4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * cl code shared between vvp and liblustre (and other Lustre clients in the
39 * Author: Nikita Danilov <nikita.danilov@sun.com>
42 #define DEBUG_SUBSYSTEM S_LLITE
44 #include <libcfs/libcfs.h>
46 #include <linux/sched.h>
48 #include <linux/quotaops.h>
49 #include <linux/highmem.h>
50 #include <linux/pagemap.h>
51 #include <linux/rbtree.h>
54 #include <obd_support.h>
55 #include <lustre_fid.h>
56 #include <lustre_dlm.h>
57 #include <lustre_ver.h>
58 #include <lustre_mdc.h>
59 #include <cl_object.h>
61 #include "llite_internal.h"
63 static const struct cl_req_operations ccc_req_ops;
66 * ccc_ prefix stands for "Common Client Code".
69 static struct kmem_cache *ccc_thread_kmem;
70 static struct kmem_cache *ccc_req_kmem;
72 static struct lu_kmem_descr ccc_caches[] = {
74 .ckd_cache = &ccc_thread_kmem,
75 .ckd_name = "ccc_thread_kmem",
76 .ckd_size = sizeof (struct ccc_thread_info),
79 .ckd_cache = &ccc_req_kmem,
80 .ckd_name = "ccc_req_kmem",
81 .ckd_size = sizeof (struct ccc_req)
88 /*****************************************************************************
90 * Vvp device and device type functions.
94 void *ccc_key_init(const struct lu_context *ctx, struct lu_context_key *key)
96 struct ccc_thread_info *info;
98 OBD_SLAB_ALLOC_PTR_GFP(info, ccc_thread_kmem, GFP_NOFS);
100 info = ERR_PTR(-ENOMEM);
104 void ccc_key_fini(const struct lu_context *ctx,
105 struct lu_context_key *key, void *data)
107 struct ccc_thread_info *info = data;
108 OBD_SLAB_FREE_PTR(info, ccc_thread_kmem);
111 struct lu_context_key ccc_key = {
112 .lct_tags = LCT_CL_THREAD,
113 .lct_init = ccc_key_init,
114 .lct_fini = ccc_key_fini
117 int ccc_req_init(const struct lu_env *env, struct cl_device *dev,
123 OBD_SLAB_ALLOC_PTR_GFP(vrq, ccc_req_kmem, GFP_NOFS);
125 cl_req_slice_add(req, &vrq->crq_cl, dev, &ccc_req_ops);
133 * An `emergency' environment used by ccc_inode_fini() when cl_env_get()
134 * fails. Access to this environment is serialized by ccc_inode_fini_guard
137 static struct lu_env *ccc_inode_fini_env = NULL;
140 * A mutex serializing calls to slp_inode_fini() under extreme memory
141 * pressure, when environments cannot be allocated.
143 static DEFINE_MUTEX(ccc_inode_fini_guard);
144 static int dummy_refcheck;
146 int ccc_global_init(struct lu_device_type *device_type)
150 result = lu_kmem_init(ccc_caches);
154 result = lu_device_type_init(device_type);
158 ccc_inode_fini_env = cl_env_alloc(&dummy_refcheck,
159 LCT_REMEMBER|LCT_NOREF);
160 if (IS_ERR(ccc_inode_fini_env)) {
161 result = PTR_ERR(ccc_inode_fini_env);
165 ccc_inode_fini_env->le_ctx.lc_cookie = 0x4;
168 lu_device_type_fini(device_type);
170 lu_kmem_fini(ccc_caches);
174 void ccc_global_fini(struct lu_device_type *device_type)
176 if (ccc_inode_fini_env != NULL) {
177 cl_env_put(ccc_inode_fini_env, &dummy_refcheck);
178 ccc_inode_fini_env = NULL;
180 lu_device_type_fini(device_type);
181 lu_kmem_fini(ccc_caches);
184 /*****************************************************************************
186 * Transfer operations.
