Skip to contentSkip to conversation

headers_check cleanups break the whole world

RFC In review Kyle McMartin 6 participants 20 messages Feb 25, 2009 – Feb 26, 2009

Subsystems (8)
lore
Share & tools
lore
#0 lore raw

[names omitted to protect the innocent, hpa@ on the CC wrt klibc maybe
using these? ]

Hi,

Commits like

headers_check fix: foo.h

fix the following 'make headers_check' warnings:
usr/include/linux/foo.h:29: include of <linux/types.h> is preferred usr/include/linux/foo.h:102: found __[us]{8,16,32,64} type without

have proved problematic...

I've had to point out at least two userspace fixes[1] for a variety of reasons that these patches exacerbated. Note however that I didn't say they were wrong.

The reason for this is you cannot intermix glibc header <sys/*.h> includes with <linux/*.h> includes for most things without defining the __KERNEL_STRICT_NAMES guard. If you fail to define this, you end up with multiple definitions of things like dev_t.

Software was able to get by, because things that used the headers, dvb for example were not getting <linux/types.h> into the include chain, because they were using <asm/types.h> directly.

I propose we invert that logic, so the presumable libc that makes use of the <linux/types.h> header can just define that it wants these types. (test __KERNEL__ as well so the kernel doesn't need a pointless #define.)

If this isn't tenable, how about moving the {,__}[su]{8,16,32,64} integer types into their own header, so we can avoid this mess ever occuring in the future. I'm sure the janitors can have a field day with that... :)

That said, who exactly is the userspace consumer for those

	typedef __kernel_dev_t	dev_t;

defines? Can we just include them all in #ifdef __KERNEL__?

Thoughts?

cheers, Kyle

1. Ok, one of them was libcap playing utterly stupid games with <linux/capability.h> and header guards, but it was exacerbated by a similar patch...

#1 lore raw
Quoted 6 lines · Kyle McMartin

The reason for this is you cannot intermix glibc header <sys/*.h> includes with <linux/*.h> includes for most things without defining the __KERNEL_STRICT_NAMES guard. If you fail to define this, you end up with multiple definitions of things like dev_t.

Quoted 9 lines

Software was able to get by, because things that used the headers, dvb for example were not getting <linux/types.h> into the include chain, because they were using <asm/types.h> directly.

I propose we invert that logic, so the presumable libc that makes use of the <linux/types.h> header can just define that it wants these types. (test __KERNEL__ as well so the kernel doesn't need a pointless #define.)

The Right Thing[TM] here is to change the exported headers so that *only* the __kernel* names are exported, and then remove the non-__KERNEL_STRICT_NAMES version of <linux/types.h>. The *only* libc for which non-__KERNEL_STRICT_NAMES ever made sense was for the long-since-obsolete libc5.

This, by the way, is not frivolous work. The work that has been done in this area already has smoked out a number of bugs where exported headers were using user space types, which meant they were using the *WRONG TYPES* on glibc. Completely.

Quoted 8 lines

If this isn't tenable, how about moving the {,__}[su]{8,16,32,64} integer types into their own header, so we can avoid this mess ever occuring in the future. I'm sure the janitors can have a field day with that... :)

That said, who exactly is the userspace consumer for those

	typedef __kernel_dev_t	dev_t;

defines? Can we just include them all in #ifdef __KERNEL__?

That is exactly what we should do.

For what it's worth, not even klibc uses these types.

-hpa

Signature · 2 lines
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.
#2 lore raw
In reply to Kyle McMartin
Quoted 45 lines · Kyle McMartin

[names omitted to protect the innocent, hpa@ on the CC wrt klibc maybe
using these? ]

Hi,

Commits like

headers_check fix: foo.h

fix the following 'make headers_check' warnings:
usr/include/linux/foo.h:29: include of <linux/types.h> is preferred usr/include/linux/foo.h:102: found __[us]{8,16,32,64} type without

have proved problematic...

I've had to point out at least two userspace fixes[1] for a variety of reasons that these patches exacerbated. Note however that I didn't say they were wrong.

The reason for this is you cannot intermix glibc header <sys/*.h> includes with <linux/*.h> includes for most things without defining the __KERNEL_STRICT_NAMES guard. If you fail to define this, you end up with multiple definitions of things like dev_t.

Software was able to get by, because things that used the headers, dvb for example were not getting <linux/types.h> into the include chain, because they were using <asm/types.h> directly.

I propose we invert that logic, so the presumable libc that makes use of the <linux/types.h> header can just define that it wants these types. (test __KERNEL__ as well so the kernel doesn't need a pointless
#define.)

