*
* There are important applications whose performance or correctness
* depends on uncached access to file data. Database clusters
- * (multiple copies of the same instance running on separate hosts)
+ * (multiple copies of the same instance running on separate hosts)
* implement their own cache coherency protocol that subsumes file
- * system cache protocols. Applications that process datasets
- * considerably larger than the client's memory do not always benefit
- * from a local cache. A streaming video server, for instance, has no
+ * system cache protocols. Applications that process datasets
+ * considerably larger than the client's memory do not always benefit
+ * from a local cache. A streaming video server, for instance, has no
* need to cache the contents of a file.
*
* When an application requests uncached I/O, all read and write requests
* 08 Jun 2003 Port to 2.5 APIs --cel
* 31 Mar 2004 Handle direct I/O without VFS support --cel
* 15 Sep 2004 Parallel async reads --cel
+ * 04 May 2005 support O_DIRECT with aio --cel
*
*/
-#include <linux/config.h>
#include <linux/errno.h>
#include <linux/sched.h>
#include <linux/kernel.h>
-#include <linux/smp_lock.h>
#include <linux/file.h>
#include <linux/pagemap.h>
#include <linux/kref.h>
#include <asm/uaccess.h>
#include <asm/atomic.h>
+#include "internal.h"
#include "iostat.h"
#define NFSDBG_FACILITY NFSDBG_VFS
-#define MAX_DIRECTIO_SIZE (4096UL << PAGE_SHIFT)
-static void nfs_free_user_pages(struct page **pages, int npages, int do_dirty);
-static kmem_cache_t *nfs_direct_cachep;
+static struct kmem_cache *nfs_direct_cachep;
/*
* This represents a set of asynchronous requests that we're waiting on
*/
struct nfs_direct_req {
struct kref kref; /* release manager */
- struct list_head list; /* nfs_read_data structs */
- struct file * filp; /* file descriptor */
+
+ /* I/O parameters */
+ struct nfs_open_context *ctx; /* file open context info */
struct kiocb * iocb; /* controlling i/o request */
- wait_queue_head_t wait; /* wait for i/o completion */
- struct inode * inode; /* target file of I/O */
- struct page ** pages; /* pages in our buffer */
- unsigned int npages; /* count of pages */
- atomic_t complete, /* i/os we're waiting for */
- count, /* bytes actually processed */
+ struct inode * inode; /* target file of i/o */
+
+ /* completion state */
+ atomic_t io_count; /* i/os we're waiting for */
+ spinlock_t lock; /* protect completion state */
+ ssize_t count, /* bytes actually processed */
error; /* any reported error */
+ struct completion completion; /* wait for i/o completion */
+
+ /* commit state */
+ struct list_head rewrite_list; /* saved nfs_write_data structs */
+ struct nfs_write_data * commit_data; /* special write_data for commits */
+ int flags;
+#define NFS_ODIRECT_DO_COMMIT (1) /* an unstable reply was received */
+#define NFS_ODIRECT_RESCHED_WRITES (2) /* write verification failed */
+ struct nfs_writeverf verf; /* unstable write verifier */
};
+static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode);
+static const struct rpc_call_ops nfs_write_direct_ops;
+
+static inline void get_dreq(struct nfs_direct_req *dreq)
+{
+ atomic_inc(&dreq->io_count);
+}
+
+static inline int put_dreq(struct nfs_direct_req *dreq)
+{
+ return atomic_dec_and_test(&dreq->io_count);
+}
/**
* nfs_direct_IO - NFS address space operation for direct I/O
*/
ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t pos, unsigned long nr_segs)
{
- struct dentry *dentry = iocb->ki_filp->f_dentry;
-
dprintk("NFS: nfs_direct_IO (%s) off/no(%Ld/%lu) EINVAL\n",
- dentry->d_name.name, (long long) pos, nr_segs);
+ iocb->ki_filp->f_path.dentry->d_name.name,
+ (long long) pos, nr_segs);
return -EINVAL;
}
-static inline int nfs_get_user_pages(int rw, unsigned long user_addr, size_t size, struct page ***pages)
+static void nfs_direct_dirty_pages(struct page **pages, unsigned int pgbase, size_t count)
{
- int result = -ENOMEM;
- unsigned long page_count;
- size_t array_size;
-
- /* set an arbitrary limit to prevent type overflow */
- /* XXX: this can probably be as large as INT_MAX */
- if (size > MAX_DIRECTIO_SIZE) {
- *pages = NULL;
- return -EFBIG;
- }
+ unsigned int npages;
+ unsigned int i;
- 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 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 && result < page_count) {
