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 <obd_support.h>
  45 #include <obd_class.h>
  46 #include <lustre_lib.h>
  47 #include <lustre/lustre_idl.h>
  48 #include <lustre_lite.h>
  49 #include <lustre_dlm.h>
  50 #include "llite_internal.h"
  51
  52 typedef struct ext2_dir_entry_2 ext2_dirent;
  53
  54 #define PageChecked(page)        test_bit(PG_checked, &(page)->flags)
  55 #define SetPageChecked(page)     set_bit(PG_checked, &(page)->flags)
  56
  57 /* returns the page unlocked, but with a reference */
  58 static int ll_dir_readpage(struct file *file, struct page *page)
  59 {
  60         struct inode *inode = page->mapping->host;
  61         struct ll_fid mdc_fid;
  62         __u64 offset;
  63         struct ptlrpc_request *request;
  64         struct mds_body *body;
  65         int rc = 0;
  66         ENTRY;
  67
  68         offset = (__u64)page->index << CFS_PAGE_SHIFT;
  69         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) off "LPU64"\n",
  70                inode->i_ino, inode->i_generation, inode, offset);
  71
  72         mdc_pack_fid(&mdc_fid, inode->i_ino, inode->i_generation, S_IFDIR);
  73
  74         rc = mdc_readpage(ll_i2sbi(inode)->ll_mdc_exp, &mdc_fid,
  75                           offset, page, &request);
  76         if (!rc) {
  77                 body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF,
  78                                       sizeof(*body));
  79                 LASSERT(body != NULL); /* checked by mdc_readpage() */
  80                 /* swabbed by mdc_readpage() */
  81                 LASSERT_REPSWABBED(request, REPLY_REC_OFF);
  82
  83                 inode->i_size = body->size;
  84                 SetPageUptodate(page);
  85         }
  86         ptlrpc_req_finished(request);
  87
  88         unlock_page(page);
  89         EXIT;
  90         return rc;
  91 }
  92
  93 struct address_space_operations ll_dir_aops = {
  94         .readpage  = ll_dir_readpage,
  95 };
  96
  97 /*
  98  * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
  99  * more robust, but we have what we have
 100  */
 101 static inline unsigned ext2_chunk_size(struct inode *inode)
 102 {
 103         return inode->i_sb->s_blocksize;
 104 }
 105
 106 static inline void ext2_put_page(struct page *page)
 107 {
 108         kunmap(page);
 109         page_cache_release(page);
 110 }
 111
 112 static inline unsigned long dir_pages(struct inode *inode)
 113 {
 114         return (inode->i_size+CFS_PAGE_SIZE-1) >> CFS_PAGE_SHIFT;
 115 }
 116
 117
 118 static void ext2_check_page(struct inode *dir, struct page *page)
 119 {
 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 rec_len;
 124         __u64 offs, limit = CFS_PAGE_SIZE;
 125         ext2_dirent *p;
 126         char *error;
 127
 128         if ((dir->i_size >> CFS_PAGE_SHIFT) == (__u64)page->index) {
 129                 limit = dir->i_size & ~CFS_PAGE_MASK;
 130                 if (limit & (chunk_size - 1)) {
 131                         CERROR("limit "LPU64" dir size %lld index "LPU64"\n",
 132                                limit, dir->i_size, (__u64)page->index);
 133                         goto Ebadsize;
 134                 }
 135                 for (offs = limit; offs < CFS_PAGE_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("%s: directory %lu/%u size %llu is not a multiple of %u\n",
 169                ll_i2mdcexp(dir)->exp_obd->obd_name, dir->i_ino,
 170                dir->i_generation, dir->i_size, chunk_size);
 171         goto fail;
 172 Eshort:
 173         error = "rec_len is smaller than minimal";
 174         goto bad_entry;
 175 Ealign:
 176         error = "unaligned directory entry";
 177         goto bad_entry;
 178 Enamelen:
 179         error = "rec_len is too small for name_len";
 180         goto bad_entry;
 181 Espan:
 182         error = "directory entry across blocks";
 183         goto bad_entry;
 184         //Einumber:
 185         // error = "inode out of bounds";
 186 bad_entry:
 187         CERROR("%s: bad entry in directory %lu/%u: %s - "
 188                "offset="LPU64"+"LPU64", inode=%lu, rec_len=%d, name_len=%d\n",
 189                ll_i2mdcexp(dir)->exp_obd->obd_name, dir->i_ino,
 190                dir->i_generation, error, (__u64)page->index << CFS_PAGE_SHIFT,
 191                offs, (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("%s: entry in directory %lu/%u spans the page boundary "
