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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/dir.c
  37  *
  38  * Directory code for lustre client.
  39  */
  40
  41 #include <linux/fs.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <linux/version.h>
  45 #include <asm/uaccess.h>
  46 #include <linux/buffer_head.h>   // for wait_on_buffer
  47 #include <linux/pagevec.h>
  48
  49 #define DEBUG_SUBSYSTEM S_LLITE
  50
  51 #include <lustre/lustre_idl.h>
  52 #include <obd_support.h>
  53 #include <obd_class.h>
  54 #include <lustre_lib.h>
  55 #include <lustre/lustre_idl.h>
  56 #include <lustre_lite.h>
  57 #include <lustre_dlm.h>
  58 #include <lustre_fid.h>
  59 #include "llite_internal.h"
  60
  61 /*
  62  * (new) readdir implementation overview.
  63  *
  64  * Original lustre readdir implementation cached exact copy of raw directory
  65  * pages on the client. These pages were indexed in client page cache by
  66  * logical offset in the directory file. This design, while very simple and
  67  * intuitive had some inherent problems:
  68  *
  69  *     . it implies that byte offset to the directory entry serves as a
  70  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  71  *     ext3/htree directory entries may move due to splits, and more
  72  *     importantly,
  73  *
  74  *     . it is incompatible with the design of split directories for cmd3,
  75  *     that assumes that names are distributed across nodes based on their
  76  *     hash, and so readdir should be done in hash order.
  77  *
  78  * New readdir implementation does readdir in hash order, and uses hash of a
  79  * file name as a telldir/seekdir cookie. This led to number of complications:
  80  *
  81  *     . hash is not unique, so it cannot be used to index cached directory
  82  *     pages on the client (note, that it requires a whole pageful of hash
  83  *     collided entries to cause two pages to have identical hashes);
  84  *
  85  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  86  *     entry cookie. ext3/htree has the same problem and lustre implementation
  87  *     mimics their solution: seekdir(hash) positions directory at the first
  88  *     entry with the given hash.
  89  *
  90  * Client side.
  91  *
  92  * 0. caching
  93  *
  94  * Client caches directory pages using hash of the first entry as an index. As
  95  * noted above hash is not unique, so this solution doesn't work as is:
  96  * special processing is needed for "page hash chains" (i.e., sequences of
  97  * pages filled with entries all having the same hash value).
  98  *
  99  * First, such chains have to be detected. To this end, server returns to the
 100  * client the hash of the first entry on the page next to one returned. When
 101  * client detects that this hash is the same as hash of the first entry on the
 102  * returned page, page hash collision has to be handled. Pages in the
 103  * hash chain, except first one, are termed "overflow pages".
 104  *
 105  * Solution to index uniqueness problem is to not cache overflow
 106  * pages. Instead, when page hash collision is detected, all overflow pages
 107  * from emerging chain are immediately requested from the server and placed in
 108  * a special data structure (struct ll_dir_chain). This data structure is used
 109  * by ll_readdir() to process entries from overflow pages. When readdir
 110  * invocation finishes, overflow pages are discarded. If page hash collision
 111  * chain weren't completely processed, next call to readdir will again detect
 112  * page hash collision, again read overflow pages in, process next portion of
 113  * entries and again discard the pages. This is not as wasteful as it looks,
 114  * because, given reasonable hash, page hash collisions are extremely rare.
 115  *
 116  * 1. directory positioning
 117  *
 118  * When seekdir(hash) is called, original
 119  *
 120  *
 121  *
 122  *
 123  *
 124  *
 125  *
 126  *
 127  * Server.
 128  *
 129  * identification of and access to overflow pages
 130  *
 131  * page format
 132  *
 133  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 134  * a header lu_dirpage which describes the start/end hash, and whether this
 135  * page is empty (contains no dir entry) or hash collide with next page.
 136  * After client receives reply, several pages will be integrated into dir page
 137  * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 138  * lu_dirpage for this integrated page will be adjusted. See
 139  * lmv_adjust_dirpages().
 140  *
 141  */
 142 /**
 143  * The following three APIs will be used by llite to iterate directory
 144  * entries from MDC dir page caches.
 145  *
 146  * ll_dir_entry_start(next) will lookup(return) entry by op_hash_offset.
 147  * To avoid extra memory allocation, the @entry will be pointed to
 148  * the dir entries in MDC page directly, so these pages can not be released
 149  * until the entry has been accessed in ll_readdir(or statahead).
 150  *
 151  * The iterate process will be
 152  *
 153  * ll_dir_entry_start: locate the page in MDC, and return the first entry.
 154  *                     hold the page.
 155  *
 156  * ll_dir_entry_next: return the next entry in the current page, if it reaches
 157  *                    to the end, release current page.
 158  *
 159  * ll_dir_entry_end: release the last page.
 160  **/
 161 struct lu_dirent *ll_dir_entry_start(struct inode *dir,
 162                                      struct md_op_data *op_data,
 163                                      struct page **ppage)
 164 {
 165         struct lu_dirent *entry;
 166         struct md_callback cb_op;
 167         int rc;
 168
 169         LASSERT(*ppage == NULL);
 170         cb_op.md_blocking_ast = ll_md_blocking_ast;
 171         rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
 172         if (rc != 0)
 173                 entry = ERR_PTR(rc);
 174         return entry;
 175 }
 176
 177 struct lu_dirent *ll_dir_entry_next(struct inode *dir,
 178                                     struct md_op_data *op_data,
 179                                     struct lu_dirent *ent,
 180                                     struct page **ppage)
 181 {
 182         struct lu_dirent *entry;
 183         struct md_callback cb_op;
 184         int rc;
 185
 186         LASSERT(*ppage != NULL);
 187         cb_op.md_blocking_ast = ll_md_blocking_ast;
 188         op_data->op_hash_offset = le64_to_cpu(ent->lde_hash);
 189         kunmap(*ppage);
 190         page_cache_release(*ppage);
 191         *ppage = NULL;
 192         rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
 193         if (rc != 0)
 194                 entry = ERR_PTR(rc);
 195         return entry;
 196 }
 197
 198 int ll_dir_read(struct inode *inode, struct md_op_data *op_data,
 199                 void *cookie, filldir_t filldir)
 200 {
 201         struct ll_sb_info       *sbi = ll_i2sbi(inode);
 202         struct ll_dir_chain     chain;
 203         struct lu_dirent        *ent;
 204         int                     api32 = ll_need_32bit_api(sbi);
 205         int                     hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 206         int                     done = 0;
 207         int                     rc = 0;
 208         __u64                   hash = MDS_DIR_END_OFF;
 209         __u64                   last_hash = MDS_DIR_END_OFF;
 210         struct page             *page = NULL;
 211         ENTRY;
 212
 213         ll_dir_chain_init(&chain);
 214         for (ent = ll_dir_entry_start(inode, op_data, &page);
 215              ent != NULL && !IS_ERR(ent) && !done;
 216              ent = ll_dir_entry_next(inode, op_data, ent, &page)) {
 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 < op_data->op_hash_offset)
 225                         /*
 226                          * Skip until we find target hash
 227                          * value.
