This is the first attempt at this specific task so please be gentle.
The thing I am unhappy about is the extra vmalloc/copy in read_kmem().
Ultimately, it is not a problem because /dev/kmem access is not
performance critical whilst the use of vmalloc/vfree/ioremap (esp in
drivers) may be very critical. But if you see how my patch can be
improved, please let me know before I send it to Linus.
A copy of the patch is on:
http://www.ocston.org/~tigran/patches/vmalloc-2.3.34.patch
regards,
Tigran.
PS. Btw, the patch also contains the page->index doc in <linux/mm.h> and
optimized sys_sync/fsync which I was going to send to Linus
anyway - hope it is not too much strain on your eyes...
diff -urN -X dontdiff linux/drivers/char/mem.c work/drivers/char/mem.c
--- linux/drivers/char/mem.c Thu Dec 16 21:59:38 1999
+++ work/drivers/char/mem.c Tue Dec 21 09:10:25 1999
@@ -227,6 +227,7 @@
unsigned long p = *ppos;
ssize_t read = 0;
ssize_t virtr;
+ char * kbuf; /* k-addr because vread() takes vmlist_lock spinlock */
if (p < (unsigned long) high_memory) {
read = count;
@@ -238,22 +239,34 @@
if (p < PAGE_SIZE && read > 0) {
size_t tmp = PAGE_SIZE - p;
if (tmp > read) tmp = read;
- clear_user(buf, tmp);
+ if (clear_user(buf, tmp))
+ return -EFAULT;
buf += tmp;
p += tmp;
read -= tmp;
count -= tmp;
}
#endif
- copy_to_user(buf, (char *)p, read);
+ if(copy_to_user(buf, (char *)p, read))
+ return -EFAULT;
p += read;
buf += read;
count -= read;
}
- virtr = vread(buf, (char *)p, count);
- if (virtr < 0)
+ kbuf = vmalloc(count);
+ if (kbuf == NULL)
+ return -ENOMEM;
+ virtr = vread(kbuf, (char *)p, count);
+ if (virtr < 0) {
+ vfree(kbuf);
return virtr;
+ }
+ if (copy_to_user(buf, kbuf, virtr)) {
+ vfree(kbuf);
+ return -EFAULT;
+ }
+ vfree(kbuf);
*ppos += p + virtr;
return virtr + read;
}
diff -urN -X dontdiff linux/fs/buffer.c work/fs/buffer.c
--- linux/fs/buffer.c Mon Dec 20 22:52:57 1999
+++ work/fs/buffer.c Tue Dec 21 08:14:50 1999
@@ -345,7 +345,6 @@
struct inode * inode;
int err;
- lock_kernel();
err = -EBADF;
file = fget(fd);
if (!file)
@@ -365,13 +364,14 @@
/* We need to protect against concurrent writers.. */
down(&inode->i_sem);
+ lock_kernel();
err = file->f_op->fsync(file, dentry);
+ unlock_kernel();
up(&inode->i_sem);
out_putf:
fput(file);
out:
- unlock_kernel();
return err;
}
@@ -382,7 +382,6 @@
struct inode * inode;
int err;
- lock_kernel();
err = -EBADF;
file = fget(fd);
if (!file)
@@ -402,13 +401,14 @@
/* this needs further work, at the moment it is identical to fsync() */
down(&inode->i_sem);
+ lock_kernel();
err = file->f_op->fsync(file, dentry);
+ unlock_kernel();
up(&inode->i_sem);
out_putf:
fput(file);
out:
- unlock_kernel();
return err;
}
diff -urN -X dontdiff linux/include/linux/mm.h work/include/linux/mm.h
--- linux/include/linux/mm.h Tue Dec 21 00:01:14 1999
+++ work/include/linux/mm.h Tue Dec 21 08:13:57 1999
@@ -244,8 +244,14 @@
* The following discussion applies only to them.
*
* A page may belong to an inode's memory mapping. In this case,
- * page->inode is the pointer to the inode, and page->offset is the
- * file offset of the page (not necessarily a multiple of PAGE_SIZE).
+ * page->inode is the pointer to the inode, and page->index is the
+ * file offset of the page in PAGE_CACHE_SIZE (not PAGE_SIZE!) units.
