netfs, cachefiles, 9p: Additional patches
Subsystems (11)
David Howells
maintainercover netfs, cachefiles, 9p: Additional patches
Hi Christian, Jeff, Gao, Dominique,
Here are some additional patches for my netfs-lib tree:
(1) Fix __cachefiles_prepare_write() to correctly validate against the DIO alignment.
(2) 9p: Fix initialisation of the netfs_inode so that i_size is set before netfs_inode_init() is called.
(3) 9p: Do a couple of cleanups (remove a couple of unused vars and turn a BUG_ON() into a warning).
(4) 9p: Always update remote_i_size, even if we're asked not to update i_size in stat2inode.
(5) 9p: Return the amount written in preference to an error if we wrote something.
David
The netfslib postings:
David Howells (5):
cachefiles: Fix __cachefiles_prepare_write()
9p: Fix initialisation of netfs_inode for 9p
9p: Do a couple of cleanups
9p: Always update remote_i_size in stat2inode
9p: Use length of data written to the server in preference to error
fs/9p/v9fs_vfs.h | 1 + fs/9p/vfs_addr.c | 24 ++++++++++++------------ fs/9p/vfs_inode.c | 6 +++--- fs/9p/vfs_inode_dotl.c | 7 ++++--- fs/cachefiles/io.c | 28 +++++++++++++++++----------- 5 files changed, 37 insertions(+), 29 deletions(-)
David Howells
maintainerPATCH 1/5 cachefiles: Fix __cachefiles_prepare_write()
Fix __cachefiles_prepare_write() to correctly determine whether the requested write will fit correctly with the DIO alignment.
--- fs/cachefiles/io.c | 28 +++++++++++++++++----------- 1 file changed, 17 insertions(+), 11 deletions(-)
fs/cachefiles/io.c +17 -11
@@ -522,16 +522,22 @@ int __cachefiles_prepare_write(struct cachefiles_object *object, bool no_space_allocated_yet) {struct cachefiles_cache *cache = object->volume->cache; -loff_t start = *_start, pos; -size_t len = *_len, down; +unsigned long long start = *_start, pos; +size_t len = *_len;int ret;/* Round to DIO size */ - down = start -round_down (start,PAGE_SIZE ); - *_start = start - down; - *_len =round_up (down + len,PAGE_SIZE ); -if (down < start || *_len > upper_len) + start =round_down (*_start,PAGE_SIZE ); +if (start != *_start) { +kleave (" = -ENOBUFS [down]" ); +return -ENOBUFS ; + } +if (*_len > upper_len) { +kleave (" = -ENOBUFS [up]" );return -ENOBUFS ; + } + + *_len =round_up (len,PAGE_SIZE );/* We need to work out whether there's sufficient disk space to perform * the write - but we can skip that check if we have space already
@@ -542,7 +548,7 @@ int __cachefiles_prepare_write(struct cachefiles_object *object, pos =cachefiles_inject_read_error ();if (pos ==0 ) - pos =vfs_llseek (file, *_start,SEEK_DATA ); + pos =vfs_llseek (file, start,SEEK_DATA );if (pos <0 && pos >= (loff_t)-MAX_ERRNO ) {if (pos == -ENXIO ) goto check_space;/* Unallocated tail */
@@ -550,7 +556,7 @@ int __cachefiles_prepare_write(struct cachefiles_object *object, cachefiles_trace_seek_error);return pos; } -if ((u64 )pos >= (u64)*_start + *_len) +if (pos >= start + *_len) goto check_space;/* Unallocated region */ /* We have a block that's at least partially filled - if we're low on
@@ -563,13 +569,13 @@ int __cachefiles_prepare_write(struct cachefiles_object *object, pos =cachefiles_inject_read_error ();if (pos ==0 ) - pos =vfs_llseek (file, *_start,SEEK_HOLE ); + pos =vfs_llseek (file, start,SEEK_HOLE );if (pos <0 && pos >= (loff_t)-MAX_ERRNO ) {trace_cachefiles_io_error (object,file_inode (file), pos, cachefiles_trace_seek_error);return pos; } -if ((u64 )pos >= (u64)*_start + *_len) +if (pos >= start + *_len)return 0 ;/* Fully allocated */ /* Partially allocated, but insufficient space: cull. */
