[patch 1/1] audit_send_reply(): fix error-path memory leak
[safe/jmp/linux-2.6] / kernel / acct.c
index 44dd6bd..91e1cfd 100644 (file)
@@ -43,7 +43,6 @@
  * a struct file opened for write. Fixed. 2/6/2000, AV.
  */
 
-#include <linux/config.h>
 #include <linux/mm.h>
 #include <linux/slab.h>
 #include <linux/acct.h>
@@ -59,6 +58,7 @@
 #include <asm/uaccess.h>
 #include <asm/div64.h>
 #include <linux/blkdev.h> /* sector_div */
+#include <linux/pid_namespace.h>
 
 /*
  * These constants control the amount of freespace that suspend and
@@ -75,7 +75,7 @@ int acct_parm[3] = {4, 2, 30};
 /*
  * External references and all of the globals.
  */
-static void do_acct_process(long, struct file *);
+static void do_acct_process(struct pid_namespace *ns, struct file *);
 
 /*
  * This structure is used so that all the data protected by lock
@@ -87,10 +87,12 @@ struct acct_glbs {
        volatile int            active;
        volatile int            needcheck;
        struct file             *file;
+       struct pid_namespace    *ns;
        struct timer_list       timer;
 };
 
-static struct acct_glbs acct_globals __cacheline_aligned = {SPIN_LOCK_UNLOCKED};
+static struct acct_glbs acct_globals __cacheline_aligned =
+       {__SPIN_LOCK_UNLOCKED(acct_globals.lock)};
 
 /*
  * Called whenever the timer says to check the free space.
@@ -118,7 +120,7 @@ static int check_free_space(struct file *file)
        spin_unlock(&acct_globals.lock);
 
        /* May block */
-       if (vfs_statfs(file->f_dentry, &sbuf))
+       if (vfs_statfs(file->f_path.dentry, &sbuf))
                return res;
        suspend = sbuf.f_blocks * SUSPEND;
        resume = sbuf.f_blocks * RESUME;
@@ -175,9 +177,11 @@ out:
 static void acct_file_reopen(struct file *file)
 {
        struct file *old_acct = NULL;
+       struct pid_namespace *old_ns = NULL;
 
        if (acct_globals.file) {
                old_acct = acct_globals.file;
+               old_ns = acct_globals.ns;
                del_timer(&acct_globals.timer);
                acct_globals.active = 0;
                acct_globals.needcheck = 0;
@@ -185,6 +189,7 @@ static void acct_file_reopen(struct file *file)
        }
        if (file) {
                acct_globals.file = file;
+               acct_globals.ns = get_pid_ns(task_active_pid_ns(current));
                acct_globals.needcheck = 0;
                acct_globals.active = 1;
                /* It's been deleted if it was used before so this is safe */
@@ -194,10 +199,11 @@ static void acct_file_reopen(struct file *file)
                add_timer(&acct_globals.timer);
        }
        if (old_acct) {
-               mnt_unpin(old_acct->f_vfsmnt);
+               mnt_unpin(old_acct->f_path.mnt);
                spin_unlock(&acct_globals.lock);
-               do_acct_process(0, old_acct);
+               do_acct_process(old_ns, old_acct);
                filp_close(old_acct, NULL);
+               put_pid_ns(old_ns);
                spin_lock(&acct_globals.lock);
        }
 }
@@ -212,7 +218,7 @@ static int acct_on(char *name)
        if (IS_ERR(file))
                return PTR_ERR(file);
 
-       if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
+       if (!S_ISREG(file->f_path.dentry->d_inode->i_mode)) {
                filp_close(file, NULL);
                return -EACCES;
        }
@@ -229,11 +235,11 @@ static int acct_on(char *name)
        }
 
        spin_lock(&acct_globals.lock);
-       mnt_pin(file->f_vfsmnt);
+       mnt_pin(file->f_path.mnt);
        acct_file_reopen(file);
        spin_unlock(&acct_globals.lock);
 
