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, 2013, 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_lite.h>
  57 #include <lustre_dlm.h>
  58 #include <lustre_fid.h>
  59 #include "llite_internal.h"
  60
  61 /*
  62  * (new) readdir implementation overview.
  63  *
  64  * Original lustre readdir implementation cached exact copy of raw directory
  65  * pages on the client. These pages were indexed in client page cache by
  66  * logical offset in the directory file. This design, while very simple and
  67  * intuitive had some inherent problems:
  68  *
  69  *     . it implies that byte offset to the directory entry serves as a
  70  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  71  *     ext3/htree directory entries may move due to splits, and more
  72  *     importantly,
  73  *
  74  *     . it is incompatible with the design of split directories for cmd3,
  75  *     that assumes that names are distributed across nodes based on their
  76  *     hash, and so readdir should be done in hash order.
  77  *
  78  * New readdir implementation does readdir in hash order, and uses hash of a
  79  * file name as a telldir/seekdir cookie. This led to number of complications:
  80  *
  81  *     . hash is not unique, so it cannot be used to index cached directory
  82  *     pages on the client (note, that it requires a whole pageful of hash
  83  *     collided entries to cause two pages to have identical hashes);
  84  *
  85  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  86  *     entry cookie. ext3/htree has the same problem and lustre implementation
  87  *     mimics their solution: seekdir(hash) positions directory at the first
  88  *     entry with the given hash.
  89  *
  90  * Client side.
  91  *
  92  * 0. caching
  93  *
  94  * Client caches directory pages using hash of the first entry as an index. As
  95  * noted above hash is not unique, so this solution doesn't work as is:
  96  * special processing is needed for "page hash chains" (i.e., sequences of
  97  * pages filled with entries all having the same hash value).
  98  *
  99  * First, such chains have to be detected. To this end, server returns to the
 100  * client the hash of the first entry on the page next to one returned. When
 101  * client detects that this hash is the same as hash of the first entry on the
 102  * returned page, page hash collision has to be handled. Pages in the
 103  * hash chain, except first one, are termed "overflow pages".
 104  *
 105  * Solution to index uniqueness problem is to not cache overflow
 106  * pages. Instead, when page hash collision is detected, all overflow pages
 107  * from emerging chain are immediately requested from the server and placed in
 108  * a special data structure (struct ll_dir_chain). This data structure is used
 109  * by ll_readdir() to process entries from overflow pages. When readdir
 110  * invocation finishes, overflow pages are discarded. If page hash collision
 111  * chain weren't completely processed, next call to readdir will again detect
 112  * page hash collision, again read overflow pages in, process next portion of
 113  * entries and again discard the pages. This is not as wasteful as it looks,
 114  * because, given reasonable hash, page hash collisions are extremely rare.
 115  *
 116  * 1. directory positioning
 117  *
 118  * When seekdir(hash) is called, original
 119  *
 120  *
 121  *
 122  *
 123  *
 124  *
 125  *
 126  *
 127  * Server.
 128  *
 129  * identification of and access to overflow pages
 130  *
 131  * page format
 132  *
 133  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 134  * a header lu_dirpage which describes the start/end hash, and whether this
 135  * page is empty (contains no dir entry) or hash collide with next page.
 136  * After client receives reply, several pages will be integrated into dir page
 137  * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 138  * lu_dirpage for this integrated page will be adjusted. See
 139  * lmv_adjust_dirpages().
 140  *
 141  */
 142 /**
 143  * The following three APIs will be used by llite to iterate directory
 144  * entries from MDC dir page caches.
 145  *
 146  * ll_dir_entry_start(next) will lookup(return) entry by op_hash_offset.
 147  * To avoid extra memory allocation, the @entry will be pointed to
 148  * the dir entries in MDC page directly, so these pages can not be released
 149  * until the entry has been accessed in ll_readdir(or statahead).
 150  *
 151  * The iterate process will be
 152  *
 153  * ll_dir_entry_start: locate the page in MDC, and return the first entry.
 154  *                     hold the page.
 155  *
 156  * ll_dir_entry_next: return the next entry in the current page, if it reaches
 157  *                    to the end, release current page.
 158  *
 159  * ll_dir_entry_end: release the last page.
 160  **/
 161 struct lu_dirent *ll_dir_entry_start(struct inode *dir,
 162                                      struct md_op_data *op_data,
 163                                      struct page **ppage)
 164 {
 165         struct lu_dirent *entry;
 166         struct md_callback cb_op;
 167         int rc;
 168
 169         LASSERT(*ppage == NULL);
 170         cb_op.md_blocking_ast = ll_md_blocking_ast;
 171         rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
 172         if (rc != 0)
 173                 entry = ERR_PTR(rc);
 174         return entry;
 175 }
 176
 177 struct lu_dirent *ll_dir_entry_next(struct inode *dir,
 178                                     struct md_op_data *op_data,
 179                                     struct lu_dirent *ent,
 180                                     struct page **ppage)
 181 {
 182         struct lu_dirent *entry;
 183         struct md_callback cb_op;
 184         int rc;
 185
 186         LASSERT(*ppage != NULL);
 187         cb_op.md_blocking_ast = ll_md_blocking_ast;
 188         op_data->op_hash_offset = le64_to_cpu(ent->lde_hash);
 189         kunmap(*ppage);
 190         page_cache_release(*ppage);
 191         *ppage = NULL;
 192         rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
 193         if (rc != 0)
 194                 entry = ERR_PTR(rc);
 195         return entry;
 196 }
 197
 198 #ifdef HAVE_DIR_CONTEXT
 199 int ll_dir_read(struct inode *inode, struct md_op_data *op_data,
 200                 struct dir_context *ctx)
 201 {
 202 #else
 203 int ll_dir_read(struct inode *inode, struct md_op_data *op_data,
 204                 void *cookie, filldir_t filldir)
 205 {
 206 #endif
 207         struct ll_sb_info       *sbi = ll_i2sbi(inode);
 208         struct ll_dir_chain     chain;
 209         struct lu_dirent        *ent;
 210         int                     api32 = ll_need_32bit_api(sbi);
 211         int                     hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 212         int                     done = 0;
 213         int                     rc = 0;
 214         __u64                   hash = MDS_DIR_END_OFF;
 215         __u64                   last_hash = MDS_DIR_END_OFF;
 216         struct page             *page = NULL;
 217         ENTRY;
 218
 219         ll_dir_chain_init(&chain);
 220         for (ent = ll_dir_entry_start(inode, op_data, &page);
 221              ent != NULL && !IS_ERR(ent) && !done;
 222              ent = ll_dir_entry_next(inode, op_data, ent, &page)) {
 223                 __u16          type;
 224                 int            namelen;
 225                 struct lu_fid  fid;
 226                 __u64          lhash;
 227                 __u64          ino;
 228
 229                 hash = le64_to_cpu(ent->lde_hash);
 230                 if (hash < op_data->op_hash_offset)
 231                         /*
 232                          * Skip until we find target hash
 233                          * value.
 234                          */
 235                         continue;
 236                 namelen = le16_to_cpu(ent->lde_namelen);
 237                 if (namelen == 0)
 238                         /*
 239                          * Skip dummy record.
