-/* -*- 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.
*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* 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, 2017, 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/buffer_head.h>
+#include <linux/errno.h>
+#include <linux/fs.h>
#include <linux/kernel.h>
#include <linux/mm.h>
+#include <linux/mpage.h>
+#include <linux/pagemap.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>
-
-#include <linux/fs.h>
-#include <linux/buffer_head.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>
+
+#ifdef HAVE_MIGRATE_H
+#include <linux/migrate.h>
+#elif defined(HAVE_MIGRATE_MODE_H)
+#include <linux/migrate_mode.h>
+#endif
#define DEBUG_SUBSYSTEM S_LLITE
-//#include <lustre_mdc.h>
-#include <lustre_lite.h>
#include "llite_internal.h"
-#include <linux/lustre_compat25.h>
+#include <lustre_compat.h>
/**
* Implements Linux VM address_space::invalidatepage() method. This method is
* 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,
+#ifdef HAVE_INVALIDATE_RANGE
+ unsigned int offset, unsigned int length
+#else
+ unsigned long offset
+#endif
+ )
{
struct inode *inode;
struct lu_env *env;
struct cl_page *page;
struct cl_object *obj;
- int result;
- int refcheck;
-
LASSERT(PageLocked(vmpage));
LASSERT(!PageWriteback(vmpage));
- /*
- * It is safe to not check anything in invalidatepage/releasepage
- * 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)) {
- inode = vmpage->mapping->host;
- obj = ll_i2info(inode)->lli_clob;
- if (obj != NULL) {
- page = cl_vmpage_page(vmpage, obj);
- if (page != NULL) {
- 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);
- }
- } else
- LASSERT(vmpage->private == 0);
- cl_env_put(env, &refcheck);
- }
+ /*
+ * It is safe to not check anything in invalidatepage/releasepage
+ * below because they are run with page locked and all our io is
+ * happening with locked page too
+ */
+#ifdef HAVE_INVALIDATE_RANGE
+ if (offset == 0 && length == PAGE_SIZE) {
+#else
+ if (offset == 0) {
+#endif
+ /* See the comment in ll_releasepage() */
+ env = cl_env_percpu_get();
+ LASSERT(!IS_ERR(env));
+
+ inode = vmpage->mapping->host;
+ obj = ll_i2info(inode)->lli_clob;
+ if (obj != NULL) {
+ page = cl_vmpage_page(vmpage, obj);
+ if (page != NULL) {
+ cl_page_delete(env, page);
+ cl_page_put(env, page);
+ }
+ } else
+ LASSERT(vmpage->private == 0);
+
+ cl_env_percpu_put(env);
}
- 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;
-
- cookie = cl_env_reenter();
- ll_invalidatepage(page, 0);
- cl_env_reexit(cookie);
- return 1;
-}
-
-static int ll_set_page_dirty(struct page *vmpage)
-{
-#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));
+ struct lu_env *env;
+ struct cl_object *obj;
+ struct cl_page *page;
+ struct address_space *mapping;
+ int result = 0;
+
+ 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;
+
+ page = cl_vmpage_page(vmpage, obj);
+ if (page == NULL)
+ return 1;
+
+ env = cl_env_percpu_get();
+ LASSERT(!IS_ERR(env));
+
+ if (!cl_page_in_use(page)) {
+ result = 1;
+ cl_page_delete(env, page);
+ }
+
+ /* To use percpu env array, the call path can not be rescheduled;
+ * otherwise percpu array will be messed if ll_releaspage() called
+ * again on the same CPU.
+ *
+ * If this page holds the last refc of cl_object, the following
+ * call path may cause reschedule:
+ * cl_page_put -> cl_page_free -> cl_object_put ->
+ * lu_object_put -> lu_object_free -> lov_delete_raid0.
+ *
+ * However, the kernel can't get rid of this inode until all pages have
+ * been cleaned up. Now that we hold page lock here, it's pretty safe
+ * that we won't get into object delete path.
