Rewriting pup_event in C

What features/apps/bugfixes needed in a future Puppy
Message
Author
User avatar
jemimah
Posts: 4307
Joined: Wed 26 Aug 2009, 19:56
Location: Tampa, FL
Contact:

#41 Post by jemimah »

No luck. I'm still getting the same error. :(

akash_rawal
Posts: 229
Joined: Wed 25 Aug 2010, 15:38
Location: ISM Dhanbad, Jharkhand, India

#42 Post by akash_rawal »

Well, I had guessed that on reading the corresponding source code of thunar on git.

I downloaded the latest thunar, compiled and ran it myself, and found that it was destroying the mount object after every access. So I modified my module to duplicate the object before starting unmount operation ... and this works!

See first post.

User avatar
jemimah
Posts: 4307
Joined: Wed 26 Aug 2009, 19:56
Location: Tampa, FL
Contact:

#43 Post by jemimah »

Perfect! It works with xfdesktop too.

Saluki will have this in the next release and I'll be able to simplify pup_event_frontend_d to only deal with saving and such.

Fantastic!

I am getting error code 4 sometimes when mounting. Can you add a sleep or something somewhere to work around it?

User avatar
jemimah
Posts: 4307
Joined: Wed 26 Aug 2009, 19:56
Location: Tampa, FL
Contact:

#44 Post by jemimah »

Something's still a little screwy with unmounting.

If you compile thunar with notification support, you get a notification that the unmount is flushing to disk. But the notification never goes away when the unmount completes.

I can definitely work around this but I think the function call is not returning like it is supposed to.

akash_rawal
Posts: 229
Joined: Wed 25 Aug 2010, 15:38
Location: ISM Dhanbad, Jharkhand, India

#45 Post by akash_rawal »

jemimah wrote: I am getting error code 4 sometimes when mounting. Can you add a sleep or something somewhere to work around it?
I found a workaround.

Although /proc/mounts isn't updated immediately, the mounted drive is accessible as soon as mount exits.

See first post for update. The error code 4 no longer appears for successful mount. (I am relying on mount to exit with non-zero exit code on failure.)
jemimah wrote: If you compile thunar with notification support, you get a notification that the unmount is flushing to disk. But the notification never goes away when the unmount completes.
But I don't get a notification at all.
On unmounting I get this output:

Code: Select all

libnotify-Message: GetServerInformation call failed: The name org.freedesktop.Notifications was not provided by any .service files
libnotify-Message: Error getting spec version

(thunar:14747): libnotify-CRITICAL **: notify_get_server_info: assertion `proxy != NULL' failed

User avatar
jemimah
Posts: 4307
Joined: Wed 26 Aug 2009, 19:56
Location: Tampa, FL
Contact:

#46 Post by jemimah »

If you want to see the notifications you also need to install xfce-notifyd.

I ended up compiling thunar without notification support and disabling thunar's error checking. I have the mount script doing all the error checking and user interaction.

It seems to work ok.

I hacked thunar and xfdesktop to set the environment variable so the gio module is only used for those two programs. I got some hangs with other parts of xfce when I set it globally.

Saluki-022 should be out soon and it will have this stuff to play with.

akash_rawal
Posts: 229
Joined: Wed 25 Aug 2010, 15:38
Location: ISM Dhanbad, Jharkhand, India

#47 Post by akash_rawal »

jemimah wrote: If you want to see the notifications you also need to install xfce-notifyd.
Thanks, now I can reproduce your problem.

I recalled that old bug:
jemimah wrote: But unmounting is segfaulting.

Code: Select all

Program received signal SIGSEGV, Segmentation fault. 
0x0809e8f3 in ?? () 
(gdb) bt 
#0  0x0809e8f3 in ?? () 
#1  0xb78ecac5 in g_simple_async_result_complete () from /usr/lib/libgio-2.0.so.0 
#2  0xb7144524 in pup_volume_monitor_generic_cb (conv=0x8220b80, rcvd_data=0x8183f10, is_new=0, user_data=0x0, conv_user_data=0xb5e07598) 
    at volume_monitor.c:324 
#3  0xb713e3f9 in ps_conv_mgr_sorter_cb (sock=0x812f120, data=0xbffff514, user_data=0x81364c0) at conv.c:156 
#4  0xb713be02 in pup_sock_common_marshaller (hook=0x0, marshal_data=0xbffff458) at core.c:250 
#5  0xb774d91e in g_hook_list_marshal_check () from /usr/lib/libglib-2.0.so.0 
#6  0xb713c22b in pup_sock_raise (sock=0x812f120, event=1, has_data=1, cb_data=0xbffff514) at core.c:240 
#7  0xb713cff9 in pup_sock_try_receive_block (sock=0x812f120, timer=0x8183f90, data_read=0xbffff558, error=0xbffff5a8) at transfer.c:262 
#8  0xb713d172 in pup_sock_receive (sock=0x812f120, timeout=0, num_blocks=4294967295, error=0xbffff5a8) at transfer.c:296 
#9  0xb713d33f in pup_sock_input_callback (sock=0x812f120) at transfer.c:324 
#10 0xb713c0e9 in pup_sock_event_source_dispatch (source=0x81467e8, callback=0, data=0x0) at core.c:376 
#11 0xb775d2cd in g_main_dispatch () from /usr/lib/libglib-2.0.so.0 
#12 0xb775def4 in g_main_context_dispatch () from /usr/lib/libglib-2.0.so.0 
#13 0xb775e0dc in g_main_context_iterate () from /usr/lib/libglib-2.0.so.0 
#14 0xb775e54d in g_main_loop_run () from /usr/lib/libglib-2.0.so.0 
akash_rawal wrote: I downloaded the latest thunar, compiled and ran it myself, and found that it was destroying the mount object after every access.
Then this function in thunar opened my eyes.

Code: Select all

void
thunar_notify_unmount_finish (GMount *mount)
{
  NotifyNotification *notification;

  g_return_if_fail (G_IS_MOUNT (mount));

  notification = g_object_get_data (G_OBJECT (mount), "thunar-notification");
  if (notification != NULL)
    {
      notify_notification_close (notification, NULL);
      g_object_set_data (G_OBJECT (mount), "thunar-notification", NULL);
    }
}
Thunar must be assuming a reference on the mount object, else it won't store its data in those objects.

It was a reference counting problem, due to which mount objects were destroyed so often. It is fixed now. See first post.

But we had forgotten the question for which the thread was meant.

Should we handle entire event management in C, including kernel module loading?

akash_rawal
Posts: 229
Joined: Wed 25 Aug 2010, 15:38
Location: ISM Dhanbad, Jharkhand, India

#48 Post by akash_rawal »

I just observed ubuntu lucid setting up my pendrive faster than puppy had ever did.

I am now scanning ubuntu to see what it does. I haven't found anything yet but there are high chances that we'd get something to learn, I believe.

User avatar
technosaurus
Posts: 4853
Joined: Mon 19 May 2008, 01:24
Location: Blue Springs, MO
Contact:

#49 Post by technosaurus »

that is likely because puppy has several sleeps to ensure that older boxes have time to settle and pick up the drive (it also reduce cpu load of the daemon process) - *buntu really doesn't care about them.

if you have a good understanding of the kernel processes that occur when a new device is inserted, its much easier to get a simple solution

from my only basic understanding, I first started by monitoring /sys/block/* ... but further poking around reveals /proc/partitions is also updated for "drives"

a simple shell solution is to do something like

Code: Select all

tail -f /proc/partitions |while read LINE; do
<code here>
done;
in C this could just be equated to fopen /proc/partitions +while fgets line||sleep 1+<code>

the C for looking in /sys/block/*/uevent (the way I do it in shell in jwm_tools) would need dirent, lstat and strtok (to get the = value)

Sorry if this wasn't helpful, i am still writing C as a second language - have to write it shell first and then translate to C
Check out my [url=https://github.com/technosaurus]github repositories[/url]. I may eventually get around to updating my [url=http://bashismal.blogspot.com]blogspot[/url].

akash_rawal
Posts: 229
Joined: Wed 25 Aug 2010, 15:38
Location: ISM Dhanbad, Jharkhand, India

#50 Post by akash_rawal »

technosaurus wrote: that is likely because puppy has several sleeps to ensure that older boxes have time to settle and pick up the drive (it also reduce cpu load of the daemon process) - *buntu really doesn't care about them.
We've got to check whether ubuntu really doesn't care or doesn't need to care.

In most cases it is possible to wait precisely till the device is ready for use. This is usually the fastest solution, as no periodic checks are involved to see whether the device is ready. Also the response is immediate.

Anyways I have very little idea about what to do when a new device is plugged.

e.g. for drives we have to setup corresponding block devices (again, don't know how). After that things are easy: wait for udevd to setup device nodes, use blkid to get partitions and display them on desktop or sidebar, etc.