@@ -577,7 +583,7 @@ int __cachefiles_prepare_write(struct cachefiles_object *object, ret =cachefiles_inject_remove_error ();if (ret ==0 ) ret =vfs_fallocate (file,FALLOC_FL_PUNCH_HOLE |FALLOC_FL_KEEP_SIZE , - *_start, *_len); + start, *_len);if (ret <0 ) {trace_cachefiles_io_error (object,file_inode (file), ret, cachefiles_trace_fallocate_error);
Simon Horman
reviewerRe: [PATCH 1/5] cachefiles: Fix __cachefiles_prepare_write()
Quoted 50 lines · David Howells
Fix
__cachefiles_prepare_write()to correctly determine whether the requested write will fit correctly with the DIO alignment.Reported-by Gao Xiang <[email protected]>Signed-off-by David Howells <[email protected]>Tested-by Yiqun Leng <[email protected]>Tested-by Jia Zhu <[email protected]>cc Jeff Layton <[email protected]>--- fs/cachefiles/io.c | 28 +++++++++++++++++----------- 1 file changed, 17 insertions(+), 11 deletions(-)fs/cachefiles/io.c +12
-6@@ -522,16 +522,22 @@ int __cachefiles_prepare_write(struct cachefiles_object *object, bool no_space_allocated_yet) {struct cachefiles_cache *cache = object->volume->cache; -loff_t start = *_start, pos; -size_t len = *_len, down; +unsigned long long start = *_start, pos; +size_t len = *_len;int ret;/* Round to DIO size */ - down = start -round_down (start,PAGE_SIZE ); - *_start = start - down; - *_len =round_up (down + len,PAGE_SIZE ); -if (down < start || *_len > upper_len) + start =round_down (*_start,PAGE_SIZE ); +if (start != *_start) { +kleave (" = -ENOBUFS [down]" ); +return -ENOBUFS ; + } +if (*_len > upper_len) { +kleave (" = -ENOBUFS [up]" );return -ENOBUFS ; + } + + *_len =round_up (len,PAGE_SIZE );/* We need to work out whether there's sufficient disk space to perform * the write - but we can skip that check if we have space already
fs/cachefiles/io.c int __cachefiles_prepare_write(struct cachefiles_object *object, · line 552
Quoted 6 lines
fs/cachefiles/io.c +1
-1@@ -542,7 +548,7 @@ int __cachefiles_prepare_write(struct cachefiles_object *object, pos =cachefiles_inject_read_error ();if (pos ==0 ) - pos =vfs_llseek (file, *_start,SEEK_DATA ); + pos =vfs_llseek (file, start,SEEK_DATA );if (pos <0 && pos >= (loff_t)-MAX_ERRNO ) {
Hi David,
I realise these patches have been accepted, but I have a minor nit:
pos is now unsigned, and so cannot be less than zero.
Flagged by Smatch and Coccinelle.
Quoted 12 lines
fs/cachefiles/io.c +1
-1if (pos == -ENXIO ) goto check_space;/* Unallocated tail */ @@ -550,7 +556,7 @@ int __cachefiles_prepare_write(struct cachefiles_object *object, cachefiles_trace_seek_error);return pos; } -if ((u64 )pos >= (u64)*_start + *_len) +if (pos >= start + *_len) goto check_space;/* Unallocated region */ /* We have a block that's at least partially filled - if we're low on
fs/cachefiles/io.c int __cachefiles_prepare_write(struct cachefiles_object *object, · line 573
Quoted 6 lines
fs/cachefiles/io.c +1
-1@@ -563,13 +569,13 @@ int __cachefiles_prepare_write(struct cachefiles_object *object, pos =cachefiles_inject_read_error ();if (pos ==0 ) - pos =vfs_llseek (file, *_start,SEEK_HOLE ); + pos =vfs_llseek (file, start,SEEK_HOLE );if (pos <0 && pos >= (loff_t)-MAX_ERRNO ) {
Ditto.
