lustre/llite/dir.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/dir.c
  12  *  linux/fs/ext2/dir.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  ext2 directory handling functions
  17  *
  18  *  Big-endian to little-endian byte-swapping/bitmaps by
  19  *        David S. Miller (davem@caip.rutgers.edu), 1995
  20  *
  21  *  All code that works with directory layout had been switched to pagecache
  22  *  and moved here. AV
  23  *
  24  *  Adapted for Lustre Light
  25  *  Copyright (C) 2002-2003, Cluster File Systems, Inc.
  26  *
  27  */
  28
  29 #include <linux/fs.h>
  30 #include <linux/ext2_fs.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/mm.h>
  33 #include <linux/version.h>
  34 #include <linux/smp_lock.h>
  35 #include <asm/uaccess.h>
  36 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  37 # include <linux/locks.h>   // for wait_on_buffer
  38 #else
  39 # include <linux/buffer_head.h>   // for wait_on_buffer
  40 #endif
  41
  42 #define DEBUG_SUBSYSTEM S_LLITE
  43
  44 #include <linux/obd_support.h>
  45 #include <linux/obd_class.h>
  46 #include <linux/lustre_lib.h>
  47 #include <linux/lustre_idl.h>
  48 #include <linux/lustre_mdc.h>
  49 #include <linux/lustre_lite.h>
  50 #include <linux/lustre_dlm.h>
  51 #include "llite_internal.h"
  52
  53 typedef struct ext2_dir_entry_2 ext2_dirent;
  54
  55 #define PageChecked(page)        test_bit(PG_checked, &(page)->flags)
  56 #define SetPageChecked(page)     set_bit(PG_checked, &(page)->flags)
  57
  58 /* returns the page unlocked, but with a reference */
  59 static int ll_dir_readpage(struct file *file, struct page *page)
  60 {
  61         struct inode *inode = page->mapping->host;
  62         struct lu_fid fid;
  63         __u64 offset;
  64         struct ptlrpc_request *request;
  65         struct mdt_body *body;
  66         int rc = 0;
  67         ENTRY;
  68
  69         offset = (__u64)page->index << PAGE_SHIFT;
  70         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) off "LPU64"\n",
  71                inode->i_ino, inode->i_generation, inode, offset);
  72
  73         ll_inode2fid(&fid, inode);
  74
  75         rc = mdc_readpage(ll_i2sbi(inode)->ll_mdc_exp, &fid,
  76                           offset, page, &request);
  77         if (!rc) {
  78                 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
  79                 LASSERT (body != NULL);         /* checked by mdc_readpage() */
  80                 LASSERT_REPSWABBED (request, 0); /* swabbed by mdc_readpage() */
  81
  82                 inode->i_size = body->size;
  83                 SetPageUptodate(page);
  84         }
  85         ptlrpc_req_finished(request);
  86
  87         unlock_page(page);
  88         EXIT;
  89         return rc;
  90 }
  91
  92 struct address_space_operations ll_dir_aops = {
  93         .readpage  = ll_dir_readpage,
  94 };
  95
  96 /*
  97  * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
  98  * more robust, but we have what we have
  99  */
 100 static inline unsigned ext2_chunk_size(struct inode *inode)
 101 {
 102         return inode->i_sb->s_blocksize;
 103 }
 104
 105 static inline void ext2_put_page(struct page *page)
 106 {
 107         kunmap(page);
 108         page_cache_release(page);
 109 }
 110
 111 static inline unsigned long dir_pages(struct inode *inode)
 112 {
 113         return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
 114 }
 115
 116
 117 static void ext2_check_page(struct page *page)
 118 {
 119         struct inode *dir = page->mapping->host;
 120         unsigned chunk_size = ext2_chunk_size(dir);
 121         char *kaddr = page_address(page);
 122         //      u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
 123         unsigned offs, rec_len;
 124         unsigned limit = PAGE_CACHE_SIZE;
 125         ext2_dirent *p;
 126         char *error;
 127
 128         if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
 129                 limit = dir->i_size & ~PAGE_CACHE_MASK;
 130                 if (limit & (chunk_size - 1)) {
 131                         CERROR("limit %d dir size %lld index %ld\n",
 132                                limit, dir->i_size, page->index);
 133                         goto Ebadsize;
 134                 }
 135                 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
 136                         ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
 137                         p->rec_len = cpu_to_le16(chunk_size);
 138                         p->name_len = 0;
 139                         p->inode = 0;
 140                 }
 141                 if (!limit)
 142                         goto out;
 143         }
 144         for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
 145                 p = (ext2_dirent *)(kaddr + offs);
 146                 rec_len = le16_to_cpu(p->rec_len);
 147
 148                 if (rec_len < EXT2_DIR_REC_LEN(1))
 149                         goto Eshort;
 150                 if (rec_len & 3)
 151                         goto Ealign;
 152                 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
 153                         goto Enamelen;
