lustre/llite/dir.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/dir.c
  33  *
  34  * Directory code for lustre client.
  35  */
  36
  37 #include <linux/fs.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/mm.h>
  40 #include <linux/version.h>
  41 #include <linux/security.h>
  42 #include <linux/user_namespace.h>
  43 #ifdef HAVE_UIDGID_HEADER
  44 # include <linux/uidgid.h>
  45 #endif
  46 #include <linux/uaccess.h>
  47 #include <linux/buffer_head.h>   // for wait_on_buffer
  48 #include <linux/pagevec.h>
  49
  50 #define DEBUG_SUBSYSTEM S_LLITE
  51
  52 #include <obd_support.h>
  53 #include <obd_class.h>
  54 #include <uapi/linux/lustre/lustre_ioctl.h>
  55 #include <lustre_lib.h>
  56 #include <lustre_dlm.h>
  57 #include <lustre_fid.h>
  58 #include <lustre_kernelcomm.h>
  59 #include <lustre_swab.h>
  60
  61 #include "llite_internal.h"
  62
  63 /*
  64  * (new) readdir implementation overview.
  65  *
  66  * Original lustre readdir implementation cached exact copy of raw directory
  67  * pages on the client. These pages were indexed in client page cache by
  68  * logical offset in the directory file. This design, while very simple and
  69  * intuitive had some inherent problems:
  70  *
  71  *     . it implies that byte offset to the directory entry serves as a
  72  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  73  *     ext3/htree directory entries may move due to splits, and more
  74  *     importantly,
  75  *
  76  *     . it is incompatible with the design of split directories for cmd3,
  77  *     that assumes that names are distributed across nodes based on their
  78  *     hash, and so readdir should be done in hash order.
  79  *
  80  * New readdir implementation does readdir in hash order, and uses hash of a
  81  * file name as a telldir/seekdir cookie. This led to number of complications:
  82  *
  83  *     . hash is not unique, so it cannot be used to index cached directory
  84  *     pages on the client (note, that it requires a whole pageful of hash
  85  *     collided entries to cause two pages to have identical hashes);
  86  *
  87  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  88  *     entry cookie. ext3/htree has the same problem and lustre implementation
  89  *     mimics their solution: seekdir(hash) positions directory at the first
  90  *     entry with the given hash.
  91  *
  92  * Client side.
  93  *
  94  * 0. caching
  95  *
  96  * Client caches directory pages using hash of the first entry as an index. As
  97  * noted above hash is not unique, so this solution doesn't work as is:
  98  * special processing is needed for "page hash chains" (i.e., sequences of
  99  * pages filled with entries all having the same hash value).
 100  *
 101  * First, such chains have to be detected. To this end, server returns to the
 102  * client the hash of the first entry on the page next to one returned. When
 103  * client detects that this hash is the same as hash of the first entry on the
 104  * returned page, page hash collision has to be handled. Pages in the
 105  * hash chain, except first one, are termed "overflow pages".
 106  *
 107  * Solution to index uniqueness problem is to not cache overflow
 108  * pages. Instead, when page hash collision is detected, all overflow pages
 109  * from emerging chain are immediately requested from the server and placed in
 110  * a special data structure (struct ll_dir_chain). This data structure is used
 111  * by ll_readdir() to process entries from overflow pages. When readdir
 112  * invocation finishes, overflow pages are discarded. If page hash collision
 113  * chain weren't completely processed, next call to readdir will again detect
 114  * page hash collision, again read overflow pages in, process next portion of
 115  * entries and again discard the pages. This is not as wasteful as it looks,
 116  * because, given reasonable hash, page hash collisions are extremely rare.
 117  *
 118  * 1. directory positioning
 119  *
 120  * When seekdir(hash) is called, original
 121  *
 122  *
 123  *
 124  *
 125  *
 126  *
 127  *
 128  *
 129  * Server.
 130  *
 131  * identification of and access to overflow pages
 132  *
 133  * page format
 134  *
 135  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 136  * a header lu_dirpage which describes the start/end hash, and whether this
 137  * page is empty (contains no dir entry) or hash collide with next page.
 138  * After client receives reply, several pages will be integrated into dir page
 139  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
 140  * lu_dirpage for this integrated page will be adjusted. See
 141  * mdc_adjust_dirpages().
 142  *
 143  */
 144 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
 145                              __u64 offset, struct ll_dir_chain *chain)
 146 {
 147         struct md_callback      cb_op;
 148         struct page             *page;
 149         int                     rc;
 150
 151         cb_op.md_blocking_ast = ll_md_blocking_ast;
 152         rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
 153         if (rc != 0)
 154                 return ERR_PTR(rc);
 155
 156         return page;
 157 }
 158
 159 void ll_release_page(struct inode *inode, struct page *page,
 160                      bool remove)
 161 {
 162         kunmap(page);
 163
 164         /* Always remove the page for striped dir, because the page is
 165          * built from temporarily in LMV layer */
 166         if (inode != NULL && S_ISDIR(inode->i_mode) &&
 167             ll_i2info(inode)->lli_lsm_md != NULL) {
 168                 __free_page(page);
 169                 return;
 170         }
 171
 172         if (remove) {
 173                 lock_page(page);
 174                 if (likely(page->mapping != NULL))
 175                         truncate_complete_page(page->mapping, page);
 176                 unlock_page(page);
 177         }
 178         put_page(page);
 179 }
 180
 181 /**
 182  * return IF_* type for given lu_dirent entry.
 183  * IF_* flag shld be converted to particular OS file type in
 184  * platform llite module.
 185  */
 186 static u16 ll_dirent_type_get(struct lu_dirent *ent)
 187 {
 188         u16 type = 0;
 189         struct luda_type *lt;
 190         int len = 0;
 191
 192         if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
 193                 const unsigned align = sizeof(struct luda_type) - 1;
 194
 195                 len = le16_to_cpu(ent->lde_namelen);
 196                 len = (len + align) & ~align;
 197                 lt = (void *)ent->lde_name + len;
 198                 type = IFTODT(le16_to_cpu(lt->lt_type));
 199         }
 200
 201         return type;
 202 }
 203
 204 #ifdef HAVE_DIR_CONTEXT
 205 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 206                 struct dir_context *ctx)
 207 {
 208 #else
 209 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 210                 void *cookie, filldir_t filldir)
 211 {
 212 #endif
 213         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 214         __u64                 pos        = *ppos;
 215         bool                  is_api32 = ll_need_32bit_api(sbi);
 216         bool                  is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 217         struct page          *page;
 218         struct ll_dir_chain   chain;
 219         bool                  done = false;
 220         int                   rc = 0;
 221         ENTRY;
 222
 223         ll_dir_chain_init(&chain);
 224
 225         page = ll_get_dir_page(inode, op_data, pos, &chain);
 226
 227         while (rc == 0 && !done) {
 228                 struct lu_dirpage *dp;
 229                 struct lu_dirent  *ent;
 230                 __u64 hash;
 231                 __u64 next;
 232
 233                 if (IS_ERR(page)) {
 234                         rc = PTR_ERR(page);
 235                         break;
 236                 }
 237
 238                 hash = MDS_DIR_END_OFF;
 239                 dp = page_address(page);
 240                 for (ent = lu_dirent_start(dp); ent != NULL && !done;
 241                      ent = lu_dirent_next(ent)) {
 242                         __u16          type;
 243                         int            namelen;
 244                         struct lu_fid  fid;
 245                         __u64          lhash;
 246                         __u64          ino;
 247
 248                         hash = le64_to_cpu(ent->lde_hash);
 249                         if (hash < pos) /* Skip until we find target hash */
 250                                 continue;
 251
 252                         namelen = le16_to_cpu(ent->lde_namelen);
 253                         if (namelen == 0) /* Skip dummy record */
 254                                 continue;
 255
 256                         if (is_api32 && is_hash64)
 257                                 lhash = hash >> 32;
 258                         else
 259                                 lhash = hash;
 260                         fid_le_to_cpu(&fid, &ent->lde_fid);
 261                         ino = cl_fid_build_ino(&fid, is_api32);
 262                         type = ll_dirent_type_get(ent);
 263                         /* For ll_nfs_get_name_filldir(), it will try to access
 264                          * 'ent' through 'lde_name', so the parameter 'name'
 265                          * for 'filldir()' must be part of the 'ent'. */
 266 #ifdef HAVE_DIR_CONTEXT
 267                         ctx->pos = lhash;
 268                         done = !dir_emit(ctx, ent->lde_name, namelen, ino,
 269                                          type);
 270 #else
 271                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 272                                        ino, type);
 273 #endif
 274                 }
 275
 276                 if (done) {
 277                         pos = hash;
 278                         ll_release_page(inode, page, false);
 279                         break;
 280                 }
 281
 282                 next = le64_to_cpu(dp->ldp_hash_end);
 283                 pos = next;
 284                 if (pos == MDS_DIR_END_OFF) {
 285                         /*
 286                          * End of directory reached.
