Whamcloud - gitweb
LU-13783 procfs: fix improper prop_ops fields
[fs/lustre-release.git] / lustre / llite / vvp_dev.c
index 6ffd719..7db2a47 100644 (file)
@@ -27,7 +27,6 @@
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
  *
  * cl_device and cl_device_type implementation for VVP layer.
  *
@@ -40,6 +39,7 @@
 #include <obd.h>
 #include "llite_internal.h"
 #include "vvp_internal.h"
+#include <linux/kallsyms.h>
 
 /*****************************************************************************
  *
@@ -265,6 +265,23 @@ struct lu_device_type vvp_device_type = {
         .ldt_ctx_tags = LCT_CL_THREAD
 };
 
+#ifndef HAVE_ACCOUNT_PAGE_DIRTIED_EXPORT
+unsigned int (*vvp_account_page_dirtied)(struct page *page,
+                                        struct address_space *mapping);
+
+unsigned int ll_account_page_dirtied(struct page *page,
+                                    struct address_space *mapping)
+{
+       /* must use __set_page_dirty, which means unlocking and
+        * relocking, which hurts performance.
+        */
+       ll_xa_unlock(&mapping->i_pages);
+       __set_page_dirty(page, mapping, 0);
+       ll_xa_lock(&mapping->i_pages);
+       return 0;
+}
+#endif
+
 /**
  * A mutex serializing calls to vvp_inode_fini() under extreme memory
  * pressure, when environments cannot be allocated.
@@ -281,6 +298,16 @@ int vvp_global_init(void)
        if (rc != 0)
                goto out_kmem;
 
+#ifndef HAVE_ACCOUNT_PAGE_DIRTIED_EXPORT
+       /*
+        * Kernel v5.2-5678-gac1c3e4 no longer exports account_page_dirtied
+        */
+       vvp_account_page_dirtied = (void *)
+               cfs_kallsyms_lookup_name("account_page_dirtied");
+       if (!vvp_account_page_dirtied)
+               vvp_account_page_dirtied = ll_account_page_dirtied;
+#endif
+
        return 0;
 
 out_kmem:
@@ -355,147 +382,61 @@ int cl_sb_fini(struct super_block *sb)
 
 /****************************************************************************
  *
- * /proc/fs/lustre/llite/$MNT/dump_page_cache
+ * debugfs/lustre/llite/$MNT/dump_page_cache
  *
  ****************************************************************************/
 
-/*
- * To represent contents of a page cache as a byte stream, following
- * information if encoded in 64bit offset:
- *
- *       - file hash bucket in lu_site::ls_hash[]       28bits
- *
- *       - how far file is from bucket head              4bits
- *
- *       - page index                                   32bits
- *
- * First two data identify a file in the cache uniquely.
- */
-
-#define PGC_OBJ_SHIFT (32 + 4)
-#define PGC_DEPTH_SHIFT (32)
-
-struct vvp_pgcache_id {
-        unsigned                 vpi_bucket;
-        unsigned                 vpi_depth;
-        uint32_t                 vpi_index;
-
-        unsigned                 vpi_curdep;
-        struct lu_object_header *vpi_obj;
-};
-
 struct vvp_seq_private {
        struct ll_sb_info       *vsp_sbi;
        struct lu_env           *vsp_env;
        u16                     vsp_refcheck;
        struct cl_object        *vsp_clob;
+       struct rhashtable_iter  vsp_iter;
+       u32                     vsp_page_index;
+       /*
+        * prev_pos is the 'pos' of the last object returned
+        * by ->start of ->next.
+        */
+       loff_t                  vvp_prev_pos;
 };
 
-static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
-{
-       CLASSERT(sizeof(pos) == sizeof(__u64));
-
-       id->vpi_index  = pos & 0xffffffff;
-       id->vpi_depth  = (pos >> PGC_DEPTH_SHIFT) & 0xf;
-       id->vpi_bucket = ((unsigned long long)pos >> PGC_OBJ_SHIFT);
-}
-
-static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
-{
-        return
-                ((__u64)id->vpi_index) |
-                ((__u64)id->vpi_depth  << PGC_DEPTH_SHIFT) |
-                ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
-}
-
-static int vvp_pgcache_obj_get(struct cfs_hash *hs, struct cfs_hash_bd *bd,
-                              struct hlist_node *hnode, void *data)
+static struct page *vvp_pgcache_current(struct vvp_seq_private *priv)
 {
-        struct vvp_pgcache_id   *id  = data;
-        struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
+       struct lu_device *dev = &priv->vsp_sbi->ll_cl->cd_lu_dev;
+       struct lu_object_header *h;
+       struct page *vmpage = NULL;
 
