From: Mikulas Patocka <mpatocka@xxxxxxxxxx>
Support the collection of I/O statistics on user-defined regions of
a DM device. Only bio-based DM devices are currently supported.
Each user-defined region specifies a starting sector, ending sector
and step. Individual statistics will be collected for each step-sized
area between the starting and ending sector.
The I/O statistics counters for each step-sized area of a region are
in the same format as /sys/block/*/stat or /proc/diskstats but extra
counters (12 and 13) are provided: total time spent reading and
writing in milliseconds. These counters may be accessed by sending
the @stats_print message to the appropriate DM device via dmsetup.
DM statistics creation will allocate memory via kmalloc or fallback to
using vmalloc space. At most, 1/4 of the overall system memory may be
allocated by DM statistics. The admin can see how much memory is used
by reading /sys/module/dm_mod/parameters/stats_current_allocated_bytes
See Documentation/device-mapper/statistics.txt for more details.
Signed-off-by: Mikulas Patocka <mpatocka@xxxxxxxxxx>
Signed-off-by: Mike Snitzer <snitzer@xxxxxxxxxx>
---
Documentation/device-mapper/statistics.txt | 121 ++++
drivers/md/Makefile | 2 +-
drivers/md/dm-ioctl.c | 22 +-
drivers/md/dm-stats.c | 946 ++++++++++++++++++++++++++++
drivers/md/dm-stats.h | 40 ++
drivers/md/dm.c | 63 ++-
drivers/md/dm.h | 17 +
7 files changed, 1200 insertions(+), 11 deletions(-)
diff --git a/Documentation/device-mapper/statistics.txt b/Documentation/device-mapper/statistics.txt
new file mode 100644
index 0000000..381df75
--- /dev/null
+++ b/Documentation/device-mapper/statistics.txt
@@ -0,0 +1,121 @@
+DM statistics
+=============
+
+Device Mapper supporta the collection of I/O statistics on user-defined
+regions of a DM device. Only bio-based DM devices are currently
+supported.
+
+Each user-defined region specifies a starting sector, ending sector
+and step. Individual statistics will be collected for each step-sized
+area between the starting and ending sector.
+
+The I/O statistics counters for each step-sized area of a region are
+in the same format as /sys/block/*/stat or /proc/diskstats but extra
+counters (12 and 13) are provided: total time spent reading and
+writing in milliseconds. These counters may be accessed by sending the
+@stats_print message to the appropriate DM device via dmsetup.
+
+Each region has a corresponding unique identifier that is assigned when
+the region is created. The region id must be supplied when querying
+statistics about the region, deleting the region, etc. Unique region
+ids enable multiple userspace programs to request and process statistics
+for the same DM device without stepping over each other's data.
+
+DM statistics creation will allocate memory via kmalloc or fallback to
+using vmalloc space. At most, 1/4 of the overall system memory may be
+allocated by DM statistics. The admin can see how much memory is used
+by reading /sys/module/dm_mod/parameters/stats_current_allocated_bytes
+
+Messages
+========
+
+ @stats_create <range> <step> [<program_id> [<program_params>]]
+
+ Create a new region and return the region id.
+ <range>
+ "-" - whole device
+ "<start>-<end>" - a specified range in 512-byte sectors
+ <step>
+ "<number>" - the number of sectors in each area
+ "/<number>" - the range is subdivided into the specified
+ number of areas
+ <program_id>
+ An optional parameter. The string that identifies a program
+ that created this range. The kernel returns this string back
+ in the output of @stats_list message, but it doesn't use this
+ value for anything.
+ <program_params>
+ An optional parameter. The string that identifies parameters
+ for a program that created this range. The kernel returns this
+ string back in the output of @stats_list message, but it
+ doesn't use this value for anything.
+
+ @stats_clear <region id>
+
+ Clear all the counters except the in-flight i/o counters.
+ <region id>
+ region id returned from @stats_create
+
+ @stats_print <region id> [<starting_line> <number_of_lines>]
+
+ Print counters for each step-sized area of a region.
+ <region id>
+ region id returned from @stats_create
+ <starting_line>
+ The index of the starting line in the output. If omitted, all
+ lines are returned.
+ <number_of_lines>
+ The number of lines in the output. If omitted, all lines are
+ returned.
+ Output format for each step-sized area of a region:
+ <starting sector>-<ending sector> counters
+ The first 11 counters have the same meaning as
+ /sys/block/*/stat or /proc/diskstats.
+ Additional counters:
+ 12. the total time spent reading in milliseconds
+ 13. the total time spent writing in milliseconds
+
+ @stats_print_clear <region id> [<starting_line> <number_of_lines>]
+
+ Atomically print and then clear all the counters except the
+ in-flight i/o counters. Useful when client consuming the
+ statistics does not want to lose any statistics (those updated
+ between printing and clearing).
+ <region id>
+ region id returned from @stats_create
+ <starting_line>
+ The index of the starting line in the output. If omitted, all
+ lines are returned/cleared.
+ <number_of_lines>
+ The number of lines in the output. If omitted, all lines are
+ returned/cleared.
+
+ @stats_delete <region id>
+
+ Delete the region with the specified id.
+ <region id>
+ region id returned from @stats_create
+
+ @stats_list [<program_id>]
+
+ List all regions registered with @stats_create.
+ <program_id>
+ An optional parameter specifying the program that queries the
+ parameters. If this parameter is specified, only region
+ belonging to this program are returned. If it is not
+ specified, all regions are returned.
