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_ioctl.h>
  55 #include <lustre_lib.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 = NULL;
 166         struct md_callback cb_op;
 167         int rc;
 168         ENTRY;
 169
 170         LASSERT(*ppage == NULL);
 171         cb_op.md_blocking_ast = ll_md_blocking_ast;
 172         op_data->op_cli_flags &= ~CLI_NEXT_ENTRY;
 173         rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
 174         if (rc != 0)
 175                 entry = ERR_PTR(rc);
 176         RETURN(entry);
 177 }
 178
 179 struct lu_dirent *ll_dir_entry_next(struct inode *dir,
 180                                     struct md_op_data *op_data,
 181                                     struct lu_dirent *ent,
 182                                     struct page **ppage)
 183 {
 184         struct lu_dirent *entry = NULL;
 185         struct md_callback cb_op;
 186         int rc;
 187         ENTRY;
 188
 189         op_data->op_hash_offset = le64_to_cpu(ent->lde_hash);
 190
 191         /* release last page */
 192         LASSERT(*ppage != NULL);
 193         kunmap(*ppage);
 194         page_cache_release(*ppage);
 195
 196         cb_op.md_blocking_ast = ll_md_blocking_ast;
 197         op_data->op_cli_flags |= CLI_NEXT_ENTRY;
 198         rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
 199         if (rc != 0)
 200                 entry = ERR_PTR(rc);
 201
 202         RETURN(entry);
 203 }
 204
 205 #ifdef HAVE_DIR_CONTEXT
 206 int ll_dir_read(struct inode *inode, struct md_op_data *op_data,
 207                 struct dir_context *ctx)
 208 {
 209 #else
 210 int ll_dir_read(struct inode *inode, 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         struct ll_dir_chain     chain;
 216         struct lu_dirent        *ent;
 217         int                     api32 = ll_need_32bit_api(sbi);
 218         int                     hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 219         int                     done = 0;
 220         int                     rc = 0;
 221         __u64                   hash = MDS_DIR_END_OFF;
 222         struct page             *page = NULL;
 223         ENTRY;
 224
 225         ll_dir_chain_init(&chain);
 226         for (ent = ll_dir_entry_start(inode, op_data, &page);
 227              ent != NULL && !IS_ERR(ent) && !done;
 228              ent = ll_dir_entry_next(inode, op_data, ent, &page)) {
 229                 __u16          type;
 230                 int            namelen;
 231                 struct lu_fid  fid;
 232                 __u64          lhash;
 233                 __u64          ino;
 234
 235                 hash = le64_to_cpu(ent->lde_hash);
 236                 if (hash < op_data->op_hash_offset)
 237                         /*
 238                          * Skip until we find target hash
 239                          * value.
 240                          */
 241                         continue;
 242                 namelen = le16_to_cpu(ent->lde_namelen);
 243                 if (namelen == 0)
 244                         /*
 245                          * Skip dummy record.
 246                          */
 247                         continue;
 248
 249                 if (api32 && hash64)
 250                         lhash = hash >> 32;
 251                 else
 252                         lhash = hash;
 253                 fid_le_to_cpu(&fid, &ent->lde_fid);
 254                 ino = cl_fid_build_ino(&fid, api32);
 255                 type = ll_dirent_type_get(ent);
 256
 257 #ifdef HAVE_DIR_CONTEXT
 258                 /* For 'll_nfs_get_name_filldir()', it will try
 259                  * to access the 'ent' through its 'lde_name',
 260                  * so the parameter 'name' for 'filldir()' must
 261                  * be part of the 'ent'. */
 262                 done = !dir_emit(ctx, ent->lde_name, namelen, ino, type);
 263 #else
 264                 done = filldir(cookie, ent->lde_name, namelen, lhash,
 265                                ino, type);
 266 #endif
 267                 if (done) {
 268                         if (op_data->op_hash_offset != MDS_DIR_END_OFF)
 269                                 op_data->op_hash_offset = hash;
 270                         break;
 271                 }
 272         }
 273
 274         if (IS_ERR(ent))
 275                 rc = PTR_ERR(ent);
 276         else if (ent == NULL)
 277                 op_data->op_hash_offset = MDS_DIR_END_OFF;
 278
 279         if (page != NULL) {
 280                 kunmap(page);
 281                 page_cache_release(page);
 282         }
 283
 284         ll_dir_chain_fini(&chain);
 285         RETURN(rc);
 286 }
 287
 288 #ifdef HAVE_DIR_CONTEXT
 289 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 290 #else
 291 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 292 #endif
 293 {
 294         struct inode            *inode  = filp->f_dentry->d_inode;
 295         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 296         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 297         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 298         int                     api32   = ll_need_32bit_api(sbi);
 299         struct md_op_data       *op_data;
 300         __u64                   pos;
 301         int                     rc;
 302         ENTRY;
 303
 304         if (lfd != NULL)
 305                 pos = lfd->lfd_pos;
 306         else
 307                 pos = 0;
 308
 309         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 310                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 311                inode, (unsigned long)pos, i_size_read(inode), api32);
 312
 313         if (pos == MDS_DIR_END_OFF)
 314                 /*
 315                  * end-of-file.
