2 This file is part of systemd.
4 Copyright 2013 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
23 #include "alloc-util.h"
24 #include "cgroup-util.h"
29 #include "parse-util.h"
30 #include "path-util.h"
31 #include "process-util.h"
33 #include "string-table.h"
34 #include "string-util.h"
35 #include "stdio-util.h"
37 #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
39 static void cgroup_compat_warn(void) {
40 static bool cgroup_compat_warned
= false;
42 if (cgroup_compat_warned
)
45 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. See cgroup-compat debug messages for details.");
46 cgroup_compat_warned
= true;
49 #define log_cgroup_compat(unit, fmt, ...) do { \
50 cgroup_compat_warn(); \
51 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
54 void cgroup_context_init(CGroupContext
*c
) {
57 /* Initialize everything to the kernel defaults, assuming the
58 * structure is preinitialized to 0 */
60 c
->cpu_shares
= CGROUP_CPU_SHARES_INVALID
;
61 c
->startup_cpu_shares
= CGROUP_CPU_SHARES_INVALID
;
62 c
->cpu_quota_per_sec_usec
= USEC_INFINITY
;
64 c
->memory_high
= CGROUP_LIMIT_MAX
;
65 c
->memory_max
= CGROUP_LIMIT_MAX
;
67 c
->memory_limit
= CGROUP_LIMIT_MAX
;
69 c
->io_weight
= CGROUP_WEIGHT_INVALID
;
70 c
->startup_io_weight
= CGROUP_WEIGHT_INVALID
;
72 c
->blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
;
73 c
->startup_blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
;
75 c
->tasks_max
= (uint64_t) -1;
78 void cgroup_context_free_device_allow(CGroupContext
*c
, CGroupDeviceAllow
*a
) {
82 LIST_REMOVE(device_allow
, c
->device_allow
, a
);
87 void cgroup_context_free_io_device_weight(CGroupContext
*c
, CGroupIODeviceWeight
*w
) {
91 LIST_REMOVE(device_weights
, c
->io_device_weights
, w
);
96 void cgroup_context_free_io_device_limit(CGroupContext
*c
, CGroupIODeviceLimit
*l
) {
100 LIST_REMOVE(device_limits
, c
->io_device_limits
, l
);
105 void cgroup_context_free_blockio_device_weight(CGroupContext
*c
, CGroupBlockIODeviceWeight
*w
) {
109 LIST_REMOVE(device_weights
, c
->blockio_device_weights
, w
);
114 void cgroup_context_free_blockio_device_bandwidth(CGroupContext
*c
, CGroupBlockIODeviceBandwidth
*b
) {
118 LIST_REMOVE(device_bandwidths
, c
->blockio_device_bandwidths
, b
);
123 void cgroup_context_done(CGroupContext
*c
) {
126 while (c
->io_device_weights
)
127 cgroup_context_free_io_device_weight(c
, c
->io_device_weights
);
129 while (c
->io_device_limits
)
130 cgroup_context_free_io_device_limit(c
, c
->io_device_limits
);
132 while (c
->blockio_device_weights
)
133 cgroup_context_free_blockio_device_weight(c
, c
->blockio_device_weights
);
135 while (c
->blockio_device_bandwidths
)
136 cgroup_context_free_blockio_device_bandwidth(c
, c
->blockio_device_bandwidths
);
138 while (c
->device_allow
)
139 cgroup_context_free_device_allow(c
, c
->device_allow
);
142 void cgroup_context_dump(CGroupContext
*c
, FILE* f
, const char *prefix
) {
143 CGroupIODeviceLimit
*il
;
144 CGroupIODeviceWeight
*iw
;
145 CGroupBlockIODeviceBandwidth
*b
;
146 CGroupBlockIODeviceWeight
*w
;
147 CGroupDeviceAllow
*a
;
148 char u
[FORMAT_TIMESPAN_MAX
];
153 prefix
= strempty(prefix
);
156 "%sCPUAccounting=%s\n"
157 "%sIOAccounting=%s\n"
158 "%sBlockIOAccounting=%s\n"
159 "%sMemoryAccounting=%s\n"
160 "%sTasksAccounting=%s\n"
161 "%sCPUShares=%" PRIu64
"\n"
162 "%sStartupCPUShares=%" PRIu64
"\n"
163 "%sCPUQuotaPerSecSec=%s\n"
164 "%sIOWeight=%" PRIu64
"\n"
165 "%sStartupIOWeight=%" PRIu64
"\n"
166 "%sBlockIOWeight=%" PRIu64
"\n"
167 "%sStartupBlockIOWeight=%" PRIu64
"\n"
168 "%sMemoryLow=%" PRIu64
"\n"
169 "%sMemoryHigh=%" PRIu64
"\n"
170 "%sMemoryMax=%" PRIu64
"\n"
171 "%sMemoryLimit=%" PRIu64
"\n"
172 "%sTasksMax=%" PRIu64
"\n"
173 "%sDevicePolicy=%s\n"
175 prefix
, yes_no(c
->cpu_accounting
),
176 prefix
, yes_no(c
->io_accounting
),
177 prefix
, yes_no(c
->blockio_accounting
),
178 prefix
, yes_no(c
->memory_accounting
),
179 prefix
, yes_no(c
->tasks_accounting
),
180 prefix
, c
->cpu_shares
,
181 prefix
, c
->startup_cpu_shares
,
182 prefix
, format_timespan(u
, sizeof(u
), c
->cpu_quota_per_sec_usec
, 1),
183 prefix
, c
->io_weight
,
184 prefix
, c
->startup_io_weight
,
185 prefix
, c
->blockio_weight
,
186 prefix
, c
->startup_blockio_weight
,
187 prefix
, c
->memory_low
,
188 prefix
, c
->memory_high
,
189 prefix
, c
->memory_max
,
190 prefix
, c
->memory_limit
,
191 prefix
, c
->tasks_max
,
192 prefix
, cgroup_device_policy_to_string(c
->device_policy
),
193 prefix
, yes_no(c
->delegate
));
195 LIST_FOREACH(device_allow
, a
, c
->device_allow
)
197 "%sDeviceAllow=%s %s%s%s\n",
200 a
->r
? "r" : "", a
->w
? "w" : "", a
->m
? "m" : "");
202 LIST_FOREACH(device_weights
, iw
, c
->io_device_weights
)
204 "%sIODeviceWeight=%s %" PRIu64
,
209 LIST_FOREACH(device_limits
, il
, c
->io_device_limits
) {
210 char buf
[FORMAT_BYTES_MAX
];
211 CGroupIOLimitType type
;
213 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
214 if (il
->limits
[type
] != cgroup_io_limit_defaults
[type
])
218 cgroup_io_limit_type_to_string(type
),
220 format_bytes(buf
, sizeof(buf
), il
->limits
[type
]));
223 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
225 "%sBlockIODeviceWeight=%s %" PRIu64
,
230 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
231 char buf
[FORMAT_BYTES_MAX
];
233 if (b
->rbps
!= CGROUP_LIMIT_MAX
)
235 "%sBlockIOReadBandwidth=%s %s\n",
238 format_bytes(buf
, sizeof(buf
), b
->rbps
));
239 if (b
->wbps
!= CGROUP_LIMIT_MAX
)
241 "%sBlockIOWriteBandwidth=%s %s\n",
244 format_bytes(buf
, sizeof(buf
), b
->wbps
));
248 static int lookup_block_device(const char *p
, dev_t
*dev
) {
257 return log_warning_errno(errno
, "Couldn't stat device %s: %m", p
);
259 if (S_ISBLK(st
.st_mode
))
261 else if (major(st
.st_dev
) != 0) {
262 /* If this is not a device node then find the block