+ Output format:
+ <id>: <start>-<end> <step size> <program_id> <program_params>
+
+Examples
+========
+
+Subdivide the DM device 'vol' into 100 pieces and start collecting
+statistics on them:
+dmsetup message vol 0 @stats_create - /100
+
+Print the statistics:
+dmsetup message vol 0 @stats_print 0
+
+Delete the statistics:
+dmsetup message vol 0 @stats_delete 0
diff --git a/drivers/md/Makefile b/drivers/md/Makefile
index 5ef78ef..2acc43f 100644
--- a/drivers/md/Makefile
+++ b/drivers/md/Makefile
@@ -3,7 +3,7 @@
#
dm-mod-y += dm.o dm-table.o dm-target.o dm-linear.o dm-stripe.o \
- dm-ioctl.o dm-io.o dm-kcopyd.o dm-sysfs.o
+ dm-ioctl.o dm-io.o dm-kcopyd.o dm-sysfs.o dm-stats.o
dm-multipath-y += dm-path-selector.o dm-mpath.o
dm-snapshot-y += dm-snap.o dm-exception-store.o dm-snap-transient.o \
dm-snap-persistent.o
diff --git a/drivers/md/dm-ioctl.c b/drivers/md/dm-ioctl.c
index f1b7586..40e6dce 100644
--- a/drivers/md/dm-ioctl.c
+++ b/drivers/md/dm-ioctl.c
@@ -1455,20 +1455,26 @@ static int table_status(struct dm_ioctl *param, size_t param_size)
return 0;
}
-static bool buffer_test_overflow(char *result, unsigned maxlen)
-{
- return !maxlen || strlen(result) + 1 >= maxlen;
-}
-
/*
- * Process device-mapper dependent messages.
+ * Process device-mapper dependent messages. Messages prefixed with '@'
+ * are processed by the DM core. All others are delivered to the target.
* Returns a number <= 1 if message was processed by device mapper.
* Returns 2 if message should be delivered to the target.
*/
static int message_for_md(struct mapped_device *md, unsigned argc, char **argv,
char *result, unsigned maxlen)
{
- return 2;
+ int r;
+
+ if (**argv != '@')
+ return 2; /* no '@' prefix, deliver to target */
+
+ r = dm_stats_message(md, argc, argv, result, maxlen);
+ if (r < 2)
+ return r;
+
+ DMERR("Unsupported message sent to DM core: %s", argv[0]);
+ return -EINVAL;
}
/*
@@ -1542,7 +1548,7 @@ static int target_message(struct dm_ioctl *param, size_t param_size)
if (r == 1) {
param->flags |= DM_DATA_OUT_FLAG;
- if (buffer_test_overflow(result, maxlen))
+ if (dm_message_test_buffer_overflow(result, maxlen))
param->flags |= DM_BUFFER_FULL_FLAG;
else
param->data_size = param->data_start + strlen(result) + 1;
diff --git a/drivers/md/dm-stats.c b/drivers/md/dm-stats.c
new file mode 100644
index 0000000..35b9037
--- /dev/null
+++ b/drivers/md/dm-stats.c
@@ -0,0 +1,946 @@
+#include <linux/errno.h>
+#include <linux/numa.h>
+#include <linux/slab.h>
+#include <linux/rculist.h>
+#include <linux/threads.h>
+#include <linux/preempt.h>
+#include <linux/irqflags.h>
+#include <linux/vmalloc.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/device-mapper.h>
+
+#include "dm.h"
+#include "dm-stats.h"
+
+#define DM_MSG_PREFIX "stats"
+
+static int dm_stat_need_rcu_barrier;
+
+/*
+ * Using 64-bit values to avoid overflow (which is a
+ * problem that block/genhd.c's IO accounting has).
+ */
+struct dm_stat_percpu {
+ unsigned long long sectors[2];
+ unsigned long long ios[2];
+ unsigned long long merges[2];
+ unsigned long long ticks[2];
+ unsigned long long io_ticks[2];
+ unsigned long long io_ticks_total;
+ unsigned long long time_in_queue;
+};
+
+struct dm_stat_shared {
+ atomic_t in_flight[2];
+ unsigned long stamp;
+ struct dm_stat_percpu tmp;
+};
+
+struct dm_stat {
+ struct list_head list_entry;
+ int id;
+ size_t n_entries;
+ sector_t start;
+ sector_t end;
+ sector_t step;
+ const char *program;
+ const char *aux;
+ struct rcu_head rcu_head;
+ size_t shared_alloc_size;
+ size_t percpu_alloc_size;
+ struct dm_stat_percpu *stat_percpu[NR_CPUS];
+ struct dm_stat_shared stat_shared[0];
+};
+
+struct dm_stats_last_position {
+ sector_t last_sector;
+ unsigned last_rw;
+};
+
+/*
+ * A typo on the command line could possibly make the kernel run out of memory
+ * and crash. To prevent the crash we account all used memory. We fail if we
+ * exhaust 1/4 of all memory or 1/2 of vmalloc space.
