lustre/lvfs/fsfilt_snap_ext3.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Lustre filesystem abstraction routines
   5  *
   6  *  Copyright (C) 2002, 2003 Cluster File Systems, Inc.
   7  *   Author: Andreas Dilger <adilger@clusterfs.com>
   8  *
   9  *   This file is part of Lustre, http://www.lustre.org.
  10  *
  11  *   Lustre is free software; you can redistribute it and/or
  12  *   modify it under the terms of version 2 of the GNU General Public
  13  *   License as published by the Free Software Foundation.
  14  *
  15  *   Lustre is distributed in the hope that it will be useful,
  16  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  *   GNU General Public License for more details.
  19  *
  20  *   You should have received a copy of the GNU General Public License
  21  *   along with Lustre; if not, write to the Free Software
  22  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  23  */
  24 #define DEBUG_SUBSYSTEM S_FILTER
  25
  26 #include <linux/init.h>
  27 #include <linux/module.h>
  28 #include <linux/fs.h>
  29 #include <linux/jbd.h>
  30 #include <linux/slab.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/quotaops.h>
  33 #include <linux/ext3_fs.h>
  34 #include <linux/ext3_jbd.h>
  35 #include <linux/ext3_extents.h>
  36 #include <linux/locks.h>
  37 #include <linux/version.h>
  38 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  39 #include <linux/ext3_xattr.h>
  40 #else
  41 #include <ext3/xattr.h>
  42 #endif
  43
  44 #include <linux/kp30.h>
  45 #include <linux/lustre_fsfilt.h>
  46 #include <linux/obd.h>
  47 #include <linux/obd_class.h>
  48 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  49 #include <linux/module.h>
  50 #include <linux/iobuf.h>
  51 #endif
  52 #include <linux/lustre_snap.h>
  53
  54 /* For snapfs in EXT3 flags --- FIXME will find other ways to store it*/
  55 #define EXT3_COW_FL                     0x00100000 /* inode is snapshot cow */
  56 #define EXT3_DEL_FL                     0x00200000 /* inode is deleting in snapshot */
  57
  58 #define EXT3_SNAP_ATTR "@snap"
  59 #define EXT3_SNAP_GENERATION "@snap_generation"
  60 #define EXT3_MAX_SNAPS 20
  61 #define EXT3_MAX_SNAP_DATA (sizeof(struct snap_ea))
  62 #define EXT3_SNAP_INDEX EXT3_XATTR_INDEX_LUSTRE
  63
  64 #define SB_SNAPTABLE_INO(sb)   (EXT3_SB(sb)->s_es->s_snaptable_ino)
  65 #define SB_FEATURE_COMPAT(sb)  (EXT3_SB(sb)->s_es->s_feature_compat)
  66
  67 #define SNAP_HAS_COMPAT_FEATURE(sb,mask)        \
  68         (SB_FEATURE_COMPAT(sb) & cpu_to_le32(mask))
  69
  70 #define EXT3_FEATURE_COMPAT_SNAPFS             0x0010
  71 #define EXT3_FEATURE_COMPAT_BLOCKCOW           0x0020
  72 /*snaptable info for EXT3*/
  73 #define EXT3_SNAPTABLE_EA       "@snaptable"
  74
  75 /* NOTE: these macros are close dependant on the structure of snap ea */
  76 #define SNAP_CNT_FROM_SIZE(size)       ((((size)-sizeof(ino_t)*2)/2)/sizeof(ino_t))
  77 #define SNAP_EA_SIZE_FROM_INDEX(index) (sizeof(ino_t)*2 + 2*sizeof(ino_t)*((index)+1))
  78
  79 #define SNAP_EA_INO_BLOCK_SIZE(size)   (((size)-sizeof(ino_t)*2)/2)
  80 #define SNAP_EA_PARENT_OFFSET(size)    (sizeof(ino_t)*2 + SNAP_EA_INO_BLOCK_SIZE((size)))
  81
  82 /* helper functions to manipulate field 'parent' in snap_ea */
  83 static inline int
  84 set_parent_ino(struct snap_ea *pea, int size, int index, ino_t val)
  85 {
  86        char * p = (char*) pea;
  87        int offset;
  88
  89        offset = sizeof(ino_t)*2 + (size - sizeof(ino_t)*2)/2;
  90        offset += sizeof(ino_t) * index;
  91        *(ino_t*)(p+offset) = val;
  92
  93        return 0;
  94 }
  95 /**
  96  * fsfilt_ext3_get_indirect - get a specific indirect inode from a primary inode
  97  * @primary: primary (direct) inode
  98  * @table: table of @slot + 1 indices in reverse chronological order
  99  * @slot: starting slot number to check for indirect inode number
 100  *
 101  * We locate an indirect inode from a primary inode using the redirection
 102  * table stored in the primary inode.  Because the desired inode may actually
 103  * be in a "newer" slot number than the supplied slot, we are given a table
 104  * of indices in chronological order to search for the correct inode number.
 105  * We walk table from @slot to 0 looking for a non-zero inode to load.
 106  *
 107  * To only load a specific index (and fail if it does not exist), you can
 108  * pass @table = NULL, and the index number in @slot.  If @slot == 0, the
 109  * primary inode data is returned.
 110  *
 111  * We return a pointer to an inode, or an error.  If the indirect inode for
 112  * the given index does not exist, NULL is returned.
 113  */
 114 static struct inode *fsfilt_ext3_get_indirect(struct inode *primary, int *table,
 115                                               int slot)
 116 {
 117         char buf[EXT3_MAX_SNAP_DATA];
 118         struct snap_ea *snaps;
 119         ino_t ino;
 120         struct inode *inode = NULL;
 121         int rc = 0, index = 0;
 122
 123         ENTRY;
 124
 125         if (slot < 0 || slot > EXT3_MAX_SNAPS || !primary)
 126                 RETURN(NULL);
 127
 128         CDEBUG(D_INODE, "ino %lu, table %p, slot %d\n", primary->i_ino, table,
 129                slot);
 130         rc = ext3_xattr_get(primary, EXT3_SNAP_INDEX, EXT3_SNAP_ATTR, buf,
 131                              EXT3_MAX_SNAP_DATA);
 132         if (rc == -ENODATA) {
 133                 slot = -1;
 134         } else if (rc < 0) {
 135                 CERROR("attribute read rc=%d \n", rc);
 136                 RETURN(NULL);
 137         }
 138         snaps = (struct snap_ea *)buf;
 139
 140         /* if table is NULL and there is a slot */
 141         if( !table && slot >= 0) {
 142                 index = slot;
 143                 ino = le32_to_cpu(snaps->ino[index]);
 144                 if(ino)
 145                         inode = iget(primary->i_sb, ino);
 146                 GOTO(err_free, rc);
 147         }
 148         /* if table is not NULL */
 149         while (!inode && slot >= 0 && table) {
 150                 index = table[slot];
 151                 ino = le32_to_cpu(snaps->ino[index]);
 152
 153                 CDEBUG(D_INODE, "snap inode at slot %d is %lu\n", slot, ino);
 154                 if (!ino) {
 155                         --slot;
 156                         continue;
 157                 }
 158                 inode = iget(primary->i_sb, ino);
 159                 GOTO(err_free, rc);
 160         }
 161         if( slot == -1 && table ) {
 162                 CDEBUG(D_INODE, "redirector not found, using primary\n");
 163                 inode = iget(primary->i_sb, primary->i_ino);
 164         }
 165 err_free:
 166         RETURN(inode);
 167 }
 168
 169 /* Save the indirect inode in the snapshot table of the primary inode. */
 170 static int fsfilt_ext3_set_indirect(struct inode *pri, int index, ino_t ind_ino,
 171                                     ino_t parent_ino )
 172 {
 173         char buf[EXT3_MAX_SNAP_DATA];
 174         struct snap_ea *snaps;
 175         int err = 0, inlist = 1;
 176         int ea_size;
 177         handle_t *handle = NULL;
 178         ENTRY;
 179
 180         CDEBUG(D_INODE, "(ino %lu, parent %lu): saving ind %lu to index %d\n",
 181                pri->i_ino, parent_ino, ind_ino, index);
 182
 183         if (index < 0 || index > MAX_SNAPS || !pri)
 184                 RETURN(-EINVAL);
 185         /* need lock the list before get_attr() to avoid race */
 186         /* read ea at first */
 187         err = ext3_xattr_get(pri, EXT3_SNAP_INDEX ,EXT3_SNAP_ATTR,
 188                                           buf, EXT3_MAX_SNAP_DATA);
 189         if (err == -ENODATA || err == -ENOATTR) {
 190                 CDEBUG(D_INODE, "no extended attributes - zeroing\n");
 191                 memset(buf, 0, EXT3_MAX_SNAP_DATA);
 192                 /* XXX
 193                  * To judge a inode in list, we only see if it has snap ea.
