Skip to contentSkip to conversation

Coverity fixes for libxl

PATCH In review Matthew Daley 5 participants 75 messages Dec 1, 2013 – Jan 9, 2014

lore
Share & tools
lore
#0 lore raw

cover Coverity fixes for libxl

Matthew Daley (13):
libxl: fix unsigned less-than-0 comparison in e820_sanitize
libxl: check for xc_domain_setmaxmem failure in libxl__build_pre
libxl: correct file open success check in libxl__device_pci_reset
libxl: don't leak p in libxl__wait_for_backend
libxl: remove unsigned less-than-0 comparison
libxl: actually abort if initializing a ctx's lock fails
libxl: don't leak output vcpu info on error in libxl_list_vcpu
libxl: don't leak ptr in libxl_list_vm error case
libxl: don't leak pcidevs in libxl_pcidev_assignable
libxl: don't try to fclose file twice on error in
libxl_userdata_store
libxl: use pipe instead of temporary file for VNC viewer --autopass
libxl: don't leak buf in libxl_xen_console_read_start error handling
libxl: replace for loop with more idiomatic do-while loop

 tools/libxl/libxl.c        |   64 +++++++++++++++++++++-----------------------
 tools/libxl/libxl_cpuid.c  |    2 +-
 tools/libxl/libxl_device.c |    3 +++
 tools/libxl/libxl_dom.c    |   42 ++++++++++++++---------------
 tools/libxl/libxl_pci.c    |    9 +++----
 tools/libxl/libxl_x86.c    |    2 +-
 6 files changed, 59 insertions(+), 63 deletions(-)
Signature · 1 line
1.7.10.4
#1 lore raw

PATCH 1/13 libxl: fix unsigned less-than-0 comparison in e820_sanitize

Both src[i].size and delta are unsigned, so checking their difference for being less than 0 doesn't work.

Coverity-ID: 1055615

---
 tools/libxl/libxl_x86.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/libxl/libxl_x86.c +1 -1
index e1c183f..b11d036 100644
@@ -125,7 +125,7 @@ static int e820_sanitize(libxl_ctx *ctx, struct e820entry src[],
             src[i].type = E820_UNUSABLE;
             delta = ram_end - src[i].addr;
             /* The end < ram_end should weed this out */
-            if (src[i].size - delta < 0)
+            if (src[i].size < delta)
                 src[i].type = 0;
             else {
                 src[i].size -= delta;
Signature · 1 line
1.7.10.4
#2 lore raw

Re: [PATCH 01/13] libxl: fix unsigned less-than-0 comparison in e820_sanitize

Ping?

Quoted 25 lines · Matthew Daley

Both src[i].size and delta are unsigned, so checking their difference for being less than 0 doesn't work.

Coverity-ID: 1055615

---
 tools/libxl/libxl_x86.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/libxl/libxl_x86.c +1 -2
index e1c183f..b11d036 100644
@@ -125,7 +125,7 @@ static int e820_sanitize(libxl_ctx *ctx, struct e820entry src[],
             src[i].type = E820_UNUSABLE;
             delta = ram_end - src[i].addr;
             /* The end < ram_end should weed this out */
-            if (src[i].size - delta < 0)
+            if (src[i].size < delta)
                 src[i].type = 0;
             else {
                 src[i].size -= delta;
--

1.7.10.4

Reviewed-by
#3 lore raw

Re: [PATCH 01/13] libxl: fix unsigned less-than-0 comparison in e820_sanitize

Quoted 8 lines · Matthew Daley

Ping?

Quoted 5 lines · Matthew Daley

Both src[i].size and delta are unsigned, so checking their difference for being less than 0 doesn't work.

Coverity-ID: 1055615

Quoted 24 lines
Quoted 18 lines
---
 tools/libxl/libxl_x86.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/libxl/libxl_x86.c +1 -2
index e1c183f..b11d036 100644
@@ -125,7 +125,7 @@ static int e820_sanitize(libxl_ctx *ctx, struct e820entry src[],
             src[i].type = E820_UNUSABLE;
             delta = ram_end - src[i].addr;
             /* The end < ram_end should weed this out */
-            if (src[i].size - delta < 0)
+            if (src[i].size < delta)
                 src[i].type = 0;
             else {
                 src[i].size -= delta;
--
Quoted 2 lines

1.7.10.4

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

Acked-by on v1 1/13
#4 lore raw

Re: [PATCH 01/13] libxl: fix unsigned less-than-0 comparison in e820_sanitize

In reply to Matthew Daley

Matthew Daley writes ("[PATCH 01/13] libxl: fix unsigned less-than-0 comparison in e820_sanitize"):

Quoted 5 lines

Both src[i].size and delta are unsigned, so checking their difference for being less than 0 doesn't work.

Coverity-ID: 1055615

Sorry to have dropped this.

Thanks,
Ian.

#5 lore raw

PATCH 2/13 libxl: check for xc_domain_setmaxmem failure in libxl__build_pre

In reply to Matthew Daley

Coverity-ID: 1087115

---
 tools/libxl/libxl_dom.c |    7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
tools/libxl/libxl_dom.c +6 -1
index 72489f8..e9ac8f3 100644
@@ -238,7 +238,12 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,
     libxl_domain_set_nodeaffinity(ctx, domid, &info->nodemap);
     libxl_set_vcpuaffinity_all(ctx, domid, info->max_vcpus, &info->cpumap);
 
-    xc_domain_setmaxmem(ctx->xch, domid, info->target_memkb + LIBXL_MAXMEM_CONSTANT);
+    rc = xc_domain_setmaxmem(ctx->xch, domid, info->target_memkb + LIBXL_MAXMEM_CONSTANT);
+    if (rc != 0) {
+        LOG(ERROR, "Couldn't set max memory");
+        return rc;
+    }
+
     xs_domid = xs_read(ctx->xsh, XBT_NULL, "/tool/xenstored/domid", NULL);
     state->store_domid = xs_domid ? atoi(xs_domid) : 0;
     free(xs_domid);
Signature · 1 line
1.7.10.4
#6 lore raw

Re: [PATCH 02/13] libxl: check for xc_domain_setmaxmem failure in libxl__build_pre

Matthew Daley writes ("[PATCH 02/13] libxl: check for xc_domain_setmaxmem failure in libxl__build_pre"):

Quoted 1 line

Coverity-ID: 1087115

...

tools/libxl/libxl_dom.c int libxl__build_pre(libxl__gc *gc, uint32_t domid, · line 245

Quoted 6 lines
tools/libxl/libxl_dom.c +5 -1
-    xc_domain_setmaxmem(ctx->xch, domid, info->target_memkb + LIBXL_MAXMEM_CONSTANT);
+    rc = xc_domain_setmaxmem(ctx->xch, domid, info->target_memkb + LIBXL_MAXMEM_CONSTANT);
+    if (rc != 0) {
+        LOG(ERROR, "Couldn't set max memory");
+        return rc;
+    }

xc_domain_setmaxmem returns -1 on error, setting errno. rc is supposed to be a libxl error number.

There are a lot of wrong (and also deviant) error handling patterns in libxl_dom.c. I would recommend using the idiom found in libxl_domain_unpause in libxl.c.

Ian.

#7 lore raw

PATCH v2 2/13 libxl: check for xc_domain_setmaxmem failure in libxl__build_pre

Coverity-ID: 1087115

---
v2: Use LIBXL__LOG_ERRNO instead of LOG and return ERROR_FAIL instead of -1

 tools/libxl/libxl_dom.c |    8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
tools/libxl/libxl_dom.c +6 -2
index e95eff8..e696fee 100644
@@ -224,7 +224,6 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,
      * whatever that turns out to be.
      */
     if (libxl_defbool_val(info->numa_placement)) {
-
         if (!libxl_bitmap_is_full(&info->cpumap)) {
             LOG(ERROR, "Can run NUMA placement only if no vcpu "
                        "affinity is specified");
@@ -238,7 +237,12 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,
     libxl_domain_set_nodeaffinity(ctx, domid, &info->nodemap);
     libxl_set_vcpuaffinity_all(ctx, domid, info->max_vcpus, &info->cpumap);
 
-    xc_domain_setmaxmem(ctx->xch, domid, info->target_memkb + LIBXL_MAXMEM_CONSTANT);
+    if (xc_domain_setmaxmem(ctx->xch, domid, info->target_memkb +
+        LIBXL_MAXMEM_CONSTANT) < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "Couldn't set max memory");
+        return ERROR_FAIL;
+    }
+
     xs_domid = xs_read(ctx->xsh, XBT_NULL, "/tool/xenstored/domid", NULL);
     state->store_domid = xs_domid ? atoi(xs_domid) : 0;
     free(xs_domid);
Signature · 1 line
1.7.10.4
#8 lore raw

Re: [PATCH 02/13 v2] libxl: check for xc_domain_setmaxmem failure in libxl__build_pre

Ping?

Quoted 37 lines · Matthew Daley

Coverity-ID: 1087115

---
v2: Use LIBXL__LOG_ERRNO instead of LOG and return ERROR_FAIL instead of -1

 tools/libxl/libxl_dom.c |    8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
tools/libxl/libxl_dom.c +6 -3
index e95eff8..e696fee 100644
@@ -224,7 +224,6 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,
      * whatever that turns out to be.
      */
     if (libxl_defbool_val(info->numa_placement)) {
-
         if (!libxl_bitmap_is_full(&info->cpumap)) {
             LOG(ERROR, "Can run NUMA placement only if no vcpu "
                        "affinity is specified");
@@ -238,7 +237,12 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,
     libxl_domain_set_nodeaffinity(ctx, domid, &info->nodemap);
     libxl_set_vcpuaffinity_all(ctx, domid, info->max_vcpus, &info->cpumap);
 
-    xc_domain_setmaxmem(ctx->xch, domid, info->target_memkb + LIBXL_MAXMEM_CONSTANT);
+    if (xc_domain_setmaxmem(ctx->xch, domid, info->target_memkb +
+        LIBXL_MAXMEM_CONSTANT) < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "Couldn't set max memory");
+        return ERROR_FAIL;
+    }
+
     xs_domid = xs_read(ctx->xsh, XBT_NULL, "/tool/xenstored/domid", NULL);
     state->store_domid = xs_domid ? atoi(xs_domid) : 0;
     free(xs_domid);
--

1.7.10.4

Reviewed-by
#9 lore raw

Re: [PATCH 02/13 v2] libxl: check for xc_domain_setmaxmem failure in libxl__build_pre

Quoted 6 lines · Matthew Daley

Ping?

Quoted 2 lines · Matthew Daley

Coverity-ID: 1087115

Regards,
Dario

Quoted 40 lines
Quoted 33 lines

---
v2: Use LIBXL__LOG_ERRNO instead of LOG and return ERROR_FAIL instead of -1

 tools/libxl/libxl_dom.c |    8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
tools/libxl/libxl_dom.c +6 -3
index e95eff8..e696fee 100644
@@ -224,7 +224,6 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,
      * whatever that turns out to be.
      */
     if (libxl_defbool_val(info->numa_placement)) {
-
         if (!libxl_bitmap_is_full(&info->cpumap)) {
             LOG(ERROR, "Can run NUMA placement only if no vcpu "
                        "affinity is specified");
@@ -238,7 +237,12 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,
     libxl_domain_set_nodeaffinity(ctx, domid, &info->nodemap);
     libxl_set_vcpuaffinity_all(ctx, domid, info->max_vcpus, &info->cpumap);
 
-    xc_domain_setmaxmem(ctx->xch, domid, info->target_memkb + LIBXL_MAXMEM_CONSTANT);
+    if (xc_domain_setmaxmem(ctx->xch, domid, info->target_memkb +
+        LIBXL_MAXMEM_CONSTANT) < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "Couldn't set max memory");
+        return ERROR_FAIL;
+    }
+
     xs_domid = xs_read(ctx->xsh, XBT_NULL, "/tool/xenstored/domid", NULL);
     state->store_domid = xs_domid ? atoi(xs_domid) : 0;
     free(xs_domid);
--
Quoted 2 lines

1.7.10.4

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere)
-----------------------------------------------------------------
Dario Faggioli, Ph.D, http://about.me/dario.faggioli
Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

Acked-by
#10 lore raw

Re: [PATCH 02/13 v2] libxl: check for xc_domain_setmaxmem failure in libxl__build_pre

Dario Faggioli writes ("Re: [Xen-devel] [PATCH 02/13 v2] libxl: check for xc_domain_setmaxmem failure in libxl__build_pre"):

Quoted 8 lines · Matthew Daley
Quoted 6 lines

Ping?