+ */
+ LASSERT(cl_object_refc(obj) > 1);
+ cl_page_put(env, page);
+
+ cl_env_percpu_put(env);
+ return result;
}
#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)
+static ssize_t
+ll_direct_IO_seg(const struct lu_env *env, struct cl_io *io, int rw,
+ struct inode *inode, size_t size, loff_t file_offset,
+ struct page **pages, int page_count)
{
- 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;
+ struct cl_page *clp;
+ struct cl_2queue *queue;
+ struct cl_object *obj = io->ci_obj;
+ int i;
+ ssize_t rc = 0;
+ size_t page_size = cl_page_size(obj);
+ size_t orig_size = size;
+ bool do_io;
+ int io_pages = 0;
+
+ ENTRY;
+ queue = &io->ci_queue;
+ cl_2queue_init(queue);
+ for (i = 0; i < page_count; i++) {
+ LASSERT(!(file_offset & (page_size - 1)));
+ clp = cl_page_find(env, obj, cl_index(obj, file_offset),
+ pages[i], CPT_TRANSIENT);
+ if (IS_ERR(clp)) {
+ rc = PTR_ERR(clp);
+ 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
+ */
+ if (clp->cp_type == CPT_CACHEABLE) {
+ struct page *vmpage = cl_page_vmpage(clp);
+ struct page *src_page;
+ struct page *dst_page;
+ void *src;
+ void *dst;
+
+ src_page = (rw == WRITE) ? pages[i] : vmpage;
+ dst_page = (rw == WRITE) ? vmpage : pages[i];
+
+ 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);
+
+ /*
+ * 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));
+
+ ++io_pages;
+ }
+
+ /* drop the reference count for cl_page_find */
+ cl_page_put(env, clp);
+ size -= page_size;
+ file_offset += page_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 = orig_size;
+
+ cl_2queue_discard(env, io, queue);
+ cl_2queue_disown(env, io, queue);
+ cl_2queue_fini(env, queue);
+ RETURN(rc);
}
/* 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));
+ int i;
+
+ for (i = 0; i < npages; i++) {
+ if (pages[i] == NULL)
+ break;
+ if (do_dirty)
+ set_page_dirty_lock(pages[i]);
+ put_page(pages[i]);
+ }
+
+#if defined(HAVE_DIRECTIO_ITER) || defined(HAVE_IOV_ITER_RW)
+ kvfree(pages);
+#else
+ OBD_FREE_LARGE(pages, npages * sizeof(*pages));
+#endif
}
-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;
- size_t size = pv->ldp_size;
- int page_count = pv->ldp_nr;
- struct page **pages = pv->ldp_pages;
- size_t 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;
- }
+#ifdef KMALLOC_MAX_SIZE
+#define MAX_MALLOC KMALLOC_MAX_SIZE
+#else
+#define MAX_MALLOC (128 * 1024)
+#endif
- /* 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, (const void *)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.
