static inline uint8_t flock_lock_new(inodelocks *il,uint8_t ltype,uint32_t sessionid,uint32_t msgid,uint32_t reqid,uint64_t owner) {
lock *l;
l = malloc(sizeof(lock));
l->owner = owner;
l->sessionid = sessionid;
l->state = STATE_ACTIVE;
l->ltype = ltype;
l->lock_instances = NULL;
l->parent = il;
l->next = NULL;
l->prev = NULL;
if (flock_check(il,ltype)) {
l->state = STATE_WAITING;
flock_lock_append_req(l,msgid,reqid);
flock_lock_inode_attach(l);
return MFS_ERROR_WAITING;
}
flock_lock_inode_attach(l);
return MFS_STATUS_OK;
}
uint8_t flock_locks_cmd(uint32_t sessionid,uint32_t msgid,uint32_t reqid,uint32_t inode,uint64_t owner,uint8_t op) {
inodelocks *il;
lock *l,*nl;
uint8_t ltype;
// flock_dump();
// syslog(LOG_NOTICE,"flock op: sessionid:%"PRIu32",msgid:%"PRIu32",reqid:%"PRIu32",inode:%"PRIu32",owner:%"PRIu64",op:%u",sessionid,msgid,reqid,inode,owner,op);
if (op!=FLOCK_INTERRUPT && op!=FLOCK_RELEASE) {
if (of_checknode(sessionid,inode)==0) {
return MFS_ERROR_EINVAL; // EBADF ?
}
}
il = flock_inode_find(inode);
if (il==NULL) {
if (op==FLOCK_UNLOCK || op==FLOCK_INTERRUPT || op==FLOCK_RELEASE) {
return MFS_STATUS_OK;
}
il = flock_inode_new(inode);
}
if (op==FLOCK_INTERRUPT) {
l = il->waiting_head;
while (l) {
nl = l->next;
if (l->sessionid==sessionid && l->owner==owner) {
flock_lock_wake_up_one(l,reqid,MFS_ERROR_EINTR);
if (l->lock_instances==NULL) { // remove
flock_lock_remove(l);
}
}
l = nl;
}
return MFS_STATUS_OK;
}
for (l=il->active ; l ; l=l->next) {
if (l->sessionid==sessionid && l->owner==owner) {
if (op==FLOCK_UNLOCK || op==FLOCK_RELEASE) {
flock_lock_unlock(l);
return MFS_STATUS_OK;
} else if (op==FLOCK_TRY_SHARED) {
if (l->ltype==LTYPE_READER) {
return MFS_STATUS_OK;
} else { // l->ltype==LTYPE_WRITER
l->ltype=LTYPE_READER;
flock_lock_check_waiting(il);
return MFS_STATUS_OK;
}
} else if (op==FLOCK_LOCK_SHARED) {
if (l->ltype==LTYPE_READER) {
return MFS_STATUS_OK;
} else { // l->ltype==LTYPE_WRITER
flock_lock_unlock(l);
return flock_lock_new(il,LTYPE_READER,sessionid,msgid,reqid,owner);
}
} else if (op==FLOCK_TRY_EXCLUSIVE) {
if (l->ltype==LTYPE_WRITER) {
return MFS_STATUS_OK;
} else { // l->ltype==LTYPE_READER
if (il->active->next==NULL) { // this lock is the only one
l->ltype=LTYPE_WRITER;
return MFS_STATUS_OK;
}
return MFS_ERROR_EAGAIN;
}
} else if (op==FLOCK_LOCK_EXCLUSIVE) {
if (l->ltype==LTYPE_WRITER) {
return MFS_STATUS_OK;
} else { // l->ltype==LTYPE_READER
flock_lock_unlock(l);
return flock_lock_new(il,LTYPE_WRITER,sessionid,msgid,reqid,owner);
}
}
return MFS_ERROR_EINVAL;
}
}
for (l=il->waiting_head ; l ; l=l->next) {
if (l->sessionid==sessionid && l->owner==owner) {
if (op==FLOCK_RELEASE) {
flock_lock_wake_up_all(l,MFS_ERROR_ECANCELED);
flock_lock_remove(l);
return MFS_STATUS_OK;
} else if (op==FLOCK_UNLOCK) {