 228                          */
 229                         continue;
 230                 namelen = le16_to_cpu(ent->lde_namelen);
 231                 if (namelen == 0)
 232                         /*
 233                          * Skip dummy record.
 234                          */
 235                         continue;
 236
 237                 if (api32 && hash64)
 238                         lhash = hash >> 32;
 239                 else
 240                         lhash = hash;
 241                 fid_le_to_cpu(&fid, &ent->lde_fid);
 242                 ino = cl_fid_build_ino(&fid, api32);
 243                 type = ll_dirent_type_get(ent);
 244                 /* For 'll_nfs_get_name_filldir()', it will try
 245                  * to access the 'ent' through its 'lde_name',
 246                  * so the parameter 'name' for 'filldir()' must
 247                  * be part of the 'ent'. */
 248                 done = filldir(cookie, ent->lde_name, namelen, lhash,
 249                                ino, type);
 250                 if (done) {
 251                         if (op_data->op_hash_offset != MDS_DIR_END_OFF)
 252                                 op_data->op_hash_offset = last_hash;
 253                         break;
 254                 } else {
 255                         last_hash = hash;
 256                 }
 257         }
 258
 259         if (IS_ERR(ent))
 260                 rc = PTR_ERR(ent);
 261
 262         if (page != NULL) {
 263                 kunmap(page);
 264                 page_cache_release(page);
 265         }
 266
 267         ll_dir_chain_fini(&chain);
 268         RETURN(rc);
 269 }
 270
 271 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 272 {
 273         struct inode            *inode  = filp->f_dentry->d_inode;
 274         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 275         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 276         __u64                   pos     = lfd->lfd_pos;
 277         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 278         int                     api32   = ll_need_32bit_api(sbi);
 279         struct md_op_data       *op_data;
 280         int                     rc;
 281 #ifdef HAVE_TOUCH_ATIME_1ARG
 282         struct path             path;
 283 #endif
 284         ENTRY;
 285
 286         if (lfd != NULL)
 287                 pos = lfd->lfd_pos;
 288         else
 289                 pos = 0;
 290
 291         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 292                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 293                inode, (unsigned long)pos, i_size_read(inode), api32);
 294
 295         if (pos == MDS_DIR_END_OFF)
 296                 /*
 297                  * end-of-file.
 298                  */
 299                 GOTO(out, rc = 0);
 300
 301         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 302                                      LUSTRE_OPC_ANY, inode);
 303         if (IS_ERR(op_data))
 304                 GOTO(out, rc = PTR_ERR(op_data));
 305
 306         op_data->op_hash_offset = pos;
 307         op_data->op_max_pages = sbi->ll_md_brw_size >> PAGE_CACHE_SHIFT;
 308         rc = ll_dir_read(inode, op_data, cookie, filldir);
 309         if (lfd != NULL)
 310                 lfd->lfd_pos = op_data->op_hash_offset;
 311
 312         if (pos == MDS_DIR_END_OFF) {
 313                 if (api32)
 314                         filp->f_pos = LL_DIR_END_OFF_32BIT;
 315                 else
 316                         filp->f_pos = LL_DIR_END_OFF;
 317         } else {
 318                 if (api32 && hash64)
 319                         filp->f_pos = op_data->op_hash_offset >> 32;
 320                 else
 321                         filp->f_pos = op_data->op_hash_offset;
 322         }
 323
 324         ll_finish_md_op_data(op_data);
 325         filp->f_version = inode->i_version;
 326 #ifdef HAVE_TOUCH_ATIME_1ARG
 327 #ifdef HAVE_F_PATH_MNT
 328         path.mnt = filp->f_path.mnt;
 329 #else
 330         path.mnt = filp->f_vfsmnt;
 331 #endif
 332         path.dentry = filp->f_dentry;
 333         touch_atime(&path);
 334 #else
 335         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 336 #endif
 337
 338 out:
 339         if (!rc)
 340                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 341
 342         RETURN(rc);
 343 }
 344
 345 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 346 {
 347         struct mgs_send_param *msp;
 348         int rc = 0;
 349
 350         OBD_ALLOC_PTR(msp);
 351         if (!msp)
 352                 return -ENOMEM;
 353
 354         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 355         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 356                                 sizeof(struct mgs_send_param), msp, NULL);
 357         if (rc)
 358                 CERROR("Failed to set parameter: %d\n", rc);
 359         OBD_FREE_PTR(msp);
 360
 361         return rc;
 362 }
 363
 364 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 365                                const char *filename)
 366 {
 367         struct ptlrpc_request *request = NULL;
 368         struct md_op_data *op_data;
 369         struct ll_sb_info *sbi = ll_i2sbi(dir);
 370         int mode;
 371         int err;
 372         ENTRY;
 373
 374         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 375                 RETURN(-EINVAL);
 376
 377         if (lump->lum_stripe_offset == (__u32)-1) {
 378                 int mdtidx;
 379
 380                 mdtidx = ll_get_mdt_idx(dir);
 381                 if (mdtidx < 0)
 382                         RETURN(mdtidx);
 383
 384                 lump->lum_stripe_offset = mdtidx;
 385         }
 386
 387         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
 388                "stripe_offset %d, stripe_count: %u\n",
 389                PFID(ll_inode2fid(dir)), dir, filename,
 390                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 391
 392         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 393                 lustre_swab_lmv_user_md(lump);
 394
 395         mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
 396         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 397                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 398                                      lump);
 399         if (IS_ERR(op_data))
 400                 GOTO(err_exit, err = PTR_ERR(op_data));
 401
 402         op_data->op_cli_flags |= CLI_SET_MEA;
 403         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 404                         current_fsuid(), current_fsgid(),
 405                         cfs_curproc_cap_pack(), 0, &request);
 406         ll_finish_md_op_data(op_data);
 407         if (err)
 408                 GOTO(err_exit, err);
 409 err_exit:
 410         ptlrpc_req_finished(request);
 411         return err;
 412 }
 413
 414 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 415                      int set_default)
 416 {
 417         struct ll_sb_info *sbi = ll_i2sbi(inode);
 418         struct md_op_data *op_data;
 419         struct ptlrpc_request *req = NULL;
 420         int rc = 0;
 421         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 422         struct obd_device *mgc = lsi->lsi_mgc;
 423         int lum_size;
 424         ENTRY;
 425
 426         if (lump != NULL) {
 427                 /*
 428                  * This is coming from userspace, so should be in
 429                  * local endian.  But the MDS would like it in little
 430                  * endian, so we swab it before we send it.