- */
- }
+/* 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. 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) * PAGE_SIZE) & \
+ ~(DT_MAX_BRW_SIZE - 1))
- rc = cl_page_own(env, io, clp);
- if (rc) {
- LASSERT(clp->cp_state == CPS_FREEING);
- cl_page_put(env, clp);
- break;
- }
+#ifndef HAVE_IOV_ITER_RW
+# define iov_iter_rw(iter) rw
+#endif
- 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 defined(HAVE_DIRECTIO_ITER) || defined(HAVE_IOV_ITER_RW)
+static ssize_t
+ll_direct_IO(
+# ifndef HAVE_IOV_ITER_RW
+ int rw,
+# endif
+ struct kiocb *iocb, struct iov_iter *iter
+# ifndef HAVE_DIRECTIO_2ARGS
+ , loff_t file_offset
+# endif
+ )
+{
+#ifdef HAVE_DIRECTIO_2ARGS
+ loff_t file_offset = iocb->ki_pos;
+#endif
+ struct ll_cl_context *lcc;
+ const struct lu_env *env;
+ struct cl_io *io;
+ struct file *file = iocb->ki_filp;
+ struct inode *inode = file->f_mapping->host;
+ ssize_t count = iov_iter_count(iter);
+ ssize_t tot_bytes = 0, result = 0;
+ size_t size = MAX_DIO_SIZE;
+
+ /* Check EOF by ourselves */
+ if (iov_iter_rw(iter) == READ && file_offset >= i_size_read(inode))
+ return 0;
+ /* FIXME: io smaller than PAGE_SIZE is broken on ia64 ??? */
+ if ((file_offset & ~PAGE_MASK) || (count & ~PAGE_MASK))
+ return -EINVAL;
+
+ CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), size=%zd (max %lu), "
+ "offset=%lld=%llx, pages %zd (max %lu)\n",
+ PFID(ll_inode2fid(inode)), inode, count, MAX_DIO_SIZE,
+ file_offset, file_offset, count >> PAGE_SHIFT,
+ MAX_DIO_SIZE >> PAGE_SHIFT);
+
+ /* Check that all user buffers are aligned as well */
+ if (iov_iter_alignment(iter) & ~PAGE_MASK)
+ return -EINVAL;
+
+ lcc = ll_cl_find(file);
+ if (lcc == NULL)
+ RETURN(-EIO);
+
+ env = lcc->lcc_env;
+ LASSERT(!IS_ERR(env));
+ io = lcc->lcc_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 mutex to operate transient pages.
+ */
+ if (iov_iter_rw(iter) == READ)
+ inode_lock(inode);
+
+ while (iov_iter_count(iter)) {
+ struct page **pages;
+ size_t offs;
+
+ count = min_t(size_t, iov_iter_count(iter), size);
+ if (iov_iter_rw(iter) == READ) {
+ if (file_offset >= i_size_read(inode))
+ break;
+
+ if (file_offset + count > i_size_read(inode))
+ count = i_size_read(inode) - file_offset;
+ }
+
+ result = iov_iter_get_pages_alloc(iter, &pages, count, &offs);
+ if (likely(result > 0)) {
+ int n = DIV_ROUND_UP(result + offs, PAGE_SIZE);
+
+ result = ll_direct_IO_seg(env, io, iov_iter_rw(iter),
+ inode, result, file_offset,
+ pages, n);
+ ll_free_user_pages(pages, n,
+ iov_iter_rw(iter) == READ);
+
+ }
+ 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 > (PAGE_SIZE / sizeof(*pages)) *
+ PAGE_SIZE) {
+ size = ((((size / 2) - 1) |
+ ~PAGE_MASK) + 1) & PAGE_MASK;
+ CDEBUG(D_VFSTRACE, "DIO size now %zu\n",
+ size);
+ continue;
+ }
+
+ GOTO(out, result);
+ }
+
+ iov_iter_advance(iter, result);
+ tot_bytes += result;
+ file_offset += result;
+ }
+out:
+ if (iov_iter_rw(iter) == READ)
+ inode_unlock(inode);
- 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;
- }
+ if (tot_bytes > 0) {
+ struct vvp_io *vio = vvp_env_io(env);
+
+ /* no commit async for direct IO */
+ vio->u.write.vui_written += tot_bytes;
+ }
- cl_2queue_discard(env, io, queue);
- cl_2queue_disown(env, io, queue);
- cl_2queue_fini(env, queue);
- RETURN(rc);
+ return tot_bytes ? : result;
}
-EXPORT_SYMBOL(ll_direct_rw_pages);
+#else /* !HAVE_DIRECTIO_ITER && !HAVE_IOV_ITER_RW */
-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)
+static inline int ll_get_user_pages(int rw, unsigned long user_addr,
