-/* -*- 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) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2012, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Lustre Lite I/O page cache routines for the 2.5/2.6 kernel version
*/
-#ifndef AUTOCONF_INCLUDED
-#include <linux/config.h>
-#endif
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/string.h>
#include <linux/stat.h>
#include <linux/errno.h>
-#include <linux/smp_lock.h>
#include <linux/unistd.h>
-#include <linux/version.h>
-#include <asm/system.h>
#include <asm/uaccess.h>
+#ifdef HAVE_MIGRATE_H
+#include <linux/migrate.h>
+#elif defined(HAVE_MIGRATE_MODE_H)
+#include <linux/migrate_mode.h>
+#endif
#include <linux/fs.h>
#include <linux/buffer_head.h>
+#include <linux/mpage.h>
#include <linux/writeback.h>
#include <linux/stat.h>
#include <asm/uaccess.h>
#include <linux/mm.h>
#include <linux/pagemap.h>
-#include <linux/smp_lock.h>
#define DEBUG_SUBSYSTEM S_LLITE
-//#include <lustre_mdc.h>
#include <lustre_lite.h>
#include "llite_internal.h"
#include <linux/lustre_compat25.h>
* aligned truncate). Lustre leaves partially truncated page in the cache,
* relying on struct inode::i_size to limit further accesses.
*/
-static int cl_invalidatepage(struct page *vmpage, unsigned long offset)
+static void ll_invalidatepage(struct page *vmpage, unsigned long offset)
{
struct inode *inode;
struct lu_env *env;
struct cl_page *page;
struct cl_object *obj;
- int result;
int refcheck;
LASSERT(PageLocked(vmpage));
* below because they are run with page locked and all our io is
* happening with locked page too
*/
- result = 0;
if (offset == 0) {
env = cl_env_get(&refcheck);
if (!IS_ERR(env)) {
lu_ref_add(&page->cp_reference,
"delete", vmpage);
cl_page_delete(env, page);
- result = 1;
lu_ref_del(&page->cp_reference,
"delete", vmpage);
cl_page_put(env, page);
cl_env_put(env, &refcheck);
}
}
- return result;
}
-#ifdef HAVE_INVALIDATEPAGE_RETURN_INT
-static int ll_invalidatepage(struct page *page, unsigned long offset)
-{
- return cl_invalidatepage(page, offset);
-}
-#else /* !HAVE_INVALIDATEPAGE_RETURN_INT */
-static void ll_invalidatepage(struct page *page, unsigned long offset)
-{
- cl_invalidatepage(page, offset);
-}
-#endif
-
#ifdef HAVE_RELEASEPAGE_WITH_INT
#define RELEASEPAGE_ARG_TYPE int
#else
#define RELEASEPAGE_ARG_TYPE gfp_t
#endif
-static int ll_releasepage(struct page *page, RELEASEPAGE_ARG_TYPE gfp_mask)
+static int ll_releasepage(struct page *vmpage, RELEASEPAGE_ARG_TYPE gfp_mask)
{
- void *cookie;
+ struct cl_env_nest nest;
+ struct lu_env *env;
+ struct cl_object *obj;
+ struct cl_page *page;
+ struct address_space *mapping;
+ int result;
- cookie = cl_env_reenter();
- ll_invalidatepage(page, 0);
- cl_env_reexit(cookie);
- return 1;
+ LASSERT(PageLocked(vmpage));
+ if (PageWriteback(vmpage) || PageDirty(vmpage))
+ return 0;
+
+ mapping = vmpage->mapping;
+ if (mapping == NULL)
+ return 1;
+
+ obj = ll_i2info(mapping->host)->lli_clob;
+ if (obj == NULL)
+ return 1;
+
+ /* 1 for page allocator, 1 for cl_page and 1 for page cache */
+ if (page_count(vmpage) > 3)
+ return 0;
+
+ /* TODO: determine what gfp should be used by @gfp_mask. */
+ env = cl_env_nested_get(&nest);
+ if (IS_ERR(env))
+ /* If we can't allocate an env we won't call cl_page_put()
