-static int vvp_page_sync_io(const struct lu_env *env, struct cl_io *io,
- struct cl_page *page, struct ccc_page *cp,
- enum cl_req_type crt)
-{
- struct cl_2queue *queue;
- int result;
-
- LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
-
- queue = &io->ci_queue;
- cl_2queue_init_page(queue, page);
-
- result = cl_io_submit_sync(env, io, crt, queue, 0);
- LASSERT(cl_page_is_owned(page, io));
-
- if (crt == CRT_READ)
- /*
- * in CRT_WRITE case page is left locked even in case of
- * error.
- */
- cl_page_list_disown(env, io, &queue->c2_qin);
- cl_2queue_fini(env, queue);
-
- return result;
-}
-
-/**
- * Prepare partially written-to page for a write.
- */
-static int vvp_io_prepare_partial(const struct lu_env *env, struct cl_io *io,
- struct cl_object *obj, struct cl_page *pg,
- struct ccc_page *cp,
- unsigned from, unsigned to)
-{
- struct cl_attr *attr = ccc_env_thread_attr(env);
- loff_t offset = cl_offset(obj, pg->cp_index);
- int result;
-
- cl_object_attr_lock(obj);
- result = cl_object_attr_get(env, obj, attr);
- cl_object_attr_unlock(obj);
- if (result == 0) {
- /*
- * If are writing to a new page, no need to read old data.
- * The extent locking will have updated the KMS, and for our
- * purposes here we can treat it like i_size.
- */
- if (attr->cat_kms <= offset) {
- char *kaddr = ll_kmap_atomic(cp->cpg_page, KM_USER0);
-
- memset(kaddr, 0, cl_page_size(obj));
- ll_kunmap_atomic(kaddr, KM_USER0);
- } else if (cp->cpg_defer_uptodate)
- cp->cpg_ra_used = 1;
- else
- result = vvp_page_sync_io(env, io, pg, cp, CRT_READ);
- /*
- * In older implementations, obdo_refresh_inode is called here
- * to update the inode because the write might modify the
- * object info at OST. However, this has been proven useless,
- * since LVB functions will be called when user space program
- * tries to retrieve inode attribute. Also, see bug 15909 for
- * details. -jay
- */
- if (result == 0)
- cl_page_export(env, pg, 1);
- }
- return result;
-}
-
-static int vvp_io_prepare_write(const struct lu_env *env,
- const struct cl_io_slice *ios,
- const struct cl_page_slice *slice,
- unsigned from, unsigned to)
-{
- struct cl_object *obj = slice->cpl_obj;
- struct ccc_page *cp = cl2ccc_page(slice);
- struct cl_page *pg = slice->cpl_page;
- struct page *vmpage = cp->cpg_page;
-
- int result;
-
- ENTRY;
-
- LINVRNT(cl_page_is_vmlocked(env, pg));
- LASSERT(vmpage->mapping->host == ccc_object_inode(obj));
-
- result = 0;
-
- CL_PAGE_HEADER(D_PAGE, env, pg, "preparing: [%d, %d]\n", from, to);
- if (!PageUptodate(vmpage)) {
- /*
- * We're completely overwriting an existing page, so _don't_
- * set it up to date until commit_write
- */
- if (from == 0 && to == PAGE_CACHE_SIZE) {
- CL_PAGE_HEADER(D_PAGE, env, pg, "full page write\n");
- POISON_PAGE(page, 0x11);
- } else
- result = vvp_io_prepare_partial(env, ios->cis_io, obj,
- pg, cp, from, to);
- } else
- CL_PAGE_HEADER(D_PAGE, env, pg, "uptodate\n");
- RETURN(result);
-}
-
-static int vvp_io_commit_write(const struct lu_env *env,
- const struct cl_io_slice *ios,
- const struct cl_page_slice *slice,
- unsigned from, unsigned to)
-{
- struct cl_object *obj = slice->cpl_obj;
- struct cl_io *io = ios->cis_io;
- struct ccc_page *cp = cl2ccc_page(slice);
- struct cl_page *pg = slice->cpl_page;
- struct inode *inode = ccc_object_inode(obj);
- struct ll_sb_info *sbi = ll_i2sbi(inode);
- struct ll_inode_info *lli = ll_i2info(inode);
- struct page *vmpage = cp->cpg_page;
-
- int result;
- int tallyop;
- loff_t size;
-
- ENTRY;
-
- LINVRNT(cl_page_is_vmlocked(env, pg));
- LASSERT(vmpage->mapping->host == inode);
-
- LU_OBJECT_HEADER(D_INODE, env, &obj->co_lu, "commiting page write\n");
- CL_PAGE_HEADER(D_PAGE, env, pg, "committing: [%d, %d]\n", from, to);
-
- /*
- * queue a write for some time in the future the first time we
- * dirty the page.
