* This file is part of Lustre, http://www.lustre.org/
* Lustre is a trademark of Sun Microsystems, Inc.
*
- * cl code shared between vvp and liblustre (and other Lustre clients in the
- * future).
- *
* Author: Nikita Danilov <nikita.danilov@sun.com>
*/
#include <cl_object.h>
#include "llite_internal.h"
-
-static const struct cl_req_operations ccc_req_ops;
-
-/*
- * ccc_ prefix stands for "Common Client Code".
- */
-
-static struct kmem_cache *ccc_thread_kmem;
-static struct kmem_cache *ccc_session_kmem;
-static struct kmem_cache *ccc_req_kmem;
-
-static struct lu_kmem_descr ccc_caches[] = {
- {
- .ckd_cache = &ccc_thread_kmem,
- .ckd_name = "ccc_thread_kmem",
- .ckd_size = sizeof (struct ccc_thread_info),
- },
- {
- .ckd_cache = &ccc_session_kmem,
- .ckd_name = "ccc_session_kmem",
- .ckd_size = sizeof (struct ccc_session)
- },
- {
- .ckd_cache = &ccc_req_kmem,
- .ckd_name = "ccc_req_kmem",
- .ckd_size = sizeof (struct ccc_req)
- },
- {
- .ckd_cache = NULL
- }
-};
-
-/*****************************************************************************
- *
- * Vvp device and device type functions.
- *
- */
-
-void *ccc_key_init(const struct lu_context *ctx, struct lu_context_key *key)
-{
- struct ccc_thread_info *info;
-
- OBD_SLAB_ALLOC_PTR_GFP(info, ccc_thread_kmem, GFP_NOFS);
- if (info == NULL)
- info = ERR_PTR(-ENOMEM);
- return info;
-}
-
-void ccc_key_fini(const struct lu_context *ctx,
- struct lu_context_key *key, void *data)
-{
- struct ccc_thread_info *info = data;
- OBD_SLAB_FREE_PTR(info, ccc_thread_kmem);
-}
-
-void *ccc_session_key_init(const struct lu_context *ctx,
- struct lu_context_key *key)
-{
- struct ccc_session *session;
-
- OBD_SLAB_ALLOC_PTR_GFP(session, ccc_session_kmem, GFP_NOFS);
- if (session == NULL)
- session = ERR_PTR(-ENOMEM);
- return session;
-}
-
-void ccc_session_key_fini(const struct lu_context *ctx,
- struct lu_context_key *key, void *data)
-{
- struct ccc_session *session = data;
- OBD_SLAB_FREE_PTR(session, ccc_session_kmem);
-}
-
-struct lu_context_key ccc_key = {
- .lct_tags = LCT_CL_THREAD,
- .lct_init = ccc_key_init,
- .lct_fini = ccc_key_fini
-};
-
-struct lu_context_key ccc_session_key = {
- .lct_tags = LCT_SESSION,
- .lct_init = ccc_session_key_init,
- .lct_fini = ccc_session_key_fini
-};
-
-int ccc_req_init(const struct lu_env *env, struct cl_device *dev,
- struct cl_req *req)
-{
- struct ccc_req *vrq;
- int result;
-
- OBD_SLAB_ALLOC_PTR_GFP(vrq, ccc_req_kmem, GFP_NOFS);
- if (vrq != NULL) {
- cl_req_slice_add(req, &vrq->crq_cl, dev, &ccc_req_ops);
- result = 0;
- } else
- result = -ENOMEM;
- return result;
-}
+#include "vvp_internal.h"
/**
- * An `emergency' environment used by ccc_inode_fini() when cl_env_get()
- * fails. Access to this environment is serialized by ccc_inode_fini_guard
+ * An `emergency' environment used by cl_inode_fini() when cl_env_get()
+ * fails. Access to this environment is serialized by cl_inode_fini_guard
* mutex.
*/
-static struct lu_env *ccc_inode_fini_env = NULL;
+struct lu_env *cl_inode_fini_env;
+int cl_inode_fini_refcheck;
/**
* A mutex serializing calls to slp_inode_fini() under extreme memory
* pressure, when environments cannot be allocated.