 431                  */
 432                 switch (lump->lmm_magic) {
 433                 case LOV_USER_MAGIC_V1: {
 434                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 435                                 lustre_swab_lov_user_md_v1(lump);
 436                         lum_size = sizeof(struct lov_user_md_v1);
 437                         break;
 438                 }
 439                 case LOV_USER_MAGIC_V3: {
 440                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 441                                 lustre_swab_lov_user_md_v3(
 442                                         (struct lov_user_md_v3 *)lump);
 443                         lum_size = sizeof(struct lov_user_md_v3);
 444                         break;
 445                 }
 446                 default: {
 447                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 448                                         " %#08x != %#08x nor %#08x\n",
 449                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 450                                         LOV_USER_MAGIC_V3);
 451                         RETURN(-EINVAL);
 452                 }
 453                 }
 454         } else {
 455                 lum_size = sizeof(struct lov_user_md_v1);
 456         }
 457
 458         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 459                                      LUSTRE_OPC_ANY, NULL);
 460         if (IS_ERR(op_data))
 461                 RETURN(PTR_ERR(op_data));
 462
 463         /* swabbing is done in lov_setstripe() on server side */
 464         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 465                         NULL, 0, &req, NULL);
 466         ll_finish_md_op_data(op_data);
 467         ptlrpc_req_finished(req);
 468         if (rc) {
 469                 if (rc != -EPERM && rc != -EACCES)
 470                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 471         }
 472
 473         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 474          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 475          need the make the distiction between the 2 versions */
 476         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 477                 char *param = NULL;
 478                 char *buf;
 479
 480                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 481                 if (param == NULL)
 482                         GOTO(end, rc = -ENOMEM);
 483
 484                 buf = param;
 485                 /* Get fsname and assume devname to be -MDT0000. */
 486                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 487                 strcat(buf, "-MDT0000.lov");
 488                 buf += strlen(buf);
 489
 490                 /* Set root stripesize */
 491                 sprintf(buf, ".stripesize=%u",
 492                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 493                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 494                 if (rc)
 495                         GOTO(end, rc);
 496
 497                 /* Set root stripecount */
 498                 sprintf(buf, ".stripecount=%hd",
 499                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 500                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 501                 if (rc)
 502                         GOTO(end, rc);
 503
 504                 /* Set root stripeoffset */
 505                 sprintf(buf, ".stripeoffset=%hd",
 506                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 507                         (typeof(lump->lmm_stripe_offset))(-1));
 508                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 509
 510 end:
 511                 if (param != NULL)
 512                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 513         }
 514         RETURN(rc);
 515 }
 516
 517 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 518                      int *lmm_size, struct ptlrpc_request **request)
 519 {
 520         struct ll_sb_info *sbi = ll_i2sbi(inode);
 521         struct mdt_body   *body;
 522         struct lov_mds_md *lmm = NULL;
 523         struct ptlrpc_request *req = NULL;
 524         int rc, lmmsize;
 525         struct md_op_data *op_data;
 526
 527         rc = ll_get_max_mdsize(sbi, &lmmsize);
 528         if (rc)
 529                 RETURN(rc);
 530
 531         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 532                                      0, lmmsize, LUSTRE_OPC_ANY,
 533                                      NULL);
 534         if (IS_ERR(op_data))
 535                 RETURN(PTR_ERR(op_data));
 536
 537         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 538         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 539         ll_finish_md_op_data(op_data);
 540         if (rc < 0) {
 541                 CDEBUG(D_INFO, "md_getattr failed on inode "
 542                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 543                 GOTO(out, rc);
 544         }
 545
 546         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 547         LASSERT(body != NULL);
 548
 549         lmmsize = body->eadatasize;
 550
 551         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 552             lmmsize == 0) {
 553                 GOTO(out, rc = -ENODATA);
 554         }
 555
 556         lmm = req_capsule_server_sized_get(&req->rq_pill,
 557                                            &RMF_MDT_MD, lmmsize);
 558         LASSERT(lmm != NULL);
 559
 560         /*
 561          * This is coming from the MDS, so is probably in
 562          * little endian.  We convert it to host endian before
 563          * passing it to userspace.
 564          */
 565         /* We don't swab objects for directories */
 566         switch (le32_to_cpu(lmm->lmm_magic)) {
 567         case LOV_MAGIC_V1:
 568                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 569                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 570                 break;
 571         case LOV_MAGIC_V3:
 572                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 573                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 574                 break;
 575         default:
 576                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 577                 rc = -EPROTO;
 578         }
 579 out:
 580         *lmmp = lmm;
 581         *lmm_size = lmmsize;
 582         *request = req;
 583         return rc;
 584 }
 585
 586 /*
 587  *  Get MDT index for the inode.
 588  */
 589 int ll_get_mdt_idx(struct inode *inode)
 590 {
 591         struct ll_sb_info *sbi = ll_i2sbi(inode);
 592         struct md_op_data *op_data;
 593         int rc, mdtidx;
 594         ENTRY;
 595
 596         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 597                                      0, LUSTRE_OPC_ANY, NULL);
 598         if (IS_ERR(op_data))
 599                 RETURN(PTR_ERR(op_data));
 600
 601         op_data->op_flags |= MF_GET_MDT_IDX;
 602         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 603         mdtidx = op_data->op_mds;
 604         ll_finish_md_op_data(op_data);
 605         if (rc < 0) {
 606                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 607                 RETURN(rc);
 608         }
 609         return mdtidx;
 610 }
 611
 612 /**
 613  * Generic handler to do any pre-copy work.
 614  *
 615  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 616  * first information for it that real work has started.
 617  *
 618  * Moreover, for a ARCHIVE request, it will sample the file data version and
 619  * store it in \a copy.
 620  *
 621  * \return 0 on success.
 622  */
 623 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 624 {
 625         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 626         struct hsm_progress_kernel       hpk;
 627         int                              rc;
 628         ENTRY;
 629
 630         /* Forge a hsm_progress based on data from copy. */
 631         hpk.hpk_fid = copy->hc_hai.hai_fid;
 632         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 633         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 634         hpk.hpk_extent.length = 0;
 635         hpk.hpk_flags = 0;
 636         hpk.hpk_errval = 0;
 637         hpk.hpk_data_version = 0;
 638
 639
 640         /* For archive request, we need to read the current file version. */
 641         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 642                 struct inode    *inode;
 643                 __u64            data_version = 0;
 644
 645                 /* Get inode for this fid */
 646                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 647                 if (IS_ERR(inode)) {
 648                         hpk.hpk_flags |= HP_FLAG_RETRY;
 649                         /* hpk_errval is >= 0 */
 650                         hpk.hpk_errval = -PTR_ERR(inode);
 651                         GOTO(progress, rc = PTR_ERR(inode));
 652                 }
 653
 654                 /* Read current file data version */
 655                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 656                 iput(inode);
 657                 if (rc != 0) {
 658                         CDEBUG(D_HSM, "Could not read file data version of "
 659                                       DFID" (rc = %d). Archive request ("
 660                                       LPX64") could not be done.\n",
 661                                       PFID(&copy->hc_hai.hai_fid), rc,
 662                                       copy->hc_hai.hai_cookie);
 663                         hpk.hpk_flags |= HP_FLAG_RETRY;
 664                         /* hpk_errval must be >= 0 */
 665                         hpk.hpk_errval = -rc;
 666                         GOTO(progress, rc);
 667                 }
 668
 669                 /* Store it the hsm_copy for later copytool use.
 670                  * Always modified even if no lsm. */
 671                 copy->hc_data_version = data_version;
 672         }
 673
 674 progress:
 675         /* On error, the request should be considered as completed */
 676         if (hpk.hpk_errval > 0)
 677                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 678         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 679                            &hpk, NULL);
 680
 681         RETURN(rc);
 682 }
 683
 684 /**
 685  * Generic handler to do any post-copy work.
 686  *
 687  * It will send the last hsm_progress update to coordinator to inform it
 688  * that copy is finished and whether it was successful or not.
 689  *
 690  * Moreover,
 691  * - for ARCHIVE request, it will sample the file data version and compare it
 692  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 693  *   will be considered as failed.
 694  * - for RESTORE request, it will sample the file data version and send it to
 695  *   coordinator which is useful if the file was imported as 'released'.
 696  *
 697  * \return 0 on success.
 698  */
 699 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 700 {
 701         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 702         struct hsm_progress_kernel       hpk;
 703         int                              rc;
 704         ENTRY;
 705
 706         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 707         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 708          * initialized if copy_end was called with copy == NULL.
 709          */
 710
 711         /* Forge a hsm_progress based on data from copy. */
 712         hpk.hpk_fid = copy->hc_hai.hai_fid;
 713         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 714         hpk.hpk_extent = copy->hc_hai.hai_extent;
 715         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 716         hpk.hpk_errval = copy->hc_errval;
 717         hpk.hpk_data_version = 0;
 718
 719         /* For archive request, we need to check the file data was not changed.
 720          *
 721          * For restore request, we need to send the file data version, this is
 722          * useful when the file was created using hsm_import.
 723          */
 724         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 725              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 726             (copy->hc_errval == 0)) {
 727                 struct inode    *inode;
 728                 __u64            data_version = 0;
 729
 730                 /* Get lsm for this fid */
 731                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 732                 if (IS_ERR(inode)) {
 733                         hpk.hpk_flags |= HP_FLAG_RETRY;
 734                         /* hpk_errval must be >= 0 */
 735                         hpk.hpk_errval = -PTR_ERR(inode);
 736                         GOTO(progress, rc = PTR_ERR(inode));
 737                 }
 738
 739                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 740                 iput(inode);
 741                 if (rc) {
 742                         CDEBUG(D_HSM, "Could not read file data version. "
 743                                       "Request could not be confirmed.\n");
 744                         if (hpk.hpk_errval == 0)
 745                                 hpk.hpk_errval = -rc;
 746                         GOTO(progress, rc);
 747                 }
 748
 749                 /* Store it the hsm_copy for later copytool use.
 750                  * Always modified even if no lsm. */
 751                 hpk.hpk_data_version = data_version;
 752
 753                 /* File could have been stripped during archiving, so we need
 754                  * to check anyway. */
 755                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 756                     (copy->hc_data_version != data_version)) {
 757                         CDEBUG(D_HSM, "File data version mismatched. "
 758                               "File content was changed during archiving. "
 759                                DFID", start:"LPX64" current:"LPX64"\n",
 760                                PFID(&copy->hc_hai.hai_fid),
 761                                copy->hc_data_version, data_version);
 762                         /* File was changed, send error to cdt. Do not ask for
 763                          * retry because if a file is modified frequently,
 764                          * the cdt will loop on retried archive requests.
 765                          * The policy engine will ask for a new archive later
 766                          * when the file will not be modified for some tunable
 767                          * time */
 768                         /* we do not notify caller */
 769                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 770                         /* hpk_errval must be >= 0 */
 771                         hpk.hpk_errval = EBUSY;
 772                 }
 773
 774         }
 775
 776 progress:
 777         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 778                            &hpk, NULL);
 779
 780         RETURN(rc);
 781 }
 782
 783
 784 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 785                           const void __user *data, size_t size)
 786 {
 787         void *copy;
 788         int rc;
 789
 790         OBD_ALLOC(copy, size);
 791         if (copy == NULL)
 792                 return -ENOMEM;
 793
 794         if (copy_from_user(copy, data, size)) {
 795                 rc = -EFAULT;
 796                 goto out;
 797         }
 798
 799         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 800 out:
 801         OBD_FREE(copy, size);
 802
 803         return rc;
 804 }
 805
 806 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 807 {
 808         int cmd = qctl->qc_cmd;
 809         int type = qctl->qc_type;
 810         int id = qctl->qc_id;
 811         int valid = qctl->qc_valid;
 812         int rc = 0;
 813         ENTRY;
 814
 815         switch (cmd) {
 816         case LUSTRE_Q_INVALIDATE:
 817         case LUSTRE_Q_FINVALIDATE:
 818         case Q_QUOTAON:
 819         case Q_QUOTAOFF:
 820         case Q_SETQUOTA:
 821         case Q_SETINFO:
 822                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 823                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 824                         RETURN(-EPERM);
 825                 break;
 826         case Q_GETQUOTA:
 827                 if (((type == USRQUOTA && current_euid() != id) ||
 828                      (type == GRPQUOTA && !in_egroup_p(id))) &&
 829                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 830                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 831                         RETURN(-EPERM);
 832                 break;
 833         case Q_GETINFO:
 834                 break;
 835         default:
 836                 CERROR("unsupported quotactl op: %#x\n", cmd);
 837                 RETURN(-ENOTTY);
 838         }
 839
 840         if (valid != QC_GENERAL) {
 841                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 842                         RETURN(-EOPNOTSUPP);
 843
 844                 if (cmd == Q_GETINFO)
 845                         qctl->qc_cmd = Q_GETOINFO;
 846                 else if (cmd == Q_GETQUOTA)
 847                         qctl->qc_cmd = Q_GETOQUOTA;
 848                 else
 849                         RETURN(-EINVAL);
 850
 851                 switch (valid) {
 852                 case QC_MDTIDX:
 853                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 854                                            sizeof(*qctl), qctl, NULL);
 855                         break;
 856                 case QC_OSTIDX:
 857                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 858                                            sizeof(*qctl), qctl, NULL);
 859                         break;
 860                 case QC_UUID:
 861                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 862                                            sizeof(*qctl), qctl, NULL);
 863                         if (rc == -EAGAIN)
 864                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 865                                                    sbi->ll_dt_exp,
 866                                                    sizeof(*qctl), qctl, NULL);
 867                         break;
 868                 default:
 869                         rc = -EINVAL;
 870                         break;
 871                 }
 872
 873                 if (rc)
 874                         RETURN(rc);
 875
 876                 qctl->qc_cmd = cmd;
 877         } else {
 878                 struct obd_quotactl *oqctl;
 879
 880                 OBD_ALLOC_PTR(oqctl);
 881                 if (oqctl == NULL)
 882                         RETURN(-ENOMEM);
 883
