Whamcloud - gitweb
LU-3321 osc: Adjustment on osc LRU for performance
[fs/lustre-release.git] / lustre / osc / lproc_osc.c
index 742733c..e765e48 100644 (file)
@@ -27,7 +27,7 @@
  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, Whamcloud, Inc.
+ * Copyright (c) 2011, 2013, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -68,14 +68,12 @@ static int osc_wr_active(struct file *file, const char *buffer,
         if (val < 0 || val > 1)
                 return -ERANGE;
 
-        LPROCFS_CLIMP_CHECK(dev);
         /* opposite senses */
         if (dev->u.cli.cl_import->imp_deactive == val)
                 rc = ptlrpc_set_import_active(dev->u.cli.cl_import, val);
         else
                 CDEBUG(D_CONFIG, "activate %d: ignoring repeat request\n", val);
 
-        LPROCFS_CLIMP_EXIT(dev);
         return count;
 }
 
@@ -136,28 +134,70 @@ static int osc_rd_max_dirty_mb(char *page, char **start, off_t off, int count,
 }
 
 static int osc_wr_max_dirty_mb(struct file *file, const char *buffer,
-                               unsigned long count, void *data)
+                              unsigned long count, void *data)
 {
-        struct obd_device *dev = data;
-        struct client_obd *cli = &dev->u.cli;
-        int pages_number, mult, rc;
+       struct obd_device *dev = data;
+       struct client_obd *cli = &dev->u.cli;
+       int pages_number, mult, rc;
 
-        mult = 1 << (20 - CFS_PAGE_SHIFT);
-        rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
-        if (rc)
-                return rc;
+       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 > OSC_MAX_DIRTY_MB_MAX << (20 - CFS_PAGE_SHIFT) ||
-            pages_number > cfs_num_physpages / 4) /* 1/4 of RAM */
-                return -ERANGE;
+       if (pages_number <= 0 ||
+           pages_number > OSC_MAX_DIRTY_MB_MAX << (20 - PAGE_CACHE_SHIFT) ||
+           pages_number > totalram_pages / 4) /* 1/4 of RAM */
+               return -ERANGE;
 
-        client_obd_list_lock(&cli->cl_loi_list_lock);
-        cli->cl_dirty_max = (obd_count)(pages_number << CFS_PAGE_SHIFT);
-        osc_wake_cache_waiters(cli);
-        client_obd_list_unlock(&cli->cl_loi_list_lock);
+       client_obd_list_lock(&cli->cl_loi_list_lock);
+       cli->cl_dirty_max = (obd_count)(pages_number << PAGE_CACHE_SHIFT);
+       osc_wake_cache_waiters(cli);
+       client_obd_list_unlock(&cli->cl_loi_list_lock);
 
-        return count;
+       return count;
+}
+
+static int osc_rd_cached_mb(char *page, char **start, off_t off, int count,
+                           int *eof, void *data)
+{
+       struct obd_device *dev = data;
+       struct client_obd *cli = &dev->u.cli;
+       int shift = 20 - PAGE_CACHE_SHIFT;
+       int rc;
+
+       rc = snprintf(page, count,
+                     "used_mb: %d\n"
+                     "busy_cnt: %d\n",
+                     (cfs_atomic_read(&cli->cl_lru_in_list) +
+                       cfs_atomic_read(&cli->cl_lru_busy)) >> shift,
+                     cfs_atomic_read(&cli->cl_lru_busy));
+
+       return rc;
+}
+
+/* shrink the number of caching pages to a specific number */
+static int osc_wr_cached_mb(struct file *file, const char *buffer,
+                           unsigned long count, void *data)
+{
+       struct obd_device *dev = data;
+       struct client_obd *cli = &dev->u.cli;
+       int pages_number, mult, rc;
+
+       mult = 1 << (20 - PAGE_CACHE_SHIFT);
+       buffer = lprocfs_find_named_value(buffer, "used_mb:", &count);
+       rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
+       if (rc)
+               return rc;
+
+       if (pages_number < 0)
+               return -ERANGE;
+
+       rc = cfs_atomic_read(&cli->cl_lru_in_list) - pages_number;
+       if (rc > 0)
+               (void)osc_lru_shrink(cli, rc, true);
+
+       return count;
 }
 
 static int osc_rd_cur_dirty_bytes(char *page, char **start, off_t off,
@@ -263,149 +303,6 @@ static int osc_wr_grant_shrink_interval(struct file *file, const char *buffer,
         return count;
 }
 
