1/*
  2 * Copyright (c) 2014 Christoph Hellwig.
  3 */
  4#include <linux/iomap.h>
  5#include "xfs.h"
  6#include "xfs_format.h"
  7#include "xfs_log_format.h"
  8#include "xfs_trans_resv.h"
  9#include "xfs_sb.h"
 10#include "xfs_mount.h"
 11#include "xfs_inode.h"
 12#include "xfs_trans.h"
 13#include "xfs_log.h"
 14#include "xfs_bmap.h"
 15#include "xfs_bmap_util.h"
 16#include "xfs_error.h"
 17#include "xfs_iomap.h"
 18#include "xfs_shared.h"
 19#include "xfs_bit.h"
 20#include "xfs_pnfs.h"
 21
 22/*
 23 * Ensure that we do not have any outstanding pNFS layouts that can be used by
 24 * clients to directly read from or write to this inode.  This must be called
 25 * before every operation that can remove blocks from the extent map.
 26 * Additionally we call it during the write operation, where aren't concerned
 27 * about exposing unallocated blocks but just want to provide basic
 28 * synchronization between a local writer and pNFS clients.  mmap writes would
 29 * also benefit from this sort of synchronization, but due to the tricky locking
 30 * rules in the page fault path we don't bother.
 31 */
 32int
 33xfs_break_layouts(
 34	struct inode		*inode,
 35	uint			*iolock)
 36{
 37	struct xfs_inode	*ip = XFS_I(inode);
 38	int			error;
 39
 40	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL));
 41
 42	while ((error = break_layout(inode, false) == -EWOULDBLOCK)) {
 43		xfs_iunlock(ip, *iolock);
 44		error = break_layout(inode, true);
 45		*iolock = XFS_IOLOCK_EXCL;
 46		xfs_ilock(ip, *iolock);
 47	}
 48
 49	return error;
 50}
 51
 52/*
 53 * Get a unique ID including its location so that the client can identify
 54 * the exported device.
 55 */
 56int
 57xfs_fs_get_uuid(
 58	struct super_block	*sb,
 59	u8			*buf,
 60	u32			*len,
 61	u64			*offset)
 62{
 63	struct xfs_mount	*mp = XFS_M(sb);
 64
 65	printk_once(KERN_NOTICE
 66"XFS (%s): using experimental pNFS feature, use at your own risk!\n",
 67		mp->m_fsname);
 68
 69	if (*len < sizeof(uuid_t))
 70		return -EINVAL;
 71
 72	memcpy(buf, &mp->m_sb.sb_uuid, sizeof(uuid_t));
 73	*len = sizeof(uuid_t);
 74	*offset = offsetof(struct xfs_dsb, sb_uuid);
 75	return 0;
 76}
 77
 78/*
 79 * Get a layout for the pNFS client.
 80 */
 81int
 82xfs_fs_map_blocks(
 83	struct inode		*inode,
 84	loff_t			offset,
 85	u64			length,
 86	struct iomap		*iomap,
 87	bool			write,
 88	u32			*device_generation)
 89{
 90	struct xfs_inode	*ip = XFS_I(inode);
 91	struct xfs_mount	*mp = ip->i_mount;
 92	struct xfs_bmbt_irec	imap;
 93	xfs_fileoff_t		offset_fsb, end_fsb;
 94	loff_t			limit;
 95	int			bmapi_flags = XFS_BMAPI_ENTIRE;
 96	int			nimaps = 1;
 97	uint			lock_flags;
 98	int			error = 0;
 99
100	if (XFS_FORCED_SHUTDOWN(mp))
101		return -EIO;
102
103	/*
104	 * We can't export inodes residing on the realtime device.  The realtime
105	 * device doesn't have a UUID to identify it, so the client has no way
106	 * to find it.
107	 */
108	if (XFS_IS_REALTIME_INODE(ip))
109		return -ENXIO;
110
111	/*
112	 * The pNFS block layout spec actually supports reflink like
113	 * functionality, but the Linux pNFS server doesn't implement it yet.
114	 */
115	if (xfs_is_reflink_inode(ip))
116		return -ENXIO;
117
118	/*
119	 * Lock out any other I/O before we flush and invalidate the pagecache,
120	 * and then hand out a layout to the remote system.  This is very
121	 * similar to direct I/O, except that the synchronization is much more
122	 * complicated.  See the comment near xfs_break_layouts for a detailed
123	 * explanation.
124	 */
125	xfs_ilock(ip, XFS_IOLOCK_EXCL);
126
127	error = -EINVAL;
128	limit = mp->m_super->s_maxbytes;
129	if (!write)
130		limit = max(limit, round_up(i_size_read(inode),
131				     inode->i_sb->s_blocksize));
132	if (offset > limit)
133		goto out_unlock;
134	if (offset > limit - length)
135		length = limit - offset;
136
137	error = filemap_write_and_wait(inode->i_mapping);
138	if (error)
139		goto out_unlock;
140	error = invalidate_inode_pages2(inode->i_mapping);
141	if (WARN_ON_ONCE(error))
142		return error;
143
144	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + length);
145	offset_fsb = XFS_B_TO_FSBT(mp, offset);
146
147	lock_flags = xfs_ilock_data_map_shared(ip);
148	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
149				&imap, &nimaps, bmapi_flags);
150	xfs_iunlock(ip, lock_flags);
151
152	if (error)
153		goto out_unlock;
154
155	if (write) {
156		enum xfs_prealloc_flags	flags = 0;
157
158		ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
159
160		if (!nimaps || imap.br_startblock == HOLESTARTBLOCK) {
161			/*
162			 * xfs_iomap_write_direct() expects to take ownership of
163			 * the shared ilock.
