X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;ds=sidebyside;f=lib%2Fplanetlab.c;h=69f1fb07ab2614f50dc7f2a34b55ad64f35df323;hb=ec4370f7ebd7fb0ce7f002f5bf2c74f03acd3ec1;hp=fbed1ebb9ffce66a2e5c8eef177e9e7d99cd3745;hpb=6bbeb3e9837789f7813f1a38d39ac2e4df3a570b;p=util-vserver.git diff --git a/lib/planetlab.c b/lib/planetlab.c index fbed1eb..69f1fb0 100644 --- a/lib/planetlab.c +++ b/lib/planetlab.c @@ -31,33 +31,54 @@ POSSIBILITY OF SUCH DAMAGE. */ +#ifdef HAVE_CONFIG_H +# include +#endif +#include +#include +#include #include #include -#include +#include #include +#include #include +#include -#include "config.h" -#include "planetlab.h" -#include "sched_cmd.h" -#include "virtual.h" #include "vserver.h" +#include "planetlab.h" static int -create_context(xid_t ctx, uint32_t flags, uint64_t bcaps) +create_context(xid_t ctx, uint64_t bcaps, struct sliver_resources *slr) { struct vc_ctx_caps vc_caps; + struct vc_net_nx vc_net; + struct vc_net_flags vc_nf; + + /* Create network context */ + if (vc_net_create(ctx) == VC_NOCTX) + return -1; + + /* Make the network context persistent */ + vc_nf.mask = vc_nf.flagword = VC_NXF_PERSISTENT; + if (vc_set_nflags(ctx, &vc_nf)) + return -1; + + /* XXX: Allow access to all IPv4 addresses (for now) */ + vc_net.type = vcNET_IPV4; + vc_net.count = 1; + vc_net.ip[0] = 0; + vc_net.mask[0] = 0; + if (vc_net_add(ctx, &vc_net) == -1) + return -1; /* * Create context info - this sets the STATE_SETUP and STATE_INIT flags. - * Don't ever clear the STATE_INIT flag, that makes us the init task. - * - * XXX - the kernel code allows initial flags to be passed as an arg. */ if (vc_ctx_create(ctx) == VC_NOCTX) return -1; - /* set capabilities - these don't take effect until SETUP flag is unset */ + /* Set capabilities - these don't take effect until SETUP flag is unset */ vc_caps.bcaps = bcaps; vc_caps.bmask = ~0ULL; /* currently unused */ vc_caps.ccaps = 0; /* don't want any of these */ @@ -65,8 +86,7 @@ create_context(xid_t ctx, uint32_t flags, uint64_t bcaps) if (vc_set_ccaps(ctx, &vc_caps)) return -1; - /* set default scheduling parameters */ - pl_setsched(ctx, 32, 0); + pl_set_limits(ctx, slr); return 0; } @@ -77,8 +97,10 @@ pl_setup_done(xid_t ctx) struct vc_ctx_flags vc_flags; /* unset SETUP flag - this allows other processes to migrate */ - vc_flags.mask = VC_VXF_STATE_SETUP; - vc_flags.flagword = 0; + /* set the PERSISTENT flag - so the context doesn't vanish */ + /* Don't clear the STATE_INIT flag, as that would make us the init task. */ + vc_flags.mask = VC_VXF_STATE_SETUP|VC_VXF_PERSISTENT; + vc_flags.flagword = VC_VXF_PERSISTENT; if (vc_set_cflags(ctx, &vc_flags)) return -1; @@ -88,9 +110,10 @@ pl_setup_done(xid_t ctx) #define RETRY_LIMIT 10 int -pl_chcontext(xid_t ctx, uint32_t flags, uint64_t bcaps) +pl_chcontext(xid_t ctx, uint64_t bcaps, struct sliver_resources *slr) { int retry_count = 0; + int net_migrated = 0; for (;;) { @@ -102,7 +125,7 @@ pl_chcontext(xid_t ctx, uint32_t flags, uint64_t bcaps) return -1; /* context doesn't exist - create it */ - if (create_context(ctx, flags, bcaps)) + if (create_context(ctx, bcaps,slr)) { if (errno == EEXIST) /* another process beat us in a race */ @@ -132,8 +155,12 @@ pl_chcontext(xid_t ctx, uint32_t flags, uint64_t bcaps) /* context has been setup */ migrate: - if (!vc_ctx_migrate(ctx)) - break; /* done */ + if (net_migrated || !vc_net_migrate(ctx)) + { + if (!vc_ctx_migrate(ctx, 0)) + break; /* done */ + net_migrated = 1; + } /* context disappeared - retry */ } @@ -158,22 +185,32 @@ pl_setsched(xid_t ctx, uint32_t cpu_share, uint32_t cpu_sched_flags) uint32_t new_flags; vc_sched.set_mask = (VC_VXSM_FILL_RATE | VC_VXSM_INTERVAL | VC_VXSM_TOKENS | - VC_VXSM_TOKENS_MIN | VC_VXSM_TOKENS_MAX); - vc_sched.fill_rate = cpu_share; /* tokens accumulated per interval */ - vc_sched.interval = 1000; /* milliseconds */ + VC_VXSM_TOKENS_MIN | VC_VXSM_TOKENS_MAX | VC_VXSM_MSEC | + VC_VXSM_FILL_RATE2 | VC_VXSM_INTERVAL2 | VC_VXSM_FORCE | + VC_VXSM_IDLE_TIME); + vc_sched.fill_rate = 0; + vc_sched.fill_rate2 = cpu_share; /* tokens accumulated per interval */ + vc_sched.interval = vc_sched.interval2 = 1000; /* milliseconds */ vc_sched.tokens = 100; /* initial allocation of tokens */ vc_sched.tokens_min = 50; /* need this many tokens to run */ vc_sched.tokens_max = 100; /* max accumulated number of tokens */ + if (cpu_share == VC_LIM_KEEP) + vc_sched.set_mask &= ~(VC_VXSM_FILL_RATE|VC_VXSM_FILL_RATE2); + VC_SYSCALL(vc_set_sched(ctx, &vc_sched)); /* get current flag values */ VC_SYSCALL(vc_get_cflags(ctx, &vc_flags)); /* guaranteed CPU corresponds to SCHED_SHARE flag being cleared */ - new_flags = (cpu_sched_flags & VS_SCHED_CPU_GUARANTEED - ? 0 - : VC_VXF_SCHED_SHARE); + if (cpu_sched_flags & VS_SCHED_CPU_GUARANTEED) { + new_flags = VC_VXF_SCHED_SHARE; + vc_sched.fill_rate = vc_sched.fill_rate2; + } + else + new_flags = 0; + if ((vc_flags.flagword & VC_VXF_SCHED_SHARE) != new_flags) { vc_flags.mask = VC_VXF_SCHED_FLAGS; @@ -183,3 +220,253 @@ pl_setsched(xid_t ctx, uint32_t cpu_share, uint32_t cpu_sched_flags) return 0; } + +struct pl_resources { + char *name; + unsigned long long *limit; +}; + +#define WHITESPACE(buffer,index,len) \ + while(isspace((int)buffer[index])) \ + if (index < len) index++; else goto out; + +#define VSERVERCONF "/etc/vservers/" +void +pl_get_limits(char *context, struct sliver_resources *slr) +{ + FILE *fb; + int cwd; + size_t len = strlen(VSERVERCONF) + strlen(context) + NULLBYTE_SIZE; + char *conf = (char *)malloc(len + strlen("rlimits/openfd.hard")); + struct pl_resources *r; + struct pl_resources sliver_list[] = { + {"sched/fill-rate2", &slr->vs_cpu}, + {"sched/fill-rate", &slr->vs_cpuguaranteed}, + + {"rlimits/nproc.hard", &slr->vs_nproc.hard}, + {"rlimits/nproc.soft", &slr->vs_nproc.soft}, + {"rlimits/nproc.min", &slr->vs_nproc.min}, + + {"rlimits/rss.hard", &slr->vs_rss.hard}, + {"rlimits/rss.soft", &slr->vs_rss.soft}, + {"rlimits/rss.min", &slr->vs_rss.min}, + + {"rlimits/as.hard", &slr->vs_as.hard}, + {"rlimits/as.soft", &slr->vs_as.soft}, + {"rlimits/as.min", &slr->vs_as.min}, + + {"rlimits/openfd.hard", &slr->vs_openfd.hard}, + {"rlimits/openfd.soft", &slr->vs_openfd.soft}, + {"rlimits/openfd.min", &slr->vs_openfd.min}, + + {"whitelisted", &slr->vs_whitelisted}, + {0,0} + }; + + sprintf(conf, "%s%s", VSERVERCONF, context); + + slr->vs_cpu = VC_LIM_KEEP; + slr->vs_cpuguaranteed = 0; + + slr->vs_rss.hard = VC_LIM_KEEP; + slr->vs_rss.soft = VC_LIM_KEEP; + slr->vs_rss.min = VC_LIM_KEEP; + + slr->vs_as.hard = VC_LIM_KEEP; + slr->vs_as.soft = VC_LIM_KEEP; + slr->vs_as.min = VC_LIM_KEEP; + + + slr->vs_nproc.hard = VC_LIM_KEEP; + slr->vs_nproc.soft = VC_LIM_KEEP; + slr->vs_nproc.min = VC_LIM_KEEP; + + slr->vs_openfd.hard = VC_LIM_KEEP; + slr->vs_openfd.soft = VC_LIM_KEEP; + slr->vs_openfd.min = VC_LIM_KEEP; + + slr->vs_whitelisted = 1; + + cwd = open(".", O_RDONLY); + if (cwd == -1) { + perror("cannot get a handle on ."); + goto out; + } + if (chdir(conf) == -1) { + fprintf(stderr, "cannot chdir to "); + perror(conf); + goto out_fd; + } + + for (r = &sliver_list[0]; r->name; r++) { + char buf[1000]; + fb = fopen(r->name, "r"); + if (fb == NULL) + continue; + if (fgets(buf, sizeof(buf), fb) != NULL && isdigit(*buf)) { + *r->limit = atoi(buf); + } + fclose(fb); + } + + fchdir(cwd); +out_fd: + close(cwd); +out: +#if 0 + /* open the conf file for reading */ + fb = fopen(conf,"r"); + if (fb != NULL) { + size_t index; + char *buffer = malloc(1000); + char *p; + + /* the conf file exist */ + while((p=fgets(buffer,1000-1,fb))!=NULL) { + index = 0; + len = strnlen(buffer,1000); + WHITESPACE(buffer,index,len); + if (buffer[index] == '#') + continue; + + for (r=&sliver_list[0]; r->name; r++) + if ((p=strstr(&buffer[index],r->name))!=NULL) { + /* adjust index into buffer */ + index+= (p-&buffer[index])+strlen(r->name); + + /* skip over whitespace */ + WHITESPACE(buffer,index,len); + + /* expecting to see = sign */ + if (buffer[index++]!='=') goto out; + + /* skip over whitespace */ + WHITESPACE(buffer,index,len); + + /* expecting to see a digit for number */ + if (!isdigit((int)buffer[index])) goto out; + + *r->limit = atoi(&buffer[index]); + if (0) /* for debugging only */ + fprintf(stderr,"pl_get_limits found %s=%lld\n", + r->name,*r->limit); + break; + } + } + out: + fclose(fb); + free(buffer); + } else { + fprintf(stderr,"cannot open %s\n",conf); + } +#endif + free(conf); +} + +int +adjust_lim(struct vc_rlimit *vcr, struct rlimit *lim) +{ + int adjusted = 0; + if (vcr->min != VC_LIM_KEEP) { + if (vcr->min > lim->rlim_cur) { + lim->rlim_cur = vcr->min; + adjusted = 1; + } + if (vcr->min > lim->rlim_max) { + lim->rlim_max = vcr->min; + adjusted = 1; + } + } + + if (vcr->soft != VC_LIM_KEEP) { + switch (vcr->min != VC_LIM_KEEP) { + case 1: + if (vcr->soft < vcr->min) + break; + case 0: + lim->rlim_cur = vcr->soft; + adjusted = 1; + } + } + + if (vcr->hard != VC_LIM_KEEP) { + switch (vcr->min != VC_LIM_KEEP) { + case 1: + if (vcr->hard < vcr->min) + break; + case 0: + lim->rlim_cur = vcr->hard; + adjusted = 1; + } + } + return adjusted; +} + + +void +pl_set_limits(xid_t ctx, struct sliver_resources *slr) +{ + struct rlimit lim; /* getrlimit values */ + unsigned long long vs_cpu; + uint32_t cpu_sched_flags; + + if (slr != 0) { + /* set memory limits */ + getrlimit(RLIMIT_RSS,&lim); + if (adjust_lim(&slr->vs_rss, &lim)) { + setrlimit(RLIMIT_RSS, &lim); + if (vc_set_rlimit(ctx, RLIMIT_RSS, &slr->vs_rss)) + { + PERROR("pl_setrlimit(%u, RLIMIT_RSS)", ctx); + exit(1); + } + } + + /* set address space limits */ + getrlimit(RLIMIT_AS,&lim); + if (adjust_lim(&slr->vs_as, &lim)) { + setrlimit(RLIMIT_AS, &lim); + if (vc_set_rlimit(ctx, RLIMIT_AS, &slr->vs_as)) + { + PERROR("pl_setrlimit(%u, RLIMIT_AS)", ctx); + exit(1); + } + } + /* set nrpoc limit */ + getrlimit(RLIMIT_NPROC,&lim); + if (adjust_lim(&slr->vs_nproc, &lim)) { + setrlimit(RLIMIT_NPROC, &lim); + if (vc_set_rlimit(ctx, RLIMIT_NPROC, &slr->vs_nproc)) + { + PERROR("pl_setrlimit(%u, RLIMIT_NPROC)", ctx); + exit(1); + } + } + + /* set openfd limit */ + getrlimit(RLIMIT_NOFILE,&lim); + if (adjust_lim(&slr->vs_openfd, &lim)) { + setrlimit(RLIMIT_NOFILE, &lim); + if (vc_set_rlimit(ctx, RLIMIT_NOFILE, &slr->vs_openfd)) + { + PERROR("pl_setrlimit(%u, RLIMIT_NOFILE)", ctx); + exit(1); + } + if (vc_set_rlimit(ctx, VC_VLIMIT_OPENFD, &slr->vs_openfd)) + { + PERROR("pl_setrlimit(%u, VLIMIT_OPENFD)", ctx); + exit(1); + } + } + vs_cpu = slr->vs_cpu; + cpu_sched_flags = slr->vs_cpuguaranteed & VS_SCHED_CPU_GUARANTEED; + } else { + vs_cpu = 1; + cpu_sched_flags = 0; + } + + if (pl_setsched(ctx, vs_cpu, cpu_sched_flags) < 0) { + PERROR("pl_setsched(&u)", ctx); + exit(1); + } +}