-static int osc_rd_create_count(char *page, char **start, off_t off, int count,
-                               int *eof, void *data)
-{
-        struct obd_device *obd = data;
-
-        if (obd == NULL)
-                return 0;
-
-        return snprintf(page, count, "%d\n",
-                        obd->u.cli.cl_oscc.oscc_grow_count);
-}
-
-/**
- * Set OSC creator's osc_creator::oscc_grow_count
- *
- * \param file   proc file
- * \param buffer buffer containing the value
- * \param count  buffer size
- * \param data   obd device
- *
- * \retval \a count
- */
-static int osc_wr_create_count(struct file *file, const char *buffer,
-                               unsigned long count, void *data)
-{
-        struct obd_device *obd = data;
-        int val, rc, i;
-
-        if (obd == NULL)
-                return 0;
-
-        rc = lprocfs_write_helper(buffer, count, &val);
-        if (rc)
-                return rc;
-
-        /* The MDT ALWAYS needs to limit the precreate count to
-         * OST_MAX_PRECREATE, and the constant cannot be changed
-         * because it is a value shared between the OSC and OST
-         * that is the maximum possible number of objects that will
-         * ever be handled by MDT->OST recovery processing.
-         *
-         * If the OST ever gets a request to delete more orphans,
-         * this implies that something has gone badly on the MDT
-         * and the OST will refuse to delete so much data from the
-         * filesystem as a safety measure. */
-        if (val < OST_MIN_PRECREATE || val > OST_MAX_PRECREATE)
-                return -ERANGE;
-        if (val > obd->u.cli.cl_oscc.oscc_max_grow_count)
-                return -ERANGE;
-
-        for (i = 1; (i << 1) <= val; i <<= 1)
-                ;
-        obd->u.cli.cl_oscc.oscc_grow_count = i;
-
-        return count;
-}
-
-/**
- * Read OSC creator's osc_creator::oscc_max_grow_count
- *
- * \param page       buffer to hold the returning string
- * \param start
- * \param off
- * \param count
- * \param eof
- *              proc read function parameters, please refer to kernel
- *              code fs/proc/generic.c proc_file_read()
- * \param data   obd device
- *
- * \retval number of characters printed.
- */
-static int osc_rd_max_create_count(char *page, char **start, off_t off,
-                                   int count, int *eof, void *data)
-{
-        struct obd_device *obd = data;
-
-        if (obd == NULL)
-                return 0;
-
-        return snprintf(page, count, "%d\n",
-                        obd->u.cli.cl_oscc.oscc_max_grow_count);
-}
-
-/**
- * Set OSC creator's osc_creator::oscc_max_grow_count
- *
- * \param file   proc file
- * \param buffer buffer containing the value
- * \param count  buffer size
- * \param data   obd device
- *
- * \retval \a count
- */
-static int osc_wr_max_create_count(struct file *file, const char *buffer,
-                                   unsigned long count, void *data)
-{
-        struct obd_device *obd = data;
-        int val, rc;
-
-        if (obd == NULL)
-                return 0;
-
-        rc = lprocfs_write_helper(buffer, count, &val);
-        if (rc)
-                return rc;
-
-        if (val < 0)
-                return -ERANGE;
-        if (val > OST_MAX_PRECREATE)
-                return -ERANGE;
-
-        if (obd->u.cli.cl_oscc.oscc_grow_count > val)
-                obd->u.cli.cl_oscc.oscc_grow_count = val;
-
-        obd->u.cli.cl_oscc.oscc_max_grow_count = val;
-
-        return count;
-}
-
-static int osc_rd_prealloc_next_id(char *page, char **start, off_t off,
-                                   int count, int *eof, void *data)
-{
-        struct obd_device *obd = data;
-
-        if (obd == NULL)
-                return 0;
-
-        return snprintf(page, count, LPU64"\n",
-                        obd->u.cli.cl_oscc.oscc_next_id);
-}
-
-static int osc_rd_prealloc_last_id(char *page, char **start, off_t off,
-                                   int count, int *eof, void *data)
-{
-        struct obd_device *obd = data;
-
-        if (obd == NULL)
-                return 0;
-
-        return snprintf(page, count, LPU64"\n",
-                        obd->u.cli.cl_oscc.oscc_last_id);
-}
-
 static int osc_rd_checksum(char *page, char **start, off_t off, int count,
                            int *eof, void *data)
 {
@@ -474,7 +371,7 @@ static int osc_wd_checksum_type(struct file *file, const char *buffer,
 