+ */
+#define DM_STATS_MEMORY_RATIO 1 / 4
+#define DM_STATS_VMALLOC_RATIO 1 / 2
+
+static DEFINE_SPINLOCK(shared_memory_lock);
+
+static unsigned long shared_memory_amount;
+
+static bool __check_shared_memory(size_t alloc_size)
+{
+ size_t a;
+
+ a = shared_memory_amount + alloc_size;
+ if (a < shared_memory_amount)
+ return false;
+ if (a >> PAGE_SHIFT > totalram_pages * DM_STATS_MEMORY_RATIO)
+ return false;
+#ifdef CONFIG_MMU
+ if (a > (VMALLOC_END - VMALLOC_START) * DM_STATS_VMALLOC_RATIO)
+ return false;
+#endif
+ return true;
+}
+
+static bool check_shared_memory(size_t alloc_size)
+{
+ bool ret;
+
+ spin_lock_irq(&shared_memory_lock);
+
+ ret = __check_shared_memory(alloc_size);
+
+ spin_unlock_irq(&shared_memory_lock);
+
+ return ret;
+}
+
+static bool claim_shared_memory(size_t alloc_size)
+{
+ spin_lock_irq(&shared_memory_lock);
+
+ if (!__check_shared_memory(alloc_size)) {
+ spin_unlock_irq(&shared_memory_lock);
+ return false;
+ }
+
+ shared_memory_amount += alloc_size;
+
+ spin_unlock_irq(&shared_memory_lock);
+
+ return true;
+}
+
+static void free_shared_memory(size_t alloc_size)
+{
+ unsigned long flags;
+
+ if (WARN_ON_ONCE(shared_memory_amount < alloc_size)) {
+ DMCRIT("Memory usage accounting bug.");
+ return;
+ }
+
+ spin_lock_irqsave(&shared_memory_lock, flags);
+
+ shared_memory_amount -= alloc_size;
+
+ spin_unlock_irqrestore(&shared_memory_lock, flags);
+}
+
+static void *dm_kvzalloc(size_t alloc_size, int node)
+{
+ void *p;
+
+ if (!claim_shared_memory(alloc_size))
+ return NULL;
+
+ if (alloc_size <= KMALLOC_MAX_SIZE) {
+ p = kzalloc_node(alloc_size, GFP_KERNEL | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN, node);
+ if (p)
+ return p;
+ }
+ p = vzalloc_node(alloc_size, node);
+ if (p)
+ return p;
+
+ free_shared_memory(alloc_size);
+
+ return NULL;
+}
+
+static void dm_kvfree(void *ptr, size_t alloc_size)
+{
+ if (!ptr)
+ return;
+
+ free_shared_memory(alloc_size);
+
+ if (is_vmalloc_addr(ptr))
+ vfree(ptr);
+ else
+ kfree(ptr);
+}
+
+static void dm_stat_free(struct rcu_head *head)
+{
+ int cpu;
+ struct dm_stat *s = container_of(head, struct dm_stat, rcu_head);
+
+ kfree(s->program);
+ kfree(s->aux);
+ for_each_possible_cpu(cpu)
+ dm_kvfree(s->stat_percpu[cpu], s->percpu_alloc_size);
+ dm_kvfree(s, s->shared_alloc_size);
+}
+
+static int dm_stat_in_flight(struct dm_stat_shared *shared)
+{
+ return atomic_read(&shared->in_flight[READ]) +
+ atomic_read(&shared->in_flight[WRITE]);
+}
+
+void dm_stats_init(struct dm_stats *stats)
+{
+ int cpu;
+ struct dm_stats_last_position *last;
+
+ mutex_init(&stats->mutex);
+ INIT_LIST_HEAD(&stats->list);
+ stats->last = alloc_percpu(struct dm_stats_last_position);
+ for_each_possible_cpu(cpu) {
+ last = per_cpu_ptr(stats->last, cpu);
+ last->last_sector = -1;
+ last->last_rw = -1;
+ }
+}
+
+void dm_stats_cleanup(struct dm_stats *stats)
+{
+ size_t ni;
+ struct dm_stat *s;
+ struct dm_stat_shared *shared;
+
+ while (!list_empty(&stats->list)) {
+ s = container_of(stats->list.next, struct dm_stat, list_entry);
+ list_del(&s->list_entry);
+ for (ni = 0; ni < s->n_entries; ni++) {
+ shared = &s->stat_shared[ni];
+ if (WARN_ON(dm_stat_in_flight(shared))) {
+ DMCRIT("leaked in-flight counter at index %lu "
+ "(start %llu, end %llu, step %llu): reads %d, writes %d",
+ (unsigned long)ni,
+ (unsigned long long)s->start,
+ (unsigned long long)s->end,
+ (unsigned long long)s->step,
+ atomic_read(&shared->in_flight[READ]),
+ atomic_read(&shared->in_flight[WRITE]));
+ }
+ }
+ dm_stat_free(&s->rcu_head);
+ }
+ free_percpu(stats->last);
+}
+
+static int dm_stats_create(struct dm_stats *stats, sector_t start, sector_t end,
+ sector_t step, const char *program, const char *aux,
+ void (*suspend_callback)(struct mapped_device *),
+ void (*resume_callback)(struct mapped_device *),
+ struct mapped_device *md)
+{
+ struct list_head *l;
+ struct dm_stat *s, *tmp_s;
+ sector_t n_entries;
+ size_t ni;
+ size_t shared_alloc_size;
+ size_t percpu_alloc_size;
+ struct dm_stat_percpu *p;
+ int cpu;
+ int ret_id;
+ int r;
+
+ if (end < start || !step)
+ return -EINVAL;
+
+ n_entries = end - start;
+ if (sector_div(n_entries, step))
+ n_entries++;
+
+ if (n_entries != (size_t)n_entries || !(n_entries + 1))
+ return -EOVERFLOW;
+
+ shared_alloc_size = sizeof(struct dm_stat) + (size_t)n_entries * sizeof(struct dm_stat_shared);
+ if ((shared_alloc_size - sizeof(struct dm_stat)) / sizeof(struct dm_stat_shared) != n_entries)
+ return -EOVERFLOW;
+
+ percpu_alloc_size = (size_t)n_entries * sizeof(struct dm_stat_percpu);
+ if (percpu_alloc_size / sizeof(struct dm_stat_percpu) != n_entries)
+ return -EOVERFLOW;
+
+ if (!check_shared_memory(shared_alloc_size + num_possible_cpus() * percpu_alloc_size))
+ return -ENOMEM;
+
+ s = dm_kvzalloc(shared_alloc_size, NUMA_NO_NODE);
+ if (!s)
+ return -ENOMEM;
+
+ s->n_entries = n_entries;
+ s->start = start;
+ s->end = end;
+ s->step = step;
+ s->shared_alloc_size = shared_alloc_size;
+ s->percpu_alloc_size = percpu_alloc_size;
+
+ s->program = kstrdup(program, GFP_KERNEL);
+ if (!s->program) {
+ r = -ENOMEM;
+ goto out;
+ }
+ s->aux = kstrdup(aux, GFP_KERNEL);
+ if (!s->aux) {
+ r = -ENOMEM;
+ goto out;
+ }
+
+ for (ni = 0; ni < n_entries; ni++) {
+ atomic_set(&s->stat_shared[ni].in_flight[READ], 0);
+ atomic_set(&s->stat_shared[ni].in_flight[WRITE], 0);
+ }
+
+ for_each_possible_cpu(cpu) {
+ p = dm_kvzalloc(percpu_alloc_size, cpu_to_node(cpu));
+ if (!p) {
+ r = -ENOMEM;
+ goto out;
+ }
+ s->stat_percpu[cpu] = p;
+ }
+
+ /*
+ * Suspend/resume to make sure there is no i/o in flight,
+ * so that newly created statistics will be exact.
