write.c

讲述linux的初始化过程

	status = 0;
	if (synchronous) {
		int error;

		error = nfs_sync_file(inode, file, page_index(page), 1, FLUSH_SYNC|FLUSH_STABLE);
		if (error < 0 || (error = file->f_error) < 0)
			status = error;
		file->f_error = 0;
	} else {
		/* If we wrote past the end of the page.
		 * Call the strategy routine so it can send out a bunch
		 * of requests.
		 */
		if (req->wb_offset == 0 && req->wb_bytes == PAGE_CACHE_SIZE)
			nfs_strategy(inode);
	}
	nfs_release_request(req);
done:
        dprintk("NFS:      nfs_updatepage returns %d (isize %Ld)\n",
                                                status, (long long)inode->i_size);
	if (status < 0)
		ClearPageUptodate(page);
	return status;
}

/*
 * Set up the argument/result storage required for the RPC call.
 */
static void
nfs_write_rpcsetup(struct list_head *head, struct nfs_write_data *data)
{
	struct nfs_page		*req;
	struct iovec		*iov;
	unsigned int		count;

	/* Set up the RPC argument and reply structs
	 * NB: take care not to mess about with data->commit et al. */

	iov = data->args.iov;
	count = 0;
	while (!list_empty(head)) {
		struct nfs_page *req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
		nfs_list_add_request(req, &data->pages);
		iov->iov_base = kmap(req->wb_page) + req->wb_offset;
		iov->iov_len = req->wb_bytes;
		count += req->wb_bytes;
		iov++;
		data->args.nriov++;
	}
	req = nfs_list_entry(data->pages.next);
	data->inode = req->wb_inode;
	data->cred = req->wb_cred;
	data->args.fh     = NFS_FH(req->wb_inode);
	data->args.offset = page_offset(req->wb_page) + req->wb_offset;
	data->args.count  = count;
	data->res.fattr   = &data->fattr;
	data->res.count   = count;
	data->res.verf    = &data->verf;
}


/*
 * Create an RPC task for the given write request and kick it.
 * The page must have been locked by the caller.
 *
 * It may happen that the page we're passed is not marked dirty.
 * This is the case if nfs_updatepage detects a conflicting request
 * that has been written but not committed.
 */
static int
nfs_flush_one(struct list_head *head, struct inode *inode, int how)
{
	struct rpc_clnt 	*clnt = NFS_CLIENT(inode);
	struct nfs_write_data	*data;
	struct rpc_task		*task;
	struct rpc_message	msg;
	int                     flags,
				async = !(how & FLUSH_SYNC),
				stable = (how & FLUSH_STABLE);
	sigset_t		oldset;


	data = nfs_writedata_alloc();
	if (!data)
		goto out_bad;
	task = &data->task;

	/* Set the initial flags for the task.  */
	flags = (async) ? RPC_TASK_ASYNC : 0;

	/* Set up the argument struct */
	nfs_write_rpcsetup(head, data);
	if (stable) {
		if (!inode->u.nfs_i.ncommit)
			data->args.stable = NFS_FILE_SYNC;
		else
			data->args.stable = NFS_DATA_SYNC;
	} else
		data->args.stable = NFS_UNSTABLE;

	/* Finalize the task. */
	rpc_init_task(task, clnt, nfs_writeback_done, flags);
	task->tk_calldata = data;
	/* Release requests */
	task->tk_release = nfs_writedata_release;

#ifdef CONFIG_NFS_V3
	msg.rpc_proc = (NFS_PROTO(inode)->version == 3) ? NFS3PROC_WRITE : NFSPROC_WRITE;
#else
	msg.rpc_proc = NFSPROC_WRITE;
#endif
	msg.rpc_argp = &data->args;
	msg.rpc_resp = &data->res;
	msg.rpc_cred = data->cred;

	dprintk("NFS: %4d initiated write call (req %x/%Ld count %d nriov %d)\n",
		task->tk_pid, 
		inode->i_dev,
		(long long)NFS_FILEID(inode),
		data->args.count, data->args.nriov);

