lustre/llite/dir.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/dir.c
  33  *
  34  * Directory code for lustre client.
  35  */
  36
  37 #include <linux/fs.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/mm.h>
  40 #include <linux/version.h>
  41 #include <linux/security.h>
  42 #include <linux/user_namespace.h>
  43 #ifdef HAVE_UIDGID_HEADER
  44 # include <linux/uidgid.h>
  45 #endif
  46 #include <linux/uaccess.h>
  47 #include <linux/buffer_head.h>   // for wait_on_buffer
  48 #include <linux/pagevec.h>
  49
  50 #define DEBUG_SUBSYSTEM S_LLITE
  51
  52 #include <obd_support.h>
  53 #include <obd_class.h>
  54 #include <uapi/linux/lustre/lustre_ioctl.h>
  55 #include <lustre_lib.h>
  56 #include <lustre_dlm.h>
  57 #include <lustre_fid.h>
  58 #include <lustre_kernelcomm.h>
  59 #include <lustre_swab.h>
  60
  61 #include "llite_internal.h"
  62
  63 /*
  64  * (new) readdir implementation overview.
  65  *
  66  * Original lustre readdir implementation cached exact copy of raw directory
  67  * pages on the client. These pages were indexed in client page cache by
  68  * logical offset in the directory file. This design, while very simple and
  69  * intuitive had some inherent problems:
  70  *
  71  *     . it implies that byte offset to the directory entry serves as a
  72  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  73  *     ext3/htree directory entries may move due to splits, and more
  74  *     importantly,
  75  *
  76  *     . it is incompatible with the design of split directories for cmd3,
  77  *     that assumes that names are distributed across nodes based on their
  78  *     hash, and so readdir should be done in hash order.
  79  *
  80  * New readdir implementation does readdir in hash order, and uses hash of a
  81  * file name as a telldir/seekdir cookie. This led to number of complications:
  82  *
  83  *     . hash is not unique, so it cannot be used to index cached directory
  84  *     pages on the client (note, that it requires a whole pageful of hash
  85  *     collided entries to cause two pages to have identical hashes);
  86  *
  87  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  88  *     entry cookie. ext3/htree has the same problem and lustre implementation
  89  *     mimics their solution: seekdir(hash) positions directory at the first
  90  *     entry with the given hash.
  91  *
  92  * Client side.
  93  *
  94  * 0. caching
  95  *
  96  * Client caches directory pages using hash of the first entry as an index. As
  97  * noted above hash is not unique, so this solution doesn't work as is:
  98  * special processing is needed for "page hash chains" (i.e., sequences of
  99  * pages filled with entries all having the same hash value).
 100  *
 101  * First, such chains have to be detected. To this end, server returns to the
 102  * client the hash of the first entry on the page next to one returned. When
 103  * client detects that this hash is the same as hash of the first entry on the
 104  * returned page, page hash collision has to be handled. Pages in the
 105  * hash chain, except first one, are termed "overflow pages".
 106  *
 107  * Solution to index uniqueness problem is to not cache overflow
 108  * pages. Instead, when page hash collision is detected, all overflow pages
 109  * from emerging chain are immediately requested from the server and placed in
 110  * a special data structure (struct ll_dir_chain). This data structure is used
 111  * by ll_readdir() to process entries from overflow pages. When readdir
 112  * invocation finishes, overflow pages are discarded. If page hash collision
 113  * chain weren't completely processed, next call to readdir will again detect
 114  * page hash collision, again read overflow pages in, process next portion of
 115  * entries and again discard the pages. This is not as wasteful as it looks,
 116  * because, given reasonable hash, page hash collisions are extremely rare.
 117  *
 118  * 1. directory positioning
 119  *
 120  * When seekdir(hash) is called, original
 121  *
 122  *
 123  *
 124  *
 125  *
 126  *
 127  *
 128  *
 129  * Server.
 130  *
 131  * identification of and access to overflow pages
 132  *
 133  * page format
 134  *
 135  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 136  * a header lu_dirpage which describes the start/end hash, and whether this
 137  * page is empty (contains no dir entry) or hash collide with next page.
 138  * After client receives reply, several pages will be integrated into dir page
 139  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
 140  * lu_dirpage for this integrated page will be adjusted. See
 141  * mdc_adjust_dirpages().
 142  *
 143  */
 144 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
 145                              __u64 offset, struct ll_dir_chain *chain)
 146 {
 147         struct md_callback      cb_op;
 148         struct page             *page;
 149         int                     rc;
 150
 151         cb_op.md_blocking_ast = ll_md_blocking_ast;
 152         rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
 153         if (rc != 0)
 154                 return ERR_PTR(rc);
 155
 156         return page;
 157 }
 158
 159 void ll_release_page(struct inode *inode, struct page *page,
 160                      bool remove)
 161 {
 162         kunmap(page);
 163
 164         /* Always remove the page for striped dir, because the page is
 165          * built from temporarily in LMV layer */
 166         if (inode != NULL && S_ISDIR(inode->i_mode) &&
 167             ll_i2info(inode)->lli_lsm_md != NULL) {
 168                 __free_page(page);
 169                 return;
 170         }
 171
 172         if (remove) {
 173                 lock_page(page);
 174                 if (likely(page->mapping != NULL))
 175                         truncate_complete_page(page->mapping, page);
 176                 unlock_page(page);
 177         }
 178         put_page(page);
 179 }
 180
 181 /**
 182  * return IF_* type for given lu_dirent entry.
 183  * IF_* flag shld be converted to particular OS file type in
 184  * platform llite module.
 185  */
 186 static u16 ll_dirent_type_get(struct lu_dirent *ent)
 187 {
 188         u16 type = 0;
 189         struct luda_type *lt;
 190         int len = 0;
 191
 192         if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
 193                 const unsigned align = sizeof(struct luda_type) - 1;
 194
 195                 len = le16_to_cpu(ent->lde_namelen);
 196                 len = (len + align) & ~align;
 197                 lt = (void *)ent->lde_name + len;
 198                 type = IFTODT(le16_to_cpu(lt->lt_type));
 199         }
 200
 201         return type;
 202 }
 203
 204 #ifdef HAVE_DIR_CONTEXT
 205 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 206                 struct dir_context *ctx)
 207 {
 208 #else
 209 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 210                 void *cookie, filldir_t filldir)
 211 {
 212 #endif
 213         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 214         __u64                 pos        = *ppos;
 215         bool                  is_api32 = ll_need_32bit_api(sbi);
 216         bool                  is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 217         struct page          *page;
 218         struct ll_dir_chain   chain;
 219         bool                  done = false;
 220         int                   rc = 0;
 221         ENTRY;
 222
 223         ll_dir_chain_init(&chain);
 224
 225         page = ll_get_dir_page(inode, op_data, pos, &chain);
 226
 227         while (rc == 0 && !done) {
 228                 struct lu_dirpage *dp;
 229                 struct lu_dirent  *ent;
 230                 __u64 hash;
 231                 __u64 next;
 232
 233                 if (IS_ERR(page)) {
 234                         rc = PTR_ERR(page);
 235                         break;
 236                 }
 237
 238                 hash = MDS_DIR_END_OFF;
 239                 dp = page_address(page);
 240                 for (ent = lu_dirent_start(dp); ent != NULL && !done;
 241                      ent = lu_dirent_next(ent)) {
 242                         __u16          type;
 243                         int            namelen;
 244                         struct lu_fid  fid;
 245                         __u64          lhash;
 246                         __u64          ino;
 247
 248                         hash = le64_to_cpu(ent->lde_hash);
 249                         if (hash < pos) /* Skip until we find target hash */
 250                                 continue;
 251
 252                         namelen = le16_to_cpu(ent->lde_namelen);
 253                         if (namelen == 0) /* Skip dummy record */
 254                                 continue;
 255
 256                         if (is_api32 && is_hash64)
 257                                 lhash = hash >> 32;
 258                         else
 259                                 lhash = hash;
 260                         fid_le_to_cpu(&fid, &ent->lde_fid);
 261                         ino = cl_fid_build_ino(&fid, is_api32);
 262                         type = ll_dirent_type_get(ent);
 263                         /* For ll_nfs_get_name_filldir(), it will try to access
 264                          * 'ent' through 'lde_name', so the parameter 'name'
 265                          * for 'filldir()' must be part of the 'ent'. */
 266 #ifdef HAVE_DIR_CONTEXT
 267                         ctx->pos = lhash;
 268                         done = !dir_emit(ctx, ent->lde_name, namelen, ino,
 269                                          type);
 270 #else
 271                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 272                                        ino, type);
 273 #endif
 274                 }
 275
 276                 if (done) {
 277                         pos = hash;
 278                         ll_release_page(inode, page, false);
 279                         break;
 280                 }
 281
 282                 next = le64_to_cpu(dp->ldp_hash_end);
 283                 pos = next;
 284                 if (pos == MDS_DIR_END_OFF) {
 285                         /*
 286                          * End of directory reached.
