xref: /openbmc/linux/fs/xfs/xfs_pnfs.c (revision e2c75e76)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2014 Christoph Hellwig.
4  */
5 #include <linux/iomap.h>
6 #include "xfs.h"
7 #include "xfs_format.h"
8 #include "xfs_log_format.h"
9 #include "xfs_trans_resv.h"
10 #include "xfs_sb.h"
11 #include "xfs_mount.h"
12 #include "xfs_inode.h"
13 #include "xfs_trans.h"
14 #include "xfs_log.h"
15 #include "xfs_bmap.h"
16 #include "xfs_bmap_util.h"
17 #include "xfs_error.h"
18 #include "xfs_iomap.h"
19 #include "xfs_shared.h"
20 #include "xfs_bit.h"
21 #include "xfs_pnfs.h"
22 
23 /*
24  * Ensure that we do not have any outstanding pNFS layouts that can be used by
25  * clients to directly read from or write to this inode.  This must be called
26  * before every operation that can remove blocks from the extent map.
27  * Additionally we call it during the write operation, where aren't concerned
28  * about exposing unallocated blocks but just want to provide basic
29  * synchronization between a local writer and pNFS clients.  mmap writes would
30  * also benefit from this sort of synchronization, but due to the tricky locking
31  * rules in the page fault path we don't bother.
32  */
33 int
34 xfs_break_layouts(
35 	struct inode		*inode,
36 	uint			*iolock)
37 {
38 	struct xfs_inode	*ip = XFS_I(inode);
39 	int			error;
40 
41 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL));
42 
43 	while ((error = break_layout(inode, false) == -EWOULDBLOCK)) {
44 		xfs_iunlock(ip, *iolock);
45 		error = break_layout(inode, true);
46 		*iolock = XFS_IOLOCK_EXCL;
47 		xfs_ilock(ip, *iolock);
48 	}
49 
50 	return error;
51 }
52 
53 /*
54  * Get a unique ID including its location so that the client can identify
55  * the exported device.
56  */
57 int
58 xfs_fs_get_uuid(
59 	struct super_block	*sb,
60 	u8			*buf,
61 	u32			*len,
62 	u64			*offset)
63 {
64 	struct xfs_mount	*mp = XFS_M(sb);
65 
66 	printk_once(KERN_NOTICE
67 "XFS (%s): using experimental pNFS feature, use at your own risk!\n",
68 		mp->m_fsname);
69 
70 	if (*len < sizeof(uuid_t))
71 		return -EINVAL;
72 
73 	memcpy(buf, &mp->m_sb.sb_uuid, sizeof(uuid_t));
74 	*len = sizeof(uuid_t);
75 	*offset = offsetof(struct xfs_dsb, sb_uuid);
76 	return 0;
77 }
78 
79 /*
80  * Get a layout for the pNFS client.
81  */
82 int
83 xfs_fs_map_blocks(
84 	struct inode		*inode,
85 	loff_t			offset,
86 	u64			length,
87 	struct iomap		*iomap,
88 	bool			write,
89 	u32			*device_generation)
90 {
91 	struct xfs_inode	*ip = XFS_I(inode);
92 	struct xfs_mount	*mp = ip->i_mount;
93 	struct xfs_bmbt_irec	imap;
94 	xfs_fileoff_t		offset_fsb, end_fsb;
95 	loff_t			limit;
96 	int			bmapi_flags = XFS_BMAPI_ENTIRE;
97 	int			nimaps = 1;
98 	uint			lock_flags;
99 	int			error = 0;
100 
101 	if (XFS_FORCED_SHUTDOWN(mp))
102 		return -EIO;
103 
104 	/*
105 	 * We can't export inodes residing on the realtime device.  The realtime
106 	 * device doesn't have a UUID to identify it, so the client has no way
107 	 * to find it.
108 	 */
109 	if (XFS_IS_REALTIME_INODE(ip))
110 		return -ENXIO;
111 
112 	/*
113 	 * The pNFS block layout spec actually supports reflink like
114 	 * functionality, but the Linux pNFS server doesn't implement it yet.
115 	 */
116 	if (xfs_is_reflink_inode(ip))
117 		return -ENXIO;
118 
119 	/*
120 	 * Lock out any other I/O before we flush and invalidate the pagecache,
121 	 * and then hand out a layout to the remote system.  This is very
122 	 * similar to direct I/O, except that the synchronization is much more
123 	 * complicated.  See the comment near xfs_break_layouts for a detailed
124 	 * explanation.
125 	 */
126 	xfs_ilock(ip, XFS_IOLOCK_EXCL);
127 
128 	error = -EINVAL;
129 	limit = mp->m_super->s_maxbytes;
130 	if (!write)
131 		limit = max(limit, round_up(i_size_read(inode),
132 				     inode->i_sb->s_blocksize));
133 	if (offset > limit)
134 		goto out_unlock;
135 	if (offset > limit - length)
136 		length = limit - offset;
137 
138 	error = filemap_write_and_wait(inode->i_mapping);
139 	if (error)
140 		goto out_unlock;
141 	error = invalidate_inode_pages2(inode->i_mapping);
142 	if (WARN_ON_ONCE(error))
143 		return error;
144 
145 	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + length);
146 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
147 
148 	lock_flags = xfs_ilock_data_map_shared(ip);
149 	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
150 				&imap, &nimaps, bmapi_flags);
151 	xfs_iunlock(ip, lock_flags);
152 
153 	if (error)
154 		goto out_unlock;
155 
156 	if (write) {
157 		enum xfs_prealloc_flags	flags = 0;
158 
159 		ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
160 
161 		if (!nimaps || imap.br_startblock == HOLESTARTBLOCK) {
162 			/*
163 			 * xfs_iomap_write_direct() expects to take ownership of
164 			 * the shared ilock.