-       mntput(file->f_vfsmnt); /* it's pinned, now give up active reference */
+       mntput(file->f_path.mnt); /* it's pinned, now give up active reference */
 
        return 0;
 }
@@ -283,7 +289,7 @@ asmlinkage long sys_acct(const char __user *name)
 void acct_auto_close_mnt(struct vfsmount *m)
 {
        spin_lock(&acct_globals.lock);
-       if (acct_globals.file && acct_globals.file->f_vfsmnt == m)
+       if (acct_globals.file && acct_globals.file->f_path.mnt == m)
                acct_file_reopen(NULL);
        spin_unlock(&acct_globals.lock);
 }
@@ -299,7 +305,7 @@ void acct_auto_close(struct super_block *sb)
 {
        spin_lock(&acct_globals.lock);
        if (acct_globals.file &&
-           acct_globals.file->f_vfsmnt->mnt_sb == sb) {
+           acct_globals.file->f_path.mnt->mnt_sb == sb) {
                acct_file_reopen(NULL);
        }
        spin_unlock(&acct_globals.lock);
@@ -329,16 +335,16 @@ static comp_t encode_comp_t(unsigned long value)
        }
 
        /*
-         * If we need to round up, do it (and handle overflow correctly).
-         */
+        * If we need to round up, do it (and handle overflow correctly).
+        */
        if (rnd && (++value > MAXFRACT)) {
                value >>= EXPSIZE;
                exp++;
        }
 
        /*
-         * Clean it up and polish it off.
-         */
+        * Clean it up and polish it off.
+        */
        exp <<= MANTSIZE;               /* Shift the exponent into place */
        exp += value;                   /* and add on the mantissa. */
        return exp;
@@ -361,30 +367,30 @@ static comp_t encode_comp_t(unsigned long value)
 
 static comp2_t encode_comp2_t(u64 value)
 {
-        int exp, rnd;
-
-        exp = (value > (MAXFRACT2>>1));
-        rnd = 0;
-        while (value > MAXFRACT2) {
-                rnd = value & 1;
-                value >>= 1;
-                exp++;
-        }
-
-        /*
-         * If we need to round up, do it (and handle overflow correctly).
-         */
-        if (rnd && (++value > MAXFRACT2)) {
-                value >>= 1;
-                exp++;
-        }
-
-        if (exp > MAXEXP2) {
-                /* Overflow. Return largest representable number instead. */
-                return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1;
-        } else {
-                return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1));
-        }
+       int exp, rnd;
+
+       exp = (value > (MAXFRACT2>>1));
+       rnd = 0;
+       while (value > MAXFRACT2) {
+               rnd = value & 1;
+               value >>= 1;
+               exp++;
+       }
+
+       /*
+        * If we need to round up, do it (and handle overflow correctly).
+        */
+       if (rnd && (++value > MAXFRACT2)) {
+               value >>= 1;
+               exp++;
+       }
+
+       if (exp > MAXEXP2) {
+               /* Overflow. Return largest representable number instead. */
+               return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1;
+       } else {
+               return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1));
+       }
 }
 #endif
 
@@ -413,22 +419,22 @@ static u32 encode_float(u64 value)
  *  The acct_process() call is the workhorse of the process
  *  accounting system. The struct acct is built here and then written
  *  into the accounting file. This function should only be called from
- *  do_exit().
+ *  do_exit() or when switching to a different output file.
  */
 
 /*
  *  do_acct_process does all actual work. Caller holds the reference to file.
  */
-static void do_acct_process(long exitcode, struct file *file)
+static void do_acct_process(struct pid_namespace *ns, struct file *file)
 {
+       struct pacct_struct *pacct = &current->signal->pacct;
        acct_t ac;
        mm_segment_t fs;
-       unsigned long vsize;
        unsigned long flim;
        u64 elapsed;
        u64 run_time;
        struct timespec uptime;
-       unsigned long jiffies;
+       struct tty_struct *tty;
 
