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                         /* we do not notify caller */
 893                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 894                         /* hpk_errval must be >= 0 */
 895                         hpk.hpk_errval = EBUSY;
 896                 }
 897
 898         }
 899
 900 progress:
 901         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 902                             &hpk, NULL);
 903
 904         /* Return first error */
 905         RETURN(rc != 0 ? rc : rc2);
 906 }
 907
 908
 909 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 910                           const void __user *data, size_t size)
 911 {
 912         void *copy;
 913         int rc;
 914
 915         OBD_ALLOC(copy, size);
 916         if (copy == NULL)
 917                 return -ENOMEM;
 918
 919         if (copy_from_user(copy, data, size)) {
 920                 rc = -EFAULT;
 921                 goto out;
 922         }
 923
 924         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 925 out:
 926         OBD_FREE(copy, size);
 927
 928         return rc;
 929 }
 930
 931 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 932 {
 933         int cmd = qctl->qc_cmd;
 934         int type = qctl->qc_type;
 935         int id = qctl->qc_id;
 936         int valid = qctl->qc_valid;
 937         int rc = 0;
 938         ENTRY;
 939
 940         switch (cmd) {
 941         case Q_SETQUOTA:
 942         case Q_SETINFO:
 943                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 944                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 945                         RETURN(-EPERM);
 946                 break;
 947         case Q_GETQUOTA:
 948                 if (((type == USRQUOTA &&
 949                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
 950                      (type == GRPQUOTA &&
 951                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
 952                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 953                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 954                         RETURN(-EPERM);
 955                 break;
 956         case Q_GETINFO:
 957                 break;
 958         default:
 959                 CERROR("unsupported quotactl op: %#x\n", cmd);
 960                 RETURN(-ENOTTY);
 961         }
 962
 963         if (valid != QC_GENERAL) {
 964                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 965                         RETURN(-EOPNOTSUPP);
 966
 967                 if (cmd == Q_GETINFO)
 968                         qctl->qc_cmd = Q_GETOINFO;
 969                 else if (cmd == Q_GETQUOTA)
 970                         qctl->qc_cmd = Q_GETOQUOTA;
 971                 else
 972                         RETURN(-EINVAL);
 973
 974                 switch (valid) {
 975                 case QC_MDTIDX:
 976                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 977                                            sizeof(*qctl), qctl, NULL);
 978                         break;
 979                 case QC_OSTIDX:
 980                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 981                                            sizeof(*qctl), qctl, NULL);
 982                         break;
 983                 case QC_UUID:
 984                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 985                                            sizeof(*qctl), qctl, NULL);
 986                         if (rc == -EAGAIN)
 987                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 988                                                    sbi->ll_dt_exp,
 989                                                    sizeof(*qctl), qctl, NULL);
 990                         break;
 991                 default:
 992                         rc = -EINVAL;
 993                         break;
 994                 }
 995
 996                 if (rc)
 997                         RETURN(rc);
 998
 999                 qctl->qc_cmd = cmd;
1000         } else {
1001                 struct obd_quotactl *oqctl;
1002
1003                 OBD_ALLOC_PTR(oqctl);
1004                 if (oqctl == NULL)
1005                         RETURN(-ENOMEM);
1006
1007                 QCTL_COPY(oqctl, qctl);
1008                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1009                 if (rc) {
1010                         OBD_FREE_PTR(oqctl);
1011                         RETURN(rc);
1012                 }
1013                 /* If QIF_SPACE is not set, client should collect the
1014                  * space usage from OSSs by itself */
1015                 if (cmd == Q_GETQUOTA &&
1016                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1017                     !oqctl->qc_dqblk.dqb_curspace) {
1018                         struct obd_quotactl *oqctl_tmp;
1019
1020                         OBD_ALLOC_PTR(oqctl_tmp);
1021                         if (oqctl_tmp == NULL)
1022                                 GOTO(out, rc = -ENOMEM);
1023
1024                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1025                         oqctl_tmp->qc_id = oqctl->qc_id;
1026                         oqctl_tmp->qc_type = oqctl->qc_type;
1027
1028                         /* collect space usage from OSTs */