+ * later on which further means it's impossible to drop
+ * page refcount by cl_page, so ask kernel to not free
+ * this page. */
+ return 0;
+
+ page = cl_vmpage_page(vmpage, obj);
+ result = page == NULL;
+ if (page != NULL) {
+ if (!cl_page_in_use(page)) {
+ result = 1;
+ cl_page_delete(env, page);
+ }
+ cl_page_put(env, page);
+ }
+ cl_env_nested_put(&nest, env);
+ return result;
}
static int ll_set_page_dirty(struct page *vmpage)
*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));
+ OBD_ALLOC_LARGE(*pages, *max_pages * sizeof(**pages));
if (*pages) {
down_read(¤t->mm->mmap_sem);
result = get_user_pages(current, current->mm, user_addr,
NULL);
up_read(¤t->mm->mmap_sem);
if (unlikely(result <= 0))
- OBD_FREE(*pages, *max_pages * sizeof(**pages));
+ OBD_FREE_LARGE(*pages, *max_pages * sizeof(**pages));
}
return result;
page_cache_release(pages[i]);
}
- OBD_FREE(pages, npages * sizeof(*pages));
+ OBD_FREE_LARGE(pages, npages * sizeof(*pages));
}
ssize_t ll_direct_rw_pages(const struct lu_env *env, struct cl_io *io,
struct ll_dio_pages *pv)
{
struct cl_page *clp;
- struct ccc_page *clup;
struct cl_2queue *queue;
struct cl_object *obj = io->ci_obj;
- struct cl_sync_io *anchor = &ccc_env_info(env)->cti_sync_io;
int i;
ssize_t rc = 0;
loff_t file_offset = pv->ldp_start_offset;
- size_t size = pv->ldp_size;
+ long size = pv->ldp_size;
int page_count = pv->ldp_nr;
struct page **pages = pv->ldp_pages;
- size_t page_size = cl_page_size(obj);
+ long page_size = cl_page_size(obj);
+ bool do_io;
+ int io_pages = 0;
ENTRY;
- cl_sync_io_init(anchor, page_count);
-
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);
break;
}
+ rc = cl_page_own(env, io, clp);
+ if (rc) {
+ LASSERT(clp->cp_state == CPS_FREEING);
+ cl_page_put(env, clp);
+ break;
+ }
+
+ do_io = true;
+
/* 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;
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, (const void *)src, min(page_size, size));
- kunmap_atomic(dst, KM_USER1);
- kunmap_atomic(src, KM_USER0);
+ src = ll_kmap_atomic(src_page, KM_USER0);
+ dst = ll_kmap_atomic(dst_page, KM_USER1);
+ memcpy(dst, src, min(page_size, size));
+ ll_kunmap_atomic(dst, KM_USER1);
+ ll_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 (rw == READ) {
+ /* do not issue the page for read, since it
+ * may reread a ra page which has NOT uptodate
+ * bit set. */
+ cl_page_disown(env, io, clp);
+ do_io = false;
+ }
+ }
+
+ if (likely(do_io)) {
+ cl_2queue_add(queue, clp);
+
/*
- * 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.
+ * 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));
- rc = cl_page_own(env, io, clp);
- if (rc) {
- LASSERT(clp->cp_state == CPS_FREEING);
- cl_page_put(env, clp);
- break;
+ ++io_pages;
}
- clup = cl2ccc_page(cl_page_at(clp, &vvp_device_type));
- clup->cpg_sync_io = anchor;
- cl_2queue_add(queue, clp);
-
- /* drop the reference count for cl_page_find, so that the page
- * will be freed in cl_2queue_fini. */
+ /* drop the reference count for cl_page_find */
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_rw(env, io, rw == READ ? CRT_READ : CRT_WRITE,
- queue, CRP_NORMAL);
- if (rc == 0) {
- /*
- * If some pages weren't sent for any reason (e.g.,
- * direct-io read found up-to-date pages in the
- * cache), count them as completed to avoid infinite
- * wait.
