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                 int rc2 = ll_dir_get_default_layout(inode, (void **)&lmm,
 755                                                     &lmm_size, &root_req, valid,
 756                                                     GET_DEFAULT_LAYOUT_ROOT);
 757                 if (rc2 == 0)
 758                         rc = 0;
 759         }
 760
 761         *plmm = lmm;
 762         *plmm_size = lmm_size;
 763         *request = req;
 764         if (root_request != NULL)
 765                 *root_request = root_req;
 766
 767         RETURN(rc);
 768 }
 769
 770 /**
 771  * This function will be used to get default LOV/LMV/Default LMV
 772  * @valid will be used to indicate which stripe it will retrieve
 773  *      OBD_MD_MEA              LMV stripe EA
 774  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 775  *      otherwise               Default LOV EA.
 776  * Each time, it can only retrieve 1 stripe EA
 777  **/
 778 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 779                      struct ptlrpc_request **request, u64 valid)
 780 {
 781         struct ptlrpc_request *req = NULL;
 782         struct lov_mds_md *lmm = NULL;
 783         int lmm_size = 0;
 784         int rc = 0;
 785         ENTRY;
 786
 787         rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
 788                                        &req, valid, 0);
 789
 790         *plmm = lmm;
 791         *plmm_size = lmm_size;
 792         *request = req;
 793
 794         RETURN(rc);
 795 }
 796
 797 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 798 {
 799         struct md_op_data       *op_data;
 800         int                     rc;
 801         int                     mdt_index;
 802         ENTRY;
 803
 804         OBD_ALLOC_PTR(op_data);
 805         if (op_data == NULL)
 806                 RETURN(-ENOMEM);
 807
 808         op_data->op_flags |= MF_GET_MDT_IDX;
 809         op_data->op_fid1 = *fid;
 810         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 811         mdt_index = op_data->op_mds;
 812         OBD_FREE_PTR(op_data);
 813         if (rc < 0)
 814                 RETURN(rc);
 815
 816         RETURN(mdt_index);
 817 }
 818
 819 /*
 820  *  Get MDT index for the inode.
 821  */
 822 int ll_get_mdt_idx(struct inode *inode)
 823 {
 824         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 825 }
 826
 827 /**
 828  * Generic handler to do any pre-copy work.
 829  *
 830  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 831  * first information for it that real work has started.
 832  *
 833  * Moreover, for a ARCHIVE request, it will sample the file data version and
 834  * store it in \a copy.
 835  *
 836  * \return 0 on success.
 837  */
 838 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 839 {
 840         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 841         struct hsm_progress_kernel       hpk;
 842         int                              rc = 0;
 843         int                              rc2;
 844         ENTRY;
 845
 846         /* Forge a hsm_progress based on data from copy. */
 847         hpk.hpk_fid = copy->hc_hai.hai_fid;
 848         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 849         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 850         hpk.hpk_extent.length = 0;
 851         hpk.hpk_flags = 0;
 852         hpk.hpk_errval = 0;
 853         hpk.hpk_data_version = 0;
 854
 855
 856         /* For archive request, we need to read the current file version. */
 857         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 858                 struct inode    *inode;
 859                 __u64            data_version = 0;
 860
 861                 /* Get inode for this fid */
 862                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 863                 if (IS_ERR(inode)) {
 864                         hpk.hpk_flags |= HP_FLAG_RETRY;
 865                         /* hpk_errval is >= 0 */
 866                         hpk.hpk_errval = -PTR_ERR(inode);
 867                         GOTO(progress, rc = PTR_ERR(inode));
 868                 }
 869
 870                 /* Read current file data version */
 871                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 872                 iput(inode);
 873                 if (rc != 0) {
 874                         CDEBUG(D_HSM, "Could not read file data version of "
 875                                       DFID" (rc = %d). Archive request ("
 876                                       "%#llx) could not be done.\n",
 877                                       PFID(&copy->hc_hai.hai_fid), rc,
 878                                       copy->hc_hai.hai_cookie);
 879                         hpk.hpk_flags |= HP_FLAG_RETRY;
 880                         /* hpk_errval must be >= 0 */
 881                         hpk.hpk_errval = -rc;
 882                         GOTO(progress, rc);
 883                 }
 884
 885                 /* Store in the hsm_copy for later copytool use.
 886                  * Always modified even if no lsm. */
 887                 copy->hc_data_version = data_version;
 888         }
 889
 890 progress:
 891         /* On error, the request should be considered as completed */
 892         if (hpk.hpk_errval > 0)
 893                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 894
 895         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 896                             &hpk, NULL);
 897
 898         /* Return first error */
 899         RETURN(rc != 0 ? rc : rc2);
 900 }
 901
 902 /**
 903  * Generic handler to do any post-copy work.
 904  *
 905  * It will send the last hsm_progress update to coordinator to inform it
 906  * that copy is finished and whether it was successful or not.
 907  *
 908  * Moreover,
 909  * - for ARCHIVE request, it will sample the file data version and compare it
 910  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 911  *   will be considered as failed.
 912  * - for RESTORE request, it will sample the file data version and send it to
 913  *   coordinator which is useful if the file was imported as 'released'.
 914  *
 915  * \return 0 on success.
 916  */
 917 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 918 {
 919         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 920         struct hsm_progress_kernel       hpk;
 921         int                              rc = 0;
 922         int                              rc2;
 923         ENTRY;
 924
 925         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 926         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 927          * initialized if copy_end was called with copy == NULL.
 928          */
 929
 930         /* Forge a hsm_progress based on data from copy. */
 931         hpk.hpk_fid = copy->hc_hai.hai_fid;
 932         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 933         hpk.hpk_extent = copy->hc_hai.hai_extent;
 934         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 935         hpk.hpk_errval = copy->hc_errval;
 936         hpk.hpk_data_version = 0;
 937
 938         /* For archive request, we need to check the file data was not changed.
 939          *
 940          * For restore request, we need to send the file data version, this is
 941          * useful when the file was created using hsm_import.
 942          */
 943         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 944              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 945             (copy->hc_errval == 0)) {
 946                 struct inode    *inode;
 947                 __u64            data_version = 0;
 948
 949                 /* Get lsm for this fid */
 950                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 951                 if (IS_ERR(inode)) {
 952                         hpk.hpk_flags |= HP_FLAG_RETRY;
 953                         /* hpk_errval must be >= 0 */
 954                         hpk.hpk_errval = -PTR_ERR(inode);
 955                         GOTO(progress, rc = PTR_ERR(inode));
 956                 }
 957
 958                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 959                 iput(inode);
 960                 if (rc) {
 961                         CDEBUG(D_HSM, "Could not read file data version. "
 962                                       "Request could not be confirmed.\n");
 963                         if (hpk.hpk_errval == 0)
 964                                 hpk.hpk_errval = -rc;
 965                         GOTO(progress, rc);
 966                 }
 967
 968                 /* Store in the hsm_copy for later copytool use.
