                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1192                                 RETURN(-EFAULT);
1193                 }
1194
1195                 if (inode->i_sb->s_root == file->f_dentry)
1196                         set_default = 1;
1197
1198                 /* in v1 and v3 cases lumv1 points to data */
1199                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1200
1201                 RETURN(rc);
1202         }
1203         case LL_IOC_LMV_GETSTRIPE: {
1204                 struct lmv_user_md __user *ulmv =
1205                                         (struct lmv_user_md __user *)arg;
1206                 struct lmv_user_md      lum;
1207                 struct ptlrpc_request   *request = NULL;
1208                 union lmv_mds_md        *lmm = NULL;
1209                 int                     lmmsize;
1210                 u64                     valid = 0;
1211                 struct lmv_user_md      *tmp = NULL;
1212                 int                     mdt_index;
1213                 int                     lum_size;
1214                 int                     stripe_count;
1215                 int                     i;
1216                 int                     rc;
1217
1218                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1219                         RETURN(-EFAULT);
1220
1221                 /* lum_magic will indicate which stripe the ioctl will like
1222                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1223                  * is for default LMV stripe */
1224                 if (lum.lum_magic == LMV_MAGIC_V1)
1225                         valid |= OBD_MD_MEA;
1226                 else if (lum.lum_magic == LMV_USER_MAGIC)
1227                         valid |= OBD_MD_DEFAULT_MEA;
1228                 else
1229                         RETURN(-EINVAL);
1230
1231                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1232                                       valid);
1233                 if (rc != 0)
1234                         GOTO(finish_req, rc);
1235
1236                 /* Get default LMV EA */
1237                 if (lum.lum_magic == LMV_USER_MAGIC) {
1238                         if (rc != 0)
1239                                 GOTO(finish_req, rc);
1240
1241                         if (lmmsize > sizeof(*ulmv))
1242                                 GOTO(finish_req, rc = -EINVAL);
1243
1244                         if (copy_to_user(ulmv, lmm, lmmsize))
1245                                 GOTO(finish_req, rc = -EFAULT);
1246
1247                         GOTO(finish_req, rc);
1248                 }
1249
1250                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1251                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1252                 OBD_ALLOC(tmp, lum_size);
1253                 if (tmp == NULL)
1254                         GOTO(finish_req, rc = -ENOMEM);
1255
1256                 mdt_index = ll_get_mdt_idx(inode);
1257                 if (mdt_index < 0)
1258                         GOTO(out_tmp, rc = -ENOMEM);
1259
1260                 tmp->lum_magic = LMV_MAGIC_V1;
1261                 tmp->lum_stripe_count = 0;
1262                 tmp->lum_stripe_offset = mdt_index;
1263                 for (i = 0; i < stripe_count; i++) {
1264                         struct lu_fid   fid;
1265
1266                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1267                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1268                         if (mdt_index < 0)
1269                                 GOTO(out_tmp, rc = mdt_index);
1270
1271                         tmp->lum_objects[i].lum_mds = mdt_index;
1272                         tmp->lum_objects[i].lum_fid = fid;
1273                         tmp->lum_stripe_count++;
1274                 }
1275
1276                 if (copy_to_user(ulmv, tmp, lum_size))
1277                         GOTO(out_tmp, rc = -EFAULT);
1278 out_tmp:
1279                 OBD_FREE(tmp, lum_size);
1280 finish_req:
1281                 ptlrpc_req_finished(request);
1282                 return rc;
1283         }
1284
1285         case LL_IOC_REMOVE_ENTRY: {
1286                 char            *filename = NULL;
1287                 int              namelen = 0;
1288                 int              rc;
1289
1290                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1291                  * unsupported server, which might crash the server(LU-2730),
1292                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1293                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1294                  * server will support REINT_RMENTRY XXX*/
1295                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1296                         RETURN(-ENOTSUPP);
1297
1298                 filename = ll_getname((const char __user *)arg);
1299                 if (IS_ERR(filename))
1300                         RETURN(PTR_ERR(filename));
1301
1302                 namelen = strlen(filename);
1303                 if (namelen < 1)
1304                         GOTO(out_rmdir, rc = -EINVAL);
1305
1306                 rc = ll_rmdir_entry(inode, filename, namelen);
1307 out_rmdir:
1308                 if (filename)
1309                         ll_putname(filename);
1310                 RETURN(rc);
1311         }
1312         case LL_IOC_LOV_SWAP_LAYOUTS:
1313                 RETURN(-EPERM);
1314         case IOC_OBD_STATFS:
1315                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1316         case LL_IOC_LOV_GETSTRIPE:
1317         case LL_IOC_MDC_GETINFO:
1318         case IOC_MDC_GETFILEINFO:
