X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fcore%2Fcgroup.c;h=b53b35dad0ac5767205fcc7ee08d81a5a6340423;hp=ec53ce73f15e965853762526befd13b5efdd0498;hb=e916c1378b697b1f267d021a823bbfcb5ac8376c;hpb=a46d97abe772733d301fc1e6fb7bdfc4f0eba1ab diff --git a/src/core/cgroup.c b/src/core/cgroup.c index ec53ce73f..b53b35dad 100644 --- a/src/core/cgroup.c +++ b/src/core/cgroup.c @@ -32,19 +32,38 @@ //#include "special.h" #include "string-table.h" #include "string-util.h" +#include "stdio-util.h" #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) #if 0 /// UNNEEDED by elogind +static void cgroup_compat_warn(void) { + static bool cgroup_compat_warned = false; + + if (cgroup_compat_warned) + return; + + log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. See cgroup-compat debug messages for details."); + cgroup_compat_warned = true; +} + +#define log_cgroup_compat(unit, fmt, ...) do { \ + cgroup_compat_warn(); \ + log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ + } while (false) + void cgroup_context_init(CGroupContext *c) { assert(c); /* Initialize everything to the kernel defaults, assuming the * structure is preinitialized to 0 */ + c->cpu_weight = CGROUP_WEIGHT_INVALID; + c->startup_cpu_weight = CGROUP_WEIGHT_INVALID; + c->cpu_quota_per_sec_usec = USEC_INFINITY; + c->cpu_shares = CGROUP_CPU_SHARES_INVALID; c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID; - c->cpu_quota_per_sec_usec = USEC_INFINITY; c->memory_high = CGROUP_LIMIT_MAX; c->memory_max = CGROUP_LIMIT_MAX; @@ -143,6 +162,8 @@ void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { "%sBlockIOAccounting=%s\n" "%sMemoryAccounting=%s\n" "%sTasksAccounting=%s\n" + "%sCPUWeight=%" PRIu64 "\n" + "%sStartupCPUWeight=%" PRIu64 "\n" "%sCPUShares=%" PRIu64 "\n" "%sStartupCPUShares=%" PRIu64 "\n" "%sCPUQuotaPerSecSec=%s\n" @@ -162,6 +183,8 @@ void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { prefix, yes_no(c->blockio_accounting), prefix, yes_no(c->memory_accounting), prefix, yes_no(c->tasks_accounting), + prefix, c->cpu_weight, + prefix, c->startup_cpu_weight, prefix, c->cpu_shares, prefix, c->startup_cpu_shares, prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1), @@ -367,6 +390,95 @@ fail: return -errno; } +static bool cgroup_context_has_cpu_weight(CGroupContext *c) { + return c->cpu_weight != CGROUP_WEIGHT_INVALID || + c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; +} + +static bool cgroup_context_has_cpu_shares(CGroupContext *c) { + return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || + c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; +} + +static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { + if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && + c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) + return c->startup_cpu_weight; + else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) + return c->cpu_weight; + else + return CGROUP_WEIGHT_DEFAULT; +} + +static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { + if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && + c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) + return c->startup_cpu_shares; + else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) + return c->cpu_shares; + else + return CGROUP_CPU_SHARES_DEFAULT; +} + +static void cgroup_apply_unified_cpu_config(Unit *u, uint64_t weight, uint64_t quota) { + char buf[MAX(DECIMAL_STR_MAX(uint64_t) + 1, (DECIMAL_STR_MAX(usec_t) + 1) * 2)]; + int r; + + xsprintf(buf, "%" PRIu64 "\n", weight); + r = cg_set_attribute("cpu", u->cgroup_path, "cpu.weight", buf); + if (r < 0) + log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, + "Failed to set cpu.weight: %m"); + + if (quota != USEC_INFINITY) + xsprintf(buf, USEC_FMT " " USEC_FMT "\n", + quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC); + else + xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); + + r = cg_set_attribute("cpu", u->cgroup_path, "cpu.max", buf); + + if (r < 0) + log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, + "Failed to set cpu.max: %m"); +} + +static void cgroup_apply_legacy_cpu_config(Unit *u, uint64_t shares, uint64_t quota) { + char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1]; + int r; + + xsprintf(buf, "%" PRIu64 "\n", shares); + r = cg_set_attribute("cpu", u->cgroup_path, "cpu.shares", buf); + if (r < 0) + log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, + "Failed to set cpu.shares: %m"); + + xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); + r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_period_us", buf); + if (r < 0) + log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, + "Failed to set cpu.cfs_period_us: %m"); + + if (quota != USEC_INFINITY) { + xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC); + r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", buf); + } else + r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", "-1"); + if (r < 0) + log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, + "Failed to set cpu.cfs_quota_us: %m"); +} + +static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { + return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, + CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); +} + +static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { + return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, + CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); +} + static bool cgroup_context_has_io_config(CGroupContext *c) { return c->io_accounting || c->io_weight != CGROUP_WEIGHT_INVALID || @@ -551,30 +663,42 @@ static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) { * and missing cgroups, i.e. EROFS and ENOENT. */ if ((mask & CGROUP_MASK_CPU) && !is_root) { - char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1]; + bool has_weight = cgroup_context_has_cpu_weight(c); + bool has_shares = cgroup_context_has_cpu_shares(c); - sprintf(buf, "%" PRIu64 "\n", - IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID ? c->startup_cpu_shares : - c->cpu_shares != CGROUP_CPU_SHARES_INVALID ? c->cpu_shares : CGROUP_CPU_SHARES_DEFAULT); - r = cg_set_attribute("cpu", path, "cpu.shares", buf); - if (r < 0) - log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, - "Failed to set cpu.shares: %m"); + if (cg_unified() > 0) { + uint64_t weight; - sprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); - r = cg_set_attribute("cpu", path, "cpu.cfs_period_us", buf); - if (r < 0) - log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, - "Failed to set cpu.cfs_period_us: %m"); + if (has_weight) + weight = cgroup_context_cpu_weight(c, state); + else if (has_shares) { + uint64_t shares = cgroup_context_cpu_shares(c, state); - if (c->cpu_quota_per_sec_usec != USEC_INFINITY) { - sprintf(buf, USEC_FMT "\n", c->cpu_quota_per_sec_usec * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC); - r = cg_set_attribute("cpu", path, "cpu.cfs_quota_us", buf); - } else - r = cg_set_attribute("cpu", path, "cpu.cfs_quota_us", "-1"); - if (r < 0) - log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, - "Failed to set cpu.cfs_quota_us: %m"); + weight = cgroup_cpu_shares_to_weight(shares); + + log_cgroup_compat(u, "Applying [Startup]CpuShares %" PRIu64 " as [Startup]CpuWeight %" PRIu64 " on %s", + shares, weight, path); + } else + weight = CGROUP_WEIGHT_DEFAULT; + + cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec); + } else { + uint64_t shares; + + if (has_shares) + shares = cgroup_context_cpu_shares(c, state); + else if (has_weight) { + uint64_t weight = cgroup_context_cpu_weight(c, state); + + shares = cgroup_cpu_weight_to_shares(weight); + + log_cgroup_compat(u, "Applying [Startup]CpuWeight %" PRIu64 " as [Startup]CpuShares %" PRIu64 " on %s", + weight, shares, path); + } else + shares = CGROUP_CPU_SHARES_DEFAULT; + + cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec); + } } if (mask & CGROUP_MASK_IO) { @@ -587,9 +711,14 @@ static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) { if (has_io) weight = cgroup_context_io_weight(c, state); - else if (has_blockio) - weight = cgroup_weight_blkio_to_io(cgroup_context_blkio_weight(c, state)); - else + else if (has_blockio) { + uint64_t blkio_weight = cgroup_context_blkio_weight(c, state); + + weight = cgroup_weight_blkio_to_io(blkio_weight); + + log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64, + blkio_weight, weight); + } else weight = CGROUP_WEIGHT_DEFAULT; xsprintf(buf, "default %" PRIu64 "\n", weight); @@ -608,8 +737,14 @@ static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) { CGroupBlockIODeviceWeight *w; /* FIXME: no way to reset this list */ - LIST_FOREACH(device_weights, w, c->blockio_device_weights) - cgroup_apply_io_device_weight(u, w->path, cgroup_weight_blkio_to_io(w->weight)); + LIST_FOREACH(device_weights, w, c->blockio_device_weights) { + weight = cgroup_weight_blkio_to_io(w->weight); + + log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s", + w->weight, weight, w->path); + + cgroup_apply_io_device_weight(u, w->path, weight); + } } } @@ -634,6 +769,9 @@ static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) { limits[CGROUP_IO_RBPS_MAX] = b->rbps; limits[CGROUP_IO_WBPS_MAX] = b->wbps; + log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s", + b->rbps, b->wbps, b->path); + if (!cgroup_apply_io_device_limit(u, b->path, limits)) cgroup_context_free_blockio_device_bandwidth(c, b); } @@ -650,9 +788,14 @@ static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) { if (has_blockio) weight = cgroup_context_blkio_weight(c, state); - else if (has_io) + else if (has_io) { + uint64_t io_weight = cgroup_context_io_weight(c, state); + weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); - else + + log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64, + io_weight, weight); + } else weight = CGROUP_BLKIO_WEIGHT_DEFAULT; xsprintf(buf, "%" PRIu64 "\n", weight); @@ -671,8 +814,14 @@ static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) { CGroupIODeviceWeight *w; /* FIXME: no way to reset this list */ - LIST_FOREACH(device_weights, w, c->io_device_weights) - cgroup_apply_blkio_device_weight(u, w->path, cgroup_weight_io_to_blkio(w->weight)); + LIST_FOREACH(device_weights, w, c->io_device_weights) { + weight = cgroup_weight_io_to_blkio(w->weight); + + log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s", + w->weight, weight, w->path); + + cgroup_apply_blkio_device_weight(u, w->path, weight); + } } } @@ -688,6 +837,9 @@ static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) { CGroupIODeviceLimit *l, *next; LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) { + log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s", + l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); + if (!cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX])) cgroup_context_free_io_device_limit(c, l); } @@ -700,19 +852,31 @@ static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) { if (cgroup_context_has_unified_memory_config(c)) max = c->memory_max; - else + else { max = c->memory_limit; + if (max != CGROUP_LIMIT_MAX) + log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max); + } + cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low); cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); cgroup_apply_unified_memory_limit(u, "memory.max", max); } else { char buf[DECIMAL_STR_MAX(uint64_t) + 1]; + uint64_t val = c->memory_limit; + + if (val == CGROUP_LIMIT_MAX) { + val = c->memory_max; + + if (val != CGROUP_LIMIT_MAX) + log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", c->memory_max); + } - if (c->memory_limit != CGROUP_LIMIT_MAX) - xsprintf(buf, "%" PRIu64 "\n", c->memory_limit); + if (val == CGROUP_LIMIT_MAX) + strncpy(buf, "-1\n", sizeof(buf)); else - xsprintf(buf, "%" PRIu64 "\n", c->memory_max); + xsprintf(buf, "%" PRIu64 "\n", val); r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf); if (r < 0) @@ -745,7 +909,11 @@ static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) { "/dev/random\0" "rwm\0" "/dev/urandom\0" "rwm\0" "/dev/tty\0" "rwm\0" - "/dev/pts/ptmx\0" "rw\0"; /* /dev/pts/ptmx may not be duplicated, but accessed */ + "/dev/pts/ptmx\0" "rw\0" /* /dev/pts/ptmx may not be duplicated, but accessed */ + /* Allow /run/elogind/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ + /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ + "/run/systemd/inaccessible/chr\0" "rwm\0" + "/run/systemd/inaccessible/blk\0" "rwm\0"; const char *x, *y; @@ -806,8 +974,8 @@ CGroupMask cgroup_context_get_mask(CGroupContext *c) { /* Figure out which controllers we need */ if (c->cpu_accounting || - c->cpu_shares != CGROUP_CPU_SHARES_INVALID || - c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID || + cgroup_context_has_cpu_weight(c) || + cgroup_context_has_cpu_shares(c) || c->cpu_quota_per_sec_usec != USEC_INFINITY) mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU; @@ -1608,7 +1776,7 @@ int manager_setup_cgroup(Manager *m) { /* 3. Install agent */ if (unified) { - /* In the unified hierarchy we can can get + /* In the unified hierarchy we can get * cgroup empty notifications via inotify. */ #if 0 /// elogind does not support the unified hierarchy, yet. @@ -1784,7 +1952,9 @@ Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { return manager_get_unit_by_pid_cgroup(m, pid); } +#endif // 0 +#if 0 /// elogind must substitute this with its own variant int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { Unit *u; @@ -1799,7 +1969,28 @@ int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { return unit_notify_cgroup_empty(u); } +#else +int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { + Session *s; + + assert(m); + assert(cgroup); + + log_debug("Got cgroup empty notification for: %s", cgroup); + s = hashmap_get(m->sessions, cgroup); + + if (s) { + session_finalize(s); + session_free(s); + } else + log_warning("Session not found: %s", cgroup); + + return 0; +} +#endif // 0 + +#if 0 /// UNNEEDED by elogind int unit_get_memory_current(Unit *u, uint64_t *ret) { _cleanup_free_ char *v = NULL; int r; @@ -1858,18 +2049,37 @@ static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { if (!u->cgroup_path) return -ENODATA; - if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0) - return -ENODATA; + if (cg_unified() > 0) { + const char *keys[] = { "usage_usec", NULL }; + _cleanup_free_ char *val = NULL; + uint64_t us; - r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); - if (r == -ENOENT) - return -ENODATA; - if (r < 0) - return r; + if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0) + return -ENODATA; - r = safe_atou64(v, &ns); - if (r < 0) - return r; + r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", keys, &val); + if (r < 0) + return r; + + r = safe_atou64(val, &us); + if (r < 0) + return r; + + ns = us * NSEC_PER_USEC; + } else { + if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0) + return -ENODATA; + + r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); + if (r == -ENOENT) + return -ENODATA; + if (r < 0) + return r; + + r = safe_atou64(v, &ns); + if (r < 0) + return r; + } *ret = ns; return 0; @@ -1883,8 +2093,8 @@ int unit_get_cpu_usage(Unit *u, nsec_t *ret) { if (r < 0) return r; - if (ns > u->cpuacct_usage_base) - ns -= u->cpuacct_usage_base; + if (ns > u->cpu_usage_base) + ns -= u->cpu_usage_base; else ns = 0; @@ -1900,11 +2110,11 @@ int unit_reset_cpu_usage(Unit *u) { r = unit_get_cpu_usage_raw(u, &ns); if (r < 0) { - u->cpuacct_usage_base = 0; + u->cpu_usage_base = 0; return r; } - u->cpuacct_usage_base = ns; + u->cpu_usage_base = ns; return 0; }