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  *
  31  * lustre/llite/dir.c
  32  *
  33  * Directory code for lustre client.
  34  */
  35
  36 #include <linux/fs.h>
  37 #include <linux/pagemap.h>
  38 #include <linux/mm.h>
  39 #include <linux/version.h>
  40 #include <linux/security.h>
  41 #include <linux/user_namespace.h>
  42 #include <linux/uidgid.h>
  43 #include <linux/uaccess.h>
  44 #include <linux/buffer_head.h>   // for wait_on_buffer
  45 #include <linux/pagevec.h>
  46
  47 #define DEBUG_SUBSYSTEM S_LLITE
  48
  49 #include <obd_support.h>
  50 #include <obd_class.h>
  51 #include <uapi/linux/lustre/lustre_ioctl.h>
  52 #include <uapi/linux/llcrypt.h>
  53 #include <lustre_lib.h>
  54 #include <lustre_dlm.h>
  55 #include <lustre_fid.h>
  56 #include <lustre_kernelcomm.h>
  57 #include <lustre_swab.h>
  58 #include <libcfs/libcfs_crypto.h>
  59
  60 #include "llite_internal.h"
  61
  62 /*
  63  * (new) readdir implementation overview.
  64  *
  65  * Original lustre readdir implementation cached exact copy of raw directory
  66  * pages on the client. These pages were indexed in client page cache by
  67  * logical offset in the directory file. This design, while very simple and
  68  * intuitive had some inherent problems:
  69  *
  70  *     . it implies that byte offset to the directory entry serves as a
  71  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  72  *     ext3/htree directory entries may move due to splits, and more
  73  *     importantly,
  74  *
  75  *     . it is incompatible with the design of split directories for cmd3,
  76  *     that assumes that names are distributed across nodes based on their
  77  *     hash, and so readdir should be done in hash order.
  78  *
  79  * New readdir implementation does readdir in hash order, and uses hash of a
  80  * file name as a telldir/seekdir cookie. This led to number of complications:
  81  *
  82  *     . hash is not unique, so it cannot be used to index cached directory
  83  *     pages on the client (note, that it requires a whole pageful of hash
  84  *     collided entries to cause two pages to have identical hashes);
  85  *
  86  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  87  *     entry cookie. ext3/htree has the same problem and lustre implementation
  88  *     mimics their solution: seekdir(hash) positions directory at the first
  89  *     entry with the given hash.
  90  *
  91  * Client side.
  92  *
  93  * 0. caching
  94  *
  95  * Client caches directory pages using hash of the first entry as an index. As
  96  * noted above hash is not unique, so this solution doesn't work as is:
  97  * special processing is needed for "page hash chains" (i.e., sequences of
  98  * pages filled with entries all having the same hash value).
  99  *
 100  * First, such chains have to be detected. To this end, server returns to the
 101  * client the hash of the first entry on the page next to one returned. When
 102  * client detects that this hash is the same as hash of the first entry on the
 103  * returned page, page hash collision has to be handled. Pages in the
 104  * hash chain, except first one, are termed "overflow pages".
 105  *
 106  * Proposed (unimplimented) solution to index uniqueness problem is to
 107  * not cache overflow pages.  Instead, when page hash collision is
 108  * detected, all overflow pages from emerging chain should be
 109  * immediately requested from the server and placed in a special data
 110  * structure.  This data structure can be used by ll_readdir() to
 111  * process entries from overflow pages.  When readdir invocation
 112  * finishes, overflow pages are discarded.  If page hash collision chain
 113  * weren't completely processed, next call to readdir will again detect
 114  * page hash collision, again read overflow pages in, process next
 115  * portion of entries and again discard the pages.  This is not as
 116  * wasteful as it looks, because, given reasonable hash, page hash
 117  * collisions are extremely rare.
 118  *
 119  * 1. directory positioning
 120  *
 121  * When seekdir(hash) is called, original
 122  *
 123  *
 124  *
 125  *
 126  *
 127  *
 128  *
 129  *
 130  * Server.
 131  *
 132  * identification of and access to overflow pages
 133  *
 134  * page format
 135  *
 136  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 137  * a header lu_dirpage which describes the start/end hash, and whether this
 138  * page is empty (contains no dir entry) or hash collide with next page.
 139  * After client receives reply, several pages will be integrated into dir page
 140  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
 141  * lu_dirpage for this integrated page will be adjusted. See
 142  * mdc_adjust_dirpages().
 143  *
 144  */
 145 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
 146                              __u64 offset, int *partial_readdir_rc)
 147 {
 148         struct md_readdir_info mrinfo = {
 149                                         .mr_blocking_ast = ll_md_blocking_ast };
 150         struct page *page;
 151         int rc;
 152
 153         rc = md_read_page(ll_i2mdexp(dir), op_data, &mrinfo, offset, &page);
 154         if (rc != 0)
 155                 return ERR_PTR(rc);
 156
 157         if (partial_readdir_rc && mrinfo.mr_partial_readdir_rc)
 158                 *partial_readdir_rc = mrinfo.mr_partial_readdir_rc;
 159
 160         return page;
 161 }
 162
 163 void ll_release_page(struct inode *inode, struct page *page,
 164                      bool remove)
 165 {
 166         kunmap(page);
 167
 168         /* Always remove the page for striped dir, because the page is
 169          * built from temporarily in LMV layer */
 170         if (inode && S_ISDIR(inode->i_mode) &&
 171             lmv_dir_striped(ll_i2info(inode)->lli_lsm_md)) {
 172                 __free_page(page);
 173                 return;
 174         }
 175
 176         if (remove) {
 177                 lock_page(page);
 178                 if (likely(page->mapping != NULL))
 179                         delete_from_page_cache(page);
 180                 unlock_page(page);
 181         }
 182         put_page(page);
 183 }
 184
 185 #ifdef HAVE_DIR_CONTEXT
 186 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 187                 struct dir_context *ctx, int *partial_readdir_rc)
 188 {
 189 #else
 190 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 191                 void *cookie, filldir_t filldir, int *partial_readdir_rc)
 192 {
 193 #endif
 194         struct ll_sb_info *sbi = ll_i2sbi(inode);
 195         __u64 pos = *ppos;
 196         bool is_api32 = ll_need_32bit_api(sbi);
 197         bool is_hash64 = test_bit(LL_SBI_64BIT_HASH, sbi->ll_flags);
 198         struct page *page;
 199         bool done = false;
 200         struct llcrypt_str lltr = LLTR_INIT(NULL, 0);
 201         int rc = 0;
 202         ENTRY;
 203
 204         if (IS_ENCRYPTED(inode)) {
 205                 rc = llcrypt_fname_alloc_buffer(inode, NAME_MAX, &lltr);
 206                 if (rc < 0)
 207                         RETURN(rc);
 208         }
 209
 210         page = ll_get_dir_page(inode, op_data, pos, partial_readdir_rc);
 211
 212         while (rc == 0 && !done) {
 213                 struct lu_dirpage *dp;
 214                 struct lu_dirent  *ent;
 215                 __u64 hash;
 216                 __u64 next;
 217
 218                 if (IS_ERR(page)) {
 219                         rc = PTR_ERR(page);
 220                         break;
 221                 }
 222
 223                 hash = MDS_DIR_END_OFF;
 224                 dp = page_address(page);
 225                 for (ent = lu_dirent_start(dp); ent != NULL && !done;
 226                      ent = lu_dirent_next(ent)) {
 227                         __u16          type;
 228                         int            namelen;
 229                         struct lu_fid  fid;
 230                         __u64          lhash;
 231                         __u64          ino;
 232
 233                         hash = le64_to_cpu(ent->lde_hash);
 234                         if (hash < pos) /* Skip until we find target hash */
 235                                 continue;
 236
 237                         namelen = le16_to_cpu(ent->lde_namelen);
 238                         if (namelen == 0) /* Skip dummy record */
 239                                 continue;
 240
 241                         if (is_api32 && is_hash64)
 242                                 lhash = hash >> 32;
 243                         else
 244                                 lhash = hash;
 245                         fid_le_to_cpu(&fid, &ent->lde_fid);
 246                         ino = cl_fid_build_ino(&fid, is_api32);
 247                         type = S_DT(lu_dirent_type_get(ent));
 248                         /* For ll_nfs_get_name_filldir(), it will try to access
 249                          * 'ent' through 'lde_name', so the parameter 'name'
 250                          * for 'filldir()' must be part of the 'ent'. */
 251 #ifdef HAVE_DIR_CONTEXT
 252                         ctx->pos = lhash;
 253                         if (!IS_ENCRYPTED(inode)) {
 254                                 done = !dir_emit(ctx, ent->lde_name, namelen,
 255                                                  ino, type);
 256                         } else {
 257                                 /* Directory is encrypted */
 258                                 int save_len = lltr.len;
 259                                 struct llcrypt_str de_name =
 260                                         LLTR_INIT(ent->lde_name, namelen);
 261
 262                                 rc = ll_fname_disk_to_usr(inode, 0, 0, &de_name,
 263                                                           &lltr, &fid);
 264                                 de_name = lltr;
 265                                 lltr.len = save_len;
 266                                 if (rc) {
 267                                         done = 1;
 268                                         break;
 269                                 }
 270                                 done = !dir_emit(ctx, de_name.name, de_name.len,
 271                                                  ino, type);
 272                         }
 273 #else
 274                         /* HAVE_DIR_CONTEXT is defined from kernel 3.11, whereas
 275                          * IS_ENCRYPTED is brought by kernel 4.14.
 276                          * So there is no need to handle encryption case here.
 277                          */
 278                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 279                                        ino, type);
 280 #endif
 281                 }
 282
 283                 if (done) {
 284                         pos = hash;
 285                         ll_release_page(inode, page, false);
 286                         break;
 287                 }
 288
 289                 next = le64_to_cpu(dp->ldp_hash_end);
 290                 pos = next;
 291                 if (pos == MDS_DIR_END_OFF) {
 292                         /*
 293                          * End of directory reached.
 294                          */
 295                         done = 1;
 296                         ll_release_page(inode, page, false);
 297                 } else {
 298                         /*
 299                          * Normal case: continue to the next
 300                          * page.
 301                          */
 302                         ll_release_page(inode, page,
 303                                         le32_to_cpu(dp->ldp_flags) &
 304                                         LDF_COLLIDE);
 305                         next = pos;
 306                         page = ll_get_dir_page(inode, op_data, pos,
 307                                                partial_readdir_rc);
 308                 }
 309         }
 310 #ifdef HAVE_DIR_CONTEXT
 311         ctx->pos = pos;
 312 #else
 313         *ppos = pos;
 314 #endif
 315         llcrypt_fname_free_buffer(&lltr);
 316         RETURN(rc);
 317 }
 318
 319 #ifdef HAVE_DIR_CONTEXT
 320 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 321 #else
 322 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 323 #endif
 324 {
 325         struct inode *inode = file_inode(filp);
 326         struct ll_file_data *lfd = filp->private_data;
 327         struct ll_sb_info *sbi = ll_i2sbi(inode);
 328         bool hash64 = test_bit(LL_SBI_64BIT_HASH, sbi->ll_flags);
 329         int api32 = ll_need_32bit_api(sbi);
 330         struct md_op_data *op_data;
 331         struct lu_fid pfid = { 0 };
 332         ktime_t kstart = ktime_get();
 333         /* result of possible partial readdir */
 334         int partial_readdir_rc = 0;
 335         __u64 pos;
 336         int rc;
 337
 338         ENTRY;
 339
 340         LASSERT(lfd != NULL);
 341         pos = lfd->lfd_pos;
 342
 343         CDEBUG(D_VFSTRACE,
 344                "VFS Op:inode="DFID"(%p) pos/size%lu/%llu 32bit_api %d\n",
 345                PFID(ll_inode2fid(inode)),
 346                inode, (unsigned long)pos, i_size_read(inode), api32);
 347
 348         if (IS_ENCRYPTED(inode)) {
 349                 rc = llcrypt_get_encryption_info(inode);
 350                 if (rc && rc != -ENOKEY)
 351                         GOTO(out, rc);
 352         }
 353
 354         if (pos == MDS_DIR_END_OFF)
 355                 /*
 356                  * end-of-file.