Quoted 2 lines · Matthew Daley

Coverity-ID: 1087115

Sorry, this one seems to have been dropped.

Ian.

#11 lore raw

PATCH 3/13 libxl: correct file open success check in libxl__device_pci_reset

In reply to Matthew Daley

It could, even if only in theory, be fd 0.

(This is not the same as commit 4b62556b57!)

Coverity-ID: 1055895

---
 tools/libxl/libxl_pci.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/libxl/libxl_pci.c +1 -1
index c8eceb4..095d564 100644
@@ -979,7 +979,7 @@ static int libxl__device_pci_reset(libxl__gc *gc, unsigned int domain, unsigned
         LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "Failed to access pciback path %s", reset);
     reset = libxl__sprintf(gc, "%s/"PCI_BDF"/reset", SYSFS_PCI_DEV, domain, bus, dev, func);
     fd = open(reset, O_WRONLY);
-    if (fd > 0) {
+    if (fd >= 0) {
         rc = write(fd, "1", 1);
         if (rc < 0)
             LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "write to %s returned %d", reset, rc);
Signature · 1 line
1.7.10.4
Acked-by on v1 3/13
#12 lore raw

Re: [PATCH 03/13] libxl: correct file open success check in libxl__device_pci_reset

Matthew Daley writes ("[PATCH 03/13] libxl: correct file open success check in libxl__device_pci_reset"):

Quoted 6 lines

It could, even if only in theory, be fd 0.

(This is not the same as commit 4b62556b57!)

Coverity-ID: 1055895

Thanks,
Ian.

#13 lore raw

PATCH 4/13 libxl: don't leak p in libxl__wait_for_backend

In reply to Matthew Daley

Coverity-ID: 1055891

---
 tools/libxl/libxl_device.c |    3 +++
 1 file changed, 3 insertions(+)
tools/libxl/libxl_device.c +3 -0
index d995c83..072b5e3 100644
@@ -1219,6 +1219,7 @@ int libxl__wait_for_backend(libxl__gc *gc, char *be_path, char *state)
             goto out;
         } else {
             if (!strcmp(p, state)) {
+                free(p);
                 rc = 0;
                 goto out;
             } else {
@@ -1226,6 +1227,8 @@ int libxl__wait_for_backend(libxl__gc *gc, char *be_path, char *state)
                 watchdog--;
             }
         }
+
+        free(p);
     }
     LOG(ERROR, "Backend %s not ready", be_path);
 out:
Signature · 1 line
1.7.10.4
#14 lore raw

Re: [PATCH 04/13] libxl: don't leak p in libxl__wait_for_backend

Quoted 2 lines · Matthew Daley

Coverity-ID: 1055891

As with my " tools/libxl: Fix libxl__device_nic_from_xs_be()" patch, I feel the suggestion will be to use libxl__xs_read_checked() instead, given all the libxl GC machinery already around.

~Andrew

Quoted 25 lines
---
 tools/libxl/libxl_device.c |    3 +++
 1 file changed, 3 insertions(+)
tools/libxl/libxl_device.c +3 -0
index d995c83..072b5e3 100644
@@ -1219,6 +1219,7 @@ int libxl__wait_for_backend(libxl__gc *gc, char *be_path, char *state)
             goto out;
         } else {
             if (!strcmp(p, state)) {
+                free(p);
                 rc = 0;
                 goto out;
             } else {
@@ -1226,6 +1227,8 @@ int libxl__wait_for_backend(libxl__gc *gc, char *be_path, char *state)
                 watchdog--;
             }
         }
+
+        free(p);
     }
     LOG(ERROR, "Backend %s not ready", be_path);
 out:
#15 lore raw

Re: [PATCH 04/13] libxl: don't leak p in libxl__wait_for_backend

Quoted 7 lines · Andrew Cooper
Quoted 2 lines · Matthew Daley

Coverity-ID: 1055891

As with my " tools/libxl: Fix libxl__device_nic_from_xs_be()" patch, I feel the suggestion will be to use libxl__xs_read_checked() instead, given all the libxl GC machinery already around.

Good point, I didn't notice it was allocated by xs_read in the first place. I was holding back on such cases with the intent to produce one big patch to convert all xs_reads in libxl to their GC'd counterparts, but since I still haven't done this, and since it makes sense to individually patch the ones that are leaking their results right now, I'll make a v2.

- Matthew

#16 lore raw

PATCH v2 4/13 libxl: don't leak p in libxl__wait_for_backend

Use libxl__xs_read_checked instead of xs_read. While at it, tidy up the function as well.

Coverity-ID: 1055891

---
v2: Use libxl__xs_read_checked instead of xs_read. Tidy up the function.

 tools/libxl/libxl_device.c   |   41 +++++++++++++++++------------------------
 tools/libxl/libxl_internal.h |    3 ++-
 2 files changed, 19 insertions(+), 25 deletions(-)
tools/libxl/libxl_device.c +17 -24
index d995c83..ba7d100 100644
@@ -1199,37 +1199,30 @@ int libxl__wait_for_device_model_deprecated(libxl__gc *gc,
                                      check_callback, check_callback_userdata);
 }
 
-int libxl__wait_for_backend(libxl__gc *gc, char *be_path, char *state)
+int libxl__wait_for_backend(libxl__gc *gc, const char *be_path,
+                            const char *state)
 {
-    libxl_ctx *ctx = libxl__gc_owner(gc);
     int watchdog = 100;
-    unsigned int len;
-    char *p;
-    char *path = GCSPRINTF("%s/state", be_path);
-    int rc = -1;
+    const char *p, *path = GCSPRINTF("%s/state", be_path);
+    int rc;
+
+    while (watchdog-- > 0) {
+        rc = libxl__xs_read_checked(gc, XBT_NULL, path, &p);
+        if (rc) return rc;
 
-    while (watchdog > 0) {
-        p = xs_read(ctx->xsh, XBT_NULL, path, &len);
         if (p == NULL) {
-            if (errno == ENOENT) {
-                LOG(ERROR, "Backend %s does not exist", be_path);
-            } else {
-                LOGE(ERROR, "Failed to access backend %s", be_path);
-            }
-            goto out;
-        } else {
-            if (!strcmp(p, state)) {
-                rc = 0;
-                goto out;
-            } else {
-                usleep(100000);
-                watchdog--;
-            }
+            LOG(ERROR, "Backend %s does not exist", be_path);
+            return ERROR_FAIL;
         }
+
+        if (!strcmp(p, state))
+            return 0;
+
+        usleep(100000);
     }
+
     LOG(ERROR, "Backend %s not ready", be_path);
-out:
-    return rc;
+    return ERROR_FAIL;
 }
 
 /*
tools/libxl/libxl_internal.h +2 -1
index a2d8247..1bd23ff 100644
@@ -944,7 +944,8 @@ _hidden char *libxl__device_frontend_path(libxl__gc *gc, libxl__device *device);
 _hidden int libxl__parse_backend_path(libxl__gc *gc, const char *path,
                                       libxl__device *dev);
 _hidden int libxl__device_destroy(libxl__gc *gc, libxl__device *dev);
-_hidden int libxl__wait_for_backend(libxl__gc *gc, char *be_path, char *state);
+_hidden int libxl__wait_for_backend(libxl__gc *gc, const char *be_path,
+                                    const char *state);
 _hidden int libxl__nic_type(libxl__gc *gc, libxl__device *dev,
                             libxl_nic_type *nictype);
 
Signature · 1 line
1.7.10.4
#17 lore raw

Re: [PATCH 04/13 v2] libxl: don't leak p in libxl__wait_for_backend

Quoted 25 lines · Matthew Daley

Use libxl__xs_read_checked instead of xs_read. While at it, tidy up the function as well.

Coverity-ID: 1055891

---
v2: Use libxl__xs_read_checked instead of xs_read. Tidy up the function.

 tools/libxl/libxl_device.c   |   41 +++++++++++++++++------------------------
 tools/libxl/libxl_internal.h |    3 ++-
 2 files changed, 19 insertions(+), 25 deletions(-)
tools/libxl/libxl_device.c +2 -2
index d995c83..ba7d100 100644
@@ -1199,37 +1199,30 @@ int libxl__wait_for_device_model_deprecated(libxl__gc *gc,
                                      check_callback, check_callback_userdata);
 }
 
-int libxl__wait_for_backend(libxl__gc *gc, char *be_path, char *state)
+int libxl__wait_for_backend(libxl__gc *gc, const char *be_path,
+                            const char *state)
 {
-    libxl_ctx *ctx = libxl__gc_owner(gc);

tools/libxl/libxl_device.c int libxl__wait_for_device_model_deprecated(libxl__gc *gc, · line 1210

Quoted 10 lines
tools/libxl/libxl_device.c +5 -4
     int watchdog = 100;
-    unsigned int len;
-    char *p;
-    char *path = GCSPRINTF("%s/state", be_path);
-    int rc = -1;
+    const char *p, *path = GCSPRINTF("%s/state", be_path);
+    int rc;
+
+    while (watchdog-- > 0) {
+        rc = libxl__xs_read_checked(gc, XBT_NULL, path, &p);

libxl__xs_read_checked() can return 0, with p set to NULL in the case of an ENOENT from xenstore.

I think you need a NULL check before strcmp.

~Andrew

Quoted 50 lines
tools/libxl/libxl_device.c +10 -18
+        if (rc) return rc;
 
-    while (watchdog > 0) {
-        p = xs_read(ctx->xsh, XBT_NULL, path, &len);
         if (p == NULL) {
-            if (errno == ENOENT) {
-                LOG(ERROR, "Backend %s does not exist", be_path);
-            } else {
-                LOGE(ERROR, "Failed to access backend %s", be_path);
-            }
-            goto out;
-        } else {
-            if (!strcmp(p, state)) {
-                rc = 0;
-                goto out;
-            } else {
-                usleep(100000);
-                watchdog--;
-            }
+            LOG(ERROR, "Backend %s does not exist", be_path);
+            return ERROR_FAIL;
         }
+
+        if (!strcmp(p, state))
+            return 0;
+
+        usleep(100000);
     }
+
     LOG(ERROR, "Backend %s not ready", be_path);
-out:
-    return rc;
+    return ERROR_FAIL;
 }
 
 /*
tools/libxl/libxl_internal.h +2 -1
index a2d8247..1bd23ff 100644
@@ -944,7 +944,8 @@ _hidden char *libxl__device_frontend_path(libxl__gc *gc, libxl__device *device);
 _hidden int libxl__parse_backend_path(libxl__gc *gc, const char *path,
                                       libxl__device *dev);
 _hidden int libxl__device_destroy(libxl__gc *gc, libxl__device *dev);
-_hidden int libxl__wait_for_backend(libxl__gc *gc, char *be_path, char *state);
+_hidden int libxl__wait_for_backend(libxl__gc *gc, const char *be_path,
+                                    const char *state);
 _hidden int libxl__nic_type(libxl__gc *gc, libxl__device *dev,
                             libxl_nic_type *nictype);
 
#18 lore raw

Re: [PATCH 04/13 v2] libxl: don't leak p in libxl__wait_for_backend

Quoted 40 lines · Andrew Cooper
Quoted 25 lines · Matthew Daley

Use libxl__xs_read_checked instead of xs_read. While at it, tidy up the function as well.

Coverity-ID: 1055891

---
v2: Use libxl__xs_read_checked instead of xs_read. Tidy up the function.

 tools/libxl/libxl_device.c   |   41 +++++++++++++++++------------------------
 tools/libxl/libxl_internal.h |    3 ++-
 2 files changed, 19 insertions(+), 25 deletions(-)
tools/libxl/libxl_device.c +2 -2
index d995c83..ba7d100 100644
@@ -1199,37 +1199,30 @@ int libxl__wait_for_device_model_deprecated(libxl__gc *gc,
                                      check_callback, check_callback_userdata);
 }
 
-int libxl__wait_for_backend(libxl__gc *gc, char *be_path, char *state)
+int libxl__wait_for_backend(libxl__gc *gc, const char *be_path,
+                            const char *state)
 {
-    libxl_ctx *ctx = libxl__gc_owner(gc);

tools/libxl/libxl_device.c int libxl__wait_for_device_model_deprecated(libxl__gc *gc, · line 1210

Quoted 10 lines
tools/libxl/libxl_device.c +5 -4
     int watchdog = 100;
-    unsigned int len;
-    char *p;
-    char *path = GCSPRINTF("%s/state", be_path);
-    int rc = -1;
+    const char *p, *path = GCSPRINTF("%s/state", be_path);
+    int rc;
+
+    while (watchdog-- > 0) {
+        rc = libxl__xs_read_checked(gc, XBT_NULL, path, &p);

libxl__xs_read_checked() can return 0, with p set to NULL in the case of an ENOENT from xenstore.

