Fedora kernel-2.6.17-1.2142_FC4 patched with stable patch-2.6.17.4-vs2.0.2-rc26.diff
[linux-2.6.git] / ipc / sem.c
index 11ce967..11b2960 100644 (file)
--- a/ipc/sem.c
+++ b/ipc/sem.c
  * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
  *
  * SMP-threaded, sysctl's added
- * (c) 1999 Manfred Spraul <manfreds@colorfullife.com>
+ * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
  * Enforced range limit on SEM_UNDO
  * (c) 2001 Red Hat Inc <alan@redhat.com>
  * Lockless wakeup
  * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
+ *
+ * support for audit of ipc object properties and permission changes
+ * Dustin Kirkland <dustin.kirkland@us.ibm.com>
  */
 
 #include <linux/config.h>
 #include <linux/time.h>
 #include <linux/smp_lock.h>
 #include <linux/security.h>
+#include <linux/syscalls.h>
+#include <linux/audit.h>
+#include <linux/capability.h>
+#include <linux/seq_file.h>
+#include <linux/mutex.h>
+
 #include <asm/uaccess.h>
 #include "util.h"
 
@@ -87,7 +96,7 @@ static struct ipc_ids sem_ids;
 static int newary (key_t, int, int);
 static void freeary (struct sem_array *sma, int id);
 #ifdef CONFIG_PROC_FS
-static int sysvipc_sem_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data);
+static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
 #endif
 
 #define SEMMSL_FAST    256 /* 512 bytes on stack */
@@ -114,10 +123,10 @@ void __init sem_init (void)
 {
        used_sems = 0;
        ipc_init_ids(&sem_ids,sc_semmni);
-
-#ifdef CONFIG_PROC_FS
-       create_proc_read_entry("sysvipc/sem", 0, 0, sysvipc_sem_read_proc, NULL);
-#endif
+       ipc_init_proc_interface("sysvipc/sem",
+                               "       key      semid perms      nsems   uid   gid  cuid  cgid      otime      ctime\n",
+                               &sem_ids,
+                               sysvipc_sem_proc_show);
 }
 
 /*
@@ -135,7 +144,7 @@ void __init sem_init (void)
  *     * if it's IN_WAKEUP, then it must wait until the value changes
  *     * if it's not -EINTR, then the operation was completed by
  *       update_queue. semtimedop can return queue.status without
- *       performing any operation on the semaphore array.
+ *       performing any operation on the sem array.
  *     * otherwise it must acquire the spinlock and check what's up.
  *
  * The two-stage algorithm is necessary to protect against the following
@@ -175,23 +184,24 @@ static int newary (key_t key, int nsems, int semflg)
 
        sma->sem_perm.mode = (semflg & S_IRWXUGO);
        sma->sem_perm.key = key;
-       sma->sem_perm.xid = current->xid;
+       sma->sem_perm.xid = vx_current_xid();
 
        sma->sem_perm.security = NULL;
        retval = security_sem_alloc(sma);
        if (retval) {
-               ipc_rcu_free(sma, size);
+               ipc_rcu_putref(sma);
                return retval;
        }
 
        id = ipc_addid(&sem_ids, &sma->sem_perm, sc_semmni);
        if(id == -1) {
                security_sem_free(sma);
-               ipc_rcu_free(sma, size);
+               ipc_rcu_putref(sma);
                return -ENOSPC;
        }
        used_sems += nsems;
 
+       sma->sem_id = sem_buildid(id, sma->sem_perm.seq);
        sma->sem_base = (struct sem *) &sma[1];
        /* sma->sem_pending = NULL; */
        sma->sem_pending_last = &sma->sem_pending;
@@ -200,7 +210,7 @@ static int newary (key_t key, int nsems, int semflg)
        sma->sem_ctime = get_seconds();
        sem_unlock(sma);
 
-       return sem_buildid(id, sma->sem_perm.seq);
+       return sma->sem_id;
 }
 
 asmlinkage long sys_semget (key_t key, int nsems, int semflg)
@@ -210,7 +220,7 @@ asmlinkage long sys_semget (key_t key, int nsems, int semflg)
 
        if (nsems < 0 || nsems > sc_semmsl)
                return -EINVAL;
-       down(&sem_ids.sem);
+       mutex_lock(&sem_ids.mutex);
        
        if (key == IPC_PRIVATE) {
                err = newary(key, nsems, semflg);
@@ -223,8 +233,7 @@ asmlinkage long sys_semget (key_t key, int nsems, int semflg)
                err = -EEXIST;
        } else {
                sma = sem_lock(id);
-               if(sma==NULL)
-                       BUG();
+               BUG_ON(sma==NULL);
                if (nsems > sma->sem_nsems)
                        err = -EINVAL;
                else if (ipcperms(&sma->sem_perm, semflg))
@@ -238,29 +247,10 @@ asmlinkage long sys_semget (key_t key, int nsems, int semflg)
                sem_unlock(sma);
        }
 