+ *
+ * (note: we couldn't suspend earlier because we must not
+ * allocate memory while suspended)
+ */
+ suspend_callback(md);
+
+ mutex_lock(&stats->mutex);
+ s->id = 0;
+ list_for_each(l, &stats->list) {
+ tmp_s = container_of(l, struct dm_stat, list_entry);
+ if (WARN_ON(tmp_s->id < s->id)) {
+ r = -EINVAL;
+ goto out_unlock_resume;
+ }
+ if (tmp_s->id > s->id)
+ break;
+ if (unlikely(s->id == INT_MAX)) {
+ r = -ENFILE;
+ goto out_unlock_resume;
+ }
+ s->id++;
+ }
+ ret_id = s->id;
+ list_add_tail_rcu(&s->list_entry, l);
+ mutex_unlock(&stats->mutex);
+
+ resume_callback(md);
+
+ return ret_id;
+
+out_unlock_resume:
+ mutex_unlock(&stats->mutex);
+ resume_callback(md);
+out:
+ dm_stat_free(&s->rcu_head);
+ return r;
+}
+
+static struct dm_stat *__dm_stats_find(struct dm_stats *stats, int id)
+{
+ struct dm_stat *s;
+
+ list_for_each_entry(s, &stats->list, list_entry) {
+ if (s->id > id)
+ break;
+ if (s->id == id)
+ return s;
+ }
+
+ return NULL;
+}
+
+static int dm_stats_delete(struct dm_stats *stats, int id)
+{
+ struct dm_stat *s;
+ int cpu;
+
+ mutex_lock(&stats->mutex);
+
+ s = __dm_stats_find(stats, id);
+ if (!s) {
+ mutex_unlock(&stats->mutex);
+ return -ENOENT;
+ }
+
+ list_del_rcu(&s->list_entry);
+ mutex_unlock(&stats->mutex);
+
+ /*
+ * vfree can't be called from RCU callback
+ */
+ for_each_possible_cpu(cpu)
+ if (is_vmalloc_addr(s->stat_percpu))
+ goto do_sync_free;
+ if (is_vmalloc_addr(s)) {
+do_sync_free:
+ synchronize_rcu_expedited();
+ dm_stat_free(&s->rcu_head);
+ } else {
+ ACCESS_ONCE(dm_stat_need_rcu_barrier) = 1;
+ call_rcu(&s->rcu_head, dm_stat_free);
+ }
+ return 0;
+}
+
+static int dm_stats_list(struct dm_stats *stats, const char *program,
+ char *result, unsigned maxlen)
+{
+ struct dm_stat *s;
+ unsigned sz = 0;
+
+ mutex_lock(&stats->mutex);
+ list_for_each_entry(s, &stats->list, list_entry) {
+ if (!program || !strcmp(program, s->program))
+ DMEMIT("%d: %llu-%llu %llu %s %s\n", s->id,
+ (unsigned long long)s->start,
+ (unsigned long long)s->end,
+ (unsigned long long)s->step,
+ s->program,
+ s->aux);
+ }
+ mutex_unlock(&stats->mutex);
+
+ return 1;
+}
+
+static void dm_stat_round(struct dm_stat_shared *shared, struct dm_stat_percpu *p)
+{
+ /*
+ * This is racy, but so is part_round_stats_single.
+ */
+ unsigned long now = jiffies;
+ unsigned in_flight_read;
+ unsigned in_flight_write;
+ unsigned long difference = now - shared->stamp;
+
+ if (!difference)
+ return;
+ in_flight_read = atomic_read(&shared->in_flight[READ]);
+ in_flight_write = atomic_read(&shared->in_flight[WRITE]);
+ if (in_flight_read)
+ p->io_ticks[READ] += difference;
+ if (in_flight_write)
+ p->io_ticks[WRITE] += difference;
+ if (in_flight_read + in_flight_write) {
+ p->io_ticks_total += difference;
+ p->time_in_queue += (in_flight_read + in_flight_write) * difference;
+ }
+ shared->stamp = now;
+}
+
+static void dm_stat_for_entry(struct dm_stat *s, size_t entry,
+ unsigned long bi_rw, unsigned len, bool merged,
+ bool end, unsigned long duration)
+{
+ unsigned long idx = bi_rw & REQ_WRITE;
+ struct dm_stat_shared *shared = &s->stat_shared[entry];
+ struct dm_stat_percpu *p;
+
+ /*
+ * For strict correctness we should use local_irq_disable/enable
+ * instead of preempt_disable/enable.
+ *
+ * This is racy if the driver finishes bios from non-interrupt
+ * context as well as from interrupt context or from more different
+ * interrupts.
+ *
+ * However, the race only results in not counting some events,
+ * so it is acceptable.
+ *
+ * part_stat_lock()/part_stat_unlock() have this race too.
