]> git.ipfire.org Git - thirdparty/systemd.git/commitdiff
core: add io controller support on the unified hierarchy 3193/head
authorTejun Heo <htejun@fb.com>
Thu, 5 May 2016 20:42:55 +0000 (16:42 -0400)
committerTejun Heo <tj@kernel.org>
Thu, 5 May 2016 20:43:06 +0000 (16:43 -0400)
On the unified hierarchy, blkio controller is renamed to io and the interface
is changed significantly.

* blkio.weight and blkio.weight_device are consolidated into io.weight which
  uses the standardized weight range [1, 10000] with 100 as the default value.

* blkio.throttle.{read|write}_{bps|iops}_device are consolidated into io.max.
  Expansion of throttling features is being worked on to support
  work-conserving absolute limits (io.low and io.high).

* All stats are consolidated into io.stats.

This patchset adds support for the new interface.  As the interface has been
revamped and new features are expected to be added, it seems best to treat it
as a separate controller rather than trying to expand the blkio settings
although we might add automatic translation if only blkio settings are
specified.

* io.weight handling is mostly identical to blkio.weight[_device] handling
  except that the weight range is different.

* Both read and write bandwidth settings are consolidated into
  CGroupIODeviceLimit which describes all limits applicable to the device.
  This makes it less painful to add new limits.

* "max" can be used to specify the maximum limit which is equivalent to no
  config for max limits and treated as such.  If a given CGroupIODeviceLimit
  doesn't contain any non-default configs, the config struct is discarded once
  the no limit config is applied to cgroup.

* lookup_blkio_device() is renamed to lookup_block_device().

Signed-off-by: Tejun Heo <htejun@fb.com>
16 files changed:
man/systemd.resource-control.xml
src/basic/cgroup-util.c
src/basic/cgroup-util.h
src/cgtop/cgtop.c
src/core/cgroup.c
src/core/cgroup.h
src/core/dbus-cgroup.c
src/core/load-fragment-gperf.gperf.m4
src/core/load-fragment.c
src/core/load-fragment.h
src/core/main.c
src/core/manager.h
src/core/system.conf
src/core/unit.c
src/shared/bus-unit-util.c
src/systemctl/systemctl.c

index fd6f7a1b69b2660ed7260e4011c0bdb58f905b93..4edb1a25a8867205af9fb9dda72d027a3e2e1247 100644 (file)
         </listitem>
       </varlistentry>
 
+      <varlistentry>
+        <term><varname>IOAccounting=</varname></term>
+
+        <listitem>
+          <para>Turn on Block I/O accounting for this unit on unified
+          hierarchy. Takes a boolean argument. Note that turning on
+          block I/O accounting for one unit will also implicitly turn
+          it on for all units contained in the same slice and all for
+          its parent slices and the units contained therein. The
+          system default for this setting may be controlled with
+          <varname>DefaultIOAccounting=</varname> in
+          <citerefentry><refentrytitle>systemd-system.conf</refentrytitle><manvolnum>5</manvolnum></citerefentry>.</para>
+        </listitem>
+      </varlistentry>
+
+      <varlistentry>
+        <term><varname>IOWeight=<replaceable>weight</replaceable></varname></term>
+        <term><varname>StartupIOWeight=<replaceable>weight</replaceable></varname></term>
+
+        <listitem>
+          <para>Set the default overall block I/O weight for the
+          executed processes on unified hierarchy. Takes a single
+          weight value (between 1 and 10000) to set the default block
+          I/O weight. This controls the <literal>io.weight</literal>
+          control group attribute, which defaults to 100. For details
+          about this control group attribute, see <ulink
+          url="https://www.kernel.org/doc/Documentation/cgroup-v2.txt">cgroup-v2.txt</ulink>.
+          The available I/O bandwidth is split up among all units
+          within one slice relative to their block I/O weight.</para>
+
+          <para>While <varname>StartupIOWeight=</varname> only applies
+          to the startup phase of the system,
+          <varname>IOWeight=</varname> applies to the later runtime of
+          the system, and if the former is not set also to the startup
+          phase. This allows prioritizing specific services at boot-up
+          differently than during runtime.</para>
+
+          <para>Implies <literal>IOAccounting=true</literal>.</para>
+        </listitem>
+      </varlistentry>
+
+      <varlistentry>
+        <term><varname>IODeviceWeight=<replaceable>device</replaceable> <replaceable>weight</replaceable></varname></term>
+
+        <listitem>
+          <para>Set the per-device overall block I/O weight for the
+          executed processes on unified hierarchy. Takes a
+          space-separated pair of a file path and a weight value to
+          specify the device specific weight value, between 1 and
+          10000. (Example: "/dev/sda 1000"). The file path may be
+          specified as path to a block device node or as any other
+          file, in which case the backing block device of the file
+          system of the file is determined. This controls the
+          <literal>io.weight</literal> control group attribute, which
+          defaults to 100. Use this option multiple times to set
+          weights for multiple devices. For details about this control
+          group attribute, see <ulink
+          url="https://www.kernel.org/doc/Documentation/cgroup-v2.txt">cgroup-v2.txt</ulink>.</para>
+
+          <para>Implies <literal>IOAccounting=true</literal>.</para>
+        </listitem>
+      </varlistentry>
+
+      <varlistentry>
+        <term><varname>IOReadBandwidthMax=<replaceable>device</replaceable> <replaceable>bytes</replaceable></varname></term>
+        <term><varname>IOWriteBandwidthMax=<replaceable>device</replaceable> <replaceable>bytes</replaceable></varname></term>
+
+        <listitem>
+          <para>Set the per-device overall block I/O bandwidth maximum
+          limit for the executed processes on unified hierarchy. This
+          limit is not work-conserving and the executed processes are
+          not allowed to use more even if the device has idle
+          capacity.  Takes a space-separated pair of a file path and a
+          bandwidth value (in bytes per second) to specify the device
+          specific bandwidth. The file path may be a path to a block
+          device node, or as any other file in which case the backing
+          block device of the file system of the file is used. If the
+          bandwidth is suffixed with K, M, G, or T, the specified
+          bandwidth is parsed as Kilobytes, Megabytes, Gigabytes, or
+          Terabytes, respectively, to the base of 1000. (Example:
+          "/dev/disk/by-path/pci-0000:00:1f.2-scsi-0:0:0:0 5M"). This
+          controls the <literal>io.max</literal> control group
+          attributes. Use this option multiple times to set bandwidth
+          limits for multiple devices. For details about this control
+          group attribute, see <ulink
+          url="https://www.kernel.org/doc/Documentation/cgroup-v2.txt">cgroup-v2.txt</ulink>.
+          </para>
+
+          <para>Implies <literal>IOAccounting=true</literal>.</para>
+        </listitem>
+      </varlistentry>
+
       <varlistentry>
         <term><varname>BlockIOAccounting=</varname></term>
 
         <listitem>
+          <para>Use IOAccounting on unified hierarchy.</para>
+
           <para>Turn on Block I/O accounting for this unit. Takes a
           boolean argument. Note that turning on block I/O accounting
           for one unit will also implicitly turn it on for all units
         <term><varname>BlockIOWeight=<replaceable>weight</replaceable></varname></term>
         <term><varname>StartupBlockIOWeight=<replaceable>weight</replaceable></varname></term>
 