 287                          */
 288                         done = 1;
 289                         ll_release_page(inode, page, false);
 290                 } else {
 291                         /*
 292                          * Normal case: continue to the next
 293                          * page.
 294                          */
 295                         ll_release_page(inode, page,
 296                                         le32_to_cpu(dp->ldp_flags) &
 297                                         LDF_COLLIDE);
 298                         next = pos;
 299                         page = ll_get_dir_page(inode, op_data, pos,
 300                                                &chain);
 301                 }
 302         }
 303 #ifdef HAVE_DIR_CONTEXT
 304         ctx->pos = pos;
 305 #else
 306         *ppos = pos;
 307 #endif
 308         ll_dir_chain_fini(&chain);
 309         RETURN(rc);
 310 }
 311
 312 #ifdef HAVE_DIR_CONTEXT
 313 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 314 #else
 315 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 316 #endif
 317 {
 318         struct inode            *inode  = file_inode(filp);
 319         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 320         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 321         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 322         int                     api32   = ll_need_32bit_api(sbi);
 323         struct md_op_data       *op_data;
 324         struct lu_fid           pfid = { 0 };
 325         __u64                   pos;
 326         int                     rc;
 327         ENTRY;
 328
 329         if (lfd != NULL)
 330                 pos = lfd->lfd_pos;
 331         else
 332                 pos = 0;
 333
 334         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 335                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 336                inode, (unsigned long)pos, i_size_read(inode), api32);
 337
 338         if (pos == MDS_DIR_END_OFF)
 339                 /*
 340                  * end-of-file.
 341                  */
 342                 GOTO(out, rc = 0);
 343
 344         if (unlikely(ll_i2info(inode)->lli_lsm_md != NULL)) {
 345                 /* This is only needed for striped dir to fill ..,
 346                  * see lmv_read_entry */
 347                 if (file_dentry(filp)->d_parent != NULL &&
 348                     file_dentry(filp)->d_parent->d_inode != NULL) {
 349                         __u64 ibits = MDS_INODELOCK_UPDATE;
 350                         struct inode *parent =
 351                                 file_dentry(filp)->d_parent->d_inode;
 352
 353                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 354                                 pfid = *ll_inode2fid(parent);
 355                 }
 356
 357                 /* If it can not find in cache, do lookup .. on the master
 358                  * object */
 359                 if (fid_is_zero(&pfid)) {
 360                         rc = ll_dir_get_parent_fid(inode, &pfid);
 361                         if (rc != 0)
 362                                 RETURN(rc);
 363                 }
 364         }
 365
 366         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 367                                      LUSTRE_OPC_ANY, inode);
 368         if (IS_ERR(op_data))
 369                 GOTO(out, rc = PTR_ERR(op_data));
 370         op_data->op_fid3 = pfid;
 371
 372 #ifdef HAVE_DIR_CONTEXT
 373         ctx->pos = pos;
 374         rc = ll_dir_read(inode, &pos, op_data, ctx);
 375         pos = ctx->pos;
 376 #else
 377         rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
 378 #endif
 379         if (lfd != NULL)
 380                 lfd->lfd_pos = pos;
 381
 382         if (pos == MDS_DIR_END_OFF) {
 383                 if (api32)
 384                         pos = LL_DIR_END_OFF_32BIT;
 385                 else
 386                         pos = LL_DIR_END_OFF;
 387         } else {
 388                 if (api32 && hash64)
 389                         pos = pos >> 32;
 390         }
 391 #ifdef HAVE_DIR_CONTEXT
 392         ctx->pos = pos;
 393 #else
 394         filp->f_pos = pos;
 395 #endif
 396         ll_finish_md_op_data(op_data);
 397
 398 out:
 399         if (!rc)
 400                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 401
 402         RETURN(rc);
 403 }
 404
 405 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 406 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 407 {
 408         struct mgs_send_param *msp;
 409         int rc = 0;
 410
 411         OBD_ALLOC_PTR(msp);
 412         if (!msp)
 413                 return -ENOMEM;
 414
 415         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 416         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 417                                 sizeof(struct mgs_send_param), msp, NULL);
 418         if (rc)
 419                 CERROR("Failed to set parameter: %d\n", rc);
 420         OBD_FREE_PTR(msp);
 421
 422         return rc;
 423 }
 424 #endif
 425
 426 /**
 427  * Create striped directory with specified stripe(@lump)
 428  *
 429  * \param[in] dparent   the parent of the directory.
 430  * \param[in] lump      the specified stripes.
 431  * \param[in] dirname   the name of the directory.
 432  * \param[in] mode      the specified mode of the directory.
 433  *
 434  * \retval              =0 if striped directory is being created successfully.
 435  *                      <0 if the creation is failed.
 436  */
 437 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
 438                                size_t len, const char *dirname, umode_t mode)
 439 {
 440         struct inode *parent = dparent->d_inode;
 441         struct ptlrpc_request *request = NULL;
 442         struct md_op_data *op_data;
 443         struct ll_sb_info *sbi = ll_i2sbi(parent);
 444         struct inode *inode = NULL;
 445         struct dentry dentry = {
 446                 .d_parent = dparent,
 447                 .d_name = {
 448                         .name = dirname,
 449                         .len = strlen(dirname),
 450                         .hash = ll_full_name_hash(dparent, dirname,
 451                                                   strlen(dirname)),
 452                 },
 453         };
 454         int err;
 455         ENTRY;
 456
 457         if (unlikely(lump->lum_magic != LMV_USER_MAGIC &&
 458                      lump->lum_magic != LMV_USER_MAGIC_SPECIFIC))
 459                 RETURN(-EINVAL);
 460
 461         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s "
 462                "stripe_offset %d, stripe_count: %u\n",
 463                PFID(ll_inode2fid(parent)), parent, dirname,
 464                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 465
 466         if (lump->lum_stripe_count > 1 &&
 467             !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
 468                 RETURN(-EINVAL);
 469
 470         if (IS_DEADDIR(parent) &&
 471             !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
 472                 RETURN(-ENOENT);
 473
 474         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
 475             lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
 476                 lustre_swab_lmv_user_md(lump);
 477
 478         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 479                 mode &= ~current_umask();
 480         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 481         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 482                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 483                                      lump);
 484         if (IS_ERR(op_data))
 485                 RETURN(PTR_ERR(op_data));
 486
 487         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 488                 /* selinux_dentry_init_security() uses dentry->d_parent and name
 489                  * to determine the security context for the file. So our fake
 490                  * dentry should be real enough for this purpose. */
 491                 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
 492                                               &op_data->op_file_secctx_name,
 493                                               &op_data->op_file_secctx,
 494                                               &op_data->op_file_secctx_size);
 495                 if (err < 0)
 496                         GOTO(out_op_data, err);
 497         }
 498
 499         op_data->op_cli_flags |= CLI_SET_MEA;
 500         err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
 501                         from_kuid(&init_user_ns, current_fsuid()),
 502                         from_kgid(&init_user_ns, current_fsgid()),
 503                         cfs_curproc_cap_pack(), 0, &request);
 504         if (err)
 505                 GOTO(out_request, err);
 506
 507         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
 508
 509         err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
 510         if (err)
 511                 GOTO(out_inode, err);
 512
 513         dentry.d_inode = inode;
 514
 515         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 516                 inode_lock(inode);
 517                 err = security_inode_notifysecctx(inode,
 518                                                   op_data->op_file_secctx,
 519                                                   op_data->op_file_secctx_size);
 520                 inode_unlock(inode);
 521         } else {
 522                 err = ll_inode_init_security(&dentry, inode, parent);
 523         }
 524         if (err)
 525                 GOTO(out_inode, err);
 526
 527 out_inode:
 528         if (inode != NULL)
 529                 iput(inode);
 530 out_request:
 531         ptlrpc_req_finished(request);
 532 out_op_data:
 533         ll_finish_md_op_data(op_data);
 534
 535         return err;
 536 }
 537