+ */
+ preempt_disable();
+ p = &s->stat_percpu[smp_processor_id()][entry];
+
+ if (!end) {
+ dm_stat_round(shared, p);
+ atomic_inc(&shared->in_flight[idx]);
+ } else {
+ dm_stat_round(shared, p);
+ atomic_dec(&shared->in_flight[idx]);
+ p->sectors[idx] += len;
+ p->ios[idx] += 1;
+ p->merges[idx] += merged;
+ p->ticks[idx] += duration;
+ }
+
+ preempt_enable();
+}
+
+static void __dm_stat_bio(struct dm_stat *s, unsigned long bi_rw,
+ sector_t bi_sector, sector_t end_sector,
+ bool end, unsigned long duration,
+ struct dm_stats_aux *aux)
+{
+ sector_t rel_sector, offset;
+ unsigned todo;
+ size_t entry;
+ unsigned fragment_len;
+
+ if (end_sector <= s->start || bi_sector >= s->end)
+ return;
+ if (unlikely(bi_sector < s->start)) {
+ rel_sector = 0;
+ todo = end_sector - s->start;
+ } else {
+ rel_sector = bi_sector - s->start;
+ todo = end_sector - bi_sector;
+ }
+ if (unlikely(end_sector > s->end))
+ todo -= end_sector - s->end;
+
+ offset = sector_div(rel_sector, s->step);
+ entry = rel_sector;
+ do {
+ if (WARN_ON_ONCE(entry >= s->n_entries)) {
+ DMCRIT("Invalid area access in region id %d", s->id);
+ return;
+ }
+ fragment_len = todo;
+ if (fragment_len > s->step - offset)
+ fragment_len = s->step - offset;
+ dm_stat_for_entry(s, entry, bi_rw, fragment_len,
+ aux->merged, end, duration);
+ todo -= fragment_len;
+ entry++;
+ offset = 0;
+ } while (unlikely(todo != 0));
+}
+
+void dm_stats_account_io(struct dm_stats *stats, unsigned long bi_rw,
+ sector_t bi_sector, unsigned bi_sectors, bool end,
+ unsigned long duration, struct dm_stats_aux *aux)
+{
+ struct dm_stat *s;
+ sector_t end_sector;
+ struct dm_stats_last_position *last;
+
+ if (unlikely(!bi_sectors))
+ return;
+
+ end_sector = bi_sector + bi_sectors;
+
+ if (!end) {
+ /*
+ * A race condition can at worst cause that the merged flag is
+ * misrepresented, so we don't have to disable preempt here.
+ */
+ last = __this_cpu_ptr(stats->last);
+ aux->merged = bi_sector == ACCESS_ONCE(last->last_sector) &&
+ (bi_rw & (REQ_WRITE | REQ_DISCARD)) ==
+ (ACCESS_ONCE(last->last_rw) & (REQ_WRITE | REQ_DISCARD));
+ ACCESS_ONCE(last->last_sector) = end_sector;
+ ACCESS_ONCE(last->last_rw) = bi_rw;
+ }
+
+ rcu_read_lock();
+
+ list_for_each_entry_rcu(s, &stats->list, list_entry)
+ __dm_stat_bio(s, bi_rw, bi_sector, end_sector, end, duration, aux);
+
+ rcu_read_unlock();
+}
+
+static void __dm_stat_init_temporary_percpu_totals(struct dm_stat_shared *shared,
+ struct dm_stat *s, size_t x)
+{
+ int cpu;
+ struct dm_stat_percpu *p;
+
+ local_irq_disable();
+ p = &s->stat_percpu[smp_processor_id()][x];
+ dm_stat_round(shared, p);
+ local_irq_enable();
+
+ memset(&shared->tmp, 0, sizeof(shared->tmp));
+ for_each_possible_cpu(cpu) {
+ p = &s->stat_percpu[cpu][x];
+ shared->tmp.sectors[READ] += ACCESS_ONCE(p->sectors[READ]);
+ shared->tmp.sectors[WRITE] += ACCESS_ONCE(p->sectors[WRITE]);
+ shared->tmp.ios[READ] += ACCESS_ONCE(p->ios[READ]);
+ shared->tmp.ios[WRITE] += ACCESS_ONCE(p->ios[WRITE]);
+ shared->tmp.merges[READ] += ACCESS_ONCE(p->merges[READ]);
+ shared->tmp.merges[WRITE] += ACCESS_ONCE(p->merges[WRITE]);
+ shared->tmp.ticks[READ] += ACCESS_ONCE(p->ticks[READ]);
+ shared->tmp.ticks[WRITE] += ACCESS_ONCE(p->ticks[WRITE]);
+ shared->tmp.io_ticks[READ] += ACCESS_ONCE(p->io_ticks[READ]);
+ shared->tmp.io_ticks[WRITE] += ACCESS_ONCE(p->io_ticks[WRITE]);
+ shared->tmp.io_ticks_total += ACCESS_ONCE(p->io_ticks_total);
+ shared->tmp.time_in_queue += ACCESS_ONCE(p->time_in_queue);
+ }
+}
+
+static void __dm_stat_clear(struct dm_stat *s, size_t idx_start, size_t idx_end,
+ bool init_tmp_percpu_totals)
+{
+ size_t x;
+ struct dm_stat_shared *shared;
+ struct dm_stat_percpu *p;
+
+ for (x = idx_start; x < idx_end; x++) {
+ shared = &s->stat_shared[x];
+ if (init_tmp_percpu_totals)
+ __dm_stat_init_temporary_percpu_totals(shared, s, x);
+ local_irq_disable();
+ p = &s->stat_percpu[smp_processor_id()][x];
+ p->sectors[READ] -= shared->tmp.sectors[READ];
+ p->sectors[WRITE] -= shared->tmp.sectors[WRITE];
+ p->ios[READ] -= shared->tmp.ios[READ];
+ p->ios[WRITE] -= shared->tmp.ios[WRITE];
+ p->merges[READ] -= shared->tmp.merges[READ];
+ p->merges[WRITE] -= shared->tmp.merges[WRITE];
+ p->ticks[READ] -= shared->tmp.ticks[READ];
+ p->ticks[WRITE] -= shared->tmp.ticks[WRITE];
+ p->io_ticks[READ] -= shared->tmp.io_ticks[READ];
+ p->io_ticks[WRITE] -= shared->tmp.io_ticks[WRITE];
+ p->io_ticks_total -= shared->tmp.io_ticks_total;
+ p->time_in_queue -= shared->tmp.time_in_queue;
+ local_irq_enable();
+ }
+}
+
+static int dm_stats_clear(struct dm_stats *stats, int id)
+{
+ struct dm_stat *s;
+
+ mutex_lock(&stats->mutex);
+
+ s = __dm_stats_find(stats, id);
+ if (!s) {
+ mutex_unlock(&stats->mutex);
+ return -ENOENT;
+ }
+
+ __dm_stat_clear(s, 0, s->n_entries, true);
+
+ mutex_unlock(&stats->mutex);
+
+ return 1;
+}
+
+/*
+ * This is like jiffies_to_msec, but works for 64-bit values.
