On Thu, Nov 24, 2022 at 11:54:28AM +0800, Gary Lin via Grub-devel wrote:
> On Wed, Nov 23, 2022 at 03:44:54PM +0100, Daniel Kiper wrote:
> > On Wed, Nov 23, 2022 at 02:40:21PM +0800, Gary Lin via Grub-devel wrote:
> > > Per "man 5 cpio", the namesize in the cpio header includes the trailing
> > > NULL byte of the pathname and the pathname is followed by NULL bytes, but
> >
> > s/NULL/NUL/ and below please...
> >
> Will fix in v2.
Thanks!
> > > the current implementation excludes the trailing NULL byte when making
> > > the newc header plus the pathname. Although make_header() would pad the
> > > pathname string, the padding won't happen when strlen(name) +
> > > sizeof(struct newc_head) is a multiple of 4, and the non-NULL-terminated
> > > pathname may lead to unexpected results.
> > >
> > > For example, when loading linux with the following command:
> > >
> > > linux /boot/vmlinuz
> > > initrd newc:test12:/boot/test12 /boot/initrd
> > >
> > > Since strlen("test12") + sizeof(struct newc_head) is 116 = 29 * 4, there
> > > is no padding for the pathname, and the following error may show during
> > > linux boot:
> > >
> > > Initramfs unpacking failed: ZSTD-compressed data is trunc
> > >
> > > To avoid the potential problems, this commit includes the NULL byte when
> > > calling make_header() and adjusts the initrd size accordingly.
> > >
> > > Signed-off-by: Gary Lin <g...@suse.com>
> > > ---
> > > grub-core/loader/linux.c | 16 +++++++++-------
> > > 1 file changed, 9 insertions(+), 7 deletions(-)
> > >
> > > diff --git a/grub-core/loader/linux.c b/grub-core/loader/linux.c
> > > index 830360172..2b26d293f 100644
> > > --- a/grub-core/loader/linux.c
> > > +++ b/grub-core/loader/linux.c
> > > @@ -127,12 +127,14 @@ insert_dir (const char *name, struct dir **root,
> > > n->name = grub_strndup (cb, ce - cb);
> > > if (ptr)
> > > {
> > > + char *tmp_name = grub_strndup (name, ce - name);
> >
> > Why is this needed?
> Because I need the extra '\0' for make_header(). For example, if "name"
> is "dir123/file", insert_dir() only sends 'd''i''r''1''2''3' to
> make_header() to create the header like this:
>
> 00000070 30 37 30 37 30 31 32 31 38 33 32 41 37 43 30 30 |07070121832A7C00|
> 00000080 30 30 34 31 45 44 30 30 30 30 30 33 45 38 30 30 |0041ED000003E800|
> 00000090 30 30 30 30 36 34 30 30 30 30 30 30 30 33 36 33 |0000640000000363|
> 000000a0 37 37 32 44 43 38 30 30 30 30 30 30 30 30 30 30 |772DC80000000000|
> 000000b0 30 30 30 30 30 38 30 30 30 30 30 30 31 33 30 30 |0000080000001300|
> 000000c0 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 |0000000000000000|
> 000000d0 30 30 30 30 30 36 30 30 30 30 30 30 30 30 64 69 |00000600000000di|
> 000000e0 72 31 32 33 30 37 30 37 30 31 31 31 34 44 37 46 |r123070701114D7F|
> ^
> end of "dir123"
>
> As you can see, the pathname is ended without '\0', and it violates the
> cpio format. If we give make_header() the extra '\0':
>
> 00000070 30 37 30 37 30 31 32 31 38 33 32 41 37 43 30 30 |07070121832A7C00|
> 00000080 30 30 34 31 45 44 30 30 30 30 30 33 45 38 30 30 |0041ED000003E800|
> 00000090 30 30 30 30 36 34 30 30 30 30 30 30 30 33 36 33 |0000640000000363|
> 000000a0 37 37 32 44 43 38 30 30 30 30 30 30 30 30 30 30 |772DC80000000000|
> 000000b0 30 30 30 30 30 38 30 30 30 30 30 30 31 33 30 30 |0000080000001300|
> 000000c0 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 |0000000000000000|
> 000000d0 30 30 30 30 30 37 30 30 30 30 30 30 30 30 64 69 |00000700000000di|
> 000000e0 72 31 32 33 00 00 00 00 30 37 30 37 30 31 31 31 |r123....07070111|
>
> There are 3 more NUL bytes appended, and the user can safely read the
> pathname without a further check.
OK, this is what I expected. Though please add a comment before
grub_strndup() call and shortly say what is happening here.
> > And if it is needed you have to check for NULL here.
> Okay. Will add the check in v2.
>
> > I do not mention that formatting is broken too. You use spaces instead
> > of tabs here and below.
> >
> I thought the indentation is 2-space without tab. Will convert 8 spaces
> into one tab.
Cool! Thank you!
> > > grub_dprintf ("linux", "Creating directory %s, %s\n", name, ce);
> > > - ptr = make_header (ptr, name, ce - name,
> > > + ptr = make_header (ptr, tmp_name, ce - name + 1,
> > > 040777, 0);
> > > + grub_free (tmp_name);
> > > }
> > > if (grub_add (*size,
> > > - ALIGN_UP ((ce - (char *) name)
> > > + ALIGN_UP ((ce - (char *) name + 1)
> > > + sizeof (struct newc_head), 4),
> > > size))
> > > {
> > > @@ -191,7 +193,7 @@ grub_initrd_init (int argc, char *argv[],
> > > grub_initrd_close (initrd_ctx);
> > > return grub_errno;
> > > }
> > > - name_len = grub_strlen (initrd_ctx->components[i].newc_name);
> > > + name_len = grub_strlen (initrd_ctx->components[i].newc_name) + 1;
> > > if (grub_add (initrd_ctx->size,
> > > ALIGN_UP (sizeof (struct newc_head) + name_len, 4),
> > > &initrd_ctx->size) ||
> > > @@ -205,7 +207,7 @@ grub_initrd_init (int argc, char *argv[],
> > > {
> > > if (grub_add (initrd_ctx->size,
> > > ALIGN_UP (sizeof (struct newc_head)
> > > - + sizeof ("TRAILER!!!") - 1, 4),
> > > + + sizeof ("TRAILER!!!"), 4),
> >
> > This change looks strange and begs for explanation in the patch.
> >
> The adjustment is to match the change of "TRAILER!!!" below.
>
> > > &initrd_ctx->size))
> > > goto overflow;
> > > free_dir (root);
> > > @@ -233,7 +235,7 @@ grub_initrd_init (int argc, char *argv[],
> > > initrd_ctx->size = ALIGN_UP (initrd_ctx->size, 4);
> > > if (grub_add (initrd_ctx->size,
> > > ALIGN_UP (sizeof (struct newc_head)
> > > - + sizeof ("TRAILER!!!") - 1, 4),
> > > + + sizeof ("TRAILER!!!"), 4),
> >
> > Ditto and bellow...
> >
> > > &initrd_ctx->size))
> > > goto overflow;
> > > free_dir (root);
> > > @@ -304,7 +306,7 @@ grub_initrd_load (struct grub_linux_initrd_context
> > > *initrd_ctx,
> > > }
> > > else if (newc)
> > > {
> > > - ptr = make_header (ptr, "TRAILER!!!", sizeof ("TRAILER!!!") - 1,
> > > + ptr = make_header (ptr, "TRAILER!!!", sizeof ("TRAILER!!!"),
> > > 0, 0);
> The "TRAILER!!!" header should count the NUL byte, not ignore it.
Again, I expected this. However, this is not obvious at first sight. So, please
mention these changes in the commit message too. One sentence is enough.
Daniel
_______________________________________________
Grub-devel mailing list
Grub-devel@gnu.org
https://lists.gnu.org/mailman/listinfo/grub-devel