+ * Although currently (2.3.34) PAGE_SIZE == PAGE_CACHE_SIZE, i.e. there
+ * happens to be one page per page cache entry and MM code can't hanlde
+ * anything else, this may well change. The link to the old page->offset
+ * is given by:
+ *
+ * page->index == (page->offset >> PAGE_CACHE_SHIFT);
*
* A page may have buffers allocated to it. In this case,
* page->buffers is a circular list of these buffer heads. Else,
diff -urN -X dontdiff linux/ipc/util.c work/ipc/util.c
--- linux/ipc/util.c Tue Dec 14 18:01:21 1999
+++ work/ipc/util.c Tue Dec 21 09:15:44 1999
@@ -161,13 +161,10 @@
void* ipc_alloc(int size)
{
void* out;
- if(size > PAGE_SIZE) {
- lock_kernel();
+ if(size > PAGE_SIZE)
out = vmalloc(size);
- unlock_kernel();
- } else {
+ else
out = kmalloc(size, GFP_KERNEL);
- }
return out;
}
diff -urN -X dontdiff linux/mm/vmalloc.c work/mm/vmalloc.c
--- linux/mm/vmalloc.c Sat Nov 20 18:09:05 1999
+++ work/mm/vmalloc.c Tue Dec 21 09:12:12 1999
@@ -7,10 +7,12 @@
#include <linux/malloc.h>
#include <linux/vmalloc.h>
+#include <linux/spinlock.h>
#include <asm/uaccess.h>
#include <asm/pgalloc.h>
+spinlock_t vmlist_lock = SPIN_LOCK_UNLOCKED;
struct vm_struct * vmlist = NULL;
static inline void free_area_pte(pmd_t * pmd, unsigned long address, unsigned long size)
@@ -162,11 +164,13 @@
if (!area)
return NULL;
addr = VMALLOC_START;
+ spin_lock(&vmlist_lock);
for (p = &vmlist; (tmp = *p) ; p = &tmp->next) {
if (size + addr < (unsigned long) tmp->addr)
break;
addr = tmp->size + (unsigned long) tmp->addr;
if (addr > VMALLOC_END-size) {
+ spin_unlock(&vmlist_lock);
kfree(area);
return NULL;
}
@@ -176,6 +180,7 @@
area->size = size + PAGE_SIZE;
area->next = *p;
*p = area;
+ spin_unlock(&vmlist_lock);
return area;
}
@@ -189,14 +194,17 @@
printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
return;
}
+ spin_lock(&vmlist_lock);
for (p = &vmlist ; (tmp = *p) ; p = &tmp->next) {
if (tmp->addr == addr) {
*p = tmp->next;
vmfree_area_pages(VMALLOC_VMADDR(tmp->addr), tmp->size);
+ spin_unlock(&vmlist_lock);
kfree(tmp);
return;
}
}
+ spin_unlock(&vmlist_lock);
printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", addr);
}
@@ -224,16 +232,17 @@
return addr;
}
-long vread(char *buf, char *addr, unsigned long count)
+long vread(char *kbuf, char *addr, unsigned long count)
{
struct vm_struct *tmp;
- char *vaddr, *buf_start = buf;
+ char *vaddr, *buf_start = kbuf;
unsigned long n;
/* Don't allow overflow */
if ((unsigned long) addr + count < count)
count = -(unsigned long) addr;
+ spin_lock(&vmlist_lock);
for (tmp = vmlist; tmp; tmp = tmp->next) {
vaddr = (char *) tmp->addr;
if (addr >= vaddr + tmp->size - PAGE_SIZE)
@@ -241,8 +250,8 @@
while (addr < vaddr) {
if (count == 0)
goto finished;
- put_user('\0', buf);
- buf++;
+ *kbuf = '\0';
+ kbuf++;
addr++;
count--;
}
@@ -250,12 +259,13 @@
do {
if (count == 0)
goto finished;
- put_user(*addr, buf);
- buf++;
+ *kbuf = *addr;
+ kbuf++;
addr++;
count--;
} while (--n > 0);
}
finished:
- return buf - buf_start;
+ spin_unlock(&vmlist_lock);
+ return kbuf - buf_start;
}
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/