Re: Ordering of directory operations maintained across system crashes in Btrfs?
On Mon, Mar 03, 2014 at 11:56:49AM -0600, thanumalayan mad wrote: Chris, Great, thanks. Any guesses whether other filesystems (disk-based) do things similar to the last two examples you pointed out? Saying we think 3 normal filesystems reorder stuff seems to motivate application developers to fix bugs ... Also, just for more information, the sequence we observed was, Thread A: unlink(foo) rename(somefile X, somefile Y) fsync(somefile Z) The source and destination of the renamed file are unrelated to the fsync. But the rename happens in the fsync()'s transaction, while unlink() is delayed. I guess this has something to do with backrefs too. Thanks, Thanu On Mon, Mar 3, 2014 at 11:43 AM, Chris Mason c...@fb.com wrote: On 02/25/2014 09:01 PM, thanumalayan mad wrote: Hi all, Slightly complicated question. Assume I do two directory operations in a Btrfs partition (such as an unlink() and a rename()), one after the other, and a crash happens after the rename(). Can Btrfs (the current version) send the second operation to the disk first, so that after the crash, I observe the effects of rename() but not the effects of the unlink()? I think I am observing Btrfs re-ordering an unlink() and a rename(), and I just want to confirm that my observation is true. Also, if Btrfs does send directory operations to disk out of order, is there some limitation on this? Like, is this restricted to only unlink() and rename()? I am looking at some (buggy) applications that use Btrfs, and this behavior seems to affect them. There isn't a single answer for this one. You might have Thread A: ulink(foo); rename(somefile, somefile2); crash This should always have the rename happen before or in the same transaction as the rename. Thread A: ulink(dirA/foo); rename(dirB/somefile, dirB/somefile2); Here you're at the mercy of what is happening in dirB. If someone fsyncs that directory, it may hit the disk before the unlink. Thread A: ulink(foo); rename(somefile, somefile2); fsync(somefile); This one is even fuzzier. Backrefs allow us to do some file fsyncs without touching the directory, making it possible the unlink will hit disk after the fsync. -chris As I understand it POSIX only garanties that the in-core data is updated by the syscalls in-order. On crash anything can happen. If the application needs something to be commited to disk then it needs to fsync(). Specifically it needs to fsync() the changed files AND directories. From man fsync: Calling fsync() does not necessarily ensure that the entry in the directory containing the file has also reached disk. For that an explicit fsync() on a file descriptor for the directory is also needed. So the fsync(somefile) above doesn't necessarily force the rename to disk. My experience with fuse tells me that at least fuse handles operations in parallel and only blocks a later operation if it is affected by an earlier operation. An unlink in one directory can (and will) run in parallel to a rename in another directory. Then, depending on how threads get scheduled, the rename can complete before the unlink. My conclusion is that you need to fsync() the directory to ensure the metadata update has made it to the disk if you require that. Otherwise you have to be able to cope with (meta)data loss on crash. Note: https://code.google.com/p/leveldb/issues/detail?id=189 talks a lot about journaling and that any yournaling filesystem should preserve the order. I think that is rather pointless for two reasons: 1) The journal gets replayed after a crash so in whatever order the two journal entries are written doesn't matter. They both make it to disk. You can't see one without the other. This is assuming you fsync()ed the dirs so force the metadata change into the journal in the first place. 2) btrfs afaik doesn't have any journal since COW already garanties atomic updates and crash protection. Overall I also think the fear of fsync() is overrated for this issue. This would only happen on programm start or whenever you open a database. Not somthing that happens every second. MfG Goswin -- To unsubscribe from this list: send the line unsubscribe linux-btrfs in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: What to do about df and btrfs fi df
