lustre/llite/dir.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/dir.c
  37  *
  38  * Directory code for lustre client.
  39  */
  40
  41 #include <linux/fs.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <linux/version.h>
  45 #include <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 #include <obd_support.h>
  53 #include <obd_class.h>
  54 #include <lustre_lib.h>
  55 #include <lustre/lustre_idl.h>
  56 #include <lustre_lite.h>
  57 #include <lustre_dlm.h>
  58 #include <lustre_fid.h>
  59 #include "llite_internal.h"
  60
  61 /*
  62  * (new) readdir implementation overview.
  63  *
  64  * Original lustre readdir implementation cached exact copy of raw directory
  65  * pages on the client. These pages were indexed in client page cache by
  66  * logical offset in the directory file. This design, while very simple and
  67  * intuitive had some inherent problems:
  68  *
  69  *     . it implies that byte offset to the directory entry serves as a
  70  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  71  *     ext3/htree directory entries may move due to splits, and more
  72  *     importantly,
  73  *
  74  *     . it is incompatible with the design of split directories for cmd3,
  75  *     that assumes that names are distributed across nodes based on their
  76  *     hash, and so readdir should be done in hash order.
  77  *
  78  * New readdir implementation does readdir in hash order, and uses hash of a
  79  * file name as a telldir/seekdir cookie. This led to number of complications:
  80  *
  81  *     . hash is not unique, so it cannot be used to index cached directory
  82  *     pages on the client (note, that it requires a whole pageful of hash
  83  *     collided entries to cause two pages to have identical hashes);
  84  *
  85  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  86  *     entry cookie. ext3/htree has the same problem and lustre implementation
  87  *     mimics their solution: seekdir(hash) positions directory at the first
  88  *     entry with the given hash.
  89  *
  90  * Client side.
  91  *
  92  * 0. caching
  93  *
  94  * Client caches directory pages using hash of the first entry as an index. As
  95  * noted above hash is not unique, so this solution doesn't work as is:
  96  * special processing is needed for "page hash chains" (i.e., sequences of
  97  * pages filled with entries all having the same hash value).
  98  *
  99  * First, such chains have to be detected. To this end, server returns to the
 100  * client the hash of the first entry on the page next to one returned. When
 101  * client detects that this hash is the same as hash of the first entry on the
 102  * returned page, page hash collision has to be handled. Pages in the
 103  * hash chain, except first one, are termed "overflow pages".
 104  *
 105  * Solution to index uniqueness problem is to not cache overflow
 106  * pages. Instead, when page hash collision is detected, all overflow pages
 107  * from emerging chain are immediately requested from the server and placed in
 108  * a special data structure (struct ll_dir_chain). This data structure is used
 109  * by ll_readdir() to process entries from overflow pages. When readdir
 110  * invocation finishes, overflow pages are discarded. If page hash collision
 111  * chain weren't completely processed, next call to readdir will again detect
 112  * page hash collision, again read overflow pages in, process next portion of
 113  * entries and again discard the pages. This is not as wasteful as it looks,
 114  * because, given reasonable hash, page hash collisions are extremely rare.
 115  *
 116  * 1. directory positioning
 117  *
 118  * When seekdir(hash) is called, original
 119  *
 120  *
 121  *
 122  *
 123  *
 124  *
 125  *
 126  *
 127  * Server.
 128  *
 129  * identification of and access to overflow pages
 130  *
 131  * page format
 132  *
 133  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 134  * a header lu_dirpage which describes the start/end hash, and whether this
 135  * page is empty (contains no dir entry) or hash collide with next page.
 136  * After client receives reply, several pages will be integrated into dir page
 137  * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 138  * lu_dirpage for this integrated page will be adjusted. See
 139  * lmv_adjust_dirpages().
 140  *
 141  */
 142 /**
 143  * The following three APIs will be used by llite to iterate directory
 144  * entries from MDC dir page caches.
 145  *
 146  * ll_dir_entry_start(next) will lookup(return) entry by op_hash_offset.
 147  * To avoid extra memory allocation, the @entry will be pointed to
 148  * the dir entries in MDC page directly, so these pages can not be released
 149  * until the entry has been accessed in ll_readdir(or statahead).
 150  *
 151  * The iterate process will be
 152  *
 153  * ll_dir_entry_start: locate the page in MDC, and return the first entry.
 154  *                     hold the page.
 155  *
 156  * ll_dir_entry_next: return the next entry in the current page, if it reaches
 157  *                    to the end, release current page.
 158  *
 159  * ll_dir_entry_end: release the last page.
 160  **/
 161 struct lu_dirent *ll_dir_entry_start(struct inode *dir,
 162                                      struct md_op_data *op_data,
 163                                      struct page **ppage)
 164 {
 165         struct lu_dirent *entry;
 166         struct md_callback cb_op;
 167         int rc;
 168
 169         LASSERT(*ppage == NULL);
 170         cb_op.md_blocking_ast = ll_md_blocking_ast;
 171         rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
 172         if (rc != 0)
 173                 entry = ERR_PTR(rc);
 174         return entry;
 175 }
 176
 177 struct lu_dirent *ll_dir_entry_next(struct inode *dir,
 178                                     struct md_op_data *op_data,
 179                                     struct lu_dirent *ent,
 180                                     struct page **ppage)
 181 {
 182         struct lu_dirent *entry;
 183         struct md_callback cb_op;
 184         int rc;
 185
 186         LASSERT(*ppage != NULL);
 187         cb_op.md_blocking_ast = ll_md_blocking_ast;
 188         op_data->op_hash_offset = le64_to_cpu(ent->lde_hash);
 189         kunmap(*ppage);
 190         page_cache_release(*ppage);
 191         *ppage = NULL;
 192         rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
 193         if (rc != 0)
 194                 entry = ERR_PTR(rc);
 195         return entry;
 196 }
 197
 198 #ifdef HAVE_DIR_CONTEXT
 199 int ll_dir_read(struct inode *inode, struct md_op_data *op_data,
 200                 struct dir_context *ctx)
 201 {
 202 #else
 203 int ll_dir_read(struct inode *inode, struct md_op_data *op_data,
 204                 void *cookie, filldir_t filldir)
 205 {
 206 #endif
 207         struct ll_sb_info       *sbi = ll_i2sbi(inode);
 208         struct ll_dir_chain     chain;
 209         struct lu_dirent        *ent;
 210         int                     api32 = ll_need_32bit_api(sbi);
 211         int                     hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 212         int                     done = 0;
 213         int                     rc = 0;
 214         __u64                   hash = MDS_DIR_END_OFF;
 215         __u64                   last_hash = MDS_DIR_END_OFF;
 216         struct page             *page = NULL;
 217         ENTRY;
 218
 219         ll_dir_chain_init(&chain);
 220         for (ent = ll_dir_entry_start(inode, op_data, &page);
 221              ent != NULL && !IS_ERR(ent) && !done;
 222              ent = ll_dir_entry_next(inode, op_data, ent, &page)) {
 223                 __u16          type;
 224                 int            namelen;
 225                 struct lu_fid  fid;
 226                 __u64          lhash;
 227                 __u64          ino;
 228
 229                 hash = le64_to_cpu(ent->lde_hash);
 230                 if (hash < op_data->op_hash_offset)
 231                         /*
 232                          * Skip until we find target hash
 233                          * value.
 234                          */
 235                         continue;
 236                 namelen = le16_to_cpu(ent->lde_namelen);
 237                 if (namelen == 0)
 238                         /*
 239                          * Skip dummy record.
