Patch(2.5.50): Simplify crypto memory allocation

Patch(2.5.50): Simplify crypto memory allocation

[Adam J. Richter]

[-- Attachment #1: Type: text/plain, Size: 835 bytes --]

	The following patch deletes the unused
crypto_tfm.crt_work_block field and combines the remaining
two kmallocs done by crypto_alloc_tfm into one, a net
deletion of 25 lines.

	I've only verified that the kernel and the crpypto modules
still build.  I don't currently use this code, although I'm
considering making a version of loop.c which would, which is why I
noticed this.

	Anyhow, if this patch turns out to work and looks OK, then
please integrate, queue it for Linus, etc., or let me know if you
would prefer that you or I follow some other course of action.

-- 
Adam J. Richter     __     ______________   575 Oroville Road
adam@yggdrasil.com     \ /                  Milpitas, California 95035
+1 408 309-6081         | g g d r a s i l   United States of America
                         "Free Software For The Rest Of Us."

[-- Attachment #2: crypt.diff --]
[-- Type: text/plain, Size: 1747 bytes --]

--- linux-2.5.50/include/linux/crypto.h	2002-11-27 14:35:54.000000000 -0800
+++ linux/include/linux/crypto.h	2002-12-08 01:06:24.000000000 -0800
@@ -161,7 +161,6 @@
 struct crypto_tfm {
 
 	void *crt_ctx;
-	void *crt_work_block;
 	u32 crt_flags;
 	
 	union {
--- linux-2.5.50/crypto/api.c	2002-11-27 14:36:21.000000000 -0800
+++ linux/crypto/api.c	2002-12-08 01:06:35.000000000 -0800
@@ -124,44 +124,26 @@
 	if (alg == NULL)
 		goto out;
 	
-	tfm = kmalloc(sizeof(*tfm), GFP_KERNEL);
+	tfm = kmalloc(sizeof(*tfm) + alg->cra_ctxsize, GFP_KERNEL);
 	if (tfm == NULL)
 		goto out_put;
 
 	memset(tfm, 0, sizeof(*tfm));
 	
-	if (alg->cra_ctxsize) {
-		tfm->crt_ctx = kmalloc(alg->cra_ctxsize, GFP_KERNEL);
-		if (tfm->crt_ctx == NULL)
-			goto out_free_tfm;
-	}
+	tfm->crt_ctx = (void*) &tfm[1];
 
 	tfm->__crt_alg = alg;
 	
-	if (alg->cra_blocksize) {
-		tfm->crt_work_block = kmalloc(alg->cra_blocksize + 1,
-					      GFP_KERNEL);
-		if (tfm->crt_work_block == NULL)
-			goto out_free_ctx;
-	}
-
 	if (crypto_init_flags(tfm, flags))
-		goto out_free_work_block;
+		goto out_free_tfm;
 		
 	if (crypto_init_ops(tfm)) {
 		crypto_exit_ops(tfm);
-		goto out_free_ctx;
+		goto out_free_tfm;
 	}
 
 	goto out;
 
-out_free_work_block:
-	if (tfm->__crt_alg->cra_blocksize)
-		kfree(tfm->crt_work_block);
-
-out_free_ctx:
-	if (tfm->__crt_alg->cra_ctxsize)
-		kfree(tfm->crt_ctx);
 out_free_tfm:
 	kfree(tfm);
 	tfm = NULL;
@@ -173,12 +155,6 @@
 
 void crypto_free_tfm(struct crypto_tfm *tfm)
 {
-	if (tfm->__crt_alg->cra_ctxsize)
-		kfree(tfm->crt_ctx);
-		
-	if (tfm->__crt_alg->cra_blocksize)
-		kfree(tfm->crt_work_block);
-		
 	if (crypto_tfm_alg_type(tfm) == CRYPTO_ALG_TYPE_CIPHER)
 		if (tfm->crt_cipher.cit_iv)
 			kfree(tfm->crt_cipher.cit_iv);

Re: Patch(2.5.50): Simplify crypto memory allocation

[James Morris]

On Sun, 8 Dec 2002, Adam J. Richter wrote:

> 	The following patch deletes the unused
> crypto_tfm.crt_work_block field and combines the remaining
> two kmallocs done by crypto_alloc_tfm into one, a net
> deletion of 25 lines.
> 
> 	I've only verified that the kernel and the crpypto modules
> still build.  I don't currently use this code, although I'm
> considering making a version of loop.c which would, which is why I
> noticed this.
> 
> 	Anyhow, if this patch turns out to work and looks OK, then
> please integrate, queue it for Linus, etc., or let me know if you
> would prefer that you or I follow some other course of action.
> 

Looks good and tests ok.  Thanks.  

(The work_block field and associated management code should have
disappeared long ago, not sure why it was still there).


- James
-- 
James Morris
<jmorris@intercode.com.au>

Re: Patch(2.5.50): Simplify crypto memory allocation

[David S. Miller]

   From: James Morris <jmorris@intercode.com.au>
   Date: Sun, 8 Dec 2002 21:55:15 +1100 (EST)

   On Sun, 8 Dec 2002, Adam J. Richter wrote:
   
   > 	Anyhow, if this patch turns out to work and looks OK, then
   > please integrate, queue it for Linus, etc., or let me know if you
   > would prefer that you or I follow some other course of action.
   
   Looks good and tests ok.  Thanks.  
   
I'll apply this, thanks Adam.

   (The work_block field and associated management code should have
   disappeared long ago, not sure why it was still there).

We just didn't do this part when the conversion was made :-)