On Mon, Feb 17, 2014 at 06:08:20PM +0100, David Sterba wrote: On Mon, Feb 10, 2014 at 01:41:23PM -0500, Josef Bacik wrote: On 02/10/2014 01:36 PM, cwillu wrote: IMO, used should definitely include metadata, especially given that we inline small files. I can convince myself both that this implies that we should roll it into b_avail, and that we should go the other way and only report the actual used number for metadata as well, so I might just plead insanity here. I could be convinced to do this. So we have total: (total disk bytes) / (raid multiplier) used: (total used in data block groups) + (total used in metadata block groups) avail: total - (total used in data block groups + total metadata block groups) The size of global block reserve should be IMO subtracted from 'avail', this reports the space as free, but is in fact not. How much global block reserve is there? Does that explain why I can't use the last 270G of my 19TB btrfs? The used amount of the global reserve might be included into filesystem 'used', but I've observed the global reserve used for short periods of time under some heavy stress, I'm convinced it needs to be accounted in the df report. As a comparison the ext2/3/4 filesystem has a % reserved for root and does not show this in available. So you get filesystem with 0 bytes free but root can still write to them. I would argue that available should not include the reserve. It is not available for normal operations, right? MfG Goswin -- To unsubscribe from this list: send the line unsubscribe linux-btrfs in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: know mount location with in FS
On Tue, Feb 18, 2014 at 11:06:38AM +0800, Anand Jain wrote: For what reason? Remember that a single block device can be mounted in multiple places (or bind-mounted, etc), so there is not even necessarily a single answer to that question. -Eric Yes indeed. (the attempt is should we be able to maintain all the mount points as a list saved/updated under per fs_devices. ?) some of the exported symbols at fs/namei.c looks closely related to the purpose here, but it didn't help unless I missed something. any comment is helpful.. The reason: First of all btrfs-progs has used scan-all-disks very liberally which isn't a scalable design (imagine a data center with 1000's of LUN). Even a simple check_mounted() does scan-all-disks (when total_disk 1), that isn't necessary if the kernel could let it know. Scan for btrfs has expensive steps of reading each super-block, and the effect is, in general most of the btrfs-progs commands are very very slow when things like scrub is running. check_mounted() fails when seeding is used (since /proc/self/mounts would show disk with lowest devid and in most common scenario it will be a seed disk. (which has different FSID from the actual disk in question). and Further most severe problem is some btrfs-progs threads has been scan-all-disks more than once during the thread's life time. So a total revamp of this design has become an immediate need. What I am planning is - btrfs-progs to init btrfs-disk-list once per required thread (mostly use BTRFS_IOC_GET_DEVS, which would dump anything and everything about the btrfs devices) - the btrfs-disk-list is obtained from kernel first, and will fill with the remaining disks which kernel isn't aware of. - If the step one also provides the mount point(s) from the kernel that would complete the loop with what end user would want to know. Thanks, Anand What about mountpoints outside the current filesystem namespace or ones that should be shortened to the filesystem namespace (e.g. in a chroot the leading dirs need to be cut)? MfG Goswin -- To unsubscribe from this list: send the line unsubscribe linux-btrfs in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: ENOSPC with 270GiB free
On Tue, Feb 18, 2014 at 08:58:10AM -0500, Josef Bacik wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 02/16/2014 08:58 AM, Goswin von Brederlow wrote: Hi, I'm getting a ENOSPC error from btrfs despite there still being plenty of space left: % df -m /mnt/nas3 Filesystem 1M-blocks Used Available Use% Mounted on /dev/mapper/nas3-a 19077220 18805132270773 99% /mnt/nas3 % btrfs fi show Label: none uuid: 4b18f84e-2499-41ca-81ff-fe1783c11491 Total devices 1 FS bytes used 17.91TiB devid1 size 18.19TiB used 17.94TiB path /dev/mapper/nas3-a Btrfs v3.12 % btrfs fi df Data, single: total=17.89TiB, used=17.88TiB System, DUP: total=32.00MiB, used=1.92MiB Metadata, DUP: total=25.50GiB, used=24.89GiB As you can see there are still 270GiB free and plenty of block groups free on the device too. So why isn't btrfs allocating a new block group to store more data? What kernel? Can you give btrfs-next a try? Mount with -o enospc_debug and when you get enospc send the dmesg. Thanks, Josef Standard Debian linux kernel: Linux nas3 3.12-1-amd64 #1 SMP Debian 3.12.9-1 (2014-02-01) x86_64 GNU/Linux Compiling new kernel and rebooting will take some time. But it looks like I can remount with enospc_debug. I will try if that outputs anything usefull first. So far I got: [258988.006643] btrfs: disk space caching is enabled MfG Goswin -- To unsubscribe from this list: send the line unsubscribe linux-btrfs in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: ENOSPC with 270GiB free
