Skip to contentSkip to conversation

fs: make d_path-like functions all have unsigned size

PATCH Superseded Greg Kroah-Hartman 4 participants 9 messages Jul 27, 2021 VFS

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fs: make d_path-like functions all have unsigned size

When running static analysis tools to find where signed values could potentially wrap the family of d_path() functions turn out to trigger a lot of mess. In evaluating the code, all of these usages seem safe, but pointer math is involved so if a negative number is ever somehow passed into these functions, memory can be traversed backwards in ways not intended.

Resolve all of the abuguity by just making "size" an unsigned value, which takes the guesswork out of everything involved.

Cc Alexander Viro <[email protected]>
Cc Andy Shevchenko <[email protected]>
Cc "Ahmed S. Darwish" <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Eric Biggers <[email protected]>
---
 fs/d_path.c            | 14 +++++++-------
 include/linux/dcache.h | 12 ++++++------
 2 files changed, 13 insertions(+), 13 deletions(-)
fs/d_path.c +7 -7
index 23a53f7b5c71..7876b741a47e 100644
@@ -182,7 +182,7 @@ static int prepend_path(const struct path *path,
  */
 char *__d_path(const struct path *path,
 	       const struct path *root,
-	       char *buf, int buflen)
+	       char *buf, unsigned int buflen)
 {
 	DECLARE_BUFFER(b, buf, buflen);
 
@@ -193,7 +193,7 @@ char *__d_path(const struct path *path,
 }
 
 char *d_absolute_path(const struct path *path,
-	       char *buf, int buflen)
+	       char *buf, unsigned int buflen)
 {
 	struct path root = {};
 	DECLARE_BUFFER(b, buf, buflen);
@@ -230,7 +230,7 @@ static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
  *
  * "buflen" should be positive.
  */
-char *d_path(const struct path *path, char *buf, int buflen)
+char *d_path(const struct path *path, char *buf, unsigned int buflen)
 {
 	DECLARE_BUFFER(b, buf, buflen);
 	struct path root;
@@ -266,7 +266,7 @@ EXPORT_SYMBOL(d_path);
 /*
  * Helper function for dentry_operations.d_dname() members
  */
-char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
+char *dynamic_dname(struct dentry *dentry, char *buffer, unsigned int buflen,
 			const char *fmt, ...)
 {
 	va_list args;
@@ -284,7 +284,7 @@ char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
 	return memcpy(buffer, temp, sz);
 }
 
-char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
+char *simple_dname(struct dentry *dentry, char *buffer, unsigned int buflen)
 {
 	DECLARE_BUFFER(b, buffer, buflen);
 	/* these dentries are never renamed, so d_lock is not needed */
@@ -328,7 +328,7 @@ static char *__dentry_path(const struct dentry *d, struct prepend_buffer *p)
 	return extract_string(&b);
 }
 
-char *dentry_path_raw(const struct dentry *dentry, char *buf, int buflen)
+char *dentry_path_raw(const struct dentry *dentry, char *buf, unsigned int buflen)
 {
 	DECLARE_BUFFER(b, buf, buflen);
 
@@ -337,7 +337,7 @@ char *dentry_path_raw(const struct dentry *dentry, char *buf, int buflen)
 }
 EXPORT_SYMBOL(dentry_path_raw);
 
-char *dentry_path(const struct dentry *dentry, char *buf, int buflen)
+char *dentry_path(const struct dentry *dentry, char *buf, unsigned int buflen)
 {
 	DECLARE_BUFFER(b, buf, buflen);
 
include/linux/dcache.h +6 -6
index 9e23d33bb6f1..1a9838dc66fe 100644
@@ -296,13 +296,13 @@ static inline unsigned d_count(const struct dentry *dentry)
  * helper function for dentry_operations.d_dname() members
  */
 extern __printf(4, 5)
-char *dynamic_dname(struct dentry *, char *, int, const char *, ...);
+char *dynamic_dname(struct dentry *, char *, unsigned int, const char *, ...);
 
-extern char *__d_path(const struct path *, const struct path *, char *, int);
-extern char *d_absolute_path(const struct path *, char *, int);
-extern char *d_path(const struct path *, char *, int);
-extern char *dentry_path_raw(const struct dentry *, char *, int);
-extern char *dentry_path(const struct dentry *, char *, int);
+char *__d_path(const struct path *, const struct path *, char *, unsigned int);
+char *d_absolute_path(const struct path *, char *, unsigned int);
+char *d_path(const struct path *, char *, unsigned int);
+char *dentry_path_raw(const struct dentry *, char *, unsigned int);
+char *dentry_path(const struct dentry *, char *, unsigned int);
 
 /* Allocation counts.. */
 
Signature · 1 line
2.32.0
#1 lore raw
Quoted 4 lines · Greg Kroah-Hartman

Resolve all of the abuguity by just making "size" an unsigned value, which takes the guesswork out of everything involved.