 316                  */
 317                 GOTO(out, rc = 0);
 318
 319         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 320                                      LUSTRE_OPC_ANY, inode);
 321         if (IS_ERR(op_data))
 322                 GOTO(out, rc = PTR_ERR(op_data));
 323
 324         if (unlikely(op_data->op_mea1 != NULL)) {
 325                 /* This is only needed for striped dir to fill ..,
 326                  * see lmv_read_entry */
 327                 if (filp->f_dentry->d_parent != NULL &&
 328                     filp->f_dentry->d_parent->d_inode != NULL) {
 329                         __u64 ibits = MDS_INODELOCK_UPDATE;
 330                         struct inode *parent =
 331                                 filp->f_dentry->d_parent->d_inode;
 332
 333                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 334                                 op_data->op_fid3 = *ll_inode2fid(parent);
 335                 }
 336
 337                 /* If it can not find in cache, do lookup .. on the master
 338                  * object */
 339                 if (fid_is_zero(&op_data->op_fid3)) {
 340                         rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
 341                         if (rc != 0)
 342                                 RETURN(rc);
 343                 }
 344         }
 345         op_data->op_hash_offset = pos;
 346         op_data->op_max_pages = sbi->ll_md_brw_pages;
 347 #ifdef HAVE_DIR_CONTEXT
 348         ctx->pos = pos;
 349         rc = ll_dir_read(inode, op_data, ctx);
 350         pos = ctx->pos;
 351 #else
 352         rc = ll_dir_read(inode, op_data, cookie, filldir);
 353 #endif
 354         if (lfd != NULL)
 355                 lfd->lfd_pos = op_data->op_hash_offset;
 356
 357         if (pos == MDS_DIR_END_OFF) {
 358                 if (api32)
 359                         pos = LL_DIR_END_OFF_32BIT;
 360                 else
 361                         pos = LL_DIR_END_OFF;
 362         } else {
 363                 if (api32 && hash64)
 364                         pos = op_data->op_hash_offset >> 32;
 365                 else
 366                         pos = op_data->op_hash_offset;
 367         }
 368 #ifdef HAVE_DIR_CONTEXT
 369         ctx->pos = pos;
 370 #else
 371         filp->f_pos = pos;
 372 #endif
 373         ll_finish_md_op_data(op_data);
 374         filp->f_version = inode->i_version;
 375
 376 out:
 377         if (!rc)
 378                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 379
 380         RETURN(rc);
 381 }
 382
 383 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 384 {
 385         struct mgs_send_param *msp;
 386         int rc = 0;
 387
 388         OBD_ALLOC_PTR(msp);
 389         if (!msp)
 390                 return -ENOMEM;
 391
 392         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 393         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 394                                 sizeof(struct mgs_send_param), msp, NULL);
 395         if (rc)
 396                 CERROR("Failed to set parameter: %d\n", rc);
 397         OBD_FREE_PTR(msp);
 398
 399         return rc;
 400 }
 401
 402 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 403                                const char *filename)
 404 {
 405         struct ptlrpc_request *request = NULL;
 406         struct md_op_data *op_data;
 407         struct ll_sb_info *sbi = ll_i2sbi(dir);
 408         int mode;
 409         int err;
 410         ENTRY;
 411
 412         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 413                 RETURN(-EINVAL);
 414
 415         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
 416                "stripe_offset %d, stripe_count: %u\n",
 417                PFID(ll_inode2fid(dir)), dir, filename,
 418                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 419
 420         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 421                 lustre_swab_lmv_user_md(lump);
 422
 423         mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
 424         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 425                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 426                                      lump);
 427         if (IS_ERR(op_data))
 428                 GOTO(err_exit, err = PTR_ERR(op_data));
 429
 430         op_data->op_cli_flags |= CLI_SET_MEA;
 431         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 432                         from_kuid(&init_user_ns, current_fsuid()),
 433                         from_kgid(&init_user_ns, current_fsgid()),
 434                         cfs_curproc_cap_pack(), 0, &request);
 435         ll_finish_md_op_data(op_data);
 436         if (err)
 437                 GOTO(err_exit, err);
 438 err_exit:
 439         ptlrpc_req_finished(request);
 440         return err;
 441 }
 442
 443 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 444                      int set_default)
 445 {
 446         struct ll_sb_info *sbi = ll_i2sbi(inode);
 447         struct md_op_data *op_data;
 448         struct ptlrpc_request *req = NULL;
 449         int rc = 0;
 450         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 451         struct obd_device *mgc = lsi->lsi_mgc;
 452         int lum_size;
 453         ENTRY;
 454
 455         if (lump != NULL) {
 456                 /*
 457                  * This is coming from userspace, so should be in
 458                  * local endian.  But the MDS would like it in little
 459                  * endian, so we swab it before we send it.
 460                  */
 461                 switch (lump->lmm_magic) {
 462                 case LOV_USER_MAGIC_V1: {
 463                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 464                                 lustre_swab_lov_user_md_v1(lump);
 465                         lum_size = sizeof(struct lov_user_md_v1);
 466                         break;
 467                 }
 468                 case LOV_USER_MAGIC_V3: {
 469                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 470                                 lustre_swab_lov_user_md_v3(
 471                                         (struct lov_user_md_v3 *)lump);
 472                         lum_size = sizeof(struct lov_user_md_v3);
 473                         break;
 474                 }
 475                 case LMV_USER_MAGIC: {
 476                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 477                                 lustre_swab_lmv_user_md(
 478                                         (struct lmv_user_md *)lump);
 479                         lum_size = sizeof(struct lmv_user_md);
 480                         break;
 481                 }
 482                 default: {
 483                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 484                                         " %#08x != %#08x nor %#08x\n",
 485                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 486                                         LOV_USER_MAGIC_V3);
 487                         RETURN(-EINVAL);
 488                 }
 489                 }
 490         } else {
 491                 lum_size = sizeof(struct lov_user_md_v1);
 492         }
 493
 494         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 495                                      LUSTRE_OPC_ANY, NULL);
 496         if (IS_ERR(op_data))
 497                 RETURN(PTR_ERR(op_data));
 498
 499         /* swabbing is done in lov_setstripe() on server side */
 500         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 501                         NULL, 0, &req, NULL);
 502         ll_finish_md_op_data(op_data);
 503         ptlrpc_req_finished(req);
 504         if (rc) {
 505                 if (rc != -EPERM && rc != -EACCES)
 506                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 507         }
 508
 509         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 510          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 511          need the make the distiction between the 2 versions */
 512         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 513                 char *param = NULL;
 514                 char *buf;
 515
 516                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 517                 if (param == NULL)
 518                         GOTO(end, rc = -ENOMEM);
 519
 520                 buf = param;
 521                 /* Get fsname and assume devname to be -MDT0000. */
 522                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 523                 strcat(buf, "-MDT0000.lov");
 524                 buf += strlen(buf);
 525
 526                 /* Set root stripesize */
 527                 sprintf(buf, ".stripesize=%u",
 528                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 529                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 530                 if (rc)
 531                         GOTO(end, rc);
 532
 533                 /* Set root stripecount */
 534                 sprintf(buf, ".stripecount=%hd",
 535                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 536                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 537                 if (rc)
 538                         GOTO(end, rc);
 539
 540                 /* Set root stripeoffset */
 541                 sprintf(buf, ".stripeoffset=%hd",
 542                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 543                         (typeof(lump->lmm_stripe_offset))(-1));
 544                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 545
 546 end:
 547                 if (param != NULL)
 548                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 549         }
 550         RETURN(rc);
 551 }
 552
 553 /**
 554  * This function will be used to get default LOV/LMV/Default LMV
 555  * @valid will be used to indicate which stripe it will retrieve
 556  *      OBD_MD_MEA              LMV stripe EA
 557  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 558  *      otherwise               Default LOV EA.
 559  * Each time, it can only retrieve 1 stripe EA
 560  **/
 561 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 562                      struct ptlrpc_request **request, obd_valid valid)
 563 {
 564         struct ll_sb_info *sbi = ll_i2sbi(inode);
 565         struct mdt_body   *body;
 566         struct lov_mds_md *lmm = NULL;
 567         struct ptlrpc_request *req = NULL;
 568         int rc, lmm_size;
 569         struct md_op_data *op_data;
 570         ENTRY;
 571
 572         rc = ll_get_default_mdsize(sbi, &lmm_size);
 573         if (rc)
 574                 RETURN(rc);
 575
 576         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 577                                      0, lmm_size, LUSTRE_OPC_ANY,
 578                                      NULL);
 579         if (IS_ERR(op_data))
 580                 RETURN(PTR_ERR(op_data));
 581
 582         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 583         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 584         ll_finish_md_op_data(op_data);
 585         if (rc < 0) {
 586                 CDEBUG(D_INFO, "md_getattr failed on inode "
 587                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 588                 GOTO(out, rc);
 589         }
 590
 591         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 592         LASSERT(body != NULL);
 593
 594         lmm_size = body->eadatasize;
 595
 596         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 597             lmm_size == 0) {
 598                 GOTO(out, rc = -ENODATA);
 599         }
 600
 601         lmm = req_capsule_server_sized_get(&req->rq_pill,
 602                                            &RMF_MDT_MD, lmm_size);
 603         LASSERT(lmm != NULL);
 604
 605         /*
 606          * This is coming from the MDS, so is probably in
 607          * little endian.  We convert it to host endian before
 608          * passing it to userspace.