 884                 QCTL_COPY(oqctl, qctl);
 885                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
 886                 if (rc) {
 887                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
 888                                 oqctl->qc_cmd = Q_QUOTAOFF;
 889                                 obd_quotactl(sbi->ll_md_exp, oqctl);
 890                         }
 891                         OBD_FREE_PTR(oqctl);
 892                         RETURN(rc);
 893                 }
 894                 /* If QIF_SPACE is not set, client should collect the
 895                  * space usage from OSSs by itself */
 896                 if (cmd == Q_GETQUOTA &&
 897                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
 898                     !oqctl->qc_dqblk.dqb_curspace) {
 899                         struct obd_quotactl *oqctl_tmp;
 900
 901                         OBD_ALLOC_PTR(oqctl_tmp);
 902                         if (oqctl_tmp == NULL)
 903                                 GOTO(out, rc = -ENOMEM);
 904
 905                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
 906                         oqctl_tmp->qc_id = oqctl->qc_id;
 907                         oqctl_tmp->qc_type = oqctl->qc_type;
 908
 909                         /* collect space usage from OSTs */
 910                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 911                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
 912                         if (!rc || rc == -EREMOTEIO) {
 913                                 oqctl->qc_dqblk.dqb_curspace =
 914                                         oqctl_tmp->qc_dqblk.dqb_curspace;
 915                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
 916                         }
 917
 918                         /* collect space & inode usage from MDTs */
 919                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 920                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
 921                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
 922                         if (!rc || rc == -EREMOTEIO) {
 923                                 oqctl->qc_dqblk.dqb_curspace +=
 924                                         oqctl_tmp->qc_dqblk.dqb_curspace;
 925                                 oqctl->qc_dqblk.dqb_curinodes =
 926                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
 927                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
 928                         } else {
 929                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
 930                         }
 931
 932                         OBD_FREE_PTR(oqctl_tmp);
 933                 }
 934 out:
 935                 QCTL_COPY(qctl, oqctl);
 936                 OBD_FREE_PTR(oqctl);
 937         }
 938
 939         RETURN(rc);
 940 }
 941
 942 static char *
 943 ll_getname(const char __user *filename)
 944 {
 945         int ret = 0, len;
 946         char *tmp = __getname();
 947
 948         if (!tmp)
 949                 return ERR_PTR(-ENOMEM);
 950
 951         len = strncpy_from_user(tmp, filename, PATH_MAX);
 952         if (len == 0)
 953                 ret = -ENOENT;
 954         else if (len > PATH_MAX)
 955                 ret = -ENAMETOOLONG;
 956
 957         if (ret) {
 958                 __putname(tmp);
 959                 tmp =  ERR_PTR(ret);
 960         }
 961         return tmp;
 962 }
 963
 964 #define ll_putname(filename) __putname(filename)
 965
 966 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 967 {
 968         struct inode *inode = file->f_dentry->d_inode;
 969         struct ll_sb_info *sbi = ll_i2sbi(inode);
 970         struct obd_ioctl_data *data;
 971         int rc = 0;
 972         ENTRY;
 973
 974         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
 975                PFID(ll_inode2fid(inode)), inode, cmd);
 976
 977         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
 978         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
 979                 return -ENOTTY;
 980
 981         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
 982         switch(cmd) {
 983         case FSFILT_IOC_GETFLAGS:
 984         case FSFILT_IOC_SETFLAGS:
 985                 RETURN(ll_iocontrol(inode, file, cmd, arg));
 986         case FSFILT_IOC_GETVERSION_OLD:
 987         case FSFILT_IOC_GETVERSION:
 988                 RETURN(put_user(inode->i_generation, (int *)arg));
 989         /* We need to special case any other ioctls we want to handle,
 990          * to send them to the MDS/OST as appropriate and to properly
 991          * network encode the arg field.
 992         case FSFILT_IOC_SETVERSION_OLD:
 993         case FSFILT_IOC_SETVERSION:
 994         */
 995         case LL_IOC_GET_MDTIDX: {
 996                 int mdtidx;
 997
 998                 mdtidx = ll_get_mdt_idx(inode);
 999                 if (mdtidx < 0)
1000                         RETURN(mdtidx);
1001
1002                 if (put_user((int)mdtidx, (int*)arg))
1003                         RETURN(-EFAULT);
1004
1005                 return 0;
1006         }
1007         case IOC_MDC_LOOKUP: {
1008                 struct ptlrpc_request *request = NULL;
1009                 int namelen, len = 0;
1010                 char *buf = NULL;
1011                 char *filename;
1012                 struct md_op_data *op_data;
1013
1014                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1015                 if (rc)
1016                         RETURN(rc);
1017                 data = (void *)buf;
1018
1019                 filename = data->ioc_inlbuf1;
1020                 namelen = strlen(filename);
1021
1022                 if (namelen < 1) {
1023                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1024                         GOTO(out_free, rc = -EINVAL);
1025                 }
1026
1027                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1028                                              0, LUSTRE_OPC_ANY, NULL);
1029                 if (IS_ERR(op_data))
1030                         GOTO(out_free, rc = PTR_ERR(op_data));
1031
1032                 op_data->op_valid = OBD_MD_FLID;
1033                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1034                 ll_finish_md_op_data(op_data);
1035                 if (rc < 0) {
1036                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1037                         GOTO(out_free, rc);
1038                 }
1039                 ptlrpc_req_finished(request);
1040                 EXIT;
1041 out_free:
1042                 obd_ioctl_freedata(buf, len);
1043                 return rc;
1044         }
1045         case LL_IOC_LMV_SETSTRIPE: {
1046                 struct lmv_user_md  *lum;
1047                 char            *buf = NULL;
1048                 char            *filename;
1049                 int              namelen = 0;
1050                 int              lumlen = 0;
1051                 int              len;
1052                 int              rc;
1053
1054                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1055                 if (rc)
1056                         RETURN(rc);
1057
1058                 data = (void *)buf;
1059                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1060                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1061                         GOTO(lmv_out_free, rc = -EINVAL);
1062
1063                 filename = data->ioc_inlbuf1;
1064                 namelen = data->ioc_inllen1;
1065
1066                 if (namelen < 1) {
1067                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1068                         GOTO(lmv_out_free, rc = -EINVAL);
1069                 }
1070                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1071                 lumlen = data->ioc_inllen2;
1072
1073                 if (lum->lum_magic != LMV_USER_MAGIC ||
1074                     lumlen != sizeof(*lum)) {
1075                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1076                                filename, lum->lum_magic, lumlen, -EFAULT);
1077                         GOTO(lmv_out_free, rc = -EINVAL);
1078                 }
1079
1080                 /**
1081                  * ll_dir_setdirstripe will be used to set dir stripe
1082                  *  mdc_create--->mdt_reint_create (with dirstripe)
1083                  */
1084                 rc = ll_dir_setdirstripe(inode, lum, filename);
1085 lmv_out_free:
1086                 obd_ioctl_freedata(buf, len);
1087                 RETURN(rc);
1088
1089         }
1090         case LL_IOC_LOV_SETSTRIPE: {
1091                 struct lov_user_md_v3 lumv3;
1092                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1093                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1094                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1095
