Whamcloud - gitweb
LU-2139 osc: Track and limit "unstable" pages
[fs/lustre-release.git] / lustre / llite / lproc_llite.c
index 5a362a3..5f995af 100644 (file)
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -16,8 +14,8 @@
  * in the LICENSE file that accompanied this code).
  *
  * You should have received a copy of the GNU General Public License
- * version 2 along with this program; If not, see [sun.com URL with a
- * copy of GPLv2].
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  *
  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  * CA 95054 USA or visit www.sun.com if you need additional information or
  * GPL HEADER END
  */
 /*
- * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
+ *
+ * Copyright (c) 2012, 2013, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -47,8 +47,7 @@ struct proc_dir_entry *proc_lustre_fs_root;
 
 #ifdef LPROCFS
 /* /proc/lustre/llite mount point registration */
-struct file_operations llite_dump_pgcache_fops;
-struct file_operations ll_ra_stats_fops;
+extern struct file_operations vvp_dump_pgcache_file_ops;
 struct file_operations ll_rw_extents_stats_fops;
 struct file_operations ll_rw_extents_stats_pp_fops;
 struct file_operations ll_rw_offset_stats_fops;
@@ -61,7 +60,8 @@ static int ll_rd_blksize(char *page, char **start, off_t off, int count,
         int rc;
 
         LASSERT(sb != NULL);
-        rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
+        rc = ll_statfs_internal(sb, &osfs,
+                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
                                 OBD_STATFS_NODELAY);
         if (!rc) {
               *eof = 1;
@@ -79,7 +79,8 @@ static int ll_rd_kbytestotal(char *page, char **start, off_t off, int count,
         int rc;
 
         LASSERT(sb != NULL);
-        rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
+        rc = ll_statfs_internal(sb, &osfs,
+                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
                                 OBD_STATFS_NODELAY);
         if (!rc) {
                 __u32 blk_size = osfs.os_bsize >> 10;
@@ -103,7 +104,8 @@ static int ll_rd_kbytesfree(char *page, char **start, off_t off, int count,
         int rc;
 
         LASSERT(sb != NULL);
-        rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
+        rc = ll_statfs_internal(sb, &osfs,
+                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
                                 OBD_STATFS_NODELAY);
         if (!rc) {
                 __u32 blk_size = osfs.os_bsize >> 10;
@@ -126,7 +128,8 @@ static int ll_rd_kbytesavail(char *page, char **start, off_t off, int count,
         int rc;
 
         LASSERT(sb != NULL);
-        rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
+        rc = ll_statfs_internal(sb, &osfs,
+                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
                                 OBD_STATFS_NODELAY);
         if (!rc) {
                 __u32 blk_size = osfs.os_bsize >> 10;
@@ -149,7 +152,8 @@ static int ll_rd_filestotal(char *page, char **start, off_t off, int count,
         int rc;
 
         LASSERT(sb != NULL);
-        rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
+        rc = ll_statfs_internal(sb, &osfs,
+                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
                                 OBD_STATFS_NODELAY);
         if (!rc) {
                  *eof = 1;
@@ -166,7 +170,8 @@ static int ll_rd_filesfree(char *page, char **start, off_t off, int count,
         int rc;
 
         LASSERT(sb != NULL);
-        rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
+        rc = ll_statfs_internal(sb, &osfs,
+                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
                                 OBD_STATFS_NODELAY);
         if (!rc) {
                  *eof = 1;
@@ -213,137 +218,291 @@ static int ll_rd_sb_uuid(char *page, char **start, off_t off, int count,
         return snprintf(page, count, "%s\n", ll_s2sbi(sb)->ll_sb_uuid.uuid);
 }
 
-static int ll_rd_max_readahead_mb(char *page, char **start, off_t off,
-                                   int count, int *eof, void *data)
+static int ll_rd_xattr_cache(char *page, char **start, off_t off,
+                               int count, int *eof, void *data)
 {
-        struct super_block *sb = data;
-        struct ll_sb_info *sbi = ll_s2sbi(sb);
-        long pages_number;
-        int mult;
+       struct super_block *sb = (struct super_block *)data;
+       struct ll_sb_info *sbi = ll_s2sbi(sb);
+       int rc;
 
-        spin_lock(&sbi->ll_lock);
-        pages_number = sbi->ll_ra_info.ra_max_pages;
-        spin_unlock(&sbi->ll_lock);
+       rc = snprintf(page, count, "%u\n", sbi->ll_xattr_cache_enabled);
 
-        mult = 1 << (20 - PAGE_CACHE_SHIFT);
-        return lprocfs_read_frac_helper(page, count, pages_number, mult);
+       return rc;
 }
 
-static int ll_wr_max_readahead_mb(struct file *file, const char *buffer,
-                                   unsigned long count, void *data)
+static int ll_wr_xattr_cache(struct file *file, const char *buffer,
+                               unsigned long count, void *data)
 {
-        struct super_block *sb = data;
-        struct ll_sb_info *sbi = ll_s2sbi(sb);
-        int mult, rc, pages_number;
+       struct super_block *sb = (struct super_block *)data;
+       struct ll_sb_info *sbi = ll_s2sbi(sb);
+       int val, rc;
 
-        mult = 1 << (20 - CFS_PAGE_SHIFT);
-        rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
-        if (rc)
-                return rc;
+       rc = lprocfs_write_helper(buffer, count, &val);
+       if (rc)
+               return rc;
 
-        if (pages_number < 0 || pages_number > num_physpages / 2) {
-                CERROR("can't set file readahead more than %lu MB\n",
-                        num_physpages >> (20 - CFS_PAGE_SHIFT + 1)); /*1/2 of RAM*/
-                return -ERANGE;
-        }
+       if (val != 0 && val != 1)
+               return -ERANGE;
 
-        spin_lock(&sbi->ll_lock);
-        sbi->ll_ra_info.ra_max_pages = pages_number;
-        spin_unlock(&sbi->ll_lock);
+       if (val == 1 && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
+               return -ENOTSUPP;
 
-        return count;
+       sbi->ll_xattr_cache_enabled = val;
+
+       return count;
 }
 
-static int ll_rd_max_read_ahead_whole_mb(char *page, char **start, off_t off,
-                                       int count, int *eof, void *data)
+static int ll_rd_site_stats(char *page, char **start, off_t off,
+                            int count, int *eof, void *data)
 {
         struct super_block *sb = data;
-        struct ll_sb_info *sbi = ll_s2sbi(sb);
-        long pages_number;
-        int mult;
-
-        spin_lock(&sbi->ll_lock);
-        pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
-        spin_unlock(&sbi->ll_lock);
 
-        mult = 1 << (20 - CFS_PAGE_SHIFT);
-        return lprocfs_read_frac_helper(page, count, pages_number, mult);
+        /*
+         * See description of statistical counters in struct cl_site, and
+         * struct lu_site.
+         */
+        return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site),
+                                   page, count);
 }
 
-static int ll_wr_max_read_ahead_whole_mb(struct file *file, const char *buffer,
-                                       unsigned long count, void *data)
+static int ll_rd_max_readahead_mb(char *page, char **start, off_t off,
+                                   int count, int *eof, void *data)
 {
-        struct super_block *sb = data;
-        struct ll_sb_info *sbi = ll_s2sbi(sb);
-        int mult, rc, pages_number;
-
-        mult = 1 << (20 - CFS_PAGE_SHIFT);
-        rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
-        if (rc)
-                return rc;
+       struct super_block *sb = data;
+       struct ll_sb_info *sbi = ll_s2sbi(sb);
+       long pages_number;
+       int mult;
 
