[XFS] Introduce two new mount options (nolargeio/largeio) to allow

filesystems to expose the filesystem stripe width in stat(2) rather than
the page cache size. This allows applications requiring high bandwidth to
easily determine the optimum I/O size for the underlying filesystem. The
default is to report the page cache size (i.e. "nolargeio").

SGI-PV: 942818
SGI-Modid: xfs-linux:xfs-kern:23830a

Signed-off-by: David Chinner <dgc@sgi.com>
Signed-off-by: Nathan Scott <nathans@sgi.com>

6 files changed, 43 insertions, 35 deletions

diff --git a/fs/xfs/linux-2.6/xfs_super.c b/fs/xfs/linux-2.6/xfs_super.c
index fa87279405d8..f6f6b6750d4f 100644
--- a/fs/xfs/linux-2.6/xfs_super.c
+++ b/fs/xfs/linux-2.6/xfs_super.c
@@ -189,7 +189,7 @@ xfs_revalidate_inode(
                 break;
         }
 
-        inode->i_blksize = PAGE_CACHE_SIZE;
+        inode->i_blksize = xfs_preferred_iosize(mp);
         inode->i_generation = ip->i_d.di_gen;
         i_size_write(inode, ip->i_d.di_size);
         inode->i_blocks =
diff --git a/fs/xfs/linux-2.6/xfs_vnode.c b/fs/xfs/linux-2.6/xfs_vnode.c
index 268f45bf6a9a..61999649ecef 100644
--- a/fs/xfs/linux-2.6/xfs_vnode.c
+++ b/fs/xfs/linux-2.6/xfs_vnode.c
@@ -124,6 +124,7 @@ vn_revalidate_core(
         inode->i_mtime      = vap->va_mtime;
         inode->i_ctime      = vap->va_ctime;
         inode->i_atime      = vap->va_atime;
+        inode->i_blksize    = vap->va_blocksize;
         if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
                 inode->i_flags |= S_IMMUTABLE;
         else
diff --git a/fs/xfs/xfs_clnt.h b/fs/xfs/xfs_clnt.h
index c93cb282f3d8..90d9d56c4dc8 100644
--- a/fs/xfs/xfs_clnt.h
+++ b/fs/xfs/xfs_clnt.h
@@ -106,5 +106,7 @@ struct xfs_mount_args {
 #define XFSMNT_IHASHSIZE        0x20000000      /* inode hash table size */
 #define XFSMNT_DIRSYNC          0x40000000      /* sync creat,link,unlink,rename
                                                  * symlink,mkdir,rmdir,mknod */
+#define XFSMNT_COMPAT_IOSIZE    0x80000000      /* don't report large preferred
+                                                 * I/O size in stat() */
 
 #endif  /* __XFS_CLNT_H__ */
diff --git a/fs/xfs/xfs_mount.h b/fs/xfs/xfs_mount.h
index 0653beecf93a..b71af184aea6 100644
--- a/fs/xfs/xfs_mount.h
+++ b/fs/xfs/xfs_mount.h
@@ -421,6 +421,9 @@ typedef struct xfs_mount {
                                                  * allocation */
 #define XFS_MOUNT_IHASHSIZE     0x00100000      /* inode hash table size */
 #define XFS_MOUNT_DIRSYNC       0x00200000      /* synchronous directory ops */
+#define XFS_MOUNT_COMPAT_IOSIZE 0x00400000      /* don't report large preferred
+                                                 * I/O size in stat() */
+
 
 /*
  * Default minimum read and write sizes.
@@ -442,6 +445,30 @@ typedef struct xfs_mount {
 #define XFS_WSYNC_READIO_LOG    15      /* 32K */
 #define XFS_WSYNC_WRITEIO_LOG   14      /* 16K */
 