Quoted 9 lines

I think you need a NULL check before strcmp.

~Andrew

Quoted 5 lines
tools/libxl/libxl_device.c +1 -2
+        if (rc) return rc;
 
-    while (watchdog > 0) {
-        p = xs_read(ctx->xsh, XBT_NULL, path, &len);
         if (p == NULL) {

^ That's checked here, isn't it? (The diff has sneakily left it here inbetween a bunch of removals)

- Matthew

Quoted 46 lines
Quoted 45 lines
tools/libxl/libxl_device.c +9 -16
-            if (errno == ENOENT) {
-                LOG(ERROR, "Backend %s does not exist", be_path);
-            } else {
-                LOGE(ERROR, "Failed to access backend %s", be_path);
-            }
-            goto out;
-        } else {
-            if (!strcmp(p, state)) {
-                rc = 0;
-                goto out;
-            } else {
-                usleep(100000);
-                watchdog--;
-            }
+            LOG(ERROR, "Backend %s does not exist", be_path);
+            return ERROR_FAIL;
         }
+
+        if (!strcmp(p, state))
+            return 0;
+
+        usleep(100000);
     }
+
     LOG(ERROR, "Backend %s not ready", be_path);
-out:
-    return rc;
+    return ERROR_FAIL;
 }
 
 /*
tools/libxl/libxl_internal.h +2 -1
index a2d8247..1bd23ff 100644
@@ -944,7 +944,8 @@ _hidden char *libxl__device_frontend_path(libxl__gc *gc, libxl__device *device);
 _hidden int libxl__parse_backend_path(libxl__gc *gc, const char *path,
                                       libxl__device *dev);
 _hidden int libxl__device_destroy(libxl__gc *gc, libxl__device *dev);
-_hidden int libxl__wait_for_backend(libxl__gc *gc, char *be_path, char *state);
+_hidden int libxl__wait_for_backend(libxl__gc *gc, const char *be_path,
+                                    const char *state);
 _hidden int libxl__nic_type(libxl__gc *gc, libxl__device *dev,
                             libxl_nic_type *nictype);
 
Reviewed-by
#19 lore raw

Re: [PATCH 04/13 v2] libxl: don't leak p in libxl__wait_for_backend

Quoted 46 lines · Matthew Daley
Quoted 45 lines · Andrew Cooper
Quoted 25 lines · Matthew Daley

Use libxl__xs_read_checked instead of xs_read. While at it, tidy up the function as well.

Coverity-ID: 1055891

---
v2: Use libxl__xs_read_checked instead of xs_read. Tidy up the function.

 tools/libxl/libxl_device.c   |   41 +++++++++++++++++------------------------
 tools/libxl/libxl_internal.h |    3 ++-
 2 files changed, 19 insertions(+), 25 deletions(-)
tools/libxl/libxl_device.c +2 -2
index d995c83..ba7d100 100644
@@ -1199,37 +1199,30 @@ int libxl__wait_for_device_model_deprecated(libxl__gc *gc,
                                      check_callback, check_callback_userdata);
 }
 
-int libxl__wait_for_backend(libxl__gc *gc, char *be_path, char *state)
+int libxl__wait_for_backend(libxl__gc *gc, const char *be_path,
+                            const char *state)
 {
-    libxl_ctx *ctx = libxl__gc_owner(gc);

tools/libxl/libxl_device.c int libxl__wait_for_device_model_deprecated(libxl__gc *gc, · line 1210

Quoted 10 lines
tools/libxl/libxl_device.c +5 -4
     int watchdog = 100;
-    unsigned int len;
-    char *p;
-    char *path = GCSPRINTF("%s/state", be_path);
-    int rc = -1;
+    const char *p, *path = GCSPRINTF("%s/state", be_path);
+    int rc;
+
+    while (watchdog-- > 0) {
+        rc = libxl__xs_read_checked(gc, XBT_NULL, path, &p);

libxl__xs_read_checked() can return 0, with p set to NULL in the case of an ENOENT from xenstore.

I think you need a NULL check before strcmp.

~Andrew

Quoted 2 lines
tools/libxl/libxl_device.c +1 -0
+        if (rc) return rc;
 
Quoted 7 lines
Quoted 3 lines

tools/libxl/libxl_device.c int libxl__wait_for_device_model_deprecated(libxl__gc *gc, · line 1213

Quoted 3 lines
tools/libxl/libxl_device.c +0 -2
-    while (watchdog > 0) {
-        p = xs_read(ctx->xsh, XBT_NULL, path, &len);
         if (p == NULL) {

^ That's checked here, isn't it? (The diff has sneakily left it here inbetween a bunch of removals)

- Matthew

So it is. That was sneaky.

Quoted 46 lines
Quoted 46 lines
Quoted 46 lines
tools/libxl/libxl_device.c +9 -16
-            if (errno == ENOENT) {
-                LOG(ERROR, "Backend %s does not exist", be_path);
-            } else {
-                LOGE(ERROR, "Failed to access backend %s", be_path);
-            }
-            goto out;
-        } else {
-            if (!strcmp(p, state)) {
-                rc = 0;
-                goto out;
-            } else {
-                usleep(100000);
-                watchdog--;
-            }
+            LOG(ERROR, "Backend %s does not exist", be_path);
+            return ERROR_FAIL;
         }
+
+        if (!strcmp(p, state))
+            return 0;
+
+        usleep(100000);
     }
+
     LOG(ERROR, "Backend %s not ready", be_path);
-out:
-    return rc;
+    return ERROR_FAIL;
 }
 
 /*
tools/libxl/libxl_internal.h +2 -1
index a2d8247..1bd23ff 100644
@@ -944,7 +944,8 @@ _hidden char *libxl__device_frontend_path(libxl__gc *gc, libxl__device *device);
 _hidden int libxl__parse_backend_path(libxl__gc *gc, const char *path,
                                       libxl__device *dev);
 _hidden int libxl__device_destroy(libxl__gc *gc, libxl__device *dev);
-_hidden int libxl__wait_for_backend(libxl__gc *gc, char *be_path, char *state);
+_hidden int libxl__wait_for_backend(libxl__gc *gc, const char *be_path,
+                                    const char *state);
 _hidden int libxl__nic_type(libxl__gc *gc, libxl__device *dev,
                             libxl_nic_type *nictype);
 
#21 lore raw

Re: [PATCH 04/13 v2] libxl: don't leak p in libxl__wait_for_backend

In reply to Matthew Daley

Matthew Daley writes ("[PATCH 04/13 v2] libxl: don't leak p in libxl__wait_for_backend"):

Quoted 2 lines

Use libxl__xs_read_checked instead of xs_read. While at it, tidy up the function as well.

This was applied to staging, passed the tests, etc., and I also put it on my backport list.

However it doesn't apply cleanly to 4.3 and the conflicts weren't trivial. Would anyone care to backport it, or to fix the same bug another way ?

Thanks,
Ian.

#22 lore raw

PATCH 5/13 libxl: remove unsigned less-than-0 comparison

In reply to Matthew Daley

...from libxl_cpuid_parse_config_xend. value is unsigned so this doesn't work, and either way the following comparison on it being bigger than 3 does what was intended here anyway.

Coverity-ID: 1055614

---
 tools/libxl/libxl_cpuid.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/libxl/libxl_cpuid.c +1 -1
index e1c406c..dd21b78 100644
@@ -298,7 +298,7 @@ int libxl_cpuid_parse_config_xend(libxl_cpuid_policy_list *cpuid,
         }
         value = str[1] - 'a';
         endptr = strchr(str, '=');
-        if (value < 0 || value > 3 || endptr == NULL) {
+        if (value > 3 || endptr == NULL) {
             return 4;
         }
         str = endptr + 1;
Signature · 1 line
1.7.10.4
Acked-by on v1 5/13
#23 lore raw

Re: [PATCH 05/13] libxl: remove unsigned less-than-0 comparison

Matthew Daley writes ("[PATCH 05/13] libxl: remove unsigned less-than-0 comparison"):

Quoted 6 lines

...from libxl_cpuid_parse_config_xend. value is unsigned so this doesn't work, and either way the following comparison on it being bigger than 3 does what was intended here anyway.

Coverity-ID: 1055614

#24 lore raw

PATCH 6/13 libxl: actually abort if initializing a ctx's lock fails

In reply to Matthew Daley

If initializing the ctx's lock fails, don't keep going, but instead error out.

Coverity-ID: 1055289

---
 tools/libxl/libxl.c |    2 ++
 1 file changed, 2 insertions(+)
tools/libxl/libxl.c +2 -0
index 2b847ef..26eaee4 100644
@@ -74,6 +74,8 @@ int libxl_ctx_alloc(libxl_ctx **pctx, int version,
         LIBXL__LOG(ctx, LIBXL__LOG_ERROR, "Failed to initialize mutex");
         free(ctx);
         ctx = 0;
+        rc = ERROR_FAIL;
+        goto out;
     }
 
     /* Now ctx is safe for ctx_free; failures simply set rc and "goto out" */
Signature · 1 line
1.7.10.4
#26 lore raw

PATCH 7/13 libxl: don't leak output vcpu info on error in libxl_list_vcpu

In reply to Matthew Daley

Coverity-ID: 1055887

---
 tools/libxl/libxl.c |   10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
tools/libxl/libxl.c +7 -3
index 26eaee4..a57d571 100644
@@ -4544,15 +4544,15 @@ libxl_vcpuinfo *libxl_list_vcpu(libxl_ctx *ctx, uint32_t domid,
     for (*nb_vcpu = 0; *nb_vcpu <= domaininfo.max_vcpu_id; ++*nb_vcpu, ++ptr) {
         if (libxl_cpu_bitmap_alloc(ctx, &ptr->cpumap, 0)) {
             LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "allocating cpumap");
-            return NULL;
+            goto err;
         }
         if (xc_vcpu_getinfo(ctx->xch, domid, *nb_vcpu, &vcpuinfo) == -1) {
             LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting vcpu info");
-            return NULL;
+            goto err;
         }
         if (xc_vcpu_getaffinity(ctx->xch, domid, *nb_vcpu, ptr->cpumap.map) == -1) {
             LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting vcpu affinity");
-            return NULL;
+            goto err;
         }
         ptr->vcpuid = *nb_vcpu;
         ptr->cpu = vcpuinfo.cpu;
@@ -4562,6 +4562,10 @@ libxl_vcpuinfo *libxl_list_vcpu(libxl_ctx *ctx, uint32_t domid,
         ptr->vcpu_time = vcpuinfo.cpu_time;
     }
     return ret;
+
+err:
+    free(ret);
+    return NULL;
 }
 
 int libxl_set_vcpuaffinity(libxl_ctx *ctx, uint32_t domid, uint32_t vcpuid,
Signature · 1 line
1.7.10.4
#28 lore raw

PATCH 8/13 libxl: don't leak ptr in libxl_list_vm error case

In reply to Matthew Daley

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist.

Coverity-ID: 1055888

---
 tools/libxl/libxl.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
tools/libxl/libxl.c +7 -7
index a57d571..ca4c2cd 100644
@@ -674,17 +674,17 @@ libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
     libxl_vminfo *ptr;
     int idx, i, ret;
     xc_domaininfo_t info[1024];
-    int size = 1024;
 
-    ptr = calloc(size, sizeof(libxl_vminfo));
-    if (!ptr)
+    ret = xc_domain_getinfolist(ctx->xch, 1, ARRAY_SIZE(info), info);
+    if (ret < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting domain info list");
         return NULL;
+    }
 
-    ret = xc_domain_getinfolist(ctx->xch, 1, 1024, info);
-    if (ret<0) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "geting domain info list");
+    ptr = calloc(ret, sizeof(libxl_vminfo));
+    if (!ptr)
         return NULL;
-    }
+
     for (idx = i = 0; i < ret; i++) {
         if (libxl_is_stubdom(ctx, info[i].domain, NULL))
             continue;
Signature · 1 line
1.7.10.4
#29 lore raw

Re: [PATCH 08/13] libxl: don't leak ptr in libxl_list_vm error case

Quoted 21 lines · Matthew Daley

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist.

Coverity-ID: 1055888

---
 tools/libxl/libxl.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
tools/libxl/libxl.c +0 -3
index a57d571..ca4c2cd 100644
@@ -674,17 +674,17 @@ libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
     libxl_vminfo *ptr;
     int idx, i, ret;
     xc_domaininfo_t info[1024];
-    int size = 1024;
 
-    ptr = calloc(size, sizeof(libxl_vminfo));
-    if (!ptr)

tools/libxl/libxl.c libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out) · line 684

Quoted 10 lines
tools/libxl/libxl.c +5 -3
+    ret = xc_domain_getinfolist(ctx->xch, 1, ARRAY_SIZE(info), info);
+    if (ret < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting domain info list");
         return NULL;
+    }
 
-    ret = xc_domain_getinfolist(ctx->xch, 1, 1024, info);
-    if (ret<0) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "geting domain info list");
+    ptr = calloc(ret, sizeof(libxl_vminfo));

We now have a possible case of calling calloc(0, sizeof(libxl_vminfo));

The implementation is free to return NULL which will cause this function to fail in the eyes of its callers.

Doing a calloc(min(1,ret), sizeof(libxl_vminfo)); will suffice, as the callers already have to correctly deal with 0 domains but some allocated memory as a result of this function.

~Andrew

Quoted 7 lines
tools/libxl/libxl.c +2 -1
+    if (!ptr)
         return NULL;
-    }
+
     for (idx = i = 0; i < ret; i++) {
         if (libxl_is_stubdom(ctx, info[i].domain, NULL))
             continue;
#30 lore raw

Re: [PATCH 08/13] libxl: don't leak ptr in libxl_list_vm error case

Quoted 35 lines · Andrew Cooper
Quoted 21 lines · Matthew Daley

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist.

Coverity-ID: 1055888

---
 tools/libxl/libxl.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
tools/libxl/libxl.c +0 -3
index a57d571..ca4c2cd 100644
@@ -674,17 +674,17 @@ libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
     libxl_vminfo *ptr;
     int idx, i, ret;
     xc_domaininfo_t info[1024];
-    int size = 1024;
 
-    ptr = calloc(size, sizeof(libxl_vminfo));
-    if (!ptr)

tools/libxl/libxl.c libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out) · line 684

Quoted 10 lines
tools/libxl/libxl.c +5 -3
+    ret = xc_domain_getinfolist(ctx->xch, 1, ARRAY_SIZE(info), info);
+    if (ret < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting domain info list");
         return NULL;
+    }
 
-    ret = xc_domain_getinfolist(ctx->xch, 1, 1024, info);
-    if (ret<0) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "geting domain info list");
+    ptr = calloc(ret, sizeof(libxl_vminfo));

We now have a possible case of calling calloc(0, sizeof(libxl_vminfo));

Quoted 2 lines

The implementation is free to return NULL which will cause this function to fail in the eyes of its callers.

Good catch.

Quoted 4 lines

Doing a calloc(min(1,ret), sizeof(libxl_vminfo)); will suffice, as the callers already have to correctly deal with 0 domains but some allocated memory as a result of this function.

v2 on its way...

- Matthew

#31 lore raw

PATCH v2 8/13 libxl: don't leak ptr in libxl_list_vm error case

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist (unless 0 domains are returned, in which case we should still allocate one libxl_vminfo struct so we can return a non-NULL result and not appear to have failed from the caller's perspective.)

Coverity-ID: 1055888

---
v2: Always allocate at least one libxl_vminfo struct for the reason given in the commit description.

 tools/libxl/libxl.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
tools/libxl/libxl.c +7 -7
index b112294..944194b 100644
@@ -674,17 +674,17 @@ libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
     libxl_vminfo *ptr;
     int idx, i, ret;
     xc_domaininfo_t info[1024];
-    int size = 1024;
 
-    ptr = calloc(size, sizeof(libxl_vminfo));
-    if (!ptr)
+    ret = xc_domain_getinfolist(ctx->xch, 1, ARRAY_SIZE(info), info);
+    if (ret < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting domain info list");
         return NULL;
+    }
 
-    ret = xc_domain_getinfolist(ctx->xch, 1, 1024, info);
-    if (ret<0) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "geting domain info list");
+    ptr = calloc(min(1, ret), sizeof(libxl_vminfo));
+    if (!ptr)
         return NULL;
-    }
+
     for (idx = i = 0; i < ret; i++) {
         if (libxl_is_stubdom(ctx, info[i].domain, NULL))
             continue;
Signature · 1 line
1.7.10.4
Reviewed-by on v2 8/13
#32 lore raw

Re: [PATCH 08/13 v2] libxl: don't leak ptr in libxl_list_vm error case

Quoted 8 lines · Matthew Daley

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist (unless 0 domains are returned, in which case we should still allocate one libxl_vminfo struct so we can return a non-NULL result and not appear to have failed from the caller's perspective.)

Coverity-ID: 1055888

Quoted 36 lines

---
v2: Always allocate at least one libxl_vminfo struct for the reason given in the commit description.

 tools/libxl/libxl.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
tools/libxl/libxl.c +7 -7
index b112294..944194b 100644
@@ -674,17 +674,17 @@ libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
     libxl_vminfo *ptr;
     int idx, i, ret;
     xc_domaininfo_t info[1024];
-    int size = 1024;
 
-    ptr = calloc(size, sizeof(libxl_vminfo));
-    if (!ptr)
+    ret = xc_domain_getinfolist(ctx->xch, 1, ARRAY_SIZE(info), info);
+    if (ret < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting domain info list");
         return NULL;
+    }
 
-    ret = xc_domain_getinfolist(ctx->xch, 1, 1024, info);
-    if (ret<0) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "geting domain info list");
+    ptr = calloc(min(1, ret), sizeof(libxl_vminfo));
+    if (!ptr)
         return NULL;
-    }
+
     for (idx = i = 0; i < ret; i++) {
         if (libxl_is_stubdom(ctx, info[i].domain, NULL))
             continue;
#33 lore raw

PATCH v3 8/13 libxl: don't leak ptr in libxl_list_vm error case

In reply to Matthew Daley

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist (unless 0 domains are returned, in which case we should still allocate one libxl_vminfo struct so we can return a non-NULL result and not appear to have failed from the caller's perspective.)

Coverity-ID: 1055888

---
v3: Oops, min() isn't defined or used anywhere in libxl. Just use a ternary expression instead.

v2: Always allocate at least one libxl_vminfo struct for the reason given in the commit description.

 tools/libxl/libxl.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
tools/libxl/libxl.c +7 -7
index b112294..1a2462c 100644
@@ -674,17 +674,17 @@ libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
     libxl_vminfo *ptr;
     int idx, i, ret;
     xc_domaininfo_t info[1024];
-    int size = 1024;
 
-    ptr = calloc(size, sizeof(libxl_vminfo));
-    if (!ptr)
+    ret = xc_domain_getinfolist(ctx->xch, 1, ARRAY_SIZE(info), info);
+    if (ret < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting domain info list");
         return NULL;
+    }
 
-    ret = xc_domain_getinfolist(ctx->xch, 1, 1024, info);
-    if (ret<0) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "geting domain info list");
+    ptr = calloc(ret ? ret : 1, sizeof(libxl_vminfo));
+    if (!ptr)
         return NULL;
-    }
+
     for (idx = i = 0; i < ret; i++) {
         if (libxl_is_stubdom(ctx, info[i].domain, NULL))
             continue;
Signature · 1 line
1.7.10.4
#34 lore raw

Re: [PATCH 08/13 v3] libxl: don't leak ptr in libxl_list_vm error case

Quoted 30 lines · Matthew Daley

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist (unless 0 domains are returned, in which case we should still allocate one libxl_vminfo struct so we can return a non-NULL result and not appear to have failed from the caller's perspective.)

Coverity-ID: 1055888

---
v3: Oops, min() isn't defined or used anywhere in libxl. Just use a ternary expression instead.

v2: Always allocate at least one libxl_vminfo struct for the reason given in the commit description.

 tools/libxl/libxl.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
tools/libxl/libxl.c +0 -3
index b112294..1a2462c 100644
@@ -674,17 +674,17 @@ libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
     libxl_vminfo *ptr;
     int idx, i, ret;
     xc_domaininfo_t info[1024];
-    int size = 1024;
 
-    ptr = calloc(size, sizeof(libxl_vminfo));
-    if (!ptr)

tools/libxl/libxl.c libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out) · line 684

Quoted 10 lines
tools/libxl/libxl.c +5 -3
+    ret = xc_domain_getinfolist(ctx->xch, 1, ARRAY_SIZE(info), info);
+    if (ret < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting domain info list");
         return NULL;
+    }
 
-    ret = xc_domain_getinfolist(ctx->xch, 1, 1024, info);
-    if (ret<0) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "geting domain info list");
+    ptr = calloc(ret ? ret : 1, sizeof(libxl_vminfo));

Now I reconsider this, a comment in the code wouldn’t go amis, given the obscurity.

Also, given -std=gnu99, you can use the gcc-ism of "ret ?: 1"

Quoted 7 lines
tools/libxl/libxl.c +2 -1
+    if (!ptr)
         return NULL;
-    }
+
     for (idx = i = 0; i < ret; i++) {
         if (libxl_is_stubdom(ctx, info[i].domain, NULL))
             continue;
#35 lore raw

Re: [PATCH 08/13 v3] libxl: don't leak ptr in libxl_list_vm error case

Quoted 37 lines · Andrew Cooper
Quoted 36 lines · Matthew Daley

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist (unless 0 domains are returned, in which case we should still allocate one libxl_vminfo struct so we can return a non-NULL result and not appear to have failed from the caller's perspective.)

Coverity-ID: 1055888

---
v3: Oops, min() isn't defined or used anywhere in libxl. Just use a ternary expression instead.

v2: Always allocate at least one libxl_vminfo struct for the reason given in the commit description.

 tools/libxl/libxl.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
tools/libxl/libxl.c +4 -3
index b112294..1a2462c 100644
@@ -674,17 +674,17 @@ libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
     libxl_vminfo *ptr;
     int idx, i, ret;
     xc_domaininfo_t info[1024];
-    int size = 1024;
 
-    ptr = calloc(size, sizeof(libxl_vminfo));
-    if (!ptr)
+    ret = xc_domain_getinfolist(ctx->xch, 1, ARRAY_SIZE(info), info);
+    if (ret < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting domain info list");
         return NULL;
+    }
 
Quoted 7 lines

tools/libxl/libxl.c libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out) · line 684

Quoted 4 lines
tools/libxl/libxl.c +1 -3
-    ret = xc_domain_getinfolist(ctx->xch, 1, 1024, info);
-    if (ret<0) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "geting domain info list");
+    ptr = calloc(ret ? ret : 1, sizeof(libxl_vminfo));

Now I reconsider this, a comment in the code wouldn’t go amis, given the obscurity.

Agreed.

Quoted 2 lines

Also, given -std=gnu99, you can use the gcc-ism of "ret ?: 1"

Eww, but OK.

- Matthew

Quoted 9 lines
Quoted 8 lines
tools/libxl/libxl.c +2 -1
+    if (!ptr)
         return NULL;
-    }
+
     for (idx = i = 0; i < ret; i++) {
         if (libxl_is_stubdom(ctx, info[i].domain, NULL))
             continue;
#36 lore raw

Re: [PATCH 08/13 v3] libxl: don't leak ptr in libxl_list_vm error case

Quoted 3 lines · Matthew Daley
Quoted 1 line

Also, given -std=gnu99, you can use the gcc-ism of "ret ?: 1"

Eww, but OK.

"Can" but not "must" ;-)

#37 lore raw

PATCH v4 8/13 libxl: don't leak ptr in libxl_list_vm error case

In reply to Matthew Daley

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist (unless 0 domains are returned, in which case we should still allocate one libxl_vminfo struct so we can return a non-NULL result and not appear to have failed from the caller's perspective.)

Coverity-ID: 1055888

---
v4: Add a comment describing the calloc malarkey

 tools/libxl/libxl.c |   19 ++++++++++++-------
 1 file changed, 12 insertions(+), 7 deletions(-)
tools/libxl/libxl.c +12 -7
index b112294..7308d44 100644
@@ -674,17 +674,22 @@ libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
     libxl_vminfo *ptr;
     int idx, i, ret;
     xc_domaininfo_t info[1024];
-    int size = 1024;
 
-    ptr = calloc(size, sizeof(libxl_vminfo));
-    if (!ptr)
+    ret = xc_domain_getinfolist(ctx->xch, 1, ARRAY_SIZE(info), info);
+    if (ret < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting domain info list");
         return NULL;
+    }
 
-    ret = xc_domain_getinfolist(ctx->xch, 1, 1024, info);
-    if (ret<0) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "geting domain info list");
+    /*
+     * Always make sure to allocate at least one element; if we don't and we
+     * request zero, we (might) get back a null pointer, which if returned
+     * to our caller will make them think we've failed
+     */
+    ptr = calloc(ret ? ret : 1, sizeof(libxl_vminfo));
+    if (!ptr)
         return NULL;
-    }
+
     for (idx = i = 0; i < ret; i++) {
         if (libxl_is_stubdom(ctx, info[i].domain, NULL))
             continue;
Signature · 1 line
1.7.10.4
Reviewed-by on v4 8/13
#38 lore raw

Re: [PATCH 08/13 v4] libxl: don't leak ptr in libxl_list_vm error case

Quoted 8 lines · Matthew Daley

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist (unless 0 domains are returned, in which case we should still allocate one libxl_vminfo struct so we can return a non-NULL result and not appear to have failed from the caller's perspective.)

Coverity-ID: 1055888

Quoted 40 lines

---
v4: Add a comment describing the calloc malarkey

 tools/libxl/libxl.c |   19 ++++++++++++-------
 1 file changed, 12 insertions(+), 7 deletions(-)
tools/libxl/libxl.c +12 -7
index b112294..7308d44 100644
@@ -674,17 +674,22 @@ libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
     libxl_vminfo *ptr;
     int idx, i, ret;
     xc_domaininfo_t info[1024];
-    int size = 1024;
 
-    ptr = calloc(size, sizeof(libxl_vminfo));
-    if (!ptr)
+    ret = xc_domain_getinfolist(ctx->xch, 1, ARRAY_SIZE(info), info);
+    if (ret < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting domain info list");
         return NULL;
+    }
 
-    ret = xc_domain_getinfolist(ctx->xch, 1, 1024, info);
-    if (ret<0) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "geting domain info list");
+    /*
+     * Always make sure to allocate at least one element; if we don't and we
+     * request zero, we (might) get back a null pointer, which if returned
+     * to our caller will make them think we've failed
+     */
+    ptr = calloc(ret ? ret : 1, sizeof(libxl_vminfo));
+    if (!ptr)
         return NULL;
-    }
+
     for (idx = i = 0; i < ret; i++) {
         if (libxl_is_stubdom(ctx, info[i].domain, NULL))
             continue;
#39 lore raw

Re: [PATCH 08/13 v4] libxl: don't leak ptr in libxl_list_vm error case

In reply to Matthew Daley

Matthew Daley writes ("[PATCH 08/13 v4] libxl: don't leak ptr in libxl_list_vm error case"):

Quoted 5 lines

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist (unless 0 domains are returned, in which case we should still allocate one libxl_vminfo struct so we can return a non-NULL result and not appear to have failed from the caller's perspective.)

...

Quoted 1 line

+ ptr = calloc(ret ? ret : 1, sizeof(libxl_vminfo));

I think this should be a call to libxl__calloc, which cannot fail.

You still need to prat about with the ?: though I'm afraid.

Ian.

#40 lore raw

Re: [PATCH 08/13 v4] libxl: don't leak ptr in libxl_list_vm error case

Quoted 10 lines · Ian Jackson

Matthew Daley writes ("[PATCH 08/13 v4] libxl: don't leak ptr in libxl_list_vm error case"):

Quoted 5 lines

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist (unless 0 domains are returned, in which case we should still allocate one libxl_vminfo struct so we can return a non-NULL result and not appear to have failed from the caller's perspective.)

...

Quoted 1 line

+ ptr = calloc(ret ? ret : 1, sizeof(libxl_vminfo));

I think this should be a call to libxl__calloc, which cannot fail.

That adds the returned pointer to the GC, and since libxl_list_vm is part of the public libxl interface, I don't think that would be an acceptable change?

Quoted 2 lines

You still need to prat about with the ?: though I'm afraid.

You mean, you require the use of "a ?: b" over "a ? a : b"?

- Matthew

#41 lore raw

Re: [PATCH 08/13 v4] libxl: don't leak ptr in libxl_list_vm error case

Matthew Daley writes ("Re: [PATCH 08/13 v4] libxl: don't leak ptr in libxl_list_vm error case"):

Quoted 8 lines · Ian Jackson
Quoted 3 lines
Quoted 1 line

+ ptr = calloc(ret ? ret : 1, sizeof(libxl_vminfo));

I think this should be a call to libxl__calloc, which cannot fail.

That adds the returned pointer to the GC, and since libxl_list_vm is part of the public libxl interface, I don't think that would be an acceptable change?

That's what NOGC is for. You pass it to libxl__calloc et al, instead of gc.

Quoted 3 lines
Quoted 1 line

You still need to prat about with the ?: though I'm afraid.

You mean, you require the use of "a ?: b" over "a ? a : b"?

I mean only that libxl__calloc has the same unfortunate semantics as calloc, when the size is zero, so you need to keep your existing ? :. I agree that use of "a ?: b" in this case is a bad thing so please don't change that.

Thanks,
Ian.

#42 lore raw

PATCH v5 8/13 libxl: don't leak ptr in libxl_list_vm error case

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist (barring the caveat described in the newly-added comment)

Coverity-ID: 1055888

---
v5: Use libxl__calloc instead of calloc

 tools/libxl/libxl.c |   27 +++++++++++++++++----------
 1 file changed, 17 insertions(+), 10 deletions(-)
tools/libxl/libxl.c +17 -10
index 67a8e0e..3b73d99 100644
@@ -671,20 +671,24 @@ out:
  * be an aggregate of multiple domains. */
 libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
 {
-    libxl_vminfo *ptr;
+    GC_INIT(ctx);
+    libxl_vminfo *ptr = NULL;
     int idx, i, ret;
     xc_domaininfo_t info[1024];
-    int size = 1024;
 
-    ptr = calloc(size, sizeof(libxl_vminfo));
-    if (!ptr)
-        return NULL;
-
-    ret = xc_domain_getinfolist(ctx->xch, 1, 1024, info);
-    if (ret<0) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "geting domain info list");
-        return NULL;
+    ret = xc_domain_getinfolist(ctx->xch, 1, ARRAY_SIZE(info), info);
+    if (ret < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting domain info list");
+        goto out;
     }
+
+    /*
+     * Always make sure to allocate at least one element; if we don't and we
+     * request zero, libxl__calloc (might) think its internal call to calloc
+     * has failed (if it returns null), if so it would kill our process.
+     */
+    ptr = libxl__calloc(NOGC, ret ? ret : 1, sizeof(libxl_vminfo));
+
     for (idx = i = 0; i < ret; i++) {
         if (libxl_is_stubdom(ctx, info[i].domain, NULL))
             continue;
@@ -694,6 +698,9 @@ libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
         idx++;
     }
     *nb_vm_out = idx;
+
+out:
+    GC_FREE;
     return ptr;
 }
 
Signature · 1 line
1.7.10.4
#43 lore raw

Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case

Quoted 35 lines · Matthew Daley

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist (barring the caveat described in the newly-added comment)

Coverity-ID: 1055888

---
v5: Use libxl__calloc instead of calloc

 tools/libxl/libxl.c |   27 +++++++++++++++++----------
 1 file changed, 17 insertions(+), 10 deletions(-)
tools/libxl/libxl.c +2 -10
index 67a8e0e..3b73d99 100644
@@ -671,20 +671,24 @@ out:
  * be an aggregate of multiple domains. */
 libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
 {
-    libxl_vminfo *ptr;
+    GC_INIT(ctx);
+    libxl_vminfo *ptr = NULL;
     int idx, i, ret;
     xc_domaininfo_t info[1024];
-    int size = 1024;
 
-    ptr = calloc(size, sizeof(libxl_vminfo));
-    if (!ptr)
-        return NULL;
-
-    ret = xc_domain_getinfolist(ctx->xch, 1, 1024, info);
-    if (ret<0) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "geting domain info list");
-        return NULL;

tools/libxl/libxl.c out: · line 688

Quoted 10 lines
tools/libxl/libxl.c +9 -0
+    ret = xc_domain_getinfolist(ctx->xch, 1, ARRAY_SIZE(info), info);
+    if (ret < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting domain info list");
+        goto out;
     }
+
+    /*
+     * Always make sure to allocate at least one element; if we don't and we
+     * request zero, libxl__calloc (might) think its internal call to calloc
+     * has failed (if it returns null), if so it would kill our process.

Is size==0 something we could/should handle in our libxl__*alloc wrappers?

Or maybe this is something we should handle here e.g. by returning NULL, except perhaps our API doesn't allow for that?

Quoted 16 lines
tools/libxl/libxl.c +6 -0
+     */
+    ptr = libxl__calloc(NOGC, ret ? ret : 1, sizeof(libxl_vminfo));
+
     for (idx = i = 0; i < ret; i++) {
         if (libxl_is_stubdom(ctx, info[i].domain, NULL))
             continue;
@@ -694,6 +698,9 @@ libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
         idx++;
     }
     *nb_vm_out = idx;
+
+out:
+    GC_FREE;
     return ptr;
 }
 
#44 lore raw

Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case

Quoted 49 lines · Ian Campbell
Quoted 35 lines · Matthew Daley

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist (barring the caveat described in the newly-added comment)

Coverity-ID: 1055888

---
v5: Use libxl__calloc instead of calloc

 tools/libxl/libxl.c |   27 +++++++++++++++++----------
 1 file changed, 17 insertions(+), 10 deletions(-)
tools/libxl/libxl.c +2 -10
index 67a8e0e..3b73d99 100644
@@ -671,20 +671,24 @@ out:
  * be an aggregate of multiple domains. */
 libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
 {
-    libxl_vminfo *ptr;
+    GC_INIT(ctx);
+    libxl_vminfo *ptr = NULL;
     int idx, i, ret;
     xc_domaininfo_t info[1024];
-    int size = 1024;
 
-    ptr = calloc(size, sizeof(libxl_vminfo));
-    if (!ptr)
-        return NULL;
-
-    ret = xc_domain_getinfolist(ctx->xch, 1, 1024, info);
-    if (ret<0) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "geting domain info list");
-        return NULL;

tools/libxl/libxl.c out: · line 688

Quoted 10 lines
tools/libxl/libxl.c +9 -0
+    ret = xc_domain_getinfolist(ctx->xch, 1, ARRAY_SIZE(info), info);
+    if (ret < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting domain info list");
+        goto out;
     }
+
+    /*
+     * Always make sure to allocate at least one element; if we don't and we
+     * request zero, libxl__calloc (might) think its internal call to calloc
+     * has failed (if it returns null), if so it would kill our process.

Is size==0 something we could/should handle in our libxl__*alloc wrappers?

Quoted 2 lines

Or maybe this is something we should handle here e.g. by returning NULL, except perhaps our API doesn't allow for that?

The current API means that returning NULL from here constitutes a failure, which needs to be distinct from "I did what you asked and there are no domains".

*nb_vm_out is a second return parameter from this function.