-        /* Cap this at the current max readahead window size, the readahead
-         * algorithm does this anyway so it's pointless to set it larger. */
-        if (pages_number < 0 || pages_number > sbi->ll_ra_info.ra_max_pages) {
-                CERROR("can't set max_read_ahead_whole_mb more than "
-                       "max_read_ahead_mb: %lu\n",
-                       sbi->ll_ra_info.ra_max_pages >> (20 - CFS_PAGE_SHIFT));
-                return -ERANGE;
-        }
+       spin_lock(&sbi->ll_lock);
+       pages_number = sbi->ll_ra_info.ra_max_pages;
+       spin_unlock(&sbi->ll_lock);
 
-        spin_lock(&sbi->ll_lock);
-        sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
-        spin_unlock(&sbi->ll_lock);
+       mult = 1 << (20 - PAGE_CACHE_SHIFT);
+       return lprocfs_read_frac_helper(page, count, pages_number, mult);
+}
 
-        return count;
+static int ll_wr_max_readahead_mb(struct file *file, const char *buffer,
+                                 unsigned long count, void *data)
+{
+       struct super_block *sb = data;
+       struct ll_sb_info *sbi = ll_s2sbi(sb);
+       int mult, rc, pages_number;
+
+       mult = 1 << (20 - PAGE_CACHE_SHIFT);
+       rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
+       if (rc)
+               return rc;
+
+       if (pages_number < 0 || pages_number > totalram_pages / 2) {
+               /* 1/2 of RAM */
+               CERROR("can't set file readahead more than %lu MB\n",
+                      totalram_pages >> (20 - PAGE_CACHE_SHIFT + 1));
+               return -ERANGE;
+       }
+
+       spin_lock(&sbi->ll_lock);
+       sbi->ll_ra_info.ra_max_pages = pages_number;
+       spin_unlock(&sbi->ll_lock);
+
+       return count;
 }
 
-static int ll_rd_max_cached_mb(char *page, char **start, off_t off,
-                               int count, int *eof, void *data)
+static int ll_rd_max_readahead_per_file_mb(char *page, char **start, off_t off,
+                                           int count, int *eof, void *data)
 {
-        struct super_block *sb = data;
-        struct ll_sb_info *sbi = ll_s2sbi(sb);
-        long pages_number;
-        int mult;
+       struct super_block *sb = data;
+       struct ll_sb_info *sbi = ll_s2sbi(sb);
+       long pages_number;
+       int mult;
 
-        spin_lock(&sbi->ll_lock);
-        pages_number = sbi->ll_async_page_max;
-        spin_unlock(&sbi->ll_lock);
+       spin_lock(&sbi->ll_lock);
+       pages_number = sbi->ll_ra_info.ra_max_pages_per_file;
+       spin_unlock(&sbi->ll_lock);
 
-        mult = 1 << (20 - CFS_PAGE_SHIFT);
-        return lprocfs_read_frac_helper(page, count, pages_number, mult);;
+       mult = 1 << (20 - PAGE_CACHE_SHIFT);
+       return lprocfs_read_frac_helper(page, count, pages_number, mult);
 }
 
-static int ll_wr_max_cached_mb(struct file *file, const char *buffer,
-                                  unsigned long count, void *data)
+static int ll_wr_max_readahead_per_file_mb(struct file *file, const char *buffer,
+                                          unsigned long count, void *data)
 {
         struct super_block *sb = data;
         struct ll_sb_info *sbi = ll_s2sbi(sb);
         int mult, rc, pages_number;
 
-        mult = 1 << (20 - CFS_PAGE_SHIFT);
+       mult = 1 << (20 - PAGE_CACHE_SHIFT);
         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
         if (rc)
                 return rc;
 
-        if (pages_number < 0 || pages_number > num_physpages) {
-                CERROR("can't set max cache more than %lu MB\n",
-                        num_physpages >> (20 - CFS_PAGE_SHIFT));
+        if (pages_number < 0 ||
+                pages_number > sbi->ll_ra_info.ra_max_pages) {
+                CERROR("can't set file readahead more than"
+                       "max_read_ahead_mb %lu MB\n",
+                       sbi->ll_ra_info.ra_max_pages);
                 return -ERANGE;
         }
 
-        spin_lock(&sbi->ll_lock);
-        sbi->ll_async_page_max = pages_number ;
-        spin_unlock(&sbi->ll_lock);
-        
-        if (!sbi->ll_dt_exp)
-                /* Not set up yet, don't call llap_shrink_cache */
-                return count;
+       spin_lock(&sbi->ll_lock);
+       sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
+       spin_unlock(&sbi->ll_lock);
 
-        if (sbi->ll_async_page_count >= sbi->ll_async_page_max)
-                llap_shrink_cache(sbi, 0);
+       return count;
+}
 