263 * device this file is stored on */
266 /* If this is a partition, try to get the originating
268 block_get_whole_disk(*dev
, dev
);
270 log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p
);
277 static int whitelist_device(const char *path
, const char *node
, const char *acc
) {
278 char buf
[2+DECIMAL_STR_MAX(dev_t
)*2+2+4];
285 if (stat(node
, &st
) < 0) {
286 log_warning("Couldn't stat device %s", node
);
290 if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
)) {
291 log_warning("%s is not a device.", node
);
297 S_ISCHR(st
.st_mode
) ? 'c' : 'b',
298 major(st
.st_rdev
), minor(st
.st_rdev
),
301 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
303 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
304 "Failed to set devices.allow on %s: %m", path
);
309 static int whitelist_major(const char *path
, const char *name
, char type
, const char *acc
) {
310 _cleanup_fclose_
FILE *f
= NULL
;
317 assert(type
== 'b' || type
== 'c');
319 f
= fopen("/proc/devices", "re");
321 return log_warning_errno(errno
, "Cannot open /proc/devices to resolve %s (%c): %m", name
, type
);
323 FOREACH_LINE(line
, f
, goto fail
) {
324 char buf
[2+DECIMAL_STR_MAX(unsigned)+3+4], *p
, *w
;
329 if (type
== 'c' && streq(line
, "Character devices:")) {
334 if (type
== 'b' && streq(line
, "Block devices:")) {
349 w
= strpbrk(p
, WHITESPACE
);
354 r
= safe_atou(p
, &maj
);
361 w
+= strspn(w
, WHITESPACE
);
363 if (fnmatch(name
, w
, 0) != 0)
372 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
374 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
375 "Failed to set devices.allow on %s: %m", path
);
381 log_warning_errno(errno
, "Failed to read /proc/devices: %m");
385 static bool cgroup_context_has_io_config(CGroupContext
*c
) {
386 return c
->io_accounting
||
387 c
->io_weight
!= CGROUP_WEIGHT_INVALID
||
388 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
||
389 c
->io_device_weights
||
393 static bool cgroup_context_has_blockio_config(CGroupContext
*c
) {
394 return c
->blockio_accounting
||
395 c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
396 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
397 c
->blockio_device_weights
||
398 c
->blockio_device_bandwidths
;
401 static uint64_t cgroup_context_io_weight(CGroupContext
*c
, ManagerState state
) {
402 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
403 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
)
404 return c
->startup_io_weight
;
405 else if (c
->io_weight
!= CGROUP_WEIGHT_INVALID
)
408 return CGROUP_WEIGHT_DEFAULT
;
411 static uint64_t cgroup_context_blkio_weight(CGroupContext
*c
, ManagerState state
) {
412 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
413 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
414 return c
->startup_blockio_weight
;
415 else if (c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
416 return c
->blockio_weight
;
418 return CGROUP_BLKIO_WEIGHT_DEFAULT
;
421 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight
) {
422 return CLAMP(blkio_weight
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_BLKIO_WEIGHT_DEFAULT
,
423 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
426 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight
) {
427 return CLAMP(io_weight
* CGROUP_BLKIO_WEIGHT_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
428 CGROUP_BLKIO_WEIGHT_MIN
, CGROUP_BLKIO_WEIGHT_MAX
);
431 static void cgroup_apply_io_device_weight(Unit
*u
, const char *dev_path
, uint64_t io_weight
) {
432 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
436 r
= lookup_block_device(dev_path
, &dev
);
440 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), io_weight
);
441 r
= cg_set_attribute("io", u
->cgroup_path
, "io.weight", buf
);
443 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
444 "Failed to set io.weight: %m");
447 static void cgroup_apply_blkio_device_weight(Unit
*u
, const char *dev_path
, uint64_t blkio_weight
) {
448 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
452 r
= lookup_block_device(dev_path
, &dev
);
456 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), blkio_weight
);
457 r
= cg_set_attribute("blkio", u
->cgroup_path
, "blkio.weight_device", buf
);
459 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
460 "Failed to set blkio.weight_device: %m");
463 static unsigned cgroup_apply_io_device_limit(Unit
*u
, const char *dev_path
, uint64_t *limits
) {
464 char limit_bufs
[_CGROUP_IO_LIMIT_TYPE_MAX
][DECIMAL_STR_MAX(uint64_t)];
465 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
466 CGroupIOLimitType type
;
471 r
= lookup_block_device(dev_path
, &dev
);
475 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++) {
476 if (limits
[type
] != cgroup_io_limit_defaults
[type
]) {
477 xsprintf(limit_bufs
[type
], "%" PRIu64
, limits
[type
]);
480 xsprintf(limit_bufs
[type
], "%s", limits
[type
] == CGROUP_LIMIT_MAX
? "max" : "0");
484 xsprintf(buf
, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev
), minor(dev
),
485 limit_bufs
[CGROUP_IO_RBPS_MAX
], limit_bufs
[CGROUP_IO_WBPS_MAX
],
486 limit_bufs
[CGROUP_IO_RIOPS_MAX
], limit_bufs
[CGROUP_IO_WIOPS_MAX
]);
487 r
= cg_set_attribute("io", u
->cgroup_path
, "io.max", buf
);
489 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
490 "Failed to set io.max: %m");
494 static unsigned cgroup_apply_blkio_device_limit(Unit
*u
, const char *dev_path
, uint64_t rbps
, uint64_t wbps
) {
495 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
500 r
= lookup_block_device(dev_path
, &dev
);
504 if (rbps
!= CGROUP_LIMIT_MAX
)
506 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), rbps
);
507 r
= cg_set_attribute("blkio", u
->cgroup_path
, "blkio.throttle.read_bps_device", buf
);
509 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
510 "Failed to set blkio.throttle.read_bps_device: %m");
