Slashdot Mirror
Ext4 Data Losses Explained, Worked Around
ddfall writes "H-Online has a follow-up on the Ext4 file system — Last week's news about data loss with the Linux Ext4 file system is explained and new solutions have been provided by Ted Ts'o to allow Ext4 to behave more like Ext3."
User: My data, it's gone!
EXT4:"Ext4 developer Ted Ts'o stresses in his answer to the bug report that Ext4 behaves precisely as demanded by the POSIX standard for file operations."
Solution: WORKS AS DESIGNED
Sent from your iPad.
FTFA, this is the problem:
Ext4, on the other hand, has another mechanism: delayed block allocation. After a file has been closed, up to a minute may elapse before data blocks on the disk are actually allocated. Delayed block allocation allows the filing system to optimise its write processes, but at the price that the metadata of a newly created file will display a size of 0 bytes and occupy no data blocks until the delayed allocation takes place. If the system crashes during this time, the rename() operation may already be committed in the journal, even though the new file still contains no data. The result is that after a crash the file is empty: both the old and the new data have been lost.
And now my question: Why did the Ext4 developers make the same mistakes Reiser and XFS both made (and later corrected) years ago? Before you get to write any filesystem code, you should have to study how other people have done it, including all the change history. Seriously.
Those who fail to learn the lessons of [change] history are doomed to repeat it.
My blog
Ext4 developer Ted Ts'o stresses in his answer to the bug report that Ext4 behaves precisely as demanded by the POSIX standard for file operations.
I couldn't disagree more:
When applications want to overwrite an existing file with new or changed data [...] they first create a temporary file for the new data and then rename it with the system call - rename(). [...] Delayed block allocation allows the filing system to optimise its write processes, but at the price that the metadata of a newly created file will display a size of 0 bytes and occupy no data blocks until [up to 60 seconds later].
Application developers reasonably expect that writes to the disk which happen far apart in time will happen in order. If I write to a file and then rename the file, I expect that the rename will not complete significantly before the write. Certainly not 60 seconds before the write. It seems dead obvious, at least to me, that the update of the directory entry should be deferred until after ext4 flushes that part of the file written prior to the change in the directory entry.
Moderating "-1, Disagree" is simple censorship. Have the guts to post your opinion.
That's General Ts'o to you!
This post climbed Mt. Washington.
This is the problem with new features - the users have problems using them until they fully understands and appreciates the advantages and disadvantages.
And also consider - ext4 is relatively new, so it will improve over time. If you want stability stick to ext3 or ext2. If you want a really stupid filesystem go FAT and prepare for a patent attack.
If builders built buildings the way programmers wrote programs, then the first woodpecker would destroy civilization.
But... those of us who learned the Ancient And Most Wise ways always triple-sync. We also sacrifice Peeps and use red food colouring in voodoo ceremonies (hey, it really is blood, so it should work) to keep the hardware from failing.
On next week's Slashdot, there will be a brief tutorial on the right way to burn a Windows CD at the stake, and how to align the standing stones known as RAM Chips to points of astronomical significance.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
The workaround (flushing everything to disk before the rename) is a disaster for laptops or anything else which might wish to spin down a disk drive.
The write-replace idiom is used when a program is updating a file and can tolerate the update being lost in a crash, but wants either the old or the new to be intact and uncorrupted. The proposed sync solution accomplishes this, but at the cost of spinning up the drive and writing the blocks at each write-replace. How often does your browser update a file while you surf? Every cache entry? Every history entry? What about your music player? Desktop manager? All of these will be spin up your disk drive.
Hiding behind POSIX is not the solution. There needs to be a solution that supports write-replace without spinning up the disk drive.
The ext4 people have kindly illuminated the problem. Now it is time to define a solution. Maybe it will be some sort of barrier logic, maybe a new kind of sync syscall. But it needs to be done.
but if you want a write later file system shouldn't it be restricted to hardware that can preserve it?
