Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge tag 'nfs-for-6.17-2' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
Pull NFS client fix from Trond Myklebust:
- NFS: Fix a data corrupting race when updating an existing write
* tag 'nfs-for-6.17-2' of git://git.linux-nfs.org/projects/trondmy/linux-nfs:
NFS: Fix a race when updating an existing write
+16
-23
+5
-4
fs/nfs/pagelist.c
+5
-4
fs/nfs/pagelist.c
···
253
253
nfs_page_clear_headlock(req);
254
254
}
255
255
256
-
/*
257
-
* nfs_page_group_sync_on_bit_locked
256
+
/**
257
+
* nfs_page_group_sync_on_bit_locked - Test if all requests have @bit set
258
+
* @req: request in page group
259
+
* @bit: PG_* bit that is used to sync page group
258
260
*
259
261
* must be called with page group lock held
260
262
*/
261
-
static bool
262
-
nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
263
+
bool nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
263
264
{
264
265
struct nfs_page *head = req->wb_head;
265
266
struct nfs_page *tmp;
+10
-19
fs/nfs/write.c
+10
-19
fs/nfs/write.c
···
153
153
}
154
154
}
155
155
156
-
static int
157
-
nfs_cancel_remove_inode(struct nfs_page *req, struct inode *inode)
156
+
static void nfs_cancel_remove_inode(struct nfs_page *req, struct inode *inode)
158
157
{
159
-
int ret;
160
-
161
-
if (!test_bit(PG_REMOVE, &req->wb_flags))
162
-
return 0;
163
-
ret = nfs_page_group_lock(req);
164
-
if (ret)
165
-
return ret;
166
158
if (test_and_clear_bit(PG_REMOVE, &req->wb_flags))
167
159
nfs_page_set_inode_ref(req, inode);
168
-
nfs_page_group_unlock(req);
169
-
return 0;
170
160
}
171
161
172
162
/**
···
575
585
}
576
586
}
577
587
588
+
ret = nfs_page_group_lock(head);
589
+
if (ret < 0)
590
+
goto out_unlock;
591
+
578
592
/* Ensure that nobody removed the request before we locked it */
579
593
if (head != folio->private) {
594
+
nfs_page_group_unlock(head);
580
595
nfs_unlock_and_release_request(head);
581
596
goto retry;
582
597
}
583
598
584
-
ret = nfs_cancel_remove_inode(head, inode);
585
-
if (ret < 0)
586
-
goto out_unlock;
587
-
588
-
ret = nfs_page_group_lock(head);
589
-
if (ret < 0)
590
-
goto out_unlock;
599
+
nfs_cancel_remove_inode(head, inode);
591
600
592
601
/* lock each request in the page group */
593
602
for (subreq = head->wb_this_page;
···
775
786
{
776
787
struct nfs_inode *nfsi = NFS_I(nfs_page_to_inode(req));
777
788
778
-
if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) {
789
+
nfs_page_group_lock(req);
790
+
if (nfs_page_group_sync_on_bit_locked(req, PG_REMOVE)) {
779
791
struct folio *folio = nfs_page_to_folio(req->wb_head);
780
792
struct address_space *mapping = folio->mapping;
781
793
···
788
798
}
789
799
spin_unlock(&mapping->i_private_lock);
790
800
}
801
+
nfs_page_group_unlock(req);
791
802
792
803
if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) {
793
804
atomic_long_dec(&nfsi->nrequests);
+1
include/linux/nfs_page.h
+1
include/linux/nfs_page.h
···
160
160
extern int nfs_page_group_lock(struct nfs_page *);
161
161
extern void nfs_page_group_unlock(struct nfs_page *);
162
162
extern bool nfs_page_group_sync_on_bit(struct nfs_page *, unsigned int);
163
+
extern bool nfs_page_group_sync_on_bit_locked(struct nfs_page *, unsigned int);
163
164
extern int nfs_page_set_headlock(struct nfs_page *req);
164
165
extern void nfs_page_clear_headlock(struct nfs_page *req);
165
166
extern bool nfs_async_iocounter_wait(struct rpc_task *, struct nfs_lock_context *);