 357                  */
 358                 GOTO(out, rc = 0);
 359
 360         if (unlikely(ll_dir_striped(inode))) {
 361                 /*
 362                  * This is only needed for striped dir to fill ..,
 363                  * see lmv_read_page()
 364                  */
 365                 if (file_dentry(filp)->d_parent != NULL &&
 366                     file_dentry(filp)->d_parent->d_inode != NULL) {
 367                         __u64 ibits = MDS_INODELOCK_LOOKUP;
 368                         struct inode *parent =
 369                                 file_dentry(filp)->d_parent->d_inode;
 370
 371                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 372                                 pfid = *ll_inode2fid(parent);
 373                 }
 374
 375                 /* If it can not find in cache, do lookup .. on the master
 376                  * object */
 377                 if (fid_is_zero(&pfid)) {
 378                         rc = ll_dir_get_parent_fid(inode, &pfid);
 379                         if (rc != 0)
 380                                 RETURN(rc);
 381                 }
 382         }
 383
 384         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 385                                      LUSTRE_OPC_ANY, inode);
 386         if (IS_ERR(op_data))
 387                 GOTO(out, rc = PTR_ERR(op_data));
 388
 389         /* foreign dirs are browsed out of Lustre */
 390         if (unlikely(op_data->op_mea1 != NULL &&
 391                      op_data->op_mea1->lsm_md_magic == LMV_MAGIC_FOREIGN)) {
 392                 ll_finish_md_op_data(op_data);
 393                 RETURN(-ENODATA);
 394         }
 395
 396         op_data->op_fid3 = pfid;
 397
 398 #ifdef HAVE_DIR_CONTEXT
 399         ctx->pos = pos;
 400         rc = ll_dir_read(inode, &pos, op_data, ctx, &partial_readdir_rc);
 401         pos = ctx->pos;
 402 #else
 403         rc = ll_dir_read(inode, &pos, op_data, cookie, filldir,
 404                          &partial_readdir_rc);
 405 #endif
 406         lfd->lfd_pos = pos;
 407         if (!lfd->fd_partial_readdir_rc)
 408                 lfd->fd_partial_readdir_rc = partial_readdir_rc;
 409
 410         if (pos == MDS_DIR_END_OFF) {
 411                 if (api32)
 412                         pos = LL_DIR_END_OFF_32BIT;
 413                 else
 414                         pos = LL_DIR_END_OFF;
 415         } else {
 416                 if (api32 && hash64)
 417                         pos = pos >> 32;
 418         }
 419 #ifdef HAVE_DIR_CONTEXT
 420         ctx->pos = pos;
 421 #else
 422         filp->f_pos = pos;
 423 #endif
 424         ll_finish_md_op_data(op_data);
 425
 426 out:
 427         if (!rc)
 428                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR,
 429                                    ktime_us_delta(ktime_get(), kstart));
 430
 431         RETURN(rc);
 432 }
 433
 434 /**
 435  * Create striped directory with specified stripe(@lump)
 436  *
 437  * \param[in] dparent   the parent of the directory.
 438  * \param[in] lump      the specified stripes.
 439  * \param[in] dirname   the name of the directory.
 440  * \param[in] mode      the specified mode of the directory.
 441  *
 442  * \retval              =0 if striped directory is being created successfully.
 443  *                      <0 if the creation is failed.
 444  */
 445 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
 446                                size_t len, const char *dirname, umode_t mode,
 447                                bool createonly)
 448 {
 449         struct inode *parent = dparent->d_inode;
 450         struct ptlrpc_request *request = NULL;
 451         struct md_op_data *op_data;
 452         struct ll_sb_info *sbi = ll_i2sbi(parent);
 453         struct inode *inode = NULL;
 454         struct dentry dentry = {
 455                 .d_parent = dparent,
 456                 .d_name = {
 457                         .name = dirname,
 458                         .len = strlen(dirname),
 459                         .hash = ll_full_name_hash(dparent, dirname,
 460                                                   strlen(dirname)),
 461                 },
 462         };
 463         bool encrypt = false;
 464         int hash_flags;
 465         int err;
 466
 467         ENTRY;
 468         if (unlikely(!lmv_user_magic_supported(lump->lum_magic)))
 469                 RETURN(-EINVAL);
 470
 471         if (lump->lum_magic != LMV_MAGIC_FOREIGN) {
 472                 CDEBUG(D_VFSTRACE,
 473                        "VFS Op:inode="DFID"(%p) name %s stripe_offset %d, stripe_count: %u\n",
 474                        PFID(ll_inode2fid(parent)), parent, dirname,
 475                        (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 476         } else {
 477                 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lump;
 478
 479                 CDEBUG(D_VFSTRACE,
 480                        "VFS Op:inode="DFID"(%p) name %s foreign, length %u, value '%.*s'\n",
 481                        PFID(ll_inode2fid(parent)), parent, dirname,
 482                        lfm->lfm_length, lfm->lfm_length, lfm->lfm_value);
 483         }
 484
 485         if (lump->lum_stripe_count > 1 &&
 486             !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
 487                 RETURN(-EINVAL);
 488
 489         if (IS_DEADDIR(parent) &&
 490             !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
 491                 RETURN(-ENOENT);
 492
 493         /* MDS < 2.14 doesn't support 'crush' hash type, and cannot handle
 494          * unknown hash if client doesn't set a valid one. switch to fnv_1a_64.
 495          */
 496         if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_CRUSH)) {
 497                 enum lmv_hash_type type = lump->lum_hash_type &
 498                                           LMV_HASH_TYPE_MASK;
 499
 500                 if (type >= LMV_HASH_TYPE_CRUSH ||
 501                     type == LMV_HASH_TYPE_UNKNOWN)
 502                         lump->lum_hash_type = (lump->lum_hash_type ^ type) |
 503                                               LMV_HASH_TYPE_FNV_1A_64;
 504         }
 505
 506         hash_flags = lump->lum_hash_type & ~LMV_HASH_TYPE_MASK;
 507         if (hash_flags & ~LMV_HASH_FLAG_KNOWN)
 508                 RETURN(-EINVAL);
 509
 510         if (unlikely(!lmv_user_magic_supported(cpu_to_le32(lump->lum_magic))))
 511                 lustre_swab_lmv_user_md(lump);
 512
 513         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 514                 mode &= ~current_umask();
 515         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 516         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 517                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 518                                      lump);
 519         if (IS_ERR(op_data))
 520                 RETURN(PTR_ERR(op_data));
 521
 522         op_data->op_dir_depth = ll_i2info(parent)->lli_dir_depth;
 523
 524         if (ll_sbi_has_encrypt(sbi) &&
 525             (IS_ENCRYPTED(parent) ||
 526             unlikely(llcrypt_dummy_context_enabled(parent)))) {
 527                 err = llcrypt_get_encryption_info(parent);
 528                 if (err)
 529                         GOTO(out_op_data, err);
 530                 if (!llcrypt_has_encryption_key(parent))
 531                         GOTO(out_op_data, err = -ENOKEY);
 532                 encrypt = true;
 533         }
 534
 535         if (test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
 536                 /* selinux_dentry_init_security() uses dentry->d_parent and name
 537                  * to determine the security context for the file. So our fake
 538                  * dentry should be real enough for this purpose. */
 539                 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
 540                                               &op_data->op_file_secctx_name,
 541                                               &op_data->op_file_secctx,
 542                                               &op_data->op_file_secctx_size);
 543                 if (err < 0)
 544                         GOTO(out_op_data, err);
 545         }
 546
 547         if (encrypt) {
 548                 err = llcrypt_inherit_context(parent, NULL, op_data, false);
 549                 if (err)
 550                         GOTO(out_op_data, err);
 551         }
 552
 553         op_data->op_cli_flags |= CLI_SET_MEA;
 554         if (createonly)
 555                 op_data->op_bias |= MDS_SETSTRIPE_CREATE;
 556
 557         err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
 558                         from_kuid(&init_user_ns, current_fsuid()),
 559                         from_kgid(&init_user_ns, current_fsgid()),
 560                         current_cap(), 0, &request);
 561         if (err)
 562                 GOTO(out_request, err);
 563
 564         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
 565
 566         err = ll_prep_inode(&inode, &request->rq_pill, parent->i_sb, NULL);
 567         if (err)
 568                 GOTO(out_inode, err);
 569
 570         dentry.d_inode = inode;
 571
 572         if (test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
 573                 /* no need to protect selinux_inode_setsecurity() by
 574                  * inode_lock. Taking it would lead to a client deadlock
 575                  * LU-13617
 576                  */
 577                 err = security_inode_notifysecctx(inode,
 578                                                   op_data->op_file_secctx,
 579                                                   op_data->op_file_secctx_size);
 580         } else {
 581                 err = ll_inode_init_security(&dentry, inode, parent);
 582         }
 583         if (err)
 584                 GOTO(out_inode, err);
 585
 586         if (encrypt) {
 587                 err = ll_set_encflags(inode, op_data->op_file_encctx,
 588                                       op_data->op_file_encctx_size, false);
 589                 if (err)
 590                         GOTO(out_inode, err);
 591         }
 592
 593 out_inode:
 594         iput(inode);
 595 out_request:
 596         ptlrpc_req_finished(request);
 597 out_op_data:
 598         ll_finish_md_op_data(op_data);
 599
 600         return err;
 601 }
 602
 603 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 604                      int set_default)
 605 {
 606         struct ll_sb_info *sbi = ll_i2sbi(inode);
 607         struct md_op_data *op_data;
 608         struct ptlrpc_request *req = NULL;
 609         int rc = 0;
 610         int lum_size;
 611         ENTRY;
 612
 613         if (lump != NULL) {
 614                 switch (lump->lmm_magic) {
 615                 case LOV_USER_MAGIC_V1:
 616                         lum_size = sizeof(struct lov_user_md_v1);
 617                         break;
 618                 case LOV_USER_MAGIC_V3:
 619                         lum_size = sizeof(struct lov_user_md_v3);
 620                         break;
 621                 case LOV_USER_MAGIC_COMP_V1:
 622                         lum_size = ((struct lov_comp_md_v1 *)lump)->lcm_size;
 623                         break;
 624                 case LMV_USER_MAGIC: {
 625                         struct lmv_user_md *lmv = (struct lmv_user_md *)lump;
 626
 627                         /* MDS < 2.14 doesn't support 'crush' hash type, and
 628                          * cannot handle unknown hash if client doesn't set a
 629                          * valid one. switch to fnv_1a_64.
