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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/dir.c
  33  *
  34  * Directory code for lustre client.
  35  */
  36
  37 #include <linux/fs.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/mm.h>
  40 #include <linux/version.h>
  41 #include <linux/security.h>
  42 #include <linux/user_namespace.h>
  43 #ifdef HAVE_UIDGID_HEADER
  44 # include <linux/uidgid.h>
  45 #endif
  46 #include <linux/uaccess.h>
  47 #include <linux/buffer_head.h>   // for wait_on_buffer
  48 #include <linux/pagevec.h>
  49
  50 #define DEBUG_SUBSYSTEM S_LLITE
  51
  52 #include <obd_support.h>
  53 #include <obd_class.h>
  54 #include <uapi/linux/lustre/lustre_ioctl.h>
  55 #include <lustre_lib.h>
  56 #include <lustre_dlm.h>
  57 #include <lustre_fid.h>
  58 #include <lustre_kernelcomm.h>
  59 #include <lustre_swab.h>
  60
  61 #include "llite_internal.h"
  62
  63 /*
  64  * (new) readdir implementation overview.
  65  *
  66  * Original lustre readdir implementation cached exact copy of raw directory
  67  * pages on the client. These pages were indexed in client page cache by
  68  * logical offset in the directory file. This design, while very simple and
  69  * intuitive had some inherent problems:
  70  *
  71  *     . it implies that byte offset to the directory entry serves as a
  72  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  73  *     ext3/htree directory entries may move due to splits, and more
  74  *     importantly,
  75  *
  76  *     . it is incompatible with the design of split directories for cmd3,
  77  *     that assumes that names are distributed across nodes based on their
  78  *     hash, and so readdir should be done in hash order.
  79  *
  80  * New readdir implementation does readdir in hash order, and uses hash of a
  81  * file name as a telldir/seekdir cookie. This led to number of complications:
  82  *
  83  *     . hash is not unique, so it cannot be used to index cached directory
  84  *     pages on the client (note, that it requires a whole pageful of hash
  85  *     collided entries to cause two pages to have identical hashes);
  86  *
  87  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  88  *     entry cookie. ext3/htree has the same problem and lustre implementation
  89  *     mimics their solution: seekdir(hash) positions directory at the first
  90  *     entry with the given hash.
  91  *
  92  * Client side.
  93  *
  94  * 0. caching
  95  *
  96  * Client caches directory pages using hash of the first entry as an index. As
  97  * noted above hash is not unique, so this solution doesn't work as is:
  98  * special processing is needed for "page hash chains" (i.e., sequences of
  99  * pages filled with entries all having the same hash value).
 100  *
 101  * First, such chains have to be detected. To this end, server returns to the
 102  * client the hash of the first entry on the page next to one returned. When
 103  * client detects that this hash is the same as hash of the first entry on the
 104  * returned page, page hash collision has to be handled. Pages in the
 105  * hash chain, except first one, are termed "overflow pages".
 106  *
 107  * Solution to index uniqueness problem is to not cache overflow
 108  * pages. Instead, when page hash collision is detected, all overflow pages
 109  * from emerging chain are immediately requested from the server and placed in
 110  * a special data structure (struct ll_dir_chain). This data structure is used
 111  * by ll_readdir() to process entries from overflow pages. When readdir
 112  * invocation finishes, overflow pages are discarded. If page hash collision
 113  * chain weren't completely processed, next call to readdir will again detect
 114  * page hash collision, again read overflow pages in, process next portion of
 115  * entries and again discard the pages. This is not as wasteful as it looks,
 116  * because, given reasonable hash, page hash collisions are extremely rare.
 117  *
 118  * 1. directory positioning
 119  *
 120  * When seekdir(hash) is called, original
 121  *
 122  *
 123  *
 124  *
 125  *
 126  *
 127  *
 128  *
 129  * Server.
 130  *
 131  * identification of and access to overflow pages
 132  *
 133  * page format
 134  *
 135  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 136  * a header lu_dirpage which describes the start/end hash, and whether this
 137  * page is empty (contains no dir entry) or hash collide with next page.
 138  * After client receives reply, several pages will be integrated into dir page
 139  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
 140  * lu_dirpage for this integrated page will be adjusted. See
 141  * mdc_adjust_dirpages().
 142  *
 143  */
 144 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
 145                              __u64 offset, struct ll_dir_chain *chain)
 146 {
 147         struct md_callback      cb_op;
 148         struct page             *page;
 149         int                     rc;
 150
 151         cb_op.md_blocking_ast = ll_md_blocking_ast;
 152         rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
 153         if (rc != 0)
 154                 return ERR_PTR(rc);
 155
 156         return page;
 157 }
 158
 159 void ll_release_page(struct inode *inode, struct page *page,
 160                      bool remove)
 161 {
 162         kunmap(page);
 163
 164         /* Always remove the page for striped dir, because the page is
 165          * built from temporarily in LMV layer */
 166         if (inode != NULL && S_ISDIR(inode->i_mode) &&
 167             ll_i2info(inode)->lli_lsm_md != NULL) {
 168                 __free_page(page);
 169                 return;
 170         }
 171
 172         if (remove) {
 173                 lock_page(page);
 174                 if (likely(page->mapping != NULL))
 175                         truncate_complete_page(page->mapping, page);
 176                 unlock_page(page);
 177         }
 178         put_page(page);
 179 }
 180
 181 /**
 182  * return IF_* type for given lu_dirent entry.
 183  * IF_* flag shld be converted to particular OS file type in
 184  * platform llite module.
 185  */
 186 static u16 ll_dirent_type_get(struct lu_dirent *ent)
 187 {
 188         u16 type = 0;
 189         struct luda_type *lt;
 190         int len = 0;
 191
 192         if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
 193                 const unsigned align = sizeof(struct luda_type) - 1;
 194
 195                 len = le16_to_cpu(ent->lde_namelen);
 196                 len = (len + align) & ~align;
 197                 lt = (void *)ent->lde_name + len;
 198                 type = IFTODT(le16_to_cpu(lt->lt_type));
 199         }
 200
 201         return type;
 202 }
 203
 204 #ifdef HAVE_DIR_CONTEXT
 205 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 206                 struct dir_context *ctx)
 207 {
 208 #else
 209 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 210                 void *cookie, filldir_t filldir)
 211 {
 212 #endif
 213         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 214         __u64                 pos        = *ppos;
 215         bool                  is_api32 = ll_need_32bit_api(sbi);
 216         bool                  is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 217         struct page          *page;
 218         struct ll_dir_chain   chain;
 219         bool                  done = false;
 220         int                   rc = 0;
 221         ENTRY;
 222
 223         ll_dir_chain_init(&chain);
 224
 225         page = ll_get_dir_page(inode, op_data, pos, &chain);
 226
 227         while (rc == 0 && !done) {
 228                 struct lu_dirpage *dp;
 229                 struct lu_dirent  *ent;
 230                 __u64 hash;
 231                 __u64 next;
 232
 233                 if (IS_ERR(page)) {
 234                         rc = PTR_ERR(page);
 235                         break;
 236                 }
 237
 238                 hash = MDS_DIR_END_OFF;
 239                 dp = page_address(page);
 240                 for (ent = lu_dirent_start(dp); ent != NULL && !done;
 241                      ent = lu_dirent_next(ent)) {
 242                         __u16          type;
 243                         int            namelen;
 244                         struct lu_fid  fid;
 245                         __u64          lhash;
 246                         __u64          ino;
 247
 248                         hash = le64_to_cpu(ent->lde_hash);
 249                         if (hash < pos) /* Skip until we find target hash */
 250                                 continue;
 251
 252                         namelen = le16_to_cpu(ent->lde_namelen);
 253                         if (namelen == 0) /* Skip dummy record */
 254                                 continue;
 255
 256                         if (is_api32 && is_hash64)
 257                                 lhash = hash >> 32;
 258                         else
 259                                 lhash = hash;
 260                         fid_le_to_cpu(&fid, &ent->lde_fid);
 261                         ino = cl_fid_build_ino(&fid, is_api32);
 262                         type = ll_dirent_type_get(ent);
 263                         /* For ll_nfs_get_name_filldir(), it will try to access
 264                          * 'ent' through 'lde_name', so the parameter 'name'
 265                          * for 'filldir()' must be part of the 'ent'. */
 266 #ifdef HAVE_DIR_CONTEXT
 267                         ctx->pos = lhash;
 268                         done = !dir_emit(ctx, ent->lde_name, namelen, ino,
 269                                          type);
 270 #else
 271                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 272                                        ino, type);
 273 #endif
 274                 }
 275
 276                 if (done) {
 277                         pos = hash;
 278                         ll_release_page(inode, page, false);
 279                         break;
 280                 }
 281
 282                 next = le64_to_cpu(dp->ldp_hash_end);
 283                 pos = next;
 284                 if (pos == MDS_DIR_END_OFF) {
 285                         /*
 286                          * End of directory reached.
