don't use SEM_UNDO because of ridiculously small undo limits on some

[Samba.git] / source / locking / shmem_sysv.c

/* 
   Unix SMB/Netbios implementation.
   Version 1.9.
   Shared memory functions - SYSV IPC implementation
   Copyright (C) Andrew Tridgell 1997
   
   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; either version 2 of the License, or
   (at your option) any later version.
   
   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.
   
   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.

*/

#include "includes.h"


#ifdef USE_SYSV_IPC

extern int DEBUGLEVEL;

#define SHMEM_KEY ((key_t)0x280267)
#define SEMAPHORE_KEY (SHMEM_KEY+2)

#define SHM_MAGIC 0x53484100
#define SHM_VERSION 2

#ifdef SHM_R
#define IPC_PERMS ((SHM_R | SHM_W) | (SHM_R>>3) | (SHM_R>>6))
#else
#define IPC_PERMS 0644
#endif

#ifdef SECURE_SEMAPHORES
/* secure semaphores are slow because we have to do a become_root()
   on every call! */
#define SEMAPHORE_PERMS IPC_PERMS
#else
#define SEMAPHORE_PERMS 0666
#endif

#ifdef SEMMSL
#define SHMEM_HASH_SIZE (SEMMSL-1)
#else
#define SHMEM_HASH_SIZE 63
#endif

#define MIN_SHM_SIZE 0x1000

static int shm_id;
static int sem_id;
static int shm_size;
static int hash_size;
static int global_lock_count;

struct ShmHeader {
   int shm_magic;
   int shm_version;
   int total_size;      /* in bytes */
   BOOL consistent;
   int first_free_off;
   int userdef_off;    /* a userdefined offset. can be used to store
                          root of tree or list */
   struct {             /* a cell is a range of bytes of sizeof(struct
                           ShmBlockDesc) size */
           int cells_free;
           int cells_used;
           int cells_system; /* number of cells used as allocated
                                block descriptors */
   } statistics;
};

#define SHM_NOT_FREE_OFF (-1)
struct ShmBlockDesc
{
   int next;    /* offset of next block in the free list or
                   SHM_NOT_FREE_OFF when block in use */
   int size;   /* user size in BlockDescSize units */
};

#define EOList_Addr     NULL
#define EOList_Off      (0)

#define CellSize        sizeof(struct ShmBlockDesc)

/* HeaderSize aligned on a 8 byte boundary */
#define AlignedHeaderSize ((sizeof(struct ShmHeader)+7) & ~7)

static struct ShmHeader *shm_header_p = NULL;
static BOOL shm_initialize_called = False;

static int read_only;

static BOOL sem_change(int i, int op)
{
#ifdef SECURE_SEMAPHORES
        extern struct current_user current_user;
        int became_root=0;
#endif
        struct sembuf sb;
        int ret;

        if (read_only) return True;

#ifdef SECURE_SEMAPHORES
        if (current_user.uid != 0) {
                become_root(0);
                became_root = 1;
        }
#endif

        sb.sem_num = i;
        sb.sem_op = op;
        sb.sem_flg = 0;

        ret = semop(sem_id, &sb, 1);

        if (ret != 0) {
                DEBUG(0,("ERROR: sem_change(%d,%d) failed (%s)\n", 
                         i, op, strerror(errno)));
        }

#ifdef SECURE_SEMAPHORES
        if (became_root) {
                unbecome_root(0);
        }
#endif

        return ret == 0;
}

static BOOL global_lock(void)
{
        global_lock_count++;
        if (global_lock_count == 1)
                return sem_change(0, -1);
        return True;
}

static BOOL global_unlock(void)
{
        global_lock_count--;
        if (global_lock_count == 0)
                return sem_change(0, 1);
        return True;
}

static void *shm_offset2addr(int offset)
{
   if (offset == 0 )
      return (void *)(0);
   
   if (!shm_header_p)
      return (void *)(0);
   
   return (void *)((char *)shm_header_p + offset);
}

static int shm_addr2offset(void *addr)
{
   if (!addr)
      return 0;
   
   if (!shm_header_p)
      return 0;
   