-        return count;
+static int ll_rd_max_read_ahead_whole_mb(char *page, char **start, off_t off,
+                                         int count, int *eof, void *data)
+{
+       struct super_block *sb = data;
+       struct ll_sb_info *sbi = ll_s2sbi(sb);
+       long pages_number;
+       int mult;
+
+       spin_lock(&sbi->ll_lock);
+       pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
+       spin_unlock(&sbi->ll_lock);
+
+       mult = 1 << (20 - PAGE_CACHE_SHIFT);
+       return lprocfs_read_frac_helper(page, count, pages_number, mult);
+}
+
+static int ll_wr_max_read_ahead_whole_mb(struct file *file, const char *buffer,
+                                        unsigned long count, void *data)
+{
+       struct super_block *sb = data;
+       struct ll_sb_info *sbi = ll_s2sbi(sb);
+       int mult, rc, pages_number;
+
+       mult = 1 << (20 - PAGE_CACHE_SHIFT);
+       rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
+       if (rc)
+               return rc;
+
+       /* Cap this at the current max readahead window size, the readahead
+        * algorithm does this anyway so it's pointless to set it larger. */
+       if (pages_number < 0 ||
+           pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
+               CERROR("can't set max_read_ahead_whole_mb more than "
+                      "max_read_ahead_per_file_mb: %lu\n",
+                       sbi->ll_ra_info.ra_max_pages_per_file >>
+                       (20 - PAGE_CACHE_SHIFT));
+               return -ERANGE;
+       }
+
+       spin_lock(&sbi->ll_lock);
+       sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
+       spin_unlock(&sbi->ll_lock);
+
+       return count;
+}
+
+static int ll_rd_max_cached_mb(char *page, char **start, off_t off,
+                               int count, int *eof, void *data)
+{
+       struct super_block     *sb    = data;
+       struct ll_sb_info      *sbi   = ll_s2sbi(sb);
+       struct cl_client_cache *cache = &sbi->ll_cache;
+       int shift = 20 - PAGE_CACHE_SHIFT;
+       int max_cached_mb;
+       int unused_mb;
+
+       *eof = 1;
+       max_cached_mb = cache->ccc_lru_max >> shift;
+       unused_mb = cfs_atomic_read(&cache->ccc_lru_left) >> shift;
+       return snprintf(page, count,
+                       "users: %d\n"
+                       "max_cached_mb: %d\n"
+                       "used_mb: %d\n"
+                       "unused_mb: %d\n"
+                       "reclaim_count: %u\n",
+                       cfs_atomic_read(&cache->ccc_users),
+                       max_cached_mb,
+                       max_cached_mb - unused_mb,
+                       unused_mb,
+                       cache->ccc_lru_shrinkers);
+}
+
+static int ll_wr_max_cached_mb(struct file *file, const char *buffer,
+                               unsigned long count, void *data)
+{
+       struct super_block *sb = data;
+       struct ll_sb_info *sbi = ll_s2sbi(sb);
+       struct cl_client_cache *cache = &sbi->ll_cache;
+       int mult, rc, pages_number;
+       int diff = 0;
+       int nrpages = 0;
+       ENTRY;
+
+       mult = 1 << (20 - PAGE_CACHE_SHIFT);
+       buffer = lprocfs_find_named_value(buffer, "max_cached_mb:", &count);
+       rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
+       if (rc)
+               RETURN(rc);
+
+       if (pages_number < 0 || pages_number > totalram_pages) {
+               CERROR("%s: can't set max cache more than %lu MB\n",
+                      ll_get_fsname(sb, NULL, 0),
+                      totalram_pages >> (20 - PAGE_CACHE_SHIFT));
+               RETURN(-ERANGE);
+       }
+
+       if (sbi->ll_dt_exp == NULL) /* being initialized */
+               GOTO(out, rc = 0);
+
+       spin_lock(&sbi->ll_lock);
+       diff = pages_number - cache->ccc_lru_max;
+       spin_unlock(&sbi->ll_lock);
+
+       /* easy - add more LRU slots. */
+       if (diff >= 0) {
+               cfs_atomic_add(diff, &cache->ccc_lru_left);
+               GOTO(out, rc = 0);
+       }
+
+       diff = -diff;
+       while (diff > 0) {
+               int tmp;
+
+               /* reduce LRU budget from free slots. */
+               do {
+                       int ov, nv;
+
+                       ov = cfs_atomic_read(&cache->ccc_lru_left);
+                       if (ov == 0)
+                               break;
+
+                       nv = ov > diff ? ov - diff : 0;
+                       rc = cfs_atomic_cmpxchg(&cache->ccc_lru_left, ov, nv);
+                       if (likely(ov == rc)) {
+                               diff -= ov - nv;
+                               nrpages += ov - nv;
+                               break;
+                       }
+               } while (1);
+
+               if (diff <= 0)
+                       break;
+
+               /* difficult - have to ask OSCs to drop LRU slots. */
+               tmp = diff << 1;
+               rc = obd_set_info_async(NULL, sbi->ll_dt_exp,
+                               sizeof(KEY_CACHE_LRU_SHRINK),
+                               KEY_CACHE_LRU_SHRINK,
+                               sizeof(tmp), &tmp, NULL);
+               if (rc < 0)
+                       break;
+       }
+
+out:
+       if (rc >= 0) {
+               spin_lock(&sbi->ll_lock);
+               cache->ccc_lru_max = pages_number;
+               spin_unlock(&sbi->ll_lock);
+               rc = count;
+       } else {
+               cfs_atomic_add(nrpages, &cache->ccc_lru_left);
+       }
+       return rc;
 }
 
 static int ll_rd_checksum(char *page, char **start, off_t off,
@@ -375,7 +534,7 @@ static int ll_wr_checksum(struct file *file, const char *buffer,
         else
                 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
 
-        rc = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
+        rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
                                 KEY_CHECKSUM, sizeof(val), &val, NULL);
         if (rc)
                 CWARN("Failed to set OSC checksum flags: %d\n", rc);
@@ -404,7 +563,7 @@ static int ll_wr_max_rw_chunk(struct file *file, const char *buffer,
         return count;
 }
 
-static int ll_rd_track_id(char *page, int count, void *data, 
+static int ll_rd_track_id(char *page, int count, void *data,
                           enum stats_track_type type)
 {
         struct super_block *sb = data;
@@ -412,7 +571,7 @@ static int ll_rd_track_id(char *page, int count, void *data,
         if (ll_s2sbi(sb)->ll_stats_track_type == type) {
                 return snprintf(page, count, "%d\n",
                                 ll_s2sbi(sb)->ll_stats_track_id);
-        
+
         } else if (ll_s2sbi(sb)->ll_stats_track_type == STATS_TRACK_ALL) {
                 return snprintf(page, count, "0 (all)\n");
         } else {
@@ -470,7 +629,7 @@ static int ll_rd_track_gid(char *page, char **start, off_t off,
 
 static int ll_wr_track_gid(struct file *file, const char *buffer,
                           unsigned long count, void *data)
-{                                                                 
+{
         return (ll_wr_track_id(buffer, count, data, STATS_TRACK_GID));
 }
 
@@ -503,72 +662,142 @@ static int ll_wr_statahead_max(struct file *file, const char *buffer,
         return count;
 }
 
-static int ll_rd_statahead_stats(char *page, char **start, off_t off,
-                                 int count, int *eof, void *data)
+static int ll_rd_statahead_agl(char *page, char **start, off_t off,
+                               int count, int *eof, void *data)
 {
         struct super_block *sb = data;
         struct ll_sb_info *sbi = ll_s2sbi(sb);
 
-        return snprintf(page, count,
-                        "statahead wrong: %u\n"
-                        "statahead total: %u\n"
-                        "ls blocked:      %llu\n"
-                        "ls cached:       %llu\n"
-                        "hit count:       %llu\n"
-                        "miss count:      %llu\n",
-                        sbi->ll_sa_wrong,
-                        sbi->ll_sa_total,
-                        sbi->ll_sa_blocked,
-                        sbi->ll_sa_cached,
-                        sbi->ll_sa_hit,
-                        sbi->ll_sa_miss);
+        return snprintf(page, count, "%u\n",
+                        sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
 }
 
-static int ll_rd_contention_time(char *page, char **start, off_t off,
-                                 int count, int *eof, void *data)
+static int ll_wr_statahead_agl(struct file *file, const char *buffer,
+                               unsigned long count, void *data)
 {
         struct super_block *sb = data;
+        struct ll_sb_info *sbi = ll_s2sbi(sb);
+        int val, rc;
 
-        *eof = 1;
-        return snprintf(page, count, "%u\n", ll_s2sbi(sb)->ll_contention_time);
+        rc = lprocfs_write_helper(buffer, count, &val);
+        if (rc)
+                return rc;
+
+        if (val)
+                sbi->ll_flags |= LL_SBI_AGL_ENABLED;
+        else
+                sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
 
+        return count;
 }
 
-static int ll_wr_contention_time(struct file *file, const char *buffer,
-                                 unsigned long count, void *data)
+static int ll_rd_statahead_stats(char *page, char **start, off_t off,
+                                 int count, int *eof, void *data)
 {
         struct super_block *sb = data;
         struct ll_sb_info *sbi = ll_s2sbi(sb);
 
-        return lprocfs_write_helper(buffer, count,&sbi->ll_contention_time) ?:
-                count;
+        return snprintf(page, count,
+                        "statahead total: %u\n"
+                        "statahead wrong: %u\n"
+                        "agl total: %u\n",
+                        atomic_read(&sbi->ll_sa_total),
+                        atomic_read(&sbi->ll_sa_wrong),
+                        atomic_read(&sbi->ll_agl_total));
 }
 
-static int ll_rd_lockless_truncate(char *page, char **start, off_t off,
-                                   int count, int *eof, void *data)
+static int ll_rd_lazystatfs(char *page, char **start, off_t off,
+                            int count, int *eof, void *data)
 {
         struct super_block *sb = data;
+        struct ll_sb_info *sbi = ll_s2sbi(sb);
 
