Back to home page

OSCL-LXR

 
 

    


0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  *  linux/fs/ext2/balloc.c
0004  *
0005  * Copyright (C) 1992, 1993, 1994, 1995
0006  * Remy Card (card@masi.ibp.fr)
0007  * Laboratoire MASI - Institut Blaise Pascal
0008  * Universite Pierre et Marie Curie (Paris VI)
0009  *
0010  *  Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
0011  *  Big-endian to little-endian byte-swapping/bitmaps by
0012  *        David S. Miller (davem@caip.rutgers.edu), 1995
0013  */
0014 
0015 #include "ext2.h"
0016 #include <linux/quotaops.h>
0017 #include <linux/slab.h>
0018 #include <linux/sched.h>
0019 #include <linux/cred.h>
0020 #include <linux/buffer_head.h>
0021 #include <linux/capability.h>
0022 
0023 /*
0024  * balloc.c contains the blocks allocation and deallocation routines
0025  */
0026 
0027 /*
0028  * The free blocks are managed by bitmaps.  A file system contains several
0029  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
0030  * block for inodes, N blocks for the inode table and data blocks.
0031  *
0032  * The file system contains group descriptors which are located after the
0033  * super block.  Each descriptor contains the number of the bitmap block and
0034  * the free blocks count in the block.  The descriptors are loaded in memory
0035  * when a file system is mounted (see ext2_fill_super).
0036  */
0037 
0038 
0039 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
0040 
0041 struct ext2_group_desc * ext2_get_group_desc(struct super_block * sb,
0042                          unsigned int block_group,
0043                          struct buffer_head ** bh)
0044 {
0045     unsigned long group_desc;
0046     unsigned long offset;
0047     struct ext2_group_desc * desc;
0048     struct ext2_sb_info *sbi = EXT2_SB(sb);
0049 
0050     if (block_group >= sbi->s_groups_count) {
0051         WARN(1, "block_group >= groups_count - "
0052              "block_group = %d, groups_count = %lu",
0053              block_group, sbi->s_groups_count);
0054 
0055         return NULL;
0056     }
0057 
0058     group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(sb);
0059     offset = block_group & (EXT2_DESC_PER_BLOCK(sb) - 1);
0060     if (!sbi->s_group_desc[group_desc]) {
0061         WARN(1, "Group descriptor not loaded - "
0062              "block_group = %d, group_desc = %lu, desc = %lu",
0063               block_group, group_desc, offset);
0064         return NULL;
0065     }
0066 
0067     desc = (struct ext2_group_desc *) sbi->s_group_desc[group_desc]->b_data;
0068     if (bh)
0069         *bh = sbi->s_group_desc[group_desc];
0070     return desc + offset;
0071 }
0072 
0073 static int ext2_valid_block_bitmap(struct super_block *sb,
0074                     struct ext2_group_desc *desc,
0075                     unsigned int block_group,
0076                     struct buffer_head *bh)
0077 {
0078     ext2_grpblk_t offset;
0079     ext2_grpblk_t next_zero_bit;
0080     ext2_fsblk_t bitmap_blk;
0081     ext2_fsblk_t group_first_block;
0082 
0083     group_first_block = ext2_group_first_block_no(sb, block_group);
0084 
0085     /* check whether block bitmap block number is set */
0086     bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
0087     offset = bitmap_blk - group_first_block;
0088     if (!ext2_test_bit(offset, bh->b_data))
0089         /* bad block bitmap */
0090         goto err_out;
0091 
0092     /* check whether the inode bitmap block number is set */
0093     bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
0094     offset = bitmap_blk - group_first_block;
0095     if (!ext2_test_bit(offset, bh->b_data))
0096         /* bad block bitmap */
0097         goto err_out;
0098 
0099     /* check whether the inode table block number is set */
0100     bitmap_blk = le32_to_cpu(desc->bg_inode_table);
0101     offset = bitmap_blk - group_first_block;
0102     next_zero_bit = ext2_find_next_zero_bit(bh->b_data,
0103                 offset + EXT2_SB(sb)->s_itb_per_group,
0104                 offset);
0105     if (next_zero_bit >= offset + EXT2_SB(sb)->s_itb_per_group)
0106         /* good bitmap for inode tables */
0107         return 1;
0108 
0109 err_out:
0110     ext2_error(sb, __func__,
0111             "Invalid block bitmap - "
0112             "block_group = %d, block = %lu",
0113             block_group, bitmap_blk);
0114     return 0;
0115 }
0116 
0117 /*
0118  * Read the bitmap for a given block_group,and validate the
0119  * bits for block/inode/inode tables are set in the bitmaps
0120  *
0121  * Return buffer_head on success or NULL in case of failure.
0122  */
0123 static struct buffer_head *
0124 read_block_bitmap(struct super_block *sb, unsigned int block_group)
0125 {
0126     struct ext2_group_desc * desc;
0127     struct buffer_head * bh = NULL;
0128     ext2_fsblk_t bitmap_blk;
0129 
0130     desc = ext2_get_group_desc(sb, block_group, NULL);
0131     if (!desc)
0132         return NULL;
0133     bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
0134     bh = sb_getblk(sb, bitmap_blk);
0135     if (unlikely(!bh)) {
0136         ext2_error(sb, __func__,
0137                 "Cannot read block bitmap - "
0138                 "block_group = %d, block_bitmap = %u",
0139                 block_group, le32_to_cpu(desc->bg_block_bitmap));
0140         return NULL;
0141     }
0142     if (likely(bh_uptodate_or_lock(bh)))
0143         return bh;
0144 
0145     if (bh_submit_read(bh) < 0) {
0146         brelse(bh);
0147         ext2_error(sb, __func__,
0148                 "Cannot read block bitmap - "
0149                 "block_group = %d, block_bitmap = %u",
0150                 block_group, le32_to_cpu(desc->bg_block_bitmap));
0151         return NULL;
0152     }
0153 
0154     ext2_valid_block_bitmap(sb, desc, block_group, bh);
0155     /*
0156      * file system mounted not to panic on error, continue with corrupt
0157      * bitmap
0158      */
0159     return bh;
0160 }
0161 
0162 static void group_adjust_blocks(struct super_block *sb, int group_no,
0163     struct ext2_group_desc *desc, struct buffer_head *bh, int count)
0164 {
0165     if (count) {
0166         struct ext2_sb_info *sbi = EXT2_SB(sb);
0167         unsigned free_blocks;
0168 
0169         spin_lock(sb_bgl_lock(sbi, group_no));
0170         free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
0171         desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
0172         spin_unlock(sb_bgl_lock(sbi, group_no));
0173         mark_buffer_dirty(bh);
0174     }
0175 }
0176 
0177 /*
0178  * The reservation window structure operations
0179  * --------------------------------------------
0180  * Operations include:
0181  * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
0182  *
0183  * We use a red-black tree to represent per-filesystem reservation
0184  * windows.