 630                          */
 631                         if (!(exp_connect_flags2(sbi->ll_md_exp) &
 632                               OBD_CONNECT2_CRUSH)) {
 633                                 enum lmv_hash_type type = lmv->lum_hash_type &
 634                                                           LMV_HASH_TYPE_MASK;
 635
 636                                 if (type >= LMV_HASH_TYPE_CRUSH ||
 637                                     type == LMV_HASH_TYPE_UNKNOWN)
 638                                         lmv->lum_hash_type =
 639                                                 (lmv->lum_hash_type ^ type) |
 640                                                 LMV_HASH_TYPE_FNV_1A_64;
 641                         }
 642                         if (lmv->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 643                                 lustre_swab_lmv_user_md(lmv);
 644                         lum_size = sizeof(*lmv);
 645                         break;
 646                 }
 647                 case LOV_USER_MAGIC_SPECIFIC: {
 648                         struct lov_user_md_v3 *v3 =
 649                                 (struct lov_user_md_v3 *)lump;
 650                         if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
 651                                 RETURN(-EINVAL);
 652                         lum_size = lov_user_md_size(v3->lmm_stripe_count,
 653                                                     LOV_USER_MAGIC_SPECIFIC);
 654                         break;
 655                 }
 656                 default:
 657                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 658                                         " %#08x != %#08x nor %#08x\n",
 659                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 660                                         LOV_USER_MAGIC_V3);
 661                         RETURN(-EINVAL);
 662                 }
 663
 664                 /*
 665                  * This is coming from userspace, so should be in
 666                  * local endian.  But the MDS would like it in little
 667                  * endian, so we swab it before we send it.
 668                  */
 669                 if ((__swab32(lump->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
 670                     le32_to_cpu(LOV_MAGIC_MAGIC))
 671                         lustre_swab_lov_user_md(lump, 0);
 672         } else {
 673                 lum_size = sizeof(struct lov_user_md_v1);
 674         }
 675
 676         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 677                                      LUSTRE_OPC_ANY, NULL);
 678         if (IS_ERR(op_data))
 679                 RETURN(PTR_ERR(op_data));
 680
 681         /* swabbing is done in lov_setstripe() on server side */
 682         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
 683         ll_finish_md_op_data(op_data);
 684         ptlrpc_req_finished(req);
 685         if (rc)
 686                 RETURN(rc);
 687
 688         RETURN(rc);
 689 }
 690
 691 static int ll_dir_get_default_layout(struct inode *inode, void **plmm,
 692                                      int *plmm_size,
 693                                      struct ptlrpc_request **request, u64 valid,
 694                                      enum get_default_layout_type type)
 695 {
 696         struct ll_sb_info *sbi = ll_i2sbi(inode);
 697         struct mdt_body   *body;
 698         struct lov_mds_md *lmm = NULL;
 699         struct ptlrpc_request *req = NULL;
 700         int rc, lmm_size;
 701         struct md_op_data *op_data;
 702         struct lu_fid fid;
 703         ENTRY;
 704
 705         rc = ll_get_default_mdsize(sbi, &lmm_size);
 706         if (rc)
 707                 RETURN(rc);
 708
 709         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 710                                      0, lmm_size, LUSTRE_OPC_ANY,
 711                                      NULL);
 712         if (IS_ERR(op_data))
 713                 RETURN(PTR_ERR(op_data));
 714
 715         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 716
 717         if (type == GET_DEFAULT_LAYOUT_ROOT) {
 718                 lu_root_fid(&op_data->op_fid1);
 719                 fid = op_data->op_fid1;
 720         } else {
 721                 fid = *ll_inode2fid(inode);
 722         }
 723
 724         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 725         ll_finish_md_op_data(op_data);
 726         if (rc < 0) {
 727                 CDEBUG(D_INFO, "md_getattr failed on inode "DFID": rc %d\n",
 728                        PFID(&fid), rc);
 729                 GOTO(out, rc);
 730         }
 731
 732         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 733         LASSERT(body != NULL);
 734
 735         lmm_size = body->mbo_eadatasize;
 736
 737         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 738             lmm_size == 0) {
 739                 GOTO(out, rc = -ENODATA);
 740         }
 741
 742         lmm = req_capsule_server_sized_get(&req->rq_pill,
 743                                            &RMF_MDT_MD, lmm_size);
 744         LASSERT(lmm != NULL);
 745
 746         /*
 747          * This is coming from the MDS, so is probably in
 748          * little endian.  We convert it to host endian before
 749          * passing it to userspace.
 750          */
 751         /* We don't swab objects for directories */
 752         switch (le32_to_cpu(lmm->lmm_magic)) {
 753         case LOV_MAGIC_V1:
 754         case LOV_MAGIC_V3:
 755         case LOV_MAGIC_COMP_V1:
 756         case LOV_USER_MAGIC_SPECIFIC:
 757                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 758                         lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
 759                 break;
 760         case LMV_MAGIC_V1:
 761                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 762                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 763                 break;
 764         case LMV_USER_MAGIC:
 765                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 766                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 767                 break;
 768         case LMV_MAGIC_FOREIGN: {
 769                 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lmm;
 770
 771                 if (LMV_MAGIC_FOREIGN != cpu_to_le32(LMV_MAGIC_FOREIGN)) {
 772                         __swab32s(&lfm->lfm_magic);
 773                         __swab32s(&lfm->lfm_length);
 774                         __swab32s(&lfm->lfm_type);
 775                         __swab32s(&lfm->lfm_flags);
 776                 }
 777                 break;
 778         }
 779         default:
 780                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 781                 rc = -EPROTO;
 782         }
 783 out:
 784         *plmm = lmm;
 785         *plmm_size = lmm_size;
 786         *request = req;
 787         return rc;
 788 }
 789
 790 /**
 791  * This function will be used to get default LOV/LMV/Default LMV
 792  * @valid will be used to indicate which stripe it will retrieve.
 793  * If the directory does not have its own default layout, then the
 794  * function will request the default layout from root FID.
 795  *      OBD_MD_MEA              LMV stripe EA
 796  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 797  *      otherwise               Default LOV EA.
 798  * Each time, it can only retrieve 1 stripe EA
 799  **/
 800 int ll_dir_getstripe_default(struct inode *inode, void **plmm, int *plmm_size,
 801                              struct ptlrpc_request **request,
 802                              struct ptlrpc_request **root_request,
 803                              u64 valid)
 804 {
 805         struct ptlrpc_request *req = NULL;
 806         struct ptlrpc_request *root_req = NULL;
 807         struct lov_mds_md *lmm = NULL;
 808         int lmm_size = 0;
 809         int rc = 0;
 810         ENTRY;
 811
 812         rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
 813                                        &req, valid, 0);
 814         if (rc == -ENODATA && !fid_is_root(ll_inode2fid(inode)) &&
 815             !(valid & OBD_MD_MEA) && root_request != NULL) {
 816                 int rc2 = ll_dir_get_default_layout(inode, (void **)&lmm,
 817                                                     &lmm_size, &root_req, valid,
 818                                                     GET_DEFAULT_LAYOUT_ROOT);
 819                 if (rc2 == 0)
 820                         rc = 0;
 821         }
 822
 823         *plmm = lmm;
 824         *plmm_size = lmm_size;
 825         *request = req;
 826         if (root_request != NULL)
 827                 *root_request = root_req;
 828
 829         RETURN(rc);
 830 }
 831
 832 /**
 833  * This function will be used to get default LOV/LMV/Default LMV
 834  * @valid will be used to indicate which stripe it will retrieve
 835  *      OBD_MD_MEA              LMV stripe EA
 836  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 837  *      otherwise               Default LOV EA.
 838  * Each time, it can only retrieve 1 stripe EA
 839  **/
 840 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 841                      struct ptlrpc_request **request, u64 valid)
 842 {
 843         struct ptlrpc_request *req = NULL;
 844         struct lov_mds_md *lmm = NULL;
 845         int lmm_size = 0;
 846         int rc = 0;
 847         ENTRY;
 848
 849         rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
 850                                        &req, valid, 0);
 851
 852         *plmm = lmm;
 853         *plmm_size = lmm_size;
 854         *request = req;
 855
 856         RETURN(rc);
 857 }
 858
 859 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 860 {
 861         struct md_op_data       *op_data;
 862         int                     rc;
 863         int                     mdt_index;
 864         ENTRY;
 865
 866         OBD_ALLOC_PTR(op_data);
 867         if (op_data == NULL)
 868                 RETURN(-ENOMEM);
 869
 870         op_data->op_flags |= MF_GET_MDT_IDX;
 871         op_data->op_fid1 = *fid;
 872         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 873         mdt_index = op_data->op_mds;
 874         OBD_FREE_PTR(op_data);
 875         if (rc < 0)
 876                 RETURN(rc);
 877
 878         RETURN(mdt_index);
 879 }
 880
 881 /*
 882  *  Get MDT index for the inode.
 883  */
 884 int ll_get_mdt_idx(struct inode *inode)
 885 {
 886         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 887 }
 888
 889 /**
 890  * Generic handler to do any pre-copy work.
 891  *
 892  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 893  * first information for it that real work has started.
 894  *
 895  * Moreover, for a ARCHIVE request, it will sample the file data version and
 896  * store it in \a copy.
 897  *
 898  * \return 0 on success.
 899  */
 900 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 901 {
 902         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 903         struct hsm_progress_kernel       hpk;
 904         int                              rc = 0;
 905         int                              rc2;
 906         ENTRY;
 907
 908         /* Forge a hsm_progress based on data from copy. */
 909         hpk.hpk_fid = copy->hc_hai.hai_fid;
 910         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 911         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 912         hpk.hpk_extent.length = 0;
 913         hpk.hpk_flags = 0;
 914         hpk.hpk_errval = 0;
 915         hpk.hpk_data_version = 0;
 916
 917
 918         /* For archive request, we need to read the current file version. */
 919         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 920                 struct inode    *inode;
 921                 __u64            data_version = 0;
 922
 923                 /* Get inode for this fid */
 924                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 925                 if (IS_ERR(inode)) {
 926                         hpk.hpk_flags |= HP_FLAG_RETRY;
 927                         /* hpk_errval is >= 0 */
 928                         hpk.hpk_errval = -PTR_ERR(inode);
 929                         GOTO(progress, rc = PTR_ERR(inode));
 930                 }
 931
 932                 /* Read current file data version */
 933                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 934                 iput(inode);
 935                 if (rc != 0) {
 936                         CDEBUG(D_HSM, "Could not read file data version of "
 937                                       DFID" (rc = %d). Archive request ("
 938                                       "%#llx) could not be done.\n",
 939                                       PFID(&copy->hc_hai.hai_fid), rc,
 940                                       copy->hc_hai.hai_cookie);
 941                         hpk.hpk_flags |= HP_FLAG_RETRY;
 942                         /* hpk_errval must be >= 0 */
 943                         hpk.hpk_errval = -rc;
 944                         GOTO(progress, rc);
 945                 }
 946
 947                 /* Store in the hsm_copy for later copytool use.
