minix/servers/vfs/vnode.c
Thomas Veerman 7c8b3ddfed VFS: fix locking bugs
.sync and fsync used unnecessarily restrictive locking type
.fsync violated locking order by obtaining a vmnt lock after a filp lock
.fsync contained a TOCTOU bug
.new_node violated locking rules (didn't upgrade lock upon file creation)
.do_pipe used unnecessarily restrictive locking type
.always lock pipes exclusively; even a read operation might require to do
 a write on a vnode object (update pipe size)
.when opening a file with O_TRUNC, upgrade vnode lock when truncating
.utime used unnecessarily restrictive locking type
.path parsing:
  .always acquire VMNT_WRITE or VMNT_EXCL on vmnt and downgrade to
   VMNT_READ if that was what was actually requested. This prevents the
   following deadlock scenario:
   thread A:
     lock_vmnt(vmp, TLL_READSER);
     lock_vnode(vp, TLL_READSER);
     upgrade_vmnt_lock(vmp, TLL_WRITE);

   thread B:
     lock_vmnt(vmp, TLL_READ);
     lock_vnode(vp, TLL_READSER);

   thread A will be stuck in upgrade_vmnt_lock and thread B is stuck in
   lock_vnode. This happens when, for example, thread A tries create a
   new node (open.c:new_node) and thread B tries to do eat_path to
   change dir (stadir.c:do_chdir). When the path is being resolved, a
   vnode is always locked with VNODE_OPCL (TLL_READSER) and then
   downgraded to VNODE_READ if read-only is actually requested. Thread
   A locks the vmnt with VMNT_WRITE (TLL_READSER) which still allows
   VMNT_READ locks. Thread B can't acquire a lock on the vnode because
   thread A has it; Thread A can't upgrade its vmnt lock to VMNT_WRITE
   (TLL_WRITE) because thread B has a VMNT_READ lock on it.

   By serializing vmnt locks during path parsing, thread B can only
   acquire a lock on vmp when thread A has completely finished its
   operation.
2013-01-11 09:18:35 +00:00