512 if (wbps
!= CGROUP_LIMIT_MAX
)
514 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), wbps
);
515 r
= cg_set_attribute("blkio", u
->cgroup_path
, "blkio.throttle.write_bps_device", buf
);
517 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
518 "Failed to set blkio.throttle.write_bps_device: %m");
523 static bool cgroup_context_has_unified_memory_config(CGroupContext
*c
) {
524 return c
->memory_low
> 0 || c
->memory_high
!= CGROUP_LIMIT_MAX
|| c
->memory_max
!= CGROUP_LIMIT_MAX
;
527 static void cgroup_apply_unified_memory_limit(Unit
*u
, const char *file
, uint64_t v
) {
528 char buf
[DECIMAL_STR_MAX(uint64_t) + 1] = "max";
531 if (v
!= CGROUP_LIMIT_MAX
)
532 xsprintf(buf
, "%" PRIu64
"\n", v
);
534 r
= cg_set_attribute("memory", u
->cgroup_path
, file
, buf
);
536 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
537 "Failed to set %s: %m", file
);
540 static void cgroup_context_apply(Unit
*u
, CGroupMask mask
, ManagerState state
) {
548 c
= unit_get_cgroup_context(u
);
549 path
= u
->cgroup_path
;
557 /* Some cgroup attributes are not supported on the root cgroup,
558 * hence silently ignore */
559 is_root
= isempty(path
) || path_equal(path
, "/");
561 /* Make sure we don't try to display messages with an empty path. */
564 /* We generally ignore errors caused by read-only mounted
565 * cgroup trees (assuming we are running in a container then),
566 * and missing cgroups, i.e. EROFS and ENOENT. */
568 if ((mask
& CGROUP_MASK_CPU
) && !is_root
) {
569 char buf
[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t
)) + 1];
571 sprintf(buf
, "%" PRIu64
"\n",
572 IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) && c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
? c
->startup_cpu_shares
:
573 c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
? c
->cpu_shares
: CGROUP_CPU_SHARES_DEFAULT
);
574 r
= cg_set_attribute("cpu", path
, "cpu.shares", buf
);
576 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
577 "Failed to set cpu.shares: %m");
579 sprintf(buf
, USEC_FMT
"\n", CGROUP_CPU_QUOTA_PERIOD_USEC
);
580 r
= cg_set_attribute("cpu", path
, "cpu.cfs_period_us", buf
);
582 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
583 "Failed to set cpu.cfs_period_us: %m");
585 if (c
->cpu_quota_per_sec_usec
!= USEC_INFINITY
) {
586 sprintf(buf
, USEC_FMT
"\n", c
->cpu_quota_per_sec_usec
* CGROUP_CPU_QUOTA_PERIOD_USEC
/ USEC_PER_SEC
);
587 r
= cg_set_attribute("cpu", path
, "cpu.cfs_quota_us", buf
);
589 r
= cg_set_attribute("cpu", path
, "cpu.cfs_quota_us", "-1");
591 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
592 "Failed to set cpu.cfs_quota_us: %m");
595 if (mask
& CGROUP_MASK_IO
) {
596 bool has_io
= cgroup_context_has_io_config(c
);
597 bool has_blockio
= cgroup_context_has_blockio_config(c
);
600 char buf
[8+DECIMAL_STR_MAX(uint64_t)+1];
604 weight
= cgroup_context_io_weight(c
, state
);
605 else if (has_blockio
) {
606 uint64_t blkio_weight
= cgroup_context_blkio_weight(c
, state
);
608 weight
= cgroup_weight_blkio_to_io(blkio_weight
);
610 log_cgroup_compat(u
, "Applying [Startup]BlockIOWeight %" PRIu64
" as [Startup]IOWeight %" PRIu64
,
611 blkio_weight
, weight
);
613 weight
= CGROUP_WEIGHT_DEFAULT
;
615 xsprintf(buf
, "default %" PRIu64
"\n", weight
);
616 r
= cg_set_attribute("io", path
, "io.weight", buf
);
618 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
619 "Failed to set io.weight: %m");
622 CGroupIODeviceWeight
*w
;
624 /* FIXME: no way to reset this list */
625 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
)
626 cgroup_apply_io_device_weight(u
, w
->path
, w
->weight
);
627 } else if (has_blockio
) {
628 CGroupBlockIODeviceWeight
*w
;
630 /* FIXME: no way to reset this list */
631 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
) {
632 weight
= cgroup_weight_blkio_to_io(w
->weight
);
634 log_cgroup_compat(u
, "Applying BlockIODeviceWeight %" PRIu64
" as IODeviceWeight %" PRIu64
" for %s",
635 w
->weight
, weight
, w
->path
);
637 cgroup_apply_io_device_weight(u
, w
->path
, weight
);
642 /* Apply limits and free ones without config. */
644 CGroupIODeviceLimit
*l
, *next
;
646 LIST_FOREACH_SAFE(device_limits
, l
, next
, c
->io_device_limits
) {
647 if (!cgroup_apply_io_device_limit(u
, l
->path
, l
->limits
))
648 cgroup_context_free_io_device_limit(c
, l
);
650 } else if (has_blockio
) {
651 CGroupBlockIODeviceBandwidth
*b
, *next
;
653 LIST_FOREACH_SAFE(device_bandwidths
, b
, next
, c
->blockio_device_bandwidths
) {
654 uint64_t limits
[_CGROUP_IO_LIMIT_TYPE_MAX
];
655 CGroupIOLimitType type
;
657 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
658 limits
[type
] = cgroup_io_limit_defaults
[type
];
660 limits
[CGROUP_IO_RBPS_MAX
] = b
->rbps
;
661 limits
[CGROUP_IO_WBPS_MAX
] = b
->wbps
;
663 log_cgroup_compat(u
, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64
" %" PRIu64
" as IO{Read|Write}BandwidthMax for %s",
664 b
->rbps
, b
->wbps
, b
->path
);
666 if (!cgroup_apply_io_device_limit(u
, b
->path
, limits
))
667 cgroup_context_free_blockio_device_bandwidth(c
, b
);
672 if (mask
& CGROUP_MASK_BLKIO
) {
673 bool has_io
= cgroup_context_has_io_config(c
);
674 bool has_blockio
= cgroup_context_has_blockio_config(c
);
677 char buf
[DECIMAL_STR_MAX(uint64_t)+1];
681 weight
= cgroup_context_blkio_weight(c
, state
);
683 uint64_t io_weight
= cgroup_context_io_weight(c
, state
);
685 weight
= cgroup_weight_io_to_blkio(cgroup_context_io_weight(c
, state
));
687 log_cgroup_compat(u
, "Applying [Startup]IOWeight %" PRIu64
" as [Startup]BlockIOWeight %" PRIu64
,
690 weight
= CGROUP_BLKIO_WEIGHT_DEFAULT
;
692 xsprintf(buf
, "%" PRIu64
"\n", weight
);
693 r
= cg_set_attribute("blkio", path
, "blkio.weight", buf
);
695 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
696 "Failed to set blkio.weight: %m");
