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 <linux/user_namespace.h>
  46 #ifdef HAVE_UIDGID_HEADER
  47 # include <linux/uidgid.h>
  48 #endif
  49 #include <asm/uaccess.h>
  50 #include <linux/buffer_head.h>   // for wait_on_buffer
  51 #include <linux/pagevec.h>
  52
  53 #define DEBUG_SUBSYSTEM S_LLITE
  54
  55 #include <lustre/lustre_idl.h>
  56 #include <obd_support.h>
  57 #include <obd_class.h>
  58 #include <lustre_ioctl.h>
  59 #include <lustre_lib.h>
  60 #include <lustre_dlm.h>
  61 #include <lustre_fid.h>
  62 #include <lustre_kernelcomm.h>
  63 #include "llite_internal.h"
  64
  65 /*
  66  * (new) readdir implementation overview.
  67  *
  68  * Original lustre readdir implementation cached exact copy of raw directory
  69  * pages on the client. These pages were indexed in client page cache by
  70  * logical offset in the directory file. This design, while very simple and
  71  * intuitive had some inherent problems:
  72  *
  73  *     . it implies that byte offset to the directory entry serves as a
  74  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  75  *     ext3/htree directory entries may move due to splits, and more
  76  *     importantly,
  77  *
  78  *     . it is incompatible with the design of split directories for cmd3,
  79  *     that assumes that names are distributed across nodes based on their
  80  *     hash, and so readdir should be done in hash order.
  81  *
  82  * New readdir implementation does readdir in hash order, and uses hash of a
  83  * file name as a telldir/seekdir cookie. This led to number of complications:
  84  *
  85  *     . hash is not unique, so it cannot be used to index cached directory
  86  *     pages on the client (note, that it requires a whole pageful of hash
  87  *     collided entries to cause two pages to have identical hashes);
  88  *
  89  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  90  *     entry cookie. ext3/htree has the same problem and lustre implementation
  91  *     mimics their solution: seekdir(hash) positions directory at the first
  92  *     entry with the given hash.
  93  *
  94  * Client side.
  95  *
  96  * 0. caching
  97  *
  98  * Client caches directory pages using hash of the first entry as an index. As
  99  * noted above hash is not unique, so this solution doesn't work as is:
 100  * special processing is needed for "page hash chains" (i.e., sequences of
 101  * pages filled with entries all having the same hash value).
 102  *
 103  * First, such chains have to be detected. To this end, server returns to the
 104  * client the hash of the first entry on the page next to one returned. When
 105  * client detects that this hash is the same as hash of the first entry on the
 106  * returned page, page hash collision has to be handled. Pages in the
 107  * hash chain, except first one, are termed "overflow pages".
 108  *
 109  * Solution to index uniqueness problem is to not cache overflow
 110  * pages. Instead, when page hash collision is detected, all overflow pages
 111  * from emerging chain are immediately requested from the server and placed in
 112  * a special data structure (struct ll_dir_chain). This data structure is used
 113  * by ll_readdir() to process entries from overflow pages. When readdir
 114  * invocation finishes, overflow pages are discarded. If page hash collision
 115  * chain weren't completely processed, next call to readdir will again detect
 116  * page hash collision, again read overflow pages in, process next portion of
 117  * entries and again discard the pages. This is not as wasteful as it looks,
 118  * because, given reasonable hash, page hash collisions are extremely rare.
 119  *
 120  * 1. directory positioning
 121  *
 122  * When seekdir(hash) is called, original
 123  *
 124  *
 125  *
 126  *
 127  *
 128  *
 129  *
 130  *
 131  * Server.
 132  *
 133  * identification of and access to overflow pages
 134  *
 135  * page format
 136  *
 137  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 138  * a header lu_dirpage which describes the start/end hash, and whether this
 139  * page is empty (contains no dir entry) or hash collide with next page.
 140  * After client receives reply, several pages will be integrated into dir page
 141  * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 142  * lu_dirpage for this integrated page will be adjusted. See
 143  * mdc_adjust_dirpages().
 144  *
 145  */
 146 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
 147                              __u64 offset, struct ll_dir_chain *chain)
 148 {
 149         struct md_callback      cb_op;
 150         struct page             *page;
 151         int                     rc;
 152
 153         cb_op.md_blocking_ast = ll_md_blocking_ast;
 154         rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
 155         if (rc != 0)
 156                 return ERR_PTR(rc);
 157
 158         return page;
 159 }
 160
 161 void ll_release_page(struct inode *inode, struct page *page,
 162                      bool remove)
 163 {
 164         kunmap(page);
 165
 166         /* Always remove the page for striped dir, because the page is
 167          * built from temporarily in LMV layer */
 168         if (inode != NULL && S_ISDIR(inode->i_mode) &&
 169             ll_i2info(inode)->lli_lsm_md != NULL) {
 170                 __free_page(page);
 171                 return;
 172         }
 173
 174         if (remove) {
 175                 lock_page(page);
 176                 if (likely(page->mapping != NULL))
 177                         truncate_complete_page(page->mapping, page);
 178                 unlock_page(page);
 179         }
 180         page_cache_release(page);
 181 }
 182
 183 /**
 184  * return IF_* type for given lu_dirent entry.
 185  * IF_* flag shld be converted to particular OS file type in
 186  * platform llite module.
 187  */
 188 static u16 ll_dirent_type_get(struct lu_dirent *ent)
 189 {
 190         u16 type = 0;
 191         struct luda_type *lt;
 192         int len = 0;
 193
 194         if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
 195                 const unsigned align = sizeof(struct luda_type) - 1;
 196
 197                 len = le16_to_cpu(ent->lde_namelen);
 198                 len = (len + align) & ~align;
 199                 lt = (void *)ent->lde_name + len;
 200                 type = IFTODT(le16_to_cpu(lt->lt_type));
 201         }
 202
 203         return type;
 204 }
 205
 206 #ifdef HAVE_DIR_CONTEXT
 207 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 208                 struct dir_context *ctx)
 209 {
 210 #else
 211 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 212                 void *cookie, filldir_t filldir)
 213 {
 214 #endif
 215         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 216         __u64                 pos        = *ppos;
 217         bool                  is_api32 = ll_need_32bit_api(sbi);
 218         bool                  is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 219         struct page          *page;
 220         struct ll_dir_chain   chain;
 221         bool                  done = false;
 222         int                   rc = 0;
 223         ENTRY;
 224
 225         ll_dir_chain_init(&chain);
 226
 227         page = ll_get_dir_page(inode, op_data, pos, &chain);
 228
 229         while (rc == 0 && !done) {
 230                 struct lu_dirpage *dp;
 231                 struct lu_dirent  *ent;
 232                 __u64 hash;
 233                 __u64 next;
 234
 235                 if (IS_ERR(page)) {
 236                         rc = PTR_ERR(page);
 237                         break;
 238                 }
 239
 240                 hash = MDS_DIR_END_OFF;
 241                 dp = page_address(page);
 242                 for (ent = lu_dirent_start(dp); ent != NULL && !done;
 243                      ent = lu_dirent_next(ent)) {
 244                         __u16          type;
 245                         int            namelen;
 246                         struct lu_fid  fid;
 247                         __u64          lhash;
 248                         __u64          ino;
 249
 250                         hash = le64_to_cpu(ent->lde_hash);
 251                         if (hash < pos)
 252                                 /*
 253                                  * Skip until we find target hash
 254                                  * value.
