/**
* 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;
restart:
io = vvp_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 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;
cl_env_nested_put(nest, env);
io = ERR_PTR(rc);
}
return io;
}
/**
* Lock upcall function that is executed either when a reply to ENQUEUE rpc is
* received from a server, or after osc_enqueue_base() matched a local DLM
* lock.
*/
static int osc_lock_upcall(void *cookie, int errcode)
{
struct osc_lock *olck = cookie;
struct cl_lock_slice *slice = &olck->ols_cl;
struct cl_lock *lock = slice->cls_lock;
struct lu_env *env;
struct cl_env_nest nest;
ENTRY;
env = cl_env_nested_get(&nest);
if (!IS_ERR(env)) {
int rc;
cl_lock_mutex_get(env, lock);
LASSERT(lock->cll_state >= CLS_QUEUING);
if (olck->ols_state == OLS_ENQUEUED) {
olck->ols_state = OLS_UPCALL_RECEIVED;
rc = ldlm_error2errno(errcode);
} else if (olck->ols_state == OLS_CANCELLED) {
rc = -EIO;
} else {
CERROR("Impossible state: %d\n", olck->ols_state);
LBUG();
}
if (rc) {
struct ldlm_lock *dlmlock;
dlmlock = ldlm_handle2lock(&olck->ols_handle);
if (dlmlock != NULL) {
lock_res_and_lock(dlmlock);
spin_lock(&osc_ast_guard);
LASSERT(olck->ols_lock == NULL);
dlmlock->l_ast_data = NULL;
olck->ols_handle.cookie = 0ULL;
spin_unlock(&osc_ast_guard);
ldlm_lock_fail_match_locked(dlmlock);
unlock_res_and_lock(dlmlock);
LDLM_LOCK_PUT(dlmlock);
}
} else {
if (olck->ols_glimpse)
olck->ols_glimpse = 0;
osc_lock_upcall0(env, olck);
}
/* Error handling, some errors are tolerable. */
if (olck->ols_locklessable && rc == -EUSERS) {
/* This is a tolerable error, turn this lock into
* lockless lock.
*/
osc_object_set_contended(cl2osc(slice->cls_obj));
LASSERT(slice->cls_ops == &osc_lock_ops);
/* Change this lock to ldlmlock-less lock. */
osc_lock_to_lockless(env, olck, 1);
olck->ols_state = OLS_GRANTED;
rc = 0;
} else if (olck->ols_glimpse && rc == -ENAVAIL) {
osc_lock_lvb_update(env, olck, rc);
cl_lock_delete(env, lock);
/* Hide the error. */
rc = 0;
}
if (rc == 0) {
/* For AGL case, the RPC sponsor may exits the cl_lock
* processing without wait() called before related OSC
* lock upcall(). So update the lock status according
* to the enqueue result inside AGL upcall(). */
if (olck->ols_agl) {
lock->cll_flags |= CLF_FROM_UPCALL;
cl_wait_try(env, lock);
lock->cll_flags &= ~CLF_FROM_UPCALL;
if (!olck->ols_glimpse)
olck->ols_agl = 0;
}
cl_lock_signal(env, lock);
/* del user for lock upcall cookie */
cl_unuse_try(env, lock);
} else {
/* del user for lock upcall cookie */
cl_lock_user_del(env, lock);
cl_lock_error(env, lock, rc);
}
/* release cookie reference, acquired by osc_lock_enqueue() */
cl_lock_hold_release(env, lock, "upcall", lock);
cl_lock_mutex_put(env, lock);
lu_ref_del(&lock->cll_reference, "upcall", lock);
/* This maybe the last reference, so must be called after
* cl_lock_mutex_put(). */
cl_lock_put(env, lock);
cl_env_nested_put(&nest, env);
} else {
/* should never happen, similar to osc_ldlm_blocking_ast(). */
LBUG();
}
RETURN(errcode);
}
/* Sharing code of page_mkwrite method for rhel5 and rhel6 */
static int ll_page_mkwrite0(struct vm_area_struct *vma, struct page *vmpage,
bool *retry)
{
struct lu_env *env;
struct cl_io *io;
struct vvp_io *vio;
struct cl_env_nest nest;
int result;
sigset_t set;
struct inode *inode;
struct ll_inode_info *lli;
io = ll_fault_io_init(vma, &env, &nest, vmpage->index, NULL);
if (IS_ERR(io)) {
result = PTR_ERR(io);
goto out;
}
result = io->ci_result;
if (result < 0)
goto out_io;
io->u.ci_fault.ft_mkwrite = 1;
io->u.ci_fault.ft_writable = 1;
vio = vvp_env_io(env);
vio->u.fault.ft_vma = vma;
vio->u.fault.ft_vmpage = vmpage;
set = cfs_block_sigsinv(sigmask(SIGKILL) | sigmask(SIGTERM));
inode = vvp_object_inode(io->ci_obj);
lli = ll_i2info(inode);
result = cl_io_loop(env, io);
cfs_restore_sigs(set);
if (result == 0) {
struct inode *inode = file_inode(vma->vm_file);
struct ll_inode_info *lli = ll_i2info(inode);
lock_page(vmpage);
if (!vmpage->mapping) {
unlock_page(vmpage);
/* page was truncated and lock was cancelled, return
* ENODATA so that VM_FAULT_NOPAGE will be returned
* to handle_mm_fault().
*/
if (result == 0)
result = -ENODATA;
} else if (!PageDirty(vmpage)) {
/* race, the page has been cleaned by ptlrpcd after
* it was unlocked, it has to be added into dirty
* cache again otherwise this soon-to-dirty page won't
* consume any grants, even worse if this page is being
* transferred because it will break RPC checksum.
*/
unlock_page(vmpage);
CDEBUG(D_MMAP, "Race on page_mkwrite %p/%lu, page has been written out, retry.\n",
vmpage, vmpage->index);
*retry = true;
result = -EAGAIN;
}
if (result == 0) {
spin_lock(&lli->lli_lock);
lli->lli_flags |= LLIF_DATA_MODIFIED;
spin_unlock(&lli->lli_lock);
}
}
out_io:
cl_io_fini(env, io);
cl_env_nested_put(&nest, env);
out:
CDEBUG(D_MMAP, "%s mkwrite with %d\n", current->comm, result);
LASSERT(ergo(result == 0, PageLocked(vmpage)));
return result;
}
