-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
+ * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2011, Whamcloud, Inc.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Implementation of cl_page for VVP layer.
*
* Author: Nikita Danilov <nikita.danilov@sun.com>
+ * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
*/
#define DEBUG_SUBSYSTEM S_LLITE
struct ccc_page *vpg = cl2ccc_page(slice);
cfs_page_t *vmpage = vpg->cpg_page;
- LASSERT(vmpage != NULL);
- if (nonblock) {
- if (TestSetPageLocked(vmpage))
- return -EAGAIN;
+ LASSERT(vmpage != NULL);
+ if (nonblock) {
+ if (!trylock_page(vmpage))
+ return -EAGAIN;
if (unlikely(PageWriteback(vmpage))) {
unlock_page(vmpage);
const struct cl_page_slice *slice,
struct cl_io *unused)
{
- cfs_page_t *vmpage = cl2vm_page(slice);
- int result;
+ cfs_page_t *vmpage = cl2vm_page(slice);
- if (clear_page_dirty_for_io(vmpage)) {
- set_page_writeback(vmpage);
- vvp_write_pending(cl2ccc(slice->cpl_obj), cl2ccc_page(slice));
- result = 0;
- } else
- result = -EALREADY;
- return result;
+ LASSERT(PageLocked(vmpage));
+ LASSERT(!PageDirty(vmpage));
+
+ set_page_writeback(vmpage);
+ vvp_write_pending(cl2ccc(slice->cpl_obj), cl2ccc_page(slice));
+
+ return 0;
}
/**
*/
static void vvp_vmpage_error(struct inode *inode, cfs_page_t *vmpage, int ioret)
{
- if (ioret == 0)
- ClearPageError(vmpage);
- else if (ioret != -EINTR) {
- SetPageError(vmpage);
- if (ioret == -ENOSPC)
- set_bit(AS_ENOSPC, &inode->i_mapping->flags);
- else
- set_bit(AS_EIO, &inode->i_mapping->flags);
- }
-}
-
-static void vvp_page_completion_common(const struct lu_env *env,
- struct ccc_page *cp, int ioret)
-{
- struct cl_page *clp = cp->cpg_cl.cpl_page;
- cfs_page_t *vmpage = cp->cpg_page;
- struct inode *inode = ccc_object_inode(clp->cp_obj);
-
- LINVRNT(cl_page_is_vmlocked(env, clp));
-
- if (!clp->cp_sync_io && clp->cp_type == CPT_CACHEABLE) {
- /*
- * Only mark the page error only when it's a cacheable page
- * and NOT a sync io.
- *
- * For sync IO and direct IO(CPT_TRANSIENT), the error is able
- * to be seen by application, so we don't need to mark a page
- * as error at all.
- */
- vvp_vmpage_error(inode, vmpage, ioret);
- unlock_page(vmpage);
- }
+ struct ccc_object *obj = cl_inode2ccc(inode);
+
+ if (ioret == 0) {
+ ClearPageError(vmpage);
+ obj->cob_discard_page_warned = 0;
+ } else {
+ SetPageError(vmpage);
+ if (ioret == -ENOSPC)
+ set_bit(AS_ENOSPC, &inode->i_mapping->flags);
+ else
+ set_bit(AS_EIO, &inode->i_mapping->flags);
+
+ if ((ioret == -ESHUTDOWN || ioret == -EINTR) &&
+ obj->cob_discard_page_warned == 0) {
+ obj->cob_discard_page_warned = 1;
+ ll_dirty_page_discard_warn(vmpage, ioret);
+ }
+ }
}
static void vvp_page_completion_read(const struct lu_env *env,
const struct cl_page_slice *slice,
int ioret)
{
- struct ccc_page *cp = cl2ccc_page(slice);
- struct cl_page *page = cl_page_top(slice->cpl_page);
- struct inode *inode = ccc_object_inode(page->cp_obj);
+ struct ccc_page *cp = cl2ccc_page(slice);
+ cfs_page_t *vmpage = cp->cpg_page;
+ struct cl_page *page = cl_page_top(slice->cpl_page);
+ struct inode *inode = ccc_object_inode(page->cp_obj);
ENTRY;
+ LASSERT(PageLocked(vmpage));
CL_PAGE_HEADER(D_PAGE, env, page, "completing READ with %d\n", ioret);
if (cp->cpg_defer_uptodate)
ll_ra_count_put(ll_i2sbi(inode), 1);
if (ioret == 0) {
- /* XXX: do we need this for transient pages? */
if (!cp->cpg_defer_uptodate)
cl_page_export(env, page, 1);
} else
cp->cpg_defer_uptodate = 0;
- vvp_page_completion_common(env, cp, ioret);
+
+ if (page->cp_sync_io == NULL)
+ unlock_page(vmpage);
EXIT;
