Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / fs / reiserfs / file.c
diff --git a/kernel/fs/reiserfs/file.c b/kernel/fs/reiserfs/file.c
new file mode 100644 (file)
index 0000000..96a1bcf
--- /dev/null
@@ -0,0 +1,270 @@
+/*
+ * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
+ */
+
+#include <linux/time.h>
+#include "reiserfs.h"
+#include "acl.h"
+#include "xattr.h"
+#include <linux/uaccess.h>
+#include <linux/pagemap.h>
+#include <linux/swap.h>
+#include <linux/writeback.h>
+#include <linux/blkdev.h>
+#include <linux/buffer_head.h>
+#include <linux/quotaops.h>
+
+/*
+ * We pack the tails of files on file close, not at the time they are written.
+ * This implies an unnecessary copy of the tail and an unnecessary indirect item
+ * insertion/balancing, for files that are written in one write.
+ * It avoids unnecessary tail packings (balances) for files that are written in
+ * multiple writes and are small enough to have tails.
+ *
+ * file_release is called by the VFS layer when the file is closed.  If
+ * this is the last open file descriptor, and the file
+ * small enough to have a tail, and the tail is currently in an
+ * unformatted node, the tail is converted back into a direct item.
+ *
+ * We use reiserfs_truncate_file to pack the tail, since it already has
+ * all the conditions coded.
+ */
+static int reiserfs_file_release(struct inode *inode, struct file *filp)
+{
+
+       struct reiserfs_transaction_handle th;
+       int err;
+       int jbegin_failure = 0;
+
+       BUG_ON(!S_ISREG(inode->i_mode));
+
+        if (atomic_add_unless(&REISERFS_I(inode)->openers, -1, 1))
+               return 0;
+
+       mutex_lock(&REISERFS_I(inode)->tailpack);
+
+        if (!atomic_dec_and_test(&REISERFS_I(inode)->openers)) {
+               mutex_unlock(&REISERFS_I(inode)->tailpack);
+               return 0;
+       }
+
+       /* fast out for when nothing needs to be done */
+       if ((!(REISERFS_I(inode)->i_flags & i_pack_on_close_mask) ||
+            !tail_has_to_be_packed(inode)) &&
+           REISERFS_I(inode)->i_prealloc_count <= 0) {
+               mutex_unlock(&REISERFS_I(inode)->tailpack);
+               return 0;
+       }
+
+       reiserfs_write_lock(inode->i_sb);
+       /*
+        * freeing preallocation only involves relogging blocks that
+        * are already in the current transaction.  preallocation gets
+        * freed at the end of each transaction, so it is impossible for
+        * us to log any additional blocks (including quota blocks)
+        */
+       err = journal_begin(&th, inode->i_sb, 1);
+       if (err) {
+               /*
+                * uh oh, we can't allow the inode to go away while there
+                * is still preallocation blocks pending.  Try to join the
+                * aborted transaction
+                */
+               jbegin_failure = err;
+               err = journal_join_abort(&th, inode->i_sb);
+
+               if (err) {
+                       /*
+                        * hmpf, our choices here aren't good.  We can pin
+                        * the inode which will disallow unmount from ever
+                        * happening, we can do nothing, which will corrupt
+                        * random memory on unmount, or we can forcibly
+                        * remove the file from the preallocation list, which
+                        * will leak blocks on disk.  Lets pin the inode
+                        * and let the admin know what is going on.
+                        */
+                       igrab(inode);
+                       reiserfs_warning(inode->i_sb, "clm-9001",
+                                        "pinning inode %lu because the "
+                                        "preallocation can't be freed",
+                                        inode->i_ino);
+                       goto out;
+               }
+       }
+       reiserfs_update_inode_transaction(inode);
+
+#ifdef REISERFS_PREALLOCATE
+       reiserfs_discard_prealloc(&th, inode);
+#endif
+       err = journal_end(&th);
+
+       /* copy back the error code from journal_begin */
+       if (!err)
+               err = jbegin_failure;
+
+       if (!err &&
+           (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) &&
+           tail_has_to_be_packed(inode)) {
+
+               /*
+                * if regular file is released by last holder and it has been
+                * appended (we append by unformatted node only) or its direct
+                * item(s) had to be converted, then it may have to be
+                * indirect2direct converted
+                */
+               err = reiserfs_truncate_file(inode, 0);
+       }
+out:
+       reiserfs_write_unlock(inode->i_sb);
+       mutex_unlock(&REISERFS_I(inode)->tailpack);
+       return err;
+}
+
+static int reiserfs_file_open(struct inode *inode, struct file *file)
+{
+       int err = dquot_file_open(inode, file);
+
+       /* somebody might be tailpacking on final close; wait for it */
+        if (!atomic_inc_not_zero(&REISERFS_I(inode)->openers)) {
+               mutex_lock(&REISERFS_I(inode)->tailpack);
+               atomic_inc(&REISERFS_I(inode)->openers);
+               mutex_unlock(&REISERFS_I(inode)->tailpack);
+       }
+       return err;
+}
+
+void reiserfs_vfs_truncate_file(struct inode *inode)
+{
+       mutex_lock(&REISERFS_I(inode)->tailpack);
+       reiserfs_truncate_file(inode, 1);
+       mutex_unlock(&REISERFS_I(inode)->tailpack);
+}
+
+/* Sync a reiserfs file. */
+
+/*
+ * FIXME: sync_mapping_buffers() never has anything to sync.  Can
+ * be removed...