-        <listitem><para>Set the default overall block I/O weight for
-        the executed processes. Takes a single weight value (between
-        10 and 1000) to set the default block I/O weight. This controls
+        <listitem><para>Use IOWeight and StartupIOWeight on unified
+        hierarchy.</para>
+
+        <para>Set the default overall block I/O weight for the
+        executed processes. Takes a single weight value (between 10
+        and 1000) to set the default block I/O weight. This controls
         the <literal>blkio.weight</literal> control group attribute,
         which defaults to 500. For details about this control group
         attribute, see <ulink
         <term><varname>BlockIODeviceWeight=<replaceable>device</replaceable> <replaceable>weight</replaceable></varname></term>
 
         <listitem>
+          <para>Use IODeviceWeight on unified hierarchy.</para>
+
           <para>Set the per-device overall block I/O weight for the
           executed processes. Takes a space-separated pair of a file
           path and a weight value to specify the device specific
         <term><varname>BlockIOWriteBandwidth=<replaceable>device</replaceable> <replaceable>bytes</replaceable></varname></term>
 
         <listitem>
+          <para>Use IOReadBandwidthMax and IOWriteBandwidthMax on
+          unified hierarchy.</para>
+
           <para>Set the per-device overall block I/O bandwidth limit
           for the executed processes. Takes a space-separated pair of
           a file path and a bandwidth value (in bytes per second) to
index 504318074750d824d7a9d47beac3813e7fbb3d38..ff57cf30b72ad649e101ba075065712938326789 100644 (file)
@@ -2060,10 +2060,10 @@ int cg_mask_supported(CGroupMask *ret) {
                         mask |= CGROUP_CONTROLLER_TO_MASK(v);
                 }
 
-                /* Currently, we only support the memory and pids
+                /* Currently, we only support the memory, io and pids
                  * controller in the unified hierarchy, mask
                  * everything else off. */
-                mask &= CGROUP_MASK_MEMORY | CGROUP_MASK_PIDS;
+                mask &= CGROUP_MASK_MEMORY | CGROUP_MASK_IO | CGROUP_MASK_PIDS;
 
         } else {
                 CGroupController c;
@@ -2249,6 +2249,26 @@ bool cg_is_legacy_wanted(void) {
         return !cg_is_unified_wanted();
 }
 
+int cg_weight_parse(const char *s, uint64_t *ret) {
+        uint64_t u;
+        int r;
+
+        if (isempty(s)) {
+                *ret = CGROUP_WEIGHT_INVALID;
+                return 0;
+        }
+
+        r = safe_atou64(s, &u);
+        if (r < 0)
+                return r;
+
+        if (u < CGROUP_WEIGHT_MIN || u > CGROUP_WEIGHT_MAX)
+                return -ERANGE;
+
+        *ret = u;
+        return 0;
+}
+
 int cg_cpu_shares_parse(const char *s, uint64_t *ret) {
         uint64_t u;
         int r;
@@ -2292,6 +2312,7 @@ int cg_blkio_weight_parse(const char *s, uint64_t *ret) {
 static const char *cgroup_controller_table[_CGROUP_CONTROLLER_MAX] = {
         [CGROUP_CONTROLLER_CPU] = "cpu",
         [CGROUP_CONTROLLER_CPUACCT] = "cpuacct",
+        [CGROUP_CONTROLLER_IO] = "io",
         [CGROUP_CONTROLLER_BLKIO] = "blkio",
         [CGROUP_CONTROLLER_MEMORY] = "memory",
         [CGROUP_CONTROLLER_DEVICES] = "devices",
index 4254e51e5d9bdbcec2d4bcf120f386de404d832e..a696c1fa60c146c5a157d39796c6e4ccfb098251 100644 (file)
@@ -34,6 +34,7 @@
 typedef enum CGroupController {
         CGROUP_CONTROLLER_CPU,
         CGROUP_CONTROLLER_CPUACCT,
+        CGROUP_CONTROLLER_IO,
         CGROUP_CONTROLLER_BLKIO,
         CGROUP_CONTROLLER_MEMORY,
         CGROUP_CONTROLLER_DEVICES,
@@ -48,6 +49,7 @@ typedef enum CGroupController {
 typedef enum CGroupMask {
         CGROUP_MASK_CPU = CGROUP_CONTROLLER_TO_MASK(CGROUP_CONTROLLER_CPU),
         CGROUP_MASK_CPUACCT = CGROUP_CONTROLLER_TO_MASK(CGROUP_CONTROLLER_CPUACCT),
+        CGROUP_MASK_IO = CGROUP_CONTROLLER_TO_MASK(CGROUP_CONTROLLER_IO),
         CGROUP_MASK_BLKIO = CGROUP_CONTROLLER_TO_MASK(CGROUP_CONTROLLER_BLKIO),
         CGROUP_MASK_MEMORY = CGROUP_CONTROLLER_TO_MASK(CGROUP_CONTROLLER_MEMORY),
         CGROUP_MASK_DEVICES = CGROUP_CONTROLLER_TO_MASK(CGROUP_CONTROLLER_DEVICES),
@@ -55,6 +57,21 @@ typedef enum CGroupMask {
         _CGROUP_MASK_ALL = CGROUP_CONTROLLER_TO_MASK(_CGROUP_CONTROLLER_MAX) - 1
 } CGroupMask;
 
+/* Special values for all weight knobs on unified hierarchy */
+#define CGROUP_WEIGHT_INVALID ((uint64_t) -1)
+#define CGROUP_WEIGHT_MIN UINT64_C(1)
+#define CGROUP_WEIGHT_MAX UINT64_C(10000)
+#define CGROUP_WEIGHT_DEFAULT UINT64_C(100)
+
+#define CGROUP_LIMIT_MIN UINT64_C(0)
+#define CGROUP_LIMIT_MAX ((uint64_t) -1)
+
+static inline bool CGROUP_WEIGHT_IS_OK(uint64_t x) {
+        return
+            x == CGROUP_WEIGHT_INVALID ||
+            (x >= CGROUP_WEIGHT_MIN && x <= CGROUP_WEIGHT_MAX);
+}
+
 /* Special values for the cpu.shares attribute */
 #define CGROUP_CPU_SHARES_INVALID ((uint64_t) -1)
 #define CGROUP_CPU_SHARES_MIN UINT64_C(2)
@@ -190,5 +207,6 @@ bool cg_is_legacy_wanted(void);
 const char* cgroup_controller_to_string(CGroupController c) _const_;
 CGroupController cgroup_controller_from_string(const char *s) _pure_;
 
+int cg_weight_parse(const char *s, uint64_t *ret);
 int cg_cpu_shares_parse(const char *s, uint64_t *ret);
 int cg_blkio_weight_parse(const char *s, uint64_t *ret);
index 14eb46c8dbd459e83cfb97af806545e1f88184af..e088e4b197de4b91629da980b01a444483f5f272 100644 (file)
@@ -269,13 +269,15 @@ static int process(
                 if (g->memory > 0)
                         g->memory_valid = true;
 
-        } else if (streq(controller, "blkio") && cg_unified() <= 0) {
+        } else if ((streq(controller, "io") && cg_unified() > 0) ||
+                   (streq(controller, "blkio") && cg_unified() <= 0)) {
                 _cleanup_fclose_ FILE *f = NULL;
                 _cleanup_free_ char *p = NULL;
+                bool unified = cg_unified() > 0;
                 uint64_t wr = 0, rd = 0;
                 nsec_t timestamp;
 