 197                "offset="LPU64"+"LPU64", inode=%lu\n",ll_i2mdcexp(dir)->exp_obd->obd_name,
 198                dir->i_ino, dir->i_generation,
 199                (__u64)page->index << CFS_PAGE_SHIFT,
 200                offs, (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                 GOTO(out_unlock, page);
 245
 246         wait_on_page(page);
 247         (void)kmap(page);
 248         if (!PageUptodate(page))
 249                 goto fail;
 250         if (!PageChecked(page))
 251                 ext2_check_page(dir, page);
 252         if (PageError(page))
 253                 goto fail;
 254
 255 out_unlock:
 256         ldlm_lock_decref(&lockh, LCK_CR);
 257         return page;
 258
 259 fail:
 260         ext2_put_page(page);
 261         page = ERR_PTR(-EIO);
 262         goto out_unlock;
 263 }
 264
 265 /*
 266  * p is at least 6 bytes before the end of page
 267  */
 268 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
 269 {
 270         return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
 271 }
 272
 273 static inline unsigned
 274 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
 275 {
 276         ext2_dirent *de = (ext2_dirent*)(base + offset);
 277         ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
 278         while ((char*)p < (char*)de)
 279                 p = ext2_next_entry(p);
 280         return (char *)p - base;
 281 }
 282
 283 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
 284         [EXT2_FT_UNKNOWN]       DT_UNKNOWN,
 285         [EXT2_FT_REG_FILE]      DT_REG,
 286         [EXT2_FT_DIR]           DT_DIR,
 287         [EXT2_FT_CHRDEV]        DT_CHR,
 288         [EXT2_FT_BLKDEV]        DT_BLK,
 289         [EXT2_FT_FIFO]          DT_FIFO,
 290         [EXT2_FT_SOCK]          DT_SOCK,
 291         [EXT2_FT_SYMLINK]       DT_LNK,
 292 };
 293
 294
 295 int ll_readdir(struct file *filp, void *dirent, filldir_t filldir)
 296 {
 297         struct inode *inode = filp->f_dentry->d_inode;
 298         loff_t pos = filp->f_pos;
 299         // XXX struct super_block *sb = inode->i_sb;
 300         __u64 offset = pos & ~CFS_PAGE_MASK;
 301         __u64 n = pos >> CFS_PAGE_SHIFT;
 302         unsigned long npages = dir_pages(inode);
 303         unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
 304         unsigned char *types = ext2_filetype_table;
 305         int need_revalidate = (filp->f_version != inode->i_version);
 306         int rc = 0;
 307         ENTRY;
 308
 309         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %llu/%llu\n",
 310                inode->i_ino, inode->i_generation, inode, pos, inode->i_size);
 311
 312         if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
 313                 RETURN(0);
 314
 315         for ( ; n < npages; n++, offset = 0) {
 316                 char *kaddr, *limit;
 317                 ext2_dirent *de;
 318                 struct page *page;
 319
 320                 CDEBUG(D_EXT2,"read %lu of dir %lu/%u page "LPU64"/%lu "
 321                        "size %llu\n",
 322                        CFS_PAGE_SIZE, inode->i_ino, inode->i_generation,
 323                        n, npages, inode->i_size);
 324                 page = ll_get_dir_page(inode, n);
 325
 326                 /* size might have been updated by mdc_readpage */
 327                 npages = dir_pages(inode);
 328
 329                 if (IS_ERR(page)) {
 330                         rc = PTR_ERR(page);
 331                         CERROR("error reading dir %lu/%u page "LPU64": rc %d\n",
 332                                inode->i_ino, inode->i_generation, n, rc);
 333                         continue;
 334                 }
 335
 336                 kaddr = page_address(page);
 337                 if (need_revalidate) {
 338                         /* page already checked from ll_get_dir_page() */
 339                         offset = ext2_validate_entry(kaddr, offset, chunk_mask);
 340                         need_revalidate = 0;
 341                 }
 342                 de = (ext2_dirent *)(kaddr+offset);
 343                 limit = kaddr + CFS_PAGE_SIZE - EXT2_DIR_REC_LEN(1);
 344                 for ( ;(char*)de <= limit; de = ext2_next_entry(de)) {
 345                         if (de->inode) {
 346                                 int over;
 347
 348                                 rc = 0; /* no error if we return something */
 349
 350                                 offset = (char *)de - kaddr;
