struct nfs_open_context *ctx; /* file open context info */
struct kiocb * iocb; /* controlling i/o request */
struct inode * inode; /* target file of i/o */
- unsigned long user_addr; /* location of user's buffer */
- size_t user_count; /* total bytes to move */
- loff_t pos; /* starting offset in file */
- struct page ** pages; /* pages in our buffer */
- unsigned int npages; /* count of pages */
/* completion state */
atomic_t io_count; /* i/os we're waiting for */
return atomic_dec_and_test(&dreq->io_count);
}
+/*
+ * "size" is never larger than rsize or wsize.
+ */
+static inline int nfs_direct_count_pages(unsigned long user_addr, size_t size)
+{
+ int page_count;
+
+ page_count = (user_addr + size + PAGE_SIZE - 1) >> PAGE_SHIFT;
+ page_count -= user_addr >> PAGE_SHIFT;
+ BUG_ON(page_count < 0);
+
+ return page_count;
+}
+
/**
* nfs_direct_IO - NFS address space operation for direct I/O
* @rw: direction (read or write)
return -EINVAL;
}
-static void nfs_free_user_pages(struct page **pages, int npages, int do_dirty)
+static void nfs_direct_dirty_pages(struct page **pages, int npages)
{
int i;
for (i = 0; i < npages; i++) {
struct page *page = pages[i];
- if (do_dirty && !PageCompound(page))
+ if (!PageCompound(page))
set_page_dirty_lock(page);
- page_cache_release(page);
}
- kfree(pages);
}
-static inline int nfs_get_user_pages(int rw, unsigned long user_addr, size_t size, struct page ***pages)
+static void nfs_direct_release_pages(struct page **pages, int npages)
{
- int result = -ENOMEM;
- unsigned long page_count;
- size_t array_size;
-
- page_count = (user_addr + size + PAGE_SIZE - 1) >> PAGE_SHIFT;
- page_count -= user_addr >> PAGE_SHIFT;
-
- array_size = (page_count * sizeof(struct page *));
- *pages = kmalloc(array_size, GFP_KERNEL);
- if (*pages) {
- down_read(¤t->mm->mmap_sem);
- result = get_user_pages(current, current->mm, user_addr,
- page_count, (rw == READ), 0,
- *pages, NULL);
- up_read(¤t->mm->mmap_sem);
- if (result != page_count) {
- /*
- * If we got fewer pages than expected from
- * get_user_pages(), the user buffer runs off the
- * end of a mapping; return EFAULT.
- */
- if (result >= 0) {
- nfs_free_user_pages(*pages, result, 0);
- result = -EFAULT;
- } else
- kfree(*pages);
- *pages = NULL;
- }
- }
- return result;
+ int i;
+ for (i = 0; i < npages; i++)
+ page_cache_release(pages[i]);
}
static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
}
/*
- * We must hold a reference to all the pages in this direct read request
- * until the RPCs complete. This could be long *after* we are woken up in
- * nfs_direct_wait (for instance, if someone hits ^C on a slow server).
- *
- * In addition, synchronous I/O uses a stack-allocated iocb. Thus we
- * can't trust the iocb is still valid here if this is a synchronous
- * request. If the waiter is woken prematurely, the iocb is long gone.
+ * Synchronous I/O uses a stack-allocated iocb. Thus we can't trust
+ * the iocb is still valid here if this is a synchronous request.
*/
static void nfs_direct_complete(struct nfs_direct_req *dreq)
{
- nfs_free_user_pages(dreq->pages, dreq->npages, 1);
-
if (dreq->iocb) {
long res = (long) dreq->error;
if (!res)
return dreq;
}
+/*
+ * We must hold a reference to all the pages in this direct read request
+ * until the RPCs complete. This could be long *after* we are woken up in
+ * nfs_direct_wait (for instance, if someone hits ^C on a slow server).
+ */
static void nfs_direct_read_result(struct rpc_task *task, void *calldata)
{
struct nfs_read_data *data = calldata;
if (nfs_readpage_result(task, data) != 0)
return;
+ nfs_direct_dirty_pages(data->pagevec, data->npages);
+ nfs_direct_release_pages(data->pagevec, data->npages);
+
spin_lock(&dreq->lock);
if (likely(task->tk_status >= 0))
/*
* For each nfs_read_data struct that was allocated on the list, dispatch
- * an NFS READ operation
+ * an NFS READ operation. If get_user_pages() fails, we stop sending reads.
+ * Read length accounting is handled by nfs_direct_read_result().
+ * Otherwise, if no requests have been sent, just return an error.
