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, 2016, 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/user_namespace.h>
  42 #ifdef HAVE_UIDGID_HEADER
  43 # include <linux/uidgid.h>
  44 #endif
  45 #include <asm/uaccess.h>
  46 #include <linux/buffer_head.h>   // for wait_on_buffer
  47 #include <linux/pagevec.h>
  48
  49 #define DEBUG_SUBSYSTEM S_LLITE
  50
  51 #include <obd_support.h>
  52 #include <obd_class.h>
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_lib.h>
  55 #include <lustre_dlm.h>
  56 #include <lustre_fid.h>
  57 #include <lustre_kernelcomm.h>
  58 #include <lustre_swab.h>
  59
  60 #include "llite_internal.h"
  61
  62 /*
  63  * (new) readdir implementation overview.
  64  *
  65  * Original lustre readdir implementation cached exact copy of raw directory
  66  * pages on the client. These pages were indexed in client page cache by
  67  * logical offset in the directory file. This design, while very simple and
  68  * intuitive had some inherent problems:
  69  *
  70  *     . it implies that byte offset to the directory entry serves as a
  71  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  72  *     ext3/htree directory entries may move due to splits, and more
  73  *     importantly,
  74  *
  75  *     . it is incompatible with the design of split directories for cmd3,
  76  *     that assumes that names are distributed across nodes based on their
  77  *     hash, and so readdir should be done in hash order.
  78  *
  79  * New readdir implementation does readdir in hash order, and uses hash of a
  80  * file name as a telldir/seekdir cookie. This led to number of complications:
  81  *
  82  *     . hash is not unique, so it cannot be used to index cached directory
  83  *     pages on the client (note, that it requires a whole pageful of hash
  84  *     collided entries to cause two pages to have identical hashes);
  85  *
  86  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  87  *     entry cookie. ext3/htree has the same problem and lustre implementation
  88  *     mimics their solution: seekdir(hash) positions directory at the first
  89  *     entry with the given hash.
  90  *
  91  * Client side.
  92  *
  93  * 0. caching
  94  *
  95  * Client caches directory pages using hash of the first entry as an index. As
  96  * noted above hash is not unique, so this solution doesn't work as is:
  97  * special processing is needed for "page hash chains" (i.e., sequences of
  98  * pages filled with entries all having the same hash value).
  99  *
 100  * First, such chains have to be detected. To this end, server returns to the
 101  * client the hash of the first entry on the page next to one returned. When
 102  * client detects that this hash is the same as hash of the first entry on the
 103  * returned page, page hash collision has to be handled. Pages in the
 104  * hash chain, except first one, are termed "overflow pages".
 105  *
 106  * Solution to index uniqueness problem is to not cache overflow
 107  * pages. Instead, when page hash collision is detected, all overflow pages
 108  * from emerging chain are immediately requested from the server and placed in
 109  * a special data structure (struct ll_dir_chain). This data structure is used
 110  * by ll_readdir() to process entries from overflow pages. When readdir
 111  * invocation finishes, overflow pages are discarded. If page hash collision
 112  * chain weren't completely processed, next call to readdir will again detect
 113  * page hash collision, again read overflow pages in, process next portion of
 114  * entries and again discard the pages. This is not as wasteful as it looks,
 115  * because, given reasonable hash, page hash collisions are extremely rare.
 116  *
 117  * 1. directory positioning
 118  *
 119  * When seekdir(hash) is called, original
 120  *
 121  *
 122  *
 123  *
 124  *
 125  *
 126  *
 127  *
 128  * Server.
 129  *
 130  * identification of and access to overflow pages
 131  *
 132  * page format
 133  *
 134  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 135  * a header lu_dirpage which describes the start/end hash, and whether this
 136  * page is empty (contains no dir entry) or hash collide with next page.
 137  * After client receives reply, several pages will be integrated into dir page
 138  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
 139  * lu_dirpage for this integrated page will be adjusted. See
 140  * mdc_adjust_dirpages().
 141  *
 142  */
 143 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
 144                              __u64 offset, struct ll_dir_chain *chain)
 145 {
 146         struct md_callback      cb_op;
 147         struct page             *page;
 148         int                     rc;
 149
 150         cb_op.md_blocking_ast = ll_md_blocking_ast;
 151         rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
 152         if (rc != 0)
 153                 return ERR_PTR(rc);
 154
 155         return page;
 156 }
 157
 158 void ll_release_page(struct inode *inode, struct page *page,
 159                      bool remove)
 160 {
 161         kunmap(page);
 162
 163         /* Always remove the page for striped dir, because the page is
 164          * built from temporarily in LMV layer */
 165         if (inode != NULL && S_ISDIR(inode->i_mode) &&
 166             ll_i2info(inode)->lli_lsm_md != NULL) {
 167                 __free_page(page);
 168                 return;
 169         }
 170
 171         if (remove) {
 172                 lock_page(page);
 173                 if (likely(page->mapping != NULL))
 174                         truncate_complete_page(page->mapping, page);
 175                 unlock_page(page);
 176         }
 177         put_page(page);
 178 }
 179
 180 /**
 181  * return IF_* type for given lu_dirent entry.
 182  * IF_* flag shld be converted to particular OS file type in
 183  * platform llite module.
 184  */
 185 static u16 ll_dirent_type_get(struct lu_dirent *ent)
 186 {
 187         u16 type = 0;
 188         struct luda_type *lt;
 189         int len = 0;
 190
 191         if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
 192                 const unsigned align = sizeof(struct luda_type) - 1;
 193
 194                 len = le16_to_cpu(ent->lde_namelen);
 195                 len = (len + align) & ~align;
 196                 lt = (void *)ent->lde_name + len;
 197                 type = IFTODT(le16_to_cpu(lt->lt_type));
 198         }
 199
 200         return type;
 201 }
 202
 203 #ifdef HAVE_DIR_CONTEXT
 204 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 205                 struct dir_context *ctx)
 206 {
 207 #else
 208 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 209                 void *cookie, filldir_t filldir)
 210 {
 211 #endif
 212         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 213         __u64                 pos        = *ppos;
 214         bool                  is_api32 = ll_need_32bit_api(sbi);
 215         bool                  is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 216         struct page          *page;
 217         struct ll_dir_chain   chain;
 218         bool                  done = false;
 219         int                   rc = 0;
 220         ENTRY;
 221
 222         ll_dir_chain_init(&chain);
 223
 224         page = ll_get_dir_page(inode, op_data, pos, &chain);
 225
 226         while (rc == 0 && !done) {
 227                 struct lu_dirpage *dp;
 228                 struct lu_dirent  *ent;
 229                 __u64 hash;
 230                 __u64 next;
 231
 232                 if (IS_ERR(page)) {
 233                         rc = PTR_ERR(page);
 234                         break;
 235                 }
 236
 237                 hash = MDS_DIR_END_OFF;
 238                 dp = page_address(page);
 239                 for (ent = lu_dirent_start(dp); ent != NULL && !done;
 240                      ent = lu_dirent_next(ent)) {
 241                         __u16          type;
 242                         int            namelen;
 243                         struct lu_fid  fid;
 244                         __u64          lhash;
 245                         __u64          ino;
 246
 247                         hash = le64_to_cpu(ent->lde_hash);
 248                         if (hash < pos) /* Skip until we find target hash */
 249                                 continue;
 250
 251                         namelen = le16_to_cpu(ent->lde_namelen);
 252                         if (namelen == 0) /* Skip dummy record */
 253                                 continue;
 254
 255                         if (is_api32 && is_hash64)
 256                                 lhash = hash >> 32;
 257                         else
 258                                 lhash = hash;
 259                         fid_le_to_cpu(&fid, &ent->lde_fid);
 260                         ino = cl_fid_build_ino(&fid, is_api32);
 261                         type = ll_dirent_type_get(ent);
 262                         /* For ll_nfs_get_name_filldir(), it will try to access
 263                          * 'ent' through 'lde_name', so the parameter 'name'
 264                          * for 'filldir()' must be part of the 'ent'. */
 265 #ifdef HAVE_DIR_CONTEXT
 266                         ctx->pos = lhash;
 267                         done = !dir_emit(ctx, ent->lde_name, namelen, ino,
 268                                          type);
 269 #else
 270                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 271                                        ino, type);
 272 #endif
 273                 }
 274
 275                 if (done) {
 276                         pos = hash;
 277                         ll_release_page(inode, page, false);
 278                         break;
 279                 }
 280
 281                 next = le64_to_cpu(dp->ldp_hash_end);
 282                 pos = next;
 283                 if (pos == MDS_DIR_END_OFF) {
 284                         /*
 285                          * End of directory reached.