+ size_t size, struct page ***pages,
+ int *max_pages)
{
- 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);
+ 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 + PAGE_SIZE - 1) >>
+ PAGE_SHIFT;
+ *max_pages -= user_addr >> PAGE_SHIFT;
+
+ 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,
+ *max_pages, (rw == READ), 0, *pages,
+ NULL);
+ up_read(¤t->mm->mmap_sem);
+ if (unlikely(result <= 0))
+ OBD_FREE_LARGE(*pages, *max_pages * sizeof(**pages));
+ }
+
+ return result;
}
-/* 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)
+static ssize_t
+ll_direct_IO(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);
+ struct ll_cl_context *lcc;
+ const struct lu_env *env;
+ struct cl_io *io;
+ struct file *file = iocb->ki_filp;
+ struct inode *inode = file->f_mapping->host;
+ ssize_t count = iov_length(iov, nr_segs);
+ ssize_t tot_bytes = 0, result = 0;
+ unsigned long seg = 0;
+ size_t size = MAX_DIO_SIZE;
+ ENTRY;
/* FIXME: io smaller than PAGE_SIZE is broken on ia64 ??? */
- if ((file_offset & ~CFS_PAGE_MASK) || (count & ~CFS_PAGE_MASK))
+ if ((file_offset & ~PAGE_MASK) || (count & ~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);
+ CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), size=%zd (max %lu), "
+ "offset=%lld=%llx, pages %zd (max %lu)\n",
+ PFID(ll_inode2fid(inode)), inode, count, MAX_DIO_SIZE,
+ file_offset, file_offset, count >> PAGE_SHIFT,
+ MAX_DIO_SIZE >> 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))
+ if (((unsigned long)iov[seg].iov_base & ~PAGE_MASK) ||
+ (iov[seg].iov_len & ~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);
+ lcc = ll_cl_find(file);
+ if (lcc == NULL)
+ RETURN(-EIO);
- /* 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);
+ env = lcc->lcc_env;
+ LASSERT(!IS_ERR(env));
+ io = lcc->lcc_io;
+ LASSERT(io != NULL);
- LASSERT(obj->cob_transient_pages == 0);
for (seg = 0; seg < nr_segs; seg++) {
- size_t iov_left = iov[seg].iov_len;
+ 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)
+ 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;
+ size_t 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 = page_count << PAGE_SHIFT;
+ result = ll_direct_IO_seg(env, io, rw, inode,
+ bytes, file_offset,
+ pages, page_count);
ll_free_user_pages(pages, max_pages, rw==READ);
} else if (page_count == 0) {
GOTO(out, result = -EFAULT);
* 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 > (PAGE_SIZE / sizeof(*pages)) *
+ PAGE_SIZE) {
size = ((((size / 2) - 1) |
- ~CFS_PAGE_MASK) + 1) &
- CFS_PAGE_MASK;
- CDEBUG(D_VFSTRACE, "DIO size now %u\n",
- (int)size);
+ ~PAGE_MASK) + 1) &
+ PAGE_MASK;
+ CDEBUG(D_VFSTRACE, "DIO size now %zu\n",
+ size);
continue;
}
}
}
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);
- }
- }
+ struct vvp_io *vio = vvp_env_io(env);
+
+ /* no commit async for direct IO */
+ vio->u.write.vui_written += tot_bytes;
+ }
- cl_env_put(env, &refcheck);
- RETURN(tot_bytes ? : result);
+ RETURN(tot_bytes ? tot_bytes : result);
}
+#endif /* HAVE_DIRECTIO_ITER || HAVE_IOV_ITER_RW */
-struct address_space_operations ll_aops = {
- .readpage = ll_readpage,
-// .readpages = ll_readpages,
- .direct_IO = ll_direct_IO_26,
- .writepage = ll_writepage,
- .writepages = generic_writepages,
- .set_page_dirty = ll_set_page_dirty,
- .sync_page = NULL,
- .prepare_write = ll_prepare_write,
- .commit_write = ll_commit_write,
- .invalidatepage = ll_invalidatepage,
- .releasepage = (void *)ll_releasepage,
- .bmap = NULL
+/**
+ * Prepare partially written-to page for a write.