 287                          */
 288                         done = 1;
 289                         ll_release_page(inode, page, false);
 290                 } else {
 291                         /*
 292                          * Normal case: continue to the next
 293                          * page.
 294                          */
 295                         ll_release_page(inode, page,
 296                                         le32_to_cpu(dp->ldp_flags) &
 297                                         LDF_COLLIDE);
 298                         next = pos;
 299                         page = ll_get_dir_page(inode, op_data, pos,
 300                                                &chain);
 301                 }
 302         }
 303 #ifdef HAVE_DIR_CONTEXT
 304         ctx->pos = pos;
 305 #else
 306         *ppos = pos;
 307 #endif
 308         ll_dir_chain_fini(&chain);
 309         RETURN(rc);
 310 }
 311
 312 #ifdef HAVE_DIR_CONTEXT
 313 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 314 #else
 315 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 316 #endif
 317 {
 318         struct inode            *inode  = file_inode(filp);
 319         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 320         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 321         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 322         int                     api32   = ll_need_32bit_api(sbi);
 323         struct md_op_data       *op_data;
 324         __u64                   pos;
 325         int                     rc;
 326         ENTRY;
 327
 328         if (lfd != NULL)
 329                 pos = lfd->lfd_pos;
 330         else
 331                 pos = 0;
 332
 333         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 334                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 335                inode, (unsigned long)pos, i_size_read(inode), api32);
 336
 337         if (pos == MDS_DIR_END_OFF)
 338                 /*
 339                  * end-of-file.
 340                  */
 341                 GOTO(out, rc = 0);
 342
 343         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 344                                      LUSTRE_OPC_ANY, inode);
 345         if (IS_ERR(op_data))
 346                 GOTO(out, rc = PTR_ERR(op_data));
 347
 348         if (unlikely(op_data->op_mea1 != NULL)) {
 349                 /* This is only needed for striped dir to fill ..,
 350                  * see lmv_read_entry */
 351                 if (file_dentry(filp)->d_parent != NULL &&
 352                     file_dentry(filp)->d_parent->d_inode != NULL) {
 353                         __u64 ibits = MDS_INODELOCK_UPDATE;
 354                         struct inode *parent =
 355                                 file_dentry(filp)->d_parent->d_inode;
 356
 357                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 358                                 op_data->op_fid3 = *ll_inode2fid(parent);
 359                 }
 360
 361                 /* If it can not find in cache, do lookup .. on the master
 362                  * object */
 363                 if (fid_is_zero(&op_data->op_fid3)) {
 364                         rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
 365                         if (rc != 0) {
 366                                 ll_finish_md_op_data(op_data);
 367                                 RETURN(rc);
 368                         }
 369                 }
 370         }
 371 #ifdef HAVE_DIR_CONTEXT
 372         ctx->pos = pos;
 373         rc = ll_dir_read(inode, &pos, op_data, ctx);
 374         pos = ctx->pos;
 375 #else
 376         rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
 377 #endif
 378         if (lfd != NULL)
 379                 lfd->lfd_pos = pos;
 380
 381         if (pos == MDS_DIR_END_OFF) {
 382                 if (api32)
 383                         pos = LL_DIR_END_OFF_32BIT;
 384                 else
 385                         pos = LL_DIR_END_OFF;
 386         } else {
 387                 if (api32 && hash64)
 388                         pos = pos >> 32;
 389         }
 390 #ifdef HAVE_DIR_CONTEXT
 391         ctx->pos = pos;
 392 #else
 393         filp->f_pos = pos;
 394 #endif
 395         ll_finish_md_op_data(op_data);
 396
 397 out:
 398         if (!rc)
 399                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 400
 401         RETURN(rc);
 402 }
 403
 404 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 405 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 406 {
 407         struct mgs_send_param *msp;
 408         int rc = 0;
 409
 410         OBD_ALLOC_PTR(msp);
 411         if (!msp)
 412                 return -ENOMEM;
 413
 414         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 415         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 416                                 sizeof(struct mgs_send_param), msp, NULL);
 417         if (rc)
 418                 CERROR("Failed to set parameter: %d\n", rc);
 419         OBD_FREE_PTR(msp);
 420
 421         return rc;
 422 }
 423 #endif
 424
 425 /**
 426  * Create striped directory with specified stripe(@lump)
 427  *
 428  * \param[in] dparent   the parent of the directory.
 429  * \param[in] lump      the specified stripes.
 430  * \param[in] dirname   the name of the directory.
 431  * \param[in] mode      the specified mode of the directory.
 432  *
 433  * \retval              =0 if striped directory is being created successfully.
 434  *                      <0 if the creation is failed.
 435  */
 436 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
 437                                size_t len, const char *dirname, umode_t mode)
 438 {
 439         struct inode *parent = dparent->d_inode;
 440         struct ptlrpc_request *request = NULL;
 441         struct md_op_data *op_data;
 442         struct ll_sb_info *sbi = ll_i2sbi(parent);
 443         struct inode *inode = NULL;
 444         struct dentry dentry = {
 445                 .d_parent = dparent,
 446                 .d_name = {
 447                         .name = dirname,
 448                         .len = strlen(dirname),
 449                         .hash = ll_full_name_hash(dparent, dirname,
 450                                                   strlen(dirname)),
 451                 },
 452         };
 453         int err;
 454         ENTRY;
 455
 456         if (unlikely(lump->lum_magic != LMV_USER_MAGIC &&
 457                      lump->lum_magic != LMV_USER_MAGIC_SPECIFIC))
 458                 RETURN(-EINVAL);
 459
 460         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s "
 461                "stripe_offset %d, stripe_count: %u\n",
 462                PFID(ll_inode2fid(parent)), parent, dirname,
 463                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 464
 465         if (lump->lum_stripe_count > 1 &&
 466             !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
 467                 RETURN(-EINVAL);
 468
 469         if (IS_DEADDIR(parent) &&
 470             !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
 471                 RETURN(-ENOENT);
 472
 473         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
 474             lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
 475                 lustre_swab_lmv_user_md(lump);
 476
 477         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 478                 mode &= ~current_umask();
 479         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 480         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 481                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 482                                      lump);
 483         if (IS_ERR(op_data))
 484                 RETURN(PTR_ERR(op_data));
 485
 486         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 487                 /* selinux_dentry_init_security() uses dentry->d_parent and name
 488                  * to determine the security context for the file. So our fake
 489                  * dentry should be real enough for this purpose. */
 490                 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
 491                                               &op_data->op_file_secctx_name,
 492                                               &op_data->op_file_secctx,
 493                                               &op_data->op_file_secctx_size);
 494                 if (err < 0)
 495                         GOTO(out_op_data, err);
 496         }
 497
 498         op_data->op_cli_flags |= CLI_SET_MEA;
 499         err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
 500                         from_kuid(&init_user_ns, current_fsuid()),
 501                         from_kgid(&init_user_ns, current_fsgid()),
 502                         cfs_curproc_cap_pack(), 0, &request);
 503         if (err)
 504                 GOTO(out_request, err);
 505
 506         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
 507
 508         err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
 509         if (err)
 510                 GOTO(out_inode, err);
 511
 512         dentry.d_inode = inode;
 513
 514         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 515                 inode_lock(inode);
 516                 err = security_inode_notifysecctx(inode,
 517                                                   op_data->op_file_secctx,
 518                                                   op_data->op_file_secctx_size);
 519                 inode_unlock(inode);
 520         } else {
 521                 err = ll_inode_init_security(&dentry, inode, parent);
 522         }
 523         if (err)
 524                 GOTO(out_inode, err);
 525
 526 out_inode:
 527         if (inode != NULL)
 528                 iput(inode);
 529 out_request:
 530         ptlrpc_req_finished(request);
 531 out_op_data:
 532         ll_finish_md_op_data(op_data);
 533
 534         return err;
 535 }
 536
 537 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 538                      int set_default)
 539 {
 540         struct ll_sb_info *sbi = ll_i2sbi(inode);
 541         struct md_op_data *op_data;
 542         struct ptlrpc_request *req = NULL;
 543         int rc = 0;
 544 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 545         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 546         struct obd_device *mgc = lsi->lsi_mgc;
 547 #endif
 548         int lum_size;
 549         ENTRY;
 550
 551         if (lump != NULL) {
 552                 /*
 553                  * This is coming from userspace, so should be in
 554                  * local endian.  But the MDS would like it in little
 555                  * endian, so we swab it before we send it.
