RE: [PATCH RFCv3 2/3] lib/vsprintf.c: make %pD print full path for file

[Justin He]

Hi Rasmus

> -----Original Message-----
> From: Rasmus Villemoes <linux@xxxxxxxxxxxxxxxxxx>
> Sent: Saturday, June 12, 2021 5:28 AM
> To: Justin He <Justin.He@xxxxxxx>; Petr Mladek <pmladek@xxxxxxxx>; Steven
> Rostedt <rostedt@xxxxxxxxxxx>; Sergey Senozhatsky
> <senozhatsky@xxxxxxxxxxxx>; Andy Shevchenko
> <andriy.shevchenko@xxxxxxxxxxxxxxx>; Jonathan Corbet <corbet@xxxxxxx>;
> Alexander Viro <viro@xxxxxxxxxxxxxxxxxx>; Linus Torvalds <torvalds@linux-
> foundation.org>
> Cc: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>; Eric Biggers
> <ebiggers@xxxxxxxxxx>; Ahmed S. Darwish <a.darwish@xxxxxxxxxxxxx>; linux-
> doc@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; linux-
> fsdevel@xxxxxxxxxxxxxxx
> Subject: Re: [PATCH RFCv3 2/3] lib/vsprintf.c: make %pD print full path for
> file
>
> On 11/06/2021 17.59, Jia He wrote:
> > We have '%pD' for printing a filename. It may not be perfect (by
> > default it only prints one component.)
> >
> > As suggested by Linus at [1]:
> > A dentry has a parent, but at the same time, a dentry really does
> > inherently have "one name" (and given just the dentry pointers, you
> > can't show mount-related parenthood, so in many ways the "show just
> > one name" makes sense for "%pd" in ways it doesn't necessarily for
> > "%pD"). But while a dentry arguably has that "one primary component",
> > a _file_ is certainly not exclusively about that last component.
> >
> > Hence change the behavior of '%pD' to print full path of that file.
> >
> > Things become more complicated when spec.precision and spec.field_width
> > is added in. string_truncate() is to handle the small space case for
> > '%pD' precision and field_width.
> >
> > [1] https://lore.kernel.org/lkml/CAHk-=wimsMqGdzik187YWLb-
> ru+iktb4MYbMQG1rnZ81dXYFVg@xxxxxxxxxxxxxx/
> >
> > Suggested-by: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
> > Signed-off-by: Jia He <justin.he@xxxxxxx>
> > ---
> >  Documentation/core-api/printk-formats.rst |  5 ++-
> >  lib/vsprintf.c                            | 47 +++++++++++++++++++++--
> >  2 files changed, 46 insertions(+), 6 deletions(-)
> >
> > diff --git a/Documentation/core-api/printk-formats.rst
> b/Documentation/core-api/printk-formats.rst
> > index f063a384c7c8..95ba14dc529b 100644
> > --- a/Documentation/core-api/printk-formats.rst
> > +++ b/Documentation/core-api/printk-formats.rst
> > @@ -408,12 +408,13 @@ dentry names
> >  ::
> >
> >     %pd{,2,3,4}
> > -   %pD{,2,3,4}
> > +   %pD
> >
> >  For printing dentry name; if we race with :c:func:`d_move`, the name
> might
> >  be a mix of old and new ones, but it won't oops.  %pd dentry is a safer
> >  equivalent of %s dentry->d_name.name we used to use, %pd<n> prints ``n``
> > -last components.  %pD does the same thing for struct file.
> > +last components.  %pD prints full file path together with mount-related
> > +parenthood.
> >
> >  Passed by reference.
> >
> > diff --git a/lib/vsprintf.c b/lib/vsprintf.c
> > index f0c35d9b65bf..317b65280252 100644
> > --- a/lib/vsprintf.c
> > +++ b/lib/vsprintf.c
> > @@ -27,6 +27,7 @@
> >  #include <linux/string.h>
> >  #include <linux/ctype.h>
> >  #include <linux/kernel.h>
> > +#include <linux/dcache.h>
> >  #include <linux/kallsyms.h>
> >  #include <linux/math64.h>
> >  #include <linux/uaccess.h>
> > @@ -601,6 +602,20 @@ char *widen_string(char *buf, int n, char *end,
> struct printf_spec spec)
> >  }
> >
> >  /* Handle string from a well known address. */
> > +static char *string_truncate(char *buf, char *end, const char *s,
> > +                        u32 full_len, struct printf_spec spec)
> > +{
> > +   int lim = 0;
> > +
> > +   if (buf < end) {
>
> See below, I think the sole caller guarantees this,