 351                                 over = filldir(dirent, de->name, de->name_len,
 352                                                (n << CFS_PAGE_SHIFT) | offset,
 353                                                le32_to_cpu(de->inode),
 354                                                types[de->file_type &
 355                                                      (EXT2_FT_MAX - 1)]);
 356                                 if (over) {
 357                                         ext2_put_page(page);
 358                                         GOTO(done, rc);
 359                                 }
 360                         }
 361                 }
 362                 ext2_put_page(page);
 363         }
 364
 365 done:
 366         filp->f_pos = (n << CFS_PAGE_SHIFT) | offset;
 367         filp->f_version = inode->i_version;
 368         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 369
 370         RETURN(rc);
 371 }
 372
 373 #define QCTL_COPY(out, in)              \
 374 do {                                    \
 375         Q_COPY(out, in, qc_cmd);        \
 376         Q_COPY(out, in, qc_type);       \
 377         Q_COPY(out, in, qc_id);         \
 378         Q_COPY(out, in, qc_stat);       \
 379         Q_COPY(out, in, qc_dqinfo);     \
 380         Q_COPY(out, in, qc_dqblk);      \
 381 } while (0)
 382
 383 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump)
 384 {
 385         struct ll_sb_info *sbi = ll_i2sbi(inode);
 386         struct mdc_op_data data;
 387         struct ptlrpc_request *req = NULL;
 388
 389         struct iattr attr = { 0 };
 390         int rc = 0;
 391
 392         /*
 393          * This is coming from userspace, so should be in
 394          * local endian.  But the MDS would like it in little
 395          * endian, so we swab it before we send it.
 396          */
 397         if (lump->lmm_magic != LOV_USER_MAGIC)
 398                 RETURN(-EINVAL);
 399
 400         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC))
 401                 lustre_swab_lov_user_md(lump);
 402
 403         ll_prepare_mdc_op_data(&data, inode, NULL, NULL, 0, 0);
 404
 405         /* swabbing is done in lov_setstripe() on server side */
 406         rc = mdc_setattr(sbi->ll_mdc_exp, &data,
 407                          &attr, lump, sizeof(*lump), NULL, 0, &req);
 408         if (rc) {
 409                 ptlrpc_req_finished(req);
 410                 if (rc != -EPERM && rc != -EACCES)
 411                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 412                 return rc;
 413         }
 414         ptlrpc_req_finished(req);
 415
 416         return rc;
 417
 418 }
 419
 420 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 421                      int *lmm_size, struct ptlrpc_request **request)
 422 {
 423         struct ll_sb_info *sbi = ll_i2sbi(inode);
 424         struct ll_fid     fid;
 425         struct mds_body   *body;
 426         struct lov_mds_md *lmm = NULL;
 427         struct ptlrpc_request *req = NULL;
 428         int rc, lmmsize;
 429
 430         ll_inode2fid(&fid, inode);
 431
 432         rc = ll_get_max_mdsize(sbi, &lmmsize);
 433         if (rc)
 434                 RETURN(rc);
 435
 436         rc = mdc_getattr(sbi->ll_mdc_exp, &fid,
 437                         OBD_MD_FLEASIZE|OBD_MD_FLDIREA,
 438                         lmmsize, &req);
 439         if (rc < 0) {
 440                 CDEBUG(D_INFO, "mdc_getattr failed on inode "
 441                        "%lu/%u: rc %d\n", inode->i_ino,
 442                        inode->i_generation, rc);
 443                 GOTO(out, rc);
 444         }
 445         body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
 446                         sizeof(*body));
 447         LASSERT(body != NULL); /* checked by mdc_getattr_name */
 448         /* swabbed by mdc_getattr_name */
 449         LASSERT_REPSWABBED(req, REPLY_REC_OFF);
 450
 451         lmmsize = body->eadatasize;
 452
 453         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 454             lmmsize == 0) {
 455                 GOTO(out, rc = -ENODATA);
 456         }
 457
 458         lmm = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1, lmmsize);
 459         LASSERT(lmm != NULL);
 460         LASSERT_REPSWABBED(req, REPLY_REC_OFF + 1);
 461
 462         /*
 463          * This is coming from the MDS, so is probably in
 464          * little endian.  We convert it to host endian before
 465          * passing it to userspace.