 287                          */
 288                         done = 1;
 289                         ll_release_page(inode, page, false);
 290                 } else {
 291                         /*
 292                          * Normal case: continue to the next
 293                          * page.
 294                          */
 295                         ll_release_page(inode, page,
 296                                         le32_to_cpu(dp->ldp_flags) &
 297                                         LDF_COLLIDE);
 298                         next = pos;
 299                         page = ll_get_dir_page(inode, op_data, pos,
 300                                                &chain);
 301                 }
 302         }
 303 #ifdef HAVE_DIR_CONTEXT
 304         ctx->pos = pos;
 305 #else
 306         *ppos = pos;
 307 #endif
 308         ll_dir_chain_fini(&chain);
 309         RETURN(rc);
 310 }
 311
 312 #ifdef HAVE_DIR_CONTEXT
 313 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 314 #else
 315 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 316 #endif
 317 {
 318         struct inode            *inode  = file_inode(filp);
 319         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 320         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 321         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 322         int                     api32   = ll_need_32bit_api(sbi);
 323         struct md_op_data       *op_data;
 324         __u64                   pos;
 325         int                     rc;
 326         ENTRY;
 327
 328         if (lfd != NULL)
 329                 pos = lfd->lfd_pos;
 330         else
 331                 pos = 0;
 332
 333         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 334                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 335                inode, (unsigned long)pos, i_size_read(inode), api32);
 336
 337         if (pos == MDS_DIR_END_OFF)
 338                 /*
 339                  * end-of-file.
 340                  */
 341                 GOTO(out, rc = 0);
 342
 343         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 344                                      LUSTRE_OPC_ANY, inode);
 345         if (IS_ERR(op_data))
 346                 GOTO(out, rc = PTR_ERR(op_data));
 347
 348         if (unlikely(op_data->op_mea1 != NULL)) {
 349                 /* This is only needed for striped dir to fill ..,
 350                  * see lmv_read_entry */
 351                 if (file_dentry(filp)->d_parent != NULL &&
 352                     file_dentry(filp)->d_parent->d_inode != NULL) {
 353                         __u64 ibits = MDS_INODELOCK_UPDATE;
 354                         struct inode *parent =
 355                                 file_dentry(filp)->d_parent->d_inode;
 356
 357                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 358                                 op_data->op_fid3 = *ll_inode2fid(parent);
 359                 }
 360
 361                 /* If it can not find in cache, do lookup .. on the master
 362                  * object */
 363                 if (fid_is_zero(&op_data->op_fid3)) {
 364                         rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
 365                         if (rc != 0) {
 366                                 ll_finish_md_op_data(op_data);
 367                                 RETURN(rc);
 368                         }
 369                 }
 370         }
 371 #ifdef HAVE_DIR_CONTEXT
 372         ctx->pos = pos;
 373         rc = ll_dir_read(inode, &pos, op_data, ctx);
 374         pos = ctx->pos;
 375 #else
 376         rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
 377 #endif
 378         if (lfd != NULL)
 379                 lfd->lfd_pos = pos;
 380
 381         if (pos == MDS_DIR_END_OFF) {
 382                 if (api32)
 383                         pos = LL_DIR_END_OFF_32BIT;
 384                 else
 385                         pos = LL_DIR_END_OFF;
 386         } else {
 387                 if (api32 && hash64)
 388                         pos = pos >> 32;
 389         }
 390 #ifdef HAVE_DIR_CONTEXT
 391         ctx->pos = pos;
 392 #else
 393         filp->f_pos = pos;
 394 #endif
 395         ll_finish_md_op_data(op_data);
 396         filp->f_version = inode->i_version;
 397
 398 out:
 399         if (!rc)
 400                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 401
 402         RETURN(rc);
 403 }
 404
 405 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 406 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 407 {
 408         struct mgs_send_param *msp;
 409         int rc = 0;
 410
 411         OBD_ALLOC_PTR(msp);
 412         if (!msp)
 413                 return -ENOMEM;
 414
 415         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 416         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 417                                 sizeof(struct mgs_send_param), msp, NULL);
 418         if (rc)
 419                 CERROR("Failed to set parameter: %d\n", rc);
 420         OBD_FREE_PTR(msp);
 421
 422         return rc;
 423 }
 424 #endif
 425
 426 /**
 427  * Create striped directory with specified stripe(@lump)
 428  *
 429  * \param[in] dparent   the parent of the directory.
 430  * \param[in] lump      the specified stripes.
 431  * \param[in] dirname   the name of the directory.
 432  * \param[in] mode      the specified mode of the directory.
 433  *
 434  * \retval              =0 if striped directory is being created successfully.
 435  *                      <0 if the creation is failed.