	rpc_clnt_sigmask(clnt, &oldset);
	rpc_call_setup(task, &msg, 0);
	rpc_execute(task);
	rpc_clnt_sigunmask(clnt, &oldset);
	return 0;
 out_bad:
	while (!list_empty(head)) {
		struct nfs_page *req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
		nfs_mark_request_dirty(req);
		nfs_unlock_request(req);
	}
	return -ENOMEM;
}

static int
nfs_flush_list(struct inode *inode, struct list_head *head, int how)
{
	LIST_HEAD(one_request);
	struct nfs_page		*req;
	int			error = 0;
	unsigned int		pages = 0,
				wpages = NFS_SERVER(inode)->wpages;

	while (!list_empty(head)) {
		pages += nfs_coalesce_requests(head, &one_request, wpages);
		req = nfs_list_entry(one_request.next);
		error = nfs_flush_one(&one_request, req->wb_inode, how);
		if (error < 0)
			break;
	}
	if (error >= 0)
		return pages;

	while (!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
		nfs_mark_request_dirty(req);
		nfs_unlock_request(req);
	}
	return error;
}


/*
 * This function is called when the WRITE call is complete.
 */
static void
nfs_writeback_done(struct rpc_task *task)
{
	struct nfs_write_data	*data = (struct nfs_write_data *) task->tk_calldata;
	struct nfs_writeargs	*argp = &data->args;
	struct nfs_writeres	*resp = &data->res;
	struct inode		*inode = data->inode;
	struct nfs_page		*req;
	struct page		*page;

	dprintk("NFS: %4d nfs_writeback_done (status %d)\n",
		task->tk_pid, task->tk_status);

	/* We can't handle that yet but we check for it nevertheless */
	if (resp->count < argp->count && task->tk_status >= 0) {
		static unsigned long    complain;
		if (time_before(complain, jiffies)) {
			printk(KERN_WARNING
			       "NFS: Server wrote less than requested.\n");
			complain = jiffies + 300 * HZ;
		}
		/* Can't do anything about it right now except throw
		 * an error. */
		task->tk_status = -EIO;
	}
#ifdef CONFIG_NFS_V3
	if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
		/* We tried a write call, but the server did not
		 * commit data to stable storage even though we
		 * requested it.
		 * Note: There is a known bug in Tru64 < 5.0 in which
		 *	 the server reports NFS_DATA_SYNC, but performs
		 *	 NFS_FILE_SYNC. We therefore implement this checking
		 *	 as a dprintk() in order to avoid filling syslog.
		 */
		static unsigned long    complain;

		if (time_before(complain, jiffies)) {
			dprintk("NFS: faulty NFSv3 server %s:"
				" (committed = %d) != (stable = %d)\n",
				NFS_SERVER(inode)->hostname,
				resp->verf->committed, argp->stable);
			complain = jiffies + 300 * HZ;
		}
	}
#endif

	/*
	 * Update attributes as result of writeback.
	 * FIXME: There is an inherent race with invalidate_inode_pages and
	 *	  writebacks since the page->count is kept > 1 for as long
	 *	  as the page has a write request pending.
	 */
	nfs_write_attributes(inode, resp->fattr);
	while (!list_empty(&data->pages)) {
		req = nfs_list_entry(data->pages.next);
		nfs_list_remove_request(req);
		page = req->wb_page;

		kunmap(page);

		dprintk("NFS: write (%x/%Ld %d@%Ld)",
			req->wb_inode->i_dev,
			(long long)NFS_FILEID(req->wb_inode),
			req->wb_bytes,
			(long long)(page_offset(page) + req->wb_offset));

		if (task->tk_status < 0) {
			ClearPageUptodate(page);
			SetPageError(page);
			if (req->wb_file)
				req->wb_file->f_error = task->tk_status;
			nfs_inode_remove_request(req);
			dprintk(", error = %d\n", task->tk_status);
			goto next;
		}

