Replace a QVector for the wordlist with a QSet. This removes all duplicate entries in a given wordlist.
Thus, it hinders a malicious wordlist that has the proper length (>4000 entries) but with repetitions (effectively << 4000 entries) to be used and potentially create weaker passphrases than estimated.
Example:
List with 4000 items but only 64 unique words would lead to only 48 bit of Entropy instead of ~95 bit!
It turns out that the previous implementation, based on installing an event filter in every QAction instance, does not work on macOS, likely due to a Qt bug.
Attempt to work around this by using a different implementation of the same idea, by reacting to ShortcutOverride events in the MainWindow object.
Fixes#10929.
Fix a bug in the database key settings dialog, where it was previously
always incorrectly applying an empty password if the password was not
changed but some other change was made (e.g. adding or removing a key
file).
Previously, in a pattern like "{TIME:yy} {TIME}",
substituteBackupFilePath() would greedily use the entire string
"yy} {TIME" as the format specifier for the first TIME template, instead
of just "yy". Fix this, by adjusting the regular expression.
This ends up changing the behaviour of a weird corner case that is
covered in the tests, so change the test. I don't think anyone cares
about that case, and I think the current behaviour is better there.
Fixes#10505 (proved by adding a test case very similar to what was
reported there).
When selecting "Database → Import Passkey", we show a file picker.
Previously, we did not specify a parent widget for it. This could have
undesirable effects on its presentation. (For example, with the Sway
tiling Wayland compositor, it would show the file picker as a tiled
window rather than a floating one.)
Fix the issue by passing in the parent widget. This is also in line with
all other usages of FileDialog::getOpenFileName() in this project.
When the keepassxc window is shown something generetes a hide event, and it is hidden again immediately.
The 50ms interval for avoiding hiding the window when shown is not enough, even on modern systems.
Make the interval longer.
Change type of Handle on FreeBSD. On FreeBSD the libusb_hotplug_register_callback() function uses a pointer to a struct as a handle.
---------
Co-authored-by: Janek Bevendorff <janek@keepassxc.org>
* Snap: Improve Web-browser Native Messaging host functionality
This commit allows for the snap distribution of KeepassXC to self-manage native messaging manifests
This is done by making the binary aware of the snapd environment changes that currently prevent this.
Furthermore, the snap sandbox is expanded to the bare minimum needed to access these privileged files.
Please note if running a self-compiled / untrusted KeepassXC snap build (I.E, installed with --dangerous)
that you must manually run `sudo snap connect keepassxc:browser-native-messaging` to grant permissions.
This will work on all distributions that expose `/snap/bin/` - such as Ubuntu, Debian, etc.
For systems which don't provide `/snap/`, such as Fedora, follow instructions for enabling "Classic" snaps.
e.g., `sudo ln -s /var/lib/snapd/snap /snap`
---------
Co-authored-by: Jonathan White <support@dmapps.us>
When the user chooses to copy the password for an entry to the clipboard, previously there was logic to check if text was selected, and if so, that text was instead copied to the clipboard. That made sense if
(a) the user invoked the Copy Password action via its keyboard shortcut, and (b) that keyboard shortcut was configured (as per default) to be Ctrl-C, i.e. the same as the system action for copy-to-clipboard.
However, it made no sense if the user invoked that action in some other way, for example by clicking the corresponding toolbar button.
It also made no sense in the case that the Copy Password action had some other keyboard shortcut assigned. Also, if some other action had Ctrl-C assigned, the logic would not kick in then.
Fix all of the above by modifying the keyboard shortcut logic to intervene precisely in the case where a shortcut is pressed that matches the system copy-to-clipboard shortcut; only in that case do we now check if text is selected and if so copy that to the clipboard instead of the action we would otherwise take.
Fixes#10734.
If the system Copy key sequence (i.e. Ctrl+C or Cmd+C) is pressed while
inside the search entry without any text being selected, previously we
would copy the currently selected entry's password. This made sense when
keyboard shortcuts were fixed. Now that they are configurable, change it
to re-route the event to the main window, which can then take the
appropriate action (i.e. Ctrl+C might be bound to some other action).
* Entry placeholder resolution: don't overdo it
After resolving placeholders, previously the code would do it all over again if anything had changed, multiple times up to the recursion limit. This would have the effect of applying a much greater recursion limit, which is confusing and unnecessary, and probably undesired.
* Entry tweaks and minor refactoring
- Entry::size(): when computing tag size, use same delimiter set as in other places in the code
- Factor tag delimiter set regex out into global constant
- Placeholder resolution: remove unnecessary special casing for self-referential placeholders (these are taken care of by existing recursion depth limit)
- Placeholder resolution: less wasteful string building loop
- Move some constants from being public static data members of Entry to being local to Entry.cpp (in anonymous namespace)
- Migrate some QRegEx instances to QRegularExpression, the modern alternative
- Miscellanous minor code cleanups
* Entry: fix hitting recursion limit with {braces}
When encountering a {brace-enclosed} substring, the placeholder resolution logic would previously keep recursing until it hit the recursion depth limit (currently 10). This would lead to "Maximum depth of replacement has been reached" messages, and was also wasting CPU cycles.
Fixes#1741
---------
Co-authored-by: Jonathan White <support@dmapps.us>
* Fixes#10723 - only display password strength warning when actively editing the password
* Also improve behavior of minimum quality warning
* Improve behavior and handling of password changes with the database settings dialog
* Prevents loss of newly entered password when toggling between elements in the settings page
* On error, switch to tab that prevents saving database settings for easier correction
A warning is issued from Qt when the path is empty. This happens most often during test runs, but can also occur when closing a database before everything gets cleaned up.
* Fix database view splitters resize behaviour
* Set default ratio sizes for first-run based on the size of the database widget itself
* Fix setting splitter sizes before database widget has had a chance to render for the first time
* Disallow collapsing the entry view (source of several bug reports)
Fixes: #10613
---------
Co-authored-by: Jonathan White <support@dmapps.us>
* Fix passphrase generator test
Previously, the test case was assuming the wrong regex. In particular, the default word list (eff_large.wordlist) contains several words that contain dashes. Adjust the regex used in the test to reflect this. This should fix rare test failures