-                r = cg_get_path(controller, path, "blkio.io_service_bytes", &p);
+                r = cg_get_path(controller, path, unified ? "io.stat" : "blkio.io_service_bytes", &p);
                 if (r < 0)
                         return r;
 
@@ -293,25 +295,38 @@ static int process(
                         if (!fgets(line, sizeof(line), f))
                                 break;
 
+                        /* Trim and skip the device */
                         l = strstrip(line);
                         l += strcspn(l, WHITESPACE);
                         l += strspn(l, WHITESPACE);
 
-                        if (first_word(l, "Read")) {
-                                l += 4;
-                                q = &rd;
-                        } else if (first_word(l, "Write")) {
-                                l += 5;
-                                q = &wr;
-                        } else
-                                continue;
-
-                        l += strspn(l, WHITESPACE);
-                        r = safe_atou64(l, &k);
-                        if (r < 0)
-                                continue;
+                        if (unified) {
+                                while (!isempty(l)) {
+                                        if (sscanf(l, "rbytes=%" SCNu64, &k))
+                                                rd += k;
+                                        else if (sscanf(l, "wbytes=%" SCNu64, &k))
+                                                wr += k;
 
-                        *q += k;
+                                        l += strcspn(l, WHITESPACE);
+                                        l += strspn(l, WHITESPACE);
+                                }
+                        } else {
+                                if (first_word(l, "Read")) {
+                                        l += 4;
+                                        q = &rd;
+                                } else if (first_word(l, "Write")) {
+                                        l += 5;
+                                        q = &wr;
+                                } else
+                                        continue;
+
+                                l += strspn(l, WHITESPACE);
+                                r = safe_atou64(l, &k);
+                                if (r < 0)
+                                        continue;
+
+                                *q += k;
+                        }
                 }
 
                 timestamp = now_nsec(CLOCK_MONOTONIC);
@@ -437,6 +452,9 @@ static int refresh(const char *root, Hashmap *a, Hashmap *b, unsigned iteration)
         if (r < 0)
                 return r;
         r = refresh_one("memory", root, a, b, iteration, 0, NULL);
+        if (r < 0)
+                return r;
+        r = refresh_one("io", root, a, b, iteration, 0, NULL);
         if (r < 0)
                 return r;
         r = refresh_one("blkio", root, a, b, iteration, 0, NULL);
index 25cc6962f9be52451cc1ca59309ea46e1978e114..44106e52ea9a25393318ad96fc382b0a9df793ae 100644 (file)
@@ -32,6 +32,7 @@
 #include "special.h"
 #include "string-table.h"
 #include "string-util.h"
+#include "stdio-util.h"
 
 #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
 
@@ -47,6 +48,9 @@ void cgroup_context_init(CGroupContext *c) {
 
         c->memory_limit = (uint64_t) -1;
 
+        c->io_weight = CGROUP_WEIGHT_INVALID;
+        c->startup_io_weight = CGROUP_WEIGHT_INVALID;
+
         c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
         c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
 
@@ -62,6 +66,24 @@ void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) {
         free(a);
 }
 
+void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) {
+        assert(c);
+        assert(w);
+
+        LIST_REMOVE(device_weights, c->io_device_weights, w);
+        free(w->path);
+        free(w);
+}
+
+void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) {
+        assert(c);
+        assert(l);
+
+        LIST_REMOVE(device_limits, c->io_device_limits, l);
+        free(l->path);
+        free(l);
+}
+
 void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) {
         assert(c);
         assert(w);
@@ -83,6 +105,12 @@ void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockI
 void cgroup_context_done(CGroupContext *c) {
         assert(c);
 
+        while (c->io_device_weights)
+                cgroup_context_free_io_device_weight(c, c->io_device_weights);
+
+        while (c->io_device_limits)
+                cgroup_context_free_io_device_limit(c, c->io_device_limits);
+
         while (c->blockio_device_weights)
                 cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights);
 
@@ -94,6 +122,8 @@ void cgroup_context_done(CGroupContext *c) {
 }
 
 void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
+        CGroupIODeviceLimit *il;
+        CGroupIODeviceWeight *iw;
         CGroupBlockIODeviceBandwidth *b;
         CGroupBlockIODeviceWeight *w;
         CGroupDeviceAllow *a;
@@ -106,12 +136,15 @@ void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
 
         fprintf(f,
                 "%sCPUAccounting=%s\n"
+                "%sIOAccounting=%s\n"
                 "%sBlockIOAccounting=%s\n"
                 "%sMemoryAccounting=%s\n"
                 "%sTasksAccounting=%s\n"
                 "%sCPUShares=%" PRIu64 "\n"
                 "%sStartupCPUShares=%" PRIu64 "\n"
                 "%sCPUQuotaPerSecSec=%s\n"
+                "%sIOWeight=%" PRIu64 "\n"
+                "%sStartupIOWeight=%" PRIu64 "\n"
                 "%sBlockIOWeight=%" PRIu64 "\n"
                 "%sStartupBlockIOWeight=%" PRIu64 "\n"
                 "%sMemoryLimit=%" PRIu64 "\n"
@@ -119,12 +152,15 @@ void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
                 "%sDevicePolicy=%s\n"
                 "%sDelegate=%s\n",
                 prefix, yes_no(c->cpu_accounting),
+                prefix, yes_no(c->io_accounting),
                 prefix, yes_no(c->blockio_accounting),
                 prefix, yes_no(c->memory_accounting),
                 prefix, yes_no(c->tasks_accounting),
                 prefix, c->cpu_shares,
                 prefix, c->startup_cpu_shares,
                 prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1),
+                prefix, c->io_weight,
+                prefix, c->startup_io_weight,
                 prefix, c->blockio_weight,
                 prefix, c->startup_blockio_weight,
                 prefix, c->memory_limit,
@@ -139,6 +175,31 @@ void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
                         a->path,
                         a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : "");
 
+        LIST_FOREACH(device_weights, iw, c->io_device_weights)
+                fprintf(f,
+                        "%sIODeviceWeight=%s %" PRIu64,
+                        prefix,
+                        iw->path,
+                        iw->weight);
+
+        LIST_FOREACH(device_limits, il, c->io_device_limits) {
+                char buf[FORMAT_BYTES_MAX];
+
+                if (il->rbps_max != CGROUP_LIMIT_MAX)
+                        fprintf(f,
+                                "%sIOReadBandwidthMax=%s %s\n",
+                                prefix,
+                                il->path,
+                                format_bytes(buf, sizeof(buf), il->rbps_max));
+
+                if (il->wbps_max != CGROUP_LIMIT_MAX)
+                        fprintf(f,
+                                "%sIOWriteBandwidthMax=%s %s\n",
+                                prefix,
+                                il->path,
+                                format_bytes(buf, sizeof(buf), il->wbps_max));
+        }
+
         LIST_FOREACH(device_weights, w, c->blockio_device_weights)
                 fprintf(f,
                         "%sBlockIODeviceWeight=%s %" PRIu64,
@@ -158,7 +219,7 @@ void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
         }
 }
 
-static int lookup_blkio_device(const char *p, dev_t *dev) {
+static int lookup_block_device(const char *p, dev_t *dev) {
         struct stat st;
         int r;
 
@@ -343,6 +404,77 @@ void cgroup_context_apply(CGroupContext *c, CGroupMask mask, const char *path, M
                                        "Failed to set cpu.cfs_quota_us on %s: %m", path);
         }
 