 194                  * So take care of snap ea of primary inodes very carefully.
 195                  * Is it right in snapfs EXT3, check it later?
 196                  */
 197                 inlist = 0;
 198         } else if (err < 0 || err > EXT3_MAX_SNAP_DATA) {
 199                 GOTO(out_unlock, err);
 200         }
 201
 202         handle = ext3_journal_start(pri, SNAP_SETIND_TRANS_BLOCKS);
 203         if(!handle)
 204                 GOTO(out_unlock, err = PTR_ERR(handle));
 205
 206         snaps = (struct snap_ea *)buf;
 207         snaps->ino[index] = cpu_to_le32 (ind_ino);
 208         ea_size = EXT3_MAX_SNAP_DATA;
 209
 210         set_parent_ino(snaps, ea_size, index, cpu_to_le32(parent_ino));
 211
 212         err = ext3_xattr_set(handle, pri, EXT3_SNAP_INDEX, EXT3_SNAP_ATTR,
 213                                      buf, EXT3_MAX_SNAP_DATA, 0);
 214         ext3_mark_inode_dirty(handle, pri);
 215         ext3_journal_stop(handle, pri);
 216 out_unlock:
 217         return err;
 218 }
 219
 220 static int ext3_set_generation(struct inode *inode, unsigned long gen)
 221 {
 222         handle_t *handle;
 223         int err = 0;
 224         ENTRY;
 225
 226         handle = ext3_journal_start(inode, EXT3_XATTR_TRANS_BLOCKS);
 227         if( !handle )
 228                 RETURN(-EINVAL);
 229
 230         err = ext3_xattr_set(handle, inode, EXT3_SNAP_INDEX,
 231                              EXT3_SNAP_GENERATION,
 232                              (char*)&gen, sizeof(int), 0);
 233         if (err < 0) {
 234                 CERROR("ino %lu, set_ext_attr err %d\n", inode->i_ino, err);
 235                 RETURN(err);
 236         }
 237
 238         ext3_journal_stop(handle, inode);
 239         RETURN(0);
 240 }
 241
 242 /*
 243  * Copy inode metadata from one inode to another, excluding blocks and size.
 244  * FIXME do we copy EA data - ACLs and such (excluding snapshot data)?
 245  */
 246 static void ext3_copy_meta(handle_t *handle, struct inode *dst, struct inode *src)
 247 {
 248         int size;
 249
 250         dst->i_mode = src->i_mode;
 251         dst->i_nlink = src->i_nlink;
 252         dst->i_uid = src->i_uid;
 253         dst->i_gid = src->i_gid;
 254         dst->i_atime = src->i_atime;
 255         dst->i_mtime = src->i_mtime;
 256         dst->i_ctime = src->i_ctime;
 257 //      dst->i_version = src->i_version;
 258         dst->i_attr_flags = src->i_attr_flags;
 259         dst->i_generation = src->i_generation;
 260         dst->u.ext3_i.i_dtime = src->u.ext3_i.i_dtime;
 261         dst->u.ext3_i.i_flags = src->u.ext3_i.i_flags | EXT3_COW_FL;
 262 #ifdef EXT3_FRAGMENTS
 263         dst->u.ext3_i.i_faddr = src->u.ext3_i.i_faddr;
 264         dst->u.ext3_i.i_frag_no = src->u.ext3_i.i_frag_no;
 265         dst->u.ext3_i.i_frag_size = src->u.ext3_i.i_frag_size;
 266 #endif
 267         if ((size = ext3_xattr_list(src, NULL, 0)) > 0) {
 268                 char names[size];
 269                 char *name;
 270                 int namelen;
 271
 272                 if (ext3_xattr_list(src, names, 0) < 0)
 273                         return;
 274                 /*
 275                  * the list of attribute names are stored as NUL terminated
 276                  * strings, with a double NUL string at the end.
 277                  */
 278                 name = names;
 279                 while ((namelen = strlen(name))) {
 280                         int attrlen;
 281                         char *buf;
 282
 283                         /* don't copy snap data */
 284                         if (!strcmp(name, EXT3_SNAP_ATTR)) {
 285                                 CDEBUG(D_INFO, "skipping %s item\n", name);
 286                                 continue;
 287                         }
 288                         CDEBUG(D_INODE, "copying %s item\n", name);
 289                         attrlen = ext3_xattr_get(src, EXT3_SNAP_INDEX,
 290                                                  EXT3_SNAP_ATTR, NULL, 0);
 291                         if (attrlen < 0)
 292                                 continue;
 293                         OBD_ALLOC(buf, attrlen);
 294                                 break;
 295                         if (!buf) {
 296                                 CERROR("No MEM\n");
 297                                 break;
 298                         }
 299                         if (ext3_xattr_get(src, EXT3_SNAP_INDEX,
 300                                            EXT3_SNAP_ATTR, buf, attrlen) < 0)
 301                                 continue;
 302                         if (ext3_xattr_set(handle, dst, EXT3_SNAP_INDEX,
 303                                            EXT3_SNAP_ATTR, buf, attrlen, 0) < 0)
 304                                 break;
 305                         OBD_FREE(buf, attrlen);
 306                         name += namelen + 1; /* skip name and trailing NUL */
 307                 }
 308         }
 309 }
 310 static int ext3_copy_reg_block(struct inode *dst, struct inode *src, int blk)
 311 {
 312         struct page     *src_page, *dst_page;
 313         loff_t          offset = blk << src->i_sb->s_blocksize_bits;
 314         unsigned long   index = offset >> PAGE_CACHE_SHIFT;
 315         int             rc = 0;
 316         ENTRY;
 317
 318         /*read the src page*/
 319         src_page = grab_cache_page(src->i_mapping, index);
 320         if (src_page == NULL)
 321                 RETURN(-ENOMEM);
 322
 323         if (!PageUptodate(src_page)) {
 324                 rc = src->i_mapping->a_ops->readpage(NULL, src_page);
 325                 if (rc < 0) {
 326                         page_cache_release(src_page);
 327                         RETURN(rc);
 328                 }
 329         }
 330         kmap(src_page);
 331         /*get dst page*/
 332
 333         dst_page = grab_cache_page(dst->i_mapping, index);
 334         if (dst_page == NULL)
 335                 GOTO(src_page_unlock, rc = -ENOMEM);
 336         kmap(dst_page);
 337
 338         rc = dst->i_mapping->a_ops->prepare_write(NULL, dst_page, 0,
 339                                                   PAGE_CACHE_SIZE - 1);
 340         if (rc)
 341                 GOTO(dst_page_unlock, rc = -EFAULT);
 342         memcpy(page_address(dst_page), page_address(src_page), PAGE_CACHE_SIZE);
 343
 344         flush_dcache_page(dst_page);
 345
 346         rc = dst->i_mapping->a_ops->commit_write(NULL, dst_page, 0,
 347                                                  PAGE_CACHE_SIZE - 1);
 348         if (!rc)
 349                 rc = 1;
 350 dst_page_unlock:
 351         kunmap(dst_page);
 352         UnlockPage(dst_page);
 353         page_cache_release(dst_page);
 354 src_page_unlock:
 355         kunmap(src_page);
 356         page_cache_release(src_page);
 357         RETURN(rc);
 358 }
 359 static int ext3_copy_dir_block(struct inode *dst, struct inode *src, int blk)
 360 {
 361         struct buffer_head *bh_dst = NULL, *bh_src = NULL;
 362         int rc = 0;
 363         handle_t *handle = NULL;
 364         ENTRY;
 365         handle = ext3_journal_start(dst, SNAP_COPYBLOCK_TRANS_BLOCKS);
 366         if( !handle )
 367                 RETURN(-EINVAL);
 368
 369         bh_src = ext3_bread(handle, src, blk, 0, &rc);
 370         if (!bh_src) {
 371                 CERROR("rcor for src blk %d, rcor %d\n", blk, rc);
 372                 GOTO(exit_relese, rc);
 373         }
 374         bh_dst = ext3_getblk(handle, dst, blk, 1, &rc);
 375         if (!bh_dst) {
 376                 CERROR("rcor for dst blk %d, rcor %d\n", blk, rc);
 377                 GOTO(exit_relese, rc);
 378         }
 379         CDEBUG(D_INODE, "copy block %lu to %lu (%ld bytes)\n",
 380                bh_src->b_blocknr, bh_dst->b_blocknr, src->i_sb->s_blocksize);
 381
 382         ext3_journal_get_write_access(handle, bh_dst);
 383         memcpy(bh_dst->b_data, bh_src->b_data, src->i_sb->s_blocksize);
 384         ext3_journal_dirty_metadata(handle, bh_dst);
 385         rc = 1;
 386
 387 exit_relese:
 388         if (bh_src) brelse(bh_src);
 389         if (bh_dst) brelse(bh_dst);
 390         if (handle)
 391                 ext3_journal_stop(handle, dst);
 392         RETURN(rc);
 393 }
 394 /* fsfilt_ext3_copy_block - copy one data block from inode @src to @dst.
