2 * Copyright (c) 2014 Christoph Hellwig.
5 #include "xfs_format.h"
6 #include "xfs_log_format.h"
7 #include "xfs_trans_resv.h"
10 #include "xfs_inode.h"
11 #include "xfs_trans.h"
14 #include "xfs_bmap_util.h"
15 #include "xfs_error.h"
16 #include "xfs_iomap.h"
17 #include "xfs_shared.h"
22 * Ensure that we do not have any outstanding pNFS layouts that can be used by
23 * clients to directly read from or write to this inode. This must be called
24 * before every operation that can remove blocks from the extent map.
25 * Additionally we call it during the write operation, where aren't concerned
26 * about exposing unallocated blocks but just want to provide basic
27 * synchronization between a local writer and pNFS clients. mmap writes would
28 * also benefit from this sort of synchronization, but due to the tricky locking
29 * rules in the page fault path we don't bother.
37 struct xfs_inode *ip = XFS_I(inode);
40 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL));
42 while ((error = break_layout(inode, false) == -EWOULDBLOCK)) {
43 xfs_iunlock(ip, *iolock);
44 if (with_imutex && (*iolock & XFS_IOLOCK_EXCL))
46 error = break_layout(inode, true);
47 *iolock = XFS_IOLOCK_EXCL;
50 xfs_ilock(ip, *iolock);
57 * Get a unique ID including its location so that the client can identify
58 * the exported device.
62 struct super_block *sb,
67 struct xfs_mount *mp = XFS_M(sb);
69 printk_once(KERN_NOTICE
70 "XFS (%s): using experimental pNFS feature, use at your own risk!\n",
73 if (*len < sizeof(uuid_t))
76 memcpy(buf, &mp->m_sb.sb_uuid, sizeof(uuid_t));
77 *len = sizeof(uuid_t);
78 *offset = offsetof(struct xfs_dsb, sb_uuid);
86 struct xfs_bmbt_irec *imap)
88 struct xfs_mount *mp = ip->i_mount;
90 if (imap->br_startblock == HOLESTARTBLOCK) {
91 iomap->blkno = IOMAP_NULL_BLOCK;
92 iomap->type = IOMAP_HOLE;
93 } else if (imap->br_startblock == DELAYSTARTBLOCK) {
94 iomap->blkno = IOMAP_NULL_BLOCK;
95 iomap->type = IOMAP_DELALLOC;
98 XFS_FSB_TO_DADDR(ip->i_mount, imap->br_startblock);
99 if (imap->br_state == XFS_EXT_UNWRITTEN)
100 iomap->type = IOMAP_UNWRITTEN;
102 iomap->type = IOMAP_MAPPED;
104 iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
105 iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
109 * Get a layout for the pNFS client.
118 u32 *device_generation)
120 struct xfs_inode *ip = XFS_I(inode);
121 struct xfs_mount *mp = ip->i_mount;
122 struct xfs_bmbt_irec imap;
123 xfs_fileoff_t offset_fsb, end_fsb;
125 int bmapi_flags = XFS_BMAPI_ENTIRE;
130 if (XFS_FORCED_SHUTDOWN(mp))
134 * We can't export inodes residing on the realtime device. The realtime
135 * device doesn't have a UUID to identify it, so the client has no way
138 if (XFS_IS_REALTIME_INODE(ip))
142 * Lock out any other I/O before we flush and invalidate the pagecache,
143 * and then hand out a layout to the remote system. This is very
144 * similar to direct I/O, except that the synchronization is much more
145 * complicated. See the comment near xfs_break_layouts for a detailed
148 xfs_ilock(ip, XFS_IOLOCK_EXCL);
151 limit = mp->m_super->s_maxbytes;
153 limit = max(limit, round_up(i_size_read(inode),
154 inode->i_sb->s_blocksize));
157 if (offset > limit - length)
158 length = limit - offset;
160 error = filemap_write_and_wait(inode->i_mapping);
163 error = invalidate_inode_pages2(inode->i_mapping);
164 if (WARN_ON_ONCE(error))
167 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + length);
168 offset_fsb = XFS_B_TO_FSBT(mp, offset);
170 lock_flags = xfs_ilock_data_map_shared(ip);
171 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
172 &imap, &nimaps, bmapi_flags);
173 xfs_iunlock(ip, lock_flags);
179 enum xfs_prealloc_flags flags = 0;
181 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
183 if (!nimaps || imap.br_startblock == HOLESTARTBLOCK) {
185 * xfs_iomap_write_direct() expects to take ownership of
188 xfs_ilock(ip, XFS_ILOCK_SHARED);
189 error = xfs_iomap_write_direct(ip, offset, length,
195 * Ensure the next transaction is committed
196 * synchronously so that the blocks allocated and
197 * handed out to the client are guaranteed to be
198 * present even after a server crash.
