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