From: Sam Vilain <[email protected]>
Export the UTS information to a per-process directory /proc/PID/uts,
that has individual nodes for hostname, ostype, etc - similar to
those in /proc/sys/kernel
This duplicates the approach used for /proc/PID/attr, which involves a
lot of duplication of similar functions. Much room for maintenance
optimisation of both implementations remains.
---
Sorry for the duplication of this to the list, stuffed up the stgit
command.
After doing this I noticed that the whole way this is done via sysctls
in /proc/sys is much, much nicer. I was going there to make
/proc/sys/kernel/osname -> /proc/self/uts/sysname (etc), but it seems
that symlinks from /proc/sys are not a done thing.
Is there an argument here perhaps for some integration between the way
this is done for /proc/sys and /proc/PID/xxx ?
fs/proc/base.c | 236 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
1 files changed, 236 insertions(+), 0 deletions(-)
diff --git a/fs/proc/base.c b/fs/proc/base.c
index 2031913..76f5acb 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -73,6 +73,7 @@ #include <linux/cpuset.h>
#include <linux/audit.h>
#include <linux/poll.h>
#include <linux/nsproxy.h>
+#include <linux/utsname.h>
#include "internal.h"
/* NOTE:
@@ -179,6 +180,22 @@ #ifdef CONFIG_AUDITSYSCALL
#endif
PROC_TID_OOM_SCORE,
PROC_TID_OOM_ADJUST,
+#ifdef CONFIG_UTS_NS
+ PROC_TID_UTS,
+ PROC_TGID_UTS,
+ PROC_TGID_UTS_SYSNAME,
+ PROC_TGID_UTS_NODENAME,
+ PROC_TGID_UTS_RELEASE,
+ PROC_TGID_UTS_VERSION,
+ PROC_TGID_UTS_MACHINE,
+ PROC_TGID_UTS_DOMAINNAME,
+ PROC_TID_UTS_SYSNAME,
+ PROC_TID_UTS_NODENAME,
+ PROC_TID_UTS_RELEASE,
+ PROC_TID_UTS_VERSION,
+ PROC_TID_UTS_MACHINE,
+ PROC_TID_UTS_DOMAINNAME,
+#endif
/* Add new entries before this */
PROC_TID_FD_DIR = 0x8000, /* 0x8000-0xffff */
@@ -238,6 +255,9 @@ #endif
#ifdef CONFIG_AUDITSYSCALL
E(PROC_TGID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
#endif
+#ifdef CONFIG_UTS_NS
+ E(PROC_TGID_UTS, "uts", S_IFDIR|S_IRUGO|S_IXUGO),
+#endif
{0,0,NULL,0}
};
static struct pid_entry tid_base_stuff[] = {
@@ -280,6 +300,9 @@ #endif
#ifdef CONFIG_AUDITSYSCALL
E(PROC_TID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
#endif
+#ifdef CONFIG_UTS_NS
+ E(PROC_TID_UTS, "uts", S_IFDIR|S_IRUGO|S_IXUGO),
+#endif
{0,0,NULL,0}
};
@@ -300,6 +323,27 @@ static struct pid_entry tid_attr_stuff[]
};
#endif
+#ifdef CONFIG_UTS_NS
+static struct pid_entry tgid_uts_stuff[] = {
+ E(PROC_TGID_UTS_SYSNAME, "sysname", S_IFREG|S_IRUGO|S_IWUGO),
+ E(PROC_TGID_UTS_NODENAME, "nodename", S_IFREG|S_IRUGO|S_IWUGO),
+ E(PROC_TGID_UTS_RELEASE, "release", S_IFREG|S_IRUGO|S_IWUGO),
+ E(PROC_TGID_UTS_VERSION, "version", S_IFREG|S_IRUGO|S_IWUGO),
+ E(PROC_TGID_UTS_MACHINE, "machine", S_IFREG|S_IRUGO|S_IWUGO),
+ E(PROC_TGID_UTS_DOMAINNAME, "domainname", S_IFREG|S_IRUGO|S_IWUGO),
+ {0,0,NULL,0}
+};
+static struct pid_entry tid_uts_stuff[] = {