I understand that doing writes immediately when requested leads to performance degradation but that is why business systems which defer writes to disk only do so when the hardware can guarantee it. In other words, we have a battery backed cache, if the battery is low or nearing end of life the cache is turned off and all writes are made when the data changes.
Trying to make performance gains to overcome limitations of the hardware never wins out.
* Winners compare their achievements to their goals, losers compare theirs to that of others.
Rubbish. Sorry, if the syncs were implicit, app developers would just be demanding a way to to turn them off most of the time because they were killing performance.
And also consider - ext4 is relatively new, so it will improve over time. If you want stability stick to ext3 or ext2.
QFT
The filesystem was first released sometime towards the end of December 2008. The Linux distros that incorporated it, gave it as an option, but the default for /root and /home was always EXT3.
In addition, this problem is not a week old like the article states. People have been discussing this problem on forums ever since mid-January, when the benchmarks for EXT4 were published and several people decided to try it out to see how it fares. I have been using EXT4 for my /root partition since January. Fortunately I haven't had any data loss, but if I do end up losing some data, I'd understand that since I have been using a brand new file-system which has not been thoroughly tested by users, nor has it been used on any servers that I know of.
Face your daemons!
If you mount your ext4 partitions with nodelalloc you should be fine. You will of course no longer benefit from the performance enhancements that delayed allocation bring, but at least you'll have all of your freaking data. I'm running Debian on Linux 2.6.29-rc8-git4, and so far my limited testing has shown this to be very effective.
Never eat more than you can lift -- Miss Piggy
All the stuff with Ext4 strikes me as amazingly arrogant, and ignorant of the past. The issue that FS authors, well any authors of any system programs/tools/etc need to understand is that your tool being usable is the #1 important thing. In the case of a file system, that means that it reliably stores data on the drive. So, if you do something that really screws that over, well then you probably did it wrong. Doesn't matter if you fully documented it, doesn't matter if it technically "follows the spec" what matters is that it isn't usable.
I mean I could write a spec for a file system that says "No write is guaranteed to be written to disk until the OS is shut down, everything can be cached in RAM for an indefinite amount of time." However that'd be real flaky and lead to data loss. That makes my FS useless. Doesn't matter if it is well documented, what matters is that the damn thing loses data on a regular basis.
I'd give these guys more credit if I was aware of any other major OS/FS combo that did shit like this, but I'm not. Linux/Ext3 doesn't, Windows/NTFS doesn't, OS-X/HFS+ doesn't, Solaris/ZFS doesn't, etc. Well that tells me something. That says that the way they are doing things isn't a good idea. If it is causing problems AND it is something else nobody else does, then probably you ought not do it.
This is just bad design, in my opinion.
This is the problem with new features - the users have problems using them until they fully understands and appreciates the advantages and disadvantages.
Advantages: Filesystem benchmarks improve. Real performance... I guess that improves, too. Does anybody know?
Disadvantages: You risk data loss with 95% of the apps you use on a daily basis. This will persist until the apps are rewritten to force data commits at appropriate times, but hopefully not frequently enough to eat up all the performance improvements and more.
Ext4 might be great for servers (where crucial data is stored in databases, which are presumably written by storage experts who read the Posix spec), but what is the rationale for using it on the desktop? Ext4 has been coming for years, and everyone assumed it was the natural successor to ext3 for *all* contexts where ext3 is used, including desktops. I hope distros don't start using or recommending ext4 by default until they figure out how to configure it for safe usage on the desktop. (That will happen long before the apps are rewritten.) Filesystem benchmarks be damned.
If this were Microsoft, we'd give them a healthy helping of humble pie, but because it's Linux and the magic word "POSIX" gets used, I'm sure we'll forgive them for it.
You must be reading a different slashdot than I am. The popular opinion I see is that this is very bad design. If the spec allows this behavior, it's time to revisit the spec.
