[PATCH] reiser4: prevent system lockups.

∙

From reiserfs-devel Wed Feb 05 00:28:06 2020
From: Edward Shishkin <edward.shishkin () gmail ! com>
Date: Wed, 05 Feb 2020 00:28:06 +0000
To: reiserfs-devel
Subject: [PATCH] reiser4: prevent system lockups
Message-Id: <94861335-0ff3-a8cf-8d73-e0f521fa0ffa () gmail ! com>
X-MARC-Message: https://marc.info/?l=reiserfs-devel&m=158086248927420
MIME-Version: 1
Content-Type: multipart/mixed; boundary="--------------F3FF70D4737419259F49BAF5"

This is a multi-part message in MIME format.
--------------F3FF70D4737419259F49BAF5
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 7bit

Hi all,

Starting from Linux-4.15, VFS introduces optimizations, which
are incompatible with Reiser4 (as well as with any FS possessing
an advanced transaction manager). Here is the problematic commit:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=aa65c29ce1b6e1990cd2c7d8004bbea7ff3aff38

The attached patch fixes the problem. This patch is against
Linux-5.4+reiser4. Please, apply.

Later I'll prepare backports for all the mentioned kernels and will
put them on Sourceforge. If someone urgently needs a backport, then
let me know.

Thanks,
Edward.

--------------F3FF70D4737419259F49BAF5
Content-Type: text/x-patch;
name="reiser4-fix-lockup.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="reiser4-fix-lockup.patch"

diff -urN linux-5.4.6/fs/reiser4/page_cache.c  linux-5.4.6.patched/fs/reiser4/page_cache.c
---  linux-5.4.6/fs/reiser4/page_cache.c   2020-02-05 00:08:07.121592727 +0100
+++  linux-5.4.6.patched/fs/reiser4/page_cache.c   2020-02-05 00:09:44.508645047 +0100
@@ -549,7 +549,8 @@
                        */
                        .releasepage = reiser4_releasepage,
                        .direct_IO = NULL,
-                      .migratepage = reiser4_migratepage
+                      .migratepage = reiser4_migratepage,
+                      .batch_lock_tabu = 1
   };

/* called just before page is released (no longer used by reiser4). Callers:
diff -urN linux-5.4.6/fs/reiser4/plugin/object.c  linux-5.4.6.patched/fs/reiser4/plugin/object.c
---  linux-5.4.6/fs/reiser4/plugin/object.c     2020-02-05 00:08:07.129592814 +0100
+++  linux-5.4.6.patched/fs/reiser4/plugin/object.c     2020-02-05 00:09:44.508645047 +0100
@@ -150,7 +150,8 @@
                      .bmap = reiser4_bmap_dispatch,
                      .invalidatepage = reiser4_invalidatepage,
                      .releasepage = reiser4_releasepage,
-                    .migratepage = reiser4_migratepage
+                    .migratepage = reiser4_migratepage,
+                    .batch_lock_tabu = 1
   };

/* VFS methods for symlink files */
diff -urN linux-5.4.6/include/linux/fs.h  linux-5.4.6.patched/include/linux/fs.h
---  linux-5.4.6/include/linux/fs.h     2020-02-05 00:08:07.135592878 +0100
+++  linux-5.4.6.patched/include/linux/fs.h     2020-02-05 00:09:21.339394689 +0100
@@ -409,6 +409,7 @@
                     int (*swap_activate)(struct swap_info_struct *sis, struct file *file,
                                                                  sector_t *span);
                     void (*swap_deactivate)(struct file *file);
+                   int batch_lock_tabu;
   };

extern const struct address_space_operations empty_aops;
diff -urN linux-5.4.6/mm/truncate.c  linux-5.4.6.patched/mm/truncate.c
---  linux-5.4.6/mm/truncate.c     2019-12-21 11:05:23.000000000 +0100
+++  linux-5.4.6.patched/mm/truncate.c     2020-02-05 00:09:21.339394689 +0100
@@ -360,6 +360,13 @@
                                                                    continue;
                                                 }
                                                 pagevec_add(&locked_pvec, page);
+                                               if (mapping->a_ops->batch_lock_tabu)
+                                                                  /*
+                                                                    * the file system doesn't allow to hold
+                                                                    * many pages locked, while calling
+                                                                    * ->invalidatepage() for one of them
+                                                                  */
+                                                                   break;
                                }
                                for (i = 0; i < pagevec_count(&locked_pvec); i++)
                                                 truncate_cleanup_page(mapping, locked_pvec.pages[i]);