-       up(&sem_ids.sem);
+       mutex_unlock(&sem_ids.mutex);
        return err;
 }
 
-/* doesn't acquire the sem_lock on error! */
-static int sem_revalidate(int semid, struct sem_array* sma, int nsems, short flg)
-{
-       struct sem_array* smanew;
-
-       smanew = sem_lock(semid);
-       if(smanew==NULL)
-               return -EIDRM;
-       if(smanew != sma || sem_checkid(sma,semid) || sma->sem_nsems != nsems) {
-               sem_unlock(smanew);
-               return -EIDRM;
-       }
-
-       if (flg && ipcperms(&sma->sem_perm, flg)) {
-               sem_unlock(smanew);
-               return -EACCES;
-       }
-       return 0;
-}
 /* Manage the doubly linked list sma->sem_pending as a FIFO:
  * insert new queue elements at the tail sma->sem_pending_last.
  */
@@ -377,12 +367,27 @@ static void update_queue (struct sem_array * sma)
                if (error <= 0) {
                        struct sem_queue *n;
                        remove_from_queue(sma,q);
-                       n = q->next;
                        q->status = IN_WAKEUP;
+                       /*
+                        * Continue scanning. The next operation
+                        * that must be checked depends on the type of the
+                        * completed operation:
+                        * - if the operation modified the array, then
+                        *   restart from the head of the queue and
+                        *   check for threads that might be waiting
+                        *   for semaphore values to become 0.
+                        * - if the operation didn't modify the array,
+                        *   then just continue.
+                        */
+                       if (q->alter)
+                               n = sma->sem_pending;
+                       else
+                               n = q->next;
                        wake_up_process(q->sleeper);
                        /* hands-off: q will disappear immediately after
                         * writing q->status.
                         */
+                       smp_wmb();
                        q->status = error;
                        q = n;
                } else {
@@ -437,8 +442,8 @@ static int count_semzcnt (struct sem_array * sma, ushort semnum)
        return semzcnt;
 }
 
-/* Free a semaphore set. freeary() is called with sem_ids.sem down and
- * the spinlock for this semaphore set hold. sem_ids.sem remains locked
+/* Free a semaphore set. freeary() is called with sem_ids.mutex locked and
+ * the spinlock for this semaphore set hold. sem_ids.mutex remains locked
  * on exit.
  */
 static void freeary (struct sem_array *sma, int id)
@@ -463,6 +468,7 @@ static void freeary (struct sem_array *sma, int id)
                n = q->next;
                q->status = IN_WAKEUP;
                wake_up_process(q->sleeper); /* doesn't sleep */
+               smp_wmb();
                q->status = -EIDRM;     /* hands-off q */
                q = n;
        }
@@ -474,7 +480,7 @@ static void freeary (struct sem_array *sma, int id)
        used_sems -= sma->sem_nsems;
        size = sizeof (*sma) + sma->sem_nsems * sizeof (struct sem);
        security_sem_free(sma);
-       ipc_rcu_free(sma, size);
+       ipc_rcu_putref(sma);
 }
 
 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
@@ -524,7 +530,7 @@ static int semctl_nolock(int semid, int semnum, int cmd, int version, union semu
                seminfo.semmnu = SEMMNU;
                seminfo.semmap = SEMMAP;
                seminfo.semume = SEMUME;
-               down(&sem_ids.sem);
+               mutex_lock(&sem_ids.mutex);
                if (cmd == SEM_INFO) {
                        seminfo.semusz = sem_ids.in_use;
                        seminfo.semaem = used_sems;
@@ -533,7 +539,7 @@ static int semctl_nolock(int semid, int semnum, int cmd, int version, union semu
                        seminfo.semaem = SEMAEM;
                }
                max_id = sem_ids.max_id;
-               up(&sem_ids.sem);
+               mutex_unlock(&sem_ids.mutex);
                if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo))) 
                        return -EFAULT;
                return (max_id < 0) ? 0: max_id;
@@ -543,7 +549,7 @@ static int semctl_nolock(int semid, int semnum, int cmd, int version, union semu
                struct semid64_ds tbuf;
                int id;
 