+ * @pg is owned when passed in and disowned when it returns non-zero result to
+ * the caller.
+ */
+static int ll_prepare_partial_page(const struct lu_env *env, struct cl_io *io,
+ struct cl_page *pg, struct file *file)
+{
+ struct cl_attr *attr = vvp_env_thread_attr(env);
+ struct cl_object *obj = io->ci_obj;
+ struct vvp_page *vpg = cl_object_page_slice(obj, pg);
+ loff_t offset = cl_offset(obj, vvp_index(vpg));
+ int result;
+ ENTRY;
+
+ cl_object_attr_lock(obj);
+ result = cl_object_attr_get(env, obj, attr);
+ cl_object_attr_unlock(obj);
+ if (result) {
+ cl_page_disown(env, io, pg);
+ GOTO(out, result);
+ }
+
+ /*
+ * 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(vpg->vpg_page, KM_USER0);
+
+ memset(kaddr, 0, cl_page_size(obj));
+ ll_kunmap_atomic(kaddr, KM_USER0);
+ GOTO(out, result = 0);
+ }
+
+ if (vpg->vpg_defer_uptodate) {
+ vpg->vpg_ra_used = 1;
+ GOTO(out, result = 0);
+ }
+
+ result = ll_io_read_page(env, io, pg, file);
+ if (result)
+ GOTO(out, result);
+
+ /* ll_io_read_page() disowns the page */
+ result = cl_page_own(env, io, pg);
+ if (!result) {
+ if (!PageUptodate(cl_page_vmpage(pg))) {
+ cl_page_disown(env, io, pg);
+ result = -EIO;
+ }
+ } else if (result == -ENOENT) {
+ /* page was truncated */
+ result = -EAGAIN;
+ }
+ EXIT;
+
+out:
+ return result;
+}
+
+static int ll_tiny_write_begin(struct page *vmpage)
+{
+ /* Page must be present, up to date, dirty, and not in writeback. */
+ if (!vmpage || !PageUptodate(vmpage) || !PageDirty(vmpage) ||
+ PageWriteback(vmpage))
+ return -ENODATA;
+
+ return 0;
+}
+
+static int ll_write_begin(struct file *file, struct address_space *mapping,
+ loff_t pos, unsigned len, unsigned flags,
+ struct page **pagep, void **fsdata)
+{
+ struct ll_cl_context *lcc = NULL;
+ const struct lu_env *env = NULL;
+ struct cl_io *io = NULL;
+ struct cl_page *page = NULL;
+
+ struct cl_object *clob = ll_i2info(mapping->host)->lli_clob;
+ pgoff_t index = pos >> PAGE_SHIFT;
+ struct page *vmpage = NULL;
+ unsigned from = pos & (PAGE_SIZE - 1);
+ unsigned to = from + len;
+ int result = 0;
+ ENTRY;
+
+ CDEBUG(D_VFSTRACE, "Writing %lu of %d to %d bytes\n", index, from, len);
+
+ lcc = ll_cl_find(file);
+ if (lcc == NULL) {
+ vmpage = grab_cache_page_nowait(mapping, index);
+ result = ll_tiny_write_begin(vmpage);
+ GOTO(out, result);
+ }
+
+ env = lcc->lcc_env;
+ io = lcc->lcc_io;
+
+ if (file->f_flags & O_DIRECT) {
+ /* direct IO failed because it couldn't clean up cached pages,
+ * this causes a problem for mirror write because the cached
+ * page may belong to another mirror, which will result in
+ * problem submitting the I/O. */
+ if (io->ci_designated_mirror > 0)
+ GOTO(out, result = -EBUSY);
+
+ /**
+ * Direct read can fall back to buffered read, but DIO is done
+ * with lockless i/o, and buffered requires LDLM locking, so
+ * in this case we must restart without lockless.
