Matthew Bobrowski <repnop@xxxxxxxxxx> writes:
> On Tue, Apr 12, 2022 at 02:50:28PM -0400, Gabriel Krisman Bertazi wrote:
>>
>> FAN_FS_ERROR is a new event for fanotify to report filesystem errors.
>> This documents how to use the feature and specific caveats.
>>
>> Signed-off-by: Gabriel Krisman Bertazi <krisman@xxxxxxxxxxxxx>
>>
>> + Michael, linux-man
>>
>> Matthew,
>>
>> as discussed, this is rebased on top of the PIDFD documentation: commit
>> 207080c7f7f5 ("fanotify: Document FAN_REPORT_PIDFD Feature").
>
> Mostly looks OK. I've left a couple suggestions that I think could be
> adopted.
>
>> Changes since v3:
>> (Matthew)
>> - Rewording and fixes from github)
>> (amir)
>> - 5.15 -> 5.16
>>
>> Changes since v2:
>> (matthew)
>> - Grammar
>> - List filesystems that support the feature
>> - file system -> filesystem
>> Changes since v1:
>> (Matthew)
>> - Grammar fixes
>> - Don't use the term "submitted" for events sent to the listener
>> - Clarify the kind of information that is file system specific
>>
>> ---
>> man2/fanotify_mark.2 | 28 +++++++++++++++++++
>> man7/fanotify.7 | 64 ++++++++++++++++++++++++++++++++++++++++++++
>> 2 files changed, 92 insertions(+)
>>
>> diff --git a/man2/fanotify_mark.2 b/man2/fanotify_mark.2
>> index 9a45cbb77893..8f9bb863980b 100644
>> --- a/man2/fanotify_mark.2
>> +++ b/man2/fanotify_mark.2
>> @@ -203,6 +203,27 @@ Create an event when a marked file or directory itself is deleted.
>> An fanotify group that identifies filesystem objects by file handles
>> is required.
>> .TP
>> +.BR FAN_FS_ERROR " (since Linux 5.16)"
>> +.\" commit 9709bd548f11a092d124698118013f66e1740f9b
>> +Create an event when a filesystem error is detected.
>> +An fanotify group that identifies filesystem objects by file handles
>> +is required.
>> +.IP
>> +An additional information record of type
>> +.BR FAN_EVENT_INFO_TYPE_ERROR
>> +is returned for each event in the read buffer.
>> +.IP
>
> I'd update this to:
>
> Create an event when a filesystem error is detected.
> An additional information record of type
> .B FAN_EVENT_INFO_TYPE_ERROR
> is returned for each event in the read buffer.
> An fanotify group that identifies filesystem objects by file handles
> is required.
>
>> +Events of such type are dependent on support
>> +from the underlying filesystem.
>> +At the time of this writing,
>
> At the time of writing,
>
>> +only the
>> +.B ext4
>> +filesystem supports this feature.
>> +.IP
>> +See
>> +.BR fanotify (7)
>> +for additional details.
>> +.TP
>> .BR FAN_MOVED_FROM " (since Linux 5.1)"
>> .\" commit 235328d1fa4251c6dcb32351219bb553a58838d2
>> Create an event when a file or directory has been moved from a marked
>> @@ -391,6 +412,13 @@ and mask contains a flag for permission events
>> or
>> .BR FAN_ACCESS_PERM ).
>> .TP
>> +.B EINVAL
>> +The group was initialized without
>> +.B FAN_REPORT_FID
>> +but one or more events in
>> +.I mask
>> +require it.
>> +.TP
>
> I'd update this to:
>
> The fanotify group was initialized without
> .BR FAN_REPORT_FID .
> One or more event types specified in the
> .I mask
> require it.