- nfs_free_user_pages(*pages, result, 0);
- *pages = NULL;
- result = -EFAULT;
- }
+ if (count == 0)
+ return;
+ pages += (pgbase >> PAGE_SHIFT);
+ npages = (count + (pgbase & ~PAGE_MASK) + PAGE_SIZE - 1) >> PAGE_SHIFT;
+ for (i = 0; i < npages; i++) {
+ struct page *page = pages[i];
+ if (!PageCompound(page))
+ set_page_dirty(page);
}
- return result;
}
-static void nfs_free_user_pages(struct page **pages, int npages, int do_dirty)
+static void nfs_direct_release_pages(struct page **pages, unsigned int npages)
{
- int i;
- for (i = 0; i < npages; i++) {
- struct page *page = pages[i];
- if (do_dirty && !PageCompound(page))
- set_page_dirty_lock(page);
- page_cache_release(page);
- }
- kfree(pages);
+ unsigned int i;
+ for (i = 0; i < npages; i++)
+ page_cache_release(pages[i]);
}
static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
{
struct nfs_direct_req *dreq;
- dreq = kmem_cache_alloc(nfs_direct_cachep, SLAB_KERNEL);
+ dreq = kmem_cache_alloc(nfs_direct_cachep, GFP_KERNEL);
if (!dreq)
return NULL;
kref_init(&dreq->kref);
- init_waitqueue_head(&dreq->wait);
- INIT_LIST_HEAD(&dreq->list);
+ kref_get(&dreq->kref);
+ init_completion(&dreq->completion);
+ INIT_LIST_HEAD(&dreq->rewrite_list);
dreq->iocb = NULL;
- atomic_set(&dreq->count, 0);
- atomic_set(&dreq->error, 0);
+ dreq->ctx = NULL;
+ spin_lock_init(&dreq->lock);
+ atomic_set(&dreq->io_count, 0);
+ dreq->count = 0;
+ dreq->error = 0;
+ dreq->flags = 0;
return dreq;
}
-static void nfs_direct_req_release(struct kref *kref)
+static void nfs_direct_req_free(struct kref *kref)
{
struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref);
+
+ if (dreq->ctx != NULL)
+ put_nfs_open_context(dreq->ctx);
kmem_cache_free(nfs_direct_cachep, dreq);
}
+static void nfs_direct_req_release(struct nfs_direct_req *dreq)
+{
+ kref_put(&dreq->kref, nfs_direct_req_free);
+}
+
/*
* Collects and returns the final error value/byte-count.
*/
static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq)
{
- int result = -EIOCBQUEUED;
+ ssize_t result = -EIOCBQUEUED;
+ struct rpc_clnt *clnt;
+ sigset_t oldset;
/* Async requests don't wait here */
if (dreq->iocb)
goto out;
- result = wait_event_interruptible(dreq->wait,
- (atomic_read(&dreq->complete) == 0));
+ clnt = NFS_CLIENT(dreq->inode);
+ rpc_clnt_sigmask(clnt, &oldset);
+ result = wait_for_completion_interruptible(&dreq->completion);
+ rpc_clnt_sigunmask(clnt, &oldset);
if (!result)
- result = atomic_read(&dreq->error);
+ result = dreq->error;
if (!result)
- result = atomic_read(&dreq->count);
+ result = dreq->count;
out:
- kref_put(&dreq->kref, nfs_direct_req_release);
return (ssize_t) result;
}
/*
- * 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 = atomic_read(&dreq->error);
+ long res = (long) dreq->error;
if (!res)
- res = atomic_read(&dreq->count);
+ res = (long) dreq->count;
aio_complete(dreq->iocb, res, 0);
- } else
- wake_up(&dreq->wait);
+ }
+ complete_all(&dreq->completion);
- kref_put(&dreq->kref, nfs_direct_req_release);
+ nfs_direct_req_release(dreq);
}
/*
- * Note we also set the number of requests we have in the dreq when we are
- * done. This prevents races with I/O completion so we will always wait
- * until all requests have been dispatched and completed.
+ * 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 struct nfs_direct_req *nfs_direct_read_alloc(size_t nbytes, size_t rsize)
-{
- struct list_head *list;
- struct nfs_direct_req *dreq;
- unsigned int reads = 0;
- unsigned int rpages = (rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
-
- dreq = nfs_direct_req_alloc();
- if (!dreq)
- return NULL;
-
- list = &dreq->list;
- for(;;) {
- struct nfs_read_data *data = nfs_readdata_alloc(rpages);
-
- if (unlikely(!data)) {
- while (!list_empty(list)) {
- data = list_entry(list->next,
- struct nfs_read_data, pages);
- list_del(&data->pages);
- nfs_readdata_free(data);
- }
- kref_put(&dreq->kref, nfs_direct_req_release);
- return NULL;
- }
-
- INIT_LIST_HEAD(&data->pages);
- list_add(&data->pages, list);
-
- data->req = (struct nfs_page *) dreq;
- reads++;
- if (nbytes <= rsize)
- break;
- nbytes -= rsize;
- }
- kref_get(&dreq->kref);
- atomic_set(&dreq->complete, reads);
- return dreq;
-}
-
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;
- if (likely(task->tk_status >= 0))
- atomic_add(data->res.count, &dreq->count);
- else
- atomic_set(&dreq->error, task->tk_status);