         if (count > sizeof(kernbuf) - 1)
                 return -EINVAL;
-        if (cfs_copy_from_user(kernbuf, buffer, count))
+       if (copy_from_user(kernbuf, buffer, count))
                 return -EFAULT;
         if (count > 0 && kernbuf[count - 1] == '\n')
                 kernbuf[count - 1] = '\0';
@@ -571,18 +468,23 @@ static int lprocfs_osc_wr_max_pages_per_rpc(struct file *file,
        struct obd_device *dev = data;
        struct client_obd *cli = &dev->u.cli;
        struct obd_connect_data *ocd = &cli->cl_import->imp_connect_data;
-       int chunk_mask, val, rc;
+       int chunk_mask, rc;
+       __u64 val;
 
-       rc = lprocfs_write_helper(buffer, count, &val);
+       rc = lprocfs_write_u64_helper(buffer, count, &val);
        if (rc)
                return rc;
 
+       /* if the max_pages is specified in bytes, convert to pages */
+       if (val >= ONE_MB_BRW_SIZE)
+               val >>= PAGE_CACHE_SHIFT;
+
        LPROCFS_CLIMP_CHECK(dev);
 
-       chunk_mask = ~((1 << (cli->cl_chunkbits - CFS_PAGE_SHIFT)) - 1);
+       chunk_mask = ~((1 << (cli->cl_chunkbits - PAGE_CACHE_SHIFT)) - 1);
        /* max_pages_per_rpc must be chunk aligned */
        val = (val + ~chunk_mask) & chunk_mask;
-       if (val == 0 || val > ocd->ocd_brw_size >> CFS_PAGE_SHIFT) {
+       if (val == 0 || val > ocd->ocd_brw_size >> PAGE_CACHE_SHIFT) {
                LPROCFS_CLIMP_EXIT(dev);
                return -ERANGE;
        }
@@ -594,6 +496,22 @@ static int lprocfs_osc_wr_max_pages_per_rpc(struct file *file,
        return count;
 }
 
+static int osc_rd_unstable_stats(char *page, char **start, off_t off,
+                               int count, int *eof, void *data)
+{
+       struct obd_device *dev = data;
+       struct client_obd *cli = &dev->u.cli;
+       int pages, mb;
+
+       pages = cfs_atomic_read(&cli->cl_unstable_count);
+       mb    = (pages * PAGE_CACHE_SIZE) >> 20;
+
+       return snprintf(page, count,
+                       "unstable_pages: %8d\n"
+                       "unstable_mb:    %8d\n",
+                       pages, mb);
+}
+
 static struct lprocfs_vars lprocfs_osc_obd_vars[] = {
         { "uuid",            lprocfs_rd_uuid,        0, 0 },
         { "ping",            0, lprocfs_wr_ping,     0, 0, 0222 },
@@ -615,17 +533,13 @@ static struct lprocfs_vars lprocfs_osc_obd_vars[] = {
                                 osc_wr_max_rpcs_in_flight, 0 },
         { "destroys_in_flight", osc_rd_destroys_in_flight, 0, 0 },
         { "max_dirty_mb",    osc_rd_max_dirty_mb, osc_wr_max_dirty_mb, 0 },
+       { "osc_cached_mb",   osc_rd_cached_mb,     osc_wr_cached_mb, 0 },
         { "cur_dirty_bytes", osc_rd_cur_dirty_bytes, 0, 0 },
         { "cur_grant_bytes", osc_rd_cur_grant_bytes,
                              osc_wr_cur_grant_bytes, 0 },
         { "cur_lost_grant_bytes", osc_rd_cur_lost_grant_bytes, 0, 0},
         { "grant_shrink_interval", osc_rd_grant_shrink_interval,
                                    osc_wr_grant_shrink_interval, 0 },
-        { "create_count",    osc_rd_create_count, osc_wr_create_count, 0 },
-        { "max_create_count", osc_rd_max_create_count,
-                              osc_wr_max_create_count, 0},
-        { "prealloc_next_id", osc_rd_prealloc_next_id, 0, 0 },
-        { "prealloc_last_id", osc_rd_prealloc_last_id, 0, 0 },
         { "checksums",       osc_rd_checksum, osc_wr_checksum, 0 },
         { "checksum_type",   osc_rd_checksum_type, osc_wd_checksum_type, 0 },
         { "resend_count",    osc_rd_resend_count, osc_wr_resend_count, 0},
@@ -638,6 +552,8 @@ static struct lprocfs_vars lprocfs_osc_obd_vars[] = {
         { "state",           lprocfs_rd_state,         0, 0 },
         { "pinger_recov",    lprocfs_rd_pinger_recov,
                              lprocfs_wr_pinger_recov,  0, 0 },
+        { "unstable_stats",  osc_rd_unstable_stats, 0, 0},
+
         { 0 }
 };
 