 286                          */
 287                         done = 1;
 288                         ll_release_page(inode, page, false);
 289                 } else {
 290                         /*
 291                          * Normal case: continue to the next
 292                          * page.
 293                          */
 294                         ll_release_page(inode, page,
 295                                         le32_to_cpu(dp->ldp_flags) &
 296                                         LDF_COLLIDE);
 297                         next = pos;
 298                         page = ll_get_dir_page(inode, op_data, pos,
 299                                                &chain);
 300                 }
 301         }
 302 #ifdef HAVE_DIR_CONTEXT
 303         ctx->pos = pos;
 304 #else
 305         *ppos = pos;
 306 #endif
 307         ll_dir_chain_fini(&chain);
 308         RETURN(rc);
 309 }
 310
 311 #ifdef HAVE_DIR_CONTEXT
 312 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 313 #else
 314 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 315 #endif
 316 {
 317         struct inode            *inode  = file_inode(filp);
 318         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 319         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 320         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 321         int                     api32   = ll_need_32bit_api(sbi);
 322         struct md_op_data       *op_data;
 323         __u64                   pos;
 324         int                     rc;
 325         ENTRY;
 326
 327         if (lfd != NULL)
 328                 pos = lfd->lfd_pos;
 329         else
 330                 pos = 0;
 331
 332         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 333                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 334                inode, (unsigned long)pos, i_size_read(inode), api32);
 335
 336         if (pos == MDS_DIR_END_OFF)
 337                 /*
 338                  * end-of-file.
 339                  */
 340                 GOTO(out, rc = 0);
 341
 342         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 343                                      LUSTRE_OPC_ANY, inode);
 344         if (IS_ERR(op_data))
 345                 GOTO(out, rc = PTR_ERR(op_data));
 346
 347         if (unlikely(op_data->op_mea1 != NULL)) {
 348                 /* This is only needed for striped dir to fill ..,
 349                  * see lmv_read_entry */
 350                 if (file_dentry(filp)->d_parent != NULL &&
 351                     file_dentry(filp)->d_parent->d_inode != NULL) {
 352                         __u64 ibits = MDS_INODELOCK_UPDATE;
 353                         struct inode *parent =
 354                                 file_dentry(filp)->d_parent->d_inode;
 355
 356                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 357                                 op_data->op_fid3 = *ll_inode2fid(parent);
 358                 }
 359
 360                 /* If it can not find in cache, do lookup .. on the master
 361                  * object */
 362                 if (fid_is_zero(&op_data->op_fid3)) {
 363                         rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
 364                         if (rc != 0) {
 365                                 ll_finish_md_op_data(op_data);
 366                                 RETURN(rc);
 367                         }
 368                 }
 369         }
 370 #ifdef HAVE_DIR_CONTEXT
 371         ctx->pos = pos;
 372         rc = ll_dir_read(inode, &pos, op_data, ctx);
 373         pos = ctx->pos;
 374 #else
 375         rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
 376 #endif
 377         if (lfd != NULL)
 378                 lfd->lfd_pos = pos;
 379
 380         if (pos == MDS_DIR_END_OFF) {
 381                 if (api32)
 382                         pos = LL_DIR_END_OFF_32BIT;
 383                 else
 384                         pos = LL_DIR_END_OFF;
 385         } else {
 386                 if (api32 && hash64)
 387                         pos = pos >> 32;
 388         }
 389 #ifdef HAVE_DIR_CONTEXT
 390         ctx->pos = pos;
 391 #else
 392         filp->f_pos = pos;
 393 #endif
 394         ll_finish_md_op_data(op_data);
 395         filp->f_version = inode->i_version;
 396
 397 out:
 398         if (!rc)
 399                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 400
 401         RETURN(rc);
 402 }
 403
 404 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 405 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 406 {
 407         struct mgs_send_param *msp;
 408         int rc = 0;
 409
 410         OBD_ALLOC_PTR(msp);
 411         if (!msp)
 412                 return -ENOMEM;
 413
 414         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 415         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 416                                 sizeof(struct mgs_send_param), msp, NULL);
 417         if (rc)
 418                 CERROR("Failed to set parameter: %d\n", rc);
 419         OBD_FREE_PTR(msp);
 420
 421         return rc;
 422 }
 423 #endif
 424
 425 /**
 426  * Create striped directory with specified stripe(@lump)
 427  *
 428  * \param[in] dparent   the parent of the directory.
 429  * \param[in] lump      the specified stripes.
 430  * \param[in] dirname   the name of the directory.
 431  * \param[in] mode      the specified mode of the directory.
 432  *
 433  * \retval              =0 if striped directory is being created successfully.
 434  *                      <0 if the creation is failed.