165 			 */
166 			xfs_ilock(ip, XFS_ILOCK_SHARED);
167 			error = xfs_iomap_write_direct(ip, offset, length,
168 						       &imap, nimaps);
169 			if (error)
170 				goto out_unlock;
171 
172 			/*
173 			 * Ensure the next transaction is committed
174 			 * synchronously so that the blocks allocated and
175 			 * handed out to the client are guaranteed to be
176 			 * present even after a server crash.
177 			 */
178 			flags |= XFS_PREALLOC_SET | XFS_PREALLOC_SYNC;
179 		}
180 
181 		error = xfs_update_prealloc_flags(ip, flags);
182 		if (error)
183 			goto out_unlock;
184 	}
185 	xfs_iunlock(ip, XFS_IOLOCK_EXCL);
186 
187 	xfs_bmbt_to_iomap(ip, iomap, &imap);
188 	*device_generation = mp->m_generation;
189 	return error;
190 out_unlock:
191 	xfs_iunlock(ip, XFS_IOLOCK_EXCL);
192 	return error;
193 }
194 
195 /*
196  * Ensure the size update falls into a valid allocated block.
197  */
198 static int
199 xfs_pnfs_validate_isize(
200 	struct xfs_inode	*ip,
201 	xfs_off_t		isize)
202 {
203 	struct xfs_bmbt_irec	imap;
204 	int			nimaps = 1;
205 	int			error = 0;
206 
207 	xfs_ilock(ip, XFS_ILOCK_SHARED);
208 	error = xfs_bmapi_read(ip, XFS_B_TO_FSBT(ip->i_mount, isize - 1), 1,
209 				&imap, &nimaps, 0);
210 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
211 	if (error)
212 		return error;
213 
214 	if (imap.br_startblock == HOLESTARTBLOCK ||
215 	    imap.br_startblock == DELAYSTARTBLOCK ||
216 	    imap.br_state == XFS_EXT_UNWRITTEN)
217 		return -EIO;
218 	return 0;
219 }
220 
221 /*
222  * Make sure the blocks described by maps are stable on disk.  This includes
223  * converting any unwritten extents, flushing the disk cache and updating the
224  * time stamps.
225  *
226  * Note that we rely on the caller to always send us a timestamp update so that
227  * we always commit a transaction here.  If that stops being true we will have
228  * to manually flush the cache here similar to what the fsync code path does
229  * for datasyncs on files that have no dirty metadata.
230  */
231 int
232 xfs_fs_commit_blocks(
233 	struct inode		*inode,
234 	struct iomap		*maps,
235 	int			nr_maps,
236 	struct iattr		*iattr)
237 {
238 	struct xfs_inode	*ip = XFS_I(inode);
239 	struct xfs_mount	*mp = ip->i_mount;
240 	struct xfs_trans	*tp;
241 	bool			update_isize = false;
242 	int			error, i;
243 	loff_t			size;
244 
245 	ASSERT(iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME));
246 
247 	xfs_ilock(ip, XFS_IOLOCK_EXCL);
248 
249 	size = i_size_read(inode);
250 	if ((iattr->ia_valid & ATTR_SIZE) && iattr->ia_size > size) {
251 		update_isize = true;
252 		size = iattr->ia_size;
253 	}
254 
255 	for (i = 0; i < nr_maps; i++) {
256 		u64 start, length, end;
257 
258 		start = maps[i].offset;
259 		if (start > size)
260 			continue;
261 
262 		end = start + maps[i].length;
263 		if (end > size)
264 			end = size;
265 
266 		length = end - start;
267 		if (!length)
268 			continue;
269 
270 		/*
271 		 * Make sure reads through the pagecache see the new data.
272 		 */
273 		error = invalidate_inode_pages2_range(inode->i_mapping,
274 					start >> PAGE_SHIFT,
275 					(end - 1) >> PAGE_SHIFT);
276 		WARN_ON_ONCE(error);
277 
278 		error = xfs_iomap_write_unwritten(ip, start, length, false);
279 		if (error)
280 			goto out_drop_iolock;
281 	}
282 
283 	if (update_isize) {
284 		error = xfs_pnfs_validate_isize(ip, size);
285 		if (error)
286 			goto out_drop_iolock;
287 	}
288 
289 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
290 	if (error)
291 		goto out_drop_iolock;
292 
293 	xfs_ilock(ip, XFS_ILOCK_EXCL);
294 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
295 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
296 
297 	xfs_setattr_time(ip, iattr);
298 	if (update_isize) {
299 		i_size_write(inode, iattr->ia_size);
300 		ip->i_d.di_size = iattr->ia_size;
301 	}
302 
303 	xfs_trans_set_sync(tp);
304 	error = xfs_trans_commit(tp);
305 
306 out_drop_iolock:
307 	xfs_iunlock(ip, XFS_IOLOCK_EXCL);
308 	return error;
309 }
310