public inbox for pbs-devel@lists.proxmox.com
 help / color / mirror / Atom feed
From: Christian Ebner <c.ebner@proxmox.com>
To: "Proxmox Backup Server development discussion"
	<pbs-devel@lists.proxmox.com>,
	"Fabian Grünbichler" <f.gruenbichler@proxmox.com>
Subject: Re: [pbs-devel] [PATCH proxmox-backup 4/4] fix #5853: client: pxar: exclude stale files on metadata read
Date: Wed, 13 Nov 2024 14:45:55 +0100	[thread overview]
Message-ID: <39def709-3df0-4182-8179-8be4022f0f9e@proxmox.com> (raw)
In-Reply-To: <1731330267.220wgkqhtc.astroid@yuna.none>

On 11/11/24 14:37, Fabian Grünbichler wrote:
> behaviour wise this seems okay to me, but if possible, I'd avoid all the
> return value tuples, see detailed comments below..

Agreed, I am not a fan of passing the stale file handle error info along 
a the boolean in the return value as well.

But unfortunately passing along the error without loosing pre-existing 
error context and switching all the `get_metadata` related functions to 
return an `Errno` is not possible as is.

E.g. `process_acl` returns an `anyhow::Error` (could be defined to 
return an e.g. `Errno::EINVALID` instead?), special handling of 
`Errno::E2BIG` for the xattr case only, ...

The current approach was choosen to keep the current anyhow error 
context close to the actual errors when they occur.
  > On November 5, 2024 3:01 pm, Christian Ebner wrote:
>> Skip and warn the user for files which returned a stale file handle
>> error while reading the metadata associated to that file.
>>
>> Instead of returning with an error when getting the metadata, return
>> a boolean flag signaling if a stale file handle has been encountered.
>>
>> Link to issue in bugtracker:
>> https://bugzilla.proxmox.com/show_bug.cgi?id=5853
>>
>> Link to thread in community forum:
>> https://forum.proxmox.com/threads/156822/
>>
>> Signed-off-by: Christian Ebner <c.ebner@proxmox.com>
>> ---
>>   pbs-client/src/pxar/create.rs | 100 ++++++++++++++++++++++------------
>>   1 file changed, 66 insertions(+), 34 deletions(-)
>>
>> diff --git a/pbs-client/src/pxar/create.rs b/pbs-client/src/pxar/create.rs
>> index 2a844922c..85be00db4 100644
>> --- a/pbs-client/src/pxar/create.rs
>> +++ b/pbs-client/src/pxar/create.rs
>> @@ -228,7 +228,7 @@ where
>>       let mut fs_feature_flags = Flags::from_magic(fs_magic);
>>   
>>       let stat = nix::sys::stat::fstat(source_dir.as_raw_fd())?;
>> -    let metadata = get_metadata(
>> +    let (metadata, stale_fd) = get_metadata(
> 
> stale_fd here is not used at all..

Yes, that one should lead to a hard error as you mentioned, so should be 
handled accordingly. I will adapt this to be a hard error.

> 
>>           source_dir.as_raw_fd(),
>>           &stat,
>>           feature_flags & fs_feature_flags,
>> @@ -744,7 +744,7 @@ impl Archiver {
>>               return Ok(());
>>           }
>>   
>> -        let metadata = get_metadata(
>> +        let (metadata, stale_fd) = get_metadata(
> 
> this one is used
> 
>>               fd.as_raw_fd(),
>>               stat,
>>               self.flags(),
>> @@ -753,6 +753,11 @@ impl Archiver {
>>               self.skip_e2big_xattr,
>>           )?;
>>   
>> +        if stale_fd {
>> +            log::warn!("Stale filehandle encountered, skip {:?}", self.path);
>> +            return Ok(());
>> +        }
> 
> for this warning.. but get_metadata already logs (potentially multiple
> times ;)) that things are incomplete cause of the stale filehandle, this
> only adds the path context..

but here there is also an early return, not just the log... this skips 
over adding this entry, and any sub entries if the entry is a directory.

The logging could however be moved to the get_metadata call and only be 
logged once, agreed.

> 
>> +
>>           if self.previous_payload_index.is_none() {
>>               return self
>>                   .add_entry_to_archive(encoder, &mut None, c_file_name, stat, fd, &metadata, None)
>> @@ -1301,7 +1306,14 @@ impl Archiver {
>>           file_name: &Path,
>>           metadata: &Metadata,
>>       ) -> Result<(), Error> {
>> -        let dest = nix::fcntl::readlinkat(fd.as_raw_fd(), &b""[..])?;
>> +        let dest = match nix::fcntl::readlinkat(fd.as_raw_fd(), &b""[..]) {
>> +            Ok(dest) => dest,
>> +            Err(Errno::ESTALE) => {
>> +                log::warn!("Stale file handle encountered, skip {file_name:?}");
>> +                return Ok(());
>> +            }
>> +            Err(err) => return Err(err.into()),
>> +        };
>>           encoder.add_symlink(metadata, file_name, dest).await?;
>>           Ok(())
>>       }
>> @@ -1397,9 +1409,10 @@ fn get_metadata(
>>       fs_magic: i64,
>>       fs_feature_flags: &mut Flags,
>>       skip_e2big_xattr: bool,
>> -) -> Result<Metadata, Error> {
>> +) -> Result<(Metadata, bool), Error> {
>>       // required for some of these
>>       let proc_path = Path::new("/proc/self/fd/").join(fd.to_string());
>> +    let mut stale_fd = false;
>>   
>>       let mut meta = Metadata {
>>           stat: pxar::Stat {
>> @@ -1412,18 +1425,27 @@ fn get_metadata(
>>           ..Default::default()
>>       };
>>   
>> -    get_xattr_fcaps_acl(
>> +    if get_xattr_fcaps_acl(
> 
> only call site, could just bubble up ESTALE

As mentioned, this has 2 issues: Loss of anyhow error context for which 
sub-function the Errno occurred and sub-functions like `process_acl` 
which do not rely on ffi calls at all, returning plain `anyhow::Error`, 
which granted could be redefined to return an Errno.

> 
>>           &mut meta,
>>           fd,
>>           &proc_path,
>>           flags,
>>           fs_feature_flags,
>>           skip_e2big_xattr,
>> -    )?;
>> -    get_chattr(&mut meta, fd)?;
>> +    )? {
>> +        stale_fd = true;
>> +        log::warn!("Stale filehandle, xattrs incomplete");
>> +    }
>> +    if get_chattr(&mut meta, fd)? {
> 
> same
> 
>> +        stale_fd = true;
>> +        log::warn!("Stale filehandle, chattr incomplete");
>> +    }
>>       get_fat_attr(&mut meta, fd, fs_magic)?;
>> -    get_quota_project_id(&mut meta, fd, flags, fs_magic)?;
>> -    Ok(meta)
>> +    if get_quota_project_id(&mut meta, fd, flags, fs_magic)? {
> 
> same
> 
>> +        stale_fd = true;
>> +        log::warn!("Stale filehandle, quota  project id incomplete");
>> +    }
> 
> see above and way down below, IMHO all of these could just bubble up the error..
> 
>> +    Ok((meta, stale_fd))
>>   }
>>   
>>   fn get_fcaps(
>> @@ -1431,22 +1453,23 @@ fn get_fcaps(
>>       fd: RawFd,
>>       flags: Flags,
>>       fs_feature_flags: &mut Flags,
>> -) -> Result<(), Error> {
>> +) -> Result<bool, Error> {
> 
> this is only called by get_xattr_fcaps_acl, so could just bubble up
> ESTALE as well..
> 
>>       if !flags.contains(Flags::WITH_FCAPS) {
>> -        return Ok(());
>> +        return Ok(false);
>>       }
>>   
>>       match xattr::fgetxattr(fd, xattr::XATTR_NAME_FCAPS) {
>>           Ok(data) => {
>>               meta.fcaps = Some(pxar::format::FCaps { data });
>> -            Ok(())
>> +            Ok(false)
>>           }
>> -        Err(Errno::ENODATA) => Ok(()),
>> +        Err(Errno::ENODATA) => Ok(false),
>>           Err(Errno::EOPNOTSUPP) => {
>>               fs_feature_flags.remove(Flags::WITH_FCAPS);
>> -            Ok(())
>> +            Ok(false)
>>           }
>> -        Err(Errno::EBADF) => Ok(()), // symlinks
>> +        Err(Errno::EBADF) => Ok(false), // symlinks
>> +        Err(Errno::ESTALE) => Ok(true),
>>           Err(err) => Err(err).context("failed to read file capabilities"),
>>       }
>>   }
>> @@ -1458,32 +1481,35 @@ fn get_xattr_fcaps_acl(
>>       flags: Flags,
>>       fs_feature_flags: &mut Flags,
>>       skip_e2big_xattr: bool,
>> -) -> Result<(), Error> {
>> +) -> Result<bool, Error> {
>>       if !flags.contains(Flags::WITH_XATTRS) {
>> -        return Ok(());
>> +        return Ok(false);
>>       }
>>   
>>       let xattrs = match xattr::flistxattr(fd) {
>>           Ok(names) => names,
>>           Err(Errno::EOPNOTSUPP) => {
>>               fs_feature_flags.remove(Flags::WITH_XATTRS);
>> -            return Ok(());
>> +            return Ok(false);
>>           }
>>           Err(Errno::E2BIG) => {
>>               match skip_e2big_xattr {
>> -                true => return Ok(()),
>> +                true => return Ok(false),
>>                   false => {
>>                       bail!("{} (try --skip-e2big-xattr)", Errno::E2BIG.to_string());
>>                   }
>>               };
>>           }
>> -        Err(Errno::EBADF) => return Ok(()), // symlinks
>> +        Err(Errno::EBADF) => return Ok(false), // symlinks
>> +        Err(Errno::ESTALE) => return Ok(true),
> 
> see above
> 
>>           Err(err) => return Err(err).context("failed to read xattrs"),
>>       };
>>   
>>       for attr in &xattrs {
>>           if xattr::is_security_capability(attr) {
>> -            get_fcaps(meta, fd, flags, fs_feature_flags)?;
>> +            if get_fcaps(meta, fd, flags, fs_feature_flags)? {
>> +                return Ok(true);
> 
> see above
> 
>> +            }
>>               continue;
>>           }
>>   
>> @@ -1505,35 +1531,37 @@ fn get_xattr_fcaps_acl(
>>               Err(Errno::EBADF) => (),   // symlinks, shouldn't be able to reach this either
>>               Err(Errno::E2BIG) => {
>>                   match skip_e2big_xattr {
>> -                    true => return Ok(()),
>> +                    true => return Ok(false),
>>                       false => {
>>                           bail!("{} (try --skip-e2big-xattr)", Errno::E2BIG.to_string());
>>                       }
>>                   };
>>               }
>> +            Err(Errno::ESTALE) => return Ok(true), // symlinks
> 
> same here (and stray copy-paste comment I guess?)
> 
>>               Err(err) => {
>>                   return Err(err).context(format!("error reading extended attribute {attr:?}"))
>>               }
>>           }
>>       }
>>   
>> -    Ok(())
>> +    Ok(false)
>>   }
>>   
>> -fn get_chattr(metadata: &mut Metadata, fd: RawFd) -> Result<(), Error> {
>> +fn get_chattr(metadata: &mut Metadata, fd: RawFd) -> Result<bool, Error> {
>>       let mut attr: libc::c_long = 0;
>>   
>>       match unsafe { fs::read_attr_fd(fd, &mut attr) } {
>>           Ok(_) => (),
>> +        Err(Errno::ESTALE) => return Ok(true),
>>           Err(errno) if errno_is_unsupported(errno) => {
>> -            return Ok(());
>> +            return Ok(false);
>>           }
>>           Err(err) => return Err(err).context("failed to read file attributes"),
>>       }
>>   
>>       metadata.stat.flags |= Flags::from_chattr(attr).bits();
>>   
>> -    Ok(())
>> +    Ok(false)
>>   }
>>   
>>   fn get_fat_attr(metadata: &mut Metadata, fd: RawFd, fs_magic: i64) -> Result<(), Error> {
>> @@ -1564,30 +1592,34 @@ fn get_quota_project_id(
>>       fd: RawFd,
>>       flags: Flags,
>>       magic: i64,
>> -) -> Result<(), Error> {
>> +) -> Result<bool, Error> {
> 
> see above
> 
>>       if !(metadata.is_dir() || metadata.is_regular_file()) {
>> -        return Ok(());
>> +        return Ok(false);
>>       }
>>   
>>       if !flags.contains(Flags::WITH_QUOTA_PROJID) {
>> -        return Ok(());
>> +        return Ok(false);
>>       }
>>   
>>       use proxmox_sys::linux::magic::*;
>>   
>>       match magic {
>>           EXT4_SUPER_MAGIC | XFS_SUPER_MAGIC | FUSE_SUPER_MAGIC | ZFS_SUPER_MAGIC => (),
>> -        _ => return Ok(()),
>> +        _ => return Ok(false),
>>       }
>>   
>>       let mut fsxattr = fs::FSXAttr::default();
>>       let res = unsafe { fs::fs_ioc_fsgetxattr(fd, &mut fsxattr) };
>>   
>> +    if let Err(Errno::ESTALE) = res {
>> +        return Ok(true);
>> +    }
>> +
>>       // On some FUSE filesystems it can happen that ioctl is not supported.
>>       // For these cases projid is set to 0 while the error is ignored.
>>       if let Err(errno) = res {
>>           if errno_is_unsupported(errno) {
>> -            return Ok(());
>> +            return Ok(false);
>>           } else {
>>               return Err(errno).context("error while reading quota project id");
>>           }
>> @@ -1597,7 +1629,7 @@ fn get_quota_project_id(
>>       if projid != 0 {
>>           metadata.quota_project_id = Some(pxar::format::QuotaProjectId { projid });
>>       }
>> -    Ok(())
>> +    Ok(false)
>>   }
>>   
>>   fn get_acl(
>> @@ -1840,7 +1872,7 @@ mod tests {
>>           let fs_magic = detect_fs_type(dir.as_raw_fd()).unwrap();
>>           let stat = nix::sys::stat::fstat(dir.as_raw_fd()).unwrap();
>>           let mut fs_feature_flags = Flags::from_magic(fs_magic);
>> -        let metadata = get_metadata(
>> +        let (metadata, _) = get_metadata(
> 
> no use of the new return value
> 
>>               dir.as_raw_fd(),
>>               &stat,
>>               fs_feature_flags,
>> @@ -1937,7 +1969,7 @@ mod tests {
>>               let stat = nix::sys::stat::fstat(source_dir.as_raw_fd()).unwrap();
>>               let mut fs_feature_flags = Flags::from_magic(fs_magic);
>>   
>> -            let metadata = get_metadata(
>> +            let (metadata, _) = get_metadata(
> 
> no use either.. so wouldn't it make more sense to pass in a path and log
> the context right in get_metadata? or treat the stale FD as an error,
> and add the context/path as part of error handling?

The first approach seems better, will however not help to differentiate 
the (hard) errors from the soft error ESTALE, which requires to skip 
over entries at the `get_metadata` call side conditionally.

Returning the stale file handle error as `Anyhow::Error` also does not 
allow to distinguish from other (hard) errors, so again it cannot be 
handled as soft error at the call site.

And returning all errors as `Errno` has the loss of error context issue 
as described above.

I will see if I can cover this better by refactoring the code, as most 
of the helpers have a single call side, so it should be possible to 
reorganize without much side effects.

> 
> the four call sites are:
> - two related to tests, we can probably treat ESTALE as hard error there
> - the one for obtaining the metadata of the source dir of the archive,
>    if that is stale we can't create an archive -> hard error as well
> - adding an entry: for the stale case, we already log a warning and
>    proceed with the next entry, so we don't benefit from the fact that
>    (incomplete) metadata and the staleness is returned, as opposed to
>    just treating ESTALE as an error that we can "catch" and handle..
> 
>>                   source_dir.as_raw_fd(),
>>                   &stat,
>>                   fs_feature_flags,
>> -- 
>> 2.39.5
>>
>>
>>
>> _______________________________________________
>> pbs-devel mailing list
>> pbs-devel@lists.proxmox.com
>> https://lists.proxmox.com/cgi-bin/mailman/listinfo/pbs-devel
>>
>>
>>
> 
> 
> _______________________________________________
> pbs-devel mailing list
> pbs-devel@lists.proxmox.com
> https://lists.proxmox.com/cgi-bin/mailman/listinfo/pbs-devel
> 
> 



_______________________________________________
pbs-devel mailing list
pbs-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pbs-devel

  reply	other threads:[~2024-11-13 13:46 UTC|newest]

Thread overview: 11+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-11-05 14:01 [pbs-devel] [PATCH proxmox-backup 0/4] fix #5853: ignore stale files Christian Ebner
2024-11-05 14:01 ` [pbs-devel] [PATCH proxmox-backup 1/4] client: pxar: skip directories on stale file handle Christian Ebner
2024-11-05 14:01 ` [pbs-devel] [PATCH proxmox-backup 2/4] client: pxar: skip directory entries " Christian Ebner
2024-11-05 14:01 ` [pbs-devel] [PATCH proxmox-backup 3/4] client: pxar: warn user and ignore stale file handles on file open Christian Ebner
2024-11-11 13:37   ` Fabian Grünbichler
2024-11-05 14:01 ` [pbs-devel] [PATCH proxmox-backup 4/4] fix #5853: client: pxar: exclude stale files on metadata read Christian Ebner
2024-11-11 13:37   ` Fabian Grünbichler
2024-11-13 13:45     ` Christian Ebner [this message]
2024-11-13 13:55       ` Fabian Grünbichler
2024-11-13 14:04         ` Christian Ebner
2024-11-14 14:43 ` [pbs-devel] [PATCH proxmox-backup 0/4] fix #5853: ignore stale files Christian Ebner

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=39def709-3df0-4182-8179-8be4022f0f9e@proxmox.com \
    --to=c.ebner@proxmox.com \
    --cc=f.gruenbichler@proxmox.com \
    --cc=pbs-devel@lists.proxmox.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
Service provided by Proxmox Server Solutions GmbH | Privacy | Legal