1029                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1030                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1031                         if (!rc || rc == -EREMOTEIO) {
1032                                 oqctl->qc_dqblk.dqb_curspace =
1033                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1034                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1035                         }
1036
1037                         /* collect space & inode usage from MDTs */
1038                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1039                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1040                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1041                         if (!rc || rc == -EREMOTEIO) {
1042                                 oqctl->qc_dqblk.dqb_curspace +=
1043                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1044                                 oqctl->qc_dqblk.dqb_curinodes =
1045                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1046                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1047                         } else {
1048                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1049                         }
1050
1051                         OBD_FREE_PTR(oqctl_tmp);
1052                 }
1053 out:
1054                 QCTL_COPY(qctl, oqctl);
1055                 OBD_FREE_PTR(oqctl);
1056         }
1057
1058         RETURN(rc);
1059 }
1060
1061 static char *
1062 ll_getname(const char __user *filename)
1063 {
1064         int ret = 0, len;
1065         char *tmp = __getname();
1066
1067         if (!tmp)
1068                 return ERR_PTR(-ENOMEM);
1069
1070         len = strncpy_from_user(tmp, filename, PATH_MAX);
1071         if (len == 0)
1072                 ret = -ENOENT;
1073         else if (len > PATH_MAX)
1074                 ret = -ENAMETOOLONG;
1075
1076         if (ret) {
1077                 __putname(tmp);
1078                 tmp =  ERR_PTR(ret);
1079         }
1080         return tmp;
1081 }
1082
1083 #define ll_putname(filename) __putname(filename)
1084
1085 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1086 {
1087         struct inode *inode = file->f_path.dentry->d_inode;
1088         struct ll_sb_info *sbi = ll_i2sbi(inode);
1089         struct obd_ioctl_data *data;
1090         int rc = 0;
1091         ENTRY;
1092
1093         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1094                PFID(ll_inode2fid(inode)), inode, cmd);
1095
1096         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1097         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1098                 return -ENOTTY;
1099
1100         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1101         switch(cmd) {
1102         case FSFILT_IOC_GETFLAGS:
1103         case FSFILT_IOC_SETFLAGS:
1104                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1105         case FSFILT_IOC_GETVERSION_OLD:
1106         case FSFILT_IOC_GETVERSION:
1107                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1108         /* We need to special case any other ioctls we want to handle,
1109          * to send them to the MDS/OST as appropriate and to properly
1110          * network encode the arg field.
1111         case FSFILT_IOC_SETVERSION_OLD:
1112         case FSFILT_IOC_SETVERSION:
1113         */
1114         case LL_IOC_GET_MDTIDX: {
1115                 int mdtidx;
1116
1117                 mdtidx = ll_get_mdt_idx(inode);
1118                 if (mdtidx < 0)
1119                         RETURN(mdtidx);
1120
1121                 if (put_user((int)mdtidx, (int __user *)arg))
1122                         RETURN(-EFAULT);
1123
1124                 return 0;
1125         }
1126         case IOC_MDC_LOOKUP: {
1127                 int namelen, len = 0;
1128                 char *buf = NULL;
1129                 char *filename;
1130
1131                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1132                 if (rc != 0)
1133                         RETURN(rc);
1134                 data = (void *)buf;
1135
1136                 filename = data->ioc_inlbuf1;
1137                 namelen = strlen(filename);
1138                 if (namelen < 1) {
1139                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1140                         GOTO(out_free, rc = -EINVAL);
1141                 }
1142
1143                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL);
1144                 if (rc < 0) {
1145                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1146                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1147                                filename, rc);
1148                         GOTO(out_free, rc);
1149                 }
1150 out_free:
1151                 obd_ioctl_freedata(buf, len);
1152                 return rc;
1153         }
1154         case LL_IOC_LMV_SETSTRIPE: {
1155                 struct lmv_user_md  *lum;
1156                 char            *buf = NULL;
1157                 char            *filename;
1158                 int              namelen = 0;
1159                 int              lumlen = 0;
1160                 umode_t          mode;
1161                 int              len;
1162                 int              rc;
1163
1164                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1165                 if (rc)