+ */
+ if (!io->ci_ignore_lockless) {
+ io->ci_ignore_lockless = 1;
+ io->ci_need_restart = 1;
+ GOTO(out, result = -ENOLCK);
+ }
+ }
+again:
+ /* To avoid deadlock, try to lock page first. */
+ vmpage = grab_cache_page_nowait(mapping, index);
+
+ if (unlikely(vmpage == NULL ||
+ PageDirty(vmpage) || PageWriteback(vmpage))) {
+ struct vvp_io *vio = vvp_env_io(env);
+ struct cl_page_list *plist = &vio->u.write.vui_queue;
+
+ /* if the page is already in dirty cache, we have to commit
+ * the pages right now; otherwise, it may cause deadlock
+ * because it holds page lock of a dirty page and request for
+ * more grants. It's okay for the dirty page to be the first
+ * one in commit page list, though. */
+ if (vmpage != NULL && plist->pl_nr > 0) {
+ unlock_page(vmpage);
+ put_page(vmpage);
+ vmpage = NULL;
+ }
+
+ /* commit pages and then wait for page lock */
+ result = vvp_io_write_commit(env, io);
+ if (result < 0)
+ GOTO(out, result);
+
+ if (vmpage == NULL) {
+ vmpage = grab_cache_page_write_begin(mapping, index,
+ flags);
+ if (vmpage == NULL)
+ GOTO(out, result = -ENOMEM);
+ }
+ }
+
+ page = cl_page_find(env, clob, vmpage->index, vmpage, CPT_CACHEABLE);
+ if (IS_ERR(page))
+ GOTO(out, result = PTR_ERR(page));
+
+ lcc->lcc_page = page;
+ lu_ref_add(&page->cp_reference, "cl_io", io);
+
+ cl_page_assume(env, io, page);
+ 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_SIZE) {
+ CL_PAGE_HEADER(D_PAGE, env, page, "full page write\n");
+ POISON_PAGE(vmpage, 0x11);
+ } else {
+ /* TODO: can be optimized at OSC layer to check if it
+ * is a lockless IO. In that case, it's not necessary
+ * to read the data. */
+ result = ll_prepare_partial_page(env, io, page, file);
+ if (result) {
+ /* vmpage should have been unlocked */
+ put_page(vmpage);
+ vmpage = NULL;
+
+ if (result == -EAGAIN)
+ goto again;
+ GOTO(out, result);
+ }
+ }
+ }
+ EXIT;
+out:
+ if (result < 0) {
+ if (vmpage != NULL) {
+ unlock_page(vmpage);
+ put_page(vmpage);
+ }
+ /* On tiny_write failure, page and io are always null. */
+ if (!IS_ERR_OR_NULL(page)) {
+ lu_ref_del(&page->cp_reference, "cl_io", io);
+ cl_page_put(env, page);
+ }
+ if (io)
+ io->ci_result = result;
+ } else {
+ *pagep = vmpage;
+ *fsdata = lcc;
+ }
+ RETURN(result);
+}
+
+static int ll_tiny_write_end(struct file *file, struct address_space *mapping,
+ loff_t pos, unsigned int len, unsigned int copied,
+ struct page *vmpage)
+{
+ struct cl_page *clpage = (struct cl_page *) vmpage->private;
+ loff_t kms = pos+copied;
+ loff_t to = kms & (PAGE_SIZE-1) ? kms & (PAGE_SIZE-1) : PAGE_SIZE;
+ __u16 refcheck;
+ struct lu_env *env = cl_env_get(&refcheck);
+ int rc = 0;
+
+ ENTRY;
+
+ if (IS_ERR(env)) {
+ rc = PTR_ERR(env);
+ goto out;
+ }
+
+ /* This page is dirty in cache, so it should have a cl_page pointer
+ * set in vmpage->private.