 538 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 539                      int set_default)
 540 {
 541         struct ll_sb_info *sbi = ll_i2sbi(inode);
 542         struct md_op_data *op_data;
 543         struct ptlrpc_request *req = NULL;
 544         int rc = 0;
 545 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 546         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 547         struct obd_device *mgc = lsi->lsi_mgc;
 548 #endif
 549         int lum_size;
 550         ENTRY;
 551
 552         if (lump != NULL) {
 553                 switch (lump->lmm_magic) {
 554                 case LOV_USER_MAGIC_V1:
 555                         lum_size = sizeof(struct lov_user_md_v1);
 556                         break;
 557                 case LOV_USER_MAGIC_V3:
 558                         lum_size = sizeof(struct lov_user_md_v3);
 559                         break;
 560                 case LOV_USER_MAGIC_COMP_V1:
 561                         lum_size = ((struct lov_comp_md_v1 *)lump)->lcm_size;
 562                         break;
 563                 case LMV_USER_MAGIC:
 564                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 565                                 lustre_swab_lmv_user_md(
 566                                         (struct lmv_user_md *)lump);
 567                         lum_size = sizeof(struct lmv_user_md);
 568                         break;
 569                 case LOV_USER_MAGIC_SPECIFIC: {
 570                         struct lov_user_md_v3 *v3 =
 571                                 (struct lov_user_md_v3 *)lump;
 572                         if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
 573                                 RETURN(-EINVAL);
 574                         lum_size = lov_user_md_size(v3->lmm_stripe_count,
 575                                                     LOV_USER_MAGIC_SPECIFIC);
 576                         break;
 577                 }
 578                 default:
 579                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 580                                         " %#08x != %#08x nor %#08x\n",
 581                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 582                                         LOV_USER_MAGIC_V3);
 583                         RETURN(-EINVAL);
 584                 }
 585
 586                 /*
 587                  * This is coming from userspace, so should be in
 588                  * local endian.  But the MDS would like it in little
 589                  * endian, so we swab it before we send it.
 590                  */
 591                 if ((__swab32(lump->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
 592                     le32_to_cpu(LOV_MAGIC_MAGIC))
 593                         lustre_swab_lov_user_md(lump);
 594         } else {
 595                 lum_size = sizeof(struct lov_user_md_v1);
 596         }
 597
 598         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 599                                      LUSTRE_OPC_ANY, NULL);
 600         if (IS_ERR(op_data))
 601                 RETURN(PTR_ERR(op_data));
 602
 603         /* swabbing is done in lov_setstripe() on server side */
 604         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
 605         ll_finish_md_op_data(op_data);
 606         ptlrpc_req_finished(req);
 607         if (rc)
 608                 RETURN(rc);
 609
 610 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 611         /*
 612          * 2.9 server has stored filesystem default stripe in ROOT xattr,
 613          * and it's stored into system config for backward compatibility.
 614          *
 615          * In the following we use the fact that LOV_USER_MAGIC_V1 and
 616          * LOV_USER_MAGIC_V3 have the same initial fields so we do not
 617          * need the make the distiction between the 2 versions
 618          */
 619         if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
 620             (lump == NULL ||
 621              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
 622              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
 623                 char *param = NULL;
 624                 char *buf;
 625
 626                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 627                 if (param == NULL)
 628                         GOTO(end, rc = -ENOMEM);
 629
 630                 buf = param;
 631                 /* Get fsname and assume devname to be -MDT0000. */
 632                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 633                 strcat(buf, "-MDT0000.lov");
 634                 buf += strlen(buf);
 635
 636                 /* Set root stripesize */
 637                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
 638                          lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 639                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 640                 if (rc)
 641                         GOTO(end, rc);
 642
 643                 /* Set root stripecount */
 644                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
 645                          lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 646                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 647                 if (rc)
 648                         GOTO(end, rc);
 649
 650                 /* Set root stripeoffset */
 651                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
 652                          lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 653                                 (typeof(lump->lmm_stripe_offset))(-1));
 654                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 655
 656 end:
 657                 if (param != NULL)
 658                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 659         }
 660 #endif
 661         RETURN(rc);
 662 }
 663
 664 /**
 665  * This function will be used to get default LOV/LMV/Default LMV
 666  * @valid will be used to indicate which stripe it will retrieve
 667  *      OBD_MD_MEA              LMV stripe EA
 668  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 669  *      otherwise               Default LOV EA.
 670  * Each time, it can only retrieve 1 stripe EA
 671  **/
 672 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 673                      struct ptlrpc_request **request, u64 valid)
 674 {
 675         struct ll_sb_info *sbi = ll_i2sbi(inode);
 676         struct mdt_body   *body;
 677         struct lov_mds_md *lmm = NULL;
 678         struct ptlrpc_request *req = NULL;
 679         int rc, lmm_size;
 680         struct md_op_data *op_data;
 681         ENTRY;
 682
 683         rc = ll_get_default_mdsize(sbi, &lmm_size);
 684         if (rc)
 685                 RETURN(rc);
 686
 687         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 688                                      0, lmm_size, LUSTRE_OPC_ANY,
 689                                      NULL);
 690         if (IS_ERR(op_data))
 691                 RETURN(PTR_ERR(op_data));
 692
 693         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 694         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 695         ll_finish_md_op_data(op_data);
 696         if (rc < 0) {
 697                 CDEBUG(D_INFO, "md_getattr failed on inode "
 698                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 699                 GOTO(out, rc);
 700         }
 701
 702         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 703         LASSERT(body != NULL);
 704
 705         lmm_size = body->mbo_eadatasize;
 706
 707         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 708             lmm_size == 0) {
 709                 GOTO(out, rc = -ENODATA);
 710         }
 711
 712         lmm = req_capsule_server_sized_get(&req->rq_pill,
 713                                            &RMF_MDT_MD, lmm_size);
 714         LASSERT(lmm != NULL);
 715
 716         /*
 717          * This is coming from the MDS, so is probably in
 718          * little endian.  We convert it to host endian before
 719          * passing it to userspace.
 720          */
 721         /* We don't swab objects for directories */
 722         switch (le32_to_cpu(lmm->lmm_magic)) {
 723         case LOV_MAGIC_V1:
 724         case LOV_MAGIC_V3:
 725         case LOV_MAGIC_COMP_V1:
 726         case LOV_USER_MAGIC_SPECIFIC:
 727                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 728                         lustre_swab_lov_user_md((struct lov_user_md *)lmm);
 729                 break;
 730         case LMV_MAGIC_V1:
 731                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 732                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 733                 break;
 734         case LMV_USER_MAGIC:
 735                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 736                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 737                 break;
 738         default:
 739                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 740                 rc = -EPROTO;
 741         }
 742 out:
 743         *plmm = lmm;
 744         *plmm_size = lmm_size;
 745         *request = req;
 746         return rc;
 747 }
 748
 749 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 750 {
 751         struct md_op_data       *op_data;
 752         int                     rc;
 753         int                     mdt_index;
 754         ENTRY;
 755
 756         OBD_ALLOC_PTR(op_data);
 757         if (op_data == NULL)
 758                 RETURN(-ENOMEM);
 759
 760         op_data->op_flags |= MF_GET_MDT_IDX;
 761         op_data->op_fid1 = *fid;
 762         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 763         mdt_index = op_data->op_mds;
 764         OBD_FREE_PTR(op_data);
 765         if (rc < 0)
 766                 RETURN(rc);
 767
 768         RETURN(mdt_index);
 769 }
 770
 771 /*
 772  *  Get MDT index for the inode.
