1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2014 Christoph Hellwig.
5 #include <linux/iomap.h>
7 #include "xfs_shared.h"
8 #include "xfs_format.h"
9 #include "xfs_log_format.h"
10 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_inode.h"
14 #include "xfs_trans.h"
17 #include "xfs_bmap_util.h"
18 #include "xfs_error.h"
19 #include "xfs_iomap.h"
20 #include "xfs_shared.h"
25 * Ensure that we do not have any outstanding pNFS layouts that can be used by
26 * clients to directly read from or write to this inode. This must be called
27 * before every operation that can remove blocks from the extent map.
28 * Additionally we call it during the write operation, where aren't concerned
29 * about exposing unallocated blocks but just want to provide basic
30 * synchronization between a local writer and pNFS clients. mmap writes would
31 * also benefit from this sort of synchronization, but due to the tricky locking
32 * rules in the page fault path we don't bother.
35 xfs_break_leased_layouts(
40 struct xfs_inode *ip = XFS_I(inode);
43 while ((error = break_layout(inode, false) == -EWOULDBLOCK)) {
44 xfs_iunlock(ip, *iolock);
46 error = break_layout(inode, true);
47 *iolock &= ~XFS_IOLOCK_SHARED;
48 *iolock |= XFS_IOLOCK_EXCL;
49 xfs_ilock(ip, *iolock);
56 * Get a unique ID including its location so that the client can identify
57 * the exported device.
61 struct super_block *sb,
66 struct xfs_mount *mp = XFS_M(sb);
68 printk_once(KERN_NOTICE
69 "XFS (%s): using experimental pNFS feature, use at your own risk!\n",
72 if (*len < sizeof(uuid_t))
75 memcpy(buf, &mp->m_sb.sb_uuid, sizeof(uuid_t));
76 *len = sizeof(uuid_t);
77 *offset = offsetof(struct xfs_dsb, sb_uuid);
82 * Get a layout for the pNFS client.
91 u32 *device_generation)
93 struct xfs_inode *ip = XFS_I(inode);
94 struct xfs_mount *mp = ip->i_mount;
95 struct xfs_bmbt_irec imap;
96 xfs_fileoff_t offset_fsb, end_fsb;
98 int bmapi_flags = XFS_BMAPI_ENTIRE;
103 if (XFS_FORCED_SHUTDOWN(mp))
107 * We can't export inodes residing on the realtime device. The realtime
108 * device doesn't have a UUID to identify it, so the client has no way
111 if (XFS_IS_REALTIME_INODE(ip))
115 * The pNFS block layout spec actually supports reflink like
116 * functionality, but the Linux pNFS server doesn't implement it yet.
118 if (xfs_is_reflink_inode(ip))
122 * Lock out any other I/O before we flush and invalidate the pagecache,
123 * and then hand out a layout to the remote system. This is very
124 * similar to direct I/O, except that the synchronization is much more
125 * complicated. See the comment near xfs_break_leased_layouts
126 * for a detailed explanation.
128 xfs_ilock(ip, XFS_IOLOCK_EXCL);
131 limit = mp->m_super->s_maxbytes;
133 limit = max(limit, round_up(i_size_read(inode),
134 inode->i_sb->s_blocksize));
137 if (offset > limit - length)
138 length = limit - offset;
140 error = filemap_write_and_wait(inode->i_mapping);
143 error = invalidate_inode_pages2(inode->i_mapping);
144 if (WARN_ON_ONCE(error))
147 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + length);
148 offset_fsb = XFS_B_TO_FSBT(mp, offset);
150 lock_flags = xfs_ilock_data_map_shared(ip);
151 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
152 &imap, &nimaps, bmapi_flags);
153 xfs_iunlock(ip, lock_flags);
159 enum xfs_prealloc_flags flags = 0;
161 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
163 if (!nimaps || imap.br_startblock == HOLESTARTBLOCK) {
165 * xfs_iomap_write_direct() expects to take ownership of
168 xfs_ilock(ip, XFS_ILOCK_SHARED);
169 error = xfs_iomap_write_direct(ip, offset, length,
175 * Ensure the next transaction is committed
176 * synchronously so that the blocks allocated and
177 * handed out to the client are guaranteed to be
178 * present even after a server crash.
180 flags |= XFS_PREALLOC_SET | XFS_PREALLOC_SYNC;
183 error = xfs_update_prealloc_flags(ip, flags);
187 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
189 error = xfs_bmbt_to_iomap(ip, iomap, &imap, false);
190 *device_generation = mp->m_generation;
193 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
198 * Ensure the size update falls into a valid allocated block.
201 xfs_pnfs_validate_isize(
202 struct xfs_inode *ip,
205 struct xfs_bmbt_irec imap;
209 xfs_ilock(ip, XFS_ILOCK_SHARED);
210 error = xfs_bmapi_read(ip, XFS_B_TO_FSBT(ip->i_mount, isize - 1), 1,
212 xfs_iunlock(ip, XFS_ILOCK_SHARED);
216 if (imap.br_startblock == HOLESTARTBLOCK ||
217 imap.br_startblock == DELAYSTARTBLOCK ||
218 imap.br_state == XFS_EXT_UNWRITTEN)
224 * Make sure the blocks described by maps are stable on disk. This includes
225 * converting any unwritten extents, flushing the disk cache and updating the
228 * Note that we rely on the caller to always send us a timestamp update so that
229 * we always commit a transaction here. If that stops being true we will have
230 * to manually flush the cache here similar to what the fsync code path does
231 * for datasyncs on files that have no dirty metadata.
234 xfs_fs_commit_blocks(
240 struct xfs_inode *ip = XFS_I(inode);
241 struct xfs_mount *mp = ip->i_mount;
242 struct xfs_trans *tp;
243 bool update_isize = false;
247 ASSERT(iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME));
249 xfs_ilock(ip, XFS_IOLOCK_EXCL);
251 size = i_size_read(inode);
252 if ((iattr->ia_valid & ATTR_SIZE) && iattr->ia_size > size) {
254 size = iattr->ia_size;
257 for (i = 0; i < nr_maps; i++) {
258 u64 start, length, end;
260 start = maps[i].offset;
264 end = start + maps[i].length;
268 length = end - start;
273 * Make sure reads through the pagecache see the new data.
275 error = invalidate_inode_pages2_range(inode->i_mapping,
277 (end - 1) >> PAGE_SHIFT);
280 error = xfs_iomap_write_unwritten(ip, start, length, false);
282 goto out_drop_iolock;
286 error = xfs_pnfs_validate_isize(ip, size);
288 goto out_drop_iolock;
291 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
293 goto out_drop_iolock;
295 xfs_ilock(ip, XFS_ILOCK_EXCL);
296 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
297 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
299 xfs_setattr_time(ip, iattr);
301 i_size_write(inode, iattr->ia_size);
302 ip->i_d.di_size = iattr->ia_size;
305 xfs_trans_set_sync(tp);
306 error = xfs_trans_commit(tp);
309 xfs_iunlock(ip, XFS_IOLOCK_EXCL);