+        if (mask & CGROUP_MASK_IO) {
+                CGroupIODeviceWeight *w;
+                CGroupIODeviceLimit *l, *next;
+
+                if (!is_root) {
+                        char buf[MAX(8+DECIMAL_STR_MAX(uint64_t)+1,
+                                     DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1)];
+                        uint64_t weight = CGROUP_WEIGHT_DEFAULT;
+
+                        if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
+                            c->startup_io_weight != CGROUP_WEIGHT_INVALID)
+                                weight = c->startup_io_weight;
+                        else if (c->io_weight != CGROUP_WEIGHT_INVALID)
+                                weight = c->io_weight;
+
+                        xsprintf(buf, "default %" PRIu64 "\n", weight);
+                        r = cg_set_attribute("io", path, "io.weight", buf);
+                        if (r < 0)
+                                log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
+                                               "Failed to set io.weight on %s: %m", path);
+
+                        /* FIXME: no way to reset this list */
+                        LIST_FOREACH(device_weights, w, c->io_device_weights) {
+                                dev_t dev;
+
+                                r = lookup_block_device(w->path, &dev);
+                                if (r < 0)
+                                        continue;
+
+                                xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), w->weight);
+                                r = cg_set_attribute("io", path, "io.weight", buf);
+                                if (r < 0)
+                                        log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
+                                                       "Failed to set io.weight on %s: %m", path);
+                        }
+                }
+
+                LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) {
+                        char rbps_buf[DECIMAL_STR_MAX(uint64_t)] = "max";
+                        char wbps_buf[DECIMAL_STR_MAX(uint64_t)] = "max";
+                        char buf[DECIMAL_STR_MAX(dev_t)*2+2+(5+DECIMAL_STR_MAX(uint64_t)+1)*2];
+                        dev_t dev;
+                        unsigned n = 0;
+
+                        r = lookup_block_device(l->path, &dev);
+                        if (r < 0)
+                                continue;
+
+                        if (l->rbps_max != CGROUP_LIMIT_MAX) {
+                                xsprintf(rbps_buf, "%" PRIu64, l->rbps_max);
+                                n++;
+                        }
+
+                        if (l->wbps_max != CGROUP_LIMIT_MAX) {
+                                xsprintf(wbps_buf, "%" PRIu64, l->wbps_max);
+                                n++;
+                        }
+
+                        xsprintf(buf, "%u:%u rbps=%s wbps=%s\n", major(dev), minor(dev), rbps_buf, wbps_buf);
+                        r = cg_set_attribute("io", path, "io.max", buf);
+                        if (r < 0)
+                                log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
+                                               "Failed to set io.max on %s: %m", path);
+
+                        /* If @l contained no config, we just cleared the kernel
+                           counterpart too. No reason to keep @l around. */
+                        if (!n)
+                                cgroup_context_free_io_device_limit(c, l);
+                }
+        }
+
         if (mask & CGROUP_MASK_BLKIO) {
                 char buf[MAX(DECIMAL_STR_MAX(uint64_t)+1,
                              DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1)];
@@ -362,7 +494,7 @@ void cgroup_context_apply(CGroupContext *c, CGroupMask mask, const char *path, M
                         LIST_FOREACH(device_weights, w, c->blockio_device_weights) {
                                 dev_t dev;
 
-                                r = lookup_blkio_device(w->path, &dev);
+                                r = lookup_block_device(w->path, &dev);
                                 if (r < 0)
                                         continue;
 
@@ -379,7 +511,7 @@ void cgroup_context_apply(CGroupContext *c, CGroupMask mask, const char *path, M
                         const char *a;
                         dev_t dev;
 
-                        r = lookup_blkio_device(b->path, &dev);
+                        r = lookup_block_device(b->path, &dev);
                         if (r < 0)
                                 continue;
 
@@ -506,6 +638,13 @@ CGroupMask cgroup_context_get_mask(CGroupContext *c) {
             c->cpu_quota_per_sec_usec != USEC_INFINITY)
                 mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU;
 
+        if (c->io_accounting ||
+            c->io_weight != CGROUP_WEIGHT_INVALID ||
+            c->startup_io_weight != CGROUP_WEIGHT_INVALID ||
+            c->io_device_weights ||
+            c->io_device_limits)
+                mask |= CGROUP_MASK_IO;
+
         if (c->blockio_accounting ||
             c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
             c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
@@ -1608,7 +1747,7 @@ void manager_invalidate_startup_units(Manager *m) {
         assert(m);
 
         SET_FOREACH(u, m->startup_units, i)
-                unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_BLKIO);
+                unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO);
 }
 
 static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = {
index 360bbca30ff1c1f2b1c6c1787b161952c4f84494..a533923072bffc990a098e2bcadf06be05f38738 100644 (file)
@@ -26,6 +26,8 @@
 
 typedef struct CGroupContext CGroupContext;
 typedef struct CGroupDeviceAllow CGroupDeviceAllow;
+typedef struct CGroupIODeviceWeight CGroupIODeviceWeight;
+typedef struct CGroupIODeviceLimit CGroupIODeviceLimit;
 typedef struct CGroupBlockIODeviceWeight CGroupBlockIODeviceWeight;
 typedef struct CGroupBlockIODeviceBandwidth CGroupBlockIODeviceBandwidth;
 
@@ -53,6 +55,19 @@ struct CGroupDeviceAllow {
         bool m:1;
 };
 
+struct CGroupIODeviceWeight {
+        LIST_FIELDS(CGroupIODeviceWeight, device_weights);
+        char *path;
+        uint64_t weight;
+};
+
+struct CGroupIODeviceLimit {
+        LIST_FIELDS(CGroupIODeviceLimit, device_limits);
+        char *path;
+        uint64_t rbps_max;
+        uint64_t wbps_max;
+};
+
 struct CGroupBlockIODeviceWeight {
         LIST_FIELDS(CGroupBlockIODeviceWeight, device_weights);
         char *path;
@@ -68,10 +83,18 @@ struct CGroupBlockIODeviceBandwidth {
 
 struct CGroupContext {
         bool cpu_accounting;
+        bool io_accounting;
         bool blockio_accounting;
         bool memory_accounting;
         bool tasks_accounting;
 
+        /* For unified hierarchy */
+        uint64_t io_weight;
+        uint64_t startup_io_weight;
+        LIST_HEAD(CGroupIODeviceWeight, io_device_weights);
+        LIST_HEAD(CGroupIODeviceLimit, io_device_limits);
+
+        /* For legacy hierarchies */
         uint64_t cpu_shares;
         uint64_t startup_cpu_shares;
         usec_t cpu_quota_per_sec_usec;
@@ -86,6 +109,7 @@ struct CGroupContext {
         CGroupDevicePolicy device_policy;
         LIST_HEAD(CGroupDeviceAllow, device_allow);
 
+        /* Common */
         uint64_t tasks_max;
 
         bool delegate;
@@ -102,6 +126,8 @@ void cgroup_context_apply(CGroupContext *c, CGroupMask mask, const char *path, M
 CGroupMask cgroup_context_get_mask(CGroupContext *c);
 
 void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a);
+void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w);
+void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l);
 void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w);
 void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b);
 
index 859d155ec1daf375e275ab49dcd26ac962fa1a95..a2a4a6249cd6b265e6916df49aecf3a48380e135 100644 (file)
 
 static BUS_DEFINE_PROPERTY_GET_ENUM(property_get_cgroup_device_policy, cgroup_device_policy, CGroupDevicePolicy);
 
