| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
Pull 9p fixes from Dominique Martinet:
"Two fixes headed to stable trees:
- a trace event was dumping uninitialized values
- a missing lock that was thought to have exclusive access, and it
turned out not to"
* tag '9p-for-6.10-rc2' of https://github.com/martinetd/linux:
9p: add missing locking around taking dentry fid list
net/9p: fix uninit-value in p9_client_rpc()
|
|
The demote_ok glock operation is only still used to prevent the inode
glocks of the "jindex" and "rindex" directories from getting recycled
while they are still referenced by sdp->sd_jindex and sdp->sd_rindex.
However, the LRU walking code will no longer recycle glocks which are
referenced, so the demote_ok glock operation is obsolete and can be
removed.
Each of a glock's holders in the gl_holders list is holding a reference
on the glock, so when the list of holders isn't empty in demote_ok(),
the existing reference count check will already prevent the glock from
getting released. This means that demote_ok() is obsolete as well.
Signed-off-by: Andreas Gruenbacher <[email protected]>
|
|
This reverts commit e7ccaf5fe1590667b3fa2f8df5c5ec9ba0dc5b85.
Before commit e7ccaf5fe159, every time a resource group glock was
dequeued by gfs2_glock_dq(), it was added to the glock LRU list even
though the glock was still referenced by the resource group and could
never be evicted, anyway. Commit e7ccaf5fe159 added a GLOF_LRU hack to
avoid that overhead for resource group glocks, and that hack was since
adopted for some other types of glocks as well.
We now no longer add glocks to the glock LRU list while they are still
referenced. This solves the underlying problem, and obsoletes the
GLOF_LRU hack.
Signed-off-by: Andreas Gruenbacher <[email protected]>
(cherry picked from commit 3e5257c810cba91e274d07f3db5cf013c7c830be)
|
|
In the current glock reference counting model, a bias of one is added to
a glock's refcount when it is locked (gl->gl_state != LM_ST_UNLOCKED).
A glock is removed from the lru_list when it is enqueued, and added back
when it is dequeued. This isn't a very appropriate model because most
glocks are held for long periods of time (for example, the inode "owns"
references to its inode and iopen glocks as long as the inode is cached
even when the glock state changes to LM_ST_UNLOCKED), and they can only
be freed when they are no longer referenced, anyway.
Fix this by getting rid of the refcount bias for locked glocks. That
way, we can use lockref_put_or_lock() to efficiently drop all but the
last glock reference, and put the glock onto the lru_list when the last
reference is dropped. When find_insert_glock() returns a reference to a
cached glock, it removes the glock from the lru_list.
Dumping the "glocks" and "glstats" debugfs files also takes glock
references, but instead of removing the glocks from the lru_list in that
case as well, we leave them on the list. This ensures that dumping
those files won't perturb the order of the glocks on the lru_list.
In addition, when the last reference to an *unlocked* glock is dropped,
we immediately free it; this preserves the preexisting behavior. If it
later turns out that caching unlocked glocks is useful in some
situations, we can change the caching strategy.
It is currently unclear if a glock that has no active references can
have the GLF_LFLUSH flag set. To make sure that such a glock won't
accidentally be evicted due to memory pressure, we add a GLF_LFLUSH
check to gfs2_dispose_glock_lru().
Signed-off-by: Andreas Gruenbacher <[email protected]>
|
|
Switch to a per-filesystem glock workqueue. Additional workqueues are
cheap nowadays, and keeping separate workqueues allows to flush the work
of each filesystem without affecting the others.
Signed-off-by: Andreas Gruenbacher <[email protected]>
|
|
When glocks cannot be freed for a long time, avoid the "task blocked for
more than N seconds" messages and report how many glocks are still
outstanding, instead.
Signed-off-by: Andreas Gruenbacher <[email protected]>
|
|
Clean up the messy code in gfs2_glock_get(). No change in
functionality.
