Whamcloud - gitweb
LU-1644 mgc: remove obsolete IR swabbing workaround
[fs/lustre-release.git] / lustre / obdclass / lprocfs_status.c
index 08db676..0c434f2 100644 (file)
@@ -23,7 +23,7 @@
  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2015, Intel Corporation.
+ * Copyright (c) 2011, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -38,7 +38,6 @@
 
 #include <obd_class.h>
 #include <lprocfs_status.h>
-#include <lustre/lustre_idl.h>
 
 #ifdef CONFIG_PROC_FS
 
@@ -60,6 +59,29 @@ int lprocfs_seq_release(struct inode *inode, struct file *file)
 }
 EXPORT_SYMBOL(lprocfs_seq_release);
 
+struct dentry *ldebugfs_add_simple(struct dentry *root,
+                                  char *name, void *data,
+                                  const struct file_operations *fops)
+{
+       struct dentry *entry;
+       umode_t mode = 0;
+
+       if (!root || !name || !fops)
+               return ERR_PTR(-EINVAL);
+
+       if (fops->read)
+               mode = 0444;
+       if (fops->write)
+               mode |= 0200;
+       entry = debugfs_create_file(name, mode, root, data, fops);
+       if (IS_ERR_OR_NULL(entry)) {
+               CERROR("LprocFS: No memory to create <debugfs> entry %s", name);
+               return entry ?: ERR_PTR(-ENOMEM);
+       }
+       return entry;
+}
+EXPORT_SYMBOL(ldebugfs_add_simple);
+
 struct proc_dir_entry *
 lprocfs_add_simple(struct proc_dir_entry *root, char *name,
                   void *data, const struct file_operations *fops)
@@ -114,6 +136,35 @@ EXPORT_SYMBOL(lprocfs_add_symlink);
 
 static const struct file_operations lprocfs_generic_fops = { };
 
+int ldebugfs_add_vars(struct dentry *parent, struct lprocfs_vars *list,
+                     void *data)
+{
+       if (IS_ERR_OR_NULL(parent) || IS_ERR_OR_NULL(list))
+               return -EINVAL;
+
+       while (list->name) {
+               struct dentry *entry;
+               umode_t mode = 0;
+
+               if (list->proc_mode != 0000) {
+                       mode = list->proc_mode;
+               } else if (list->fops) {
+                       if (list->fops->read)
+                               mode = 0444;
+                       if (list->fops->write)
+                               mode |= 0200;
+               }
+               entry = debugfs_create_file(list->name, mode, parent,
+                                           list->data ? : data,
+                                           list->fops ? : &lprocfs_generic_fops);
+               if (IS_ERR_OR_NULL(entry))
+                       return entry ? PTR_ERR(entry) : -ENOMEM;
+               list++;
+       }
+       return 0;
+}
+EXPORT_SYMBOL_GPL(ldebugfs_add_vars);
+
 /**
  * Add /proc entries.
  *
@@ -155,6 +206,13 @@ lprocfs_add_vars(struct proc_dir_entry *root, struct lprocfs_vars *list,
 }
 EXPORT_SYMBOL(lprocfs_add_vars);
 
+void ldebugfs_remove(struct dentry **entryp)
+{
+       debugfs_remove_recursive(*entryp);
+       *entryp = NULL;
+}
+EXPORT_SYMBOL_GPL(ldebugfs_remove);
+
 #ifndef HAVE_REMOVE_PROC_SUBTREE
 /* for b=10866, global variable */
 DECLARE_RWSEM(_lprocfs_lock);
@@ -264,6 +322,31 @@ void lprocfs_remove_proc_entry(const char *name, struct proc_dir_entry *parent)
 }
 EXPORT_SYMBOL(lprocfs_remove_proc_entry);
 
+struct dentry *ldebugfs_register(const char *name, struct dentry *parent,
+                                struct lprocfs_vars *list, void *data)
+{
+       struct dentry *entry;
+
+       entry = debugfs_create_dir(name, parent);
+       if (IS_ERR_OR_NULL(entry)) {
+               entry = entry ?: ERR_PTR(-ENOMEM);
+               goto out;
+       }
+
+       if (!IS_ERR_OR_NULL(list)) {
+               int rc;
+
+               rc = ldebugfs_add_vars(entry, list, data);
+               if (rc) {
+                       debugfs_remove(entry);
+                       entry = ERR_PTR(rc);
+               }
+       }
+out:
+       return entry;
+}
+EXPORT_SYMBOL_GPL(ldebugfs_register);
+
 struct proc_dir_entry *
 lprocfs_register(const char *name, struct proc_dir_entry *parent,
                 struct lprocfs_vars *list, void *data)
@@ -286,93 +369,6 @@ lprocfs_register(const char *name, struct proc_dir_entry *parent,
 EXPORT_SYMBOL(lprocfs_register);
 
