1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2014 Christoph Hellwig.
6 #include "xfs_shared.h"
7 #include "xfs_format.h"
8 #include "xfs_log_format.h"
9 #include "xfs_trans_resv.h"
10 #include "xfs_mount.h"
11 #include "xfs_inode.h"
12 #include "xfs_trans.h"
14 #include "xfs_iomap.h"
18 * Ensure that we do not have any outstanding pNFS layouts that can be used by
19 * clients to directly read from or write to this inode. This must be called
20 * before every operation that can remove blocks from the extent map.
21 * Additionally we call it during the write operation, where aren't concerned
22 * about exposing unallocated blocks but just want to provide basic
23 * synchronization between a local writer and pNFS clients. mmap writes would
24 * also benefit from this sort of synchronization, but due to the tricky locking
25 * rules in the page fault path we don't bother.
28 xfs_break_leased_layouts(
33 struct xfs_inode *ip = XFS_I(inode);
36 while ((error = break_layout(inode, false)) == -EWOULDBLOCK) {
37 xfs_iunlock(ip, *iolock);
39 error = break_layout(inode, true);
40 *iolock &= ~XFS_IOLOCK_SHARED;
41 *iolock |= XFS_IOLOCK_EXCL;
42 xfs_ilock(ip, *iolock);
49 * Get a unique ID including its location so that the client can identify
50 * the exported device.
54 struct super_block *sb,
59 struct xfs_mount *mp = XFS_M(sb);
62 "Using experimental pNFS feature, use at your own risk!");
64 if (*len < sizeof(uuid_t))
67 memcpy(buf, &mp->m_sb.sb_uuid, sizeof(uuid_t));
68 *len = sizeof(uuid_t);
69 *offset = offsetof(struct xfs_dsb, sb_uuid);
74 * We cannot use file based VFS helpers such as file_modified() to update
75 * inode state as we modify the data/metadata in the inode here. Hence we have
76 * to open code the timestamp updates and SUID/SGID stripping. We also need
77 * to set the inode prealloc flag to ensure that the extents we allocate are not
78 * removed if the inode is reclaimed from memory before xfs_fs_block_commit()
79 * is from the client to indicate that data has been written and the file size
83 xfs_fs_map_update_inode(
89 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid,
94 xfs_ilock(ip, XFS_ILOCK_EXCL);
95 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
97 VFS_I(ip)->i_mode &= ~S_ISUID;
98 if (VFS_I(ip)->i_mode & S_IXGRP)
99 VFS_I(ip)->i_mode &= ~S_ISGID;
100 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
101 ip->i_diflags |= XFS_DIFLAG_PREALLOC;
103 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
104 return xfs_trans_commit(tp);
108 * Get a layout for the pNFS client.
117 u32 *device_generation)
119 struct xfs_inode *ip = XFS_I(inode);
120 struct xfs_mount *mp = ip->i_mount;
121 struct xfs_bmbt_irec imap;
122 xfs_fileoff_t offset_fsb, end_fsb;
124 int bmapi_flags = XFS_BMAPI_ENTIRE;
130 if (xfs_is_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 * The pNFS block layout spec actually supports reflink like
143 * functionality, but the Linux pNFS server doesn't implement it yet.
145 if (xfs_is_reflink_inode(ip))
149 * Lock out any other I/O before we flush and invalidate the pagecache,
150 * and then hand out a layout to the remote system. This is very
151 * similar to direct I/O, except that the synchronization is much more
152 * complicated. See the comment near xfs_break_leased_layouts
153 * for a detailed explanation.