 969                  * Always modified even if no lsm. */
 970                 hpk.hpk_data_version = data_version;
 971
 972                 /* File could have been stripped during archiving, so we need
 973                  * to check anyway. */
 974                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 975                     (copy->hc_data_version != data_version)) {
 976                         CDEBUG(D_HSM, "File data version mismatched. "
 977                               "File content was changed during archiving. "
 978                                DFID", start:%#llx current:%#llx\n",
 979                                PFID(&copy->hc_hai.hai_fid),
 980                                copy->hc_data_version, data_version);
 981                         /* File was changed, send error to cdt. Do not ask for
 982                          * retry because if a file is modified frequently,
 983                          * the cdt will loop on retried archive requests.
 984                          * The policy engine will ask for a new archive later
 985                          * when the file will not be modified for some tunable
 986                          * time */
 987                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 988                         rc = -EBUSY;
 989                         /* hpk_errval must be >= 0 */
 990                         hpk.hpk_errval = -rc;
 991                         GOTO(progress, rc);
 992                 }
 993
 994         }
 995
 996 progress:
 997         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 998                             &hpk, NULL);
 999
1000         /* Return first error */
1001         RETURN(rc != 0 ? rc : rc2);
1002 }
1003
1004
1005 static int copy_and_ct_start(int cmd, struct obd_export *exp,
1006                              const struct lustre_kernelcomm __user *data)
1007 {
1008         struct lustre_kernelcomm *lk;
1009         struct lustre_kernelcomm *tmp;
1010         size_t size = sizeof(*lk);
1011         size_t new_size;
1012         int i;
1013         int rc;
1014
1015         /* copy data from userspace to get numbers of archive_id */
1016         OBD_ALLOC(lk, size);
1017         if (lk == NULL)
1018                 return -ENOMEM;
1019
1020         if (copy_from_user(lk, data, size))
1021                 GOTO(out_lk, rc = -EFAULT);
1022
1023         if (lk->lk_flags & LK_FLG_STOP)
1024                 goto do_ioctl;
1025
1026         if (!(lk->lk_flags & LK_FLG_DATANR)) {
1027                 __u32 archive_mask = lk->lk_data_count;
1028                 int count;
1029
1030                 /* old hsm agent to old MDS */
1031                 if (!exp_connect_archive_id_array(exp))
1032                         goto do_ioctl;
1033
1034                 /* old hsm agent to new MDS */
1035                 lk->lk_flags |= LK_FLG_DATANR;
1036
1037                 if (archive_mask == 0)
1038                         goto do_ioctl;
1039
1040                 count = hweight32(archive_mask);
1041                 new_size = offsetof(struct lustre_kernelcomm, lk_data[count]);
1042                 OBD_ALLOC(tmp, new_size);
1043                 if (tmp == NULL)
1044                         GOTO(out_lk, rc = -ENOMEM);
1045
1046                 memcpy(tmp, lk, size);
1047                 tmp->lk_data_count = count;
1048                 OBD_FREE(lk, size);
1049                 lk = tmp;
1050                 size = new_size;
1051
1052                 count = 0;
1053                 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
1054                         if (BIT(i) & archive_mask) {
1055                                 lk->lk_data[count] = i + 1;
1056                                 count++;
1057                         }
1058                 }
1059                 goto do_ioctl;
1060         }
1061
1062         /* new hsm agent to new mds */
1063         if (lk->lk_data_count > 0) {
1064                 new_size = offsetof(struct lustre_kernelcomm,
1065                                     lk_data[lk->lk_data_count]);
1066                 OBD_ALLOC(tmp, new_size);
1067                 if (tmp == NULL)
1068                         GOTO(out_lk, rc = -ENOMEM);
1069
1070                 OBD_FREE(lk, size);
1071                 lk = tmp;
1072                 size = new_size;
1073
1074                 if (copy_from_user(lk, data, size))
1075                         GOTO(out_lk, rc = -EFAULT);
1076         }
1077
1078         /* new hsm agent to old MDS */
1079         if (!exp_connect_archive_id_array(exp)) {
1080                 __u32 archives = 0;
1081
1082                 if (lk->lk_data_count > LL_HSM_ORIGIN_MAX_ARCHIVE)
1083                         GOTO(out_lk, rc = -EINVAL);
1084
1085                 for (i = 0; i < lk->lk_data_count; i++) {
1086                         if (lk->lk_data[i] > LL_HSM_ORIGIN_MAX_ARCHIVE) {
1087                                 rc = -EINVAL;
1088                                 CERROR("%s: archive id %d requested but only "
1089                                        "[0 - %zu] supported: rc = %d\n",
1090                                        exp->exp_obd->obd_name, lk->lk_data[i],
1091                                        LL_HSM_ORIGIN_MAX_ARCHIVE, rc);
1092                                 GOTO(out_lk, rc);
1093                         }
1094
1095                         if (lk->lk_data[i] == 0) {
1096                                 archives = 0;
1097                                 break;
1098                         }
1099
1100                         archives |= (1 << (lk->lk_data[i] - 1));
1101                 }
1102                 lk->lk_flags &= ~LK_FLG_DATANR;
1103                 lk->lk_data_count = archives;
1104         }
1105 do_ioctl:
1106         rc = obd_iocontrol(cmd, exp, size, lk, NULL);
1107 out_lk:
1108         OBD_FREE(lk, size);
1109         return rc;
1110 }
1111
1112 static int check_owner(int type, int id)
1113 {
1114         switch (type) {
1115         case USRQUOTA:
1116                 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1117                         return -EPERM;
1118                 break;
1119         case GRPQUOTA:
1120                 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1121                         return -EPERM;
1122                 break;
1123         case PRJQUOTA:
1124                 break;
1125         }
1126         return 0;
1127 }
1128
1129 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1130 {
1131         int cmd = qctl->qc_cmd;
1132         int type = qctl->qc_type;
1133         int id = qctl->qc_id;
1134         int valid = qctl->qc_valid;
1135         int rc = 0;
1136         ENTRY;
1137
1138         switch (cmd) {
1139         case Q_SETQUOTA:
1140         case Q_SETINFO:
1141         case LUSTRE_Q_SETDEFAULT:
1142         case LUSTRE_Q_SETQUOTAPOOL:
1143         case LUSTRE_Q_SETINFOPOOL:
1144                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1145                         RETURN(-EPERM);
1146                 break;
1147         case Q_GETQUOTA:
1148         case LUSTRE_Q_GETDEFAULT:
1149         case LUSTRE_Q_GETQUOTAPOOL:
1150                 if (check_owner(type, id) &&
1151                     (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1152                         RETURN(-EPERM);
1153                 break;
1154         case Q_GETINFO:
1155         case LUSTRE_Q_GETINFOPOOL:
1156                 break;
1157         default:
1158                 CERROR("unsupported quotactl op: %#x\n", cmd);
1159                 RETURN(-ENOTSUPP);
1160         }
1161
1162         if (valid != QC_GENERAL) {
1163                 if (cmd == Q_GETINFO)
1164                         qctl->qc_cmd = Q_GETOINFO;
1165                 else if (cmd == Q_GETQUOTA ||