699 CGroupBlockIODeviceWeight
*w
;
701 /* FIXME: no way to reset this list */
702 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
703 cgroup_apply_blkio_device_weight(u
, w
->path
, w
->weight
);
705 CGroupIODeviceWeight
*w
;
707 /* FIXME: no way to reset this list */
708 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
) {
709 weight
= cgroup_weight_io_to_blkio(w
->weight
);
711 log_cgroup_compat(u
, "Applying IODeviceWeight %" PRIu64
" as BlockIODeviceWeight %" PRIu64
" for %s",
712 w
->weight
, weight
, w
->path
);
714 cgroup_apply_blkio_device_weight(u
, w
->path
, weight
);
719 /* Apply limits and free ones without config. */
721 CGroupBlockIODeviceBandwidth
*b
, *next
;
723 LIST_FOREACH_SAFE(device_bandwidths
, b
, next
, c
->blockio_device_bandwidths
) {
724 if (!cgroup_apply_blkio_device_limit(u
, b
->path
, b
->rbps
, b
->wbps
))
725 cgroup_context_free_blockio_device_bandwidth(c
, b
);
728 CGroupIODeviceLimit
*l
, *next
;
730 LIST_FOREACH_SAFE(device_limits
, l
, next
, c
->io_device_limits
) {
731 log_cgroup_compat(u
, "Applying IO{Read|Write}Bandwidth %" PRIu64
" %" PRIu64
" as BlockIO{Read|Write}BandwidthMax for %s",
732 l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
], l
->path
);
734 if (!cgroup_apply_blkio_device_limit(u
, l
->path
, l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
]))
735 cgroup_context_free_io_device_limit(c
, l
);
740 if ((mask
& CGROUP_MASK_MEMORY
) && !is_root
) {
741 if (cg_unified() > 0) {
742 uint64_t max
= c
->memory_max
;
744 if (cgroup_context_has_unified_memory_config(c
))
747 max
= c
->memory_limit
;
749 if (max
!= CGROUP_LIMIT_MAX
)
750 log_cgroup_compat(u
, "Applying MemoryLimit %" PRIu64
" as MemoryMax", max
);
753 cgroup_apply_unified_memory_limit(u
, "memory.low", c
->memory_low
);
754 cgroup_apply_unified_memory_limit(u
, "memory.high", c
->memory_high
);
755 cgroup_apply_unified_memory_limit(u
, "memory.max", max
);
757 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
758 uint64_t val
= c
->memory_limit
;
760 if (val
== CGROUP_LIMIT_MAX
) {
763 if (val
!= CGROUP_LIMIT_MAX
)
764 log_cgroup_compat(u
, "Applying MemoryMax %" PRIi64
" as MemoryLimit", c
->memory_max
);
767 if (val
== CGROUP_LIMIT_MAX
)
768 strncpy(buf
, "-1\n", sizeof(buf
));
770 xsprintf(buf
, "%" PRIu64
"\n", val
);
772 r
= cg_set_attribute("memory", path
, "memory.limit_in_bytes", buf
);
774 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
775 "Failed to set memory.limit_in_bytes: %m");
779 if ((mask
& CGROUP_MASK_DEVICES
) && !is_root
) {
780 CGroupDeviceAllow
*a
;
782 /* Changing the devices list of a populated cgroup
783 * might result in EINVAL, hence ignore EINVAL
786 if (c
->device_allow
|| c
->device_policy
!= CGROUP_AUTO
)
787 r
= cg_set_attribute("devices", path
, "devices.deny", "a");
789 r
= cg_set_attribute("devices", path
, "devices.allow", "a");
791 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
792 "Failed to reset devices.list: %m");
794 if (c
->device_policy
== CGROUP_CLOSED
||
795 (c
->device_policy
== CGROUP_AUTO
&& c
->device_allow
)) {
796 static const char auto_devices
[] =
797 "/dev/null\0" "rwm\0"
798 "/dev/zero\0" "rwm\0"
799 "/dev/full\0" "rwm\0"
800 "/dev/random\0" "rwm\0"
801 "/dev/urandom\0" "rwm\0"
803 "/dev/pts/ptmx\0" "rw\0" /* /dev/pts/ptmx may not be duplicated, but accessed */
804 /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */
805 "/run/systemd/inaccessible/chr\0" "rwm\0"
806 "/run/systemd/inaccessible/blk\0" "rwm\0";
810 NULSTR_FOREACH_PAIR(x
, y
, auto_devices
)
811 whitelist_device(path
, x
, y
);
813 whitelist_major(path
, "pts", 'c', "rw");
814 whitelist_major(path
, "kdbus", 'c', "rw");
815 whitelist_major(path
, "kdbus/*", 'c', "rw");
818 LIST_FOREACH(device_allow
, a
, c
->device_allow
) {
834 if (startswith(a
->path
, "/dev/"))
835 whitelist_device(path
, a
->path
, acc
);
836 else if (startswith(a
->path
, "block-"))
837 whitelist_major(path
, a
->path
+ 6, 'b', acc
);
838 else if (startswith(a
->path
, "char-"))
839 whitelist_major(path
, a
->path
+ 5, 'c', acc
);
841 log_unit_debug(u
, "Ignoring device %s while writing cgroup attribute.", a
->path
);
845 if ((mask
& CGROUP_MASK_PIDS
) && !is_root
) {
847 if (c
->tasks_max
!= (uint64_t) -1) {
848 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
850 sprintf(buf
, "%" PRIu64
"\n", c
->tasks_max
);
851 r
= cg_set_attribute("pids", path
, "pids.max", buf
);
853 r
= cg_set_attribute("pids", path
, "pids.max", "max");
856 log_unit_full(u
, IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
857 "Failed to set pids.max: %m");
861 CGroupMask
cgroup_context_get_mask(CGroupContext
*c
) {
864 /* Figure out which controllers we need */
866 if (c
->cpu_accounting
||
867 c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
868 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
869 c
->cpu_quota_per_sec_usec
!= USEC_INFINITY
)
870 mask
|= CGROUP_MASK_CPUACCT
| CGROUP_MASK_CPU
;
872 if (cgroup_context_has_io_config(c
) || cgroup_context_has_blockio_config(c
))
873 mask
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
875 if (c
->memory_accounting
||
876 c
->memory_limit
!= CGROUP_LIMIT_MAX
||
877 cgroup_context_has_unified_memory_config(c
))
878 mask
|= CGROUP_MASK_MEMORY
;
880 if (c
->device_allow
||
881 c
->device_policy
!= CGROUP_AUTO
)
882 mask
|= CGROUP_MASK_DEVICES
;
884 if (c
->tasks_accounting
||
885 c
->tasks_max
!= (uint64_t) -1)
886 mask
|= CGROUP_MASK_PIDS
;
891 CGroupMask
unit_get_own_mask(Unit
*u
) {
894 /* Returns the mask of controllers the unit needs for itself */
896 c
= unit_get_cgroup_context(u
);
900 /* If delegation is turned on, then turn on all cgroups,
901 * unless we are on the legacy hierarchy and the process we
902 * fork into it is known to drop privileges, and hence
903 * shouldn't get access to the controllers.