If this isn't tenable, how about moving the {,__}[su]{8,16,32,64} integer types into their own header, so we can avoid this mess ever occuring in the future. I'm sure the janitors can have a field day with that... :)

That said, who exactly is the userspace consumer for those

	typedef __kernel_dev_t	dev_t;

defines? Can we just include them all in #ifdef __KERNEL__?

Quoted 7 lines

Thoughts?

cheers, Kyle

1. Ok, one of them was libcap playing utterly stupid games with <linux/capability.h> and header guards, but it was exacerbated by a similar patch...

Well, the intention is to clean up the situation somewhat.

__KERNEL_STRICT_NAMES is a really old construct that has been with us forever. It's not widely used ... i dont know how widely it's being relied on. Sam, should we get rid of it, or should user-space define __KERNEL_STRICT_NAMES in cases the glibc definition collides with the kernel's definition?

Note that if user-space is "playing utterly stupid games", it can cause trouble no matter what scheme we pick - so we have to filter out the reasonable problems that we should and can fix in the kernel.

Ingo

#3 lore raw
Quoted 9 lines · Ingo Molnar

Well, the intention is to clean up the situation somewhat.

__KERNEL_STRICT_NAMES is a really old construct that has been with us forever. It's not widely used ... i dont know how widely it's being relied on. Sam, should we get rid of it, or should user-space define __KERNEL_STRICT_NAMES in cases the glibc definition collides with the kernel's definition?

Quoted 5 lines

Note that if user-space is "playing utterly stupid games", it can cause trouble no matter what scheme we pick - so we have to filter out the reasonable problems that we should and can fix in the kernel.

__KERNEL_STRICT_NAMES is an anachronism that was put in to not break libc5. It has long outlived its usefulness, together with all the other libc5 support crap in the kernel headers -- which do nothing but make the kernel headers useless for any sane purposes.

Please let's just axe it.

-hpa

Signature · 2 lines
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.
#4 lore raw
In reply to Ingo Molnar
Quoted 5 lines · Ingo Molnar

Note that if user-space is "playing utterly stupid games", it can cause trouble no matter what scheme we pick - so we have to filter out the reasonable problems that we should and can fix in the kernel.

Yeah, sorry, I should have been clearer, libcap does stupid #define _LINUX_TYPES_H_
typedef unsigned int __u32;
#include <linux/header.h>

Which is just horribly broken when it's included early in a file and then we try to #include <asm/sigcontext.h> (and doesn't get the rest of those types since it now includes <linux/types.h> instead of <asm/types.h>) (coreutils was doing this.)

The real case is something like using the dvb headers, which legitimately is trying to include both <sys/*.h> and then <linux/dvb/*.h> and expecting it to work.

The footnote in the prior mail was really just an example of why not having just the plain __u32 et al types in their own header.

regards, Kyle

#5 lore raw
Quoted 11 lines · Kyle McMartin

Yeah, sorry, I should have been clearer, libcap does stupid #define _LINUX_TYPES_H_
typedef unsigned int __u32;
#include <linux/header.h>

Which is just horribly broken when it's included early in a file and then we try to #include <asm/sigcontext.h> (and doesn't get the rest of those types since it now includes <linux/types.h> instead of <asm/types.h>) (coreutils was doing this.)

Quoted 7 lines

The real case is something like using the dvb headers, which legitimately is trying to include both <sys/*.h> and then <linux/dvb/*.h> and expecting it to work.

The footnote in the prior mail was really just an example of why not having just the plain __u32 et al types in their own header.

Because <linux/types.h> should work just fine once we get rid of the __KERNEL_STRICT_NAMES idiocy?

FWIW, Arnd Bergmann has been working on exactly this cleanup.

-hpa

Signature · 2 lines
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.
#6 lore raw
Quoted 5 lines · H. Peter Anvin

Because <linux/types.h> should work just fine once we get rid of the __KERNEL_STRICT_NAMES idiocy?

FWIW, Arnd Bergmann has been working on exactly this cleanup.

Ah good, I was about to cook a patch, then I realized there's probably quite a mess of userspace things depending on, say, sector_t and such.

This is a pretty ugly problem, all things considered. :/

regards, Kyle

#7 lore raw
Quoted 4 lines · Kyle McMartin
Quoted 3 lines · H. Peter Anvin

Because <linux/types.h> should work just fine once we get rid of the __KERNEL_STRICT_NAMES idiocy?

Quoted 8 lines
Quoted 2 lines

FWIW, Arnd Bergmann has been working on exactly this cleanup.

Ah good, I was about to cook a patch, then I realized there's probably quite a mess of userspace things depending on, say, sector_t and such.

This is a pretty ugly problem, all things considered. :/

Not all that bad, actually.

This is the last patch I've seen on the subject:

http://marc.info/[email protected]

-hpa

Signature · 2 lines
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.
#8 lore raw
Quoted 6 lines · H. Peter Anvin

Not all that bad, actually.