 948                  * Always modified even if no lsm. */
 949                 copy->hc_data_version = data_version;
 950         }
 951
 952 progress:
 953         /* On error, the request should be considered as completed */
 954         if (hpk.hpk_errval > 0)
 955                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 956
 957         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 958                             &hpk, NULL);
 959
 960         /* Return first error */
 961         RETURN(rc != 0 ? rc : rc2);
 962 }
 963
 964 /**
 965  * Generic handler to do any post-copy work.
 966  *
 967  * It will send the last hsm_progress update to coordinator to inform it
 968  * that copy is finished and whether it was successful or not.
 969  *
 970  * Moreover,
 971  * - for ARCHIVE request, it will sample the file data version and compare it
 972  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 973  *   will be considered as failed.
 974  * - for RESTORE request, it will sample the file data version and send it to
 975  *   coordinator which is useful if the file was imported as 'released'.
 976  *
 977  * \return 0 on success.
 978  */
 979 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 980 {
 981         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 982         struct hsm_progress_kernel       hpk;
 983         int                              rc = 0;
 984         int                              rc2;
 985         ENTRY;
 986
 987         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 988         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 989          * initialized if copy_end was called with copy == NULL.
 990          */
 991
 992         /* Forge a hsm_progress based on data from copy. */
 993         hpk.hpk_fid = copy->hc_hai.hai_fid;
 994         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 995         hpk.hpk_extent = copy->hc_hai.hai_extent;
 996         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 997         hpk.hpk_errval = copy->hc_errval;
 998         hpk.hpk_data_version = 0;
 999
1000         /* For archive request, we need to check the file data was not changed.
1001          *
1002          * For restore request, we need to send the file data version, this is
1003          * useful when the file was created using hsm_import.
1004          */
1005         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
1006              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
1007             (copy->hc_errval == 0)) {
1008                 struct inode    *inode;
1009                 __u64            data_version = 0;
1010
1011                 /* Get lsm for this fid */
1012                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
1013                 if (IS_ERR(inode)) {
1014                         hpk.hpk_flags |= HP_FLAG_RETRY;
1015                         /* hpk_errval must be >= 0 */
1016                         hpk.hpk_errval = -PTR_ERR(inode);
1017                         GOTO(progress, rc = PTR_ERR(inode));
1018                 }
1019
1020                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
1021                 iput(inode);
1022                 if (rc) {
1023                         CDEBUG(D_HSM, "Could not read file data version. "
1024                                       "Request could not be confirmed.\n");
1025                         if (hpk.hpk_errval == 0)
1026                                 hpk.hpk_errval = -rc;
1027                         GOTO(progress, rc);
1028                 }
1029
1030                 /* Store in the hsm_copy for later copytool use.
1031                  * Always modified even if no lsm. */
1032                 hpk.hpk_data_version = data_version;
1033
1034                 /* File could have been stripped during archiving, so we need
1035                  * to check anyway. */
1036                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1037                     (copy->hc_data_version != data_version)) {
1038                         CDEBUG(D_HSM, "File data version mismatched. "
1039                               "File content was changed during archiving. "
1040                                DFID", start:%#llx current:%#llx\n",
1041                                PFID(&copy->hc_hai.hai_fid),
1042                                copy->hc_data_version, data_version);
1043                         /* File was changed, send error to cdt. Do not ask for
1044                          * retry because if a file is modified frequently,
1045                          * the cdt will loop on retried archive requests.
1046                          * The policy engine will ask for a new archive later
1047                          * when the file will not be modified for some tunable
1048                          * time */
1049                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
1050                         rc = -EBUSY;
1051                         /* hpk_errval must be >= 0 */
1052                         hpk.hpk_errval = -rc;
1053                         GOTO(progress, rc);
1054                 }
1055
1056         }
1057
1058 progress:
1059         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1060                             &hpk, NULL);
1061
1062         /* Return first error */
1063         RETURN(rc != 0 ? rc : rc2);
1064 }
1065
1066
1067 static int copy_and_ct_start(int cmd, struct obd_export *exp,
1068                              const struct lustre_kernelcomm __user *data)
1069 {
1070         struct lustre_kernelcomm *lk;
1071         struct lustre_kernelcomm *tmp;
1072         size_t size = sizeof(*lk);
1073         size_t new_size;
1074         int i;
1075         int rc;
1076
1077         /* copy data from userspace to get numbers of archive_id */
1078         OBD_ALLOC(lk, size);
1079         if (lk == NULL)
1080                 return -ENOMEM;
1081
1082         if (copy_from_user(lk, data, size))
1083                 GOTO(out_lk, rc = -EFAULT);
1084
1085         if (lk->lk_flags & LK_FLG_STOP)
1086                 goto do_ioctl;
1087
1088         if (!(lk->lk_flags & LK_FLG_DATANR)) {
1089                 __u32 archive_mask = lk->lk_data_count;
1090                 int count;
1091
1092                 /* old hsm agent to old MDS */
1093                 if (!exp_connect_archive_id_array(exp))
1094                         goto do_ioctl;
1095
1096                 /* old hsm agent to new MDS */
1097                 lk->lk_flags |= LK_FLG_DATANR;
1098
1099                 if (archive_mask == 0)
1100                         goto do_ioctl;
1101
1102                 count = hweight32(archive_mask);
1103                 new_size = offsetof(struct lustre_kernelcomm, lk_data[count]);
1104                 OBD_ALLOC(tmp, new_size);
1105                 if (tmp == NULL)
1106                         GOTO(out_lk, rc = -ENOMEM);
1107
1108                 memcpy(tmp, lk, size);
1109                 tmp->lk_data_count = count;
1110                 OBD_FREE(lk, size);
1111                 lk = tmp;
1112                 size = new_size;
1113
1114                 count = 0;
1115                 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
1116                         if (BIT(i) & archive_mask) {
1117                                 lk->lk_data[count] = i + 1;
1118                                 count++;
1119                         }
1120                 }
1121                 goto do_ioctl;
1122         }
1123
1124         /* new hsm agent to new mds */
1125         if (lk->lk_data_count > 0) {
1126                 new_size = offsetof(struct lustre_kernelcomm,
1127                                     lk_data[lk->lk_data_count]);
1128                 OBD_ALLOC(tmp, new_size);
1129                 if (tmp == NULL)
1130                         GOTO(out_lk, rc = -ENOMEM);
1131
1132                 OBD_FREE(lk, size);
1133                 lk = tmp;
1134                 size = new_size;
1135
1136                 if (copy_from_user(lk, data, size))
1137                         GOTO(out_lk, rc = -EFAULT);
1138         }
1139
1140         /* new hsm agent to old MDS */
1141         if (!exp_connect_archive_id_array(exp)) {
1142                 __u32 archives = 0;
1143
1144                 if (lk->lk_data_count > LL_HSM_ORIGIN_MAX_ARCHIVE)
1145                         GOTO(out_lk, rc = -EINVAL);
1146
1147                 for (i = 0; i < lk->lk_data_count; i++) {
1148                         if (lk->lk_data[i] > LL_HSM_ORIGIN_MAX_ARCHIVE) {
1149                                 rc = -EINVAL;
1150                                 CERROR("%s: archive id %d requested but only "
1151                                        "[0 - %zu] supported: rc = %d\n",
1152                                        exp->exp_obd->obd_name, lk->lk_data[i],
1153                                        LL_HSM_ORIGIN_MAX_ARCHIVE, rc);
1154                                 GOTO(out_lk, rc);
1155                         }
1156
1157                         if (lk->lk_data[i] == 0) {
1158                                 archives = 0;
1159                                 break;
1160                         }
1161
1162                         archives |= (1 << (lk->lk_data[i] - 1));
1163                 }
1164                 lk->lk_flags &= ~LK_FLG_DATANR;
1165                 lk->lk_data_count = archives;
1166         }
1167 do_ioctl:
1168         rc = obd_iocontrol(cmd, exp, size, lk, NULL);
1169 out_lk:
1170         OBD_FREE(lk, size);
1171         return rc;
1172 }
1173
1174 static int check_owner(int type, int id)
1175 {
1176         switch (type) {
1177         case USRQUOTA:
1178                 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1179                         return -EPERM;
1180                 break;
1181         case GRPQUOTA:
1182                 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1183                         return -EPERM;
1184                 break;
1185         case PRJQUOTA:
1186                 break;
1187         }
1188         return 0;
1189 }
1190
1191 int quotactl_ioctl(struct super_block *sb, struct if_quotactl *qctl)
1192 {
1193         struct ll_sb_info *sbi = ll_s2sbi(sb);
1194         int cmd = qctl->qc_cmd;
1195         int type = qctl->qc_type;
1196         int id = qctl->qc_id;
1197         int valid = qctl->qc_valid;
1198         int rc = 0;
1199
1200         ENTRY;
1201
1202         switch (cmd) {
1203         case Q_SETQUOTA:
1204         case Q_SETINFO:
1205         case LUSTRE_Q_SETDEFAULT:
1206         case LUSTRE_Q_SETQUOTAPOOL:
1207         case LUSTRE_Q_SETINFOPOOL:
1208         case LUSTRE_Q_SETDEFAULT_POOL:
1209         case LUSTRE_Q_DELETEQID:
1210                 if (!capable(CAP_SYS_ADMIN))
1211                         RETURN(-EPERM);
1212
1213                 if (sb->s_flags & SB_RDONLY)
1214                         RETURN(-EROFS);
1215                 break;
1216         case Q_GETQUOTA:
1217         case LUSTRE_Q_GETDEFAULT:
1218         case LUSTRE_Q_GETQUOTAPOOL:
1219         case LUSTRE_Q_GETDEFAULT_POOL:
1220                 if (check_owner(type, id) &&
1221                     (!capable(CAP_SYS_ADMIN)))
1222                         RETURN(-EPERM);
1223                 break;
1224         case Q_GETINFO:
1225         case LUSTRE_Q_GETINFOPOOL:
1226                 break;
1227         default:
1228                 CERROR("unsupported quotactl op: %#x\n", cmd);
1229                 RETURN(-ENOTSUPP);
1230         }
1231
1232         if (valid != QC_GENERAL) {
1233                 if (cmd == Q_GETINFO)
1234                         qctl->qc_cmd = Q_GETOINFO;
1235                 else if (cmd == Q_GETQUOTA ||
1236                          cmd == LUSTRE_Q_GETQUOTAPOOL)
1237                         qctl->qc_cmd = Q_GETOQUOTA;
1238                 else
1239                         RETURN(-EINVAL);
1240
1241                 switch (valid) {
1242                 case QC_MDTIDX:
1243                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1244                                            sizeof(*qctl), qctl, NULL);
1245                         break;
1246                 case QC_OSTIDX:
1247                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1248                                            sizeof(*qctl), qctl, NULL);
1249                         break;
1250                 case QC_UUID:
1251                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1252                                            sizeof(*qctl), qctl, NULL);
1253                         if (rc == -EAGAIN)
1254                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1255                                                    sbi->ll_dt_exp,
1256                                                    sizeof(*qctl), qctl, NULL);
1257                         break;
1258                 default:
1259                         rc = -EINVAL;
1260                         break;
1261                 }
1262
1263                 if (rc)