 240                          */
 241                         continue;
 242
 243                 if (api32 && hash64)
 244                         lhash = hash >> 32;
 245                 else
 246                         lhash = hash;
 247                 fid_le_to_cpu(&fid, &ent->lde_fid);
 248                 ino = cl_fid_build_ino(&fid, api32);
 249                 type = ll_dirent_type_get(ent);
 250
 251 #ifdef HAVE_DIR_CONTEXT
 252                 /* For 'll_nfs_get_name_filldir()', it will try
 253                  * to access the 'ent' through its 'lde_name',
 254                  * so the parameter 'name' for 'filldir()' must
 255                  * be part of the 'ent'. */
 256                 done = !dir_emit(ctx, ent->lde_name, namelen, ino, type);
 257 #else
 258                 done = filldir(cookie, ent->lde_name, namelen, lhash,
 259                                ino, type);
 260 #endif
 261                 if (done) {
 262                         if (op_data->op_hash_offset != MDS_DIR_END_OFF)
 263                                 op_data->op_hash_offset = last_hash;
 264                         break;
 265                 } else {
 266                         last_hash = hash;
 267                 }
 268         }
 269
 270         if (IS_ERR(ent))
 271                 rc = PTR_ERR(ent);
 272
 273         if (page != NULL) {
 274                 kunmap(page);
 275                 page_cache_release(page);
 276         }
 277
 278         ll_dir_chain_fini(&chain);
 279         RETURN(rc);
 280 }
 281
 282 #ifdef HAVE_DIR_CONTEXT
 283 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 284 #else
 285 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 286 #endif
 287 {
 288         struct inode            *inode  = filp->f_dentry->d_inode;
 289         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 290         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 291         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 292         int                     api32   = ll_need_32bit_api(sbi);
 293         struct md_op_data       *op_data;
 294         __u64                   pos;
 295         int                     rc;
 296         ENTRY;
 297
 298         if (lfd != NULL)
 299                 pos = lfd->lfd_pos;
 300         else
 301                 pos = 0;
 302
 303         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 304                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 305                inode, (unsigned long)pos, i_size_read(inode), api32);
 306
 307         if (pos == MDS_DIR_END_OFF)
 308                 /*
 309                  * end-of-file.
 310                  */
 311                 GOTO(out, rc = 0);
 312
 313         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 314                                      LUSTRE_OPC_ANY, inode);
 315         if (IS_ERR(op_data))
 316                 GOTO(out, rc = PTR_ERR(op_data));
 317
 318         op_data->op_hash_offset = pos;
 319         op_data->op_max_pages = sbi->ll_md_brw_pages;
 320 #ifdef HAVE_DIR_CONTEXT
 321         ctx->pos = pos;
 322         rc = ll_dir_read(inode, op_data, ctx);
 323         pos = ctx->pos;
 324 #else
 325         rc = ll_dir_read(inode, op_data, cookie, filldir);
 326 #endif
 327         if (lfd != NULL)
 328                 lfd->lfd_pos = op_data->op_hash_offset;
 329
 330         if (pos == MDS_DIR_END_OFF) {
 331                 if (api32)
 332                         pos = LL_DIR_END_OFF_32BIT;
 333                 else
 334                         pos = LL_DIR_END_OFF;
 335         } else {
 336                 if (api32 && hash64)
 337                         pos = op_data->op_hash_offset >> 32;
 338                 else
 339                         pos = op_data->op_hash_offset;
 340         }
 341 #ifdef HAVE_DIR_CONTEXT
 342         ctx->pos = pos;
 343 #else
 344         filp->f_pos = pos;
 345 #endif
 346         ll_finish_md_op_data(op_data);
 347         filp->f_version = inode->i_version;
 348
 349 out:
 350         if (!rc)
 351                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 352
 353         RETURN(rc);
 354 }
 355
 356 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 357 {
 358         struct mgs_send_param *msp;
 359         int rc = 0;
 360
 361         OBD_ALLOC_PTR(msp);
 362         if (!msp)
 363                 return -ENOMEM;
 364
 365         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 366         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 367                                 sizeof(struct mgs_send_param), msp, NULL);
 368         if (rc)
 369                 CERROR("Failed to set parameter: %d\n", rc);
 370         OBD_FREE_PTR(msp);
 371
 372         return rc;
 373 }
 374
 375 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 376                                const char *filename)
 377 {
 378         struct ptlrpc_request *request = NULL;
 379         struct md_op_data *op_data;
 380         struct ll_sb_info *sbi = ll_i2sbi(dir);
 381         int mode;
 382         int err;
 383         ENTRY;
 384
 385         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 386                 RETURN(-EINVAL);
 387
 388         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
 389                "stripe_offset %d, stripe_count: %u\n",
 390                PFID(ll_inode2fid(dir)), dir, filename,
 391                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 392
 393         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 394                 lustre_swab_lmv_user_md(lump);
 395
 396         mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
 397         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 398                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 399                                      lump);
 400         if (IS_ERR(op_data))
 401                 GOTO(err_exit, err = PTR_ERR(op_data));
 402
 403         op_data->op_cli_flags |= CLI_SET_MEA;
 404         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 405                         from_kuid(&init_user_ns, current_fsuid()),
 406                         from_kgid(&init_user_ns, current_fsgid()),
 407                         cfs_curproc_cap_pack(), 0, &request);
 408         ll_finish_md_op_data(op_data);
 409         if (err)
 410                 GOTO(err_exit, err);
 411 err_exit:
 412         ptlrpc_req_finished(request);
 413         return err;
 414 }
 415
 416 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 417                      int set_default)
 418 {
 419         struct ll_sb_info *sbi = ll_i2sbi(inode);
 420         struct md_op_data *op_data;
 421         struct ptlrpc_request *req = NULL;
 422         int rc = 0;
 423         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 424         struct obd_device *mgc = lsi->lsi_mgc;
 425         int lum_size;
 426         ENTRY;
 427
 428         if (lump != NULL) {
 429                 /*
 430                  * This is coming from userspace, so should be in
 431                  * local endian.  But the MDS would like it in little
 432                  * endian, so we swab it before we send it.
 433                  */
 434                 switch (lump->lmm_magic) {
 435                 case LOV_USER_MAGIC_V1: {
 436                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 437                                 lustre_swab_lov_user_md_v1(lump);
 438                         lum_size = sizeof(struct lov_user_md_v1);
 439                         break;
 440                 }
 441                 case LOV_USER_MAGIC_V3: {
 442                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 443                                 lustre_swab_lov_user_md_v3(
 444                                         (struct lov_user_md_v3 *)lump);
 445                         lum_size = sizeof(struct lov_user_md_v3);
 446                         break;
 447                 }
 448                 case LMV_USER_MAGIC: {
 449                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 450                                 lustre_swab_lmv_user_md(
 451                                         (struct lmv_user_md *)lump);
 452                         lum_size = sizeof(struct lmv_user_md);
 453                         break;
 454                 }
 455                 default: {
 456                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 457                                         " %#08x != %#08x nor %#08x\n",
 458                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 459                                         LOV_USER_MAGIC_V3);
 460                         RETURN(-EINVAL);
 461                 }
 462                 }
 463         } else {
 464                 lum_size = sizeof(struct lov_user_md_v1);
 465         }
 466
 467         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 468                                      LUSTRE_OPC_ANY, NULL);
 469         if (IS_ERR(op_data))
 470                 RETURN(PTR_ERR(op_data));
 471
 472         /* swabbing is done in lov_setstripe() on server side */
 473         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 474                         NULL, 0, &req, NULL);
 475         ll_finish_md_op_data(op_data);
 476         ptlrpc_req_finished(req);
 477         if (rc) {
 478                 if (rc != -EPERM && rc != -EACCES)
 479                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 480         }
 481
 482         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 483          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 484          need the make the distiction between the 2 versions */
 485         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 486                 char *param = NULL;
 487                 char *buf;
 488
 489                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 490                 if (param == NULL)
 491                         GOTO(end, rc = -ENOMEM);
 492
 493                 buf = param;
 494                 /* Get fsname and assume devname to be -MDT0000. */
 495                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 496                 strcat(buf, "-MDT0000.lov");
 497                 buf += strlen(buf);
 498
 499                 /* Set root stripesize */
 500                 sprintf(buf, ".stripesize=%u",
 501                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 502                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 503                 if (rc)
 504                         GOTO(end, rc);
 505
 506                 /* Set root stripecount */
 507                 sprintf(buf, ".stripecount=%hd",
 508                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 509                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 510                 if (rc)
 511                         GOTO(end, rc);
 512
 513                 /* Set root stripeoffset */
 514                 sprintf(buf, ".stripeoffset=%hd",
 515                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 516                         (typeof(lump->lmm_stripe_offset))(-1));
 517                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 518
 519 end:
 520                 if (param != NULL)
 521                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 522         }
 523         RETURN(rc);
 524 }
 525
 526 /**
 527  * This function will be used to get default LOV/LMV/Default LMV
 528  * @valid will be used to indicate which stripe it will retrieve
 529  *      OBD_MD_MEA              LMV stripe EA
 530  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 531  *      otherwise               Default LOV EA.