0185  *
0186  */
0187 
0188 /**
0189  * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
0190  * @root:       root of per-filesystem reservation rb tree
0191  * @verbose:        verbose mode
0192  * @fn:         function which wishes to dump the reservation map
0193  *
0194  * If verbose is turned on, it will print the whole block reservation
0195  * windows(start, end). Otherwise, it will only print out the "bad" windows,
0196  * those windows that overlap with their immediate neighbors.
0197  */
0198 #if 1
0199 static void __rsv_window_dump(struct rb_root *root, int verbose,
0200                   const char *fn)
0201 {
0202     struct rb_node *n;
0203     struct ext2_reserve_window_node *rsv, *prev;
0204     int bad;
0205 
0206 restart:
0207     n = rb_first(root);
0208     bad = 0;
0209     prev = NULL;
0210 
0211     printk("Block Allocation Reservation Windows Map (%s):\n", fn);
0212     while (n) {
0213         rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
0214         if (verbose)
0215             printk("reservation window 0x%p "
0216                 "start: %lu, end: %lu\n",
0217                 rsv, rsv->rsv_start, rsv->rsv_end);
0218         if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
0219             printk("Bad reservation %p (start >= end)\n",
0220                    rsv);
0221             bad = 1;
0222         }
0223         if (prev && prev->rsv_end >= rsv->rsv_start) {
0224             printk("Bad reservation %p (prev->end >= start)\n",
0225                    rsv);
0226             bad = 1;
0227         }
0228         if (bad) {
0229             if (!verbose) {
0230                 printk("Restarting reservation walk in verbose mode\n");
0231                 verbose = 1;
0232                 goto restart;
0233             }
0234         }
0235         n = rb_next(n);
0236         prev = rsv;
0237     }
0238     printk("Window map complete.\n");
0239     BUG_ON(bad);
0240 }
0241 #define rsv_window_dump(root, verbose) \
0242     __rsv_window_dump((root), (verbose), __func__)
0243 #else
0244 #define rsv_window_dump(root, verbose) do {} while (0)
0245 #endif
0246 
0247 /**
0248  * goal_in_my_reservation()
0249  * @rsv:        inode's reservation window
0250  * @grp_goal:       given goal block relative to the allocation block group
0251  * @group:      the current allocation block group
0252  * @sb:         filesystem super block
0253  *
0254  * Test if the given goal block (group relative) is within the file's
0255  * own block reservation window range.
0256  *
0257  * If the reservation window is outside the goal allocation group, return 0;
0258  * grp_goal (given goal block) could be -1, which means no specific
0259  * goal block. In this case, always return 1.
0260  * If the goal block is within the reservation window, return 1;
0261  * otherwise, return 0;
0262  */
0263 static int
0264 goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
0265             unsigned int group, struct super_block * sb)
0266 {
0267     ext2_fsblk_t group_first_block, group_last_block;
0268 
0269     group_first_block = ext2_group_first_block_no(sb, group);
0270     group_last_block = ext2_group_last_block_no(sb, group);
0271 
0272     if ((rsv->_rsv_start > group_last_block) ||
0273         (rsv->_rsv_end < group_first_block))
0274         return 0;
0275     if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
0276         || (grp_goal + group_first_block > rsv->_rsv_end)))
0277         return 0;
0278     return 1;
0279 }
0280 
0281 /**
0282  * search_reserve_window()
0283  * @root:       root of reservation tree
0284  * @goal:       target allocation block
0285  *
0286  * Find the reserved window which includes the goal, or the previous one
0287  * if the goal is not in any window.
0288  * Returns NULL if there are no windows or if all windows start after the goal.
0289  */
0290 static struct ext2_reserve_window_node *
0291 search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
0292 {
0293     struct rb_node *n = root->rb_node;
0294     struct ext2_reserve_window_node *rsv;
0295 
0296     if (!n)
0297         return NULL;
0298 
0299     do {
0300         rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
0301 
0302         if (goal < rsv->rsv_start)
0303             n = n->rb_left;
0304         else if (goal > rsv->rsv_end)
0305             n = n->rb_right;
0306         else
0307             return rsv;
0308     } while (n);
0309     /*
0310      * We've fallen off the end of the tree: the goal wasn't inside
0311      * any particular node.  OK, the previous node must be to one
0312      * side of the interval containing the goal.  If it's the RHS,
0313      * we need to back up one.
0314      */
0315     if (rsv->rsv_start > goal) {
0316         n = rb_prev(&rsv->rsv_node);
0317         rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
0318     }
0319     return rsv;
0320 }
0321 
0322 /*
0323  * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
0324  * @sb:         super block
0325  * @rsv:        reservation window to add
0326  *
0327  * Must be called with rsv_lock held.
0328  */
0329 void ext2_rsv_window_add(struct super_block *sb,
0330             struct ext2_reserve_window_node *rsv)
0331 {
0332     struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
0333     struct rb_node *node = &rsv->rsv_node;
0334     ext2_fsblk_t start = rsv->rsv_start;
0335 
0336     struct rb_node ** p = &root->rb_node;
0337     struct rb_node * parent = NULL;
0338     struct ext2_reserve_window_node *this;
0339 
0340     while (*p)
0341     {
0342         parent = *p;
0343         this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
0344 
0345         if (start < this->rsv_start)
0346             p = &(*p)->rb_left;
0347         else if (start > this->rsv_end)
0348             p = &(*p)->rb_right;
0349         else {
0350             rsv_window_dump(root, 1);
0351             BUG();
0352         }
0353     }
0354 
0355     rb_link_node(node, parent, p);
0356     rb_insert_color(node, root);
0357 }
0358 
0359 /**
0360  * rsv_window_remove() -- unlink a window from the reservation rb tree
0361  * @sb:         super block
0362  * @rsv:        reservation window to remove
0363  *
0364  * Mark the block reservation window as not allocated, and unlink it
0365  * from the filesystem reservation window rb tree. Must be called with
0366  * rsv_lock held.
0367  */
0368 static void rsv_window_remove(struct super_block *sb,
0369                   struct ext2_reserve_window_node *rsv)
0370 {
0371     rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
0372     rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
0373     rsv->rsv_alloc_hit = 0;
0374     rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
0375 }
0376 
0377 /*
0378  * rsv_is_empty() -- Check if the reservation window is allocated.
0379  * @rsv:        given reservation window to check
0380  *
0381  * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
0382  */
0383 static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
0384 {
0385     /* a valid reservation end block could not be 0 */
0386     return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
0387 }
0388 
0389 /**
0390  * ext2_init_block_alloc_info()
0391  * @inode:      file inode structure
0392  *
0393  * Allocate and initialize the  reservation window structure, and
0394  * link the window to the ext2 inode structure at last
0395  *
0396  * The reservation window structure is only dynamically allocated
0397  * and linked to ext2 inode the first time the open file
0398  * needs a new block. So, before every ext2_new_block(s) call, for
0399  * regular files, we should check whether the reservation window
0400  * structure exists or not. In the latter case, this function is called.
0401  * Fail to do so will result in block reservation being turned off for that
0402  * open file.
0403  *
0404  * This function is called from ext2_get_blocks_handle(), also called
0405  * when setting the reservation window size through ioctl before the file
0406  * is open for write (needs block allocation).
0407  *
0408  * Needs truncate_mutex protection prior to calling this function.
0409  */
0410 void ext2_init_block_alloc_info(struct inode *inode)
0411 {
0412     struct ext2_inode_info *ei = EXT2_I(inode);
0413     struct ext2_block_alloc_info *block_i;
0414     struct super_block *sb = inode->i_sb;
0415 
0416     block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
0417     if (block_i) {
0418         struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node;
0419 
0420         rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
0421         rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
0422 
0423         /*
0424          * if filesystem is mounted with NORESERVATION, the goal
0425          * reservation window size is set to zero to indicate
0426          * block reservation is off
0427          */
0428         if (!test_opt(sb, RESERVATION))
0429             rsv->rsv_goal_size = 0;
0430         else
0431             rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS;
0432         rsv->rsv_alloc_hit = 0;
0433         block_i->last_alloc_logical_block = 0;
0434         block_i->last_alloc_physical_block = 0;
0435     }
0436     ei->i_block_alloc_info = block_i;
0437 }
0438 
0439 /**
0440  * ext2_discard_reservation()
0441  * @inode:      inode
0442  *
0443  * Discard(free) block reservation window on last file close, or truncate
0444  * or at last iput().
0445  *
0446  * It is being called in three cases:
0447  *  ext2_release_file(): last writer closes the file
0448  *  ext2_clear_inode(): last iput(), when nobody links to this file.
0449  *  ext2_truncate(): when the block indirect map is about to change.
0450  */
0451 void ext2_discard_reservation(struct inode *inode)
0452 {
0453     struct ext2_inode_info *ei = EXT2_I(inode);
0454     struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
0455     struct ext2_reserve_window_node *rsv;
0456     spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock;
0457 
0458     if (!block_i)
0459         return;
0460 
0461     rsv = &block_i->rsv_window_node;
0462     if (!rsv_is_empty(&rsv->rsv_window)) {
0463         spin_lock(rsv_lock);
0464         if (!rsv_is_empty(&rsv->rsv_window))
0465             rsv_window_remove(inode->i_sb, rsv);
0466         spin_unlock(rsv_lock);
0467     }
0468 }
0469 
0470 /**
0471  * ext2_free_blocks() -- Free given blocks and update quota and i_blocks
0472  * @inode:      inode
0473  * @block:      start physical block to free
0474  * @count:      number of blocks to free
0475  */
0476 void ext2_free_blocks (struct inode * inode, unsigned long block,
0477                unsigned long count)
0478 {
0479     struct buffer_head *bitmap_bh = NULL;
0480     struct buffer_head * bh2;
0481     unsigned long block_group;
0482     unsigned long bit;
0483     unsigned long i;
0484     unsigned long overflow;
0485     struct super_block * sb = inode->i_sb;
0486     struct ext2_sb_info * sbi = EXT2_SB(sb);
0487     struct ext2_group_desc * desc;
0488     struct ext2_super_block * es = sbi->s_es;
0489     unsigned freed = 0, group_freed;
0490 
0491     if (!ext2_data_block_valid(sbi, block, count)) {
0492         ext2_error (sb, "ext2_free_blocks",
0493                 "Freeing blocks not in datazone - "
0494                 "block = %lu, count = %lu", block, count);
0495         goto error_return;
0496     }
0497 
0498     ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
0499 
0500 do_more:
0501     overflow = 0;
0502     block_group = (block - le32_to_cpu(es->s_first_data_block)) /
0503               EXT2_BLOCKS_PER_GROUP(sb);
0504     bit = (block - le32_to_cpu(es->s_first_data_block)) %
0505               EXT2_BLOCKS_PER_GROUP(sb);
0506     /*
0507      * Check to see if we are freeing blocks across a group
0508      * boundary.
0509      */
0510     if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
0511         overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
0512         count -= overflow;
0513     }
0514     brelse(bitmap_bh);
0515     bitmap_bh = read_block_bitmap(sb, block_group);
0516     if (!bitmap_bh)
0517         goto error_return;
0518 
0519     desc = ext2_get_group_desc (sb, block_group, &bh2);
0520     if (!desc)
0521         goto error_return;
0522 
0523     if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
0524         in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
0525         in_range (block, le32_to_cpu(desc->bg_inode_table),
0526               sbi->s_itb_per_group) ||
0527         in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
0528               sbi->s_itb_per_group)) {
0529         ext2_error (sb, "ext2_free_blocks",
0530                 "Freeing blocks in system zones - "
0531                 "Block = %lu, count = %lu",
0532                 block, count);
0533         goto error_return;
0534     }
0535 
0536     for (i = 0, group_freed = 0; i < count; i++) {
0537         if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
0538                         bit + i, bitmap_bh->b_data)) {
0539             ext2_error(sb, __func__,
0540                 "bit already cleared for block %lu", block + i);
0541         } else {
0542             group_freed++;
0543         }
0544     }
0545 
0546     mark_buffer_dirty(bitmap_bh);
0547     if (sb->s_flags & SB_SYNCHRONOUS)
0548         sync_dirty_buffer(bitmap_bh);
0549 
0550     group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
0551     freed += group_freed;
0552 
0553     if (overflow) {
0554         block += count;
0555         count = overflow;
0556         goto do_more;
0557     }
0558 error_return:
0559     brelse(bitmap_bh);
0560     if (freed) {
0561         percpu_counter_add(&sbi->s_freeblocks_counter, freed);
0562         dquot_free_block_nodirty(inode, freed);
0563         mark_inode_dirty(inode);
0564     }
0565 }
0566 
0567 /**
0568  * bitmap_search_next_usable_block()
0569  * @start:      the starting block (group relative) of the search
0570  * @bh:         bufferhead contains the block group bitmap
0571  * @maxblocks:      the ending block (group relative) of the reservation
0572  *
0573  * The bitmap search --- search forward through the actual bitmap on disk until
0574  * we find a bit free.
0575  */
0576 static ext2_grpblk_t
0577 bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
0578                     ext2_grpblk_t maxblocks)
0579 {
0580     ext2_grpblk_t next;
0581 
0582     next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
0583     if (next >= maxblocks)
0584         return -1;
0585     return next;
0586 }
0587 
0588 /**
0589  * find_next_usable_block()
0590  * @start:      the starting block (group relative) to find next
0591  *          allocatable block in bitmap.
0592  * @bh:         bufferhead contains the block group bitmap
0593  * @maxblocks:      the ending block (group relative) for the search
0594  *
0595  * Find an allocatable block in a bitmap.  We perform the "most
0596  * appropriate allocation" algorithm of looking for a free block near
0597  * the initial goal; then for a free byte somewhere in the bitmap;
0598  * then for any free bit in the bitmap.
0599  */
0600 static ext2_grpblk_t
0601 find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
0602 {
0603     ext2_grpblk_t here, next;
0604     char *p, *r;
0605 
0606     if (start > 0) {
0607         /*
0608          * The goal was occupied; search forward for a free 
0609          * block within the next XX blocks.
0610          *
0611          * end_goal is more or less random, but it has to be
0612          * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
0613          * next 64-bit boundary is simple..
0614          */
0615         ext2_grpblk_t end_goal = (start + 63) & ~63;
0616         if (end_goal > maxblocks)
0617             end_goal = maxblocks;
0618         here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
0619         if (here < end_goal)
0620             return here;
0621         ext2_debug("Bit not found near goal\n");
0622     }
0623 
0624     here = start;
0625     if (here < 0)
0626         here = 0;
0627 
0628     p = ((char *)bh->b_data) + (here >> 3);
0629     r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
0630     next = (r - ((char *)bh->b_data)) << 3;
0631 
0632     if (next < maxblocks && next >= here)
0633         return next;
0634 
0635     here = bitmap_search_next_usable_block(here, bh, maxblocks);
0636     return here;
0637 }
0638 
0639 /**
0640  * ext2_try_to_allocate()
0641  * @sb:         superblock
0642  * @group:      given allocation block group
0643  * @bitmap_bh:      bufferhead holds the block bitmap
0644  * @grp_goal:       given target block within the group
0645  * @count:      target number of blocks to allocate
0646  * @my_rsv:     reservation window
0647  *
0648  * Attempt to allocate blocks within a give range. Set the range of allocation
0649  * first, then find the first free bit(s) from the bitmap (within the range),
0650  * and at last, allocate the blocks by claiming the found free bit as allocated.
0651  *
0652  * To set the range of this allocation:
0653  *  if there is a reservation window, only try to allocate block(s)
0654  *  from the file's own reservation window;
0655  *  Otherwise, the allocation range starts from the give goal block,
0656  *  ends at the block group's last block.
0657  *
0658  * If we failed to allocate the desired block then we may end up crossing to a
0659  * new bitmap.
0660  */
0661 static int
0662 ext2_try_to_allocate(struct super_block *sb, int group,
0663             struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
0664             unsigned long *count,
0665             struct ext2_reserve_window *my_rsv)
0666 {
0667     ext2_fsblk_t group_first_block = ext2_group_first_block_no(sb, group);
0668     ext2_fsblk_t group_last_block = ext2_group_last_block_no(sb, group);
0669         ext2_grpblk_t start, end;
0670     unsigned long num = 0;
0671 
0672     start = 0;
0673     end = group_last_block - group_first_block + 1;
0674     /* we do allocation within the reservation window if we have a window */
0675     if (my_rsv) {
0676         if (my_rsv->_rsv_start >= group_first_block)
0677             start = my_rsv->_rsv_start - group_first_block;
0678         if (my_rsv->_rsv_end < group_last_block)
0679             end = my_rsv->_rsv_end - group_first_block + 1;
0680         if (grp_goal < start || grp_goal >= end)
0681             grp_goal = -1;
0682     }
0683     BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
0684 
0685     if (grp_goal < 0) {
0686         grp_goal = find_next_usable_block(start, bitmap_bh, end);
0687         if (grp_goal < 0)
0688             goto fail_access;
0689         if (!my_rsv) {
0690             int i;
0691 
0692             for (i = 0; i < 7 && grp_goal > start &&
0693                     !ext2_test_bit(grp_goal - 1,
0694                                 bitmap_bh->b_data);
0695                         i++, grp_goal--)
0696                 ;
0697         }
0698     }
0699 
0700     for (; num < *count && grp_goal < end; grp_goal++) {
0701         if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
0702                     grp_goal, bitmap_bh->b_data)) {
0703             if (num == 0)
0704                 continue;
0705             break;
0706         }
0707         num++;
0708     }
0709 
0710     if (num == 0)
0711         goto fail_access;
0712 
0713     *count = num;
0714     return grp_goal - num;
0715 fail_access:
0716     return -1;
0717 }
0718 
0719 /**
0720  *  find_next_reservable_window():
0721  *      find a reservable space within the given range.
0722  *      It does not allocate the reservation window for now:
0723  *      alloc_new_reservation() will do the work later.
0724  *
0725  *  @search_head: the head of the searching list;
0726  *      This is not necessarily the list head of the whole filesystem
0727  *
0728  *      We have both head and start_block to assist the search
0729  *      for the reservable space. The list starts from head,
0730  *      but we will shift to the place where start_block is,
0731  *      then start from there, when looking for a reservable space.
0732  *
0733  *  @sb: the super block.
0734  *
0735  *  @start_block: the first block we consider to start the real search from
0736  *
0737  *  @last_block:
0738  *      the maximum block number that our goal reservable space
0739  *      could start from. This is normally the last block in this
0740  *      group. The search will end when we found the start of next
0741  *      possible reservable space is out of this boundary.
0742  *      This could handle the cross boundary reservation window
0743  *      request.
0744  *
0745  *  basically we search from the given range, rather than the whole
0746  *  reservation double linked list, (start_block, last_block)
0747  *  to find a free region that is of my size and has not
0748  *  been reserved.
0749  *
0750  */
0751 static int find_next_reservable_window(
0752                 struct ext2_reserve_window_node *search_head,
0753                 struct ext2_reserve_window_node *my_rsv,
0754                 struct super_block * sb,
0755                 ext2_fsblk_t start_block,
0756                 ext2_fsblk_t last_block)
0757 {
0758     struct rb_node *next;
0759     struct ext2_reserve_window_node *rsv, *prev;
0760     ext2_fsblk_t cur;
0761     int size = my_rsv->rsv_goal_size;
0762 
0763     /* TODO: make the start of the reservation window byte-aligned */
0764     /* cur = *start_block & ~7;*/
0765     cur = start_block;
0766     rsv = search_head;
0767     if (!rsv)
0768         return -1;
0769 
0770     while (1) {
0771         if (cur <= rsv->rsv_end)
0772             cur = rsv->rsv_end + 1;
0773 
0774         /* TODO?