Signed-off-by: Andreas Gruenbacher <[email protected]>
|
|
Invert the meaning of the GLF_INITIAL flag: right now, when GLF_INITIAL
is set, a DLM lock exists and we have a valid identifier for it; when
GLF_INITIAL is cleared, no DLM lock exists (yet). This is confusing.
In addition, it makes more sense to highlight the exceptional case
(i.e., no DLM lock exists yet) in glock dumps and trace points than to
highlight the common case.
To avoid confusion between the "old" and the "new" meaning of the flag,
use 'a' instead of 'I' to represent the flag.
For improved code consistency, check if the GLF_INITIAL flag is cleared
to determine whether a DLM lock exists instead of checking if the lock
identifier is non-zero.
Document what the flag is used for.
Signed-off-by: Andreas Gruenbacher <[email protected]>
|
|
Signed-off-by: Andreas Gruenbacher <[email protected]>
|
|
requests"
[email protected] <[email protected]> says:
We've been testing ondemand mode for cachefiles since January, and we're
almost done. We hit a lot of issues during the testing period, and this
patch set fixes some of the issues related to ondemand requests.
The patches have passed internal testing without regression.
The following is a brief overview of the patches, see the patches for
more details.
Patch 1-5: Holding reference counts of reqs and objects on read requests
to avoid malicious restore leading to use-after-free.
Patch 6-10: Add some consistency checks to copen/cread/get_fd to avoid
malicious copen/cread/close fd injections causing use-after-free or hung.
Patch 11: When cache is marked as CACHEFILES_DEAD, flush all requests,
otherwise the kernel may be hung. since this state is irreversible, the
daemon can read open requests but cannot copen.
Patch 12: Allow interrupting a read request being processed by killing
the read process as a way of avoiding hung in some special cases.
fs/cachefiles/daemon.c | 3 +-
fs/cachefiles/internal.h | 5 +
fs/cachefiles/ondemand.c | 217 ++++++++++++++++++++++--------
include/trace/events/cachefiles.h | 8 +-
4 files changed, 176 insertions(+), 57 deletions(-)
* patches from https://lore.kernel.org/r/[email protected]:
cachefiles: make on-demand read killable
cachefiles: flush all requests after setting CACHEFILES_DEAD
cachefiles: Set object to close if ondemand_id < 0 in copen
cachefiles: defer exposing anon_fd until after copy_to_user() succeeds
cachefiles: never get a new anonymous fd if ondemand_id is valid
cachefiles: add spin_lock for cachefiles_ondemand_info
cachefiles: add consistency check for copen/cread
cachefiles: remove err_put_fd label in cachefiles_ondemand_daemon_read()
cachefiles: fix slab-use-after-free in cachefiles_ondemand_daemon_read()
cachefiles: fix slab-use-after-free in cachefiles_ondemand_get_fd()
cachefiles: remove requests from xarray during flushing requests
cachefiles: add output string to cachefiles_obj_[get|put]_ondemand_fd
Signed-off-by: Christian Brauner <[email protected]>
|
|
Replacing wait_for_completion() with wait_for_completion_killable() in
cachefiles_ondemand_send_req() allows us to kill processes that might
trigger a hunk_task if the daemon is abnormal.
But now only CACHEFILES_OP_READ is killable, because OP_CLOSE and OP_OPEN
is initiated from kworker context and the signal is prohibited in these
kworker.
Note that when the req in xas changes, i.e. xas_load(&xas) != req, it
means that a process will complete the current request soon, so wait
again for the request to be completed.
In addition, add the cachefiles_ondemand_finish_req() helper function to
simplify the code.
Suggested-by: Hou Tao <[email protected]>
Signed-off-by: Baokun Li <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Acked-by: Jeff Layton <[email protected]>
Reviewed-by: Jia Zhu <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|
|
In ondemand mode, when the daemon is processing an open request, if the
kernel flags the cache as CACHEFILES_DEAD, the cachefiles_daemon_write()
will always return -EIO, so the daemon can't pass the copen to the kernel.
Then the kernel process that is waiting for the copen triggers a hung_task.