1319         case IOC_MDC_GETFILESTRIPE: {
1320                 struct ptlrpc_request *request = NULL;
1321                 struct lov_user_md __user *lump;
1322                 struct lov_mds_md *lmm = NULL;
1323                 struct mdt_body *body;
1324                 char *filename = NULL;
1325                 int lmmsize;
1326
1327                 if (cmd == IOC_MDC_GETFILEINFO ||
1328                     cmd == IOC_MDC_GETFILESTRIPE) {
1329                         filename = ll_getname((const char __user *)arg);
1330                         if (IS_ERR(filename))
1331                                 RETURN(PTR_ERR(filename));
1332
1333                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1334                                                       &lmmsize, &request);
1335                 } else {
1336                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1337                                               &request, 0);
1338                 }
1339
1340                 if (request) {
1341                         body = req_capsule_server_get(&request->rq_pill,
1342                                                       &RMF_MDT_BODY);
1343                         LASSERT(body != NULL);
1344                 } else {
1345                         GOTO(out_req, rc);
1346                 }
1347
1348                 if (rc < 0) {
1349                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1350                                                cmd == LL_IOC_MDC_GETINFO))
1351                                 GOTO(skip_lmm, rc = 0);
1352                         else
1353                                 GOTO(out_req, rc);
1354                 }
1355
1356                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1357                     cmd == LL_IOC_LOV_GETSTRIPE) {
1358                         lump = (struct lov_user_md __user *)arg;
1359                 } else {
1360                         struct lov_user_mds_data __user *lmdp;
1361                         lmdp = (struct lov_user_mds_data __user *)arg;
1362                         lump = &lmdp->lmd_lmm;
1363                 }
1364                 if (copy_to_user(lump, lmm, lmmsize)) {
1365                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1366                                 GOTO(out_req, rc = -EFAULT);
1367                         rc = -EOVERFLOW;
1368                 }
1369         skip_lmm:
1370                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1371                         struct lov_user_mds_data __user *lmdp;
1372                         lstat_t st = { 0 };
1373
1374                         st.st_dev       = inode->i_sb->s_dev;
1375                         st.st_mode      = body->mbo_mode;
1376                         st.st_nlink     = body->mbo_nlink;
1377                         st.st_uid       = body->mbo_uid;
1378                         st.st_gid       = body->mbo_gid;
1379                         st.st_rdev      = body->mbo_rdev;
1380                         st.st_size      = body->mbo_size;
1381                         st.st_blksize   = PAGE_CACHE_SIZE;
1382                         st.st_blocks    = body->mbo_blocks;
1383                         st.st_atime     = body->mbo_atime;
1384                         st.st_mtime     = body->mbo_mtime;
1385                         st.st_ctime     = body->mbo_ctime;
1386                         st.st_ino       = inode->i_ino;
1387
1388                         lmdp = (struct lov_user_mds_data __user *)arg;
1389                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1390                                 GOTO(out_req, rc = -EFAULT);
1391                 }
1392
1393                 EXIT;
1394         out_req:
1395                 ptlrpc_req_finished(request);
1396                 if (filename)
1397                         ll_putname(filename);
1398                 return rc;
1399         }
1400         case OBD_IOC_QUOTACHECK: {
1401                 struct obd_quotactl *oqctl;
1402                 int error = 0;
1403
1404                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1405                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1406                         RETURN(-EPERM);
1407
1408                 OBD_ALLOC_PTR(oqctl);
1409                 if (!oqctl)
1410                         RETURN(-ENOMEM);
1411                 oqctl->qc_type = arg;
1412                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1413                 if (rc < 0) {
1414                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1415                         error = rc;
1416                 }
1417
1418                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1419                 if (rc < 0)
1420                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1421
1422                 OBD_FREE_PTR(oqctl);
1423                 return error ?: rc;
1424         }
1425         case OBD_IOC_POLL_QUOTACHECK: {
1426                 struct if_quotacheck *check;
1427
1428                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1429                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1430                         RETURN(-EPERM);
1431
1432                 OBD_ALLOC_PTR(check);
1433                 if (!check)
1434                         RETURN(-ENOMEM);
1435
1436                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1437                                    NULL);
1438                 if (rc) {
1439                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1440                         if (copy_to_user((void __user *)arg, check,