if (FlocksMode==MODE_CORRECT) {
// logically this call should do this:
flock_lock_wake_up_all(l,MFS_ERROR_ECANCELED);
flock_lock_remove(l);
}
// but in all tested kernels it was just ignored
return MFS_STATUS_OK;
} else if (op==FLOCK_TRY_SHARED || op==FLOCK_TRY_EXCLUSIVE) {
return MFS_ERROR_EAGAIN;
} else if (op==FLOCK_LOCK_SHARED) {
if (l->ltype==LTYPE_READER) {
flock_lock_append_req(l,msgid,reqid);
return MFS_ERROR_WAITING;
} else {
// return MFS_ERROR_EINVAL;
flock_lock_wake_up_all(l,MFS_ERROR_ECANCELED);
l->ltype=LTYPE_READER;
flock_lock_append_req(l,msgid,reqid);
return MFS_ERROR_WAITING;
}
} else if (op==FLOCK_LOCK_EXCLUSIVE) {
if (l->ltype==LTYPE_WRITER) {
flock_lock_append_req(l,msgid,reqid);
return MFS_ERROR_WAITING;
} else {
// return MFS_ERROR_EINVAL;
flock_lock_wake_up_all(l,MFS_ERROR_ECANCELED);
l->ltype=LTYPE_WRITER;
flock_lock_append_req(l,msgid,reqid);
return MFS_ERROR_WAITING;
}
}
return MFS_ERROR_EINVAL;
}
}
if (op==FLOCK_UNLOCK || op==FLOCK_RELEASE) {
return MFS_STATUS_OK;
}
ltype = (op==FLOCK_TRY_SHARED || op==FLOCK_LOCK_SHARED)?LTYPE_READER:LTYPE_WRITER;
if (op==FLOCK_TRY_SHARED || op==FLOCK_TRY_EXCLUSIVE) {
if (flock_check(il,ltype)) {
return MFS_ERROR_EAGAIN;
}
}
return flock_lock_new(il,ltype,sessionid,msgid,reqid,owner);
}
/*
* File control.
*/
int
fcntl(int fdes, int cmd, intptr_t arg)
{
int iarg;
int error = 0;
int retval;
proc_t *p;
file_t *fp;
vnode_t *vp;
u_offset_t offset;
u_offset_t start;
struct vattr vattr;
int in_crit;
int flag;
struct flock sbf;
struct flock64 bf;
struct o_flock obf;
struct flock64_32 bf64_32;
struct fshare fsh;
struct shrlock shr;
struct shr_locowner shr_own;
offset_t maxoffset;
model_t datamodel;
int fdres;
#if defined(_ILP32) && !defined(lint) && defined(_SYSCALL32)
ASSERT(sizeof (struct flock) == sizeof (struct flock32));
ASSERT(sizeof (struct flock64) == sizeof (struct flock64_32));
#endif
#if defined(_LP64) && !defined(lint) && defined(_SYSCALL32)
ASSERT(sizeof (struct flock) == sizeof (struct flock64_64));
ASSERT(sizeof (struct flock64) == sizeof (struct flock64_64));
#endif
/*
* First, for speed, deal with the subset of cases
* that do not require getf() / releasef().
*/
switch (cmd) {
case F_GETFD:
if ((error = f_getfd_error(fdes, &flag)) == 0)
retval = flag;
goto out;
case F_SETFD:
error = f_setfd_error(fdes, (int)arg);
retval = 0;
goto out;
case F_GETFL:
if ((error = f_getfl(fdes, &flag)) == 0)
retval = (flag & (FMASK | FASYNC)) + FOPEN;
goto out;
case F_GETXFL:
if ((error = f_getfl(fdes, &flag)) == 0)
retval = flag + FOPEN;
goto out;
case F_BADFD:
if ((error = f_badfd(fdes, &fdres, (int)arg)) == 0)
retval = fdres;
goto out;
}
/*
* Second, for speed, deal with the subset of cases that
* require getf() / releasef() but do not require copyin.