 436  */
 437 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
 438                                size_t len, const char *dirname, umode_t mode)
 439 {
 440         struct inode *parent = dparent->d_inode;
 441         struct ptlrpc_request *request = NULL;
 442         struct md_op_data *op_data;
 443         struct ll_sb_info *sbi = ll_i2sbi(parent);
 444         struct inode *inode = NULL;
 445         struct dentry dentry = {
 446                 .d_parent = dparent,
 447                 .d_name = {
 448                         .name = dirname,
 449                         .len = strlen(dirname),
 450                         .hash = ll_full_name_hash(dparent, dirname,
 451                                                   strlen(dirname)),
 452                 },
 453         };
 454         int err;
 455         ENTRY;
 456
 457         if (unlikely(lump->lum_magic != LMV_USER_MAGIC &&
 458                      lump->lum_magic != LMV_USER_MAGIC_SPECIFIC))
 459                 RETURN(-EINVAL);
 460
 461         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s "
 462                "stripe_offset %d, stripe_count: %u\n",
 463                PFID(ll_inode2fid(parent)), parent, dirname,
 464                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 465
 466         if (lump->lum_stripe_count > 1 &&
 467             !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
 468                 RETURN(-EINVAL);
 469
 470         if (IS_DEADDIR(parent) &&
 471             !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
 472                 RETURN(-ENOENT);
 473
 474         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
 475             lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
 476                 lustre_swab_lmv_user_md(lump);
 477
 478         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 479                 mode &= ~current_umask();
 480         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 481         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 482                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 483                                      lump);
 484         if (IS_ERR(op_data))
 485                 RETURN(PTR_ERR(op_data));
 486
 487         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 488                 /* selinux_dentry_init_security() uses dentry->d_parent and name
 489                  * to determine the security context for the file. So our fake
 490                  * dentry should be real enough for this purpose. */
 491                 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
 492                                               &op_data->op_file_secctx_name,
 493                                               &op_data->op_file_secctx,
 494                                               &op_data->op_file_secctx_size);
 495                 if (err < 0)
 496                         GOTO(out_op_data, err);
 497         }
 498
 499         op_data->op_cli_flags |= CLI_SET_MEA;
 500         err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
 501                         from_kuid(&init_user_ns, current_fsuid()),
 502                         from_kgid(&init_user_ns, current_fsgid()),
 503                         cfs_curproc_cap_pack(), 0, &request);
 504         if (err)
 505                 GOTO(out_request, err);
 506
 507         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
 508
 509         err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
 510         if (err)
 511                 GOTO(out_inode, err);
 512
 513         dentry.d_inode = inode;
 514
 515         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 516                 inode_lock(inode);
 517                 err = security_inode_notifysecctx(inode,
 518                                                   op_data->op_file_secctx,
 519                                                   op_data->op_file_secctx_size);
 520                 inode_unlock(inode);
 521         } else {
 522                 err = ll_inode_init_security(&dentry, inode, parent);
 523         }
 524         if (err)
 525                 GOTO(out_inode, err);
 526
 527 out_inode:
 528         if (inode != NULL)
 529                 iput(inode);
 530 out_request:
 531         ptlrpc_req_finished(request);
 532 out_op_data:
 533         ll_finish_md_op_data(op_data);
 534
 535         return err;
 536 }
 537
 538 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 539                      int set_default)
 540 {
 541         struct ll_sb_info *sbi = ll_i2sbi(inode);
 542         struct md_op_data *op_data;
 543         struct ptlrpc_request *req = NULL;
 544         int rc = 0;
 545 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 546         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 547         struct obd_device *mgc = lsi->lsi_mgc;
 548 #endif
 549         int lum_size;
 550         ENTRY;
 551
 552         if (lump != NULL) {
 553                 /*
 554                  * This is coming from userspace, so should be in
 555                  * local endian.  But the MDS would like it in little
 556                  * endian, so we swab it before we send it.
 557                  */
 558                 switch (lump->lmm_magic) {
 559                 case LOV_USER_MAGIC_V1: {
 560                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 561                                 lustre_swab_lov_user_md_v1(lump);
 562                         lum_size = sizeof(struct lov_user_md_v1);
 563                         break;
 564                 }
 565                 case LOV_USER_MAGIC_V3: {
 566                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 567                                 lustre_swab_lov_user_md_v3(
 568                                         (struct lov_user_md_v3 *)lump);
 569                         lum_size = sizeof(struct lov_user_md_v3);
 570                         break;
 571                 }
 572                 case LOV_USER_MAGIC_COMP_V1: {
 573                         if (lump->lmm_magic !=
 574                             cpu_to_le32(LOV_USER_MAGIC_COMP_V1))
 575                                 lustre_swab_lov_comp_md_v1(
 576                                         (struct lov_comp_md_v1 *)lump);
 577                         lum_size = le32_to_cpu(
 578                                 ((struct lov_comp_md_v1 *)lump)->lcm_size);
 579                         break;
 580                 }
 581                 case LMV_USER_MAGIC: {
 582                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 583                                 lustre_swab_lmv_user_md(
 584                                         (struct lmv_user_md *)lump);
 585                         lum_size = sizeof(struct lmv_user_md);
 586                         break;
 587                 }
 588                 default: {
 589                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 590                                         " %#08x != %#08x nor %#08x\n",
 591                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 592                                         LOV_USER_MAGIC_V3);
 593                         RETURN(-EINVAL);
 594                 }
 595                 }
 596         } else {
 597                 lum_size = sizeof(struct lov_user_md_v1);
 598         }
 599
 600         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 601                                      LUSTRE_OPC_ANY, NULL);
 602         if (IS_ERR(op_data))
 603                 RETURN(PTR_ERR(op_data));
 604
 605         /* swabbing is done in lov_setstripe() on server side */
 606         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
 607         ll_finish_md_op_data(op_data);
 608         ptlrpc_req_finished(req);
 609         if (rc)
 610                 RETURN(rc);
 611
 612 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 613         /*
 614          * 2.9 server has stored filesystem default stripe in ROOT xattr,
 615          * and it's stored into system config for backward compatibility.
 616          *
 617          * In the following we use the fact that LOV_USER_MAGIC_V1 and
 618          * LOV_USER_MAGIC_V3 have the same initial fields so we do not
 619          * need the make the distiction between the 2 versions
 620          */
 621         if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
 622             (lump == NULL ||
 623              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
 624              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
 625                 char *param = NULL;
 626                 char *buf;
 627
 628                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 629                 if (param == NULL)
 630                         GOTO(end, rc = -ENOMEM);
 631
 632                 buf = param;
 633                 /* Get fsname and assume devname to be -MDT0000. */
 634                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 635                 strcat(buf, "-MDT0000.lov");
 636                 buf += strlen(buf);
 637
 638                 /* Set root stripesize */
 639                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
 640                          lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 641                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 642                 if (rc)
 643                         GOTO(end, rc);
 644
 645                 /* Set root stripecount */
 646                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
 647                          lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 648                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 649                 if (rc)
 650                         GOTO(end, rc);
 651
 652                 /* Set root stripeoffset */
 653                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
 654                          lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 655                                 (typeof(lump->lmm_stripe_offset))(-1));
 656                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 657
 658 end:
 659                 if (param != NULL)
 660                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 661         }
 662 #endif
 663         RETURN(rc);
 664 }
 665
 666 /**
 667  * This function will be used to get default LOV/LMV/Default LMV
 668  * @valid will be used to indicate which stripe it will retrieve
 669  *      OBD_MD_MEA              LMV stripe EA
 670  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 671  *      otherwise               Default LOV EA.
 672  * Each time, it can only retrieve 1 stripe EA
 673  **/
 674 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 675                      struct ptlrpc_request **request, u64 valid)
 676 {
 677         struct ll_sb_info *sbi = ll_i2sbi(inode);
 678         struct mdt_body   *body;
 679         struct lov_mds_md *lmm = NULL;
 680         struct ptlrpc_request *req = NULL;
 681         int rc, lmm_size;
 682         struct md_op_data *op_data;
 683         ENTRY;
 684
 685         rc = ll_get_default_mdsize(sbi, &lmm_size);
 686         if (rc)
 687                 RETURN(rc);
 688
 689         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 690                                      0, lmm_size, LUSTRE_OPC_ANY,
 691                                      NULL);
 692         if (IS_ERR(op_data))
 693                 RETURN(PTR_ERR(op_data));
 694
 695         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 696         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 697         ll_finish_md_op_data(op_data);
 698         if (rc < 0) {
 699                 CDEBUG(D_INFO, "md_getattr failed on inode "
 700                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 701                 GOTO(out, rc);
 702         }
 703
 704         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 705         LASSERT(body != NULL);
 706
 707         lmm_size = body->mbo_eadatasize;
 708
 709         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 710             lmm_size == 0) {
 711                 GOTO(out, rc = -ENODATA);
 712         }
 713
 714         lmm = req_capsule_server_sized_get(&req->rq_pill,
 715                                            &RMF_MDT_MD, lmm_size);
 716         LASSERT(lmm != NULL);
 717
 718         /*
 719          * This is coming from the MDS, so is probably in
 720          * little endian.  We convert it to host endian before
 721          * passing it to userspace.