Give me Classic Slashdot or give me death!
Ext4, on the other hand, has another mechanism: delayed block allocation. After a file has been closed, up to a minute may elapse before data blocks on the disk are actually allocated. Delayed block allocation allows the filing system to optimise its write processes, but at the price that the metadata of a newly created file will display a size of 0 bytes and occupy no data blocks until the delayed allocation takes place. If the system crashes during this time, the rename() operation may already be committed in the journal, even though the new file still contains no data. The result is that after a crash the file is empty: both the old and the new data have been lost.
Ext4 developer Ted Ts'o stresses in his answer to the bug report that Ext4 behaves precisely as demanded by the POSIX standard for file operations.
If that is true, then to the extent that is true, POSIX is "broken". Related changes to a file system really need to take place in an orderly way. Creating a file, writing its data, and renaming it, are related. Letting the latter change persist while the former change is lost, is just wrong. Does POSIX really require this behavior, or just allow it? If it requires it, then IMHO, POSIX is indeed broken. And if POSIX is broken, then companies like Microsoft are vindicated in their non-conformance.
now we need to go OSS in diesel cars
The first question that came to mind when I read that is "why would the average application need to concern itself with filesystem details?"
They don't. Applications just need to concern themselves with the details of of the APIs they use, and the guarantees those APIs do or don't provide.
The POSIX file APIs specify quite clearly that there is no guarantee that your data is on the disk until you call fsync(). The problem is with applications that assumed they could ignore what the specification said just because it always seemed to work okay on the file systems they tested with.
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
"No write is guaranteed to be written to disk until the OS is shut down, everything can be cached in RAM for an indefinite amount of time." However that'd be real flaky and lead to data loss. That makes my FS useless. Doesn't matter if it is well documented, what matters is that the damn thing loses data on a regular basis.
It turns out that all the modern operative systems work exactly like that. In ALL of them you need to use explicit syncronization (fsync and friends) to get a notification that your data has really been written to disk (and that's all what you get, a notification, because the system could oops before fsync finishes). You also can mount your filesystem as "sync", which sucks.
Journaling, COW/transaction-based filesystems like ZFS only guarantee the integrity, not that your data is safe. It turns out that Ext3 has the same problem, it's just that the window is smaller (5 seconds). And I wouldn't bet that HFS and ZFS have not the same problem (btrfs is COW and transaction based, like ZFS, and has the same problem).
Welcome to the real world...
The first question that came to mind when I read that is "why would the average application need to concern itself with filesystem details?"
They don't. Applications just need to concern themselves with the details of of the APIs they use, and the guarantees those APIs do or don't provide.
The POSIX file APIs specify quite clearly that there is no guarantee that your data is on the disk until you call fsync(). The problem is with applications that assumed they could ignore what the specification said just because it always seemed to work okay on the file systems they tested with.
Thanks for explaining that. In that case, I salute Mr. Tso and others for telling the truth and not caving in to pressure when they are in fact correctly following the specification. Too often people who are correct don't have the fortitude to value that more than immediate convenience, so this is a refreshing thing to see. Perhaps this will become the sort of history with which developers are expected to be familiar.
I imagine it will take a lot of work but at least with Free Software this can be fixed. That's definitely what should happen, anyway. There are sometimes when things just go wrong no matter how correct your effort was; in those cases, it makes sense to just deal with the problem in the most hassle-free manner possible. This, however, is not one of those times. Thinking that you can selectively adhere to a standard and then claim that you are compliant with that standard is just the sort of thing that really should cause problems. Correcting the applications that made faulty assumptions is therefore the right way to deal with this, daunting and inconvenient though that may be.
Removing this delayed-allocation feature from ext4 or placing limits on it that are not required by the POSIX standard is definitely the wrong way to deal with this. To do so would surely invite more of the same. It would only encourage developers to believe that the standards aren't really important, that they'll just be "bailed out" if they fail to implement them. You don't need any sort of programming or system design expertise to understand that, just an understanding of how human beings operate and what they do with precedents that are set.
It is a miracle that curiosity survives formal education. - Einstein
Microsoft Patent
1) Modern filesystems are expected behave better than POSIX demands.
2) POSIX does not cover what should happen in a system crash at all.
3) The issue is not about saving data, but the atomicity of updates so that either the new data or the old data would be saved at all times.
4) fsync is not a solution, because ir forces the operation to complete *now*, which is counterproductive to write performance, cache coherence, laptop battery life, excessive SSD wear and a bunch of other reasons.
We don't need reliable data-on-disk-now, we need reliable old-or-new data without using a sledgehammer of fsync.
The POSIX standard is just fine. The problem is application assumptions that aren't up to snuff.
Read the qmail source code sometime. Every time the author wants to assure himself that data has been written to the disk, it calls fsync.
If you don't, you risk losing data. Plain and simple.
- Michael T. Babcock (Yes, I blog)
From the explanations I received and some reading I've done, I don't think the data is just getting "thrown away" so that isn't really a valid question. The issue seems to be that unless fsync is called, the changes requested by the application may happen in a sequence that is other than what the application programmer expected. The example I saw in this discussion involved first writing data to a file and then renaming it soon afterwards. If I understand this correctly, the application is assuming that the rename cannot possibly happen before the writing of the data is done even though the specification has no such requirement. If the application needs this to happen in the order in which it was requested, it needs to write the data, then call fsync, then rename the file. You could probably fill a library with what I don't know about low-level filesystem details, so please correct me if I have misunderstood this.
The example I found in the Wikipedia entry on ext4 was different. That one involved data loss because the application updates/overwrites an existing file and does not call fsync and then the system crashes. The Wiki article states that this leads to undefined behavior (which, afaik, is correct per the spec). The article also states that a typical result is that the file was set to zero-length in preparation for being overwritten but because of the crash, the new data was never written so it remains zero-length, causing the loss of the old version of the file. Under ext3 you would usually find either the old version of the file or the new version.
What I don't understand and hope that a more knowledgable person could explain is why this can't be done a slightly different way. This is where I can apply reason to come up with something that sounds preferable to me but I simply don't have the background knowledge of filesystems to understand the "why". If the overwrite of the file is delayed, why isn't the truncation of the file to zero-length also delayed? That is, instead of doing it this way:
Step 1: Truncate file length to zero in preparation of overwriting it.
Step 2: Delay the writing of the new data for performance reasons.
Step 3: After the delay has elapsed, actually write the data to the disk.
Why can't it be done this way instead?
Step 1: Delay the truncation of the file length to zero in preparation of overwriting it.
Step 2: Delay the writing of the new data.
Step 3: After the delay has elapsed, set the file length to zero and immediately write the new data, as a single operation if that is possible, or as one operation immediately followed by the other.
That way if there is a crash, you'd still get either the old version or the new one and not a zero-length file where data used to be. The only disadvantage I can see is that this might continue to enable developers to make assumptions that are not found in the standard because the buggy behavior ext4 is now exposing may continue to work. If there's no technical reason why it cannot be done that way, perhaps the bad precedent alone is a good reason to either not handle it this way or to change the spec.
It is a miracle that curiosity survives formal education. - Einstein
They don't. Applications just need to concern themselves with the details of of the APIs they use, and the guarantees those APIs do or don't provide.
Yup, and the problem has existed with KDE startup for years. I remember the startup files getting trashed when Mandrake first came out and I tried KDE for long enough to get hooked, and it's happened to me a few times a year ever since with every filesystem I've used. I just make my own backups of the .kde directory and fix this manually when it happens. I'm pretty good at this restore by now. Hopefully this bug in KDE will get fixed now that it is causing the KDE project such great embarrassment. I had a silent wish Tso would increase the default commit interval to 10 minutes when the first defenders of the KDE bug started squawking, but he's was too gracious for that.
