On Thu, 2017-09-14 at 09:50 +1000, NeilBrown wrote: > Since 4.13, errors from writeback are more reliably reported > to all file descriptors that might be relevant. > > Add notes to this effect, and also add details about ENOSPC and EDQUOT > which can be delayed in a similar manner to EIO - for NFS in particular. > > Signed-off-by: NeilBrown <neilb@xxxxxxxx> > --- > > This is my summary of recent changes, and details that have been made > clear during the exploration of those changes. > > I haven't mentioned the fact that EPERM can be returned by > write/fsync/close on NFS if the permissions on the server are changed. > We probably should ... are there other errors that are worth mentioning > along with EPERM, ENOSPC, EDQUOT ?? > > Thanks, > NeilBronw > Many thanks for doing this! It was on my to-do list. Comments below: > > man2/close.2 | 9 +++++++++ > man2/fsync.2 | 19 ++++++++++++++++++- > man2/write.2 | 20 +++++++++++++++++--- > 3 files changed, 44 insertions(+), 4 deletions(-) > > diff --git a/man2/close.2 b/man2/close.2 > index 751ec322b1f1..9776c839b8b6 100644 > --- a/man2/close.2 > +++ b/man2/close.2 > @@ -82,6 +82,15 @@ call was interrupted by a signal; see > .TP > .B EIO > An I/O error occurred. > +.TP > +.BR ENOSPC ", " EDQUOT > +On NFS, these errors are not normally reported against the first write > +which exceeds the available storage space, but instead against a > +subsequent > +.BR write (2), > +.BR fsync (2), > +or > +.BR close (2). > .PP > See NOTES for a discussion of why > .BR close () > diff --git a/man2/fsync.2 b/man2/fsync.2 > index f1a01301da0f..e706a08d360d 100644 > --- a/man2/fsync.2 > +++ b/man2/fsync.2 > @@ -120,12 +120,29 @@ is set appropriately. > is not a valid open file descriptor. > .TP > .B EIO > -An error occurred during synchronization. > +An error occurred during synchronization. This error may relate > +to data written to some other file descriptor on the same file. > +.\" commit 088737f44bbf6378745f5b57b035e57ee3dc4750 > +Since Linux 4.13 errors from write-back will be reported to > +all file descriptors that might have written the data which triggered > +the error, and which are still open. This is a little awkward. How could we report to a fd that was no longer open? How about: "Since Linux 4.13, errors from write-back will be reported to all file descriptors that were open at the time that the error was recorded." > Some filesystems (e.g. NFS) > +keep close track of which data came through which file descriptor, > +and give more precise reporting. Other filesystems (e.g. most local > +filesystems) will report errors to all file descriptors on the same > +file. > .TP > .BR EROFS ", " EINVAL > .I fd > is bound to a special file (e.g., a pipe, FIFO, or socket) > which does not support synchronization. > +.TP > +.BR ENOSPC ", " EDQUOT > +.I fd > +is bound to a file on NFS or another filesystem which does not allocate > +space at the time of a > +.BR write (2) > +system call, and some previous write failed due to insufficient > +storage space. > .SH CONFORMING TO > POSIX.1-2001, POSIX.1-2008, 4.3BSD. > .SH AVAILABILITY > diff --git a/man2/write.2 b/man2/write.2 > index 6a39b5b5541d..1a9a86b03b04 100644 > --- a/man2/write.2 > +++ b/man2/write.2 > @@ -47,7 +47,7 @@ write \- write to a file descriptor > .BR write () > writes up to > .I count > -bytes from the buffer pointed > +bytes from the buffer starting at > .I buf > to the file referred to by the file descriptor > .IR fd . > @@ -181,6 +181,14 @@ or the file offset is not suitably aligned. > .TP > .B EIO > A low-level I/O error occurred while modifying the inode. > +This error may relate to data written by an earlier > +.BR write (2), > +which may have been issued to a different file descriptor on > +the same file. Since Linux 4.13 errors from write-back will > +be reported to all file descriptors that might have > +written the data which triggered the error, and which are still > +open. This is where things get a little more vague. Some filesystems will return errors on a subsequent write(2) when previous writeback has failed -- some don't. In either case though, write(2) should never advance your errseq_t cursor, so only an fsync will "clear" an earlier error. I'm not sure how best to convey that in the manpages though. > +.\" commit 088737f44bbf6378745f5b57b035e57ee3dc4750 > .TP > .B ENOSPC > The device containing the file referred to by > @@ -222,8 +230,14 @@ unsigned and signed integer data types specified by POSIX.1. > A successful return from > .BR write () > does not make any guarantee that data has been committed to disk. > -In fact, on some buggy implementations, it does not even guarantee > -that space has successfully been reserved for the data. > +On some filesystems, including NFS, it does not even guarantee > +that space has successfully been reserved for the data. In the case, > +some errors might be delayed to a future > +.BR write (2) > +or to > +.BR fsync (2) > +or even > +.BR close (2). > The only way to be sure is to call > .BR fsync (2) > after you are done writing all your data. -- Jeff Layton <jlayton@xxxxxxxxxx>