This is the last patch I've seen on the subject:

http://marc.info/[email protected]

Ah, excellent. I must have missed it (looks like Arnd didn't send it to linux-arch@ as well. :)

Big ACK from me.

regards, Kyle

maintainer
#9 lore raw

PATCH 1/1 Make exported headers use strict posix types

A number of standard posix types are used in exported headers, which is not allowed if __STRICT_KERNEL_NAMES is defined. In order to get rid of the non-__STRICT_KERNEL_NAMES part and to make sane headers the default, we have to change them all to safe types.

There are also still some leftovers in reiserfs_fs.h, elfcore.h and coda.h, but these files have not compiled in user space for a long time.

This leaves out the various integer types (u_int32_t, uint32_t, u32, uint and the signed and 8/16/64 bit variants thereof), which we may want to do in an automated script instead.

---

Quoted 10 lines · Kyle McMartin
Quoted 6 lines · H. Peter Anvin

Not all that bad, actually.

This is the last patch I've seen on the subject:

http://marc.info/[email protected]

Ah, excellent. I must have missed it (looks like Arnd didn't send it to linux-arch@ as well. :)

This is my previous patch, ported to today's git. It's my last day before my vacation, so I can't do the usual testing etc. on it. Hopefully somebody else can take care of this.

	Arnd <><
include/asm-generic/fcntl.h +6 -6
index b847741..4d3e483 100644
@@ -117,9 +117,9 @@
 struct flock {
 	short	l_type;
 	short	l_whence;
-	off_t	l_start;
-	off_t	l_len;
-	pid_t	l_pid;
+	__kernel_off_t	l_start;
+	__kernel_off_t	l_len;
+	__kernel_pid_t	l_pid;
 	__ARCH_FLOCK_PAD
 };
 #endif
@@ -140,9 +140,9 @@ struct flock {
 struct flock64 {
 	short  l_type;
 	short  l_whence;
-	loff_t l_start;
-	loff_t l_len;
-	pid_t  l_pid;
+	__kernel_loff_t l_start;
+	__kernel_loff_t l_len;
+	__kernel_pid_t  l_pid;
 	__ARCH_FLOCK64_PAD
 };
 #endif
include/asm-generic/siginfo.h +7 -7
index 9695701..35f75a3 100644
@@ -23,7 +23,7 @@ typedef union sigval {
 #endif
 
 #ifndef __ARCH_SI_UID_T
-#define __ARCH_SI_UID_T	uid_t
+#define __ARCH_SI_UID_T	__kernel_uid_t
 #endif
 
 /*
@@ -47,13 +47,13 @@ typedef struct siginfo {
 
 		/* kill() */
 		struct {
-			pid_t _pid;		/* sender's pid */
+			__kernel_pid_t _pid;	/* sender's pid */
 			__ARCH_SI_UID_T _uid;	/* sender's uid */
 		} _kill;
 
 		/* POSIX.1b timers */
 		struct {
-			timer_t _tid;		/* timer id */
+			__kernel_timer_t _tid;	/* timer id */
 			int _overrun;		/* overrun count */
 			char _pad[sizeof( __ARCH_SI_UID_T) - sizeof(int)];
 			sigval_t _sigval;	/* same as below */
@@ -62,18 +62,18 @@ typedef struct siginfo {
 
 		/* POSIX.1b signals */
 		struct {
-			pid_t _pid;		/* sender's pid */
+			__kernel_pid_t _pid;	/* sender's pid */
 			__ARCH_SI_UID_T _uid;	/* sender's uid */
 			sigval_t _sigval;
 		} _rt;
 
 		/* SIGCHLD */
 		struct {
-			pid_t _pid;		/* which child */
+			__kernel_pid_t _pid;	/* which child */
 			__ARCH_SI_UID_T _uid;	/* sender's uid */
 			int _status;		/* exit code */
-			clock_t _utime;
-			clock_t _stime;
+			__kernel_clock_t _utime;
+			__kernel_clock_t _stime;
 		} _sigchld;
 
 		/* SIGILL, SIGFPE, SIGSEGV, SIGBUS */
include/linux/agpgart.h +7 -7
index 110c600..f6778ec 100644
@@ -77,20 +77,20 @@ typedef struct _agp_setup {
  * The "prot" down below needs still a "sleep" flag somehow ...
  */
 typedef struct _agp_segment {
-	off_t pg_start;		/* starting page to populate    */
-	size_t pg_count;	/* number of pages              */
-	int prot;		/* prot flags for mmap          */
+	__kernel_off_t pg_start;	/* starting page to populate    */
+	__kernel_size_t pg_count;	/* number of pages              */
+	int prot;			/* prot flags for mmap          */
 } agp_segment;
 