0775          * in the case we could not find a reservable space
0776          * that is what is expected, during the re-search, we could
0777          * remember what's the largest reservable space we could have
0778          * and return that one.
0779          *
0780          * For now it will fail if we could not find the reservable
0781          * space with expected-size (or more)...
0782          */
0783         if (cur > last_block)
0784             return -1;      /* fail */
0785 
0786         prev = rsv;
0787         next = rb_next(&rsv->rsv_node);
0788         rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
0789 
0790         /*
0791          * Reached the last reservation, we can just append to the
0792          * previous one.
0793          */
0794         if (!next)
0795             break;
0796 
0797         if (cur + size <= rsv->rsv_start) {
0798             /*
0799              * Found a reserveable space big enough.  We could
0800              * have a reservation across the group boundary here
0801              */
0802             break;
0803         }
0804     }
0805     /*
0806      * we come here either :
0807      * when we reach the end of the whole list,
0808      * and there is empty reservable space after last entry in the list.
0809      * append it to the end of the list.
0810      *
0811      * or we found one reservable space in the middle of the list,
0812      * return the reservation window that we could append to.
0813      * succeed.
0814      */
0815 
0816     if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
0817         rsv_window_remove(sb, my_rsv);
0818 
0819     /*
0820      * Let's book the whole available window for now.  We will check the
0821      * disk bitmap later and then, if there are free blocks then we adjust
0822      * the window size if it's larger than requested.
0823      * Otherwise, we will remove this node from the tree next time
0824      * call find_next_reservable_window.
0825      */
0826     my_rsv->rsv_start = cur;
0827     my_rsv->rsv_end = cur + size - 1;
0828     my_rsv->rsv_alloc_hit = 0;
0829 
0830     if (prev != my_rsv)
0831         ext2_rsv_window_add(sb, my_rsv);
0832 
0833     return 0;
0834 }
0835 
0836 /**
0837  *  alloc_new_reservation()--allocate a new reservation window
0838  *
0839  *      To make a new reservation, we search part of the filesystem
0840  *      reservation list (the list that inside the group). We try to
0841  *      allocate a new reservation window near the allocation goal,
0842  *      or the beginning of the group, if there is no goal.
0843  *
0844  *      We first find a reservable space after the goal, then from
0845  *      there, we check the bitmap for the first free block after
0846  *      it. If there is no free block until the end of group, then the
0847  *      whole group is full, we failed. Otherwise, check if the free
0848  *      block is inside the expected reservable space, if so, we
0849  *      succeed.
0850  *      If the first free block is outside the reservable space, then
0851  *      start from the first free block, we search for next available
0852  *      space, and go on.
0853  *
0854  *  on succeed, a new reservation will be found and inserted into the list
0855  *  It contains at least one free block, and it does not overlap with other
0856  *  reservation windows.
0857  *
0858  *  failed: we failed to find a reservation window in this group
0859  *
0860  *  @my_rsv: the reservation
0861  *
0862  *  @grp_goal: The goal (group-relative).  It is where the search for a
0863  *      free reservable space should start from.
0864  *      if we have a goal(goal >0 ), then start from there,
0865  *      no goal(goal = -1), we start from the first block
0866  *      of the group.
0867  *
0868  *  @sb: the super block
0869  *  @group: the group we are trying to allocate in
0870  *  @bitmap_bh: the block group block bitmap
0871  *
0872  */
0873 static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
0874         ext2_grpblk_t grp_goal, struct super_block *sb,
0875         unsigned int group, struct buffer_head *bitmap_bh)
0876 {
0877     struct ext2_reserve_window_node *search_head;
0878     ext2_fsblk_t group_first_block, group_end_block, start_block;
0879     ext2_grpblk_t first_free_block;
0880     struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
0881     unsigned long size;
0882     int ret;
0883     spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
0884 
0885     group_first_block = ext2_group_first_block_no(sb, group);
0886     group_end_block = ext2_group_last_block_no(sb, group);
0887 
0888     if (grp_goal < 0)
0889         start_block = group_first_block;
0890     else
0891         start_block = grp_goal + group_first_block;
0892 
0893     size = my_rsv->rsv_goal_size;
0894 
0895     if (!rsv_is_empty(&my_rsv->rsv_window)) {
0896         /*
0897          * if the old reservation is cross group boundary
0898          * and if the goal is inside the old reservation window,
0899          * we will come here when we just failed to allocate from
0900          * the first part of the window. We still have another part
0901          * that belongs to the next group. In this case, there is no
0902          * point to discard our window and try to allocate a new one
0903          * in this group(which will fail). we should
0904          * keep the reservation window, just simply move on.
0905          *
0906          * Maybe we could shift the start block of the reservation
0907          * window to the first block of next group.
0908          */
0909 
0910         if ((my_rsv->rsv_start <= group_end_block) &&
0911                 (my_rsv->rsv_end > group_end_block) &&
0912                 (start_block >= my_rsv->rsv_start))
0913             return -1;
0914 
0915         if ((my_rsv->rsv_alloc_hit >
0916              (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
0917             /*
0918              * if the previously allocation hit ratio is
0919              * greater than 1/2, then we double the size of
0920              * the reservation window the next time,
0921              * otherwise we keep the same size window
0922              */
0923             size = size * 2;
0924             if (size > EXT2_MAX_RESERVE_BLOCKS)
0925                 size = EXT2_MAX_RESERVE_BLOCKS;
0926             my_rsv->rsv_goal_size= size;
0927         }
0928     }
0929 
0930     spin_lock(rsv_lock);
0931     /*
0932      * shift the search start to the window near the goal block
0933      */
0934     search_head = search_reserve_window(fs_rsv_root, start_block);
0935 
0936     /*
0937      * find_next_reservable_window() simply finds a reservable window
0938      * inside the given range(start_block, group_end_block).
0939      *
0940      * To make sure the reservation window has a free bit inside it, we
0941      * need to check the bitmap after we found a reservable window.
0942      */
0943 retry:
0944     ret = find_next_reservable_window(search_head, my_rsv, sb,
0945                         start_block, group_end_block);
0946 
0947     if (ret == -1) {
0948         if (!rsv_is_empty(&my_rsv->rsv_window))
0949             rsv_window_remove(sb, my_rsv);
0950         spin_unlock(rsv_lock);
0951         return -1;
0952     }
0953 
0954     /*
0955      * On success, find_next_reservable_window() returns the
0956      * reservation window where there is a reservable space after it.
