drivers/block/drbd/drbd_proc.c

   1 /*
   2    drbd_proc.c
   3
   4    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
   5
   6    Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
   7    Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
   8    Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
   9
  10    drbd is free software; you can redistribute it and/or modify
  11    it under the terms of the GNU General Public License as published by
  12    the Free Software Foundation; either version 2, or (at your option)
  13    any later version.
  14
  15    drbd is distributed in the hope that it will be useful,
  16    but WITHOUT ANY WARRANTY; without even the implied warranty of
  17    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18    GNU General Public License for more details.
  19
  20    You should have received a copy of the GNU General Public License
  21    along with drbd; see the file COPYING.  If not, write to
  22    the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  23
  24  */
  25
  26 #include <linux/module.h>
  27
  28 #include <linux/uaccess.h>
  29 #include <linux/fs.h>
  30 #include <linux/file.h>
  31 #include <linux/proc_fs.h>
  32 #include <linux/seq_file.h>
  33 #include <linux/drbd.h>
  34 #include "drbd_int.h"
  35
  36 struct proc_dir_entry *drbd_proc;
  37
  38 static void seq_printf_with_thousands_grouping(struct seq_file *seq, long v)
  39 {
  40         /* v is in kB/sec. We don't expect TiByte/sec yet. */
  41         if (unlikely(v >= 1000000)) {
  42                 /* cool: > GiByte/s */
  43                 seq_printf(seq, "%ld,", v / 1000000);
  44                 v %= 1000000;
  45                 seq_printf(seq, "%03ld,%03ld", v/1000, v % 1000);
  46         } else if (likely(v >= 1000))
  47                 seq_printf(seq, "%ld,%03ld", v/1000, v % 1000);
  48         else
  49                 seq_printf(seq, "%ld", v);
  50 }
  51
  52 static void drbd_get_syncer_progress(struct drbd_device *device,
  53                 union drbd_dev_state state, unsigned long *rs_total,
  54                 unsigned long *bits_left, unsigned int *per_mil_done)
  55 {
  56         /* this is to break it at compile time when we change that, in case we
  57          * want to support more than (1<<32) bits on a 32bit arch. */
  58         typecheck(unsigned long, device->rs_total);
  59         *rs_total = device->rs_total;
  60
  61         /* note: both rs_total and rs_left are in bits, i.e. in
  62          * units of BM_BLOCK_SIZE.
  63          * for the percentage, we don't care. */
  64
  65         if (state.conn == C_VERIFY_S || state.conn == C_VERIFY_T)
  66                 *bits_left = device->ov_left;
  67         else
  68                 *bits_left = drbd_bm_total_weight(device) - device->rs_failed;
  69         /* >> 10 to prevent overflow,
  70          * +1 to prevent division by zero */
  71         if (*bits_left > *rs_total) {
  72                 /* D'oh. Maybe a logic bug somewhere.  More likely just a race
  73                  * between state change and reset of rs_total.
  74                  */
  75                 *bits_left = *rs_total;
  76                 *per_mil_done = *rs_total ? 0 : 1000;
  77         } else {
  78                 /* Make sure the division happens in long context.
  79                  * We allow up to one petabyte storage right now,
  80                  * at a granularity of 4k per bit that is 2**38 bits.
  81                  * After shift right and multiplication by 1000,
  82                  * this should still fit easily into a 32bit long,
  83                  * so we don't need a 64bit division on 32bit arch.
  84                  * Note: currently we don't support such large bitmaps on 32bit
  85                  * arch anyways, but no harm done to be prepared for it here.
  86                  */
  87                 unsigned int shift = *rs_total > UINT_MAX ? 16 : 10;
  88                 unsigned long left = *bits_left >> shift;
  89                 unsigned long total = 1UL + (*rs_total >> shift);
  90                 unsigned long tmp = 1000UL - left * 1000UL/total;
  91                 *per_mil_done = tmp;
  92         }
  93 }
  94
  95
  96 /*lge
  97  * progress bars shamelessly adapted from driver/md/md.c
  98  * output looks like
  99  *      [=====>..............] 33.5% (23456/123456)
 100  *      finish: 2:20:20 speed: 6,345 (6,456) K/sec
 101  */
 102 static void drbd_syncer_progress(struct drbd_device *device, struct seq_file *seq,
 103                 union drbd_dev_state state)
 104 {
 105         unsigned long db, dt, dbdt, rt, rs_total, rs_left;
 106         unsigned int res;
 107         int i, x, y;
 108         int stalled = 0;
 109
 110         drbd_get_syncer_progress(device, state, &rs_total, &rs_left, &res);
 111
 112         x = res/50;
 113         y = 20-x;
 114         seq_puts(seq, "\t[");
 115         for (i = 1; i < x; i++)
 116                 seq_putc(seq, '=');
 117         seq_putc(seq, '>');
 118         for (i = 0; i < y; i++)
 119                 seq_putc(seq, '.');
 120         seq_puts(seq, "] ");
 121
 122         if (state.conn == C_VERIFY_S || state.conn == C_VERIFY_T)
 123                 seq_puts(seq, "verified:");
 124         else
 125                 seq_puts(seq, "sync'ed:");
 126         seq_printf(seq, "%3u.%u%% ", res / 10, res % 10);
 127
 128         /* if more than a few GB, display in MB */
 129         if (rs_total > (4UL << (30 - BM_BLOCK_SHIFT)))
 130                 seq_printf(seq, "(%lu/%lu)M",
 131                             (unsigned long) Bit2KB(rs_left >> 10),
 132                             (unsigned long) Bit2KB(rs_total >> 10));
 133         else
 134                 seq_printf(seq, "(%lu/%lu)K",
 135                             (unsigned long) Bit2KB(rs_left),
 136                             (unsigned long) Bit2KB(rs_total));
 137
 138         seq_puts(seq, "\n\t");
 139
 140         /* see drivers/md/md.c
 141          * We do not want to overflow, so the order of operands and
 142          * the * 100 / 100 trick are important. We do a +1 to be
 143          * safe against division by zero. We only estimate anyway.