- */
- cl_page_list_for_each(clp, &queue->c2_qin)
- cl_sync_io_note(anchor, +1);
- /* wait for the IO to be finished. */
- rc = cl_sync_io_wait(env, io, &queue->c2_qout,
- anchor) ?: pv->ldp_size;
- }
+ if (rc == 0 && io_pages) {
+ rc = cl_io_submit_sync(env, io,
+ rw == READ ? CRT_READ : CRT_WRITE,
+ queue, 0);
}
+ if (rc == 0)
+ rc = pv->ldp_size;
cl_2queue_discard(env, io, queue);
cl_2queue_disown(env, io, queue);
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
+#ifdef KMALLOC_MAX_SIZE
+#define MAX_MALLOC KMALLOC_MAX_SIZE
+#else
+#define MAX_MALLOC (128 * 1024)
+#endif
+
+/* This is the maximum size of a single O_DIRECT request, based on the
* 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))
+ * then truncate this to be a full-sized RPC. For 4kB PAGE_SIZE this is
+ * up to 22MB for 128kB kmalloc and up to 682MB for 4MB kmalloc. */
+#define MAX_DIO_SIZE ((MAX_MALLOC / sizeof(struct brw_page) * CFS_PAGE_SIZE) & \
+ ~(DT_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 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;
+ 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;
- size_t size = MAX_DIO_SIZE;
+ long size = MAX_DIO_SIZE;
int refcheck;
ENTRY;
- if (!lli->lli_smd || !lli->lli_smd->lsm_object_id)
+ if (!lli->lli_has_smd)
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",
+ 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);
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);
+ /* 0. Need locking between buffered and direct access. and race with
+ * size changing by concurrent truncates and writes.
+ * 1. Need inode mutex to operate transient pages.
+ */
+ if (rw == READ)
+ mutex_lock(&inode->i_mutex);
LASSERT(obj->cob_transient_pages == 0);
for (seg = 0; seg < nr_segs; seg++) {
- size_t iov_left = iov[seg].iov_len;
+ 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)
+ if (file_offset >= i_size_read(inode))
break;
- if (file_offset + iov_left > inode->i_size)
- iov_left = inode->i_size - file_offset;
+ if (file_offset + iov_left > i_size_read(inode))
+ iov_left = i_size_read(inode) - file_offset;
}
while (iov_left > 0) {
struct page **pages;
int page_count, max_pages = 0;
- size_t bytes;
+ long bytes;
- bytes = min(size,iov_left);
- page_count = ll_get_user_pages(rw, user_addr,
- 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);
+ bytes, file_offset,
+ pages, page_count);
ll_free_user_pages(pages, max_pages, rw==READ);
} else if (page_count == 0) {
GOTO(out, result = -EFAULT);
size = ((((size / 2) - 1) |
~CFS_PAGE_MASK) + 1) &
CFS_PAGE_MASK;
- CDEBUG(D_VFSTRACE, "DIO size now %u\n",
- (int)size);
+ CDEBUG(D_VFSTRACE,"DIO size now %lu\n",
+ size);
continue;
}
}
}
out:
- LASSERT(obj->cob_transient_pages == 0);
- if (rw == READ)
- UNLOCK_INODE_MUTEX(inode);
+ LASSERT(obj->cob_transient_pages == 0);
+ if (rw == READ)
+ mutex_unlock(&inode->i_mutex);
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);
- }
+ struct lov_stripe_md *lsm;
+
