Re: [PATCH: eCryptfs] Encrypt on writepage()

From: Michael Halcrow
Date: Mon Nov 07 2005 - 15:43:45 EST

On Wed, Nov 02, 2005 at 11:32:58PM -0600, Michael Halcrow wrote:
> On Wed, Nov 02, 2005 at 08:55:30PM -0700, Phillip Hellewell wrote:
> > +static int ecryptfs_writepage(struct page *page, struct writeback_control *wbc)
> > +{
> > ...
> > + ecryptfs_printk(1, KERN_NOTICE, "Copying page\n");
> > + memcpy(lower_kaddr, kaddr, crypt_stats->extent_size);
>
> This part will, of course, need to be replaced with the same crypto
> operations being done in ecryptfs_commit_write() in order to encrypt
> the data on its way out via ecryptfs_writepage().

This patch encrypts on the writepage() execution path. It also removes
the initialization vector read/rotate/write code into its own
function, which simplifies commit_write().

I also took the opportunity to remove an unused macro from the
header.

This applies on top of the 12 original patches sent out, plus the
de-debug patch in the message with the subject ``Re: [PATCH: eCryptfs]
Remove debug wrappers''.

Signed-off-by: Michael Halcrow <mhalcrow@xxxxxxxxxx>
Signed-off-by: Phillip Hellewell <phillip@xxxxxxxxxxxxxxxxxxxx>
Signed-off-by: Michael Thompson <mcthomps@xxxxxxxxxx>
Signed-off-by: Kent Yoder <yoder1@xxxxxxxxxx>

ecryptfs_kernel.h | 3
mmap.c | 344 +++++++++++++++++++++++++++++++-----------------------
2 files changed, 202 insertions(+), 145 deletions(-)
Index: linux-2.6.14-rc5-ecryptfs/fs/ecryptfs/mmap.c
===================================================================
--- linux-2.6.14-rc5-ecryptfs.orig/fs/ecryptfs/mmap.c 2005-11-03 16:12:46.000000000 -0600
+++ linux-2.6.14-rc5-ecryptfs/fs/ecryptfs/mmap.c 2005-11-07 13:44:54.000000000 -0600
@@ -30,6 +30,8 @@
#include <linux/pagemap.h>
#include <linux/writeback.h>
#include <linux/page-flags.h>
+#include <linux/mount.h>
+#include <linux/file.h>
#include <linux/crypto.h>
#include <asm/scatterlist.h>
#include "ecryptfs_kernel.h"
@@ -318,90 +320,224 @@
}