317 lines
8.7 KiB
C

/* This file contains the routines related to vnodes.
* The entry points are:
*
* get_vnode - increase counter and get details of an inode
* get_free_vnode - get a pointer to a free vnode obj
* find_vnode - find a vnode according to the FS endpoint and the inode num.
* dup_vnode - duplicate vnode (i.e. increase counter)
* put_vnode - drop vnode (i.e. decrease counter)
*/
#include "fs.h"
#include "threads.h"
#include "vnode.h"
#include "vmnt.h"
#include "fproc.h"
#include "file.h"
#include <minix/vfsif.h>
#include <assert.h>
/* Is vnode pointer reasonable? */
#if NDEBUG
#define SANEVP(v)
#define CHECKVN(v)
#define ASSERTVP(v)
#else
#define SANEVP(v) ((((v) >= &vnode[0] && (v) < &vnode[NR_VNODES])))
#define BADVP(v, f, l) printf("%s:%d: bad vp %p\n", f, l, v)
/* vp check that returns 0 for use in check_vrefs() */
#define CHECKVN(v) if(!SANEVP(v)) { \
BADVP(v, __FILE__, __LINE__); \
return 0; \
}
/* vp check that panics */
#define ASSERTVP(v) if(!SANEVP(v)) { \
BADVP(v, __FILE__, __LINE__); panic("bad vp"); }
#endif
#if LOCK_DEBUG
/*===========================================================================*
* check_vnode_locks_by_me *
*===========================================================================*/
void check_vnode_locks_by_me(struct fproc *rfp)
{
/* Check whether this thread still has locks held on vnodes */
struct vnode *vp;
for (vp = &vnode[0]; vp < &vnode[NR_VNODES]; vp++) {
if (tll_locked_by_me(&vp->v_lock)) {
panic("Thread %d still holds vnode lock on vp %x call_nr=%d\n",
mthread_self(), vp, job_call_nr);
}
}
if (rfp->fp_vp_rdlocks != 0)
panic("Thread %d still holds read locks on a vnode (%d) call_nr=%d\n",
mthread_self(), rfp->fp_vp_rdlocks, job_call_nr);
}
#endif
/*===========================================================================*
* check_vnode_locks *
*===========================================================================*/
void check_vnode_locks()
{
struct vnode *vp;
int count = 0;
for (vp = &vnode[0]; vp < &vnode[NR_VNODES]; vp++)
if (is_vnode_locked(vp)) {
count++;
}
if (count) panic("%d locked vnodes\n", count);
#if 0
printf("check_vnode_locks OK\n");
#endif
}
/*===========================================================================*
* get_free_vnode *
*===========================================================================*/
struct vnode *get_free_vnode()
{
/* Find a free vnode slot in the vnode table (it's not actually allocated) */
struct vnode *vp;
for (vp = &vnode[0]; vp < &vnode[NR_VNODES]; ++vp) {
if (vp->v_ref_count == 0 && !is_vnode_locked(vp)) {
vp->v_uid = -1;
vp->v_gid = -1;
vp->v_sdev = NO_DEV;
vp->v_mapfs_e = NONE;
vp->v_mapfs_count = 0;
vp->v_mapinode_nr = 0;
return(vp);
}
}
err_code = ENFILE;
return(NULL);
}
/*===========================================================================*
* find_vnode *
*===========================================================================*/
struct vnode *find_vnode(int fs_e, ino_t ino)
{
/* Find a specified (FS endpoint and inode number) vnode in the
* vnode table */
struct vnode *vp;
for (vp = &vnode[0]; vp < &vnode[NR_VNODES]; ++vp)
if (vp->v_ref_count > 0 && vp->v_inode_nr == ino && vp->v_fs_e == fs_e)
return(vp);
return(NULL);
}
/*===========================================================================*
* is_vnode_locked *
*===========================================================================*/
int is_vnode_locked(struct vnode *vp)
{
/* Find out whether a thread holds a lock on this vnode or is trying to obtain
* a lock. */
ASSERTVP(vp);
return(tll_islocked(&vp->v_lock) || tll_haspendinglock(&vp->v_lock));
}
/*===========================================================================*
* init_vnodes *
*===========================================================================*/
void init_vnodes(void)
{
struct vnode *vp;
for (vp = &vnode[0]; vp < &vnode[NR_VNODES]; ++vp) {
vp->v_fs_e = NONE;
vp->v_mapfs_e = NONE;
vp->v_inode_nr = 0;
vp->v_ref_count = 0;
vp->v_fs_count = 0;
vp->v_mapfs_count = 0;
tll_init(&vp->v_lock);
}
}
/*===========================================================================*
* lock_vnode *
*===========================================================================*/
int lock_vnode(struct vnode *vp, tll_access_t locktype)
{
int r;
ASSERTVP(vp);
r = tll_lock(&vp->v_lock, locktype);
#if LOCK_DEBUG
if (locktype == VNODE_READ) {
fp->fp_vp_rdlocks++;
}
#endif
if (r == EBUSY) return(r);
return(OK);
}
/*===========================================================================*
* unlock_vnode *
*===========================================================================*/
void unlock_vnode(struct vnode *vp)
{
#if LOCK_DEBUG
int i;
register struct vnode *rvp;
struct worker_thread *w;
#endif
ASSERTVP(vp);
#if LOCK_DEBUG
/* Decrease read-only lock counter when not locked as VNODE_OPCL or
* VNODE_WRITE */
if (!tll_locked_by_me(&vp->v_lock)) {
fp->fp_vp_rdlocks--;
}
for (i = 0; i < NR_VNODES; i++) {
rvp = &vnode[i];
w = rvp->v_lock.t_write;
assert(w != self);
while (w && w->w_next != NULL) {
w = w->w_next;
assert(w != self);
}
w = rvp->v_lock.t_serial;
assert(w != self);
while (w && w->w_next != NULL) {
w = w->w_next;
assert(w != self);
}
}
#endif
tll_unlock(&vp->v_lock);
}
/*===========================================================================*
* vnode *
*===========================================================================*/
void upgrade_vnode_lock(struct vnode *vp)
{
ASSERTVP(vp);
tll_upgrade(&vp->v_lock);
}
/*===========================================================================*
* dup_vnode *
*===========================================================================*/
void dup_vnode(struct vnode *vp)
{
/* dup_vnode() is called to increment the vnode and therefore the
* referred inode's counter.
*/
ASSERTVP(vp);
vp->v_ref_count++;
}
/*===========================================================================*
* put_vnode *
*===========================================================================*/
void put_vnode(struct vnode *vp)
{
/* Decrease vnode's usage counter and decrease inode's usage counter in the
* corresponding FS process. Decreasing the fs_count each time we decrease the
* ref count would lead to poor performance. Instead, only decrease fs_count
* when the ref count hits zero. However, this could lead to fs_count to wrap.
* To prevent this, we drop the counter to 1 when the counter hits 256.
* We maintain fs_count as a sanity check to make sure VFS and the FS are in
* sync.
*/
int r, lock_vp;
ASSERTVP(vp);
/* Lock vnode. It's quite possible this thread already has a lock on this
* vnode. That's no problem, because the reference counter will not decrease
* to zero in that case. However, if the counter does decrease to zero *and*
* is already locked, we have a consistency problem somewhere. */
lock_vp = lock_vnode(vp, VNODE_OPCL);
if (vp->v_ref_count > 1) {
/* Decrease counter */
vp->v_ref_count--;
if (vp->v_fs_count > 256)
vnode_clean_refs(vp);
if (lock_vp != EBUSY) unlock_vnode(vp);
return;
}
/* If we already had a lock, there is a consistency problem */
assert(lock_vp != EBUSY);
upgrade_vnode_lock(vp); /* Acquire exclusive access */
/* A vnode that's not in use can't be put back. */
if (vp->v_ref_count <= 0)
panic("put_vnode failed: bad v_ref_count %d\n", vp->v_ref_count);
/* fs_count should indicate that the file is in use. */
if (vp->v_fs_count <= 0)
panic("put_vnode failed: bad v_fs_count %d\n", vp->v_fs_count);
/* Tell FS we don't need this inode to be open anymore. */
r = req_putnode(vp->v_fs_e, vp->v_inode_nr, vp->v_fs_count);
if (r != OK) {
printf("VFS: putnode failed: %d\n", r);
util_stacktrace();
}
/* This inode could've been mapped. If so, tell mapped FS to close it as
* well. If mapped onto same FS, this putnode is not needed. */
if (vp->v_mapfs_e != NONE && vp->v_mapfs_e != vp->v_fs_e)
req_putnode(vp->v_mapfs_e, vp->v_mapinode_nr, vp->v_mapfs_count);
vp->v_fs_count = 0;
vp->v_ref_count = 0;
vp->v_mapfs_count = 0;
unlock_vnode(vp);
}
/*===========================================================================*
* vnode_clean_refs *
*===========================================================================*/
void vnode_clean_refs(struct vnode *vp)
{
/* Tell the underlying FS to drop all reference but one. */
if (vp == NULL) return;
if (vp->v_fs_count <= 1) return; /* Nothing to do */
/* Drop all references except one */
req_putnode(vp->v_fs_e, vp->v_inode_nr, vp->v_fs_count - 1);
vp->v_fs_count = 1;
}