Ok, will remove this check statement

>
> > +           if (spec.precision >= 0)
> > +                   lim = strlen(s) - min_t(int, spec.precision, strlen(s));
> > +
> > +           return widen_string(buf + full_len, full_len, end - lim, spec);
> > +   }
> > +
> > +   return buf;
>
> which is good because this would almost certainly be wrong (violating
> the "always forward buf appropriately regardless of whether you wrote
> something" rule).
>
> > +}
> >  static char *string_nocheck(char *buf, char *end, const char *s,
> >                         struct printf_spec spec)
> >  {
> > @@ -920,13 +935,37 @@ char *dentry_name(char *buf, char *end, const
> struct dentry *d, struct printf_sp
> >  }
> >
> >  static noinline_for_stack
> > -char *file_dentry_name(char *buf, char *end, const struct file *f,
> > +char *file_d_path_name(char *buf, char *end, const struct file *f,
> >                     struct printf_spec spec, const char *fmt)
> >  {
> > +   const struct path *path;
> > +   char *p;
> > +   int prepend_len, reserved_size, dpath_len;
> > +
> >     if (check_pointer(&buf, end, f, spec))
> >             return buf;
> >
> > -   return dentry_name(buf, end, f->f_path.dentry, spec, fmt);
> > +   path = &f->f_path;
> > +   if (check_pointer(&buf, end, path, spec))
> > +           return buf;
> > +
> > +   p = d_path_unsafe(path, buf, end - buf, &prepend_len);
>
> If I'm reading this right, you're using buf as scratch space to write
> however much of the path fits. Then [*]
>
> > +   /* Minus 1 byte for '\0' */
> > +   dpath_len = end - buf - prepend_len - 1;
> > +
> > +   reserved_size = max_t(int, dpath_len, spec.field_width);
> > +
> > +   /* no filling space at all */
> > +   if (buf >= end || !buf)
> > +           return buf + reserved_size;
>
> Why the !buf check? The only way we can have that is the snprintf(NULL,
> 0, ...) case of asking how much space we'd need to malloc, right? In
> which case end would be NULL+0 == NULL, so buf >= end automatically,
> regardless of how much have been "printed" before %pD.

My original purpose is to avoid any memory copy/move for kvasprintf->
vsnprintf(NULL, 0,...). But as you said, this can be folded into the case
buf >= end.
Do you think whether following case should be forbidden?:
vsnprintf(NULL, 8,...).
Sorry if it is too verbose. If above invoking is valid, !buf should
still be checked.

>
> > +
> > +   /* small space for long name */
> > +   if (buf < end && prepend_len < 0)
>
> So if we did an early return for buf >= end, we now know buf < end and
> hence the first part here is redundant.
>
> Anyway, as for [*]:
>
> > +           return string_truncate(buf, end, p, dpath_len, spec);
> > +
> > +   /* space is enough */
> > +   return string_nocheck(buf, end, p, spec);
>
> Now you're passing p to string_truncate or string_nocheck, while p
> points somewhere into buf itself. I can't convince myself that would be
> safe. At the very least, it deserves a couple of comments.

When code goes here, the buffer space must be as follows:
|.........|.........|
buf       p         end

So string_nocheck is safe because essential it would byte-to-byte copy p to buf.

But I agree comments are needed here.


--
Cheers,
Justin (Jia He)


IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.