Since the DEAD state is irreversible, it can only be exited by closing
/dev/cachefiles. Therefore, after calling cachefiles_io_error() to mark
the cache as CACHEFILES_DEAD, if in ondemand mode, flush all requests to
avoid the above hungtask. We may still be able to read some of the cached
data before closing the fd of /dev/cachefiles.
Note that this relies on the patch that adds reference counting to the req,
otherwise it may UAF.
Fixes: c8383054506c ("cachefiles: notify the user daemon when looking up cookie")
Signed-off-by: Baokun Li <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Acked-by: Jeff Layton <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|
|
If copen is maliciously called in the user mode, it may delete the request
corresponding to the random id. And the request may have not been read yet.
Note that when the object is set to reopen, the open request will be done
with the still reopen state in above case. As a result, the request
corresponding to this object is always skipped in select_req function, so
the read request is never completed and blocks other process.
Fix this issue by simply set object to close if its id < 0 in copen.
Signed-off-by: Zizhi Wo <[email protected]>
Signed-off-by: Baokun Li <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Acked-by: Jeff Layton <[email protected]>
Reviewed-by: Jia Zhu <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|
|
After installing the anonymous fd, we can now see it in userland and close
it. However, at this point we may not have gotten the reference count of
the cache, but we will put it during colse fd, so this may cause a cache
UAF.
So grab the cache reference count before fd_install(). In addition, by
kernel convention, fd is taken over by the user land after fd_install(),
and the kernel should not call close_fd() after that, i.e., it should call
fd_install() after everything is ready, thus fd_install() is called after
copy_to_user() succeeds.
Fixes: c8383054506c ("cachefiles: notify the user daemon when looking up cookie")
Suggested-by: Hou Tao <[email protected]>
Signed-off-by: Baokun Li <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Acked-by: Jeff Layton <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|
|
Now every time the daemon reads an open request, it gets a new anonymous fd
and ondemand_id. With the introduction of "restore", it is possible to read
the same open request more than once, and therefore an object can have more
than one anonymous fd.
If the anonymous fd is not unique, the following concurrencies will result
in an fd leak:
t1 | t2 | t3
------------------------------------------------------------
cachefiles_ondemand_init_object
cachefiles_ondemand_send_req
REQ_A = kzalloc(sizeof(*req) + data_len)
wait_for_completion(&REQ_A->done)
cachefiles_daemon_read
cachefiles_ondemand_daemon_read
REQ_A = cachefiles_ondemand_select_req
cachefiles_ondemand_get_fd
load->fd = fd0
ondemand_id = object_id0
------ restore ------
cachefiles_ondemand_restore
// restore REQ_A
cachefiles_daemon_read
cachefiles_ondemand_daemon_read
REQ_A = cachefiles_ondemand_select_req
cachefiles_ondemand_get_fd
load->fd = fd1
ondemand_id = object_id1
process_open_req(REQ_A)
write(devfd, ("copen %u,%llu", msg->msg_id, size))
cachefiles_ondemand_copen
xa_erase(&cache->reqs, id)
complete(&REQ_A->done)
kfree(REQ_A)
process_open_req(REQ_A)
// copen fails due to no req
// daemon close(fd1)
cachefiles_ondemand_fd_release
// set object closed
-- umount --
cachefiles_withdraw_cookie
cachefiles_ondemand_clean_object
cachefiles_ondemand_init_close_req
if (!cachefiles_ondemand_object_is_open(object))
return -ENOENT;
// The fd0 is not closed until the daemon exits.
However, the anonymous fd holds the reference count of the object and the
object holds the reference count of the cookie. So even though the cookie
has been relinquished, it will not be unhashed and freed until the daemon
exits.
In fscache_hash_cookie(), when the same cookie is found in the hash list,
if the cookie is set with the FSCACHE_COOKIE_RELINQUISHED bit, then the new
cookie waits for the old cookie to be unhashed, while the old cookie is
waiting for the leaked fd to be closed, if the daemon does not exit in time
it will trigger a hung task.