   return (int)((char *)addr - (char *)shm_header_p);
}


static int shm_alloc(int size)
{
        unsigned num_cells ;
        struct ShmBlockDesc *scanner_p;
        struct ShmBlockDesc *prev_p;
        struct ShmBlockDesc *new_p;
        int result_offset;
   
   
        if (!shm_header_p) {
                /* not mapped yet */
                DEBUG(0,("ERROR shm_alloc : shmem not mapped\n"));
                return 0;
        }
        
        global_lock();
        
        if (!shm_header_p->consistent) {
                DEBUG(0,("ERROR shm_alloc : shmem not consistent\n"));
                global_unlock();
                return 0;
        }
        
        /* calculate the number of cells */
        num_cells = (size + (CellSize-1)) / CellSize;
        
        /* set start of scan */
        prev_p = (struct ShmBlockDesc *)shm_offset2addr(shm_header_p->first_free_off);
        scanner_p =     prev_p ;
        
        /* scan the free list to find a matching free space */
        while ((scanner_p != EOList_Addr) && (scanner_p->size < num_cells)) {
                prev_p = scanner_p;
                scanner_p = (struct ShmBlockDesc *)shm_offset2addr(scanner_p->next);
        }
   
        /* at this point scanner point to a block header or to the end of
           the list */
        if (scanner_p == EOList_Addr) {
                DEBUG(0,("ERROR shm_alloc : alloc of %d bytes failed\n",size));
                global_unlock();
                return (0);
        }
   
        /* going to modify shared mem */
        shm_header_p->consistent = False;
        
        /* if we found a good one : scanner == the good one */
        if (scanner_p->size > num_cells + 2) {
                /* Make a new one */
                new_p = scanner_p + 1 + num_cells;
                new_p->size = scanner_p->size - (num_cells + 1);
                new_p->next = scanner_p->next;
                scanner_p->size = num_cells;
                scanner_p->next = shm_addr2offset(new_p);

                shm_header_p->statistics.cells_free -= 1;
                shm_header_p->statistics.cells_system += 1;
        }

        /* take it from the free list */
        if (prev_p == scanner_p) {
                shm_header_p->first_free_off = scanner_p->next;
        } else {
                prev_p->next = scanner_p->next;
        }
        shm_header_p->statistics.cells_free -= scanner_p->size;
        shm_header_p->statistics.cells_used += scanner_p->size;

        result_offset = shm_addr2offset(&(scanner_p[1]));
        scanner_p->next = SHM_NOT_FREE_OFF;

        /* end modification of shared mem */
        shm_header_p->consistent = True;

        global_unlock();
        
        DEBUG(6,("shm_alloc : allocated %d bytes at offset %d\n",
                 size,result_offset));

        return result_offset;
}   

static void shm_solve_neighbors(struct ShmBlockDesc *head_p )
{
        struct ShmBlockDesc *next_p;
   
        /* Check if head_p and head_p->next are neighbors and if so
           join them */
        if ( head_p == EOList_Addr ) return ;
        if ( head_p->next == EOList_Off ) return ;
   
        next_p = (struct ShmBlockDesc *)shm_offset2addr(head_p->next);
        if ((head_p + head_p->size + 1) == next_p) {
                head_p->size += next_p->size + 1; /* adapt size */
                head_p->next = next_p->next; /* link out */
      
                shm_header_p->statistics.cells_free += 1;
                shm_header_p->statistics.cells_system -= 1;
        }
}


static BOOL shm_free(int offset)
{
        struct ShmBlockDesc *header_p; /* pointer to header of
                                          block to free */
        struct ShmBlockDesc *scanner_p; /* used to scan the list */
        struct ShmBlockDesc *prev_p; /* holds previous in the
                                        list */
   
        if (!shm_header_p) {
                /* not mapped yet */
                DEBUG(0,("ERROR shm_free : shmem not mapped\n"));
                return False;
        }
        
        global_lock();
        
        if (!shm_header_p->consistent) {
                DEBUG(0,("ERROR shm_free : shmem not consistent\n"));
                global_unlock();
                return False;
        }
        
        /* make pointer to header of block */
        header_p = ((struct ShmBlockDesc *)shm_offset2addr(offset) - 1); 
        
        if (header_p->next != SHM_NOT_FREE_OFF) {
                DEBUG(0,("ERROR shm_free : bad offset (%d)\n",offset));
                global_unlock();
                return False;
        }
        
        /* find a place in the free_list to put the header in */
        
        /* set scanner and previous pointer to start of list */
        prev_p = (struct ShmBlockDesc *)
                shm_offset2addr(shm_header_p->first_free_off);
        scanner_p = prev_p ;
        
        while ((scanner_p != EOList_Addr) && 
               (scanner_p < header_p)) { 
                /* while we didn't scan past its position */
                prev_p = scanner_p ;
                scanner_p = (struct ShmBlockDesc *)
                        shm_offset2addr(scanner_p->next);
        }
        
        shm_header_p->consistent = False;
        
        DEBUG(6,("shm_free : freeing %d bytes at offset %d\n",
                 header_p->size*CellSize,offset));
        
        if (scanner_p == prev_p) {
                shm_header_p->statistics.cells_free += header_p->size;
                shm_header_p->statistics.cells_used -= header_p->size;
                
                /* we must free it at the beginning of the list */
                shm_header_p->first_free_off = shm_addr2offset(header_p);
                /* set the free_list_pointer to this block_header */
                
                /* scanner is the one that was first in the list */
                header_p->next = shm_addr2offset(scanner_p);
                shm_solve_neighbors(header_p); 
                
                shm_header_p->consistent = True;
        } else {
                shm_header_p->statistics.cells_free += header_p->size;
                shm_header_p->statistics.cells_used -= header_p->size;
                
                prev_p->next = shm_addr2offset(header_p);
                header_p->next = shm_addr2offset(scanner_p);
                shm_solve_neighbors(header_p) ;
                shm_solve_neighbors(prev_p) ;
           
                shm_header_p->consistent = True;
        }

        global_unlock();
        return True;
}


/* 
 * Function to create the hash table for the share mode entries. Called
 * when smb shared memory is global locked.
 */
static BOOL shm_create_hash_table(unsigned int hash_entries)
{
        int size = hash_entries * sizeof(int);

        global_lock();
        shm_header_p->userdef_off = shm_alloc(size);

        if(shm_header_p->userdef_off == 0) {
                DEBUG(0,("shm_create_hash_table: Failed to create hash table of size %d\n",
                         size));
                global_unlock();
                return False;
        }

        /* Clear hash buckets. */
        memset(shm_offset2addr(shm_header_p->userdef_off), '\0', size);
        global_unlock();
        return True;
}


static BOOL shm_validate_header(int size)
{
        if(!shm_header_p) {
                /* not mapped yet */
                DEBUG(0,("ERROR shm_validate_header : shmem not mapped\n"));
                return False;
        }
   
        if(shm_header_p->shm_magic != SHM_MAGIC) {
                DEBUG(0,("ERROR shm_validate_header : bad magic\n"));
                return False;
        }

        if(shm_header_p->shm_version != SHM_VERSION) {
                DEBUG(0,("ERROR shm_validate_header : bad version %X\n",
                         shm_header_p->shm_version));
                return False;
        }
   
        if(shm_header_p->total_size != size) {
                DEBUG(0,("ERROR shmem size mismatch (old = %d, new = %d)\n",
                         shm_header_p->total_size,size));
                return False;
        }

        if(!shm_header_p->consistent) {
                DEBUG(0,("ERROR shmem not consistent\n"));
                return False;
        }
        return True;
}


static BOOL shm_initialize(int size)
{
        struct ShmBlockDesc * first_free_block_p;
        