 154                 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
 155                         goto Espan;
 156                 //              if (le32_to_cpu(p->inode) > max_inumber)
 157                 //goto Einumber;
 158         }
 159         if (offs != limit)
 160                 goto Eend;
 161 out:
 162         SetPageChecked(page);
 163         return;
 164
 165         /* Too bad, we had an error */
 166
 167 Ebadsize:
 168         CERROR("ext2_check_page"
 169                 "size of directory #%lu is not a multiple of chunk size\n",
 170                 dir->i_ino
 171         );
 172         goto fail;
 173 Eshort:
 174         error = "rec_len is smaller than minimal";
 175         goto bad_entry;
 176 Ealign:
 177         error = "unaligned directory entry";
 178         goto bad_entry;
 179 Enamelen:
 180         error = "rec_len is too small for name_len";
 181         goto bad_entry;
 182 Espan:
 183         error = "directory entry across blocks";
 184         goto bad_entry;
 185         //Einumber:
 186         // error = "inode out of bounds";
 187 bad_entry:
 188         CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
 189                 "offset=%lu+%u, inode=%lu, rec_len=%d, name_len=%d",
 190                 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT), offs,
 191                 (unsigned long) le32_to_cpu(p->inode),
 192                 rec_len, p->name_len);
 193         goto fail;
 194 Eend:
 195         p = (ext2_dirent *)(kaddr + offs);
 196         CERROR("ext2_check_page"
 197                 "entry in directory #%lu spans the page boundary"
 198                 "offset=%lu, inode=%lu",
 199                 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
 200                 (unsigned long) le32_to_cpu(p->inode));
 201 fail:
 202         SetPageChecked(page);
 203         SetPageError(page);
 204 }
 205
 206 static struct page *ll_get_dir_page(struct inode *dir, unsigned long n)
 207 {
 208         struct ldlm_res_id res_id =
 209                 { .name = { dir->i_ino, (__u64)dir->i_generation} };
 210         struct lustre_handle lockh;
 211         struct obd_device *obddev = class_exp2obd(ll_i2sbi(dir)->ll_mdc_exp);
 212         struct address_space *mapping = dir->i_mapping;
 213         struct page *page;
 214         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 215         int rc;
 216
 217         rc = ldlm_lock_match(obddev->obd_namespace, LDLM_FL_BLOCK_GRANTED,
 218                              &res_id, LDLM_IBITS, &policy, LCK_CR, &lockh);
 219         if (!rc) {
 220                 struct lookup_intent it = { .it_op = IT_READDIR };
 221                 struct ptlrpc_request *request;
 222                 struct mdc_op_data data;
 223
 224                 ll_prepare_mdc_op_data(&data, dir, NULL, NULL, 0, 0);
 225
 226                 rc = mdc_enqueue(ll_i2sbi(dir)->ll_mdc_exp, LDLM_IBITS, &it,
 227                                  LCK_CR, &data, &lockh, NULL, 0,
 228                                  ldlm_completion_ast, ll_mdc_blocking_ast, dir,
 229                                  0);
 230
 231                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 232                 if (request)
 233                         ptlrpc_req_finished(request);
 234                 if (rc < 0) {
 235                         CERROR("lock enqueue: rc: %d\n", rc);
 236                         return ERR_PTR(rc);
 237                 }
 238         }
 239         ldlm_lock_dump_handle(D_OTHER, &lockh);
 240
 241         page = read_cache_page(mapping, n,
 242                                (filler_t*)mapping->a_ops->readpage, NULL);
 243         if (!IS_ERR(page)) {
 244                 wait_on_page(page);
 245                 (void)kmap(page);
 246                 if (!PageUptodate(page))
 247                         goto fail;
 248                 if (!PageChecked(page))
 249                         ext2_check_page(page);
 250                 if (PageError(page))
 251                         goto fail;
 252         }
 253
 254 out_unlock:
 255         ldlm_lock_decref(&lockh, LCK_CR);
 256         return page;
 257
 258 fail:
 259         ext2_put_page(page);
 260         page = ERR_PTR(-EIO);
 261         goto out_unlock;
 262 }
 263
 264 /*
 265  * p is at least 6 bytes before the end of page
 266  */
 267 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
 268 {
 269         return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
 270 }
 271
 272 static inline unsigned
 273 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
 274 {
 275         ext2_dirent *de = (ext2_dirent*)(base + offset);
 276         ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
 277         while ((char*)p < (char*)de)
 278                 p = ext2_next_entry(p);
 279         return (char *)p - base;
 280 }
 281
 282 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
 283         [EXT2_FT_UNKNOWN]       DT_UNKNOWN,
 284         [EXT2_FT_REG_FILE]      DT_REG,
 285         [EXT2_FT_DIR]           DT_DIR,
 286         [EXT2_FT_CHRDEV]        DT_CHR,
 287         [EXT2_FT_BLKDEV]        DT_BLK,