 255                                  */
 256                                 continue;
 257
 258                         namelen = le16_to_cpu(ent->lde_namelen);
 259                         if (namelen == 0)
 260                                 /*
 261                                  * Skip dummy record.
 262                                  */
 263                                 continue;
 264
 265                         if (is_api32 && is_hash64)
 266                                 lhash = hash >> 32;
 267                         else
 268                                 lhash = hash;
 269                         fid_le_to_cpu(&fid, &ent->lde_fid);
 270                         ino = cl_fid_build_ino(&fid, is_api32);
 271                         type = ll_dirent_type_get(ent);
 272                         /* For 'll_nfs_get_name_filldir()', it will try
 273                          * to access the 'ent' through its 'lde_name',
 274                          * so the parameter 'name' for 'filldir()' must
 275                          * be part of the 'ent'. */
 276 #ifdef HAVE_DIR_CONTEXT
 277                         ctx->pos = lhash;
 278                         done = !dir_emit(ctx, ent->lde_name, namelen, ino,
 279                                          type);
 280 #else
 281                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 282                                        ino, type);
 283 #endif
 284                 }
 285
 286                 if (done) {
 287                         pos = hash;
 288                         ll_release_page(inode, page, false);
 289                         break;
 290                 }
 291
 292                 next = le64_to_cpu(dp->ldp_hash_end);
 293                 pos = next;
 294                 if (pos == MDS_DIR_END_OFF) {
 295                         /*
 296                          * End of directory reached.
 297                          */
 298                         done = 1;
 299                         ll_release_page(inode, page, false);
 300                 } else {
 301                         /*
 302                          * Normal case: continue to the next
 303                          * page.
 304                          */
 305                         ll_release_page(inode, page,
 306                                         le32_to_cpu(dp->ldp_flags) &
 307                                         LDF_COLLIDE);
 308                         next = pos;
 309                         page = ll_get_dir_page(inode, op_data, pos,
 310                                                &chain);
 311                 }
 312         }
 313 #ifdef HAVE_DIR_CONTEXT
 314         ctx->pos = pos;
 315 #else
 316         *ppos = pos;
 317 #endif
 318         ll_dir_chain_fini(&chain);
 319         RETURN(rc);
 320 }
 321
 322 #ifdef HAVE_DIR_CONTEXT
 323 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 324 #else
 325 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 326 #endif
 327 {
 328         struct inode            *inode  = filp->f_path.dentry->d_inode;
 329         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 330         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 331         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 332         int                     api32   = ll_need_32bit_api(sbi);
 333         struct md_op_data       *op_data;
 334         __u64                   pos;
 335         int                     rc;
 336         ENTRY;
 337
 338         if (lfd != NULL)
 339                 pos = lfd->lfd_pos;
 340         else
 341                 pos = 0;
 342
 343         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 344                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 345                inode, (unsigned long)pos, i_size_read(inode), api32);
 346
 347         if (pos == MDS_DIR_END_OFF)
 348                 /*
 349                  * end-of-file.
 350                  */
 351                 GOTO(out, rc = 0);
 352
 353         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 354                                      LUSTRE_OPC_ANY, inode);
 355         if (IS_ERR(op_data))
 356                 GOTO(out, rc = PTR_ERR(op_data));
 357
 358         if (unlikely(op_data->op_mea1 != NULL)) {
 359                 /* This is only needed for striped dir to fill ..,
 360                  * see lmv_read_entry */
 361                 if (filp->f_path.dentry->d_parent != NULL &&
 362                     filp->f_path.dentry->d_parent->d_inode != NULL) {
 363                         __u64 ibits = MDS_INODELOCK_UPDATE;
 364                         struct inode *parent =
 365                                 filp->f_path.dentry->d_parent->d_inode;
 366
 367                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 368                                 op_data->op_fid3 = *ll_inode2fid(parent);
 369                 }
 370
 371                 /* If it can not find in cache, do lookup .. on the master
 372                  * object */
 373                 if (fid_is_zero(&op_data->op_fid3)) {
 374                         rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
 375                         if (rc != 0) {
 376                                 ll_finish_md_op_data(op_data);
 377                                 RETURN(rc);
 378                         }
 379                 }
 380         }
 381         op_data->op_max_pages = sbi->ll_md_brw_pages;
 382 #ifdef HAVE_DIR_CONTEXT
 383         ctx->pos = pos;
 384         rc = ll_dir_read(inode, &pos, op_data, ctx);
 385         pos = ctx->pos;
 386 #else
 387         rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
 388 #endif
 389         if (lfd != NULL)
 390                 lfd->lfd_pos = pos;
 391
 392         if (pos == MDS_DIR_END_OFF) {
 393                 if (api32)
 394                         pos = LL_DIR_END_OFF_32BIT;
 395                 else
 396                         pos = LL_DIR_END_OFF;
 397         } else {
 398                 if (api32 && hash64)
 399                         pos = pos >> 32;
 400         }
 401 #ifdef HAVE_DIR_CONTEXT
 402         ctx->pos = pos;
 403 #else
 404         filp->f_pos = pos;
 405 #endif
 406         ll_finish_md_op_data(op_data);
 407         filp->f_version = inode->i_version;
 408
 409 out:
 410         if (!rc)
 411                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 412
 413         RETURN(rc);
 414 }
 415
 416 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 417 {
 418         struct mgs_send_param *msp;
 419         int rc = 0;
 420
 421         OBD_ALLOC_PTR(msp);
 422         if (!msp)
 423                 return -ENOMEM;
 424
 425         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 426         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 427                                 sizeof(struct mgs_send_param), msp, NULL);
 428         if (rc)
 429                 CERROR("Failed to set parameter: %d\n", rc);
 430         OBD_FREE_PTR(msp);
 431
 432         return rc;
 433 }
 434
 435 /**
 436  * Create striped directory with specified stripe(@lump)
 437  *
 438  * param[in]parent      the parent of the directory.
 439  * param[in]lump        the specified stripes.
 440  * param[in]dirname     the name of the directory.
 441  * param[in]mode        the specified mode of the directory.
 442  *
 443  * retval               =0 if striped directory is being created successfully.
 444  *                      <0 if the creation is failed.