1264                         RETURN(rc);
1265
1266                 qctl->qc_cmd = cmd;
1267         } else {
1268                 struct obd_quotactl *oqctl;
1269                 int oqctl_len = sizeof(*oqctl);
1270
1271                 if (LUSTRE_Q_CMD_IS_POOL(cmd))
1272                         oqctl_len += LOV_MAXPOOLNAME + 1;
1273
1274                 OBD_ALLOC(oqctl, oqctl_len);
1275                 if (oqctl == NULL)
1276                         RETURN(-ENOMEM);
1277
1278                 QCTL_COPY(oqctl, qctl);
1279                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1280                 if (rc) {
1281                         OBD_FREE(oqctl, oqctl_len);
1282                         RETURN(rc);
1283                 }
1284                 /* If QIF_SPACE is not set, client should collect the
1285                  * space usage from OSSs by itself */
1286                 if ((cmd == Q_GETQUOTA || cmd == LUSTRE_Q_GETQUOTAPOOL) &&
1287                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1288                     !oqctl->qc_dqblk.dqb_curspace) {
1289                         struct obd_quotactl *oqctl_tmp;
1290                         int qctl_len = sizeof(*oqctl_tmp) + LOV_MAXPOOLNAME + 1;
1291
1292                         OBD_ALLOC(oqctl_tmp, qctl_len);
1293                         if (oqctl_tmp == NULL)
1294                                 GOTO(out, rc = -ENOMEM);
1295
1296                         if (cmd == LUSTRE_Q_GETQUOTAPOOL) {
1297                                 oqctl_tmp->qc_cmd = LUSTRE_Q_GETQUOTAPOOL;
1298                                 memcpy(oqctl_tmp->qc_poolname,
1299                                        qctl->qc_poolname,
1300                                        LOV_MAXPOOLNAME + 1);
1301                         } else {
1302                                 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1303                         }
1304                         oqctl_tmp->qc_id = oqctl->qc_id;
1305                         oqctl_tmp->qc_type = oqctl->qc_type;
1306
1307                         /* collect space usage from OSTs */
1308                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1309                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1310                         if (!rc || rc == -EREMOTEIO) {
1311                                 oqctl->qc_dqblk.dqb_curspace =
1312                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1313                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1314                         }
1315
1316                         /* collect space & inode usage from MDTs */
1317                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1318                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1319                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1320                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1321                         if (!rc || rc == -EREMOTEIO) {
1322                                 oqctl->qc_dqblk.dqb_curspace +=
1323                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1324                                 oqctl->qc_dqblk.dqb_curinodes =
1325                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1326                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1327                         } else {
1328                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1329                         }
1330
1331                         OBD_FREE(oqctl_tmp, qctl_len);
1332                 }
1333 out:
1334                 QCTL_COPY(qctl, oqctl);
1335                 OBD_FREE(oqctl, oqctl_len);
1336         }
1337
1338         RETURN(rc);
1339 }
1340
1341 int ll_rmfid(struct file *file, void __user *arg)
1342 {
1343         const struct fid_array __user *ufa = arg;
1344         struct inode *inode = file_inode(file);
1345         struct fid_array *lfa = NULL;
1346         size_t size;
1347         unsigned nr;
1348         int i, rc, *rcs = NULL;
1349         ENTRY;
1350
1351         if (!capable(CAP_DAC_READ_SEARCH) &&
1352             !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
1353                 RETURN(-EPERM);
1354         /* Only need to get the buflen */
1355         if (get_user(nr, &ufa->fa_nr))
1356                 RETURN(-EFAULT);
1357         /* DoS protection */
1358         if (nr > OBD_MAX_FIDS_IN_ARRAY)
1359                 RETURN(-E2BIG);
1360
1361         size = offsetof(struct fid_array, fa_fids[nr]);
1362         OBD_ALLOC(lfa, size);
1363         if (!lfa)
1364                 RETURN(-ENOMEM);
1365         OBD_ALLOC_PTR_ARRAY(rcs, nr);
1366         if (!rcs)
1367                 GOTO(free_lfa, rc = -ENOMEM);
1368
1369         if (copy_from_user(lfa, arg, size))
1370                 GOTO(free_rcs, rc = -EFAULT);
1371
1372         /* Call mdc_iocontrol */
1373         rc = md_rmfid(ll_i2mdexp(file_inode(file)), lfa, rcs, NULL);
1374         if (!rc) {
1375                 for (i = 0; i < nr; i++)
1376                         if (rcs[i])
1377                                 lfa->fa_fids[i].f_ver = rcs[i];
1378                 if (copy_to_user(arg, lfa, size))
1379                         rc = -EFAULT;
1380         }
1381
1382 free_rcs:
1383         OBD_FREE_PTR_ARRAY(rcs, nr);
1384 free_lfa:
1385         OBD_FREE(lfa, size);
1386
1387         RETURN(rc);
1388 }
1389
1390 /* This function tries to get a single name component,
1391  * to send to the server. No actual path traversal involved,
1392  * so we limit to NAME_MAX */
1393 static char *ll_getname(const char __user *filename)
1394 {
1395         int ret = 0, len;
1396         char *tmp;
1397
1398         OBD_ALLOC(tmp, NAME_MAX + 1);
1399
1400         if (!tmp)
1401                 return ERR_PTR(-ENOMEM);
1402
1403         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1404         if (len < 0)
1405                 ret = -ENOENT;
1406         else if (len > NAME_MAX)
1407                 ret = -ENAMETOOLONG;
1408
1409         if (ret) {
1410                 OBD_FREE(tmp, NAME_MAX + 1);
1411                 tmp =  ERR_PTR(ret);
1412         }
1413         return tmp;
1414 }
1415
1416 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1417
1418 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1419 {
1420         struct dentry *dentry = file_dentry(file);
1421         struct inode *inode = file_inode(file);
1422         struct ll_sb_info *sbi = ll_i2sbi(inode);
1423         struct obd_ioctl_data *data = NULL;
1424         int rc = 0;
1425         ENTRY;
1426
1427         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1428                PFID(ll_inode2fid(inode)), inode, cmd);
1429
1430         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1431         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1432                 return -ENOTTY;
1433
1434         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1435         switch (cmd) {
1436         case FS_IOC_GETFLAGS:
1437         case FS_IOC_SETFLAGS:
1438                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1439         case FSFILT_IOC_GETVERSION:
1440         case FS_IOC_GETVERSION:
1441                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1442         /* We need to special case any other ioctls we want to handle,
1443          * to send them to the MDS/OST as appropriate and to properly
1444          * network encode the arg field. */
1445         case FS_IOC_SETVERSION:
1446                 RETURN(-ENOTSUPP);
1447
1448         case LL_IOC_GET_MDTIDX: {
1449                 int mdtidx;
1450
1451                 mdtidx = ll_get_mdt_idx(inode);
1452                 if (mdtidx < 0)
1453                         RETURN(mdtidx);
1454
1455                 if (put_user((int)mdtidx, (int __user *)arg))
1456                         RETURN(-EFAULT);
1457
1458                 return 0;
1459         }
1460         case IOC_MDC_LOOKUP: {
1461                 int namelen, len = 0;
1462                 char *filename;
1463
1464                 rc = obd_ioctl_getdata(&data, &len, (void __user *)arg);
1465                 if (rc != 0)
1466                         RETURN(rc);
1467
1468                 filename = data->ioc_inlbuf1;
1469                 namelen = strlen(filename);
1470                 if (namelen < 1) {
1471                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1472                         GOTO(out_free, rc = -EINVAL);
1473                 }
1474
1475                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1476                 if (rc < 0) {
1477                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1478                                sbi->ll_fsname, namelen, filename, rc);
1479                         GOTO(out_free, rc);
1480                 }
1481 out_free:
1482                 OBD_FREE_LARGE(data, len);
1483                 return rc;
1484         }
1485         case LL_IOC_LMV_SETSTRIPE: {
1486                 struct lmv_user_md  *lum;
1487                 char *filename;
1488                 int namelen = 0;
1489                 int lumlen = 0;
1490                 umode_t mode;
1491                 bool createonly = false;
1492                 int len;
1493                 int rc;
1494
1495                 rc = obd_ioctl_getdata(&data, &len, (void __user *)arg);
1496                 if (rc)
1497                         RETURN(rc);
1498
1499                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1500                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1501                         GOTO(lmv_out_free, rc = -EINVAL);
1502
1503                 filename = data->ioc_inlbuf1;
1504                 namelen = data->ioc_inllen1;
1505
1506                 if (namelen < 1) {
1507                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1508                         GOTO(lmv_out_free, rc = -EINVAL);
1509                 }
1510                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1511                 lumlen = data->ioc_inllen2;
1512
1513                 if (!lmv_user_magic_supported(lum->lum_magic)) {
1514                         CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
1515                                lum->lum_magic, -EINVAL);
1516                         GOTO(lmv_out_free, rc = -EINVAL);
1517                 }
1518
1519                 if ((lum->lum_magic == LMV_USER_MAGIC ||
1520                      lum->lum_magic == LMV_USER_MAGIC_SPECIFIC) &&
1521                     lumlen < sizeof(*lum)) {
1522                         CERROR("%s: wrong lum size %d for magic %x : rc = %d\n",
1523                                filename, lumlen, lum->lum_magic, -EINVAL);
1524                         GOTO(lmv_out_free, rc = -EINVAL);
1525                 }
1526
1527                 if (lum->lum_magic == LMV_MAGIC_FOREIGN &&
1528                     lumlen < sizeof(struct lmv_foreign_md)) {
1529                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1530                                filename, lum->lum_magic, lumlen, -EFAULT);
1531                         GOTO(lmv_out_free, rc = -EINVAL);
1532                 }
1533
1534                 mode = data->ioc_type;
1535                 createonly = data->ioc_obdo1.o_flags & OBD_FL_OBDMDEXISTS;
1536                 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode,
1537                                          createonly);
1538 lmv_out_free:
1539                 OBD_FREE_LARGE(data, len);
1540                 RETURN(rc);
1541
1542         }
1543         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1544                 struct lmv_user_md        lum;
1545                 struct lmv_user_md __user *ulump =
1546                                         (struct lmv_user_md __user *)arg;
1547                 int                       rc;
1548
1549                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1550                         RETURN(-EFAULT);
1551
1552                 if (lum.lum_magic != LMV_USER_MAGIC)
1553                         RETURN(-EINVAL);
1554
1555                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1556
1557                 RETURN(rc);
1558         }
1559         case LL_IOC_LOV_SETSTRIPE_NEW:
1560         case LL_IOC_LOV_SETSTRIPE: {
1561                 struct lov_user_md_v3 *lumv3 = NULL;
1562                 struct lov_user_md_v1 lumv1;
1563                 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1564                 struct lov_user_md_v1 __user *lumv1p =
1565                         (struct lov_user_md_v1 __user *)arg;
1566                 struct lov_user_md_v3 __user *lumv3p =
1567                         (struct lov_user_md_v3 __user *)arg;
1568                 int lum_size = 0;
1569
1570                 int set_default = 0;
1571
1572                 BUILD_BUG_ON(sizeof(struct lov_user_md_v3) <=
1573                              sizeof(struct lov_comp_md_v1));
1574                 BUILD_BUG_ON(sizeof(*lumv3) != sizeof(*lumv3p));