905 * Note that on the unified hierarchy it is safe to delegate
906 * controllers to unprivileged services. */
911 e
= unit_get_exec_context(u
);
913 exec_context_maintains_privileges(e
) ||
915 return _CGROUP_MASK_ALL
;
918 return cgroup_context_get_mask(c
);
921 CGroupMask
unit_get_members_mask(Unit
*u
) {
924 /* Returns the mask of controllers all of the unit's children
927 if (u
->cgroup_members_mask_valid
)
928 return u
->cgroup_members_mask
;
930 u
->cgroup_members_mask
= 0;
932 if (u
->type
== UNIT_SLICE
) {
936 SET_FOREACH(member
, u
->dependencies
[UNIT_BEFORE
], i
) {
941 if (UNIT_DEREF(member
->slice
) != u
)
944 u
->cgroup_members_mask
|=
945 unit_get_own_mask(member
) |
946 unit_get_members_mask(member
);
950 u
->cgroup_members_mask_valid
= true;
951 return u
->cgroup_members_mask
;
954 CGroupMask
unit_get_siblings_mask(Unit
*u
) {
957 /* Returns the mask of controllers all of the unit's siblings
958 * require, i.e. the members mask of the unit's parent slice
959 * if there is one. */
961 if (UNIT_ISSET(u
->slice
))
962 return unit_get_members_mask(UNIT_DEREF(u
->slice
));
964 return unit_get_own_mask(u
) | unit_get_members_mask(u
);
967 CGroupMask
unit_get_subtree_mask(Unit
*u
) {
969 /* Returns the mask of this subtree, meaning of the group
970 * itself and its children. */
972 return unit_get_own_mask(u
) | unit_get_members_mask(u
);
975 CGroupMask
unit_get_target_mask(Unit
*u
) {
978 /* This returns the cgroup mask of all controllers to enable
979 * for a specific cgroup, i.e. everything it needs itself,
980 * plus all that its children need, plus all that its siblings
981 * need. This is primarily useful on the legacy cgroup
982 * hierarchy, where we need to duplicate each cgroup in each
983 * hierarchy that shall be enabled for it. */
985 mask
= unit_get_own_mask(u
) | unit_get_members_mask(u
) | unit_get_siblings_mask(u
);
986 mask
&= u
->manager
->cgroup_supported
;
991 CGroupMask
unit_get_enable_mask(Unit
*u
) {
994 /* This returns the cgroup mask of all controllers to enable
995 * for the children of a specific cgroup. This is primarily
996 * useful for the unified cgroup hierarchy, where each cgroup
997 * controls which controllers are enabled for its children. */
999 mask
= unit_get_members_mask(u
);
1000 mask
&= u
->manager
->cgroup_supported
;
1005 /* Recurse from a unit up through its containing slices, propagating
1006 * mask bits upward. A unit is also member of itself. */
1007 void unit_update_cgroup_members_masks(Unit
*u
) {
1013 /* Calculate subtree mask */
1014 m
= unit_get_subtree_mask(u
);
1016 /* See if anything changed from the previous invocation. If
1017 * not, we're done. */
1018 if (u
->cgroup_subtree_mask_valid
&& m
== u
->cgroup_subtree_mask
)
1022 u
->cgroup_subtree_mask_valid
&&
1023 ((m
& ~u
->cgroup_subtree_mask
) != 0) &&
1024 ((~m
& u
->cgroup_subtree_mask
) == 0);
1026 u
->cgroup_subtree_mask
= m
;
1027 u
->cgroup_subtree_mask_valid
= true;
1029 if (UNIT_ISSET(u
->slice
)) {
1030 Unit
*s
= UNIT_DEREF(u
->slice
);
1033 /* There's more set now than before. We
1034 * propagate the new mask to the parent's mask
1035 * (not caring if it actually was valid or
1038 s
->cgroup_members_mask
|= m
;
1041 /* There's less set now than before (or we
1042 * don't know), we need to recalculate
1043 * everything, so let's invalidate the
1044 * parent's members mask */
1046 s
->cgroup_members_mask_valid
= false;
1048 /* And now make sure that this change also hits our
1050 unit_update_cgroup_members_masks(s
);
1054 static const char *migrate_callback(CGroupMask mask
, void *userdata
) {
1061 if (u
->cgroup_path
&&
1062 u
->cgroup_realized
&&
1063 (u
->cgroup_realized_mask
& mask
) == mask
)
1064 return u
->cgroup_path
;
1066 u
= UNIT_DEREF(u
->slice
);
1072 char *unit_default_cgroup_path(Unit
*u
) {
1073 _cleanup_free_
char *escaped
= NULL
, *slice
= NULL
;
1078 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1079 return strdup(u
->manager
->cgroup_root
);
1081 if (UNIT_ISSET(u
->slice
) && !unit_has_name(UNIT_DEREF(u
->slice
), SPECIAL_ROOT_SLICE
)) {
1082 r
= cg_slice_to_path(UNIT_DEREF(u
->slice
)->id
, &slice
);
1087 escaped
= cg_escape(u
->id
);
1092 return strjoin(u
->manager
->cgroup_root
, "/", slice
, "/", escaped
, NULL
);
1094 return strjoin(u
->manager
->cgroup_root
, "/", escaped
, NULL
);
1097 int unit_set_cgroup_path(Unit
*u
, const char *path
) {
1098 _cleanup_free_
char *p
= NULL
;
1110 if (streq_ptr(u
->cgroup_path
, p
))
1114 r
= hashmap_put(u
->manager
->cgroup_unit
, p
, u
);
1119 unit_release_cgroup(u
);
1127 int unit_watch_cgroup(Unit
*u
) {
1128 _cleanup_free_
char *events
= NULL
;
1133 if (!u
->cgroup_path
)
1136 if (u
->cgroup_inotify_wd
>= 0)
1139 /* Only applies to the unified hierarchy */
1142 return log_unit_error_errno(u
, r
, "Failed detect whether the unified hierarchy is used: %m");
1146 /* Don't watch the root slice, it's pointless. */
1147 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1150 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_inotify_wd_unit
, &trivial_hash_ops
);
1154 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events", &events
);
1158 u
->cgroup_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1159 if (u
->cgroup_inotify_wd
< 0) {
1161 if (errno
== ENOENT
) /* If the directory is already
1162 * gone we don't need to track
1163 * it, so this is not an error */
1166 return log_unit_error_errno(u
, errno
, "Failed to add inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1169 r
= hashmap_put(u
->manager
->cgroup_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_inotify_wd
), u
);
1171 return log_unit_error_errno(u
, r
, "Failed to add inotify watch descriptor to hash map: %m");
1176 static int unit_create_cgroup(
1178 CGroupMask target_mask
,
1179 CGroupMask enable_mask
) {
1186 c