 /* Generic callbacks */
-int lprocfs_uint_seq_show(struct seq_file *m, void *data)
-{
-       seq_printf(m, "%u\n", *(unsigned int *)data);
-       return 0;
-}
-EXPORT_SYMBOL(lprocfs_uint_seq_show);
-
-int lprocfs_wr_uint(struct file *file, const char __user *buffer,
-                    unsigned long count, void *data)
-{
-       unsigned        *p = data;
-       char             dummy[MAX_STRING_SIZE + 1];
-       char            *end;
-       unsigned long    tmp;
-
-       if (count >= sizeof(dummy))
-               return -EINVAL;
-
-       if (count == 0)
-               return 0;
-
-       if (copy_from_user(dummy, buffer, count))
-               return -EFAULT;
-
-       dummy[count] = 0;
-
-       tmp = simple_strtoul(dummy, &end, 0);
-       if (dummy == end)
-               return -EINVAL;
-
-       *p = (unsigned int)tmp;
-       return count;
-}
-EXPORT_SYMBOL(lprocfs_wr_uint);
-
-ssize_t lprocfs_uint_seq_write(struct file *file, const char __user *buffer,
-                              size_t count, loff_t *off)
-{
-       int *data = ((struct seq_file *)file->private_data)->private;
-       int rc;
-       __s64 val = 0;
-
-       rc = lprocfs_str_to_s64(buffer, count, &val);
-       if (rc < 0)
-               return rc;
-
-       return lprocfs_wr_uint(file, buffer, count, data);
-}
-EXPORT_SYMBOL(lprocfs_uint_seq_write);
-
-int lprocfs_u64_seq_show(struct seq_file *m, void *data)
-{
-       LASSERT(data != NULL);
-       seq_printf(m, "%llu\n", *(__u64 *)data);
-       return 0;
-}
-EXPORT_SYMBOL(lprocfs_u64_seq_show);
-
-int lprocfs_atomic_seq_show(struct seq_file *m, void *data)
-{
-       atomic_t *atom = data;
-       LASSERT(atom != NULL);
-       seq_printf(m, "%d\n", atomic_read(atom));
-       return 0;
-}
-EXPORT_SYMBOL(lprocfs_atomic_seq_show);
-
-ssize_t
-lprocfs_atomic_seq_write(struct file *file, const char __user *buffer,
-                       size_t count, loff_t *off)
-{
-       atomic_t *atm = ((struct seq_file *)file->private_data)->private;
-       __s64 val = 0;
-       int rc;
-
-       rc = lprocfs_str_to_s64(buffer, count, &val);
-       if (rc < 0)
-               return rc;
-
-       if (val <= 0 || val > INT_MAX)
-               return -ERANGE;
-
-       atomic_set(atm, val);
-       return count;
-}
-EXPORT_SYMBOL(lprocfs_atomic_seq_write);
-
 int lprocfs_uuid_seq_show(struct seq_file *m, void *data)
 {
        struct obd_device *obd = data;
@@ -383,6 +379,16 @@ int lprocfs_uuid_seq_show(struct seq_file *m, void *data)
 }
 EXPORT_SYMBOL(lprocfs_uuid_seq_show);
 
+static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
+                        char *buf)
+{
+       struct obd_device *obd = container_of(kobj, struct obd_device,
+                                             obd_kset.kobj);
+
+       return sprintf(buf, "%s\n", obd->obd_uuid.uuid);
+}
+LUSTRE_RO_ATTR(uuid);
+
 int lprocfs_name_seq_show(struct seq_file *m, void *data)
 {
        struct obd_device *dev = data;
@@ -393,104 +399,153 @@ int lprocfs_name_seq_show(struct seq_file *m, void *data)
 }
 EXPORT_SYMBOL(lprocfs_name_seq_show);
 
-int lprocfs_blksize_seq_show(struct seq_file *m, void *data)
+static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
+                             char *buf)
 {
-       struct obd_device *obd = data;
-       struct obd_statfs  osfs;
-       int rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
-                           cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
-                           OBD_STATFS_NODELAY);
+       struct obd_device *obd = container_of(kobj, struct obd_device,
+                                             obd_kset.kobj);
+       struct obd_statfs osfs;
+       int rc;
+
+       rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
+                       ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
+                       OBD_STATFS_NODELAY);
        if (!rc)
-               seq_printf(m, "%u\n", osfs.os_bsize);
+               return sprintf(buf, "%u\n", osfs.os_bsize);
+
        return rc;
 }
-EXPORT_SYMBOL(lprocfs_blksize_seq_show);
+LUSTRE_RO_ATTR(blocksize);
 
-int lprocfs_kbytestotal_seq_show(struct seq_file *m, void *data)
+static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
+                               char *buf)
 {
-       struct obd_device *obd = data;
-       struct obd_statfs  osfs;
-       int rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
-                           cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
-                           OBD_STATFS_NODELAY);
+       struct obd_device *obd = container_of(kobj, struct obd_device,
+                                             obd_kset.kobj);
+       struct obd_statfs osfs;
+       int rc;
+
+       rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
+                       ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
+                       OBD_STATFS_NODELAY);
        if (!rc) {
-               __u32 blk_size = osfs.os_bsize >> 10;
-               __u64 result = osfs.os_blocks;
+               u32 blk_size = osfs.os_bsize >> 10;
+               u64 result = osfs.os_blocks;
 
                while (blk_size >>= 1)
                        result <<= 1;
 
-               seq_printf(m, "%llu\n", result);
+               return sprintf(buf, "%llu\n", result);
        }
+
        return rc;
 }
-EXPORT_SYMBOL(lprocfs_kbytestotal_seq_show);
+LUSTRE_RO_ATTR(kbytestotal);
 