Pardon my ignorance, but why not size_t instead of an unsigned int? I feel it will be more clear this way; but, yes, on 64-bit machines this will extend the buflen param to 64-bit.

Thanks,

--
Ahmed S. Darwish
Linutronix GmbH

#2 lore raw
Quoted 9 lines · Ahmed S. Darwish
Quoted 4 lines · Greg Kroah-Hartman

Resolve all of the abuguity by just making "size" an unsigned value, which takes the guesswork out of everything involved.

Pardon my ignorance, but why not size_t instead of an unsigned int? I feel it will be more clear this way; but, yes, on 64-bit machines this will extend the buflen param to 64-bit.

I have no objection moving it to size_t, but as you say, I don't think it's needed to make the buffer get that big.

thanks,

greg k-h

#3 lore raw
In reply to Greg Kroah-Hartman
Quoted 12 lines · Greg Kroah-Hartman

When running static analysis tools to find where signed values could potentially wrap the family of d_path() functions turn out to trigger a lot of mess. In evaluating the code, all of these usages seem safe, but pointer math is involved so if a negative number is ever somehow passed into these functions, memory can be traversed backwards in ways not intended.

fs/d_path.c +0 -0
index 23a53f7b5c71..7876b741a47e 100644

fs/d_path.c static int prepend_path(const struct path *path, · line 187

Quoted 7 lines
fs/d_path.c +1 -1
@@ -182,7 +182,7 @@ static int prepend_path(const struct path *path,
  */
 char *__d_path(const struct path *path,
 	       const struct path *root,
-	       char *buf, int buflen)
+	       char *buf, unsigned int buflen)
 {
 	DECLARE_BUFFER(b, buf, buflen);

I have questions about the quality of the analysis tool you're using.

struct prepend_buffer {
        char *buf;
        int len;
};
#define DECLARE_BUFFER(__name, __buf, __len) \
        struct prepend_buffer __name = {.buf = __buf + __len, .len = __len}

Why is it not flagging the assignment of an unsigned int buflen to a signed int len?

fs/d_path.c static int prepend_path(const struct path *path,

Quoted 5 lines
fs/d_path.c +5 -0
+char *__d_path(const struct path *, const struct path *, char *, unsigned int);
+char *d_absolute_path(const struct path *, char *, unsigned int);
+char *d_path(const struct path *, char *, unsigned int);
+char *dentry_path_raw(const struct dentry *, char *, unsigned int);
+char *dentry_path(const struct dentry *, char *, unsigned int);

While you're touching these declarations, please name the 'unsigned int' parameter. I don't care about the others; they are obvious, but an unsigned int might be flags, a length, or a small grey walrus.

#4 lore raw
Quoted 30 lines · Matthew Wilcox
Quoted 12 lines · Greg Kroah-Hartman

When running static analysis tools to find where signed values could potentially wrap the family of d_path() functions turn out to trigger a lot of mess. In evaluating the code, all of these usages seem safe, but pointer math is involved so if a negative number is ever somehow passed into these functions, memory can be traversed backwards in ways not intended.

fs/d_path.c +0 -0
index 23a53f7b5c71..7876b741a47e 100644

fs/d_path.c static int prepend_path(const struct path *path, · line 187

Quoted 7 lines
fs/d_path.c +1 -1
@@ -182,7 +182,7 @@ static int prepend_path(const struct path *path,
  */
 char *__d_path(const struct path *path,
 	       const struct path *root,
-	       char *buf, int buflen)
+	       char *buf, unsigned int buflen)
 {
 	DECLARE_BUFFER(b, buf, buflen);

I have questions about the quality of the analysis tool you're using.

struct prepend_buffer {
        char *buf;
        int len;
};
#define DECLARE_BUFFER(__name, __buf, __len) \
        struct prepend_buffer __name = {.buf = __buf + __len, .len = __len}
Quoted 2 lines

Why is it not flagging the assignment of an unsigned int buflen to a signed int len?

Ah, I could not run the tool after I made this change. I can change len in prepend_buffer as well.

