lustre/llite/dir.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/dir.c
  33  *
  34  * Directory code for lustre client.
  35  */
  36
  37 #include <linux/fs.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/mm.h>
  40 #include <linux/version.h>
  41 #include <linux/security.h>
  42 #include <linux/user_namespace.h>
  43 #ifdef HAVE_UIDGID_HEADER
  44 # include <linux/uidgid.h>
  45 #endif
  46 #include <linux/uaccess.h>
  47 #include <linux/buffer_head.h>   // for wait_on_buffer
  48 #include <linux/pagevec.h>
  49
  50 #define DEBUG_SUBSYSTEM S_LLITE
  51
  52 #include <obd_support.h>
  53 #include <obd_class.h>
  54 #include <uapi/linux/lustre/lustre_ioctl.h>
  55 #include <lustre_lib.h>
  56 #include <lustre_dlm.h>
  57 #include <lustre_fid.h>
  58 #include <lustre_kernelcomm.h>
  59 #include <lustre_swab.h>
  60
  61 #include "llite_internal.h"
  62
  63 /*
  64  * (new) readdir implementation overview.
  65  *
  66  * Original lustre readdir implementation cached exact copy of raw directory
  67  * pages on the client. These pages were indexed in client page cache by
  68  * logical offset in the directory file. This design, while very simple and
  69  * intuitive had some inherent problems:
  70  *
  71  *     . it implies that byte offset to the directory entry serves as a
  72  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  73  *     ext3/htree directory entries may move due to splits, and more
  74  *     importantly,
  75  *
  76  *     . it is incompatible with the design of split directories for cmd3,
  77  *     that assumes that names are distributed across nodes based on their
  78  *     hash, and so readdir should be done in hash order.
  79  *
  80  * New readdir implementation does readdir in hash order, and uses hash of a
  81  * file name as a telldir/seekdir cookie. This led to number of complications:
  82  *
  83  *     . hash is not unique, so it cannot be used to index cached directory
  84  *     pages on the client (note, that it requires a whole pageful of hash
  85  *     collided entries to cause two pages to have identical hashes);
  86  *
  87  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  88  *     entry cookie. ext3/htree has the same problem and lustre implementation
  89  *     mimics their solution: seekdir(hash) positions directory at the first
  90  *     entry with the given hash.
  91  *
  92  * Client side.
  93  *
  94  * 0. caching
  95  *
  96  * Client caches directory pages using hash of the first entry as an index. As
  97  * noted above hash is not unique, so this solution doesn't work as is:
  98  * special processing is needed for "page hash chains" (i.e., sequences of
  99  * pages filled with entries all having the same hash value).
 100  *
 101  * First, such chains have to be detected. To this end, server returns to the
 102  * client the hash of the first entry on the page next to one returned. When
 103  * client detects that this hash is the same as hash of the first entry on the
 104  * returned page, page hash collision has to be handled. Pages in the
 105  * hash chain, except first one, are termed "overflow pages".
 106  *
 107  * Solution to index uniqueness problem is to not cache overflow
 108  * pages. Instead, when page hash collision is detected, all overflow pages
 109  * from emerging chain are immediately requested from the server and placed in
 110  * a special data structure (struct ll_dir_chain). This data structure is used
 111  * by ll_readdir() to process entries from overflow pages. When readdir
 112  * invocation finishes, overflow pages are discarded. If page hash collision
 113  * chain weren't completely processed, next call to readdir will again detect
 114  * page hash collision, again read overflow pages in, process next portion of
 115  * entries and again discard the pages. This is not as wasteful as it looks,
 116  * because, given reasonable hash, page hash collisions are extremely rare.
 117  *
 118  * 1. directory positioning
 119  *
 120  * When seekdir(hash) is called, original
 121  *
 122  *
 123  *
 124  *
 125  *
 126  *
 127  *
 128  *
 129  * Server.
 130  *
 131  * identification of and access to overflow pages
 132  *
 133  * page format
 134  *
 135  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 136  * a header lu_dirpage which describes the start/end hash, and whether this
 137  * page is empty (contains no dir entry) or hash collide with next page.
 138  * After client receives reply, several pages will be integrated into dir page
 139  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
 140  * lu_dirpage for this integrated page will be adjusted. See
 141  * mdc_adjust_dirpages().
 142  *
 143  */
 144 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
 145                              __u64 offset, struct ll_dir_chain *chain)
 146 {
 147         struct md_callback      cb_op;
 148         struct page             *page;
 149         int                     rc;
 150
 151         cb_op.md_blocking_ast = ll_md_blocking_ast;
 152         rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
 153         if (rc != 0)
 154                 return ERR_PTR(rc);
 155
 156         return page;
 157 }
 158
 159 void ll_release_page(struct inode *inode, struct page *page,
 160                      bool remove)
 161 {
 162         kunmap(page);
 163
 164         /* Always remove the page for striped dir, because the page is
 165          * built from temporarily in LMV layer */
 166         if (inode != NULL && S_ISDIR(inode->i_mode) &&
 167             ll_i2info(inode)->lli_lsm_md != NULL) {
 168                 __free_page(page);
 169                 return;
 170         }
 171
 172         if (remove) {
 173                 lock_page(page);
 174                 if (likely(page->mapping != NULL))
 175                         truncate_complete_page(page->mapping, page);
 176                 unlock_page(page);
 177         }
 178         put_page(page);
 179 }
 180
 181 /**
 182  * return IF_* type for given lu_dirent entry.
 183  * IF_* flag shld be converted to particular OS file type in
 184  * platform llite module.
 185  */
 186 static u16 ll_dirent_type_get(struct lu_dirent *ent)
 187 {
 188         u16 type = 0;
 189         struct luda_type *lt;
 190         int len = 0;
 191
 192         if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
 193                 const unsigned align = sizeof(struct luda_type) - 1;
 194
 195                 len = le16_to_cpu(ent->lde_namelen);
 196                 len = (len + align) & ~align;
 197                 lt = (void *)ent->lde_name + len;
 198                 type = IFTODT(le16_to_cpu(lt->lt_type));
 199         }
 200
 201         return type;
 202 }
 203
 204 #ifdef HAVE_DIR_CONTEXT
 205 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 206                 struct dir_context *ctx)
 207 {
 208 #else
 209 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 210                 void *cookie, filldir_t filldir)
 211 {
 212 #endif
 213         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 214         __u64                 pos        = *ppos;
 215         bool                  is_api32 = ll_need_32bit_api(sbi);
 216         bool                  is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 217         struct page          *page;
 218         struct ll_dir_chain   chain;
 219         bool                  done = false;
 220         int                   rc = 0;
 221         ENTRY;
 222
 223         ll_dir_chain_init(&chain);
 224
 225         page = ll_get_dir_page(inode, op_data, pos, &chain);
 226
 227         while (rc == 0 && !done) {
 228                 struct lu_dirpage *dp;
 229                 struct lu_dirent  *ent;
 230                 __u64 hash;
 231                 __u64 next;
 232
 233                 if (IS_ERR(page)) {
 234                         rc = PTR_ERR(page);
 235                         break;
 236                 }
 237
 238                 hash = MDS_DIR_END_OFF;
 239                 dp = page_address(page);
 240                 for (ent = lu_dirent_start(dp); ent != NULL && !done;
 241                      ent = lu_dirent_next(ent)) {
 242                         __u16          type;
 243                         int            namelen;
 244                         struct lu_fid  fid;
 245                         __u64          lhash;
 246                         __u64          ino;
 247
 248                         hash = le64_to_cpu(ent->lde_hash);
 249                         if (hash < pos) /* Skip until we find target hash */
 250                                 continue;
 251
 252                         namelen = le16_to_cpu(ent->lde_namelen);
 253                         if (namelen == 0) /* Skip dummy record */
 254                                 continue;
 255
 256                         if (is_api32 && is_hash64)
 257                                 lhash = hash >> 32;
 258                         else
 259                                 lhash = hash;
 260                         fid_le_to_cpu(&fid, &ent->lde_fid);
 261                         ino = cl_fid_build_ino(&fid, is_api32);
 262                         type = ll_dirent_type_get(ent);
 263                         /* For ll_nfs_get_name_filldir(), it will try to access
 264                          * 'ent' through 'lde_name', so the parameter 'name'
 265                          * for 'filldir()' must be part of the 'ent'. */
 266 #ifdef HAVE_DIR_CONTEXT
 267                         ctx->pos = lhash;
 268                         done = !dir_emit(ctx, ent->lde_name, namelen, ino,
 269                                          type);
 270 #else
 271                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 272                                        ino, type);
 273 #endif
 274                 }
 275
 276                 if (done) {
 277                         pos = hash;
 278                         ll_release_page(inode, page, false);
 279                         break;
 280                 }
 281
 282                 next = le64_to_cpu(dp->ldp_hash_end);
 283                 pos = next;
 284                 if (pos == MDS_DIR_END_OFF) {
 285                         /*
 286                          * End of directory reached.