User avatar
jemimah
Posts: 4307
Joined: Wed 26 Aug 2009, 19:56
Location: Tampa, FL
Contact:

#51 Post by jemimah »

I updated to 0.0.7 and it seems a lot more stable.

However one weird issue that wasn't there in 0.0.5.

In saluki, if a volume has a label, I mount it on /mnt/<label> instead of /mnt/<devname>.

Thunar still works fine , but in 007 when I click on a drive icon in xfdesktop, it tries to open /mnt/<devname> instead of the correct mount point.

I looked at the code but I couldn't figure out what might have changed this behavior. Actually, I can't figure out why it worked in 005 - it looks like it shouldn't.

akash_rawal
Posts: 229
Joined: Wed 25 Aug 2010, 15:38
Location: ISM Dhanbad, Jharkhand, India

#52 Post by akash_rawal »

Strange, I cannot reproduce your problem despite all my attempts.

Anyways the volume monitor is designed to handle issues like this. The volume monitor gets all info from /proc/mounts (If that isn't found then it falls back to /etc/mtab and prints a warning to stderr). The only exception (since 0.0.7) is while mounting the drive, but then the monitor is itself executing the mount command with the generated mountpoint.

On mounting a drive with different mountpoint I get correct event:

Code: Select all

# pup-volume-monitor-admin --listen-drv
Volume mounted
Volume sda5:
	unix_dev=/dev/sda5
	label=WinXP
	fstype=ntfs
	uuid=545CC7085CC6E438
	drv_sysname=sda
	mntpnt=/mnt/WinXP
	mntflags = 0
Even on unmounting and mounting in quick succession I get correct mountpoint (although not correct event):

Code: Select all

# pup-volume-monitor-admin --listen-drv
Volume remounted
Volume sda5:
	unix_dev=/dev/sda5
	label=WinXP
	fstype=ntfs
	uuid=545CC7085CC6E438
	drv_sysname=sda
	mntpnt=/mnt/WinXP
	mntflags = 0
The source code for all this is found at daemon/mtabdata.c and daemon/monitor.c .
Attachments
screenshot1.jpg
thunar too works properly
(84.83 KiB) Downloaded 421 times

akash_rawal
Posts: 229
Joined: Wed 25 Aug 2010, 15:38
Location: ISM Dhanbad, Jharkhand, India

#53 Post by akash_rawal »

akash_rawal wrote: I just observed ubuntu lucid setting up my pendrive faster than puppy had ever did.
I see the performance issue solved in slacko 5.3.1. It detects my pendrive faster than even ubuntu 10.04 did.

Seems as if no need to rewrite full event management? Or are there any more performance issues?

User avatar
jemimah
Posts: 4307
Joined: Wed 26 Aug 2009, 19:56
Location: Tampa, FL
Contact:

#54 Post by jemimah »

Here's the code for mounting.

Code: Select all

gchar *mntpnt = g_build_filename(PUP_EVENT_MNTPNT, (gchar *) input, NULL);
gchar *cmd = g_strdup_printf("tmount '%s'", vol->unix_dev);
g_mkdir_with_parents(mntpnt, S_IRWXU | S_IRWXG | S_IRWXO);
g_spawn_command_line_sync(cmd, NULL, &stderr_data, &exitcode, &error);
//g_chmod(mntpnt, S_IRWXU | S_IRWXG | S_IRWXO);
g_free(cmd);
I know that *input is the device name at this point because the mkdir creates the directory /mnt/<devname>.

I don't use the regular mount command - because I need it to do a few extra things - like mounting in a nonstandard location.

Since the drive has not been mounted yet, it won't be in the mtab.

Once it is in the mtab then the monitor admin reports the correct value.

I've worked around this issue by symlinking /mnt/devname to /mnt/label in the tmount script. But maybe there is a better solution?


EDIT:

I've thought about this some more and I think I'm going to go back to mounting by device name again and symlink to the label name. It doesn't matter much any more since thunar and xfdesktop show the label either way.
Last edited by jemimah on Sun 06 May 2012, 20:06, edited 1 time in total.

User avatar
jemimah
Posts: 4307
Joined: Wed 26 Aug 2009, 19:56
Location: Tampa, FL
Contact:

#55 Post by jemimah »

My personal thought on rewriting the event manager are this.

You need to get buy-in from Barry to use it in woof if you want it to go mainstream. The best way to do this, IMO, is to start small, debug the heck out of it, and keep it modular.

Then I think udev rules could be developed to handle the desktop icons for ROX.

I'm working on udev rules to keep the fstab up to date (which is all Saluki needs from the event frontend now), but for some reason the blkid command returns incomplete output when run from a script called by udev as opposed to the command line. I can't figure out why.

jamesbond
Posts: 3433
Joined: Mon 26 Feb 2007, 05:02
Location: The Blue Marble

#56 Post by jamesbond »

jemimah wrote:Then I think udev rules could be developed to handle the desktop icons for ROX.
I'm doing exactly this in Fatdog-next. Not yet there yet, but I can already get the icons to display / removed when devices are inserted/removed.
I'm working on udev rules to keep the fstab up to date (which is all Saluki needs from the event frontend now), but for some reason the blkid command returns incomplete output when run from a script called by udev as opposed to the command line. I can't figure out why.
Check your PATH, make sure blkid you're executing is the full blkid and not busybox blkid.
Fatdog64 forum links: [url=http://murga-linux.com/puppy/viewtopic.php?t=117546]Latest version[/url] | [url=https://cutt.ly/ke8sn5H]Contributed packages[/url] | [url=https://cutt.ly/se8scrb]ISO builder[/url]

User avatar
jemimah
Posts: 4307
Joined: Wed 26 Aug 2009, 19:56
Location: Tampa, FL
Contact:

#57 Post by jemimah »

It didn't work with the full blkid either. I think udev itself is doing something weird to keep events in sync.

I figured out I can use guess_fstype instead and that works ok. Doesn't get me the drive label though.

Not a big deal now though since I changed the way my whole mounting system works.

akash_rawal
Posts: 229
Joined: Wed 25 Aug 2010, 15:38
Location: ISM Dhanbad, Jharkhand, India

#58 Post by akash_rawal »

jemimah wrote: Here's the code for mounting.

Code: Select all

gchar *mntpnt = g_build_filename(PUP_EVENT_MNTPNT, (gchar *) input, NULL);
gchar *cmd = g_strdup_printf("tmount '%s'", vol->unix_dev);
g_mkdir_with_parents(mntpnt, S_IRWXU | S_IRWXG | S_IRWXO);
g_spawn_command_line_sync(cmd, NULL, &stderr_data, &exitcode, &error);
//g_chmod(mntpnt, S_IRWXU | S_IRWXG | S_IRWXO);
g_free(cmd);
The correct code is:

Code: Select all

gchar *mntpnt;
if (vol->label && (g_strcmp0(vol->label, "") != 0))
    mntpnt = g_build_filename(PUP_EVENT_MNTPNT, vol->label, NULL);
else
    mntpnt = g_build_filename(PUP_EVENT_MNTPNT, (gchar *) input, NULL);

Rest of the code remains as original. This will also save you from extra hassle of finding labels.

akash_rawal
Posts: 229
Joined: Wed 25 Aug 2010, 15:38
Location: ISM Dhanbad, Jharkhand, India

#59 Post by akash_rawal »

jemimah wrote: but for some reason the blkid command returns incomplete output when run from a script called by udev as opposed to the command line. I can't figure out why.
udev removes the entire environment , including PATH variable. You have to re-export it yourself or use full path to blkid.

User avatar
jemimah
Posts: 4307
Joined: Wed 26 Aug 2009, 19:56
Location: Tampa, FL
Contact:

#60 Post by jemimah »

I get a segfault during startup occaisionally.

Code: Select all

Program received signal SIGSEGV, Segmentation fault.
0xb7d77358 in strstr () from /lib/libc.so.6
(gdb) bt
#0  0xb7d77358 in strstr () from /lib/libc.so.6
#1  0x0804b5d0 in volume_new_from_blkid_dev (volumes=0x804fc70, 
    drives=0x804fc38, dev=0x8053110, cache=0x8052890, udev_ctx=0x8050600)
    at volume.c:72
#2  0x0804bf1c in monitor_fill_data (mon=0x8051800) at monitor.c:86
#3  0x0804c6f8 in server_new (
    address=0x804cb3c "/tmp/pup_volume_monitor_socket", ctx=0x0) at server.c:49
#4  0x0804c77c in main (argc=<value optimized out>, argv=<value optimized out>)
    at main.c:12

Using g_strrstr instead of strstr seems to resolve whatever the issue is.

Post Reply