1166                          cmd == LUSTRE_Q_GETQUOTAPOOL)
1167                         qctl->qc_cmd = Q_GETOQUOTA;
1168                 else
1169                         RETURN(-EINVAL);
1170
1171                 switch (valid) {
1172                 case QC_MDTIDX:
1173                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1174                                            sizeof(*qctl), qctl, NULL);
1175                         break;
1176                 case QC_OSTIDX:
1177                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1178                                            sizeof(*qctl), qctl, NULL);
1179                         break;
1180                 case QC_UUID:
1181                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1182                                            sizeof(*qctl), qctl, NULL);
1183                         if (rc == -EAGAIN)
1184                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1185                                                    sbi->ll_dt_exp,
1186                                                    sizeof(*qctl), qctl, NULL);
1187                         break;
1188                 default:
1189                         rc = -EINVAL;
1190                         break;
1191                 }
1192
1193                 if (rc)
1194                         RETURN(rc);
1195
1196                 qctl->qc_cmd = cmd;
1197         } else {
1198                 struct obd_quotactl *oqctl;
1199                 int oqctl_len = sizeof(*oqctl);
1200
1201                 if (LUSTRE_Q_CMD_IS_POOL(cmd))
1202                         oqctl_len += LOV_MAXPOOLNAME + 1;
1203
1204                 OBD_ALLOC(oqctl, oqctl_len);
1205                 if (oqctl == NULL)
1206                         RETURN(-ENOMEM);
1207
1208                 QCTL_COPY(oqctl, qctl);
1209                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1210                 if (rc) {
1211                         OBD_FREE(oqctl, oqctl_len);
1212                         RETURN(rc);
1213                 }
1214                 /* If QIF_SPACE is not set, client should collect the
1215                  * space usage from OSSs by itself */
1216                 if ((cmd == Q_GETQUOTA || cmd == LUSTRE_Q_GETQUOTAPOOL) &&
1217                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1218                     !oqctl->qc_dqblk.dqb_curspace) {
1219                         struct obd_quotactl *oqctl_tmp;
1220                         int qctl_len = sizeof(*oqctl_tmp) + LOV_MAXPOOLNAME + 1;
1221
1222                         OBD_ALLOC(oqctl_tmp, qctl_len);
1223                         if (oqctl_tmp == NULL)
1224                                 GOTO(out, rc = -ENOMEM);
1225
1226                         if (cmd == LUSTRE_Q_GETQUOTAPOOL) {
1227                                 oqctl_tmp->qc_cmd = LUSTRE_Q_GETQUOTAPOOL;
1228                                 memcpy(oqctl_tmp->qc_poolname,
1229                                        qctl->qc_poolname,
1230                                        LOV_MAXPOOLNAME + 1);
1231                         } else {
1232                                 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1233                         }
1234                         oqctl_tmp->qc_id = oqctl->qc_id;
1235                         oqctl_tmp->qc_type = oqctl->qc_type;
1236
1237                         /* collect space usage from OSTs */
1238                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1239                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1240                         if (!rc || rc == -EREMOTEIO) {
1241                                 oqctl->qc_dqblk.dqb_curspace =
1242                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1243                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1244                         }
1245
1246                         /* collect space & inode usage from MDTs */
1247                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1248                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1249                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1250                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1251                         if (!rc || rc == -EREMOTEIO) {
1252                                 oqctl->qc_dqblk.dqb_curspace +=
1253                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1254                                 oqctl->qc_dqblk.dqb_curinodes =
1255                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1256                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1257                         } else {
1258                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1259                         }
1260
1261                         OBD_FREE(oqctl_tmp, qctl_len);
1262                 }
1263 out:
1264                 QCTL_COPY(qctl, oqctl);
1265                 OBD_FREE(oqctl, oqctl_len);
1266         }
1267
1268         RETURN(rc);
1269 }
1270
1271 int ll_rmfid(struct file *file, void __user *arg)
1272 {
1273         const struct fid_array __user *ufa = arg;
1274         struct fid_array *lfa = NULL;
1275         size_t size;
1276         unsigned nr;
1277         int i, rc, *rcs = NULL;
1278         ENTRY;
1279
1280         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
1281             !(ll_i2sbi(file_inode(file))->ll_flags & LL_SBI_USER_FID2PATH))
1282                 RETURN(-EPERM);
1283         /* Only need to get the buflen */
1284         if (get_user(nr, &ufa->fa_nr))
1285                 RETURN(-EFAULT);
1286         /* DoS protection */
1287         if (nr > OBD_MAX_FIDS_IN_ARRAY)
1288                 RETURN(-E2BIG);
1289
1290         size = offsetof(struct fid_array, fa_fids[nr]);
1291         OBD_ALLOC(lfa, size);
1292         if (!lfa)
1293                 RETURN(-ENOMEM);
1294         OBD_ALLOC_PTR_ARRAY(rcs, nr);
1295         if (!rcs)
1296                 GOTO(free_lfa, rc = -ENOMEM);
1297
1298         if (copy_from_user(lfa, arg, size))
1299                 GOTO(free_rcs, rc = -EFAULT);
1300
1301         /* Call mdc_iocontrol */
1302         rc = md_rmfid(ll_i2mdexp(file_inode(file)), lfa, rcs, NULL);
1303         if (!rc) {
1304                 for (i = 0; i < nr; i++)
1305                         if (rcs[i])
1306                                 lfa->fa_fids[i].f_ver = rcs[i];
1307                 if (copy_to_user(arg, lfa, size))
1308                         rc = -EFAULT;
1309         }
1310
1311 free_rcs:
1312         OBD_FREE_PTR_ARRAY(rcs, nr);
1313 free_lfa:
1314         OBD_FREE(lfa, size);
1315
1316         RETURN(rc);
1317 }
1318
1319 /* This function tries to get a single name component,
1320  * to send to the server. No actual path traversal involved,
1321  * so we limit to NAME_MAX */
1322 static char *ll_getname(const char __user *filename)
1323 {
1324         int ret = 0, len;
1325         char *tmp;
1326
1327         OBD_ALLOC(tmp, NAME_MAX + 1);
1328
1329         if (!tmp)
1330                 return ERR_PTR(-ENOMEM);
1331
1332         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1333         if (len < 0)
1334                 ret = -ENOENT;
1335         else if (len > NAME_MAX)
1336                 ret = -ENAMETOOLONG;
1337
1338         if (ret) {
1339                 OBD_FREE(tmp, NAME_MAX + 1);
1340                 tmp =  ERR_PTR(ret);
1341         }
1342         return tmp;
1343 }
1344
1345 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1346
1347 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1348 {
1349         struct dentry *dentry = file_dentry(file);
1350         struct inode *inode = file_inode(file);
1351         struct ll_sb_info *sbi = ll_i2sbi(inode);
1352         struct obd_ioctl_data *data;
1353         int rc = 0;