1096                 int set_default = 0;
1097
1098                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1099                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1100                         sizeof(lumv3p->lmm_objects[0]));
1101                 /* first try with v1 which is smaller than v3 */
1102                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1103                         RETURN(-EFAULT);
1104
1105                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1106                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1107                                 RETURN(-EFAULT);
1108                 }
1109
1110                 if (inode->i_sb->s_root == file->f_dentry)
1111                         set_default = 1;
1112
1113                 /* in v1 and v3 cases lumv1 points to data */
1114                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1115
1116                 RETURN(rc);
1117         }
1118         case LL_IOC_LMV_GETSTRIPE: {
1119                 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1120                 struct lmv_user_md lum;
1121                 struct lmv_user_md *tmp;
1122                 int lum_size;
1123                 int rc = 0;
1124                 int mdtindex;
1125
1126                 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1127                         RETURN(-EFAULT);
1128
1129                 if (lum.lum_magic != LMV_MAGIC_V1)
1130                         RETURN(-EINVAL);
1131
1132                 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1133                 OBD_ALLOC(tmp, lum_size);
1134                 if (tmp == NULL)
1135                         GOTO(free_lmv, rc = -ENOMEM);
1136
1137                 memcpy(tmp, &lum, sizeof(lum));
1138                 tmp->lum_stripe_count = 1;
1139                 mdtindex = ll_get_mdt_idx(inode);
1140                 if (mdtindex < 0)
1141                         GOTO(free_lmv, rc = -ENOMEM);
1142
1143                 tmp->lum_stripe_offset = mdtindex;
1144                 tmp->lum_objects[0].lum_mds = mdtindex;
1145                 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1146                        sizeof(struct lu_fid));
1147                 if (copy_to_user((void *)arg, tmp, lum_size))
1148                         GOTO(free_lmv, rc = -EFAULT);
1149 free_lmv:
1150                 if (tmp)
1151                         OBD_FREE(tmp, lum_size);
1152                 RETURN(rc);
1153         }
1154         case LL_IOC_REMOVE_ENTRY: {
1155                 char            *filename = NULL;
1156                 int              namelen = 0;
1157                 int              rc;
1158
1159                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1160                  * unsupported server, which might crash the server(LU-2730),
1161                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1162                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1163                  * server will support REINT_RMENTRY XXX*/
1164                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1165                         RETURN(-ENOTSUPP);
1166
1167                 filename = ll_getname((const char *)arg);
1168                 if (IS_ERR(filename))
1169                         RETURN(PTR_ERR(filename));
1170
1171                 namelen = strlen(filename);
1172                 if (namelen < 1)
1173                         GOTO(out_rmdir, rc = -EINVAL);
1174
1175                 rc = ll_rmdir_entry(inode, filename, namelen);
1176 out_rmdir:
1177                 if (filename)
1178                         ll_putname(filename);
1179                 RETURN(rc);
1180         }
1181         case LL_IOC_LOV_SWAP_LAYOUTS:
1182                 RETURN(-EPERM);
1183         case LL_IOC_OBD_STATFS:
1184                 RETURN(ll_obd_statfs(inode, (void *)arg));
1185         case LL_IOC_LOV_GETSTRIPE:
1186         case LL_IOC_MDC_GETINFO:
1187         case IOC_MDC_GETFILEINFO:
1188         case IOC_MDC_GETFILESTRIPE: {
1189                 struct ptlrpc_request *request = NULL;
1190                 struct lov_user_md *lump;
1191                 struct lov_mds_md *lmm = NULL;
1192                 struct mdt_body *body;
1193                 char *filename = NULL;
1194                 int lmmsize;
1195
1196                 if (cmd == IOC_MDC_GETFILEINFO ||
1197                     cmd == IOC_MDC_GETFILESTRIPE) {
1198                         filename = ll_getname((const char *)arg);
1199                         if (IS_ERR(filename))
1200                                 RETURN(PTR_ERR(filename));
1201
1202                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1203                                                       &lmmsize, &request);
1204                 } else {
1205                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1206                 }
1207
1208                 if (request) {
1209                         body = req_capsule_server_get(&request->rq_pill,
1210                                                       &RMF_MDT_BODY);
1211                         LASSERT(body != NULL);
1212                 } else {
1213                         GOTO(out_req, rc);
1214                 }
1215
1216                 if (rc < 0) {
1217                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1218                                                cmd == LL_IOC_MDC_GETINFO))
1219                                 GOTO(skip_lmm, rc = 0);
1220                         else
1221                                 GOTO(out_req, rc);
1222                 }
1223
1224                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1225                     cmd == LL_IOC_LOV_GETSTRIPE) {
1226                         lump = (struct lov_user_md *)arg;
1227                 } else {
1228                         struct lov_user_mds_data *lmdp;
1229                         lmdp = (struct lov_user_mds_data *)arg;
1230                         lump = &lmdp->lmd_lmm;
1231                 }
1232                 if (copy_to_user(lump, lmm, lmmsize)) {
1233                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1234                                 GOTO(out_req, rc = -EFAULT);
1235                         rc = -EOVERFLOW;
1236                 }
1237         skip_lmm:
1238                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1239                         struct lov_user_mds_data *lmdp;
1240                         lstat_t st = { 0 };
1241
1242                         st.st_dev     = inode->i_sb->s_dev;
1243                         st.st_mode    = body->mode;
1244                         st.st_nlink   = body->nlink;
1245                         st.st_uid     = body->uid;
1246                         st.st_gid     = body->gid;
1247                         st.st_rdev    = body->rdev;
1248                         st.st_size    = body->size;
1249                         st.st_blksize = PAGE_CACHE_SIZE;
1250                         st.st_blocks  = body->blocks;
1251                         st.st_atime   = body->atime;
1252                         st.st_mtime   = body->mtime;
1253                         st.st_ctime   = body->ctime;
1254                         st.st_ino     = inode->i_ino;
1255
1256                         lmdp = (struct lov_user_mds_data *)arg;
1257                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1258                                 GOTO(out_req, rc = -EFAULT);
1259                 }
1260
1261                 EXIT;
1262         out_req:
1263                 ptlrpc_req_finished(request);
1264                 if (filename)
1265                         ll_putname(filename);
1266                 return rc;
1267         }
1268         case IOC_LOV_GETINFO: {
1269                 struct lov_user_mds_data *lumd;
1270                 struct lov_stripe_md *lsm;
1271                 struct lov_user_md *lum;
1272                 struct lov_mds_md *lmm;
1273                 int lmmsize;
1274                 lstat_t st;
1275
1276                 lumd = (struct lov_user_mds_data *)arg;
1277                 lum = &lumd->lmd_lmm;
1278
1279                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1280                 if (rc)
1281                         RETURN(rc);
1282
1283                 OBD_ALLOC_LARGE(lmm, lmmsize);
1284                 if (lmm == NULL)
1285                         RETURN(-ENOMEM);
1286
1287                 if (copy_from_user(lmm, lum, lmmsize))
1288                         GOTO(free_lmm, rc = -EFAULT);
1289
1290                 switch (lmm->lmm_magic) {
1291                 case LOV_USER_MAGIC_V1:
1292                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1293                                 break;
1294                         /* swab objects first so that stripes num will be sane */
1295                         lustre_swab_lov_user_md_objects(