 556                  */
 557                 switch (lump->lmm_magic) {
 558                 case LOV_USER_MAGIC_V1: {
 559                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 560                                 lustre_swab_lov_user_md_v1(lump);
 561                         lum_size = sizeof(struct lov_user_md_v1);
 562                         break;
 563                 }
 564                 case LOV_USER_MAGIC_V3: {
 565                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 566                                 lustre_swab_lov_user_md_v3(
 567                                         (struct lov_user_md_v3 *)lump);
 568                         lum_size = sizeof(struct lov_user_md_v3);
 569                         break;
 570                 }
 571                 case LOV_USER_MAGIC_COMP_V1: {
 572                         if (lump->lmm_magic !=
 573                             cpu_to_le32(LOV_USER_MAGIC_COMP_V1))
 574                                 lustre_swab_lov_comp_md_v1(
 575                                         (struct lov_comp_md_v1 *)lump);
 576                         lum_size = le32_to_cpu(
 577                                 ((struct lov_comp_md_v1 *)lump)->lcm_size);
 578                         break;
 579                 }
 580                 case LMV_USER_MAGIC: {
 581                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 582                                 lustre_swab_lmv_user_md(
 583                                         (struct lmv_user_md *)lump);
 584                         lum_size = sizeof(struct lmv_user_md);
 585                         break;
 586                 }
 587                 case LOV_USER_MAGIC_SPECIFIC: {
 588                         struct lov_user_md_v3 *v3 =
 589                                         (struct lov_user_md_v3 *)lump;
 590                         if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
 591                                 RETURN(-EINVAL);
 592                         if (lump->lmm_magic !=
 593                             cpu_to_le32(LOV_USER_MAGIC_SPECIFIC)) {
 594                                 lustre_swab_lov_user_md_v3(v3);
 595                                 lustre_swab_lov_user_md_objects(v3->lmm_objects,
 596                                                 v3->lmm_stripe_count);
 597                         }
 598                         lum_size = lov_user_md_size(v3->lmm_stripe_count,
 599                                                     LOV_USER_MAGIC_SPECIFIC);
 600                         break;
 601                 }
 602                 default: {
 603                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 604                                         " %#08x != %#08x nor %#08x\n",
 605                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 606                                         LOV_USER_MAGIC_V3);
 607                         RETURN(-EINVAL);
 608                 }
 609                 }
 610         } else {
 611                 lum_size = sizeof(struct lov_user_md_v1);
 612         }
 613
 614         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 615                                      LUSTRE_OPC_ANY, NULL);
 616         if (IS_ERR(op_data))
 617                 RETURN(PTR_ERR(op_data));
 618
 619         /* swabbing is done in lov_setstripe() on server side */
 620         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
 621         ll_finish_md_op_data(op_data);
 622         ptlrpc_req_finished(req);
 623         if (rc)
 624                 RETURN(rc);
 625
 626 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 627         /*
 628          * 2.9 server has stored filesystem default stripe in ROOT xattr,
 629          * and it's stored into system config for backward compatibility.
 630          *
 631          * In the following we use the fact that LOV_USER_MAGIC_V1 and
 632          * LOV_USER_MAGIC_V3 have the same initial fields so we do not
 633          * need the make the distiction between the 2 versions
 634          */
 635         if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
 636             (lump == NULL ||
 637              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
 638              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
 639                 char *param = NULL;
 640                 char *buf;
 641
 642                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 643                 if (param == NULL)
 644                         GOTO(end, rc = -ENOMEM);
 645
 646                 buf = param;
 647                 /* Get fsname and assume devname to be -MDT0000. */
 648                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 649                 strcat(buf, "-MDT0000.lov");
 650                 buf += strlen(buf);
 651
 652                 /* Set root stripesize */
 653                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
 654                          lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 655                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 656                 if (rc)
 657                         GOTO(end, rc);
 658
 659                 /* Set root stripecount */
 660                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
 661                          lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 662                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 663                 if (rc)
 664                         GOTO(end, rc);
 665
 666                 /* Set root stripeoffset */
 667                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
 668                          lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 669                                 (typeof(lump->lmm_stripe_offset))(-1));
 670                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 671
 672 end:
 673                 if (param != NULL)
 674                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 675         }
 676 #endif
 677         RETURN(rc);
 678 }
 679
 680 /**
 681  * This function will be used to get default LOV/LMV/Default LMV
 682  * @valid will be used to indicate which stripe it will retrieve
 683  *      OBD_MD_MEA              LMV stripe EA
 684  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 685  *      otherwise               Default LOV EA.
 686  * Each time, it can only retrieve 1 stripe EA
 687  **/
 688 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 689                      struct ptlrpc_request **request, u64 valid)
 690 {
 691         struct ll_sb_info *sbi = ll_i2sbi(inode);
 692         struct mdt_body   *body;
 693         struct lov_mds_md *lmm = NULL;
 694         struct ptlrpc_request *req = NULL;
 695         int rc, lmm_size;
 696         struct md_op_data *op_data;
 697         ENTRY;
 698
 699         rc = ll_get_default_mdsize(sbi, &lmm_size);
 700         if (rc)
 701                 RETURN(rc);
 702
 703         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 704                                      0, lmm_size, LUSTRE_OPC_ANY,
 705                                      NULL);
 706         if (IS_ERR(op_data))
 707                 RETURN(PTR_ERR(op_data));
 708
 709         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 710         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 711         ll_finish_md_op_data(op_data);
 712         if (rc < 0) {
 713                 CDEBUG(D_INFO, "md_getattr failed on inode "
 714                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 715                 GOTO(out, rc);
 716         }
 717
 718         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 719         LASSERT(body != NULL);
 720
 721         lmm_size = body->mbo_eadatasize;
 722
 723         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 724             lmm_size == 0) {
 725                 GOTO(out, rc = -ENODATA);
 726         }
 727
 728         lmm = req_capsule_server_sized_get(&req->rq_pill,
 729                                            &RMF_MDT_MD, lmm_size);
 730         LASSERT(lmm != NULL);
 731
 732         /*
 733          * This is coming from the MDS, so is probably in
 734          * little endian.  We convert it to host endian before
 735          * passing it to userspace.