-int lprocfs_kbytesfree_seq_show(struct seq_file *m, void *data)
+static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
+                              char *buf)
 {
-       struct obd_device *obd = data;
-       struct obd_statfs  osfs;
-       int rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
-                           cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
-                           OBD_STATFS_NODELAY);
+       struct obd_device *obd = container_of(kobj, struct obd_device,
+                                             obd_kset.kobj);
+       struct obd_statfs osfs;
+       int rc;
+
+       rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
+                       ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
+                       OBD_STATFS_NODELAY);
        if (!rc) {
-               __u32 blk_size = osfs.os_bsize >> 10;
-               __u64 result = osfs.os_bfree;
+               u32 blk_size = osfs.os_bsize >> 10;
+               u64 result = osfs.os_bfree;
 
                while (blk_size >>= 1)
                        result <<= 1;
 
-               seq_printf(m, "%llu\n", result);
+               return sprintf(buf, "%llu\n", result);
        }
+
        return rc;
 }
-EXPORT_SYMBOL(lprocfs_kbytesfree_seq_show);
+LUSTRE_RO_ATTR(kbytesfree);
 
-int lprocfs_kbytesavail_seq_show(struct seq_file *m, void *data)
+static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
+                               char *buf)
 {
-       struct obd_device *obd = data;
-       struct obd_statfs  osfs;
-       int rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
-                           cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
-                           OBD_STATFS_NODELAY);
+       struct obd_device *obd = container_of(kobj, struct obd_device,
+                                             obd_kset.kobj);
+       struct obd_statfs osfs;
+       int rc;
+
+       rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
+                       ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
+                       OBD_STATFS_NODELAY);
        if (!rc) {
-               __u32 blk_size = osfs.os_bsize >> 10;
-               __u64 result = osfs.os_bavail;
+               u32 blk_size = osfs.os_bsize >> 10;
+               u64 result = osfs.os_bavail;
 
                while (blk_size >>= 1)
                        result <<= 1;
 
-               seq_printf(m, "%llu\n", result);
+               return sprintf(buf, "%llu\n", result);
        }
+
        return rc;
 }
-EXPORT_SYMBOL(lprocfs_kbytesavail_seq_show);
+LUSTRE_RO_ATTR(kbytesavail);
 
-int lprocfs_filestotal_seq_show(struct seq_file *m, void *data)
+static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
+                              char *buf)
 {
-       struct obd_device *obd = data;
-       struct obd_statfs  osfs;
-       int rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
-                           cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
-                           OBD_STATFS_NODELAY);
+       struct obd_device *obd = container_of(kobj, struct obd_device,
+                                             obd_kset.kobj);
+       struct obd_statfs osfs;
+       int rc;
+
+       rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
+                       ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
+                       OBD_STATFS_NODELAY);
        if (!rc)
-               seq_printf(m, "%llu\n", osfs.os_files);
+               return sprintf(buf, "%llu\n", osfs.os_files);
+
        return rc;
 }
-EXPORT_SYMBOL(lprocfs_filestotal_seq_show);
+LUSTRE_RO_ATTR(filestotal);
 
-int lprocfs_filesfree_seq_show(struct seq_file *m, void *data)
+static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
+                             char *buf)
 {
-       struct obd_device *obd = data;
-       struct obd_statfs  osfs;
-       int rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
-                           cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
-                           OBD_STATFS_NODELAY);
+       struct obd_device *obd = container_of(kobj, struct obd_device,
+                                             obd_kset.kobj);
+       struct obd_statfs osfs;
+       int rc;
+
+       rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
+                       ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
+                       OBD_STATFS_NODELAY);
        if (!rc)
-               seq_printf(m, "%llu\n", osfs.os_ffree);
+               return sprintf(buf, "%llu\n", osfs.os_ffree);
+
        return rc;
 }
-EXPORT_SYMBOL(lprocfs_filesfree_seq_show);
+LUSTRE_RO_ATTR(filesfree);
+
+ssize_t conn_uuid_show(struct kobject *kobj, struct attribute *attr, char *buf)
+{
+       struct obd_device *obd = container_of(kobj, struct obd_device,
+                                             obd_kset.kobj);
+       struct ptlrpc_connection *conn;
+       ssize_t count;
+
+       LPROCFS_CLIMP_CHECK(obd);
+       conn = obd->u.cli.cl_import->imp_connection;
+       if (conn && obd->u.cli.cl_import)
+               count = sprintf(buf, "%s\n", conn->c_remote_uuid.uuid);
+       else
+               count = sprintf(buf, "%s\n", "<none>");
+
+       LPROCFS_CLIMP_EXIT(obd);
+       return count;
+}
+EXPORT_SYMBOL(conn_uuid_show);
 
 int lprocfs_server_uuid_seq_show(struct seq_file *m, void *data)
 {
@@ -532,6 +587,95 @@ int lprocfs_conn_uuid_seq_show(struct seq_file *m, void *data)
 EXPORT_SYMBOL(lprocfs_conn_uuid_seq_show);
 