 287                          */
 288                         done = 1;
 289                         ll_release_page(inode, page, false);
 290                 } else {
 291                         /*
 292                          * Normal case: continue to the next
 293                          * page.
 294                          */
 295                         ll_release_page(inode, page,
 296                                         le32_to_cpu(dp->ldp_flags) &
 297                                         LDF_COLLIDE);
 298                         next = pos;
 299                         page = ll_get_dir_page(inode, op_data, pos,
 300                                                &chain);
 301                 }
 302         }
 303 #ifdef HAVE_DIR_CONTEXT
 304         ctx->pos = pos;
 305 #else
 306         *ppos = pos;
 307 #endif
 308         ll_dir_chain_fini(&chain);
 309         RETURN(rc);
 310 }
 311
 312 #ifdef HAVE_DIR_CONTEXT
 313 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 314 #else
 315 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 316 #endif
 317 {
 318         struct inode            *inode  = file_inode(filp);
 319         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 320         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 321         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 322         int                     api32   = ll_need_32bit_api(sbi);
 323         struct md_op_data       *op_data;
 324         struct lu_fid           pfid = { 0 };
 325         __u64                   pos;
 326         int                     rc;
 327         ENTRY;
 328
 329         if (lfd != NULL)
 330                 pos = lfd->lfd_pos;
 331         else
 332                 pos = 0;
 333
 334         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 335                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 336                inode, (unsigned long)pos, i_size_read(inode), api32);
 337
 338         if (pos == MDS_DIR_END_OFF)
 339                 /*
 340                  * end-of-file.
 341                  */
 342                 GOTO(out, rc = 0);
 343
 344         if (unlikely(ll_i2info(inode)->lli_lsm_md != NULL)) {
 345                 /* This is only needed for striped dir to fill ..,
 346                  * see lmv_read_entry */
 347                 if (file_dentry(filp)->d_parent != NULL &&
 348                     file_dentry(filp)->d_parent->d_inode != NULL) {
 349                         __u64 ibits = MDS_INODELOCK_UPDATE;
 350                         struct inode *parent =
 351                                 file_dentry(filp)->d_parent->d_inode;
 352
 353                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 354                                 pfid = *ll_inode2fid(parent);
 355                 }
 356
 357                 /* If it can not find in cache, do lookup .. on the master
 358                  * object */
 359                 if (fid_is_zero(&pfid)) {
 360                         rc = ll_dir_get_parent_fid(inode, &pfid);
 361                         if (rc != 0)
 362                                 RETURN(rc);
 363                 }
 364         }
 365
 366         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 367                                      LUSTRE_OPC_ANY, inode);
 368         if (IS_ERR(op_data))
 369                 GOTO(out, rc = PTR_ERR(op_data));
 370
 371         /* foreign dirs are browsed out of Lustre */
 372         if (unlikely(op_data->op_mea1 != NULL &&
 373                      op_data->op_mea1->lsm_md_magic == LMV_MAGIC_FOREIGN)) {
 374                 ll_finish_md_op_data(op_data);
 375                 RETURN(-ENODATA);
 376         }
 377
 378         op_data->op_fid3 = pfid;
 379
 380 #ifdef HAVE_DIR_CONTEXT
 381         ctx->pos = pos;
 382         rc = ll_dir_read(inode, &pos, op_data, ctx);
 383         pos = ctx->pos;
 384 #else
 385         rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
 386 #endif
 387         if (lfd != NULL)
 388                 lfd->lfd_pos = pos;
 389
 390         if (pos == MDS_DIR_END_OFF) {
 391                 if (api32)
 392                         pos = LL_DIR_END_OFF_32BIT;
 393                 else
 394                         pos = LL_DIR_END_OFF;
 395         } else {
 396                 if (api32 && hash64)
 397                         pos = pos >> 32;
 398         }
 399 #ifdef HAVE_DIR_CONTEXT
 400         ctx->pos = pos;
 401 #else
 402         filp->f_pos = pos;
 403 #endif
 404         ll_finish_md_op_data(op_data);
 405
 406 out:
 407         if (!rc)
 408                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 409
 410         RETURN(rc);
 411 }
 412
 413 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 414 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 415 {
 416         struct mgs_send_param *msp;
 417         int rc = 0;
 418
 419         OBD_ALLOC_PTR(msp);
 420         if (!msp)
 421                 return -ENOMEM;
 422
 423         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 424         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 425                                 sizeof(struct mgs_send_param), msp, NULL);
 426         if (rc)
 427                 CERROR("Failed to set parameter: %d\n", rc);
 428         OBD_FREE_PTR(msp);
 429
 430         return rc;
 431 }
 432 #endif
 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 {
 448         struct inode *parent = dparent->d_inode;
 449         struct ptlrpc_request *request = NULL;
 450         struct md_op_data *op_data;
 451         struct ll_sb_info *sbi = ll_i2sbi(parent);
 452         struct inode *inode = NULL;
 453         struct dentry dentry = {
 454                 .d_parent = dparent,
 455                 .d_name = {
 456                         .name = dirname,
 457                         .len = strlen(dirname),
 458                         .hash = ll_full_name_hash(dparent, dirname,
 459                                                   strlen(dirname)),
 460                 },
 461         };
 462         int err;
 463         ENTRY;
 464
 465         if (unlikely(!lmv_magic_supported(lump->lum_magic)))
 466                 RETURN(-EINVAL);
 467
 468         if (lump->lum_magic != LMV_MAGIC_FOREIGN) {
 469                 CDEBUG(D_VFSTRACE,
 470                        "VFS Op:inode="DFID"(%p) name %s stripe_offset %d, stripe_count: %u\n",
 471                        PFID(ll_inode2fid(parent)), parent, dirname,
 472                        (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 473         } else {
 474                 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lump;
 475
 476                 CDEBUG(D_VFSTRACE,
 477                        "VFS Op:inode="DFID"(%p) name %s foreign, length %u, value '%.*s'\n",
 478                        PFID(ll_inode2fid(parent)), parent, dirname,
 479                        lfm->lfm_length, lfm->lfm_length, lfm->lfm_value);
 480         }
 481
 482         if (lump->lum_stripe_count > 1 &&
 483             !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
 484                 RETURN(-EINVAL);
 485
 486         if (IS_DEADDIR(parent) &&
 487             !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
 488                 RETURN(-ENOENT);
 489
 490         if (unlikely(!lmv_magic_supported(cpu_to_le32(lump->lum_magic))))
 491                 lustre_swab_lmv_user_md(lump);
 492
 493         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 494                 mode &= ~current_umask();
 495         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 496         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 497                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 498                                      lump);
 499         if (IS_ERR(op_data))
 500                 RETURN(PTR_ERR(op_data));
 501
 502         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 503                 /* selinux_dentry_init_security() uses dentry->d_parent and name
 504                  * to determine the security context for the file. So our fake
 505                  * dentry should be real enough for this purpose. */
 506                 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
 507                                               &op_data->op_file_secctx_name,
 508                                               &op_data->op_file_secctx,
 509                                               &op_data->op_file_secctx_size);
 510                 if (err < 0)
 511                         GOTO(out_op_data, err);
 512         }
 513
 514         op_data->op_cli_flags |= CLI_SET_MEA;
 515         err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
 516                         from_kuid(&init_user_ns, current_fsuid()),
 517                         from_kgid(&init_user_ns, current_fsgid()),
 518                         cfs_curproc_cap_pack(), 0, &request);
 519         if (err)
 520                 GOTO(out_request, err);
 521
 522         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
 523
 524         err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
 525         if (err)
 526                 GOTO(out_inode, err);
 527
 528         dentry.d_inode = inode;
 529
 530         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 531                 inode_lock(inode);
 532                 err = security_inode_notifysecctx(inode,
 533                                                   op_data->op_file_secctx,
 534                                                   op_data->op_file_secctx_size);
 535                 inode_unlock(inode);
 536         } else {
 537                 err = ll_inode_init_security(&dentry, inode, parent);
 538         }
 539         if (err)
 540                 GOTO(out_inode, err);
 541
 542 out_inode:
 543         if (inode != NULL)
 544                 iput(inode);
 545 out_request:
 546         ptlrpc_req_finished(request);
 547 out_op_data:
 548         ll_finish_md_op_data(op_data);
 549
 550         return err;
 551 }
 552
 553 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 554                      int set_default)
 555 {
 556         struct ll_sb_info *sbi = ll_i2sbi(inode);
 557         struct md_op_data *op_data;
 558         struct ptlrpc_request *req = NULL;
 559         int rc = 0;
 560 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 561         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 562         struct obd_device *mgc = lsi->lsi_mgc;
 563 #endif
 564         int lum_size;
 565         ENTRY;
 566
 567         if (lump != NULL) {
 568                 /*
 569                  * This is coming from userspace, so should be in
 570                  * local endian.  But the MDS would like it in little
 571                  * endian, so we swab it before we send it.