*/
if ((fp = getf(fdes)) == NULL) {
error = EBADF;
goto out;
}
iarg = (int)arg;
switch (cmd) {
/* ONC_PLUS EXTRACT END */
case F_DUPFD:
p = curproc;
if ((uint_t)iarg >= p->p_fno_ctl) {
if (iarg >= 0)
fd_too_big(p);
error = EINVAL;
} else if ((retval = ufalloc_file(iarg, fp)) == -1) {
error = EMFILE;
} else {
mutex_enter(&fp->f_tlock);
fp->f_count++;
mutex_exit(&fp->f_tlock);
}
goto done;
case F_DUP2FD:
p = curproc;
if (fdes == iarg) {
retval = iarg;
} else if ((uint_t)iarg >= p->p_fno_ctl) {
if (iarg >= 0)
fd_too_big(p);
error = EBADF;
} else {
/*
* We can't hold our getf(fdes) across the call to
* closeandsetf() because it creates a window for
* deadlock: if one thread is doing dup2(a, b) while
* another is doing dup2(b, a), each one will block
* waiting for the other to call releasef(). The
* solution is to increment the file reference count
* (which we have to do anyway), then releasef(fdes),
* then closeandsetf(). Incrementing f_count ensures
* that fp won't disappear after we call releasef().
* When closeandsetf() fails, we try avoid calling
* closef() because of all the side effects.
*/
mutex_enter(&fp->f_tlock);
fp->f_count++;
mutex_exit(&fp->f_tlock);
releasef(fdes);
if ((error = closeandsetf(iarg, fp)) == 0) {
retval = iarg;
} else {
mutex_enter(&fp->f_tlock);
if (fp->f_count > 1) {
fp->f_count--;
mutex_exit(&fp->f_tlock);
} else {
mutex_exit(&fp->f_tlock);
(void) closef(fp);
}
}
goto out;
}
goto done;
case F_SETFL:
vp = fp->f_vnode;
flag = fp->f_flag;
if ((iarg & (FNONBLOCK|FNDELAY)) == (FNONBLOCK|FNDELAY))
iarg &= ~FNDELAY;
if ((error = VOP_SETFL(vp, flag, iarg, fp->f_cred)) == 0) {
iarg &= FMASK;
mutex_enter(&fp->f_tlock);
fp->f_flag &= ~FMASK | (FREAD|FWRITE);
fp->f_flag |= (iarg - FOPEN) & ~(FREAD|FWRITE);
mutex_exit(&fp->f_tlock);
}
retval = 0;
goto done;
}
/*
* Finally, deal with the expensive cases.
*/
retval = 0;
in_crit = 0;
maxoffset = MAXOFF_T;
datamodel = DATAMODEL_NATIVE;
#if defined(_SYSCALL32_IMPL)
if ((datamodel = get_udatamodel()) == DATAMODEL_ILP32)
maxoffset = MAXOFF32_T;
#endif
vp = fp->f_vnode;
flag = fp->f_flag;
offset = fp->f_offset;
switch (cmd) {
/* ONC_PLUS EXTRACT START */
/*
* The file system and vnode layers understand and implement
* locking with flock64 structures. So here once we pass through
* the test for compatibility as defined by LFS API, (for F_SETLK,
* F_SETLKW, F_GETLK, F_GETLKW, F_FREESP) we transform
* the flock structure to a flock64 structure and send it to the
* lower layers. Similarly in case of GETLK the returned flock64
* structure is transformed to a flock structure if everything fits
* in nicely, otherwise we return EOVERFLOW.
*/
case F_GETLK:
case F_O_GETLK:
case F_SETLK:
case F_SETLKW:
case F_SETLK_NBMAND:
/*
* Copy in input fields only.
*/
if (cmd == F_O_GETLK) {
if (datamodel != DATAMODEL_ILP32) {
error = EINVAL;
break;
}
if (copyin((void *)arg, &obf, sizeof (obf))) {
error = EFAULT;
break;
}
bf.l_type = obf.l_type;
bf.l_whence = obf.l_whence;
bf.l_start = (off64_t)obf.l_start;
bf.l_len = (off64_t)obf.l_len;
bf.l_sysid = (int)obf.l_sysid;
bf.l_pid = obf.l_pid;
} else if (datamodel == DATAMODEL_NATIVE) {
if (copyin((void *)arg, &sbf, sizeof (sbf))) {
error = EFAULT;
break;
}
/*
* XXX In an LP64 kernel with an LP64 application
* there's no need to do a structure copy here
* struct flock == struct flock64. However,
* we did it this way to avoid more conditional
* compilation.
