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 void cgroup_context_init(CGroupContext
*c
) {
42 /* Initialize everything to the kernel defaults, assuming the
43 * structure is preinitialized to 0 */
45 c
->cpu_shares
= CGROUP_CPU_SHARES_INVALID
;
46 c
->startup_cpu_shares
= CGROUP_CPU_SHARES_INVALID
;
47 c
->cpu_quota_per_sec_usec
= USEC_INFINITY
;
49 c
->memory_high
= CGROUP_LIMIT_MAX
;
50 c
->memory_max
= CGROUP_LIMIT_MAX
;
52 c
->memory_limit
= CGROUP_LIMIT_MAX
;
54 c
->io_weight
= CGROUP_WEIGHT_INVALID
;
55 c
->startup_io_weight
= CGROUP_WEIGHT_INVALID
;
57 c
->blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
;
58 c
->startup_blockio_weight
= CGROUP_BLKIO_WEIGHT_INVALID
;
60 c
->tasks_max
= (uint64_t) -1;
63 void cgroup_context_free_device_allow(CGroupContext
*c
, CGroupDeviceAllow
*a
) {
67 LIST_REMOVE(device_allow
, c
->device_allow
, a
);
72 void cgroup_context_free_io_device_weight(CGroupContext
*c
, CGroupIODeviceWeight
*w
) {
76 LIST_REMOVE(device_weights
, c
->io_device_weights
, w
);
81 void cgroup_context_free_io_device_limit(CGroupContext
*c
, CGroupIODeviceLimit
*l
) {
85 LIST_REMOVE(device_limits
, c
->io_device_limits
, l
);
90 void cgroup_context_free_blockio_device_weight(CGroupContext
*c
, CGroupBlockIODeviceWeight
*w
) {
94 LIST_REMOVE(device_weights
, c
->blockio_device_weights
, w
);
99 void cgroup_context_free_blockio_device_bandwidth(CGroupContext
*c
, CGroupBlockIODeviceBandwidth
*b
) {
103 LIST_REMOVE(device_bandwidths
, c
->blockio_device_bandwidths
, b
);
108 void cgroup_context_done(CGroupContext
*c
) {
111 while (c
->io_device_weights
)
112 cgroup_context_free_io_device_weight(c
, c
->io_device_weights
);
114 while (c
->io_device_limits
)
115 cgroup_context_free_io_device_limit(c
, c
->io_device_limits
);
117 while (c
->blockio_device_weights
)
118 cgroup_context_free_blockio_device_weight(c
, c
->blockio_device_weights
);
120 while (c
->blockio_device_bandwidths
)
121 cgroup_context_free_blockio_device_bandwidth(c
, c
->blockio_device_bandwidths
);
123 while (c
->device_allow
)
124 cgroup_context_free_device_allow(c
, c
->device_allow
);
127 void cgroup_context_dump(CGroupContext
*c
, FILE* f
, const char *prefix
) {
128 CGroupIODeviceLimit
*il
;
129 CGroupIODeviceWeight
*iw
;
130 CGroupBlockIODeviceBandwidth
*b
;
131 CGroupBlockIODeviceWeight
*w
;
132 CGroupDeviceAllow
*a
;
133 char u
[FORMAT_TIMESPAN_MAX
];
138 prefix
= strempty(prefix
);
141 "%sCPUAccounting=%s\n"
142 "%sIOAccounting=%s\n"
143 "%sBlockIOAccounting=%s\n"
144 "%sMemoryAccounting=%s\n"
145 "%sTasksAccounting=%s\n"
146 "%sCPUShares=%" PRIu64
"\n"
147 "%sStartupCPUShares=%" PRIu64
"\n"
148 "%sCPUQuotaPerSecSec=%s\n"
149 "%sIOWeight=%" PRIu64
"\n"
150 "%sStartupIOWeight=%" PRIu64
"\n"
151 "%sBlockIOWeight=%" PRIu64
"\n"
152 "%sStartupBlockIOWeight=%" PRIu64
"\n"
153 "%sMemoryLow=%" PRIu64
"\n"
154 "%sMemoryHigh=%" PRIu64
"\n"
155 "%sMemoryMax=%" PRIu64
"\n"
156 "%sMemoryLimit=%" PRIu64
"\n"
157 "%sTasksMax=%" PRIu64
"\n"
158 "%sDevicePolicy=%s\n"
160 prefix
, yes_no(c
->cpu_accounting
),
161 prefix
, yes_no(c
->io_accounting
),
162 prefix
, yes_no(c
->blockio_accounting
),
163 prefix
, yes_no(c
->memory_accounting
),
164 prefix
, yes_no(c
->tasks_accounting
),
165 prefix
, c
->cpu_shares
,
166 prefix
, c
->startup_cpu_shares
,
167 prefix
, format_timespan(u
, sizeof(u
), c
->cpu_quota_per_sec_usec
, 1),
168 prefix
, c
->io_weight
,
169 prefix
, c
->startup_io_weight
,
170 prefix
, c
->blockio_weight
,
171 prefix
, c
->startup_blockio_weight
,
172 prefix
, c
->memory_low
,
173 prefix
, c
->memory_high
,
174 prefix
, c
->memory_max
,
175 prefix
, c
->memory_limit
,
176 prefix
, c
->tasks_max
,
177 prefix
, cgroup_device_policy_to_string(c
->device_policy
),
178 prefix
, yes_no(c
->delegate
));
180 LIST_FOREACH(device_allow
, a
, c
->device_allow
)
182 "%sDeviceAllow=%s %s%s%s\n",
185 a
->r
? "r" : "", a
->w
? "w" : "", a
->m
? "m" : "");
187 LIST_FOREACH(device_weights
, iw
, c
->io_device_weights
)
189 "%sIODeviceWeight=%s %" PRIu64
,
194 LIST_FOREACH(device_limits
, il
, c
->io_device_limits
) {
195 char buf
[FORMAT_BYTES_MAX
];
196 CGroupIOLimitType type
;
198 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
199 if (il
->limits
[type
] != cgroup_io_limit_defaults
[type
])
203 cgroup_io_limit_type_to_string(type
),
205 format_bytes(buf
, sizeof(buf
), il
->limits
[type
]));
208 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
210 "%sBlockIODeviceWeight=%s %" PRIu64
,
215 LIST_FOREACH(device_bandwidths
, b
, c
->blockio_device_bandwidths
) {
216 char buf
[FORMAT_BYTES_MAX
];
218 if (b
->rbps
!= CGROUP_LIMIT_MAX
)
220 "%sBlockIOReadBandwidth=%s %s\n",
223 format_bytes(buf
, sizeof(buf
), b
->rbps
));
224 if (b
->wbps
!= CGROUP_LIMIT_MAX
)
226 "%sBlockIOWriteBandwidth=%s %s\n",
229 format_bytes(buf
, sizeof(buf
), b
->wbps
));
233 static int lookup_block_device(const char *p
, dev_t
*dev
) {
242 return log_warning_errno(errno
, "Couldn't stat device %s: %m", p
);
244 if (S_ISBLK(st
.st_mode
))
246 else if (major(st
.st_dev
) != 0) {
247 /* If this is not a device node then find the block
248 * device this file is stored on */
251 /* If this is a partition, try to get the originating
253 block_get_whole_disk(*dev
, dev
);
255 log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p
);
262 static int whitelist_device(const char *path