1575                 /* first try with v1 which is smaller than v3 */
1576                 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1577                         RETURN(-EFAULT);
1578
1579                 if (is_root_inode(inode))
1580                         set_default = 1;
1581
1582                 switch (lumv1.lmm_magic) {
1583                 case LOV_USER_MAGIC_V3:
1584                 case LOV_USER_MAGIC_SPECIFIC:
1585                         lum_size = ll_lov_user_md_size(&lumv1);
1586                         if (lum_size < 0)
1587                                 RETURN(lum_size);
1588                         OBD_ALLOC(lumv3, lum_size);
1589                         if (!lumv3)
1590                                 RETURN(-ENOMEM);
1591                         if (copy_from_user(lumv3, lumv3p, lum_size))
1592                                 GOTO(out, rc = -EFAULT);
1593                         lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1594                         break;
1595                 case LOV_USER_MAGIC_V1:
1596                         break;
1597                 default:
1598                         GOTO(out, rc = -ENOTSUPP);
1599                 }
1600
1601                 /* in v1 and v3 cases lumv1 points to data */
1602                 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1603 out:
1604                 if (lumv3)
1605                         OBD_FREE(lumv3, lum_size);
1606                 RETURN(rc);
1607         }
1608         case LL_IOC_LMV_GETSTRIPE: {
1609                 struct lmv_user_md __user *ulmv =
1610                                         (struct lmv_user_md __user *)arg;
1611                 struct lmv_user_md      lum;
1612                 struct ptlrpc_request   *request = NULL;
1613                 struct ptlrpc_request   *root_request = NULL;
1614                 union lmv_mds_md        *lmm = NULL;
1615                 int                     lmmsize;
1616                 u64                     valid = 0;
1617                 struct lmv_user_md      *tmp = NULL;
1618                 int                     mdt_index;
1619                 int                     lum_size;
1620                 int                     stripe_count;
1621                 int                     max_stripe_count;
1622                 int                     i;
1623                 int                     rc;
1624
1625                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1626                         RETURN(-EFAULT);
1627
1628                 max_stripe_count = lum.lum_stripe_count;
1629                 /* lum_magic will indicate which stripe the ioctl will like
1630                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1631                  * is for default LMV stripe */
1632                 if (lum.lum_magic == LMV_MAGIC_V1)
1633                         valid |= OBD_MD_MEA;
1634                 else if (lum.lum_magic == LMV_USER_MAGIC)
1635                         valid |= OBD_MD_DEFAULT_MEA;
1636                 else
1637                         RETURN(-EINVAL);
1638
1639                 rc = ll_dir_getstripe_default(inode, (void **)&lmm, &lmmsize,
1640                                               &request, &root_request, valid);
1641                 if (rc != 0)
1642                         GOTO(finish_req, rc);
1643
1644                 /* Get default LMV EA */
1645                 if (lum.lum_magic == LMV_USER_MAGIC) {
1646                         if (lmmsize > sizeof(*ulmv))
1647                                 GOTO(finish_req, rc = -EINVAL);
1648
1649                         if (root_request != NULL) {
1650                                 struct lmv_user_md *lum;
1651                                 struct ll_inode_info *lli;
1652
1653                                 lum = (struct lmv_user_md *)lmm;
1654                                 lli = ll_i2info(inode);
1655                                 if (lum->lum_max_inherit == LMV_INHERIT_NONE ||
1656                                     (lum->lum_max_inherit > 0 &&
1657                                      lum->lum_max_inherit < lli->lli_dir_depth))
1658                                         GOTO(finish_req, rc = -ENODATA);
1659
1660                                 if (lum->lum_max_inherit ==
1661                                     lli->lli_dir_depth) {
1662                                         lum->lum_max_inherit = LMV_INHERIT_NONE;
1663                                         lum->lum_max_inherit_rr =
1664                                                 LMV_INHERIT_RR_NONE;
1665                                         goto out_copy;
1666                                 }
1667                                 if (lum->lum_max_inherit > lli->lli_dir_depth &&
1668                                     lum->lum_max_inherit <= LMV_INHERIT_MAX)
1669                                         lum->lum_max_inherit -=
1670                                                 lli->lli_dir_depth;
1671
1672                                 if (lum->lum_max_inherit_rr >
1673                                         lli->lli_dir_depth &&
1674                                     lum->lum_max_inherit_rr <=
1675                                         LMV_INHERIT_RR_MAX)
1676                                         lum->lum_max_inherit_rr -=
1677                                                 lli->lli_dir_depth;
1678                                 else if (lum->lum_max_inherit_rr ==
1679                                                 lli->lli_dir_depth)
1680                                         lum->lum_max_inherit_rr =
1681                                                 LMV_INHERIT_RR_NONE;
1682                         }
1683 out_copy:
1684                         if (copy_to_user(ulmv, lmm, lmmsize))
1685                                 GOTO(finish_req, rc = -EFAULT);
1686
1687                         GOTO(finish_req, rc);
1688                 }
1689
1690                 /* if foreign LMV case, fake stripes number */
1691                 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1692                         struct lmv_foreign_md *lfm;
1693
1694                         lfm = (struct lmv_foreign_md *)lmm;
1695                         if (lfm->lfm_length < XATTR_SIZE_MAX -
1696                             offsetof(typeof(*lfm), lfm_value)) {
1697                                 __u32 size = lfm->lfm_length +
1698                                              offsetof(typeof(*lfm), lfm_value);
1699
1700                                 stripe_count = lmv_foreign_to_md_stripes(size);
1701                         } else {
1702                                 CERROR("invalid %d foreign size returned\n",
1703                                             lfm->lfm_length);
1704                                 return -EINVAL;
1705                         }
1706                 } else {
1707                         stripe_count = lmv_mds_md_stripe_count_get(lmm);
1708                 }
1709                 if (max_stripe_count < stripe_count) {
1710                         lum.lum_stripe_count = stripe_count;
1711                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1712                                 GOTO(finish_req, rc = -EFAULT);
1713                         GOTO(finish_req, rc = -E2BIG);
1714                 }
1715
1716                 /* enough room on user side and foreign case */
1717                 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1718                         struct lmv_foreign_md *lfm;
1719                         __u32 size;
1720
1721                         lfm = (struct lmv_foreign_md *)lmm;
1722                         size = lfm->lfm_length +
1723                                offsetof(struct lmv_foreign_md, lfm_value);
1724                         if (copy_to_user(ulmv, lfm, size))
1725                                 GOTO(finish_req, rc = -EFAULT);
1726                         GOTO(finish_req, rc);
1727                 }
1728
1729                 lum_size = lmv_user_md_size(stripe_count,
1730                                             LMV_USER_MAGIC_SPECIFIC);
1731                 OBD_ALLOC(tmp, lum_size);
1732                 if (tmp == NULL)
1733                         GOTO(finish_req, rc = -ENOMEM);
1734
1735                 mdt_index = ll_get_mdt_idx(inode);
1736                 if (mdt_index < 0)
1737                         GOTO(out_tmp, rc = -ENOMEM);
1738
1739                 tmp->lum_magic = LMV_MAGIC_V1;
1740                 tmp->lum_stripe_count = 0;
1741                 tmp->lum_stripe_offset = mdt_index;
1742                 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1743                 for (i = 0; i < stripe_count; i++) {
1744                         struct lu_fid   fid;
1745
1746                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1747                         if (fid_is_sane(&fid)) {
1748                                 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1749                                 if (mdt_index < 0)
1750                                         GOTO(out_tmp, rc = mdt_index);
1751
1752                                 tmp->lum_objects[i].lum_mds = mdt_index;
1753                                 tmp->lum_objects[i].lum_fid = fid;
1754                         }
1755
1756                         tmp->lum_stripe_count++;
1757                 }
1758
1759                 if (copy_to_user(ulmv, tmp, lum_size))
1760                         GOTO(out_tmp, rc = -EFAULT);
1761 out_tmp:
1762                 OBD_FREE(tmp, lum_size);
1763 finish_req:
1764                 ptlrpc_req_finished(request);
1765                 ptlrpc_req_finished(root_request);
1766                 return rc;
1767         }
1768
1769         case LL_IOC_UNLOCK_FOREIGN:
1770                 /* if not a foreign symlink do nothing */
1771                 if (ll_foreign_is_removable(dentry, true)) {
1772                         CDEBUG(D_INFO,
1773                                "prevent rmdir of non-foreign dir ("DFID")\n",
1774                                PFID(ll_inode2fid(inode)));
1775                         RETURN(-EOPNOTSUPP);
1776                 }
1777                 RETURN(0);
1778
1779         case LL_IOC_REMOVE_ENTRY: {
1780                 char            *filename = NULL;
1781                 int              namelen = 0;
1782                 int              rc;
1783
1784                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1785                  * unsupported server, which might crash the server(LU-2730),
1786                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1787                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1788                  * server will support REINT_RMENTRY XXX*/
1789                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1790                         RETURN(-ENOTSUPP);
1791
1792                 filename = ll_getname((const char __user *)arg);
1793                 if (IS_ERR(filename))
1794                         RETURN(PTR_ERR(filename));
1795
1796                 namelen = strlen(filename);
1797                 if (namelen < 1)
1798                         GOTO(out_rmdir, rc = -EINVAL);
1799
1800                 rc = ll_rmdir_entry(inode, filename, namelen);
1801 out_rmdir:
1802                 if (filename)
1803                         ll_putname(filename);
1804                 RETURN(rc);
1805         }
1806         case LL_IOC_RMFID:
1807                 RETURN(ll_rmfid(file, (void __user *)arg));
1808         case LL_IOC_LOV_SWAP_LAYOUTS:
1809                 RETURN(-EPERM);
1810         case IOC_OBD_STATFS:
1811                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1812         case LL_IOC_LOV_GETSTRIPE:
1813         case LL_IOC_LOV_GETSTRIPE_NEW:
1814         case LL_IOC_MDC_GETINFO_V1:
1815         case LL_IOC_MDC_GETINFO_V2:
1816         case IOC_MDC_GETFILEINFO_V1:
1817         case IOC_MDC_GETFILEINFO_V2:
1818         case IOC_MDC_GETFILESTRIPE: {
1819                 struct ptlrpc_request *request = NULL;
1820                 struct ptlrpc_request *root_request = NULL;
1821                 struct lov_user_md __user *lump;
1822                 struct lov_mds_md *lmm = NULL;
1823                 struct mdt_body *body;
1824                 char *filename = NULL;
1825                 lstat_t __user *statp = NULL;
1826                 lstatx_t __user *stxp = NULL;
1827                 __u64 __user *flagsp = NULL;
1828                 __u32 __user *lmmsizep = NULL;
1829                 struct lu_fid __user *fidp = NULL;
1830                 int lmmsize;
1831                 bool api32;
1832
1833                 if (cmd == IOC_MDC_GETFILEINFO_V1 ||
1834                     cmd == IOC_MDC_GETFILEINFO_V2 ||
1835                     cmd == IOC_MDC_GETFILESTRIPE) {
1836                         filename = ll_getname((const char __user *)arg);
1837                         if (IS_ERR(filename))
1838                                 RETURN(PTR_ERR(filename));
1839