 736          */
 737         /* We don't swab objects for directories */
 738         switch (le32_to_cpu(lmm->lmm_magic)) {
 739         case LOV_MAGIC_V1:
 740                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 741                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 742                 break;
 743         case LOV_MAGIC_V3:
 744                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 745                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 746                 break;
 747         case LOV_MAGIC_COMP_V1:
 748                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 749                         lustre_swab_lov_comp_md_v1(
 750                                         (struct lov_comp_md_v1 *)lmm);
 751                 break;
 752         case LMV_MAGIC_V1:
 753                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 754                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 755                 break;
 756         case LMV_USER_MAGIC:
 757                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 758                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 759                 break;
 760         case LOV_USER_MAGIC_SPECIFIC: {
 761                 struct lov_user_md_v3 *v3 = (struct lov_user_md_v3 *)lmm;
 762
 763                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
 764                         lustre_swab_lov_user_md_v3(v3);
 765                         lustre_swab_lov_user_md_objects(v3->lmm_objects,
 766                                                         v3->lmm_stripe_count);
 767                         }
 768                 }
 769                 break;
 770         default:
 771                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 772                 rc = -EPROTO;
 773         }
 774 out:
 775         *plmm = lmm;
 776         *plmm_size = lmm_size;
 777         *request = req;
 778         return rc;
 779 }
 780
 781 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 782 {
 783         struct md_op_data       *op_data;
 784         int                     rc;
 785         int                     mdt_index;
 786         ENTRY;
 787
 788         OBD_ALLOC_PTR(op_data);
 789         if (op_data == NULL)
 790                 RETURN(-ENOMEM);
 791
 792         op_data->op_flags |= MF_GET_MDT_IDX;
 793         op_data->op_fid1 = *fid;
 794         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 795         mdt_index = op_data->op_mds;
 796         OBD_FREE_PTR(op_data);
 797         if (rc < 0)
 798                 RETURN(rc);
 799
 800         RETURN(mdt_index);
 801 }
 802
 803 /*
 804  *  Get MDT index for the inode.
 805  */
 806 int ll_get_mdt_idx(struct inode *inode)
 807 {
 808         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 809 }
 810
 811 /**
 812  * Generic handler to do any pre-copy work.
 813  *
 814  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 815  * first information for it that real work has started.
 816  *
 817  * Moreover, for a ARCHIVE request, it will sample the file data version and
 818  * store it in \a copy.
 819  *
 820  * \return 0 on success.
 821  */
 822 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 823 {
 824         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 825         struct hsm_progress_kernel       hpk;
 826         int                              rc = 0;
 827         int                              rc2;
 828         ENTRY;
 829
 830         /* Forge a hsm_progress based on data from copy. */
 831         hpk.hpk_fid = copy->hc_hai.hai_fid;
 832         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 833         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 834         hpk.hpk_extent.length = 0;
 835         hpk.hpk_flags = 0;
 836         hpk.hpk_errval = 0;
 837         hpk.hpk_data_version = 0;
 838
 839
 840         /* For archive request, we need to read the current file version. */
 841         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 842                 struct inode    *inode;
 843                 __u64            data_version = 0;
 844
 845                 /* Get inode for this fid */
 846                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 847                 if (IS_ERR(inode)) {
 848                         hpk.hpk_flags |= HP_FLAG_RETRY;
 849                         /* hpk_errval is >= 0 */
 850                         hpk.hpk_errval = -PTR_ERR(inode);
 851                         GOTO(progress, rc = PTR_ERR(inode));
 852                 }
 853
 854                 /* Read current file data version */
 855                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 856                 iput(inode);
 857                 if (rc != 0) {
 858                         CDEBUG(D_HSM, "Could not read file data version of "
 859                                       DFID" (rc = %d). Archive request ("
 860                                       "%#llx) could not be done.\n",
 861                                       PFID(&copy->hc_hai.hai_fid), rc,
 862                                       copy->hc_hai.hai_cookie);
 863                         hpk.hpk_flags |= HP_FLAG_RETRY;
 864                         /* hpk_errval must be >= 0 */
 865                         hpk.hpk_errval = -rc;
 866                         GOTO(progress, rc);
 867                 }
 868
 869                 /* Store in the hsm_copy for later copytool use.
 870                  * Always modified even if no lsm. */
 871                 copy->hc_data_version = data_version;
 872         }
 873
 874 progress:
 875         /* On error, the request should be considered as completed */
 876         if (hpk.hpk_errval > 0)
 877                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 878
 879         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 880                             &hpk, NULL);
 881
 882         /* Return first error */
 883         RETURN(rc != 0 ? rc : rc2);
 884 }
 885
 886 /**
 887  * Generic handler to do any post-copy work.
 888  *
 889  * It will send the last hsm_progress update to coordinator to inform it
 890  * that copy is finished and whether it was successful or not.
 891  *
 892  * Moreover,
 893  * - for ARCHIVE request, it will sample the file data version and compare it
 894  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 895  *   will be considered as failed.
 896  * - for RESTORE request, it will sample the file data version and send it to
 897  *   coordinator which is useful if the file was imported as 'released'.
 898  *
 899  * \return 0 on success.
 900  */
 901 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 902 {
 903         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 904         struct hsm_progress_kernel       hpk;
 905         int                              rc = 0;
 906         int                              rc2;
 907         ENTRY;
 908
 909         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 910         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 911          * initialized if copy_end was called with copy == NULL.
 912          */
 913
 914         /* Forge a hsm_progress based on data from copy. */
 915         hpk.hpk_fid = copy->hc_hai.hai_fid;
 916         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 917         hpk.hpk_extent = copy->hc_hai.hai_extent;
 918         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 919         hpk.hpk_errval = copy->hc_errval;
 920         hpk.hpk_data_version = 0;
 921
 922         /* For archive request, we need to check the file data was not changed.
 923          *
 924          * For restore request, we need to send the file data version, this is
 925          * useful when the file was created using hsm_import.
 926          */
 927         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 928              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 929             (copy->hc_errval == 0)) {
 930                 struct inode    *inode;
 931                 __u64            data_version = 0;
 932
 933                 /* Get lsm for this fid */
 934                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 935                 if (IS_ERR(inode)) {
 936                         hpk.hpk_flags |= HP_FLAG_RETRY;
 937                         /* hpk_errval must be >= 0 */
 938                         hpk.hpk_errval = -PTR_ERR(inode);
 939                         GOTO(progress, rc = PTR_ERR(inode));
 940                 }
 941
 942                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 943                 iput(inode);
 944                 if (rc) {
 945                         CDEBUG(D_HSM, "Could not read file data version. "
 946                                       "Request could not be confirmed.\n");
 947                         if (hpk.hpk_errval == 0)
 948                                 hpk.hpk_errval = -rc;
 949                         GOTO(progress, rc);
 950                 }
 951
 952                 /* Store in the hsm_copy for later copytool use.
 953                  * Always modified even if no lsm. */
 954                 hpk.hpk_data_version = data_version;
 955
 956                 /* File could have been stripped during archiving, so we need
 957                  * to check anyway. */
 958                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 959                     (copy->hc_data_version != data_version)) {
 960                         CDEBUG(D_HSM, "File data version mismatched. "
 961                               "File content was changed during archiving. "
 962                                DFID", start:%#llx current:%#llx\n",
 963                                PFID(&copy->hc_hai.hai_fid),
 964                                copy->hc_data_version, data_version);
 965                         /* File was changed, send error to cdt. Do not ask for
 966                          * retry because if a file is modified frequently,
 967                          * the cdt will loop on retried archive requests.