 609          */
 610         /* We don't swab objects for directories */
 611         switch (le32_to_cpu(lmm->lmm_magic)) {
 612         case LOV_MAGIC_V1:
 613                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 614                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 615                 break;
 616         case LOV_MAGIC_V3:
 617                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 618                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 619                 break;
 620         case LMV_MAGIC_V1:
 621                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 622                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 623                 break;
 624         case LMV_USER_MAGIC:
 625                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 626                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 627                 break;
 628         default:
 629                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 630                 rc = -EPROTO;
 631         }
 632 out:
 633         *plmm = lmm;
 634         *plmm_size = lmm_size;
 635         *request = req;
 636         return rc;
 637 }
 638
 639 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 640 {
 641         struct md_op_data       *op_data;
 642         int                     rc;
 643         int                     mdt_index;
 644         ENTRY;
 645
 646         OBD_ALLOC_PTR(op_data);
 647         if (op_data == NULL)
 648                 RETURN(-ENOMEM);
 649
 650         op_data->op_flags |= MF_GET_MDT_IDX;
 651         op_data->op_fid1 = *fid;
 652         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 653         mdt_index = op_data->op_mds;
 654         OBD_FREE_PTR(op_data);
 655         if (rc < 0)
 656                 RETURN(rc);
 657
 658         RETURN(mdt_index);
 659 }
 660
 661 /*
 662  *  Get MDT index for the inode.
 663  */
 664 int ll_get_mdt_idx(struct inode *inode)
 665 {
 666         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 667 }
 668
 669 /**
 670  * Generic handler to do any pre-copy work.
 671  *
 672  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 673  * first information for it that real work has started.
 674  *
 675  * Moreover, for a ARCHIVE request, it will sample the file data version and
 676  * store it in \a copy.
 677  *
 678  * \return 0 on success.
 679  */
 680 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 681 {
 682         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 683         struct hsm_progress_kernel       hpk;
 684         int                              rc;
 685         ENTRY;
 686
 687         /* Forge a hsm_progress based on data from copy. */
 688         hpk.hpk_fid = copy->hc_hai.hai_fid;
 689         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 690         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 691         hpk.hpk_extent.length = 0;
 692         hpk.hpk_flags = 0;
 693         hpk.hpk_errval = 0;
 694         hpk.hpk_data_version = 0;
 695
 696
 697         /* For archive request, we need to read the current file version. */
 698         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 699                 struct inode    *inode;
 700                 __u64            data_version = 0;
 701
 702                 /* Get inode for this fid */
 703                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 704                 if (IS_ERR(inode)) {
 705                         hpk.hpk_flags |= HP_FLAG_RETRY;
 706                         /* hpk_errval is >= 0 */
 707                         hpk.hpk_errval = -PTR_ERR(inode);
 708                         GOTO(progress, rc = PTR_ERR(inode));
 709                 }
 710
 711                 /* Read current file data version */
 712                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 713                 iput(inode);
 714                 if (rc != 0) {
 715                         CDEBUG(D_HSM, "Could not read file data version of "
 716                                       DFID" (rc = %d). Archive request ("
 717                                       LPX64") could not be done.\n",
 718                                       PFID(&copy->hc_hai.hai_fid), rc,
 719                                       copy->hc_hai.hai_cookie);
 720                         hpk.hpk_flags |= HP_FLAG_RETRY;
 721                         /* hpk_errval must be >= 0 */
 722                         hpk.hpk_errval = -rc;
 723                         GOTO(progress, rc);
 724                 }
 725
 726                 /* Store it the hsm_copy for later copytool use.
 727                  * Always modified even if no lsm. */
 728                 copy->hc_data_version = data_version;
 729         }
 730
 731 progress:
 732         /* On error, the request should be considered as completed */
 733         if (hpk.hpk_errval > 0)
 734                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 735         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 736                            &hpk, NULL);
 737
 738         RETURN(rc);
 739 }
 740
 741 /**
 742  * Generic handler to do any post-copy work.
 743  *
 744  * It will send the last hsm_progress update to coordinator to inform it
 745  * that copy is finished and whether it was successful or not.
 746  *
 747  * Moreover,
 748  * - for ARCHIVE request, it will sample the file data version and compare it
 749  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 750  *   will be considered as failed.
 751  * - for RESTORE request, it will sample the file data version and send it to
 752  *   coordinator which is useful if the file was imported as 'released'.
 753  *
 754  * \return 0 on success.
 755  */
 756 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 757 {
 758         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 759         struct hsm_progress_kernel       hpk;
 760         int                              rc;
 761         ENTRY;
 762
 763         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 764         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 765          * initialized if copy_end was called with copy == NULL.
 766          */
 767
 768         /* Forge a hsm_progress based on data from copy. */
 769         hpk.hpk_fid = copy->hc_hai.hai_fid;
 770         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 771         hpk.hpk_extent = copy->hc_hai.hai_extent;
 772         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 773         hpk.hpk_errval = copy->hc_errval;
 774         hpk.hpk_data_version = 0;
 775
 776         /* For archive request, we need to check the file data was not changed.
 777          *
 778          * For restore request, we need to send the file data version, this is
 779          * useful when the file was created using hsm_import.
 780          */
 781         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 782              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 783             (copy->hc_errval == 0)) {
 784                 struct inode    *inode;
 785                 __u64            data_version = 0;
 786
 787                 /* Get lsm for this fid */
 788                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 789                 if (IS_ERR(inode)) {
 790                         hpk.hpk_flags |= HP_FLAG_RETRY;
 791                         /* hpk_errval must be >= 0 */
 792                         hpk.hpk_errval = -PTR_ERR(inode);
 793                         GOTO(progress, rc = PTR_ERR(inode));
 794                 }
 795
 796                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 797                 iput(inode);
 798                 if (rc) {
 799                         CDEBUG(D_HSM, "Could not read file data version. "
 800                                       "Request could not be confirmed.\n");
 801                         if (hpk.hpk_errval == 0)
 802                                 hpk.hpk_errval = -rc;
 803                         GOTO(progress, rc);
 804                 }
 805
 806                 /* Store it the hsm_copy for later copytool use.
