Re: git: 47a57144af25 - main - cpuset: Byte swap cpuset for compat32 on big endian architectures

From: Dmitry Chagin <dchagin_at_freebsd.org>
Date: Thu, 19 May 2022 16:49:39 UTC
On Thu, May 19, 2022 at 03:48:42PM +0000, Justin Hibbits wrote:
> The branch main has been updated by jhibbits:
> 
> URL: https://cgit.FreeBSD.org/src/commit/?id=47a57144af25a7bd768b29272d50a36fdf2874ba
> 
> commit 47a57144af25a7bd768b29272d50a36fdf2874ba
> Author:     Justin Hibbits <jhibbits@FreeBSD.org>
> AuthorDate: 2022-05-12 20:38:22 +0000
> Commit:     Justin Hibbits <jhibbits@FreeBSD.org>
> CommitDate: 2022-05-19 15:49:55 +0000
> 
>     cpuset: Byte swap cpuset for compat32 on big endian architectures
>     
>     Summary:
>     BITSET uses long as its basic underlying type, which is dependent on the
>     compile type, meaning on 32-bit builds the basic type is 32 bits, but on
>     64-bit builds it's 64 bits.  On little endian architectures this doesn't
>     matter, because the LSB is always at the low bit, so the words get
>     effectively concatenated moving between 32-bit and 64-bit, but on
>     big-endian architectures it throws a wrench in, as setting bit 0 in
>     32-bit mode is equivalent to setting bit 32 in 64-bit mode.  To
>     demonstrate:
>     
>     32-bit mode:
>     
>     BIT_SET(foo, 0):        0x00000001
>     
>     64-bit sees: 0x0000000100000000
>     
>     cpuset is the only system interface that uses bitsets, so solve this
>     by swapping the integer sub-components at the copyin/copyout points.
>     
>     Reviewed by:    kib
>     MFC after:      3 days
>     Sponsored by:   Juniper Networks, Inc.
>     Differential Revision:  https://reviews.freebsd.org/D35225
> ---
>  sys/compat/freebsd32/freebsd32_misc.c | 70 +++++++++++++++++++++++++++++++++--
>  sys/kern/kern_cpuset.c                | 32 ++++++++++------
>  sys/sys/cpuset.h                      |  9 +++++
>  sys/sys/syscallsubr.h                 | 10 +++--
>  4 files changed, 101 insertions(+), 20 deletions(-)
> 
> diff --git a/sys/compat/freebsd32/freebsd32_misc.c b/sys/compat/freebsd32/freebsd32_misc.c
> index 25c7a4e17dc7..d4206fdd24aa 100644
> --- a/sys/compat/freebsd32/freebsd32_misc.c
> +++ b/sys/compat/freebsd32/freebsd32_misc.c
> @@ -3310,13 +3310,72 @@ freebsd32_cpuset_getid(struct thread *td,
>  	    PAIR32TO64(id_t, uap->id), uap->setid));
>  }
>  
> +static int
> +copyin32_set(const void *u, void *k, size_t size)
> +{
> +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
> +	int rv;
> +	struct bitset *kb = k;
> +	int *p;
> +
> +	rv = copyin(u, k, size);
> +	if (rv != 0)
> +		return (rv);
> +
> +	p = (int *)kb->__bits;
> +	/* Loop through swapping words.