 240                          */
 241                         continue;
 242
 243                 if (api32 && hash64)
 244                         lhash = hash >> 32;
 245                 else
 246                         lhash = hash;
 247                 fid_le_to_cpu(&fid, &ent->lde_fid);
 248                 ino = cl_fid_build_ino(&fid, api32);
 249                 type = ll_dirent_type_get(ent);
 250
 251 #ifdef HAVE_DIR_CONTEXT
 252                 /* For 'll_nfs_get_name_filldir()', it will try
 253                  * to access the 'ent' through its 'lde_name',
 254                  * so the parameter 'name' for 'filldir()' must
 255                  * be part of the 'ent'. */
 256                 done = !dir_emit(ctx, ent->lde_name, namelen, ino, type);
 257 #else
 258                 done = filldir(cookie, ent->lde_name, namelen, lhash,
 259                                ino, type);
 260 #endif
 261                 if (done) {
 262                         if (op_data->op_hash_offset != MDS_DIR_END_OFF)
 263                                 op_data->op_hash_offset = last_hash;
 264                         break;
 265                 } else {
 266                         last_hash = hash;
 267                 }
 268         }
 269
 270         if (IS_ERR(ent))
 271                 rc = PTR_ERR(ent);
 272
 273         if (page != NULL) {
 274                 kunmap(page);
 275                 page_cache_release(page);
 276         }
 277
 278         ll_dir_chain_fini(&chain);
 279         RETURN(rc);
 280 }
 281
 282 #ifdef HAVE_DIR_CONTEXT
 283 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 284 #else
 285 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 286 #endif
 287 {
 288         struct inode            *inode  = filp->f_dentry->d_inode;
 289         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 290         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 291         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 292         int                     api32   = ll_need_32bit_api(sbi);
 293         struct md_op_data       *op_data;
 294         __u64                   pos;
 295         int                     rc;
 296         ENTRY;
 297
 298         if (lfd != NULL)
 299                 pos = lfd->lfd_pos;
 300         else
 301                 pos = 0;
 302
 303         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 304                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 305                inode, (unsigned long)pos, i_size_read(inode), api32);
 306
 307         if (pos == MDS_DIR_END_OFF)
 308                 /*
 309                  * end-of-file.
 310                  */
 311                 GOTO(out, rc = 0);
 312
 313         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 314                                      LUSTRE_OPC_ANY, inode);
 315         if (IS_ERR(op_data))
 316                 GOTO(out, rc = PTR_ERR(op_data));
 317
 318         op_data->op_hash_offset = pos;
 319         op_data->op_max_pages = sbi->ll_md_brw_size >> PAGE_CACHE_SHIFT;
 320 #ifdef HAVE_DIR_CONTEXT
 321         ctx->pos = pos;
 322         rc = ll_dir_read(inode, op_data, ctx);
 323         pos = ctx->pos;
 324 #else
 325         rc = ll_dir_read(inode, op_data, cookie, filldir);
 326 #endif
 327         if (lfd != NULL)
 328                 lfd->lfd_pos = op_data->op_hash_offset;
 329
 330         if (pos == MDS_DIR_END_OFF) {
 331                 if (api32)
 332                         pos = LL_DIR_END_OFF_32BIT;
 333                 else
 334                         pos = LL_DIR_END_OFF;
 335         } else {
 336                 if (api32 && hash64)
 337                         pos = op_data->op_hash_offset >> 32;
 338                 else
 339                         pos = op_data->op_hash_offset;
 340         }
 341 #ifdef HAVE_DIR_CONTEXT
 342         ctx->pos = pos;
 343 #else
 344         filp->f_pos = pos;
 345 #endif
 346         ll_finish_md_op_data(op_data);
 347         filp->f_version = inode->i_version;
 348
 349 out:
 350         if (!rc)
 351                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 352
 353         RETURN(rc);
 354 }
 355
 356 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 357 {
 358         struct mgs_send_param *msp;
 359         int rc = 0;
 360
 361         OBD_ALLOC_PTR(msp);
 362         if (!msp)
 363                 return -ENOMEM;
 364
 365         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 366         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 367                                 sizeof(struct mgs_send_param), msp, NULL);
 368         if (rc)
 369                 CERROR("Failed to set parameter: %d\n", rc);
 370         OBD_FREE_PTR(msp);
 371
 372         return rc;
 373 }
 374
 375 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 376                                const char *filename)
 377 {
 378         struct ptlrpc_request *request = NULL;
 379         struct md_op_data *op_data;
 380         struct ll_sb_info *sbi = ll_i2sbi(dir);
 381         int mode;
 382         int err;
 383         ENTRY;
 384
 385         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 386                 RETURN(-EINVAL);
 387
 388         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
 389                "stripe_offset %d, stripe_count: %u\n",
 390                PFID(ll_inode2fid(dir)), dir, filename,
 391                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 392
 393         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 394                 lustre_swab_lmv_user_md(lump);
 395
 396         mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
 397         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 398                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 399                                      lump);
 400         if (IS_ERR(op_data))
 401                 GOTO(err_exit, err = PTR_ERR(op_data));
 402
 403         op_data->op_cli_flags |= CLI_SET_MEA;
 404         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 405                         current_fsuid(), current_fsgid(),
 406                         cfs_curproc_cap_pack(), 0, &request);
 407         ll_finish_md_op_data(op_data);
 408         if (err)
 409                 GOTO(err_exit, err);
 410 err_exit:
 411         ptlrpc_req_finished(request);
 412         return err;
 413 }
 414
 415 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 416                      int set_default)
 417 {
 418         struct ll_sb_info *sbi = ll_i2sbi(inode);
 419         struct md_op_data *op_data;
 420         struct ptlrpc_request *req = NULL;
 421         int rc = 0;
 422         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 423         struct obd_device *mgc = lsi->lsi_mgc;
 424         int lum_size;
 425         ENTRY;
 426
 427         if (lump != NULL) {
 428                 /*
 429                  * This is coming from userspace, so should be in
 430                  * local endian.  But the MDS would like it in little
 431                  * endian, so we swab it before we send it.
 432                  */
 433                 switch (lump->lmm_magic) {
 434                 case LOV_USER_MAGIC_V1: {
 435                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 436                                 lustre_swab_lov_user_md_v1(lump);
 437                         lum_size = sizeof(struct lov_user_md_v1);
 438                         break;
 439                 }
 440                 case LOV_USER_MAGIC_V3: {
 441                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 442                                 lustre_swab_lov_user_md_v3(
 443                                         (struct lov_user_md_v3 *)lump);
 444                         lum_size = sizeof(struct lov_user_md_v3);
 445                         break;
 446                 }
 447                 case LMV_USER_MAGIC: {
 448                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 449                                 lustre_swab_lmv_user_md(
 450                                         (struct lmv_user_md *)lump);
 451                         lum_size = sizeof(struct lmv_user_md);
 452                         break;
 453                 }
 454                 default: {
 455                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 456                                         " %#08x != %#08x nor %#08x\n",
 457                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 458                                         LOV_USER_MAGIC_V3);
 459                         RETURN(-EINVAL);
 460                 }
 461                 }
 462         } else {
 463                 lum_size = sizeof(struct lov_user_md_v1);
 464         }
 465
 466         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 467                                      LUSTRE_OPC_ANY, NULL);
 468         if (IS_ERR(op_data))
 469                 RETURN(PTR_ERR(op_data));
 470
 471         /* swabbing is done in lov_setstripe() on server side */
 472         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 473                         NULL, 0, &req, NULL);
 474         ll_finish_md_op_data(op_data);
 475         ptlrpc_req_finished(req);
 476         if (rc) {
 477                 if (rc != -EPERM && rc != -EACCES)
 478                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 479         }
 480
 481         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 482          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 483          need the make the distiction between the 2 versions */
 484         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 485                 char *param = NULL;
 486                 char *buf;
 487
 488                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 489                 if (param == NULL)
 490                         GOTO(end, rc = -ENOMEM);
 491
 492                 buf = param;
 493                 /* Get fsname and assume devname to be -MDT0000. */
 494                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 495                 strcat(buf, "-MDT0000.lov");
 496                 buf += strlen(buf);
 497
 498                 /* Set root stripesize */
 499                 sprintf(buf, ".stripesize=%u",
 500                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 501                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 502                 if (rc)
 503                         GOTO(end, rc);
 504
 505                 /* Set root stripecount */
 506                 sprintf(buf, ".stripecount=%hd",
 507                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 508                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 509                 if (rc)
 510                         GOTO(end, rc);
 511
 512                 /* Set root stripeoffset */
 513                 sprintf(buf, ".stripeoffset=%hd",
 514                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 515                         (typeof(lump->lmm_stripe_offset))(-1));
 516                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 517
 518 end:
 519                 if (param != NULL)
 520                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 521         }
 522         RETURN(rc);
 523 }
 524
 525 /**
 526  * This function will be used to get default LOV/LMV/Default LMV
 527  * @valid will be used to indicate which stripe it will retrieve
 528  *      OBD_MD_MEA              LMV stripe EA
 529  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 530  *      otherwise               Default LOV EA.