- if (unlikely(atomic_dec_and_test(&dreq->complete)))
+ spin_lock(&dreq->lock);
+ if (unlikely(task->tk_status < 0)) {
+ dreq->error = task->tk_status;
+ spin_unlock(&dreq->lock);
+ } else {
+ dreq->count += data->res.count;
+ spin_unlock(&dreq->lock);
+ nfs_direct_dirty_pages(data->pagevec,
+ data->args.pgbase,
+ data->res.count);
+ }
+ nfs_direct_release_pages(data->pagevec, data->npages);
+
+ if (put_dreq(dreq))
nfs_direct_complete(dreq);
}
};
/*
- * For each nfs_read_data struct that was allocated on the list, dispatch
- * an NFS READ operation
+ * For each rsize'd chunk of the user's buffer, dispatch an NFS READ
+ * operation. If nfs_readdata_alloc() or get_user_pages() fails,
+ * bail and stop sending more reads. Read length accounting is
+ * handled automatically 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, unsigned long user_addr, size_t count, loff_t file_offset)
+static ssize_t nfs_direct_read_schedule_segment(struct nfs_direct_req *dreq,
+ const struct iovec *iov,
+ loff_t pos)
{
- struct file *file = dreq->filp;
- struct inode *inode = file->f_mapping->host;
- struct nfs_open_context *ctx = (struct nfs_open_context *)
- file->private_data;
- struct list_head *list = &dreq->list;
- struct page **pages = dreq->pages;
+ struct nfs_open_context *ctx = dreq->ctx;
+ struct inode *inode = ctx->path.dentry->d_inode;
+ unsigned long user_addr = (unsigned long)iov->iov_base;
+ size_t count = iov->iov_len;
size_t rsize = NFS_SERVER(inode)->rsize;
- unsigned int curpage, pgbase;
+ struct rpc_task *task;
+ struct rpc_message msg = {
+ .rpc_cred = ctx->cred,
+ };
+ struct rpc_task_setup task_setup_data = {
+ .rpc_client = NFS_CLIENT(inode),
+ .rpc_message = &msg,
+ .callback_ops = &nfs_read_direct_ops,
+ .flags = RPC_TASK_ASYNC,
+ };
+ unsigned int pgbase;
+ int result;
+ ssize_t started = 0;
- curpage = 0;
- pgbase = user_addr & ~PAGE_MASK;
do {
struct nfs_read_data *data;
size_t bytes;
- bytes = rsize;
- if (count < rsize)
- bytes = count;
+ pgbase = user_addr & ~PAGE_MASK;
+ bytes = min(rsize,count);
+
+ result = -ENOMEM;
+ data = nfs_readdata_alloc(nfs_page_array_len(pgbase, bytes));
+ if (unlikely(!data))
+ break;
+
+ 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 (result < 0) {
+ nfs_readdata_release(data);
+ break;
+ }
+ if ((unsigned)result < data->npages) {
+ bytes = result * PAGE_SIZE;
+ if (bytes <= pgbase) {
+ nfs_direct_release_pages(data->pagevec, result);
+ nfs_readdata_release(data);
+ break;
+ }
+ bytes -= pgbase;
+ data->npages = result;
+ }
- data = list_entry(list->next, struct nfs_read_data, pages);
- list_del_init(&data->pages);
+ get_dreq(dreq);
+ data->req = (struct nfs_page *) dreq;
data->inode = inode;
- data->cred = ctx->cred;
+ data->cred = msg.rpc_cred;
data->args.fh = NFS_FH(inode);
data->args.context = ctx;
- data->args.offset = file_offset;
+ 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;
data->res.count = bytes;
+ msg.rpc_argp = &data->args;
+ msg.rpc_resp = &data->res;
- rpc_init_task(&data->task, NFS_CLIENT(inode), RPC_TASK_ASYNC,
- &nfs_read_direct_ops, data);
- NFS_PROTO(inode)->read_setup(data);
+ task_setup_data.task = &data->task;
+ task_setup_data.callback_data = data;
+ NFS_PROTO(inode)->read_setup(data, &msg);
- data->task.tk_cookie = (unsigned long) inode;
+ task = rpc_run_task(&task_setup_data);
+ if (!IS_ERR(task))
+ rpc_put_task(task);
- lock_kernel();
- rpc_execute(&data->task);
- unlock_kernel();
-
- dfprintk(VFS, "NFS: %4d initiated direct read call (req %s/%Ld, %u bytes @ offset %Lu)\n",
+ dprintk("NFS: %5u initiated direct read call "
+ "(req %s/%Ld, %zu bytes @ offset %Lu)\n",
data->task.tk_pid,
inode->i_sb->s_id,
(long long)NFS_FILEID(inode),
bytes,
(unsigned long long)data->args.offset);
- file_offset += bytes;
+ started += bytes;
+ user_addr += bytes;
+ pos += bytes;
+ /* FIXME: Remove this unnecessary math from final patch */
pgbase += bytes;
- curpage += pgbase >> PAGE_SHIFT;
pgbase &= ~PAGE_MASK;
+ BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
count -= bytes;
} while (count != 0);
+
+ if (started)
+ return started;
+ 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 file_offset, struct page **pages, unsigned int nr_pages)
+static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq,
+ const struct iovec *iov,
+ unsigned long nr_segs,
+ loff_t pos)
{
- ssize_t result;
- sigset_t oldset;