1840                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1841                                                       &lmmsize, &request);
1842                 } else {
1843                         rc = ll_dir_getstripe_default(inode, (void **)&lmm,
1844                                                       &lmmsize, &request,
1845                                                       &root_request, 0);
1846                 }
1847
1848                 if (request) {
1849                         body = req_capsule_server_get(&request->rq_pill,
1850                                                       &RMF_MDT_BODY);
1851                         LASSERT(body != NULL);
1852                 } else {
1853                         GOTO(out_req, rc);
1854                 }
1855
1856                 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO_V1 ||
1857                                        cmd == LL_IOC_MDC_GETINFO_V1 ||
1858                                        cmd == IOC_MDC_GETFILEINFO_V2 ||
1859                                        cmd == LL_IOC_MDC_GETINFO_V2)) {
1860                         lmmsize = 0;
1861                         rc = 0;
1862                 }
1863
1864                 if (rc < 0)
1865                         GOTO(out_req, rc);
1866
1867                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1868                     cmd == LL_IOC_LOV_GETSTRIPE ||
1869                     cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1870                         lump = (struct lov_user_md __user *)arg;
1871                 } else if (cmd == IOC_MDC_GETFILEINFO_V1 ||
1872                            cmd == LL_IOC_MDC_GETINFO_V1){
1873                         struct lov_user_mds_data_v1 __user *lmdp;
1874
1875                         lmdp = (struct lov_user_mds_data_v1 __user *)arg;
1876                         statp = &lmdp->lmd_st;
1877                         lump = &lmdp->lmd_lmm;
1878                 } else {
1879                         struct lov_user_mds_data __user *lmdp;
1880
1881                         lmdp = (struct lov_user_mds_data __user *)arg;
1882                         fidp = &lmdp->lmd_fid;
1883                         stxp = &lmdp->lmd_stx;
1884                         flagsp = &lmdp->lmd_flags;
1885                         lmmsizep = &lmdp->lmd_lmmsize;
1886                         lump = &lmdp->lmd_lmm;
1887                 }
1888
1889                 if (lmmsize == 0) {
1890                         /* If the file has no striping then zero out *lump so
1891                          * that the caller isn't confused by garbage. */
1892                         if (clear_user(lump, sizeof(*lump)))
1893                                 GOTO(out_req, rc = -EFAULT);
1894                 } else if (copy_to_user(lump, lmm, lmmsize)) {
1895                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1896                                 GOTO(out_req, rc = -EFAULT);
1897                         rc = -EOVERFLOW;
1898                 }
1899                 api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
1900
1901                 if (cmd == IOC_MDC_GETFILEINFO_V1 ||
1902                     cmd == LL_IOC_MDC_GETINFO_V1) {
1903                         lstat_t st = { 0 };
1904
1905                         st.st_dev       = inode->i_sb->s_dev;
1906                         st.st_mode      = body->mbo_mode;
1907                         st.st_nlink     = body->mbo_nlink;
1908                         st.st_uid       = body->mbo_uid;
1909                         st.st_gid       = body->mbo_gid;
1910                         st.st_rdev      = body->mbo_rdev;
1911                         if (llcrypt_require_key(inode) == -ENOKEY)
1912                                 st.st_size = round_up(st.st_size,
1913                                                    LUSTRE_ENCRYPTION_UNIT_SIZE);
1914                         else
1915                                 st.st_size = body->mbo_size;
1916                         st.st_blksize   = PAGE_SIZE;
1917                         st.st_blocks    = body->mbo_blocks;
1918                         st.st_atime     = body->mbo_atime;
1919                         st.st_mtime     = body->mbo_mtime;
1920                         st.st_ctime     = body->mbo_ctime;
1921                         st.st_ino       = cl_fid_build_ino(&body->mbo_fid1,
1922                                                            api32);
1923
1924                         if (copy_to_user(statp, &st, sizeof(st)))
1925                                 GOTO(out_req, rc = -EFAULT);
1926                 } else if (cmd == IOC_MDC_GETFILEINFO_V2 ||
1927                            cmd == LL_IOC_MDC_GETINFO_V2) {
1928                         lstatx_t stx = { 0 };
1929                         __u64 valid = body->mbo_valid;
1930
1931                         stx.stx_blksize = PAGE_SIZE;
1932                         stx.stx_nlink = body->mbo_nlink;
1933                         stx.stx_uid = body->mbo_uid;
1934                         stx.stx_gid = body->mbo_gid;
1935                         stx.stx_mode = body->mbo_mode;
1936                         stx.stx_ino = cl_fid_build_ino(&body->mbo_fid1,
1937                                                        api32);
1938                         if (llcrypt_require_key(inode) == -ENOKEY)
1939                                 stx.stx_size = round_up(stx.stx_size,
1940                                                    LUSTRE_ENCRYPTION_UNIT_SIZE);
1941                         else
1942                                 stx.stx_size = body->mbo_size;
1943                         stx.stx_blocks = body->mbo_blocks;
1944                         stx.stx_atime.tv_sec = body->mbo_atime;
1945                         stx.stx_ctime.tv_sec = body->mbo_ctime;
1946                         stx.stx_mtime.tv_sec = body->mbo_mtime;
1947                         stx.stx_btime.tv_sec = body->mbo_btime;
1948                         stx.stx_rdev_major = MAJOR(body->mbo_rdev);
1949                         stx.stx_rdev_minor = MINOR(body->mbo_rdev);
1950                         stx.stx_dev_major = MAJOR(inode->i_sb->s_dev);
1951                         stx.stx_dev_minor = MINOR(inode->i_sb->s_dev);
1952                         stx.stx_mask |= STATX_BASIC_STATS | STATX_BTIME;
1953
1954                         /*
1955                          * For a striped directory, the size and blocks returned
1956                          * from MDT is not correct.
1957                          * The size and blocks are aggregated by client across
1958                          * all stripes.
1959                          * Thus for a striped directory, do not return the valid
1960                          * FLSIZE and FLBLOCKS flags to the caller.
1961                          * However, this whould be better decided by the MDS
1962                          * instead of the client.
1963                          */
1964                         if (cmd == LL_IOC_MDC_GETINFO_V2 &&
1965                             ll_i2info(inode)->lli_lsm_md != NULL)
1966                                 valid &= ~(OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
1967
1968                         if (flagsp && copy_to_user(flagsp, &valid,
1969                                                    sizeof(*flagsp)))
1970                                 GOTO(out_req, rc = -EFAULT);
1971
1972                         if (fidp && copy_to_user(fidp, &body->mbo_fid1,
1973                                                  sizeof(*fidp)))
1974                                 GOTO(out_req, rc = -EFAULT);
1975
1976                         if (!(valid & OBD_MD_FLSIZE))
1977                                 stx.stx_mask &= ~STATX_SIZE;
1978                         if (!(valid & OBD_MD_FLBLOCKS))
1979                                 stx.stx_mask &= ~STATX_BLOCKS;
1980
1981                         if (stxp && copy_to_user(stxp, &stx, sizeof(stx)))
1982                                 GOTO(out_req, rc = -EFAULT);
1983
1984                         if (lmmsizep && copy_to_user(lmmsizep, &lmmsize,
1985                                                      sizeof(*lmmsizep)))
1986                                 GOTO(out_req, rc = -EFAULT);
1987                 }
1988
1989                 EXIT;
1990 out_req:
1991                 ptlrpc_req_finished(request);
1992                 ptlrpc_req_finished(root_request);
1993                 if (filename)
1994                         ll_putname(filename);
1995                 return rc;
1996         }
1997         case OBD_IOC_QUOTACTL: {
1998                 struct if_quotactl *qctl;
1999                 int qctl_len = sizeof(*qctl) + LOV_MAXPOOLNAME + 1;
2000
2001                 OBD_ALLOC(qctl, qctl_len);
2002                 if (!qctl)
2003                         RETURN(-ENOMEM);
2004
2005                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
2006                         GOTO(out_quotactl, rc = -EFAULT);
2007
2008                 if (LUSTRE_Q_CMD_IS_POOL(qctl->qc_cmd)) {
2009                         char __user *from = (char __user *)arg +
2010                                         offsetof(typeof(*qctl), qc_poolname);
2011                         if (copy_from_user(qctl->qc_poolname, from,
2012                                            LOV_MAXPOOLNAME + 1))
2013                                 GOTO(out_quotactl, rc = -EFAULT);
2014                 }
2015
2016                 rc = quotactl_ioctl(inode->i_sb, qctl);
2017                 if (rc == 0 &&
2018                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
2019                         rc = -EFAULT;
2020
2021 out_quotactl:
2022                 OBD_FREE(qctl, qctl_len);
2023                 RETURN(rc);
2024         }
2025         case OBD_IOC_GETNAME_OLD:
2026         case OBD_IOC_GETDTNAME:
2027         case OBD_IOC_GETMDNAME:
2028                 RETURN(ll_get_obd_name(inode, cmd, arg));
2029         case LL_IOC_FLUSHCTX:
2030                 RETURN(ll_flush_ctx(inode));
2031         case LL_IOC_GETOBDCOUNT: {
2032                 u32 count, vallen;
2033                 struct obd_export *exp;
2034
2035                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
2036                         RETURN(-EFAULT);
2037
2038                 /* get ost count when count is zero, get mdt count otherwise */
2039                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
2040                 vallen = sizeof(count);
2041                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
2042                                   KEY_TGT_COUNT, &vallen, &count);
2043                 if (rc) {
2044                         CERROR("get target count failed: %d\n", rc);
2045                         RETURN(rc);
2046                 }
2047
2048                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
2049                         RETURN(-EFAULT);
2050
2051                 RETURN(0);
2052         }
2053         case LL_IOC_PATH2FID:
2054                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
2055                                      sizeof(struct lu_fid)))
2056                         RETURN(-EFAULT);
2057                 RETURN(0);
2058         case LL_IOC_GET_CONNECT_FLAGS: {
2059                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
2060                                      (void __user *)arg));
2061         }
2062         case OBD_IOC_FID2PATH:
2063                 RETURN(ll_fid2path(inode, (void __user *)arg));
2064         case LL_IOC_GETPARENT:
2065                 RETURN(ll_getparent(file, (void __user *)arg));
2066         case LL_IOC_FID2MDTIDX: {
2067                 struct obd_export *exp = ll_i2mdexp(inode);
2068                 struct lu_fid     fid;
2069                 __u32             index;
2070
2071                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
2072                                    sizeof(fid)))
2073                         RETURN(-EFAULT);
2074
2075                 /* Call mdc_iocontrol */
2076                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
2077                                    (__u32 __user *)&index);
2078                 if (rc != 0)
2079                         RETURN(rc);
2080
2081                 RETURN(index);
2082         }
2083         case LL_IOC_HSM_REQUEST: {
2084                 struct hsm_user_request *hur;
2085                 ssize_t                  totalsize;
2086
2087                 OBD_ALLOC_PTR(hur);
2088                 if (hur == NULL)
2089                         RETURN(-ENOMEM);
2090
2091                 /* We don't know the true size yet; copy the fixed-size part */
2092                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
2093                         OBD_FREE_PTR(hur);
2094                         RETURN(-EFAULT);
2095                 }
2096
2097                 /* Compute the whole struct size */
2098                 totalsize = hur_len(hur);
2099                 OBD_FREE_PTR(hur);
