vfs: fix possible deadlock in ext2, ext3, ext4 when using xattrs
[sfrench/cifs-2.6.git] / fs / ext4 / xattr.c
index e832e96095b33c177e880db6370e1686153439f6..e9054c1c7d9366ce36f1d140b125c5db8915cd15 100644 (file)
 #define BFIRST(bh) ENTRY(BHDR(bh)+1)
 #define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
 
-#define IHDR(inode, raw_inode) \
-       ((struct ext4_xattr_ibody_header *) \
-               ((void *)raw_inode + \
-                EXT4_GOOD_OLD_INODE_SIZE + \
-                EXT4_I(inode)->i_extra_isize))
-#define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
-
 #ifdef EXT4_XATTR_DEBUG
 # define ea_idebug(inode, f...) do { \
                printk(KERN_DEBUG "inode %s:%lu: ", \
@@ -487,7 +480,7 @@ ext4_xattr_release_block(handle_t *handle, struct inode *inode,
                ea_bdebug(bh, "refcount now=0; freeing");
                if (ce)
                        mb_cache_entry_free(ce);
-               ext4_free_blocks(handle, inode, bh->b_blocknr, 1);
+               ext4_free_blocks(handle, inode, bh->b_blocknr, 1, 1);
                get_bh(bh);
                ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
        } else {
@@ -508,6 +501,24 @@ out:
        return;
 }
 
+/*
+ * Find the available free space for EAs. This also returns the total number of
+ * bytes used by EA entries.
+ */
+static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
+                                   size_t *min_offs, void *base, int *total)
+{
+       for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
+               *total += EXT4_XATTR_LEN(last->e_name_len);
+               if (!last->e_value_block && last->e_value_size) {
+                       size_t offs = le16_to_cpu(last->e_value_offs);
+                       if (offs < *min_offs)
+                               *min_offs = offs;
+               }
+       }
+       return (*min_offs - ((void *)last - base) - sizeof(__u32));
+}
+
 struct ext4_xattr_info {
        int name_index;
        const char *name;
@@ -739,12 +750,11 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
                }
        } else {
                /* Allocate a buffer where we construct the new block. */
-               s->base = kmalloc(sb->s_blocksize, GFP_KERNEL);
+               s->base = kzalloc(sb->s_blocksize, GFP_KERNEL);
                /* assert(header == s->base) */
                error = -ENOMEM;
                if (s->base == NULL)
                        goto cleanup;
-               memset(s->base, 0, sb->s_blocksize);
                header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
                header(s->base)->h_blocks = cpu_to_le32(1);
                header(s->base)->h_refcount = cpu_to_le32(1);
@@ -811,7 +821,7 @@ inserted:
                        new_bh = sb_getblk(sb, block);
                        if (!new_bh) {
 getblk_failed:
-                               ext4_free_blocks(handle, inode, block, 1);
+                               ext4_free_blocks(handle, inode, block, 1, 1);
                                error = -EIO;
                                goto cleanup;
                        }
@@ -1013,7 +1023,9 @@ ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
        }
        if (!error) {
                ext4_xattr_update_super_block(handle, inode->i_sb);
-               inode->i_ctime = CURRENT_TIME_SEC;
+               inode->i_ctime = ext4_current_time(inode);
+               if (!value)
+                       EXT4_I(inode)->i_state &= ~EXT4_STATE_NO_EXPAND;
                error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
                /*
                 * The bh is consumed by ext4_mark_iloc_dirty, even with
@@ -1066,6 +1078,253 @@ retry:
        return error;
 }
 
+/*
+ * Shift the EA entries in the inode to create space for the increased
+ * i_extra_isize.
+ */
+static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
+                                    int value_offs_shift, void *to,
+                                    void *from, size_t n, int blocksize)
+{
+       struct ext4_xattr_entry *last = entry;
+       int new_offs;
+
+       /* Adjust the value offsets of the entries */
+       for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
+               if (!last->e_value_block && last->e_value_size) {
+                       new_offs = le16_to_cpu(last->e_value_offs) +
+                                                       value_offs_shift;
+                       BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
+                                > blocksize);
+                       last->e_value_offs = cpu_to_le16(new_offs);
+               }
+       }
+       /* Shift the entries by n bytes */
+       memmove(to, from, n);
+}
+
+/*
+ * Expand an inode by new_extra_isize bytes when EAs are present.
