Greg,
Here's a reworked version of shm to sysfs.
Summary :
-shm destruction moved to kobject release function
-id seek was trivial.Attrshow rewritten.
-Tested against recursive fgreps-kde-personal benchmark program
-Patched might_sleep problem
-Checking release code in front of all shm_destroy calls
-No more MAJ charset for attributes
-Adding EOL to attr values.
-Working against 2.6.0.test2
Is it ok for you ?
Thanks for your help,
Regards,
Fabian
diff -Naur orig/ipc/shm.c edited/ipc/shm.c
--- orig/ipc/shm.c 2003-07-27 17:02:04.000000000 +0000
+++ edited/ipc/shm.c 2003-07-31 22:34:59.000000000 +0000
@@ -13,6 +13,7 @@
* Shared /dev/zero support, Kanoj Sarcar <[email protected]>
* Move the mm functionality over to mm/shmem.c, Christoph Rohland <[email protected]>
*
+ * sysfs support (c) 2003 Fabian Frederick <[email protected]>
*/
#include <linux/config.h>
@@ -27,6 +28,8 @@
#include <linux/shmem_fs.h>
#include <linux/security.h>
#include <asm/uaccess.h>
+#include <linux/kobject.h>
+#include <linux/kobj_map.h>
#include "util.h"
@@ -62,6 +65,12 @@
#ifdef CONFIG_PROC_FS
create_proc_read_entry("sysvipc/shm", 0, 0, sysvipc_shm_read_proc, NULL);
#endif
+ strcpy(shm_ids.kobj.name, "shm");
+ shm_ids.kobj.parent = NULL;
+ shm_ids.kobj.kset = NULL;
+ shm_ids.kobj.ktype = NULL;
+ kobject_register(&shm_ids.kobj);
+
}
static inline int shm_checkid(struct shmid_kernel *s, int id)
@@ -81,8 +90,6 @@
return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni+1);
}
-
-
static inline void shm_inc (int id) {
struct shmid_kernel *shp;
@@ -94,6 +101,80 @@
shm_unlock(shp);
}
+/*
+ * sysfs exports
+ */
+static ssize_t shm_attr_show(struct kobject *kobj, struct attribute *attr, char *buf)
+{
+ unsigned int id=simple_strtoul(kobj->name,NULL,10)%SEQ_MULTIPLIER;
+ struct shmid_kernel *shp;
+ down (&shm_ids.sem);
+ shp=shm_lock(id);
+ if(!strcmp(attr->name,"key"))
+ snprintf(buf, PAGE_SIZE,"%ld\n", (long)shp->shm_perm.key);
+ else if(!strcmp(attr->name, "owner"))
+ snprintf(buf, PAGE_SIZE, "%d\n", shp->shm_perm.uid);
+ else if(!strcmp(attr->name,"perms"))
+ snprintf(buf, PAGE_SIZE, "%4o\n", shp->shm_flags);
+ else if(!strcmp(attr->name,"bytes"))
+ snprintf(buf, PAGE_SIZE, "%d\n", (int)shp->shm_segsz);
+ else if (!strcmp(attr->name, "nattach"))
+ snprintf(buf, PAGE_SIZE, "%ld\n", \
+ is_file_hugepages(shp->shm_file) ? \
+ (file_count(shp->shm_file)-1):shp->shm_nattch);
+ else if (!strcmp(attr->name,"cpid"))
+ snprintf(buf,PAGE_SIZE, "%d\n", shp->shm_cprid);
+ else if (!strcmp(attr->name,"lpid"))
+ snprintf(buf,PAGE_SIZE, "%d\n",shp->shm_lprid);
+ else if (!strcmp(attr->name,"gid"))
+ snprintf(buf,PAGE_SIZE, "%d\n", shp->shm_perm.gid);
+ else if (!strcmp(attr->name,"cuid"))
+ snprintf(buf,PAGE_SIZE, "%d\n", shp->shm_perm.cuid);
+ else if (!strcmp(attr->name,"cgid"))
+ snprintf(buf,PAGE_SIZE, "%d\n", shp->shm_perm.cgid);
+ else if (!strcmp(attr->name,"atime"))
+ snprintf(buf,PAGE_SIZE, "%ld\n", (long)shp->shm_atim);
+ else if (!strcmp(attr->name,"dtime"))
+ snprintf(buf,PAGE_SIZE, "%ld\n", (long)shp->shm_dtim);
+ else if (!strcmp(attr->name,"ctime"))
+ snprintf(buf,PAGE_SIZE, "%ld\n", (long) shp->shm_ctim);
+ shm_unlock(shp);
+ up (&shm_ids.sem);
+ return strlen(buf);
+}
+
+#define SHM_ATTR(_name) \
+static struct attribute shm_attr_##_name={ \
+ .name=__stringify(_name), \
+ .mode=0444, \
+}; \
+sysfs_create_file(&shm_ids.entries[id].kobj, &shm_attr_##_name );
+
+void shm_release (struct kobject *kobj)
+{
+ struct shmid_kernel *shp;
+ int id;
+ id=(int)simple_strtoul(kobj->name, NULL, 10);
+ shp = shm_lock(id);
+ shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
+ shm_rmid (id);
+ shm_unlock(shp);
+ if (!is_file_hugepages(shp->shm_file))
+ shmem_lock(shp->shm_file, 0);
+ fput (shp->shm_file);
+ security_shm_free(shp);
+ ipc_rcu_free(shp, sizeof(struct shmid_kernel));
+}
+
+static struct sysfs_ops shm_sysfs_ops ={
+ .show = shm_attr_show,
+};
+
+static struct kobj_type ktype_shm ={
+ .sysfs_ops = &shm_sysfs_ops,