Quoted 19 lines
fs/cachefiles/io.c +2
-2trace_cachefiles_io_error (object,file_inode (file), pos, cachefiles_trace_seek_error);return pos; } -if ((u64 )pos >= (u64)*_start + *_len) +if (pos >= start + *_len)return 0 ;/* Fully allocated */ /* Partially allocated, but insufficient space: cull. */ @@ -577,7 +583,7 @@ int __cachefiles_prepare_write(struct cachefiles_object *object, ret =cachefiles_inject_remove_error ();if (ret ==0 ) ret =vfs_fallocate (file,FALLOC_FL_PUNCH_HOLE |FALLOC_FL_KEEP_SIZE , - *_start, *_len); + start, *_len);if (ret <0 ) {trace_cachefiles_io_error (object,file_inode (file), ret, cachefiles_trace_fallocate_error);
David Howells
maintainerRe: [PATCH 1/5] cachefiles: Fix __cachefiles_prepare_write()
Quoted 2 lines · Simon Horman
I realise these patches have been accepted, but I have a minor nit:
pos is now unsigned, and so cannot be less than zero.
Good point. How about the attached patch. Whilst I would prefer to use unsigned long long to avoid the casts, it might
David
---
cachefiles: Fix signed/unsigned mixup
In __cachefiles_prepare_write(), the start and pos variables were made unsigned 64-bit so that the casts in the checking could be got rid of - which should be fine since absolute file offsets can't be negative, except that an error code may be obtained from vfs_llseek(), which *would* be negative. This breaks the error check.
Fix this for now by reverting pos and start to be signed and putting back the casts. Unfortunately, the error value checks cannot be replaced with IS_ERR_VALUE() as long might be 32-bits.
--- fs/cachefiles/io.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
fs/cachefiles/io.c +3 -3
@@ -522,7 +522,7 @@ int __cachefiles_prepare_write(struct cachefiles_object *object, bool no_space_allocated_yet) {struct cachefiles_cache *cache = object->volume->cache; -unsigned long long start = *_start, pos; +loff_t start = *_start, pos;size_t len = *_len;int ret;
@@ -556,7 +556,7 @@ int __cachefiles_prepare_write(struct cachefiles_object *object, cachefiles_trace_seek_error);return pos; } -if (pos >= start + *_len) +if ((u64 )pos >= (u64 )start + *_len) goto check_space;/* Unallocated region */ /* We have a block that's at least partially filled - if we're low on
@@ -575,7 +575,7 @@ int __cachefiles_prepare_write(struct cachefiles_object *object, cachefiles_trace_seek_error);return pos; } -if (pos >= start + *_len) +if ((u64 )pos >= (u64 )start + *_len)return 0 ;/* Fully allocated */ /* Partially allocated, but insufficient space: cull. */
Re: [PATCH 1/5] cachefiles: Fix __cachefiles_prepare_write()
Quoted 7 lines · David Howells
Quoted 2 lines · Simon Horman
I realise these patches have been accepted, but I have a minor nit:
pos is now unsigned, and so cannot be less than zero.Good point. How about the attached patch. Whilst I would prefer to use unsigned long long to avoid the casts, it might
Hi David,
I would also prefer to avoid casts, but I agree this is a good way forward. Thanks for the quick fix.
David Howells
maintainerPATCH 2/5 9p: Fix initialisation of netfs_inode for 9p
In reply to David HowellsThe 9p filesystem is calling netfs_inode_init() in v9fs_init_inode() - before the struct inode fields have been initialised from the obtained file stats (ie. after v9fs_stat2inode*() has been called), but netfslib wants to set a couple of its fields from i_size.