 395    No lock here.  User should do the lock.
 396    User should check the return value to see if the result is correct.
 397    Return value:
 398    1:    The block has been copied successfully
 399    0:    No block is copied, usually this is because src has no such blk
 400   -1:    Error
 401 */
 402
 403 static int fsfilt_ext3_copy_block (struct inode *dst, struct inode *src, int blk)
 404 {
 405         int rc = 0;
 406         ENTRY;
 407         CDEBUG(D_INODE, "copy blk %d from %lu to %lu \n", blk, src->i_ino,
 408                dst->i_ino);
 409         /*
 410          * ext3_getblk() require handle!=NULL
 411          */
 412         if (S_ISREG(src->i_mode)) {
 413                 rc = ext3_copy_reg_block(dst, src, blk);
 414         } else {
 415                 rc = ext3_copy_dir_block(dst, src, blk);
 416         }
 417
 418         RETURN(rc);
 419 }
 420
 421 static inline int ext3_has_ea(struct inode *inode)
 422 {
 423        return (EXT3_I(inode)->i_file_acl != 0);
 424 }
 425 /* XXXThis function has a very bad effect to
 426  * the performance of filesystem,
 427  * will find another way to fix it
 428  */
 429 static void fs_flushinval_pages(handle_t *handle, struct inode* inode)
 430 {
 431         if (inode->i_blocks > 0 && inode->i_mapping) {
 432                 fsync_inode_data_buffers(inode);
 433                 truncate_inode_pages(inode->i_mapping, 0);
 434         }
 435 }
 436 /*  ext3_migrate_data:
 437  *  MOVE all the data blocks from inode src to inode dst as well as
 438  *  COPY all attributes(meta data) from inode src to inode dst.
 439  *  For extended attributes(EA), we COPY all the EAs but skip the Snap EA from
 440  *  src to dst. If the dst has Snap EA, then we CAN'T overwrite it. We CAN'T
 441  *  copy the src Snap EA. XXX for EA, can we change it to MOVE all the EAs
 442  *  (exclude Snap EA) to dst and copy it back to src ? This is for LAN free
 443  *  backup later.
 444  */
 445 static int ext3_migrate_data(handle_t *handle, struct inode *dst,
 446                              struct inode *src)
 447 {
 448         unsigned long err = 0;
 449         /* 512 byte disk blocks per inode block */
 450         int bpib = src->i_sb->s_blocksize >> 9;
 451         ENTRY;
 452
 453
 454         if((!dst) || (!src))
 455                 RETURN(-EINVAL);
 456
 457         if (dst->i_ino == src->i_ino)
 458                 RETURN(0);
 459
 460         fs_flushinval_pages(handle, src);
 461
 462         ext3_copy_meta(handle, dst, src);
 463
 464         CDEBUG(D_INODE, "migrating data blocks from %lu to %lu\n",
 465                src->i_ino, dst->i_ino);
 466         /* Can't check blocks in case of EAs */
 467
 468         memcpy(EXT3_I(dst)->i_data, EXT3_I(src)->i_data,
 469                sizeof(EXT3_I(src)->i_data));
 470         memset(EXT3_I(src)->i_data, 0, sizeof(EXT3_I(src)->i_data));
 471
 472         ext3_discard_prealloc(src);
 473
 474         dst->i_size = EXT3_I(dst)->i_disksize = EXT3_I(src)->i_disksize;
 475         src->i_size = EXT3_I(src)->i_disksize = 0;
 476
 477         dst->i_blocks = src->i_blocks;
 478         src->i_blocks = 0;
 479         /*  Check EA blocks here to modify i_blocks correctly */
 480         if(ext3_has_ea (src)) {
 481                 src->i_blocks += bpib;
 482                 if( ! ext3_has_ea (dst) )
 483                         if( dst->i_blocks >= bpib )
 484                                 dst->i_blocks -= bpib;
 485         } else {
 486                 if( ext3_has_ea (dst))
 487                         dst->i_blocks += bpib;
 488         }
 489
 490         CDEBUG(D_INODE, "migrate data from ino %lu to ino %lu\n", src->i_ino,
 491                dst->i_ino);
 492         ext3_mark_inode_dirty(handle, src);
 493         ext3_mark_inode_dirty(handle, dst);
 494         RETURN(err);
 495 }
 496
 497 static handle_t * ext3_copy_data(handle_t *handle, struct inode *dst,
 498                                  struct inode *src, int *has_orphan)
 499 {
 500         unsigned long blocks, blk, cur_blks;
 501         int low_credits, save_ref;
 502         ENTRY;
 503
 504         blocks =(src->i_size + src->i_sb->s_blocksize-1) >>
 505                  src->i_sb->s_blocksize_bits;
 506         low_credits = handle->h_buffer_credits - SNAP_BIGCOPY_TRANS_BLOCKS;
 507
 508         CDEBUG(D_INODE, "%lu blocks need to be copied,low credits limit %d\n",
 509                blocks, low_credits);
 510
 511         for (blk = 0, cur_blks= dst->i_blocks; blk < blocks; blk++) {
 512                 if (!ext3_bmap(src->i_mapping, blk))
 513                         continue;
 514                 if(handle->h_buffer_credits <= low_credits) {
 515                         int needed = (blocks - blk) * EXT3_DATA_TRANS_BLOCKS;
 516                         if (needed > 4 * SNAP_COPYBLOCK_TRANS_BLOCKS)
 517                                 needed = 4 * SNAP_COPYBLOCK_TRANS_BLOCKS;
 518                         if (journal_extend(handle, needed)) {
 519                                 CDEBUG(D_INFO, "create_indirect:fail to extend "
 520                                        "journal, restart trans\n");
 521
 522                                 if(!*has_orphan) {
 523                                         CDEBUG(D_INODE, "add orphan ino %lu"
 524                                                "nlink %d to orphan list \n",
 525                                                 dst->i_ino, dst->i_nlink);
 526                                         ext3_orphan_add(handle, dst);
 527                                         *has_orphan = 1;
 528                                 }
 529                                 dst->u.ext3_i.i_disksize =
 530                                         blk * dst->i_sb->s_blocksize;
 531                                 dst->i_blocks = cur_blks;
 532                                 dst->i_mtime = CURRENT_TIME;
 533                                 ext3_mark_inode_dirty(handle, dst);
 534                                 /*
 535                                  * We can be sure the last handle was stoped
 536                                  * ONLY if the handle's reference count is 1
 537                                  */
 538                                 save_ref = handle->h_ref;
 539                                 handle->h_ref = 1;
 540                                 if( ext3_journal_stop(handle, dst) ){
 541                                         CERROR("fail to stop journal\n");
 542                                         handle = NULL;
 543                                         break;
 544                                 }
 545                                 handle = ext3_journal_start(dst,
 546                                                 low_credits + needed);
 547                                 if( !handle ){
 548                                         CERROR("fail to restart handle\n");
 549                                         break;
 550                                 }
 551                                 handle->h_ref = save_ref;
 552                         }
 553                 }
 554                 if (fsfilt_ext3_copy_block( dst, src, blk) < 0 )
 555                         break;
 556                 cur_blks += dst->i_sb->s_blocksize / 512;
 557         }
 558
 559         dst->i_size = dst->u.ext3_i.i_disksize = src->i_size;
 560         RETURN(handle);
 561 }
 562 /*Here delete the data of that pri inode
 563  *FIXME later, should throw the blocks of
 564  *primary inode directly
 565  */
 566 static int ext3_throw_inode_data(handle_t *handle, struct inode *inode)
 567 {
 568         struct inode *tmp = NULL;
 569         ENTRY;
 570
 571         tmp = ext3_new_inode(handle, inode, (int)inode->i_mode, 0);
 572         if(tmp) {
 573                 CERROR("ext3_new_inode error\n");
 574                 RETURN(-EIO);
 575         }
 576         double_down(&inode->i_sem, &tmp->i_sem);
 577         ext3_migrate_data(handle, tmp, inode);
 578         double_up(&inode->i_sem, &tmp->i_sem);
 579         tmp->i_nlink = 0;
 580         iput(tmp);
 581         RETURN(0);
 582 }
 583 /**
 584  * fsfilt_ext3_create_indirect - copy data, attributes from primary to new indir inode
 585  * @pri: primary (source) inode
 586  * @index: index in snapshot table where indirect inode should be stored
 587  * @delete: flag that the primary inode is being deleted
 588  *
 589  * We copy all of the data blocks from the @*src inode to the @*dst inode, as
 590  * well as copying the attributes from @*src to @*dst.  If @delete == 1, then
 591  * the primary inode will only be a redirector and will appear deleted.
 592  *
 593  * FIXME do we move EAs, only non-snap EAs, what?