-        *eof = 1;
         return snprintf(page, count, "%u\n",
-                        ll_s2sbi(sb)->ll_lockless_truncate_enable);
+                        (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
 }
 
-static int ll_wr_lockless_truncate(struct file *file, const char *buffer,
-                                   unsigned long count, void *data)
+static int ll_wr_lazystatfs(struct file *file, const char *buffer,
+                            unsigned long count, void *data)
 {
         struct super_block *sb = data;
         struct ll_sb_info *sbi = ll_s2sbi(sb);
+        int val, rc;
+
+        rc = lprocfs_write_helper(buffer, count, &val);
+        if (rc)
+                return rc;
+
+        if (val)
+                sbi->ll_flags |= LL_SBI_LAZYSTATFS;
+        else
+                sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
 
-        return lprocfs_write_helper(buffer, count,
-                                    &sbi->ll_lockless_truncate_enable)
-                                    ?: count;
+        return count;
+}
+
+static int ll_rd_maxea_size(char *page, char **start, off_t off,
+                            int count, int *eof, void *data)
+{
+        struct super_block *sb = data;
+        struct ll_sb_info *sbi = ll_s2sbi(sb);
+        unsigned int ealen;
+        int rc;
+
+        rc = ll_get_max_mdsize(sbi, &ealen);
+        if (rc)
+                return rc;
+
+        return snprintf(page, count, "%u\n", ealen);
+}
+
+static int ll_rd_sbi_flags(char *page, char **start, off_t off,
+                               int count, int *eof, void *data)
+{
+       const char *str[] = LL_SBI_FLAGS;
+       struct super_block *sb = data;
+       int flags = ll_s2sbi(sb)->ll_flags;
+       int i = 0;
+       int rc = 0;
+
+       while (flags != 0) {
+               if (ARRAY_SIZE(str) <= i) {
+                       CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
+                               "flags please.\n", ll_get_fsname(sb, NULL, 0));
+                       return -EINVAL;
+               }
+
+               if (flags & 0x1)
+                       rc += snprintf(page + rc, count - rc, "%s ", str[i]);
+               flags >>= 1;
+               ++i;
+       }
+       if (rc > 0)
+               rc += snprintf(page + rc, count - rc, "\b\n");
+       return rc;
+}
+
+static int ll_rd_unstable_stats(char *page, char **start, off_t off,
+                             int count, int *eof, void *data)
+{
+       struct super_block      *sb    = data;
+       struct ll_sb_info       *sbi   = ll_s2sbi(sb);
+       struct cl_client_cache  *cache = &sbi->ll_cache;
+       int pages, mb, rc;
+
+       pages = cfs_atomic_read(&cache->ccc_unstable_nr);
+       mb    = (pages * PAGE_CACHE_SIZE) >> 20;
+
+       rc = snprintf(page, count, "unstable_pages: %8d\n"
+                                  "unstable_mb:    %8d\n", pages, mb);
+
+       return rc;
 }
 
 static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
         { "uuid",         ll_rd_sb_uuid,          0, 0 },
         //{ "mntpt_path",   ll_rd_path,             0, 0 },
         { "fstype",       ll_rd_fstype,           0, 0 },
+        { "site",         ll_rd_site_stats,       0, 0 },
         { "blocksize",    ll_rd_blksize,          0, 0 },
         { "kbytestotal",  ll_rd_kbytestotal,      0, 0 },
         { "kbytesfree",   ll_rd_kbytesfree,       0, 0 },
@@ -579,6 +808,8 @@ static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
         //{ "filegroups",   lprocfs_rd_filegroups,  0, 0 },
         { "max_read_ahead_mb", ll_rd_max_readahead_mb,
                                ll_wr_max_readahead_mb, 0 },
+        { "max_read_ahead_per_file_mb", ll_rd_max_readahead_per_file_mb,
+                                        ll_wr_max_readahead_per_file_mb, 0 },
         { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
                                      ll_wr_max_read_ahead_whole_mb, 0 },
         { "max_cached_mb",    ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
@@ -588,10 +819,13 @@ static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
         { "stats_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
         { "stats_track_gid",  ll_rd_track_gid, ll_wr_track_gid, 0 },
         { "statahead_max",    ll_rd_statahead_max, ll_wr_statahead_max, 0 },
+        { "statahead_agl",    ll_rd_statahead_agl, ll_wr_statahead_agl, 0 },
         { "statahead_stats",  ll_rd_statahead_stats, 0, 0 },
-        { "contention_seconds", ll_rd_contention_time, ll_wr_contention_time, 0},
-        { "lockless_truncate", ll_rd_lockless_truncate,
-                               ll_wr_lockless_truncate, 0},
+        { "lazystatfs",       ll_rd_lazystatfs, ll_wr_lazystatfs, 0 },
+        { "max_easize",       ll_rd_maxea_size, 0, 0 },
+       { "sbi_flags",        ll_rd_sbi_flags, 0, 0 },
+       { "xattr_cache",      ll_rd_xattr_cache, ll_wr_xattr_cache, 0 },
+       { "unstable_stats",   ll_rd_unstable_stats, 0, 0},
         { 0 }
 };
 
@@ -605,14 +839,6 @@ struct llite_file_opcode {
         /* file operation */
         { LPROC_LL_DIRTY_HITS,     LPROCFS_TYPE_REGS, "dirty_pages_hits" },
         { LPROC_LL_DIRTY_MISSES,   LPROCFS_TYPE_REGS, "dirty_pages_misses" },
-        { LPROC_LL_WB_WRITEPAGE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
-                                   "writeback_from_writepage" },
-        { LPROC_LL_WB_PRESSURE,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
-                                   "writeback_from_pressure" },
-        { LPROC_LL_WB_OK,          LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
-                                   "writeback_ok_pages" },
-        { LPROC_LL_WB_FAIL,        LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
-                                   "writeback_failed_pages" },
         { LPROC_LL_READ_BYTES,     LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
                                    "read_bytes" },
         { LPROC_LL_WRITE_BYTES,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
@@ -621,36 +847,40 @@ struct llite_file_opcode {
                                    "brw_read" },
         { LPROC_LL_BRW_WRITE,      LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
                                    "brw_write" },
-
+        { LPROC_LL_OSC_READ,       LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
+                                   "osc_read" },
+        { LPROC_LL_OSC_WRITE,      LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
+                                   "osc_write" },
         { LPROC_LL_IOCTL,          LPROCFS_TYPE_REGS, "ioctl" },
         { LPROC_LL_OPEN,           LPROCFS_TYPE_REGS, "open" },
         { LPROC_LL_RELEASE,        LPROCFS_TYPE_REGS, "close" },
         { LPROC_LL_MAP,            LPROCFS_TYPE_REGS, "mmap" },
         { LPROC_LL_LLSEEK,         LPROCFS_TYPE_REGS, "seek" },
         { LPROC_LL_FSYNC,          LPROCFS_TYPE_REGS, "fsync" },
+        { LPROC_LL_READDIR,        LPROCFS_TYPE_REGS, "readdir" },
         /* inode operation */
         { LPROC_LL_SETATTR,        LPROCFS_TYPE_REGS, "setattr" },
         { LPROC_LL_TRUNC,          LPROCFS_TYPE_REGS, "truncate" },
-        { LPROC_LL_LOCKLESS_TRUNC, LPROCFS_TYPE_REGS, "lockless_truncate"},
         { LPROC_LL_FLOCK,          LPROCFS_TYPE_REGS, "flock" },
         { LPROC_LL_GETATTR,        LPROCFS_TYPE_REGS, "getattr" },
+        /* dir inode operation */
+        { LPROC_LL_CREATE,         LPROCFS_TYPE_REGS, "create" },
+        { LPROC_LL_LINK,           LPROCFS_TYPE_REGS, "link" },
+        { LPROC_LL_UNLINK,         LPROCFS_TYPE_REGS, "unlink" },
+        { LPROC_LL_SYMLINK,        LPROCFS_TYPE_REGS, "symlink" },
+        { LPROC_LL_MKDIR,          LPROCFS_TYPE_REGS, "mkdir" },
+        { LPROC_LL_RMDIR,          LPROCFS_TYPE_REGS, "rmdir" },
+        { LPROC_LL_MKNOD,          LPROCFS_TYPE_REGS, "mknod" },
+        { LPROC_LL_RENAME,         LPROCFS_TYPE_REGS, "rename" },
         /* special inode operation */
         { LPROC_LL_STAFS,          LPROCFS_TYPE_REGS, "statfs" },
         { LPROC_LL_ALLOC_INODE,    LPROCFS_TYPE_REGS, "alloc_inode" },
         { LPROC_LL_SETXATTR,       LPROCFS_TYPE_REGS, "setxattr" },
         { LPROC_LL_GETXATTR,       LPROCFS_TYPE_REGS, "getxattr" },
+       { LPROC_LL_GETXATTR_HITS,  LPROCFS_TYPE_REGS, "getxattr_hits" },
         { LPROC_LL_LISTXATTR,      LPROCFS_TYPE_REGS, "listxattr" },
         { LPROC_LL_REMOVEXATTR,    LPROCFS_TYPE_REGS, "removexattr" },
         { LPROC_LL_INODE_PERM,     LPROCFS_TYPE_REGS, "inode_permission" },
-        { LPROC_LL_DIRECT_READ,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
-                                   "direct_read" },
-        { LPROC_LL_DIRECT_WRITE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
-                                   "direct_write" },
-        { LPROC_LL_LOCKLESS_READ,  LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
-                                   "lockless_read_bytes" },
-        { LPROC_LL_LOCKLESS_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
-                                   "lockless_write_bytes" },
-
 };
 