 typedef struct _agp_region {
-	pid_t pid;		/* pid of process               */
-	size_t seg_count;	/* number of segments           */
+	__kernel_pid_t pid;		/* pid of process       */
+	__kernel_size_t seg_count;	/* number of segments   */
 	struct _agp_segment *seg_list;
 } agp_region;
 
 typedef struct _agp_allocate {
 	int key;		/* tag of allocation            */
-	size_t pg_count;	/* number of pages              */
+	__kernel_size_t pg_count;/* number of pages             */
 	__u32 type;		/* 0 == normal, other devspec   */
    	__u32 physical;         /* device specific (some devices  
 				 * need a phys address of the     
@@ -100,7 +100,7 @@ typedef struct _agp_allocate {
 
 typedef struct _agp_bind {
 	int key;		/* tag of allocation            */
-	off_t pg_start;		/* starting page to populate    */
+	__kernel_off_t pg_start;/* starting page to populate    */
 } agp_bind;
 
 typedef struct _agp_unbind {
include/linux/cn_proc.h +10 -10
index 1c86d65..b8125b2 100644
@@ -65,20 +65,20 @@ struct proc_event {
 		} ack;
 
 		struct fork_proc_event {
-			pid_t parent_pid;
-			pid_t parent_tgid;
-			pid_t child_pid;
-			pid_t child_tgid;
+			__kernel_pid_t parent_pid;
+			__kernel_pid_t parent_tgid;
+			__kernel_pid_t child_pid;
+			__kernel_pid_t child_tgid;
 		} fork;
 
 		struct exec_proc_event {
-			pid_t process_pid;
-			pid_t process_tgid;
+			__kernel_pid_t process_pid;
+			__kernel_pid_t process_tgid;
 		} exec;
 
 		struct id_proc_event {
-			pid_t process_pid;
-			pid_t process_tgid;
+			__kernel_pid_t process_pid;
+			__kernel_pid_t process_tgid;
 			union {
 				__u32 ruid; /* task uid */
 				__u32 rgid; /* task gid */
@@ -90,8 +90,8 @@ struct proc_event {
 		} id;
 
 		struct exit_proc_event {
-			pid_t process_pid;
-			pid_t process_tgid;
+			__kernel_pid_t process_pid;
+			__kernel_pid_t process_tgid;
 			__u32 exit_code, exit_signal;
 		} exit;
 	} event_data;
include/linux/coda_psdev.h +2 -0
index 07ae8f8..ebdc055 100644
@@ -6,6 +6,7 @@
 #define CODA_PSDEV_MAJOR 67
 #define MAX_CODADEVS  5	   /* how many do we allow */
 
+#ifdef __KERNEL__
 struct kstatfs;
 
 /* communication pending/processing queues */
@@ -89,4 +90,5 @@ struct upc_req {
 
 extern struct venus_comm coda_comms[];
 
+#endif /* __KERNEL__ */
 #endif
include/linux/compiler-gcc4.h +1 -1
index 0999271..f6dcc29 100644
@@ -4,7 +4,7 @@
 
 /* GCC 4.1.[01] miscompiles __weak */
 #if __GNUC_MINOR__ == 1 && __GNUC_PATCHLEVEL__ <= 1
-# error Your version of gcc miscompiles the __weak directive
+// # error Your version of gcc miscompiles the __weak directive
 #endif
 
 #define __used			__attribute__((__used__))
include/linux/cyclades.h +3 -3
index d06fbf2..788850b 100644
@@ -82,9 +82,9 @@ struct cyclades_monitor {
  * open)
  */
 struct cyclades_idle_stats {
-    time_t	   in_use;	/* Time device has been in use (secs) */
-    time_t	   recv_idle;	/* Time since last char received (secs) */
-    time_t	   xmit_idle;	/* Time since last char transmitted (secs) */
+    __kernel_time_t in_use;	/* Time device has been in use (secs) */
+    __kernel_time_t recv_idle;	/* Time since last char received (secs) */
+    __kernel_time_t xmit_idle;	/* Time since last char transmitted (secs) */
     unsigned long  recv_bytes;	/* Bytes received */
     unsigned long  xmit_bytes;	/* Bytes transmitted */
     unsigned long  overruns;	/* Input overruns */
include/linux/dvb/video.h +1 -1
index bd49c3e..ee5d2df 100644
@@ -137,7 +137,7 @@ struct video_event {
 #define VIDEO_EVENT_FRAME_RATE_CHANGED	2
 #define VIDEO_EVENT_DECODER_STOPPED 	3
 #define VIDEO_EVENT_VSYNC 		4
-	time_t timestamp;
+	__kernel_time_t timestamp;
 	union {
 		video_size_t size;
 		unsigned int frame_rate;	/* in frames per 1000sec */
include/linux/if_pppol2tp.h +1 -1
index c7a6688..3a14b08 100644
@@ -26,7 +26,7 @@
  */
 struct pppol2tp_addr
 {
-	pid_t	pid;			/* pid that owns the fd.
+	__kernel_pid_t	pid;		/* pid that owns the fd.
 					 * 0 => current */
 	int	fd;			/* FD of UDP socket to use */
 