 531  * Each time, it can only retrieve 1 stripe EA
 532  **/
 533 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 534                      struct ptlrpc_request **request, obd_valid valid)
 535 {
 536         struct ll_sb_info *sbi = ll_i2sbi(inode);
 537         struct mdt_body   *body;
 538         struct lov_mds_md *lmm = NULL;
 539         struct ptlrpc_request *req = NULL;
 540         int rc, lmm_size;
 541         struct md_op_data *op_data;
 542         ENTRY;
 543
 544         rc = ll_get_max_mdsize(sbi, &lmm_size);
 545         if (rc)
 546                 RETURN(rc);
 547
 548         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 549                                      0, lmm_size, LUSTRE_OPC_ANY,
 550                                      NULL);
 551         if (IS_ERR(op_data))
 552                 RETURN(PTR_ERR(op_data));
 553
 554         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 555         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 556         ll_finish_md_op_data(op_data);
 557         if (rc < 0) {
 558                 CDEBUG(D_INFO, "md_getattr failed on inode "
 559                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 560                 GOTO(out, rc);
 561         }
 562
 563         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 564         LASSERT(body != NULL);
 565
 566         lmm_size = body->eadatasize;
 567
 568         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 569             lmm_size == 0) {
 570                 GOTO(out, rc = -ENODATA);
 571         }
 572
 573         lmm = req_capsule_server_sized_get(&req->rq_pill,
 574                                            &RMF_MDT_MD, lmm_size);
 575         LASSERT(lmm != NULL);
 576
 577         /*
 578          * This is coming from the MDS, so is probably in
 579          * little endian.  We convert it to host endian before
 580          * passing it to userspace.
 581          */
 582         /* We don't swab objects for directories */
 583         switch (le32_to_cpu(lmm->lmm_magic)) {
 584         case LOV_MAGIC_V1:
 585                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 586                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 587                 break;
 588         case LOV_MAGIC_V3:
 589                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 590                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 591                 break;
 592         case LMV_MAGIC:
 593         case LMV_MAGIC_MIGRATE:
 594                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 595                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 596                 break;
 597         case LMV_USER_MAGIC:
 598                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 599                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 600                 break;
 601         default:
 602                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 603                 rc = -EPROTO;
 604         }
 605 out:
 606         *plmm = lmm;
 607         *plmm_size = lmm_size;
 608         *request = req;
 609         return rc;
 610 }
 611
 612 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 613 {
 614         struct md_op_data       *op_data;
 615         int                     rc;
 616         int                     mdt_index;
 617         ENTRY;
 618
 619         OBD_ALLOC_PTR(op_data);
 620         if (op_data == NULL)
 621                 RETURN(-ENOMEM);
 622
 623         op_data->op_flags |= MF_GET_MDT_IDX;
 624         op_data->op_fid1 = *fid;
 625         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 626         mdt_index = op_data->op_mds;
 627         OBD_FREE_PTR(op_data);
 628         if (rc < 0)
 629                 RETURN(rc);
 630
 631         RETURN(mdt_index);
 632 }
 633
 634 /*
 635  *  Get MDT index for the inode.
 636  */
 637 int ll_get_mdt_idx(struct inode *inode)
 638 {
 639         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 640 }
 641
 642 /**
 643  * Generic handler to do any pre-copy work.
 644  *
 645  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 646  * first information for it that real work has started.
 647  *
 648  * Moreover, for a ARCHIVE request, it will sample the file data version and
 649  * store it in \a copy.
 650  *
 651  * \return 0 on success.
 652  */
 653 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 654 {
 655         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 656         struct hsm_progress_kernel       hpk;
 657         int                              rc;
 658         ENTRY;
 659
 660         /* Forge a hsm_progress based on data from copy. */
 661         hpk.hpk_fid = copy->hc_hai.hai_fid;
 662         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 663         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 664         hpk.hpk_extent.length = 0;
 665         hpk.hpk_flags = 0;
 666         hpk.hpk_errval = 0;
 667         hpk.hpk_data_version = 0;
 668
 669
 670         /* For archive request, we need to read the current file version. */
 671         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 672                 struct inode    *inode;
 673                 __u64            data_version = 0;
 674
 675                 /* Get inode for this fid */
 676                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 677                 if (IS_ERR(inode)) {
 678                         hpk.hpk_flags |= HP_FLAG_RETRY;
 679                         /* hpk_errval is >= 0 */
 680                         hpk.hpk_errval = -PTR_ERR(inode);
 681                         GOTO(progress, rc = PTR_ERR(inode));
 682                 }
 683
 684                 /* Read current file data version */
 685                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 686                 iput(inode);
 687                 if (rc != 0) {
 688                         CDEBUG(D_HSM, "Could not read file data version of "
 689                                       DFID" (rc = %d). Archive request ("
 690                                       LPX64") could not be done.\n",
 691                                       PFID(&copy->hc_hai.hai_fid), rc,
 692                                       copy->hc_hai.hai_cookie);
 693                         hpk.hpk_flags |= HP_FLAG_RETRY;
 694                         /* hpk_errval must be >= 0 */
 695                         hpk.hpk_errval = -rc;
 696                         GOTO(progress, rc);
 697                 }
 698
 699                 /* Store it the hsm_copy for later copytool use.
 700                  * Always modified even if no lsm. */
 701                 copy->hc_data_version = data_version;
 702         }
 703
 704 progress:
 705         /* On error, the request should be considered as completed */
 706         if (hpk.hpk_errval > 0)
 707                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 708         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 709                            &hpk, NULL);
 710
 711         RETURN(rc);
 712 }
 713
 714 /**
 715  * Generic handler to do any post-copy work.
 716  *
 717  * It will send the last hsm_progress update to coordinator to inform it
 718  * that copy is finished and whether it was successful or not.
 719  *
 720  * Moreover,
 721  * - for ARCHIVE request, it will sample the file data version and compare it
 722  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 723  *   will be considered as failed.
 724  * - for RESTORE request, it will sample the file data version and send it to
 725  *   coordinator which is useful if the file was imported as 'released'.
 726  *
 727  * \return 0 on success.