1166                         RETURN(rc);
1167
1168                 data = (void *)buf;
1169                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1170                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1171                         GOTO(lmv_out_free, rc = -EINVAL);
1172
1173                 filename = data->ioc_inlbuf1;
1174                 namelen = data->ioc_inllen1;
1175
1176                 if (namelen < 1) {
1177                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1178                         GOTO(lmv_out_free, rc = -EINVAL);
1179                 }
1180                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1181                 lumlen = data->ioc_inllen2;
1182
1183                 if (lum->lum_magic != LMV_USER_MAGIC ||
1184                     lumlen != sizeof(*lum)) {
1185                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1186                                filename, lum->lum_magic, lumlen, -EFAULT);
1187                         GOTO(lmv_out_free, rc = -EINVAL);
1188                 }
1189
1190 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1191                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1192 #else
1193                 mode = data->ioc_type;
1194 #endif
1195                 rc = ll_dir_setdirstripe(inode, lum, filename, mode);
1196 lmv_out_free:
1197                 obd_ioctl_freedata(buf, len);
1198                 RETURN(rc);
1199
1200         }
1201         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1202                 struct lmv_user_md        lum;
1203                 struct lmv_user_md __user *ulump =
1204                                         (struct lmv_user_md __user *)arg;
1205                 int                       rc;
1206
1207                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1208                         RETURN(-EFAULT);
1209
1210                 if (lum.lum_magic != LMV_USER_MAGIC)
1211                         RETURN(-EINVAL);
1212
1213                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1214
1215                 RETURN(rc);
1216         }
1217         case LL_IOC_LOV_SETSTRIPE: {
1218                 struct lov_user_md_v3 lumv3;
1219                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1220                 struct lov_user_md_v1 __user *lumv1p =
1221                         (struct lov_user_md_v1 __user *)arg;
1222                 struct lov_user_md_v3 __user *lumv3p =
1223                         (struct lov_user_md_v3 __user *)arg;
1224
1225                 int set_default = 0;
1226
1227                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1228                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1229                         sizeof(lumv3p->lmm_objects[0]));
1230                 /* first try with v1 which is smaller than v3 */
1231                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1232                         RETURN(-EFAULT);
1233
1234                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1235                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1236                                 RETURN(-EFAULT);
1237                 }
1238
1239                 if (inode->i_sb->s_root == file->f_path.dentry)
1240                         set_default = 1;
1241
1242                 /* in v1 and v3 cases lumv1 points to data */
1243                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1244
1245                 RETURN(rc);
1246         }
1247         case LL_IOC_LMV_GETSTRIPE: {
1248                 struct lmv_user_md __user *ulmv =
1249                                         (struct lmv_user_md __user *)arg;
1250                 struct lmv_user_md      lum;
1251                 struct ptlrpc_request   *request = NULL;
1252                 union lmv_mds_md        *lmm = NULL;
1253                 int                     lmmsize;
1254                 u64                     valid = 0;
1255                 struct lmv_user_md      *tmp = NULL;
1256                 int                     mdt_index;
1257                 int                     lum_size;
1258                 int                     stripe_count;
1259                 int                     max_stripe_count;
1260                 int                     i;
1261                 int                     rc;
1262
1263                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1264                         RETURN(-EFAULT);
1265
1266                 max_stripe_count = lum.lum_stripe_count;
1267                 /* lum_magic will indicate which stripe the ioctl will like
1268                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1269                  * is for default LMV stripe */
1270                 if (lum.lum_magic == LMV_MAGIC_V1)
1271                         valid |= OBD_MD_MEA;
1272                 else if (lum.lum_magic == LMV_USER_MAGIC)
1273                         valid |= OBD_MD_DEFAULT_MEA;
1274                 else
1275                         RETURN(-EINVAL);
1276
1277                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1278                                       valid);
1279                 if (rc != 0)
1280                         GOTO(finish_req, rc);
1281
1282                 /* Get default LMV EA */