 572                  */
 573                 switch (lump->lmm_magic) {
 574                 case LOV_USER_MAGIC_V1: {
 575                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 576                                 lustre_swab_lov_user_md_v1(lump);
 577                         lum_size = sizeof(struct lov_user_md_v1);
 578                         break;
 579                 }
 580                 case LOV_USER_MAGIC_V3: {
 581                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 582                                 lustre_swab_lov_user_md_v3(
 583                                         (struct lov_user_md_v3 *)lump);
 584                         lum_size = sizeof(struct lov_user_md_v3);
 585                         break;
 586                 }
 587                 case LOV_USER_MAGIC_COMP_V1: {
 588                         if (lump->lmm_magic !=
 589                             cpu_to_le32(LOV_USER_MAGIC_COMP_V1))
 590                                 lustre_swab_lov_comp_md_v1(
 591                                         (struct lov_comp_md_v1 *)lump);
 592                         lum_size = le32_to_cpu(
 593                                 ((struct lov_comp_md_v1 *)lump)->lcm_size);
 594                         break;
 595                 }
 596                 case LMV_USER_MAGIC: {
 597                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 598                                 lustre_swab_lmv_user_md(
 599                                         (struct lmv_user_md *)lump);
 600                         lum_size = sizeof(struct lmv_user_md);
 601                         break;
 602                 }
 603                 case LOV_USER_MAGIC_SPECIFIC: {
 604                         struct lov_user_md_v3 *v3 =
 605                                         (struct lov_user_md_v3 *)lump;
 606                         if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
 607                                 RETURN(-EINVAL);
 608                         if (lump->lmm_magic !=
 609                             cpu_to_le32(LOV_USER_MAGIC_SPECIFIC)) {
 610                                 lustre_swab_lov_user_md_v3(v3);
 611                                 lustre_swab_lov_user_md_objects(v3->lmm_objects,
 612                                                 v3->lmm_stripe_count);
 613                         }
 614                         lum_size = lov_user_md_size(v3->lmm_stripe_count,
 615                                                     LOV_USER_MAGIC_SPECIFIC);
 616                         break;
 617                 }
 618                 default: {
 619                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 620                                         " %#08x != %#08x nor %#08x\n",
 621                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 622                                         LOV_USER_MAGIC_V3);
 623                         RETURN(-EINVAL);
 624                 }
 625                 }
 626         } else {
 627                 lum_size = sizeof(struct lov_user_md_v1);
 628         }
 629
 630         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 631                                      LUSTRE_OPC_ANY, NULL);
 632         if (IS_ERR(op_data))
 633                 RETURN(PTR_ERR(op_data));
 634
 635         /* swabbing is done in lov_setstripe() on server side */
 636         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
 637         ll_finish_md_op_data(op_data);
 638         ptlrpc_req_finished(req);
 639         if (rc)
 640                 RETURN(rc);
 641
 642 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 643         /*
 644          * 2.9 server has stored filesystem default stripe in ROOT xattr,
 645          * and it's stored into system config for backward compatibility.
 646          *
 647          * In the following we use the fact that LOV_USER_MAGIC_V1 and
 648          * LOV_USER_MAGIC_V3 have the same initial fields so we do not
 649          * need the make the distiction between the 2 versions
 650          */
 651         if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
 652             (lump == NULL ||
 653              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
 654              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
 655                 char *param = NULL;
 656                 char *buf;
 657
 658                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 659                 if (param == NULL)
 660                         GOTO(end, rc = -ENOMEM);
 661
 662                 buf = param;
 663                 /* Get fsname and assume devname to be -MDT0000. */
 664                 snprintf(buf, MGS_PARAM_MAXLEN, "%s-MDT0000.lov",
 665                          sbi->ll_fsname);
 666                 buf += strlen(buf);
 667
 668                 /* Set root stripesize */
 669                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
 670                          lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 671                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 672                 if (rc)
 673                         GOTO(end, rc);
 674
 675                 /* Set root stripecount */
 676                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
 677                          lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 678                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 679                 if (rc)
 680                         GOTO(end, rc);
 681
 682                 /* Set root stripeoffset */
 683                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
 684                          lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 685                                 (typeof(lump->lmm_stripe_offset))(-1));
 686                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 687
 688 end:
 689                 if (param != NULL)
 690                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 691         }
 692 #endif
 693         RETURN(rc);
 694 }
 695
 696 /**
 697  * This function will be used to get default LOV/LMV/Default LMV
 698  * @valid will be used to indicate which stripe it will retrieve
 699  *      OBD_MD_MEA              LMV stripe EA
 700  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 701  *      otherwise               Default LOV EA.
 702  * Each time, it can only retrieve 1 stripe EA
 703  **/
 704 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 705                      struct ptlrpc_request **request, u64 valid)
 706 {
 707         struct ll_sb_info *sbi = ll_i2sbi(inode);
 708         struct mdt_body   *body;
 709         struct lov_mds_md *lmm = NULL;
 710         struct ptlrpc_request *req = NULL;
 711         int rc, lmm_size;
 712         struct md_op_data *op_data;
 713         ENTRY;
 714
 715         rc = ll_get_default_mdsize(sbi, &lmm_size);
 716         if (rc)
 717                 RETURN(rc);
 718
 719         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 720                                      0, lmm_size, LUSTRE_OPC_ANY,
 721                                      NULL);
 722         if (IS_ERR(op_data))
 723                 RETURN(PTR_ERR(op_data));
 724
 725         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 726         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 727         ll_finish_md_op_data(op_data);
 728         if (rc < 0) {
 729                 CDEBUG(D_INFO, "md_getattr failed on inode "
 730                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 731                 GOTO(out, rc);
 732         }
 733
 734         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 735         LASSERT(body != NULL);
 736
 737         lmm_size = body->mbo_eadatasize;
 738
 739         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 740             lmm_size == 0) {
 741                 GOTO(out, rc = -ENODATA);
 742         }
 743
 744         lmm = req_capsule_server_sized_get(&req->rq_pill,
 745                                            &RMF_MDT_MD, lmm_size);
 746         LASSERT(lmm != NULL);
 747
 748         /*
 749          * This is coming from the MDS, so is probably in
 750          * little endian.  We convert it to host endian before
 751          * passing it to userspace.
 752          */
 753         /* We don't swab objects for directories */
 754         switch (le32_to_cpu(lmm->lmm_magic)) {
 755         case LOV_MAGIC_V1:
 756                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 757                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 758                 break;
 759         case LOV_MAGIC_V3:
 760                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 761                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 762                 break;
 763         case LOV_MAGIC_COMP_V1:
 764                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 765                         lustre_swab_lov_comp_md_v1(
 766                                         (struct lov_comp_md_v1 *)lmm);
 767                 break;
 768         case LMV_MAGIC_V1:
 769                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 770                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 771                 break;
 772         case LMV_USER_MAGIC:
 773                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 774                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 775                 break;
 776         case LOV_USER_MAGIC_SPECIFIC: {
 777                 struct lov_user_md_v3 *v3 = (struct lov_user_md_v3 *)lmm;
 778
 779                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
 780                         lustre_swab_lov_user_md_v3(v3);
 781                         lustre_swab_lov_user_md_objects(v3->lmm_objects,
 782                                                         v3->lmm_stripe_count);
 783                         }
 784                 }
 785                 break;
 786         case LMV_MAGIC_FOREIGN: {
 787                 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lmm;
 788
 789                 if (LMV_MAGIC_FOREIGN != cpu_to_le32(LMV_MAGIC_FOREIGN)) {
 790                         __swab32s(&lfm->lfm_magic);
 791                         __swab32s(&lfm->lfm_length);
 792                         __swab32s(&lfm->lfm_type);
 793                         __swab32s(&lfm->lfm_flags);
 794                 }
 795                 break;
 796         }
 797         default:
 798                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 799                 rc = -EPROTO;
 800         }
 801 out:
 802         *plmm = lmm;
 803         *plmm_size = lmm_size;
 804         *request = req;
 805         return rc;
 806 }
 807
 808 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 809 {
 810         struct md_op_data       *op_data;
 811         int                     rc;
 812         int                     mdt_index;
 813         ENTRY;
 814
 815         OBD_ALLOC_PTR(op_data);
 816         if (op_data == NULL)
 817                 RETURN(-ENOMEM);
 818
 819         op_data->op_flags |= MF_GET_MDT_IDX;
 820         op_data->op_fid1 = *fid;
 821         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 822         mdt_index = op_data->op_mds;
 823         OBD_FREE_PTR(op_data);
 824         if (rc < 0)
 825                 RETURN(rc);
 826
 827         RETURN(mdt_index);
 828 }
 829
 830 /*
 831  *  Get MDT index for the inode.