+/*
+ * Allow large block sizes to be reported to userspace programs if the
+ * "largeio" mount option is used. 
+ *
+ * If compatibility mode is specified, simply return the basic unit of caching
+ * so that we don't get inefficient read/modify/write I/O from user apps.
+ * Otherwise....
+ *
+ * If the underlying volume is a stripe, then return the stripe width in bytes
+ * as the recommended I/O size. It is not a stripe and we've set a default
+ * buffered I/O size, return that, otherwise return the compat default.
+ */
+static inline unsigned long
+xfs_preferred_iosize(xfs_mount_t *mp)
+{
+        if (mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE)
+                return PAGE_CACHE_SIZE;
+        return (mp->m_swidth ?
+                (mp->m_swidth << mp->m_sb.sb_blocklog) :
+                ((mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) ?
+                        (1 << (int)MAX(mp->m_readio_log, mp->m_writeio_log)) :
+                        PAGE_CACHE_SIZE));
+}
+
 #define XFS_MAXIOFFSET(mp)      ((mp)->m_maxioffset)
 
 #define XFS_FORCED_SHUTDOWN(mp) ((mp)->m_flags & XFS_MOUNT_FS_SHUTDOWN)
diff --git a/fs/xfs/xfs_vfsops.c b/fs/xfs/xfs_vfsops.c
index 9142351df515..7227baee8994 100644
--- a/fs/xfs/xfs_vfsops.c
+++ b/fs/xfs/xfs_vfsops.c
@@ -307,6 +307,9 @@ xfs_start_flags(
         if (ap->flags & XFSMNT_DIRSYNC)
                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 
+        if (ap->flags & XFSMNT_COMPAT_IOSIZE)
+                mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
+
         /*
          * no recovery flag requires a read-only mount
          */
@@ -1645,6 +1648,9 @@ xfs_vget(
 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
+#define MNTOPT_LARGEIO     "largeio"    /* report large I/O sizes in stat() */
+#define MNTOPT_NOLARGEIO   "nolargeio"  /* do not report large I/O sizes
+                                         * in stat(). */
 
 STATIC unsigned long
 suffix_strtoul(const char *cp, char **endp, unsigned int base)
@@ -1681,6 +1687,7 @@ xfs_parseargs(
         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
         int                     iosize;
 
+        args->flags |= XFSMNT_COMPAT_IOSIZE;
 #if 0   /* XXX: off by default, until some remaining issues ironed out */
         args->flags |= XFSMNT_IDELETE; /* default to on */
 #endif
@@ -1809,6 +1816,10 @@ xfs_parseargs(
                         args->flags &= ~XFSMNT_IDELETE;
                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
                         args->flags |= XFSMNT_IDELETE;
+                } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
+                        args->flags &= ~XFSMNT_COMPAT_IOSIZE;
+                } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
+                        args->flags |= XFSMNT_COMPAT_IOSIZE;
                 } else if (!strcmp(this_char, "osyncisdsync")) {
                         /* no-op, this is now the default */
 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
diff --git a/fs/xfs/xfs_vnodeops.c b/fs/xfs/xfs_vnodeops.c
index e2bf2ef58b66..8221b11a48c0 100644
--- a/fs/xfs/xfs_vnodeops.c
+++ b/fs/xfs/xfs_vnodeops.c
@@ -181,40 +181,7 @@ xfs_getattr(
                 vap->va_rdev = 0;
 
                 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
-
-#if 0
-                        /* Large block sizes confuse various
-                         * user space programs, so letting the
-                         * stripe size through is not a good
-                         * idea for now.
-                         */
-                        vap->va_blocksize = mp->m_swidth ?
-                                /*
-                                 * If the underlying volume is a stripe, then
-                                 * return the stripe width in bytes as the
-                                 * recommended I/O size.
-                                 */
-                                (mp->m_swidth << mp->m_sb.sb_blocklog) :
-                                /*
-                                 * Return the largest of the preferred buffer
-                                 * sizes since doing small I/Os into larger
-                                 * buffers causes buffers to be decommissioned.
-                                 * The value returned is in bytes.
-                                 */
-                                (1 << (int)MAX(mp->m_readio_log,
-                                               mp->m_writeio_log));
-
-#else
-                        vap->va_blocksize =
-                                /*
-                                 * Return the largest of the preferred buffer
-                                 * sizes since doing small I/Os into larger
-                                 * buffers causes buffers to be decommissioned.
-                                 * The value returned is in bytes.
-                                 */
-                                1 << (int)MAX(mp->m_readio_log,
-                                               mp->m_writeio_log);
-#endif
+                        vap->va_blocksize = xfs_preferred_iosize(mp);
                 } else {
 
                         /*