To avoid this, allocate a new anonymous fd only if no anonymous fd has
been allocated (ondemand_id == 0) or if the previously allocated anonymous
fd has been closed (ondemand_id == -1). Moreover, returns an error if
ondemand_id is valid, letting the daemon know that the current userland
restore logic is abnormal and needs to be checked.
Fixes: c8383054506c ("cachefiles: notify the user daemon when looking up cookie")
Signed-off-by: Baokun Li <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Acked-by: Jeff Layton <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|
|
The following concurrency may cause a read request to fail to be completed
and result in a hung:
t1 | t2
---------------------------------------------------------
cachefiles_ondemand_copen
req = xa_erase(&cache->reqs, id)
// Anon fd is maliciously closed.
cachefiles_ondemand_fd_release
xa_lock(&cache->reqs)
cachefiles_ondemand_set_object_close(object)
xa_unlock(&cache->reqs)
cachefiles_ondemand_set_object_open
// No one will ever close it again.
cachefiles_ondemand_daemon_read
cachefiles_ondemand_select_req
// Get a read req but its fd is already closed.
// The daemon can't issue a cread ioctl with an closed fd, then hung.
So add spin_lock for cachefiles_ondemand_info to protect ondemand_id and
state, thus we can avoid the above problem in cachefiles_ondemand_copen()
by using ondemand_id to determine if fd has been closed.
Fixes: c8383054506c ("cachefiles: notify the user daemon when looking up cookie")
Signed-off-by: Baokun Li <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Acked-by: Jeff Layton <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|
|
This prevents malicious processes from completing random copen/cread
requests and crashing the system. Added checks are listed below:
* Generic, copen can only complete open requests, and cread can only
complete read requests.
* For copen, ondemand_id must not be 0, because this indicates that the
request has not been read by the daemon.
* For cread, the object corresponding to fd and req should be the same.
Signed-off-by: Baokun Li <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Acked-by: Jeff Layton <[email protected]>
Reviewed-by: Jingbo Xu <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|
|
The err_put_fd label is only used once, so remove it to make the code
more readable. In addition, the logic for deleting error request and
CLOSE request is merged to simplify the code.
Signed-off-by: Baokun Li <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Acked-by: Jeff Layton <[email protected]>
Reviewed-by: Jia Zhu <[email protected]>
Reviewed-by: Gao Xiang <[email protected]>
Reviewed-by: Jingbo Xu <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|
|
We got the following issue in a fuzz test of randomly issuing the restore
command:
==================================================================
BUG: KASAN: slab-use-after-free in cachefiles_ondemand_daemon_read+0xb41/0xb60
Read of size 8 at addr ffff888122e84088 by task ondemand-04-dae/963
CPU: 13 PID: 963 Comm: ondemand-04-dae Not tainted 6.8.0-dirty #564
Call Trace:
kasan_report+0x93/0xc0
cachefiles_ondemand_daemon_read+0xb41/0xb60
vfs_read+0x169/0xb50
ksys_read+0xf5/0x1e0
Allocated by task 116:
kmem_cache_alloc+0x140/0x3a0
cachefiles_lookup_cookie+0x140/0xcd0
fscache_cookie_state_machine+0x43c/0x1230
[...]
Freed by task 792:
kmem_cache_free+0xfe/0x390
cachefiles_put_object+0x241/0x480
fscache_cookie_state_machine+0x5c8/0x1230
[...]
==================================================================
Following is the process that triggers the issue:
mount | daemon_thread1 | daemon_thread2
------------------------------------------------------------
cachefiles_withdraw_cookie
cachefiles_ondemand_clean_object(object)
cachefiles_ondemand_send_req
REQ_A = kzalloc(sizeof(*req) + data_len)
wait_for_completion(&REQ_A->done)
cachefiles_daemon_read
cachefiles_ondemand_daemon_read
REQ_A = cachefiles_ondemand_select_req
msg->object_id = req->object->ondemand->ondemand_id
------ restore ------
cachefiles_ondemand_restore
xas_for_each(&xas, req, ULONG_MAX)
xas_set_mark(&xas, CACHEFILES_REQ_NEW)
cachefiles_daemon_read
cachefiles_ondemand_daemon_read
REQ_A = cachefiles_ondemand_select_req
copy_to_user(_buffer, msg, n)
xa_erase(&cache->reqs, id)
complete(&REQ_A->done)
------ close(fd) ------
cachefiles_ondemand_fd_release
cachefiles_put_object
cachefiles_put_object
kmem_cache_free(cachefiles_object_jar, object)
REQ_A->object->ondemand->ondemand_id
// object UAF !!!