+ */
+ LASSERT(clpage != NULL);
+
+ if (copied == 0)
+ goto out_env;
+
+ /* Update the underlying size information in the OSC/LOV objects this
+ * page is part of.
+ */
+ cl_page_touch(env, clpage, to);
+
+out_env:
+ cl_env_put(env, &refcheck);
+
+out:
+ /* Must return page unlocked. */
+ unlock_page(vmpage);
+
+ RETURN(rc);
+}
+
+static int ll_write_end(struct file *file, struct address_space *mapping,
+ loff_t pos, unsigned len, unsigned copied,
+ struct page *vmpage, void *fsdata)
+{
+ struct ll_cl_context *lcc = fsdata;
+ const struct lu_env *env;
+ struct cl_io *io;
+ struct vvp_io *vio;
+ struct cl_page *page;
+ unsigned from = pos & (PAGE_SIZE - 1);
+ bool unplug = false;
+ int result = 0;
+ ENTRY;
+
+ put_page(vmpage);
+
+ CDEBUG(D_VFSTRACE, "pos %llu, len %u, copied %u\n", pos, len, copied);
+
+ if (lcc == NULL) {
+ result = ll_tiny_write_end(file, mapping, pos, len, copied,
+ vmpage);
+ GOTO(out, result);
+ }
+
+ LASSERT(lcc != NULL);
+ env = lcc->lcc_env;
+ page = lcc->lcc_page;
+ io = lcc->lcc_io;
+ vio = vvp_env_io(env);
+
+ LASSERT(cl_page_is_owned(page, io));
+ if (copied > 0) {
+ struct cl_page_list *plist = &vio->u.write.vui_queue;
+
+ lcc->lcc_page = NULL; /* page will be queued */
+
+ /* Add it into write queue */
+ cl_page_list_add(plist, page);
+ if (plist->pl_nr == 1) /* first page */
+ vio->u.write.vui_from = from;
+ else
+ LASSERT(from == 0);
+ vio->u.write.vui_to = from + copied;
+
+ /* To address the deadlock in balance_dirty_pages() where
+ * this dirty page may be written back in the same thread. */
+ if (PageDirty(vmpage))
+ unplug = true;
+
+ /* We may have one full RPC, commit it soon */
+ if (plist->pl_nr >= PTLRPC_MAX_BRW_PAGES)
+ unplug = true;
+
+ CL_PAGE_DEBUG(D_VFSTRACE, env, page,
+ "queued page: %d.\n", plist->pl_nr);
+ } else {
+ cl_page_disown(env, io, page);
+
+ lcc->lcc_page = NULL;
+ lu_ref_del(&page->cp_reference, "cl_io", io);
+ cl_page_put(env, page);
+
+ /* page list is not contiguous now, commit it now */
+ unplug = true;
+ }
+ if (unplug || io->u.ci_wr.wr_sync)
+ result = vvp_io_write_commit(env, io);
+
+ if (result < 0)
+ io->ci_result = result;
+
+
+out:
+ RETURN(result >= 0 ? copied : result);
+}
+
+#ifdef CONFIG_MIGRATION
+static 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
+
+const struct address_space_operations ll_aops = {
+ .readpage = ll_readpage,
+ .direct_IO = ll_direct_IO,
+ .writepage = ll_writepage,
+ .writepages = ll_writepages,
+ .set_page_dirty = __set_page_dirty_nobuffers,
+ .write_begin = ll_write_begin,
+ .write_end = ll_write_end,
+ .invalidatepage = ll_invalidatepage,
+ .releasepage = (void *)ll_releasepage,
+#ifdef CONFIG_MIGRATION
+ .migratepage = ll_migratepage,
+#endif
};
git://git.whamcloud.com / fs / lustre-release.git / blobdiff