+EXPORT_SYMBOL(ll_direct_rw_pages);
+
+static ssize_t ll_direct_IO_26_seg(const struct lu_env *env, struct cl_io *io,
+ int rw, struct inode *inode,
+ struct address_space *mapping,
+ size_t size, loff_t file_offset,
+ struct page **pages, int page_count)
+{
+ struct ll_dio_pages pvec = { .ldp_pages = pages,
+ .ldp_nr = page_count,
+ .ldp_size = size,
+ .ldp_offsets = NULL,
+ .ldp_start_offset = file_offset
+ };
+
+ return ll_direct_rw_pages(env, io, rw, inode, &pvec);
+}
+
+/* This is the maximum size of a single O_DIRECT request, based on a 128kB
+ * kmalloc limit. We need to fit all of the brw_page structs, each one
+ * representing PAGE_SIZE worth of user data, into a single buffer, and
+ * then truncate this to be a full-sized RPC. This is 22MB for 4kB pages. */
+#define MAX_DIO_SIZE ((128 * 1024 / sizeof(struct brw_page) * CFS_PAGE_SIZE) & \
+ ~(PTLRPC_MAX_BRW_SIZE - 1))
+static ssize_t ll_direct_IO_26(int rw, struct kiocb *iocb,
+ const struct iovec *iov, loff_t file_offset,
+ unsigned long nr_segs)
+{
+ struct lu_env *env;
+ struct cl_io *io;
+ struct file *file = iocb->ki_filp;
+ struct inode *inode = file->f_mapping->host;
+ struct ccc_object *obj = cl_inode2ccc(inode);
+ ssize_t count = iov_length(iov, nr_segs);
+ ssize_t tot_bytes = 0, result = 0;
+ struct ll_inode_info *lli = ll_i2info(inode);
+ struct lov_stripe_md *lsm = lli->lli_smd;
+ unsigned long seg = 0;
+ size_t size = MAX_DIO_SIZE;
+ int refcheck;
+ ENTRY;
+
+ if (!lli->lli_smd || !lli->lli_smd->lsm_object_id)
+ RETURN(-EBADF);
+
+ /* FIXME: io smaller than PAGE_SIZE is broken on ia64 ??? */
+ if ((file_offset & ~CFS_PAGE_MASK) || (count & ~CFS_PAGE_MASK))
+ RETURN(-EINVAL);
+
+ CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), size="LPSZ" (max %lu), "
+ "offset=%lld=%llx, pages "LPSZ" (max %lu)\n",
+ inode->i_ino, inode->i_generation, inode, count, MAX_DIO_SIZE,
+ file_offset, file_offset, count >> CFS_PAGE_SHIFT,
+ MAX_DIO_SIZE >> CFS_PAGE_SHIFT);
+
+ /* Check that all user buffers are aligned as well */
+ for (seg = 0; seg < nr_segs; seg++) {
+ if (((unsigned long)iov[seg].iov_base & ~CFS_PAGE_MASK) ||
+ (iov[seg].iov_len & ~CFS_PAGE_MASK))
+ RETURN(-EINVAL);
+ }
+
+ env = cl_env_get(&refcheck);
+ LASSERT(!IS_ERR(env));
+ io = ccc_env_io(env)->cui_cl.cis_io;
+ LASSERT(io != NULL);
+
+ /* 0. Need locking between buffered and direct access. and race with
+ *size changing by concurrent truncates and writes.
+ * 1. Need inode sem to operate transient pages. */
+ if (rw == READ)
+ LOCK_INODE_MUTEX(inode);
+
+ LASSERT(obj->cob_transient_pages == 0);
+ for (seg = 0; seg < nr_segs; seg++) {
+ size_t iov_left = iov[seg].iov_len;
+ unsigned long user_addr = (unsigned long)iov[seg].iov_base;
+
+ if (rw == READ) {
+ if (file_offset >= inode->i_size)
+ break;
+ if (file_offset + iov_left > inode->i_size)
+ iov_left = inode->i_size - file_offset;
+ }
+
+ while (iov_left > 0) {
+ struct page **pages;
+ int page_count, max_pages = 0;
+ size_t bytes;
+
+ bytes = min(size,iov_left);
+ page_count = ll_get_user_pages(rw, user_addr,
+ bytes,
+ &pages, &max_pages);
+ if (likely(page_count > 0)) {
+ if (unlikely(page_count < max_pages))
+ bytes = page_count << CFS_PAGE_SHIFT;
+ result = ll_direct_IO_26_seg(env, io, rw, inode,
+ file->f_mapping,
+ bytes,
+ file_offset, pages,
+ page_count);
+ ll_free_user_pages(pages, max_pages, rw==READ);
+ } else if (page_count == 0) {
+ GOTO(out, result = -EFAULT);
+ } else {
+ result = page_count;
+ }
+ if (unlikely(result <= 0)) {
+ /* If we can't allocate a large enough buffer
+ * for the request, shrink it to a smaller
+ * PAGE_SIZE multiple and try again.
+ * We should always be able to kmalloc for a
+ * page worth of page pointers = 4MB on i386. */
+ if (result == -ENOMEM &&
+ size > (CFS_PAGE_SIZE / sizeof(*pages)) *
+ CFS_PAGE_SIZE) {
+ size = ((((size / 2) - 1) |
+ ~CFS_PAGE_MASK) + 1) &
+ CFS_PAGE_MASK;
+ CDEBUG(D_VFSTRACE, "DIO size now %u\n",
+ (int)size);
+ continue;
+ }
+
+ GOTO(out, result);
+ }
+
+ tot_bytes += result;
+ file_offset += result;
+ iov_left -= result;
+ user_addr += result;
+ }
+ }
+out:
+ LASSERT(obj->cob_transient_pages == 0);
+ if (rw == READ)
+ UNLOCK_INODE_MUTEX(inode);
+
+ if (tot_bytes > 0) {
+ if (rw == WRITE) {
+ lov_stripe_lock(lsm);
+ obd_adjust_kms(ll_i2dtexp(inode), lsm, file_offset, 0);
+ lov_stripe_unlock(lsm);
+ }
+ }
+
+ cl_env_put(env, &refcheck);
+ RETURN(tot_bytes ? : result);
+}