 968                          * The policy engine will ask for a new archive later
 969                          * when the file will not be modified for some tunable
 970                          * time */
 971                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 972                         rc = -EBUSY;
 973                         /* hpk_errval must be >= 0 */
 974                         hpk.hpk_errval = -rc;
 975                         GOTO(progress, rc);
 976                 }
 977
 978         }
 979
 980 progress:
 981         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 982                             &hpk, NULL);
 983
 984         /* Return first error */
 985         RETURN(rc != 0 ? rc : rc2);
 986 }
 987
 988
 989 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 990                           const void __user *data, size_t size)
 991 {
 992         void *copy;
 993         int rc;
 994
 995         OBD_ALLOC(copy, size);
 996         if (copy == NULL)
 997                 return -ENOMEM;
 998
 999         if (copy_from_user(copy, data, size)) {
1000                 rc = -EFAULT;
1001                 goto out;
1002         }
1003
1004         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1005 out:
1006         OBD_FREE(copy, size);
1007
1008         return rc;
1009 }
1010
1011 static int check_owner(int type, int id)
1012 {
1013         switch (type) {
1014         case USRQUOTA:
1015                 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1016                         return -EPERM;
1017                 break;
1018         case GRPQUOTA:
1019                 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1020                         return -EPERM;
1021                 break;
1022         case PRJQUOTA:
1023                 break;
1024         }
1025         return 0;
1026 }
1027
1028 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1029 {
1030         int cmd = qctl->qc_cmd;
1031         int type = qctl->qc_type;
1032         int id = qctl->qc_id;
1033         int valid = qctl->qc_valid;
1034         int rc = 0;
1035         ENTRY;
1036
1037         switch (cmd) {
1038         case Q_SETQUOTA:
1039         case Q_SETINFO:
1040         case LUSTRE_Q_SETDEFAULT:
1041                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1042                         RETURN(-EPERM);
1043                 break;
1044         case Q_GETQUOTA:
1045         case LUSTRE_Q_GETDEFAULT:
1046                 if (check_owner(type, id) &&
1047                     (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1048                         RETURN(-EPERM);
1049                 break;
1050         case Q_GETINFO:
1051                 break;
1052         default:
1053                 CERROR("unsupported quotactl op: %#x\n", cmd);
1054                 RETURN(-ENOTSUPP);
1055         }
1056
1057         if (valid != QC_GENERAL) {
1058                 if (cmd == Q_GETINFO)
1059                         qctl->qc_cmd = Q_GETOINFO;
1060                 else if (cmd == Q_GETQUOTA)
1061                         qctl->qc_cmd = Q_GETOQUOTA;
1062                 else
1063                         RETURN(-EINVAL);
1064
1065                 switch (valid) {
1066                 case QC_MDTIDX:
1067                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1068                                            sizeof(*qctl), qctl, NULL);
1069                         break;
1070                 case QC_OSTIDX:
1071                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1072                                            sizeof(*qctl), qctl, NULL);
1073                         break;
1074                 case QC_UUID:
1075                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1076                                            sizeof(*qctl), qctl, NULL);
1077                         if (rc == -EAGAIN)
1078                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1079                                                    sbi->ll_dt_exp,
1080                                                    sizeof(*qctl), qctl, NULL);
1081                         break;
1082                 default:
1083                         rc = -EINVAL;
1084                         break;
1085                 }
1086
1087                 if (rc)
1088                         RETURN(rc);
1089
1090                 qctl->qc_cmd = cmd;
1091         } else {
1092                 struct obd_quotactl *oqctl;
1093
1094                 OBD_ALLOC_PTR(oqctl);
1095                 if (oqctl == NULL)
1096                         RETURN(-ENOMEM);
1097
1098                 QCTL_COPY(oqctl, qctl);
1099                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1100                 if (rc) {
1101                         OBD_FREE_PTR(oqctl);
1102                         RETURN(rc);
1103                 }
1104                 /* If QIF_SPACE is not set, client should collect the
1105                  * space usage from OSSs by itself */
1106                 if (cmd == Q_GETQUOTA &&
1107                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1108                     !oqctl->qc_dqblk.dqb_curspace) {
1109                         struct obd_quotactl *oqctl_tmp;
1110
1111                         OBD_ALLOC_PTR(oqctl_tmp);
1112                         if (oqctl_tmp == NULL)
1113                                 GOTO(out, rc = -ENOMEM);
1114
1115                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1116                         oqctl_tmp->qc_id = oqctl->qc_id;
1117                         oqctl_tmp->qc_type = oqctl->qc_type;
1118
1119                         /* collect space usage from OSTs */
1120                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1121                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1122                         if (!rc || rc == -EREMOTEIO) {
1123                                 oqctl->qc_dqblk.dqb_curspace =
1124                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1125                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1126                         }
1127
1128                         /* collect space & inode usage from MDTs */
1129                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1130                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1131                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1132                         if (!rc || rc == -EREMOTEIO) {
1133                                 oqctl->qc_dqblk.dqb_curspace +=
1134                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1135                                 oqctl->qc_dqblk.dqb_curinodes =
1136                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1137                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1138                         } else {
1139                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1140                         }
1141
1142                         OBD_FREE_PTR(oqctl_tmp);
1143                 }
1144 out:
1145                 QCTL_COPY(qctl, oqctl);
1146                 OBD_FREE_PTR(oqctl);
1147         }
1148
1149         RETURN(rc);
1150 }
1151
1152 /* This function tries to get a single name component,
1153  * to send to the server. No actual path traversal involved,
1154  * so we limit to NAME_MAX */
1155 static char *ll_getname(const char __user *filename)
1156 {
1157         int ret = 0, len;
1158         char *tmp;
1159
1160         OBD_ALLOC(tmp, NAME_MAX + 1);
1161
1162         if (!tmp)
1163                 return ERR_PTR(-ENOMEM);
1164
1165         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1166         if (len < 0)
1167                 ret = -ENOENT;
1168         else if (len > NAME_MAX)
1169                 ret = -ENAMETOOLONG;
1170
1171         if (ret) {
1172                 OBD_FREE(tmp, NAME_MAX + 1);
1173                 tmp =  ERR_PTR(ret);
1174         }
1175         return tmp;
1176 }
1177
1178 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1179
1180 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1181 {
1182         struct dentry *dentry = file_dentry(file);
1183         struct inode *inode = file_inode(file);
1184         struct ll_sb_info *sbi = ll_i2sbi(inode);
1185         struct obd_ioctl_data *data;
1186         int rc = 0;
1187         ENTRY;
1188
1189         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1190                PFID(ll_inode2fid(inode)), inode, cmd);
1191
1192         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1193         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1194                 return -ENOTTY;
1195
1196         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1197         switch (cmd) {
1198         case FS_IOC_GETFLAGS:
1199         case FS_IOC_SETFLAGS:
1200                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1201         case FSFILT_IOC_GETVERSION:
1202         case FS_IOC_GETVERSION:
1203                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1204         /* We need to special case any other ioctls we want to handle,
1205          * to send them to the MDS/OST as appropriate and to properly
1206          * network encode the arg field. */
1207         case FS_IOC_SETVERSION:
1208                 RETURN(-ENOTSUPP);
1209
1210         case LL_IOC_GET_MDTIDX: {
1211                 int mdtidx;
1212
1213                 mdtidx = ll_get_mdt_idx(inode);
1214                 if (mdtidx < 0)
1215                         RETURN(mdtidx);
1216
1217                 if (put_user((int)mdtidx, (int __user *)arg))
1218                         RETURN(-EFAULT);
1219
1220                 return 0;
1221         }
1222         case IOC_MDC_LOOKUP: {