1354         ENTRY;
1355
1356         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1357                PFID(ll_inode2fid(inode)), inode, cmd);
1358
1359         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1360         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1361                 return -ENOTTY;
1362
1363         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1364         switch (cmd) {
1365         case FS_IOC_GETFLAGS:
1366         case FS_IOC_SETFLAGS:
1367                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1368         case FSFILT_IOC_GETVERSION:
1369         case FS_IOC_GETVERSION:
1370                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1371         /* We need to special case any other ioctls we want to handle,
1372          * to send them to the MDS/OST as appropriate and to properly
1373          * network encode the arg field. */
1374         case FS_IOC_SETVERSION:
1375                 RETURN(-ENOTSUPP);
1376
1377         case LL_IOC_GET_MDTIDX: {
1378                 int mdtidx;
1379
1380                 mdtidx = ll_get_mdt_idx(inode);
1381                 if (mdtidx < 0)
1382                         RETURN(mdtidx);
1383
1384                 if (put_user((int)mdtidx, (int __user *)arg))
1385                         RETURN(-EFAULT);
1386
1387                 return 0;
1388         }
1389         case IOC_MDC_LOOKUP: {
1390                                      int namelen, len = 0;
1391                 char *buf = NULL;
1392                 char *filename;
1393
1394                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1395                 if (rc != 0)
1396                         RETURN(rc);
1397                 data = (void *)buf;
1398
1399                 filename = data->ioc_inlbuf1;
1400                 namelen = strlen(filename);
1401                 if (namelen < 1) {
1402                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1403                         GOTO(out_free, rc = -EINVAL);
1404                 }
1405
1406                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1407                 if (rc < 0) {
1408                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1409                                sbi->ll_fsname, namelen, filename, rc);
1410                         GOTO(out_free, rc);
1411                 }
1412 out_free:
1413                 OBD_FREE_LARGE(buf, len);
1414                 return rc;
1415         }
1416         case LL_IOC_LMV_SETSTRIPE: {
1417                 struct lmv_user_md  *lum;
1418                 char            *buf = NULL;
1419                 char            *filename;
1420                 int              namelen = 0;
1421                 int              lumlen = 0;
1422                 umode_t          mode;
1423                 int              len;
1424                 int              rc;
1425
1426                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1427                 if (rc)
1428                         RETURN(rc);
1429
1430                 data = (void *)buf;
1431                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1432                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1433                         GOTO(lmv_out_free, rc = -EINVAL);
1434
1435                 filename = data->ioc_inlbuf1;
1436                 namelen = data->ioc_inllen1;
1437
1438                 if (namelen < 1) {
1439                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1440                         GOTO(lmv_out_free, rc = -EINVAL);
1441                 }
1442                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1443                 lumlen = data->ioc_inllen2;
1444
1445                 if (!lmv_user_magic_supported(lum->lum_magic)) {
1446                         CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
1447                                lum->lum_magic, -EINVAL);
1448                         GOTO(lmv_out_free, rc = -EINVAL);
1449                 }
1450
1451                 if ((lum->lum_magic == LMV_USER_MAGIC ||
1452                      lum->lum_magic == LMV_USER_MAGIC_SPECIFIC) &&
1453                     lumlen < sizeof(*lum)) {
1454                         CERROR("%s: wrong lum size %d for magic %x : rc = %d\n",
1455                                filename, lumlen, lum->lum_magic, -EINVAL);
1456                         GOTO(lmv_out_free, rc = -EINVAL);
1457                 }
1458
1459                 if (lum->lum_magic == LMV_MAGIC_FOREIGN &&
1460                     lumlen < sizeof(struct lmv_foreign_md)) {
1461                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1462                                filename, lum->lum_magic, lumlen, -EFAULT);
1463                         GOTO(lmv_out_free, rc = -EINVAL);
1464                 }
1465
1466                 mode = data->ioc_type;
1467                 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1468 lmv_out_free:
1469                 OBD_FREE_LARGE(buf, len);
1470                 RETURN(rc);
1471
1472         }
1473         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1474                 struct lmv_user_md        lum;
1475                 struct lmv_user_md __user *ulump =
1476                                         (struct lmv_user_md __user *)arg;
1477                 int                       rc;
1478
1479                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1480                         RETURN(-EFAULT);
1481
1482                 if (lum.lum_magic != LMV_USER_MAGIC)
1483                         RETURN(-EINVAL);
1484
1485                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1486
1487                 RETURN(rc);
1488         }
1489         case LL_IOC_LOV_SETSTRIPE_NEW:
1490         case LL_IOC_LOV_SETSTRIPE: {
1491                 struct lov_user_md_v3 *lumv3 = NULL;
1492                 struct lov_user_md_v1 lumv1;
1493                 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1494                 struct lov_user_md_v1 __user *lumv1p =
1495                         (struct lov_user_md_v1 __user *)arg;
1496                 struct lov_user_md_v3 __user *lumv3p =
1497                         (struct lov_user_md_v3 __user *)arg;
1498                 int lum_size = 0;
1499
1500                 int set_default = 0;
1501
1502                 BUILD_BUG_ON(sizeof(struct lov_user_md_v3) <=
1503                              sizeof(struct lov_comp_md_v1));
1504                 BUILD_BUG_ON(sizeof(*lumv3) != sizeof(*lumv3p));
1505                 /* first try with v1 which is smaller than v3 */
1506                 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1507                         RETURN(-EFAULT);
1508
1509                 if (inode->i_sb->s_root == file_dentry(file))
1510                         set_default = 1;
1511
1512                 switch (lumv1.lmm_magic) {
1513                 case LOV_USER_MAGIC_V3:
1514                 case LOV_USER_MAGIC_SPECIFIC:
1515                         lum_size = ll_lov_user_md_size(&lumv1);
1516                         if (lum_size < 0)
1517                                 RETURN(lum_size);
1518                         OBD_ALLOC(lumv3, lum_size);
1519                         if (!lumv3)
1520                                 RETURN(-ENOMEM);
1521                         if (copy_from_user(lumv3, lumv3p, lum_size))
1522                                 GOTO(out, rc = -EFAULT);
1523                         lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1524                         break;
1525                 case LOV_USER_MAGIC_V1:
1526                         break;
1527                 default:
1528                         GOTO(out, rc = -ENOTSUPP);
1529                 }
1530
1531                 /* in v1 and v3 cases lumv1 points to data */