+ ssize_t result = -EINVAL;
+ size_t requested_bytes = 0;
+ unsigned long seg;
+
+ get_dreq(dreq);
+
+ for (seg = 0; seg < nr_segs; seg++) {
+ const struct iovec *vec = &iov[seg];
+ result = nfs_direct_read_schedule_segment(dreq, vec, pos);
+ if (result < 0)
+ break;
+ requested_bytes += result;
+ if ((size_t)result < vec->iov_len)
+ break;
+ pos += vec->iov_len;
+ }
+
+ if (put_dreq(dreq))
+ nfs_direct_complete(dreq);
+
+ if (requested_bytes != 0)
+ return 0;
+
+ if (result < 0)
+ return result;
+ return -EIO;
+}
+
+static ssize_t nfs_direct_read(struct kiocb *iocb, const struct iovec *iov,
+ unsigned long nr_segs, loff_t pos)
+{
+ ssize_t result = 0;
struct inode *inode = iocb->ki_filp->f_mapping->host;
- struct rpc_clnt *clnt = NFS_CLIENT(inode);
struct nfs_direct_req *dreq;
- dreq = nfs_direct_read_alloc(count, NFS_SERVER(inode)->rsize);
+ dreq = nfs_direct_req_alloc();
if (!dreq)
return -ENOMEM;
- dreq->pages = pages;
- dreq->npages = nr_pages;
dreq->inode = inode;
- dreq->filp = iocb->ki_filp;
+ dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
if (!is_sync_kiocb(iocb))
dreq->iocb = iocb;
- nfs_add_stats(inode, NFSIOS_DIRECTREADBYTES, count);
- rpc_clnt_sigmask(clnt, &oldset);
- nfs_direct_read_schedule(dreq, user_addr, count, file_offset);
- result = nfs_direct_wait(dreq);
- rpc_clnt_sigunmask(clnt, &oldset);
+ result = nfs_direct_read_schedule_iovec(dreq, iov, nr_segs, pos);
+ if (!result)
+ result = nfs_direct_wait(dreq);
+ nfs_direct_req_release(dreq);
return result;
}
-static struct nfs_direct_req *nfs_direct_write_alloc(size_t nbytes, size_t wsize)
+static void nfs_direct_free_writedata(struct nfs_direct_req *dreq)
{
- struct list_head *list;
- struct nfs_direct_req *dreq;
- unsigned int writes = 0;
- unsigned int wpages = (wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
+ while (!list_empty(&dreq->rewrite_list)) {
+ struct nfs_write_data *data = list_entry(dreq->rewrite_list.next, struct nfs_write_data, pages);
+ list_del(&data->pages);
+ nfs_direct_release_pages(data->pagevec, data->npages);
+ nfs_writedata_release(data);
+ }
+}
- dreq = nfs_direct_req_alloc();
- if (!dreq)
- return NULL;
+#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
+static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
+{
+ struct inode *inode = dreq->inode;
+ struct list_head *p;
+ struct nfs_write_data *data;
+ struct rpc_task *task;
+ struct rpc_message msg = {
+ .rpc_cred = dreq->ctx->cred,
+ };
+ struct rpc_task_setup task_setup_data = {
+ .rpc_client = NFS_CLIENT(inode),
+ .callback_ops = &nfs_write_direct_ops,
+ .flags = RPC_TASK_ASYNC,
+ };
- list = &dreq->list;
- for(;;) {
- struct nfs_write_data *data = nfs_writedata_alloc(wpages);
+ dreq->count = 0;
+ get_dreq(dreq);
- if (unlikely(!data)) {
- while (!list_empty(list)) {
- data = list_entry(list->next,
- struct nfs_write_data, pages);
- list_del(&data->pages);
- nfs_writedata_free(data);
- }
- kref_put(&dreq->kref, nfs_direct_req_release);
- return NULL;
- }
+ list_for_each(p, &dreq->rewrite_list) {
+ data = list_entry(p, struct nfs_write_data, pages);
- INIT_LIST_HEAD(&data->pages);
- list_add(&data->pages, list);
+ get_dreq(dreq);
- data->req = (struct nfs_page *) dreq;
- writes++;
- if (nbytes <= wsize)
- break;
- nbytes -= wsize;
+ /* Use stable writes */
+ data->args.stable = NFS_FILE_SYNC;
+
+ /*
+ * Reset data->res.
+ */
+ nfs_fattr_init(&data->fattr);
+ data->res.count = data->args.count;
+ memset(&data->verf, 0, sizeof(data->verf));
+
+ /*
+ * Reuse data->task; data->args should not have changed
+ * since the original request was sent.
+ */
+ task_setup_data.task = &data->task;
+ task_setup_data.callback_data = data;
+ msg.rpc_argp = &data->args;
+ msg.rpc_resp = &data->res;
+ NFS_PROTO(inode)->write_setup(data, &msg);
+
+ /*
+ * We're called via an RPC callback, so BKL is already held.
+ */
+ task = rpc_run_task(&task_setup_data);
+ if (!IS_ERR(task))
+ rpc_put_task(task);
+
+ dprintk("NFS: %5u rescheduled direct write call (req %s/%Ld, %u bytes @ offset %Lu)\n",
+ data->task.tk_pid,
+ inode->i_sb->s_id,
+ (long long)NFS_FILEID(inode),
+ data->args.count,
+ (unsigned long long)data->args.offset);
}
- kref_get(&dreq->kref);
- atomic_set(&dreq->complete, writes);
- return dreq;
+
+ if (put_dreq(dreq))
+ nfs_direct_write_complete(dreq, inode);
}
-/*
- * Collects and returns the final error value/byte-count.