 144          *
 145          * dt: time from mark until now
 146          * db: blocks written from mark until now
 147          * rt: remaining time
 148          */
 149         /* Rolling marks. last_mark+1 may just now be modified.  last_mark+2 is
 150          * at least (DRBD_SYNC_MARKS-2)*DRBD_SYNC_MARK_STEP old, and has at
 151          * least DRBD_SYNC_MARK_STEP time before it will be modified. */
 152         /* ------------------------ ~18s average ------------------------ */
 153         i = (device->rs_last_mark + 2) % DRBD_SYNC_MARKS;
 154         dt = (jiffies - device->rs_mark_time[i]) / HZ;
 155         if (dt > 180)
 156                 stalled = 1;
 157
 158         if (!dt)
 159                 dt++;
 160         db = device->rs_mark_left[i] - rs_left;
 161         rt = (dt * (rs_left / (db/100+1)))/100; /* seconds */
 162
 163         seq_printf(seq, "finish: %lu:%02lu:%02lu",
 164                 rt / 3600, (rt % 3600) / 60, rt % 60);
 165
 166         dbdt = Bit2KB(db/dt);
 167         seq_puts(seq, " speed: ");
 168         seq_printf_with_thousands_grouping(seq, dbdt);
 169         seq_puts(seq, " (");
 170         /* ------------------------- ~3s average ------------------------ */
 171         if (drbd_proc_details >= 1) {
 172                 /* this is what drbd_rs_should_slow_down() uses */
 173                 i = (device->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS;
 174                 dt = (jiffies - device->rs_mark_time[i]) / HZ;
 175                 if (!dt)
 176                         dt++;
 177                 db = device->rs_mark_left[i] - rs_left;
 178                 dbdt = Bit2KB(db/dt);
 179                 seq_printf_with_thousands_grouping(seq, dbdt);
 180                 seq_puts(seq, " -- ");
 181         }
 182
 183         /* --------------------- long term average ---------------------- */
 184         /* mean speed since syncer started
 185          * we do account for PausedSync periods */
 186         dt = (jiffies - device->rs_start - device->rs_paused) / HZ;
 187         if (dt == 0)
 188                 dt = 1;
 189         db = rs_total - rs_left;
 190         dbdt = Bit2KB(db/dt);
 191         seq_printf_with_thousands_grouping(seq, dbdt);
 192         seq_putc(seq, ')');
 193
 194         if (state.conn == C_SYNC_TARGET ||
 195             state.conn == C_VERIFY_S) {
 196                 seq_puts(seq, " want: ");
 197                 seq_printf_with_thousands_grouping(seq, device->c_sync_rate);
 198         }
 199         seq_printf(seq, " K/sec%s\n", stalled ? " (stalled)" : "");
 200
 201         if (drbd_proc_details >= 1) {
 202                 /* 64 bit:
 203                  * we convert to sectors in the display below. */
 204                 unsigned long bm_bits = drbd_bm_bits(device);
 205                 unsigned long bit_pos;
 206                 unsigned long long stop_sector = 0;
 207                 if (state.conn == C_VERIFY_S ||
 208                     state.conn == C_VERIFY_T) {
 209                         bit_pos = bm_bits - device->ov_left;
 210                         if (verify_can_do_stop_sector(device))
 211                                 stop_sector = device->ov_stop_sector;
 212                 } else
 213                         bit_pos = device->bm_resync_fo;
 214                 /* Total sectors may be slightly off for oddly
 215                  * sized devices. So what. */
 216                 seq_printf(seq,
 217                         "\t%3d%% sector pos: %llu/%llu",
 218                         (int)(bit_pos / (bm_bits/100+1)),
 219                         (unsigned long long)bit_pos * BM_SECT_PER_BIT,
 220                         (unsigned long long)bm_bits * BM_SECT_PER_BIT);
 221                 if (stop_sector != 0 && stop_sector != ULLONG_MAX)
 222                         seq_printf(seq, " stop sector: %llu", stop_sector);
 223                 seq_putc(seq, '\n');
 224         }
 225 }
 226
 227 int drbd_seq_show(struct seq_file *seq, void *v)
 228 {
 229         int i, prev_i = -1;
 230         const char *sn;
 231         struct drbd_device *device;
 232         struct net_conf *nc;
 233         union drbd_dev_state state;
 234         char wp;
 235
 236         static char write_ordering_chars[] = {
 237                 [WO_NONE] = 'n',
 238                 [WO_DRAIN_IO] = 'd',
 239                 [WO_BDEV_FLUSH] = 'f',
 240         };
 241
 242         seq_printf(seq, "version: " REL_VERSION " (api:%d/proto:%d-%d)\n%s\n",
 243                    API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX, drbd_buildtag());
 244
 245         /*