#ifdef CONFIG_NFS_V3
		if (resp->verf->committed != NFS_UNSTABLE) {
			nfs_inode_remove_request(req);
			dprintk(" OK\n");
			goto next;
		}
		memcpy(&req->wb_verf, resp->verf, sizeof(req->wb_verf));
		req->wb_timeout = jiffies + NFS_COMMIT_DELAY;
		nfs_mark_request_commit(req);
		dprintk(" marked for commit\n");
#else
		nfs_inode_remove_request(req);
#endif
	next:
		nfs_unlock_request(req);
	}
}


#ifdef CONFIG_NFS_V3
/*
 * Set up the argument/result storage required for the RPC call.
 */
static void
nfs_commit_rpcsetup(struct list_head *head, struct nfs_write_data *data)
{
	struct nfs_page		*first, *last;
	struct inode		*inode;
	loff_t			start, end, len;

	/* Set up the RPC argument and reply structs
	 * NB: take care not to mess about with data->commit et al. */

	list_splice(head, &data->pages);
	INIT_LIST_HEAD(head);
	first = nfs_list_entry(data->pages.next);
	last = nfs_list_entry(data->pages.prev);
	inode = first->wb_inode;

	/*
	 * Determine the offset range of requests in the COMMIT call.
	 * We rely on the fact that data->pages is an ordered list...
	 */
	start = page_offset(first->wb_page) + first->wb_offset;
	end = page_offset(last->wb_page) + (last->wb_offset + last->wb_bytes);
	len = end - start;
	/* If 'len' is not a 32-bit quantity, pass '0' in the COMMIT call */
	if (end >= inode->i_size || len < 0 || len > (~((u32)0) >> 1))
		len = 0;

	data->inode	  = inode;
	data->cred	  = first->wb_cred;
	data->args.fh     = NFS_FH(inode);
	data->args.offset = start;
	data->res.count   = data->args.count = (u32)len;
	data->res.fattr   = &data->fattr;
	data->res.verf    = &data->verf;
}

/*
 * Commit dirty pages
 */
static int
nfs_commit_list(struct list_head *head, int how)
{
	struct rpc_message	msg;
	struct rpc_clnt		*clnt;
	struct nfs_write_data	*data;
	struct rpc_task         *task;
	struct nfs_page         *req;
	int                     flags,
				async = !(how & FLUSH_SYNC);
	sigset_t		oldset;

	data = nfs_writedata_alloc();

	if (!data)
		goto out_bad;
	task = &data->task;

	flags = (async) ? RPC_TASK_ASYNC : 0;

	/* Set up the argument struct */
	nfs_commit_rpcsetup(head, data);
	req = nfs_list_entry(data->pages.next);
	clnt = NFS_CLIENT(req->wb_inode);

	rpc_init_task(task, clnt, nfs_commit_done, flags);
	task->tk_calldata = data;
	/* Release requests */
	task->tk_release = nfs_writedata_release;

	msg.rpc_proc = NFS3PROC_COMMIT;
	msg.rpc_argp = &data->args;
	msg.rpc_resp = &data->res;
	msg.rpc_cred = data->cred;

	dprintk("NFS: %4d initiated commit call\n", task->tk_pid);
	rpc_clnt_sigmask(clnt, &oldset);
	rpc_call_setup(task, &msg, 0);
	rpc_execute(task);
	rpc_clnt_sigunmask(clnt, &oldset);
	return 0;
 out_bad:
	while (!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
		nfs_mark_request_commit(req);
		nfs_unlock_request(req);
	}
	return -ENOMEM;
}

/*
 * COMMIT call returned
 */
static void
nfs_commit_done(struct rpc_task *task)
{
	struct nfs_write_data	*data = (struct nfs_write_data *)task->tk_calldata;
	struct nfs_writeres	*resp = &data->res;
	struct nfs_page		*req;
	struct inode		*inode = data->inode;

        dprintk("NFS: %4d nfs_commit_done (status %d)\n",
                                task->tk_pid, task->tk_status);

	nfs_write_attributes(inode, resp->fattr);
	while (!list_empty(&data->pages)) {
		req = nfs_list_entry(data->pages.next);
		nfs_list_remove_request(req);

		dprintk("NFS: commit (%x/%Ld %d@%Ld)",
			req->wb_inode->i_dev,
			(long long)NFS_FILEID(req->wb_inode),
			req->wb_bytes,
			(long long)(page_offset(req->wb_page) + req->wb_offset));
		if (task->tk_status < 0) {
			if (req->wb_file)
				req->wb_file->f_error = task->tk_status;
			nfs_inode_remove_request(req);
			dprintk(", error = %d\n", task->tk_status);
			goto next;
		}