+ lsm = ccc_inode_lsm_get(inode);
+ LASSERT(lsm != NULL);
+ lov_stripe_lock(lsm);
+ obd_adjust_kms(ll_i2dtexp(inode), lsm, file_offset, 0);
+ lov_stripe_unlock(lsm);
+ ccc_inode_lsm_put(inode, lsm);
+ }
+ }
+
+ cl_env_put(env, &refcheck);
+ RETURN(tot_bytes ? : result);
+}
+
+#if defined(HAVE_KERNEL_WRITE_BEGIN_END) || defined(MS_HAS_NEW_AOPS)
+static int ll_write_begin(struct file *file, struct address_space *mapping,
+ loff_t pos, unsigned len, unsigned flags,
+ struct page **pagep, void **fsdata)
+{
+ pgoff_t index = pos >> PAGE_CACHE_SHIFT;
+ struct page *page;
+ int rc;
+ unsigned from = pos & (PAGE_CACHE_SIZE - 1);
+ ENTRY;
+
+ page = grab_cache_page_write_begin(mapping, index, flags);
+ if (!page)
+ RETURN(-ENOMEM);
+
+ *pagep = page;
+
+ rc = ll_prepare_write(file, page, from, from + len);
+ if (rc) {
+ unlock_page(page);
+ page_cache_release(page);
}
+ RETURN(rc);
+}
+
+static int ll_write_end(struct file *file, struct address_space *mapping,
+ loff_t pos, unsigned len, unsigned copied,
+ struct page *page, void *fsdata)
+{
+ unsigned from = pos & (PAGE_CACHE_SIZE - 1);
+ int rc;
+
+ rc = ll_commit_write(file, page, from, from + copied);
+ unlock_page(page);
+ page_cache_release(page);
- cl_env_put(env, &refcheck);
- RETURN(tot_bytes ? : result);
+ return rc ?: copied;
+}
+#endif
+
+#ifdef CONFIG_MIGRATION
+int ll_migratepage(struct address_space *mapping,
+ struct page *newpage, struct page *page
+#ifdef HAVE_MIGRATEPAGE_4ARGS
+ , enum migrate_mode mode
+#endif
+ )
+{
+ /* Always fail page migration until we have a proper implementation */
+ return -EIO;
}
+#endif
+#ifndef MS_HAS_NEW_AOPS
struct address_space_operations ll_aops = {
.readpage = ll_readpage,
// .readpages = ll_readpages,
.direct_IO = ll_direct_IO_26,
.writepage = ll_writepage,
- .writepages = generic_writepages,
+ .writepages = ll_writepages,
.set_page_dirty = ll_set_page_dirty,
- .sync_page = NULL,
+#ifdef HAVE_KERNEL_WRITE_BEGIN_END
+ .write_begin = ll_write_begin,
+ .write_end = ll_write_end,
+#else
.prepare_write = ll_prepare_write,
.commit_write = ll_commit_write,
+#endif
.invalidatepage = ll_invalidatepage,
.releasepage = (void *)ll_releasepage,
+#ifdef CONFIG_MIGRATION
+ .migratepage = ll_migratepage,
+#endif
.bmap = NULL
};
+#else
+struct address_space_operations_ext ll_aops = {
+ .orig_aops.readpage = ll_readpage,
+// .orig_aops.readpages = ll_readpages,
+ .orig_aops.direct_IO = ll_direct_IO_26,
+ .orig_aops.writepage = ll_writepage,
+ .orig_aops.writepages = ll_writepages,
+ .orig_aops.set_page_dirty = ll_set_page_dirty,
+ .orig_aops.prepare_write = ll_prepare_write,
+ .orig_aops.commit_write = ll_commit_write,
+ .orig_aops.invalidatepage = ll_invalidatepage,
+ .orig_aops.releasepage = ll_releasepage,
+#ifdef CONFIG_MIGRATION
+ .orig_aops.migratepage = ll_migratepage,
+#endif
+ .orig_aops.bmap = NULL,
+ .write_begin = ll_write_begin,
+ .write_end = ll_write_end
+};
+#endif
git://git.whamcloud.com / fs / lustre-release.git / blobdiff