*/
bf.l_type = sbf.l_type;
bf.l_whence = sbf.l_whence;
bf.l_start = (off64_t)sbf.l_start;
bf.l_len = (off64_t)sbf.l_len;
bf.l_sysid = sbf.l_sysid;
bf.l_pid = sbf.l_pid;
}
#if defined(_SYSCALL32_IMPL)
else {
struct flock32 sbf32;
if (copyin((void *)arg, &sbf32, sizeof (sbf32))) {
error = EFAULT;
break;
}
bf.l_type = sbf32.l_type;
bf.l_whence = sbf32.l_whence;
bf.l_start = (off64_t)sbf32.l_start;
bf.l_len = (off64_t)sbf32.l_len;
bf.l_sysid = sbf32.l_sysid;
bf.l_pid = sbf32.l_pid;
}
#endif /* _SYSCALL32_IMPL */
/*
* 64-bit support: check for overflow for 32-bit lock ops
*/
if ((error = flock_check(vp, &bf, offset, maxoffset)) != 0)
break;
/*
* Not all of the filesystems understand F_O_GETLK, and
* there's no need for them to know. Map it to F_GETLK.
*/
if ((error = VOP_FRLOCK(vp, (cmd == F_O_GETLK) ? F_GETLK : cmd,
&bf, flag, offset, NULL, fp->f_cred)) != 0)
break;
/*
* If command is GETLK and no lock is found, only
* the type field is changed.
*/
if ((cmd == F_O_GETLK || cmd == F_GETLK) &&
bf.l_type == F_UNLCK) {
/* l_type always first entry, always a short */
if (copyout(&bf.l_type, &((struct flock *)arg)->l_type,
sizeof (bf.l_type)))
error = EFAULT;
break;
}
if (cmd == F_O_GETLK) {
/*
* Return an SVR3 flock structure to the user.
*/
obf.l_type = (int16_t)bf.l_type;
obf.l_whence = (int16_t)bf.l_whence;
obf.l_start = (int32_t)bf.l_start;
obf.l_len = (int32_t)bf.l_len;
if (bf.l_sysid > SHRT_MAX || bf.l_pid > SHRT_MAX) {
/*
* One or both values for the above fields
* is too large to store in an SVR3 flock
* structure.
*/
error = EOVERFLOW;
break;
}
obf.l_sysid = (int16_t)bf.l_sysid;
obf.l_pid = (int16_t)bf.l_pid;
if (copyout(&obf, (void *)arg, sizeof (obf)))
error = EFAULT;
} else if (cmd == F_GETLK) {
/*
* Copy out SVR4 flock.
*/
int i;
if (bf.l_start > maxoffset || bf.l_len > maxoffset) {
error = EOVERFLOW;
break;
}
if (datamodel == DATAMODEL_NATIVE) {
for (i = 0; i < 4; i++)
sbf.l_pad[i] = 0;
/*
* XXX In an LP64 kernel with an LP64
* application there's no need to do a
* structure copy here as currently
* struct flock == struct flock64.
* We did it this way to avoid more
* conditional compilation.
*/
sbf.l_type = bf.l_type;
sbf.l_whence = bf.l_whence;
sbf.l_start = (off_t)bf.l_start;
sbf.l_len = (off_t)bf.l_len;
sbf.l_sysid = bf.l_sysid;
sbf.l_pid = bf.l_pid;
if (copyout(&sbf, (void *)arg, sizeof (sbf)))
error = EFAULT;
}
#if defined(_SYSCALL32_IMPL)
else {
struct flock32 sbf32;
if (bf.l_start > MAXOFF32_T ||
bf.l_len > MAXOFF32_T) {
error = EOVERFLOW;
break;
}
for (i = 0; i < 4; i++)
sbf32.l_pad[i] = 0;
sbf32.l_type = (int16_t)bf.l_type;
sbf32.l_whence = (int16_t)bf.l_whence;
sbf32.l_start = (off32_t)bf.l_start;
sbf32.l_len = (off32_t)bf.l_len;
sbf32.l_sysid = (int32_t)bf.l_sysid;
sbf32.l_pid = (pid32_t)bf.l_pid;
if (copyout(&sbf32,
(void *)arg, sizeof (sbf32)))
error = EFAULT;
}
#endif
}
break;
/* ONC_PLUS EXTRACT END */
case F_CHKFL:
/*
* This is for internal use only, to allow the vnode layer
* to validate a flags setting before applying it. User
* programs can't issue it.