 728  */
 729 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 730 {
 731         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 732         struct hsm_progress_kernel       hpk;
 733         int                              rc;
 734         ENTRY;
 735
 736         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 737         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 738          * initialized if copy_end was called with copy == NULL.
 739          */
 740
 741         /* Forge a hsm_progress based on data from copy. */
 742         hpk.hpk_fid = copy->hc_hai.hai_fid;
 743         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 744         hpk.hpk_extent = copy->hc_hai.hai_extent;
 745         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 746         hpk.hpk_errval = copy->hc_errval;
 747         hpk.hpk_data_version = 0;
 748
 749         /* For archive request, we need to check the file data was not changed.
 750          *
 751          * For restore request, we need to send the file data version, this is
 752          * useful when the file was created using hsm_import.
 753          */
 754         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 755              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 756             (copy->hc_errval == 0)) {
 757                 struct inode    *inode;
 758                 __u64            data_version = 0;
 759
 760                 /* Get lsm for this fid */
 761                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 762                 if (IS_ERR(inode)) {
 763                         hpk.hpk_flags |= HP_FLAG_RETRY;
 764                         /* hpk_errval must be >= 0 */
 765                         hpk.hpk_errval = -PTR_ERR(inode);
 766                         GOTO(progress, rc = PTR_ERR(inode));
 767                 }
 768
 769                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 770                 iput(inode);
 771                 if (rc) {
 772                         CDEBUG(D_HSM, "Could not read file data version. "
 773                                       "Request could not be confirmed.\n");
 774                         if (hpk.hpk_errval == 0)
 775                                 hpk.hpk_errval = -rc;
 776                         GOTO(progress, rc);
 777                 }
 778
 779                 /* Store it the hsm_copy for later copytool use.
 780                  * Always modified even if no lsm. */
 781                 hpk.hpk_data_version = data_version;
 782
 783                 /* File could have been stripped during archiving, so we need
 784                  * to check anyway. */
 785                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 786                     (copy->hc_data_version != data_version)) {
 787                         CDEBUG(D_HSM, "File data version mismatched. "
 788                               "File content was changed during archiving. "
 789                                DFID", start:"LPX64" current:"LPX64"\n",
 790                                PFID(&copy->hc_hai.hai_fid),
 791                                copy->hc_data_version, data_version);
 792                         /* File was changed, send error to cdt. Do not ask for
 793                          * retry because if a file is modified frequently,
 794                          * the cdt will loop on retried archive requests.
 795                          * The policy engine will ask for a new archive later
 796                          * when the file will not be modified for some tunable
 797                          * time */
 798                         /* we do not notify caller */
 799                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 800                         /* hpk_errval must be >= 0 */
 801                         hpk.hpk_errval = EBUSY;
 802                 }
 803
 804         }
 805
 806 progress:
 807         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 808                            &hpk, NULL);
 809
 810         RETURN(rc);
 811 }
 812
 813
 814 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 815                           const void __user *data, size_t size)
 816 {
 817         void *copy;
 818         int rc;
 819
 820         OBD_ALLOC(copy, size);
 821         if (copy == NULL)
 822                 return -ENOMEM;
 823
 824         if (copy_from_user(copy, data, size)) {
 825                 rc = -EFAULT;
 826                 goto out;
 827         }
 828
 829         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 830 out:
 831         OBD_FREE(copy, size);
 832
 833         return rc;
 834 }
 835
 836 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 837 {
 838         int cmd = qctl->qc_cmd;
 839         int type = qctl->qc_type;
 840         int id = qctl->qc_id;
 841         int valid = qctl->qc_valid;
 842         int rc = 0;
 843         ENTRY;
 844
 845         switch (cmd) {
 846         case LUSTRE_Q_INVALIDATE:
 847         case LUSTRE_Q_FINVALIDATE:
 848         case Q_QUOTAON:
 849         case Q_QUOTAOFF:
 850         case Q_SETQUOTA:
 851         case Q_SETINFO:
 852                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 853                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 854                         RETURN(-EPERM);
 855                 break;
 856         case Q_GETQUOTA:
 857                 if (((type == USRQUOTA && current_euid() != id) ||
 858                      (type == GRPQUOTA && !in_egroup_p(id))) &&
 859                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 860                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 861                         RETURN(-EPERM);
 862                 break;
 863         case Q_GETINFO:
 864                 break;
 865         default:
 866                 CERROR("unsupported quotactl op: %#x\n", cmd);
 867                 RETURN(-ENOTTY);
 868         }
 869
 870         if (valid != QC_GENERAL) {
 871                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 872                         RETURN(-EOPNOTSUPP);
 873
 874                 if (cmd == Q_GETINFO)
 875                         qctl->qc_cmd = Q_GETOINFO;
 876                 else if (cmd == Q_GETQUOTA)
 877                         qctl->qc_cmd = Q_GETOQUOTA;
 878                 else
 879                         RETURN(-EINVAL);
 880
 881                 switch (valid) {
 882                 case QC_MDTIDX:
 883                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 884                                            sizeof(*qctl), qctl, NULL);
 885                         break;
 886                 case QC_OSTIDX:
 887                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 888                                            sizeof(*qctl), qctl, NULL);
 889                         break;
 890                 case QC_UUID:
 891                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 892                                            sizeof(*qctl), qctl, NULL);
 893                         if (rc == -EAGAIN)
 894                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 895                                                    sbi->ll_dt_exp,
 896                                                    sizeof(*qctl), qctl, NULL);
 897                         break;
 898                 default:
 899                         rc = -EINVAL;
 900                         break;
 901                 }
 902
 903                 if (rc)
 904                         RETURN(rc);
 905
 906                 qctl->qc_cmd = cmd;
 907         } else {
 908                 struct obd_quotactl *oqctl;
 909
 910                 OBD_ALLOC_PTR(oqctl);
 911                 if (oqctl == NULL)
 912                         RETURN(-ENOMEM);
 913
 914                 QCTL_COPY(oqctl, qctl);
 915                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
 916                 if (rc) {
 917                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
 918                                 oqctl->qc_cmd = Q_QUOTAOFF;
 919                                 obd_quotactl(sbi->ll_md_exp, oqctl);
 920                         }
 921                         OBD_FREE_PTR(oqctl);
 922                         RETURN(rc);
 923                 }
 924                 /* If QIF_SPACE is not set, client should collect the
 925                  * space usage from OSSs by itself */
 926                 if (cmd == Q_GETQUOTA &&
 927                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
 928                     !oqctl->qc_dqblk.dqb_curspace) {
 929                         struct obd_quotactl *oqctl_tmp;
 930
 931                         OBD_ALLOC_PTR(oqctl_tmp);
 932                         if (oqctl_tmp == NULL)
 933                                 GOTO(out, rc = -ENOMEM);
 934
 935                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
 936                         oqctl_tmp->qc_id = oqctl->qc_id;
 937                         oqctl_tmp->qc_type = oqctl->qc_type;
 938
 939                         /* collect space usage from OSTs */
 940                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 941                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
 942                         if (!rc || rc == -EREMOTEIO) {
 943                                 oqctl->qc_dqblk.dqb_curspace =
 944                                         oqctl_tmp->qc_dqblk.dqb_curspace;
 945                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
 946                         }
 947
 948                         /* collect space & inode usage from MDTs */
 949                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 950                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
 951                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
 952                         if (!rc || rc == -EREMOTEIO) {
 953                                 oqctl->qc_dqblk.dqb_curspace +=
 954                                         oqctl_tmp->qc_dqblk.dqb_curspace;
 955                                 oqctl->qc_dqblk.dqb_curinodes =
 956                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
 957                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
 958                         } else {
 959                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
 960                         }
 961
 962                         OBD_FREE_PTR(oqctl_tmp);
 963                 }
 964 out:
 965                 QCTL_COPY(qctl, oqctl);
 966                 OBD_FREE_PTR(oqctl);
 967         }
 968
 969         RETURN(rc);
 970 }
 971
 972 static char *
 973 ll_getname(const char __user *filename)
 974 {
 975         int ret = 0, len;
 976         char *tmp = __getname();
 977
 978         if (!tmp)
 979                 return ERR_PTR(-ENOMEM);
 980
 981         len = strncpy_from_user(tmp, filename, PATH_MAX);
 982         if (len == 0)
 983                 ret = -ENOENT;
 984         else if (len > PATH_MAX)
 985                 ret = -ENAMETOOLONG;
 986
 987         if (ret) {
 988                 __putname(tmp);
 989                 tmp =  ERR_PTR(ret);
 990         }
 991         return tmp;
 992 }
 993
 994 #define ll_putname(filename) __putname(filename)
 995
 996 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 997 {
 998         struct inode *inode = file->f_dentry->d_inode;
 999         struct ll_sb_info *sbi = ll_i2sbi(inode);
1000         struct obd_ioctl_data *data;
1001         int rc = 0;
1002         ENTRY;
1003
1004         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1005                PFID(ll_inode2fid(inode)), inode, cmd);
1006
1007         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1008         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1009                 return -ENOTTY;
1010
1011         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1012         switch(cmd) {
1013         case FSFILT_IOC_GETFLAGS:
1014         case FSFILT_IOC_SETFLAGS:
1015                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1016         case FSFILT_IOC_GETVERSION_OLD:
1017         case FSFILT_IOC_GETVERSION:
1018                 RETURN(put_user(inode->i_generation, (int *)arg));
1019         /* We need to special case any other ioctls we want to handle,
1020          * to send them to the MDS/OST as appropriate and to properly
1021          * network encode the arg field.
