#endif
/* for kernels 2.6.18 and later */
+#define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
+
#define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
ext3_ext_insert_extent(handle, inode, path, newext, flag)
#define ext3_mb_discard_inode_preallocations(inode) \
ext3_discard_preallocations(inode)
+#define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
+#define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
+
+static struct kmem_cache *fcb_cache;
+
+struct fsfilt_cb_data {
+ struct ext4_journal_cb_entry cb_jcb; /* private data - MUST BE FIRST */
+ fsfilt_cb_t cb_func; /* MDS/OBD completion function */
+ struct obd_device *cb_obd; /* MDS/OBD completion device */
+ __u64 cb_last_rcvd; /* MDS/OST last committed operation */
+ void *cb_data; /* MDS/OST completion function data */
+};
+
+static char *fsfilt_ext3_get_label(struct super_block *sb)
+{
+ return EXT3_SB(sb)->s_es->s_volume_name;
+}
+
+/* kernel has ext4_blocks_for_truncate since linux-3.1.1 */
+#ifdef HAVE_BLOCKS_FOR_TRUNCATE
+# include <ext4/truncate.h>
+#else
+static inline unsigned long ext4_blocks_for_truncate(struct inode *inode)
+{
+ ext4_lblk_t needed;
+
+ needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
+ if (needed < 2)
+ needed = 2;
+ if (needed > EXT4_MAX_TRANS_DATA)
+ needed = EXT4_MAX_TRANS_DATA;
+ return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
+}
+#endif
+
+/*
+ * We don't currently need any additional blocks for rmdir and
+ * unlink transactions because we are storing the OST oa_id inside
+ * the inode (which we will be changing anyways as part of this
+ * transaction).
+ */
+static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
+ int logs)
+{
+ /* For updates to the last received file */
+ int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
+ journal_t *journal;
+ void *handle;
+
+ if (current->journal_info) {
+ CDEBUG(D_INODE, "increasing refcount on %p\n",
+ current->journal_info);
+ goto journal_start;
+ }
+
+ switch(op) {
+ case FSFILT_OP_UNLINK:
+ /* delete one file + create/update logs for each stripe */
+ nblocks += EXT3_DELETE_TRANS_BLOCKS(inode->i_sb);
+ nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
+ FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
+ break;
+ case FSFILT_OP_CANCEL_UNLINK:
+ LASSERT(logs == 1);
+
+ /* blocks for log header bitmap update OR
+ * blocks for catalog header bitmap update + unlink of logs +
+ * blocks for delete the inode (include blocks truncating). */
+ nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
+ EXT3_DELETE_TRANS_BLOCKS(inode->i_sb) +
+ ext4_blocks_for_truncate(inode) + 3;
+ break;
+ default: CERROR("unknown transaction start op %d\n", op);
+ LBUG();
+ }
+
+ LASSERT(current->journal_info == desc_private);
+ journal = EXT3_SB(inode->i_sb)->s_journal;
+ if (nblocks > journal->j_max_transaction_buffers) {
+ CWARN("too many credits %d for op %ux%u using %d instead\n",
+ nblocks, op, logs, journal->j_max_transaction_buffers);
+ nblocks = journal->j_max_transaction_buffers;
+ }
+
+ journal_start:
+ LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
+ handle = ext3_journal_start(inode, nblocks);
+
+ if (!IS_ERR(handle))
+ LASSERT(current->journal_info == handle);
+ else
+ CERROR("error starting handle for op %u (%u credits): rc %ld\n",
+ op, nblocks, PTR_ERR(handle));
+ return handle;
+}
+
+static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
+{
+ int rc;
+ handle_t *handle = h;
+
+ LASSERT(current->journal_info == handle);
+ if (force_sync)
+ handle->h_sync = 1; /* recovery likes this */
+
+ rc = ext3_journal_stop(handle);
+
+ return rc;
+}
+
#ifndef EXT3_EXTENTS_FL
#define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
#endif
return rc;
}
+int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
+{
+ unsigned long block;
+ struct buffer_head *bh;
+ int err, blocksize, csize, boffs, osize = size;
+
+ /* prevent reading after eof */
+ spin_lock(&inode->i_lock);
+ if (i_size_read(inode) < *offs + size) {
+ size = i_size_read(inode) - *offs;
+ spin_unlock(&inode->i_lock);
+ if (size < 0) {
+ CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
+ i_size_read(inode), *offs);
+ return -EBADR;
+ } else if (size == 0) {
+ return 0;
+ }
+ } else {
+ spin_unlock(&inode->i_lock);
+ }
+
+ blocksize = 1 << inode->i_blkbits;
+
+ while (size > 0) {
+ block = *offs >> inode->i_blkbits;
+ boffs = *offs & (blocksize - 1);
+ csize = min(blocksize - boffs, size);
+ bh = ext3_bread(NULL, inode, block, 0, &err);
+ if (!bh) {
+ CERROR("can't read block: %d\n", err);
+ return err;
+ }
+