/**
- * @param page Page that is locked before this call is made
+ * @param lower_file Can be NULL
+ * @return Zero on success
+ */
+static int
+ecryptfs_read_rotate_write_iv(char *iv, struct inode *inode,
+ int lower_iv_idx, struct file *lower_file,
+ struct page *page)
+{
+ int rc = 0;
+ pgoff_t records_page_index;
+ struct ecryptfs_crypt_stats *crypt_stats;
+ struct page *records_page;
+ char *records_virt;
+ int lower_file_needs_fput = 0;
+ struct address_space_operations *lower_a_ops;
+ struct inode *lower_inode;
+
+ crypt_stats = &(INODE_TO_PRIVATE(inode)->crypt_stats);
+ lower_inode = INODE_TO_LOWER(inode);
+ lower_a_ops = lower_inode->i_mapping->a_ops;
+ records_page_index = LAST_RECORDS_PAGE_IDX(crypt_stats, page->index);
+ ecryptfs_printk(1, KERN_NOTICE, "records_page_index = [%lu]\n",
+ records_page_index);
+ records_page = grab_cache_page(lower_inode->i_mapping,
+ records_page_index);
+ if (!records_page) {
+ ecryptfs_printk(0, KERN_ERR, "records_page == NULL "
+ "after grab_cache_page at index [%lu]"
+ "\n", records_page_index);
+ rc = -EIO;
+ goto out;
+ }
+ /* TODO: Assume encrypted only for version 0.1 */
+ ecryptfs_printk(1, KERN_NOTICE, "lower_iv_idx = [%d]\n", lower_iv_idx);
+ records_virt = kmap(records_page);
+ if (!records_virt) {
+ rc = -ENOMEM;
+ ecryptfs_printk(1, KERN_NOTICE, "Error in kmap\n");
+ goto out_unlock_and_release;
+ }
+ if (!lower_file) {
+ struct dentry *lower_dentry;
+ struct vfsmount *lower_mnt;
+
+ if (!lower_inode->i_dentry.next) {
+ rc = -EINVAL;
+ ecryptfs_printk(1, KERN_NOTICE, "No dentry for "
+ "lower_inode\n");
+ goto out_unmap;
+ }
+ lower_dentry = list_entry(lower_inode->i_dentry.next,
+ struct dentry, d_alias);
+ mntget(SUPERBLOCK_TO_PRIVATE(inode->i_sb)->lower_mnt);
+ lower_mnt = SUPERBLOCK_TO_PRIVATE(inode->i_sb)->lower_mnt;
+ lower_file = dentry_open(lower_dentry, lower_mnt, FMODE_READ);
+ if (IS_ERR(lower_file)) {
+ rc = PTR_ERR(lower_file);
+ ecryptfs_printk(0, KERN_ERR,
+ "Error opening dentry; rc = [%i]\n",
+ rc);
+ mntput(SUPERBLOCK_TO_PRIVATE(inode->i_sb)->lower_mnt);
+ goto out_unmap;
+ }
+ lower_file_needs_fput = 1;
+ }
+ rc = lower_a_ops->prepare_write(lower_file, records_page, lower_iv_idx,
+ (lower_iv_idx + crypt_stats->iv_bytes));
+ if (rc) {
+ ecryptfs_printk(0, KERN_ERR, "Error in lower prepare_write() "
+ "call: rc = [%d]\n", rc);
+ goto out_unmap;
+ }
+ down(&crypt_stats->iv_sem);
+ ecryptfs_rotate_iv(crypt_stats->iv);
+ memcpy(iv, crypt_stats->iv, crypt_stats->iv_bytes);
+ up(&crypt_stats->iv_sem);
+ memcpy(records_virt + lower_iv_idx, crypt_stats->iv,
+ crypt_stats->iv_bytes);
+ rc = lower_a_ops->commit_write(lower_file, records_page, lower_iv_idx,
+ (lower_iv_idx + crypt_stats->iv_bytes));
+ if (rc) {
+ ecryptfs_printk(0, KERN_ERR, "Error in lower commit_write() "
+ "call: rc = [%d]\n", rc);
+ goto out_unmap;
+ }
+ lower_inode->i_mtime = lower_inode->i_ctime = CURRENT_TIME;
+ ecryptfs_printk(1, KERN_NOTICE, "Unlocking page with index = "
+ "[%lu]\n", records_page->index);
+out_unmap:
+ kunmap(records_page);
+out_unlock_and_release:
+ unlock_page(records_page);
+ page_cache_release(records_page);
+ ecryptfs_printk(1, KERN_NOTICE, "After committing IV write, "
+ "lower_inode->i_blocks = [%lu]\n",
+ lower_inode->i_blocks);
+out:
+ if (lower_file_needs_fput)
+ fput(lower_file);
+ return rc;
+}
+
+/**
+ * @return Zero on success; negative on error
+ */
+static int encrypt_page(struct ecryptfs_crypt_stats *crypt_stats,
+ struct page *page, struct page *lower_page, char *iv)
+{
+ int rc = 0;
+ ecryptfs_printk(1, KERN_NOTICE, "Calling do_encrypt_page()\n");
+ ecryptfs_printk(1, KERN_NOTICE, "Encrypting page with iv:\n");
+ if (unlikely(ecryptfs_verbosity > 0)) {
+ ecryptfs_dump_hex(iv, crypt_stats->iv_bytes);
+ }
+ ecryptfs_printk(1, KERN_NOTICE, "First 8 bytes before "
+ "encryption:\n");
+ if (ecryptfs_verbosity > 0) {
+ ecryptfs_dump_hex((char *)page_address(page), 8);
+ }
+ rc = do_encrypt_page_offset(crypt_stats, lower_page, 0,
+ page, 0, crypt_stats->extent_size, iv);
+ ecryptfs_printk(1, KERN_NOTICE, "Encrypted [%d] bytes\n", rc);
+ ecryptfs_printk(1, KERN_NOTICE, "First 8 bytes after " "encryption:\n");