~Andrew

Quoted 18 lines
Quoted 17 lines
tools/libxl/libxl.c +6 -0
+     */
+    ptr = libxl__calloc(NOGC, ret ? ret : 1, sizeof(libxl_vminfo));
+
     for (idx = i = 0; i < ret; i++) {
         if (libxl_is_stubdom(ctx, info[i].domain, NULL))
             continue;
@@ -694,6 +698,9 @@ libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
         idx++;
     }
     *nb_vm_out = idx;
+
+out:
+    GC_FREE;
     return ptr;
 }
 
#45 lore raw

Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case [and 1 more messages]

In reply to Ian Campbell

Ian Campbell writes ("Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case"):

Quoted 5 lines · Matthew Daley

tools/libxl/libxl.c out: · line 688

Quoted 4 lines
tools/libxl/libxl.c +4 -0
+    /*
+  * Always make sure to allocate at least one element; if we don't and we
+  * request zero, libxl__calloc (might) think its internal call to calloc
+  * has failed (if it returns null), if so it would kill our process.

[rewrapped -iwj]

Quoted 3 lines

Is size==0 something we could/should handle in our libxl__*alloc wrappers?

I think so. I think they should promise that if you pass size==0 you get a non-null pointer. Calling realloc with size==0 should crash.

Matthew Daley writes ("Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case"):

Quoted 1 line

Ping?

See Ian C's comment above, which AFAICT hasn't been answered.

Thanks,
Ian.

#46 lore raw

Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case [and 1 more messages]

Quoted 11 lines · Ian Jackson

Ian Campbell writes ("Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case"):

Quoted 5 lines · Matthew Daley

tools/libxl/libxl.c out: · line 688

Quoted 4 lines
tools/libxl/libxl.c +4 -0
+    /*
+  * Always make sure to allocate at least one element; if we don't and we
+  * request zero, libxl__calloc (might) think its internal call to calloc
+  * has failed (if it returns null), if so it would kill our process.

[rewrapped -iwj]

Quoted 2 lines

Is size==0 something we could/should handle in our libxl__*alloc wrappers?

I think so. I think they should promise that if you pass size==0 you get a non-null pointer. Calling realloc with size==0 should crash.

Can we not?

Having a non-NULL pointer to a 0 length buffer is madness, whose use should not be further encouraged.

Furthermore, code which ends calling libxl__*alloc() with a size of 0 *is* buggy, and should suffer an abort(), just as much as attempting to realloc to a size of 0.

Quoted 7 lines

Matthew Daley writes ("Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case"):

Quoted 1 line

Ping?

See Ian C's comment above, which AFAICT hasn't been answered.

Thanks,
Ian.

I believe I suitably answered that question, and justified why it had to stay.

There is an API difference between returning NULL (Call to list domains failed), and non NULL but with nb_domains = 0 (Call to list domains succeeded but there are no domains).

~Andrew

#47 lore raw

Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case [and 1 more messages]

Andrew Cooper writes ("Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case [and 1 more messages]"):

Quoted 8 lines · Ian Jackson
Quoted 2 lines

I think so. I think they should promise that if you pass size==0 you get a non-null pointer. Calling realloc with size==0 should crash.

Can we not?

Having a non-NULL pointer to a 0 length buffer is madness, whose use should not be further encouraged.

I don't know where you get the idea that such a pointer is "madness". It's a permitted response from malloc() according to the spec and indeed it's what glibc's malloc does!

Quoted 2 lines

Furthermore, code which ends calling libxl__*alloc() with a size of 0 *is* buggy, and should suffer an abort(),

Calling malloc(0) is certainly not buggy. It has behaviour which is clearly defined by the spec, and if you are careful with your error handling it is possible to use it correct. It's just a bit awkward.

libxl__*alloc should have an interface at least as capable as malloc()
- that is, libxl__zalloc(,0) should be fine.

Or to put it another way calling libxl_*alloc(,0) is only buggy because we've made it be buggy (or defined it to be so).

I wouldn't mind if libxl_*alloc(,0) succeeded and returned NULL, but I think having them return non-NULL is probably better.

Quoted 1 line

just as much as attempting to realloc to a size of 0.

realloc(,0) is problematic in the standard C API because it's not possible to tell, if you get NULL back, whether the original block was freed.

The problem with libxl__realloc(,,0) is that it's not clear whether the caller intends to free the block, or resize it to size 0. I think none of our use patterns will end up calling it. But I am open to arguments that libxl__realloc(,,0) should be defined to succeed and return NULL, or to succeed and return non-NULL.

But since this is controversial I should probably just apply Matthew's patch, which is a clear improvement, while we argue about it. So:

Ian.

#48 lore raw

Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case [and 1 more messages]

In reply to Ian Jackson
Quoted 11 lines · Ian Jackson

Ian Campbell writes ("Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case"):

Quoted 5 lines · Matthew Daley

tools/libxl/libxl.c out: · line 688

Quoted 4 lines
tools/libxl/libxl.c +4 -0
+    /*
+  * Always make sure to allocate at least one element; if we don't and we
+  * request zero, libxl__calloc (might) think its internal call to calloc
+  * has failed (if it returns null), if so it would kill our process.

[rewrapped -iwj]

Quoted 3 lines

Is size==0 something we could/should handle in our libxl__*alloc wrappers?

Quoted 7 lines

I think so. I think they should promise that if you pass size==0 you get a non-null pointer. Calling realloc with size==0 should crash.

Matthew Daley writes ("Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case"):

Quoted 1 line

Ping?

See Ian C's comment above, which AFAICT hasn't been answered.

Sorry, I implicitly agreed with Andrew's comment (at least, the second part...)

For the record, for custom wrapper functions like these I usually prefer to stick as close to the existing functions' semantics as possible, not so much out of love for standards but out of desiring the absence of surprising behaviour.

As for the second part of the comment, as Andrew already mentioned, the nb_domains output argument is sufficient to distinguish failure vs. 0 domains from libxl_list_vm already.

Out of curiosity, looking through libxl_list_vm's existing callers:
- libxl__domain_make is only (ab)using libxl_list_vm to get the number of vms via nb_domains... it just frees the vm list right afterward!
- xl_cmdimpl.c:print_uptime doesn't even check libxl_list_vm's result :D I will patch this shortly.

(Also, I see you have applied my patch regardless of this discussion; thanks.)

- Matthew

Quoted 3 lines

Thanks,
Ian.

#49 lore raw

Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case [and 1 more messages]

Quoted 18 lines · Matthew Daley
Quoted 17 lines · Ian Jackson

Ian Campbell writes ("Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case"):

Quoted 5 lines · Matthew Daley

tools/libxl/libxl.c out: · line 688

Quoted 4 lines
tools/libxl/libxl.c +4 -0
+    /*
+  * Always make sure to allocate at least one element; if we don't and we
+  * request zero, libxl__calloc (might) think its internal call to calloc
+  * has failed (if it returns null), if so it would kill our process.

[rewrapped -iwj]

Quoted 3 lines

Is size==0 something we could/should handle in our libxl__*alloc wrappers?

I think so. I think they should promise that if you pass size==0 you get a non-null pointer. Calling realloc with size==0 should crash.

Matthew Daley writes ("Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case"):

Quoted 1 line

Ping?

Quoted 8 lines
Quoted 1 line

See Ian C's comment above, which AFAICT hasn't been answered.

Sorry, I implicitly agreed with Andrew's comment (at least, the second part...)

For the record, for custom wrapper functions like these I usually prefer to stick as close to the existing functions' semantics as possible, not so much out of love for standards but out of desiring the absence of surprising behaviour.

Although, I suppose changing things wrt. the standards would just introduce new slightly-surprising behaviours vs. having existing really-surprising ones...

- Matthew

Quoted 18 lines

As for the second part of the comment, as Andrew already mentioned, the nb_domains output argument is sufficient to distinguish failure vs. 0 domains from libxl_list_vm already.

Out of curiosity, looking through libxl_list_vm's existing callers:
- libxl__domain_make is only (ab)using libxl_list_vm to get the number of vms via nb_domains... it just frees the vm list right afterward!
- xl_cmdimpl.c:print_uptime doesn't even check libxl_list_vm's result :D I will patch this shortly.

(Also, I see you have applied my patch regardless of this discussion; thanks.)

- Matthew

Quoted 3 lines

Thanks,
Ian.

#50 lore raw

Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case [and 1 more messages]

Matthew Daley writes ("Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case [and 1 more messages]"):

Quoted 9 lines · Matthew Daley
Quoted 4 lines

For the record, for custom wrapper functions like these I usually prefer to stick as close to the existing functions' semantics as possible, not so much out of love for standards but out of desiring the absence of surprising behaviour.

Although, I suppose changing things wrt. the standards would just introduce new slightly-surprising behaviours vs. having existing really-surprising ones...

Precisely. And anyone surprised by the lack of a workaround for some standard misbeaviour of malloc or realloc would hopefully go and find the doc comment for the wrapper and be reassured.

Ian.

#51 lore raw

PATCH 1/1 xl: check for libxl_list_vm failure in print_uptime

In reply to Matthew Daley
---
 tools/libxl/xl_cmdimpl.c |    4 ++++
 1 file changed, 4 insertions(+)
tools/libxl/xl_cmdimpl.c +4 -0
index bd26bcc..7c5b9ec 100644
@@ -6245,6 +6245,10 @@ static void print_uptime(int short_mode, uint32_t doms[], int nb_doms)
     if (nb_doms == 0) {
         print_dom0_uptime(short_mode, now);
         info = libxl_list_vm(ctx, &nb_vm);
+        if (info == NULL) {
+            fprintf(stderr, "Could not list vms.\n");
+            return;
+        }
         for (i = 0; i < nb_vm; i++)
             print_domU_uptime(info[i].domid, short_mode, now);
         libxl_vminfo_list_free(info, nb_vm);
Signature · 1 line
1.7.10.4
#54 lore raw

Re: [PATCH 08/13 v5] libxl: don't leak ptr in libxl_list_vm error case

In reply to Matthew Daley

Ping?

Quoted 64 lines · Matthew Daley

While at it, tidy up the function; there's no point in allocating more than the amount of domains actually returned by xc_domain_getinfolist (barring the caveat described in the newly-added comment)

Coverity-ID: 1055888

---
v5: Use libxl__calloc instead of calloc

 tools/libxl/libxl.c |   27 +++++++++++++++++----------
 1 file changed, 17 insertions(+), 10 deletions(-)
tools/libxl/libxl.c +17 -11
index 67a8e0e..3b73d99 100644
@@ -671,20 +671,24 @@ out:
  * be an aggregate of multiple domains. */
 libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
 {
-    libxl_vminfo *ptr;
+    GC_INIT(ctx);
+    libxl_vminfo *ptr = NULL;
     int idx, i, ret;
     xc_domaininfo_t info[1024];
-    int size = 1024;
 
-    ptr = calloc(size, sizeof(libxl_vminfo));
-    if (!ptr)
-        return NULL;
-
-    ret = xc_domain_getinfolist(ctx->xch, 1, 1024, info);
-    if (ret<0) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "geting domain info list");
-        return NULL;
+    ret = xc_domain_getinfolist(ctx->xch, 1, ARRAY_SIZE(info), info);
+    if (ret < 0) {
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "getting domain info list");
+        goto out;
     }
+
+    /*
+     * Always make sure to allocate at least one element; if we don't and we
+     * request zero, libxl__calloc (might) think its internal call to calloc
+     * has failed (if it returns null), if so it would kill our process.
+     */
+    ptr = libxl__calloc(NOGC, ret ? ret : 1, sizeof(libxl_vminfo));
+
     for (idx = i = 0; i < ret; i++) {
         if (libxl_is_stubdom(ctx, info[i].domain, NULL))
             continue;
@@ -694,6 +698,9 @@ libxl_vminfo * libxl_list_vm(libxl_ctx *ctx, int *nb_vm_out)
         idx++;
     }
     *nb_vm_out = idx;
+
+out:
+    GC_FREE;
     return ptr;
 }
 
--

1.7.10.4

#55 lore raw

PATCH 9/13 libxl: don't leak pcidevs in libxl_pcidev_assignable

In reply to Matthew Daley

Coverity-ID: 1055896

---
 tools/libxl/libxl_pci.c |    7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
tools/libxl/libxl_pci.c +3 -4
index 095d564..2e52470 100644
@@ -1021,11 +1021,10 @@ static int libxl_pcidev_assignable(libxl_ctx *ctx, libxl_device_pci *pcidev)
             pcidevs[i].bus == pcidev->bus &&
             pcidevs[i].dev == pcidev->dev &&
             pcidevs[i].func == pcidev->func)
-        {
-            return 1;
-        }
+            break;
     }
-    return 0;
+    free(pcidevs);
+    return i != num;
 }
 
 int libxl__device_pci_add(libxl__gc *gc, uint32_t domid, libxl_device_pci *pcidev, int starting)
Signature · 1 line
1.7.10.4
#57 lore raw

PATCH 10/13 libxl: don't try to fclose file twice on error in libxl_userdata_store

In reply to Matthew Daley

Do this by changing the function to not use stdio file operations, but just use write(2) directly.