 288         [EXT2_FT_FIFO]          DT_FIFO,
 289         [EXT2_FT_SOCK]          DT_SOCK,
 290         [EXT2_FT_SYMLINK]       DT_LNK,
 291 };
 292
 293
 294 int ll_readdir(struct file * filp, void * dirent, filldir_t filldir)
 295 {
 296         struct inode *inode = filp->f_dentry->d_inode;
 297         loff_t pos = filp->f_pos;
 298         // XXX struct super_block *sb = inode->i_sb;
 299         unsigned offset = pos & ~PAGE_CACHE_MASK;
 300         unsigned long n = pos >> PAGE_CACHE_SHIFT;
 301         unsigned long npages = dir_pages(inode);
 302         unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
 303         unsigned char *types = ext2_filetype_table;
 304         int need_revalidate = (filp->f_version != inode->i_version);
 305         int rc = 0;
 306         ENTRY;
 307
 308         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %llu/%llu\n",
 309                inode->i_ino, inode->i_generation, inode, pos, inode->i_size);
 310
 311         if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
 312                 RETURN(0);
 313
 314         for ( ; n < npages; n++, offset = 0) {
 315                 char *kaddr, *limit;
 316                 ext2_dirent *de;
 317                 struct page *page;
 318
 319                 CDEBUG(D_EXT2,"read %lu of dir %lu/%u page %lu/%lu size %llu\n",
 320                        PAGE_CACHE_SIZE, inode->i_ino, inode->i_generation,
 321                        n, npages, inode->i_size);
 322                 page = ll_get_dir_page(inode, n);
 323
 324                 /* size might have been updated by mdc_readpage */
 325                 npages = dir_pages(inode);
 326
 327                 if (IS_ERR(page)) {
 328                         rc = PTR_ERR(page);
 329                         CERROR("error reading dir %lu/%u page %lu: rc %d\n",
 330                                inode->i_ino, inode->i_generation, n, rc);
 331                         continue;
 332                 }
 333
 334                 kaddr = page_address(page);
 335                 if (need_revalidate) {
 336                         offset = ext2_validate_entry(kaddr, offset, chunk_mask);
 337                         need_revalidate = 0;
 338                 }
 339                 de = (ext2_dirent *)(kaddr+offset);
 340                 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
 341                 for ( ;(char*)de <= limit; de = ext2_next_entry(de)) {
 342                         if (de->inode) {
 343                                 int over;
 344
 345                                 rc = 0; /* no error if we return something */
 346
 347                                 offset = (char *)de - kaddr;
 348                                 over = filldir(dirent, de->name, de->name_len,
 349                                                (n<<PAGE_CACHE_SHIFT) | offset,
 350                                                le32_to_cpu(de->inode),
 351                                                types[de->file_type &
 352                                                      (EXT2_FT_MAX - 1)]);
 353                                 if (over) {
 354                                         ext2_put_page(page);
 355                                         GOTO(done, rc);
 356                                 }
 357                         }
 358                 }
 359                 ext2_put_page(page);
 360         }
 361
 362 done:
 363         filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
 364         filp->f_version = inode->i_version;
 365         update_atime(inode);
 366         RETURN(rc);
 367 }
 368
 369 #define QCTL_COPY(out, in)              \
 370 do {                                    \
 371         Q_COPY(out, in, qc_cmd);        \
 372         Q_COPY(out, in, qc_type);       \
 373         Q_COPY(out, in, qc_id);         \
 374         Q_COPY(out, in, qc_stat);       \
 375         Q_COPY(out, in, qc_dqinfo);     \
 376         Q_COPY(out, in, qc_dqblk);      \
 377 } while (0)
 378
 379 static int ll_dir_ioctl(struct inode *inode, struct file *file,
 380                         unsigned int cmd, unsigned long arg)
 381 {
 382         struct ll_sb_info *sbi = ll_i2sbi(inode);
 383         struct obd_ioctl_data *data;
 384         ENTRY;
 385
 386         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
 387                inode->i_ino, inode->i_generation, inode, cmd);
 388
 389         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
 390         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
 391                 return -ENOTTY;
 392
 393         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_IOCTL);
 394         switch(cmd) {
 395         case EXT3_IOC_GETFLAGS:
 396         case EXT3_IOC_SETFLAGS:
 397                 RETURN(ll_iocontrol(inode, file, cmd, arg));
 398         case EXT3_IOC_GETVERSION_OLD:
 399         case EXT3_IOC_GETVERSION:
 400                 RETURN(put_user(inode->i_generation, (int *)arg));
 401         /* We need to special case any other ioctls we want to handle,
 402          * to send them to the MDS/OST as appropriate and to properly
 403          * network encode the arg field.