 832  */
 833 int ll_get_mdt_idx(struct inode *inode)
 834 {
 835         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 836 }
 837
 838 /**
 839  * Generic handler to do any pre-copy work.
 840  *
 841  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 842  * first information for it that real work has started.
 843  *
 844  * Moreover, for a ARCHIVE request, it will sample the file data version and
 845  * store it in \a copy.
 846  *
 847  * \return 0 on success.
 848  */
 849 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 850 {
 851         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 852         struct hsm_progress_kernel       hpk;
 853         int                              rc = 0;
 854         int                              rc2;
 855         ENTRY;
 856
 857         /* Forge a hsm_progress based on data from copy. */
 858         hpk.hpk_fid = copy->hc_hai.hai_fid;
 859         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 860         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 861         hpk.hpk_extent.length = 0;
 862         hpk.hpk_flags = 0;
 863         hpk.hpk_errval = 0;
 864         hpk.hpk_data_version = 0;
 865
 866
 867         /* For archive request, we need to read the current file version. */
 868         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 869                 struct inode    *inode;
 870                 __u64            data_version = 0;
 871
 872                 /* Get inode for this fid */
 873                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 874                 if (IS_ERR(inode)) {
 875                         hpk.hpk_flags |= HP_FLAG_RETRY;
 876                         /* hpk_errval is >= 0 */
 877                         hpk.hpk_errval = -PTR_ERR(inode);
 878                         GOTO(progress, rc = PTR_ERR(inode));
 879                 }
 880
 881                 /* Read current file data version */
 882                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 883                 iput(inode);
 884                 if (rc != 0) {
 885                         CDEBUG(D_HSM, "Could not read file data version of "
 886                                       DFID" (rc = %d). Archive request ("
 887                                       "%#llx) could not be done.\n",
 888                                       PFID(&copy->hc_hai.hai_fid), rc,
 889                                       copy->hc_hai.hai_cookie);
 890                         hpk.hpk_flags |= HP_FLAG_RETRY;
 891                         /* hpk_errval must be >= 0 */
 892                         hpk.hpk_errval = -rc;
 893                         GOTO(progress, rc);
 894                 }
 895
 896                 /* Store in the hsm_copy for later copytool use.
 897                  * Always modified even if no lsm. */
 898                 copy->hc_data_version = data_version;
 899         }
 900
 901 progress:
 902         /* On error, the request should be considered as completed */
 903         if (hpk.hpk_errval > 0)
 904                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 905
 906         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 907                             &hpk, NULL);
 908
 909         /* Return first error */
 910         RETURN(rc != 0 ? rc : rc2);
 911 }
 912
 913 /**
 914  * Generic handler to do any post-copy work.
 915  *
 916  * It will send the last hsm_progress update to coordinator to inform it
 917  * that copy is finished and whether it was successful or not.
 918  *
 919  * Moreover,
 920  * - for ARCHIVE request, it will sample the file data version and compare it
 921  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 922  *   will be considered as failed.
 923  * - for RESTORE request, it will sample the file data version and send it to
 924  *   coordinator which is useful if the file was imported as 'released'.
 925  *
 926  * \return 0 on success.
 927  */
 928 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 929 {
 930         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 931         struct hsm_progress_kernel       hpk;
 932         int                              rc = 0;
 933         int                              rc2;
 934         ENTRY;
 935
 936         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 937         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 938          * initialized if copy_end was called with copy == NULL.
 939          */
 940
 941         /* Forge a hsm_progress based on data from copy. */
 942         hpk.hpk_fid = copy->hc_hai.hai_fid;
 943         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 944         hpk.hpk_extent = copy->hc_hai.hai_extent;
 945         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 946         hpk.hpk_errval = copy->hc_errval;
 947         hpk.hpk_data_version = 0;
 948
 949         /* For archive request, we need to check the file data was not changed.
 950          *
 951          * For restore request, we need to send the file data version, this is
 952          * useful when the file was created using hsm_import.
 953          */
 954         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 955              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 956             (copy->hc_errval == 0)) {
 957                 struct inode    *inode;
 958                 __u64            data_version = 0;
 959
 960                 /* Get lsm for this fid */
 961                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 962                 if (IS_ERR(inode)) {
 963                         hpk.hpk_flags |= HP_FLAG_RETRY;
 964                         /* hpk_errval must be >= 0 */
 965                         hpk.hpk_errval = -PTR_ERR(inode);
 966                         GOTO(progress, rc = PTR_ERR(inode));
 967                 }
 968
 969                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 970                 iput(inode);
 971                 if (rc) {
 972                         CDEBUG(D_HSM, "Could not read file data version. "
 973                                       "Request could not be confirmed.\n");
 974                         if (hpk.hpk_errval == 0)
 975                                 hpk.hpk_errval = -rc;
 976                         GOTO(progress, rc);
 977                 }
 978
 979                 /* Store in the hsm_copy for later copytool use.
 980                  * Always modified even if no lsm. */
 981                 hpk.hpk_data_version = data_version;
 982
 983                 /* File could have been stripped during archiving, so we need
 984                  * to check anyway. */
 985                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 986                     (copy->hc_data_version != data_version)) {
 987                         CDEBUG(D_HSM, "File data version mismatched. "
 988                               "File content was changed during archiving. "
 989                                DFID", start:%#llx current:%#llx\n",
 990                                PFID(&copy->hc_hai.hai_fid),
 991                                copy->hc_data_version, data_version);
 992                         /* File was changed, send error to cdt. Do not ask for
 993                          * retry because if a file is modified frequently,
 994                          * the cdt will loop on retried archive requests.