-               if(semid >= sem_ids.size)
+               if(semid >= sem_ids.entries->size)
                        return -EINVAL;
 
                memset(&tbuf,0,sizeof(tbuf));
@@ -611,17 +617,28 @@ static int semctl_main(int semid, int semnum, int cmd, int version, union semun
        switch (cmd) {
        case GETALL:
        {
-               ushort *array = arg.array;
+               ushort __user *array = arg.array;
                int i;
 
                if(nsems > SEMMSL_FAST) {
+                       ipc_rcu_getref(sma);
                        sem_unlock(sma);                        
+
                        sem_io = ipc_alloc(sizeof(ushort)*nsems);
-                       if(sem_io == NULL)
+                       if(sem_io == NULL) {
+                               ipc_lock_by_ptr(&sma->sem_perm);
+                               ipc_rcu_putref(sma);
+                               sem_unlock(sma);
                                return -ENOMEM;
-                       err = sem_revalidate(semid, sma, nsems, S_IRUGO);
-                       if(err)
+                       }
+
+                       ipc_lock_by_ptr(&sma->sem_perm);
+                       ipc_rcu_putref(sma);
+                       if (sma->sem_perm.deleted) {
+                               sem_unlock(sma);
+                               err = -EIDRM;
                                goto out_free;
+                       }
                }
 
                for (i = 0; i < sma->sem_nsems; i++)
@@ -637,28 +654,43 @@ static int semctl_main(int semid, int semnum, int cmd, int version, union semun
                int i;
                struct sem_undo *un;
 
+               ipc_rcu_getref(sma);
                sem_unlock(sma);
 
                if(nsems > SEMMSL_FAST) {
                        sem_io = ipc_alloc(sizeof(ushort)*nsems);
-                       if(sem_io == NULL)
+                       if(sem_io == NULL) {
+                               ipc_lock_by_ptr(&sma->sem_perm);
+                               ipc_rcu_putref(sma);
+                               sem_unlock(sma);
                                return -ENOMEM;
+                       }
                }
 
                if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
+                       ipc_lock_by_ptr(&sma->sem_perm);
+                       ipc_rcu_putref(sma);
+                       sem_unlock(sma);
                        err = -EFAULT;
                        goto out_free;
                }
 
                for (i = 0; i < nsems; i++) {
                        if (sem_io[i] > SEMVMX) {
+                               ipc_lock_by_ptr(&sma->sem_perm);
+                               ipc_rcu_putref(sma);
+                               sem_unlock(sma);
                                err = -ERANGE;
                                goto out_free;
                        }
                }
-               err = sem_revalidate(semid, sma, nsems, S_IWUGO);
-               if(err)
+               ipc_lock_by_ptr(&sma->sem_perm);
+               ipc_rcu_putref(sma);
+               if (sma->sem_perm.deleted) {
+                       sem_unlock(sma);
+                       err = -EIDRM;
                        goto out_free;
+               }
 
                for (i = 0; i < nsems; i++)
                        sma->sem_base[i].semval = sem_io[i];
@@ -792,7 +824,11 @@ static int semctl_down(int semid, int semnum, int cmd, int version, union semun
                goto out_unlock;
        }       
        ipcp = &sma->sem_perm;
-       
+
+       err = audit_ipc_obj(ipcp);
+       if (err)
+               goto out_unlock;
+
        if (current->euid != ipcp->cuid && 
            current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN)) {
                err=-EPERM;
@@ -809,6 +845,9 @@ static int semctl_down(int semid, int semnum, int cmd, int version, union semun
                err = 0;
                break;
        case IPC_SET:
+               err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode, ipcp);
+               if (err)
+                       goto out_unlock;
                ipcp->uid = setbuf.uid;
                ipcp->gid = setbuf.gid;
                ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
@@ -857,9 +896,9 @@ asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg)
                return err;
        case IPC_RMID:
        case IPC_SET:
-               down(&sem_ids.sem);
+               mutex_lock(&sem_ids.mutex);
                err = semctl_down(semid,semnum,cmd,version,arg);
-               up(&sem_ids.sem);
+               mutex_unlock(&sem_ids.mutex);
                return err;
        default:
                return -EINVAL;
@@ -871,7 +910,7 @@ static inline void lock_semundo(void)
        struct sem_undo_list *undo_list;
 
        undo_list = current->sysvsem.undo_list;
-       if ((undo_list != NULL) && (atomic_read(&undo_list->refcnt) != 1))
+       if (undo_list)
                spin_lock(&undo_list->lock);
 }
 
@@ -891,7 +930,7 @@ static inline void unlock_semundo(void)
        struct sem_undo_list *undo_list;
 
        undo_list = current->sysvsem.undo_list;
-       if ((undo_list != NULL) && (atomic_read(&undo_list->refcnt) != 1))
+       if (undo_list)
                spin_unlock(&undo_list->lock);
 }
 
@@ -919,9 +958,7 @@ static inline int get_undo_list(struct sem_undo_list **undo_listp)
                if (undo_list == NULL)
                        return -ENOMEM;
                memset(undo_list, 0, size);
-               /* don't initialize unodhd->lock here.  It's done
-                * in copy_semundo() instead.
-                */
+               spin_lock_init(&undo_list->lock);
                atomic_set(&undo_list->refcnt, 1);
                current->sysvsem.undo_list = undo_list;
        }
@@ -978,11 +1015,16 @@ static struct sem_undo *find_undo(int semid)
                goto out;
        }
        nsems = sma->sem_nsems;
+       ipc_rcu_getref(sma);
        sem_unlock(sma);
 
        new = (struct sem_undo *) kmalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
-       if (!new)
+       if (!new) {
+               ipc_lock_by_ptr(&sma->sem_perm);
+               ipc_rcu_putref(sma);
+               sem_unlock(sma);
                return ERR_PTR(-ENOMEM);
+       }
        memset(new, 0, sizeof(struct sem_undo) + sizeof(short)*nsems);
        new->semadj = (short *) &new[1];
        new->semid = semid;
@@ -992,13 +1034,18 @@ static struct sem_undo *find_undo(int semid)
        if (un) {
                unlock_semundo();
                kfree(new);
+               ipc_lock_by_ptr(&sma->sem_perm);
+               ipc_rcu_putref(sma);
+               sem_unlock(sma);
                goto out;
        }
-       error = sem_revalidate(semid, sma, nsems, 0);
-       if (error) {
+       ipc_lock_by_ptr(&sma->sem_perm);
+       ipc_rcu_putref(sma);
+       if (sma->sem_perm.deleted) {
+               sem_unlock(sma);
                unlock_semundo();
                kfree(new);
-               un = ERR_PTR(error);
+               un = ERR_PTR(-EIDRM);
                goto out;
        }
        new->proc_next = ulp->proc_list;
@@ -1020,7 +1067,7 @@ asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
        struct sembuf fast_sops[SEMOPM_FAST];
        struct sembuf* sops = fast_sops, *sop;
        struct sem_undo *un;
-       int undos = 0, decrease = 0, alter = 0, max;
+       int undos = 0, alter = 0, max;
        struct sem_queue queue;
        unsigned long jiffies_left = 0;
 
@@ -1055,13 +1102,10 @@ asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
                if (sop->sem_num >= max)
                        max = sop->sem_num;
                if (sop->sem_flg & SEM_UNDO)
-                       undos++;
-               if (sop->sem_op < 0)
-                       decrease = 1;
-               if (sop->sem_op > 0)
+                       undos = 1;
+               if (sop->sem_op != 0)
                        alter = 1;
        }
-       alter |= decrease;
 
 retry_undos:
        if (undos) {
@@ -1102,8 +1146,11 @@ retry_undos:
                goto out_unlock_free;
 
        error = try_atomic_semop (sma, sops, nsops, un, current->tgid);
-       if (error <= 0)
-               goto update;
+       if (error <= 0) {
+               if (alter && error == 0)
+                       update_queue (sma);
+               goto out_unlock_free;
+       }
 
        /* We need to sleep on this operation, so we put the current
         * task into the pending queue and go to sleep.
@@ -1115,6 +1162,7 @@ retry_undos:
        queue.undo = un;
        queue.pid = current->tgid;
        queue.id = semid;
+       queue.alter = alter;
        if (alter)
                append_to_queue(sma ,&queue);
        else
@@ -1144,8 +1192,7 @@ retry_undos:
 
        sma = sem_lock(semid);
        if(sma==NULL) {
-               if(queue.prev != NULL)
-                       BUG();
+               BUG_ON(queue.prev != NULL);
                error = -EIDRM;
                goto out_free;
        }
@@ -1166,9 +1213,6 @@ retry_undos:
        remove_from_queue(sma,&queue);
        goto out_unlock_free;
 