 773  */
 774 int ll_get_mdt_idx(struct inode *inode)
 775 {
 776         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 777 }
 778
 779 /**
 780  * Generic handler to do any pre-copy work.
 781  *
 782  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 783  * first information for it that real work has started.
 784  *
 785  * Moreover, for a ARCHIVE request, it will sample the file data version and
 786  * store it in \a copy.
 787  *
 788  * \return 0 on success.
 789  */
 790 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 791 {
 792         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 793         struct hsm_progress_kernel       hpk;
 794         int                              rc = 0;
 795         int                              rc2;
 796         ENTRY;
 797
 798         /* Forge a hsm_progress based on data from copy. */
 799         hpk.hpk_fid = copy->hc_hai.hai_fid;
 800         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 801         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 802         hpk.hpk_extent.length = 0;
 803         hpk.hpk_flags = 0;
 804         hpk.hpk_errval = 0;
 805         hpk.hpk_data_version = 0;
 806
 807
 808         /* For archive request, we need to read the current file version. */
 809         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 810                 struct inode    *inode;
 811                 __u64            data_version = 0;
 812
 813                 /* Get inode for this fid */
 814                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 815                 if (IS_ERR(inode)) {
 816                         hpk.hpk_flags |= HP_FLAG_RETRY;
 817                         /* hpk_errval is >= 0 */
 818                         hpk.hpk_errval = -PTR_ERR(inode);
 819                         GOTO(progress, rc = PTR_ERR(inode));
 820                 }
 821
 822                 /* Read current file data version */
 823                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 824                 iput(inode);
 825                 if (rc != 0) {
 826                         CDEBUG(D_HSM, "Could not read file data version of "
 827                                       DFID" (rc = %d). Archive request ("
 828                                       "%#llx) could not be done.\n",
 829                                       PFID(&copy->hc_hai.hai_fid), rc,
 830                                       copy->hc_hai.hai_cookie);
 831                         hpk.hpk_flags |= HP_FLAG_RETRY;
 832                         /* hpk_errval must be >= 0 */
 833                         hpk.hpk_errval = -rc;
 834                         GOTO(progress, rc);
 835                 }
 836
 837                 /* Store in the hsm_copy for later copytool use.
 838                  * Always modified even if no lsm. */
 839                 copy->hc_data_version = data_version;
 840         }
 841
 842 progress:
 843         /* On error, the request should be considered as completed */
 844         if (hpk.hpk_errval > 0)
 845                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 846
 847         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 848                             &hpk, NULL);
 849
 850         /* Return first error */
 851         RETURN(rc != 0 ? rc : rc2);
 852 }
 853
 854 /**
 855  * Generic handler to do any post-copy work.
 856  *
 857  * It will send the last hsm_progress update to coordinator to inform it
 858  * that copy is finished and whether it was successful or not.
 859  *
 860  * Moreover,
 861  * - for ARCHIVE request, it will sample the file data version and compare it
 862  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 863  *   will be considered as failed.
 864  * - for RESTORE request, it will sample the file data version and send it to
 865  *   coordinator which is useful if the file was imported as 'released'.
 866  *
 867  * \return 0 on success.
 868  */
 869 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 870 {
 871         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 872         struct hsm_progress_kernel       hpk;
 873         int                              rc = 0;
 874         int                              rc2;
 875         ENTRY;
 876
 877         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 878         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 879          * initialized if copy_end was called with copy == NULL.
 880          */
 881
 882         /* Forge a hsm_progress based on data from copy. */
 883         hpk.hpk_fid = copy->hc_hai.hai_fid;
 884         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 885         hpk.hpk_extent = copy->hc_hai.hai_extent;
 886         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 887         hpk.hpk_errval = copy->hc_errval;
 888         hpk.hpk_data_version = 0;
 889
 890         /* For archive request, we need to check the file data was not changed.
 891          *
 892          * For restore request, we need to send the file data version, this is
 893          * useful when the file was created using hsm_import.
 894          */
 895         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 896              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 897             (copy->hc_errval == 0)) {
 898                 struct inode    *inode;
 899                 __u64            data_version = 0;
 900
 901                 /* Get lsm for this fid */
 902                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 903                 if (IS_ERR(inode)) {
 904                         hpk.hpk_flags |= HP_FLAG_RETRY;
 905                         /* hpk_errval must be >= 0 */
 906                         hpk.hpk_errval = -PTR_ERR(inode);
 907                         GOTO(progress, rc = PTR_ERR(inode));
 908                 }
 909
 910                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 911                 iput(inode);
 912                 if (rc) {
 913                         CDEBUG(D_HSM, "Could not read file data version. "
 914                                       "Request could not be confirmed.\n");
 915                         if (hpk.hpk_errval == 0)
 916                                 hpk.hpk_errval = -rc;
 917                         GOTO(progress, rc);
 918                 }
 919
 920                 /* Store in the hsm_copy for later copytool use.
 921                  * Always modified even if no lsm. */
 922                 hpk.hpk_data_version = data_version;
 923
 924                 /* File could have been stripped during archiving, so we need
 925                  * to check anyway. */
 926                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 927                     (copy->hc_data_version != data_version)) {
 928                         CDEBUG(D_HSM, "File data version mismatched. "
 929                               "File content was changed during archiving. "
 930                                DFID", start:%#llx current:%#llx\n",
 931                                PFID(&copy->hc_hai.hai_fid),
 932                                copy->hc_data_version, data_version);
 933                         /* File was changed, send error to cdt. Do not ask for
 934                          * retry because if a file is modified frequently,
 935                          * the cdt will loop on retried archive requests.
 936                          * The policy engine will ask for a new archive later
 937                          * when the file will not be modified for some tunable
 938                          * time */
 939                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 940                         rc = -EBUSY;
 941                         /* hpk_errval must be >= 0 */
 942                         hpk.hpk_errval = -rc;
 943                         GOTO(progress, rc);
 944                 }
 945
 946         }
 947
 948 progress:
 949         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 950                             &hpk, NULL);
 951
 952         /* Return first error */
 953         RETURN(rc != 0 ? rc : rc2);
 954 }
 955
 956
 957 static int copy_and_ct_start(int cmd, struct obd_export *exp,
 958                              const struct lustre_kernelcomm __user *data)
 959 {
 960         struct lustre_kernelcomm *lk;
 961         struct lustre_kernelcomm *tmp;
 962         size_t size = sizeof(*lk);
 963         size_t new_size;
 964         int i;
 965         int rc;
 966
 967         /* copy data from userspace to get numbers of archive_id */
 968         OBD_ALLOC(lk, size);
 969         if (lk == NULL)
 970                 return -ENOMEM;
 971
 972         if (copy_from_user(lk, data, size))
 973                 GOTO(out_lk, rc = -EFAULT);
 974
 975         if (lk->lk_flags & LK_FLG_STOP)
 976                 goto do_ioctl;
 977
 978         if (!(lk->lk_flags & LK_FLG_DATANR)) {
 979                 __u32 archive_mask = lk->lk_data_count;
 980                 int count;
 981
 982                 /* old hsm agent to old MDS */
 983                 if (!exp_connect_archive_id_array(exp))
 984                         goto do_ioctl;
 985
 986                 /* old hsm agent to new MDS */
 987                 lk->lk_flags |= LK_FLG_DATANR;
 988
 989                 if (archive_mask == 0)
 990                         goto do_ioctl;
 991
 992                 count = hweight32(archive_mask);
 993                 new_size = offsetof(struct lustre_kernelcomm, lk_data[count]);
 994                 OBD_ALLOC(tmp, new_size);
 995                 if (tmp == NULL)
 996                         GOTO(out_lk, rc = -ENOMEM);
 997
 998                 memcpy(tmp, lk, size);
 999                 tmp->lk_data_count = count;
1000                 OBD_FREE(lk, size);
1001                 lk = tmp;
1002                 size = new_size;
1003
1004                 count = 0;
1005                 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
1006                         if ((1 << i) & archive_mask) {
1007                                 lk->lk_data[count] = i + 1;