--- fs/9p/v9fs_vfs.h | 1 + fs/9p/vfs_inode.c | 6 +++--- fs/9p/vfs_inode_dotl.c | 1 + 3 files changed, 5 insertions(+), 3 deletions(-)
fs/9p/v9fs_vfs.h +1 -0
@@ -42,6 +42,7 @@ struct inode *v9fs_alloc_inode(struct super_block *sb);void v9fs_free_inode (struct inode *inode);struct inode *v9fs_get_inode (struct super_block *sb,umode_t mode,dev_t rdev); +void v9fs_set_netfs_context (struct inode *inode);int v9fs_init_inode (struct v9fs_session_info *v9ses,struct inode *inode,umode_t mode,dev_t rdev);void v9fs_evict_inode (struct inode *inode);
fs/9p/vfs_inode.c +3 -3
@@ -246,7 +246,7 @@ void v9fs_free_inode(struct inode *inode)/* * Set parameters for the netfs library */ -static void v9fs_set_netfs_context (struct inode *inode) +void v9fs_set_netfs_context (struct inode *inode) {struct v9fs_inode *v9inode =V9FS_I (inode);netfs_inode_init (&v9inode->netfs, &v9fs_req_ops, true);
@@ -326,8 +326,6 @@ int v9fs_init_inode(struct v9fs_session_info *v9ses, err = -EINVAL ; goto error; } - -v9fs_set_netfs_context (inode); error:return err;
@@ -359,6 +357,7 @@ struct inode *v9fs_get_inode(struct super_block *sb, umode_t mode, dev_t rdev)iput (inode);return ERR_PTR (err); } +v9fs_set_netfs_context (inode);return inode; }
@@ -461,6 +460,7 @@ static struct inode *v9fs_qid_iget(struct super_block *sb, goto error;v9fs_stat2inode (st, inode, sb,0 ); +v9fs_set_netfs_context (inode);v9fs_cache_inode_get_cookie (inode);unlock_new_inode (inode);return inode;
fs/9p/vfs_inode_dotl.c +1 -0
@@ -128,6 +128,7 @@ static struct inode *v9fs_qid_iget_dotl(struct super_block *sb, goto error;v9fs_stat2inode_dotl (st, inode,0 ); +v9fs_set_netfs_context (inode);v9fs_cache_inode_get_cookie (inode); retval =v9fs_get_acl (inode, fid);if (retval)
David Howells
maintainerPATCH 3/5 9p: Do a couple of cleanups
In reply to David HowellsDo a couple of cleanups to 9p:
(1) Remove a couple of unused variables.
(2) Turn a BUG_ON() into a warning, consolidate with another warning and make the warning message include the inode number rather than whatever's in i_private (which will get hashed anyway).
--- fs/9p/vfs_addr.c | 17 +++++++++-------- 1 file changed, 9 insertions(+), 8 deletions(-)
fs/9p/vfs_addr.c +9 -8
@@ -28,8 +28,6 @@static void v9fs_upload_to_server (struct netfs_io_subrequest *subreq) { -struct inode *inode = subreq->rreq->inode; -struct v9fs_inode __maybe_unused *v9inode =V9FS_I (inode);struct p9_fid *fid = subreq->rreq->netfs_priv;int err;
@@ -98,15 +96,13 @@ static int v9fs_init_request(struct netfs_io_request *rreq, struct file *file)if (file) { fid = file->private_data; -BUG_ON (!fid); +if (!fid) + goto no_fid;p9_fid_get (fid); }else { fid =v9fs_fid_find_inode (rreq->inode, writing,INVALID_UID , true); -if (!fid) { -WARN_ONCE (1 ,"folio expected an open fid inode->i_private=%p\n" , - rreq->inode->i_private); -return -EINVAL ; - } +if (!fid) + goto no_fid; }/* we might need to read from a fid that was opened write-only
@@ -115,6 +111,11 @@ static int v9fs_init_request(struct netfs_io_request *rreq, struct file *file)WARN_ON (rreq->origin ==NETFS_READ_FOR_WRITE && !(fid->mode &P9_ORDWR )); rreq->netfs_priv = fid;return 0 ; + +no_fid: +WARN_ONCE (1 ,"folio expected an open fid inode->i_ino=%lx\n" , + rreq->inode->i_ino); +return -EINVAL ; }/**
Re: [PATCH 3/5] 9p: Do a couple of cleanups
David Howells wrote on Wed, Jan 03, 2024 at 02:59:27PM +0000:
Quoted 9 lines
Do a couple of cleanups to 9p:
(1) Remove a couple of unused variables.