 246           cs .. connection state
 247           ro .. node role (local/remote)
 248           ds .. disk state (local/remote)
 249              protocol
 250              various flags
 251           ns .. network send
 252           nr .. network receive
 253           dw .. disk write
 254           dr .. disk read
 255           al .. activity log write count
 256           bm .. bitmap update write count
 257           pe .. pending (waiting for ack or data reply)
 258           ua .. unack'd (still need to send ack or data reply)
 259           ap .. application requests accepted, but not yet completed
 260           ep .. number of epochs currently "on the fly", P_BARRIER_ACK pending
 261           wo .. write ordering mode currently in use
 262          oos .. known out-of-sync kB
 263         */
 264
 265         rcu_read_lock();
 266         idr_for_each_entry(&drbd_devices, device, i) {
 267                 if (prev_i != i - 1)
 268                         seq_putc(seq, '\n');
 269                 prev_i = i;
 270
 271                 state = device->state;
 272                 sn = drbd_conn_str(state.conn);
 273
 274                 if (state.conn == C_STANDALONE &&
 275                     state.disk == D_DISKLESS &&
 276                     state.role == R_SECONDARY) {
 277                         seq_printf(seq, "%2d: cs:Unconfigured\n", i);
 278                 } else {
 279                         /* reset device->congestion_reason */
 280                         bdi_rw_congested(device->rq_queue->backing_dev_info);
 281
 282                         nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
 283                         wp = nc ? nc->wire_protocol - DRBD_PROT_A + 'A' : ' ';
 284                         seq_printf(seq,
 285                            "%2d: cs:%s ro:%s/%s ds:%s/%s %c %c%c%c%c%c%c\n"
 286                            "    ns:%u nr:%u dw:%u dr:%u al:%u bm:%u "
 287                            "lo:%d pe:%d ua:%d ap:%d ep:%d wo:%c",
 288                            i, sn,
 289                            drbd_role_str(state.role),
 290                            drbd_role_str(state.peer),
 291                            drbd_disk_str(state.disk),
 292                            drbd_disk_str(state.pdsk),
 293                            wp,
 294                            drbd_suspended(device) ? 's' : 'r',
 295                            state.aftr_isp ? 'a' : '-',
 296                            state.peer_isp ? 'p' : '-',
 297                            state.user_isp ? 'u' : '-',
 298                            device->congestion_reason ?: '-',
 299                            test_bit(AL_SUSPENDED, &device->flags) ? 's' : '-',
 300                            device->send_cnt/2,
 301                            device->recv_cnt/2,
 302                            device->writ_cnt/2,
 303                            device->read_cnt/2,
 304                            device->al_writ_cnt,
 305                            device->bm_writ_cnt,
 306                            atomic_read(&device->local_cnt),
 307                            atomic_read(&device->ap_pending_cnt) +
 308                            atomic_read(&device->rs_pending_cnt),
 309                            atomic_read(&device->unacked_cnt),
 310                            atomic_read(&device->ap_bio_cnt),
 311                            first_peer_device(device)->connection->epochs,
 312                            write_ordering_chars[device->resource->write_ordering]
 313                         );
 314                         seq_printf(seq, " oos:%llu\n",
 315                                    Bit2KB((unsigned long long)
 316                                            drbd_bm_total_weight(device)));
 317                 }
 318                 if (state.conn == C_SYNC_SOURCE ||
 319                     state.conn == C_SYNC_TARGET ||
 320                     state.conn == C_VERIFY_S ||
 321                     state.conn == C_VERIFY_T)
 322                         drbd_syncer_progress(device, seq, state);
 323
 324                 if (drbd_proc_details >= 1 && get_ldev_if_state(device, D_FAILED)) {
 325                         lc_seq_printf_stats(seq, device->resync);
 326                         lc_seq_printf_stats(seq, device->act_log);
 327                         put_ldev(device);
 328                 }
 329
 330                 if (drbd_proc_details >= 2)
 331                         seq_printf(seq, "\tblocked on activity log: %d\n", atomic_read(&device->ap_actlog_cnt));
 332         }
 333         rcu_read_unlock();
 334
 335         return 0;
 336 }