= unit_get_cgroup_context(u
);
1190 if (!u
->cgroup_path
) {
1191 _cleanup_free_
char *path
= NULL
;
1193 path
= unit_default_cgroup_path(u
);
1197 r
= unit_set_cgroup_path(u
, path
);
1199 return log_unit_error_errno(u
, r
, "Control group %s exists already.", path
);
1201 return log_unit_error_errno(u
, r
, "Failed to set unit's control group path to %s: %m", path
);
1204 /* First, create our own group */
1205 r
= cg_create_everywhere(u
->manager
->cgroup_supported
, target_mask
, u
->cgroup_path
);
1207 return log_unit_error_errno(u
, r
, "Failed to create cgroup %s: %m", u
->cgroup_path
);
1209 /* Start watching it */
1210 (void) unit_watch_cgroup(u
);
1212 /* Enable all controllers we need */
1213 r
= cg_enable_everywhere(u
->manager
->cgroup_supported
, enable_mask
, u
->cgroup_path
);
1215 log_unit_warning_errno(u
, r
, "Failed to enable controllers on cgroup %s, ignoring: %m", u
->cgroup_path
);
1217 /* Keep track that this is now realized */
1218 u
->cgroup_realized
= true;
1219 u
->cgroup_realized_mask
= target_mask
;
1220 u
->cgroup_enabled_mask
= enable_mask
;
1222 if (u
->type
!= UNIT_SLICE
&& !c
->delegate
) {
1224 /* Then, possibly move things over, but not if
1225 * subgroups may contain processes, which is the case
1226 * for slice and delegation units. */
1227 r
= cg_migrate_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, u
->cgroup_path
, migrate_callback
, u
);
1229 log_unit_warning_errno(u
, r
, "Failed to migrate cgroup from to %s, ignoring: %m", u
->cgroup_path
);
1235 int unit_attach_pids_to_cgroup(Unit
*u
) {
1239 r
= unit_realize_cgroup(u
);
1243 r
= cg_attach_many_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, u
->pids
, migrate_callback
, u
);
1250 static bool unit_has_mask_realized(Unit
*u
, CGroupMask target_mask
, CGroupMask enable_mask
) {
1253 return u
->cgroup_realized
&& u
->cgroup_realized_mask
== target_mask
&& u
->cgroup_enabled_mask
== enable_mask
;
1256 /* Check if necessary controllers and attributes for a unit are in place.
1258 * If so, do nothing.
1259 * If not, create paths, move processes over, and set attributes.
1261 * Returns 0 on success and < 0 on failure. */
1262 static int unit_realize_cgroup_now(Unit
*u
, ManagerState state
) {
1263 CGroupMask target_mask
, enable_mask
;
1268 if (u
->in_cgroup_queue
) {
1269 LIST_REMOVE(cgroup_queue
, u
->manager
->cgroup_queue
, u
);
1270 u
->in_cgroup_queue
= false;
1273 target_mask
= unit_get_target_mask(u
);
1274 enable_mask
= unit_get_enable_mask(u
);
1276 if (unit_has_mask_realized(u
, target_mask
, enable_mask
))
1279 /* First, realize parents */
1280 if (UNIT_ISSET(u
->slice
)) {
1281 r
= unit_realize_cgroup_now(UNIT_DEREF(u
->slice
), state
);
1286 /* And then do the real work */
1287 r
= unit_create_cgroup(u
, target_mask
, enable_mask
);
1291 /* Finally, apply the necessary attributes. */
1292 cgroup_context_apply(u
, target_mask
, state
);
1297 static void unit_add_to_cgroup_queue(Unit
*u
) {
1299 if (u
->in_cgroup_queue
)
1302 LIST_PREPEND(cgroup_queue
, u
->manager
->cgroup_queue
, u
);
1303 u
->in_cgroup_queue
= true;
1306 unsigned manager_dispatch_cgroup_queue(Manager
*m
) {
1312 state
= manager_state(m
);
1314 while ((i
= m
->cgroup_queue
)) {
1315 assert(i
->in_cgroup_queue
);
1317 r
= unit_realize_cgroup_now(i
, state
);
1319 log_warning_errno(r
, "Failed to realize cgroups for queued unit %s, ignoring: %m", i
->id
);
1327 static void unit_queue_siblings(Unit
*u
) {
1330 /* This adds the siblings of the specified unit and the
1331 * siblings of all parent units to the cgroup queue. (But
1332 * neither the specified unit itself nor the parents.) */
1334 while ((slice
= UNIT_DEREF(u
->slice
))) {
1338 SET_FOREACH(m
, slice
->dependencies
[UNIT_BEFORE
], i
) {
1342 /* Skip units that have a dependency on the slice
1343 * but aren't actually in it. */
1344 if (UNIT_DEREF(m
->slice
) != slice
)
1347 /* No point in doing cgroup application for units
1348 * without active processes. */
1349 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m
)))
1352 /* If the unit doesn't need any new controllers
1353 * and has current ones realized, it doesn't need
1355 if (unit_has_mask_realized(m
, unit_get_target_mask(m
), unit_get_enable_mask(m
)))
1358 unit_add_to_cgroup_queue(m
);
1365 int unit_realize_cgroup(Unit
*u
) {
1368 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1371 /* So, here's the deal: when realizing the cgroups for this
1372 * unit, we need to first create all parents, but there's more
1373 * actually: for the weight-based controllers we also need to
1374 * make sure that all our siblings (i.e. units that are in the
1375 * same slice as we are) have cgroups, too. Otherwise, things
1376 * would become very uneven as each of their processes would
1377 * get as much resources as all our group together. This call
1378 * will synchronously create the parent cgroups, but will
1379 * defer work on the siblings to the next event loop
1382 /* Add all sibling slices to the cgroup queue. */
1383 unit_queue_siblings(u
);
1385 /* And realize this one now (and apply the values) */
1386 return unit_realize_cgroup_now(u
, manager_state(u
->manager
));
1389 void unit_release_cgroup(Unit
*u
) {
1392 /* Forgets all cgroup details for this cgroup */
1394 if (u
->cgroup_path
) {
1395 (void) hashmap_remove(u
->manager
->cgroup_unit
, u
->cgroup_path
);
1396 u
->cgroup_path
= mfree(u
->cgroup_path
);
1399 if (u
->cgroup_inotify_wd
>= 0) {
1400 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_inotify_wd
) < 0)
1401 log_unit_debug_errno(u
, errno
, "Failed to remove cgroup inotify watch %i for %s, ignoring", u
->cgroup_inotify_wd
, u
->id
);
1403 (void) hashmap_remove(u
->manager
->cgroup_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_inotify_wd
));
1404 u
->cgroup_inotify_wd
= -1;
1408 void unit_prune_cgroup(Unit