On Mon, Feb 17, 2014 at 08:42:23AM +0100, Dan van der Ster wrote: Did you already try this?? [1]: btrfs fi balance start -dusage=5 /mnt/nas3 Cheers, dan [1] https://btrfs.wiki.kernel.org/index.php/Problem_FAQ#I_get_.22No_space_left_on_device.22_errors.2C_but_df_says_I.27ve_got_lots_of_space On Sun, Feb 16, 2014 at 2:58 PM, Goswin von Brederlow goswin-...@web.de wrote: Hi, I'm getting a ENOSPC error from btrfs despite there still being plenty of space left: % df -m /mnt/nas3 Filesystem 1M-blocks Used Available Use% Mounted on /dev/mapper/nas3-a 19077220 18805132270773 99% /mnt/nas3 % btrfs fi show Label: none uuid: 4b18f84e-2499-41ca-81ff-fe1783c11491 Total devices 1 FS bytes used 17.91TiB devid1 size 18.19TiB used 17.94TiB path /dev/mapper/nas3-a Btrfs v3.12 % btrfs fi df Data, single: total=17.89TiB, used=17.88TiB System, DUP: total=32.00MiB, used=1.92MiB Metadata, DUP: total=25.50GiB, used=24.89GiB As you can see there are still 270GiB free and plenty of block groups free on the device too. So why isn't btrfs allocating a new block group to store more data? MfG Goswin I did and that isn't the problem. Balancing only frees up partially used block groups so they can be reused. But the problem is that the remaining free block groups are not getting used. MfG Goswin -- To unsubscribe from this list: send the line unsubscribe linux-btrfs in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: btrfsck does not fix
On Mon, Feb 17, 2014 at 03:20:58AM +, Duncan wrote: Chris Murphy posted on Sun, 16 Feb 2014 12:54:44 -0700 as excerpted: Also, 10 hours to balance two disks at 2.3TB seems like a long time. I'm not sure if that's expected. FWIW, I think you may not realize how big 2.3 TiB is, and/or how slow spinning rust can be when dealing with TiBs of potentially fragmented data... 2.3TiB * 1024GiB/TiB * 1024 MiB/GiB / 10 hours / 60 min/hr / 60 sec/min = 66.99... real close to 67 MiB/sec Since it's multiple TiB we're talking and only two devices, that's almost certainly spinning rust, not SSD, and on spinning rust, 67 MiB/sec really isn't /that/ bad, especially if the filesystem wasn't new and had been reasonably used, thus likely had some fragmentation to deal with. Don't forget that that is 67MiB/s reading data and 67MiB/s writing data giving a total of 134MiB/s. Still, on a good system each disk should have about that speed so it's about 50% of theoretical maximum. Which is quite good given that the disks will need to seek between every read and write. In comparison moving data with LVM gets only about half that speed and that doesn't even have the overhead of a filesystem to deal with. MfG Goswin -- To unsubscribe from this list: send the line unsubscribe linux-btrfs in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
ENOSPC with 270GiB free
Hi, I'm getting a ENOSPC error from btrfs despite there still being plenty of space left: % df -m /mnt/nas3 Filesystem 1M-blocks Used Available Use% Mounted on /dev/mapper/nas3-a 19077220 18805132270773 99% /mnt/nas3 % btrfs fi show Label: none uuid: 4b18f84e-2499-41ca-81ff-fe1783c11491 Total devices 1 FS bytes used 17.91TiB devid1 size 18.19TiB used 17.94TiB path /dev/mapper/nas3-a Btrfs v3.12 % btrfs fi df Data, single: total=17.89TiB, used=17.88TiB System, DUP: total=32.00MiB, used=1.92MiB Metadata, DUP: total=25.50GiB, used=24.89GiB As you can see there are still 270GiB free and plenty of block groups free on the device too. So why isn't btrfs allocating a new block group to store more data? MfG Goswin -- To unsubscribe from this list: send the line unsubscribe linux-btrfs in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: ENOSPC with 270GiB free
On Sun, Feb 16, 2014 at 06:18:38PM +, Duncan wrote: Goswin von Brederlow posted on Sun, 16 Feb 2014 14:58:08 +0100 as excerpted: As you can see there are still 270GiB free and plenty of block groups free on the device too. So why isn't btrfs allocating a new block group to store more data? I saw this on a much (much) smaller filesystem a few weeks ago, when I redid my /boot. In my case it was under a gig total, so mixed-mode, but copying files over in a particular order errored some of them out with ENOSPC. But the way I was copying (using mc) left the ones that hadn't copied selected, and I tried a copy of them again, and/or used mc's directory-diff to find the missing files and copy them over again. After about three times, they all copied. So some combination of size and metadata wasn't triggering a new block allocation, but coming in a different order, it triggered fine. Again, this was mixed-mode, so data/metadata blocks mixed, and it didn't matter which ran out first since they were combined. I wonder if you're running into something similar. Can you try doing the copy in a different order, or is it one big file? I'm using rsync and towards the last few GB before it gives ENOSPC the filesystem gets realy slow and eats more and more cpu time. I'm copying multi gigabyte files and the second last file managed 2MB/s and the failed one came down to ~100K/s at the end. So trying a lot of different files isn't realy feasable, timewise. MfG Goswin -- To unsubscribe from this list: send the line unsubscribe linux-btrfs in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html