1441                                              sizeof(*check)))
1442                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1443                         GOTO(out_poll, rc);
1444                 }
1445
1446                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1447                                    NULL);
1448                 if (rc) {
1449                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1450                         if (copy_to_user((void __user *)arg, check,
1451                                              sizeof(*check)))
1452                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1453                         GOTO(out_poll, rc);
1454                 }
1455         out_poll:
1456                 OBD_FREE_PTR(check);
1457                 RETURN(rc);
1458         }
1459 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0)
1460         case LL_IOC_QUOTACTL_18: {
1461                 /* copy the old 1.x quota struct for internal use, then copy
1462                  * back into old format struct.  For 1.8 compatibility. */
1463                 struct if_quotactl_18 *qctl_18;
1464                 struct if_quotactl *qctl_20;
1465
1466                 OBD_ALLOC_PTR(qctl_18);
1467                 if (!qctl_18)
1468                         RETURN(-ENOMEM);
1469
1470                 OBD_ALLOC_PTR(qctl_20);
1471                 if (!qctl_20)
1472                         GOTO(out_quotactl_18, rc = -ENOMEM);
1473
1474                 if (copy_from_user(qctl_18, (void __user *)arg,
1475                                    sizeof(*qctl_18)))
1476                         GOTO(out_quotactl_20, rc = -ENOMEM);
1477
1478                 QCTL_COPY(qctl_20, qctl_18);
1479                 qctl_20->qc_idx = 0;
1480
1481                 /* XXX: dqb_valid was borrowed as a flag to mark that
1482                  *      only mds quota is wanted */
1483                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1484                     qctl_18->qc_dqblk.dqb_valid) {
1485                         qctl_20->qc_valid = QC_MDTIDX;
1486                         qctl_20->qc_dqblk.dqb_valid = 0;
1487                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1488                         qctl_20->qc_valid = QC_UUID;
1489                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1490                 } else {
1491                         qctl_20->qc_valid = QC_GENERAL;
1492                 }
1493
1494                 rc = quotactl_ioctl(sbi, qctl_20);
1495
1496                 if (rc == 0) {
1497                         QCTL_COPY(qctl_18, qctl_20);
1498                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1499
1500                         if (copy_to_user((void __user *)arg, qctl_18,
1501                                              sizeof(*qctl_18)))
1502                                 rc = -EFAULT;
1503                 }
1504
1505         out_quotactl_20:
1506                 OBD_FREE_PTR(qctl_20);
1507         out_quotactl_18:
1508                 OBD_FREE_PTR(qctl_18);
1509                 RETURN(rc);
1510         }
1511 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0) */
1512         case OBD_IOC_QUOTACTL: {
1513                 struct if_quotactl *qctl;
1514
1515                 OBD_ALLOC_PTR(qctl);
1516                 if (!qctl)
1517                         RETURN(-ENOMEM);
1518
1519                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1520                         GOTO(out_quotactl, rc = -EFAULT);
1521
1522                 rc = quotactl_ioctl(sbi, qctl);
1523
1524                 if (rc == 0 &&
1525                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1526                         rc = -EFAULT;
1527
1528         out_quotactl:
1529                 OBD_FREE_PTR(qctl);
1530                 RETURN(rc);
1531         }
1532         case OBD_IOC_GETDTNAME:
1533         case OBD_IOC_GETMDNAME:
1534                 RETURN(ll_get_obd_name(inode, cmd, arg));
1535         case LL_IOC_FLUSHCTX:
1536                 RETURN(ll_flush_ctx(inode));
1537 #ifdef CONFIG_FS_POSIX_ACL
1538         case LL_IOC_RMTACL: {
1539             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1540                 inode == inode->i_sb->s_root->d_inode) {
1541                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1542
1543                 LASSERT(fd != NULL);
1544                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1545                 if (!rc)
1546                         fd->fd_flags |= LL_FILE_RMTACL;
1547                 RETURN(rc);
1548             } else
1549                 RETURN(0);
1550         }
1551 #endif
1552         case LL_IOC_GETOBDCOUNT: {
1553                 u32 count, vallen;
1554                 struct obd_export *exp;
1555
1556                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1557                         RETURN(-EFAULT);
1558
1559                 /* get ost count when count is zero, get mdt count otherwise */
1560                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1561                 vallen = sizeof(count);
1562                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1563                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1564                 if (rc) {
1565                         CERROR("get target count failed: %d\n", rc);
1566                         RETURN(rc);
1567                 }
1568
1569                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1570                         RETURN(-EFAULT);
1571
1572                 RETURN(0);
