/**
* API independent part for page fault initialization.
* \param env - corespondent lu_env to processing
* \param vma - virtual memory area addressed to page fault
* \param index - page index corespondent to fault.
* \parm ra_flags - vma readahead flags.
*
* \return error codes from cl_io_init.
*/
static struct cl_io *
ll_fault_io_init(struct lu_env *env, struct vm_area_struct *vma,
pgoff_t index, unsigned long *ra_flags)
{
struct file *file = vma->vm_file;
struct inode *inode = file_inode(file);
struct cl_io *io;
struct cl_fault_io *fio;
int rc;
ENTRY;
if (ll_file_nolock(file))
RETURN(ERR_PTR(-EOPNOTSUPP));
restart:
io = vvp_env_thread_io(env);
io->ci_obj = ll_i2info(inode)->lli_clob;
LASSERT(io->ci_obj != NULL);
fio = &io->u.ci_fault;
fio->ft_index = index;
fio->ft_executable = vma->vm_flags&VM_EXEC;
/*
* disable VM_SEQ_READ and use VM_RAND_READ to make sure that
* the kernel will not read other pages not covered by ldlm in
* filemap_nopage. we do our readahead in ll_readpage.
*/
if (ra_flags != NULL)
*ra_flags = vma->vm_flags & (VM_RAND_READ|VM_SEQ_READ);
vma->vm_flags &= ~VM_SEQ_READ;
vma->vm_flags |= VM_RAND_READ;
CDEBUG(D_MMAP, "vm_flags: %lx (%lu %d)\n", vma->vm_flags,
fio->ft_index, fio->ft_executable);
rc = cl_io_init(env, io, CIT_FAULT, io->ci_obj);
if (rc == 0) {
struct vvp_io *vio = vvp_env_io(env);
struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
LASSERT(vio->vui_cl.cis_io == io);
/* mmap lock must be MANDATORY it has to cache
* pages. */
io->ci_lockreq = CILR_MANDATORY;
vio->vui_fd = fd;
} else {
LASSERT(rc < 0);
cl_io_fini(env, io);
if (io->ci_need_restart)
goto restart;
io = ERR_PTR(rc);
}
RETURN(io);
}
int ll_file_mmap(struct file *file, struct vm_area_struct *vma)
{
struct inode *inode = file_inode(file);
int rc;
if (ll_file_nolock(file))
return -EOPNOTSUPP;
ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_MAP, 1);
rc = generic_file_mmap(file, vma);
if (rc == 0) {
vma->vm_ops = &ll_file_vm_ops;
vma->vm_ops->open(vma);
/* update the inode's size and mtime */
rc = ll_glimpse_size(inode);
}
return rc;
}
/**
* API independent part for page fault initialization.
* \param vma - virtual memory area addressed to page fault
* \param env - corespondent lu_env to processing
* \param nest - nested level
* \param index - page index corespondent to fault.
* \parm ra_flags - vma readahead flags.
*
* \return allocated and initialized env for fault operation.
* \retval EINVAL if env can't allocated
* \return other error codes from cl_io_init.
*/
static struct cl_io *
ll_fault_io_init(struct vm_area_struct *vma, struct lu_env **env_ret,
struct cl_env_nest *nest, pgoff_t index,
unsigned long *ra_flags)
{
struct file *file = vma->vm_file;
struct inode *inode = file_inode(file);
struct cl_io *io;
struct cl_fault_io *fio;
struct lu_env *env;
int rc;
*env_ret = NULL;
if (ll_file_nolock(file))
return ERR_PTR(-EOPNOTSUPP);
/*
* page fault can be called when lustre IO is
* already active for the current thread, e.g., when doing read/write
* against user level buffer mapped from Lustre buffer. To avoid
* stomping on existing context, optionally force an allocation of a new
* one.