*/
-static void nfs_direct_read_schedule(struct nfs_direct_req *dreq)
+static ssize_t nfs_direct_read_schedule(struct nfs_direct_req *dreq, unsigned long user_addr, size_t count, loff_t pos)
{
struct nfs_open_context *ctx = dreq->ctx;
struct inode *inode = ctx->dentry->d_inode;
struct list_head *list = &dreq->list;
- struct page **pages = dreq->pages;
- size_t count = dreq->user_count;
- loff_t pos = dreq->pos;
size_t rsize = NFS_SERVER(inode)->rsize;
- unsigned int curpage, pgbase;
+ unsigned int pgbase;
+ int result;
+ ssize_t started = 0;
+ struct nfs_read_data *data;
- curpage = 0;
- pgbase = dreq->user_addr & ~PAGE_MASK;
+ pgbase = user_addr & ~PAGE_MASK;
do {
- struct nfs_read_data *data;
size_t bytes;
bytes = rsize;
data = list_entry(list->next, struct nfs_read_data, pages);
list_del_init(&data->pages);
+ data->npages = nfs_direct_count_pages(user_addr, bytes);
+ down_read(¤t->mm->mmap_sem);
+ result = get_user_pages(current, current->mm, user_addr,
+ data->npages, 1, 0, data->pagevec, NULL);
+ up_read(¤t->mm->mmap_sem);
+ if (unlikely(result < data->npages))
+ goto out_err;
+
data->inode = inode;
data->cred = ctx->cred;
data->args.fh = NFS_FH(inode);
data->args.context = ctx;
data->args.offset = pos;
data->args.pgbase = pgbase;
- data->args.pages = &pages[curpage];
+ data->args.pages = data->pagevec;
data->args.count = bytes;
data->res.fattr = &data->fattr;
data->res.eof = 0;
bytes,
(unsigned long long)data->args.offset);
+ started += bytes;
+ user_addr += bytes;
pos += bytes;
pgbase += bytes;
- curpage += pgbase >> PAGE_SHIFT;
pgbase &= ~PAGE_MASK;
count -= bytes;
} while (count != 0);
BUG_ON(!list_empty(list));
+ return 0;
+
+out_err:
+ if (result > 0)
+ nfs_direct_release_pages(data->pagevec, result);
+
+ list_add(&data->pages, list);
+ while (!list_empty(list)) {
+ data = list_entry(list->next, struct nfs_read_data, pages);
+ list_del(&data->pages);
+ nfs_readdata_free(data);
+ if (put_dreq(dreq))
+ nfs_direct_complete(dreq);
+ }
+
+ if (started)
+ return 0;
+ return result < 0 ? (ssize_t) result : -EFAULT;
}
-static ssize_t nfs_direct_read(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos, struct page **pages, unsigned int nr_pages)
+static ssize_t nfs_direct_read(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos)
{
ssize_t result;
sigset_t oldset;
if (!dreq)
return -ENOMEM;
- dreq->user_addr = user_addr;
- dreq->user_count = count;
- dreq->pos = pos;
- dreq->pages = pages;
- dreq->npages = nr_pages;
dreq->inode = inode;
dreq->ctx = get_nfs_open_context((struct nfs_open_context *)iocb->ki_filp->private_data);
if (!is_sync_kiocb(iocb))
nfs_add_stats(inode, NFSIOS_DIRECTREADBYTES, count);
rpc_clnt_sigmask(clnt, &oldset);
- nfs_direct_read_schedule(dreq);
- result = nfs_direct_wait(dreq);
+ result = nfs_direct_read_schedule(dreq, user_addr, count, pos);
+ if (!result)
+ result = nfs_direct_wait(dreq);
rpc_clnt_sigunmask(clnt, &oldset);
return result;
while (!list_empty(&dreq->list)) {
struct nfs_write_data *data = list_entry(dreq->list.next, struct nfs_write_data, pages);
list_del(&data->pages);
+ nfs_direct_release_pages(data->pagevec, data->npages);
nfs_writedata_release(data);
}
}
data->cred = dreq->ctx->cred;
data->args.fh = NFS_FH(data->inode);
- data->args.offset = dreq->pos;
- data->args.count = dreq->user_count;
+ data->args.offset = 0;
+ data->args.count = 0;
data->res.count = 0;
data->res.fattr = &data->fattr;
data->res.verf = &data->verf;
/*
* For each nfs_write_data struct that was allocated on the list, dispatch
- * an NFS WRITE operation
+ * an NFS WRITE operation. If get_user_pages() fails, we stop sending writes.
+ * Write length accounting is handled by nfs_direct_write_result().
+ * Otherwise, if no requests have been sent, just return an error.