+ * Returns 0 on success or negative error number on failure.
+ */
+int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
+                              struct ext4_inode *raw_inode, handle_t *handle)
+{
+       struct ext4_xattr_ibody_header *header;
+       struct ext4_xattr_entry *entry, *last, *first;
+       struct buffer_head *bh = NULL;
+       struct ext4_xattr_ibody_find *is = NULL;
+       struct ext4_xattr_block_find *bs = NULL;
+       char *buffer = NULL, *b_entry_name = NULL;
+       size_t min_offs, free;
+       int total_ino, total_blk;
+       void *base, *start, *end;
+       int extra_isize = 0, error = 0, tried_min_extra_isize = 0;
+       int s_min_extra_isize = le16_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize);
+
+       down_write(&EXT4_I(inode)->xattr_sem);
+retry:
+       if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) {
+               up_write(&EXT4_I(inode)->xattr_sem);
+               return 0;
+       }
+
+       header = IHDR(inode, raw_inode);
+       entry = IFIRST(header);
+
+       /*
+        * Check if enough free space is available in the inode to shift the
+        * entries ahead by new_extra_isize.
+        */
+
+       base = start = entry;
+       end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
+       min_offs = end - base;
+       last = entry;
+       total_ino = sizeof(struct ext4_xattr_ibody_header);
+
+       free = ext4_xattr_free_space(last, &min_offs, base, &total_ino);
+       if (free >= new_extra_isize) {
+               entry = IFIRST(header);
+               ext4_xattr_shift_entries(entry, EXT4_I(inode)->i_extra_isize
+                               - new_extra_isize, (void *)raw_inode +
+                               EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
+                               (void *)header, total_ino,
+                               inode->i_sb->s_blocksize);
+               EXT4_I(inode)->i_extra_isize = new_extra_isize;
+               error = 0;
+               goto cleanup;
+       }
+
+       /*
+        * Enough free space isn't available in the inode, check if
+        * EA block can hold new_extra_isize bytes.
+        */
+       if (EXT4_I(inode)->i_file_acl) {
+               bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
+               error = -EIO;
+               if (!bh)
+                       goto cleanup;
+               if (ext4_xattr_check_block(bh)) {
+                       ext4_error(inode->i_sb, __FUNCTION__,
+                               "inode %lu: bad block %llu", inode->i_ino,
+                               EXT4_I(inode)->i_file_acl);
+                       error = -EIO;
+                       goto cleanup;
+               }
+               base = BHDR(bh);
+               first = BFIRST(bh);
+               end = bh->b_data + bh->b_size;
+               min_offs = end - base;
+               free = ext4_xattr_free_space(first, &min_offs, base,
+                                            &total_blk);
+               if (free < new_extra_isize) {
+                       if (!tried_min_extra_isize && s_min_extra_isize) {
+                               tried_min_extra_isize++;
+                               new_extra_isize = s_min_extra_isize;
+                               brelse(bh);
+                               goto retry;
+                       }
+                       error = -1;
+                       goto cleanup;
+               }
+       } else {
+               free = inode->i_sb->s_blocksize;
+       }
+
+       while (new_extra_isize > 0) {
+               size_t offs, size, entry_size;
+               struct ext4_xattr_entry *small_entry = NULL;
+               struct ext4_xattr_info i = {
+                       .value = NULL,
+                       .value_len = 0,
+               };
+               unsigned int total_size;  /* EA entry size + value size */
+               unsigned int shift_bytes; /* No. of bytes to shift EAs by? */
+               unsigned int min_total_size = ~0U;
+
+               is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
+               bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
+               if (!is || !bs) {
+                       error = -ENOMEM;
+                       goto cleanup;
+               }
+
+               is->s.not_found = -ENODATA;
+               bs->s.not_found = -ENODATA;
+               is->iloc.bh = NULL;
+               bs->bh = NULL;
+
+               last = IFIRST(header);
+               /* Find the entry best suited to be pushed into EA block */
+               entry = NULL;
+               for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
