direct-io: merge direct_io_walker into __blockdev_direct_IO
authorAndi Kleen <ak@linux.intel.com>
Tue, 2 Aug 2011 04:38:09 +0000 (21:38 -0700)
committerroot <root@serles.lst.de>
Fri, 28 Oct 2011 12:58:58 +0000 (14:58 +0200)
This doesn't change anything for the compiler, but hch thought it would
make the code clearer.

I moved the reference counting into its own little inline.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
fs/direct-io.c

index 6d425821be6673c8195561365edfadd3b2f7cc84..d740ab67ff6e1bd581218fb149eb7a58c40fa3f3 100644 (file)
@@ -1043,136 +1043,10 @@ out:
        return ret;
 }
 
-static inline ssize_t
-direct_io_worker(int rw, struct kiocb *iocb, struct inode *inode, 
-       const struct iovec *iov, loff_t offset, unsigned long nr_segs, 
-       unsigned blkbits, get_block_t get_block, dio_iodone_t end_io,
-       dio_submit_t submit_io, struct dio *dio, struct dio_submit *sdio)
+static inline int drop_refcount(struct dio *dio)
 {
-       unsigned long user_addr; 
+       int ret2;
        unsigned long flags;
-       int seg;
-       ssize_t ret = 0;
-       ssize_t ret2;
-       size_t bytes;
-       struct buffer_head map_bh = { 0, };
-
-       dio->inode = inode;
-       dio->rw = rw;
-       sdio->blkbits = blkbits;
-       sdio->blkfactor = inode->i_blkbits - blkbits;
-       sdio->block_in_file = offset >> blkbits;
-
-       sdio->get_block = get_block;
-       dio->end_io = end_io;
-       sdio->submit_io = submit_io;
-       sdio->final_block_in_bio = -1;
-       sdio->next_block_for_io = -1;
-
-       dio->iocb = iocb;
-       dio->i_size = i_size_read(inode);
-
-       spin_lock_init(&dio->bio_lock);
-       dio->refcount = 1;
-
-       /*
-        * In case of non-aligned buffers, we may need 2 more
-        * pages since we need to zero out first and last block.
-        */
-       if (unlikely(sdio->blkfactor))
-               sdio->pages_in_io = 2;
-
-       for (seg = 0; seg < nr_segs; seg++) {
-               user_addr = (unsigned long)iov[seg].iov_base;
-               sdio->pages_in_io +=
-                       ((user_addr+iov[seg].iov_len +PAGE_SIZE-1)/PAGE_SIZE
-                               - user_addr/PAGE_SIZE);
-       }
-
-       for (seg = 0; seg < nr_segs; seg++) {
-               user_addr = (unsigned long)iov[seg].iov_base;
-               sdio->size += bytes = iov[seg].iov_len;
-
-               /* Index into the first page of the first block */
-               sdio->first_block_in_page = (user_addr & ~PAGE_MASK) >> blkbits;
-               sdio->final_block_in_request = sdio->block_in_file +
-                                               (bytes >> blkbits);
-               /* Page fetching state */
-               sdio->head = 0;
-               sdio->tail = 0;
-               sdio->curr_page = 0;
-
-               sdio->total_pages = 0;
-               if (user_addr & (PAGE_SIZE-1)) {
-                       sdio->total_pages++;
-                       bytes -= PAGE_SIZE - (user_addr & (PAGE_SIZE - 1));
-               }
-               sdio->total_pages += (bytes + PAGE_SIZE - 1) / PAGE_SIZE;
-               sdio->curr_user_address = user_addr;
-       
-               ret = do_direct_IO(dio, sdio, &map_bh);
-
-               dio->result += iov[seg].iov_len -
-                       ((sdio->final_block_in_request - sdio->block_in_file) <<
-                                       blkbits);
-
-               if (ret) {
-                       dio_cleanup(dio, sdio);
-                       break;
-               }
-       } /* end iovec loop */
-
-       if (ret == -ENOTBLK) {
-               /*
-                * The remaining part of the request will be
-                * be handled by buffered I/O when we return
-                */
-               ret = 0;
-       }
-       /*
-        * There may be some unwritten disk at the end of a part-written
-        * fs-block-sized block.  Go zero that now.
-        */
-       dio_zero_block(dio, sdio, 1, &map_bh);
-
-       if (sdio->cur_page) {
-               ret2 = dio_send_cur_page(dio, sdio, &map_bh);
-               if (ret == 0)
-                       ret = ret2;
-               page_cache_release(sdio->cur_page);
-               sdio->cur_page = NULL;
-       }
-       if (sdio->bio)
-               dio_bio_submit(dio, sdio);
-
-       /*
-        * It is possible that, we return short IO due to end of file.
-        * In that case, we need to release all the pages we got hold on.
-        */
-       dio_cleanup(dio, sdio);
-
-       /*
-        * All block lookups have been performed. For READ requests
-        * we can let i_mutex go now that its achieved its purpose
-        * of protecting us from looking up uninitialized blocks.
-        */
-       if (rw == READ && (dio->flags & DIO_LOCKING))
-               mutex_unlock(&dio->inode->i_mutex);
-
-       /*
-        * The only time we want to leave bios in flight is when a successful
-        * partial aio read or full aio write have been setup.  In that case
-        * bio completion will call aio_complete.  The only time it's safe to
-        * call aio_complete is when we return -EIOCBQUEUED, so we key on that.
-        * This had *better* be the only place that raises -EIOCBQUEUED.
-        */
-       BUG_ON(ret == -EIOCBQUEUED);
-       if (dio->is_async && ret == 0 && dio->result &&
-           ((rw & READ) || (dio->result == sdio->size)))
-               ret = -EIOCBQUEUED;
-
-       if (ret != -EIOCBQUEUED)
-               dio_await_completion(dio);
 