*/
-static void nfs_direct_write_schedule(struct nfs_direct_req *dreq, int sync)
+static ssize_t nfs_direct_write_schedule(struct nfs_direct_req *dreq, unsigned long user_addr, size_t count, loff_t pos, int sync)
{
struct nfs_open_context *ctx = dreq->ctx;
struct inode *inode = ctx->dentry->d_inode;
struct list_head *list = &dreq->list;
- struct page **pages = dreq->pages;
- size_t count = dreq->user_count;
- loff_t pos = dreq->pos;
size_t wsize = NFS_SERVER(inode)->wsize;
- unsigned int curpage, pgbase;
+ unsigned int pgbase;
+ int result;
+ ssize_t started = 0;
+ struct nfs_write_data *data;
- curpage = 0;
- pgbase = dreq->user_addr & ~PAGE_MASK;
+ pgbase = user_addr & ~PAGE_MASK;
do {
- struct nfs_write_data *data;
size_t bytes;
bytes = wsize;
BUG_ON(list_empty(list));
data = list_entry(list->next, struct nfs_write_data, pages);
+
+ data->npages = nfs_direct_count_pages(user_addr, bytes);
+ down_read(¤t->mm->mmap_sem);
+ result = get_user_pages(current, current->mm, user_addr,
+ data->npages, 0, 0, data->pagevec, NULL);
+ up_read(¤t->mm->mmap_sem);
+ if (unlikely(result < data->npages))
+ goto out_err;
+
list_move_tail(&data->pages, &dreq->rewrite_list);
data->inode = inode;
data->args.context = ctx;
data->args.offset = pos;
data->args.pgbase = pgbase;
- data->args.pages = &pages[curpage];
+ data->args.pages = data->pagevec;
data->args.count = bytes;
data->res.fattr = &data->fattr;
data->res.count = bytes;
bytes,
(unsigned long long)data->args.offset);
+ started += bytes;
+ user_addr += bytes;
pos += bytes;
pgbase += bytes;
- curpage += pgbase >> PAGE_SHIFT;
pgbase &= ~PAGE_MASK;
count -= bytes;
} while (count != 0);
BUG_ON(!list_empty(list));
+ return 0;
+
+out_err:
+ if (result > 0)
+ nfs_direct_release_pages(data->pagevec, result);
+
+ list_add(&data->pages, list);
+ while (!list_empty(list)) {
+ data = list_entry(list->next, struct nfs_write_data, pages);
+ list_del(&data->pages);
+ nfs_writedata_free(data);
+ if (put_dreq(dreq))
+ nfs_direct_write_complete(dreq, inode);
+ }
+
+ if (started)
+ return 0;
+ return result < 0 ? (ssize_t) result : -EFAULT;
}
-static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos, struct page **pages, int nr_pages)
+static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos)
{
ssize_t result;
sigset_t oldset;
if (dreq->commit_data == NULL || count < wsize)
sync = FLUSH_STABLE;
- dreq->user_addr = user_addr;
- dreq->user_count = count;
- dreq->pos = pos;
- dreq->pages = pages;
- dreq->npages = nr_pages;
dreq->inode = inode;
dreq->ctx = get_nfs_open_context((struct nfs_open_context *)iocb->ki_filp->private_data);
if (!is_sync_kiocb(iocb))
nfs_begin_data_update(inode);
rpc_clnt_sigmask(clnt, &oldset);
- nfs_direct_write_schedule(dreq, sync);
- result = nfs_direct_wait(dreq);
+ result = nfs_direct_write_schedule(dreq, user_addr, count, pos, sync);
+ if (!result)
+ result = nfs_direct_wait(dreq);
rpc_clnt_sigunmask(clnt, &oldset);
return result;
ssize_t nfs_file_direct_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
{
ssize_t retval = -EINVAL;
- int page_count;
- struct page **pages;
struct file *file = iocb->ki_filp;
struct address_space *mapping = file->f_mapping;
if (retval)
goto out;
- retval = nfs_get_user_pages(READ, (unsigned long) buf,
- count, &pages);
- if (retval < 0)
- goto out;
- page_count = retval;
-
- retval = nfs_direct_read(iocb, (unsigned long) buf, count, pos,
- pages, page_count);
+ retval = nfs_direct_read(iocb, (unsigned long) buf, count, pos);
if (retval > 0)
iocb->ki_pos = pos + retval;
ssize_t nfs_file_direct_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
{
ssize_t retval;
- int page_count;
- struct page **pages;
struct file *file = iocb->ki_filp;
struct address_space *mapping = file->f_mapping;
if (retval)
goto out;
- retval = nfs_get_user_pages(WRITE, (unsigned long) buf,
- count, &pages);
- if (retval < 0)
- goto out;
- page_count = retval;
-
- retval = nfs_direct_write(iocb, (unsigned long) buf, count,
- pos, pages, page_count);
+ retval = nfs_direct_write(iocb, (unsigned long) buf, count, pos);
/*
* XXX: nfs_end_data_update() already ensures this file's