Quoted 9 lines

fs/d_path.c static int prepend_path(const struct path *path,

Quoted 5 lines
fs/d_path.c +5 -0
+char *__d_path(const struct path *, const struct path *, char *, unsigned int);
+char *d_absolute_path(const struct path *, char *, unsigned int);
+char *d_path(const struct path *, char *, unsigned int);
+char *dentry_path_raw(const struct dentry *, char *, unsigned int);
+char *dentry_path(const struct dentry *, char *, unsigned int);

While you're touching these declarations, please name the 'unsigned int' parameter. I don't care about the others; they are obvious, but an unsigned int might be flags, a length, or a small grey walrus.

Sure, will respin this with both of those changes. Thanks for the review.

greg k-h

maintainer
#5 lore raw
In reply to Greg Kroah-Hartman
Quoted 9 lines · Greg Kroah-Hartman

When running static analysis tools to find where signed values could potentially wrap the family of d_path() functions turn out to trigger a lot of mess. In evaluating the code, all of these usages seem safe, but pointer math is involved so if a negative number is ever somehow passed into these functions, memory can be traversed backwards in ways not intended.

Resolve all of the abuguity by just making "size" an unsigned value, which takes the guesswork out of everything involved.

TBH, I'm not sure it's the right approach. Huge argument passed to d_path() is a bad idea, no matter what. Do you really want to have the damn thing try and fill 3Gb of buffer, all while holding rcu_read_lock() and a global spinlock or two? Hell, s/3Gb/1Gb/ and it won't get any better...

How about we do this instead:

d_path(const struct path *path, char *buf, int buflen)

{
	if (unlikely((unsigned)buflen > 0x8000)) {
		buf += (unsigned)buflen - 0x8000;
		buflen = 0x8000;
	}
	as in mainline
}

and take care of both issues?

#6 lore raw
Quoted 28 lines · Al Viro
Quoted 9 lines · Greg Kroah-Hartman

When running static analysis tools to find where signed values could potentially wrap the family of d_path() functions turn out to trigger a lot of mess. In evaluating the code, all of these usages seem safe, but pointer math is involved so if a negative number is ever somehow passed into these functions, memory can be traversed backwards in ways not intended.

Resolve all of the abuguity by just making "size" an unsigned value, which takes the guesswork out of everything involved.

TBH, I'm not sure it's the right approach. Huge argument passed to d_path() is a bad idea, no matter what. Do you really want to have the damn thing try and fill 3Gb of buffer, all while holding rcu_read_lock() and a global spinlock or two? Hell, s/3Gb/1Gb/ and it won't get any better...

How about we do this instead:

d_path(const struct path *path, char *buf, int buflen)

{
	if (unlikely((unsigned)buflen > 0x8000)) {
		buf += (unsigned)buflen - 0x8000;
		buflen = 0x8000;
	}
	as in mainline
}
Quoted 1 line

and take care of both issues?

umm ... what if someone passes in -ENOMEM as buflen? Not saying we have such a path right now, but I could imagine it happening.

	if (unlikely(buflen < 0))
		return ERR_PTR(buflen);
	if (unlikely(buflen > 0x8000)) {
		buf += buflen - 0x8000;
		buflen = 0x8000;
	}
	...
maintainer
#7 lore raw
Quoted 9 lines · Matthew Wilcox

umm ... what if someone passes in -ENOMEM as buflen? Not saying we have such a path right now, but I could imagine it happening.

	if (unlikely(buflen < 0))
		return ERR_PTR(buflen);
	if (unlikely(buflen > 0x8000)) {
		buf += buflen - 0x8000;
		buflen = 0x8000;
	}

Not really. You don't want ERR_PTR() of random negative numbers to start flying around...

#8 lore raw
Quoted 12 lines · Al Viro
Quoted 9 lines · Matthew Wilcox

umm ... what if someone passes in -ENOMEM as buflen? Not saying we have such a path right now, but I could imagine it happening.

	if (unlikely(buflen < 0))
		return ERR_PTR(buflen);
	if (unlikely(buflen > 0x8000)) {
		buf += buflen - 0x8000;
		buflen = 0x8000;
	}
Quoted 2 lines

Not really. You don't want ERR_PTR() of random negative numbers to start flying around...

yeah. the problem is that we're trying to infer what's actually going on when the user has (potentially) passed us complete crap. so do we assume that 'buffer' is good if 'buflen' is >32KB? plausible it might be. is it still plausibly good if buflen is >4MB? i would say 'no'.

	if (unlikely((unsigned)buflen > 4096U * 1024))
		return ERR_PTR(-EINVAL);
	if (unlikely(buflen > 0x8000)) {
		buf += buflen - 0x8000;
		buflen = 0x8000;
	}