turbulentFather of Linux, Linus Torvalds, he jumped into the public eye again. The "opening method" is still a familiar formula——curse. Let’s first take a look at Linus’s famous angry scene: Your code is garbage. I'm going to throw you in spam for a week.
And this time the "victim" is fromGoogleOne of the programmers, Steven Rostedt.
And he is not just a developer, in the words of a netizen, "he can be considered a big boss."
△
Picture source: "OSC Open Source Community" comment area
Not only that, judging from the timeline, the two sides have already clashed for a long time.4 daysfor a long time...
So what is going on?
An "inodes", four days of quarrel
This heated debate took place on the Linux kernel mailing list.
Steven originally posted a post about the eventfs (event file system) patch.
Specifically, I want to discuss whether inodes (index nodes) should remain unique.
(Note: inodes are a core concept in Linux file systems. It is a data structure used to store metadata about a file or directory, rather than the actual contents of the file.)
Steven thinks:
Linus previously suggested using the same inode in eventfs to simplify the implementation of getdents(), which means that all files and directories will use the same inode.
However, this practice was later found to cause problems with the "find" command because the directory and file have the same inode.
Linus later discovered that on 64-bit machines, there was a hole in the eventfs_inode structure due to alignment, which could be used to store the inode of the directory. This solved the directory problem, but the file still retained its own inode.
In Steven's opinion, since the tar command relies on the inode to determine the uniqueness of the file, this approach will destroy the function of the tar command:
Currently, the tar command is already having issues with tracefs (a variant of the event file system) as it shows all files as having a size of zero, causing tar to copy nothing.
In addition, Steven also gave his own solution - it is recommended to copy the get_next_ino() function of the VFS layer to the tracefs_get_next_ino() function of tracefs and add a "files" parameter.
This way, when the eventfs directory is created, the required number of inodes is known in advance. tracefs_get_next_ino() will return a new inode and reserve the next "files" inodes for the caller to use.
When a file's inode is created, its inode will be the inode of its parent directory plus its index into that directory's array of files, giving each file a unique inode.
However, such a proposal was strongly opposed by Linus.
Linus's core point is"The inode is no longer the only descriptor, and we should not continue to rely on this old mechanism.".
However, Steven did not buy Linus' reply. He insisted:
All files and directories should have unique inodes, which simplifies certain aspects of the file system.
However, after several rounds of discussions, Linus couldn't sit still, and then the famous scene of anger occurred:
Don't make things so complicated.
You do not fully understand the purpose and necessity of these functions
Your code is garbage.
Both sides seemed to have their own opinions, and the game went back and forth for a long time. The battle lasted from January 26 to January 29...
But the dramatic thing is that while Linus was arguing, he later released the Linux kernel version 6.8-rc2.
He hopes this version will resolve issues found in previous versions and encourages users to test it.
This is not the first public “confrontation”
In fact, before this, Steven also publicly "confronted" Linus in an event speech at the beginning of 2020.
He even directly positioned the theme of his speech"ArguingwithLinusTorvalds", the content is still focused on suggestions on how to improve Linux efficiency.
However, netizens also expressed their opinions on this latest battle.
Some people think that historical baggage should be abandoned, while others think that it is just a difference in their design concepts:
△Picture source: "OSC Open Source Community" comment area