On Wed, 2009-09-02 at 09:07 -0400, Simo Sorce wrote:
> On Wed, 2009-09-02 at 14:53 +0200, Sumit Bose wrote:
> > On Tue, Sep 01, 2009 at 04:36:26PM -0400, Simo Sorce wrote:
> > > newer tevent versions (correctly) fail if loops are nested.
> > > fix the code to never nest loops.
> > >
> > > Simo.
> > >
> >
> > If during a enumeration an uid/gid is found which is not in the range,
> > the whole transaction is canceled and nothing is cached. Is this
> > expected?
>
> No it's not, I'll check how to handle that.
Ok I have fixed it by ignoring errors on the save operation.
I also found a copy&paste error where we were calling enum_users_process
from enum_groups_process
Simo.
--
Simo Sorce * Red Hat, Inc * New York
>From ea005368d9432f7e56334d39f6731db311e28bd9 Mon Sep 17 00:00:00 2001
From: Simo Sorce <sso...@redhat.com>
Date: Tue, 1 Sep 2009 15:44:52 -0400
Subject: [PATCH] Fix proxy enumeration
New tevent library finally outlawed nested loops.
---
server/providers/proxy.c | 209 +++++++++++++++++++++++++++-------------------
1 files changed, 123 insertions(+), 86 deletions(-)
diff --git a/server/providers/proxy.c b/server/providers/proxy.c
index d361191..c308d3f 100644
--- a/server/providers/proxy.c
+++ b/server/providers/proxy.c
@@ -684,6 +684,21 @@ static void get_pw_uid_remove_done(struct tevent_req *subreq)
/* =Getpwent-wrapper======================================================*/
+struct enum_users_state {
+ struct tevent_context *ev;
+ struct proxy_ctx *ctx;
+ struct sysdb_ctx *sysdb;
+ struct sss_domain_info *domain;
+ struct sysdb_handle *handle;
+
+ struct passwd *pwd;
+
+ size_t buflen;
+ char *buffer;
+
+ bool in_transaction;
+};
+
static void enum_users_process(struct tevent_req *subreq);
static struct tevent_req *enum_users_send(TALLOC_CTX *mem_ctx,
@@ -693,18 +708,32 @@ static struct tevent_req *enum_users_send(TALLOC_CTX *mem_ctx,
struct sss_domain_info *domain)
{
struct tevent_req *req, *subreq;
- struct proxy_state *state;
+ struct enum_users_state *state;
enum nss_status status;
- req = tevent_req_create(mem_ctx, &state, struct proxy_state);
+ req = tevent_req_create(mem_ctx, &state, struct enum_users_state);
if (!req) return NULL;
- memset(state, 0, sizeof(struct proxy_state));
-
state->ev = ev;
state->ctx = ctx;
state->sysdb = sysdb;
state->domain = domain;
+ state->handle = NULL;
+
+ state->pwd = talloc(state, struct passwd);
+ if (!state->pwd) {
+ tevent_req_error(req, ENOMEM);
+ goto fail;
+ }
+
+ state->buflen = DEFAULT_BUFSIZE;
+ state->buffer = talloc_size(state, state->buflen);
+ if (!state->buffer) {
+ tevent_req_error(req, ENOMEM);
+ goto fail;
+ }
+
+ state->in_transaction = false;
status = ctx->ops.setpwent();
if (status != NSS_STATUS_SUCCESS) {
@@ -730,56 +759,51 @@ static void enum_users_process(struct tevent_req *subreq)
{
struct tevent_req *req = tevent_req_callback_data(subreq,
struct tevent_req);
- struct proxy_state *state = tevent_req_data(req,
- struct proxy_state);
+ struct enum_users_state *state = tevent_req_data(req,
+ struct enum_users_state);
struct proxy_ctx *ctx = state->ctx;
enum nss_status status;
- char *buffer;
char *newbuf;
- size_t buflen;
int ret;
- ret = sysdb_transaction_recv(subreq, state, &state->handle);
- if (ret) {
- tevent_req_error(req, ret);
- return;
- }
- talloc_zfree(subreq);
-
- state->pwd = talloc(state, struct passwd);
- if (!state->pwd) {
- tevent_req_error(req, ENOMEM);
- return;
- }
+ if (!state->in_transaction) {
+ ret = sysdb_transaction_recv(subreq, state, &state->handle);
+ if (ret) {
+ goto fail;
+ }
+ talloc_zfree(subreq);
- buflen = DEFAULT_BUFSIZE;
- buffer = talloc_size(state, buflen);
- if (!buffer) {
- tevent_req_error(req, ENOMEM);
- return;
+ state->in_transaction = true;
+ } else {
+ ret = sysdb_store_user_recv(subreq);
+ if (ret) {
+ /* Do not fail completely on errors.