(2) Turn a
BUG_ON()into a warning, consolidate with another warning and make the warning message include the inode number rather than whatever's ini_private(which will get hashed anyway).Suggested-by Dominique Martinet <[email protected]>
Thanks,
Signature · 1 line
Dominique Martinet | Asmadeus
David Howells
maintainerPATCH 4/5 9p: Always update remote_i_size in stat2inode
In reply to David HowellsAlways update remote_i_size in v9fs_stat2inode*() if the size is available, even if we are asked not to update i_isize
--- fs/9p/vfs_inode_dotl.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
fs/9p/vfs_inode_dotl.c +3 -3
@@ -684,10 +684,10 @@ v9fs_stat2inode_dotl(struct p9_stat_dotl *stat, struct inode *inode, mode |= inode->i_mode & ~S_IALLUGO ; inode->i_mode = mode; } -if (!(flags &V9FS_STAT2INODE_KEEP_ISIZE ) && - stat->st_result_mask &P9_STATS_SIZE ) { +if (stat->st_result_mask &P9_STATS_SIZE ) { v9inode->netfs.remote_i_size = stat->st_size; -v9fs_i_size_write (inode, stat->st_size); +if (!(flags &V9FS_STAT2INODE_KEEP_ISIZE )) +v9fs_i_size_write (inode, stat->st_size); }if (stat->st_result_mask &P9_STATS_BLOCKS ) inode->i_blocks = stat->st_blocks;
Dominique Martinet
maintainerRe: [PATCH 4/5] 9p: Always update remote_i_size in stat2inode
David Howells wrote on Wed, Jan 03, 2024 at 02:59:28PM +0000:
Quoted 2 lines
Always update
remote_i_sizein v9fs_stat2inode*() if the size is available, even if we are asked not to updatei_isize
Sorry -- hold on for this patch, let's drop it for now and take it more slowly through next cycle.
I had mostly forgotten about V9FS_STAT2INODE_KEEP_ISIZE and not paying enough attention yesterday evening, but it's not innocent -- I assume netfs will do the right thing if we update the *remote* i_size when there is cached data, but the inode's i_size cannot be updated as easily.
It's hard to notice because the comment got split in 5e3cc1ee1405a7 ("9p: use inode->i_lock to protect i_size_write() under 32-bit"), but v9fs_refresh_inode* still have it:
/*
* We don't want to refresh inode->i_size,
* because we may have cached data
*/
I assume refreshing i_size at a bad time would act like a truncation of cached memory.
(To answer the other thread's comment that v9fs_i_size_write is useless; it's far from obvious enough but I'm afraid it is needed:
- include/linux/fs.h has a comment saying i_size_write does need locking around it for 32bit to avoid breaking i_size_seqcount; that's still true in today's tree.
- we could use any lock as long as it's coherent within the 9p subsystem, but we don't need a whole mutex so i_lock it is.)
Signature · 1 line
Dominique Martinet | Asmadeus
David Howells
maintainerPATCH 5/5 9p: Use length of data written to the server in preference to error
In reply to David HowellsIn v9fs_upload_to_server(), we pass the error to netfslib to terminate the subreq rather than the amount of data written - even if we did actually write something.
Further, we assume that the write is always entirely done if successful - but it might have been partially complete - as returned by p9_client_write(), but we ignore that.
Fix this by indicating the amount written by preference and only returning the error if we didn't write anything.
(We might want to return both in future if both are available as this might be useful as to whether we retry or not.)
