[PATCH v3] staging: lustre: llog: fix wrong offset in llog_process_thread()

James Simmons jsimmons at infradead.org
Fri Nov 18 16:42:26 UTC 2016


> On Thu, Nov 17, 2016 at 06:29:08PM -0500, James Simmons wrote:
> > From: Mikhail Pershin <mike.pershin at intel.com>
> > 
> > - llh_cat_idx may become bigger than llog bitmap size in
> >   llog_cat_set_first_idx() function
> > - it is wrong to use previous cur_offset as new buffer offset,
> >   new offset should be calculated from value returned by
> >   llog_next_block().
> > - optimize llog_skip_over() to find llog entry offset by index
> >   for llog with fixed-size records.
> > 
> > Signed-off-by: Mikhail Pershin <mike.pershin at intel.com>
> > Signed-off-by: Bob Glossman <bob.glossman at intel.com>
> > Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-6714
> > Reviewed-on: http://review.whamcloud.com/15316
> > Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-6163
> > Reviewed-on: http://review.whamcloud.com/18819
> > Reviewed-by: John L. Hammond <john.hammond at intel.com>
> > Reviewed-by: James Simmons <uja.ornl at yahoo.com>
> > Reviewed-by: Oleg Drokin <oleg.drokin at intel.com>
> > Signed-off-by: James Simmons <jsimmons at infradead.org>
> > ---
> > 
> > ChangeLog:
> > 
> > v1) Initial patch with umoddi issue
> > v2) Included fix from patch LU-6163 that fixed umoddi problem
> > v3) Remove no longer needed last_offset variable
> > 
> >  drivers/staging/lustre/lustre/obdclass/llog.c |   82 +++++++++++++++++-------
> >  include/linux/fs.h                            |    2 +-
> >  2 files changed, 59 insertions(+), 25 deletions(-)
> > 
> > diff --git a/drivers/staging/lustre/lustre/obdclass/llog.c b/drivers/staging/lustre/lustre/obdclass/llog.c
> > index 3bc1789..ae63047 100644
> > --- a/drivers/staging/lustre/lustre/obdclass/llog.c
> > +++ b/drivers/staging/lustre/lustre/obdclass/llog.c
> > @@ -217,8 +217,7 @@ static int llog_process_thread(void *arg)
> >  	struct llog_log_hdr		*llh = loghandle->lgh_hdr;
> >  	struct llog_process_cat_data	*cd  = lpi->lpi_catdata;
> >  	char				*buf;
> > -	__u64				 cur_offset;
> > -	__u64				 last_offset;
> > +	u64 cur_offset, tmp_offset;
> >  	int chunk_size;
> >  	int				 rc = 0, index = 1, last_index;
> >  	int				 saved_index = 0;
> > @@ -229,6 +228,8 @@ static int llog_process_thread(void *arg)
> >  
> >  	cur_offset = llh->llh_hdr.lrh_len;
> >  	chunk_size = llh->llh_hdr.lrh_len;
> > +	/* expect chunk_size to be power of two */
> > +	LASSERT(is_power_of_2(chunk_size));
> >  
> >  	buf = libcfs_kvzalloc(chunk_size, GFP_NOFS);
> >  	if (!buf) {
> > @@ -245,38 +246,50 @@ static int llog_process_thread(void *arg)
> >  	else
> >  		last_index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
> >  
> > -	/* Record is not in this buffer. */
> > -	if (index > last_index)
> > -		goto out;
> > -
> >  	while (rc == 0) {
> > +		unsigned int buf_offset = 0;
> >  		struct llog_rec_hdr *rec;
> > +		bool partial_chunk;
> > +		off_t chunk_offset;
> >  
> >  		/* skip records not set in bitmap */
> >  		while (index <= last_index &&
> >  		       !ext2_test_bit(index, LLOG_HDR_BITMAP(llh)))
> >  			++index;
> >  
> > -		LASSERT(index <= last_index + 1);
> > -		if (index == last_index + 1)
> > +		if (index > last_index)
> >  			break;
> > -repeat:
> > +
> >  		CDEBUG(D_OTHER, "index: %d last_index %d\n",
> >  		       index, last_index);
> > -
> > +repeat:
> >  		/* get the buf with our target record; avoid old garbage */
> >  		memset(buf, 0, chunk_size);
> > -		last_offset = cur_offset;
> >  		rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
> >  				     index, &cur_offset, buf, chunk_size);
> >  		if (rc)
> >  			goto out;
> >  
> > +		/*
> > +		 * NB: after llog_next_block() call the cur_offset is the
> > +		 * offset of the next block after read one.
> > +		 * The absolute offset of the current chunk is calculated
> > +		 * from cur_offset value and stored in chunk_offset variable.
> > +		 */
> > +		tmp_offset = cur_offset;
> > +		if (do_div(tmp_offset, chunk_size)) {
> > +			partial_chunk = true;
> > +			chunk_offset = cur_offset & ~(chunk_size - 1);
> > +		} else {
> > +			partial_chunk = false;