1296                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1297                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1298                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1299                         break;
1300                 case LOV_USER_MAGIC_V3:
1301                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1302                                 break;
1303                         /* swab objects first so that stripes num will be sane */
1304                         lustre_swab_lov_user_md_objects(
1305                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1306                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1307                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1308                         break;
1309                 default:
1310                         GOTO(free_lmm, rc = -EINVAL);
1311                 }
1312
1313                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1314                 if (rc < 0)
1315                         GOTO(free_lmm, rc = -ENOMEM);
1316
1317                 /* Perform glimpse_size operation. */
1318                 memset(&st, 0, sizeof(st));
1319
1320                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1321                 if (rc)
1322                         GOTO(free_lsm, rc);
1323
1324                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1325                         GOTO(free_lsm, rc = -EFAULT);
1326
1327                 EXIT;
1328         free_lsm:
1329                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1330         free_lmm:
1331                 OBD_FREE_LARGE(lmm, lmmsize);
1332                 return rc;
1333         }
1334         case OBD_IOC_LLOG_CATINFO: {
1335                 RETURN(-EOPNOTSUPP);
1336         }
1337         case OBD_IOC_QUOTACHECK: {
1338                 struct obd_quotactl *oqctl;
1339                 int error = 0;
1340
1341                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1342                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1343                         RETURN(-EPERM);
1344
1345                 OBD_ALLOC_PTR(oqctl);
1346                 if (!oqctl)
1347                         RETURN(-ENOMEM);
1348                 oqctl->qc_type = arg;
1349                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1350                 if (rc < 0) {
1351                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1352                         error = rc;
1353                 }
1354
1355                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1356                 if (rc < 0)
1357                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1358
1359                 OBD_FREE_PTR(oqctl);
1360                 return error ?: rc;
1361         }
1362         case OBD_IOC_POLL_QUOTACHECK: {
1363                 struct if_quotacheck *check;
1364
1365                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1366                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1367                         RETURN(-EPERM);
1368
1369                 OBD_ALLOC_PTR(check);
1370                 if (!check)
1371                         RETURN(-ENOMEM);
1372
1373                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1374                                    NULL);
1375                 if (rc) {
1376                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1377                         if (copy_to_user((void *)arg, check,
1378                                              sizeof(*check)))
1379                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1380                         GOTO(out_poll, rc);
1381                 }
1382
1383                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1384                                    NULL);
1385                 if (rc) {
1386                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1387                         if (copy_to_user((void *)arg, check,
1388                                              sizeof(*check)))
1389                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1390                         GOTO(out_poll, rc);
1391                 }
1392         out_poll:
1393                 OBD_FREE_PTR(check);
1394                 RETURN(rc);
1395         }
1396 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1397         case LL_IOC_QUOTACTL_18: {
1398                 /* copy the old 1.x quota struct for internal use, then copy
1399                  * back into old format struct.  For 1.8 compatibility. */
1400                 struct if_quotactl_18 *qctl_18;
1401                 struct if_quotactl *qctl_20;
1402
1403                 OBD_ALLOC_PTR(qctl_18);
1404                 if (!qctl_18)
1405                         RETURN(-ENOMEM);
1406
1407                 OBD_ALLOC_PTR(qctl_20);
1408                 if (!qctl_20)
1409                         GOTO(out_quotactl_18, rc = -ENOMEM);
1410
1411                 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1412                         GOTO(out_quotactl_20, rc = -ENOMEM);
1413
1414                 QCTL_COPY(qctl_20, qctl_18);
1415                 qctl_20->qc_idx = 0;
1416
1417                 /* XXX: dqb_valid was borrowed as a flag to mark that
1418                  *      only mds quota is wanted */
1419                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1420                     qctl_18->qc_dqblk.dqb_valid) {
1421                         qctl_20->qc_valid = QC_MDTIDX;
1422                         qctl_20->qc_dqblk.dqb_valid = 0;
1423                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1424                         qctl_20->qc_valid = QC_UUID;
1425                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1426                 } else {
1427                         qctl_20->qc_valid = QC_GENERAL;
1428                 }
1429
1430                 rc = quotactl_ioctl(sbi, qctl_20);
1431
1432                 if (rc == 0) {
1433                         QCTL_COPY(qctl_18, qctl_20);
1434                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1435
1436                         if (copy_to_user((void *)arg, qctl_18,
1437                                              sizeof(*qctl_18)))
1438                                 rc = -EFAULT;
1439                 }
1440
1441         out_quotactl_20:
1442                 OBD_FREE_PTR(qctl_20);
1443         out_quotactl_18:
1444                 OBD_FREE_PTR(qctl_18);
1445                 RETURN(rc);
1446         }
1447 #else
1448 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1449 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1450         case LL_IOC_QUOTACTL: {
1451                 struct if_quotactl *qctl;
1452
1453                 OBD_ALLOC_PTR(qctl);
1454                 if (!qctl)
1455                         RETURN(-ENOMEM);
1456
1457                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1458                         GOTO(out_quotactl, rc = -EFAULT);
1459
1460                 rc = quotactl_ioctl(sbi, qctl);
1461
1462                 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1463                         rc = -EFAULT;
1464
1465         out_quotactl:
1466                 OBD_FREE_PTR(qctl);
1467                 RETURN(rc);
1468         }
1469         case OBD_IOC_GETDTNAME:
1470         case OBD_IOC_GETMDNAME:
1471                 RETURN(ll_get_obd_name(inode, cmd, arg));
1472         case LL_IOC_FLUSHCTX:
1473                 RETURN(ll_flush_ctx(inode));
1474 #ifdef CONFIG_FS_POSIX_ACL
1475         case LL_IOC_RMTACL: {
1476             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1477                 inode == inode->i_sb->s_root->d_inode) {
1478                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1479
1480                 LASSERT(fd != NULL);
1481                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1482                 if (!rc)
1483                         fd->fd_flags |= LL_FILE_RMTACL;
1484                 RETURN(rc);
1485             } else
1486                 RETURN(0);
1487         }
1488 #endif
1489         case LL_IOC_GETOBDCOUNT: {
1490                 int count, vallen;
1491                 struct obd_export *exp;
1492
1493                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1494                         RETURN(-EFAULT);
1495
1496                 /* get ost count when count is zero, get mdt count otherwise */
1497                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1498                 vallen = sizeof(count);
1499                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1500                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1501                 if (rc) {
1502                         CERROR("get target count failed: %d\n", rc);