 435  */
 436 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
 437                                const char *dirname, umode_t mode)
 438 {
 439         struct inode *parent = dparent->d_inode;
 440         struct ptlrpc_request *request = NULL;
 441         struct md_op_data *op_data;
 442         struct ll_sb_info *sbi = ll_i2sbi(parent);
 443         struct inode *inode = NULL;
 444         struct dentry dentry = {
 445                 .d_parent = dparent,
 446                 .d_name = {
 447                         .name = dirname,
 448                         .len = strlen(dirname),
 449                         .hash = ll_full_name_hash(dparent, dirname,
 450                                                   strlen(dirname)),
 451                 },
 452         };
 453         int err;
 454         ENTRY;
 455
 456         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 457                 RETURN(-EINVAL);
 458
 459         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s "
 460                "stripe_offset %d, stripe_count: %u\n",
 461                PFID(ll_inode2fid(parent)), parent, dirname,
 462                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 463
 464         if (lump->lum_stripe_count > 1 &&
 465             !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
 466                 RETURN(-EINVAL);
 467
 468         if (IS_DEADDIR(parent) &&
 469             !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
 470                 RETURN(-ENOENT);
 471
 472         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 473                 lustre_swab_lmv_user_md(lump);
 474
 475         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 476                 mode &= ~current_umask();
 477         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 478         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 479                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 480                                      lump);
 481         if (IS_ERR(op_data))
 482                 RETURN(PTR_ERR(op_data));
 483
 484         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 485                 /* selinux_dentry_init_security() uses dentry->d_parent and name
 486                  * to determine the security context for the file. So our fake
 487                  * dentry should be real enough for this purpose. */
 488                 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
 489                                               &op_data->op_file_secctx_name,
 490                                               &op_data->op_file_secctx,
 491                                               &op_data->op_file_secctx_size);
 492                 if (err < 0)
 493                         GOTO(out_op_data, err);
 494         }
 495
 496         op_data->op_cli_flags |= CLI_SET_MEA;
 497         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 498                         from_kuid(&init_user_ns, current_fsuid()),
 499                         from_kgid(&init_user_ns, current_fsgid()),
 500                         cfs_curproc_cap_pack(), 0, &request);
 501         if (err)
 502                 GOTO(out_request, err);
 503
 504         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
 505
 506         err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
 507         if (err)
 508                 GOTO(out_inode, err);
 509
 510         dentry.d_inode = inode;
 511
 512         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 513                 inode_lock(inode);
 514                 err = security_inode_notifysecctx(inode,
 515                                                   op_data->op_file_secctx,
 516                                                   op_data->op_file_secctx_size);
 517                 inode_unlock(inode);
 518         } else {
 519                 err = ll_inode_init_security(&dentry, inode, parent);
 520         }
 521         if (err)
 522                 GOTO(out_inode, err);
 523
 524 out_inode:
 525         if (inode != NULL)
 526                 iput(inode);
 527 out_request:
 528         ptlrpc_req_finished(request);
 529 out_op_data:
 530         ll_finish_md_op_data(op_data);
 531
 532         return err;
 533 }
 534
 535 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 536                      int set_default)
 537 {
 538         struct ll_sb_info *sbi = ll_i2sbi(inode);
 539         struct md_op_data *op_data;
 540         struct ptlrpc_request *req = NULL;
 541         int rc = 0;
 542 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 543         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 544         struct obd_device *mgc = lsi->lsi_mgc;
 545 #endif
 546         int lum_size;
 547         ENTRY;
 548
 549         if (lump != NULL) {
 550                 /*
 551                  * This is coming from userspace, so should be in
 552                  * local endian.  But the MDS would like it in little
 553                  * endian, so we swab it before we send it.
 554                  */
 555                 switch (lump->lmm_magic) {
 556                 case LOV_USER_MAGIC_V1: {
 557                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 558                                 lustre_swab_lov_user_md_v1(lump);
 559                         lum_size = sizeof(struct lov_user_md_v1);
 560                         break;
 561                 }
 562                 case LOV_USER_MAGIC_V3: {
 563                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 564                                 lustre_swab_lov_user_md_v3(
 565                                         (struct lov_user_md_v3 *)lump);
 566                         lum_size = sizeof(struct lov_user_md_v3);
 567                         break;
 568                 }
 569                 case LOV_USER_MAGIC_COMP_V1: {
 570                         if (lump->lmm_magic !=
 571                             cpu_to_le32(LOV_USER_MAGIC_COMP_V1))
 572                                 lustre_swab_lov_comp_md_v1(
 573                                         (struct lov_comp_md_v1 *)lump);
 574                         lum_size = le32_to_cpu(
 575                                 ((struct lov_comp_md_v1 *)lump)->lcm_size);
 576                         break;
 577                 }
 578                 case LMV_USER_MAGIC: {
 579                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 580                                 lustre_swab_lmv_user_md(
 581                                         (struct lmv_user_md *)lump);
 582                         lum_size = sizeof(struct lmv_user_md);
 583                         break;
 584                 }
 585                 default: {
 586                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 587                                         " %#08x != %#08x nor %#08x\n",
 588                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 589                                         LOV_USER_MAGIC_V3);
 590                         RETURN(-EINVAL);
 591                 }
 592                 }
 593         } else {
 594                 lum_size = sizeof(struct lov_user_md_v1);
 595         }
 596
 597         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 598                                      LUSTRE_OPC_ANY, NULL);
 599         if (IS_ERR(op_data))
 600                 RETURN(PTR_ERR(op_data));
 601
 602         /* swabbing is done in lov_setstripe() on server side */
 603         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
 604         ll_finish_md_op_data(op_data);
 605         ptlrpc_req_finished(req);
 606         if (rc)
 607                 RETURN(rc);
 608
 609 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 610         /*
 611          * 2.9 server has stored filesystem default stripe in ROOT xattr,
 612          * and it's stored into system config for backward compatibility.
 613          *
 614          * In the following we use the fact that LOV_USER_MAGIC_V1 and
 615          * LOV_USER_MAGIC_V3 have the same initial fields so we do not
 616          * need the make the distiction between the 2 versions
 617          */
 618         if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
 619             (lump == NULL ||
 620              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
 621              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
 622                 char *param = NULL;
 623                 char *buf;
 624
 625                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 626                 if (param == NULL)
 627                         GOTO(end, rc = -ENOMEM);
 628
 629                 buf = param;
 630                 /* Get fsname and assume devname to be -MDT0000. */
 631                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 632                 strcat(buf, "-MDT0000.lov");
 633                 buf += strlen(buf);
 634
 635                 /* Set root stripesize */
 636                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
 637                          lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 638                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 639                 if (rc)
 640                         GOTO(end, rc);
 641
 642                 /* Set root stripecount */
 643                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
 644                          lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 645                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 646                 if (rc)
 647                         GOTO(end, rc);
 648
 649                 /* Set root stripeoffset */
 650                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
 651                          lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 652                                 (typeof(lump->lmm_stripe_offset))(-1));
 653                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 654
 655 end:
 656                 if (param != NULL)
 657                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 658         }
 659 #endif
 660         RETURN(rc);
 661 }
 662
 663 /**
 664  * This function will be used to get default LOV/LMV/Default LMV
 665  * @valid will be used to indicate which stripe it will retrieve
 666  *      OBD_MD_MEA              LMV stripe EA
 667  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 668  *      otherwise               Default LOV EA.
 669  * Each time, it can only retrieve 1 stripe EA
 670  **/
 671 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 672                      struct ptlrpc_request **request, u64 valid)
 673 {
 674         struct ll_sb_info *sbi = ll_i2sbi(inode);
 675         struct mdt_body   *body;
 676         struct lov_mds_md *lmm = NULL;
 677         struct ptlrpc_request *req = NULL;
 678         int rc, lmm_size;
 679         struct md_op_data *op_data;
 680         ENTRY;
 681
 682         rc = ll_get_default_mdsize(sbi, &lmm_size);
 683         if (rc)
 684                 RETURN(rc);
 685
 686         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 687                                      0, lmm_size, LUSTRE_OPC_ANY,
 688                                      NULL);
 689         if (IS_ERR(op_data))
 690                 RETURN(PTR_ERR(op_data));
 691
 692         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 693         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 694         ll_finish_md_op_data(op_data);
 695         if (rc < 0) {
 696                 CDEBUG(D_INFO, "md_getattr failed on inode "
 697                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 698                 GOTO(out, rc);
 699         }
 700
 701         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 702         LASSERT(body != NULL);
 703
 704         lmm_size = body->mbo_eadatasize;
 705
 706         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 707             lmm_size == 0) {
 708                 GOTO(out, rc = -ENODATA);
 709         }
 710
 711         lmm = req_capsule_server_sized_get(&req->rq_pill,
 712                                            &RMF_MDT_MD, lmm_size);
 713         LASSERT(lmm != NULL);
 714
 715         /*
 716          * This is coming from the MDS, so is probably in
 717          * little endian.  We convert it to host endian before
 718          * passing it to userspace.