+ .release = &shm_release,
+};
+
/* This is called by fork, once for every shm attach. */
static void shm_open (struct vm_area_struct *shmd)
{
@@ -110,14 +191,8 @@
*/
static void shm_destroy (struct shmid_kernel *shp)
{
- shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
- shm_rmid (shp->id);
shm_unlock(shp);
- if (!is_file_hugepages(shp->shm_file))
- shmem_lock(shp->shm_file, 0);
- fput (shp->shm_file);
- security_shm_free(shp);
- ipc_rcu_free(shp, sizeof(struct shmid_kernel));
+ kobject_unregister(&shm_ids.entries[shp->id%SEQ_MULTIPLIER].kobj);
}
/*
@@ -165,6 +240,8 @@
.nopage = shmem_nopage,
};
+ /* Call with shm_ids.sem locked
+ */
static int newseg (key_t key, int shmflg, size_t size)
{
int error;
@@ -223,7 +300,29 @@
else
file->f_op = &shm_file_operations;
shm_tot += numpages;
+ sprintf(shm_ids.entries[id].kobj.name, "%ld", \
+ (long)shm_buildid(id,shp->shm_perm.seq));
+ shm_ids.entries[id].kobj.parent=kobject_get(&shm_ids.kobj);
+ shm_ids.entries[id].kobj.kset=NULL;
+ shm_ids.entries[id].kobj.ktype=&ktype_shm;
shm_unlock(shp);
+ if(!kobject_register(&(shm_ids.entries[id].kobj))){
+ SHM_ATTR(key);
+ SHM_ATTR(owner);
+ SHM_ATTR(bytes);
+ SHM_ATTR(perms);
+ SHM_ATTR(nattach);
+ SHM_ATTR(cpid);
+ SHM_ATTR(lpid);
+ SHM_ATTR(gid);
+ SHM_ATTR(cuid);
+ SHM_ATTR(cgid);
+ SHM_ATTR(atime);
+ SHM_ATTR(dtime);
+ SHM_ATTR(ctime);
+ }else{
+ return -EINVAL;
+ }
return shp->id;
no_id:
diff -Naur orig/ipc/util.h edited/ipc/util.h
--- orig/ipc/util.h 2003-07-27 17:03:11.000000000 +0000
+++ edited/ipc/util.h 2003-07-29 06:51:51.000000000 +0000
@@ -19,9 +19,11 @@
unsigned short seq_max;
struct semaphore sem;
struct ipc_id* entries;
+ struct kobject kobj; /* sysfs/ipc/shm {,sem,msg...} */
};
struct ipc_id {
+ struct kobject kobj; /* sysfs/ipc/xxx/yyy */
struct kern_ipc_perm* p;
};
___________________________________
> Summary :
>
> -shm destruction moved to kobject release function
> -id seek was trivial.Attrshow rewritten.
> -Tested against recursive fgreps-kde-personal benchmark program
> -Patched might_sleep problem
> -Checking release code in front of all shm_destroy calls
> -No more MAJ charset for attributes
> -Adding EOL to attr values.
> -Working against 2.6.0.test2
>
> Is it ok for you ?
A few comments below.
> +#include <linux/kobj_map.h>
You shouldn't need to include this.
> + strcpy(shm_ids.kobj.name, "shm");
> + shm_ids.kobj.parent = NULL;
> + shm_ids.kobj.kset = NULL;
> + shm_ids.kobj.ktype = NULL;
> + kobject_register(&shm_ids.kobj);
The kobject internals rely on the entire kobject being zero-filled. You
should be memset'ing it (and probably the entire object) earlier, which
saves you from manually initializing the fields above to 0, and prevents
you from missing any in the future.
> +static ssize_t shm_attr_show(struct kobject *kobj, struct attribute *attr, char *buf)
> +{
> + unsigned int id=simple_strtoul(kobj->name,NULL,10)%SEQ_MULTIPLIER;
> + struct shmid_kernel *shp;
> + down (&shm_ids.sem);
> + shp=shm_lock(id);
> + if(!strcmp(attr->name,"key"))
> + snprintf(buf, PAGE_SIZE,"%ld\n", (long)shp->shm_perm.key);
...
If you're going to do it this way, I suggest declaring the attributes
before the show method, so you can do a pointer comparison, rather than a
string compare:
+ if(attr == &shm_attr_key))
+ snprintf(buf, PAGE_SIZE,"%ld\n", (long)shp->shm_perm.key);
...
> +#define SHM_ATTR(_name) \
> +static struct attribute shm_attr_##_name={ \
> + .name=__stringify(_name), \
> + .mode=0444, \
> +}; \
Also, you should declare all the attributes here, rather than in the
registration function. They will only be defined once, and will not put
unnecessary pressure on the stack.
struct kobj_type::default_attrs is designed to hold an array of attributes
that are added to each kobject registered of that type. This is done
automatically for you, so you don't have to do it manually. (It also
checks the errors as it adds each attribute). I suggest using that.
Thanks,
-pat