1008                                 count++;
1009                         }
1010                 }
1011                 goto do_ioctl;
1012         }
1013
1014         /* new hsm agent to new mds */
1015         if (lk->lk_data_count > 0) {
1016                 new_size = offsetof(struct lustre_kernelcomm,
1017                                     lk_data[lk->lk_data_count]);
1018                 OBD_ALLOC(tmp, new_size);
1019                 if (tmp == NULL)
1020                         GOTO(out_lk, rc = -ENOMEM);
1021
1022                 OBD_FREE(lk, size);
1023                 lk = tmp;
1024                 size = new_size;
1025
1026                 if (copy_from_user(lk, data, size))
1027                         GOTO(out_lk, rc = -EFAULT);
1028         }
1029
1030         /* new hsm agent to old MDS */
1031         if (!exp_connect_archive_id_array(exp)) {
1032                 __u32 archives = 0;
1033
1034                 if (lk->lk_data_count > LL_HSM_ORIGIN_MAX_ARCHIVE)
1035                         GOTO(out_lk, rc = -EINVAL);
1036
1037                 for (i = 0; i < lk->lk_data_count; i++) {
1038                         if (lk->lk_data[i] > LL_HSM_ORIGIN_MAX_ARCHIVE) {
1039                                 rc = -EINVAL;
1040                                 CERROR("%s: archive id %d requested but only "
1041                                        "[0 - %zu] supported: rc = %d\n",
1042                                        exp->exp_obd->obd_name, lk->lk_data[i],
1043                                        LL_HSM_ORIGIN_MAX_ARCHIVE, rc);
1044                                 GOTO(out_lk, rc);
1045                         }
1046
1047                         if (lk->lk_data[i] == 0) {
1048                                 archives = 0;
1049                                 break;
1050                         }
1051
1052                         archives |= (1 << (lk->lk_data[i] - 1));
1053                 }
1054                 lk->lk_flags &= ~LK_FLG_DATANR;
1055                 lk->lk_data_count = archives;
1056         }
1057 do_ioctl:
1058         rc = obd_iocontrol(cmd, exp, size, lk, NULL);
1059 out_lk:
1060         OBD_FREE(lk, size);
1061         return rc;
1062 }
1063
1064 static int check_owner(int type, int id)
1065 {
1066         switch (type) {
1067         case USRQUOTA:
1068                 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1069                         return -EPERM;
1070                 break;
1071         case GRPQUOTA:
1072                 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1073                         return -EPERM;
1074                 break;
1075         case PRJQUOTA:
1076                 break;
1077         }
1078         return 0;
1079 }
1080
1081 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1082 {
1083         int cmd = qctl->qc_cmd;
1084         int type = qctl->qc_type;
1085         int id = qctl->qc_id;
1086         int valid = qctl->qc_valid;
1087         int rc = 0;
1088         ENTRY;
1089
1090         switch (cmd) {
1091         case Q_SETQUOTA:
1092         case Q_SETINFO:
1093         case LUSTRE_Q_SETDEFAULT:
1094                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1095                         RETURN(-EPERM);
1096                 break;
1097         case Q_GETQUOTA:
1098         case LUSTRE_Q_GETDEFAULT:
1099                 if (check_owner(type, id) &&
1100                     (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1101                         RETURN(-EPERM);
1102                 break;
1103         case Q_GETINFO:
1104                 break;
1105         default:
1106                 CERROR("unsupported quotactl op: %#x\n", cmd);
1107                 RETURN(-ENOTSUPP);
1108         }
1109
1110         if (valid != QC_GENERAL) {
1111                 if (cmd == Q_GETINFO)
1112                         qctl->qc_cmd = Q_GETOINFO;
1113                 else if (cmd == Q_GETQUOTA)
1114                         qctl->qc_cmd = Q_GETOQUOTA;
1115                 else
1116                         RETURN(-EINVAL);
1117
1118                 switch (valid) {
1119                 case QC_MDTIDX:
1120                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1121                                            sizeof(*qctl), qctl, NULL);
1122                         break;
1123                 case QC_OSTIDX:
1124                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1125                                            sizeof(*qctl), qctl, NULL);
1126                         break;
1127                 case QC_UUID:
1128                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1129                                            sizeof(*qctl), qctl, NULL);
1130                         if (rc == -EAGAIN)
1131                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1132                                                    sbi->ll_dt_exp,
1133                                                    sizeof(*qctl), qctl, NULL);
1134                         break;
1135                 default:
1136                         rc = -EINVAL;
1137                         break;
1138                 }
1139
1140                 if (rc)
1141                         RETURN(rc);
1142
1143                 qctl->qc_cmd = cmd;
1144         } else {
1145                 struct obd_quotactl *oqctl;
1146
1147                 OBD_ALLOC_PTR(oqctl);
1148                 if (oqctl == NULL)
1149                         RETURN(-ENOMEM);
1150
1151                 QCTL_COPY(oqctl, qctl);
1152                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1153                 if (rc) {
1154                         OBD_FREE_PTR(oqctl);
1155                         RETURN(rc);
1156                 }
1157                 /* If QIF_SPACE is not set, client should collect the
1158                  * space usage from OSSs by itself */
1159                 if (cmd == Q_GETQUOTA &&
1160                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1161                     !oqctl->qc_dqblk.dqb_curspace) {
1162                         struct obd_quotactl *oqctl_tmp;
1163
1164                         OBD_ALLOC_PTR(oqctl_tmp);
1165                         if (oqctl_tmp == NULL)
1166                                 GOTO(out, rc = -ENOMEM);
1167
1168                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1169                         oqctl_tmp->qc_id = oqctl->qc_id;
1170                         oqctl_tmp->qc_type = oqctl->qc_type;
1171
1172                         /* collect space usage from OSTs */
1173                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1174                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1175                         if (!rc || rc == -EREMOTEIO) {
1176                                 oqctl->qc_dqblk.dqb_curspace =
1177                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1178                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1179                         }
1180
1181                         /* collect space & inode usage from MDTs */
1182                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1183                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1184                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1185                         if (!rc || rc == -EREMOTEIO) {
1186                                 oqctl->qc_dqblk.dqb_curspace +=
1187                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1188                                 oqctl->qc_dqblk.dqb_curinodes =
1189                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1190                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1191                         } else {
1192                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1193                         }
1194
1195                         OBD_FREE_PTR(oqctl_tmp);
1196                 }
1197 out:
1198                 QCTL_COPY(qctl, oqctl);
1199                 OBD_FREE_PTR(oqctl);
1200         }
1201
1202         RETURN(rc);
1203 }
1204
1205 /* This function tries to get a single name component,
1206  * to send to the server. No actual path traversal involved,
1207  * so we limit to NAME_MAX */
1208 static char *ll_getname(const char __user *filename)
1209 {
1210         int ret = 0, len;
1211         char *tmp;
1212
1213         OBD_ALLOC(tmp, NAME_MAX + 1);
1214
1215         if (!tmp)
1216                 return ERR_PTR(-ENOMEM);
1217
1218         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1219         if (len < 0)
1220                 ret = -ENOENT;
1221         else if (len > NAME_MAX)
1222                 ret = -ENAMETOOLONG;
1223
1224         if (ret) {
1225                 OBD_FREE(tmp, NAME_MAX + 1);
1226                 tmp =  ERR_PTR(ret);
1227         }
1228         return tmp;
1229 }
1230
1231 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1232
1233 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1234 {
1235         struct dentry *dentry = file_dentry(file);
1236         struct inode *inode = file_inode(file);
1237         struct ll_sb_info *sbi = ll_i2sbi(inode);
1238         struct obd_ioctl_data *data;
1239         int rc = 0;
1240         ENTRY;
1241
1242         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1243                PFID(ll_inode2fid(inode)), inode, cmd);
1244
1245         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1246         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1247                 return -ENOTTY;
1248
1249         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1250         switch (cmd) {
1251         case FS_IOC_GETFLAGS:
1252         case FS_IOC_SETFLAGS:
1253                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1254         case FSFILT_IOC_GETVERSION:
1255         case FS_IOC_GETVERSION:
1256                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1257         /* We need to special case any other ioctls we want to handle,