- *
- * This is different from what other file systems do: they usually
- * just mark page (and some of its buffers) dirty and rely on
- * balance_dirty_pages() to start a write-back. Lustre wants write-back
- * to be started earlier for the following reasons:
- *
- * (1) with a large number of clients we need to limit the amount
- * of cached data on the clients a lot;
- *
- * (2) large compute jobs generally want compute-only then io-only
- * and the IO should complete as quickly as possible;
- *
- * (3) IO is batched up to the RPC size and is async until the
- * client max cache is hit
- * (/proc/fs/lustre/osc/OSC.../max_dirty_mb)
- *
- */
- if (!PageDirty(vmpage)) {
- tallyop = LPROC_LL_DIRTY_MISSES;
- result = cl_page_cache_add(env, io, pg, CRT_WRITE);
- if (result == 0) {
- /* page was added into cache successfully. */
- set_page_dirty(vmpage);
- vvp_write_pending(cl2ccc(obj), cp);
- } else if (result == -EDQUOT) {
- pgoff_t last_index = i_size_read(inode) >> PAGE_CACHE_SHIFT;
- bool need_clip = true;
-
- /*
- * Client ran out of disk space grant. Possible
- * strategies are:
- *
- * (a) do a sync write, renewing grant;
- *
- * (b) stop writing on this stripe, switch to the
- * next one.
- *
- * (b) is a part of "parallel io" design that is the
- * ultimate goal. (a) is what "old" client did, and
- * what the new code continues to do for the time
- * being.
- */
- if (last_index > pg->cp_index) {
- to = PAGE_CACHE_SIZE;
- need_clip = false;
- } else if (last_index == pg->cp_index) {
- int size_to = i_size_read(inode) & ~CFS_PAGE_MASK;
- if (to < size_to)
- to = size_to;
- }
- if (need_clip)
- cl_page_clip(env, pg, 0, to);
- result = vvp_page_sync_io(env, io, pg, cp, CRT_WRITE);
- if (result)
- CERROR("Write page %lu of inode %p failed %d\n",
- pg->cp_index, inode, result);
- }
- } else {
- tallyop = LPROC_LL_DIRTY_HITS;
- result = 0;
- }
- ll_stats_ops_tally(sbi, tallyop, 1);
-
- /* Inode should be marked DIRTY even if no new page was marked DIRTY
- * because page could have been not flushed between 2 modifications.
- * It is important the file is marked DIRTY as soon as the I/O is done
- * Indeed, when cache is flushed, file could be already closed and it
- * is too late to warn the MDT.
- * It is acceptable that file is marked DIRTY even if I/O is dropped
- * for some reasons before being flushed to OST.
- */
- if (result == 0) {
- spin_lock(&lli->lli_lock);
- lli->lli_flags |= LLIF_DATA_MODIFIED;
- spin_unlock(&lli->lli_lock);
- }
-
- size = cl_offset(obj, pg->cp_index) + to;
-
- ll_inode_size_lock(inode);
- if (result == 0) {
- if (size > i_size_read(inode)) {
- cl_isize_write_nolock(inode, size);
- CDEBUG(D_VFSTRACE, DFID" updating i_size %lu\n",
- PFID(lu_object_fid(&obj->co_lu)),
- (unsigned long)size);
- }
- cl_page_export(env, pg, 1);
- } else {
- if (size > i_size_read(inode))
- cl_page_discard(env, io, pg);
- }
- ll_inode_size_unlock(inode);
- RETURN(result);
-}
-