 404         case EXT3_IOC_SETVERSION_OLD:
 405         case EXT3_IOC_SETVERSION:
 406         */
 407         case IOC_MDC_LOOKUP: {
 408                 struct ptlrpc_request *request = NULL;
 409                 struct lu_fid fid;
 410                 char *buf = NULL;
 411                 char *filename;
 412                 int namelen, rc, len = 0;
 413
 414                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 415                 if (rc)
 416                         RETURN(rc);
 417                 data = (void *)buf;
 418
 419                 filename = data->ioc_inlbuf1;
 420                 namelen = data->ioc_inllen1;
 421
 422                 if (namelen < 1) {
 423                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
 424                         GOTO(out, rc = -EINVAL);
 425                 }
 426
 427                 ll_inode2fid(&fid, inode);
 428                 rc = mdc_getattr_name(sbi->ll_mdc_exp, &fid, filename, namelen,
 429                                       OBD_MD_FLID, 0, &request);
 430                 if (rc < 0) {
 431                         CDEBUG(D_INFO, "mdc_getattr_name: %d\n", rc);
 432                         GOTO(out, rc);
 433                 }
 434
 435                 ptlrpc_req_finished(request);
 436
 437                 EXIT;
 438         out:
 439                 obd_ioctl_freedata(buf, len);
 440                 return rc;
 441         }
 442         case LL_IOC_LOV_SETSTRIPE: {
 443                 struct ptlrpc_request *request = NULL;
 444                 struct mdc_op_data op_data;
 445                 struct iattr attr = { 0 };
 446                 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
 447                 int rc = 0;
 448
 449                 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
 450
 451                 LASSERT(sizeof(lum) == sizeof(*lump));
 452                 LASSERT(sizeof(lum.lmm_objects[0]) ==
 453                         sizeof(lump->lmm_objects[0]));
 454                 rc = copy_from_user(&lum, lump, sizeof(lum));
 455                 if (rc)
 456                         return(-EFAULT);
 457
 458                 /*
 459                  * This is coming from userspace, so should be in
 460                  * local endian.  But the MDS would like it in little
 461                  * endian, so we swab it before we send it.
 462                  */
 463                 if (lum.lmm_magic != LOV_USER_MAGIC)
 464                         RETURN(-EINVAL);
 465
 466                 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC))
 467                         lustre_swab_lov_user_md(&lum);
 468
 469                 /* swabbing is done in lov_setstripe() on server side */
 470                 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
 471                                  &attr, &lum, sizeof(lum), NULL, 0, &request);
 472                 if (rc) {
 473                         ptlrpc_req_finished(request);
 474                         if (rc != -EPERM && rc != -EACCES)
 475                                 CERROR("mdc_setattr fails: rc = %d\n", rc);
 476                         return rc;
 477                 }
 478                 ptlrpc_req_finished(request);
 479
 480                 return rc;
 481         }
 482         case LL_IOC_LOV_GETSTRIPE: {
 483                 struct ptlrpc_request *request = NULL;
 484                 struct lov_user_md *lump = (struct lov_user_md *)arg;
 485                 struct lov_mds_md *lmm;
 486                 struct lu_fid fid;
 487                 struct mdt_body *body;
 488                 int rc, lmmsize;
 489
 490                 ll_inode2fid(&fid, inode);
 491
 492                 rc = ll_get_max_mdsize(sbi, &lmmsize);
 493                 if (rc)
 494                         RETURN(rc);
 495
 496                 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, OBD_MD_FLDIREA,
 497                                  lmmsize, &request);
 498                 if (rc < 0) {
 499                         CDEBUG(D_INFO, "mdc_getattr failed: rc = %d\n", rc);
 500                         RETURN(rc);
 501                 }
 502
 503                 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
 504                 LASSERT(body != NULL);         /* checked by mdc_getattr_name */
 505                 LASSERT_REPSWABBED(request, 0);/* swabbed by mdc_getattr_name */
 506
 507                 lmmsize = body->eadatasize;
 508                 if (lmmsize == 0)
 509                         GOTO(out_get, rc = -ENODATA);
 510
 511                 lmm = lustre_msg_buf(request->rq_repmsg, 1, lmmsize);
 512                 LASSERT(lmm != NULL);
 513                 LASSERT_REPSWABBED(request, 1);
 514
 515                 /*
 516                  * This is coming from the MDS, so is probably in
 517                  * little endian.  We convert it to host endian before
 518                  * passing it to userspace.
