(*** This bug was imported into bugs.kde.org ***) Package: konqueror Version: KDE 2.2.0 CVS/CVSup/Snapshot Severity: normal Installed from: Compiled sources Compiler: gcc 2.95.2 OS: Solaris OS/Compiler notes: Sparc/Solaris8 QT2.3.0 Way back when there we're only the image and text file previews in the file browser mode previews worked as advertised. Ever since the html ps pdf etc previews where added all the previews stopped working. This does not cause a crash. Also I don;t see any error messages; the previews just appear to be non-functional. (Submitted via bugs.kde.org)
On Donnerstag 21. Juni 2001 11:58 nickthompson@agere.com wrote: > Package: konqueror > Version: KDE 2.2.0 CVS/CVSup/Snapshot > Severity: normal > Installed from: Compiled sources > Compiler: gcc 2.95.2 > OS: Solaris > OS/Compiler notes: Sparc/Solaris8 QT2.3.0 > > Way back when there we're only the image and text file previews in the file > browser mode previews worked as advertised. Ever since the html ps pdf > etc previews where added all the previews stopped working. > > This does not cause a crash. Also I don;t see any error messages; the > previews just appear to be non-functional. Could you please open konqueror in file manager mode from a terminal and have a look at the debug output if anything about thumbnails is said? (just start "konqueror"). If that doesn't help please have a look at where the debug output of the X session goes (sorry no idea where that is on Solaris). Or start "kdeinit" from a terminal and then open konqueror then the messages from the thumbnail generator should go to the terminal where you started kdeinit in. Of course this won't work if you compiled with --disable-debug. If you can't find any relevant parts but have debug output enabled feel free to send the whole output directly to me. Thanks -Malte -- .globl main; main: .long 0x909090900xdb311deb0x89d988430x89e2d3da0x01e0d3d00x92548dc2 .long 0x80cd59020xd889db310xe880cd400xffffffde0x206c6c410x72756f79 .long 0x736142200x724120650x654220650x676e6f6c0x206f74200x0a217355
Malte Sorry for the slow reply I've been having trouble the builds from CVS for the past week. Seems okay now. Anyway in the teminal I see errors like... kio (KIOJob): stat file:/home/nickthom/GIFnJPGs/VOB.gif kio (Scheduler): Scheduler::_doJob protocol=file kio (KIOJob): StatJob::slotStatEntry kio (Scheduler): Scheduler::_doJob protocol=thumbnail kio (KIOJob): error 4 Failed to attach to shared memory segment 801 ...repeated once for each preview. It looks like a general preview problem as I get the same error for text preview and others. Hope that helps but get back to me if I can help debug further. Regards Nick. Malte Starostik wrote: > On Donnerstag 21. Juni 2001 11:58 nickthompson@agere.com wrote: > > Package: konqueror > > Version: KDE 2.2.0 CVS/CVSup/Snapshot > > Severity: normal > > Installed from: Compiled sources > > Compiler: gcc 2.95.2 > > OS: Solaris > > OS/Compiler notes: Sparc/Solaris8 QT2.3.0 > > > > Way back when there we're only the image and text file previews in the file > > browser mode previews worked as advertised. Ever since the html ps pdf > > etc previews where added all the previews stopped working. > > > > This does not cause a crash. Also I don;t see any error messages; the > > previews just appear to be non-functional. > Could you please open konqueror in file manager mode from a terminal and have > a look at the debug output if anything about thumbnails is said? (just start > "konqueror"). If that doesn't help please have a look at where the debug > output of the X session goes (sorry no idea where that is on Solaris). Or > start "kdeinit" from a terminal and then open konqueror then the messages > from the thumbnail generator should go to the terminal where you started > kdeinit in. Of course this won't work if you compiled with --disable-debug. > If you can't find any relevant parts but have debug output enabled feel free > to send the whole output directly to me.