 /** add up per-cpu counters */
+
+/**
+ * Lock statistics structure for access, possibly only on this CPU.
+ *
+ * The statistics struct may be allocated with per-CPU structures for
+ * efficient concurrent update (usually only on server-wide stats), or
+ * as a single global struct (e.g. for per-client or per-job statistics),
+ * so the required locking depends on the type of structure allocated.
+ *
+ * For per-CPU statistics, pin the thread to the current cpuid so that
+ * will only access the statistics for that CPU.  If the stats structure
+ * for the current CPU has not been allocated (or previously freed),
+ * allocate it now.  The per-CPU statistics do not need locking since
+ * the thread is pinned to the CPU during update.
+ *
+ * For global statistics, lock the stats structure to prevent concurrent update.
+ *
+ * \param[in] stats    statistics structure to lock
+ * \param[in] opc      type of operation:
+ *                     LPROCFS_GET_SMP_ID: "lock" and return current CPU index
+ *                             for incrementing statistics for that CPU
+ *                     LPROCFS_GET_NUM_CPU: "lock" and return number of used
+ *                             CPU indices to iterate over all indices
+ * \param[out] flags   CPU interrupt saved state for IRQ-safe locking
+ *
+ * \retval cpuid of current thread or number of allocated structs
+ * \retval negative on error (only for opc LPROCFS_GET_SMP_ID + per-CPU stats)
+ */
+int lprocfs_stats_lock(struct lprocfs_stats *stats,
+                      enum lprocfs_stats_lock_ops opc,
+                      unsigned long *flags)
+{
+       if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) {
+               if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
+                       spin_lock_irqsave(&stats->ls_lock, *flags);
+               else
+                       spin_lock(&stats->ls_lock);
+               return opc == LPROCFS_GET_NUM_CPU ? 1 : 0;
+       }
+
+       switch (opc) {
+       case LPROCFS_GET_SMP_ID: {
+               unsigned int cpuid = get_cpu();
+
+               if (unlikely(!stats->ls_percpu[cpuid])) {
+                       int rc = lprocfs_stats_alloc_one(stats, cpuid);
+
+                       if (rc < 0) {
+                               put_cpu();
+                               return rc;
+                       }
+               }
+               return cpuid;
+       }
+       case LPROCFS_GET_NUM_CPU:
+               return stats->ls_biggest_alloc_num;
+       default:
+               LBUG();
+       }
+}
+
+/**
+ * Unlock statistics structure after access.
+ *
+ * Unlock the lock acquired via lprocfs_stats_lock() for global statistics,
+ * or unpin this thread from the current cpuid for per-CPU statistics.
+ *
+ * This function must be called using the same arguments as used when calling
+ * lprocfs_stats_lock() so that the correct operation can be performed.
+ *
+ * \param[in] stats    statistics structure to unlock
+ * \param[in] opc      type of operation (current cpuid or number of structs)
+ * \param[in] flags    CPU interrupt saved state for IRQ-safe locking
+ */
+void lprocfs_stats_unlock(struct lprocfs_stats *stats,
+                         enum lprocfs_stats_lock_ops opc,
+                         unsigned long *flags)
+{
+       if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) {
+               if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
+                       spin_unlock_irqrestore(&stats->ls_lock, *flags);
+               else
+                       spin_unlock(&stats->ls_lock);
+       } else if (opc == LPROCFS_GET_SMP_ID) {
+               put_cpu();
+       }
+}
+
+/** add up per-cpu counters */
 void lprocfs_stats_collect(struct lprocfs_stats *stats, int idx,
                           struct lprocfs_counter *cnt)
 {
@@ -569,16 +713,6 @@ void lprocfs_stats_collect(struct lprocfs_stats *stats, int idx,
        lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
 }
 
-/**
- * Append a space separated list of current set flags to str.
- */
-#define flag2str(flag)                                         \
-       do {                                                            \
-               if (imp->imp_##flag) {                                  \
-                       seq_printf(m, "%s" #flag, first ? "" : ", ");   \
-                       first = false;                                  \
-               }                                                       \
-       } while (0)
 static void obd_import_flags2str(struct obd_import *imp, struct seq_file *m)
 {
        bool first = true;
@@ -588,19 +722,16 @@ static void obd_import_flags2str(struct obd_import *imp, struct seq_file *m)
                first = false;
        }
 
-       flag2str(invalid);
-       flag2str(deactive);
-       flag2str(replayable);
-       flag2str(delayed_recovery);
-       flag2str(no_lock_replay);
-       flag2str(vbr_failed);
-       flag2str(pingable);
-       flag2str(resend_replay);
-       flag2str(no_pinger_recover);
-       flag2str(need_mne_swab);
-       flag2str(connect_tried);
+       flag2str(imp, invalid);
+       flag2str(imp, deactive);
+       flag2str(imp, replayable);
+       flag2str(imp, delayed_recovery);
+       flag2str(imp, vbr_failed);
+       flag2str(imp, pingable);
+       flag2str(imp, resend_replay);
+       flag2str(imp, no_pinger_recover);
+       flag2str(imp, connect_tried);
 }
-#undef flag2str
 
 static const char *obd_connect_names[] = {
        /* flags names  */
@@ -617,7 +748,7 @@ static const char *obd_connect_names[] = {
        "truncate_lock",
        "initial_transno",
        "inode_bit_locks",
-       "join_file(obsolete)",
+       "barrier",
        "getattr_by_fid",
        "no_oh_for_devices",
        "remote_client",
@@ -664,17 +795,18 @@ static const char *obd_connect_names[] = {
        "multi_mod_rpcs",
        "dir_stripe",
        "subtree",
-       "lock_ahead",
+       "lockahead",
        "bulk_mbits",
        "compact_obdo",
        "second_flags",
        /* flags2 names */
        "file_secctx",
+       "lockaheadv2",
        NULL
 };
 
-static void obd_connect_seq_flags2str(struct seq_file *m, __u64 flags,
-                                     __u64 flags2, const char *sep)
+void obd_connect_seq_flags2str(struct seq_file *m, __u64 flags, __u64 flags2,
+                              const char *sep)
 {
        bool first = true;
        __u64 mask;
@@ -711,6 +843,7 @@ static void obd_connect_seq_flags2str(struct seq_file *m, __u64 flags,
                first = false;
        }
 }
+EXPORT_SYMBOL(obd_connect_seq_flags2str);
 
 int obd_connect_flags2str(char *page, int count, __u64 flags, __u64 flags2,
                          const char *sep)
@@ -747,8 +880,8 @@ int obd_connect_flags2str(char *page, int count, __u64 flags, __u64 flags2,
 }
 EXPORT_SYMBOL(obd_connect_flags2str);
 