, const char *node
, const char *acc
) {
263 char buf
[2+DECIMAL_STR_MAX(dev_t
)*2+2+4];
270 if (stat(node
, &st
) < 0) {
271 log_warning("Couldn't stat device %s", node
);
275 if (!S_ISCHR(st
.st_mode
) && !S_ISBLK(st
.st_mode
)) {
276 log_warning("%s is not a device.", node
);
282 S_ISCHR(st
.st_mode
) ? 'c' : 'b',
283 major(st
.st_rdev
), minor(st
.st_rdev
),
286 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
288 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
289 "Failed to set devices.allow on %s: %m", path
);
294 static int whitelist_major(const char *path
, const char *name
, char type
, const char *acc
) {
295 _cleanup_fclose_
FILE *f
= NULL
;
302 assert(type
== 'b' || type
== 'c');
304 f
= fopen("/proc/devices", "re");
306 return log_warning_errno(errno
, "Cannot open /proc/devices to resolve %s (%c): %m", name
, type
);
308 FOREACH_LINE(line
, f
, goto fail
) {
309 char buf
[2+DECIMAL_STR_MAX(unsigned)+3+4], *p
, *w
;
314 if (type
== 'c' && streq(line
, "Character devices:")) {
319 if (type
== 'b' && streq(line
, "Block devices:")) {
334 w
= strpbrk(p
, WHITESPACE
);
339 r
= safe_atou(p
, &maj
);
346 w
+= strspn(w
, WHITESPACE
);
348 if (fnmatch(name
, w
, 0) != 0)
357 r
= cg_set_attribute("devices", path
, "devices.allow", buf
);
359 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
360 "Failed to set devices.allow on %s: %m", path
);
366 log_warning_errno(errno
, "Failed to read /proc/devices: %m");
370 static bool cgroup_context_has_io_config(CGroupContext
*c
) {
371 return c
->io_accounting
||
372 c
->io_weight
!= CGROUP_WEIGHT_INVALID
||
373 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
||
374 c
->io_device_weights
||
378 static bool cgroup_context_has_blockio_config(CGroupContext
*c
) {
379 return c
->blockio_accounting
||
380 c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
381 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
||
382 c
->blockio_device_weights
||
383 c
->blockio_device_bandwidths
;
386 static uint64_t cgroup_context_io_weight(CGroupContext
*c
, ManagerState state
) {
387 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
388 c
->startup_io_weight
!= CGROUP_WEIGHT_INVALID
)
389 return c
->startup_io_weight
;
390 else if (c
->io_weight
!= CGROUP_WEIGHT_INVALID
)
393 return CGROUP_WEIGHT_DEFAULT
;
396 static uint64_t cgroup_context_blkio_weight(CGroupContext
*c
, ManagerState state
) {
397 if (IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) &&
398 c
->startup_blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
399 return c
->startup_blockio_weight
;
400 else if (c
->blockio_weight
!= CGROUP_BLKIO_WEIGHT_INVALID
)
401 return c
->blockio_weight
;
403 return CGROUP_BLKIO_WEIGHT_DEFAULT
;
406 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight
) {
407 return CLAMP(blkio_weight
* CGROUP_WEIGHT_DEFAULT
/ CGROUP_BLKIO_WEIGHT_DEFAULT
,
408 CGROUP_WEIGHT_MIN
, CGROUP_WEIGHT_MAX
);
411 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight
) {
412 return CLAMP(io_weight
* CGROUP_BLKIO_WEIGHT_DEFAULT
/ CGROUP_WEIGHT_DEFAULT
,
413 CGROUP_BLKIO_WEIGHT_MIN
, CGROUP_BLKIO_WEIGHT_MAX
);
416 static void cgroup_apply_io_device_weight(const char *path
, const char *dev_path
, uint64_t io_weight
) {
417 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
421 r
= lookup_block_device(dev_path
, &dev
);
425 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), io_weight
);
426 r
= cg_set_attribute("io", path
, "io.weight", buf
);
428 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
429 "Failed to set io.weight on %s: %m", path
);
432 static void cgroup_apply_blkio_device_weight(const char *path
, const char *dev_path
, uint64_t blkio_weight
) {
433 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
437 r
= lookup_block_device(dev_path
, &dev
);
441 xsprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), blkio_weight
);
442 r
= cg_set_attribute("blkio", path
, "blkio.weight_device", buf
);
444 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
445 "Failed to set blkio.weight_device on %s: %m", path
);
448 static unsigned cgroup_apply_io_device_limit(const char *path
, const char *dev_path
, uint64_t *limits
) {
449 char limit_bufs
[_CGROUP_IO_LIMIT_TYPE_MAX
][DECIMAL_STR_MAX(uint64_t)];
450 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
451 CGroupIOLimitType type
;
456 r
= lookup_block_device(dev_path
, &dev
);
460 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++) {
461 if (limits
[type
] != cgroup_io_limit_defaults
[type
]) {
462 xsprintf(limit_bufs
[type
], "%" PRIu64
, limits
[type
]);
465 xsprintf(limit_bufs
[type
], "%s", limits
[type
] == CGROUP_LIMIT_MAX
? "max" : "0");
469 xsprintf(buf
, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev
), minor(dev
),
470 limit_bufs
[CGROUP_IO_RBPS_MAX
], limit_bufs
[CGROUP_IO_WBPS_MAX
],
471 limit_bufs
[CGROUP_IO_RIOPS_MAX
], limit_bufs
[CGROUP_IO_WIOPS_MAX
]);
472 r
= cg_set_attribute("io", path
, "io.max", buf
);
474 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
475 "Failed to set io.max on %s: %m", path
);
479 static unsigned cgroup_apply_blkio_device_limit(const char *path
, const char *dev_path
, uint64_t rbps
, uint64_t wbps
) {
480 char buf
[DECIMAL_STR_MAX(dev_t
)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
485 r
= lookup_block_device(dev_path
, &dev
);
489 if (rbps
!= CGROUP_LIMIT_MAX
)
491 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), rbps
);
492 r
= cg_set_attribute("blkio", path
, "blkio.throttle.read_bps_device", buf
);
494 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
495 "Failed to set blkio.throttle.read_bps_device on %s: %m", path
);
497 if (wbps
!= CGROUP_LIMIT_MAX
)
499 sprintf(buf
, "%u:%u %" PRIu64
"\n", major(dev
), minor(dev
), wbps
);
500 r
= cg_set_attribute("blkio", path
, "blkio.throttle.write_bps_device", buf
);
502 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
503 "Failed to set blkio.throttle.write_bps_device on %s: %m", path
);
508 static bool cgroup_context_has_unified_memory_config(CGroupContext
*c
) {
509 return c
->memory_low
> 0 || c
->memory_high
!= CGROUP_LIMIT_MAX
|| c
->memory_max
!= CGROUP_LIMIT_MAX
;
512 static void cgroup_apply_unified_memory_limit(const char *path
, const char *file
, uint64_t v
) {
513 char buf
[DECIMAL_STR_MAX(uint64_t) + 1] = "max";
516 if (v
!= CGROUP_LIMIT_MAX
)
517 xsprintf(buf
, "%" PRIu64
"\n", v
);
519 r
= cg_set_attribute("memory", path
, file
, buf
);
521 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
522 "Failed to set %s on %s: %m", file
, path
);
525 void cgroup_context_apply(CGroupContext
*c
, CGroupMask mask
, const char *path
, ManagerState state
) {
535 /* Some cgroup attributes are not supported on the root cgroup,
536 * hence silently ignore */
537 is_root
= isempty(path
) || path_equal(path
, "/");
539 /* Make sure we don't try to display messages with an empty path. */
542 /* We generally ignore errors caused by read-only mounted
543 * cgroup trees (assuming we are running in a container then),
544 * and missing cgroups, i.e. EROFS and ENOENT. */
546 if ((mask
& CGROUP_MASK_CPU
) && !is_root
) {
547 char buf
[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t
)) + 1];
549 sprintf(buf
, "%" PRIu64
"\n",
550 IN_SET(state
, MANAGER_STARTING
, MANAGER_INITIALIZING
) && c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
? c
->startup_cpu_shares
:
551 c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
? c
->cpu_shares
: CGROUP_CPU_SHARES_DEFAULT
);
552 r
= cg_set_attribute("cpu", path
, "cpu.shares", buf
);
554 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
555 "Failed to set cpu.shares on %s: %m", path
);
557 sprintf(buf
, USEC_FMT
"\n", CGROUP_CPU_QUOTA_PERIOD_USEC
);
558 r
= cg_set_attribute("cpu", path
, "cpu.cfs_period_us", buf
);
560 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
561 "Failed to set cpu.cfs_period_us on %s: %m", path
);
563 if (c
->cpu_quota_per_sec_usec
!= USEC_INFINITY
) {
564 sprintf(buf
, USEC_FMT
"\n", c
->cpu_quota_per_sec_usec
* CGROUP_CPU_QUOTA_PERIOD_USEC
/ USEC_PER_SEC
);
565 r
= cg_set_attribute("cpu", path
, "cpu.cfs_quota_us", buf
);
567 r
= cg_set_attribute("cpu", path
, "cpu.cfs_quota_us", "-1");
569 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
570 "Failed to set cpu.cfs_quota_us on %s: %m", path
);
573 if (mask
& CGROUP_MASK_IO
) {
574 bool has_io
= cgroup_context_has_io_config(c
);
575 bool has_blockio
= cgroup_context_has_blockio_config(c
);
578 char buf
[8+DECIMAL_STR_MAX(uint64_t)+1];
582 weight
= cgroup_context_io_weight(c
, state
);
583 else if (has_blockio
)
584 weight
= cgroup_weight_blkio_to_io(cgroup_context_blkio_weight(c
, state
));
586 weight
= CGROUP_WEIGHT_DEFAULT
;
588 xsprintf(buf
, "default %" PRIu64
"\n", weight
);
589 r
= cg_set_attribute("io", path
, "io.weight", buf
);
591 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
592 "Failed to set io.weight on %s: %m", path
);
595 CGroupIODeviceWeight
*w
;
597 /* FIXME: no way to reset this list */
598 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
)
599 cgroup_apply_io_device_weight(path
, w
->path
, w
->weight
);
600 } else if (has_blockio
) {
601 CGroupBlockIODeviceWeight
*w
;
603 /* FIXME: no way to reset this list */
604 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
605 cgroup_apply_io_device_weight(path
, w
->path
, cgroup_weight_blkio_to_io(w
->weight
));
609 /* Apply limits and free ones without config. */
611 CGroupIODeviceLimit
*l
, *next
;
613 LIST_FOREACH_SAFE(device_limits
, l
, next
, c
->io_device_limits
) {
614 if (!cgroup_apply_io_device_limit(path
, l
->path
, l
->limits
))
615 cgroup_context_free_io_device_limit(c
, l
);
617 } else if (has_blockio
) {
618 CGroupBlockIODeviceBandwidth
*b
, *next
;
620 LIST_FOREACH_SAFE(device_bandwidths
, b
, next
, c
->blockio_device_bandwidths
) {
621 uint64_t limits
[_CGROUP_IO_LIMIT_TYPE_MAX
];
622 CGroupIOLimitType type
;
624 for (type
= 0; type
< _CGROUP_IO_LIMIT_TYPE_MAX
; type
++)
625 limits
[type
] = cgroup_io_limit_defaults
[type
];
627 limits
[CGROUP_IO_RBPS_MAX
] = b
->rbps
;
628 limits
[CGROUP_IO_WBPS_MAX
] = b
->wbps
;
630 if (!cgroup_apply_io_device_limit(path
, b
->path
, limits
))
631 cgroup_context_free_blockio_device_bandwidth(c
, b
);
636 if (mask
& CGROUP_MASK_BLKIO
) {
637 bool has_io
= cgroup_context_has_io_config(c
);
638 bool has_blockio
= cgroup_context_has_blockio_config(c
);
641 char buf
[DECIMAL_STR_MAX(uint64_t)+1];
645 weight
= cgroup_context_blkio_weight(c
, state
);
647 weight
= cgroup_weight_io_to_blkio(cgroup_context_io_weight(c
, state
));
649 weight
= CGROUP_BLKIO_WEIGHT_DEFAULT
;
651 xsprintf(buf
, "%" PRIu64
"\n", weight
);
652 r
= cg_set_attribute("blkio", path
, "blkio.weight", buf
);
654 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
655 "Failed to set blkio.weight on %s: %m", path
);
658 CGroupBlockIODeviceWeight
*w
;
660 /* FIXME: no way to reset this list */
661 LIST_FOREACH(device_weights
, w
, c
->blockio_device_weights
)
662 cgroup_apply_blkio_device_weight(path
, w
->path
, w
->weight
);
664 CGroupIODeviceWeight
*w
;
666 /* FIXME: no way to reset this list */
667 LIST_FOREACH(device_weights
, w
, c
->io_device_weights
)
668 cgroup_apply_blkio_device_weight(path
, w
->path
, cgroup_weight_io_to_blkio(w
->weight
));
672 /* Apply limits and free ones without config. */
674 CGroupBlockIODeviceBandwidth
*b
, *next
;
676 LIST_FOREACH_SAFE(device_bandwidths
, b
, next
, c
->blockio_device_bandwidths
) {
677 if (!cgroup_apply_blkio_device_limit(path
, b
->path
, b
->rbps
, b
->wbps
))
678 cgroup_context_free_blockio_device_bandwidth(c
, b
);
681 CGroupIODeviceLimit
*l
, *next
;
683 LIST_FOREACH_SAFE(device_limits
, l
, next
, c
->io_device_limits
) {
684 if (!cgroup_apply_blkio_device_limit(path
, l
->path
, l
->limits
[CGROUP_IO_RBPS_MAX
], l
->limits
[CGROUP_IO_WBPS_MAX
]))
685 cgroup_context_free_io_device_limit(c
, l
);
690 if ((mask
& CGROUP_MASK_MEMORY
) && !is_root
) {
691 if (cg_unified() > 0) {
692 uint64_t max
= c
->memory_max
;
694 if (cgroup_context_has_unified_memory_config(c
))
697 max
= c
->memory_limit
;
699 cgroup_apply_unified_memory_limit(path
, "memory.low", c
->memory_low
);
700 cgroup_apply_unified_memory_limit(path
, "memory.high", c
->memory_high
);
701 cgroup_apply_unified_memory_limit(path
, "memory.max", max
);
703 char buf
[DECIMAL_STR_MAX(uint64_t) + 1];
705 if (c
->memory_limit
!= CGROUP_LIMIT_MAX
)
706 xsprintf(buf
, "%" PRIu64
"\n", c
->memory_limit
);
708 xsprintf(buf
, "%" PRIu64
"\n", c
->memory_max
);
710 r
= cg_set_attribute("memory", path
, "memory.limit_in_bytes", buf
);
712 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
713 "Failed to set memory.limit_in_bytes on %s: %m", path
);
717 if ((mask
& CGROUP_MASK_DEVICES
) && !is_root
) {
718 CGroupDeviceAllow
*a
;
720 /* Changing the devices list of a populated cgroup
721 * might result in EINVAL, hence ignore EINVAL
724 if (c
->device_allow
|| c
->device_policy
!= CGROUP_AUTO
)
725 r
= cg_set_attribute("devices", path
, "devices.deny", "a");
727 r
= cg_set_attribute("devices", path
, "devices.allow", "a");
729 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EINVAL
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
730 "Failed to reset devices.list on %s: %m", path
);
732 if (c
->device_policy
== CGROUP_CLOSED
||
733 (c
->device_policy
== CGROUP_AUTO
&& c
->device_allow
)) {
734 static const char auto_devices
[] =
735 "/dev/null\0" "rwm\0"
736 "/dev/zero\0" "rwm\0"
737 "/dev/full\0" "rwm\0"
738 "/dev/random\0" "rwm\0"
739 "/dev/urandom\0" "rwm\0"
741 "/dev/pts/ptmx\0" "rw\0"; /* /dev/pts/ptmx may not be duplicated, but accessed */
745 NULSTR_FOREACH_PAIR(x
, y
, auto_devices
)
746 whitelist_device(path
, x
, y
);
748 whitelist_major(path
, "pts", 'c', "rw");
749 whitelist_major(path
, "kdbus", 'c', "rw");
750 whitelist_major(path
, "kdbus/*", 'c', "rw");
753 LIST_FOREACH(device_allow
, a
, c
->device_allow
) {
769 if (startswith(a
->path
, "/dev/"))
770 whitelist_device(path
, a
->path
, acc
);
771 else if (startswith(a
->path
, "block-"))
772 whitelist_major(path
, a
->path
+ 6, 'b', acc
);
773 else if (startswith(a
->path
, "char-"))
774 whitelist_major(path
, a
->path
+ 5, 'c', acc
);
776 log_debug("Ignoring device %s while writing cgroup attribute.", a
->path
);
780 if ((mask
& CGROUP_MASK_PIDS
) && !is_root
) {
782 if (c
->tasks_max
!= (uint64_t) -1) {
783 char buf
[DECIMAL_STR_MAX(uint64_t) + 2];
785 sprintf(buf
, "%" PRIu64
"\n", c
->tasks_max
);
786 r
= cg_set_attribute("pids", path
, "pids.max", buf
);
788 r
= cg_set_attribute("pids", path
, "pids.max", "max");
791 log_full_errno(IN_SET(r
, -ENOENT
, -EROFS
, -EACCES
) ? LOG_DEBUG
: LOG_WARNING
, r
,
792 "Failed to set pids.max on %s: %m", path
);
796 CGroupMask
cgroup_context_get_mask(CGroupContext
*c
) {
799 /* Figure out which controllers we need */
801 if (c
->cpu_accounting
||
802 c
->cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
803 c
->startup_cpu_shares
!= CGROUP_CPU_SHARES_INVALID
||
804 c
->cpu_quota_per_sec_usec
!= USEC_INFINITY
)
805 mask
|= CGROUP_MASK_CPUACCT
| CGROUP_MASK_CPU
;
807 if (cgroup_context_has_io_config(c
) || cgroup_context_has_blockio_config(c
))
808 mask
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
810 if (c
->memory_accounting
||
811 c
->memory_limit
!= CGROUP_LIMIT_MAX
||
812 cgroup_context_has_unified_memory_config(c
))
813 mask
|= CGROUP_MASK_MEMORY
;
815 if (c
->device_allow
||
816 c
->device_policy
!= CGROUP_AUTO
)
817 mask
|= CGROUP_MASK_DEVICES
;
819 if (c
->tasks_accounting
||
820 c
->tasks_max
!= (uint64_t) -1)
821 mask
|= CGROUP_MASK_PIDS
;
826 CGroupMask
unit_get_own_mask(Unit
*u
) {
829 /* Returns the mask of controllers the unit needs for itself */
831 c
= unit_get_cgroup_context(u
);
835 /* If delegation is turned on, then turn on all cgroups,
836 * unless we are on the legacy hierarchy and the process we
837 * fork into it is known to drop privileges, and hence
838 * shouldn't get access to the controllers.
840 * Note that on the unified hierarchy it is safe to delegate
841 * controllers to unprivileged services. */
846 e
= unit_get_exec_context(u
);
848 exec_context_maintains_privileges(e
) ||
850 return _CGROUP_MASK_ALL
;
853 return cgroup_context_get_mask(c
);
856 CGroupMask
unit_get_members_mask(Unit
*u
) {
859 /* Returns the mask of controllers all of the unit's children
862 if (u
->cgroup_members_mask_valid
)
863 return u
->cgroup_members_mask
;
865 u
->cgroup_members_mask
= 0;
867 if (u
->type
== UNIT_SLICE
) {
871 SET_FOREACH(member
, u
->dependencies
[UNIT_BEFORE
], i
) {
876 if (UNIT_DEREF(member
->slice
) != u
)
879 u
->cgroup_members_mask
|=
880 unit_get_own_mask(member
) |
881 unit_get_members_mask(member
);
885 u
->cgroup_members_mask_valid
= true;
886 return u
->cgroup_members_mask
;
889 CGroupMask
unit_get_siblings_mask(Unit
*u
) {
892 /* Returns the mask of controllers all of the unit's siblings
893 * require, i.e. the members mask of the unit's parent slice
894 * if there is one. */
896 if (UNIT_ISSET(u
->slice
))
897 return unit_get_members_mask(UNIT_DEREF(u
->slice
));
899 return unit_get_own_mask(u
) | unit_get_members_mask(u
);
902 CGroupMask
unit_get_subtree_mask(Unit
*u
) {
904 /* Returns the mask of this subtree, meaning of the group
905 * itself and its children. */
907 return unit_get_own_mask(u
) | unit_get_members_mask(u
);
910 CGroupMask
unit_get_target_mask(Unit
*u
) {
913 /* This returns the cgroup mask of all controllers to enable
914 * for a specific cgroup, i.e. everything it needs itself,
915 * plus all that its children need, plus all that its siblings
916 * need. This is primarily useful on the legacy cgroup
917 * hierarchy, where we need to duplicate each cgroup in each
918 * hierarchy that shall be enabled for it. */
920 mask
= unit_get_own_mask(u
) | unit_get_members_mask(u
) | unit_get_siblings_mask(u
);
921 mask
&= u
->manager
->cgroup_supported
;
926 CGroupMask
unit_get_enable_mask(Unit
*u
) {
929 /* This returns the cgroup mask of all controllers to enable
930 * for the children of a specific cgroup. This is primarily
931 * useful for the unified cgroup hierarchy, where each cgroup
932 * controls which controllers are enabled for its children. */
934 mask
= unit_get_members_mask(u
);
935 mask
&= u
->manager
->cgroup_supported
;
940 /* Recurse from a unit up through its containing slices, propagating
941 * mask bits upward. A unit is also member of itself. */
942 void unit_update_cgroup_members_masks(Unit
*u
) {
948 /* Calculate subtree mask */
949 m
= unit_get_subtree_mask(u
);
951 /* See if anything changed from the previous invocation. If
952 * not, we're done. */
953 if (u
->cgroup_subtree_mask_valid
&& m
== u
->cgroup_subtree_mask
)
957 u
->cgroup_subtree_mask_valid
&&
958 ((m
& ~u
->cgroup_subtree_mask
) != 0) &&
959 ((~m
& u
->cgroup_subtree_mask
) == 0);
961 u
->cgroup_subtree_mask
= m
;
962 u
->cgroup_subtree_mask_valid
= true;
964 if (UNIT_ISSET(u
->slice
)) {
965 Unit
*s
= UNIT_DEREF(u
->slice
);
968 /* There's more set now than before. We
969 * propagate the new mask to the parent's mask
970 * (not caring if it actually was valid or
973 s
->cgroup_members_mask
|= m
;
976 /* There's less set now than before (or we
977 * don't know), we need to recalculate
978 * everything, so let's invalidate the
979 * parent's members mask */
981 s
->cgroup_members_mask_valid
= false;
983 /* And now make sure that this change also hits our
985 unit_update_cgroup_members_masks(s
);
989 static const char *migrate_callback(CGroupMask mask
, void *userdata
) {
996 if (u
->cgroup_path
&&
997 u
->cgroup_realized
&&
998 (u
->cgroup_realized_mask
& mask
) == mask
)
999 return u
->cgroup_path
;
1001 u
= UNIT_DEREF(u
->slice
);
1007 char *unit_default_cgroup_path(Unit
*u
) {
1008 _cleanup_free_
char *escaped
= NULL
, *slice
= NULL
;
1013 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1014 return strdup(u
->manager
->cgroup_root
);
1016 if (UNIT_ISSET(u
->slice
) && !unit_has_name(UNIT_DEREF(u
->slice
), SPECIAL_ROOT_SLICE
)) {
1017 r
= cg_slice_to_path(UNIT_DEREF(u
->slice
)->id
, &slice
);
1022 escaped
= cg_escape(u
->id
);
1027 return strjoin(u
->manager
->cgroup_root
, "/", slice
, "/", escaped
, NULL
);
1029 return strjoin(u
->manager
->cgroup_root
, "/", escaped
, NULL
);
1032 int unit_set_cgroup_path(Unit
*u
, const char *path
) {
1033 _cleanup_free_
char *p
= NULL
;
1045 if (streq_ptr(u
->cgroup_path
, p
))
1049 r
= hashmap_put(u
->manager
->cgroup_unit
, p
, u
);
1054 unit_release_cgroup(u
);
1062 int unit_watch_cgroup(Unit
*u
) {
1063 _cleanup_free_
char *events
= NULL
;
1068 if (!u
->cgroup_path
)
1071 if (u
->cgroup_inotify_wd
>= 0)
1074 /* Only applies to the unified hierarchy */
1077 return log_unit_error_errno(u
, r
, "Failed detect wether the unified hierarchy is used: %m");
1081 /* Don't watch the root slice, it's pointless. */
1082 if (unit_has_name(u
, SPECIAL_ROOT_SLICE
))
1085 r
= hashmap_ensure_allocated(&u
->manager
->cgroup_inotify_wd_unit
, &trivial_hash_ops
);
1089 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, "cgroup.events", &events
);
1093 u
->cgroup_inotify_wd
= inotify_add_watch(u
->manager
->cgroup_inotify_fd
, events
, IN_MODIFY
);
1094 if (u
->cgroup_inotify_wd
< 0) {
1096 if (errno
== ENOENT
) /* If the directory is already
1097 * gone we don't need to track
1098 * it, so this is not an error */
1101 return log_unit_error_errno(u
, errno
, "Failed to add inotify watch descriptor for control group %s: %m", u
->cgroup_path
);
1104 r
= hashmap_put(u
->manager
->cgroup_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_inotify_wd
), u
);
1106 return log_unit_error_errno(u
, r
, "Failed to add inotify watch descriptor to hash map: %m");
1111 static int unit_create_cgroup(
1113 CGroupMask target_mask
,
1114 CGroupMask enable_mask
) {
1121 c
= unit_get_cgroup_context(u
);
1125 if (!u
->cgroup_path
) {
1126 _cleanup_free_
char *path
= NULL
;
1128 path
= unit_default_cgroup_path(u
);
1132 r
= unit_set_cgroup_path(u
, path
);
1134 return log_unit_error_errno(u
, r
, "Control group %s exists already.", path
);
1136 return log_unit_error_errno(u
, r
, "Failed to set unit's control group path to %s: %m", path
);
1139 /* First, create our own group */
1140 r
= cg_create_everywhere(u
->manager
->cgroup_supported
, target_mask
, u
->cgroup_path
);
1142 return log_unit_error_errno(u
, r
, "Failed to create cgroup %s: %m", u
->cgroup_path
);
1144 /* Start watching it */
1145 (void) unit_watch_cgroup(u
);
1147 /* Enable all controllers we need */
1148 r
= cg_enable_everywhere(u
->manager
->cgroup_supported
, enable_mask
, u
->cgroup_path
);
1150 log_unit_warning_errno(u
, r
, "Failed to enable controllers on cgroup %s, ignoring: %m", u
->cgroup_path
);
1152 /* Keep track that this is now realized */
1153 u
->cgroup_realized
= true;
1154 u
->cgroup_realized_mask
= target_mask
;
1155 u
->cgroup_enabled_mask
= enable_mask
;
1157 if (u
->type
!= UNIT_SLICE
&& !c
->delegate
) {
1159 /* Then, possibly move things over, but not if
1160 * subgroups may contain processes, which is the case
1161 * for slice and delegation units. */
1162 r
= cg_migrate_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, u
->cgroup_path
, migrate_callback
, u
);
1164 log_unit_warning_errno(u
, r
, "Failed to migrate cgroup from to %s, ignoring: %m", u
->cgroup_path
);
1170 int unit_attach_pids_to_cgroup(Unit
*u
) {
1174 r
= unit_realize_cgroup(u
);
1178 r
= cg_attach_many_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, u
->pids
, migrate_callback
, u
);
1185 static bool unit_has_mask_realized(Unit
*u
, CGroupMask target_mask
, CGroupMask enable_mask
) {
1188 return u
->cgroup_realized
&& u
->cgroup_realized_mask
== target_mask
&& u
->cgroup_enabled_mask
== enable_mask
;
1191 /* Check if necessary controllers and attributes for a unit are in place.
1193 * If so, do nothing.
1194 * If not, create paths, move processes over, and set attributes.
1196 * Returns 0 on success and < 0 on failure. */
1197 static int unit_realize_cgroup_now(Unit
*u
, ManagerState state
) {
1198 CGroupMask target_mask
, enable_mask
;
1203 if (u
->in_cgroup_queue
) {
1204 LIST_REMOVE(cgroup_queue
, u
->manager
->cgroup_queue
, u
);
1205 u
->in_cgroup_queue
= false;
1208 target_mask
= unit_get_target_mask(u
);
1209 enable_mask
= unit_get_enable_mask(u
);
1211 if (unit_has_mask_realized(u
, target_mask
, enable_mask
))
1214 /* First, realize parents */
1215 if (UNIT_ISSET(u
->slice
)) {
1216 r
= unit_realize_cgroup_now(UNIT_DEREF(u
->slice
), state
);
1221 /* And then do the real work */
1222 r
= unit_create_cgroup(u
, target_mask
, enable_mask
);
1226 /* Finally, apply the necessary attributes. */
1227 cgroup_context_apply(unit_get_cgroup_context(u
), target_mask
, u
->cgroup_path
, state
);
1232 static void unit_add_to_cgroup_queue(Unit
*u
) {
1234 if (u
->in_cgroup_queue
)
1237 LIST_PREPEND(cgroup_queue
, u
->manager
->cgroup_queue
, u
);
1238 u
->in_cgroup_queue
= true;
1241 unsigned manager_dispatch_cgroup_queue(Manager
*m
) {
1247 state
= manager_state(m
);
1249 while ((i
= m
->cgroup_queue
)) {
1250 assert(i
->in_cgroup_queue
);
1252 r
= unit_realize_cgroup_now(i
, state
);
1254 log_warning_errno(r
, "Failed to realize cgroups for queued unit %s, ignoring: %m", i
->id
);
1262 static void unit_queue_siblings(Unit
*u
) {
1265 /* This adds the siblings of the specified unit and the
1266 * siblings of all parent units to the cgroup queue. (But
1267 * neither the specified unit itself nor the parents.) */
1269 while ((slice
= UNIT_DEREF(u
->slice
))) {
1273 SET_FOREACH(m
, slice
->dependencies
[UNIT_BEFORE
], i
) {
1277 /* Skip units that have a dependency on the slice
1278 * but aren't actually in it. */
1279 if (UNIT_DEREF(m
->slice
) != slice
)
1282 /* No point in doing cgroup application for units
1283 * without active processes. */
1284 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m
)))
1287 /* If the unit doesn't need any new controllers
1288 * and has current ones realized, it doesn't need
1290 if (unit_has_mask_realized(m
, unit_get_target_mask(m
), unit_get_enable_mask(m
)))
1293 unit_add_to_cgroup_queue(m
);
1300 int unit_realize_cgroup(Unit
*u
) {
1303 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1306 /* So, here's the deal: when realizing the cgroups for this
1307 * unit, we need to first create all parents, but there's more
1308 * actually: for the weight-based controllers we also need to
1309 * make sure that all our siblings (i.e. units that are in the
1310 * same slice as we are) have cgroups, too. Otherwise, things
1311 * would become very uneven as each of their processes would
1312 * get as much resources as all our group together. This call
1313 * will synchronously create the parent cgroups, but will
1314 * defer work on the siblings to the next event loop
1317 /* Add all sibling slices to the cgroup queue. */
1318 unit_queue_siblings(u
);
1320 /* And realize this one now (and apply the values) */
1321 return unit_realize_cgroup_now(u
, manager_state(u
->manager
));
1324 void unit_release_cgroup(Unit
*u
) {
1327 /* Forgets all cgroup details for this cgroup */
1329 if (u
->cgroup_path
) {
1330 (void) hashmap_remove(u
->manager
->cgroup_unit
, u
->cgroup_path
);
1331 u
->cgroup_path
= mfree(u
->cgroup_path
);
1334 if (u
->cgroup_inotify_wd
>= 0) {
1335 if (inotify_rm_watch(u
->manager
->cgroup_inotify_fd
, u
->cgroup_inotify_wd
) < 0)
1336 log_unit_debug_errno(u
, errno
, "Failed to remove cgroup inotify watch %i for %s, ignoring", u
->cgroup_inotify_wd
, u
->id
);
1338 (void) hashmap_remove(u
->manager
->cgroup_inotify_wd_unit
, INT_TO_PTR(u
->cgroup_inotify_wd
));
1339 u
->cgroup_inotify_wd
= -1;
1343 void unit_prune_cgroup(Unit
*u
) {
1349 /* Removes the cgroup, if empty and possible, and stops watching it. */
1351 if (!u
->cgroup_path
)
1354 is_root_slice
= unit_has_name(u
, SPECIAL_ROOT_SLICE
);
1356 r
= cg_trim_everywhere(u
->manager
->cgroup_supported
, u
->cgroup_path
, !is_root_slice
);
1358 log_debug_errno(r
, "Failed to destroy cgroup %s, ignoring: %m", u
->cgroup_path
);
1365 unit_release_cgroup(u
);
1367 u
->cgroup_realized
= false;
1368 u
->cgroup_realized_mask
= 0;
1369 u
->cgroup_enabled_mask
= 0;
1372 int unit_search_main_pid(Unit
*u
, pid_t
*ret
) {
1373 _cleanup_fclose_
FILE *f
= NULL
;
1374 pid_t pid
= 0, npid
, mypid
;
1380 if (!u
->cgroup_path
)
1383 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
, &f
);
1388 while (cg_read_pid(f
, &npid
) > 0) {
1394 /* Ignore processes that aren't our kids */
1395 if (get_process_ppid(npid
, &ppid
) >= 0 && ppid
!= mypid
)
1399 /* Dang, there's more than one daemonized PID
1400 in this group, so we don't know what process
1401 is the main process. */
1412 static int unit_watch_pids_in_path(Unit
*u
, const char *path
) {
1413 _cleanup_closedir_
DIR *d
= NULL
;
1414 _cleanup_fclose_
FILE *f
= NULL
;
1420 r
= cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER
, path
, &f
);
1426 while ((r
= cg_read_pid(f
, &pid
)) > 0) {
1427 r
= unit_watch_pid(u
, pid
);
1428 if (r
< 0 && ret
>= 0)
1432 if (r
< 0 && ret
>= 0)
1436 r
= cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER
, path
, &d
);
1443 while ((r
= cg_read_subgroup(d
, &fn
)) > 0) {
1444 _cleanup_free_
char *p
= NULL
;
1446 p
= strjoin(path
, "/", fn
, NULL
);
1452 r
= unit_watch_pids_in_path(u
, p
);
1453 if (r
< 0 && ret
>= 0)
1457 if (r
< 0 && ret
>= 0)
1464 int unit_watch_all_pids(Unit
*u
) {
1467 /* Adds all PIDs from our cgroup to the set of PIDs we
1468 * watch. This is a fallback logic for cases where we do not
1469 * get reliable cgroup empty notifications: we try to use
1470 * SIGCHLD as replacement. */
1472 if (!u
->cgroup_path
)
1475 if (cg_unified() > 0) /* On unified we can use proper notifications */
1478 return unit_watch_pids_in_path(u
, u
->cgroup_path
);
1481 int unit_notify_cgroup_empty(Unit
*u
) {
1486 if (!u
->cgroup_path
)
1489 r
= cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER
, u
->cgroup_path
);
1493 unit_add_to_gc_queue(u
);
1495 if (UNIT_VTABLE(u
)->notify_cgroup_empty
)
1496 UNIT_VTABLE(u
)->notify_cgroup_empty(u
);
1501 static int on_cgroup_inotify_event(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
1502 Manager
*m
= userdata
;
1509 union inotify_event_buffer buffer
;
1510 struct inotify_event
*e
;
1513 l
= read(fd
, &buffer
, sizeof(buffer
));
1515 if (errno
== EINTR
|| errno
== EAGAIN
)
1518 return log_error_errno(errno
, "Failed to read control group inotify events: %m");
1521 FOREACH_INOTIFY_EVENT(e
, buffer
, l
) {
1525 /* Queue overflow has no watch descriptor */
1528 if (e
->mask
& IN_IGNORED
)
1529 /* The watch was just removed */
1532 u
= hashmap_get(m
->cgroup_inotify_wd_unit
, INT_TO_PTR(e
->wd
));
1533 if (!u
) /* Not that inotify might deliver
1534 * events for a watch even after it
1535 * was removed, because it was queued
1536 * before the removal. Let's ignore
1537 * this here safely. */
1540 (void) unit_notify_cgroup_empty(u
);
1545 int manager_setup_cgroup(Manager
*m
) {
1546 _cleanup_free_
char *path
= NULL
;
1553 /* 1. Determine hierarchy */
1554 m
->cgroup_root
= mfree(m
->cgroup_root
);
1555 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 0, &m
->cgroup_root
);
1557 return log_error_errno(r
, "Cannot determine cgroup we are running in: %m");
1559 /* Chop off the init scope, if we are already located in it */
1560 e
= endswith(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
1562 /* LEGACY: Also chop off the system slice if we are in
1563 * it. This is to support live upgrades from older systemd
1564 * versions where PID 1 was moved there. Also see
1565 * cg_get_root_path(). */
1566 if (!e
&& MANAGER_IS_SYSTEM(m
)) {
1567 e
= endswith(m
->cgroup_root
, "/" SPECIAL_SYSTEM_SLICE
);
1569 e
= endswith(m
->cgroup_root
, "/system"); /* even more legacy */
1574 /* And make sure to store away the root value without trailing
1575 * slash, even for the root dir, so that we can easily prepend
1577 while ((e
= endswith(m
->cgroup_root
, "/")))
1581 r
= cg_get_path(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, NULL
, &path
);
1583 return log_error_errno(r
, "Cannot find cgroup mount point: %m");
1585 unified
= cg_unified();
1587 return log_error_errno(r
, "Couldn't determine if we are running in the unified hierarchy: %m");
1589 log_debug("Unified cgroup hierarchy is located at %s.", path
);
1591 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER
". File system hierarchy is at %s.", path
);
1594 const char *scope_path
;
1596 /* 3. Install agent */
1599 /* In the unified hierarchy we can can get
1600 * cgroup empty notifications via inotify. */
1602 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
1603 safe_close(m
->cgroup_inotify_fd
);
1605 m
->cgroup_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
1606 if (m
->cgroup_inotify_fd
< 0)
1607 return log_error_errno(errno
, "Failed to create control group inotify object: %m");
1609 r
= sd_event_add_io(m
->event
, &m
->cgroup_inotify_event_source
, m
->cgroup_inotify_fd
, EPOLLIN
, on_cgroup_inotify_event
, m
);
1611 return log_error_errno(r
, "Failed to watch control group inotify object: %m");
1613 /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also
1614 * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
1615 r
= sd_event_source_set_priority(m
->cgroup_inotify_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
1617 return log_error_errno(r
, "Failed to set priority of inotify event source: %m");
1619 (void) sd_event_source_set_description(m
->cgroup_inotify_event_source
, "cgroup-inotify");
1621 } else if (MANAGER_IS_SYSTEM(m
)) {
1623 /* On the legacy hierarchy we only get
1624 * notifications via cgroup agents. (Which
1625 * isn't really reliable, since it does not
1626 * generate events when control groups with
1627 * children run empty. */
1629 r
= cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER
, SYSTEMD_CGROUP_AGENT_PATH
);
1631 log_warning_errno(r
, "Failed to install release agent, ignoring: %m");
1633 log_debug("Installed release agent.");
1635 log_debug("Release agent already installed.");
1638 /* 4. Make sure we are in the special "init.scope" unit in the root slice. */
1639 scope_path
= strjoina(m
->cgroup_root
, "/" SPECIAL_INIT_SCOPE
);
1640 r
= cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER
, scope_path
, 0);
1642 return log_error_errno(r
, "Failed to create %s control group: %m", scope_path
);
1644 /* also, move all other userspace processes remaining
1645 * in the root cgroup into that scope. */
1646 r
= cg_migrate(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, SYSTEMD_CGROUP_CONTROLLER
, scope_path
, false);
1648 log_warning_errno(r
, "Couldn't move remaining userspace processes, ignoring: %m");
1650 /* 5. And pin it, so that it cannot be unmounted */
1651 safe_close(m
->pin_cgroupfs_fd
);
1652 m
->pin_cgroupfs_fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOCTTY
|O_NONBLOCK
);
1653 if (m
->pin_cgroupfs_fd
< 0)
1654 return log_error_errno(errno
, "Failed to open pin file: %m");
1656 /* 6. Always enable hierarchical support if it exists... */
1658 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
1661 /* 7. Figure out which controllers are supported */
1662 r
= cg_mask_supported(&m
->cgroup_supported
);
1664 return log_error_errno(r
, "Failed to determine supported controllers: %m");
1666 for (c
= 0; c
< _CGROUP_CONTROLLER_MAX
; c
++)
1667 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c
), yes_no(m
->cgroup_supported
& c
));
1672 void manager_shutdown_cgroup(Manager
*m
, bool delete) {
1675 /* We can't really delete the group, since we are in it. But
1677 if (delete && m
->cgroup_root
)
1678 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER
, m
->cgroup_root
, false);
1680 m
->cgroup_inotify_wd_unit
= hashmap_free(m
->cgroup_inotify_wd_unit
);
1682 m
->cgroup_inotify_event_source
= sd_event_source_unref(m
->cgroup_inotify_event_source
);
1683 m
->cgroup_inotify_fd
= safe_close(m
->cgroup_inotify_fd
);
1685 m
->pin_cgroupfs_fd
= safe_close(m
->pin_cgroupfs_fd
);
1687 m
->cgroup_root
= mfree(m
->cgroup_root
);
1690 Unit
* manager_get_unit_by_cgroup(Manager
*m
, const char *cgroup
) {
1697 u
= hashmap_get(m
->cgroup_unit
, cgroup
);
1701 p
= strdupa(cgroup
);
1705 e
= strrchr(p
, '/');
1707 return hashmap_get(m
->cgroup_unit
, SPECIAL_ROOT_SLICE
);
1711 u
= hashmap_get(m
->cgroup_unit
, p
);
1717 Unit
*manager_get_unit_by_pid_cgroup(Manager
*m
, pid_t pid
) {
1718 _cleanup_free_
char *cgroup
= NULL
;
1726 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, pid
, &cgroup
);
1730 return manager_get_unit_by_cgroup(m
, cgroup
);
1733 Unit
*manager_get_unit_by_pid(Manager
*m
, pid_t pid
) {
1742 return hashmap_get(m
->units
, SPECIAL_INIT_SCOPE
);
1744 u
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(pid
));
1748 u
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(pid
));
1752 return manager_get_unit_by_pid_cgroup(m
, pid
);
1755 int manager_notify_cgroup_empty(Manager
*m
, const char *cgroup
) {
1761 log_debug("Got cgroup empty notification for: %s", cgroup
);
1763 u
= manager_get_unit_by_cgroup(m
, cgroup
);
1767 return unit_notify_cgroup_empty(u
);
1770 int unit_get_memory_current(Unit
*u
, uint64_t *ret
) {
1771 _cleanup_free_
char *v
= NULL
;
1777 if (!u
->cgroup_path
)
1780 if ((u
->cgroup_realized_mask
& CGROUP_MASK_MEMORY
) == 0)
1783 if (cg_unified() <= 0)
1784 r
= cg_get_attribute("memory", u
->cgroup_path
, "memory.usage_in_bytes", &v
);
1786 r
= cg_get_attribute("memory", u
->cgroup_path
, "memory.current", &v
);
1792 return safe_atou64(v
, ret
);
1795 int unit_get_tasks_current(Unit
*u
, uint64_t *ret
) {
1796 _cleanup_free_
char *v
= NULL
;
1802 if (!u
->cgroup_path
)
1805 if ((u
->cgroup_realized_mask
& CGROUP_MASK_PIDS
) == 0)
1808 r
= cg_get_attribute("pids", u
->cgroup_path
, "pids.current", &v
);
1814 return safe_atou64(v
, ret
);
1817 static int unit_get_cpu_usage_raw(Unit
*u
, nsec_t
*ret
) {
1818 _cleanup_free_
char *v
= NULL
;
1825 if (!u
->cgroup_path
)
1828 if ((u
->cgroup_realized_mask
& CGROUP_MASK_CPUACCT
) == 0)
1831 r
= cg_get_attribute("cpuacct", u
->cgroup_path
, "cpuacct.usage", &v
);
1837 r
= safe_atou64(v
, &ns
);
1845 int unit_get_cpu_usage(Unit
*u
, nsec_t
*ret
) {
1849 r
= unit_get_cpu_usage_raw(u
, &ns
);
1853 if (ns
> u
->cpuacct_usage_base
)
1854 ns
-= u
->cpuacct_usage_base
;
1862 int unit_reset_cpu_usage(Unit
*u
) {
1868 r
= unit_get_cpu_usage_raw(u
, &ns
);
1870 u
->cpuacct_usage_base
= 0;
1874 u
->cpuacct_usage_base
= ns
;
1878 bool unit_cgroup_delegate(Unit
*u
) {
1883 c
= unit_get_cgroup_context(u
);
1890 void unit_invalidate_cgroup(Unit
*u
, CGroupMask m
) {
1893 if (!UNIT_HAS_CGROUP_CONTEXT(u
))
1899 /* always invalidate compat pairs together */
1900 if (m
& (CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
))
1901 m
|= CGROUP_MASK_IO
| CGROUP_MASK_BLKIO
;
1903 if ((u
->cgroup_realized_mask
& m
) == 0)
1906 u
->cgroup_realized_mask
&= ~m
;
1907 unit_add_to_cgroup_queue(u
);
1910 void manager_invalidate_startup_units(Manager
*m
) {
1916 SET_FOREACH(u
, m
->startup_units
, i
)
1917 unit_invalidate_cgroup(u
, CGROUP_MASK_CPU
|CGROUP_MASK_IO
|CGROUP_MASK_BLKIO
);
1920 static const char* const cgroup_device_policy_table
[_CGROUP_DEVICE_POLICY_MAX
] = {
1921 [CGROUP_AUTO
] = "auto",
1922 [CGROUP_CLOSED
] = "closed",
1923 [CGROUP_STRICT
] = "strict",
1926 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy
, CGroupDevicePolicy
);