 532  * Each time, it can only retrieve 1 stripe EA
 533  **/
 534 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 535                      struct ptlrpc_request **request, obd_valid valid)
 536 {
 537         struct ll_sb_info *sbi = ll_i2sbi(inode);
 538         struct mdt_body   *body;
 539         struct lov_mds_md *lmm = NULL;
 540         struct ptlrpc_request *req = NULL;
 541         int rc, lmm_size;
 542         struct md_op_data *op_data;
 543         ENTRY;
 544
 545         rc = ll_get_default_mdsize(sbi, &lmm_size);
 546         if (rc)
 547                 RETURN(rc);
 548
 549         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 550                                      0, lmm_size, LUSTRE_OPC_ANY,
 551                                      NULL);
 552         if (IS_ERR(op_data))
 553                 RETURN(PTR_ERR(op_data));
 554
 555         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 556         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 557         ll_finish_md_op_data(op_data);
 558         if (rc < 0) {
 559                 CDEBUG(D_INFO, "md_getattr failed on inode "
 560                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 561                 GOTO(out, rc);
 562         }
 563
 564         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 565         LASSERT(body != NULL);
 566
 567         lmm_size = body->eadatasize;
 568
 569         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 570             lmm_size == 0) {
 571                 GOTO(out, rc = -ENODATA);
 572         }
 573
 574         lmm = req_capsule_server_sized_get(&req->rq_pill,
 575                                            &RMF_MDT_MD, lmm_size);
 576         LASSERT(lmm != NULL);
 577
 578         /*
 579          * This is coming from the MDS, so is probably in
 580          * little endian.  We convert it to host endian before
 581          * passing it to userspace.
 582          */
 583         /* We don't swab objects for directories */
 584         switch (le32_to_cpu(lmm->lmm_magic)) {
 585         case LOV_MAGIC_V1:
 586                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 587                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 588                 break;
 589         case LOV_MAGIC_V3:
 590                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 591                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 592                 break;
 593         case LMV_MAGIC_V1:
 594                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 595                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 596                 break;
 597         case LMV_USER_MAGIC:
 598                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 599                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 600                 break;
 601         default:
 602                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 603                 rc = -EPROTO;
 604         }
 605 out:
 606         *plmm = lmm;
 607         *plmm_size = lmm_size;
 608         *request = req;
 609         return rc;
 610 }
 611
 612 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 613 {
 614         struct md_op_data       *op_data;
 615         int                     rc;
 616         int                     mdt_index;
 617         ENTRY;
 618
 619         OBD_ALLOC_PTR(op_data);
 620         if (op_data == NULL)
 621                 RETURN(-ENOMEM);
 622
 623         op_data->op_flags |= MF_GET_MDT_IDX;
 624         op_data->op_fid1 = *fid;
 625         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 626         mdt_index = op_data->op_mds;
 627         OBD_FREE_PTR(op_data);
 628         if (rc < 0)
 629                 RETURN(rc);
 630
 631         RETURN(mdt_index);
 632 }
 633
 634 /*
 635  *  Get MDT index for the inode.
 636  */
 637 int ll_get_mdt_idx(struct inode *inode)
 638 {
 639         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 640 }
 641
 642 /**
 643  * Generic handler to do any pre-copy work.
 644  *
 645  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 646  * first information for it that real work has started.
 647  *
 648  * Moreover, for a ARCHIVE request, it will sample the file data version and
 649  * store it in \a copy.
 650  *
 651  * \return 0 on success.
 652  */
 653 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 654 {
 655         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 656         struct hsm_progress_kernel       hpk;
 657         int                              rc;
 658         ENTRY;
 659
 660         /* Forge a hsm_progress based on data from copy. */
 661         hpk.hpk_fid = copy->hc_hai.hai_fid;
 662         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 663         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 664         hpk.hpk_extent.length = 0;
 665         hpk.hpk_flags = 0;
 666         hpk.hpk_errval = 0;
 667         hpk.hpk_data_version = 0;
 668
 669
 670         /* For archive request, we need to read the current file version. */
 671         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 672                 struct inode    *inode;
 673                 __u64            data_version = 0;
 674
 675                 /* Get inode for this fid */
 676                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 677                 if (IS_ERR(inode)) {
 678                         hpk.hpk_flags |= HP_FLAG_RETRY;
 679                         /* hpk_errval is >= 0 */
 680                         hpk.hpk_errval = -PTR_ERR(inode);
 681                         GOTO(progress, rc = PTR_ERR(inode));
 682                 }
 683
 684                 /* Read current file data version */
 685                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 686                 iput(inode);
 687                 if (rc != 0) {
 688                         CDEBUG(D_HSM, "Could not read file data version of "
 689                                       DFID" (rc = %d). Archive request ("
 690                                       LPX64") could not be done.\n",
 691                                       PFID(&copy->hc_hai.hai_fid), rc,
 692                                       copy->hc_hai.hai_cookie);
 693                         hpk.hpk_flags |= HP_FLAG_RETRY;
 694                         /* hpk_errval must be >= 0 */
 695                         hpk.hpk_errval = -rc;
 696                         GOTO(progress, rc);
 697                 }
 698
 699                 /* Store it the hsm_copy for later copytool use.
 700                  * Always modified even if no lsm. */
 701                 copy->hc_data_version = data_version;
 702         }
 703
 704 progress:
 705         /* On error, the request should be considered as completed */
 706         if (hpk.hpk_errval > 0)
 707                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 708         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 709                            &hpk, NULL);
 710
 711         RETURN(rc);
 712 }
 713
 714 /**
 715  * Generic handler to do any post-copy work.
 716  *
 717  * It will send the last hsm_progress update to coordinator to inform it
 718  * that copy is finished and whether it was successful or not.
 719  *
 720  * Moreover,
 721  * - for ARCHIVE request, it will sample the file data version and compare it
 722  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 723  *   will be considered as failed.