1022         case FSFILT_IOC_SETVERSION_OLD:
1023         case FSFILT_IOC_SETVERSION:
1024         */
1025         case LL_IOC_GET_MDTIDX: {
1026                 int mdtidx;
1027
1028                 mdtidx = ll_get_mdt_idx(inode);
1029                 if (mdtidx < 0)
1030                         RETURN(mdtidx);
1031
1032                 if (put_user((int)mdtidx, (int*)arg))
1033                         RETURN(-EFAULT);
1034
1035                 return 0;
1036         }
1037         case IOC_MDC_LOOKUP: {
1038                 struct ptlrpc_request *request = NULL;
1039                 int namelen, len = 0;
1040                 char *buf = NULL;
1041                 char *filename;
1042                 struct md_op_data *op_data;
1043
1044                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1045                 if (rc)
1046                         RETURN(rc);
1047                 data = (void *)buf;
1048
1049                 filename = data->ioc_inlbuf1;
1050                 namelen = strlen(filename);
1051
1052                 if (namelen < 1) {
1053                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1054                         GOTO(out_free, rc = -EINVAL);
1055                 }
1056
1057                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1058                                              0, LUSTRE_OPC_ANY, NULL);
1059                 if (IS_ERR(op_data))
1060                         GOTO(out_free, rc = PTR_ERR(op_data));
1061
1062                 op_data->op_valid = OBD_MD_FLID;
1063                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1064                 ll_finish_md_op_data(op_data);
1065                 if (rc < 0) {
1066                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1067                         GOTO(out_free, rc);
1068                 }
1069                 ptlrpc_req_finished(request);
1070                 EXIT;
1071 out_free:
1072                 obd_ioctl_freedata(buf, len);
1073                 return rc;
1074         }
1075         case LL_IOC_LMV_SETSTRIPE: {
1076                 struct lmv_user_md  *lum;
1077                 char            *buf = NULL;
1078                 char            *filename;
1079                 int              namelen = 0;
1080                 int              lumlen = 0;
1081                 int              len;
1082                 int              rc;
1083
1084                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1085                 if (rc)
1086                         RETURN(rc);
1087
1088                 data = (void *)buf;
1089                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1090                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1091                         GOTO(lmv_out_free, rc = -EINVAL);
1092
1093                 filename = data->ioc_inlbuf1;
1094                 namelen = data->ioc_inllen1;
1095
1096                 if (namelen < 1) {
1097                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1098                         GOTO(lmv_out_free, rc = -EINVAL);
1099                 }
1100                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1101                 lumlen = data->ioc_inllen2;
1102
1103                 if (lum->lum_magic != LMV_USER_MAGIC ||
1104                     lumlen != sizeof(*lum)) {
1105                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1106                                filename, lum->lum_magic, lumlen, -EFAULT);
1107                         GOTO(lmv_out_free, rc = -EINVAL);
1108                 }
1109
1110                 /**
1111                  * ll_dir_setdirstripe will be used to set dir stripe
1112                  *  mdc_create--->mdt_reint_create (with dirstripe)
1113                  */
1114                 rc = ll_dir_setdirstripe(inode, lum, filename);
1115 lmv_out_free:
1116                 obd_ioctl_freedata(buf, len);
1117                 RETURN(rc);
1118
1119         }
1120         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1121                 struct lmv_user_md        lum;
1122                 struct lmv_user_md __user *ulump =
1123                                         (struct lmv_user_md __user *)arg;
1124                 int                       rc;
1125
1126                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1127                         RETURN(-EFAULT);
1128
1129                 if (lum.lum_magic != LMV_USER_MAGIC)
1130                         RETURN(-EINVAL);
1131
1132                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1133
1134                 RETURN(rc);
1135         }
1136         case LL_IOC_LOV_SETSTRIPE: {
1137                 struct lov_user_md_v3 lumv3;
1138                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1139                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1140                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1141
1142                 int set_default = 0;
1143
1144                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1145                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1146                         sizeof(lumv3p->lmm_objects[0]));
1147                 /* first try with v1 which is smaller than v3 */
1148                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1149                         RETURN(-EFAULT);
1150
1151                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1152                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1153                                 RETURN(-EFAULT);
1154                 }
1155
1156                 if (inode->i_sb->s_root == file->f_dentry)
1157                         set_default = 1;
1158
1159                 /* in v1 and v3 cases lumv1 points to data */
1160                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1161
1162                 RETURN(rc);
1163         }
1164         case LL_IOC_LMV_GETSTRIPE: {
1165                 struct lmv_user_md __user *ulmv =
1166                                         (struct lmv_user_md __user *)arg;
1167                 struct lmv_user_md      lum;
1168                 struct ptlrpc_request   *request = NULL;
1169                 union lmv_mds_md        *lmm = NULL;
1170                 int                     lmmsize;
1171                 obd_valid               valid = 0;
1172                 struct lmv_user_md      *tmp = NULL;
1173                 int                     mdt_index;
1174                 int                     lum_size;
1175                 int                     stripe_count;
1176                 int                     i;
1177                 int                     rc;
1178
1179                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1180                         RETURN(-EFAULT);
1181
1182                 /* lum_magic will indicate which stripe the ioctl will like
1183                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1184                  * is for default LMV stripe */
1185                 if (lum.lum_magic == LMV_MAGIC_V1)
1186                         valid |= OBD_MD_MEA;
1187                 else if (lum.lum_magic == LMV_USER_MAGIC)
1188                         valid |= OBD_MD_DEFAULT_MEA;
1189                 else
1190                         RETURN(-EINVAL);
1191
1192                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1193                                       valid);
1194                 if (rc != 0 && rc != -ENODATA)
1195                         GOTO(finish_req, rc);
1196
1197                 /* Get default LMV EA */
1198                 if (lum.lum_magic == LMV_USER_MAGIC) {
1199                         if (rc != 0)
1200                                 GOTO(finish_req, rc);
1201
1202                         if (lmmsize > sizeof(*ulmv))
1203                                 GOTO(finish_req, rc = -EINVAL);
1204
1205                         if (copy_to_user(ulmv, lmm, lmmsize))
1206                                 GOTO(finish_req, rc = -EFAULT);
1207
1208                         GOTO(finish_req, rc);
1209                 }
1210
1211                 /* Get normal LMV EA */
1212                 if (rc == -ENODATA) {
1213                         stripe_count = 1;
1214                 } else {
1215                         LASSERT(lmm != NULL);
1216                         stripe_count = lmv_mds_md_stripe_count_get(lmm);
1217                 }
1218
1219                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1220                 OBD_ALLOC(tmp, lum_size);
1221                 if (tmp == NULL)
1222                         GOTO(finish_req, rc = -ENOMEM);
1223
1224                 tmp->lum_magic = LMV_MAGIC_V1;
1225                 tmp->lum_stripe_count = 1;
1226                 mdt_index = ll_get_mdt_idx(inode);
1227                 if (mdt_index < 0)
1228                         GOTO(out_tmp, rc = -ENOMEM);
1229                 tmp->lum_stripe_offset = mdt_index;