 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
@@ -663,14 +893,30 @@ void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
                  sbi->ll_stats_track_id == current->pid)
                 lprocfs_counter_add(sbi->ll_stats, op, count);
         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
-                 sbi->ll_stats_track_id == current->p_pptr->pid)
-                lprocfs_counter_add(sbi->ll_stats, op, count);
-        else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
-                 sbi->ll_stats_track_id == current->gid)
+                 sbi->ll_stats_track_id == current->parent->pid)
                 lprocfs_counter_add(sbi->ll_stats, op, count);
+       else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
+                sbi->ll_stats_track_id == current_gid())
+               lprocfs_counter_add(sbi->ll_stats, op, count);
 }
 EXPORT_SYMBOL(ll_stats_ops_tally);
 
+static const char *ra_stat_string[] = {
+        [RA_STAT_HIT] = "hits",
+        [RA_STAT_MISS] = "misses",
+        [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
+        [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
+        [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
+        [RA_STAT_FAILED_MATCH] = "failed lock match",
+        [RA_STAT_DISCARDED] = "read but discarded",
+        [RA_STAT_ZERO_LEN] = "zero length file",
+        [RA_STAT_ZERO_WINDOW] = "zero size window",
+        [RA_STAT_EOF] = "read-ahead to EOF",
+        [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
+        [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
+};
+
+
 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
                                 struct super_block *sb, char *osc, char *mdc)
 {
@@ -695,12 +941,12 @@ int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
         len = strlen(lsi->lsi_lmd->lmd_profile);
         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
         if (ptr && (strcmp(ptr, "-client") == 0))
-                len -= 7; 
-        
+                len -= 7;
+
         /* Mount info */
         snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
                  lsi->lsi_lmd->lmd_profile, sb);
-        
+
         sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
         if (IS_ERR(sbi->ll_proc_root)) {
                 err = PTR_ERR(sbi->ll_proc_root);
@@ -708,17 +954,11 @@ int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
                 RETURN(err);
         }
 
-
         rc = lprocfs_seq_create(sbi->ll_proc_root, "dump_page_cache", 0444,
-                                &llite_dump_pgcache_fops, sbi);
+                                &vvp_dump_pgcache_file_ops, sbi);
         if (rc)
                 CWARN("Error adding the dump_page_cache file\n");
 
-        rc = lprocfs_seq_create(sbi->ll_proc_root, "read_ahead_stats", 0644,
-                                &ll_ra_stats_fops, sbi);
-        if (rc)
-                CWARN("Error adding the read_ahead_stats file\n");
-
         rc = lprocfs_seq_create(sbi->ll_proc_root, "extents_stats", 0644,
                                 &ll_rw_extents_stats_fops, sbi);
         if (rc)
@@ -735,8 +975,8 @@ int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
                 CWARN("Error adding the offset_stats file\n");
 
         /* File operations stats */
-        sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES, 
-                                            LPROCFS_STATS_FLAG_PERCPU);
+        sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
+                                            LPROCFS_STATS_FLAG_NONE);
         if (sbi->ll_stats == NULL)
                 GOTO(out, err = -ENOMEM);
         /* do counter init */
@@ -758,6 +998,20 @@ int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
         if (err)
                 GOTO(out, err);
 
+        sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
+                                               LPROCFS_STATS_FLAG_NONE);
+        if (sbi->ll_ra_stats == NULL)
+                GOTO(out, err = -ENOMEM);
+
+        for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
+                lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
+                                     ra_stat_string[id], "pages");
+        err = lprocfs_register_stats(sbi->ll_proc_root, "read_ahead_stats",
+                                     sbi->ll_ra_stats);
+        if (err)
+                GOTO(out, err);
+
+
         err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_llite_obd_vars, sb);
         if (err)
                 GOTO(out, err);
@@ -802,6 +1056,7 @@ int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
 out:
         if (err) {
                 lprocfs_remove(&sbi->ll_proc_root);
+                lprocfs_free_stats(&sbi->ll_ra_stats);
                 lprocfs_free_stats(&sbi->ll_stats);
         }
         RETURN(err);
@@ -811,253 +1066,12 @@ void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
 {
         if (sbi->ll_proc_root) {
                 lprocfs_remove(&sbi->ll_proc_root);
+                lprocfs_free_stats(&sbi->ll_ra_stats);
                 lprocfs_free_stats(&sbi->ll_stats);
         }
 }
 #undef MAX_STRING_SIZE
 