 995                          * The policy engine will ask for a new archive later
 996                          * when the file will not be modified for some tunable
 997                          * time */
 998                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 999                         rc = -EBUSY;
1000                         /* hpk_errval must be >= 0 */
1001                         hpk.hpk_errval = -rc;
1002                         GOTO(progress, rc);
1003                 }
1004
1005         }
1006
1007 progress:
1008         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1009                             &hpk, NULL);
1010
1011         /* Return first error */
1012         RETURN(rc != 0 ? rc : rc2);
1013 }
1014
1015
1016 static int copy_and_ct_start(int cmd, struct obd_export *exp,
1017                              const struct lustre_kernelcomm __user *data)
1018 {
1019         struct lustre_kernelcomm *lk;
1020         struct lustre_kernelcomm *tmp;
1021         size_t size = sizeof(*lk);
1022         size_t new_size;
1023         int i;
1024         int rc;
1025
1026         /* copy data from userspace to get numbers of archive_id */
1027         OBD_ALLOC(lk, size);
1028         if (lk == NULL)
1029                 return -ENOMEM;
1030
1031         if (copy_from_user(lk, data, size))
1032                 GOTO(out_lk, rc = -EFAULT);
1033
1034         if (lk->lk_flags & LK_FLG_STOP)
1035                 goto do_ioctl;
1036
1037         if (!(lk->lk_flags & LK_FLG_DATANR)) {
1038                 __u32 archive_mask = lk->lk_data_count;
1039                 int count;
1040
1041                 /* old hsm agent to old MDS */
1042                 if (!exp_connect_archive_id_array(exp))
1043                         goto do_ioctl;
1044
1045                 /* old hsm agent to new MDS */
1046                 lk->lk_flags |= LK_FLG_DATANR;
1047
1048                 if (archive_mask == 0)
1049                         goto do_ioctl;
1050
1051                 count = hweight32(archive_mask);
1052                 new_size = offsetof(struct lustre_kernelcomm, lk_data[count]);
1053                 OBD_ALLOC(tmp, new_size);
1054                 if (tmp == NULL)
1055                         GOTO(out_lk, rc = -ENOMEM);
1056
1057                 memcpy(tmp, lk, size);
1058                 tmp->lk_data_count = count;
1059                 OBD_FREE(lk, size);
1060                 lk = tmp;
1061                 size = new_size;
1062
1063                 count = 0;
1064                 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
1065                         if ((1 << i) & archive_mask) {
1066                                 lk->lk_data[count] = i + 1;
1067                                 count++;
1068                         }
1069                 }
1070                 goto do_ioctl;
1071         }
1072
1073         /* new hsm agent to new mds */
1074         if (lk->lk_data_count > 0) {
1075                 new_size = offsetof(struct lustre_kernelcomm,
1076                                     lk_data[lk->lk_data_count]);
1077                 OBD_ALLOC(tmp, new_size);
1078                 if (tmp == NULL)
1079                         GOTO(out_lk, rc = -ENOMEM);
1080
1081                 OBD_FREE(lk, size);
1082                 lk = tmp;
1083                 size = new_size;
1084
1085                 if (copy_from_user(lk, data, size))
1086                         GOTO(out_lk, rc = -EFAULT);
1087         }
1088
1089         /* new hsm agent to old MDS */
1090         if (!exp_connect_archive_id_array(exp)) {
1091                 __u32 archives = 0;
1092
1093                 if (lk->lk_data_count > LL_HSM_ORIGIN_MAX_ARCHIVE)
1094                         GOTO(out_lk, rc = -EINVAL);
1095
1096                 for (i = 0; i < lk->lk_data_count; i++) {
1097                         if (lk->lk_data[i] > LL_HSM_ORIGIN_MAX_ARCHIVE) {
1098                                 rc = -EINVAL;
1099                                 CERROR("%s: archive id %d requested but only "
1100                                        "[0 - %zu] supported: rc = %d\n",
1101                                        exp->exp_obd->obd_name, lk->lk_data[i],
1102                                        LL_HSM_ORIGIN_MAX_ARCHIVE, rc);
1103                                 GOTO(out_lk, rc);
1104                         }
1105
1106                         if (lk->lk_data[i] == 0) {
1107                                 archives = 0;
1108                                 break;
1109                         }
1110
1111                         archives |= (1 << (lk->lk_data[i] - 1));
1112                 }
1113                 lk->lk_flags &= ~LK_FLG_DATANR;
1114                 lk->lk_data_count = archives;
1115         }
1116 do_ioctl:
1117         rc = obd_iocontrol(cmd, exp, size, lk, NULL);
1118 out_lk:
1119         OBD_FREE(lk, size);
1120         return rc;
1121 }
1122
1123 static int check_owner(int type, int id)
1124 {
1125         switch (type) {
1126         case USRQUOTA:
1127                 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1128                         return -EPERM;
1129                 break;
1130         case GRPQUOTA:
1131                 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1132                         return -EPERM;
1133                 break;
1134         case PRJQUOTA:
1135                 break;
1136         }
1137         return 0;
1138 }
1139
1140 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1141 {
1142         int cmd = qctl->qc_cmd;
1143         int type = qctl->qc_type;
1144         int id = qctl->qc_id;
1145         int valid = qctl->qc_valid;
1146         int rc = 0;
1147         ENTRY;
1148
1149         switch (cmd) {
1150         case Q_SETQUOTA:
1151         case Q_SETINFO:
1152         case LUSTRE_Q_SETDEFAULT:
1153                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1154                         RETURN(-EPERM);
1155                 break;
1156         case Q_GETQUOTA:
1157         case LUSTRE_Q_GETDEFAULT:
1158                 if (check_owner(type, id) &&
1159                     (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1160                         RETURN(-EPERM);
1161                 break;
1162         case Q_GETINFO:
1163                 break;
1164         default:
1165                 CERROR("unsupported quotactl op: %#x\n", cmd);
1166                 RETURN(-ENOTSUPP);
1167         }
1168
1169         if (valid != QC_GENERAL) {
1170                 if (cmd == Q_GETINFO)
1171                         qctl->qc_cmd = Q_GETOINFO;
1172                 else if (cmd == Q_GETQUOTA)
1173                         qctl->qc_cmd = Q_GETOQUOTA;
1174                 else
1175                         RETURN(-EINVAL);
1176
1177                 switch (valid) {
1178                 case QC_MDTIDX:
1179                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1180                                            sizeof(*qctl), qctl, NULL);
1181                         break;
1182                 case QC_OSTIDX:
1183                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1184                                            sizeof(*qctl), qctl, NULL);
1185                         break;
1186                 case QC_UUID:
1187                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1188                                            sizeof(*qctl), qctl, NULL);
1189                         if (rc == -EAGAIN)
1190                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1191                                                    sbi->ll_dt_exp,
1192                                                    sizeof(*qctl), qctl, NULL);
1193                         break;
1194                 default:
1195                         rc = -EINVAL;
1196                         break;
1197                 }
1198
1199                 if (rc)
1200                         RETURN(rc);
1201
1202                 qctl->qc_cmd = cmd;
1203         } else {
1204                 struct obd_quotactl *oqctl;
1205
1206                 OBD_ALLOC_PTR(oqctl);
1207                 if (oqctl == NULL)
1208                         RETURN(-ENOMEM);
1209
1210                 QCTL_COPY(oqctl, qctl);
1211                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1212                 if (rc) {
1213                         OBD_FREE_PTR(oqctl);
1214                         RETURN(rc);
1215                 }
1216                 /* If QIF_SPACE is not set, client should collect the
1217                  * space usage from OSSs by itself */
1218                 if (cmd == Q_GETQUOTA &&
1219                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1220                     !oqctl->qc_dqblk.dqb_curspace) {
1221                         struct obd_quotactl *oqctl_tmp;
1222
1223                         OBD_ALLOC_PTR(oqctl_tmp);
1224                         if (oqctl_tmp == NULL)
1225                                 GOTO(out, rc = -ENOMEM);
1226
1227                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1228                         oqctl_tmp->qc_id = oqctl->qc_id;
1229                         oqctl_tmp->qc_type = oqctl->qc_type;
1230
1231                         /* collect space usage from OSTs */
1232                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1233                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1234                         if (!rc || rc == -EREMOTEIO) {
1235                                 oqctl->qc_dqblk.dqb_curspace =
1236                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1237                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1238                         }
1239
1240                         /* collect space & inode usage from MDTs */
1241                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1242                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1243                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1244                         if (!rc || rc == -EREMOTEIO) {
1245                                 oqctl->qc_dqblk.dqb_curspace +=
1246                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1247                                 oqctl->qc_dqblk.dqb_curinodes =
1248                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1249                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1250                         } else {
1251                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1252                         }
1253
1254                         OBD_FREE_PTR(oqctl_tmp);
1255                 }
1256 out:
1257                 QCTL_COPY(qctl, oqctl);
1258                 OBD_FREE_PTR(oqctl);
1259         }
1260
1261         RETURN(rc);
1262 }
1263
1264 /* This function tries to get a single name component,
1265  * to send to the server. No actual path traversal involved,
1266  * so we limit to NAME_MAX */
1267 static char *ll_getname(const char __user *filename)
1268 {
1269         int ret = 0, len;
1270         char *tmp;
1271
1272         OBD_ALLOC(tmp, NAME_MAX + 1);
1273
1274         if (!tmp)
1275                 return ERR_PTR(-ENOMEM);
1276
1277         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1278         if (len < 0)