1230                 tmp->lum_objects[0].lum_mds = mdt_index;
1231                 tmp->lum_objects[0].lum_fid = *ll_inode2fid(inode);
1232                 for (i = 1; i < stripe_count; i++) {
1233                         struct lmv_mds_md_v1 *lmm1;
1234
1235                         lmm1 = &lmm->lmv_md_v1;
1236                         mdt_index = ll_get_mdt_idx_by_fid(sbi,
1237                                                      &lmm1->lmv_stripe_fids[i]);
1238                         if (mdt_index < 0)
1239                                 GOTO(out_tmp, rc = mdt_index);
1240
1241                         tmp->lum_objects[i].lum_mds = mdt_index;
1242                         tmp->lum_objects[i].lum_fid = lmm1->lmv_stripe_fids[i];
1243                         tmp->lum_stripe_count++;
1244                 }
1245
1246                 if (copy_to_user(ulmv, tmp, lum_size))
1247                         GOTO(out_tmp, rc = -EFAULT);
1248 out_tmp:
1249                 OBD_FREE(tmp, lum_size);
1250 finish_req:
1251                 ptlrpc_req_finished(request);
1252                 return rc;
1253         }
1254
1255         case LL_IOC_REMOVE_ENTRY: {
1256                 char            *filename = NULL;
1257                 int              namelen = 0;
1258                 int              rc;
1259
1260                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1261                  * unsupported server, which might crash the server(LU-2730),
1262                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1263                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1264                  * server will support REINT_RMENTRY XXX*/
1265                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1266                         RETURN(-ENOTSUPP);
1267
1268                 filename = ll_getname((const char *)arg);
1269                 if (IS_ERR(filename))
1270                         RETURN(PTR_ERR(filename));
1271
1272                 namelen = strlen(filename);
1273                 if (namelen < 1)
1274                         GOTO(out_rmdir, rc = -EINVAL);
1275
1276                 rc = ll_rmdir_entry(inode, filename, namelen);
1277 out_rmdir:
1278                 if (filename)
1279                         ll_putname(filename);
1280                 RETURN(rc);
1281         }
1282         case LL_IOC_LOV_SWAP_LAYOUTS:
1283                 RETURN(-EPERM);
1284         case LL_IOC_OBD_STATFS:
1285                 RETURN(ll_obd_statfs(inode, (void *)arg));
1286         case LL_IOC_LOV_GETSTRIPE:
1287         case LL_IOC_MDC_GETINFO:
1288         case IOC_MDC_GETFILEINFO:
1289         case IOC_MDC_GETFILESTRIPE: {
1290                 struct ptlrpc_request *request = NULL;
1291                 struct lov_user_md *lump;
1292                 struct lov_mds_md *lmm = NULL;
1293                 struct mdt_body *body;
1294                 char *filename = NULL;
1295                 int lmmsize;
1296
1297                 if (cmd == IOC_MDC_GETFILEINFO ||
1298                     cmd == IOC_MDC_GETFILESTRIPE) {
1299                         filename = ll_getname((const char *)arg);
1300                         if (IS_ERR(filename))
1301                                 RETURN(PTR_ERR(filename));
1302
1303                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1304                                                       &lmmsize, &request);
1305                 } else {
1306                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1307                                               &request, 0);
1308                 }
1309
1310                 if (request) {
1311                         body = req_capsule_server_get(&request->rq_pill,
1312                                                       &RMF_MDT_BODY);
1313                         LASSERT(body != NULL);
1314                 } else {
1315                         GOTO(out_req, rc);
1316                 }
1317
1318                 if (rc < 0) {
1319                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1320                                                cmd == LL_IOC_MDC_GETINFO))
1321                                 GOTO(skip_lmm, rc = 0);
1322                         else
1323                                 GOTO(out_req, rc);
1324                 }
1325
1326                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1327                     cmd == LL_IOC_LOV_GETSTRIPE) {
1328                         lump = (struct lov_user_md *)arg;
1329                 } else {
1330                         struct lov_user_mds_data *lmdp;
1331                         lmdp = (struct lov_user_mds_data *)arg;
1332                         lump = &lmdp->lmd_lmm;
1333                 }
1334                 if (copy_to_user(lump, lmm, lmmsize)) {
1335                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1336                                 GOTO(out_req, rc = -EFAULT);
1337                         rc = -EOVERFLOW;
1338                 }
1339         skip_lmm:
1340                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1341                         struct lov_user_mds_data *lmdp;
1342                         lstat_t st = { 0 };
1343
1344                         st.st_dev     = inode->i_sb->s_dev;
1345                         st.st_mode    = body->mode;
1346                         st.st_nlink   = body->nlink;
1347                         st.st_uid     = body->uid;
1348                         st.st_gid     = body->gid;
1349                         st.st_rdev    = body->rdev;
1350                         st.st_size    = body->size;
1351                         st.st_blksize = PAGE_CACHE_SIZE;
1352                         st.st_blocks  = body->blocks;
1353                         st.st_atime   = body->atime;
1354                         st.st_mtime   = body->mtime;
1355                         st.st_ctime   = body->ctime;
1356                         st.st_ino     = inode->i_ino;
1357
1358                         lmdp = (struct lov_user_mds_data *)arg;
1359                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1360                                 GOTO(out_req, rc = -EFAULT);
1361                 }
1362
1363                 EXIT;
1364         out_req:
1365                 ptlrpc_req_finished(request);
1366                 if (filename)
1367                         ll_putname(filename);
1368                 return rc;
1369         }
1370         case IOC_LOV_GETINFO: {
1371                 struct lov_user_mds_data *lumd;
1372                 struct lov_stripe_md *lsm;
1373                 struct lov_user_md *lum;
1374                 struct lov_mds_md *lmm;
1375                 int lmmsize;
1376                 lstat_t st;
1377
1378                 lumd = (struct lov_user_mds_data *)arg;
1379                 lum = &lumd->lmd_lmm;
1380
1381                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1382                 if (rc)
1383                         RETURN(rc);
1384
1385                 OBD_ALLOC_LARGE(lmm, lmmsize);
1386                 if (lmm == NULL)
1387                         RETURN(-ENOMEM);
1388
1389                 if (copy_from_user(lmm, lum, lmmsize))
1390                         GOTO(free_lmm, rc = -EFAULT);
1391
1392                 switch (lmm->lmm_magic) {
1393                 case LOV_USER_MAGIC_V1:
1394                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1395                                 break;
1396                         /* swab objects first so that stripes num will be sane */
1397                         lustre_swab_lov_user_md_objects(
1398                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1399                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1400                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1401                         break;
1402                 case LOV_USER_MAGIC_V3:
1403                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1404                                 break;
1405                         /* swab objects first so that stripes num will be sane */
1406                         lustre_swab_lov_user_md_objects(
1407                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1408                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1409                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1410                         break;
1411                 default:
1412                         GOTO(free_lmm, rc = -EINVAL);
1413                 }
1414
1415                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1416                 if (rc < 0)
1417                         GOTO(free_lmm, rc = -ENOMEM);
1418
1419                 /* Perform glimpse_size operation. */
1420                 memset(&st, 0, sizeof(st));
1421
1422                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1423                 if (rc)
1424                         GOTO(free_lsm, rc);
1425