 719          */
 720         /* We don't swab objects for directories */
 721         switch (le32_to_cpu(lmm->lmm_magic)) {
 722         case LOV_MAGIC_V1:
 723                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 724                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 725                 break;
 726         case LOV_MAGIC_V3:
 727                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 728                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 729                 break;
 730         case LOV_MAGIC_COMP_V1:
 731                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 732                         lustre_swab_lov_comp_md_v1(
 733                                         (struct lov_comp_md_v1 *)lmm);
 734                 break;
 735         case LMV_MAGIC_V1:
 736                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 737                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 738                 break;
 739         case LMV_USER_MAGIC:
 740                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 741                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 742                 break;
 743         default:
 744                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 745                 rc = -EPROTO;
 746         }
 747 out:
 748         *plmm = lmm;
 749         *plmm_size = lmm_size;
 750         *request = req;
 751         return rc;
 752 }
 753
 754 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 755 {
 756         struct md_op_data       *op_data;
 757         int                     rc;
 758         int                     mdt_index;
 759         ENTRY;
 760
 761         OBD_ALLOC_PTR(op_data);
 762         if (op_data == NULL)
 763                 RETURN(-ENOMEM);
 764
 765         op_data->op_flags |= MF_GET_MDT_IDX;
 766         op_data->op_fid1 = *fid;
 767         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 768         mdt_index = op_data->op_mds;
 769         OBD_FREE_PTR(op_data);
 770         if (rc < 0)
 771                 RETURN(rc);
 772
 773         RETURN(mdt_index);
 774 }
 775
 776 /*
 777  *  Get MDT index for the inode.
 778  */
 779 int ll_get_mdt_idx(struct inode *inode)
 780 {
 781         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 782 }
 783
 784 /**
 785  * Generic handler to do any pre-copy work.
 786  *
 787  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 788  * first information for it that real work has started.
 789  *
 790  * Moreover, for a ARCHIVE request, it will sample the file data version and
 791  * store it in \a copy.
 792  *
 793  * \return 0 on success.
 794  */
 795 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 796 {
 797         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 798         struct hsm_progress_kernel       hpk;
 799         int                              rc = 0;
 800         int                              rc2;
 801         ENTRY;
 802
 803         /* Forge a hsm_progress based on data from copy. */
 804         hpk.hpk_fid = copy->hc_hai.hai_fid;
 805         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 806         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 807         hpk.hpk_extent.length = 0;
 808         hpk.hpk_flags = 0;
 809         hpk.hpk_errval = 0;
 810         hpk.hpk_data_version = 0;
 811
 812
 813         /* For archive request, we need to read the current file version. */
 814         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 815                 struct inode    *inode;
 816                 __u64            data_version = 0;
 817
 818                 /* Get inode for this fid */
 819                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 820                 if (IS_ERR(inode)) {
 821                         hpk.hpk_flags |= HP_FLAG_RETRY;
 822                         /* hpk_errval is >= 0 */
 823                         hpk.hpk_errval = -PTR_ERR(inode);
 824                         GOTO(progress, rc = PTR_ERR(inode));
 825                 }
 826
 827                 /* Read current file data version */
 828                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 829                 iput(inode);
 830                 if (rc != 0) {
 831                         CDEBUG(D_HSM, "Could not read file data version of "
 832                                       DFID" (rc = %d). Archive request ("
 833                                       "%#llx) could not be done.\n",
 834                                       PFID(&copy->hc_hai.hai_fid), rc,
 835                                       copy->hc_hai.hai_cookie);
 836                         hpk.hpk_flags |= HP_FLAG_RETRY;
 837                         /* hpk_errval must be >= 0 */
 838                         hpk.hpk_errval = -rc;
 839                         GOTO(progress, rc);
 840                 }
 841
 842                 /* Store in the hsm_copy for later copytool use.
 843                  * Always modified even if no lsm. */
 844                 copy->hc_data_version = data_version;
 845         }
 846
 847 progress:
 848         /* On error, the request should be considered as completed */
 849         if (hpk.hpk_errval > 0)
 850                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 851
 852         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 853                             &hpk, NULL);
 854
 855         /* Return first error */
 856         RETURN(rc != 0 ? rc : rc2);
 857 }
 858
 859 /**
 860  * Generic handler to do any post-copy work.
 861  *
 862  * It will send the last hsm_progress update to coordinator to inform it
 863  * that copy is finished and whether it was successful or not.
 864  *
 865  * Moreover,
 866  * - for ARCHIVE request, it will sample the file data version and compare it
 867  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 868  *   will be considered as failed.
 869  * - for RESTORE request, it will sample the file data version and send it to
 870  *   coordinator which is useful if the file was imported as 'released'.
 871  *
 872  * \return 0 on success.
 873  */
 874 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 875 {
 876         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 877         struct hsm_progress_kernel       hpk;
 878         int                              rc = 0;
 879         int                              rc2;
 880         ENTRY;
 881
 882         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 883         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 884          * initialized if copy_end was called with copy == NULL.
 885          */
 886
 887         /* Forge a hsm_progress based on data from copy. */
 888         hpk.hpk_fid = copy->hc_hai.hai_fid;
 889         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 890         hpk.hpk_extent = copy->hc_hai.hai_extent;
 891         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED |
 892                         HP_FLAG_COMPLETE_DELAYED;
 893         hpk.hpk_errval = copy->hc_errval;
 894         hpk.hpk_data_version = 0;
 895
 896         /* For archive request, we need to check the file data was not changed.
 897          *
 898          * For restore request, we need to send the file data version, this is
 899          * useful when the file was created using hsm_import.
 900          */
 901         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 902              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 903             (copy->hc_errval == 0)) {
 904                 struct inode    *inode;
 905                 __u64            data_version = 0;
 906
 907                 /* Get lsm for this fid */
 908                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 909                 if (IS_ERR(inode)) {
 910                         hpk.hpk_flags |= HP_FLAG_RETRY;
 911                         /* hpk_errval must be >= 0 */
 912                         hpk.hpk_errval = -PTR_ERR(inode);
 913                         GOTO(progress, rc = PTR_ERR(inode));
 914                 }
 915
 916                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 917                 iput(inode);
 918                 if (rc) {
 919                         CDEBUG(D_HSM, "Could not read file data version. "
 920                                       "Request could not be confirmed.\n");
 921                         if (hpk.hpk_errval == 0)
 922                                 hpk.hpk_errval = -rc;
 923                         GOTO(progress, rc);
 924                 }
 925
 926                 /* Store in the hsm_copy for later copytool use.
 927                  * Always modified even if no lsm. */
 928                 hpk.hpk_data_version = data_version;
 929
 930                 /* File could have been stripped during archiving, so we need
 931                  * to check anyway. */
 932                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 933                     (copy->hc_data_version != data_version)) {
 934                         CDEBUG(D_HSM, "File data version mismatched. "
 935                               "File content was changed during archiving. "
 936                                DFID", start:%#llx current:%#llx\n",
 937                                PFID(&copy->hc_hai.hai_fid),
 938                                copy->hc_data_version, data_version);
 939                         /* File was changed, send error to cdt. Do not ask for
 940                          * retry because if a file is modified frequently,
 941                          * the cdt will loop on retried archive requests.