 594  * FIXME we could do readpage/writepage, but we would have to handle block
 595  *       allocation then, and it ruins sparse files for 1k/2k filesystems,
 596  *       at the expense of doing a memcpy.
 597  */
 598 static struct inode* fsfilt_ext3_create_indirect(struct inode *pri, int index,
 599                                                  unsigned int gen,
 600                                                  struct inode* parent,
 601                                                  int del)
 602 {
 603         struct inode *ind = NULL;
 604         handle_t *handle = NULL;
 605         int err = 0;
 606         int has_orphan = 0;
 607         ENTRY;
 608
 609         if( pri == pri->i_sb->u.ext3_sb.s_journal_inode ){
 610                 CERROR("TRY TO COW JOUNRAL\n");
 611                 RETURN(ERR_PTR(-EINVAL));
 612         }
 613         CDEBUG(D_INODE, "creating indirect inode for %lu at index %d, %s pri\n",
 614                pri->i_ino, index, del ? "deleting" : "preserve");
 615
 616         ind = fsfilt_ext3_get_indirect(pri, NULL, index);
 617
 618         handle = ext3_journal_start(pri, SNAP_CREATEIND_TRANS_BLOCKS);
 619         if( !handle ) {
 620                 CERROR("handle not NULL\n");
 621                 RETURN(ERR_PTR(-EINVAL));
 622         }
 623         /* XXX ? We should pass an err argument to get_indirect and precisely
 624          * detect the errors, for some errors, we should exit right away.
 625          */
 626
 627         /* if the option is SNAP_DEL_PRI_WITH_IND and there is an indirect,
 628          * we just free the primary data blocks and mark this inode delete
 629          */
 630         if((del) && ind && !IS_ERR(ind)) {
 631                 /* for directory, we don't free the data blocks,
 632                  * or ext3_rmdir will report errors "bad dir, no data blocks"
 633                  */
 634                 CDEBUG(D_INODE, "del==SNAP_DEL_PRI_WITH_IND && ind\n");
 635                 if(!S_ISDIR(pri->i_mode)) {
 636                         err = ext3_throw_inode_data(handle, pri);
 637                         if (err)
 638                                 GOTO(exit, err);
 639                         pri->i_nlink = 1;
 640                 }
 641                 pri->u.ext3_i.i_dtime = CURRENT_TIME;
 642                 ext3_mark_inode_dirty(handle, pri);
 643                 GOTO(exit, err=0);
 644         }
 645
 646         if (ind && !IS_ERR(ind)) {
 647                 CDEBUG(D_INODE, "existing indirect ino %lu for %lu: index %d\n",
 648                        ind->i_ino, pri->i_ino, index);
 649
 650                 GOTO(exit, err=0);
 651         }
 652
 653         /* XXX: check this, ext3_new_inode, the first arg should be "dir" */
 654         ind = ext3_new_inode(handle, pri, (int)pri->i_mode, 0);
 655         if (IS_ERR(ind))
 656                 GOTO(exit, err);
 657         CDEBUG(D_INODE, "got new inode %lu\n", ind->i_ino);
 658         ind->i_rdev = pri->i_rdev;
 659         ind->i_op = pri->i_op;
 660
 661         /*init ind ops*/
 662         memcpy(ind->i_op, pri->i_op, sizeof(*pri->i_op));
 663         memcpy(ind->i_fop, pri->i_fop, sizeof(*pri->i_fop));
 664         memcpy(ind->i_mapping->a_ops, pri->i_mapping->a_ops,
 665                sizeof(*pri->i_mapping->a_ops));
 666
 667         ext3_set_generation(ind, (unsigned long)gen);
 668         /* If we are deleting the primary inode, we want to ensure that it is
 669          * written to disk with a non-zero link count, otherwise the next iget
 670          * and iput will mark the inode as free (which we don't want, we want
 671          * it to stay a redirector).  We fix this in ext3_destroy_indirect()
 672          * when the last indirect inode is removed.
 673          *
 674          * We then do what ext3_delete_inode() does so that the metadata will
 675          * appear the same as a deleted inode, and we can detect it later.
 676          */
 677         if (del) {
 678                 CDEBUG(D_INODE, "deleting primary inode\n");
 679
 680                 down(&ind->i_sem);
 681                 err = ext3_migrate_data(handle, ind, pri);
 682                 if (err)
 683                         GOTO(exit_unlock, err);
 684
 685                 err = fsfilt_ext3_set_indirect(pri, index, ind->i_ino, parent->i_ino);
 686                 if (err)
 687                         GOTO(exit_unlock, err);
 688
 689                 /* XXX for directory, we copy the block back
 690                  * or ext3_rmdir will report errors "bad dir, no data blocks"
 691                  */
 692                 if( S_ISDIR(pri->i_mode)) {
 693                         handle = ext3_copy_data(handle, pri, ind, &has_orphan);
 694                         if(!handle)
 695                                 GOTO(exit_unlock, err= -EINVAL);
 696                 }
 697
 698                 pri->u.ext3_i.i_flags |= EXT3_DEL_FL;
 699                 ind->u.ext3_i.i_flags |= EXT3_COW_FL;
 700                 if(S_ISREG(pri->i_mode)) pri->i_nlink = 1;
 701                 pri->u.ext3_i.i_dtime = CURRENT_TIME;
 702                 //pri->u.ext3_i.i_generation++;
 703                 ext3_mark_inode_dirty(handle, pri);
 704                 ext3_mark_inode_dirty(handle, ind);
 705                 up(&ind->i_sem);
 706         } else {
 707                 down(&ind->i_sem);
 708                 err = ext3_migrate_data(handle, ind, pri);
 709                 if (err)
 710                         goto exit_unlock;
 711
 712                 /* for regular files we do blocklevel COW's maybe */
 713                 if (EXT3_HAS_COMPAT_FEATURE(pri->i_sb, EXT3_FEATURE_COMPAT_BLOCKCOW)
 714                     && S_ISREG(pri->i_mode)) {
 715
 716                         CDEBUG(D_INODE, "ino %lu, do block cow\n", pri->i_ino);
 717                         /* because after migrate_data , pri->i_size is 0 */
 718                         pri->i_size = ind->i_size;
 719                 }
 720                 else {
 721                         int bpib = pri->i_sb->s_blocksize >> 9;
 722                         CDEBUG(D_INODE, "ino %lu, do file cow\n", pri->i_ino);
 723
 724                         /* XXX: can we do this better?
 725                          * If it's a fast symlink, we should copy i_data back!