When we see the request within xa_lock, req->object must not have been
freed yet, so grab the reference count of object before xa_unlock to
avoid the above issue.
Fixes: 0a7e54c1959c ("cachefiles: resend an open request if the read request's object is closed")
Signed-off-by: Baokun Li <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Acked-by: Jeff Layton <[email protected]>
Reviewed-by: Jia Zhu <[email protected]>
Reviewed-by: Jingbo Xu <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|
|
We got the following issue in a fuzz test of randomly issuing the restore
command:
==================================================================
BUG: KASAN: slab-use-after-free in cachefiles_ondemand_daemon_read+0x609/0xab0
Write of size 4 at addr ffff888109164a80 by task ondemand-04-dae/4962
CPU: 11 PID: 4962 Comm: ondemand-04-dae Not tainted 6.8.0-rc7-dirty #542
Call Trace:
kasan_report+0x94/0xc0
cachefiles_ondemand_daemon_read+0x609/0xab0
vfs_read+0x169/0xb50
ksys_read+0xf5/0x1e0
Allocated by task 626:
__kmalloc+0x1df/0x4b0
cachefiles_ondemand_send_req+0x24d/0x690
cachefiles_create_tmpfile+0x249/0xb30
cachefiles_create_file+0x6f/0x140
cachefiles_look_up_object+0x29c/0xa60
cachefiles_lookup_cookie+0x37d/0xca0
fscache_cookie_state_machine+0x43c/0x1230
[...]
Freed by task 626:
kfree+0xf1/0x2c0
cachefiles_ondemand_send_req+0x568/0x690
cachefiles_create_tmpfile+0x249/0xb30
cachefiles_create_file+0x6f/0x140
cachefiles_look_up_object+0x29c/0xa60
cachefiles_lookup_cookie+0x37d/0xca0
fscache_cookie_state_machine+0x43c/0x1230
[...]
==================================================================
Following is the process that triggers the issue:
mount | daemon_thread1 | daemon_thread2
------------------------------------------------------------
cachefiles_ondemand_init_object
cachefiles_ondemand_send_req
REQ_A = kzalloc(sizeof(*req) + data_len)
wait_for_completion(&REQ_A->done)
cachefiles_daemon_read
cachefiles_ondemand_daemon_read
REQ_A = cachefiles_ondemand_select_req
cachefiles_ondemand_get_fd
copy_to_user(_buffer, msg, n)
process_open_req(REQ_A)
------ restore ------
cachefiles_ondemand_restore
xas_for_each(&xas, req, ULONG_MAX)
xas_set_mark(&xas, CACHEFILES_REQ_NEW);
cachefiles_daemon_read
cachefiles_ondemand_daemon_read
REQ_A = cachefiles_ondemand_select_req
write(devfd, ("copen %u,%llu", msg->msg_id, size));
cachefiles_ondemand_copen
xa_erase(&cache->reqs, id)
complete(&REQ_A->done)
kfree(REQ_A)
cachefiles_ondemand_get_fd(REQ_A)
fd = get_unused_fd_flags
file = anon_inode_getfile
fd_install(fd, file)
load = (void *)REQ_A->msg.data;
load->fd = fd;
// load UAF !!!
This issue is caused by issuing a restore command when the daemon is still
alive, which results in a request being processed multiple times thus
triggering a UAF. So to avoid this problem, add an additional reference
count to cachefiles_req, which is held while waiting and reading, and then
released when the waiting and reading is over.
