Skip to contentSkip to conversation

file struct's dentry being null?

DISCUSSION Shaya Potter 3 participants 6 messages Jul 2, 2003 VFS

lore
Share & tools
lore

Shaya Potter

author
#0 lore raw

I decided to do an experiment, in fput, I did a

struct file * next;
next = (struct file *) file->f_list.next;
if ( !next->dentry )

	printk("file struct: next's dentry is null, %d\n", next);

and I get a bunch of kernel output along these lines, with 5 different next memory locations. I only have 3 mount points, so it doesn't seem to be related to that (though I would have thought it might have been the list_head on the superblock). anyways, just trying to figure out what it's pointing to?

thanks,

shaya

#1 lore raw
Quoted 4 lines · Shaya Potter

I decided to do an experiment, in fput, I did a

struct file * next;
next = (struct file *) file->f_list.next;

->f_list.next points to the ->f_list member of next struct file rather than to the struct file itself. This is how it works for all struct list_head embedded into objects.

Quoted 9 lines

if ( !next->dentry )

	printk("file struct: next's dentry is null, %d\n", next);

and I get a bunch of kernel output along these lines, with 5 different next memory locations. I only have 3 mount points, so it doesn't seem to be related to that (though I would have thought it might have been the list_head on the superblock). anyways, just trying to figure out what it's pointing to?

Quoted 4 lines

thanks,

shaya

Nikita.

Quoted 1 line
author
#2 lore raw
Quoted 9 lines · Nikita Danilov
Quoted 4 lines · Shaya Potter

I decided to do an experiment, in fput, I did a

struct file * next;
next = (struct file *) file->f_list.next;

->f_list.next points to the ->f_list member of next struct file rather than to the struct file itself. This is how it works for all struct list_head embedded into objects.

yes, I understand that, the idea being that it's the first entry in the struct that you want to link list, so can get the next struct by just dereferencing the list_head (as &list_head == &struct)

I was just cheating, but that explains my point, I could see it being null if I hit the real list_head (i.e. on the superblock), but I am getting 5 distinct "hits", with only 3 mount points, so trying to figure out why.

the issue I'm trying to figure out is (posted once b4) that I am getting corrupt entries off of ext3's file list in it's superblock, so that when fs_may_remount_ro() is called, it oops with a null pointer exception.

#3 lore raw
Quoted 8 lines · Shaya Potter
Quoted 6 lines · Nikita Danilov
Quoted 4 lines · Shaya Potter

I decided to do an experiment, in fput, I did a

struct file * next;
next = (struct file *) file->f_list.next;

Quoted 8 lines
Quoted 3 lines

->f_list.next points to the ->f_list member of next struct file rather than to the struct file itself. This is how it works for all struct list_head embedded into objects.

yes, I understand that, the idea being that it's the first entry in the struct that you want to link list, so can get the next struct by just dereferencing the list_head (as &list_head == &struct)

No, his point was that &list_head != $struct, but rather &struct == &list_head - some_offset.

So the code you'd need would look like:
struct file *next;
next = list_entry( <address of the list_head>, file, f_list );

Eli
--------------------. "If it ain't broke now,
Eli Carter \ it will be soon." -- crypto-gram
eli.carter(a)inet.com `-------------------------------------------------

author
#4 lore raw
Quoted 18 lines · Eli Carter
Quoted 8 lines · Shaya Potter
Quoted 6 lines · Nikita Danilov
Quoted 4 lines · Shaya Potter

I decided to do an experiment, in fput, I did a

struct file * next;
next = (struct file *) file->f_list.next;

Quoted 8 lines
Quoted 3 lines

->f_list.next points to the ->f_list member of next struct file rather than to the struct file itself. This is how it works for all struct list_head embedded into objects.

yes, I understand that, the idea being that it's the first entry in the struct that you want to link list, so can get the next struct by just dereferencing the list_head (as &list_head == &struct)

Quoted 6 lines

No, his point was that &list_head != $struct, but rather &struct == &list_head - some_offset.

So the code you'd need would look like:
struct file *next;
next = list_entry( <address of the list_head>, file, f_list );

ok, right, looked at list_entry again, and I tried to cheat too much.

that's for the pointer.

author
#5 lore raw
In reply to Eli Carter
Quoted 18 lines · Eli Carter
Quoted 8 lines · Shaya Potter
Quoted 6 lines · Nikita Danilov
Quoted 4 lines · Shaya Potter

I decided to do an experiment, in fput, I did a

struct file * next;
next = (struct file *) file->f_list.next;

Quoted 8 lines
Quoted 3 lines

->f_list.next points to the ->f_list member of next struct file rather than to the struct file itself. This is how it works for all struct list_head embedded into objects.

yes, I understand that, the idea being that it's the first entry in the struct that you want to link list, so can get the next struct by just dereferencing the list_head (as &list_head == &struct)

Quoted 6 lines

No, his point was that &list_head != $struct, but rather &struct == &list_head - some_offset.

So the code you'd need would look like:
struct file *next;
next = list_entry( <address of the list_head>, file, f_list );

ok, even with that fix, I'm still getting the same sort of output

I'm printing out 6 different distinct memory locations whose next entry in the file list has a null dentry.

The system operates normally like this, so just trying to figure out what its dereferencing.

this is the block of code I stick in fput()

{
	struct file * next;
	next = list_entry(file->f_list.next, struct file, f_list);
	if ( !next->f_dentry )
	{
		printk("deleting dentry whose next is null!, next = %d\n",next);
	}

}