 807                  * Always modified even if no lsm. */
 808                 hpk.hpk_data_version = data_version;
 809
 810                 /* File could have been stripped during archiving, so we need
 811                  * to check anyway. */
 812                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 813                     (copy->hc_data_version != data_version)) {
 814                         CDEBUG(D_HSM, "File data version mismatched. "
 815                               "File content was changed during archiving. "
 816                                DFID", start:"LPX64" current:"LPX64"\n",
 817                                PFID(&copy->hc_hai.hai_fid),
 818                                copy->hc_data_version, data_version);
 819                         /* File was changed, send error to cdt. Do not ask for
 820                          * retry because if a file is modified frequently,
 821                          * the cdt will loop on retried archive requests.
 822                          * The policy engine will ask for a new archive later
 823                          * when the file will not be modified for some tunable
 824                          * time */
 825                         /* we do not notify caller */
 826                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 827                         /* hpk_errval must be >= 0 */
 828                         hpk.hpk_errval = EBUSY;
 829                 }
 830
 831         }
 832
 833 progress:
 834         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 835                            &hpk, NULL);
 836
 837         RETURN(rc);
 838 }
 839
 840
 841 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 842                           const void __user *data, size_t size)
 843 {
 844         void *copy;
 845         int rc;
 846
 847         OBD_ALLOC(copy, size);
 848         if (copy == NULL)
 849                 return -ENOMEM;
 850
 851         if (copy_from_user(copy, data, size)) {
 852                 rc = -EFAULT;
 853                 goto out;
 854         }
 855
 856         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 857 out:
 858         OBD_FREE(copy, size);
 859
 860         return rc;
 861 }
 862
 863 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 864 {
 865         int cmd = qctl->qc_cmd;
 866         int type = qctl->qc_type;
 867         int id = qctl->qc_id;
 868         int valid = qctl->qc_valid;
 869         int rc = 0;
 870         ENTRY;
 871
 872         switch (cmd) {
 873         case LUSTRE_Q_INVALIDATE:
 874         case LUSTRE_Q_FINVALIDATE:
 875         case Q_QUOTAON:
 876         case Q_QUOTAOFF:
 877         case Q_SETQUOTA:
 878         case Q_SETINFO:
 879                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 880                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 881                         RETURN(-EPERM);
 882                 break;
 883         case Q_GETQUOTA:
 884                 if (((type == USRQUOTA &&
 885                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
 886                      (type == GRPQUOTA &&
 887                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
 888                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 889                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 890                         RETURN(-EPERM);
 891                 break;
 892         case Q_GETINFO:
 893                 break;
 894         default:
 895                 CERROR("unsupported quotactl op: %#x\n", cmd);
 896                 RETURN(-ENOTTY);
 897         }
 898
 899         if (valid != QC_GENERAL) {
 900                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 901                         RETURN(-EOPNOTSUPP);
 902
 903                 if (cmd == Q_GETINFO)
 904                         qctl->qc_cmd = Q_GETOINFO;
 905                 else if (cmd == Q_GETQUOTA)
 906                         qctl->qc_cmd = Q_GETOQUOTA;
 907                 else
 908                         RETURN(-EINVAL);
 909
 910                 switch (valid) {
 911                 case QC_MDTIDX:
 912                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 913                                            sizeof(*qctl), qctl, NULL);
 914                         break;
 915                 case QC_OSTIDX:
 916                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 917                                            sizeof(*qctl), qctl, NULL);
 918                         break;
 919                 case QC_UUID:
 920                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 921                                            sizeof(*qctl), qctl, NULL);
 922                         if (rc == -EAGAIN)
 923                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 924                                                    sbi->ll_dt_exp,
 925                                                    sizeof(*qctl), qctl, NULL);
 926                         break;
 927                 default:
 928                         rc = -EINVAL;
 929                         break;
 930                 }
 931
 932                 if (rc)
 933                         RETURN(rc);
 934
 935                 qctl->qc_cmd = cmd;
 936         } else {
 937                 struct obd_quotactl *oqctl;
 938
 939                 OBD_ALLOC_PTR(oqctl);
 940                 if (oqctl == NULL)
 941                         RETURN(-ENOMEM);
 942
 943                 QCTL_COPY(oqctl, qctl);
 944                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
 945                 if (rc) {
 946                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
 947                                 oqctl->qc_cmd = Q_QUOTAOFF;
 948                                 obd_quotactl(sbi->ll_md_exp, oqctl);
 949                         }
 950                         OBD_FREE_PTR(oqctl);
 951                         RETURN(rc);
 952                 }
 953                 /* If QIF_SPACE is not set, client should collect the
 954                  * space usage from OSSs by itself */
 955                 if (cmd == Q_GETQUOTA &&
 956                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
 957                     !oqctl->qc_dqblk.dqb_curspace) {
 958                         struct obd_quotactl *oqctl_tmp;
 959
 960                         OBD_ALLOC_PTR(oqctl_tmp);
 961                         if (oqctl_tmp == NULL)
 962                                 GOTO(out, rc = -ENOMEM);
 963
 964                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
 965                         oqctl_tmp->qc_id = oqctl->qc_id;
 966                         oqctl_tmp->qc_type = oqctl->qc_type;
 967
 968                         /* collect space usage from OSTs */
 969                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 970                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
 971                         if (!rc || rc == -EREMOTEIO) {
 972                                 oqctl->qc_dqblk.dqb_curspace =
 973                                         oqctl_tmp->qc_dqblk.dqb_curspace;
 974                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
 975                         }
 976
 977                         /* collect space & inode usage from MDTs */
 978                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 979                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
 980                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
 981                         if (!rc || rc == -EREMOTEIO) {
 982                                 oqctl->qc_dqblk.dqb_curspace +=
 983                                         oqctl_tmp->qc_dqblk.dqb_curspace;
 984                                 oqctl->qc_dqblk.dqb_curinodes =
 985                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
 986                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
 987                         } else {
 988                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
 989                         }
 990
 991                         OBD_FREE_PTR(oqctl_tmp);
 992                 }
 993 out:
 994                 QCTL_COPY(qctl, oqctl);
 995                 OBD_FREE_PTR(oqctl);
 996         }
 997
 998         RETURN(rc);
 999 }
1000
1001 static char *
1002 ll_getname(const char __user *filename)
1003 {
1004         int ret = 0, len;
1005         char *tmp = __getname();
1006
1007         if (!tmp)
1008                 return ERR_PTR(-ENOMEM);
1009
1010         len = strncpy_from_user(tmp, filename, PATH_MAX);
1011         if (len == 0)
1012                 ret = -ENOENT;
1013         else if (len > PATH_MAX)
1014                 ret = -ENAMETOOLONG;
1015
1016         if (ret) {
1017                 __putname(tmp);
1018                 tmp =  ERR_PTR(ret);
1019         }
1020         return tmp;
1021 }
1022
1023 #define ll_putname(filename) __putname(filename)
1024
1025 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1026 {
1027         struct inode *inode = file->f_dentry->d_inode;
1028         struct ll_sb_info *sbi = ll_i2sbi(inode);
1029         struct obd_ioctl_data *data;
1030         int rc = 0;
1031         ENTRY;
1032
1033         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1034                PFID(ll_inode2fid(inode)), inode, cmd);
1035
1036         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1037         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1038                 return -ENOTTY;
1039
1040         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1041         switch(cmd) {
1042         case FSFILT_IOC_GETFLAGS:
1043         case FSFILT_IOC_SETFLAGS:
1044                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1045         case FSFILT_IOC_GETVERSION_OLD:
1046         case FSFILT_IOC_GETVERSION:
1047                 RETURN(put_user(inode->i_generation, (int *)arg));
1048         /* We need to special case any other ioctls we want to handle,
1049          * to send them to the MDS/OST as appropriate and to properly
1050          * network encode the arg field.