		/* Okay, COMMIT succeeded, apparently. Check the verifier
		 * returned by the server against all stored verfs. */
		if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
			/* We have a match */
			nfs_inode_remove_request(req);
			dprintk(" OK\n");
			goto next;
		}
		/* We have a mismatch. Write the page again */
		dprintk(" mismatch\n");
		nfs_mark_request_dirty(req);
	next:
		nfs_unlock_request(req);
	}
}
#endif

int nfs_flush_file(struct inode *inode, struct file *file, unsigned long idx_start,
		   unsigned int npages, int how)
{
	LIST_HEAD(head);
	int			res,
				error = 0;

	res = nfs_scan_dirty(inode, &head, file, idx_start, npages);
	if (res)
		error = nfs_flush_list(inode, &head, how);
	if (error < 0)
		return error;
	return res;
}

int nfs_flush_timeout(struct inode *inode, int how)
{
	LIST_HEAD(head);
	int			pages,
				error = 0;

	pages = nfs_scan_dirty_timeout(inode, &head);
	if (pages)
		error = nfs_flush_list(inode, &head, how);
	if (error < 0)
		return error;
	return pages;
}

#ifdef CONFIG_NFS_V3
int nfs_commit_file(struct inode *inode, struct file *file, unsigned long idx_start,
		    unsigned int npages, int how)
{
	LIST_HEAD(head);
	int			res,
				error = 0;

	res = nfs_scan_commit(inode, &head, file, idx_start, npages);
	if (res)
		error = nfs_commit_list(&head, how);
	if (error < 0)
		return error;
	return res;
}

int nfs_commit_timeout(struct inode *inode, int how)
{
	LIST_HEAD(head);
	int			pages,
				error = 0;

	pages = nfs_scan_commit_timeout(inode, &head);
	if (pages) {
		pages += nfs_scan_commit(inode, &head, NULL, 0, 0);
		error = nfs_commit_list(&head, how);
	}
	if (error < 0)
		return error;
	return pages;
}
#endif

int nfs_sync_file(struct inode *inode, struct file *file, unsigned long idx_start,
		  unsigned int npages, int how)
{
	int	error,
		wait;

	wait = how & FLUSH_WAIT;
	how &= ~FLUSH_WAIT;

	if (!inode && file)
		inode = file->f_dentry->d_inode;

	do {
		error = 0;
		if (wait)
			error = nfs_wait_on_requests(inode, file, idx_start, npages);
		if (error == 0)
			error = nfs_flush_file(inode, file, idx_start, npages, how);
#ifdef CONFIG_NFS_V3
		if (error == 0)
			error = nfs_commit_file(inode, file, idx_start, npages, how);
#endif
	} while (error > 0);
	return error;
}

int nfs_init_nfspagecache(void)
{
	nfs_page_cachep = kmem_cache_create("nfs_page",
					    sizeof(struct nfs_page),
					    0, SLAB_HWCACHE_ALIGN,
					    NULL, NULL);
	if (nfs_page_cachep == NULL)
		return -ENOMEM;

	nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
					     sizeof(struct nfs_write_data),
					     0, SLAB_HWCACHE_ALIGN,
					     NULL, NULL);
	if (nfs_wdata_cachep == NULL)
		return -ENOMEM;

	return 0;
}

void nfs_destroy_nfspagecache(void)
{
	if (kmem_cache_destroy(nfs_page_cachep))
		printk(KERN_INFO "nfs_page: not all structures were freed\n");
	if (kmem_cache_destroy(nfs_wdata_cachep))
		printk(KERN_INFO "nfs_write_data: not all structures were freed\n");
}