- */
-static ssize_t nfs_direct_write_wait(struct nfs_direct_req *dreq, int intr)
+static void nfs_direct_commit_result(struct rpc_task *task, void *calldata)
{
- int result = 0;
+ struct nfs_write_data *data = calldata;
+ struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
- if (intr) {
- result = wait_event_interruptible(dreq->wait,
- (atomic_read(&dreq->complete) == 0));
- } else {
- wait_event(dreq->wait, (atomic_read(&dreq->complete) == 0));
+ /* Call the NFS version-specific code */
+ if (NFS_PROTO(data->inode)->commit_done(task, data) != 0)
+ return;
+ if (unlikely(task->tk_status < 0)) {
+ dprintk("NFS: %5u commit failed with error %d.\n",
+ task->tk_pid, task->tk_status);
+ dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
+ } else if (memcmp(&dreq->verf, &data->verf, sizeof(data->verf))) {
+ dprintk("NFS: %5u commit verify failed\n", task->tk_pid);
+ dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
}
- if (!result)
- result = atomic_read(&dreq->error);
- if (!result)
- result = atomic_read(&dreq->count);
+ dprintk("NFS: %5u commit returned %d\n", task->tk_pid, task->tk_status);
+ nfs_direct_write_complete(dreq, data->inode);
+}
- kref_put(&dreq->kref, nfs_direct_req_release);
- return (ssize_t) result;
+static const struct rpc_call_ops nfs_commit_direct_ops = {
+ .rpc_call_done = nfs_direct_commit_result,
+ .rpc_release = nfs_commit_release,
+};
+
+static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
+{
+ struct nfs_write_data *data = dreq->commit_data;
+ struct rpc_task *task;
+ struct rpc_message msg = {
+ .rpc_argp = &data->args,
+ .rpc_resp = &data->res,
+ .rpc_cred = dreq->ctx->cred,
+ };
+ struct rpc_task_setup task_setup_data = {
+ .task = &data->task,
+ .rpc_client = NFS_CLIENT(dreq->inode),
+ .rpc_message = &msg,
+ .callback_ops = &nfs_commit_direct_ops,
+ .callback_data = data,
+ .flags = RPC_TASK_ASYNC,
+ };
+
+ data->inode = dreq->inode;
+ data->cred = msg.rpc_cred;
+
+ data->args.fh = NFS_FH(data->inode);
+ data->args.offset = 0;
+ data->args.count = 0;
+ data->res.count = 0;
+ data->res.fattr = &data->fattr;
+ data->res.verf = &data->verf;
+
+ NFS_PROTO(data->inode)->commit_setup(data, &msg);
+
+ /* Note: task.tk_ops->rpc_release will free dreq->commit_data */
+ dreq->commit_data = NULL;
+
+ dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
+
+ task = rpc_run_task(&task_setup_data);
+ if (!IS_ERR(task))
+ rpc_put_task(task);
+}
+
+static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
+{
+ int flags = dreq->flags;
+
+ dreq->flags = 0;
+ switch (flags) {
+ case NFS_ODIRECT_DO_COMMIT:
+ nfs_direct_commit_schedule(dreq);
+ break;
+ case NFS_ODIRECT_RESCHED_WRITES:
+ nfs_direct_write_reschedule(dreq);
+ break;
+ default:
+ if (dreq->commit_data != NULL)
+ nfs_commit_free(dreq->commit_data);
+ nfs_direct_free_writedata(dreq);
+ nfs_zap_mapping(inode, inode->i_mapping);
+ nfs_direct_complete(dreq);
+ }
}
+static void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
+{
+ dreq->commit_data = nfs_commit_alloc();
+ if (dreq->commit_data != NULL)
+ dreq->commit_data->req = (struct nfs_page *) dreq;
+}
+#else
+static inline void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
+{
+ dreq->commit_data = NULL;
+}
+
+static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
+{
+ nfs_direct_free_writedata(dreq);
+ nfs_zap_mapping(inode, inode->i_mapping);
+ nfs_direct_complete(dreq);
+}
+#endif
+
static void nfs_direct_write_result(struct rpc_task *task, void *calldata)
{
struct nfs_write_data *data = calldata;
if (nfs_writeback_done(task, data) != 0)
return;
- /* If the server fell back to an UNSTABLE write, it's an error. */
- if (unlikely(data->res.verf->committed != NFS_FILE_SYNC))
- status = -EIO;
- if (likely(status >= 0))
- atomic_add(data->res.count, &dreq->count);
- else
- atomic_set(&dreq->error, status);
+ spin_lock(&dreq->lock);
- if (unlikely(atomic_dec_and_test(&dreq->complete)))
- nfs_direct_complete(dreq);
+ if (unlikely(status < 0)) {
+ /* An error has occurred, so we should not commit */
+ dreq->flags = 0;
+ dreq->error = status;
+ }
+ if (unlikely(dreq->error != 0))
+ goto out_unlock;
+
+ dreq->count += data->res.count;
+
+ if (data->res.verf->committed != NFS_FILE_SYNC) {
+ switch (dreq->flags) {
+ case 0:
+ memcpy(&dreq->verf, &data->verf, sizeof(dreq->verf));
+ dreq->flags = NFS_ODIRECT_DO_COMMIT;
+ break;
+ case NFS_ODIRECT_DO_COMMIT:
+ if (memcmp(&dreq->verf, &data->verf, sizeof(dreq->verf))) {
+ dprintk("NFS: %5u write verify failed\n", task->tk_pid);
+ dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
+ }
+ }
+ }
+out_unlock:
+ spin_unlock(&dreq->lock);
+}
+
+/*
+ * NB: Return the value of the first error return code. Subsequent
+ * errors after the first one are ignored.
