ipc.texi

Util-linux 软件包包含许多工具。其中比较重要的是加载、卸载、格式化、分区和管理硬盘驱动器

IPC_SET, SETVAL, SETALL : sem_ctime is updated.
@item
SETVAL, SETALL : Undo entries are cleared for altered semaphores in
all processes. Processes sleeping on the wait queues are
awakened if a semval becomes 0 or increases.@refill
@item
IPC_SET : sem_perm.uid, sem_perm.gid, sem_perm.mode are updated from
user supplied values.@refill
@end itemize

@noindent
Errors:
@noindent
EACCES : do not have permission for specified access.@*
@noindent
EFAULT : arg is not accessible.@*
@noindent
EIDRM  : The resource was removed.@*
@noindent
EINVAL : semid < 0 or semnum < 0 or semnum >= nsems.@*
@noindent
EPERM  : IPC_RMID, IPC_SET ... not creator, owner or super-user.@*
@noindent
ERANGE : arg.array[i].semval > SEMVMX or < 0 for some i.




@node semlimits, Shared Memory, semctl, Semaphores
@subsection Limits on Semaphore Resources

@noindent
Sizeof various structures:
@example
semid_ds    44   /* 1 per semaphore array .. dynamic */
sem          8   /* 1 for each semaphore in system .. dynamic */
sembuf       6   /* allocated by user */
sem_undo    20   /* 1 for each undo request .. dynamic */
@end example

@noindent
Limits :@*
@itemize @bullet
@item
SEMVMX  32767  semaphore maximum value (short).
@item
SEMMNI  number of semaphore identifiers (or arrays) system wide...policy.
@item
SEMMSL  maximum  number  of semaphores per id.
1 semid_ds per array, 1 struct sem per semaphore
=> SEMMSL =  (PAGE_SIZE - sizeof(semid_ds)) / sizeof(sem).
Implementation maximum SEMMSL = 500.@refill
@item
SEMMNS  maximum number of semaphores system wide ... policy.
Setting SEMMNS >= SEMMSL*SEMMNI makes it irrelevent.@refill
@item
SEMOPM 	Maximum number of operations in one semop call...policy.
@end itemize

@noindent
Unused or unimplemented:@*
@noindent
SEMAEM  adjust on exit max value.@*
@noindent
SEMMNU  number of undo structures system-wide.@*
@noindent
SEMUME  maximum number of undo entries per process.



@node Shared Memory, shmget, semlimits, Top
@section Shared Memory

Shared memory is distinct from the sharing of read-only code pages or
the sharing of unaltered data pages that is available due to the
copy-on-write mechanism. The essential difference is that the
shared pages are dirty (in the case of Shared memory) and can be
made to appear at a convenient location in the process' address space.@refill

@noindent
A shared segment is described by :
@example
struct shmid_ds
    struct  ipc_perm shm_perm;
    int     shm_segsz;              /* size of segment (bytes) */
    time_t  shm_atime;              /* last attach time */
    time_t  shm_dtime;              /* last detach time */
    time_t  shm_ctime;              /* last change time */
    ulong   *shm_pages;             /* internal page table */
    ushort  shm_cpid;               /* pid, creator */
    ushort  shm_lpid;               /* pid, last operation */
    short   shm_nattch;             /* no. of current attaches */
@end example

A shmget allocates a shmid_ds and an internal page table. A shmat
maps the segment into the process' address space with pointers
into the internal page table and the actual pages are faulted in
as needed. The memory associated with the segment must be explicitly
destroyed by calling shmctl with IPC_RMID.@refill

@menu
* shmget::
* shmat::
* shmdt::
* shmctl::
* shmlimits:: Limits imposed by this implementation.
@end menu


@node shmget, shmat, Shared Memory, Shared Memory
@subsection shmget

@noindent
A shared memory segment is allocated by a shmget system call:

@example
int shmget(key_t key, int size, int shmflg);
@end example

@itemize @bullet
@item
key : an integer usually got from @code{ftok} or IPC_PRIVATE
@item
size : size of the segment in bytes (SHMMIN <= size <= SHMMAX).
@item
shmflg :
@itemize @asis
@item
IPC_CREAT used to create a new resource
@item
IPC_EXCL used with IPC_CREAT to ensure failure if the resource exists.
@item
rwxrwxrwx  access permissions.
@end itemize
@item
returns : shmid on success. -1 on failure.
@end itemize

A descriptor for a shared memory segment is allocated if there isn't one
corresponding to the given key. The access permissions specified are
then copied into the @code{shm_perm} struct for the segment along with the
user-id etc. The user must use the IPC_CREAT flag or key = IPC_PRIVATE
to allocate a new segment.@refill

If the segment already exists, the access permissions are verified,
and a check is made to see that it is not marked for destruction.@refill

