Fix "allocating allocated segment" panic

Gunnar Beutner did all the hard work on this one by correctly
identifying that this issue is a race between dmu_sync() and
dbuf_dirty().

Now in all cases the caller is responsible for preventing this
race by making sure the zfs_range_lock() is held when dirtying
a buffer which may be referenced in a log record.  The mmap
case which relies on zfs_putpage() was not taking the range
lock.  This code was accidentally dropped when the function
was rewritten for the Linux VFS.

This patch adds the required range locking to zfs_putpage().

It also adds the missing ZFS_ENTER()/ZFS_EXIT() macros which
aren't strictly required due to the VFS holding a reference.
However, this makes the code more consistent with the upsteam
code and there's no harm in being extra careful here.

Original-patch-by: Gunnar Beutner <gunnar@beutner.name>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #541

diff --git a/module/zfs/zfs_vnops.c b/module/zfs/zfs_vnops.c
index 89f0f60..5765c9a 100644 (file)
--- a/module/zfs/zfs_vnops.c
+++ b/module/zfs/zfs_vnops.c
@@ -3790,7 +3790,8 @@ zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
        zfs_sb_t        *zsb = ITOZSB(ip);
        loff_t          offset;
        loff_t          pgoff;
-        unsigned int   pglen;
+       unsigned int    pglen;
+       rl_t            *rl;
        dmu_tx_t        *tx;
        caddr_t         va;
        int             err = 0;
@@ -3799,6 +3800,8 @@ zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
        int             cnt = 0;
        int             sync;
 
+       ZFS_ENTER(zsb);
+       ZFS_VERIFY_ZP(zp);
 
        ASSERT(PageLocked(pp));
 
@@ -3810,6 +3813,7 @@ zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
        /* Page is beyond end of file */
        if (pgoff >= offset) {
                unlock_page(pp);
+               ZFS_EXIT(zsb);
                return (0);
        }
 
@@ -3832,6 +3836,7 @@ zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
        set_page_writeback(pp);
        unlock_page(pp);
 
+       rl = zfs_range_lock(zp, pgoff, pglen, RL_WRITER);
        tx = dmu_tx_create(zsb->z_os);
 
        sync = ((zsb->z_os->os_sync == ZFS_SYNC_ALWAYS) ||
@@ -3858,6 +3863,8 @@ zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
                if (sync)
                        zfs_putpage_commit_cb(pp, ECANCELED);
 
+               zfs_range_unlock(rl);
+               ZFS_EXIT(zsb);
                return (err);
        }
 
@@ -3873,6 +3880,7 @@ zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
        zfs_log_write(zsb->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0);
 
        dmu_tx_commit(tx);
+       zfs_range_unlock(rl);
        ASSERT3S(err, ==, 0);
 
        if (sync) {
@@ -3880,6 +3888,7 @@ zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
                zfs_putpage_commit_cb(pp, err);
        }
 
+       ZFS_EXIT(zsb);
        return (err);
 }