+ */
+static void nfs_direct_write_release(void *calldata)
+{
+ struct nfs_write_data *data = calldata;
+ struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
+
+ if (put_dreq(dreq))
+ nfs_direct_write_complete(dreq, data->inode);
}
static const struct rpc_call_ops nfs_write_direct_ops = {
.rpc_call_done = nfs_direct_write_result,
- .rpc_release = nfs_writedata_release,
+ .rpc_release = nfs_direct_write_release,
};
/*
- * For each nfs_write_data struct that was allocated on the list, dispatch
- * an NFS WRITE operation
- *
- * XXX: For now, support only FILE_SYNC writes. Later we may add
- * support for UNSTABLE + COMMIT.
+ * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE
+ * operation. If nfs_writedata_alloc() or get_user_pages() fails,
+ * bail and stop sending more writes. Write length accounting is
+ * handled automatically 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, struct inode *inode, struct nfs_open_context *ctx, unsigned long user_addr, size_t count, loff_t file_offset)
+static ssize_t nfs_direct_write_schedule_segment(struct nfs_direct_req *dreq,
+ const struct iovec *iov,
+ loff_t pos, int sync)
{
- struct list_head *list = &dreq->list;
- struct page **pages = dreq->pages;
+ struct nfs_open_context *ctx = dreq->ctx;
+ struct inode *inode = ctx->path.dentry->d_inode;
+ unsigned long user_addr = (unsigned long)iov->iov_base;
+ size_t count = iov->iov_len;
+ struct rpc_task *task;
+ struct rpc_message msg = {
+ .rpc_cred = ctx->cred,
+ };
+ struct rpc_task_setup task_setup_data = {
+ .rpc_client = NFS_CLIENT(inode),
+ .rpc_message = &msg,
+ .callback_ops = &nfs_write_direct_ops,
+ .flags = RPC_TASK_ASYNC,
+ };
size_t wsize = NFS_SERVER(inode)->wsize;
- unsigned int curpage, pgbase;
+ unsigned int pgbase;
+ int result;
+ ssize_t started = 0;
- curpage = 0;
- pgbase = user_addr & ~PAGE_MASK;
do {
struct nfs_write_data *data;
size_t bytes;
- bytes = wsize;
- if (count < wsize)
- bytes = count;
+ pgbase = user_addr & ~PAGE_MASK;
+ bytes = min(wsize,count);
- data = list_entry(list->next, struct nfs_write_data, pages);
- list_del_init(&data->pages);
+ result = -ENOMEM;
+ data = nfs_writedata_alloc(nfs_page_array_len(pgbase, bytes));
+ if (unlikely(!data))
+ break;
+ 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 (result < 0) {
+ nfs_writedata_release(data);
+ break;
+ }
+ if ((unsigned)result < data->npages) {
+ bytes = result * PAGE_SIZE;
+ if (bytes <= pgbase) {
+ nfs_direct_release_pages(data->pagevec, result);
+ nfs_writedata_release(data);
+ break;
+ }
+ bytes -= pgbase;
+ data->npages = result;
+ }
+
+ get_dreq(dreq);
+
+ list_move_tail(&data->pages, &dreq->rewrite_list);
+
+ data->req = (struct nfs_page *) dreq;
data->inode = inode;
- data->cred = ctx->cred;
+ data->cred = msg.rpc_cred;
data->args.fh = NFS_FH(inode);
data->args.context = ctx;
- data->args.offset = file_offset;
+ data->args.offset = pos;
data->args.pgbase = pgbase;
- data->args.pages = &pages[curpage];
+ data->args.pages = data->pagevec;
data->args.count = bytes;
+ data->args.stable = sync;
data->res.fattr = &data->fattr;
data->res.count = bytes;
+ data->res.verf = &data->verf;
- rpc_init_task(&data->task, NFS_CLIENT(inode), RPC_TASK_ASYNC,
- &nfs_write_direct_ops, data);
- NFS_PROTO(inode)->write_setup(data, FLUSH_STABLE);
+ task_setup_data.task = &data->task;
+ task_setup_data.callback_data = data;
+ msg.rpc_argp = &data->args;
+ msg.rpc_resp = &data->res;
+ NFS_PROTO(inode)->write_setup(data, &msg);
- data->task.tk_priority = RPC_PRIORITY_NORMAL;
- data->task.tk_cookie = (unsigned long) inode;
+ task = rpc_run_task(&task_setup_data);
+ if (!IS_ERR(task))
+ rpc_put_task(task);
- lock_kernel();
- rpc_execute(&data->task);
- unlock_kernel();
-
- dfprintk(VFS, "NFS: %4d initiated direct write call (req %s/%Ld, %u bytes @ offset %Lu)\n",
+ dprintk("NFS: %5u initiated direct write call "
+ "(req %s/%Ld, %zu bytes @ offset %Lu)\n",
data->task.tk_pid,
inode->i_sb->s_id,
(long long)NFS_FILEID(inode),
bytes,
(unsigned long long)data->args.offset);
- file_offset += bytes;
+ started += bytes;
+ user_addr += bytes;
+ pos += bytes;
+
+ /* FIXME: Remove this useless math from the final patch */
pgbase += bytes;
- curpage += pgbase >> PAGE_SHIFT;
pgbase &= ~PAGE_MASK;
+ BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
count -= bytes;
} while (count != 0);
+
+ if (started)
+ return started;