+ E(PROC_TID_UTS_SYSNAME, "sysname", S_IFREG|S_IRUGO|S_IWUGO),
+ E(PROC_TID_UTS_NODENAME, "nodename", S_IFREG|S_IRUGO|S_IWUGO),
+ E(PROC_TID_UTS_RELEASE, "release", S_IFREG|S_IRUGO|S_IWUGO),
+ E(PROC_TID_UTS_VERSION, "version", S_IFREG|S_IRUGO|S_IWUGO),
+ E(PROC_TID_UTS_MACHINE, "machine", S_IFREG|S_IRUGO|S_IWUGO),
+ E(PROC_TID_UTS_DOMAINNAME, "domainname", S_IFREG|S_IRUGO|S_IWUGO),
+ {0,0,NULL,0}
+};
+#endif
+
#undef E
static int proc_fd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
@@ -1608,6 +1652,148 @@ static struct file_operations proc_tgid_
static struct inode_operations proc_tgid_attr_inode_operations;
#endif
+#ifdef CONFIG_UTS_NS
+static ssize_t proc_pid_uts_read(struct file * file, char __user * buf,
+ size_t count, loff_t *ppos)
+{
+ struct inode * inode = file->f_dentry->d_inode;
+ ssize_t length;
+ loff_t __ppos = *ppos;
+ struct task_struct *task = get_proc_task(inode);
+ char __buf[__NEW_UTS_LEN+1];
+ char *which;
+
+ length = -ESRCH;
+ if (!task)
+ goto out_no_task;
+
+ switch (file->f_dentry->d_name.name[0]) {
+ case 's':
+ which = task->nsproxy->uts_ns->name.sysname;
+ break;
+ case 'n':
+ which = task->nsproxy->uts_ns->name.nodename;
+ break;
+ case 'r':
+ which = task->nsproxy->uts_ns->name.release;
+ break;
+ case 'v':
+ which = task->nsproxy->uts_ns->name.version;
+ break;
+ case 'm':
+ which = task->nsproxy->uts_ns->name.machine;
+ break;
+ case 'd':
+ which = task->nsproxy->uts_ns->name.domainname;
+ break;
+ default:
+ printk("procfs: impossible uts part '%s'",
+ (char*)file->f_dentry->d_name.name);
+ length = -EINVAL;
+ goto out;
+ }
+
+ length = strlen(which);
+ strcpy(__buf, which);
+ __buf[length++] = '\n';
+
+ if (__ppos >= length)
+ return 0;
+ if (count > length - __ppos)
+ count = length - __ppos;
+ if (copy_to_user(buf, __buf + __ppos, count))
+ return -EFAULT;
+
+out:
+ put_task_struct(task);
+out_no_task:
+ return length;
+}
+
+static ssize_t proc_pid_uts_write(struct file * file, const char __user * buf,
+ size_t count, loff_t *ppos)
+{
+ struct inode * inode = file->f_dentry->d_inode;
+ ssize_t length;
+ struct task_struct *task = get_proc_task(inode);
+ char *which;
+ char __buf[__NEW_UTS_LEN+1];
+
+ length = -ESRCH;
+ if (!task)
+ goto out_no_task;
+ if (count > PAGE_SIZE)
+ count = PAGE_SIZE;
+
+ /* No partial writes. */
+ length = -EINVAL;
+ if (*ppos != 0)
+ goto out;
+ if (count > __NEW_UTS_LEN)
+ goto out;
+
+ length = -EPERM;
+ if (!capable(CAP_SYS_ADMIN))
+ goto out;
+
+ length = -EFAULT;
+ if (copy_from_user(__buf, buf, count))
+ goto out;
+
+ length = -EINVAL;
+ length = strnlen(__buf, count);
+ if (count != length)
+ goto out;
+
+ if (__buf[length-1] == '\n')
+ length = length - 1;
+ __buf[length] = '\0';
+
+ switch (file->f_dentry->d_name.name[0]) {
+ case 's':
+ which = task->nsproxy->uts_ns->name.sysname;
+ break;
+ case 'n':