+static int property_get_io_device_weight(
+                sd_bus *bus,
+                const char *path,
+                const char *interface,
+                const char *property,
+                sd_bus_message *reply,
+                void *userdata,
+                sd_bus_error *error) {
+
+        CGroupContext *c = userdata;
+        CGroupIODeviceWeight *w;
+        int r;
+
+        assert(bus);
+        assert(reply);
+        assert(c);
+
+        r = sd_bus_message_open_container(reply, 'a', "(st)");
+        if (r < 0)
+                return r;
+
+        LIST_FOREACH(device_weights, w, c->io_device_weights) {
+                r = sd_bus_message_append(reply, "(st)", w->path, w->weight);
+                if (r < 0)
+                        return r;
+        }
+
+        return sd_bus_message_close_container(reply);
+}
+
+static int property_get_io_device_limits(
+                sd_bus *bus,
+                const char *path,
+                const char *interface,
+                const char *property,
+                sd_bus_message *reply,
+                void *userdata,
+                sd_bus_error *error) {
+
+        CGroupContext *c = userdata;
+        CGroupIODeviceLimit *l;
+        int r;
+
+        assert(bus);
+        assert(reply);
+        assert(c);
+
+        r = sd_bus_message_open_container(reply, 'a', "(st)");
+        if (r < 0)
+                return r;
+
+        LIST_FOREACH(device_limits, l, c->io_device_limits) {
+                uint64_t v;
+
+                if (streq(property, "IOReadBandwidthMax"))
+                        v = l->rbps_max;
+                else
+                        v = l->wbps_max;
+
+                if (v == CGROUP_LIMIT_MAX)
+                        continue;
+
+                r = sd_bus_message_append(reply, "(st)", l->path, v);
+                if (r < 0)
+                        return r;
+        }
+
+        return sd_bus_message_close_container(reply);
+}
+
 static int property_get_blockio_device_weight(
                 sd_bus *bus,
                 const char *path,
@@ -141,6 +211,12 @@ const sd_bus_vtable bus_cgroup_vtable[] = {
         SD_BUS_PROPERTY("CPUShares", "t", NULL, offsetof(CGroupContext, cpu_shares), 0),
         SD_BUS_PROPERTY("StartupCPUShares", "t", NULL, offsetof(CGroupContext, startup_cpu_shares), 0),
         SD_BUS_PROPERTY("CPUQuotaPerSecUSec", "t", bus_property_get_usec, offsetof(CGroupContext, cpu_quota_per_sec_usec), 0),
+        SD_BUS_PROPERTY("IOAccounting", "b", bus_property_get_bool, offsetof(CGroupContext, io_accounting), 0),
+        SD_BUS_PROPERTY("IOWeight", "t", NULL, offsetof(CGroupContext, io_weight), 0),
+        SD_BUS_PROPERTY("StartupIOWeight", "t", NULL, offsetof(CGroupContext, startup_io_weight), 0),
+        SD_BUS_PROPERTY("IODeviceWeight", "a(st)", property_get_io_device_weight, 0, 0),
+        SD_BUS_PROPERTY("IOReadBandwidthMax", "a(st)", property_get_io_device_limits, 0, 0),
+        SD_BUS_PROPERTY("IOWriteBandwidthMax", "a(st)", property_get_io_device_limits, 0, 0),
         SD_BUS_PROPERTY("BlockIOAccounting", "b", bus_property_get_bool, offsetof(CGroupContext, blockio_accounting), 0),
         SD_BUS_PROPERTY("BlockIOWeight", "t", NULL, offsetof(CGroupContext, blockio_weight), 0),
         SD_BUS_PROPERTY("StartupBlockIOWeight", "t", NULL, offsetof(CGroupContext, startup_blockio_weight), 0),
@@ -281,6 +357,238 @@ int bus_cgroup_set_property(
 
                 return 1;
 