        /*
         * First check to see if there is enough free_space to continue
@@ -468,13 +474,7 @@ static void do_acct_process(long exitcode, struct file *file)
        }
 #endif
        do_div(elapsed, AHZ);
-       ac.ac_btime = xtime.tv_sec - elapsed;
-       jiffies = cputime_to_jiffies(cputime_add(current->utime,
-                                                current->signal->utime));
-       ac.ac_utime = encode_comp_t(jiffies_to_AHZ(jiffies));
-       jiffies = cputime_to_jiffies(cputime_add(current->stime,
-                                                current->signal->stime));
-       ac.ac_stime = encode_comp_t(jiffies_to_AHZ(jiffies));
+       ac.ac_btime = get_seconds() - elapsed;
        /* we really need to bite the bullet and change layout */
        ac.ac_uid = current->uid;
        ac.ac_gid = current->gid;
@@ -487,56 +487,36 @@ static void do_acct_process(long exitcode, struct file *file)
        ac.ac_gid16 = current->gid;
 #endif
 #if ACCT_VERSION==3
-       ac.ac_pid = current->tgid;
-       ac.ac_ppid = current->parent->tgid;
+       ac.ac_pid = task_tgid_nr_ns(current, ns);
+       rcu_read_lock();
+       ac.ac_ppid = task_tgid_nr_ns(rcu_dereference(current->real_parent), ns);
+       rcu_read_unlock();
 #endif
 
-       read_lock(&tasklist_lock);      /* pin current->signal */
-       ac.ac_tty = current->signal->tty ?
-               old_encode_dev(tty_devnum(current->signal->tty)) : 0;
-       read_unlock(&tasklist_lock);
-
-       ac.ac_flag = 0;
-       if (current->flags & PF_FORKNOEXEC)
-               ac.ac_flag |= AFORK;
-       if (current->flags & PF_SUPERPRIV)
-               ac.ac_flag |= ASU;
-       if (current->flags & PF_DUMPCORE)
-               ac.ac_flag |= ACORE;
-       if (current->flags & PF_SIGNALED)
-               ac.ac_flag |= AXSIG;
-
-       vsize = 0;
-       if (current->mm) {
-               struct vm_area_struct *vma;
-               down_read(&current->mm->mmap_sem);
-               vma = current->mm->mmap;
-               while (vma) {
-                       vsize += vma->vm_end - vma->vm_start;
-                       vma = vma->vm_next;
-               }
-               up_read(&current->mm->mmap_sem);
-       }
-       vsize = vsize / 1024;
-       ac.ac_mem = encode_comp_t(vsize);
+       spin_lock_irq(&current->sighand->siglock);
+       tty = current->signal->tty;
+       ac.ac_tty = tty ? old_encode_dev(tty_devnum(tty)) : 0;
+       ac.ac_utime = encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct->ac_utime)));
+       ac.ac_stime = encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct->ac_stime)));
+       ac.ac_flag = pacct->ac_flag;
+       ac.ac_mem = encode_comp_t(pacct->ac_mem);
+       ac.ac_minflt = encode_comp_t(pacct->ac_minflt);
+       ac.ac_majflt = encode_comp_t(pacct->ac_majflt);
+       ac.ac_exitcode = pacct->ac_exitcode;
+       spin_unlock_irq(&current->sighand->siglock);
        ac.ac_io = encode_comp_t(0 /* current->io_usage */);    /* %% */
        ac.ac_rw = encode_comp_t(ac.ac_io / 1024);
-       ac.ac_minflt = encode_comp_t(current->signal->min_flt +
-                                    current->min_flt);
-       ac.ac_majflt = encode_comp_t(current->signal->maj_flt +
-                                    current->maj_flt);
        ac.ac_swaps = encode_comp_t(0);
-       ac.ac_exitcode = exitcode;
 
        /*
-         * Kernel segment override to datasegment and write it
-         * to the accounting file.
-         */
+        * Kernel segment override to datasegment and write it
+        * to the accounting file.
+        */
        fs = get_fs();
        set_fs(KERNEL_DS);
        /*
-        * Accounting records are not subject to resource limits.
-        */
+        * Accounting records are not subject to resource limits.
+        */
        flim = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
        current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
        file->f_op->write(file, (char *)&ac,
@@ -546,14 +526,67 @@ static void do_acct_process(long exitcode, struct file *file)
 }
 