+ which = task->nsproxy->uts_ns->name.nodename;
+ break;
+ case 'r':
+ which = task->nsproxy->uts_ns->name.release;
+ break;
+ case 'v':
+ which = task->nsproxy->uts_ns->name.version;
+ break;
+ case 'm':
+ which = task->nsproxy->uts_ns->name.machine;
+ break;
+ case 'd':
+ which = task->nsproxy->uts_ns->name.domainname;
+ break;
+ default:
+ printk("procfs: impossible uts part '%s'",
+ (char*)file->f_dentry->d_name.name);
+ length = -EINVAL;
+ goto out;
+ }
+
+ strcpy(which, __buf);
+
+out:
+ put_task_struct(task);
+out_no_task:
+ return length;
+}
+
+static struct file_operations proc_pid_uts_operations = {
+ .read = proc_pid_uts_read,
+ .write = proc_pid_uts_write,
+};
+
+static struct file_operations proc_tid_uts_operations;
+static struct inode_operations proc_tid_uts_inode_operations;
+static struct file_operations proc_tgid_uts_operations;
+static struct inode_operations proc_tgid_uts_inode_operations;
+#endif
+
/* SMP-safe */
static struct dentry *proc_pident_lookup(struct inode *dir,
struct dentry *dentry,
@@ -1760,6 +1946,30 @@ #ifdef CONFIG_SECURITY
case PROC_TGID_ATTR_FSCREATE:
inode->i_fop = &proc_pid_attr_operations;
break;
+ case PROC_TID_UTS:
+ inode->i_nlink = 2;
+ inode->i_op = &proc_tid_uts_inode_operations;
+ inode->i_fop = &proc_tid_uts_operations;
+ break;
+ case PROC_TGID_UTS:
+ inode->i_nlink = 2;
+ inode->i_op = &proc_tgid_uts_inode_operations;
+ inode->i_fop = &proc_tgid_uts_operations;
+ break;
+ case PROC_TGID_UTS_SYSNAME:
+ case PROC_TGID_UTS_NODENAME:
+ case PROC_TGID_UTS_RELEASE:
+ case PROC_TGID_UTS_VERSION:
+ case PROC_TGID_UTS_MACHINE:
+ case PROC_TGID_UTS_DOMAINNAME:
+ case PROC_TID_UTS_SYSNAME:
+ case PROC_TID_UTS_NODENAME:
+ case PROC_TID_UTS_RELEASE:
+ case PROC_TID_UTS_VERSION:
+ case PROC_TID_UTS_MACHINE:
+ case PROC_TID_UTS_DOMAINNAME:
+ inode->i_fop = &proc_pid_uts_operations;
+ break;
#endif
#ifdef CONFIG_KALLSYMS
case PROC_TID_WCHAN:
@@ -1889,6 +2099,32 @@ static struct inode_operations proc_tid_
};
#endif
+#ifdef CONFIG_UTS_NS
+static int proc_tgid_uts_readdir(struct file * filp,
+ void * dirent, filldir_t filldir)
+{
+ return proc_pident_readdir(filp,dirent,filldir,
+ tgid_uts_stuff,ARRAY_SIZE(tgid_uts_stuff));
+}
+
+static int proc_tid_uts_readdir(struct file * filp,
+ void * dirent, filldir_t filldir)
+{
+ return proc_pident_readdir(filp,dirent,filldir,
+ tid_uts_stuff,ARRAY_SIZE(tid_uts_stuff));
+}
+
+static struct file_operations proc_tgid_uts_operations = {
+ .read = generic_read_dir,
+ .readdir = proc_tgid_uts_readdir,
+};
+
+static struct file_operations proc_tid_uts_operations = {
+ .read = generic_read_dir,
+ .readdir = proc_tid_uts_readdir,
+};
+#endif
+
/*
* /proc/self:
*/
Sam Vilain <[email protected]> wrote:
>
> Export the UTS information to a per-process directory /proc/PID/uts,
> that has individual nodes for hostname, ostype, etc - similar to
> those in /proc/sys/kernel
umm, why?