 519                  */
 520                 if (lmm->lmm_magic == __swab32(LOV_MAGIC)) {
 521                         lustre_swab_lov_user_md((struct lov_user_md *)lmm);
 522                         lustre_swab_lov_user_md_objects((struct lov_user_md *)lmm);
 523                 }
 524
 525                 rc = copy_to_user(lump, lmm, lmmsize);
 526                 if (rc)
 527                         GOTO(out_get, rc = -EFAULT);
 528
 529                 EXIT;
 530         out_get:
 531                 ptlrpc_req_finished(request);
 532                 return rc;
 533         }
 534         case LL_IOC_OBD_STATFS:
 535                 RETURN(ll_obd_statfs(inode, (void *)arg));
 536         case IOC_MDC_GETFILEINFO:
 537         case IOC_MDC_GETSTRIPE: {
 538                 struct ptlrpc_request *request = NULL;
 539                 struct lu_fid fid;
 540                 struct mdt_body *body;
 541                 struct lov_user_md *lump;
 542                 struct lov_mds_md *lmm;
 543                 char *filename;
 544                 int rc, lmmsize;
 545
 546                 filename = getname((const char *)arg);
 547                 if (IS_ERR(filename))
 548                         RETURN(PTR_ERR(filename));
 549
 550                 ll_inode2fid(&fid, inode);
 551
 552                 rc = ll_get_max_mdsize(sbi, &lmmsize);
 553                 if (rc)
 554                         RETURN(rc);
 555
 556                 rc = mdc_getattr_name(sbi->ll_mdc_exp, &fid, filename,
 557                                       strlen(filename) + 1, OBD_MD_FLEASIZE,
 558                                       lmmsize, &request);
 559                 if (rc < 0) {
 560                         CDEBUG(D_INFO, "mdc_getattr_name failed on %s: rc %d\n",
 561                                filename, rc);
 562                         GOTO(out_name, rc);
 563                 }
 564
 565                 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
 566                 LASSERT(body != NULL);         /* checked by mdc_getattr_name */
 567                 LASSERT_REPSWABBED(request, 0);/* swabbed by mdc_getattr_name */
 568
 569                 lmmsize = body->eadatasize;
 570
 571                 if (!(body->valid & OBD_MD_FLEASIZE) || lmmsize == 0)
 572                         GOTO(out_req, rc = -ENODATA);
 573
 574                 if (lmmsize > 4096)
 575                         GOTO(out_req, rc = -EFBIG);
 576
 577                 lmm = lustre_msg_buf(request->rq_repmsg, 1, lmmsize);
 578                 LASSERT(lmm != NULL);
 579                 LASSERT_REPSWABBED(request, 1);
 580
 581                 /*
 582                  * This is coming from the MDS, so is probably in
 583                  * little endian.  We convert it to host endian before
 584                  * passing it to userspace.
 585                  */
 586                 if (lmm->lmm_magic == __swab32(LOV_MAGIC)) {
 587                         lustre_swab_lov_user_md((struct lov_user_md *)lmm);
 588                         lustre_swab_lov_user_md_objects((struct lov_user_md *)lmm);
 589                 } else if (lmm->lmm_magic == __swab32(LOV_MAGIC_JOIN)) {
 590                         lustre_swab_lov_user_md_join((struct lov_user_md_join *)lmm);
 591                 }
 592                 if (lmm->lmm_magic == LOV_MAGIC_JOIN) {
 593                         struct lov_stripe_md *lsm;
 594                         struct lov_user_md_join *lmj;
 595                         int lmj_size, i, aindex = 0, rc;
 596
 597                         rc = obd_unpackmd(sbi->ll_osc_exp, &lsm, lmm, lmmsize);
 598                         if (rc < 0)
 599                                 GOTO(out_req, rc = -ENOMEM);
 600                         rc = obd_checkmd(sbi->ll_osc_exp, sbi->ll_mdc_exp, lsm);
 601                         if (rc)
 602                                 GOTO(out_free_memmd, rc);
 603
 604                         lmj_size = sizeof(struct lov_user_md_join) +
 605                                    lsm->lsm_stripe_count *
 606                                    sizeof(struct lov_user_ost_data_join);
 607                         OBD_ALLOC(lmj, lmj_size);
 608                         if (!lmj)
 609                                 GOTO(out_free_memmd, rc = -ENOMEM);
 610
 611                         memcpy(lmj, lmm, sizeof(struct lov_user_md_join));
 612                         for(i = 0; i < lsm->lsm_stripe_count; i++) {
 613                                 struct lov_array_info *lai = lsm->lsm_array;
 614                                 if ((lai->lai_ext_array[aindex].le_loi_idx +
 615                                      lai->lai_ext_array[aindex].le_stripe_count)<=i){
 616                                         aindex ++;
 617                                 }