+ if (ecryptfs_verbosity > 0) {
+ ecryptfs_dump_hex((char *)page_address(lower_page), 8);
+ }
+ if (rc > 0)
+ rc = 0;
+ return rc;
+}
+
+/**
+ * @param page Page that is locked before this call is made
*/
static int ecryptfs_writepage(struct page *page, struct writeback_control *wbc)
{
- int err = -EIO;
+ int rc = -EIO;
+ int err = 0;
unsigned long lower_index;
struct inode *inode;
struct inode *lower_inode;
struct page *lower_page;
char *kaddr, *lower_kaddr;
struct ecryptfs_crypt_stats *crypt_stats;
+
ecryptfs_printk(1, KERN_NOTICE, "Enter; page->index = [%ld]; "
"page->mapping->host = [%p]\n", page->index,
page->mapping->host);
-
- /* Sanity checkers */
- if (1) {
- struct address_space *mapping = page->mapping;
- loff_t offset = (loff_t) page->index << PAGE_CACHE_SHIFT;
- if (wbc->sync_mode != WB_SYNC_NONE) {
- ecryptfs_printk(1, KERN_NOTICE,
- "wbc->sync_mode != WB_SYNC_NONE\n");
- } else {
- ecryptfs_printk(1, KERN_NOTICE,
- "wbc->sync_mode == WB_SYNC_NONE\n");
- }
- /* racing with truncate? */
- if (offset > mapping->host->i_size) {
- ecryptfs_printk(1, KERN_NOTICE, "Racing truncate\n");
- }
- BUG_ON(PageWriteback(page));
- }
-
- /* End sanity */
-
inode = page->mapping->host;
crypt_stats = &(INODE_TO_PRIVATE(inode)->crypt_stats);
lower_inode = INODE_TO_LOWER(inode);
- /* The 2nd argument here should be the index of the lower file, which
- * should be the calculated result of our index interpolation */
lower_index = PG_IDX_TO_LWR_PG_IDX(crypt_stats, page->index);
ecryptfs_printk(1, KERN_NOTICE, "Grab lower_idx = [%ld]\n",
lower_index);
lower_page = grab_cache_page(lower_inode->i_mapping, lower_index);
if (!lower_page) {
- err = -ENOMEM;
+ rc = -ENOMEM;
goto out;
}
- if (PageWriteback(lower_page)) {
- ecryptfs_printk(1, KERN_NOTICE,
- "lower_page is under writeback\n");
- } else {
- ecryptfs_printk(1, KERN_NOTICE,
- "lower_page is avail for writeback\n");
- }
- if (!PageLocked(page)) {
- ecryptfs_printk(1, KERN_NOTICE, "Page not locked\n");
- } else {
- ecryptfs_printk(1, KERN_NOTICE, "Page is already locked\n");
- }
kaddr = (char *)kmap(page);
+ if (!kaddr) {
+ rc = -ENOMEM;
+ goto out;
+ }
lower_kaddr = (char *)kmap(lower_page);
- ecryptfs_printk(1, KERN_NOTICE, "Copying page\n");
- memcpy(lower_kaddr, kaddr, crypt_stats->extent_size);
+ if (!lower_kaddr) {
+ rc = -ENOMEM;
+ goto out;
+ }
+ if (crypt_stats->encrypted) {
+ char iv[ECRYPTFS_MAX_IV_BYTES];
+ int record_byte_offset;
+ /* TODO: HMAC: Include HMAC bytes in the record size */
+ record_byte_offset = (RECORD_IDX(crypt_stats, page->index)
+ * crypt_stats->iv_bytes);
+ rc = ecryptfs_read_rotate_write_iv(iv, inode,
+ record_byte_offset, NULL,
+ page);
+ if (rc) {
+ ecryptfs_printk(0, KERN_ERR, "Error rotating "
+ "IV; write failure. Assuming "
+ "IV page corruption; writing "
+ "0's to associated data extent"
+ ".\n");
+ memset(lower_kaddr, 0, crypt_stats->extent_size);
+ err = -EIO;
+ goto do_lower_write;
+ }
+ ecryptfs_printk(1, KERN_NOTICE, "Encrypting page with iv:\n");
+ if (unlikely(ecryptfs_verbosity > 0))
+ ecryptfs_dump_hex(iv, crypt_stats->iv_bytes);
+ err = encrypt_page(crypt_stats, page, lower_page, iv);
+ if (err)
+ ecryptfs_printk(0, KERN_WARNING, "Error encrypting "
+ "page (upper index [%llu])\n",
+ page->index);
+ } else
+ memcpy(lower_kaddr, kaddr, crypt_stats->extent_size);
+do_lower_write:
kunmap(page);
kunmap(lower_page);
- if (PageWriteback(lower_page)) {
- ecryptfs_printk(1, KERN_NOTICE,
- "lower_page is under writeback\n");
- } else {
- ecryptfs_printk(1, KERN_NOTICE,
- "lower_page is avail for writeback\n");
+ rc = lower_inode->i_mapping->a_ops->writepage(lower_page, wbc);
+ if (rc) {
+ ecryptfs_printk(0, KERN_ERR, "Error calling lower writepage(); "
+ "rc = [%d]\n", rc);
}
- err = lower_inode->i_mapping->a_ops->writepage(lower_page, wbc);
lower_inode->i_mtime = lower_inode->i_ctime = CURRENT_TIME;
page_cache_release(lower_page);
- if (err)
+ if (rc)
ClearPageUptodate(page);
else
SetPageUptodate(page);
unlock_page(page);
out:
- ecryptfs_printk(1, KERN_NOTICE, "Exit; err = [%d]\n", err);
- return err;
+ if (err)
+ rc = err;
+ ecryptfs_printk(1, KERN_NOTICE, "Exit; rc = [%d]\n", rc);
+ return rc;
}