 722          */
 723         /* We don't swab objects for directories */
 724         switch (le32_to_cpu(lmm->lmm_magic)) {
 725         case LOV_MAGIC_V1:
 726                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 727                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 728                 break;
 729         case LOV_MAGIC_V3:
 730                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 731                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 732                 break;
 733         case LOV_MAGIC_COMP_V1:
 734                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 735                         lustre_swab_lov_comp_md_v1(
 736                                         (struct lov_comp_md_v1 *)lmm);
 737                 break;
 738         case LMV_MAGIC_V1:
 739                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 740                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 741                 break;
 742         case LMV_USER_MAGIC:
 743                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 744                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 745                 break;
 746         default:
 747                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 748                 rc = -EPROTO;
 749         }
 750 out:
 751         *plmm = lmm;
 752         *plmm_size = lmm_size;
 753         *request = req;
 754         return rc;
 755 }
 756
 757 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 758 {
 759         struct md_op_data       *op_data;
 760         int                     rc;
 761         int                     mdt_index;
 762         ENTRY;
 763
 764         OBD_ALLOC_PTR(op_data);
 765         if (op_data == NULL)
 766                 RETURN(-ENOMEM);
 767
 768         op_data->op_flags |= MF_GET_MDT_IDX;
 769         op_data->op_fid1 = *fid;
 770         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 771         mdt_index = op_data->op_mds;
 772         OBD_FREE_PTR(op_data);
 773         if (rc < 0)
 774                 RETURN(rc);
 775
 776         RETURN(mdt_index);
 777 }
 778
 779 /*
 780  *  Get MDT index for the inode.
 781  */
 782 int ll_get_mdt_idx(struct inode *inode)
 783 {
 784         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 785 }
 786
 787 /**
 788  * Generic handler to do any pre-copy work.
 789  *
 790  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 791  * first information for it that real work has started.
 792  *
 793  * Moreover, for a ARCHIVE request, it will sample the file data version and
 794  * store it in \a copy.
 795  *
 796  * \return 0 on success.
 797  */
 798 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 799 {
 800         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 801         struct hsm_progress_kernel       hpk;
 802         int                              rc = 0;
 803         int                              rc2;
 804         ENTRY;
 805
 806         /* Forge a hsm_progress based on data from copy. */
 807         hpk.hpk_fid = copy->hc_hai.hai_fid;
 808         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 809         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 810         hpk.hpk_extent.length = 0;
 811         hpk.hpk_flags = 0;
 812         hpk.hpk_errval = 0;
 813         hpk.hpk_data_version = 0;
 814
 815
 816         /* For archive request, we need to read the current file version. */
 817         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 818                 struct inode    *inode;
 819                 __u64            data_version = 0;
 820
 821                 /* Get inode for this fid */
 822                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 823                 if (IS_ERR(inode)) {
 824                         hpk.hpk_flags |= HP_FLAG_RETRY;
 825                         /* hpk_errval is >= 0 */
 826                         hpk.hpk_errval = -PTR_ERR(inode);
 827                         GOTO(progress, rc = PTR_ERR(inode));
 828                 }
 829
 830                 /* Read current file data version */
 831                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 832                 iput(inode);
 833                 if (rc != 0) {
 834                         CDEBUG(D_HSM, "Could not read file data version of "
 835                                       DFID" (rc = %d). Archive request ("
 836                                       "%#llx) could not be done.\n",
 837                                       PFID(&copy->hc_hai.hai_fid), rc,
 838                                       copy->hc_hai.hai_cookie);
 839                         hpk.hpk_flags |= HP_FLAG_RETRY;
 840                         /* hpk_errval must be >= 0 */
 841                         hpk.hpk_errval = -rc;
 842                         GOTO(progress, rc);
 843                 }
 844
 845                 /* Store in the hsm_copy for later copytool use.
 846                  * Always modified even if no lsm. */
 847                 copy->hc_data_version = data_version;
 848         }
 849
 850 progress:
 851         /* On error, the request should be considered as completed */
 852         if (hpk.hpk_errval > 0)
 853                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 854
 855         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 856                             &hpk, NULL);
 857
 858         /* Return first error */
 859         RETURN(rc != 0 ? rc : rc2);
 860 }
 861
 862 /**
 863  * Generic handler to do any post-copy work.
 864  *
 865  * It will send the last hsm_progress update to coordinator to inform it
 866  * that copy is finished and whether it was successful or not.
 867  *
 868  * Moreover,
 869  * - for ARCHIVE request, it will sample the file data version and compare it
 870  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 871  *   will be considered as failed.
 872  * - for RESTORE request, it will sample the file data version and send it to
 873  *   coordinator which is useful if the file was imported as 'released'.
 874  *
 875  * \return 0 on success.
 876  */
 877 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 878 {
 879         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 880         struct hsm_progress_kernel       hpk;
 881         int                              rc = 0;
 882         int                              rc2;
 883         ENTRY;
 884
 885         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 886         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 887          * initialized if copy_end was called with copy == NULL.
 888          */
 889
 890         /* Forge a hsm_progress based on data from copy. */
 891         hpk.hpk_fid = copy->hc_hai.hai_fid;
 892         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 893         hpk.hpk_extent = copy->hc_hai.hai_extent;
 894         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 895         hpk.hpk_errval = copy->hc_errval;
 896         hpk.hpk_data_version = 0;
 897
 898         /* For archive request, we need to check the file data was not changed.
 899          *
 900          * For restore request, we need to send the file data version, this is
 901          * useful when the file was created using hsm_import.
 902          */
 903         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 904              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 905             (copy->hc_errval == 0)) {
 906                 struct inode    *inode;
 907                 __u64            data_version = 0;
 908
 909                 /* Get lsm for this fid */
 910                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 911                 if (IS_ERR(inode)) {
 912                         hpk.hpk_flags |= HP_FLAG_RETRY;
 913                         /* hpk_errval must be >= 0 */
 914                         hpk.hpk_errval = -PTR_ERR(inode);
 915                         GOTO(progress, rc = PTR_ERR(inode));
 916                 }
 917
 918                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 919                 iput(inode);
 920                 if (rc) {
 921                         CDEBUG(D_HSM, "Could not read file data version. "
 922                                       "Request could not be confirmed.\n");
 923                         if (hpk.hpk_errval == 0)
 924                                 hpk.hpk_errval = -rc;
 925                         GOTO(progress, rc);
 926                 }
 927
 928                 /* Store in the hsm_copy for later copytool use.
 929                  * Always modified even if no lsm. */
 930                 hpk.hpk_data_version = data_version;
 931
 932                 /* File could have been stripped during archiving, so we need
 933                  * to check anyway. */
 934                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 935                     (copy->hc_data_version != data_version)) {
 936                         CDEBUG(D_HSM, "File data version mismatched. "
 937                               "File content was changed during archiving. "
 938                                DFID", start:%#llx current:%#llx\n",
 939                                PFID(&copy->hc_hai.hai_fid),
 940                                copy->hc_data_version, data_version);
 941                         /* File was changed, send error to cdt. Do not ask for
 942                          * retry because if a file is modified frequently,
 943                          * the cdt will loop on retried archive requests.