1279                 ret = -ENOENT;
1280         else if (len > NAME_MAX)
1281                 ret = -ENAMETOOLONG;
1282
1283         if (ret) {
1284                 OBD_FREE(tmp, NAME_MAX + 1);
1285                 tmp =  ERR_PTR(ret);
1286         }
1287         return tmp;
1288 }
1289
1290 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1291
1292 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1293 {
1294         struct dentry *dentry = file_dentry(file);
1295         struct inode *inode = file_inode(file);
1296         struct ll_sb_info *sbi = ll_i2sbi(inode);
1297         struct obd_ioctl_data *data;
1298         int rc = 0;
1299         ENTRY;
1300
1301         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1302                PFID(ll_inode2fid(inode)), inode, cmd);
1303
1304         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1305         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1306                 return -ENOTTY;
1307
1308         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1309         switch (cmd) {
1310         case FS_IOC_GETFLAGS:
1311         case FS_IOC_SETFLAGS:
1312                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1313         case FSFILT_IOC_GETVERSION:
1314         case FS_IOC_GETVERSION:
1315                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1316         /* We need to special case any other ioctls we want to handle,
1317          * to send them to the MDS/OST as appropriate and to properly
1318          * network encode the arg field. */
1319         case FS_IOC_SETVERSION:
1320                 RETURN(-ENOTSUPP);
1321
1322         case LL_IOC_GET_MDTIDX: {
1323                 int mdtidx;
1324
1325                 mdtidx = ll_get_mdt_idx(inode);
1326                 if (mdtidx < 0)
1327                         RETURN(mdtidx);
1328
1329                 if (put_user((int)mdtidx, (int __user *)arg))
1330                         RETURN(-EFAULT);
1331
1332                 return 0;
1333         }
1334         case IOC_MDC_LOOKUP: {
1335                                      int namelen, len = 0;
1336                 char *buf = NULL;
1337                 char *filename;
1338
1339                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1340                 if (rc != 0)
1341                         RETURN(rc);
1342                 data = (void *)buf;
1343
1344                 filename = data->ioc_inlbuf1;
1345                 namelen = strlen(filename);
1346                 if (namelen < 1) {
1347                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1348                         GOTO(out_free, rc = -EINVAL);
1349                 }
1350
1351                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1352                 if (rc < 0) {
1353                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1354                                sbi->ll_fsname, namelen, filename, rc);
1355                         GOTO(out_free, rc);
1356                 }
1357 out_free:
1358                 OBD_FREE_LARGE(buf, len);
1359                 return rc;
1360         }
1361         case LL_IOC_LMV_SETSTRIPE: {
1362                 struct lmv_user_md  *lum;
1363                 char            *buf = NULL;
1364                 char            *filename;
1365                 int              namelen = 0;
1366                 int              lumlen = 0;
1367                 umode_t          mode;
1368                 int              len;
1369                 int              rc;
1370
1371                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1372                 if (rc)
1373                         RETURN(rc);
1374
1375                 data = (void *)buf;
1376                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1377                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1378                         GOTO(lmv_out_free, rc = -EINVAL);
1379
1380                 filename = data->ioc_inlbuf1;
1381                 namelen = data->ioc_inllen1;
1382
1383                 if (namelen < 1) {
1384                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1385                         GOTO(lmv_out_free, rc = -EINVAL);
1386                 }
1387                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1388                 lumlen = data->ioc_inllen2;
1389
1390                 if (!lmv_magic_supported(lum->lum_magic)) {
1391                         CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
1392                                lum->lum_magic, -EINVAL);
1393                         GOTO(lmv_out_free, rc = -EINVAL);
1394                 }
1395
1396                 if ((lum->lum_magic == LMV_USER_MAGIC ||
1397                      lum->lum_magic == LMV_USER_MAGIC_SPECIFIC) &&
1398                     lumlen < sizeof(*lum)) {
1399                         CERROR("%s: wrong lum size %d for magic %x : rc = %d\n",
1400                                filename, lumlen, lum->lum_magic, -EINVAL);
1401                         GOTO(lmv_out_free, rc = -EINVAL);
1402                 }
1403
1404                 if (lum->lum_magic == LMV_MAGIC_FOREIGN &&
1405                     lumlen < sizeof(struct lmv_foreign_md)) {
1406                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1407                                filename, lum->lum_magic, lumlen, -EFAULT);
1408                         GOTO(lmv_out_free, rc = -EINVAL);
1409                 }
1410
1411 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1412                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1413 #else
1414                 mode = data->ioc_type;
1415 #endif
1416                 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1417 lmv_out_free:
1418                 OBD_FREE_LARGE(buf, len);
1419                 RETURN(rc);
1420
1421         }
1422         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1423                 struct lmv_user_md        lum;
1424                 struct lmv_user_md __user *ulump =
1425                                         (struct lmv_user_md __user *)arg;
1426                 int                       rc;
1427
1428                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1429                         RETURN(-EFAULT);
1430
1431                 if (lum.lum_magic != LMV_USER_MAGIC)
1432                         RETURN(-EINVAL);
1433
1434                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1435
1436                 RETURN(rc);
1437         }
1438         case LL_IOC_LOV_SETSTRIPE_NEW:
1439         case LL_IOC_LOV_SETSTRIPE: {
1440                 struct lov_user_md_v3 *lumv3 = NULL;
1441                 struct lov_user_md_v1 lumv1;
1442                 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1443                 struct lov_user_md_v1 __user *lumv1p =
1444                         (struct lov_user_md_v1 __user *)arg;
1445                 struct lov_user_md_v3 __user *lumv3p =
1446                         (struct lov_user_md_v3 __user *)arg;
1447                 int lum_size = 0;
1448
1449                 int set_default = 0;
1450
1451                 CLASSERT(sizeof(struct lov_user_md_v3) >
1452                          sizeof(struct lov_comp_md_v1));
1453                 CLASSERT(sizeof(*lumv3) == sizeof(*lumv3p));
1454                 /* first try with v1 which is smaller than v3 */
1455                 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1456                         RETURN(-EFAULT);
1457
1458                 if (inode->i_sb->s_root == file_dentry(file))
1459                         set_default = 1;
1460
1461                 switch (lumv1.lmm_magic) {
1462                 case LOV_USER_MAGIC_V3:
1463                 case LOV_USER_MAGIC_SPECIFIC:
1464                         lum_size = ll_lov_user_md_size(&lumv1);
1465                         if (lum_size < 0)
1466                                 RETURN(lum_size);
1467                         OBD_ALLOC(lumv3, lum_size);
1468                         if (!lumv3)
1469                                 RETURN(-ENOMEM);
1470                         if (copy_from_user(lumv3, lumv3p, lum_size))
1471                                 GOTO(out, rc = -EFAULT);
1472                         lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1473                         break;
1474                 case LOV_USER_MAGIC_V1:
1475                         break;
1476                 default:
1477                         GOTO(out, rc = -ENOTSUPP);
1478                 }
1479
1480                 /* in v1 and v3 cases lumv1 points to data */
1481                 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1482 out:
1483                 if (lumv3)
1484                         OBD_FREE(lumv3, lum_size);
1485                 RETURN(rc);
1486         }
1487         case LL_IOC_LMV_GETSTRIPE: {
1488                 struct lmv_user_md __user *ulmv =
1489                                         (struct lmv_user_md __user *)arg;
1490                 struct lmv_user_md      lum;
1491                 struct ptlrpc_request   *request = NULL;
1492                 union lmv_mds_md        *lmm = NULL;
1493                 int                     lmmsize;
1494                 u64                     valid = 0;
1495                 struct lmv_user_md      *tmp = NULL;
1496                 int                     mdt_index;
1497                 int                     lum_size;
1498                 int                     stripe_count;
1499                 int                     max_stripe_count;
1500                 int                     i;
1501                 int                     rc;
1502
1503                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1504                         RETURN(-EFAULT);
1505
1506                 max_stripe_count = lum.lum_stripe_count;
1507                 /* lum_magic will indicate which stripe the ioctl will like
1508                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1509                  * is for default LMV stripe */
1510                 if (lum.lum_magic == LMV_MAGIC_V1)
1511                         valid |= OBD_MD_MEA;
1512                 else if (lum.lum_magic == LMV_USER_MAGIC)
1513                         valid |= OBD_MD_DEFAULT_MEA;
1514                 else
1515                         RETURN(-EINVAL);
1516
1517                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1518                                       valid);
1519                 if (rc != 0)
1520                         GOTO(finish_req, rc);
1521
1522                 /* Get default LMV EA */
1523                 if (lum.lum_magic == LMV_USER_MAGIC) {
1524                         if (lmmsize > sizeof(*ulmv))
1525                                 GOTO(finish_req, rc = -EINVAL);
1526
1527                         if (copy_to_user(ulmv, lmm, lmmsize))
1528                                 GOTO(finish_req, rc = -EFAULT);
1529
1530                         GOTO(finish_req, rc);
1531                 }
1532
1533                 /* if foreign LMV case, fake stripes number */
1534                 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1535                         struct lmv_foreign_md *lfm;
1536
1537                         lfm = (struct lmv_foreign_md *)lmm;