1051         case FSFILT_IOC_SETVERSION_OLD:
1052         case FSFILT_IOC_SETVERSION:
1053         */
1054         case LL_IOC_GET_MDTIDX: {
1055                 int mdtidx;
1056
1057                 mdtidx = ll_get_mdt_idx(inode);
1058                 if (mdtidx < 0)
1059                         RETURN(mdtidx);
1060
1061                 if (put_user((int)mdtidx, (int*)arg))
1062                         RETURN(-EFAULT);
1063
1064                 return 0;
1065         }
1066         case IOC_MDC_LOOKUP: {
1067                 int namelen, len = 0;
1068                 char *buf = NULL;
1069                 char *filename;
1070
1071                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1072                 if (rc != 0)
1073                         RETURN(rc);
1074                 data = (void *)buf;
1075
1076                 filename = data->ioc_inlbuf1;
1077                 namelen = strlen(filename);
1078                 if (namelen < 1) {
1079                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1080                         GOTO(out_free, rc = -EINVAL);
1081                 }
1082
1083                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL);
1084                 if (rc < 0) {
1085                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1086                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1087                                filename, rc);
1088                         GOTO(out_free, rc);
1089                 }
1090 out_free:
1091                 obd_ioctl_freedata(buf, len);
1092                 return rc;
1093         }
1094         case LL_IOC_LMV_SETSTRIPE: {
1095                 struct lmv_user_md  *lum;
1096                 char            *buf = NULL;
1097                 char            *filename;
1098                 int              namelen = 0;
1099                 int              lumlen = 0;
1100                 int              len;
1101                 int              rc;
1102
1103                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1104                 if (rc)
1105                         RETURN(rc);
1106
1107                 data = (void *)buf;
1108                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1109                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1110                         GOTO(lmv_out_free, rc = -EINVAL);
1111
1112                 filename = data->ioc_inlbuf1;
1113                 namelen = data->ioc_inllen1;
1114
1115                 if (namelen < 1) {
1116                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1117                         GOTO(lmv_out_free, rc = -EINVAL);
1118                 }
1119                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1120                 lumlen = data->ioc_inllen2;
1121
1122                 if (lum->lum_magic != LMV_USER_MAGIC ||
1123                     lumlen != sizeof(*lum)) {
1124                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1125                                filename, lum->lum_magic, lumlen, -EFAULT);
1126                         GOTO(lmv_out_free, rc = -EINVAL);
1127                 }
1128
1129                 /**
1130                  * ll_dir_setdirstripe will be used to set dir stripe
1131                  *  mdc_create--->mdt_reint_create (with dirstripe)
1132                  */
1133                 rc = ll_dir_setdirstripe(inode, lum, filename);
1134 lmv_out_free:
1135                 obd_ioctl_freedata(buf, len);
1136                 RETURN(rc);
1137
1138         }
1139         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1140                 struct lmv_user_md        lum;
1141                 struct lmv_user_md __user *ulump =
1142                                         (struct lmv_user_md __user *)arg;
1143                 int                       rc;
1144
1145                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1146                         RETURN(-EFAULT);
1147
1148                 if (lum.lum_magic != LMV_USER_MAGIC)
1149                         RETURN(-EINVAL);
1150
1151                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1152
1153                 RETURN(rc);
1154         }
1155         case LL_IOC_LOV_SETSTRIPE: {
1156                 struct lov_user_md_v3 lumv3;
1157                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1158                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1159                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1160
1161                 int set_default = 0;
1162
1163                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1164                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1165                         sizeof(lumv3p->lmm_objects[0]));
1166                 /* first try with v1 which is smaller than v3 */
1167                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1168                         RETURN(-EFAULT);
1169
1170                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1171                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1172                                 RETURN(-EFAULT);
1173                 }
1174
1175                 if (inode->i_sb->s_root == file->f_dentry)
1176                         set_default = 1;
1177
1178                 /* in v1 and v3 cases lumv1 points to data */
1179                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1180
1181                 RETURN(rc);
1182         }
1183         case LL_IOC_LMV_GETSTRIPE: {
1184                 struct lmv_user_md __user *ulmv =
1185                                         (struct lmv_user_md __user *)arg;
1186                 struct lmv_user_md      lum;
1187                 struct ptlrpc_request   *request = NULL;
1188                 union lmv_mds_md        *lmm = NULL;
1189                 int                     lmmsize;
1190                 obd_valid               valid = 0;
1191                 struct lmv_user_md      *tmp = NULL;
1192                 int                     mdt_index;
1193                 int                     lum_size;
1194                 int                     stripe_count;
1195                 int                     i;
1196                 int                     rc;
1197
1198                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1199                         RETURN(-EFAULT);
1200
1201                 /* lum_magic will indicate which stripe the ioctl will like
1202                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1203                  * is for default LMV stripe */
1204                 if (lum.lum_magic == LMV_MAGIC_V1)
1205                         valid |= OBD_MD_MEA;
1206                 else if (lum.lum_magic == LMV_USER_MAGIC)
1207                         valid |= OBD_MD_DEFAULT_MEA;
1208                 else
1209                         RETURN(-EINVAL);
1210
1211                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1212                                       valid);
1213                 if (rc != 0)
1214                         GOTO(finish_req, rc);
1215
1216                 /* Get default LMV EA */
1217                 if (lum.lum_magic == LMV_USER_MAGIC) {
1218                         if (rc != 0)
1219                                 GOTO(finish_req, rc);
1220
1221                         if (lmmsize > sizeof(*ulmv))
1222                                 GOTO(finish_req, rc = -EINVAL);
1223
1224                         if (copy_to_user(ulmv, lmm, lmmsize))
1225                                 GOTO(finish_req, rc = -EFAULT);
1226
1227                         GOTO(finish_req, rc);
1228                 }
1229
1230                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1231                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1232                 OBD_ALLOC(tmp, lum_size);
1233                 if (tmp == NULL)
1234                         GOTO(finish_req, rc = -ENOMEM);
1235
1236                 mdt_index = ll_get_mdt_idx(inode);
1237                 if (mdt_index < 0)
1238                         GOTO(out_tmp, rc = -ENOMEM);
1239
1240                 tmp->lum_magic = LMV_MAGIC_V1;
1241                 tmp->lum_stripe_count = 0;
1242                 tmp->lum_stripe_offset = mdt_index;
1243                 for (i = 0; i < stripe_count; i++) {
1244                         struct lu_fid   *fid;
1245
1246                         fid = &lmm->lmv_md_v1.lmv_stripe_fids[i];
1247                         mdt_index = ll_get_mdt_idx_by_fid(sbi, fid);
1248                         if (mdt_index < 0)
1249                                 GOTO(out_tmp, rc = mdt_index);
1250
1251                         tmp->lum_objects[i].lum_mds = mdt_index;