include/linux/mroute6.h +1 -1
index 5375fac..43dc97e 100644
@@ -65,7 +65,7 @@ struct mif6ctl {
 	mifi_t	mif6c_mifi;		/* Index of MIF */
 	unsigned char mif6c_flags;	/* MIFF_ flags */
 	unsigned char vifc_threshold;	/* ttl limit */
-	u_short	 mif6c_pifi;		/* the index of the physical IF */
+	__u16	 mif6c_pifi;		/* the index of the physical IF */
 	unsigned int vifc_rate_limit;	/* Rate limiter values (NI) */
 };
 
include/linux/netfilter_ipv4/ipt_owner.h +4 -4
index 92f4bda..60e941f 100644
@@ -9,10 +9,10 @@
 #define IPT_OWNER_COMM	0x10
 
 struct ipt_owner_info {
-    uid_t uid;
-    gid_t gid;
-    pid_t pid;
-    pid_t sid;
+    __kernel_uid_t uid;
+    __kernel_gid_t gid;
+    __kernel_pid_t pid;
+    __kernel_pid_t sid;
     char comm[16];
     u_int8_t match, invert;	/* flags */
 };
include/linux/netfilter_ipv6/ip6t_owner.h +4 -4
index 19937da..dc2cbcb 100644
@@ -8,10 +8,10 @@
 #define IP6T_OWNER_SID	0x08
 
 struct ip6t_owner_info {
-    uid_t uid;
-    gid_t gid;
-    pid_t pid;
-    pid_t sid;
+    __kernel_uid_t uid;
+    __kernel_gid_t gid;
+    __kernel_pid_t pid;
+    __kernel_pid_t sid;
     u_int8_t match, invert;	/* flags */
 };
 
include/linux/ppp_defs.h +2 -2
index 1c866bd..0f93ed6 100644
@@ -177,8 +177,8 @@ struct ppp_comp_stats {
  * the last NP packet was sent or received.
  */
 struct ppp_idle {
-    time_t xmit_idle;		/* time since last NP packet sent */
-    time_t recv_idle;		/* time since last NP packet received */
+    __kernel_time_t xmit_idle;	/* time since last NP packet sent */
+    __kernel_time_t recv_idle;	/* time since last NP packet received */
 };
 
 #endif /* _PPP_DEFS_H_ */
include/linux/suspend_ioctls.h +5 -5
index 2c6faec..40d4605 100644
@@ -7,8 +7,8 @@
  * SNAPSHOT_SET_SWAP_AREA ioctl
  */
 struct resume_swap_area {
-	loff_t offset;
-	u_int32_t dev;
+	__kernel_loff_t offset;
+	__u32 dev;
 } __attribute__((packed));
 
 #define SNAPSHOT_IOC_MAGIC	'3'
@@ -20,13 +20,13 @@ struct resume_swap_area {
 #define SNAPSHOT_S2RAM			_IO(SNAPSHOT_IOC_MAGIC, 11)
 #define SNAPSHOT_SET_SWAP_AREA		_IOW(SNAPSHOT_IOC_MAGIC, 13, \
 							struct resume_swap_area)
-#define SNAPSHOT_GET_IMAGE_SIZE		_IOR(SNAPSHOT_IOC_MAGIC, 14, loff_t)
+#define SNAPSHOT_GET_IMAGE_SIZE		_IOR(SNAPSHOT_IOC_MAGIC, 14, __kernel_loff_t)
 #define SNAPSHOT_PLATFORM_SUPPORT	_IO(SNAPSHOT_IOC_MAGIC, 15)
 #define SNAPSHOT_POWER_OFF		_IO(SNAPSHOT_IOC_MAGIC, 16)
 #define SNAPSHOT_CREATE_IMAGE		_IOW(SNAPSHOT_IOC_MAGIC, 17, int)
 #define SNAPSHOT_PREF_IMAGE_SIZE	_IO(SNAPSHOT_IOC_MAGIC, 18)
-#define SNAPSHOT_AVAIL_SWAP_SIZE	_IOR(SNAPSHOT_IOC_MAGIC, 19, loff_t)
-#define SNAPSHOT_ALLOC_SWAP_PAGE	_IOR(SNAPSHOT_IOC_MAGIC, 20, loff_t)
+#define SNAPSHOT_AVAIL_SWAP_SIZE	_IOR(SNAPSHOT_IOC_MAGIC, 19, __kernel_loff_t)
+#define SNAPSHOT_ALLOC_SWAP_PAGE	_IOR(SNAPSHOT_IOC_MAGIC, 20, __kernel_loff_t)
 #define SNAPSHOT_IOC_MAXNR	20
 
 #endif /* _LINUX_SUSPEND_IOCTLS_H */
include/linux/time.h +4 -4
index fbbd2a1..242f624 100644
@@ -12,14 +12,14 @@
 #ifndef _STRUCT_TIMESPEC
 #define _STRUCT_TIMESPEC
 struct timespec {
-	time_t	tv_sec;		/* seconds */
-	long	tv_nsec;	/* nanoseconds */
+	__kernel_time_t	tv_sec;			/* seconds */
+	long		tv_nsec;		/* nanoseconds */
 };
 #endif
 
 struct timeval {
-	time_t		tv_sec;		/* seconds */
-	suseconds_t	tv_usec;	/* microseconds */
+	__kernel_time_t		tv_sec;		/* seconds */
+	__kernel_suseconds_t	tv_usec;	/* microseconds */
 };
 
 struct timezone {
include/linux/times.h +4 -4
index e2d3020..87b6261 100644
@@ -4,10 +4,10 @@
 #include <linux/types.h>
 
 struct tms {
-	clock_t tms_utime;
-	clock_t tms_stime;
-	clock_t tms_cutime;
-	clock_t tms_cstime;
+	__kernel_clock_t tms_utime;
+	__kernel_clock_t tms_stime;
+	__kernel_clock_t tms_cutime;
+	__kernel_clock_t tms_cstime;
 };
 
 #endif
include/linux/utime.h +2 -2
index 640be6a..5cdf673 100644
@@ -4,8 +4,8 @@
 #include <linux/types.h>
 
 struct utimbuf {
-	time_t actime;
-	time_t modtime;
+	__kernel_time_t actime;
+	__kernel_time_t modtime;
 };
 
 #endif
include/linux/xfrm.h +1 -1
index 52f3abd..a58fe2b 100644
@@ -58,7 +58,7 @@ struct xfrm_selector
 	__u8	prefixlen_s;
 	__u8	proto;
 	int	ifindex;
-	uid_t	user;
+	__kernel_uid_t	user;
 };
 
 #define XFRM_INF (~(__u64)0)
include/mtd/mtd-abi.h +2 -2
index c6c61cd..fb67201 100644
@@ -84,8 +84,8 @@ struct otp_info {
 #define MEMGETREGIONINFO	_IOWR('M', 8, struct region_info_user)
 #define MEMSETOOBSEL		_IOW('M', 9, struct nand_oobinfo)
 #define MEMGETOOBSEL		_IOR('M', 10, struct nand_oobinfo)
-#define MEMGETBADBLOCK		_IOW('M', 11, loff_t)
-#define MEMSETBADBLOCK		_IOW('M', 12, loff_t)
+#define MEMGETBADBLOCK		_IOW('M', 11, __kernel_loff_t)
+#define MEMSETBADBLOCK		_IOW('M', 12, __kernel_loff_t)
 #define OTPSELECT		_IOR('M', 13, int)
 #define OTPGETREGIONCOUNT	_IOW('M', 14, int)
 #define OTPGETREGIONINFO	_IOW('M', 15, struct otp_info)
include/sound/asound.h +2 -2
index 1c02ed1..16684c5 100644
@@ -385,7 +385,7 @@ struct snd_pcm_sw_params {
 
 struct snd_pcm_channel_info {
 	unsigned int channel;
-	off_t offset;			/* mmap offset */
+	__kernel_off_t offset;		/* mmap offset */
 	unsigned int first;		/* offset to first sample in bits */
 	unsigned int step;		/* samples distance in bits */
 };
@@ -789,7 +789,7 @@ struct snd_ctl_elem_info {
 	snd_ctl_elem_type_t type;	/* R: value type - SNDRV_CTL_ELEM_TYPE_* */
 	unsigned int access;		/* R: value access (bitmask) - SNDRV_CTL_ELEM_ACCESS_* */
 	unsigned int count;		/* count of values */
-	pid_t owner;			/* owner's PID of this control */
+	__kernel_pid_t owner;		/* owner's PID of this control */
 	union {
 		struct {
 			long min;		/* R: minimum value */
#10 lore raw

Re: [PATCH] Make exported headers use strict posix types

Quoted 13 lines · Arnd Bergmann

A number of standard posix types are used in exported headers, which is not allowed if __STRICT_KERNEL_NAMES is defined. In order to get rid of the non-__STRICT_KERNEL_NAMES part and to make sane headers the default, we have to change them all to safe types.

There are also still some leftovers in reiserfs_fs.h, elfcore.h and coda.h, but these files have not compiled in user space for a long time.

This leaves out the various integer types (u_int32_t, uint32_t, u32, uint and the signed and 8/16/64 bit variants thereof), which we may want to do in an automated script instead.

Quoted 6 lines

---

This is my previous patch, ported to today's git. It's my last day before my vacation, so I can't do the usual testing etc. on it. Hopefully somebody else can take care of this.

I hesitate to put it in kbuild.git... Any better place?

Quoted 13 lines
include/linux/coda_psdev.h +1 -0
index 07ae8f8..ebdc055 100644
@@ -6,6 +6,7 @@
 #define CODA_PSDEV_MAJOR 67
 #define MAX_CODADEVS  5	   /* how many do we allow */
 
+#ifdef __KERNEL__
 struct kstatfs;
 
 /* communication pending/processing queues */

include/linux/coda_psdev.h struct upc_req { · line 94

Quoted 5 lines
include/linux/coda_psdev.h +1 -0
@@ -89,4 +90,5 @@ struct upc_req {
 
 extern struct venus_comm coda_comms[];
 
+#endif /* __KERNEL__ */
 #endif

Does this change really belong in this patchset?

include/linux/compiler-gcc4.h line 10

Quoted 13 lines
include/linux/compiler-gcc4.h +1 -1
index 0999271..f6dcc29 100644
@@ -4,7 +4,7 @@
 
 /* GCC 4.1.[01] miscompiles __weak */
 #if __GNUC_MINOR__ == 1 && __GNUC_PATCHLEVEL__ <= 1
-# error Your version of gcc miscompiles the __weak directive
+// # error Your version of gcc miscompiles the __weak directive
 #endif
 
 #define __used			__attribute__((__used__))

Yes - I can build kernels again. But I think we should skip this change..

include/linux/mroute6.h struct mif6ctl { · line 71

Quoted 13 lines
include/linux/mroute6.h +1 -1
index 5375fac..43dc97e 100644
@@ -65,7 +65,7 @@ struct mif6ctl {
 	mifi_t	mif6c_mifi;		/* Index of MIF */
 	unsigned char mif6c_flags;	/* MIFF_ flags */
 	unsigned char vifc_threshold;	/* ttl limit */
-	u_short	 mif6c_pifi;		/* the index of the physical IF */
+	__u16	 mif6c_pifi;		/* the index of the physical IF */
 	unsigned int vifc_rate_limit;	/* Rate limiter values (NI) */
 };
 

This change looks correct - but I assume this is a separate patch.

Rest looks good.

Sam

maintainer
#11 lore raw

Re: [PATCH] Make exported headers use strict posix types

Quoted 7 lines · Sam Ravnborg

Does this change really belong in this patchset?

Yes - I can build kernels again. But I think we should skip this change..

This change looks correct - but I assume this is a separate patch.

Rest looks good.

Agreed on all points, I had the changes sitting in a dirty git tree and got sloppy. Thanks for looking through it!

	Arnd <><
maintainer
#12 lore raw

Re: [PATCH] Make exported headers use strict posix types

In reply to Sam Ravnborg
Quoted 6 lines · Sam Ravnborg
Quoted 6 lines
include/linux/mroute6.h +0 -0
index 5375fac..43dc97e 100644
@@ -65,7 +65,7 @@ struct mif6ctl {

      mifi_t  mif6c_mifi;             /* Index of MIF */

Quoted 9 lines
Quoted 7 lines

      unsigned char mif6c_flags;      /* MIFF_ flags */ unsigned char vifc_threshold;   /* ttl limit */
-     u_short  mif6c_pifi;            /* the index of the physical IF */ +     __u16    mif6c_pifi;            /* the index of the physical IF */ unsigned int vifc_rate_limit;   /* Rate limiter values (NI) */ };

This change looks correct - but I assume this is a separate patch.

Actually, this change does belong in here: This is the only user of u_short in exported headers, and nothing uses u_char, u_int or u_long, so it gets rid of a whole class of types. The only questionable types left after this are the standard stdint.h types, many of which are probably legitimate:

grep '\<\(u_\|u\|\)int\(8\|16\|32\|64\)_t\>' obj/usr/include/ -r | \
		 cut -f 1 -d: | uniq -c | grep -v netfilter
      2 obj/usr/include/linux/atmlec.h
     13 obj/usr/include/linux/jffs2.h
      1 obj/usr/include/linux/ivtvfb.h
     12 obj/usr/include/linux/types.h
    121 obj/usr/include/linux/pfkeyv2.h
     13 obj/usr/include/linux/ip_vs.h
     21 obj/usr/include/linux/dm-ioctl.h
      4 obj/usr/include/linux/cm4000_cs.h
      8 obj/usr/include/linux/dlm_netlink.h
      2 obj/usr/include/linux/selinux_netlink.h
     11 obj/usr/include/linux/dvb/video.h
      1 obj/usr/include/linux/dvb/audio.h
     22 obj/usr/include/linux/atmmpc.h
     23 obj/usr/include/linux/coda.h
      1 obj/usr/include/linux/matroxfb.h
     13 obj/usr/include/linux/if_arcnet.h
      8 obj/usr/include/drm/mga_drm.h
      1 obj/usr/include/drm/radeon_drm.h
     20 obj/usr/include/drm/via_drm.h
      9 obj/usr/include/drm/drm.h
     74 obj/usr/include/drm/drm_mode.h
     69 obj/usr/include/drm/i915_drm.h
     15 obj/usr/include/mtd/nftl-user.h
      2 obj/usr/include/mtd/jffs2-user.h
     18 obj/usr/include/mtd/inftl-user.h
     32 obj/usr/include/mtd/mtd-abi.h
     35 obj/usr/include/mtd/ubi-user.h
      4 obj/usr/include/sound/emu10k1.h
      1 obj/usr/include/sound/asound.h

	Arnd <><
#13 lore raw

Re: [PATCH] Make exported headers use strict posix types

Quoted 6 lines · Arnd Bergmann

Actually, this change does belong in here: This is the only user of u_short in exported headers, and nothing uses u_char, u_int or u_long, so it gets rid of a whole class of types. The only questionable types left after this are the standard stdint.h types, many of which are probably legitimate:

Not legitimate in *exported* headers.

-hpa

#14 lore raw

Re: [PATCH] Make exported headers use strict posix types

Quoted 9 lines · H. Peter Anvin
Quoted 6 lines · Arnd Bergmann

Actually, this change does belong in here: This is the only user of u_short in exported headers, and nothing uses u_char, u_int or u_long, so it gets rid of a whole class of types. The only questionable types left after this are the standard stdint.h types, many of which are probably legitimate:

Not legitimate in *exported* headers.

Not in the ones which might be included implicitly by glibc, perhaps -- but other non-POSIX headers are perfectly entitled to require C standard types, surely?

Signature · 2 lines
David Woodhouse                            Open Source Technology Centre
[email protected]                              Intel Corporation
#15 lore raw

Re: [PATCH] Make exported headers use strict posix types

Quoted 5 lines · David Woodhouse
Quoted 1 line

Not legitimate in *exported* headers.

Not in the ones which might be included implicitly by glibc, perhaps -- but other non-POSIX headers are perfectly entitled to require C standard types, surely?

It doesn't make any sense, though, when there is a straighforward way to deal without.

Now, for source code maintainability reasons I still believe this should be done mechanically instead of having to edit the source all over the place.

-hpa

#16 lore raw

Re: [PATCH] Make exported headers use strict posix types

In reply to Sam Ravnborg
Quoted 7 lines · Sam Ravnborg
Quoted 4 lines

This is my previous patch, ported to today's git. It's my last day before my vacation, so I can't do the usual testing etc. on it. Hopefully somebody else can take care of this.

I hesitate to put it in kbuild.git... Any better place?

I'll put it in tip:core/headers-fixes.

-hpa

maintainer
#17 lore raw

Re: [PATCH] Make exported headers use strict posix types

Quoted 10 lines · H. Peter Anvin
Quoted 7 lines · Sam Ravnborg
Quoted 4 lines

This is my previous patch, ported to today's git. It's my last day before my vacation, so I can't do the usual testing etc. on it. Hopefully somebody else can take care of this.

I hesitate to put it in kbuild.git... Any better place?

I'll put it in tip:core/headers-fixes.

Just in the minute I notice a bug :(

Unfortunately, I mistakenly replaced uid_t with __kernel_uid_t when it had to be __kernel_uid32_t. New series follows, including the changes for all integer types.

	Arnd <><
#18 lore raw

Re: [PATCH] Make exported headers use strict posix types

Quoted 7 lines · Arnd Bergmann

Just in the minute I notice a bug :(

Unfortunately, I mistakenly replaced uid_t with __kernel_uid_t when it had to be __kernel_uid32_t. New series follows, including the changes for all integer types.

S'okay... it's easy enough to back out at this time.

-hpa

#19 lore raw
In reply to Ingo Molnar
Quoted 6 lines

__KERNEL_STRICT_NAMES is a really old construct that has been with us forever. It's not widely used ... i dont know how widely it's being relied on. Sam, should we get rid of it, or should user-space define __KERNEL_STRICT_NAMES in cases the glibc definition collides with the kernel's definition?

hpa already answered this question and he knows
the userland<->kernel stuff better than I do.

IMO the best thing would be that Arnd rebased his original patch and we get that rewieved and in the end applied.

Sam