1532                 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1533 out:
1534                 if (lumv3)
1535                         OBD_FREE(lumv3, lum_size);
1536                 RETURN(rc);
1537         }
1538         case LL_IOC_LMV_GETSTRIPE: {
1539                 struct lmv_user_md __user *ulmv =
1540                                         (struct lmv_user_md __user *)arg;
1541                 struct lmv_user_md      lum;
1542                 struct ptlrpc_request   *request = NULL;
1543                 struct ptlrpc_request   *root_request = NULL;
1544                 union lmv_mds_md        *lmm = NULL;
1545                 int                     lmmsize;
1546                 u64                     valid = 0;
1547                 struct lmv_user_md      *tmp = NULL;
1548                 int                     mdt_index;
1549                 int                     lum_size;
1550                 int                     stripe_count;
1551                 int                     max_stripe_count;
1552                 int                     i;
1553                 int                     rc;
1554
1555                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1556                         RETURN(-EFAULT);
1557
1558                 max_stripe_count = lum.lum_stripe_count;
1559                 /* lum_magic will indicate which stripe the ioctl will like
1560                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1561                  * is for default LMV stripe */
1562                 if (lum.lum_magic == LMV_MAGIC_V1)
1563                         valid |= OBD_MD_MEA;
1564                 else if (lum.lum_magic == LMV_USER_MAGIC)
1565                         valid |= OBD_MD_DEFAULT_MEA;
1566                 else
1567                         RETURN(-EINVAL);
1568
1569                 rc = ll_dir_getstripe_default(inode, (void **)&lmm, &lmmsize,
1570                                               &request, &root_request, valid);
1571                 if (rc != 0)
1572                         GOTO(finish_req, rc);
1573
1574                 /* Get default LMV EA */
1575                 if (lum.lum_magic == LMV_USER_MAGIC) {
1576                         if (lmmsize > sizeof(*ulmv))
1577                                 GOTO(finish_req, rc = -EINVAL);
1578
1579                         if (copy_to_user(ulmv, lmm, lmmsize))
1580                                 GOTO(finish_req, rc = -EFAULT);
1581
1582                         GOTO(finish_req, rc);
1583                 }
1584
1585                 /* if foreign LMV case, fake stripes number */
1586                 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1587                         struct lmv_foreign_md *lfm;
1588
1589                         lfm = (struct lmv_foreign_md *)lmm;
1590                         if (lfm->lfm_length < XATTR_SIZE_MAX -
1591                             offsetof(typeof(*lfm), lfm_value)) {
1592                                 __u32 size = lfm->lfm_length +
1593                                              offsetof(typeof(*lfm), lfm_value);
1594
1595                                 stripe_count = lmv_foreign_to_md_stripes(size);
1596                         } else {
1597                                 CERROR("invalid %d foreign size returned\n",
1598                                             lfm->lfm_length);
1599                                 return -EINVAL;
1600                         }
1601                 } else {
1602                         stripe_count = lmv_mds_md_stripe_count_get(lmm);
1603                 }
1604                 if (max_stripe_count < stripe_count) {
1605                         lum.lum_stripe_count = stripe_count;
1606                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1607                                 GOTO(finish_req, rc = -EFAULT);
1608                         GOTO(finish_req, rc = -E2BIG);
1609                 }
1610
1611                 /* enough room on user side and foreign case */
1612                 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1613                         struct lmv_foreign_md *lfm;
1614                         __u32 size;
1615
1616                         lfm = (struct lmv_foreign_md *)lmm;
1617                         size = lfm->lfm_length +
1618                                offsetof(struct lmv_foreign_md, lfm_value);
1619                         if (copy_to_user(ulmv, lfm, size))
1620                                 GOTO(finish_req, rc = -EFAULT);
1621                         GOTO(finish_req, rc);
1622                 }
1623
1624                 lum_size = lmv_user_md_size(stripe_count,
1625                                             LMV_USER_MAGIC_SPECIFIC);
1626                 OBD_ALLOC(tmp, lum_size);
1627                 if (tmp == NULL)
1628                         GOTO(finish_req, rc = -ENOMEM);
1629
1630                 mdt_index = ll_get_mdt_idx(inode);
1631                 if (mdt_index < 0)
1632                         GOTO(out_tmp, rc = -ENOMEM);
1633
1634                 tmp->lum_magic = LMV_MAGIC_V1;
1635                 tmp->lum_stripe_count = 0;
1636                 tmp->lum_stripe_offset = mdt_index;
1637                 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1638                 for (i = 0; i < stripe_count; i++) {
1639                         struct lu_fid   fid;
1640
1641                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1642                         if (fid_is_sane(&fid)) {
1643                                 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1644                                 if (mdt_index < 0)
1645                                         GOTO(out_tmp, rc = mdt_index);
1646
1647                                 tmp->lum_objects[i].lum_mds = mdt_index;
1648                                 tmp->lum_objects[i].lum_fid = fid;
1649                         }
1650
1651                         tmp->lum_stripe_count++;
1652                 }
1653
1654                 if (copy_to_user(ulmv, tmp, lum_size))
1655                         GOTO(out_tmp, rc = -EFAULT);
1656 out_tmp:
1657                 OBD_FREE(tmp, lum_size);
1658 finish_req:
1659                 ptlrpc_req_finished(request);
1660                 ptlrpc_req_finished(root_request);
1661                 return rc;
1662         }
1663
1664         case LL_IOC_REMOVE_ENTRY: {
1665                 char            *filename = NULL;
1666                 int              namelen = 0;
1667                 int              rc;
1668
1669                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1670                  * unsupported server, which might crash the server(LU-2730),
1671                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1672                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1673                  * server will support REINT_RMENTRY XXX*/
1674                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1675                         RETURN(-ENOTSUPP);
1676
1677                 filename = ll_getname((const char __user *)arg);
1678                 if (IS_ERR(filename))
1679                         RETURN(PTR_ERR(filename));
1680
1681                 namelen = strlen(filename);
1682                 if (namelen < 1)
1683                         GOTO(out_rmdir, rc = -EINVAL);
1684
1685                 rc = ll_rmdir_entry(inode, filename, namelen);
1686 out_rmdir:
1687                 if (filename)
1688                         ll_putname(filename);
1689                 RETURN(rc);
1690         }
1691         case LL_IOC_RMFID:
1692                 RETURN(ll_rmfid(file, (void __user *)arg));
1693         case LL_IOC_LOV_SWAP_LAYOUTS:
1694                 RETURN(-EPERM);
1695         case IOC_OBD_STATFS:
1696                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1697         case LL_IOC_LOV_GETSTRIPE:
1698         case LL_IOC_LOV_GETSTRIPE_NEW:
1699         case LL_IOC_MDC_GETINFO:
1700         case LL_IOC_MDC_GETINFO_OLD:
1701         case IOC_MDC_GETFILEINFO:
1702         case IOC_MDC_GETFILEINFO_OLD:
1703         case IOC_MDC_GETFILESTRIPE: {
1704                 struct ptlrpc_request *request = NULL;
1705                 struct ptlrpc_request *root_request = NULL;
1706                 struct lov_user_md __user *lump;
1707                 struct lov_mds_md *lmm = NULL;
1708                 struct mdt_body *body;
1709                 char *filename = NULL;
1710                 lstat_t __user *statp = NULL;
1711                 lstatx_t __user *stxp = NULL;
1712                 __u64 __user *flagsp = NULL;
1713                 __u32 __user *lmmsizep = NULL;
1714                 struct lu_fid __user *fidp = NULL;
1715                 int lmmsize;
1716
1717                 if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1718                     cmd == IOC_MDC_GETFILEINFO ||
1719                     cmd == IOC_MDC_GETFILESTRIPE) {
1720                         filename = ll_getname((const char __user *)arg);
1721                         if (IS_ERR(filename))
1722                                 RETURN(PTR_ERR(filename));
1723
1724                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1725                                                       &lmmsize, &request);
1726                 } else {
1727                         rc = ll_dir_getstripe_default(inode, (void **)&lmm,
1728                                                       &lmmsize, &request,
1729                                                       &root_request, 0);
1730                 }
1731
1732                 if (request) {
1733                         body = req_capsule_server_get(&request->rq_pill,
1734                                                       &RMF_MDT_BODY);
1735                         LASSERT(body != NULL);
1736                 } else {
1737                         GOTO(out_req, rc);
1738                 }
1739
1740                 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1741                                        cmd == LL_IOC_MDC_GETINFO ||
1742                                        cmd == IOC_MDC_GETFILEINFO_OLD ||
1743                                        cmd == LL_IOC_MDC_GETINFO_OLD)) {
1744                         lmmsize = 0;
1745                         rc = 0;
1746                 }
1747
1748                 if (rc < 0)
1749                         GOTO(out_req, rc);
1750
1751                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1752                     cmd == LL_IOC_LOV_GETSTRIPE ||
1753                     cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1754                         lump = (struct lov_user_md __user *)arg;
1755                 } else if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1756                            cmd == LL_IOC_MDC_GETINFO_OLD){
1757                         struct lov_user_mds_data_v1 __user *lmdp;
1758
1759                         lmdp = (struct lov_user_mds_data_v1 __user *)arg;
1760                         statp = &lmdp->lmd_st;
1761                         lump = &lmdp->lmd_lmm;
1762                 } else {
1763                         struct lov_user_mds_data __user *lmdp;
1764
1765                         lmdp = (struct lov_user_mds_data __user *)arg;
1766                         fidp = &lmdp->lmd_fid;
1767                         stxp = &lmdp->lmd_stx;
1768                         flagsp = &lmdp->lmd_flags;
1769                         lmmsizep = &lmdp->lmd_lmmsize;
1770                         lump = &lmdp->lmd_lmm;
1771                 }
1772
1773                 if (lmmsize == 0) {
1774                         /* If the file has no striping then zero out *lump so
1775                          * that the caller isn't confused by garbage. */
1776                         if (clear_user(lump, sizeof(*lump)))
1777                                 GOTO(out_req, rc = -EFAULT);
1778                 } else if (copy_to_user(lump, lmm, lmmsize)) {
1779                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1780                                 GOTO(out_req, rc = -EFAULT);
1781                         rc = -EOVERFLOW;
1782                 }
1783
1784                 if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1785                     cmd == LL_IOC_MDC_GETINFO_OLD) {
1786                         lstat_t st = { 0 };
1787
1788                         st.st_dev       = inode->i_sb->s_dev;
1789                         st.st_mode      = body->mbo_mode;
1790                         st.st_nlink     = body->mbo_nlink;
1791                         st.st_uid       = body->mbo_uid;
1792                         st.st_gid       = body->mbo_gid;
1793                         st.st_rdev      = body->mbo_rdev;
1794                         st.st_size      = body->mbo_size;
1795                         st.st_blksize   = PAGE_SIZE;
1796                         st.st_blocks    = body->mbo_blocks;
1797                         st.st_atime     = body->mbo_atime;
1798                         st.st_mtime     = body->mbo_mtime;
1799                         st.st_ctime     = body->mbo_ctime;
1800                         st.st_ino       = cl_fid_build_ino(&body->mbo_fid1,
1801                                                 sbi->ll_flags &
1802                                                 LL_SBI_32BIT_API);
1803
1804                         if (copy_to_user(statp, &st, sizeof(st)))
1805                                 GOTO(out_req, rc = -EFAULT);
1806                 } else if (cmd == IOC_MDC_GETFILEINFO ||
1807                            cmd == LL_IOC_MDC_GETINFO) {
1808                         lstatx_t stx = { 0 };
1809                         __u64 valid = body->mbo_valid;
1810
1811                         stx.stx_blksize = PAGE_SIZE;
1812                         stx.stx_nlink = body->mbo_nlink;
1813                         stx.stx_uid = body->mbo_uid;
1814                         stx.stx_gid = body->mbo_gid;
1815                         stx.stx_mode = body->mbo_mode;
1816                         stx.stx_ino = cl_fid_build_ino(&body->mbo_fid1,
1817                                                        sbi->ll_flags &
1818                                                        LL_SBI_32BIT_API);
1819                         stx.stx_size = body->mbo_size;
1820                         stx.stx_blocks = body->mbo_blocks;
1821                         stx.stx_atime.tv_sec = body->mbo_atime;
1822                         stx.stx_ctime.tv_sec = body->mbo_ctime;
1823                         stx.stx_mtime.tv_sec = body->mbo_mtime;
1824                         stx.stx_btime.tv_sec = body->mbo_btime;
1825                         stx.stx_rdev_major = MAJOR(body->mbo_rdev);
1826                         stx.stx_rdev_minor = MINOR(body->mbo_rdev);
1827                         stx.stx_dev_major = MAJOR(inode->i_sb->s_dev);
1828                         stx.stx_dev_minor = MINOR(inode->i_sb->s_dev);
1829                         stx.stx_mask |= STATX_BASIC_STATS | STATX_BTIME;
1830
1831                         /*
1832                          * For a striped directory, the size and blocks returned
1833                          * from MDT is not correct.
1834                          * The size and blocks are aggregated by client across
1835                          * all stripes.
1836                          * Thus for a striped directory, do not return the valid
1837                          * FLSIZE and FLBLOCKS flags to the caller.
1838                          * However, this whould be better decided by the MDS
1839                          * instead of the client.