1223                                      int namelen, len = 0;
1224                 char *buf = NULL;
1225                 char *filename;
1226
1227                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1228                 if (rc != 0)
1229                         RETURN(rc);
1230                 data = (void *)buf;
1231
1232                 filename = data->ioc_inlbuf1;
1233                 namelen = strlen(filename);
1234                 if (namelen < 1) {
1235                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1236                         GOTO(out_free, rc = -EINVAL);
1237                 }
1238
1239                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1240                 if (rc < 0) {
1241                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1242                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1243                                filename, rc);
1244                         GOTO(out_free, rc);
1245                 }
1246 out_free:
1247                 OBD_FREE_LARGE(buf, len);
1248                 return rc;
1249         }
1250         case LL_IOC_LMV_SETSTRIPE: {
1251                 struct lmv_user_md  *lum;
1252                 char            *buf = NULL;
1253                 char            *filename;
1254                 int              namelen = 0;
1255                 int              lumlen = 0;
1256                 umode_t          mode;
1257                 int              len;
1258                 int              rc;
1259
1260                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1261                 if (rc)
1262                         RETURN(rc);
1263
1264                 data = (void *)buf;
1265                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1266                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1267                         GOTO(lmv_out_free, rc = -EINVAL);
1268
1269                 filename = data->ioc_inlbuf1;
1270                 namelen = data->ioc_inllen1;
1271
1272                 if (namelen < 1) {
1273                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1274                         GOTO(lmv_out_free, rc = -EINVAL);
1275                 }
1276                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1277                 lumlen = data->ioc_inllen2;
1278
1279                 if ((lum->lum_magic != LMV_USER_MAGIC &&
1280                      lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) ||
1281                     lumlen < sizeof(*lum)) {
1282                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1283                                filename, lum->lum_magic, lumlen, -EFAULT);
1284                         GOTO(lmv_out_free, rc = -EINVAL);
1285                 }
1286
1287 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1288                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1289 #else
1290                 mode = data->ioc_type;
1291 #endif
1292                 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1293 lmv_out_free:
1294                 OBD_FREE_LARGE(buf, len);
1295                 RETURN(rc);
1296
1297         }
1298         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1299                 struct lmv_user_md        lum;
1300                 struct lmv_user_md __user *ulump =
1301                                         (struct lmv_user_md __user *)arg;
1302                 int                       rc;
1303
1304                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1305                         RETURN(-EFAULT);
1306
1307                 if (lum.lum_magic != LMV_USER_MAGIC)
1308                         RETURN(-EINVAL);
1309
1310                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1311
1312                 RETURN(rc);
1313         }
1314         case LL_IOC_LOV_SETSTRIPE_NEW:
1315         case LL_IOC_LOV_SETSTRIPE: {
1316                 struct lov_user_md_v3 *lumv3 = NULL;
1317                 struct lov_user_md_v1 lumv1;
1318                 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1319                 struct lov_user_md_v1 __user *lumv1p =
1320                         (struct lov_user_md_v1 __user *)arg;
1321                 struct lov_user_md_v3 __user *lumv3p =
1322                         (struct lov_user_md_v3 __user *)arg;
1323                 int lum_size;
1324
1325                 int set_default = 0;
1326
1327                 CLASSERT(sizeof(struct lov_user_md_v3) >
1328                          sizeof(struct lov_comp_md_v1));
1329                 CLASSERT(sizeof(*lumv3) == sizeof(*lumv3p));
1330                 /* first try with v1 which is smaller than v3 */
1331                 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1332                         RETURN(-EFAULT);
1333
1334                 if (inode->i_sb->s_root == file_dentry(file))
1335                         set_default = 1;
1336
1337                 switch (lumv1.lmm_magic) {
1338                 case LOV_USER_MAGIC_V3:
1339                 case LOV_USER_MAGIC_SPECIFIC:
1340                         lum_size = ll_lov_user_md_size(&lumv1);
1341                         if (lum_size < 0)
1342                                 RETURN(lum_size);
1343                         OBD_ALLOC(lumv3, lum_size);
1344                         if (!lumv3)
1345                                 RETURN(-ENOMEM);
1346                         if (copy_from_user(lumv3, lumv3p, lum_size))
1347                                 GOTO(out, rc = -EFAULT);
1348                         lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1349                         break;
1350                 case LOV_USER_MAGIC_V1:
1351                         break;
1352                 default:
1353                         GOTO(out, rc = -ENOTSUPP);
1354                 }
1355
1356                 /* in v1 and v3 cases lumv1 points to data */
1357                 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1358 out:
1359                 if (lumv3)
1360                         OBD_FREE(lumv3, lum_size);
1361                 RETURN(rc);
1362         }
1363         case LL_IOC_LMV_GETSTRIPE: {
1364                 struct lmv_user_md __user *ulmv =
1365                                         (struct lmv_user_md __user *)arg;
1366                 struct lmv_user_md      lum;
1367                 struct ptlrpc_request   *request = NULL;
1368                 union lmv_mds_md        *lmm = NULL;
1369                 int                     lmmsize;
1370                 u64                     valid = 0;
1371                 struct lmv_user_md      *tmp = NULL;
1372                 int                     mdt_index;
1373                 int                     lum_size;
1374                 int                     stripe_count;
1375                 int                     max_stripe_count;
1376                 int                     i;
1377                 int                     rc;
1378
1379                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1380                         RETURN(-EFAULT);
1381
1382                 max_stripe_count = lum.lum_stripe_count;
1383                 /* lum_magic will indicate which stripe the ioctl will like
1384                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1385                  * is for default LMV stripe */
1386                 if (lum.lum_magic == LMV_MAGIC_V1)
1387                         valid |= OBD_MD_MEA;
1388                 else if (lum.lum_magic == LMV_USER_MAGIC)
1389                         valid |= OBD_MD_DEFAULT_MEA;
1390                 else
1391                         RETURN(-EINVAL);
1392
1393                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1394                                       valid);
1395                 if (rc != 0)
1396                         GOTO(finish_req, rc);
1397
1398                 /* Get default LMV EA */
1399                 if (lum.lum_magic == LMV_USER_MAGIC) {
1400                         if (lmmsize > sizeof(*ulmv))
1401                                 GOTO(finish_req, rc = -EINVAL);
1402
1403                         if (copy_to_user(ulmv, lmm, lmmsize))
1404                                 GOTO(finish_req, rc = -EFAULT);
1405
1406                         GOTO(finish_req, rc);
1407                 }
1408
1409                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1410                 if (max_stripe_count < stripe_count) {
1411                         lum.lum_stripe_count = stripe_count;
1412                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1413                                 GOTO(finish_req, rc = -EFAULT);
1414                         GOTO(finish_req, rc = -E2BIG);
1415                 }
1416
1417                 lum_size = lmv_user_md_size(stripe_count,
1418                                             LMV_USER_MAGIC_SPECIFIC);
1419                 OBD_ALLOC(tmp, lum_size);
1420                 if (tmp == NULL)
1421                         GOTO(finish_req, rc = -ENOMEM);
1422
1423                 mdt_index = ll_get_mdt_idx(inode);
1424                 if (mdt_index < 0)
1425                         GOTO(out_tmp, rc = -ENOMEM);
1426
1427                 tmp->lum_magic = LMV_MAGIC_V1;
1428                 tmp->lum_stripe_count = 0;
1429                 tmp->lum_stripe_offset = mdt_index;
1430                 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1431                 for (i = 0; i < stripe_count; i++) {
1432                         struct lu_fid   fid;
1433
1434                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1435                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1436                         if (mdt_index < 0)