2100                 if (totalsize < 0)
2101                         RETURN(-E2BIG);
2102
2103                 /* Final size will be more than double totalsize */
2104                 if (totalsize >= MDS_MAXREQSIZE / 3)
2105                         RETURN(-E2BIG);
2106
2107                 OBD_ALLOC_LARGE(hur, totalsize);
2108                 if (hur == NULL)
2109                         RETURN(-ENOMEM);
2110
2111                 /* Copy the whole struct */
2112                 if (copy_from_user(hur, (void __user *)arg, totalsize))
2113                         GOTO(out_hur, rc = -EFAULT);
2114
2115                 if (hur->hur_request.hr_action == HUA_RELEASE) {
2116                         const struct lu_fid *fid;
2117                         struct inode *f;
2118                         int i;
2119
2120                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
2121                                 fid = &hur->hur_user_item[i].hui_fid;
2122                                 f = search_inode_for_lustre(inode->i_sb, fid);
2123                                 if (IS_ERR(f)) {
2124                                         rc = PTR_ERR(f);
2125                                         break;
2126                                 }
2127
2128                                 rc = ll_hsm_release(f);
2129                                 iput(f);
2130                                 if (rc != 0)
2131                                         break;
2132                         }
2133                 } else {
2134                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
2135                                            hur, NULL);
2136                 }
2137
2138 out_hur:
2139                 OBD_FREE_LARGE(hur, totalsize);
2140
2141                 RETURN(rc);
2142         }
2143         case LL_IOC_HSM_PROGRESS: {
2144                 struct hsm_progress_kernel      hpk;
2145                 struct hsm_progress             hp;
2146
2147                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
2148                         RETURN(-EFAULT);
2149
2150                 hpk.hpk_fid = hp.hp_fid;
2151                 hpk.hpk_cookie = hp.hp_cookie;
2152                 hpk.hpk_extent = hp.hp_extent;
2153                 hpk.hpk_flags = hp.hp_flags;
2154                 hpk.hpk_errval = hp.hp_errval;
2155                 hpk.hpk_data_version = 0;
2156
2157                 /* File may not exist in Lustre; all progress
2158                  * reported to Lustre root */
2159                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
2160                                    NULL);
2161                 RETURN(rc);
2162         }
2163         case LL_IOC_HSM_CT_START:
2164                 if (!capable(CAP_SYS_ADMIN))
2165                         RETURN(-EPERM);
2166
2167                 rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
2168                                        (struct lustre_kernelcomm __user *)arg);
2169                 RETURN(rc);
2170
2171         case LL_IOC_HSM_COPY_START: {
2172                 struct hsm_copy *copy;
2173                 int              rc;
2174
2175                 OBD_ALLOC_PTR(copy);
2176                 if (copy == NULL)
2177                         RETURN(-ENOMEM);
2178                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2179                         OBD_FREE_PTR(copy);
2180                         RETURN(-EFAULT);
2181                 }
2182
2183                 rc = ll_ioc_copy_start(inode->i_sb, copy);
2184                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2185                         rc = -EFAULT;
2186
2187                 OBD_FREE_PTR(copy);
2188                 RETURN(rc);
2189         }
2190         case LL_IOC_HSM_COPY_END: {
2191                 struct hsm_copy *copy;
2192                 int              rc;
2193
2194                 OBD_ALLOC_PTR(copy);
2195                 if (copy == NULL)
2196                         RETURN(-ENOMEM);
2197                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2198                         OBD_FREE_PTR(copy);
2199                         RETURN(-EFAULT);
2200                 }
2201
2202                 rc = ll_ioc_copy_end(inode->i_sb, copy);
2203                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2204                         rc = -EFAULT;
2205
2206                 OBD_FREE_PTR(copy);
2207                 RETURN(rc);
2208         }
2209         case LL_IOC_MIGRATE: {
2210                 struct lmv_user_md *lum;
2211                 int len;
2212                 char *filename;
2213                 int namelen = 0;
2214                 __u32 flags;
2215                 int rc;
2216
2217                 rc = obd_ioctl_getdata(&data, &len, (void __user *)arg);
2218                 if (rc)
2219                         RETURN(rc);
2220
2221                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
2222                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
2223                         GOTO(migrate_free, rc = -EINVAL);
2224
2225                 filename = data->ioc_inlbuf1;
2226                 namelen = data->ioc_inllen1;
2227                 flags = data->ioc_type;
2228
2229                 if (namelen < 1 || namelen != strlen(filename) + 1) {
2230                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
2231                         GOTO(migrate_free, rc = -EINVAL);
2232                 }
2233
2234                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
2235                 if (lum->lum_magic != LMV_USER_MAGIC &&
2236                     lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
2237                         rc = -EINVAL;
2238                         CERROR("%s: wrong lum magic %x: rc = %d\n",
2239                                filename, lum->lum_magic, rc);
2240                         GOTO(migrate_free, rc);
2241                 }
2242
2243                 rc = ll_migrate(inode, file, lum, filename, flags);
2244 migrate_free:
2245                 OBD_FREE_LARGE(data, len);
2246
2247                 RETURN(rc);
2248         }
2249         case FS_IOC_FSGETXATTR:
2250                 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
2251         case FS_IOC_FSSETXATTR:
2252                 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
2253         case LL_IOC_PROJECT:
2254                 RETURN(ll_ioctl_project(file, cmd, arg));
2255         case LL_IOC_PCC_DETACH_BY_FID: {
2256                 struct lu_pcc_detach_fid *detach;
2257                 struct lu_fid *fid;
2258                 struct inode *inode2;
2259                 unsigned long ino;
2260
2261                 OBD_ALLOC_PTR(detach);
2262                 if (detach == NULL)
2263                         RETURN(-ENOMEM);
2264
2265                 if (copy_from_user(detach,
2266                                    (const struct lu_pcc_detach_fid __user *)arg,
2267                                    sizeof(*detach)))
2268                         GOTO(out_detach, rc = -EFAULT);
2269
2270                 fid = &detach->pccd_fid;
2271                 ino = cl_fid_build_ino(fid, ll_need_32bit_api(sbi));
2272                 inode2 = ilookup5(inode->i_sb, ino, ll_test_inode_by_fid, fid);
2273                 if (inode2 == NULL)
2274                         /* Target inode is not in inode cache, and PCC file
2275                          * has aleady released, return immdiately.
2276                          */
2277                         GOTO(out_detach, rc = 0);
2278
2279                 if (!S_ISREG(inode2->i_mode))
2280                         GOTO(out_iput, rc = -EINVAL);
2281
2282                 if (!inode_owner_or_capable(&init_user_ns, inode2))
2283                         GOTO(out_iput, rc = -EPERM);
2284
2285                 rc = pcc_ioctl_detach(inode2, detach->pccd_opt);
2286 out_iput:
2287                 iput(inode2);
2288 out_detach:
2289                 OBD_FREE_PTR(detach);
2290                 RETURN(rc);
2291         }
2292 #ifdef HAVE_LUSTRE_CRYPTO
2293         case LL_IOC_SET_ENCRYPTION_POLICY:
2294                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2295                         return -EOPNOTSUPP;
2296                 return llcrypt_ioctl_set_policy(file, (const void __user *)arg);
2297         case LL_IOC_GET_ENCRYPTION_POLICY_EX:
2298                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2299                         return -EOPNOTSUPP;
2300                 return llcrypt_ioctl_get_policy_ex(file, (void __user *)arg);
2301         case LL_IOC_ADD_ENCRYPTION_KEY:
2302                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2303                         return -EOPNOTSUPP;
2304                 return llcrypt_ioctl_add_key(file, (void __user *)arg);
2305         case LL_IOC_REMOVE_ENCRYPTION_KEY:
2306                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2307                         return -EOPNOTSUPP;
2308                 return llcrypt_ioctl_remove_key(file, (void __user *)arg);
2309         case LL_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS:
2310                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2311                         return -EOPNOTSUPP;
2312                 return llcrypt_ioctl_remove_key_all_users(file,
2313                                                           (void __user *)arg);
2314         case LL_IOC_GET_ENCRYPTION_KEY_STATUS:
2315                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2316                         return -EOPNOTSUPP;
2317                 return llcrypt_ioctl_get_key_status(file, (void __user *)arg);
2318 #endif
2319         default:
2320                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
2321                                      (void __user *)arg));
2322         }
2323 }
2324
2325 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
2326 {
2327         struct inode *inode = file->f_mapping->host;
2328         struct ll_file_data *fd = file->private_data;
2329         struct ll_sb_info *sbi = ll_i2sbi(inode);
2330         int api32 = ll_need_32bit_api(sbi);
2331         loff_t ret = -EINVAL;
2332         ENTRY;
2333
2334         inode_lock(inode);
2335         switch (origin) {
2336         case SEEK_SET:
2337                 break;
2338         case SEEK_CUR:
2339                 offset += file->f_pos;
2340                 break;
2341         case SEEK_END:
2342                 if (offset > 0)
2343                         GOTO(out, ret);
2344                 if (api32)
2345                         offset += LL_DIR_END_OFF_32BIT;
2346                 else
2347                         offset += LL_DIR_END_OFF;
2348                 break;
2349         default:
2350                 GOTO(out, ret);
2351         }
2352
2353         if (offset >= 0 &&
2354             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
2355              (!api32 && offset <= LL_DIR_END_OFF))) {
2356                 if (offset != file->f_pos) {
2357                         bool hash64;
2358
2359                         hash64 = test_bit(LL_SBI_64BIT_HASH, sbi->ll_flags);
2360                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
2361                             (!api32 && offset == LL_DIR_END_OFF))
2362                                 fd->lfd_pos = MDS_DIR_END_OFF;
2363                         else if (api32 && hash64)
2364                                 fd->lfd_pos = offset << 32;
2365                         else
2366                                 fd->lfd_pos = offset;
2367                         file->f_pos = offset;
2368                         file->f_version = 0;
2369                 }
2370                 ret = offset;
2371         }
2372         GOTO(out, ret);
2373
2374 out:
2375         inode_unlock(inode);
2376         return ret;
2377 }
2378
2379 static int ll_dir_open(struct inode *inode, struct file *file)
2380 {
2381         ENTRY;
2382         RETURN(ll_file_open(inode, file));
2383 }
2384
2385 static int ll_dir_release(struct inode *inode, struct file *file)
2386 {
2387         ENTRY;
2388         RETURN(ll_file_release(inode, file));
2389 }
2390
2391 /* notify error if partially read striped directory */
2392 static int ll_dir_flush(struct file *file, fl_owner_t id)
2393 {
2394         struct ll_file_data *lfd = file->private_data;
2395         int rc = lfd->fd_partial_readdir_rc;
2396
2397         lfd->fd_partial_readdir_rc = 0;
2398
2399         return rc;
2400 }
2401
2402 const struct file_operations ll_dir_operations = {
2403         .llseek         = ll_dir_seek,
2404         .open           = ll_dir_open,
2405         .release        = ll_dir_release,
2406         .read           = generic_read_dir,
2407 #ifdef HAVE_DIR_CONTEXT
2408         .iterate_shared = ll_iterate,
2409 #else
2410         .readdir        = ll_readdir,
2411 #endif
2412         .unlocked_ioctl = ll_dir_ioctl,
2413         .fsync          = ll_fsync,
2414         .flush          = ll_dir_flush,
2415 };