@@ -654,11 +570,11 @@ static int osc_rpc_stats_seq_show(struct seq_file *seq, void *v)
         struct obd_device *dev = seq->private;
         struct client_obd *cli = &dev->u.cli;
         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
-        int i;
+       int i;
 
-        cfs_gettimeofday(&now);
+       do_gettimeofday(&now);
 
-        client_obd_list_lock(&cli->cl_loi_list_lock);
+       client_obd_list_lock(&cli->cl_loi_list_lock);
 
         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
                    now.tv_sec, now.tv_usec);
@@ -766,21 +682,21 @@ LPROC_SEQ_FOPS(osc_rpc_stats);
 
 static int osc_stats_seq_show(struct seq_file *seq, void *v)
 {
-        struct timeval now;
-        struct obd_device *dev = seq->private;
-        struct osc_stats *stats = &obd2osc_dev(dev)->od_stats;
+       struct timeval now;
+       struct obd_device *dev = seq->private;
+       struct osc_stats *stats = &obd2osc_dev(dev)->od_stats;
 
-        cfs_gettimeofday(&now);
+       do_gettimeofday(&now);
 
-        seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
-                   now.tv_sec, now.tv_usec);
-        seq_printf(seq, "lockless_write_bytes\t\t"LPU64"\n",
-                   stats->os_lockless_writes);
-        seq_printf(seq, "lockless_read_bytes\t\t"LPU64"\n",
-                   stats->os_lockless_reads);
-        seq_printf(seq, "lockless_truncate\t\t"LPU64"\n",
-                   stats->os_lockless_truncates);
-        return 0;
+       seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
+                  now.tv_sec, now.tv_usec);
+       seq_printf(seq, "lockless_write_bytes\t\t"LPU64"\n",
+                  stats->os_lockless_writes);
+       seq_printf(seq, "lockless_read_bytes\t\t"LPU64"\n",
+                  stats->os_lockless_reads);
+       seq_printf(seq, "lockless_truncate\t\t"LPU64"\n",
+                  stats->os_lockless_truncates);
+       return 0;
 }
 
 static ssize_t osc_stats_seq_write(struct file *file, const char *buf,
@@ -798,14 +714,15 @@ LPROC_SEQ_FOPS(osc_stats);
 
 int lproc_osc_attach_seqstat(struct obd_device *dev)
 {
-        int rc;
+       int rc;
 
-        rc = lprocfs_seq_create(dev->obd_proc_entry, "osc_stats", 0444,
-                                &osc_stats_fops, dev);
-        if (rc == 0)
-                rc = lprocfs_obd_seq_create(dev, "rpc_stats", 0444,
-                                            &osc_rpc_stats_fops, dev);
-        return rc;
+       rc = lprocfs_seq_create(dev->obd_proc_entry, "osc_stats", 0644,
+                               &osc_stats_fops, dev);
+       if (rc == 0)
+               rc = lprocfs_obd_seq_create(dev, "rpc_stats", 0644,
+                                           &osc_rpc_stats_fops, dev);
+
+       return rc;
 }
 
 void lprocfs_osc_init_vars(struct lprocfs_static_vars *lvars)