+ */
+static unsigned long long jiffies_to_msec64(unsigned long long j)
+{
+ unsigned long long result = 0;
+ unsigned mult;
+
+ if (j)
+ result = jiffies_to_msecs(j & 0x3fffff);
+ if (j >= 1 << 22) {
+ mult = jiffies_to_msecs(1 << 22);
+ result += (unsigned long long)mult * (unsigned long long)jiffies_to_msecs((j >> 22) & 0x3fffff);
+ }
+ if (j >= 1ULL << 44)
+ result += (unsigned long long)mult * (unsigned long long)mult * (unsigned long long)jiffies_to_msecs(j >> 44);
+
+ return result;
+}
+
+static int dm_stats_print(struct dm_stats *stats, int id,
+ size_t idx_start, size_t idx_len,
+ bool clear, char *result, unsigned maxlen)
+{
+ unsigned sz = 0;
+ struct dm_stat *s;
+ size_t x;
+ sector_t start, end;
+ size_t idx_end;
+ struct dm_stat_shared *shared;
+
+ mutex_lock(&stats->mutex);
+
+ s = __dm_stats_find(stats, id);
+ if (!s) {
+ mutex_unlock(&stats->mutex);
+ return -ENOENT;
+ }
+
+ idx_end = idx_start + idx_len;
+ if (idx_end < idx_start ||
+ idx_end > s->n_entries)
+ idx_end = s->n_entries;
+
+ if (idx_start > idx_end)
+ idx_start = idx_end;
+
+ start = s->start + s->step * idx_start;
+
+ for (x = idx_start; x < idx_end; x++, start = end) {
+ shared = &s->stat_shared[x];
+ end = start + s->step;
+ if (unlikely(end > s->end))
+ end = s->end;
+
+ __dm_stat_init_temporary_percpu_totals(shared, s, x);
+
+ DMEMIT("%llu-%llu %llu %llu %llu %llu %llu %llu %llu %llu %d %llu %llu %llu %llu\n",
+ (unsigned long long)start,
+ (unsigned long long)end,
+ shared->tmp.ios[READ],
+ shared->tmp.merges[READ],
+ shared->tmp.sectors[READ],
+ jiffies_to_msec64(shared->tmp.ticks[READ]),
+ shared->tmp.ios[WRITE],
+ shared->tmp.merges[WRITE],
+ shared->tmp.sectors[WRITE],
+ jiffies_to_msec64(shared->tmp.ticks[WRITE]),
+ dm_stat_in_flight(shared),
+ jiffies_to_msec64(shared->tmp.io_ticks_total),
+ jiffies_to_msec64(shared->tmp.time_in_queue),
+ jiffies_to_msec64(shared->tmp.io_ticks[READ]),
+ jiffies_to_msec64(shared->tmp.io_ticks[WRITE]));
+
+ if (unlikely(sz + 1 >= maxlen))
+ goto buffer_overflow;
+ }
+
+ if (clear)
+ __dm_stat_clear(s, idx_start, idx_end, false);
+
+buffer_overflow:
+ mutex_unlock(&stats->mutex);
+
+ return 1;
+}
+
+static int dm_stats_set_aux(struct dm_stats *stats, int id, const char *aux)
+{
+ struct dm_stat *s;
+
+ mutex_lock(&stats->mutex);
+
+ s = __dm_stats_find(stats, id);
+ if (!s) {
+ mutex_unlock(&stats->mutex);
+ return -ENOENT;
+ }
+
+ aux = kstrdup(aux, GFP_KERNEL);
+ if (!aux) {
+ mutex_unlock(&stats->mutex);
+ return -ENOMEM;
+ }
+
+ kfree(s->aux);
+ s->aux = aux;
+
+ mutex_unlock(&stats->mutex);
+
+ return 0;
+}
+
+static int message_stats_create(struct mapped_device *md,
+ unsigned argc, char **argv,
+ char *result, unsigned maxlen)
+{
+ int id;
+ char dummy;
+ unsigned long long start, end, step;
+ unsigned div;
+ const char *program, *aux;
+
+ if (argc < 3 || argc > 5)
+ return -EINVAL;
+
+ if (!strcmp(argv[1], "-")) {
+ start = 0;
+ end = dm_get_size(md);
+ if (!end)
+ end = 1;
+ } else if (sscanf(argv[1], "%llu-%llu%c", &start, &end, &dummy) != 2 ||
+ start != (sector_t)start || end != (sector_t)end)
+ return -EINVAL;
+
+ if (start >= end)
+ return -EINVAL;
+
+ if (sscanf(argv[2], "/%u%c", &div, &dummy) == 1) {
+ step = end - start;
+ if (do_div(step, div))
+ step++;
+ if (!step)
+ step = 1;
+ } else if (sscanf(argv[2], "%llu%c", &step, &dummy) != 1 ||
+ step != (sector_t)step || !step)
+ return -EINVAL;
+
+ program = "-";
+ aux = "-";
+
+ if (argc > 3)
+ program = argv[3];
+
+ if (argc > 4)
+ aux = argv[4];
+
+ /*
+ * If a buffer overflow happens after we created the region,
+ * it's too late (the userspace would retry with a larger
+ * buffer, but the region id that caused the overflow is already
+ * leaked). So we must detect buffer overflow in advance.