1437                                 GOTO(out_tmp, rc = mdt_index);
1438
1439                         tmp->lum_objects[i].lum_mds = mdt_index;
1440                         tmp->lum_objects[i].lum_fid = fid;
1441                         tmp->lum_stripe_count++;
1442                 }
1443
1444                 if (copy_to_user(ulmv, tmp, lum_size))
1445                         GOTO(out_tmp, rc = -EFAULT);
1446 out_tmp:
1447                 OBD_FREE(tmp, lum_size);
1448 finish_req:
1449                 ptlrpc_req_finished(request);
1450                 return rc;
1451         }
1452
1453         case LL_IOC_REMOVE_ENTRY: {
1454                 char            *filename = NULL;
1455                 int              namelen = 0;
1456                 int              rc;
1457
1458                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1459                  * unsupported server, which might crash the server(LU-2730),
1460                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1461                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1462                  * server will support REINT_RMENTRY XXX*/
1463                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1464                         RETURN(-ENOTSUPP);
1465
1466                 filename = ll_getname((const char __user *)arg);
1467                 if (IS_ERR(filename))
1468                         RETURN(PTR_ERR(filename));
1469
1470                 namelen = strlen(filename);
1471                 if (namelen < 1)
1472                         GOTO(out_rmdir, rc = -EINVAL);
1473
1474                 rc = ll_rmdir_entry(inode, filename, namelen);
1475 out_rmdir:
1476                 if (filename)
1477                         ll_putname(filename);
1478                 RETURN(rc);
1479         }
1480         case LL_IOC_LOV_SWAP_LAYOUTS:
1481                 RETURN(-EPERM);
1482         case IOC_OBD_STATFS:
1483                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1484         case LL_IOC_LOV_GETSTRIPE:
1485         case LL_IOC_LOV_GETSTRIPE_NEW:
1486         case LL_IOC_MDC_GETINFO:
1487         case IOC_MDC_GETFILEINFO:
1488         case IOC_MDC_GETFILESTRIPE: {
1489                 struct ptlrpc_request *request = NULL;
1490                 struct lov_user_md __user *lump;
1491                 struct lov_mds_md *lmm = NULL;
1492                 struct mdt_body *body;
1493                 char *filename = NULL;
1494                 int lmmsize;
1495
1496                 if (cmd == IOC_MDC_GETFILEINFO ||
1497                     cmd == IOC_MDC_GETFILESTRIPE) {
1498                         filename = ll_getname((const char __user *)arg);
1499                         if (IS_ERR(filename))
1500                                 RETURN(PTR_ERR(filename));
1501
1502                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1503                                                       &lmmsize, &request);
1504                 } else {
1505                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1506                                               &request, 0);
1507                 }
1508
1509                 if (request) {
1510                         body = req_capsule_server_get(&request->rq_pill,
1511                                                       &RMF_MDT_BODY);
1512                         LASSERT(body != NULL);
1513                 } else {
1514                         GOTO(out_req, rc);
1515                 }
1516
1517                 if (rc < 0) {
1518                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1519                                                cmd == LL_IOC_MDC_GETINFO))
1520                                 GOTO(skip_lmm, rc = 0);
1521                         else
1522                                 GOTO(out_req, rc);
1523                 }
1524
1525                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1526                     cmd == LL_IOC_LOV_GETSTRIPE ||
1527                     cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1528                         lump = (struct lov_user_md __user *)arg;
1529                 } else {
1530                         struct lov_user_mds_data __user *lmdp;
1531                         lmdp = (struct lov_user_mds_data __user *)arg;
1532                         lump = &lmdp->lmd_lmm;
1533                 }
1534                 if (copy_to_user(lump, lmm, lmmsize)) {
1535                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1536                                 GOTO(out_req, rc = -EFAULT);
1537                         rc = -EOVERFLOW;
1538                 }
1539         skip_lmm:
1540                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1541                         struct lov_user_mds_data __user *lmdp;
1542                         lstat_t st = { 0 };
1543
1544                         st.st_dev       = inode->i_sb->s_dev;
1545                         st.st_mode      = body->mbo_mode;
1546                         st.st_nlink     = body->mbo_nlink;
1547                         st.st_uid       = body->mbo_uid;
1548                         st.st_gid       = body->mbo_gid;
1549                         st.st_rdev      = body->mbo_rdev;
1550                         st.st_size      = body->mbo_size;
1551                         st.st_blksize   = PAGE_SIZE;
1552                         st.st_blocks    = body->mbo_blocks;
1553                         st.st_atime     = body->mbo_atime;
1554                         st.st_mtime     = body->mbo_mtime;
1555                         st.st_ctime     = body->mbo_ctime;
1556                         st.st_ino       = cl_fid_build_ino(&body->mbo_fid1,
1557                                                 sbi->ll_flags &
1558                                                 LL_SBI_32BIT_API);
1559
1560                         lmdp = (struct lov_user_mds_data __user *)arg;
1561                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1562                                 GOTO(out_req, rc = -EFAULT);
1563                 }
1564
1565                 EXIT;
1566         out_req:
1567                 ptlrpc_req_finished(request);
1568                 if (filename)
1569                         ll_putname(filename);
1570                 return rc;
1571         }
1572         case OBD_IOC_QUOTACTL: {
1573                 struct if_quotactl *qctl;
1574
1575                 OBD_ALLOC_PTR(qctl);
1576                 if (!qctl)
1577                         RETURN(-ENOMEM);
1578
1579                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1580                         GOTO(out_quotactl, rc = -EFAULT);
1581
1582                 rc = quotactl_ioctl(sbi, qctl);
1583
1584                 if (rc == 0 &&
1585                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1586                         rc = -EFAULT;
1587
1588         out_quotactl:
1589                 OBD_FREE_PTR(qctl);
1590                 RETURN(rc);
1591         }
1592         case OBD_IOC_GETDTNAME:
1593         case OBD_IOC_GETMDNAME:
1594                 RETURN(ll_get_obd_name(inode, cmd, arg));
1595         case LL_IOC_FLUSHCTX:
1596                 RETURN(ll_flush_ctx(inode));
1597         case LL_IOC_GETOBDCOUNT: {
1598                 u32 count, vallen;
1599                 struct obd_export *exp;
1600
1601                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1602                         RETURN(-EFAULT);
1603
1604                 /* get ost count when count is zero, get mdt count otherwise */
1605                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1606                 vallen = sizeof(count);
1607                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1608                                   KEY_TGT_COUNT, &vallen, &count);
1609                 if (rc) {
1610                         CERROR("get target count failed: %d\n", rc);
1611                         RETURN(rc);
1612                 }
1613
1614                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1615                         RETURN(-EFAULT);
1616
1617                 RETURN(0);
1618         }
1619         case LL_IOC_PATH2FID:
1620                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1621                                      sizeof(struct lu_fid)))
1622                         RETURN(-EFAULT);
1623                 RETURN(0);
1624         case LL_IOC_GET_CONNECT_FLAGS: {
1625                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1626                                      (void __user *)arg));
1627         }
1628         case OBD_IOC_FID2PATH:
1629                 RETURN(ll_fid2path(inode, (void __user *)arg));
1630         case LL_IOC_GETPARENT:
1631                 RETURN(ll_getparent(file, (void __user *)arg));
1632         case LL_IOC_FID2MDTIDX: {
1633                 struct obd_export *exp = ll_i2mdexp(inode);
1634                 struct lu_fid     fid;
1635                 __u32             index;
1636
1637                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1638                                    sizeof(fid)))
1639                         RETURN(-EFAULT);
1640
1641                 /* Call mdc_iocontrol */
1642                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1643                                    (__u32 __user *)&index);
1644                 if (rc != 0)
1645                         RETURN(rc);
1646
1647                 RETURN(index);
1648         }
1649         case LL_IOC_HSM_REQUEST: {
1650                 struct hsm_user_request *hur;
1651                 ssize_t                  totalsize;
1652
1653                 OBD_ALLOC_PTR(hur);
1654                 if (hur == NULL)
1655                         RETURN(-ENOMEM);
1656