}
-static void vvp_page_completion_write_common(const struct lu_env *env,
- const struct cl_page_slice *slice,
- int ioret)
+static void vvp_page_completion_write(const struct lu_env *env,
+ const struct cl_page_slice *slice,
+ int ioret)
{
- struct ccc_page *cp = cl2ccc_page(slice);
+ struct ccc_page *cp = cl2ccc_page(slice);
+ struct cl_page *pg = slice->cpl_page;
+ cfs_page_t *vmpage = cp->cpg_page;
+ ENTRY;
+
+ LASSERT(ergo(pg->cp_sync_io != NULL, PageLocked(vmpage)));
+ LASSERT(PageWriteback(vmpage));
+
+ CL_PAGE_HEADER(D_PAGE, env, pg, "completing WRITE with %d\n", ioret);
/*
* TODO: Actually it makes sense to add the page into oap pending
* ->cpo_completion method. The underlying transfer should be notified
* and then re-add the page into pending transfer queue. -jay
*/
+
cp->cpg_write_queued = 0;
vvp_write_complete(cl2ccc(slice->cpl_obj), cp);
- vvp_page_completion_common(env, cp, ioret);
-}
-
-static void vvp_page_completion_write(const struct lu_env *env,
- const struct cl_page_slice *slice,
- int ioret)
-{
- struct ccc_page *cp = cl2ccc_page(slice);
- struct cl_page *pg = slice->cpl_page;
- cfs_page_t *vmpage = cp->cpg_page;
-
- ENTRY;
-
- LINVRNT(cl_page_is_vmlocked(env, pg));
- LASSERT(PageWriteback(vmpage));
-
- CL_PAGE_HEADER(D_PAGE, env, pg, "completing WRITE with %d\n", ioret);
+ /*
+ * Only mark the page error only when it's an async write because
+ * applications won't wait for IO to finish.
+ */
+ if (pg->cp_sync_io == NULL)
+ vvp_vmpage_error(ccc_object_inode(pg->cp_obj), vmpage, ioret);
- vvp_page_completion_write_common(env, slice, ioret);
end_page_writeback(vmpage);
EXIT;
}
* truncated. Skip it.
*/
static int vvp_page_make_ready(const struct lu_env *env,
- const struct cl_page_slice *slice)
+ const struct cl_page_slice *slice)
{
- cfs_page_t *vmpage = cl2vm_page(slice);
- struct cl_page *pg = slice->cpl_page;
- int result;
-
- result = -EAGAIN;
- /* we're trying to write, but the page is locked.. come back later */
- if (!TestSetPageLocked(vmpage)) {
- if (pg->cp_state == CPS_CACHED) {
- /*
- * We can cancel IO if page wasn't dirty after all.
- */
- clear_page_dirty_for_io(vmpage);
- /*
- * This actually clears the dirty bit in the radix
- * tree.
- */
- set_page_writeback(vmpage);
- vvp_write_pending(cl2ccc(slice->cpl_obj),
- cl2ccc_page(slice));
- CL_PAGE_HEADER(D_PAGE, env, pg, "readied\n");
- result = 0;
- } else
- /*
- * Page was concurrently truncated.
- */
- LASSERT(pg->cp_state == CPS_FREEING);
- }
- RETURN(result);
+ cfs_page_t *vmpage = cl2vm_page(slice);
+ struct cl_page *pg = slice->cpl_page;
+ int result = 0;
+
+ lock_page(vmpage);
+ if (clear_page_dirty_for_io(vmpage)) {
+ LASSERT(pg->cp_state == CPS_CACHED);
+ /* This actually clears the dirty bit in the radix
+ * tree. */
+ set_page_writeback(vmpage);
+ vvp_write_pending(cl2ccc(slice->cpl_obj),
+ cl2ccc_page(slice));
+ CL_PAGE_HEADER(D_PAGE, env, pg, "readied\n");
+ } else if (pg->cp_state == CPS_PAGEOUT) {
+ /* is it possible for osc_flush_async_page() to already
+ * make it ready? */
+ result = -EALREADY;
+ } else {
+ CL_PAGE_DEBUG(D_ERROR, env, pg, "Unexpecting page state %d.\n",
+ pg->cp_state);
+ LBUG();
+ }
+ unlock_page(vmpage);
+ RETURN(result);
}
static int vvp_page_print(const struct lu_env *env,
static void vvp_transient_page_verify(const struct cl_page *page)
{
- struct inode *inode = ccc_object_inode(page->cp_obj);
+ struct inode *inode = ccc_object_inode(page->cp_obj);
- LASSERT(!TRYLOCK_INODE_MUTEX(inode));
- /* LASSERT_SEM_LOCKED(&inode->i_alloc_sem); */
+ LASSERT(!mutex_trylock(&inode->i_mutex));
}