PS I use a lot of experimental graphics drivers for work, hence lockups during startup are common enough that I probably see this KDE bug more than most KDE users. But they really violate every rule of using config files: 1st. open with minimum permission needed, in this case read only, unless a write is absolutely necessary. 2nd. only update a file when it needs updating. 3rd. when updating a config file make a copy, commit it to disk, and then replace the original, making sure file permissions and ownership are unchanged, then commit the rename if necessary.
PS2 Those computer users saying an fsync will kill performance need to get cluebat applied to them by the nearest programmer. 1st. There will be no fsyncs of config files at startup once the KDE startup is fixed. 2nd. fsyncs on modern filesystems are pretty fast, ext3 is the rare exception to that norm; this will be non-noticable when you apply a settings change. 3rd. These types of programming errors are not the norm; I've graded first and second year computer science classes and each of the three major mistakes made would have lost you 20-30% of your score for the assignment.
Kirk McKusick spent a lot of time working out the right order to write metadata and file data in FFS and the resulting file system, FFS with Soft Updates, gets high performance and high reliability... even after a crash.
The application programmers aren't at fault here, the POSIX spec is. A filesystem is essentially a hierarchical database, yet POSIX doesn't include a way to make atomic updates to it. The only tool provided is fsync, which kills performance if used. And even with fsync some things - such as rewriting a configuration file - are either outright impossible or complex and fragile.
The real solution is to come up with a transactional API for filesystem. Until that's done, problems like this will persist. Calling fsync - which forces a disk write - or playing around with temporary files isn't reasonable when all you want to do is make sure that the file will be updated properly or left alone.
The alternative is to have every program call fsync constantly, which not only kills performance, but ironically enough also negates some of Ext4's advantages, such as delayed block allocation, since it essentially disables write caching. And it doesn't work if you are doing more complex things, such as, say, mass renaming files in a directory; you have no way of ensuring that either they are all renamed, or none are.
Forget magic. Any technology distinguishable from divine power is insufficiently advanced.
1) Modern filesystems are expected behave better than POSIX demands.
2) POSIX does not cover what should happen in a system crash at all.
3) The issue is not about saving data, but the atomicity of updates so that either the new data or the old data would be saved at all times.
4) fsync is not a solution, because ir forces the operation to complete *now*, which is counterproductive to write performance, cache coherence, laptop battery life, excessive SSD wear and a bunch of other reasons.
We don't need reliable data-on-disk-now, we need reliable old-or-new data without using a sledgehammer of fsync.
1. POSIX is an API. It tries not to force the filesystem into being anything at all. So, for instance, you can write a filesystem that waits to do writes more efficiently to cut down on the wear of SSDs.
2. Ext3 has a max 5 second delay. That means this bug exists in Ext3 as well.
3. If you have important data that if not written to the hard drive will cause catastrophic failure, then you use the part of the API that forces that write.
4. Atomicity does not guarantee the filesystem be synchronized with cache. It means that during the update no other process can alter the affected file and that after the update the change will be seen by all other processes.
We don't need a filesystem that sledgehammers each and every byte of data to the hard drive just in case there is a crash. What we DO need is a filesystem that can flexibly handle important data when told it is important, and less important data very efficiently.
What you are asking is that the filesystem be some kind of sentient all knowing being that can tell when data is important or not and then can write important data immediately and non-important data efficiently. I think that it is a little better to have the application be the one that knows when it's dealing with important data or not.
fsyncs have other nasty side effects other than performance. For example, in Firefox 3, places.sqlite is fsynced after every page is loaded. For a laptop user, this behavior is unacceptable as it prevents the disks from staying spun down (not to mention the infuriating whine it creates to spin the disk up after every or nearly every page load). The use of fsync in Firefox 3 has actually caused some people (myself included), to mount ~/.mozilla as tmpfs and just write a cron job to write changed files back to disk once every 10 minutes.