-static void obd_connect_data_seqprint(struct seq_file *m,
-                                     struct obd_connect_data *ocd)
+void
+obd_connect_data_seqprint(struct seq_file *m, struct obd_connect_data *ocd)
 {
        __u64 flags;
 
@@ -964,9 +1097,8 @@ int lprocfs_state_seq_show(struct seq_file *m, void *data)
                        &imp->imp_state_hist[(k + j) % IMP_STATE_HIST_LEN];
                if (ish->ish_state == 0)
                        continue;
-               seq_printf(m, " - [ "CFS_TIME_T", %s ]\n",
-                          ish->ish_time,
-               ptlrpc_import_state_name(ish->ish_state));
+               seq_printf(m, " - [ %lld, %s ]\n", (s64)ish->ish_time,
+                          ptlrpc_import_state_name(ish->ish_state));
        }
 
        LPROCFS_CLIMP_EXIT(obd);
@@ -990,27 +1122,25 @@ int lprocfs_timeouts_seq_show(struct seq_file *m, void *data)
        struct obd_device *obd = (struct obd_device *)data;
        struct obd_import *imp;
        unsigned int cur, worst;
-       time_t now, worstt;
-       struct dhms ts;
+       time64_t now, worstt;
        int i;
 
        LASSERT(obd != NULL);
        LPROCFS_CLIMP_CHECK(obd);
        imp = obd->u.cli.cl_import;
 
-       now = cfs_time_current_sec();
+       now = ktime_get_real_seconds();
 
        /* Some network health info for kicks */
-       s2dhms(&ts, now - imp->imp_last_reply_time);
-       seq_printf(m, "%-10s : %ld, "DHMS_FMT" ago\n",
-                  "last reply", imp->imp_last_reply_time, DHMS_VARS(&ts));
+       seq_printf(m, "%-10s : %lld, %llds ago\n",
+                  "last reply", (s64)imp->imp_last_reply_time,
+                  (s64)(now - imp->imp_last_reply_time));
 
        cur = at_get(&imp->imp_at.iat_net_latency);
        worst = imp->imp_at.iat_net_latency.at_worst_ever;
        worstt = imp->imp_at.iat_net_latency.at_worst_time;
-       s2dhms(&ts, now - worstt);
-       seq_printf(m, "%-10s : cur %3u  worst %3u (at %ld, "DHMS_FMT" ago) ",
-                  "network", cur, worst, worstt, DHMS_VARS(&ts));
+       seq_printf(m, "%-10s : cur %3u  worst %3u (at %lld, %llds ago) ",
+                  "network", cur, worst, (s64)worstt, (s64)(now - worstt));
        lprocfs_at_hist_helper(m, &imp->imp_at.iat_net_latency);
 
        for(i = 0; i < IMP_AT_MAX_PORTALS; i++) {
@@ -1019,10 +1149,9 @@ int lprocfs_timeouts_seq_show(struct seq_file *m, void *data)
                cur = at_get(&imp->imp_at.iat_service_estimate[i]);
                worst = imp->imp_at.iat_service_estimate[i].at_worst_ever;
                worstt = imp->imp_at.iat_service_estimate[i].at_worst_time;
-               s2dhms(&ts, now - worstt);
-               seq_printf(m, "portal %-2d  : cur %3u  worst %3u (at %ld, "
-                          DHMS_FMT" ago) ", imp->imp_at.iat_portal[i],
-                          cur, worst, worstt, DHMS_VARS(&ts));
+               seq_printf(m, "portal %-2d  : cur %3u  worst %3u (at %lld, %llds ago) ",
+                          imp->imp_at.iat_portal[i], cur, worst, (s64)worstt,
+                          (s64)(now - worstt));
                lprocfs_at_hist_helper(m, &imp->imp_at.iat_service_estimate[i]);
        }
 
@@ -1049,14 +1178,82 @@ int lprocfs_connect_flags_seq_show(struct seq_file *m, void *data)
 }
 EXPORT_SYMBOL(lprocfs_connect_flags_seq_show);
 
-int
-lprocfs_obd_setup(struct obd_device *obd)
+static struct attribute *obd_def_uuid_attrs[] = {
+       &lustre_attr_uuid.attr,
+       NULL,
+};
+
+static struct attribute *obd_def_attrs[] = {
+       &lustre_attr_blocksize.attr,
+       &lustre_attr_kbytestotal.attr,
+       &lustre_attr_kbytesfree.attr,
+       &lustre_attr_kbytesavail.attr,
+       &lustre_attr_filestotal.attr,
+       &lustre_attr_filesfree.attr,
+       &lustre_attr_uuid.attr,
+       NULL,
+};
+
+static void obd_sysfs_release(struct kobject *kobj)
 {
-       int rc = 0;
+       struct obd_device *obd = container_of(kobj, struct obd_device,
+                                             obd_kset.kobj);
 