> This duplicates the approach used for /proc/PID/attr, which involves a
> lot of duplication of similar functions. Much room for maintenance
> optimisation of both implementations remains.
>
> ...
>
> fs/proc/base.c | 236 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
ouch.
On Llu, 2006-05-22 at 17:24 +1200, Sam Vilain wrote:
> From: Sam Vilain <[email protected]>
>
> Export the UTS information to a per-process directory /proc/PID/uts,
> that has individual nodes for hostname, ostype, etc - similar to
> those in /proc/sys/kernel
Can you explain the locking being used here against the name being
changed at the same moment ?
On Mon, May 22, 2006 at 05:24:25PM +1200, Sam Vilain wrote:
> From: Sam Vilain <[email protected]>
>
> Export the UTS information to a per-process directory /proc/PID/uts,
> that has individual nodes for hostname, ostype, etc - similar to
> those in /proc/sys/kernel
>
> This duplicates the approach used for /proc/PID/attr, which involves a
> lot of duplication of similar functions. Much room for maintenance
> optimisation of both implementations remains.
hmm, IMHO what we want/need in the future is some
entries for the namespaces/nsproxies to make them
accessible from outside (kind of handle)
the processes could then use relative symlinks
to point to those entries, but I'm not sure they
really belong in /proc although that is probably
more approriate than the other 100 entries not
belonging to procfs :)
> ---
> Sorry for the duplication of this to the list, stuffed up the stgit
> command.
>
> After doing this I noticed that the whole way this is done via sysctls
> in /proc/sys is much, much nicer. I was going there to make
> /proc/sys/kernel/osname -> /proc/self/uts/sysname (etc), but it seems
> that symlinks from /proc/sys are not a done thing.
>
> Is there an argument here perhaps for some integration between the way
> this is done for /proc/sys and /proc/PID/xxx ?
I'm not very happy about the /sys and /proc/sys
mixup, let's avoid further mixing there ...
best,
Herbert
> fs/proc/base.c | 236 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> 1 files changed, 236 insertions(+), 0 deletions(-)
>
> diff --git a/fs/proc/base.c b/fs/proc/base.c
> index 2031913..76f5acb 100644
> --- a/fs/proc/base.c
> +++ b/fs/proc/base.c
> @@ -73,6 +73,7 @@ #include <linux/cpuset.h>
> #include <linux/audit.h>
> #include <linux/poll.h>
> #include <linux/nsproxy.h>
> +#include <linux/utsname.h>
> #include "internal.h"
>
> /* NOTE:
> @@ -179,6 +180,22 @@ #ifdef CONFIG_AUDITSYSCALL
> #endif
> PROC_TID_OOM_SCORE,
> PROC_TID_OOM_ADJUST,
> +#ifdef CONFIG_UTS_NS
> + PROC_TID_UTS,
> + PROC_TGID_UTS,
> + PROC_TGID_UTS_SYSNAME,
> + PROC_TGID_UTS_NODENAME,
> + PROC_TGID_UTS_RELEASE,
> + PROC_TGID_UTS_VERSION,
> + PROC_TGID_UTS_MACHINE,
> + PROC_TGID_UTS_DOMAINNAME,
> + PROC_TID_UTS_SYSNAME,