+ */
+ snprintf(result, maxlen, "%d", INT_MAX);
+ if (dm_message_test_buffer_overflow(result, maxlen))
+ return 1;
+
+ id = dm_stats_create(dm_get_stats(md), start, end, step, program, aux,
+ dm_internal_suspend, dm_internal_resume, md);
+ if (id < 0)
+ return id;
+
+ snprintf(result, maxlen, "%d", id);
+
+ return 1;
+}
+
+static int message_stats_delete(struct mapped_device *md,
+ unsigned argc, char **argv)
+{
+ int id;
+ char dummy;
+
+ if (argc != 2)
+ return -EINVAL;
+
+ if (sscanf(argv[1], "%d%c", &id, &dummy) != 1 || id < 0)
+ return -EINVAL;
+
+ return dm_stats_delete(dm_get_stats(md), id);
+}
+
+static int message_stats_clear(struct mapped_device *md,
+ unsigned argc, char **argv)
+{
+ int id;
+ char dummy;
+
+ if (argc != 2)
+ return -EINVAL;
+
+ if (sscanf(argv[1], "%d%c", &id, &dummy) != 1 || id < 0)
+ return -EINVAL;
+
+ return dm_stats_clear(dm_get_stats(md), id);
+}
+
+static int message_stats_list(struct mapped_device *md,
+ unsigned argc, char **argv,
+ char *result, unsigned maxlen)
+{
+ int r;
+ const char *program = NULL;
+
+ if (argc < 1 || argc > 2)
+ return -EINVAL;
+
+ if (argc > 1) {
+ program = kstrdup(argv[1], GFP_KERNEL);
+ if (!program)
+ return -ENOMEM;
+ }
+
+ r = dm_stats_list(dm_get_stats(md), program, result, maxlen);
+
+ kfree(program);
+
+ return r;
+}
+
+static int message_stats_print(struct mapped_device *md,
+ unsigned argc, char **argv, bool clear,
+ char *result, unsigned maxlen)
+{
+ int id;
+ char dummy;
+ unsigned long idx_start = 0, idx_len = -1;
+
+ if (argc != 2 && argc != 4)
+ return -EINVAL;
+
+ if (sscanf(argv[1], "%d%c", &id, &dummy) != 1 || id < 0)
+ return -EINVAL;
+
+ if (argc > 3) {
+ if (strcmp(argv[2], "-") &&
+ sscanf(argv[2], "%lu%c", &idx_start, &dummy) != 1)
+ return -EINVAL;
+ if (strcmp(argv[3], "-") &&
+ sscanf(argv[3], "%lu%c", &idx_len, &dummy) != 1)
+ return -EINVAL;
+ }
+
+ return dm_stats_print(dm_get_stats(md), id, idx_start, idx_len, clear,
+ result, maxlen);
+}
+
+static int message_stats_set_aux(struct mapped_device *md,
+ unsigned argc, char **argv)
+{
+ int id;
+ char dummy;
+
+ if (argc != 3)
+ return -EINVAL;
+
+ if (sscanf(argv[1], "%d%c", &id, &dummy) != 1 || id < 0)
+ return -EINVAL;
+
+ return dm_stats_set_aux(dm_get_stats(md), id, argv[2]);
+}
+
+int dm_stats_message(struct mapped_device *md, unsigned argc, char **argv,
+ char *result, unsigned maxlen)
+{
+ int r;
+
+ if (dm_request_based(md)) {
+ DMWARN("Statistics are only supported for bio-based devices");
+ return -EOPNOTSUPP;
+ }
+
+ if (!strcasecmp(argv[0], "@stats_create"))
+ r = message_stats_create(md, argc, argv, result, maxlen);
+ else if (!strcasecmp(argv[0], "@stats_delete"))
+ r = message_stats_delete(md, argc, argv);
+ else if (!strcasecmp(argv[0], "@stats_clear"))
+ r = message_stats_clear(md, argc, argv);
+ else if (!strcasecmp(argv[0], "@stats_list"))
+ r = message_stats_list(md, argc, argv, result, maxlen);
+ else if (!strcasecmp(argv[0], "@stats_print"))
+ r = message_stats_print(md, argc, argv, false, result, maxlen);
+ else if (!strcasecmp(argv[0], "@stats_print_clear"))
+ r = message_stats_print(md, argc, argv, true, result, maxlen);
+ else if (!strcasecmp(argv[0], "@stats_set_aux"))
+ r = message_stats_set_aux(md, argc, argv);
+ else
+ return 2; /* this wasn't a stats message */
+
+ if (r == -EINVAL)
+ DMWARN("Invalid parameters for message %s", argv[0]);
+
+ return r;
+}
+
+int __init dm_statistics_init(void)
+{
+ dm_stat_need_rcu_barrier = 0;
+ return 0;
+}
+
+void dm_statistics_exit(void)
+{
+ if (dm_stat_need_rcu_barrier)
+ rcu_barrier();
+ if (WARN_ON(shared_memory_amount))
+ DMCRIT("shared_memory_amount leaked: %lu", shared_memory_amount);
+}
+
+module_param_named(stats_current_allocated_bytes, shared_memory_amount, ulong, S_IRUGO);
+MODULE_PARM_DESC(stats_current_allocated_bytes, "Memory currently used by statistics");
diff --git a/drivers/md/dm-stats.h b/drivers/md/dm-stats.h
new file mode 100644
index 0000000..e7c4984
--- /dev/null
+++ b/drivers/md/dm-stats.h
@@ -0,0 +1,40 @@
+#ifndef DM_STATS_H
+#define DM_STATS_H
+
+#include <linux/types.h>
+#include <linux/mutex.h>
+#include <linux/list.h>
+
+int dm_statistics_init(void);
+void dm_statistics_exit(void);
+
+struct dm_stats {
+ struct mutex mutex;
+ struct list_head list; /* list of struct dm_stat */
+ struct dm_stats_last_position __percpu *last;
+ sector_t last_sector;
+ unsigned last_rw;
+};
+
+struct dm_stats_aux {
+ bool merged;
+};
+
+void dm_stats_init(struct dm_stats *st);
+void dm_stats_cleanup(struct dm_stats *st);
+
+struct mapped_device;
+
+int dm_stats_message(struct mapped_device *md, unsigned argc, char **argv,
+ char *result, unsigned maxlen);
+
+void dm_stats_account_io(struct dm_stats *stats, unsigned long bi_rw,
+ sector_t bi_sector, unsigned bi_sectors, bool end,
+ unsigned long duration, struct dm_stats_aux *aux);
+
+static inline bool dm_stats_used(struct dm_stats *st)
+{
+ return !list_empty(&st->list);
+}
+
+#endif
diff --git a/drivers/md/dm.c b/drivers/md/dm.c
index 9e39d2b..2044f91 100644
--- a/drivers/md/dm.c