*/
-static DEFINE_MUTEX(ccc_inode_fini_guard);
-static int dummy_refcheck;
-
-int ccc_global_init(struct lu_device_type *device_type)
-{
- int result;
-
- result = lu_kmem_init(ccc_caches);
- if (result)
- return result;
-
- result = lu_device_type_init(device_type);
- if (result)
- goto out_kmem;
-
- ccc_inode_fini_env = cl_env_alloc(&dummy_refcheck,
- LCT_REMEMBER|LCT_NOREF);
- if (IS_ERR(ccc_inode_fini_env)) {
- result = PTR_ERR(ccc_inode_fini_env);
- goto out_device;
- }
-
- ccc_inode_fini_env->le_ctx.lc_cookie = 0x4;
- return 0;
-out_device:
- lu_device_type_fini(device_type);
-out_kmem:
- lu_kmem_fini(ccc_caches);
- return result;
-}
-
-void ccc_global_fini(struct lu_device_type *device_type)
-{
- if (ccc_inode_fini_env != NULL) {
- cl_env_put(ccc_inode_fini_env, &dummy_refcheck);
- ccc_inode_fini_env = NULL;
- }
- lu_device_type_fini(device_type);
- lu_kmem_fini(ccc_caches);
-}
-
-static void vvp_object_size_lock(struct cl_object *obj)
-{
- struct inode *inode = vvp_object_inode(obj);
-
- ll_inode_size_lock(inode);
- cl_object_attr_lock(obj);
-}
-
-static void vvp_object_size_unlock(struct cl_object *obj)
-{
- struct inode *inode = vvp_object_inode(obj);
-
- cl_object_attr_unlock(obj);
- ll_inode_size_unlock(inode);
-}
-
-/*****************************************************************************
- *
- * io operations.
- *
- */
-
-void ccc_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
-{
- struct cl_io *io = ios->cis_io;
-
- CLOBINVRNT(env, io->ci_obj, vvp_object_invariant(io->ci_obj));
-}
-
-int ccc_io_one_lock_index(const struct lu_env *env, struct cl_io *io,
- __u32 enqflags, enum cl_lock_mode mode,
- pgoff_t start, pgoff_t end)
-{
- struct ccc_io *cio = ccc_env_io(env);
- struct cl_lock_descr *descr = &cio->cui_link.cill_descr;
- struct cl_object *obj = io->ci_obj;
-
- CLOBINVRNT(env, obj, vvp_object_invariant(obj));
- ENTRY;
-
- CDEBUG(D_VFSTRACE, "lock: %d [%lu, %lu]\n", mode, start, end);
-
- memset(&cio->cui_link, 0, sizeof cio->cui_link);
-
- if (cio->cui_fd && (cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
- descr->cld_mode = CLM_GROUP;
- descr->cld_gid = cio->cui_fd->fd_grouplock.cg_gid;
- } else {
- descr->cld_mode = mode;
- }
- descr->cld_obj = obj;
- descr->cld_start = start;
- descr->cld_end = end;
- descr->cld_enq_flags = enqflags;
-
- cl_io_lock_add(env, io, &cio->cui_link);
- RETURN(0);
-}
-
-void ccc_io_update_iov(const struct lu_env *env,
- struct ccc_io *cio, struct cl_io *io)
-{
- int i;
- size_t size = io->u.ci_rw.crw_count;
-
- cio->cui_iov_olen = 0;
- if (!cl_is_normalio(env, io) || cio->cui_tot_nrsegs == 0)
- return;
-
- for (i = 0; i < cio->cui_tot_nrsegs; i++) {
- struct iovec *iv = &cio->cui_iov[i];
-
- if (iv->iov_len < size)
- size -= iv->iov_len;
- else {
- if (iv->iov_len > size) {
- cio->cui_iov_olen = iv->iov_len;
- iv->iov_len = size;
- }
- break;
- }
- }
-
- cio->cui_nrsegs = i + 1;
- LASSERTF(cio->cui_tot_nrsegs >= cio->cui_nrsegs,
- "tot_nrsegs: %lu, nrsegs: %lu\n",
- cio->cui_tot_nrsegs, cio->cui_nrsegs);
-}
-
-int ccc_io_one_lock(const struct lu_env *env, struct cl_io *io,
- __u32 enqflags, enum cl_lock_mode mode,
- loff_t start, loff_t end)
-{
- struct cl_object *obj = io->ci_obj;
- return ccc_io_one_lock_index(env, io, enqflags, mode,
- cl_index(obj, start), cl_index(obj, end));
-}
-
-void ccc_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
-{
- CLOBINVRNT(env, ios->cis_io->ci_obj,
- vvp_object_invariant(ios->cis_io->ci_obj));
-}
-
-void ccc_io_advance(const struct lu_env *env,
- const struct cl_io_slice *ios,
- size_t nob)
-{
- struct ccc_io *cio = cl2ccc_io(env, ios);
- struct cl_io *io = ios->cis_io;
- struct cl_object *obj = ios->cis_io->ci_obj;
-
- CLOBINVRNT(env, obj, vvp_object_invariant(obj));
-
- if (!cl_is_normalio(env, io))
- return;
-
- LASSERT(cio->cui_tot_nrsegs >= cio->cui_nrsegs);
- LASSERT(cio->cui_tot_count >= nob);
-
- cio->cui_iov += cio->cui_nrsegs;
- cio->cui_tot_nrsegs -= cio->cui_nrsegs;
- cio->cui_tot_count -= nob;
-
- /* update the iov */
- if (cio->cui_iov_olen > 0) {
- struct iovec *iv;
-
- cio->cui_iov--;
- cio->cui_tot_nrsegs++;
- iv = &cio->cui_iov[0];
- if (io->ci_continue) {
- iv->iov_base += iv->iov_len;
- LASSERT(cio->cui_iov_olen > iv->iov_len);
- iv->iov_len = cio->cui_iov_olen - iv->iov_len;
- } else {
- /* restore the iov_len, in case of restart io. */
- iv->iov_len = cio->cui_iov_olen;
- }
- cio->cui_iov_olen = 0;
- }
-}
-
-/**
- * Helper function that if necessary adjusts file size (inode->i_size), when
- * position at the offset \a pos is accessed. File size can be arbitrary stale
- * on a Lustre client, but client at least knows KMS. If accessed area is
- * inside [0, KMS], set file size to KMS, otherwise glimpse file size.
- *
- * Locking: cl_isize_lock is used to serialize changes to inode size and to
- * protect consistency between inode size and cl_object
- * attributes. cl_object_size_lock() protects consistency between cl_attr's of
- * top-object and sub-objects.
- */
-int ccc_prep_size(const struct lu_env *env, struct cl_object *obj,
- struct cl_io *io, loff_t start, size_t count, int *exceed)
-{
- struct cl_attr *attr = ccc_env_thread_attr(env);
- struct inode *inode = vvp_object_inode(obj);
- loff_t pos = start + count - 1;
- loff_t kms;
- int result;
-
- /*
- * Consistency guarantees: following possibilities exist for the
- * relation between region being accessed and real file size at this
- * moment:
- *
- * (A): the region is completely inside of the file;
- *
- * (B-x): x bytes of region are inside of the file, the rest is
- * outside;
- *
- * (C): the region is completely outside of the file.
- *
- * This classification is stable under DLM lock already acquired by
- * the caller, because to change the class, other client has to take
- * DLM lock conflicting with our lock. Also, any updates to ->i_size
- * by other threads on this client are serialized by
- * ll_inode_size_lock(). This guarantees that short reads are handled
- * correctly in the face of concurrent writes and truncates.
- */
- vvp_object_size_lock(obj);
- result = cl_object_attr_get(env, obj, attr);
- if (result == 0) {
- kms = attr->cat_kms;
- if (pos > kms) {
- /*
- * A glimpse is necessary to determine whether we
- * return a short read (B) or some zeroes at the end
- * of the buffer (C)
- */
- vvp_object_size_unlock(obj);
- result = cl_glimpse_lock(env, io, inode, obj, 0);
- if (result == 0 && exceed != NULL) {
- /* If objective page index exceed end-of-file
- * page index, return directly. Do not expect
- * kernel will check such case correctly.
- * linux-2.6.18-128.1.1 miss to do that.
- * --bug 17336 */
- loff_t size = i_size_read(inode);
- unsigned long cur_index = start >>
- PAGE_CACHE_SHIFT;
-
- if ((size == 0 && cur_index != 0) ||
- (((size - 1) >> PAGE_CACHE_SHIFT) <
- cur_index))
- *exceed = 1;
- }
- return result;
- } else {
- /*
- * region is within kms and, hence, within real file
- * size (A). We need to increase i_size to cover the
- * read region so that generic_file_read() will do its
- * job, but that doesn't mean the kms size is
- * _correct_, it is only the _minimum_ size. If
- * someone does a stat they will get the correct size
- * which will always be >= the kms value here.
- * b=11081
- */
- if (i_size_read(inode) < kms) {
- i_size_write(inode, kms);
- CDEBUG(D_VFSTRACE,
- DFID" updating i_size "LPU64"\n",
- PFID(lu_object_fid(&obj->co_lu)),
- (__u64)i_size_read(inode));
-
- }
- }
- }
+static DEFINE_MUTEX(cl_inode_fini_guard);
- vvp_object_size_unlock(obj);
-
- return result;
-}
-
-/*****************************************************************************
- *
- * Transfer operations.