/**
@@ -574,7 +710,6 @@
int record_byte_offset;
int iv_byte_offset;
int i;
- int record_size;
pgoff_t records_page_index;
pgoff_t lower_page_index;

@@ -590,7 +725,6 @@
rc = ecryptfs_do_readpage(file, page, page->index);
goto out;
}
- record_size = crypt_stats->iv_bytes;
/* The file is encrypted, hmac verified, or both. */
ecryptfs_printk(1, KERN_NOTICE,
"crypt_stats->iv_bytes = [%d]\n",
@@ -617,8 +751,9 @@
ClearPageUptodate(page);
goto out;
}
-
- record_byte_offset = RECORD_IDX(crypt_stats, page->index) * record_size;
+ /* TODO: HMAC: Include HMAC bytes in the record size */
+ record_byte_offset = (RECORD_IDX(crypt_stats, page->index)
+ * crypt_stats->iv_bytes);
iv_byte_offset = -1;
if (crypt_stats->encrypted) {
iv_byte_offset = record_byte_offset;
@@ -721,37 +856,6 @@
}

/**
- * @return Zero on success; negative on error
- */
-static int encrypt_page(struct ecryptfs_crypt_stats *crypt_stats,
- struct page *page, struct page *lower_page, char *iv)
-{
- int rc = 0;
- ecryptfs_printk(1, KERN_NOTICE, "Calling do_encrypt_page()\n");
- if (iv) {
- ecryptfs_printk(1, KERN_NOTICE, "Encrypting page with iv:\n");
- if (ecryptfs_verbosity > 0) {
- ecryptfs_dump_hex(iv, crypt_stats->iv_bytes);
- }
- }
- ecryptfs_printk(1, KERN_NOTICE, "First 8 bytes before "
- "encryption:\n");
- if (ecryptfs_verbosity > 0) {
- ecryptfs_dump_hex((char *)page_address(page), 8);
- }
- rc = do_encrypt_page_offset(crypt_stats, lower_page, 0,
- page, 0, crypt_stats->extent_size, iv);
- ecryptfs_printk(1, KERN_NOTICE, "Encrypted [%d] bytes\n", rc);
- ecryptfs_printk(1, KERN_NOTICE, "First 8 bytes after " "encryption:\n");
- if (ecryptfs_verbosity > 0) {
- ecryptfs_dump_hex((char *)page_address(lower_page), 8);
- }
- if (rc > 0)
- rc = 0;
- return rc;
-}
-
-/**
* This is where we encrypt the data and pass the encrypted data to
* the lower filesystem. In OpenPGP-compatible mode, we operate on
* entire underlying packets.
@@ -774,9 +878,9 @@
unsigned bytes = to - from;
struct ecryptfs_crypt_stats *crypt_stats;
pgoff_t lower_page_index;
- struct page *records_page;
struct address_space_operations *lower_a_ops;
struct dentry *ecryptfs_dentry;
+
ecryptfs_printk(1, KERN_NOTICE,
"Enter; page->index = [%lu]; from = [%d]; to = [%d]\n",
page->index, from, to);
@@ -875,73 +979,29 @@
(char *)page_address(page), crypt_stats->extent_size);
} else {
/* The file is either encrypted or HMAC'd */
- pgoff_t records_page_index;
- int record_byte_offset = -1;
- int iv_byte_offset = -1;
- char *records_virt;
char iv[ECRYPTFS_MAX_IV_BYTES];
- int record_size;
+ int record_byte_offset;
ecryptfs_printk(1, KERN_NOTICE,
"crypt_stats->iv_bytes = [%d]\n",
crypt_stats->iv_bytes);
- ecryptfs_printk(1, KERN_NOTICE,
- "crypt_stats->records_per_page = [%d]\n",
- crypt_stats->records_per_page);
- record_size = crypt_stats->iv_bytes;
- records_page_index =
- LAST_RECORDS_PAGE_IDX(crypt_stats, page->index);
- ecryptfs_printk(1, KERN_NOTICE, "records_page_index = [%lu]\n",