1426                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1427                         GOTO(free_lsm, rc = -EFAULT);
1428
1429                 EXIT;
1430         free_lsm:
1431                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1432         free_lmm:
1433                 OBD_FREE_LARGE(lmm, lmmsize);
1434                 return rc;
1435         }
1436         case OBD_IOC_LLOG_CATINFO: {
1437                 RETURN(-EOPNOTSUPP);
1438         }
1439         case OBD_IOC_QUOTACHECK: {
1440                 struct obd_quotactl *oqctl;
1441                 int error = 0;
1442
1443                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1444                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1445                         RETURN(-EPERM);
1446
1447                 OBD_ALLOC_PTR(oqctl);
1448                 if (!oqctl)
1449                         RETURN(-ENOMEM);
1450                 oqctl->qc_type = arg;
1451                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1452                 if (rc < 0) {
1453                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1454                         error = rc;
1455                 }
1456
1457                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1458                 if (rc < 0)
1459                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1460
1461                 OBD_FREE_PTR(oqctl);
1462                 return error ?: rc;
1463         }
1464         case OBD_IOC_POLL_QUOTACHECK: {
1465                 struct if_quotacheck *check;
1466
1467                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1468                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1469                         RETURN(-EPERM);
1470
1471                 OBD_ALLOC_PTR(check);
1472                 if (!check)
1473                         RETURN(-ENOMEM);
1474
1475                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1476                                    NULL);
1477                 if (rc) {
1478                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1479                         if (copy_to_user((void *)arg, check,
1480                                              sizeof(*check)))
1481                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1482                         GOTO(out_poll, rc);
1483                 }
1484
1485                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1486                                    NULL);
1487                 if (rc) {
1488                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1489                         if (copy_to_user((void *)arg, check,
1490                                              sizeof(*check)))
1491                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1492                         GOTO(out_poll, rc);
1493                 }
1494         out_poll:
1495                 OBD_FREE_PTR(check);
1496                 RETURN(rc);
1497         }
1498 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1499         case LL_IOC_QUOTACTL_18: {
1500                 /* copy the old 1.x quota struct for internal use, then copy
1501                  * back into old format struct.  For 1.8 compatibility. */
1502                 struct if_quotactl_18 *qctl_18;
1503                 struct if_quotactl *qctl_20;
1504
1505                 OBD_ALLOC_PTR(qctl_18);
1506                 if (!qctl_18)
1507                         RETURN(-ENOMEM);
1508
1509                 OBD_ALLOC_PTR(qctl_20);
1510                 if (!qctl_20)
1511                         GOTO(out_quotactl_18, rc = -ENOMEM);
1512
1513                 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1514                         GOTO(out_quotactl_20, rc = -ENOMEM);
1515
1516                 QCTL_COPY(qctl_20, qctl_18);
1517                 qctl_20->qc_idx = 0;
1518
1519                 /* XXX: dqb_valid was borrowed as a flag to mark that
1520                  *      only mds quota is wanted */
1521                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1522                     qctl_18->qc_dqblk.dqb_valid) {
1523                         qctl_20->qc_valid = QC_MDTIDX;
1524                         qctl_20->qc_dqblk.dqb_valid = 0;
1525                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1526                         qctl_20->qc_valid = QC_UUID;
1527                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1528                 } else {
1529                         qctl_20->qc_valid = QC_GENERAL;
1530                 }
1531
1532                 rc = quotactl_ioctl(sbi, qctl_20);
1533
1534                 if (rc == 0) {
1535                         QCTL_COPY(qctl_18, qctl_20);
1536                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1537
1538                         if (copy_to_user((void *)arg, qctl_18,
1539                                              sizeof(*qctl_18)))
1540                                 rc = -EFAULT;
1541                 }
1542
1543         out_quotactl_20:
1544                 OBD_FREE_PTR(qctl_20);
1545         out_quotactl_18:
1546                 OBD_FREE_PTR(qctl_18);
1547                 RETURN(rc);
1548         }
1549 #else
1550 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1551 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1552         case LL_IOC_QUOTACTL: {
1553                 struct if_quotactl *qctl;
1554
1555                 OBD_ALLOC_PTR(qctl);
1556                 if (!qctl)
1557                         RETURN(-ENOMEM);
1558
1559                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1560                         GOTO(out_quotactl, rc = -EFAULT);
1561
1562                 rc = quotactl_ioctl(sbi, qctl);
1563
1564                 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1565                         rc = -EFAULT;
1566
1567         out_quotactl:
1568                 OBD_FREE_PTR(qctl);
1569                 RETURN(rc);
1570         }
1571         case OBD_IOC_GETDTNAME:
1572         case OBD_IOC_GETMDNAME:
1573                 RETURN(ll_get_obd_name(inode, cmd, arg));
1574         case LL_IOC_FLUSHCTX:
1575                 RETURN(ll_flush_ctx(inode));
1576 #ifdef CONFIG_FS_POSIX_ACL
1577         case LL_IOC_RMTACL: {
1578             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1579                 inode == inode->i_sb->s_root->d_inode) {
1580                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1581
1582                 LASSERT(fd != NULL);
1583                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1584                 if (!rc)
1585                         fd->fd_flags |= LL_FILE_RMTACL;
1586                 RETURN(rc);
1587             } else
1588                 RETURN(0);
1589         }
1590 #endif
1591         case LL_IOC_GETOBDCOUNT: {
1592                 int count, vallen;
1593                 struct obd_export *exp;
1594
1595                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1596                         RETURN(-EFAULT);
1597
1598                 /* get ost count when count is zero, get mdt count otherwise */
1599                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1600                 vallen = sizeof(count);
1601                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1602                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1603                 if (rc) {
1604                         CERROR("get target count failed: %d\n", rc);
1605                         RETURN(rc);
1606                 }
1607
1608                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1609                         RETURN(-EFAULT);
1610
1611                 RETURN(0);
1612         }
1613         case LL_IOC_PATH2FID:
1614                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1615                                      sizeof(struct lu_fid)))
1616                         RETURN(-EFAULT);
1617                 RETURN(0);
1618         case LL_IOC_GET_CONNECT_FLAGS: {
1619                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1620         }
1621         case OBD_IOC_CHANGELOG_SEND:
1622         case OBD_IOC_CHANGELOG_CLEAR:
1623                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1624                                     sizeof(struct ioc_changelog));
1625                 RETURN(rc);
1626         case OBD_IOC_FID2PATH:
1627                 RETURN(ll_fid2path(inode, (void *)arg));
1628         case LL_IOC_HSM_REQUEST: {
1629                 struct hsm_user_request *hur;
1630                 int                      totalsize;
1631
1632                 OBD_ALLOC_PTR(hur);
1633                 if (hur == NULL)
1634                         RETURN(-ENOMEM);
1635
1636                 /* We don't know the true size yet; copy the fixed-size part */
1637                 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1638                         OBD_FREE_PTR(hur);