 466          */
 467         if (lmm->lmm_magic == __swab32(LOV_MAGIC)) {
 468                 lustre_swab_lov_user_md((struct lov_user_md *)lmm);
 469                 lustre_swab_lov_user_md_objects((struct lov_user_md *)lmm);
 470         }
 471 out:
 472         *lmmp = lmm;
 473         *lmm_size = lmmsize;
 474         *request = req;
 475         return rc;
 476 }
 477
 478 static int ll_dir_ioctl(struct inode *inode, struct file *file,
 479                         unsigned int cmd, unsigned long arg)
 480 {
 481         struct ll_sb_info *sbi = ll_i2sbi(inode);
 482         struct obd_ioctl_data *data;
 483         ENTRY;
 484
 485         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
 486                inode->i_ino, inode->i_generation, inode, cmd);
 487
 488         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
 489         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
 490                 return -ENOTTY;
 491
 492         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_IOCTL);
 493         switch(cmd) {
 494         case EXT3_IOC_GETFLAGS:
 495         case EXT3_IOC_SETFLAGS:
 496                 RETURN(ll_iocontrol(inode, file, cmd, arg));
 497         case EXT3_IOC_GETVERSION_OLD:
 498         case EXT3_IOC_GETVERSION:
 499                 RETURN(put_user(inode->i_generation, (int *)arg));
 500         /* We need to special case any other ioctls we want to handle,
 501          * to send them to the MDS/OST as appropriate and to properly
 502          * network encode the arg field.
 503         case EXT3_IOC_SETVERSION_OLD:
 504         case EXT3_IOC_SETVERSION:
 505         */
 506         case IOC_MDC_LOOKUP: {
 507                 struct ptlrpc_request *request = NULL;
 508                 struct ll_fid fid;
 509                 char *buf = NULL;
 510                 char *filename;
 511                 int namelen, rc, len = 0;
 512
 513                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 514                 if (rc)
 515                         RETURN(rc);
 516                 data = (void *)buf;
 517
 518                 filename = data->ioc_inlbuf1;
 519                 namelen = data->ioc_inllen1;
 520
 521                 if (namelen < 1) {
 522                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
 523                         GOTO(out, rc = -EINVAL);
 524                 }
 525
 526                 ll_inode2fid(&fid, inode);
 527                 rc = mdc_getattr_name(sbi->ll_mdc_exp, &fid, filename, namelen,
 528                                       OBD_MD_FLID, 0, &request);
 529                 if (rc < 0) {
 530                         CDEBUG(D_INFO, "mdc_getattr_name: %d\n", rc);
 531                         GOTO(out, rc);
 532                 }
 533
 534                 ptlrpc_req_finished(request);
 535
 536                 EXIT;
 537         out:
 538                 obd_ioctl_freedata(buf, len);
 539                 return rc;
 540         }
 541         case LL_IOC_LOV_SETSTRIPE: {
 542                 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
 543                 int rc = 0;
 544
 545                 LASSERT(sizeof(lum) == sizeof(*lump));
 546                 LASSERT(sizeof(lum.lmm_objects[0]) ==
 547                         sizeof(lump->lmm_objects[0]));
 548                 rc = copy_from_user(&lum, lump, sizeof(lum));
 549                 if (rc)
 550                         return(-EFAULT);
 551
 552                 rc = ll_dir_setstripe(inode, &lum);
 553
 554                 return rc;
 555         }
 556         case LL_IOC_OBD_STATFS:
 557                 RETURN(ll_obd_statfs(inode, (void *)arg));
 558         case LL_IOC_LOV_GETSTRIPE:
 559         case LL_IOC_MDC_GETINFO:
 560         case IOC_MDC_GETFILEINFO:
 561         case IOC_MDC_GETFILESTRIPE: {
 562                 struct ptlrpc_request *request = NULL;
 563                 struct mds_body *body;
 564                 struct lov_user_md *lump;
 565                 struct lov_mds_md *lmm = NULL;