+                       total_size =
+                       EXT4_XATTR_SIZE(le32_to_cpu(last->e_value_size)) +
+                                       EXT4_XATTR_LEN(last->e_name_len);
+                       if (total_size <= free && total_size < min_total_size) {
+                               if (total_size < new_extra_isize) {
+                                       small_entry = last;
+                               } else {
+                                       entry = last;
+                                       min_total_size = total_size;
+                               }
+                       }
+               }
+
+               if (entry == NULL) {
+                       if (small_entry) {
+                               entry = small_entry;
+                       } else {
+                               if (!tried_min_extra_isize &&
+                                   s_min_extra_isize) {
+                                       tried_min_extra_isize++;
+                                       new_extra_isize = s_min_extra_isize;
+                                       goto retry;
+                               }
+                               error = -1;
+                               goto cleanup;
+                       }
+               }
+               offs = le16_to_cpu(entry->e_value_offs);
+               size = le32_to_cpu(entry->e_value_size);
+               entry_size = EXT4_XATTR_LEN(entry->e_name_len);
+               i.name_index = entry->e_name_index,
+               buffer = kmalloc(EXT4_XATTR_SIZE(size), GFP_NOFS);
+               b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
+               if (!buffer || !b_entry_name) {
+                       error = -ENOMEM;
+                       goto cleanup;
+               }
+               /* Save the entry name and the entry value */
+               memcpy(buffer, (void *)IFIRST(header) + offs,
+                      EXT4_XATTR_SIZE(size));
+               memcpy(b_entry_name, entry->e_name, entry->e_name_len);
+               b_entry_name[entry->e_name_len] = '\0';
+               i.name = b_entry_name;
+
+               error = ext4_get_inode_loc(inode, &is->iloc);
+               if (error)
+                       goto cleanup;
+
+               error = ext4_xattr_ibody_find(inode, &i, is);
+               if (error)
+                       goto cleanup;
+
+               /* Remove the chosen entry from the inode */
+               error = ext4_xattr_ibody_set(handle, inode, &i, is);
+
+               entry = IFIRST(header);
+               if (entry_size + EXT4_XATTR_SIZE(size) >= new_extra_isize)
+                       shift_bytes = new_extra_isize;
+               else
+                       shift_bytes = entry_size + size;
+               /* Adjust the offsets and shift the remaining entries ahead */
+               ext4_xattr_shift_entries(entry, EXT4_I(inode)->i_extra_isize -
+                       shift_bytes, (void *)raw_inode +
+                       EXT4_GOOD_OLD_INODE_SIZE + extra_isize + shift_bytes,
+                       (void *)header, total_ino - entry_size,
+                       inode->i_sb->s_blocksize);
+
+               extra_isize += shift_bytes;
+               new_extra_isize -= shift_bytes;
+               EXT4_I(inode)->i_extra_isize = extra_isize;
+
+               i.name = b_entry_name;
+               i.value = buffer;
+               i.value_len = size;
+               error = ext4_xattr_block_find(inode, &i, bs);
+               if (error)
+                       goto cleanup;
+
+               /* Add entry which was removed from the inode into the block */
+               error = ext4_xattr_block_set(handle, inode, &i, bs);
+               if (error)
+                       goto cleanup;
+               kfree(b_entry_name);
+               kfree(buffer);
+               brelse(is->iloc.bh);
+               kfree(is);
+               kfree(bs);
+       }
+       brelse(bh);
+       up_write(&EXT4_I(inode)->xattr_sem);
+       return 0;
+
+cleanup:
+       kfree(b_entry_name);
+       kfree(buffer);
+       if (is)
+               brelse(is->iloc.bh);
+       kfree(is);
+       kfree(bs);
+       brelse(bh);
+       up_write(&EXT4_I(inode)->xattr_sem);
+       return error;
+}
+
+
+
 /*
  * ext4_xattr_delete_inode()
  *
@@ -1127,7 +1386,7 @@ ext4_xattr_cache_insert(struct buffer_head *bh)
        struct mb_cache_entry *ce;
        int error;
 
-       ce = mb_cache_entry_alloc(ext4_xattr_cache);
+       ce = mb_cache_entry_alloc(ext4_xattr_cache, GFP_NOFS);
        if (!ce) {
                ea_bdebug(bh, "out of memory");
                return;