> + PROC_TID_UTS_NODENAME,
> + PROC_TID_UTS_RELEASE,
> + PROC_TID_UTS_VERSION,
> + PROC_TID_UTS_MACHINE,
> + PROC_TID_UTS_DOMAINNAME,
> +#endif
>
> /* Add new entries before this */
> PROC_TID_FD_DIR = 0x8000, /* 0x8000-0xffff */
> @@ -238,6 +255,9 @@ #endif
> #ifdef CONFIG_AUDITSYSCALL
> E(PROC_TGID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
> #endif
> +#ifdef CONFIG_UTS_NS
> + E(PROC_TGID_UTS, "uts", S_IFDIR|S_IRUGO|S_IXUGO),
> +#endif
> {0,0,NULL,0}
> };
> static struct pid_entry tid_base_stuff[] = {
> @@ -280,6 +300,9 @@ #endif
> #ifdef CONFIG_AUDITSYSCALL
> E(PROC_TID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
> #endif
> +#ifdef CONFIG_UTS_NS
> + E(PROC_TID_UTS, "uts", S_IFDIR|S_IRUGO|S_IXUGO),
> +#endif
> {0,0,NULL,0}
> };
>
> @@ -300,6 +323,27 @@ static struct pid_entry tid_attr_stuff[]
> };
> #endif
>
> +#ifdef CONFIG_UTS_NS
> +static struct pid_entry tgid_uts_stuff[] = {
> + E(PROC_TGID_UTS_SYSNAME, "sysname", S_IFREG|S_IRUGO|S_IWUGO),
> + E(PROC_TGID_UTS_NODENAME, "nodename", S_IFREG|S_IRUGO|S_IWUGO),
> + E(PROC_TGID_UTS_RELEASE, "release", S_IFREG|S_IRUGO|S_IWUGO),
> + E(PROC_TGID_UTS_VERSION, "version", S_IFREG|S_IRUGO|S_IWUGO),
> + E(PROC_TGID_UTS_MACHINE, "machine", S_IFREG|S_IRUGO|S_IWUGO),
> + E(PROC_TGID_UTS_DOMAINNAME, "domainname", S_IFREG|S_IRUGO|S_IWUGO),
> + {0,0,NULL,0}
> +};
> +static struct pid_entry tid_uts_stuff[] = {
> + E(PROC_TID_UTS_SYSNAME, "sysname", S_IFREG|S_IRUGO|S_IWUGO),
> + E(PROC_TID_UTS_NODENAME, "nodename", S_IFREG|S_IRUGO|S_IWUGO),
> + E(PROC_TID_UTS_RELEASE, "release", S_IFREG|S_IRUGO|S_IWUGO),
> + E(PROC_TID_UTS_VERSION, "version", S_IFREG|S_IRUGO|S_IWUGO),
> + E(PROC_TID_UTS_MACHINE, "machine", S_IFREG|S_IRUGO|S_IWUGO),
> + E(PROC_TID_UTS_DOMAINNAME, "domainname", S_IFREG|S_IRUGO|S_IWUGO),
> + {0,0,NULL,0}
> +};
> +#endif
> +
> #undef E
>
> static int proc_fd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
> @@ -1608,6 +1652,148 @@ static struct file_operations proc_tgid_
> static struct inode_operations proc_tgid_attr_inode_operations;
> #endif
>
> +#ifdef CONFIG_UTS_NS
> +static ssize_t proc_pid_uts_read(struct file * file, char __user * buf,
> + size_t count, loff_t *ppos)
> +{
> + struct inode * inode = file->f_dentry->d_inode;
> + ssize_t length;
> + loff_t __ppos = *ppos;
> + struct task_struct *task = get_proc_task(inode);
> + char __buf[__NEW_UTS_LEN+1];
> + char *which;
> +
> + length = -ESRCH;
> + if (!task)
> + goto out_no_task;
> +
> + switch (file->f_dentry->d_name.name[0]) {
> + case 's':
> + which = task->nsproxy->uts_ns->name.sysname;
> + break;
> + case 'n':
> + which = task->nsproxy->uts_ns->name.nodename;
> + break;
> + case 'r':
> + which = task->nsproxy->uts_ns->name.release;
> + break;
> + case 'v':
> + which = task->nsproxy->uts_ns->name.version;
> + break;
> + case 'm':
> + which = task->nsproxy->uts_ns->name.machine;