0957      * Before we reserve this reservable space, we need
0958      * to make sure there is at least a free block inside this region.
0959      *
0960      * Search the first free bit on the block bitmap.  Search starts from
0961      * the start block of the reservable space we just found.
0962      */
0963     spin_unlock(rsv_lock);
0964     first_free_block = bitmap_search_next_usable_block(
0965             my_rsv->rsv_start - group_first_block,
0966             bitmap_bh, group_end_block - group_first_block + 1);
0967 
0968     if (first_free_block < 0) {
0969         /*
0970          * no free block left on the bitmap, no point
0971          * to reserve the space. return failed.
0972          */
0973         spin_lock(rsv_lock);
0974         if (!rsv_is_empty(&my_rsv->rsv_window))
0975             rsv_window_remove(sb, my_rsv);
0976         spin_unlock(rsv_lock);
0977         return -1;      /* failed */
0978     }
0979 
0980     start_block = first_free_block + group_first_block;
0981     /*
0982      * check if the first free block is within the
0983      * free space we just reserved
0984      */
0985     if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
0986         return 0;       /* success */
0987     /*
0988      * if the first free bit we found is out of the reservable space
0989      * continue search for next reservable space,
0990      * start from where the free block is,
0991      * we also shift the list head to where we stopped last time
0992      */
0993     search_head = my_rsv;
0994     spin_lock(rsv_lock);
0995     goto retry;
0996 }
0997 
0998 /**
0999  * try_to_extend_reservation()
1000  * @my_rsv:     given reservation window
1001  * @sb:         super block
1002  * @size:       the delta to extend
1003  *
1004  * Attempt to expand the reservation window large enough to have
1005  * required number of free blocks
1006  *
1007  * Since ext2_try_to_allocate() will always allocate blocks within
1008  * the reservation window range, if the window size is too small,
1009  * multiple blocks allocation has to stop at the end of the reservation
1010  * window. To make this more efficient, given the total number of
1011  * blocks needed and the current size of the window, we try to
1012  * expand the reservation window size if necessary on a best-effort
1013  * basis before ext2_new_blocks() tries to allocate blocks.
1014  */
1015 static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
1016             struct super_block *sb, int size)
1017 {
1018     struct ext2_reserve_window_node *next_rsv;
1019     struct rb_node *next;
1020     spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
1021 
1022     if (!spin_trylock(rsv_lock))
1023         return;
1024 
1025     next = rb_next(&my_rsv->rsv_node);
1026 
1027     if (!next)
1028         my_rsv->rsv_end += size;
1029     else {
1030         next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
1031 
1032         if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1033             my_rsv->rsv_end += size;
1034         else
1035             my_rsv->rsv_end = next_rsv->rsv_start - 1;
1036     }
1037     spin_unlock(rsv_lock);
1038 }
1039 
1040 /**
1041  * ext2_try_to_allocate_with_rsv()
1042  * @sb:         superblock
1043  * @group:      given allocation block group
1044  * @bitmap_bh:      bufferhead holds the block bitmap
1045  * @grp_goal:       given target block within the group
1046  * @count:      target number of blocks to allocate
1047  * @my_rsv:     reservation window
1048  *
1049  * This is the main function used to allocate a new block and its reservation
1050  * window.
1051  *
1052  * Each time when a new block allocation is need, first try to allocate from
1053  * its own reservation.  If it does not have a reservation window, instead of
1054  * looking for a free bit on bitmap first, then look up the reservation list to
1055  * see if it is inside somebody else's reservation window, we try to allocate a
1056  * reservation window for it starting from the goal first. Then do the block
1057  * allocation within the reservation window.
1058  *
1059  * This will avoid keeping on searching the reservation list again and
1060  * again when somebody is looking for a free block (without
1061  * reservation), and there are lots of free blocks, but they are all
1062  * being reserved.
1063  *
1064  * We use a red-black tree for the per-filesystem reservation list.
1065  */
1066 static ext2_grpblk_t
1067 ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
1068             struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
1069             struct ext2_reserve_window_node * my_rsv,
1070             unsigned long *count)
1071 {
1072     ext2_fsblk_t group_first_block, group_last_block;
1073     ext2_grpblk_t ret = 0;
1074     unsigned long num = *count;
1075 
1076     /*
1077      * we don't deal with reservation when
1078      * filesystem is mounted without reservation
1079      * or the file is not a regular file
1080      * or last attempt to allocate a block with reservation turned on failed
1081      */
1082     if (my_rsv == NULL) {
1083         return ext2_try_to_allocate(sb, group, bitmap_bh,
1084                         grp_goal, count, NULL);
1085     }
1086     /*
1087      * grp_goal is a group relative block number (if there is a goal)
1088      * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
1089      * first block is a filesystem wide block number
1090      * first block is the block number of the first block in this group
1091      */
1092     group_first_block = ext2_group_first_block_no(sb, group);
1093     group_last_block = ext2_group_last_block_no(sb, group);
1094 
1095     /*
1096      * Basically we will allocate a new block from inode's reservation
1097      * window.
1098      *
1099      * We need to allocate a new reservation window, if:
1100      * a) inode does not have a reservation window; or
1101      * b) last attempt to allocate a block from existing reservation
1102      *    failed; or
1103      * c) we come here with a goal and with a reservation window
1104      *
1105      * We do not need to allocate a new reservation window if we come here
1106      * at the beginning with a goal and the goal is inside the window, or
1107      * we don't have a goal but already have a reservation window.
1108      * then we could go to allocate from the reservation window directly.
1109      */
1110     while (1) {
1111         if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1112             !goal_in_my_reservation(&my_rsv->rsv_window,
1113                         grp_goal, group, sb)) {
1114             if (my_rsv->rsv_goal_size < *count)
1115                 my_rsv->rsv_goal_size = *count;
1116             ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1117                             group, bitmap_bh);
1118             if (ret < 0)
1119                 break;          /* failed */
1120 
1121             if (!goal_in_my_reservation(&my_rsv->rsv_window,
1122                             grp_goal, group, sb))
1123                 grp_goal = -1;
1124         } else if (grp_goal >= 0) {
1125             int curr = my_rsv->rsv_end -
1126                     (grp_goal + group_first_block) + 1;
1127 
1128             if (curr < *count)
1129                 try_to_extend_reservation(my_rsv, sb,
1130                             *count - curr);
1131         }
1132 
1133         if ((my_rsv->rsv_start > group_last_block) ||
1134                 (my_rsv->rsv_end < group_first_block)) {
1135             rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
1136             BUG();
1137         }
1138         ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
1139                        &num, &my_rsv->rsv_window);
1140         if (ret >= 0) {
1141             my_rsv->rsv_alloc_hit += num;
1142             *count = num;
1143             break;              /* succeed */
1144         }
1145         num = *count;
1146     }
1147     return ret;
1148 }
1149 
1150 /**
1151  * ext2_has_free_blocks()
1152  * @sbi:        in-core super block structure.