> +	 * `size' is in bytes, we need bits. */
> +	for (int i = 0; i < __bitset_words(size * 8); i++) {
> +		int tmp = p[0];
> +		p[0] = p[1];
> +		p[1] = tmp;
> +		p += 2;
> +	}
> +	return (0);
> +#else
> +	return (copyin(u, k, size));
> +#endif
> +}
> +
> +static int
> +copyout32_set(const void *k, void *u, size_t size)
> +{
> +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
> +	const struct bitset *kb = k;
> +	struct bitset *ub = u;
> +	const int *kp = (const int *)kb->__bits;
> +	int *up = (int *)ub->__bits;
> +	int rv;
> +
> +	for (int i = 0; i < __bitset_words(CPU_SETSIZE); i++) {
> +		/* `size' is in bytes, we need bits. */
> +		for (int i = 0; i < __bitset_words(size * 8); i++) {
> +			rv = suword32(up, kp[1]);
> +			if (rv == 0)
> +				rv = suword32(up + 1, kp[0]);
> +			if (rv != 0)
> +				return (EFAULT);
> +		}
> +	}
> +	return (0);
> +#else
> +	return (copyout(k, u, size));
> +#endif
> +}
> +
> +static const struct cpuset_copy_cb cpuset_copy32_cb = {
> +	.copyin = copyin32_set,
> +	.copyout = copyout32_set
> +};
> +
>  int
>  freebsd32_cpuset_getaffinity(struct thread *td,
>      struct freebsd32_cpuset_getaffinity_args *uap)
>  {
>  
>  	return (kern_cpuset_getaffinity(td, uap->level, uap->which,
> -	    PAIR32TO64(id_t,uap->id), uap->cpusetsize, uap->mask));
> +	    PAIR32TO64(id_t,uap->id), uap->cpusetsize, uap->mask,
> +	    &cpuset_copy32_cb));
>  }
>  
>  int
> @@ -3325,7 +3384,8 @@ freebsd32_cpuset_setaffinity(struct thread *td,
>  {
>  
>  	return (user_cpuset_setaffinity(td, uap->level, uap->which,
> -	    PAIR32TO64(id_t,uap->id), uap->cpusetsize, uap->mask));
> +	    PAIR32TO64(id_t,uap->id), uap->cpusetsize, uap->mask,
> +	    &cpuset_copy32_cb));
>  }
>  
>  int
> @@ -3334,7 +3394,8 @@ freebsd32_cpuset_getdomain(struct thread *td,
>  {
>  
>  	return (kern_cpuset_getdomain(td, uap->level, uap->which,
> -	    PAIR32TO64(id_t,uap->id), uap->domainsetsize, uap->mask, uap->policy));
> +	    PAIR32TO64(id_t,uap->id), uap->domainsetsize, uap->mask, uap->policy,
> +	    &cpuset_copy32_cb));
>  }
>  
>  int
> @@ -3343,7 +3404,8 @@ freebsd32_cpuset_setdomain(struct thread *td,
>  {
>  
>  	return (kern_cpuset_setdomain(td, uap->level, uap->which,
> -	    PAIR32TO64(id_t,uap->id), uap->domainsetsize, uap->mask, uap->policy));
> +	    PAIR32TO64(id_t,uap->id), uap->domainsetsize, uap->mask, uap->policy,
> +	    &cpuset_copy32_cb));
>  }
>  
>  int
> diff --git a/sys/kern/kern_cpuset.c b/sys/kern/kern_cpuset.c
> index 56d92687a693..e21ec146bb00 100644
> --- a/sys/kern/kern_cpuset.c
> +++ b/sys/kern/kern_cpuset.c
> @@ -52,6 +52,7 @@ __FBSDID("$FreeBSD$");
>  #include <sys/sched.h>
>  #include <sys/smp.h>
>  #include <sys/syscallsubr.h>
> +#include <sys/sysent.h>
>  #include <sys/capsicum.h>
>  #include <sys/cpuset.h>
>  #include <sys/domainset.h>
> @@ -1742,6 +1743,11 @@ cpuset_check_capabilities(struct thread *td, cpulevel_t level, cpuwhich_t which,
>  	return (0);
>  }
>  
> +static const struct cpuset_copy_cb copy_set = {
> +	.copyin = copyin,
> +	.copyout = copyout
> +};
> +
>  #ifndef _SYS_SYSPROTO_H_
>  struct cpuset_args {
>  	cpusetid_t	*setid;
> @@ -1883,12 +1889,12 @@ sys_cpuset_getaffinity(struct thread *td, struct cpuset_getaffinity_args *uap)
>  {
>  
>  	return (kern_cpuset_getaffinity(td, uap->level, uap->which,
> -	    uap->id, uap->cpusetsize, uap->mask));
> +	    uap->id, uap->cpusetsize, uap->mask, &copy_set));
>  }
>  
>  int
>  kern_cpuset_getaffinity(struct thread *td, cpulevel_t level, cpuwhich_t which,
> -    id_t id, size_t cpusetsize, cpuset_t *maskp)
> +    id_t id, size_t cpusetsize, cpuset_t *maskp, const struct cpuset_copy_cb *cb)
>  {

this broke the build, as Linuxulator also uses kern_cpuset_getaffinity,
may be we need to move out the userspace related code from
kern_cpuset_getaffinity to user_cpuset_getaffinity?


>  	struct thread *ttd;
>  	struct cpuset *nset;
> @@ -1977,7 +1983,7 @@ kern_cpuset_getaffinity(struct thread *td, cpulevel_t level, cpuwhich_t which,
>  			goto out;
>  		}
>  		size = min(cpusetsize, sizeof(cpuset_t));
> -		error = copyout(mask, maskp, size);
> +		error = cb->copyout(mask, maskp, size);
>  		if (error != 0)
>  			goto out;
>  		if (cpusetsize > size) {
> @@ -2021,7 +2027,7 @@ sys_cpuset_setaffinity(struct thread *td, struct cpuset_setaffinity_args *uap)
>  {
>  
>  	return (user_cpuset_setaffinity(td, uap->level, uap->which,
> -	    uap->id, uap->cpusetsize, uap->mask));
> +	    uap->id, uap->cpusetsize, uap->mask, &copy_set));
>  }
>  
>  int
> @@ -2109,7 +2115,7 @@ kern_cpuset_setaffinity(struct thread *td, cpulevel_t level, cpuwhich_t which,
>  
>  int
>  user_cpuset_setaffinity(struct thread *td, cpulevel_t level, cpuwhich_t which,
> -    id_t id, size_t cpusetsize, const cpuset_t *maskp)
> +    id_t id, size_t cpusetsize, const cpuset_t *maskp, const struct cpuset_copy_cb *cb)
>  {
>  	cpuset_t *mask;
>  	int error;
> @@ -2117,7 +2123,7 @@ user_cpuset_setaffinity(struct thread *td, cpulevel_t level, cpuwhich_t which,
>  
>  	size = min(cpusetsize, sizeof(cpuset_t));
>  	mask = malloc(sizeof(cpuset_t), M_TEMP, M_WAITOK | M_ZERO);
> -	error = copyin(maskp, mask, size);
> +	error = cb->copyin(maskp, mask, size);
>  	if (error)
>  		goto out;
>  	/*
> @@ -2165,12 +2171,13 @@ sys_cpuset_getdomain(struct thread *td, struct cpuset_getdomain_args *uap)
>  {
>  
>  	return (kern_cpuset_getdomain(td, uap->level, uap->which,
> -	    uap->id, uap->domainsetsize, uap->mask, uap->policy));
> +	    uap->id, uap->domainsetsize, uap->mask, uap->policy, &copy_set));
>  }
>  
>  int
>  kern_cpuset_getdomain(struct thread *td, cpulevel_t level, cpuwhich_t which,
> -    id_t id, size_t domainsetsize, domainset_t *maskp, int *policyp)
> +    id_t id, size_t domainsetsize, domainset_t *maskp, int *policyp,
> +    const struct cpuset_copy_cb *cb)
>  {
>  	struct domainset outset;
>  	struct thread *ttd;
> @@ -2268,7 +2275,7 @@ kern_cpuset_getdomain(struct thread *td, cpulevel_t level, cpuwhich_t which,
>  	}
>  	DOMAINSET_COPY(&outset.ds_mask, mask);
>  	if (error == 0)
> -		error = copyout(mask, maskp, domainsetsize);
> +		error = cb->copyout(mask, maskp, domainsetsize);
>  	if (error == 0)
>  		if (suword32(policyp, outset.ds_policy) != 0)
>  			error = EFAULT;
> @@ -2292,12 +2299,13 @@ sys_cpuset_setdomain(struct thread *td, struct cpuset_setdomain_args *uap)
>  {
>  
>  	return (kern_cpuset_setdomain(td, uap->level, uap->which,
> -	    uap->id, uap->domainsetsize, uap->mask, uap->policy));
> +	    uap->id, uap->domainsetsize, uap->mask, uap->policy, &copy_set));
>  }
>  
>  int
>  kern_cpuset_setdomain(struct thread *td, cpulevel_t level, cpuwhich_t which,
> -    id_t id, size_t domainsetsize, const domainset_t *maskp, int policy)
> +    id_t id, size_t domainsetsize, const domainset_t *maskp, int policy,
> +    const struct cpuset_copy_cb *cb)
>  {
>  	struct cpuset *nset;
>  	struct cpuset *set;
> @@ -2318,7 +2326,7 @@ kern_cpuset_setdomain(struct thread *td, cpulevel_t level, cpuwhich_t which,
>  		return (error);
>  	memset(&domain, 0, sizeof(domain));
>  	mask = malloc(domainsetsize, M_TEMP, M_WAITOK | M_ZERO);
> -	error = copyin(maskp, mask, domainsetsize);
> +	error = cb->copyin(maskp, mask, domainsetsize);
>  	if (error)
>  		goto out;
>  	/*
> diff --git a/sys/sys/cpuset.h b/sys/sys/cpuset.h
> index 0ce470d5f569..4f55cdc27103 100644
> --- a/sys/sys/cpuset.h
> +++ b/sys/sys/cpuset.h
> @@ -154,6 +154,15 @@ struct prison;
>  struct proc;
>  struct thread;
>  
> +/*
> + * Callbacks for copying in/out a cpuset or domainset.  Used for alternate
> + * ABIs, like compat32.
> + */
> +struct cpuset_copy_cb {
> +	int (*copyin)(const void *, void *, size_t);
> +	int (*copyout)(const void *, void *, size_t);
> +};
> +
>  struct cpuset *cpuset_thread0(void);
>  struct cpuset *cpuset_ref(struct cpuset *);
>  void	cpuset_rel(struct cpuset *);
> diff --git a/sys/sys/syscallsubr.h b/sys/sys/syscallsubr.h
> index 294539c6593c..70e33e9244c4 100644
> --- a/sys/sys/syscallsubr.h
> +++ b/sys/sys/syscallsubr.h
> @@ -39,6 +39,7 @@
>  #include <sys/_uio.h>
>  
>  struct __wrusage;
> +struct cpuset_copy_cb;
>  struct file;
>  struct filecaps;
>  enum idtype;
> @@ -119,18 +120,19 @@ int	kern_connectat(struct thread *td, int dirfd, int fd,
>  int	kern_copy_file_range(struct thread *td, int infd, off_t *inoffp,
>  	    int outfd, off_t *outoffp, size_t len, unsigned int flags);
>  int	kern_cpuset_getaffinity(struct thread *td, cpulevel_t level,
> -	    cpuwhich_t which, id_t id, size_t cpusetsize, cpuset_t *maskp);
> +	    cpuwhich_t which, id_t id, size_t cpusetsize, cpuset_t *maskp,
> +	    const struct cpuset_copy_cb *cb);
>  int	kern_cpuset_setaffinity(struct thread *td, cpulevel_t level,
>  	    cpuwhich_t which, id_t id, cpuset_t *maskp);
>  int	user_cpuset_setaffinity(struct thread *td, cpulevel_t level,
>  	    cpuwhich_t which, id_t id, size_t cpusetsize,
> -	    const cpuset_t *maskp);
> +	    const cpuset_t *maskp, const struct cpuset_copy_cb *cb);
>  int	kern_cpuset_getdomain(struct thread *td, cpulevel_t level,
>  	    cpuwhich_t which, id_t id, size_t domainsetsize,
> -	    domainset_t *maskp, int *policyp);
> +	    domainset_t *maskp, int *policyp, const struct cpuset_copy_cb *cb);
>  int	kern_cpuset_setdomain(struct thread *td, cpulevel_t level,
>  	    cpuwhich_t which, id_t id, size_t domainsetsize,
> -	    const domainset_t *maskp, int policy);
> +	    const domainset_t *maskp, int policy, const struct cpuset_copy_cb *cb);
>  int	kern_cpuset_getid(struct thread *td, cpulevel_t level,
>  	    cpuwhich_t which, id_t id, cpusetid_t *setid);
>  int	kern_cpuset_setid(struct thread *td, cpuwhich_t which,