+ * Just report the failure to save and go on */
+ DEBUG(2, ("Failed to store user. Ignoring.\n"));
+ }
+ talloc_zfree(subreq);
}
- /* getpwent is synchronous so instead of artifically wrapping it
- * in fake async calls let's just loop here */
-
again:
- status = ctx->ops.getpwent_r(state->pwd, buffer, buflen, &ret);
+ /* get entry */
+ status = ctx->ops.getpwent_r(state->pwd,
+ state->buffer, state->buflen, &ret);
switch (status) {
case NSS_STATUS_TRYAGAIN:
/* buffer too small ? */
- if (buflen < MAX_BUF_SIZE) {
- buflen *= 2;
+ if (state->buflen < MAX_BUF_SIZE) {
+ state->buflen *= 2;
}
- if (buflen > MAX_BUF_SIZE) {
- buflen = MAX_BUF_SIZE;
+ if (state->buflen > MAX_BUF_SIZE) {
+ state->buflen = MAX_BUF_SIZE;
}
- newbuf = talloc_realloc_size(state, buffer, buflen);
+ newbuf = talloc_realloc_size(state, state->buffer, state->buflen);
if (!newbuf) {
ret = ENOMEM;
goto fail;
}
- buffer = newbuf;
+ state->buffer = newbuf;
goto again;
case NSS_STATUS_NOTFOUND:
@@ -813,14 +837,8 @@ again:
tevent_req_error(req, ENOMEM);
return;
}
- /* SYNC WAIT: (polls until req is done) */
- if (!tevent_req_poll(subreq, state->ev)) {
- tevent_req_error(req, EFAULT);
- return;
- }
- ret = sysdb_store_user_recv(subreq);
-
- goto again; /* next one */
+ tevent_req_set_callback(subreq, enum_users_process, req);
+ return;
case NSS_STATUS_UNAVAIL:
/* "remote" backend unavailable. Enter offline mode */
@@ -1203,6 +1221,21 @@ static void get_gr_gid_remove_done(struct tevent_req *subreq)
/* =Getgrent-wrapper======================================================*/
+struct enum_groups_state {
+ struct tevent_context *ev;
+ struct proxy_ctx *ctx;
+ struct sysdb_ctx *sysdb;
+ struct sss_domain_info *domain;
+ struct sysdb_handle *handle;
+
+ struct group *grp;
+
+ size_t buflen;
+ char *buffer;
+
+ bool in_transaction;
+};
+
static void enum_groups_process(struct tevent_req *subreq);
static struct tevent_req *enum_groups_send(TALLOC_CTX *mem_ctx,
@@ -1212,18 +1245,32 @@ static struct tevent_req *enum_groups_send(TALLOC_CTX *mem_ctx,
struct sss_domain_info *domain)
{
struct tevent_req *req, *subreq;
- struct proxy_state *state;
+ struct enum_groups_state *state;
enum nss_status status;
- req = tevent_req_create(mem_ctx, &state, struct proxy_state);
+ req = tevent_req_create(mem_ctx, &state, struct enum_groups_state);
if (!req) return NULL;
- memset(state, 0, sizeof(struct proxy_state));
-
state->ev = ev;
state->ctx = ctx;
state->sysdb = sysdb;
state->domain = domain;
+ state->handle = NULL;
+
+ state->grp = talloc(state, struct group);
+ if (!state->grp) {
+ tevent_req_error(req, ENOMEM);
+ goto fail;
+ }
+
+ state->buflen = DEFAULT_BUFSIZE;
+ state->buffer = talloc_size(state, state->buflen);
+ if (!state->buffer) {
+ tevent_req_error(req, ENOMEM);
+ goto fail;
+ }
+