Note that since there is only one reference count for waiting, we need to
avoid the same request being completed multiple times, so we can only
complete the request if it is successfully removed from the xarray.
Fixes: e73fa11a356c ("cachefiles: add restore command to recover inflight ondemand read requests")
Suggested-by: Hou Tao <[email protected]>
Signed-off-by: Baokun Li <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Acked-by: Jeff Layton <[email protected]>
Reviewed-by: Jia Zhu <[email protected]>
Reviewed-by: Jingbo Xu <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|
|
Even with CACHEFILES_DEAD set, we can still read the requests, so in the
following concurrency the request may be used after it has been freed:
mount | daemon_thread1 | daemon_thread2
------------------------------------------------------------
cachefiles_ondemand_init_object
cachefiles_ondemand_send_req
REQ_A = kzalloc(sizeof(*req) + data_len)
wait_for_completion(&REQ_A->done)
cachefiles_daemon_read
cachefiles_ondemand_daemon_read
// close dev fd
cachefiles_flush_reqs
complete(&REQ_A->done)
kfree(REQ_A)
xa_lock(&cache->reqs);
cachefiles_ondemand_select_req
req->msg.opcode != CACHEFILES_OP_READ
// req use-after-free !!!
xa_unlock(&cache->reqs);
xa_destroy(&cache->reqs)
Hence remove requests from cache->reqs when flushing them to avoid
accessing freed requests.
Fixes: c8383054506c ("cachefiles: notify the user daemon when looking up cookie")
Signed-off-by: Baokun Li <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Acked-by: Jeff Layton <[email protected]>
Reviewed-by: Jia Zhu <[email protected]>
Reviewed-by: Gao Xiang <[email protected]>
Reviewed-by: Jingbo Xu <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|
|
We were accidentally returning -EROFS during recovery on filesystem
inconsistency - since this is what the journal returns on emergency
shutdown.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Can't actually be used uninitialized, but gcc was being silly.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Compatibility fix - we no longer have a separate table for which order
gc walks btrees in, and special case the stripes btree directly.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Fix the 'make W=1' warning:
WARNING: modpost: missing MODULE_DESCRIPTION() in fs/bcachefs/mean_and_variance_test.o
Signed-off-by: Jeff Johnson <[email protected]>
Signed-off-by: Kent Overstreet <[email protected]>
|
|
bch2_check_version_downgrade() was setting c->sb.version, which
bch2_sb_set_downgrade() expects to be at the previous version; and it
shouldn't even have been set directly because c->sb.version is updated
by write_super().
Signed-off-by: Kent Overstreet <[email protected]>
|
|
delete_dead_snapshots now runs before the main fsck.c passes which check
for keys for invalid snapshots; thus, it needs those checks as well.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Consolidate per-key work into delete_dead_snapshots_process_key(), so we
now walk all keys once, not twice.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
We now track whether a transaction is locked, and verify that we don't
have nodes locked when the transaction isn't locked; reorder relocks to
not pop the new assert.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
This function is used for finding the hash seed (which is the same in
all versions of an inode in different snapshots): ff an inode has been
deleted in a child snapshot we need to iterate until we find a live
version.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
It can be useful to know the exact byte offset within a btree node where
an error occured.
Signed-off-by: Kent Overstreet <[email protected]>
|
|
Function handle_callback() is used to request a glock demote. This
often happens in response to a conflicting remote locking request and
subsequent bast callback from DLM, but there are other reasons for
triggering a demote request as well, such as when trying to release a
glock in response to memory pressure. To clarify that, rename the
function to request_demote().
Signed-off-by: Andreas Gruenbacher <[email protected]>
|
|
The GLF_FROZEN flag indicates that a reply to a DLM locking request has
been received, but should not be processed at this time. To clarify
that meaning, rename the flag to GLF_HAVE_FROZEN_REPLY.