-        if (id->vpi_curdep-- > 0)
-                return 0; /* continue */
+       rhashtable_walk_start(&priv->vsp_iter);
+       while ((h = rhashtable_walk_next(&priv->vsp_iter)) != NULL) {
+               struct inode *inode;
+               int nr;
 
-        if (lu_object_is_dying(hdr))
-                return 1;
+               if (!priv->vsp_clob) {
+                       struct lu_object *lu_obj;
 
-        cfs_hash_get(hs, hnode);
-        id->vpi_obj = hdr;
-        return 1;
-}
-
-static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
-                                        struct lu_device *dev,
-                                        struct vvp_pgcache_id *id)
-{
-       LASSERT(lu_device_is_cl(dev));
+                       lu_obj = lu_object_get_first(h, dev);
+                       if (!lu_obj)
+                               continue;
 
-       id->vpi_depth &= 0xf;
-       id->vpi_obj    = NULL;
-       id->vpi_curdep = id->vpi_depth;
-
-       cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
-                               vvp_pgcache_obj_get, id);
-       if (id->vpi_obj != NULL) {
-               struct lu_object *lu_obj;
-
-               lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
-               if (lu_obj != NULL) {
+                       priv->vsp_clob = lu2cl(lu_obj);
                        lu_object_ref_add(lu_obj, "dump", current);
-                       return lu2cl(lu_obj);
+                       priv->vsp_page_index = 0;
                }
-               lu_object_put(env, lu_object_top(id->vpi_obj));
-
-       } else if (id->vpi_curdep > 0) {
-               id->vpi_depth = 0xf;
-       }
-       return NULL;
-}
 
-static struct page *vvp_pgcache_find(const struct lu_env *env,
-                                    struct lu_device *dev,
-                                    struct cl_object **clobp, loff_t *pos)
-{
-       struct cl_object     *clob;
-       struct lu_site       *site;
-       struct vvp_pgcache_id id;
-
-       site = dev->ld_site;
-       vvp_pgcache_id_unpack(*pos, &id);
-
-       while (1) {
-               if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
-                       return NULL;
-               clob = vvp_pgcache_obj(env, dev, &id);
-               if (clob) {
-                       struct inode *inode = vvp_object_inode(clob);
-                       struct page *vmpage;
-                       int nr;
-
-                       nr = find_get_pages_contig(inode->i_mapping,
-                                                  id.vpi_index, 1, &vmpage);
-                       if (nr > 0) {
-                               id.vpi_index = vmpage->index;
-                               /* Cant support over 16T file */
-                               if (vmpage->index <= 0xffffffff) {
-                                       *clobp = clob;
-                                       *pos = vvp_pgcache_id_pack(&id);
-                                       return vmpage;
-                               }
-                               put_page(vmpage);
-                       }
-
-                       lu_object_ref_del(&clob->co_lu, "dump", current);
-                       cl_object_put(env, clob);
+               inode = vvp_object_inode(priv->vsp_clob);
+               nr = find_get_pages_contig(inode->i_mapping,
+                                          priv->vsp_page_index, 1, &vmpage);
+               if (nr > 0) {
+                       priv->vsp_page_index = vmpage->index;
+                       break;
                }
-               /* to the next object. */
-               ++id.vpi_depth;
-               id.vpi_depth &= 0xf;
-               if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
-                       return NULL;
-               id.vpi_index = 0;
+               lu_object_ref_del(&priv->vsp_clob->co_lu, "dump", current);
+               cl_object_put(priv->vsp_env, priv->vsp_clob);
+               priv->vsp_clob = NULL;
+               priv->vsp_page_index = 0;
        }
+       rhashtable_walk_stop(&priv->vsp_iter);
+       return vmpage;
 }
 
