This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / fs / gfs2 / glops.c
diff --git a/fs/gfs2/glops.c b/fs/gfs2/glops.c
new file mode 100644 (file)
index 0000000..39c8ae2
--- /dev/null
@@ -0,0 +1,497 @@
+/*
+ * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
+ * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License version 2.
+ */
+
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/completion.h>
+#include <linux/buffer_head.h>
+#include <linux/gfs2_ondisk.h>
+#include <linux/lm_interface.h>
+
+#include "gfs2.h"
+#include "incore.h"
+#include "bmap.h"
+#include "glock.h"
+#include "glops.h"
+#include "inode.h"
+#include "log.h"
+#include "meta_io.h"
+#include "recovery.h"
+#include "rgrp.h"
+#include "util.h"
+#include "trans.h"
+
+/**
+ * ail_empty_gl - remove all buffers for a given lock from the AIL
+ * @gl: the glock
+ *
+ * None of the buffers should be dirty, locked, or pinned.
+ */
+
+static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
+{
+       struct gfs2_sbd *sdp = gl->gl_sbd;
+       unsigned int blocks;
+       struct list_head *head = &gl->gl_ail_list;
+       struct gfs2_bufdata *bd;
+       struct buffer_head *bh;
+       u64 blkno;
+       int error;
+
+       blocks = atomic_read(&gl->gl_ail_count);
+       if (!blocks)
+               return;
+
+       error = gfs2_trans_begin(sdp, 0, blocks);
+       if (gfs2_assert_withdraw(sdp, !error))
+               return;
+
+       gfs2_log_lock(sdp);
+       while (!list_empty(head)) {
+               bd = list_entry(head->next, struct gfs2_bufdata,
+                               bd_ail_gl_list);
+               bh = bd->bd_bh;
+               blkno = bh->b_blocknr;
+               gfs2_assert_withdraw(sdp, !buffer_busy(bh));
+
+               bd->bd_ail = NULL;
+               list_del(&bd->bd_ail_st_list);
+               list_del(&bd->bd_ail_gl_list);
+               atomic_dec(&gl->gl_ail_count);
+               brelse(bh);
+               gfs2_log_unlock(sdp);
+
+               gfs2_trans_add_revoke(sdp, blkno);
+
+               gfs2_log_lock(sdp);
+       }
+       gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
+       gfs2_log_unlock(sdp);
+
+       gfs2_trans_end(sdp);
+       gfs2_log_flush(sdp, NULL);
+}
+
+/**
+ * gfs2_pte_inval - Sync and invalidate all PTEs associated with a glock
+ * @gl: the glock
+ *
+ */
+
+static void gfs2_pte_inval(struct gfs2_glock *gl)
+{
+       struct gfs2_inode *ip;
+       struct inode *inode;
+
+       ip = gl->gl_object;
+       inode = &ip->i_inode;
+       if (!ip || !S_ISREG(inode->i_mode))
+               return;
+
+       if (!test_bit(GIF_PAGED, &ip->i_flags))
+               return;
+
+       unmap_shared_mapping_range(inode->i_mapping, 0, 0);
+
+       if (test_bit(GIF_SW_PAGED, &ip->i_flags))
+               set_bit(GLF_DIRTY, &gl->gl_flags);
+
+       clear_bit(GIF_SW_PAGED, &ip->i_flags);
+}
+
+/**
+ * meta_go_sync - sync out the metadata for this glock
+ * @gl: the glock
+ *
+ * Called when demoting or unlocking an EX glock.  We must flush
+ * to disk all dirty buffers/pages relating to this glock, and must not
+ * not return to caller to demote/unlock the glock until I/O is complete.
+ */
+
+static void meta_go_sync(struct gfs2_glock *gl)
+{
+       if (gl->gl_state != LM_ST_EXCLUSIVE)
+               return;
+
+       if (test_and_clear_bit(GLF_DIRTY, &gl->gl_flags)) {
+               gfs2_log_flush(gl->gl_sbd, gl);
+               gfs2_meta_sync(gl);
+               gfs2_ail_empty_gl(gl);
+       }
+}
+
+/**
+ * meta_go_inval - invalidate the metadata for this glock
+ * @gl: the glock
+ * @flags:
+ *
+ */
+
+static void meta_go_inval(struct gfs2_glock *gl, int flags)
+{
+       if (!(flags & DIO_METADATA))
+               return;
+
+       gfs2_meta_inval(gl);
+       gl->gl_vn++;
+}
+
+/**
+ * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
+ * @gl: the glock protecting the inode
+ *
+ */
+
+static void inode_go_sync(struct gfs2_glock *gl)
+{
+       struct gfs2_inode *ip = gl->gl_object;
+
+       if (ip && !S_ISREG(ip->i_inode.i_mode))
+               ip = NULL;
+
+       if (test_bit(GLF_DIRTY, &gl->gl_flags)) {
+               gfs2_log_flush(gl->gl_sbd, gl);
+               if (ip)
+                       filemap_fdatawrite(ip->i_inode.i_mapping);
+               gfs2_meta_sync(gl);
+               if (ip) {
+                       struct address_space *mapping = ip->i_inode.i_mapping;
+                       int error = filemap_fdatawait(mapping);
+                       if (error == -ENOSPC)
+                               set_bit(AS_ENOSPC, &mapping->flags);
+                       else if (error)
+                               set_bit(AS_EIO, &mapping->flags);
+               }
+               clear_bit(GLF_DIRTY, &gl->gl_flags);
+               gfs2_ail_empty_gl(gl);
+       }
+}
+
+/**
+ * inode_go_xmote_th - promote/demote a glock
+ * @gl: the glock
+ * @state: the requested state
+ * @flags:
+ *
+ */
+
+static void inode_go_xmote_th(struct gfs2_glock *gl)
+{
+       if (gl->gl_state != LM_ST_UNLOCKED)
+               gfs2_pte_inval(gl);
+       if (gl->gl_state == LM_ST_EXCLUSIVE)
+               inode_go_sync(gl);
+}
+
+/**
+ * inode_go_xmote_bh - After promoting/demoting a glock
+ * @gl: the glock
+ *
+ */
+
+static void inode_go_xmote_bh(struct gfs2_glock *gl)
+{
+       struct gfs2_holder *gh = gl->gl_req_gh;
+       struct buffer_head *bh;
+       int error;
+
+       if (gl->gl_state != LM_ST_UNLOCKED &&
+           (!gh || !(gh->gh_flags & GL_SKIP))) {
+               error = gfs2_meta_read(gl, gl->gl_name.ln_number, 0, &bh);
+               if (!error)
+                       brelse(bh);
+       }
+}
+
+/**
+ * inode_go_drop_th - unlock a glock
+ * @gl: the glock
+ *
+ * Invoked from rq_demote().
+ * Another node needs the lock in EXCLUSIVE mode, or lock (unused for too long)
+ * is being purged from our node's glock cache; we're dropping lock.
+ */
+
+static void inode_go_drop_th(struct gfs2_glock *gl)
+{
+       gfs2_pte_inval(gl);
+       if (gl->gl_state == LM_ST_EXCLUSIVE)
+               inode_go_sync(gl);
+}
+
+/**
+ * inode_go_inval - prepare a inode glock to be released
+ * @gl: the glock
+ * @flags:
+ *
+ */
+
+static void inode_go_inval(struct gfs2_glock *gl, int flags)
+{
+       struct gfs2_inode *ip = gl->gl_object;
+       int meta = (flags & DIO_METADATA);
+
+       if (meta) {
+               gfs2_meta_inval(gl);
+               if (ip)
+                       set_bit(GIF_INVALID, &ip->i_flags);
+       }
+
+       if (ip && S_ISREG(ip->i_inode.i_mode)) {
+               truncate_inode_pages(ip->i_inode.i_mapping, 0);
+               clear_bit(GIF_PAGED, &ip->i_flags);
+       }
+}
+
+/**
+ * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
+ * @gl: the glock
+ *
+ * Returns: 1 if it's ok
+ */
+
+static int inode_go_demote_ok(struct gfs2_glock *gl)
+{
+       struct gfs2_sbd *sdp = gl->gl_sbd;
+       int demote = 0;
+
+       if (!gl->gl_object && !gl->gl_aspace->i_mapping->nrpages)
+               demote = 1;
+       else if (!sdp->sd_args.ar_localcaching &&
+                time_after_eq(jiffies, gl->gl_stamp +
+                              gfs2_tune_get(sdp, gt_demote_secs) * HZ))
+               demote = 1;
+
+       return demote;
+}
+
+/**
+ * inode_go_lock - operation done after an inode lock is locked by a process
+ * @gl: the glock
+ * @flags:
+ *
+ * Returns: errno
+ */
+
+static int inode_go_lock(struct gfs2_holder *gh)
+{
+       struct gfs2_glock *gl = gh->gh_gl;
+       struct gfs2_inode *ip = gl->gl_object;
+       int error = 0;
+
+       if (!ip)
+               return 0;
+
+       if (test_bit(GIF_INVALID, &ip->i_flags)) {
+               error = gfs2_inode_refresh(ip);
+               if (error)
+                       return error;
+       }
+
+       if ((ip->i_di.di_flags & GFS2_DIF_TRUNC_IN_PROG) &&
+           (gl->gl_state == LM_ST_EXCLUSIVE) &&
+           (gh->gh_state == LM_ST_EXCLUSIVE))
+               error = gfs2_truncatei_resume(ip);
+
+       return error;
+}
+
+/**
+ * inode_go_unlock - operation done before an inode lock is unlocked by a
+ *                  process
+ * @gl: the glock
+ * @flags:
+ *
+ */
+
+static void inode_go_unlock(struct gfs2_holder *gh)
+{
+       struct gfs2_glock *gl = gh->gh_gl;
+       struct gfs2_inode *ip = gl->gl_object;
+
+       if (ip)
+               gfs2_meta_cache_flush(ip);
+}
+
+/**
+ * rgrp_go_demote_ok - Check to see if it's ok to unlock a RG's glock
+ * @gl: the glock
+ *
+ * Returns: 1 if it's ok
+ */
+
+static int rgrp_go_demote_ok(struct gfs2_glock *gl)
+{
+       return !gl->gl_aspace->i_mapping->nrpages;
+}
+
+/**
+ * rgrp_go_lock - operation done after an rgrp lock is locked by
+ *    a first holder on this node.
+ * @gl: the glock
+ * @flags:
+ *
+ * Returns: errno
+ */
+
+static int rgrp_go_lock(struct gfs2_holder *gh)
+{
+       return gfs2_rgrp_bh_get(gh->gh_gl->gl_object);
+}
+
+/**
+ * rgrp_go_unlock - operation done before an rgrp lock is unlocked by
+ *    a last holder on this node.