+        } else if (streq(name, "IOAccounting")) {
+                int b;
+
+                r = sd_bus_message_read(message, "b", &b);
+                if (r < 0)
+                        return r;
+
+                if (mode != UNIT_CHECK) {
+                        c->io_accounting = b;
+                        unit_invalidate_cgroup(u, CGROUP_MASK_IO);
+                        unit_write_drop_in_private(u, mode, name, b ? "IOAccounting=yes" : "IOAccounting=no");
+                }
+
+                return 1;
+
+        } else if (streq(name, "IOWeight")) {
+                uint64_t weight;
+
+                r = sd_bus_message_read(message, "t", &weight);
+                if (r < 0)
+                        return r;
+
+                if (!CGROUP_WEIGHT_IS_OK(weight))
+                        return sd_bus_error_set_errnof(error, EINVAL, "IOWeight value out of range");
+
+                if (mode != UNIT_CHECK) {
+                        c->io_weight = weight;
+                        unit_invalidate_cgroup(u, CGROUP_MASK_IO);
+
+                        if (weight == CGROUP_WEIGHT_INVALID)
+                                unit_write_drop_in_private(u, mode, name, "IOWeight=");
+                        else
+                                unit_write_drop_in_private_format(u, mode, name, "IOWeight=%" PRIu64, weight);
+                }
+
+                return 1;
+
+        } else if (streq(name, "StartupIOWeight")) {
+                uint64_t weight;
+
+                r = sd_bus_message_read(message, "t", &weight);
+                if (r < 0)
+                        return r;
+
+                if (CGROUP_WEIGHT_IS_OK(weight))
+                        return sd_bus_error_set_errnof(error, EINVAL, "StartupIOWeight value out of range");
+
+                if (mode != UNIT_CHECK) {
+                        c->startup_io_weight = weight;
+                        unit_invalidate_cgroup(u, CGROUP_MASK_IO);
+
+                        if (weight == CGROUP_WEIGHT_INVALID)
+                                unit_write_drop_in_private(u, mode, name, "StartupIOWeight=");
+                        else
+                                unit_write_drop_in_private_format(u, mode, name, "StartupIOWeight=%" PRIu64, weight);
+                }
+
+                return 1;
+
+        } else if (streq(name, "IOReadBandwidthMax") || streq(name, "IOWriteBandwidthMax")) {
+                const char *path;
+                bool read = true;
+                unsigned n = 0;
+                uint64_t u64;
+
+                if (streq(name, "IOWriteBandwidthMax"))
+                        read = false;
+
+                r = sd_bus_message_enter_container(message, 'a', "(st)");
+                if (r < 0)
+                        return r;
+
+                while ((r = sd_bus_message_read(message, "(st)", &path, &u64)) > 0) {
+
+                        if (mode != UNIT_CHECK) {
+                                CGroupIODeviceLimit *a = NULL, *b;
+
+                                LIST_FOREACH(device_limits, b, c->io_device_limits) {
+                                        if (path_equal(path, b->path)) {
+                                                a = b;
+                                                break;
+                                        }
+                                }
+
+                                if (!a) {
+                                        a = new0(CGroupIODeviceLimit, 1);
+                                        if (!a)
+                                                return -ENOMEM;
+
+                                        a->path = strdup(path);
+                                        if (!a->path) {
+                                                free(a);
+                                                return -ENOMEM;
+                                        }
+
+                                        a->rbps_max = CGROUP_LIMIT_MAX;
+                                        a->wbps_max = CGROUP_LIMIT_MAX;
+
+                                        LIST_PREPEND(device_limits, c->io_device_limits, a);
+                                }
+
+                                if (read)
+                                        a->rbps_max = u64;
+                                else
+                                        a->wbps_max = u64;
+                        }
+
+                        n++;
+                }
+                if (r < 0)
+                        return r;
+
+                r = sd_bus_message_exit_container(message);
+                if (r < 0)
+                        return r;
+
+                if (mode != UNIT_CHECK) {
+                        CGroupIODeviceLimit *a;
+                        _cleanup_free_ char *buf = NULL;
+                        _cleanup_fclose_ FILE *f = NULL;
+                        size_t size = 0;
+
+                        if (n == 0) {
+                                LIST_FOREACH(device_limits, a, c->io_device_limits)
+                                        if (read)
+                                                a->rbps_max = CGROUP_LIMIT_MAX;
+                                        else
+                                                a->wbps_max = CGROUP_LIMIT_MAX;
+                        }
+
+                        unit_invalidate_cgroup(u, CGROUP_MASK_IO);
+
+                        f = open_memstream(&buf, &size);
+                        if (!f)
+                                return -ENOMEM;
+
+                        if (read) {
+                                fputs("IOReadBandwidthMax=\n", f);
+                                LIST_FOREACH(device_limits, a, c->io_device_limits)
+                                        if (a->rbps_max != CGROUP_LIMIT_MAX)
+                                                fprintf(f, "IOReadBandwidthMax=%s %" PRIu64 "\n", a->path, a->rbps_max);
+                        } else {
+                                fputs("IOWriteBandwidthMax=\n", f);
+                                LIST_FOREACH(device_limits, a, c->io_device_limits)
+                                        if (a->wbps_max != CGROUP_LIMIT_MAX)
+                                                fprintf(f, "IOWriteBandwidthMax=%s %" PRIu64 "\n", a->path, a->wbps_max);
+                        }
+
+                        r = fflush_and_check(f);
+                        if (r < 0)
+                                return r;
+                        unit_write_drop_in_private(u, mode, name, buf);
+                }
+
+                return 1;
+
+        } else if (streq(name, "IODeviceWeight")) {
+                const char *path;
+                uint64_t weight;
+                unsigned n = 0;
+
+                r = sd_bus_message_enter_container(message, 'a', "(st)");
+                if (r < 0)
+                        return r;
+
+                while ((r = sd_bus_message_read(message, "(st)", &path, &weight)) > 0) {
+
+                        if (!CGROUP_WEIGHT_IS_OK(weight) || weight == CGROUP_WEIGHT_INVALID)
+                                return sd_bus_error_set_errnof(error, EINVAL, "IODeviceWeight out of range");
+
+                        if (mode != UNIT_CHECK) {
+                                CGroupIODeviceWeight *a = NULL, *b;
+
+                                LIST_FOREACH(device_weights, b, c->io_device_weights) {
+                                        if (path_equal(b->path, path)) {
+                                                a = b;
+                                                break;
+                                        }
+                                }
+
+                                if (!a) {
+                                        a = new0(CGroupIODeviceWeight, 1);
+                                        if (!a)
+                                                return -ENOMEM;
+
+                                        a->path = strdup(path);
+                                        if (!a->path) {
+                                                free(a);
+                                                return -ENOMEM;
+                                        }
+                                        LIST_PREPEND(device_weights,c->io_device_weights, a);
+                                }
+
+                                a->weight = weight;
+                        }
+
+                        n++;
+                }
+
+                r = sd_bus_message_exit_container(message);
+                if (r < 0)
+                        return r;
+
+                if (mode != UNIT_CHECK) {
+                        _cleanup_free_ char *buf = NULL;
+                        _cleanup_fclose_ FILE *f = NULL;
+                        CGroupIODeviceWeight *a;
+                        size_t size = 0;
+
+                        if (n == 0) {
+                                while (c->io_device_weights)
+                                        cgroup_context_free_io_device_weight(c, c->io_device_weights);
+                        }
+
+                        unit_invalidate_cgroup(u, CGROUP_MASK_IO);
+
+                        f = open_memstream(&buf, &size);
+                        if (!f)
+                                return -ENOMEM;
+
+                        fputs("IODeviceWeight=\n", f);
+                        LIST_FOREACH(device_weights, a, c->io_device_weights)
+                                fprintf(f, "IODeviceWeight=%s %" PRIu64 "\n", a->path, a->weight);
+
+                        r = fflush_and_check(f);
+                        if (r < 0)
+                                return r;
+                        unit_write_drop_in_private(u, mode, name, buf);
+                }
+
+                return 1;
+
         } else if (streq(name, "BlockIOAccounting")) {
                 int b;
 
index 928b913c7b88b7e3028acc191db364da96f187a0..ad45611d9d72bf9d3a081091d592ed272ff86a05 100644 (file)
@@ -120,6 +120,12 @@ $1.MemoryAccounting,             config_parse_bool,                  0,
 $1.MemoryLimit,                  config_parse_memory_limit,          0,                             offsetof($1, cgroup_context)
 $1.DeviceAllow,                  config_parse_device_allow,          0,                             offsetof($1, cgroup_context)
 $1.DevicePolicy,                 config_parse_device_policy,         0,                             offsetof($1, cgroup_context.device_policy)
+$1.IOAccounting,                 config_parse_bool,                  0,                             offsetof($1, cgroup_context.io_accounting)
+$1.IOWeight,                     config_parse_io_weight,             0,                             offsetof($1, cgroup_context.io_weight)
+$1.StartupIOWeight,              config_parse_io_weight,             0,                             offsetof($1, cgroup_context.startup_io_weight)
+$1.IODeviceWeight,               config_parse_io_device_weight,      0,                             offsetof($1, cgroup_context)
+$1.IOReadBandwidthMax,           config_parse_io_limit,              0,                             offsetof($1, cgroup_context)
+$1.IOWriteBandwidthMax,          config_parse_io_limit,              0,                             offsetof($1, cgroup_context)
 $1.BlockIOAccounting,            config_parse_bool,                  0,                             offsetof($1, cgroup_context.blockio_accounting)
 $1.BlockIOWeight,                config_parse_blockio_weight,        0,                             offsetof($1, cgroup_context.blockio_weight)
 $1.StartupBlockIOWeight,         config_parse_blockio_weight,        0,                             offsetof($1, cgroup_context.startup_blockio_weight)
index 1a8c03904cd1d799bca119b3ebf4d36ecd3d0c90..9d7329e924aae2cb339706aff68417ab47549fcd 100644 (file)
@@ -2904,6 +2904,196 @@ int config_parse_device_allow(
         return 0;
 }
 