1258          * to send them to the MDS/OST as appropriate and to properly
1259          * network encode the arg field. */
1260         case FS_IOC_SETVERSION:
1261                 RETURN(-ENOTSUPP);
1262
1263         case LL_IOC_GET_MDTIDX: {
1264                 int mdtidx;
1265
1266                 mdtidx = ll_get_mdt_idx(inode);
1267                 if (mdtidx < 0)
1268                         RETURN(mdtidx);
1269
1270                 if (put_user((int)mdtidx, (int __user *)arg))
1271                         RETURN(-EFAULT);
1272
1273                 return 0;
1274         }
1275         case IOC_MDC_LOOKUP: {
1276                                      int namelen, len = 0;
1277                 char *buf = NULL;
1278                 char *filename;
1279
1280                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1281                 if (rc != 0)
1282                         RETURN(rc);
1283                 data = (void *)buf;
1284
1285                 filename = data->ioc_inlbuf1;
1286                 namelen = strlen(filename);
1287                 if (namelen < 1) {
1288                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1289                         GOTO(out_free, rc = -EINVAL);
1290                 }
1291
1292                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1293                 if (rc < 0) {
1294                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1295                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1296                                filename, rc);
1297                         GOTO(out_free, rc);
1298                 }
1299 out_free:
1300                 OBD_FREE_LARGE(buf, len);
1301                 return rc;
1302         }
1303         case LL_IOC_LMV_SETSTRIPE: {
1304                 struct lmv_user_md  *lum;
1305                 char            *buf = NULL;
1306                 char            *filename;
1307                 int              namelen = 0;
1308                 int              lumlen = 0;
1309                 umode_t          mode;
1310                 int              len;
1311                 int              rc;
1312
1313                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1314                 if (rc)
1315                         RETURN(rc);
1316
1317                 data = (void *)buf;
1318                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1319                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1320                         GOTO(lmv_out_free, rc = -EINVAL);
1321
1322                 filename = data->ioc_inlbuf1;
1323                 namelen = data->ioc_inllen1;
1324
1325                 if (namelen < 1) {
1326                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1327                         GOTO(lmv_out_free, rc = -EINVAL);
1328                 }
1329                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1330                 lumlen = data->ioc_inllen2;
1331
1332                 if ((lum->lum_magic != LMV_USER_MAGIC &&
1333                      lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) ||
1334                     lumlen < sizeof(*lum)) {
1335                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1336                                filename, lum->lum_magic, lumlen, -EFAULT);
1337                         GOTO(lmv_out_free, rc = -EINVAL);
1338                 }
1339
1340 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1341                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1342 #else
1343                 mode = data->ioc_type;
1344 #endif
1345                 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1346 lmv_out_free:
1347                 OBD_FREE_LARGE(buf, len);
1348                 RETURN(rc);
1349
1350         }
1351         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1352                 struct lmv_user_md        lum;
1353                 struct lmv_user_md __user *ulump =
1354                                         (struct lmv_user_md __user *)arg;
1355                 int                       rc;
1356
1357                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1358                         RETURN(-EFAULT);
1359
1360                 if (lum.lum_magic != LMV_USER_MAGIC)
1361                         RETURN(-EINVAL);
1362
1363                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1364
1365                 RETURN(rc);
1366         }
1367         case LL_IOC_LOV_SETSTRIPE_NEW:
1368         case LL_IOC_LOV_SETSTRIPE: {
1369                 struct lov_user_md_v3 *lumv3 = NULL;
1370                 struct lov_user_md_v1 lumv1;
1371                 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1372                 struct lov_user_md_v1 __user *lumv1p =
1373                         (struct lov_user_md_v1 __user *)arg;
1374                 struct lov_user_md_v3 __user *lumv3p =
1375                         (struct lov_user_md_v3 __user *)arg;
1376                 int lum_size = 0;
1377
1378                 int set_default = 0;
1379
1380                 CLASSERT(sizeof(struct lov_user_md_v3) >
1381                          sizeof(struct lov_comp_md_v1));
1382                 CLASSERT(sizeof(*lumv3) == sizeof(*lumv3p));
1383                 /* first try with v1 which is smaller than v3 */
1384                 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1385                         RETURN(-EFAULT);
1386
1387                 if (inode->i_sb->s_root == file_dentry(file))
1388                         set_default = 1;
1389
1390                 switch (lumv1.lmm_magic) {
1391                 case LOV_USER_MAGIC_V3:
1392                 case LOV_USER_MAGIC_SPECIFIC:
1393                         lum_size = ll_lov_user_md_size(&lumv1);
1394                         if (lum_size < 0)
1395                                 RETURN(lum_size);
1396                         OBD_ALLOC(lumv3, lum_size);
1397                         if (!lumv3)
1398                                 RETURN(-ENOMEM);
1399                         if (copy_from_user(lumv3, lumv3p, lum_size))
1400                                 GOTO(out, rc = -EFAULT);
1401                         lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1402                         break;
1403                 case LOV_USER_MAGIC_V1:
1404                         break;
1405                 default:
1406                         GOTO(out, rc = -ENOTSUPP);
1407                 }
1408
1409                 /* in v1 and v3 cases lumv1 points to data */
1410                 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1411 out:
1412                 if (lumv3)
1413                         OBD_FREE(lumv3, lum_size);
1414                 RETURN(rc);
1415         }
1416         case LL_IOC_LMV_GETSTRIPE: {
1417                 struct lmv_user_md __user *ulmv =
1418                                         (struct lmv_user_md __user *)arg;
1419                 struct lmv_user_md      lum;
1420                 struct ptlrpc_request   *request = NULL;
1421                 union lmv_mds_md        *lmm = NULL;
1422                 int                     lmmsize;
1423                 u64                     valid = 0;
1424                 struct lmv_user_md      *tmp = NULL;
1425                 int                     mdt_index;
1426                 int                     lum_size;
1427                 int                     stripe_count;
1428                 int                     max_stripe_count;
1429                 int                     i;
1430                 int                     rc;
1431
1432                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1433                         RETURN(-EFAULT);
1434
1435                 max_stripe_count = lum.lum_stripe_count;
1436                 /* lum_magic will indicate which stripe the ioctl will like
1437                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1438                  * is for default LMV stripe */
1439                 if (lum.lum_magic == LMV_MAGIC_V1)
1440                         valid |= OBD_MD_MEA;
1441                 else if (lum.lum_magic == LMV_USER_MAGIC)
1442                         valid |= OBD_MD_DEFAULT_MEA;
1443                 else
1444                         RETURN(-EINVAL);
1445
1446                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1447                                       valid);
1448                 if (rc != 0)
1449                         GOTO(finish_req, rc);
1450
1451                 /* Get default LMV EA */
1452                 if (lum.lum_magic == LMV_USER_MAGIC) {
1453                         if (lmmsize > sizeof(*ulmv))
1454                                 GOTO(finish_req, rc = -EINVAL);
1455
1456                         if (copy_to_user(ulmv, lmm, lmmsize))
1457                                 GOTO(finish_req, rc = -EFAULT);
1458
1459                         GOTO(finish_req, rc);
1460                 }
1461
1462                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1463                 if (max_stripe_count < stripe_count) {
1464                         lum.lum_stripe_count = stripe_count;
1465                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1466                                 GOTO(finish_req, rc = -EFAULT);
1467                         GOTO(finish_req, rc = -E2BIG);
1468                 }
1469
1470                 lum_size = lmv_user_md_size(stripe_count,
1471                                             LMV_USER_MAGIC_SPECIFIC);
1472                 OBD_ALLOC(tmp, lum_size);
1473                 if (tmp == NULL)
1474                         GOTO(finish_req, rc = -ENOMEM);
1475
1476                 mdt_index = ll_get_mdt_idx(inode);
1477                 if (mdt_index < 0)
1478                         GOTO(out_tmp, rc = -ENOMEM);
1479
1480                 tmp->lum_magic = LMV_MAGIC_V1;
1481                 tmp->lum_stripe_count = 0;
1482                 tmp->lum_stripe_offset = mdt_index;
1483                 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1484                 for (i = 0; i < stripe_count; i++) {