-#define seq_page_flag(seq, page, flag, has_flags) do {                  \
-                if (test_bit(PG_##flag, &(page)->flags)) {              \
-                        if (!has_flags)                                 \
-                                has_flags = 1;                          \
-                        else                                            \
-                                seq_putc(seq, '|');                     \
-                        seq_puts(seq, #flag);                           \
-                }                                                       \
-        } while(0);
-
-static void *llite_dump_pgcache_seq_start(struct seq_file *seq, loff_t *pos)
-{
-        struct ll_async_page *dummy_llap = seq->private;
-
-        if (dummy_llap->llap_magic == 2)
-                return NULL;
-
-        return (void *)1;
-}
-
-static int llite_dump_pgcache_seq_show(struct seq_file *seq, void *v)
-{
-        struct ll_async_page *llap, *dummy_llap = seq->private;
-        struct ll_sb_info *sbi = dummy_llap->llap_cookie;
-
-        /* 2.4 doesn't seem to have SEQ_START_TOKEN, so we implement
-         * it in our own state */
-        if (dummy_llap->llap_magic == 0) {
-                seq_printf(seq, "gener |  llap  cookie  origin wq du wb | page "
-                                "inode index count [ page flags ]\n");
-                return 0;
-        }
-
-        spin_lock(&sbi->ll_lock);
-
-        llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
-        if (llap != NULL)  {
-                int has_flags = 0;
-                struct page *page = llap->llap_page;
-
-                LASSERTF(llap->llap_origin < LLAP__ORIGIN_MAX, "%u\n",
-                         llap->llap_origin);
-
-                seq_printf(seq," %5lu | %p %p %s %s %s %s | %p %lu/%u(%p) "
-                           "%lu %u [",
-                           sbi->ll_pglist_gen,
-                           llap, llap->llap_cookie,
-                           llap_origins[llap->llap_origin],
-                           llap->llap_write_queued ? "wq" : "- ",
-                           llap->llap_defer_uptodate ? "du" : "- ",
-                           PageWriteback(page) ? "wb" : "-",
-                           page, page->mapping->host->i_ino,
-                           page->mapping->host->i_generation,
-                           page->mapping->host, page->index,
-                           page_count(page));
-                seq_page_flag(seq, page, locked, has_flags);
-                seq_page_flag(seq, page, error, has_flags);
-                seq_page_flag(seq, page, referenced, has_flags);
-                seq_page_flag(seq, page, uptodate, has_flags);
-                seq_page_flag(seq, page, dirty, has_flags);
-#if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,12))
-                seq_page_flag(seq, page, highmem, has_flags);
-#endif
-                seq_page_flag(seq, page, writeback, has_flags);
-                if (!has_flags)
-                        seq_puts(seq, "-]\n");
-                else
-                        seq_puts(seq, "]\n");
-        }
-
-        spin_unlock(&sbi->ll_lock);
-
-        return 0;
-}
-
-static void *llite_dump_pgcache_seq_next(struct seq_file *seq, void *v, 
-                                         loff_t *pos)
-{
-        struct ll_async_page *llap, *dummy_llap = seq->private;
-        struct ll_sb_info *sbi = dummy_llap->llap_cookie;
-
-        /* bail if we just displayed the banner */
-        if (dummy_llap->llap_magic == 0) {
-                dummy_llap->llap_magic = 1;
-                return dummy_llap;
-        }
-
-        /* we've just displayed the llap that is after us in the list.
-         * we advance to a position beyond it, returning null if there
-         * isn't another llap in the list beyond that new position. */
-        spin_lock(&sbi->ll_lock);
-        llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
-        list_del_init(&dummy_llap->llap_pglist_item);
-        if (llap) {
-                list_add(&dummy_llap->llap_pglist_item,&llap->llap_pglist_item);
-                llap =llite_pglist_next_llap(sbi,&dummy_llap->llap_pglist_item);
-        }
-        spin_unlock(&sbi->ll_lock);
-
-        ++*pos;
-        if (llap == NULL) {
-                dummy_llap->llap_magic = 2;
-                return NULL;
-        }
-        return dummy_llap;
-}
-
-static void null_stop(struct seq_file *seq, void *v)
-{
-}
-
-struct seq_operations llite_dump_pgcache_seq_sops = {
-        .start = llite_dump_pgcache_seq_start,
-        .stop = null_stop,
-        .next = llite_dump_pgcache_seq_next,
-        .show = llite_dump_pgcache_seq_show,
-};
-
-/* we're displaying llaps in a list_head list.  we don't want to hold a lock
- * while we walk the entire list, and we don't want to have to seek into
- * the right position in the list as an app advances with many syscalls.  we
- * allocate a dummy llap and hang it off file->private.  its position in
- * the list records where the app is currently displaying.  this way our
- * seq .start and .stop don't actually do anything.  .next returns null
- * when the dummy hits the end of the list which eventually leads to .release
- * where we tear down.  this kind of displaying is super-racey, so we put
- * a generation counter on the list so the output shows when the list
- * changes between reads.
- */
-static int llite_dump_pgcache_seq_open(struct inode *inode, struct file *file)
-{
-        struct proc_dir_entry *dp = PDE(inode);
-        struct ll_async_page *dummy_llap;
-        struct seq_file *seq;
-        struct ll_sb_info *sbi = dp->data;
-        int rc = -ENOMEM;
-
-        LPROCFS_ENTRY_AND_CHECK(dp);
-
-        OBD_ALLOC_PTR_WAIT(dummy_llap);
-        if (dummy_llap == NULL)
-                GOTO(out, rc);
-        dummy_llap->llap_page = NULL;
-        dummy_llap->llap_cookie = sbi;
-        dummy_llap->llap_magic = 0;
-
-        rc = seq_open(file, &llite_dump_pgcache_seq_sops);
-        if (rc) {
-                OBD_FREE(dummy_llap, sizeof(*dummy_llap));
-                GOTO(out, rc);
-        }
-        seq = file->private_data;
-        seq->private = dummy_llap;
-
-        spin_lock(&sbi->ll_lock);
-        list_add(&dummy_llap->llap_pglist_item, &sbi->ll_pglist);
-        spin_unlock(&sbi->ll_lock);
-
-out:
-        if (rc)
-                LPROCFS_EXIT();
-        return rc;
-}
-
-static int llite_dump_pgcache_seq_release(struct inode *inode,
-                                          struct file *file)
-{
-        struct seq_file *seq = file->private_data;
-        struct ll_async_page *dummy_llap = seq->private;
-        struct ll_sb_info *sbi = dummy_llap->llap_cookie;
-
-        spin_lock(&sbi->ll_lock);
-        if (!list_empty(&dummy_llap->llap_pglist_item))
-                list_del_init(&dummy_llap->llap_pglist_item);
-        spin_unlock(&sbi->ll_lock);
-        OBD_FREE(dummy_llap, sizeof(*dummy_llap));
-
-        return lprocfs_seq_release(inode, file);
-}
-
-struct file_operations llite_dump_pgcache_fops = {
-        .owner   = THIS_MODULE,
-        .open    = llite_dump_pgcache_seq_open,
-        .read    = seq_read,
-        .release = llite_dump_pgcache_seq_release,
-};
-
-static int ll_ra_stats_seq_show(struct seq_file *seq, void *v)
-{
-        struct timeval now;
-        struct ll_sb_info *sbi = seq->private;
-        struct ll_ra_info *ra = &sbi->ll_ra_info;
-        int i;
-        static char *ra_stat_strings[] = {
-                [RA_STAT_HIT] = "hits",
-                [RA_STAT_MISS] = "misses",
-                [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
-                [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
-                [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
-                [RA_STAT_FAILED_MATCH] = "failed lock match",
-                [RA_STAT_DISCARDED] = "read but discarded",
-                [RA_STAT_ZERO_LEN] = "zero length file",
-                [RA_STAT_ZERO_WINDOW] = "zero size window",
-                [RA_STAT_EOF] = "read-ahead to EOF",
-                [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
-                [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
-        };
-
-        do_gettimeofday(&now);
-
-        spin_lock(&sbi->ll_lock);
-
-        seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
-                   now.tv_sec, now.tv_usec);
-        seq_printf(seq, "pending issued pages:           %lu\n",
-                   ra->ra_cur_pages);
-
-        for(i = 0; i < _NR_RA_STAT; i++)
-                seq_printf(seq, "%-25s %lu\n", ra_stat_strings[i], 
-                           ra->ra_stats[i]);
-
-        spin_unlock(&sbi->ll_lock);
-
-        return 0;
-}
-
-static ssize_t ll_ra_stats_seq_write(struct file *file, const char *buf,
-                                       size_t len, loff_t *off)
-{
-        struct seq_file *seq = file->private_data;
-        struct ll_sb_info *sbi = seq->private;
-        struct ll_ra_info *ra = &sbi->ll_ra_info;
-
-        spin_lock(&sbi->ll_lock);
-        memset(ra->ra_stats, 0, sizeof(ra->ra_stats));
-        spin_unlock(&sbi->ll_lock);
-
-        return len;
-}
-
-LPROC_SEQ_FOPS(ll_ra_stats);
-
 #define pct(a,b) (b ? a * 100 / b : 0)
 