> > +			chunk_offset = cur_offset - chunk_size;
> > +		}
> > +
> >  		/* NB: when rec->lrh_len is accessed it is already swabbed
> >  		 * since it is used at the "end" of the loop and the rec
> >  		 * swabbing is done at the beginning of the loop.
> >  		 */
> > -		for (rec = (struct llog_rec_hdr *)buf;
> > +		for (rec = (struct llog_rec_hdr *)(buf + buf_offset);
> >  		     (char *)rec < buf + chunk_size;
> >  		     rec = llog_rec_hdr_next(rec)) {
> >  			CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
> > @@ -288,13 +301,28 @@ static int llog_process_thread(void *arg)
> >  			CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
> >  			       rec->lrh_type, rec->lrh_index);
> >  
> > -			if (rec->lrh_index == 0) {
> > -				/* probably another rec just got added? */
> > -				rc = 0;
> > -				if (index <= loghandle->lgh_last_idx)
> > -					goto repeat;
> > -				goto out; /* no more records */
> > +			/*
> > +			 * for partial chunk the end of it is zeroed, check
> > +			 * for index 0 to distinguish it.
> > +			 */
> > +			if (partial_chunk && !rec->lrh_index) {
> > +				/* concurrent llog_add() might add new records
> > +				 * while llog_processing, check this is not
> > +				 * the case and re-read the current chunk
> > +				 * otherwise.
> > +				 */
> > +				if (index > loghandle->lgh_last_idx) {
> > +					rc = 0;
> > +					goto out;
> > +				}
> > +				CDEBUG(D_OTHER, "Re-read last llog buffer for new records, index %u, last %u\n",
> > +				       index, loghandle->lgh_last_idx);
> > +				/* save offset inside buffer for the re-read */
> > +				buf_offset = (char *)rec - (char *)buf;
> > +				cur_offset = chunk_offset;
> > +				goto repeat;
> >  			}
> > +
> >  			if (!rec->lrh_len || rec->lrh_len > chunk_size) {
> >  				CWARN("invalid length %d in llog record for index %d/%d\n",
> >  				      rec->lrh_len,
> > @@ -309,6 +337,14 @@ static int llog_process_thread(void *arg)
> >  				continue;
> >  			}
> >  
> > +			if (rec->lrh_index != index) {
> > +				CERROR("%s: Invalid record: index %u but expected %u\n",
> > +				       loghandle->lgh_ctxt->loc_obd->obd_name,
> > +				       rec->lrh_index, index);
> > +				rc = -ERANGE;
> > +				goto out;
> > +			}
> > +
> >  			CDEBUG(D_OTHER,
> >  			       "lrh_index: %d lrh_len: %d (%d remains)\n",
> >  			       rec->lrh_index, rec->lrh_len,
> > @@ -316,7 +352,7 @@ static int llog_process_thread(void *arg)
> >  
> >  			loghandle->lgh_cur_idx = rec->lrh_index;
> >  			loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
> > -						    last_offset;
> > +						    chunk_offset;
> >  
> >  			/* if set, process the callback on this record */
> >  			if (ext2_test_bit(index, LLOG_HDR_BITMAP(llh))) {
> > @@ -325,16 +361,14 @@ static int llog_process_thread(void *arg)
> >  				last_called_index = index;
> >  				if (rc)
> >  					goto out;
> > -			} else {
> > -				CDEBUG(D_OTHER, "Skipped index %d\n", index);
> >  			}
> >  
> > -			/* next record, still in buffer? */
> > -			++index;
> > -			if (index > last_index) {
> > +			/* exit if the last index is reached */
> > +			if (index >= last_index) {
> >  				rc = 0;
> >  				goto out;
> >  			}
> > +			index++;
> >  		}
> >  	}
> >  
> > diff --git a/include/linux/fs.h b/include/linux/fs.h
> > index dc0478c..3f70ec3 100644
> > --- a/include/linux/fs.h
> > +++ b/include/linux/fs.h
> > @@ -2690,7 +2690,7 @@ enum kernel_read_file_id {
> >  
> >  static inline const char *kernel_read_file_id_str(enum kernel_read_file_id id)
> >  {
> > -	if (id < 0 || id >= READING_MAX_ID)
> > +	if (id >= READING_MAX_ID)
> >  		return kernel_read_file_str[READING_UNKNOWN];
> >  
> >  	return kernel_read_file_str[id];
> 
> Why are you modifying this fs.h function in this patch?  Your changelog
> does not talk about it at all.  Nor does lustre even call this function,
> only a security module, loadpin.
> 
> totally confused,

Oh that is my mistake. So when I build lustre with W=1 it spits out 
warnings about id being unsigned and testing for < 0. I patched it 
to remove the noise to see the real errors. I included it by mistake.



More information about the devel mailing list