+ state->in_transaction = false;
status = ctx->ops.setgrent();
if (status != NSS_STATUS_SUCCESS) {
@@ -1249,56 +1296,52 @@ static void enum_groups_process(struct tevent_req *subreq)
{
struct tevent_req *req = tevent_req_callback_data(subreq,
struct tevent_req);
- struct proxy_state *state = tevent_req_data(req,
- struct proxy_state);
+ struct enum_groups_state *state = tevent_req_data(req,
+ struct enum_groups_state);
struct proxy_ctx *ctx = state->ctx;
enum nss_status status;
- char *buffer;
char *newbuf;
- size_t buflen;
int ret;
- ret = sysdb_transaction_recv(subreq, state, &state->handle);
- if (ret) {
- tevent_req_error(req, ret);
- return;
- }
- talloc_zfree(subreq);
-
- state->grp = talloc(state, struct group);
- if (!state->grp) {
- tevent_req_error(req, ENOMEM);
- return;
- }
+ if (!state->in_transaction) {
+ ret = sysdb_transaction_recv(subreq, state, &state->handle);
+ if (ret) {
+ tevent_req_error(req, ret);
+ return;
+ }
+ talloc_zfree(subreq);
- buflen = DEFAULT_BUFSIZE;
- buffer = talloc_size(state, buflen);
- if (!buffer) {
- tevent_req_error(req, ENOMEM);
- return;
+ state->in_transaction = true;
+ } else {
+ ret = sysdb_store_group_recv(subreq);
+ if (ret) {
+ /* Do not fail completely on errors.
+ * Just report the failure to save and go on */
+ DEBUG(2, ("Failed to store group. Ignoring.\n"));
+ }
+ talloc_zfree(subreq);
}
- /* getgrent is synchronous so instead of artifically wrapping it
- * in fake async calls let's just loop here */
-
again:
- status = ctx->ops.getgrent_r(state->grp, buffer, buflen, &ret);
+ /* get entry */
+ status = ctx->ops.getgrent_r(state->grp,
+ state->buffer, state->buflen, &ret);
switch (status) {
case NSS_STATUS_TRYAGAIN:
/* buffer too small ? */
- if (buflen < MAX_BUF_SIZE) {
- buflen *= 2;
+ if (state->buflen < MAX_BUF_SIZE) {
+ state->buflen *= 2;
}
- if (buflen > MAX_BUF_SIZE) {
- buflen = MAX_BUF_SIZE;
+ if (state->buflen > MAX_BUF_SIZE) {
+ state->buflen = MAX_BUF_SIZE;
}
- newbuf = talloc_realloc_size(state, buffer, buflen);
+ newbuf = talloc_realloc_size(state, state->buffer, state->buflen);
if (!newbuf) {
ret = ENOMEM;
goto fail;
}
- buffer = newbuf;
+ state->buffer = newbuf;
goto again;
case NSS_STATUS_NOTFOUND:
@@ -1329,14 +1372,8 @@ again:
tevent_req_error(req, ENOMEM);
return;
}
- /* SYNC WAIT: (polls until req is done) */
- if (!tevent_req_poll(subreq, state->ev)) {
- tevent_req_error(req, EFAULT);
- return;
- }
- ret = sysdb_store_group_recv(subreq);
-
- goto again; /* next one */
+ tevent_req_set_callback(subreq, enum_groups_process, req);
+ return;
case NSS_STATUS_UNAVAIL:
/* "remote" backend unavailable. Enter offline mode */
--
1.6.2.5
_______________________________________________
sssd-devel mailing list
sssd-devel@lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/sssd-devel