 724  * - for RESTORE request, it will sample the file data version and send it to
 725  *   coordinator which is useful if the file was imported as 'released'.
 726  *
 727  * \return 0 on success.
 728  */
 729 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 730 {
 731         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 732         struct hsm_progress_kernel       hpk;
 733         int                              rc;
 734         ENTRY;
 735
 736         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 737         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 738          * initialized if copy_end was called with copy == NULL.
 739          */
 740
 741         /* Forge a hsm_progress based on data from copy. */
 742         hpk.hpk_fid = copy->hc_hai.hai_fid;
 743         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 744         hpk.hpk_extent = copy->hc_hai.hai_extent;
 745         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 746         hpk.hpk_errval = copy->hc_errval;
 747         hpk.hpk_data_version = 0;
 748
 749         /* For archive request, we need to check the file data was not changed.
 750          *
 751          * For restore request, we need to send the file data version, this is
 752          * useful when the file was created using hsm_import.
 753          */
 754         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 755              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 756             (copy->hc_errval == 0)) {
 757                 struct inode    *inode;
 758                 __u64            data_version = 0;
 759
 760                 /* Get lsm for this fid */
 761                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 762                 if (IS_ERR(inode)) {
 763                         hpk.hpk_flags |= HP_FLAG_RETRY;
 764                         /* hpk_errval must be >= 0 */
 765                         hpk.hpk_errval = -PTR_ERR(inode);
 766                         GOTO(progress, rc = PTR_ERR(inode));
 767                 }
 768
 769                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 770                 iput(inode);
 771                 if (rc) {
 772                         CDEBUG(D_HSM, "Could not read file data version. "
 773                                       "Request could not be confirmed.\n");
 774                         if (hpk.hpk_errval == 0)
 775                                 hpk.hpk_errval = -rc;
 776                         GOTO(progress, rc);
 777                 }
 778
 779                 /* Store it the hsm_copy for later copytool use.
 780                  * Always modified even if no lsm. */
 781                 hpk.hpk_data_version = data_version;
 782
 783                 /* File could have been stripped during archiving, so we need
 784                  * to check anyway. */
 785                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 786                     (copy->hc_data_version != data_version)) {
 787                         CDEBUG(D_HSM, "File data version mismatched. "
 788                               "File content was changed during archiving. "
 789                                DFID", start:"LPX64" current:"LPX64"\n",
 790                                PFID(&copy->hc_hai.hai_fid),
 791                                copy->hc_data_version, data_version);
 792                         /* File was changed, send error to cdt. Do not ask for
 793                          * retry because if a file is modified frequently,
 794                          * the cdt will loop on retried archive requests.
 795                          * The policy engine will ask for a new archive later
 796                          * when the file will not be modified for some tunable
 797                          * time */
 798                         /* we do not notify caller */
 799                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 800                         /* hpk_errval must be >= 0 */
 801                         hpk.hpk_errval = EBUSY;
 802                 }
 803
 804         }
 805
 806 progress:
 807         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 808                            &hpk, NULL);
 809
 810         RETURN(rc);
 811 }
 812
 813
 814 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 815                           const void __user *data, size_t size)
 816 {
 817         void *copy;
 818         int rc;
 819
 820         OBD_ALLOC(copy, size);
 821         if (copy == NULL)
 822                 return -ENOMEM;
 823
 824         if (copy_from_user(copy, data, size)) {
 825                 rc = -EFAULT;
 826                 goto out;
 827         }
 828
 829         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 830 out:
 831         OBD_FREE(copy, size);
 832
 833         return rc;
 834 }
 835
 836 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 837 {
 838         int cmd = qctl->qc_cmd;
 839         int type = qctl->qc_type;
 840         int id = qctl->qc_id;
 841         int valid = qctl->qc_valid;
 842         int rc = 0;
 843         ENTRY;
 844
 845         switch (cmd) {
 846         case LUSTRE_Q_INVALIDATE:
 847         case LUSTRE_Q_FINVALIDATE:
 848         case Q_QUOTAON:
 849         case Q_QUOTAOFF:
 850         case Q_SETQUOTA:
 851         case Q_SETINFO:
 852                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 853                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 854                         RETURN(-EPERM);
 855                 break;
 856         case Q_GETQUOTA:
 857                 if (((type == USRQUOTA &&
 858                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
 859                      (type == GRPQUOTA &&
 860                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
 861                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 862                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 863                         RETURN(-EPERM);
 864                 break;
 865         case Q_GETINFO:
 866                 break;
 867         default:
 868                 CERROR("unsupported quotactl op: %#x\n", cmd);
 869                 RETURN(-ENOTTY);
 870         }
 871
 872         if (valid != QC_GENERAL) {
 873                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 874                         RETURN(-EOPNOTSUPP);
 875
 876                 if (cmd == Q_GETINFO)
 877                         qctl->qc_cmd = Q_GETOINFO;
 878                 else if (cmd == Q_GETQUOTA)
 879                         qctl->qc_cmd = Q_GETOQUOTA;
 880                 else
 881                         RETURN(-EINVAL);
 882
 883                 switch (valid) {
 884                 case QC_MDTIDX:
 885                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 886                                            sizeof(*qctl), qctl, NULL);
 887                         break;
 888                 case QC_OSTIDX:
 889                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 890                                            sizeof(*qctl), qctl, NULL);
 891                         break;
 892                 case QC_UUID:
 893                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 894                                            sizeof(*qctl), qctl, NULL);
 895                         if (rc == -EAGAIN)
 896                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 897                                                    sbi->ll_dt_exp,
 898                                                    sizeof(*qctl), qctl, NULL);
 899                         break;
 900                 default:
 901                         rc = -EINVAL;
 902                         break;
 903                 }
 904
 905                 if (rc)
 906                         RETURN(rc);
 907
 908                 qctl->qc_cmd = cmd;
 909         } else {
 910                 struct obd_quotactl *oqctl;
 911
 912                 OBD_ALLOC_PTR(oqctl);
 913                 if (oqctl == NULL)
 914                         RETURN(-ENOMEM);
 915
 916                 QCTL_COPY(oqctl, qctl);
 917                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
 918                 if (rc) {
 919                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
 920                                 oqctl->qc_cmd = Q_QUOTAOFF;
 921                                 obd_quotactl(sbi->ll_md_exp, oqctl);
 922                         }
 923                         OBD_FREE_PTR(oqctl);
 924                         RETURN(rc);
 925                 }
 926                 /* If QIF_SPACE is not set, client should collect the
 927                  * space usage from OSSs by itself */
 928                 if (cmd == Q_GETQUOTA &&
 929                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
 930                     !oqctl->qc_dqblk.dqb_curspace) {
 931                         struct obd_quotactl *oqctl_tmp;
 932
 933                         OBD_ALLOC_PTR(oqctl_tmp);
 934                         if (oqctl_tmp == NULL)
 935                                 GOTO(out, rc = -ENOMEM);
 936
 937                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
 938                         oqctl_tmp->qc_id = oqctl->qc_id;
 939                         oqctl_tmp->qc_type = oqctl->qc_type;
 940
 941                         /* collect space usage from OSTs */
 942                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 943                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