1538                         if (lfm->lfm_length < XATTR_SIZE_MAX -
1539                             offsetof(typeof(*lfm), lfm_value)) {
1540                                 __u32 size = lfm->lfm_length +
1541                                              offsetof(typeof(*lfm), lfm_value);
1542
1543                                 stripe_count = lmv_foreign_to_md_stripes(size);
1544                         } else {
1545                                 CERROR("invalid %d foreign size returned\n",
1546                                             lfm->lfm_length);
1547                                 return -EINVAL;
1548                         }
1549                 } else {
1550                         stripe_count = lmv_mds_md_stripe_count_get(lmm);
1551                 }
1552                 if (max_stripe_count < stripe_count) {
1553                         lum.lum_stripe_count = stripe_count;
1554                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1555                                 GOTO(finish_req, rc = -EFAULT);
1556                         GOTO(finish_req, rc = -E2BIG);
1557                 }
1558
1559                 /* enough room on user side and foreign case */
1560                 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1561                         struct lmv_foreign_md *lfm;
1562                         __u32 size;
1563
1564                         lfm = (struct lmv_foreign_md *)lmm;
1565                         size = lfm->lfm_length +
1566                                offsetof(struct lmv_foreign_md, lfm_value);
1567                         if (copy_to_user(ulmv, lfm, size))
1568                                 GOTO(finish_req, rc = -EFAULT);
1569                         GOTO(finish_req, rc);
1570                 }
1571
1572                 lum_size = lmv_user_md_size(stripe_count,
1573                                             LMV_USER_MAGIC_SPECIFIC);
1574                 OBD_ALLOC(tmp, lum_size);
1575                 if (tmp == NULL)
1576                         GOTO(finish_req, rc = -ENOMEM);
1577
1578                 mdt_index = ll_get_mdt_idx(inode);
1579                 if (mdt_index < 0)
1580                         GOTO(out_tmp, rc = -ENOMEM);
1581
1582                 tmp->lum_magic = LMV_MAGIC_V1;
1583                 tmp->lum_stripe_count = 0;
1584                 tmp->lum_stripe_offset = mdt_index;
1585                 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1586                 for (i = 0; i < stripe_count; i++) {
1587                         struct lu_fid   fid;
1588
1589                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1590                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1591                         if (mdt_index < 0)
1592                                 GOTO(out_tmp, rc = mdt_index);
1593
1594                         tmp->lum_objects[i].lum_mds = mdt_index;
1595                         tmp->lum_objects[i].lum_fid = fid;
1596                         tmp->lum_stripe_count++;
1597                 }
1598
1599                 if (copy_to_user(ulmv, tmp, lum_size))
1600                         GOTO(out_tmp, rc = -EFAULT);
1601 out_tmp:
1602                 OBD_FREE(tmp, lum_size);
1603 finish_req:
1604                 ptlrpc_req_finished(request);
1605                 return rc;
1606         }
1607
1608         case LL_IOC_REMOVE_ENTRY: {
1609                 char            *filename = NULL;
1610                 int              namelen = 0;
1611                 int              rc;
1612
1613                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1614                  * unsupported server, which might crash the server(LU-2730),
1615                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1616                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1617                  * server will support REINT_RMENTRY XXX*/
1618                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1619                         RETURN(-ENOTSUPP);
1620
1621                 filename = ll_getname((const char __user *)arg);
1622                 if (IS_ERR(filename))
1623                         RETURN(PTR_ERR(filename));
1624
1625                 namelen = strlen(filename);
1626                 if (namelen < 1)
1627                         GOTO(out_rmdir, rc = -EINVAL);
1628
1629                 rc = ll_rmdir_entry(inode, filename, namelen);
1630 out_rmdir:
1631                 if (filename)
1632                         ll_putname(filename);
1633                 RETURN(rc);
1634         }
1635         case LL_IOC_LOV_SWAP_LAYOUTS:
1636                 RETURN(-EPERM);
1637         case IOC_OBD_STATFS:
1638                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1639         case LL_IOC_LOV_GETSTRIPE:
1640         case LL_IOC_LOV_GETSTRIPE_NEW:
1641         case LL_IOC_MDC_GETINFO:
1642         case IOC_MDC_GETFILEINFO:
1643         case IOC_MDC_GETFILESTRIPE: {
1644                 struct ptlrpc_request *request = NULL;
1645                 struct lov_user_md __user *lump;
1646                 struct lov_mds_md *lmm = NULL;
1647                 struct mdt_body *body;
1648                 char *filename = NULL;
1649                 int lmmsize;
1650
1651                 if (cmd == IOC_MDC_GETFILEINFO ||
1652                     cmd == IOC_MDC_GETFILESTRIPE) {
1653                         filename = ll_getname((const char __user *)arg);
1654                         if (IS_ERR(filename))
1655                                 RETURN(PTR_ERR(filename));
1656
1657                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1658                                                       &lmmsize, &request);
1659                 } else {
1660                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1661                                               &request, 0);
1662                 }
1663
1664                 if (request) {
1665                         body = req_capsule_server_get(&request->rq_pill,
1666                                                       &RMF_MDT_BODY);
1667                         LASSERT(body != NULL);
1668                 } else {
1669                         GOTO(out_req, rc);
1670                 }
1671
1672                 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1673                                        cmd == LL_IOC_MDC_GETINFO)) {
1674                         lmmsize = 0;
1675                         rc = 0;
1676                 }
1677
1678                 if (rc < 0)
1679                         GOTO(out_req, rc);
1680
1681                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1682                     cmd == LL_IOC_LOV_GETSTRIPE ||
1683                     cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1684                         lump = (struct lov_user_md __user *)arg;
1685                 } else {
1686                         struct lov_user_mds_data __user *lmdp;
1687                         lmdp = (struct lov_user_mds_data __user *)arg;
1688                         lump = &lmdp->lmd_lmm;
1689                 }
1690
1691                 if (lmmsize == 0) {
1692                         /* If the file has no striping then zero out *lump so
1693                          * that the caller isn't confused by garbage. */
1694                         if (clear_user(lump, sizeof(*lump)))
1695                                 GOTO(out_req, rc = -EFAULT);
1696                 } else if (copy_to_user(lump, lmm, lmmsize)) {
1697                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1698                                 GOTO(out_req, rc = -EFAULT);
1699                         rc = -EOVERFLOW;
1700                 }
1701
1702                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1703                         struct lov_user_mds_data __user *lmdp;
1704                         lstat_t st = { 0 };
1705
1706                         st.st_dev       = inode->i_sb->s_dev;
1707                         st.st_mode      = body->mbo_mode;
1708                         st.st_nlink     = body->mbo_nlink;
1709                         st.st_uid       = body->mbo_uid;
1710                         st.st_gid       = body->mbo_gid;
1711                         st.st_rdev      = body->mbo_rdev;
1712                         st.st_size      = body->mbo_size;
1713                         st.st_blksize   = PAGE_SIZE;
1714                         st.st_blocks    = body->mbo_blocks;
1715                         st.st_atime     = body->mbo_atime;
1716                         st.st_mtime     = body->mbo_mtime;
1717                         st.st_ctime     = body->mbo_ctime;
1718                         st.st_ino       = cl_fid_build_ino(&body->mbo_fid1,
1719                                                 sbi->ll_flags &
1720                                                 LL_SBI_32BIT_API);
1721
1722                         lmdp = (struct lov_user_mds_data __user *)arg;
1723                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1724                                 GOTO(out_req, rc = -EFAULT);
1725                 }
1726
1727                 EXIT;
1728         out_req:
1729                 ptlrpc_req_finished(request);
1730                 if (filename)
1731                         ll_putname(filename);
1732                 return rc;
1733         }
1734         case OBD_IOC_QUOTACTL: {
1735                 struct if_quotactl *qctl;
1736
1737                 OBD_ALLOC_PTR(qctl);
1738                 if (!qctl)
1739                         RETURN(-ENOMEM);
1740
1741                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1742                         GOTO(out_quotactl, rc = -EFAULT);
1743
1744                 rc = quotactl_ioctl(sbi, qctl);
1745
1746                 if (rc == 0 &&
1747                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1748                         rc = -EFAULT;
1749
1750         out_quotactl:
1751                 OBD_FREE_PTR(qctl);
1752                 RETURN(rc);
1753         }
1754         case OBD_IOC_GETDTNAME:
1755         case OBD_IOC_GETMDNAME:
1756                 RETURN(ll_get_obd_name(inode, cmd, arg));
1757         case LL_IOC_FLUSHCTX:
1758                 RETURN(ll_flush_ctx(inode));
1759         case LL_IOC_GETOBDCOUNT: {
1760                 u32 count, vallen;
1761                 struct obd_export *exp;
1762
1763                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1764                         RETURN(-EFAULT);
1765
1766                 /* get ost count when count is zero, get mdt count otherwise */
1767                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1768                 vallen = sizeof(count);
1769                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1770                                   KEY_TGT_COUNT, &vallen, &count);
1771                 if (rc) {
1772                         CERROR("get target count failed: %d\n", rc);
1773                         RETURN(rc);
1774                 }
1775
1776                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1777                         RETURN(-EFAULT);
1778
1779                 RETURN(0);
1780         }
1781         case LL_IOC_PATH2FID:
1782                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1783                                      sizeof(struct lu_fid)))