*/
error = EINVAL;
break;
case F_ALLOCSP:
case F_FREESP:
case F_ALLOCSP64:
case F_FREESP64:
if ((flag & FWRITE) == 0) {
error = EBADF;
break;
}
if (vp->v_type != VREG) {
error = EINVAL;
break;
}
if (datamodel != DATAMODEL_ILP32 &&
(cmd == F_ALLOCSP64 || cmd == F_FREESP64)) {
error = EINVAL;
break;
}
#if defined(_ILP32) || defined(_SYSCALL32_IMPL)
if (datamodel == DATAMODEL_ILP32 &&
(cmd == F_ALLOCSP || cmd == F_FREESP)) {
struct flock32 sbf32;
/*
* For compatibility we overlay an SVR3 flock on an SVR4
* flock. This works because the input field offsets
* in "struct flock" were preserved.
*/
if (copyin((void *)arg, &sbf32, sizeof (sbf32))) {
error = EFAULT;
break;
} else {
bf.l_type = sbf32.l_type;
bf.l_whence = sbf32.l_whence;
bf.l_start = (off64_t)sbf32.l_start;
bf.l_len = (off64_t)sbf32.l_len;
bf.l_sysid = sbf32.l_sysid;
bf.l_pid = sbf32.l_pid;
}
}
#endif /* _ILP32 || _SYSCALL32_IMPL */
#if defined(_LP64)
if (datamodel == DATAMODEL_LP64 &&
(cmd == F_ALLOCSP || cmd == F_FREESP)) {
if (copyin((void *)arg, &bf, sizeof (bf))) {
error = EFAULT;
break;
}
}
#endif /* defined(_LP64) */
#if !defined(_LP64) || defined(_SYSCALL32_IMPL)
if (datamodel == DATAMODEL_ILP32 &&
(cmd == F_ALLOCSP64 || cmd == F_FREESP64)) {
if (copyin((void *)arg, &bf64_32, sizeof (bf64_32))) {
error = EFAULT;
break;
} else {
/*
* Note that the size of flock64 is different in
* the ILP32 and LP64 models, due to the l_pad
* field. We do not want to assume that the
* flock64 structure is laid out the same in
* ILP32 and LP64 environments, so we will
* copy in the ILP32 version of flock64
* explicitly and copy it to the native
* flock64 structure.
*/
bf.l_type = (short)bf64_32.l_type;
bf.l_whence = (short)bf64_32.l_whence;
bf.l_start = bf64_32.l_start;
bf.l_len = bf64_32.l_len;
bf.l_sysid = (int)bf64_32.l_sysid;
bf.l_pid = (pid_t)bf64_32.l_pid;
}
}
#endif /* !defined(_LP64) || defined(_SYSCALL32_IMPL) */
if (cmd == F_ALLOCSP || cmd == F_FREESP)
error = flock_check(vp, &bf, offset, maxoffset);
else if (cmd == F_ALLOCSP64 || cmd == F_FREESP64)
error = flock_check(vp, &bf, offset, MAXOFFSET_T);
if (error)
break;
if (vp->v_type == VREG && bf.l_len == 0 &&
bf.l_start > OFFSET_MAX(fp)) {
error = EFBIG;
break;
}
/*
* Make sure that there are no conflicting non-blocking
* mandatory locks in the region being manipulated. If
* there are such locks then return EACCES.
*/
if ((error = flock_get_start(vp, &bf, offset, &start)) != 0)
break;
if (nbl_need_check(vp)) {
u_offset_t begin;
ssize_t length;
nbl_start_crit(vp, RW_READER);
in_crit = 1;
vattr.va_mask = AT_SIZE;
if ((error = VOP_GETATTR(vp, &vattr, 0, CRED())) != 0)
break;
begin = start > vattr.va_size ? vattr.va_size : start;
length = vattr.va_size > start ? vattr.va_size - start :
start - vattr.va_size;
if (nbl_conflict(vp, NBL_WRITE, begin, length, 0)) {
error = EACCES;
break;
}
}
if (cmd == F_ALLOCSP64)
cmd = F_ALLOCSP;
else if (cmd == F_FREESP64)
cmd = F_FREESP;
error = VOP_SPACE(vp, cmd, &bf, flag, offset, fp->f_cred, NULL);
break;
#if !defined(_LP64) || defined(_SYSCALL32_IMPL)
/* ONC_PLUS EXTRACT START */
case F_GETLK64:
case F_SETLK64:
case F_SETLKW64:
case F_SETLK64_NBMAND:
/*
* Large Files: Here we set cmd as *LK and send it to
* lower layers. *LK64 is only for the user land.