While at it, tidy up the function's style issues.

Coverity-ID: 1056195

---
 tools/libxl/libxl_dom.c |   35 ++++++++++++++---------------------
 1 file changed, 14 insertions(+), 21 deletions(-)
tools/libxl/libxl_dom.c +14 -21
index e9ac8f3..0fe0a77 100644
@@ -1604,8 +1604,6 @@ int libxl_userdata_store(libxl_ctx *ctx, uint32_t domid,
     const char *newfilename;
     int e, rc;
     int fd = -1;
-    FILE *f = NULL;
-    size_t rs;
 
     filename = userdata_path(gc, domid, userdata_userid, "d");
     if (!filename) {
@@ -1626,38 +1624,33 @@ int libxl_userdata_store(libxl_ctx *ctx, uint32_t domid,
 
     rc = ERROR_FAIL;
 
-    fd= open(newfilename, O_RDWR|O_CREAT|O_TRUNC, 0600);
-    if (fd<0)
+    fd = open(newfilename, O_RDWR | O_CREAT | O_TRUNC, 0600);
+    if (fd < 0)
         goto err;
 
-    f= fdopen(fd, "wb");
-    if (!f)
+    if (write(fd, data, datalen) != datalen)
         goto err;
-    fd = -1;
 
-    rs = fwrite(data, 1, datalen, f);
-    if (rs != datalen) {
-        assert(ferror(f));
+    if (close(fd) < 0) {
+        fd = -1;
         goto err;
     }
+    fd = -1;
 
-    if (fclose(f))
-        goto err;
-    f = 0;
-
-    if (rename(newfilename,filename))
+    if (rename(newfilename, filename))
         goto err;
 
     rc = 0;
 
 err:
-    e = errno;
-    if (f) fclose(f);
-    if (fd>=0) close(fd);
+    if (fd >= 0) {
+        e = errno;
+        close(fd);
+        errno = e;
+    }
 
-    errno = e;
-    if ( rc )
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "cannot write %s for %s",
+    if (rc)
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "cannot write/rename %s for %s",
                  newfilename, filename);
 out:
     GC_FREE;
Signature · 1 line
1.7.10.4
#58 lore raw

Re: [PATCH 10/13] libxl: don't try to fclose file twice on error in libxl_userdata_store

Matthew Daley writes ("[PATCH 10/13] libxl: don't try to fclose file twice on error in libxl_userdata_store"):

Quoted 2 lines

Do this by changing the function to not use stdio file operations, but just use write(2) directly.

This is not correct. If you use write(2) directly you have to call it in a loop, in case of partial writes. This is quite tiresome. I think it would be better to simply fix the function not to "retry" the fclose if it fails the first time.

Quoted 1 line

While at it, tidy up the function's style issues.

This makes the result very hard to review. It is better to split nonfunctional changes out into their own patch.

Ian.

#59 lore raw

Re: [PATCH 10/13] libxl: don't try to fclose file twice on error in libxl_userdata_store

Quoted 8 lines · Ian Jackson

Matthew Daley writes ("[PATCH 10/13] libxl: don't try to fclose file twice on error in libxl_userdata_store"):

Quoted 2 lines

Do this by changing the function to not use stdio file operations, but just use write(2) directly.

This is not correct. If you use write(2) directly you have to call it in a loop, in case of partial writes. This is quite tiresome. I think it would be better to simply fix the function not to "retry" the fclose if it fails the first time.

Would using libxl_write_exactly be acceptable instead?

Quoted 5 lines
Quoted 2 lines

While at it, tidy up the function's style issues.

This makes the result very hard to review. It is better to split nonfunctional changes out into their own patch.

Sorry, will do.

- Matthew

#60 lore raw

Re: [PATCH 10/13] libxl: don't try to fclose file twice on error in libxl_userdata_store

Matthew Daley writes ("Re: [PATCH 10/13] libxl: don't try to fclose file twice on error in libxl_userdata_store"):

Quoted 7 lines · Ian Jackson
Quoted 4 lines

This is not correct. If you use write(2) directly you have to call it in a loop, in case of partial writes. This is quite tiresome. I think it would be better to simply fix the function not to "retry" the fclose if it fails the first time.

Would using libxl_write_exactly be acceptable instead?

Yes, I guess, if you prefer.

Ian.

#61 lore raw

PATCH v2 10/13 libxl: don't try to fclose file twice on error in libxl_userdata_store

Do this by changing the function to not use stdio file operations, but just use write(2) directly.

While at it, tidy up the function's style issues.

Coverity-ID: 1056195

---
v2: Use libxl_write_exactly instead of write to ensure all the userdata is written out

 tools/libxl/libxl_dom.c |   35 ++++++++++++++---------------------
 1 file changed, 14 insertions(+), 21 deletions(-)
tools/libxl/libxl_dom.c +14 -21
index faba32f..55f74b2 100644
@@ -1603,8 +1603,6 @@ int libxl_userdata_store(libxl_ctx *ctx, uint32_t domid,
     const char *newfilename;
     int e, rc;
     int fd = -1;
-    FILE *f = NULL;
-    size_t rs;
 
     filename = userdata_path(gc, domid, userdata_userid, "d");
     if (!filename) {
@@ -1625,38 +1623,33 @@ int libxl_userdata_store(libxl_ctx *ctx, uint32_t domid,
 
     rc = ERROR_FAIL;
 
-    fd= open(newfilename, O_RDWR|O_CREAT|O_TRUNC, 0600);
-    if (fd<0)
+    fd = open(newfilename, O_RDWR | O_CREAT | O_TRUNC, 0600);
+    if (fd < 0)
         goto err;
 
-    f= fdopen(fd, "wb");
-    if (!f)
+    if (libxl_write_exactly(ctx, fd, data, datalen, "userdata", newfilename))
         goto err;
-    fd = -1;
 
-    rs = fwrite(data, 1, datalen, f);
-    if (rs != datalen) {
-        assert(ferror(f));
+    if (close(fd) < 0) {
+        fd = -1;
         goto err;
     }
+    fd = -1;
 
-    if (fclose(f))
-        goto err;
-    f = 0;
-
-    if (rename(newfilename,filename))
+    if (rename(newfilename, filename))
         goto err;
 
     rc = 0;
 
 err:
-    e = errno;
-    if (f) fclose(f);
-    if (fd>=0) close(fd);
+    if (fd >= 0) {
+        e = errno;
+        close(fd);
+        errno = e;
+    }
 
-    errno = e;
-    if ( rc )
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "cannot write %s for %s",
+    if (rc)
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "cannot write/rename %s for %s",
                  newfilename, filename);
 out:
     GC_FREE;
Signature · 1 line
1.7.10.4
#62 lore raw

PATCH v3 10/13 libxl: don't try to fclose file twice on error in libxl_userdata_store

Do this by changing the function to not use stdio file operations, but just use the fd directly with libxl_write_exactly.

While at it, tidy up the function's style issues.

Coverity-ID: 1056195

---
v3: Fix the commit message

v2: Use libxl_write_exactly instead of write to ensure all the userdata is written out

 tools/libxl/libxl_dom.c |   35 ++++++++++++++---------------------
 1 file changed, 14 insertions(+), 21 deletions(-)
tools/libxl/libxl_dom.c +14 -21
index faba32f..55f74b2 100644
@@ -1603,8 +1603,6 @@ int libxl_userdata_store(libxl_ctx *ctx, uint32_t domid,
     const char *newfilename;
     int e, rc;
     int fd = -1;
-    FILE *f = NULL;
-    size_t rs;
 
     filename = userdata_path(gc, domid, userdata_userid, "d");
     if (!filename) {
@@ -1625,38 +1623,33 @@ int libxl_userdata_store(libxl_ctx *ctx, uint32_t domid,
 
     rc = ERROR_FAIL;
 
-    fd= open(newfilename, O_RDWR|O_CREAT|O_TRUNC, 0600);
-    if (fd<0)
+    fd = open(newfilename, O_RDWR | O_CREAT | O_TRUNC, 0600);
+    if (fd < 0)
         goto err;
 
-    f= fdopen(fd, "wb");
-    if (!f)
+    if (libxl_write_exactly(ctx, fd, data, datalen, "userdata", newfilename))
         goto err;
-    fd = -1;
 
-    rs = fwrite(data, 1, datalen, f);
-    if (rs != datalen) {
-        assert(ferror(f));
+    if (close(fd) < 0) {
+        fd = -1;
         goto err;
     }
+    fd = -1;
 
-    if (fclose(f))
-        goto err;
-    f = 0;
-
-    if (rename(newfilename,filename))
+    if (rename(newfilename, filename))
         goto err;
 
     rc = 0;
 
 err:
-    e = errno;
-    if (f) fclose(f);
-    if (fd>=0) close(fd);
+    if (fd >= 0) {
+        e = errno;
+        close(fd);
+        errno = e;
+    }
 
-    errno = e;
-    if ( rc )
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "cannot write %s for %s",
+    if (rc)
+        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "cannot write/rename %s for %s",
                  newfilename, filename);
 out:
     GC_FREE;
Signature · 1 line
1.7.10.4
Acked-by on v3 10/13
#63 lore raw

Re: [PATCH 10/13 v3] libxl: don't try to fclose file twice on error in libxl_userdata_store

Matthew Daley writes ("[PATCH 10/13 v3] libxl: don't try to fclose file twice on error in libxl_userdata_store"):

Quoted 4 lines

Do this by changing the function to not use stdio file operations, but just use the fd directly with libxl_write_exactly.

While at it, tidy up the function's style issues.

#64 lore raw

PATCH 11/13 libxl: use pipe instead of temporary file for VNC viewer --autopass

In reply to Matthew Daley

Coverity was complaining about the permissions implicitly set on the temporary file used to pass the VNC password to the viewer when using the --autopass feature. By replacing the use of the temporary file with a pipe, we fix the problem (well, quiesce Coverity at least), tidy the code and remove the buildup of temporary file cruft all at once.

Tested with TightVNC.

Coverity-ID: 1055958

---
 tools/libxl/libxl.c |   30 +++++++++++-------------------
 1 file changed, 11 insertions(+), 19 deletions(-)
tools/libxl/libxl.c +11 -19
index ca4c2cd..41b8f60 100644
@@ -1623,7 +1623,7 @@ int libxl_vncviewer_exec(libxl_ctx *ctx, uint32_t domid, int autopass)
     GC_INIT(ctx);
     const char *vnc_port;
     const char *vnc_listen = NULL, *vnc_pass = NULL;
-    int port = 0, autopass_fd = -1;
+    int port = 0, autopass_fds[2] = {-1, -1};
     char *vnc_bin, *args[] = {
         "vncviewer",
         NULL, /* hostname:display */
@@ -1655,38 +1655,30 @@ int libxl_vncviewer_exec(libxl_ctx *ctx, uint32_t domid, int autopass)
     args[1] = libxl__sprintf(gc, "%s:%d", vnc_listen, port);
 
     if ( vnc_pass ) {
-        char tmpname[] = "/tmp/vncautopass.XXXXXX";
-        autopass_fd = mkstemp(tmpname);
-        if ( autopass_fd < 0 ) {
-            LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR,
-                             "mkstemp %s failed", tmpname);
-            goto x_fail;
-        }
-
-        if ( unlink(tmpname) ) {
-            /* should never happen */
-            LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR,
-                             "unlink %s failed", tmpname);
+        if ( pipe(autopass_fds) < 0 ) {
+            LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "pipe failed");
             goto x_fail;
         }
 
-        if ( libxl_write_exactly(ctx, autopass_fd, vnc_pass, strlen(vnc_pass),
-                                    tmpname, "vnc password") )
+        if ( libxl_write_exactly(ctx, autopass_fds[1], vnc_pass, strlen(vnc_pass),
+                                    "(pipe)", "vnc password") )
             goto x_fail;
 
-        if ( lseek(autopass_fd, SEEK_SET, 0) ) {
-            LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR,
-                             "rewind %s (autopass) failed", tmpname);
+        if ( close(autopass_fds[1]) < 0 ) {
+            autopass_fds[1] = -1;
             goto x_fail;
         }
+        autopass_fds[1] = -1;
 
         args[2] = "-autopass";
     }
 
-    libxl__exec(gc, autopass_fd, -1, -1, args[0], args, NULL);
+    libxl__exec(gc, autopass_fds[0], -1, -1, args[0], args, NULL);
     abort();
 
  x_fail:
+    if ( autopass_fds[0] >= 0 ) close(autopass_fds[0]);
+    if ( autopass_fds[1] >= 0 ) close(autopass_fds[1]);
     GC_FREE;
     return ERROR_FAIL;
 }
Signature · 1 line
1.7.10.4
#65 lore raw

Re: [PATCH 11/13] libxl: use pipe instead of temporary file for VNC viewer --autopass

Matthew Daley writes ("[PATCH 11/13] libxl: use pipe instead of temporary file for VNC viewer --autopass"):

Quoted 5 lines

Coverity was complaining about the permissions implicitly set on the temporary file used to pass the VNC password to the viewer when using the --autopass feature. By replacing the use of the temporary file with a pipe, we fix the problem (well, quiesce Coverity at least), tidy the code and remove the buildup of temporary file cruft all at once.