        /*
         * Sync will always be dropping the final ref and completing the
@@ -1188,14 +1062,7 @@ direct_io_worker(int rw, struct kiocb *iocb, struct inode *inode,
        spin_lock_irqsave(&dio->bio_lock, flags);
        ret2 = --dio->refcount;
        spin_unlock_irqrestore(&dio->bio_lock, flags);
-
-       if (ret2 == 0) {
-               ret = dio_complete(dio, offset, ret, false);
-               kmem_cache_free(dio_cache, dio);
-       } else
-               BUG_ON(ret != -EIOCBQUEUED);
-
-       return ret;
+       return ret2;
 }
 
 /*
@@ -1239,6 +1106,9 @@ __blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
        loff_t end = offset;
        struct dio *dio;
        struct dio_submit sdio = { 0, };
+       unsigned long user_addr;
+       size_t bytes;
+       struct buffer_head map_bh = { 0, };
 
        if (rw & WRITE)
                rw = WRITE_ODIRECT;
@@ -1316,9 +1186,132 @@ __blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
        dio->is_async = !is_sync_kiocb(iocb) && !((rw & WRITE) &&
                (end > i_size_read(inode)));
 
-       retval = direct_io_worker(rw, iocb, inode, iov, offset,
-                               nr_segs, blkbits, get_block, end_io,
-                                 submit_io, dio, &sdio);
+       retval = 0;
+
+       dio->inode = inode;
+       dio->rw = rw;
+       sdio.blkbits = blkbits;
+       sdio.blkfactor = inode->i_blkbits - blkbits;
+       sdio.block_in_file = offset >> blkbits;
+
+       sdio.get_block = get_block;
+       dio->end_io = end_io;
+       sdio.submit_io = submit_io;
+       sdio.final_block_in_bio = -1;
+       sdio.next_block_for_io = -1;
+
+       dio->iocb = iocb;
+       dio->i_size = i_size_read(inode);
+
+       spin_lock_init(&dio->bio_lock);
+       dio->refcount = 1;
+
+       /*
+        * In case of non-aligned buffers, we may need 2 more
+        * pages since we need to zero out first and last block.
+        */
+       if (unlikely(sdio.blkfactor))
+               sdio.pages_in_io = 2;
+
+       for (seg = 0; seg < nr_segs; seg++) {
+               user_addr = (unsigned long)iov[seg].iov_base;
+               sdio.pages_in_io +=
+                       ((user_addr + iov[seg].iov_len + PAGE_SIZE-1) /
+                               PAGE_SIZE - user_addr / PAGE_SIZE);
+       }
+
+       for (seg = 0; seg < nr_segs; seg++) {
+               user_addr = (unsigned long)iov[seg].iov_base;
+               sdio.size += bytes = iov[seg].iov_len;
+
+               /* Index into the first page of the first block */
+               sdio.first_block_in_page = (user_addr & ~PAGE_MASK) >> blkbits;
+               sdio.final_block_in_request = sdio.block_in_file +
+                                               (bytes >> blkbits);
+               /* Page fetching state */
+               sdio.head = 0;
+               sdio.tail = 0;
+               sdio.curr_page = 0;
+
+               sdio.total_pages = 0;