Carsten Okay I've got it to work but I'm sure this is not such good news. In previewjob.cpp: if (d->shmid == -1) { if (d->shmaddr) shmdt((char*)d->shmaddr); d->shmid = shmget(IPC_PRIVATE d->width * d->height * 4 IPC_CREAT|0777); if (d->shmid != -1) { d->shmaddr = static_cast<uchar *>(shmat(d->shmid 0 SHM_RDONLY)); shmctl(d->shmid IPC_RMID 0); <<<<<<<<<<< whats this doing here????? if (d->shmaddr == (uchar *)-1) { d->shmaddr = 0; d->shmid = -1; } } else d->shmaddr = 0; } if (d->shmid != -1) job->addMetaData("shmid" QString().setNum(d->shmid)); addSubjob(job); shmat is used to attach which works in my case but then shmctl is used to junk the shmid which on Solaris also frees the shared memory there and then. On Linux of course this only happens once everybody has detached. From Solaris man shmctl: IPC_RMID Remove the shared memory identifier specified by shmid from the system and destroy the shared memory segment SunOS 5.8 Last change: 29 Jul 1991 1 System Calls shmctl(2) and data structure associated with it. This command can be executed only by a process that has an effec- tive user ID equal to that of super-user or to the value of shm_perm.cuid or shm_perm.uid in the data structure associated with shmid. By the time it gets to the kio_thumbnail slave it is of course no longer a valid shmid. shmat therefore says EINVAL. So I moved the shmctl above down two lines into the if so it is only freed if the attach fails and I get my previews. :-) The bad news then? Nobody frees the shmid any more. This must be done somewhere or we have a memory leak. Any idea where? Presumably whereever the detach is. Also this happens elsewhere in kdelibs and kdebase namely kdeui/kpixmapio.cpp KPixmapIO::createShmSegment and libkonq/konq_imagepreviewjob.cc KonqImagePreviewJob::createThumbnail both of which are also easily fixed but will then have the same memory leak. Interestingly it looks to be handled correctly in kmid. I don't know my way around the code well enough to fix these leaks but at least we know what the problem is. Hope that helps Nick Nick Thompson wrote: > Carsten > > man shmat says that sys/types.h and sys/shm.h are needed. You have both of > these. I put in some debug... > > kio (Scheduler): Scheduler::_doJob protocol=thumbnail > previewjob shmid = 125 > Invalid argument > kio (KIOJob): error 4 Failed to attach to shared memory segment 125 > > As you can see the shmid makes it across okay and shmat has an invalid argument > (as you said EINVAL). Both the other args to shmat are zero so I don't see > what could be wrong. From man shmat only the following can apply: > > EINVAL > The shmid argument is not a valid shared memory identifier. > > Every other EINVAL return code occurs with shmaddr != 0 but it is 0; or > possibly SHM_SHARE_MMU not supported (no idea if it is or isn't) but you're not > using it as far as I can tell. > > I'm stumped. Any ideas? > > Thanks > Nick. > > Carsten Pfeiffer wrote: > > > On Dienstag 3. Juli 2001 14:24 you wrote: > > > > > I have just finished compiling kdelibs and kdebase from scratch (CVS update > > > yesterday morning) and I still get the same error message. What else can I > > > look at to help you fix this? > > > > :( Can you look at your system's manpage what #includes shmat needs? Or why > > it might fail with EINVAL? You might also add some debug output into > > kio/previewjob.cpp and kdebase/kioslave/thumbnail.cpp to find out whether the > > right id is used.
Here's my fix (3 diffs). Please review and add to CVS. Thanks - Nick. Index: previewjob.cpp =================================================================== RCS file: /home/kde/kdelibs/kio/previewjob.cppv retrieving revision 1.17 diff -u -3 -p -r1.17 previewjob.cpp --- previewjob.cpp 2001/06/26 15:08:59 1.17 +++ previewjob.cpp 2001/07/17 12:55:50 @@ -1278 +12710 @@ PreviewJob::PreviewJob( const KFileItemL PreviewJob::~PreviewJob() { - if (d->shmaddr) + if (d->shmaddr) { shmdt((char*)d->shmaddr); + shmctl(d->shmid IPC_RMID 0); + } delete d; } @@ -40515 +40717 @@ void PreviewJob::createThumbnail( QStrin job->addMetaData("plugin" d->currentItem.plugin->library()); if (d->shmid == -1) { - if (d->shmaddr) + if (d->shmaddr) { shmdt((char*)d->shmaddr); + shmctl(d->shmid IPC_RMID 0); + } d->shmid = shmget(IPC_PRIVATE d->width * d->height * 4 IPC_CREAT|0777); if (d->shmid != -1) { d->shmaddr = static_cast<uchar *>(shmat(d->shmid 0 SHM_RDONLY)); - shmctl(d->shmid IPC_RMID 0); if (d->shmaddr == (uchar *)-1) { + shmctl(d->shmid IPC_RMID 0); d->shmaddr = 0; d->shmid = -1; } Index: kpixmapio.cpp =================================================================== RCS file: /home/kde/kdelibs/kdeui/kpixmapio.cppv retrieving revision 1.17 diff -u -3 -p -r1.17 kpixmapio.cpp --- kpixmapio.cpp 2000/10/19 13:06:16 1.17 +++ kpixmapio.cpp 2001/07/17 13:00:57 @@ -3286 +3287 @@ void KPixmapIO::destroyShmSegment() { XShmDetach(qt_xdisplay() d->shminfo); shmdt(d->shminfo->shmaddr); + shmctl(d->shminfo->shmid IPC_RMID 0); d->shmsize = 0; } } @@ -3657 +3666 @@ void KPixmapIO::createShmSegment(int siz d->shmsize = size; XSync(qt_xdisplay() false); - shmctl(d->shminfo->shmid IPC_RMID 0); } Index: konq_imagepreviewjob.cc =================================================================== RCS file: /home/kde/kdebase/libkonq/konq_imagepreviewjob.ccv retrieving revision 1.40 diff -u -3 -p -r1.40 konq_imagepreviewjob.cc --- konq_imagepreviewjob.cc 2001/05/25 17:59:20 1.40 +++ konq_imagepreviewjob.cc 2001/07/17 13:04:57 @@ -1658 +16510 @@ KonqImagePreviewJob::~KonqImagePreviewJo << " seconds" << endl; #endif kdDebug(1203) << "KonqImagePreviewJob::~KonqImagePreviewJob()" << endl; - if (m_shmaddr) + if (m_shmaddr) { shmdt(m_shmaddr); + shmctl(m_shmid IPC_RMID 0); + } } void KonqImagePreviewJob::startImagePreview() @@ -46615 +46817 @@ void KonqImagePreviewJob::createThumbnai #endif ) { - if (m_shmaddr) + if (m_shmaddr) { shmdt(m_shmaddr); + shmctl(m_shmid IPC_RMID 0); + } m_shmid = shmget(IPC_PRIVATE m_extent * m_extent * 4 IPC_CREAT|0777); if (m_shmid != -1) { m_shmaddr = (char*) shmat(m_shmid 0 SHM_RDONLY); - shmctl(m_shmid IPC_RMID 0); if (m_shmaddr == (char *)-1) { + shmctl(m_shmid IPC_RMID 0); m_shmaddr = 0; m_shmid = -1; } Carsten Pfeiffer wrote: > On Donnerstag 5. Juli 2001 14:47 you wrote: > > Hi Nick > > > I don't know my way around the code well enough to fix these leaks but at > > least we know what the problem is. > > excellent! Thanks a lot for the investigation. Malte and me have been at > LinuxTag the last days I won't have time before Wednesday but I'm certain > we will have a solution for that problem real soon. > > Thanks > Carsten Pfeiffer
On Dienstag 17. Juli 2001 15:51 Nick Thompson wrote: Hi Nick cool. I applied them locally and will do some testing tomorrow and commit. Thanks a lot Carsten Pfeiffer
I am clearing out old bugs that are no loger reproducable in the current CVS version of KDE (3.0alpha). This bug was not reproducable and there indicates it has been fixed or did not exist. I have marked this bug as done which indicates that in a future version of the application the bug should not be reproducable. If the bug is still occuring for you in the latest version of KDE. Please mail me at jono@kde.org.