 944                         if (!rc || rc == -EREMOTEIO) {
 945                                 oqctl->qc_dqblk.dqb_curspace =
 946                                         oqctl_tmp->qc_dqblk.dqb_curspace;
 947                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
 948                         }
 949
 950                         /* collect space & inode usage from MDTs */
 951                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 952                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
 953                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
 954                         if (!rc || rc == -EREMOTEIO) {
 955                                 oqctl->qc_dqblk.dqb_curspace +=
 956                                         oqctl_tmp->qc_dqblk.dqb_curspace;
 957                                 oqctl->qc_dqblk.dqb_curinodes =
 958                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
 959                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
 960                         } else {
 961                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
 962                         }
 963
 964                         OBD_FREE_PTR(oqctl_tmp);
 965                 }
 966 out:
 967                 QCTL_COPY(qctl, oqctl);
 968                 OBD_FREE_PTR(oqctl);
 969         }
 970
 971         RETURN(rc);
 972 }
 973
 974 static char *
 975 ll_getname(const char __user *filename)
 976 {
 977         int ret = 0, len;
 978         char *tmp = __getname();
 979
 980         if (!tmp)
 981                 return ERR_PTR(-ENOMEM);
 982
 983         len = strncpy_from_user(tmp, filename, PATH_MAX);
 984         if (len == 0)
 985                 ret = -ENOENT;
 986         else if (len > PATH_MAX)
 987                 ret = -ENAMETOOLONG;
 988
 989         if (ret) {
 990                 __putname(tmp);
 991                 tmp =  ERR_PTR(ret);
 992         }
 993         return tmp;
 994 }
 995
 996 #define ll_putname(filename) __putname(filename)
 997
 998 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 999 {
1000         struct inode *inode = file->f_dentry->d_inode;
1001         struct ll_sb_info *sbi = ll_i2sbi(inode);
1002         struct obd_ioctl_data *data;
1003         int rc = 0;
1004         ENTRY;
1005
1006         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1007                PFID(ll_inode2fid(inode)), inode, cmd);
1008
1009         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1010         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1011                 return -ENOTTY;
1012
1013         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1014         switch(cmd) {
1015         case FSFILT_IOC_GETFLAGS:
1016         case FSFILT_IOC_SETFLAGS:
1017                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1018         case FSFILT_IOC_GETVERSION_OLD:
1019         case FSFILT_IOC_GETVERSION:
1020                 RETURN(put_user(inode->i_generation, (int *)arg));
1021         /* We need to special case any other ioctls we want to handle,
1022          * to send them to the MDS/OST as appropriate and to properly
1023          * network encode the arg field.
1024         case FSFILT_IOC_SETVERSION_OLD:
1025         case FSFILT_IOC_SETVERSION:
1026         */
1027         case LL_IOC_GET_MDTIDX: {
1028                 int mdtidx;
1029
1030                 mdtidx = ll_get_mdt_idx(inode);
1031                 if (mdtidx < 0)
1032                         RETURN(mdtidx);
1033
1034                 if (put_user((int)mdtidx, (int*)arg))
1035                         RETURN(-EFAULT);
1036
1037                 return 0;
1038         }
1039         case IOC_MDC_LOOKUP: {
1040                 int namelen, len = 0;
1041                 char *buf = NULL;
1042                 char *filename;
1043
1044                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1045                 if (rc != 0)
1046                         RETURN(rc);
1047                 data = (void *)buf;
1048
1049                 filename = data->ioc_inlbuf1;
1050                 namelen = strlen(filename);
1051                 if (namelen < 1) {
1052                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1053                         GOTO(out_free, rc = -EINVAL);
1054                 }
1055
1056                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL);
1057                 if (rc < 0) {
1058                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1059                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1060                                filename, rc);
1061                         GOTO(out_free, rc);
1062                 }
1063 out_free:
1064                 obd_ioctl_freedata(buf, len);
1065                 return rc;
1066         }
1067         case LL_IOC_LMV_SETSTRIPE: {
1068                 struct lmv_user_md  *lum;
1069                 char            *buf = NULL;
1070                 char            *filename;
1071                 int              namelen = 0;
1072                 int              lumlen = 0;
1073                 int              len;
1074                 int              rc;
1075
1076                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1077                 if (rc)
1078                         RETURN(rc);
1079
1080                 data = (void *)buf;
1081                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1082                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1083                         GOTO(lmv_out_free, rc = -EINVAL);
1084
1085                 filename = data->ioc_inlbuf1;
1086                 namelen = data->ioc_inllen1;
1087
1088                 if (namelen < 1) {
1089                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1090                         GOTO(lmv_out_free, rc = -EINVAL);
1091                 }
1092                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1093                 lumlen = data->ioc_inllen2;
1094
1095                 if (lum->lum_magic != LMV_USER_MAGIC ||
1096                     lumlen != sizeof(*lum)) {
1097                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1098                                filename, lum->lum_magic, lumlen, -EFAULT);
1099                         GOTO(lmv_out_free, rc = -EINVAL);
1100                 }
1101
1102                 /**
1103                  * ll_dir_setdirstripe will be used to set dir stripe
1104                  *  mdc_create--->mdt_reint_create (with dirstripe)
1105                  */
1106                 rc = ll_dir_setdirstripe(inode, lum, filename);
1107 lmv_out_free:
1108                 obd_ioctl_freedata(buf, len);
1109                 RETURN(rc);
1110
1111         }
1112         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1113                 struct lmv_user_md        lum;
1114                 struct lmv_user_md __user *ulump =
1115                                         (struct lmv_user_md __user *)arg;
1116                 int                       rc;
1117
1118                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1119                         RETURN(-EFAULT);
1120
1121                 if (lum.lum_magic != LMV_USER_MAGIC)
1122                         RETURN(-EINVAL);
1123
1124                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1125
1126                 RETURN(rc);
1127         }
1128         case LL_IOC_LOV_SETSTRIPE: {
1129                 struct lov_user_md_v3 lumv3;
1130                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1131                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1132                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1133
1134                 int set_default = 0;
1135
1136                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1137                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1138                         sizeof(lumv3p->lmm_objects[0]));
1139                 /* first try with v1 which is smaller than v3 */
1140                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1141                         RETURN(-EFAULT);
1142
1143                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1144                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1145                                 RETURN(-EFAULT);
1146                 }
1147
1148                 if (inode->i_sb->s_root == file->f_dentry)
1149                         set_default = 1;
1150
1151                 /* in v1 and v3 cases lumv1 points to data */
1152                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1153
1154                 RETURN(rc);
1155         }
1156         case LL_IOC_LMV_GETSTRIPE: {
1157                 struct lmv_user_md __user *ulmv =
1158                                         (struct lmv_user_md __user *)arg;
1159                 struct lmv_user_md      lum;
1160                 struct ptlrpc_request   *request = NULL;
1161                 union lmv_mds_md        *lmm = NULL;
1162                 int                     lmmsize;
1163                 obd_valid               valid = 0;
1164                 struct lmv_user_md      *tmp = NULL;
1165                 int                     mdt_index;
1166                 int                     lum_size;
1167                 int                     stripe_count;
1168                 int                     i;
1169                 int                     rc;
1170