+ */
+
+static int reiserfs_sync_file(struct file *filp, loff_t start, loff_t end,
+                             int datasync)
+{
+       struct inode *inode = filp->f_mapping->host;
+       int err;
+       int barrier_done;
+
+       err = filemap_write_and_wait_range(inode->i_mapping, start, end);
+       if (err)
+               return err;
+
+       mutex_lock(&inode->i_mutex);
+       BUG_ON(!S_ISREG(inode->i_mode));
+       err = sync_mapping_buffers(inode->i_mapping);
+       reiserfs_write_lock(inode->i_sb);
+       barrier_done = reiserfs_commit_for_inode(inode);
+       reiserfs_write_unlock(inode->i_sb);
+       if (barrier_done != 1 && reiserfs_barrier_flush(inode->i_sb))
+               blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
+       mutex_unlock(&inode->i_mutex);
+       if (barrier_done < 0)
+               return barrier_done;
+       return (err < 0) ? -EIO : 0;
+}
+
+/* taken fs/buffer.c:__block_commit_write */
+int reiserfs_commit_page(struct inode *inode, struct page *page,
+                        unsigned from, unsigned to)
+{
+       unsigned block_start, block_end;
+       int partial = 0;
+       unsigned blocksize;
+       struct buffer_head *bh, *head;
+       unsigned long i_size_index = inode->i_size >> PAGE_CACHE_SHIFT;
+       int new;
+       int logit = reiserfs_file_data_log(inode);
+       struct super_block *s = inode->i_sb;
+       int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
+       struct reiserfs_transaction_handle th;
+       int ret = 0;
+
+       th.t_trans_id = 0;
+       blocksize = 1 << inode->i_blkbits;
+
+       if (logit) {
+               reiserfs_write_lock(s);
+               ret = journal_begin(&th, s, bh_per_page + 1);
+               if (ret)
+                       goto drop_write_lock;
+               reiserfs_update_inode_transaction(inode);
+       }
+       for (bh = head = page_buffers(page), block_start = 0;
+            bh != head || !block_start;
+            block_start = block_end, bh = bh->b_this_page) {
+
+               new = buffer_new(bh);
+               clear_buffer_new(bh);
+               block_end = block_start + blocksize;
+               if (block_end <= from || block_start >= to) {
+                       if (!buffer_uptodate(bh))
+                               partial = 1;
+               } else {
+                       set_buffer_uptodate(bh);
+                       if (logit) {
+                               reiserfs_prepare_for_journal(s, bh, 1);
+                               journal_mark_dirty(&th, bh);
+                       } else if (!buffer_dirty(bh)) {
+                               mark_buffer_dirty(bh);
+                               /*
+                                * do data=ordered on any page past the end
+                                * of file and any buffer marked BH_New.
+                                */
+                               if (reiserfs_data_ordered(inode->i_sb) &&
+                                   (new || page->index >= i_size_index)) {
+                                       reiserfs_add_ordered_list(inode, bh);
+                               }
+                       }
+               }
+       }
+       if (logit) {
+               ret = journal_end(&th);
+drop_write_lock:
+               reiserfs_write_unlock(s);
+       }
+       /*
+        * If this is a partial write which happened to make all buffers
+        * uptodate then we can optimize away a bogus readpage() for
+        * the next read(). Here we 'discover' whether the page went
+        * uptodate as a result of this (potentially partial) write.
+        */
+       if (!partial)
+               SetPageUptodate(page);
+       return ret;
+}
+
+const struct file_operations reiserfs_file_operations = {
+       .unlocked_ioctl = reiserfs_ioctl,
+#ifdef CONFIG_COMPAT
+       .compat_ioctl = reiserfs_compat_ioctl,
+#endif
+       .mmap = generic_file_mmap,
+       .open = reiserfs_file_open,
+       .release = reiserfs_file_release,
+       .fsync = reiserfs_sync_file,
+       .read_iter = generic_file_read_iter,
+       .write_iter = generic_file_write_iter,
+       .splice_read = generic_file_splice_read,
+       .splice_write = iter_file_splice_write,
+       .llseek = generic_file_llseek,
+};
+
+const struct inode_operations reiserfs_file_inode_operations = {
+       .setattr = reiserfs_setattr,
+       .setxattr = reiserfs_setxattr,
+       .getxattr = reiserfs_getxattr,
+       .listxattr = reiserfs_listxattr,
+       .removexattr = reiserfs_removexattr,
+       .permission = reiserfs_permission,
+       .get_acl = reiserfs_get_acl,
+       .set_acl = reiserfs_set_acl,
+};