- *
- */
-
-void ccc_req_completion(const struct lu_env *env,
- const struct cl_req_slice *slice, int ioret)
-{
- struct ccc_req *vrq;
-
- if (ioret > 0)
- cl_stats_tally(slice->crs_dev, slice->crs_req->crq_type, ioret);
-
- vrq = cl2ccc_req(slice);
- OBD_SLAB_FREE_PTR(vrq, ccc_req_kmem);
-}
-
-/**
- * Implementation of struct cl_req_operations::cro_attr_set() for ccc
- * layer. ccc is responsible for
- *
- * - o_[mac]time
- *
- * - o_mode
- *
- * - o_parent_seq
- *
- * - o_[ug]id
- *
- * - o_parent_oid
- *
- * - o_parent_ver
- *
- * - o_ioepoch,
- *
- * and capability.
- */
-void ccc_req_attr_set(const struct lu_env *env,
- const struct cl_req_slice *slice,
- const struct cl_object *obj,
- struct cl_req_attr *attr, obd_valid flags)
-{
- struct inode *inode;
- struct obdo *oa;
- obd_flag valid_flags;
-
- oa = attr->cra_oa;
- inode = vvp_object_inode(obj);
- valid_flags = OBD_MD_FLTYPE;
-
- if ((flags & OBD_MD_FLOSSCAPA) != 0) {
- LASSERT(attr->cra_capa == NULL);
- attr->cra_capa = cl_capa_lookup(inode,
- slice->crs_req->crq_type);
- }
-
- 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);
-}
-
-static const struct cl_req_operations ccc_req_ops = {
- .cro_attr_set = ccc_req_attr_set,
- .cro_completion = ccc_req_completion
-};
-
-int cl_setattr_ost(struct inode *inode, const struct iattr *attr,
- struct obd_capa *capa)
+int cl_setattr_ost(struct cl_object *obj, const struct iattr *attr,
+ unsigned int attr_flags)
{
struct lu_env *env;
struct cl_io *io;
if (IS_ERR(env))
RETURN(PTR_ERR(env));
- io = ccc_env_thread_io(env);
- io->ci_obj = ll_i2info(inode)->lli_clob;
+ io = vvp_env_thread_io(env);
+ io->ci_obj = obj;
- io->u.ci_setattr.sa_attr.lvb_atime = LTIME_S(attr->ia_atime);
- io->u.ci_setattr.sa_attr.lvb_mtime = LTIME_S(attr->ia_mtime);
- io->u.ci_setattr.sa_attr.lvb_ctime = LTIME_S(attr->ia_ctime);
- io->u.ci_setattr.sa_attr.lvb_size = attr->ia_size;
- io->u.ci_setattr.sa_valid = attr->ia_valid;
- io->u.ci_setattr.sa_capa = capa;
+ io->u.ci_setattr.sa_attr.lvb_atime = LTIME_S(attr->ia_atime);
+ io->u.ci_setattr.sa_attr.lvb_mtime = LTIME_S(attr->ia_mtime);
+ io->u.ci_setattr.sa_attr.lvb_ctime = LTIME_S(attr->ia_ctime);
+ io->u.ci_setattr.sa_attr.lvb_size = attr->ia_size;
+ io->u.ci_setattr.sa_attr_flags = attr_flags;
+ io->u.ci_setattr.sa_valid = attr->ia_valid;
+ io->u.ci_setattr.sa_parent_fid = lu_object_fid(&obj->co_lu);
again:
if (cl_io_init(env, io, CIT_SETATTR, io->ci_obj) == 0) {
- struct ccc_io *cio = ccc_env_io(env);
+ struct vvp_io *vio = vvp_env_io(env);
- if (attr->ia_valid & ATTR_FILE)
- /* populate the file descriptor for ftruncate to honor
- * group lock - see LU-787 */
- cio->cui_fd = LUSTRE_FPRIVATE(attr->ia_file);
+ if (attr->ia_valid & ATTR_FILE)
+ /* populate the file descriptor for ftruncate to honor
+ * group lock - see LU-787 */
+ vio->vui_fd = LUSTRE_FPRIVATE(attr->ia_file);
result = cl_io_loop(env, io);
} else {
RETURN(result);
}
-/*****************************************************************************
- *
- * Type conversions.