 726                          * The criteria to determine a fast symlink is:
 727                          * 1) it's a link and its i_blocks is 0
 728                          * 2) it's a link and its i_blocks is bpib ( the case
 729                          *    it has been cowed and has ea )
 730                          */
 731                         if( S_ISLNK(ind->i_mode) && ((ind->i_blocks == 0) ||
 732                             (ext3_has_ea(ind) && ind->i_blocks == bpib))) {
 733                                 CDEBUG(D_INODE, "ino %lu is fast symlink\n", pri->i_ino);
 734                                 memcpy(EXT3_I(pri)->i_data, EXT3_I(ind)->i_data,
 735                                        sizeof(EXT3_I(ind)->i_data));
 736                                 pri->i_size = ind->i_size;
 737                         }
 738                         else {
 739                                 handle = ext3_copy_data(handle, pri, ind, &has_orphan);
 740                                 if (!handle)
 741                                         GOTO(exit_unlock, err);
 742                         }
 743                 }
 744                 /* set cow flag for ind */
 745                 ind->u.ext3_i.i_flags |= EXT3_COW_FL;
 746                 pri->u.ext3_i.i_flags &= ~EXT3_COW_FL;
 747
 748                 ext3_mark_inode_dirty(handle, pri);
 749                 ext3_mark_inode_dirty(handle, ind);
 750
 751                 err = fsfilt_ext3_set_indirect(pri, index, ind->i_ino, parent->i_ino);
 752                 if (err)
 753                         GOTO(exit_unlock, err);
 754                 up(&ind->i_sem);
 755         }
 756
 757         if (!EXT3_HAS_COMPAT_FEATURE(pri->i_sb,
 758                                      EXT3_FEATURE_COMPAT_SNAPFS)) {
 759                 lock_super(pri->i_sb);
 760                 ext3_journal_get_write_access(handle, pri->i_sb->u.ext3_sb.s_sbh);
 761                 pri->i_sb->u.ext3_sb.s_es->s_feature_compat |=
 762                         cpu_to_le32(EXT3_FEATURE_COMPAT_SNAPFS);
 763                 ext3_journal_dirty_metadata(handle, pri->i_sb->u.ext3_sb.s_sbh);
 764                 pri->i_sb->s_dirt = 1;
 765                 unlock_super(pri->i_sb);
 766         }
 767         if (has_orphan) {
 768                 CDEBUG(D_INODE, "del %lu nlink %d from orphan list\n",
 769                        ind->i_ino, ind->i_nlink);
 770                 ext3_orphan_del(handle, ind);
 771         }
 772         ext3_journal_stop(handle, pri);
 773
 774         RETURN(ind);
 775
 776 exit_unlock:
 777         up(&ind->i_sem);
 778         ind->i_nlink = 0;
 779 exit:
 780         if (has_orphan) {
 781                 CDEBUG(D_INODE, "del %lu nlink %d from orphan list\n",
 782                        ind->i_ino, ind->i_nlink);
 783                 ext3_orphan_del(handle, ind);
 784         }
 785         iput(ind);
 786         ext3_journal_stop(handle, pri);
 787
 788         RETURN(ERR_PTR(err));
 789 }
 790
 791 static int fsfilt_ext3_snap_feature (struct super_block *sb, int feature, int op) {
 792
 793         int rc = -EINVAL;
 794         handle_t *handle;
 795         ENTRY;
 796
 797         switch (op) {
 798                 case SNAP_SET_FEATURE:
 799                         handle = ext3_journal_start(sb->s_root->d_inode, 1);
 800                         lock_super(sb);
 801                         ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 802                         SB_FEATURE_COMPAT(sb) |= cpu_to_le32(feature);
 803                         sb->s_dirt = 1;
 804                         ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 805                         unlock_super(sb);
 806                         ext3_journal_stop(handle, sb->s_root->d_inode);
 807                         break;
 808                 case SNAP_CLEAR_FEATURE:
 809                         handle = ext3_journal_start(sb->s_root->d_inode, 1);
 810                         lock_super(sb);
 811                         ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
 812                         SB_FEATURE_COMPAT(sb) &= ~cpu_to_le32(feature);
 813                         ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 814                         sb->s_dirt = 1;
 815                         unlock_super(sb);
 816                         ext3_journal_stop(handle, sb->s_root->d_inode);
 817                         break;
 818                 case SNAP_HAS_FEATURE:
 819                         /*FIXME should lock super or not*/
 820                         rc = SNAP_HAS_COMPAT_FEATURE(sb, feature);
 821                         break;
 822                 default:
 823                         break;
 824         }
 825         RETURN(rc);
 826 }
 827 /*
 828  * is_redirector - determines if a primary inode is a redirector
 829  * @inode: primary inode to test
 830  *
 831  * Returns 1 if the inode is a redirector, 0 otherwise.
 832  */
 833 static int fsfilt_ext3_is_redirector(struct inode *inode)
 834 {
 835         int is_redirector = 0;
 836         int rc;
 837         ENTRY;
 838
 839         rc = ext3_xattr_get(inode, EXT3_SNAP_INDEX ,EXT3_SNAP_ATTR,
 840                                           NULL, 0);
 841         if (rc > 0 && rc <= MAX_SNAP_DATA)
 842                 is_redirector = 1;
 843         CDEBUG(D_INODE, "inode %lu %s redirector\n", inode->i_ino,
 844                is_redirector ? "is" : "isn't");
 845         RETURN(is_redirector);
 846 }
 847 /*if it's indirect inode or not */
 848 static int fsfilt_ext3_is_indirect(struct inode *inode)
 849 {
 850         if (EXT3_I(inode)->i_flags |= EXT3_COW_FL)
 851                 return 1;
 852         else
 853                 return 0;
 854 }
 855
 856 /* get the indirect ino at index of the primary inode
 857  * return value:        postive:        indirect ino number
 858  *                      negative or 0:  error
 859  */
 860 static ino_t fsfilt_ext3_get_indirect_ino(struct inode *primary, int index)
 861 {
 862         char buf[EXT3_MAX_SNAP_DATA];
 863         struct snap_ea *snaps;
 864         ino_t ino = 0;
 865         int err;
 866         ENTRY;
 867         if (index < 0 || index > EXT3_MAX_SNAPS || !primary)
 868                 RETURN(0);
 869
 870         err = ext3_xattr_get(primary, EXT3_SNAP_INDEX, EXT3_SNAP_ATTR,
 871                              buf, EXT3_MAX_SNAP_DATA);
 872         if (err == -ENOATTR) {
 873                 GOTO(err_free, ino = -ENOATTR);
 874         } else if (err < 0) {
 875                 CERROR(" attribute read error err=%d\n", err);
 876                 GOTO(err_free, ino = err);
 877         }
 878         snaps = (struct snap_ea *)buf;
 879         ino = le32_to_cpu (snaps->ino[index]);
 880         CDEBUG(D_INODE, "snap ino for %ld at index %d is %lu\n",
 881                primary->i_ino, index, ino);
 882 err_free:
 883         RETURN(ino);
 884 }
 885
 886
 887 /* The following functions are used by destroy_indirect */
 888 #define inode_bmap(inode, nr) (EXT3_I(inode)->i_data[(nr)])
 889 #define inode_setbmap(inode, nr, physical) (EXT3_I(inode)->i_data[(nr)]=(physical))
 890 static inline int block_bmap(struct buffer_head * bh, int nr)
 891 {
 892         int tmp;
 893
 894         if (!bh)
 895                 return 0;
 896         tmp = le32_to_cpu(((u32 *) bh->b_data)[nr]);
 897         brelse (bh);
 898         return tmp;
 899 }
 900
 901 static inline int block_setbmap(handle_t *handle, struct buffer_head * bh,
 902                                  int nr, int physical)
 903 {
 904
 905         if (!bh)
 906                 return 0;
 907         ext3_journal_get_write_access(handle, bh);
 908         ((u32 *) bh->b_data)[nr] = cpu_to_le32(physical);
 909         ext3_journal_dirty_metadata(handle, bh);
 910         brelse (bh);
 911         return 1;
 912 }
 913
 914 static int ext3_migrate_block(handle_t *handle, struct inode * dst,
 915                               struct inode *src, int block)
 916 {
 917         int i1_d=0, i1_s=0, i2_d=0, i2_s=0, i3_d=0, i3_s=0;
 918         int addr_per_block = EXT3_ADDR_PER_BLOCK(src->i_sb);
 919         int addr_per_block_bits = EXT3_ADDR_PER_BLOCK_BITS(src->i_sb);
 920         unsigned long blksz = src->i_sb->s_blocksize;
 921         kdev_t ddev = dst->i_dev;
 922         kdev_t sdev = src->i_dev;
 923         int physical = 0;
 924         ENTRY;
 925
 926         if (block < 0) {
 927                 CWARN("ext3_migrate_block block < 0 %p \n", src->i_sb);
 928                 RETURN(0);
 929         }
 930         if (block >= EXT3_NDIR_BLOCKS + addr_per_block +
 931                 (1 << (addr_per_block_bits * 2)) +
 932                 ((1 << (addr_per_block_bits * 2)) << addr_per_block_bits)) {
 933                 CWARN("ext3_migrate_block block > big %p \n", src->i_sb);
 934                 RETURN(0);
 935         }
 936         /* EXT3_NDIR_BLOCK */
 937         if (block < EXT3_NDIR_BLOCKS) {
 938                 if(inode_bmap(dst, block))
 939                         RETURN(0);
 940                 else {
 941                         if( (physical = inode_bmap(src, block)) ) {
 942                                 inode_setbmap (dst, block, physical);
 943                                 inode_setbmap (src, block, 0);
 944                                 RETURN(1);
 945                         }
 946                         else
 947                                 RETURN(0);
 948                 }
 949         }
 950         /* EXT3_IND_BLOCK */
 951         block -= EXT3_NDIR_BLOCKS;
 952         if (block < addr_per_block) {
 953                 i1_d = inode_bmap (dst, EXT3_IND_BLOCK);
 954                 if (!i1_d) {
 955                         physical = inode_bmap(src, EXT3_IND_BLOCK);
 956                         if( physical ) {
 957                                 inode_setbmap (dst, EXT3_IND_BLOCK, physical);
 958                                 inode_setbmap (src, EXT3_IND_BLOCK, 0);
 959                                 RETURN(1);
 960                         }
 961                         else
 962                                 RETURN(0);
 963                 }
 964                 if(block_bmap(bread(ddev, i1_d, blksz), block))
 965                         RETURN(0);
 966
 967                 i1_s = inode_bmap (src, EXT3_IND_BLOCK);
 968                 if( !i1_s)      RETURN(0);
 969
 970                 physical = block_bmap(bread(sdev, i1_s, blksz), block);
 971
 972                 if( physical) {
 973                         block_setbmap(handle, bread(ddev, i1_d, blksz),block,
 974                                       physical);
 975                         block_setbmap(handle, bread(sdev, i1_s, blksz),block,0);
 976                         RETURN(1);
 977                 }
 978                 else
 979                         RETURN(0);
 980         }
 981         /* EXT3_DIND_BLOCK */
 982         block -= addr_per_block;
 983         if (block < (1 << (addr_per_block_bits * 2))) {
 984                 i1_d = inode_bmap (dst, EXT3_DIND_BLOCK);
 985                 i1_s = inode_bmap (src, EXT3_DIND_BLOCK);
 986                 if (!i1_d) {
 987                         if( (physical = inode_bmap(src, EXT3_DIND_BLOCK)) ) {
 988                                 inode_setbmap (dst, EXT3_DIND_BLOCK, physical);
 989                                 inode_setbmap (src, EXT3_DIND_BLOCK, 0);
 990                                 RETURN(1);
 991                         }
 992                         else
 993                                 RETURN(0);
 994                 }
 995                 i2_d = block_bmap (bread (ddev, i1_d, blksz),
 996                                 block >> addr_per_block_bits);
 997
 998                 if (!i2_d) {
 999
1000                         if(!i1_s)       RETURN(0);
1001
1002                         physical = block_bmap(bread (sdev, i1_s, blksz),
1003                                                block >> addr_per_block_bits);
1004                         if(physical) {
1005                                 block_setbmap(handle, bread (ddev, i1_d,blksz),
1006                                               block >> addr_per_block_bits,
1007                                               physical);
1008                                 block_setbmap(handle, bread (sdev, i1_s,blksz),
1009                                               block >> addr_per_block_bits, 0);
1010                                 RETURN(1);
1011                         }
1012                         else
1013                                 RETURN(0);
1014                 }
1015                 physical = block_bmap(bread (ddev, i2_d, blksz),
1016                                       block & (addr_per_block - 1));
1017                 if(physical)
1018                                 RETURN(0);
1019                 else {
1020                         i2_s =  block_bmap (bread (sdev, i1_s, blksz),
1021                                 block >> addr_per_block_bits);
1022                         if(!i2_s)       RETURN(0);
1023
1024                         physical = block_bmap(bread (sdev, i2_s, blksz),
1025                                    block & (addr_per_block - 1));
1026                         if(physical) {
1027                                 block_setbmap(handle, bread (ddev, i2_d, blksz),
1028                                    block & (addr_per_block - 1), physical);
1029                                 block_setbmap(handle, bread (sdev, i2_s, blksz),
1030                                    block & (addr_per_block - 1), 0);
1031                                 RETURN(1);
1032                         }
1033                         else
1034                                 RETURN(0);
1035                 }
1036
1037         }
1038         /* EXT3_TIND_BLOCK */
1039         block -= (1 << (addr_per_block_bits * 2));
1040         i1_d = inode_bmap (dst, EXT3_TIND_BLOCK);
1041         i1_s = inode_bmap (src, EXT3_TIND_BLOCK);
1042         if (!i1_d) {
1043                 if((physical = inode_bmap(src, EXT3_TIND_BLOCK)) )
1044                         inode_setbmap (dst, EXT3_TIND_BLOCK, physical);
1045                 else
1046                         RETURN(0);
1047         }
1048         i2_d = block_bmap(bread (ddev, i1_d, blksz),
1049                            block >> (addr_per_block_bits * 2));
1050
1051         if(i1_s) i2_s = block_bmap(bread(sdev, i1_s, blksz),
1052                                    block >> (addr_per_block_bits * 2));
1053
1054         if (!i2_d) {
1055                 if( !i1_s)      RETURN(0);
1056
1057                 physical = block_bmap(bread (sdev, i1_s, blksz),
1058                                        block >> (addr_per_block_bits * 2));
1059                 if(physical) {
1060                         block_setbmap(handle, bread (ddev, i1_d, blksz),
1061                                       block >> (addr_per_block_bits * 2), physical);
1062                         block_setbmap(handle, bread (sdev, i1_s, blksz),
1063                                       block >> (addr_per_block_bits * 2), 0);
1064                         RETURN(1);
1065                 }
1066                 else
1067                         RETURN(0);
1068         }
1069         i3_d = block_bmap (bread (ddev, i2_d, blksz),
1070                         (block >> addr_per_block_bits) & (addr_per_block - 1));
1071         if( i2_s) i3_s = block_bmap (bread (sdev, i2_s, blksz),
1072                         (block >> addr_per_block_bits) & (addr_per_block - 1));
1073
1074         if (!i3_d) {
1075                 if (!i2_s)      RETURN(0);
1076                 physical = block_bmap (bread (sdev, i2_s, blksz),
1077                         (block >> addr_per_block_bits) & (addr_per_block - 1));
1078                 if( physical) {
1079                         block_setbmap (handle, bread (ddev, i2_d, blksz),
1080                                        (block >> addr_per_block_bits) &
1081                                        (addr_per_block - 1), physical);
1082                         block_setbmap (handle, bread (sdev, i2_s, blksz),
1083                                        (block >> addr_per_block_bits) &
1084                                        (addr_per_block - 1),0);
1085                         RETURN(1);
1086                 }
1087                 else
1088                         RETURN(0);
1089         }
1090         physical = block_bmap (bread (ddev, i3_d, blksz),
1091                            block & (addr_per_block - 1)) ;
1092         if(physical)
1093                 RETURN(0);
1094         else {
1095                 if(!i3_s)
1096                         RETURN(0);
1097                 physical = block_bmap(bread(sdev, i3_s, blksz),
1098                                       block & (addr_per_block - 1));
1099                 if(physical) {
1100                         block_setbmap (handle, bread (ddev, i3_d, blksz),
1101                                        block & (addr_per_block - 1), physical);
1102                         block_setbmap (handle, bread (sdev, i3_s, blksz),
1103                                        block & (addr_per_block - 1), 0);
1104                         RETURN(1);
1105                 }
1106                 else
1107                         RETURN(0);
1108         }
1109 }
1110
1111 /* Generate i_blocks from blocks for an inode .
1112  * We also calculate EA block here.
1113  */
1114 static unsigned long calculate_i_blocks(struct inode *inode, int blocks)
1115 {
1116         /* 512 byte disk blocks per inode block */
1117         int bpib = inode->i_sb->s_blocksize >> 9;
1118         int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
1119         unsigned long i_blocks = 0;
1120         int i=0, j=0, meta_blocks = 0;
1121         ENTRY;
1122         if(!inode)
1123                 RETURN(0);
1124
1125         if( blocks < 0 ) {
1126                 /* re-calculate blocks here */
1127                 blocks = (inode->i_size + inode->i_sb->s_blocksize-1)
1128                           >> inode->i_sb->s_blocksize_bits;
1129         }
1130
1131         /* calculate data blocks */
1132         for(i = 0; i < blocks; i++) {
1133                 if(ext3_bmap(inode->i_mapping, i))
1134                         i_blocks += bpib;
1135         }
1136         /* calculate meta blocks */
1137         blocks -= EXT3_NDIR_BLOCKS;
1138         if(blocks > 0) {
1139                 meta_blocks++;
1140                 blocks -= addr_per_block;
1141         }
1142         if( blocks > 0 ) meta_blocks++;
1143         i=0;
1144
1145         while( (blocks > 0) && (i < addr_per_block) ) {
1146                 meta_blocks++;
1147                 blocks -= addr_per_block;
1148                 i++;
1149         }
1150
1151         if ( blocks > 0 ) meta_blocks += 2;
1152         i=0; j=0;
1153
1154         while( blocks > 0) {
1155                 meta_blocks++;
1156                 blocks -= addr_per_block;
1157                 i++;
1158                 if(i >= addr_per_block  ) {
1159                         i=0;
1160                         j++;
1161                 }
1162                 if( j >= addr_per_block) {
1163                         j=0;
1164                         meta_blocks++;
1165                 }
1166         }
1167         /* calculate EA blocks */
1168         if(ext3_has_ea(inode))
1169                 meta_blocks++;
1170
1171         i_blocks += meta_blocks * bpib;
1172         CDEBUG(D_INODE, "ino %lu, get i_blocks %lu\n", inode->i_ino, i_blocks);
1173
1174         RETURN(i_blocks);
1175 }
1176
1177 /**
1178  * fsfilt_ext3_destroy_indirect - delete an indirect inode from the table
1179  * @pri: primary inode
1180  * @ind: indirect inode
1181  * @index: index of inode that should be deleted
1182  *
1183  * We delete the @*ind inode, and remove it from the snapshot table.  If @*ind
1184  * is NULL, we use the inode at @index.