+++ b/drivers/md/dm.c
@@ -60,6 +60,7 @@ struct dm_io {
struct bio *bio;
unsigned long start_time;
spinlock_t endio_lock;
+ struct dm_stats_aux stats_aux;
};
/*
@@ -198,6 +199,8 @@ struct mapped_device {
/* zero-length flush that will be cloned and submitted to targets */
struct bio flush_bio;
+
+ struct dm_stats stats;
};
/*
@@ -269,6 +272,7 @@ static int (*_inits[])(void) __initdata = {
dm_io_init,
dm_kcopyd_init,
dm_interface_init,
+ dm_statistics_init,
};
static void (*_exits[])(void) = {
@@ -279,6 +283,7 @@ static void (*_exits[])(void) = {
dm_io_exit,
dm_kcopyd_exit,
dm_interface_exit,
+ dm_statistics_exit,
};
static int __init dm_init(void)
@@ -384,6 +389,16 @@ int dm_lock_for_deletion(struct mapped_device *md)
return r;
}
+sector_t dm_get_size(struct mapped_device *md)
+{
+ return get_capacity(md->disk);
+}
+
+struct dm_stats *dm_get_stats(struct mapped_device *md)
+{
+ return &md->stats;
+}
+
static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
{
struct mapped_device *md = bdev->bd_disk->private_data;
@@ -466,8 +481,9 @@ static int md_in_flight(struct mapped_device *md)
static void start_io_acct(struct dm_io *io)
{
struct mapped_device *md = io->md;
+ struct bio *bio = io->bio;
int cpu;
- int rw = bio_data_dir(io->bio);
+ int rw = bio_data_dir(bio);
io->start_time = jiffies;
@@ -476,6 +492,10 @@ static void start_io_acct(struct dm_io *io)
part_stat_unlock();
atomic_set(&dm_disk(md)->part0.in_flight[rw],
atomic_inc_return(&md->pending[rw]));
+
+ if (unlikely(dm_stats_used(&md->stats)))
+ dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_sector,
+ bio_sectors(bio), false, 0, &io->stats_aux);
}
static void end_io_acct(struct dm_io *io)
@@ -491,6 +511,10 @@ static void end_io_acct(struct dm_io *io)
part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
part_stat_unlock();
+ if (unlikely(dm_stats_used(&md->stats)))
+ dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_sector,
+ bio_sectors(bio), true, duration, &io->stats_aux);
+
/*
* After this is decremented the bio must not be touched if it is
* a flush.
@@ -1519,7 +1543,7 @@ static void _dm_request(struct request_queue *q, struct bio *bio)
return;
}
-static int dm_request_based(struct mapped_device *md)
+int dm_request_based(struct mapped_device *md)
{
return blk_queue_stackable(md->queue);
}
@@ -1959,6 +1983,8 @@ static struct mapped_device *alloc_dev(int minor)
md->flush_bio.bi_bdev = md->bdev;
md->flush_bio.bi_rw = WRITE_FLUSH;
+ dm_stats_init(&md->stats);
+
/* Populate the mapping, nobody knows we exist yet */
spin_lock(&_minor_lock);
old_md = idr_replace(&_minor_idr, md, minor);
@@ -2010,6 +2036,7 @@ static void free_dev(struct mapped_device *md)
put_disk(md->disk);
blk_cleanup_queue(md->queue);
+ dm_stats_cleanup(&md->stats);
module_put(THIS_MODULE);
kfree(md);
}
@@ -2695,6 +2722,38 @@ out:
return r;
}
+/*
+ * Internal suspend/resume works like userspace-driven suspend. It waits
+ * until all bios finish and prevents issuing new bios to the target drivers.
+ * It may be used only from the kernel.
+ *
+ * Internal suspend holds md->suspend_lock, which prevents interaction with
+ * userspace-driven suspend.
+ */
+
+void dm_internal_suspend(struct mapped_device *md)
+{
+ mutex_lock(&md->suspend_lock);
+ if (dm_suspended_md(md))
+ return;
+
+ set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
+ synchronize_srcu(&md->io_barrier);
+ flush_workqueue(md->wq);
+ dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
+}
+
+void dm_internal_resume(struct mapped_device *md)
+{
+ if (dm_suspended_md(md))
+ goto done;
+
+ dm_queue_flush(md);
+
+done:
+ mutex_unlock(&md->suspend_lock);
+}
+
/*-----------------------------------------------------------------
* Event notification.
*---------------------------------------------------------------*/
diff --git a/drivers/md/dm.h b/drivers/md/dm.h
index 45b97da..9236e7c 100644
--- a/drivers/md/dm.h
+++ b/drivers/md/dm.h
@@ -16,6 +16,8 @@
#include <linux/blkdev.h>
#include <linux/hdreg.h>
+#include "dm-stats.h"
+
/*
* Suspend feature flags
*/
@@ -146,10 +148,16 @@ void dm_destroy(struct mapped_device *md);
void dm_destroy_immediate(struct mapped_device *md);
int dm_open_count(struct mapped_device *md);
int dm_lock_for_deletion(struct mapped_device *md);
+int dm_request_based(struct mapped_device *md);
+sector_t dm_get_size(struct mapped_device *md);
+struct dm_stats *dm_get_stats(struct mapped_device *md);
int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
unsigned cookie);
+void dm_internal_suspend(struct mapped_device *md);
+void dm_internal_resume(struct mapped_device *md);
+
int dm_io_init(void);
void dm_io_exit(void);
@@ -162,4 +170,13 @@ void dm_kcopyd_exit(void);
struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size);
void dm_free_md_mempools(struct dm_md_mempools *pools);
+
+/*
+ * Helpers that are used by DM core
+ */
+static inline bool dm_message_test_buffer_overflow(char *result, unsigned maxlen)
+{
+ return !maxlen || strlen(result) + 1 >= maxlen;
+}
+
#endif
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel
