misaligned memory access in cmdlinepart.c
Hi,
in mtdpart_setup_real()/newpart() command line parser 'this_mtd' structure can be misaligned,
it may cause exception on some kind of CPU.
That happened because the structure allocated in a variable length area and got mixed with partition names.
The patch follows.
Regards.
Timofei.
drivers/mtd/cmdlinepart.c renamed from cmdlinepart.c +8 -2
@@ -234,12 +234,14 @@ static int mtdpart_setup_real(char *s)* parse one mtd. have it reserve memory for the * struct cmdline_mtd_partition and the mtd-id string. */ +#define THIS_MTD_ALIGN_CONST (sizeof(void*)-1) parts =newpart (p +1 ,/* cmdline */ &s,/* out: updated cmdline ptr */ &num_parts,/* out: number of parts */ 0 ,/* first partition */ (unsigned char **)&this_mtd,/* out: extra mem */ - mtd_id_len +1 +sizeof (*this_mtd)); + mtd_id_len +1 +sizeof (*this_mtd) + +THIS_MTD_ALIGN_CONST );if (!parts) {/*
@@ -252,7 +254,11 @@ static int mtdpart_setup_real(char *s)return 0 ; } -/* enter results */ +/* align this_mtd */ + this_mtd = (struct cmdline_mtd_partition *) + (~THIS_MTD_ALIGN_CONST & +THIS_MTD_ALIGN_CONST + (unsigned long )(char *)this_mtd); +/* enter results */ this_mtd->parts = parts; this_mtd->num_parts = num_parts; this_mtd->mtd_id = (char *)(this_mtd +1 );
Hi.
27 Apr I've submitted the tiny patch.
No answer since then.
Did the patch rejected?
Or any chance it could be accepted?
--
Regards.
Timofei.
Thanks for bugging again.
Quoted 7 lines · Timofei V. Bondarenko
in mtdpart_setup_real()/newpart() command line parser '
this_mtd' structure can be misaligned,
it may cause exception on some kind of CPU.That happened because the structure allocated in a variable length area and got mixed with partition names.
Your description appears to make sense (I don't use cmdlinepart). The patch, however, should be changed to use ALIGN() from include/linux/kernel.h instead.
Quoted 35 lines
drivers/mtd/cmdlinepart.c renamed from cmdlinepart.c +0
-018:40:02.971778640 +0400
drivers/mtd/cmdlinepart.c renamed from cmdlinepart.c +3
-1@@ -234,12 +234,14 @@ static int mtdpart_setup_real(char *s)* parse one mtd. have it reserve memory for the * struct cmdline_mtd_partition and the mtd-id string. */ +#define THIS_MTD_ALIGN_CONST (sizeof(void*)-1) parts =newpart (p +1 ,/* cmdline */ &s,/* out: updated cmdline ptr */ &num_parts,/* out: number of parts */ 0 ,/* first partition */ (unsigned char **)&this_mtd,/* out: extra mem */ - mtd_id_len +1 +sizeof (*this_mtd)); + mtd_id_len +1 +sizeof (*this_mtd) + +THIS_MTD_ALIGN_CONST );if (!parts)drivers/mtd/cmdlinepart.c renamed from cmdlinepart.c +5
-1{/* @@ -252,7 +254,11 @@ static int mtdpart_setup_real(char *s)return 0 ; } -/* enter results */ +/* align this_mtd */ + this_mtd = (struct cmdline_mtd_partition *) + (~THIS_MTD_ALIGN_CONST & +THIS_MTD_ALIGN_CONST + (unsigned long )(char *)this_mtd); +/* enter results */ this_mtd->parts = parts; this_mtd->num_parts = num_parts; this_mtd->mtd_id = (char *)(this_mtd +1 );
Quoted 3 lines
______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/
Jörn
Signature · 4 lines
There are two ways of constructing a software design: one way is to make it so simple that there are obviously no deficiencies, and the other is to make it so complicated that there are no obvious deficiencies. -- C. A. R. Hoare
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 17 lines · Jörn Engel
cmdlinepart.c static int mtdpart_setup_real(char *s) · line 237
Quoted 12 lines · Timofei V. Bondarenko
Index:
cmdlinepart.c
===================================================================
RCS file:/home/cvs/mtd/drivers/mtd/cmdlinepart.c,vretrieving revision 1.17
diff -u -p -r1.17cmdlinepart.ccmdlinepart.c +1
-0@@ -234,12 +234,14 @@ static int mtdpart_setup_real(char *s)* parse one mtd. have it reserve memory for the * struct cmdline_mtd_partition and the mtd-id string. */ +#define THIS_MTD_ALIGN_CONST sizeof(void*)Not sure if this macro is worth the pain. If it is, you should declare it outside the function. Just above is what most people do.
Ok.. That's matter of taste. I'm rather like to complicate things.
Quoted 7 lines
cmdlinepart.c static int mtdpart_setup_real(char *s) · line 241
Quoted 4 lines
cmdlinepart.c +4
-0+/* align this_mtd */ + this_mtd = (struct cmdline_mtd_partition *) +ALIGN ((unsigned long )this_mtd,THIS_MTD_ALIGN_CONST ); +/* enter results */ Gcc allows arithmetic with (void*) exactly for code like this:
Quoted 5 lines
this_mtd = ALIGN((void*)this_mtd, THIS_MTD_ALIGN_CONST); or: this_mtd = ALIGN((void*)this_mtd, sizeof(void*));You should retest it, though.
error: invalid operands to binary &
Then we've to cast it from integer to a pointer type.
--
Timofei.
Index: cmdlinepart.c
===================================================================
RCS file: /home/cvs/mtd/drivers/mtd/cmdlinepart.c,v retrieving revision 1.17
cmdlinepart.c +6 -2
@@ -239,7 +239,8 @@ static int mtdpart_setup_real(char *s) &num_parts,/* out: number of parts */ 0 ,/* first partition */ (unsigned char **)&this_mtd,/* out: extra mem */ - mtd_id_len + 1 + sizeof(*this_mtd)); + mtd_id_len +1 +sizeof (*this_mtd) + + sizeof(void*)-1 /*alignment*/);if (!parts) {/*
@@ -252,7 +253,10 @@ static int mtdpart_setup_real(char *s)return 0 ; } -/* enter results */ +/* align this_mtd */ + this_mtd = (struct cmdline_mtd_partition *) +ALIGN ((unsigned long )this_mtd,sizeof (void *)); +/* enter results */ this_mtd->parts = parts; this_mtd->num_parts = num_parts; this_mtd->mtd_id = (char *)(this_mtd +1 );
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 5 lines · Jörn Engel
Hmm. Well, I just don't care enough, let's keep it.
Can you test the current patch and see if it works for you? Once you've confirmed that, I'll commit it.
Yes. I've tested it on the uclinux-bf533.
--
Timofei.
Quoted 8 lines · Timofei V. Bondarenko
Quoted 5 lines · Jörn Engel
Hmm. Well, I just don't care enough, let's keep it.
Can you test the current patch and see if it works for you? Once you've confirmed that, I'll commit it.
Yes. I've tested it on the uclinux-bf533.
Doesn't apply with patch -p1. Can you respin the patch against mtd cvs and make sure it does? If in doubt, read:
http://www.zip.com.au/~akpm/linux/patches/stuff/tpp.txt
Jörn
Signature · 3 lines
Fools ignore complexity. Pragmatists suffer it. Some can avoid it. Geniuses remove it. -- Perlis's Programming Proverb #58, SIGPLAN Notices, Sept. 1982
Quoted 9 lines · Jörn Engel
Quoted 7 lines · Timofei V. Bondarenko
Quoted 4 lines · Jörn Engel
Hmm. Well, I just don't care enough, let's keep it.
Can you test the current patch and see if it works for you? Once you've confirmed that, I'll commit it.
Quoted 6 lines
Quoted 1 line
Yes. I've tested it on the uclinux-bf533.
Doesn't apply with patch -p1. Can you respin the patch against mtd cvs and make sure it does? If in doubt, read:
http://www.zip.com.au/~akpm/linux/patches/stuff/tpp.txt
Ok, sorry for noise.
Is it good enough?
drivers/mtd/cmdlinepart.c +6 -2
@@ -239,7 +239,8 @@ static int mtdpart_setup_real(char *s) &num_parts,/* out: number of parts */ 0 ,/* first partition */ (unsigned char **)&this_mtd,/* out: extra mem */ - mtd_id_len + 1 + sizeof(*this_mtd)); + mtd_id_len +1 +sizeof (*this_mtd) + + sizeof(void*)-1 /*alignment*/);if (!parts) {/*
@@ -252,7 +253,10 @@ static int mtdpart_setup_real(char *s)return 0 ; } -/* enter results */ +/* align this_mtd */ + this_mtd = (struct cmdline_mtd_partition *) +ALIGN ((unsigned long )this_mtd,sizeof (void *)); +/* enter results */ this_mtd->parts = parts; this_mtd->num_parts = num_parts; this_mtd->mtd_id = (char *)(this_mtd +1 );
Hi Timofei,
I guess you should use a mail client that doesn't expand tabs to spaces, like sylpheed or pine.
Best regards,
Vitaly
Quoted 58 lines · Timofei V. Bondarenko
Quoted 10 lines · JЖrn Engel
Quoted 8 lines · Timofei V. Bondarenko
Quoted 4 lines · JЖrn Engel
Hmm. Well, I just don't care enough, let's keep it.
Can you test the current patch and see if it works for you? Once you've confirmed that, I'll commit it.
Quoted 7 lines
Quoted 1 line
Yes. I've tested it on the uclinux-bf533.
Doesn't apply with patch -p1. Can you respin the patch against mtd cvs and make sure it does? If in doubt, read:
http://www.zip.com.au/~akpm/linux/patches/stuff/tpp.txtOk, sorry for noise.
Is it good enough?diff -u -p -r1.17
cmdlinepart.cdrivers/mtd/cmdlinepart.c 7 Jun 2005 08:02:00 -0000 renamed from drivers/mtd/cmdlinepart.c 26 Nov 2004 11:18:47 -0000 1.17 +6
-2@@ -239,7 +239,8 @@ static int mtdpart_setup_real(char *s) &num_parts, /* out: number of parts */ 0, /* first partition */ (unsigned char**)&this_mtd, /* out: extra mem */ - mtd_id_len + 1 + sizeof(*this_mtd)); + mtd_id_len + 1 + sizeof(*this_mtd) + + sizeof(void*)-1 /*alignment*/); if(!parts) { /*@@ -252,7 +253,10 @@ static int mtdpart_setup_real(char *s) return 0; } - /* enter results */ + /* align this_mtd */ + this_mtd = (struct cmdline_mtd_partition *) + ALIGN((unsigned long)this_mtd, sizeof(void*)); + /* enter results */ this_mtd->parts = parts; this_mtd->num_parts = num_parts; this_mtd->mtd_id = (char*)(this_mtd + 1);______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/
Quoted 28 lines · Vitaly Wool
Hi Timofei,
I guess you should use a mail client that doesn't expand tabs to spaces, like sylpheed or pine.Best regards,
VitalyQuoted 20 lines · Timofei V. Bondarenko
Quoted 9 lines · JЖrn Engel
Quoted 7 lines · Timofei V. Bondarenko
Quoted 4 lines · JЖrn Engel
Hmm. Well, I just don't care enough, let's keep it.
Can you test the current patch and see if it works for you? Once you've confirmed that, I'll commit it.
Quoted 6 lines
Quoted 1 line
Yes. I've tested it on the uclinux-bf533.
Doesn't apply with patch -p1. Can you respin the patch against mtd cvs and make sure it does? If in doubt, read:
http://www.zip.com.au/~akpm/linux/patches/stuff/tpp.txtOk, sorry for noise.
Is it good enough?
Ohh. That's not silly mail, that's poor me.
drivers/mtd/cmdlinepart.c +6 -2
@@ -239,7 +239,8 @@ static int mtdpart_setup_real(char *s) &num_parts,/* out: number of parts */ 0 ,/* first partition */ (unsigned char **)&this_mtd,/* out: extra mem */ - mtd_id_len + 1 + sizeof(*this_mtd)); + mtd_id_len +1 +sizeof (*this_mtd) + + sizeof(void*)-1 /*alignment*/);if (!parts) {/*
@@ -252,7 +253,10 @@ static int mtdpart_setup_real(char *s)return 0 ; } -/* enter results */ +/* align this_mtd */ + this_mtd = (struct cmdline_mtd_partition *) +ALIGN ((unsigned long )this_mtd,sizeof (void *)); +/* enter results */ this_mtd->parts = parts; this_mtd->num_parts = num_parts; this_mtd->mtd_id = (char *)(this_mtd +1 );
Quoted 9 lines · Vitaly Wool
Hi Timofei,
I guess you should use a mail client that doesn't expand tabs to spaces, like sylpheed or pine.Quoted 5 lines
Quoted 5 lines
JЖrn Engel wrote:
Doesn't apply with patch -p1. Can you respin the patch against mtd cvs and make sure it does? If in doubt, read:
http://www.zip.com.au/~akpm/linux/patches/stuff/tpp.txt
I'm apologize for my unusual inaccuracy. Now, the tabs are preserved; patch -p1 looks happy. For cvslog:
---
the this_mtd structure should be aligned
---
drivers/mtd/cmdlinepart.c +6 -2
@@ -239,7 +239,8 @@ static int mtdpart_setup_real(char *s) &num_parts,/* out: number of parts */ 0 ,/* first partition */ (unsigned char **)&this_mtd,/* out: extra mem */ - mtd_id_len + 1 + sizeof(*this_mtd)); + mtd_id_len +1 +sizeof (*this_mtd) + + sizeof(void*)-1 /*alignment*/);if (!parts) {/*
@@ -252,7 +253,10 @@ static int mtdpart_setup_real(char *s)return 0 ; } -/* enter results */ +/* align this_mtd */ + this_mtd = (struct cmdline_mtd_partition *) +ALIGN ((unsigned long )this_mtd,sizeof (void *)); +/* enter results */ this_mtd->parts = parts; this_mtd->num_parts = num_parts; this_mtd->mtd_id = (char *)(this_mtd +1 );
Quoted 3 lines · Timofei V. Bondarenko
Ok, sorry for noise.
Is it good enough?
Still broken. You have some whitespace noise in it.
I could easily fix it up, but I feel like a bastard and prefer to share the pain. If your mail reader is screwing it up, you can send the patch as an attachment. Inline would be preferred, though.
Quoted 18 lines
diff -u -p -r1.17
cmdlinepart.cdrivers/mtd/cmdlinepart.c +2
-1@@ -239,7 +239,8 @@ static int mtdpart_setup_real(char *s) &num_parts,/* out: number of parts */ 0 ,/* first partition */ (unsigned char **)&this_mtd,/* out: extra mem */ - mtd_id_len + 1 + sizeof(*this_mtd)); + mtd_id_len +1 +sizeof (*this_mtd) + + sizeof(void*)-1 /*alignment*/);if (!parts) {drivers/mtd/cmdlinepart.c +0
-0/* @@ -252,7 +253,10 @@ static int mtdpart_setup_real(char *s)return 0 ;
drivers/mtd/cmdlinepart.c static int mtdpart_setup_real(char *s) · line 262
Quoted 10 lines
drivers/mtd/cmdlinepart.c +4
-1} -/* enter results */ +/* align this_mtd */ + this_mtd = (struct cmdline_mtd_partition *) +ALIGN ((unsigned long )this_mtd,sizeof (void *)); +/* enter results */ this_mtd->parts = parts; this_mtd->num_parts = num_parts; this_mtd->mtd_id = (char *)(this_mtd +1 );
Jörn
Signature · 4 lines
I've never met a human being who would want to read 17,000 pages of documentation, and if there was, I'd kill him to get him out of the gene pool. -- Joseph Costello