*/
env = cl_env_nested_get(nest);
if (IS_ERR(env))
return ERR_PTR(-EINVAL);
*env_ret = env;
io = ccc_env_thread_io(env);
io->ci_obj = ll_i2info(inode)->lli_clob;
LASSERT(io->ci_obj);
fio = &io->u.ci_fault;
fio->ft_index = index;
fio->ft_executable = vma->vm_flags&VM_EXEC;
/*
* disable VM_SEQ_READ and use VM_RAND_READ to make sure that
* the kernel will not read other pages not covered by ldlm in
* filemap_nopage. we do our readahead in ll_readpage.
*/
if (ra_flags)
*ra_flags = vma->vm_flags & (VM_RAND_READ|VM_SEQ_READ);
vma->vm_flags &= ~VM_SEQ_READ;
vma->vm_flags |= VM_RAND_READ;
CDEBUG(D_MMAP, "vm_flags: %lx (%lu %d)\n", vma->vm_flags,
fio->ft_index, fio->ft_executable);
rc = cl_io_init(env, io, CIT_FAULT, io->ci_obj);
if (rc == 0) {
struct ccc_io *cio = ccc_env_io(env);
struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
LASSERT(cio->cui_cl.cis_io == io);
/* mmap lock must be MANDATORY it has to cache pages. */
io->ci_lockreq = CILR_MANDATORY;
cio->cui_fd = fd;
} else {
LASSERT(rc < 0);
cl_io_fini(env, io);
cl_env_nested_put(nest, env);
io = ERR_PTR(rc);
}
return io;
}
static int vvp_mmap_locks(const struct lu_env *env,
struct ccc_io *vio, struct cl_io *io)
{
struct ccc_thread_info *cti = ccc_env_info(env);
struct mm_struct *mm = current->mm;
struct vm_area_struct *vma;
struct cl_lock_descr *descr = &cti->cti_descr;
ldlm_policy_data_t policy;
unsigned long addr;
unsigned long seg;
ssize_t count;
int result;
ENTRY;
LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
if (!cl_is_normalio(env, io))
RETURN(0);
if (vio->cui_iov == NULL) /* nfs or loop back device write */
RETURN(0);
/* No MM (e.g. NFS)? No vmas too. */
if (mm == NULL)
RETURN(0);
for (seg = 0; seg < vio->cui_nrsegs; seg++) {
const struct iovec *iv = &vio->cui_iov[seg];
addr = (unsigned long)iv->iov_base;
count = iv->iov_len;
if (count == 0)
continue;
count += addr & (~CFS_PAGE_MASK);
addr &= CFS_PAGE_MASK;
down_read(&mm->mmap_sem);
while((vma = our_vma(mm, addr, count)) != NULL) {
struct inode *inode = vma->vm_file->f_dentry->d_inode;
int flags = CEF_MUST;
if (ll_file_nolock(vma->vm_file)) {
/*
* For no lock case, a lockless lock will be
* generated.
*/
flags = CEF_NEVER;
}
/*
* XXX: Required lock mode can be weakened: CIT_WRITE
* io only ever reads user level buffer, and CIT_READ
* only writes on it.
*/
policy_from_vma(&policy, vma, addr, count);
descr->cld_mode = vvp_mode_from_vma(vma);
descr->cld_obj = ll_i2info(inode)->lli_clob;
descr->cld_start = cl_index(descr->cld_obj,
policy.l_extent.start);
descr->cld_end = cl_index(descr->cld_obj,
policy.l_extent.end);
descr->cld_enq_flags = flags;
result = cl_io_lock_alloc_add(env, io, descr);
CDEBUG(D_VFSTRACE, "lock: %d: [%lu, %lu]\n",
descr->cld_mode, descr->cld_start,
descr->cld_end);
if (result < 0)
RETURN(result);
if (vma->vm_end - addr >= count)
break;
count -= vma->vm_end - addr;
addr = vma->vm_end;
}
up_read(&mm->mmap_sem);
}
RETURN(0);
}