* Most of the comments described above for F_SETLK
* applies here too.
* Large File support is only needed for ILP32 apps!
*/
if (datamodel != DATAMODEL_ILP32) {
error = EINVAL;
break;
}
if (cmd == F_GETLK64)
cmd = F_GETLK;
else if (cmd == F_SETLK64)
cmd = F_SETLK;
else if (cmd == F_SETLKW64)
cmd = F_SETLKW;
else if (cmd == F_SETLK64_NBMAND)
cmd = F_SETLK_NBMAND;
/*
* Note that the size of flock64 is different in the ILP32
* and LP64 models, due to the sucking l_pad field.
* We do not want to assume that the flock64 structure is
* laid out in the same in ILP32 and LP64 environments, so
* we will copy in the ILP32 version of flock64 explicitly
* and copy it to the native flock64 structure.
*/
if (copyin((void *)arg, &bf64_32, sizeof (bf64_32))) {
error = EFAULT;
break;
}
bf.l_type = (short)bf64_32.l_type;
bf.l_whence = (short)bf64_32.l_whence;
bf.l_start = bf64_32.l_start;
bf.l_len = bf64_32.l_len;
bf.l_sysid = (int)bf64_32.l_sysid;
bf.l_pid = (pid_t)bf64_32.l_pid;
if ((error = flock_check(vp, &bf, offset, MAXOFFSET_T)) != 0)
break;
if ((error = VOP_FRLOCK(vp, cmd, &bf, flag, offset,
NULL, fp->f_cred)) != 0)
break;
if ((cmd == F_GETLK) && bf.l_type == F_UNLCK) {
if (copyout(&bf.l_type, &((struct flock *)arg)->l_type,
sizeof (bf.l_type)))
error = EFAULT;
break;
}
if (cmd == F_GETLK) {
int i;
/*
* We do not want to assume that the flock64 structure
* is laid out in the same in ILP32 and LP64
* environments, so we will copy out the ILP32 version
* of flock64 explicitly after copying the native
* flock64 structure to it.
*/
for (i = 0; i < 4; i++)
bf64_32.l_pad[i] = 0;
bf64_32.l_type = (int16_t)bf.l_type;
bf64_32.l_whence = (int16_t)bf.l_whence;
bf64_32.l_start = bf.l_start;
bf64_32.l_len = bf.l_len;
bf64_32.l_sysid = (int32_t)bf.l_sysid;
bf64_32.l_pid = (pid32_t)bf.l_pid;
if (copyout(&bf64_32, (void *)arg, sizeof (bf64_32)))
error = EFAULT;
}
break;
/* ONC_PLUS EXTRACT END */
#endif /* !defined(_LP64) || defined(_SYSCALL32_IMPL) */
/* ONC_PLUS EXTRACT START */
case F_SHARE:
case F_SHARE_NBMAND:
case F_UNSHARE:
/*
* Copy in input fields only.
*/
if (copyin((void *)arg, &fsh, sizeof (fsh))) {
error = EFAULT;
break;
}
/*
* Local share reservations always have this simple form
*/
shr.s_access = fsh.f_access;
shr.s_deny = fsh.f_deny;
shr.s_sysid = 0;
shr.s_pid = ttoproc(curthread)->p_pid;
shr_own.sl_pid = shr.s_pid;
shr_own.sl_id = fsh.f_id;
shr.s_own_len = sizeof (shr_own);
shr.s_owner = (caddr_t)&shr_own;
error = VOP_SHRLOCK(vp, cmd, &shr, flag, fp->f_cred);
/* ONC_PLUS EXTRACT END */
break;
default:
error = EINVAL;
break;
}
if (in_crit)
nbl_end_crit(vp);
done:
releasef(fdes);
out:
if (error)
return (set_errno(error));
return (retval);
}