1252                         tmp->lum_objects[i].lum_fid = *fid;
1253                         tmp->lum_stripe_count++;
1254                 }
1255
1256                 if (copy_to_user(ulmv, tmp, lum_size))
1257                         GOTO(out_tmp, rc = -EFAULT);
1258 out_tmp:
1259                 OBD_FREE(tmp, lum_size);
1260 finish_req:
1261                 ptlrpc_req_finished(request);
1262                 return rc;
1263         }
1264
1265         case LL_IOC_REMOVE_ENTRY: {
1266                 char            *filename = NULL;
1267                 int              namelen = 0;
1268                 int              rc;
1269
1270                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1271                  * unsupported server, which might crash the server(LU-2730),
1272                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1273                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1274                  * server will support REINT_RMENTRY XXX*/
1275                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1276                         RETURN(-ENOTSUPP);
1277
1278                 filename = ll_getname((const char *)arg);
1279                 if (IS_ERR(filename))
1280                         RETURN(PTR_ERR(filename));
1281
1282                 namelen = strlen(filename);
1283                 if (namelen < 1)
1284                         GOTO(out_rmdir, rc = -EINVAL);
1285
1286                 rc = ll_rmdir_entry(inode, filename, namelen);
1287 out_rmdir:
1288                 if (filename)
1289                         ll_putname(filename);
1290                 RETURN(rc);
1291         }
1292         case LL_IOC_LOV_SWAP_LAYOUTS:
1293                 RETURN(-EPERM);
1294         case IOC_OBD_STATFS:
1295                 RETURN(ll_obd_statfs(inode, (void *)arg));
1296         case LL_IOC_LOV_GETSTRIPE:
1297         case LL_IOC_MDC_GETINFO:
1298         case IOC_MDC_GETFILEINFO:
1299         case IOC_MDC_GETFILESTRIPE: {
1300                 struct ptlrpc_request *request = NULL;
1301                 struct lov_user_md *lump;
1302                 struct lov_mds_md *lmm = NULL;
1303                 struct mdt_body *body;
1304                 char *filename = NULL;
1305                 int lmmsize;
1306
1307                 if (cmd == IOC_MDC_GETFILEINFO ||
1308                     cmd == IOC_MDC_GETFILESTRIPE) {
1309                         filename = ll_getname((const char *)arg);
1310                         if (IS_ERR(filename))
1311                                 RETURN(PTR_ERR(filename));
1312
1313                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1314                                                       &lmmsize, &request);
1315                 } else {
1316                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1317                                               &request, 0);
1318                 }
1319
1320                 if (request) {
1321                         body = req_capsule_server_get(&request->rq_pill,
1322                                                       &RMF_MDT_BODY);
1323                         LASSERT(body != NULL);
1324                 } else {
1325                         GOTO(out_req, rc);
1326                 }
1327
1328                 if (rc < 0) {
1329                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1330                                                cmd == LL_IOC_MDC_GETINFO))
1331                                 GOTO(skip_lmm, rc = 0);
1332                         else
1333                                 GOTO(out_req, rc);
1334                 }
1335
1336                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1337                     cmd == LL_IOC_LOV_GETSTRIPE) {
1338                         lump = (struct lov_user_md *)arg;
1339                 } else {
1340                         struct lov_user_mds_data *lmdp;
1341                         lmdp = (struct lov_user_mds_data *)arg;
1342                         lump = &lmdp->lmd_lmm;
1343                 }
1344                 if (copy_to_user(lump, lmm, lmmsize)) {
1345                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1346                                 GOTO(out_req, rc = -EFAULT);
1347                         rc = -EOVERFLOW;
1348                 }
1349         skip_lmm:
1350                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1351                         struct lov_user_mds_data *lmdp;
1352                         lstat_t st = { 0 };
1353
1354                         st.st_dev     = inode->i_sb->s_dev;
1355                         st.st_mode    = body->mode;
1356                         st.st_nlink   = body->nlink;
1357                         st.st_uid     = body->uid;
1358                         st.st_gid     = body->gid;
1359                         st.st_rdev    = body->rdev;
1360                         st.st_size    = body->size;
1361                         st.st_blksize = PAGE_CACHE_SIZE;
1362                         st.st_blocks  = body->blocks;
1363                         st.st_atime   = body->atime;
1364                         st.st_mtime   = body->mtime;
1365                         st.st_ctime   = body->ctime;
1366                         st.st_ino     = inode->i_ino;
1367
1368                         lmdp = (struct lov_user_mds_data *)arg;
1369                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1370                                 GOTO(out_req, rc = -EFAULT);
1371                 }
1372
1373                 EXIT;
1374         out_req:
1375                 ptlrpc_req_finished(request);
1376                 if (filename)
1377                         ll_putname(filename);
1378                 return rc;
1379         }
1380         case IOC_LOV_GETINFO: {
1381                 struct lov_user_mds_data *lumd;
1382                 struct lov_stripe_md *lsm;
1383                 struct lov_user_md *lum;
1384                 struct lov_mds_md *lmm;
1385                 int lmmsize;
1386                 lstat_t st;
1387
1388                 lumd = (struct lov_user_mds_data *)arg;
1389                 lum = &lumd->lmd_lmm;
1390
1391                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1392                 if (rc)
1393                         RETURN(rc);
1394
1395                 OBD_ALLOC_LARGE(lmm, lmmsize);
1396                 if (lmm == NULL)
1397                         RETURN(-ENOMEM);
1398
1399                 if (copy_from_user(lmm, lum, lmmsize))
1400                         GOTO(free_lmm, rc = -EFAULT);
1401
1402                 switch (lmm->lmm_magic) {
1403                 case LOV_USER_MAGIC_V1:
1404                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1405                                 break;
1406                         /* swab objects first so that stripes num will be sane */
1407                         lustre_swab_lov_user_md_objects(
1408                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1409                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1410                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1411                         break;
1412                 case LOV_USER_MAGIC_V3:
1413                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1414                                 break;
1415                         /* swab objects first so that stripes num will be sane */
1416                         lustre_swab_lov_user_md_objects(
1417                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1418                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1419                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1420                         break;
1421                 default:
1422                         GOTO(free_lmm, rc = -EINVAL);
1423                 }
1424
1425                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1426                 if (rc < 0)
1427                         GOTO(free_lmm, rc = -ENOMEM);
1428
1429                 /* Perform glimpse_size operation. */
1430                 memset(&st, 0, sizeof(st));
1431
1432                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1433                 if (rc)
1434                         GOTO(free_lsm, rc);
1435
1436                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1437                         GOTO(free_lsm, rc = -EFAULT);
1438
1439                 EXIT;
1440         free_lsm:
1441                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1442         free_lmm:
1443                 OBD_FREE_LARGE(lmm, lmmsize);
1444                 return rc;
1445         }
1446         case OBD_IOC_QUOTACHECK: {
1447                 struct obd_quotactl *oqctl;