+ memcpy(buf, bh->b_data + boffs, csize);
+ brelse(bh);
+
+ *offs += csize;
+ buf += csize;
+ size -= csize;
+ }
+ return osize;
+}
+EXPORT_SYMBOL(fsfilt_ext3_read);
+
+static int fsfilt_ext3_read_record(struct file * file, void *buf,
+ int size, loff_t *offs)
+{
+ int rc;
+ rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
+ if (rc > 0)
+ rc = 0;
+ return rc;
+}
+
+int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
+ loff_t *offs, handle_t *handle)
+{
+ struct buffer_head *bh = NULL;
+ loff_t old_size = i_size_read(inode), offset = *offs;
+ loff_t new_size = i_size_read(inode);
+ unsigned long block;
+ int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
+
+ while (bufsize > 0) {
+ if (bh != NULL)
+ brelse(bh);
+
+ block = offset >> inode->i_blkbits;
+ boffs = offset & (blocksize - 1);
+ size = min(blocksize - boffs, bufsize);
+ bh = ext3_bread(handle, inode, block, 1, &err);
+ if (!bh) {
+ CERROR("can't read/create block: %d\n", err);
+ break;
+ }
+
+ err = ext3_journal_get_write_access(handle, bh);
+ if (err) {
+ CERROR("journal_get_write_access() returned error %d\n",
+ err);
+ break;
+ }
+ LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
+ memcpy(bh->b_data + boffs, buf, size);
+ err = ext3_journal_dirty_metadata(handle, bh);
+ if (err) {
+ CERROR("journal_dirty_metadata() returned error %d\n",
+ err);
+ break;
+ }
+ if (offset + size > new_size)
+ new_size = offset + size;
+ offset += size;
+ bufsize -= size;
+ buf += size;
+ }
+ if (bh)
+ brelse(bh);
+
+ /* correct in-core and on-disk sizes */
+ if (new_size > i_size_read(inode)) {
+ spin_lock(&inode->i_lock);
+ if (new_size > i_size_read(inode))
+ i_size_write(inode, new_size);
+ if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
+ EXT3_I(inode)->i_disksize = i_size_read(inode);
+ if (i_size_read(inode) > old_size) {
+ spin_unlock(&inode->i_lock);
+ mark_inode_dirty(inode);
+ } else {
+ spin_unlock(&inode->i_lock);
+ }
+ }
+
+ if (err == 0)
+ *offs = offset;
+ return err;
+}
+EXPORT_SYMBOL(fsfilt_ext3_write_handle);
+
+static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
+ loff_t *offs, int force_sync)
+{
+ struct inode *inode = file->f_dentry->d_inode;
+ handle_t *handle;
+ int err, block_count = 0, blocksize;
+
+ /* Determine how many transaction credits are needed */
+ blocksize = 1 << inode->i_blkbits;
+ block_count = (*offs & (blocksize - 1)) + bufsize;
+ block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
+
+ handle = ext3_journal_start(inode,
+ block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
+ if (IS_ERR(handle)) {
+ CERROR("can't start transaction for %d blocks (%d bytes)\n",
+ block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2,
+ bufsize);
+ return PTR_ERR(handle);
+ }
+
+ err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
+
+ if (!err && force_sync)
+ handle->h_sync = 1; /* recovery likes this */
+
+ ext3_journal_stop(handle);
+
+ return err;
+}
+
+static int fsfilt_ext3_setup(struct super_block *sb)
+{
+ if (!EXT3_HAS_COMPAT_FEATURE(sb,
+ EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
+ CERROR("ext3 mounted without journal\n");
+ return -EINVAL;
+ }
+
+#ifdef S_PDIROPS
+ CWARN("Enabling PDIROPS\n");
+ set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
+ sb->s_flags |= S_PDIROPS;
+#endif
+ if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
+ CWARN("filesystem doesn't have dir_index feature enabled\n");
+ return 0;
+}
static struct fsfilt_operations fsfilt_ext3_ops = {
- .fs_type = "ext3",
- .fs_owner = THIS_MODULE,
- .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
+ .fs_type = "ext3",
+ .fs_owner = THIS_MODULE,
+ .fs_getlabel = fsfilt_ext3_get_label,
+ .fs_start = fsfilt_ext3_start,
+ .fs_commit = fsfilt_ext3_commit,
+ .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
+ .fs_write_record = fsfilt_ext3_write_record,
+ .fs_read_record = fsfilt_ext3_read_record,
+ .fs_setup = fsfilt_ext3_setup,
};
static int __init fsfilt_ext3_init(void)
{
- return fsfilt_register_ops(&fsfilt_ext3_ops);
+ int rc;
+
+ fcb_cache = kmem_cache_create("fsfilt_ext3_fcb",
+ sizeof(struct fsfilt_cb_data),
+ 0, 0, NULL);
+ if (!fcb_cache) {
+ CERROR("error allocating fsfilt journal callback cache\n");
+ GOTO(out, rc = -ENOMEM);
+ }
+
+ rc = fsfilt_register_ops(&fsfilt_ext3_ops);
+
+ if (rc)
+ kmem_cache_destroy(fcb_cache);
+out:
+ return rc;
}
static void __exit fsfilt_ext3_exit(void)
{
fsfilt_unregister_ops(&fsfilt_ext3_ops);
+ kmem_cache_destroy(fcb_cache);
}
module_init(fsfilt_ext3_init);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff