shm.c

Linux 1.0 内核C源代码 Linux最早版本代码 由Linus Torvalds亲自书写的

/*
 * linux/ipc/shm.c
 * Copyright (C) 1992, 1993 Krishna Balasubramanian 
 *         Many improvements/fixes by Bruno Haible.
 * assume user segments start at 0x0
 */

#include <linux/errno.h>
#include <asm/segment.h>
#include <linux/sched.h>
#include <linux/ipc.h> 
#include <linux/shm.h>
#include <linux/stat.h>
#include <linux/malloc.h>

extern int ipcperms (struct ipc_perm *ipcp, short semflg);
extern unsigned int get_swap_page(void);
static int findkey (key_t key);
static int newseg (key_t key, int shmflg, int size);
static int shm_map (struct shm_desc *shmd, int remap);
static void killseg (int id);

static int shm_tot = 0;  /* total number of shared memory pages */
static int shm_rss = 0; /* number of shared memory pages that are in memory */
static int shm_swp = 0; /* number of shared memory pages that are in swap */
static int max_shmid = 0; /* every used id is <= max_shmid */
static struct wait_queue *shm_lock = NULL;
static struct shmid_ds *shm_segs[SHMMNI];

static unsigned short shm_seq = 0; /* incremented, for recognizing stale ids */

/* some statistics */
static ulong swap_attempts = 0;
static ulong swap_successes = 0;
static ulong used_segs = 0;

void shm_init (void)
{
	int id;
    
       	for (id = 0; id < SHMMNI; id++) 
		shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
	shm_tot = shm_rss = shm_seq = max_shmid = used_segs = 0;
	shm_lock = NULL;
	return;
}

static int findkey (key_t key)    
{
	int id;
	struct shmid_ds *shp;
	
	for (id=0; id <= max_shmid; id++) {
		while ((shp = shm_segs[id]) == IPC_NOID) 
			sleep_on (&shm_lock);
		if (shp == IPC_UNUSED)
			continue;
		if (key == shp->shm_perm.key) 
			return id;
	}
	return -1;
}

/* 
 * allocate new shmid_ds and pgtable. protected by shm_segs[id] = NOID.
 */
static int newseg (key_t key, int shmflg, int size)
{
	struct shmid_ds *shp;
	int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
	int id, i;

	if (size < SHMMIN)
		return -EINVAL;
	if (shm_tot + numpages >= SHMALL)
		return -ENOSPC;
	for (id=0; id < SHMMNI; id++)
		if (shm_segs[id] == IPC_UNUSED) {
			shm_segs[id] = (struct shmid_ds *) IPC_NOID;
			goto found;
		}
	return -ENOSPC;

found:
	shp = (struct shmid_ds *) kmalloc (sizeof (*shp), GFP_KERNEL);
	if (!shp) {
		shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
		if (shm_lock)
			wake_up (&shm_lock);
		return -ENOMEM;
	}

	shp->shm_pages = (ulong *) kmalloc (numpages*sizeof(ulong),GFP_KERNEL);
	if (!shp->shm_pages) {
		shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
		if (shm_lock)
			wake_up (&shm_lock);
		kfree_s (shp, sizeof (*shp));
		return -ENOMEM;
	}

	for (i=0; i< numpages; shp->shm_pages[i++] = 0);
	shm_tot += numpages;
	shp->shm_perm.key = key;
	shp->shm_perm.mode = (shmflg & S_IRWXUGO);
	shp->shm_perm.cuid = shp->shm_perm.uid = current->euid;
	shp->shm_perm.cgid = shp->shm_perm.gid = current->egid;
	shp->shm_perm.seq = shm_seq;
	shp->shm_segsz = size;
	shp->shm_cpid = current->pid;
	shp->attaches = NULL;
	shp->shm_lpid = shp->shm_nattch = 0;
	shp->shm_atime = shp->shm_dtime = 0;
	shp->shm_ctime = CURRENT_TIME;
	shp->shm_npages = numpages;

	if (id > max_shmid)
		max_shmid = id;
	shm_segs[id] = shp;
	used_segs++;
	if (shm_lock)
		wake_up (&shm_lock);
	return id + (int)shm_seq*SHMMNI;
}

