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