 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
@@ -1105,31 +1119,32 @@ static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
         struct timeval now;
         struct ll_sb_info *sbi = seq->private;
         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
-        int k;
+       int k;
 
-        do_gettimeofday(&now);
+       do_gettimeofday(&now);
 
-        if (!sbi->ll_rw_stats_on) {
-                seq_printf(seq, "Disabled\n"
-                                "Write anything in this file to activate\n");
+       if (!sbi->ll_rw_stats_on) {
+               seq_printf(seq, "disabled\n"
+                                "write anything in this file to activate, "
+                                "then 0 or \"[D/d]isabled\" to deactivate\n");
                 return 0;
         }
         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
                    now.tv_sec, now.tv_usec);
         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
-        seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n", 
+        seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
                    "extents", "calls", "%", "cum%",
                    "calls", "%", "cum%");
-        spin_lock(&sbi->ll_pp_extent_lock);
-        for(k = 0; k < LL_PROCESS_HIST_MAX; k++) {
-                if(io_extents->pp_extents[k].pid != 0) {
-                        seq_printf(seq, "\nPID: %d\n",
-                                   io_extents->pp_extents[k].pid);
-                        ll_display_extents_info(io_extents, seq, k);
-                }
-        }
-        spin_unlock(&sbi->ll_pp_extent_lock);
-        return 0;
+       spin_lock(&sbi->ll_pp_extent_lock);
+       for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
+               if (io_extents->pp_extents[k].pid != 0) {
+                       seq_printf(seq, "\nPID: %d\n",
+                                  io_extents->pp_extents[k].pid);
+                       ll_display_extents_info(io_extents, seq, k);
+               }
+       }
+       spin_unlock(&sbi->ll_pp_extent_lock);
+       return 0;
 }
 
 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
@@ -1140,45 +1155,56 @@ static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
         struct ll_sb_info *sbi = seq->private;
         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
         int i;
+        int value = 1, rc = 0;
 
-        sbi->ll_rw_stats_on = 1;
-        spin_lock(&sbi->ll_pp_extent_lock);
-        for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
-                io_extents->pp_extents[i].pid = 0;
-                lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
-                lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
-        }
-        spin_unlock(&sbi->ll_pp_extent_lock);
-        return len;
+        rc = lprocfs_write_helper(buf, len, &value);
+        if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
+                       strcmp(buf, "Disabled") == 0))
+                value = 0;
+
+        if (value == 0)
+                sbi->ll_rw_stats_on = 0;
+        else
+                sbi->ll_rw_stats_on = 1;
+
+       spin_lock(&sbi->ll_pp_extent_lock);
+       for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
+               io_extents->pp_extents[i].pid = 0;
+               lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
+               lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
+       }
+       spin_unlock(&sbi->ll_pp_extent_lock);
+       return len;
 }
 
 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
 
 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
 {
-        struct timeval now;
-        struct ll_sb_info *sbi = seq->private;
-        struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
+       struct timeval now;
+       struct ll_sb_info *sbi = seq->private;
+       struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
 
-        do_gettimeofday(&now);
+       do_gettimeofday(&now);
 
-        if (!sbi->ll_rw_stats_on) {
-                seq_printf(seq, "Disabled\n"
-                                "Write anything in this file to activate\n");
+       if (!sbi->ll_rw_stats_on) {
+                seq_printf(seq, "disabled\n"
+                                "write anything in this file to activate, "
+                                "then 0 or \"[D/d]isabled\" to deactivate\n");
                 return 0;
         }
         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
                    now.tv_sec, now.tv_usec);
 
         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
-        seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n", 
+        seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
                    "extents", "calls", "%", "cum%",
                    "calls", "%", "cum%");
-        spin_lock(&sbi->ll_lock);
-        ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
-        spin_unlock(&sbi->ll_lock);
+       spin_lock(&sbi->ll_lock);
+       ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
+       spin_unlock(&sbi->ll_lock);
 
-        return 0;
+       return 0;
 }
 
 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
@@ -1188,24 +1214,33 @@ static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
         struct ll_sb_info *sbi = seq->private;
         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
         int i;
+        int value = 1, rc = 0;
 
-        sbi->ll_rw_stats_on = 1;
-        spin_lock(&sbi->ll_pp_extent_lock);
-        for(i = 0; i <= LL_PROCESS_HIST_MAX; i++)
-        {
-                io_extents->pp_extents[i].pid = 0;
-                lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
-                lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
-        }
-        spin_unlock(&sbi->ll_pp_extent_lock);
+        rc = lprocfs_write_helper(buf, len, &value);
+        if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
+                       strcmp(buf, "Disabled") == 0))
+                value = 0;
 
-        return len;
+        if (value == 0)
+                sbi->ll_rw_stats_on = 0;
+        else
+                sbi->ll_rw_stats_on = 1;
+       spin_lock(&sbi->ll_pp_extent_lock);
+       for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
+               io_extents->pp_extents[i].pid = 0;
+               lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
+               lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
+       }
+       spin_unlock(&sbi->ll_pp_extent_lock);
+
+       return len;
 }
 
 LPROC_SEQ_FOPS(ll_rw_extents_stats);
 
-void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
-                               *file, size_t count, int rw)
+void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
+                       struct ll_file_data *file, loff_t pos,
+                       size_t count, int rw)
 {
         int i, cur = -1;
         struct ll_rw_process_info *process;
@@ -1219,7 +1254,7 @@ void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
         process = sbi->ll_rw_process_info;
         offset = sbi->ll_rw_offset_info;
 
-        spin_lock(&sbi->ll_pp_extent_lock);
+       spin_lock(&sbi->ll_pp_extent_lock);
         /* Extent statistics */
         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
                 if(io_extents->pp_extents[i].pid == pid) {
@@ -1230,7 +1265,7 @@ void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
 
         if (cur == -1) {
                 /* new process */
-                sbi->ll_extent_process_count = 
+                sbi->ll_extent_process_count =
                         (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
                 cur = sbi->ll_extent_process_count;
                 io_extents->pp_extents[cur].pid = pid;
@@ -1238,7 +1273,7 @@ void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
         }
 
-        for(i = 0; (count >= (1 << LL_HIST_START << i)) && 
+        for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
              (i < (LL_HIST_MAX - 1)); i++);
         if (rw == 0) {
                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
@@ -1247,24 +1282,23 @@ void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
         }
-        spin_unlock(&sbi->ll_pp_extent_lock);
+       spin_unlock(&sbi->ll_pp_extent_lock);
 