        DEBUG(5,("shm_initialize : initializing shmem size %d\n",size));
   
        if( !shm_header_p ) {
                /* not mapped yet */
                DEBUG(0,("ERROR shm_initialize : shmem not mapped\n"));
                return False;
        }
   
        shm_header_p->shm_magic = SHM_MAGIC;
        shm_header_p->shm_version = SHM_VERSION;
        shm_header_p->total_size = size;
        shm_header_p->first_free_off = AlignedHeaderSize;
        shm_header_p->userdef_off = 0;
        
        first_free_block_p = (struct ShmBlockDesc *)
                shm_offset2addr(shm_header_p->first_free_off);
        first_free_block_p->next = EOList_Off;
        first_free_block_p->size = 
                (size - (AlignedHeaderSize+CellSize))/CellSize;   
        shm_header_p->statistics.cells_free = first_free_block_p->size;
        shm_header_p->statistics.cells_used = 0;
        shm_header_p->statistics.cells_system = 1;
   
        shm_header_p->consistent = True;
   
        shm_initialize_called = True;

        return True;
}
   
static BOOL shm_close( void )
{
        return True;
}


static int shm_get_userdef_off(void)
{
   if (!shm_header_p)
      return 0;
   else
      return shm_header_p->userdef_off;
}


/*******************************************************************
  Lock a particular hash bucket entry.
  ******************************************************************/
static BOOL shm_lock_hash_entry(unsigned int entry)
{
        return sem_change(entry+1, -1);
}

/*******************************************************************
  Unlock a particular hash bucket entry.
  ******************************************************************/
static BOOL shm_unlock_hash_entry(unsigned int entry)
{
        return sem_change(entry+1, 1);
}


/*******************************************************************
  Gather statistics on shared memory usage.
  ******************************************************************/
static BOOL shm_get_usage(int *bytes_free,
                          int *bytes_used,
                          int *bytes_overhead)
{
        if(!shm_header_p) {
                /* not mapped yet */
                DEBUG(0,("ERROR shm_free : shmem not mapped\n"));
                return False;
        }

        *bytes_free = shm_header_p->statistics.cells_free * CellSize;
        *bytes_used = shm_header_p->statistics.cells_used * CellSize;
        *bytes_overhead = shm_header_p->statistics.cells_system * CellSize + 
                AlignedHeaderSize;
        
        return True;
}


/*******************************************************************
hash a number into a hash_entry
  ******************************************************************/
static unsigned shm_hash_size(void)
{
        return hash_size;
}


static struct shmem_ops shmops = {
        shm_close,
        shm_alloc,
        shm_free,
        shm_get_userdef_off,
        shm_offset2addr,
        shm_addr2offset,
        shm_lock_hash_entry,
        shm_unlock_hash_entry,
        shm_get_usage,
        shm_hash_size,
};

/*******************************************************************
  open the shared memory
  ******************************************************************/
struct shmem_ops *sysv_shm_open(int ronly)
{
        BOOL created_new = False;
        BOOL other_processes;
        struct shmid_ds shm_ds;
        struct semid_ds sem_ds;
        union semun su;
        int i;
        int pid;

        read_only = ronly;

        shm_size = lp_shmem_size();

        DEBUG(4,("Trying sysv shmem open of size %d\n", shm_size));

        /* first the semaphore */
        sem_id = semget(SEMAPHORE_KEY, 0, 0);
        if (sem_id == -1) {
                if (read_only) return NULL;

                hash_size = SHMEM_HASH_SIZE;

                while (hash_size > 1) {
                        sem_id = semget(SEMAPHORE_KEY, hash_size+1, 
                                        IPC_CREAT|IPC_EXCL| SEMAPHORE_PERMS);
                        if (sem_id != -1 || errno != EINVAL) break;
                        hash_size--;
                }

                if (sem_id == -1) {
                        DEBUG(0,("Can't create or use semaphore %s\n", 
                                 strerror(errno)));
                }   

                if (sem_id != -1) {
                        su.val = 1;
                        for (i=0;i<hash_size+1;i++) {
                                if (semctl(sem_id, i, SETVAL, su) != 0) {
                                        DEBUG(1,("Failed to init semaphore %d\n", i));
                                }
                        }
                }
        }
        if (shm_id == -1) {
                sem_id = semget(SEMAPHORE_KEY, 0, 0);
        }
        if (sem_id == -1) {
                DEBUG(0,("Can't create or use semaphore %s\n", 
                         strerror(errno)));
                return NULL;
        }   

        su.buf = &sem_ds;
        if (semctl(sem_id, 0, IPC_STAT, su) != 0) {
                DEBUG(0,("ERROR shm_open : can't IPC_STAT\n"));
        }
        hash_size = sem_ds.sem_nsems-1;

        if (!read_only) {
                if (sem_ds.sem_perm.cuid != 0 || sem_ds.sem_perm.cgid != 0) {
                        DEBUG(0,("ERROR: root did not create the semaphore\n"));
                        return NULL;
                }

                if (semctl(sem_id, 0, GETVAL, su) == 0 &&
                    !process_exists((pid=semctl(sem_id, 0, GETPID, su)))) {
                        DEBUG(0,("WARNING: clearing global IPC lock set by dead process %d\n",
                                 pid));
                        su.val = 1;
                        if (semctl(sem_id, 0, SETVAL, su) != 0) {
                                DEBUG(0,("ERROR: Failed to clear global lock\n"));
                        }
                }

                sem_ds.sem_perm.mode = SEMAPHORE_PERMS;
                if (semctl(sem_id, 0, IPC_SET, su) != 0) {
                        DEBUG(0,("ERROR shm_open : can't IPC_SET\n"));
                }
        }

        
        
        if (!global_lock())
                return NULL;


        for (i=1;i<hash_size+1;i++) {
                if (semctl(sem_id, i, GETVAL, su) == 0 && 
                    !process_exists((pid=semctl(sem_id, i, GETPID, su)))) {
                        DEBUG(1,("WARNING: clearing IPC lock %d set by dead process %d\n", 
                                 i, pid));
                        su.val = 1;
                        if (semctl(sem_id, i, SETVAL, su) != 0) {
                                DEBUG(0,("ERROR: Failed to clear IPC lock %d\n", i));
                        }
                }
        }
        
        /* try to use an existing key */
        shm_id = shmget(SHMEM_KEY, shm_size, 0);
        
        /* if that failed then create one */
        if (shm_id == -1) {
                if (read_only) return NULL;
                while (shm_size > MIN_SHM_SIZE) {
                        shm_id = shmget(SHMEM_KEY, shm_size, 
                                        IPC_CREAT | IPC_EXCL | IPC_PERMS);
                        if (shm_id != -1 || errno != EINVAL) break;
                        shm_size *= 0.9;
                }
                created_new = (shm_id != -1);
        }
        
        if (shm_id == -1) {
                DEBUG(0,("Can't create or use IPC area\n"));
                global_unlock();
                return NULL;
        }   
        
        
        shm_header_p = (struct ShmHeader *)shmat(shm_id, 0, 
                                                 read_only?SHM_RDONLY:0);
        if ((int)shm_header_p == -1) {
                DEBUG(0,("Can't attach to IPC area\n"));
                global_unlock();
                return NULL;
        }

        /* to find out if some other process is already mapping the file,
           we use a registration file containing the processids of the file
           mapping processes */
        if (shmctl(shm_id, IPC_STAT, &shm_ds) != 0) {
                DEBUG(0,("ERROR shm_open : can't IPC_STAT\n"));
        }

        if (!read_only) {
                if (shm_ds.shm_perm.cuid != 0 || shm_ds.shm_perm.cgid != 0) {
                        DEBUG(0,("ERROR: root did not create the shmem\n"));
                        global_unlock();
                        return NULL;
                }
        }

        shm_size = shm_ds.shm_segsz;

        other_processes = (shm_ds.shm_nattch > 1);

        if (!read_only && !other_processes) {
                memset((char *)shm_header_p, 0, shm_size);
                shm_initialize(shm_size);
                shm_create_hash_table(hash_size);
                DEBUG(3,("Initialised IPC area of size %d\n", shm_size));
        } else if (!shm_validate_header(shm_size)) {
                /* existing file is corrupt, samba admin should remove
                   it by hand */
                DEBUG(0,("ERROR shm_open : corrupt IPC area - remove it!\n"));
                global_unlock();
                return NULL;
        }
   
        global_unlock();
        return &shmops;
}



#else 
 int ipc_dummy_procedure(void)
{return 0;}
#endif