-       LASSERT(obd != NULL);
-       LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
-       LASSERT(obd->obd_type->typ_procroot != NULL);
+       complete(&obd->obd_kobj_unregister);
+}
+
+int lprocfs_obd_setup(struct obd_device *obd, bool uuid_only)
+{
+       struct lprocfs_vars *debugfs_vars = NULL;
+       int rc;
+
+       if (!obd || obd->obd_magic != OBD_DEVICE_MAGIC)
+               return -ENODEV;
+
+       rc = kobject_set_name(&obd->obd_kset.kobj, "%s", obd->obd_name);
+       if (rc)
+               return rc;
+
+       obd->obd_ktype.sysfs_ops = &lustre_sysfs_ops;
+       obd->obd_ktype.release = obd_sysfs_release;
+       if (obd->obd_attrs)
+               obd->obd_ktype.default_attrs = obd->obd_attrs;
+
+       obd->obd_kset.kobj.parent = obd->obd_type->typ_kobj;
+       obd->obd_kset.kobj.ktype = &obd->obd_ktype;
+       init_completion(&obd->obd_kobj_unregister);
+       rc = kset_register(&obd->obd_kset);
+       if (rc)
+               return rc;
+
+       if (uuid_only)
+               obd->obd_attrs_group.attrs = obd_def_uuid_attrs;
+       else
+               obd->obd_attrs_group.attrs = obd_def_attrs;
+
+       rc = sysfs_create_group(&obd->obd_kset.kobj, &obd->obd_attrs_group);
+       if (rc) {
+               kset_unregister(&obd->obd_kset);
+               return rc;
+       }
+
+       if (!obd->obd_type->typ_procroot)
+               debugfs_vars = obd->obd_vars;
+       obd->obd_debugfs_entry = ldebugfs_register(obd->obd_name,
+                                                  obd->obd_type->typ_debugfs_entry,
+                                                  debugfs_vars, obd);
+       if (IS_ERR_OR_NULL(obd->obd_debugfs_entry)) {
+               rc = obd->obd_debugfs_entry ? PTR_ERR(obd->obd_debugfs_entry)
+                                           : -ENOMEM;
+               CERROR("error %d setting up debugfs for %s\n",
+                      rc, obd->obd_name);
+               obd->obd_debugfs_entry = NULL;
+               lprocfs_obd_cleanup(obd);
+               return rc;
+       }
+
+       if (obd->obd_proc_entry || !obd->obd_type->typ_procroot)
+               GOTO(already_registered, rc);
 
        obd->obd_proc_entry = lprocfs_register(obd->obd_name,
                                               obd->obd_type->typ_procroot,
@@ -1065,25 +1262,37 @@ lprocfs_obd_setup(struct obd_device *obd)
                rc = PTR_ERR(obd->obd_proc_entry);
                CERROR("error %d setting up lprocfs for %s\n",rc,obd->obd_name);
                obd->obd_proc_entry = NULL;
+               lprocfs_obd_cleanup(obd);
        }
+already_registered:
        return rc;
 }
 EXPORT_SYMBOL(lprocfs_obd_setup);
 
 int lprocfs_obd_cleanup(struct obd_device *obd)
 {
-        if (!obd)
-                return -EINVAL;
-        if (obd->obd_proc_exports_entry) {
-                /* Should be no exports left */
-                lprocfs_remove(&obd->obd_proc_exports_entry);
-                obd->obd_proc_exports_entry = NULL;
-        }
-        if (obd->obd_proc_entry) {
-                lprocfs_remove(&obd->obd_proc_entry);
-                obd->obd_proc_entry = NULL;
-        }
-        return 0;
+       if (!obd)
+               return -EINVAL;
+
+       if (obd->obd_proc_exports_entry) {
+               /* Should be no exports left */
+               lprocfs_remove(&obd->obd_proc_exports_entry);
+               obd->obd_proc_exports_entry = NULL;
+       }
+
+       if (obd->obd_proc_entry) {
+               lprocfs_remove(&obd->obd_proc_entry);
+               obd->obd_proc_entry = NULL;
+       }
+
+       if (!IS_ERR_OR_NULL(obd->obd_debugfs_entry))
+               ldebugfs_remove(&obd->obd_debugfs_entry);
+
+       sysfs_remove_group(&obd->obd_kset.kobj, &obd->obd_attrs_group);
+       kset_unregister(&obd->obd_kset);
+       wait_for_completion(&obd->obd_kobj_unregister);
+
+       return 0;
 }
 EXPORT_SYMBOL(lprocfs_obd_cleanup);
 
@@ -1123,7 +1332,6 @@ int lprocfs_stats_alloc_one(struct lprocfs_stats *stats, unsigned int cpuid)
        }
        return rc;
 }
-EXPORT_SYMBOL(lprocfs_stats_alloc_one);
 
 struct lprocfs_stats *lprocfs_alloc_stats(unsigned int num,
                                           enum lprocfs_stats_flags flags)
@@ -1208,6 +1416,32 @@ void lprocfs_free_stats(struct lprocfs_stats **statsh)
 }
 EXPORT_SYMBOL(lprocfs_free_stats);
 