Signed-off-by: Andreas Gruenbacher <[email protected]>
|
|
The GLF_REPLY_PENDING flag indicates to glock_work_func() that in
response to a locking request, DLM has sent a reply that needs to be
processed. A flag with that name could as well indicate that we are
waiting on a reply from DLM, however. To disambiguate these two cases,
rename the flag to GLF_HAVE_REPLY.
Signed-off-by: Andreas Gruenbacher <[email protected]>
|
|
Rename the GLF_FREEING flag to GLF_UNLOCKED, and the ->go_free glock
operation to ->go_unlocked. This mechanism is used to wait for the
underlying DLM lock to be unlocked; being able to free the glock is a
consequence of the DLM lock being unlocked.
Signed-off-by: Andreas Gruenbacher <[email protected]>
|
|
The return statement at the end of run_queue() is useless.
Signed-off-by: Andreas Gruenbacher <[email protected]>
|
|
Function __gfs2_glock_dq() gets defined before it is used, so there is
no need for a separate function declaration.
Signed-off-by: Andreas Gruenbacher <[email protected]>
|
|
If a write path in COW mode fails, either before submitting a bio for the
new extents or an actual IO error happens, we can end up allowing a fast
fsync to log file extent items that point to unwritten extents.
This is because dropping the extent maps happens when completing ordered
extents, at btrfs_finish_one_ordered(), and the completion of an ordered
extent is executed in a work queue.
This can result in a fast fsync to start logging file extent items based
on existing extent maps before the ordered extents complete, therefore
resulting in a log that has file extent items that point to unwritten
extents, resulting in a corrupt file if a crash happens after and the log
tree is replayed the next time the fs is mounted.
This can happen for both direct IO writes and buffered writes.
For example consider a direct IO write, in COW mode, that fails at
btrfs_dio_submit_io() because btrfs_extract_ordered_extent() returned an
error:
1) We call btrfs_finish_ordered_extent() with the 'uptodate' parameter
set to false, meaning an error happened;
2) That results in marking the ordered extent with the BTRFS_ORDERED_IOERR
flag;
3) btrfs_finish_ordered_extent() queues the completion of the ordered
extent - so that btrfs_finish_one_ordered() will be executed later in
a work queue. That function will drop extent maps in the range when
it's executed, since the extent maps point to unwritten locations
(signaled by the BTRFS_ORDERED_IOERR flag);
4) After calling btrfs_finish_ordered_extent() we keep going down the
write path and unlock the inode;
5) After that a fast fsync starts and locks the inode;
6) Before the work queue executes btrfs_finish_one_ordered(), the fsync
task sees the extent maps that point to the unwritten locations and
logs file extent items based on them - it does not know they are
unwritten, and the fast fsync path does not wait for ordered extents
to complete, which is an intentional behaviour in order to reduce
latency.
For the buffered write case, here's one example:
1) A fast fsync begins, and it starts by flushing delalloc and waiting for
the writeback to complete by calling filemap_fdatawait_range();
2) Flushing the dellaloc created a new extent map X;
3) During the writeback some IO error happened, and at the end io callback
(end_bbio_data_write()) we call btrfs_finish_ordered_extent(), which
sets the BTRFS_ORDERED_IOERR flag in the ordered extent and queues its
completion;
4) After queuing the ordered extent completion, the end io callback clears
the writeback flag from all pages (or folios), and from that moment the
fast fsync can proceed;
5) The fast fsync proceeds sees extent map X and logs a file extent item
based on extent map X, resulting in a log that points to an unwritten
data extent - because the ordered extent completion hasn't run yet, it
happens only after the logging.
To fix this make btrfs_finish_ordered_extent() set the inode flag
BTRFS_INODE_NEEDS_FULL_SYNC in case an error happened for a COW write,
so that a fast fsync will wait for ordered extent completion.
Note that this issues of using extent maps that point to unwritten
locations can not happen for reads, because in read paths we start by
locking the extent range and wait for any ordered extents in the range
to complete before looking for extent maps.