1153  *
1154  * Check if filesystem has at least 1 free block available for allocation.
1155  */
1156 static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
1157 {
1158     ext2_fsblk_t free_blocks, root_blocks;
1159 
1160     free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1161     root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1162     if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1163         !uid_eq(sbi->s_resuid, current_fsuid()) &&
1164         (gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) ||
1165          !in_group_p (sbi->s_resgid))) {
1166         return 0;
1167     }
1168     return 1;
1169 }
1170 
1171 /*
1172  * Returns 1 if the passed-in block region is valid; 0 if some part overlaps
1173  * with filesystem metadata blocks.
1174  */
1175 int ext2_data_block_valid(struct ext2_sb_info *sbi, ext2_fsblk_t start_blk,
1176               unsigned int count)
1177 {
1178     if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
1179         (start_blk + count - 1 < start_blk) ||
1180         (start_blk + count - 1 >= le32_to_cpu(sbi->s_es->s_blocks_count)))
1181         return 0;
1182 
1183     /* Ensure we do not step over superblock */
1184     if ((start_blk <= sbi->s_sb_block) &&
1185         (start_blk + count - 1 >= sbi->s_sb_block))
1186         return 0;
1187 
1188     return 1;
1189 }
1190 
1191 /*
1192  * ext2_new_blocks() -- core block(s) allocation function
1193  * @inode:      file inode
1194  * @goal:       given target block(filesystem wide)
1195  * @count:      target number of blocks to allocate
1196  * @errp:       error code
1197  *
1198  * ext2_new_blocks uses a goal block to assist allocation.  If the goal is
1199  * free, or there is a free block within 32 blocks of the goal, that block
1200  * is allocated.  Otherwise a forward search is made for a free block; within 
1201  * each block group the search first looks for an entire free byte in the block
1202  * bitmap, and then for any free bit if that fails.
1203  * This function also updates quota and i_blocks field.
1204  */
1205 ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
1206             unsigned long *count, int *errp)
1207 {
1208     struct buffer_head *bitmap_bh = NULL;
1209     struct buffer_head *gdp_bh;
1210     int group_no;
1211     int goal_group;
1212     ext2_grpblk_t grp_target_blk;   /* blockgroup relative goal block */
1213     ext2_grpblk_t grp_alloc_blk;    /* blockgroup-relative allocated block*/
1214     ext2_fsblk_t ret_block;     /* filesyetem-wide allocated block */
1215     int bgi;            /* blockgroup iteration index */
1216     int performed_allocation = 0;
1217     ext2_grpblk_t free_blocks;  /* number of free blocks in a group */
1218     struct super_block *sb;
1219     struct ext2_group_desc *gdp;
1220     struct ext2_super_block *es;
1221     struct ext2_sb_info *sbi;
1222     struct ext2_reserve_window_node *my_rsv = NULL;
1223     struct ext2_block_alloc_info *block_i;
1224     unsigned short windowsz = 0;
1225     unsigned long ngroups;
1226     unsigned long num = *count;
1227     int ret;
1228 
1229     *errp = -ENOSPC;
1230     sb = inode->i_sb;
1231 
1232     /*
1233      * Check quota for allocation of this block.
1234      */
1235     ret = dquot_alloc_block(inode, num);
1236     if (ret) {
1237         *errp = ret;
1238         return 0;
1239     }
1240 
1241     sbi = EXT2_SB(sb);
1242     es = EXT2_SB(sb)->s_es;
1243     ext2_debug("goal=%lu.\n", goal);
1244     /*
1245      * Allocate a block from reservation only when
1246      * filesystem is mounted with reservation(default,-o reservation), and
1247      * it's a regular file, and
1248      * the desired window size is greater than 0 (One could use ioctl
1249      * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
1250      * reservation on that particular file)
1251      */
1252     block_i = EXT2_I(inode)->i_block_alloc_info;
1253     if (block_i) {
1254         windowsz = block_i->rsv_window_node.rsv_goal_size;
1255         if (windowsz > 0)
1256             my_rsv = &block_i->rsv_window_node;
1257     }
1258 
1259     if (!ext2_has_free_blocks(sbi)) {
1260         *errp = -ENOSPC;
1261         goto out;
1262     }
1263 
1264     /*
1265      * First, test whether the goal block is free.
1266      */
1267     if (goal < le32_to_cpu(es->s_first_data_block) ||
1268         goal >= le32_to_cpu(es->s_blocks_count))
1269         goal = le32_to_cpu(es->s_first_data_block);
1270     group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1271             EXT2_BLOCKS_PER_GROUP(sb);
1272     goal_group = group_no;
1273 retry_alloc:
1274     gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1275     if (!gdp)
1276         goto io_error;
1277 
1278     free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1279     /*
1280      * if there is not enough free blocks to make a new resevation
1281      * turn off reservation for this allocation
1282      */
1283     if (my_rsv && (free_blocks < windowsz)
1284         && (free_blocks > 0)
1285         && (rsv_is_empty(&my_rsv->rsv_window)))
1286         my_rsv = NULL;
1287 
1288     if (free_blocks > 0) {
1289         grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1290                 EXT2_BLOCKS_PER_GROUP(sb));
1291         /*
1292          * In case we retry allocation (due to fs reservation not
1293          * working out or fs corruption), the bitmap_bh is non-null
1294          * pointer and we have to release it before calling
1295          * read_block_bitmap().
1296          */
1297         brelse(bitmap_bh);
1298         bitmap_bh = read_block_bitmap(sb, group_no);
1299         if (!bitmap_bh)
1300             goto io_error;
1301         grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1302                     bitmap_bh, grp_target_blk,
1303                     my_rsv, &num);
1304         if (grp_alloc_blk >= 0)
1305             goto allocated;
1306     }
1307 
1308     ngroups = EXT2_SB(sb)->s_groups_count;
1309     smp_rmb();
1310 
1311     /*
1312      * Now search the rest of the groups.  We assume that
1313      * group_no and gdp correctly point to the last group visited.
1314      */
1315     for (bgi = 0; bgi < ngroups; bgi++) {
1316         group_no++;
1317         if (group_no >= ngroups)
1318             group_no = 0;
1319         gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
1320         if (!gdp)
1321             goto io_error;
1322 
1323         free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1324         /*
1325          * skip this group (and avoid loading bitmap) if there
1326          * are no free blocks
1327          */
1328         if (!free_blocks)
1329             continue;
1330         /*
1331          * skip this group if the number of
1332          * free blocks is less than half of the reservation
1333          * window size.