>
>> .B ENODEV
>> The filesystem object indicated by
>> .I pathname
>> diff --git a/man7/fanotify.7 b/man7/fanotify.7
>> index 57dd2b040380..59c928cf233f 100644
>> --- a/man7/fanotify.7
>> +++ b/man7/fanotify.7
>> @@ -217,6 +217,25 @@ struct fanotify_event_info_pidfd {
>> .EE
>> .in
>> .PP
>> +In case of a
>> +.B FAN_FS_ERROR
>> +event,
>> +an additional record describing the error that occurred
> ^
> information
>
>> +is returned alongside the generic
>> +.I fanotify_event_metadata
>> +structure within the read buffer.
>> +This structure is defined as follows:
>> +.PP
>> +.in +4n
>> +.EX
>> +struct fanotify_event_info_error {
>> + struct fanotify_event_info_header hdr;
>> + __s32 error;
>> + __u32 error_count;
>> +};
>> +.EE
>> +.in
>> +.PP
>> All information records contain a nested structure of type
>> .IR fanotify_event_info_header .
>> This structure holds meta-information about the information record
>> @@ -358,6 +377,9 @@ A child file or directory was deleted in a watched parent.
>> .B FAN_DELETE_SELF
>> A watched file or directory was deleted.
>> .TP
>> +.B FAN_FS_ERROR
>> +A filesystem error was detected.
>> +.TP
>> .B FAN_MOVED_FROM
>> A file or directory has been moved from a watched parent directory.
>> .TP
>> @@ -630,6 +652,25 @@ Once the event listener has dealt with an event and the pidfd is no
>> longer required, the pidfd should be closed via
>> .BR close(2) .
>> .PP
>> +The fields of the
>> +.I fanotify_event_info_error
>> +structure are as follows:
>> +.TP
>> +.I hdr
>> +This is a structure of type
>> +.IR fanotify_event_info_header .
>> +The
>> +.I info_type
>> +field is set to
>> +.BR FAN_EVENT_INFO_TYPE_ERROR .
>> +.TP
>> +.I error
>> +Identifies the type of error that occurred.
>> +.TP
>> +.I error_count
>> +This is a counter of the number of errors suppressed
>> +since the last error was read.
>> +.PP
>> The following macros are provided to iterate over a buffer containing
>> fanotify event metadata returned by a
>> .BR read (2)
>> @@ -719,6 +760,29 @@ field.
>> In that case, the audit subsystem will log information about the access
>> decision to the audit logs.
>> .\"
>> +.SS Monitoring filesystems for errors
>> +A single FAN_FS_ERROR event is stored per filesystem at once.
>
> This is just for my understanding. So, it isn't until the event
> listener reads and clears the FAN_FS_ERROR event from the event queue
> that you're able to receive subsequent filesystem errors for
> potentially different objects on the same filesystem. Based on this,
> it sounds like if there were to be two separate objects on the same
> filesystem that experienced some kind of error at the same time, then
> only one of those errors encoded in FAN_FS_ERROR would be returned?
Matthew,
That is correct. we queue up to one event per filesystem. The
reasoning is that, when an error occurs, the most important piece of
information is the first failure, but the filesystem is very likely to
spew dozens of extra errors after the first one, usually not very
interesting. A proper response would be using some fsck tool to repair
the entire filesystem.
>> +Extra error messages are suppressed and accounted for
>> +in the
>> +.I error_count
>> +field of the existing FAN_FS_ERROR event record,
>> +but details about the errors are lost.
>> +.PP
>> +Errors types reported by FAN_FS_ERROR are generic errno values
>
> I'd update this to:
>
> Errors reported by FAN_FS_ERROR ...
>
>> +but not all kinds of errors types are reported by all filesystems.
> ^
> s/errors/error
>
>> +.PP
>> +Errors not directly related to a file (i.e. super block corruption)
>> +are reported with an invalid
>> +.IR file_handle .
>> +For these errors, the
>> +.I file_handle
>> +will have the field
>> +.I handle_type
>> +set to
>> +.BR FILEID_INVALID ,
>> +and the handle buffer size set to
>> +.BR 0 .
>> +.\"
>> .SS Closing the fanotify file descriptor
>> When all file descriptors referring to the fanotify notification group are
>> closed, the fanotify group is released and its resources
>
> /M
--
Gabriel Krisman Bertazi