So, while I'm all for applications using fsync when it's really needed, the last thing I'd like to see every application on the planet sprinkling their code with fsync "just to be sure".
KDE isn't fixed right now. Additionally, KDE is not the only application that generates lots of write activity. I work with real-time systems, and write performance on data collection systems is important.
I did some benchmarks on the ext3 file system, the ext4 system without the patch, and the ext4 system with the patch. Code followed the open(), write(), close() sequence was 76% faster than the code with fsync(). Code that followed the open(), write(), close(), rename() sequence was 28% faster than code with that followed the open(), write(), fsync(), close(), rename() sequence. Additionally, the benchmarks were not significantly affected by the presence which file system was used (ext3, ext4, or ext4 patched.) You can look up the spreadsheet and the discussion at the launchpad discussion.
Major Linux file backup utilities, like tar, gzip, and rsync don't use fsync as part of normal operations. The only application of the three, tar, that uses fsync, only uses it when verifying data is physically written to disk. In that situation, it writes the data, calls fsync, calls ioctl(FDFLUSH), and the reads the data back. Strictly speaking, that is the only way to make sure the file is written to disk, and is readable.
Finally, as Theodore Ts'o has pointed out, if you really want to make sure the file is saved to disk, you also have to fsync() the directory too. I have never seen anyone do that, as part of a normal file save. Most C programming textbooks simply have fopen, fwrite, fclose as the recommended way to save files. Calling fsync this often is unusual for most C programmers.
I would hate to be in your programming class. Your enforcing programming standards that aren't followed by key Linux utilities, aren't in most textbooks, and aren't portable to non-Linux file systems.
If you require your students to fsync() the file and the directory, as part of a normal assignment, you are requiring them to do things that aren't done by any Linux utility out there. Further, if you are that paranoid, you better follow the example from the tar utility, and after the fsync completes, read all the data back to verify it was successfully written.
You don't understand the problem.
You are wrong when you say EXT3 has this problem. It does not have it. If the EXT3 system crashes during those 5 seconds, you either get the old file or the new one. For EXT4, if it crashes, you can get a zero-length file, with both the old and new data lost.
The long delay is irrelevant and is confusing people about this bug. In fact the long delay is very nice in EXT4 as it means it is much more efficient and will use less power. I don't really mind if a crash during this time means I lose the new version of a file. But PLEASE don't lose the old one as well!!! That is inexcusable, and I don't care if the delay is .1 second.
It's really depressing that there are so many clueless comments in Slashdot --- but I guess I shouldn't be surprised.
Patches to work around buggy applications which don't call fsync() have been around long before this issue got slashdotted, and before the Ubuntu Laundpad page got slammed with comments. In fact, I commented very early in the Ubuntu log that patches that detected the buggy applications and implicitly forced the disk blocks to disk were already available. Since then, both Fedora and Ubuntu are shipping with these workaround patches.
And yet, people are still saying that ext4 is broken, and will never work, and that I'm saying all of this so that I don't have to change my code, etc ---- when in fact I created the patches to work around the broken applications *first*, and only then started trying to advocate that people fix their d*mn broken applications.
If you want to make your applications such that they are only safe on Linux and ext3/ext4, be my guest. The workaround patches are all you need for ext4. The fixes have been queued for 2.6.30 as soon as its merge window opens (probably in a week or so), and Fedora and Ubuntu have already merged them into their kernels for their beta releases which will be released in April/May. They will slow down filesystem performance in a few rare cases for properly written applications, so if you have a system that is reliable, and runs on a UPS, you can turn off the workaround patches with a mount option.
Applications that rely on this behaviour won't necessarily work well on other operating systems, and on other filesystems. But if you only care about Linux and ext3/ext4 file systems, you don't have to change anything. I will still reserve the right to call them broken, though.