/**
* Implementation of struct cl_req_operations::cro_attr_set() for VVP
* layer. VVP is responsible for
*
* - o_[mac]time
*
* - o_mode
*
* - o_parent_seq
*
* - o_[ug]id
*
* - o_parent_oid
*
* - o_parent_ver
*
* - o_ioepoch,
*
*/
static void vvp_req_attr_set(const struct lu_env *env,
const struct cl_req_slice *slice,
const struct cl_object *obj,
struct cl_req_attr *attr, u64 flags)
{
struct inode *inode;
struct obdo *oa;
u32 valid_flags;
oa = attr->cra_oa;
inode = vvp_object_inode(obj);
valid_flags = OBD_MD_FLTYPE;
if (slice->crs_req->crq_type == CRT_WRITE) {
if (flags & OBD_MD_FLEPOCH) {
oa->o_valid |= OBD_MD_FLEPOCH;
oa->o_ioepoch = ll_i2info(inode)->lli_ioepoch;
valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME |
OBD_MD_FLUID | OBD_MD_FLGID;
}
}
obdo_from_inode(oa, inode, valid_flags & flags);
obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_INVALID_PFID))
oa->o_parent_oid++;
memcpy(attr->cra_jobid, ll_i2info(inode)->lli_jobid,
LUSTRE_JOBID_SIZE);
}
/**
* For swapping layout. The file's layout may have changed.
* To avoid populating pages to a wrong stripe, we have to verify the
* correctness of layout. It works because swapping layout processes
* have to acquire group lock.
*/
static bool can_populate_pages(const struct lu_env *env, struct cl_io *io,
struct inode *inode)
{
struct ll_inode_info *lli = ll_i2info(inode);
struct ccc_io *cio = ccc_env_io(env);
bool rc = true;
switch (io->ci_type) {
case CIT_READ:
case CIT_WRITE:
/* don't need lock here to check lli_layout_gen as we have held
* extent lock and GROUP lock has to hold to swap layout */
if (lli->lli_layout_gen != cio->cui_layout_gen) {
io->ci_need_restart = 1;
/* this will return application a short read/write */
io->ci_continue = 0;
rc = false;
}
case CIT_FAULT:
/* fault is okay because we've already had a page. */
default:
break;
}
return rc;
}
int vvp_conf_set(const struct lu_env *env, struct cl_object *obj,
const struct cl_object_conf *conf)
{
struct ll_inode_info *lli = ll_i2info(conf->coc_inode);
if (conf->coc_opc == OBJECT_CONF_INVALIDATE) {
lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
return 0;
}
if (conf->coc_opc != OBJECT_CONF_SET)
return 0;
if (conf->u.coc_md != NULL && conf->u.coc_md->lsm != NULL) {
CDEBUG(D_VFSTRACE, "layout lock change: %u -> %u\n",
lli->lli_layout_gen,
conf->u.coc_md->lsm->lsm_layout_gen);
lli->lli_has_smd = true;
lli->lli_layout_gen = conf->u.coc_md->lsm->lsm_layout_gen;
} else {
CDEBUG(D_VFSTRACE, "layout lock destroyed: %u.\n",
lli->lli_layout_gen);
lli->lli_has_smd = false;
lli->lli_layout_gen = LL_LAYOUT_GEN_EMPTY;
}
return 0;
}
static int cl_io_get(struct inode *inode, struct lu_env **envout,
struct cl_io **ioout, int *refcheck)
{
struct lu_env *env;
struct cl_io *io;
struct ll_inode_info *lli = ll_i2info(inode);
struct cl_object *clob = lli->lli_clob;
int result;
if (S_ISREG(inode->i_mode)) {
env = cl_env_get(refcheck);
if (!IS_ERR(env)) {
io = vvp_env_thread_io(env);
io->ci_obj = clob;
*envout = env;
*ioout = io;
result = 1;
} else {
result = PTR_ERR(env);
}
} else {
result = 0;
}
return result;
}
static void ll_inode_destroy_callback(struct rcu_head *head)
{
struct inode *inode = container_of(head, struct inode, i_rcu);
struct ll_inode_info *ptr = ll_i2info(inode);
kmem_cache_free(ll_inode_cachep, ptr);
}
/**
* Implements Linux VM address_space::invalidatepage() method. This method is
* called when the page is truncate from a file, either as a result of
* explicit truncate, or when inode is removed from memory (as a result of
* final iput(), umount, or memory pressure induced icache shrinking).
*
* [0, offset] bytes of the page remain valid (this is for a case of not-page
* aligned truncate). Lustre leaves partially truncated page in the cache,
* relying on struct inode::i_size to limit further accesses.
*/
static void ll_invalidatepage(struct page *vmpage, unsigned int offset,
unsigned int length)
{
struct inode *inode;
struct lu_env *env;
struct cl_page *page;
struct cl_object *obj;
LASSERT(PageLocked(vmpage));
LASSERT(!PageWriteback(vmpage));
/*
* It is safe to not check anything in invalidatepage/releasepage
* below because they are run with page locked and all our io is
* happening with locked page too
*/
if (offset == 0 && length == PAGE_SIZE) {
/* See the comment in ll_releasepage() */
env = cl_env_percpu_get();
LASSERT(!IS_ERR(env));
inode = vmpage->mapping->host;
obj = ll_i2info(inode)->lli_clob;
if (obj) {
page = cl_vmpage_page(vmpage, obj);
if (page) {
cl_page_delete(env, page);
cl_page_put(env, page);
}
} else {
LASSERT(vmpage->private == 0);
}
cl_env_percpu_put(env);
}
int vvp_object_invariant(const struct cl_object *obj)
{
struct inode *inode = vvp_object_inode(obj);
struct ll_inode_info *lli = ll_i2info(inode);
return (S_ISREG(inode->i_mode) || inode->i_mode == 0) &&
lli->lli_clob == obj;
}
/**
* 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);
}
static void vvp_req_attr_set(const struct lu_env *env, struct cl_object *obj,
struct cl_req_attr *attr)
{
struct inode *inode;
struct obdo *oa;
u64 valid_flags = OBD_MD_FLTYPE;
oa = attr->cra_oa;
inode = vvp_object_inode(obj);
if (attr->cra_type == CRT_WRITE)
valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME |
OBD_MD_FLUID | OBD_MD_FLGID;
obdo_from_inode(oa, inode, valid_flags & attr->cra_flags);
obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_INVALID_PFID))
oa->o_parent_oid++;
memcpy(attr->cra_jobid, ll_i2info(inode)->lli_jobid, LUSTRE_JOBID_SIZE);
}
struct vvp_object *cl_inode2vvp(struct inode *inode)
{
struct ll_inode_info *lli = ll_i2info(inode);
struct cl_object *obj = lli->lli_clob;
struct lu_object *lu;
lu = lu_object_locate(obj->co_lu.lo_header, &vvp_device_type);
LASSERT(lu);
return lu2vvp(lu);
}
static int ll_releasepage(struct page *vmpage, RELEASEPAGE_ARG_TYPE gfp_mask)
{
struct lu_env *env;
struct cl_object *obj;
struct cl_page *page;
struct address_space *mapping;
int result = 0;
LASSERT(PageLocked(vmpage));
if (PageWriteback(vmpage) || PageDirty(vmpage))
return 0;
mapping = vmpage->mapping;
if (mapping == NULL)
return 1;
obj = ll_i2info(mapping->host)->lli_clob;
if (obj == NULL)
return 1;
/* 1 for caller, 1 for cl_page and 1 for page cache */
if (page_count(vmpage) > 3)
return 0;
page = cl_vmpage_page(vmpage, obj);
if (page == NULL)
return 1;
env = cl_env_percpu_get();
LASSERT(!IS_ERR(env));
if (!cl_page_in_use(page)) {
result = 1;
cl_page_delete(env, page);
}
/* To use percpu env array, the call path can not be rescheduled;
* otherwise percpu array will be messed if ll_releaspage() called
* again on the same CPU.
*
* If this page holds the last refc of cl_object, the following
* call path may cause reschedule:
* cl_page_put -> cl_page_free -> cl_object_put ->
* lu_object_put -> lu_object_free -> lov_delete_raid0.
*
* However, the kernel can't get rid of this inode until all pages have
* been cleaned up. Now that we hold page lock here, it's pretty safe
* that we won't get into object delete path.
*/
LASSERT(cl_object_refc(obj) > 1);
cl_page_put(env, page);
cl_env_percpu_put(env);
return result;
}
/** records that a write is in flight */
void vvp_write_pending(struct ccc_object *club, struct ccc_page *page)
{
struct ll_inode_info *lli = ll_i2info(club->cob_inode);
spin_lock(&lli->lli_lock);
lli->lli_flags |= LLIF_SOM_DIRTY;
if (page != NULL && list_empty(&page->cpg_pending_linkage))
list_add(&page->cpg_pending_linkage,
&club->cob_pending_list);
spin_unlock(&lli->lli_lock);
}
static struct dentry *
ll_iget_for_nfs(struct super_block *sb, struct lu_fid *fid, struct lu_fid *parent)
{
struct inode *inode;
struct dentry *result;
CDEBUG(D_INFO, "Get dentry for fid: "DFID"\n", PFID(fid));
if (!fid_is_sane(fid))
return ERR_PTR(-ESTALE);
inode = search_inode_for_lustre(sb, fid);
if (IS_ERR(inode))
return ERR_CAST(inode);
if (is_bad_inode(inode)) {
/* we didn't find the right inode.. */
iput(inode);
return ERR_PTR(-ESTALE);
}
/**
* It is an anonymous dentry without OST objects created yet.
* We have to find the parent to tell MDS how to init lov objects.
*/
if (S_ISREG(inode->i_mode) && !ll_i2info(inode)->lli_has_smd &&
parent != NULL) {
struct ll_inode_info *lli = ll_i2info(inode);
spin_lock(&lli->lli_lock);
lli->lli_pfid = *parent;
spin_unlock(&lli->lli_lock);
}
result = d_obtain_alias(inode);
if (IS_ERR(result)) {
iput(inode);
return result;
}
return result;
}
/** Queues DONE_WRITING if
* - done writing is allowed;
* - inode has no no dirty pages; */
void ll_queue_done_writing(struct inode *inode, unsigned long flags)
{
struct ll_inode_info *lli = ll_i2info(inode);
struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob);
ENTRY;
spin_lock(&lli->lli_lock);
lli->lli_flags |= flags;
if ((lli->lli_flags & LLIF_DONE_WRITING) &&
cfs_list_empty(&club->cob_pending_list)) {
struct ll_close_queue *lcq = ll_i2sbi(inode)->ll_lcq;
if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
CWARN("ino %lu/%u(flags %u) som valid it just after "
"recovery\n",
inode->i_ino, inode->i_generation,
lli->lli_flags);
/* DONE_WRITING is allowed and inode has no dirty page. */
spin_lock(&lcq->lcq_lock);
LASSERT(cfs_list_empty(&lli->lli_close_list));
CDEBUG(D_INODE, "adding inode %lu/%u to close list\n",
inode->i_ino, inode->i_generation);
cfs_list_add_tail(&lli->lli_close_list, &lcq->lcq_head);
/* Avoid a concurrent insertion into the close thread queue:
* an inode is already in the close thread, open(), write(),
* close() happen, epoch is closed as the inode is marked as
* LLIF_EPOCH_PENDING. When pages are written inode should not
* be inserted into the queue again, clear this flag to avoid
* it. */
lli->lli_flags &= ~LLIF_DONE_WRITING;
cfs_waitq_signal(&lcq->lcq_waitq);
spin_unlock(&lcq->lcq_lock);
}
spin_unlock(&lli->lli_lock);
EXIT;
}
/** records that a write has completed */
void vvp_write_complete(struct ccc_object *club, struct ccc_page *page)
{
struct ll_inode_info *lli = ll_i2info(club->cob_inode);
int rc = 0;
spin_lock(&lli->lli_lock);
if (page != NULL && !list_empty(&page->cpg_pending_linkage)) {
list_del_init(&page->cpg_pending_linkage);
rc = 1;
}
spin_unlock(&lli->lli_lock);
if (rc)
ll_queue_done_writing(club->cob_inode, 0);
}
/**
* Cliens updates SOM attributes on MDS (including llog cookies):
* obd_getattr with no lock and md_setattr.
*/
int ll_som_update(struct inode *inode, struct md_op_data *op_data)
{
struct ll_inode_info *lli = ll_i2info(inode);
struct ptlrpc_request *request = NULL;
__u32 old_flags;
struct obdo *oa;
int rc;
ENTRY;
LASSERT(op_data != NULL);
if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
CERROR("ino %lu/%u(flags %u) som valid it just after "
"recovery\n", inode->i_ino, inode->i_generation,
lli->lli_flags);
OBDO_ALLOC(oa);
if (!oa) {
CERROR("can't allocate memory for Size-on-MDS update.\n");
RETURN(-ENOMEM);
}
old_flags = op_data->op_flags;
op_data->op_flags = MF_SOM_CHANGE;
/* If inode is already in another epoch, skip getattr from OSTs. */
if (lli->lli_ioepoch == op_data->op_ioepoch) {
rc = ll_inode_getattr(inode, oa, op_data->op_ioepoch,
old_flags & MF_GETATTR_LOCK);
if (rc) {
oa->o_valid = 0;
if (rc != -ENOENT)
CERROR("inode_getattr failed (%d): unable to "
"send a Size-on-MDS attribute update "
"for inode %lu/%u\n", rc, inode->i_ino,
inode->i_generation);
} else {
CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n",
PFID(&lli->lli_fid));
}
/* Install attributes into op_data. */
md_from_obdo(op_data, oa, oa->o_valid);
}
rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data,
NULL, 0, NULL, 0, &request, NULL);
ptlrpc_req_finished(request);
OBDO_FREE(oa);
RETURN(rc);
}
/** Pack SOM attributes info @opdata for CLOSE, DONE_WRITING rpc. */
void ll_done_writing_attr(struct inode *inode, struct md_op_data *op_data)
{
struct ll_inode_info *lli = ll_i2info(inode);
op_data->op_flags |= MF_SOM_CHANGE;
/* Check if Size-on-MDS attributes are valid. */
if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
CERROR("ino %lu/%u(flags %u) som valid it just after "
"recovery\n", inode->i_ino, inode->i_generation,
lli->lli_flags);
if (!cl_local_size(inode)) {
/* Send Size-on-MDS Attributes if valid. */
op_data->op_attr.ia_valid |= ATTR_MTIME_SET | ATTR_CTIME_SET |
ATTR_ATIME_SET | ATTR_SIZE | ATTR_BLOCKS;
}
}
static int vvp_conf_set(const struct lu_env *env, struct cl_object *obj,
const struct cl_object_conf *conf)
{
struct ll_inode_info *lli = ll_i2info(conf->coc_inode);
if (conf->coc_opc == OBJECT_CONF_INVALIDATE) {
CDEBUG(D_VFSTRACE, DFID ": losing layout lock\n",
PFID(&lli->lli_fid));
ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
/* Clean up page mmap for this inode.
* The reason for us to do this is that if the page has
* already been installed into memory space, the process
* can access it without interacting with lustre, so this
* page may be stale due to layout change, and the process
* will never be notified.
* This operation is expensive but mmap processes have to pay
* a price themselves.
*/
unmap_mapping_range(conf->coc_inode->i_mapping,
0, OBD_OBJECT_EOF, 0);
return 0;
}
if (conf->coc_opc != OBJECT_CONF_SET)
return 0;
if (conf->u.coc_md && conf->u.coc_md->lsm) {
CDEBUG(D_VFSTRACE, DFID ": layout version change: %u -> %u\n",
PFID(&lli->lli_fid), lli->lli_layout_gen,
conf->u.coc_md->lsm->lsm_layout_gen);
lli->lli_has_smd = lsm_has_objects(conf->u.coc_md->lsm);
ll_layout_version_set(lli, conf->u.coc_md->lsm->lsm_layout_gen);
} else {
CDEBUG(D_VFSTRACE, DFID ": layout nuked: %u.\n",
PFID(&lli->lli_fid), lli->lli_layout_gen);
lli->lli_has_smd = false;
ll_layout_version_set(lli, LL_LAYOUT_GEN_EMPTY);
}
return 0;
}
static int vvp_io_read_lock(const struct lu_env *env,
const struct cl_io_slice *ios)
{
struct cl_io *io = ios->cis_io;
struct ll_inode_info *lli = ll_i2info(ccc_object_inode(io->ci_obj));
int result;
ENTRY;
/* XXX: Layer violation, we shouldn't see lsm at llite level. */
if (lli->lli_has_smd) /* lsm-less file doesn't need to lock */
result = vvp_io_rw_lock(env, io, CLM_READ,
io->u.ci_rd.rd.crw_pos,
io->u.ci_rd.rd.crw_pos +
io->u.ci_rd.rd.crw_count - 1);
else
result = 0;
RETURN(result);
}
/* find any ldlm lock of the inode in mdc and lov
* return 0 not find
* 1 find one
* < 0 error */
static int find_cbdata(struct inode *inode)
{
struct ll_inode_info *lli = ll_i2info(inode);
struct ll_sb_info *sbi = ll_i2sbi(inode);
int rc = 0;
ENTRY;
LASSERT(inode);
rc = md_find_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
return_if_equal, NULL);
if (rc != 0)
RETURN(rc);
if (lli->lli_smd)
rc = obd_find_cbdata(sbi->ll_dt_exp, lli->lli_smd,
return_if_equal, NULL);
RETURN(rc);
}
/**
* Implements Linux VM address_space::invalidatepage() method. This method is
* called when the page is truncate from a file, either as a result of
* explicit truncate, or when inode is removed from memory (as a result of
* final iput(), umount, or memory pressure induced icache shrinking).
*
* [0, offset] bytes of the page remain valid (this is for a case of not-page
* aligned truncate). Lustre leaves partially truncated page in the cache,
* relying on struct inode::i_size to limit further accesses.
*/
static void ll_invalidatepage(struct page *vmpage,
#ifdef HAVE_INVALIDATE_RANGE
unsigned int offset, unsigned int length
#else
unsigned long offset
#endif
)
{
struct inode *inode;
struct lu_env *env;
struct cl_page *page;
struct cl_object *obj;
int refcheck;
LASSERT(PageLocked(vmpage));
LASSERT(!PageWriteback(vmpage));
/*
* It is safe to not check anything in invalidatepage/releasepage
* below because they are run with page locked and all our io is
* happening with locked page too
*/
#ifdef HAVE_INVALIDATE_RANGE
if (offset == 0 && length == PAGE_CACHE_SIZE) {
#else
if (offset == 0) {
#endif
env = cl_env_get(&refcheck);
if (!IS_ERR(env)) {
inode = vmpage->mapping->host;
obj = ll_i2info(inode)->lli_clob;
if (obj != NULL) {
page = cl_vmpage_page(vmpage, obj);
if (page != NULL) {
cl_page_delete(env, page);
cl_page_put(env, page);
}
} else
LASSERT(vmpage->private == 0);
cl_env_put(env, &refcheck);
}
}
static int vvp_object_print(const struct lu_env *env, void *cookie,
lu_printer_t p, const struct lu_object *o)
{
struct vvp_object *obj = lu2vvp(o);
struct inode *inode = obj->vob_inode;
struct ll_inode_info *lli;
(*p)(env, cookie, "(%d %d) inode: %p ",
atomic_read(&obj->vob_transient_pages),
atomic_read(&obj->vob_mmap_cnt),
inode);
if (inode) {
lli = ll_i2info(inode);
(*p)(env, cookie, "%lu/%u %o %u %d %p "DFID,
inode->i_ino, inode->i_generation, inode->i_mode,
inode->i_nlink, atomic_read(&inode->i_count),
lli->lli_clob, PFID(&lli->lli_fid));
}
return 0;
}
/* check again */
if (save != lli->lli_rmtperm_time) {
rc = do_check_remote_perm(lli, mask);
if (!rc || (rc != -ENOENT && i)) {
mutex_unlock(&lli->lli_rmtperm_mutex);
break;
}
}
if (i++ > 5) {
CERROR("check remote perm falls in dead loop!\n");
LBUG();
}
oc = ll_mdscapa_get(inode);
rc = md_get_remote_perm(sbi->ll_md_exp, ll_inode2fid(inode), oc,
ll_i2suppgid(inode), &req);
capa_put(oc);
if (rc) {
mutex_unlock(&lli->lli_rmtperm_mutex);
break;
}
perm = req_capsule_server_swab_get(&req->rq_pill, &RMF_ACL,
lustre_swab_mdt_remote_perm);
if (unlikely(perm == NULL)) {
mutex_unlock(&lli->lli_rmtperm_mutex);
rc = -EPROTO;
break;
}
rc = ll_update_remote_perm(inode, perm);
mutex_unlock(&lli->lli_rmtperm_mutex);
if (rc == -ENOMEM)
break;
ptlrpc_req_finished(req);
req = NULL;
} while (1);
ptlrpc_req_finished(req);
return rc;
}
#if 0 /* NB: remote perms can't be freed in ll_mdc_blocking_ast of UPDATE lock,
* because it will fail sanity test 48.
*/
void ll_free_remote_perms(struct inode *inode)
{
struct ll_inode_info *lli = ll_i2info(inode);
struct hlist_head *hash = lli->lli_remote_perms;
struct ll_remote_perm *lrp;
struct hlist_node *node, *next;
int i;
LASSERT(hash);
spin_lock(&lli->lli_lock);
for (i = 0; i < REMOTE_PERM_HASHSIZE; i++) {
hlist_for_each_entry_safe(lrp, node, next, hash + i,
lrp_list)
free_ll_remote_perm(lrp);
}
spin_unlock(&lli->lli_lock);
}
/**
* Implements Linux VM address_space::invalidatepage() method. This method is
* called when the page is truncate from a file, either as a result of
* explicit truncate, or when inode is removed from memory (as a result of
* final iput(), umount, or memory pressure induced icache shrinking).
*
* [0, offset] bytes of the page remain valid (this is for a case of not-page
* aligned truncate). Lustre leaves partially truncated page in the cache,
* relying on struct inode::i_size to limit further accesses.
*/
static int cl_invalidatepage(struct page *vmpage, unsigned long offset)
{
struct inode *inode;
struct lu_env *env;
struct cl_page *page;
struct cl_object *obj;
int result;
int refcheck;
LASSERT(PageLocked(vmpage));
LASSERT(!PageWriteback(vmpage));
/*
* It is safe to not check anything in invalidatepage/releasepage
* below because they are run with page locked and all our io is
* happening with locked page too
*/
result = 0;
if (offset == 0) {
env = cl_env_get(&refcheck);
if (!IS_ERR(env)) {
inode = vmpage->mapping->host;
obj = ll_i2info(inode)->lli_clob;
if (obj != NULL) {
page = cl_vmpage_page(vmpage, obj);
if (page != NULL) {
lu_ref_add(&page->cp_reference,
"delete", vmpage);
cl_page_delete(env, page);
result = 1;
lu_ref_del(&page->cp_reference,
"delete", vmpage);
cl_page_put(env, page);
}
} else
LASSERT(vmpage->private == 0);
cl_env_put(env, &refcheck);
}
static int vvp_attr_get(const struct lu_env *env, struct cl_object *obj,
struct cl_attr *attr)
{
struct inode *inode = vvp_object_inode(obj);
/*
* lov overwrites most of these fields in
* lov_attr_get()->...lov_merge_lvb_kms(), except when inode
* attributes are newer.
*/
attr->cat_size = i_size_read(inode);
attr->cat_mtime = inode->i_mtime.tv_sec;
attr->cat_atime = inode->i_atime.tv_sec;
attr->cat_ctime = inode->i_ctime.tv_sec;
attr->cat_blocks = inode->i_blocks;
attr->cat_uid = from_kuid(&init_user_ns, inode->i_uid);
attr->cat_gid = from_kgid(&init_user_ns, inode->i_gid);
attr->cat_projid = ll_i2info(inode)->lli_projid;
/* KMS is not known by this layer */
return 0; /* layers below have to fill in the rest */
}
void ll_release_page(struct inode *inode, struct page *page, bool remove)
{
kunmap(page);
/*
* Always remove the page for striped dir, because the page is
* built from temporarily in LMV layer
*/
if (inode && S_ISDIR(inode->i_mode) &&
ll_i2info(inode)->lli_lsm_md) {
__free_page(page);
return;
}
if (remove) {
lock_page(page);
if (likely(page->mapping))
truncate_complete_page(page->mapping, page);
unlock_page(page);
}
put_page(page);
}
static int ll_readlink_internal(struct inode *inode,
struct ptlrpc_request **request, char **symname)
{
struct ll_inode_info *lli = ll_i2info(inode);
struct ll_sb_info *sbi = ll_i2sbi(inode);
int rc, symlen = i_size_read(inode) + 1;
struct mdt_body *body;
struct md_op_data *op_data;
*request = NULL;
if (lli->lli_symlink_name) {
int print_limit = min_t(int, PAGE_SIZE - 128, symlen);
*symname = lli->lli_symlink_name;
/* If the total CDEBUG() size is larger than a page, it
* will print a warning to the console, avoid this by
* printing just the last part of the symlink. */
CDEBUG(D_INODE, "using cached symlink %s%.*s, len = %d\n",
print_limit < symlen ? "..." : "", print_limit,
(*symname) + symlen - print_limit, symlen);
return 0;
}
static int vvp_attr_update(const struct lu_env *env, struct cl_object *obj,
const struct cl_attr *attr, unsigned valid)
{
struct inode *inode = vvp_object_inode(obj);
if (valid & CAT_UID)
inode->i_uid = make_kuid(&init_user_ns, attr->cat_uid);
if (valid & CAT_GID)
inode->i_gid = make_kgid(&init_user_ns, attr->cat_gid);
if (valid & CAT_ATIME)
inode->i_atime.tv_sec = attr->cat_atime;
if (valid & CAT_MTIME)
inode->i_mtime.tv_sec = attr->cat_mtime;
if (valid & CAT_CTIME)
inode->i_ctime.tv_sec = attr->cat_ctime;
if (0 && valid & CAT_SIZE)
i_size_write(inode, attr->cat_size);
if (valid & CAT_PROJID)
ll_i2info(inode)->lli_projid = attr->cat_projid;
/* not currently necessary */
if (0 && valid & (CAT_UID|CAT_GID|CAT_SIZE|CAT_PROJID))
mark_inode_dirty(inode);
return 0;
}
/** Closes ioepoch and packs Size-on-MDS attribute if needed into @op_data. */
void ll_ioepoch_close(struct inode *inode, struct md_op_data *op_data,
struct obd_client_handle **och, unsigned long flags)
{
struct ll_inode_info *lli = ll_i2info(inode);
struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob);
spin_lock(&lli->lli_lock);
if (!(list_empty(&club->cob_pending_list))) {
if (!(lli->lli_flags & LLIF_EPOCH_PENDING)) {
LASSERT(*och != NULL);
LASSERT(lli->lli_pending_och == NULL);
/* Inode is dirty and there is no pending write done
* request yet, DONE_WRITE is to be sent later. */
lli->lli_flags |= LLIF_EPOCH_PENDING;
lli->lli_pending_och = *och;
spin_unlock(&lli->lli_lock);
inode = igrab(inode);
LASSERT(inode);
GOTO(out, 0);
}
if (flags & LLIF_DONE_WRITING) {
/* Some pages are still dirty, it is early to send
* DONE_WRITE. Wait untill all pages will be flushed
* and try DONE_WRITE again later. */
LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
lli->lli_flags |= LLIF_DONE_WRITING;
spin_unlock(&lli->lli_lock);
inode = igrab(inode);
LASSERT(inode);
GOTO(out, 0);
}
}
CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID"\n",
ll_i2info(inode)->lli_ioepoch, PFID(&lli->lli_fid));
op_data->op_flags |= MF_EPOCH_CLOSE;
if (flags & LLIF_DONE_WRITING) {
LASSERT(lli->lli_flags & LLIF_SOM_DIRTY);
LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
*och = lli->lli_pending_och;
lli->lli_pending_och = NULL;
lli->lli_flags &= ~LLIF_EPOCH_PENDING;
} else {
/* Pack Size-on-MDS inode attributes only if they has changed */
if (!(lli->lli_flags & LLIF_SOM_DIRTY)) {
spin_unlock(&lli->lli_lock);
GOTO(out, 0);
}
/* There is a pending DONE_WRITE -- close epoch with no
* attribute change. */
if (lli->lli_flags & LLIF_EPOCH_PENDING) {
spin_unlock(&lli->lli_lock);
GOTO(out, 0);
}
}
LASSERT(list_empty(&club->cob_pending_list));
lli->lli_flags &= ~LLIF_SOM_DIRTY;
spin_unlock(&lli->lli_lock);
ll_done_writing_attr(inode, op_data);
out:
return;
}
ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size)
{
struct inode *inode = dentry->d_inode;
int rc = 0, rc2 = 0;
struct lov_mds_md *lmm = NULL;
struct ptlrpc_request *request = NULL;
int lmmsize;
LASSERT(inode);
CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
inode->i_ino, inode->i_generation, inode);
ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LISTXATTR, 1);
rc = ll_getxattr_common(inode, NULL, buffer, size, OBD_MD_FLXATTRLS);
if (rc < 0)
GOTO(out, rc);
if (buffer != NULL) {
struct ll_sb_info *sbi = ll_i2sbi(inode);
char *xattr_name = buffer;
int xlen, rem = rc;
while (rem > 0) {
xlen = strnlen(xattr_name, rem - 1) + 1;
rem -= xlen;
if (xattr_type_filter(sbi,
get_xattr_type(xattr_name)) == 0) {
/* skip OK xattr type
* leave it in buffer
*/
xattr_name += xlen;
continue;
}
/* move up remaining xattrs in buffer
* removing the xattr that is not OK
*/
memmove(xattr_name, xattr_name + xlen, rem);
rc -= xlen;
}
}
if (S_ISREG(inode->i_mode)) {
if (!ll_i2info(inode)->lli_has_smd)
rc2 = -1;
} else if (S_ISDIR(inode->i_mode)) {
rc2 = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
}
if (rc2 < 0) {
GOTO(out, rc2 = 0);
} else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)) {
const int prefix_len = sizeof(XATTR_LUSTRE_PREFIX) - 1;
const size_t name_len = sizeof("lov") - 1;
const size_t total_len = prefix_len + name_len + 1;
if (((rc + total_len) > size) && (buffer != NULL)) {
ptlrpc_req_finished(request);
return -ERANGE;
}
if (buffer != NULL) {
buffer += rc;
memcpy(buffer, XATTR_LUSTRE_PREFIX, prefix_len);
memcpy(buffer + prefix_len, "lov", name_len);
buffer[prefix_len + name_len] = '\0';
}
rc2 = total_len;
}
out:
ptlrpc_req_finished(request);
rc = rc + rc2;
return rc;
}