Reviewed-by: Qu Wenruo <[email protected]>
Signed-off-by: Filipe Manana <[email protected]>
Signed-off-by: David Sterba <[email protected]>
|
|
A current limitation of open_by_handle_at() is that it's currently not possible
to use it from within containers at all because we require CAP_DAC_READ_SEARCH
in the initial namespace. That's unfortunate because there are scenarios where
using open_by_handle_at() from within containers.
Two examples:
(1) cgroupfs allows to encode cgroups to file handles and reopen them with
open_by_handle_at().
(2) Fanotify allows placing filesystem watches they currently aren't usable in
containers because the returned file handles cannot be used.
Here's a proposal for relaxing the permission check for open_by_handle_at().
(1) Opening file handles when the caller has privileges over the filesystem
(1.1) The caller has an unobstructed view of the filesystem.
(1.2) The caller has permissions to follow a path to the file handle.
This doesn't address the problem of opening a file handle when only a portion
of a filesystem is exposed as is common in containers by e.g., bind-mounting a
subtree. The proposal to solve this use-case is:
(2) Opening file handles when the caller has privileges over a subtree
(2.1) The caller is able to reach the file from the provided mount fd.
(2.2) The caller has permissions to construct an unobstructed path to the
file handle.
(2.3) The caller has permissions to follow a path to the file handle.
The relaxed permission checks are currently restricted to directory file
handles which are what both cgroupfs and fanotify need. Handling disconnected
non-directory file handles would lead to a potentially non-deterministic api.
If a disconnected non-directory file handle is provided we may fail to decode
a valid path that we could use for permission checking. That in itself isn't a
problem as we would just return EACCES in that case. However, confusion may
arise if a non-disconnected dentry ends up in the cache later and those opening
the file handle would suddenly succeed.
* It's potentially possible to use timing information (side-channel) to infer
whether a given inode exists. I don't think that's particularly
problematic. Thanks to Jann for bringing this to my attention.
* An unrelated note (IOW, these are thoughts that apply to
open_by_handle_at() generically and are unrelated to the changes here):
Jann pointed out that we should verify whether deleted files could
potentially be reopened through open_by_handle_at(). I don't think that's
possible though.
Another potential thing to check is whether open_by_handle_at() could be
abused to open internal stuff like memfds or gpu stuff. I don't think so
but I haven't had the time to completely verify this.
This dates back to discussions Amir and I had quite some time ago and thanks to
him for providing a lot of details around the export code and related patches!
Link: https://lore.kernel.org/r/[email protected]
Reviewed-by: Amir Goldstein <[email protected]>
Reviewed-by: Jeff Layton <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|
|
Wolfram reported that debugfs remained empty on some of his boards
triggering the message "debugfs: Unknown parameter 'auto'".
The root of the issue is that we ignored unknown mount options in the
old mount api but we started rejecting unknown mount options in the new
mount api. Continue to ignore unknown mount options to not regress
userspace.
Fixes: a20971c18752 ("vfs: Convert debugfs to use the new mount API")
Link: https://lore.kernel.org/r/20240527100618.np2wqiw5mz7as3vk@ninjato
Reported-by: Wolfram Sang <[email protected]>
Tested-by: Wolfram Sang <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|
|
Fix the 'make W=1' warning:
WARNING: modpost: missing MODULE_DESCRIPTION() in fs/autofs/autofs4.o
Signed-off-by: Jeff Johnson <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Acked-by: Al Viro <[email protected]>
Acked-by: Ian Kent <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|
|
Fix the 'make W=1' warnings:
WARNING: modpost: missing MODULE_DESCRIPTION() in fs/fat/fat.o
WARNING: modpost: missing MODULE_DESCRIPTION() in fs/fat/fat_test.o
Signed-off-by: Jeff Johnson <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Acked-by: OGAWA Hirofumi <[email protected]>
Acked-by: Al Viro <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|
|
Fix the 'make W=1' warnings:
WARNING: modpost: missing MODULE_DESCRIPTION() in fs/binfmt_misc.o
WARNING: modpost: missing MODULE_DESCRIPTION() in fs/binfmt_script.o
Signed-off-by: Jeff Johnson <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Acked-by: Al Viro <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
|