+ * @gl: the glock
+ * @flags:
+ *
+ */
+
+static void rgrp_go_unlock(struct gfs2_holder *gh)
+{
+       gfs2_rgrp_bh_put(gh->gh_gl->gl_object);
+}
+
+/**
+ * trans_go_xmote_th - promote/demote the transaction glock
+ * @gl: the glock
+ * @state: the requested state
+ * @flags:
+ *
+ */
+
+static void trans_go_xmote_th(struct gfs2_glock *gl)
+{
+       struct gfs2_sbd *sdp = gl->gl_sbd;
+
+       if (gl->gl_state != LM_ST_UNLOCKED &&
+           test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
+               gfs2_meta_syncfs(sdp);
+               gfs2_log_shutdown(sdp);
+       }
+}
+
+/**
+ * trans_go_xmote_bh - After promoting/demoting the transaction glock
+ * @gl: the glock
+ *
+ */
+
+static void trans_go_xmote_bh(struct gfs2_glock *gl)
+{
+       struct gfs2_sbd *sdp = gl->gl_sbd;
+       struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
+       struct gfs2_glock *j_gl = ip->i_gl;
+       struct gfs2_log_header_host head;
+       int error;
+
+       if (gl->gl_state != LM_ST_UNLOCKED &&
+           test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
+               gfs2_meta_cache_flush(GFS2_I(sdp->sd_jdesc->jd_inode));
+               j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
+
+               error = gfs2_find_jhead(sdp->sd_jdesc, &head);
+               if (error)
+                       gfs2_consist(sdp);
+               if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
+                       gfs2_consist(sdp);
+
+               /*  Initialize some head of the log stuff  */
+               if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
+                       sdp->sd_log_sequence = head.lh_sequence + 1;
+                       gfs2_log_pointers_init(sdp, head.lh_blkno);
+               }
+       }
+}
+
+/**
+ * trans_go_drop_th - unlock the transaction glock
+ * @gl: the glock
+ *
+ * We want to sync the device even with localcaching.  Remember
+ * that localcaching journal replay only marks buffers dirty.
+ */
+
+static void trans_go_drop_th(struct gfs2_glock *gl)
+{
+       struct gfs2_sbd *sdp = gl->gl_sbd;
+
+       if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
+               gfs2_meta_syncfs(sdp);
+               gfs2_log_shutdown(sdp);
+       }
+}
+
+/**
+ * quota_go_demote_ok - Check to see if it's ok to unlock a quota glock
+ * @gl: the glock
+ *
+ * Returns: 1 if it's ok
+ */
+
+static int quota_go_demote_ok(struct gfs2_glock *gl)
+{
+       return !atomic_read(&gl->gl_lvb_count);
+}
+
+const struct gfs2_glock_operations gfs2_meta_glops = {
+       .go_xmote_th = meta_go_sync,
+       .go_drop_th = meta_go_sync,
+       .go_type = LM_TYPE_META,
+};
+
+const struct gfs2_glock_operations gfs2_inode_glops = {
+       .go_xmote_th = inode_go_xmote_th,
+       .go_xmote_bh = inode_go_xmote_bh,
+       .go_drop_th = inode_go_drop_th,
+       .go_inval = inode_go_inval,
+       .go_demote_ok = inode_go_demote_ok,
+       .go_lock = inode_go_lock,
+       .go_unlock = inode_go_unlock,
+       .go_type = LM_TYPE_INODE,
+};
+
+const struct gfs2_glock_operations gfs2_rgrp_glops = {
+       .go_xmote_th = meta_go_sync,
+       .go_drop_th = meta_go_sync,
+       .go_inval = meta_go_inval,
+       .go_demote_ok = rgrp_go_demote_ok,
+       .go_lock = rgrp_go_lock,
+       .go_unlock = rgrp_go_unlock,
+       .go_type = LM_TYPE_RGRP,
+};
+
+const struct gfs2_glock_operations gfs2_trans_glops = {
+       .go_xmote_th = trans_go_xmote_th,
+       .go_xmote_bh = trans_go_xmote_bh,
+       .go_drop_th = trans_go_drop_th,
+       .go_type = LM_TYPE_NONDISK,
+};
+
+const struct gfs2_glock_operations gfs2_iopen_glops = {
+       .go_type = LM_TYPE_IOPEN,
+};
+
+const struct gfs2_glock_operations gfs2_flock_glops = {
+       .go_type = LM_TYPE_FLOCK,
+};
+
+const struct gfs2_glock_operations gfs2_nondisk_glops = {
+       .go_type = LM_TYPE_NONDISK,
+};
+
+const struct gfs2_glock_operations gfs2_quota_glops = {
+       .go_demote_ok = quota_go_demote_ok,
+       .go_type = LM_TYPE_QUOTA,
+};
+
+const struct gfs2_glock_operations gfs2_journal_glops = {
+       .go_type = LM_TYPE_JOURNAL,
+};
+