git-annex/Annex/Link.hs
Joey Hess 6a9e923c74
fix handling of linked worktrees on filesystems w/o symlinks
Fix bug in handling of linked worktrees on filesystems not supporting
symlinks, that caused annexed file content to be stored in the wrong
location inside the git directory, and also caused pointer files to not get
populated.

This parameterizes functions in Annex.Locations with a GitLocationMaker.
The uses of standardGitLocationMaker are in cases where the path returned
by a function should not change when in a linked worktree. For example,
gitAnnexLink uses standardGitLocationMaker because symlink targets should
always be to ".git/annex/objects" paths, even when in a linked worktree.
Hopefully I have gotten all uses of standardGitLocationMaker right.

This also assumes that all path construction to the annex directory
is done via the functions in Annex.Locations, and there is no other,
ad-hoc construction elsewhere. Thankfully, Annex.Locations has been around
since the beginning, and has been used consistently. I think.

---

In fixupUnusualRepos, when symlinks are supported, the .git file is replaced
with a symlink to the linked worktree git directory. And in that directory,
an "annex" symlink points to the main annex directory. In that case,
it's not necessary to set mainWorkTreePath. It would be ok to set it,
but not setting it in that case allows an optimisation of avoiding reading
the "commondir" file.

The change to make fixupUnusualRepos set mainWorkTreePath when the
repository is not initialized yet is done in case the initialization itself
writes to the annex directory. If that were the case, without setting
mainWorkTreePath, the annex symlink would not be set up yet, and so
it might have created the annex directory in the wrong place. Currently
that didn't happen, but now that mainWorkTreePath is available, using it
here avoids any such later problem.

---

This commit does not deal with the mess of a worktree that has
experienced this bug before. In particular, if `git-annex get` were
run in such a worktree, it would have stored the object files in the
linked worktree's git directory, rather than in the main git directory.
Such misplaced object files need to be dealt with; the plan is to make
git-annex fsck notice and fix them.

A worktree that has experienced this bug before will contain unpopulated
pointer files. Those may eventually get fixed up in regular usage of
git-annex, but git-annex fsck will also fix them up.

---

Finally, this has me pondering if all of git-annex's state files should
really be stored in one common place across all linked worktrees. Should
perhaps state files that are specific to the worktree be stored per-worktree?
That has not been the case when using git-annex on filesystems supporting
symlinks, but it *has* been the case on filesystems not supporting
symlinks. Perhaps this leads to some other buggy behavior in some cases.
Or perhaps to extra work being done.

For example, the keys database has an associated files table. Which depends
on the worktree. But reconcileStaged updates that table, so when git-annex
is used first in one worktree and then in another one, reconcileStaged will
update the table to reflect the current worktree. Which is extra work each
time a different worktree is used. But also, what if two git-annex
processes are running at the same time, in separate worktrees? Probably
this needs more thought and investigation.

So there is a risk that this commit exposes such buggy behavior in a
situation where it didn't happen before, due to the filesystem not
supporting symlinks. But, given how much this bug crippled using linked
worktrees in such a situation, I doubt that many people have been doing
that.
2025-07-14 13:20:39 -04:00

486 lines
17 KiB
Haskell

{- git-annex links to content
-
- On file systems that support them, symlinks are used.
-
- On other filesystems, git instead stores the symlink target in a regular
- file.
-
- Pointer files are used instead of symlinks for unlocked files.
-
- Copyright 2013-2022 Joey Hess <id@joeyh.name>
-
- Licensed under the GNU AGPL version 3 or higher.
-}
{-# LANGUAGE CPP, BangPatterns, OverloadedStrings #-}
module Annex.Link where
import Annex.Common
import qualified Annex
import qualified Annex.Queue
import qualified Git.Queue
import qualified Git.UpdateIndex
import qualified Git.Index
import qualified Git.LockFile
import qualified Git.Env
import qualified Git
import Logs.Restage
import Git.Types
import Git.FilePath
import Git.Config
import Annex.HashObject
import Annex.InodeSentinal
import Annex.PidLock
import Utility.FileMode
import Utility.InodeCache
import Utility.Tmp.Dir
import Utility.CopyFile
import qualified Database.Keys.Handle
import qualified Utility.RawFilePath as R
import qualified Utility.FileIO as F
import qualified Utility.OsString as OS
import qualified Data.ByteString as S
import qualified Data.ByteString.Char8 as S8
import qualified Data.ByteString.Lazy as L
#ifndef mingw32_HOST_OS
#if MIN_VERSION_unix(2,8,0)
#else
import System.PosixCompat.Files (isSymbolicLink)
#endif
#endif
type LinkTarget = S.ByteString
{- Checks if a file is a link to a key. -}
isAnnexLink :: OsPath -> Annex (Maybe Key)
isAnnexLink file = maybe Nothing parseLinkTargetOrPointer <$> getAnnexLinkTarget file
{- Gets the link target of a symlink.
-
- On a filesystem that does not support symlinks, fall back to getting the
- link target by looking inside the file.
-
- Returns Nothing if the file is not a symlink, or not a link to annex
- content.
-}
getAnnexLinkTarget :: OsPath -> Annex (Maybe LinkTarget)
getAnnexLinkTarget f = getAnnexLinkTarget' f
=<< (coreSymlinks <$> Annex.getGitConfig)
{- Pass False to force looking inside file, for when git checks out
- symlinks as plain files. -}
getAnnexLinkTarget' :: OsPath -> Bool -> Annex (Maybe LinkTarget)
getAnnexLinkTarget' file coresymlinks = if coresymlinks
then check probesymlink $
return Nothing
else check probesymlink $
check probefilecontent $
return Nothing
where
check getlinktarget fallback =
liftIO (catchMaybeIO getlinktarget) >>= \case
Just l
| isLinkToAnnex l -> return (Just l)
| otherwise -> return Nothing
Nothing -> fallback
probesymlink = R.readSymbolicLink (fromOsPath file)
probefilecontent = F.withFile file ReadMode $ \h -> do
s <- S.hGet h maxSymlinkSz
-- If we got the full amount, the file is too large
-- to be a symlink target.
return $ if S.length s == maxSymlinkSz
then mempty
else
-- If there are any NUL or newline
-- characters, or whitespace, we
-- certainly don't have a symlink to a
-- git-annex key.
if any (`S8.elem` s) ("\0\n\r \t" :: [Char])
then mempty
else s
makeAnnexLink :: LinkTarget -> OsPath -> Annex ()
makeAnnexLink = makeGitLink
{- Creates a link on disk.
-
- On a filesystem that does not support symlinks, writes the link target
- to a file. Note that git will only treat the file as a symlink if
- it's staged as such, so use addAnnexLink when adding a new file or
- modified link to git.
-}
makeGitLink :: LinkTarget -> OsPath -> Annex ()
makeGitLink linktarget file = ifM (coreSymlinks <$> Annex.getGitConfig)
( liftIO $ do
void $ tryIO $ removeFile file
R.createSymbolicLink linktarget (fromOsPath file)
, liftIO $ F.writeFile' file linktarget
)
{- Creates a link on disk, and additionally stages it in git. -}
addAnnexLink :: LinkTarget -> OsPath -> Annex ()
addAnnexLink linktarget file = do
makeAnnexLink linktarget file
stageSymlink file =<< hashSymlink linktarget
{- Injects a symlink target into git, returning its Sha. -}
hashSymlink :: LinkTarget -> Annex Sha
hashSymlink = go . fromOsPath . toInternalGitPath . toOsPath
where
go :: LinkTarget -> Annex Sha
go = hashBlob
{- Stages a symlink to an annexed object, using a Sha of its target. -}
stageSymlink :: OsPath -> Sha -> Annex ()
stageSymlink file sha =
Annex.Queue.addUpdateIndex =<<
inRepo (Git.UpdateIndex.stageSymlink file sha)
{- Injects a pointer file content into git, returning its Sha. -}
hashPointerFile :: Key -> Annex Sha
hashPointerFile key = hashBlob $ formatPointer key
{- Stages a pointer file, using a Sha of its content -}
stagePointerFile :: OsPath -> Maybe FileMode -> Sha -> Annex ()
stagePointerFile file mode sha =
Annex.Queue.addUpdateIndex =<<
inRepo (Git.UpdateIndex.stageFile sha treeitemtype file)
where
treeitemtype
| maybe False isExecutable mode = TreeExecutable
| otherwise = TreeFile
writePointerFile :: OsPath -> Key -> Maybe FileMode -> IO ()
writePointerFile file k mode = do
F.writeFile' file (formatPointer k)
maybe noop (R.setFileMode (fromOsPath file)) mode
newtype Restage = Restage Bool
{- Restage pointer file. This is used after updating a worktree file
- when content is added/removed, to prevent git status from showing
- it as modified.
-
- The InodeCache is for the worktree file. It is used to detect when
- the worktree file is changed by something else before git update-index
- gets to look at it.
-
- Asks git to refresh its index information for the file.
- That in turn runs the clean filter on the file; when the clean
- filter produces the same pointer that was in the index before, git
- realizes that the file has not actually been modified.
-
- Note that, if the pointer file is staged for deletion, or has different
- content than the current worktree content staged, this won't change
- that. So it's safe to call at any time and any situation.
-
- If the index is known to be locked (eg, git add has run git-annex),
- that would fail. Restage False will prevent the index being updated,
- and will store it in the restage log. Displays a message to help the
- user understand why the file will appear to be modified.
-
- This uses the git queue, so the update is not performed immediately,
- and this can be run multiple times cheaply. Using the git queue also
- prevents building up too large a number of updates when many files
- are being processed. It's also recorded in the restage log so that,
- if the process is interrupted before the git queue is fulushed, the
- restage will be taken care of later.
-}
restagePointerFile :: Restage -> OsPath -> InodeCache -> Annex ()
restagePointerFile (Restage False) f orig = do
flip writeRestageLog orig =<< inRepo (toTopFilePath f)
toplevelWarning True $ unableToRestage $ Just f
restagePointerFile (Restage True) f orig = do
flip writeRestageLog orig =<< inRepo (toTopFilePath f)
-- Avoid refreshing the index if run by the
-- smudge clean filter, because git uses that when
-- it's already refreshing the index, probably because
-- this very action is running. Running it again would likely
-- deadlock.
unlessM (Annex.getState Annex.insmudgecleanfilter) $
Annex.Queue.addFlushAction restagePointerFileRunner [f]
restagePointerFileRunner :: Git.Queue.FlushActionRunner Annex
restagePointerFileRunner =
Git.Queue.FlushActionRunner "restagePointerFiles" $ \r _fs ->
restagePointerFiles r
-- Restage all files in the restage log that have not been modified.
--
-- Other changes to the files may have been staged before this
-- gets a chance to run. To avoid a race with any staging of
-- changes, first lock the index file. Then run git update-index
-- on all still-unmodified files, using a copy of the index file,
-- to bypass the lock. Then replace the old index file with the new
-- updated index file.
restagePointerFiles :: Git.Repo -> Annex ()
restagePointerFiles r = unlessM (Annex.getState Annex.insmudgecleanfilter) $ do
-- Flush any queued changes to the keys database, so they
-- are visible to child processes.
-- The database is closed because that may improve behavior
-- when run in Windows's WSL1, which has issues with
-- multiple writers to SQL databases.
liftIO . Database.Keys.Handle.closeDbHandle
=<< Annex.getRead Annex.keysdbhandle
realindex <- liftIO $ Git.Index.currentIndexFile r
numsz@(numfiles, _) <- calcnumsz
let lock = Git.Index.indexFileLock realindex
lockindex = liftIO $ catchMaybeIO $ Git.LockFile.openLock' lock
unlockindex = liftIO . maybe noop Git.LockFile.closeLock
showwarning = warning $ unableToRestage Nothing
go Nothing = showwarning
go (Just _) = withtmpdir $ \tmpdir -> do
tsd <- getTSDelta
let tmpindex = tmpdir </> literalOsPath "index"
let replaceindex = liftIO $ moveFile tmpindex realindex
let updatetmpindex = do
r' <- liftIO $ Git.Env.addGitEnv r Git.Index.indexEnv
. fromOsPath
=<< Git.Index.indexEnvVal tmpindex
configfilterprocess numsz $
runupdateindex tsd r' replaceindex
return True
ok <- liftIO (createLinkOrCopy realindex tmpindex)
<&&> catchBoolIO updatetmpindex
unless ok showwarning
when (numfiles > 0) $
bracket lockindex unlockindex go
where
withtmpdir = withTmpDirIn
(Git.localGitDir r)
(literalOsPath "annexindex")
isunmodified tsd f orig =
genInodeCache f tsd >>= return . \case
Nothing -> False
Just new -> compareStrong orig new
{- Avoid git warning about CRLF munging -}
avoidcrlfwarning r' = r' { gitGlobalOpts = gitGlobalOpts r' ++
[ Param "-c"
, Param $ "core.safecrlf=" ++ boolConfig False
] }
runupdateindex tsd r' replaceindex =
runsGitAnnexChildProcessViaGit' (avoidcrlfwarning r') $ \r'' ->
Git.UpdateIndex.refreshIndex r'' $ \feeder -> do
let atend = do
-- wait for index write
liftIO $ feeder Nothing
replaceindex
streamRestageLog atend $ \topf ic -> do
let f = fromTopFilePath topf r''
liftIO $ whenM (isunmodified tsd f ic) $
feedupdateindex f feeder
{- update-index is documented as picky about "./file" and it
- fails on "../../repo/path/file" when cwd is not in the repo
- being acted on. Avoid these problems with an absolute path.
-}
feedupdateindex f feeder = do
absf <- absPath f
feeder (Just absf)
calcnumsz = calcRestageLog (0, 0) $ \(_f, ic) (numfiles, sizefiles) ->
(numfiles+1, sizefiles + inodeCacheFileSize ic)
{- filter.annex.process configured to use git-annex filter-process
- is sometimes faster and sometimes slower than using
- git-annex smudge. The latter is run once per file, while
- the former has the content of files piped to it.
-}
filterprocessfaster :: (Integer, FileSize) -> Bool
filterprocessfaster (numfiles, sizefiles) =
let estimate_enabled = sizefiles `div` 191739611
estimate_disabled = numfiles `div` 7
in estimate_enabled <= estimate_disabled
{- This disables filter.annex.process if it's set when it would
- probably not be faster to use it. Unfortunately, simply
- passing -c filter.annex.process= also prevents git from
- running the smudge filter, so .git/config has to be modified
- to disable it. The modification is reversed at the end. In
- case this process is terminated early, the next time this
- runs it will take care of reversing the modification.
-}
configfilterprocess numsz = bracket setup cleanup . const
where
setup
| filterprocessfaster numsz = return Nothing
| otherwise = fromRepo (Git.Config.getMaybe ck) >>= \case
Nothing -> return Nothing
Just v -> do
void $ inRepo (Git.Config.change ckd (fromConfigValue v))
void $ inRepo (Git.Config.unset ck)
return (Just v)
cleanup (Just v) = do
void $ inRepo $ Git.Config.change ck (fromConfigValue v)
void $ inRepo (Git.Config.unset ckd)
cleanup Nothing = fromRepo (Git.Config.getMaybe ckd) >>= \case
Nothing -> return ()
Just v -> do
whenM (isNothing <$> fromRepo (Git.Config.getMaybe ck)) $
void $ inRepo (Git.Config.change ck (fromConfigValue v))
void $ inRepo (Git.Config.unset ckd)
ck = ConfigKey "filter.annex.process"
ckd = ConfigKey "filter.annex.process-temp-disabled"
unableToRestage :: Maybe OsPath -> StringContainingQuotedPath
unableToRestage mf =
"git status will show " <> maybe "some files" QuotedPath mf
<> " to be modified, since content availability has changed"
<> " and git-annex was unable to update the index."
<> " This is only a cosmetic problem affecting git status; git add,"
<> " git commit, etc won't be affected."
<> " To fix the git status display, you can run:"
<> " git-annex restage"
{- Parses a symlink target or a pointer file to a Key.
-
- Makes sure that the pointer file is valid, including not being longer
- than the maximum allowed size of a valid pointer file, and that any
- subsequent lines after the first contain the validPointerLineTag.
- If a valid pointer file gets some other data appended to it, it should
- never be considered valid, unless that data happened to itself be a
- valid pointer file.
-}
parseLinkTargetOrPointer :: S.ByteString -> Maybe Key
parseLinkTargetOrPointer = either (const Nothing) id
. parseLinkTargetOrPointer'
data InvalidAppendedPointerFile = InvalidAppendedPointerFile
parseLinkTargetOrPointer' :: S.ByteString -> Either InvalidAppendedPointerFile (Maybe Key)
parseLinkTargetOrPointer' b =
let (firstline, rest) = S8.span (/= '\n') b
in case parsekey $ droptrailing '\r' firstline of
Just k
| S.length b > maxValidPointerSz -> Left InvalidAppendedPointerFile
| restvalid (dropleading '\n' rest) -> Right (Just k)
| otherwise -> Left InvalidAppendedPointerFile
Nothing -> Right Nothing
where
parsekey l
| isLinkToAnnex l = fileKey $ toOsPath $
snd $ S8.breakEnd pathsep l
| otherwise = Nothing
restvalid r
| S.null r = True
| otherwise =
let (l, r') = S8.span (/= '\n') r
in validPointerLineTag `S.isInfixOf` l
&& (not (S8.null r') && S8.head r' == '\n')
&& restvalid (S8.tail r')
dropleading c l
| S.null l = l
| S8.head l == c = S8.tail l
| otherwise = l
droptrailing c l
| S.null l = l
| S8.last l == c = S8.init l
| otherwise = l
pathsep '/' = True
#ifdef mingw32_HOST_OS
pathsep '\\' = True
#endif
pathsep _ = False
{- Avoid looking at more of the lazy ByteString than necessary since it
- could be reading from a large file that is not a pointer file. -}
parseLinkTargetOrPointerLazy :: L.ByteString -> Maybe Key
parseLinkTargetOrPointerLazy = either (const Nothing) id
. parseLinkTargetOrPointerLazy'
parseLinkTargetOrPointerLazy' :: L.ByteString -> Either InvalidAppendedPointerFile (Maybe Key)
parseLinkTargetOrPointerLazy' b =
let b' = L.take (fromIntegral maxPointerSz) b
in parseLinkTargetOrPointer' (L.toStrict b')
formatPointer :: Key -> S.ByteString
formatPointer k = fromOsPath prefix <> fromOsPath (keyFile k) <> nl
where
prefix = toInternalGitPath $
pathSeparator `OS.cons` objectDir standardGitLocationMaker
nl = S8.singleton '\n'
{- Maximum size of a file that could be a pointer to a key.
- Check to avoid buffering really big files in git into
- memory when reading files that may be pointers.
-
- 8192 bytes is plenty for a pointer to a key. This adds some additional
- padding to allow for pointer files that have lines of additional data
- after the key.
-
- One additional byte is used to detect when a valid pointer file
- got something else appended to it.
-}
maxPointerSz :: Int
maxPointerSz = maxValidPointerSz + 1
{- Maximum size of a valid pointer files is 32kb. -}
maxValidPointerSz :: Int
maxValidPointerSz = 32768
maxSymlinkSz :: Int
maxSymlinkSz = 8192
{- Checks if a worktree file is a pointer to a key.
-
- Unlocked files whose content is present are not detected by this.
-
- It's possible, though unlikely, that an annex symlink points to
- an object that looks like a pointer file. Or that a non-annex
- symlink does. Avoids a false positive in those cases.
- -}
isPointerFile :: OsPath -> IO (Maybe Key)
isPointerFile f = catchDefaultIO Nothing $
#if defined(mingw32_HOST_OS)
F.withFile f ReadMode readhandle
#else
#if MIN_VERSION_unix(2,8,0)
let open = do
fd <- openFd (fromOsPath f) ReadOnly
(defaultFileFlags { nofollow = True })
fdToHandle fd
in bracket open hClose readhandle
#else
ifM (isSymbolicLink <$> R.getSymbolicLinkStatus (fromOsPath f))
( return Nothing
, F.withFile f ReadMode readhandle
)
#endif
#endif
where
readhandle h = parseLinkTargetOrPointer <$> S.hGet h maxPointerSz
{- Checks a symlink target or pointer file first line to see if it
- appears to point to annexed content.
-
- We only look for paths inside the .git directory, and not at the .git
- directory itself, because GIT_DIR may cause a directory name other
- than .git to be used.
-}
isLinkToAnnex :: S.ByteString -> Bool
isLinkToAnnex s = p `OS.isInfixOf` s'
#ifdef mingw32_HOST_OS
-- '/' is used inside pointer files on Windows, not the native '\'
|| p' `OS.isInfixOf` s'
#endif
where
s' = toOsPath s
p = pathSeparator `OS.cons` objectDir standardGitLocationMaker
#ifdef mingw32_HOST_OS
p' = toInternalGitPath p
#endif
{- String that must appear on every line of a valid pointer file. -}
validPointerLineTag :: S.ByteString
validPointerLineTag = "/annex/"