 942                          * The policy engine will ask for a new archive later
 943                          * when the file will not be modified for some tunable
 944                          * time */
 945                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 946                         rc = -EBUSY;
 947                         /* hpk_errval must be >= 0 */
 948                         hpk.hpk_errval = -rc;
 949                         GOTO(progress, rc);
 950                 }
 951
 952         }
 953
 954 progress:
 955         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 956                             &hpk, NULL);
 957
 958         /* Return first error */
 959         RETURN(rc != 0 ? rc : rc2);
 960 }
 961
 962
 963 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 964                           const void __user *data, size_t size)
 965 {
 966         void *copy;
 967         int rc;
 968
 969         OBD_ALLOC(copy, size);
 970         if (copy == NULL)
 971                 return -ENOMEM;
 972
 973         if (copy_from_user(copy, data, size)) {
 974                 rc = -EFAULT;
 975                 goto out;
 976         }
 977
 978         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 979 out:
 980         OBD_FREE(copy, size);
 981
 982         return rc;
 983 }
 984
 985 static int check_owner(int type, int id)
 986 {
 987         switch (type) {
 988         case USRQUOTA:
 989                 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
 990                         return -EPERM;
 991                 break;
 992         case GRPQUOTA:
 993                 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
 994                         return -EPERM;
 995                 break;
 996         case PRJQUOTA:
 997                 break;
 998         }
 999         return 0;
1000 }
1001
1002 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1003 {
1004         int cmd = qctl->qc_cmd;
1005         int type = qctl->qc_type;
1006         int id = qctl->qc_id;
1007         int valid = qctl->qc_valid;
1008         int rc = 0;
1009         ENTRY;
1010
1011         switch (cmd) {
1012         case Q_SETQUOTA:
1013         case Q_SETINFO:
1014                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1015                         RETURN(-EPERM);
1016                 break;
1017         case Q_GETQUOTA:
1018                 if (check_owner(type, id) &&
1019                     (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1020                         RETURN(-EPERM);
1021                 break;
1022         case Q_GETINFO:
1023                 break;
1024         default:
1025                 CERROR("unsupported quotactl op: %#x\n", cmd);
1026                 RETURN(-ENOTTY);
1027         }
1028
1029         if (valid != QC_GENERAL) {
1030                 if (cmd == Q_GETINFO)
1031                         qctl->qc_cmd = Q_GETOINFO;
1032                 else if (cmd == Q_GETQUOTA)
1033                         qctl->qc_cmd = Q_GETOQUOTA;
1034                 else
1035                         RETURN(-EINVAL);
1036
1037                 switch (valid) {
1038                 case QC_MDTIDX:
1039                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1040                                            sizeof(*qctl), qctl, NULL);
1041                         break;
1042                 case QC_OSTIDX:
1043                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1044                                            sizeof(*qctl), qctl, NULL);
1045                         break;
1046                 case QC_UUID:
1047                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1048                                            sizeof(*qctl), qctl, NULL);
1049                         if (rc == -EAGAIN)
1050                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1051                                                    sbi->ll_dt_exp,
1052                                                    sizeof(*qctl), qctl, NULL);
1053                         break;
1054                 default:
1055                         rc = -EINVAL;
1056                         break;
1057                 }
1058
1059                 if (rc)
1060                         RETURN(rc);
1061
1062                 qctl->qc_cmd = cmd;
1063         } else {
1064                 struct obd_quotactl *oqctl;
1065
1066                 OBD_ALLOC_PTR(oqctl);
1067                 if (oqctl == NULL)
1068                         RETURN(-ENOMEM);
1069
1070                 QCTL_COPY(oqctl, qctl);
1071                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1072                 if (rc) {
1073                         OBD_FREE_PTR(oqctl);
1074                         RETURN(rc);
1075                 }
1076                 /* If QIF_SPACE is not set, client should collect the
1077                  * space usage from OSSs by itself */
1078                 if (cmd == Q_GETQUOTA &&
1079                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1080                     !oqctl->qc_dqblk.dqb_curspace) {
1081                         struct obd_quotactl *oqctl_tmp;
1082
1083                         OBD_ALLOC_PTR(oqctl_tmp);
1084                         if (oqctl_tmp == NULL)
1085                                 GOTO(out, rc = -ENOMEM);
1086
1087                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1088                         oqctl_tmp->qc_id = oqctl->qc_id;
1089                         oqctl_tmp->qc_type = oqctl->qc_type;
1090
1091                         /* collect space usage from OSTs */
1092                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1093                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1094                         if (!rc || rc == -EREMOTEIO) {
1095                                 oqctl->qc_dqblk.dqb_curspace =
1096                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1097                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1098                         }
1099
1100                         /* collect space & inode usage from MDTs */
1101                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1102                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1103                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1104                         if (!rc || rc == -EREMOTEIO) {
1105                                 oqctl->qc_dqblk.dqb_curspace +=
1106                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1107                                 oqctl->qc_dqblk.dqb_curinodes =
1108                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1109                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1110                         } else {
1111                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1112                         }
1113
1114                         OBD_FREE_PTR(oqctl_tmp);
1115                 }
1116 out:
1117                 QCTL_COPY(qctl, oqctl);
1118                 OBD_FREE_PTR(oqctl);
1119         }
1120
1121         RETURN(rc);
1122 }
1123
1124 /* This function tries to get a single name component,
1125  * to send to the server. No actual path traversal involved,
1126  * so we limit to NAME_MAX */
1127 static char *ll_getname(const char __user *filename)
1128 {
1129         int ret = 0, len;
1130         char *tmp;
1131
1132         OBD_ALLOC(tmp, NAME_MAX + 1);
1133
1134         if (!tmp)
1135                 return ERR_PTR(-ENOMEM);
1136
1137         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1138         if (len < 0)
1139                 ret = -ENOENT;
1140         else if (len > NAME_MAX)
1141                 ret = -ENAMETOOLONG;
1142
1143         if (ret) {
1144                 OBD_FREE(tmp, NAME_MAX + 1);
1145                 tmp =  ERR_PTR(ret);
1146         }
1147         return tmp;
1148 }
1149
1150 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1151
1152 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1153 {
1154         struct dentry *dentry = file_dentry(file);
1155         struct inode *inode = file_inode(file);
1156         struct ll_sb_info *sbi = ll_i2sbi(inode);
1157         struct obd_ioctl_data *data;
1158         int rc = 0;
1159         ENTRY;
1160
1161         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1162                PFID(ll_inode2fid(inode)), inode, cmd);
1163
1164         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1165         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1166                 return -ENOTTY;
1167
1168         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1169         switch(cmd) {
1170         case FSFILT_IOC_GETFLAGS:
1171         case FSFILT_IOC_SETFLAGS:
1172                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1173         case FSFILT_IOC_GETVERSION_OLD:
1174         case FSFILT_IOC_GETVERSION:
1175                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1176         /* We need to special case any other ioctls we want to handle,
1177          * to send them to the MDS/OST as appropriate and to properly
1178          * network encode the arg field.