+u64 lprocfs_stats_collector(struct lprocfs_stats *stats, int idx,
+                           enum lprocfs_fields_flags field)
+{
+       unsigned long flags = 0;
+       unsigned int num_cpu;
+       unsigned int i;
+       u64 ret = 0;
+
+       LASSERT(stats);
+
+       num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
+       for (i = 0; i < num_cpu; i++) {
+               struct lprocfs_counter *cntr;
+
+               if (!stats->ls_percpu[i])
+                       continue;
+
+               cntr = lprocfs_stats_counter_get(stats, i, idx);
+               ret += lprocfs_read_helper(cntr, &stats->ls_cnt_header[idx],
+                                          stats->ls_flags, field);
+       }
+       lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
+       return ret;
+}
+EXPORT_SYMBOL(lprocfs_stats_collector);
+
 void lprocfs_clear_stats(struct lprocfs_stats *stats)
 {
        struct lprocfs_counter          *percpu_cntr;
@@ -1276,11 +1510,11 @@ static int lprocfs_stats_seq_show(struct seq_file *p, void *v)
        int                              idx    = *(loff_t *)v;
 
        if (idx == 0) {
-               struct timeval now;
+               struct timespec64 now;
 
-               do_gettimeofday(&now);
-               seq_printf(p, "%-25s %lu.%lu secs.usecs\n",
-                          "snapshot_time", now.tv_sec, now.tv_usec);
+               ktime_get_real_ts64(&now);
+               seq_printf(p, "%-25s %llu.%09lu secs.nsecs\n",
+                          "snapshot_time", (s64)now.tv_sec, now.tv_nsec);
        }
 
        hdr = &stats->ls_cnt_header[idx];
@@ -1322,7 +1556,7 @@ static int lprocfs_stats_seq_open(struct inode *inode, struct file *file)
        if (rc)
                return rc;
        seq = file->private_data;
-       seq->private = PDE_DATA(inode);
+       seq->private = inode->i_private ? inode->i_private : PDE_DATA(inode);
        return 0;
 }
 
@@ -1335,6 +1569,22 @@ static const struct file_operations lprocfs_stats_seq_fops = {
         .release = lprocfs_seq_release,
 };
 
+int ldebugfs_register_stats(struct dentry *parent, const char *name,
+                           struct lprocfs_stats *stats)
+{
+       struct dentry *entry;
+
+       LASSERT(!IS_ERR_OR_NULL(parent));
+
+       entry = debugfs_create_file(name, 0644, parent, stats,
+                                   &lprocfs_stats_seq_fops);
+       if (IS_ERR_OR_NULL(entry))
+               return entry ? PTR_ERR(entry) : -ENOMEM;
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(ldebugfs_register_stats);
+
 int lprocfs_register_stats(struct proc_dir_entry *root, const char *name,
                            struct lprocfs_stats *stats)
 {
@@ -1609,6 +1859,7 @@ int lprocfs_read_frac_helper(char *buffer, unsigned long count, long val,
         buffer[prtn++] ='\n';
         return prtn;
 }
+EXPORT_SYMBOL(lprocfs_read_frac_helper);
 
 int lprocfs_seq_read_frac_helper(struct seq_file *m, long val, int mult)
 {
@@ -1890,26 +2141,6 @@ static int str_to_s64_internal(const char __user *buffer, unsigned long count,
 
 /**
  * Convert a user string into a signed 64 bit number. This function produces
- * an error when the value parsed from the string underflows or
- * overflows. This function accepts strings which contain digits and
- * optionally a decimal or hex strings which are prefixed with "0x".
- *
- * \param[in] buffer   string consisting of numbers and optionally a decimal
- * \param[in] count    buffer length
- * \param[in] val      if successful, the value represented by the string
- *
- * \retval             0 on success
- * \retval             negative number on error
- */
-int lprocfs_str_to_s64(const char __user *buffer, unsigned long count,
-                      __s64 *val)
-{
-       return str_to_s64_internal(buffer, count, val, 1, false);
-}
-EXPORT_SYMBOL(lprocfs_str_to_s64);
-
-/**
- * Convert a user string into a signed 64 bit number. This function produces
  * an error when the value parsed from the string times multiplier underflows or
  * overflows. This function only accepts strings that contains digits, an
  * optional decimal, and a char representing a unit at the end. If a unit is
@@ -1941,7 +2172,7 @@ int lprocfs_str_with_units_to_s64(const char __user *buffer,
 }
 EXPORT_SYMBOL(lprocfs_str_with_units_to_s64);
 
-static char *lprocfs_strnstr(const char *s1, const char *s2, size_t len)
+char *lprocfs_strnstr(const char *s1, const char *s2, size_t len)
 {
        size_t l2;
 
@@ -1956,6 +2187,7 @@ static char *lprocfs_strnstr(const char *s1, const char *s2, size_t len)
        }
        return NULL;
 }
+EXPORT_SYMBOL(lprocfs_strnstr);
 
 /**
  * Find the string \a name in the input \a buffer, and return a pointer to the
@@ -1987,6 +2219,22 @@ char *lprocfs_find_named_value(const char *buffer, const char *name,
 }
 EXPORT_SYMBOL(lprocfs_find_named_value);
 
+int ldebugfs_seq_create(struct dentry *parent, const char *name, umode_t mode,
+                       const struct file_operations *seq_fops, void *data)
+{
+       struct dentry *entry;
+
+       /* Disallow secretly (un)writable entries. */
+       LASSERT((!seq_fops->write) == (!(mode & 0222)));
+
+       entry = debugfs_create_file(name, mode, parent, data, seq_fops);
+       if (IS_ERR_OR_NULL(entry))
+               return entry ? PTR_ERR(entry) : -ENOMEM;
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(ldebugfs_seq_create);
+
 int lprocfs_seq_create(struct proc_dir_entry *parent,
                       const char *name,
                       mode_t mode,
@@ -2060,19 +2308,29 @@ void lprocfs_oh_clear(struct obd_histogram *oh)
 }
 EXPORT_SYMBOL(lprocfs_oh_clear);
 