 /**
+ * acct_init_pacct - initialize a new pacct_struct
+ * @pacct: per-process accounting info struct to initialize
+ */
+void acct_init_pacct(struct pacct_struct *pacct)
+{
+       memset(pacct, 0, sizeof(struct pacct_struct));
+       pacct->ac_utime = pacct->ac_stime = cputime_zero;
+}
+
+/**
+ * acct_collect - collect accounting information into pacct_struct
+ * @exitcode: task exit code
+ * @group_dead: not 0, if this thread is the last one in the process.
+ */
+void acct_collect(long exitcode, int group_dead)
+{
+       struct pacct_struct *pacct = &current->signal->pacct;
+       unsigned long vsize = 0;
+
+       if (group_dead && current->mm) {
+               struct vm_area_struct *vma;
+               down_read(&current->mm->mmap_sem);
+               vma = current->mm->mmap;
+               while (vma) {
+                       vsize += vma->vm_end - vma->vm_start;
+                       vma = vma->vm_next;
+               }
+               up_read(&current->mm->mmap_sem);
+       }
+
+       spin_lock_irq(&current->sighand->siglock);
+       if (group_dead)
+               pacct->ac_mem = vsize / 1024;
+       if (thread_group_leader(current)) {
+               pacct->ac_exitcode = exitcode;
+               if (current->flags & PF_FORKNOEXEC)
+                       pacct->ac_flag |= AFORK;
+       }
+       if (current->flags & PF_SUPERPRIV)
+               pacct->ac_flag |= ASU;
+       if (current->flags & PF_DUMPCORE)
+               pacct->ac_flag |= ACORE;
+       if (current->flags & PF_SIGNALED)
+               pacct->ac_flag |= AXSIG;
+       pacct->ac_utime = cputime_add(pacct->ac_utime, current->utime);
+       pacct->ac_stime = cputime_add(pacct->ac_stime, current->stime);
+       pacct->ac_minflt += current->min_flt;
+       pacct->ac_majflt += current->maj_flt;
+       spin_unlock_irq(&current->sighand->siglock);
+}
+
+/**
  * acct_process - now just a wrapper around do_acct_process
  * @exitcode: task exit code
  *
  * handles process accounting for an exiting task
  */
-void acct_process(long exitcode)
+void acct_process(void)
 {
        struct file *file = NULL;
+       struct pid_namespace *ns;
 
        /*
         * accelerate the common fastpath:
@@ -568,38 +601,10 @@ void acct_process(long exitcode)
                return;
        }
        get_file(file);
+       ns = get_pid_ns(acct_globals.ns);
        spin_unlock(&acct_globals.lock);
 
-       do_acct_process(exitcode, file);
+       do_acct_process(ns, file);
        fput(file);
-}
-
-
-/**
- * acct_update_integrals - update mm integral fields in task_struct
- * @tsk: task_struct for accounting
- */
-void acct_update_integrals(struct task_struct *tsk)
-{
-       if (likely(tsk->mm)) {
-               long delta =
-                       cputime_to_jiffies(tsk->stime) - tsk->acct_stimexpd;
-
-               if (delta == 0)
-                       return;
-               tsk->acct_stimexpd = tsk->stime;
-               tsk->acct_rss_mem1 += delta * get_mm_rss(tsk->mm);
-               tsk->acct_vm_mem1 += delta * tsk->mm->total_vm;
-       }
-}
-
-/**
- * acct_clear_integrals - clear the mm integral fields in task_struct
- * @tsk: task_struct whose accounting fields are cleared
- */
-void acct_clear_integrals(struct task_struct *tsk)
-{
-       tsk->acct_stimexpd = 0;
-       tsk->acct_rss_mem1 = 0;
-       tsk->acct_vm_mem1 = 0;
+       put_pid_ns(ns);
 }