 618                                 CDEBUG(D_INFO, "aindex %d i %d l_extent_start"LPU64""
 619                                                "len %d \n", aindex, i,
 620                                                lai->lai_ext_array[aindex].le_start,
 621                                                (int)lai->lai_ext_array[aindex].le_len);
 622                                 lmj->lmm_objects[i].l_extent_start =
 623                                         lai->lai_ext_array[aindex].le_start;
 624
 625                                 if ((int)lai->lai_ext_array[aindex].le_len == -1) {
 626                                         lmj->lmm_objects[i].l_extent_end = -1;
 627                                 } else {
 628                                         lmj->lmm_objects[i].l_extent_end =
 629                                         lai->lai_ext_array[aindex].le_start +
 630                                         lai->lai_ext_array[aindex].le_len;
 631                                 }
 632                                 lmj->lmm_objects[i].l_object_id =
 633                                         lsm->lsm_oinfo[i].loi_id;
 634                                 lmj->lmm_objects[i].l_object_gr =
 635                                         lsm->lsm_oinfo[i].loi_gr;
 636                                 lmj->lmm_objects[i].l_ost_gen =
 637                                         lsm->lsm_oinfo[i].loi_ost_gen;
 638                                 lmj->lmm_objects[i].l_ost_idx =
 639                                         lsm->lsm_oinfo[i].loi_ost_idx;
 640                         }
 641                         lmm = (struct lov_mds_md *)lmj;
 642                         lmmsize = lmj_size;
 643 out_free_memmd:
 644                         obd_free_memmd(sbi->ll_osc_exp, &lsm);
 645                         if (rc)
 646                                 GOTO(out_req, rc);
 647                 }
 648                 if (cmd == IOC_MDC_GETFILEINFO) {
 649                         struct lov_user_mds_data *lmdp;
 650                         lstat_t st = { 0 };
 651
 652                         st.st_dev     = 0;
 653                         st.st_mode    = body->mode;
 654                         st.st_nlink   = body->nlink;
 655                         st.st_uid     = body->uid;
 656                         st.st_gid     = body->gid;
 657                         st.st_rdev    = body->rdev;
 658                         st.st_size    = body->size;
 659                         st.st_blksize = PAGE_SIZE;
 660                         st.st_blocks  = body->blocks;
 661                         st.st_atime   = body->atime;
 662                         st.st_mtime   = body->mtime;
 663                         st.st_ctime   = body->ctime;
 664                         st.st_ino     = body->ino;
 665
 666                         lmdp = (struct lov_user_mds_data *)arg;
 667                         rc = copy_to_user(&lmdp->lmd_st, &st, sizeof(st));
 668                         if (rc)
 669                                 GOTO(out_req, rc = -EFAULT);
 670                         lump = &lmdp->lmd_lmm;
 671                 } else {
 672                         lump = (struct lov_user_md *)arg;
 673                 }
 674
 675                 rc = copy_to_user(lump, lmm, lmmsize);
 676                 if (lmm->lmm_magic == LOV_MAGIC_JOIN)
 677                         OBD_FREE(lmm, lmmsize);
 678                 if (rc)
 679                         GOTO(out_req, rc = -EFAULT);
 680
 681                 EXIT;
 682         out_req:
 683                 ptlrpc_req_finished(request);
 684         out_name:
 685                 putname(filename);
 686                 return rc;
 687         }
 688         case OBD_IOC_LLOG_CATINFO: {
 689                 struct ptlrpc_request *req = NULL;
 690                 char *buf = NULL;
 691                 int rc, len = 0;
 692                 char *bufs[2], *str;
 693                 int lens[2], size;
 694
 695                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 696                 if (rc)
 697                         RETURN(rc);
 698                 data = (void *)buf;
 699
 700                 if (!data->ioc_inlbuf1) {
 701                         obd_ioctl_freedata(buf, len);
 702                         RETURN(-EINVAL);
 703                 }
 704
 705                 lens[0] = data->ioc_inllen1;
 706                 bufs[0] = data->ioc_inlbuf1;
 707                 if (data->ioc_inllen2) {
 708                         lens[1] = data->ioc_inllen2;
 709                         bufs[1] = data->ioc_inlbuf2;
 710                 } else {
 711                         lens[1] = 0;
 712                         bufs[1] = NULL;
 713                 }
 714                 size = data->ioc_plen1;
 715                 req = ptlrpc_prep_req(sbi2mdc(sbi)->cl_import,