+int config_parse_io_weight(
+                const char *unit,
+                const char *filename,
+                unsigned line,
+                const char *section,
+                unsigned section_line,
+                const char *lvalue,
+                int ltype,
+                const char *rvalue,
+                void *data,
+                void *userdata) {
+
+        uint64_t *weight = data;
+        int r;
+
+        assert(filename);
+        assert(lvalue);
+        assert(rvalue);
+
+        r = cg_weight_parse(rvalue, weight);
+        if (r < 0) {
+                log_syntax(unit, LOG_ERR, filename, line, r, "IO weight '%s' invalid. Ignoring.", rvalue);
+                return 0;
+        }
+
+        return 0;
+}
+
+int config_parse_io_device_weight(
+                const char *unit,
+                const char *filename,
+                unsigned line,
+                const char *section,
+                unsigned section_line,
+                const char *lvalue,
+                int ltype,
+                const char *rvalue,
+                void *data,
+                void *userdata) {
+
+        _cleanup_free_ char *path = NULL;
+        CGroupIODeviceWeight *w;
+        CGroupContext *c = data;
+        const char *weight;
+        uint64_t u;
+        size_t n;
+        int r;
+
+        assert(filename);
+        assert(lvalue);
+        assert(rvalue);
+
+        if (isempty(rvalue)) {
+                while (c->io_device_weights)
+                        cgroup_context_free_io_device_weight(c, c->io_device_weights);
+
+                return 0;
+        }
+
+        n = strcspn(rvalue, WHITESPACE);
+        weight = rvalue + n;
+        weight += strspn(weight, WHITESPACE);
+
+        if (isempty(weight)) {
+                log_syntax(unit, LOG_ERR, filename, line, 0, "Expected block device and device weight. Ignoring.");
+                return 0;
+        }
+
+        path = strndup(rvalue, n);
+        if (!path)
+                return log_oom();
+
+        if (!path_startswith(path, "/dev")) {
+                log_syntax(unit, LOG_ERR, filename, line, 0, "Invalid device node path '%s'. Ignoring.", path);
+                return 0;
+        }
+
+        r = cg_weight_parse(weight, &u);
+        if (r < 0) {
+                log_syntax(unit, LOG_ERR, filename, line, r, "IO weight '%s' invalid. Ignoring.", weight);
+                return 0;
+        }
+
+        assert(u != CGROUP_WEIGHT_INVALID);
+
+        w = new0(CGroupIODeviceWeight, 1);
+        if (!w)
+                return log_oom();
+
+        w->path = path;
+        path = NULL;
+
+        w->weight = u;
+
+        LIST_PREPEND(device_weights, c->io_device_weights, w);
+        return 0;
+}
+
+int config_parse_io_limit(
+                const char *unit,
+                const char *filename,
+                unsigned line,
+                const char *section,
+                unsigned section_line,
+                const char *lvalue,
+                int ltype,
+                const char *rvalue,
+                void *data,
+                void *userdata) {
+
+        _cleanup_free_ char *path = NULL;
+        CGroupIODeviceLimit *l = NULL, *t;
+        CGroupContext *c = data;
+        const char *limit;
+        uint64_t num;
+        bool read;
+        size_t n;
+        int r;
+
+        assert(filename);
+        assert(lvalue);
+        assert(rvalue);
+
+        read = streq("IOReadBandwidthMax", lvalue);
+
+        if (isempty(rvalue)) {
+                LIST_FOREACH(device_limits, l, c->io_device_limits)
+                        if (read)
+                                l->rbps_max = CGROUP_LIMIT_MAX;
+                        else
+                                l->wbps_max = CGROUP_LIMIT_MAX;
+                return 0;
+        }
+
+        n = strcspn(rvalue, WHITESPACE);
+        limit = rvalue + n;
+        limit += strspn(limit, WHITESPACE);
+
+        if (!*limit) {
+                log_syntax(unit, LOG_ERR, filename, line, 0, "Expected space separated pair of device node and bandwidth. Ignoring.");
+                return 0;
+        }
+
+        path = strndup(rvalue, n);
+        if (!path)
+                return log_oom();
+
+        if (!path_startswith(path, "/dev")) {
+                log_syntax(unit, LOG_ERR, filename, line, 0, "Invalid device node path '%s'. Ignoring.", path);
+                return 0;
+        }
+
+        if (streq("max", limit)) {
+                num = CGROUP_LIMIT_MAX;
+        } else {
+                r = parse_size(limit, 1000, &num);
+                if (r < 0 || num <= 0) {
+                        log_syntax(unit, LOG_ERR, filename, line, r, "IO Limit '%s' invalid. Ignoring.", rvalue);
+                        return 0;
+                }
+        }
+
+        LIST_FOREACH(device_limits, t, c->io_device_limits) {
+                if (path_equal(path, t->path)) {
+                        l = t;
+                        break;
+                }
+        }
+
+        if (!l) {
+                l = new0(CGroupIODeviceLimit, 1);
+                if (!l)
+                        return log_oom();
+
+                l->path = path;
+                path = NULL;
+                l->rbps_max = CGROUP_LIMIT_MAX;
+                l->wbps_max = CGROUP_LIMIT_MAX;
+
+                LIST_PREPEND(device_limits, c->io_device_limits, l);
+        }
+
+        if (read)
+                l->rbps_max = num;
+        else
+                l->wbps_max = num;
+
+        return 0;
+}
+
 int config_parse_blockio_weight(
                 const char *unit,
                 const char *filename,
@@ -3824,6 +4014,9 @@ void unit_dump_config_items(FILE *f) {
                 { config_parse_memory_limit,          "LIMIT" },
                 { config_parse_device_allow,          "DEVICE" },
                 { config_parse_device_policy,         "POLICY" },
+                { config_parse_io_limit,              "LIMIT" },
+                { config_parse_io_weight,             "WEIGHT" },
+                { config_parse_io_device_weight,      "DEVICEWEIGHT" },
                 { config_parse_blockio_bandwidth,     "BANDWIDTH" },
                 { config_parse_blockio_weight,        "WEIGHT" },
                 { config_parse_blockio_device_weight, "DEVICEWEIGHT" },
index 34f15afa6288dc38b24ddd28f9b7c544e94d90c1..b36a2e3a028fdb366881a81abda0b84fbe6e5658 100644 (file)
@@ -86,6 +86,9 @@ int config_parse_memory_limit(const char *unit, const char *filename, unsigned l
 int config_parse_tasks_max(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
 int config_parse_device_policy(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
 int config_parse_device_allow(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
+int config_parse_io_weight(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
+int config_parse_io_device_weight(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
+int config_parse_io_limit(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
 int config_parse_blockio_weight(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
 int config_parse_blockio_device_weight(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
 int config_parse_blockio_bandwidth(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
index ed4d42c8ccd90a17b7c0c5835c9f96441fbcf52f..6397aadc73e5f82d525e8475e3cc6bb2a1d12696 100644 (file)
@@ -122,6 +122,7 @@ static usec_t arg_default_timer_accuracy_usec = 1 * USEC_PER_MINUTE;
 static Set* arg_syscall_archs = NULL;
 static FILE* arg_serialization = NULL;
 static bool arg_default_cpu_accounting = false;
+static bool arg_default_io_accounting = false;
 static bool arg_default_blockio_accounting = false;
 static bool arg_default_memory_accounting = false;
 static bool arg_default_tasks_accounting = true;
@@ -691,6 +692,7 @@ static int parse_config_file(void) {
                 { "Manager", "DefaultLimitRTPRIO",        config_parse_limit,            RLIMIT_RTPRIO, arg_default_rlimit         },
                 { "Manager", "DefaultLimitRTTIME",        config_parse_limit,            RLIMIT_RTTIME, arg_default_rlimit         },
                 { "Manager", "DefaultCPUAccounting",      config_parse_bool,             0, &arg_default_cpu_accounting            },
+                { "Manager", "DefaultIOAccounting",       config_parse_bool,             0, &arg_default_io_accounting             },
                 { "Manager", "DefaultBlockIOAccounting",  config_parse_bool,             0, &arg_default_blockio_accounting        },
                 { "Manager", "DefaultMemoryAccounting",   config_parse_bool,             0, &arg_default_memory_accounting         },
                 { "Manager", "DefaultTasksAccounting",    config_parse_bool,             0, &arg_default_tasks_accounting          },
@@ -733,6 +735,7 @@ static void manager_set_defaults(Manager *m) {
         m->default_start_limit_interval = arg_default_start_limit_interval;
         m->default_start_limit_burst = arg_default_start_limit_burst;
         m->default_cpu_accounting = arg_default_cpu_accounting;
+        m->default_io_accounting = arg_default_io_accounting;
         m->default_blockio_accounting = arg_default_blockio_accounting;
         m->default_memory_accounting = arg_default_memory_accounting;
         m->default_tasks_accounting = arg_default_tasks_accounting;
index 17f84e6963068b392ab96e1e5a69122e29fc6cbf..f23e4056c41aecf5f13272c37e25e094e7fe672d 100644 (file)
@@ -252,6 +252,7 @@ struct Manager {
 