1503                         RETURN(rc);
1504                 }
1505
1506                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1507                         RETURN(-EFAULT);
1508
1509                 RETURN(0);
1510         }
1511         case LL_IOC_PATH2FID:
1512                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1513                                      sizeof(struct lu_fid)))
1514                         RETURN(-EFAULT);
1515                 RETURN(0);
1516         case LL_IOC_GET_CONNECT_FLAGS: {
1517                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1518         }
1519         case OBD_IOC_CHANGELOG_SEND:
1520         case OBD_IOC_CHANGELOG_CLEAR:
1521                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1522                                     sizeof(struct ioc_changelog));
1523                 RETURN(rc);
1524         case OBD_IOC_FID2PATH:
1525                 RETURN(ll_fid2path(inode, (void *)arg));
1526         case LL_IOC_HSM_REQUEST: {
1527                 struct hsm_user_request *hur;
1528                 int                      totalsize;
1529
1530                 OBD_ALLOC_PTR(hur);
1531                 if (hur == NULL)
1532                         RETURN(-ENOMEM);
1533
1534                 /* We don't know the true size yet; copy the fixed-size part */
1535                 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1536                         OBD_FREE_PTR(hur);
1537                         RETURN(-EFAULT);
1538                 }
1539
1540                 /* Compute the whole struct size */
1541                 totalsize = hur_len(hur);
1542                 OBD_FREE_PTR(hur);
1543
1544                 /* Make sure the size is reasonable */
1545                 if (totalsize >= MDS_MAXREQSIZE)
1546                         RETURN(-E2BIG);
1547
1548                 OBD_ALLOC_LARGE(hur, totalsize);
1549                 if (hur == NULL)
1550                         RETURN(-ENOMEM);
1551
1552                 /* Copy the whole struct */
1553                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1554                         OBD_FREE_LARGE(hur, totalsize);
1555                         RETURN(-EFAULT);
1556                 }
1557
1558                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1559                         const struct lu_fid *fid;
1560                         struct inode *f;
1561                         int i;
1562
1563                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1564                                 fid = &hur->hur_user_item[i].hui_fid;
1565                                 f = search_inode_for_lustre(inode->i_sb, fid);
1566                                 if (IS_ERR(f)) {
1567                                         rc = PTR_ERR(f);
1568                                         break;
1569                                 }
1570
1571                                 rc = ll_hsm_release(f);
1572                                 iput(f);
1573                                 if (rc != 0)
1574                                         break;
1575                         }
1576                 } else {
1577                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1578                                            hur, NULL);
1579                 }
1580
1581                 OBD_FREE_LARGE(hur, totalsize);
1582
1583                 RETURN(rc);
1584         }
1585         case LL_IOC_HSM_PROGRESS: {
1586                 struct hsm_progress_kernel      hpk;
1587                 struct hsm_progress             hp;
1588
1589                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1590                         RETURN(-EFAULT);
1591
1592                 hpk.hpk_fid = hp.hp_fid;
1593                 hpk.hpk_cookie = hp.hp_cookie;
1594                 hpk.hpk_extent = hp.hp_extent;
1595                 hpk.hpk_flags = hp.hp_flags;
1596                 hpk.hpk_errval = hp.hp_errval;
1597                 hpk.hpk_data_version = 0;
1598
1599                 /* File may not exist in Lustre; all progress
1600                  * reported to Lustre root */
1601                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1602                                    NULL);
1603                 RETURN(rc);
1604         }
1605         case LL_IOC_HSM_CT_START:
1606                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1607                         RETURN(-EPERM);
1608
1609                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1610                                     sizeof(struct lustre_kernelcomm));
1611                 RETURN(rc);
1612
1613         case LL_IOC_HSM_COPY_START: {
1614                 struct hsm_copy *copy;
1615                 int              rc;
1616
1617                 OBD_ALLOC_PTR(copy);
1618                 if (copy == NULL)
1619                         RETURN(-ENOMEM);
1620                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1621                         OBD_FREE_PTR(copy);
1622                         RETURN(-EFAULT);
1623                 }
1624
1625                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1626                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1627                         rc = -EFAULT;
1628
1629                 OBD_FREE_PTR(copy);
1630                 RETURN(rc);
1631         }
1632         case LL_IOC_HSM_COPY_END: {
1633                 struct hsm_copy *copy;
1634                 int              rc;
1635
1636                 OBD_ALLOC_PTR(copy);
1637                 if (copy == NULL)
1638                         RETURN(-ENOMEM);
1639                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1640                         OBD_FREE_PTR(copy);
1641                         RETURN(-EFAULT);
1642                 }
1643
1644                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1645                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1646                         rc = -EFAULT;
1647
1648                 OBD_FREE_PTR(copy);
1649                 RETURN(rc);
1650         }
1651         default:
1652                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1653                                      (void *)arg));
1654         }
1655 }
1656
1657 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1658 {
1659         struct inode *inode = file->f_mapping->host;
1660         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1661         struct ll_sb_info *sbi = ll_i2sbi(inode);
1662         int api32 = ll_need_32bit_api(sbi);
1663         loff_t ret = -EINVAL;
1664         ENTRY;
1665
1666         mutex_lock(&inode->i_mutex);
1667         switch (origin) {
1668                 case SEEK_SET:
1669                         break;
1670                 case SEEK_CUR:
1671                         offset += file->f_pos;
1672                         break;
1673                 case SEEK_END:
1674                         if (offset > 0)
1675                                 GOTO(out, ret);
1676                         if (api32)
1677                                 offset += LL_DIR_END_OFF_32BIT;
1678                         else
1679                                 offset += LL_DIR_END_OFF;
1680                         break;
1681                 default:
1682                         GOTO(out, ret);
1683         }
1684
1685         if (offset >= 0 &&
1686             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1687              (!api32 && offset <= LL_DIR_END_OFF))) {
1688                 if (offset != file->f_pos) {
1689                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1690                             (!api32 && offset == LL_DIR_END_OFF))
1691                                 fd->lfd_pos = MDS_DIR_END_OFF;
1692                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1693                                 fd->lfd_pos = offset << 32;
1694                         else
1695                                 fd->lfd_pos = offset;
1696                         file->f_pos = offset;
1697                         file->f_version = 0;
1698                 }
1699                 ret = offset;
1700         }
1701         GOTO(out, ret);
1702
1703 out:
1704         mutex_unlock(&inode->i_mutex);
1705         return ret;
1706 }
1707
1708 int ll_dir_open(struct inode *inode, struct file *file)
1709 {
1710         ENTRY;
1711         RETURN(ll_file_open(inode, file));
1712 }
1713
1714 int ll_dir_release(struct inode *inode, struct file *file)
1715 {
1716         ENTRY;
1717         RETURN(ll_file_release(inode, file));
1718 }
1719
1720 struct file_operations ll_dir_operations = {
1721         .llseek   = ll_dir_seek,
1722         .open     = ll_dir_open,
1723         .release  = ll_dir_release,
1724         .read     = generic_read_dir,
1725         .readdir  = ll_readdir,
1726         .unlocked_ioctl   = ll_dir_ioctl,
1727         .fsync    = ll_fsync,
1728 };