1784                         RETURN(-EFAULT);
1785                 RETURN(0);
1786         case LL_IOC_GET_CONNECT_FLAGS: {
1787                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1788                                      (void __user *)arg));
1789         }
1790         case OBD_IOC_FID2PATH:
1791                 RETURN(ll_fid2path(inode, (void __user *)arg));
1792         case LL_IOC_GETPARENT:
1793                 RETURN(ll_getparent(file, (void __user *)arg));
1794         case LL_IOC_FID2MDTIDX: {
1795                 struct obd_export *exp = ll_i2mdexp(inode);
1796                 struct lu_fid     fid;
1797                 __u32             index;
1798
1799                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1800                                    sizeof(fid)))
1801                         RETURN(-EFAULT);
1802
1803                 /* Call mdc_iocontrol */
1804                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1805                                    (__u32 __user *)&index);
1806                 if (rc != 0)
1807                         RETURN(rc);
1808
1809                 RETURN(index);
1810         }
1811         case LL_IOC_HSM_REQUEST: {
1812                 struct hsm_user_request *hur;
1813                 ssize_t                  totalsize;
1814
1815                 OBD_ALLOC_PTR(hur);
1816                 if (hur == NULL)
1817                         RETURN(-ENOMEM);
1818
1819                 /* We don't know the true size yet; copy the fixed-size part */
1820                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1821                         OBD_FREE_PTR(hur);
1822                         RETURN(-EFAULT);
1823                 }
1824
1825                 /* Compute the whole struct size */
1826                 totalsize = hur_len(hur);
1827                 OBD_FREE_PTR(hur);
1828                 if (totalsize < 0)
1829                         RETURN(-E2BIG);
1830
1831                 /* Final size will be more than double totalsize */
1832                 if (totalsize >= MDS_MAXREQSIZE / 3)
1833                         RETURN(-E2BIG);
1834
1835                 OBD_ALLOC_LARGE(hur, totalsize);
1836                 if (hur == NULL)
1837                         RETURN(-ENOMEM);
1838
1839                 /* Copy the whole struct */
1840                 if (copy_from_user(hur, (void __user *)arg, totalsize))
1841                         GOTO(out_hur, rc = -EFAULT);
1842
1843                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1844                         const struct lu_fid *fid;
1845                         struct inode *f;
1846                         int i;
1847
1848                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1849                                 fid = &hur->hur_user_item[i].hui_fid;
1850                                 f = search_inode_for_lustre(inode->i_sb, fid);
1851                                 if (IS_ERR(f)) {
1852                                         rc = PTR_ERR(f);
1853                                         break;
1854                                 }
1855
1856                                 rc = ll_hsm_release(f);
1857                                 iput(f);
1858                                 if (rc != 0)
1859                                         break;
1860                         }
1861                 } else {
1862                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1863                                            hur, NULL);
1864                 }
1865
1866 out_hur:
1867                 OBD_FREE_LARGE(hur, totalsize);
1868
1869                 RETURN(rc);
1870         }
1871         case LL_IOC_HSM_PROGRESS: {
1872                 struct hsm_progress_kernel      hpk;
1873                 struct hsm_progress             hp;
1874
1875                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1876                         RETURN(-EFAULT);
1877
1878                 hpk.hpk_fid = hp.hp_fid;
1879                 hpk.hpk_cookie = hp.hp_cookie;
1880                 hpk.hpk_extent = hp.hp_extent;
1881                 hpk.hpk_flags = hp.hp_flags;
1882                 hpk.hpk_errval = hp.hp_errval;
1883                 hpk.hpk_data_version = 0;
1884
1885                 /* File may not exist in Lustre; all progress
1886                  * reported to Lustre root */
1887                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1888                                    NULL);
1889                 RETURN(rc);
1890         }
1891         case LL_IOC_HSM_CT_START:
1892                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1893                         RETURN(-EPERM);
1894
1895                 rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
1896                                        (struct lustre_kernelcomm __user *)arg);
1897                 RETURN(rc);
1898
1899         case LL_IOC_HSM_COPY_START: {
1900                 struct hsm_copy *copy;
1901                 int              rc;
1902
1903                 OBD_ALLOC_PTR(copy);
1904                 if (copy == NULL)
1905                         RETURN(-ENOMEM);
1906                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1907                         OBD_FREE_PTR(copy);
1908                         RETURN(-EFAULT);
1909                 }
1910
1911                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1912                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1913                         rc = -EFAULT;
1914
1915                 OBD_FREE_PTR(copy);
1916                 RETURN(rc);
1917         }
1918         case LL_IOC_HSM_COPY_END: {
1919                 struct hsm_copy *copy;
1920                 int              rc;
1921
1922                 OBD_ALLOC_PTR(copy);
1923                 if (copy == NULL)
1924                         RETURN(-ENOMEM);
1925                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1926                         OBD_FREE_PTR(copy);
1927                         RETURN(-EFAULT);
1928                 }
1929
1930                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1931                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1932                         rc = -EFAULT;
1933
1934                 OBD_FREE_PTR(copy);
1935                 RETURN(rc);
1936         }
1937         case LL_IOC_MIGRATE: {
1938                 struct lmv_user_md *lum;
1939                 char *buf = NULL;
1940                 int len;
1941                 char *filename;
1942                 int namelen = 0;
1943                 int rc;
1944
1945                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1946                 if (rc)
1947                         RETURN(rc);
1948
1949                 data = (struct obd_ioctl_data *)buf;
1950                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1951                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1952                         GOTO(migrate_free, rc = -EINVAL);
1953
1954                 filename = data->ioc_inlbuf1;
1955                 namelen = data->ioc_inllen1;
1956
1957                 if (namelen < 1 || namelen != strlen(filename) + 1) {
1958                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1959                         GOTO(migrate_free, rc = -EINVAL);
1960                 }
1961
1962                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1963                 if (lum->lum_magic != LMV_USER_MAGIC &&
1964                     lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
1965                         rc = -EINVAL;
1966                         CERROR("%s: wrong lum magic %x: rc = %d\n",
1967                                filename, lum->lum_magic, rc);
1968                         GOTO(migrate_free, rc);
1969                 }
1970
1971                 rc = ll_migrate(inode, file, lum, filename);
1972 migrate_free:
1973                 OBD_FREE_LARGE(buf, len);
1974
1975                 RETURN(rc);
1976         }
1977         case LL_IOC_FSGETXATTR:
1978                 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
1979         case LL_IOC_FSSETXATTR:
1980                 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
1981         default:
1982                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1983                                      (void __user *)arg));
1984         }
1985 }
1986
1987 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1988 {
1989         struct inode *inode = file->f_mapping->host;
1990         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1991         struct ll_sb_info *sbi = ll_i2sbi(inode);
1992         int api32 = ll_need_32bit_api(sbi);
1993         loff_t ret = -EINVAL;
1994         ENTRY;
1995
1996         inode_lock(inode);
1997         switch (origin) {
1998                 case SEEK_SET:
1999                         break;
2000                 case SEEK_CUR:
2001                         offset += file->f_pos;
2002                         break;
2003                 case SEEK_END:
2004                         if (offset > 0)
2005                                 GOTO(out, ret);
2006                         if (api32)
2007                                 offset += LL_DIR_END_OFF_32BIT;
2008                         else
2009                                 offset += LL_DIR_END_OFF;
2010                         break;
2011                 default:
2012                         GOTO(out, ret);
2013         }
2014
2015         if (offset >= 0 &&
2016             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
2017              (!api32 && offset <= LL_DIR_END_OFF))) {
2018                 if (offset != file->f_pos) {
2019                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
2020                             (!api32 && offset == LL_DIR_END_OFF))
2021                                 fd->lfd_pos = MDS_DIR_END_OFF;
2022                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
2023                                 fd->lfd_pos = offset << 32;
2024                         else
2025                                 fd->lfd_pos = offset;
2026                         file->f_pos = offset;
2027                         file->f_version = 0;
2028                 }
2029                 ret = offset;
2030         }
2031         GOTO(out, ret);
2032
2033 out:
2034         inode_unlock(inode);
2035         return ret;
2036 }
2037
2038 static int ll_dir_open(struct inode *inode, struct file *file)
2039 {
2040         ENTRY;
2041         RETURN(ll_file_open(inode, file));
2042 }
2043
2044 static int ll_dir_release(struct inode *inode, struct file *file)
2045 {
2046         ENTRY;
2047         RETURN(ll_file_release(inode, file));
2048 }
2049
2050 const struct file_operations ll_dir_operations = {
2051         .llseek         = ll_dir_seek,
2052         .open           = ll_dir_open,
2053         .release        = ll_dir_release,
2054         .read           = generic_read_dir,
2055 #ifdef HAVE_DIR_CONTEXT
2056         .iterate        = ll_iterate,
2057 #else
2058         .readdir        = ll_readdir,
2059 #endif
2060         .unlocked_ioctl = ll_dir_ioctl,
2061         .fsync          = ll_fsync,
2062 };