int sys_shmget (key_t key, int size, int shmflg)
{
	struct shmid_ds *shp;
	int id = 0;
	
	if (size < 0 || size > SHMMAX)
		return -EINVAL;
	if (key == IPC_PRIVATE) 
		return newseg(key, shmflg, size);
	if ((id = findkey (key)) == -1) {
		if (!(shmflg & IPC_CREAT))
			return -ENOENT;
		return newseg(key, shmflg, size);
	} 
	if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL))
		return -EEXIST;
	shp = shm_segs[id];
	if (shp->shm_perm.mode & SHM_DEST)
		return -EIDRM;
	if (size > shp->shm_segsz)
		return -EINVAL;
	if (ipcperms (&shp->shm_perm, shmflg))
		return -EACCES;
	return shp->shm_perm.seq*SHMMNI + id;
}

/* 
 * Only called after testing nattch and SHM_DEST.
 * Here pages, pgtable and shmid_ds are freed.
 */
static void killseg (int id)
{
	struct shmid_ds *shp;
	int i, numpages;
	ulong page;

	shp = shm_segs[id];
	if (shp == IPC_NOID || shp == IPC_UNUSED) {
		printk ("shm nono: killseg called on unused seg id=%d\n", id);
		return;
	}
	shp->shm_perm.seq++;     /* for shmat */
	numpages = shp->shm_npages; 
	shm_seq++;
	shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
	used_segs--;
	if (id == max_shmid) 
		while (max_shmid && (shm_segs[--max_shmid] == IPC_UNUSED));
	if (!shp->shm_pages) {
		printk ("shm nono: killseg shp->pages=NULL. id=%d\n", id);
		return;
	}
	for (i=0; i< numpages ; i++) {
		if (!(page = shp->shm_pages[i]))
			continue;
		if (page & 1) {
			free_page (page & PAGE_MASK);
			shm_rss--;
		} else {
			swap_free (page);
			shm_swp--;
		}
	}
	kfree_s (shp->shm_pages, numpages * sizeof (ulong));
	shm_tot -= numpages;
	kfree_s (shp, sizeof (*shp));
	return;
}

int sys_shmctl (int shmid, int cmd, struct shmid_ds *buf)
{
	struct shmid_ds *shp, tbuf;
	struct ipc_perm *ipcp;
	int id, err;
	
	if (cmd < 0 || shmid < 0)
		return -EINVAL;
	if (cmd == IPC_SET) {
		if (!buf)
			return -EFAULT;
		err = verify_area (VERIFY_READ, buf, sizeof (*buf));
		if (err)
			return err;
		memcpy_fromfs (&tbuf, buf, sizeof (*buf));
	}

	switch (cmd) { /* replace with proc interface ? */
	case IPC_INFO: 
	{
		struct shminfo shminfo;
		if (!buf)
			return -EFAULT;
		shminfo.shmmni = SHMMNI;
		shminfo.shmmax = SHMMAX;
		shminfo.shmmin = SHMMIN;
		shminfo.shmall = SHMALL;
		shminfo.shmseg = SHMSEG;
		err = verify_area (VERIFY_WRITE, buf, sizeof (struct shminfo));
		if (err)
			return err;
		memcpy_tofs (buf, &shminfo, sizeof(struct shminfo));
		return max_shmid;
	}
	case SHM_INFO: 
	{ 
		struct shm_info shm_info;
		if (!buf)
			return -EFAULT;
		err = verify_area (VERIFY_WRITE, buf, sizeof (shm_info));
		if (err)
			return err;
		shm_info.used_ids = used_segs; 
		shm_info.shm_rss = shm_rss;
		shm_info.shm_tot = shm_tot;
		shm_info.shm_swp = shm_swp;
		shm_info.swap_attempts = swap_attempts;
		shm_info.swap_successes = swap_successes;
		memcpy_tofs (buf, &shm_info, sizeof(shm_info));
		return max_shmid;
	}
	case SHM_STAT:
		if (!buf)
			return -EFAULT;
		err = verify_area (VERIFY_WRITE, buf, sizeof (*shp));
		if (err)
			return err;
		if (shmid > max_shmid)
			return -EINVAL;
		shp = shm_segs[shmid];
		if (shp == IPC_UNUSED || shp == IPC_NOID)
			return -EINVAL;
		if (ipcperms (&shp->shm_perm, S_IRUGO))
			return -EACCES;
		id = shmid + shp->shm_perm.seq * SHMMNI; 
		memcpy_tofs (buf, shp, sizeof(*shp));
		return id;
	}
	
	shp = shm_segs[id = shmid % SHMMNI];
	if (shp == IPC_UNUSED || shp == IPC_NOID)
		return -EINVAL;
	ipcp = &shp->shm_perm;
	if (ipcp->seq != shmid / SHMMNI) 
		return -EIDRM;
	
	switch (cmd) {
	case SHM_UNLOCK:
		if (!suser())
			return -EPERM;
		if (!(ipcp->mode & SHM_LOCKED))
			return -EINVAL;
		ipcp->mode &= ~SHM_LOCKED;
		break;
	case SHM_LOCK:
/* Allow superuser to lock segment in memory */
/* Should the pages be faulted in here or leave it to user? */
/* need to determine interaction with current->swappable */
		if (!suser())
			return -EPERM;
		if (ipcp->mode & SHM_LOCKED)
			return -EINVAL;
		ipcp->mode |= SHM_LOCKED;
		break;
	case IPC_STAT:
		if (ipcperms (ipcp, S_IRUGO))
			return -EACCES;
		if (!buf)
			return -EFAULT;
		err = verify_area (VERIFY_WRITE, buf, sizeof (*shp));
		if (err)
			return err;
		memcpy_tofs (buf, shp, sizeof(*shp));
		break;
	case IPC_SET:
		if (suser() || current->euid == shp->shm_perm.uid ||
		    current->euid == shp->shm_perm.cuid) {
			ipcp->uid = tbuf.shm_perm.uid;
			ipcp->gid = tbuf.shm_perm.gid;
			ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
				| (tbuf.shm_perm.mode & S_IRWXUGO);
			shp->shm_ctime = CURRENT_TIME;
			break;
		}
		return -EPERM;
	case IPC_RMID:
		if (suser() || current->euid == shp->shm_perm.uid ||
		    current->euid == shp->shm_perm.cuid) {
			shp->shm_perm.mode |= SHM_DEST;
			if (shp->shm_nattch <= 0) 
				killseg (id);
			break;
		}
		return -EPERM;
	default:
		return -EINVAL;
	}
	return 0;
}

/*
 * check range is unmapped, ensure page tables exist
 * mark page table entries with shm_sgn.
 * if remap != 0 the range is remapped.
 */
static int shm_map (struct shm_desc *shmd, int remap)
{
	unsigned long invalid = 0;
	unsigned long *page_table;
	unsigned long tmp, shm_sgn;
	unsigned long page_dir = shmd->task->tss.cr3;
	
	/* check that the range is unmapped and has page_tables */
	for (tmp = shmd->start; tmp < shmd->end; tmp += PAGE_SIZE) { 
		page_table = PAGE_DIR_OFFSET(page_dir,tmp);
		if (*page_table & PAGE_PRESENT) {
			page_table = (ulong *) (PAGE_MASK & *page_table);
			page_table += ((tmp >> PAGE_SHIFT) & (PTRS_PER_PAGE-1));
			if (*page_table) {
				if (!remap)
					return -EINVAL;
				if (*page_table & PAGE_PRESENT) {
					--current->rss;
					free_page (*page_table & PAGE_MASK);
				}
				else
					swap_free (*page_table);
				invalid++;
			}
			continue;
		}  
	      {
		unsigned long new_pt;
		if(!(new_pt = get_free_page(GFP_KERNEL)))	/* clearing needed?  SRB. */
			return -ENOMEM;
		*page_table = new_pt | PAGE_TABLE;
		tmp |= ((PAGE_SIZE << 10) - PAGE_SIZE);
	}}
	if (invalid)
		invalidate();

	/* map page range */
	shm_sgn = shmd->shm_sgn;
	for (tmp = shmd->start; tmp < shmd->end; tmp += PAGE_SIZE, 
	     shm_sgn += (1 << SHM_IDX_SHIFT)) {