+ return result < 0 ? (ssize_t) result : -EFAULT;
}
-static ssize_t nfs_direct_write_seg(struct inode *inode, struct nfs_open_context *ctx, unsigned long user_addr, size_t count, loff_t file_offset, struct page **pages, int nr_pages)
+static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq,
+ const struct iovec *iov,
+ unsigned long nr_segs,
+ loff_t pos, int sync)
{
- ssize_t result;
- sigset_t oldset;
- struct rpc_clnt *clnt = NFS_CLIENT(inode);
- struct nfs_direct_req *dreq;
-
- dreq = nfs_direct_write_alloc(count, NFS_SERVER(inode)->wsize);
- if (!dreq)
- return -ENOMEM;
+ ssize_t result = 0;
+ size_t requested_bytes = 0;
+ unsigned long seg;
- dreq->pages = pages;
- dreq->npages = nr_pages;
+ get_dreq(dreq);
- nfs_begin_data_update(inode);
+ for (seg = 0; seg < nr_segs; seg++) {
+ const struct iovec *vec = &iov[seg];
+ result = nfs_direct_write_schedule_segment(dreq, vec,
+ pos, sync);
+ if (result < 0)
+ break;
+ requested_bytes += result;
+ if ((size_t)result < vec->iov_len)
+ break;
+ pos += vec->iov_len;
+ }
- rpc_clnt_sigmask(clnt, &oldset);
- nfs_direct_write_schedule(dreq, inode, ctx, user_addr, count,
- file_offset);
- result = nfs_direct_write_wait(dreq, clnt->cl_intr);
- rpc_clnt_sigunmask(clnt, &oldset);
+ if (put_dreq(dreq))
+ nfs_direct_write_complete(dreq, dreq->inode);
- nfs_end_data_update(inode);
+ if (requested_bytes != 0)
+ return 0;
- return result;
+ if (result < 0)
+ return result;
+ return -EIO;
}
-/*
- * Upon return, generic_file_direct_IO invalidates any cached pages
- * that non-direct readers might access, so they will pick up these
- * writes immediately.
- */
-static ssize_t nfs_direct_write(struct inode *inode, struct nfs_open_context *ctx, const struct iovec *iov, loff_t file_offset, unsigned long nr_segs)
+static ssize_t nfs_direct_write(struct kiocb *iocb, const struct iovec *iov,
+ unsigned long nr_segs, loff_t pos,
+ size_t count)
{
- ssize_t tot_bytes = 0;
- unsigned long seg = 0;
-
- while ((seg < nr_segs) && (tot_bytes >= 0)) {
- ssize_t result;
- int page_count;
- struct page **pages;
- const struct iovec *vec = &iov[seg++];
- unsigned long user_addr = (unsigned long) vec->iov_base;
- size_t size = vec->iov_len;
-
- page_count = nfs_get_user_pages(WRITE, user_addr, size, &pages);
- if (page_count < 0) {
- nfs_free_user_pages(pages, 0, 0);
- if (tot_bytes > 0)
- break;
- return page_count;
- }
+ ssize_t result = 0;
+ struct inode *inode = iocb->ki_filp->f_mapping->host;
+ struct nfs_direct_req *dreq;
+ size_t wsize = NFS_SERVER(inode)->wsize;
+ int sync = NFS_UNSTABLE;
- nfs_add_stats(inode, NFSIOS_DIRECTWRITTENBYTES, size);
- result = nfs_direct_write_seg(inode, ctx, user_addr, size,
- file_offset, pages, page_count);
+ dreq = nfs_direct_req_alloc();
+ if (!dreq)
+ return -ENOMEM;
+ nfs_alloc_commit_data(dreq);
- if (result <= 0) {
- if (tot_bytes > 0)
- break;
- return result;
- }
- nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, result);
- tot_bytes += result;
- file_offset += result;
- if (result < size)
- break;
- }
- return tot_bytes;
+ if (dreq->commit_data == NULL || count < wsize)
+ sync = NFS_FILE_SYNC;
+
+ dreq->inode = inode;
+ dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
+ if (!is_sync_kiocb(iocb))
+ dreq->iocb = iocb;
+
+ result = nfs_direct_write_schedule_iovec(dreq, iov, nr_segs, pos, sync);
+ if (!result)
+ result = nfs_direct_wait(dreq);
+ nfs_direct_req_release(dreq);
+
+ return result;
}
/**
* nfs_file_direct_read - file direct read operation for NFS files
* @iocb: target I/O control block
- * @buf: user's buffer into which to read data
- * count: number of bytes to read
- * pos: byte offset in file where reading starts
+ * @iov: vector of user buffers into which to read data
+ * @nr_segs: size of iov vector
+ * @pos: byte offset in file where reading starts
*
* We use this function for direct reads instead of calling
* generic_file_aio_read() in order to avoid gfar's check to see if
* the request starts before the end of the file. For that check
* to work, we must generate a GETATTR before each direct read, and
* even then there is a window between the GETATTR and the subsequent
- * READ where the file size could change. So our preference is simply
+ * READ where the file size could change. Our preference is simply
* to do all reads the application wants, and the server will take
* care of managing the end of file boundary.