164			 */
165			xfs_ilock(ip, XFS_ILOCK_SHARED);
166			error = xfs_iomap_write_direct(ip, offset, length,
167						       &imap, nimaps);
168			if (error)
169				goto out_unlock;
170
171			/*
172			 * Ensure the next transaction is committed
173			 * synchronously so that the blocks allocated and
174			 * handed out to the client are guaranteed to be
175			 * present even after a server crash.
176			 */
177			flags |= XFS_PREALLOC_SET | XFS_PREALLOC_SYNC;
178		}
179
180		error = xfs_update_prealloc_flags(ip, flags);
181		if (error)
182			goto out_unlock;
183	}
184	xfs_iunlock(ip, XFS_IOLOCK_EXCL);
185
186	xfs_bmbt_to_iomap(ip, iomap, &imap);
187	*device_generation = mp->m_generation;
188	return error;
189out_unlock:
190	xfs_iunlock(ip, XFS_IOLOCK_EXCL);
191	return error;
192}
193
194/*
195 * Ensure the size update falls into a valid allocated block.
196 */
197static int
198xfs_pnfs_validate_isize(
199	struct xfs_inode	*ip,
200	xfs_off_t		isize)
201{
202	struct xfs_bmbt_irec	imap;
203	int			nimaps = 1;
204	int			error = 0;
205
206	xfs_ilock(ip, XFS_ILOCK_SHARED);
207	error = xfs_bmapi_read(ip, XFS_B_TO_FSBT(ip->i_mount, isize - 1), 1,
208				&imap, &nimaps, 0);
209	xfs_iunlock(ip, XFS_ILOCK_SHARED);
210	if (error)
211		return error;
212
213	if (imap.br_startblock == HOLESTARTBLOCK ||
214	    imap.br_startblock == DELAYSTARTBLOCK ||
215	    imap.br_state == XFS_EXT_UNWRITTEN)
216		return -EIO;
217	return 0;
218}
219
220/*
221 * Make sure the blocks described by maps are stable on disk.  This includes
222 * converting any unwritten extents, flushing the disk cache and updating the
223 * time stamps.
224 *
225 * Note that we rely on the caller to always send us a timestamp update so that
226 * we always commit a transaction here.  If that stops being true we will have
227 * to manually flush the cache here similar to what the fsync code path does
228 * for datasyncs on files that have no dirty metadata.
229 */
230int
231xfs_fs_commit_blocks(
232	struct inode		*inode,
233	struct iomap		*maps,
234	int			nr_maps,
235	struct iattr		*iattr)
236{
237	struct xfs_inode	*ip = XFS_I(inode);
238	struct xfs_mount	*mp = ip->i_mount;
239	struct xfs_trans	*tp;
240	bool			update_isize = false;
241	int			error, i;
242	loff_t			size;
243
244	ASSERT(iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME));
245
246	xfs_ilock(ip, XFS_IOLOCK_EXCL);
247
248	size = i_size_read(inode);
249	if ((iattr->ia_valid & ATTR_SIZE) && iattr->ia_size > size) {
250		update_isize = true;
251		size = iattr->ia_size;
252	}
253
254	for (i = 0; i < nr_maps; i++) {
255		u64 start, length, end;
256
257		start = maps[i].offset;
258		if (start > size)
259			continue;
260
261		end = start + maps[i].length;
262		if (end > size)
263			end = size;
264
265		length = end - start;
266		if (!length)
267			continue;
268	
269		/*
270		 * Make sure reads through the pagecache see the new data.
271		 */
272		error = invalidate_inode_pages2_range(inode->i_mapping,
273					start >> PAGE_SHIFT,
274					(end - 1) >> PAGE_SHIFT);
275		WARN_ON_ONCE(error);
276
277		error = xfs_iomap_write_unwritten(ip, start, length);
278		if (error)
279			goto out_drop_iolock;
280	}
281
282	if (update_isize) {
283		error = xfs_pnfs_validate_isize(ip, size);
284		if (error)
285			goto out_drop_iolock;
286	}
287
288	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
289	if (error)
290		goto out_drop_iolock;
291
292	xfs_ilock(ip, XFS_ILOCK_EXCL);
293	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
294	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
295
296	xfs_setattr_time(ip, iattr);
297	if (update_isize) {
298		i_size_write(inode, iattr->ia_size);
299		ip->i_d.di_size = iattr->ia_size;
300	}
301
302	xfs_trans_set_sync(tp);
303	error = xfs_trans_commit(tp);
304
305out_drop_iolock:
306	xfs_iunlock(ip, XFS_IOLOCK_EXCL);
307	return error;
308}