> + break;
> + case 'd':
> + which = task->nsproxy->uts_ns->name.domainname;
> + break;
> + default:
> + printk("procfs: impossible uts part '%s'",
> + (char*)file->f_dentry->d_name.name);
> + length = -EINVAL;
> + goto out;
> + }
> +
> + length = strlen(which);
> + strcpy(__buf, which);
> + __buf[length++] = '\n';
> +
> + if (__ppos >= length)
> + return 0;
> + if (count > length - __ppos)
> + count = length - __ppos;
> + if (copy_to_user(buf, __buf + __ppos, count))
> + return -EFAULT;
> +
> +out:
> + put_task_struct(task);
> +out_no_task:
> + return length;
> +}
> +
> +static ssize_t proc_pid_uts_write(struct file * file, const char __user * buf,
> + size_t count, loff_t *ppos)
> +{
> + struct inode * inode = file->f_dentry->d_inode;
> + ssize_t length;
> + struct task_struct *task = get_proc_task(inode);
> + char *which;
> + char __buf[__NEW_UTS_LEN+1];
> +
> + length = -ESRCH;
> + if (!task)
> + goto out_no_task;
> + if (count > PAGE_SIZE)
> + count = PAGE_SIZE;
> +
> + /* No partial writes. */
> + length = -EINVAL;
> + if (*ppos != 0)
> + goto out;
> + if (count > __NEW_UTS_LEN)
> + goto out;
> +
> + length = -EPERM;
> + if (!capable(CAP_SYS_ADMIN))
> + goto out;
> +
> + length = -EFAULT;
> + if (copy_from_user(__buf, buf, count))
> + goto out;
> +
> + length = -EINVAL;
> + length = strnlen(__buf, count);
> + if (count != length)
> + goto out;
> +
> + if (__buf[length-1] == '\n')
> + length = length - 1;
> + __buf[length] = '\0';
> +
> + switch (file->f_dentry->d_name.name[0]) {
> + case 's':
> + which = task->nsproxy->uts_ns->name.sysname;
> + break;
> + case 'n':
> + which = task->nsproxy->uts_ns->name.nodename;
> + break;
> + case 'r':
> + which = task->nsproxy->uts_ns->name.release;
> + break;
> + case 'v':
> + which = task->nsproxy->uts_ns->name.version;
> + break;
> + case 'm':
> + which = task->nsproxy->uts_ns->name.machine;
> + break;
> + case 'd':
> + which = task->nsproxy->uts_ns->name.domainname;
> + break;
> + default:
> + printk("procfs: impossible uts part '%s'",
> + (char*)file->f_dentry->d_name.name);
> + length = -EINVAL;
> + goto out;
> + }
> +
> + strcpy(which, __buf);
> +
> +out:
> + put_task_struct(task);
> +out_no_task:
> + return length;
> +}
> +
> +static struct file_operations proc_pid_uts_operations = {
> + .read = proc_pid_uts_read,
> + .write = proc_pid_uts_write,
> +};
> +
> +static struct file_operations proc_tid_uts_operations;
> +static struct inode_operations proc_tid_uts_inode_operations;
> +static struct file_operations proc_tgid_uts_operations;
> +static struct inode_operations proc_tgid_uts_inode_operations;
> +#endif
> +
> /* SMP-safe */
> static struct dentry *proc_pident_lookup(struct inode *dir,
> struct dentry *dentry,
> @@ -1760,6 +1946,30 @@ #ifdef CONFIG_SECURITY
> case PROC_TGID_ATTR_FSCREATE:
> inode->i_fop = &proc_pid_attr_operations;