@code{size} is effectively rounded up to a multiple of PAGE_SIZE as shared
memory is allocated in pages.@refill

@noindent
Errors:@*
@noindent
EINVAL : (allocate) Size not in range specified above.@*
         (procure) Size greater than size of segment.@*
@noindent
EEXIST : (allocate) IPC_CREAT | IPC_EXCL specified and resource exists.@*
@noindent
EIDRM  : (procure) The resource is marked destroyed or was removed.@*
@noindent
ENOSPC : (allocate) All id's are taken (max of SHMMNI id's system-wide).
Allocating a segment of the requested size would exceed the
system wide limit on total shared memory (SHMALL).@refill
@*
@noindent
ENOENT : (procure) Resource does not exist and IPC_CREAT not specified.@*
@noindent
EACCES : (procure) Do not have permission for specified access.@*
@noindent
ENOMEM : (allocate) Could not allocate memory for shmid_ds or pg_table.



@node shmat, shmdt, shmget, Shared Memory
@subsection shmat

@noindent
Maps a shared segment into the process' address space.

@example
char *virt_addr;
virt_addr =  shmat (int shmid, char *shmaddr, int shmflg);
@end example

@itemize @bullet
@item
shmid : id got from call to shmget.
@item
shmaddr : requested attach address.@*
    If shmaddr is 0 the system finds an unmapped region.@*
    If a non-zero value is indicated the value must be page
    aligned or the user must specify the SHM_RND flag.@refill
@item
shmflg :@*
     SHM_RDONLY : request read-only attach.@*
     SHM_RND : attach address is rounded DOWN to a multiple of SHMLBA.
@item
returns: virtual address of attached segment. -1 on failure.
@end itemize

When shmaddr is 0, the attach address is determined by finding an
unmapped region in the address range 1G to 1.5G, starting at 1.5G
and coming down from there. The algorithm is very simple so you
are encouraged to avoid non-specific attaches.

@noindent
Algorithm:
@display
Determine attach address as described above.
Check region (shmaddr, shmaddr + size) is not mapped and allocate
page tables (undocumented SHM_REMAP flag!).
Map the region by setting up pointers into the internal page table.
Add a descriptor for the attach to the task struct for the process.
@code{shm_nattch}, @code{shm_lpid}, @code{shm_atime} are updated.
@end display

@noindent
Notes:@*
The @code{brk} value is not altered.
The segment is automatically detached when the process exits.
The same segment may be attached as read-only or read-write and
more than once in the process' address space.
A shmat can succeed on a segment marked for destruction.
The request for a particular type of attach is made using the SHM_RDONLY flag.
There is no notion of a write-only attach. The requested attach
permissions must fall within those allowed by @code{shm_perm.mode}.

@noindent
Errors:@*
@noindent
EACCES : Do not have permission for requested access.@*
@noindent
EINVAL : shmid < 0 or unused, shmaddr not aligned, attach at brk failed.@*
@noindent
EIDRM  : resource was removed.@*
@noindent
ENOMEM : Could not allocate memory for descriptor or page tables.


@node shmdt, shmctl, shmat, Shared Memory
@subsection shmdt

@example
int shmdt (char *shmaddr);
@end example

@itemize @bullet
@item
shmaddr : attach address of segment (returned by shmat).
@item
returns : 0 on success. -1 on failure.
@end itemize

An attached segment is detached and @code{shm_nattch} decremented. The
occupied region in user space is unmapped. The segment is destroyed
if it is marked for destruction and @code{shm_nattch} is 0.
@code{shm_lpid} and @code{shm_dtime} are updated.@refill

@noindent
Errors:@*
@noindent
EINVAL : No shared memory segment attached at shmaddr.


@node shmctl, shmlimits, shmdt, Shared Memory
@subsection shmctl

@noindent
Destroys allocated segments. Reads/Writes the control structures.

@example
int shmctl (int shmid, int cmd, struct shmid_ds *buf);
@end example

@itemize @bullet
@item
shmid : id got from call to shmget.
@item
cmd : IPC_STAT, IPC_SET, IPC_RMID (@xref{syscalls}).
@itemize @asis
@item
IPC_SET : Used to set the owner uid, gid, and shm_perms.mode field.
@item
IPC_RMID : The segment is marked destroyed. It is only destroyed
on the last detach.@refill
@item
IPC_STAT : The shmid_ds structure is copied into the user allocated buffer.
@end itemize
@item
buf : used to read (IPC_STAT) or write (IPC_SET) information.
@item
returns : 0 on success, -1 on failure.
@end itemize

The user must execute an IPC_RMID shmctl call to free the memory
allocated by the shared segment. Otherwise all the pages faulted in
will continue to live in memory or swap.@refill

@noindent
Errors:@*
@noindent
EACCES : Do not have permission for requested access.@*
@noindent
EFAULT : buf is not accessible.@*
@noindent
EINVAL : shmid < 0 or unused.@*
@noindent
EIDRM  : identifier destroyed.@*
@noindent
EPERM  : not creator, owner or super-user (IPC_SET, IPC_RMID).


@node shmlimits, Notes, shmctl, Shared Memory
@subsection Limits on Shared Memory Resources

@noindent
Limits:
@itemize @bullet
@item
SHMMNI  max num of shared segments system wide ... 4096.
@item
SHMMAX  max shared memory segment size (bytes) ... 4M
@item
SHMMIN  min shared memory segment size (bytes).
1 byte (though PAGE_SIZE is the effective minimum size).@refill
@item
SHMALL  max shared mem system wide (in pages) ... policy.
@item
SHMLBA  segment low boundary address multiple.
Must be page aligned. SHMLBA = PAGE_SIZE.@refill
@end itemize
@noindent
Unused or unimplemented:@*
SHMSEG : maximum number of shared segments per process.



@node Notes, Top, shmlimits, Top
@section Miscellaneous Notes

The system calls are mapped into one -- @code{sys_ipc}. This should be
transparent to the user.@refill

@subsection Semaphore @code{undo} requests

There is one sem_undo structure associated with a process for
each semaphore which was altered (with an undo request) by the process.
@code{sem_undo} structures are freed only when the process exits.

One major cause for unhappiness with the undo mechanism is that
it does not fit in with the notion of having an atomic set of
operations on an array. The undo requests for an array and each
semaphore therein may have been accumulated over many @code{semop}
calls. Thus use the undo mechanism with private semaphores only.@refill

Should the process sleep in @code{exit} or should all undo
operations be applied with the IPC_NOWAIT flag in effect?
Currently  those undo operations which go through immediately are
applied and those that require a wait are ignored silently.@refill

@subsection Shared memory, @code{malloc} and the @code{brk}.
Note that since this section was written the implementation was
changed so that non-specific attaches are done in the region
1G - 1.5G. However much of the following is still worth thinking
about so I left it in.

On many systems, the shared memory is allocated in a special region
of the address space ... way up somewhere. As mentioned earlier,
this implementation attaches shared segments at the lowest possible
address. Thus if you plan to use @code{malloc}, it is wise to malloc a
large space and then proceed to attach the shared segments. This way
malloc sets the brk sufficiently above the region it will use.@refill

Alternatively you can use @code{sbrk} to adjust the @code{brk} value
as you make shared memory attaches. The implementation is not very
smart about selecting attach addresses. Using the system default
addresses will result in fragmentation if detaches do not occur
in the reverse sequence as attaches.@refill

Taking control of the matter is probably best. The rule applied
is that attaches are allowed in unmapped regions other than
in the text space (see <a.out.h>). Also remember that attach addresses
and segment sizes are multiples of PAGE_SIZE.@refill

One more trap (I quote Bruno on this). If you use malloc() to get space
for your shared memory (ie. to fix the @code{brk}), you must ensure you
get an unmapped address range. This means you must mallocate more memory
than you had ever allocated before. Memory returned by malloc(), used,
then freed by free() and then again returned by malloc is no good.
Neither is calloced memory.@refill

Note that a shared memory region remains a shared memory region until
you unmap it. Attaching a segment at the @code{brk} and calling malloc
after that will result in an overlap of what malloc thinks is its
space with what is really a shared memory region. For example in the case
of a read-only attach, you will not be able to write to the overlapped
portion.@refill


@subsection Fork, exec and exit

On a fork, the child inherits attached shared memory segments but
not the semaphore undo information.@refill

In the case of an exec, the attached shared segments are detached.
The sem undo information however remains intact.@refill

Upon exit, all attached shared memory segments are detached.
The adjust values in the undo structures are added to the relevant semvals
if the operations are permitted. Disallowed operations are ignored.@refill


@subsection Other Features

These features of the current implementation are
likely to be modified in the future.

The SHM_LOCK and SHM_UNLOCK flag are available (super-user) for use with the
@code{shmctl} call to prevent swapping of a shared segment. The user
must fault in any pages that are required to be present after locking
is enabled.

The IPC_INFO, MSG_STAT, MSG_INFO, SHM_STAT, SHM_INFO, SEM_STAT, SEMINFO
@code{ctl} calls are used by the @code{ipcs} program to provide information
on allocated resources. These can be modified as needed or moved to a proc
file system interface.


@sp 3
Thanks to Ove Ewerlid, Bruno Haible, Ulrich Pegelow and Linus Torvalds
for ideas, tutorials, bug reports and fixes, and merriment. And more
thanks to Bruno.


@contents
@bye