The query args of POST requests become the request body. If there's a
redirect, the redirected url will therefore not contain the query
arguments. Use an explicit request body to make the redirection work.
Also add logging of extended error codes for this IO error, maybe we can
become more specific about which situations should trigger a journal
mode switch.
This is the first time the account url may update outside of
account setup.
Summary of redirection handling:
1. During account setup (wizard)
- status.php gets permanently redirected -> adjust url
- authed PROPFIND gets *any* redirection -> adjust url
2. During connectivity ping (ConnectionValidator)
- status.php gets permanently redirected -> adjust url (new!)
All other redirections should be followed transparently and
don't update the account url in the settings.
Merge csync_create and csync_init into the constructor and
replace csync_destroy with the destructor.
Also use a QByteArray for csync_s::root_perms and flatten
csync_rename_s as a rename sub-struct of csync_s since it
can now handle C++ types.
Just expose csync_file_stat_t since we don't need an abstraction layer
anymore. Also pass the nodes of both trees directly to the visitor
function.
Issue #1817
Also move csync_normalize_etag to common/utility since we
don't need the char* function anymore.
Remove the single space file_stat->remotePerm codepath since
this won't be used in csync anymore since
8de3bda0b1.
Issue #1817
This is the first commit trying to unify csync_file_stat_s,
csync_vio_file_stat_s and csync_tree_walk_file_s. Use QByteArray
and unique_ptr already since I'm not used to track memory allocations
and this will make the transition easier.
Issue #1817
Now that csync builds as C++, this will avoid having to implement
functionalities needed by csync mandatorily in csync itself.
This library is built as part of libocsync and symbols exported
through it.
This requires a relicense of Utility as LGPL. All classes moved into
this library from src/libsync will need to be relicensed as well.
This will allow us to unify data structures between csync and libsync.
Utility functions like csync_time and c_std are still compiled as C
since we won't need to be coupled with Qt in the short term.
By setting the icon in Desktop.ini of the root folder, this adds the icon
both when browsing the folder directly and to the sidebar shortcut.
To avoid overwriting any user setting that could exist in Desktop.ini,
only do this if the file doesn't exist. Editing .ini files on Windows
isn't trivial and isn't worth it given that this file won't exist most
of the time.
Fixes#2446
Allow upgrade path when the server removes support for oauth
Relates: https://github.com/owncloud/client/issues/5848#issuecomment-317353049
We also need to force the account to commit the config to the disk,
otherwise we may not register we are no longer using owncloud and we
risk sending the password as the token to the token refresh API call
Before commit d3b00532b1,
fetchFromKeychain was called everytime we detect that the creds are
invalid (in AccountState::slotInvalidCredentials)
But since that commit, AccountState was calling askFromUser directly,
breaking the refresh of the token.
So I made sure AccountState::slotInvalidCredentials still calls
refreshAccessToken.
Another change that was made was too be sure to clear the cookies
in HttpCredentials::invalidateToken even when we are only clearing the
access_token. That's because the session with a cookie may stay valid
longer than the access_token
We only want to know if they were touched within the last 15 seconds,
so change the data structure to use a QMultiMap, and sort them by
QElapsedTimer. This allows us to iterate over old entries ordered by
time and to stop once we find a recent entry.
This makes the look-up slower but in most cases the folder watcher
will report any change within milliseconds, and we start from the
most recent. What this really makes slower are actual user file
changes while a fast sync is underways which will need to iterate
over the whole map to find out the file isn't there.
This reduces the growth of the memory usage when downloading a large
amount of files.
We need Qt 5.9 for HTTP2 because, even if Qt 5.8 already has support
for it, there is some critical bug in the HTTP2 implementation which
make it unusable [ https://codereview.qt-project.org/186050 and
https://codereview.qt-project.org/186066 ]
When using HTTP2, we can use many more parallel network request, this
is especially good for small file handling
Lower the priority of the GET and PUT propagation jobs, so the quota
or selective sync ui PROPFIND will not be blocked by them
Since these errors are blacklisted, it can take up to 24h to retry items
that had a 507 error for a while. This way users can intervene and cause
an upload attempt immediately.
It now produces a summary error message indicating the problem.
Adjust blacklist database table to contain 'errorCategory'. This is
useful for two things:
- Reestablishing summary messages based on blacklisted errors. For
example if we don't retry a 507ed file, we still want to show the
message about space on the server
- Selectively wiping the blacklist: When we have ui for something like
"I deleted some files, please retry all files now!", we want to
delete all blacklist entries of a specific category only.
* A bunch of code was determining sync status by ad-hoc comparing some
progress info fields. It can now just check the status, making it
easier to comprehend.
* There's a clear indication for "a new sync is starting", which helps
wiping the issues tab at the right time.
For now we use them for:
* csync errors: This allows them to appear in the sync issues tab
* insufficient local disk space, as a summary of individual file errors
Insufficient remote space will use them too, as might other issues that
are bigger than a single sync item.
The QNAM may continue to outlive both.
Rename Credentials::getQNAM() to createQNAM() while we're at it - it's
used to make a new QNAM that will subsequently be owned by the Account
object.
See d01065b9a1 for rationale.
Relates to
d40c56eda5147cf798a6
* SocketAPI has COPL_LOCAL_LINK / EMAIL_LOCAL_LINK commands
* The nautilus and dolphing shell integrations show a submenu from which
one can share as well as access the private link.
* The SocketAPI provides a new GET_STRINGS command to access localized
strings.
* The private link can also be accessed from the user/group sharing
dialog.
* The numeric file id is extracted from the full id to create the
private link url.
Calling forgetSensitiveData() on account deletion leads to a timer for
clearQNAMCache() being queued. Then the Account object is deleted. The
Credentials object stays alive for now because it has a deleteLater
deleter.
If the timer calls into a slot on the Credentials object, the _account
pointer will be invalid at this time.
As a workaround, move the target slot to Account - that way it will not
be called as the account object is already destroyed.
However since Account and Credentials are mutually dependent, it would
be much preferable if their lifetimes were linked, avoiding this
category of bugs.
The current behavior was introduced in
d40c56eda5 and I currently don't
understand why - maybe there's another way of dealing with the problem
that existed then.
Before, blacklisted errors were set to FileIgnored status and hence
displayed as warnings. Now, they have their own BlacklistedError
category which allows them to appear as errors in the issues list and in
the shell integration icons.
When synchronizing a folder on a samba share, creating files that begin
with ._ is often forbidden. This prevented the client from creating
its ._sync_abcdef.db file.
Now, it'll check whether the preferred filename is creatable, and if
it isn't it'll use .sync_abcdef.db instead.
The disadvantage is that this alternative path won't be ignored by
older clients - that was the reason for the ._ prefix.
* For conflicts where mtime and size are identical:
a) If there's no remote checksum, skip (unchanged)
b) If there's a remote checksum that's a useful hash, create a
PropagateDownload job and compute the local hash. If the hashes
are identical, don't download the file and just update metadata.
* Avoid exposing the existence of checksumTypeId beyond the database
layer. This makes handling checksums easier in general because they
can usually be treated as a single blob.
This change was prompted by the difficulty of producing file_stat_t
entries uniformly from PROPFINDs and the database.
- Add category to the all messages (they did not have it was merged right after
the patch to add category everywhere, but this code did not have it.)
- Make sure there is no warnings in the normal flow. (The wizard does a request
without authentication to determine the auth type)
All our crypto code is handled by qt nodaways.
No need to carry this dependency.
Especially since it causes warnings on system where there are
twp openssl version installed:
/usr/bin/ld: warning: libcrypto.so.1.0.0, needed by /usr/lib/libQt5Network.so.5.9.0, may conflict with libcrypto.so.1.1
Issue #5783
When the directry that should be removed by selective sync contains changes,
we ignore the whole sub tree instead of only ignoreing new files.
We cannot ignore the whole directory, we need to ignore only the directory
that do not have files to remove
See owncloud/enterprise#1966
If the server and the client's database go out of sync, there could be
persistent 404 errors. This change ensures that the problem corrects
itself eventually by triggering a remote discovery of the file's
parent folders.
It does not address the root cause that might have lead to the
divergence.
These would otherwise be line-wrapped by clang-format,
and then consecutive reformattings remove the aligned
comment indentation
Example:
int a; // too long comment
->
int a; // too long
// comment
->
int a; // too long
// comment
When a new folder becomes selective-sync excluded, we already mark it
and all its parent folders with _invalid_ etags to force rediscovery.
That's not enough however. Later calls to csync_statedb_get_below_path
could still pull data about the excluded files into the remote tree.
That lead to incorrect behavior, such as uploads happening for folders
that had been explicitly excluded from sync.
To fix the problem, statedb_get_below_path is adjusted to not read the
data about excluded folders from the database.
Currently we can't wipe this data from the database outright because we
need it to determine whether the files in the excluded folder can be
wiped away or not.
See owncloud/enterprise#1965
Use qCInfo for anything that has general value for support and
development. Use qCWarning for any recoverable error and qCCritical
for anything that could result in data loss or would identify a serious
issue with the code.
Issue #5647
This gives more insight about the logs and allow setting fine-tuned
logging rules. The categories are set to only output Info by default
so this allows us to provide more concise logging while keeping the
ability to extract more information for a specific category when
developping or debugging customer issues.
Issue #5647
Add the log level and category name in the output. Only output the
thread ID and function name for qCDebug statements as they are not
necessary for general use and make the log harder to read.
Also make sure that the message pattern is set when NO_MSG_HANDLER is
used. Using an environment variable should have priority over it anyway.
When we first detect a 503 (probably from a PROPFIND) and enter the
ServiceUnavailable state, we new trigger a status.php query that will
switch the state to MaintenanceMode if necessary.
Before this patch, to deep folder would just be ignored, without any feedback.
This patch makes it so deep folder are properly shown as ignored in the UI.
Also increase the MAX_DEPTH
Issue: #1067
I'm confident this is unnecessary. The original bug in #3283 was
to call ignoreSslErrors() without an argument in the 'accept'
case, which meant ignoring *all* subsequent SSL errors.
With that fixed, explicitly aborting the reply and resetting QNAM
is not needed since not ignoring the error will lead to the SSL
handshake failing.
See also:
75b38d1a2f (workaround introduced)
89376e14d6 (real fix)
76ce5adbf0 (cherry-pick of workaround)
Newer servers will have the option of hiding version, versionstring, edition
and productname. They will always send the full information in the capabilities.
By default QNetworkReply::errorString() often produces messages like
"Error downloading <url> - server replied: <reason>"
but the "downloading" part invariably confuses people since the
error might very well have been produced by a PUT request.
This commit produces clearer error messages for HTTP errors.
Additionally:
* Remove some unnecessary null checks from slots connected to
network job signals and document that these signals never send
null replies.
* There was a bug where AbstractNetworkJob::_timedout wasn't
set when derived classes overrode slotTimeout. We now ensure
it's always set by disallowing overrides of slotTimeout.
Instead it now calls onTimedOut, which allows custom handling.
* Several subclasses declared errorString, isTimedOut. Move
these to AbstractNetworkJob.
* Unify handling of OC-ErrorString (via the new, general
Job::errorString)
* Add documentation in various places.
Backtrace looks like this:
Crash: EXCEPTION_ACCESS_VIOLATION_READ at 0x0
File "propagatedownload.cpp", line 234, in OCC::GETFileJob::slotReadyRead
File "moc_propagatedownl_CA5CFSHZDTX34X.cpp", line 86, in OCC::GETFileJob::qt_static_metacall
File "qobject.cpp", line 495, in QMetaCallEvent::placeMetaCall
File "qobject.cpp", line 1256, in QObject::event
File "qapplication.cpp", line 3804, in QApplicationPrivate::notify_helper
GETFileJob::slotReadyRead can be called with a QueuedConnection when the
bendwith manager is involved. In that case, if the QNAM was reset
in between, the reply might have been destroyed.
(This is only speculation based on the backtrace)
* make target duration a client option instead of a capability
* simplify algorithm for determining chunk size significantly
* preserve chunk size for the whole propagation, not just per upload
* move options to SyncOptions to avoid depending on ConfigFile
in the propagator
* move chunk-size adjustment to after a chunk finishes, not when
a new chunk starts
The destructor of the PropagateItemJob will access the propagator's
_activeJobList. So the _rootJob needs to be destroyed before it.
Order of destruction is the reverse of the order of the members in
the class. So put it at the end so it can be destroyed first.
(This made TestSyncEngine::testDirDownloadWithError crash sometimes
in the master branch)
I didn't want to touch this visible translated string in a
patch release so I just switched the data.
For Qt versions before 5, we continue to show the build-time version
of OpenSSL but it's clearly marked as such.
In 8ef11a38c9, we started blacklisting
SoftError for 0 seconds. But if the two sync happen with less than
1s interval, we would still prevent them to happen.
So make sure we expire if 0 seconds have expired
It is possible to create files with filenames that differ
only by case in NTFS, but most operations such as stat and
open only target one of these by default.
When that happens, we want to avoid uploading incorrect data
and give up on the file.
Typically this situation should never occurr during normal use
of Windows. It can happen, however, when a NTFS partition is
mounted in another OS.
The crash reporter shows many crashes in OwncloudPropagator::scheduleNextJob.
We don't really know what could be the cause, but it's probably because
the _activeJobList contains dangling pointer.
So this patch makes sure to remove all the jobs from this list as they get
destroyed.
* For requests:
- reuse the original QNetworkRequest, so headers and attributes
are the same as in the original request
- determine the original http method from the reply and the request
attributes
- keep the original request body around such that it can be sent
again in case the request is redirected
* Simplify the interface that is used for creating new requests in
AbstractNetworkJob.
This leads to crashes since we changed the connection to the parent
jobs not to be queued anymore.
We don't really need to bubble up the finished state through
parents in that case, and it would also mean that we'd recurse
all the way through leaves as we go up to each parent. So just call
abort directly on the OwncloudPropagator and make sure the abortion
call is posted to the event loop.
Avoid using connections to report up the job tree for signals
that we can directly communicate to the OwncloudPropagator.
This slightly reduces the memory usage and avoid passing those calls
through the whole parent chain.
In preparation for the PropagateDirectory refactoring, simplify things
by removing WaitForFinishedInParentDirectory, which is currently
implemented as a one-level check.
This value is important for directory items, but is however never
used since a directory CSYNC_INSTRUCTION_RENAME item will always be in
PropagateDirectory::_firstJob, which will have to pass through its own
PropagateDirectory job's parallelism() before reaching the parent's
_subJobs optimization.
Since PropagateDirectory::parallelism can only return WaitForFinished
or FullParallelism, that value is lost. So this commit doesn't
change the behavior for directories, and allow file renames to be
scheduled in parallel across directories (which isn't a problem).
It could be possible that _firstJob is marked as finished if
aborted before its parent PropagateDirectory was marked as finished,
allowing a posted scheduleNextJob call to schedule the child job
in-between.
- I checked every occurence of a '%2' and make correct use of the
QString::arg overload that takes several argument instead of chaining
them, because the first argument can contains a '%1'
- I tried to look for every label that they either use plain text or richtext
and escape the user provided strings in there.
By default, followRedirects is true for all requests, to transparently
handle redirections. In the wizard, we have special redirect-handling
code though and that was being skipped.
Setting the flag to false allows the wizard to be aware of redirects
and to handle them in the correct way. Tested with the server described
in
https://github.com/owncloud/administration/tree/master/redirectServer
There's a second bug here, where followRedirects always converts
redirected requests to the GET verb. That means redirected PROPFINDs
will never have worked. This change un-breaks them for the wizard only.
There should be no case that previously worked that stops working now.
The test sets OWNCLOUD_MAX_PARALLEL to 1 to disable parallelism.
But since the max amount of parallelism is twice as much, that does not
work.
So change the way we compute the hardMaximumActiveJob: Use the value of
OWNCLOUD_MAX_PARALLEL to maximize this amount and base the maximum amount
of transfer jobs on it instead of the other way.
A result of this change is that, in case of bandwidth limit, we keep the
default of 6 non-transfer jobs in parallel. I believe that's fine since
the short jobs do not really use bandwidth, so we can still keep the same
amount of small jobs.
This reverts commit e1f5a49c21.
Retrying uploads with insufficent storage errors frequently leads to
high server traffic. See #5537 for links and a sketch of a correct
solution.
If we call
setConfiguration(QNetworkConfiguration());
This sets an invalid configuration on the QNAM.
But later, when we really go online because interfaces are discovered,
QNetworkAccessManagerPrivate::_q_onlineStateChanged is called (with isOnline=true).
And this will set the state to disconnected because customNetworkConfiguration is
true, and the networkConfiguration state is disabled.
The workaround we to fix another bug on Windows in which the default network
configuration was not behaving properly.
The issue on linux is hard to reproduce and only happen in some condition,
but it was reproduced on smashbox when they run two owncloudcmd at the same time.
Issues: #4720 , #3600
We were removing the wholme journal db when the user wanted to keep all files,
But that would also remove the selective sync lists.
We should only remove the metadata table.
Issue #5484
- Put all tests in the bin directory so that DLLs can be loaded
- Add missing exports
- Skip tests that use code depending on zlib
- The "GMT" timezone is named differently, use the int constructor instead
5 tests are still failing, it's not really worth fixing at the moment
since no developper is currently using Windows as its main platform.
It could be possible that _firstJob is marked as finished if
aborted before its parent PropagateDirectory was marked as finished,
allowing a posted scheduleNextJob call to schedule the child job
in-between.
We can do that because the only changes that were in master but not in 2.3 were the
translations change and documentation change, and the support for the 'M' permission
which we want in 2.3.
The sync engine rely on the 'M' in premission to ask for confirmation
(As requested in issue #5340)
But we only want to ask the premission for the 'root' of the mounting point and not
for every subfolders within it.
So we change the discovery phase in a way that it does not keep the 'M' for
children within the external storage.
Added two checkboxes in the Account Wizard in the advanced page to change the first options.
Also added a checkbox in the general settings to ask for confirmation for external storages.
Theme options allow to hide the checkboxes in the wizard.
As described in issue #5340