1283                 if (lum.lum_magic == LMV_USER_MAGIC) {
1284                         if (rc != 0)
1285                                 GOTO(finish_req, rc);
1286
1287                         if (lmmsize > sizeof(*ulmv))
1288                                 GOTO(finish_req, rc = -EINVAL);
1289
1290                         if (copy_to_user(ulmv, lmm, lmmsize))
1291                                 GOTO(finish_req, rc = -EFAULT);
1292
1293                         GOTO(finish_req, rc);
1294                 }
1295
1296                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1297                 if (max_stripe_count < stripe_count) {
1298                         lum.lum_stripe_count = stripe_count;
1299                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1300                                 GOTO(finish_req, rc = -EFAULT);
1301                         GOTO(finish_req, rc = -E2BIG);
1302                 }
1303
1304                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1305                 OBD_ALLOC(tmp, lum_size);
1306                 if (tmp == NULL)
1307                         GOTO(finish_req, rc = -ENOMEM);
1308
1309                 mdt_index = ll_get_mdt_idx(inode);
1310                 if (mdt_index < 0)
1311                         GOTO(out_tmp, rc = -ENOMEM);
1312
1313                 tmp->lum_magic = LMV_MAGIC_V1;
1314                 tmp->lum_stripe_count = 0;
1315                 tmp->lum_stripe_offset = mdt_index;
1316                 for (i = 0; i < stripe_count; i++) {
1317                         struct lu_fid   fid;
1318
1319                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1320                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1321                         if (mdt_index < 0)
1322                                 GOTO(out_tmp, rc = mdt_index);
1323
1324                         tmp->lum_objects[i].lum_mds = mdt_index;
1325                         tmp->lum_objects[i].lum_fid = fid;
1326                         tmp->lum_stripe_count++;
1327                 }
1328
1329                 if (copy_to_user(ulmv, tmp, lum_size))
1330                         GOTO(out_tmp, rc = -EFAULT);
1331 out_tmp:
1332                 OBD_FREE(tmp, lum_size);
1333 finish_req:
1334                 ptlrpc_req_finished(request);
1335                 return rc;
1336         }
1337
1338         case LL_IOC_REMOVE_ENTRY: {
1339                 char            *filename = NULL;
1340                 int              namelen = 0;
1341                 int              rc;
1342
1343                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1344                  * unsupported server, which might crash the server(LU-2730),
1345                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1346                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1347                  * server will support REINT_RMENTRY XXX*/
1348                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1349                         RETURN(-ENOTSUPP);
1350
1351                 filename = ll_getname((const char __user *)arg);
1352                 if (IS_ERR(filename))
1353                         RETURN(PTR_ERR(filename));
1354
1355                 namelen = strlen(filename);
1356                 if (namelen < 1)
1357                         GOTO(out_rmdir, rc = -EINVAL);
1358
1359                 rc = ll_rmdir_entry(inode, filename, namelen);
1360 out_rmdir:
1361                 if (filename)
1362                         ll_putname(filename);
1363                 RETURN(rc);
1364         }
1365         case LL_IOC_LOV_SWAP_LAYOUTS:
1366                 RETURN(-EPERM);
1367         case IOC_OBD_STATFS:
1368                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1369         case LL_IOC_LOV_GETSTRIPE:
1370         case LL_IOC_MDC_GETINFO:
1371         case IOC_MDC_GETFILEINFO:
1372         case IOC_MDC_GETFILESTRIPE: {
1373                 struct ptlrpc_request *request = NULL;
1374                 struct lov_user_md __user *lump;
1375                 struct lov_mds_md *lmm = NULL;
1376                 struct mdt_body *body;
1377                 char *filename = NULL;
1378                 int lmmsize;
1379
1380                 if (cmd == IOC_MDC_GETFILEINFO ||
1381                     cmd == IOC_MDC_GETFILESTRIPE) {
1382                         filename = ll_getname((const char __user *)arg);
1383                         if (IS_ERR(filename))
1384                                 RETURN(PTR_ERR(filename));
1385
1386                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1387                                                       &lmmsize, &request);
1388                 } else {
1389                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1390                                               &request, 0);
1391                 }
1392
1393                 if (request) {
1394                         body = req_capsule_server_get(&request->rq_pill,
1395                                                       &RMF_MDT_BODY);
1396                         LASSERT(body != NULL);
1397                 } else {
1398                         GOTO(out_req, rc);
1399                 }
1400
1401                 if (rc < 0) {
1402                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1403                                                cmd == LL_IOC_MDC_GETINFO))