 566                 char *filename = NULL;
 567                 int rc, lmmsize;
 568
 569                 if (cmd == IOC_MDC_GETFILEINFO ||
 570                     cmd == IOC_MDC_GETFILESTRIPE) {
 571                         filename = getname((const char *)arg);
 572                         if (IS_ERR(filename))
 573                                 RETURN(PTR_ERR(filename));
 574
 575                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
 576                                                       &lmmsize, &request);
 577                 } else {
 578                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
 579                 }
 580
 581                 if (request) {
 582                         body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF,
 583                                               sizeof(*body));
 584                         LASSERT(body != NULL); /* checked by mdc_getattr_name */
 585                         /* swabbed by mdc_getattr_name */
 586                         LASSERT_REPSWABBED(request, REPLY_REC_OFF);
 587                 } else {
 588                         GOTO(out_req, rc);
 589                 }
 590
 591                 if (rc < 0) {
 592                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
 593                                                cmd == LL_IOC_MDC_GETINFO))
 594                                 GOTO(skip_lmm, rc = 0);
 595                         else
 596                                 GOTO(out_req, rc);
 597                 }
 598
 599                 if (cmd == IOC_MDC_GETFILESTRIPE ||
 600                     cmd == LL_IOC_LOV_GETSTRIPE) {
 601                         lump = (struct lov_user_md *)arg;
 602                 } else {
 603                         struct lov_user_mds_data *lmdp;
 604                         lmdp = (struct lov_user_mds_data *)arg;
 605                         lump = &lmdp->lmd_lmm;
 606                 }
 607                 rc = copy_to_user(lump, lmm, lmmsize);
 608                 if (rc)
 609                         GOTO(out_lmm, rc = -EFAULT);
 610         skip_lmm:
 611                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
 612                         struct lov_user_mds_data *lmdp;
 613                         lstat_t st = { 0 };
 614
 615                         st.st_dev     = inode->i_sb->s_dev;
 616                         st.st_mode    = body->mode;
 617                         st.st_nlink   = body->nlink;
 618                         st.st_uid     = body->uid;
 619                         st.st_gid     = body->gid;
 620                         st.st_rdev    = body->rdev;
 621                         st.st_size    = body->size;
 622                         st.st_blksize = CFS_PAGE_SIZE;
 623                         st.st_blocks  = body->blocks;
 624                         st.st_atime   = body->atime;
 625                         st.st_mtime   = body->mtime;
 626                         st.st_ctime   = body->ctime;
 627                         st.st_ino     = body->ino;
 628
 629                         lmdp = (struct lov_user_mds_data *)arg;
 630                         rc = copy_to_user(&lmdp->lmd_st, &st, sizeof(st));
 631                         if (rc)
 632                                 GOTO(out_lmm, rc = -EFAULT);
 633                 }
 634
 635                 EXIT;
 636         out_lmm:
 637                 if (lmm && lmm->lmm_magic == LOV_MAGIC_JOIN)
 638                         OBD_FREE(lmm, lmmsize);
 639         out_req:
 640                 ptlrpc_req_finished(request);
 641                 if (filename)
 642                         putname(filename);
 643                 return rc;
 644         }
 645         case IOC_LOV_GETINFO: {
 646                 struct lov_user_mds_data *lumd;
 647                 struct lov_stripe_md *lsm;
 648                 struct lov_user_md *lum;
 649                 struct lov_mds_md *lmm;
 650                 int lmmsize;
 651                 lstat_t st;