1485                         struct lu_fid   fid;
1486
1487                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1488                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1489                         if (mdt_index < 0)
1490                                 GOTO(out_tmp, rc = mdt_index);
1491
1492                         tmp->lum_objects[i].lum_mds = mdt_index;
1493                         tmp->lum_objects[i].lum_fid = fid;
1494                         tmp->lum_stripe_count++;
1495                 }
1496
1497                 if (copy_to_user(ulmv, tmp, lum_size))
1498                         GOTO(out_tmp, rc = -EFAULT);
1499 out_tmp:
1500                 OBD_FREE(tmp, lum_size);
1501 finish_req:
1502                 ptlrpc_req_finished(request);
1503                 return rc;
1504         }
1505
1506         case LL_IOC_REMOVE_ENTRY: {
1507                 char            *filename = NULL;
1508                 int              namelen = 0;
1509                 int              rc;
1510
1511                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1512                  * unsupported server, which might crash the server(LU-2730),
1513                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1514                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1515                  * server will support REINT_RMENTRY XXX*/
1516                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1517                         RETURN(-ENOTSUPP);
1518
1519                 filename = ll_getname((const char __user *)arg);
1520                 if (IS_ERR(filename))
1521                         RETURN(PTR_ERR(filename));
1522
1523                 namelen = strlen(filename);
1524                 if (namelen < 1)
1525                         GOTO(out_rmdir, rc = -EINVAL);
1526
1527                 rc = ll_rmdir_entry(inode, filename, namelen);
1528 out_rmdir:
1529                 if (filename)
1530                         ll_putname(filename);
1531                 RETURN(rc);
1532         }
1533         case LL_IOC_LOV_SWAP_LAYOUTS:
1534                 RETURN(-EPERM);
1535         case IOC_OBD_STATFS:
1536                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1537         case LL_IOC_LOV_GETSTRIPE:
1538         case LL_IOC_LOV_GETSTRIPE_NEW:
1539         case LL_IOC_MDC_GETINFO:
1540         case IOC_MDC_GETFILEINFO:
1541         case IOC_MDC_GETFILESTRIPE: {
1542                 struct ptlrpc_request *request = NULL;
1543                 struct lov_user_md __user *lump;
1544                 struct lov_mds_md *lmm = NULL;
1545                 struct mdt_body *body;
1546                 char *filename = NULL;
1547                 int lmmsize;
1548
1549                 if (cmd == IOC_MDC_GETFILEINFO ||
1550                     cmd == IOC_MDC_GETFILESTRIPE) {
1551                         filename = ll_getname((const char __user *)arg);
1552                         if (IS_ERR(filename))
1553                                 RETURN(PTR_ERR(filename));
1554
1555                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1556                                                       &lmmsize, &request);
1557                 } else {
1558                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1559                                               &request, 0);
1560                 }
1561
1562                 if (request) {
1563                         body = req_capsule_server_get(&request->rq_pill,
1564                                                       &RMF_MDT_BODY);
1565                         LASSERT(body != NULL);
1566                 } else {
1567                         GOTO(out_req, rc);
1568                 }
1569
1570                 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1571                                        cmd == LL_IOC_MDC_GETINFO)) {
1572                         lmmsize = 0;
1573                         rc = 0;
1574                 }
1575
1576                 if (rc < 0)
1577                         GOTO(out_req, rc);
1578
1579                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1580                     cmd == LL_IOC_LOV_GETSTRIPE ||
1581                     cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1582                         lump = (struct lov_user_md __user *)arg;
1583                 } else {
1584                         struct lov_user_mds_data __user *lmdp;
1585                         lmdp = (struct lov_user_mds_data __user *)arg;
1586                         lump = &lmdp->lmd_lmm;
1587                 }
1588
1589                 if (lmmsize == 0) {
1590                         /* If the file has no striping then zero out *lump so
1591                          * that the caller isn't confused by garbage. */
1592                         if (clear_user(lump, sizeof(*lump)))
1593                                 GOTO(out_req, rc = -EFAULT);
1594                 } else if (copy_to_user(lump, lmm, lmmsize)) {
1595                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1596                                 GOTO(out_req, rc = -EFAULT);
1597                         rc = -EOVERFLOW;
1598                 }
1599
1600                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1601                         struct lov_user_mds_data __user *lmdp;
1602                         lstat_t st = { 0 };
1603
1604                         st.st_dev       = inode->i_sb->s_dev;
1605                         st.st_mode      = body->mbo_mode;
1606                         st.st_nlink     = body->mbo_nlink;
1607                         st.st_uid       = body->mbo_uid;
1608                         st.st_gid       = body->mbo_gid;
1609                         st.st_rdev      = body->mbo_rdev;
1610                         st.st_size      = body->mbo_size;
1611                         st.st_blksize   = PAGE_SIZE;
1612                         st.st_blocks    = body->mbo_blocks;
1613                         st.st_atime     = body->mbo_atime;
1614                         st.st_mtime     = body->mbo_mtime;
1615                         st.st_ctime     = body->mbo_ctime;
1616                         st.st_ino       = cl_fid_build_ino(&body->mbo_fid1,
1617                                                 sbi->ll_flags &
1618                                                 LL_SBI_32BIT_API);
1619
1620                         lmdp = (struct lov_user_mds_data __user *)arg;
1621                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1622                                 GOTO(out_req, rc = -EFAULT);
1623                 }
1624
1625                 EXIT;
1626         out_req:
1627                 ptlrpc_req_finished(request);
1628                 if (filename)
1629                         ll_putname(filename);
1630                 return rc;
1631         }
1632         case OBD_IOC_QUOTACTL: {
1633                 struct if_quotactl *qctl;
1634
1635                 OBD_ALLOC_PTR(qctl);
1636                 if (!qctl)
1637                         RETURN(-ENOMEM);
1638
1639                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1640                         GOTO(out_quotactl, rc = -EFAULT);
1641
1642                 rc = quotactl_ioctl(sbi, qctl);
1643
1644                 if (rc == 0 &&
1645                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1646                         rc = -EFAULT;
1647
1648         out_quotactl:
1649                 OBD_FREE_PTR(qctl);
1650                 RETURN(rc);
1651         }
1652         case OBD_IOC_GETDTNAME:
1653         case OBD_IOC_GETMDNAME:
1654                 RETURN(ll_get_obd_name(inode, cmd, arg));
1655         case LL_IOC_FLUSHCTX:
1656                 RETURN(ll_flush_ctx(inode));
1657         case LL_IOC_GETOBDCOUNT: {
1658                 u32 count, vallen;
1659                 struct obd_export *exp;
1660
1661                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1662                         RETURN(-EFAULT);
1663
1664                 /* get ost count when count is zero, get mdt count otherwise */
1665                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1666                 vallen = sizeof(count);
1667                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1668                                   KEY_TGT_COUNT, &vallen, &count);
1669                 if (rc) {
1670                         CERROR("get target count failed: %d\n", rc);
1671                         RETURN(rc);
1672                 }
1673
1674                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1675                         RETURN(-EFAULT);
1676
1677                 RETURN(0);
1678         }
1679         case LL_IOC_PATH2FID:
1680                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1681                                      sizeof(struct lu_fid)))
1682                         RETURN(-EFAULT);
1683                 RETURN(0);
1684         case LL_IOC_GET_CONNECT_FLAGS: {
1685                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1686                                      (void __user *)arg));
1687         }
1688         case OBD_IOC_FID2PATH:
1689                 RETURN(ll_fid2path(inode, (void __user *)arg));
1690         case LL_IOC_GETPARENT:
1691                 RETURN(ll_getparent(file, (void __user *)arg));
1692         case LL_IOC_FID2MDTIDX: {
1693                 struct obd_export *exp = ll_i2mdexp(inode);
1694                 struct lu_fid     fid;
1695                 __u32             index;
1696
1697                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1698                                    sizeof(fid)))
1699                         RETURN(-EFAULT);
1700
1701                 /* Call mdc_iocontrol */
1702                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1703                                    (__u32 __user *)&index);
1704                 if (rc != 0)
1705                         RETURN(rc);
1706
1707                 RETURN(index);
1708         }