I don't think this is a good idea. The VNC client may want to read the file more than once.

Quoted 1 line

Coverity-ID: 1055958

I think this is a false positive.

Quoted 1 line

From mkstemp(3) on Debian wheezy:

The file is created with permissions 0600, that is, read plus write for owner only. (In glibc versions 2.06 and earlier, the file is created with permissions 0666, that is, read and write for all users.) The returned file descriptor provides both read and write access to the
file. The file is opened with the open(2) O_EXCL flag, guaranteeing
that the caller is the process that creates the file.

Quoted 1 line

From mkstemp(3) on FreeBSD 9.2-RELEASE:

The mkstemp() function makes the same replacement to the template and creates the template file, mode 0600, returning a file descriptor opened for reading and writing. [...]

I was going to say that if the standard spec doesn't have this behaviour, it's a bug. But SuS agrees too. From
http://pubs.opengroup.org/onlinepubs/9699919799/functions/mkstemp.html:

The mkstemp() function shall create the file, and obtain a file descriptor for it, as if by a call to:

open(pathname, O_RDWR|O_CREAT|O_EXCL, S_IRUSR|S_IWUSR)

So I think Coverity is just wrong about this. Perhaps it thinks you are using an old version of glibc with an (IMO outrageous) security bug.

Ian.

#66 lore raw

Re: [PATCH 11/13] libxl: use pipe instead of temporary file for VNC viewer --autopass

Quoted 9 lines · Ian Jackson

Matthew Daley writes ("[PATCH 11/13] libxl: use pipe instead of temporary file for VNC viewer --autopass"):

Quoted 5 lines

Coverity was complaining about the permissions implicitly set on the temporary file used to pass the VNC password to the viewer when using the --autopass feature. By replacing the use of the temporary file with a pipe, we fix the problem (well, quiesce Coverity at least), tidy the code and remove the buildup of temporary file cruft all at once.

I don't think this is a good idea. The VNC client may want to read the file more than once.

I know it's a handwavey argument, but doesn't passing the password through stdin (vs. giving an explicit filename) imply that the file could very well be a pipe (or at least unseekable)?

Quoted 24 lines
Quoted 2 lines

Coverity-ID: 1055958

I think this is a false positive.

From mkstemp(3) on Debian wheezy:

The file is created with permissions 0600, that is, read plus write for owner only. (In glibc versions 2.06 and earlier, the file is created with permissions 0666, that is, read and write for all users.) The returned file descriptor provides both read and write access to the
file. The file is opened with the open(2) O_EXCL flag, guaranteeing
that the caller is the process that creates the file.

From mkstemp(3) on FreeBSD 9.2-RELEASE:

The mkstemp() function makes the same replacement to the template and creates the template file, mode 0600, returning a file descriptor opened for reading and writing. [...]

I was going to say that if the standard spec doesn't have this behaviour, it's a bug. But SuS agrees too. From
http://pubs.opengroup.org/onlinepubs/9699919799/functions/mkstemp.html:

Quoted 8 lines

The mkstemp() function shall create the file, and obtain a file descriptor for it, as if by a call to:

open(pathname, O_RDWR|O_CREAT|O_EXCL, S_IRUSR|S_IWUSR)

So I think Coverity is just wrong about this. Perhaps it thinks you are using an old version of glibc with an (IMO outrageous) security bug.

My man mkstemp says: "In glibc versions 2.06 and earlier, the file is created with permissions 0666, that is, read and write for all users.". So, a long time ago, you could very well have been outraged.

That said, I'm happy to drop this patch and tell Coverity to hush if that's preferable.

- Matthew

#67 lore raw

PATCH 12/13 libxl: don't leak buf in libxl_xen_console_read_start error handling

In reply to Matthew Daley

Coverity-ID: 1055889

---
 tools/libxl/libxl.c |    1 +
 1 file changed, 1 insertion(+)
tools/libxl/libxl.c +1 -0
index 41b8f60..4cd54a8 100644
@@ -5116,6 +5116,7 @@ libxl_xen_console_reader *
     cr = malloc(sizeof(libxl_xen_console_reader));
     if (!cr) {
         LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "cannot malloc libxl_xen_console_reader");
+        free(buf);
         return NULL;
     }
 
Signature · 1 line
1.7.10.4
#68 lore raw

Re: [PATCH 12/13] libxl: don't leak buf in libxl_xen_console_read_start error handling

Matthew Daley writes ("[PATCH 12/13] libxl: don't leak buf in libxl_xen_console_read_start error handling"):

Quoted 2 lines

Coverity-ID: 1055889

...

tools/libxl/libxl.c libxl_xen_console_reader * · line 5121

Quoted 11 lines
tools/libxl/libxl.c +1 -0
index 41b8f60..4cd54a8 100644
@@ -5116,6 +5116,7 @@ libxl_xen_console_reader *
     cr = malloc(sizeof(libxl_xen_console_reader));
     if (!cr) {
         LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "cannot malloc libxl_xen_console_reader");
+        free(buf);
         return NULL;
     }

In general we prefer to use the initialise everything and "goto out" style of error handling all the time.

But in this case the only errors are allocation failures which we have deemed fatal: so the function should simply be changed to use libxl__zalloc, which cannot fail.

Ian.

#69 lore raw

PATCH v2 12/13 libxl: don't leak buf in libxl_xen_console_read_start error handling

Use libxl__zallocs instead of plain mallocs + memset.

Coverity-ID: 1055889

---
v2: Use libxl__zallocs instead of custom error handling. I'm not sure if this is what you had in mind, Ian, specifically the use of GC_INIT/GC_FREE and NOGC vs. just using &ctx->nogc_gc

 tools/libxl/libxl.c |   19 ++++---------------
 1 file changed, 4 insertions(+), 15 deletions(-)
tools/libxl/libxl.c +4 -15
index a64186a..bdd721e 100644
@@ -5108,29 +5108,18 @@ int libxl_send_debug_keys(libxl_ctx *ctx, char *keys)
 libxl_xen_console_reader *
     libxl_xen_console_read_start(libxl_ctx *ctx, int clear)
 {
+    GC_INIT(ctx);
     libxl_xen_console_reader *cr;
     unsigned int size = 16384;
-    char *buf = malloc(size);
-
-    if (!buf) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "cannot malloc buffer for libxl_xen_console_reader,"
-            " size is %u", size);
-        return NULL;
-    }
 
-    cr = malloc(sizeof(libxl_xen_console_reader));
-    if (!cr) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "cannot malloc libxl_xen_console_reader");
-        return NULL;
-    }
-
-    memset(cr, 0, sizeof(libxl_xen_console_reader));
-    cr->buffer = buf;
+    cr = libxl__zalloc(NOGC, sizeof(libxl_xen_console_reader));
+    cr->buffer = libxl__zalloc(NOGC, size);
     cr->size = size;
     cr->count = size;
     cr->clear = clear;
     cr->incremental = 1;
 
+    GC_FREE;
     return cr;
 }
 
Signature · 1 line
1.7.10.4
Acked-by on v2 12/13
#70 lore raw

Re: [PATCH 12/13 v2] libxl: don't leak buf in libxl_xen_console_read_start error handling

Matthew Daley writes ("[PATCH 12/13 v2] libxl: don't leak buf in libxl_xen_console_read_start error handling"):

Quoted 4 lines

Use libxl__zallocs instead of plain mallocs + memset.

Coverity-ID: 1055889

Quoted 3 lines

v2: Use libxl__zallocs instead of custom error handling. I'm not sure if this is what you had in mind, Ian, specifically the use of GC_INIT/GC_FREE and NOGC vs. just using &ctx->nogc_gc

Yes, this was exactly what I meant.

As a bonus it's now much shorter and also more obviously correct :-).

Thanks,
Ian.

#71 lore raw

PATCH 13/13 libxl: replace for loop with more idiomatic do-while loop

In reply to Matthew Daley
---
 tools/libxl/libxl.c |    7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)
tools/libxl/libxl.c +2 -5
index 4cd54a8..293e14a 100644
@@ -5413,14 +5413,11 @@ int libxl_cpupool_destroy(libxl_ctx *ctx, uint32_t poolid)
         goto out1;
     }
 
-    for (;;) {
+    do {
         t = xs_transaction_start(ctx->xsh);
 
         xs_rm(ctx->xsh, XBT_NULL, libxl__sprintf(gc, "/local/pool/%d", poolid));
-
-        if (xs_transaction_end(ctx->xsh, t, 0) || (errno != EAGAIN))
-            break;
-    }
+    } while (!xs_transaction_end(ctx->xsh, t, 0) && errno == EAGAIN);
 
     rc = 0;
 
Signature · 1 line
1.7.10.4
#72 lore raw

Re: [PATCH 13/13] libxl: replace for loop with more idiomatic do-while loop

Matthew Daley writes ("[PATCH 13/13] libxl: replace for loop with more idiomatic do-while loop"):

Quoted 1 line

Personally I think the for loop is clearer. I'm not a fan of do loops.

Ian.

#73 lore raw

Re: [PATCH 13/13] libxl: replace for loop with more idiomatic do-while loop

Quoted 5 lines · Ian Jackson

Matthew Daley writes ("[PATCH 13/13] libxl: replace for loop with more idiomatic do-while loop"):

Quoted 1 line

Personally I think the for loop is clearer. I'm not a fan of do loops.

Hrmn, really? I guess I'll go find another bikeshed then...

- Matthew

Reviewed-by on the v1 series
#74 lore raw
In reply to Matthew Daley

On 01/12/2013 10:14, Matthew Daley wrote:

All except for patches 4 and 8 (which have followups),

Quoted 24 lines

Matthew Daley (13):
libxl: fix unsigned less-than-0 comparison in e820_sanitize
libxl: check for xc_domain_setmaxmem failure in libxl__build_pre
libxl: correct file open success check in libxl__device_pci_reset
libxl: don't leak p in libxl__wait_for_backend
libxl: remove unsigned less-than-0 comparison
libxl: actually abort if initializing a ctx's lock fails
libxl: don't leak output vcpu info on error in libxl_list_vcpu
libxl: don't leak ptr in libxl_list_vm error case
libxl: don't leak pcidevs in libxl_pcidev_assignable
libxl: don't try to fclose file twice on error in
libxl_userdata_store
libxl: use pipe instead of temporary file for VNC viewer --autopass
libxl: don't leak buf in libxl_xen_console_read_start error handling
libxl: replace for loop with more idiomatic do-while loop

 tools/libxl/libxl.c        |   64 +++++++++++++++++++++-----------------------
 tools/libxl/libxl_cpuid.c  |    2 +-
 tools/libxl/libxl_device.c |    3 +++
 tools/libxl/libxl_dom.c    |   42 ++++++++++++++---------------
 tools/libxl/libxl_pci.c    |    9 +++----
 tools/libxl/libxl_x86.c    |    2 +-
 6 files changed, 59 insertions(+), 63 deletions(-)