 445  */
 446 static int ll_dir_setdirstripe(struct inode *parent, struct lmv_user_md *lump,
 447                                const char *dirname, umode_t mode)
 448 {
 449         struct ptlrpc_request *request = NULL;
 450         struct md_op_data *op_data;
 451         struct ll_sb_info *sbi = ll_i2sbi(parent);
 452         struct inode *inode = NULL;
 453         struct dentry dentry;
 454         int err;
 455         ENTRY;
 456
 457         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 458                 RETURN(-EINVAL);
 459
 460         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
 461                "stripe_offset %d, stripe_count: %u\n",
 462                PFID(ll_inode2fid(parent)), parent, dirname,
 463                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 464
 465         if (lump->lum_stripe_count > 1 &&
 466             !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
 467                 RETURN(-EINVAL);
 468
 469         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 470                 lustre_swab_lmv_user_md(lump);
 471
 472         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 473                 mode &= ~current_umask();
 474         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 475         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 476                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 477                                      lump);
 478         if (IS_ERR(op_data))
 479                 GOTO(err_exit, err = PTR_ERR(op_data));
 480
 481         op_data->op_cli_flags |= CLI_SET_MEA;
 482         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 483                         from_kuid(&init_user_ns, current_fsuid()),
 484                         from_kgid(&init_user_ns, current_fsgid()),
 485                         cfs_curproc_cap_pack(), 0, &request);
 486         ll_finish_md_op_data(op_data);
 487         if (err)
 488                 GOTO(err_exit, err);
 489
 490         err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
 491         if (err)
 492                 GOTO(err_exit, err);
 493
 494         memset(&dentry, 0, sizeof(dentry));
 495         dentry.d_inode = inode;
 496
 497         err = ll_init_security(&dentry, inode, parent);
 498         iput(inode);
 499         if (err)
 500                 GOTO(err_exit, err);
 501
 502 err_exit:
 503         ptlrpc_req_finished(request);
 504         return err;
 505 }
 506
 507 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 508                      int set_default)
 509 {
 510         struct ll_sb_info *sbi = ll_i2sbi(inode);
 511         struct md_op_data *op_data;
 512         struct ptlrpc_request *req = NULL;
 513         int rc = 0;
 514         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 515         struct obd_device *mgc = lsi->lsi_mgc;
 516         int lum_size;
 517         ENTRY;
 518
 519         if (lump != NULL) {
 520                 /*
 521                  * This is coming from userspace, so should be in
 522                  * local endian.  But the MDS would like it in little
 523                  * endian, so we swab it before we send it.
 524                  */
 525                 switch (lump->lmm_magic) {
 526                 case LOV_USER_MAGIC_V1: {
 527                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 528                                 lustre_swab_lov_user_md_v1(lump);
 529                         lum_size = sizeof(struct lov_user_md_v1);
 530                         break;
 531                 }
 532                 case LOV_USER_MAGIC_V3: {
 533                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 534                                 lustre_swab_lov_user_md_v3(
 535                                         (struct lov_user_md_v3 *)lump);
 536                         lum_size = sizeof(struct lov_user_md_v3);
 537                         break;
 538                 }
 539                 case LMV_USER_MAGIC: {
 540                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 541                                 lustre_swab_lmv_user_md(
 542                                         (struct lmv_user_md *)lump);
 543                         lum_size = sizeof(struct lmv_user_md);
 544                         break;
 545                 }
 546                 default: {
 547                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 548                                         " %#08x != %#08x nor %#08x\n",
 549                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 550                                         LOV_USER_MAGIC_V3);
 551                         RETURN(-EINVAL);
 552                 }
 553                 }
 554         } else {
 555                 lum_size = sizeof(struct lov_user_md_v1);
 556         }
 557
 558         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 559                                      LUSTRE_OPC_ANY, NULL);
 560         if (IS_ERR(op_data))
 561                 RETURN(PTR_ERR(op_data));
 562
 563         /* swabbing is done in lov_setstripe() on server side */
 564         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
 565         ll_finish_md_op_data(op_data);
 566         ptlrpc_req_finished(req);
 567         if (rc) {
 568                 if (rc != -EPERM && rc != -EACCES)
 569                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 570         }
 571
 572         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 573          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 574          need the make the distiction between the 2 versions */
 575         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 576                 char *param = NULL;
 577                 char *buf;
 578
 579                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 580                 if (param == NULL)
 581                         GOTO(end, rc = -ENOMEM);
 582
 583                 buf = param;
 584                 /* Get fsname and assume devname to be -MDT0000. */
 585                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 586                 strcat(buf, "-MDT0000.lov");
 587                 buf += strlen(buf);
 588
 589                 /* Set root stripesize */
 590                 sprintf(buf, ".stripesize=%u",
 591                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 592                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 593                 if (rc)
 594                         GOTO(end, rc);
 595
 596                 /* Set root stripecount */
 597                 sprintf(buf, ".stripecount=%hd",
 598                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 599                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 600                 if (rc)
 601                         GOTO(end, rc);
 602
 603                 /* Set root stripeoffset */
 604                 sprintf(buf, ".stripeoffset=%hd",
 605                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 606                         (typeof(lump->lmm_stripe_offset))(-1));
 607                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 608
 609 end:
 610                 if (param != NULL)
 611                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 612         }
 613         RETURN(rc);
 614 }
 615
 616 /**
 617  * This function will be used to get default LOV/LMV/Default LMV
 618  * @valid will be used to indicate which stripe it will retrieve
 619  *      OBD_MD_MEA              LMV stripe EA
 620  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 621  *      otherwise               Default LOV EA.
 622  * Each time, it can only retrieve 1 stripe EA
 623  **/
 624 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 625                      struct ptlrpc_request **request, u64 valid)
 626 {
 627         struct ll_sb_info *sbi = ll_i2sbi(inode);
 628         struct mdt_body   *body;
 629         struct lov_mds_md *lmm = NULL;
 630         struct ptlrpc_request *req = NULL;
 631         int rc, lmm_size;
 632         struct md_op_data *op_data;
 633         ENTRY;
 634
 635         rc = ll_get_default_mdsize(sbi, &lmm_size);
 636         if (rc)
 637                 RETURN(rc);
 638
 639         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 640                                      0, lmm_size, LUSTRE_OPC_ANY,
 641                                      NULL);
 642         if (IS_ERR(op_data))
 643                 RETURN(PTR_ERR(op_data));
 644
 645         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 646         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 647         ll_finish_md_op_data(op_data);
 648         if (rc < 0) {
 649                 CDEBUG(D_INFO, "md_getattr failed on inode "
 650                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 651                 GOTO(out, rc);
 652         }
 653
 654         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 655         LASSERT(body != NULL);
 656
 657         lmm_size = body->mbo_eadatasize;
 658
 659         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 660             lmm_size == 0) {
 661                 GOTO(out, rc = -ENODATA);
 662         }
 663
 664         lmm = req_capsule_server_sized_get(&req->rq_pill,
 665                                            &RMF_MDT_MD, lmm_size);
 666         LASSERT(lmm != NULL);
 667
 668         /*
 669          * This is coming from the MDS, so is probably in
 670          * little endian.  We convert it to host endian before
 671          * passing it to userspace.
 672          */
 673         /* We don't swab objects for directories */
 674         switch (le32_to_cpu(lmm->lmm_magic)) {
 675         case LOV_MAGIC_V1:
 676                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 677                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 678                 break;
 679         case LOV_MAGIC_V3:
 680                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 681                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 682                 break;
 683         case LMV_MAGIC_V1:
 684                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 685                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 686                 break;
 687         case LMV_USER_MAGIC:
 688                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 689                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 690                 break;
 691         default:
 692                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 693                 rc = -EPROTO;
 694         }
 695 out:
 696         *plmm = lmm;
 697         *plmm_size = lmm_size;
 698         *request = req;
 699         return rc;
 700 }
 701
 702 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 703 {
 704         struct md_op_data       *op_data;
 705         int                     rc;
 706         int                     mdt_index;
 707         ENTRY;
 708
 709         OBD_ALLOC_PTR(op_data);
 710         if (op_data == NULL)
 711                 RETURN(-ENOMEM);
 712
 713         op_data->op_flags |= MF_GET_MDT_IDX;
 714         op_data->op_fid1 = *fid;
 715         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 716         mdt_index = op_data->op_mds;
 717         OBD_FREE_PTR(op_data);
 718         if (rc < 0)
 719                 RETURN(rc);
 720
 721         RETURN(mdt_index);
 722 }
 723
 724 /*
 725  *  Get MDT index for the inode.
 726  */
 727 int ll_get_mdt_idx(struct inode *inode)
 728 {
 729         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 730 }
 731
 732 /**
 733  * Generic handler to do any pre-copy work.
 734  *
 735  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 736  * first information for it that real work has started.
 737  *
 738  * Moreover, for a ARCHIVE request, it will sample the file data version and
 739  * store it in \a copy.
 740  *
 741  * \return 0 on success.
 742  */
 743 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 744 {
 745         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 746         struct hsm_progress_kernel       hpk;
 747         int                              rc = 0;
 748         int                              rc2;
 749         ENTRY;
 750
 751         /* Forge a hsm_progress based on data from copy. */
 752         hpk.hpk_fid = copy->hc_hai.hai_fid;
 753         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 754         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 755         hpk.hpk_extent.length = 0;
 756         hpk.hpk_flags = 0;
 757         hpk.hpk_errval = 0;
 758         hpk.hpk_data_version = 0;
 759
 760
 761         /* For archive request, we need to read the current file version. */
 762         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 763                 struct inode    *inode;
 764                 __u64            data_version = 0;
 765
 766                 /* Get inode for this fid */
 767                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 768                 if (IS_ERR(inode)) {
 769                         hpk.hpk_flags |= HP_FLAG_RETRY;
 770                         /* hpk_errval is >= 0 */
 771                         hpk.hpk_errval = -PTR_ERR(inode);
 772                         GOTO(progress, rc = PTR_ERR(inode));
 773                 }
 774
 775                 /* Read current file data version */
 776                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 777                 iput(inode);
 778                 if (rc != 0) {
 779                         CDEBUG(D_HSM, "Could not read file data version of "
 780                                       DFID" (rc = %d). Archive request ("
 781                                       LPX64") could not be done.\n",
 782                                       PFID(&copy->hc_hai.hai_fid), rc,
 783                                       copy->hc_hai.hai_cookie);
 784                         hpk.hpk_flags |= HP_FLAG_RETRY;
 785                         /* hpk_errval must be >= 0 */
 786                         hpk.hpk_errval = -rc;
 787                         GOTO(progress, rc);
 788                 }
 789
 790                 /* Store in the hsm_copy for later copytool use.
 791                  * Always modified even if no lsm. */
 792                 copy->hc_data_version = data_version;
 793         }
 794
 795 progress:
 796         /* On error, the request should be considered as completed */
 797         if (hpk.hpk_errval > 0)
 798                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 799
 800         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 801                             &hpk, NULL);
 802
 803         /* Return first error */
 804         RETURN(rc != 0 ? rc : rc2);
 805 }
 806
 807 /**
 808  * Generic handler to do any post-copy work.
 809  *
 810  * It will send the last hsm_progress update to coordinator to inform it
 811  * that copy is finished and whether it was successful or not.
 812  *
 813  * Moreover,
 814  * - for ARCHIVE request, it will sample the file data version and compare it
 815  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 816  *   will be considered as failed.
 817  * - for RESTORE request, it will sample the file data version and send it to
 818  *   coordinator which is useful if the file was imported as 'released'.
 819  *
 820  * \return 0 on success.
 821  */
 822 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 823 {
 824         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 825         struct hsm_progress_kernel       hpk;
 826         int                              rc = 0;
 827         int                              rc2;
 828         ENTRY;
 829
 830         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 831         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 832          * initialized if copy_end was called with copy == NULL.
 833          */
 834
 835         /* Forge a hsm_progress based on data from copy. */
 836         hpk.hpk_fid = copy->hc_hai.hai_fid;
 837         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 838         hpk.hpk_extent = copy->hc_hai.hai_extent;
 839         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 840         hpk.hpk_errval = copy->hc_errval;
 841         hpk.hpk_data_version = 0;
 842
 843         /* For archive request, we need to check the file data was not changed.
 844          *
 845          * For restore request, we need to send the file data version, this is
 846          * useful when the file was created using hsm_import.
 847          */
 848         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 849              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 850             (copy->hc_errval == 0)) {
 851                 struct inode    *inode;
 852                 __u64            data_version = 0;
 853
 854                 /* Get lsm for this fid */
 855                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 856                 if (IS_ERR(inode)) {
 857                         hpk.hpk_flags |= HP_FLAG_RETRY;
 858                         /* hpk_errval must be >= 0 */
 859                         hpk.hpk_errval = -PTR_ERR(inode);
 860                         GOTO(progress, rc = PTR_ERR(inode));
 861                 }
 862
 863                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 864                 iput(inode);
 865                 if (rc) {
 866                         CDEBUG(D_HSM, "Could not read file data version. "
 867                                       "Request could not be confirmed.\n");
 868                         if (hpk.hpk_errval == 0)
 869                                 hpk.hpk_errval = -rc;
 870                         GOTO(progress, rc);
 871                 }
 872
 873                 /* Store in the hsm_copy for later copytool use.
 874                  * Always modified even if no lsm. */
 875                 hpk.hpk_data_version = data_version;
 876
 877                 /* File could have been stripped during archiving, so we need
 878                  * to check anyway. */
 879                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 880                     (copy->hc_data_version != data_version)) {
 881                         CDEBUG(D_HSM, "File data version mismatched. "
 882                               "File content was changed during archiving. "
 883                                DFID", start:"LPX64" current:"LPX64"\n",
 884                                PFID(&copy->hc_hai.hai_fid),
 885                                copy->hc_data_version, data_version);
 886                         /* File was changed, send error to cdt. Do not ask for
 887                          * retry because if a file is modified frequently,
 888                          * the cdt will loop on retried archive requests.