+               if (user_addr & (PAGE_SIZE-1)) {
+                       sdio.total_pages++;
+                       bytes -= PAGE_SIZE - (user_addr & (PAGE_SIZE - 1));
+               }
+               sdio.total_pages += (bytes + PAGE_SIZE - 1) / PAGE_SIZE;
+               sdio.curr_user_address = user_addr;
+
+               retval = do_direct_IO(dio, &sdio, &map_bh);
+
+               dio->result += iov[seg].iov_len -
+                       ((sdio.final_block_in_request - sdio.block_in_file) <<
+                                       blkbits);
+
+               if (retval) {
+                       dio_cleanup(dio, &sdio);
+                       break;
+               }
+       } /* end iovec loop */
+
+       if (retval == -ENOTBLK) {
+               /*
+                * The remaining part of the request will be
+                * be handled by buffered I/O when we return
+                */
+               retval = 0;
+       }
+       /*
+        * There may be some unwritten disk at the end of a part-written
+        * fs-block-sized block.  Go zero that now.
+        */
+       dio_zero_block(dio, &sdio, 1, &map_bh);
+
+       if (sdio.cur_page) {
+               ssize_t ret2;
+
+               ret2 = dio_send_cur_page(dio, &sdio, &map_bh);
+               if (retval == 0)
+                       retval = ret2;
+               page_cache_release(sdio.cur_page);
+               sdio.cur_page = NULL;
+       }
+       if (sdio.bio)
+               dio_bio_submit(dio, &sdio);
+
+       /*
+        * It is possible that, we return short IO due to end of file.
+        * In that case, we need to release all the pages we got hold on.
+        */
+       dio_cleanup(dio, &sdio);
+
+       /*
+        * All block lookups have been performed. For READ requests
+        * we can let i_mutex go now that its achieved its purpose
+        * of protecting us from looking up uninitialized blocks.
+        */
+       if (rw == READ && (dio->flags & DIO_LOCKING))
+               mutex_unlock(&dio->inode->i_mutex);
+
+       /*
+        * The only time we want to leave bios in flight is when a successful
+        * partial aio read or full aio write have been setup.  In that case
+        * bio completion will call aio_complete.  The only time it's safe to
+        * call aio_complete is when we return -EIOCBQUEUED, so we key on that.
+        * This had *better* be the only place that raises -EIOCBQUEUED.
+        */
+       BUG_ON(retval == -EIOCBQUEUED);
+       if (dio->is_async && retval == 0 && dio->result &&
+           ((rw & READ) || (dio->result == sdio.size)))
+               retval = -EIOCBQUEUED;
+
+       if (retval != -EIOCBQUEUED)
+               dio_await_completion(dio);
+
+       if (drop_refcount(dio) == 0) {
+               retval = dio_complete(dio, offset, retval, false);
+               kmem_cache_free(dio_cache, dio);
+       } else
+               BUG_ON(retval != -EIOCBQUEUED);
 
 out:
        return retval;