1334          */
1335         if (my_rsv && (free_blocks <= (windowsz/2)))
1336             continue;
1337 
1338         brelse(bitmap_bh);
1339         bitmap_bh = read_block_bitmap(sb, group_no);
1340         if (!bitmap_bh)
1341             goto io_error;
1342         /*
1343          * try to allocate block(s) from this group, without a goal(-1).
1344          */
1345         grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
1346                     bitmap_bh, -1, my_rsv, &num);
1347         if (grp_alloc_blk >= 0)
1348             goto allocated;
1349     }
1350     /*
1351      * We may end up a bogus earlier ENOSPC error due to
1352      * filesystem is "full" of reservations, but
1353      * there maybe indeed free blocks available on disk
1354      * In this case, we just forget about the reservations
1355      * just do block allocation as without reservations.
1356      */
1357     if (my_rsv) {
1358         my_rsv = NULL;
1359         windowsz = 0;
1360         group_no = goal_group;
1361         goto retry_alloc;
1362     }
1363     /* No space left on the device */
1364     *errp = -ENOSPC;
1365     goto out;
1366 
1367 allocated:
1368 
1369     ext2_debug("using block group %d(%d)\n",
1370             group_no, gdp->bg_free_blocks_count);
1371 
1372     ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
1373 
1374     if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1375         in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1376         in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1377               EXT2_SB(sb)->s_itb_per_group) ||
1378         in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1379               EXT2_SB(sb)->s_itb_per_group)) {
1380         ext2_error(sb, "ext2_new_blocks",
1381                 "Allocating block in system zone - "
1382                 "blocks from "E2FSBLK", length %lu",
1383                 ret_block, num);
1384         /*
1385          * ext2_try_to_allocate marked the blocks we allocated as in
1386          * use.  So we may want to selectively mark some of the blocks
1387          * as free
1388          */
1389         num = *count;
1390         goto retry_alloc;
1391     }
1392 
1393     performed_allocation = 1;
1394 
1395     if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1396         ext2_error(sb, "ext2_new_blocks",
1397                 "block("E2FSBLK") >= blocks count(%d) - "
1398                 "block_group = %d, es == %p ", ret_block,
1399             le32_to_cpu(es->s_blocks_count), group_no, es);
1400         goto out;
1401     }
1402 
1403     group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
1404     percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1405 
1406     mark_buffer_dirty(bitmap_bh);
1407     if (sb->s_flags & SB_SYNCHRONOUS)
1408         sync_dirty_buffer(bitmap_bh);
1409 
1410     *errp = 0;
1411     brelse(bitmap_bh);
1412     if (num < *count) {
1413         dquot_free_block_nodirty(inode, *count-num);
1414         mark_inode_dirty(inode);
1415         *count = num;
1416     }
1417     return ret_block;
1418 
1419 io_error:
1420     *errp = -EIO;
1421 out:
1422     /*
1423      * Undo the block allocation
1424      */
1425     if (!performed_allocation) {
1426         dquot_free_block_nodirty(inode, *count);
1427         mark_inode_dirty(inode);
1428     }
1429     brelse(bitmap_bh);
1430     return 0;
1431 }
1432 
1433 ext2_fsblk_t ext2_new_block(struct inode *inode, unsigned long goal, int *errp)
1434 {
1435     unsigned long count = 1;
1436 
1437     return ext2_new_blocks(inode, goal, &count, errp);
1438 }
1439 
1440 #ifdef EXT2FS_DEBUG
1441 
1442 unsigned long ext2_count_free(struct buffer_head *map, unsigned int numchars)
1443 {
1444     return numchars * BITS_PER_BYTE - memweight(map->b_data, numchars);
1445 }
1446 
1447 #endif  /*  EXT2FS_DEBUG  */
1448 
1449 unsigned long ext2_count_free_blocks (struct super_block * sb)
1450 {
1451     struct ext2_group_desc * desc;
1452     unsigned long desc_count = 0;
1453     int i;
1454 #ifdef EXT2FS_DEBUG
1455     unsigned long bitmap_count, x;
1456     struct ext2_super_block *es;
1457 
1458     es = EXT2_SB(sb)->s_es;
1459     desc_count = 0;
1460     bitmap_count = 0;
1461     desc = NULL;
1462     for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1463         struct buffer_head *bitmap_bh;
1464         desc = ext2_get_group_desc (sb, i, NULL);
1465         if (!desc)
1466             continue;
1467         desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1468         bitmap_bh = read_block_bitmap(sb, i);
1469         if (!bitmap_bh)
1470             continue;
1471         
1472         x = ext2_count_free(bitmap_bh, sb->s_blocksize);
1473         printk ("group %d: stored = %d, counted = %lu\n",
1474             i, le16_to_cpu(desc->bg_free_blocks_count), x);
1475         bitmap_count += x;
1476         brelse(bitmap_bh);
1477     }
1478     printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
1479         (long)le32_to_cpu(es->s_free_blocks_count),
1480         desc_count, bitmap_count);
1481     return bitmap_count;
1482 #else
1483         for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
1484                 desc = ext2_get_group_desc (sb, i, NULL);
1485                 if (!desc)
1486                         continue;
1487                 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
1488     }
1489     return desc_count;
1490 #endif
1491 }
1492 
1493 static inline int test_root(int a, int b)
1494 {
1495     int num = b;
1496 
1497     while (a > num)
1498         num *= b;
1499     return num == a;
1500 }
1501 
1502 static int ext2_group_sparse(int group)
1503 {
1504     if (group <= 1)
1505         return 1;
1506     return (test_root(group, 3) || test_root(group, 5) ||
1507         test_root(group, 7));
1508 }
1509 
1510 /**
1511  *  ext2_bg_has_super - number of blocks used by the superblock in group
1512  *  @sb: superblock for filesystem
1513  *  @group: group number to check
1514  *
1515  *  Return the number of blocks used by the superblock (primary or backup)
1516  *  in this group.  Currently this will be only 0 or 1.
1517  */
1518 int ext2_bg_has_super(struct super_block *sb, int group)
1519 {
1520     if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
1521         !ext2_group_sparse(group))
1522         return 0;
1523     return 1;
1524 }
1525 
1526 /**
1527  *  ext2_bg_num_gdb - number of blocks used by the group table in group
1528  *  @sb: superblock for filesystem
1529  *  @group: group number to check
1530  *
1531  *  Return the number of blocks used by the group descriptor table
1532  *  (primary or backup) in this group.  In the future there may be a
1533  *  different number of descriptor blocks in each group.
1534  */
1535 unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
1536 {
1537     return ext2_bg_has_super(sb, group) ? EXT2_SB(sb)->s_gdb_count : 0;
1538 }
1539