-update:
-       if (alter)
-               update_queue (sma);
 out_unlock_free:
        sem_unlock(sma);
 out_free:
@@ -1199,8 +1243,6 @@ int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
                error = get_undo_list(&undo_list);
                if (error)
                        return error;
-               if (atomic_read(&undo_list->refcnt) == 1)
-                       spin_lock_init(&undo_list->lock);
                atomic_inc(&undo_list->refcnt);
                tsk->sysvsem.undo_list = undo_list;
        } else 
@@ -1267,12 +1309,27 @@ found:
                /* perform adjustments registered in u */
                nsems = sma->sem_nsems;
                for (i = 0; i < nsems; i++) {
-                       struct sem * sem = &sma->sem_base[i];
+                       struct sem * semaphore = &sma->sem_base[i];
                        if (u->semadj[i]) {
-                               sem->semval += u->semadj[i];
-                               if (sem->semval < 0)
-                                       sem->semval = 0; /* shouldn't happen */
-                               sem->sempid = current->tgid;
+                               semaphore->semval += u->semadj[i];
+                               /*
+                                * Range checks of the new semaphore value,
+                                * not defined by sus:
+                                * - Some unices ignore the undo entirely
+                                *   (e.g. HP UX 11i 11.22, Tru64 V5.1)
+                                * - some cap the value (e.g. FreeBSD caps
+                                *   at 0, but doesn't enforce SEMVMX)
+                                *
+                                * Linux caps the semaphore value, both at 0
+                                * and at SEMVMX.
+                                *
+                                *      Manfred <manfred@colorfullife.com>
+                                */
+                               if (semaphore->semval < 0)
+                                       semaphore->semval = 0;
+                               if (semaphore->semval > SEMVMX)
+                                       semaphore->semval = SEMVMX;
+                               semaphore->sempid = current->tgid;
                        }
                }
                sma->sem_otime = get_seconds();
@@ -1285,54 +1342,24 @@ next_entry:
 }
 
 #ifdef CONFIG_PROC_FS
-static int sysvipc_sem_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data)
+static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
 {
-       off_t pos = 0;
-       off_t begin = 0;
-       int i, len = 0;
+       struct sem_array *sma = it;
 
-       len += sprintf(buffer, "       key      semid perms      nsems   uid   gid  cuid  cgid      otime      ctime\n");
-       down(&sem_ids.sem);
-
-       for(i = 0; i <= sem_ids.max_id; i++) {
-               struct sem_array *sma;
-               sma = sem_lock(i);
-               if (sma) {
-                       if (!vx_check(sma->sem_perm.xid, VX_IDENT)) {
-                               sem_unlock(sma);
-                               continue;
-                       }
-                       len += sprintf(buffer + len, "%10d %10d  %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",
-                               sma->sem_perm.key,
-                               sem_buildid(i,sma->sem_perm.seq),
-                               sma->sem_perm.mode,
-                               sma->sem_nsems,
-                               sma->sem_perm.uid,
-                               sma->sem_perm.gid,
-                               sma->sem_perm.cuid,
-                               sma->sem_perm.cgid,
-                               sma->sem_otime,
-                               sma->sem_ctime);
-                       sem_unlock(sma);
+       if (!vx_check(sma->sem_perm.xid, VX_IDENT))
+               return 0;
 
-                       pos += len;
-                       if(pos < offset) {
-                               len = 0;
-                               begin = pos;
-                       }
-                       if(pos > offset + length)
-                               goto done;
-               }
-       }
-       *eof = 1;
-done:
-       up(&sem_ids.sem);
-       *start = buffer + (offset - begin);
-       len -= (offset - begin);
-       if(len > length)
-               len = length;
-       if(len < 0)
-               len = 0;
-       return len;
+       return seq_printf(s,
+                         "%10d %10d  %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",
+                         sma->sem_perm.key,
+                         sma->sem_id,
+                         sma->sem_perm.mode,
+                         sma->sem_nsems,
+                         sma->sem_perm.uid,
+                         sma->sem_perm.gid,
+                         sma->sem_perm.cuid,
+                         sma->sem_perm.cgid,
+                         sma->sem_otime,
+                         sma->sem_ctime);
 }
 #endif