summaryrefslogtreecommitdiff
path: root/usr/src/uts/common/vm
diff options
context:
space:
mode:
Diffstat (limited to 'usr/src/uts/common/vm')
-rw-r--r--usr/src/uts/common/vm/hat.h2
-rw-r--r--usr/src/uts/common/vm/seg_vn.c12
-rw-r--r--usr/src/uts/common/vm/vm_anon.c2
-rw-r--r--usr/src/uts/common/vm/vm_as.c20
-rw-r--r--usr/src/uts/common/vm/vm_pvn.c28
-rw-r--r--usr/src/uts/common/vm/vm_usage.c198
6 files changed, 201 insertions, 61 deletions
diff --git a/usr/src/uts/common/vm/hat.h b/usr/src/uts/common/vm/hat.h
index 1d91475e38..156b810046 100644
--- a/usr/src/uts/common/vm/hat.h
+++ b/usr/src/uts/common/vm/hat.h
@@ -21,6 +21,7 @@
/*
* Copyright 2009 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
+ * Copyright 2012 Joyent, Inc. All rights reserved.
*/
/* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
@@ -460,6 +461,7 @@ void hat_setstat(struct as *, caddr_t, size_t, uint_t);
*/
#define HAT_ADV_PGUNLOAD 0x00
#define HAT_FORCE_PGUNLOAD 0x01
+#define HAT_CURPROC_PGUNLOAD 0x02
/*
* Attributes for hat_page_*attr, hat_setstats and
diff --git a/usr/src/uts/common/vm/seg_vn.c b/usr/src/uts/common/vm/seg_vn.c
index 31c293d416..5f106f6c06 100644
--- a/usr/src/uts/common/vm/seg_vn.c
+++ b/usr/src/uts/common/vm/seg_vn.c
@@ -20,6 +20,7 @@
*/
/*
* Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2012, Joyent, Inc. All rights reserved.
*/
/* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
@@ -7254,7 +7255,8 @@ segvn_sync(struct seg *seg, caddr_t addr, size_t len, int attr, uint_t flags)
vpp = svd->vpage;
offset = svd->offset + (uintptr_t)(addr - seg->s_base);
bflags = ((flags & MS_ASYNC) ? B_ASYNC : 0) |
- ((flags & MS_INVALIDATE) ? B_INVAL : 0);
+ ((flags & MS_INVALIDATE) ? B_INVAL : 0) |
+ ((flags & MS_INVALCURPROC) ? (B_INVALCURONLY | B_INVAL) : 0);
if (attr) {
pageprot = attr & ~(SHARED|PRIVATE);
@@ -7279,11 +7281,11 @@ segvn_sync(struct seg *seg, caddr_t addr, size_t len, int attr, uint_t flags)
vpp = &svd->vpage[seg_page(seg, addr)];
} else if (svd->vp && svd->amp == NULL &&
- (flags & MS_INVALIDATE) == 0) {
+ (flags & (MS_INVALIDATE | MS_INVALCURPROC)) == 0) {
/*
- * No attributes, no anonymous pages and MS_INVALIDATE flag
- * is not on, just use one big request.
+ * No attributes, no anonymous pages and MS_INVAL* flags
+ * are not on, just use one big request.
*/
err = VOP_PUTPAGE(svd->vp, (offset_t)offset, len,
bflags, svd->cred, NULL);
@@ -7335,7 +7337,7 @@ segvn_sync(struct seg *seg, caddr_t addr, size_t len, int attr, uint_t flags)
* might race in and lock the page after we unlock and before
* we do the PUTPAGE, then PUTPAGE simply does nothing.
*/
- if (flags & MS_INVALIDATE) {
+ if (flags & (MS_INVALIDATE | MS_INVALCURPROC)) {
if ((pp = page_lookup(vp, off, SE_SHARED)) != NULL) {
if (pp->p_lckcnt != 0 || pp->p_cowcnt != 0) {
page_unlock(pp);
diff --git a/usr/src/uts/common/vm/vm_anon.c b/usr/src/uts/common/vm/vm_anon.c
index fdf9f7790c..f30ba7ef2e 100644
--- a/usr/src/uts/common/vm/vm_anon.c
+++ b/usr/src/uts/common/vm/vm_anon.c
@@ -20,6 +20,7 @@
*/
/*
* Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1986, 2010, Joyent, Inc. All rights reserved.
*/
/* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
@@ -792,6 +793,7 @@ anon_resvmem(size_t size, boolean_t takemem, zone_t *zone, int tryhard)
mutex_enter(&p->p_lock);
if (rctl_incr_swap(p, zone, ptob(npages)) != 0) {
mutex_exit(&p->p_lock);
+ atomic_add_64(&zone->zone_anon_alloc_fail, 1);
return (0);
}
mutex_exit(&p->p_lock);
diff --git a/usr/src/uts/common/vm/vm_as.c b/usr/src/uts/common/vm/vm_as.c
index 01ad32e0b1..8caa257486 100644
--- a/usr/src/uts/common/vm/vm_as.c
+++ b/usr/src/uts/common/vm/vm_as.c
@@ -21,6 +21,7 @@
/*
* Copyright 2010 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
+ * Copyright 2012, Joyent, Inc. All rights reserved.
*/
/* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
@@ -56,6 +57,7 @@
#include <sys/debug.h>
#include <sys/tnf_probe.h>
#include <sys/vtrace.h>
+#include <sys/ddi.h>
#include <vm/hat.h>
#include <vm/xhat.h>
@@ -879,6 +881,7 @@ as_fault(struct hat *hat, struct as *as, caddr_t addr, size_t size,
struct seg *segsav;
int as_lock_held;
klwp_t *lwp = ttolwp(curthread);
+ zone_t *zonep = curzone;
int is_xhat = 0;
int holding_wpage = 0;
extern struct seg_ops segdev_ops;
@@ -928,6 +931,23 @@ retry:
if (as == &kas)
CPU_STATS_ADDQ(CPU, vm, kernel_asflt, 1);
CPU_STATS_EXIT_K();
+ if (zonep->zone_pg_flt_delay != 0) {
+ /*
+ * The zone in which this process is running
+ * is currently over it's physical memory cap.
+ * Throttle page faults to help the user-land
+ * memory capper catch up. Note that
+ * drv_usectohz() rounds up.
+ */
+ atomic_add_64(&zonep->zone_pf_throttle, 1);
+ atomic_add_64(&zonep->zone_pf_throttle_usec,
+ zonep->zone_pg_flt_delay);
+ if (zonep->zone_pg_flt_delay < TICK_TO_USEC(1))
+ drv_usecwait(zonep->zone_pg_flt_delay);
+ else
+ delay(drv_usectohz(
+ zonep->zone_pg_flt_delay));
+ }
break;
}
}
diff --git a/usr/src/uts/common/vm/vm_pvn.c b/usr/src/uts/common/vm/vm_pvn.c
index 7233581227..39ace0b3c2 100644
--- a/usr/src/uts/common/vm/vm_pvn.c
+++ b/usr/src/uts/common/vm/vm_pvn.c
@@ -20,6 +20,7 @@
*/
/*
* Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, Joyent, Inc. All rights reserved.
*/
/* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
@@ -431,7 +432,14 @@ pvn_write_done(page_t *plist, int flags)
page_io_unlock(pp);
page_unlock(pp);
}
- } else if (flags & B_INVAL) {
+ } else if ((flags & (B_INVAL | B_INVALCURONLY)) == B_INVAL) {
+ /*
+ * If B_INVALCURONLY is set, then we handle that case
+ * in the next conditional if hat_page_is_mapped()
+ * indicates that there are no additional mappings
+ * to the page.
+ */
+
/*
* XXX - Failed writes with B_INVAL set are
* not handled appropriately.
@@ -572,8 +580,9 @@ pvn_write_done(page_t *plist, int flags)
}
/*
- * Flags are composed of {B_ASYNC, B_INVAL, B_FREE, B_DONTNEED, B_DELWRI,
- * B_TRUNC, B_FORCE}. B_DELWRI indicates that this page is part of a kluster
+ * Flags are composed of {B_ASYNC, B_INVAL, B_INVALCURONLY, B_FREE,
+ * B_DONTNEED, B_DELWRI, B_TRUNC, B_FORCE}.
+ * B_DELWRI indicates that this page is part of a kluster
* operation and is only to be considered if it doesn't involve any
* waiting here. B_TRUNC indicates that the file is being truncated
* and so no i/o needs to be done. B_FORCE indicates that the page
@@ -627,13 +636,17 @@ pvn_getdirty(page_t *pp, int flags)
* If we want to free or invalidate the page then
* we need to unload it so that anyone who wants
* it will have to take a minor fault to get it.
+ * If we are only invalidating the page for the
+ * current process, then pass in a different flag.
* Otherwise, we're just writing the page back so we
* need to sync up the hardwre and software mod bit to
* detect any future modifications. We clear the
* software mod bit when we put the page on the dirty
* list.
*/
- if (flags & (B_INVAL | B_FREE)) {
+ if (flags & B_INVALCURONLY) {
+ (void) hat_pageunload(pp, HAT_CURPROC_PGUNLOAD);
+ } else if (flags & (B_INVAL | B_FREE)) {
(void) hat_pageunload(pp, HAT_FORCE_PGUNLOAD);
} else {
(void) hat_pagesync(pp, HAT_SYNC_ZERORM);
@@ -645,7 +658,7 @@ pvn_getdirty(page_t *pp, int flags)
* list after all.
*/
page_io_unlock(pp);
- if (flags & B_INVAL) {
+ if ((flags & (B_INVAL | B_INVALCURONLY)) == B_INVAL) {
/*LINTED: constant in conditional context*/
VN_DISPOSE(pp, B_INVAL, 0, kcred);
} else if (flags & B_FREE) {
@@ -657,6 +670,9 @@ pvn_getdirty(page_t *pp, int flags)
* of VOP_PUTPAGE() who prefer freeing the
* page _only_ if no one else is accessing it.
* E.g. segmap_release()
+ * We also take this path for B_INVALCURONLY and
+ * let page_release call VN_DISPOSE if no one else is
+ * using the page.
*
* The above hat_ismod() check is useless because:
* (1) we may not be holding SE_EXCL lock;
@@ -681,7 +697,7 @@ pvn_getdirty(page_t *pp, int flags)
* We'll detect the fact that they used it when the
* i/o is done and avoid freeing the page.
*/
- if (flags & B_FREE)
+ if (flags & (B_FREE | B_INVALCURONLY))
page_downgrade(pp);
diff --git a/usr/src/uts/common/vm/vm_usage.c b/usr/src/uts/common/vm/vm_usage.c
index d422f8d0e8..8f425e9e4f 100644
--- a/usr/src/uts/common/vm/vm_usage.c
+++ b/usr/src/uts/common/vm/vm_usage.c
@@ -25,6 +25,10 @@
*/
/*
+ * Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ */
+
+/*
* vm_usage
*
* This file implements the getvmusage() private system call.
@@ -114,7 +118,7 @@
* For accurate counting of map-shared and COW-shared pages.
*
* - visited private anons (refcnt > 1) for each collective.
- * (entity->vme_anon_hash)
+ * (entity->vme_anon)
* For accurate counting of COW-shared pages.
*
* The common accounting structure is the vmu_entity_t, which represents
@@ -152,6 +156,7 @@
#include <sys/vm_usage.h>
#include <sys/zone.h>
#include <sys/sunddi.h>
+#include <sys/sysmacros.h>
#include <sys/avl.h>
#include <vm/anon.h>
#include <vm/as.h>
@@ -199,6 +204,14 @@ typedef struct vmu_object {
} vmu_object_t;
/*
+ * Node for tree of visited COW anons.
+ */
+typedef struct vmu_anon {
+ avl_node_t vma_node;
+ uintptr_t vma_addr;
+} vmu_anon_t;
+
+/*
* Entity by which to count results.
*
* The entity structure keeps the current rss/swap counts for each entity
@@ -221,7 +234,7 @@ typedef struct vmu_entity {
struct vmu_entity *vme_next_calc;
mod_hash_t *vme_vnode_hash; /* vnodes visited for entity */
mod_hash_t *vme_amp_hash; /* shared amps visited for entity */
- mod_hash_t *vme_anon_hash; /* COW anons visited for entity */
+ avl_tree_t vme_anon; /* COW anons visited for entity */
vmusage_t vme_result; /* identifies entity and results */
} vmu_entity_t;
@@ -324,6 +337,23 @@ bounds_cmp(const void *bnd1, const void *bnd2)
}
/*
+ * Comparison routine for our AVL tree of anon structures.
+ */
+static int
+vmu_anon_cmp(const void *lhs, const void *rhs)
+{
+ const vmu_anon_t *l = lhs, *r = rhs;
+
+ if (l->vma_addr == r->vma_addr)
+ return (0);
+
+ if (l->vma_addr < r->vma_addr)
+ return (-1);
+
+ return (1);
+}
+
+/*
* Save a bound on the free list.
*/
static void
@@ -363,13 +393,18 @@ static void
vmu_free_entity(mod_hash_val_t val)
{
vmu_entity_t *entity = (vmu_entity_t *)val;
+ vmu_anon_t *anon;
+ void *cookie = NULL;
if (entity->vme_vnode_hash != NULL)
i_mod_hash_clear_nosync(entity->vme_vnode_hash);
if (entity->vme_amp_hash != NULL)
i_mod_hash_clear_nosync(entity->vme_amp_hash);
- if (entity->vme_anon_hash != NULL)
- i_mod_hash_clear_nosync(entity->vme_anon_hash);
+
+ while ((anon = avl_destroy_nodes(&entity->vme_anon, &cookie)) != NULL)
+ kmem_free(anon, sizeof (vmu_anon_t));
+
+ avl_destroy(&entity->vme_anon);
entity->vme_next = vmu_data.vmu_free_entities;
vmu_data.vmu_free_entities = entity;
@@ -485,10 +520,10 @@ vmu_alloc_entity(id_t id, int type, id_t zoneid)
"vmusage amp hash", VMUSAGE_HASH_SIZE, vmu_free_object,
sizeof (struct anon_map));
- if (entity->vme_anon_hash == NULL)
- entity->vme_anon_hash = mod_hash_create_ptrhash(
- "vmusage anon hash", VMUSAGE_HASH_SIZE,
- mod_hash_null_valdtor, sizeof (struct anon));
+ VERIFY(avl_first(&entity->vme_anon) == NULL);
+
+ avl_create(&entity->vme_anon, vmu_anon_cmp, sizeof (struct vmu_anon),
+ offsetof(struct vmu_anon, vma_node));
entity->vme_next = vmu_data.vmu_entities;
vmu_data.vmu_entities = entity;
@@ -518,7 +553,8 @@ vmu_alloc_zone(id_t id)
zone->vmz_id = id;
- if ((vmu_data.vmu_calc_flags & (VMUSAGE_ZONE | VMUSAGE_ALL_ZONES)) != 0)
+ if ((vmu_data.vmu_calc_flags &
+ (VMUSAGE_ZONE | VMUSAGE_ALL_ZONES | VMUSAGE_A_ZONE)) != 0)
zone->vmz_zone = vmu_alloc_entity(id, VMUSAGE_ZONE, id);
if ((vmu_data.vmu_calc_flags & (VMUSAGE_PROJECTS |
@@ -613,21 +649,19 @@ vmu_find_insert_object(mod_hash_t *hash, caddr_t key, uint_t type)
}
static int
-vmu_find_insert_anon(mod_hash_t *hash, caddr_t key)
+vmu_find_insert_anon(vmu_entity_t *entity, void *key)
{
- int ret;
- caddr_t val;
+ vmu_anon_t anon, *ap;
- ret = i_mod_hash_find_nosync(hash, (mod_hash_key_t)key,
- (mod_hash_val_t *)&val);
+ anon.vma_addr = (uintptr_t)key;
- if (ret == 0)
+ if (avl_find(&entity->vme_anon, &anon, NULL) != NULL)
return (0);
- ret = i_mod_hash_insert_nosync(hash, (mod_hash_key_t)key,
- (mod_hash_val_t)key, (mod_hash_hndl_t)0);
+ ap = kmem_alloc(sizeof (vmu_anon_t), KM_SLEEP);
+ ap->vma_addr = (uintptr_t)key;
- ASSERT(ret == 0);
+ avl_add(&entity->vme_anon, ap);
return (1);
}
@@ -937,7 +971,10 @@ vmu_amp_update_incore_bounds(avl_tree_t *tree, struct anon_map *amp,
if (ap != NULL && vn != NULL && vn->v_pages != NULL &&
(page = page_exists(vn, off)) != NULL) {
- page_type = VMUSAGE_BOUND_INCORE;
+ if (PP_ISFREE(page))
+ page_type = VMUSAGE_BOUND_NOT_INCORE;
+ else
+ page_type = VMUSAGE_BOUND_INCORE;
if (page->p_szc > 0) {
pgcnt = page_get_pagecnt(page->p_szc);
pgshft = page_get_shift(page->p_szc);
@@ -1024,7 +1061,10 @@ vmu_vnode_update_incore_bounds(avl_tree_t *tree, vnode_t *vnode,
if (vnode->v_pages != NULL &&
(page = page_exists(vnode, ptob(index))) != NULL) {
- page_type = VMUSAGE_BOUND_INCORE;
+ if (PP_ISFREE(page))
+ page_type = VMUSAGE_BOUND_NOT_INCORE;
+ else
+ page_type = VMUSAGE_BOUND_INCORE;
if (page->p_szc > 0) {
pgcnt = page_get_pagecnt(page->p_szc);
pgshft = page_get_shift(page->p_szc);
@@ -1304,6 +1344,12 @@ vmu_calculate_seg(vmu_entity_t *vmu_entities, struct seg *seg)
}
/*
+ * Pages on the free list aren't counted for the rss.
+ */
+ if (PP_ISFREE(page))
+ continue;
+
+ /*
* Assume anon structs with a refcnt
* of 1 are not COW shared, so there
* is no reason to track them per entity.
@@ -1320,8 +1366,7 @@ vmu_calculate_seg(vmu_entity_t *vmu_entities, struct seg *seg)
* Track COW anons per entity so
* they are not double counted.
*/
- if (vmu_find_insert_anon(entity->vme_anon_hash,
- (caddr_t)ap) == 0)
+ if (vmu_find_insert_anon(entity, ap) == 0)
continue;
result->vmu_rss_all += (pgcnt << PAGESHIFT);
@@ -1461,8 +1506,9 @@ vmu_calculate_proc(proc_t *p)
entities = tmp;
}
if (vmu_data.vmu_calc_flags &
- (VMUSAGE_ZONE | VMUSAGE_ALL_ZONES | VMUSAGE_PROJECTS |
- VMUSAGE_ALL_PROJECTS | VMUSAGE_TASKS | VMUSAGE_ALL_TASKS |
+ (VMUSAGE_ZONE | VMUSAGE_ALL_ZONES | VMUSAGE_A_ZONE |
+ VMUSAGE_PROJECTS | VMUSAGE_ALL_PROJECTS |
+ VMUSAGE_TASKS | VMUSAGE_ALL_TASKS |
VMUSAGE_RUSERS | VMUSAGE_ALL_RUSERS | VMUSAGE_EUSERS |
VMUSAGE_ALL_EUSERS)) {
ret = i_mod_hash_find_nosync(vmu_data.vmu_zones_hash,
@@ -1594,8 +1640,7 @@ vmu_free_extra()
mod_hash_destroy_hash(te->vme_vnode_hash);
if (te->vme_amp_hash != NULL)
mod_hash_destroy_hash(te->vme_amp_hash);
- if (te->vme_anon_hash != NULL)
- mod_hash_destroy_hash(te->vme_anon_hash);
+ VERIFY(avl_first(&te->vme_anon) == NULL);
kmem_free(te, sizeof (vmu_entity_t));
}
while (vmu_data.vmu_free_zones != NULL) {
@@ -1739,12 +1784,34 @@ vmu_cache_rele(vmu_cache_t *cache)
}
/*
+ * When new data is calculated, update the phys_mem rctl usage value in the
+ * zones.
+ */
+static void
+vmu_update_zone_rctls(vmu_cache_t *cache)
+{
+ vmusage_t *rp;
+ size_t i = 0;
+ zone_t *zp;
+
+ for (rp = cache->vmc_results; i < cache->vmc_nresults; rp++, i++) {
+ if (rp->vmu_type == VMUSAGE_ZONE &&
+ rp->vmu_zoneid != ALL_ZONES) {
+ if ((zp = zone_find_by_id(rp->vmu_zoneid)) != NULL) {
+ zp->zone_phys_mem = rp->vmu_rss_all;
+ zone_rele(zp);
+ }
+ }
+ }
+}
+
+/*
* Copy out the cached results to a caller. Inspect the callers flags
* and zone to determine which cached results should be copied.
*/
static int
vmu_copyout_results(vmu_cache_t *cache, vmusage_t *buf, size_t *nres,
- uint_t flags, int cpflg)
+ uint_t flags, id_t req_zone_id, int cpflg)
{
vmusage_t *result, *out_result;
vmusage_t dummy;
@@ -1763,7 +1830,7 @@ vmu_copyout_results(vmu_cache_t *cache, vmusage_t *buf, size_t *nres,
/* figure out what results the caller is interested in. */
if ((flags & VMUSAGE_SYSTEM) && curproc->p_zone == global_zone)
types |= VMUSAGE_SYSTEM;
- if (flags & (VMUSAGE_ZONE | VMUSAGE_ALL_ZONES))
+ if (flags & (VMUSAGE_ZONE | VMUSAGE_ALL_ZONES | VMUSAGE_A_ZONE))
types |= VMUSAGE_ZONE;
if (flags & (VMUSAGE_PROJECTS | VMUSAGE_ALL_PROJECTS |
VMUSAGE_COL_PROJECTS))
@@ -1826,26 +1893,33 @@ vmu_copyout_results(vmu_cache_t *cache, vmusage_t *buf, size_t *nres,
continue;
}
- /* Skip "other zone" results if not requested */
- if (result->vmu_zoneid != curproc->p_zone->zone_id) {
- if (result->vmu_type == VMUSAGE_ZONE &&
- (flags & VMUSAGE_ALL_ZONES) == 0)
- continue;
- if (result->vmu_type == VMUSAGE_PROJECTS &&
- (flags & (VMUSAGE_ALL_PROJECTS |
- VMUSAGE_COL_PROJECTS)) == 0)
- continue;
- if (result->vmu_type == VMUSAGE_TASKS &&
- (flags & VMUSAGE_ALL_TASKS) == 0)
- continue;
- if (result->vmu_type == VMUSAGE_RUSERS &&
- (flags & (VMUSAGE_ALL_RUSERS |
- VMUSAGE_COL_RUSERS)) == 0)
- continue;
- if (result->vmu_type == VMUSAGE_EUSERS &&
- (flags & (VMUSAGE_ALL_EUSERS |
- VMUSAGE_COL_EUSERS)) == 0)
+ if (result->vmu_type == VMUSAGE_ZONE &&
+ flags & VMUSAGE_A_ZONE) {
+ /* Skip non-requested zone results */
+ if (result->vmu_zoneid != req_zone_id)
continue;
+ } else {
+ /* Skip "other zone" results if not requested */
+ if (result->vmu_zoneid != curproc->p_zone->zone_id) {
+ if (result->vmu_type == VMUSAGE_ZONE &&
+ (flags & VMUSAGE_ALL_ZONES) == 0)
+ continue;
+ if (result->vmu_type == VMUSAGE_PROJECTS &&
+ (flags & (VMUSAGE_ALL_PROJECTS |
+ VMUSAGE_COL_PROJECTS)) == 0)
+ continue;
+ if (result->vmu_type == VMUSAGE_TASKS &&
+ (flags & VMUSAGE_ALL_TASKS) == 0)
+ continue;
+ if (result->vmu_type == VMUSAGE_RUSERS &&
+ (flags & (VMUSAGE_ALL_RUSERS |
+ VMUSAGE_COL_RUSERS)) == 0)
+ continue;
+ if (result->vmu_type == VMUSAGE_EUSERS &&
+ (flags & (VMUSAGE_ALL_EUSERS |
+ VMUSAGE_COL_EUSERS)) == 0)
+ continue;
+ }
}
count++;
if (out_result != NULL) {
@@ -1901,10 +1975,12 @@ vm_getusage(uint_t flags, time_t age, vmusage_t *buf, size_t *nres, int cpflg)
int cacherecent = 0;
hrtime_t now;
uint_t flags_orig;
+ id_t req_zone_id;
/*
* Non-global zones cannot request system wide and/or collated
- * results, or the system result, so munge the flags accordingly.
+ * results, or the system result, or usage of another zone, so munge
+ * the flags accordingly.
*/
flags_orig = flags;
if (curproc->p_zone != global_zone) {
@@ -1924,6 +2000,10 @@ vm_getusage(uint_t flags, time_t age, vmusage_t *buf, size_t *nres, int cpflg)
flags &= ~VMUSAGE_SYSTEM;
flags |= VMUSAGE_ZONE;
}
+ if (flags & VMUSAGE_A_ZONE) {
+ flags &= ~VMUSAGE_A_ZONE;
+ flags |= VMUSAGE_ZONE;
+ }
}
/* Check for unknown flags */
@@ -1934,6 +2014,21 @@ vm_getusage(uint_t flags, time_t age, vmusage_t *buf, size_t *nres, int cpflg)
if ((flags & VMUSAGE_MASK) == 0)
return (set_errno(EINVAL));
+ /* If requesting results for a specific zone, get the zone ID */
+ if (flags & VMUSAGE_A_ZONE) {
+ size_t bufsize;
+ vmusage_t zreq;
+
+ if (ddi_copyin((caddr_t)nres, &bufsize, sizeof (size_t), cpflg))
+ return (set_errno(EFAULT));
+ /* Requested zone ID is passed in buf, so 0 len not allowed */
+ if (bufsize == 0)
+ return (set_errno(EINVAL));
+ if (ddi_copyin((caddr_t)buf, &zreq, sizeof (vmusage_t), cpflg))
+ return (set_errno(EFAULT));
+ req_zone_id = zreq.vmu_id;
+ }
+
mutex_enter(&vmu_data.vmu_lock);
now = gethrtime();
@@ -1953,7 +2048,7 @@ start:
mutex_exit(&vmu_data.vmu_lock);
ret = vmu_copyout_results(cache, buf, nres, flags_orig,
- cpflg);
+ req_zone_id, cpflg);
mutex_enter(&vmu_data.vmu_lock);
vmu_cache_rele(cache);
if (vmu_data.vmu_pending_waiters > 0)
@@ -2009,8 +2104,11 @@ start:
mutex_exit(&vmu_data.vmu_lock);
+ /* update zone's phys. mem. rctl usage */
+ vmu_update_zone_rctls(cache);
/* copy cache */
- ret = vmu_copyout_results(cache, buf, nres, flags_orig, cpflg);
+ ret = vmu_copyout_results(cache, buf, nres, flags_orig,
+ req_zone_id, cpflg);
mutex_enter(&vmu_data.vmu_lock);
vmu_cache_rele(cache);
mutex_exit(&vmu_data.vmu_lock);