 652                 int rc;
 653
 654                 lumd = (struct lov_user_mds_data *)arg;
 655                 lum = &lumd->lmd_lmm;
 656
 657                 rc = ll_get_max_mdsize(sbi, &lmmsize);
 658                 if (rc)
 659                         RETURN(rc);
 660
 661                 OBD_ALLOC(lmm, lmmsize);
 662                 rc = copy_from_user(lmm, lum, lmmsize);
 663                 if (rc)
 664                         GOTO(free_lmm, rc = -EFAULT);
 665
 666                 rc = obd_unpackmd(sbi->ll_osc_exp, &lsm, lmm, lmmsize);
 667                 if (rc < 0)
 668                         GOTO(free_lmm, rc = -ENOMEM);
 669
 670                 rc = obd_checkmd(sbi->ll_osc_exp, sbi->ll_mdc_exp, lsm);
 671                 if (rc)
 672                         GOTO(free_lsm, rc);
 673
 674                 /* Perform glimpse_size operation. */
 675                 memset(&st, 0, sizeof(st));
 676
 677                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
 678                 if (rc)
 679                         GOTO(free_lsm, rc);
 680
 681                 rc = copy_to_user(&lumd->lmd_st, &st, sizeof(st));
 682                 if (rc)
 683                         GOTO(free_lsm, rc = -EFAULT);
 684
 685                 EXIT;
 686         free_lsm:
 687                 obd_free_memmd(sbi->ll_osc_exp, &lsm);
 688         free_lmm:
 689                 OBD_FREE(lmm, lmmsize);
 690                 return rc;
 691         }
 692         case OBD_IOC_LLOG_CATINFO: {
 693                 struct ptlrpc_request *req = NULL;
 694                 char *buf = NULL;
 695                 int rc, len = 0;
 696                 char *bufs[3] = { NULL }, *str;
 697                 int lens[3] = { sizeof(struct ptlrpc_body) };
 698                 int size[2] = { sizeof(struct ptlrpc_body) };
 699
 700                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 701                 if (rc)
 702                         RETURN(rc);
 703                 data = (void *)buf;
 704
 705                 if (!data->ioc_inlbuf1) {
 706                         obd_ioctl_freedata(buf, len);
 707                         RETURN(-EINVAL);
 708                 }
 709
 710                 lens[REQ_REC_OFF] = data->ioc_inllen1;
 711                 bufs[REQ_REC_OFF] = data->ioc_inlbuf1;
 712                 if (data->ioc_inllen2) {
 713                         lens[REQ_REC_OFF + 1] = data->ioc_inllen2;
 714                         bufs[REQ_REC_OFF + 1] = data->ioc_inlbuf2;
 715                 } else {
 716                         lens[REQ_REC_OFF + 1] = 0;
 717                         bufs[REQ_REC_OFF + 1] = NULL;
 718                 }
 719
 720                 req = ptlrpc_prep_req(sbi2mdc(sbi)->cl_import,
 721                                       LUSTRE_LOG_VERSION, LLOG_CATINFO, 3, lens,
 722                                       bufs);
 723                 if (!req)
 724                         GOTO(out_catinfo, rc = -ENOMEM);
 725
 726                 size[REPLY_REC_OFF] = data->ioc_plen1;
 727                 ptlrpc_req_set_repsize(req, 2, size);
 728
 729                 rc = ptlrpc_queue_wait(req);
 730                 str = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF,
 731                                         data->ioc_plen1);
 732                 if (!rc)
 733                         rc = copy_to_user(data->ioc_pbuf1, str,data->ioc_plen1);
 734                 ptlrpc_req_finished(req);
 735         out_catinfo:
 736                 obd_ioctl_freedata(buf, len);
 737                 RETURN(rc);
 738         }
 739         case OBD_IOC_QUOTACHECK: {
 740                 struct obd_quotactl *oqctl;
 741                 int rc, error = 0;
 742
 743                 if (!capable(CAP_SYS_ADMIN))
 744                         RETURN(-EPERM);
 745
 746                 OBD_ALLOC_PTR(oqctl);
 747                 if (!oqctl)
 748                         RETURN(-ENOMEM);
 749                 oqctl->qc_type = arg;
 750                 rc = obd_quotacheck(sbi->ll_mdc_exp, oqctl);