1448                 int error = 0;
1449
1450                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1451                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1452                         RETURN(-EPERM);
1453
1454                 OBD_ALLOC_PTR(oqctl);
1455                 if (!oqctl)
1456                         RETURN(-ENOMEM);
1457                 oqctl->qc_type = arg;
1458                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1459                 if (rc < 0) {
1460                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1461                         error = rc;
1462                 }
1463
1464                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1465                 if (rc < 0)
1466                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1467
1468                 OBD_FREE_PTR(oqctl);
1469                 return error ?: rc;
1470         }
1471         case OBD_IOC_POLL_QUOTACHECK: {
1472                 struct if_quotacheck *check;
1473
1474                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1475                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1476                         RETURN(-EPERM);
1477
1478                 OBD_ALLOC_PTR(check);
1479                 if (!check)
1480                         RETURN(-ENOMEM);
1481
1482                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1483                                    NULL);
1484                 if (rc) {
1485                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1486                         if (copy_to_user((void *)arg, check,
1487                                              sizeof(*check)))
1488                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1489                         GOTO(out_poll, rc);
1490                 }
1491
1492                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1493                                    NULL);
1494                 if (rc) {
1495                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1496                         if (copy_to_user((void *)arg, check,
1497                                              sizeof(*check)))
1498                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1499                         GOTO(out_poll, rc);
1500                 }
1501         out_poll:
1502                 OBD_FREE_PTR(check);
1503                 RETURN(rc);
1504         }
1505 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1506         case LL_IOC_QUOTACTL_18: {
1507                 /* copy the old 1.x quota struct for internal use, then copy
1508                  * back into old format struct.  For 1.8 compatibility. */
1509                 struct if_quotactl_18 *qctl_18;
1510                 struct if_quotactl *qctl_20;
1511
1512                 OBD_ALLOC_PTR(qctl_18);
1513                 if (!qctl_18)
1514                         RETURN(-ENOMEM);
1515
1516                 OBD_ALLOC_PTR(qctl_20);
1517                 if (!qctl_20)
1518                         GOTO(out_quotactl_18, rc = -ENOMEM);
1519
1520                 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1521                         GOTO(out_quotactl_20, rc = -ENOMEM);
1522
1523                 QCTL_COPY(qctl_20, qctl_18);
1524                 qctl_20->qc_idx = 0;
1525
1526                 /* XXX: dqb_valid was borrowed as a flag to mark that
1527                  *      only mds quota is wanted */
1528                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1529                     qctl_18->qc_dqblk.dqb_valid) {
1530                         qctl_20->qc_valid = QC_MDTIDX;
1531                         qctl_20->qc_dqblk.dqb_valid = 0;
1532                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1533                         qctl_20->qc_valid = QC_UUID;
1534                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1535                 } else {
1536                         qctl_20->qc_valid = QC_GENERAL;
1537                 }
1538
1539                 rc = quotactl_ioctl(sbi, qctl_20);
1540
1541                 if (rc == 0) {
1542                         QCTL_COPY(qctl_18, qctl_20);
1543                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1544
1545                         if (copy_to_user((void *)arg, qctl_18,
1546                                              sizeof(*qctl_18)))
1547                                 rc = -EFAULT;
1548                 }
1549
1550         out_quotactl_20:
1551                 OBD_FREE_PTR(qctl_20);
1552         out_quotactl_18:
1553                 OBD_FREE_PTR(qctl_18);
1554                 RETURN(rc);
1555         }
1556 #else
1557 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1558 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1559         case OBD_IOC_QUOTACTL: {
1560                 struct if_quotactl *qctl;
1561
1562                 OBD_ALLOC_PTR(qctl);
1563                 if (!qctl)
1564                         RETURN(-ENOMEM);
1565
1566                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1567                         GOTO(out_quotactl, rc = -EFAULT);
1568
1569                 rc = quotactl_ioctl(sbi, qctl);
1570
1571                 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1572                         rc = -EFAULT;
1573
1574         out_quotactl:
1575                 OBD_FREE_PTR(qctl);
1576                 RETURN(rc);
1577         }
1578         case OBD_IOC_GETDTNAME:
1579         case OBD_IOC_GETMDNAME:
1580                 RETURN(ll_get_obd_name(inode, cmd, arg));
1581         case LL_IOC_FLUSHCTX:
1582                 RETURN(ll_flush_ctx(inode));
1583 #ifdef CONFIG_FS_POSIX_ACL
1584         case LL_IOC_RMTACL: {
1585             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1586                 inode == inode->i_sb->s_root->d_inode) {
1587                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1588
1589                 LASSERT(fd != NULL);
1590                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1591                 if (!rc)
1592                         fd->fd_flags |= LL_FILE_RMTACL;
1593                 RETURN(rc);
1594             } else
1595                 RETURN(0);
1596         }
1597 #endif
1598         case LL_IOC_GETOBDCOUNT: {
1599                 int count, vallen;
1600                 struct obd_export *exp;
1601
1602                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1603                         RETURN(-EFAULT);
1604
1605                 /* get ost count when count is zero, get mdt count otherwise */
1606                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1607                 vallen = sizeof(count);
1608                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1609                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1610                 if (rc) {
1611                         CERROR("get target count failed: %d\n", rc);
1612                         RETURN(rc);
1613                 }
1614
1615                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1616                         RETURN(-EFAULT);
1617
1618                 RETURN(0);
1619         }
1620         case LL_IOC_PATH2FID:
1621                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1622                                      sizeof(struct lu_fid)))
1623                         RETURN(-EFAULT);
1624                 RETURN(0);
1625         case LL_IOC_GET_CONNECT_FLAGS: {
1626                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1627         }
1628         case OBD_IOC_CHANGELOG_SEND:
1629         case OBD_IOC_CHANGELOG_CLEAR:
1630                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1631                                     sizeof(struct ioc_changelog));
1632                 RETURN(rc);
1633         case OBD_IOC_FID2PATH:
1634                 RETURN(ll_fid2path(inode, (void *)arg));
1635         case LL_IOC_HSM_REQUEST: {
1636                 struct hsm_user_request *hur;
1637                 int                      totalsize;
1638
1639                 OBD_ALLOC_PTR(hur);
1640                 if (hur == NULL)
1641                         RETURN(-ENOMEM);
1642
1643                 /* We don't know the true size yet; copy the fixed-size part */
1644                 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1645                         OBD_FREE_PTR(hur);
1646                         RETURN(-EFAULT);
1647                 }
1648
1649                 /* Compute the whole struct size */
1650                 totalsize = hur_len(hur);
1651                 OBD_FREE_PTR(hur);
1652
1653                 /* Final size will be more than double totalsize */