*u
) {
1414 /* Removes the cgroup, if empty and possible, and stops watching it. */
1416 if (!u
->cgroup_path
)
1419 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
1421 r
= cg_trim_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, !is_root_slice
);
1423 log_unit_debug_errno(u
, r
, "Failed to destroy cgroup %s, ignoring: %m", u
->cgroup_path
);
1430 unit_release_cgroup(u
);
1432 u
->cgroup_realized
= false;
1433 u
->cgroup_realized_mask
= 0;
1434 u
->cgroup_enabled_mask
= 0;
1437 int unit_search_main_pid(Unit
*u
, pid_t
*ret
) {
1438 _cleanup_fclose_
FILE *f
= NULL
;
1439 pid_t pid
= 0, npid
, mypid
;
1445 if (!u
->cgroup_path
)
1448 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, &f
);
1453 while (cg_read_pid(f
, &npid
) > 0) {
1459 /* Ignore processes that aren't our kids */
1460 if (get_process_ppid(npid
, &ppid
) >= 0 && ppid
!= mypid
)
1464 /* Dang, there's more than one daemonized PID
1465 in this group, so we don't know what process
1466 is the main process. */
1477 static int unit_watch_pids_in_path(Unit
*u
, const char *path
) {
1478 _cleanup_closedir_
DIR *d
= NULL
;
1479 _cleanup_fclose_
FILE *f
= NULL
;
1485 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, path
, &f
);
1491 while ((r
= cg_read_pid(f
, &pid
)) > 0) {
1492 r
= unit_watch_pid(u
, pid
);
1493 if (r
< 0 && ret
>= 0)
1497 if (r
< 0 && ret
>= 0)
1501 r
= cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER
, path
, &d
);
1508 while ((r
= cg_read_subgroup(d
, &fn
)) > 0) {
1509 _cleanup_free_
char *p
= NULL
;
1511 p
= strjoin(path
, "/", fn
, NULL
);
1517 r
= unit_watch_pids_in_path(u
, p
);
1518 if (r
< 0 && ret
>= 0)
1522 if (r
< 0 && ret
>= 0)
1529 int unit_watch_all_pids(Unit
*u
) {
1532 /* Adds all PIDs from our cgroup to the set of PIDs we
1533 * watch. This is a fallback logic for cases where we do not
1534 * get reliable cgroup empty notifications: we try to use
1535 * SIGCHLD as replacement. */
1537 if (!u
->cgroup_path
)
1540 if (cg_unified() > 0) /* On unified we can use proper notifications */
1543 return unit_watch_pids_in_path(u
, u
->cgroup_path
);
1546 int unit_notify_cgroup_empty(Unit
*u
) {
1551 if (!u
->cgroup_path
)
1554 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
1558 unit_add_to_gc_queue(u
);
1560 if (UNIT_VTABLE(u
)->notify_cgroup_empty
)
1561 UNIT_VTABLE(u
)->notify_cgroup_empty(u
);
1566 static int on_cgroup_inotify_event(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
1567 Manager
*m
= userdata
;
1574 union inotify_event_buffer buffer
;
1575 struct inotify_event
*e
;
1578 l
= read(fd
, &buffer
, sizeof(buffer
));
1580 if (errno
== EINTR
|| errno
== EAGAIN
)
1583 return log_error_errno(errno
, "Failed to read control group inotify events: %m");
1586 FOREACH_INOTIFY_EVENT(e
, buffer
, l
) {
1590 /* Queue overflow has no watch descriptor */
1593 if (e
->mask
& IN_IGNORED
)
1594 /* The watch was just removed */
1597 u
= hashmap_get(m
->cgroup_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
1598 if (!u
) /* Not that inotify might deliver
1599 * events for a watch even after it
1600 * was removed, because it was queued
1601 * before the removal. Let's ignore
1602 * this here safely. */
1605 (void) unit_notify_cgroup_empty(u
);
1610 int manager_setup_cgroup(Manager
*m
) {
1611 _cleanup_free_
char *path
= NULL
;
1618 /* 1. Determine hierarchy */
1619 m
->cgroup_root
= mfree(m
->cgroup_root
);
1620 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 0, &m
->cgroup_root
);
1622 return log_error_errno(r
, "Cannot determine cgroup we are running in: %m");
1624 /* Chop off the init scope, if we are already located in it */
1625 e
= endswith(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
1627 /* LEGACY: Also chop off the system slice if we are in
1628 * it. This is to support live upgrades from older systemd
1629 * versions where PID 1 was moved there. Also see
1630 * cg_get_root_path(). */
1631 if (!e
&& MANAGER_IS_SYSTEM(m
)) {
1632 e
= endswith(m
->cgroup_root
, "/" SPECIAL_SYSTEM_SLICE
);
1634 e
= endswith(m
->cgroup_root
, "/system"); /* even more legacy */
1639 /* And make sure to store away the root value without trailing
1640 * slash, even for the root dir, so that we can easily prepend
1642 while ((e
= endswith(m
->cgroup_root
, "/")))
1646 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, NULL
, &path
);
1648 return log_error_errno(r
, "Cannot find cgroup mount point: %m");
1650 unified
= cg_unified();
1652 return log_error_errno(r
, "Couldn't determine if we are running in the unified hierarchy: %m");
1654 log_debug("Unified cgroup hierarchy is located at %s.", path
);
1656 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER
". File system hierarchy is at %s.", path
);
1659 const char *scope_path
;
1661 /* 3. Install agent */
1664 /* In the unified hierarchy we can get
1665 * cgroup empty notifications via inotify. */
1667 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
1668 safe_close(m
->cgroup_inotify_fd
);
1670 m
->cgroup_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
1671 if (m
->cgroup_inotify_fd
< 0)
1672 return log_error_errno(errno
, "Failed to create control group inotify object: %m");
1674 r
= sd_event_add_io(m
->event
, &m
->cgroup_inotify_event_source
, m
->cgroup_inotify_fd
, EPOLLIN
, on_cgroup_inotify_event
, m
);
1676 return log_error_errno(r
, "Failed to watch control group inotify object: %m");
1678 /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also
1679 * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
1680 r
= sd_event_source_set_priority(m
->cgroup_inotify_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
1682 return log_error_errno(r
, "Failed to set priority of inotify event source: %m");
1684 (void) sd_event_source_set_description(m
->cgroup_inotify_event_source
, "cgroup-inotify");