 751                 if (rc < 0) {
 752                         CDEBUG(D_INFO, "mdc_quotacheck failed: rc %d\n", rc);
 753                         error = rc;
 754                 }
 755
 756                 rc = obd_quotacheck(sbi->ll_osc_exp, oqctl);
 757                 if (rc < 0)
 758                         CDEBUG(D_INFO, "osc_quotacheck failed: rc %d\n", rc);
 759
 760                 OBD_FREE_PTR(oqctl);
 761                 return error ?: rc;
 762         }
 763         case OBD_IOC_POLL_QUOTACHECK: {
 764                 struct if_quotacheck *check;
 765                 int rc;
 766
 767                 if (!capable(CAP_SYS_ADMIN))
 768                         RETURN(-EPERM);
 769
 770                 OBD_ALLOC_PTR(check);
 771                 if (!check)
 772                         RETURN(-ENOMEM);
 773
 774                 rc = obd_iocontrol(cmd, sbi->ll_mdc_exp, 0, (void *)check,
 775                                    NULL);
 776                 if (rc) {
 777                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
 778                         if (copy_to_user((void *)arg, check, sizeof(*check)))
 779                                 rc = -EFAULT;
 780                         GOTO(out_poll, rc);
 781                 }
 782
 783                 rc = obd_iocontrol(cmd, sbi->ll_osc_exp, 0, (void *)check,
 784                                    NULL);
 785                 if (rc) {
 786                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
 787                         if (copy_to_user((void *)arg, check, sizeof(*check)))
 788                                 rc = -EFAULT;
 789                         GOTO(out_poll, rc);
 790                 }
 791         out_poll:
 792                 OBD_FREE_PTR(check);
 793                 RETURN(rc);
 794         }
 795 #ifdef HAVE_QUOTA_SUPPORT
 796         case OBD_IOC_QUOTACTL: {
 797                 struct if_quotactl *qctl;
 798                 struct obd_quotactl *oqctl;
 799
 800                 int cmd, type, id, rc = 0;
 801
 802                 OBD_ALLOC_PTR(qctl);
 803                 if (!qctl)
 804                         RETURN(-ENOMEM);
 805
 806                 OBD_ALLOC_PTR(oqctl);
 807                 if (!oqctl) {
 808                         OBD_FREE_PTR(qctl);
 809                         RETURN(-ENOMEM);
 810                 }
 811                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
 812                         GOTO(out_quotactl, rc = -EFAULT);
 813
 814                 cmd = qctl->qc_cmd;
 815                 type = qctl->qc_type;
 816                 id = qctl->qc_id;
 817                 switch (cmd) {
 818                 case Q_QUOTAON:
 819                 case Q_QUOTAOFF:
 820                 case Q_SETQUOTA:
 821                 case Q_SETINFO:
 822                         if (!capable(CAP_SYS_ADMIN))
 823                                 GOTO(out_quotactl, rc = -EPERM);
 824                         break;
 825                 case Q_GETQUOTA:
 826                         if (((type == USRQUOTA && current->euid != id) ||
 827                              (type == GRPQUOTA && !in_egroup_p(id))) &&
 828                             !capable(CAP_SYS_ADMIN))
 829                                 GOTO(out_quotactl, rc = -EPERM);
 830
 831                         /* XXX: dqb_valid is borrowed as a flag to mark that
 832                          *      only mds quota is wanted */
 833                         if (qctl->qc_dqblk.dqb_valid)
 834                                 qctl->obd_uuid = sbi->ll_mdc_exp->exp_obd->
 835                                                         u.cli.cl_target_uuid;
 836                         break;
 837                 case Q_GETINFO:
 838                         break;
 839                 default:
 840                         CERROR("unsupported quotactl op: %#x\n", cmd);
 841                         GOTO(out_quotactl, -ENOTTY);
 842                 }
 843
 844                 QCTL_COPY(oqctl, qctl);
 845