1840                          */
1841                         if (cmd == LL_IOC_MDC_GETINFO &&
1842                             ll_i2info(inode)->lli_lsm_md != NULL)
1843                                 valid &= ~(OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
1844
1845                         if (flagsp && copy_to_user(flagsp, &valid,
1846                                                    sizeof(*flagsp)))
1847                                 GOTO(out_req, rc = -EFAULT);
1848
1849                         if (fidp && copy_to_user(fidp, &body->mbo_fid1,
1850                                                  sizeof(*fidp)))
1851                                 GOTO(out_req, rc = -EFAULT);
1852
1853                         if (!(valid & OBD_MD_FLSIZE))
1854                                 stx.stx_mask &= ~STATX_SIZE;
1855                         if (!(valid & OBD_MD_FLBLOCKS))
1856                                 stx.stx_mask &= ~STATX_BLOCKS;
1857
1858                         if (stxp && copy_to_user(stxp, &stx, sizeof(stx)))
1859                                 GOTO(out_req, rc = -EFAULT);
1860
1861                         if (lmmsizep && copy_to_user(lmmsizep, &lmmsize,
1862                                                      sizeof(*lmmsizep)))
1863                                 GOTO(out_req, rc = -EFAULT);
1864                 }
1865
1866                 EXIT;
1867 out_req:
1868                 ptlrpc_req_finished(request);
1869                 ptlrpc_req_finished(root_request);
1870                 if (filename)
1871                         ll_putname(filename);
1872                 return rc;
1873         }
1874         case OBD_IOC_QUOTACTL: {
1875                 struct if_quotactl *qctl;
1876                 int qctl_len = sizeof(*qctl) + LOV_MAXPOOLNAME + 1;
1877
1878                 OBD_ALLOC(qctl, qctl_len);
1879                 if (!qctl)
1880                         RETURN(-ENOMEM);
1881
1882                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1883                         GOTO(out_quotactl, rc = -EFAULT);
1884
1885                 if (LUSTRE_Q_CMD_IS_POOL(qctl->qc_cmd)) {
1886                         char __user *from = (char __user *)arg +
1887                                         offsetof(typeof(*qctl), qc_poolname);
1888                         if (copy_from_user(qctl->qc_poolname, from,
1889                                            LOV_MAXPOOLNAME + 1))
1890                                 GOTO(out_quotactl, rc = -EFAULT);
1891                 }
1892
1893                 rc = quotactl_ioctl(sbi, qctl);
1894                 if (rc == 0 &&
1895                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1896                         rc = -EFAULT;
1897
1898 out_quotactl:
1899                 OBD_FREE(qctl, qctl_len);
1900                 RETURN(rc);
1901         }
1902         case OBD_IOC_GETDTNAME:
1903         case OBD_IOC_GETMDNAME:
1904                 RETURN(ll_get_obd_name(inode, cmd, arg));
1905         case LL_IOC_FLUSHCTX:
1906                 RETURN(ll_flush_ctx(inode));
1907         case LL_IOC_GETOBDCOUNT: {
1908                 u32 count, vallen;
1909                 struct obd_export *exp;
1910
1911                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1912                         RETURN(-EFAULT);
1913
1914                 /* get ost count when count is zero, get mdt count otherwise */
1915                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1916                 vallen = sizeof(count);
1917                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1918                                   KEY_TGT_COUNT, &vallen, &count);
1919                 if (rc) {
1920                         CERROR("get target count failed: %d\n", rc);
1921                         RETURN(rc);
1922                 }
1923
1924                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1925                         RETURN(-EFAULT);
1926
1927                 RETURN(0);
1928         }
1929         case LL_IOC_PATH2FID:
1930                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1931                                      sizeof(struct lu_fid)))
1932                         RETURN(-EFAULT);
1933                 RETURN(0);
1934         case LL_IOC_GET_CONNECT_FLAGS: {
1935                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1936                                      (void __user *)arg));
1937         }
1938         case OBD_IOC_FID2PATH:
1939                 RETURN(ll_fid2path(inode, (void __user *)arg));
1940         case LL_IOC_GETPARENT:
1941                 RETURN(ll_getparent(file, (void __user *)arg));
1942         case LL_IOC_FID2MDTIDX: {
1943                 struct obd_export *exp = ll_i2mdexp(inode);
1944                 struct lu_fid     fid;
1945                 __u32             index;
1946
1947                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1948                                    sizeof(fid)))
1949                         RETURN(-EFAULT);
1950
1951                 /* Call mdc_iocontrol */
1952                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1953                                    (__u32 __user *)&index);
1954                 if (rc != 0)
1955                         RETURN(rc);
1956
1957                 RETURN(index);
1958         }
1959         case LL_IOC_HSM_REQUEST: {
1960                 struct hsm_user_request *hur;
1961                 ssize_t                  totalsize;
1962
1963                 OBD_ALLOC_PTR(hur);
1964                 if (hur == NULL)
1965                         RETURN(-ENOMEM);
1966
1967                 /* We don't know the true size yet; copy the fixed-size part */
1968                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1969                         OBD_FREE_PTR(hur);
1970                         RETURN(-EFAULT);
1971                 }
1972
1973                 /* Compute the whole struct size */
1974                 totalsize = hur_len(hur);
1975                 OBD_FREE_PTR(hur);
1976                 if (totalsize < 0)
1977                         RETURN(-E2BIG);
1978
1979                 /* Final size will be more than double totalsize */
1980                 if (totalsize >= MDS_MAXREQSIZE / 3)
1981                         RETURN(-E2BIG);
1982
1983                 OBD_ALLOC_LARGE(hur, totalsize);
1984                 if (hur == NULL)
1985                         RETURN(-ENOMEM);
1986
1987                 /* Copy the whole struct */
1988                 if (copy_from_user(hur, (void __user *)arg, totalsize))
1989                         GOTO(out_hur, rc = -EFAULT);
1990
1991                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1992                         const struct lu_fid *fid;
1993                         struct inode *f;
1994                         int i;
1995
1996                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1997                                 fid = &hur->hur_user_item[i].hui_fid;
1998                                 f = search_inode_for_lustre(inode->i_sb, fid);
1999                                 if (IS_ERR(f)) {
2000                                         rc = PTR_ERR(f);
2001                                         break;
2002                                 }
2003
2004                                 rc = ll_hsm_release(f);
2005                                 iput(f);
2006                                 if (rc != 0)
2007                                         break;
2008                         }
2009                 } else {
2010                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
2011                                            hur, NULL);
2012                 }
2013
2014 out_hur:
2015                 OBD_FREE_LARGE(hur, totalsize);
2016
2017                 RETURN(rc);
2018         }
2019         case LL_IOC_HSM_PROGRESS: {
2020                 struct hsm_progress_kernel      hpk;
2021                 struct hsm_progress             hp;
2022
2023                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
2024                         RETURN(-EFAULT);
2025
2026                 hpk.hpk_fid = hp.hp_fid;
2027                 hpk.hpk_cookie = hp.hp_cookie;
2028                 hpk.hpk_extent = hp.hp_extent;
2029                 hpk.hpk_flags = hp.hp_flags;
2030                 hpk.hpk_errval = hp.hp_errval;
2031                 hpk.hpk_data_version = 0;
2032
2033                 /* File may not exist in Lustre; all progress
2034                  * reported to Lustre root */
2035                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
2036                                    NULL);
2037                 RETURN(rc);
2038         }
2039         case LL_IOC_HSM_CT_START:
2040                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2041                         RETURN(-EPERM);
2042
2043                 rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
2044                                        (struct lustre_kernelcomm __user *)arg);
2045                 RETURN(rc);
2046
2047         case LL_IOC_HSM_COPY_START: {
2048                 struct hsm_copy *copy;
2049                 int              rc;
2050
2051                 OBD_ALLOC_PTR(copy);