- *
+ *
* This function also eliminates unnecessarily updating the file's
* atime locally, as the NFS server sets the file's atime, and this
* client must read the updated atime from the server back into its
* cache.
*/
-ssize_t nfs_file_direct_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
+ssize_t nfs_file_direct_read(struct kiocb *iocb, const struct iovec *iov,
+ unsigned long nr_segs, 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;
+ size_t count;
- dprintk("nfs: direct read(%s/%s, %lu@%Ld)\n",
- file->f_dentry->d_parent->d_name.name,
- file->f_dentry->d_name.name,
- (unsigned long) count, (long long) pos);
+ count = iov_length(iov, nr_segs);
+ nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count);
+
+ dprintk("nfs: direct read(%s/%s, %zd@%Ld)\n",
+ file->f_path.dentry->d_parent->d_name.name,
+ file->f_path.dentry->d_name.name,
+ count, (long long) pos);
- if (count < 0)
- goto out;
- retval = -EFAULT;
- if (!access_ok(VERIFY_WRITE, buf, count))
- goto out;
retval = 0;
if (!count)
goto out;
if (retval)
goto out;
- page_count = nfs_get_user_pages(READ, (unsigned long) buf,
- count, &pages);
- if (page_count < 0) {
- nfs_free_user_pages(pages, 0, 0);
- retval = page_count;
- goto out;
- }
-
- retval = nfs_direct_read(iocb, (unsigned long) buf, count, pos,
- pages, page_count);
+ retval = nfs_direct_read(iocb, iov, nr_segs, pos);
if (retval > 0)
iocb->ki_pos = pos + retval;
/**
* nfs_file_direct_write - file direct write operation for NFS files
* @iocb: target I/O control block
- * @buf: user's buffer from which to write data
- * count: number of bytes to write
- * pos: byte offset in file where writing starts
+ * @iov: vector of user buffers from which to write data
+ * @nr_segs: size of iov vector
+ * @pos: byte offset in file where writing starts
*
* We use this function for direct writes instead of calling
* generic_file_aio_write() in order to avoid taking the inode
* Note that O_APPEND is not supported for NFS direct writes, as there
* is no atomic O_APPEND write facility in the NFS protocol.
*/
-ssize_t nfs_file_direct_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
+ssize_t nfs_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
+ unsigned long nr_segs, loff_t pos)
{
- ssize_t retval;
+ ssize_t retval = -EINVAL;
struct file *file = iocb->ki_filp;
- struct nfs_open_context *ctx =
- (struct nfs_open_context *) file->private_data;
struct address_space *mapping = file->f_mapping;
- struct inode *inode = mapping->host;
- struct iovec iov = {
- .iov_base = (char __user *)buf,
- };
+ size_t count;
- dfprintk(VFS, "nfs: direct write(%s/%s, %lu@%Ld)\n",
- file->f_dentry->d_parent->d_name.name,
- file->f_dentry->d_name.name,
- (unsigned long) count, (long long) pos);
+ count = iov_length(iov, nr_segs);
+ nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count);
- retval = -EINVAL;
- if (!is_sync_kiocb(iocb))
- goto out;
+ dfprintk(VFS, "nfs: direct write(%s/%s, %zd@%Ld)\n",
+ file->f_path.dentry->d_parent->d_name.name,
+ file->f_path.dentry->d_name.name,
+ count, (long long) pos);
retval = generic_write_checks(file, &pos, &count, 0);
if (retval)
retval = 0;
if (!count)
goto out;
- iov.iov_len = count,
-
- retval = -EFAULT;
- if (!access_ok(VERIFY_READ, iov.iov_base, iov.iov_len))
- goto out;
retval = nfs_sync_mapping(mapping);
if (retval)
goto out;
- retval = nfs_direct_write(inode, ctx, &iov, pos, 1);
- if (mapping->nrpages)
- invalidate_inode_pages2(mapping);
+ retval = nfs_direct_write(iocb, iov, nr_segs, pos, count);
+
if (retval > 0)
iocb->ki_pos = pos + retval;
return retval;
}
-int nfs_init_directcache(void)
+/**
+ * nfs_init_directcache - create a slab cache for nfs_direct_req structures
+ *
+ */
+int __init nfs_init_directcache(void)
{
nfs_direct_cachep = kmem_cache_create("nfs_direct_cache",
sizeof(struct nfs_direct_req),
- 0, SLAB_RECLAIM_ACCOUNT,
- NULL, NULL);
+ 0, (SLAB_RECLAIM_ACCOUNT|
+ SLAB_MEM_SPREAD),
+ NULL);
if (nfs_direct_cachep == NULL)
return -ENOMEM;
return 0;
}
+/**
+ * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures
+ *
+ */
void nfs_destroy_directcache(void)
{
- if (kmem_cache_destroy(nfs_direct_cachep))
- printk(KERN_INFO "nfs_direct_cache: not all structures were freed\n");
+ kmem_cache_destroy(nfs_direct_cachep);
}