1171                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1172                         RETURN(-EFAULT);
1173
1174                 /* lum_magic will indicate which stripe the ioctl will like
1175                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1176                  * is for default LMV stripe */
1177                 if (lum.lum_magic == LMV_MAGIC_V1)
1178                         valid |= OBD_MD_MEA;
1179                 else if (lum.lum_magic == LMV_USER_MAGIC)
1180                         valid |= OBD_MD_DEFAULT_MEA;
1181                 else
1182                         RETURN(-EINVAL);
1183
1184                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1185                                       valid);
1186                 if (rc != 0)
1187                         GOTO(finish_req, rc);
1188
1189                 /* Get default LMV EA */
1190                 if (lum.lum_magic == LMV_USER_MAGIC) {
1191                         if (rc != 0)
1192                                 GOTO(finish_req, rc);
1193
1194                         if (lmmsize > sizeof(*ulmv))
1195                                 GOTO(finish_req, rc = -EINVAL);
1196
1197                         if (copy_to_user(ulmv, lmm, lmmsize))
1198                                 GOTO(finish_req, rc = -EFAULT);
1199
1200                         GOTO(finish_req, rc);
1201                 }
1202
1203                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1204                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1205                 OBD_ALLOC(tmp, lum_size);
1206                 if (tmp == NULL)
1207                         GOTO(finish_req, rc = -ENOMEM);
1208
1209                 mdt_index = ll_get_mdt_idx(inode);
1210                 if (mdt_index < 0)
1211                         GOTO(out_tmp, rc = -ENOMEM);
1212
1213                 tmp->lum_magic = LMV_MAGIC_V1;
1214                 tmp->lum_stripe_count = 0;
1215                 tmp->lum_stripe_offset = mdt_index;
1216                 for (i = 0; i < stripe_count; i++) {
1217                         struct lu_fid   *fid;
1218
1219                         fid = &lmm->lmv_md_v1.lmv_stripe_fids[i];
1220                         mdt_index = ll_get_mdt_idx_by_fid(sbi, fid);
1221                         if (mdt_index < 0)
1222                                 GOTO(out_tmp, rc = mdt_index);
1223
1224                         tmp->lum_objects[i].lum_mds = mdt_index;
1225                         tmp->lum_objects[i].lum_fid = *fid;
1226                         tmp->lum_stripe_count++;
1227                 }
1228
1229                 if (copy_to_user(ulmv, tmp, lum_size))
1230                         GOTO(out_tmp, rc = -EFAULT);
1231 out_tmp:
1232                 OBD_FREE(tmp, lum_size);
1233 finish_req:
1234                 ptlrpc_req_finished(request);
1235                 return rc;
1236         }
1237
1238         case LL_IOC_REMOVE_ENTRY: {
1239                 char            *filename = NULL;
1240                 int              namelen = 0;
1241                 int              rc;
1242
1243                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1244                  * unsupported server, which might crash the server(LU-2730),
1245                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1246                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1247                  * server will support REINT_RMENTRY XXX*/
1248                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1249                         RETURN(-ENOTSUPP);
1250
1251                 filename = ll_getname((const char *)arg);
1252                 if (IS_ERR(filename))
1253                         RETURN(PTR_ERR(filename));
1254
1255                 namelen = strlen(filename);
1256                 if (namelen < 1)
1257                         GOTO(out_rmdir, rc = -EINVAL);
1258
1259                 rc = ll_rmdir_entry(inode, filename, namelen);
1260 out_rmdir:
1261                 if (filename)
1262                         ll_putname(filename);
1263                 RETURN(rc);
1264         }
1265         case LL_IOC_LOV_SWAP_LAYOUTS:
1266                 RETURN(-EPERM);
1267         case IOC_OBD_STATFS:
1268                 RETURN(ll_obd_statfs(inode, (void *)arg));
1269         case LL_IOC_LOV_GETSTRIPE:
1270         case LL_IOC_MDC_GETINFO:
1271         case IOC_MDC_GETFILEINFO:
1272         case IOC_MDC_GETFILESTRIPE: {
1273                 struct ptlrpc_request *request = NULL;
1274                 struct lov_user_md *lump;
1275                 struct lov_mds_md *lmm = NULL;
1276                 struct mdt_body *body;
1277                 char *filename = NULL;
1278                 int lmmsize;
1279
1280                 if (cmd == IOC_MDC_GETFILEINFO ||
1281                     cmd == IOC_MDC_GETFILESTRIPE) {
1282                         filename = ll_getname((const char *)arg);
1283                         if (IS_ERR(filename))
1284                                 RETURN(PTR_ERR(filename));
1285
1286                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1287                                                       &lmmsize, &request);
1288                 } else {
1289                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1290                                               &request, 0);
1291                 }
1292
1293                 if (request) {
1294                         body = req_capsule_server_get(&request->rq_pill,
1295                                                       &RMF_MDT_BODY);
1296                         LASSERT(body != NULL);
1297                 } else {
1298                         GOTO(out_req, rc);
1299                 }
1300
1301                 if (rc < 0) {
1302                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1303                                                cmd == LL_IOC_MDC_GETINFO))
1304                                 GOTO(skip_lmm, rc = 0);
1305                         else
1306                                 GOTO(out_req, rc);
1307                 }
1308
1309                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1310                     cmd == LL_IOC_LOV_GETSTRIPE) {
1311                         lump = (struct lov_user_md *)arg;
1312                 } else {
1313                         struct lov_user_mds_data *lmdp;
1314                         lmdp = (struct lov_user_mds_data *)arg;
1315                         lump = &lmdp->lmd_lmm;
1316                 }
1317                 if (copy_to_user(lump, lmm, lmmsize)) {
1318                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1319                                 GOTO(out_req, rc = -EFAULT);
1320                         rc = -EOVERFLOW;
1321                 }
1322         skip_lmm:
1323                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1324                         struct lov_user_mds_data *lmdp;
1325                         lstat_t st = { 0 };
1326
1327                         st.st_dev     = inode->i_sb->s_dev;
1328                         st.st_mode    = body->mode;
1329                         st.st_nlink   = body->nlink;
1330                         st.st_uid     = body->uid;
1331                         st.st_gid     = body->gid;
1332                         st.st_rdev    = body->rdev;
1333                         st.st_size    = body->size;
1334                         st.st_blksize = PAGE_CACHE_SIZE;
1335                         st.st_blocks  = body->blocks;
1336                         st.st_atime   = body->atime;
1337                         st.st_mtime   = body->mtime;
1338                         st.st_ctime   = body->ctime;
1339                         st.st_ino     = inode->i_ino;
1340
1341                         lmdp = (struct lov_user_mds_data *)arg;
1342                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1343                                 GOTO(out_req, rc = -EFAULT);
1344                 }
1345
1346                 EXIT;
1347         out_req:
1348                 ptlrpc_req_finished(request);
1349                 if (filename)
1350                         ll_putname(filename);
1351                 return rc;
1352         }
1353         case IOC_LOV_GETINFO: {
1354                 struct lov_user_mds_data *lumd;
1355                 struct lov_stripe_md *lsm;
1356                 struct lov_user_md *lum;
1357                 struct lov_mds_md *lmm;
1358                 int lmmsize;
1359                 lstat_t st;
1360
1361                 lumd = (struct lov_user_mds_data *)arg;
1362                 lum = &lumd->lmd_lmm;
1363
1364                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1365                 if (rc)
1366                         RETURN(rc);
1367
1368                 OBD_ALLOC_LARGE(lmm, lmmsize);
1369                 if (lmm == NULL)
1370                         RETURN(-ENOMEM);
1371
1372                 if (copy_from_user(lmm, lum, lmmsize))
1373                         GOTO(free_lmm, rc = -EFAULT);
1374
1375                 switch (lmm->lmm_magic) {
1376                 case LOV_USER_MAGIC_V1:
1377                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1378                                 break;
1379                         /* swab objects first so that stripes num will be sane */
1380                         lustre_swab_lov_user_md_objects(
1381                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1382                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1383                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1384                         break;