 716                                       LUSTRE_LOG_VERSION, LLOG_CATINFO,
 717                                       2, lens, bufs);
 718                 if (!req)
 719                         GOTO(out_catinfo, rc = -ENOMEM);
 720                 req->rq_replen = lustre_msg_size(1, &size);
 721
 722                 rc = ptlrpc_queue_wait(req);
 723                 str = lustre_msg_string(req->rq_repmsg, 0, data->ioc_plen1);
 724                 if (!rc)
 725                         rc = copy_to_user(data->ioc_pbuf1, str,
 726                                           data->ioc_plen1);
 727                 ptlrpc_req_finished(req);
 728         out_catinfo:
 729                 obd_ioctl_freedata(buf, len);
 730                 RETURN(rc);
 731         }
 732         case OBD_IOC_QUOTACHECK: {
 733                 struct obd_quotactl *oqctl;
 734                 int rc, error = 0;
 735
 736                 if (!capable(CAP_SYS_ADMIN))
 737                         RETURN(-EPERM);
 738
 739                 OBD_ALLOC_PTR(oqctl);
 740                 if (!oqctl)
 741                         RETURN(-ENOMEM);
 742                 oqctl->qc_type = arg;
 743                 rc = obd_quotacheck(sbi->ll_mdc_exp, oqctl);
 744                 if (rc < 0) {
 745                         CDEBUG(D_INFO, "mdc_quotacheck failed: rc %d\n", rc);
 746                         error = rc;
 747                 }
 748
 749                 rc = obd_quotacheck(sbi->ll_osc_exp, oqctl);
 750                 if (rc < 0)
 751                         CDEBUG(D_INFO, "osc_quotacheck failed: rc %d\n", rc);
 752
 753                 OBD_FREE_PTR(oqctl);
 754                 return error ?: rc;
 755         }
 756         case OBD_IOC_POLL_QUOTACHECK: {
 757                 struct if_quotacheck *check;
 758                 int rc;
 759
 760                 if (!capable(CAP_SYS_ADMIN))
 761                         RETURN(-EPERM);
 762
 763                 OBD_ALLOC_PTR(check);
 764                 if (!check)
 765                         RETURN(-ENOMEM);
 766
 767                 rc = obd_iocontrol(cmd, sbi->ll_mdc_exp, 0, (void *)check,
 768                                    NULL);
 769                 if (rc) {
 770                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
 771                         if (copy_to_user((void *)arg, check, sizeof(*check)))
 772                                 rc = -EFAULT;
 773                         GOTO(out_poll, rc);
 774                 }
 775
 776                 rc = obd_iocontrol(cmd, sbi->ll_osc_exp, 0, (void *)check,
 777                                    NULL);
 778                 if (rc) {
 779                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
 780                         if (copy_to_user((void *)arg, check, sizeof(*check)))
 781                                 rc = -EFAULT;
 782                         GOTO(out_poll, rc);
 783                 }
 784         out_poll:
 785                 OBD_FREE_PTR(check);
 786                 RETURN(rc);
 787         }
 788 #if HAVE_QUOTA_SUPPORT
 789         case OBD_IOC_QUOTACTL: {
 790                 struct if_quotactl *qctl;
 791                 struct obd_quotactl *oqctl;
 792
 793                 int cmd, type, id, rc = 0;
 794
 795                 OBD_ALLOC_PTR(qctl);
 796                 if (!qctl)
 797                         RETURN(-ENOMEM);
 798
 799                 OBD_ALLOC_PTR(oqctl);
 800                 if (!oqctl) {
 801                         OBD_FREE_PTR(qctl);
 802                         RETURN(-ENOMEM);
 803                 }
 804                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
 805                         GOTO(out_quotactl, rc = -EFAULT);
 806
 807                 cmd = qctl->qc_cmd;
 808                 type = qctl->qc_type;
 809                 id = qctl->qc_id;
 810                 switch (cmd) {
 811                 case Q_QUOTAON:
 812                 case Q_QUOTAOFF:
 813                 case Q_SETQUOTA:
 814                 case Q_SETINFO:
 815                         if (!capable(CAP_SYS_ADMIN))
 816                                 GOTO(out_quotactl, rc = -EPERM);
 817                         break;
 818                 case Q_GETQUOTA:
 819                         if (((type == USRQUOTA && current->euid != id) ||
 820                              (type == GRPQUOTA && !in_egroup_p(id))) &&
 821                             !capable(CAP_SYS_ADMIN))
 822                                 GOTO(out_quotactl, rc = -EPERM);
 823
 824                         /* XXX: dqb_valid is borrowed as a flag to mark that
 825                          *      only mds quota is wanted */
 826                         if (qctl->qc_dqblk.dqb_valid)
 827                                 qctl->obd_uuid =
 828                                        sbi->ll_mdc_exp->exp_obd->u.cli.