1639                         RETURN(-EFAULT);
1640                 }
1641
1642                 /* Compute the whole struct size */
1643                 totalsize = hur_len(hur);
1644                 OBD_FREE_PTR(hur);
1645
1646                 /* Final size will be more than double totalsize */
1647                 if (totalsize >= MDS_MAXREQSIZE / 3)
1648                         RETURN(-E2BIG);
1649
1650                 OBD_ALLOC_LARGE(hur, totalsize);
1651                 if (hur == NULL)
1652                         RETURN(-ENOMEM);
1653
1654                 /* Copy the whole struct */
1655                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1656                         OBD_FREE_LARGE(hur, totalsize);
1657                         RETURN(-EFAULT);
1658                 }
1659
1660                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1661                         const struct lu_fid *fid;
1662                         struct inode *f;
1663                         int i;
1664
1665                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1666                                 fid = &hur->hur_user_item[i].hui_fid;
1667                                 f = search_inode_for_lustre(inode->i_sb, fid);
1668                                 if (IS_ERR(f)) {
1669                                         rc = PTR_ERR(f);
1670                                         break;
1671                                 }
1672
1673                                 rc = ll_hsm_release(f);
1674                                 iput(f);
1675                                 if (rc != 0)
1676                                         break;
1677                         }
1678                 } else {
1679                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1680                                            hur, NULL);
1681                 }
1682
1683                 OBD_FREE_LARGE(hur, totalsize);
1684
1685                 RETURN(rc);
1686         }
1687         case LL_IOC_HSM_PROGRESS: {
1688                 struct hsm_progress_kernel      hpk;
1689                 struct hsm_progress             hp;
1690
1691                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1692                         RETURN(-EFAULT);
1693
1694                 hpk.hpk_fid = hp.hp_fid;
1695                 hpk.hpk_cookie = hp.hp_cookie;
1696                 hpk.hpk_extent = hp.hp_extent;
1697                 hpk.hpk_flags = hp.hp_flags;
1698                 hpk.hpk_errval = hp.hp_errval;
1699                 hpk.hpk_data_version = 0;
1700
1701                 /* File may not exist in Lustre; all progress
1702                  * reported to Lustre root */
1703                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1704                                    NULL);
1705                 RETURN(rc);
1706         }
1707         case LL_IOC_HSM_CT_START:
1708                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1709                         RETURN(-EPERM);
1710
1711                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1712                                     sizeof(struct lustre_kernelcomm));
1713                 RETURN(rc);
1714
1715         case LL_IOC_HSM_COPY_START: {
1716                 struct hsm_copy *copy;
1717                 int              rc;
1718
1719                 OBD_ALLOC_PTR(copy);
1720                 if (copy == NULL)
1721                         RETURN(-ENOMEM);
1722                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1723                         OBD_FREE_PTR(copy);
1724                         RETURN(-EFAULT);
1725                 }
1726
1727                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1728                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1729                         rc = -EFAULT;
1730
1731                 OBD_FREE_PTR(copy);
1732                 RETURN(rc);
1733         }
1734         case LL_IOC_HSM_COPY_END: {
1735                 struct hsm_copy *copy;
1736                 int              rc;
1737
1738                 OBD_ALLOC_PTR(copy);
1739                 if (copy == NULL)
1740                         RETURN(-ENOMEM);
1741                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1742                         OBD_FREE_PTR(copy);
1743                         RETURN(-EFAULT);
1744                 }
1745
1746                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1747                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1748                         rc = -EFAULT;
1749
1750                 OBD_FREE_PTR(copy);
1751                 RETURN(rc);
1752         }
1753         case LL_IOC_MIGRATE: {
1754                 char            *buf = NULL;
1755                 const char      *filename;
1756                 int             namelen = 0;
1757                 int             len;
1758                 int             rc;
1759                 int             mdtidx;
1760
1761                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1762                 if (rc < 0)
1763                         RETURN(rc);
1764
1765                 data = (struct obd_ioctl_data *)buf;
1766                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1767                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1768                         GOTO(migrate_free, rc = -EINVAL);
1769
1770                 filename = data->ioc_inlbuf1;
1771                 namelen = data->ioc_inllen1;
1772                 if (namelen < 1)
1773                         GOTO(migrate_free, rc = -EINVAL);
1774
1775                 if (data->ioc_inllen2 != sizeof(mdtidx))
1776                         GOTO(migrate_free, rc = -EINVAL);
1777                 mdtidx = *(int *)data->ioc_inlbuf2;
1778
1779                 rc = ll_migrate(inode, file, mdtidx, filename, namelen);
1780 migrate_free:
1781                 obd_ioctl_freedata(buf, len);
1782
1783                 RETURN(rc);
1784         }
1785         default:
1786                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1787                                      (void *)arg));
1788         }
1789 }
1790
1791 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1792 {
1793         struct inode *inode = file->f_mapping->host;
1794         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1795         struct ll_sb_info *sbi = ll_i2sbi(inode);
1796         int api32 = ll_need_32bit_api(sbi);
1797         loff_t ret = -EINVAL;
1798         ENTRY;
1799
1800         mutex_lock(&inode->i_mutex);
1801         switch (origin) {
1802                 case SEEK_SET:
1803                         break;
1804                 case SEEK_CUR:
1805                         offset += file->f_pos;
1806                         break;
1807                 case SEEK_END:
1808                         if (offset > 0)
1809                                 GOTO(out, ret);
1810                         if (api32)
1811                                 offset += LL_DIR_END_OFF_32BIT;
1812                         else
1813                                 offset += LL_DIR_END_OFF;
1814                         break;
1815                 default:
1816                         GOTO(out, ret);
1817         }
1818
1819         if (offset >= 0 &&
1820             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1821              (!api32 && offset <= LL_DIR_END_OFF))) {
1822                 if (offset != file->f_pos) {
1823                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1824                             (!api32 && offset == LL_DIR_END_OFF))
1825                                 fd->lfd_pos = MDS_DIR_END_OFF;
1826                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1827                                 fd->lfd_pos = offset << 32;
1828                         else
1829                                 fd->lfd_pos = offset;
1830                         file->f_pos = offset;
1831                         file->f_version = 0;
1832                 }
1833                 ret = offset;
1834         }
1835         GOTO(out, ret);
1836
1837 out:
1838         mutex_unlock(&inode->i_mutex);
1839         return ret;
1840 }
1841
1842 int ll_dir_open(struct inode *inode, struct file *file)
1843 {
1844         ENTRY;
1845         RETURN(ll_file_open(inode, file));
1846 }
1847
1848 int ll_dir_release(struct inode *inode, struct file *file)
1849 {
1850         ENTRY;
1851         RETURN(ll_file_release(inode, file));
1852 }
1853
1854 struct file_operations ll_dir_operations = {
1855         .llseek         = ll_dir_seek,
1856         .open           = ll_dir_open,
1857         .release        = ll_dir_release,
1858         .read           = generic_read_dir,
1859 #ifdef HAVE_DIR_CONTEXT
1860         .iterate        = ll_iterate,
1861 #else
1862         .readdir        = ll_readdir,
1863 #endif
1864         .unlocked_ioctl = ll_dir_ioctl,
1865         .fsync          = ll_fsync,
1866 };