1657                 /* We don't know the true size yet; copy the fixed-size part */
1658                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1659                         OBD_FREE_PTR(hur);
1660                         RETURN(-EFAULT);
1661                 }
1662
1663                 /* Compute the whole struct size */
1664                 totalsize = hur_len(hur);
1665                 OBD_FREE_PTR(hur);
1666                 if (totalsize < 0)
1667                         RETURN(-E2BIG);
1668
1669                 /* Final size will be more than double totalsize */
1670                 if (totalsize >= MDS_MAXREQSIZE / 3)
1671                         RETURN(-E2BIG);
1672
1673                 OBD_ALLOC_LARGE(hur, totalsize);
1674                 if (hur == NULL)
1675                         RETURN(-ENOMEM);
1676
1677                 /* Copy the whole struct */
1678                 if (copy_from_user(hur, (void __user *)arg, totalsize))
1679                         GOTO(out_hur, rc = -EFAULT);
1680
1681                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1682                         const struct lu_fid *fid;
1683                         struct inode *f;
1684                         int i;
1685
1686                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1687                                 fid = &hur->hur_user_item[i].hui_fid;
1688                                 f = search_inode_for_lustre(inode->i_sb, fid);
1689                                 if (IS_ERR(f)) {
1690                                         rc = PTR_ERR(f);
1691                                         break;
1692                                 }
1693
1694                                 rc = ll_hsm_release(f);
1695                                 iput(f);
1696                                 if (rc != 0)
1697                                         break;
1698                         }
1699                 } else {
1700                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1701                                            hur, NULL);
1702                 }
1703
1704 out_hur:
1705                 OBD_FREE_LARGE(hur, totalsize);
1706
1707                 RETURN(rc);
1708         }
1709         case LL_IOC_HSM_PROGRESS: {
1710                 struct hsm_progress_kernel      hpk;
1711                 struct hsm_progress             hp;
1712
1713                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1714                         RETURN(-EFAULT);
1715
1716                 hpk.hpk_fid = hp.hp_fid;
1717                 hpk.hpk_cookie = hp.hp_cookie;
1718                 hpk.hpk_extent = hp.hp_extent;
1719                 hpk.hpk_flags = hp.hp_flags;
1720                 hpk.hpk_errval = hp.hp_errval;
1721                 hpk.hpk_data_version = 0;
1722
1723                 /* File may not exist in Lustre; all progress
1724                  * reported to Lustre root */
1725                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1726                                    NULL);
1727                 RETURN(rc);
1728         }
1729         case LL_IOC_HSM_CT_START:
1730                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1731                         RETURN(-EPERM);
1732
1733                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1734                                     sizeof(struct lustre_kernelcomm));
1735                 RETURN(rc);
1736
1737         case LL_IOC_HSM_COPY_START: {
1738                 struct hsm_copy *copy;
1739                 int              rc;
1740
1741                 OBD_ALLOC_PTR(copy);
1742                 if (copy == NULL)
1743                         RETURN(-ENOMEM);
1744                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1745                         OBD_FREE_PTR(copy);
1746                         RETURN(-EFAULT);
1747                 }
1748
1749                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1750                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1751                         rc = -EFAULT;
1752
1753                 OBD_FREE_PTR(copy);
1754                 RETURN(rc);
1755         }
1756         case LL_IOC_HSM_COPY_END: {
1757                 struct hsm_copy *copy;
1758                 int              rc;
1759
1760                 OBD_ALLOC_PTR(copy);
1761                 if (copy == NULL)
1762                         RETURN(-ENOMEM);
1763                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1764                         OBD_FREE_PTR(copy);
1765                         RETURN(-EFAULT);
1766                 }
1767
1768                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1769                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1770                         rc = -EFAULT;
1771
1772                 OBD_FREE_PTR(copy);
1773                 RETURN(rc);
1774         }
1775         case LL_IOC_MIGRATE: {
1776                 struct lmv_user_md *lum;
1777                 char *buf = NULL;
1778                 int len;
1779                 char *filename;
1780                 int namelen = 0;
1781                 int rc;
1782
1783                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1784                 if (rc)
1785                         RETURN(rc);
1786
1787                 data = (struct obd_ioctl_data *)buf;
1788                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1789                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1790                         GOTO(migrate_free, rc = -EINVAL);
1791
1792                 filename = data->ioc_inlbuf1;
1793                 namelen = data->ioc_inllen1;
1794
1795                 if (namelen < 1 || namelen != strlen(filename) + 1) {
1796                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1797                         GOTO(migrate_free, rc = -EINVAL);
1798                 }
1799
1800                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1801                 if (lum->lum_magic != LMV_USER_MAGIC &&
1802                     lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
1803                         rc = -EINVAL;
1804                         CERROR("%s: wrong lum magic %x: rc = %d\n",
1805                                filename, lum->lum_magic, rc);
1806                         GOTO(migrate_free, rc);
1807                 }
1808
1809                 rc = ll_migrate(inode, file, lum, filename);
1810 migrate_free:
1811                 OBD_FREE_LARGE(buf, len);
1812
1813                 RETURN(rc);
1814         }
1815         case LL_IOC_FSGETXATTR:
1816                 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
1817         case LL_IOC_FSSETXATTR:
1818                 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
1819         default:
1820                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1821                                      (void __user *)arg));
1822         }
1823 }
1824
1825 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1826 {
1827         struct inode *inode = file->f_mapping->host;
1828         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1829         struct ll_sb_info *sbi = ll_i2sbi(inode);
1830         int api32 = ll_need_32bit_api(sbi);
1831         loff_t ret = -EINVAL;
1832         ENTRY;
1833
1834         inode_lock(inode);
1835         switch (origin) {
1836                 case SEEK_SET:
1837                         break;
1838                 case SEEK_CUR:
1839                         offset += file->f_pos;
1840                         break;
1841                 case SEEK_END:
1842                         if (offset > 0)
1843                                 GOTO(out, ret);
1844                         if (api32)
1845                                 offset += LL_DIR_END_OFF_32BIT;
1846                         else
1847                                 offset += LL_DIR_END_OFF;
1848                         break;
1849                 default:
1850                         GOTO(out, ret);
1851         }
1852
1853         if (offset >= 0 &&
1854             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1855              (!api32 && offset <= LL_DIR_END_OFF))) {
1856                 if (offset != file->f_pos) {
1857                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1858                             (!api32 && offset == LL_DIR_END_OFF))
1859                                 fd->lfd_pos = MDS_DIR_END_OFF;
1860                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1861                                 fd->lfd_pos = offset << 32;
1862                         else
1863                                 fd->lfd_pos = offset;
1864                         file->f_pos = offset;
1865                         file->f_version = 0;
1866                 }
1867                 ret = offset;
1868         }
1869         GOTO(out, ret);
1870
1871 out:
1872         inode_unlock(inode);
1873         return ret;
1874 }
1875
1876 static int ll_dir_open(struct inode *inode, struct file *file)
1877 {
1878         ENTRY;
1879         RETURN(ll_file_open(inode, file));
1880 }
1881
1882 static int ll_dir_release(struct inode *inode, struct file *file)
1883 {
1884         ENTRY;
1885         RETURN(ll_file_release(inode, file));
1886 }
1887
1888 const struct file_operations ll_dir_operations = {
1889         .llseek         = ll_dir_seek,
1890         .open           = ll_dir_open,
1891         .release        = ll_dir_release,
1892         .read           = generic_read_dir,
1893 #ifdef HAVE_DIR_CONTEXT
1894         .iterate        = ll_iterate,
1895 #else
1896         .readdir        = ll_readdir,
1897 #endif
1898         .unlocked_ioctl = ll_dir_ioctl,
1899         .fsync          = ll_fsync,
1900 };