1709         case LL_IOC_HSM_REQUEST: {
1710                 struct hsm_user_request *hur;
1711                 ssize_t                  totalsize;
1712
1713                 OBD_ALLOC_PTR(hur);
1714                 if (hur == NULL)
1715                         RETURN(-ENOMEM);
1716
1717                 /* We don't know the true size yet; copy the fixed-size part */
1718                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1719                         OBD_FREE_PTR(hur);
1720                         RETURN(-EFAULT);
1721                 }
1722
1723                 /* Compute the whole struct size */
1724                 totalsize = hur_len(hur);
1725                 OBD_FREE_PTR(hur);
1726                 if (totalsize < 0)
1727                         RETURN(-E2BIG);
1728
1729                 /* Final size will be more than double totalsize */
1730                 if (totalsize >= MDS_MAXREQSIZE / 3)
1731                         RETURN(-E2BIG);
1732
1733                 OBD_ALLOC_LARGE(hur, totalsize);
1734                 if (hur == NULL)
1735                         RETURN(-ENOMEM);
1736
1737                 /* Copy the whole struct */
1738                 if (copy_from_user(hur, (void __user *)arg, totalsize))
1739                         GOTO(out_hur, rc = -EFAULT);
1740
1741                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1742                         const struct lu_fid *fid;
1743                         struct inode *f;
1744                         int i;
1745
1746                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1747                                 fid = &hur->hur_user_item[i].hui_fid;
1748                                 f = search_inode_for_lustre(inode->i_sb, fid);
1749                                 if (IS_ERR(f)) {
1750                                         rc = PTR_ERR(f);
1751                                         break;
1752                                 }
1753
1754                                 rc = ll_hsm_release(f);
1755                                 iput(f);
1756                                 if (rc != 0)
1757                                         break;
1758                         }
1759                 } else {
1760                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1761                                            hur, NULL);
1762                 }
1763
1764 out_hur:
1765                 OBD_FREE_LARGE(hur, totalsize);
1766
1767                 RETURN(rc);
1768         }
1769         case LL_IOC_HSM_PROGRESS: {
1770                 struct hsm_progress_kernel      hpk;
1771                 struct hsm_progress             hp;
1772
1773                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1774                         RETURN(-EFAULT);
1775
1776                 hpk.hpk_fid = hp.hp_fid;
1777                 hpk.hpk_cookie = hp.hp_cookie;
1778                 hpk.hpk_extent = hp.hp_extent;
1779                 hpk.hpk_flags = hp.hp_flags;
1780                 hpk.hpk_errval = hp.hp_errval;
1781                 hpk.hpk_data_version = 0;
1782
1783                 /* File may not exist in Lustre; all progress
1784                  * reported to Lustre root */
1785                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1786                                    NULL);
1787                 RETURN(rc);
1788         }
1789         case LL_IOC_HSM_CT_START:
1790                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1791                         RETURN(-EPERM);
1792
1793                 rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
1794                                        (struct lustre_kernelcomm __user *)arg);
1795                 RETURN(rc);
1796
1797         case LL_IOC_HSM_COPY_START: {
1798                 struct hsm_copy *copy;
1799                 int              rc;
1800
1801                 OBD_ALLOC_PTR(copy);
1802                 if (copy == NULL)
1803                         RETURN(-ENOMEM);
1804                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1805                         OBD_FREE_PTR(copy);
1806                         RETURN(-EFAULT);
1807                 }
1808
1809                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1810                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1811                         rc = -EFAULT;
1812
1813                 OBD_FREE_PTR(copy);
1814                 RETURN(rc);
1815         }
1816         case LL_IOC_HSM_COPY_END: {
1817                 struct hsm_copy *copy;
1818                 int              rc;
1819
1820                 OBD_ALLOC_PTR(copy);
1821                 if (copy == NULL)
1822                         RETURN(-ENOMEM);
1823                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1824                         OBD_FREE_PTR(copy);
1825                         RETURN(-EFAULT);
1826                 }
1827
1828                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1829                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1830                         rc = -EFAULT;
1831
1832                 OBD_FREE_PTR(copy);
1833                 RETURN(rc);
1834         }
1835         case LL_IOC_MIGRATE: {
1836                 struct lmv_user_md *lum;
1837                 char *buf = NULL;
1838                 int len;
1839                 char *filename;
1840                 int namelen = 0;
1841                 int rc;
1842
1843                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1844                 if (rc)
1845                         RETURN(rc);
1846
1847                 data = (struct obd_ioctl_data *)buf;
1848                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1849                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1850                         GOTO(migrate_free, rc = -EINVAL);
1851
1852                 filename = data->ioc_inlbuf1;
1853                 namelen = data->ioc_inllen1;
1854
1855                 if (namelen < 1 || namelen != strlen(filename) + 1) {
1856                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1857                         GOTO(migrate_free, rc = -EINVAL);
1858                 }
1859
1860                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1861                 if (lum->lum_magic != LMV_USER_MAGIC &&
1862                     lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
1863                         rc = -EINVAL;
1864                         CERROR("%s: wrong lum magic %x: rc = %d\n",
1865                                filename, lum->lum_magic, rc);
1866                         GOTO(migrate_free, rc);
1867                 }
1868
1869                 rc = ll_migrate(inode, file, lum, filename);
1870 migrate_free:
1871                 OBD_FREE_LARGE(buf, len);
1872
1873                 RETURN(rc);
1874         }
1875         case LL_IOC_FSGETXATTR:
1876                 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
1877         case LL_IOC_FSSETXATTR:
1878                 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
1879         default:
1880                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1881                                      (void __user *)arg));
1882         }
1883 }
1884
1885 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1886 {
1887         struct inode *inode = file->f_mapping->host;
1888         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1889         struct ll_sb_info *sbi = ll_i2sbi(inode);
1890         int api32 = ll_need_32bit_api(sbi);
1891         loff_t ret = -EINVAL;
1892         ENTRY;
1893
1894         inode_lock(inode);
1895         switch (origin) {
1896                 case SEEK_SET:
1897                         break;
1898                 case SEEK_CUR:
1899                         offset += file->f_pos;
1900                         break;
1901                 case SEEK_END:
1902                         if (offset > 0)
1903                                 GOTO(out, ret);
1904                         if (api32)
1905                                 offset += LL_DIR_END_OFF_32BIT;
1906                         else
1907                                 offset += LL_DIR_END_OFF;
1908                         break;
1909                 default:
1910                         GOTO(out, ret);
1911         }
1912
1913         if (offset >= 0 &&
1914             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1915              (!api32 && offset <= LL_DIR_END_OFF))) {
1916                 if (offset != file->f_pos) {
1917                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1918                             (!api32 && offset == LL_DIR_END_OFF))
1919                                 fd->lfd_pos = MDS_DIR_END_OFF;
1920                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1921                                 fd->lfd_pos = offset << 32;
1922                         else
1923                                 fd->lfd_pos = offset;
1924                         file->f_pos = offset;
1925                         file->f_version = 0;
1926                 }
1927                 ret = offset;
1928         }
1929         GOTO(out, ret);
1930
1931 out:
1932         inode_unlock(inode);
1933         return ret;
1934 }
1935
1936 static int ll_dir_open(struct inode *inode, struct file *file)
1937 {
1938         ENTRY;
1939         RETURN(ll_file_open(inode, file));
1940 }
1941
1942 static int ll_dir_release(struct inode *inode, struct file *file)
1943 {
1944         ENTRY;
1945         RETURN(ll_file_release(inode, file));
1946 }
1947
1948 const struct file_operations ll_dir_operations = {
1949         .llseek         = ll_dir_seek,
1950         .open           = ll_dir_open,
1951         .release        = ll_dir_release,
1952         .read           = generic_read_dir,
1953 #ifdef HAVE_DIR_CONTEXT
1954         .iterate        = ll_iterate,
1955 #else
1956         .readdir        = ll_readdir,
1957 #endif
1958         .unlocked_ioctl = ll_dir_ioctl,
1959         .fsync          = ll_fsync,
1960 };