Ext3 journaling patches from 2.4.21:

- jbd-flushtime: fixes dirty buffer starvation during IO (may help perf!)
- jbd-ctx_switch: avoids thousands of spurious context switches on busy buffers
- jbd-get_write_access: avoids assertion failure after IO error (bug 1871)

diff --git a/lustre/kernel_patches/patches/jbd-ctx_switch.patch b/lustre/kernel_patches/patches/jbd-ctx_switch.patch
new file mode 100644 (file)
index 0000000..1b27c2d
--- /dev/null
+++ b/lustre/kernel_patches/patches/jbd-ctx_switch.patch
@@ -0,0 +1,13 @@
+diff -rup linux-2.4.20-uml linux-2.4.21-rc2
+--- linux-2.4.20-uml/fs/jbd/transaction.c      Thu Nov 28 16:53:15 2002
++++ linux-2.4.21-rc2/fs/jbd/transaction.c      Fri May 16 11:00:40 2003
+@@ -666,7 +673,8 @@ repeat:
+                       spin_unlock(&journal_datalist_lock);
+                       unlock_journal(journal);
+                       /* commit wakes up all shadow buffers after IO */
+-                      sleep_on(&jh2bh(jh)->b_wait);
++                      wait_event(jh2bh(jh)->b_wait,
++                                              jh->b_jlist != BJ_Shadow);
+                       lock_journal(journal);
+                       goto repeat;
+               }

diff --git a/lustre/kernel_patches/patches/jbd-flushtime.patch b/lustre/kernel_patches/patches/jbd-flushtime.patch
new file mode 100644 (file)
index 0000000..7e2fbd4
--- /dev/null
+++ b/lustre/kernel_patches/patches/jbd-flushtime.patch
@@ -0,0 +1,34 @@
+diff -rup linux-2.4.20-uml linux-2.4.21-rc2
+--- linux-2.4.20-uml/fs/jbd/transaction.c      Thu Nov 28 16:53:15 2002
++++ linux-2.4.21-rc2/fs/jbd/transaction.c      Fri May 16 11:00:40 2003
+@@ -1109,7 +1142,6 @@ int journal_dirty_metadata (handle_t *ha
+       
+       spin_lock(&journal_datalist_lock);
+       set_bit(BH_JBDDirty, &bh->b_state);
+-      set_buffer_flushtime(bh);
+ 
+       J_ASSERT_JH(jh, jh->b_transaction != NULL);
+       
+@@ -2035,6 +2093,13 @@ void journal_file_buffer(struct journal_
+       spin_unlock(&journal_datalist_lock);
+ }
+ 
++static void jbd_refile_buffer(struct buffer_head *bh)
++{
++      if (buffer_dirty(bh) && (bh->b_list != BUF_DIRTY))
++              set_buffer_flushtime(bh);
++      refile_buffer(bh);
++}
++
+ /* 
+  * Remove a buffer from its current buffer list in preparation for
+  * dropping it from its current transaction entirely.  If the buffer has
+@@ -2055,7 +2120,7 @@ void __journal_refile_buffer(struct jour
+               __journal_unfile_buffer(jh);
+               jh->b_transaction = NULL;
+               /* Onto BUF_DIRTY for writeback */
+-              refile_buffer(jh2bh(jh));
++              jbd_refile_buffer(jh2bh(jh));
+               return;
+       }
+       

diff --git a/lustre/kernel_patches/patches/jbd-get_write_access.patch b/lustre/kernel_patches/patches/jbd-get_write_access.patch
new file mode 100644 (file)
index 0000000..345a3f9
--- /dev/null
+++ b/lustre/kernel_patches/patches/jbd-get_write_access.patch
@@ -0,0 +1,56 @@
+diff -rup linux-2.4.20-uml linux-2.4.21-rc2
+--- linux-2.4.20-uml/fs/jbd/transaction.c      Thu Nov 28 16:53:15 2002
++++ linux-2.4.21-rc2/fs/jbd/transaction.c      Fri May 16 11:00:40 2003
+@@ -735,7 +743,8 @@ done_locked:
+               int offset;
+               char *source;
+ 
+-              J_ASSERT_JH(jh, buffer_uptodate(jh2bh(jh)));
++              J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)),
++                          "Possible IO failure.\n");
+               page = jh2bh(jh)->b_page;
+               offset = ((unsigned long) jh2bh(jh)->b_data) & ~PAGE_MASK;
+               source = kmap(page);
+diff -rup linux-2.4.20-uml linux-2.4.21-rc2
+--- linux-2.4.20-uml/include/linux/jbd.h       Mon Aug 25 16:16:57 2003
++++ linux-2.4.21-rc2/include/linux/jbd.h       Tue May 13 13:28:29 2003
+@@ -40,6 +40,15 @@
+  */
+ #undef JBD_PARANOID_WRITES
+ 
++/*
++ * Define JBD_PARANIOD_IOFAIL to cause a kernel BUG() if ext3 finds
++ * certain classes of error which can occur due to failed IOs.  Under
++ * normal use we want ext3 to continue after such errors, because
++ * hardware _can_ fail, but for debugging purposes when running tests on
++ * known-good hardware we may want to trap these errors.
++ */
++#undef JBD_PARANOID_IOFAIL
++
+ #ifdef CONFIG_JBD_DEBUG
+ /*
+  * Define JBD_EXPENSIVE_CHECKING to enable more expensive internal
+@@ -232,6 +272,23 @@ void buffer_assertion_failure(struct buf
+ #define J_ASSERT(assert)      do { } while (0)
+ #endif                /* JBD_ASSERTIONS */
+ 
++#if defined(JBD_PARANOID_IOFAIL)
++#define J_EXPECT(expr, why...)                J_ASSERT(expr)
++#define J_EXPECT_BH(bh, expr, why...) J_ASSERT_BH(bh, expr)
++#define J_EXPECT_JH(jh, expr, why...) J_ASSERT_JH(jh, expr)
++#else
++#define __journal_expect(expr, why...)                                             \
++      do {                                                                 \
++              if (!(expr)) {                                               \
++                      printk(KERN_ERR "EXT3-fs unexpected failure: %s;\n", # expr); \
++                      printk(KERN_ERR why);                                \
++              }                                                            \
++      } while (0)
++#define J_EXPECT(expr, why...)                __journal_expect(expr, ## why)
++#define J_EXPECT_BH(bh, expr, why...) __journal_expect(expr, ## why)
++#define J_EXPECT_JH(jh, expr, why...) __journal_expect(expr, ## why)
++#endif
++
+ enum jbd_state_bits {
+       BH_JWrite
+         = BH_PrivateStart,    /* 1 if being written to log (@@@ DEBUGGING) */