1404                                 GOTO(skip_lmm, rc = 0);
1405                         else
1406                                 GOTO(out_req, rc);
1407                 }
1408
1409                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1410                     cmd == LL_IOC_LOV_GETSTRIPE) {
1411                         lump = (struct lov_user_md __user *)arg;
1412                 } else {
1413                         struct lov_user_mds_data __user *lmdp;
1414                         lmdp = (struct lov_user_mds_data __user *)arg;
1415                         lump = &lmdp->lmd_lmm;
1416                 }
1417                 if (copy_to_user(lump, lmm, lmmsize)) {
1418                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1419                                 GOTO(out_req, rc = -EFAULT);
1420                         rc = -EOVERFLOW;
1421                 }
1422         skip_lmm:
1423                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1424                         struct lov_user_mds_data __user *lmdp;
1425                         lstat_t st = { 0 };
1426
1427                         st.st_dev       = inode->i_sb->s_dev;
1428                         st.st_mode      = body->mbo_mode;
1429                         st.st_nlink     = body->mbo_nlink;
1430                         st.st_uid       = body->mbo_uid;
1431                         st.st_gid       = body->mbo_gid;
1432                         st.st_rdev      = body->mbo_rdev;
1433                         st.st_size      = body->mbo_size;
1434                         st.st_blksize   = PAGE_CACHE_SIZE;
1435                         st.st_blocks    = body->mbo_blocks;
1436                         st.st_atime     = body->mbo_atime;
1437                         st.st_mtime     = body->mbo_mtime;
1438                         st.st_ctime     = body->mbo_ctime;
1439                         st.st_ino       = inode->i_ino;
1440
1441                         lmdp = (struct lov_user_mds_data __user *)arg;
1442                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1443                                 GOTO(out_req, rc = -EFAULT);
1444                 }
1445
1446                 EXIT;
1447         out_req:
1448                 ptlrpc_req_finished(request);
1449                 if (filename)
1450                         ll_putname(filename);
1451                 return rc;
1452         }
1453         case OBD_IOC_QUOTACTL: {
1454                 struct if_quotactl *qctl;
1455
1456                 OBD_ALLOC_PTR(qctl);
1457                 if (!qctl)
1458                         RETURN(-ENOMEM);
1459
1460                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1461                         GOTO(out_quotactl, rc = -EFAULT);
1462
1463                 rc = quotactl_ioctl(sbi, qctl);
1464
1465                 if (rc == 0 &&
1466                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1467                         rc = -EFAULT;
1468
1469         out_quotactl:
1470                 OBD_FREE_PTR(qctl);
1471                 RETURN(rc);
1472         }
1473         case OBD_IOC_GETDTNAME:
1474         case OBD_IOC_GETMDNAME:
1475                 RETURN(ll_get_obd_name(inode, cmd, arg));
1476         case LL_IOC_FLUSHCTX:
1477                 RETURN(ll_flush_ctx(inode));
1478 #ifdef CONFIG_FS_POSIX_ACL
1479         case LL_IOC_RMTACL: {
1480             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1481                 inode == inode->i_sb->s_root->d_inode) {
1482                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1483
1484                 LASSERT(fd != NULL);
1485                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1486                 if (!rc)
1487                         fd->fd_flags |= LL_FILE_RMTACL;
1488                 RETURN(rc);
1489             } else
1490                 RETURN(0);
1491         }
1492 #endif
1493         case LL_IOC_GETOBDCOUNT: {
1494                 u32 count, vallen;
1495                 struct obd_export *exp;
1496
1497                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1498                         RETURN(-EFAULT);
1499
1500                 /* get ost count when count is zero, get mdt count otherwise */
1501                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1502                 vallen = sizeof(count);
1503                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1504                                   KEY_TGT_COUNT, &vallen, &count);
1505                 if (rc) {
1506                         CERROR("get target count failed: %d\n", rc);
1507                         RETURN(rc);
1508                 }
1509
1510                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1511                         RETURN(-EFAULT);
1512
1513                 RETURN(0);
1514         }
1515         case LL_IOC_PATH2FID:
1516                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1517                                      sizeof(struct lu_fid)))
1518                         RETURN(-EFAULT);
1519                 RETURN(0);
1520         case LL_IOC_GET_CONNECT_FLAGS: {
1521                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1522                                      (void __user *)arg));
1523         }
1524         case OBD_IOC_CHANGELOG_SEND:
1525         case OBD_IOC_CHANGELOG_CLEAR:
1526                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1527                         RETURN(-EPERM);
1528
1529                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1530                                     sizeof(struct ioc_changelog));
1531                 RETURN(rc);
1532         case OBD_IOC_FID2PATH:
1533                 RETURN(ll_fid2path(inode, (void __user *)arg));
1534         case LL_IOC_GETPARENT:
1535                 RETURN(ll_getparent(file, (void __user *)arg));
1536         case LL_IOC_FID2MDTIDX: {
1537                 struct obd_export *exp = ll_i2mdexp(inode);
1538                 struct lu_fid     fid;
1539                 __u32             index;
1540
1541                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1542                                    sizeof(fid)))
1543                         RETURN(-EFAULT);
1544
1545                 /* Call mdc_iocontrol */
1546                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1547                                    (__u32 __user *)&index);
1548                 if (rc != 0)
1549                         RETURN(rc);
1550
1551                 RETURN(index);
1552         }
1553         case LL_IOC_HSM_REQUEST: {
1554                 struct hsm_user_request *hur;
1555                 ssize_t                  totalsize;
1556
1557                 OBD_ALLOC_PTR(hur);
1558                 if (hur == NULL)
1559                         RETURN(-ENOMEM);
1560
1561                 /* We don't know the true size yet; copy the fixed-size part */
1562                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1563                         OBD_FREE_PTR(hur);
1564                         RETURN(-EFAULT);
1565                 }
1566
1567                 /* Compute the whole struct size */
1568                 totalsize = hur_len(hur);
1569                 OBD_FREE_PTR(hur);
1570                 if (totalsize < 0)
1571                         RETURN(-E2BIG);
1572
1573                 /* Final size will be more than double totalsize */
1574                 if (totalsize >= MDS_MAXREQSIZE / 3)
1575                         RETURN(-E2BIG);
1576
1577                 OBD_ALLOC_LARGE(hur, totalsize);
1578                 if (hur == NULL)
1579                         RETURN(-ENOMEM);
1580
1581                 /* Copy the whole struct */
1582                 if (copy_from_user(hur, (void __user *)arg, totalsize)) {
1583                         OBD_FREE_LARGE(hur, totalsize);
1584                         RETURN(-EFAULT);
1585                 }
1586
1587                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1588                         const struct lu_fid *fid;
1589                         struct inode *f;
1590                         int i;
1591
1592                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1593                                 fid = &hur->hur_user_item[i].hui_fid;
1594                                 f = search_inode_for_lustre(inode->i_sb, fid);
1595                                 if (IS_ERR(f)) {
1596                                         rc = PTR_ERR(f);
1597                                         break;
1598                                 }
1599
1600                                 rc = ll_hsm_release(f);
1601                                 iput(f);
1602                                 if (rc != 0)
1603                                         break;
1604                         }
1605                 } else {
1606                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1607                                            hur, NULL);
1608                 }
1609
1610                 OBD_FREE_LARGE(hur, totalsize);
1611
1612                 RETURN(rc);
1613         }
1614         case LL_IOC_HSM_PROGRESS: {
1615                 struct hsm_progress_kernel      hpk;
1616                 struct hsm_progress             hp;
1617
1618                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1619                         RETURN(-EFAULT);
1620
1621                 hpk.hpk_fid = hp.hp_fid;
1622                 hpk.hpk_cookie = hp.hp_cookie;
1623                 hpk.hpk_extent = hp.hp_extent;
1624                 hpk.hpk_flags = hp.hp_flags;
1625                 hpk.hpk_errval = hp.hp_errval;
1626                 hpk.hpk_data_version = 0;
1627
1628                 /* File may not exist in Lustre; all progress
1629                  * reported to Lustre root */
1630                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1631                                    NULL);
1632                 RETURN(rc);
1633         }
1634         case LL_IOC_HSM_CT_START:
1635                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1636                         RETURN(-EPERM);
1637
1638                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1639                                     sizeof(struct lustre_kernelcomm));
1640                 RETURN(rc);
1641
1642         case LL_IOC_HSM_COPY_START: {
1643                 struct hsm_copy *copy;
1644                 int              rc;
1645
1646                 OBD_ALLOC_PTR(copy);
1647                 if (copy == NULL)
1648                         RETURN(-ENOMEM);
1649                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1650                         OBD_FREE_PTR(copy);
1651                         RETURN(-EFAULT);
1652                 }
1653
