From: Daniel Thompson <daniel.thompson@linaro.org>
Date: Mon, 27 Oct 2014 18:51:43 +0000 (+0000)
Subject: staging: android: logger: Fix log corruption regression
X-Git-Tag: v3.18-rc3~6^2~8
X-Git-Url: https://openfabrics.org/gitweb/?a=commitdiff_plain;h=5fecf3a1e1a0af61eb34eb6976ec9f59cca65d3f;p=~emulex%2Finfiniband.git

staging: android: logger: Fix log corruption regression

Since commit cd678fce4280 ("switch logger to ->write_iter()"), any
attempt to write to the log results in the log data being written over
its own metadata, thus rendering the log unreadable.

The problem was first detected when I ran an Android userspace on the
v3.18-rc1 kernel. However the issue can also be observed with a
non-Android userspace by using echo/cat to write to/from /dev/log_main .

This patch resolves the problem by using a temporary to track the status
of not-yet-committed writes to the log buffer.

Signed-off-by: Daniel Thompson <daniel.thompson@linaro.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---

diff --git a/drivers/staging/android/logger.c b/drivers/staging/android/logger.c
index 28b93d39a94..a673ffa34aa 100644
--- a/drivers/staging/android/logger.c
+++ b/drivers/staging/android/logger.c
@@ -420,7 +420,7 @@ static ssize_t logger_write_iter(struct kiocb *iocb, struct iov_iter *from)
 	struct logger_log *log = file_get_log(iocb->ki_filp);
 	struct logger_entry header;
 	struct timespec now;
-	size_t len, count;
+	size_t len, count, w_off;
 
 	count = min_t(size_t, iocb->ki_nbytes, LOGGER_ENTRY_MAX_PAYLOAD);
 
@@ -452,11 +452,14 @@ static ssize_t logger_write_iter(struct kiocb *iocb, struct iov_iter *from)
 	memcpy(log->buffer + log->w_off, &header, len);
 	memcpy(log->buffer, (char *)&header + len, sizeof(header) - len);
 
-	len = min(count, log->size - log->w_off);
+	/* Work with a copy until we are ready to commit the whole entry */
+	w_off =  logger_offset(log, log->w_off + sizeof(struct logger_entry));
 
-	if (copy_from_iter(log->buffer + log->w_off, len, from) != len) {
+	len = min(count, log->size - w_off);
+
+	if (copy_from_iter(log->buffer + w_off, len, from) != len) {
 		/*
-		 * Note that by not updating w_off, this abandons the
+		 * Note that by not updating log->w_off, this abandons the
 		 * portion of the new entry that *was* successfully
 		 * copied, just above.  This is intentional to avoid
 		 * message corruption from missing fragments.
@@ -470,7 +473,7 @@ static ssize_t logger_write_iter(struct kiocb *iocb, struct iov_iter *from)
 		return -EFAULT;
 	}
 
-	log->w_off = logger_offset(log, log->w_off + count);
+	log->w_off = logger_offset(log, w_off + count);
 	mutex_unlock(&log->mutex);
 
 	/* wake up any blocked readers */