190 void ccc_req_completion(const struct lu_env *env,
191 const struct cl_req_slice *slice, int ioret)
196 cl_stats_tally(slice->crs_dev, slice->crs_req->crq_type, ioret);
198 vrq = cl2ccc_req(slice);
199 OBD_SLAB_FREE_PTR(vrq, ccc_req_kmem);
203 * Implementation of struct cl_req_operations::cro_attr_set() for ccc
204 * layer. ccc is responsible for
222 void ccc_req_attr_set(const struct lu_env *env,
223 const struct cl_req_slice *slice,
224 const struct cl_object *obj,
225 struct cl_req_attr *attr, u64 flags)
232 inode = vvp_object_inode(obj);
233 valid_flags = OBD_MD_FLTYPE;
235 if ((flags & OBD_MD_FLOSSCAPA) != 0) {
236 LASSERT(attr->cra_capa == NULL);
237 attr->cra_capa = cl_capa_lookup(inode,
238 slice->crs_req->crq_type);
241 if (slice->crs_req->crq_type == CRT_WRITE) {
242 if (flags & OBD_MD_FLEPOCH) {
243 oa->o_valid |= OBD_MD_FLEPOCH;
244 oa->o_ioepoch = ll_i2info(inode)->lli_ioepoch;
245 valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME |
246 OBD_MD_FLUID | OBD_MD_FLGID;
249 obdo_from_inode(oa, inode, valid_flags & flags);
250 obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
251 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_INVALID_PFID))
253 memcpy(attr->cra_jobid, ll_i2info(inode)->lli_jobid,
257 static const struct cl_req_operations ccc_req_ops = {
258 .cro_attr_set = ccc_req_attr_set,
259 .cro_completion = ccc_req_completion
262 int cl_setattr_ost(struct inode *inode, const struct iattr *attr,
263 struct obd_capa *capa)
272 env = cl_env_get(&refcheck);
274 RETURN(PTR_ERR(env));
276 io = ccc_env_thread_io(env);
277 io->ci_obj = ll_i2info(inode)->lli_clob;
279 io->u.ci_setattr.sa_attr.lvb_atime = LTIME_S(attr->ia_atime);
280 io->u.ci_setattr.sa_attr.lvb_mtime = LTIME_S(attr->ia_mtime);
281 io->u.ci_setattr.sa_attr.lvb_ctime = LTIME_S(attr->ia_ctime);
282 io->u.ci_setattr.sa_attr.lvb_size = attr->ia_size;
283 io->u.ci_setattr.sa_valid = attr->ia_valid;
284 io->u.ci_setattr.sa_parent_fid = ll_inode2fid(inode);
285 io->u.ci_setattr.sa_capa = capa;
288 if (cl_io_init(env, io, CIT_SETATTR, io->ci_obj) == 0) {
289 struct vvp_io *vio = vvp_env_io(env);
291 if (attr->ia_valid & ATTR_FILE)
292 /* populate the file descriptor for ftruncate to honor
293 * group lock - see LU-787 */
294 vio->vui_fd = LUSTRE_FPRIVATE(attr->ia_file);
296 result = cl_io_loop(env, io);
298 result = io->ci_result;
301 if (unlikely(io->ci_need_restart))
303 /* HSM import case: file is released, cannot be restored
304 * no need to fail except if restore registration failed
306 if (result == -ENODATA && io->ci_restore_needed &&
307 io->ci_result != -ENODATA)
309 cl_env_put(env, &refcheck);
313 /*****************************************************************************
319 struct ccc_req *cl2ccc_req(const struct cl_req_slice *slice)
321 return container_of0(slice, struct ccc_req, crq_cl);
325 * Initialize or update CLIO structures for regular files when new
326 * meta-data arrives from the server.
328 * \param inode regular file inode
329 * \param md new file metadata from MDS
330 * - allocates cl_object if necessary,
331 * - updated layout, if object was already here.
333 int cl_file_inode_init(struct inode *inode, struct lustre_md *md)
336 struct ll_inode_info *lli;
337 struct cl_object *clob;
338 struct lu_site *site;
340 struct cl_object_conf conf = {
349 LASSERT(md->body->mbo_valid & OBD_MD_FLID);
350 LASSERT(S_ISREG(inode->i_mode));
352 env = cl_env_get(&refcheck);
356 site = ll_i2sbi(inode)->ll_site;
357 lli = ll_i2info(inode);
359 LASSERT(fid_is_sane(fid));
361 if (lli->lli_clob == NULL) {
362 /* clob is slave of inode, empty lli_clob means for new inode,
363 * there is no clob in cache with the given fid, so it is
364 * unnecessary to perform lookup-alloc-lookup-insert, just
365 * alloc and insert directly. */
366 LASSERT(inode->i_state & I_NEW);
367 conf.coc_lu.loc_flags = LOC_F_NEW;
368 clob = cl_object_find(env, lu2cl_dev(site->ls_top_dev),
372 * No locking is necessary, as new inode is
373 * locked by I_NEW bit.