1573         }
1574         case LL_IOC_PATH2FID:
1575                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1576                                      sizeof(struct lu_fid)))
1577                         RETURN(-EFAULT);
1578                 RETURN(0);
1579         case LL_IOC_GET_CONNECT_FLAGS: {
1580                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1581                                      (void __user *)arg));
1582         }
1583         case OBD_IOC_CHANGELOG_SEND:
1584         case OBD_IOC_CHANGELOG_CLEAR:
1585                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1586                                     sizeof(struct ioc_changelog));
1587                 RETURN(rc);
1588         case OBD_IOC_FID2PATH:
1589                 RETURN(ll_fid2path(inode, (void __user *)arg));
1590         case LL_IOC_GETPARENT:
1591                 RETURN(ll_getparent(file, (void __user *)arg));
1592         case LL_IOC_FID2MDTIDX: {
1593                 struct obd_export *exp = ll_i2mdexp(inode);
1594                 struct lu_fid     fid;
1595                 __u32             index;
1596
1597                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1598                                    sizeof(fid)))
1599                         RETURN(-EFAULT);
1600
1601                 /* Call mdc_iocontrol */
1602                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1603                                    (__u32 __user *)&index);
1604                 if (rc != 0)
1605                         RETURN(rc);
1606
1607                 RETURN(index);
1608         }
1609         case LL_IOC_HSM_REQUEST: {
1610                 struct hsm_user_request *hur;
1611                 ssize_t                  totalsize;
1612
1613                 OBD_ALLOC_PTR(hur);
1614                 if (hur == NULL)
1615                         RETURN(-ENOMEM);
1616
1617                 /* We don't know the true size yet; copy the fixed-size part */
1618                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1619                         OBD_FREE_PTR(hur);
1620                         RETURN(-EFAULT);
1621                 }
1622
1623                 /* Compute the whole struct size */
1624                 totalsize = hur_len(hur);
1625                 OBD_FREE_PTR(hur);
1626                 if (totalsize < 0)
1627                         RETURN(-E2BIG);
1628
1629                 /* Final size will be more than double totalsize */
1630                 if (totalsize >= MDS_MAXREQSIZE / 3)
1631                         RETURN(-E2BIG);
1632
1633                 OBD_ALLOC_LARGE(hur, totalsize);
1634                 if (hur == NULL)
1635                         RETURN(-ENOMEM);
1636
1637                 /* Copy the whole struct */
1638                 if (copy_from_user(hur, (void __user *)arg, totalsize)) {
1639                         OBD_FREE_LARGE(hur, totalsize);
1640                         RETURN(-EFAULT);
1641                 }
1642
1643                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1644                         const struct lu_fid *fid;
1645                         struct inode *f;
1646                         int i;
1647
1648                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1649                                 fid = &hur->hur_user_item[i].hui_fid;
1650                                 f = search_inode_for_lustre(inode->i_sb, fid);
1651                                 if (IS_ERR(f)) {
1652                                         rc = PTR_ERR(f);
1653                                         break;
1654                                 }
1655
1656                                 rc = ll_hsm_release(f);
1657                                 iput(f);
1658                                 if (rc != 0)
1659                                         break;
1660                         }
1661                 } else {
1662                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1663                                            hur, NULL);
1664                 }
1665
1666                 OBD_FREE_LARGE(hur, totalsize);
1667
1668                 RETURN(rc);
1669         }
1670         case LL_IOC_HSM_PROGRESS: {
1671                 struct hsm_progress_kernel      hpk;
1672                 struct hsm_progress             hp;
1673
1674                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1675                         RETURN(-EFAULT);
1676
1677                 hpk.hpk_fid = hp.hp_fid;
1678                 hpk.hpk_cookie = hp.hp_cookie;
1679                 hpk.hpk_extent = hp.hp_extent;
1680                 hpk.hpk_flags = hp.hp_flags;
1681                 hpk.hpk_errval = hp.hp_errval;
1682                 hpk.hpk_data_version = 0;
1683
1684                 /* File may not exist in Lustre; all progress
1685                  * reported to Lustre root */
1686                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1687                                    NULL);
1688                 RETURN(rc);
1689         }
1690         case LL_IOC_HSM_CT_START:
1691                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1692                         RETURN(-EPERM);
1693
1694                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1695                                     sizeof(struct lustre_kernelcomm));
1696                 RETURN(rc);
1697
1698         case LL_IOC_HSM_COPY_START: {
1699                 struct hsm_copy *copy;
1700                 int              rc;
1701
1702                 OBD_ALLOC_PTR(copy);
1703                 if (copy == NULL)
1704                         RETURN(-ENOMEM);