- records_page_index);
- records_page = grab_cache_page(lower_inode->i_mapping,
- records_page_index);
- if (!records_page) {
- ecryptfs_printk(0, KERN_ERR, "records_page == NULL "
- "after grab_cache_page at index [%lu]"
- "\n", records_page_index);
- rc = -EIO;
+ /* TODO: HMAC: Include HMAC bytes in the record size */
+ record_byte_offset = (RECORD_IDX(crypt_stats, page->index)
+ * crypt_stats->iv_bytes);
+ rc = ecryptfs_read_rotate_write_iv(iv, inode,
+ record_byte_offset,
+ lower_file, page);
+ if (rc) {
+ ecryptfs_printk(0, KERN_ERR, "Error rotating IV\n");
goto out_unlock_lower;
}
- iv_byte_offset = -1;
- record_byte_offset =
- RECORD_IDX(crypt_stats, page->index) * record_size;
- if (crypt_stats->encrypted) {
- iv_byte_offset = record_byte_offset;
- }
- ecryptfs_printk(1, KERN_NOTICE, "iv_byte_offset = [%d]\n",
- iv_byte_offset);
- records_virt = kmap(records_page);
- rc = lower_a_ops->prepare_write(lower_file, records_page,
- iv_byte_offset,
- iv_byte_offset
- + crypt_stats->iv_bytes);
- down(&crypt_stats->iv_sem);
- ecryptfs_rotate_iv(crypt_stats->iv);
- memcpy(iv, crypt_stats->iv, crypt_stats->iv_bytes);
- up(&crypt_stats->iv_sem);
- memcpy(records_virt + iv_byte_offset, crypt_stats->iv,
- crypt_stats->iv_bytes);
- rc = lower_a_ops->commit_write(lower_file, records_page,
- iv_byte_offset,
- iv_byte_offset +
- crypt_stats->iv_bytes);
- kunmap(records_page);
- lower_inode->i_mtime = lower_inode->i_ctime = CURRENT_TIME;
- mark_inode_dirty_sync(inode);
- ecryptfs_printk(1, KERN_NOTICE, "Unlocking page with index = "
- "[%lu]\n", records_page->index);
- unlock_page(records_page);
- page_cache_release(records_page);
- ecryptfs_printk(1, KERN_NOTICE, "After committing IV write, "
- "lower_inode->i_blocks = [%lu]\n",
- lower_inode->i_blocks);
ecryptfs_printk(1, KERN_NOTICE, "Encrypting page with iv:\n");
- if (ecryptfs_verbosity > 0) {
+ if (unlikely(ecryptfs_verbosity > 0))
ecryptfs_dump_hex(iv, crypt_stats->iv_bytes);
- }
rc = encrypt_page(crypt_stats, page, lower_page, iv);
if (rc) {
ecryptfs_printk(0, KERN_WARNING, "Error encrypting "
- "page\n");
+ "page (upper index [%llu])\n",
+ page->index);
goto out;
}
}
Index: linux-2.6.14-rc5-ecryptfs/fs/ecryptfs/ecryptfs_kernel.h
===================================================================
--- linux-2.6.14-rc5-ecryptfs.orig/fs/ecryptfs/ecryptfs_kernel.h 2005-11-03 15:36:39.000000000 -0600
+++ linux-2.6.14-rc5-ecryptfs/fs/ecryptfs/ecryptfs_kernel.h 2005-11-07 13:37:31.000000000 -0600
@@ -173,9 +173,6 @@
((idx / crypt_stats->records_per_page) + idx \
+ crypt_stats->num_header_pages + 1)
#define RECORD_IDX(crypt_stats, idx) (idx % crypt_stats->records_per_page)
-#define RECORD_OFFSET(crypt_stats, idx) \
- (RECORD_IDX(crypt_stats, idx) * (crypt_stats->iv_bytes \
- + crypt_stats->hmac_bytes))
#define LAST_RECORDS_PAGE_IDX(crypt_stats, idx) \
(PG_IDX_TO_LWR_PG_IDX(crypt_stats, idx) \
- RECORD_IDX(crypt_stats,idx) - 1)
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/