1185  */
1186 static int fsfilt_ext3_destroy_indirect(struct inode *pri, int index,
1187                                         struct inode *next_ind)
1188 {
1189         char buf[EXT3_MAX_SNAP_DATA];
1190         struct snap_ea *snaps;
1191         struct inode *ind;
1192         int save = 0, i=0, err = 0;
1193         handle_t *handle=NULL;
1194         time_t ctime;
1195         ENTRY;
1196
1197         if (index < 0 || index > EXT3_MAX_SNAPS)
1198                 RETURN(0);
1199
1200         if( pri == pri->i_sb->u.ext3_sb.s_journal_inode ){
1201                 CERROR("TRY TO DESTROY JOURNAL'S IND\n");
1202                 RETURN(-EINVAL);
1203         }
1204
1205         err = ext3_xattr_get(pri, EXT3_SNAP_INDEX, EXT3_SNAP_ATTR,
1206                              buf, EXT3_MAX_SNAP_DATA);
1207         if (err < 0) {
1208                 CERROR("inode %lu attribute read error\n", pri->i_ino);
1209                 RETURN(err);
1210         }
1211
1212         snaps = (struct snap_ea *)buf;
1213         if ( !snaps->ino[index] ) {
1214                 CERROR("for pri ino %lu, index %d, redirect ino is 0\n",
1215                        pri->i_ino, index);
1216                 RETURN(-EINVAL);
1217         }
1218
1219         CDEBUG(D_INODE, "for pri ino %lu, reading inode %lu at index %d\n",
1220                pri->i_ino, (ulong)le32_to_cpu(snaps->ino[index]), index);
1221
1222         ind = iget(pri->i_sb, le32_to_cpu (snaps->ino[index]));
1223
1224         if ( !ind || IS_ERR(ind) || is_bad_inode(ind))
1225                 RETURN(-EINVAL);
1226
1227         CDEBUG(D_INODE, "iget ind %lu, ref count = %d\n",
1228                ind->i_ino, atomic_read(&ind->i_count));
1229
1230         handle = ext3_journal_start(pri, SNAP_DESTROY_TRANS_BLOCKS);
1231         if (!handle) {
1232                 iput(ind);
1233                 RETURN(-EINVAL);
1234         }
1235         /* if it's block level cow, first copy the blocks back */
1236         if (EXT3_HAS_COMPAT_FEATURE(pri->i_sb, EXT3_FEATURE_COMPAT_BLOCKCOW) &&
1237             S_ISREG(pri->i_mode)) {
1238                 int blocks;
1239
1240                 if (!next_ind) {
1241                         next_ind = pri;
1242                         down(&ind->i_sem);
1243                 } else {
1244                         double_down(&next_ind->i_sem, &ind->i_sem);
1245                 }
1246                 blocks = (next_ind->i_size + next_ind->i_sb->s_blocksize-1)
1247                           >> next_ind->i_sb->s_blocksize_bits;
1248
1249                 CDEBUG(D_INODE, "migrate block back from ino %lu to %lu\n",
1250                        ind->i_ino, next_ind->i_ino);
1251
1252                 for(i = 0; i < blocks; i++) {
1253                         if( ext3_bmap(next_ind->i_mapping, i) )
1254                                 continue;
1255                         if( !ext3_bmap(ind->i_mapping, i) )
1256                                 continue;
1257                         ext3_migrate_block(handle, next_ind, ind, i) ;
1258                 }
1259                 /* Now re-compute the i_blocks */
1260                 /* XXX shall we take care of ind here? probably not */
1261                 next_ind->i_blocks = calculate_i_blocks( next_ind, blocks);
1262                 ext3_mark_inode_dirty(handle, next_ind);
1263
1264                 if (next_ind == pri)
1265                         up(&ind->i_sem);
1266                 else
1267                         double_up(&next_ind->i_sem, &ind->i_sem);
1268
1269         }
1270
1271         CDEBUG(D_INODE, "delete indirect ino %lu\n", ind->i_ino);
1272         CDEBUG(D_INODE, "iput ind %lu, ref count = %d\n", ind->i_ino,
1273                atomic_read(&ind->i_count));
1274
1275         ind->i_nlink = 0;
1276         iput (ind);
1277
1278         snaps->ino[index] = cpu_to_le32(0);
1279         for (i = 0; i < EXT3_MAX_SNAPS; i++)
1280                 save += snaps->ino[i];
1281
1282
1283         /*Should we remove snap feature here*/
1284         /*
1285          * If we are deleting the last indirect inode, and the primary inode
1286          * has already been deleted, then mark the primary for deletion also.
1287          * Otherwise, if we are deleting the last indirect inode remove the
1288          * snaptable from the inode.    XXX
1289          */
1290         if (!save && pri->u.ext3_i.i_dtime) {
1291                 CDEBUG(D_INODE, "deleting primary %lu\n", pri->i_ino);
1292                 pri->i_nlink = 0;
1293                 /* reset err to 0 now */
1294                 err = 0;
1295         } else {
1296                 CDEBUG(D_INODE, "%s redirector table\n",
1297                        save ? "saving" : "deleting");
1298                 /* XXX: since set ea will modify i_ctime of pri,
1299                         so save/restore i_ctime. Need this necessary ? */
1300                 ctime = pri->i_ctime;
1301                 err = ext3_xattr_set(handle, pri, EXT3_SNAP_INDEX, EXT3_SNAP_ATTR,
1302                                      save ? buf : NULL, EXT3_MAX_SNAP_DATA, 0);
1303                 pri->i_ctime = ctime;
1304                 ext3_mark_inode_dirty(handle, pri);
1305         }
1306         ext3_journal_stop(handle, pri);
1307
1308         RETURN(err);
1309 }
1310
1311 /* restore a primary inode with the indirect inode at index */
1312 static int fsfilt_ext3_restore_indirect(struct inode *pri, int index)
1313 {
1314         struct inode *ind;
1315         int err = 0;
1316         handle_t *handle = NULL;
1317         ENTRY;
1318
1319         if (index < 0 || index > EXT3_MAX_SNAPS)
1320                 RETURN(-EINVAL);
1321
1322         if( pri == pri->i_sb->u.ext3_sb.s_journal_inode ){
1323                 CERROR("TRY TO RESTORE JOURNAL\n");
1324                 RETURN(-EINVAL);
1325         }
1326         CDEBUG(D_INODE, "pri ino %lu, index %d\n", pri->i_ino, index);
1327
1328         ind = fsfilt_ext3_get_indirect(pri, NULL, index);
1329
1330         if (!ind)
1331                 RETURN(-EINVAL);
1332
1333         CDEBUG(D_INODE, "restore ino %lu to %lu\n", pri->i_ino, ind->i_ino);
1334
1335         handle = ext3_journal_start(pri, SNAP_RESTORE_TRANS_BLOCKS);
1336         if( !handle )
1337                 RETURN(-EINVAL);
1338         /* first destroy all the data blocks in primary inode */
1339         /* XXX: check this, ext3_new_inode, the first arg should be "dir" */
1340         err = ext3_throw_inode_data(handle, pri);
1341         if (err) {
1342                 CERROR("restore_indirect, new_inode err\n");
1343                 RETURN(err);
1344         }
1345         double_down(&pri->i_sem, &ind->i_sem);
1346         ext3_migrate_data(handle, pri, ind);
1347         pri->u.ext3_i.i_flags &= ~EXT3_COW_FL;
1348         ext3_mark_inode_dirty(handle, pri);
1349         double_up(&pri->i_sem, &ind->i_sem);
1350         iput(ind);
1351
1352         //fsfilt_ext3_destroy_indirect(pri, index);
1353         ext3_journal_stop(handle, pri);
1354
1355         RETURN(err);
1356 }
1357
1358 /**
1359  * ext3_snap_iterate - iterate through all of the inodes
1360  * @sb: filesystem superblock
1361  * @repeat: pointer to function called on each valid inode
1362  * @start: inode to start iterating at
1363  * @priv: private data to the caller/repeat function
1364  *
1365  * If @start is NULL, then we do not return an inode pointer.  If @*start is
1366  * NULL, then we start at the beginning of the filesystem, and iterate over
1367  * all of the inodes in the system.  If @*start is non-NULL, then we start
1368  * iterating at this inode.
1369  *
1370  * We call the repeat function for each inode that is in use.  The repeat
1371  * function must check if this is a redirector (with is_redirector) if it
1372  * only wants to operate on redirector inodes.  If there is an error or
1373  * the repeat function returns non-zero, we return the last inode operated
1374  * on in the @*start parameter.  This allows the caller to restart the
1375  * iteration at this inode if desired, by returning a positive value.
1376  * Negative return values indicate an error.
1377  *
1378  * NOTE we cannot simply traverse the existing filesystem tree from the root
1379  *      inode, as there may be disconnected trees from deleted files/dirs
1380  *
1381  * FIXME If there was a list of inodes with EAs, we could simply walk the list
1382  * intead of reading every inode.  This is an internal implementation issue.