1385                 case LOV_USER_MAGIC_V3:
1386                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1387                                 break;
1388                         /* swab objects first so that stripes num will be sane */
1389                         lustre_swab_lov_user_md_objects(
1390                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1391                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1392                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1393                         break;
1394                 default:
1395                         GOTO(free_lmm, rc = -EINVAL);
1396                 }
1397
1398                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1399                 if (rc < 0)
1400                         GOTO(free_lmm, rc = -ENOMEM);
1401
1402                 /* Perform glimpse_size operation. */
1403                 memset(&st, 0, sizeof(st));
1404
1405                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1406                 if (rc)
1407                         GOTO(free_lsm, rc);
1408
1409                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1410                         GOTO(free_lsm, rc = -EFAULT);
1411
1412                 EXIT;
1413         free_lsm:
1414                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1415         free_lmm:
1416                 OBD_FREE_LARGE(lmm, lmmsize);
1417                 return rc;
1418         }
1419         case OBD_IOC_QUOTACHECK: {
1420                 struct obd_quotactl *oqctl;
1421                 int error = 0;
1422
1423                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1424                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1425                         RETURN(-EPERM);
1426
1427                 OBD_ALLOC_PTR(oqctl);
1428                 if (!oqctl)
1429                         RETURN(-ENOMEM);
1430                 oqctl->qc_type = arg;
1431                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1432                 if (rc < 0) {
1433                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1434                         error = rc;
1435                 }
1436
1437                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1438                 if (rc < 0)
1439                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1440
1441                 OBD_FREE_PTR(oqctl);
1442                 return error ?: rc;
1443         }
1444         case OBD_IOC_POLL_QUOTACHECK: {
1445                 struct if_quotacheck *check;
1446
1447                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1448                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1449                         RETURN(-EPERM);
1450
1451                 OBD_ALLOC_PTR(check);
1452                 if (!check)
1453                         RETURN(-ENOMEM);
1454
1455                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1456                                    NULL);
1457                 if (rc) {
1458                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1459                         if (copy_to_user((void *)arg, check,
1460                                              sizeof(*check)))
1461                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1462                         GOTO(out_poll, rc);
1463                 }
1464
1465                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1466                                    NULL);
1467                 if (rc) {
1468                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1469                         if (copy_to_user((void *)arg, check,
1470                                              sizeof(*check)))
1471                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1472                         GOTO(out_poll, rc);
1473                 }
1474         out_poll:
1475                 OBD_FREE_PTR(check);
1476                 RETURN(rc);
1477         }
1478 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1479         case LL_IOC_QUOTACTL_18: {
1480                 /* copy the old 1.x quota struct for internal use, then copy
1481                  * back into old format struct.  For 1.8 compatibility. */
1482                 struct if_quotactl_18 *qctl_18;
1483                 struct if_quotactl *qctl_20;
1484
1485                 OBD_ALLOC_PTR(qctl_18);
1486                 if (!qctl_18)
1487                         RETURN(-ENOMEM);
1488
1489                 OBD_ALLOC_PTR(qctl_20);
1490                 if (!qctl_20)
1491                         GOTO(out_quotactl_18, rc = -ENOMEM);
1492
1493                 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1494                         GOTO(out_quotactl_20, rc = -ENOMEM);
1495
1496                 QCTL_COPY(qctl_20, qctl_18);
1497                 qctl_20->qc_idx = 0;
1498
1499                 /* XXX: dqb_valid was borrowed as a flag to mark that
1500                  *      only mds quota is wanted */
1501                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1502                     qctl_18->qc_dqblk.dqb_valid) {
1503                         qctl_20->qc_valid = QC_MDTIDX;
1504                         qctl_20->qc_dqblk.dqb_valid = 0;
1505                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1506                         qctl_20->qc_valid = QC_UUID;
1507                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1508                 } else {
1509                         qctl_20->qc_valid = QC_GENERAL;
1510                 }
1511
1512                 rc = quotactl_ioctl(sbi, qctl_20);
1513
1514                 if (rc == 0) {
1515                         QCTL_COPY(qctl_18, qctl_20);
1516                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1517
1518                         if (copy_to_user((void *)arg, qctl_18,
1519                                              sizeof(*qctl_18)))
1520                                 rc = -EFAULT;
1521                 }
1522
1523         out_quotactl_20:
1524                 OBD_FREE_PTR(qctl_20);
1525         out_quotactl_18:
1526                 OBD_FREE_PTR(qctl_18);
1527                 RETURN(rc);
1528         }
1529 #else
1530 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1531 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1532         case OBD_IOC_QUOTACTL: {
1533                 struct if_quotactl *qctl;
1534
1535                 OBD_ALLOC_PTR(qctl);
1536                 if (!qctl)
1537                         RETURN(-ENOMEM);
1538
1539                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1540                         GOTO(out_quotactl, rc = -EFAULT);
1541
1542                 rc = quotactl_ioctl(sbi, qctl);
1543
1544                 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1545                         rc = -EFAULT;
1546
1547         out_quotactl:
1548                 OBD_FREE_PTR(qctl);
1549                 RETURN(rc);
1550         }
1551         case OBD_IOC_GETDTNAME:
1552         case OBD_IOC_GETMDNAME:
1553                 RETURN(ll_get_obd_name(inode, cmd, arg));
1554         case LL_IOC_FLUSHCTX:
1555                 RETURN(ll_flush_ctx(inode));
1556 #ifdef CONFIG_FS_POSIX_ACL
1557         case LL_IOC_RMTACL: {
1558             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1559                 inode == inode->i_sb->s_root->d_inode) {
1560                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1561
1562                 LASSERT(fd != NULL);
1563                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1564                 if (!rc)
1565                         fd->fd_flags |= LL_FILE_RMTACL;
1566                 RETURN(rc);
1567             } else
1568                 RETURN(0);
1569         }
1570 #endif
1571         case LL_IOC_GETOBDCOUNT: {
1572                 int count, vallen;
1573                 struct obd_export *exp;
1574
1575                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1576                         RETURN(-EFAULT);
1577
1578                 /* get ost count when count is zero, get mdt count otherwise */
1579                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1580                 vallen = sizeof(count);
1581                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1582                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1583                 if (rc) {
1584                         CERROR("get target count failed: %d\n", rc);
1585                         RETURN(rc);
1586                 }
1587
1588                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1589                         RETURN(-EFAULT);
1590
1591                 RETURN(0);
1592         }
1593         case LL_IOC_PATH2FID:
1594                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1595                                      sizeof(struct lu_fid)))
1596                         RETURN(-EFAULT);
1597                 RETURN(0);
1598         case LL_IOC_GET_CONNECT_FLAGS: {
1599                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1600         }
1601         case OBD_IOC_CHANGELOG_SEND:
1602         case OBD_IOC_CHANGELOG_CLEAR:
1603                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1604                                     sizeof(struct ioc_changelog));
1605                 RETURN(rc);
1606         case OBD_IOC_FID2PATH:
1607                 RETURN(ll_fid2path(inode, (void *)arg));