1179         case FSFILT_IOC_SETVERSION_OLD:
1180         case FSFILT_IOC_SETVERSION:
1181         */
1182         case LL_IOC_GET_MDTIDX: {
1183                 int mdtidx;
1184
1185                 mdtidx = ll_get_mdt_idx(inode);
1186                 if (mdtidx < 0)
1187                         RETURN(mdtidx);
1188
1189                 if (put_user((int)mdtidx, (int __user *)arg))
1190                         RETURN(-EFAULT);
1191
1192                 return 0;
1193         }
1194         case IOC_MDC_LOOKUP: {
1195                 int namelen, len = 0;
1196                 char *buf = NULL;
1197                 char *filename;
1198
1199                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1200                 if (rc != 0)
1201                         RETURN(rc);
1202                 data = (void *)buf;
1203
1204                 filename = data->ioc_inlbuf1;
1205                 namelen = strlen(filename);
1206                 if (namelen < 1) {
1207                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1208                         GOTO(out_free, rc = -EINVAL);
1209                 }
1210
1211                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1212                 if (rc < 0) {
1213                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1214                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1215                                filename, rc);
1216                         GOTO(out_free, rc);
1217                 }
1218 out_free:
1219                 OBD_FREE_LARGE(buf, len);
1220                 return rc;
1221         }
1222         case LL_IOC_LMV_SETSTRIPE: {
1223                 struct lmv_user_md  *lum;
1224                 char            *buf = NULL;
1225                 char            *filename;
1226                 int              namelen = 0;
1227                 int              lumlen = 0;
1228                 umode_t          mode;
1229                 int              len;
1230                 int              rc;
1231
1232                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1233                 if (rc)
1234                         RETURN(rc);
1235
1236                 data = (void *)buf;
1237                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1238                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1239                         GOTO(lmv_out_free, rc = -EINVAL);
1240
1241                 filename = data->ioc_inlbuf1;
1242                 namelen = data->ioc_inllen1;
1243
1244                 if (namelen < 1) {
1245                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1246                         GOTO(lmv_out_free, rc = -EINVAL);
1247                 }
1248                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1249                 lumlen = data->ioc_inllen2;
1250
1251                 if (lum->lum_magic != LMV_USER_MAGIC ||
1252                     lumlen != sizeof(*lum)) {
1253                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1254                                filename, lum->lum_magic, lumlen, -EFAULT);
1255                         GOTO(lmv_out_free, rc = -EINVAL);
1256                 }
1257
1258 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1259                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1260 #else
1261                 mode = data->ioc_type;
1262 #endif
1263                 rc = ll_dir_setdirstripe(dentry, lum, filename, mode);
1264 lmv_out_free:
1265                 OBD_FREE_LARGE(buf, len);
1266                 RETURN(rc);
1267
1268         }
1269         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1270                 struct lmv_user_md        lum;
1271                 struct lmv_user_md __user *ulump =
1272                                         (struct lmv_user_md __user *)arg;
1273                 int                       rc;
1274
1275                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1276                         RETURN(-EFAULT);
1277
1278                 if (lum.lum_magic != LMV_USER_MAGIC)
1279                         RETURN(-EINVAL);
1280
1281                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1282
1283                 RETURN(rc);
1284         }
1285         case LL_IOC_LOV_SETSTRIPE_NEW:
1286         case LL_IOC_LOV_SETSTRIPE: {
1287                 struct lov_user_md_v3 lumv3;
1288                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1289                 struct lov_user_md_v1 __user *lumv1p =
1290                         (struct lov_user_md_v1 __user *)arg;
1291                 struct lov_user_md_v3 __user *lumv3p =
1292                         (struct lov_user_md_v3 __user *)arg;
1293
1294                 int set_default = 0;
1295
1296                 CLASSERT(sizeof(struct lov_user_md_v3) >
1297                          sizeof(struct lov_comp_md_v1));
1298                 CLASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1299                 CLASSERT(sizeof(lumv3.lmm_objects[0]) ==
1300                          sizeof(lumv3p->lmm_objects[0]));
1301                 /* first try with v1 which is smaller than v3 */
1302                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1303                         RETURN(-EFAULT);
1304
1305                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3)
1306                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1307                                 RETURN(-EFAULT);
1308
1309                 if (inode->i_sb->s_root == file_dentry(file))
1310                         set_default = 1;
1311
1312                 /* in v1 and v3 cases lumv1 points to data */
1313                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1314
1315                 RETURN(rc);
1316         }
1317         case LL_IOC_LMV_GETSTRIPE: {
1318                 struct lmv_user_md __user *ulmv =
1319                                         (struct lmv_user_md __user *)arg;
1320                 struct lmv_user_md      lum;
1321                 struct ptlrpc_request   *request = NULL;
1322                 union lmv_mds_md        *lmm = NULL;
1323                 int                     lmmsize;
1324                 u64                     valid = 0;
1325                 struct lmv_user_md      *tmp = NULL;
1326                 int                     mdt_index;
1327                 int                     lum_size;
1328                 int                     stripe_count;
1329                 int                     max_stripe_count;
1330                 int                     i;
1331                 int                     rc;
1332
1333                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1334                         RETURN(-EFAULT);
1335
1336                 max_stripe_count = lum.lum_stripe_count;
1337                 /* lum_magic will indicate which stripe the ioctl will like
1338                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1339                  * is for default LMV stripe */
1340                 if (lum.lum_magic == LMV_MAGIC_V1)
1341                         valid |= OBD_MD_MEA;
1342                 else if (lum.lum_magic == LMV_USER_MAGIC)
1343                         valid |= OBD_MD_DEFAULT_MEA;
1344                 else
1345                         RETURN(-EINVAL);
1346
1347                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1348                                       valid);
1349                 if (rc != 0)
1350                         GOTO(finish_req, rc);
1351
1352                 /* Get default LMV EA */
1353                 if (lum.lum_magic == LMV_USER_MAGIC) {
1354                         if (lmmsize > sizeof(*ulmv))
1355                                 GOTO(finish_req, rc = -EINVAL);
1356
1357                         if (copy_to_user(ulmv, lmm, lmmsize))
1358                                 GOTO(finish_req, rc = -EFAULT);
1359
1360                         GOTO(finish_req, rc);
1361                 }
1362
1363                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1364                 if (max_stripe_count < stripe_count) {
1365                         lum.lum_stripe_count = stripe_count;
1366                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1367                                 GOTO(finish_req, rc = -EFAULT);
1368                         GOTO(finish_req, rc = -E2BIG);
1369                 }
1370
1371                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1372                 OBD_ALLOC(tmp, lum_size);
1373                 if (tmp == NULL)
1374                         GOTO(finish_req, rc = -ENOMEM);
1375
1376                 mdt_index = ll_get_mdt_idx(inode);
1377                 if (mdt_index < 0)
1378                         GOTO(out_tmp, rc = -ENOMEM);
1379
1380                 tmp->lum_magic = LMV_MAGIC_V1;
1381                 tmp->lum_stripe_count = 0;
1382                 tmp->lum_stripe_offset = mdt_index;
1383                 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1384                 for (i = 0; i < stripe_count; i++) {
1385                         struct lu_fid   fid;
1386
1387                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1388                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1389                         if (mdt_index < 0)
1390                                 GOTO(out_tmp, rc = mdt_index);
1391
1392                         tmp->lum_objects[i].lum_mds = mdt_index;
1393                         tmp->lum_objects[i].lum_fid = fid;
1394                         tmp->lum_stripe_count++;
1395                 }
1396
1397                 if (copy_to_user(ulmv, tmp, lum_size))
1398                         GOTO(out_tmp, rc = -EFAULT);
1399 out_tmp:
1400                 OBD_FREE(tmp, lum_size);
1401 finish_req:
1402                 ptlrpc_req_finished(request);
1403                 return rc;
1404         }
1405
1406         case LL_IOC_REMOVE_ENTRY: {
1407                 char            *filename = NULL;
1408                 int              namelen = 0;
1409                 int              rc;
1410
1411                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1412                  * unsupported server, which might crash the server(LU-2730),
1413                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1414                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1415                  * server will support REINT_RMENTRY XXX*/
1416                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1417                         RETURN(-ENOTSUPP);
1418
1419                 filename = ll_getname((const char __user *)arg);
1420                 if (IS_ERR(filename))
1421                         RETURN(PTR_ERR(filename));
1422
1423                 namelen = strlen(filename);
1424                 if (namelen < 1)
1425                         GOTO(out_rmdir, rc = -EINVAL);
1426
1427                 rc = ll_rmdir_entry(inode, filename, namelen);
1428 out_rmdir:
1429                 if (filename)
1430                         ll_putname(filename);
1431                 RETURN(rc);
1432         }
1433         case LL_IOC_LOV_SWAP_LAYOUTS:
1434                 RETURN(-EPERM);
1435         case IOC_OBD_STATFS:
1436                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1437         case LL_IOC_LOV_GETSTRIPE:
1438         case LL_IOC_LOV_GETSTRIPE_NEW:
1439         case LL_IOC_MDC_GETINFO:
1440         case IOC_MDC_GETFILEINFO:
1441         case IOC_MDC_GETFILESTRIPE: {
1442                 struct ptlrpc_request *request = NULL;
1443                 struct lov_user_md __user *lump;
1444                 struct lov_mds_md *lmm = NULL;
1445                 struct mdt_body *body;
1446                 char *filename = NULL;
1447                 int lmmsize;
1448
1449                 if (cmd == IOC_MDC_GETFILEINFO ||
1450                     cmd == IOC_MDC_GETFILESTRIPE) {
1451                         filename = ll_getname((const char __user *)arg);
1452                         if (IS_ERR(filename))
1453                                 RETURN(PTR_ERR(filename));
1454
1455                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1456                                                       &lmmsize, &request);
1457                 } else {
1458                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1459                                               &request, 0);
1460                 }
1461
1462                 if (request) {
1463                         body = req_capsule_server_get(&request->rq_pill,
1464                                                       &RMF_MDT_BODY);
1465                         LASSERT(body != NULL);
1466                 } else {
1467                         GOTO(out_req, rc);
1468                 }
1469
1470                 if (rc < 0) {
1471                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1472                                                cmd == LL_IOC_MDC_GETINFO))
1473                                 GOTO(skip_lmm, rc = 0);
1474                         else
1475                                 GOTO(out_req, rc);
1476                 }
1477
1478                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1479                     cmd == LL_IOC_LOV_GETSTRIPE ||
1480                     cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1481                         lump = (struct lov_user_md __user *)arg;
1482                 } else {
1483                         struct lov_user_mds_data __user *lmdp;
1484                         lmdp = (struct lov_user_mds_data __user *)arg;
1485                         lump = &lmdp->lmd_lmm;
1486                 }
1487                 if (copy_to_user(lump, lmm, lmmsize)) {
1488                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1489                                 GOTO(out_req, rc = -EFAULT);
1490                         rc = -EOVERFLOW;
1491                 }
1492         skip_lmm:
1493                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1494                         struct lov_user_mds_data __user *lmdp;
1495                         lstat_t st = { 0 };
1496
1497                         st.st_dev       = inode->i_sb->s_dev;
1498                         st.st_mode      = body->mbo_mode;
1499                         st.st_nlink     = body->mbo_nlink;
1500                         st.st_uid       = body->mbo_uid;
1501                         st.st_gid       = body->mbo_gid;
1502                         st.st_rdev      = body->mbo_rdev;
1503                         st.st_size      = body->mbo_size;
1504                         st.st_blksize   = PAGE_SIZE;
1505                         st.st_blocks    = body->mbo_blocks;
1506                         st.st_atime     = body->mbo_atime;
1507                         st.st_mtime     = body->mbo_mtime;
1508                         st.st_ctime     = body->mbo_ctime;
1509                         st.st_ino       = cl_fid_build_ino(&body->mbo_fid1,
1510                                                 sbi->ll_flags &
1511                                                 LL_SBI_32BIT_API);
1512
1513                         lmdp = (struct lov_user_mds_data __user *)arg;
1514                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1515                                 GOTO(out_req, rc = -EFAULT);
1516                 }
1517
1518                 EXIT;
1519         out_req:
1520                 ptlrpc_req_finished(request);
1521                 if (filename)
1522                         ll_putname(filename);
1523                 return rc;
1524         }
1525         case OBD_IOC_QUOTACTL: {
1526                 struct if_quotactl *qctl;
1527
1528                 OBD_ALLOC_PTR(qctl);
1529                 if (!qctl)
1530                         RETURN(-ENOMEM);
1531
1532                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1533                         GOTO(out_quotactl, rc = -EFAULT);
1534
1535                 rc = quotactl_ioctl(sbi, qctl);
1536
1537                 if (rc == 0 &&
1538                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1539                         rc = -EFAULT;
1540
1541         out_quotactl:
1542                 OBD_FREE_PTR(qctl);
1543                 RETURN(rc);
1544         }
1545         case OBD_IOC_GETDTNAME:
1546         case OBD_IOC_GETMDNAME:
1547                 RETURN(ll_get_obd_name(inode, cmd, arg));
1548         case LL_IOC_FLUSHCTX:
1549                 RETURN(ll_flush_ctx(inode));
1550         case LL_IOC_GETOBDCOUNT: {
1551                 u32 count, vallen;
1552                 struct obd_export *exp;
1553
1554                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1555                         RETURN(-EFAULT);
1556
1557                 /* get ost count when count is zero, get mdt count otherwise */
1558                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1559                 vallen = sizeof(count);
1560                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1561                                   KEY_TGT_COUNT, &vallen, &count);
1562                 if (rc) {
1563                         CERROR("get target count failed: %d\n", rc);
1564                         RETURN(rc);
1565                 }
1566
1567                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1568                         RETURN(-EFAULT);
1569
1570                 RETURN(0);
1571         }
1572         case LL_IOC_PATH2FID:
1573                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1574                                      sizeof(struct lu_fid)))
1575                         RETURN(-EFAULT);
1576                 RETURN(0);
1577         case LL_IOC_GET_CONNECT_FLAGS: {
1578                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1579                                      (void __user *)arg));
1580         }
1581         case OBD_IOC_FID2PATH:
1582                 RETURN(ll_fid2path(inode, (void __user *)arg));
1583         case LL_IOC_GETPARENT:
1584                 RETURN(ll_getparent(file, (void __user *)arg));
1585         case LL_IOC_FID2MDTIDX: {
1586                 struct obd_export *exp = ll_i2mdexp(inode);
1587                 struct lu_fid     fid;
1588                 __u32             index;
1589
1590                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1591                                    sizeof(fid)))
1592                         RETURN(-EFAULT);
1593
1594                 /* Call mdc_iocontrol */
1595                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1596                                    (__u32 __user *)&index);
1597                 if (rc != 0)
1598                         RETURN(rc);
1599
1600                 RETURN(index);
1601         }
1602         case LL_IOC_HSM_REQUEST: {
1603                 struct hsm_user_request *hur;
1604                 ssize_t                  totalsize;
1605
1606                 OBD_ALLOC_PTR(hur);
1607                 if (hur == NULL)
1608                         RETURN(-ENOMEM);
1609
1610                 /* We don't know the true size yet; copy the fixed-size part */
1611                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1612                         OBD_FREE_PTR(hur);
1613                         RETURN(-EFAULT);
1614                 }
1615
1616                 /* Compute the whole struct size */
1617                 totalsize = hur_len(hur);
1618                 OBD_FREE_PTR(hur);
1619                 if (totalsize < 0)
1620                         RETURN(-E2BIG);
1621
1622                 /* Final size will be more than double totalsize */
1623                 if (totalsize >= MDS_MAXREQSIZE / 3)
1624                         RETURN(-E2BIG);
1625
1626                 OBD_ALLOC_LARGE(hur, totalsize);
1627                 if (hur == NULL)
1628                         RETURN(-ENOMEM);
1629
1630                 /* Copy the whole struct */
1631                 if (copy_from_user(hur, (void __user *)arg, totalsize))
1632                         GOTO(out_hur, rc = -EFAULT);
1633
1634                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1635                         const struct lu_fid *fid;
1636                         struct inode *f;
1637                         int i;
1638
1639                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1640                                 fid = &hur->hur_user_item[i].hui_fid;
1641                                 f = search_inode_for_lustre(inode->i_sb, fid);
1642                                 if (IS_ERR(f)) {
1643                                         rc = PTR_ERR(f);
1644                                         break;
1645                                 }
1646
1647                                 rc = ll_hsm_release(f);
1648                                 iput(f);
1649                                 if (rc != 0)
1650                                         break;
1651                         }
1652                 } else {
1653                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1654                                            hur, NULL);
1655                 }
1656
1657 out_hur:
1658                 OBD_FREE_LARGE(hur, totalsize);
1659
1660                 RETURN(rc);
1661         }
1662         case LL_IOC_HSM_PROGRESS: {
1663                 struct hsm_progress_kernel      hpk;
1664                 struct hsm_progress             hp;
1665
1666                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1667                         RETURN(-EFAULT);
1668
1669                 hpk.hpk_fid = hp.hp_fid;
1670                 hpk.hpk_cookie = hp.hp_cookie;
1671                 hpk.hpk_extent = hp.hp_extent;
1672                 hpk.hpk_flags = hp.hp_flags;
1673                 hpk.hpk_errval = hp.hp_errval;
1674                 hpk.hpk_data_version = 0;
1675
1676                 /* File may not exist in Lustre; all progress
1677                  * reported to Lustre root */
1678                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1679                                    NULL);
1680                 RETURN(rc);
1681         }
1682         case LL_IOC_HSM_CT_START:
1683                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1684                         RETURN(-EPERM);
1685
1686                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1687                                     sizeof(struct lustre_kernelcomm));
1688                 RETURN(rc);
1689
1690         case LL_IOC_HSM_COPY_START: {
1691                 struct hsm_copy *copy;
1692                 int              rc;
1693
1694                 OBD_ALLOC_PTR(copy);
1695                 if (copy == NULL)
1696                         RETURN(-ENOMEM);
1697                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1698                         OBD_FREE_PTR(copy);
1699                         RETURN(-EFAULT);
1700                 }
1701
1702                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1703                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1704                         rc = -EFAULT;
1705
1706                 OBD_FREE_PTR(copy);
1707                 RETURN(rc);
1708         }
1709         case LL_IOC_HSM_COPY_END: {
1710                 struct hsm_copy *copy;
1711                 int              rc;
1712
1713                 OBD_ALLOC_PTR(copy);
1714                 if (copy == NULL)
1715                         RETURN(-ENOMEM);
1716                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1717                         OBD_FREE_PTR(copy);
1718                         RETURN(-EFAULT);
1719                 }
1720
1721                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1722                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1723                         rc = -EFAULT;
1724
1725                 OBD_FREE_PTR(copy);
1726                 RETURN(rc);
1727         }
1728         case LL_IOC_MIGRATE: {
1729                 char            *buf = NULL;
1730                 const char      *filename;
1731                 int             namelen = 0;
1732                 int             len;
1733                 int             rc;
1734                 int             mdtidx;
1735
1736                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1737                 if (rc < 0)
1738                         RETURN(rc);
1739
1740                 data = (struct obd_ioctl_data *)buf;
1741                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1742                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1743                         GOTO(migrate_free, rc = -EINVAL);
1744
1745                 filename = data->ioc_inlbuf1;
1746                 namelen = data->ioc_inllen1;
1747                 /* \0 is packed at the end of filename */
1748                 if (namelen < 1 || namelen != strlen(filename) + 1)
1749                         GOTO(migrate_free, rc = -EINVAL);
1750
1751                 if (data->ioc_inllen2 != sizeof(mdtidx))
1752                         GOTO(migrate_free, rc = -EINVAL);
1753                 mdtidx = *(int *)data->ioc_inlbuf2;
1754
1755                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1756 migrate_free:
1757                 OBD_FREE_LARGE(buf, len);
1758
1759                 RETURN(rc);
1760         }
1761         case LL_IOC_FSGETXATTR:
1762                 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
1763         case LL_IOC_FSSETXATTR:
1764                 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
1765         default:
1766                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1767                                      (void __user *)arg));
1768         }
1769 }
1770
1771 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1772 {
1773         struct inode *inode = file->f_mapping->host;
1774         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1775         struct ll_sb_info *sbi = ll_i2sbi(inode);
1776         int api32 = ll_need_32bit_api(sbi);
1777         loff_t ret = -EINVAL;
1778         ENTRY;
1779
1780         inode_lock(inode);
1781         switch (origin) {
1782                 case SEEK_SET:
1783                         break;
1784                 case SEEK_CUR:
1785                         offset += file->f_pos;
1786                         break;
1787                 case SEEK_END:
1788                         if (offset > 0)
1789                                 GOTO(out, ret);
1790                         if (api32)
1791                                 offset += LL_DIR_END_OFF_32BIT;
1792                         else
1793                                 offset += LL_DIR_END_OFF;
1794                         break;
1795                 default:
1796                         GOTO(out, ret);
1797         }
1798
1799         if (offset >= 0 &&
1800             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1801              (!api32 && offset <= LL_DIR_END_OFF))) {
1802                 if (offset != file->f_pos) {
1803                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1804                             (!api32 && offset == LL_DIR_END_OFF))
1805                                 fd->lfd_pos = MDS_DIR_END_OFF;
1806                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1807                                 fd->lfd_pos = offset << 32;
1808                         else
1809                                 fd->lfd_pos = offset;
1810                         file->f_pos = offset;
1811                         file->f_version = 0;
1812                 }
1813                 ret = offset;
1814         }
1815         GOTO(out, ret);
1816
1817 out:
1818         inode_unlock(inode);
1819         return ret;
1820 }
1821
1822 static int ll_dir_open(struct inode *inode, struct file *file)
1823 {
1824         ENTRY;
1825         RETURN(ll_file_open(inode, file));
1826 }
1827
1828 static int ll_dir_release(struct inode *inode, struct file *file)
1829 {
1830         ENTRY;
1831         RETURN(ll_file_release(inode, file));
1832 }
1833
1834 const struct file_operations ll_dir_operations = {
1835         .llseek         = ll_dir_seek,
1836         .open           = ll_dir_open,
1837         .release        = ll_dir_release,
1838         .read           = generic_read_dir,
1839 #ifdef HAVE_DIR_CONTEXT
1840         .iterate        = ll_iterate,
1841 #else
1842         .readdir        = ll_readdir,
1843 #endif
1844         .unlocked_ioctl = ll_dir_ioctl,
1845         .fsync          = ll_fsync,
1846 };