1383  */
1384
1385 static int ext3_iterate_all(struct super_block *sb,
1386                             int (*repeat)(struct inode *inode,void *priv),
1387                             struct inode **start, void *priv)
1388 {
1389         struct inode *tmp = NULL;
1390         int gstart, gnum, err = 0;
1391         ino_t istart, ibase;
1392         ENTRY;
1393
1394         if (!start)
1395                 start = &tmp;
1396         if (!*start) {
1397                 *start = iget(sb, EXT3_ROOT_INO);
1398                 if (!*start)
1399                         GOTO(exit, err = -ENOMEM);
1400
1401                 if (is_bad_inode(*start))
1402                         GOTO(exit, err = -EIO);
1403         }
1404         if ((*start)->i_ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count)) {
1405                 CERROR("invalid starting inode %ld\n",(*start)->i_ino);
1406                 GOTO(exit, err = -EINVAL);
1407         }
1408         if ((*start)->i_ino < EXT3_FIRST_INO(sb)) {
1409                 if ((err = (*repeat)(*start, priv) != 0))
1410                         GOTO(exit, err);
1411                 iput(*start);
1412                 *start = iget(sb, EXT3_FIRST_INO(sb));
1413                 if (!*start)
1414                         GOTO(exit, err = -ENOMEM);
1415                 if (is_bad_inode(*start))
1416                         GOTO(exit, err = -EIO);
1417         }
1418
1419         gstart = ((*start)->i_ino - 1) / EXT3_INODES_PER_GROUP(sb);
1420         istart = ((*start)->i_ino - 1) % EXT3_INODES_PER_GROUP(sb);
1421         ibase = gstart * EXT3_INODES_PER_GROUP(sb);
1422         for (gnum = gstart; gnum < EXT3_SB(sb)->s_groups_count;
1423              gnum++, ibase += EXT3_INODES_PER_GROUP(sb)) {
1424                 struct ext3_group_desc * gdp;
1425                 int bitmap_nr, ibyte;
1426                 char *bitmap;
1427
1428                 gdp = ext3_get_group_desc (sb, gnum, NULL);
1429                 if (!gdp || le16_to_cpu(gdp->bg_free_inodes_count) ==
1430                     EXT3_INODES_PER_GROUP(sb))
1431                         continue;
1432
1433                 bitmap_nr = ext3_load_inode_bitmap(sb, gnum);
1434                 if (bitmap_nr < 0)
1435                         continue;
1436
1437                 bitmap = EXT3_SB(sb)->s_inode_bitmap[bitmap_nr]->b_data;
1438                 for (ibyte = istart >> 3; ibyte < EXT3_INODES_PER_GROUP(sb) >> 3;
1439                      ibyte++) {
1440                         int i, bit;
1441
1442                         if (!bitmap[ibyte])
1443                                 continue;
1444
1445                         /* FIXME need to verify if bit endianness will
1446                          *       work properly here for all architectures.
1447                          */
1448                         for (i = 1, bit = 1; i <= 8; i++, bit <<= 1) {
1449                                 ino_t ino = ibase + (ibyte << 3) + i;
1450
1451                                 if ((bitmap[ibyte] & bit) == 0)
1452                                         continue;
1453                                 if (*start) {
1454                                         if (ino < (*start)->i_ino)
1455                                                 continue;
1456                                 } else {
1457                                         *start = iget(sb, ino);
1458                                         if (!*start)
1459                                                 GOTO(exit, err = -ENOMEM);
1460                                         if (is_bad_inode(*start))
1461                                                 GOTO(exit, err = -EIO);
1462                                 }
1463                                 if ((err = (*repeat)(*start, priv)) != 0)
1464                                         GOTO(exit, err);
1465                                 iput(*start);
1466                                 *start = NULL;
1467                         }
1468                 }
1469                 istart = 0;
1470         }
1471 exit:
1472         iput(tmp);
1473         RETURN(err);
1474 }
1475
1476 static int fsfilt_ext3_iterate(struct super_block *sb,
1477                                int (*repeat)(struct inode *inode, void *priv),
1478                                struct inode **start, void *priv, int flag)
1479 {
1480         switch(flag) {
1481                 case SNAP_ITERATE_ALL_INODE:
1482                         return ext3_iterate_all (sb, repeat, start, priv);
1483                 default:
1484                         return -EINVAL;
1485         }
1486 }
1487
1488 static int fsfilt_ext3_get_snap_info(struct inode *inode, void *key,
1489                                      __u32 keylen, void *val,
1490                                      __u32 *vallen)
1491 {
1492         int rc = 0;
1493         ENTRY;
1494
1495         if (!vallen || !val) {
1496                 CERROR("val and val_size is 0!\n");
1497                 RETURN(-EFAULT);
1498         }
1499         if (keylen >= strlen(MAX_SNAPTABLE_COUNT)
1500             && strcmp(key, MAX_SNAPTABLE_COUNT) == 0) {
1501                 /*FIXME should get it from the EA_size*/
1502                *((__u32 *)val) = EXT3_MAX_SNAPS;
1503                *vallen = sizeof(int);
1504                RETURN(rc);
1505         } else if (keylen >= strlen(SNAPTABLE_INFO)
1506                    && strcmp(key, SNAPTABLE_INFO) == 0) {
1507                 rc = ext3_xattr_get(inode, EXT3_SNAP_INDEX,
1508                                     EXT3_SNAPTABLE_EA, val, *vallen);
1509                 RETURN(rc);
1510         } else if (keylen >= strlen(SNAP_GENERATION)
1511                    && strcmp(key, SNAP_GENERATION) == 0) {
1512
1513                 rc = ext3_xattr_get(inode, EXT3_SNAP_INDEX,EXT3_SNAP_GENERATION,
1514                                     (char *)val, *vallen);
1515                 if (rc == -ENOATTR) {
1516                         *((__u32 *)val) = 0;
1517                         *vallen = sizeof(int);
1518                         rc = 0;
1519                 }
1520                 RETURN(rc);
1521         }
1522         RETURN(-EINVAL);
1523 }
1524
1525 static int fsfilt_ext3_set_snap_info(struct inode *inode, void *key,
1526                                      __u32 keylen, void *val,
1527                                      __u32 *vallen)
1528 {
1529         int rc = 0;
1530         ENTRY;
1531
1532         if (!vallen || !val) {
1533                 CERROR("val and val_size is 0!\n");
1534                 RETURN(-EFAULT);
1535         }
1536
1537         if (keylen >= strlen(SNAPTABLE_INFO)
1538             && strcmp(key, SNAPTABLE_INFO) == 0) {
1539                 handle_t *handle;
1540
1541                 handle = ext3_journal_start(inode, EXT3_XATTR_TRANS_BLOCKS);
1542                 if( !handle )
1543                         RETURN(-EINVAL);
1544                 rc = ext3_xattr_set(handle, inode, EXT3_SNAP_INDEX,
1545                                     EXT3_SNAPTABLE_EA, val, *vallen, 0);
1546                 ext3_journal_stop(handle, inode);
1547
1548                 RETURN(rc);
1549         } else if (keylen >= strlen(SNAP_GENERATION)
1550                    && strcmp(key, SNAP_GENERATION) == 0) {
1551                 LASSERT(inode);
1552                 rc = ext3_set_generation(inode, *(int*)val);
1553
1554                 RETURN(rc);
1555         }
1556         RETURN(-EINVAL);
1557 }
1558
1559 struct fsfilt_operations fsfilt_ext3_snap_ops = {
1560         .fs_type                = "ext3_snap",
1561         .fs_owner               = THIS_MODULE,
1562         .fs_create_indirect     = fsfilt_ext3_create_indirect,
1563         .fs_get_indirect        = fsfilt_ext3_get_indirect,
1564         .fs_set_indirect        = fsfilt_ext3_set_indirect,
1565         .fs_snap_feature        = fsfilt_ext3_snap_feature,
1566         .fs_is_redirector       = fsfilt_ext3_is_redirector,
1567         .fs_is_indirect         = fsfilt_ext3_is_indirect,
1568         .fs_get_indirect_ino    = fsfilt_ext3_get_indirect_ino,
1569         .fs_destroy_indirect    = fsfilt_ext3_destroy_indirect,
1570         .fs_restore_indirect    = fsfilt_ext3_restore_indirect,
1571         .fs_iterate             = fsfilt_ext3_iterate,
1572         .fs_copy_block          = fsfilt_ext3_copy_block,
1573         .fs_set_snap_info       = fsfilt_ext3_set_snap_info,
1574         .fs_get_snap_info       = fsfilt_ext3_get_snap_info,
1575 };
1576
1577 static int __init fsfilt_ext3_snap_init(void)
1578 {
1579         int rc;
1580
1581         rc = fsfilt_register_ops(&fsfilt_ext3_snap_ops);
1582
1583         return rc;
1584 }
1585
1586 static void __exit fsfilt_ext3_snap_exit(void)
1587 {
1588
1589         fsfilt_unregister_ops(&fsfilt_ext3_snap_ops);
1590 }
1591
1592 module_init(fsfilt_ext3_snap_init);
1593 module_exit(fsfilt_ext3_snap_exit);
1594
1595 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1596 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
1597 MODULE_LICENSE("GPL");