2052                 if (copy == NULL)
2053                         RETURN(-ENOMEM);
2054                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2055                         OBD_FREE_PTR(copy);
2056                         RETURN(-EFAULT);
2057                 }
2058
2059                 rc = ll_ioc_copy_start(inode->i_sb, copy);
2060                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2061                         rc = -EFAULT;
2062
2063                 OBD_FREE_PTR(copy);
2064                 RETURN(rc);
2065         }
2066         case LL_IOC_HSM_COPY_END: {
2067                 struct hsm_copy *copy;
2068                 int              rc;
2069
2070                 OBD_ALLOC_PTR(copy);
2071                 if (copy == NULL)
2072                         RETURN(-ENOMEM);
2073                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2074                         OBD_FREE_PTR(copy);
2075                         RETURN(-EFAULT);
2076                 }
2077
2078                 rc = ll_ioc_copy_end(inode->i_sb, copy);
2079                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2080                         rc = -EFAULT;
2081
2082                 OBD_FREE_PTR(copy);
2083                 RETURN(rc);
2084         }
2085         case LL_IOC_MIGRATE: {
2086                 struct lmv_user_md *lum;
2087                 char *buf = NULL;
2088                 int len;
2089                 char *filename;
2090                 int namelen = 0;
2091                 int rc;
2092
2093                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
2094                 if (rc)
2095                         RETURN(rc);
2096
2097                 data = (struct obd_ioctl_data *)buf;
2098                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
2099                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
2100                         GOTO(migrate_free, rc = -EINVAL);
2101
2102                 filename = data->ioc_inlbuf1;
2103                 namelen = data->ioc_inllen1;
2104
2105                 if (namelen < 1 || namelen != strlen(filename) + 1) {
2106                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
2107                         GOTO(migrate_free, rc = -EINVAL);
2108                 }
2109
2110                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
2111                 if (lum->lum_magic != LMV_USER_MAGIC &&
2112                     lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
2113                         rc = -EINVAL;
2114                         CERROR("%s: wrong lum magic %x: rc = %d\n",
2115                                filename, lum->lum_magic, rc);
2116                         GOTO(migrate_free, rc);
2117                 }
2118
2119                 rc = ll_migrate(inode, file, lum, filename);
2120 migrate_free:
2121                 OBD_FREE_LARGE(buf, len);
2122
2123                 RETURN(rc);
2124         }
2125         case FS_IOC_FSGETXATTR:
2126                 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
2127         case FS_IOC_FSSETXATTR:
2128                 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
2129         case LL_IOC_PCC_DETACH_BY_FID: {
2130                 struct lu_pcc_detach_fid *detach;
2131                 struct lu_fid *fid;
2132                 struct inode *inode2;
2133                 unsigned long ino;
2134
2135                 OBD_ALLOC_PTR(detach);
2136                 if (detach == NULL)
2137                         RETURN(-ENOMEM);
2138
2139                 if (copy_from_user(detach,
2140                                    (const struct lu_pcc_detach_fid __user *)arg,
2141                                    sizeof(*detach)))
2142                         GOTO(out_detach, rc = -EFAULT);
2143
2144                 fid = &detach->pccd_fid;
2145                 ino = cl_fid_build_ino(fid, ll_need_32bit_api(sbi));
2146                 inode2 = ilookup5(inode->i_sb, ino, ll_test_inode_by_fid, fid);
2147                 if (inode2 == NULL)
2148                         /* Target inode is not in inode cache, and PCC file
2149                          * has aleady released, return immdiately.
2150                          */
2151                         GOTO(out_detach, rc = 0);
2152
2153                 if (!S_ISREG(inode2->i_mode))
2154                         GOTO(out_iput, rc = -EINVAL);
2155
2156                 if (!inode_owner_or_capable(inode2))
2157                         GOTO(out_iput, rc = -EPERM);
2158
2159                 rc = pcc_ioctl_detach(inode2, detach->pccd_opt);
2160 out_iput:
2161                 iput(inode2);
2162 out_detach:
2163                 OBD_FREE_PTR(detach);
2164                 RETURN(rc);
2165         }
2166 #ifdef HAVE_LUSTRE_CRYPTO
2167         case LL_IOC_SET_ENCRYPTION_POLICY:
2168                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2169                         return -EOPNOTSUPP;
2170                 return llcrypt_ioctl_set_policy(file, (const void __user *)arg);
2171         case LL_IOC_GET_ENCRYPTION_POLICY_EX:
2172                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2173                         return -EOPNOTSUPP;
2174                 return llcrypt_ioctl_get_policy_ex(file, (void __user *)arg);
2175         case LL_IOC_ADD_ENCRYPTION_KEY:
2176                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2177                         return -EOPNOTSUPP;
2178                 return llcrypt_ioctl_add_key(file, (void __user *)arg);
2179         case LL_IOC_REMOVE_ENCRYPTION_KEY:
2180                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2181                         return -EOPNOTSUPP;
2182                 return llcrypt_ioctl_remove_key(file, (void __user *)arg);
2183         case LL_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS:
2184                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2185                         return -EOPNOTSUPP;
2186                 return llcrypt_ioctl_remove_key_all_users(file,
2187                                                           (void __user *)arg);
2188         case LL_IOC_GET_ENCRYPTION_KEY_STATUS:
2189                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2190                         return -EOPNOTSUPP;
2191                 return llcrypt_ioctl_get_key_status(file, (void __user *)arg);
2192 #endif
2193         default:
2194                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
2195                                      (void __user *)arg));
2196         }
2197 }
2198
2199 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
2200 {
2201         struct inode *inode = file->f_mapping->host;
2202         struct ll_file_data *fd = file->private_data;
2203         struct ll_sb_info *sbi = ll_i2sbi(inode);
2204         int api32 = ll_need_32bit_api(sbi);
2205         loff_t ret = -EINVAL;
2206         ENTRY;
2207
2208         inode_lock(inode);
2209         switch (origin) {
2210         case SEEK_SET:
2211                 break;
2212         case SEEK_CUR:
2213                 offset += file->f_pos;
2214                 break;
2215         case SEEK_END:
2216                 if (offset > 0)
2217                         GOTO(out, ret);
2218                 if (api32)
2219                         offset += LL_DIR_END_OFF_32BIT;
2220                 else
2221                         offset += LL_DIR_END_OFF;
2222                 break;
2223         default:
2224                 GOTO(out, ret);
2225         }
2226
2227         if (offset >= 0 &&
2228             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
2229              (!api32 && offset <= LL_DIR_END_OFF))) {
2230                 if (offset != file->f_pos) {
2231                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
2232                             (!api32 && offset == LL_DIR_END_OFF))
2233                                 fd->lfd_pos = MDS_DIR_END_OFF;
2234                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
2235                                 fd->lfd_pos = offset << 32;
2236                         else
2237                                 fd->lfd_pos = offset;
2238                         file->f_pos = offset;
2239                         file->f_version = 0;
2240                 }
2241                 ret = offset;
2242         }
2243         GOTO(out, ret);
2244
2245 out:
2246         inode_unlock(inode);
2247         return ret;
2248 }
2249
2250 static int ll_dir_open(struct inode *inode, struct file *file)
2251 {
2252         ENTRY;
2253         RETURN(ll_file_open(inode, file));
2254 }
2255
2256 static int ll_dir_release(struct inode *inode, struct file *file)
2257 {
2258         ENTRY;
2259         RETURN(ll_file_release(inode, file));
2260 }
2261
2262 const struct file_operations ll_dir_operations = {
2263         .llseek         = ll_dir_seek,
2264         .open           = ll_dir_open,
2265         .release        = ll_dir_release,
2266         .read           = generic_read_dir,
2267 #ifdef HAVE_DIR_CONTEXT
2268         .iterate_shared = ll_iterate,
2269 #else
2270         .readdir        = ll_readdir,
2271 #endif
2272         .unlocked_ioctl = ll_dir_ioctl,
2273         .fsync          = ll_fsync,
2274 };