- *
- */
-
-struct ccc_io *cl2ccc_io(const struct lu_env *env,
- const struct cl_io_slice *slice)
-{
- struct ccc_io *cio;
-
- cio = container_of(slice, struct ccc_io, cui_cl);
- LASSERT(cio == ccc_env_io(env));
- return cio;
-}
-
-struct ccc_req *cl2ccc_req(const struct cl_req_slice *slice)
-{
- return container_of0(slice, struct ccc_req, crq_cl);
-}
-
/**
* Initialize or update CLIO structures for regular files when new
* meta-data arrives from the server.
* locked by I_NEW bit.
*/
lli->lli_clob = clob;
- lli->lli_has_smd = lsm_has_objects(md->lsm);
lu_object_ref_add(&clob->co_lu, "inode", inode);
} else
result = PTR_ERR(clob);
bkt = lu_site_bkt_from_fid(site, &header->loh_fid);
- init_waitqueue_entry_current(&waiter);
+ init_waitqueue_entry(&waiter, current);
add_wait_queue(&bkt->lsb_marche_funebre, &waiter);
while (1) {
set_current_state(TASK_UNINTERRUPTIBLE);
if (atomic_read(&header->loh_ref) == 1)
break;
- waitq_wait(&waiter, TASK_UNINTERRUPTIBLE);
+ schedule();
}
set_current_state(TASK_RUNNING);
cookie = cl_env_reenter();
env = cl_env_get(&refcheck);
emergency = IS_ERR(env);
- if (emergency) {
- mutex_lock(&ccc_inode_fini_guard);
- LASSERT(ccc_inode_fini_env != NULL);
- cl_env_implant(ccc_inode_fini_env, &refcheck);
- env = ccc_inode_fini_env;
- }
+ if (emergency) {
+ mutex_lock(&cl_inode_fini_guard);
+ LASSERT(cl_inode_fini_env != NULL);
+ cl_env_implant(cl_inode_fini_env, &refcheck);
+ env = cl_inode_fini_env;
+ }
+
/*
* cl_object cache is a slave to inode cache (which, in turn
* is a slave to dentry cache), don't keep cl_object in memory
lu_object_ref_del(&clob->co_lu, "inode", inode);
cl_object_put_last(env, clob);
lli->lli_clob = NULL;
- if (emergency) {
- cl_env_unplant(ccc_inode_fini_env, &refcheck);
- mutex_unlock(&ccc_inode_fini_guard);
- } else
- cl_env_put(env, &refcheck);
+ if (emergency) {
+ cl_env_unplant(cl_inode_fini_env, &refcheck);
+ mutex_unlock(&cl_inode_fini_guard);
+ } else {
+ cl_env_put(env, &refcheck);
+ }
+
cl_env_reexit(cookie);
}
}
/**
- * return IF_* type for given lu_dirent entry.
- * IF_* flag shld be converted to particular OS file type in
- * platform llite module.
- */
-__u16 ll_dirent_type_get(struct lu_dirent *ent)
-{
- __u16 type = 0;
- struct luda_type *lt;
- int len = 0;
-
- if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
- const unsigned align = sizeof(struct luda_type) - 1;
-
- len = le16_to_cpu(ent->lde_namelen);
- len = (len + align) & ~align;
- lt = (void *)ent->lde_name + len;
- type = IFTODT(le16_to_cpu(lt->lt_type));
- }
- return type;
-}
-
-/**
* build inode number from passed @fid */
__u64 cl_fid_build_ino(const struct lu_fid *fid, int api32)
{
gen = (fid_flatten(fid) >> 32);
RETURN(gen);
}
-
-/* lsm is unreliable after hsm implementation as layout can be changed at
- * any time. This is only to support old, non-clio-ized interfaces. It will
- * cause deadlock if clio operations are called with this extra layout refcount
- * because in case the layout changed during the IO, ll_layout_refresh() will
- * have to wait for the refcount to become zero to destroy the older layout.
- *
- * Notice that the lsm returned by this function may not be valid unless called
- * inside layout lock - MDS_INODELOCK_LAYOUT. */
-struct lov_stripe_md *ccc_inode_lsm_get(struct inode *inode)
-{
- return lov_lsm_get(ll_i2info(inode)->lli_clob);
-}
-
-void inline ccc_inode_lsm_put(struct inode *inode, struct lov_stripe_md *lsm)
-{
- lov_lsm_put(ll_i2info(inode)->lli_clob, lsm);
-}