static int vvp_transient_page_own(const struct lu_env *env,
}
static int vvp_transient_page_is_vmlocked(const struct lu_env *env,
- const struct cl_page_slice *slice)
+ const struct cl_page_slice *slice)
{
- struct inode *inode = ccc_object_inode(slice->cpl_obj);
- int locked;
+ struct inode *inode = ccc_object_inode(slice->cpl_obj);
+ int locked;
- locked = !TRYLOCK_INODE_MUTEX(inode);
- if (!locked)
- UNLOCK_INODE_MUTEX(inode);
- return locked ? -EBUSY : -ENODATA;
+ locked = !mutex_trylock(&inode->i_mutex);
+ if (!locked)
+ mutex_unlock(&inode->i_mutex);
+ return locked ? -EBUSY : -ENODATA;
}
static void
-vvp_transient_page_completion_write(const struct lu_env *env,
- const struct cl_page_slice *slice,
- int ioret)
+vvp_transient_page_completion(const struct lu_env *env,
+ const struct cl_page_slice *slice,
+ int ioret)
{
vvp_transient_page_verify(slice->cpl_page);
- vvp_page_completion_write_common(env, slice, ioret);
}
-
static void vvp_transient_page_fini(const struct lu_env *env,
- struct cl_page_slice *slice)
+ struct cl_page_slice *slice)
{
- struct ccc_page *cp = cl2ccc_page(slice);
- struct cl_page *clp = slice->cpl_page;
- struct ccc_object *clobj = cl2ccc(clp->cp_obj);
+ struct ccc_page *cp = cl2ccc_page(slice);
+ struct cl_page *clp = slice->cpl_page;
+ struct ccc_object *clobj = cl2ccc(clp->cp_obj);
- vvp_page_fini_common(cp);
- LASSERT(!TRYLOCK_INODE_MUTEX(clobj->cob_inode));
- clobj->cob_transient_pages--;
+ vvp_page_fini_common(cp);
+ LASSERT(!mutex_trylock(&clobj->cob_inode->i_mutex));
+ clobj->cob_transient_pages--;
}
static const struct cl_page_operations vvp_transient_page_ops = {
.io = {
[CRT_READ] = {
.cpo_prep = ccc_transient_page_prep,
- .cpo_completion = vvp_page_completion_read,
+ .cpo_completion = vvp_transient_page_completion,
},
[CRT_WRITE] = {
.cpo_prep = ccc_transient_page_prep,
- .cpo_completion = vvp_transient_page_completion_write,
+ .cpo_completion = vvp_transient_page_completion,
}
}
};
struct cl_page *vvp_page_init(const struct lu_env *env, struct cl_object *obj,
- struct cl_page *page, cfs_page_t *vmpage)
+ struct cl_page *page, cfs_page_t *vmpage)
{
- struct ccc_page *cpg;
- int result;
-
- CLOBINVRNT(env, obj, ccc_object_invariant(obj));
-
- OBD_SLAB_ALLOC_PTR_GFP(cpg, vvp_page_kmem, CFS_ALLOC_IO);
- if (cpg != NULL) {
- cpg->cpg_page = vmpage;
- page_cache_get(vmpage);
-
- CFS_INIT_LIST_HEAD(&cpg->cpg_pending_linkage);
- if (page->cp_type == CPT_CACHEABLE) {
- SetPagePrivate(vmpage);
- vmpage->private = (unsigned long)page;
- cl_page_slice_add(page, &cpg->cpg_cl, obj,
- &vvp_page_ops);
- } else {
- struct ccc_object *clobj = cl2ccc(obj);
-
- LASSERT(!TRYLOCK_INODE_MUTEX(clobj->cob_inode));
- cl_page_slice_add(page, &cpg->cpg_cl, obj,
- &vvp_transient_page_ops);
- clobj->cob_transient_pages++;
- }
- result = 0;
- } else
- result = -ENOMEM;
- return ERR_PTR(result);
+ struct ccc_page *cpg;
+ int result;
+
+ CLOBINVRNT(env, obj, ccc_object_invariant(obj));
+
+ OBD_SLAB_ALLOC_PTR_GFP(cpg, vvp_page_kmem, CFS_ALLOC_IO);
+ if (cpg != NULL) {
+ cpg->cpg_page = vmpage;
+ page_cache_get(vmpage);
+
+ CFS_INIT_LIST_HEAD(&cpg->cpg_pending_linkage);
+ if (page->cp_type == CPT_CACHEABLE) {
+ SetPagePrivate(vmpage);
+ vmpage->private = (unsigned long)page;
+ cl_page_slice_add(page, &cpg->cpg_cl, obj,
+ &vvp_page_ops);
+ } else {
+ struct ccc_object *clobj = cl2ccc(obj);
+
+ LASSERT(!mutex_trylock(&clobj->cob_inode->i_mutex));
+ cl_page_slice_add(page, &cpg->cpg_cl, obj,
+ &vvp_transient_page_ops);
+ clobj->cob_transient_pages++;
+ }
+ result = 0;
+ } else
+ result = -ENOMEM;
+ return ERR_PTR(result);
}