1654                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1655                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1656                         rc = -EFAULT;
1657
1658                 OBD_FREE_PTR(copy);
1659                 RETURN(rc);
1660         }
1661         case LL_IOC_HSM_COPY_END: {
1662                 struct hsm_copy *copy;
1663                 int              rc;
1664
1665                 OBD_ALLOC_PTR(copy);
1666                 if (copy == NULL)
1667                         RETURN(-ENOMEM);
1668                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1669                         OBD_FREE_PTR(copy);
1670                         RETURN(-EFAULT);
1671                 }
1672
1673                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1674                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1675                         rc = -EFAULT;
1676
1677                 OBD_FREE_PTR(copy);
1678                 RETURN(rc);
1679         }
1680         case LL_IOC_MIGRATE: {
1681                 char            *buf = NULL;
1682                 const char      *filename;
1683                 int             namelen = 0;
1684                 int             len;
1685                 int             rc;
1686                 int             mdtidx;
1687
1688                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1689                 if (rc < 0)
1690                         RETURN(rc);
1691
1692                 data = (struct obd_ioctl_data *)buf;
1693                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1694                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1695                         GOTO(migrate_free, rc = -EINVAL);
1696
1697                 filename = data->ioc_inlbuf1;
1698                 namelen = data->ioc_inllen1;
1699                 /* \0 is packed at the end of filename */
1700                 if (namelen < 1 || namelen != strlen(filename) + 1)
1701                         GOTO(migrate_free, rc = -EINVAL);
1702
1703                 if (data->ioc_inllen2 != sizeof(mdtidx))
1704                         GOTO(migrate_free, rc = -EINVAL);
1705                 mdtidx = *(int *)data->ioc_inlbuf2;
1706
1707                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1708 migrate_free:
1709                 obd_ioctl_freedata(buf, len);
1710
1711                 RETURN(rc);
1712         }
1713         default:
1714                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1715                                      (void __user *)arg));
1716         }
1717 }
1718
1719 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1720 {
1721         struct inode *inode = file->f_mapping->host;
1722         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1723         struct ll_sb_info *sbi = ll_i2sbi(inode);
1724         int api32 = ll_need_32bit_api(sbi);
1725         loff_t ret = -EINVAL;
1726         ENTRY;
1727
1728         mutex_lock(&inode->i_mutex);
1729         switch (origin) {
1730                 case SEEK_SET:
1731                         break;
1732                 case SEEK_CUR:
1733                         offset += file->f_pos;
1734                         break;
1735                 case SEEK_END:
1736                         if (offset > 0)
1737                                 GOTO(out, ret);
1738                         if (api32)
1739                                 offset += LL_DIR_END_OFF_32BIT;
1740                         else
1741                                 offset += LL_DIR_END_OFF;
1742                         break;
1743                 default:
1744                         GOTO(out, ret);
1745         }
1746
1747         if (offset >= 0 &&
1748             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1749              (!api32 && offset <= LL_DIR_END_OFF))) {
1750                 if (offset != file->f_pos) {
1751                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1752                             (!api32 && offset == LL_DIR_END_OFF))
1753                                 fd->lfd_pos = MDS_DIR_END_OFF;
1754                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1755                                 fd->lfd_pos = offset << 32;
1756                         else
1757                                 fd->lfd_pos = offset;
1758                         file->f_pos = offset;
1759                         file->f_version = 0;
1760                 }
1761                 ret = offset;
1762         }
1763         GOTO(out, ret);
1764
1765 out:
1766         mutex_unlock(&inode->i_mutex);
1767         return ret;
1768 }
1769
1770 static int ll_dir_open(struct inode *inode, struct file *file)
1771 {
1772         ENTRY;
1773         RETURN(ll_file_open(inode, file));
1774 }
1775
1776 static int ll_dir_release(struct inode *inode, struct file *file)
1777 {
1778         ENTRY;
1779         RETURN(ll_file_release(inode, file));
1780 }
1781
1782 const struct file_operations ll_dir_operations = {
1783         .llseek         = ll_dir_seek,
1784         .open           = ll_dir_open,
1785         .release        = ll_dir_release,
1786         .read           = generic_read_dir,
1787 #ifdef HAVE_DIR_CONTEXT
1788         .iterate        = ll_iterate,
1789 #else
1790         .readdir        = ll_readdir,
1791 #endif
1792         .unlocked_ioctl = ll_dir_ioctl,
1793         .fsync          = ll_fsync,
1794 };