 889                          * The policy engine will ask for a new archive later
 890                          * when the file will not be modified for some tunable
 891                          * time */
 892                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 893                         rc = -EBUSY;
 894                         /* hpk_errval must be >= 0 */
 895                         hpk.hpk_errval = -rc;
 896                         GOTO(progress, rc);
 897                 }
 898
 899         }
 900
 901 progress:
 902         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 903                             &hpk, NULL);
 904
 905         /* Return first error */
 906         RETURN(rc != 0 ? rc : rc2);
 907 }
 908
 909
 910 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 911                           const void __user *data, size_t size)
 912 {
 913         void *copy;
 914         int rc;
 915
 916         OBD_ALLOC(copy, size);
 917         if (copy == NULL)
 918                 return -ENOMEM;
 919
 920         if (copy_from_user(copy, data, size)) {
 921                 rc = -EFAULT;
 922                 goto out;
 923         }
 924
 925         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 926 out:
 927         OBD_FREE(copy, size);
 928
 929         return rc;
 930 }
 931
 932 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 933 {
 934         int cmd = qctl->qc_cmd;
 935         int type = qctl->qc_type;
 936         int id = qctl->qc_id;
 937         int valid = qctl->qc_valid;
 938         int rc = 0;
 939         ENTRY;
 940
 941         switch (cmd) {
 942         case Q_SETQUOTA:
 943         case Q_SETINFO:
 944                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 945                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 946                         RETURN(-EPERM);
 947                 break;
 948         case Q_GETQUOTA:
 949                 if (((type == USRQUOTA &&
 950                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
 951                      (type == GRPQUOTA &&
 952                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
 953                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 954                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 955                         RETURN(-EPERM);
 956                 break;
 957         case Q_GETINFO:
 958                 break;
 959         default:
 960                 CERROR("unsupported quotactl op: %#x\n", cmd);
 961                 RETURN(-ENOTTY);
 962         }
 963
 964         if (valid != QC_GENERAL) {
 965                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 966                         RETURN(-EOPNOTSUPP);
 967
 968                 if (cmd == Q_GETINFO)
 969                         qctl->qc_cmd = Q_GETOINFO;
 970                 else if (cmd == Q_GETQUOTA)
 971                         qctl->qc_cmd = Q_GETOQUOTA;
 972                 else
 973                         RETURN(-EINVAL);
 974
 975                 switch (valid) {
 976                 case QC_MDTIDX:
 977                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 978                                            sizeof(*qctl), qctl, NULL);
 979                         break;
 980                 case QC_OSTIDX:
 981                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 982                                            sizeof(*qctl), qctl, NULL);
 983                         break;
 984                 case QC_UUID:
 985                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 986                                            sizeof(*qctl), qctl, NULL);
 987                         if (rc == -EAGAIN)
 988                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 989                                                    sbi->ll_dt_exp,
 990                                                    sizeof(*qctl), qctl, NULL);
 991                         break;
 992                 default:
 993                         rc = -EINVAL;
 994                         break;
 995                 }
 996
 997                 if (rc)
 998                         RETURN(rc);
 999
1000                 qctl->qc_cmd = cmd;
1001         } else {
1002                 struct obd_quotactl *oqctl;
1003
1004                 OBD_ALLOC_PTR(oqctl);
1005                 if (oqctl == NULL)
1006                         RETURN(-ENOMEM);
1007
1008                 QCTL_COPY(oqctl, qctl);
1009                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1010                 if (rc) {
1011                         OBD_FREE_PTR(oqctl);
1012                         RETURN(rc);
1013                 }
1014                 /* If QIF_SPACE is not set, client should collect the
1015                  * space usage from OSSs by itself */
1016                 if (cmd == Q_GETQUOTA &&
1017                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1018                     !oqctl->qc_dqblk.dqb_curspace) {
1019                         struct obd_quotactl *oqctl_tmp;
1020
1021                         OBD_ALLOC_PTR(oqctl_tmp);
1022                         if (oqctl_tmp == NULL)
1023                                 GOTO(out, rc = -ENOMEM);
1024
1025                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1026                         oqctl_tmp->qc_id = oqctl->qc_id;
1027                         oqctl_tmp->qc_type = oqctl->qc_type;
1028
1029                         /* collect space usage from OSTs */
1030                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1031                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1032                         if (!rc || rc == -EREMOTEIO) {
1033                                 oqctl->qc_dqblk.dqb_curspace =
1034                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1035                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1036                         }
1037
1038                         /* collect space & inode usage from MDTs */
1039                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1040                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1041                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1042                         if (!rc || rc == -EREMOTEIO) {
1043                                 oqctl->qc_dqblk.dqb_curspace +=
1044                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1045                                 oqctl->qc_dqblk.dqb_curinodes =
1046                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1047                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1048                         } else {
1049                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1050                         }
1051
1052                         OBD_FREE_PTR(oqctl_tmp);
1053                 }
1054 out:
1055                 QCTL_COPY(qctl, oqctl);
1056                 OBD_FREE_PTR(oqctl);
1057         }
1058
1059         RETURN(rc);
1060 }
1061
1062 static char *
1063 ll_getname(const char __user *filename)
1064 {
1065         int ret = 0, len;
1066         char *tmp = __getname();
1067
1068         if (!tmp)
1069                 return ERR_PTR(-ENOMEM);
1070
1071         len = strncpy_from_user(tmp, filename, PATH_MAX);
1072         if (len == 0)
1073                 ret = -ENOENT;
1074         else if (len > PATH_MAX)
1075                 ret = -ENAMETOOLONG;
1076
1077         if (ret) {
1078                 __putname(tmp);
1079                 tmp =  ERR_PTR(ret);
1080         }
1081         return tmp;
1082 }
1083
1084 #define ll_putname(filename) __putname(filename)
1085
1086 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1087 {
1088         struct inode *inode = file->f_path.dentry->d_inode;
1089         struct ll_sb_info *sbi = ll_i2sbi(inode);
1090         struct obd_ioctl_data *data;
1091         int rc = 0;
1092         ENTRY;
1093
1094         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1095                PFID(ll_inode2fid(inode)), inode, cmd);
1096
1097         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1098         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1099                 return -ENOTTY;
1100
1101         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1102         switch(cmd) {
1103         case FSFILT_IOC_GETFLAGS:
1104         case FSFILT_IOC_SETFLAGS:
1105                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1106         case FSFILT_IOC_GETVERSION_OLD:
1107         case FSFILT_IOC_GETVERSION:
1108                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1109         /* We need to special case any other ioctls we want to handle,
1110          * to send them to the MDS/OST as appropriate and to properly
1111          * network encode the arg field.
