[GFS2] Fix lock ordering bug in page fault path

Mmapped files were able to trigger a lock ordering bug. Private
maps do not need to take the glock so early on. Shared maps do
unfortunately, however we can get around that by adding a flag
into the flags for the struct gfs2_file. This only works because
we are taking an exclusive lock at this point, so we know that
nobody else can be racing with us.

Fixes Red Hat bugzilla: #201196

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>

fs/gfs2/incore.h

@@ -279,6 +279,7 @@ static inline struct gfs2_sbd *GFS2_SB(struct inode *inode)
 
 enum {
 	GFF_DID_DIRECT_ALLOC	= 0,
+	GFF_EXLOCK = 1,
 };
 
 struct gfs2_file {

fs/gfs2/log.c

@@ -256,8 +256,8 @@ static unsigned int current_tail(struct gfs2_sbd *sdp)
 	if (list_empty(&sdp->sd_ail1_list))
 		tail = sdp->sd_log_head;
 	else {
-		ai = list_entry(sdp->sd_ail1_list.prev,
-				struct gfs2_ail, ai_list);
+		ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail,
+				ai_list);
 		tail = ai->ai_first;
 	}
 

fs/gfs2/ops_address.c

@@ -237,14 +237,22 @@ static int gfs2_readpage(struct file *file, struct page *page)
 	struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
 	struct gfs2_holder gh;
 	int error;
+	int do_unlock = 0;
 
 	if (likely(file != &gfs2_internal_file_sentinal)) {
+		if (file) {
+			struct gfs2_file *gf = file->private_data;
+			if (test_bit(GFF_EXLOCK, &gf->f_flags))
+				goto skip_lock;
+		}
 		gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|GL_AOP, &gh);
+		do_unlock = 1;
 		error = gfs2_glock_nq_m_atime(1, &gh);
 		if (unlikely(error))
 			goto out_unlock;
 	}
 
+skip_lock:
 	if (gfs2_is_stuffed(ip)) {
 		error = stuffed_readpage(ip, page);
 		unlock_page(page);
@@ -262,7 +270,7 @@ out:
 	return error;
 out_unlock:
 	unlock_page(page);
-	if (file != &gfs2_internal_file_sentinal)
+	if (do_unlock)
 		gfs2_holder_uninit(&gh);
 	goto out;
 }
@@ -291,17 +299,24 @@ static int gfs2_readpages(struct file *file, struct address_space *mapping,
 	struct gfs2_holder gh;
 	unsigned page_idx;
 	int ret;
+	int do_unlock = 0;
 
 	if (likely(file != &gfs2_internal_file_sentinal)) {
+		if (file) {
+			struct gfs2_file *gf = file->private_data;
+			if (test_bit(GFF_EXLOCK, &gf->f_flags))
+				goto skip_lock;
+		}
 		gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
 				 LM_FLAG_TRY_1CB|GL_ATIME|GL_AOP, &gh);
+		do_unlock = 1;
 		ret = gfs2_glock_nq_m_atime(1, &gh);
 		if (ret == GLR_TRYFAILED) 
 			goto out_noerror;
 		if (unlikely(ret))
 			goto out_unlock;
 	}
-
+skip_lock:
 	if (gfs2_is_stuffed(ip)) {
 		struct pagevec lru_pvec;
 		pagevec_init(&lru_pvec, 0);
@@ -326,7 +341,7 @@ static int gfs2_readpages(struct file *file, struct address_space *mapping,
 		ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
 	}
 
-	if (likely(file != &gfs2_internal_file_sentinal)) {
+	if (do_unlock) {
 		gfs2_glock_dq_m(1, &gh);
 		gfs2_holder_uninit(&gh);
 	}
@@ -344,7 +359,7 @@ out_unlock:
 		unlock_page(page);
 		page_cache_release(page);
 	}
-	if (likely(file != &gfs2_internal_file_sentinal))
+	if (do_unlock)
 		gfs2_holder_uninit(&gh);
 	goto out;
 }

fs/gfs2/ops_vm.c

@@ -46,13 +46,7 @@ static struct page *gfs2_private_nopage(struct vm_area_struct *area,
 					unsigned long address, int *type)
 {
 	struct gfs2_inode *ip = GFS2_I(area->vm_file->f_mapping->host);
-	struct gfs2_holder i_gh;
 	struct page *result;
-	int error;
-
-	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
-	if (error)
-		return NULL;
 
 	set_bit(GIF_PAGED, &ip->i_flags);
 
@@ -61,8 +55,6 @@ static struct page *gfs2_private_nopage(struct vm_area_struct *area,
 	if (result && result != NOPAGE_OOM)
 		pfault_be_greedy(ip);
 
-	gfs2_glock_dq_uninit(&i_gh);
-
 	return result;
 }
 
@@ -141,7 +133,9 @@ static int alloc_page_backing(struct gfs2_inode *ip, struct page *page)
 static struct page *gfs2_sharewrite_nopage(struct vm_area_struct *area,
 					   unsigned long address, int *type)
 {
-	struct gfs2_inode *ip = GFS2_I(area->vm_file->f_mapping->host);
+	struct file *file = area->vm_file;
+	struct gfs2_file *gf = file->private_data;
+	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 	struct gfs2_holder i_gh;
 	struct page *result = NULL;
 	unsigned long index = ((address - area->vm_start) >> PAGE_CACHE_SHIFT) +
@@ -156,13 +150,14 @@ static struct page *gfs2_sharewrite_nopage(struct vm_area_struct *area,
 	set_bit(GIF_PAGED, &ip->i_flags);
 	set_bit(GIF_SW_PAGED, &ip->i_flags);
 
-	error = gfs2_write_alloc_required(ip,
-					  (uint64_t)index << PAGE_CACHE_SHIFT,
+	error = gfs2_write_alloc_required(ip, (u64)index << PAGE_CACHE_SHIFT,
 					  PAGE_CACHE_SIZE, &alloc_required);
 	if (error)
 		goto out;
 
+	set_bit(GFF_EXLOCK, &gf->f_flags);
 	result = filemap_nopage(area, address, type);
+	clear_bit(GFF_EXLOCK, &gf->f_flags);
 	if (!result || result == NOPAGE_OOM)
 		goto out;
 
@@ -177,8 +172,7 @@ static struct page *gfs2_sharewrite_nopage(struct vm_area_struct *area,
 	}
 
 	pfault_be_greedy(ip);
-
- out:
+out:
 	gfs2_glock_dq_uninit(&i_gh);
 
 	return result;

fs/gfs2/recovery.c

@@ -153,8 +153,7 @@ static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
 
 	if (lh.lh_header.mh_magic != GFS2_MAGIC ||
 	    lh.lh_header.mh_type != GFS2_METATYPE_LH ||
-	    lh.lh_blkno != blk ||
-	    lh.lh_hash != hash)
+	    lh.lh_blkno != blk || lh.lh_hash != hash)
 		return 1;
 
 	*head = lh;
@@ -482,11 +481,9 @@ int gfs2_recover_journal(struct gfs2_jdesc *jd)
 
 		/* Acquire a shared hold on the transaction lock */
 
-		error = gfs2_glock_nq_init(sdp->sd_trans_gl,
-					   LM_ST_SHARED,
+		error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
 					   LM_FLAG_NOEXP | LM_FLAG_PRIORITY |
-					   GL_NOCANCEL | GL_NOCACHE,
-					   &t_gh);
+					   GL_NOCANCEL | GL_NOCACHE, &t_gh);
 		if (error)
 			goto fail_gunlock_ji;