 829                                        cl_import->imp_target_uuid;
 830                         break;
 831                 case Q_GETINFO:
 832                         break;
 833                 default:
 834                         CERROR("unsupported quotactl op: %#x\n", cmd);
 835                         GOTO(out_quotactl, -ENOTTY);
 836                 }
 837
 838                 QCTL_COPY(oqctl, qctl);
 839
 840                 if (qctl->obd_uuid.uuid[0]) {
 841                         struct obd_device *obd;
 842                         struct obd_uuid *uuid = &qctl->obd_uuid;
 843
 844                         obd = class_find_client_notype(uuid,
 845                                          &sbi->ll_osc_exp->exp_obd->obd_uuid);
 846                         if (!obd)
 847                                 GOTO(out_quotactl, rc = -ENOENT);
 848
 849                         if (cmd == Q_GETINFO)
 850                                 oqctl->qc_cmd = Q_GETOINFO;
 851                         else if (cmd == Q_GETQUOTA)
 852                                 oqctl->qc_cmd = Q_GETOQUOTA;
 853                         else
 854                                 GOTO(out_quotactl, rc = -EINVAL);
 855
 856                         if (sbi->ll_mdc_exp->exp_obd == obd) {
 857                                 rc = obd_quotactl(sbi->ll_mdc_exp, oqctl);
 858                         } else {
 859                                 int i;
 860                                 struct obd_export *exp;
 861                                 struct lov_obd *lov = &sbi->ll_osc_exp->
 862                                                             exp_obd->u.lov;
 863
 864                                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 865                                         exp = lov->tgts[i].ltd_exp;
 866
 867                                         if (!lov->tgts[i].active)
 868                                                 continue;
 869
 870                                         if (exp->exp_obd == obd) {
 871                                                 rc = obd_quotactl(exp, oqctl);
 872                                                 break;
 873                                         }
 874                                 }
 875                         }
 876
 877                         oqctl->qc_cmd = cmd;
 878                         QCTL_COPY(qctl, oqctl);
 879
 880                         if (copy_to_user((void *)arg, qctl, sizeof(*qctl)))
 881                                 rc = -EFAULT;
 882
 883                         GOTO(out_quotactl, rc);
 884                 }
 885
 886                 rc = obd_quotactl(sbi->ll_mdc_exp, oqctl);
 887                 if (rc && rc != -EBUSY && cmd == Q_QUOTAON) {
 888                         oqctl->qc_cmd = Q_QUOTAOFF;
 889                         obd_quotactl(sbi->ll_mdc_exp, oqctl);
 890                 }
 891
 892                 QCTL_COPY(qctl, oqctl);
 893
 894                 if (copy_to_user((void *)arg, qctl, sizeof(*qctl)))
 895                         rc = -EFAULT;
 896         out_quotactl:
 897                 OBD_FREE_PTR(qctl);
 898                 OBD_FREE_PTR(oqctl);
 899                 RETURN(rc);
 900         }
 901 #endif /* HAVE_QUOTA_SUPPORT */
 902         case OBD_IOC_GETNAME: {
 903                 struct obd_device *obd = class_exp2obd(sbi->ll_osc_exp);
 904                 if (!obd)
 905                         RETURN(-EFAULT);
 906                 if (copy_to_user((void *)arg, obd->obd_name,
 907                                 strlen(obd->obd_name) + 1))
 908                         RETURN (-EFAULT);
 909                 RETURN(0);
 910         }
 911         default:
 912                 RETURN(obd_iocontrol(cmd, sbi->ll_osc_exp,0,NULL,(void *)arg));
 913         }
 914 }
 915
 916 int ll_dir_open(struct inode *inode, struct file *file)
 917 {
 918         ENTRY;
 919         RETURN(ll_file_open(inode, file));
 920 }
 921
 922 int ll_dir_release(struct inode *inode, struct file *file)
 923 {
 924         ENTRY;
 925         RETURN(ll_file_release(inode, file));
 926 }
 927
 928 struct file_operations ll_dir_operations = {
 929         .open     = ll_dir_open,
 930         .release  = ll_dir_release,
 931         .read     = generic_read_dir,
 932         .readdir  = ll_readdir,
 933         .ioctl    = ll_dir_ioctl
 934 };
 935