1608         case LL_IOC_HSM_REQUEST: {
1609                 struct hsm_user_request *hur;
1610                 int                      totalsize;
1611
1612                 OBD_ALLOC_PTR(hur);
1613                 if (hur == NULL)
1614                         RETURN(-ENOMEM);
1615
1616                 /* We don't know the true size yet; copy the fixed-size part */
1617                 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1618                         OBD_FREE_PTR(hur);
1619                         RETURN(-EFAULT);
1620                 }
1621
1622                 /* Compute the whole struct size */
1623                 totalsize = hur_len(hur);
1624                 OBD_FREE_PTR(hur);
1625
1626                 /* Final size will be more than double totalsize */
1627                 if (totalsize >= MDS_MAXREQSIZE / 3)
1628                         RETURN(-E2BIG);
1629
1630                 OBD_ALLOC_LARGE(hur, totalsize);
1631                 if (hur == NULL)
1632                         RETURN(-ENOMEM);
1633
1634                 /* Copy the whole struct */
1635                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1636                         OBD_FREE_LARGE(hur, totalsize);
1637                         RETURN(-EFAULT);
1638                 }
1639
1640                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1641                         const struct lu_fid *fid;
1642                         struct inode *f;
1643                         int i;
1644
1645                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1646                                 fid = &hur->hur_user_item[i].hui_fid;
1647                                 f = search_inode_for_lustre(inode->i_sb, fid);
1648                                 if (IS_ERR(f)) {
1649                                         rc = PTR_ERR(f);
1650                                         break;
1651                                 }
1652
1653                                 rc = ll_hsm_release(f);
1654                                 iput(f);
1655                                 if (rc != 0)
1656                                         break;
1657                         }
1658                 } else {
1659                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1660                                            hur, NULL);
1661                 }
1662
1663                 OBD_FREE_LARGE(hur, totalsize);
1664
1665                 RETURN(rc);
1666         }
1667         case LL_IOC_HSM_PROGRESS: {
1668                 struct hsm_progress_kernel      hpk;
1669                 struct hsm_progress             hp;
1670
1671                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1672                         RETURN(-EFAULT);
1673
1674                 hpk.hpk_fid = hp.hp_fid;
1675                 hpk.hpk_cookie = hp.hp_cookie;
1676                 hpk.hpk_extent = hp.hp_extent;
1677                 hpk.hpk_flags = hp.hp_flags;
1678                 hpk.hpk_errval = hp.hp_errval;
1679                 hpk.hpk_data_version = 0;
1680
1681                 /* File may not exist in Lustre; all progress
1682                  * reported to Lustre root */
1683                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1684                                    NULL);
1685                 RETURN(rc);
1686         }
1687         case LL_IOC_HSM_CT_START:
1688                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1689                         RETURN(-EPERM);
1690
1691                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1692                                     sizeof(struct lustre_kernelcomm));
1693                 RETURN(rc);
1694
1695         case LL_IOC_HSM_COPY_START: {
1696                 struct hsm_copy *copy;
1697                 int              rc;
1698
1699                 OBD_ALLOC_PTR(copy);
1700                 if (copy == NULL)
1701                         RETURN(-ENOMEM);
1702                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1703                         OBD_FREE_PTR(copy);
1704                         RETURN(-EFAULT);
1705                 }
1706
1707                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1708                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1709                         rc = -EFAULT;
1710
1711                 OBD_FREE_PTR(copy);
1712                 RETURN(rc);
1713         }
1714         case LL_IOC_HSM_COPY_END: {
1715                 struct hsm_copy *copy;
1716                 int              rc;
1717
1718                 OBD_ALLOC_PTR(copy);
1719                 if (copy == NULL)
1720                         RETURN(-ENOMEM);
1721                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1722                         OBD_FREE_PTR(copy);
1723                         RETURN(-EFAULT);
1724                 }
1725
1726                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1727                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1728                         rc = -EFAULT;
1729
1730                 OBD_FREE_PTR(copy);
1731                 RETURN(rc);
1732         }
1733         case LL_IOC_MIGRATE: {
1734                 char            *buf = NULL;
1735                 const char      *filename;
1736                 int             namelen = 0;
1737                 int             len;
1738                 int             rc;
1739                 int             mdtidx;
1740
1741                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1742                 if (rc < 0)
1743                         RETURN(rc);
1744
1745                 data = (struct obd_ioctl_data *)buf;
1746                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1747                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1748                         GOTO(migrate_free, rc = -EINVAL);
1749
1750                 filename = data->ioc_inlbuf1;
1751                 namelen = data->ioc_inllen1;
1752                 /* \0 is packed at the end of filename */
1753                 if (namelen < 1 || namelen != strlen(filename) + 1)
1754                         GOTO(migrate_free, rc = -EINVAL);
1755
1756                 if (data->ioc_inllen2 != sizeof(mdtidx))
1757                         GOTO(migrate_free, rc = -EINVAL);
1758                 mdtidx = *(int *)data->ioc_inlbuf2;
1759
1760                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1761 migrate_free:
1762                 obd_ioctl_freedata(buf, len);
1763
1764                 RETURN(rc);
1765         }
1766         default:
1767                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1768                                      (void *)arg));
1769         }
1770 }
1771
1772 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1773 {
1774         struct inode *inode = file->f_mapping->host;
1775         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1776         struct ll_sb_info *sbi = ll_i2sbi(inode);
1777         int api32 = ll_need_32bit_api(sbi);
1778         loff_t ret = -EINVAL;
1779         ENTRY;
1780
1781         mutex_lock(&inode->i_mutex);
1782         switch (origin) {
1783                 case SEEK_SET:
1784                         break;
1785                 case SEEK_CUR:
1786                         offset += file->f_pos;
1787                         break;
1788                 case SEEK_END:
1789                         if (offset > 0)
1790                                 GOTO(out, ret);
1791                         if (api32)
1792                                 offset += LL_DIR_END_OFF_32BIT;
1793                         else
1794                                 offset += LL_DIR_END_OFF;
1795                         break;
1796                 default:
1797                         GOTO(out, ret);
1798         }
1799
1800         if (offset >= 0 &&
1801             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1802              (!api32 && offset <= LL_DIR_END_OFF))) {
1803                 if (offset != file->f_pos) {
1804                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1805                             (!api32 && offset == LL_DIR_END_OFF))
1806                                 fd->lfd_pos = MDS_DIR_END_OFF;
1807                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1808                                 fd->lfd_pos = offset << 32;
1809                         else
1810                                 fd->lfd_pos = offset;
1811                         file->f_pos = offset;
1812                         file->f_version = 0;
1813                 }
1814                 ret = offset;
1815         }
1816         GOTO(out, ret);
1817
1818 out:
1819         mutex_unlock(&inode->i_mutex);
1820         return ret;
1821 }
1822
1823 static int ll_dir_open(struct inode *inode, struct file *file)
1824 {
1825         ENTRY;
1826         RETURN(ll_file_open(inode, file));
1827 }
1828
1829 static int ll_dir_release(struct inode *inode, struct file *file)
1830 {
1831         ENTRY;
1832         RETURN(ll_file_release(inode, file));
1833 }
1834
1835 const struct file_operations ll_dir_operations = {
1836         .llseek         = ll_dir_seek,
1837         .open           = ll_dir_open,
1838         .release        = ll_dir_release,
1839         .read           = generic_read_dir,
1840 #ifdef HAVE_DIR_CONTEXT
1841         .iterate        = ll_iterate,
1842 #else
1843         .readdir        = ll_readdir,
1844 #endif
1845         .unlocked_ioctl = ll_dir_ioctl,
1846         .fsync          = ll_fsync,
1847 };