         bool default_cpu_accounting;
         bool default_memory_accounting;
+        bool default_io_accounting;
         bool default_blockio_accounting;
         bool default_tasks_accounting;
 
index eacd7ee2822af102ea77faa8f44adf9d4b0ecd5a..db8b7acd7866db5e1908dd6744327200b5e280fe 100644 (file)
@@ -38,6 +38,7 @@
 #DefaultStartLimitBurst=5
 #DefaultEnvironment=
 #DefaultCPUAccounting=no
+#DefaultIOAccounting=no
 #DefaultBlockIOAccounting=no
 #DefaultMemoryAccounting=no
 #DefaultTasksAccounting=yes
index 8153515e894b964d9b3d4433847a5eb26f5f89fc..04addc1f70308a69b91cfa1ab22d41c92353d20c 100644 (file)
@@ -132,6 +132,7 @@ static void unit_init(Unit *u) {
                  * been initialized */
 
                 cc->cpu_accounting = u->manager->default_cpu_accounting;
+                cc->io_accounting = u->manager->default_io_accounting;
                 cc->blockio_accounting = u->manager->default_blockio_accounting;
                 cc->memory_accounting = u->manager->default_memory_accounting;
                 cc->tasks_accounting = u->manager->default_tasks_accounting;
@@ -1213,6 +1214,7 @@ static int unit_add_startup_units(Unit *u) {
                 return 0;
 
         if (c->startup_cpu_shares == CGROUP_CPU_SHARES_INVALID &&
+            c->startup_io_weight == CGROUP_WEIGHT_INVALID &&
             c->startup_blockio_weight == CGROUP_BLKIO_WEIGHT_INVALID)
                 return 0;
 
index 2b755cea28bc390981454158f903106446c3fe8e..94d1c1d63cac210317235bae3661dd91b6566d3f 100644 (file)
@@ -154,7 +154,7 @@ int bus_append_unit_property_assignment(sd_bus_message *m, const char *assignmen
                 r = sd_bus_message_append(m, "sv", sn, "t", l.rlim_cur);
 
         } else if (STR_IN_SET(field,
-                       "CPUAccounting", "MemoryAccounting", "BlockIOAccounting", "TasksAccounting",
+                       "CPUAccounting", "MemoryAccounting", "IOAccounting", "BlockIOAccounting", "TasksAccounting",
                        "SendSIGHUP", "SendSIGKILL", "WakeSystem", "DefaultDependencies",
                        "IgnoreSIGPIPE", "TTYVHangup", "TTYReset", "RemainAfterExit",
                        "PrivateTmp", "PrivateDevices", "PrivateNetwork", "NoNewPrivileges",
@@ -207,6 +207,17 @@ int bus_append_unit_property_assignment(sd_bus_message *m, const char *assignmen
 
                 r = sd_bus_message_append(m, "v", "t", u);
 
+        } else if (STR_IN_SET(field, "IOWeight", "StartupIOWeight")) {
+                uint64_t u;
+
+                r = cg_weight_parse(eq, &u);
+                if (r < 0) {
+                        log_error("Failed to parse %s value %s.", field, eq);
+                        return -EINVAL;
+                }
+
+                r = sd_bus_message_append(m, "v", "t", u);
+
         } else if (STR_IN_SET(field, "BlockIOWeight", "StartupBlockIOWeight")) {
                 uint64_t u;
 
@@ -273,7 +284,8 @@ int bus_append_unit_property_assignment(sd_bus_message *m, const char *assignmen
                         r = sd_bus_message_append(m, "v", "a(ss)", 1, path, rwm);
                 }
 
-        } else if (STR_IN_SET(field, "BlockIOReadBandwidth", "BlockIOWriteBandwidth")) {
+        } else if (STR_IN_SET(field, "IOReadBandwidthMax", "IOWriteBandwidthMax",
+                              "BlockIOReadBandwidth", "BlockIOWriteBandwidth")) {
 
                 if (isempty(eq))
                         r = sd_bus_message_append(m, "v", "a(st)", 0);
@@ -295,16 +307,20 @@ int bus_append_unit_property_assignment(sd_bus_message *m, const char *assignmen
                                 return -EINVAL;
                         }
 
-                        r = parse_size(bandwidth, 1000, &bytes);
-                        if (r < 0) {
-                                log_error("Failed to parse byte value %s.", bandwidth);
-                                return -EINVAL;
+                        if (streq(bandwidth, "max")) {
+                                bytes = CGROUP_LIMIT_MAX;
+                        } else {
+                                r = parse_size(bandwidth, 1000, &bytes);
+                                if (r < 0) {
+                                        log_error("Failed to parse byte value %s.", bandwidth);
+                                        return -EINVAL;
+                                }
                         }
 
                         r = sd_bus_message_append(m, "v", "a(st)", 1, path, bytes);
                 }
 
-        } else if (streq(field, "BlockIODeviceWeight")) {
+        } else if (STR_IN_SET(field, "IODeviceWeight", "BlockIODeviceWeight")) {
 
                 if (isempty(eq))
                         r = sd_bus_message_append(m, "v", "a(st)", 0);
index bec4f31b3994480cd6d19ced27840510bbecc656..387de025c5e4ab35607922727e07570d40d901e1 100644 (file)
@@ -4428,7 +4428,7 @@ static int print_property(const char *name, sd_bus_message *m, const char *conte
 
                         return 0;
 
-                } else if (contents[1] == SD_BUS_TYPE_STRUCT_BEGIN && streq(name, "BlockIODeviceWeight")) {
+                } else if (contents[1] == SD_BUS_TYPE_STRUCT_BEGIN && (streq(name, "IODeviceWeight") || streq(name, "BlockIODeviceWeight"))) {
                         const char *path;
                         uint64_t weight;
 
@@ -4447,7 +4447,8 @@ static int print_property(const char *name, sd_bus_message *m, const char *conte
 
                         return 0;
 
-                } else if (contents[1] == SD_BUS_TYPE_STRUCT_BEGIN && (streq(name, "BlockIOReadBandwidth") || streq(name, "BlockIOWriteBandwidth"))) {
+                } else if (contents[1] == SD_BUS_TYPE_STRUCT_BEGIN && (streq(name, "IOReadBandwidthMax") || streq(name, "IOWriteBandwidthMax") ||
+                                                                       streq(name, "BlockIOReadBandwidth") || streq(name, "BlockIOWriteBandwidth"))) {
                         const char *path;
                         uint64_t bandwidth;