> break;
> + case PROC_TID_UTS:
> + inode->i_nlink = 2;
> + inode->i_op = &proc_tid_uts_inode_operations;
> + inode->i_fop = &proc_tid_uts_operations;
> + break;
> + case PROC_TGID_UTS:
> + inode->i_nlink = 2;
> + inode->i_op = &proc_tgid_uts_inode_operations;
> + inode->i_fop = &proc_tgid_uts_operations;
> + break;
> + case PROC_TGID_UTS_SYSNAME:
> + case PROC_TGID_UTS_NODENAME:
> + case PROC_TGID_UTS_RELEASE:
> + case PROC_TGID_UTS_VERSION:
> + case PROC_TGID_UTS_MACHINE:
> + case PROC_TGID_UTS_DOMAINNAME:
> + case PROC_TID_UTS_SYSNAME:
> + case PROC_TID_UTS_NODENAME:
> + case PROC_TID_UTS_RELEASE:
> + case PROC_TID_UTS_VERSION:
> + case PROC_TID_UTS_MACHINE:
> + case PROC_TID_UTS_DOMAINNAME:
> + inode->i_fop = &proc_pid_uts_operations;
> + break;
> #endif
> #ifdef CONFIG_KALLSYMS
> case PROC_TID_WCHAN:
> @@ -1889,6 +2099,32 @@ static struct inode_operations proc_tid_
> };
> #endif
>
> +#ifdef CONFIG_UTS_NS
> +static int proc_tgid_uts_readdir(struct file * filp,
> + void * dirent, filldir_t filldir)
> +{
> + return proc_pident_readdir(filp,dirent,filldir,
> + tgid_uts_stuff,ARRAY_SIZE(tgid_uts_stuff));
> +}
> +
> +static int proc_tid_uts_readdir(struct file * filp,
> + void * dirent, filldir_t filldir)
> +{
> + return proc_pident_readdir(filp,dirent,filldir,
> + tid_uts_stuff,ARRAY_SIZE(tid_uts_stuff));
> +}
> +
> +static struct file_operations proc_tgid_uts_operations = {
> + .read = generic_read_dir,
> + .readdir = proc_tgid_uts_readdir,
> +};
> +
> +static struct file_operations proc_tid_uts_operations = {
> + .read = generic_read_dir,
> + .readdir = proc_tid_uts_readdir,
> +};
> +#endif
> +
> /*
> * /proc/self:
> */
Sam Vilain <[email protected]> writes:
> From: Sam Vilain <[email protected]>
>
> Export the UTS information to a per-process directory /proc/PID/uts,
> that has individual nodes for hostname, ostype, etc - similar to
> those in /proc/sys/kernel
>
> This duplicates the approach used for /proc/PID/attr, which involves a
> lot of duplication of similar functions. Much room for maintenance
> optimisation of both implementations remains.
> ---
> Sorry for the duplication of this to the list, stuffed up the stgit
> command.
>
> After doing this I noticed that the whole way this is done via sysctls
> in /proc/sys is much, much nicer. I was going there to make
> /proc/sys/kernel/osname -> /proc/self/uts/sysname (etc), but it seems
> that symlinks from /proc/sys are not a done thing.
>
> Is there an argument here perhaps for some integration between the way
> this is done for /proc/sys and /proc/PID/xxx ?
>
> fs/proc/base.c | 236 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> 1 files changed, 236 insertions(+), 0 deletions(-)
Good intentions :)
But since this doesn't actually fix /proc/sys/kernel/osname and friends.
I would call this implementation a failure.