--- fs/9p/vfs_addr.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)
fs/9p/vfs_addr.c +3 -4
@@ -29,12 +29,11 @@static void v9fs_upload_to_server (struct netfs_io_subrequest *subreq) {struct p9_fid *fid = subreq->rreq->netfs_priv; -int err; +int err, len;trace_netfs_sreq (subreq, netfs_sreq_trace_submit); -p9_client_write (fid, subreq->start, &subreq->io_iter, &err); -netfs_write_subrequest_terminated (subreq, err <0 ? err : subreq->len, - false); + len =p9_client_write (fid, subreq->start, &subreq->io_iter, &err); +netfs_write_subrequest_terminated (subreq, len ?: err, false); }static void v9fs_upload_to_server_worker (struct work_struct *work)
Re: [PATCH 5/5] 9p: Use length of data written to the server in preference to error
David Howells wrote on Wed, Jan 03, 2024 at 02:59:29PM +0000:
Quoted 8 lines
In
v9fs_upload_to_server(), we pass the error to netfslib to terminate the subreq rather than the amount of data written - even if we did actually write something.Further, we assume that the write is always entirely done if successful - but it might have been partially complete - as returned by
p9_client_write(), but we ignore that.
Quoted 7 lines
Fix this by indicating the amount written by preference and only returning the error if we didn't write anything.
(We might want to return both in future if both are available as this might be useful as to whether we retry or not.)
Suggested-by Dominique Martinet <[email protected]>
Thanks,
Signature · 1 line
Dominique Martinet | Asmadeus
David Howells
maintainer[PATCH 6/5] netfs: Rearrange netfs_io_subrequest to put request pointer first
In reply to David Howellsnetfs: Rearrange netfs_io_subrequest to put request pointer first
Rearrange the netfs_io_subrequest struct to put the netfs_io_request pointer (rreq) first. This then allows netfs_io_subrequest to be put in a union with a pointer to a wrapper around netfs_io_request. This will be useful in the future for cifs and maybe ceph.
--- include/linux/netfs.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/netfs.h +1 -1
@@ -204,8 +204,8 @@ struct netfs_cache_resources {* the pages it points to can be relied on to exist for the duration. */ struct netfs_io_subrequest { -struct work_struct work;struct netfs_io_request *rreq;/* Supervising I/O request */ +struct work_struct work;struct list_head rreq_link;/* Link in rreq->subrequests */ struct iov_iter io_iter;/* Iterator for this subrequest */ loff_t start;/* Where to start the I/O */
David Howells
maintainer[PATCH 7/5] netfs: Fix proc/fs/fscache symlink to point to "netfs" not "../netfs"
In reply to David HowellsFix the proc/fs/fscache symlink to point to "netfs" not "../netfs".
--- fs/netfs/fscache_proc.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
fs/netfs/fscache_proc.c +1 -1
@@ -16,7 +16,7 @@*/ int __init fscache_proc_init(void) { - if (!proc_symlink("fs/fscache", NULL, "../netfs")) + if (!proc_symlink("fs/fscache", NULL, "netfs")) goto error_sym; if (!proc_create_seq("fs/netfs/caches", S_IFREG | 0444, NULL,
Christian Brauner
maintainerQuoted 19 lines · David Howells
Hi Christian, Jeff, Gao, Dominique,
Here are some additional patches for my netfs-lib tree:
(1) Fix
__cachefiles_prepare_write()to correctly validate against the DIO alignment.(2) 9p: Fix initialisation of the
netfs_inodeso thati_sizeis set beforenetfs_inode_init()is called.(3) 9p: Do a couple of cleanups (remove a couple of unused vars and turn a
BUG_ON()into a warning).(4) 9p: Always update
remote_i_size, even if we're asked not to updatei_sizein stat2inode.(5) 9p: Return the amount written in preference to an error if we wrote something.
Quoted 8 lines
David
The netfslib postings:
Pulled this into vfs.netfs. Thanks, David.