-int lprocfs_obd_rd_max_pages_per_rpc(char *page, char **start, off_t off,
-                                    int count, int *eof, void *data)
+ssize_t lustre_attr_show(struct kobject *kobj,
+                        struct attribute *attr, char *buf)
 {
-       struct obd_device *dev = data;
-       struct client_obd *cli = &dev->u.cli;
-       int rc;
+       struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
 
-       spin_lock(&cli->cl_loi_list_lock);
-       rc = snprintf(page, count, "%d\n", cli->cl_max_pages_per_rpc);
-       spin_unlock(&cli->cl_loi_list_lock);
+       return a->show ? a->show(kobj, attr, buf) : 0;
+}
+EXPORT_SYMBOL_GPL(lustre_attr_show);
 
-       return rc;
+ssize_t lustre_attr_store(struct kobject *kobj, struct attribute *attr,
+                         const char *buf, size_t len)
+{
+       struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
+
+       return a->store ? a->store(kobj, attr, buf, len) : len;
 }
+EXPORT_SYMBOL_GPL(lustre_attr_store);
+
+const struct sysfs_ops lustre_sysfs_ops = {
+       .show  = lustre_attr_show,
+       .store = lustre_attr_store,
+};
+EXPORT_SYMBOL_GPL(lustre_sysfs_ops);
 
 int lprocfs_obd_max_pages_per_rpc_seq_show(struct seq_file *m, void *data)
 {
@@ -2086,6 +2344,98 @@ int lprocfs_obd_max_pages_per_rpc_seq_show(struct seq_file *m, void *data)
 }
 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_show);
 
+ssize_t lprocfs_obd_max_pages_per_rpc_seq_write(struct file *file,
+                                               const char __user *buffer,
+                                               size_t count, loff_t *off)
+{
+       struct obd_device *dev =
+               ((struct seq_file *)file->private_data)->private;
+       struct client_obd *cli = &dev->u.cli;
+       struct obd_connect_data *ocd = &cli->cl_import->imp_connect_data;
+       int chunk_mask, rc;
+       s64 val;
+
+       rc = lprocfs_str_with_units_to_s64(buffer, count, &val, '1');
+       if (rc)
+               return rc;
+       if (val < 0)
+               return -ERANGE;
+
+       /* if the max_pages is specified in bytes, convert to pages */
+       if (val >= ONE_MB_BRW_SIZE)
+               val >>= PAGE_SHIFT;
+
+       LPROCFS_CLIMP_CHECK(dev);
+
+       chunk_mask = ~((1 << (cli->cl_chunkbits - PAGE_SHIFT)) - 1);
+       /* max_pages_per_rpc must be chunk aligned */
+       val = (val + ~chunk_mask) & chunk_mask;
+       if (val == 0 || (ocd->ocd_brw_size != 0 &&
+                        val > ocd->ocd_brw_size >> PAGE_SHIFT)) {
+               LPROCFS_CLIMP_EXIT(dev);
+               return -ERANGE;
+       }
+       spin_lock(&cli->cl_loi_list_lock);
+       cli->cl_max_pages_per_rpc = val;
+       client_adjust_max_dirty(cli);
+       spin_unlock(&cli->cl_loi_list_lock);
+
+       LPROCFS_CLIMP_EXIT(dev);
+       return count;
+}
+EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_write);
+
+ssize_t short_io_bytes_show(struct kobject *kobj, struct attribute *attr,
+                           char *buf)
+{
+       struct obd_device *dev = container_of(kobj, struct obd_device,
+                                             obd_kset.kobj);
+       struct client_obd *cli = &dev->u.cli;
+       int rc;
+
+       spin_lock(&cli->cl_loi_list_lock);
+       rc = sprintf(buf, "%d\n", cli->cl_short_io_bytes);
+       spin_unlock(&cli->cl_loi_list_lock);
+       return rc;
+}
+EXPORT_SYMBOL(short_io_bytes_show);
+
+/* Used to catch people who think they're specifying pages. */
+#define MIN_SHORT_IO_BYTES 64
+
+ssize_t short_io_bytes_store(struct kobject *kobj, struct attribute *attr,
+                            const char *buffer, size_t count)
+{
+       struct obd_device *dev = container_of(kobj, struct obd_device,
+                                             obd_kset.kobj);
+       struct client_obd *cli = &dev->u.cli;
+       u32 val;
+       int rc;
+
+       LPROCFS_CLIMP_CHECK(dev);
+
+       rc = kstrtouint(buffer, 0, &val);
+       if (rc)
+               GOTO(out, rc);
+
+       if (val < MIN_SHORT_IO_BYTES || val > OBD_MAX_SHORT_IO_BYTES)
+               GOTO(out, rc = -ERANGE);
+
+       rc = count;
+
+       spin_lock(&cli->cl_loi_list_lock);
+       if (val > (cli->cl_max_pages_per_rpc << PAGE_SHIFT))
+               rc = -ERANGE;
+       else
+               cli->cl_short_io_bytes = val;
+       spin_unlock(&cli->cl_loi_list_lock);
+
+out:
+       LPROCFS_CLIMP_EXIT(dev);
+       return rc;
+}
+EXPORT_SYMBOL(short_io_bytes_store);
+
 int lprocfs_wr_root_squash(const char __user *buffer, unsigned long count,
                           struct root_squash_info *squash, char *name)
 {