 944                          * The policy engine will ask for a new archive later
 945                          * when the file will not be modified for some tunable
 946                          * time */
 947                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 948                         rc = -EBUSY;
 949                         /* hpk_errval must be >= 0 */
 950                         hpk.hpk_errval = -rc;
 951                         GOTO(progress, rc);
 952                 }
 953
 954         }
 955
 956 progress:
 957         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 958                             &hpk, NULL);
 959
 960         /* Return first error */
 961         RETURN(rc != 0 ? rc : rc2);
 962 }
 963
 964
 965 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 966                           const void __user *data, size_t size)
 967 {
 968         void *copy;
 969         int rc;
 970
 971         OBD_ALLOC(copy, size);
 972         if (copy == NULL)
 973                 return -ENOMEM;
 974
 975         if (copy_from_user(copy, data, size)) {
 976                 rc = -EFAULT;
 977                 goto out;
 978         }
 979
 980         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 981 out:
 982         OBD_FREE(copy, size);
 983
 984         return rc;
 985 }
 986
 987 static int check_owner(int type, int id)
 988 {
 989         switch (type) {
 990         case USRQUOTA:
 991                 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
 992                         return -EPERM;
 993                 break;
 994         case GRPQUOTA:
 995                 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
 996                         return -EPERM;
 997                 break;
 998         case PRJQUOTA:
 999                 break;
1000         }
1001         return 0;
1002 }
1003
1004 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1005 {
1006         int cmd = qctl->qc_cmd;
1007         int type = qctl->qc_type;
1008         int id = qctl->qc_id;
1009         int valid = qctl->qc_valid;
1010         int rc = 0;
1011         ENTRY;
1012
1013         switch (cmd) {
1014         case Q_SETQUOTA:
1015         case Q_SETINFO:
1016                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1017                         RETURN(-EPERM);
1018                 break;
1019         case Q_GETQUOTA:
1020                 if (check_owner(type, id) &&
1021                     (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1022                         RETURN(-EPERM);
1023                 break;
1024         case Q_GETINFO:
1025                 break;
1026         default:
1027                 CERROR("unsupported quotactl op: %#x\n", cmd);
1028                 RETURN(-ENOTTY);
1029         }
1030
1031         if (valid != QC_GENERAL) {
1032                 if (cmd == Q_GETINFO)
1033                         qctl->qc_cmd = Q_GETOINFO;
1034                 else if (cmd == Q_GETQUOTA)
1035                         qctl->qc_cmd = Q_GETOQUOTA;
1036                 else
1037                         RETURN(-EINVAL);
1038
1039                 switch (valid) {
1040                 case QC_MDTIDX:
1041                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1042                                            sizeof(*qctl), qctl, NULL);
1043                         break;
1044                 case QC_OSTIDX:
1045                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1046                                            sizeof(*qctl), qctl, NULL);
1047                         break;
1048                 case QC_UUID:
1049                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1050                                            sizeof(*qctl), qctl, NULL);
1051                         if (rc == -EAGAIN)
1052                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1053                                                    sbi->ll_dt_exp,
1054                                                    sizeof(*qctl), qctl, NULL);
1055                         break;
1056                 default:
1057                         rc = -EINVAL;
1058                         break;
1059                 }
1060
1061                 if (rc)
1062                         RETURN(rc);
1063
1064                 qctl->qc_cmd = cmd;
1065         } else {
1066                 struct obd_quotactl *oqctl;
1067
1068                 OBD_ALLOC_PTR(oqctl);
1069                 if (oqctl == NULL)
1070                         RETURN(-ENOMEM);
1071
1072                 QCTL_COPY(oqctl, qctl);
1073                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1074                 if (rc) {
1075                         OBD_FREE_PTR(oqctl);
1076                         RETURN(rc);
1077                 }
1078                 /* If QIF_SPACE is not set, client should collect the
1079                  * space usage from OSSs by itself */
1080                 if (cmd == Q_GETQUOTA &&
1081                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1082                     !oqctl->qc_dqblk.dqb_curspace) {
1083                         struct obd_quotactl *oqctl_tmp;
1084
1085                         OBD_ALLOC_PTR(oqctl_tmp);
1086                         if (oqctl_tmp == NULL)
1087                                 GOTO(out, rc = -ENOMEM);
1088
1089                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1090                         oqctl_tmp->qc_id = oqctl->qc_id;
1091                         oqctl_tmp->qc_type = oqctl->qc_type;
1092
1093                         /* collect space usage from OSTs */
1094                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1095                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1096                         if (!rc || rc == -EREMOTEIO) {
1097                                 oqctl->qc_dqblk.dqb_curspace =
1098                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1099                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1100                         }
1101
1102                         /* collect space & inode usage from MDTs */
1103                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1104                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1105                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1106                         if (!rc || rc == -EREMOTEIO) {
1107                                 oqctl->qc_dqblk.dqb_curspace +=
1108                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1109                                 oqctl->qc_dqblk.dqb_curinodes =
1110                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1111                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1112                         } else {
1113                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1114                         }
1115
1116                         OBD_FREE_PTR(oqctl_tmp);
1117                 }
1118 out:
1119                 QCTL_COPY(qctl, oqctl);
1120                 OBD_FREE_PTR(oqctl);
1121         }
1122
1123         RETURN(rc);
1124 }
1125
1126 /* This function tries to get a single name component,
1127  * to send to the server. No actual path traversal involved,
1128  * so we limit to NAME_MAX */
1129 static char *ll_getname(const char __user *filename)
1130 {
1131         int ret = 0, len;
1132         char *tmp;
1133
1134         OBD_ALLOC(tmp, NAME_MAX + 1);
1135
1136         if (!tmp)
1137                 return ERR_PTR(-ENOMEM);
1138
1139         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1140         if (len < 0)
1141                 ret = -ENOENT;
1142         else if (len > NAME_MAX)
1143                 ret = -ENAMETOOLONG;
1144
1145         if (ret) {
1146                 OBD_FREE(tmp, NAME_MAX + 1);
1147                 tmp =  ERR_PTR(ret);
1148         }
1149         return tmp;
1150 }
1151
1152 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1153
1154 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1155 {
1156         struct dentry *dentry = file_dentry(file);
1157         struct inode *inode = file_inode(file);
1158         struct ll_sb_info *sbi = ll_i2sbi(inode);
1159         struct obd_ioctl_data *data;
1160         int rc = 0;
1161         ENTRY;
1162
1163         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1164                PFID(ll_inode2fid(inode)), inode, cmd);
1165
1166         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1167         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1168                 return -ENOTTY;
1169
1170         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1171         switch (cmd) {
1172         case FS_IOC_GETFLAGS:
1173         case FS_IOC_SETFLAGS:
1174                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1175         case FSFILT_IOC_GETVERSION:
1176         case FS_IOC_GETVERSION:
1177                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1178         /* We need to special case any other ioctls we want to handle,
1179          * to send them to the MDS/OST as appropriate and to properly
1180          * network encode the arg field. */
1181         case FS_IOC_SETVERSION:
1182                 RETURN(-ENOTSUPP);
1183
1184         case LL_IOC_GET_MDTIDX: {
1185                 int mdtidx;
1186
1187                 mdtidx = ll_get_mdt_idx(inode);
1188                 if (mdtidx < 0)
1189                         RETURN(mdtidx);
1190
1191                 if (put_user((int)mdtidx, (int __user *)arg))
1192                         RETURN(-EFAULT);
1193
1194                 return 0;
1195         }
1196         case IOC_MDC_LOOKUP: {
1197                                      int namelen, len = 0;
1198                 char *buf = NULL;
1199                 char *filename;
1200
1201                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1202                 if (rc != 0)
1203                         RETURN(rc);
1204                 data = (void *)buf;
1205
1206                 filename = data->ioc_inlbuf1;
1207                 namelen = strlen(filename);
1208                 if (namelen < 1) {
1209                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1210                         GOTO(out_free, rc = -EINVAL);
1211                 }
1212
1213                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1214                 if (rc < 0) {
1215                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1216                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1217                                filename, rc);
1218                         GOTO(out_free, rc);
1219                 }
1220 out_free:
1221                 OBD_FREE_LARGE(buf, len);
1222                 return rc;
1223         }
1224         case LL_IOC_LMV_SETSTRIPE: {
1225                 struct lmv_user_md  *lum;
1226                 char            *buf = NULL;
1227                 char            *filename;
1228                 int              namelen = 0;
1229                 int              lumlen = 0;
1230                 umode_t          mode;
1231                 int              len;
1232                 int              rc;
1233
1234                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1235                 if (rc)
1236                         RETURN(rc);
1237
1238                 data = (void *)buf;
1239                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1240                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1241                         GOTO(lmv_out_free, rc = -EINVAL);
1242
1243                 filename = data->ioc_inlbuf1;
1244                 namelen = data->ioc_inllen1;
1245
1246                 if (namelen < 1) {
1247                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1248                         GOTO(lmv_out_free, rc = -EINVAL);
1249                 }
1250                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1251                 lumlen = data->ioc_inllen2;
1252
1253                 if ((lum->lum_magic != LMV_USER_MAGIC &&
1254                      lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) ||
1255                     lumlen < sizeof(*lum)) {
1256                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1257                                filename, lum->lum_magic, lumlen, -EFAULT);
1258                         GOTO(lmv_out_free, rc = -EINVAL);
1259                 }
1260
1261 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1262                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1263 #else
1264                 mode = data->ioc_type;
1265 #endif
1266                 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1267 lmv_out_free:
1268                 OBD_FREE_LARGE(buf, len);
1269                 RETURN(rc);
1270
1271         }
1272         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1273                 struct lmv_user_md        lum;
1274                 struct lmv_user_md __user *ulump =
1275                                         (struct lmv_user_md __user *)arg;
1276                 int                       rc;
1277
1278                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1279                         RETURN(-EFAULT);
1280
1281                 if (lum.lum_magic != LMV_USER_MAGIC)
1282                         RETURN(-EINVAL);
1283
1284                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1285
1286                 RETURN(rc);
1287         }
1288         case LL_IOC_LOV_SETSTRIPE_NEW:
1289         case LL_IOC_LOV_SETSTRIPE: {
1290                 struct lov_user_md_v3 lumv3;
1291                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1292                 struct lov_user_md_v1 __user *lumv1p =
1293                         (struct lov_user_md_v1 __user *)arg;
1294                 struct lov_user_md_v3 __user *lumv3p =
1295                         (struct lov_user_md_v3 __user *)arg;
1296
1297                 int set_default = 0;
1298
1299                 CLASSERT(sizeof(struct lov_user_md_v3) >
1300                          sizeof(struct lov_comp_md_v1));
1301                 CLASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1302                 CLASSERT(sizeof(lumv3.lmm_objects[0]) ==
1303                          sizeof(lumv3p->lmm_objects[0]));
1304                 /* first try with v1 which is smaller than v3 */
1305                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1306                         RETURN(-EFAULT);
1307
1308                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3)
1309                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1310                                 RETURN(-EFAULT);
1311
1312                 if (inode->i_sb->s_root == file_dentry(file))
1313                         set_default = 1;
1314
1315                 /* in v1 and v3 cases lumv1 points to data */
1316                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1317
1318                 RETURN(rc);
1319         }
1320         case LL_IOC_LMV_GETSTRIPE: {
1321                 struct lmv_user_md __user *ulmv =
1322                                         (struct lmv_user_md __user *)arg;
1323                 struct lmv_user_md      lum;
1324                 struct ptlrpc_request   *request = NULL;
1325                 union lmv_mds_md        *lmm = NULL;
1326                 int                     lmmsize;
1327                 u64                     valid = 0;
1328                 struct lmv_user_md      *tmp = NULL;
1329                 int                     mdt_index;
1330                 int                     lum_size;
1331                 int                     stripe_count;
1332                 int                     max_stripe_count;
1333                 int                     i;
1334                 int                     rc;
1335
1336                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1337                         RETURN(-EFAULT);
1338
1339                 max_stripe_count = lum.lum_stripe_count;
1340                 /* lum_magic will indicate which stripe the ioctl will like
1341                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1342                  * is for default LMV stripe */
1343                 if (lum.lum_magic == LMV_MAGIC_V1)
1344                         valid |= OBD_MD_MEA;
1345                 else if (lum.lum_magic == LMV_USER_MAGIC)
1346                         valid |= OBD_MD_DEFAULT_MEA;
1347                 else
1348                         RETURN(-EINVAL);
1349
1350                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1351                                       valid);
1352                 if (rc != 0)
1353                         GOTO(finish_req, rc);
1354
1355                 /* Get default LMV EA */
1356                 if (lum.lum_magic == LMV_USER_MAGIC) {
1357                         if (lmmsize > sizeof(*ulmv))
1358                                 GOTO(finish_req, rc = -EINVAL);
1359
1360                         if (copy_to_user(ulmv, lmm, lmmsize))
1361                                 GOTO(finish_req, rc = -EFAULT);
1362
1363                         GOTO(finish_req, rc);
1364                 }
1365
1366                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1367                 if (max_stripe_count < stripe_count) {
1368                         lum.lum_stripe_count = stripe_count;
1369                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1370                                 GOTO(finish_req, rc = -EFAULT);
1371                         GOTO(finish_req, rc = -E2BIG);
1372                 }
1373
1374                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1375                 OBD_ALLOC(tmp, lum_size);
1376                 if (tmp == NULL)
1377                         GOTO(finish_req, rc = -ENOMEM);
1378
1379                 mdt_index = ll_get_mdt_idx(inode);
1380                 if (mdt_index < 0)
1381                         GOTO(out_tmp, rc = -ENOMEM);
1382
1383                 tmp->lum_magic = LMV_MAGIC_V1;
1384                 tmp->lum_stripe_count = 0;
1385                 tmp->lum_stripe_offset = mdt_index;
1386                 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1387                 for (i = 0; i < stripe_count; i++) {
1388                         struct lu_fid   fid;
1389
1390                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1391                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1392                         if (mdt_index < 0)
1393                                 GOTO(out_tmp, rc = mdt_index);
1394
1395                         tmp->lum_objects[i].lum_mds = mdt_index;
1396                         tmp->lum_objects[i].lum_fid = fid;
1397                         tmp->lum_stripe_count++;
1398                 }
1399
1400                 if (copy_to_user(ulmv, tmp, lum_size))
1401                         GOTO(out_tmp, rc = -EFAULT);
1402 out_tmp:
1403                 OBD_FREE(tmp, lum_size);
1404 finish_req:
1405                 ptlrpc_req_finished(request);
1406                 return rc;
1407         }
1408
1409         case LL_IOC_REMOVE_ENTRY: {
1410                 char            *filename = NULL;
1411                 int              namelen = 0;
1412                 int              rc;
1413
1414                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1415                  * unsupported server, which might crash the server(LU-2730),
1416                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1417                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1418                  * server will support REINT_RMENTRY XXX*/
1419                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1420                         RETURN(-ENOTSUPP);
1421
1422                 filename = ll_getname((const char __user *)arg);
1423                 if (IS_ERR(filename))
1424                         RETURN(PTR_ERR(filename));
1425
1426                 namelen = strlen(filename);
1427                 if (namelen < 1)
1428                         GOTO(out_rmdir, rc = -EINVAL);
1429
1430                 rc = ll_rmdir_entry(inode, filename, namelen);
1431 out_rmdir:
1432                 if (filename)
1433                         ll_putname(filename);
1434                 RETURN(rc);
1435         }
1436         case LL_IOC_LOV_SWAP_LAYOUTS:
1437                 RETURN(-EPERM);
1438         case IOC_OBD_STATFS:
1439                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1440         case LL_IOC_LOV_GETSTRIPE:
1441         case LL_IOC_LOV_GETSTRIPE_NEW:
1442         case LL_IOC_MDC_GETINFO:
1443         case IOC_MDC_GETFILEINFO:
1444         case IOC_MDC_GETFILESTRIPE: {
1445                 struct ptlrpc_request *request = NULL;
1446                 struct lov_user_md __user *lump;
1447                 struct lov_mds_md *lmm = NULL;
1448                 struct mdt_body *body;
1449                 char *filename = NULL;
1450                 int lmmsize;
1451
1452                 if (cmd == IOC_MDC_GETFILEINFO ||
1453                     cmd == IOC_MDC_GETFILESTRIPE) {
1454                         filename = ll_getname((const char __user *)arg);
1455                         if (IS_ERR(filename))
1456                                 RETURN(PTR_ERR(filename));
1457
1458                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1459                                                       &lmmsize, &request);
1460                 } else {
1461                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1462                                               &request, 0);
1463                 }
1464
1465                 if (request) {
1466                         body = req_capsule_server_get(&request->rq_pill,
1467                                                       &RMF_MDT_BODY);
1468                         LASSERT(body != NULL);
1469                 } else {
1470                         GOTO(out_req, rc);
1471                 }
1472
1473                 if (rc < 0) {
1474                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1475                                                cmd == LL_IOC_MDC_GETINFO))
1476                                 GOTO(skip_lmm, rc = 0);
1477                         else
1478                                 GOTO(out_req, rc);
1479                 }
1480
1481                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1482                     cmd == LL_IOC_LOV_GETSTRIPE ||
1483                     cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1484                         lump = (struct lov_user_md __user *)arg;
1485                 } else {
1486                         struct lov_user_mds_data __user *lmdp;
1487                         lmdp = (struct lov_user_mds_data __user *)arg;
1488                         lump = &lmdp->lmd_lmm;
1489                 }
1490                 if (copy_to_user(lump, lmm, lmmsize)) {
1491                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1492                                 GOTO(out_req, rc = -EFAULT);
1493                         rc = -EOVERFLOW;
1494                 }
1495         skip_lmm:
1496                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1497                         struct lov_user_mds_data __user *lmdp;
1498                         lstat_t st = { 0 };
1499
1500                         st.st_dev       = inode->i_sb->s_dev;
1501                         st.st_mode      = body->mbo_mode;
1502                         st.st_nlink     = body->mbo_nlink;
1503                         st.st_uid       = body->mbo_uid;
1504                         st.st_gid       = body->mbo_gid;
1505                         st.st_rdev      = body->mbo_rdev;
1506                         st.st_size      = body->mbo_size;
1507                         st.st_blksize   = PAGE_SIZE;
1508                         st.st_blocks    = body->mbo_blocks;
1509                         st.st_atime     = body->mbo_atime;
1510                         st.st_mtime     = body->mbo_mtime;
1511                         st.st_ctime     = body->mbo_ctime;
1512                         st.st_ino       = cl_fid_build_ino(&body->mbo_fid1,
1513                                                 sbi->ll_flags &
1514                                                 LL_SBI_32BIT_API);
1515
1516                         lmdp = (struct lov_user_mds_data __user *)arg;
1517                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1518                                 GOTO(out_req, rc = -EFAULT);
1519                 }
1520
1521                 EXIT;
1522         out_req:
1523                 ptlrpc_req_finished(request);
1524                 if (filename)
1525                         ll_putname(filename);
1526                 return rc;
1527         }
1528         case OBD_IOC_QUOTACTL: {
1529                 struct if_quotactl *qctl;
1530
1531                 OBD_ALLOC_PTR(qctl);
1532                 if (!qctl)
1533                         RETURN(-ENOMEM);
1534
1535                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1536                         GOTO(out_quotactl, rc = -EFAULT);
1537
1538                 rc = quotactl_ioctl(sbi, qctl);
1539
1540                 if (rc == 0 &&
1541                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1542                         rc = -EFAULT;
1543
1544         out_quotactl:
1545                 OBD_FREE_PTR(qctl);
1546                 RETURN(rc);
1547         }
1548         case OBD_IOC_GETDTNAME:
1549         case OBD_IOC_GETMDNAME:
1550                 RETURN(ll_get_obd_name(inode, cmd, arg));
1551         case LL_IOC_FLUSHCTX:
1552                 RETURN(ll_flush_ctx(inode));
1553         case LL_IOC_GETOBDCOUNT: {
1554                 u32 count, vallen;
1555                 struct obd_export *exp;
1556
1557                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1558                         RETURN(-EFAULT);
1559
1560                 /* get ost count when count is zero, get mdt count otherwise */
1561                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1562                 vallen = sizeof(count);
1563                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1564                                   KEY_TGT_COUNT, &vallen, &count);
1565                 if (rc) {
1566                         CERROR("get target count failed: %d\n", rc);
1567                         RETURN(rc);
1568                 }
1569
1570                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1571                         RETURN(-EFAULT);
1572
1573                 RETURN(0);
1574         }
1575         case LL_IOC_PATH2FID:
1576                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1577                                      sizeof(struct lu_fid)))
1578                         RETURN(-EFAULT);
1579                 RETURN(0);
1580         case LL_IOC_GET_CONNECT_FLAGS: {
1581                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1582                                      (void __user *)arg));
1583         }
1584         case OBD_IOC_FID2PATH:
1585                 RETURN(ll_fid2path(inode, (void __user *)arg));
1586         case LL_IOC_GETPARENT:
1587                 RETURN(ll_getparent(file, (void __user *)arg));
1588         case LL_IOC_FID2MDTIDX: {
1589                 struct obd_export *exp = ll_i2mdexp(inode);
1590                 struct lu_fid     fid;
1591                 __u32             index;
1592
1593                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1594                                    sizeof(fid)))
1595                         RETURN(-EFAULT);
1596
1597                 /* Call mdc_iocontrol */
1598                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1599                                    (__u32 __user *)&index);
1600                 if (rc != 0)
1601                         RETURN(rc);
1602
1603                 RETURN(index);
1604         }
1605         case LL_IOC_HSM_REQUEST: {
1606                 struct hsm_user_request *hur;
1607                 ssize_t                  totalsize;
1608
1609                 OBD_ALLOC_PTR(hur);
1610                 if (hur == NULL)
1611                         RETURN(-ENOMEM);
1612