1686 } else if (MANAGER_IS_SYSTEM(m
)) {
1688 /* On the legacy hierarchy we only get
1689 * notifications via cgroup agents. (Which
1690 * isn't really reliable, since it does not
1691 * generate events when control groups with
1692 * children run empty. */
1694 r
= cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER
, SYSTEMD_CGROUP_AGENT_PATH
);
1696 log_warning_errno(r
, "Failed to install release agent, ignoring: %m");
1698 log_debug("Installed release agent.");
1700 log_debug("Release agent already installed.");
1703 /* 4. Make sure we are in the special "init.scope" unit in the root slice. */
1704 scope_path
= strjoina(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
1705 r
= cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
1707 return log_error_errno(r
, "Failed to create %s control group: %m", scope_path
);
1709 /* also, move all other userspace processes remaining
1710 * in the root cgroup into that scope. */
1711 r
= cg_migrate(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
1713 log_warning_errno(r
, "Couldn't move remaining userspace processes, ignoring: %m");
1715 /* 5. And pin it, so that it cannot be unmounted */
1716 safe_close(m
->pin_cgroupfs_fd
);
1717 m
->pin_cgroupfs_fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOCTTY
|O_NONBLOCK
);
1718 if (m
->pin_cgroupfs_fd
< 0)
1719 return log_error_errno(errno
, "Failed to open pin file: %m");
1721 /* 6. Always enable hierarchical support if it exists... */
1723 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
1726 /* 7. Figure out which controllers are supported */
1727 r
= cg_mask_supported(&m
->cgroup_supported
);
1729 return log_error_errno(r
, "Failed to determine supported controllers: %m");
1731 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++)
1732 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c
), yes_no(m
->cgroup_supported
& CGROUP_CONTROLLER_TO_MASK(c
)));
1737 void manager_shutdown_cgroup(Manager
*m
, bool delete) {
1740 /* We can't really delete the group, since we are in it. But
1742 if (delete && m
->cgroup_root
)
1743 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, false);
1745 m
->cgroup_inotify_wd_unit
= hashmap_free(m
->cgroup_inotify_wd_unit
);
1747 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
1748 m
->cgroup_inotify_fd
= safe_close(m
->cgroup_inotify_fd
);
1750 m
->pin_cgroupfs_fd
= safe_close(m
->pin_cgroupfs_fd
);
1752 m
->cgroup_root
= mfree(m
->cgroup_root
);
1755 Unit
* manager_get_unit_by_cgroup(Manager
*m
, const char *cgroup
) {
1762 u
= hashmap_get(m
->cgroup_unit
, cgroup
);
1766 p
= strdupa(cgroup
);
1770 e
= strrchr(p
, '/');
1772 return hashmap_get(m
->cgroup_unit
, SPECIAL_ROOT_SLICE
);
1776 u
= hashmap_get(m
->cgroup_unit
, p
);
1782 Unit
*manager_get_unit_by_pid_cgroup(Manager
*m
, pid_t pid
) {
1783 _cleanup_free_
char *cgroup
= NULL
;
1791 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, pid
, &cgroup
);
1795 return manager_get_unit_by_cgroup(m
, cgroup
);
1798 Unit
*manager_get_unit_by_pid(Manager
*m
, pid_t pid
) {
1807 return hashmap_get(m
->units
, SPECIAL_INIT_SCOPE
);
1809 u
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(pid
));
1813 u
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(pid
));
1817 return manager_get_unit_by_pid_cgroup(m
, pid
);
1820 int manager_notify_cgroup_empty(Manager
*m
, const char *cgroup
) {
1826 log_debug("Got cgroup empty notification for: %s", cgroup
);
1828 u
= manager_get_unit_by_cgroup(m
, cgroup
);
1832 return unit_notify_cgroup_empty(u
);
1835 int unit_get_memory_current(Unit
*u
, uint64_t *ret
) {
1836 _cleanup_free_
char *v
= NULL
;
1842 if (!u
->cgroup_path
)
1845 if ((u
->cgroup_realized_mask
& CGROUP_MASK_MEMORY
) == 0)
1848 if (cg_unified() <= 0)
1849 r
= cg_get_attribute("memory", u
->cgroup_path
, "memory.usage_in_bytes", &v
);
1851 r
= cg_get_attribute("memory", u
->cgroup_path
, "memory.current", &v
);
1857 return safe_atou64(v
, ret
);
1860 int unit_get_tasks_current(Unit
*u
, uint64_t *ret
) {
1861 _cleanup_free_
char *v
= NULL
;
1867 if (!u
->cgroup_path
)
1870 if ((u
->cgroup_realized_mask
& CGROUP_MASK_PIDS
) == 0)
1873 r
= cg_get_attribute("pids", u
->cgroup_path
, "pids.current", &v
);
1879 return safe_atou64(v
, ret
);
1882 static int unit_get_cpu_usage_raw(Unit
*u
, nsec_t
*ret
) {
1883 _cleanup_free_
char *v
= NULL
;
1890 if (!u
->cgroup_path
)
1893 if ((u
->cgroup_realized_mask
& CGROUP_MASK_CPUACCT
) == 0)
1896 r
= cg_get_attribute("cpuacct", u
->cgroup_path
, "cpuacct.usage", &v
);
1902 r
= safe_atou64(v
, &ns
);
1910 int unit_get_cpu_usage(Unit
*u
, nsec_t
*ret
) {
1914 r
= unit_get_cpu_usage_raw(u
, &ns
);
1918 if (ns
> u
->cpuacct_usage_base
)
1919 ns
-= u
->cpuacct_usage_base
;
1927 int unit_reset_cpu_usage(Unit
*u
) {
1933 r
= unit_get_cpu_usage_raw(u
, &ns
);
1935 u
->cpuacct_usage_base
= 0;
1939 u
->cpuacct_usage_base
= ns
;
1943 bool unit_cgroup_delegate(Unit
*u
) {
1948 c
= unit_get_cgroup_context(u
);
1955 void unit_invalidate_cgroup(Unit
*u
, CGroupMask m
) {
1958 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1964 /* always invalidate compat pairs together */
1965 if (m
& (CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
))
1966 m
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
1968 if ((u
->cgroup_realized_mask
& m
) == 0)
1971 u
->cgroup_realized_mask
&= ~m
;
1972 unit_add_to_cgroup_queue(u
);
1975 void manager_invalidate_startup_units(Manager
*m
) {
1981 SET_FOREACH(u
, m
->startup_units
, i
)
1982 unit_invalidate_cgroup(u
, CGROUP_MASK_CPU
|CGROUP_MASK_IO
|CGROUP_MASK_BLKIO
);
1985 static const char* const cgroup_device_policy_table
[_CGROUP_DEVICE_POLICY_MAX
] = {
1986 [CGROUP_AUTO
] = "auto",
1987 [CGROUP_CLOSED
] = "closed",
1988 [CGROUP_STRICT
] = "strict",
1991 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy
, CGroupDevicePolicy
);