Shared memory woes

I'm searched the Web and this board, and I've read the man pages, and I've read two books that touch on this subject. I am getting a tad fustrated.

The man pages say it's best to pass in NULL for the second argument of shmat(), but if I do that, the pointer that's returned to me could be *anywhere* in the segment of shared memory. How could I find the beginning of the segment with only a pointer to some random spot in the middle?

Conversely, I could pass in the memory address to shmat() that was returned to me from the first time I attached shared memory. This had been working well for me for a while until I tried attaching to two seperate segments of shared memory that were created by two seperate processes. I received the EINVAL error. I don't understand what's invalid about what I'm trying to do. It used to work. Could it have something to do with the fact that both segments seem to have the same address? They have different IDs.

I feel like I'm missing something because every example I've looked at passes in NULL for the address.

If shmaddr isn't NULL and SHM_RND is asserted in shmflg,
the attach occurs at the address equal to shmaddr rounded
down to the nearest multiple of SHMLBA. Otherwise shmaddr
must be a page-aligned address at which the attach occurs.

Are you meeting the bolded requirement? If you're passing a non-page-aligned address then that could be the problem. You can try using the SHM_RND flag if you don't want to bother with making sure the address is page-aligned. It's hard to say what's wrong without seeing your code.

Hmm. The VERY FIRST call to shmat() must use NULL. The first call actually creates the segment, so the address is not defined prior to the call.

From my man page :

The segment is attached for reading if (shmflg & SHM_RDONLY) is
"true"; otherwise, it is attached for reading and writing. It is not
possible to attach a segment for write only.

If the shared memory segment has never been attached to by any process
prior to the current shmat() call, shmaddr must be specified as zero
and the segment is attached at a location selected by the operating
system. That location is identical in all processes accessing that
shared memory object. Once the operating system selects a location
for a shared memory segment, the same location will be used across any
subsequent shmat() and shmdt() calls on the segment until it is
removed by the IPC_RMID operation of shmctl().

If this is not the first shmat() call on the shared memory segment
throughout the system, shmaddr must either be zero or contain a
nonzero address that is identical to the one returned from previous
shmat() calls for that segment. Even if no processes are currently
attached to the segment, as long as the segment has been attached
before, the same rule applies.

Beej
any help?

I've read Beej's work and the man pages. I was sending NULL during the first attachment, and the address I was sending the second time was page-aligned.

It turns out that it works properly when sending in NULL for the address when attaching the second time. That it works is good enough for my employer, but I don't understand how it works. How is it that two pointers with different values point to the same segment of shared memory?