375 lli->lli_clob = clob;
376 lli->lli_has_smd = lsm_has_objects(md->lsm);
377 lu_object_ref_add(&clob->co_lu, "inode", inode);
379 result = PTR_ERR(clob);
381 result = cl_conf_set(env, lli->lli_clob, &conf);
384 cl_env_put(env, &refcheck);
387 CERROR("Failure to initialize cl object "DFID": %d\n",
393 * Wait for others drop their references of the object at first, then we drop
394 * the last one, which will lead to the object be destroyed immediately.
395 * Must be called after cl_object_kill() against this object.
397 * The reason we want to do this is: destroying top object will wait for sub
398 * objects being destroyed first, so we can't let bottom layer (e.g. from ASTs)
399 * to initiate top object destroying which may deadlock. See bz22520.
401 static void cl_object_put_last(struct lu_env *env, struct cl_object *obj)
403 struct lu_object_header *header = obj->co_lu.lo_header;
406 if (unlikely(atomic_read(&header->loh_ref) != 1)) {
407 struct lu_site *site = obj->co_lu.lo_dev->ld_site;
408 struct lu_site_bkt_data *bkt;
410 bkt = lu_site_bkt_from_fid(site, &header->loh_fid);
412 init_waitqueue_entry_current(&waiter);
413 add_wait_queue(&bkt->lsb_marche_funebre, &waiter);
416 set_current_state(TASK_UNINTERRUPTIBLE);
417 if (atomic_read(&header->loh_ref) == 1)
419 waitq_wait(&waiter, TASK_UNINTERRUPTIBLE);
422 set_current_state(TASK_RUNNING);
423 remove_wait_queue(&bkt->lsb_marche_funebre, &waiter);
426 cl_object_put(env, obj);
429 void cl_inode_fini(struct inode *inode)
432 struct ll_inode_info *lli = ll_i2info(inode);
433 struct cl_object *clob = lli->lli_clob;
440 cookie = cl_env_reenter();
441 env = cl_env_get(&refcheck);
442 emergency = IS_ERR(env);
444 mutex_lock(&ccc_inode_fini_guard);
445 LASSERT(ccc_inode_fini_env != NULL);
446 cl_env_implant(ccc_inode_fini_env, &refcheck);
447 env = ccc_inode_fini_env;
450 * cl_object cache is a slave to inode cache (which, in turn
451 * is a slave to dentry cache), don't keep cl_object in memory
452 * when its master is evicted.
454 cl_object_kill(env, clob);
455 lu_object_ref_del(&clob->co_lu, "inode", inode);
456 cl_object_put_last(env, clob);
457 lli->lli_clob = NULL;
459 cl_env_unplant(ccc_inode_fini_env, &refcheck);
460 mutex_unlock(&ccc_inode_fini_guard);
462 cl_env_put(env, &refcheck);
463 cl_env_reexit(cookie);
468 * return IF_* type for given lu_dirent entry.
469 * IF_* flag shld be converted to particular OS file type in
470 * platform llite module.
472 __u16 ll_dirent_type_get(struct lu_dirent *ent)
475 struct luda_type *lt;
478 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
479 const unsigned align = sizeof(struct luda_type) - 1;
481 len = le16_to_cpu(ent->lde_namelen);
482 len = (len + align) & ~align;
483 lt = (void *)ent->lde_name + len;
484 type = IFTODT(le16_to_cpu(lt->lt_type));
490 * build inode number from passed @fid */
491 __u64 cl_fid_build_ino(const struct lu_fid *fid, int api32)
493 if (BITS_PER_LONG == 32 || api32)
494 RETURN(fid_flatten32(fid));
496 RETURN(fid_flatten(fid));
500 * build inode generation from passed @fid. If our FID overflows the 32-bit
501 * inode number then return a non-zero generation to distinguish them. */
502 __u32 cl_fid_build_gen(const struct lu_fid *fid)
507 if (fid_is_igif(fid)) {
508 gen = lu_igif_gen(fid);
512 gen = (fid_flatten(fid) >> 32);
516 /* lsm is unreliable after hsm implementation as layout can be changed at
517 * any time. This is only to support old, non-clio-ized interfaces. It will
518 * cause deadlock if clio operations are called with this extra layout refcount
519 * because in case the layout changed during the IO, ll_layout_refresh() will
520 * have to wait for the refcount to become zero to destroy the older layout.
522 * Notice that the lsm returned by this function may not be valid unless called
523 * inside layout lock - MDS_INODELOCK_LAYOUT. */
524 struct lov_stripe_md *ccc_inode_lsm_get(struct inode *inode)
526 return lov_lsm_get(ll_i2info(inode)->lli_clob);
529 void inline ccc_inode_lsm_put(struct inode *inode, struct lov_stripe_md *lsm)
531 lov_lsm_put(ll_i2info(inode)->lli_clob, lsm);