200 flags |= XFS_PREALLOC_SET | XFS_PREALLOC_SYNC;
203 error = xfs_update_prealloc_flags(ip, flags);
207 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
209 xfs_bmbt_to_iomap(ip, iomap, &imap);
210 *device_generation = mp->m_generation;
213 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
218 * Ensure the size update falls into a valid allocated block.
221 xfs_pnfs_validate_isize(
222 struct xfs_inode *ip,
225 struct xfs_bmbt_irec imap;
229 xfs_ilock(ip, XFS_ILOCK_SHARED);
230 error = xfs_bmapi_read(ip, XFS_B_TO_FSBT(ip->i_mount, isize - 1), 1,
232 xfs_iunlock(ip, XFS_ILOCK_SHARED);
236 if (imap.br_startblock == HOLESTARTBLOCK ||
237 imap.br_startblock == DELAYSTARTBLOCK ||
238 imap.br_state == XFS_EXT_UNWRITTEN)
244 * Make sure the blocks described by maps are stable on disk. This includes
245 * converting any unwritten extents, flushing the disk cache and updating the
248 * Note that we rely on the caller to always send us a timestamp update so that
249 * we always commit a transaction here. If that stops being true we will have
250 * to manually flush the cache here similar to what the fsync code path does
251 * for datasyncs on files that have no dirty metadata.
254 xfs_fs_commit_blocks(
260 struct xfs_inode *ip = XFS_I(inode);
261 struct xfs_mount *mp = ip->i_mount;
262 struct xfs_trans *tp;
263 bool update_isize = false;
267 ASSERT(iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME));
269 xfs_ilock(ip, XFS_IOLOCK_EXCL);
271 size = i_size_read(inode);
272 if ((iattr->ia_valid & ATTR_SIZE) && iattr->ia_size > size) {
274 size = iattr->ia_size;
277 for (i = 0; i < nr_maps; i++) {
278 u64 start, length, end;
280 start = maps[i].offset;
284 end = start + maps[i].length;
288 length = end - start;
293 * Make sure reads through the pagecache see the new data.
295 error = invalidate_inode_pages2_range(inode->i_mapping,
297 (end - 1) >> PAGE_SHIFT);
300 error = xfs_iomap_write_unwritten(ip, start, length);
302 goto out_drop_iolock;
306 error = xfs_pnfs_validate_isize(ip, size);
308 goto out_drop_iolock;
311 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
312 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
314 xfs_trans_cancel(tp);
315 goto out_drop_iolock;
318 xfs_ilock(ip, XFS_ILOCK_EXCL);
319 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
320 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
322 xfs_setattr_time(ip, iattr);
324 i_size_write(inode, iattr->ia_size);
325 ip->i_d.di_size = iattr->ia_size;
328 xfs_trans_set_sync(tp);
329 error = xfs_trans_commit(tp);
332 xfs_iunlock(ip, XFS_IOLOCK_EXCL);