155 xfs_ilock(ip, XFS_IOLOCK_EXCL);
158 limit = mp->m_super->s_maxbytes;
160 limit = max(limit, round_up(i_size_read(inode),
161 inode->i_sb->s_blocksize));
164 if (offset > limit - length)
165 length = limit - offset;
167 error = filemap_write_and_wait(inode->i_mapping);
170 error = invalidate_inode_pages2(inode->i_mapping);
171 if (WARN_ON_ONCE(error))
174 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + length);
175 offset_fsb = XFS_B_TO_FSBT(mp, offset);
177 lock_flags = xfs_ilock_data_map_shared(ip);
178 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
179 &imap, &nimaps, bmapi_flags);
180 seq = xfs_iomap_inode_sequence(ip, 0);
182 ASSERT(!nimaps || imap.br_startblock != DELAYSTARTBLOCK);
184 if (!error && write &&
185 (!nimaps || imap.br_startblock == HOLESTARTBLOCK)) {
186 if (offset + length > XFS_ISIZE(ip))
187 end_fsb = xfs_iomap_eof_align_last_fsb(ip, end_fsb);
188 else if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
189 end_fsb = min(end_fsb, imap.br_startoff +
191 xfs_iunlock(ip, lock_flags);
193 error = xfs_iomap_write_direct(ip, offset_fsb,
194 end_fsb - offset_fsb, 0, &imap, &seq);
199 * Ensure the next transaction is committed synchronously so
200 * that the blocks allocated and handed out to the client are
201 * guaranteed to be present even after a server crash.
203 error = xfs_fs_map_update_inode(ip);
205 error = xfs_log_force_inode(ip);
210 xfs_iunlock(ip, lock_flags);
212 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
214 error = xfs_bmbt_to_iomap(ip, iomap, &imap, 0, 0, seq);
215 *device_generation = mp->m_generation;
218 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
223 * Ensure the size update falls into a valid allocated block.
226 xfs_pnfs_validate_isize(
227 struct xfs_inode *ip,
230 struct xfs_bmbt_irec imap;
234 xfs_ilock(ip, XFS_ILOCK_SHARED);
235 error = xfs_bmapi_read(ip, XFS_B_TO_FSBT(ip->i_mount, isize - 1), 1,
237 xfs_iunlock(ip, XFS_ILOCK_SHARED);
241 if (imap.br_startblock == HOLESTARTBLOCK ||
242 imap.br_startblock == DELAYSTARTBLOCK ||
243 imap.br_state == XFS_EXT_UNWRITTEN)
249 * Make sure the blocks described by maps are stable on disk. This includes
250 * converting any unwritten extents, flushing the disk cache and updating the
253 * Note that we rely on the caller to always send us a timestamp update so that
254 * we always commit a transaction here. If that stops being true we will have
255 * to manually flush the cache here similar to what the fsync code path does
256 * for datasyncs on files that have no dirty metadata.
259 xfs_fs_commit_blocks(
265 struct xfs_inode *ip = XFS_I(inode);
266 struct xfs_mount *mp = ip->i_mount;
267 struct xfs_trans *tp;
268 bool update_isize = false;
272 ASSERT(iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME));
274 xfs_ilock(ip, XFS_IOLOCK_EXCL);
276 size = i_size_read(inode);
277 if ((iattr->ia_valid & ATTR_SIZE) && iattr->ia_size > size) {
279 size = iattr->ia_size;
282 for (i = 0; i < nr_maps; i++) {
283 u64 start, length, end;
285 start = maps[i].offset;
289 end = start + maps[i].length;
293 length = end - start;
298 * Make sure reads through the pagecache see the new data.
300 error = invalidate_inode_pages2_range(inode->i_mapping,
302 (end - 1) >> PAGE_SHIFT);
305 error = xfs_iomap_write_unwritten(ip, start, length, false);
307 goto out_drop_iolock;
311 error = xfs_pnfs_validate_isize(ip, size);
313 goto out_drop_iolock;
316 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
318 goto out_drop_iolock;
320 xfs_ilock(ip, XFS_ILOCK_EXCL);
321 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
322 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
324 ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID)));
325 setattr_copy(&nop_mnt_idmap, inode, iattr);
327 i_size_write(inode, iattr->ia_size);
328 ip->i_disk_size = iattr->ia_size;
331 xfs_trans_set_sync(tp);
332 error = xfs_trans_commit(tp);
335 xfs_iunlock(ip, XFS_IOLOCK_EXCL);