1613                 /* We don't know the true size yet; copy the fixed-size part */
1614                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1615                         OBD_FREE_PTR(hur);
1616                         RETURN(-EFAULT);
1617                 }
1618
1619                 /* Compute the whole struct size */
1620                 totalsize = hur_len(hur);
1621                 OBD_FREE_PTR(hur);
1622                 if (totalsize < 0)
1623                         RETURN(-E2BIG);
1624
1625                 /* Final size will be more than double totalsize */
1626                 if (totalsize >= MDS_MAXREQSIZE / 3)
1627                         RETURN(-E2BIG);
1628
1629                 OBD_ALLOC_LARGE(hur, totalsize);
1630                 if (hur == NULL)
1631                         RETURN(-ENOMEM);
1632
1633                 /* Copy the whole struct */
1634                 if (copy_from_user(hur, (void __user *)arg, totalsize))
1635                         GOTO(out_hur, rc = -EFAULT);
1636
1637                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1638                         const struct lu_fid *fid;
1639                         struct inode *f;
1640                         int i;
1641
1642                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1643                                 fid = &hur->hur_user_item[i].hui_fid;
1644                                 f = search_inode_for_lustre(inode->i_sb, fid);
1645                                 if (IS_ERR(f)) {
1646                                         rc = PTR_ERR(f);
1647                                         break;
1648                                 }
1649
1650                                 rc = ll_hsm_release(f);
1651                                 iput(f);
1652                                 if (rc != 0)
1653                                         break;
1654                         }
1655                 } else {
1656                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1657                                            hur, NULL);
1658                 }
1659
1660 out_hur:
1661                 OBD_FREE_LARGE(hur, totalsize);
1662
1663                 RETURN(rc);
1664         }
1665         case LL_IOC_HSM_PROGRESS: {
1666                 struct hsm_progress_kernel      hpk;
1667                 struct hsm_progress             hp;
1668
1669                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1670                         RETURN(-EFAULT);
1671
1672                 hpk.hpk_fid = hp.hp_fid;
1673                 hpk.hpk_cookie = hp.hp_cookie;
1674                 hpk.hpk_extent = hp.hp_extent;
1675                 hpk.hpk_flags = hp.hp_flags;
1676                 hpk.hpk_errval = hp.hp_errval;
1677                 hpk.hpk_data_version = 0;
1678
1679                 /* File may not exist in Lustre; all progress
1680                  * reported to Lustre root */
1681                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1682                                    NULL);
1683                 RETURN(rc);
1684         }
1685         case LL_IOC_HSM_CT_START:
1686                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1687                         RETURN(-EPERM);
1688
1689                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1690                                     sizeof(struct lustre_kernelcomm));
1691                 RETURN(rc);
1692
1693         case LL_IOC_HSM_COPY_START: {
1694                 struct hsm_copy *copy;
1695                 int              rc;
1696
1697                 OBD_ALLOC_PTR(copy);
1698                 if (copy == NULL)
1699                         RETURN(-ENOMEM);
1700                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1701                         OBD_FREE_PTR(copy);
1702                         RETURN(-EFAULT);
1703                 }
1704
1705                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1706                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1707                         rc = -EFAULT;
1708
1709                 OBD_FREE_PTR(copy);
1710                 RETURN(rc);
1711         }
1712         case LL_IOC_HSM_COPY_END: {
1713                 struct hsm_copy *copy;
1714                 int              rc;
1715
1716                 OBD_ALLOC_PTR(copy);
1717                 if (copy == NULL)
1718                         RETURN(-ENOMEM);
1719                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1720                         OBD_FREE_PTR(copy);
1721                         RETURN(-EFAULT);
1722                 }
1723
1724                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1725                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1726                         rc = -EFAULT;
1727
1728                 OBD_FREE_PTR(copy);
1729                 RETURN(rc);
1730         }
1731         case LL_IOC_MIGRATE: {
1732                 char            *buf = NULL;
1733                 const char      *filename;
1734                 int             namelen = 0;
1735                 int             len;
1736                 int             rc;
1737                 int             mdtidx;
1738
1739                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1740                 if (rc < 0)
1741                         RETURN(rc);
1742
1743                 data = (struct obd_ioctl_data *)buf;
1744                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1745                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1746                         GOTO(migrate_free, rc = -EINVAL);
1747
1748                 filename = data->ioc_inlbuf1;
1749                 namelen = data->ioc_inllen1;
1750                 /* \0 is packed at the end of filename */
1751                 if (namelen < 1 || namelen != strlen(filename) + 1)
1752                         GOTO(migrate_free, rc = -EINVAL);
1753
1754                 if (data->ioc_inllen2 != sizeof(mdtidx))
1755                         GOTO(migrate_free, rc = -EINVAL);
1756                 mdtidx = *(int *)data->ioc_inlbuf2;
1757
1758                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1759 migrate_free:
1760                 OBD_FREE_LARGE(buf, len);
1761
1762                 RETURN(rc);
1763         }
1764         case LL_IOC_FSGETXATTR:
1765                 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
1766         case LL_IOC_FSSETXATTR:
1767                 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
1768         default:
1769                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1770                                      (void __user *)arg));
1771         }
1772 }
1773
1774 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1775 {
1776         struct inode *inode = file->f_mapping->host;
1777         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1778         struct ll_sb_info *sbi = ll_i2sbi(inode);
1779         int api32 = ll_need_32bit_api(sbi);
1780         loff_t ret = -EINVAL;
1781         ENTRY;
1782
1783         inode_lock(inode);
1784         switch (origin) {
1785                 case SEEK_SET:
1786                         break;
1787                 case SEEK_CUR:
1788                         offset += file->f_pos;
1789                         break;
1790                 case SEEK_END:
1791                         if (offset > 0)
1792                                 GOTO(out, ret);
1793                         if (api32)
1794                                 offset += LL_DIR_END_OFF_32BIT;
1795                         else
1796                                 offset += LL_DIR_END_OFF;
1797                         break;
1798                 default:
1799                         GOTO(out, ret);
1800         }
1801
1802         if (offset >= 0 &&
1803             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1804              (!api32 && offset <= LL_DIR_END_OFF))) {
1805                 if (offset != file->f_pos) {
1806                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1807                             (!api32 && offset == LL_DIR_END_OFF))
1808                                 fd->lfd_pos = MDS_DIR_END_OFF;
1809                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1810                                 fd->lfd_pos = offset << 32;
1811                         else
1812                                 fd->lfd_pos = offset;
1813                         file->f_pos = offset;
1814                         file->f_version = 0;
1815                 }
1816                 ret = offset;
1817         }
1818         GOTO(out, ret);
1819
1820 out:
1821         inode_unlock(inode);
1822         return ret;
1823 }
1824
1825 static int ll_dir_open(struct inode *inode, struct file *file)
1826 {
1827         ENTRY;
1828         RETURN(ll_file_open(inode, file));
1829 }
1830
1831 static int ll_dir_release(struct inode *inode, struct file *file)
1832 {
1833         ENTRY;
1834         RETURN(ll_file_release(inode, file));
1835 }
1836
1837 const struct file_operations ll_dir_operations = {
1838         .llseek         = ll_dir_seek,
1839         .open           = ll_dir_open,
1840         .release        = ll_dir_release,
1841         .read           = generic_read_dir,
1842 #ifdef HAVE_DIR_CONTEXT
1843         .iterate        = ll_iterate,
1844 #else
1845         .readdir        = ll_readdir,
1846 #endif
1847         .unlocked_ioctl = ll_dir_ioctl,
1848         .fsync          = ll_fsync,
1849 };