 846                 if (qctl->obd_uuid.uuid[0]) {
 847                         struct obd_device *obd;
 848                         struct obd_uuid *uuid = &qctl->obd_uuid;
 849
 850                         obd = class_find_client_notype(uuid,
 851                                          &sbi->ll_osc_exp->exp_obd->obd_uuid);
 852                         if (!obd)
 853                                 GOTO(out_quotactl, rc = -ENOENT);
 854
 855                         if (cmd == Q_GETINFO)
 856                                 oqctl->qc_cmd = Q_GETOINFO;
 857                         else if (cmd == Q_GETQUOTA)
 858                                 oqctl->qc_cmd = Q_GETOQUOTA;
 859                         else
 860                                 GOTO(out_quotactl, rc = -EINVAL);
 861
 862                         if (sbi->ll_mdc_exp->exp_obd == obd) {
 863                                 rc = obd_quotactl(sbi->ll_mdc_exp, oqctl);
 864                         } else {
 865                                 int i;
 866                                 struct obd_export *exp;
 867                                 struct lov_obd *lov = &sbi->ll_osc_exp->
 868                                                             exp_obd->u.lov;
 869
 870                                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 871                                         if (!lov->lov_tgts[i] ||
 872                                             !lov->lov_tgts[i]->ltd_active)
 873                                                 continue;
 874                                         exp = lov->lov_tgts[i]->ltd_exp;
 875                                         if (exp->exp_obd == obd) {
 876                                                 rc = obd_quotactl(exp, oqctl);
 877                                                 break;
 878                                         }
 879                                 }
 880                         }
 881
 882                         oqctl->qc_cmd = cmd;
 883                         QCTL_COPY(qctl, oqctl);
 884
 885                         if (copy_to_user((void *)arg, qctl, sizeof(*qctl)))
 886                                 rc = -EFAULT;
 887
 888                         GOTO(out_quotactl, rc);
 889                 }
 890
 891                 rc = obd_quotactl(sbi->ll_mdc_exp, oqctl);
 892                 if (rc && rc != -EBUSY && cmd == Q_QUOTAON) {
 893                         oqctl->qc_cmd = Q_QUOTAOFF;
 894                         obd_quotactl(sbi->ll_mdc_exp, oqctl);
 895                 }
 896
 897                 QCTL_COPY(qctl, oqctl);
 898
 899                 if (copy_to_user((void *)arg, qctl, sizeof(*qctl)))
 900                         rc = -EFAULT;
 901         out_quotactl:
 902                 OBD_FREE_PTR(qctl);
 903                 OBD_FREE_PTR(oqctl);
 904                 RETURN(rc);
 905         }
 906 #endif /* HAVE_QUOTA_SUPPORT */
 907         case OBD_IOC_GETNAME: {
 908                 struct obd_device *obd = class_exp2obd(sbi->ll_osc_exp);
 909                 if (!obd)
 910                         RETURN(-EFAULT);
 911                 if (copy_to_user((void *)arg, obd->obd_name,
 912                                 strlen(obd->obd_name) + 1))
 913                         RETURN (-EFAULT);
 914                 RETURN(0);
 915         }
 916         default:
 917                 RETURN(obd_iocontrol(cmd, sbi->ll_osc_exp,0,NULL,(void *)arg));
 918         }
 919 }
 920
 921 int ll_dir_open(struct inode *inode, struct file *file)
 922 {
 923         ENTRY;
 924         RETURN(ll_file_open(inode, file));
 925 }
 926
 927 int ll_dir_release(struct inode *inode, struct file *file)
 928 {
 929         ENTRY;
 930         RETURN(ll_file_release(inode, file));
 931 }
 932
 933 struct file_operations ll_dir_operations = {
 934         .open     = ll_dir_open,
 935         .release  = ll_dir_release,
 936         .read     = generic_read_dir,
 937         .readdir  = ll_readdir,
 938         .ioctl    = ll_dir_ioctl
 939 };
 940