On Thu, Feb 22, 2024 at 09:58:24AM -0600, Bill O'Donnell wrote: > On Thu, Feb 22, 2024 at 09:22:52AM -0600, Eric Sandeen wrote: > > As filesystems are converted to the new mount API, informational messages, > > errors, and warnings are being routed through infof, errorf, and warnf > > type functions provided by the mount API, which places these messages in > > the log buffer associated with the filesystem context rather than > > in the kernel log / dmesg. > > > > However, userspace is not yet extracting these messages, so they are > > essentially getting lost. mount(8) still refers the user to dmesg(1) > > on failure. Eric pointed me at a sample program (over irc) that revealed that one can read() the fd returned by fsopen() to obtain these messages. I had no idea that was possible... because there are no manpages for these system calls! Can we /please/ document and write (more) fstests for these things? [that was mostly aimed at dhowells] --D > > At least for now, modify logfc() to always log these messages to dmesg > > as well as to the fileystem context. This way we can continue converting > > filesystems to the new mount API interfaces without worrying about losing > > this information until userspace can retrieve it. > > > > Signed-off-by: Eric Sandeen <sandeen@xxxxxxxxxx> > > Reviewed-by: Bill O'Donnell <bodonnel@xxxxxxxxxx> > > > --- > > > > A few considerations/concerns: > > > > * viro suggested that maybe this should be conditional - possibly config > > > > * systemd is currently probing with a dummy mount option which will > > generate noise, see > > https://github.com/systemd/systemd/blob/main/src/basic/mountpoint-util.c#L759 > > i.e. - > > [ 10.689256] proc: Unknown parameter 'adefinitelynotexistingmountoption' > > [ 10.801045] tmpfs: Unknown parameter 'adefinitelynotexistingmountoption' > > [ 11.119431] proc: Unknown parameter 'adefinitelynotexistingmountoption' > > [ 11.692032] proc: Unknown parameter 'adefinitelynotexistingmountoption' > > > > * will this generate other dmesg noise in general if the mount api messages > > are more noisy in general? (spot-checking old conversions, I don't think so.) > > > > fs/fs_context.c | 38 ++++++++++++++++++++------------------ > > 1 file changed, 20 insertions(+), 18 deletions(-) > > > > diff --git a/fs/fs_context.c b/fs/fs_context.c > > index 98589aae5208..3c78b99d5cae 100644 > > --- a/fs/fs_context.c > > +++ b/fs/fs_context.c > > @@ -427,8 +427,8 @@ struct fs_context *vfs_dup_fs_context(struct fs_context *src_fc) > > EXPORT_SYMBOL(vfs_dup_fs_context); > > > > /** > > - * logfc - Log a message to a filesystem context > > - * @log: The filesystem context to log to, or NULL to use printk. > > + * logfc - Log a message to dmesg and optionally a filesystem context > > + * @log: The filesystem context to log to, or NULL to use printk alone > > * @prefix: A string to prefix the output with, or NULL. > > * @level: 'w' for a warning, 'e' for an error. Anything else is a notice. > > * @fmt: The format of the buffer. > > @@ -439,22 +439,24 @@ void logfc(struct fc_log *log, const char *prefix, char level, const char *fmt, > > struct va_format vaf = {.fmt = fmt, .va = &va}; > > > > va_start(va, fmt); > > - if (!log) { > > - switch (level) { > > - case 'w': > > - printk(KERN_WARNING "%s%s%pV\n", prefix ? prefix : "", > > - prefix ? ": " : "", &vaf); > > - break; > > - case 'e': > > - printk(KERN_ERR "%s%s%pV\n", prefix ? prefix : "", > > - prefix ? ": " : "", &vaf); > > - break; > > - default: > > - printk(KERN_NOTICE "%s%s%pV\n", prefix ? prefix : "", > > - prefix ? ": " : "", &vaf); > > - break; > > - } > > - } else { > > + switch (level) { > > + case 'w': > > + printk(KERN_WARNING "%s%s%pV\n", prefix ? prefix : "", > > + prefix ? ": " : "", &vaf); > > + break; > > + case 'e': > > + printk(KERN_ERR "%s%s%pV\n", prefix ? prefix : "", > > + prefix ? ": " : "", &vaf); > > + break; > > + default: > > + printk(KERN_NOTICE "%s%s%pV\n", prefix ? prefix : "", > > + prefix ? ": " : "", &vaf); > > + break; > > + } > > + va_end(va); > > + > > + va_start(va, fmt); > > + if (log) { > > unsigned int logsize = ARRAY_SIZE(log->buffer); > > u8 index; > > char *q = kasprintf(GFP_KERNEL, "%c %s%s%pV\n", level, > > -- > > 2.43.0 > > > >