Let's first fix /proc/sys/kernel/osname to be sensitive to the caller,
and then see if we can make /proc/sys a symlink to /proc/<pid>/sys
Eric
Andrew Morton wrote:
>Sam Vilain <[email protected]> wrote:
>
>
>>Export the UTS information to a per-process directory /proc/PID/uts,
>> that has individual nodes for hostname, ostype, etc - similar to
>> those in /proc/sys/kernel
>>
>>
>umm, why?
>
>
Er, for the same reason we have /proc/PID/mounts ?
>> This duplicates the approach used for /proc/PID/attr, which involves a
>> lot of duplication of similar functions. Much room for maintenance
>> optimisation of both implementations remains.
>> fs/proc/base.c | 236 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
>>
>ouch.
>
>
Also yow. Refer above gross understatement.
Sam.
Alan Cox wrote:
>On Llu, 2006-05-22 at 17:24 +1200, Sam Vilain wrote:
>
>
>>From: Sam Vilain <[email protected]>
>>
>>Export the UTS information to a per-process directory /proc/PID/uts,
>>that has individual nodes for hostname, ostype, etc - similar to
>>those in /proc/sys/kernel
>>
>>
>
>Can you explain the locking being used here against the name being
>changed at the same moment ?
>
Is this a test? :-)
Let's see, get_task_pid locks the task struct (so that it doesn't go
away while we're de-referencing the nsproxy and uts_ns etc), and the
kobj references are assumed to be enough to avoid the references
dropping away in the meantime.
I didn't grab uts_sem. That semaphore could be made per-uts_ns, in
theory. Whether anyone cares about contention that much is another
question.
I intended this to be exploratory, it wasn't really mergable. Should
have added an [RFC] tag, sorry about that.
Sam.
Sam Vilain wrote:
>I didn't grab uts_sem. That semaphore could be made per-uts_ns, in
>theory. Whether anyone cares about contention that much is another
>question.
>
>
FWIW, the uts_sem isn't mentioned anywhere in the
/proc/sys/kernel/osname sysctl, either. So that interface probably
isn't safe on SMP/preempt.
Sam.
Eric W. Biederman wrote:
>>Sorry for the duplication of this to the list, stuffed up the stgit
>>command.
>>
>>After doing this I noticed that the whole way this is done via sysctls
>>in /proc/sys is much, much nicer. I was going there to make
>>/proc/sys/kernel/osname -> /proc/self/uts/sysname (etc), but it seems
>>that symlinks from /proc/sys are not a done thing.
>>
>>Is there an argument here perhaps for some integration between the way
>>this is done for /proc/sys and /proc/PID/xxx ?
>>
>> fs/proc/base.c | 236 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
>> 1 files changed, 236 insertions(+), 0 deletions(-)
>>
>>
>
>Good intentions :)
>But since this doesn't actually fix /proc/sys/kernel/osname and friends.
>I would call this implementation a failure.
>
>
Yes, actually my second failure from yesterday's efforts. Thought I'd
better send something though.
>Let's first fix /proc/sys/kernel/osname to be sensitive to the caller,
>and then see if we can make /proc/sys a symlink to /proc/<pid>/sys
>
>
Ok, that sounds like a much simpler starting point, I'll see what I can
cook up.
The only thing was, the names were odd and "machine" was missing.
Isn't everything under /proc/<pid> effectively a sysctl? Wouldn't it be
much nicer to re-use that infrastructure for everything under there?
Sam.
Sam Vilain wrote:
>Sam Vilain wrote:
>
>
>
>>I didn't grab uts_sem. That semaphore could be made per-uts_ns, in
>>theory. Whether anyone cares about contention that much is another
>>question.
>>
>>
>>
>>
>
>FWIW, the uts_sem isn't mentioned anywhere in the
>/proc/sys/kernel/osname sysctl, either. So that interface probably
>isn't safe on SMP/preempt.
>
>
No, there it is, in proc_doutsstring. I didn't parse the "doutsstring"
as "do_uts_string".
Sam.