 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
@@ -553,35 +494,58 @@ static int vvp_pgcache_show(struct seq_file *f, void *v)
        } else {
                seq_puts(f, "missing\n");
        }
-       lu_object_ref_del(&priv->vsp_clob->co_lu, "dump", current);
-       cl_object_put(priv->vsp_env, priv->vsp_clob);
 
        return 0;
 }
 
+static void vvp_pgcache_rewind(struct vvp_seq_private *priv)
+{
+       if (priv->vvp_prev_pos) {
+               struct lu_site *s = priv->vsp_sbi->ll_cl->cd_lu_dev.ld_site;
+
+               rhashtable_walk_exit(&priv->vsp_iter);
+               rhashtable_walk_enter(&s->ls_obj_hash, &priv->vsp_iter);
+               priv->vvp_prev_pos = 0;
+               if (priv->vsp_clob) {
+                       lu_object_ref_del(&priv->vsp_clob->co_lu, "dump",
+                                         current);
+                       cl_object_put(priv->vsp_env, priv->vsp_clob);
+               }
+               priv->vsp_clob = NULL;
+       }
+}
+
+static struct page *vvp_pgcache_next_page(struct vvp_seq_private *priv)
+{
+       priv->vsp_page_index += 1;
+       return vvp_pgcache_current(priv);
+}
+
 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 {
        struct vvp_seq_private *priv = f->private;
-       struct page *ret;
-
-       if (priv->vsp_sbi->ll_site->ls_obj_hash->hs_cur_bits >
-           64 - PGC_OBJ_SHIFT)
-               ret = ERR_PTR(-EFBIG);
-       else
-               ret = vvp_pgcache_find(priv->vsp_env,
-                                      &priv->vsp_sbi->ll_cl->cd_lu_dev,
-                                      &priv->vsp_clob, pos);
-       return ret;
+
+       if (*pos == 0) {
+               vvp_pgcache_rewind(priv);
+       } else if (*pos == priv->vvp_prev_pos) {
+               /* Return the current item */;
+       } else {
+               WARN_ON(*pos != priv->vvp_prev_pos + 1);
+               priv->vsp_page_index += 1;
+       }
+
+       priv->vvp_prev_pos = *pos;
+       return vvp_pgcache_current(priv);
 }
 
 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 {
        struct vvp_seq_private *priv = f->private;
 
+       WARN_ON(*pos != priv->vvp_prev_pos);
        *pos += 1;
-       return vvp_pgcache_find(priv->vsp_env,
-                               &priv->vsp_sbi->ll_cl->cd_lu_dev,
-                               &priv->vsp_clob, pos);
+       priv->vvp_prev_pos = *pos;
+       return vvp_pgcache_next_page(priv);
 }
 
 static void vvp_pgcache_stop(struct seq_file *f, void *v)
@@ -589,29 +553,35 @@ static void vvp_pgcache_stop(struct seq_file *f, void *v)
         /* Nothing to do */
 }
 
-static struct seq_operations vvp_pgcache_ops = {
-        .start = vvp_pgcache_start,
-        .next  = vvp_pgcache_next,
-        .stop  = vvp_pgcache_stop,
-        .show  = vvp_pgcache_show
+static const struct seq_operations vvp_pgcache_ops = {
+       .start = vvp_pgcache_start,
+       .next  = vvp_pgcache_next,
+       .stop  = vvp_pgcache_stop,
+       .show  = vvp_pgcache_show
 };
 
 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 {
        struct vvp_seq_private *priv;
+       struct lu_site *s;
 
        priv = __seq_open_private(filp, &vvp_pgcache_ops, sizeof(*priv));
        if (!priv)
                return -ENOMEM;
 
-       priv->vsp_sbi = PDE_DATA(inode);
+       priv->vsp_sbi = inode->i_private;
        priv->vsp_env = cl_env_get(&priv->vsp_refcheck);
+       priv->vsp_clob = NULL;
        if (IS_ERR(priv->vsp_env)) {
                int err = PTR_ERR(priv->vsp_env);
 
                seq_release_private(inode, filp);
                return err;
        }
+
+       s = priv->vsp_sbi->ll_cl->cd_lu_dev.ld_site;
+       rhashtable_walk_enter(&s->ls_obj_hash, &priv->vsp_iter);
+
        return 0;
 }
 
@@ -620,7 +590,12 @@ static int vvp_dump_pgcache_seq_release(struct inode *inode, struct file *file)
        struct seq_file *seq = file->private_data;
        struct vvp_seq_private *priv = seq->private;
 
+       if (priv->vsp_clob) {
+               lu_object_ref_del(&priv->vsp_clob->co_lu, "dump", current);
+               cl_object_put(priv->vsp_env, priv->vsp_clob);
+       }
        cl_env_put(priv->vsp_env, &priv->vsp_refcheck);
+       rhashtable_walk_exit(&priv->vsp_iter);
        return seq_release_private(inode, file);
 }