1112         case FSFILT_IOC_SETVERSION_OLD:
1113         case FSFILT_IOC_SETVERSION:
1114         */
1115         case LL_IOC_GET_MDTIDX: {
1116                 int mdtidx;
1117
1118                 mdtidx = ll_get_mdt_idx(inode);
1119                 if (mdtidx < 0)
1120                         RETURN(mdtidx);
1121
1122                 if (put_user((int)mdtidx, (int __user *)arg))
1123                         RETURN(-EFAULT);
1124
1125                 return 0;
1126         }
1127         case IOC_MDC_LOOKUP: {
1128                 int namelen, len = 0;
1129                 char *buf = NULL;
1130                 char *filename;
1131
1132                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1133                 if (rc != 0)
1134                         RETURN(rc);
1135                 data = (void *)buf;
1136
1137                 filename = data->ioc_inlbuf1;
1138                 namelen = strlen(filename);
1139                 if (namelen < 1) {
1140                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1141                         GOTO(out_free, rc = -EINVAL);
1142                 }
1143
1144                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1145                 if (rc < 0) {
1146                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1147                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1148                                filename, rc);
1149                         GOTO(out_free, rc);
1150                 }
1151 out_free:
1152                 obd_ioctl_freedata(buf, len);
1153                 return rc;
1154         }
1155         case LL_IOC_LMV_SETSTRIPE: {
1156                 struct lmv_user_md  *lum;
1157                 char            *buf = NULL;
1158                 char            *filename;
1159                 int              namelen = 0;
1160                 int              lumlen = 0;
1161                 umode_t          mode;
1162                 int              len;
1163                 int              rc;
1164
1165                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1166                 if (rc)
1167                         RETURN(rc);
1168
1169                 data = (void *)buf;
1170                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1171                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1172                         GOTO(lmv_out_free, rc = -EINVAL);
1173
1174                 filename = data->ioc_inlbuf1;
1175                 namelen = data->ioc_inllen1;
1176
1177                 if (namelen < 1) {
1178                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1179                         GOTO(lmv_out_free, rc = -EINVAL);
1180                 }
1181                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1182                 lumlen = data->ioc_inllen2;
1183
1184                 if (lum->lum_magic != LMV_USER_MAGIC ||
1185                     lumlen != sizeof(*lum)) {
1186                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1187                                filename, lum->lum_magic, lumlen, -EFAULT);
1188                         GOTO(lmv_out_free, rc = -EINVAL);
1189                 }
1190
1191 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1192                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1193 #else
1194                 mode = data->ioc_type;
1195 #endif
1196                 rc = ll_dir_setdirstripe(inode, lum, filename, mode);
1197 lmv_out_free:
1198                 obd_ioctl_freedata(buf, len);
1199                 RETURN(rc);
1200
1201         }
1202         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1203                 struct lmv_user_md        lum;
1204                 struct lmv_user_md __user *ulump =
1205                                         (struct lmv_user_md __user *)arg;
1206                 int                       rc;
1207
1208                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1209                         RETURN(-EFAULT);
1210
1211                 if (lum.lum_magic != LMV_USER_MAGIC)
1212                         RETURN(-EINVAL);
1213
1214                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1215
1216                 RETURN(rc);
1217         }
1218         case LL_IOC_LOV_SETSTRIPE: {
1219                 struct lov_user_md_v3 lumv3;
1220                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1221                 struct lov_user_md_v1 __user *lumv1p =
1222                         (struct lov_user_md_v1 __user *)arg;
1223                 struct lov_user_md_v3 __user *lumv3p =
1224                         (struct lov_user_md_v3 __user *)arg;
1225
1226                 int set_default = 0;
1227
1228                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1229                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1230                         sizeof(lumv3p->lmm_objects[0]));
1231                 /* first try with v1 which is smaller than v3 */
1232                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1233                         RETURN(-EFAULT);
1234
1235                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1236                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1237                                 RETURN(-EFAULT);
1238                 }
1239
1240                 if (inode->i_sb->s_root == file->f_path.dentry)
1241                         set_default = 1;
1242
1243                 /* in v1 and v3 cases lumv1 points to data */
1244                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1245
1246                 RETURN(rc);
1247         }
1248         case LL_IOC_LMV_GETSTRIPE: {
1249                 struct lmv_user_md __user *ulmv =
1250                                         (struct lmv_user_md __user *)arg;
1251                 struct lmv_user_md      lum;
1252                 struct ptlrpc_request   *request = NULL;
1253                 union lmv_mds_md        *lmm = NULL;
1254                 int                     lmmsize;
1255                 u64                     valid = 0;
1256                 struct lmv_user_md      *tmp = NULL;
1257                 int                     mdt_index;
1258                 int                     lum_size;
1259                 int                     stripe_count;
1260                 int                     max_stripe_count;
1261                 int                     i;
1262                 int                     rc;
1263
1264                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1265                         RETURN(-EFAULT);
1266
1267                 max_stripe_count = lum.lum_stripe_count;
1268                 /* lum_magic will indicate which stripe the ioctl will like
1269                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1270                  * is for default LMV stripe */
1271                 if (lum.lum_magic == LMV_MAGIC_V1)
1272                         valid |= OBD_MD_MEA;
1273                 else if (lum.lum_magic == LMV_USER_MAGIC)
1274                         valid |= OBD_MD_DEFAULT_MEA;
1275                 else
1276                         RETURN(-EINVAL);
1277
1278                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1279                                       valid);
1280                 if (rc != 0)
1281                         GOTO(finish_req, rc);
1282
1283                 /* Get default LMV EA */
1284                 if (lum.lum_magic == LMV_USER_MAGIC) {
1285                         if (rc != 0)
1286                                 GOTO(finish_req, rc);
1287
1288                         if (lmmsize > sizeof(*ulmv))
1289                                 GOTO(finish_req, rc = -EINVAL);
1290
1291                         if (copy_to_user(ulmv, lmm, lmmsize))
1292                                 GOTO(finish_req, rc = -EFAULT);
1293
1294                         GOTO(finish_req, rc);
1295                 }
1296
1297                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1298                 if (max_stripe_count < stripe_count) {
1299                         lum.lum_stripe_count = stripe_count;
1300                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1301                                 GOTO(finish_req, rc = -EFAULT);
1302                         GOTO(finish_req, rc = -E2BIG);
1303                 }
1304
1305                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1306                 OBD_ALLOC(tmp, lum_size);
1307                 if (tmp == NULL)
1308                         GOTO(finish_req, rc = -ENOMEM);
1309
1310                 mdt_index = ll_get_mdt_idx(inode);
1311                 if (mdt_index < 0)
1312                         GOTO(out_tmp, rc = -ENOMEM);
1313
1314                 tmp->lum_magic = LMV_MAGIC_V1;
1315                 tmp->lum_stripe_count = 0;
1316                 tmp->lum_stripe_offset = mdt_index;
1317                 for (i = 0; i < stripe_count; i++) {
1318                         struct lu_fid   fid;
1319
1320                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1321                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1322                         if (mdt_index < 0)
1323                                 GOTO(out_tmp, rc = mdt_index);
1324
1325                         tmp->lum_objects[i].lum_mds = mdt_index;
1326                         tmp->lum_objects[i].lum_fid = fid;
1327                         tmp->lum_stripe_count++;
1328                 }
1329
1330                 if (copy_to_user(ulmv, tmp, lum_size))
1331                         GOTO(out_tmp, rc = -EFAULT);
1332 out_tmp:
1333                 OBD_FREE(tmp, lum_size);