1654                 if (totalsize >= MDS_MAXREQSIZE / 3)
1655                         RETURN(-E2BIG);
1656
1657                 OBD_ALLOC_LARGE(hur, totalsize);
1658                 if (hur == NULL)
1659                         RETURN(-ENOMEM);
1660
1661                 /* Copy the whole struct */
1662                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1663                         OBD_FREE_LARGE(hur, totalsize);
1664                         RETURN(-EFAULT);
1665                 }
1666
1667                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1668                         const struct lu_fid *fid;
1669                         struct inode *f;
1670                         int i;
1671
1672                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1673                                 fid = &hur->hur_user_item[i].hui_fid;
1674                                 f = search_inode_for_lustre(inode->i_sb, fid);
1675                                 if (IS_ERR(f)) {
1676                                         rc = PTR_ERR(f);
1677                                         break;
1678                                 }
1679
1680                                 rc = ll_hsm_release(f);
1681                                 iput(f);
1682                                 if (rc != 0)
1683                                         break;
1684                         }
1685                 } else {
1686                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1687                                            hur, NULL);
1688                 }
1689
1690                 OBD_FREE_LARGE(hur, totalsize);
1691
1692                 RETURN(rc);
1693         }
1694         case LL_IOC_HSM_PROGRESS: {
1695                 struct hsm_progress_kernel      hpk;
1696                 struct hsm_progress             hp;
1697
1698                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1699                         RETURN(-EFAULT);
1700
1701                 hpk.hpk_fid = hp.hp_fid;
1702                 hpk.hpk_cookie = hp.hp_cookie;
1703                 hpk.hpk_extent = hp.hp_extent;
1704                 hpk.hpk_flags = hp.hp_flags;
1705                 hpk.hpk_errval = hp.hp_errval;
1706                 hpk.hpk_data_version = 0;
1707
1708                 /* File may not exist in Lustre; all progress
1709                  * reported to Lustre root */
1710                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1711                                    NULL);
1712                 RETURN(rc);
1713         }
1714         case LL_IOC_HSM_CT_START:
1715                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1716                         RETURN(-EPERM);
1717
1718                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1719                                     sizeof(struct lustre_kernelcomm));
1720                 RETURN(rc);
1721
1722         case LL_IOC_HSM_COPY_START: {
1723                 struct hsm_copy *copy;
1724                 int              rc;
1725
1726                 OBD_ALLOC_PTR(copy);
1727                 if (copy == NULL)
1728                         RETURN(-ENOMEM);
1729                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1730                         OBD_FREE_PTR(copy);
1731                         RETURN(-EFAULT);
1732                 }
1733
1734                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1735                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1736                         rc = -EFAULT;
1737
1738                 OBD_FREE_PTR(copy);
1739                 RETURN(rc);
1740         }
1741         case LL_IOC_HSM_COPY_END: {
1742                 struct hsm_copy *copy;
1743                 int              rc;
1744
1745                 OBD_ALLOC_PTR(copy);
1746                 if (copy == NULL)
1747                         RETURN(-ENOMEM);
1748                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1749                         OBD_FREE_PTR(copy);
1750                         RETURN(-EFAULT);
1751                 }
1752
1753                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1754                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1755                         rc = -EFAULT;
1756
1757                 OBD_FREE_PTR(copy);
1758                 RETURN(rc);
1759         }
1760         case LL_IOC_MIGRATE: {
1761                 char            *buf = NULL;
1762                 const char      *filename;
1763                 int             namelen = 0;
1764                 int             len;
1765                 int             rc;
1766                 int             mdtidx;
1767
1768                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1769                 if (rc < 0)
1770                         RETURN(rc);
1771
1772                 data = (struct obd_ioctl_data *)buf;
1773                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1774                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1775                         GOTO(migrate_free, rc = -EINVAL);
1776
1777                 filename = data->ioc_inlbuf1;
1778                 namelen = data->ioc_inllen1;
1779                 /* \0 is packed at the end of filename */
1780                 if (namelen < 1 || namelen != strlen(filename) + 1)
1781                         GOTO(migrate_free, rc = -EINVAL);
1782
1783                 if (data->ioc_inllen2 != sizeof(mdtidx))
1784                         GOTO(migrate_free, rc = -EINVAL);
1785                 mdtidx = *(int *)data->ioc_inlbuf2;
1786
1787                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1788 migrate_free:
1789                 obd_ioctl_freedata(buf, len);
1790
1791                 RETURN(rc);
1792         }
1793         default:
1794                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1795                                      (void *)arg));
1796         }
1797 }
1798
1799 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1800 {
1801         struct inode *inode = file->f_mapping->host;
1802         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1803         struct ll_sb_info *sbi = ll_i2sbi(inode);
1804         int api32 = ll_need_32bit_api(sbi);
1805         loff_t ret = -EINVAL;
1806         ENTRY;
1807
1808         mutex_lock(&inode->i_mutex);
1809         switch (origin) {
1810                 case SEEK_SET:
1811                         break;
1812                 case SEEK_CUR:
1813                         offset += file->f_pos;
1814                         break;
1815                 case SEEK_END:
1816                         if (offset > 0)
1817                                 GOTO(out, ret);
1818                         if (api32)
1819                                 offset += LL_DIR_END_OFF_32BIT;
1820                         else
1821                                 offset += LL_DIR_END_OFF;
1822                         break;
1823                 default:
1824                         GOTO(out, ret);
1825         }
1826
1827         if (offset >= 0 &&
1828             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1829              (!api32 && offset <= LL_DIR_END_OFF))) {
1830                 if (offset != file->f_pos) {
1831                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1832                             (!api32 && offset == LL_DIR_END_OFF))
1833                                 fd->lfd_pos = MDS_DIR_END_OFF;
1834                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1835                                 fd->lfd_pos = offset << 32;
1836                         else
1837                                 fd->lfd_pos = offset;
1838                         file->f_pos = offset;
1839                         file->f_version = 0;
1840                 }
1841                 ret = offset;
1842         }
1843         GOTO(out, ret);
1844
1845 out:
1846         mutex_unlock(&inode->i_mutex);
1847         return ret;
1848 }
1849
1850 static int ll_dir_open(struct inode *inode, struct file *file)
1851 {
1852         ENTRY;
1853         RETURN(ll_file_open(inode, file));
1854 }
1855
1856 static int ll_dir_release(struct inode *inode, struct file *file)
1857 {
1858         ENTRY;
1859         RETURN(ll_file_release(inode, file));
1860 }
1861
1862 const struct file_operations ll_dir_operations = {
1863         .llseek         = ll_dir_seek,
1864         .open           = ll_dir_open,
1865         .release        = ll_dir_release,
1866         .read           = generic_read_dir,
1867 #ifdef HAVE_DIR_CONTEXT
1868         .iterate        = ll_iterate,
1869 #else
1870         .readdir        = ll_readdir,
1871 #endif
1872         .unlocked_ioctl = ll_dir_ioctl,
1873         .fsync          = ll_fsync,
1874 };