+#if 0
+ struct cl_page *page = vvp_vmpage_page_transient(vmpage);
+ struct vvp_object *obj = cl_inode2vvp(vmpage->mapping->host);
+ struct vvp_page *cpg;
+
+ /*
+ * XXX should page method be called here?
+ */
+ LASSERT(&obj->co_cl == page->cp_obj);
+ cpg = cl2vvp_page(cl_page_at(page, &vvp_device_type));
+ /*
+ * XXX cannot do much here, because page is possibly not locked:
+ * sys_munmap()->...
+ * ->unmap_page_range()->zap_pte_range()->set_page_dirty().
+ */
+ vvp_write_pending(obj, cpg);
+#endif
+ RETURN(__set_page_dirty_nobuffers(vmpage));
+}
+
+#define MAX_DIRECTIO_SIZE 2*1024*1024*1024UL
+
+static inline int ll_get_user_pages(int rw, unsigned long user_addr,
+ size_t size, struct page ***pages,
+ int *max_pages)
+{
+ int result = -ENOMEM;
+
+ /* set an arbitrary limit to prevent arithmetic overflow */
+ if (size > MAX_DIRECTIO_SIZE) {
+ *pages = NULL;
+ return -EFBIG;
+ }
+
+ *max_pages = (user_addr + size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
+ *max_pages -= user_addr >> CFS_PAGE_SHIFT;
+
+ OBD_ALLOC_WAIT(*pages, *max_pages * sizeof(**pages));
+ if (*pages) {
+ down_read(¤t->mm->mmap_sem);
+ result = get_user_pages(current, current->mm, user_addr,
+ *max_pages, (rw == READ), 0, *pages,
+ NULL);
+ up_read(¤t->mm->mmap_sem);
+ if (unlikely(result <= 0))
+ OBD_FREE(*pages, *max_pages * sizeof(**pages));
+ }
+
+ return result;
+}
+
+/* ll_free_user_pages - tear down page struct array
+ * @pages: array of page struct pointers underlying target buffer */
+static void ll_free_user_pages(struct page **pages, int npages, int do_dirty)
+{
+ int i;
+
+ for (i = 0; i < npages; i++) {
+ if (pages[i] == NULL)
+ break;
+ if (do_dirty)
+ set_page_dirty_lock(pages[i]);
+ page_cache_release(pages[i]);
+ }
+
+ OBD_FREE(pages, npages * sizeof(*pages));
+}
+
+ssize_t ll_direct_rw_pages(const struct lu_env *env, struct cl_io *io,
+ int rw, struct inode *inode,
+ struct ll_dio_pages *pv)
+{
+ struct cl_page *clp;
+ struct cl_2queue *queue;
+ struct cl_object *obj = io->ci_obj;
+ int i;
+ ssize_t rc = 0;
+ loff_t file_offset = pv->ldp_start_offset;
+ long size = pv->ldp_size;
+ int page_count = pv->ldp_nr;
+ struct page **pages = pv->ldp_pages;
+ long page_size = cl_page_size(obj);
+ ENTRY;
+
+ queue = &io->ci_queue;
+ cl_2queue_init(queue);
+ for (i = 0; i < page_count; i++) {
+ if (pv->ldp_offsets)
+ file_offset = pv->ldp_offsets[i];
+ LASSERT(!(file_offset & (page_size - 1)));
+ clp = cl_page_find(env, obj, cl_index(obj, file_offset),
+ pv->ldp_pages[i], CPT_TRANSIENT);
+ if (IS_ERR(clp)) {
+ rc = PTR_ERR(clp);
+ break;
+ }
+
+ /* check the page type: if the page is a host page, then do
+ * write directly */
+ /*
+ * Very rare case that the host pages can be found for
+ * directIO case, since linux kernel truncated all covered
+ * pages before getting here. So, to make the OST happy(to
+ * write a contiguous region), all pages are issued
+ * here. -jay */
+ if (clp->cp_type == CPT_CACHEABLE) {
+ cfs_page_t *vmpage = cl_page_vmpage(env, clp);
+ cfs_page_t *src_page;
+ cfs_page_t *dst_page;
+ void *src;
+ void *dst;
+
+ src_page = (rw == WRITE) ? pages[i] : vmpage;
+ dst_page = (rw == WRITE) ? vmpage : pages[i];
+
+ src = kmap_atomic(src_page, KM_USER0);
+ dst = kmap_atomic(dst_page, KM_USER1);
+ memcpy(dst, src, min(page_size, size));
+ kunmap_atomic(dst, KM_USER1);
+ kunmap_atomic(src, KM_USER0);
+
+ /* make sure page will be added to the transfer by
+ * cl_io_submit()->...->vvp_page_prep_write(). */
+ if (rw == WRITE)
+ set_page_dirty(vmpage);
+ /*
+ * If direct-io read finds up-to-date page in the
+ * cache, just copy it to the user space. Page will be
+ * filtered out by vvp_page_prep_read(). This
+ * preserves an invariant, that page is read at most
+ * once, see cl_page_flags::CPF_READ_COMPLETED.
+ */
+ }
+
+ rc = cl_page_own(env, io, clp);
+ if (rc) {
+ LASSERT(clp->cp_state == CPS_FREEING);
+ cl_page_put(env, clp);
+ break;
+ }
+
+ cl_2queue_add(queue, clp);
+
+ /* drop the reference count for cl_page_find, so that the page
+ * will be freed in cl_2queue_fini. */
+ cl_page_put(env, clp);
+ /*
+ * Set page clip to tell transfer formation engine that page
+ * has to be sent even if it is beyond KMS.
+ */
+ cl_page_clip(env, clp, 0, min(size, page_size));
+ size -= page_size;
+ file_offset += page_size;
+ }
+
+ if (rc == 0) {
+ rc = cl_io_submit_sync(env, io,
+ rw == READ ? CRT_READ : CRT_WRITE,
+ queue, CRP_NORMAL, 0);
+ if (rc == 0)
+ rc = pv->ldp_size;
+ }
+
+ cl_2queue_discard(env, io, queue);
+ cl_2queue_disown(env, io, queue);
+ cl_2queue_fini(env, queue);
+ RETURN(rc);
+}
+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);
+ long count = iov_length(iov, nr_segs);
+ long 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;
+ long 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=%lu (max %lu), "
+ "offset=%lld=%llx, pages %lu (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++) {
+ long 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;
+ long 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 %lu\n",
+ size);
+ continue;
+ }
+
+ GOTO(out, result);
+ }
+
+ tot_bytes += result;
+ file_offset += result;
+ iov_left -= result;
+ user_addr += result;