1334 finish_req:
1335                 ptlrpc_req_finished(request);
1336                 return rc;
1337         }
1338
1339         case LL_IOC_REMOVE_ENTRY: {
1340                 char            *filename = NULL;
1341                 int              namelen = 0;
1342                 int              rc;
1343
1344                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1345                  * unsupported server, which might crash the server(LU-2730),
1346                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1347                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1348                  * server will support REINT_RMENTRY XXX*/
1349                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1350                         RETURN(-ENOTSUPP);
1351
1352                 filename = ll_getname((const char __user *)arg);
1353                 if (IS_ERR(filename))
1354                         RETURN(PTR_ERR(filename));
1355
1356                 namelen = strlen(filename);
1357                 if (namelen < 1)
1358                         GOTO(out_rmdir, rc = -EINVAL);
1359
1360                 rc = ll_rmdir_entry(inode, filename, namelen);
1361 out_rmdir:
1362                 if (filename)
1363                         ll_putname(filename);
1364                 RETURN(rc);
1365         }
1366         case LL_IOC_LOV_SWAP_LAYOUTS:
1367                 RETURN(-EPERM);
1368         case IOC_OBD_STATFS:
1369                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1370         case LL_IOC_LOV_GETSTRIPE:
1371         case LL_IOC_MDC_GETINFO:
1372         case IOC_MDC_GETFILEINFO:
1373         case IOC_MDC_GETFILESTRIPE: {
1374                 struct ptlrpc_request *request = NULL;
1375                 struct lov_user_md __user *lump;
1376                 struct lov_mds_md *lmm = NULL;
1377                 struct mdt_body *body;
1378                 char *filename = NULL;
1379                 int lmmsize;
1380
1381                 if (cmd == IOC_MDC_GETFILEINFO ||
1382                     cmd == IOC_MDC_GETFILESTRIPE) {
1383                         filename = ll_getname((const char __user *)arg);
1384                         if (IS_ERR(filename))
1385                                 RETURN(PTR_ERR(filename));
1386
1387                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1388                                                       &lmmsize, &request);
1389                 } else {
1390                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1391                                               &request, 0);
1392                 }
1393
1394                 if (request) {
1395                         body = req_capsule_server_get(&request->rq_pill,
1396                                                       &RMF_MDT_BODY);
1397                         LASSERT(body != NULL);
1398                 } else {
1399                         GOTO(out_req, rc);
1400                 }
1401
1402                 if (rc < 0) {
1403                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1404                                                cmd == LL_IOC_MDC_GETINFO))
1405                                 GOTO(skip_lmm, rc = 0);
1406                         else
1407                                 GOTO(out_req, rc);
1408                 }
1409
1410                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1411                     cmd == LL_IOC_LOV_GETSTRIPE) {
1412                         lump = (struct lov_user_md __user *)arg;
1413                 } else {
1414                         struct lov_user_mds_data __user *lmdp;
1415                         lmdp = (struct lov_user_mds_data __user *)arg;
1416                         lump = &lmdp->lmd_lmm;
1417                 }
1418                 if (copy_to_user(lump, lmm, lmmsize)) {
1419                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1420                                 GOTO(out_req, rc = -EFAULT);
1421                         rc = -EOVERFLOW;
1422                 }
1423         skip_lmm:
1424                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1425                         struct lov_user_mds_data __user *lmdp;
1426                         lstat_t st = { 0 };
1427
1428                         st.st_dev       = inode->i_sb->s_dev;
1429                         st.st_mode      = body->mbo_mode;
1430                         st.st_nlink     = body->mbo_nlink;
1431                         st.st_uid       = body->mbo_uid;
1432                         st.st_gid       = body->mbo_gid;
1433                         st.st_rdev      = body->mbo_rdev;
1434                         st.st_size      = body->mbo_size;
1435                         st.st_blksize   = PAGE_CACHE_SIZE;
1436                         st.st_blocks    = body->mbo_blocks;
1437                         st.st_atime     = body->mbo_atime;
1438                         st.st_mtime     = body->mbo_mtime;
1439                         st.st_ctime     = body->mbo_ctime;
1440                         st.st_ino       = inode->i_ino;
1441
1442                         lmdp = (struct lov_user_mds_data __user *)arg;
1443                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1444                                 GOTO(out_req, rc = -EFAULT);
1445                 }
1446
1447                 EXIT;
1448         out_req:
1449                 ptlrpc_req_finished(request);
1450                 if (filename)
1451                         ll_putname(filename);
1452                 return rc;
1453         }
1454         case OBD_IOC_QUOTACTL: {
1455                 struct if_quotactl *qctl;
1456
1457                 OBD_ALLOC_PTR(qctl);
1458                 if (!qctl)
1459                         RETURN(-ENOMEM);
1460
1461                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1462                         GOTO(out_quotactl, rc = -EFAULT);
1463
1464                 rc = quotactl_ioctl(sbi, qctl);
1465
1466                 if (rc == 0 &&
1467                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1468                         rc = -EFAULT;
1469
1470         out_quotactl:
1471                 OBD_FREE_PTR(qctl);
1472                 RETURN(rc);
1473         }
1474         case OBD_IOC_GETDTNAME:
1475         case OBD_IOC_GETMDNAME:
1476                 RETURN(ll_get_obd_name(inode, cmd, arg));
1477         case LL_IOC_FLUSHCTX:
1478                 RETURN(ll_flush_ctx(inode));
1479 #ifdef CONFIG_FS_POSIX_ACL
1480         case LL_IOC_RMTACL: {
1481             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1482                 inode == inode->i_sb->s_root->d_inode) {
1483                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1484
1485                 LASSERT(fd != NULL);
1486                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1487                 if (!rc)
1488                         fd->fd_flags |= LL_FILE_RMTACL;
1489                 RETURN(rc);
1490             } else
1491                 RETURN(0);
1492         }
1493 #endif
1494         case LL_IOC_GETOBDCOUNT: {
1495                 u32 count, vallen;
1496                 struct obd_export *exp;
1497
1498                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1499                         RETURN(-EFAULT);
1500
1501                 /* get ost count when count is zero, get mdt count otherwise */
1502                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1503                 vallen = sizeof(count);
1504                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1505                                   KEY_TGT_COUNT, &vallen, &count);
1506                 if (rc) {
1507                         CERROR("get target count failed: %d\n", rc);
1508                         RETURN(rc);
1509                 }
1510
1511                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1512                         RETURN(-EFAULT);
1513
1514                 RETURN(0);
1515         }
1516         case LL_IOC_PATH2FID:
1517                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1518                                      sizeof(struct lu_fid)))
1519                         RETURN(-EFAULT);
1520                 RETURN(0);
1521         case LL_IOC_GET_CONNECT_FLAGS: {
1522                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1523                                      (void __user *)arg));
1524         }
1525         case OBD_IOC_CHANGELOG_SEND:
1526         case OBD_IOC_CHANGELOG_CLEAR:
1527                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1528                         RETURN(-EPERM);
1529
1530                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1531                                     sizeof(struct ioc_changelog));
1532                 RETURN(rc);
1533         case OBD_IOC_FID2PATH:
1534                 RETURN(ll_fid2path(inode, (void __user *)arg));
1535         case LL_IOC_GETPARENT:
1536                 RETURN(ll_getparent(file, (void __user *)arg));
1537         case LL_IOC_FID2MDTIDX: {
1538                 struct obd_export *exp = ll_i2mdexp(inode);
1539                 struct lu_fid     fid;
1540                 __u32             index;
1541
1542                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1543                                    sizeof(fid)))
1544                         RETURN(-EFAULT);
1545
1546                 /* Call mdc_iocontrol */
1547                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1548                                    (__u32 __user *)&index);
1549                 if (rc != 0)
1550                         RETURN(rc);
1551
1552                 RETURN(index);
1553         }
1554         case LL_IOC_HSM_REQUEST: {
1555                 struct hsm_user_request *hur;
1556                 ssize_t                  totalsize;
1557
1558                 OBD_ALLOC_PTR(hur);
1559                 if (hur == NULL)
1560                         RETURN(-ENOMEM);
1561
1562                 /* We don't know the true size yet; copy the fixed-size part */
1563                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1564                         OBD_FREE_PTR(hur);
1565                         RETURN(-EFAULT);
1566                 }
1567
1568                 /* Compute the whole struct size */
1569                 totalsize = hur_len(hur);
1570                 OBD_FREE_PTR(hur);
1571                 if (totalsize < 0)
1572                         RETURN(-E2BIG);
1573
1574                 /* Final size will be more than double totalsize */
1575                 if (totalsize >= MDS_MAXREQSIZE / 3)
1576                         RETURN(-E2BIG);
1577
1578                 OBD_ALLOC_LARGE(hur, totalsize);
1579                 if (hur == NULL)
1580                         RETURN(-ENOMEM);
1581
1582                 /* Copy the whole struct */
1583                 if (copy_from_user(hur, (void __user *)arg, totalsize)) {
1584                         OBD_FREE_LARGE(hur, totalsize);
1585                         RETURN(-EFAULT);
1586                 }
1587
1588                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1589                         const struct lu_fid *fid;
1590                         struct inode *f;
1591                         int i;
1592
1593                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1594                                 fid = &hur->hur_user_item[i].hui_fid;
1595                                 f = search_inode_for_lustre(inode->i_sb, fid);
1596                                 if (IS_ERR(f)) {
1597                                         rc = PTR_ERR(f);
1598                                         break;
1599                                 }
1600
1601                                 rc = ll_hsm_release(f);
1602                                 iput(f);
1603                                 if (rc != 0)
1604                                         break;
1605                         }
1606                 } else {
1607                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1608                                            hur, NULL);
1609                 }
1610
1611                 OBD_FREE_LARGE(hur, totalsize);
1612
1613                 RETURN(rc);
1614         }
1615         case LL_IOC_HSM_PROGRESS: {
1616                 struct hsm_progress_kernel      hpk;
1617                 struct hsm_progress             hp;
1618
1619                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1620                         RETURN(-EFAULT);
1621
1622                 hpk.hpk_fid = hp.hp_fid;
1623                 hpk.hpk_cookie = hp.hp_cookie;
1624                 hpk.hpk_extent = hp.hp_extent;
1625                 hpk.hpk_flags = hp.hp_flags;
1626                 hpk.hpk_errval = hp.hp_errval;
1627                 hpk.hpk_data_version = 0;
1628
1629                 /* File may not exist in Lustre; all progress
1630                  * reported to Lustre root */
1631                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1632                                    NULL);
1633                 RETURN(rc);
1634         }
1635         case LL_IOC_HSM_CT_START:
1636                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1637                         RETURN(-EPERM);
1638
1639                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1640                                     sizeof(struct lustre_kernelcomm));
1641                 RETURN(rc);
1642
1643         case LL_IOC_HSM_COPY_START: {
1644                 struct hsm_copy *copy;
1645                 int              rc;
1646
1647                 OBD_ALLOC_PTR(copy);
1648                 if (copy == NULL)
1649                         RETURN(-ENOMEM);
1650                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1651                         OBD_FREE_PTR(copy);
1652                         RETURN(-EFAULT);
1653                 }
1654
1655                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1656                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1657                         rc = -EFAULT;
1658
1659                 OBD_FREE_PTR(copy);
1660                 RETURN(rc);
1661         }
1662         case LL_IOC_HSM_COPY_END: {
1663                 struct hsm_copy *copy;
1664                 int              rc;
1665
1666                 OBD_ALLOC_PTR(copy);
1667                 if (copy == NULL)
1668                         RETURN(-ENOMEM);
1669                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1670                         OBD_FREE_PTR(copy);
1671                         RETURN(-EFAULT);
1672                 }
1673
1674                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1675                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1676                         rc = -EFAULT;
1677
1678                 OBD_FREE_PTR(copy);
1679                 RETURN(rc);
1680         }
1681         case LL_IOC_MIGRATE: {
1682                 char            *buf = NULL;
1683                 const char      *filename;
1684                 int             namelen = 0;
1685                 int             len;
1686                 int             rc;
1687                 int             mdtidx;
1688
1689                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1690                 if (rc < 0)
1691                         RETURN(rc);
1692
1693                 data = (struct obd_ioctl_data *)buf;
1694                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1695                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1696                         GOTO(migrate_free, rc = -EINVAL);
1697
1698                 filename = data->ioc_inlbuf1;
1699                 namelen = data->ioc_inllen1;
1700                 /* \0 is packed at the end of filename */
1701                 if (namelen < 1 || namelen != strlen(filename) + 1)
1702                         GOTO(migrate_free, rc = -EINVAL);
1703
1704                 if (data->ioc_inllen2 != sizeof(mdtidx))
1705                         GOTO(migrate_free, rc = -EINVAL);
1706                 mdtidx = *(int *)data->ioc_inlbuf2;
1707
1708                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1709 migrate_free:
1710                 obd_ioctl_freedata(buf, len);
1711
1712                 RETURN(rc);
1713         }
1714         default:
1715                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1716                                      (void __user *)arg));
1717         }
1718 }
1719
1720 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1721 {
1722         struct inode *inode = file->f_mapping->host;
1723         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1724         struct ll_sb_info *sbi = ll_i2sbi(inode);
1725         int api32 = ll_need_32bit_api(sbi);
1726         loff_t ret = -EINVAL;
1727         ENTRY;
1728
1729         mutex_lock(&inode->i_mutex);
1730         switch (origin) {
1731                 case SEEK_SET:
1732                         break;
1733                 case SEEK_CUR:
1734                         offset += file->f_pos;
1735                         break;
1736                 case SEEK_END:
1737                         if (offset > 0)
1738                                 GOTO(out, ret);
1739                         if (api32)
1740                                 offset += LL_DIR_END_OFF_32BIT;
1741                         else
1742                                 offset += LL_DIR_END_OFF;
1743                         break;
1744                 default:
1745                         GOTO(out, ret);
1746         }
1747
1748         if (offset >= 0 &&
1749             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1750              (!api32 && offset <= LL_DIR_END_OFF))) {
1751                 if (offset != file->f_pos) {
1752                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1753                             (!api32 && offset == LL_DIR_END_OFF))
1754                                 fd->lfd_pos = MDS_DIR_END_OFF;
1755                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1756                                 fd->lfd_pos = offset << 32;
1757                         else
1758                                 fd->lfd_pos = offset;
1759                         file->f_pos = offset;
1760                         file->f_version = 0;
1761                 }
1762                 ret = offset;
1763         }
1764         GOTO(out, ret);
1765
1766 out:
1767         mutex_unlock(&inode->i_mutex);
1768         return ret;
1769 }
1770
1771 static int ll_dir_open(struct inode *inode, struct file *file)
1772 {
1773         ENTRY;
1774         RETURN(ll_file_open(inode, file));
1775 }
1776
1777 static int ll_dir_release(struct inode *inode, struct file *file)
1778 {
1779         ENTRY;
1780         RETURN(ll_file_release(inode, file));
1781 }
1782
1783 const struct file_operations ll_dir_operations = {
1784         .llseek         = ll_dir_seek,
1785         .open           = ll_dir_open,
1786         .release        = ll_dir_release,
1787         .read           = generic_read_dir,
1788 #ifdef HAVE_DIR_CONTEXT
1789         .iterate        = ll_iterate,
1790 #else
1791         .readdir        = ll_readdir,
1792 #endif
1793         .unlocked_ioctl = ll_dir_ioctl,
1794         .fsync          = ll_fsync,
1795 };