1705                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1706                         OBD_FREE_PTR(copy);
1707                         RETURN(-EFAULT);
1708                 }
1709
1710                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1711                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1712                         rc = -EFAULT;
1713
1714                 OBD_FREE_PTR(copy);
1715                 RETURN(rc);
1716         }
1717         case LL_IOC_HSM_COPY_END: {
1718                 struct hsm_copy *copy;
1719                 int              rc;
1720
1721                 OBD_ALLOC_PTR(copy);
1722                 if (copy == NULL)
1723                         RETURN(-ENOMEM);
1724                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1725                         OBD_FREE_PTR(copy);
1726                         RETURN(-EFAULT);
1727                 }
1728
1729                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1730                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1731                         rc = -EFAULT;
1732
1733                 OBD_FREE_PTR(copy);
1734                 RETURN(rc);
1735         }
1736         case LL_IOC_MIGRATE: {
1737                 char            *buf = NULL;
1738                 const char      *filename;
1739                 int             namelen = 0;
1740                 int             len;
1741                 int             rc;
1742                 int             mdtidx;
1743
1744                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1745                 if (rc < 0)
1746                         RETURN(rc);
1747
1748                 data = (struct obd_ioctl_data *)buf;
1749                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1750                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1751                         GOTO(migrate_free, rc = -EINVAL);
1752
1753                 filename = data->ioc_inlbuf1;
1754                 namelen = data->ioc_inllen1;
1755                 /* \0 is packed at the end of filename */
1756                 if (namelen < 1 || namelen != strlen(filename) + 1)
1757                         GOTO(migrate_free, rc = -EINVAL);
1758
1759                 if (data->ioc_inllen2 != sizeof(mdtidx))
1760                         GOTO(migrate_free, rc = -EINVAL);
1761                 mdtidx = *(int *)data->ioc_inlbuf2;
1762
1763                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1764 migrate_free:
1765                 obd_ioctl_freedata(buf, len);
1766
1767                 RETURN(rc);
1768         }
1769         default:
1770                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1771                                      (void __user *)arg));
1772         }
1773 }
1774
1775 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1776 {
1777         struct inode *inode = file->f_mapping->host;
1778         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1779         struct ll_sb_info *sbi = ll_i2sbi(inode);
1780         int api32 = ll_need_32bit_api(sbi);
1781         loff_t ret = -EINVAL;
1782         ENTRY;
1783
1784         mutex_lock(&inode->i_mutex);
1785         switch (origin) {
1786                 case SEEK_SET:
1787                         break;
1788                 case SEEK_CUR:
1789                         offset += file->f_pos;
1790                         break;
1791                 case SEEK_END:
1792                         if (offset > 0)
1793                                 GOTO(out, ret);
1794                         if (api32)
1795                                 offset += LL_DIR_END_OFF_32BIT;
1796                         else
1797                                 offset += LL_DIR_END_OFF;
1798                         break;
1799                 default:
1800                         GOTO(out, ret);
1801         }
1802
1803         if (offset >= 0 &&
1804             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1805              (!api32 && offset <= LL_DIR_END_OFF))) {
1806                 if (offset != file->f_pos) {
1807                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1808                             (!api32 && offset == LL_DIR_END_OFF))
1809                                 fd->lfd_pos = MDS_DIR_END_OFF;
1810                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1811                                 fd->lfd_pos = offset << 32;
1812                         else
1813                                 fd->lfd_pos = offset;
1814                         file->f_pos = offset;
1815                         file->f_version = 0;
1816                 }
1817                 ret = offset;
1818         }
1819         GOTO(out, ret);
1820
1821 out:
1822         mutex_unlock(&inode->i_mutex);
1823         return ret;
1824 }
1825
1826 static int ll_dir_open(struct inode *inode, struct file *file)
1827 {
1828         ENTRY;
1829         RETURN(ll_file_open(inode, file));
1830 }
1831
1832 static int ll_dir_release(struct inode *inode, struct file *file)
1833 {
1834         ENTRY;
1835         RETURN(ll_file_release(inode, file));
1836 }
1837
1838 const struct file_operations ll_dir_operations = {
1839         .llseek         = ll_dir_seek,
1840         .open           = ll_dir_open,
1841         .release        = ll_dir_release,
1842         .read           = generic_read_dir,
1843 #ifdef HAVE_DIR_CONTEXT
1844         .iterate        = ll_iterate,
1845 #else
1846         .readdir        = ll_readdir,
1847 #endif
1848         .unlocked_ioctl = ll_dir_ioctl,
1849         .fsync          = ll_fsync,
1850 };