Imagelibrary/rtems
1
0
Fork 1
mirror of https://gitlab.rtems.org/rtems/rtos/rtems.git synced 2025-12-25 13:57:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

cpukit/jffs2: Fix unaligned writes

Browse Source 
The code in flashio.c had a special case for handling unaligned buffers
during writes. That code did roughly the following:

If the size or start address of the buffer is not aligned to a word:
* align the size up to the next word
* create a temp buffer on heap or stack with the new size
* copy the data from write buffer to this temp buffer
* write the temp buffer with the _new_ size to the flash
* tell the code above, that the original size has been written

That means, that in certain cases, one to three random bytes have been
written to the flash. That is definitively not correct.

The special cases that trigger the behaviour seem to happen quite often
when using the RTEMS functions to unpack a .tar.gz with files in the
range of a few megabytes stored on the JFFS2 file system to the same
file system.

The RTEMS interface for flash drivers doesn't define any alignment
requirements. Therefore that code is not necessary and can just be
removed to solve these issues.
This commit is contained in:
Christian Mauderer
2025-09-17 13:33:19 +02:00
parent ba3acfd3d0
commit 1bb370b644
1 changed files with 3 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files

66
cpukit/libfs/src/jffs2/src/flashio.c
View File

@@ -65,74 +65,14 @@ jffs2_flash_direct_writev(struct jffs2_sb_info *c, const struct iovec *vecs,
int ret = 0;
for (i = 0; i < count; i++) {
// writes need to be aligned but the data we're passed may not be
// Observation suggests most unaligned writes are small, so we
// optimize for that case.
if (((vecs[i].iov_len & (sizeof (int) - 1))) ||
(((unsigned long) vecs[i].
iov_base & (sizeof (unsigned long) - 1)))) {
// are there iov's after this one? Or is it so much we'd need
// to do multiple writes anyway?
if ((i + 1) < count || vecs[i].iov_len > 256) {
// cop out and malloc
unsigned long j;
ssize_t sizetomalloc = 0, totvecsize = 0;
char *cbuf, *cbufptr;
for (j = i; j < count; j++)
totvecsize += vecs[j].iov_len;
// pad up in case unaligned
sizetomalloc = totvecsize + sizeof (int) - 1;
sizetomalloc &= ~(sizeof (int) - 1);
cbuf = (char *) malloc(sizetomalloc);
// malloc returns aligned memory
if (!cbuf) {
ret = -ENOMEM;
goto writev_out;
}
cbufptr = cbuf;
for (j = i; j < count; j++) {
memcpy(cbufptr, vecs[j].iov_base,
vecs[j].iov_len);
cbufptr += vecs[j].iov_len;
}
ret =
jffs2_flash_write(c, to, sizetomalloc,
&thislen, cbuf);
if (thislen > totvecsize) // in case it was aligned up
thislen = totvecsize;
totlen += thislen;
free(cbuf);
goto writev_out;
} else {
// otherwise optimize for the common case
int buf[256 / sizeof (int)]; // int, so int aligned
size_t lentowrite;
lentowrite = vecs[i].iov_len;
// pad up in case its unaligned
lentowrite += sizeof (int) - 1;
lentowrite &= ~(sizeof (int) - 1);
memcpy(buf, vecs[i].iov_base, lentowrite);
ret =
jffs2_flash_write(c, to, lentowrite,
&thislen, (char *) &buf);
if (thislen > vecs[i].iov_len)
thislen = vecs[i].iov_len;
} // else
} else
ret =
jffs2_flash_write(c, to, vecs[i].iov_len, &thislen,
vecs[i].iov_base);
ret = jffs2_flash_write(c, to, vecs[i].iov_len, &thislen,
vecs[i].iov_base);
totlen += thislen;
if (ret || thislen != vecs[i].iov_len)
break;
to += vecs[i].iov_len;
}
writev_out:
if (retlen)
*retlen = totlen;