-        spin_lock(&sbi->ll_process_lock);
+       spin_lock(&sbi->ll_process_lock);
         /* Offset statistics */
         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
                 if (process[i].rw_pid == pid) {
                         if (process[i].rw_last_file != file) {
-                                process[i].rw_range_start = file->f_pos;
-                                process[i].rw_last_file_pos =
-                                                        file->f_pos + count;
+                                process[i].rw_range_start = pos;
+                                process[i].rw_last_file_pos = pos + count;
                                 process[i].rw_smallest_extent = count;
                                 process[i].rw_largest_extent = count;
                                 process[i].rw_offset = 0;
                                 process[i].rw_last_file = file;
-                                spin_unlock(&sbi->ll_process_lock);
+                               spin_unlock(&sbi->ll_process_lock);
                                 return;
                         }
-                        if (process[i].rw_last_file_pos != file->f_pos) {
+                        if (process[i].rw_last_file_pos != pos) {
                                 *off_count =
                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
                                 offset[*off_count].rw_op = process[i].rw_op;
@@ -1280,88 +1314,87 @@ void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
                                 offset[*off_count].rw_offset =
                                         process[i].rw_offset;
                                 process[i].rw_op = rw;
-                                process[i].rw_range_start = file->f_pos;
+                                process[i].rw_range_start = pos;
                                 process[i].rw_smallest_extent = count;
                                 process[i].rw_largest_extent = count;
-                                process[i].rw_offset = file->f_pos -
+                                process[i].rw_offset = pos -
                                         process[i].rw_last_file_pos;
                         }
                         if(process[i].rw_smallest_extent > count)
                                 process[i].rw_smallest_extent = count;
                         if(process[i].rw_largest_extent < count)
                                 process[i].rw_largest_extent = count;
-                        process[i].rw_last_file_pos = file->f_pos + count;
-                        spin_unlock(&sbi->ll_process_lock);
+                        process[i].rw_last_file_pos = pos + count;
+                       spin_unlock(&sbi->ll_process_lock);
                         return;
                 }
         }
         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
         process[*process_count].rw_pid = pid;
         process[*process_count].rw_op = rw;
-        process[*process_count].rw_range_start = file->f_pos;
-        process[*process_count].rw_last_file_pos = file->f_pos + count;
+        process[*process_count].rw_range_start = pos;
+        process[*process_count].rw_last_file_pos = pos + count;
         process[*process_count].rw_smallest_extent = count;
         process[*process_count].rw_largest_extent = count;
         process[*process_count].rw_offset = 0;
         process[*process_count].rw_last_file = file;
-        spin_unlock(&sbi->ll_process_lock);
+       spin_unlock(&sbi->ll_process_lock);
 }
 
-char lpszt[] = LPSZ;
-
 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
 {
         struct timeval now;
         struct ll_sb_info *sbi = seq->private;
         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
-        char format[50];
-        int i;
+       int i;
 
-        do_gettimeofday(&now);
+       do_gettimeofday(&now);
 
-        if (!sbi->ll_rw_stats_on) {
-                seq_printf(seq, "Disabled\n"
-                                "Write anything in this file to activate\n");
+       if (!sbi->ll_rw_stats_on) {
+                seq_printf(seq, "disabled\n"
+                                "write anything in this file to activate, "
+                                "then 0 or \"[D/d]isabled\" to deactivate\n");
                 return 0;
         }
-        spin_lock(&sbi->ll_process_lock);
+       spin_lock(&sbi->ll_process_lock);
 
         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
                    now.tv_sec, now.tv_usec);
         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
                    "R/W", "PID", "RANGE START", "RANGE END",
                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
-        sprintf(format, "%s%s%s%s%s\n",
-                "%3c %10d %14Lu %14Lu %17", lpszt+1, " %17", lpszt+1, " %14Ld");
-        /* We stored the discontiguous offsets here; print them first */
-        for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
-                if (offset[i].rw_pid != 0)
-                        /* Is there a way to snip the '%' off of LPSZ? */
-                        seq_printf(seq, format,
-                                   offset[i].rw_op ? 'W' : 'R',
-                                   offset[i].rw_pid,
-                                   offset[i].rw_range_start,
-                                   offset[i].rw_range_end,
-                                   offset[i].rw_smallest_extent,
-                                   offset[i].rw_largest_extent,
-                                   offset[i].rw_offset);
-        }
-        /* Then print the current offsets for each process */
-        for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
-                if (process[i].rw_pid != 0)
-                        seq_printf(seq, format,
-                                   process[i].rw_op ? 'W' : 'R',
-                                   process[i].rw_pid,
-                                   process[i].rw_range_start,
-                                   process[i].rw_last_file_pos,
-                                   process[i].rw_smallest_extent,
-                                   process[i].rw_largest_extent,
-                                   process[i].rw_offset);
-        }
-        spin_unlock(&sbi->ll_process_lock);
 
-        return 0;
+       /* We stored the discontiguous offsets here; print them first */
+       for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
+               if (offset[i].rw_pid != 0)
+                       seq_printf(seq,
+                                  "%3c %10d %14Lu %14Lu %17lu %17lu %14Lu",
+                                  offset[i].rw_op == READ ? 'R' : 'W',
+                                  offset[i].rw_pid,
+                                  offset[i].rw_range_start,
+                                  offset[i].rw_range_end,
+                                  (unsigned long)offset[i].rw_smallest_extent,
+                                  (unsigned long)offset[i].rw_largest_extent,
+                                  offset[i].rw_offset);
+       }
+
+       /* Then print the current offsets for each process */
+       for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
+               if (process[i].rw_pid != 0)
+                       seq_printf(seq,
+                                  "%3c %10d %14Lu %14Lu %17lu %17lu %14Lu",
+                                  process[i].rw_op == READ ? 'R' : 'W',
+                                  process[i].rw_pid,
+                                  process[i].rw_range_start,
+                                  process[i].rw_last_file_pos,
+                                  (unsigned long)process[i].rw_smallest_extent,
+                                  (unsigned long)process[i].rw_largest_extent,
+                                  process[i].rw_offset);
+       }
+       spin_unlock(&sbi->ll_process_lock);
+
+       return 0;
 }
 
 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
@@ -1371,19 +1404,29 @@ static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
         struct ll_sb_info *sbi = seq->private;
         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
+        int value = 1, rc = 0;
 
-        sbi->ll_rw_stats_on = 1;
+        rc = lprocfs_write_helper(buf, len, &value);
 
-        spin_lock(&sbi->ll_process_lock);
-        sbi->ll_offset_process_count = 0;
-        sbi->ll_rw_offset_entry_count = 0;
-        memset(process_info, 0, sizeof(struct ll_rw_process_info) *
-               LL_PROCESS_HIST_MAX);
-        memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
-               LL_OFFSET_HIST_MAX);
-        spin_unlock(&sbi->ll_process_lock);
+        if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
+                           strcmp(buf, "Disabled") == 0))
+                value = 0;
 
-        return len;
+        if (value == 0)
+                sbi->ll_rw_stats_on = 0;
+        else
+                sbi->ll_rw_stats_on = 1;
+
+       spin_lock(&sbi->ll_process_lock);
+       sbi->ll_offset_process_count = 0;
+       sbi->ll_rw_offset_entry_count = 0;
+       memset(process_info, 0, sizeof(struct ll_rw_process_info) *
+              LL_PROCESS_HIST_MAX);
+       memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
+              LL_OFFSET_HIST_MAX);
+       spin_unlock(&sbi->ll_process_lock);
+
+       return len;
 }
 
 LPROC_SEQ_FOPS(ll_rw_offset_stats);