Mercurial > hg
view tests/test-basic.t @ 46607:e9901d01d135
revlog: add a mechanism to verify expected file position before appending
If someone uses `hg debuglocks`, or some non-hg process writes to the .hg
directory without respecting the locks, or if the repo's on a networked
filesystem, it's possible for the revlog code to write out corrupted data.
The form of this corruption can vary depending on what data was written and how
that happened. We are in the "networked filesystem" case (though I've had users
also do this to themselves with the "`hg debuglocks`" scenario), and most often
see this with the changelog. What ends up happening is we produce two items
(let's call them rev1 and rev2) in the .i file that have the same linkrev,
baserev, and offset into the .d file, while the data in the .d file is appended
properly. rev2's compressed_size is accurate for rev2, but when we go to
decompress the data in the .d file, we use the offset that's recorded in the
index file, which is the same as rev1, and attempt to decompress
rev2.compressed_size bytes of rev1's data. This usually does not succeed. :)
When using inline data, this also fails, though I haven't investigated why too
closely. This shows up as a "patch decode" error. I believe what's happening
there is that we're basically ignoring the offset field, getting the data
properly, but since baserev != rev, it thinks this is a delta based on rev
(instead of a full text) and can't actually apply it as such.
For now, I'm going to make this an optional component and default it to entirely
off. I may increase the default severity of this in the future, once I've
enabled it for my users and we gain more experience with it. Luckily, most of my
users have a versioned filesystem and can roll back to before the corruption has
been written, it's just a hassle to do so and not everyone knows how (so it's a
support burden). Users on other filesystems will not have that luxury, and this
can cause them to have a corrupted repository that they are unlikely to know how
to resolve, and they'll see this as a data-loss event. Refusing to create the
corruption is a much better user experience.
This mechanism is not perfect. There may be false-negatives (racy writes that
are not detected). There should not be any false-positives (non-racy writes that
are detected as such). This is not a mechanism that makes putting a repo on a
networked filesystem "safe" or "supported", just *less* likely to cause
corruption.
Differential Revision: https://phab.mercurial-scm.org/D9952
author | Kyle Lippincott <spectral@google.com> |
---|---|
date | Wed, 03 Feb 2021 16:33:10 -0800 |
parents | 9c9e0b4b2ca7 |
children | 40b51c28b242 |
line wrap: on
line source
Create a repository: #if no-extraextensions $ hg config devel.all-warnings=true devel.default-date=0 0 extensions.fsmonitor= (fsmonitor !) largefiles.usercache=$TESTTMP/.cache/largefiles lfs.usercache=$TESTTMP/.cache/lfs ui.slash=True ui.interactive=False ui.detailed-exit-code=True ui.merge=internal:merge ui.mergemarkers=detailed ui.promptecho=True ui.timeout.warn=15 web.address=localhost web\.ipv6=(?:True|False) (re) web.server-header=testing stub value #endif $ hg init t $ cd t Prepare a changeset: $ echo a > a $ hg add a $ hg status A a Writes to stdio succeed and fail appropriately #if devfull $ hg status 2>/dev/full A a $ hg status >/dev/full abort: No space left on device [255] #endif #if devfull $ hg status >/dev/full 2>&1 [255] $ hg status ENOENT 2>/dev/full [255] #endif On Python 3, stdio may be None: $ hg debuguiprompt --config ui.interactive=true 0<&- abort: Bad file descriptor [255] $ hg version -q 0<&- Mercurial Distributed SCM * (glob) #if py3 $ hg version -q 1>&- abort: Bad file descriptor [255] #else $ hg version -q 1>&- #endif $ hg unknown -q 1>&- hg: unknown command 'unknown' (did you mean debugknown?) [10] $ hg version -q 2>&- Mercurial Distributed SCM * (glob) $ hg unknown -q 2>&- [10] $ hg commit -m test This command is ancient: $ hg history changeset: 0:acb14030fe0a tag: tip user: test date: Thu Jan 01 00:00:00 1970 +0000 summary: test Verify that updating to revision 0 via commands.update() works properly $ cat <<EOF > update_to_rev0.py > from mercurial import commands, hg, ui as uimod > myui = uimod.ui.load() > repo = hg.repository(myui, path=b'.') > commands.update(myui, repo, rev=b"0") > EOF $ hg up null 0 files updated, 0 files merged, 1 files removed, 0 files unresolved $ "$PYTHON" ./update_to_rev0.py 1 files updated, 0 files merged, 0 files removed, 0 files unresolved $ hg identify -n 0 Poke around at hashes: $ hg manifest --debug b789fdd96dc2f3bd229c1dd8eedf0fc60e2b68e3 644 a $ hg cat a a Verify should succeed: $ hg verify checking changesets checking manifests crosschecking files in changesets and manifests checking files checked 1 changesets with 1 changes to 1 files Repository root: $ hg root $TESTTMP/t $ hg log -l1 -T '{reporoot}\n' $TESTTMP/t $ hg root -Tjson | sed 's|\\\\|\\|g' [ { "hgpath": "$TESTTMP/t/.hg", "reporoot": "$TESTTMP/t", "storepath": "$TESTTMP/t/.hg/store" } ] At the end... $ cd .. Status message redirection: $ hg init empty status messages are sent to stdout by default: $ hg outgoing -R t empty -Tjson 2>/dev/null comparing with empty searching for changes [ { "bookmarks": [], "branch": "default", "date": [0, 0], "desc": "test", "node": "acb14030fe0a21b60322c440ad2d20cf7685a376", "parents": ["0000000000000000000000000000000000000000"], "phase": "draft", "rev": 0, "tags": ["tip"], "user": "test" } ] which can be configured to send to stderr, so the output wouldn't be interleaved: $ cat <<'EOF' >> "$HGRCPATH" > [ui] > message-output = stderr > EOF $ hg outgoing -R t empty -Tjson 2>/dev/null [ { "bookmarks": [], "branch": "default", "date": [0, 0], "desc": "test", "node": "acb14030fe0a21b60322c440ad2d20cf7685a376", "parents": ["0000000000000000000000000000000000000000"], "phase": "draft", "rev": 0, "tags": ["tip"], "user": "test" } ] $ hg outgoing -R t empty -Tjson >/dev/null comparing with empty searching for changes this option should be turned off by HGPLAIN= since it may break scripting use: $ HGPLAIN= hg outgoing -R t empty -Tjson 2>/dev/null comparing with empty searching for changes [ { "bookmarks": [], "branch": "default", "date": [0, 0], "desc": "test", "node": "acb14030fe0a21b60322c440ad2d20cf7685a376", "parents": ["0000000000000000000000000000000000000000"], "phase": "draft", "rev": 0, "tags": ["tip"], "user": "test" } ] but still overridden by --config: $ HGPLAIN= hg outgoing -R t empty -Tjson --config ui.message-output=stderr \ > 2>/dev/null [ { "bookmarks": [], "branch": "default", "date": [0, 0], "desc": "test", "node": "acb14030fe0a21b60322c440ad2d20cf7685a376", "parents": ["0000000000000000000000000000000000000000"], "phase": "draft", "rev": 0, "tags": ["tip"], "user": "test" } ] Invalid ui.message-output option: $ hg log -R t --config ui.message-output=bad abort: invalid ui.message-output destination: bad [255] Underlying message streams should be updated when ui.fout/ferr are set: $ cat <<'EOF' > capui.py > from mercurial import pycompat, registrar > cmdtable = {} > command = registrar.command(cmdtable) > @command(b'capui', norepo=True) > def capui(ui): > out = ui.fout > ui.fout = pycompat.bytesio() > ui.status(b'status\n') > ui.ferr = pycompat.bytesio() > ui.warn(b'warn\n') > out.write(b'stdout: %s' % ui.fout.getvalue()) > out.write(b'stderr: %s' % ui.ferr.getvalue()) > EOF $ hg --config extensions.capui=capui.py --config ui.message-output=stdio capui stdout: status stderr: warn